J Mater Sci (2007) 42:87658770
DOI 10.1007/s10853-007-1952-8
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Nano- and micromechanical properties of hierarchical biological
materials and tissues
Markus J. Buehler
Published online: 7 August 2007
Springer Science+Business Media, LLC 2007
Abstract The mechanical properties of biological materi-
als have been the focal point of extensive studies over the past
decades, leading to formation of a new research field that
intimately connects biology, chemistry and materials sci-
ence. Significant advances have been made in many disci-
plines and research areas, ranging throughout a variety of
material scales, from atomistic, molecular up to continuum
scales. Experimental studies are now carried out with
molecular precision, including investigations of how
molecular defects such as protein mutations or protein
knockout influence larger length- and time-scales. Simula-
tion studies of biological materials now range from electronic
structure calculations of DNA, molecular simulations of
proteins and biomolecules like actin and tubulin to contin-
uum theories of bone and collagenous tissues. The integration
of predictive numerical studies with experimental methods
represents a new frontier in materials research. The field is at
a turning point when major breakthroughs in the under-
standing, synthesis, control and analysis of complex biolog-
ical systems emerge. Here we provide a brief perspective of
the state of this field and outline new research directions.
Introduction
The field of mechanical properties of biological materials
has seen an exciting development over the past several
M. J. Buehler (&)
Laboratory for Atomistic and Molecular Mechanics, Department
of Civil and Environmental Engineering, Massachusetts Institute
of Technology, 77 Massachusetts Ave. Room 1-272, Cambridge,
MA, USA
e-mail: mbuehler@MIT.EDU
years, partly due to the emergence of physical science
based approaches in the biological sciences, leading to
cross-disciplinary investigations of materials, structures,
diseases as well as the development of new treatment and
diagnostics methods.
This special issue of the Journal of Materials Science
entitled Nano- and micromechanical properties of hier-
archical biological materials: Linking mechanics, chemis-
try and biology constitutes a snapshot of the current
status, new directions and research opportunities in this
field. The focus of the collection of papers is on experi-
mental, computational and theoretical efforts that contrib-
ute to a more quantitative understanding of biological
materials and the interaction of biological materials with
their environment, across various scales. This includes the
deformation and fracture behavior of biological materials,
with a particular focus on nanoscale features and materials,
and its relationship with diseases.
The issue also contains contributions that focus on the
synthesis and growth of biological tissues and materials, in
particular such methods that advance the ability to control
nano- and micro-structural features more quantitatively.
Individual contributions emphasize on linking the chemical
or molecular, and mesoscopic structures of these materials
to macroscopic engineering properties, across various
scales, including the impact of genetic mutations, solvent
conditions and other chemical stimuli on the macroscopic
behavior of the material. Specific topics include elasticity,
deformation and fracture of biological and biomimetic
materials including bone, tissues and scaffolding materials,
interactions of cells with materials under mechanical load-
ing, mechanics of hierarchical biological materials as well
as the mechanics of single molecules. Development of
experimental protocols to study the human stratum corne-
um, addressing the role of misfolded proteins in Alzheimers
123
8766
disease, development of theories of the properties of viral
capsids and elucidation of the statistical mechanics princi-
ples of stretching biopolymers in geometric confinement
represent a broad range of topics.
Materials science of biological materials: challenges and
opportunities
Historically, the use of classes of materials has been used
to classify stages of civilizations, ranging from stone age
more than 300,000 years ago, to the bronze age, and pos-
sibly the silicon age in the late 20th and early 21st century.
However, a systematic analysis of materials in the context
of linking chemical and physical concepts with engineering
applications has not been achieved until very recently. For
instance, 50 year ago, E. Orowan, M. Polanyi and G.I.
Taylor have discovered dislocations, a concept proposed
theoretically in 1905 by V. Volterra. It was discovered that
dislocations represent the fundamental mechanism of
plastic deformation of metals [1, 2]. Remarkably, it was not
until dislocations and other nano- and microscropic
mechanisms have been understood theoretically that major
breakthroughs have been possible that utilize this knowl-
edge, to enable building airplanes, cars, space shuttles and
more recently, nanodevices, through synthesis of ultra-
strong and heat resistant materials, for instance.
Perhaps, today we stand at another cross-road: Biological
materials and systems are vital elements of life, and there-
fore, a rigorous understanding of the matter that makes life
work is in reach. This may enable us eventually to inte- randomly over the volume in crystalline materials, biolog-
grate concepts from living systems into materials design,
seamlessly. Optical, mechanical and electrical properties at
ultra-small material scales, their control, synthesis and
analysis as well as their theoretical description represent
major scientific and engineering challenges and opportuni-
ties. However, just like in the case of more conventional
materials, these breakthroughs will probably only be
accessible provided that the fundamentals are understood
very well. Characterization of the materials found in biology
within a rigorous materials science approach is aimed
towards the elucidation of these fundamental principles of
assembly, deformation and fracture of these materials.
Deformation and fracture properties are intimately
linked to the atomic microstructure of the material.
Whereas crystalline materials show mechanisms such as
dislocation spreading or crack extension, biological mate-
rials feature molecular unfolding or sliding, with a partic-
ular significance of rupture of chemical bonds such as
hydrogen bonds, covalent cross-links or intermolecular
entanglement. Much different mechanisms operate at larger
length scales, where the interaction of molecules with cells
and of cells with one another, different tissue types and the
123
J Mater Sci (2007) 42:87658770
influence of tissue remodelling become more evident. The
dominance of specific mechanisms is controlled by geo-
metrical parameters as well as the structural arrangement of
the protein elementary building blocks, across many hier-
archical scales, from nano to macro (Fig. 1).
It is known from other fields in materials science that
nano- or microscopic structures control the macroscopic
material behavior: For example, grain size reduction or
confinement leads to an increase of the strength of crystal-
line metals [36]. Deformation maps have been proposed to
characterize material properties for engineering applications
[7]. Discovering similar insight for biological structures and
materials represents and important frontier of research. A
particularly challenging question is the elucidation of the
significance and role of nanostructures for macroscopic
properties, that is, carrying out sensitivity analyses that show
how small-scale features influence larger scale properties.
A major trait of biological materials is the occurrence of
hierarchies and, at the molecular scale, the abundance of
weak interactions (e.g. H-bonds). The presence of hierar-
chies in biological materials may be vital to take advantage
of molecular and sub-molecular features, often character-
ized by weak interactions, and multiply their properties so
that they become visible at larger scales, in order to provide
a link between structural organization and function [8].
Utilization of weak interactions makes it possible to pro-
duce strong materials at moderate temperatures and thus
with limited energy use. An important distinction between
traditional and biological materials is the geometrical
occurrence of defects. While defects are often distributed
ical materials consist of an ordered structure that reaches
down to the nano-scale. In many biological materials,
defects are placed with atomistic or molecular precision,
and may play a major role in the material behavior observed
at larger scales. These features have been observed in bone,
nacre, collagenous tissue or cellular protein networks.
The mechanical properties of biological materials have
wide ranging implications for biology. In cells for instance,
mechanical sensation is used to transmit signals from the
environment to the cell nucleus in order to control tissue
formation and regeneration [9, 10]. The structural integrity
and shape of cells is controlled by the cells cytoskeleton,
which resembles an interplay of complex protein structures
and signaling cascades arranged in a hierarchical fashion
[9]. Bone and collagen, providing structure to our body, or
spider silk, used for prey procurement, are examples of
materials that have incredible elasticity, strength and
robustness unmatched by many man-made materials,
mainly attributed to its structural formation with molecular
precision [1119]. The transfer of concepts observed in
biology into technological applications and new materials
design remains a big challenge, with potential huge payoff.
J Mater Sci (2007) 42:87658770
Fig. 1 Overview over different material scales, from nano to macro,
here exemplified for collagenous tissue [1115]. Biological materials
such as collagen, skin, bone, spider silk or cytoskeletal networks in
cells feature complex, hierarchical structures. The macroscopic
mechanical material behavior is controlled by the interplay of
properties throughout various scales. In order to understand defor-
In particular, the combination of nanostructural and hier-
archical features into materials developments could lead to
significant breakthroughs.
What are the most promising strategies in order to
analyze these materials? Perhaps, an integrated approach
that uses experiment and simulation concurrently could
evolve into a new paradigm of materials research. Exper-
imental techniques have gained unparalleled accuracy in
both length- and time scales (see Fig. 2), as reflected in
development and utilization of Atomic Force Microscope
(AFM) [20, 21], optical tweezers [22, 23] or nanoinden-
tation [24] to analyze biological materials [25]. At the same
time, modeling and simulation have evolved into predictive
tools that complement experimental analyses (see Fig. 2).
It is now achievable to start from smallest scalescon-
sidering electrons and atoms, to reach all the way up to
macroscopic scales of entire tissues [26], by explicitly
considering the characteristic structural features at each
scale. Even though there are still major challenges ahead of
us, this progress is amazing and provides one with infinite
possibilities and potentials, transforming materials science
as a discipline through increased integration of computa-
tional approaches in scientific research.
Linking the scales: cross-scale interactions
A central theme of the efforts in developing the materials
science of biological materials is to appreciate the struc-
8767
mation and fracture mechanisms, it is crucial to elucidate atomistic
and molecular mechanisms at each scale, and to appreciate the cross-
scale interaction of these mechanisms. Our ability to synthesize,
characterize and control such materials in their native environment or
in technological applications depends critically on the theoretical
foundation of its mechanical behavior
ture-property or structure-processing-property paradigm,
constituting the heart of the materials science community.
This paradigm has guided materials science for many
decades. For biological materials, there are many chal-
lenges that make developing these rigorous links increas-
ingly difficult.
For example, bond energies in biological materials are
often comparable to the thermal energy, as for instance in
the case of hydrogen bonding, the most abundant chemical
bond in biology. Biological materials show highly visco-
elastic behavior, since their response to mechanical
deformation is intrinsically time-dependent. In many cases,
biological structures contain extremely compliant fila-
ments, in which entropic contributions to free energy are
important and can even control the deformation behavior.
Many material properties are also length scale dependent
and can vary significantly across various length scales.
Quite often, this can be quite perplexing, since measuring
different volumes of material lead to different values of
Youngs modulus. Size effects very strong and possibly
utilized systematically to ensure physiological functioning
of the material in its biological context. However, why and
how these size effects are exploited within this context
remains less understood. The presence of hierarchical
structures calls for new paradigms in thinking about the
structure-property paradigm, since corresponding concepts
must include an explicit notion of the cross-scale and inter-
scale interactions [27].
123
8768
Fig. 2 Overview over various computational and experimental tools.
Hierarchical coupling of different computational tools can be used to
traverse throughout a wide range of length- and time scales. Such
methods enable to provide a fundamental insight into deformation and
fracture phenomena, across various time- and length-scales. Hand-
shaking between different methods enables one to transport informa-
tion from one scale to another. Eventually, results of atomistic,
molecular or mesoscale simulation may feed into constitutive
equations or continuum models. While continuum mechanical
theories have been very successful for crystalline materials, biological
materials require statistical theories. Experimental techniques such as
Atomic Force Microscope (AFM), Molecular Force Spectroscopy
(MFS), nanoindentation or optical tweezers now overlap into
atomistic and molecular approaches, enabling direct comparison of
experiment and simulation
It has become evident that the atomistic scale, and in
particular the notion of a chemical bond, provides a very
fundamental, universal platform at which a variety of sci-
entific disciplines can interact. Chemists, through the
molecular structure of proteins, physicists, through the
statistical mechanics of a large number of atoms, and
materials scientists through analysis of phenomena such as
elasticity, optical properties, electrical properties or ther-
modynamics, linking structure and function (see Fig. 3).
Fig. 3 Chemistry is the most
fundamental language of
materials science. Many other
disciplines can link up with the
notion of a chemical bond that
defines the structure and
eventually the properties of
materials, thereby representing a
joint root for these disciplines
123
J Mater Sci (2007) 42:87658770
A particularly exciting aspect of the materials science
of biological materials is that it is interdisciplinary, by
nature. Nature doesnt know of scientific disciplines,
since they were invented by humans many centuries ago.
Performing research in this field thus often means to
overcome barriers between scientific disciplines and to
develop strategies that enable us speak to each other
more openly. Structures in universities and research
institutions may have to be modified to facilitate such
investigations.
It is vital overcome the barrier that currently separates
the scales, through development of new methods, better
model systems and an advanced appreciation for a multi-
scale view, in order to fully understand multi-scale or
cross-scale interactions. To facilitate these developments,
we must also develop a proper nomenclature to capture
the various scales involved in a material. Current ter-
minologies referring to atomistic, meso, micro and macro
are insufficient to capture the subtleties of the various
scales. Research should address the question, what are
the opportunities in integrating nanoscience and nano-
technology into biological research? What will and can
our impact be, in a long perspective, in understanding
fundamental biology? For instance, is the nanomechanics
of protein materials significant for biology, and have
biologists missed out on important effects due to lack of
consideration of the nanomechanics? How does Nature
design materials that are environmentally friendly,
lightweight and yet tough and robust and can serve
multiple objectives? How is robustness achieved? How
do universality and diversity integrate into biological
structures?
From a theoretical viewpoint, major challenges are the
development of new materials theories that include atom-
istic and statistical effects into an effective description,
J Mater Sci (2007) 42:87658770
while retaining a system theoretical perspective [2830],
maybe eventually leading to a merger between system
biology and materials science.
Similar to dislocation mechanics for metal plasticity,
what is the theoretical framework for biological materials
and structures? It is possible that statistical theories may
evolve into the theoretical language of nanomechanics.
Atomistic simulations of complex protein structures with
explicit solvents are often prohibitive, and coarse-graining
techniques are often used. However, how effective are
coarse-graining techniques? Can we indeed average out
over atomistic or mesoscale structures? How important are
atomistic features at macroscale? What are the best
numerical strategies to simulate the role of water in very
small confinement? How does confined water influence the
mechanics of natural and biological materials?
Progress in these various challenging fields will proba-
bly occur specific to problems and applications, perhaps in
those have most impact in medical or economic fields.
Eventually, we must generalize our insight into the
formulation of a holistic theory that extends the current
nomenclature, theory and experimental thinking. These
efforts will provide the scientific and engineering funda-
mentals to develop and maintain the infrastructures to
enable and evolve modern civilization. Materialsand
materials sciencewill surely play a seminal role in these
developments.
Future directions, challenges and impact
Over the last centuries, engineers have developed
understanding of how to create complex man-made
structures out of a diverse range of constituents, at var-
ious scales (machines, buildings, airplanes, nuclear
reactors and many others). Increased development and
research funding into these areas of research will lead to
breakthroughs not only on the fundamental sciences, but
also in technological applications. Research in the area
of mechanics of biological materials will extend our
ability to carry out structural engineering, as used for
buildings or bridges today, to the ultimate scalenano-
scale, and may be a vital component of the realization of
nanotechnology.
A better understanding of the mechanics of biological
and natural materials, integrated within complex techno-
logical systems will make it possible to combine living and
non-living environments to develop sustainable technolo-
gies. New materials technologies such as protein-based
materials produced by recombinant DNA techniques rep-
resent new frontiers in materials design and synthesis [31,
32]. These questions have high impact in the understanding
and design of environmentally friendly technologies and
8769
may enhance the quality of life of millions of people,
through advances in the medical sciences as well as
through improvements of the living environment. A cur-
rently pressing question is the development of new tech-
nologies to address the energy problem. Advances may be
possible by utilization of bacteria to produce and process
fuel from crops, or by enabling the synthesis of materials at
reduced processing temperature.
Nanoscience and nanotechnology enable us to make
structures at the ultimate scale (self assembly, recombinant
DNA, utilization of motor proteins for nano-machines and
many others). This will perhaps lead to novel complex
structural materials, designed from nano to macro. The
theoretical progress in understanding hierarchical biologi-
cal materials will facilitate to use an extended physical
space, through the use of multiple hierarchies, in an effi-
cient and controlled manner, that is, lead to a bottom-up
structural design on the sub-macroscopic scale, instead of
trial-and-error approaches. For example, the extended
hierarchical design space might serve as means to realize
new physical realities that are not accessible to a single
scale, such as material synthesis at moderate temperatures,
or fault tolerant hierarchical assembly pathways [33],
which enable biological systems to overcome the limita-
tions to particular chemical bonds (soft) and chemical
elements (organic) present under natural conditions [27].
The increased understanding of the hierarchical design
laws might further enable the development and application
of new organic and organic-inorganic multi-featured
composites (such as assemblies of carbon nanotubes and
proteins or polymer-protein composites [3436]), which
will mainly consist of chemical elements that appear in our
environment in an almost unlimited amount (C, H, N, O,
S). These materials might consequently help to solve
humans energy and resource problems (e.g. fossil
resources, iron etc.), and allow us to manufacture nano-
materials, which will be produced in the future by tech-
niques like recombinant DNA [31, 37, 38] or peptide self-
assembly [32, 39, 40], techniques where the borders
between materials, structures and machines vanish.
Applications of these new materials and structures are
new biomaterials, new polymers, new composites, engi-
neered spider silk, new scaffolding tissues, improved
understanding of cell-ECM interactions, cell mechanics,
hierarchical structures and self-assembly. In addition to the
long-term impact in biology, bioengineering and medicine,
this research may eventually contribute to our theoretical
understanding of how structural features at different scales
interact with one another. In light of the extended physical
design space discussed above, this may transform engi-
neering approaches not only for materials applications,
but also in manufacturing, transportation or designs of
networks.
123
8770 J Mater Sci (2007) 42:87658770

Acknowledgements Id like to sincerely thank all authors and 17. Doyle J (2007) Nature 446:860
contributors to this special issue for submitting excellent illustrations 18. Kitano H (2002) Nature 420(6912):206
of exciting research work in this field to this special issue. I am 19. Kitano H (2002) Science 295(5560):1662
delighted to have such a strong collection of papers from outstanding 20. Smith BL et al (1999) Nature 399(6738):761
contributors, from many different countries, throughout a range of 21. Prater CB, Butt HJ, Hansma PK (1990) Nature 345(6278):839
scientific disciplines. 22. Sun YL et al (2004) J Biomech 37(11):1665
23. Dao M, Lim CT, Suresh S (2003) J Mech Phys Solids 51(11
12):2259
24. Tai K, Ulm FJ, Ortiz C (2006) Nanogranular origins of the
References
strength of bone. Nano Lett 6:2520
25. Lim CT et al (2006) Mater Sci Eng C-Biomimetic Supramol Syst
1. Taylor GI (1934) Proc Roy Soc A 145:362 26(8):1278
2. Hirth JP, Lothe J (1982) Theory of dislocations. Wiley-Inter- 26. Goddard WA (2006) A Perspective of materials modeling In: Yip
science, Hoboken, NJ S (ed) Handbook of materials modeling. Springer, New York
3. Wolf D et al (2003) Z Metallk 94:1052 27. Ackbarow T, Buehler MJ Hierarchical coexistence of universality
4. Yip S (1998) Nature 391:532 and diversity controls robustness and multi-functionality in
5. von Blanckenhagen B, Gumbsch P, Arzt E (2001) Modelling protein materials (unpublished)
Simul Mater Sci Eng 9:157 28. Bell GI (1978) Science 200(4342):618
6. Nieh TG, Wadsworth J (1991) Scripta Met 25(4):955 29. Lakes R (1993) Nature 361(6412):511
7. Frost HJ, Ashby MF (1982) Deformation-mechanism maps. 30. Bustamante C et al (1994) Science 265(5178):1599
Pergamon Press, Oxford 31. Langer R, Tirrell DA (2004) Nature 428(6982):487
8. Fraser P, Bickmore W (2007) Nature 447(7143):413 32. Zhao XJ, Zhang SG (2007) Macromol Biosci 7(1):13
9. Alberts B et al (2002) Molecular biology of the cell. Taylor & 33. Holland JH (1995) Hidden orderhow adaptation builds
Francis, Boca Raton complexity. Helix Books, Reading, MA
10. Engler AJ et al (2006) Cell 126(4):677 34. Cui XQ et al (2007) J Phys Chem C 111(5):2025
11. Buehler MJ (2007) Nanotechnology 18:295102 35. Hule RA, Pochan DJ (2007) MRS Bull 32(4):5
12. Buehler MJ (2006) P Natl Acad Sci USA 103(33):12285 36. Winey KI, Vaia RA (2007) MRS Bull 32(4):5
13. Fratzl P et al (2004) J Mater Chem 14(14):2115 37. Petka WA et al (1998) Science 281(5375):389
14. An KN, Sun YL, Luo ZP (2004) Biorheology 41(34):239 38. Smeenk JM, et al (2005) Angew Chem Int Ed 44(13):1968
15. Ramachandran GN, Kartha G (1955) Nature 176:593 39. Zhao XJ, Zhang SG (2006) Chem Soc Rev 35(11):1105
16. Currey JD (2002) Bones: structure and mechanics. Princeton 40. Mershin A et al (2005) Nat Biotechnol 23(11):1379
University Press, Princeton, NJ

123
J Mater Sci (2007) 42:87718787
DOI 10.1007/s10853-007-1719-2
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Superelasticity, energy dissipation and strain hardening
of vimentin coiled-coil intermediate filaments: atomistic
and continuum studies
Theodor Ackbarow Markus J. Buehler
Received: 3 January 2007 / Accepted: 23 March 2007 / Published online: 10 July 2007

Springer Science+Business Media, LLC 2007
Abstract Vimentin coiled-coil alpha-helical dimers are
elementary protein building blocks of intermediate fila-
ments, an important component of the cells cytoskeleton
that has been shown to control the large-deformation
behavior of eukaryotic cells. Here we use a combination of
atomistic simulation and continuum theory to model tensile
and bending deformation of single alpha-helices as well
as coiled-coil double helices of the 2B segment of the
vimentin dimer. We find that vimentin dimers can be
extended to tensile strains up to 100% at forces below
50 pN, until strain hardening sets in with rapidly rising
forces, approaching 8 nN at 200% strain. We systemati-
cally explore the differences between single alpha-helical
structures and coiled-coil superhelical structures. Based on
atomistic simulation, we discover a transition in deforma-
tion mechanism under varying pulling rates, resulting in
different strength criteria for the unfolding force. Based on
an extension of Bells theory that describes the dependence
of the mechanical unfolding force on the pulling rate, we
develop a fully atomistically informed continuum model of
the mechanical properties of vimentin coiled-coils that is
capable of predicting its nanomechanical behavior over a
wide range of deformation rates that include experimental
conditions. This model enables us to describe the
mechanics of cyclic stretching experiments, suggesting a
hysteresis in the forcestrain response, leading to energy
dissipation as the protein undergoes repeated tensile
loading. We find that the dissipated energy increases
continuously with increasing pulling rate. Our atomistic
T. Ackbarow
M. J. Buehler (&)
Laboratory for Atomistic and Molecular Mechanics, Department
of Civil and Environmental Engineering, Massachusetts Institute
of Technology, Cambridge, MA 02139, USA
e-mail: mbuehler@MIT.EDU
and continuum results help to interpret experimental
studies that have provided evidence for the significnifi-
cance of vimentin intermediate filaments for the large-
deformation regime of eukaryotic cells. We conclude that
vimentin dimers are superelastic, highly dissipative protein
assemblies.
Introduction: the structure of vimentin intermediate
filaments and role in eukaryotic cells
Together with beta sheets, alpha helical (AH) structures are
the most abundant secondary structures found in proteins.
These two patterns are particularly common because they
result from hydrogen bonding between the NH and C=O
groups in the polypeptide backbone. An alpha helix is
generated when a single polypeptide chain twists around on
itself stabilized by hydrogen bonds (H-bond) made
between every fourth residue, linking the O of peptide i to
the N of peptide i + 4 in the residue chain. Consequently,
at each convolution, three H-bonds are found in parallel
arrangement that stabilize the helical configuration [1].
Another particularly stable configuration, found for the
first time in keratin intermediate filaments (IFs) about
50 years ago, are alpha helical coiled-coils, where the
primary structure reveals a pronounced seven residue
periodicity (abcdefg)n, called a heptad repeat. Within this
repeat, positions a and d are preferably occupied
with nonpolar residues [2] such as LEU, ALA, VAL or
ILE. The hydrophobic residuesconsequently concen-
trated on one sideare the reason for the coiled-coil
structure. In order to avoid contact with surrounding water
molecules, AHs assemble into coiled-coils by wrapping
123
8772
around each other and clustering the hydrophobic side
chains inside [1]. Additionally, interhelical and intrahelical
salt bridges contribute to coiled-coil thermodynamic sta-
bility and are expected to enhance resistance to mechanical
stretch in a nonstereospecific manner [3].
Coiled-coils are also the building blocks of vimentin
intermediate filament (IF) dimers, which are composed of a
head, a tail, and an extremely elongated central rod-do-
main. A schematic of the vimentin dimer structure is
shown in Fig. 1a. The rod-like structure is 310 residues
long and consists of four coiled-coil alpha helices (1A, 1B,
2A, 2B) divided by linkers (L1, L12, L2) [2, 4, 5, 7, 8].
Interestingly, all helices in the rod domain have different
lengths.
The lengths of each of the components are absolutely
conserved for all types of IFs in different types of cells.
Additionally, either ends of the rod as well as the position
of the stutter in the 2B segment are highly conserved (a
definition of various biological terms and concepts is pro-
vided in Table 1) [2, 4, 5, 9]. The stutter is an irregularity
in the periodic sequence, where four extra residues are
inserted into the continuous heptad repeat, what results in
an almost parallel run of both helices without interrupting
the coiled-coil geometry. For all types of IFs, the stutter is
spaced precisely six heptads away from the C-terminal end
of coil 2B [4]. In contrast to the conserved regions, the
head and tail domain are greatly diverse for all types of IFs
[2, 4].
Fig. 1 Geometry of vimentin intermediate filament, from atomistic
to macromolecular structure. Subplot (a): The dimers, approximately
45 nm long, are the elementary building blocks of vimentin
intermediate filaments. A dimer consists of a head, tail (plotted in
red) and an elongated rod domain which is divided into four alpha-
helical coiled coils (1A, 1B, 2A, 2B) connected through linkers L1,
L12, L2 (also red) [2]. The molecular dynamics simulations described
in this paper are performed on alpha-helices, placed in the 2B
segment (yellow). Subplot (b): Intermediate filaments assembly
hierarchically into fibrils. Dimers, the elementary building blocks,
assembly through a half-staggered, anti-parallel overlap into tetra-
mers, which associate laterally into 13 nm thick unit-length-filaments
(ULFs). In the next step, ULFs assembly longitudinally into more
compact filaments (10 nm in diameter) with a length of more than
240 nm [2, 46]
123
J Mater Sci (2007) 42:87718787
The five types of IFs and the three assembly groups are
listed in Table 2. The most widely distributed of all IF
proteins is vimentin, typically expressed in leukocytes,
blood vessel endothelial cells, some epithelial cells, and
mesenchymal cells such as fibroblasts [1].
The role of IF networks in the cytoskeleton
IFs, in addition to microtubules (MTs) and microfilaments
(MFs) are one of the three major components of the
cytoskeleton in eukaryotic cells [11]. The cytoskeleton
plays a critical role in determining the shape and the
mechanical properties of the cell, and is vital for numerous
additional functions such as cell motility or protein
synthesis [1113].
Like many other biological materials, IFs are hierar-
chical structures with highly specific features at nanoscale.
Vimentin IF dimers, shown schematically in Fig. 1a, are
the elementary building blocks of IFs and thus represent
the first level of hierarchy. Through following the different
steps of assembly (Fig. 1b) [14, 15], dimers associate to
fibrils, which build the second level of hierarchy. In vivo,
these fibers can reach a length of up to several lm and
consist of 16 dimers in cross-section. The third level of
hierarchy are three-dimensional IF-networks inside the
cell, reinforcing the plasma membrane [4, 12, 16]. Inside
the network, IF-associated proteins such as plectins gen-
erate the connection between individual filaments.
The cells IF networks are connected with other cells
networks and with the extracellular matrix at the plasma
membrane [1]. This architecture guarantees that tensile and
shear loads applied to the tissue can be carried by IF net-
works.
The fact that IFs span from the cell nucleus to the cell
membrane, and interact with IF networks of other cells
suggest that IFs play an important role in transmitting
mechanical signals from the plasma membrane to the nu-
cleus, where a specific response can be triggered by
mechanical stimulation [10, 17].
Plakin-type cross-bridging proteins, also known as cyt-
olinkers (e.g. plectins or desmoplakins) link all three
cytoskeletal networks (MTs, MFs and IFs). These proteins
attach the IFs to MTs, MFs or adhesion complexes [18].
However, to date little is known about these interaction or
the mechanical functions of plakin-proteins and their
ability to store or/and dissipate any elastic energy during
internal contraction or external deformation [2, 11].
In contrast to MTs and MFs, IFs do not participate in the
dynamic functions of the cytoskeleton. Further, they do not
support active transport of motor proteins such as myosin
and kinsesin, due to the missing polarity in the protein
structure (in contrast to MT and MF networks) [2]. They do
not participate in any cell movement [1]. These evidences
J Mater Sci (2007) 42:87718787 8773

Table 1 Summary of different biological terms and concepts used in this paper
Cytoskeleton: A composite inside the cell consisting of three different networks: Actin filaments, microtubules and intermediate
filaments. The intermediate filament network is in the focus of this paper. These networks connect the nucleus
(nuclear membrane) with the plasma membrane and are furthermore responsible for the organization inside the cell.
Intermediate filaments One of the three components of the cytoskeleton; mainly responsible for the large deformation behavior of the cell.
(IFs):
Cross bridging Cross bridging proteins form connections inside each cytoskeletal network as well as connections between different
proteins: networks.
Dimer: A dimer is the elementary building block of an IF fiber. This protein consists of a head domain, a tail domain and an
extremely elongated coiled-coil rod. A coiled-coil is a superhelix that consists of two alpha helices that twist around
each other.
Assembly: Individual IF dimers assemble systematically and hierarchically into filaments (Fig. 1b). Two dimers build a tetramer,
two tetramers build an octamer and four octamers build a unit length filament (ULF). Once this level of assembly is
reached, ULFs ally longitudinally into long fibers.
Residue: The primary structure of a protein consists of a sequence of amino acids. One residue is thus one amino acid in the
polypeptide backbone.
Conserved structure: A structure is conserved when parts of the residue sequence are similar or do not very at all between the different
species (e.g. human and fish). For example, certain parts of the IF sequence are very similar between different species
as well as inside the IF protein family (vimentin, desmin, keratin, etc.). Conserved structures often signify a
particular amino acid sequence that has proven to be particularly suitable for a specific task, and has thus been kept
identical during the evolutionary process.

Table 2 Different types of intermediate filaments, its location in the cell and its assembly group. Intermediate filaments are classified into five
different types. Most of them appear in the cytoskeleton, except lamins that are found in the nucleoskeleton. Keratins, in contrast to the other IF
types, assemble into heterodimers, consisting of one acidit and one basic keratin [6, 10]
# Type Location Assembly group

I. Acidic keratins Cytoskeleton AG I (heterodimer)

II. Basic keratins Cytoskeleton AG I (heterodimer)
III. Vimentin, desmin Cytoskeleton AG II (homodimer)
IV. Neurofilaments Cytoskeleton AG II (homodimer)
V. Lamins Nucleoskeleton AG III (homodimer)
VI. Phakinin, filesin Eye lens cells

underline the specific static-mechanical role of IFs. These
static-mechanical properties are the focus of the work de-
scribed in this paper.
Mechanical properties of IFs compared to other
cytoskeletal components
A great diversity of mechanical properties enables the vi-
mentin IFs to satisfy their specific mechanical role in cells,
such as to guarantee their structural integrity or their shape.
It has been hypothesized that IFs are critical to provide
strength to the cell under large deformation, and to absorb
large amounts of energy upon a certain load by unfolding
[19, 20]. This represents a means to reinforce the cell in
extreme deformations so that cells can withstand dramatic
loads and deformations [11, 19, 20].
IFs exhibit a highly nonlinear stressstrain relationship
with a high resistance against rupture, also known as strain
hardening. Figure 2 depicts experimental results [7] of the
stressstrain response of gels composed of the three cyto-
skeletal proteins. In these rheological experiments, gels
with equal weight concentrations were sheared and the
deformation response (strain) was measured. It is apparent
that vimentin gels are capable of sustaining large defor-
mation at large forces. In contrast, actin filaments rupture at
low strains but large forces, and MTs break at moderately
large strains, but small forces. This underlines the signifi-
cance of vimentin IFs to carry large forces at large defor-
mation.
The stiff behavior of MTs and MFs may be one reason
for much smaller breaking strains of approximately 60%
for MT and 20% for MF in networks of equal weight
concentration (see Fig. 2) [7]. In contrast, IFs feature much
higher extensibility. It has been suggested that a higher
flexibility of IFs at small strains (compared with MFs)
results in a lower mechanical resistance during cell
movement, mainly performed by MFs. It has also been
shown that the rigidity of circulating lymphocytes and fi-
broblasts primary depends on vimentin IFs, whereas MTs
play a minor role [11, 2123]. The different mechanical
properties of the cytoskeletal networks clearly indicate that
the cytoskeleton is a composite with a range of mechanical

123
8774 J Mater Sci (2007) 42:87718787

5 Observations in rupture experiments on single IFs have

shown a dramatic change in filament diameter, which re-
4 mained unchanged for hours after rupture appeared [24].
stress in Pa

This is a sign for a profound change in the molecular

3 X
2 the alpha-helical coiled-coil are converted into beta-sheet-
f-actin
1 vimentin
X microtubules
0 mechanical, molecular properties of the coiled-coil dimer
0% 20% 40% 60% 80% 100%
strain
Fig. 2 Shearing experiments carried out with gels of equal weight
concentration demonstrate the differences in the mechanical proper-
ties of various cytoskeletal networks. In contrast to vimentin that
sustains strains much larger than 80%, MT break at 60% strain, and
MF break at 20% strain. Additionally, vimentin gels exhibits
continuous significant strain hardening. We note that this stress strain
curve is different from the behavior of a single protein as reported in
this paper due to the different level of hierarchy. Data source: Janmey
et al. [7]
properties, which cannot be achieved by a polymer network
composed out of a single type of polymer.
Further, experiments have revealed that the mechanical
properties of vimentin determines the cell stiffness in
particular for high stress and large deformation [4, 11, 12,
2426]. For instance, in shear tests vimentin deficient cells
were shown to be 40% less stiff at high strains compared
with wild-type cells (see Fig. 3), while their elastic prop-
erties do not change much under small deformation [11].
These experiments strongly support the notion that the
biological significance of vimentin IFs lies in the large-
deformation regime.
8
stiffness in Pa
architecture under large deformation. It was suggested that
type structures with both strands being aligned along the
filament axis. This provided some evidence that the
mechanical properties of IFs mainly depend on the nano-
[24]. However, no direct experimental, simulation or the-
oretical prove has been reported thus far.
The parallel assembly of dimers into filaments (see
Fig. 1b) may allow for slip between dimers under large
deformation. In addition, presence of linkers in the rod
domain [26] lead to a higher flexibility of IFs and thus to a
smaller bending stiffness and persistence length of the fil-
aments (0.31.0 lm) compared to MTs (18 mm, due to
geometry) and MFs (310 lm) [4, 12].
Finally, the mechanical role of intermediate filaments is
particularly evident in diseases in which the loss of
mechanical function and integrity of various tissues is
associated with intermediate-filament-protein mutations [6,
27]. It was shown that mutations in keratin IFs reduce the
ability of these IF networks to bundle and to resist large
deformation [11]. Furthermore, it has been suggested that
point mutations lead to the aggregation of the cytoskeleton
and extensive cell fragility in epidermis, heart and skeletal
muscle after they are exposed to mechanical strain [28].
Mechanics of similar alpha helical and coiled-coil
structures
In recent years, a variety of different alpha-helical struc-
tures and coiled-coils were studied in experiment as well as
in simulation [2835].

6 AFM experiments on single molecules of double-headed

myosin, single-headed myosin as well as coiled-coil tail
4 fragments were reported in [29, 31]. It was found that the
transition to unfolding of the protein structure (in the fol-
2 lowing referred to as angular point or unfolding force)
wild-type
vimentin-deficient
begins at strains of about 20% of stretched protein length.
0 Furthermore, it was shown that myosin is a very elastic
0 2 4 6 8 protein, with almost no hysteresis at small pulling rates.
stress in Pa Even if the coiled-coil structure has been unfolded com-
pletely under mechanical forces, it refolds again to its
Fig. 3 Stiffness of cells as a function of stress state, comparing wild-
type and vimentin deficient cells. It was shown in experiments that
initial configuration in less than one second [29]. Some of
vimentin deficient cells are much less stiff at higher stresses than the characteristics of myosin deformation during tensile
wild-type cells. These results suggest that vimentin proteins play a tests were also observed in MD simulations that were
critical role in particular for the large-deformation elastic properties carried out on parts of the coiled-coil structure [31].
of cells. It also corroborates the notion that due to the progressive
strain hardening at large strains, IFs can be understood as security
A multi-scale model for human hair that mainly consists
belts of the cell that operate after MTs and MFs have ruptured [3]. of keratin IF coiled-coils was developed in [30]. MD
Data source: Wang et al. [11] simulations on coiled-coils have shown force-deformation

123
J Mater Sci (2007) 42:87718787
characteristics similar to the one observed for myosin.
Additionally, a pulling rate dependence of the coiled-coil
structure was found, and it was suggested that the non-
equilibrated system (due to the very high pulling rates in
MD) is the reason for this behavior.
Furthermore, some simulations were carried out on
single alpha helices [32, 33]. By applying tensile loads,
unfolding of the helix was observed after a short steep
increase in force. Thereby, the 16 residues long helix in
[32] started to unfold at both ends simultaneously, in
contrast to the 20-residue long amino acid chain modeled
in [33], which unfolded systematically from the side where
force was applied.
Research strategy and outline
Experiments on entire cells have provided strong evidence
for the importance of mechanical properties of vimentin
IFs on the large deformation behavior of cells (see e.g.
Figs. 2 and 3). However, it remains unclear if the origin of
these effects lies at the molecular protein level, at the level
of dimers, or if it is a consequence of larger-scale structural
features of the vimentin protein network [4, 24].
Only few alpha-helical structures have been analyzed in
experiments or in simulations. Further, until now, neither
any systematic studies on the difference between single
AHs and coiled-coils were carried out, nor the dependence
on the pulling velocity was analyzed in detail. Addition-
ally, no explanations about the highly conserved assembly
of AHs into coiled-coils were suggested that consider
mechanical aspects. A structure-property link for the three
deformation regimes and associated strength models has
not been reported. Earlier MD simulations were carried out
at extremely large strain rates, and no direct link between
simulation and experiment has been reported.
Here we perform a series of atomistic studies of tensile
and bending deformation of vimentin to arrive at a detailed
understanding of the mechanical behavior of these mole-
cules under small and large deformation and at different
pulling velocities. We systematically investigate the pull-
ing rate dependence of the mechanical properties of this
particular structure. Development of a theoretical model
enables us to develop a rigorous understanding of the
pulling rate dependence, which can be used to extrapolate
our results to pulling rates that are comparable to those
applied in experiment. This leads to a direct link between
our simulations and results obtained in experiments.
Moreover, the continuum model enables us to predict the
behavior of the protein under cyclic loading and under
varying pulling rates.
We focus on the coiled-coil structure in the 2B segment
of the vimentin rod domain, since this part of the vimen-
tin protein is highly conserved across various species,
8775
indicating a significant biological function. The goal is to
develop a structure-function relationship on the protein
level and to link these properties with biological cellular
functions of vimentin.
The outline of the paper is as follows. In Theoretical
concepts: Modified Bell theory, we present a modification
to Bells classical model that enables us to describe the
dependence of protein unfolding forces as a function of
pulling rates. In Atomistic modeling methods, we briefly
introduce our atomistic modeling procedure. In Compu-
tational results, we report results of atomistic modeling for
a variety of boundary conditions. Section Atomistically
informed continuum model is dedicated to development of
a continuum model, based on the atomistic simulation
results reported in Computational results. In the last Sect.
Discussion and conclusion we conclude with an extensive
discussion of our results in light of biological function and
earlier experimental results.
Theoretical concepts: Modified Bell theory
Mechanical loading of protein structures can result in
severe changes in the protein structure, inducing unfolding
of the protein. Typically, a variety of unfolding processes
exist for a given protein structure, each of which has a
specific reaction pathway and an associated energy barrier.
The different unfolding modes in the protein can be
understood as the interplay between different unfolding
processes with different activation barriers operating at
different activation distances.
Several theories exist that describe competing processes
due to mechanically induced instabilities of protein struc-
tures. These concepts stem primarily from the field of
physical chemistry [3640]. Most of them are derived from
a theory originally postulated by Bell in 1978 [41].
In Bells theory, the off rate v is the product of a natural
vibration frequency, x0, of the bond in vacuum and the
quasi-equilibrium likelihood of reaching the transition state
with an energy barrier Eb that is reduced by mechanical
energy f
xb, where f is the applied force along the coor-
dinate x, and xb is the distance between the equilibrated
state (minimum of the well) and the transition state.
The off rate, also known from chemical reaction kinet-
ics, is given by


Eb  f
xb
v x0
exp  : 1
kb
T
The off rate describes how often a bond is broken per
unit time and equals to the reciprocal of the lifetime of a
bond. The natural vibration frequency of a bond is
x0  1 13 s1 [41].
123
8776
However, Eq. 1 does not describe the dependence of the
speed at which a bond breaks due to the applied pulling
force. Instead, it only provides an estimate of the time scale
at which the bond will be broken.
In order to overcome this limitation, we modify Eq. 1
based on the following idea: The speed v at which a bond is
broken equals to the distance that needs to be overcome in
order to break the bond (xb) divided by the time for the
bond breaking. Consequently, v is the product of v
xb.
This leads to the following equation for the bond breaking
speed:


Eb  f
xb
v x0
xb
exp  : 2
kb
T
This equation can be rewritten in the following way:


f
xb
v v0
exp ; 3 CMARMM22 force field is a reasonable model for atom-
kb
T
with v0 as the natural bond breaking speed (when no load is
applied), defined as:


Eb
v0 x0
xb
exp  : 4 To apply load, Ca atoms at one end are fixed and the force
kb
T
This modified framework enables us to study the
dependence between the unfolding force and the bond
breaking speed or to calculate the force at which a bond
breaks, at a certain pulling rate. We can rewrite Eq. 3 in the
following way:
kb
T kb
T
f v
ln v 
ln v0 a
ln v b; 5
xb xb
where a kb
T=xb and b - kb
T=xb
ln m0 . Equa-
tion 5 predicts that the unfolding force depends logarith-
mically on the pulling speed. We note that it contains two
parameters a and b, which can be calculated exactly from
the parameters xb and Eb for a certain temperature.
Equation 5 now provides an immediate link between the
pulling rate and the pulling force that is necessary to lover
the energy barrier in such a way that the bond can be
broken with the applied velocity; increasing the pulling rate
means increasing the probability of bond rupture. This is
due to lowering of the energy barrier at the transition point,
because of the applied force f. We emphasize that typically
several unfolding mechanisms exist, each of which is
characterized uniquely by the pair xb and Eb.
We note that multi-transition state energy landscapes
were already predicted in [37]. However, the driving
parameter in the theory discussed in [37] was the loading
rate (the increase in force over time), and not the pulling
rate as done here.
123
J Mater Sci (2007) 42:87718787
A strategy to determine the dependence of the unfolding
force f on pulling speed, associated mechanisms and energy
barriers is to use atomistic modeling, as described in the
following sections.
Atomistic modeling methods
Atomistic modeling
Here we employ atomistic simulation to provide a bridge
between microscopic length- and time scales such as
quantum chemistry, and macroscopic scales such as con-
tinuum mechanics. We use classical molecular dynamics
(MD).
Our MD simulations are carried out with the program
NAMD [42] using the CHARMM22 force field [43]. The
istic interactions within proteins and between different
proteins (including covalent bonds, H-bonds, electrostatic
interactions and vdW interactions).
To apply the forces to the molecule that induce defor-
mation, we use steered molecular dynamics (SMD) [42].
is applied on the Ca atom at the other end. The SMD
technique is equivalent to attaching one end of a virtual
harmonic spring to the end of the system and pulling at the
virtual atom on the other end of the spring [16]. The SMD
method thus mimics an experiment where one end of the
molecule is fixed (e.g. on a gold surface), while the other
end is pulled at with the AFM cantilever tip. Using this
technique, different loading conditions (e.g. tensile and
bending) can be realized.
The force experienced by the virtual atom is given by
[16]:
F kv
t  x: 6
Here, x is the displacement of the pulled atom, m the
pulling velocity, t the time step, and k is the spring con-
stant.
For the tensile loading simulations described in this
paper, the SMD spring constant is k = 10 kcal/mol/A 2.
Different pulling rates (v) used for the simulations are
indicated in the corresponding sections.
In the bending simulations, we chose a fixed pulling rate
v = 0.000,002 A /fs, with a spring constant k = 0.01 kcal/
mol/A 2.
By monitoring the applied force (F) and the position of
the atom that is pulled (x) over the simulation time, we
obtain force-versus-displacement data that is used to derive
the mechanical properties such as bending stiffness or the
Youngs modulus [44].
J Mater Sci (2007) 42:87718787
Molecular strain is defined as e x  x0 =x0 , where
x0 is the initial, undeformed length, and x is the current,
deformed length of the protein structure.
Due to the time scale limitations of MD to several
nanoseconds, there is typically a huge difference between
simulation and experiment with respect to pulling rates.
Experimental rates are six to eight magnitudes smaller than
in MD simulations, which requires additional consideration
in order to interpret MD results in light of experimental
findings.
Tensile and bending simulations were performed at a
temperature of 300 K (NVT ensemble), with temperature
control using a Berendsen thermostat. The time step used
in all atomistic simulations discussed in this article is 1 fs.
We use Visual Molecular Dynamics (VMD) for visu-
alization of protein structures [45].
Initial protein structures
We take structures obtained from X-ray diffraction exper-
iments and stored in the Protein Data Bank (PDB) as the
starting point for our atomistic simulations.
The first structure taken from the PDB is a 52 residue
long coiled-coil from the 2B segment (residues 355-406;
PDB ID 1GK6). The second structure is a single alpha
helix from the same coiled-coil. To obtain the geometry of
a single alpha helix, we extract one of the helices from the
1GK6 PDB file. Hydrogen atoms and charges are assigned
according to pH 7.
The coiled-coil part considered in this paper is colored
in yellow, as shown in Fig. 1a. The structure obtained from
the PDB is solved completely in a water skin that
encompasses the entire protein.
We perform energy minimization and finite temperature
equilibration of all structures simulated before the protein
is loaded by applying the SMD technique.
Computational results
Tensile deformation of single AHs and coiled-coils
Figure 4a shows the loading case for both structures con-
sidered. Figure 4b depicts the force versus strain response
of a single AH and the coiled-coil structure, both carried
out at identical loading rates of 5 m/s. We note that the
force in Fig. 4b is normalized by the number of AHs in the
structure (that is, one for the single AH, and two for the
coiled-coil).
For both cases, we observe three distinct regimes (IIII)
characterizing the stretching dynamics of the protein.
These three regimes are now characterized first for the
8777
coiled-coil structure before the differences between the
single AH and the coiled-coil are described.
In the first regime (I), the helical structure is stretched
homogeneously. At the angular point (indicated as AP
in Fig. 4, and marked with x in the plot), the structure
begins to unfold by rupture of H-bonds, characterized by a
significant change in slope in the forcestrain plot, leading
to regime (II). The AP thus represents the critical force
necessary to induce a structural change in the protein,
corresponding to the unfolding force.
We observe that unfolding is initiated at the end where
the load is applied. In the case of the coiled-coil, each of
the two alpha helices unfolds individually, while the sec-
ondary superhelical arrangement remains intact, until
strains reach more than 100%. Unfolding of each AH
structure is characterized by sequential breaking of H-
bonds. During this regime, the force remains approxi-
mately constant while the entire protein is unfolded at
strains approaching 150%.
Once the complete helix is unfolded, the slope increases
continuously while the secondary superhelical configura-
tion is lost at strains larger than approximately 150%,
eventually leading to stretching of covalent bonds in the
Fig. 4 Tensile experiments of vimentin proteins. Subplot (a) depicts
a schematic of the applied load (left: coiled-coil, right: single AH).
Subplot (b): The forcestrain curves of a single AH-structure and an
alpha-helical coiled-coil, both at a pulling rate of 5 m/s. To enable
better comparison of both curves, the force of the coiled-coil is
divided by the number of helices. The first regime (I) consists of a
steep increase in force until a strain of approximately 25% (referred to
as angular point (AP)) for the coiled-coil and 13% for the single AH.
The first regime is followed by the second regime (II) during which
unfolding of the alpha-helices occurs. The forces at the AP are much
higher for a single AH than for the coiled-coil. In the third regime
(III), a non-linear increase in strain by stretching the backbone is
observed. Thereby the single AH has a much steeper and earlier
increase in force than the coiled-coil structure. The differences
between the single AH and the coiled-coil structure are summarized
in Table 3
123
8778
protein backbone, giving rise to rapidly increasing forces at
large deformation in regime (III).
This unfolding sequence (first unfolding of individual
AHs, then unfolding of the superhelical structure) suggests
that in the coiled-coil case, the individual AHs represent
the weakest link in the structure.
Figure 5a and b depict snapshots of the AH coiled-coil
and the single AH during the three regimes.
Even though the unfolding curves of single AHs and
coiled coil structures are qualitatively similar, there are
several significant quantitative differences.
First, the slope of the single AH in the first regime is
almost two times steeper and leads to much higher forces at
the angular point. The angular point appears at strains of
13% in the single AH structure, compared with 25% for the
coiled-coil. Second, unfolding of the single AH begins at
the residue where the pulling force is applied, but is fol-
lowed by immediate additional rupture of H-bonds at res-
idue 369 and 383 inside the protein. Third, strain hardening
of the single AH sets in at 20% lower strains; 125% in the
single AH case versus 145% in the coiled coil. The strain
hardening increases much steeper in the case of a single
AH. The reason for this difference is likely due to the
uncoiling process of the superhelical structure that is
missing in the pure AH structure. The differences between
the two structures are summarized in Table 3.
To investigate the pulling rate dependence of vimentin
mechanics, we have carried out pulling experiments with
systematically varying pulling rates. Figure 6 depicts
forcestrain curves for different pulling velocities, clearly
illustrating a strong pulling velocity (or, equivalently strain
Fig. 5 Snapshots of the structures during different regimes under
tensile loading (pulling rate of 5 m/s in all cases). Subplot (a):
Snapshots of a coiled-coil during three regimes (regime (I):
homogeneous stretching, regime (II): propagation of unfolding wave,
regime (III): uncoiling of superhelical structure and stretching of
protein backbone). Subplot (b): Snapshots of a single AH structure as
it undergoes unfolding. The coiled-coil unfolds systematically starting
at the point, where force is applied. The H-bonds of the single AH
structure rupture at several convolutions simultaneously (IIa), which
seem to be chosen randomly and is followed by unsystematic
unfolding
123
J Mater Sci (2007) 42:87718787
Table 3 Comparison of the mechanics of single AH with the coiled-
coil in different dimensions. The data were derived from simulations
with a pulling rate of 5 m/s. It clearly indicates that the single AH has
much more irregularities and instabilities compared to the coiled-coil,
suggesting a higher mechanical stability of the coiled-coil structure
Dimension Single AH Coiled-coil
Equilibrated H-bond length in A 3.08 0.29 2.97 0.16
Slope of the first regime in pN/A 186.1 94.9
Force per AH at angular point in pN 930 670
Strain at angular point 13% 20%
Breaking of H-bonds Simultaneous Sequential
Beginning strain hardening at 125% 145%
Strain hardening Very steep Continuous
Fig. 6 Forcestrain curves of a coiled-coil alpha-helical structure at
different pulling rates. We describe the behavior with the developed
continuum model and extrapolate to pulling rates, used AFM
experiments. For pulling velocities of smaller than 5 m/s the strains,
where the regime changes take place are very similar to those found in
experiments and are 20% for the angular point and 120% for the
beginning of the strain hardening. The straight lines refer to the
prediction of our continuum model, for various pulling rates
rate) dependence of the mechanics of AH coiled-coils. We
find that the forcestrain curves show a significant depen-
dence on the pulling velocity. Increases in the pulling
velocity lead to increased slopes in particular in regimes (I)
and (II).
In the following sections, we report a detailed analysis
of the different deformation regimes, linking atomistic
process with the observed forcestrain responses.
First regime (I): stretching of H-bonds
In order to understand the deformation mechanisms and
associated driving forces during the first regime, we ana-
lyze the behavior and the length of the H-bonds as the
applied force is increased continuously.
The first regime is characterized by homogeneous,
elastic stretching of the entire structure, where the elastic
J Mater Sci (2007) 42:87718787 8779

strains are contributed largely from stretching of H-bonds

in the AH arrangements.
Figure 7 depicts the average length of H-bonds, for dif-
ferent end displacements, for the case of a pulling rate of
5 m/s (results shown only for the coiled-coil structure). We
find that the H-bonds have an initial average length of
2.97 A , and show a rather homogeneous length distribution
over the entire sequence. As the lateral loading is increased,
the H-bonds are being stretched until unfolding is initiated
at displacements between 14 A and 20 A . Unfolding occurs
highly localized at the ends of the protein structure.
Under typical large MD pulling rates, most H-bonds in
the protein are never stretched close to their rupture dis-
tance since the localized unfolding mechanism is activated.
However, we observe that the slower the pulling rates, the
more homogeneous the strain distribution at the onset of
unfolding.

Second regime (II): protein unfolding

The second regime starting at the angular point (AP) is

dominated by unfolding of individual AHs due to the
continuous rupture of H-bonds.
Figure 8 depicts the length of the H-bonds of all resi-
dues at different displacements, for the coiled coil geom-
etry (Fig. 8a) and the single AH structure (Fig. 8b). In both
cases, unfolding initiates by significant stretching of H-
bonds at both ends of the proteins where the structure is
either fixed or the pulling load is applied.
Fig. 8 The H-bond rupture dominates the second regime (data from
simulation with v = 1 m/s). Once the H-bond is broken, the
convolution unfolds to a length of approximately 11 A . Subplot (a)
shows the results for the coiled-coil structure. A rupture-wave moving
away from the point of applied force can be observed. Subplot (b):
Results for the single AH structure. Here, beside the boundary effects
at both ends, rupture starts at residue 381, and is propagating first into
the direction, where load is applied and afterwards in the opposite
direction, indicating a more random process. No clear unfolding wave
is observed in this case

Unfolding of the coiled-coil (Fig. 8a) can be character-

Fig. 7 Average length of H-bonds in the coiled-coil vimentin protein.
As the loading is applied, the H-bonds in the protein are stretched
increasingly, resulting in an increase of the average distance of H-
bonds. Unfolding initiates at the end of the protein when the lateral
end displacements reaches 14 A . The italic labels at the x-axis
indicate the molecular strain; the italic labels inside the plot
correspond to the H-bond strain. Apparently, the average H-bond
strain is always smaller than the molecular strain
ized by an unfolding wave that starts at the end where the
load is applied. The unfolding wave represents propagation
of the unfolding front running through the entire protein
structure (from the right to the left in Fig. 8a).
In contrast to the coiled-coil structure, unfolding of the
single AH is less systematic (Fig. 8b). It begins with rup-
ture of H-bonds at residue 381, which is approximately in
the middle of the structure. This effect occurs at random
residues that vary from simulation to simulation.
In order to quantify the speed of the unfolding wave, in
Fig. 9a we plot the position of the H-bond in the equili-
brated protein at the moment of rupture (defined for an H-
bond length of more than 5 A ), for various pulling rates.
We define the speed of the unfolding wave as the slope of
this curve, which is determined by fitting a linear function.
Figure 9b plots this unfolding wave speed as a function of
pulling velocities. The dependence of the unfolding wave
speed on the pulling speed is almost perfectly linear.

123
8780
Fig. 9 Subplot (a): The position of the rupturing H-bond is plotted
over the time. For simple calculation of the wave speed, we defined the
point where pulling load is applied as 0 A . The rupture wave propagates
originating from this point in a very systematic order. We defined the
slope of the linearly fitted function as the wave speed. Subplot (b): The
wave speed equals to the slope of the linear fits in subplot (a). In subplot
(b) we plot the wave speed as a function of the pulling rate. A linear
function fits the relations in a excellent manner. The wave speed is of
about 67% of the pulling speed, what indicates, that the pulling is much
faster than the unfolding. Furthermore, we found the intersection of the
linear fit and the x-axis at a pulling rate of 0.161 m/s
We find that the unfolding wave speed equals to 67% of
the pulling speed, indicating that the unfolding is approx-
imately 33% slower than the pulling speed. Additionally,
Fig. 10 Snapshots of the unfolding dynamics. Breaking of H-bonds
is followed by immediate unfolding of the convolution. However,
after the unfolding, the convolution is not straightened completely,
which does not happen until higher forces are applied during strain
123
J Mater Sci (2007) 42:87718787
we observe that the linear curve intersects the zero line of
the horizontal axes at a pulling velocity of 0.161 m/s. This
means that the deformation wave vanishes at this pulling
speed of vcr = 0.161 m/s.
Vanishing of the unfolding wave at this pulling speed
has an important consequence for the unfolding mecha-
nism. For pulling rates v [ vcr , unfolding is initiated by
localized destruction of the helical structure, while strains
in the other parts of the protein remain relatively low. At
slower rates, strain distribution becomes more homoge-
neous and larger over the entire protein sequence.
Therefore, when v [ vcr , all H-bonds in the coiled-coil
are strained homogeneously, and rupture occurs at a ran-
dom location rather than at the location where the force is
applied, representing a different unfolding mechanism. We
will discuss this mechanism in Pulling rate dependence:
change in mechanism.
Third regime (III): unfolding of the coiled-coil structure
and stretching of protein backbone
In regime (III), a non-linear increase in strain by stretching
the backbone is observed, reaching forces on the order of
several nN at strains approaching 200%.
Figure 10 plots some snapshots of the unfolding pro-
cesses (for the coiled-coil geometry), indicating that the
amino acid backbone is not completely straight immedi-
ately after rupture of H-bonds. This suggests that the
straightening of the unfolded protein happens at forces that
are higher than the H-bond rupture forces. This explains the
continuous and smooth increase of the stiffness during
strain hardening (regime (III), Fig. 4).
We find that the single AH has a much steeper and
earlier increase in force than the coiled-coil structure. This
is because uncoiling of the superhelical structure is missing
for the single AH structure.
In contrast to regime (I) and (II), we observe only
negligible pulling rate dependence in regime (III), when
hardening (regime (III)). Oxygen atoms are in red, nitrogen atoms in
blue. The H-bonds are represented by the light blue lines. The
direction of the pulling force is upwards
J Mater Sci (2007) 42:87718787
deformation is primarily characterized by stretching of
covalent bonds.
Pulling rate dependence: change in mechanism
We believe there exist at least two different modes of
unfolding of coiled-coil structures. The first mode is
characterized by generation of a localized unfolding wave
as described in detail in Second regime (II): protein
unfolding. We first analyze this deformation mode.
As pointed out in Theoretical concepts: Modified Bell
theory, the accessible mechanisms and associated param-
eters Eb and xb are an immediate consequence of the
atomistic structure and atomic interactions. The results of
MD simulations shown in Fig. 6 are now used to extract
activation barrier and activation distance associated with
this mode of deformation.
Figure 11 plots the results of angular point forces from
simulations with varying pulling rates (data extracted from
Fig. 6). In agreement with the theoretical prediction, we
observe a logarithmic dependence of the force at the AP on
the pulling rate.
By directly fitting the results from atomistic simulation
to Eqs. 4 and 5, we obtain Eb = 5.6 kcal/mol and
xb = 0.17 A . These values indicate that at the angular point,
Fig. 11 Theoretical predictions for the unfolding force at the angular
point (AP) compared with simulation and experiment. Simulation
directly proves existence of the first mechanism (local unfolding,
brown data points), with Eb = 5.6 kcal/mol and xb = 0.17 A (results
obtained directly from fitting to the simulation data). The local
unfolding mode is characterized by a deformation wave that is
linearly proportional to the pulling speed (see Fig. 9). The
deformation wave vanishes at a pulling speed of vcr = 0.161 m/s
(see thicker, dotted line), suggesting that below 0.161 m/s, H-bonds
are strained homogeneously. This gives rise to a different deformation
mode characterized by simultaneous breaking of three H-bonds
(homogeneous rupture), so that Eb = 15 kcal/mol and xb = 1 A .
The continuous red line indicates the theoretical prediction based on
this theory. Experimental results [6, 29] agree well with the
theoretical prediction
8781
an individual H-bond is broken, as it is known that Eb for
H-bonds is between 4 and 5 kcal/mol.
This mechanisms ceases to operate when v [ vcr , due to
disappearance of the unfolding wave. The most significant
consequence of the disappearance of the unfolding wave is
that rather than inducing large local strains, the entire
molecule is stretched homogeneously. Therefore, the H-
bonds that stabilize the helical structure are strained
equally under the applied load.
For rupture to occur, three H-bonds representing a
complete convolution is broken simultaneously, at a ran-
dom location in the coiled-coil. Assuming that a single H-
bonds has a rupture energy of 5 kcal/mol with rupture
distance of 1 A , the total rupture energy is Eb = 15 kcal/
mol and xb = 1 A . Estimates for breaking distance and
energetics of breaking H-bonds are taken from the analysis
reported in [46].
Figure 11 plots the force at the AP as a function of
pulling speed, including the predictions for both deforma-
tion modes (continuous lines). The extended Bell theory
combined with the predicted change in unfolding mode
(rupture of one bond at high pulling rates versus rupture of
three parallel bonds at slow pulling rates) creates a rea-
sonable link between the very small pulling rates applied in
experiments (nm/s) and several magnitudes higher veloci-
ties used in MD (m/s), and thus enables to predict the
unfolding force (force at the angular point) as a function of
the pulling speed.
Interestingly, the pulling speed corresponding to van-
ishing unfolding wave speed lies in close proximity to the
intersection of the two modes of deformation (see circle in
Fig. 11). This provides additional, independent prove for
the change in mechanism, and suggests that the second
mechanismhomogeneous rupture of three H-bonds in a
convolutionis correct.
We note that these concepts also make sense from a
biological point of view: Nature forms three H-bonds in
parallel, instead of forming a single, much stronger bond,
as three H-bonds are energetically easier to create. How-
ever, this concept only makes sense if the three H-bonds
are rupturing at the same time, which indeed appears to be
the case under smaller strain rates comparable to those
present under physiological conditions.
Not much is known about the specific pulling velocities
that appear under physiological conditions. MT are rear-
ranged with pulling velocities in the magnitude of lm/min
[47], close to pulling rates used in the experiments reported
in [6, 29]. However, pulling rates sensed by the IF net-
works inside cells due to external load (e.g. during physical
activity such as running) may lie in the regime between
experiment and simulation.
Figure 12 provides an illustration of the competition
between two mechanisms, each characterized by a pair of
123
8782 J Mater Sci (2007) 42:87718787

Eb and xb, by plotting the free energy landscape under

different applied loads. When pulling is applied, the higher
barrier that is further away from the equilibrium is lowered.
At the point where a change in mechanism is predicted, the
higher barrier becomes lower due to the applied force than
the lower barrier.

Bending of single AHs and coiled-coils

In addition to the tensile tests described in Tensile

deformation of single AHs and coiled-coils we also per-
form bending simulations with the objective to calculate
the bending stiffness and the persistence length of the
coiled-coil dimer as well as of the single protein.
Using the double supported three-point bending test-
known from engineering mechanicsas a reference model,
we fix the Ca-atoms at both ends of the protein, and apply a
force at the middle of our system. As a result, we obtain the
forcedisplacement curves at the pulling point (see Fig. 13).
Taking the differential equation of a continuum Euler-
Bernoulli beam (with u as the out-of-plane displacement,
EI as bending stiffness, and F as the applied force)
EI
u000 x Fx
as a starting point and applying the boundary conditions,
we receive the following expression for the bending
stiffness:
F L3
EI ; 8
d 48
15 f= 0 pN
Energy E in kcal/mol
Fig. 13 Bending of a single AH and the coiled-coil, both feature a
length of 71 A . The pulling rate is 0.2 m/s. The slope of both curves,
here illustrated by a linear fit is proportional to the bending stiffness,
and indicates the coiled-coil to be about 4.8 times stiffer than the
single AH. We used Eq. 8 for calculating the bending stiffness and
derived for the coiled-coil a value of 2.60 1028 Nm2, and for the
single AH 0.54 1028 Nm2. We believe that this difference is a
direct consequence of the changed geometry and thus increased
second momentum of area
with L as the length of the protein, in our case L = 71 A ,
and d as the displacement at the location where the force is
7
applied.
The bending stiffness is proportional to the ratio F/d,
which is the slope from the linear fit of our simulation
results.
For the coiled-coil structure, we calculate for the
bending stiffness EI = 2.60 1028 Nm2, and for the
single AH protein, EI = 0.54 1028 Nm2.
Consequently, the coiled-coil is approximately 4.8 times
stiffer and has therefore a much higher persistence length
than the single protein. This relation is reasonable and can
be attributed to the change in the area moment of inertia. If

f= 500 pN
f= 1000 pN we assume that the single AH has a circular area, the
10
change factor of 4.8 would equal to an increase in radius of
E = E b f xb
about 50%, which is close to the geometrical change in
5 going from single AH to coiled-coil geometry.
With this result, the persistence length can be estimated
0 using the following expression:

-5
-0.2 0.0 0.2 0.4 0.6 0.8
distance x_b from equilibrium in A
Fig. 12 Fitting Bells theory with experimental and simulation
results, we predict the following energy landscape for coiled-coil
structure for different forces f at angular point. The equation
indicates, how the energy is calculated. The equilibrium energy and
distance is set to 0. The first transition state appears 0.17 A away from
the equilibrium, the second transition state appears at 1 A . We can see
that at forces of 700 pN or lesswhich we predict to appear in
experimentsthe second transition state is dominating. At higher
force (dotted lines), the first transition state, observed in simulation
has higher energies compared to the second transition state
EI
1.0
nP ; 9
kB T
where kB is the Boltzmann constant and T the absolute
temperature in Kelvin (K). We estimate a persistence
length of 63 nm for the coiled-coil and of 13 nm for the
single helix.
However, it was reported that the persistence length of
AHs is approximately 1 nm [48] and thus about one
magnitude smaller. We explain the difference between
experiment and simulation by the strain rate dependence,
which was already found for tensile loads and was also

123
J Mater Sci (2007) 42:87718787
reported for bending deformation of tropocollagen mole-
cules [49]. We address the detailed analysis of this
dependence in future work.
Atomistically informed continuum model
Based on the simulation data for varying pulling rates and
the analysis based on Bells theory, we generate a continuum
model, consisting of linear functions for the first and the
second regime and a harmonic function for the third regime.
During the first two regimes, the slope increases linearly
with increasing strain rates, while in the third regime the
curvature is proportional to the square of the pulling rate.
As mentioned above, we find that the force at the angular
points depends logarithmically on the pulling rate, fol-
lowing the predictions by the theoretical model.
Model formulation and parameter fitting
We use a set of empirical equations to describe the rate
dependent forceextension curves based on a continuum
model. All parameters in the model are determined rigor-
ously from atomistic simulation results. The form of the
equation (e.g. the dependence of AP force on pulling speed
according to Eq. 5) is based on the extended Bell theory
described in Theoretical concepts: Modified Bell theory.
The forces in the three regimes are given by
8 Through systematic comparison of a single AH with a
< F1 Fa
x=xa
>
Fx; v F2 s2
x  xa Fa ; 10 mechanical behavior of AH structures due to the first level
>
:
F3 s3
x  xs 2 Fs
where Fa maxc1
lnv c2 ; c12
lnv c13 , xa lower strain at the angular point are consequences of the
c5
v c6 , and s2 maxc7
v c8 ; 0are additional
parameters, and v is the pulling speed (in units of A /fs).
Further, F s c3
x3 c4 ;s3 c9
v c10
v
2
c11 and xs Fa
1  ss21  c4 =c3  s2 .
Note that Fa and xa are the force and the displacement at
the angular point, Fs and xs are the force and the dis-
placement at the beginning of the third regime. The
parameters ci are summarized in Table 4.
Results of this continuum model and comparison with
MD simulation results are shown in Fig. 6 (the continuous
lines are the predictions by the continuum model).
Example application: cyclic loading experiment of
vimentin coiled-coils
The continuum model given in Eq. 10 enables us to de-
scribe the behavior of this structure even if the strain rate is
varying during the stress application or cyclic loading is
applied. Figure 14a shows a prediction of the forcestrain
8783
curve of an experiment of cycling loading of a coiled-coil
vimentin structure.
In modeling cyclic loading, we assume that the relaxa-
tion curve for the coiled-coil structure equals to the one at
pulling rates in the order of nm/s (quasistatic pulling rate, in
accordance with experimental observations that show
refolding processes at time scales of seconds [29]). Similar
results were shown in experiments of cyclic loading for the
coiled-coil structure of myosin, even though different levels
of assemblies were considered [29]. Our results show
qualitative agreement with these experimental studies.
Figure 14b depicts the dissipation energy per loading
cycle as a function of the pulling rate. This plot clearly
indicates, as predicted by [4, 24], that vimentin interme-
diate filaments are a stress buffering protein for higher
deformation rates.
Discussion and conclusion
In this article, we focused on atomistic and continuum
modeling of tensile deformation of single and coiled-coil
AH structures in the rod domain of a vimentin dimer
(geometry see Fig. 1). We have observed the existence of
strain hardening and viscoelasticity, phenomena that have
previously been known to exist at the filament level [4].
Our work confirms that these also exist at the molecular,
dimer level.
coiled-coil structure, we explored differences in the nano-
of hierarchical assembly. We find that the less steep in-
crease in force during the first and third regime, and the
superhelical structure. We find that the third regime of the
coiled-coil features a smoother and less steep strain hard-
ening, since uncoiling of the coiled-coil structure appears
in addition to the stretching of the backbone. Some of the
differences between AH structures and coiled-coils are
summarized in Table 3.
We also have observed that the unfolding of the single
AH structure at large pulling velocities is much less con-
trolled, leading to simultaneous rupture of H-bonds at
several residues at the angular point (AP), in contrast to the
forcestrain curve of the coiled-coil structure, where a
systematic unfolding appears by propagation of an
unfolding wave.
We thus conclude that coiled-coils are mechanically
more stable than single AH structures. This is exemplified
by the smooth change from the first to the second regime at
the AP, the continuous unfolding during the second regime
and the less steep increase in force during the third regime.
This could also explain why such structures appear in cells
123
8784 J Mater Sci (2007) 42:87718787

Table 4 Parameters of the continuum model derived from the curve fitting to match atomistic results within a hierarchical multi-scale scheme
Numerical parameters and its units
c1 c2 c3 c4 C5 c6 c7 c8 c9 c10 c11 c12 c13
220 3.60 237 19.66 57.71 17.40 173.52 6.3 2.4E7 4.55 2.7 41 1.17
pN pN
pN/A pN
A
A 2
pN
fs/A pN/A 3
pN
fs2/A 2
pN
fs/A
pN/A pN pN

describing the nanomechanical properties of vimentin

Fig. 14 Subplot (a): Using our continuum model, we predict the
following behavior for a cyclic load experiment. In this example, we
are pulling with a speed of 1 m/s. For relaxation, we assume the same
behavior as for pulling with a speed of 10 nm/s. The blue area equals
to the dissipated energy. Subplot (b): Our continuum model enables
us to model the dissipation energy for the coiled-coil structure as a
function of the pulling rate. The dissipation energy grows rather
slowly until pulling rates of 0.1 m/s, and increases almost linearly
with increasing pulling velocities afterwards (here exponential, due to
the log-plot). This indicates that vimentin has strain buffering
properties in cells, as it features small energy dissipation rates for
small pulling rates (e.g. during cell motility), but extremely large
dissipation for large pulling rates
(e.g. IFs) and tissues (e.g. muscle myosin), where high and
continuously changing loads occur [1].
We have analyzed the pulling rate dependence of the
coiled-coil, and developed a continuum model that enables
us to predict the forcestrain curves for variations in strain
rates. This model is valid for pulling rates over ten orders of
magnitudes, and can thus be used to calculate the behavior
of vimentin proteins at pulling rates used in experiments (see
Fig. 6) as well as its behavior due to cyclic loading
(Fig. 14). To the best of our knowledge, this is the first
atomistically informed continuum model capable of
coiled-coil proteins over several magnitudes of pulling rates.
The study reported in this article illustrates how MD can
be used to analyze the complex deformation mechanism in
a structural protein material, overcoming some of the time
scale limitations of classical MD (Fig. 11). The strategy of
carrying out simulations at varying rates and interpretation
of results within the framework of Bells theory proved to
be a fruitful combination that elucidates nanomechanical
behavior of such structures.
Based on the atomistic simulation results, we propose a
change in mechanism that occurs at pulling velocities
vcr = 0.161 m/s (Fig. 11). However, even though the
change in mechanism leads to modifications in the unfold-
ing mechanism, we find that the strain at the angular point is
not influenced, since in both simulation and experiment the
strain is found to be approximately 20% [29].
We have observed that the boundary conditions can play
a key role in the unfolding behavior under large pulling
velocities, as the H-bonds ruptures first at the fixed and the
end at which the pulling force is applied. This is observed
even if the fixed and pulled atoms are several convolutions
away from the protein ends. This suggests that the linkers
between the coiled-coils in dimers, being the flexible
boundaries of the coiled-coils, might change the unfolding
behavior and enable the coiled-coils to withstand higher
forces before unfolding appears. This would give the linker
a specific mechanical role in the dimer, and may change the
energy landscape in particular at large pulling velocities.
We may explore these aspects in future studies.
Our studies illustrate how Nature realizes superelasticity
in vimentin coiled-coils and demonstrate that stress buf-
feringalready known from larger scale IF networksis
also implemented at the molecular level of individual
proteins.
Linking results to the filament level
Until now, few experimental results about the mechanical
properties of IFs at the filament level are available. AFM
studies were carried out recently for pulling desmin dimers
from the surface of a filament [6], and for performing
bending tests on 300 nm long filaments [26]. Additionally,
AFM cantilevers were moved perpendicular to the filament
axis in order to analyze the stretching and the rupture

123
J Mater Sci (2007) 42:87718787
behavior of IFs [24]. Glass micro-beam force transducer
apparatus were used to measure tensile properties of single
keratin IF from hagfish threads [19, 20].
In recent experiments [12], filaments were manipulated
with the AFM cantilever perpendicular to the filament axis
with forces between 3040 nN and velocities of 0.4 lm/s.
It was found that the breaking extension at the rupture point
is up to 3.6 times of the initial length and the average
extension over all experiments was 2.6 (160% strain).
Assuming that there are 16 dimers in one filament, the
applied force per dimer in this experiment was of about
1,8002,500 pN.
By fitting these forces and velocities in our continuum
model, we estimate protein strains of 150160%. At this
strain, the coiled-coil structure is in the third regime (the
regime of strain hardening), but still far away from rup-
turing. The strength of covalent peptide bonds was reported
to be in the order of 8 nN (and 16 nN for the coiled-coil
[50]). Therefore, in addition to the protein extension, slip
between proteins may occur, which may lead to additional
strains of up to 50%. We derived this number by taking the
length of a tetramer (approximately 60 nm; two dimer, half
staggered overlap) and by assuming that after slip has taken
place, the dimers are arranged sequentially, leading to a
maximal length of 90 nm (two times 45 nm). Relating the
difference in length (30 nm) to the initial length (60 nm)
we receive a strain due to slippage of 50%. Consequently,
the breaking strain being the sum of stretching and slip, can
be estimated to be on the order of 200%. This is in rather
good agreement with experiment (160% average, and
maximum of up to 260%).
We conclude that yield in vimentin filaments appears
due to the slip between proteins, as it is the weakest part in
the assembled system. This idea further explains the highly
reduced diameter of the ruptured filaments being a result of
the unfolded proteins. This suggests that both mechanisms
(extension and slippage) appear during failure of IFs. Up
until now this was only suggested [20, 24]. Future AFM
studies could be used to determine the forces at which slip
appears, which will help to understand the mechanical
properties of vimentin on the filament level.
Tensile tests were recently performed on wet keratin IFs
in hagfish threads [19, 20]. The curves observed in these
experiments contained four regimes, whereof the first three
regimes appear very similar to the regimes that we observed
in our simulations. Not only the shape of the curves, but also
the strains where the changes between different regimes
take place fit rather closely. For example, the angular point
appeared in experiment at a strain of 34% (our simulation
25%), and the beginning of the third regime was in exper-
iment at 100% of strain (our simulation 120%).
Additionally, if we assume that each thread (diameter of
10 nm) has in average 16 dimers (a reasonable assumption
8785
based on its diameter), we can roughly calculate the average
area per dimer to be 480 A 2, and use this ratio for comparing
Youngs Modulus between our model and experiment. Our
model predicts a modulus of approximately 25 MPa, which
is in good agreement with experiment (10 MPa). Addi-
tionally, the second regime in experiment was demonstrated
to begin with the opening of the AH (also known as the alpha
to beta transition). This was also observed in the simulation
at the beginning of the second regime.
Comparing our results with those from experiments, we
have shown that the specific mechanical properties of vi-
mentin IFs at least partly originate from the protein level.
Among others, the particular molecular nanomechanics
explains how the large extensibility known from filament
rupture experiments is possible and determined the slip-
page between dimers as the weaker link in filaments.
Our results let us suggest that the first regime does not
appear due to the entropic elasticity of the dimer head and
tail, as hypothesized in [20], but due to the stretching of the
H-bonds. As long as H-bonds are not brokenwhich
happens at the angular pointdeformation is completely
elastic. This type of elasticity was observed in experiment
and additionally supports our hypothesis.
Linking results to other coiled-coil structures
The forcestrain curves observed in our simulations are
similar to MD simulations carried out on different AHs and
coiled-coils [3033], as well as to previous experimental
studies on other alpha-helical coiled-coil structures such as
myosin [29] or desmin intermediate filaments [6], even
though the force levels are much larger due to the very high
pulling velocities.
Due to the similarity in the curve shape, the similar
strain levels in experiment and simulation, as well as the
similar force levels in simulations for different coiled-coil
structures, we conclude that the secondary structure, rein-
forced by H-bonds is mainly responsible for the mechani-
cal behavior during tensile deformation, as long as point
mutations do not destroy the structure. It was already
suggested earlier that the mechanisms underlying the
mechanics of proteins are very simple, even if the number
of possible amino-acid configurations is extremely high
[51]. Our observations corroborate this notion.
Interpretation of results in light of biological function
The mechanical properties of metazoan cells, its rigidity at
high stresses and its integrity are mainly caused by vi-
mentin IFs. However, up to now, it was not known exactly,
from which level of hierarchy the specific mechanical
properties arise. In this work we have shown which
mechanical properties appear on the protein level and
123
8786
proved the suggestion that coiled-coils mainly contribute to
the mechanical behavior, discovered on the filament level
as well as on the IF network level.
We hypothesize that Nature utilizes coiled-coils as a
simple construct in order to realize superelasticity in bio-
logical materials. The deformation behavior of coiled-coils
is not only elastic but also contains a long plateau at con-
stant force (Figs. 4 and 6). Thereby, the very high and
reversible deformation of coiled-coils is realized by a
second, stress induced phase, which equals to the rupture of
the H-bonds followed by an immediate unfolding. Once the
tensile load is reduced, the second phase becomes unstable,
or in other word, the structure folds back to the more stable
helical shape by reforming the H-bonds. This behavior is
reminiscent of a shape memory alloy.
There is another twist to this feature: The second phase
may be used simultaneously as the security belt of the scale modeling techniques could be used in future work,
cell, as covalent bonds rupture at forces of three magni-
tudes higher than H-bonds. Additionally, the transition
from the plateau regime to the stretching of the covalent
bonds is very smooth so that no shock waves appear that
could lead to uncontrolled rupture. Therefore, coiled-coils
may be considered as the elementary building blocks of IFs
that enable cells to withstand dramatic loads, large defor-
mation and very high deformation rates.
In coiled-coil structures, H-bonds are apparently not
used to generate mechanically robust structures, as can be
seen by the existence of regime (II) with low unfolding
forces. Instead, they provide a means of enabling low-force
unfolding and simple refolding into helical structures.
Strength originates from the superhelical structure and
from the covalent chemistry present in the protein back-
bone.
The hysteresis (Fig. 14) and the strong strain rate
dependence (Fig. 6) provide further evidence that coiled-
coils represent strain buffering elements in cells, with the
possibility to dissipate great amounts of energy as it
undergoes repeated stretching and relaxation cycles. At the
same time, at small deformations (smaller than approxi-
mately 20%) and small pulling velocities that appear dur-
ing cell movement, the resistance of the coiled-coil is
completely elastic, and thus do not dissipate energy.
The comparison of our results with those from simula-
tions and experiments on other coiled-coils, underline the
important role of H-bonds as a key in the realization of the
specific mechanical properties of coiled-coils. Our results
suggest that Nature creates three H-bonds in parallel (three
for single AHs, and six for coiled-coils, where three are
present in each AH). This is energetically much easier to
realize than single bonds with strength equal to three H-
bonds. Our calculations predict that during pulling veloc-
ities applied in experiments, and most likely during those
appearing in vivo, three parallel H-bonds break simulta-
123
J Mater Sci (2007) 42:87718787
neously. This clearly supports the idea of creating parallel
bonds. The coiled-coil configuration reinforces the protein
additionally and makes it much more stable.
Outlook
Future studies could be focused to develop a more com-
plete understanding of vimentin dimers, the role of the
head and tail domain, the reason for the different lengths of
the conserved coiled-coil segments in the rod domain, and
the function of the stutter. In addition, a thorough theo-
retical understanding of the driving forces during assembly
and the interaction of assembled IFs is still missing.
Furthermore, the time and length scale, which can be
realized with MD is still comparably small. In order to
overcome the limitations of length and time scale, multi-
linking the atomistic scale through mesoscale, such as
coarse graining, to the continuum scale. By using such
techniques, it may be possible to shed light on the different
levels of hierarchies and thus contribute to the under-
standing of the protein mechanics of vimentin alpha-helical
coiled-coils. Techniques such as nudged elastic band
(NEB) may be used to determine the energetic pathway of
rupturing three H-bonds simultaneously.
The combination of new theories, experimental tech-
niques on the nano-scale in addition to modern simulation
approaches, might be the key in understanding hierarchical
biological materials and thus help to heal diseases or help
designing new multifunctional materials such as biological
actuators.
We hope that our theoretical studies could motivate new
experiments on single alpha helical as well as coiled-coil
proteins. In particular, experiments are needed that produce
forceextension curves for systematically varying pulling
rates. Such results can be used directly to elucidate energy
barriers of different mechanisms, and may be used to
compare quantitatively with the theoretical concepts and
results reported here.
Acknowledgements TA acknowledges the support of the German
National Academic Foundation and the Dr.-Juergen-Ulderup Foun-
dation. This research was partly supported by the Army Research
Office (ARO), grant number W911NF-06-1-0291, program officer Dr.
Bruce LaMattina. We acknowledge fruitful discussions with Professor
Harald Herrmann (University of Heidelberg, Germany) and Professor
Laurent Kreplak (University of Basel, Switzerland).
References
1. Alberts B et al (2002) In: Molecular biology of the cell. Taylor &
Francis
2. Strelkov SV, Herrmann H, Aebi U (2003) BioEssays 25:243
3. Burkhard P, Kammerer RA, Steinmetz MO, Bourenkov GP, Aebi
U (2000) Structure 8:223
J Mater Sci (2007) 42:87718787
4. Herrmann H, Aebi U (2004) Annu Rev Biochem 73:749
5. Strelkov SV, Herrmann H, Geisler N, Wedig T, Zimbelmann R,
Aebi U, Burkhard P (2002) EMBO J 6:1255
6. Kiss B, Karsai A, Kellermayer MSZ (2006) J Struct Biol 155:327
7. Janmey PA, Euteneuer U, Traub P, Schliwa M (1991) J Cell Biol
113:155
8. Mucke N, Wedig T, Burer A, Marekov L, Steinert P, Langowski
J, Aebi U, Herrmann H (2004) J Mol Biol 340:97
9. Smith TA, Strelkov SV, Burkhard P, Aebi U, Parry DAD (2002) J
Struct Biol 137:128
10. Wilson KL, Zastrow MS, Lee KK (2001) Cell 104:647
11. Wang N, Stamenovic D (2003) J Muscle Res Cell Motil 23:535
12. Mucke N, Kreplak L, Kirmse R, Wedig T, Herrmann H, Aebi U,
Langowski J (2004) J Mol Biol 355:2342
13. Helfand BT, Chang L, Goldman RD (2004) J Cell Sci 117:133
14. Smith TA, Hempstead PD, Palliser CC, Parry DAD (2003) Pro-
teins 50:207
15. Kreplak L, Aebi U, Herrmann H (2004) Exp Cell Res 301:77
16. Strelkov SV, Schumacher J, Burkhard P, Aebi U, Herrmann H
(2004) J Mol Biol 343:1067
17. Moir RD, Spann TP (2001). Cell Mol Life Sci 58:1748
18. Coulombe PA, Bousquet O, Ma L, Yamada S, Wirtz D (2000)
Cell Biol 10:420
19. Fudge DS, Gosline JM (2004) Proc R Soc Lond 271:291
20. Fudge DS, Gardner KH, Forsyth VT, Riekel C Gosline JM (2003)
Biophys J 85:2015
21. Eckes B, Dogic D, Colucci-Guyon E, Wang N., Maniotis A,
Ingber D (1998) J Cell Sci 111:1897
22. Brown MJ, Hallam JA, Colucci-Guyon E, Shaw S (2001) J
Immunol 166:6640
23. Nieminen M, Henttinen T, Merinen M, Marttila- Ichihara F,
Eriksson JE, Jalkanen S (2006) Nat Cell Biol 8:156
24. Kreplak L, Bar H, Leterrier JF, Herrmann H, Aebi U (2005) J
Mol Biol 354:569
25. Storm C., Pastore JJ, MacKintosh FC, Lubensky TC, Janmey PA
(2005) Nature 435:191
26. Guzman C, Jeney S, Kreplak L, Kasas S, Kulik AJ, Aebi U, Forro
L (2006) J Mol Biol 360:623
8787
27. Omary MB, Coulombe PA, Irwin McLean WH (2004) N Engl J
Med 351:2087
28. Schietke R, Broehl D, Wedig T, Muecke N, Herrmann H, Magin
TM (2006) Eur J Cell Biol 85:1
29. Schwaiger I, Sattler C, Hostetter D, Rief M (2002) Nat Mater
1:232
30. Akkermans RL, Warren CPB (2004) Phil Trans R Soc Lond
362:1783
31. Root DD, Yadavalli VK, Forbes JF, Wang K (2006) Biophys J
90:2852
32. Cieplak M, Hoang TX, Robbins MO (2002) Proteins: Struct
Funct Genet 49:104
33. Rohs R, Etchebest C, Lavery R (1999) Biophys J 76:2760
34. Bornschloegl T, Rief M (2006) PRL 96:118102
35. Mitsui J, Nakajima K, Arakawa H, Hara M, Ikai A (2000) Bio-
chem Biophys Res Commun 272:55
36. Hanke F, Kreuzer HJ (2006) Phys Rev 74:031909
37. Evan E, Ritchie K (1997) Biophys J 72:1541
38. Dudko OK, Hummer G, Szabo A (2006) PRL 96:108101
39. Wiita AP, Ainavarapu SRK, Huang HH, Fernandez JM (2006)
PNAS 103:7222
40. Gilli P, Bertolasi V, Pretto L, Gilli G (2006) J Mol Struct 790:40
41. Bell GI (1978) Science 200:618
42. Lu H, Isralewitz B, Krammer A, Vogel V, Schulten K (1998)
Biophys J 75:662
43. MacKerell AD et al (1998) J Phys Chem 102:3586
44. Courtney TH (1990) In: Mechanical behaviour of materials.
McGraw-Hill
45. Humphrey W, Dalke A, Schulten K (1996) J Mol Graphs 14:33
46. Sheu SY, Yang DY, Selzle HL, Schlag EW (2003) PNAS
100:12683
47. Inoue S, Salmon ED (1995) Mol Biol Cell 6(12):1619
48. Papadopoulos P et al (2006) Biomacromolecules 7:618
49. Buehler MJ, Wand SY (in press) Entropic elasticity controls
nanomechanics of single tropocollagen molecules
50. Buehler MJ (2007) J Mech Mater Struct (in print)
51. Baker D (2000) Nature 405:39
123
J Mater Sci (2007) 42:87888794
DOI 10.1007/s10853-007-1918-x
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Nanomechanics and Raman spectroscopy of fibrillin 2 knock-out
mouse bones
N. B. Kavukcuoglu E. Arteaga-Solis
S. Lee-Arteaga F. Ramirez A. B. Mann
Received: 21 December 2006 / Accepted: 5 June 2007 / Published online: 10 July 2007

Springer Science+Business Media, LLC 2007
Abstract Absence of fibrillin 2 (Fbn2), a non-collage-
nous bone protein, causes a connective tissue disorder
called congenital contractural arachnodactyly (CCA) and
has been associated with decreased bone mineral density.
Nanoindentation and Raman microspectroscopy have been
used to compare the mechanical and chemical properties of
cortical bone from femora of Fbn2/ deficient mice and
their wild-type controls (Fbn2+/+). It was found that
Fbn2/ bones have significantly lower hardness and
elastic modulus compared to Fbn2+/+ bones, especially in
the mid-cortical section. The Raman analysis showed little
difference with genotype except for a decrease in type-B
carbonate substitution in the endosteal region of Fbn2/
bones. The results indicate that Fbn2 plays a direct role in
determining the mechanical properties of bone.
Introduction
Bone is a composite tissue comprised of a mineral phase,
an organic matrix and water. Bone mass in adults
is maintained by coupling between bone resorption by
N. B. Kavukcuoglu
A. B. Mann
Materials Science & Engineering, Rutgers University,
Piscataway, NJ 08854, USA
E. Arteaga-Solis
S. Lee-Arteaga
F. Ramirez
Child Health Institute, UMDNJ-RWJMS, New Brunswick,
NJ 08903, USA
A. B. Mann (&)
Biomedical Engineering, Rutgers University, Piscataway,
NJ 08854, USA
e-mail: abmann@rci.rutgers.edu
123
osteoclasts and bone formation by osteoblasts, however,
when the activity levels of osteoclasts exceeds osteoblasts
uncoupling takes place causing a significant loss in bone
mass and increased fragility. The reduced bone mineral
density (BMD) is the main cause of the most common bone
disease, namely osteopenia/osteoporosis, which is charac-
terized by the microarchitectural deterioration of bone
tissue and an increase in fracture risk [1]. Both osteoblastic
and osteoclastic activity are closely linked to the presence
of non-collagenous bone matrix proteins. Although the
organic phase of bone is 90% type I collagen, there are still
considerable amounts of the non-collagenous proteins
present including fibrillin 1 and 2. Investigators have pro-
posed that the overall mix of non-collagenous proteins
makes a significant contribution to the characteristic
properties of bone, for instance its ability to mineralize
compared to other tissues [2, 3]. It has also been reported
that the non-collageous proteins have a role in bone turn-
over regulation [4] and contribute to the structural integrity
of bone.
Fibrillin 1 and 2 are extracellular glycoproteins that
constitute the major structural components of connective
tissue microfibrils. Their absence in bones can have dramatic
effects on the bones properties. The connective tissue
disorders known as Marfans syndrome and congenital
contractural arachnodactyly are caused by mutations in
fibrillin 1 (Fbn1) and fibrillin 2 (Fbn2), respectively [5, 6].
In addition to other manifestations, these two heritable
disorders of the connective tissues have been associated with
a reduction in bone mineral density [710]. The findings that
fibrillin-rich microfibrils are widely distributed in the
developing mouse and human skeleton has been interpreted
to suggest key roles of these macromolecular aggregates in
bone formation, growth and mineralization [1113].
Furthermore, the fact that fibrillins are highly expressed by
J Mater Sci (2007) 42:87888794
differentiating osteoblasts may also indicate a function in
matrix adherence through interaction with the avb3 integrin
[11, 14, 15]. Finally, fibrillin-rich microfibrils have been
associated recently with modulating the activity of TGF-b,
one of the major growth factors involved in bone physiology
[16]. Consistent with these earlier lines of indirect evidence,
unpublished data indicates that both fibrillin-1 and fibrillin-2
deficient mice are osteopenic due to distinct alterations in the
balance of bone remodeling [17; Arteaga-Soils et al.
(unpublished)].
In this study Raman spectroscopy has been used to
investigate how the absence of Fbn2 affects the chemistry
of mouse bones, while nanoindentation has been used to
examine the bones mechanical properties. Raman Spec-
troscopy is a non-destructive way of analyzing the
molecular structure of the mineral components of bone at
the microscopic level. Raman can characterize both the
organic and inorganic component of bone by examining the
vibrational spectra of the chemical bonds. It provides
quantitative information on the changes in the mineral and
matrix compositon as well as the nature and amounts of
substituents in the mineral. These advantages of Raman
Spectroscopy have led to it gaining in importance as a
method to study bone [1820] and, in particular, to
examine compositional changes due to aging and diseases,
and the link between bone chemistry and mechanical
deformation.
Nanoindentation has become the standard tool for
characterizing the mechanics of materials at small length
scales because of its ability to probe a surface and map its
properties with a resolution smaller than 1 lm. Due to
small dimensions of mouse bones, this technique provides
an effective measurement method for their mechanical
properties [2123]. In the present study the average prop-
erties (mechanical and chemical) and the location depen-
dent (intra-bone) variations in the properties were
investigated and compared between cortical femora bones
from a number of knockout (Fbn2/) and wild-type
(Fbn2+/+) mice.
Materials and methods
Sample preparation
The study used 3-month old female 129/SvEv Fbn2/
mouse models (n = 5) and their age, sex and background
matched wild-type controls (n = 6). Creation of Fbn2/
mice has been already described [24]. The research was
approved by the Rutgers and UMDNJ-RWJMS Animal
Care and Facilities Committees (Protocol # I 06012-2)
and performed according to the NIH guidelines for the care
and use of laboratory animals (NIH Publication #85-23
8789
Rev. 1985). The left femur from the mice of each genotype
was used for this study. The bones were prepared for
nanoindentation testing as described by Roy et al [25].
Briefly, this involved dehydration in graded alcohol solu-
tions (70100%) and mounting in a low temperature cure
epoxy (SPI supplies, West Chester, PA). Dehydration can
affect the mechanical properties of bone, however all the
bones where prepared in the same way so these effects will
be present in bones from both genotypes. There is a small
possibility that one genotype is affected more by dehy-
dration than the other, but this was not investigated. After
mounting, the included femora were sectioned transversely
at mid-shaft using a diamond wafering saw. The surfaces
were ground with silicon carbide paper of decreasing grit
size (400, 600 and then 1,200 particles per inch) followed
by polishing with diamond paste down to 1/2 then 1/4 lm
grit size. After the polishing, all specimens were cleaned
ultrasonically to remove surface debris. Raman studies do
not require any specific kind of sample preparation [19], so
the same bones used for nanoindentation testing were used.
It should be noted that embedding samples into epoxy has
been shown to have no effect on the Raman data [26]. The
locations for the nanoindentation and Raman spectroscopy
measurements are shown in Fig. 1.
Raman microspectroscopy
A Renishaw inViaTM Raman microscope was used for this
study with a 785 nm laser beam and a grating of 1,200 1/mm.
The laser was focused on each point of interest through a
Leica DMLM, 50X/0.75 NA objective providing an
approximately 2 lm spot size. All spectral acquisitions were
performed in the 3502,000 cm1 wavenumber range and
the spots were spaced every 10 lm in the x-(transverse)
direction (see Fig. 1). Depending on the thickness of the
cortex 1520 measurements were taken. Test locations were
marked by microindents prior to testing. Signal-to-noise
ratio and fluorescence background were minimized by
employing 18 s of exposure time and 3 times of accumula-
tion (three separate spectra) at each measurement location.
The Wire 2 software provided by Renishaw was used to
remove the background using cubic spline interpolation at
each acquisition. A single spectrum for each sample point
was obtained by averaging the three accumulations. The
software was then used to analyze the peaks after Gaussian
Lorentzian curve fitting. The intensities of the following
peaks were measured: phosphate m1 at ~958 cm1 (PO3 4
symmetric stretching band), carbonate m1 at ~1,071 cm1
(CO23 symmetric stretching band), CH2 wag at ~1,450 cm
1
(CH bending band), amide ~1,667 cm1 (C=O stretching
band), amide III at ~1243 cm1 (in-phase combination of
the NH bending and CN stretching) and the width (full
width at half maximum, FWHM) of phosphate m1 at around
123
8790
Fig. 1 Optical image (Leica, 5) looking down on a mouse femurs
transverse section. Microindentations were performed to mark the test
areas. The white marks represent the grid pattern of nanoindentations.
Raman tests were performed on the same area as the nanoindentation,
but the Raman was between the lines of nanoindentations to avoid any
possibility of the two testing methods affecting each other. All tests
were started at the periosteal and ended at the endosteal regions of
bone
958 cm1. A typical Raman spectrum for cortical mouse
bone is shown in Fig. 2.
The degree of mineral to organic ratio for each sample
was calculated by taking the ratio of the intensities of the
PO34 m1/amide I peaks [27]. An increase in this ratio typ-
ically indicates a more mineralized matrix [18, 20], though
a change in the proteins could also affect the peak ratio. In
bone, two types of carbonate substitutions have been
reported. Type-A carbonate substitution is when the CO2 3
ion substitutes for the OH ion, and type-B carbonate
substitution is when the CO2
3 ion substitutes for the PO4
3
ion [18, 2830]. In the present study the type-B carbonate
substitution was quantified by the ratio of the peak inten-
sities for CO2 1 3 1
3 (~1,071 cm )/PO4 m1 (~958 cm ). Type-
A substitution, however, could not be quantified because
the wavenumber for the vibrational signal for this substi-
tution (~1,108 cm1) was very weak and could not be
123
J Mater Sci (2007) 42:87888794
observed with the Raman system. In addition, the band-
width (FWHM) of the PO3 1
4 m1 (~958 cm ) peak was
calculated to assess the crystallinity of each sample. It has
been reported by a number of researchers that as the mineral
crystallinity improves the bandwidth of the PO3 4 m1 peak
decreases [29, 31]. Mineral crystallinity in this case refers to
an increase in the crystal size and/or atomic ordering.
Nanoindentation
In this study, the nanoindentation tests were performed
using a TriboindenterTM (Hysitron Inc., MN) in load control
mode. All samples were tested in the ambient environment.
Prior to testing the bones, the contact area, Ac, was deter-
mined by analysis of the load-depth curve and calibrating
the tip on a standard material (fused quartz) as described by
Oliver and Pharr [32]. The tip was nominally a diamond
Berkovich pyramid and the loading direction was parallel to
the long axis of the femur. For each indent a maximum load
of 1,000 lN was applied with a 2 s hold time at peak load.
A grid pattern of 3 100 (~300 indents) was performed
across the cortical transverse sections of each bone (Fig. 1).
The test area was marked by microindents for each
specimen. Nanoindentation tests were performed after the
Raman and the location of the indentation arrays were
chosen to be at approximately the same location as the
Raman spectral measurements to allow a comparison
between the mechanical and chemical data for each sam-
ple. Nanoindents were spaced every 2.5 lm and 10 lm in
the x-(transverse) and y-(tangential) directions, respectively
(the minimum spacing was at least 10 times the maximum
nanoindent depth). The Oliver and Pharr method [32] was
used to determine the mechanical properties. This method
uses the following equations to find the bones elastic
modulus, Es, and hardness, H:
p  1
p S 1  m2s 1  m2i Pmax
Er p ; Er ; H
2 Ac Es Ei Ac
where, Er is the reduced elastic modulus, S is the measured
contact stiffness, Pmax is the maximum load and Ac is the
contact area. mi and ms are Poissons ratio for the indenter and
sample, respectively. The values for the diamond indenter tip
are Ei = 1,140 GPa, mi = 0.07 and the Poissons ratio for the
bone was taken to be ms = 0.3.
Statistical analysis
For each sample the Raman and nanoindentation data
were average so a single value for each property, such as
hardness, was obtained. These values were then averaged
for all of the samples of each genotype to give the mean
J Mater Sci (2007) 42:87888794
Fig. 2 Typical Raman
spectrum of cortical bone from a
mouse femur
value of the property for the genotype. ANOVA was then
used to examine the statistical significance of differences
between the genotypes.
The Raman spectra and the nanoindentation across the
transverse axis of each knock-out and wild-type bone were
examined to give the intra-bone variations for each geno-
type. Mechanical and chemical properties were calculated
at each measurement location for each individual sample.
The nanoindentation data collected across the cortical
thicknesses of each bone were then divided into seven
equal sections. Sections I and II are the bone adjacent to
periosteal (outer) surface. Sections IIIV are the mid-
cortical and sections VIVII are the bone adjacent to the
endosteal (inner) surface (Fig. 1). For the Raman spectral
measurements because of the larger spot size of the micro-
Raman compared to the nanoindent size and the lower
number of measurements (1520) across the cortical
thickness it was divided into three equal sections
(Iperiosteal, IImid-cortical and IIIendosteal). The
mechanical and chemical properties were averaged over
each section for each specimen. The mean values of the
properties for each section were then averaged over all the
bones of each genotype. ANOVA was conducted to assess
the significance of the differences between the knock-out
and wild-type mouse bones in each transverse section. All
tests used 95% minimum level of confidence and p 0.05
was considered statistically significant.
Table 1 Average bone parameters and standard errors derived from nanoindentation and Raman spectra of bones from Fbn2/ mice and
Fbn+/+ mice
Genotype Hardness, H Elastic modulus, E
(GPa) (GPa)
Fbn2/ 1.15 0.01 29.3 0.3 15.02 0.2
Fbn2+/+ 1.23 0.01 31.8 0.3
8791
Results
Nanoindentation
First, the mean values of H and E across the entire section of
cortical bone, plus the standard deviations and the standard
errors for both genotypes (Fbn2/ and Fbn2+/+) were
calculated (see Table 1). ANOVA revealed that bones from
Fbn2/ mice exhibited significantly lower hardness and
elastic modulus values compared to bones from Fbn2+/+
mice (p < 0.05). The average H and E of Fbn2/ mouse
bones were 1.15 0.01 GPa and 29.3 0.3 GPa, respec-
tively. For Fbn2+/+, H was 1.23 0.01 GPa and E was
31.8 0.3 GPa.
The intra-bone variations were plotted as bar charts to
show the variation of H and E with genotype across the
transverse axis of the cortical bone (Fig. 3). Individual
specimens from the same genotype exhibited similar vari-
ations with respect to location. The results show that
Fbn2/ mice bones have lower H and E at locations IIVII
compared to the Fbn2+/+ mice bones. This difference
between the genotypes was statistically significant
(p < 0.05) in the mid-cortical region, sections IIIV. The H
and E values of bones from Fbn2/ mice both decreased
between the periosteal (sections I and II) and mid-cortical
(sections IIIV), then increased slightly towards the end-
osteal (sections VI and VII) part of the bones cross-section.
Mineral: organic Carbonate substitution Crystallinity
PO3
4 m1/amide I CO2 3
3 /PO4 m1 1/PO3
4 m1 (FWHM)
0.110 0.0005 0.0530 0.0002
14.63 0.3 0.115 0.0023 0.0529 0.0002
123
8792 J Mater Sci (2007) 42:87888794

when considering the average over the entire cross-section

of the bone.
Intra-bone variations of the bone parameters were plotted
across the transverse axis of the cortical mouse femora
bones as shown in Fig. 4. Bones from Fbn2/ mice
exhibited a similar mineral: organic ratio to bones from
Fbn2+/+ mice at the periostal and mid-cortical regions
(I and II), but the mineral: organic ratio for Fbn2/ bones
was higher (though not statistically significant) in the near

Fig. 3 Intra-bone variations in (a) hardness, H, and (b) elastic

modulus, E, for femoral cortical bones from Fbn2/ and Fbn2+/+
mice. Sections I and II are the bone adjacent to periosteal surface.
Sections IIIV are the mid-cortical and sections VIVII are the bone
adjacent to the endosteal surface. The data were taken along a grid
traverse between the outer periosteal and inner endosteal regions of
bone as shown by Fig. 1. Fbn2/ and Fbn2+/+ bones exhibit
statistically significant differences in mechanical properties between
sections III and V (p < 0.05). All other differences were of low
statistical significance. The error bars are standard statistical errors

For the bones from Fbn2+/+ mice, the variation in the H

and E values followed an opposite trend where the values
increased from the periosteal (sections I and II) to mid-
cortical (sections IIIV), but decrease near the endosteal
(sections VI and VII). Note that the individual bone samples
of each genotype showed similar variations with location to
those seen in the average mechanical properties.

Raman analysis

The Raman spectra for the cortical bone specimens from

both genotypes were analyzed across the entire bone and
also for the three transverse sections. The mean values
(across the entire cross-section of the cortical bone) and the
standard errors of the phosphate m1/amide I ratio, type-B
carbonate/amide I ratio, and 1/Phosphate(FWHM) are given
for each genotype in Table 1. The results show the mineral:
organic ratio and crystallinity were increased, and type-B
carbonate substitution were decreased in Fbn2/ mouse
bones, but these results had low statistical significance
Fig. 4 Intra-bone variations in (a) mineral:organic ratio, (b) type-B
carbonate substitution and (c) crystallinity for femoral cortical bones
from Fbn2/ and Fbn2+/+ mice. Section I is the bone adjacent to
periosteal surface. Sections II is mid-cortical and section III is the
bone adjacent to the endosteal surface. The data were taken along a
line traverse between the outer periosteal and inner endosteal regions
of the bone. Type-B carbonate substitution of Fbn2/ and Fbn2+/+
bones exhibit a statistically significant difference in section III
(p < 0.05). All other differences were of low statistical significance.
The error bars are standard statistical errors

123
J Mater Sci (2007) 42:87888794
endosteal region (III) (Fig. 4a). Type-B carbonate substi-
tution was significantly lower in Fbn2/ mouse bones at
the near endosteal region compared to Fbn2+/+ mouse
bones (p < 0.05), as seen in Fig. 4b. The crystallinity
exhibited the greatest difference in the mid-cortical where it
was higher for Fbn2/ mouse bones compared to Fbn2+/+
mouse bones (see region II on Fig. 4c), though this was of
low statistical significance.
Overall the Raman data showed only minor differences
between the apatite phase of the bones from the two
genotypes. The only difference that was of statistical sig-
nificance was for carbonate substitution in the endosteal
region.
Discussion
This study has investigated the role of fibrillin 2 (Fbn2), on
the mechanical and chemical properties of cortical bone
from mouse femora. Specifically, the average and location
dependent differences between Fbn2/ and Fbn2+/+
mice bones have been investigated.
Fbn2 deficiency caused a statistically significant decline
in the average mechanical properties as measured by nano-
indentation of mouse bones from sex and age matched
knock-out and wild-type genotypes. Further to this, the
location dependent analyses revealed that the difference is
predominantly in the mid-cortical sections where the bones
from Fbn2/ mice exhibited significantly lower H and E
compared to the Fbn2+/+ mouse bones. The endosteal and
periosteal showed no statistically significant difference in
mechanical properties with genotype. The sections of bone
showing the most significant changes in mechanical prop-
erties with genotype correspond to the sections showing
differences in crystallinity, though the crystallinity differ-
ences were of low statistical significance (possibly due to the
size of the sample set). The effects of genotype on
mechanical properties were mostly confined to the mid-
cortical section which comprises the mature bone, while the
differences in mechanical properties with genotype in the
periosteal and endosteal regions where much smaller. This
suggests that mature lamellar bone has properties which
show a different dependence on Fbn2 than immature woven
bone. Specifically, it appears that the absence of Fbn2
reduces the mechanical properties in the fully mineralized
matrix. This finding is consistent with a previous suggestion
based solely on a gene expression data of a structural role for
microfibrils in the mature bone [11, 13, 33]. It also agrees
with unpublished data indicating an increase in bone fragility
in Fbn2/ mice (Arteaga-Soils et al. (unpublished)).
Another interesting finding of this study was the
decreased type-B carbonate substitution in Fbn/ mouse
bones. This was mostly seen in the endosteal region where
8793
the type-B carbonate substitution of Fbn2/ was signifi-
cantly lower than that of Fbn2+/+ bones.
Overall the observed effects of Fbn2 deficiency on the
mechanical properties of mouse bones indicate that the
protein plays a direct role in determining the mechanics of
the bones. The Raman data shows that the difference with
genotype in the apatite phase of the bone matrix is not of
statistical significance except for a decrease in carbonate
substitution in the absence of Fbn2 in the endosteal region.
All the other differences in the Raman data were of low
statistical significance. Based on this it seems unlikely that
the differences in mechanical properties due to the absence
of Fbn2 were due to changes in the apatite phase.
Conclusions
This study has demonstrated that at the nano-scale Fbn2, a
non-collagenous bone protein, is critical in determining the
mechanical properties and, to a much lesser extent, the
chemical properties of bones. In bones from Fbn2+/+
mice, hardness and elastic modulus were higher, especially
in the mid-cortical region compared to bones from Fbn2/
mice. Typically the mechanical properties of bone would
be expected to increase with increasing crystallinity, but
this is not seen in these results as crystallinity appeared to
be higher in the mid-cortical region of the Fbn2/ bones,
though this was of low statistical significance. This sug-
gests that the fibrillin (microfibrils) in the wildtype mouse
bones may be contributing directly to its mechanical
properties. The decreased mechanical properties did to a
limited extent correspond to a decrease in carbonate sub-
stitution in the apatite phase. However, the most significant
decrease in carbonate substitution was observed only in the
endosteal region where the mechanical properties did not
show a statistically significant difference with genotype.
In summary, the mechanical properties of mouse corti-
cal bone show a clear dependence on the presence of fi-
brillin 2. This is most noticeable in the mature mid-cortical
bone and it does not appear to be due to changes in the
bones mineral (apatite) phase. Thus, it is concluded that
fibrillin 2 and, consequently, microfibrils play a direct role
in determining the mechanical properties of bone.
Acknowledgments The authors would like to thank Prof. David
Denhardt of Rutgers University, NJ and Prof. Nejat Guzelsu of
UMDNJ, NJ for helpful discussions. Financial support for this
research has been provided by NSF, ACS PRF, DOD DURIP,
NJCHE, NIH and the Rutgers University Busch Bequest.
References
1. Ammann P (2003) Therapie 58:403
2. Gundberg CM (2003) Osteoporosis Int 14:S37
123
8794 J Mater Sci (2007) 42:87888794

3. Roach HI (1994) Cell Biol Int 18:617 on the musculoskeletal aspects of marfan syndrome: meeting
4. Kaartinen MT, Pirhonen A, LinnalaKankkunen A, Maenpaa PH report and state of the science, p n/a
(1997) J Biol Chem 272:22736 18. Akkus O, Adar F, Schaffler MB (2004) Bone 34:443
5. Putnam EA, Zhang H, Ramirez F, Milewicz DM (1995) Nat 19. Morris MD, Finney WF (2004) Spectrosc Int J 18:155
Genet 11:456 20. McCreadie BR, Morris MD, Chen T-C, SudhakerRao D, Finney
6. Quondamatteo F, Reinhardt DP, Charbonneau NL, Pophal G, WF, Widjaja E, Goldstein SA (2006) Bone 39:1190
Sakai LY, Herken R (2002) Matrix Biol 21:637 21. Akhter MP, Fan Z, Rho JY (2004) Calcified Tissue Int 75:416
7. Carter N, Duncan E, Wordsworth P (2000) Rheumatology 39:307 22. Chen Q, Rho JY, Fan Z, Laulederkind SJF, Raghow R (2003)
8. Dallas SL, Keene DR, Bruder SP, Saharinen J, Sakai LY, Mundy Calcified Tissue Int 73:387
GR, Bonewald LF (2000) J Bone Mineral Res 15:68 23. Silva MJ, Brodt MD, Fan Z, Rho J-Y (2004) J Biomech 37:1639
9. Giampietro PF, Peterson M, Schneider R, Davis JG, Raggio C, 24. Arteaga-Solis E, Gayraud B, Lee SY, Shum L, Sakai L, Ramirez
Myers E, Burke SW, Boachie-Adjei O, Mueller CM (2003) F (2001) J Cell Biol 154:275
Osteoporosis Int 14:559 25. Roy ME, Rho JY, Tsui TY, Evans ND, Pharr GM (1999)
10. Kohlmeier L, Gasner C, Bachrach LK, Marcus R (1995) J Bone J Biomed Mater Res 44:191
Mineral Res 10:1550 26. Carden A, Morris MD (2000) J Biomed Optics 5:259
11. Zhang H, Hu W, Ramirez F (1995) J Cell Biol 129:1165 27. Boskey A, Mendelsohn R (2005) J Biomed Optics 10(3):031102
12. Plantin P, Durigon M, Boileau C, Le Parc JM (2000) Annales De 28. Freeman JJ, Wopenka B, Silva MJ, Pasteris JD (2001) Calcified
Pathologie 20:115 Tissue Int 68:156
13. Keene DR, Sakai LY, Burgeson RE (1991) J Histochem Cyto- 29. Penel G, Leroy G, Rey C, Bres E (1998) Calcified Tissue Int
chem 39:59 63:475
14. Ramirez F, Pereira L (1999) Int J Biochem Cell Biol 31:255 30. Yerramshetty JS, Lind C, Akkus O (2006) Bone 39:1236
15. Sakamoto H, Broekelmann T, Cheresh DA, Ramirez F, Rosen- 31. Akkus O, Polyakova-Akkus A, Adar F, Schaffler MB (2003)
bloom J, Mecham RP (1996) J Biol Chem 271:4916 J Bone Mineral Res 18:1012
16. Neptune ER, Frischmeyer PA, Arking DE, Myers L, Bunton TE, 32. Oliver WC, Pharr GM (1992) J Mater Res 7:1564
Gayraud B, Ramirez F, Sakai LY, Dietz HC (2003) Nature Genet 33. Kitahama S, Gibson MA, Hatzinikolas G, Hay S, Kuliwaba JL,
33:407 Evdokiou A, Atkins GJ, Findlay DM (2000) Bone 27:61
17. Jones KB, Sponseller PD, Erkula G, Sakai L, Ramirez F, Dietz
HC, Kost-Byerly S, Bridwell KH, Sandell L (2006) Symposium

123
J Mater Sci (2007) 42:87958803
DOI 10.1007/s10853-007-1914-1
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Mechanics of molecular collagen is influenced by hydroxyapatite
in natural bone
Rahul Bhowmik Kalpana S. Katti Dinesh R. Katti
Received: 17 January 2007 / Accepted: 4 June 2007 / Published online: 10 July 2007

Springer Science+Business Media, LLC 2007
Abstract As often seen in biological structural mate-
rials, bone exhibits complex hierarchical structure. The
primary constituents of bone are collagen and hydroxy-
apatite (HAP). HAP mineralizes at specific locations at
collagen, in such a way that the c-axis of HAP aligns
parallel to collagen molecule. The collagen molecule is
helical overall with non-helical ends that are N- or
C-telopeptides. The collagen molecule with telopeptides
interacts with specific surfaces of mineralized HAP.
When subjected to load, the interactions at the interface
between HAP and collagen may significantly affect the
overall mechanics of the collagen molecule. Here, we
have performed molecular dynamics (MD) and steered
MD (SMD) simulations in order to understand the load
carrying behavior of collagen in the proximity of HAP.
Our simulations indicate that the load-deformation re-
sponse of collagen is different when it interacts with
HAP as compared to its response in the absence of HAP.
The interface between HAP and collagen affects the
overall load-deformation response of collagen. Further,
bone also has considerable amount of water and we have
observed that water significantly influences the load-
deformation response of collagen due to collagen-water-
HAP interactions.
R. Bhowmik
K. S. Katti (&)
D. R. Katti
Department of Civil Engineering, North Dakota State University,
Fargo, ND 58105, USA
e-mail: Kalpana.Katti@ndsu.edu
R. Bhowmik
e-mail: Rahul.Bhowmik@ndsu.edu
D. R. Katti
e-mail: Dinesh.Katti@ndsu.edu
Introduction
Bone is an important structural component of the human
body and is a part of the skeletal system. It performs var-
ious mechanical, biological, and chemical functions which
include: supporting the body structure, protecting internal
organs, producing red and white blood cells, and storing
various ions [1, 2]. It has a unique capability of self-
regeneration under appropriate conditions [3]. Bone
exhibits a distinct set of mechanical properties depending
on its location in the skeletal system [4]. This is due to the
fact that the different parts (bone) of the skeletal system
undergo different loading paths during normal daily
activities [5, 6]. These unique sets of mechanical properties
are a result of structural hierarchy in bone [79] which
encompasses molecular to macroscopic level spanning
over several orders of magnitude of length scale as shown
in Fig. 1. However, the role of mechanics of various
structures at different length scales on overall mechanical
response is not clearly understood. This understanding is
vital for designing novel implant materials with mechanical
properties similar to natural bone [1014]. Several attempts
have been made to understand the mechanical properties of
bone and its relation to the hierarchal structural organiza-
tion [1531]. Ji et al. have used finite element modeling
(FEM) methods to calculate the fundamental mechanical
properties of bone [21]. Buehler et al have explained the
reasons for the specific length of collagen molecules which
is observed in the nanostructure of bone and its mechanical
properties by multiscale modeling techniques [22, 23].
Hellmich et al. have examined the mechanical behavior of
bone through micromechanics formulations [24, 25]. It has
been observed that the youngs modulus of cortical bone is
in the range of 1420 GPa [26], whereas that of the osteon
lamellar structure is about 22 GPa [27]. However, the
123
8796
Fig. 1 Schematic showing
structural hierarchy in Bone
mechanical response of collagen when mineral has min-
eralized at specific locations is not known. This is very
difficult to obtain from experimental techniques due to
small size of mineralized hydroxyapatite crystals
(50 nm 25 nm 3 nm) and collagen molecule (300 nm)
[2831]. In this present study we have obtained the
mechanical response of collagen molecule in proximity of
mineral using steered molecular dynamics (SMD) simula-
tions. In our prior work on a biological hybrid system
nacre, [32], we have observed marked influence of prox-
imity of mineral on the mechanics of the protein molecule
at organic-inorganic interfaces.
Also, it has been observed from experiments that water
affects the overall mechanical response of bone. Bone has
about 10 % of water in body environment (Table 1) [28,
3335] and with loss of water, bone exhibits different
mechanical properties [3643]. It has been observed that
strain-at-fracture and energy-to-fracture decreases, whereas
tensile strength, stiffness, and hardness increases with loss
of water [3641]. Nyman et al. have shown that toughness
of bone decreases when water is removed from collagen,
whereas both the strength and toughness decrease when
water is lost from mineral phase [42]. Currey et al. have
shown that water affects the viscoelastic behavior of bone
[43]. Dry bone exhibits lower anelastic deformation as
Table 1 Bone Composition (adapted from reference 19)
Components Weight percentage
Hydroxyapatite 60
Collagen 20
Water 9
Ions and Non-collagenous proteins 11
123
J Mater Sci (2007) 42:87958803
compared to wet bone. These indicate that water has sig-
nificant influence on overall mechanical response of bone.
However, the role of water on the nanostructure of bone,
where the collagen is interacting with hydroxyapatite
(HAP) nanocrystals, is not well understood. Here, we
attempt to understand the mechanical response of wet
collagen (solvated collagen) and compare it with dry col-
lagen (unsolvated collagen) when interacting with HAP
surface. This has been done using SMD simulations.
In order to perform simulations, nano and molecular
structure of collagen-HAP system should be understood
clearly. At the nanoscale, bone consists of HAP mineral-
ized at specific locations on collagen molecules [7].
Between the collagen molecules, HAP mineralizes in
specific zones called as hole zones (Figs. 1 and 2) [44,
45]. The mineralized HAP in these zones is hexagonal in
structure with its c-axis aligned parallel to collagen fibers
[46]. Collagen exhibits helical structure whereas its ends
are non-helical and are known as N- or C-telopeptides
[4749]. Thus, collagen interacts with HAP through these
telopeptides as shown in Fig. 2. We have constructed the
model in a similar way as shown in Fig. 2. The main focus
of the present modeling study with SMD simulations is to
evaluate the role of mineral (HAP) and water on the
load-deformation behavior of collagen molecule in close
proximity of HAP.
Simulation Details
NAMD [50] has been used to perform MD and SMD
simulations using CHARMm (Chemistry at Harvard
Macromolecular mechanics) force field [51] and VMD [52]
has been used for all the interactive studies and visualization.
J Mater Sci (2007) 42:87958803
Fig. 2 The HAP-Collagen
Interface
NAMD has been developed by theoretical and computa-
tional biophysics group at the Beckham Institute for
Advanced Science and Technology at the University of
Illinois at Urbana-Champaign. Minimization has been
performed by conjugate gradient method on all models
before performing MD and SMD simulations. Isothermal
and isobaric ensemble (NPT) is used for MD and SMD
simulations with periodic boundary conditions. Electro-
static interactions between pairs are calculated by Particle
Mesh Ewald (PME) technique. The van der Waals cut off
distance of 9 A has been used in all simulations. Tem-
perature of all models is controlled by Langevin dynamics
and pressure is increased to 1.01 bar in steps and is cal-
culated by Nose-Hoover Langevin piston method [53, 54].
Verlet algorithm is used to integrate the Newton equation
of motion [53, 54]. MD simulations have been performed
on all models followed by SMD simulations.
Model Construction And Simulations
Collagen Molecule with N-Telopeptide
The collagen is a triple helix molecule with three poly-
peptide chains with each polypeptide chain forming a left-
handed helix that is folded in a right-handed superhelix
[55]. The structure of collagen molecule with telopeptide is
obtained from literature [49], where the (GPP)n model of
collagen was obtained from the Protein Data Bank-ID
1 k6f.pdb [56]. Further, N-terminal telopeptide sequence
used in the simulations is as follows [57, 58]:
a1-GlnLeuSerTyrGlyTyrAspGluLysSerThrGlyIleSer-
ValPro-helix (GlyProMet-)
a2-GlnPheAspAlaLysGlyGlyGlyPro-helix (GlyProMet-)
The three stranded N-telopeptide are joined to triple
helix molecule (GPP).
First, collagen molecule with N-terminal telopeptide
(N-Collagen) is geometrically optimized through energy
8797
minimization using CHARMm force field [51]. Further,
the optimized model is solvated with 1800 TIP3P
water molecules followed by energy minimization of the
solvated N-Collagen. Furthermore, the temperature is
increased to 300 K from 0 K in steps of 100 K followed
by increasing pressure from 0 bar to 1.01 bar in steps of
0.25 bar. The resulting structure has been used to per-
form SMD simulations in absence of HAP. The same
geometrically optimized N-Collagen molecule has been
used with HAP for interaction studies. In order to ana-
lyze the role of water on the overall mechanical
response of N-Collagen, unsolvated N-Collagen is also
used for SMD simulations in close proximity and in
absence of HAP. Here, again the same route has been
taken to obtain unsolvated N-Collagen molecule for
SMD simulations.
HAP crystal model
HAP mineral has a hexagonal structure with space group
P63/m [59]. The unit cell parameters are a = 9.424 A ,
b = 9.424 A, c = 6.879 A, a = 90, b = 90, c = 120. The
dimensions of the HAP used in the present study are
a = 75.392 A , b = 75.392 A , c = 20.637 A , a = 90,
b = 90, c = 120 which correspond to 192 unit cells of
HAP i.e., 8 unit cells along a-axis, 8 unit cells along b-axis
and 3 unit cells along c-axis. The dimensions of periodic
boundary conditions are the same as dimensions of the
HAP model used. CHARMm force field parameters for
HAP have been obtained from our previous study of
interfaces between HAP and polyacrylic acid [60, 61]. The
constructed HAP model is first geometrically optimized
through energy minimization then its temperature is
increased from 0 K to 300 K in steps of 100 K. Further,
pressure is increased to 1.01 bar in steps of 0.25 bar.
Finally, the model is again geometrically optimized
through energy minimization. The optimized model has
been used to create appropriate surfaces. It has been found
123
8798
from experiments that the c-axis of HAP aligns parallel to
collagen molecule, i.e. (0001) surface of HAP interacts
with telopeptide of collagen molecule (Fig. 2). The (0001)
surface of HAP has been created by extending the c-axis of
HAP model from 20.637 A to 320.637 A which creates
300 A of vacuum between (0001) surface of HAP and its
periodic image. This creates a pseudo 2D periodic
boundary condition from 3D periodic boundary condition.
The (0001) surface of HAP is rendered non-dipolar by
removing half of the surface ions (calcium atoms) from
(0001) surface to its opposite surface. The construction of
non-dipolar surface from dipolar surface has been per-
formed in accordance with Tasker et al. [62]. The non-
dipolar surface has been constructed because it has been
observed previously that these surfaces are stable as com-
pared to dipolar surfaces [63]. The HAP model with (0001)
surface is geometrically optimized through energy mini-
mization and then the temperature is increased from 0 K to
300 K in steps of 100 K. Further, pressure is increased to
1.01 bar. Finally, the model is geometrically optimized in
order to obtain the structure of the model at the global
energy minimum. This optimized model has been used
with solvated and unsolvated N-Collagen to construct
HAP-Collagen model.
Hydroxyapatite with N-Collagen molecule (HAP-
Collagen)
The optimized model of solvated and unsolvated N-Col-
lagen is brought in close proximity of minimized model of
HAP with (0001) surface separately (HAP-Collagen)
(Fig. 3). The model thus constructed is first geometrically
optimized through energy minimization. Further, the same
procedure is used to reach a temperature of 300 K at a
pressure of 1.01 bar as mentioned in the previous sections.
Finally, such created HAP-Collagen model is used for
SMD simulations.
Fig. 3 Minimized models of
(a) Solvated and (b) Unsolvated
N-Collagen interacting HAP
surface (N-Collagen molecules
are represented in Tube (a1
chains are in black color
whereas a2 chain is in white
color), Water is represented by
ball and stick representations,
and HAP is represented in
VDW)
123
J Mater Sci (2007) 42:87958803
Steered molecular dynamics (SMD) simulations
SMD simulation is a type of MD simulation technique in
which force is applied to selected atom/atoms. It provides
the dynamics of binding and unbinding of analyzed mol-
ecules with applied load [6474]. It also provides the
mechanical response of molecules. SMD simulations can
be conducted at: constant force [64, 75] and constant
velocity [6467, 76]. We have used constant velocity SMD
(v-SMD) in which constant velocity is applied to selected
atom/atoms. Force is calculated in v-SMD as F = k(vtx),
where k is the spring constant of the spring attached to the
pulled atom/atoms, and v is the pulling velocity, t is the
time and x is displacement. In order to analyze the response
of N-Collagen molecule under load, the center of mass of
N-Collagen has been pulled at a velocity of 1 A /ps. The
1 2
force constant of the spring is 4.0 kcal mol A which
corresponds to spatial (thermal)
p
 fluctuation of the center of
mass of 0.77 A kB T=k [77] at 300 K. The ends of the
telopeptides chains are fixed (carbonyl carbon atoms of
pyroglutamic acid residues of all the chains). In order to
analyze the load-deformation behavior of N-Collagen, the
center of mass of the N-Collagen is pulled in close prox-
imity of HAP and in absence of HAP. To understand the
role of water on the load-deformation behavior, another
type of v-SMD simulations have been performed in which
center of mass of solvated and unsolvated N-Collagen is
pulled in close proximity of HAP separately and the layers
of HAP opposite to (0001) surface are fixed (Fig. 4).
Interaction energy calculation
The interaction energy between the components has
been evaluated by energy evaluation tool of NAMD,
MDEnergyTM [50]. Here, energy is calculated between
defined atoms using a specified cut-off distance. This
J Mater Sci (2007) 42:87958803
Fig. 4 Pulling of solvated N-Collagen in close proximity of HAP.
Here Layers opposite to (0001) surface of HAP were fixed (as shown
in white color) (N-Collagen molecules are represented in VDW (a1
chains are in black color whereas a2 chain is in white color), Water is
represented in ball and stick, and HAP is represented in VDW)
technique uses the trajectory file obtained during simula-
tions, the topology file of the structure, and the parameter
file. The interaction energy between the atoms has been
calculated for the entire 150 ps of v-SMD simulations.
Results and discussion
Figure 5a shows the load-deformation response of solvated
N-Collagen in close proximity (grey) and in absence
(black) of HAP (Fig. 6). The load-deformation behavior of
N-Collagen in close proximity of HAP shows five distinct
regions and they are represented as A1A2, A2A3, A3A4,
A4A5, and A5A6. The slopes of these regions represent the
stiffness of N-Collagen when HAP is present. The change
in slope corresponds to unbinding and binding events
occurring at different time intervals. The region A1A2
results from the interaction of N-Collagen with water
molecules and with HAP molecule. The change in slope
from region A1A2 to A2A3 results from unbinding of
N-Collagen from water molecules that are interacting with
N-Collagen and HAP surface. The regions A3A4 and
A4A5 result from breaking of hydrogen bonds between
proline and glycine residues of helical regions of
8799
Fig. 5 Load-Deformation plots of (a) Solvated and (b) Unsolvated
collagen molecule in close proximity and in absence of HAP
N-Collagen, and also from unbinding of water molecules
from N-Collagen. Finally, the region A5A6 is attributed to
the backbone chain of the N-Collagen molecule. Similarly,
the same type of behavior is also obtained for N-Collagen
in absence of HAP as shown in Fig. 5a. However, the
region A2A3 which has been obtained for the N-Collagen
in presence of HAP is not observed for the N-Collagen in
absence of HAP. The reason for this is that at the start of
the region A2A3 (in the plot for N-Collagen in presence of
HAP), N-Collagen unbinds from the water molecules and
the water molecules interact with the charged surface of
HAP, whereas in the absence of HAP, there is absence of
interaction of water molecules with any surface.
In order to understand the role of water on the
mechanics of N-Collagen, unsolvated N-Collagen is also
used and is shown in Fig. 5b, where pulling of unsolvated
N-Collagen in close proximity of HAP is represented by
grey color whereas the pulling of unsolvated N-Collagen in
absence of HAP is represented by black color. The
response of N-Collagen in close proximity of HAP exhibits
five regions: C1C2, C2C3, C3C4, C4C5, and C5C6
whereas the N-Collagen in absence of HAP exhibits six
regions: D1D2, D2D3, D3D4, D4D5, D5D6, and D6D7. A
similar type of trend has been observed for several proteins
when pulled at constant velocity [71, 78]. The regions
C5C6 and D6D7 result from backbone chain of
N-Collagen. On comparing the slopes of regions A1A2
123
8800
Fig. 6 Pulling of Solvated
N-Collagen (a) in close
proximity of HAP and (b) in
absence of HAP (Here
telopeptide ends are fixed)
(44.93 kcal mol1A 2) and B1B2 (53.03 kcal mol1A 2), of
solvated N-Collagen with the slopes of the regions C1C2
(53.24 kcal mol1A 2) and D1D2 (57.92 kcal mol1A 2) of
unsolvated N-Collagen, it has been observed that the slopes
of these regions are similar. This indicates that the presence
of HAP and water does not affect stiffness of the backbone
chain of N-Collagen molecule. However, the peak loads of
solvated and unsolvated N-Collagen molecules in close
proximity of HAP are 10600 pN and 9700 pN respectively,
at displacements of 34.7 A and 15.6 A respectively. These
are observed as peak loads when the contribution of the
backbone of N-Collagen is not significant in the load-
deformation response. Similarly, we have observed that the
peak loads of solvated and unsolvated N-Collagen mole-
cules in absence of water are 10000 pN and 6500 pN with
displacements 27.7 A and 15.6 A respectively. This indi-
cates that water influences the load-deformation response
of N-Collagen. In order to analyze the role of water on the
interaction of N-Collagen with HAP, another type of
v-SMD simulations have been performed as mentioned in
the SMD section. Here, solvated and unsolvated N-Colla-
gen molecules are pulled in close proximity of HAP while
keeping the layers opposite to (0001) surface of HAP fixed
(Fig. 7). The load-deformation response thus obtained is
shown in Fig. 8a, b. The area under the load-deformation
plot represents the energy needed by a molecule to deform.
The area under the plots of Fig. 8a, b represents the energy
needed by solvated and unsolvated molecule to untie from
the HAP surface. On comparing the area of the
load-deformation response of solvated and unsolvated
N-Collagen in close proximity of HAP, it has been
123
J Mater Sci (2007) 42:87958803
observed that the solvated N-Collagen requires more
energy (296.87 kcal/mol) to untie from the surface as
compared to the unsolvated N-Collagen molecule
(210.34 kcal/mol). This comparison has been performed
for displacement of center of mass up to 50 A as beyond
50 A, the area does not change significantly. To understand
the attachment of N-Collagen molecule with HAP, inter-
action energy between various components during v-SMD
simulations has been calculated using MDEnergyTM as
shown in Fig. 9. During v-SMD simulations, interaction
energy between HAP-water is highest followed by inter-
action energy between collagen-water and then between
HAP-collagen. Although significant changes are apparent
in the load-deformation response as shown in Fig. 7, the
corresponding changes are not significant in the Energy-
Time plot (Fig. 9). This indicates that the interaction
between HAP and N-Collagen occurs through water i.e.,
the load is transferred to HAP from N-Collagen through the
water molecules present between HAP and N-Collagen.
Conclusions
Bone exhibits a complex structural hierarchy that spans
over nanometer to millimeter. At the nanostructural level,
bone is primarily composed of HAP and collagen. Here, we
have evaluated the load-deformation behavior of N-Colla-
gen molecules in close proximity and in absence of HAP
using v-SMD simulations. It has been observed that the
load-deformation response of solvated N-collagen in close
proximity of HAP has features which result from breaking
J Mater Sci (2007) 42:87958803
Fig. 7 Pulling of (a) Solvated
and (b) Unsolvated N-Collagen
in close proximity of HAP along
c-axis (along [0001] direction)
during different time (Here
layers opposite to (0001)
surface are fixed)
of hydrogen bonds between N-Collagen and water, where
water is interacting with HAP significantly. These features
are not present in the plot for N-Collagen in absence of
HAP due to lack of interaction between water and HAP.
The load-deformation response for solvated N-Collagen
differs from the load-deformation response of unsolvated
N-Collagen. To understand the role of water on the load-
deformation response of N-Collagen in close proximity of
HAP, solvated and unsolvated N-Collagen molecules are
pulled separately in close proximity of HAP. It has been
observed from the plots of load-deformation response that
the solvated N-Collagen requires more energy to untie
from the HAP surface as compared to unsolvated
8801
N-Collagen. We have shown from the energy-time plots of
various pairs that N-Collagen interacts with HAP through
water.
The present study is focused on the influence of mineral
and water on the load deformation behavior of N-Collagen
in HAP-Collagen model system. This model system is the
initial step towards understanding the molecular and nano
mechanical response of bone. However, in order to predict
the molecular and nano mechanical response of bone
accurately, the present model needs to incorporate several
complex molecular and nano features of bone such as
intergrowth. Also, in order to evaluate the mechanical
response of nano and molecular structure of real bone
123
8802
Fig. 8 Load-deformation plots of (a) solvated and (b) unsolvated
collagen in close proximity of HAP (Here layers opposite to (0001)
surface are fixed)
Fig. 9 Total interaction energy between HAP-collagen, collagen-
water and HAP-water
accurately, the present model would require more param-
eters to include the complex nano and molecular structural
features of real bone. However, the present study is a first
step in the process and has shown that mineral and water
influence the load-deformation behavior of N-Collagen.
123
J Mater Sci (2007) 42:87958803
Acknowledgments This research is partially funded by grant from
National Science Foundation (CAREER grant # 0132768). The
computational resources provided by National Center of Supercon-
ducting applications (NCSA) at University of Illinois at Urbana-
Champaign (UIUC) through Teragrid are acknowledged. The authors
would also like to acknowledge Dr. Gregory H. Wettstein and Francis
Larson of Center for High Performance Computing (CHPC), NDSU.
References
1. Marieb EN (1998) In: Human anatomy & physiology, 4th edn.
Benjamin/Cummings Science Publishing, Menlo Park, California
2. Tortora GJ (1989) In: Principles of human anatomy, 5th edn.
Harper & Row Publishers, New York
3. Boyle WJ, Simonet WS, Lacey DL (2003) Nature 423:337
4. Von Recum AF (1998) In: Handbook of biomaterials evaluation.
CRC Publisher
5. Ramakrishna S, Mayer J, Wintermantel E, Leong WK (2001)
Comp Sci Tech 61:1189
6. Black J (1992) In: Biological performance of materials: funda-
mentals of biocompatibility. Marcel Dekker, New York
7. Weiner S, Wagner HD (1998) Ann Rev Mater Res 28:271
8. Weiner S, Traub W (1992) FASEB 6:879
9. Landis WJ (1995) Bone 16:533
10. Katti KS, Gujjula P (2002) In: Proceedings of 15th ASCE engi-
neering mechanics conference, New York, NY
11. Verma D, Katti K, Katti D (2006) J Biomed Mater Res 77:59
12. Verma D, Katti K, Katti D (2006) J Biomed Mater Res
78A:772
13. Katti KS, Turlapati P, Verma D, Bhowmik R, Gujjula P, Katti DR
(2006) Am J Biochem Biotech 2:73
14. Katti KS (2004) Colloids Interfaces B 39:133
15. Gao H, Ji B, Jager IL, Arzt E, Fratzl P (2003) Proc Natl Acad Sci
USA 100:5597
16. Gao H, Chen S (2005) J Appl Mech 72:732
17. Ji B, Gao H (2004) Mater Sci Eng A 366:96
18. Ji B, Gao H (2004) J Mech Phys Solids 52:1963
19. Ji B, Gao H, Hsia KJ (2004) Phil Mag Lett 84:631
20. Gao H, Ji B (2003) Eng Frac Mech 70:1777
21. Ji B, Gao H (2006) Comp Sci & Tech 66:1212
22. Buehler MJ (2006) PNAS 103:12285
23. Buehler MJ (2006) J Mat Sci 21:1947
24. Hellmich C, Barthelemy J-F, Dormieux L (2004) Eur J Mech A/
Solids 23:783
25. Fritsch A, Hellmich C (2007) J Theor Biol 244:597
26. Reilly DT, Burstein AH, Frankel VH (1974) J Biomech 7:271
27. Rho JY, Tsui TY, Pharr GM (1997) Biomaterials 18:1325
28. Murugan R, Ramakrishna S (2005) Comp Sci & Tech 65:2385
29. Jae-Young Rho L. Kuhn-Spearing, Zioupos P (1998) Med Eng &
Phy 20:92
30. Landis WJ (1995) Bone 16:533
31. Ziv V, Weiner S (1994) Conn Tissue Res 30:165
32. Ghosh P, Katti DR, Katti KS (2007) Biomacromolecules 8:851
33. Legeros RZ (1994) In: Biological and synthetic apatites. CRC
Press, Boca Raton
34. Park JB (1984) In: Biomaterials science and engineering. Plenum
Press, New York
35. Murugan R, Ramakrishna S (2004) In: Encyclopedia of nano-
science and nanotechnology, vol 7. American Scientific Pub-
lishers, California, p 595
36. Dempster WT, Liddicoat RT (1952) Am J Anatomy 91:331
37. Evans FG (1973) Mechanical properties of bone. Thomas,
Springfield
38. Evans FG, Lebow M (1951) J Appl Physiol 3:563
J Mater Sci (2007) 42:87958803
39. Sedlin ED, Hirsch C (1966) Acta Orthopaedica Scandinavica
37:29
40. Smith JW, Walmsley R (1959) J Anatomy 93:503
41. Yamada H, Evans FG (1970) Strength of biological materials.
Williams & Wilkins, Baltimore
42. Nyman JS, Roy A, Shen X, Acuna RL, Tyler JH Wang X (2006) J
Biomech 39:931
43. Currey JD (1965) J Exp Biol 43:279
44. Fitton-Jackson S (1957) Proc R Soc London Ser B 146:270
45. Nylen MU, Scott DB, Mosley VM (1960) In: Sognnaes RF (ed)
Calcification in Biological Systems. Am Assoc Adv Sci p 129
46. Itoha S, Masanori K, Koyamac Y, Takakudac K, Shinomiyaa K,
Tanaka J (2002) Biomaterials 23:3919
47. Kadler KE, Holmes DF, Trotter JA, Chapman JA (1996) Bio-
chem J 316:1
48. Malone JP, Alvares K, Veis A (2005) Biochem 44:15269
49. Malone JP, George A, Veis A (2004) Proteins: Struct Funct &
Biol 54:206
50. Kale L, Skeel R, Bhandarkar M, Brunner R, Gursoy A, Krawetz
N, Phillips J, Shinozaki A, Varadarajan K, Schulten KJ (1999)
Comput Phys 151:283 http://www.ks.uiuc.edu/Research/namd/,
as on 16 January 2007
51. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan
S, Karplus MJ (1983) J Comput Chem 4:187
52. Humphrey W, Dalke A, Schulten KJ (1996) J Mol Graphics
14:33, http://www.ks.uiuc.edu/Research/vmd/, as on 16 January
2007
53. Feller SE, Zhang Y, Pastor RW, Brooks BR (1995) J Chem Phys
103:4613
54. Martyna GJ, Tobias DJ, Klein ML (1994) J Chem Phys 101:4177
55. Beck K, Brodsky B (1998) J Struct Biol 122:17
56. Berisio R, Vitagliano L, Mazzarella L, Zagari A (2002) Protein
Sci 11:262
57. Jones EY, Miller A (1987) Biopolymers 26:463
58. Otter A, Kotovych G, Scott PG (1989) Biochemistry 28:8003
8803
59. Sudarsanan K, Young RA (1969) Acta Crystallographica B
25:1534
60. Bhowmik R, Katti KS, Verma D, Katti DR (2007) Mater Sci Eng
C 27:352
61. Bhowmik R, Katti KS, Katti D (2006) Polymer 48:664
62. Tasker PW (1979) J Phys C Solid State Phys 12:4977
63. Bertaut FCR (1958) Acad Sci Paris 246:3447
64. Izrailev S, Stepaniants S, Balsera M, Oono Y, Schulten K (1997)
Biophys J 72:1568
65. Isralewitz B, Izrailev S, Schulten K (1997) Biophys J 73:2972
66. Wriggers W, Schulten K (1999) Proteins Struct Funct Genet
35:262
67. Kosztin D, Izrailev S, Schulten K (1999) Biophys J 76:188
68. Molnar F, Ben-Nun M, MartiNez TJ, Schulten K (2000) J Mol
Struct (THEOCHEM) 506:169
69. Lu H, Isralewitz B, Krammer A, Vogel V, Schulten K (1998)
Biophys J 75:662
70. Krammer A, Lu H, Isralewitz B, Schulten K, Vogel V (1999)
Proc Natl Acad Sci USA 96:1351
71. Lu H, Schulten K (1999) Proteins Struct Funct Genet 35:453
72. Lu H, Schulten K (1999) Chem Phys 247:141
73. Izrailev S, Stepaniants S, Isralewitz B, Kosztin D, Lu H, Molnar
F, Wriggers W, Schulten K (1998) In: Computational molecular
dynamics: challenges, methods, ideas. volume 4 of Lecture Notes
in Computational Science and Engineering, Springer-Verlag,
Berlin p 39
74. Schulten K, Schulten Z, Szabo A (1981) J Chem Phys 74:4426
75. Lu H, Krammer A, Isralewitz B, Vogel V, Schulten K (2000) In:
Elastic filaments of the cell. Academic/Plenum Publishers, New
York, p 143
76. Grubmuller H, Heymann B, Tavan P (1996) Science 271:997
77. Balsera M, Stepaniants S, Oono Y, Schulten K (1997) Biophys J
73:1281
78. Gao M, Wilmanns M, Schulten K (2002) Biophysical J 83:3435
123
J Mater Sci (2007) 42:88048810
DOI 10.1007/s10853-007-1916-z
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
A solid-state NMR comparison of the mineral structure in bone
from diseased joints in the horse
Sergey Maltsev Melinda J. Duer Rachel C. Murray
Christian Jaeger
Received: 19 December 2006 / Accepted: 4 June 2007 / Published online: 10 July 2007
Springer Science+Business Media, LLC 2007
Abstract In this work, subchondral cortical bone mate-
rial is investigated from the joints of five horses, three of
which presented with no clinical signs or radiographic
signs of osteoarthritis and two of which suffered osteoar-
thritis joint disease, as judged by clinical and radiographic
assessment and histological findings. The horse is a good
model for osteoarthritis in humans, so the aim of this study
is to use nuclear magnetic resonance (NMR) for a detailed
investigation of the bone structure in bone material affected
by osteoarthritis. In particular, we report on the assessment
of the mineral structure of these samples as viewed by
solid-state NMR.
Introduction
Bone as a material is interesting from a number of points of
view. From the biological perspective, it is the material
which gives higher animals their shape and form and
allows locomotion. From the materials science perspective,
it is an organicinorganic composite material which
S. Maltsev
C. Jaeger (&)
Federal Institute for Materials Research and Testing, Division
I.3, Working Group NMR Spectroscopy, Richard Willstaetter
Str. 11, D-12489 Berlin, Germany
e-mail: christian.jaeger@bam.de
M. J. Duer
Department of Chemistry, University of Cambridge, Lensfield
Road, Cambridge CB2 1EW, UK
R. C. Murray
Centre for Equine Studies, Animal Health Trust, Lanwades Park,
Kentford, Newmarket, Suffolk CB8 7UU, UK
123
exhibits remarkable strength and resilience to fracture
under most circumstances. Increasingly, the structure of
bone on molecular length scales has been studied in order
to understand its material properties in more detail and so
to generate a design template for new synthetic materials
mimicking its properties.
Bone is known to fail as a material under some
instances. Trauma generated fractures are common, as is
repetitive overloading injury in athletes. The bone dis-
eases of osteoporosis and osteoarthritis (OA) have huge
economic and welfare impact in the human population. In
the case of osteoporosis, the material strength of the bone
is clearly compromised. However, the effect of osteoar-
thritis on the integrity of bone material is less well
understood. An osteoarthritis-associated increase in sub-
chondral bone volume and trabecular thickness has been
detected in subchondral cortical and cancellous bone from
a number of species [14]. An increase in bone mineral
density has been associated with osteoarthritis in human
joints [5] and it appears that osteoarthritis is less likely in
osteoporotic individuals [57]. However, although in-
creased remodeling of subchondral bone in osteoarthritis
has been indicated in a number of reports [2, 3, 810],
hypomineralization of subchondral bone has also been
noted [2, 11]. It is reported that the stiffness of sub-
chondral cancellous bone increases in osteoarthritis [12,
13], while the stiffness of subchondral cortical bone de-
creases [1416]. It is clear from these various reports that
bone mineral structure plays a part in osteoarthritis,
though less clear how. Accordingly, we have undertaken a
solid-state NMR study in order to characterize the bone
mineral in osteoarthritis, in order to improve our under-
standing of the molecular basis of the disease.
Solid-state NMR is an ideal method for studying com-
posite materials such as bone. An NMR spectrum for a
J Mater Sci (2007) 42:88048810
particular nuclear isotope contains signals from all such
nuclear spins in the sample, distinguished by their chemical
shift, which is determined by the chemical environment of
the spin. Moreover, experiments can be designed such that
through-space magnetic dipoledipole interactions between
nuclear spins also affect the spectrum in some measurable
manner. The dipoledipole coupling between nuclear spins
has a strength which is proportional to 1/r3 where r is the
internuclear distance. Thus, experiments can be performed
in which different chemical sites can be distinguished by
the different frequencies of the signals arising from them
and for each such site distinguished, its spatial correlation
to other nuclear spins determined. The signals from nuclear
spins also relax over time, with a characteristic time con-
stant which is determined by the environment of the spin;
measurement of such relaxation times thus often provides
useful information on molecular-level structure as well. In
this study we will use the dipoledipole coupling between
1
H and 31P spins to perform heteronuclear correlation
(HETCOR) experiments which examine the ordering and
disposition of phosphate, hydroxide and water in the min-
eral component of the bone samples of interest. 1H and 31P
spin-lattice (T1) relaxation time measurements are used to
further characterize these species and their environment.
The horse is a good model for osteoarthritis in humans
[9], so the aim of this study was to use equine subchondral
cortical bone to investigate the hypothesis that mineral
structure and composition is altered in osteoarthritis com-
pared to normal bone. The objective of the study was to use
solid state NMR to compare mineral composition and
structure between subchondral bone from horses with and
without clinical, radiographic and histological evidence of
osteoarthritis.
Experimental
Samples
Osteochondral samples from the dorsal proximal aspect of
the third tarsal bone were obtained from five horses fol-
lowing humane destruction for clinical reasons unrelated to
this study. Three horses had no clinical, radiographic or
histological evidence of abnormality or pain (samples 27,
82, 85, aged 3, 6 and 7 years, respectively). Two horses
had clinical signs of pain localized to the tarsometatarsal
joint with radiographic and histological evidence of
osteoarthritis (samples 64 and 124, aged 10 and 17 years).
Naturally-hydrated samples were frozen in liquid
nitrogen and ground in a Sartorius Dismembrator ball mill.
The samples were then immediately transferred via a fun-
nel into the NMR rotors (approximately 100 mg) and
lightly pressed into place with a plastic rod.
8805
NMR
The measurements were performed on BRUKER Avance
600 spectrometer, operating at 600 MHz for 1H and
243 MHz for 31P using a magic-angle spinning (MAS)
probe which utilized 4 mm zirconia rotors. The spinning
frequency was 12.5 kHz for all experiments. The 90 1H
and 31P pulse length were 3 ls and 5 ls, respectively. The
1
H echo experiments were acquired with an echo delay of
single rotor period (80 ls) and a spectral width of 100 kHz.
The relaxation delay was 2s.
For the 1H31P cross polarization measurements (CP) a
recycle delay of 2 s was used, and the 1H spin-lock field
was ramped down to 50% to broaden the match condition.
Two-dimensional 1H31P heteronuclear correlation
(HETCOR) experiments [17] were performed using fre-
quency-switch Lee-Goldburg (FSLG) [18] cross polariza-
tion (CP) with contact times of 100 ls, 200 ls, 300 ls,
500 ls, 750 ls, 1 ms, 2 ms, 4 ms, 6 ms and 8 ms. A total
of 64 scans were averaged for each t1 time increment in the
two-dimensional experiment.
1
H T1 relaxation time measurements of the proton
species in the mineral crystals cannot be done directly,
because most of the protons are located in the organic
matrix such that these signals must be suppressed. Hence,
the overall 1H T1 mineral time can be measured either
by detecting the 31P signal after CP and varying the 1H
relaxation delay after presaturation before CP or using a 1H
echo sequence prior to the T1 experiment. In the latter case
the protons in the protein matrix dephase and only 1H
nuclei of the mineral phase remain. The relaxation delays
were 50 ms, 100 ms, 200 ms, 500 ms, 1 s, 2 s, 5 s, 10 s,
20 s and 50 s using eight scans for each measurement.
Finally, the 31P T1 relaxation times were determined
using the Torchia method [19] after cross polarization. The
delays were varied from 1 s to 2,800 s.
Histology
Osteochondral samples underwent routine histological
preparation, decalcification and paraffin-embedding.
Four lm thick sections were stained with Toluidine blue
and Haematoxylin and Eosin for evaluation of the articular
cartilage and subchondral bone. Cartilage morphology was
assessed using an Olympus DP12 microscope (Olympus
UK Ltd., London, UK) and a polarized light. Each section
was assessed with respect to chondrocyte orientation and
morphology, staining patterns, articular surface and
osteochondral junction defects; focal cartilage and sub-
chondral bone abnormalities; cartilage erosion; osteophyte
formation; abnormally increased area of dense bone and
focal cancellous bone abnormalities. Presence of osteoar-
thritic change was based on previously described histo-
123
8806
logical scoring systems for cartilage and subchondral bone
[20, 21].
Results and discussion
Histology
Samples 27, 82 and 85 had no evidence of osteoarthritic
change in either the cartilage or subchondral bone at any
site within the joint. Samples 64 and 124 had clear evi-
dence of osteoarthritic change within this joint cartilage
structural disorganization with chondrone formation and
subchondral bone fibrosis. Sample 64 had clear definition
of the subchondral bone margins. Sample 124 had marked
thickening of the subchondral bone with generalized loss of
porosity, typical of osteoarthritic changes (Fig. 1).
NMR
Fig. 2 shows the two-dimensional 1H31P HETCOR
spectra for samples 27 and 64 as typical samples from the
non-OA and OA groups. The spectra were recorded using
FSLG (frequency switched Lee-Goldburg) cross polariza-
tion and 1H31P mixing times of 0.5 ms, 2 ms and 8 ms.
These spectra show the spatial correlations between 1H
spins (from various groups) and 31P spins (primarily
phosphate species contained in the mineral component) in
the sample, and so essentially contain information about
the spatial correlation of 1H containing groups in the
mineral/closely bound to the mineral surface and mineral
phosphate, there being little phosphate elsewhere in the
bone matrix observable by NMR [22]. As most of the
protons (1H) of the protein matrix are not close to 31P, the
protons in the protein matrix are not observed in this
experiment, making it an excellent method by which to
examine the mineral component of a sample exclusively,
without resorting to deproteination of the sample, as dis-
cussed previously by Ackermann et al. [22].
Fig. 1 Stained microscopy of the osteoarthritic sample 124
123
J Mater Sci (2007) 42:88048810
The spectra shown in Fig. 2 for the two different sam-
ples are very similar in both 1H and 31P lineshapes and in
the relative intensities of the various signals to each other
for equivalent mixing times. Moreover, they are similar to
those reported previously by other workers examining the
mineral structure of bone from bovine and rat species [23],
and bovine, rat and human [22]. Very similar spectra are
found for all the other samples in this work and results
from experiments which used HartmannHahn cross
polarization rather than FSLG are similar also (a loss of the
water signal due to spin-diffusion processes can occur in
HartmannHahn experiments). The latter is not surprising
as 1H1H distances in the mineral matrix are expected to be
large so that at the MAS frequency (12.5 kHz), spin dif-
fusion between 1H spins is expected to be slow on the
timescale of the cross polarization transfer and under such
circumstances, the cross polarization dynamics are similar
for FSLG and HartmannHahn cross polarization.
The spectra in Fig. 2 show at least three broad signals in
the 1H dimension: a sharper signal at approximately 0 ppm,
due to OH hydroxyl ions, a broad peak centered around
5.5 ppm due to structural water in the mineral (or at least,
water in close proximity to mineral phosphate, otherwise
this signal would not appear in the spectrum at all) and a
very broad peak in the region 515 ppm from HPO2 4
hydrogen phosphate groups. The OH 1H signal is corre-
lated with a relatively narrow 31P resonance, indicating that
the phosphate groups in close spatial proximity to the hy-
droxyl groups are relatively well ordered, and thus belong
to a relatively crystalline hydroxyapatite structure. The
H2O and HPO2 4
1
H signals, on the other hand, are corre-
lated with relatively broad 31P lines, indicating that these
1
H and their nearby phosphate/hydrogen phosphate groups
are in relatively disordered environments.
Two-dimensional 31P31P correlation spectra for the
samples under non-spinning conditions for mixing times of
1 ms, 10 ms, 100 ms, 1 s and 10 s and room temperature
(not presented in this paper) were undertaken to establish
whether all mineral crystals are similar or whether there are
populations with different species of phosphate groups. At
short mixing times, the exchange pattern in the two-
dimensional spectrum is simply a diagonal lineshape as
there is essentially no spin diffusion between the 31P spins
on this timescale. As the mixing time increases, there are
increasing amounts of exchange intensity correlating dif-
ferent 31P signals, until at mixing times of 1 s and above,
the exchange pattern is essentially circular, indicating that
all 31P signals are correlated with each other. Moreover, the
buildup of exchange intensity with mixing time to this
position is smooth. In other words, all the 31P sites repre-
sented in these spectra (and therefore, also the HETCOR
spectra in Fig. 2) exist in every crystallite in the sample,
i.e. the different 1H sites and phosphate groups inferred
J Mater Sci (2007) 42:88048810 8807

Sample 27 Sample 27 8 msec CP

1msec CP

7 6 5 4 3 2 1 0 -1 -2 -3 7 6 5 4 3 2 1 0 -1 -2 -3
-2 -2 -2 -2
0 0 0 0
2 2 2 2
4 4 4
( H), ppm

( H), ppm
4
6 6 6 6
8 8 8 8
10 10 10 10
1

1
12 12 12 12
14 14 14 14
16 16 16 16
18 18 18 18
20 20 20 20
7 6 5 4 3 2 1 0 -1 -2 -3 7 6 5 4 3 2 1 0 -1 -2 -3
31
( P), ppm 31
( P), ppm

Sample 64 1.5 msec CP

7 6 5 4 3 2 1 0 -1 -2 -3
-2 -2
0 0
2 2
4 4
( H), ppm

6 6
8 8
10 10
12 12
1

14 14
16 16
18 18
20 20
7 6 5 4 3 2 1 0 -1 -2 -3
31
( P), ppm

Fig. 2 2D 1H31P FSLG HETCOR for sample 27 and sample 64 at mixing times of 1 ms and 1.5 ms respectively and sample 27 at mixing times
of 8 ms

from the HETCOR spectra in Fig. 2 exist in every crys-
tallite; there are not different mineral phases in different
crystallites for instance.
We thus ascribe the more disordered region containing
H2O/phosphate and HPO2 4 groups to surface regions of the
crystallites and the more ordered OH/PO3 4 groups to a more
crystalline core in the crystallites as it was shown recently for
nanocrystalline hydroxyapatite in detail [22]. This is con-
sistent with the idea that the surface regions of any crystal are
expected to be more disordered due to the lack of coordi-
nation around surface ions, and/ or coordination of surface
ions by heterogeneous molecules from the protein matrix in
the case of mineral crystals in bone. The H2O/phosphate
region may arise in part or whole from matrix water bound to
the surface of the mineral crystallites.
HETCOR spectra in Fig. 2 were recorded in fact for
between four and ten mixing times for each sample. This
allows us to determine the approximate rate constants for
the (presumed exponential) build up of 31P magnetization
by cross polarization from 1H, which in turn depends on the
1
H31P distances involved, as well as other interactions
affecting both the 1H and 31P spins and any local molecular
dynamics (for instance, local rotations or librations of
water molecules).
The rate constants were determined by fitting the
experimental data to Eq. (1) [24].
St a1  expt=sHP  expt=sd 1
where S(t) is the 31P signal intensity for a mixing time of t,
sHP is the rate constant for the 1H31P cross polarization
process and sd is a decay constant describing the decay of
the 31P signal through various relaxation processes affect-
ing both the 1H and 31P spins.
The sHP rate constants determined for the 31P signal
intensity in the three regions indicated in Fig. 3 are given
in Tables 1 and 2 for HETCOR spectra using Hartman
Hahn and FSLG cross polarization, respectively.
It should be noted that sHP for the OH/PO3 4 region is
difficult to determine accurately due to the near linear
profile of the signal buildup for the 31P intensity in this
spectral region over the range of mixing times used.
The data in Tables 1 and 2 shows that the sHP rate
constants for each 31P spectral region are very similar
within the limits of error for all samples studied. This, plus
the similarity of the two-dimensional HETCOR spectra
themselves leads us to conclude that the mineral compo-
sitions and structures are very similar for all samples and

123
8808 J Mater Sci (2007) 42:88048810

0.75 msec CP PO3

4 in spatial proximity to one type of OH ; hydroxy-
7 6 5 4 3 2 1 0 -1 -2 -3 apatite.
-2 -2
0 0 OH- The 31P lineshape for the H2O/ phosphate region shows
2 2 more complex behaviour: an initially broad 31P lineshape
4 4 decreases in linewidth as the mixing time increases from
6 6 H2O
( H), ppm

2 ms to the longest mixing time studied, 8 ms, for both

8 8
FSLG and HartmanHahn mixing (the latter not shown).
10 10
1

12 12
HPO42- One possible explanation for this effect is that at the longer
14 14
16 16
-
18 18 1msec LGCP OH
-
20 20 4msec LGCP OH
-
7 6 5 4 3 2 1 0 -1 -2 -3 1.0 8msec LGCP OH
31
( P), ppm
0.8
Fig. 3 OH, water and HPO2
4
1
H spectral assignment on the 1H31P
correlation spectra 0.6

0.4
Table 1 The rate constants sHP for the 31P intensity in the three
spectral regions shown in Fig. 2 of 1H31P HETCOR spectra recorded 0.2
using HartmanHahn cross polarization
Sample sHP (OH/PO3 sHP (H2O/PO3 sHP (HPO2 0.0
4 4 4
region)/ms region)/ms region) /ms 15 10 5 0 -5 -10
31
( P), ppm
27 2.9 0.5 0.6 0.1 0.4 0.2
64 10 4 0.6 0.1 0.6 0.1
1msec LGCP H2O
124 2.6 0.8 0.7 0.1 0.6 0.2 4msec LGCP H2O
1.0 8msec LGCP H2O
Error estimates are from curve fitting and do not take direct account
of experimental errors
0.8

0.6
Table 2 The rate constants sHP for the 31P intensity in the three
spectral regions shown in Fig. 2 of 1H31P HETCOR spectra recorded 0.4
using FSLG cross polarization
0.2
Sample sHP (OH/PO3
4 sHP (H2O/PO3
4 sHP (HPO2
4
region) /ms region)/ms region)/ms
0.0
64 1.7 0.3 0.5 0.1 0.4 0.1 15 10 5 0 -5 -10
31
124 21 0.3 0.1 0.2 0.2 ( P), ppm

Error estimates are from curve fitting and do not take direct account of 1msec LGCP HPO4
2-

experimental errors 4msec LGCP HPO4

2-

1.0 8msec LGCP HPO4

2-

thus that joint disease has not changed the mineral structure 0.8
in any way that we can measure here.
It is interesting to examine the 31P lineshapes extracted 0.6
from the two-dimensional HETCOR spectra in Fig. 2.
0.4
These are plotted for sample 27 in Fig. 4 for different
1
H31P FSLG mixing times for the three spectral regions 0.2
indicated in Fig. 3. Other samples give near identical plots,
indicating that the phosphate environment and distribution 0.0
is very similar in all samples. The 31P lineshape for the 15 10 5 0 -5 -10
31
OH/PO3 4 spectral region is invariant with mixing time in
( P), ppm
both the FSLG and HartmanHahn mixing time experi-
Fig. 4 The 31P lineshapes extracted from the two-dimensional FSLG
ments. This is the behaviour expected for a single com- HETCOR spectra for the three spectral regions defined in Fig. 2 for
ponent (crystalline) system in which there is one type of sample 27 for FSLG mixing times of 18 ms

123
J Mater Sci (2007) 42:88048810
mixing times, there is spin diffusion from H2O 1H to OH
groups in the ordered hydroxyapatite regions somewhere in
the experiment prior to cross polarization to 31P, so that
magnetization which initially started on H2O ends up on
the ordered phosphate groups (and 31P spins thereof) in the
hydroxyapatite core of the crystallites. However, FSLG
mixing does not permit spin diffusion between 1H spins
during the mixing period and the longest t1 time used in the
HETCOR experiment is 6 ms which from previous work
[25] is too shorter a time for there to be significant spin
diffusion between 1H spins (significant spin diffusion
between OH and H2O only occurs on a timescale of
200 ms [25]). Thus, the existence of the narrowed 31P line
does not arise from spin diffusion between 1H, but must
instead arise from a second phosphate site spatially corre-
lated with H2O which has different cross polarization
dynamics to the sites which account for the main cross
polarization events at shorter mixing times. The fact that
this second site does not become apparent until longer
mixing times suggests that either (a) its cross polarization
transfer rate constant sHP is significantly longer, or (b) the
decay constant governing the decay of the 31P cross
polarization signal is significantly less than, those for the
sites which account for the majority of polarization transfer
at short mixing times. Alternatively, both (a) and (b) may
occur to some extent. In case (a), the conclusion would be
that the H2O from which the cross polarization is occurring
for this second H2O/phosphate site is relatively distant
from the phosphate to which it cross polarizes and in case
(b), that the rotating-frame spin-lattice relaxation times for
the 31P and/or 1H involved in the cross polarization process
for the second H2O/phosphate site are significantly smaller
than for other H2O/phosphate groups with the faster cross
polarization rates. Moreover, the phosphate groups
involved in this second site are (from the narrowed 31P
lineshape) more ordered than for the rapidly cross polar-
izing H2O/phosphate sites. A likely explanation is that the
mineral bone crystals consists of crystalline hydroxyapatite
as core covered with an amorphous layer as found for
nanocrystalline hydroxyapatite [23]. It must then be con-
cluded that the water molecules are close to this interface
which in turn would explain the higher degree of order of
the phosphate anions due to the near hydroxyapatite
structure.
For completeness in this study, we have recorded the
mineral 1H and 31P T1 relaxation time constants for each
sample also. These are shown in Table 3. It should be
noted that a single T1 time constant describes the relaxation
of the entire 1H spectrum within experimental error. The
estimates of error are estimates of the error from fitting the
experimental data points to an exponential curve. They do
not take into account noise in the experimental data nor the
effects of any baseline errors (for instance) in the
8809
Table 3 The 1H and 31P T1 relaxation time constants for the bone
samples used in this work. The measurement method is described in
the Experimental section
Sample 27 82 85 64 124
1
H T1/s 1.1 0.3 1.1 0.2 1.0 0.2 1.1 0.2 1.2 0.3
31
P T1/s 244 4 207 3 213 3 228 3 206 6
experimental data. Thus, from the values in Table 3, we
believe that the T1 values for both mineral 1H and 31P are
very similar between all samples once all sources of error
are taken into account. Two methods were used for
determining the T1 values: (1) direct T1 measurements,
after first selecting the mineral 1H spectroscopically (which
is done here using a rotor-synchronized echo; the protein
1
H signals have relatively short transverse relaxation times
and so decay during the echo period, leaving only the
mineral 1H signal at the end of the echo experiments), and
(2) indirect measurement using cross polarization from 1H
to 31P (which automatically selects the mineral 1H, as only
these are close enough to 31P spins to cross polarize). In
first case, we obtain relaxation information of every 1H
signal independently. In the second case, we obtain the
information about 1H relaxation through the 31P nuclei. In
general, the T1 relaxation times measured in these two
ways can be different. The difference tells us about the
internal structure of proton lines: if, for example, T1
relaxation measured via cross polarization to 31P is faster,
then there are at least two components forming the 1H line
and this should be taken account of when analyzing the
experimental relaxation data. However, T1 values mea-
sured by methods (1) and (2) were very similar, strongly
suggesting that a one exponent analysis is sufficient.
Furthermore, in our work, using different cross polari-
zation contact times (100 ls and 9 ms) yields very similar
T1 behaviour, despite the fact, that for a 100 ls contact
time, there is almost no signal from hydroxyapatite OH 1H
present in the spectrum and at 9 ms, almost no HPO2 4 or
H2O signal contributes to the spectra. This further suggests
that one exponent fitting of the experimental T1 relaxation
curves may be used and, within error limits, this gave
acceptable fits in every case.
Conclusions
We have investigated the molecular-level structure of the
mineral in bone taken from the subchondral regions of
equine joints where there was clinical and histological
evidence of OA joint disease. Our study used solid-state
NMR as the primary method of structure investigation and
we employed a number of different experiments including
two-dimensional 1H-31P HETCOR spectra and relaxation
123
8810
time measurements. Our results do not find any differences
between the mineral structure and composition of the dis-
eased bone samples as compared with bone which from
clinical findings and histology is not diseased. This sug-
gests that the osteoarthritis may have more influence on the
bone organic matrix or its interaction with the mineral, than
alteration in the mineral structure itself. These results will
no doubt add to the debate about how precisely OA affects
underlying bone structure and strength.
References
1. Dequeker J, Mokassa L, Aerssens J, Boonene S (1997) Microsc
Res Tech 358:37
2. Grynpas MD, Alpert B, Katz I, Lieberman I, Pritzker KPH (1991)
Calcif Tissue Int 20:49
3. Panula HE, Nieminen J, Parkkinen JJ, Arnala I, Kroger H, Alhava
E (1998) Acta Orthop Scand 627:69
4. Radin EL, Martin BM, Burr DB, Caterson B, Boyd RD, Goodwin
J (1984) J Orthop Res 221:2
5. Burr DB (1998) Curr Opin Rheumatol 256:10
6. Burr DB, Schaffler MB (1997) Microsc Res Tech 37:343
7. Dequeker J (1985) Clin Rheum Dis 271:11
8. Matsui H, Shimizu M, Tsuhi H (1997) Microsc Res Tech 333:37
123
J Mater Sci (2007) 42:88048810
9. Murray RC, Vedi S, Birch HL, Lakhani KH, Goodship AE (2001)
J Orthop Res 1035:19
10. Shimizu M, Tsuji H, Matsui H, Katoh Y, Sano A (1993) Clin
Orthop 229:293
11. Mansell JP, Bailey AJ (1998) J Clin Invest 1596:101
12. Finlay JB, Bourne RB, Kraemer WJ, Moroz TK, Rorabeck CH
(1989) Clin Orthop 193:247
13. Li B, Aspden RM (1997) J Bone Miner Res 641:12
14. Ding M, Dalstra M, Line F, Hvid I (1998) Acta Orthop Scand
358:69
15. Ewers BJ, Newberry WN, Garcia JJ, Haut RC (1998) Trans
Orthop Res Soc 459:23
16. Li B, Aspden RM (1997) Ann Rheum Dis 247:56
17. Maudsley AA, Ernst RR (1977) Chem Phys Lett 50:368
18. Lee M, Goldburg WI (1965) Phys Rev A1261:140
19. Torchia DA (1984) J Magn Reson 613:30
20. Kim DY, Taylor HW, Moore RM, Paulsen DB, Cho DY (2003)
Vet J 166:52
21. Yagi R, McBurney D, Laverty D, Weiner S, Horton WE Jr (2005)
J Orthop Res 23:1128
22. Cho G, Wu Y, Ackermann JL (2003) Science 613:300
23. Wilson EE, Awomusi A, Morris MD, Kohn DH, Tecklenburg
MMJ, Beck LW (2006) Biophys J 3722:90
24. Wilson MA (1987) NMR techniques and applications in geo-
chemistry and soil chemistry. Pergamon Press, London
25. Jager C, Welzel T, Meyer-Zaika W, Epple M (2006) Magn Reson
Chem Chem 573:44
J Mater Sci (2007) 42:88118823
DOI 10.1007/s10853-007-1917-y
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
A continuum model for remodeling in living structures
Ellen Kuhl Gerhard A. Holzapfel
Received: 29 December 2006 / Accepted: 4 June 2007 / Published online: 14 July 2007

Springer Science+Business Media, LLC 2007
Abstract A new remodeling theory accounting for
mechanically driven collagen fiber reorientation in car-
diovascular tissues is proposed. The constitutive equations
for the living tissues are motivated by phenomenologically
based microstructural considerations on the collagen fiber
level. Homogenization from this molecular microscale to
the macroscale of the cardiovascular tissue is performed
via the concept of chain network models. In contrast to
purely invariant-based macroscopic approaches, the pres-
ent approach is thus governed by a limited set of physically
motivated material parameters. Its particular feature is the
underlying orthotropic unit cell which inherently incorpo-
rates transverse isotropy and standard isotropy as special
cases. To account for mechanically induced remodeling,
the unit cell dimensions are postulated to change gradually
in response to mechanical loading. From an algorithmic
point of view, rather than updating vector-valued micro-
structural directions, as in previously suggested models, we
update the scalar-valued dimensions of this orthotropic unit
cell with respect to the positive eigenvalues of a tensorial
driving force. This update is straightforward, experiences
no singularities and leads to a stable and robust remodeling
algorithm. Embedded in a finite element framework, the
E. Kuhl (&)
Department of Mechanical Engineering, Stanford University,
Stanford, CA 94305, USA
e-mail: ekuhl@stanford.edu
G. A. Holzapfel
Department of Solid Mechanics, School of Engineering
Sciences, Royal Institute of Technology, Stockholm 100 44,
Sweden
G. A. Holzapfel
Institute for Biomechanics, Center for Biomedical Engineering,
Graz University of Technology, Graz 8010, Austria
algorithm is applied to simulate the uniaxial loading of a
cylindrical tendon and the complex multiaxial loading
situation in a model artery. After investigating different
material and spatial stress and strain measures as potential
driving forces, we conclude that the Cauchy stress, i.e., the
true stress acting on the deformed configuration, seems to
be a reasonable candidate to drive the remodeling process.
Introduction and motivation
Living tissues are able to adapt to physiological and path-
ophysiological stimuli in order to keep adequate perfusion
according to the metabolic demand of the tissue. For
example, changes in mechanical stimuli lead to altered
cellular and extracellular activities, and typical observed
biological responses are related to growth, remodeling,
adaptation, and repair, i.e., mechanobiology (see, e.g.,
Humphrey [33], Huang et al. [31], Ingber [34], Klein-Nu-
lend et al. [35], Wang and Thampatty [55], Holzapfel and
Ogden [28, 29], Mofrad and Kamm [47] or Lehoux et al.
[40]). Changes in the material (and structural) properties of,
for example, the artery wall through alterations in its
internal microstructure constitute an active process that
occurs in response to changes of mechanical parameters, a
process called arterial remodeling. It is the endothelium
cell that sense mechanical and humoral parameters, trans-
duce signals to the underlying smooth muscle cells and to
the surrounding tissue, and relay mechanical and bio-
chemical changes into biomolecular events. Therefore, the
endothelium at the interface of the blood plays a crucial role
in the initiation of arterial remodeling. In particular, in
cardiovascular tissues, remodeling processes are important
123
8812
in the context of arterial development, atherosclerosis, and
healing in response to arterial injury. A great deal of inter-
disciplinary research effort is devoted to the (mechanical)
signaling pathways because they may enable the identification
of therapeutic targets and the development of new pharma-
cological strategies. Moreover, understanding the interplay
between the architecture of the internal microstructure and the
mechanical loading is of fundamental importance to engineer,
e.g., blood vessel substitutes, see Nerem and Seliktar [48].
The function and integrity of organs are maintained by
the tension in collagen fibers, which contribute signifi-
cantly to the stability and strength of organs. Collagen
fibers are typically considered as the main load bearing
constituent of the extracellular matrix. Accordingly,
changes in the material (and structural) properties can
primarily be attributed to variations in collagen content,
type and thickness and, of course, in the orientation within
the tissue. Leung et al. [41, 42] were amongst the first to
verify experimentally that mechanical forces relate to
pressure and flow direct medial cell biosynthesis and
modulate structural adaptations to hemodynamic changes.
Based on in vitro studies of smooth muscle cells, they
reported that aortic medial cells attached to elastic mem-
branes and subjected to cyclic stretching consistently syn-
thesized collagen of types I and III much more rapidly than
did cells growing on stationary membranes. In the present
manuscript we use the word remodeling exclusively with
respect to collagen fiber reorientation, while the type and
thickness of collagen as well as its content and its con-
centration are assumed to be constant. In addition, we do
not address adaptation in the form of volumetric growth
which is addressed in detail elsewhere in the literature, see,
e.g., Rodriguez et al. [49], Lubarda and Hoger [43] or Kuhl
et al. [38]. There is, however, strong evidence that growth
and remodeling can indeed be viewed as separate indi-
vidual processes. Stopak and Harris [50] studied the ori-
entation of collagen fibrils due to the forces exerted on
them by fibroblast in gels. Fiber reorientation was found to
take place in response to changes in the mechanical loading
although no significant growth, resorption, and production
of new fibers was reported. Motivated by these findings,
Garikipati et al. [20] provided a theoretical framework that
focuses exclusively on collagen fiber remodeling and
supported their theory by a set of remodeling experiments.
For a more sophisticated theoretical approach that captures
the interaction of the individual phenomena of growth and
remodeling, the reader is referred to Menzel [45].
To gain further insight into the complex biomechanical
phenomena related to tissue remodeling we aim at formu-
lating and implementing a novel constitutive framework
for collagen fiber remodeling with particular emphasis on
the arterial wall, in which type I collagen is the major
constituent. As such, the central focus of this study is to
123
J Mater Sci (2007) 42:88118823
capture, predict and explain basic trends observed in col-
lagen fiber remodeling and its impact on the structural
response at the tissue or organ level. There seems to be a
general agreement that the interplay between matrix stress,
fibroblast alignment and stress in the actin network is
responsible for collagen fibril reorientation as reported by
Stopak and Harris [50] and described in detail by Gariki-
pati et al. [20]. However, we do not aim at explaining the
origin of remodeling which is governed by many highly
complex interactive phenomena on the cellular level that
involve altered gene expression in response to altered
loading (i.e., gene transcription, translation, protein syn-
thesis, packing, and activation) which eventually results in
altered rates of turnover of cells and matrix. Nor do we aim
at following the classical continuum mechanics approach
and develop a purely invariant based macroscopic theory
governed by a number of abstract material parameters. Our
goal is to apply suitable homogenization techniques to
derive a sound phenomenologically and micromechani-
cally based formulation with a limited number of param-
eters that have a clear physical interpretation.
To this end, we begin our investigations on the
microstructural or rather molecular level focusing on the
mechanical description of the individual collagen fibers.
The characteristic feature of typical collagen molecules
is their long, stiff, triple-stranded helical structure in
which three collagen polypeptide chains are wound
around one another in rope-like superhelical structures
which are stabilized by numerous hydrogen bonds. The
mechanical properties of these helical structures are un-
like those of any other natural or synthetic polymers.
Collagen helices display a remarkable stiffness which
may be characterized appropriately through the so-called
wormlike chain model. The wormlike chain, or rather
Kratky and Porod model [36], imagines the polymer as a
rod that bends smoothly under thermal fluctuations.
Traditionally applied to model the DNA double helix,
see Bustamante et al. [7, 8] and Marko and Siggia [44],
the Kradky and Porod model was recently adopted to
simulate the behavior of the collagenous triple helix by
Bischoff et al. [3, 4], Garikipati et al. [19, 20] and Kuhl
et al. [37, 39].
After the collagen fibrils have formed in the extra-
cellular space, they are greatly strengthened by the for-
mation of covalent crosslinks between lysine residues of
the constituent collagen molecules. If cross-linking is
inhibited, the tensile strength of the fibrils is drastically
reduced, the collagenous tissue becomes fragile and the
structure tends to tear, see Alberts et al. [1]. To incor-
porate these characteristic cross-linking network effects,
different isotropic chain network models have been pro-
posed in the past, see, e.g., Flory [17], Treloar [53] and
Arruda and Boyce [2, 5, 6]. In order to account for the
J Mater Sci (2007) 42:88118823
anisotropic nature of cardiovascular tissues on the mes-
oscopic extracellular matrix level, we generalize the
cubic isotropic unit cell of the Arruda and Boyce model
to obtain the orthotropic eight-chain model suggested
recently by Bischoff et al. [3, 4].
Finally, it remains to incorporate the living nature of the
tissue and its ability to adapt its collagenous microstructure
to the mechanical loading environment. Naturally, fiber
directions will evolve in vivo to optimize the load bearing
capacity while keeping the required compliance. Tradi-
tionally, remodeling theories in arteries can be classified
into stress driven and strain driven approaches. The former
are typically based on the assumption that the cardiovas-
cular tissue remodels its geometry to restore circumferen-
tial wall stress due to pressurization and wall shear stress
due to blood flow to normal levels, see, e.g., Taber and
Humphrey [52], Gleason and Humphrey [22] or Hariton
et al. [25]. Alternatively, motivated by successful predic-
tions in hard tissue mechanics, the authors of the latter type
of models suggest that strain rather than stress is the rele-
vant driving force for the remodeling process, see Kuhl
et al. [37], Himpel et al. [26] or Driessen et al. [13]. Either
of the two theories is able to identify characteristic
microstructural directions which are allowed to reorient
with respect to the eigendirections of a mechanically rel-
evant second-order tensor. In terms of algorithmic proce-
dures, this vector reorientation typically leads to complex
rotational updates which usually involve singularities due
to the trigonometric nature of the underlying update
equation, see, e.g., Menzel [45, 46]. Although very elegant
from a mathematical point of view and maybe well-suited
for microstructures with one predominant orientation, these
reorientation models seem rather cumbersome in the con-
text of arterial walls where multiple fiber families need to
be accounted for.
When aiming to develop reliable constitutive theories
for remodeling in cardiovascular tissues it is crucial to have
detailed insight in the structural arrangement of the colla-
gen fiber distribution. By using the birefringent properties
of collagen, Finlay et al. [16] elaborated tangential sections
of cerebral arterial walls to examine the integrated struc-
tural order of the individual layers. Alternative techniques
providing information about the collagen fiber distribution
in arterial walls were discussed recently by Elbischger
et al. [14, 15]. Along these lines, continuously distributed
collagen fiber orientations were incorporated in the more
recent models by Driessen et al. [11, 12], Freed et al. [18]
and Gasser et al. [21].
Experimental findings suggest that at biological equi-
librium two predominant fiber orientations can be identi-
fied in each layer of the arterial wall. Typically, these two
discrete families of collagen fibers are found to be located
somewhere in between the directions of the two maximal
8813
principal stresses (or strains). Hence, as the stress (or
strain) state varies with the radial position, the orientations
of the two collagen fiber families also vary across the
thickness of the arterial wall, as reported by, e.g., Taber
and Humphrey [52] and Holzapfel et al. [30]. This varia-
tion across the wall thickness was successfully obtained
from the discrete collagen fiber reorientation models by
Driessen et al. [13] and Hariton et al. [25]. In [13] it was
assumed that the collagen fibers align along preferred
directions, situated in between the principal stretch direc-
tions, while in [25] the remodeling process was assumed to
be stress driven. Within the present manuscript, we com-
bine these basic assumptions with the fundamental concept
of chain network models to obtain a three-dimensional
remodeling theory which is general enough to predict
remodeling in complex multiaxial loading situations. In
contrast to existing theories, which strongly rely on com-
plex rotational updates, this new approach can be algo-
rithmically realized in terms of remarkably simple and
straightforward updates of scalar-valued spatial dimen-
sions.
The manuscript is organized as follows: the governing
equations of the micromechanically motivated remodeling
theory are derived in Sect. Governing equations. Starting
from the molecular level, we derive the constitutive
equations for anisotropic soft biological tissues based on
the concept of orthotropic chain network models. Section
Computational examples then focuses on two particular
model problems, a cylindrical tendon subject to uniaxial
tension, and a tube-like artery subject to uniaxial stretch in
combination with a distending pressure. Section Discus-
sion closes with some final remarks.
Governing equations
In what follows, we summarize our set of constitutive
equations for anisotropic cardiovascular tissues. To this
end, we apply the following hypotheses:
Hypothesis I: Large arteries seek to restore wall stress
to within a range of homeostatic values.
Hypothesis II: Collagen fibers as the main load bearing
constituent of the extracellular matrix adapt their
orientation and align with respect to the principal stress
directions in order to minimize wall stress.
Hypothesis III: Collagen fiber remodeling can be
modeled and simulated phenomenologically to improve
the understanding of fiber orientation and provide
further insight in the structural arrangement of the
individual arterial layers.
It should be mentioned, however, that although the pres-
ent model takes into account microstructural information,
123
8814 J Mater Sci (2007) 42:88118823

it is still based on a rather phenomenological approach in the
sense that it does not explain the mechanisms how the indi-
vidual cells actually sense changes in loading and commu-
nicate this information. Many different receptors on the
surface of endothelial cells and vascular smooth muscle cells
are able to detect subtle changes in the mechanical envi-
ronment. They initiate various different mechanotransduc-
tion cascades according to the nature of the mechanical
stimulus perceived. The cytoskeleton and other structural
components play an important role in mechanotransduction
as they are able to transmit and modulate tension between
focal adhesion sites, integrins and the extracellular matrix.
Moreover, changes in the mechanical environment may also
initiate changes in the ionic composition of the cells, medi-
ated by ion channels, stimulate various membrane receptors
and induce complex biochemical responses, see, e.g., Huang
et al. [31], Mofrad and Kamm [47] or Lehoux et al. [40] for
excellent overviews. Since we do not aim at simulating these
molecular mechanisms of mechanotransduction, all these
phenomena are modeled phenomenologically through a set
of continuum based remodeling equations which we describe
in the sequel.
We begin on the microstructural level with the con-
stitutive description of the individual collagen fibers. On
the mesolevel, we then elaborate a representative volume
element representing the extracellular matrix. On the
macroscopic level, we finally characterize the overall
tissue behavior through a micromechanically motivated
constitutive model which is able to account for micro-
structural adaptation in response to changes in the
mechanical loading.
On the collagen fiber level
constant and h is the absolute temperature. In the case of
living tissues, we suggest h = 310 K, i.e., h = 37 C.
The two parameters governing the chain behavior are
the contour length L and the persistence length A, as
illustrated in Fig. 1.
The force required to pull the ends of the chain away
from each other by a distance r thus follows straightfor-
wardly by taking the derivative of the free energy wchn with
respect to the end-to-end length.
" #
chn dwchn 1 r 1
f kh 4
1 2
dr 4A L 1
r=L2
Note that due to the particular nature of the free energy
chn
w , the end-to-end length r of a wormlike chain cannot
extend beyond its contour length L as 0 < r < L.
Remark 1 [Parameters on the collagen fiber level] The
wormlike chain model is essentially a two-parameter
model governed by the contour length L and the persistence
length A. Figure 1 illustrates the physical meaning of the
persistence length. It shows the force-displacement curves
of a single collagen fiber indicating the increase in initial
stiffness with increasing persistence length for, say,
A = 0.1, A = 0.4, and A = 0.8. Note that throughout the
entire manuscript, all lengths of the model have been
rendered non-dimensional by dividing them by the link
length of the chain, as proposed by Garikipati et al. [19].
Remark 2 [Specific data for the persistence length]
F-actin (a filamentous protein responsible for the con-
traction and relaxation of muscle) has a persistence
length A of approximately 16 lm. For nanotubes A is in

On the microscopic level, we assume that the microstruc-

ture of a collagen triple helix is represented through the
10
wormlike chain model. Wormlike chain models were
introduced within the context of DNA mechanics by Marko 9

and Siggia [44], and Bustamante et al. [7, 8], and recently 8
applied to collagen fibers by Bischoff et al. [3, 4], Gari- 7
f(r/L) times 1/(k)

kipati et al. [19, 20] and Kuhl et al. [37, 39]. In the sta-
6
tistical mechanics of long chain molecules such as collagen
fibrils, the key kinematic variable that characterizes the 5
conformation of the chain is the end-to-end length r. 4
According to the wormlike chain model, the free energy
3
wchn of a single collagen fiber can be expressed in terms of
the end-to-end length in the following form. 2

2  1
chn L r 1 r
w wchn
0 kh 2 2
1 0
0 0.2 0.4 0.6 0.8 1
4A L 1
r=L L
r/L
Herein, wchn
0 is the value of the chain energy in the Fig. 1 Collagen fiber level s Single chain force vs. chain stretch for
unperturbed state, k = 1.381 1023 J/K is the Boltzmann varying persistence lengths A

123
J Mater Sci (2007) 42:88118823
the millimeter range. Note, however, that A for DNA
in vivo has a value of ~50 nm (Hagerman [24]), and A
for synthetic polymers is typically only a few nanome-
ters. Hence, the persistence lengths A L of a typical
DNA molecule and a synthetic molecule are considerably
smaller than their contour lengths L. Recent studies
performed by means of optical tweezers seem to indicate
that under physiological conditions collagen I molecules
have a persistence length of ~14.5 nm which would be
less than 5% of their contour length of ~309 nm, see Sun
et al. [51]. Accordingly, collagen would be much more
flexible than previously assumed, yet even more flexible
than DNA. Although we suggest to stick to the wormlike
chain approach in the sequel, this is not a general limi-
tation of the overall constitutive model as such. Due to
the modular structure of the overall framework, the free
energy function for the individual fibers (1) can easily be
modified, adapted and integrated straightforwardly in the
macroscopic model.
On the extracellular matrix level
From a mechanical point of view, the extracellular matrix
is modeled as a surrounding substrate in which the indi-
vidual collagen fibers are embedded. A representative
volume element of the extracellular matrix thus consists of
a substrate of elastin, proteoglycans and cell, characterized
through the isotropic free energy wiso, and an anisotropic
contribution due to the individual chains wchn. Moreover,
we introduce a repulsive chain contribution wrep to char-
acterize the tissues behavior of the initial configuration
such that the total free energy may be written in the fol-
lowing form.
W Wiso Wchn Wrep 3
The individual terms of the free energy take on the
explicit representations.
1 1
Wiso k ln2 J lI1C
ndim
l lnJ
2

2
2
 F, u
chn nchn L r 1 r
W kh 2 2
4
4A L 1
r=L L
" #
n chn
1 1 1
rep
rep
W kh
W
4A L 4r0 1
r0 =L2 4r0
For the isotropic term Wiso, we apply a standard
neo-Hookean model expressed in terms of the first
invariant I1C C : I of the right Cauchy-Green tensor
C Ft  F;where F rX u denotes the deformation
gradient and J detF > 0 is its determinant. Moreover, k
and l are the standard Lame constants.
8815
The overall chain energy Wchn follows by summing up
the contributions wchn of eight individual chains weightened
by the overall chain number density nchn, i.e., the number of
chains per unit volume. According to the original eight-
chain model by Arruda and Boyce, each of these chains
connect the corners of a regular cuboid of dimensions 2 l1,
2 l2, and 2 l3 with its center, compare Fig. 2, left. The end-
to-end length r0 in the undeformedpconfiguration, thus,
p
follows straightforwardly as r0 l2I l21 l22 l23 :
The unit cell is postulated to deform in the principal stretch
space. Accordingly, the end-to-end length r in the deformed
configuration can be expressed in terms of the deformation
gradient F or rather in terms of the right Cauchy-Green
tensor C,
q q q
r l2I n

I  n
I l2I n0I  C  n0I l2I I
IC 5
with explicit summation over all three direction I = 1,2,3.
Here, we have introduced the non-standard invariants
I
IC n
I n0I  C  n0I with the understanding that I
IC

I  n
represents the stretch in the n0I direction squared. Thereby,
n0I are the unit normal vectors of the unit cell axes in the
undeformed reference configuration, see figure 3. After the
deformation, they map onto the vectors n
I F  n0I which
are obviously no longer of unit length.
Finally, the repulsive contribution W
rep
lnI
C I
C I
C
1 2 3
is constructed to compensate for the chain stresses in the
reference configuration caused by non-vanishing initial
Fig. 2 Kinematics of the eight-chain modelorthotropic case (left),
transversely isotropic case (middle), and isotropic case (right)
n01 =nI
l1 +
I
n02 = nII
n03 =nIII l2 +
II
F, u l3 +
III
Fig. 3 Remodeling based on changes of cell dimensions s
Instantaneous alignment of cell axes n0I with eigenvectors nrI and
gradual adaptation of cell dimensions lIwith respect to eigenvalues
kr
I
123
8816 J Mater Sci (2007) 42:88118823

end-to-end lengths r0. Classically, the free energy W assume the chain number density to be constant, however, its
introduced in (3) defines the Cauchy stress r in the fol- evolution in time due to changes in collagen content and
lowing form, thickness could be incorporated straightforwardly. The
degree of anisotropy is governed by the unit cell dimensions
1 dW t l1, l2,p
and
r  F riso rchn rrep 6 l3 which implicitly define the end-to-end length
J dF r0 l2I in the initial configuration.
whereby the individual stress contributions can be
expressed as follows.
On the tissue level
1
riso k lnJ
lI lb In cardiovascular tissues, the collagen fibers are not arbi-
J " #
chn trarily distributed in space but follow rather a particular
n 1 1 1
rchn kh

chn
r 7
pattern. To account for remodeling in the form of micro-
4AJ L 4r1
r=L2 4r structural rearrangement, we allow the fiber direction to
" #
nchn 1 1 1 rotate in response to the current mechanical stress envi-
rep
r kh

rep
r ronment. We assume the Cauchy stress tensor
4AJ L 4r0 1
r0 =L2 4r0
r krI nrI  nrI 8
Herein, the Finger deformation tensor b F  Ft repre-
sents a characteristic spatial strain measure. The tensorial to be the driving force of the remodeling process. Here, we
basis of the chain stress follows straightforwardly as have introduced its eigenvalue decomposition with krI and

chn l2I n
r
I :Note that the repulsive energy Wrep has

I  n nI being the corresponding eigenvalues and eigenvectors,
been constructed such that the corresponding tensorial respectively. For notational simplicity, we have implicitly

rep
l2I =I
IC n
basis of the repulsive stress r
I  n

I ensures a assumed the summation over all I = 1,2,3 components. Our
stress free reference configuration as rrep
rchn jrr0 for remodeling theory is based on two fundamental hypothe-
which I
IC 1 for all I = 1,2,3. ses:

Remarks 3 [Special cases of transverse isotropy and
isotropy] The orthotropic eight-chain model, as introduced
by Bischoff et al. [3, 4], can be understood as a kinematic
generalization of the original isotropic eight-chain model,
as indicated in Fig. 2.
Hypothesis I: The characteristic directions n0I of the
microstructure align instantaneously with respect to the
eigenvectors nrI :
Hypothesis II: The unit cell dimensions lI adapt
gradually with respect to the positive eigenvalues kr
I :

Transverse The first postulate is motivated by the general idea

p isotropy l1 = r0 and l2 = l3 = 0 and thus
r I
1C r0
p p
Isotropy l1 l2 l3 r0 = 3 and thus r I1C =3 r0
The case of transverse isotropy follows by choosing
l2 = l3, its yet more special case is the consideration of one
single fiber direction n01 with l1 = r0 and l2 p
=
l3 = 0. In this
particular case, the end-to-end length r I
1C r0 is obvi-
ously equivalent to the scaled stretch along the fiber
direction n01 : The special case of isotropy follows by
assuming that all p three cell dimensions are equal
l1 p l3 r0 = 3 such that the end-to-end length
l2 n0I  nrI
r I1C =3 r0 can be expressed exclusively in terms of the
trace of the right Cauchy-Green tensor I1C C : I:
Remarks 4 [Parameters on the extracellular matrix level]
On the extracellular matrix level, the mechanical behavior is
characterized through eight parameters in total, i.e., Lame
constants k and l characterizing the surrounding substrate,
the micromechanically motivated chain parameters L and A
and the chain number density nchn. For the moment, we shall
beyond all network models that the unit cell is taken to
deform in the principal stretch space, see, e.g., Boyce and
Arruda [6]. The second postulate closely follows the recent
approach of Hariton et al. [25] suggesting that the collagen
fibers in cardiovascular tissues are located between the
directions of the two maximum principal stresses.
Accordingly, the essence of our remodeling theory can be
summarized as follows,
krI
and lI ! r0 9
jjkrI jj
compare with Fig. 3. For the sake of notational simplicity,
we have introduced the notion krI with the understanding
that krI krI for positive eigenvalues krI > 0 and krI 0
for non-positive eigenvalues krI  0: From a physiological
point of view, we thus exclusively allow for tension
driven reorientation processes. It remains to formulate a
reasonable evolution equation for the microstructural cell
dimensions lI which we assume to obey the following law

123
J Mater Sci (2007) 42:88118823 8817

r 
kI l0I dt n01 x  n01
dt lI j
exp
jtr0 10
jjkr
I jj r0
which is obviously more cumbersome due to numerical
where j denotes a relaxation parameter and l0I ; I = 1,2,3, singularities introduced through trigonometric functions.
are the dimensions of the cuboid in the undeformed In contrast to the rather simple explicit equation for
configuration. Alternatively, the above equation could be the adaptation of the unit cell dimensions (11), the
integrated in time to render an explicit update equation for fiber direction can be expressed through the exponential
all cell dimensions lI. update


kr l0I
lI I

1
exp
jtr0 l0I 11
jjkr
I jj r 0
Recall that the reorientation
p process itself leaves the
initial chain length r0 l2I unaltered.
Remarks 5 [Remodeling based on fiber rotation] Previous
remodeling approaches in the literature focused on the
reorientation of a single microstructural direction, see, e.g.,
Menzel [46], Kuhl et al. [37] and Himpel et al. [26].
Applications thus focused primarily on transversely
isotropic biological tissues with one distinguished fiber
orientation such as muscles, tendons or ligaments for which
l2 = l3  l1. Similar to the previous approach, reorientation
was assumed to be driven by either strain or stress, based on
an eigenvalue decomposition, as given in (8). In contrast to
the new model suggested in this contribution, the following
assumptions were postulated for the remodeling process:
The characteristic direction n01 of the microstructure
aligns gradually with respect to the eigenvector nrmax
related to the maximum positive eigenvalue kr
max :
The unit cell dimensions lI remain unaffected by the
process of remodeling.
Consequently, the reorientation-based analogue of Eq.
(9) could be expressed as follows, compare Fig. 4.
n01 ! nrmax and lI const.
Rather than updating the scalar-valued unit cell lengths
lI, this reorientation approach requires an update of the
vector-valued fiber direction n0I
Fig. 4 Remodeling based on changes of cell orientation s Gradual
alignment of the cell axes n01 with eigenvector nrmax of related
maximum principal eigenvalue kr
max and constant cell dimensions
lI = const
1
n0k1
3
exp
Dt e x  n0k x n0k1  nrmax
1 1
j 1
3
where x and e denote the time discrete rotation vector and
the third-order permutation tensor, respectively. Recall that
in addition to potential numerical difficulties, this previous
reorientation approach is restricted to transversely isotropic
biological tissues and its extension to reorienting multiple
directions seems to be a rather complex task.
Remarks 6 [Dissipation] It is a well-accepted fact that a
purely mechanical theory is thermodynamically inadmis-
sible for remodeling processes that stiffen the material, see,
e.g., the discussions in Menzel [46], Kuhl et al. [37], and
Himpel et al. [26] or Garikipati et al. [20]. The fact that the
dissipation would be positive for stiffening materials in a
purely mechanical theory indicates that other thermody-
namic phenomena, e.g., of chemo-mechanical nature,
should indeed be taken into account. Alternatively, mixture
theories could be considered in which energy and entropy
are exchanged amongst the individual constituents of the
tissue, see, e.g., Gleason and Humphrey [22, 23].
Accordingly, to date, there is no general agreement of how
evolution laws for reorientation of microstructural direc-
tions should be formulated. In the context of linear elas-
ticity, it has been shown that the free energy attains an
extremum if strain and stress share the same principal
directions, see, e.g., Cowin [10] or Vianello [54]. In non-
linear elasticity, however, it is not even clear to date
whether stress or strain is the relevant driving force for the
remodeling process. The previous n01 ! nrmax non-linear
reorientation model addressed in Remark 5 is based on the
general paradigm that nature always tries to find the
extremum, see Menzel [46], Kuhl et al. [37] and Himpel
et al. [26]. However, the recent reorientation model for
arteries by Hariton et al. [25] just postulated a general
stress driven remodeling process and so does the model
presented herein. For detailed discussions about the impact
of the dissipation inequality in remodeling, the reader is
referred to the excellent discussion by Garikipati et al.
[20].
Remarks 7 [Parameters on the tissue level] An additional
convincing benefit of the incorporation of remodeling is

123
8818
that the number of parameters could even be reduced, as
compared to the fixed cell dimension model introduced in
Sect. Governing equations. For the overall model of
adapting living tissue, seven material parameters are
required altogether, i.e., the two Lame constants k and l
for the extracellular matrix, the chain number density nchn,
the micromechanically motivated contour length L and the
persistence length A. However, now, instead of prescribing
fixed cell dimensions l1, l2, and l3, we just have to define
the initial chain length r0, while the individual dimensions
of the cell evolve naturally in response to the given
mechanical loading environment.
Nevertheless, one additional parameter j is introduced
to account for the speed of adaptation. From a biological
point of view, the relaxation parameter j reflects the
turnover rate or rather the speed of adaptation. Note that we
have implicitly assumed that the time scale for fiber
reorientation is orders of magnitude larger than the typical
time scale of a diastolic-systolic pressure circle. The role of
j is illustrated in Fig. 5 for a simulation with
k = 27.293, l = 3.103, nchn = 7.0 1021 chains per unit
volume, L = 2.125, A = 1.82, and r0 = 1.0. In the initial
state, we assumepan isotropic fiber distribution with
l1 l2 l3 r0 = 3: The curves illustrate the angle of
alignment for a cubic specimen of unit length subject to
uniaxial tension with an incremental loading of DF = 1.0.
The load is applied in 10 steps in the dimensionless time
interval 0 < t 10. Then, at F = 10.0, the load is held
constant while the cell dimensions are allowed to remodel
progressively at 10 < t 100. Figure 5 demonstrates that
the collagen fibers align gradually with respect to the
loading axis. As time evolves, l1 r0 while l2 = l3 0
whereby n01 is instantaneously aligned with the loading
axis. As expected, the speed of the adaptation process
increases for larger values of the relaxation parameter j.
60
50
angle of alignment
40
30
20
10
0
0 20 40 60 80
time
Fig. 5 Tissue level s Angle between collagen fibers and direction of
uniaxial tension versus time for varying relaxation parameters j
123
J Mater Sci (2007) 42:88118823
Computational examples
Finally, the features of the proposed remodeling strategy is
elaborated by means of two benchmark problems, i.e., a
cylindrical tendon subject to uniaxial tension, and a tube-
like artery subject to uniaxial stretch in combination with
internal pressure.
Since we aim at elaborating inhomogeneous structures
with an initially random fiber orientation, the remodeling
strategy is embedded in a standard finite element algorithm
with the remodeling equation being evaluated at the inte-
gration point level. A NewtonRaphson solution strategy
based on a consistent linearization of the governing equa-
tions is applied throughout to solve the non-linear biome-
chanical problem efficiently in an incrementally iterative
way.
In the following examples, we choose the Lame param-
eters to k = 27.293 and l = 3.103 according to Menzel
[45]. The Boltzmann constant is k = 1.381 1023, and the
absolute temperature is h = 310, measured in Kelvin.
Recall that all chain lengths are normalized by a length
scale originating from the statistical model. Following
Garikipati et al. [19], we scale all lengths by the link length
in a chain. Accordingly, the contour length is chosen to be
L = 1.594, the corresponding persistence length that ac-
counts for a reasonable initial stiffness
p is A = 1.365, and the
initial end-to-end length is r0 l2I 1:0: The degree of
anisotropy is reflected through the chain number density
nchn, i.e., the number of chains per unit volume, which we
choose in the order of 1021, as suggested by Bischoff et al.
[3] or Garikipati et al. [19], thus nchn = 2 1021. The
relaxation parameter is chosen to be j = 0.025 per unit
time. In all examples, we start with an initially random fiber
orientation which is realized by assigning different values to
the unit cell dimensions lI in each element of the mesh. Due
to the geometric and material non-linearities of the problem,
the load is applied incrementally until the final load level is
reached. Then, the load is held constant while remodeling
occurs until convergence towards a biological equilibrium
state.
Cylindrical tendon under uniaxial tension
To elaborate the proposed remodeling algorithm in the
simple case of uniaxial loading, we analyze a cylindrical
tendon under uniaxial tension which has been studied
earlier in the context of tissue engineering of tendon
constructs by Kuhl et al. [37]. When strained along its
load-bearing axis, tendon shows a microscopical straight-
100 ening of the low amplitude wavy collagen fibers accom-
panied by an exponentially increasing resistance to
elongation. The model tendon we aim at analyzing quali-
tatively to elaborate whether the suggested model is able to
J Mater Sci (2007) 42:88118823
capture these general trends is 10 units long and has an
initial cross-section of 10 units2. It is discretized with 10
elements in height and 48 elements per cross-section, thus
with 480 tri-linear finite elements in total. The load is
increased incrementally with DF = 25 from 0 < t 12 until
a final load level of F = 300 is reached. From 12 < t 150,
we apply another 138 remodeling steps until biological
equilibrium occurs.
Figure 6 illustrates the remodeling history of the cylin-
drical tendon. At first, when the load is applied, the tendon
is stretched in the axial direction up to about 175% of its
original length, compare with Fig. 6, left. However, as time
evolves, the collagen fibers tend to align gradually with the
loading axis. Accordingly, the structure stiffens signifi-
cantly and the overall stretch decreases to about 140%. At
the final and yet stiffest state, all fibers are aligned with the
direction of the mechanical load.
During the remodeling process, the unit cell lengths
p evolved from initially random values 0 lI
have obviously
r0 with l2I r0 to l1 = r0 and l2 = l3 = 0 representing
the special case of transverse isotropy. Selected snapshots
of the remodeling history are depicted in Fig. 6, right. For
each element, the end-to-end vectors of each chain have
been projected on the tendons surface. The individual
colors represent the collagen fiber angle measured against
the loading axis. Red colors indicate a full alignment with an
angle zero, while blue colors indicate that the collagen fibers
are oriented orthogonal to the loading axis. In particular, in
the second snapshot of the series at t = 12 corresponding to
the final loading state, the spatial inhomogeneity is nicely
visible. The red elements with aligned collagen fibers display
very stiff response and deform only marginally while the soft
blue elements undergo significant stretches. As expected,
this initial inhomogeneity tends to vanish gradually in the
course of remodeling.
Tube-like artery subject to uniaxial stretch and internal
pressure
We turn now to the more challenging example of a multi-
axial loading state induced by a uniaxial stretching in
Fig. 6 Cylindrical tendon 1.8
under uniaxial tension s Stretch 1.7
versus time (left) and selected 1.6
snapshots of the remodeling 1.5
stretch
process at different times with
1.4
colors representing the
1.3
individual collagen fiber angles
(right) 1.2
1.1
1
0 50
8819
combination with an internal pressure representing the blood
flow. It has long been recognized that the layered structure of
cardiovascular tissues displays a geometric complexity way
beyond the highly organized parallel bundles of collagen
fibers that constitute the mechanical backbone of tendons.
Motivated by earlier theoretical studies of Driessen et al.
[13] and Hariton et al. [25], and supported by the experi-
mental findings of Finlay et al. [16], we analyze the
remodeling history in a tube-like artery. At this point, we
restrict ourselves to a rather qualitative analysis aiming at
including physiologically realistic data at later stages of this
project. The initial dimensions are given through a length of
8 units and an inner and outer radius of 1 and 3 units,
respectively. Along the height we apply 12 elements, we use
8 elements across the thickness and 16 elements in circum-
ferential direction. Here, we apply standard tri-linear finite
elements. Note, however, that shell elements would be more
appropriate when it comes to quantitative studies. The load is
applied incrementally in 25 load steps from 0 < t25 with
Dp = 0.192 p0 and Dl = 0.032 until a final state of p = 4.8 p0
and Dl = 0.8, i.e., an axial stretch of 10%, is reached. Again,
the load is then held constant for another 125 time steps for
the time period 25 < t150 to allow for remodeling towards
a final state of biological equilibrium. Figure 7 shows five
selected snapshots of the remodeling process. In between the
first and the second snapshot, the internal pressure and the
prescribed axial stretch are increased incrementally.
Accordingly, the tube blows up and stretches along its axial
direction occur. Due to the nearly incompressible behavior of
soft biological tissues, the tube thickness decreases drasti-
cally in response to loading. As time evolves, the collagen
fibers tend to reorient so that they are finally located between
the two directions of maximum principal stress. For this
particular case of loading, these are obviously the longitu-
dinal and the circumferential directions.
Starting from an initial random orientation
p at random
unit cell lengths 0 lI r0 with l2I r0 in the first
snapshot of the series, the unit cell lengths evolve pro-
gressively driven by the positive eigenvalues of the Cauchy
stress kr
I : As a natural consequence, the chain directions
remodel gradually and the tube stiffens with respect to the
100 150
time
123
8820
Fig. 7 Tube-like artery subject
to axial stretch and internal
pressure s Selected snapshots
of the remodeling process at
different times with colors
representing the individual
collagen fiber angles
applied loading. Accordingly, the outer radius, which had
increased during the first stages of loading, decreases
remarkably in response to remodeling, compare time t = 24
with t = 150. While the chains had originally been oriented
randomly in space at t = 1, their component in radial
direction vanishes gradually as l3 goes to zero in response
to the compressive stresses kr3  0 in the radial direction at
t = 150. Thus, the suggested remodeling algorithm proves
able to produce not only transversely isotropic but also
orthotropic microstructures in a natural way.
Recall that the final collagen fiber angles developed
naturally in response to the positive eigenvalues krI of the
Cauchy stress r which had been postulated to be the
driving force of the remodeling process. Due to the inho-
mogeneous stress state across the radial direction, different
fiber angles arise in the individual layers.
Figure 8 depicts the final collagen fiber orientations at
three different radial locations representing the intima, the
media and the adventitia. The results suggest that the
transmural pitch of the fiber orientation increases from the
inner to the outer wall. The analytically predicted double-
helix architecture of the collagen fibers agrees very well
with experimental observations by Finlay et al. [16] and
Holzapfel et al. [27, 30]. In addition, our results nicely
agree with the more recent numerical studies by Driessen
et al. [13] and Hariton et al. [25].
Fig. 8 Tube-like artery subject to axial stretch and internal pressure
s Final stage of the remodeling process with collagen fiber
orientations projected on the individual layers
123
J Mater Sci (2007) 42:88118823
The predicted inhomogeneity in the radial direction with
an almost circumferential fiber orientation at the luminal
side of the artery and an increase of the inclination towards
the outer side is closely related to the phenomenon of
prestress, typically encountered in arterial specimens. For
the considered model problem, the degree of inhomoge-
neity, obviously, strongly depends on the stretch-to-pres-
sure ratio. Motivated by a recent study by Gleason and
Humphrey [23] who analyzed the elastin, collagen and
smooth muscle cell turnover and remodeling in response to
different loading scenarios, we systematically elaborate
combinations of transmural pressure and axial stress. To
illustrate the influence of the mechanical loading situation
on the remodeling process, we display Fig. 9 to show the
final biological equilibrium stages for different stretches at
different pressure levels. For the sake of visibility, we have
virtually peeled off the individual layers after the calcula-
tion of the entire tube had been performed. Experimentally,
the definition of these specific layers could be made
apparent by the use of stains differentiating smooth mus-
cles cells from collagen fibers. For pure stretching, as
depicted in Fig. 9 (left), the fibers of all layers are aligned
with the loading axis, i.e., l1 = r0 and l2 = l3 = 0 with
n01 pointing in the axial direction. For pure pressure,
as depicted in Fig. 9 (right), all fibers are oriented in
the circumferential direction, i.e., again, l1 = r0 and
l2 = l3 = 0, however, now with n01 pointing in the circum-
ferential direction. These computational results agree well
with the experimental findings of human brain arteries by
Finlay et al. [16] who reported a significant realignment of
collagen with increasing distending pressure for all arterial
layers and a recruitment towards the circumferential
direction. Intermediate stages of combined stretch and
pressure loading, as shown in the three images between the
left and the right image of Fig. 9, reveal the usually
observed helical fiber arrangement. Based on these results,
it is obvious to suggest that the collagenous architecture
strongly depends on the mechanical environment.
In the biomechanics community there is an ongoing
debate over which quantity drives the remodeling process.
To this end, we analyze the influence of different driving
forces of tensorial nature. For purpose of comparison with
J Mater Sci (2007) 42:88118823 8821

Fig. 9 Tube-like artery subject

to axial stretch and internal % % % % %
pressure s Variations of
collagen fiber orientations in
response to different stretch-to-
pressure ratios with colors
representing the individual
collagen fiber angles

the (spatial) Cauchy stress r used in the previous examples,
we now elaborate the use of the (material) second Piola
Kirchhoff stress S JF
1  r  F
t ; as the representative
material stress measure. Figure 10 compares now the out-
come of a remodeling process driven by spatial and
material stresses. The diagram illustrates the fiber orien-
tation angle over the radius. Variations in fiber orientation
seem to be slightly more pronounced when using the sec-
ond PiolaKirchhoff stress as the driving force. Neverthe-
less, for this particular problem, it appears that differences
are rather marginal. We suggest that the Cauchy stress,
which is the true stress experienced by the deformed
structure, is the more reasonable choice.
Finally, we elaborate the difference of a stress and
strain-based remodeling criterion, as used by, e.g., Driessen
et al. [13], Menzel [46] and Humphrey [32], Hariton et al.
[25], respectively. To this end, we compare the outcome of
two different remodeling processes based on either the
Cauchy stress r as the characteristic spatial stress measure,
see Fig. 11 (left), or on the Finger deformation tensor b as
the characteristic spatial strain measure, see Fig. 11 (right).
Both the curves as well as the plots of the structure dem-
onstrate that strain-based remodeling is much less sensitive
to the radial variation. The difference between an almost
circumferential arrangement at the luminal side and a
rather helical structure towards the outermost layer is less

Fig. 10 Tube-like artery

subject to axial stretch and 40
internal pressure s Collagen 35
fiber orientation versus radial
30
position (left) and outcome of
angle

spatial versus material stress- 25

based remodeling (right) 20
15
10
5
0
2.2 2.4 2.6 2.8 3 3.2 3.4
radius

Fig. 11 Tube-like artery 40

subject to axial stretch and
35
internal pressure s Collagen
fiber orientation versus radial 30
position (left) and outcome of 25
angle

spatial stress versus spatial

20
strain-based remodeling (right)
15

10

0
2 2.2 2.4 2.6 2.8 3 3.2 3.4
radius

123
8822
pronounced for the strain driven case. This finding is, of
course, rooted in the non-linear nature of the force-
displacement behavior of the individual chains, recall
Fig. 1. While the strain varies slowly close to the locking
stretch, corresponding stresses may vary significantly.
Accordingly, the stress-based remodeling criterion predicts
a wider variation in collagen fiber angles than the strain
driven remodeling algorithm. Although for hard tissues,
such as bone, several studies including the one by Cowin
[9] suggest that strain is the critical mechanical factor in
growth and adaptation, we postulate that stress is a more
reasonable choice as the driving force in the context of soft
tissues such as arteries, tendons or ligaments. This obser-
vation is along with Taber and Humphrey [52] who suggest
that stress and not strain correlates well with growth in
arteries.
Discussion
A continuum theory of remodeling for soft biological tis-
sues has been proposed with a particular focus on cardio-
vascular tissue. In this context, remodeling is attributed to
the collagen fiber reorientation as a natural consequence of
changes in the mechanical loading environment. Hence, in
our approach the optimal configuration of collagen fibers
depends solely on the external loading acting upon the
living tissue. Based on a homogenization strategy from the
molecular microscale via the mesoscale of the extracellular
matrix to the macroscale of the overall tissue, we derived a
statistical mechanics-based chain network model governed
by a limited set of physically motivated material parame-
ters. In addition to the two (classical) Lame constants, five
additional parameters sufficed to characterize the highly
non-linear, exponentially stiffening of the collagen mor-
phology: the contour length, the persistence length and the
initial end-to-end length of the collagen chains, the chain
number density accounting for the degree of anisotropy and
the turnover rate. The concept of chain network models
was applied to transmit information from the molecular
level to the extracellular matrix level. In contrast to exist-
ing remodeling theories which are based on complex
rotational updates of characteristic microstructural direc-
tions, our theory essentially captures remodeling in the
form of changes of the dimensions of this representative
element. By means of simple model problems, we were
able to show that the theory indeed succeeded in charac-
terizing mechanically introduced remodeling of collagen
fibers.
Finally, we elaborated two biomechanically relevant
boundary-value problems, a cylindrical tendon subject to
uniaxial tension and a tube-like artery loaded by an axial
stretch in combination with different internal pressures. For
123
J Mater Sci (2007) 42:88118823
these more complex inhomogeneous (living) structures, the
remodeling algorithm was embedded in a non-linear finite
element program, the equations were linearized consis-
tently and solved with an incremental iterative Newton
Raphson solution strategy. Starting with initially random
fiber orientation, the algorithm generated collagenous
architectures which qualitatively resembled experimental
observations. In all computational examples, the suggested
algorithm convinced through its remarkable stability and
robustness. What remains, however, is the realization of
quantitative validations of the suggested theory and its
algorithmic realization. In addition to the collagen fiber
orientation, which could eventually be determined through
techniques such as microscopy, histology, magnetic reso-
nance diffusion tensor imaging, the chain number density
and the turnover rate remain to be classified. The latter is
assumed to be of particular relevance in the context of
improving cardiovascular surgery and predicting patient-
specific remodeling in response to medical treatment.
Finally we would like to point out once again, that
although the reorientation approach outlined in the present
manuscript is motivated by micromechanical consider-
ations, it is still rather phenomenological to most extend. In
contrast to purely invariant based formulations, it intro-
duces material parameters such as the contour length or the
persistence length which have a clear physical interpreta-
tion. The computational mechanics community would thus
certainly classify the model as bottom-up or rather micro-
mechanically based and non-phenomenological. In the cell
biology community, however, the same model would most
probably be considered rather phenomenological or top-
down since it does not explicitly address cell level
phenomena like mechanotransduction or the biochemical
origin of the remodeling process as such. Nevertheless, to
date it seems unmanageable to simulate large tissue
structures or organs and yet at the same time to account for
every single biochemical phenomenon individually. To the
most extend, these complex mechanisms are not even fully
understood at this point.
Successful constitutive models require an understanding
of the functional interactions between the key components
of cells up to the organs, and how these interactions change
from a physiological to a pathological state. Such infor-
mation resides neither in the individual genome nor in the
protein. This information is contained in the interactions of
proteins with cellular, organ and system structures. The
identification of these interactive relationships is clearly
within the focus of intense current research. Continuum
models like the one presented herein will certainly benefit
from gradually incorporating more and more information
from the protein, subcellular and cellular level in the future
to define precisely how mechanical forces translate into
chemical signals that initiate the process of remodeling.
J Mater Sci (2007) 42:88118823
References
1. Alberts B, Johnson A, Lewis J, Raff M, Roberts K, Walter P
(2002) Molecular biology of the cell, 4th edn. Garland Science
Taylor Francis Group, New York
2. Arruda EM, Boyce MC (1993) J Mech Phys Solids 41:389
3. Bischoff JE, Arruda EM, Grosh K (2002) J Appl Mech 69:570
4. Bischoff JE, Arruda EM, Grosh K (2002) J Appl Mech 69:198
5. Boyce MC (1996) Rubber Chem Technol 69:781
6. Boyce MC, Arruda EM (2000) Rubber Chem Technol 73:504
7. Bustamante C, Bryant Z, Smith SB (2003) Nature 421:423
8. Bustamante C, Smith S, Marko JF, Siggia ED (1994) Science
265:1599
9. Cowin SC (1984) Calc Tissue Int 36:S99
10. Cowin SC (1995) J Elast 34:45
11. Driessen NJB, Bouten CVC, Baaijens FPT (2005) J Biomech Eng
127:494
12. Driessen NJB, Peters GWM, Huyghe JM, Bouten CVC, Baaijens
FPT (2003) J Biomech 36:1151
13. Driessen NJB, Wilson W, Bouten CVC, Baaijens FPT (2004) J
Theor Biol 36:53
14. Elbischger PJ, Bischof H, Holzapfel GA, Regitnig P (2005) In:
Suri JS, Yuan C, Wilson DL, Laxminarayan S (eds) Plaque
imaging: pixel to molecular level, vol 113. Studies in Health
Technology and Informatics, IOS Press, p 97
15. Elbischger PJ, Bischof H, Regitnig P, Holzapfel GA (2004)
Pattern Anal Appl 7:269
16. Finlay HM, Mc Cullough L, Canham PB (1995) J Vasc Res
32:301
17. Flory PJ (1969) Statistical mechanics of chain molecules. John
Wiley Sons, ChichesterNew York
18. Freed AD, Einstein DR, Vesely I (2005) Biomech Model
Mechanobiol 4:100
19. Garikipati K, Arruda EM, Grosh K, Narayanan H, Calve S (2004)
J Mech Phys Solids 52:1595
20. Garikipati K, Olberding JE, Narayanan H, Arruda EM, Grosh K,
Calve S (2006) J Mech Phys Solids 54:1493
21. Gasser TC, Ogden RW, Holzapfel GA (2006) J R Soc Interface
3:15
22. Gleason RL, Humphrey JD (2004) J Vasc Res 41:352
23. Gleason RL, Humphrey JD (2005) Math Med Biol 22:347
24. Hagerman PJ (1988) Annu Rev Biophys Biophys Chem 17:265
25. Hariton I, de Botton G, Gasser TC, Holzapfel GA (2006) Bio-
mechanics and modeling in mechanobiology, pp available online,
DOI 10.1007/s10237-006-0049-7
26. Himpel G, Menzel A, Kuhl E, Steinmann P (2007) Int J Num
Meth Eng, in press
8823
27. Holzapfel GA, Gasser TC, Stadler M (2002) European J Mech A
Solids 21:441
28. Holzapfel GA, Ogden RW (2003) Biomechanics of soft tissue in
cardiovascular systems CISM courses and lectures no 441.
Springer Verlag, WienNew York
29. Holzapfel GA, Ogden RW (2006) mechanics of biological tissue.
Springer, BerlinHeidelbergNew York
30. Holzapfel GA, Stadler M, Schulze-Bauer CAJ (2002) Ann Bio-
med Eng 30:753
31. Huang H, Kamm R, Lee RT (2003) Am J Physiol Cell Physiol
287:C1
32. Humphrey JD (2001) J Biomech Eng 123:638
33. Humphrey JD (2002) Cardiovasular solid mechanics. Springer
Verlag, BerlinHeidelbergNew York
34. Ingber DE (2003) Ann Intern Med 35:564
35. Klein-Nulend J, Bacabac RG, Mullender MG (2005) Pathol Biol
(Paris) 53:576
36. Kratky O, Porod G (1949) Recl Trav Chim 68:1106
37. Kuhl E, Garikipati K, Arruda EM, Grosh K (2005) J Mech Phys
Solids 53:1552
38. Kuhl E, Maas R, Himpel G, Menzel A (2006) Biomechanics and
modeling in mechanobiology, pp available online, DOI 10.1007/
s10237-006-0062-x
39. Kuhl E, Menzel A, Garikipati K (2006) Phil Mag 86:3241
40. Lehoux S, Castier Y, Tedgui A (2006) J Intern Med 259:381
41. Leung DYM, Glagov S, Mathews MB (1976) Science 191:475
42. Leung DYM, Glagov S, Mathews MB (1977) Circ Res 41:316
43. Lubarda VA, Hoger A (2002) Int J Solids Struct 39:4627
44. Marko JF, Siggia ED (1995) Macromolecules 28:8759
45. Menzel A (2006) In: Holzapfel GA, Ogden RW (eds) IUTAM
Symposium mechanics of biological tissue. Springer Verlag, p 91
46. Menzel A (2005) Biomech Model Mechanobiol 3:147
47. Mofrad MRK, Kamm R (eds) (2006) Cytoskeletal mechanics
models and measurements. Cambridge University Press
48. Nerem RM, Seliktar D (2001) Annu Rev Biomed Eng 3:225
49. Rodriguez EK, Hoger A, Mc Culloch AD (1994) J Biomech
27:455
50. Stopak D, Harris AK (1982) Dev Biol 90:383
51. Sun YL, Luo ZP, Fertala A, An KN (2002) Biochem Biophys Res
Commun 295:382
52. Taber LA, Humphrey JD (2001) J Biomech Eng 123:528
53. Treloar LRG (1975) The physics of rubber elasticity. Clarendon
Press, Oxford
54. Vianello M (1996) J Elast 42:283
55. Wang JH, Thampatty BP (2006) Biomech Model Mechanobio 5:1
123
J Mater Sci (2007) 42:88248837
DOI 10.1007/s10853-007-1859-4
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Micromechanics of crystal interfaces in polycrystalline solid
phases of porous media: fundamentals and application to strength
of hydroxyapatite biomaterials
Andreas Fritsch Luc Dormieux Christian Hellmich
Julien Sanahuja
Received: 20 February 2007 / Accepted: 17 May 2007 / Published online: 7 August 2007

Springer Science+Business Media, LLC 2007
Abstract Interfaces are often believed to play a role in
the mechanical behavior of mineralized biological and
biomimetic materials. This motivates the micromechanical
description of the elasticity and brittle failure of interfaces
between crystals in a (dense) polycrystal, which serves as
the skeleton of a porous material defined one observation
scale above. Equilibrium and compatibility conditions,
together with a suitable matrix-inclusion problem with a
compliant interface, yield the homogenized elastic prop-
erties of the polycrystal, and of the porous material with
polycrystalline solid phase. Incompressibility of single
crystals guarantees finite shear stiffness of the polycrystal,
even for vanishing interface stiffness, while increasing the
latter generally leads to an increase of polycrystal shear
stiffness. Corresponding elastic energy expressions give
Andreas FritschOn leave from Institute for Mechanics of Materials
and Structures, Vienna University of Technology (TU Wien), 1040
Vienna, Austria.
A. Fritsch
L. Dormieux
Laboratory of Materials and Structures, National School of Civil
Engineering (ENPC), Marne-la-Vallee 77455, France
A. Fritsch
e-mail: Andreas.Fritsch@tuwien.ac.at
L. Dormieux
e-mail: dormieux@lmsgc.enpc.fr
C. Hellmich (&)
Institute for Mechanics of Materials and Structures, Vienna
University of Technology (TU Wien), Vienna 1040, Austria
e-mail: Christian.Hellmich@tuwien.ac.at
J. Sanahuja
Lafarge Research Center, Saint-Quentin Fallavier Cedex 38291,
France
e-mail: julien.sanahuja@lafarge.com
123
access to effective stresses representing the stress hetero-
geneities in the microstructures, which induce brittle fail-
ure. Thereby, Coulomb-type brittle failure of the crystalline
interfaces implies DruckerPrager-type (brittle, elastic
limit-type) failure properties at the scale of the polycrystal.
At the even higher scale of the porous material, high
interfacial rigidities or low interfacial friction angles may
result in closed elastic domains, indicating material failure
even under hydrostatic pressure. This micromechanics
model can satisfactorily reproduce the experimental
strength data of different (brittle) hydroxyapatite bioma-
terials, across largely variable porosities. Thereby, the
brittle failure criteria can be well approximated by micro-
mechanically derived criteria referring to ductile solid
matrices, both criteria being even identical if the solid
matrix is incompressible.
Introduction
Interfaces are believed to often play a fundamental role in
the mechanical behavior of hierarchically organized bio-
logical materials. Accordingly, much attention has been
paid to the polymer-filled interfaces between ceramic tab-
lets in nacre [2, 20, 32, 33, 39, 40, 41], but the importance
of interfacial behavior was also discussed for other classes
of biological materials, such as bone [48].
To gain insight into these material systems, material/
microstructure models have been developed within differ-
ent theoretical frameworks, such as fracture mechanics and
scaling laws [20, 39, 40, 41], large-scale elastoplastic
Finite Element analyses [32, 33, 48], or periodic homog-
enization on the basis of a unit cell discretized by Finite
Elements [2].
J Mater Sci (2007) 42:88248837
In addition to such periodic, FE-based (computational)
homogenization approaches, analytical and/or semianalyt-
ical approaches of random homogenization (continuum
micromechanics [51, 52]) have been recently used as to
effectively predict the elastic properties of complicated
hierarchically structured material systems (such as bone
[19, 22, 24, 25], wood [29, 30], concrete [5, 23, 49], or
shale [50]), from the elasticity and the mechanical inter-
actionsover different observation scalesof nanoscaled
elementary components. Thereby, not every single detail of
the highly random microstructures, but only the essential
morphological features are considered, in terms of homo-
geneous subdomains (material phases) inside representa-
tive volume elements (RVEs, Fig. 1), their volume
fractions, their elasticity, and their mechanical interaction.
Theoretically, it has been recently well understood how to
extend these homogenization techniques to the ductile
failure of (bulk) phases [3, 4, 6, 10, 12, 15] (while appli-
cations to real materials [37] are more rare than for the
elastic case). In comparison, the treatment of brittle failure
and of interfaces in the framework of random homogeni-
zation is still a very open field: It is the focus of this
paperboth fundamentally, and in view of the failure of
biomimetic hydroxyapatite biomaterials.
Extending very recent results [14, 44], where inclusion
coatings and interfaces in porous polycrystals were mod-
eled, we here tackle the description of the elasticity and
failure of interfaces between crystals in a (dense) poly-
crystal, which serves as the skeleton of a porous material
defined one observation scale above (Fig. 2). Thereby, we
Fig. 1 Multistep homogenization: properties of phases (with charac-
teristic lengths of d and d2, respectively) inside RVEs with
characteristic lengths of or 2 , respectively, are determined from
homogenization over smaller RVEs with characteristic lengths of
2  d and 3  d2 , respectively
8825
(a) (b)
Fig. 2 (a) Polycrystal with interfaces (schematic representation of
volume Vi of crystal i and interface I ij between crystals i and j),
serving as skeleton in a porous material at larger observation scale (b)
show characteristic features of a corresponding new
micromechanics model, which is based on matrix-inclusion
problems with compliant interfaces [21, 27, 53], and which
turns out to reasonably explain the behavior of porous
hydroxyapatite biomaterials, especially for their brittle
failure in the compressive regime.
Fundamentals of continuum
micromechanicsrepresentative volume element
In continuum micromechanics [28, 47, 51, 52], a material
is understood as a macro-homogeneous, but micro-
heterogeneous body filling a representative volume ele-
ment (RVE) with characteristic length ;  d; d standing
for the characteristic length of inhomogeneities within the
RVE (see Fig. 1), and  L; L standing for the charac-
teristic lengths of geometry or loading of a structure built
up by the material defined on the RVE (see Table 1 for a
list of all symbols in this paper). In general, the micro-
structure within one RVE is so complicated that it
cannot be described in complete detail. Therefore, quasi-
homogeneous subdomains with known physical quantities
(such as volume fractions or elastic properties) are rea-
sonably chosen. They typically include 3D subdomains,
and may also include the 2D interfaces between the 3D
subdomains. They are called material phases; bulk and
interface phases, respectively. The homogenized
mechanical behavior of the overall material, i.e., the rela-
tion between homogeneous deformations acting on the
boundary of the RVE and resulting (average) stresses, or
the ultimate stresses sustainable by the RVE, can then be
estimated from the mechanical behavior of the aforemen-
tioned homogeneous phases (representing the inhomoge-
neities within the RVE), their dosages within the RVE,
their characteristic shapes, and their interactions. If a single
phase exhibits a heterogeneous microstructure itself, its
mechanical behavior can be estimated by introduction of an
RVE within this phase, with dimensions 2  d; comprising
again smaller phases with characteristic length d2  2 ;
and so on, leading to a multistep homogenization scheme
(see Fig. 1).
123
8826 J Mater Sci (2007) 42:88248837

Table 1 List of symbols Table 1 continued

AC Surface area of spherical crystal with radius a r Radial coordinate in spherical coordinate system
Aex Constant in solution of matrix-inclusion problem with S Fourth-order Eshelby tensor for spherical inclusions
compliant interface T Traction force vector acting on surface element of
Ai Surface area of crystal i interface
Ain Constant in solution of matrix-inclusion problem with Tn Normal component of T
compliant interface Tt Tangential component of T
a Characteristic crystal radius Ttcr Critical (maximum) tangential traction bearable by
Bex Constant in solution of matrix-inclusion problem with intercrystalline interface
compliant interface t Tangential vector to surface of a single crystal
Bin Constant in solution of matrix-inclusion problem with tr Trace of a second-order tensor
compliant interface
VC Volume of spherical crystal with radius a
Cex Constant in solution of matrix-inclusion problem with
compliant interface VC Surface of spherical crystal with radius a
CC Fourth-order stiffness tensor of single crystals within the Vi Volume of crystal i
RVE Vpoly Vi Surface of crystal i
Cpoly Fourth-order homogenized stiffness tensor of polycrystal Vpoly Volume of an RVE of polycrystal with compliant
with compliant interfaces interfaces
CPORO Fourth-order homogenized stiffness tensor of a porous VPORO Volume of an RVE of porous material the solid phase of
material the solid phase of which is a polycrystal with which is a polycrystal with compliant interfaces
weak interfaces VS Volume of solid phase within the RVE VPORO
d Characteristic length of inhomogeneities within the RVE x Position vector within an RVE, either Vpoly or VPORO
Epoly Second-order macroscopic strain tensor (related to RVE a Friction angle of interfaces between single crystals
Vpoly of polycrystal with compliant interfaces)
dij Kronecker delta (components of second-order identity
E0 Uniform strain imposed at infinity of matrix surrounding tensor 1)
inclusion with compliant interface
dI Dirac distribution supported on I
Epoly,v Macroscopic volumetric strain (related to RVE Vpoly of
e Second-order strain tensor field within single crystals
polycrystal with compliant interfaces)
filling RVE Vpoly of polycrystal with compliant
Epoly,d Macroscopic equivalent deviatoric strain (related to RVE interfaces
Vpoly of polycrystal with compliant interfaces)
h Latitudinal coordinate of spherical coordinate system
er Radial unit vector
j Kt0 a=lC Dimensionless quantity related to rigidity of interface
e1 ; e2 ; e3 Unit base vectors of Cartesian base frame
lC Shear modulus of single crystals
fi Volume fraction of crystal i within the RVE Vpoly
lpoly Homogenized shear modulus of polycrystal with
h Cohesion of interfaces between single crystals compliant interfaces (RVE Vpoly)
I Fourth-order identity tensor lPORO Homogenized shear modulus of a porous material the
I Entity of interfaces within polycrystalline RVE Vpoly solid phase of which is a polycrystal with compliant
I ij Interface between crystals i and j interfaces
J Volumetric part of fourth-order identity tensor I mpoly Homogenized Poissons ratio of polycrystal with
compliant interfaces (RVE Vpoly)
K Deviatoric part of fourth-order identity tensor I
n Displacements within and at the boundary of RVE Vpoly
K Second-order interface stiffness tensor
n Displacement discontinuity at the interfaces between
K 0 2K Second-order interface stiffness tensor in matrix-inclusion crystals
problem with compliant interface
nn  Normal component of n
Kn Normal interface stiffness (component of K)
nt  Tangential component of n
Kt Tangential interface stiffness (component of K)
n Displacement discontinuity at compliant interface of
kC Bulk modulus of single crystals generalized matrix-inclusion problem
kpoly Homogenized bulk modulus of polycrystal with compliant ni ; nj Displacements along interface I ij , in crystal i and j,
interfaces (RVE Vpoly) respectively
kPORO Homogenized bulk modulus of a porous material the solid n
Mean displacement at the interface I ij
phase of which is a polycrystal with compliant interfaces
nin Displacement field inside the inclusion surrounded by
Characteristic length of the RVE compliant interface and infinite matrix (related to
L Characteristic lengths of geometry or loading of a structure generalized matrix-inclusion problem)
built up by the material defined on the RVE nex Displacement field throughout the matrix surrounding
n Normal vector onto surface of a single crystal inclusion coated by compliant interface (related to
RVE Representative volume element generalized matrix-inclusion problem)

123
J Mater Sci (2007) 42:88248837 8827

Z !
Table 1 continued 1 XZ s
Epoly exdV n  ndS
Rpoly Second-order macroscopic stress tensor (related to RVE Vpoly Vpoly ij I ij
Vpoly of polycrystal with weak interfaces) Z X fi Z
1 X s

n  ndS
s

n  ndS
Rpoly;m Macroscopic mean stress (related to RVE Vpoly of 2
polycrystal with weak interfaces) Vpoly i oVi i
V i oVi
Rpoly;d Macroscopic equivalent deviatoric stress (related to
RVE Vpoly of polycrystal with weak interfaces) with location vector x; normal n onto the spherical surface
RPORO Second-order macroscopic stress tensor (related to RVE of the crystals,
VPORO of porous material the solid phase of which is a
polycrystal with weak interfaces)
n
ni nj =2 nj  n=2 ni n=2 3
RPORO;m Macroscopic mean stress (related to RVE VPORO of
porous material the solid phase of which is a as the mean displacement at the interface I ij ; Vi and
polycrystal with weak interfaces)
fi Vi =Vpoly as the volume and the volume fraction of the
RPORO;d Macroscopic equivalent deviatoric stress (related to RVE
VPORO of porous material the solid phase of which is a ith crystal, and Vi as its surface with area Ai. For crystals
polycrystal with weak interfaces) of the same shape and size (with volume VC and surface
r Second-order stress tensor field with in single crystals VC), and indiscernible average mean displacements at
filling RVE Vpoly of polycrystal with compliant their surfaces, (2) can be transformed to
interfaces
rin Stress field inside the inclusion surrounded by compliant Z
1 s

n  ndS
interface and infinite matrix (related to generalized Epoly 4
matrix-inclusion problem) VC oVC
rex Stress field throughout the matrix surrounding inclusion
coated by compliant interface (related to generalized
The corresponding macroscopic stresses Rpoly are equal
matrix-inclusion problem) to the spatial average of the (equilibrated) local stresses
r Scattering factor in two-membered evolution strategy rx inside the RVE Vpoly,
/ Longitudinal coordinate of spherical coordinate system Z
u Volume fraction of pores within the RVE VPORO 1
Rpoly hrxi rxdV
v lC =kC Dimensionless quantity related to compressibility of Vpoly Vpoly
single crystals X fi Z
W Macroscopic energy density rxdV
i
Vi V i
1 Second-order identity tensor
X fi Z

First-order tensor contraction x  rx
nxdS 5
: Second-order tensor contraction i
Vi oVi
 Dyadic product of tensors
For spherical crystals with radius a, surface VC with
area AC 4pa2 , and volume VC 4=3pa3 , (5) can be
Micromechanics of polycrystal with weak interfaces further transformed,
X fi Z
Micromechanical representation Rpoly aer x  rx
er xdS
4 3
i 3 pa oVC

We consider an RVE with volume Vpoly (Figs. 2a, 9a) X 3fi Z

e x  rx
er xdS
hosting single crystals of typically quasi-spherical shape i
AC oVC r
and of volume Vi, separated from each other by very thin Z
1
(essentially 2D) interfaces I ij between crystals i and j, all anx  rx
nxdS
VC oVC
interfaces making up the entity of interfaces I ; [I ij I ; Z
3
see Fig. 2. Macroscopic strains Epoly are imposed at the nx  rx
nxdS
AC oVC
boundary of the RVE Vpoly in terms of displacements n; Z
1
rxdV 6
on oVpoly : nx Epoly
x 1 VC V C

with x as the position vector within the RVE. The
geometrical compatibility of (1) with the local
microscopic strains ex in the crystals and the
displacement discontinuities n nj  ni at the
interfaces I ij between the crystals i and j implies [14]
with radial unit vector er being identical to the normal n.
Since the microscopic stresses are equilibrated (div r 0),
(5) and (6) imply [9, p. 118, 28], that the macroscopic
stresses act as traction forces Rpoly
n both at the boundary
of the RVE, oVpoly ; and those of single crystals, oVC ;

123
8828 J Mater Sci (2007) 42:88248837

on oVpoly and oVC : rx
nx Rpoly
nx 7 homogenization. This was recently shown quantitatively
The relation between Rpoly and Epoly depends on the
constitutive behavior of the single crystals and of the
interfaces between them.
Constitutive behavior of interfaces and single crystals
The interfaces are the weakest locations of the material, the
load bearing capacities of which are bounded according to
a Coulomb-type law,
for polycrystals consisting of perfectly disordered needles,
being either isotropic or anisotropic [18].
Homogenized elasticity of polycrystal with compliant
interfaces
As long as the interfaces behave elastically, the relation
between Rpoly and Epoly reads as
Rpoly Cpoly : Epoly 11

8x 2 I ij : Tt x  Ttcr ah  Tn x 8 with the macroscopic homogenized stiffness tensor of the

with friction angle a, cohesion h, and Tt and Tn as the
tangential and normal components of the traction force
T Tn n Tt t acting on an infinitesimal interface area
around x, with normal n, and t as the tangential unit vector,
t
n 0: We consider brittle interface failure once a crit-
ical value Tt Ttcr is reached in (8).
Below this critical value, the interface behaves linear
elastically, i.e., the interface traction Tx is related to a
displacement discontinuity nx encountered when
crossing the interface I ij along nx:
polycrystal, Cpoly 3kpoly J 2lpoly K, with bulk modulus
kpoly and shear modulus lpoly ; depending on the local
elastic properties CC and Kt.
Following [14], the establishment of this dependence is
based on the behavior of a composite solid consisting of a
spherical inclusion of radius a and a compliant interface
coating the inclusion, being itself embedded in an infinite
matrix exhibiting the elastic properties Cpoly of the
homogenized polycrystal, and being subjected to uniform
strains E0 at infinity (Fig. 3).
Mathematically, we have

Tx K
nx r\a : r CC : e
with 9 ra : T K 0
n
K Kn n  n Kt 1  n  n; Kn ! 1 with n n=2; K 0 2K 12
r>a : r Cpoly : e
K is the second-order interface stiffness tensor with infinite
r!1 : n ! E0
x
normal component Kn (no mutual interpenetration of
crystals), and positive tangential component Kt (allowing
For determination of kpoly, a purely spherical deforma-
for relative tangential movements of crystal surfaces). Also
tion, E0 E0 1 is imposed at r ! 1. Spherical symmetry
the bulk crystal phase inside the RVE Vpoly behaves linear
of both the loading and the geometry of the considered
elastically,
solid implies vanishing tangential displacement disconti-
nuities at the inclusion interface, nt   0. Since Kn ! 1,
8x 2 Vi : rx CC : ex 10 also nn  0 (no mutual interpenetration of crystals), and
the matrix-inclusion problem with compliant interfaces
with CC 3kC J 2lC K as the isotropic elastic stiffness
of the bulk material phase comprising all single crystals;
with bulk modulus kC and shear modulus lC. J 1=31  1
and K I  J are the volumetric and the deviatoric part of
the fourth-order identity tensor I; with components
Iijkl 1=2dik djl dil dkj ; the components of the second-
order unit tensor 1, dij (Kronecker delta), read as dij = 1 for
i = j and dij = 0 for i 6 j:
The assumption of crystal isotropy deserves to be
commented, since single crystals are generally anisotropic,
including approximately transversely isotropic hydroxay-
apatite [34]. However, hydroxyapatite anisotropy is not
very pronounced [34], and in addition, the disorder Fig. 3 Matrix-inclusion problem with compliant interface (general-
ized Eshelby problem): a spherical inclusion with interface is
of crystals (and of their principal material directions)
embedded in an infinite matrix subjected to uniform strain E0 at
probably renders isotropic phase proporties as suitable infinity. The elastic properties of the matrix are those of the
approximation for the purpose of polycrystal property homogenized material

123
J Mater Sci (2007) 42:88248837 8829

reduces to the classical Eshelby-type inclusion problem nin r a . Their use for estimating the traction forces
with a perfect, rigid interface [17]. Then, consideration of at the interfaces within the polycrystalline RVE Vpoly yields
only one bulk phase (the crystals) implies that the overall the corresponding macroscopic stress Rpoly according to
bulk modulus kpoly is identical to the crystal bulk modulus (6) as
k C,
Z
1
kpoly  kC 13 Rpoly an  r
nr adS 18
VC oVC

For determination of lpoly, a purely deviatoric defor- The solution for the displacements at r = a+ turns out to
mation, E0 E0 e1  e1  e3  e3 , is imposed (see Fig. 3 be, according to (12)2 and (3), nr a ni n=2; a
for the Cartesian base frame e1 ; e2 ; e3 ). The mathematical suitable estimate for the mean displacement n
at the crystal
form of the displacement field in the exterior region, r > a interface I ij . Use of this quantity in (4) yields the corre-
(the homogenized material), nex , is established in the line sponding macroscopic strains Epoly in the form
of [27], and reads in spherical coordinates (see Fig. 3 for
Z
Eulerian angles / and h) as 1 s
Epoly n a  ndS 19
VC oVC ex
nex;r Bex 5  4mpoly Cex
Aex r 3 4 cos2 / sin2 h  cos2 h
E0 r 1  2mpoly r 2 Shear components Rpoly;12 and Epoly;12 of macroscopic
nex;h 1 Bex Cex stresses (18) and strains (19), together with (14)(17) and
Aex r  2 4 2 2 sin 2h1 cos2 /
E0 2 r r (50)(54), give access to lpoly , via lpoly R12 =2E12 ,
nex;/ 1 Bex Cex yielding (after elimination of E0) the following expression,
 Aex r  2 4 2 2 sin h sin 2/
E0 2 r r 2 !1 31
14 lC 5j lC 6kC 17lC
1 34 5 20
where mpoly is the Poissons ratio of the polycrystal with lpoly 2 8lpoly 57kC 4lC
weak interfaces,
with the dimensionless quantity j Kt0 a=lC . j ! 1
3kpoly  2lpoly relates to a rigid interface. The higher the rigidity j of
mpoly 15 the interface, the higher the overall polycrystal shear
6kpoly 2lpoly
modulus (Fig. 4), for different (dimensionless)
compressibilities v lC =kC of the single crystals.
The boundary condition in (12)4 directly implies Aex = 1,
Thereby, crystal incompressibility (v ! 0) guarantees
while the constants Bex and Cex will follow from interface
finite overall shear stiffness even for an interface with
conditions.
vanishing stiffness (j 0), while a polycrystal built up of
Inside the inclusion (r < a, the solid crystal phase), the
crystals with zero bulk modulus (v ! 1) and connected
displacement field nin reads as
through zero-stiffness interfaces (j = 0) does not exhibit
nin;r
Ain rBin r 3 cos2 /sin2 hcos2 h
E0
nin;h 1 11lC 15kC Bin r 3
Ain r sin2h1cos2 / 16 1
E0 2 33kC 2lC
nin;/ 1 11lC 15kC Bin r 3 0.8
 Ain r sinhsin2/
E0 2 33kC 2lC
0.6
The four remaining constants Bex ; Cex ; Ain and Bin are
determined by enforcing equilibrium of forces at the 0.4
interface r = a:
0.2
T rin
n rex
n K 0
n 17

together with constitutive laws (12)1, (12)2 and (12)3, see 0 6

2 4 8 10
Appendix. This solution for the displacement fields nin
and nex gives access to the traction forces at the Fig. 4 Homogenized shear modulus lpoly of polycrystal, as function
of dimensionless quantity j Kt0 a=lC (interfacial rigidity), for
interfaces Tr a r
nr a K 0
nex r a  different crystal compressibilities v lC =kC , Eq. 20

123
8830 J Mater Sci (2007) 42:88248837

any shear stiffness (Fig. 4), but still the bulk stiffness of However, use of the average tangential traction force hTt i
the single crystals according to (13). In case of an in failure criterion (8) is problematic since force peaks ini-
incompressible solid (kC ! 1; v lC =kC ! 0), it tializing failure may be cancelled out in the averaging process.
p
follows from (13) that kpoly ! 1, and (20) reduces to As a remedy, we use the second-order moment hTt2 i (also
called quadratic average) as a characteristic or effective value

 Tt x, in the line of [13, 14, 35]. The relation between
for
lpoly 2 lpoly p
485 j 114 9j  57j 0 21 hTt2 i and Rpoly is established through energy consider-
lC lC
ations: The energy stored in the RVE Vpoly can be expressed
through the global macroscopic energy density W as
Upscaled failure properties of polycrystal with weak 1
interfaces Vpoly W Vpoly Rpoly : Epoly
2
1
In order to determine the effective failure properties Vpoly Epoly : Cpoly : Epoly
2

resulting from local failure characteristics (8) and from the 1 2 2
interactions between interfaces and bulk single crystals, we Vpoly kpoly Epoly;v 2lpoly Epoly;d 24
2
are left with relating the local interface forces Tx 2 I to
the macroscopic stresses Rpoly , see (5). The tangential and with macroscopic volumetric strain Epoly;v tr Epoly
normal traction forces, Tt and Tn, occuring in the interface and equivalent deviatoric strain Epoly;d
p
failure criterion (8), are non-homogeneously distributed 1=2 Epoly;d : Epoly;d ; Epoly;d Epoly  1=3Epoly;v 1.
across the interfaces. Failure will occur where relatively In order to express W from a microstructural viewpoint,
high tangential traction forces encounter a relative low we consider the local constitutive behavior of the interface
resistance due to relatively low normal traction forces. (Eq. 9) and of the bulk phase (Eq. 10). The corresponding
Instead of trying to model the actual force fields across the macroscopic elastic energy stored in the RVE reads as
interfaces, we estimate the effect of the actual force dis- Z Z
tribution through so-called effective traction forces, as it is 1 1
Vpoly W r : edV T
ndS
commonly done for stress, strain, or force fields in the 2 Vpoly 2 I
Z Z
context of continuum micromechanics [14, 47]. In this line, 1 1
e : CC : edV n
K
ndS 25
we represent the failure-inducing interplay between mod- 2 Vpoly 2 I
erate normal traction forces and tangential traction force
R
peaks by means of two different effective measures for the In order to extract hTt2 i A1C I Tt2 dS from (25), varia-
normal and the tangential traction forces, respectively: (i) tions of W with varying Kt (holding merely Epoly fixed) are
first-order moments of normal forces, and (ii) second-order studied,
moments of tangential forces.
Z Z
The first-order moment of the normal traction forces, oW oe on
Vpoly : rdV
TdS
hTn i, is related to the macroscopic mean stress Rpoly;m oKt Vpoly oKt I oKt
through Z Z
1 oe
n
1  n  n
ndS : rdV
1 2 I Vpoly oK t
Rpoly;m trRpoly Z Z
3
Z o 1
 n  ndI : rdV nt 2 dS
1 3 Vpoly oK t 2 I
tr nx  rx
nxdS
3 AC oVC 26
Z Z
1 1
rrr xdS Tn xdS where T r
n was considered and
AC oVC AC oVC R where dI Ris the Dirac
distribution of support I ; V dI f dV I f dS. For
hTn i 22 transformation of (26), we extend Hills lemma [28] to
(22) establishes a first link between the macroscopic the case of displacement discontinuities at the interfaces
stress Rpoly and the interface tractions Tx: We use this [14]. Considering (5) and the format (2) for the
average (or first-order moment) of normal traction forces as macroscopic strains Epoly , (26) can be transformed to
to estimate the average interface resistance Ttcr in (8), Z
oW o
according to Vpoly e n  ndI : rdV
oKt Vpoly oKt
Z Z
1 oEpoly 1
Ttcr ah  hTn i 23 nt 2 dS : Rpoly n 2 dS 27
2 I oKt 2 I t

123
J Mater Sci (2007) 42:88248837 8831

Fixed macroscopic strains Epoly according to (1) imply

oEpoly =oKt 0, so that (27) becomes 0.6

Z 0.5
oW 1 ID E
Vpoly nt 2 dS nt 2 28
oKt 2 I 2 0.4

0.3
Identification of (28) with the derivation of the mac-
roscopic expression for the energy density (24) with 0.2
respect to Kt yields
0.1

I D E ok
poly 2 olpoly 2
nt 2 Epoly;v 4 E 29 0
10 20 30 40 50
Vpoly oKt oKt poly;d
  Fig. 5 Concentration factor BTt relating macroscopic deviatoric
When considering hTt2 i Kt2 n2t  according to (9), stress on polycrystal to effective tangential traction in intercrystalline
okpoly =oKt 0 according to (13), and Rpoly;d interfaces, as function of dimensionless quantity j Kt0 a=lC
2lpoly Epoly;d , (29) reduces to (interfacial rigidity), for different crystal compressibilities
v lC =kC , Eq. 32
!
I o 1 increases the peaks of tangential traction p force,
i.e., the
hTt2 i  Kt2 R2poly;d 30 proportionality factor BTt , again bounded by 2=5 (Fig. 5),
Vpoly oKt lpoly
r
where Rpoly;d is the equivalent deviatoric stress of the 2
lim BTt v 34
macroscopic second-order stress tensor Rpoly , j!1 5
r Use of the micro traction-macro stress relationships (22)
1
Rpoly;d Rpoly;d : Rpoly;d and (32) in the local interface criterion (8) yields a mac-
2
roscopic polycrystal-specific brittle-failure criterion in the
with Rpoly;d Rpoly  Rpoly;m 1; 31
form
1
and Rpoly;m trRpoly
3 BTt Rpoly;d  ah  Rpoly;m 35

Combination of (30) with I =Vpoly 3=2a and with (35) expresses that Coulomb-type brittle failure (8) in the
j Kt0 a=lC yields interfaces between spherical crystals inside the RVE results
q in DruckerPrager-type (brittle) failure properties at the
hTt2 i BTt Rpoly;d scale of the polycrystal.
v
!
u
l u 1 o lC Micromechanics of porous material with polycrystalline
with BTt v C ; j t j2 32 skeleton
kC 3 oj lpoly

We consider an RVE VPORO (Figs. 2b, 9) of a porous

Remarkably, the second-order moment p of
tangential
material (with porosity u) where the contiguous solid
tractions over all interfaces within the RVE, hTt2 i, is pro-
phase [volume VS, VS VPORO 1  u is a polycrystal
portional to the macroscopic equivalent deviatoric stress
with weak interfaces according to Section Microme-
Rpoly;d , expressed by the proportionality factor BTt . The more
chanics of polycrystal with weak interfaces. The Mori
compressible the solid crystal (the larger v lC =kC ), the
Tanaka homogenization scheme has been proven as suit-
higher the tangential traction peaks in the intercrystalline
able tool to upscale the elastic properties of the solid phase
interface, generated by an equivalent deviatoric macro-
[kpoly and lpoly defined through (13), (20), (21)] to the
scopic stress Rpoly;d . However, pthe
corresponding
concen-
stiffness of such a porous material, see e.g. [9, 11],
tration factor BTt is bounded by 2=5 (Fig. 5),
r  1
2 CPORO 1  uCpoly : 1  uI uI  S1
lim BTt j 33
v!1 5 36

On the other hand, for any constant crystal compress- with the Eshelby tensor S for spherical inclusions reading
ibility v, stiffening the interface (enlarging j Kt0 a=lC ) also as [17]

123
8832 J Mater Sci (2007) 42:88248837

"
2 #
3kpoly 6kpoly 2lpoly 3u a
S J K 37  R2PORO;m
3kpoly 4lpoly 53kpoly 4lpoly 4 B Tt
" #
so that 2u23  50mpoly 35m2poly
1 R2PORO;d
7 5mpoly 2
4kpoly lpoly 1  u
2
kPORO 38 a
3kpoly u 4lpoly 2 h1  uRPORO;m
B Tt

2
1  u9kpoly 8lpoly a
lPORO lpoly 39 h2 1  u2 44
9kpoly 1 23 u 8lpoly 1 32 u B Tt

We consider brittle failure of the overall porous medium with mpoly mpoly kpoly ; lpoly according to (15),
if the polycrystal failure criterion (35) is reached in highly lpoly lpoly kC ; lC ; j according to (20), and
stressed regions of the polycrystalline matrix. The corre- BTt BTt v lkCC ; j according to (32).
sponding (micro-) heterogeneity within the solid matrix The elastic stress domain of the porous medium the
has recently been shown [13] to be reasonably considerable matrix of which is a polycrystal with brittle interfaces in-
through so-called (homogeneous) effective (micro-) creases with decreasing crystal compressibility v (Fig. 6).
stresses, such as the square root of the spatial average over For the incompressible limit case, v ! 0, (44) reduces to
"
2 #

the solid material phase, of the squares of equivalent
3u a 2
deviatoric (micro-) stresses,  R2PORO;m 1 u R2PORO;d
4 B Tt 3

2
s
Z a
q 2 h1  uRPORO;m
1 1 B Tt
hr2d iS rd x : rd xdV 40
VS V S 2
2
a
h2 1  u2 45
1 B Tt
with rd x rx  trrx1 41
3 For a crystal compressibility of hydroxyapatite,
v 0:54 (see also Section Application to hydroxyapatite
The effective deviatoric stress (40), used to approximate biomaterials), the elastic domain increases with decreas-
Rpoly;d in (35), is accessible through energy considerations ing interfacial rigidity (Fig. 7) and with increasing friction
similar to those of (24) to (30), and result to be ([9, p. 132, angle a (Fig. 8). High interfacial rigidities j or low friction
13]) angles a result in closed elastic domains, indicating pos-
sible failure of the porous material even under hydrostatic
"
o 1 stress states R 1Rm , while low interfacial rigidities j or
R2poly;d hr2d iS  R2 high friction angles a are related to open elastic domains,
olpoly kPORO PORO;m
# related to infinite resistance of the porous material, as long
o 1 l2poly
 R2PORO;d 42
olpoly lPORO 1  u
0.5

In analogy to (23), the effective mean stress level in the 0.4

solid matrix is chosen as the stress average over the solid
phase, 0.3

Z
1 1 0.2
Rpoly;m hrm iS trrxdV
VS V S 3
RPORO;m 0.1
43
1u
0
2 1.5 1 0.5 0 0.5
Use of Eqs. 43 and 42, together with (38)(41), (13), and
(20), in (35) yields a failure criterion at the scale of the Fig. 6 Elastic limits of a porous material the matrix of which is a
polycrystal with brittle interfaces, for different crystal compressibil-
porous material with polycrystalline interfaces in the solid ities v lC =kC (Eq. 44): u 0:5; a 0:3; j ! 1. Uniaxial load
phase, path indicated (thin solid line)

123
J Mater Sci (2007) 42:88248837 8833

Table 2 Experimental data: Compressive strength fc as function of

3 porosity u, for artificial hydroxyapatite produced through different
synthesis routes
2.5
Peelen et al. [42] Akao et al. [1] Martin and Brown [38]
2
u (%) fc (MPa) u (%) fc (MPa) u (%) fc (MPa)
1.5
36 160 2.8 509 27 172.5a
1 48 114 3.9 465 39 119a
60 69 9.1 415
0.5 65 45 19.4 308
0 70 30
2 1.5 1 0.5 0 0.5 a
Mean value calculated from three experiments
Fig. 7 Elastic limits of a porous material the matrix of which is a
polycrystal with brittle interfaces, for different dimensionless section, to which extent the model developed before can
quantities j Kt0 a=lC (interfacial rigidity) (Eq. 44): u 0:5;
serve the purpose of the aforementioned prediction.
a 0:3; v 0:54. Uniaxial load path indicated (thin solid line)

Materials processing and uniaxial mechanical testing

3
We here consider the following artificially produced HA
2.5 materials: Peelen et al. [42] controlled the porosity of HA
by a variation of the sintering temperature (1,100
2 1,400 C, Table 2). Compacted commercially available
1.5 powders were used to produce HA with porosities between
36% and 70%. Cylindrical samples (diameter: 1 cm,
1 length: 11.5 cm) were tested in compression (Table 2).
Akao et al. [1] precipitated HA powder and sintered it at
0.5
different temperatures (1,1501,300 C). Porosities ranged
0 from 3% to 19% (Table 2). Compression tests were
2 1.5 1 0.5 0 0.5
performed on bars with dimensions of 5 5 10 cm3
Fig. 8 Elastic limits of a porous material the matrix of which is a (Table 2).
polycrystal with brittle interfaces, for different friction angles a (Eq. Martin and Brown [38] prepared calcium-deficient HA
44): u 0:5; j ! 1; v 0:54. Uniaxial load path indicated (thin formed in aqueous solutions at physiological temperatures.
solid line)
The authors realized two different liquid-to-solid weight
ratios, resulting in porosities of 27% and 39%, respectively
as the macroscopic stress state R contains a certain (Table 2). Cylindrical samples with diameter of ~6 mm
hydrostatic amount (Figs. 7 and 8). were tested in compression (Table 2).

Micromechanical representation of hydroxyapatite

Application to hydroxyapatite biomaterials biomaterials

Porous hydroxyapatite (HA) biomaterials are widely used
for replacement of hard tissue defects, because of their
chemical composition, microstructure and Youngs mod-
ulus being similar to the bone mineral, called carbonated or
calcium-deficient hydroxyapatite (CDHA) [26, 36, 46]. If
porous scaffolds are used as bone replacement material in
highly loaded anatomical locations, reliability of their
mechanical properties is particularly important for the
performance of the implants. Therefore, the prediction of
strength of HA biomaterials from their microstructure and
porosity is of particular interest. To the knowledge of the
authors, corresponding micromechanical approaches are
extremely rare or inexistent, so that we check in this
In the hierarchical organization of synthetic hydroxyapatite
ceramics, we identify two different scales which will be
considered in the framework of a two-step homogenization
scheme. The first homogenization step refers to an obser-
vation scale of several hundreds of microns where
hydroxyapatite crystals are separated by boundaries or
interfaces (Fig. 9a). The latter will be shown to be a po-
tential nucleus for failure of the material. The corre-
sponding homogenized material is called hydroxyapatite
polycrystal with interfaces. At the microstructural scale
with a characteristic length of some millimeters (Fig. 9b),
pores are embedded in a matrix which is made up of the
material which was homogenized in the first upscaling step.

123
8834 J Mater Sci (2007) 42:88248837

We use the two-membered evolution strategy [24, 45],

closely related to the ideas of Darwins evolution theory.
The components of a three-dimensional vector of estima-
tions for a, h and j; a; h; jparent ; representing the parent,
are slightly varied by help of a random number generator
(representing mutations), resulting in a vector
hydroxyapatite inter-crystalline
crystals interfaces a; h; jchild , representing the child,
(a) Hydroxyapatite polycrystal with weak interfaces a; h; jchild a; h; jparent
N raparent ; N rhparent ; N rjparent 47

N denotes a number produced by a standardized normally

distributed random number generator standardly available
in MATLAB [31]. r stands for a scattering factor which
will be dealt with later on.
If the child fits better in its environment than the
polycrystal (matrix) micropores
parent, i.e., if
(b) porous material with polycrystalline solid matrix
Ga; h; jchild \Ga; h; jparent  48
Fig. 9 Micromechanical representation of a porous hydroxyapatite
polycrystal by means of a two-step homogenization procedure. (a)
Hydroxyapatite polycrystal with weak interfaces and (b) porus see (46), vector a; h; jchild will be further varied, i.e., it
material with polycrystalline solid matrix then becomes the parent for the next generation. If not, the
original parent undergoes new mutations.
Elastic properties of single crystals of hydroxyapatite Based on the number of successes of the evolution,
i.e., the number of cases for which Eq. 48 holds, the
An ultrasonic interferometer technique delivers typical scattering factor r is changed: If the total number of suc-
values for bulk and shear moduli, kC kHA 82:6 GPa cesses within the last 10 mutations exceeds a certain
and lC lHA 44:9 GPa [34]. threshold (typically 4), r is enlarged, otherwise it is
reduced.
Biomaterial-independent properties of interfaces
If the difference between Ga; h; jparent  and
between hydroxyapatite crystals, a; h; jback-analysis
Ga; h; jchild  lies within a prescribed tolerance over a
certain number of mutations, the optimum aopt ; hopt ;
The expression for macroscopic admissible stress states
jopt a; h; jparent a; h; jchild has been reached.
(44) contains three material properties which are difficult to
Applying this procedure, (46)(48), to (44) and using
be directly accessed, namely the friction angle a, the
the experimental data from Table 2 yields, depending on
cohesion h, and the rigidity j of the interfaces. Therefore,
the start values of the optimization procedure, a set of
these phase properties will be determined by means of an
solution vectors aopt ; hopt ; jopt which are equal in terms of
optimization procedure providing the closest match of
the highly satisfactory correlation coefficient r 2 0:97
model predictions to experimentally determined uniaxial
between the respective model predictions and the corre-
compressive strength data of hydroxyapatite biomaterials,
sponding experimental data for uniaxial compressive
given in Table 2 [1, 38, 42].
strength (see Fig. 10). To give an example, aopt ; hopt ;
The sum of squares of relative errors between predicted
jopt 0:6750; 17:2397; 0:9119 and (0.9345, 17.7664,
strength and experimental strength values is minimized,
6.4160) (h has the dimension (MPa)) are two of these
pred
!
exp 2 solution vectors. For all calculated optimal solution
X fc;i  fc;i
Ga; h; j exp !0 vectors, we find a constant ratio a0 a=BTt 1:61 (see
fc;i 46
i Eqs. 20 and 32), implying a relationship between a and j,
) aopt ; hopt ; jopt depicted in Fig. 11.
Clearly, it would be interesting to cross-check these
pred
where fc;i denotes predicted uniaxial compressive strength interface failure parameters derived from our inverse
pred
values obtained pfrom
Eq. 44 with RPORO;m fc;i =3, method with other direct tests. Deplorably, an extensive
pred
RPORO;d fc;i = 3, together with Eqs. 13, 20, and 32, for literature check could not provide any direct in situ mea-
exp
porosity values ui according to Table 2. fc;i is the corre- surements of stresses and failure mechanisms at the inter-
sponding ith experimental strength value, see Table 2. face micro level. The only additional experimental

123
J Mater Sci (2007) 42:88248837 8835

3u 2
 a02 R2PORO;m 1 uR2PORO;d
700 4 3
Peelen et al. 1978
600 Akao et al. 1981 2a02 h1  uRPORO;m a02 h2 1  u2 49
Martin and Brown 1995
micromechanical model
500 with a0 a=BTt j and h as only two parameters being left
for an optimization procedure to match the experimental
400
data of Fig. 10 and Table 2. This procedure delivers a
300 cohesion hopt 16:51 MPa (close to the values obtained for
the brittle case in Section Biomaterial-independent prop-
200
erties of interfaces between hydroxyapatite crystals,
100 a; h; jback-analysis) and ratio a0opt 1:61 which is
quasi-identical to the one obtained for the brittle case
0 0.2 0.4 0.6 0.8 1 (Section Biomaterial-independent properties of interfaces
between hydroxyapatite crystals, a; h; jback-analysis),
Fig. 10 Uniaxial compressive strength fc of porous hydroxyapatite
biomaterial as function of porosity u: model prediction accordingpto
implying an aj-relationship quasi-identical to that of

Eq. 44 or Eq. 49, evaluated with RPORO;m fc =3; RPORO;d fc = 3, Fig. 11. This means that the failure of porous hydroxyapa-
compared to experimental data (Table 2). Correlation coefficient tite biomaterials can be equally well represented by a brittle
r2 = 0.97 elastic-limit-type micromechanics model and a ductile one
related to limit analysis. In this context, it is very interesting
to note that the ductile criterion (49) is even identical to the
1
elastic domain for incompressible solid matrices, Eq. 45.
Accordingly, one might argue that the nature of the
0.8
heterogeneity of the stresses in the solid matrix (considered
herein by quadratic averages) is far more important for the
0.6 overall failure of the material than the precise mode of
local interface failure (brittle or ductile). However, as re-
0.4 gards hydroxyapatite biomaterials, experiments [7, 38, 43]
strongly support brittle failure: A comprehensive mechan-
0.2 ical formulation for its possible origin, namely breaking of
weak interfaces between hydroxyapatite crystals, was the
0 20 40 60 80 100 main focus of the present paper.
Fig. 11 Friction angleinterface rigidity relationship a(j) suitable for
representation of strength of hydroxyapatite biomaterials (Fig. 10)

evidence are scanning electron micrographs (Fig. 2 in Ref. Appendix

[8], Figs. 57 in Ref. [38]): These images, however, clearly
show sharp, rough failure surfaces, coinciding with the
boundaries of single, micrometer-sized grains. This, to-
gether with the sharp stress drops in corresponding
(macroscopic) stressstrain diagrams indicating brittle
overall failure, strongly suggests brittle failure of the
crystal interfaces, as we have modeled herein.
Brittle versus ductile failure of solid matrix in porous
medium
From a purely mathematical standpoint, it is interesting to
compare the elastic domain (44) to the yield surface of a
porous medium, related to failure of a ductile (not a brittle)
solid matrix obeying DruckerPrager criterion (35). This
yield surface can be obtained through non-linear homoge-
nization based on effective quantities (42) and (43), as
detailed in [9, 11],
Solution of matrix-inclusion problem with compliant
interface (generalized Eshelby problem, Fig. 3)
Solution of Eqs. 1214, and 16 for the constants
Bex ; Cex ; Ain , and Bin yields them as:
Bex a5 176l3poly l2C 24l2poly l3C
 12l3C akpoly Kt  171l2C akpoly Kt kC 240l3poly kC lC
136l3poly lC aKt 48l3poly aKt kC  132al2poly Kt l2C
528l2poly kpoly l2C 9al2poly Kt kC lC
720l2poly kpoly kC lC 342l2poly l2C kC
144l2poly akpoly Kt kC 408l2poly akpoly Kt lC
27aKt kC kpoly lpoly lC  396aKt l2C kpoly lpoly
 57lpoly l2C aKt kC  4lpoly l3C aKt =N 50

123
8836
Cex 5a3 48l2poly aKt kC 240l2poly kC lC
136l2poly aKt lC 176l2poly l2C  8lpoly l3C
9lpoly aKt kC lC  114lpoly l2C kC  132lpoly aKt l2C
 57l2C aKt kC  4l3C aKt lpoly =N 51
Ain 5544l2poly aKt lC 192l2poly aKt kC 320l2poly l2C
1536l2poly kC lC 16lpoly aKt l2C 228lpoly aKt kC lC
408akpoly lpoly Kt lC 144akpoly lpoly Kt kC
240kpoly lpoly l2C 1152kpoly lpoly kC lC
12aKt l2C kpoly 171aKt kC kpoly lC lpoly =N 52
Bin 240l2poly lC 8lpoly lC 6kpoly lC
 12lpoly kC  9kpoly kC =a2 N 53
N 1408l3poly l2C 192l2poly l3C 24l3C akpoly Kt
342l2C akpoly Kt kC 1920l3poly kC lC
1088l3poly lC aKt 384l3poly aKt kC
1664al2poly Kt l2C 1584l2poly kpoly l2C
1032al2poly Kt kC lC 2160l2poly kpoly kC lC
2736l2poly l2C kC 432l2poly akpoly Kt kC
1224l2poly akpoly Kt lC 801aKt kC kpoly lpoly lC
852aKt l2C kpoly lpoly 1710lpoly l2C kpoly kC
120lpoly l3C kpoly 684lpoly l2C aKt kC
48lpoly l3C aKt 54
They define the displacement fields (14) and (16), which
give access to strains e Os n, stresses R (via (12)1, and
(12)3 respectively), mean interface displacements
n and
interface tractions t (via (12)2).
References
1. Akao M, Aoki H, Kato K (1981) J Mater Sci 16:809
2. Barthelat F, Tang H, Zavattieri PD, Li C-M, Espinosa HD (2007)
J Mech Phys Solids 55(2):306
3. Barthelemy J-F, Dormieux L (2003) Comptes Rendus Mecanique
331:271
4. Barthelemy J-F, Dormieux L (2004) Int J Numer Anal Methods
Geomech 28(78):565
5. Bernard O, Ulm FJ, Lemarchand E (2003) Cement Concr Res
33(9):1293
6. Bernaud D, Deude V, Dormieux L, Maghous S, Schmitt DP
(2002) Int J Numer Anal Methods Geomech 26:845
7. Chu T-MG, Orton DG, Hollister SJ, Feinberg SE, Halloran JW
(2002) Biomaterials 23:1283
8. de With G, van Dijk HJA, Hattu N, Prijs K (1981) J Mater Sci
16:1592
9. Dormieux L (2005) In: Dormieux and Ulm [16], p 109
123
J Mater Sci (2007) 42:88248837
10. Dormieux L, Barthelemy J-F, Maghous S (2006) Comptes Ren-
dus Mecanique 334:111, In French
11. Dormieux L, Kondo D, Ulm F-J (2006) Microporomechanics.
Wiley
12. Dormieux L, Maghous S (2000) CR Acad Sci IIb Math Mech
328(8):593, In French
13. Dormieux L, Molinari A, Kondo D (2002) J Mech Phys Solids
50:2203
14. Dormieux L, Sanahuja J, Maalej Y (2007) Comptes Rendus
Mecanique 335:25, In French
15. Dormieux L, Sanahuja J, Maghous S (2006) Comptes Rendus
Mecanique 334:19
16. Dormieux L, Ulm F-J (eds) (2005) CISM courses and lectures,
vol 480. Applied micromechanics of porous media. Springer,
Wien
17. Eshelby JD (1957) Proc Roy Soc Lond, Ser A 241:376
18. Fritsch A, Dormieux L, Hellmich Ch (2006) Comptes Rendus
Mecanique 334(3):151
19. Fritsch A, Hellmich Ch (2007) J Theor Biol 244(4):597
20. de Gennes P-G, Okumura K (2000) CR Acad Sci IV 1(2):257
21. Hashin Z (1991) J Appl Mech 58:444
22. Hellmich Ch, Barthelemy J-F, Dormieux L (2004) Eur J Mech A-
Solids 23:783
23. Hellmich Ch, Mang HA (2005) J Mater Civil Eng (ASCE)
17(3):246
24. Hellmich Ch, Ulm F-J (2002) J Eng Mech (ASCE) 128(8):898
25. Hellmich Ch, Ulm F-J, Dormieux L (2004) Biomech Model
Mechanobiol 2:219
26. Hench LL, Jones JR (eds) (2005) Biomaterials, artificial organs
and tissue engineering. Woodhead Publishing, Cambridge, UK
27. Herve E, Zaoui A (1993) Int J Eng Sci 31(1):1
28. Hill R (1963) J Mech Phys Solids 11:357
29. Hofstetter K, Hellmich Ch, Eberhardsteiner J (2005) Eur J Mech
A Solids 24(6):1030
30. Hofstetter K, Hellmich Ch, Eberhardsteiner J (2006) Comput
Assisted Mech Eng Sci 13(4):523
31. Hunt BR, Eipsman RL, Rosenberg JM (2001) A guide to
MATLAB for beginners and experienced users, 1st edn. Cam-
bridge University Press, Cambridge, UK
32. Katti DR, Katti KS (2001) J Mater Sci 36(6):1411
33. Katti DR, Katti KS, Sopp JM, Sarikaya M (2001) Comput Theor
Polym Sci 11(5):397404
34. Katz JK, Ukraincik K (1971) J Biomech 4:221
35. Kreher W (1990) J Mech Phys Solids 38(1):115
36. LeGeros RZ (2002) Clin Orthop Relat Res 395:81
37. Lemarchand E, Ulm F-J, Dormieux L (2002) J Eng Mech (ASCE)
128(8):876
38. Martin RI, Brown PW (1995) J Mater Sci: Mater Med 6:138
39. Okumura K (2002) Eur Phys JE Soft Matter 7(4):303
40. Okumura K (2003) Europhys Lett 63(5):701
41. Okumura K, de Gennes P-G (2001) The Eur Phys JE Soft
Matter 4(1):121
42. Peelen JGJ, Rejda BV, de Groot K (1978) Ceramurgia Int 4(2):71
43. Pramanik S, Agarwal AK, Rai KN, Garg A (2007) Ceram Int
33(3):419
44. Sanahuja J, Dormieux L (2005) Comptes Rendus Mecanique
333:818, In French
45. Schwefel H (1977) Numerische Optimierung von Computer-
Modellen mittels der Evolutionsstrategie. Birkhauser Basel,
Switzerland, In German
46. Suchanek W, Yoshimura M (1998) J Mater Res 13(1):94
47. Suquet P (ed) (1997) Continuum micromechanics. Springer,
Wien
48. Tai K, Ulm F-J, Ortiz Ch (2006) Nano Lett 6(11):2520
J Mater Sci (2007) 42:88248837 8837

49. Ulm F-J, Constantinides G, Heukamp FH (2004) Materials and 51. Zaoui A (1997) In: Suquet [47], p 291
structures 37(1):43 52. Zaoui A (2002) J Eng Mech (ASCE) 128(8):808
50. Ulm F-J, Delafargue A, Constantinides G (2005) In: Dormieux 53. Zhong Z, Meguid SA (1997) J Elasticity 46(2):91
and Ulm [16], p 207

123
J Mater Sci (2007) 42:88388843
DOI 10.1007/s10853-007-1846-9
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Stretching a stiff polymer in a tube
Jizeng Wang Huajian Gao
Received: 29 January 2007 / Accepted: 14 May 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract The present paper investigates the force-
extension behavior of a stiff polymer under stretching
inside a small tube. We develop a theory and perform
Brownian dynamic simulations based on a recently devel-
oped generalized bead-rod model (GBR) to show that the
force-extension relation of such a strongly confined poly-
mer chain can be described by that of an unconfined
polymer subject to an effective force which is derived
based on Odijks theory of a confined polymer chain.
Introduction
Physical properties of macromolecules and polymers in
complex environments are usually influenced by external
conditions such as geometrical confinements and applied
forces [1, 2]. Polymers in geometrical confinements that
are smaller than their unconfined molecular sizes are of
great significance in fields from polymeric liquid crystals
[3] to biological structures such as nucleosomes [4] and
viruses [5, 6]. In concentrated polymeric solutions and
melts, a single polymer can be subjected to large resis-
tances to motion from surrounding chains perpendicular to
its contour while moving along its contour with relative
ease. This situation is similar to a polymer confined in a
tube-like region [2]. Early discussions on the behaviors of a
single polymer chain confined in a tube can be found in de
Gennes work [1, 79], where simple scaling laws for the
J. Wang (&)
H. Gao
Division of Engineering, Brown University, 182 Hope Street,
Providence, RI 02912, USA
e-mail: Jizeng_Wang@brown.edu
123
confinement free energy and the longitudinal extension
were proposed as functions of the confinement cross-sec-
tion. This line of work was significantly elaborated for
polymer models such as the Gaussian and wormlike chains
[2]. For a stiff polymer chain1 tightly2 confined inside a
tube or nematic liquid, an interesting scaling law for the
confinement free energy of the chain was derived by Odijk
et al. [1016], with results later confirmed by numerical
simulations [1719].
Over the last decade, direct experimental investigations
have become possible for static and dynamic properties of
single polymer molecules subject to external forces, such
as the measurement of force-extension relation of DNAs in
electrolytes [2022]. Bustamante et al. [21] have shown
that the force-extension curve of a 97004 bps DNA mol-
ecule in 10 mM Na+ can be described by a wormlike chain
model provided that the contour length L and persistence
length p are treated as adjustable parameters to fit experi-
mental data. Marko and Siggia [23] interpolated a formula
for the extension of polymers which can almost exactly fit
the force-extension curve of a long DNA chain. For short
polymer chains, Kierfeld et al. [24] developed a discrete
harmonic chain model which has been numerically verified
[19] and shown to be valid even for a charged short chain if
an effective persistent length is introduced [25].
1
In an unconfined solution, a stiff polymer usually means that its
contour length L is smaller than its persistence length p. On the other
hand, a polymer confined in a small tube can be regarded as stiff as
long as the persistence length p of a polymer is larger than the tube
radius R, irrespective of the ratio L/p. We do not distinguish these two
situations in the current study.
2
Here, a tightly or strongly confined polymer implies that the
typical length scale of the confinement is smaller than the polymers
persistence length.
J Mater Sci (2007) 42:88388843 8839

and potential energies due to stretching and tube-confine-

r (s) u (s )
s
ment as [3, 23]
O
2R
Z  2
f f 1 L
o2 r
z H pkB T ds  f
rL
2 0 os2
Z L
Fig. 1 Coordinate system of a wormlike chain under stretching
inside a confining tube  r0 Vr? ds 5
0

in which p is the persistence length of the confined chain

In spite of the above progresses on confined polymers, and
the behavior of a stiff polymer chain under stretching
inside a small tube has not been fully investigated. In this 
0; kr? k\R
paper, we develop a theory and perform Brownian Vr? 6
1; otherwise
dynamics simulations based on a recently developed
generalized bead-rod model (GBR) [19, 25] to show that is the confinement potential per unit length due to the
the force-extension relation of a strongly confined poly- tube-wall [26, 27]. It has not been successful to obtain
mer can be described by that of an unconfined polymer analytical solutions to Eq. 5 under the hard wall boundary
subject to an effective tensile force which can be condition of Eq. 6 [2628]. However, solutions can often
analytically derived from Odijks theory on confined be obtained for a wormlike chain confined in a harmonic
polymers. potential [26, 27]
1
Vr? Nr2? : 7
Theory 2

Figure 1 shows the model system of a continuous wormlike Burkhard [26] has shown that the confinement free energy
chain confined inside a cylindrical tube with radius R. A set of a polymer with respect to the harmonic potential in
of Cartesian coordinates (x, y, z) are placed at the center of Eq. 7 has the same general form as that based on Eq. 6,
the tube so that the z-axis is along the tube axis. The chain with difference only in a dimensionless prefactor which
is stretched by a tensile force f f k where k is the unit depends on the particular potential. Following this
vector along the z-axis. The position vector along the arc approach, we will establish the basic form of the solution
length s of the chain is based on the harmonic potential in Eq. 7 and then deter-
mine a prefactor based on Odijks theory of a confined
r r? s zsk 1 polymer chain.
In Eq. 5, the potential energy with respect to the
where r? x; y is perpendicular to the z-axis, and external force can be rewritten as
r0 0: The derivatives
Z L
or? dz f
u? ; ujj k 2 f
rL  r0 fz fL  u2? ds 8
os ds 2 0

define the tangential vector. In the case of strong where we have used Eq. 3 and z(0) = 0. Inserting Eqs. 4, 7
confinement, the undulation of the chain due to thermal and 8 into Eq. 5 yields
fluctuation will be small so that ku? k\\1: The
Z L 
inextensibility condition of the wormlike chain 1 ou? 2
kuk kor=osk 1 and Eq. 2 lead to H  pkB T ds
2 0 os
Z Z Z s 2
f L 2 N L
dz 1 u? ds u? ndn ds 9
1  u2? Ou4? : 3 2 0 2 0 0
ds 2
Therefore, we have where we have dropped a constant term and used relation

 Z s
o2 r o2 r ? d 2 z o u? Ou2? k ou?
4
2 2k  : r? s u? ndn: 10
os2 os ds os os 0

The Hamiltonian of the confined wormlike chain under Following earlier studies [3, 23, 29], we introduce
stretching can be expressed as the summation of bending Fourier transform

123
8840 J Mater Sci (2007) 42:88388843

Z Z L
~? x eixs u? sds 11 1
u zL L  u2? ds; 20
2 0

to decouple Eq. 9 into normal modes as Eqs. 15, 19 and 20 give

Z 1 
H 1 2 N 1 f 1 1
px u~2 dx 12 1  \z > p 21
kB T 4p 1 kB T kB T x 2 ? 2 fe p=kB T
with average energy contributed by each mode where <z> is denoted as the average end-to-end distance of
  the wormlike chain along z-axis normalized by the contour
\Hx > 1 2 f N 1 length L, and
px u2? > :
\~ 13
kB T 2 kB T kB T x 2
According to the equipartition theorem, <Hx> is equal to c2 p
fe f kB T 22
kBT for two degrees of freedom, and Eq. 13 becomes k2
which can also be written in the form
2
u2?
\~ > : 14 p
4=3
px2 kBf T kBNT x12 fe p fp
c2 :
kB T kB T 2R
It follows that
Eq. 22 suggests that the behavior of a strongly confined
Z L polymer under stretching can be described by that of an
1
\u2? > ds unconfined polymer subject to the effective force equal to
L 0
Z fe. In other words, the tube confinement can be viewed as
1 1 2
f
dx an effective stretching force. As a consequence, the force-
2p 1 px k T kNT x12
2
B B extension relation of a long wormlike chain confined in a
1 tube can be obtained as
q
p 15
fp=kB T 2 Np3 =kB T
fe p 1 1
\z >  ; 23
In the case of f = 0, we use the following result from kB T 41  \z > 2 4
Odijks theory on confined polymers [3, 10, 11],
following Marko and Siggia [23]. It can be seen from
Eq. 23 that the normalized extension of the polymer chain
k
\u2? > 16 under combined actions of mechanical stretching and
cp
geometrical confinement only depends on the normalized
where k is the Odijk deflection length and c is a tensile force and the ratio p/R, with no dependence on the
numerically determined prefactor around 2.5 [1719]. For contour length of the chain (L>>p). In contrast, the force-
a confining tube of radius R, Odijk [10] has shown that extension relation of a discrete wormlike chain of finite
contour length L is [24]
k 4R2 p1=3 : 17 !1=2
fe p 1 B
Comparing Eqs. 15 and 16 gives 1
kB T 2B 1  \z > 2
 
1 k 1 1=2 p 1
q  1 B 1
p cp 18 2B L 1  \z >
2 Np3 =kB T  
1 p 3 1
 p  \z > 24
or 2 1B L 2 Up=L

where Ux 1  x xe1=x ; B b=2p2 ; and b is the

c4 p
N 4 kB T: 19 length of a unit bond in the discrete chain. For the discrete
4k chain, Eq. 24 shows that the normalized extension <z>
Noting that under stretch and tube-confinement is determined by the

123
J Mater Sci (2007) 42:88388843 8841

normalized tensile force, as well as p/R, p/L and b/p. Only with current position rnj under cylindrical confinement to
when L>>p and b<<p Eq. 24 can be reduced to Eq. 23. illustrate the algorithm. If the bead is located close enough
to the reflecting wall, e.g.,
q
Brownian dynamics simulations Snj  R  x2nj y2nj
p 27
 5DDt
Here we briefly describe Brownian dynamics simulations
that were performed to verify the simple force-extension the stochastic movement caused by the reflecting wall is
relations proposed in the previous section. The simulations given by
were based on the GBR model of a wormlike chain under xnj
strong confinements [19]. dxwall
nj  q
In the GBR model, a semiflexible polymer or filament is x2nj y2nj
described as N identical virtual beads of radius a connected    
Snj p Snj p
by N1 inextensible rods of length b with the unit  f1 p DDt f2 p DDU 28
DDt DDt
tangential vectors uj (juj j 1, j 1; 2; . . . ; N  1). The
contour length of the chain is L = (N1)b. The N virtual ynj
dywall
nj  q
beads with coordinates rj xj ; yj ; zj (j 1; 2; . . . ; N) are x2nj y2nj
introduced for modeling hydrodynamic interactions       29
between different chain segments. The Brownian dynamics Snj p Snj p
f1 p DDt f2 p DDU
of a discrete wormlike chain involves the collective motion DDt DDt
of N identical beads in solution. After the hard rod con-
straint is implemented via the linear constraint solver dzwall
nj 0 30
(LINCS) [30], the new position vector rn1 of the N beads
where j 1; 2; . . . ; N, Snj is the current relative radial
is determined from
position of the jthpbeadand functions f1, f2 are defined in
[19]. The value 5DDt used above was suggested by
rn1 I  Tn Bn rn
Peters et al. [31] in their treatment of a sticky wall. In the
Dt
Dn Fn nn Tn d 25 present case, the wall of the confining tube is assumed to
kB T influence the motion of a bead only if the distance between
where rn (3N vector) is the current position of the beads, the bead and the wall is smaller than this value. We
Fn is the collective vector of internal (inter-beads) and therefore set dxwall wall wall
nj dynj dznj 0 if the current
external forces, nn is the random force generated at each distance
p between the jth bead and the wall exceeds
p
time step from a Gaussian distribution with zero mean and 5DDt, i.e., Snj > 5DDt. For convenience, we can
variance equal to express the above stochastic displacements given by
Eqs. 2830 as a 3N vector v, which consists of
\nn nn0 > 2Dn Dtdnn0 : 26 vwall wall wall wall
nj dxnj ; dynj ; dznj , j 1; 2; . . . ; N. Including v
in the current position vector rn in Eq. 25, we obtain
Here, Dt is the time step and dnn0 is the Kronecker delta
symbol, I  Tn Bn is a projection matrix which sets the rn1 I  Tn Bn rn vwall
n
constraints and Dn is the translational diffusion matrix Dt
determined through hydrodynamic interactions between Dn Fn n Tn d 31
kB T
beads.
Special considerations are needed for numerical simu-
lations of spatially confined wormlike chains. Peters et al. Results and discussions
[31] introduced an efficient algorithm for the Brownian
dynamics of a particle near reflecting wall. In their method, For a confined wormlike chain subjected to a constant
the errors of discretization are kept on the order of O(Dt) to tensile force f in the z-direction on both ends of the chain,
handle the boundary conditions, while a naive treatment of the force vector Fn in Eq. 31 can be written as
identifying reflection processes by checking pboundary

crossings usually yields errors on the order of O Dt [31]. Fn Fbn Ft 32
Their approach has been adapted to polymer confinements
in the GBR model [19]. For the collective motion of many where Ft is the tensile force and Fbn is the effective force
beads in the bead-rod wormlike chain, we take the jth bead on the beads due to the bending rigidity of the chain.

123
8842
0.12 p=p0, R=10nm
p=2p0, R=20nm
p=2p0, R=10nm
0.1
p=p0, R=5nm
CWLC, p/R=5.3, c=2.5
0.08 DWLC, p/R=5.3, c=2.5
CWLC, p/R=10.6, c=2.7
1-<z>
DWLC, p/R=10.6, c=2.7
0.06
0.04
0.02
0 -2 0 1 2 3
10 10 10 10 10
fp/k T
B
Fig. 2 Comparisons of Brownian dynamics simulation results for the
relative extension of nanotube confined wormlike chains with
corresponding theoretical predictions based on the concept of an
effective force in Eq. 22. CWLC stands for the continuous
wormlike chain model and DWLC stands for the discrete wormlike
chain model. The theoretical predictions are shown as continuous
curves. In plotting the curves, we have taken the basic measure of
persistence length to be p0 53:248 nm , corresponding to that of a
DNA chain
Based on the GBR model, Brownian dynamics simula-
tions have been performed for wormlike chains in
nanotubes of different radius. In all simulations, the chains
are initially set in a straight configuration. Tube confine-
ments and constant tensile forces are then applied during
the chains relaxation. We record the normalized end-to-
end distance <z> of a chain along z-axis at each time
increment. Figure 2 compares the simulation results with
corresponding theoretical predictions based on the effec-
tive force concept in Eq. 22. The chains were simulated for
a total time of 10 ls with parameters b/p  0.19,
a 1:3 nm, T = 293 K, Dt 300 ps (in cases of
fp=kB T\100) or Dt 30 ps (in cases of fp=kB T > 100).
Each data point in Fig. 2 is obtained by averaging the
recorded values of <z> for 200 trajectories with different
random seeds. The abbreviation DWLC stands for the
discrete wormlike chain model in Eq. 24, and CWLC
stands for the continuous wormlike chain model in Eq. 23.
The simulations were conducted for chains within different
tubes under parameters p/R  5.3, p=R  10:6 and
R = 5 nm, 10 nm, 20 nm. It can be seen from Fig. 2 that
the simulation results are in good agreement with those
predicted by the discrete wormlike chain model when the
effective force expression is applied. Figure 3 illustrates
the dynamic evolution, averaged over 200 different tra-
jectories, of the relative extension of wormlike chains
confined in nanotubes of different radius over a total sim-
ulation time of 10 ls under parameters b/p = 0.19,
123
J Mater Sci (2007) 42:88388843
-1
10
~t1/2
~t
1-<z>
-2
10
p=p0, R=10nm, fp/kBT=8
p=1.5p0, R=15nm, fp/kBT=8
p=2p0, R=20 nm, fp/kBT=8
fp/kBT=31, R=Infinity
f=0, R=8nm
-3
4
10 -2 -1 0 1
10 10 10 10 10
t ( s)
Fig. 3 The dynamic evolution of the relative extension of a wormlike
chain under stretching by a force of different magnitudes inside
nanotubes of different radii, where the effective force is fixed as
31kB T=p and p0 53:248 nm corresponds to the persistence length of
a DNA chain
Dt 300 ps, a = 1.3nm and T = 293 K. The effective
force is fixed at fe p=kB T 31 for each trajectory, which
can be realized by setting p=R  5:3 and fp=kB T 8, or by
setting fp=kB T 31 for an unconfined chain, or by con-
fining a free chain in a tube. The results indicate that
equilibrium is reached within a few microseconds. It can be
observed from Fig. 3 that, under the same normalized
effective force fe p=kB T 31, the crossover times between
ballistic ( t) and diffusive ( t1=2 ) behaviors are different
for tubes with different radii. These results confirm that the
force-extension behavior of a wormlike chain confined in a
nanotube can be well described by the discrete wormlike
chain model of Eq. 24 incorporating the concept of effec-
tive force defined in Eq. 22.
Conclusion
Theories of confined polymers suggest that the effect of
geometrical confinements on the force-extension relation
of a polymer can be represented by an effective force
applied to an unconfined polymer chain. Here we have
theoretically investigated this problem and derived an
analytical expression of this effective force for a polymer
chain confined in a cylindrical tube based on Odijks theory
on confined polymers. We have also performed Brownian
dynamics simulations, based on a recently developed
generalized bead-rod model, to investigate the force-
extension behavior of a wormlike chain confined in a
nanotube, with results in excellent agreement with the
discrete wormlike chain model of Kierfeld et al. [24]
incorporating the effective force concept defined in Eq. 22.
J Mater Sci (2007) 42:88388843 8843

References 18. Bicout DJ, Burkhard TW (2001) J Phys A: Math Gen 34:5745
19. Wang J, Gao H (2005) J Chem Phys 123:084906
1. De Gennes PG (1979) Scaling concepts in polymer physics. 20. Smith S, Finzi L, Bustamante D (1992) Science 258:1122
Cornell University Press, Ithaca, NY 21. Bustamante C, Marko JF, Siggia ED, Smith S (1994) Science
2. Doi M, Edwards SF (1986) The theory of polymer dynamics. 265:1599
Oxford University Press, New York 22. Strick T, Allemand J, Bensimon D, Bensimon A, Croquette V
3. Odijk T (1986) Macromolecules 19:2313 (1996) Science 271:1835
4. Shiessel H (2003) J Phys Condens Matter 15:R699 23. Marko JF, Siggia ED (1995) Macromolecules 28:8759
5. Earshaw WC, Harrison SC (1977) Nature 268:598 24. Kierfeld J, Niampoly O, Sa-Yakanit V, Lipowsky R (2004) Eur
6. Purohit PK, Kondev J, Philips R (2003) Proc Natl Acad Sci USA Phys J E 14:17
100:3173 25. Wang J, Fan X, Gao H (2006) Mol Cell Biomech 3(1):13
7. De Gennes PG (1976) Macromolecules 9:587 26. Burkhardt TW (1995) J Phys A: Math Gen 28:L629
8. De Gennes PG (1976) Macromolecules 9:594 27. Burkhardt TW (1997) J Phys A: Math Gen 30:L167
9. De Gennes PG (1971) J Chem Phys 55:572 28. Jo K, Dhinhra DM, Odijk T, De Pablo JJ, Graham MD, Runn-
10. Odijk T (1983) Macromolecules 16:1340 heim R, Forrest D, Schwartz DC (2007) Proc Natl Acad Sci USA
11. Odijk T (1993) Macromolecules 26:6897 104:2673
12. Khokhlov AR, Semenov AN (1981) Physica A 108:546 29. Landau LD, Lifshitz EM (1958) Statistical physics. Addison-
13. Khokhlov AR, Semenov AN (1982) Physica A 112:605 Wesley, Reading, MA
14. Helfrich W, Harbich W (1985) Chem Scr 25:32 30. Hess B, Bekker H, Berendsen HJC, Fraaije JGEM (1997)
15. Helfrich W (1978) Z Naturforsch A 33:305 J Comput Chem 18:1463
16. Helfrich W, Servuss RM (1984) Nuovo Cimento D 3:137 31. Peters EAJF, Barenbrug TMAOM (2002) Phys Rev E 66:056701
17. Dijkstra M, Frenkel D, Lekkerkerker HNW (1993) Physica A
193:374

123
J Mater Sci (2007) 42:88448852
DOI 10.1007/s10853-007-1820-6

NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS

Atomistic-mesoscale coupled mechanical analysis of polymeric

nanofibers
V. U. Unnikrishnan G. U. Unnikrishnan
J. N. Reddy C. T. Lim

Received: 9 February 2007 / Accepted: 3 May 2007 / Published online: 14 July 2007

Springer Science+Business Media, LLC 2007

Abstract Theoretical analysis of Poly-(L)-Lactic Acid
(PLLA) nanofibers is a necessary step towards designing
novel biomedical applications. This paper aims to analyze
the mechanical properties of PLLA nanofibers so that
optimal scaffolds in tissue engineering applications can be
developed. We carry out analysis of PLLA nanofibers to
estimate the mechanical properties from basic building
blocks to the nanofibrous structures. A single PLLA
nanofiber is made up of ShishKebab like fibrils inter-
twined together and can contain both amorphous and
crystalline phases. The elastic modulus of the Lactic acid
monomeric formation in the crystalline phase is derived
using second-derivative of the strain energy using molec-
ular dynamics simulation. The mechanical property of the
ShishKebab fibril is derived by homogenization. The fiber
modulus is then obtained using the Northolt and van der
Houts continuous chain theory. One of the significant
contributions in this paper is the use of modified continu-
ous chain theory, where a combined multiscale approach is
used in the estimation of the mechanical properties of
V. U. Unnikrishnan
G. U. Unnikrishnan

J. N. Reddy
Advanced Computational Mechanics Laboratory,
Department of Mechanical Engineering, Texas A&M University,
College Station, TX 77843-3123, USA
J. N. Reddy (&)
Engineering Science Programme,
National University of Singapore, Singapore, Singapore
e-mail: jnreddy@tamu.edu
C. T. Lim (&)
Division of Bioengineering & Department of Mechanical
Engineering, National University of Singapore,
9 Engineering Drive 1, Singapore 117576, Singapore
e-mail: ctlim@nus.edu.sg
PLLA nanofibers. The theoretical results correlate well
with reported experimental data.
Introduction
Polymeric nanofibers are attractive materials for a wide
range of applications in the bio-medical, textile and other
emerging technologies. This is primarily due to their large
surface area to volume ratio and the unique features at the
nanometer scale [1]. Structures of fibrous polymers are
generally very flexible, and their conformation is easily
deformed against mechanical extension or induced motion
between its atoms. In any industrial application, the suit-
ability of a material and/or structure relies significantly on
their physical properties, especially their mechanical and
electrical properties. Whilst the mechanical design ensures
dimensional stability and structural integrity, the electrical
design aims to fulfill the functionality of the products.
In recent years, polymeric nanofibers have been devel-
oped for a variety of applications such as tissue engineering,
molecular filters, sensors and protective clothing [26]. For
example, polymeric nanofibers can be used to form nano-
fibrous scaffolds for tissue engineering application [7].
These polymeric scaffolds allow cells to proliferate and
grow into tissues with defined sizes and shapes for trans-
plantation purposes [8, 9]. An understanding of the struc-
tureproperty relationship is essential for the engineering
applications of polymeric nanofibers since they are affected
by the mechanical properties arising from the internal
molecular structures. Tremendous savings in cost can be
achieved if preliminary experimental designs can be eval-
uated theoretically to eliminate inferior designs and reduce

123
J Mater Sci (2007) 42:88448852
the number of experiments. The proposed theoretical work
in the analysis of nanofiber is primarily aimed at providing a
computational framework for the estimation of the
mechanical properties and to provide a strong connection
between experimental observations and theoretical analysis.
Fibers prepared from polymer solution or melt by con-
ventional methods (melt, dry and wet spinning) have
diameters in the range of 5500 mm [10]. Recently, there
has been increased interest in the fabrication of nanofibers
(with diameters in the range from tens to hundreds of
nanometers) using electrospinning [1, 10, 11] for mechan-
ical characterization studies. Using molecular dynamics
(MD) simulation, crystalline lactic acid monomer units are
equilibrated and thermostatted to the experimental condi-
tions by a series of NVE ensemble (Microcanonical
ensemble) and NVT ensemble (Canonical ensemble) anal-
ysis and subjected to isothermal strain conditions to obtain
the mechanical properties [12, 13]. To develop an optimal
scaffold for tissue engineering application, it is required to
manipulate the mechanical characteristics of the nanofi-
brous scaffolds. There has been numerous experimental
studies on the design of optimal scaffolds [8]. However,
very few theoretical studies exist in predicting the
mechanical properties and behavior of nanofibers under
external mechanical loads using multiscale simulation. This
paper aims to analyze the mechanical properties of PLLA
nanofibers via an atomistic-mesoscale stimulation method.
All molecular dynamics simulations were performed using
Cerius2 (version 4.6, Accelrys, Inc.) simulation package.
Analysis of the orientation process during uniaxial
drawing of a polymer has long been investigated in many
theoretical and experimental studies [7, 14, 15]. Based on
the deformation of cellulose fibers, analytical models were
developed for rodlets connected by crosslinks. These
models were modified with the various additions like cross-
linking with forces applied to the ends of the chains as well
as changes in material properties. This research lead to two
different formulations for the analysis of polymeric chains:
the rubber elasticity theory based on complex constitutive
relations and the orientation based mechanism for the
analysis of semi-crystalline polymers leading to the aggre-
gate model [16]. The fibrils in a nanofibrous material are
found to intertwine to form polymeric nanofibers. The fiber
modulus is obtained using the Northolt and van der Houts
continuous chain theory [14, 1720]. This is an enhance-
ment over conventional homogenization techniques,
because the effect of shear deformation of the fibrils is not
taken into consideration. The continuum chain formulation
used in this paper gives relationships between the macro-
scopic elastic constants and the orientation parameters
based on the spatial distribution of the nanofibrils.
The paper is organized as follows. Section (Atomistic
simulation) describes the atomistic simulation of crystalline
8845
lactic acid using MD simulation for the estimation of the
mechanical properties in the atomistic scale. Homogeniza-
tion and description of the ShishKebab model is discussed
in section (ShishKebab modelelectrospun nanofibers).
The multiscale transfer of quantities of interest from the
atomistic scale to the mesoscale by micromechanical
methods is discussed in sections (Micromechanical analy-
sis) and (Continuum volume averaging: micromechanical
method). In section (Continuous chain model of polymeric
fibers), the formulation of the continuum chain model
to scaleup the material properties is discussed. Section
(Results and discussion) combines the results from various
methods and finally the paper concludes with a summary in
section (Conclusion).
Atomistic simulation
The knowledge of structure and molecular motion in
polymers is essential to understand the properties of prac-
tical interest. Theoretical simulation of the physical pro-
cesses forms the first step in this work. The estimation of
the mechanical properties of the PLLA fibers needs to be
carried out. There are various methods of estimating the
physical properties of atomistic structures, molecular
dynamics (MD) simulation being one of them and is used
here. Molecular dynamics has been a very popular tool for
the determination of mechanical, thermal and other prop-
erties of interest in atomistic structures [2125]. The
starting point of a MD simulation is the non-relativistic
quantum mechanical time dependent Schrodinger equation.
The thermodynamic state characterized by the fixed num-
ber of atoms, volume and temperature called the canonical
ensemble [25] forms the basis of the MD simulation here.
The simulated system and the heat bath couple to form a
composite system. The conservation of the energy still
holds in the composite system but the total energy of the
simulated system fluctuates. The motion of the particles in
the system is governed by the Hamiltonian, which is a
function of the position and momentum of the particles [13,
25]. The Hamiltonian representing the total energy of an
isolated system is given as the sum of the potential and
kinetic energy terms and thermodynamic terms, as given by
P
Hr N ; pN 2m
1
p2i U r N
i 1
E pN ; r N Ek pN U r N
where Ur N is the potential energy from intermolecular
P 2
interactions as a function of the spatial ordinate r N ; 2m
1
pi
i
is the kinetic energy, which represents the momentum pi of
the particle i with mass mi and pi is a function of the
absolute temperature. The time derivative of the
Hamiltonian gives
123
8846 J Mater Sci (2007) 42:88448852


dH 1 X X oU oE
pi
p_ i
r_i 0 2 fi t Dt 8
dt m i i
ori ori tDt

and the spatial derivative of the Hamiltonian gives the The elastic constants can be obtained directly from the
equation of motion as variation of the potential for nanofibers [13, 23, 26]. The
total potential energy due to the strain which is the elastic
dH oU strain energy can be expanded as a Taylor series for small
3
dri ori displacements, with the initial position being represented
by the equillibrium position. The elastic moduli tensor can
The reliability of a MD simulation depends mainly on be written as
the type of potential functions used. The total potential of
2 3
the computational unit cell is given by the sum of valence
!
bond energies and the nonbonding interactions 1 X6 2
6 d U 1 dU a b c d 1 dU a c 7 7
Cabcd 6  a a a a d a a 7
X Xh i 2NXa j6i 4 drij2 rij drij ij ij ij ij
bd
rij drij ij ij 5
U tot VijB VijNB 4 |{z}
j j>i Aac 0

where VijB is the potential energy due to bonding and VijNB is
the potential energy due to nonbonding interactions. The
force of attraction and repulsion (Fab) experienced by each
molecule is obtained from the gradient of the potential field
oU tot
Fab  5
orab
This force is used in calculating the updated position of
the atoms and is carried out by the Velocity-Verlet time
integration scheme [13]. In this analysis, the time step is
chosen in such a way that the material reaches a metastable
state at a given ambient temperature. This is normally in
the range of femtoseconds (1015 s), and the local variation
of the velocity and the kinetic energy about a small
increment in time is very small. It has been shown that
without a sudden change in the ambient conditions, the
molecules vibrate about the mean energy position, which
ensures that there is no change in the inherent temperature
of the material. At the beginning of each time step of the
simulation, updated velocities vi (t) are calculated for each
particle by



Dt Dt fi t
vi t vi t  Dt 6
2 2 mi
where Dt is the time step in the MD simulation, fi t is the
total force acting on particle i at time t and mi is the mass of
the particle. The coordinates of the particles ri t are
updated from the velocities
Dt
ri t Dt ri t vi t Dt 7 where the crystal structure is analyzed under the influence
2
From the updated particle position, the interatomic for-
ces are computed from the first derivatives of the potential
energy field E with respect to the atomic coordinates ri
9
Here Cabcd is the elastic moduli; U Urij is the potential
energy as a function of the interatomic distance rij ; Aac is the
internal stress tensor, which at equilibrium is equal to zero;
Wa is the average volume of an atom; N is the number of
atoms; dab is the Dirac-Delta function; a; b; c; d are the
spatial dimensions and aaij is the undeformed value of rij in
the a -direction. The deformation distance is given by
uaij rij  aaij and it is related to the strain by ubij
aaij eab where eab are the components of the homogenous
infinitesimal strain tensor associated with atoms i and j.
The knowledge of structure and molecular motion in
polymers is essential to the understanding of mechanical
and thermal properties. The crystallization behavior of
PLLA shows that it is a semicrystalline polymer that
crystallizes from melt and from solution to form fibers [27].
Studies on crystal structure of lactide copolymers by var-
ious studies have shown that the unit cell of PLLA is a
pseudo-orthorhombic structure a 10:6 A; b 6:1 A;
and c 28:8 A, which is used here [2831]. X-ray dif-
fraction experiments and Nuclear Magnetic Resonance
(NMR) analysis for the estimation of the fibrous and crystal
structure of PLLA has shown that the crystalline structure
of PLLA differs slightly [31] from that used by Hoogsteen
et al. [28] and De Santis et al. [29]. However, these
differences are not high enough to cause a change in the
properties of the PLLA structure. MD analysis of crystal-
line PLLA is carried out with the crystal structure and the
entire computational model is equilibrated to the experi-
mental conditions (see Fig. 1). The minimum energy con-
dition is the starting point for the thermostating analysis,
of thermal energies. Isothermal strain conditions were
applied to the thermally equilibrated unit cell and the
elastic constants were obtained using second derivative
elastic constant analysis [13, 32].

123
J Mater Sci (2007) 42:88448852 8847

Fig. 2 ShishKebab model and the homogenized equivalent contin-

uum (EC) ShishKebab model
Fig. 1 Computational domain of the Crystalline PLLA Unit cell

ShishKebab modelelectrospun nanofibers constituent properties is an established method of bridging

the scales [13] and it forms the preliminary basis of mul-
Orientation and extension of molecules in a polymer melt tiscale modeling. Here, we deal with the volume averaging
affects the crystallization kinetics, structure and morphol- of the different elastic measures in the molecular level
ogy. In an entangled polymer, one of the most common drawing similarity between the macroscopic quantities,
crystallization formations is the ShishKebab structure connected by the equivalence of the strain energy due to
[3335]. The innermost portion of a ShishKebab structure deformation and the change in the potential energy in an
is a long and macroscopically smooth extended chain which isothermal mechanical straining process. For an elastic
is crystalline in nature, called a Shish. The Kebabs are composite material, the effective constitutive relations are
folded chain crystalline structures entangling the Shish. The given by the volume average of the stress and strain.
direction of growth of the Kebab is normal to the Shish [12]. Similarly, for each phase on the micro- and nanoscale, the
There are various views on the formation of the Shish constitutive relation can be given as
Kebab structure in a crystallization process; however, in
this study we are interested only in the experimentally hriktot C k heiktot 10
observed Shish structures in some of the very latest works h
iktot is the volume averaged state of phase k, including the
on crystalline PLLA nanofibers. ShishKebab structure can matrix, fiber and any interphase layers [13]. The volume
be found in many of the crystallization inducing processes averaging of the state variables are given by
like electro-spinning, melt spinning, etc. [11, 15, 36].
For the polymeric nanofiber, AFM imaging also reveals Z Z
  1   1
a ShishKebab structure [35, 36]. The elastic property r

ab EC
rab dv;
eab EC eab dv 11
V V
obtained from MD analysis is used in the homogenization X X
of the ShishKebab model. In this work, we are proposing    
the homogenization of the ShishKebab model assuming r

ab EC
C
eab EC
that the homogenized axial modulus of Shish is obtained
Similarly, for an N particle atomic ensemble
from the crystalline modulus using MD simulations and the
Kebab modulus is obtained from the average of the mod-
ulus of the RVE in all the directions (see Fig. 2). This   1X N   1X N

ab
r rab ;
eab eab 12
assumption is valid since the ShishKebab model consists N i1 N i1
of only crystalline formations.    

ab C
eab
r

Micromechanical analysis For a simple EC, the average stresses due to the atomic
ensemble is equal to the average stress due to volume
Direct application of the micromechanical methods for the averaging, establishing the relationship between the mate-
nanofiber raises several questions. Volume averaging of the rial constants derived from the MD simulation and volume

123
8848 J Mater Sci (2007) 42:88448852

averaging of the state variables for use in the structural

homogenization and micromechanical techniques. Struc-
tural models have been developed for foams and cellular
materials based on a unit cell. Though these models are
based on the information that the porosity of the material is
above 70%, this method can be used in the present analysis,
as there is a large expected range of porosity (F) of the
polymeric fiber based on experimental studies [13]. The
effective modulus (E*) by the structural model is given
as [37]

p 3
E 3
2:3 1  U 13
E 2
Fig. 3 Far field strain applied to an RVE with an inclusion
Another structural-based homogenization procedure for
porous material such as foam, called the 3D open cell conditions, the ellipsoidal inclusion experiences a uniform
material model, is from Gibson and Ashby [37]. The eigenstrain eij . By applying the eigenstrain method, the
effective modulus of the porous material is related to the effective modulus of the RVE can be calculated. MT
fiber modulus by [37] theory was originally concerned with the calculation of
E internal stress in a matrix containing inclusions with
1  U 2 14 eigenstrains. However, this theory is valid only for cases
E
where the volume fraction of the inclusion is small. The
According to Thelen et al. [37], the 3D open cell model MT method treats the different inclusions as distinct
is based on assumptions of high porosity; and it gives good regions and does not take into consideration the geometry
predictions of modulus for materials with porosities in the [40, 41]. The MT formulation used in this work follows
range of 1090%. The nanofibrous materials definitely fall closely with that of Fisher et al. [40]. In a multiphase
in this range, and therefore, this model can be used in the model (i.e., material with multiple inclusions), as in the
conservative prediction of the elastic modulus [27, 38, 39]. case of a void-PLLA RVE, the different regions are
However, the major drawback of the methods mentioned represented as distinct cylindrical phases equivalently dis-
above is that the actual amount of voids present in the persed in the matrix (see Fig. 4). This model is further used
nanofibers is not known for comparing with experimental in the study of fiber orientations. To elucidate the expres-
data. Hence the porosity based methods cannot be used for sions for MT method, we assume that the composite is
a reliable estimate of the stiffness of the nanofiber and composed of K phases. The stiffness of the matrix is
therefore we need to look at theories that take into con- denoted by Cm and the volume fraction of the matrix
sideration both the porosity and orientation of the nanofiber is denoted by vm. The kth phase (or inclusion) has a stiff-
constituents. ness of Ck and volume fraction of vk. The dilute strain
concentration factor for the kth phase, denoted by Adil k ,
relates the volume averaged strain in the kth inclusion to
that of the matrix [40, 41] and it is obtained from
Continuum volume averaging: micromechanical
method h i X
K 1
Sk Cm Ck  Cm 1 Adil
r  vn Sn Adil
n I 15
Applying the eshelby eigenstrain formulation, the effect of n
the fiber phase on the matrix stress is captured by means of
an averaged strain concentration tensor [13]. The strain
concentration tensors for various morphologies of the fiber
phase are considered to cause corresponding eigenstrains
on the matrix layer. This can be analyzed using the Mori
Tanaka (MT) method, which is detailed below.
Consider an RVE subjected to a homogenous displace-
ment boundary condition that produces a uniform strain eoij
in an infinite homogenous material containing an inclusion
as shown in Fig. 3. Eshelby has shown that under the above Fig. 4 Idealized lamellar fibril homogenized model

123
J Mater Sci (2007) 42:88448852
where k; n ff ; g; . . . ; K  1g, and Sk is the Eshelby
Tensor for the dispersed inclusions. The effective modulus
of the composite (i.e., matrix with inclusions), C, is found
from
" #
X
K1
C Cm I  vk Adil 16
k modulus Ec ; Gc of the chain used in this model by the
k1
Continuous chain model of polymeric fibers
Subsequent to the homogenization of the ShishKebab
model, it is found that a fibril intertwines around other
fibrils to form the nanofiber. Tan and Lim [11] have
reported that a fibril might terminate by connecting another
fibril or it may branch into two others. This type of com-
plex intertwining cannot be modeled by simple homoge-
nization techniques and therefore a detailed analytical
procedures need to be considered. The deformation char-
acteristics of an oriented crystalline polymeric fiber have to
take into consideration, apart from the mechanical prop-
erties of the material, the molecular arrangement in the
nanoscale, and at larger length scales [20, 42]. The model
that we use in this paper is based on the analysis of
extension of oriented crystalline fibers called the contin-
uum chain model. However, we extend this theoretical
formulation by incorporating the effect of the smaller-scale
material properties by adequate homogenization tech-
niques. It has also been experimentally shown that a
polymeric fiber experiences shear deformation when sub-
jected to a tensile test [18]. The elastic deformation of a
crystalline fibril is the result of the extension of the chain,
which is the predominant effect, and the shear between
adjacent chains is the secondary effect [14, 43].
The continuous chain model (series model) developed
by Northolt and van der Hout [14, 20, 44] is used for the
description of the tensile deformation of the fibers (see
Fig. 5). This model describes the deformation of a poly-
meric fiber as the sum of a linear extension and a rotation
Domain
Chain Segment
Fiber
Fiber Axis
Chain
Fig. 5 Schematics of a chain, a chain segment and the surrounding
domain in the analysis using a continuous chain model
8849
of the chains towards the fiber axis. The deformation of the
fiber is taken as the average deformation of a polymer
chain in the direction of the fiber axis as is shown in
Fig. 5. Detailed description of the continuous chain model
can be found in Northolt [27], and Northolt and van der
Hout [14]. We propose to modify the elastic and shear
homogenized elastic and shear modulus Ec S; vS ; vk ;
Gc S; vs ; vk , which are functions of (1) the Eshelby tensor
(S) for the circular inclusions, and (2) the volume fraction
of the Shish and Kebab vS ; vk or an equivalent homoge-
nized structure. Thus, the effective fiber modulus is
obtained by
 2 
1 1 sin h E
17
Efiber Ec S; vs ; vk 2Gc S; vs ; vk
where Efiber is the fiber modulus, hsin2 hiE is the strain
orientation parameter and is given by the following
equation. [14, 17]:
p=2
R
qhcosh sin3 h dh
  0
sin2 h E 18
p=2
R
qhcosh sinh dh
0
The modified fibril strain can subsequently be written as
the sum of the elastic strain and the strain due to elastic
rotation or shear of the fibrils as given by
 2   
r sin h E  r
ef 1  e GS;vs ;vk 19
Ef S; vs ; vk 2
Results and discussion
There is a wide range in the reported values of the
mechanical properties of PLLA fibers [14, 1820, 43, 44].
These are primarily affected by various factors like rate of
drawing of the polymer, temperature and crystallinity of
the polymer material [14, 19, 44]. Most of the studies
carried out so far do not take into consideration the porosity
of the nanofiber. There have been very few studies on the
internal structure of an electrospun nanofibrous material
and most of these studies have been aimed at providing the
factors affecting the nanofiber dimensions [45, 46]. The
mechanical stiffness of the nanofiber obtained by using
Timoshenko beam theory and ordinary beam bending
theory give conservative values. These results are not
reliable since they do not capture the inherent orientation
inhomogeneity of the nanofiber. The material modeling
123
8850
strategy used here is novel as it considers the inhomoge-
neity of the nanofiber and the orientation of the fibrils
[7, 11, 15, 36]. This modeling procedure is carried out by
using mathematically well-established multiscale modeling
simulation techniques coupling the atomistic scale to
macroscopic scales. To the best knowledge of the authors,
such a methodology of extracting the material properties
from a completely computational point of view (indepen-
dent of experimental data) for nanofibers has not been
attempted. As homogenization methods considering only
the individual aspects of material modeling at different
scales are available in literature, a multiscale computa-
tional framework is being proposed in this work. The
material properties are extracted from the molecular to the
macro level, and finally validated with independent
experimental results. The uniqueness is in the multiscale
approach proposed here.
High strength PLLA fibers of the order of 16 GPa with
high crystallinity and porosity has been produced by dry
spinning [38]. Leenslag and Pennings have reported a tensile
modulus of 14 GPa for solution-spun PLA fibers [39].
Numerous studies by researchers have produced high
modulus PLLA fibers for various uses, having elastic
modulus ranging from 1 to 20 GPa. For example, Tan and
Lim [36] reported elastic modulus values of 110 GPa,
Hoogsteen et al. [28] 16 GPa, Yuan et al. [47] 15 GPa,
Broz et al. [48] 3.0 GPa, and Cicero and Dorgan [49]
reported 1.53.0 GPa for different draw ratios. Inai et al. [1]
reported the elastic modulus in the range of 2.9 0.4 GPa
for semi-crystalline electrospun polymeric PLLA fibers.
Most of the above reported values were attained by the
estimation of elastic stiffness in tension. Flexural modulus
in the range of 69 GPa was obtained by Lim et al. [50].
The Youngs modulus obtained from MD analysis of
crystalline lactic acid should conform to the experimental
value of the modulus of crystalline PLLA [51]. The
experimentally obtained elastic modulus for a ~90% crys-
talline PLLA made by a hot drawn (melt spinning) process
is 9.2 GPa [51] and a Poissons ratio of 0.44 has been
reported by Balac et al. [52]. The MD simulations were
performed with a fixed time step of 1 fs and the interatomic
interactions were calculated using a Universal 1.02 force
field (UFF) of Cerius2 (version 4.6, Accelrys, Inc.) in all
the simulations [53]. This potential function includes van
der Waals, bond stretch, bond angle bend, and torsional
rotation terms. The computational unit cell was minimized
and equilibrated by NVE process. The temperature scaling
was carried out using NVT ensemble with the Nose
Hoover thermostat. During the minimization and NVT
processes, the atoms are allowed to equilibrate within the
fixed MD cell.
An elastic modulus of 9.44 GPa along the major axis,
and 5.71 GPa and 4.57 GPa along the minor axes with an
123
J Mater Sci (2007) 42:88448852
average Poissons ratio of 0.4 was obtained using MD
simulations. The resultant effective modulus of the nano-
fibrous structure obtained using MoriTanaka method was
5.77 GPa, by considering about 0.2 volume fraction of the
Shish in Kebab in the nanofiber. This obtained effective
modulus closely matches with many of the effective
properties obtained [28, 36, 50]. The effective modulus of
the ShishKebab configuration is now used in the modified
continuum chain model to obtain the variation of the
effective properties of the nanofiber with varying orienta-
tion parameter values.
The elastic stiffness calculated using the MoriTanaka,
3D open cell and honeycomb structure models is given in
Fig. 6. It can be seen that the predicted stiffness values has
a maximum of 5.9 GPa and decreases with an increase in
porosity. The amount of porosity is also indicative of the
diameter of the fiber, as the diameter increases the porosity
of the fiber also increases as seen from experimental
observations [11]. It can be seen from Fig. 6 that with an
increase in diameter the elastic stiffness would decrease.
This method, however, fails to provide an accurate esti-
mation of the elastic modulus of nanofibrous materials,
when compared to experiments (see, for example, Inai
et al. [1]). The average elastic property obtained by
homogenization of the ShishKebab model is used in the
modified continuum chain model. In this continuum chain
model, the homogenized elastic property is predicted by
Eq. 17 using the strain orientation parameter derived from
the birefringence data (Eq. 18). This analytically predicted
elastic modulus of PLLA fibers (as shown in Fig. 7) closely
matches with the experimental values of Mezghani [54].
The stress-strain curves, using the continuum chain model
and the experimentally determined values for different
draw velocities by Inai et al. [1] is compared in Fig. 8.
8
Mori Tanaka
3D-Open Cell
Elastic Stiff ness ( G Pa)
Honey Comb
6
4
2
0
0 0.2 0.4 0.6 0.8 1
Void Volume Fraction
Fig. 6 Variation of elastic modulus of nanofibers with void volume
fractions
J Mater Sci (2007) 42:88448852 8851

7 MD simulation to obtain the crystalline elastic modulus.

Mezghani (Expt)
6.5
The next scale of modeling is the homogenization of the
Continous Chain Model
ShishKebab model using the MoriTanaka method. Based
6
Elas tic M odulus (GPa)

on this homogenization principle, the modified elastic

5.5 constants are obtained and are subsequently used in the
5
homogenized-continuum chain model to obtain the mac-
roscale homogenized elastic modulus. The highlight here is
4.5
the multiscale method of estimating the elastic properties
4 of PLLA nanofibers, as compared to previous computa-
3.5
tional procedures that depend solely on either the con-
ventional continuum chain model or atomistic simulations
3 only. The simulation results obtained show excellent cor-
2.5 relation with that of experiments, even without involving
0 0.1 0.2 0.3 0.4 0.5 any experimental data in the analysis.
<sin2 >E
Acknowledgements The authors would like to acknowledge sup-
Fig. 7 Variation of elastic modulus with the strain orientation port of the Laboratory for Molecular Simulation at Texas A&M
parameter University for providing software and computer time. The first three
authors also gratefully acknowledge the partial support received from
Dr. C. F. Shih at the National University of Singapore.
30

25
References

1. Inai R, Kotaki M, Ramakrishna S (2005) Nanotechnology 16:208

20 2. Lim TC, Kotaki M, Yong TKJ, Yang F, Fujihara K, Ramakrishna
S (2004) Mater Technol (Nano Bio-engineering Special) 19:20
St ress ( MPa)

3. Lim TC, Ramakrishna S (2005) In: Next-generation applications

15
10
63 m/Min
5 630 m/Min
Continuous Chain Model
0 8. Freed LE, Vunjak-Novakovic G, Biron RJ, Eagles DB, Lesnoy
0 0.02 0.04 0.06
Strain
Fig. 8 Comparison of stress-strain curves with that obtained from
experiments
From the figure, it can be seen that the predicted stress-
strain curve lies closer to the higher draw velocity curve.
The modified continuum chain model predicts a stiffer fiber
due to the inadequate information on the internal structure
of the nanofiber. With a better understanding of the internal
structure a more refined estimate of the stress-strain curve
can be obtained.
Conclusions
In this study, a multiscale modeling approach is used to
obtain the effective elastic modulus of the PLLA nanofiber.
The analysis is carried out from the atomistic level using
for polymeric nanofibres. John Wiley, England, p 137
4. Ramakrishna S, Fujihara K, Teo WE, Lim TC, Ma ZW (2005) In:
An introduction to electrospinning and nanofibers. World Sci-
entific, Singapore
5. Ramakrishna S, Fujihara K, Ganesh VK, Teo WE, Lim TC
(2005) In: Science and engineering of polymer nanofibers.
American Scientific Publisher, California
6. Zammaretti P, Jaconi M (2004) Curr Opin Biotechnol 15:430
7. Tan EPS, Lim CT (2006) J Biomed Mat Res Part A 77A:526
0.08 0.1 DC, Barlow SK, Langer R (1994) Bio/Technology 12:689
9. Moran JM, Pazzano D, Bonassar LJ (2003) Tissue Eng 9:63
10. Lee KH, Kim HY, Khil MS, Ra YM, Lee DR (2003) Polymer
44:1287
11. Tan EPS, Lim CT (2005) Appl Phys Lett 87:123106
12. Dukovski I, Muthukumar M (2003) J Chem Phys 118:6648
13. Unnikrishnan VU, Reddy JN (2005) Int J Multiscale Comput Eng
3:437
14. Northolt MG, Hout RVD (1985) Polymer 26:310
15. Tan EPS, Lim CT (2006) Composites Sci Technol 66:1099
16. Richardson ID, Ward IM (1970) J Phys D: Appl Phys 3:643
17. Northolt MG (1980) Polymer 21:1199
18. Northolt MG, Baltussen JJM (2002) J Appl Polym Sci 83:508
19. Northolt MG, Baltussen JJM (2002) J Appl Polym Sci 83:508
20. Northolt MG, Roos A, Kampschreur JH (1989) J Polym Sci: Part
B: Polym Phys 27:1107
21. Frankland SJV, Brenner DW (1999) In: National meeting of the
american chemical society. New Orleans, LA, p 38953
22. Qian D, Wagner GJ, Liu WK, Yu MF, Ruoff RS (2002) Appl
Mech Rev 55(6):495
23. Chandra N, Namilae S, Shet C (2004) Phys Rev B 69:94101
24. Mylvaganam K, Zhang L (2004) Carbon 42:2025
25. Griebel M, Hamaekers J (2004) Comp Methods Appl Mech Eng
193:1773

123
8852
26. Blonski S, Brostow W, Kubyt J (1994) Phys Rev B 49:6494
27. Kalb B, Pennings AJ (1980) Polymer 21:607
28. Hoogsteen W, Postema AR, Pennings AJ, Ten Brinke G (1990)
Macromolecules 23:634
29. De-Santis P, Kovacs AJ (1968) Biopolymers 6:299
30. Yamane H, Sasai K, Takano M (2004) J Rheol 48:599
31. Fujita M, Doi Y (2003) Biomacromolecules 4:1301
32. Namilae S, Chandra N, Shet C (2004) Chem Phys Lett 387:247
33. Somania RH, Yanga L, Zhub L, Hsiao BS (2005) Polymer
46:8587
34. Welch P, Muthukumar M (2001) Phys Rev Lett 87:218302
35. Zussman E, Rittel D, Yarin AL (2003) Appl Phys Lett 82:3958
36. Tan EPS, Lim CT (2004) Appl Phys Lett 84:3400
37. Thelen S, Barthelat F, Brinson LC (2004) J Biomed Mat Res
69A:601
38. Leenslag JW, Pennings AJ (1987) Polym Commun 28:92
39. Leenslag JW, Pennings AJ (1987) Polymer 28:1695
40. Fisher FT, Bradshaw RD, Brinson LC (2003) Compo Sci Technol
63(11):1689
41. Mura T (1997) In: Micromechanics of defects in solids. Kluwer
Academic
42. Ward IM (1967) Br J Appl Phys 18:1165
123
J Mater Sci (2007) 42:88448852
43. Baltussen JJM, Northolt MG (2001) Polymer 42:3835
44. Baltussen JJM, Northolt MG, Hout RVD (1997) J Rheol 41:549
45. Ruotsalainen T, Turku J, Heikkila P, Ruokolainen J, Nykanen A,
Laitinen T, Torkkeli M, Serimaa R, Ten Brinke G, Harlin A,
Ikkala O (2005) Adv Mater 17:1048
46. Demir MM, Yilgor I, Yilgor E, Erman B (2002) Polymer 43:3303
47. Yuan XY, Mak AFT, Yao KD (2003) Polym Degrad Stability
79(1):45
48. Broz ME, Vander Hart DL, Washburn NR (2003) Biomaterials
24:4181
49. Cicero JA, Dorgan JR (2001) J Polym Environ 9:1
50. Lim JY, Kim SH, Lim S, Kim YH (2002) Macromol Chem Phys
202:2447
51. Fambri L, Pergoretti A, Fenner R, Incardona D, Migliaresi C
(1997) Polymer 38:79
52. Balac I, Milovancevic M, Tang C-Y, Uskokovic PS, Uskokovic
DP (2004) Mater Lett 58:2437
53. Rappe AK, Casewit CJ, Colwell KS, Goddard WA, Skiff WM
(1992) J Am Chem Soc 114:10024
54. Mezghani K, Spruiell JE (1998) J Polym Sci: Part B: Polym Phys
36:1005
J Mater Sci (2007) 42:88538863
DOI 10.1007/s10853-007-1812-6
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
A polyconvex hyperelastic model for fiber-reinforced materials
in application to soft tissues
Alexander E. Ehret Mikhail Itskov
Received: 15 December 2006 / Accepted: 30 April 2007 / Published online: 12 August 2007

Springer Science+Business Media, LLC 2007
Abstract In this paper a generalized anisotropic hyper-
elastic constitutive model for fiber-reinforced materials is
proposed. Collagen fiber alignment in biological tissues is
taken into account by means of structural tensors, where
orthotropic and transversely isotropic material symmetries
appear as special cases. The model is capable to describe
the anisotropic stress response of soft tissues at large
strains and is applied for example to different types of
arteries. The proposed strain energy function is polyconvex
and coercive. This guarantees the existence of a global
minimizer of the total elastic energy, which is important in
the context of a boundary value problem.
Introduction
In soft biological tissues the distribution, arrangement and
interaction of the constituents lead to a diversity of
mechanical characteristics and thus high specificity and
functionality. In particular, the extracellular matrix plays
an important role. Its main constituents are proteins, gly-
cosaminoglycans as well as bound and unbound water, see
e.g., [1]. Among those proteins, different collagen types are
crucial for the mechanical properties of soft tissues. Indeed,
some collagen types form fibers or networks and thus
provide reinforcing structures. By ordered arrangement,
this finally leads to anisotropy. The alignment of these
structures is manifold and reaches from parallel fiber
bundles in tendons to helical arrangement in arteries and
A. E. Ehret (&)
M. Itskov
Department of Continuum Mechanics, RWTH Aachen
University, Eilfschornsteinstr. 18, 52062 Aachen, Germany
e-mail: ehret@km.rwth-aachen.de
two- and three-dimensional networks in skin, see e.g., [2].
Collagen fibers are usually crimped or undulated in a nat-
ural state. With increasing strain they line-up, straighten,
and finally become the main load bearing elements. As a
consequence, the stress response of soft tissue samples is in
general characterized by highly non-linear behavior with
J-shaped or nearly exponential stress-strain curves. Hence,
from a biomechanical perspective, both anisotropy and
non-linear behavior at large elastic strains should be
regarded by a constitutive model in order to describe soft
tissues appropriately. In the elastic domain, hyperelastic
models meet these requirements and have therefore
extensively been utilized in soft tissue mechanics.
The diversity among the mechanical characteristics of
soft tissues motivated a great number of constitutive for-
mulations for different tissue types. For example, the
reader is referred to [3] for a survey on strain energy
functions for planar biological tissues in connection with
biaxial testing techniques. Modeling approaches for vari-
ous components of the cardiovascular system the reader
may find in [1] and references therein. In particular, arterial
wall mechanics as well as several hyperelastic models for
arterial tissue are studied in [4]. Several strain energy
functions for myocardium have recently been compared by
Schmid et al. [5]. Although the major part of these models
is based on phenomenological approaches, a number of
micromechanically based models have been suggested. For
example, chain network models (see e.g., [6]), taking into
account single collagen fibrils by means of statistical
mechanics, have been applied to model orthotropic and
transversely isotropic soft tissues [7, 8]. Other authors
derived constitutive relations based on sinusoidal, zig-zag
or circular helix representations for the crimped collagen
fibers, see e.g., [911]. The material parameters appearing
in microstructurally based models generally have a
123
8854
physical meaning. However, estimation of these parameters
might be a difficult task in view of the variance among
individual tissue samples. In general, phenomenological
models are able to adequately describe the macroscopic
behavior of soft tissues observed in experiments and have
therefore wide applicability. Although the related material
parameters do not usually provide a clear physical inter-
pretation, they can be obtained by fitting the model to
experimental data. Notwithstanding, even within a phe-
nomenological approach it is preferable to include as much
information about the structure as possible. Thus, motivated
by the histological structure of blood vessels, Holzapfel
et al. [4] considered each layer of an arterial wall as a fiber-
reinforced composite of an isotropic matrix and two sym-
metrically arranged fiber helices. This led to a multi-layer
constitutive model with an elastic potential composed of an
isotropic part and an anisotropic part associated with the two
fiber contributions. In a recent paper, Gasser et al. [12] have
presented an approach to replace the mean fiber directions in
the latter model by a collagen fiber dispersion as experi-
mentally observed by means of polarized light microscopy.
Besides these biomechanical aspects, the issue of
material stability should be taken into account if the tissue
is assumed to maintain its integrity throughout the elastic
deformations. Possible criteria for material stability have
recently been discussed [13]. One of those is the strong
ellipticity condition, which also implicates positive defi-
niteness of the acoustic tensor so that the speed of dis-
placement waves is real for any direction of propagation.
Contrariwise, the loss of ellipticity has recently been used
to describe several failure mechanisms in fiber-reinforced
materials like, e.g., fiber kinking, splitting and de-bonding
[14, 15]. Investigation of ellipticity for several constitutive
models (see e.g., [16]) showed, for example, that the
so-called Fung elastic model [2, 17] widely and success-
fully applied to soft biological tissues is in general not
elliptic and that enforcement of the strong ellipticity con-
dition imposes severe restrictions on the material constants.
For polyconvex strain energy functions [18] on the other
hand, ellipticity is guaranteed (see e.g., [19]). Another
advantage of polyconvex functions appears in the context
of boundary value problems. The question whether there is
a solution of the boundary value problem is bound to the
existence of a global minimizer of the total elastic energy
of a body. According to Ball [18, 20] this minimizer exists
if the strain energy function is polyconvex and satisfies a
certain growth condition referred to as coercivity. Hence,
polyconvexity provides an excellent starting point to for-
mulate strain energy functions that guarantee both ellip-
ticity and existence of the global minimizer.
Instead of checking various anisotropic hyperelastic
models for polyconvexity, one can formulate polyconvex
strain energy functions from scratch. Schroder and Neff
123
J Mater Sci (2007) 42:88538863
[19] investigated polyconvexity of numerous functions of
isotropic and anisotropic strain invariants. On this basis,
they proposed a variety of polyconvex free energy terms
for a transversely isotropic material as well as an extension
for orthotropic materials. Furthermore, for these material
symmetries, sufficient conditions for the polyconvexity and
coercivity of the strain energy function were elaborated by
Steigmann [21]. Itskov and Aksel [22] presented a class of
transversely isotropic and orthotropic polyconvex and
coercive strain energy functions. An advantage of the latter
formulation is that the functions fulfill the condition of an
energy and stress-free undeformed configuration a priori.
Their hyperelastic model is based on a power series rep-
resentation with an arbitrary number of terms. A modified
version of the model, based on an exponential function
representation was successfully utilized to describe the
mechanical behavior of soft collagenous tissues by Itskov
et al. [23]. Balzani et al. [24] followed the approach by
Holzapfel [4] and presented a number of polyconvex strain
energies for soft biological tissues consisting of an isotro-
pic part and the superposition of several transversely iso-
tropic contributions. We remark that some of the material
parameters in these models should be restricted further to
prevent the initial stiffness associated with the fiber part
from being infinite. In the models [4, 24], the transversely
isotropic parts are switched off if the associated preferred
directions are under compression. This guarantees
convexity of the strain energy function, however, it leads
to purely isotropic behavior when all fibers are in a
compressive state.
In this paper we present a polyconvex anisotropic
hyperelastic constitutive model for materials consisting of
an isotropic matrix and reinforced by an arbitrary number of
fiber families. Each fiber family is explicitly taken into
account by a structural tensor and associated with a weight
factor. The model is formulated in terms of so-called gen-
eralized structural tensors. It is shown that for a material
reinforced by a single family of fibers, these tensors coin-
cide with a recently proposed structure tensor that includes
collagen fiber dispersion [12]. The model is developed in a
generalized form offering maximum flexibility for use with
different types of engineering materials and biological
tissues. Earlier models for orthotropic and transversely
isotropic materials are included as special cases.
The paper is organized as follows: We begin with the
basic mathematical notations and definitions as well as a
short discussion of hyperelasticity. Then, the generalized
structural tensors and a functional basis for the anisotropic
strain energy function are introduced. We proceed with the
derivation of the generalized polyconvex and coercive
strain energy function. Finally, numerical examples are
presented, showing special cases of the model in applica-
tion to arterial tissue.
J Mater Sci (2007) 42:88538863 8855

Preliminaries Hyperelasticity

Basic tensor operations An elastic material is called hyperelastic if its elastic

behavior can be described by a strain energy function
Let Lin be a set of all linear mappings of a three-dimen- W = WF(F) per unit reference volume, where F denotes the
sional vector space R3 over reals into itself. Lin represents deformation gradient. According to the principle of objec-
a finite-dimensional vector space with the inner product. tivity, the strain energy has to be independent of superposed
The elements of Lin are called second-order tensors. rigid body motions. This requirement can automatically be
Subsets of Lin are defined by Sym = {M [ Lin : satisfied representing the strain energy function in terms of
M = MT}, Orth = {Q [ Lin : Q = QT}, and Inv = {A [ the right CauchyGreen tensor C = FTF so that
Lin : detA = 0}, constituted by symmetric, orthogonal,
and invertible second-order tensors, respectively, where W WF F WC C: 5
detA denotes the determinant of a second-order tensor A. For an unconstrained hyperelastic material, a
Sym+ is a subset of Sym formed by all positive definite constitutive law can be given by (see e.g., [27])
symmetric second-order tensors. The adjugate of a second-
order tensor is denoted by adjA = A1 detA and the oW
cofactor by cofA = AT detA, where A [ Inv. S2 ; 6
oC
Let Lin be a set of all linear mappings of Lin into
where S denotes the second PiolaKirchhoff stress tensor.
itself so that B D : A; B 2 Lin; 8A 2 Lin; 8D 2 Lin.
The material time derivative of Eq. (6) yields
The elements of Lin are called fourth-order tensors. They
can be constructed from second-order tensors by means of 1_
the tensor products  and  such that (see e.g., S_ C : C; 7
2
[25])
where C is the tangent tensor of fourth order defined by
A  B : C AB : C; A  B : C ACB;
1 oS o2 W
A; B 2 Lin; 8C 2 Lin: C2 4 : 8
oC oCoC
Note that these definitions differ from notations in other For constrained materials, the constitutive law takes the
works. A simple composition of fourth-order tensors with form
second-order ones is introduced by
oW of
ADB : C AD : CB; D 2 Lin; A; B 2 Lin; 8C 2 Lin: S2 q ; 9
oC oC
2
with an arbitrary scalar q and a function f accounting for
The calculation of stresses and elasticity tensors requires the kinematic constraint f(C) = 0. Accordingly, the stress
application of the derivative with respect to a tensor. This rate and the tangent tensor of fourth order are given by (see
derivative obeys the following product rules of e.g., [28])
differentiation [26]
1 of o2 W o2 W
f A;C A  f ;C f A;C ; S_ C : C_ q_ ; C4 2q : 10
2 oC oCoC oCoC
3
AB;C A;C B AB;C ;

where f, A, and B represent a scalar and two tensor-valued

differentiable tensor functions of C, respectively. The Anisotropic strain energy functions
notation ()S indicates a symmetrization operation on
fourth-order tensors defined by (see e.g., [25]) Material symmetry
1

DS : A D : A AT ; D 2 Lin; 8A 2 Lin: 4 The symmetry group G of a material is a set of all
2
orthogonal mappings which preserve the material symme-
A function f of several arguments, say x1, x2,, xn, will try. For anisotropic materials, the symmetry group can be
be abbreviated by the notation f f^xi ; i 1; 2; . . .; n. defined with the aid of structural tensors Li, i = 0,1,,n as

123
8856 J Mater Sci (2007) 42:88538863

 
G Q 2 Orth : QLi QT Li ; i 0; 1; . . .; n : 11
For a hyperelastic material, the condition of material
symmetry is written in terms of the strain energy function
(5) and the symmetry group (11) by
WC QCQT WC C; 8Q 2 G: 12
According to Rychlewskis theorem [29], this condition
is satisfied if and only if the strain energy can be
represented as an isotropic tensor function of arguments
containing the structural tensors. In view of (11) one can
thus write (see e.g., [22])
For orthotropic and transversely isotropic materials,
another set of generalized structural tensors has recently
been presented [22]. Orthotropy is characterized by sym-
metry with respect to three mutually orthogonal planes by
reflections from which the material properties remain
unchanged. The axes normal to these planes are referred to
as the principal material directions. Introducing unit vec-
tors li ; i 1; 2; 3; in these directions allows to form the
structural tensors (see e.g., [22])
b 1 l1  l1 ; L
L b 2 l2  l2 ;
16
b3 I  L
L b1  Lb 2 l3  l3 ;

WCL QCQT ; QLi QT WCL C; Li ; where a superposed hat serves to distinguish them from
13 those associated with fiber directions (14). Transverse
i 0; 1; . . .; n; 8Q 2 Orth:
isotropy represents a material symmetry with respect to
At first, we consider a general fiber-reinforced material only one preferred direction. Rotations about this axis and
consisting of an isotropic matrix and an arbitrary number n reflections from planes parallel or orthogonal to it preserve
of fiber families as schematically shown in Fig. 1. Let the the material properties. Specifying this principal material
alignment direction of each fiber family be given by a unit direction by a unit vector l1 ; two structural tensors can be
vector mi, i = 1,2,...,n. Then we define n + 1 structural defined by (see e.g., [22])
tensors Li, i = 0,1,...,n by
b 1 l1  l1 ;
L b 2 1 I  L1 :
L 17
1 2
Li mi  mi ; L0 I; i 1; 2; . . .; n; 14
3 In a similar manner as (15), linear combinations of the
where I denotes the identity tensor of second order. The tensors (16) or (17) lead to the generalized structural
tensor L0 is assumed to be associated with the isotropic tensors (cf. [22, 23])
matrix. In the next step, we form linear combinations of the
tensors (14) and thus define so-called generalized structural X
m
r b
X
m
r
~r
L wi L i; wi 1; r 1; 2; . . .; 18
tensors by i1 i1

X
n
r
X
n
r
where m = 3 for orthotropy, m = 2 for transverse isotropy
~r
L vi Li ; vi 1; r 1; 2; . . .; 15 and w(r)
i  0 denote scalar weight factors. In fiber-rein-
i0 i0
forced materials orthotropy and transverse isotropy arise as
where v(r)
i  0, i = 0,1,...,n denote scalar weight factors. special cases. Indeed, certain arrangement of two or more
Note that both the structural tensors (14) and the general- fiber families may lead to orthotropic material symmetry
ized structural tensors (15) are characterized by the prop- and one or more fiber families aligned in one single
~ r 1; i 0; 1; . . .; n; r 1; 2; . . ..
erty trLi trL direction result in transverse isotropy (see e.g., [30]). In
this case, the generalized structural tensors defined by (15)
and (18) should coincide so that consequently the weight
factors w(r)
i , i = 1,2,,m, associated with principal material
directions (16, 17) are related to the factors v(r) i ,
i = 0,1,,n, weighting the fiber directions (14) (see
Appendix).
In a recent work [12], the dispersion of collagen fibers
was included into a constitutive model for arterial tissue.
Therein, it is assumed that the fiber distribution is char-
acterized by rotational symmetry about a mean preferred
direction given by the unit vector a0 . As a measure for the
fiber distribution, a scalar parameter j is introduced such
that a so-called structure tensor H takes the form [12]

Fig. 1 General fiber-reinforced material

H jI 1  3ja0  a0 : 19

123
J Mater Sci (2007) 42:88538863 8857

It is seen that the same form is also obtained as a special

case of (15), setting n 1; m1 a0 and v(r) 0 = 3j, which
underlines the physical interpretability of the weight
factors.
WF F WF; adjF; detF: 23
The arguments F, adjF and detF describe the deformation
of line, surface and volume elements, respectively. A
subclass of polyconvex functions (23) can be obtained by
the additive representation [19]

Functional basis
1 F W
WF F W
2 adjF W

3 detF; 24

The right CauchyGreen tensor C and the structural tensors
(14) form a finite system of tensors. According to Hilberts
theorem, one can find a finite set of isotropic invariants Vk,
k = 1,2,...,t of these tensors called the functional basis, in
terms of which all other isotropic invariants can be
expressed. Thus in view of (13), one can write
where W
i ; i 1; 2; 3, are convex functions of their argu-
ment, respectively.
We recall the following basic property of convex
functions: Let uA : Inv ! R be convex and p : R ! R
be convex and monotone increasing, then p(u(B)) is con-
vex. Indeed, for some B1, B2 [ Inv and k [ [0, 1], we have

^ V k ;
W W k 1; 2; . . .; t: 20
Let us first consider the case where all fiber families are
either parallel or mutually orthogonal. Then, taking into
account the symmetry of C and Li, i = 1,2,...,n, and
applying the classical invariant theory (see e.g., [25, 31]), a
functional basis can be given by
V 1 trC; V 2 trC2 ; V 3 trC3 ;
V 3i trCLi ; V 3ni tr C2 Li ; i 1; 2; . . .; n;
21
see also [30]. For a general fiber-reinforced material with
arbitrary fiber alignment, the functional basis (21) is
completed by the invariants
     
tr CLi Lj ; tr Li Lj ; tr Li Lj Lk ; i\j\k 1; 2; . . .; n:
22
In regard to the constitutive modeling, however, it is
useful to restrict the arguments of the strain energy
function. The invariants (22)2,3 are constants and will be
neglected for this reason. Further, we omit the dependence
of the strain energy on the coupled terms (22)1 which
reflect deformations of the fiber families mp and mq
relative to each other. Accordingly, in the following, we
reduce the list of arguments of the strain energy to the
invariants (21) even in the case of an arbitrary fiber
orientation and set t = 3 + 2n.
Polyconvex strain energy functions
Polyconvexity
pukB1 1  kB2  pkuB1 1  kuB2
25
due to the convexity of u and monotonicity of p. Since the
latter is also convex we can further write
pkuB1 1  kuB2  kpuB1 1  kpuB2
26
which implies convexity of p(u(B)).
Now let us consider the convexity properties of the in-
variants Vk, k = 1,2,...,t, (21), bearing in mind the
requirements for polyconvex functions (23). Indeed it can
be shown that the invariants V3+n+i, i = 1,2,...,n, are not
convex with respect to F, while the remaining ones in (21)
satisfy this condition [19]. However, on the basis of the
CayleyHamilton theorem, the invariants (21) can be
expressed uniquely in terms of other invariants [22]
Ii trCLi ; Ji trcofCLi ; IIIC detC;
27
i 0; 1; . . .; n;
which are convex with respect to F, adjF and detF,
respectively [19], and thus provide a suitable basis to
formulate polyconvex functions. On account of the positive
semi-definiteness of the structural tensors (14), we notice
the following important property
Ii [ 0; J i [ 0; i 0; 1; . . .; n: 28
The convexity properties remain unaffected by forming
linear combinations of the invariants (27)1,2 with non-
negative weight factors (see e.g., [32]). Thus, in view of
(15) one can define generalized invariants based on the
generalized structural tensors as

X
n
r   X
n
r  
A strain energy function W W F F : Inv ! R is said to I~r ~ r ; J~r
vi Ii tr CL ~r ;
vi Ji tr cofCL
be polyconvex [18] if there exists a convex function i0 i0


W WF; adjF; detF : Inv; Inv; R ! R such that r 1; 2; . . .; 29

123
8858 J Mater Sci (2007) 42:88538863

which are likewise convex with respect to F and adjF, On the other hand, the generalized representation (31)
respectively. The definitions (27) and (29) finally allow to enables to account for various material characteristics in a
represent the strain energy function in the form straightforward manner by an appropriate choice of any
convex and monotone increasing functions fr, gr and a
~ I~r ; J~r ; IIIC ;
W W r 1; 2; . . .: 30 convex function hr satisfying (34). Clearly, the earlier
models can be obtained as special cases. Indeed, for fiber
Utilizing the additive representation (24), a set of
orientations with orthotropic or transversely isotropic
polyconvex strain energy functions can hence be
symmetry, the power functions
constructed by
h
  i 1
~ar  1
~br 
1X s
 1=2 fr I~r Ir  1 ; gr J~r Jr  1 ;
W lr fr I~r gr J~r hr IIIC ; 31 ar br
4 r1
1=2 1
cr 
hr IIIC IIIC  1 35
where lr  0 are material parameters with the dimension cr
of stress. According to the above mentioned statement recover the strain energy proposed in [22] and showing
about convexity, the functions fr and gr are convex and good agreement with experiments on calendered rubber,
monotone increasing functions of their arguments, while hr where ar  1, br  1, cr > 0 denote material constants.
are convex with respect to III1/2 C = detF. The strain energy Setting
function is given in terms of a series with an arbitrary
h i h i
number of terms s, which offers wide flexibility in appli-
fr I~r a1r ear Ir 1  1 ; gr J~r b1 ebr Jr 1  1 ;
~ ~
cation to experimental data. Nevertheless, truncating the
c r 
1=2
series after the first term, so that s = 1, is often sufficient to hr IIIC c1 IIIC r  1 ;
r
describe experimental results adequately as, e.g., shown in 36
the numerical examples presented hereinafter.
In the next step, we consider the condition of the energy where ar  0, br  0, cr > 0, yields the exponential
and stress free undeformed configuration. To this end, we model successfully utilized to describe pericardial tissue
set C = I and require and rabbit skin in [23].

1X s
WjCI l fr 1 gr 1 hr 1 0: 32 Coercivity
4 r1 r

Applying (6), the condition of a stress free undeformed For the existence of a global minimizer of the total elastic
state reads as energy of a body, polyconvexity of the strain energy
function is not sufficient. The strain energy function

oW 1Xs  0  additionally has to satisfy a growth condition referred to as

SjCI 2 l fr 1  g0r 1 L~r coercivity [18, 20]. The coercivity condition can be
oC CI 2 r1 r
  formulated as [34, 35]
1
g0r 1 h0r 1 I 0: 33
2 WC  c0 trCp trcofCq   c1 ; 8C 2 Sym 37
It is observed that the hyperelastic model (31) a priori with real, scalar valued constants c0 [ 0; p  1; q  34 and
fulfills the condition of the energy and stress free c1. Inserting (31) into (37) one finds that (37) holds if
undeformed configuration whenever
P
s

1
lr fr I~r  c0 trCp  d1 ;
fr 1 gr 1 hr 1 0; r 1; 2; . . .; s; 4
r1
34

fr0 1 g0r 1  12 h0r 1 1; r 1; 2; . . .; s: P
s
1
4 lr gr J~r  c0 trcofCq  d2 ; 38
r1
Note that the conditions (34) coincide with those Ps  
1 1=2
imposed on the so-called generalized strain measures [33]. 4 lr hr IIIC  d3 ;
r1
Equation (31) represents the natural generalization of
the strain energy functions suggested in [22, 23] from two where d1 + d2 + d3 = c1. Keeping in mind that fr and gr
perspectives. On the one hand, the set of structural tensors are convex, we can use the basic statement for convex
(14) allows to include any fiber reinforcement geometry functions that the first order approximation of a convex
and thus extends the applicability of the previous model function globally underestimates the function (see e.g.,
defined for orthotropic or transversely isotropic materials. [32]). Thus, expanding fr and gr in series around the natural

123
J Mater Sci (2007) 42:88538863 8859

state and truncating these series after the linear term we X

s


have in consideration of (34) C lr fr00 I~r L
~r  L
~r
r1






fr I~r  fr 1 fr0 1 I~r  1 I~r  1 ; g00r J~r III2C C1 L
~ r C1  C1 L
~ r C1



 

 

gr J~r  gr 1 g0r 1 J~r  1 J~r  1 ; r 1; 2; . . .; s:  g0r J~r g00r J~r J~r IIIC C1  C1 L ~ r C1

39 C1 L~ r C1  C1

Consequently, in view of (29), the conditions (38)1,2 are

 1  1
 1
g0r J~r J~r g00r J~r J~r2 h0r IIIC 2 IIIC 2
satisfied if 4

n  1  1

P
s P r h00r IIIC 2 IIIC C1  C1
1
4 lr vi Ii  1  c0 trCp  d1 ; 4
 
r1 i0  40
 1  1
 1
S
P
s P n
r  g0r J~r J~r h0r IIIC 2 IIIC 2 C1  C1
1
4 lr vi Ji  1  c0 trcofCq  d2 : 2
r1 i0

1  o
gr Jr IIIC C  C1 L
0 ~ ~ r C1 C1 L~ r C1  C1 S :
Bearing in mind (14)2 and (27)1,2 and choosing
p = q = 1, one can rearrange the latter inequalities as 44
  n  In vivo, biological soft tissues contain a large amount of
1
P
s
r 1
Ps P r
12 lr v0 c0
trC 4 lr vi Ii 1 d1 0; water and thus show a very slight compressibility. In the
 s r1  r1 
i1  constitutive modeling they are often considered as
1
P r 1
P
s Pn
r
12 l v
r 0 c 0 trcofC 4 lr v i J i 1 d2 0: incompressible. The incompressibility constraint IIIC = 1
r1 r1 i1
can be taken into account setting the constraint function f
41 in (9) to
Taking into account the positive definiteness of Ii and Ji 1

(28), it is seen that the conditions (41) can easily be fC III3C  1: 45

P r
fulfilled, setting c0 1=12 sr1 lr v0 and d1 d2
P Keeping (34) in mind, the strain energy function (31) for
1=4 sr1 lr . Finally, to satisfy the condition (38)3 we
an incompressible material is then given by
assume the functions hr(III1/2
C ) to be bounded below with a
lower bound, say jr, such that 1X s 



W l fr I~r gr K~r ; K~r tr C1 L
~r ; 46
1=2 1=2 4 r1 r
hr IIIC  jr ; r 1; 2; . . .; s; 8IIIC 2 R : 42
where the invariants K~r result from J~r under the constraint
In doing so, we find that condition (38)3 is satisfied
P IIIC = 1. In this case, straightforward insertion of (45) and
setting d3 1=4 sr1 lr jr . Thus, equations (41) and (42)
(46) into (9) and (10) yields the expressions for the second
imply (37).
PiolaKirchhoff stress tensor

Constitutive relations and tangent moduli 1X s 



~ r  g0 K~r C1 L

~ r C1 1 qC1 ;
S lr fr0 I~r L r
2 r1 3
In the following, expressions for the second PiolaKirch- 47
hoff stress and the tangent tensor are derived. Inserting the
strain energy function (31) into the constitutive relation (6), and the tangent tensor of fourth order
the second PiolaKirchhoff stress tensor calculates to
X
s 



1X s 

 C lr fr00 I~r L ~ r g00 K~r C1 L
~ r L
r
~ r C1 C1 L
~ r C1
S ~ r  g0 J~r IIIC C1 L
l f 0 I~r L ~ r C1
2 r1 r r r r1



  g0r K~r C1 C1 L
~ r C1 C1 L~ r C1 C1 S

 1  1 1  
g0r J~r J~r h0r III2C III2C C1 : 43 2 1 1
1 
2 1 1 S
q C C  C C :
3 3
Applying the product rules of differentiation (3), the
48
tangent tensor (8) is written as

123
8860
Numerical examples
In this section, the constitutive model (31) is specified and
applied to two different sets of experimental data on human
arterial tissue. Arteries are composed of three distinct
layers called intima, media and adventitia. In constitutive
modeling, each layer is often considered as an incom-
pressible fiber-reinforced composite with two mechanically
equivalent families of collagen fibers that form symmetri-
cal helices tilted by an angle u against the circumferential
direction [4, 13, 36].
Coronary arteries
In a recent work, Holzapfel et al. [36] have studied the
layer-specific mechanical properties of human coronary
arteries with non-atherosclerotic intimal thickening in
cyclic uniaxial quasi-static tension tests. Therein, strips
from left anterior descending coronary arteries were split
into the adventitial, medial and intimal layer and finally
loaded such that the principal axes of deformation coin-
cided with the circumferential, axial and radial direction of
the vessel. For each strip sample, the recorded stress
response for loading in circumferential and axial direction
was modeled using a hyperelastic constitutive model based
on the above mentioned fiber-reinforcement assumption
(see [36] for details).
In order to describe the mechanical behavior of the
arterial tissue samples, we first specialize the polyconvex
strain energy function (31). To this end, we choose
appropriate formulations for the functions fr, gr and hr. For
all three arterial layers, a clearly exponential shape of the
stress-stretch curves is observed and thus the representation
(36) appears to be suitable. Furthermore, we assume that
each layer can be described as an incompressible rein-
forced material with two equivalent families of fibers
arranged by an angle u as described above. Introducing
unit vectors eh and ez in the circumferential and axial
direction of the artery, respectively, the orientations of the
two fiber families are given by the vectors
m1 cos ueh sin uez ; m2 cos ueh  sin uez : 49 responses of samples obtained from different age groups.
Applying formulae (14, 15, 29, 462) and taking into
account the mechanical equivalence of the fiber families,
so that v(r) (r)
2 = v1 , one obtains the generalized invariants
 
r
I~r 13 1  2v1 k2h k2z kh kz 2
r
2 2 2 2

2v 1 kh cos  u kz sin u ;
r
50
K~r 13 1  2v1 k2 2 2 single layers. Thus, assuming that the principal axis of
h kz kh kz
r
 deformation coincide with the circumferential, axial and
2v1 k2 2 2
h cos u kz sin u ;
2
123
J Mater Sci (2007) 42:88538863
where kh and kz denote the principal stretches in
circumferential and axial direction, respectively.
Furthermore, the incompressibility constraint khkzkr = 1
has been taken into account, where kr denotes the radial
stretch. Considering one single term s = 1 in the series
representation (46) we have in view of (36) (cf. [23])
 l 1  
~   1 

~
 
W exp a I  1  1 exp b K  1  1 ;
4 a b
51
where the index 1 has been omitted for the sake of
clarity and I~ and K~ are given by (50). Note that by this
means, we reduce the set of unknown material constants
to l,a,b,v1 and the angle u. With these results at hand, and
assuming that the lateral directions are stress free in the
uniaxial tension test, the Cauchy stresses in circumferential
and axial direction are calculated, respectively, by
 
oW oW
rh kh ; rz kz : 52
okh okz
This model was fitted and compared to the experimental
results from one sample (specimen IX in [36]). In this way,
the parameters l,a,b,v1 and u were determined for each
layer, where u was likewise treated as a phenomenological
parameter regardless of its physical interpretability. The
experimental and model results are presented as Cauchy
stress versus stretch diagrams in Fig. 2 together with the
values of the material parameters. For all three layers
accurate agreement is observed.
Abdominal aorta
Vande Geest et al. [37] have studied the age dependency of
the mechanical behavior of the human abdominal aorta in
biaxial tension-controlled tests. To this end, infrarenal
abdominal aortic square samples were biaxially loaded in
circumferential and axial direction. Different protocols
varying in the ratio between circumferential and axial
tension, Thh:Tzz, were applied (see [37] for details). An
interesting result is the difference in the form of the stress
Given as second PiolaKirchhoff stress versus Green
Lagrange strain, the curves change from a sigmoidal shape
in the youngest group (<30 years) to an exponential nature
in the older ones. As no data on the single layers of the
arteries are available, we consider the whole artery as an
incompressible fiber-reinforced material with two sym-
metrically arranged fiber helices as described above for the
radial direction, the fiber orientation and generalized
J Mater Sci (2007) 42:88538863 8861

(a) 180 wherein a > 0, b > 0 are material parameters determining

experiment axial
160
Intima
experiment circumferential the limiting values of I~  1 and K~  1, respectively.
=71.333 kPa model
140 Assuming that the radial direction is stress free, the
=100.502
Cauchy Stress [kPa]

120 =87.594 second PiolaKirchhoff stresses in circumferential and

1=0.110 axial direction can be calculated, respectively, by
100
=17.982
80
oW oW
60 Sh k1
h ; Sz k1
z : 54
okh okz
40
20
We fitted the model (53) to the results of the test
protocols with Thh:Tzz ratios 0.5:1, 1:1, and 1:0.5,
0
1 1.05 1.1 1.15 1.2 considering specimen 3 in [37]. The results are shown in
Stretch Fig. 3 together with the data obtained from the plots in
(b) 160 [37]. In the next step, the parametrized model was used to
experiment axial predict the stress in circumferential and axial direction for
140 experiment circumferential
model the remaining protocols 0.75:1 and 1:0.75. The simulated
120 Media as well as the experimental values are presented in Fig. 4
Cauchy Stress [kPa]

100 =9.399 kPa

=25.822
and show good agreement.
80 =21.482 In both examples, identification of the material param-
1=0.062
60 eters was achieved by means of least-squares and a
=26.703
LevenbergMarquardt type algorithm. It should be noted
40
that the experimental data was obtained from printed dia-
20
grams which might have led to some slight inaccuracies.
0
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45
Stretch

(c) 100 80 (a)

2 Piola-Kirchhoff Stress S [kPa]

experiment axial
experiment circumferential
80 model
60
Cauchy Stress [kPa]

Adventitia
60 =20.265 kPa
=86.340 40
=77.521
40 1=0.082
=69.552 20 experiment 0.5:1
experiment 1:1
nd

20 experiment 1:0.5
model
0
0 0 0.2 0.4 0.6 0.8 1
1 1.05 1.1 1.15 1.2 1.25 Green-Lagrange Strain E
Stretch

Fig. 2 Comparison of the polyconvex model (51) with experimental 80 (b)

data on coronary artery layers [36]
2 Piola-Kirchhoff Stress Sz [kPa]

60

invariants are again calculated according to (49) and (50).

40
In particular, we focus on the stress response of the young
arteries. Their s-shaped stress-strain curves show a
noticeable similarity to those obtained in experiments with 20 experiment 0.5:1
experiment 1:1
nd

rubber materials (see e.g., [38]). For this reason, we choose experiment 1:0.5
model
the functions fr and gr in a logarithmic form motivated by 0
the Gent model for rubber elasticity [39]. Setting s = 1 in 0 0.2 0.4 0.6 0.8 1
Green-Lagrange Strain Ez
(36), this leads to

    Fig. 3 Comparison of the polyconvex model (53) with experimental
 l I~  1 K~  1
W a ln 1   b ln 1  ; 53 data [37]. Material constants: l = 81.048 kPa, a = 2.060, b = 0.408,
4 a b v1 = 0.498, u = 47.59

123
8862 J Mater Sci (2007) 42:88538863

80 directions (18) and the factors v(r)

i , i = 0,1,...,n, related to
2 Piola-Kirchhoff Stress S, Sz [kPa] matrix and fibers (15) are given for some basic fiber
constellations.
60
Unidirectional alignment of one family of fibers leads to
transverse isotropy with respect to the fiber direction.
40 Setting n = 1 in (15)1 and insertion of (14)1, (15)2 leads in
circ. 0.75:1 view of (17)2 to
1:0.75
20 axial 075:1  
1:0.75
prediction circ.
~ r vr L0 vr L1 1 1  vr I vr L1
L
nd

0 1 1 1
prediction axial 3
0 1 
r b 2 
r b
0 0.2 0.4 0.6 0.8 1 1 2v1 L 1 1  v1 L 2:
Green-Lagrange Strain E, Ez
3 3

Fig. 4 Experimental data [37] and prediction by the polyconvex
model (53)
These are, however, considered negligible in comparison to
the variation in data among individual tissue samples.
In a fiber reinforced material, orthotropy may be the
result of different fiber configurations. We first consider the
case where fibers are aligned in three mutually orthogonal
fiber directions coinciding with the principal material
directions, so that li mi ; i 1; 2; 3. Then, in view of
(15,16,18) one obtains

3 h
X i
Conclusions ~r 1
L
r r r r b
1  v1  v2  v3 3vi L i:
3 i1
In the present paper we have presented a generalized A material with two orthogonal fiber families is
polyconvex and coercive strain energy function for fiber- likewise orthotropic, where the principal directions are
reinforced materials. The model is able to take into account given by the two fiber directions, so that l1 m1 and
the collagen fiber structure of soft biological tissues. The l2 m2 , and the direction normal to the plane in which the
strain energy function is given in a generalized form fibers lie. The generalized structural tensors (18) are thus
offering wide flexibility in application to different types of given by
tissue. For example, the model agreed appropriately with
the mechanical behavior of arterial tissue in various h i h i
~ r 1 1 2vr  vr L
L b 1 1 1  vr 2vr Lb2
experiments. The agreement was achieved with a small 3 1 2
3 1 2

number of material constants, some of which allow for a 1h r

i
r b
1  v1  v2 L 3:
physical interpretation. Further, the model adequately 3
predicted the stress response for loading conditions which
Two equivalent families of fibers (v(r) (r)
2 = v1 ) aligned in
had not been considered for parameter identification. As a
two arbitrary directions result in orthotropy. The principal
result of polyconvexity and coercivity the model is elliptic
material directions are given by the bisectors of the two
for all admissible strains and guarantees the existence of a
fiber directions and the normal to the plane in which the
solution of a boundary value problem.
fibers lie. The fiber directions can be expressed in terms of
As previously mentioned, the mechanical characteristics
the principal material directions by
vary among different kinds of tissue. For this reason, fur-
ther evaluation of the capacities of the model in application m1 cos a l1 sin a l2 ; m2 cos a l1  sin a l2 ;
to other tissue types and various test protocols might be
useful and is going to be conducted. where the angle between the fibers is 2a. Hence, the
generalized structural tensors read as
h i
Appendix ~ r 1 1 6 cos2 a  2vr L
L b1
1
3
1h i
r b 1h i
r b
Relations between the generalized structural tensors 1 6 sin2 a  2v1 L 2 1  2v1 L 3:
(15) and (18) 3 3
Finally, if there is a third fiber family i = 3 aligned
In the following, the relations between the weight factors normal to the plane spanned by the mechanically
w(r)
i , i = 1,...,n, associated with principal material equivalent fiber families, we have

123
J Mater Sci (2007) 42:88538863 8863

h i 16. Wilber JP, Walton JR (2002) Math Mech Solids 7:217

~ r 1 1 6 cos2 a  2vr  vr L
L b1 17. Chuong CJ, Fung YC (1983) ASME J Biomech Eng 105:268
1 3
3 18. Ball JM (1977) Arch Rat Mech Anal 63:337
1h r
i
r b 19. Schroder J, Neff P (2003) Int J Solids Struct 40:401
1 6 sin2 a  2v1  v3 L 2
3 20. Ball JM (2002) Geometry, mechanics, and dynamics. Springer,
1 h i New York, p 3
r r b
1  2v1 2v3 L 3: 21. Steigmann DJ (2003) Math Mech Solids 8:497
3
22. Itskov M, Aksel N (2004) Int J Solids Struct 41:3833
Additionally, superposition of several of the given cases 23. Itskov M, Ehret AE, Mavrilas D (2006) Biomech Model
may preserve the orthotropic material symmetry. Mechanobiol 5:17
24. Balzani D, Neff P, Schroder J, Holzapfel GA (2006) Int J Solids
Struct 43:6052
25. Itskov M (2007) Tensor algebra and tensor analysis for engineers
References with application to continuum mechanics. Springer, Berlin
26. Itskov M (2002) ZAMM 82:535
1. Humphrey JD (2002) Cardiovascular Solid mechanics: cells, 27. Truesdell C, Noll W (1965) Handbuch der Physik, vol III/3.
tissues, and organs. Springer, New York Springer, Berlin
2. Fung YC (1993) Biomechanics mechanical properties of living 28. Ogden RW (1984) Non-linear elastic deformations. Ellis
tissues, 2nd ed. Springer, New York Horwood, Chichester
3. Sacks MS (2000) J Elasticity 61:199 29. Zhang JM, Rychlewski J (1990) Arch Mech 42:267
4. Holzapfel GA, Gasser TC, Ogden RW (2000) J Elasticity 61:1 30. Spencer AJM (1984) Continuum theory of the mechanics of fibre-
5. Schmid H, Nash MP, Young AA, Hunter PJ (2006) ASME J reinforced composites. Springer Wien, p 1
Biomech Eng 128:742 31. Boehler JP (1977) ZAMM 57:323
6. Arruda EM, Boyce MC (1993) J Mech Phys Solids 41:389 32. Boyd S, Vandenberghe L (2004) Convex optimization.
7. Bischoff JE, Arruda EA, Grosh K (2002) J Appl Mech 69:570 Cambridge University Press, Cambridge
8. Kuhl E, Garikipati K, Arruda EM, Grosh K (2005) Mech Phys 33. Hill R (1968) J Mech Phys Solids 16:229
Solids 53:1552 34. Muller S, Qi T, Yan BS (1994) Ann Inst Henri Poincare Analyse
9. Lanir Y (1978) Biophys J 24:541 non lineaire 11:217
10. Kastelic J, Palley I, Baer E (1980) J Biomech 13:887 35. Ball JM (1996) Bull Amer Math Soc 33:269
11. Freed AD, Doehring TC (2005) ASME J Biomech Eng 127:587 36. Holzapfel GA, Sommer G, Gasser CT, Regitnig P (2005) Am J
12. Gasser TC, Ogden RW, Holzapfel GA (2006) J R Soc Interface Physiol Heart Circ Physiol 289:2048
3:15 37. Vande Geest JP, Sacks MS, Vorp DA (2004) ASME J Biomech
13. Holzapfel GA, Gasser TC, Ogden RW (2004) ASME J Biomech Eng 126:815
Eng 126:264 38. Treloar LRG (1975) The physics of rubber elasticity, 3rd ed.
14. Merodio J, Ogden RW (2002) Arch Mech 54:525 Oxford University Press
15. Merodio J, Ogden RW (2003) Int J Solids Struct 40:4707 39. Gent A (2004) Rubber Chem Technol 96:59

123
J Mater Sci (2007) 42:88648872
DOI 10.1007/s10853-007-1784-6
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Tubulin: from atomistic structure to supramolecular mechanical
properties
Marco A. Deriu Sren Enemark Monica Soncini
Franco M. Montevecchi Alberto Redaelli
Received: 15 December 2006 / Accepted: 17 April 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract Microtubules (MTs) are fundamental structural
elements in the cytoskeleton of eukaryotic cells. Their
unique mechanical properties depend on the properties of
the tubulin dimer, its interactions with surrounding dimers
and the geometric organization within the MT. While the
geometry has already been well described in experimental
works, the mechanical characteristics of the dimer as well
as of the individual monomers have up to date not been
described. These may therefore provide new, additional
insight to the microtubule tensile properties. In this paper
we construct a mesoscale model of MT with a bottom-up
approach. First, we evaluate the elastic constants of each of
the two monomers together with the interaction force be-
tween them by means of molecular dynamics (MD) sim-
ulations carried out in an explicit water environment. Using
the MD results, we model a 1 lm long MT as a cylinder
constituted by interacting elastic elements and examine its
properties via finite element method (FEM). The obtained
results show an elastic constant value for a-tubulin of
11 N/m, while for the b-tubulin the elastic constant was
measured to be 15.6 N/m. Concerning interactions between
neighbouring monomers, the elastic constants along the
protofilament (45 N/m for the intra-dimer interface and
18 N/m for the inter-dimer interface) are more rigid than
elastic constants calculated for lateral interfaces (11 and
15 N/m). The mesoscale model provides mechanical
properties of the whole MT, thus allowing the comparison
M. A. Deriu (&)
F. M. Montevecchi
Department of Mechanical Engineering, Politecnico di Torino,
Corso Duca degli Abruzzi, 24, Torino 10129, Italy
e-mail: marco.deriu@polito.it
S. Enemark
M. Soncini
A. Redaelli
Department of Bioengineering, Politecnico di Milano,
Milan, Italy
123
with data obtained by other previous experimental and
theoretical studies. We report here a Young modulus of
1.66 GPa for the MT under axial tension. In perspective
our approach provides a simple tool for the analysis of MT
mechanical behaviour under different conditions.
Introduction
Microtubules (MTs) are long, hollow cylinders made of
protofilaments which bind together laterally along the long
axis of the cylinder and form the microtubule wall [1]. The
elemental base of the protofilaments is the tubulin hetero-
dimer which consists of an a- and a b-tubulin monomer.
Both are globular molecules with a relatively high degree
of similarity from the primary to the tertiary structure level.
Their orientation in the dimer is almost the same, and
stacked together on top of each other as they are in the
protofilament they seem very alike [2]. However, the
structures do also have individual features that separate
them. One important difference is that in the MT the
a-monomer binds a GTP molecule while the b-monomer
binds a GDP molecule.
Different MT configurations exist depending on the
number of protofilaments included in the MT wall. The
number of protofilaments in a MT observed in-vivo and
in-vitro varies widely, from 8 to 19 [3], making the MT
structure highly polymorphic. Most cellular microtubules
have 13 protofilaments.
In microtubules the protofilaments bind together later-
ally with a slight shift generating a spiral with a pitch of 2,
3 or 4 monomers length [3]. For MTs with 3 start-helices,
the neighbouring monomers form spirals of all-a- and
J Mater Sci (2007) 42:88648872
all-b-tubulins except at the seam of the microtubule where
a-tubulins bind laterally to b-tubulins and vice versa. For
MTs with 2 or 4 start-helices, there is no seam and the
spirals consist of all-a- and all-b-tubulins.
Microtubules are essential structural elements present in
all eukaryotic cells. They form motor protein tracks ment models, but so far all these models have been based
which are used for directed transport within the cell and
they are part of the spindle apparatus used to move and
segregate the chromosomes during cell division. They also
play a purely mechanical role for the cell maintaining its
shape.
Despite the many experimental and computational
efforts the mechanical properties of the microtubule are still
debated. The experimental methods included optical twee-
zers [4, 5], hydrodynamics flow [6], atomic force micros-
copy (AFM) [7], and thermal fluctuation and viscosity
measurements [8]. The reported values of Youngs modulus
vary between 0.1 GPa [7] and 2.5 GPa [5]. Gittes et al. [8]
estimated a length independent MT flexural rigidity of
2.15
1023 Nm2 based on thermal fluctuations in shape.
From this value they calculated the Youngs modulus for
MTs to be 1.2 GPa using the assumption that the MT is both
homogeneous and isotropic in structure. The length of the
MTs in the experiments ranges between 25 lm and 65 lm.
MTs of this length are typically found in cilia.
It should be remarked that both the isotropy assumption
and the length independence are now generally questioned.
In a more recent experiment by Kis et al. [7] the elastic
deformation of MTs bound to a surface with holes of dif-
ferent sizes was directly measured with AFM. The AFM tip
was used to apply a force on the MT and the deflection was
measured and used to estimate simultaneously both
Youngs and shear moduli. The results show a significant
difference between the Youngs modulus (Y) of 100 MPa
and the shear modulus (G) of 1.4 MPa, evidencing that the
MT is highly anisotropic.
These results indicate that the MT flexural rigidity is not
length independent and consequently, it does not seem to
be an appropriate parameter to describe the MT mechanical
behaviour at least not for sufficiently short MTs (few lm
long MTs). For short microtubules the sliding between
adjacent protofilaments is relevant during MT bending. On
the contrary very long MTs (2565 lm) are more rigid and
the Youngs modulus dominates the mechanical behaviour
since only a slight sliding occurs between adjacent proto-
filaments in bending [9]. Thus, the value reported by Gittes
et al. [8] might result in a correct evaluation of the Youngs
modulus of the MT, since neglecting the shearing for
sufficiently long MT merely introduces small errors.
However, even at their best, these measurements are not
able to identify the origin of the shear and Youngs mod-
ulus values on the basis of the different forces between and
within the monomers. Therefore, they cannot provide the
8865
desired resolution at the single monomer level needed for a
thorough understanding of the MTs mechanical behaviour.
On the theoretical level several attempts have been
made to produce a bottom-up approach corresponding to
the one here presented. Such attempts include finite ele-
on assumptions on how the interactions vary with distance
[911]. In particular these works use reported experimental
values of Young modulus and shear modulus to define the
interactions between the mechanical properties of the
building blocks of the MT.
In this work we use molecular dynamics (MD) to obtain
mechanical characteristics of the different types of inter-
actions present in the MT both between and within
monomers. The characteristics are used to create a simple
MT model with finite elements methods (FEM) from which
we are able to evaluate the overall MTs mechanical
properties. In particular, we demonstrate that it is possible
to make a full description of the microtubule using MD
combined with the available experimental data on the
atomistic structure.
Methods
Molecular models and equilibration of the structures
for MD simulations
With the aim to characterize the tubulin monomers and all
the monomer-to-monomer interactions involved in the MT
structure, we used the atomic structure of ab-tubulin,
1TUB.pdb [2], available from the RSCB Protein Data Bank
[12]. Due to a large variation among monomer isotypes
present in the crystallographic assay a smaller part of the
C-terminal is missing in both monomers. Other pdb
structures exist [13] but were not taking into consideration
since a larger number of amino acids are missing despite
the fact that some of them have a better resolution.
The chosen structure was modified before use. A missing
Mg2+ ion present in other structures was added based on data
contained in the refined structure 1JFF.pdb [13]. Further-
more, a TaxolTM molecule was removed. TaxolTM is used as
a stabilizing agent for crystal formation of the microtubules
but is not present under physiological conditions.
Gromacs 3.3.1 software [14] with the GROMOS96 43a1
force field was used to perform all the simulations using
cut-offs of 1 nm for non-bond interactions (van der Waals
and electrostatic). The time step was set to 2 fs for all the
MD simulations.
The structure was first energy minimized in vacuum
before it was centred and orientated along the x-axis in a
rectangular box of size 1898 nm with periodic boundary
conditions. The rest of the box was filled with Single Point
123
8866
Charge (SPC) [15] water molecules to explicitly model
water in the system. To balance the strong negative charge
of the dimer and neutralize the total system charge 40 Na+
ions were added to the solution.
The entire system was then energy minimized again and
after that heated to 300 K by coupling it to an external heat
bath for 35 ps [16]. Finally the system was left to equili-
brate for 800 ps in order to stabilize the structure in terms
J Mater Sci (2007) 42:88648872
For each system, the interaction interfaces were identi-
fied and then created on the basis of the contact surfaces
described in literature [17]. After the set up of the relative
position of the two monomers, the water molecules and
ions were added and each system was heated and equili-
brated for 200 ps adding a position restraint in the form of
a harmonic potential (Vpr) to each Ca-atom:

1  2
of temperature and energy oscillations.
The equilibrated structure was the basis for all further
dynamics done for the mechanical characterization.
Mechanical characterization of interactions between
monomers
The mechanical characterization was done with MD simu-
lations in which the interaction energy between two
monomers was measured at different distances. On the basis
of the previously mentioned equilibrated structure four
different molecular systems were generated, each consisting
of two monomers: the intra-dimer (ab-interaction), inter-
dimer (ba-interaction) and lateral (aa- and bb-interaction)
complex (Fig. 1).
Fig. 1 Different interfaces in MT lattice: longitudinal interactions
involving ab-tubulin (a) and ba-tubulin interfaces (b) and lateral
interactions involving aa-tubulin (c) and bb-tubulin interfaces (d)
123
Vpr rCa;i = kpr rCa;i t  rCa;i 0 1
2
where kpr is the elastic constant of the harmonic potential
and rCa,i is the position of the Ca atom of the i-th residue.
For each molecular system several configurations were
generated with different distances (d) between the centres
of mass (CM) of the two monomers. A pulling MD pro-
cedure was used to generate different initial configurations
moving the monomers (Dd = dd0) along the line con-
necting the two CMs (Fig. 2) closer and apart with respect
to the initial position (d0).
For each distance the interaction energy term, i.e., the
sum of the Coulomb and van der Waals terms, was ob-
tained by running a MD simulation for at least 100 ps
while the protein backbone structure was maintained
restraining the monomer Ca-atoms. This prevents marked
structural changes in the monomer shape, which are con-
sidered in the single monomer testing.
During the MD simulation, the monomer-to-monomer
interaction energy and CMs distance were sampled every
0.2 ps. The last 50 ps are considered to determine the mean
value of interaction energy and CMs distance for each
configuration.
Fig. 2 Set up of the configurations with different monomer-
to-monomer distances
J Mater Sci (2007) 42:88648872 8867

The mean value of the interaction potential energy (Vint)
between the two monomers as a function of the mean value
of the CMs distance (d) was fitted with a 3rd order poly-
nomial function. Consequently, the force (Fint) can be
obtained as first derivative with the opposite sign of the
interaction potential energy with respect to the distance.
The elastic constant (kint) is calculated as second derivative
of the interaction potential energy.
Mechanical characterization of single monomers
The mechanical properties of a- and b-monomers were
tested using an AFM-like approach in which springs were
applied to deform the monomer (Fig. 3).
First, water and ions were removed from the equili-
brated structure and two independent systems consisting of
a- or b-tubulin were extracted. Single monomer systems
were then solvated by adding SPC water molecules and
neutralized with Na+ ions. Then the system was energy
minimized, heated to 300 K and equilibrated for 200 ps
while all the Ca-atoms of the monomer were restrained (see
Eq. 1).
Two pull groups (P and P) were defined, selecting
Ca-atoms of the residues involved in the longitudinal
interactions between adjacent monomers along the pro-
tofilament. A spring was connected to the CM of each
group and characterized by an elastic constant (ks) equal to
10 nN/nm. The spring elastic constant value was chosen on
the basis of a preliminary ks sensitivity analysis. The value
of ks = 10 nN/nm represents a good compromise between
low ks values that require very time consuming pulling
simulations and high ks values which allow faster simula-
tions but introduce large oscillations in the force values.
During MD simulation, the free end of each spring (S and
S) was moved with a constant velocity (vs) along the x-axis
corresponding to the MT longitudinal direction (Fig. 3).
Compression and elongation tests were carried out for 2 ns
MD simulations moving S and S with a velocity equal to
5
104 nm/ps.
The force F(t) applied to the molecule is given by:
F t ks jxS t  xP tj jxS0 t  xP0 tj; 2
where xS(t), xS(t), xP(t) and xP(t) are the CM positions of S,
S, P and P, respectively. The spring forces at the begin-
ning of the simulation are equal to zero since the free ends
of each spring and the CM of its corresponding pull group
have the same position.
The deformation e(t) in elongation and compression
tests was calculated as:
DLt Lt  L0
et ; 3
L0 L0
where L(t) is the monomer length as function of time
calculated as the distance between the CMs of groups P and P
Lt jxP t  xP0 tj 4
and L0 is the initial monomer length, i.e., for t = 0.
In all tests structure deformations of 10% were carried
out.
The elastic constant (km) of the monomer then is given
by the Hooks law:
F t
km m a; b 5
DLt
which corresponds to the slope of the F-Dl curve at the
time t.
MT mesoscale model
The mesoscale model of an entire microtubule was built
using a finite element (FE) approach. a-tubulin and
b-tubulin monomers and all the interactions between
monomers in the microtubule lattice were represented as
springs. The mechanical properties were set on the basis of
the results obtained with MD simulations. In this way the
microtubule was simulated as a network of springs.
In order to simplify the microtubule model, thus
reducing the number of elements along the protofilament,
two different springs k~ab and k~ba were defined representing
the mechanical behaviour of ab interface (ba interface)
together with the elastic properties of half of the related
monomers (Fig. 4a, b):

Fig. 3 Scheme of mechanical test for the monomer. The CMs of pull  1
1 1 1
groups (P, P) are each connected to a spring. Constant rate k~ab 6
displacements are imposed to the free end of the springs (S, S) 2ka 2kb kab

123
8868 J Mater Sci (2007) 42:88648872

total elongation of 10% was applied. During the simulation

(a) (b) k the displacement of all nodes are free to move only in the
2k
k ~ longitudinal direction.
/2 k
CM k
2k The data obtained by the FE model were used to cal-
CM ~
~
k k k
k culate the MT Young modulus (YMT) based on force value
/2 CM
2k ~ (FMT) obtained imposing an elongation (DLMT) of about
k
k k 0.1 lm. Thus, Youngs modulus was calculated using
/2 CM
2k Hooks law:
CM CM
k k~

/2 FMT LMT
2k YMT ; 8
k (c) DLMT AMT
/2
2k
CM
~ where LMT is the MTs original length and AMT is the MTs
k k
/2 cross sectional area.
2k To avoid the boundary effects only the central part of
k
/2 the microtubule was considered for data analysis. ABA-
2k
CM QUS/Standard version 6.6-1 (ABAQUS/Standard, Hibbitt,
/2 Karlsson & Sorensen) was used to perform the numerical
analyses at the mesoscale level.

Fig. 4 Scheme of the mesoscale models. (a) Extended model of three

protofilaments, where a-tubulin and b-tubulin monomers and all the
interactions between monomers in the microtubule lattice were
represented as springs. (b) Simplified model where the interactions
along the protofilament together with related half-monomers are
condensed in a spring element. (c) Lattice model of MT modeled as a
cylinder, where tubulin monomers and their interactions are modeled
as springs based on the simplified model (b)
 1
1 1 1 potential energy calculated for the original structure
k~ba 7
2ka 2kb kba
where ka and kb are the elastic properties of tubulin
monomers, while kab and kba are the intra- and inter-dimer
interaction, respectively.
The microtubule geometry was reconstructed starting
from literature data [3] concerning the permitted structures
of microtubules. A 1 lm long microtubule constituted by 10
straight protofilaments with 2 start-helices (corresponding
to a mismatch of 2 monomers) was built (Fig. 4c). In turn, a
displacement of 0.8 nm was set between adjacent filaments.
In this case, adjacent protofilaments face a sequence of aa
and bb lateral contacts. Concerning lateral interactions, the
behaviour of the lateral contacts was simulated by springs
with elastic properties described by kaa and kbb.
The MT, created as a cylinder, had a cross section
diameter of about 14 nm, based on data provided by MD
simulations. In particular, the position of the nodes of
spring elements along and between protofilaments was set
on the basis of the distance between the CM of a and b
monomers at the minimum of potential energy calculated.
The mesoscale model of microtubules was used to
simulate an axial test. The nodes of both extremities of the
MT were moved in axial direction symmetrically until a
Results
Equilibrated structure for MD simulation
The ab-tubulin structure refined starting from 1TUB.pdb
showed a potential energy of V = 2.15
106 kJ/mol after
energy minimization in solvated environment, lower than
1TUB.pdb (V = +5.28
1010 kJ/mol). A structure analysis
with PROCHECK at this point showed that 95.1% of the
residues fall in the permitted regions.
A decrease in potential energy from about 1.76
106
kJ/mol to 1.78
106 kJ/mol was observed in the first
200300 ps of equilibration, then the potential energy is
stabilized and the standard deviation of the potential energy
is reduced to 0.1%.
Mechanical characterization of interactions between
monomers
Four different interactions were investigated, namely the
ab-, ba-, aa-, and the bb-interactions. As explained earlier,
the ab- and the ba-interactions correspond to the mono-
mer-to-monomer interactions within the dimer and
between two dimers in the longitudinal direction, respec-
tively. The aa- and the bb-interactions correspond to the
two different types of lateral contacts.
Using the method described for interaction force cal-
culations, potential energy (Vint) vs. distance (d) relations
were obtained for all the considered molecular systems.
In all the considered configurations the Vint as a
function of time resulted stable after 50 ps as reported in

123
J Mater Sci (2007) 42:88648872 8869

1000 Table 1 Minimum interaction energies (Vmin) together with the dis-
tance between the CM of the two monomers, maximum forces of
t (ps) interaction (Fmax) between the two monomers occurring when they
0 d=5.78 nm are dmax apart, and elastic constants (kint) for the interaction evaluated
20 40 60 80 100
at the interaction energy minimum distance dmin
Vint (kJ/mol)

-1000 ab ba aa bb
d= 4 . 2 n m

Vmin (103 kJ/mol) 5.38 3.05 4.00 1.89

- 2 0 00
dmin (nm) 3.78 3.48 4.35 4.48
d=4.47 nm Fmax (103 pN) 11.9 6.0 6.3 4.1
-3 0 0 0
dmax (nm) 4.31 4.14 5.16 5.17
kint (N/m) 44.7 18.3 15.5 11.9
-4 000

Fig. 5 Interaction energy as function of time as measured at three

different distances (5.78, 4.20, 4.47 nm) for the bb-interaction
15

Fig. 5 for three of the different distances tested for the

10
bb-interaction.

Fint (pN x 103)

The averaged data points are reported in Fig. 6 and were
5
fitted with a 3rd-order polynomial approximating the ex- d (nm)
pected real energy behaviour as function of the distance 3.5 4.0 4.5 5.0 5.5 6.0
0
(d). Different potential energy minima (Vmin) together with

the corresponding distance (dmin) were obtained for each of
the four types of interaction as shown in Table 1. From the -5
polynomial expression obtained by the fit forces (Fint) as a
function of the distance (d) are reported in Fig. 7. The -10
minimum value of the force corresponds to the maximum
attractive force acting between the two monomers (Fmax) -15
occurring when they are a distance dmax apart (Table 1). Fig. 7 The curves in grey tone scale from dark to light show ab, ba,
The spring constants (kint) were also evaluated (Table 1). aa, and bb monomer-to-monomer interaction forces, respectively, as
function of distance. Curves are derivative of the fitted 3rd order
Mechanical characterization of single monomers polynomials (Fig. 6.)

Output data from AFM-like MD simulations consist of the

position along the x-axis of the CMs of the P, P, S and S blocks were used to reduce the large number of untreated
groups at 2 fs intervals. Averages over 1ps non-overlapping output data from the simulations.
For the two pull groups together with the springs free
1000
ends, the relative position with respect to the initial posi-
d (nm)
tion plotted as function of time is shown in Fig. 8. Due to
3.5 4.0 4.5 5.0 5.5 6.0
0 thermal motion the relative positions of P and P fluctuate
as the groups move apart. The elongation of each spring

Vint (kJ/mol)

- 1 0 00
over time is given by the difference between the position
- 2 0 00
of the free end of the spring and the corresponding pull

group.
- 3 0 00 The monomer contraction and elongation (Dl(t); Eq. 3),
and the force (F(t); Eq. 2), imposed by the spring were
- 4 0 00
calculated based on the filtered data from the simulation
-5 00 0 output. The curves showing F(t) versus Dl(t) for a-tubulin
and b-tubulin in both pulling and compression tests are
-6 00 0 reported in Fig. 9. Positive Dl(t) values imply an elongation
of the monomer while negative Dl(t) values correspond to a
Fig. 6 Interaction energy as function of distance between monomers.
Crosses: ab. Triangles: ba. Squares: aa. Diamonds: bb. The curves in
compression.
grey tone scale from dark to light show the fitted lines in the same A straight tendency line is superimposed on the
order F(t)Dl(t) curves for each monomer. Based on Eq. 5, the

123
8870 J Mater Sci (2007) 42:88648872

S 14
0.3
12
0.2

k (pN/nm x 10 3 )
P
10
0.1
8
x(nm)

0
100 200 300 400 500 600 6

-0.1 t (ps)
4

-0.2 P 2

-0.3
S -1.0 -0 .5 0 0.5 1.0 1.5
l(mn)
Fig. 8 Relative motion (x(t)) of the two pull groups (P, light grey; P,
dark grey) and the one free spring end as a function of time during the Fig. 10 Mechanical properties of the spring elements along the
a-tubulin elongation test (black) protofilaments calculated on the basis of MD data (k~ab (diamonds) for
the inter- and k~ba (squares) for the intra-dimer interactions)

4000 Pu l l i ng MT mesoscale characterization

F (pN)
3000 Mechanical properties of the spring elements representing
the single monomer deformations and all monomer inter-
2000
actions were defined based on the elastic behaviour ob-
1000
tained by MD. Figure 10 shows the curves used as the
input parameters, k~ab and k~ba , in the FE model for the
elastic behaviour along the protofilament.
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 The axial test force-elongation curve for a 1 lm long
- 10 0 0 MT elongated until 10% shows a linear behaviour with
l(nm)
slope 335 pN/nm.
- 20 0 0 The Young modulus (Y) was calculated assuming MT as
a hollow cylinder with a cross section of 206.5 nm2 based
-3 000
on force and displacement data at 10% of MT strain. The
cross-section was estimated setting a radial dimension of
-40 00
Compression 4.6 nm for tubulins disposed on a circumference with
14.3 nm of mean diameter. Under these hypotheses Y
Fig. 9 F(t)Dl(t) curve obtained for a-tubulin (light grey) and results to be equal to 1.66 GPa, which is in agreement with
b-tubulin monomers (dark grey). Tendency lines (black) are
data reported in literature [48].
superimposed on the curves and their slope represents the elastic
behaviour of the monomers

slope of the tendency lines provides the elastic constant
values ka = 11 N/m for the a-monomer model and
kb = 15.6 N/m for the b-monomer model.
The b-tubulin model seems to be slightly more rigid
than the a-tubulin model, despite the monomer similarity.
However, the value of km is affected by the total elongation
considered for the analysis. In particular, considering a
strain of the monomer up to 5% in elongation and com-
pression the elastic constant value for both a- and b-tubulin
is in the range 1516 N/m.
Discussion
The aim of the present paper is to increase the under-
standing of the MTs mechanical behaviour starting from
available knowledge about the dimers molecular structure
and the geometry of the MT lattice. A new bottom-up
hierarchical approach was developed. The basic building
blocks and their interaction interfaces were identified con-
sidering the structure of the MT as reported by electron
crystallography analyses [2, 13]. The mechanical properties
of the building blocks and their interactions were calculated

123
J Mater Sci (2007) 42:88648872
through MD simulations. The results together with the data
on the permitted MT lattice structure obtained from litera-
ture [3] were used as input for a FE model. Using advanced
FE structural code, a spring elements model was used to
investigate the overall MT behaviour under axial tension.
The results obtained are directly comparable to results from
MT experiments [48]. In contrast, no direct experimental
measurements on the dimer are available for comparison.
Previously, experimental works have estimated the dimers
mechanical properties starting from the MT structure [7, 8].
Apart from the difficulties in separating the dimers
mechanical properties from the lateral and inter-dimer
interactions, these experiments cannot provide detailed
knowledge about the individual MT building blocks and
their specific interactions. The present work in contrast
takes the investigation of the MT to the molecular level.
Also our model has some limitations. The single
monomer AFM-like measurements are done at a particular
velocity. However, globular protein structures are known to
have viscoelastic properties, consequently the strain rate
imposed can influence the obtained results. Strain rates
imposed in typical AFM pulling/compression experiments
on proteins are in the order of 109 nm/ps [18, 19], while
we impose a velocity of 5
104 nm/ps.
However a preliminary analysis aimed at evaluating
how the velocity influences the obtained elastic constant
values (data not published) resulted in no substantial dif-
ferences between force values (and thus in elastic constant
values) calculated at a strain rate of 103 nm/ps with
respect to force values calculated at 5
104 nm/ps.
The atoms chosen for the pull groups and the use of the
position restraints introduce some arbitrary choices that
require further discussion. Different parts of the mono-
mers surface structure have different mechanical proper-
ties, and this is also valid for different parts of the pull
group which contains loops and regions weaker than
others. Without taking this into account and without a
proper pull group definition a pulling approach would only
report the mechanical behaviour of the weakest region of
the surface. In our work we try to estimate an average
behaviour of the whole monomer by maintaining the
interface surface fixed. The position restraints used serve
precisely this purpose.
Our approach differs from experimental AFM mea-
surements [18] and simple AFM simulations with the
spring attached to a single atom [19, 20]. In these studies
force values are typically of order 100 pN, while we report
values of about 1 nN. Preliminary tests demonstrated that
the choice of the Ca-atoms of residues used to set the pull
groups (P and P) influences the output force values. In
particular, moving the entire interface instead of a single
atom (as in typical AFM simulations) results in higher
calculated force values.
8871
Limitations in the method used to obtain the interactions
between monomers include the positioning and interaction
path used. Different permitted MT configurations like 10:2
or 13:3 should ideally be constructed with different posi-
tions and interaction paths of the monomers. Simplifying,
we study only the interactions along two perpendicular
paths: along the protofilament and laterally. In correspon-
dence, the original structure, 1TUB.pdb, taken from the
RCSB Protein Data Bank was obtained from protofilaments
in the 2-dimensional planar zinc-sheet configuration. A full
investigation in 3 dimensions could be interesting for a
later model refinement.
In order to measure the interaction energy a novel
approach was used rather than a constrained potential of
mean force (PMF). Preliminary docking tests demonstrated
that using a distance constraint method or a harmonic
umbrella potential acting on the CMs of the monomers has
limitations in our case due to the globular shape and size of
the investigated molecules. Indeed when arranging the two
monomers very close to each other (e.g., with a harmonic
umbrella potential acting on each monomers CM) their
structures, squeezed by the high repulsion forces, widely
deformed during the simulation. In this way the real dis-
tance between the outer surfaces of the two docked
monomers cannot be controlled.
Our approach, which restraints each Ca-atom (resulting in a
backbone restraint) allowed the structure of each monomer to
maintain its shape throughout the whole simulation while
lateral groups of each residues on the surface moved and
rearranged their positions. The spring constants describing the
interactions between the monomers correspond with knowl-
edge about the interactions as described in literature [2, 17]
and with experimental evidence concerning the dynamic
instability of the MT structure [21]. The general opinion is that
during the MTs disassembly, protofilaments break the lateral
contacts between the tubulin dimers and peel off from MTs
structure. Subsequently, free protofilament segments then
break at the inter-dimer interfaces splitting into ab-tubulin
dimers. On the contrary, the ab-tubulin dimer is known as a
very stable structure and can be dissociated only by means of
severe chemical treatments [1]. In accordance, our
results show that lateral (Fmax = 6.3 nN for aa-interaction
and Fmax = 4.1 nN for bb-interaction) and inter-dimer
(Fmax = 6.0 nN) interactions are much weaker than the
interaction between the monomers within the dimer
(Fmax = 11.9 nN). In addition, we note that no substantial
difference is found between the values measured for lateral
and inter-dimer interactions. Not surprising the energy needed
to pull the two monomers within the dimer apart (ab-inter-
action) is much greater than the energy needed to pull any of
the other dimers apart (Table 1).
It is important to observe that the force-distance relations
reported in Fig. 7 were calculated based on interaction
123
8872
potential energy curves fitted with a 3rd order polynomial 3.
function. With this approximation, the force-distance 4.
5.
curves are valid around the minimum value and are not 6.
applicable far from this minimum value. In particular, it is 7.
clear that there is no repulsion force when the monomers are
far apart, and the force should be go zero for large distances.
8.
The mesoscale model here developed is based on a
simple representation of the MT structure. Indeed, the 9.
monomers deformations and interactions are accounted
for by means of mono-dimensional elements. The meso- 10.
11.
scale model of the MT does not include thermal fluctua- 12.
tions. However considering the persistence length of the
MT about 16 mm [1] the bending effect of thermal fluc-
tuations on 1 lm long MT can be neglected. 13.
The FE analysis of a 1 lm long MT under axial tension 14.
allowed calculating mechanical characteristics of an entire 15.
MT. The Young modulus obtained using our bottom-up
approach is 1.66 GPa in agreement with data reported in
literature. Up to date, all the experimental works in the past 16.
produced a wide range of values for mechanical charac-
teristics of the microtubule (0.12.5 GPa) [48], and the 17.
most of these studies reported a Young modulus of
18.
12 GPa.
19.
Acknowledgments This research has been supported by the EST
Marie Curie programme contract number MEST-CT-2004-504465
and by the Active Biomics STREP project contract number NMP4- 20.
CT-2004-516989.
21.
References
1. Howard J (2001) In: Mechanics of motor proteins and the cyto-
skeleton. Sinauer, Sunderland, pp 119134
2. Nogales E, Wolf SG, Downing KH (1998) Nature 391:199
123
J Mater Sci (2007) 42:88648872
Chretien D, Wade RH (1991) Biol Cell 71:161
Felgner H, Frank R, Schliwa M (1996) J Cell Sci 109:509
Kurachi M, Hoshi M, Tashiro H (1997) Cell Mot Cyt 38:201
Venier P, Maggs AC, Pantaloni D (1994) J Biol Chem 269:13353
Kis A, Kasas S, Babi B, Kulik AJ, Benot W, Briggs GAD,
Schonenberger C, Catsicas S, Forro L (2002) Phys Rev Lett
89:248101
Gittes F, Mickey B, Nettleton J, Howard J (1993) J Cell Biol
120:923
Kasas S, Kis A, Riederer BM, Forro L, Dietler G, Catsicas S
(2004) Chem Phys Chem 5:252
Janosi IM, Chretien D, Flyvbjerg H (2002) Biophys J 83:1317
VanBuren V, Cassimeris L, Odde DJ (2005) Biophys J 89:2911
Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN,
Weissig H, Shindyalov IN, Bourne PE (2000) Nucl Acid Res
28:235
Lowe J, Li H, Downing KH, Nogales E (2001) J Mol Biol
313:1045
Lindahl E, Hess B, van der Spoel D (2001) J Mol Mod 7:306
Berendsen HJC, Postma JPM, van Gunstersen WF, Hermans J
(1981) In: Interaction model for water in relation to protein
hydration. Reidel Publishing Company Dordrecht, Dordrecht, p
331
Berendsen HJC, Postma JPM, DiNola A, Haak JR (1984) J Chem
Phys 81:3684
Nogales E, Whittaker M, Miligan RA, Downing KH (1999) Cell
96:79
Lenne PF, Raae AJ, Altmann SM, Saraste M, Horber JKH (2000)
FEBS Lett 476:124
Altmann SM, Grunberg RG, Lenne PF, Ylanne J, Raae AJ,
Herbert K, Saraste M, Nilges M, Horber JKH (2002) Struct
10:1085
Soncini M, Vesentini S, Ruffoni D, Orsi M, Deriu MA, Redaelli
A (2006) Biomech Model Mechanobiol Nov 18;[Epub ahead of
print]
Erickson HP, Stoffler D (1996) J Cell Biol 135:5
J Mater Sci (2007) 42:88738884
DOI 10.1007/s10853-007-1792-6
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Analysis of deformation coupled surface remodeling in porous
biomaterials
Partha Ganguly Franz-Josef Ulm
Received: 12 December 2006 / Accepted: 20 April 2007 / Published online: 7 August 2007
Springer Science+Business Media, LLC 2007
Abstract Surface remodeling of biological tissues
through tissue growth or dissolution is deemed critical to
their proper functioning, and is influenced by the defor-
mation of the tissues during physiological activities. The
present work attempts to develop a constitutive framework
for deformation modulated surface remodeling of biologi-
cal tissues. The framework is developed assuming finite
deformation of the tissue, and the effect of deformation on
the driving force for surface remodeling is determined from
thermodynamic principles. The microscopic trends are
upscaled to yield the remodeling-induced change in a
macroscopic porous tissue. By way of application, the
effect of deformation on the remodeling kinetics is deter-
mined for an incompressible elastic tissue. Depending on
the ratio of the specific elastic stiffness and the specific
Gibbs energy variation induced by the cell, the effect of
deformation on the remodeling kinetics can be significant.
It is found that both tensile and compressive deformation
aid tissue dissolution (and dissuade growth). However, the
magnitude of the effect is found to be different under
tensile and compressive loadings, and critically depends on
the reference frame used for the strain measurements. For
Lagrangian strain measures (e.g., stretch, engineering
strain), the increase in the dissolution kinetics per unit
strain is higher under compressive loadings. On the other
hand, for Eulerian strain measures (e.g., logarithmic or true
strain), the effect of tensile loading on the dissolution
P. Ganguly
Schlumberger Doll Research, Cambridge, MA 02141, USA
F.-J. Ulm (&)
Massachusetts Institute of Technology, 77 Massachusetts Ave.,
Cambridge, MA 02139, USA
e-mail: ulm@mit.edu
kinetics is higher. This reinforces the need for proper ref-
erence frame definition for experimental strain measure-
ments.
Introduction
Many biological tissues continuously undergo remodeling
(growth or resorption) during their natural life spans. For
example, about 25% of trabecular bone and 3% of cortical
bone in a mature adult is remodeled annually. The tissue
remodeling phenomena has important consequences, and is
deemed critical for a number of physiological events
including the healing of bone fracture [13], wound healing
in the skin [4, 5], hypertrophy of the heart (i.e., size in-
crease in muscles of an overloaded heart, [6]) and changes
of pulmonary blood vessels in hypertension [7]. The tissue
remodeling mechanism and kinetics are affected by the
tissue deformation during the remodeling process, e.g., the
contractile forces generated during skin wound contraction
is associated with scar tissue formation [8], and the
importance of in-vivo stress in bone healing is generally
accepted, though the exact nature of the effect is uncertain
[1, 9]. Thus modeling of deformation-coupled remodeling
processes in biological tissues is important.
The interaction of the remodeling characteristics and the
stress/deformation in biological tissues have been numeri-
cally examined by several investigators [1013]. Skalak
[10] showed that residual stress may be developed due to
non-uniform (and incompatible) growth in biological tis-
sues. Hoger and co-workers [1113] have modeled defor-
mation coupled growth by introducing a multiplicative split
of the total deformation gradient (i.e., due to both
123
8874
deformation and growth) into its elastic deformation and
growth parts, and have analyzed problems related to the
hypertrophy of the heart and growth in bones under stress.
The analysis is similar to the multiplicative split of the
elasto-plastic deformation gradient into the elastic and
plastic components in conventional finite deformation
elasto-plastic constitutive analysis (e.g., [14, 15]), and
estimate the growth phenomenon in terms of its effects on a
given set of material points (i.e., the analysis is based on a
material reference frame). On the other hand, the remod-
eling phenomenon involves growth or resorption of the
biological tissue, and thus, implicitly involves a change in
the number of material points (increase for growth,
decrease for resorption), and a continuously variable set of
material points. Hoger and co-workers have resolved this
anomaly by assuming growth to be volumetric and dis-
tributed, i.e., no volume element is destroyed or created
during the growth process, rather the existing volume
elements are extended or compressed as new material
points are added or some material points disappear. Such
an approach is effective for volumetric remodeling pro-
cesses in biological tissues. However, in surface remodel-
ing, the appearance of new material points (or the
disappearance of old ones) determine the characteristic of
the newly formed surface, and the growth process is spa-
tially discontinuous. Such situations arise in a number of
biomaterial remodeling scenarios, including trabecular
bone growth/resorption under osteoblast/osteoclast cell
action at the bone surface and wound healing in skin tissues
(which involve the adhesion of the fibroblast cells at the
surface of the skin matrix). The applicability of the mul-
tiplicative split in such situations is questionable.
Another characteristic of the trabecular bone/skin
remodeling problem deserves mention at this point. Both
bone and the skin are porous tissues, composed of struts
and pores (see Fig. 1a). The remodeling phenomenon in
Fig. 1 (a) The porous tissue at the macroscale, consisting of solid
struts and fluid saturated pores. The biological cells are shown
attached to the solid. (b) Schematic representation of the microscale,
where the multinucleated biological cells may attach at the surface of
the solid strut, and initiate the remodeling phenomena. The surface
123
J Mater Sci (2007) 42:88738884
these materials can be conveniently divided into two length
scales: (a) the microscale (see Fig. 1b), which is typically
at the length scale of individual struts or pores for the
porous structures, and the relevant scale for biochemical
phenomena related to remodeling processes, and (b) the
macroscale, which is at the scale length of the entire porous
tissue, and is the relevant length scale for determining the
effect of remodeling on tissue characteristics. Typically the
length scales corresponding to the micro and the macro
scale in porous biomaterials differ by one or two orders of
magnitude. For example, in porous trabecular bone, the
microscale corresponding to the size of individual trabec-
ula and individual pores are of the sizes 100500 lm [16,
17], while the porous structure, corresponding to the scale
of the trabecular bone samples generally used in experi-
ments, is 10100 mm in size [18]. Similarly, in problems
of growth in skin tissues, the length of individual struts and
the pore diameter is approximately 100 lm [5], while the
scaffolds used in experiments are 1100 mm in size [5]. It
is noteworthy that in both cases, the material addition (or
removal) occurs at the lower scale length. The size of the
osteoclast cells responsible for resorption in bone is 50 lm
[19], while the fibroblast cells involved in skin growth are
20 lm in size [5]. In order to predict the effect of
remodeling (which occurs at the microscale) on the mac-
roscopic tissue characteristics, consistent upscaling
schemes need to be developed. Such upscaling may be
inconvenient when a multiplicative split of the overall
deformation gradient is assumed, since the estimation of
the macroscopic deformation and growth components of
the deformation gradient from the microscopic measure-
ments is not straightforward.
The focus of the present work is the analysis of the
remodeling phenomenon (at different length scales) in
porous biological tissues. The problem arising from the
variability of the material points in such systems is
area of the solid strut is As, n is the normal to the solid surface, the
solid-cell interface area is f and u is the velocity. A negative u:n at the
solid-cell interface signifies dissolution of the solid and a positive u:n
signifies growth
J Mater Sci (2007) 42:88738884
circumvented by developing the theory in terms of a spatial
coordinate system. The model is developed both w.r.t. the
undeformed reference frame (i.e., at time t = t0), and the
deformed (or current) reference frame.The present model
follows the analytical lines earlier employed by Silva and
Ulm [2] for bone resorption, Lemarchand et al. [20] for
mechanically induced dissolution processes, and Ulm et al.
[21] for calcium leaching. However, the present work treats
the biochemical phenomena at finite strains (whereas
infinitesimal strain assumptions were used in the above
studies) in order to develop a model for biological tissue
remodeling, which may involve small deformations for
hard tissues (e.g., bone) and relatively large deformations
for soft tissues (e.g., skin).
Remodeling analysis at the microscale
Problem definition
The representative volume element (r.e.v) at this scale
contains a solid domain Vs and a cell attached along the
cell-solid interface f (see Fig. 1b). The mass transfer takes
place between the solid surface and the cellular fluid
around it, by deposition or dissolution of solid mass from
or into the cell fluid. The cell fluid is considered to be a
mixture (at pressure p) of a solvent and a solute (partial
pressure pi), generated by the biochemical activity of the
cell. The problem is defined in terms of the spatial coor-
dinates, in order to circumvent the difficulty of defining a
suitable material coordinate system for a problem with
continuous mass (and material points) addition or removal
at the cell-solid interface. Since concurrent deformation
and growth processes are assumed, the attachment or
removal of the material points occur in the deformed solid
configuration. Thus, the remodeling problem is initially
defined in terms of quantities measured in the deformed
configuration (i.e., current reference frame) x, and subse-
quently suitable transformation equations are used to
express these quantities in the original (i.e., undeformed)
reference frame, x.
Figure 2 shows the evolution of several field variables
across the solid-cell interface (x = n). The solid deforms
under the applied loading (external stress and fluid pres-
sure), and simultaneously undergoes mass addition (or re-
moval) at the solid-cell interface, which in turn changes the
volume of the solid. In the current configuration, the mass
addition (or removal) at the interface may be represented in
terms of velocity jump at the solid-cell interface. The
particle velocity in the solid bulk (ud ) is defined

by the
microscopic strain rate in the solid, d 12 gradud deforming material (without any growth/resorption) are
grad t ud , where grad t ud is the transpose of the given below.
8875
gradient of the ud vector (i.e., gradud oud =ox). This
defines the particle velocity at any spatial point inside the
solid (approaching the solid-cell interface), and un
ud
(see Fig. 2a). On the other hand, the solid-cell interface
velocity is defined by the rate of mass addition to (or
removal from) the solid as well as the solid strain rate. Let
the velocity component due to the mass addition/removal
be uc , defined such that uc  n > 0 (n is the outward
normal in the current configuration to the solid-cell inter-
face, see Fig. 2) for addition, and uc  n\0 for mass
removal. The velocity of the solid-cell interface un is
defined by the sum of the two components uc and ud ,
un ud uc . This results in a discontinuity in the
velocity field at the solid-cell interface, and the jump in
velocity u is defined by
u un
un
uc 1
In case of density, the discontinuity occurs across the
solid-cell interface. The solid density at the solid side is
defined by solid mass per unit volume qs t, while that at
the cell side is the fluid mass per unit volume, qf t.
qt qn ; t
qn
; t qf t
qs t 2
In the present work, the change in the solid density
across the interface is also significant. The solid density
jumps from qs t on the solid side to 0 in the fluid.
qs t qs n ; t
qs n
; t
qs t 3
Similarly, the jump in the free energy mass-density (free
energy per unit mass) across the interface is defined by the
difference in the free energy mass-density in the cell fluid
(wi) and the solid (ws).
w wn
wn
wi
ws 4
The discontinuities in the density and the free
energy mass-density are schematically shown in
Figs. 2b and c.
Preliminaries
The focus of the present work is the analysis of deforma-
tion coupled remodeling in porous biological tissues, and
the analysis is conducted in both the initial (i.e., unde-
formed) and the current (i.e., deformed) reference frames.
The analysis thus involves repeated transformation of
quantities between the two reference frames. The equations
for volume, density and surface area transformations for a
123
8876 J Mater Sci (2007) 42:88738884

(a) (b) (c)

u = ud + uc =s =s

n
n n
solid ( Vs ) cell ( Vf ) sosliodli(dVs ) cell c(eVlfl) sosliodli(dVs ) ceclle(lV
l f)

=f =f
u = ud

- + - + - +

Fig. 2 Schematic representation of the evolution of (a) velocity, (b)
density and (c) free energy per unit mass across the solid-cell
interface (x = n) in the current configuration. The vertical lines
corresponding to x = n and x = n+ are infinitesimally distant from the
interface, and lie in the solid domain (Vs) and the cellular domain (Vf),
respectively. The outward normal to the solid-cell interface is
represented by n. The area of the solid-cell interface is f. The
primary purpose of this figure is to show the jump in the field
quantities across the solid-cell interface. The field quantities have
been shown constant in each of the domains (Vs and Vf) for
convenience. No such assumptions have been employed in the
analysis

If vs t0 is the original (undeformed) infinitesimal vol- fields that exhibit discontinuities within the given domain.
ume of a solid, then the volume vs t after deformation is For example, if the field x exhibits a discontinuity at R
given by within the domain V of perimeter A, then
Z Z Z
vs t jvs t0 5
x  nA dA divxdV x  nR dR 9
A V R
where, j is the determinant of the deformation gradient f ,
defined in terms of the undeformed (X) and deformed (X) where, nA and nR are outward normals at the boundary A
reference frames (f Gradx ox=oX). For a solid with and the surface of discontinuity R, respectively, and x is
constant mass, the density is inversely proportional to the the jump in the field x across the surface of discontinuity.
volume, and the densities in the two reference frames are
related by Mass and volume conservation

qs t 1=jqs t0 6 The mass of solid (considering the solid domain, Vs in

Fig. 2) is defined by Ms, and is related to the solid density
The rate of change of the volume (in terms of quantities qs t and the solid volume vs t1 by
in the undeformed configuration) can be expressed by
Z Z
differentiating Eq. (5) w.r.t. time, and realizing that the
volume in undeformed configuration vs t0 is independent Ms t dms t qs tdvs t 10
Ms t Vs t
of time (i.e., dvs t0 =dt 0).
The rate of change of mass is given by
dvs t dj
vs t0 7 Z Z
dt dt dMs t dqs t
dvs t qs tu  ndas t
dt Vs t dt As t
If an infinitessimal surface area das t0 was oriented
with an unit normal n in the undeformed solid, the corre- 11
sponding area das t will be oriented with the unit normal n where As t is the area of the solid periphery in the
in the deformed solid deformed configuration and n is the outward normal to the
solid periphery. The velocity field u is discontinuous in the
ndas t jf
1 t  Ndas t0 8 domain Vs t while the density field qs t is continuous

where f
1 t is the inverse transpose of the deformation

gradient f .
Another important concept that is repeatedly applied in
the present work is the generalized divergence theorem,
which deals with the application of divergence theorem to
1
In this work, the infinitesimal variables (e.g., dvs t) and the inte-
gration domains (e.g., Vs t) have been differentiated through use of
different letters or symbols. In some cases, the same symbols have
been used to represent the physical quantities at both the micro and
the macro-scales.

123
J Mater Sci (2007) 42:88738884 8877

(see Eq. (1) and Fig. 2). Thus using the generalized Z Z
dVs t
divergence theorem (Eq. 9) on the solid domain, the rate of divud dvs t uc  ndac t 17
dt Vs t ft
change of the solid mass can be expressed by
Z   The first term reflects the change in volume due to the
dMs t dqs t d deformation undergone by the solid, while the second term
qs tdivu dvs t
dt Vs t dt signifies the corresponding volume increase (or decrease)
Z due to biological growth (or dissolution) at the solid surface.
qs tu  ndac t 12 The two activities thus can be clearly demarcated in the
ft
deformed configuration. This demarcation is less clear in
where ft is the solid-cell interface area in the deformed the undeformed configuration, as evident from the volume
configuration (the infinitesimal solid-cell interface area is change rate derived applying Eqs. (7, 8, 16) to Eq. (17).
given by dac t) and n is the outward normal (outward Z Z
from the solid) at the interface. The discontinuity of the dVs t dj
dvs t0 u c  Ndac t0
je 18
velocity field u is defined by uc (see Eq. 1). The above dt Vs t0 dt ft0
equation contains three terms. The first two terms
dqs t=dt and qs tdivud integrated over the solid In contrast to Eq. (17), the second term in Eq. (18),
volume vs t describe the mass conservation in a reflecting undeformed measures, contains both the bio-
deforming solid. The last term defines the mass loss or chemical (e u c  N) and the mechanical (j detf ) contri-
addition at the solid-cell interface. Considering mass butions. The coupling arises due to the fact that Eq. (18)
conservation in the solid bulk (i.e., neglecting diffusional measures the rate of change of current volume (i.e., volume
mass exchange between the growing/dissolving solid after coupled deformation and mass change) in the unde-
surface and the solid bulk), i.e., formed reference frame, and must account for the change
in density due to deformation. It may be convenient to
Z  
dqs t introduce the (Lagrangian) mass rate per unit (undeformed)
qs tdivud dvs t 0 13 surface involved in the deposition or dissolution process,
Vs t dt
m u c  N. Use of m

c qs t0 e
c in Eqs. (15, 18) yield the
the mass gain (or loss) is only due to the surface Lagrangian expressions in the form
contribution. Z
dMs t
Z
c dac t0
m 19
dMs t dt ft0
qs tuc  ndac t 14
dt f t
Z Z
dVs t dj
c
m
Since the rate of mass change is invariant with the dvs t0 j dac t0 20
change in configuration, the rate in the deformed and dt Vs t0 dt ft0 qs t0
undeformed configurations can be equated
Z Z
dMs t
qs tuc  ndac t u c  Ndac t0
qs t0 e Thermodynamic formulation
dt f t ft0

15 We are interested in the thermodynamic evolution, in an

c infinitessimal time interval dt, of a material domain, which at
where qs t0 ; ue and n are the solid density, the velocity of
the solid-cell front, and the normal to the solid-cell front, in
the undeformed configuration. The growth/dissolution
velocities in the undeformed and the deformed
configurations may be related by applying Eqs. (6, 8) to
Eq. (15).
uec f
1  uc
The rate of change of the solid volume in the deformed
configuration can be derived from Eq. (12) by letting
qs t 1:
time t coincides with the solid domain Vs t, and which at
time t + dt is composed of the solid domain Vs t dt and of
the solute in dV, which corresponds to the solid that dissolved
in the cell fluid. These thermodynamic evolutions are sub-
jected to the Clausius-Duhem inequality which states that for
all processes the rate of change of Helmholtz free energy (W)
is equal to (for reversible processes) or less than (for irre-
16 versible) the rate of external work provided to the system
(Pext), and results in a non-negative rate of dissipation D.
dW
D P ext
0 21
dt

123
8878 J Mater Sci (2007) 42:88738884

Z Z  
In the deformed configuration, the external work rate is  jp pi
the sum of two terms. The first term is the work rate P ext f : pdvs t0

m
c dac t0
Vs t0 ft0 qs t0 qi
developed by the traction field t r  n (with r the Cauchy
25
stress tensor and n the outward normal to the solid surface)
and the total solid velocity u (composed of velocities due to 
where f is the rate of change of the deformation gradient,
deformation and chemical growth/dissolution) along the
and p jr  t f
1 is the Boussinesq (or first Piola-Kirch-
solid surface in the deformed configuration As(t)
hoff) stress tensor.
Z Z The change of the Helmholtz free energy of the system
Pext u u  r  ndas t divud  rdvs t under consideration is due to the change of the free energy,
As t Vs t
Z between t and t + dt, in the solid bulk and along the surface
uc  tdac t 22 of discontinuity. Expressed in terms of the specific free
ft energies, this change in the deformed reference frame is
where the generalized divergence theorem Eq. (9), and Eq. given by
(1) have been used. The last term in Eq. (22) represents the Z Z
dW dws
work rate pdV, involving the fluid pressure p on the boundary qs t dvs t
wi
ws qs tuc  ndac t
dt Vs t dt ft
of the considered domain and the dissolved solid volume dV,
which is obtained by letting t
pn in Eq. (22) 26

Z Z where wi and ws are the free energy per unit mass of the
Pext u r : ddvs t
puc  ndac t 23 solute ions (in the cell fluid) and the solid, respectively.
Vs t f t While the first term in Eq. (26) is classical, the second
term expresses the spontaneous change at the solid-cell
where d is the (symmetric) Eulerian strain rate tensor, interface, induced by w wi
ws which can be
obtained from developing divud  r gradud : r, while interpreted as the discontinuity in Eq. (4) in free
considering divr 0 and r rt in Vs t. The second energy per unit mass between the solute in the cell
contribution to P ext of the considered system arises from fluid (wi ) and the solid phase (ws). The Lagrangian
the introduction (or extraction) of the solute mass dMi into counterpart of Eq. (26) is readily obtained using Eqs. (5,
the cell-fluid, which, given mass conservation of the sys- 6, 15, 19)
tem under consideration, equals with opposite sign the
change in solid mass (given by Eq. 15), i.e., Z Z
dW dw
dMs =dt
dMi =dt
qi dVi , where qi is the mass density qs t0 s dvs t0
mc dac t0
wi
ws

dt Vs t0 dt ft0
of the reactant ions in the cell-fluid (assumed for purpose of
clarity constant in dV). The resulting work contribution is 27
the work rate Piext pi dVi developed by the partial pressure
of the Rsolute pi along the rate of change of solute volume Finally, using Eqs. (24, 26) in Eq. (21) yields the
dVi ft qs t=qi uc  ndft. The total external work Eulerian expression of the dissipation rate
rate in the deformed reference frame thus reads Z  
dws
Z D r : d
qs t dvs t
Vs t dt
P ext Pext u Piext r : ddvs t Z  
p
Z 
Vs t
 gi
ws
q tuc  ndac t  0 28
p pi f t qs t s


q tuc  ndac t 24
ft qs t qi s where gi wi pi =qi is the Gibbs potential or free mass
enthalpy (per mass unit) of the solute (ions in the solution
In the undeformed configuration, the external work of the cell fluid). Analogously, Eqs. (25) and (27) yield the
terms can be evaluated through suitable transformation of Lagrangian counterpart
Eq. (24), using Eqs. (6, 8, 15, 19)2
Z  
 dws
D f : p
qs t0 dvs t0
Vs t0 dt
Z " #
2
Using grad: Grad:  f
1 and p jr  t f
1 , divud  r jp
h i h i   gi
ws

c dac t0  0
m 29
1
j Gradu  f
d
1
: p t f 1j f : p : Given that dvs t jdvs t0 , f t 0 qst0
R R  
d
Vs t divu  rdvs t Vs t0 f : p dvs t0 .

123
J Mater Sci (2007) 42:88738884
The first integral (the volume integral) in these two
expressions of the Clausius-Duhem inequality represents
the intrinsic dissipation rate in the solid bulk. It is of the
standard format classically employed in continuum
mechanics. Assuming elastic deformation in the solid bulk,
and thus zero dissipation, the terms simply signify that the
rate of change of elastic energy in the bulk is governed by
the stressstrain work. More generally, in the absence of
bio-chemical processes at the solid-cell surface (i.e.,
uc  n 0 , m
c 0), it represents the amount of exter-
nally supplied energy which is not stored in the solid
microstructure, but dissipated into heat form. Given its
intrinsic nature, associated with solid deformation, it may
be assumed to be non-negative irrespective of the phe-
nomena at the solid-cell surface. Consequently, the dissi-
pation associated with the resorption process which is
captured by the surface integrals in Eqs. (28) and (29),
must be non-negative as well, i.e.,
" #
p A0 lBGP
ls . Thus, lBGP
ls > 0 is expected to
uft gi
ws
qs tuc  n  0 30
qst
" #
jp
uft0 gi
ws
uc  N  0
qs t0 e 31
qst0
In contrast to the volumetric dissipation in the solid
bulk, the dissipation rateh ufti is a surface dissipation rate
density (of dimension uft MT
3 ). Following stan-
dard thermodynamic arguments, expressions Eqs. (30) and
(31) may be used to formally identify the term
gi
ws
p=qs t as the driving force of the influx (in the
case of mass deposition) or the outflux (in the case of
dissolution) of solid mass per unit surface occupied by
cells, which is expressed by qs tuc  n in the deformed
configuration, and m u c  N in the undeformed

c t0 qs t0 e
configuration. It is convenient to distinguish in ws the en-
ergy associated with elastic deformation of the solid from
those related to the chemical composition of the solid, by
considering ws wel el
s gs , where ws is the elastic free
energy and gs is the chemical potential per unit solid mass.
Furthermore, the dissipation expressions may be rewritten
in terms of the molar flux J qs t=Muc  n in Eq. (30)
and Je qs t0 =Me u c  N in Eq. (31), where M is the
molar mass, and M=qs t and M=qs t0 the molar vol-
umes of the solid in the deformed and reference configu-
rations
uft A  J  0; uft0 A  Je  0 32 dissolution process by a constitutive equation relating the
where A is the so-called chemical affinity (see [22, 23])
here of the cell-mediated biochemical reaction.
8879
 
p
A Mgi
gs
M wel
s 33
qs t
From Eqs. (3033) it is apparent that a positive affinity
encourages mass addition or surface growth
(uc uc  n  0), while a negative driving force encourages
mass removal or dissolution at the solid-cell surface
(uc  0). This is readily seen for an undeformed solid
phase, for which uft uft0 , and for which, following
chemical thermodynamics, the first term Mgi
gs in Eq.
(33) is identified as the Gibbs energy variation per mole;
that is the difference between the chemical potential of the
dissolved ions in the cell fluid, which is controlled by the
cell, and which is therefore conveniently referred to as
biologically generated potential lBGP [2], and the chemical
potential of the same substance bound in the solid phase ls
(which in turn is related to the solubility product of the
solid). In this case, the chemical affinity coincides with the
Gibbs energy variation (see e.g., [24]), i.e.,
encourage growth at the solid surface, as evident from the
non-negativity of the dissipation. Similarly, for a deform-
able medium, cellular fluid pressure and/or elastic defor-
mation in the solid will encourage mass dissolution at the
solid surface. Adopting the stated link between the Gibbs
potential gi and the biological generated potential lBGP for
constant solute pressure and temperature (for which
gi = const), the biochemical affinity becomes
M
el 
A lBGP
ls
v p 34
qs t v
el
where vel
v qs tws is the elastic energy per unit current
volume, while M=qs t is the current molar volume.
el
Denoting by e v el
v qs t0 ws and M=qs t0 the elastic
energy per unit volume and the molar volume, respectively,
in the undeformed configuration, the driving force can be
expressed in Lagrangian variables.
M
el 
A lBGP
ls
e
v jp 35
qs t0 v
Finally, following standard thermodynamics, the iden-
tification of affinity A as the driving force of the bio-
chemical growth/dissolution at the solid surface, (which
qualitatively defines the effect of the different components,
e.g., pressure, deformation and biochemical potential dif-
ference, on the growth/dissolution tendency at the solid
surface) leads to the definition of the kinetics of the growth/
driving force (A) to the molar flux in the deformed con-
figuration J J A, or in the undeformed configuration
Je JeA. The simplest form of such a growth/dissolution
123
8880 J Mater Sci (2007) 42:88738884

law that satisfies the non-negativity of the local dissipation biological tissue. At the macroscale, the primary objective
in Eq. (32) is a linear form which captures, similar to a is to determine the effects of the remodeling processes
discrete Ficks Law, the diffusion of the ionic species (which occurs at the microscale) on variables that are rel-
between the solid and the cellular fluid. evant at the scale of the porous material. At this scale, the
    r.e.v is composed of the solid domain (Vs t and the fluid
M e M e domain (Vf t; which occupies in the current configuration
A k J k J 36
qs t qs t0 the volume Vf t V t
Vs t; V t being the volume of
the porous tissue. The ratio of the current fluid volume over
Or equivalently, the total initial volume of the r.e.v V(t0) is referred to as the
Lagrangian porosity, and its rate of change is defined as
A kuc  keuec 37
 
d/ 1 dVt dVs t
where uc uc  n and uec uec  N. The diffusion
39
dt Vt0 dt dt
coefficient k (of dimension k LMT
1 mole
1 ) is
primarily governed by the diffusion mechanism in the
Using Vt JVt0 (where J is the determinant of the
deformed configuration, and is independent of the factors
macroscopic deformation gradient F) and defining
governing the driving force A, i.e., pressure, deformation
dVs t=dt by the upscaled version of Eq. (18), Eq. (39) can
and biochemical potential difference. In return, the
be rewritten as
diffusion coefficient ke in the reference configuration
takes into account the change of orientation of the solid Z Z !
d/ 1 dJ dj c
surface due to deformation. The two expressions for the Vt0
dvs t0
u Ndac t0
j~
diffusion coefficients k and ke can be related by using Eqs. dt Vt0 dt Vs t0 dt ft0

(16, 37) 40
where Vs t0 is now the total solid volume in the porous
ke material, and ft0 is the total cell-solid interaction
n  f  N
1 38
k surface (for the entire porous tissue being considered),
both expressed in the undeformed configuration.
Finally, it should be noted that surface curvature effects
Expressing
dj
the volume
R average of the rate
of change
have been neglected in our derivation. The surface curva- dj
of j by dt , i.e., Vs t0 dt dvs t0 Vs t0 djdts Vs t0 , Eq. (40)
ture affects both the dissolution kinetics, as well as the
can be reduced to
solid stress state, because of the associated stress discon-
htinuity
h ii at the solid-fluid interface (term of the form   Z
r  n
cjn, where j is the mean curvature of the d/ dJ dj 1

cs t0
uc  Ndac t0
j~
solid-cell surface3, and c is the surface energy per unit area, dt dt dt Vs t0 Vt0 ft0
of dimension MT2). This stress discontinuity would enter 41
through Eq. (22) the energy derivation, and would ulti-
mately lead to an effect of this curvature term on the dis- or equivalently, using the upscaled version of Eq. (20),
solution kinetics in a similar way as the fluid pressure.   Z
Indeed, it suffices to replace p in Eqs. (34, 35) by an d/ dJ dj 1
c
m

cs t0
j dac t0
effective pressure p0 p cj. It is no surprise, then, to dt dt dt Vs t0 Vt0 ft0 qs t0
confirm that a high positive (i.e., outward) curvature of a 42
solid surface leads to a higher dissolution rate, as it has
been recognized in skeletal tissue mechanics [25]. where cs t0 Vs t0 =Vt0 is the solid volume fraction in
the undeformed porous material, and m u c  N is

c qs t0 e
the mass rate per unit (undeformed) surface involved in the
Application to porous media cell-mediated deposition or dissolution process. The first
and second terms in Eqs. (41) and (42) represent the con-
Porosity change tribution of pure mechanical deformation (the entire porous
material and the solid phase, respectively) to the change in
At the microscopic scale, the formulation analyzed defor- porosity, and are the classical terms employed in the
mation coupled remodeling of a single strut of the porous Biot-Coussy theory of porous media [22]. The third term
3
signifies the contribution of the biochemical growth/dis-
2j 1=R1 1=R2 , with R1 and R2 the radius of curvature at the
solution process, amplified by deformation.
major and minor axis of the solid-cell surface.

123
J Mater Sci (2007) 42:88738884 8881

Dissipations Finally, if we add to the two contributions, given by Eqs.

(45, 46), the additional work rate related to the introduction
For upscaling of the Clausius-Duhem inequality from the (or extraction) of the solute mass dMi into the cell-fluid, we
micro-scale of the solid (see Eq. 29) to the macro-scale of the obtain the macroscopic counterpart of Eq. (25).
porous continua, the r.e.v in the macro-scale (consisting of
the porous tissue, as shown in Fig. 1a) is considered to be   Z  
d/ pi
subjected at the external boundary oVt0 to a uniform P ext Vt0 F_ : P p
c dac t0
m
dt ft0 qi
velocity boundary condition, and at the fluidsolid interface,
i.e., along oVf t, to a uniform pore fluid pressure p. The 47
presentation is inspired by the derivation of Biots porome-
Or equivalently, using Eq. (42), Eq. (47) can be
chanics theory by Dormieux et al. [26]. For purpose of
rewritten as
analysis, the boundary of the r.e.v oVt0 is assumed to be
located in the solid domain, so that the fluid domain is sur-  
rounded by solid. In this case, the boundary conditions read _ d/
P ext Vt0 F : P p
dt
oV t0 : u F_  X
d
43 Z  
jp pi



c dac t0
m 48
ft0 qs t0 qi
oVf t : t
pn 44 where d/ =dt stands for the change in Lagrangian porosity
in the absence of mass addition or removal processes.
where F_ is the macroscopic deformation gradient rate, ud
denotes the microscopic velocity field in the solid phase  
d/ dJ dj
that leads to deformation, and n is the unit normal vector
cs t0 49
dt dt dt Vs t0
oriented outward with respect to the solid at the fluidsolid
interface. The macroscopic counterpart of Eq. (22) then is
Equation (48) expresses the work rate provided to the
the sum of two work rate terms provided to the solid phase:
solid phase. Indeed, it is nothing but an application of Eq.
1. The work rate developed by the surface tractions p  N (25) using the boundary conditions given by Eqs. (43, 44).
and the velocity ud prescribed on oV t0 in the refer- The associated change of the Helmholtz free energy in the
ence configuration. Given the uniform boundary con- solid phase is given by the upscaled version of Eq. (27).
dition Eq. (43), this term can be developed with the Assuming the solid chemical potential (gs) is invariant with
help of the Hill Lemma (see Appendix) time (dgs =dt 0), and using Eq. (33), the rate of change of
Z   Helmholtz free energy at the macroscopic scale can be

 rewritten in the form
Pext ud ud  p  N b dat0 F_ : P Vt0
oV t0
 el 
45 dW vv
de
Vt0 cs t0
dt dt Vs t0
where N b Dis Ethe unit outward normal vector to oV t0 , Z  
A jp pi
and P p is the macroscopic Boussinesq ten-


c dac t0
m 50
Vt0 ft0 M qs t0 qi
sor, that is the volume average of the microscopic tensor
p over the total volume of the porous medium Vt0 . where ev el
v is the microscopic elastic energy volume-density,
2. The work rate developed by the uniform pore pressure measured in the undeformed configuration. Finally, using
p R and the rate of volume change dVf =dt Eqs. (48, 50) in Eq. (21) yields the Clausius-Duhem

oVf t u  ndat (composed of the volume change inequality at the macroscopic scale (in the undeformed
due to deformation and chemical growth/dissolution) reference frame)
along the solid-fluid interface. Since the fluid-surface
interface is entirely surrounded by the solid phase, the D Dm Dc  0 51
corresponding volume change equals the change in
pore space. Hence, with Eq. (39) where, Dm , denoting the rate of dissipation due to
mechanical factors, and Dc , the rate of dissipation
dVf d/ primarily due to growth/dissolution (but including the
p p Vt0 46 coupled terms), are respectively given by
dt dt

123
8882 J Mater Sci (2007) 42:88738884

 el  !
d/ vv
de
_
Dm Vt0 F : P p
cs t0 52
dt dt Vs t0
Z  c


m
Dc A dac t0 53
ft0 M
Equation (52) is the standard form of the mechanical
dissipation associated with the pure (finite) deformation
of the solid phase of a porous material, as described by
the classical Biot theory. The additional dissipation
caused by the biological activity of cells attached to the
solid surface is captured by Eq. (53), where A is the
affinity defined in various forms and configurations by
Eqs. (3335): it is the driving force of the mass depo-

c and of the associated solid
sition or dissolution flux m
c
growth uc  n or ue  N. Provided the value of the affin-
ity is known, the rate of the biological mediated mass
removal can be determined from kinetics laws, such as
Eqs. (36, 37), and the resulting change of the porosity
and the overall solid behavior can be evaluated.
Example of deformation coupled remodeling
The effect of imposed deformation on the dissolution rate
at the surface of a cylindrical strut at the microscale is
explored in this section. The surface of the solid strut may
be assumed to be smooth and the curvature effect on the
dissolution rate is neglected. For an undeformed solid, the
growth velocity is uc0 , the corresponding biochemical
affinity is A0 lBGP
ls , and the solid volume is V0.
From Eq. (37),
A0 kuc0 54
The solid subsequently undergoes uniaxial deformation.
Assuming the solid (a soft biological tissue) to be an
incompressible isotropic Hookean material, the change in
the elastic energy density may be written as (following
[27])
E
v el
De v Ik
3 55
41 m
where E is the elastic modulus and m the Poissons ratio for
the isotropic biological tissue, and Ik k21 k22 k23 ; k21 ,
k22 , k23 being the eigenvalues of the symmetric Cauchy
dilatation tensor (C f t  f ). Assuming that the fluid
pressure (p) and the biochemical potential term
(lBGP
ls ) are insensitive to the imposed deformation,
the driving force for growth/dissolution in the deformed
solid is given by (see Eq. 35)
123
M el
A A0
e
v 56
qs t0 v
Normalizing both sides of Eq. (56) by A0 , and using
Eqs. (37, 54, 55), the dissolution velocities in the
deforming and the undeformed solid can be related by
   
EM Ik
3
uc uc0 1
57
A0 qs t0 41 m
For uniaxial deformation (in x1-direction) of an incom-
pressible tissue, m 0:5, and Ik is given by4
Ik k21 2k
1
1 . p
Physically, k1 is an eigenvalue of C and the stretch of
a material element in the uniaxial loading direction. Thus,
Eq. (57) can be rewritten in a dimensionless form
  2
uc EM k1 2k
1
1
3
c 1
58
u0 A0 qs t0 6
The dimensionless number E EM=A0 qs t0 is the
ratio of the specific elastic modulus of the soft biological
tissue (E=qs t0 ) and the specific affinity (A0 =M) of the
biochemical process that takes place at the solid-cell
interface, and may be considered as a measure of the rel-
ative importance of the elastic deformation vis-a-vis the
pure biochemical driving force on the overall remodeling
kinetics. Estimation of E for soft tissues is unavailable to
the best of our knowledge. However, E for dissolution in
hard-tissues (bone) may be estimated from the work of
Silva and Ulm [2] (qs t0 A0 =M
31 MPa), and Rho
et al. [28] (E 17 GPa), which yields E
548 for bone
dissolution (This value may still be higher if one considers
the stiffness of the hydroxyapatite crystals that build up the
ultrastructure of bone; see e.g., [17]). Figure 3a shows the
normalized dissolution velocities (uc =uc0 ) for E values
ranging from 1 to
1; 000, for stretch ratios raging from
0.5 to 2 (i.e., engineering strains from 0.5 (compression)
to + 1.0 (tension)). It is evident that both compressive and
tensile deformation of the solid enhance the dissolution
rate, at all E values. The effects, as expected from Eq. (58),
increase with increasing jE j. Quantitatively, the effects of
compressive and tensile deformations on the dissolution
rate are dissimilar, and the rate of dissolution rate
enhancement (per unit stretch or engineering strain) is
larger under compressive loading. It should however be
noted that the dissolution enhancement effect is relative to
the chosen strain measure; here the stretcha Lagrangian
strain measure. This trend is inverted if an Eulerian strain
4
For an incompressible solid, the third invarient of the Cauchy
dilatation tensor k21 k22 k23 1. For uniaxial deformation in an isotropic
solid, k2 k3 . Thus, k22 k23 k
1 2 2 2 2
1 , and k1 k2 k3 k1 2k1 .

1
J Mater Sci (2007) 42:88738884 8883

1000 1000
(a) (b)

Normalized dissolution velocity (uc /uc )

0
Normalized dissolution velocity (uc /u c )
0
= -1000 = -1000

100 100

= -100 = -100

10 10

= -10 = -10

= -1 = -1
1 1
0 .2 5 0 .5 0 .7 5 1 1 .2 5 1 .5 1 .7 5 2 2 .2 5 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
Str e tch in load ing dir e ctio n T rue strain

Fig. 3 Representation of the normalized dissolution velocity varia- deformation measure is defined by the true strain in the loading
tion for different soft biological tissues under uniaxial loading, in (a) direction (natural log of stretch), signifying compression for negative
Lagrangian and (b) Eulerian strain measures. The Lagrangian true strain and tension for positive strain. The tissue properties are
measure of tissue deformation is defined by the stretch in the loading characterized by E, which signifies a ratio of the elastic deformation
direction (stretch = (final length)/(initial length)). This signifies energy and the biochemical energy for the tissue remodeling (see text)
compression for stretch <1 and tension for stretch >1. The Eulerian

measure is employed, such as the true (logarithmic) strain an increase in the chemical potential of the dissolving ions
which is related to the (Lagrangian) stretch by
ln lnk1 . (compared to their potential in the solid) was observed to
The results
 displayed
 in Fig. 3b indicate, for the same encourage tissue growth. The microscopic equations were
value of 
ln , a higher dissolution rate enhancement in upscaled to yield the tissue characteristics (e.g., change in
tension than in compression. This underscores the necessity Lagrangian porosity) at the macroscale. A numerical
to specify (both theoretically and experimentally) the ref- example on the effect of deformation on the dissolution
erence frame (Lagrangian or Eulerian) used for measuring rate for an incompressible soft tissue showed that the rate
field quantities. increased for both tensile and compressive deformations.
The change of the dissolution rate (per unit strain) de-
pended on the strain measure. For Lagrangian measures
Conclusion
(e.g., stretch), the rate increase per unit strain was higher
under compressive loading than under tensile loading. On
The present work attempts to develop an integrated anal-
the other hand, the rate increase per unit strain was higher
ysis for deformation coupled surface remodeling in porous
under tensile deformation when an Eulerian measure (e.g.,
biomaterials. In such materials (examples include trabec-
true strain) was used.
ular bone and skin), the biochemical phenomenon related
to remodeling (growth/dissolution) occurs at a microscopic Acknowledgments The authors wish to thank the Natural Sciences
scale and estimation of its effect on the tissue level and Engineering Research Council (NSERC) of Canada and the Es-
(macroscopic) properties is important. In the present work, ther and Harold E. Edgerton Chair at MIT for financial assistance.
The authors are thankful to Profs. Lorna Gibson (MIT), Olivier
the remodeling and deformation (at the microscale) in the
Coussy (Institute de Navier, France), and Luc Dormieux (Institute de
biological tissues is modeled through development of Navier, France), and Dr. Anirban Sain (McGill University, Canada)
suitable equations for mass, volume and the relevant for helpful discussions.
thermodynamic quantities for the growing solid, which are
then upscaled to yield the macroscopic effects. The equa-
tions are developed assuming a spatial reference frame, to Appendix: Hill Lemma
circumvent the problems of defining a suitable material
reference frame for growth/dissolution processes, charac- This demonstration of the Hill Lemma is inspired by the
terized by a continuous change in the number of material presentation of Zaoui [29]. We denote by p and f_ the
(and mass) points. The analysis examined the effects of microscopic Boussinesq tensor which (in the absence of
different field quantities on the growth/dissolution potential body forces) satisfies Divp 0 in V(t0), and the micro-
of the solid-cell interface. A positive pressure of the cel- scopic deformation gradient rate tensor f_ Gradu. If,
lular fluid and deformation of the solid was observed to either f_ satisfies a uniform deformation boundary condi-
encourage solid dissolution (and tissue resorption), while tion, or p a uniform traction boundary condition, then

123
8884 J Mater Sci (2007) 42:88738884

D E D E D E

Again, inputing f_ij V t F_ ij in Eq. (65)

f_ : p f_ : p F_ : P 59 0
Vt0 Vt0 Vt0
P V t0 F_ ij Pij V t0 F_ : P 67
(1) Consider the r.e.v V subjected to a uniform velocity
boundary condition

on oV t0 : u F_  X 60 References

where F_ represents the deformation rate of the r.e.v at the 1. Cowin SC (1983) Ann Biomed Eng 11:263
macroscopic scale, while u is defined at the microscopic 2. Silva EC, Ulm F-J (2002) In: Karihaloo BL (ed) Proc. IUTAM
scale. The work rate provided to the r.e.v by the surface Symp. on Analytical and Computational Fracture Mechanics of
Non-Homogeneous Materials, Kluwer Acad. Pub., London,
traction t p  n in the undeformed configuration reads (in p 355
components) 3. Ballarini R, Kayacan R, Ulm F-J, Belytschko T, Heuer AH
(2005) Int J Fracture 135(14):187
Z Z 4. Grinnell F (1994) J Cell Biol 124:401
P ui ti dat0 F_ ij Xj pik Nk dat0 61 5. Freyman TM, Yannas IV, Pek Y-S, Yokoo R, Gibson LJ (2001)
oV t0 oV t0 Exp Cell Res 269:140
6. Lund D, Tomanek RJ (1978) Am J Anat 152:141
Application of the divergence theorem yields, for any 7. Sobin SS, Tremer HM, Hardy JD, Chiodi HP (1983) J Appl
Physiol 55:1445
stress field pik that satisfies pik;k 0 (that is Divp 0) 8. Yannas IV, Lee E, Orgill DP, Skrabut EM, Murphy GF (1989)
yields Proc Natl Acad Sci USA 86:933
Z Z 9. Pauwels F (1980) Biomechanics of the locomotor apparatus.
Springer-Verlag, Berlin
Xj pik Nk dat0 Xj pik ;k dvt0 10. Skalak R (1981) In: Carlson DE, Shield RT (eds) Proc. IUTAM
oV t0 V t0
Z Symp. on Finite Elasticity, Martinus Nijhoff, The Hague, p 348
11. Rodriguez EK, Hoger A, McCulloch AD (1994) J Biomech
pik djk Xj pik;k dvt0
V t0 27:455
Z 12. Chen Y-C, Hoger A (2000) J Elast 59:175
pij dvt0 62 13. Lubarda VA, Hoger A (2002) Int J Solids Struct 39:4627
V t0 14. Lee EH (1969) J Appl Mech 36:1
15. Lubarda VA (1991) Int J Solids Struct 27:885

16. Gibson LJ (1985) J Biomech 18:317
Thus, inputing pij V t0 Pij in Eq. (61)
17. Hellmich C, Ulm F-J (2002) J Biomech 35:1199
18. Zysset PK, Curnier A (1996) J Biomech 29:1549
P V t0 F_ ij Pij V t0 F_ : P 63 19. Blair HC (1998) BioEssays 20:837
20. Lemarchand E, Dormieux L, Ulm F-J (2003) In: Huyghe JM (ed)
(2) Consider the r.e.v subjected to a uniform traction Proc. IUTAM Symp. on the Mechanics of Physicochemical and
Electromechanical Interactions in Porous Media, Kerkrade
boundary condition (Netherlands)
21. Ulm F-J, Lemarchand E, Heukamp FH (2003) Eng Fract Mech
on oV t0 : t p  N P  N 64 70:871
22. Coussy O (1995) Mechanics of porous continua. J. Wiley & Sons,
where P is the macroscopic Boussinesq tensor, while t is Chichester UK
the microscopic stress vector (defined on the undeformed 23. Coussy O, Ulm F-J (1996) Arch Appl Mech 66(8):523
configuration). The work rate is developed in the form: 24. Atkins PW (1994) Physical chemistry. WH Freeman and Com-
pany, New York, p 216
Z Z Z 25. Martin RB, Burr DB, Sharkey NA (1998) Skeletal tissue
mechanics. Springer-Verlag, New York
P ui ti dA ui pij Nj dA ui Nj dA Pij 65 26. Dormieux L, Djimedo K, Ulm F-J (2006) Microporomechanics.
oV oV oV
J. Wiley & Sons, Chichester UK
27. Attard MM (2003) Int J Solids and Struct 40:4563
Application of the divergence theorem yields 28. Rho JY, Roy ME, Tsui TY, Pharr GM (1999) J Biomed Mater
Z Z Z Res 45:48
29. Zaoui A (1997) Materiaux heterogene et composites (lecture
ui Nj dA ui djk Nk dA ui djk ;k dV t0 notes, Edition 1997). Departement de Mecanique, Ecole Poly-
oV oV V t0
Z Z technique, France
ui;k djk dV t0 f_ij dV t0 66
V t0 V t0

123
J Mater Sci (2007) 42:88858893
DOI 10.1007/s10853-007-1759-7
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Effects of contact shape on biological wet adhesion
Yewang Su Baohua Ji Yonggang Huang
Kehchih Hwang
Received: 18 December 2006 / Accepted: 10 April 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract Wet adhesion is widely adopted in biological
adhesion systems in nature. Wet adhesion is studied in this
paper with the focus on the effect of different contact
shapes (flat, concave, convex, and ring-like) on the adhe-
sion force. The evolution of the liquid bridge between a
fiber tip and substrate during the detaching process shows
two transition points. The first transition from the radius-
controlled to the contact-angle controlled process is critical
to influence the strength and robustness of adhesion. We
show that a concave shape is more effective than a flat one,
while a convex shape has no advantage. A ring-like contact
shape has advantages in a hydrophobic environment and on
a rough surface.
Introduction
Insects and geckos have evolved a variety of well-defined
shapes and structures for adhesion. Although often intricate
and fragile, they can nonetheless deal with extreme
mechanical loads with high efficiency. Some insects live
attached to a substrate with dry adhesion; others use wet
adhesion. These abilities are based on a variety of inge-
nious structural solutions. Understanding these is of great
Y. Su
B. Ji (&)
K. Hwang
Department of Engineering Mechanics, Tsinghua University,
Beijing 100084, China
e-mail: bhji@mail.tsinghua.edu.cn
Y. Huang
Department of Mechanical Science and Engineering, University
of Illinois, Urbana, IL 61801, USA
e-mail: huang9@uiuc.edu
scientific interest, since it can give insights into the
workings of nature for shaping the structures in evolu-
tionary processes. Also, we can discover the detailed
chemical and physical properties of the materials which
have evolved, and can learn about their use as structural
elements and their biological role and function for the
guidelines of man made adhesion systems.
Previous works [13] have studied the hierarchical
structure of the animals feet and the scaling law for dry
adhesion. The adhesive system with dry adhesion is based
on van der Waals forces or electrostatic force between the
finely structured feet and the substrate. A number of
adhesion measures [47] on the attachment devices such
as: flies, spiders, geckos, Tettigonia viridissima, etc, pro-
vide the evidence that van der Waals force plays dominant
role in dry adhesion, and theoretical analysis [1, 8, 9]
proved that biological systems indeed can use van der
Waals force to achieve strong adhesion with finely struc-
ture design of the attachment systems.
In contrast, wet adhesion is based on capillary for-
cesthrough a liquid bridge between the fiber and the
surface of the substrate [2]. Capillary force is a long range
force in comparison with the van der Waals force. The
capillary force between a hemisphere and a plane had been
studied several decades ago [10]. The determination of li-
quid-bridge profile is the key to solve a wet adhesion
problem [11]. The stability, breakage, elastic properties of
liquid bridge [12] at nanometer scale are studied by both
theoretical and experimental approaches. Experiments
showed that humidity contributes significantly to gecko
adhesion implying the importance of capillary force for
adhesion at larger scale [13, 14]. A fiber-substrate model
has been recently studied by Qian and Gao [15]. They find
that the size of the fiber plays an important role in bio-
logical wet adhesion, and the strength of wet adhesion is
123
8886
highly enhanced by radius reduction of the fiber. A scaling
law of the adhesion strength was obtained for the self-
similar fiber-substrate systems.
In previous studies, the tip of the fiber in the adhesion
system was often assumed to be flat or hemisphere-like
[16]. Experimental studies show that nature provides a
variety of contact shapes, e.g., horseshoe, suction cup, and
torus, etc., for good adhesion [1621]. What is the effect of
the contact shape on the adhesion force? For dry adhesion,
it has been found that the contact shape exerts a strong
influence on adhesion strength, e.g., the best shape is the
flat tip for perfectly smooth substrate [16]. Gao and Yao
[19] also discussed how to design an optimum contact
shape for dry adhesion at nanoscale. In this paper, we study
the effects of contact shape on the adhesion force for the
wet adhesion system. This study is orientated to help
improve the design of micro- and nano- manipulation
system based on capillary force by the biomimicking
approach. For example, recent experimental studies [21,
22] suggested that a concave tip can generate a much larger
capillary force than a flat one. However, the mechanisms
have not been well understood.
The organization of the paper is as follows: In Section
The model, we introduce the models for wet adhesion pressure difference which is constant within the meniscus
with four different contact shapes of the fiber tip, mainly
focusing on how to calculate the profile of the liquid bridge
and the adhesion force. These four contact shapes are flat,
concave, convex, and ring-like, which represent the flat
punch, suction cup, hemisphere, and torus structure
evolved in nature [18], respectively. In Section Results, cosh
these four contact shapes are studied, and the detaching
force for each contact shape is calculated and compared.
Finally, discussion and conclusions are made in Section
Discussion and conclusions.
The model
The wet adhesion of the hairs of biological attachment
system is modeled as a fiber contacting with the substrate
mediated by a liquid bridge. To study the effect of contact
shape on the adhesion strength, we introduce four different
geometries of the fiber tip, i.e. flat, concave, convex, and
ring-like, to model the flat punch, suction cup, hemisphere,
and torus structure evolved in nature [18], respectively,
shown in Fig. 1. The fiber is assumed axisymmetric. The
Fig. 1 The adhesion systems
with different contact shapes.
(a) flat; (b) concave; (c) convex;
(d) ring-like
123
J Mater Sci (2007) 42:88858893
model for the liquid bridge is shown in Fig. 2. We make
the following assumptions in order to simplify the analysis:
(a) No phase change such as liquid to gas during the
detaching process such that the volume of liquid is
conserved;
(b) Since the fiber and substrate are much stiffer than the
liquid and therefore have essentially no deformation
during the process, both the fiber and the substrate are
assumed to be rigid;
(c) The detaching process is quasi-static such that the
inertia and viscoelastic effects are negligible;
(d) Since the typical size (diameter) of the fiber is at the
micrometer scale, the effect of gravity is negligible in
comparison with the surface tension at this scale.
For this problem, the YoungLaplace equation connects
the pressure difference inside and outside the meniscus to
the local liquid profile by
1 1
DP c 1
R1 R2
where 1/R1 and 1/R2 denote the two local principal
curvatures of the liquid profile, respectively, DP is the
due to the negligible gravity, and c is interface energy
between vapor and liquid. c is connected with the contact
angle h via the Youngs equation,
cSV  cSL
2
c
where cSV and cSL are interface energy of solidvapor and
solidliquid, respectively.
The liquid profile should be obtained first in order to
calculate the adhesive force. An ordinary differential
equation can be derived from YoungLaplace equation [10,
15, 23],
du DP sinu dx dz
 ; where cosu; sinu 3
ds c x ds ds
where x and z are the coordinates of the axisymmetric li-
quid bridge, u is the angle between the local tangent of
liquid surface and the horizontal axis, and s is the arc
length of the liquid profile as shown in Fig. 2.
J Mater Sci (2007) 42:88858893 8887

surfaces at the corresponding contact angles hup = h1

Fiber (shown in Fig. 2). Again the liquid meets the substrate at
the contact angle h2. For the angle controlled process the
X up
X
boundary conditions are,
O up up
uZ 0 p  h1 and uZ D h2 5b
Liquid
D Vapor The liquid volume can be calculated by the following
S integral,
2 Z
Substrate
V A dZ 6
D
Z
And the normalized adhesion force can be calculated as
Fig. 2 The sketch map of the liquid profile and its reference frame follows [15],

Introducing the nondimensional units X x=R, A X Li

Z z=R; S s=R, and H DP
R=c, Eq. (3) can be g
H
sinhiup 7
p i
p
rewritten as,
where A is the non-dimensional contact area between liquid
du sinu dX dZ and the fiber, Li is the nondimensional perimeter of contact
H ; where cosu; sinu 4
dS X dS dS line of the circle i. For flat, concave or convex contact tip,
the contact area is a circle with one circle contact line (see
There are normally two states of the contact between the fiber Fig. 2 and 6a, b),
tip and the liquid during the pulling process of the liquid
bridge, i.e., radius controlled state and angle controlled state 2
A p
X up ; and L1 2pX up ; i1 8
for flat contact shape [15, 24]. A critical separation D (the
distance between the fiber tip and substrate, see Fig. 2), be- however, for ring-like contact shape (Fig. 6c), the contact
low which the liquid periphery becomes pinned at the edge of area is annular with two circle contact lines,
the fiber, defines the transition point between contact angle
dominated adhesion and fiber radius dominated adhesion out2 in2 out in
[15, 24]. At the beginning of the detachment, the liquid A pX up  X up ; L1 2pX up and L2 2pX up ;
profile is at the radius controlled state; as the separation i 1; 2 9
becomes larger than the critical separation D, the shrinkage
of liquid contact area induces the transition from radius Table 1 in the Appendix gives the analytical solutions for
controlled state to angle controlled state. the axisymmetric profile of the liquid bridge in curvilinear
In the radius controlled state, the liquid periphery is coordinates for the calculation of the liquid volume and the
pinned at the edge of the fiber tip, and the liquid meets the separation.
surfaces of the fiber tip at the angle hup different from their
real contact angle h1; X up is the radius of the contact area
between the fiber and the liquid assuming the contact line is Results
a circle (see Fig. 2). On the surface of the substrate, the
liquid meets the solid always at the contact angle h2 as We first study the detaching process of a fiber with flat tip
shown in Fig. 2. Therefore, the boundary conditions for the from the substrate, focusing on the details of the separating
radius controlled process are, process, such as the transition points between the two
states, and the relation between transition points and the
XZ 0 1 and uZ D h2 5a point for the maximum detaching force. We then study the
detaching process of the fiber with other three contact
where uup = p hup, and hup can be calculated according to shapes, i.e., concave, convex, and ring-like.

the separation D.
In the angle controlled state, the liquid periphery can not Flat tip
touch the circumference of the fiber due to the shrinkage of
the contact area when the separation between the fiber tip The prior study [15, 23] on the mechanics of wet adhesion
and substrate becomes large, then the liquid meets the fiber of a cylindrical fiber with flat tip to an infinite substrate

123
8888 J Mater Sci (2007) 42:88858893

Fig. 3 A typical separating

process of the fiber-liquid-
substrate system with a flat tip
which shows the evolution of
the profile of the liquid as the
contact state between fiber and
liquid changes from radius
controlled to angle controlled,
and then comes back to radius
controlled state before the liquid
collapses

focused on the scaling effects in adhesion force. The size

reduction of the fiber and liquid bridge (i.e., the radius of
fiber and volume of liquid) as well as small contact angle
increase the adhesion strength. The present work focuses
on the influence of contact shape on the adhesion force. We
first examine the evolution of liquid bridge profile, and its
contact area with both the fiber and the substrate during the
detaching process in order to identify the mechanisms of
enhancing the adhesion strength via the contact shape.
A typical separating process of liquid bridge between a
fiber with flat tip and substrate is shown in Fig. 3. There are
two transition points during the detaching process. The first
is the radius controlled to angle controlled state transition;
the second one is angle-to-radius controlled transition,
which has not been reported before. At the beginning of the
detachment, the normalized detaching force g (capillary
force) is negative because the part of total detaching force
due to the pressure difference is negative and its absolute
value is larger than the part due to the surface tension of the
liquid at small separation (see Eq. 7). With the increase of
the separation, the detaching force continuously increases
and becomes positive, then reaches its maximum value at a
critical point; afterwards, the capillary force decreases as
the separation increases, as shown in Fig. 4a. The position
of the two transition points is of great interest for small
contact angle h1. For example, for h1 = 0, the radius-to-
angle controlled transition (transition I) occurs after the
critical point for the maximum capillary force; and the
transition II, i.e., the angle-to-radius controlled transition
occurs near the critical point of breakage of the liquid
bridge. These are further illustrated in Fig. 4b, c, which
show the evolution of the contact angle and contact radius
between the fiber and liquid, respectively. In particular,
Fig. 4c clearly shows the shrinkage and then spreading
process of the contact area between the two transition
points. We define the maximum value of the detaching
force as the ideal maximum detaching force when the
transition I happens after the critical point for the maxi-
mum detaching force. When the transition I happens earlier Fig. 4 (a) The force-separation curve of the liquid bridge in a quasi-
than the critical point (at larger h1), the maximum static pulling process, showing two contact states and their transition
points; (b) the evolution of the angle between the fiber surface and
detaching force is smaller than the ideal maximum liquid uup during the pulling process; (c) the evolution of the
detaching force as shown in Fig. 5. normalized radius of contact area X up

123
J Mater Sci (2007) 42:88858893
Fig. 5 The force-separation curves for the flat shape with different
contact angle h1, which shows that the larger the h1, the earlier the
happening of the transition I, and the smaller the adhesion force is
The positions of the transition points on Fig. 4a depend on
the liquid volume and the two contact angles h1 and h2. Our
numerical results show that the increase of contact angle h1
brings transition I forward (see Fig. 5). Once the transition-I
point reaches the critical point for the ideal maximum
detaching force, the detaching force starts to decrease, and
the maximum detaching force becomes smaller than the ideal
maximum adhesion force as shown in Fig. 5. Therefore,
transition I is crucial for the detaching force, and the contact
angle h1 influences the transition-I position. For h1 smaller
than a critical value, the transition I always happens after the
critical point for the ideal maximum detaching force such
that the maximum detaching force is equal to the ideal
maximum detaching force, as shown by the red dashed line
and the curve of the flat tip in Fig. 8. We define this critical
value of the contact angle h1 as h*1 below which the detaching
force can achieve its ideal maximum value. This mechanism
is important to the understanding of detaching process of
fiber with other contact shapes, such as concave and convex
tip, as to be shown later.
Fig. 6 Schematic illustration of the fiber-liquid-substrate system with
three different shapes. (a) the concave shape with an effective contact
angle heq < h1 and an effective liquid volume Veq V  DV ; (b) the
8889
Concave and convex tip
For the study of detaching process of the fiber with the
concave shape, we first introduce the concept of the effective
contact angle and effective liquid volume for the conve-
nience of comparing with the flat shape. The effective con-
tact angle heq is defined as the angle between the tangent line
of the liquid profile (two dimensional) and the horizontal line
at the triple point (of solid, liquid and vapor), as shown in
Fig. 6a, and h eq < h1 for the concave shape. The effective
liquid volume V  eq is defined as the volume of liquid under the
horizontal plane passing the outer edge of the concave sur-
face (see Fig. 6a). However, the total volume of the liquid
bridge under the concave surface is V  V eq DV, where

DV is the volume between the horizontal plane and the
concave surface, as shown in Fig. 6a.
The numerical results show that the concave tip can
induce larger detaching force than the flat tip under the
condition of the same effective liquid volume at large
contact angle h1 (h1 > h*1). Here the effective liquid
volume for flat shape is the same as its total volume.
The detaching forces for the concave, flat and convex
tips are shown in Fig. 7 by the curves A, B and C,
respectively, for the liquid volume V  V  eq 1:0. The
maximum detaching force for the concave shape is larger
than that for the flat tip, while the convex tip gives the
smallest. What is the mechanism for the concave shape
to achieve large detaching force? It is observed that all
three curves overlap before the detaching force for the
convex shape reaches its maximum value, after which
the detaching force for the convex tip decreases, fol-
lowed by the flat tip, but the force for the concave shape
continues to increase and reaches its maximum value
higher than that of the flat shape at a larger separation
D. The underlying mechanism is that the concave shape
delays the radius-to-angle controlled transition, which
effectively increases the detaching force.
In order to further understand this mechanism, we study
the relationship between the contact angle h1 and the
convex shape with an equivalent contact angle heq > h1 and an
effective liquid volume Veq V DV ; (c) the ring-like contact
shape with two contact lines
123
8890
Fig. 7 The comparison of the detaching force among different
contact shapes, concave, convex and flat shape. The figure shows that
the concave shape can induce larger detaching force than that with flat
and the convex shape under the condition of the same effective liquid
volume
maximum detaching force for the concave shape. The
comparison of the maximum force between the concave
10 2
shape with Z X =4 and Z X =4, and the flat shape at
same effective liquid volume, is shown in Fig. 8. For the
flat shape, the maximum detaching force remains a constant
and is equal to the ideal maximum detaching force when
h1 < h*1 = 20, but starts to decrease when h1 > 20. The
concave shape achieves the same ideal maximum detaching
force, but its critical value h*1 is much larger than that of the
2
flat shape, such as h*1 = 45 for the concave shape Z X =4.
This implies that the concave shape is less sensitive to the
changing of contact angle h1, and reflects the robustness
of the adhesion system [3, 19, 25] about the surface
chemistry of material. For further increased curvature of the
10
concave-shaped fiber tip, Z X =4, the maximum force
Fig. 8 The relationship between the maximum detaching force and
10
the contact angle h1 for different contact shapes: concave Z X =4,
2 2
concave Z X =4, convex Z X =15 and the flat shape. It shows
that the concave shape make the maximum detaching force be
insensitive to the change of contact shape h1
123
J Mater Sci (2007) 42:88858893
does not decrease until the contact angle is increased to as
large as h1 = 90, and only has a minor decrease at the
contact angle h1 = 120, exhibiting stronger robustness.
This shows that the large curvature at the contact edge of the
concave shape can effectively enlarge the critical value of
h*1 (retarding effect) for the radius-to-angle controlled
transition so that the transition can not happen and therefore
the ideal maximum detaching force can always be achieved
in a very large region of h1. It can be seen from Fig. 8 that
the larger the curvature of the concave shape, the more
effective the retarding effect it has.
In the same manner of studying the detaching process of
the fiber with the concave tip, we consider the effect of the
convex contact shape on the wet adhesion. Our calculations
show that the detaching force for the convex tip is more
sensitive to the contact angle h1 in comparison with the
concave shape and flat shape, see Fig. 8. The mechanism is
that the effective contact angle between the fiber and liquid
for the convex shape is larger than their real contact angle,
heq > h1 due to the negative curvature of the contact surface
(see Fig. 6b, while heq < h1 for concave shape) and thus
promotes the transition I to occur earlier, which severely
weakens the adhesion force. For instance, the detaching force
2
of a convex tip described by Z X =15 is calculated and
compared with that of the flat and concave shape in Fig. 8.
From Fig. 8, we can see that the detaching force for the
convex tip can only achieve the ideal maximum detaching
force at very small region of the contact angle h1 and declines
much faster with the increase of the contact angle in com-
parison with the flat and concave tip. Therefore, we conclude
that the convex shape is less efficient for the wet adhesion.
Ring-like tip
Our previous calculations have shown that, at small contact
angle, the adhesion force of the fiber-substrate system with
flat, concave and convex contact shape mainly relies on the
pressure difference DP; however, for large contact angles
h1 and h2, i.e., the surface of fiber tip and the substrate are
hydrophobic, the pressure difference can be positive, which
largely decreases the detaching force. In order to enhance
the adhesion force in a hydrophobic environment, the
contact area should be reduced, but the contact line length
should be enlarged to strengthen the detaching force via
surface tension. The ring-like contact shape should be a
good candidate.
The detaching force of the ring-like contact shape in
comparison with the flat shape at various liquid volume is
shown in Fig. 9. The ring-like contact shape is clearly a
better choice than the flat shape at a hydrophobic environ-
ment (i.e.. at large contact angle h2); we can see that the ring-
like shape induces a larger detaching force at a larger sepa-
ration for small liquid volume because its force-separation
J Mater Sci (2007) 42:88858893
Fig. 9 The force-separation curve of the ring-like contact shape in a
hydrophobic environment (large h2) in comparison with the envelope
of the force-separation curves of the flat shape with various liquid
volume, where the contact angle h1 = 0. It shows that the ring-like
contact shape can achieve comparably larger detaching force at larger
separation in a hydrophobic environment
curve is above the envelop of those of the flat tip for different
liquid volume (see Fig. 9). Although the maximum detach-
ing force of the flat shape at volume V 0:2 is larger than
that of the ring-like contact shape, this maximum is achieved

at very small separation, D\0:1 ; for a rough surface with
D > 0:1, this maximum detaching force may not be achiev-
able. Consequently, the ring-like contact shape may enhance
the adhesive force in the hydrophobic environment.
Discussion and conclusions
This paper is aimed to study the effect of different contact
shapes on the detaching force in order to provide the
guidelines for the design of the micro- and nano- attach-
ment system through biomimicking approach. For a fiber
with flat tip, there are two transition points of the contact
state between the fiber tip and the liquid bridge, i.e., the
Appendix
Table 1 Analytical solution of the axisymmetric liquid profile
Meridional curvature Average Profile Profile radius
curvature radius
du
a dS
>0 H 6 0 X>0
Xu
8891
radius-to-angle controlled transition (transition I) and the
angle-to-radius controlled transition (transition II), where
the second transition is identified for the first time in this
paper. The transition I is crucial for the maximum
detaching force. The detaching force always achieves its
ideal, maximal value if the transition I occurs after a crit-
ical point. There also exists a critical value for the contact
angle between the fiber and liquid for the transition I at
which the detaching force reaches its Ideal maximal value.
In comparison with the flat tip, the concave contact shape is
more effective to delay the transition I and therefore to
achieve the maximum detaching force. The concave tip
also makes the detaching force be insensitive to the
changing of contact angle, and therefore is robust to the
chemical change of the environment. This finding is con-
sistent with the experimental observations [21, 22] of the
attachment system of flies to explain why nature designs
the concave tip for wet adhesion. Furthermore, we show
that a convex contact shape has no such advantage, and a
ring-like contact shape has advantages for a rough surface
in a hydrophobic environment. The present work represents
the first systematically study of the effect of contact shape
on wet adhesion of biological attachment systems, which
paves the way for the optimum design of the contact shape
of wet adhesion systems, and may be used in the bio-in-
spired design of new mechanical microsystems (such as
MEMS/NEMS). Knowledge gained from these studies will
not only extend our insights concerning biological adhe-
sion, but these results may also prove indispensable in
practical applications.
Acknowledgments This work is supported by the National Natural
Science Foundation of China through Grant No. 10442002,
10502031, 10628205, 10121202, Tsinghua Basic Research Founda-
tion, and National Basic Research Program of China through Grant
No. 2004CB619304, and SRF for ROCS, SEM.
Remark
q
sinu sin2 uHXup HX up 2sin2 uup
H
123
8892 J Mater Sci (2007) 42:88858893

Table 1 continued
Meridional curvature Average Profile Profile radius Remark
curvature radius
du
p
b \0 H 6 0 X>0 sinu sin2 uHX up HX up 2sinuup
dS Xu H

8 p
> 2
< sinu sin uHX up HXup 2sinuup du > 0
du p
H dS
c \0 or H 6 0 X>0 Xu
dS > 2
: sinu sin uHXup HX up 2sinuup du \0
du
dS
0 or ( H dS
2 X
du
\0 Inflexion Xc  Hup HX up  2sinuup > 0
dS du
dS
0 sin2 uc HXup HX up  2sinuup M0\M  1

du
d dS
0 (global) H 6 0 X>0 X H1 u p2

du
e dS
0 (global) H0 X>0 sin u = 0 (u = 0 oru = p)

du X up sinuup
f dS
6 0 or du
dS
0 (local) H 0 X>0 Xu sinu

123
J Mater Sci (2007) 42:88858893 8893

Table 1 continued
Meridional curvature Average Profile Profile radius Remark
curvature radius

g du
dS
6 0 H 6 0 X0 Xu 2sin
H
u
Zu  2cos
H
u

(local)

Note: Meridional curvature refers to du

dS
H  sin
X
u
, Local means position of a point or several points, and global means the whole line of the
liquid profile

References 13. Huber G, Mantz H, Spolenak R, Mecke K, Jacobs K, Gorb SN,

1. Gao H, Wang X, Yao H, Gorb S, Arzt E (2005) Mech Mater
37:275
2. Federle W, Riehle M, Curtis ASG, Full RJ (2002) Integr Comp
Biol 42:1100
3. Gao H, Ji B, Buehler MJ, Yao H (2004) Mol Cel Biol 1:37
4. Jiao YK, Gorb S, Scherge M (2000) J Exp Biol 203:1887
5. Gorb S, Gorb E, Kastner V (2001) J Exp Biol 204:1421
6. Autumn K, Peattie AM (2002) Integr Comp Biol 42:1081
7. Kesel AB, Martin A, Seidl T (2004) Smart Mater Struct 13:
512
8. Autumn K, Sitti M, Liang YCA, Peattie AM, Hansen WR,
Sponberg S, Kenny TW, Fearing R, Israelachvili JN, Full RJ
(2002) Proc Natl Acad Sci U S A 99:12252
9. Arzt E, Gorb S, Spolenak R (2003) Proc Natl Acad Sci USA
100:10603
10. Orr FM, Scriven LE (1975) J Fluid Mech 67:723
11. Rabinovich YI, Adler JJ, Esayanur MS, Ata A, Singh RK,
Moudgil BM (2002) Adv Colloid Interface Sci 96:213
12. Choe H, Hong MH, Seo Y, Lee K, Kim G, Cho Y, Ihm J, Jhe W
(2005) Phys Rev Lett 95
Arzt E (2005) Proc Natl Acad Sci USA 102:16293
14. Sun WX, Neuzil P, Kustandi TS, Oh S, Samper VD (2005)
Biophys J 89:L14
15. Qian J, Gao HJ (2006) Acta Biomater 2:51
16. Spolenak R, Gorb S, Gao H, Arzt E (2005) Proc R Soc London
Ser A-Math Phys Eng Sci 461:305
17. Federle W, Brainerd EL, McMahon TA, Holldobler B (2001)
Proc Natl Acad Sci USA 98:6215
18. Gorb S, Jiao YK, Scherge M (2000) J Comp Physiol A-Sens
Neural Behav Physiol 186:821
19. Gao H, Yao H (2004) Proc Natl Acad Sci USA 101:7851
20. Scherge M, Gorb SN (2000) J Micromech Microeng 10:359
21. Langer MG, Ruppersberg JP, Gorb S (2004) Proc R Soc Lond Ser
B-Biol Sci 271:2209
22. Saito S, Motokado T, Obata KJ, Takahashi K (2005) Appl. Phys
Lett 87
23. Obata KJ, Motokado T, Saito S, Takahashi K (2004) J Fluid
Mech 498:113
24. Fortes MA (1982) Powder Metall Int 14:96
25. Gao H, Ji B, Jager IL, Arzt E, Fratzl P (2003) Proc Natl Acad Sci
USA 100:5597

123
J Mater Sci (2007) 42:88948903
DOI 10.1007/s10853-007-1901-6
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Multiscale model of electronic behavior and localization
in stretched dry DNA
Ryan L. Barnett Paul Maragakis Ari Turner
Maria Fyta Efthimios Kaxiras
Received: 9 January 2007 / Accepted: 2 April 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract When the DNA double helix is subjected to
external forces it can stretch elastically to elongations
reaching 100% of its natural length. These distortions,
imposed at the mesoscopic or macroscopic scales, have a
dramatic effect on electronic properties at the atomic scale
and on electrical transport along DNA. Accordingly, a
multiscale approach is necessary to capture the electronic
behavior of the stretched DNA helix. To construct such a
model, we begin with accurate density-functional-theory
calculations for electronic states in DNA bases and base
pairs in various relative configurations encountered in the
equilibrium and stretched forms. These results are com-
plemented by semi-empirical quantum mechanical calcu-
lations for the states of a small size [18 base pair
poly(CG)poly(CG)] dry, neutral DNA sequence, using
R. L. Barnett
A. Turner
M. Fyta
E. Kaxiras
Department of Physics, Harvard University, Cambridge,
MA 02138, USA
Present Address:
R. L. Barnett
Department of Physics, California Institute of Technology,
Pasadena, CA 91125, USA
P. Maragakis
Department of Chemistry and Chemical Biology,
Harvard University, Cambridge, MA 02138, USA
Present Address:
P. Maragakis
D.E. Shaw Group, 120 West Forty-Fifth St., New York,
NY 10036, USA
E. Kaxiras (&)
School of Engineering and Applied Sciences,
Harvard University, Cambridge, MA 02138, USA
e-mail: kaxiras@physics.harvard.edu
123
previously published models for stretched DNA. The cal-
culated electronic states are then used to parametrize an
effective tight-binding model that can describe electron
hopping in the presence of environmental effects, such as
the presence of stray water molecules on the backbone or
structural features of the substrate. These effects introduce
disorder in the model hamiltonian which leads to electron
localization. The localization length is smaller by several
orders of magnitude in stretched DNA relative to that in the
unstretched structure.
Introduction
Soon after Watson and Cricks discovery of the DNA
double-helix structure [1], Eley and Spivey [2] introduced
the notion of efficient charge transport along the stacked p
orbitals of the bases. The mechanism of charge transport
has been the subject of numerous studies in the intervening
years, with renewed interest fuelled recently by both bio-
logical and technological considerations. Over a decade
ago, Barton and co-workers observed distance-independent
charge transfer between DNA-intercalated transition-metal
complexes [3] and argued that it would be relevant for
biology and biotechnology. More recent electron transport
experiments on DNA have yielded widely varying results,
showing alternatively insulating behavior [48], semicon-
ducting behavior [9], Ohmic conductivity [1013], and
proximity induced superconductivity [14]. The large
number of relevant variables endemic to such experiments,
like the DNA-electrode contact, and the rich variety of
structures that DNA can assume, are the causes of vari-
ability in the experimental measurements (for a recent
review of transport theory and experiments see Ref. [15]).
J Mater Sci (2007) 42:88948903
Specifically, there is a large diversity of the DNA forms
in terms of its composition, length, and structure. Experi-
ments done long ago, suggested that DNA substantially
longer than its natural length (also referred to as over-
stretched DNA) can undergo a transition to an elongated
structure up to twice the length of relaxed DNA [16]. This
was also confirmed by recent single molecule stretching
experiments [1719], which showed that the molecule can
be reversibly stretched up to 90% of its natural length. Such
important deformations of the double helix may occur in
biological environments. Stretching of DNA is also related
to cellular processes, such as transcription and replication.
For example, proteins often induce important local distor-
tions in the double helix while they diffuse along the mol-
ecule in search of their target sequences. The electronic and
transport properties of DNA are directly influenced by its
different conformations as well as by environmental factors,
such as counterions, impurities or temperature. A full
account of these effects based on a realistic, atomic scale
description of the structure and the electronic properties
challenges the capabilities of theoretical models.
Theoretical efforts to understand the electronic behavior
and transport in DNA can be divided into two general
categories:
(i) Model calculations that use effective hamiltonians
and master equations to describe the dynamics of electrons
and holes in DNA (see, for instance, Refs. [2023]). Recent
results [24] have led to considerable insights concerning
the sequence-independent delocalization of electronic
states in DNA. The main limitation of such approaches lies
in the difficulty of determining accurate values for the
parameters in the effective hamiltonians.
(ii) Ab initio calculations that can provide an accurate
and detailed description of the electronic features [2527].
These approaches are typically limited to a small number
of atoms due to computational costs, and cannot readily
handle the full complexity of DNA molecules in various
conformations. In particular, stretching of DNA can induce
a very significant deviation from the B form which is stable
under normal conditions in aqueous solution. Such struc-
tural distortions are bound to have a profound effect on the
electronic behavior. A realistic description of these effects
makes it necessary to handle both the atomic scale features
and the overall state of the macromolecule.
In the present work, we address the problem of DNA
stretching effects on the electronic states and the electron
localization by providing a bridge between the two
extremes of the length scale; a similar methodology was
recently used to study hole transfer in DNA [28]. Theo-
retically, there are different ways of pulling the opposite
ends of the DNA strands, leading to different stretched
DNA forms, which are determined largely by base pair
reorientations. Here, we use the poly(CG)poly(CG)
8895
structures obtained in the pioneering study of Lebrun and
Lavery [29] as the representative structure for stretching
effects. This study modeled the adiabatic elongation of
selected DNA molecules in two modes of stretching, cor-
responding to pulling on opposite 33 ends or 55 ends
of the molecule: In the 33 stretching mode, the DNA
helix is unwound leading to a ribbon-like structure, while
in the 55 stretching mode the DNA helix contracts.
We begin with a set of detailed calculations for the
electronic structure of DNA bases (A,T,C,G) and repre-
sentative base pairs (AT-AT, CG-CG, AT-CG, CG-GC) in
various relative configurations, as they are likely to appear
in the stretched forms, These calculations are based on
density-functional theory [30, 31] and serve to set the stage
for more extensive calculations which employ successive
levels of approximations necessary to handle the compu-
tational demands. Specifically, we extract the salient fea-
tures of electronic structure of the individual DNA bases
and base pairs from the ab initio calculations; these are
compared to an efficient and realistic semi-empirical model
[32], in order to establish the validity of the latter approach.
At this intermediate scale, we consider an 18 base pair
poly(CG)poly(CG) DNA sequence which has been stret-
ched by 30%, 60% and 90% relative to the natural length of
the unstretched B form. The atomic structure of these
forms has been established by Lebrun and Lavery [29],
using empirical interatomic potentials. We next use the
information from this approximate description to build an
effective hamiltonian for the electronic behavior at much
larger scales. This allows us to describe electron localiza-
tion, due to the combined effects of stretching and envi-
ronmental factors, over mesoscopic to macroscopic length
scales. The essence of the approach and the different scales
involved are shown schematically in Fig. 1. We emphasize
that we address here issues related only to dry and neutral
DNA structures, where the negatively charged groups on
the backbone are passivated by protons, conditions that are
relevant to the experiments we consider for comparison to
our theoretical results; water molecules or counterions
(such as Na+) are not considered in our calculations.
Theoretical methods
Ab initio calculations
As our first step toward establishing the electronic behavior
of dry, neutral DNA, we study the nature of electronic
states in individual bases and in base pairs. For these cal-
culations we used three different implementations of den-
sity-functional theory [30]: a method that uses atomic-like
orbitals as the basis [34], one that uses plane waves [35]
and a third that uses a real-space grid [36]. In all three
123
8896
nm
abinitio semiempirical effective
density electronic tightbinding
functional theory structure hamiltonian
100
10
1500 bp
18 bp
1 weight of orbital |un to the electronic wavefunction.
1 bp
30 1130 94000 atoms
Fig. 1 Schematic illustration of the different scales included in the
current multiscale model: The two pictures on the left are atomistic
systems simulated with different computational approaches (ab initio
density functional theory and semi-empirical electronic structure,
resprectively). The picture on the right represents a rope composed of
DNA molecules, as in experiments [33], which is treated by an
effective tight-binding hamiltonian constructed from the atomistic
scale calculations
approaches, we used the same exchange-correlation func-
tional in the local-density approximation [31], for consis-
tency and simplicity. More elaborate approximations to
exchange-correlation effects, such as the generalized gra-
dient approximation [37], do not provide any improvement
in describing the physics of these weakly interacting units.
In each method we used pseudopotentials to represent the
atomic cores, of the Trouiller-Martins type [38] in SIES-
TA, the Vanderbilt ultrasoft type [39] in VASP and the
HammannSchluterChiang type [40] in HARES, with
computational parameters (number of orbitals in basis,
plane-wave kinetic energy cutoff and grid spacing) that
ensure a high level of convergence. These calculations
provide a thorough check on the consistency of various
computational schemes to reproduce the electronic features
of interest. The results are in excellent agreement across
the three approaches. Since in these calculations there are
no adjustable parameters, we refer to them in the following
as ab initio results.
Construction of semi-empirical model
The stretched forms contain a large number of atoms,
typically beyond what can be efficiently treated with the
123
J Mater Sci (2007) 42:88948903
ab initio methods used for the DNA bases and base pairs.
Accordingly, for the electronic structure calculations of
these structures we use an efficient semi-empirical quan-
tum-mechanical approach which employs a minimal basis
set [32]. The consistency of this approach is then verified
against the ab initio calculations. Within the semi-
empirical scheme, the electronic eigenfunctions are
expressed as
X
jwn i cn
m jum i 1
m
where the basis set |un includes the s and p atomic
orbitals for each atom in the system. The coefficients
c(n)m are numerical constants, with |c(n)m|2 giving the
This method uses a second order expansion in the elec-
tronic density to obtain the total energy and takes into
account self-consistently charge transfer effects which
are important for biological systems. The method gives
results for the band gaps that are in excellent agreement
with those of the ab initio approaches described above
(see Refs. [5, 41]).
The highest occupied and lowest unoccupied molecular
orbitals (HOMO and LUMO, respectively, also referred
to collectively as frontier states in the following) are
extended over the entire structure in Bloch-like wave
functions. In order to describe electron hopping and
localization, we need to express these in terms of a basis of
Wannier-like states that are localized on the individual
bases. To this end, we construct maximally localized states
on single base pairs by taking linear combinations of the
HOMO and LUMO states from the wavefunctions of Eq. 1.
The maximally localized states will then be used to cal-
culate the hopping parameters in the effective 1D hamil-
tonian. Using the extended electronic states |w(n) of the
frontier states, with corresponding energies e(n), we define
the maximally localized states jw ~ i i through the unitary
transformation
X
~ i i
jw ~ i ijwn i
hwn jw 2
n
which minimizes the sum of the variances
X

f ~ i j^z2 jw
hw ~ i i  hw
~ i j^zjw
~ i i2 3
i
under the constraint hw ~ j i dij where z is the position
~ i jw
along the helical axis. Similar and more general
methodologies have been developed in the past for
obtaining maximally localized states from extended ones
[42, 43]. Due to the invariance of the trace, the first term in
Eq. 3 is independent of the unitary transformation and the
J Mater Sci (2007) 42:88948903 8897

problem is simplified to one of maximizing the second term P P


He cyn cn t1 cyn cn1 cyn1 cn
on the right-hand side with the same orthonormality n n even

P y
constraint. Carrying out the minimization, we arrive at t2 cn cn1 cyn1 cn 9
the equation n odd

P y
t3 cn cn2 cyn2 cn
n
~ n
hw j^
zj w ~ m izn  zm 0 4
where n represents the nth base pair along the helical axis
where and we have neglected spin indices because they are
unimportant for our analysis. Note that there is a difference
~ n j^
zn hw ~ n i:
zjw 5 between hopping elements connecting even and odd sites
to their neighbors (t1 and t2 terms in the effective
By inspection, we see that f is maximized when zn = zm hamiltonian of Eq. 9), due to the asymmetry in the
for all m and n, corresponding to maximally delocalized structure illustrated in Fig. 2. Performing a Fourier
states. On the other hand, f is minimized when the states transform on the electron creation and annihilation
~ n i are the eigenfunctions of the position operator ^z
jw operators
within the HOMO or LUMO subspace. Therefore, the
problem is further reduced to constructing and diagonal- 1 X ikn
izing the matrix ck p e cn 10
N n
Mnm hwn j^
zjwm i 6 gives a hamiltonian which has coupling between momenta
~ i i that provide the k and k + p/a. By doubling the unit cell (and reducing the
which has the eigenvectors hwn jw
Brillouin Zone by a factor of two), this can finally be
desired transformation given in Eq. 2. The eigenvalues zn
diagonalized to obtain the eigenvalues
are the positions of the localized states. To evaluate the
matrix elements we use the approximation p
Ek e 2t3 cos2k t12 t22 2t1 t2 cos2k 11
P n m with the momentum sum carried out over the reduced
hwn j^
zjwm i cl cm hul j^
zjum i
lm
7 Brillouin Zone. With these expressions for the band
P n m
 cl cm Slm zlm structure energies, the density of states (DOS)
lm

where Slm hul jum i is the overlap matrix between the two 1X n
z z gx dx  Ek 12
atomic orbitals and zlm l 2 m is the average z-value for the N k;n
atoms located at sites given by the labels l and m. Once the
localized states are constructed, the hopping parameters can be readily obtained. These quantities are essential in
can be computed as describing electron localization along the DNA double
helix under different conditions.
X
~ i jHjw
tij hw ~ j i ~ i jwn ihwn jw
en hw ~ j i 8
n

~ i i are determined from

recalling that the quantities hwn jw
the transformation described above.
G C
Having defined the maximally localized states in terms
of the electronic wavefunctions from the all-atom calcu- C G
lations, we next produce an effective tight-binding hamil- (2)
tonian, which allows us to study electron hopping along the (3)
G C
(1)
DNA double helix. This approach has also been used in a C G
recent study on functionalized carbon nanotubes [44]. In
our effective hamiltonian, we consider hopping between G C
first and second neighbors along the helix, and denote the
hopping matrix elements according to the scheme shown in Fig. 2 Schematic depiction of electron hopping in poly(CG)
poly(CG) DNA for the HOMO state. The hopping matrix elements
Fig. 2 for the HOMO state of the poly(CG)poly(CG)
ti are denoted by the indices (i) = (1), (2), (3). Electrons are localized
structure (all other frontier states involve exactly the same on the G bases. For the LUMO state, the hopping is similar with
type of hopping matrix elements): electrons localized on the C bases

123
8898 J Mater Sci (2007) 42:88948903

Disorder and localization length

In order to quantify the amount of localization that is

expected in stretched DNA forms, we add a term to the
hamiltonian in Eq. 9 of the form
X
Hdis Un cyn cn 13
n

which is meant to emulate disorder arising from a variety of

sources such as interaction of the DNA bases with stray water
molecules and ions, or interaction with the substrate. Un are
uncorrelated random energy variations chosen according to a
Gaussian distribution of zero mean and width c

 
1 U2
PU p exp  2 : 14
c 2p 2c
Fig. 3 The DNA base pairs AT (top) and CG (bottom), with the
Once the disorder hamiltonian is constructed with a specific atoms labeled. The purines (A, G) are on the right, the pyrimidines (T,
C) on the left. Atom labeling follows standard notation convention
set of random on-site energies, by direct diagonalization we [46]. All rotations were performed with respect to the helical axis
find the eigenstates jWi i of H Hdis (we use capital denoted by the black circle (see text)
symbols to denote the new wavefunctions from the
hamiltonian that includes the disorder term) and then
calculate the localization length defined as in Fig. 4. Specifically, the HOMO state of the AT pair is
h i1=2 exactly the same as that of the HOMO state of the isolated
Li hWi j^ njWi i2
n2 jWi i  hWi j^ 15 A, and the LUMO state of AT the same as that of the
isolated T. Similarly, the HOMO state of CG is identified
where n^ is the position operator along the DNA helix with that of the isolated G and the LUMO state with that of
defined as:
X
n^ ncyn cn 16 T
n

For a single-hopping model with weak disorder, the A

localization length scales as L ~ (t/c)2 for electrons near the
middle of the band [45], with t the hopping matrix element
which determines the band width. The more complicated AT

effective hamiltonian considered here is not amenable to

simple analytic treatment.
C

Results and discussion G

We begin our discussion with an overview of electronic

states in single bases and isolated base pairs. The structure
of the base pairs is shown in Fig. 3 with the atoms in each
base labeled for future reference. These calculations will
set the stage for a proper interpretation of the behavior in
the stretched and unstretched dry, neutral DNA helix.
Frontier states
The frontier states in the base pairs are related to only one
component of the pair for both AT and CG. This is shown
CG
Fig. 4 The frontier states in the base pairs and their identification
with corresponding orbitals in the isolated bases. The middle figure in
each panel shows the total charge density on the plane of the base
pair, with higher values of the charge density in red and lower values
in blue. The figure on the left shows the HOMO state and the figure on
the right shows the LUMO state, where pink and light blue
isosurfaces correspond to positive and negative values of the
wavefunctions. The labels on the left denote the type of bases and
base pairs

123
J Mater Sci (2007) 42:88948903
the isolated C. Thus, the purines (A or G) give rise to the
HOMO state, while the pyrimidines (T or C) are respon-
sible for the LUMO states of each pair. It is clear from the
same figure, that essentially all atomic pz orbitals which
belong to a purine or pyrimidine contribute to the respec-
tive HOMO or LUMO p state of the base pair. This is in
agreement with calculations on the optical absorption
spectra of DNA bases and base pairs [47]. A closer
inspection of Fig. 4 shows that the molecular frontier states
of both AT and CG can be identified as similar contribu-
tions (up to sign changes) from specific groups of carbon
and nitrogen atoms. Specifically, in the purines (A and G)
three distinct groups of atoms are mainly involved in
forming the HOMO orbital and include atoms (C8N7),
(C2N3) and (N1C6C5C4N9), respectively. In the
pyrimidines (T and C) the main groups involved in forming
the LUMO orbital are two, (C4C5N1) and (N3N7C6).
In both base pairs the atoms that are less involved in the
frontier molecular states are the carbon atoms that form a
double bond with an oxygen atom, such as C2 of A and C
and the four-fold bonded C7 atom of A.
The frontier states are very little affected when the two
components of the base-pair are separated along the
direction in which they are hydrogen-bonded. To demon-
strate this, we show in Fig. 5 the change in the eigenvalues
of the frontier states in AT and CG as a function of the
distance between the two atoms that are bonded to the two
backbones (we call this the backbone distance). For both
base pairs the nitrogen atoms labeled N1 and N9, are the
ones attached to the backbone (see Fig. 3). In order to
obtain realistic structures, for each value of the backbone
distance we hold the atoms of each base that are bonded to
the backbone fixed and allow all other atoms to relax fully.
2.5
TLUMO
(eV)
AT CG
0
AHOMO
2.5
7 8 9 10 11 7 8 9 10
backbone distance () backbone distance ()
o o
Fig. 5 Eigenvalues of states in the AT and CG base pairs as a
function of backbone distance. In each case three states are included
above and below the band gap. Lines are results from SIESTA
calculations, points are results from HARES calculations (see text).
The frontier orbitals in both pairs are related to one component of the
pair as indicated by the labels. The equilibrium backbone distance is
denoted by a vertical dashed line
8899
These calculations were performed with the SIESTA code
[34] and the relaxed configurations were used as input to
calculate the electronic structure with the other two
methodologies [35, 36]. In Fig. 5 we show complete results
from the SIESTA calculations and selected results from
one of the other two approaches.
The results of Fig. 5 show clearly that only in the region
where the backbone distance becomes significantly smaller
than the equilibrium value, interaction between the two
bases shifts the eigenvalues of the electronic states appre-
ciably, but even then the shifts are relatively small for the
frontier states. It is also noteworthy that the band gap of the
AT pair is significantly larger (~3 eV) than that of the CG
pair (~2 eV) and that the frontier states of CG lie within the
band gap of the AT pair. This observation is important
because it indicates that in an arbitrary sequence of base
pairs, the frontier states will be associated with those of the
CG pairs. This statement is verified by calculations of
electronic states in the AT-AT, CG-CG and AT-CG base
pair combinations, to which we turn next.
For more detailed comparisons, we collect in Table 1
the eigenvalues of the frontier states for the DNA pairs and
the pair combinations, at different equilibrium configura-
tions in the three relevant variables, the backbone distance,
the axial distance and the rotation angle. Some results on
the CG-GC base pair combination are also shown, to allow
for comparison to the poly(C)poly(G) sequence.
When two base pairs are stacked on top of each other,
there are two degrees of freedom for motion of one relative
to the other: a separation along the helical axis, which we
will call axial distance, and a relative rotation around the
helical axis. We take the helical axis to be that which
corresponds to stacking of successive base pairs in the B
form of the DNA double helix. According to the notation of
Fig. 3, the helical axis for both base-pairs is normal to the
line connecting atoms C4 and C6 and is closer (about one
third of their distance) to the purine atom C6. For each
configuration we fix the atoms that are bonded to the
CLUMO backbone at a given relative position and allow all other
atoms to relax, as was done in the calculations involving
the backbone distance discussed above. In Fig. 6 we show
the behavior of electronic eigenvalues as a function of the
GHOMO
axial distance and the rotation angle. As above, the
eigenvalues show little dependence on these two variables,
except for rather small values of the axial distance which
11 correspond to unphysically small separation between the
two base pairs.
What is also remarkable in the above results, is that in
the AT-CG combination, the frontier states are clearly
identified with those corresponding to the CG pair exclu-
sively, which has the smaller band gap (see Fig. 6).
Moreover, we note that the band gap of the poly(C)
poly(G) sequence, as calculated by the semi-empirical
123
8900 J Mater Sci (2007) 42:88948903

Table 1 Eigenvalues (in eV) of

min HOMO LUMO gap
the frontier states for the DNA
base pairs and the base-pair Backbone distance
combinations, at the equilibrium
AT
8.67 A 1.63 1.60 3.23
configurations for the backbone
distance, the axial distance (at CG
8.73 A 0.80 1.31 2.11
zero relative angle of rotation) Axial distance
and the angle of rotation (at the
AT-AT 3.67 A 1.33 1.37 2.70
equilibrium axial distance)
CG-CG
3.52 A 0.46 1.95 1.41
AT-CG
3.36 A 0.71 1.00 1.71
Rotation angle
AT-AT 36 1.48 1.58 3.06
108 1.45 1.68 3.13
180 1.55 1.63 3.18
CG-CG 36 0.52 1.22 1.74
108 0.64 1.54 2.18
180 0.94 1.60 2.54
The column labeled min CG-GC 36 0.86 1.51 2.37
gives the values of the distances 108 0.66 1.43 2.09
and the angle at the equilibrium
configurations. Due to 180 0.60 1.12 1.72
symmetry the values for the AT-CG 36 0.73 1.38 2.11
minima at rotation angles larger 108 0.59 1.27 1.86
than 180 are similar to those
180 0.81 1.25 2.06
given here and are not shown

method based on a minimal atomic orbital basis [32] is in
excellent agreement with the value obtained from the
SIESTA calculation (2.0 eV and 2.1 eV, respectively). The
band gap is expected to be significantly smaller in the case
2.5
(eV)
of wet DNA and in the presence of counterions, as shown
in Ref. [48], for a Z-DNA helix. The band gaps between all
three ab initio methods are identical within the accuracy of
these methods. The nature of electronic wavefunctions
obtained by the different methods is also in good qualita-
tive agreement. Accordingly, in the rest of this paper we
focus our attention to electron localization in the dry,

TLUMO

0 ATAT neutral poly(CG)poly(CG) sequence, and employ the

AHOMO results of the semi-empirical electronic structure method.
2.5
2.5
Hopping electrons
(eV)

CLUMO

0 CGCG
GHOMO
In Fig. 7, we show the unstretched and the three stretched
2.5 forms of the poly(CG)poly(CG) sequences at 30%, 60%,
2.5
90% elongation, along with the features of the frontier
(eV)

ATCG
CLUMO
states. For visualization purposes, we represent the calcu-
0 GHOMO
lated wavefunction magnitude of the frontier states by blue
2.5 (HOMO) and red (LUMO) spheres, centered at the sites
2.9 3.4 3.9 4.4 4.9 5.4 06 0 120 180 240 300 360
axial distance ()
o

angle (deg.)
Fig. 6 Eigenvalues of states in the AT-AT, CG-CG and AT-CG base
pair combinations as a function of the distance along the helical axis
(at zero angle of rotation) and the rotation angle around the helical
axis (at the equilibrium axial distance). Lines are results from
SIESTA calculations, points are results from VASP calculations (see
text). In each case three states are included above and below the band
gap. The value of the distance or the rotation angle that correspond to
equilibrium configurations are indicated by vertical dashed lines
(there are five almost equivalent local minima in rotation). As in
Fig. 5, frontier orbitals are identified as the corresponding orbital of
one base only
where the atomic orbitals are located. The radius of the
sphere centered on a particular atom is proportional to the
magnitude of the dominant coefficient |c(n)m|2 at this site
(see Eq. 1), which is essentially proportional to the local
electronic density. It is evident from this figure that the
nature of the orbitals themselves, represented by the radii
of the colored spheres, does not change much in the dif-
ferent stretched DNA forms, but the overlap between
orbitals at neighboring bases is affected greatly by the
amount of stretching. For the poly(CG)poly(CG)
sequence shown, the HOMO orbitals are always associated

123
J Mater Sci (2007) 42:88948903 8901

involve only the G sites; those for the LUMO state involve
only the C sites. As a consistency check, we have also
calculated matrix elements for farther neighbors and found
33 55
those to be much smaller in magnitude. We have calculated
the values of t1,t2,t3 by repeating the same procedure as
(a) above for the stretched forms of the poly(CG)poly(CG)
DNA sequence. We note that if t2 = t3 = 0 electrons will
not be able to migrate along the DNA molecule even if t1 is
quite large, because at least one of the other two hops is
(b) necessary for migration (see Fig. 2). From this simple
picture, it is evident that the conductivity will be deter-
5
C G mined by which matrix element dominates. Quantitatively,
3 the bottleneck hopping matrix element is given by
(c)
3
G C 5 t maxminjt1 j; jt2 j; jt3 j: 17

Fig. 7 The DNA structures for the unstretched (top) and the different
amounts of stretching in the 33 and the 55 modes with features of ping matrix element calculated as a function of stretching.
the frontier orbitals described by the blue (HOMO) and red (LUMO)
spheres (see text for details). For both modes the amount of stretching
is (a) 30%, (b) 60%, and (c) 90% relative to the unstretched structure,
which is the B-DNA form. The 35 orientations of the poly(CG) the molecule and that the hopping will decrease more from
poly(CG) sequence are shown in the left panel at 90% stretching,
where the structure is easier to visualize
with the G sites for all the stretching modes, while the
LUMO orbitals are related to the C sites. However, as the
DNA becomes more elongated, the orbitals overlap even
less and become localized for high stretching modes. The
elongation to the overstretched form is achieved by
changing the dihedral angle configuration of the DNA
backbone, which leaves the local part of the orbitals
essentially intact. Note how the orbitals rotate and spread
out as the structure is being ovestretched, following the
rotation of bases.
We now turn to a discussion of the results for the hop-
ping matrix elements of Eq. 9. Our discussion here is rel-
evant to what happens when the occupation of a frontier
state is changed from complete filling (for the HOMO) or
complete depletion (for the LUMO), that is, the physics of
In Fig. 8 we show the value of the bottleneck hop-
This indicates that hopping conductivity will dramatically
decrease by several orders of magnitude upon stretching
stretching in the 33 mode than in the 55 mode. This is
due to the conformational changes induced by the different
stretching modes, described earlier.
Localization length
The significant dropping of the hopping matrix elements
upon stretching as described in the previous section is
indicative of electron localization with a weak amount of
disorder. To investigate this possibility in detail, we focus
on effects of stretching in the 33 mode. The evolution of
the density of HOMO states upon stretching is shown in
101
(2)
(3)
101 (2)
(2)

small amounts of hole or electron doping. In Table 2 we (2)

(2)
give the values for e; t1 ; t2 ; t3 (see Fig. 2) for the two
t (meV)

(2)

frontier states of the unstretched poly(CG)poly(CG) DNA 103

(2)
(3)
form. The hopping matrix elements for the HOMO state
HOMO 33
HOMO 55 (3)
5
10 LUMO 33
Table 2 Parameters for the on-site (e) and hopping matrix elements
LUMO 55 (3)
(ti, i = 1,2,3), for the HOMO and LUMO states of unstretched
poly(CG)poly(CG) DNA
(2)
HOMO LUMO 107
0 30 60 90
% stretching
e (eV) 3.12 0.09
t1 (meV) 14.0 0.29 Fig. 8 The frontier state bottleneck hopping matrix elements as
t2 (meV) 2.60 0.04 given by Eq. 17 for the different types (33 or 55) and amounts of
t3 (meV) 0.09 0.26 stretching of poly(CG)poly(CG) DNA. At each value of stretching,
the dominant hopping process is indicated in parenthesis

123
8902 J Mater Sci (2007) 42:88948903

Fig. 9; similar behavior is observed for the LUMO states.
The dramatic narrowing of the DOS width (equivalent to
reduced dispersion in a band-structure picture) is strongly
suggestive of electron localization [49], in this case
induced by stretching. This localization length is controlled
by the hopping elements t, since e is the same at each site.
For a more quantitative description, we show in Fig. 9
the localization length Li for each eigenstate for a 1500
base-pair DNA strand under different amounts of stretch-
ing. The value of L(i) for each state is obtained from Eq. 15,
with disorder strength c = 0.3 meV, which determines the
width of the gaussian given in Eq. 14. This disorder
strength is much smaller than the band width of the
unstretched DNA, but becomes comparable to the band
width as the molecule is stretched. The magnitude of such
variations in on-site energies is consistent with those
1000
100
L (base pairs)
produced by the dipole potential terms, for instance, due to
the presence of a stray water molecule situated on the
substrate roughly 15 A away from the DNA bases. We find
that changing the value of c by an order of magnitude
(either smaller or larger) does not affect the qualitative
picture presented here. Note that the localization length is
not a strict function of the energy, as it depends on the
disorder near where a given state happens to be localized.
As the molecule is stretched, the localization length
dramatically decreases until, for 60% stretching, the
eigenstates are completely localized on single base pairs.
The charge localization length as a function of DNA
stretching has been recently studied in the experiment of
Heim et al. [33]. This study focuses on k-DNA which has
an irregular sequence of base pairs, and can be compared to
our theoretical results for poly(CG)poly(CG) recalling
that the frontier states even for a random sequence are
associated with those of the CG base-pairs. In the experi-
ment, ropes of k -DNA on a substrate are overstretched by
a receding meniscus technique. The DNA ropes in this
experimental setup are slightly positively charged, corre-
sponding to a depletion of a few electrons per 1000 base
pairs. We suggest that this situation is approximated by the
structures of dry and neutral DNA that we considered

above. Electrons were injected into the DNA and the

resulting localization length was measured by an electron
10
force microscope. For the unstretched DNA, the charge
was found to delocalize across the entire molecule,
extending over a length of several microns. On the other
1 hand, the charge injected into the overstretched DNA is
localized, extending over a few hundred nanometers only.
This is qualitatively consistent with the picture that emer-
ges from our theoretical analysis, and is even in reasonable
1 quantitative agreement: the degree of localization in
0% experiment, measured by the ratio of length scales going
30% from unstretched to stretched DNA structures, is approxi-
60%
0.8 mately two orders of magnitude, while the same quantity in
our calculations, going from unstretched to 60% stretched
DNA is ~103.
states / meV

0.6

0.4 Summary

We have described and implemented a multiscale method

0.2
to derive effective hamiltonian models that are able to
capture the dynamics of conduction and valence electrons
0 in stretched DNA, starting from ab initio, all-atom quantum
20 10 0 10 20
energy (meV) mechanical calculations. The ab initio simulations revealed
that the frontier states in the base pairs are related to only
Fig. 9 (bottom) The density of electronic states for the HOMO state one component of the pair. The purines were found to be
stretched in the 33 mode. For comparison, the on-site energy
parameter, e, has been set to zero. (top) The localization length Li,
associated with the HOMO states while the pyrimidines
defined in Eq. 15, is computed for each eigenstate with disorder with the LUMO states. In the AT-CG combination the
strength c = 0.3 meV frontier states are identified with those of the CG pair. For

123
J Mater Sci (2007) 42:88948903
all combinations of bases and base pairs studied here, the
nature of these states was not affected by separation of the
bases or base pairs along different directions or rotation
along the helical axis.
Turning to the next length scale and the semi-empirical
calculations, we have calculated the bottleneck matrix 18. Cluzel P, Lebrun A, Heller C, Lavery R, Viovy JL, Chatenay D,
elements for electron hopping along the DNA molecule, as
a function of stretching. These show a significant decrease
with elongation of DNA, which is stronger for stretching in
the 33 mode than in the 55 mode. We were able to 21. Iguchi K (2004) Int J Mod Phys B 18:1845
show quantitatively that stretching of DNA dramatically
narrows the DOS width of frontier states. A small amount
of disorder produced by environmental factors will natu-
rally lead to localization of the electrons along the DNA.
Our estimate for the degree of localization, based on a
reasonable (and quite small) amount of disorder in the
on-site energies for the electron states, is in very good
agreement with recent experimental observations. This
provides direct validation for the consistency and com-
pleteness of the multiscale method presented here.
Acknowledgments The authors are grateful to Richard Lavery for
providing the overstretched structures. MF acknowledges support by
Harvards Nanoscale Science and Engineering Center, funded by the
National Science Foundation, Award Number PHY-0117795.
References
1. Watson JD, Crick FHC (1953) Nature 171:737
2. Eley DD, Spivey DI (1961) Trans Faraday Soc 58:411
3. Murphy CJ, Arkin MR, Jenkins Y, Ghatlia ND, Bossmann SH,
Turro NJ, Barton JK (1993) Science 262:1025
4. Braun E, Eichen Y, Sivan U, Ben-Yoseph G (1998) Nature 391:
775
5. De pablo P, Moreno-Herrero F, Colchero J, Herrero J, Herrero P,
Baro A, Ordejon P, Soler J, Artacho E (2000) Phys Rev Lett
85:4992
6. Storm AJ, Van Noort J, Devries S, Dekker C (2001) Appl Phys
Lett 79:3881
7. Zhang Y, Austin RH, Kraeft J, Cox EC, Ong NP (2002) Phys Rev
Lett 89:198102
8. Baba Y, Sekiguchi T, Shimoyama I, Hirao N, Nath KG (2006)
Phys Rev B 74:205433
9. Porath D, Bezryadin A, Devries S, Dekker C (2000) Nature
403:635
10. Fink H-W, Schonenberger C (1999) Nature 398:407
11. Cai L, Tabata H, Kawai T (2000) Appl Phys Lett 77:3105
12. Tran P, Alavi B, Gruner G (2000) Phys Rev Lett 85:1564
13. Cohen H, Nogues C, Naaman R, Porath D (2005) Proc Natl Acad
Sci 102:11589
8903
14. Kasumov A, Kociak M, Gueron S, Reulet B, Volkov V, Klinov
D, Bouchiat H (2001) Science 291:280
15. Endres RG, Cox DL, Singh RRP (2004) Rev Mod Phys 76:195
16. Wilkins MHF, Gosling RG, Seeds WE (1951) Nature (London)
167:759
17. Smith SB, Cui YJ, Bustamante C (1996) Science 271:795
Caron F (1996) Science 271:792
19. Strick RT, Allemand JF, Bensimon D, Croquette V (2000) Annu
Rev Biophys Biomolec Struct 29:523
20. Yamada Y (2004) Int J Mod Phys B 18:1697
22. Yi JY (2003) Phys Rev B 68:193103
23. Chang CM, Neto AHC, Bishop AR (2004) Chem Phys 303:189
24. Caetano RA, Schulz PA (2005) Phys Rev Lett 95:126601
25. Barnett RN, Cleveland CL, Joy A, Landmand U, Schuster GB
(2001) Science 294:567
26. Artacho E, Machado M, Sanchez-Portal S, Ordejon P, Soler JM
(2003) Molec Phys 101:1587
27. Alexandre SS, Artacho E, Soler JM (2003) Phys Rev Lett
91:108105
28. Senthilkumar K et al (2005) J Am Chem Soc 127:14894
29. Lebrun A, Lavery R (1996) Nucl Ac Res 24:2260
30. Hohenberg P, Kohn W (1964) Phys Rev 136:B864; Kohn W,
Sham LJ (1965) Phys Rev 140:A1133
31. Perdew JP, Zunger A (1981) Phys Rev B 23:5048
32. Elstner M, Porezag D, Jungnickel G, Elsner J, Haugk M, Frau-
enheim T, Suhai S, Seifert G (1998) Phys Rev B 58:7260
33. Heim T, Melin T, Deresmes D, Vuillaume D (2004) Appl Phys
Lett 85:2637
34. Soler JM, Artacho E, Gale JD, Garca A, Junguera J, Ordejon P,
Sanchez-Portal D (2002) J Phys Condens Matter 14:2745
35. Kresse G, Furthmuller (1996) J Phys Rev B 54:11169
36. Waghmare UV, Kim H, Park IJ, Modine N, Maragakis P, Kaxiras
E (2001) Computer Phys Comm 137:341
37. Perdew JP, Wang Y (1992) Phys Rev B 45:13244; Perdew JP,
Burke K, Ernzerhof M (1996) Phys Rev Lett 77:3865
38. Trouiller N, Martins JL (1991) Phys Rev B 43:8861
39. Vanderbilt D (1990) Phys Rev B 41:7892
40. Hamann DR, Schluter M, Chiang C (1979) Phys Rev Lett
43:1494
41. Maragakis P, Barnett RL, Kaxiras E, Elstner M, Frauenheim T
(2002) Phys Rev B 66:241104(R)
42. Marzari N, Vanderbilt D (1997) Phys Rev B 56:12847
43. Sgiarovello C, Peressi M, Resta R (2001) Phys Rev B 64:115202
44. Lee Y-S, Buongiorno Nardelli M, Marzari N (2005) Phys Rev
Lett 95:076804
45. Thouless DJ (1972) J Phys C 5:77
46. Sinden RR (1994) In: DNA Structure and Function, Academic
Press, London
47. Varsano D, Felice RI, Marques MAL, Rubio A (2006) J Phys
Chem B 110:7129
48. Gervasio FL, Carloni P, Parrinello M (2002) Phys Rev Lett
89:108102
49. Anderson PW (1958) Phys Rev 109:1492
123
J Mater Sci (2007) 42:89048910
DOI 10.1007/s10853-007-1720-9
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
An lower bound on receptor density for stable cell adhesion due
to thermal undulations
Yuan Lin L. B. Freund
Received: 16 December 2006 / Accepted: 7 March 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract The adhesion of a living cell to an extracellular
matrix surface is effected through the bonding of receptor
molecules in the cell membrane to compatible ligand
molecules on the surface. In a series of experiments on
adhesion of cells to a substrate surface with a controlled
density of ligand binding sites, Arnold et al. (ChemPhys-
Chem 5:383, 2004) showed that tight cell adhesions could
form only if the areal density of binding sites on the sub-
strate was higher than some critical value. Furthermore,
this critical value was consistent across the four cell types
examined in the experiments. For ligand density below the
critical level, on the other hand, virtually no adhesions
formed. In this article, we examine the competition
between thermal undulations of the cell membrane and its
adhesion to the substrate. In particular, we show that
thermal undulations destabilize membrane bonding to the
substrate unless the bond spacing is below a certain level.
By following this line of reasoning in the context of clas-
sical statistical mechanics, we obtain an estimate of the
critical value of spacing which is in reasonable agreement
with the observations.
Introduction
One of the most extensively studied topics in modern bio-
physics is cell adhesion, a phenomenon central to cell
motility [1], endocytosis [2], and other biological processes.
Y. Lin
L. B. Freund (&)
Division of Engineering, Brown University, 184 Hope Street,
Providence, RI 02912, USA
e-mail: freund@brown.edu
123
For cell-to-cell or cell-to-extracellular matrix adhesion to
occur, binding proteins (integrins or ligands) within the cell
membrane must aggregate into compact clusters in order to
bond specifically with molecules outside the cell. These
adhesion patches can continue to grow by recruiting more
membrane bound adhesion molecules from the vicinity
[35]. This occurs naturally because the bonding of a
molecule in the membrane lowers the entropy of the binder
distribution locally, giving rise to an osmotic pressure
gradient which, in turn, induces binder transport into the
region. As is shown in [6], cell adhesion is influenced by
numerous factors including binding affinity between inte-
grins and ligands, the concentration of binding molecules,
the properties of the bonding surfaces, and so on.
Arnold and coworkers [7] have reported results of
experiments which revealed remarkable aspects of adhe-
sion at a size scale between the dimensions of the indi-
vidual bonds (a few nanometers) and the dimensions of a
tight adhesion patch (a few microns). They deposited gold
nanodots periodically on a polyethylene glycol based
substrate, chosen because it had no bonding affinity for cell
membranes. One binding molecule, an RGD ligand, was
then attached to each dot; a dot size of 8 nm or lessa
dimension below the diameter of a ligandwas chosen in
order to limit the number of molecules per dot to just one.
By controlling the spacing of the ligands in this way, it was
possible to observe the influence of density of ligands on
cell adhesion. No feature of the system other than the
spacing was modified from experiment to experiment. It
was reported in [7] that a critical spacing between dots
exists for a class of cell types. For dots spaced more closely
together than this critical spacing, tight cell adhesion pat-
ches formed. On the other hand, for spacings larger than
this critical value, almost no bonds between the cell and the
substrate were formed.
J Mater Sci (2007) 42:89048910
Motivated by this striking evidence, we examine the
relationship between the natural thermal undulations of a
cell wall immersed in a heat bath and the spacing of
binding sites necessary to effect adhesion to the substrate
surface. In particular, we show that the magnitude of
membrane fluctuations strongly depends on the average
binding energy density. Comparison of this fluctuation
magnitude at a reference point to the width of the energy
well, which is introduced to describe the binding interac-
tion, defines a natural value of spacing between binding
sites. For binding site spacing smaller than this critical
value, stable adhesion in the statistical sense can occur. For
binding site spacing larger than this critical value, however,
any bonds formed are unstable because thermal undulation
can ultimately overcome the bond resistance so the sur-
faces will not adhere.
The plan of this paper is as follows. In the following
section, a one-dimensional configuration is considered; this
relatively simple situation provides a convenient vehicle
for introducing the concepts and mathematical strategy
involved in the approach. Two models where adhesion is
treated either as being localized at discrete binding sites or
continuous across the interface are presented, and their
asymptotic behaviors as the membrane size becomes much
larger than the spacing between binding sites are examined.
The analysis is generalized to the more realistic two-
dimensional configuration in this section after that, and
comparisons between the theoretical prediction and
experimental observations are presented. Concluding
remarks are summarized in the final section.
A one-dimensional model
The phenomenon of thermal undulations of a membrane has
been studied by many authors, for example [8] and [9], in the
context of identifying the persistence length of a membrane.
Other situations in which the membrane and its substrate
interact have also been considered; see for example [10] and
[11]. The thermal fluctuations of fluid membranes in the
presence of periodic confining potentials were considered
by Gov and Safran [12] and the undulation spectra mem-
brane were examined by Merath and Seifert [13].
As a basis for arriving at the essential result in the
simplest way, consider a nominally flat membrane as
shown in Fig. 1. The membrane surface is immersed in a
thermal bath and is positioned near a substrate with
potential binding sites. For the one-dimensional model,
deflection is assumed to vary in only one direction, along
which spatial position is identified by the x coordinate. The
potential binding sites are identified by the circular sym-
bols on the membrane in the figure. Denote the total
number of binding sites by Nb. For simplicity, assume that
8905
z
x
Fig. 1 Schematic diagram of an elastic membrane with length L.
Potential binding sites are uniformly distributed along the membrane
with regular spacing D. The membrane is constrained against
transverse deflection at both ends
Nb is an odd number and that the binding sites are dis-
tributed along the membrane at equal intervals of D; see
Fig. 1. The extent of the membrane in the x direction is L,
so L = (Nb + 1)D. The reference plane of the membrane
coincides with L/2 x L/2, z = 0 as shown in
Fig. 1.
In order to describe adhesion, the interaction between
the membrane and the substrate at potential binding sites
must be prescribed. Here, adhesion is represented by means
of a potential energy well, with the depth of the well cor-
responding to the energy reduction achieved when bonding
occurs in the absence of other physical effects. In other
words, it is the chemical potential of bonding. The width of
the potential well represents the compliance of the bond.
We believe this is a realistic description of bonding when
the molecules involved in adhesion can be relatively large
and compliant. As suggested in Lin et al. [14], a conve-
nient choice for the shape of such a well is
h
Cb
 X 2 i
ub q kTln e kT  1 e2kT q 1 1
where ub is the bonding interaction energy, q is the
deflection of membrane in the z direction at this binding
site and kT is the thermal energy unit. Cb is the depth of the
energy well at deflection q = 0 and W is the curvature of
well at the same point. Notice that, in the one-dimensional
model, Cb represents the total energy reduction achieved in
bonding at the site across a unit thickness in the y (out-
of-plane) direction in Fig. 1, and has the physical
dimensions force length. The physical units of W are
force/length and, as has been pointed out in Lin et al. [14],
the interaction potential shown in (1) can be approximated
by a harmonic potential
1 2
ub q  Xq  Cb 2
2
within the well itself. Figure 2 shows the comparison
between the full potential and the harmonic approximation,
where the solid line corresponds to (1) for Cb = 10 kT and
W = 20 kT/nm2; the dashed line represents the harmonic
potential (2) for the same parameters. As is evident form
the figure, these two potentials are nearly identical when
the deflection q is deep in the well, and only when q
123
8906 J Mater Sci (2007) 42:89048910

5 small enough so that the energy of deformation arises only

from bending. If the elastic bending stiffness of the
membrane is denoted by Ce then the elastic bending
energy Ue of the membrane is
0
Z
U / kT

K=2
1
Ue Ce h00 x2 dx: 5
b

2 K=2
5
Notice that the physical dimensions of Ce are
force length2. Assuming that the binding potential is
the same for all potential binding sites, the total binding
10 energy is
2 1 0 1 2
q

Fig. 2 Shape of the energy well as described in (1) (solid line), and X
Nb

its comparison with the harmonic approximation (dashed line) Ub ub hi : 6

i1
P odd
Here hi d Nn 1;3;5... an cosnpxi =K is the deflection of
approaches the rim of the well does the quadratic potential
the membrane at binding point xi = i DL/2. In the
begin to deviate significantly from the full potential. The
framework of statistical mechanics, the partition function Z
width of the well d can be estimated from (2) as
of the system is
r
2Cb Z 1 Z 1 Z 1
d : 3 Z


eUe Ub =kT da1 da3 . . . daNodd : 7
X
1 1 1

Of course, one can proceed by using (1) to describe the Following substitution of the expressions for Ue and Ub
effect of adhesion, as was done in Lin et al. [14]. However, into (7), the partition function becomes
in the current study we mainly focus on the conditions
Z 1 Z 1 Z 1
where stable adhesions are formed. In this case, the T

deflection magnitude of the membrane at each binding Z e Nb Cb =kT

ea
Q
a
da1 da3 . . . daNodd
1 1 1
point is expected to be less than the well width d come on
8
and substantially less for a high probability of bonding.
Consequently, use of the harmonic approximation of in which the vector a is defined as a a1 ; a3 ; . . . ; aNodd T .
interaction energy should yield reasonably accurate results. The components of the matrix Q are
Furthermore, using (2) instead of (1) can greatly simplify
the analysis, the benefits of which will become evident in X
Nb
npx 
mpx  n4
i i
the analysis described next. Based on these considerations, Qmn cb cos cos ce 3 dmn ; 9
i1
K K k
(2) is chosen here to describe the interaction energy at each
binding point. where ce p4 =4Ce =dkT; k K=d and cb
Assume that the membrane is constrained against Cb =kT Xd2 =2kT: Obviously Q is a real, symmetric
transverse deflection at x = L/2 (see Fig. 1) but is matrix. Furthermore, because Ue + Ub is a positive definite
otherwise unconstrained except at the binding sites. A function of the random variables, the matrix Q can be
truncated Fourier series approximation of the transverse diagonalized by a proper basis transformation, that is,
deflection h(x) can be written as
Q AT
Q
A: 10
X
Nodd
npx
hx d an cos 4 Here A is a normalized basis transformation matrix and Q
n 1;3;5...
K
is a positive definite diagonal matrix. The transformed
in which Nodd is an odd number; the number of modes random variables spanning configuration space are given in
being considered is then (Nodd + 1)/2. The coefficients an terms of the transformation matrix by a AT a. Then the
represent a set of non-dimensional random variables integral in (8) reduces to a product of one dimensional
describing the response of the membrane to thermal Gaussian integrals. Hence, the partition function can be
excitation. Suppose the transverse deflection amplitude is evaluated exactly as

123
J Mater Sci (2007) 42:89048910 8907

Y
N odd
 1=2 0.16
N b cb p
Z e : 11
n1;3;5... Qnn
0.14

The thermal average or expectation value of any phys-

0.12
ical quantity, say g, then can be calculated by

/
Z 1 Z 1 Z 1 0.1
1
hgi ... ga1 ; a3 ; . . . ; aNodd
Z 1 1 1
0.08
 eUe Ub =kT da1 da3 . . . daNodd : 12

In the context of stability of adhesion, the expectation 0.06

0 5 10 15 20 25 30
values of the magnitude of membrane deflection at each Number of modes included in the computation
binding points are of particular interest. Here, we focus on
circumstances only at the binding point at x = 0, although Fig. 3 Dependence of the value r/d on the number of modes
included in the computation. The solid line represents the computa-
any other particular site may be examined in the same way. tional results obtained by using the harmonic interaction potential (2)
The standard deviation r in the deflection at that point is and the diamond symbols correspond to results obtained by using the
readily calculated as potential given in (1)

v
!2 +
interaction potential defined in (1). It is clear that these two
u*
u X
Nodd potentials yield nearly identical results, as expected. From
r dt an simple analysis, it can be shown that the difference between
n1;3;5...
the predictions obtained by using these two potentials is a
by means of (12), and the normalized result is power series in the dimensionless factor ecb . If cb is rela-
tively large, which is the case for real cell adhesions where
X
Nodd the binding energy per bond is about Cb = 10 kT, then this
r2 Pnn
13 difference becomes very small.
d2 n1;3;5... 2Q nn The numerical results also show that only the first few
deformation modes, corresponding to relatively large
where the matrix P is defined as P AT PA, and P is a
wavelengths, contribute significantly to the value of r/d.
matrix with each component equal to 1. The ratio r/d
This outcome can be understood by realizing that the bend-
compares the standard deviation of fluctuations in the
ing energy associated with any mode, identified by wave-
membrane deflection at the binding point to the width of
number n, is proportional to n4, as indicated by (5), so the
the energy well. For adhesion to be stable, this quantity
magnitude of this mode an necessarily decreases rapidly with
should be less than 1, and perhaps substantially less. If the
increasing n because the energy cost becomes increasingly
criterion for stable adhesion is given in the form
high. This suggests that the discrete bonding assumption can
be relaxed and that the behavior can be captured by means of
r  rcr 14 a model based on continuous bonding. In other words, at the
scale of long wavelengths the spacing D between binding
then the smaller the value of rcr, the higher the confidence
points becomes relatively small so that, effectively, adhesion
one has that the adhesion will remain stable. For example, if
is possible continuously along the membrane.
the distribution of membrane deflection at the binding point
Therefore, we re-examine the question on the basis of an
is assumed to be Gaussian, then choosing rcr/d = 1 gives us
assumption that the adhesion is continuous along the
about 68% confidence level that the adhesion will remain
membrane and that the density of interaction potential for
stable, whereas the confidence level increases to about 95%
adhesion is still given by (2). Now, ub represents the
if one chooses rcr/d = 1/2. The dependence of the ratio r/d
adhesion energy per unit length. In this case Cb has the
on other parameters, such as binding energy Cb or spacing
physical units of force and W has the dimensions force/
between two sites D, for example, can be studied numerically
length2. The total elastic energy is still given by (5) and the
on the basis of (13), as described by Lin et al. [14]. With the
total binding energy now takes the form
parameters chosen as k = 100, Nb = 41, ce = 104 and
cb = 6, Fig. 3 shows the sensitivity of r/d to the number of Z K=2
modes included in the calculation. In the figure, the solid line Ub ub dx: 15
K=2
represents the result of using the harmonic potential (2) and
the diamond symbols indicate results obtained by using the The partition function Z, defined in (7), becomes

123
8908 J Mater Sci (2007) 42:89048910

 1=2

Normalized critical binding energy density

Y
N 1
p odd
cb k ce n4
Z pe c b k 3 : 16 0.9
n1;3;5...
2 k
0.8
where cb = Cbd/kT. The non-dimensional parameters k
0.7
and ce have the same forms as previously. Similarly, the
standard deviation r of membrane fluctuations at point 0.6
x = 0 is evaluated from (12) to become 0.5

0.4
Nodd 
X 1
r2 2ce n4
cb k 3 : 17 0.3
d2 n1;3;5... k
0.2
0 5 10 15 20 25 30
Typically, the size of a cluster of molecular bonds that Number of binding sites
makes up a focal adhesion is in the micrometer range
whereas the width of the energy well is in the nanometer Fig. 4 Dependence of the critical adhesion energy density from the
discrete bonding model, normalized by the corresponding value
range, at most. Consequently, the value of k is at least on implied by the continuous bonding model, on the number of binding
the order of 100 and usually larger. Taking the fact that k is points Nb along the membrane
large and letting Nodd ! 1, the summation of the infinite
series appearing in (17) is convergent and the result is

  this reason, the model is identified as being two-dimen-

r2 p 2 1=4 sional. The potential binding sites are indicated by the
: 18
d2 8 ce c3b small circular symbols on the membrane shown in Fig. 5.
As is evident from the diagram, these sites are arranged in a
This is an extremely simple result because the length of the square pattern with spacing D << L in both the x and y
membrane k drops out, and r/d is determined solely by the directions of a rectangular coordinate frame in the nominal
two dimensionless parameters ce and cb, the bending plane of the membrane.
modulus of the membrane and the adhesion energy density. Because L >> D, we expect the continuum bonding
To ascertain the validity of (18), a comparison with results model to provide an accurate prediction on the basis of the
of the discrete model was considered. On the basis of analysis presented in the preceding section. As above,
k = 100, ce = 104 as the values of key system parameters assume the binding energy density ub to be given by (2).
and rcr/d = 1 as the criterion for stable adhesion, the
critical adhesion energy density, computed from the dis-
crete model (13) and normalized by the prediction from
continuous model (18), is plotted in Fig. 4 against the
number of binding points Nb within the membrane. Clearly,
the discrete model result approaches that of the continuous
model asymptotically as Nb increases. Therefore, the con-
tinuous adhesion model appears to be applicable to real
biological systems where hundreds of binding sites can be y
distributed in close proximity to each other along the x
membrane.

Generalization to a two-dimensional system

In this section, the foregoing analysis is extended to the

more realistic two dimensional configurations. In this case,
consider a nominally flat, square membrane with size
L L which is immersed in a thermal bath and which is
positioned near a substrate with potential binding sites. The Fig. 5 Sketch of a portion of a large membrane. The circle symbols
represent the binding sites on the substrate, and these sites are
transverse deflection of the membrane h (x,y) in this arranged in a regular square array with spacing D. The extent of the
instance depends on two independent variables x and y; for membrane in the plane is L L, where L >> D

123
J Mater Sci (2007) 42:89048910 8909

Notice that, in this case, Cb and W have the dimensions NZ odd 1

odd 1 NZ 1
force/length and force/length3, respectively. If the depth of r2 1 k2 cb 4
 k x2 y2 2 dx dy:
the potential well representing bond interaction energy at d2 4 ce 2ce
0 0
any discrete site is Gb, then
25
Cb From the form of the integrand, it is clear that Nodd must be
Cb 19
D2 substantially larger than k (cb/2ce)1/4 for the approximation
Nodd ! 1 to be valid. In terms of the model, the values of
in the present instance.
ce and cb are of the same order so the restriction implies
We proceed in a manner similar to that followed for the
that the number of modes taken into account must be large
one-dimensional case. The membrane is constrained
enough so that the shortest wavelength included will be
against transverse deflection at its edges and the deflection
less than D, a reasonable expectation. For Nodd ! 1, the
h(x,y) is represented by the truncated Fourier series
double integral is readily evaluated by elementary
X
Nodd X
Nodd methods, with the result that
mpx npy
hx; y d amn cos cos :
m 1;3;5... n 1;3;5...
K K r
r2 p2 2
20 2
: 26
d 32 ce cb
As before, the parameters amn represent a set of random
In order to draw quantitative inferences from (26) con-
variables spanning the available configuration space of the
cerning the critical value of bond site spacing below which
system. The total bending energy is
adhesion is stable, assume that the bending modulus of the
membrane is Ce = 20 kT, a typical value for bilayer lipid
Z Z K=2  2 2
1 K=2
oh o2 h membranes bilayer lipid membranes [8]. Also, the chemi-
Ue Ce dx dy 21
2 K=2 K=2 ox2 oy2 cal potential of a single bond is assumed to have the value
Gb = 10 kT, a typical chemical bonding energy for an
where the bending modulus Ce has the physical dimensions integrin-RGD ligand pair [15]. For these parameter values,
force length. The total adhesion energy Ub is the dependence of the critical spacing Dcr/d on r /d is
illustrated in Fig. 6. It is evident from the figure that the
Z K=2 Z K=2
value of a Dcr implied by the model depends quite strongly
Ub ub dx dy: 22
K=2 K=2 on the value of r/d that is taken to represent adhesion. As
noted above, for a Gaussian probability distribution, a
The standard deviation r of membrane fluctuations at variance of d2 implies a 68% probability that the membrane
point x = 0, y = 0 can then be evaluated by means of is within the adhesion well, while a variance of 1=4d2
(12) as implies a 95% likelihood of the bond having been
completed.
" #1
r2 X
Nodd X
Nodd
c b k2 ce m2 n2 2
23
d2 m1;3;5... n 1;3;5...
2 k2
160

where
/ - critical spacing

120

p4 Ce K Cb d2 Xd4
ce ; k ; cb 24 80
4 kT d kT 2kT
are dimensionless variables similar to those defined for 40
the one-dimensional case. For sufficiently large values of
Nodd, the value of the double sum in (23) can be 0
0 0.2 0.4 0.6 0.8 1
approximated accurately by a double integral. The double / - standard deviation
series is convergent so there is no disadvantage in
allowing the total number of modes Nodd to become Fig. 6 Dependence of the normalized spacing between membrane
binding sites on the normalized standard deviation in membrane
large. Accounting for the fact that the increments in both fluctuation at a particular binding site within the adhesion energy
m and n in the series is 2, the integral approximation to well, illustrated for membrane bending stiffness Ce = 20 kT and
the double series is chemical potential Gb = 10 kT

123
8910
To compare the model results to experimental results, it
is necessary to choose a value for d, the half width of the
energy well representing interaction between bonding ele-
ments. Suppose that we adopt the value of d  1 nm, a
large value relative to atomic dimensions, in order to reflect
the compliance of the adhesion molecules. From a statis-
tical point of view, the value of r/d should be less than 1,
and perhaps substantially less, in order for the behavior to
correspond to a stable adhesion. If we assume that the
critical values of spacing Dcr lies within the range for
which the value of r/d varies from 1/2 to 1 then, on the
basis of (26), the implied range of critical spacing is from
40 nm to 160 nm. This range includes spacings compara-
ble to the critical spacing range 5873 nm observed by
Arnold et al. [7]. It is noted that the spacing in the exper-
iments referred to a regular hexagonal array of binding
sites. From simple geometrical considerations and the
assumption that bond energy per unit area is the same for
either square or hexagonal arrays of binding sites, the
critical spacing for a hexagonal array falls within the range
43172 nm.
Finally, we note that the assumption that the outer
boundaries of the square membrane are constrained against
transverse deflection is somewhat arbitrary. Instead, if it is
assumed that the edges are subjected to periodic boundary
conditions, rather than conditions of vanishing deflection,
then the final result corresponding to (26) is modified
slightly, taking the form
r References
r2 p2 2
: 27
d2 8 ce cb
In this case, the critical spacing is implied to be in the
range 2080 nm for a square array of binding sites, and in
the range 2186 nm for a hexagonal array. Again, these
estimates correspond to the full range 1=2\r=d\1 for the
standard deviation in fluctuation of in transverse deflection
for the representative bond site of the membrane.
Conclusions
In this article the stability of adhesive contact between a
compliant cell membrane and a substrate has been con-
sidered. The binding sites have been assumed to be peri-
odically distributed over the substrate surface. The
membrane undergoes continual thermal fluctuations due to
its immersion in a large heat bath, and these fluctuations
are necessarily suppressed to some degree in the process of
123
J Mater Sci (2007) 42:89048910
adhesion. The competition between thermal fluctuations
and formation of adhesive bonds has been examined within
the framework of classical statistical mechanics. Mathe-
matical models for adhesion have been examined for two
cases, one in which adhesion sites are widely spaced and
therefore discrete and a second in which adhesion site
spacing is small enough so that adhesion can be viewed as
continuous over the membrane surface. The equivalence
between these two cases has been demonstrated under
circumstances for which the size of the membrane being
considered is much larger than the spacing between bind-
ing sites.
It has been found that background thermal undulations
set an upper limit on bond site spacing for stable
adhesions to form, in a statistical sense. A contact region
with adhesions sites space more widely than this upper
bound will likely not adhere; if bonds are formed, their
state is unstable and they will be overcome by thermal
fluctuations. An estimate of the critical spacing based on
reasonable values of system parameters is in good
agreement with experimental observations that have been
reported on the actual spacing for a number of cell
types.
Acknowledgments The research support of the National Science
Foundation through the Brown University MRSEC Program (DMR-
0079964) is gratefully acknowledged.
1. Harris AK, Wild P, Stopak D (1980) Science 208:177
2. Alberts B, Bray D, Lewis J, Raff M, Roberts K, Watson JD
(1989) Molecular biology of the cell, 2nd edn. Garland Publish-
ing, New York and London
3. Boulbitch A, Guttenberg Z, Sackmann E (2001) Biophys J
81:2743
4. Freund LB, Lin Y (2004) J Mech Phys Solids 52:2455
5. Shenoy VB, Freund LB (2005) Proc Natl Acad Sci USA
102:3213
6. Hersel U, Dahmen C, Kessler H (2003) Biomaterials 24:4385
7. Arnold M, Cavalcanti-Adam EA, Glass R, Blummel J, Eck W,
Kantlehner M, Kessler H, Spatz JP (2004) ChemPhysChem 5:383
8. Boal D (2002) Mechanics of the cell. Cambridge University
Press, Cambridge
9. de Gennes PG, Taupin C (1982) J Phys Chem 86:2294
10. Gov N, Zilman AG, Safran S (2003) Phys Rev Lett 90:228101
11. Lin LC-L, Brown FL (2004) Biophys J 86:764
12. Gov N, Safran SA (2004) Phys Rev E 69:011101
13. Merath R-J, Seifert U (2006) Phys Rev E 73:010401
14. Lin Y, Inamdar M, Freund LB (2007) J Mech Phys Solids 55: in
press
15. Goennenwein S, Tanaka M, Hu B, Moroder L, Sackmann E
(2003) Biophys J 85:646
J Mater Sci (2007) 42:89118918
DOI 10.1007/s10853-007-1698-3
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Fracture and repair of bone: a multiscale problem
David Taylor
Received: 7 December 2006 / Accepted: 15 March 2007 / Published online: 7 August 2007
Springer Science+Business Media, LLC 2007
Abstract This article reviews current work on the
strength and toughness of bone, its mechanisms of fracture
and its ability to repair and adapt its structure. These
properties are affected at all size levels, from the nano-
structure of collagen molecules and mineral crystals,
through the microstructure of osteons and trabeculae, up to
the macroscopic shape and density variations that occur at
the level of a whole bone.
Introduction
The strength and fracture of bone has interested scientists
for a very long time. Galileo Galilei, in his pioneering work
on mechanics, Dialogues Concerning the Two New Sci-
ences [1], discusses the shapes of bones in relation to their
strength as an example of a problem in scaling. He worked
in the University of Padova, and it can be no coincidence
that it was there, around the same time, that the first
anatomy theatre was built, allowing students to observe the
dissection of cadavers. Interdisciplinary research in this
field has a long history.
In this article I will discuss the current state of our
understanding of fracture in bone, and point out lines of
developing research. It is impossible, or at least unwise, to
discuss the fracture of this material without also discussing
its repair, because our bones are living structures, capable
D. Taylor (&)
Department of Mechanical Engineering, Trinity Centre
for Bioengineering, Trinity College Dublin, College Green,
Dublin 2, Ireland
e-mail: dtaylor@tcd.ie
of maintaining their integrity by continual repair of damage
and by adapting to changes in their stress environment.
These two aspects of boneits resistance to fracture and
its functional adaptationhave evolved hand-in-hand,
giving us a material, which is supremely well adapted to its
role.
Two themes will run through this article. The first is that
the subject is a hierarchical one which must be viewed on
different scales: it is convenient to divide these into three,
which I will call the macro, micro, and nano scales, though
inevitably these divisions are somewhat arbitrary and
overlapping. The second theme is that, despite some unique
characteristics, bone is a structural material, which can be
investigated in the same way, using the same tools and
ideas, which we use to investigate other structural materi-
als. Thus the expertise of materials scientists, who under-
stand the relationships between structure and mechanical
properties, is vital to our research activities in this field.
The macro scale
In the world of man-made materials it is normal to think of
the material and the structure (i.e. the macroscopic
structure of the object in question) as two different things,
the first being broadly the province of the materials sci-
entist whilst the second is the concern of the engineer. In
natural materials, however, there can be no such easy
distinction: in structures such as a human bone or tendon or
the stem of a plant, the properties of the material vary from
place to place within the structure in a smooth, continuous
manner which we are only beginning to imitate in, for
instance, functionally-graded materials. For example, in the
head of the femur (Fig. 1) the material varies from the solid
outer shell of cortical bone to the inner network of
123
8912
trabecular bone, in which the size and orientation of indi-
vidual trabeculae is driven by local stress. It is difficult to
know where the material stops and the structure begins, so
we must start our discussion by considering an entire bone,
for example a typical long bone such as the femur. At this
scale three factors exert an influence on the mechanical
performance of the structure, these are: shape, size and
density.
Shape
It is often said that the shapes of our bones have been
optimised for their purpose through long ages of evolution.
In fact optimised is not quite the right word, because
evolution does not act to optimise a structure but rather to
make it just good enough to give the organsm a competitive
advantage, a better chance of survival. Alexander [3] and
Martin [4] considered the various factors involved: cru-
cially, the aim here is not to avoid fracture at all costs, in
fact a certain probability of fracture (which, for the indi-
vidual, may mean death) is essential in order to reduce
weight, ensuring a gracile support structure, one which is
light and efficient. Nevertheless, the optimisation problem
is an interesting one for the student, and gives us some
useful insights. Consider the problem of designing a
bonesimplified to a regular cylinder of outer radius r and
thickness t: what will be the optimum value of the shape
factor r/t, assuming that the aim is to minimise weight for a
given strength? Long bones are subjected to a mixture of
loading types: tension, compression, bending and torsion. It
is a simple matter to show that, for the cases of tension,
compression and torsion, the factor r/t has no effectit
cannot be optimised. On the other hand, under applied
bending the optimum value will be infinity: clearly no
sensible solution arises. Pauwels [5] pointed out that there
is an extra weight term, because the bone tube is full of
marrow, and this leads to a finite optimum value for r/t.
Currey and Alexander [6] developed this idea still further,
considering different modes of failure (yielding, impact,
etc); their results suggested an optimum value of approxi-
mately 2 which, despite considerable variation, is a typical
value for long bones.
Modern approaches to this problem tend to use com-
puter simulationsusually finite element (FE) analysisto
study the development of the complex shapes of real bones.
These will be discussed in the next section since they are
also capable of simulating density variations.
Density
Bone can vary its density by changing its porosity. The
solid material (cortical bone) that makes up the outer shells
of our bones has typically 5% porosity. This can increase
123
J Mater Sci (2007) 42:89118918
under conditions of disuse or disease, forming osteoporotic
bone; further increases in porosity give rise to a spongy,
open structure known as cancellous or trabecular bone,
which is found near joints where it provides a low-weight
solution to the problem of transferring compressive stresses
from the joints down into the bone tube, as we saw above
(Fig. 1).
The prediction of bone shape (in practice largely con-
fined to predicting the local thickness of the cortex) and the
distribution of density (or, in some cases the local structure
of the trabecular bone), has been extensively studied using
FE simulations. These features are not only determined by
evolutionary pressures, but can also change within the
lifetime of an individual, as a result of changes in the stress
environment as caused by, for example, the adoption of a
more active lifestyle or the implantation of an artificial
joint [7]. The most successful simulations are those, which
take account of the development of damage over time, its
continual repair and the functional adaptation, which will
occur if the rates of damage and repair are not equal [8].
This work will not be discussed in detail here, as it takes
us out of the realm of materials science, being largely
concerned with continuum-mechanics techniques such as
damage mechanics, using control variables such as strain-
energy density. Some workers have achieved considerable
predictive success (see for example [9]), though their work
is hampered by the limited amount of experimental data
available and the large degree of scatter inevitable in such
data.
Fig. 1 Cross section through the head of a femur, showing the outer
shell of compact cortical bone and the open network of trabecular
bone [2]
J Mater Sci (2007) 42:89118918
Size
The stress to cause failure in a sample of bone will decrease
with increasing size of the sample. This is a well-known
phenomenon, apparent when failure occurs by a weakest-
link mechanism, as in the strength of brittle materials such
as ceramics and the fatigue strength of metals. It is a major
issue for bone because of the enormous changes in scale
that occur: the bones of a mouse and those of an elephant
are essentially the same shape, but vary in volume by many
orders of magnitude. Such problems are traditionally
addressed using a probabilistic technique: we showed that a
Weibull analysis is capable of predicting scaling factors in
the fatigue strength of cortical bone, and that the approach
can be extended to discuss size effects in different animals
and, with the addition of terms to describe repair and
adaptation, can be used to predict the probability of stress
fractures (i.e. fatigue failures) in athletes, military person-
nel and other vulnerable groups [10; 11].
The micro scale
Fracture
The work described above was essentially in the realm of
continuum mechanics, though the introduction of proba-
bilistic terms hints at the existence of local material vari-
ations and therefore of a microstructure. Various pieces of
evidence give us clues as to the relevant size scale on
which fracture mechanisms operate. For example, cracks
and sharp notches have almost no effect when they are as
small as 0.37 mm [12] and even circular holes as large as
10 mm diameter do not reduce the strength of the specimen
by as much as the elastic stress-concentration factor [13].
Systematic studies by Nalla et al. [14] and Malik et al. [15]
have shown that the measured fracture toughness, Kc,
increases with crack length for cracks up to several milli-
metres long. This behaviour, described by the so-called
R-curve (resistance curve, see Fig. 2), is expected to persist
until the crack is at least an order of magnitude larger than
those microstructural features which affect toughness. A
final clue comes from the theoretical analysis, commonly
used in process zone and critical distance approaches [17;
18], which derives a critical length scale, L, as a function of
the fracture toughness and static strength (ro) of the
material, as:
 
1 Kc 2
L 1
p ro
Strength and toughness vary considerably in bone, but if we
use typical values (Kc = 4 MPa(m)1/2, ro = 160 MPa) we
8913
obtain L = 0.2 mm. All of these pieces of evidence suggest
that the fracture of bone, and therefore its toughness, is con-
trolled by mechanisms operating at the 0.11 mm scale.
Several different mechanisms operate, and there is
considerable debate as to their relative importance. A
useful distinction here is between so-called intrinsinc
toughening mechanisms, which operate in front of the
crack tip, and extrinsic toughening mechanisms which
operate behind the tip, in the crack wake; the latter are
responsible for the increase in measured toughness with
crack length as shown in Fig. 2. Nalla et al. studied these
using a number of techniques, including three-dimensional
imaging by tomography [19; 16]. They concluded that a
major contribution to toughening comes from the bridging
of the crack faces by unbroken ligaments of material, an
extrinsic mechanism illustrated in Fig. 2. Much smaller
contributions were predicted to come from other mecha-
nisms such as crack deflection, bridging by collagen and
the consumption of energy in the zone of microdamage
ahead of the crack. Vashishth et al. [20] on the other hand,
predicted a major role for microdamage, which takes the
form of a cloud of small cracks ahead of the main crack,
whilst Yeni and Fyhrie argued that collagen-fibre bridging
across microcracks has a significant effect on bones
strength [21]. Another important factor is plasticity. Bone
exhibits non-linear stress/strain behaviour: taking some
typical figures (Youngs modulus = 15 GPa; tensile
strength = 160 MPa; strain to failure = 0.02) we see that
about half of the strain to failure is due to non-linear
deformation, some caused by microdamage, some by vis-
coelasticity and some by plastic deformation, which occurs
in the collagen phase (see below concerning bones
chemical make-up). Testing at high strain rates, where
collagen has less time to deform, reduces the strain to
failure [22] and also the toughness [23]. It seems likely that
plasticity is responsible for a large proportion of the
intrinsic toughness of bone, though further work is cer-
tainly needed to clarify the relative contributions of these
different mechanisms.
Individual microstructural features begin to make
themselves apparent as the crack length gets smaller. The
principal features of interest in cortical bone are osteons;
these are long, cylindrical structures, typically 200 lm in
diameter and many millimetres long, which run along the
length of the bone (Fig. 3). In the centre of each osteon is a
blood vessel, supplying nutrients to living cells, which
reside in small cavities nearby, connected by thin cellular
processes (Fig. 4).
Discontinuous crack growth is observed under constant
load for sub-millimetre cracks (see Fig. 5) [25], in contrast to
the smoother growth behaviour of longer cracks [26]. Cracks
are temporarily arrested by features such as osteons and
Volkmans canals. For crack lengths less than about 100 lm
123
8914 J Mater Sci (2007) 42:89118918

Fig. 2 Cracks in bone show increasing toughness with crack length, increasing toughness is correlated to the presence of unbroken
as the graph here demonstrates for a crack growing from a notch. The ligaments which span the crack faces [16]
photographs, taken at early and late stages in the test, show how the

Fig. 3 A transverse section through a human femur, showing (on the left) two osteons with a crack running between them. At higher
magnification (on the right) the crack is seen to be white, indicating that it has been filled in by subsequent mineralization [24]

crack growth behaviour becomes completely dominated by stitial bone between osteons (Fig. 3). Much work has gone
these features and this is especially apparent under cyclic into the detection and measurement of these cracks, using
loading [27; 28]. Small fatigue cracks initiate in the inter- penetrant dyes [29]. Unlike metallic materials (but very

123
J Mater Sci (2007) 42:89118918
Fig. 4 Two bone cells (osteocytes): note the long thin processes
through which the cells communicate; these run in narrow channels
(canaliculi) within the bone matrix
much like fibre composites), bone develops cracks not only
on its surface but internally, so the majority of the damage is
not immediately visible. Dyes of different colours can be
introduced to stain cracks at different stages during a test,
providing a picture of the development of damage over time
[30]. In shape these cracks are typically elliptical planes,
about 400 lm by 100 lm, oriented with their major axes
approximately (but not exactly) parallel to the bones lon-
gitudinal axis: Fig. 6 shows an image of a typical crack,
obtained using laser scanning confocal microscopy [31].
Again this is very different from the behaviour of fatigue
cracks in metals, which orient themselves perpendicular to
the principal stress axis, but not unlike some forms of
delamination damage in composite laminates.
These cracks grow relatively quickly until they reach
barriers in the microstructure such as the cement lines,
which surround osteons (Fig. 3). At this point the great
majority of cracks cease growing altogether. A few con-
tinue, usually growing around their first osteon but break-
ing through osteons as they become longer and more
energetic. Thus the osteon structure (or, in some other
animals, a different microstructure on a similar scale,
Fig. 5 Crack growth rate as a
function of length for a small
crack growing in cortical bone.
Growth is discontinuous:
minima in the graph (numbered)
correspond to periods of
temporary arrest when the crack
encountered features in the
microstructure, in this case
Volkmans canals (enhanced on
the photograph)
8915
known as plexiform bone) is crucial in determining the
materials resistance to fatigue failure. Recently a lot of
work has been done in this field, measuring the length
and number-density of microcracks (and also regions of
so-called diffuse damage which will be discussed below).
Correlations have been found between the amount of
damage and the applied load, the age of the subject, the
density of bone cells, the existence of osteoporosis and
other factors [3235]. Further theoretical work is needed to
make sense of all this data.
Repair
The repair process in cortical bone is ideally suited to deal
with cracks of this size. First described by Frost [36], repair
uses two different types of cells, working together: osteo-
clastscells, which dissolve boneare followed by
osteoblasts, which make new bone. The result is a so-called
Basic Multicellular Unit (BMU) taking the form of a cavity
approximately 200 lm across which moves along the
length of the bone, removing old material containing
microcracks and replacing it with new material, forming an
osteon. It has taken a lot of careful research to demonstrate
that BMUs do not occur randomly but are targeted towards
microcracks [3739] (see Fig. 7). The actual mechanisms
by which this happens are still being investigated and this
is a very exciting area of current research, involving
elements of materials science, fracture mechanics, bio-
chemistry and genetics. For example, Klein-Nulend and
co-workers [41; 42] have suggested a mechanism to control
the movement of BMUs. A crucial feature of this model is
fluid flow, which is relatively stagnant in the region just
ahead of the BMU cavity; osteoclast activity here is
envisaged to be stimulated by altered levels of nitric oxide
emitted by cells, causing the BMU to move along paths of
maximum principal stress. Hazenberg et al. [43] have
proposed a mechanism, which considers how cracks can be
detected. The main step in this model is the fracture of
cellular processes, which pass across crack faces; these are
envisaged to be cut in a scissor-like action due to local
123
8916
Fig. 6 An image of a typical crack inside a specimen of bone,
obtained using laser scanning confocal microscopy [31]
shearing motions. Once broken, these cells release a sub-
stance, which stimulates the formation of osteoclasts, cre-
ating a BMU. Other work has suggested mechanisms by
which cells might be affected by the presence of a nearby
crack and signal to other cells to initiate the repair process
[44; 45].
Some theoretical models have attempted to reproduce
these damage and repair processes in computer simula-
tions. Martin [46; 47] has incorporated the behaviour of
BMUs, creating an equilibrium crack density in which
cracks are continually being created and removed. This
equilibrium can be disturbed in various ways, leading to
adaptation or stress fractures. One interesting feature is the
fact that the BMUs themselves tend to increase stress
Fig. 7 The resorption cavity of a BMU (white object indicated by
large arrowheads) moves towards a microcrack (indicated by the
smaller arrows). The surrounding structure shows bone cells and their
processes [40]
123
J Mater Sci (2007) 42:89118918
because they increase the level of porosity, creating a
potential instability. We have developed a model in which
the growth behaviour of each individual microcrack and
BMU is described, using stochastic variables [48]. Such
models have the advantage that they can predict many
parameters, which can be directly measured, such as the
number density of cracks and BMUs, and so can be tuned
in the light of experimental data.
Trabecular Bone
Trabecular bone can be thought of as a network structure of
rods and plates of thickness typically in the 10100 lm
range. Its failure characteristics are very different from
those of solid cortical bone. As might be expected for a
foam-like material, failures of individual trabeculae (by
cracking, tensile snapping or compressive buckling) con-
tribute to a gradual deterioration in the materials integrity.
Partially broken trabeculae can be repaired by BMUs,
though in this case they operate by scouring the material
surface, rather than by tunnelling. Completely fractured
trabeculae can be repaired in much the same way as a
broken bone, by the formation of a microscopic callus [24],
as shown in Fig. 8. Sometimes this cannot happen and then
the surrounding network structure must adapt to the loss.
There is evidence for a gradual deterioration in the struc-
ture of this type of bone over long periods of time, due to
random fracture events, which may contribute to the onset
of osteoporosis.
The nano scale
This final section considers structure on the scale reaching
from 1 lm down to molecular dimensions. The processes
of damage and repair on this scale are less well understood
than at the larger size levels, though some recent work has
begun to shed light on the effects of structure and com-
position on material strength and toughness.
Bone is a composite material, made up of two fibrous
components, closely interwoven. These are collagen, a
biological polymer with a triple helix, and a mineral phase
based on calcium, which is traditionally called hydroxy-
apatite (though recent work suggests may be better
described as a carbonated apatite [49]). In the formation of
bone, long fibres of collagen (typically 0.2 lm in diameter)
are laid down first; hydroxyapatite is subsequently pre-
cipitated amongst the fibres, forming elongated crystals,
each typically 5 nm diameter and 40 nm long, joined in a
continuous network. Layers of this material are built up,
very much like the process of making fibre-composite
laminate sheets, though the laminar thickness in this case is
of the order of a few microns. Mechanical properties can be
J Mater Sci (2007) 42:89118918
Fig. 8 An individual trabecula containing a callus, which has formed
to repair a fracture [24]. Note that subsequent cracks have initiated
(arrowed) indicating an imminent re-fracture
adjusted by varying fibre orientations; for example, osteons
are built with laminae encircling the blood vessel as tubes:
varying the proportions of fibres in different orientations
creates osteons with different mechanical properties [50].
In addition to the microcracks mentioned above, damage
can occur on a smaller scale. Areas of material stained by the
penetrant dye (Fig. 9) are seen, at higher magnification, to
consist of many small cracks, each typically 1 lm in length;
this has been termed diffuse damage [51]. Other kinds of
damage presumably exist at smaller size scalesone can
imagine, for example, fracture of the hydroxyapatite crystal
network and the breakdown of the interface between it and
the collagen fibres, but to my knowledge these damage
modes have yet to be observed, partly due to the difficulty of
preparing specimens for high-magnification work without
creating damage in the process.
Some recent pieces of work have investigated the effect
that changes at the molecular level have on bones strength
and toughness: a good review on toughness effects at the
nano scale has been provided by Nyman et al. [49]. As yet
no clear, consistent picture has emerged but there are some
themes developing which merit further attention. Firstly,
we cannot forget a third constituent of bone, which is
Fig. 9 Cross section of a rat ulna after extensive fatigue loading [45]:
microcracks can be seen (labelled lCr) along with a region which has
become stained due to the presence of diffuse damage (labelled Dfdx)
8917
water; it occupies about 25% by volume of compact bone.
It is well known that if a bone sample is allowed to dry out
before testing it becomes much more brittle as well as
containing more damage, created by shrinkage. The re-
moval of even small amounts of water has been correlated
to reductions in toughness and plastic strain [52]. Water is
found both in the collagen phase, where it stabilises col-
lagen fibres via hydrogen bonding, and in the mineral
phase, where it alters the crystallographic structure.
A second factor is the quality of the collagen phase,
whose deterioration has been correlated to age-related
changes in bone toughness and strength [53; 54]. An
important parameter here is the degree of crosslinking.
Like many polymers, collagen undergoes crosslinking,
which tends to make its structure stronger, but also perhaps
more brittle. The degree and type of crosslinking changes
with the age of the tissue [55; 56] and is affected by several
other factors [57]. Gamma radiation, which denatures
collagen, causes reductions in bone strength and toughness
[58]. Changes in the mineral phase also affect mechanical
properties [59; 60], and cracks tend to accumulate prefer-
entially in regions of high mineral content [61].
Almost nothing is known about any repair processes that
may be operating at the sub-micron level, though Boyd
[24] has pointed out that macroscopic cracks can some-
times be filled in with mineralised material (Fig. 3)
suggesting an alternative to repair by BMUs.
Concluding remarks
The fracture and repair of bone is a subject for which we
have to take a hierarchical perspective, a fact, which has
not been lost on some researchers [62; 63]. Changes at
the sub-micron level, which affect the chemistry and
structure of collagen fibres and mineral crystals, deter-
mine the nature of the basic material, the material within
which cracks will grow. But significant elements of the
control of toughness occur at the microstructural scale,
where features such as osteons in cortical bone, and the
geometry and arrangement of trabeculae in cancellous
bone, strongly affect the ease of crack growth and local
fracture events. At the macroscopic scale, bones ability
to vary its material propertiesproperties such as density
and fibre orientationallow the best use to be made of
the material to create a strong, light, efficient skeletal
structure.
References
1. Galileo G (1638) Dialogues concerning the two new sciences.
Elzevir, Leyden, The Netherlands
123
8918
2. Gray H (1959) Anatomy of the human body. Lee and Febiger,
Philadelphia, USA
3. Alexander RM (1984) J Theo Biol 109:621
4. Martin RB (2003) J Theo Biol 220:271
5. Pauwels F(1980) Biomechanics of the Locomotor Apparatus.
Berlin, Springer Verlag
6. Currey JD, Alexander RM (1985) J Zool Lon A 206:453
7. Huiskes R (1988) In: Fitzgerald R (ed) Non-cemented total hip
arthroplasty, Raven Press, New York, pp 283302
8. Prendergast PJ, Taylor D (1994) J Biomech 27:1067
9. Huiskes R, Rulmerman R, Van Lenthe GH, Janssen JD (2000)
Nature 405:704
10. Taylor D (2000) J Theo Biol 206:299
11. Taylor D, Casolari E, Bignardi C (2004) J Orthopaed Res 22:487
12. Lakes RS, Nakamura S, Behiri JC, Bonfield W (1990) J Biomech
23:967
13. Seltzer KL, Stover SM, Taylor KT, Willits NH (1996) Veter Surg
25:371
14. Nalla RK, Kruzic JJ, Kinney JH, Ritchie RO (2004) Bone
35:1240
15. Malik CL, Stover SM, Martin RB, Gibeling JC (2003) J Biomech
36:191
16. Nalla RK, Kruzic JJ, Kinney JH, Ritchie RO (2005) Biomaterials
26:217
17. Bazant ZP (2004) Mater Struct 37:1
18. Taylor D (2004) Eng Frac Mech 71:2407
19. Nalla RK, Kinney JH, Ritchie RO (2003) Nat Mater 2:164
20. Vashishth D, Tanner KE, Bonfield W (2003) J Biomech 36:121
21. Yeni YN, Fyhrie DP (2003) J Biomech 36:1343
22. Natali AN, Meroi EA (1989) J Biomed Eng 11:266
23. Tanabe Y (1999) In: Takahashi HE (ed) Mechanical Loading of
Bones and Joints. Springer-Verlag, Tokyo, Japan
24. Boyde A (2003) J Anatomy 203:173
25. Hazenberg JG, Taylor D, Lee TC (2006) Biomaterials 27:2114
26. Nalla RK, Kruzic JJ, Kinney JH, Ritchie RO (2005) Biomaterials
26:2183
27. OBrien FJ, Taylor D, Lee TC (2005) J Orthopaed Res 23:475
28. Akkus O, Rimnac CM (2001) J Biomech 34:757
29. Frost HM (1960) Henry Ford Hosp Med Bull 8:25
30. OBrien F, Taylor D, Lee TC (2002) J Biomech 35:523
31. OBrien F, Taylor D, Lee TC (2000) J Anatomy 197:413
32. Qiu S, Rao DS, Fyhrie DP, Palnitkar S, Parfitt AM (2005) Bone
37:10
33. Frank JD, Ryan M, Kalscheur VL, Ruaux-Mason CP, Hozak RR,
Muir P (2002) Bone 30:201
34. Fazzalari NL, Forwood MR, Smith K, Manthey BA, Herreen P
(1998) Bone 22:381
123
J Mater Sci (2007) 42:89118918
35. Muir P, McCarthy J, Radtke CL, Markel MD, Santschi EM,
Scollay MC, Kalscheur VL (2006) Bone 38:342
36. Frost HM (1969) Calci Tissue Res 3:211
37. Burr DB (2002) Bone 30:2
38. Burr DB, Martin B (1993) J Biomech 26:613
39. Lee TC, Staines A, Taylor D (2002) J Anatomy 201:437
40. Colopy SA, Benz-Dean J, Barrett JG, Sample SJ, Lu Y, Danova
NA, Kalscheur VL, Vanderby J, Markel MD, Muir P (2004) Bone
35:881
41. Burger EH, Klein-Nulend J, Smit TH (2003) J Biomech 36:1453
42. Klein-Nulend J, Bacabac RG, Mullender MG (2005) Pathologie
Biologie 53:576580
43. Hazenberg JG, Freeley M, Foran E, Lee TC, Taylor D (2006)
J Biomech (In Press)
44. Verborgt O, Gibson GJ, Schaffler MB (2000) J Bone Miner Res
15:60
45. Bentolila V, Boyce TM, Fyhrie DP, Drumb R, Skerry TM,
Schaffler MB (1998) Bone 23:275
46. Martin B (1995) J Orthopaed Res 13:309
47. Hazelwood SJ, Bruce Martin R, Rashid MM, Rodrigo JJ (2001)
J Biomech 34:299
48. Taylor D, Lee TC (2003) J Anatomy 203:203
49. Nyman JS, Reyes M, Wang X (2005) Micron 36:566
50. Ascenzi A, Baschieri P, Benvenuti A (1990) J Biomech 23:763
51. Vashishth D, Koontz J, Qiu SJ, Lundin-Cannon D, Yeni YN,
Schaffler MB, Fyhrie DP (2000) Bone 26:147
52. Nyman JS, Roy A, Shen X, Acuna RL, Tyler JH, Wang X (2006)
J Biomech 39:931
53. Wang X, Xiaoe LI, Shen X, Agrawal CM (2003) Ann Biomed
Eng 31:1365
54. Zioupos P, Currey JD, Hamer AJ (1999) J Biomed Mater Res
45:108
55. Eyre DR, Dickson IR, Van Ness K (1988) Biochem J 252:495
56. Nalla RK, Kruzic JJ, Kinney JH, Balooch M, Ager III JW,
Ritchie RO (2006) Mater Sci Eng C 26:1251
57. Knott L, Whitehead CC, Fleming RH, Bailey AJ (1995) Biochem
J 310:1045
58. Akkus O, Rimnac CM (2001) J Orthopaed Res 19:927
59. Akkus O, Adar F, Schaffler MB (2004) Bone 34:443
60. Currey JD (1984) Biol Sci 304:509
61. Wasserman N, Yerramshetty J, Akkus O (2005) Eur J Morphol
42:43
62. Akkus O, Yeni YN, Wasserman N (2004) Crit Rev Biomed Eng
32:379
63. Vashishth D (2005) Crit Rev Euk Gene Expr 15:343357
J Mater Sci (2007) 42:89198933
DOI 10.1007/s10853-007-1693-8
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Insights into whole bone and tooth function using optical
metrology
Ron Shahar Steve Weiner
Received: 12 December 2006 / Accepted: 13 March 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract Understanding the relations between the
mechanical responses of whole entities, their materials
properties and their structures, is a challenge. This chal-
lenge is greatly enhanced when the material itself is
complex, and when the entity it forms has a convoluted
shape. It is for these reasons that it is still beyond the state-
of-the-art to predict and fully understand the mechanical
functions of whole biological entities such as bones and
teeth. Recent advances in optical metrology open up new
opportunities as they enable the precise and accurate
mapping of the manner in which the entire surface of a
whole stiff mineralized tissue deforms. Furthermore these
data can be obtained non-destructively and without contact
with the sample. Data of this kind create the exciting
possibility of relating the complex distribution of
mechanical properties of loaded biological materials such
as bone and teeth and their microstructures to deformations
and strains. Such studies could improve our understanding
of normal physiological processes such as skeletal aging, as
well as disease processes such as osteoporosis. They also
provide opportunities for engineers designing bio-inspired
materials to study the principles, advantages, and charac-
teristics of the behavior of hierarchical and multifunctional
materials.
In this manuscript we review optical metrology meth-
ods, highlight studies of whole body function for bones and
teeth, and in particular those studies that provide insights
R. Shahar (&)
Koret School of Veterinary Medicine, The Hebrew University of
Jerusalem, P.O. Box 12, Rehovot 76100, Israel
e-mail: shahar@agri.huji.ac.il
S. Weiner
Department of Structural Biology, Weizmann Institute of
Science, Rehovot 76100, Israel
into structure-function relations. We also outline the
potential for future studies.
Introduction
Understanding the relations between the mechanical
responses of whole entities, their materials properties and
their structures, is a challenge. This challenge is greatly
enhanced when the material itself is complex, and when the
entity it forms has a convoluted shape. This situation is
almost always the case in biology and was well recognized
by DArcy Thompson [1] in his classic publication entitled
On Growth and Form. He wrote in 1917 Matter as such
produces nothing, changes nothing, does nothing; and
however convenient it may afterwards be to abbreviate our
nomenclature and our descriptions, we must carefully
realize in the outset that the spermatozooan, the nucleus,
the chromosomes or the germ-plasma can never act as
matter alone, but only as seats of energy and as centers of
force (p. 20, volume I). Despite the years that have
elapsed since 1917, and the enormous technological ad-
vances made, it is still beyond the state-of-the-art to predict
and fully understand the mechanical functions of whole
biological entities such as bones and teeth.
Various attempts have been made to achieve this goal.
They were based mainly on measuring deformations of
whole bones or teeth at single points, or developing con-
stitutive theoretical models that can predict the behavior of
the whole bone or tooth based on the mechanical properties
of its individual constituent materials. However both
approaches were, for the most part, only partially suc-
cessful. Recent advances in optical metrology open up new
123
8920 J Mater Sci (2007) 42:89198933

opportunities as they enable the precise and accurate

mapping of the manner in which the entire surface of a
whole stiff mineralized tissue deforms.
Data of this kind create the exciting possibility of relating
the complex distribution of mechanical properties of loaded
biological materials such as bones and teeth and their
microstructures to deformations and strains. Such studies
could improve our understanding of normal physiological
processes such as skeletal aging, as well as disease pro-
cesses such as osteoporosis. They also provide opportunities
for engineers designing bio-inspired materials to study the
principles, advantages and characteristics of the behavior of
hierarchical and multifunctional materials [2, 3].

The materials of bones and teeth

Bones and the bulk of teeth are composed of the same basic
materials [4]. Only the thin hard outer layer of teeth
(enamel) is quite different [5]. The materials that bones are
composed of are also collectively called bone, whereas in
teeth (aside from the enamel) they are collectively called
dentin. They both have the same constituents which
include the mineral, carbonated hydroxyapatite (also known
as dahllite), the framework protein, type I collagen, many
other so-called non-collagenous proteins and water [6].
Enamel also contains carbonated hydroxyapatite, but the
crystals are orders of magnitude larger than those in bone
and dentin [7]. Furthermore, in most vertebrate teeth (except
for those of many fish), collagen is absent in enamel.
The materials of bones and teeth have hierarchical
structures, in that they change at different length scales (see
Fig. 1). They are also graded materials, as their composi-
tion, structure, and mechanical properties may vary con-
tinuously or in discrete steps, from one location to another
[8]. The combination of these two attributes results in a very
complex material type that certainly cannot be described in
terms of one value for a particular materials property. The
following is a brief description of the hierarchical structures
of bone (the material), dentin, and enamel.
The hierarchical structure of bone
Bone is a term used to denote both the organ and the sponge-like (cancellous) (level 6 Fig. 1). This entire com-
material of which it is composed. The material bone is a
bio-composite with a complex and hierarchical structure.
The various hierarchies of bone span many different
length scales. The basic building blocks of the material
bone are an organic matrix, most of which is type-I col-
lagen, and a mineral phase (level 1 Fig. 1) [4]. At a higher
length scale bone is composed of mineralized collagen
fibrils 80100 nm in diameter (level 2 Fig. 1), arranged in
a regular, staggered array of collagen molecules (level 3
Fig. 1) [9]. The crystals are located within and around
Fig. 1 Hierarchical structure of bone (from Weiner and Wagner [4],
with permission)
these fibrils. They are 3080 nm wide and 23 nm thick
[10]. At the next hierarchical level the mineralized fibrils
are organized in a variety of patterns, the most common of
which is the lamella (level 4 bottom left in Figs. 1 and 2).
Each lamella is 23 microns thick and has a complex
rotated plywood structure [11].
At the next hierarchical level, lamellae are arranged in
one of several possible ways, depending on location and
species; in mature human bone the most common
arrangement is concentric layers (called secondary ost-
eons). These form cylinders 150250 lm in diameter and
contain a central hollow tube 80 lm in diameter, which
contain blood vessels and nerves (level 5 Fig. 1) [12]. At
the next level bone can be mostly solid (cortical) or
plex arrangement, with different but characteristic 3-D
geometric morphologies, ultimately forms the organ bone.
The structures of tooth enamel and dentin
Teeth consist of a crown (the part above the gum line) and
one or more roots (the part below the gum line). The crown
has an exterior layer of enamel and the root has an exterior
layer of cementum. Both parts have a continuous interior
layer of dentin, which surrounds the pulp cavity (see Fig. 3).

123
J Mater Sci (2007) 42:89198933
Fig. 2 SEM image of the fracture surface of lamellar bone from a
baboon tibia. Different orientations of mineralized collagen fibrils
within a lamella can be noted. Canaliculi are shown aligned more or
less orthogonally to the lamellae (white arrows)
Fig. 3 Tooth anatomy (diagram obtained from: http://www.enchant-
edlearning.com/Home.html)
Enamel
Enamel consists of 96% mineral, 1% organic material, and
3% water. The basic building blocks of enamel are ribbon-
like crystals measuring 6070 nm in width, 2530 nm in
thickness and their length may span the entire thickness of
the enamel layer [13]. They are organized into cylindrical-
8921
shaped rods which are surrounded by interrod enamel [5].
Rods are made up of crystals whose long axes are aligned
with the long axis of the rod. The interrod region surrounds
the rods and in it the crystals are oriented at an angle to the
rod. The boundary between the rod and interrod contains
organic material called rod sheath [5].
Dentin
Dentin is bone-like in composition, containing by volume
around 50% mineral, 30% organic material, and 20%
water, making it softer but tougher than enamel [5]. The
bulk of the tooth is composed of dentin, which has
the same hierarchical structure as bone up to the level of
the lamellae. At this level in dentin the mineralized
collagen fibrils are all arranged in one plane (more or less
parallel to the outer surface of the tooth), but there is little
or no preferred orientation within the plane (level 4 bottom
right Fig. 1) [14]. This structure is characteristic of the
material of the tooth root and part of the crown. In the
crown, however, there is an additional dentin type, called
peritubular dentin [15]. The peritubular dentin forms a
pipe-like lining about one micron thick around the
so-called dentinal tubules (Fig. 4a). The tubule diameters
are also about one micron. There are huge numbers of such
tubules in both the root and the crown of the tooth. The
peritubular dentin has little or no collagen and is therefore
much denser than the intertubular dentin [16]. At a higher
level of organization, the 50 to 200 micron-thick zone
beneath the dentin-enamel interface (the so-called dentino-
enamel junction or DEJ) has a structure quite different
from the other forms of dentin (see Fig. 4b). It is less
mineralized than the bulk dentin of the crown and the
collagen fibrils have a different organization [17]. As a
result, it is less stiff than the bulk of dentin.
The graded structure of bone and tooth materials
The mechanical response of materials with spatial gradi-
ents in composition and structure is of considerable interest
in disciplines as diverse as tribology, geology, optoelec-
tronics, biomechanics, fracture mechanics, and nanotech-
nology [8]. The damage and failure resistance of surfaces
to normal and sliding contact or impact can be changed
substantially through such gradients. The composition,
structure, and mechanical properties of a material may vary
continuously or in discrete steps with depth beneath a free
surface. Gradations in microstructure and/or porosity are
commonly seen in biological structures such as bamboo,
plant stems, and bone, where the strongest elements are
located in regions that experience the highest stresses.
Learning from nature, materials scientists increasingly aim
123
8922
Fig. 4 (a) Crown dentin microstructure: Typical scanning electron
microscopy (SEM) micrograph of a fracture surface (dehydrated) of
crown dentin. The dentinal tubules (black arrowhead) and intertubular
dentin (white arrow) can be clearly seen. Crown dentin has an extra
phase of highly mineralized peritubular linings surrounding the
tubules (black arrow). (b) The large enamel crystals can be seen on
the right side, and the dentin (including the dentinal tubules) on the
left. The DEJ forms the border area between them, in the center of the
picture
to engineer graded materials that are more damage-resis-
tant than their conventional homogeneous counterparts [2].
This is particularly important at surfaces or at interfaces
between dissimilar materials, where contact failure
commonly occurs.
Bone and teeth are both fine examples of naturally
occurring graded materials. The graded structure of bone
can be seen at several levels. At the osteonal level for
example, differences in the level of mineralization occur
within osteons, as the inner (younger) lamellae are less
mineralized than the outer lamellae. The outer border of
an osteon (so-called cement line) is believed to contain
very little collagen and is less mineralized than the
surrounding bone [12], whereas the so-called interstitial
lamellae (remnants of old osteons) are more highly
mineralized. This graded structure critically affects the
mechanical behavior of bone, and in particular crack
propagation, as most cracks tend to advance until
reaching cement lines, at which point they stop due to
this weak interface and travel around the osteon rather
123
J Mater Sci (2007) 42:89198933
than through it. At another level, the transition between
cortical and cancellous bone is also graded, and can be
observed grossly in terms of increasing bone porosity
from the periosteal to the endosteal surface.
Teeth are also quite obviously graded. Tooth enamel
hardness decreases away from the external surface towards
the DEJ. At the DEJ hardness drops precipitously for a
distance of 200300 lm (dentinal soft zone), and then
partially recovers [18, 19]. As a result, the enamel cap
displaces primarily as a rigid body, and the soft dentinal
zone serves as a cushion between the stiff enamel and less
stiff dentin. Within the crown dentin, the materials prop-
erties continuously change from one location to another
[20], and the intertubular and peritubular dentin also
appears to have a graded interface [18, 21].
Mechanical performance of whole bone and teeth
To date most mechanical studies of bones and teeth have
investigated the mechanical properties of each of the
bulk materials that make up the bone or tooth. These
studies were limited by the minimum size of samples
needed for measurements carried out in materials testing
machines. As a result osteonal bone is well investigated,
as is plexiform bone from bovids. Measurements of other
bone types such as circumferential lamellar bone, parallel
fibered bone and also root and crown dentin, were
infrequently carried out since special circumstances or
experimental set-ups are required due to the small size of
the samples available. In the last 10 or so years the
nanoindenter has been used to measure elastic properties
at a very high spatial resolution [22, 23]. This method
has greatly increased our understanding of the graded
nature of these biological materials and the differences
that exist between the mechanical properties of different
structural types.
Most investigations of the mechanical performance of
whole bones have to date relied on in vivo or in vitro
experiments in which bones were tested in compression
or by 3- or 4-point bending. Measurements in these
experiments consisted of single values like load-to-yield
and load-to-failure. Strain measurements were made at a
very limited number of points (120) by placing strain
gauges on the surface of bones and teeth. Alternatively
theoretical numerical models (mostly finite element
models) were used, which were based on numerous
simplifying assumptions, particularly as far as material
properties were concerned, and thus yielded rough
approximations only. Optical metrology offers an alter-
native approach which yields data of substantially greater
value.
J Mater Sci (2007) 42:89198933
Optical metrology methods
The main advantages of all optical metrology techniques
are the full-field nature of the measurements they provide,
and that they are non-destructive [24]. In essence a mea-
surement is obtained from each pixel in the field of view.
The number of pixels is dependant upon optical resolution,
and is usually around 105. In this regard, results obtained
experimentally by optical methods are similar to results
that would be obtained by a very large (and practically
impossible) array of strain gauges.
Optical methods provide full-field surface measurements
of either stress (photoelasticity) or displacements and
strains (various interferometry methods and digital image
correlation (DIC)). Full field measurements, be they
deformation, strain or stress, have several advantages: they
can identify local strain or stress peaks and gradients, which
would be missed by measurements made with few strain
gauges. This is particularly significant in the case of inho-
mogeneous, anisotropic, graded biological materials. They
are also compatible with the results of finite element anal-
ysis, and therefore facilitate model validation. Furthermore,
the full-field data set makes it possible to vary finite element
models in terms of material properties until the predicted
numerical results agree with measured results, providing
deeper insight into local material properties.
All optical metrology methods are based in one way or
another on interference. Interference is a phenomenon that
occurs when two or more waves overlap each other in
space. The superposition principle states that the resulting
field is the sum of the original fields. The resultant intensity
however is not merely the algebraic sum of the intensities
but also depends on the phase difference between the dif-
ferent waves. Thus interference may be constructive or
destructive.
Here we review optical metrology methods, highlight
studies of whole body function for bones and teeth, and in
particular those studies that provide insights into structure-
function relations. We also outline the potential for future
studies.
Photoelasticity
Principle of the method
Photoelasticity was one of the first optical metrology
techniques used for the study of whole bones and teeth. The
method relies on the property of birefringence that certain
so-called photoelastic materials exhibit when subjected to
stress. Birefringent materials have two different refractive
indices when light passes through them. Furthermore, the
magnitude of the refractive indices at each point is directly
related to the state of stress at that point. When using this
8923
method, the structure under investigation is coated with a
photoelastic material, or a replica made of photoelastic
material is produced whose geometry is identical to that of
the structure investigated. When such a sample is illumi-
nated by polarized light after loading, its birefringence
causes the light to become refracted in the two orthogonal
principle stress directions. The difference in the refractive
indices leads to a relative phase retardation between the
two component waves. The two waves are then brought
together in a polariscope, and interference takes place. As a
result fringe patterns form, which depend on the relative
retardation. These fringe patterns allow the determination
of the state of stress at the surface of the object tested.
Despite its appeal as a method able to map surface stress
distributions, photoelasticity has several inherent weak-
nesses. When a sample is coated with a photoelastic
material, potential errors include a mismatch of Poissons
ratio between the coating material and the investigated
structure, incorrect light incidence, uneven coating thick-
ness and the potential reinforcing effect of the photoelastic
material [25]. When a replica is used, only the geometry is
simulated, and the complex materials properties are not
taken into account. Furthermore, replicas or coated models
cannot interact in a biologically relevant way with physi-
ologic fluids, thus in essence a dry sample is tested. This
further reduces the validity of the experimental results.
Teeth
One of the first applications of 3-D photoelasticity to study
the behavior of whole teeth was by Johnson et al. [26].
They measured the stress fields occurring in epoxy resin
replicas of human molar teeth with different types of cavity
preparations. They showed that when cavities are prepared
with rounded angles, stress concentration is minimized
compared to cavities prepared with sharp angles.
The nature of the stress distribution from the tooth root to
its supporting alveolar bone was investigated by Asundi and
Kishen using a combination of in vivo strain gauges and
in vitro photoelasticity experiments [2731]. They used
epoxy resin to create a model of a normal tooth and its sup-
porting alveolar bone, which included a thin layer of silicon
rubber between the bone and the tooth to simulate the peri-
odontal ligament. Using digital photoelasticity, they showed
the stress distribution pattern in the tooth and supporting
bone replica (see Fig. 5). The replicas were loaded at several
angles relative to the long axis of the tooth. As the loading
force direction formed a larger angle with the long axis of the
tooth, stresses in the alveolar bone increased. They also
showed that stresses decreased from the top (cervical) to the
base (apical) region of the root, and that most of the bite
forces along the tooth axis caused loading of the alveolar
bone in the cervical region, and to a lesser extent in the apical
123
8924
Fig. 5 Stress patterns in tooth and alveolar bone, showing areas of
stress concentration (SCareas of increased density of fringes)
(From Asundi and Kishen [27], with permission)
region. They suggested that this might protect the soft tissues
(blood vessels and nerves) entering the root at its apex. Thus
these studies demonstrate well the use of whole-field results
to better understand aspects of the behavior of the entire tooth
in its bony socket. This could not be achieved only with
strain-gauge measurements.
The above experiments represent the stress distribution
based solely on geometry. Kishen et al. [32] overcame this
weakness by comparing the distribution of the elastic mod-
ulus of dentin (using microindentation on tooth sections)
with the stress patterns that developed in the photoelastic
model. Dentin was shown to be a graded material, and its
properties (elastic modulus, degree of mineralization) varied
spatially. The stress pattern observed in the homogeneous
photoelastic material predicted bending with compressive
stresses on the facial side and tensile stresses on the lingual
side. The authors speculated that teeth adapt their structure to
the loads they experience, as do bones. However, since teeth
do not change their morphology, adaptation occurs in the
form of creation of material property variations. These
variations allow teeth to avoid areas of high stresses and
result in uniform stress distributions.
Kishen and Asundi [33] also used photoelasticity to
investigate the behavior under load of individual teeth with
and without endodontic treatment. They showed that end-
odontic treatment resulted in higher tensile stresses and
stress concentration in the remaining dentin than stresses
occurring under equivalent loads in intact teeth.
These studies elegantly combine photoelastic measure-
ments with other direct materials properties measurements
to produce a more integrated understanding of whole tooth
behavior.
Bone
Several studies used photoelasticity to investigate bone and
bone-implant mechanical behavior [34, 35]. However,
123
J Mater Sci (2007) 42:89198933
since they all used replicas made of photoelastic resin, the
effect of bone structure could not be evaluated and results
can only show trends, since the material tested is homo-
geneous and isotropic; a far cry from the complex, aniso-
tropic and graded bone material evaluated.
Murphy and Prendergast [25] demonstrated another
potential use of photoelastic-based experiments. They first
created a finite element model to compare two methods of
attachment of the implant used in shoulder replacement
surgery. They then used reflective photoelasticity on 5
scapulae coated with a thin photoelastic layer and loaded as
in the model to validate their predictions. The photoelastic
and finite element results were similar, creating confidence
in the results of their numerical simulation. We note that
both the numerical model and the experiment use a sim-
plified representation of the bone (isotropic and homoge-
neous) hence results must be interpreted with caution and
are likely to be different in the in vivo situation.
Deformation-measuring interferometry techniques
Interferometry techniques used for optical metrology are
based on illumination of the investigated sample by coherent
laser light and on the resulting interference phenomena that
occur in various ways. These methods are capable of
detecting sub-micron displacements. They include Moire
interferometry (MI), holographic interferometry (HI), and
electronic speckle pattern interferometry (ESPI).
Moire interferometry
Principle of the method
Moire interferometry requires the application of an adher-
ing grid of parallel lines to the surface of the sample. This
grating deforms together with the sample surface. Another
identical grating is overlaid on the sample surface. Inter-
ference between the un-deformed and deformed gratings in
reflected light produces Moire fringes, from which the two
components of displacement at each point on the surface
can be determined. Differentiation of the displacement
fields makes it possible to also determine the surface strain
tensor. Various Moire methods, such as Moire fringes and
digital MI, have frequently been used to study the strain
distribution in bones and teeth under load.
The need to apply gratings to the sample surface of
biological materials is the main limitation of this technique,
since such application is technically demanding in small
samples, may affect the mechanical behavior of the sample,
restricts access of water and limits the sensitivity and
precision of the resulting measurements. Another limitation
is the 2-D nature of this method, since out-of-plane
displacements cannot be determined. Higher resolution,
J Mater Sci (2007) 42:89198933
non-contact methods such as HI and ESPI overcome these
deficiencies (see below).
Teeth
A key insight into the design strategy of teeth was obtained
using Moire fringes on slices of human teeth [18]. The
study showed that under compression much of the strain is
taken up in a zone some 200 microns thick that separates
the stiff outer enamel layer from the crown dentin. This
zone was known from hardness measurements to be much
softer than the enamel and the underlying dentin. It was
proposed that this soft zone is an important working part direction parallel to the tubules and perpendicular to them.
of the teeth in that it acts as a cushion or a gasket during
mastication.
Huo [36] extended this study by using a finite element
model with the same geometry as the tooth slices. Four
finite element models, using different types of material
properties for dentin were tested. They ranged from
homogeneous and isotropic properties to an inhomoge-
neous and anisotropic model. The latter model yielded
numerical results that most resembled the experimental
data produced by Wang and Weiner [18]. Clearly the
model can now be used to better understand the structure
mechanical relations of teeth by systematically varying the
input parameters and the geometry. This is a very prom-
ising use of a finite element model to extend the data and
improve our understanding of the functional attributes of
this soft zone, or for that matter other design features of a
tooth or bone.
A follow-up of the Wang and Weiner experiment was
performed by Wood et al. [37] using MI. This experiment
was also designed to investigate the material properties of
the soft zone, but using a completely different approach.
Slices of the tooth crown were cut perpendicular to the long
axis of the tooth. In this way the enamel formed a con-
tinuous ring around the dentin. In other slices the enamel
was not continuous. The changes in shape of the slices
were then monitored as a function of changing humidity.
The results clearly showed that if the enamel constrained
the dentin then very little deformation occurred during
drying. If however the enamel was discontinuous, then
much of the strain was localized in the soft zone as the
dentin dried. This too therefore demonstrated the unique
materials properties of this zone. In addition, the experi-
ment showed that a continuous enamel ring prevents the
dentin from contracting in this zone during drying. This
presumably implies a very strong interface between the
dentin and the enamel.
The role of water in the mechanical behavior of dentin
was also investigated by Kishen and Asundi [38]. Several
human incisors were ground to prepare parallel-sided
longitudinal sections, and a high-frequency grating was
8925
attached to the specimen surface by a thin layer of epoxy.
The dentin was divided into an outer region (close to the
enamel) and an inner region. They showed that when hy-
drated, both regions underwent the same lateral strains
(along a line perpendicular to the long axis of the tooth).
However, when dehydrated there was a significant differ-
ence in lateral strain between the two regions (higher in the
outer region). They found that dehydration of dentin in-
duced residual compressive strains in the outer region of
dentin, as was observed by Wood et al. [37] for uncon-
strained dentin. This showed that free water has a signifi-
cant impact on the stress-strain response of dentin in a
The presence of water increased the toughness of dentin,
while loss of water increased its stiffness. The authors point
out that the dentinal collagen fibrils are made of microfi-
brils separated by spaces filled with water molecules.
Dehydration of the dentin leads to loss of these spaces and
allows the collagen polypeptide chains to contact each
other, and form molecular associations not formed in the
hydrated state. These associations stabilize the structure
and increase its stiffness.
Moire interferometry has also been used to determine
the deformation gradients of enamel and dentin [39]. The
purpose of this study was to look for a biomechanical basis
for a significant problem affecting the teeth of humans,
namely non-carious cervical lesions, by studying the strain
distributions in enamel and dentin in human teeth loaded in
compression. They observed that while the enamel under-
went marked strain gradients in the direction perpendicular
to the long axis of the tooth, the crown dentin experienced
marked strain gradients along the tooth axis (see Fig. 6).
As load increased, the strains in the enamel increased in the
area above the base of the crown (cemento-enamel junc-
tion), while strains in the dentin increased below the
cemento-enamel junction. These unique in-plane strain
patterns led the authors to hypothesize that loading caused
by biting forces contributes to strain concentration at the
junction between the crown and the root of the tooth. This
in turn could cause the formation of cavity-like lesions in
this area which are attributed to mechanics rather than
bacterial activity.
This example illustrates the ability of optical metrology
methods in investigations that aim to gain a fundamental
understanding of the design strategies of mineralized
organs like teeth and bone. In particular the mechanical
behavior that results from the complex interplay of the
organs external morphology and internal structure, com-
bined with the spatial distribution of material properties
can be measured. Such data helps explain the normal
behavior of teeth, and the occurrence of common lesions
such as non-carious defects in the enamel at the crown-root
junction.
123
8926
Fig. 6 Typical digital MI fringe
patterns in directions along (A,
B) and across (C, D) the long
axis of the tooth, for loads of
10 N (a, c) and 30 N (b, d)
(from Kishen et al. [39], with
permission)
Holographic interferometry
Principle of the method
Holography is a general term describing techniques for
recording and reconstructing wavefronts. Holograms are
patterns produced by the interference of object and refer-
ence beams and recorded on a recording medium [40].
Holographic interferometry allows the comparison of
wavefronts recorded at different instants in time. When HI
is used to measure the surface deformation of loaded
samples, a hologram of the unloaded sample is obtained
and stored. It is then allowed to interfere with a wave
scattered from the loaded, and therefore deformed sample.
The phase difference between the pre-load and post-load
wave fronts is calculated, and this describes the deforma-
tion that occurred during this time interval.
Early applications of this technique used photographic
plates as the recording medium. This process was time con-
suming, especially the development of the photographic films
required. In recent years CCD sensors and powerful com-
puting resources made it possible to directly record the holo-
grams and evaluate them digitally by computer simulation
(digital HI). Two procedures can be used: (1) Double exposure
interferometry, in which two exposures of the sample are
made on the same hologram, i.e. pre- and post-loading. By
reconstructing the hologram, the two waves scattered from the
sample in the 2 states will interfere. (2) A single recording of
the sample in the reference state is made, and then the holo-
gram is processed and replaced in the same position as in the
recording. By looking through the hologram it is now possible
to observe the interference between the non-constructed
sample and the wave from the sample in its original position.
123
J Mater Sci (2007) 42:89198933
Thus deformation can be followed in real time by observing
dynamic changes in the interference pattern.
Digital HI is very sensitive to vibrations, and consider-
able amount of noise may be present in the data. This is
especially true when relatively low-resolution recording
devices (CCD camera or complementary metal-oxide
semiconductor (CMOS) sensors) are used. When samples
are measured under water currents and vibrations in the
water increase the difficulty of obtaining reliable results.
Noise reduction can be achieved by using faster, higher
resolution recording devices, and taking all precautions
possible to damp out vibrations.
Bone
One of the first reports of the use of double exposure HI
described the measurement of strains on the surface of
artificial femora made of glass fiber epoxy composite. The
femora, with and without implants, were loaded in com-
pression along their long axes [41]. The authors showed
that implantation resulted in significant changes in the
femoral strain patterns. In particular, significant decreases
in peak strains in the upper half of the bone (the region
which contains the implant) were observed. This observa-
tion confirmed the suspected phenomenon of stress
shielding in implanted bones, where the implant, which is
much stiffer that the surrounding bone, bears most of the
stress, and the bone experiences lower than physiological
stress. This causes bone loss (the biologic response to
lower load) and makes the bone more susceptible to frac-
ture under non-physiologic impact loading, like a fall.
It should be noted that the behavior of femoral replicas
made of homogeneous and isotropic material, only yields
J Mater Sci (2007) 42:89198933
information on the effect of the bones geometry on the
strain fields.
Double-exposure holography was also used to determine
bone deformation (swelling and shrinking) due to thermal
stress [42]. Thermal stresses were used simply because
they were simple to apply. The experimental results were
compared to measurements made with a thermo-graphic
measuring system, and a good correlation was found
between the temperature and displacement distributions. It
was also shown that the deformation fields differed
depending on which point was used for the application of
the heat source.
Osten and coworkers [43, 44] reviewed the potential of
HI in biomechanics. They described a system, which uses
two laser sources and two fast CCD cameras. The latter
concurrently record the out-of-plane and in-plane defor-
mation components. The deformation fields of notched
samples of antler bone were followed dynamically at a rate
of 30 hologram pairs per second. The deformation maps
clearly demonstrated the presence of high displacement
gradients (equivalent to strains) at the tip of the notch (See
Fig. 7). Further refinement of this system, using ultra-fast
and high-resolution cameras and use of three lasers and
three fast cameras will allow experiments to follow fracture
phenomena in hitherto unattainable detail and obtain all 3
displacement components, thus significantly extending the
potential application of HI to the study of biologic samples.
Electronic speckle pattern interferometry
Principle of the method
Another optical method which allows whole-field measure-
ment of displacements of optically rough surfaces is electronic
speckle pattern interferometry. It is based on the observation
that when a rough surface is illuminated by a coherent source of
light, the reflected light beams interfere locally with each other,
creating a grainy image termed a speckle pattern. The speckle
pattern changes when the specimen is loaded. The differences
between patterns of images obtained before and after applying
load are used to detect shifts in the phases of speckles [45, 46].
These phase differences correspond to surface displacements,
and their magnitudes can be determined using one of several
available phase-shifting algorithms. Phase unwrapping allows
the determination of the entire displacement field. One of the
main advantages of this method is its extreme sensitivity,
which allows detection of displacements as small as tens of
nanometers. (The limit of detection is ~k/30, were k is the
wavelength of the laser light; typically several hundreds of
nanometers). There are several additional ways to utilize
speckle-based methods. One such method, called shearogra-
phy, deserves special mention, because it allows direct mea-
surements of displacement derivatives, which can be related to
8927
strains. This method is based on causing interference between
rays returned from two neighboring surface points. When two
such points lie on the same Cartesian axis and are very close to
each other, the relative phase change occurring before and after
deformation can be related to the partial derivative of the
deformation along that axis. One of the advantages of this setup
is its reduced sensitivity to vibrations [47].
The potential of ESPI to measure deformations of loa-
ded biomineralized tissues such as teeth and bones
increased significantly after it was demonstrated that such
measurements can be made with the sample under water
[19, 48, 49]. This is essential for the investigation of bio-
logical specimens. This technique allows not only the
investigation of deformations and strains of entire organs,
but also the determination of the elastic properties of
millimeter-sized samples with good precision and accuracy
in teeth [48] and bone [49]. Furthermore, by improving the
signal-to-noise ratio it will be possible to determine local
strain variations within these small samples and correlate
them with microstructures.
Electronic speckle pattern interferometry has several
limitations. It is extremely sensitive to ambient vibrations,
and requires an isolated environment, which usually
includes an acoustically insulated system mounted on top
of a floating optical table. Applied loads must be small to
avoid large displacements which could lead to decorrela-
tion. Therefore when larger loads are of interest, they must
be reached by a series of small and consecutive incremental
loads, with displacement fields determined for each incre-
ment, and then summed.
Another technical difficulty is the fact that fluctuations
in speckle intensities may occur during the time between
obtaining the reference image and loaded image, especially
when wet biologic samples are tested, resulting in decor-
relation. This makes data analysis very challenging. Kirk-
patrick and Brooks [50], in an attempt to overcome the
difficulties associated with speckle fluctuations, used a
modification of classical laser speckle data analysis, which
was based on a sequential sampling algorithm which is
insensitive to slow speckle decorrelation (relative to the
sampling rate). In a preliminary proof-of-concept type of
experiment, they showed that when a sample interval of
0.02 s was used, reliable results could be obtained. They
applied this technique to obtain the Youngs moduli of
samples of porcine cortical bone. Another approach was
based upon applying very small incremental loads between
successive images (reference and loaded state), thus cre-
ating displacements which are much smaller than the pixel
dimensions. In this technique each experiment consists of
many individual steps, and avoids decorrelation. Using this
approach, Youngs modulus (human dentin and equine
cortical bone), and Poissons ratio (equine cortical bone)
were measured [48, 49].
123
8928
Fig. 7 Deformation of a notched piece of antler bone immersed in
water, measured by digital HI, at three different times (0.5, 1, and
1.5 s after the beginning of the loading process). (a), (c), and (e) are
Last, obtaining the three components of displacement
requires changing the laser configuration during measure-
ments, and this process requires time. This makes ESPI less
suitable for dynamic measurements. A novel approach taken
to shorten the acquisition time of in-plane and out-of-plane
deformations was reported by Rodriguez et al. [51]. They
developed an ESPI system which allowed easy, precise and
rapid alteration of the direction of the laser beams illumi-
nating the surface of the object, by use of optical fibers
attached to a rotating platform to split and guide the light
beam. To demonstrate the abilities of their system, they
measured the surface displacements of a human lower jaw
123
J Mater Sci (2007) 42:89198933
the phase maps and (b), (d), and (f) are the corresponding
deformations in the y-direction
bone (mandible) resulting from two different loading loca-
tions. However in this preliminary study only the general
low-resolution pattern was evaluated and quantitative dis-
placement data were not obtained. It should also be noted that
this set of experiments was performed on dry specimens, and
therefore does not represent the real deformations this bone
will undergo under physiologic load.
Teeth
Zaslansky et al. [19] used ESPI to measure the deformation
distributions of whole human premolar teeth (see Fig. 8).
J Mater Sci (2007) 42:89198933
Fig. 8 RDM maps showing
deformation of a premolar
during incremental loading
(From Zaslansky et al. [19],
with permission)
Experiments were performed in a custom-designed
micromechanical compression device. The entire setup was
located within a sealed stainless-steel chamber, and the
teeth were submerged in water during the experiment (see
setup details in [48]).
Compressive load was applied incrementally either to a
localized area at the tip of the main cusp (simulating hard
food mastication) or over most of the upper region of the
main cusp (simulating soft food mastication). The same
protocol was also applied to exact replicas of the tested
teeth. It was hypothesized that any difference in the
deformation patterns between the real tooth and its replica
will be the result of the tooth structure. In order to
interpret the spatial distribution of the whole-field dis-
placements a novel approach was developed, named rel-
ative displacement maps (RDM). These provide a
reference-independent map of the 3-D surface displace-
ments (see [19] for details).
The teeth and their replicas deformed in roughly the
same manner. This indicates that the overall deformation
pattern of the premolar is mostly controlled by shape, and
to a much lesser extent by its complex internal structure.
An interesting difference was seen between the defor-
mations occurring due to soft food and hard food masti-
cation. When subjected to a more concentrated load
(simulation of hard food) the tooth undergoes substantial
bending. Another significant finding was the much more
uniform deformation of natural teeth compared to their
replicas. This was attributed to the properties of the
enamel cap that behaved mainly as a rigid body, with
only moderate deformation at higher loads. One particu-
larly interesting observation was that the location of
minimal deformation of the natural teeth, but not the
replicas, was in the region of contact between neighboring
teeth. This is presumably a built-in design feature to
minimize abrasion between teeth.
Bone
Electronic speckle pattern interferometry has been used to
study the three-dimensional strains on the surface of mice
8929
femora when load was applied either along the long axis of
the femur by compressing the femoral head, or from medial
to lateral surfaces at the knee region [52, 53]. It has shown
that the two different modes of loading resulted in very
different patterns of strain distribution on the surface of the
entire femur. In these studies the femora were tested dry,
and therefore the results may not represent their true
physiologic behavior. Studies such as this further under-
standing of the response of whole bone to load. They also
provide insight into the remodeling process that occurs
when a change in loading pattern is applied to bones.
Another potential use of ESPI was recently reported by
Mohr et al. [54]. They assessed the process of fracture
healing by harvesting mid-sagittal sections of fracture
callus (a layer of cartilage and bone produced by the body
to bridge the fracture area) from sheep 8 weeks after
fractures were created in the middle of their sheen bone
(tibia). The fractures were stabilized with an external metal
device (external fixator frames). The slices were subjected
to axial compression, and the strain distribution in the
entire callus was measured by ESPI. Slices with complete
bony bridging were shown to have higher stiffness than
those consisting mostly of connective tissue bridging.
Highest strains in all sections were seen in the fracture gap.
Continuation of this study could significantly further
understanding of the fracture healing process and provide
answers to such questions as the mechanical consequences
of the sequence of events occurring during the fracture
healing process, and the most suitable conditions required
for successful healing.
Digital image correlation
Principles of the method
Digital image correlation is an optical metrology tech-
nique which is based on entirely different principles
from those of interferometry-based techniques described
thus far. Digital image correlation is performed by
matching high-resolution structural patterns or color-
sprayed random image patterns of the surface of objects
123
8930 J Mater Sci (2007) 42:89198933

before and after they are placed under load. Displacement

fields can be determined by using photogrammetry to
compare images of the sample before and after load
application. An array of measurement points is digitally
superimposed on the image of the sample in the pre-load
state, to define the locations of displacement measure-
ment. The image processing system stores a small sub-
area based on its unique surface texture surrounding each
measurement point in the preload image (either the
inherent texture of the sample or by spraying a contrast
agent), and a search is performed in a defined area in the
post-load, deformed image. The displacement search
procedure is performed by one of a variety of image
correlation algorithms. The algorithm determines the
position of the deformed points with respect to the
undeformed image. The method is not sensitive to rota-
tions; however it detects rigid body motion and defor-
mation. The resulting deformation fields can be smoothed
and differentiated, thus yielding the surface components
of the Lagrangian strain tensor.
Digital image correlation has several advantages over
measurements by ESPI. It is much less sensitive to ambient Fig. 9 Local strain distribution overlaid on a photomicrograph of
vibrations, can detect rigid body motion, can simulta- cortical bone. Note: Principal strain peaks near osteocyte lacunae and
microcrack tip (Figure kindly provided by Prof. Nicolella)
neously measure 3-D displacements and has a high dynamic
range (microns to millimeters). Its main disadvantage
compared to ESPI is its lower sensitivity, which is deter- much higher strains are measured locally. This observation
mined by the field of view and does not exceed 0.3 microns has a major impact on the various strain-driven theories of
[47, 55]. the remodeling process.
In the last decade several research groups reported on a To date, DIC has not been used to measure
wide array of applications of DIC for the measurement of deformations of entire bones or teeth. However the
displacements and strains of loaded biological tissues. The advantages of this method over interferometry methods
most-often investigated tissues were cortical bone, can- (especially the low sensitivity to vibrations and ability to
cellous bone and articular cartilage; other tissues were also measure large deformations) make it particularly suitable
studied, from individual cells to large tissues such as the for this purpose.
wall of bovine arteries, the bovine hoof horn and the bone-
cement-implant system [56]. Discussion

Bone The study of the mechanical behavior of whole organs

The group of Nicolella and co-workers published a series of
elegant papers, in which they examined the high-resolution
strain distribution within small samples of cortical bone
[5760]. In these studies they have associated and related
local strain variations to meso-scale structural elements such
as osteocyte lacunae and microcrack tips, and compared
them to the global strain of the entire sample. They found
that for global sample strains at ~2,000 microstrain, the local
strain at the osteocyte lacunae reached values as high as
12,00015,000 microstrain, and even ~30,000 microstrain
near crack tips (see Fig. 9). These studies demonstrated that
while the global strain, as measured by strain gauges, is in
the physiologically accepted range (2,000 microstrain),
(bones and teeth) is only at its initial stages. While it seems
reasonable to assume that the extremely convolute three-
dimensional geometry and intricate material distribution
and architecture of bones and teeth, and the hierarchical,
complex and graded nature of the materials of which they
are made, all play important roles in their response to
mechanical loads, real understanding of these relationships
is currently lacking. In the past studies of such questions
were hampered by the inherent limitations of the available
methods of investigation.
The application of advanced optical metrology techniques
to the study of biomineralized tissues has therefore much
future potential. The addition of such recent technological
advances such as extremely fast CCD cameras which can

123
J Mater Sci (2007) 42:89198933
obtain images at rates of up to 10 KHz, and readily available
powerful computing power which can easily handle vast
quantities of data in real time, will allow researchers in the
field of whole-bone and whole tooth function to use inter-
ferometry techniques and DIC to answer questions which
seemed beyond our reach only a short time ago.
The most basic and intriguing question is that of the
deformation (and resulting strain) distributions of whole
organs under various types of load. Until recently it was
considered impossible to elucidate the strains and stresses
occurring in bones and teeth of complicated shape and
structure. As has been shown in the preceding sections,
while optical metrology techniques show great promise in
this respect, attempts to use these methodologies to
investigate this question are only beginning.
Bones come in an impressively large assortment of
shapes and sizes. Within each bone the different hierarchies
are themselves distributed in a very varied way. There are
areas which are more mineralized than others, areas with
secondary osteonal bone while other areas have primarily
parallel lamellae, areas with thick cortex and others with
very thin cortex, areas with cancellous bone or without
cancellous bone, and so on. It is reasonable to assume that
most of these variations can be rationalized in mechanical
terms, but until now such rationalizations could only be
based on speculation and very little experimental support.
Testing whole bones with optical metrology methods can
allow a much more detailed exploration of the mechanical
significance of the shape and structure of bones.
The organ bone must be both stiff (to resist deformation
under the action of muscles) and strong (to bear load
without breaking). A further requirement is the basic need
to minimize weight so as to decrease the energy cost of
locomotion. The solution nature found to these conflicting
demands is a structural motif common to almost all bones
in all speciesan outer cortical shell of varying thickness,
which surrounds cancellous bone in some regions. Cortical
bone is almost solid, with very few voids in it, while
cancellous bone (also termed spongy or trabecular bone)
consists of slender (50150 lm in diameter) and relatively
long (12 mm) struts surrounded by large spaces. It has
been shown that cortical and cancellous bones are made of
the same material, possibly with minor differences in the
level of mineralization. However due to the different void
contents, a volume of cortical bone is much stiffer
(10100 times) than an equal volume of cancellous bone.
Therefore it is commonly believed that most of the stiffness
of bones is due to their cortical shells. The cortico-can-
cellous structural arrangement raises the question: what is
the mechanical function of the cancellous bone compo-
nent?
Only a small number of studies tried to address this
question experimentally, and results were limited due to
8931
methodology limitations. Using optical metrology it is now
possible to obtain a quantitatively precise and accurate
answer to this question. Bones can be scanned to determine
with high resolution their structure (in terms of voids,
mineral content, sizes, and orientation of cancellous struts,
etc.), and then loaded while an optical measurement
method is used to measure surface displacements. Since the
methods are non-destructive, the load can be removed and
the bone treated to remove a predetermined amount of
cancellous bone from a selected region (quantified by a
post-removal microCT scan). The loading experiment can
now be repeated, and surface displacements of the modified
bone measured. Direct comparison can be made between
the pre- and post-removal displacement and strain fields,
showing quantitatively the effect of the removal of can-
cellous bone regions. Preliminary results of such an
experiment demonstrate an obvious difference in strain
distributions between the two states in a rat femur.
One of the most outstanding attributes of all optical
metrology methods is their compatibility with the results
obtained by finite element analysis. The finite element
method (FEM) is a commonly used engineering numerical
method which is used to determine the stresses, strains and
deformations of structures of complex geometry and
material properties when placed under load. Parametric
analysis of the contribution of each component can be
easily achieved by this technique. FEM is frequently used
to model various biomineralized tissues, such as bones and
teeth [36, 6165]. In such models, the geometry, even
when complicated, can be simulated very accurately by
advanced CAD tools. However the need to assign correct
material properties to the elements of these models poses a
much greater challenge since biologic materials are
anisotropic, inhomogeneous and graded in nature. Thus it
is possible to create a model of a bone or tooth, and then
experimentally test the same bone under conditions iden-
tical to those simulated in the numerical analysis, mea-
suring displacements with optical metrology. It is then
possible to compare the analysis results to the experimental
results. The model can be modified and fine-tuned up to the
point when the predicted and experimental results are
close, indicating by reverse engineering the correct spatial
distribution of the material properties within the bone or
tooth. Several simple examples of this approach were
provided above ([25, 36]), but the potential is far from
tapped.
Optical whole-field metrology methods can also be
applied to good advantage in studying questions arising in
clinical settings. For example, dentists are not certain of the
effect carries, or various restoration materials and methods
used to treat them, have on the mechanical behavior of
whole teeth. To answer such questions, we are currently
testing intact premolar teeth under load with ESPI. We first
123
8932
measure the deformation maps occurring as a result of load
applied to intact teeth. We then create carries-like lesions in
their crowns and retest them. We then use standard filling
materials and techniques and test them a third time (Barak
et al., submitted). It should be stressed that such an endeavor
requires a system which is able to accurately and non-
destructively measure whole-field displacements of objects
immersed in water. ESPI and other optical metrology
methods are therefore ideally suited to achieve these goals.
Other biologically intriguing questions involve load
transfer between bones through joints and between teeth
and their sockets through the periodontal ligament. Optic
metrology methods could supply detailed answers to
questions of both basic and clinical nature. Such investi-
gations are particularly difficult because of the strongly
viscoelastic nature of the soft tissues involved in load
transfer.
An area which has received much attention in recent
years is the phenomenon of microcrack propagation in
bone, up to fracture, and in particular the as yet unresolved
question of how cortical bone is able to withstand fracture,
either after long-term cyclic loading, or during trauma. The
answer to this question has far-reaching implications to
human health, since failure of protective mechanisms is
relatively common, especially in young active athletes and
in the aging population. It has become increasingly clear
that the fracture process in bone will have to be modeled
using non-linear fracture mechanics models, rather than
linear elastic ones, because in bone the size of the process
zone is very large compared with the size of the crack.
However, direct experimental evidence for these models is
lacking. This need is particularly suited to the use of ultra-
fast cameras and digital HI which will enable dynamic
measurement of the deformation and strain fields associ-
ated with crack propagation near the tip of the crack, in an
area in which linear elastic predictions fail.
Conclusions
Optical metrology methods make it possible to study and
obtain data on the mechanical behavior of entire structures
such as whole bones and whole teeth. These data combined
with numerical modeling could contribute significantly to
our understanding of the challenging questions concerning
structure-function relations in these important organs.
Acknowledgements The authors wish to thank Dr. Paul Zaslansky
and Dr. Meir Barak for helpful suggestions and discussions. S. W. is
the incumbent of the Dr. Walter and Dr. Trude Burchardt Professorial
Chair of Structural Biology. Support for this research was provided
from grant RO1 DE006954 from the National Institute of Dental and
Craniofacial Research to Dr. Stephen Weiner, Weizmann Institute of
Science.
123
J Mater Sci (2007) 42:89198933
References
1. Thompson D (1942) On growth and form. Cambridge University
Press, Cambridge
2. Mayer G (2005) Science 310:1144
3. Gao HJ (2006) Int J Frac 138:101
4. Weiner S, Wagner HD (1998) Ann Rev Mater Sci 28:271
5. Nanci A (2003) Ten Cates oral histology. Mosby, Inc., St Louis,
Missouri, USA
6. Delmas PD, Tracy RP, Riggs BL, Mann KG (1984) Calcif Tissue
Int 36:308
7. Wang HJ, Tannukit S, Zhu DH, Snead ML, Paine ML (2005)
J Bone Miner Res 20:1032
8. Suresh S (2001) Science 292:2447
9. Hodge AJ, Petruska JA (1963) In: Ramachandran GN (ed)
Aspects of protein structure. Academic Press, New York, p 289
10. Fratzl P, Schreiber S, Boyde A (1996) Calcif Tissue Int 58:341
11. Weiner S, Traub W, Wagner HD (1999) J Struct Biol 126:241
12. Currey J (2002) Bones. Princeton University Press, Princeton
13. Daculsi G, Menanteau J, Kerebel LM, Mitre D (1984) Calcif
Tissue Int 36:550
14. Houlle P, Voegel JC, Schultz P, Steuer P, Cuisinier FJG (1997)
J Dent Res 76:895
15. Weiner S, Veis A, Beniash E, Arad T, Dillon JW, Sabsay B,
Siddiqui F (1999) J Struct Biol 126:27
16. Scott DB, Simmelin JW, Nygaard V (1974) J Dent Res 53:165
17. Zaslansky P, Friesem AA, Weiner S (2006) J Struct Biol 153:188
18. Wang R, Weiner S (1998) J Biomech 31:135
19. Zaslansky P, Shahar R, Friesem AA, weiner S (2006) Adv Funct
Mater 16:1925
20. Tesch W, Eidelman N, Roschger P, Goldenberg F, Klaushofer K,
Fratzl P (2001) Calc Tissue Int 69:147
21. Wang RZ, Weiner S (1998) Connect Tissue Res 39:269
22. Cuy JL, Mann AB, Livi KJ, Teaford MF, Weihs TP (2002) Arch
Oral Biol 47:281
23. Rho JY, Pharr GM (1997) Biomaterials 18:1325
24. Gasvik KJ (2002) Optical metrology. John Wiley & Sons, LTD,
West Sussex, England
25. Murphy LA, Prendergast PJ (2005) J Biomech 38:1702
26. Johnson EW, Castaldi CR, Gau DJ, Wysocki GP (1968) J Dent
Res 47:548
27. Asundi A, Kishen A (2000) Proc Inst Mech Eng [H] 214:659
28. Asundi A, Kishen A (2000) Arch Oral Biol 45:543
29. Asundi A, Kishen A (2001) J Biomed Opt 6:224
30. Asundi A, Kishen A (1999) Endod Dent Traumatol 15:83
31. Kishen A, Asundi A (2005) J Biomed Opt 10:034010
32. Kishen A, Ramamurty U, Asundi A (2000) J Biomed Mater Res
51:650
33. Kishen A, Asundi A (2002) J Biomed Opt 7:262
34. Peindl RD, Harrow ME, Connor PM, Banks DM, Dalessandro
DF (2004) Exp Mech 44:228
35. Meyer C, Kahn JL, Boutemi P, Wilk A (2002) J Craniomaxillofac
Surg 30:160
36. Huo B (2005) J Biomech 38:587
37. Wood JD, Wang RZ, Weiner S, Pashley DH (2003) Dent Mater
19:159
38. Kishen A, Asundi A (2005) J Biomed Mater Res A 73:192
39. Kishen A, Tan KB, Asundi A (2006) J Dent 34:12
40. Pedrini G, Osten W, Gusev ME (2006) Appl Opt 45:3456
41. Katz DM, Blatcher S, Shelton JC (1998) Med Eng Phys 20:114
42. Kozuchi J, Taniguchi M, Chubachi N, Takagi T (2002) Appli-
cation of holographic interferometry measuring techniques to
deformation measurement of bone due to thermal stress. IEEE
instrumentation and measurement technology conference,
Anchorage, AK, USA
J Mater Sci (2007) 42:89198933
43. Alexeenko W, Pedrini G, Zaslansky P, Kuzmina E, Osten W,
Weiner S (2004) Digital holographic interferometry for the
investigation of the elastic properties of bone. 12th international 55.
conference on experimental mechanics, Bari, Italy 56.
44. Pedrini G, Alexeenko IV, Zaslansky P, Tiziani HJ, Osten W 57.
(2005) Proc SPIEThe Int Soc Opt Eng 5776:325
45. Jones R, Wykes C (1989) Holographic and speckle interferom- 58.
etry. Cambridge University Press, Cambridge
46. Vest CM (1979) Holographic interferometry. Wiley, New York 59.
47. Schmidt T, Tyson J, Galanulis K (2003) Exp Tech 27:47
48. Zaslansky P, Currey JD, Friesem AA, Weiner S (2005) J Biomed 60.
Opt 10:024020
49. Shahar R, Zaslansky P, Barak M, Friesem AA, Currey JD, 61.
Weiner S (2006) J Biomech 40:252
50. Kirkpatrick SJ, Brooks BW (1998) J Biomed Mater Res 39:373 62.
51. Rodriguez D, Moreno V, Gallas M, Abeleira MT, Suarez D
(2004) Med Eng Phys 26:371 63.
52. Samala PR, Su M, Liu S, Jiang HH, Yokota H, Yang LX (2005)
Strain measurement of a mouse bone by 3D-electronic speckle 64.
pattern interferometry (3d_ESPI). SPIE, Bellingham, WA
53. Su M, Samala PR, Jiang HH, Liu S, Yang L (2005) J Hologr
Speckle 2:34 65.
54. Mohr M, Simon U, Claes L, Nbottlang M (2006) Full-field strain
acquisition on ovine fracture callus with electronic speckle pat-
8933
tern interferometry. 5th world congress of biomechanics, Munich,
Germany
Schmidt T, Tyson J, Galanulis K (2003) Exp Tech 27:22
Zhang D, Arola DD (2004) J Biomed Opt 9:691
Nicolella DP, Nicholls AE, Lankford J, Davy DT (2001) J Bio-
mech 34:135
Nicolella DP, Lankford J (2002) J Musculoskelet Neuronal
Interact 2:261
Nicolella DP, Bonewald LF, Moravits DE, Lankford J (2005) Eur
J Morphol 42:23
Nicolella DP, Moravits DE, Gale AM, Bonewald LF, Lankford J
(2006) J Biomech 39:1735
Koca OL, Eskitascioglu G, Usumez A (2005) J Prosthet Dent
93:38
Kuijs RH, Fennis WMM, Kreulen CM, Barink M, Verdonschot N
(2003) J Dent Res 82:967
Simon U, Augat P, Ignatius A, Claes L (2003) J Biomech
36:1079
Jaecques SVN, Van Oosterwyck H, Muraru L, Van Cleynenb-
reugel T, De Smet E, Wevers M, Naert I, Van der Sloten J (2004)
Biomaterials 25:1683
Taylor WR, Roland E, Ploeg H, Hertig D, Klabunde R, Warner
MD, Hobatho MC, Rakotomanana L, Clift SE (2002) J Biomech
35:767
123
J Mater Sci (2007) 42:89348942
DOI 10.1007/s10853-007-1648-0
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Synergistic effects of metals and oxidants in the curing of marine
mussel adhesive
Lauren M. Hight Jonathan J. Wilker
Received: 16 December 2006 / Accepted: 1 March 2007 / Published online: 7 August 2007

Springer Science+Business Media, LLC 2007
Abstract Marine mussels produce an impressive adhe-
sive material for affixing themselves to rocks in the tur-
bulent marine environment. This glue is generated by
application of proteins to the surface followed by extensive
cross-linking to yield the final matrix. Prior studies have
shown that simple oxidation or reactivity brought about by
metal ions may be key to this protein cross-linking process.
Here we have explored protein cross-linking reactivity in
which combinations of metals and oxidants may display
synergistic effects with respect to adhesive curing.
Extracted adhesive proteins were mixed with a series of
metals, oxidants, and combinations thereof. In some cases,
synergistic curing was observed. For example, we found
that iron(II) ions with hydrogen peroxide brought about a
greater degree of protein cross-linking than the sum of the
individual components. These studies were performed as
part of our efforts to provide perspectives on the connec-
tions between biology, chemistry, and functional materials.
Introduction
The oceans abound with a fascinating collection of unique
materials, ranging from barnacle, tube worm, and oyster
cements to the adhesives of sea stars, mussels, and corals.
We cannot help but marvel at how these organisms design
and generate such functional materials for affixing them-
selves to underwater surfaces. This material development
L. M. Hight
J. J. Wilker (&)
Department of Chemistry, Purdue University, West Lafayette,
IN 47907-2084, USA
e-mail: wilker@purdue.edu
123
must account for performance within the turbulent and
harsh marine environment. We are curious to see what
links exist between small-scale chemical bonds and the
resulting macroscopic materials.
Perhaps the two most well studied marine biological
materials are barnacle cement [16] and mussel adhesive
[710]. In both cases, the material is produced by the
animal depositing a mixture of proteins onto a surface.
Cross-linking of the proteins then yields the final, cured
matrix. The exact details of the cross-linking process and
surface adhesion remain to be determined. In the case of
barnacles, the proteins contain cysteine thiols that may
become oxidized to disulfides for cross-linking [16].
Mussel adhesive proteins contain the unusual amino acid
3,4-dihydroxyphenylalanine (DOPA) [710]. These DOPA
residues are essential for cross-linking and can comprise
high levels (~1022%) of the total protein amino acid
content. Also present in mussel adhesive at remarkably
high levels are transition metals such as iron and zinc
[1116].
The majority of studies related to cross-linking of
mussel adhesive proteins have focused upon oxidation
reactions. Oxidants such as NaIO4 [1719] and the oxi-
dizing tyrosinase enzyme [1821] readily react with
DOPA-containing proteins, peptides, and synthetic poly-
mers to yield cross-linked products. A recent atomic force
microscopy study indicates that DOPA oxidation to the
quinone may be responsible for curing the bulk material
[22]. Surface adhesive bonding, by contrast, may depend
upon the reduced (i.e., not oxidized) catechol sidechain
of DOPA interacting with the animals chosen substrate
[22].
Our laboratory [2328] and others [17, 18, 2932] have
been exploring potential roles of metals in the formation of
mussel adhesive. Metal ions such as Fe3+ have the ability to
J Mater Sci (2007) 42:89348942
both become chelated by the DOPA sidechain [25] as well
as bring about DOPA oxidation [7, 33, 34]. From spec-
troscopic and reactivity studies, we found that the iron in
mussel adhesive may be bound by three DOPA-containing
protein strands [26]. Furthermore, these Fe(DOPA)3 cen-
ters can react with oxygen to generate protein-based radi-
cals [24, 26]. Such oxidized protein may go on to afford
proteinprotein coupling for curing or protein-surface
bonding for direct adhesion.
We wondered how such chemical and bonding pro-
cesses may contribute to the actual generation and perfor-
mance of mussel adhesive. In particular, we were curious
to see how well metals or oxidants might be able to bring
about cross-linking of mussel adhesive protein. Adhesive
precursor proteins were extracted from mussels prior to
cross-linking [27, 28]. After precipitation from aqueous
solutions by addition of an organic solvent and centrifu-
gation, a slightly viscous protein hydrogel was obtained.
We reacted this gel with a collection of potential cross-
linking agents [27, 28]. The various resulting matrices were
examined for materials properties employing a direct
measure of curing [27, 28]. Using an Instron materials
testing system, a rod was moved into the gel at constant
velocity, all the while recording force (c.f., Fig. 1). This
measurement of compressibility and shear properties
allows rapid analysis of many samples and provides
quantitative comparisons (c.f., Table 1) [27, 28]. We have
recently also reported on the rheological characteristics of
these hydrogels both before and after cross-linking [23].
We reacted the protein hydrogel with a collection of
metals and oxidants [27, 28]. In general, we found that
Fig. 1 Photograph of the experimental setup used for direct
measurements of curing mussel adhesive extract with various
reagents. A rod was run into the protein hydrogel at constant velocity
while force was recorded. The sample tube in the photograph is raised
for visibility
8935
simple oxidants (e.g., H2O2, Na2S2O8) cross-linked the
protein, but only to a limited degree [27]. The most
effective reagents were oxidizing metal ions such as Fe3+,
Cr6+ (in Cr2 O2
7 ), and Mn
7+
(in MnO 4 ) [27]. By contrast,
+ 2+ 2+
simple metal ions (e.g., Na , Co , Ni ) did not afford any
curing. For oxidizing reagents, no correlation was found
between curing and reduction potentials. In general, con-
centration dependences were found, with higher concen-
trations of a given reagent generally bringing about more
curing. Of the reagents examined that may typically be
available to mussels, Fe3+ provided the most pronounced
cross-linking [28]. Given the high concentrations of iron in
mussel adhesive plaques [1116], we concluded that iron
may be the reagent responsible for bringing together DOPA
proteins to form this material [26, 28].
In general, metal ions may react with oxidants and
reductants to generate reactive species. Perhaps the most
well known cases are from metalloenzymes. For example,
in the cytochrome P450cam system, a metal ion (iron in a
heme) reacts with an oxidant (O2) and a reductant (elec-
trons) to generate a reactive intermediate capable of oxi-
dizing an unactivated hydrocarbon (camphor) to a product
alcohol [35]. Small molecule systems are also well studied
in which a metal complex (e.g., Fe EDTA) can react with
an oxidant (e.g., H2O2) to produce reactive oxygen species
(e.g., hydroxyl radical) [36, 37]. Thus metals and oxidants
can react in synergistic ways to bring about chemistry
unavailable to either component alone. Perhaps both metals
and oxidants play a role in the curing of mussel adhesive.
We found oxidized protein, in the form of an organic
radical, within fresh adhesive plaques harvested from
mussels [26]. Earlier studies showed the presence of oxi-
dase activity in the plaques [37]. Thus oxidation reactions
are clearly part of the protein curing process. However, the
active oxidant generated by mussels for this cross-linking
has not yet been identified. The usual suspects list of
common biological oxidants include oxidizing enzymes
(e.g., tyrosinase) and the most common small molecule
oxidizers found in biology such as dioxygen (O2), hydro-
gen peroxide (H2O2), superoxide (O 2 ), and hydroxyl rad-


icals (OH ) [38, 39]. Generally speaking, most protein
cross-linking processes require the presence of oxidants
[33]. With regard to mussel proteins, nearly all cross-
linking studies have focused on oxidation, with periodate
(IO4 ) [1719] and the tyrosinase enzyme [1821] com-
prising the majority of work.
Another interesting issue related to oxidation within this
biological material is that of metal ion transport. Prior to
internalization by organisms, the majority of environmental
iron is in the +3 oxidation state [4042]. In open ocean
water, for example, nearly all iron is particulate and typi-
cally in various forms of Fe3+ [4042]. Mussels collect this
insoluble iron in their bivalve filtration system [1214].
123
8936
The process by which these animals transport this accu-
mulated iron to their adhesive system has not been de-
scribed in detail. Generally speaking, divalent Fe2+ is more
soluble and easier to transport than trivalent Fe3+ [43].
Thus most organisms accumulate Fe3+ and reduce the ion
down to Fe2+ for actual use [43]. Given our results indi-
cating that Fe3+, rather than Fe2+, appears to be required for
mussel protein cross-linking [27, 28], we can envision a
scenario in which mussels accumulate Fe3+ in their filters,
reduce the metal to Fe2+ for transport to the byssal adhesive
system, then reoxidize to Fe3+ for initiating the curing. If
such processes are taking place, an oxidant is required for
the final Fe2 ! Fe3 transformation. Here we report on
the ability of oxidants to influence the curing of DOPA
proteins by metal ions. At this time, we can only speculate
as to which oxidants might be pertinent to mussel adhesive.
Thus we have focused on two: hydrogen peroxide (H2O2)
and sodium periodate (NaIO4). Hydrogen peroxide is a
thermodynamically strong (E = 1.78 V H2O2/H2O) [44]
but kinetically slow oxidant found in many biological
systems. When examining the oxidation of biological
molecules, H2O2 makes for an obvious choice. Our prior
studies showed that H2O2 brings about only modest curing
of mussel adhesive proteins [27, 28]. We chose NaIO4 for a
contrast to H2O2. Previously we found that this strong
oxidant (E = 1.59 V IO4 /IO3 ) [45] cured mussel
adhesive proteins appreciably, to a similar degree as Fe3+
[27]. This reagent has also been popular with in vitro
studies of mussel adhesive protein cross-linking [1719].
As will be seen below, synergistic curing effects of metals
and oxidants were found. These results may provide in-
sights on how mussels use chemical reactivity to generate
their adhesive material.
Experimental
Protein extractions
Mussel adhesive protein was extracted from the animals
according to a literature method [46]. This extraction
method relies upon acid precipitation of most biomolecules
while leaving the DOPA-containing proteins in solution
[46]. Subsequent removal of the proteins from solution,
using an organic solvent precipitant, yields a gel containing
two of the protein variants, with a relative composition of
~80% Mefp-1 and ~20% Mefp-2 [46]. After precipitation
of these DOPA proteins from solution with acetone, cen-
trifugation yielded hydrogel pellets [46]. The pellets were
stored in a 4 C cold box under a blanket of deionized
water until used, within 7 days of extraction. Roughly 85
pellets yielded enough sample to complete one set of
penetration tests. The pellets were placed in a plastic
123
J Mater Sci (2007) 42:89348942
strainer with 3.5 mm holes. Using a ceramic container with
a diameter slightly smaller than the strainer, the pellets
were forced through the strainer for homogenization, while
chilled with ice. Deionized water was added to the strained
pellets by the following steps: strained pellets were trans-
ferred to a tared 100 mL beaker, mass was recorded and
multiplied by 1/3. This resulting value was the amount of
deionized water added in grams to the beaker of strained
pellets such that the final amount of water in the mixture
was 25% w/w. The strained pellets and water were thor-
oughly mixed with a spatula to obtain a homogenous
mixture. This mixture was then transferred in 1011 g
increments to a 10 mL plastic syringe with a 2 mm open-
ing used to deposit the homogenous material into plastic
2 mL microcentrifuge tubes (9 mm in diameter, 35 mm
deep). During the transfer process, the mixture remaining
in the beaker was kept on ice and covered with Parafilm
until placed in the syringe. The tube was capped and tapped
on the bench 67 times to eliminate air bubbles and then
placed on ice until all of the mixture was dispensed. The
sample tubes were covered with aluminum foil and stored
in a 4 C cold box until used, within 7 days of extraction.
Preparation of cross-linking reagent solutions
For all studies discussed below, the final concentration of
cross-linkers were 45 mM. This concentration was chosen
to be consistent with our prior studies [27, 28] as well as
close to that of the natural abundance in mussel adhesive.
Total iron in mussel adhesive plaques is approximately one
part per thousand [11, 12] or about 20 mM when converted
to solution concentrations. To obtain a final cross-linker
concentration of 45 mM in the gel, 1 M solutions of the
desired reagents were prepared fresh daily. In previous
work, 0.5 M solutions were used [27, 28], but experiments
discussed below require a 1 M solution to compensate for
the dilution that will occur upon addition of two reagents
(metal and oxidant) instead of only one. To prepare solu-
tions, the calculated amount of reagent required for a 1 M
solution in 5 mL volume was weighed out and placed in a
10 mL volumetric flask. After placing all compounds in
individual flasks, 5 mL of deionized water was added and
the flask vortexed to aid dissolution. For solutions that did
not completely dissolve with mixing alone, mild heat was
applied by hotplate for 710 min. The heated solutions
included NaIO4, ZnCl2, Mn(OOCCH3)3, CuCl, and TiF3.
All solutions were transferred to 15 45 mm screw thread
glass vials and capped to prevent any evaporation. For
reagents where solubility did not allow a 1 M concentra-
tion, saturated solutions were used instead. The saturated
solutions included NaIO4, Mn(OOCCH3)3, CuCl, and TiF3.
All water used in this study was purified to at least 18 MW/
cm using a Barnstead Nanopure Infinity system. In our
J Mater Sci (2007) 42:89348942 8937

prior report on cross-linking hydrogel extracts, we found 280

the chloride and nitrate salts of a given metal ion yielded 240
2+
identical cross-linking results [27]. Thus only one salt of a Cu +
200 NaIO4
given ion was examined here.

force (mN)
160
NaIO
Cross-linking reactions 4
120 2+
Cu +
80 H2O2
Into each sample tube was dispensed 1.001.05 g of the 2+
40 Cu
homogenized adhesive protein extract. Then 50 lL of a HO
1 M non-metal oxidant solution was added followed by a 0 water 2 2
50 lL solution of a 1 M metal salt. The reagents were 0 5 10 15 20 25
added to the sample tube by 100 lL glass syringes. For rod extension (mm)
controls, 100 lL of deionized water was added to the tube. Fig. 2 Extension versus force data for a rod penetrating into a protein
When adding reagent combinations, the non-metal oxidant hydrogel extracted from marine mussels. The protein gel was reacted
was consistently added first. In the case of combining a with water, Cu2+, H2O2, NaIO4, Cu2+ + H2O2, and Cu2+ + NaIO4.
metal and water, the water was added first. Each sample Higher force indicates greater cross-linking of the protein. Each trace
is averaged from multiple runs. Error bars show one standard
tube was mixed thoroughly with a microspatula for 56 s. deviation for a given data point
The spatula was then scraped on the inside rim of the tube
to return any mixture that stuck back into the sample tube.
Each tube was capped and tapped on the bench 67 times 600
to eliminate air bubbles. The sample was then allowed to 2+
react at room temperature for one hour. 500 Fe +
NaIO4
400
force (mN)

Direct measurements of curing 2+

300 Fe +
H2O2
A penetration test [4751] was used to measure the effect 200 NaIO
of various reagent combinations on cross-linking. A 5 mm 4

diameter rod (blank drill bit) was run into the gel mixtures 100 HO
2 2
2+
at constant velocity, all the while measuring force on the 0 Fe
rod. Previously we used a 3.5 mm rod, thus data in this 0 5 10 15 20 25
current study yields higher penetration force values [27, rod extension (mm)
28]. The theory behind this mechanical test predicts that
Fig. 3 Penetration plots for the mussel adhesive protein gel reacted
linear relationships will be found between penetration with Fe2+, H2O2, NaIO4, and combinations thereof. Each trace is
depth and force [48, 49]. Some deviations from this averaged from multiple runs and error bars are shown with one
expected linearity will be seen below and are a result of standard deviation
inhomogeneity in the samples resulting from cross-linking.
The starting protein hydrogel is homogeneous, as are many
of the cross-linked products. Extreme levels of cross-link- 700
ing, however, can yield hardened solids suspended in 600 3+
solutions. Thus the materials no longer behave ideally. We Fe +
500 NaIO4
discussed this phenomenon in our recent paper in which we
force (mN)

examined the rheological properties of some cross-linked 400

3+
protein gels [23]. 300 Fe +
Figure 1 shows the experimental setup used and the H2 O 2
200
Instron 5544 materials testing system fitted with a 2 N load NaIO
cell. Protein gel sample remained in the 2 mL plastic tubes 100 Fe
3+ 4
HO
and were secured in a chuck grip fastened to the base of the 0 2 2

instrument. Similarly, the 5 mm rod was held in place using 0 5 10 15 20 25

a chuck grip attached to the load cell. Prior to collecting
data, the rod was positioned ~34 mm centered above the
extract. Each sample was penetrated to a depth of 20 mm at
a rate of 20 mm/min. Load data were collected every
0.5 mm. Typical data can be seen below, in Figs. 24.
rod extension (mm)
Fig. 4 Penetration plots for the mussel adhesive protein gel reacted
with Fe3+, H2O2, NaIO4, and combinations thereof. Each trace is
averaged from multiple runs and shows error bars of one standard
deviation for each point

123
8938 J Mater Sci (2007) 42:89348942

Table 1 Penetration forces in mN observed for an extract of mussel adhesive proteins mixed with various combinations of reagents
Reagent Water n H2O2 n NaIO4 n

Water 30 12 32 34 20 10 144 57 10
NaCl 31 16 5 19 2 4 113 54 5
CaCl2 22 9 5 38 1 4 126 38 5
ZnCl2 27 12 5 42 12 5 180 42 5
AlCl3 22 15 5 29 9 5 146 51 5
Ga(NO3)3 23 10 6 35 10 5 322 89 5
CoCl2 28 16 5 24 8 5 111 4 4
TiF3 24 9 5 27 9 5 211 38 5
CuCl 30 11 5 32 6 5 221 66 5
CuCl2 29 9 5 65 22 5 215 61 5
VCl3 24 10 6 77 13 4 392 79 4
Na3VO4 1201 638 5 741 264 5 1201 619 5
Na2Cr2O7 1563 664 5 630 550 5 1859 983 6
MnCl2 26 13 6 41 12 5 626 192 5
Mn(OOCCH3)3 25 9 5 23 7 5 147 42 5
KMnO4 887 319 6 490 251 5 1646 665 6
FeCl2 28 7 5 222 25 4 478 83 4
FeCl3 134 47 5 213 63 5 560 137 5
Each value is an average of n runs and one standard deviation is shown

Reported maximum loads were taken at a final penetration
depth of 20 mm. For a given experimental run on a given
day, every potential cross-linking reagent or combination of
Table 1 was examined. The final force values shown for a
given entry in Table 1, below, were averaged from five
different runs each performed on five different days. Out-
lying data points were removed if they failed a Q test at 90%
confidence level. Reported errors denote one standard
deviation. The number of samples used for each data point
is provided in Table 1. When comparing the averaged data
in the Discussion section, standard t-tests were per-
formed on the raw data, thereby yielding probabilities that a
given pair of samples showed the same curing.
Results
Curing data
In order to provide a context for the data obtained, we start
the discussion with controls. Water added to the adhesive
extract showed a penetration force of 30 12 mN.
Hydrogen peroxide, alone, did not exhibit curing, with a
force of 34 20 mN. On its own, cross-linking was found
for NaIO4 at 144 57 mN. Figure 2 shows a penetration
test with water, H2O2, and NaIO4 each added to separate
extracts. As can be seen, there is no appreciable difference
between the water control and H2O2. Thus Figs. 3 and 4
will not show water traces in the interest of clarity. Below
we provide the effects of cross-linking with each of these
oxidants in combination with various metal ions. All curing
data are shown in Table 1. We will present the data starting
with low levels of curing and progress to higher degrees of
cross-linking.
Non-redox active metal ions
For the Na+ ion, from NaCl, no difference was observed for
protein cross-linking (31 16 mN) versus the water con-
trol (30 12 mN), as is seen in Table 1. Addition of H2O2
provided no increase (19 2 mN) and NaIO4 was higher
(113 54 mN) but no more so than NaIO4 alone
(144 57 mN). Similar results were found for CaCl2 and
ZnCl2. Aluminum(III) is often used, along with Ga3+, to
mimic the coordination chemistry of Fe3+ but without the
available redox chemistry. Aluminum(III) alone
(22 15 mN) or with H2O2 (29 9 mN) showed no cur-
ing nor did Al3+ with NaIO4 (146 51 mN) beyond that of
the individual components. Gallium(III), however,
appeared different. Alone Ga3+ (23 10 mN) or with
H2O2 (35 10 mN) exhibited no notable effect. When
combined with NaIO4, hardening from Ga3+
(322 89 mN) was near double that of the individual
components summed together (144 + 23 = 167 mN).

123
J Mater Sci (2007) 42:89348942
Redox active metal ions
Cobalt has redox activity available, with the Co2+ and Co3+
ions most common. Results found for Co2+ were similar to
that of Na+, with Co2+ alone (28 16 mN), mixed with
H2O2 (24 8 mN), or NaIO4 (111 4 mN) no greater
than any of the controls. Titanium also has multiple oxi-
dation states available (e.g., 2+, 3+, 4+). For this study we
chose a Ti3+ salt based upon prior experience showing that
TiF3 is easiest to work with in water, minimizing imme-
diate formation of TiO2 powder [52]. Titanium(III) alone
(24 9 mN) or with H2O2 (27 9 mN) did not cure the
adhesive extract to any significant degree. With both Ti3+
and NaIO4 the observed curing was high (211 38 mN)
although still close to the sum of the individual components
(24 + 144 = 168 mN).
Given that the copper-containing tyrosinase enzymes
can bring about oxidation of substrates such as tyrosine,
DOPA, phenols, catechol [53], we were curious to see
about potential cross-linking from copper in conjunction
with oxidants. Inorganic copper can also catalyze the oxi-
dation of small molecules such as catechol [54]. The Cu+
salt CuCl provided data similar to the TiF3 case-little
curing alone (30 11 mN) or with H2O2 (32 6 mN).
Periodate addition may have provided extra curing
(221 66 mN) relative to each reagent on their own
(30 + 144 = 174 mN), although the difference was not
great. Curing from the Cu2+ ion alone (29 9 mN) was
insignificant and when combined with NaIO4 (215
61 mN) was high, but not conspicuously so. For Cu2+ and
H2O2 together (65 22 mN), the observed penetration
force was no greater than the sum of the components
(29 + 34 = 63), but was higher than any of the reagents
mentioned so far with added H2O2. These data can be seen
in both Fig. 2 and Table 1.
Vanadium(III), similarly, did not cure alone
(24 10 mN) or with H2O2 (77 13 mN) much beyond
the sum of each separate reagent (24 + 34 = 58 mN). A
mild synergistic effect was observed with V3+ and NaIO4
at 392 79 mN. On its own, Na3VO4 cured the protein
extract substantially (1201 638 mN), consistent with
prior observations of V5+ being a potent cross-linker [27].
Periodate did not contribute to curing (1201 619) nor
did H2O2 (741 264). We also reported previously that
Cr6+, in the dichromate form Cr2 O2 7 , is one of the
strongest curing agents we have found to date
(1563 664 mN) [27]. Neither H2O2 (630 550 mN)
nor NaIO4 (1859 983 mN) enhanced the Cr6+ reactivity
significantly. Indeed, like what was found with V5+, the
metal with H2O2 actually showed less curing than the
sum of the two parts. However, such decreases may be a
result of limitations in the methods employed here (vide
infra).
8939
We examined the cross-linking afforded by three man-
ganese ions, Mn2+, Mn3+, and Mn7+ (Table 1). Starting
with Mn2+ (26 13 mN) and H2O2 combined
(41 12 mN) no detectable effect was noted. Pronounced
synergistic curing, however, was seen with both Mn2+ and
NaIO4 together (626 192 mN). By contrast, Mn3+
(25 9 mN) did not cure any more effectively when H2O2
(23 7 mN) or NaIO4 (147 42 mN) were added, rela-
tive to the oxidant-only solutions. As shown in Table 1,
Mn7+ brought about dramatic curing (887 319 mN). No
enhancement was found with H2O2 (490 251 mN) but a
doubling was seen with Mn7+ and NaIO4 (1646
665 mN).
Spectroscopic studies from our laboratory have impli-
cated Fe3+ in the curing of mussel adhesive [26]. Thus we
were particularly interested to explore possible influences
of oxidation upon iron-induced cross-linking of adhesive
proteins. Although Fe2+ did not yield any curing alone
(28 7 mN), in combination with H2O2 much greater
penetration forces were required (222 25 mN) as can be
seen in Fig. 3. Further enhancement of curing was found
with Fe2+ and NaIO4 (478 83 mN). Alone, Fe3+ cured
the protein significantly (134 47 mN) and this effect
increased to a small degree when combined with H2O2
(213 63 mN). Both Fe3+ and IO 4 , together, showed
synergistic cross-linking (560 137 mN) greater than that
with Fe3+ and H2O2, presented in Fig. 4. Thus oxidants do
enhance the curing ability of iron. Also worth noting is
that, of all the metal ions examined here, only Fe2+ dis-
played a dramatic enhancement of curing with H2O2
(Fig. 3; Table 1).
Visual observations of curing
The starting gel had a light tan color, a mostly thin and
liquidy consistency, with no clumping. To make a food
analogy, the material began somewhat like apple sauce. We
have reported on the rheological properties of this protein
extract, both before and after cross-linking [23]. In general,
if the numbers of Table 1 indicate little or no curing (e.g.,
~80 mN and less), no changes in color or consistency were
found except for added color in the following cases:
CuCl2 + H2O2 (light brown), CoCl2 + H2O2 (slight pink),
VCl3 + H2O2 (grey), and Mn3+ + NaIO4 (rose). Hydrogen
peroxide brought about no observable changes on its own.
For NaIO4, alone, the protein gel took on a pinkish-brown
color with no conspicuous change in consistency. Like-
wise, the metal ions and NaIO4 with curing under
~220 mN and lower, along with Ga3 IO 4 , were pink-
ish-brown. Reactions with color changes, but without ma-
jor viscosity increases were FeCl3 (black), FeCl3 + H2O2
(dark brown), FeCl2 + H2O2 (dark brown), and TiF3 +
NaIO4 (reddish brown). The latter two samples separated
123
8940
somewhat, with a darker color solid residing below a
lighter colored solution above.
The powerful oxidant Na2Cr2O7, alone or with H2O2,
generated a yellow-tan extremely viscous gel that stuck to
the spatulas when handled. Addition of NaIO4 formed an
orange-tan solid with a yellow solution above. For KMnO4,
alone or with H2O2, a colorless solution was found to
reside above a black gel containing tan flecks. With
KMnO4 and NaIO4, the clear solution was present above a
purple, then red solid. In all cases, the increase in viscosity
was immediate and extreme. With Na3VO4, alone, also a
strong oxidant and cross-linker, a brown and then dark
green, very viscous and sticky gel was noted with a clear
solution above. Similar results were found with Na3-
VO4 + NaIO4 and VCl3 + NaIO4. A combination Na3VO4
and H2O2 showed a pink to yellow to green to brown to
yellowbrown transformation within 20 s, also producing a
viscous gel with a light yellow solution above. Combined
with NaIO4, MnCl2 showed a substantial increase in
viscosity, a red-brown gel, and a colorless solution above.
Both FeCl2 and FeCl3 mixed with NaIO4 yielded viscous
yellowbrownred gels.
Discussion
Comparisons with prior data
The general trends of curing reported here agree with our
earlier data [27, 28]. For example, H2O2 cross-linking is
minimal and that of IO 4 and Fe
3+
are appreciable. Some
specific changes, however, could be noted in that reagents
such as H2O2 showed higher force values here
(34 20 mN) than earlier (20 4 mN). These differences
can be attributed to incremental improvements we made to
our test procedure. The most conspicuous change was use
of a 5 mm penetration rod here versus a 3.5 mm rod pre-
viously. Thus the force data tend to now be higher. Other
changes in the procedure worth noting include extracting
the protein from mussel feet collected in a different season
and a slightly different method of mixing reagents into the
gel extract. Previously we massed the pellets, strained them
to homogeneity, and added water for normalization of
volumes based upon the pre-strained mass. Here we
avoided variations from loss in the strainer by using the
post-strained mass for all calculations [27].
The only significant observed variation from our pre-
vious work is that of Mn3+ alone. Here curing was not
observed for this ion, but previously it was. Earlier we
noted a similar darkening and thickening of the sample for
both Mn(OOCCH3)3 and KMnO4. As will be discussed
below, Mn2+ + NaIO4 curing may be a result of initial
oxidation to Mn7+. Beyond possible Mn7+ impurities in the
123
J Mater Sci (2007) 42:89348942
earlier starting material, at this time we do not have a
satisfactory explanation for previously observed Mn3+
cross-linking. Also earlier we noted that Ca2+ (21 1 mN)
cured slightly versus the water control (14 3). Here Ca2+
(22 9 mN) was not notably different from the water
control (30 12 mN). In both cases the differences were
small. According to a Student t-test, there is a 21.5%
probability of the Ca2+ and water control results being the
same.
Potential curing synergy between metals and oxidants
The V3+ ion with IO 4 (392 79 mN) cured significantly
more than either component alone (24 10 and
144 57 mN), thereby indicating the presence of syner-
gistic curing. Only a 1.5% probability exists that V3+ with
IO 
4 cured to the same degree as IO4 alone, according to
a Student t-test. Hydrogen peroxide did not provide an
analogous effect with V3+. By contrast, the V5+ ion did
not enhance curing when H2O2 or NaIO4 were added.
However, V5+ curing was very pronounced
(1201 638 mN). As noted above, the extreme cross-
linking yielded separation between the solution and solid,
thereby making measurements of the homogeneous gel
difficult. Thus a limit of our method may be found here.
Were enhanced cross-linking to occur by addition of an
oxidant, such reactivity may not be observable. This
separation of solid and solution also became problematic
when we measured the rheological properties of protein
gels cross-linked at the extreme limits [23]. The current
lack of observed curing enhancement with oxidant
addition to V5+ does not necessarily mean such an
enhancement does not exist.
Curing observed from VCl3 + NaIO4 may be a result of
partial V3 ! V5 oxidation brought about by the perio-
date. The resulting V5+ could then cross-link in a manner
analogous to that shown for Na3VO4 alone. Prior work
showed no reactivity from V4+. Oxidation of V3+ to V5+ is
thermodynamically uphill by ~1.3 V, based upon the sum
of reduction potentials for the VO2+/V3+ and
HV2 O 5 =HV2 O7
3
couples [44, 55]. Reduction of IO 4
occurs at a comparable potential, at 1.59 V for IO 
4 /IO3
3 5
[45], thereby indicating such V ! V oxidations are
feasible. The degree of cross-linking from V3+ and NaIO4
(392 79 mN) was less than that of V5+ alone
(1201 638 mN, same as V3+ and NaIO4 at 4.5%) indi-
cating that if such metal-based oxidation took place, the
reaction did not go to completion. Alternatively, a more
complex cross-linking mechanism, possibly incorporating
enzyme catalysis principles, could be at play. We have
tested the ability of oxidizing enzymes to cure this adhesive
protein abstract [27]. The lack of observed hardening with
enzymes may have been a result of the enzymes having
J Mater Sci (2007) 42:89348942
poor access to substrate when suspended in the viscous
gelatinous extract [27].
For the manganese ions, Mn7+ exhibited curing at the
high extreme, characteristic of oxidizing metal ions. No
additional curing was noted with H2O2 or NaIO4 although
that from Mn7+, alone, may be approaching the limit of
what is observable. Manganese(III) did not cure beyond the
appropriate control reactions. By contrast, Mn2+ provided
some interesting results. Alone (26 13 mN) or with H2O2
(41 12 mN) Mn2+ induced no cross-linking. An obvious
synergistic effect was detected when NaIO4 was added
(626 192 mN), with only a 0.7% probability of this
hardening being the same as NaIO4 alone. With no curing
for Mn3+ or Mn4+ ions, oxidation of Mn2+ to Mn3+ or Mn4+
cannot be involved. Oxidation of Mn2+ all the way up to
Mn7+, at 1.51 V, is thermodynamically possible [44] with
a strong 1.59 V (for IO 
4 /IO3 ) oxidant such as NaIO4 [45].
Thus the curing seen for Mn2+ + NaIO4 may simply be a
result of partial oxidation to Mn7+. However, a lack of
added curing for Mn3+ with NaIO4 implies that this ion was
not brought to Mn7+. A more complex interplay between
metal, oxidant, and protein may well be at work here.
Perhaps the most unexpected observation found is that
of enhanced cross-linking for Ga3+ (23 10 mN alone)
with added IO 4 (322 89 mN). This apparent enhanced
curing of Ga3+ and IO 4 has a 10.2% probability of being
the same as only IO 4 . The Ga
3+
ion is inert to redox
chemistry except under extreme conditions. Typically,
when such an ion binds to oxidizable ligands such as thi-
olates, electron density is removed from the ligand and
placed onto the metal ion, thereby making the ligand more
difficult to oxidize. Thus we might expect no major
enhancement of Ga3+-induced curing with an added oxi-
dant. A contrasting case may occur when a redox active
metal ion such as Fe3+ binds to a redox active ligand. Metal
reduction and ligand oxidation can take place. If such
oxidation of the ligand takes place within the framework
of a DOPA protein, for example, cross-linking may then
begin. At this time we have no suitable explanation for the
Ga3+ + NaIO4 result.
In terms of potential synergistic curing when both a
metal and oxidant are present, Fe2+ provides a nice
example. Figure 3 shows that the combination of FeCl2 and
H2O2 at 222 25 mN at 20 mm extension is clearly
greater than the simple sum of Fe2+ (28 7 mN) and H2O2
(34 20 mN). The likelihood of overlap between these
values is only 0.1%. Similarly Fe2+ and NaIO4 at
478 83 mN is enhanced significantly over only Fe2+ and
NaIO4 summed (28 + 144 = 172 mN), with only a 0.6%
probability of the values being coincident. These oxidants
may start by bringing about an Fe2+ Fe3+ reaction.
We know that Fe3+ is an effective cross-linker (Fig. 4;
Table 1). Indeed, the cross-linking of Fe3+ + H2O2
8941
(213 63 mN) is nearly identical to that of Fe2+ + H2O2
(222 25 mN), with a 33.6% chance of the values being
identical. Also similar, at 93.4% probability, are Fe3+ with
NaIO4 (560 137 mN) and Fe2+ with NaIO4
(478 83 mN), shown in Fig. 4. However, the high degree
of curing found with Fe3+ and NaIO4 cannot be easily
attributed to simple oxidation of the metal center. Thus a
synergistic effect of protein cross-linking appears to be
present.
Although Fe3+, alone, may be a more effective curing
agent than Fe2+, the aqueous solubility of Fe2+ is generally
greater. Iron in marine environments is readily available at
~10 parts per billion [56], predominantly existing as Fe3+
and predominantly as insoluble, particulate species
[4042]. Mussels are known to collect iron from sea water
with their bivalve filtration system, accumulating high
concentrations in their filter [12, 13]. How this particulate
iron transfers to the adhesion system is not known. If the
animal were to internalize the metal for transport, reduction
to more soluble Fe2+ may be required [57]. Our results here
show that extrusion of only this Fe2+ along with the
adhesive proteins would not bring about curing and adhe-
sive production. If mussels apply oxidants along with the
Fe2+ and protein, oxidation to Fe3+ could then result,
thereby initiating the adhesive curing process. Such a
mechanism is only speculation at this stage. However, we
[26] and others [37] have found oxidative activity in
adhesive plaques.
Conclusions
Mussels are among the rich collection of marine species
that use clever and unique chemistry for the construction of
macroscopic structures. The results shown above are pre-
sented as part of our efforts to describe the links between
biology, chemistry, and materials. Combinations of metals
and oxidants do appear to exhibit synergy with respect to
adhesive curing. We find that the effects of some metal-
induced cross-linking is enhanced with added oxidants to
degrees greater than the sum of the individual components.
Metal concentrations in mussel adhesive are quite high and,
perhaps, can generate all the cross-linking needed by the
animal. Metal-induced cross-linking by metals such as iron
requires the oxidized ion (i.e., Fe3+ and not Fe2+). However
the reduced forms are often easier to transport in biological
contexts [57]. Perhaps oxidants enhance the curing ability
of metal ions by generating sufficient concentrations of
the particular ion most needed for materials generation.
Oxidants could be in the form of small molecules such as
H2O2 or even enzymes. These results raise the possibility
that mussels may use a combination of metals and oxidants
to generate their impressive adhesive material.
123
8942 J Mater Sci (2007) 42:89348942

Acknowledgements LMH was a summer researcher in the Purdue 27. Monahan J, Wilker JJ (2004) Langmuir 20:3724
University Chemical Biology Research Experience for Undergradu- 28. Monahan J, Wilker JJ (2003) Chem Commun 1672
ates program, sponsored by the National Science Foundation. This 29. Taylor SW, Luther GW III, Waite JH (1994) Inorg Chem 33:5819
work was also supported by the Lord Corporation and the Office of 30. Taylor SW, Chase DB, Emptage MH, Nelson MJ, Waite JH
Naval Research. (1996) Inorg Chem 35:7572
31. Frank BP, Belfort G (2001) Langmuir 17:1905
32. Frank BP, Belfort G (2002) Biotechnol Prog 18:580
33. Waite JH (1992) Biol Bull 183:178
References
34. Rzepecki LM, Waite JH (1991) In: Scheuer PJ (ed) Bioorganic
marine chemistry. Springer-Verlag
1. Naldrett MJ, Kaplan DL (1997) Marine Biol 127:629 35. Denisov IG, Makris TM, Sligar SG, Schlichting I (2005) Chem
2. Naldrett MJ (1993) J Mar Biol Assoc UK 73:689 Rev 105:2253
3. Dougherty WJ (1990) J Crustacean Biol 10:469 36. Gozzo F (2001) J Mol Catal A 171:1
4. Kamino K, Inoue K, Maruyama T, Takamatsu N, Harayama S, 37. Waite JH (1985) J Mar Biol Assoc UK 65:359
Shizuri Y (2000) J Biol Chem 275:27360 38. Mishra B, Hassan PA, Priyadarsini KI, Mohan H (2005) J Phys
5. Kamino K, Odo S, Maruyama Y (1996) Biol Bull 190:403 Chem B 109:12718
6. Yamamoto H, Nagai A (1992) Marine Chem 37:131 39. Halliwell B (1990) Free Radic Res Commun 9:1
7. Deming TJ (1999) Curr Opin Chem Biol 3:100 40. Sunda WG (2001) In: Turner DR, Hunter KA (ed) The biogeo-
8. Waite JH, Andersen NH, Jewhurst S, Sun C (2005) J Adhes chemistry of iron in seawater. Wiley
81:297 41. Ussher SJ, Achterberg EP, Worsfold PJ (2004) Environ Chem
9. Waite JH (2002) Integr Comp Biol 42:1172 1:67
10. Waite JH, Lichtenegger HC, Stucky GD, Hansma P (2004) 42. Achterberg EP, Holland TW, Bowie AR, Fauzi R, Mantoura C,
Biochemistry 43:7653 Worsfold PJ (2001) Anal Chim Acta 442:1
11. Coombs TL, Keller PJ (1981) Aquat Toxicol 1:291 43. Richardson DR (2005) Curr Med Chem 12:2711
12. George SG, Pirie BJS, Coombs TL (1976) J Exp Mar Biol 23:71 44. Lide DR (1992) CRC handbook of chemistry and physics, 73rd
13. Pentreath RJ (1973) J Mar Biol Assoc UK 53:127 edn. CRC Press, Boca Raton
14. Hobden DJ (1967) J Mar Biol Assoc UK 47:597 45. Bratsch SG (1989) J Phys Chem Ref Data 18:1
15. Szefer P, Geldon J, Ahmed-Ali A, Osuna FP, Ruiz-Fernandes 46. Waite JH (1995) Methods Enzymol 258:1
AC, Galvan SRG (1998) Environ Int 24:359 47. Bourne MC (1966) J Food Sci 31:282
16. Koide M, Lee DS, Goldberg ED (1982) Estuar Coastal Shelf Sci 48. Bourne MC (1967) J Food Sci 32:601
15:679 49. Bourne MC (1975) J Texture Stud 5:459
17. Fant C, Elwing H, Hook F (2002) Biomacromolecules 3:732 50. Bourne MC (1982) In: Food texture and viscosity: concept and
18. Fant C, Sott K, Elwing H, Hook F (2000) Biofouling 16:119 measurement. New York
19. Suci PA, Geesey GG (2000) J Colloid Interface Sci 230:340 51. deMan JM (1969) J Texture Stud 1:114
20. Hansen DC, Corcoran SG, Waite JH (1998) Langmuir 14:1139 52. Sever MJ, Wilker JJ (2004) Dalton Trans 1061
21. Burzio LA, Burzio VA, Pardo L, Burzio LO (2000) Comp Bio- 53. Decker H, Tuczek F (2000) Trends Biochem Sci 25:392
chem Physiol B 126:383 54. Balla J, Kiss T, Jameson RF (1992) Inorg Chem 31:58
22. Lee H, Scherer NF, Messersmith PB (2006) Proc Natl Acad Sci 55. Bard AJ, Parsons R, Jordan J (1985) In: Standard potentials in
USA 103:12999 aqueous solution. New York
23. Loizou E, Weisser JT, Dundigalla A, Porcar L, Schmidt G, 56. Donat JR, Bruland KW (1995) In: Salbu B, Steinnes E (ed) Trace
Wilker JJ (2006) Macromol Biosci 6:711 elements in natural waters. CRC Press
24. Weisser JT, Nilges MJ, Sever MJ, Wilker JJ (2006) Inorg Chem 57. Crichton RR, Wilmet S, Legssyer R, Ward RJ (2002) J Inorg
45:7736 Biochem 91:9
25. Sever MJ, Wilker JJ (2006) Dalton Trans 813
26. Sever MJ, Weisser JT, Monahan J, Srinivasan S, Wilker JJ (2004)
Angew Chem Int Ed 43:448

123
J Mater Sci (2007) 42:89438956
DOI 10.1007/s10853-007-1649-z
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Mechanics of the hysteretic large strain behavior of mussel byssus
threads
Katia Bertoldi Mary C. Boyce
Received: 21 December 2006 / Accepted: 1 March 2007 / Published online: 7 August 2007

Springer Science+Business Media, LLC 2007
Abstract Natural fibers are particularly interesting from a
materials point of view since their morphology has been
tailored to enable a wide range of macroscopic level
functions and mechanical properties. In this paper, we
focus on mussel byssal threads which possess a morphol-
ogy specifically designed to provide a hysteretic yet resil-
ient large strain deformation behavior. X-ray diffraction
studies have shown that numerous natural fibers have a
multi-domain architecture composed of folded modules
which are linked together in series along a macromolecular
chain. This microstructure leads to a strong rate and tem-
perature dependent mechanical behavior and one which
exhibits a stretch-induced softening of the mechanical
response as a result of the underlying morphology evolv-
ing with imposed stretched. This paper addresses the
development of a constitutive model for the stressstrain
behavior of the distal portion of mussel byssal threads
based on the underlying protein network structure and its
morphology evolving with imposed stretched. The model
will be shown to capture the major features of the stress
strain behavior, including the highly nonlinear stressstrain
behavior, and its dependence on strain rate and stretch-
induced softening.
K. Bertoldi
M. C. Boyce (&)
Department of Mechanical Engineering, Massachusetts Institute
of Technology, Room 1-304, 77 Massachusetts Avenue,
Cambridge, MA 02139, USA
e-mail: mcboyce@mit.edu
K. Bertoldi
e-mail: bertoldk@mit.edu
Introduction
Mussels live in dense beds, along the surf swept rocks of
the coast. Their success in aquatic habitats is based on their
ability to adhere to the rocks using the mussel byssus,
which typically consists of 2060 byssal threads. Each
thread (having a total length of . 3 mm and a diameter of
. 0.1 mm) consists of three regions, which differ in both
morphology and mechanical properties [14] (see detail in
Fig. 1): the adhesive plaque, the distal region and the
proximal region. The proximal region, which represents
one third of the total fiber length, consists of a crimped
sheath which covers loosely packed coiled fibrils, whereas
the smoother and narrower distal region is formed by
bundles of densely packed filaments. From a mechanical
point of view, the proximal region is extremely compliant,
giving a rubber-like behavior, whereas the distal region is
stiffer and less extensible (see Fig. 1) exhibiting a highly
hysteretic large deformation behavior. In this study we will
focus on the behavior of the distal segment, developing a
constitutive model for this region. The model will be
shown to capture the major features of the stressstrain
behavior of the byssal thread, including the highly non-
linear stressstrain behavior, and its dependence on strain
rate and stretch-induced softening.
Background
Microstructure of the distal region
Transmission Electron Microscopy analysis [13] reveal
that all parts of the byssus are formed by the same basic
filamentous components (diameter 79 nm) organized in
different ways at the submicroscopic level and embedded
123
8944 J Mater Sci (2007) 42:89438956

120
[8] showed that the preCOLs organize themselves into
hexagonal (6 + 1) bundles that are banana-shaped
Nominal Stress [MPa]

100
Distal ~ 1 cm Proximal
(Fig. 3b). The presence of two bumps along the preCOLs
d ~ 0.1 mm
80 has also been observed. AFM images of a stretched byssal
60 thread suggest that the macroscopic strain is accommo-
40
~ 2 cm Distal dated by both the elimination of the tilt seen in the
Proximal unstrained thread and the unraveling of the flanking
20
Plaque globular domains. The initial angular or bent structure
0
0 0.5 1 1.5 2 2.5
may act to lock the neighboring fibres and prevent
Eng.Strain intermolecular shear/sliding within the bundles.

Fig. 1 Nominal stress-engineering strain behavior in uniaxial tension

of the distal and the proximal region of Mytilus californianus at an
extension rate of 10 mm/min (data replotted from Carrington and
Gosline [5]). The unstressed length of the specimen is l0 = 5.4 mm
for the proximal region and l0 = 12.5 mm for the distal region
in a microfibrillar matrix. In the proximal portion a cor-
rugated superficial dense layer covers a central region
characterized by undulating filamentous strands and
individual coiled filaments embedded in an abundant and
loose microfilamentous matrix. In contrast, in the distal
region the outer dense layer is smooth and covers a
spongy structure with tubular cavities lined by a compact
tissue (Fig. 2). The tissue consists of filaments extended
linearly, regularly arranged nearly parallel to the major
axis of the thread and embedded in a dense matrix with a
microfilamentous structure (Fig. 2a). More recently, the
fiber has been found to be of a collagenous nature [6, 7];
byssal collagen has a block copolymer structure of
alternating hard and soft blocks (preCOLs) consisting of a
central collagen domain flanked by elastin-like or silk-like
domains and terminated at both ends by histidine- and
DOPA-sequences (Fig. 3a). In addition, Hassenkam et al.
Stressstrain behavior of the distal region
The mechanical behavior of the distal region of Mytilus
californianus has been characterized using
quasi-static tensile tests [4];
cyclical loading tests [5];
tensile test at different strain rates [5].
The mechanical tests have been performed using an
Instron-1122 tensometer and a custom-made tensometer [4],
capable of a maximum extension rate of 1,000 mm/min.
Interestingly, the stressstrain curve of the distal region
shows the typical features of polymer response to large-
strain uniaxial deformation (see Fig. 1): an initially stiff
response, a clear yielding point, followed by a stress
plateau and subsequent strain hardening.
The yielding (at approximately 15% extension) plays
an important function on the thread behavior: the material
response is stiff prior to yield, yield then provides the
extensibility needed to distribute the load over multiple
threads at relatively constant load on an individual thread.
Hence, yield also limits the load level transferred to the

Fig. 2 Distal part of the byssus

of Mytilus galloprovincialis. (a)
Longitudinal section showing
the filaments (arrows) running
parallel to the thread axis and
tubular cavities. (b) Cross
section showing the cavities
dispersed in a homogeneous
tissue. (c) Oblique section
evidencing the filaments
(arrows) and their regular
pattern. (From Bairati and
Vitellaro-Zuccarello [1])

123
J Mater Sci (2007) 42:89438956 8945

Fig. 3 (a) Block domain

structure of byssal preCOLs (as
observed in Mytilus
galloprovincialis); (b) the
hexagonal (6 + 1) bundles of
bent-core preCOLs in the
banana configuration; (c) Model
of an array of preCOLs bundles.
(From Hassenkam et al. [8])

Distal region
80

70
Nominal Stress [MPa]

60
1000 mm/min
50

40

30

20 10 mm/min

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Eng. Strain

Fig. 4 Nominal stress-engineering strain behavior in uniaxial tension Gosline [5]). The true stresstrue strain behavior of polyurethane
of the distal region of Mytilus californianus at extension rates of PU1 at quasi-static and high strain rate regimes is shown on the right
10 mm/min and 1,000/mm (left) (data replotted from Carrington and (data replotted from Yi et al. [9])

adhesive plaque which helps to maintain adhesion to the Cyclical loading

surface.
Rate dependence
Samples from the distal region of byssal threads of Mytilus
californianus have been tested in uniaxial tension at
extension rates of 10 and 1,000 mm/min (see Fig. 4),
corresponding to 0.0133/s and 1.33/s, respectively.1 The
higher the strain rate, the larger the stress and the yielding
stress, whereas the yield strain remains relatively un-
changed. In addition, at the highest strain rate, the plateau
following yielding is less distinct. Again we observe that
synthetic copolymers such as thermoplastic polyurethanes
[9] (see detail in Fig. 4) as well as ethylene methacrylic
acid [10] exhibit a microdomain or aggregated structural
features and similar mechanical behavior.
1 the same curve, independent of the cycle. Both the
The unstressed length of the specimens used in the tests is not
mentioned in Carrington and Gosline [5]. In the present paper, we use
an initial length l0 = 12.5 mm, value reported in Bell and Gosline [4].
Cyclic tests at different strain levels have been conducted
for the distal regions of threads of Mytilus californianus.
Figure 5 shows the nominal stress-engineering strain
behavior at strain levels of 0.08 (Fig. 5a), 0.16 (Fig. 5b),
0.35 (Fig. 5c) and 0.65 (Fig. 5d). When the thread is loa-
ded beyond its yield point, much of the energy is dissipated
via deformation in the distal region. Hysteresis is a crucial
property for the thread, since it reflects a shock-damping
capacity: as the wave passes, the thread unloads and due to
energy dissipation the creature is not dashed against the
rocks.
The reloading stressstrain curves show a significantly
more compliant character when compared to the initial
loading behavior; this effect is known as softening
behavior and it is typical of many copolymeric materials
(see, for example, [9, 11]). Similar again to the copolymer
behavior, the unloading paths after a given strain all follow
hysteretic and softening behaviors are strain-dependent:
hysteresis and softening increase with increasing strain.

123
8946 J Mater Sci (2007) 42:89438956

Distal region
1.2
(a) Max. strain= 0.08 (b) Max. strain= 0.16 Polyurethane
Force [N]

True Stress [MPa]

0.8

0.4

0
1.2
(c) Max. strain=0.35 (d) Max. strain= 0.65
Force [N]

0.8

0.4
True Strain
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0. 2 0. 3 0. 4 0. 5 0. 6
Eng. Strain Eng. Strain

Fig. 5 Force-engineering strain behavior of distal portions of byssal
threads of Mytilus californianus at cyclic uniaxial tests (data replotted
from Carrington and Gosline [5]). Four different strain levels are
considered: (a) 0.08 (extension rate = 5 mm/min), (b) 0.16 (extension
rate = 5 mm/min), (c) 0.35 (extension rate = 10 mm/min), (d) 0.65
(extension rate = 10 mm/min). The true stresstrue strain behavior of
polyurethane PU1 at cyclic uniaxial compression tests at engineering
strain rate
_ 0:063 /s is shown on the right (data replotted from Yi
et al. [9])

In addition, after unloading, the distal region increases material microstructure (see The representative vol-
in stiffness with time and gradually returns to its initial ume element (RVE) section).
structure and mechanical behavior [5] (see Fig. 6). the constitutive model for the stressstrain behavior of
each phase. For this study, the distal portion of the
Micromechanical model byssal thread will be regarded as a two-phase compos-
ite: the constitutive model for the tissue will be
To understand the mechanisms that govern the behavior of presented in Material constitutive behavior section,
the distal portion of byssal threads, a micromechanical whereas the fluid filling the cavities will be simply
model, representative of the underlying structure, is modeled as a hydrostatic incompressible fluid.
developed. The model framework permits us to derive the the constitutive model of the boundaries between
macroscopic mechanical response from the material different phases. Pressure equilibrium is enforced
microstructure. The development of a micromechanical across the fluid/tissue interface.
model requires: the computation of the macroscopic mechanical behav-
ior of the material based on the response of the RVE
the geometric definition of a representative volume (see RVE loading and macroscopic response
element (RVE) which captures the main features of the section).

The representative volume element (RVE)

0.5

As described in Microstructure of the distal region

0.4 section, the distal region of the byssal thread consists of a
spongy tissue with tubular water-filled cavities. This
Force [N]

0.3 microstructure is idealized by a staggered array of ellip-

soidal cavities (see Fig. 7).
0.2 In particular, a cylindrical cell is employed, character-
ized by an initial radius R0 and initial height H0. Following
0.1 Socrate and Boyce [12], when axisymmetric loading about
t=0
t=30 min
the z-axis is applied to the material, geometric compati-
0
bility of deformation in the two families of antisymmetric
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 cells requires that
Eng. Strain

Fig. 6 Force-engineering strain behavior of distal portions of byssal

Rn2 RH0  n2 2Rn 0:5H0 2 ; 1
threads of Mytilus californianus at cyclic uniaxial tests (data replotted
where R(z) = R0(z) + ur(z) and n denotes the axial
from Carrington and Gosline [5]). Two cycles were performed at
t = 0 (solid line), followed by two additional cycles after 30 min coordinate for points at the outer radius of the cell in the
delay (dashed line) undeformed configuration. In addition, to enforce

123
J Mater Sci (2007) 42:89438956 8947

idealized microstructure z RVE fiber-reinforced materials has been investigated by Bisc-

D C hoff et al. [15, 16] who developed an anisotropic molecular
network model for skin tissue, Holzapfel [17, 18] in a
model of arterial wall tissue, Horgan and Saccomandi [19]
in the developments of a model for a matrix reinforced by
H0
fibers characterized by limiting extensibility, Cantournet
et al. [20] in the investigation of the effects of carbon
nanotubes on the mechanical behavior of elastomeric
materials.
A B
r
Let F ox=ox0 be the deformation gradient, mapping a
material point from the reference position x0 to its current
AR0
location x and J be its determinant, J detF .
Fig. 7 Construction of the RVE (defined by the rectangle ABCD on For an isotropic hyperelastic material the strain energy
the right) based on a staggered array of ellipsoidal fluid pockets; z is density W is a function of the invariants of the right
the thread axis Cauchy-Green tensor C FT F (or, alternatively, also the
left Cauchy-Green tensor B FFT ),
symmetry of the axial deformation about the cell midplan
(z = H/2) for the two families of antisymmetric W WI1 ; I2 ; I3 ; 5
where
uz n uz H0  n 2uz n 0:5H0 : 2
1h i
Equations 1 and 2 define the constraints to be imposed to I1 trC; I2 trC2  trC2 ; I3 detC J 2 ;
the RVE due to the presence of the anti-symmetric families 2
of cells. 6
I1 being essentially a scalar equivalent stretch measure and
RVE loading and macroscopic response I3 the square of the volume ratio. For an isotropic material
reinforced by a single family of fibers with direction v0 in
When the RVE is subjected to an axial loading condition, the reference configuration, the strain energy density W is
the top edge CD moves in z-direction, ur |AD = 0, given by [14]
uz |AB = 0, whereas ur |BC and uz |BC satisfy Eqs. 1 and 2.
The macroscopic stretch in axial and hoop direction are W WI1 ; I2 ; I3 ; I4 ; I5 7
given by
where
uz j ur j
kz 1 CD ; kh 1 BC ; 3
H0 R0 I4 v0
Cv0 ; I 5 v0
C 2 v0 ; 8
whereas, employing the average stress theorem (e.g., [13]), I4 being a measure of the fiber stretch. If n families of
the macroscopic cartesian stress components rij can be i
reinforcing fibers each characterized by a direction v0 are
expressed as considered, Spencer [14] has shown that the strain energy
Z Z density W depends not only on I(i) (i)
4 and I5 , but also on
1 1 higher order invariants, which capture the interaction
rij rij dV rik nk xj dS; 4
V V effects between different fiber families. However, the
V S
dependence of the material response on I2, I(i) 5 and the

(where xj is the position vector and n the outward normal), higher order invariants is generally weak, so that the strain
allowing us to work in terms of tractions on the boundary. energy can be simplified as

i
Material constitutive behavior W Wiso I1 ; I3 Wani I4 : 9

The byssal thread of marine mussels is a fiber-reinforced where the isotropic term Wiso reflects the contribution of
composite material, in which a microfibrillar matrix is the microfibrillar matrix and the anisotropic term Wani
reinforced by stronger fibers. Following pioneering work of captures the stiffening effect of the preferentially aligned
Spencer [14], the mechanical response of hyper-elastic fibrous network.

123
8948 J Mater Sci (2007) 42:89438956

Microfibrillar matrix behavior 60

D
No detailed information is available on the structure of the 50

microfibrillar matrix, and the identification of its protein C

content is a topic of investigation [21]. Therefore, the 40
matrix is modeled as an isotropic, hyperelastic and B

Force [pN]
mechanically nearly-incompressible Neo-Hookean mate- 30
rial. To include compressibility a bulk strain energy term is
A
added to the strain energy [22, 23], so that the total strain
20
energy of the material is given by

l K 10
Wiso I1  3  l log J J  12 ; 10 A B C D
2 2
0
where l and K denote the shear and bulk modulus, 0 20 40 60 80 100 120 140 160 180 200
respectively. The first Piola-Kirchhoff stress tensor Siso is Extension [nm]
found by differentiation with respect to C
Fig. 8 Typical saw-tooth profile obtained from single molecule
force-extension tests of protein molecules. (a) The molecule is stretch
oW until the held force is such that unfolding becomes highly probable.
S 2F ; 11 (b) Unfolding of a domain abruptly reduces the holding force, since
oC
additional contour length is released. (cd) The molecule is stretched
yielding again until a force is reached where the next domain unfolds. (From
[24])

Siso lF JKJ  1  lFT : 12

The corresponding Cauchy stress, r 1J SFT , is given by
l
riso B  I KJ  1I: 13 is taken into account in this study.
J
Fibrous network behavior
In this section the three-dimensional constitutive model for
the highly non-linear stressstrain behavior of the fibrous
network is developed. This model is based on a mathematical
representation of the network structure together with a model
of the axial force-displacement behavior of a single fiber.
Recently, single molecule force-extension tests showed
that the behavior of many protein molecules (including
muscle protein titin [24], extracellular matrix protein
tenascin [25], the red blood cell cytoskeleton protein
spectrin [26] and lustrin A [27]) exhibit a saw-tooth
profile (Fig. 8). Proteins are often characterized by a multi-
domain architecture composed of folded modules which
are linked together in series along the macromolecule.
When a critical stress level is reached inside the folded
domains, the folded modules will begin to sequentially
unfold, releasing additional contour length that makes the
macromolecule response more compliant and leads to a
distinct drop in load.
Waite et al. [7] showed that the collagen contained in
byssal threads (from both M. edulis and M. galloprovin-
cialis) is characterized by a breaking strain that greatly
exceeds the ultimate strain for more typical collagenous
structures, such as tendons. This large increase in length of
the preCOLs structure after stretch can only be explained
by the unraveling of the globular domains (Fig. 9), which
Since the preCOLs are incorporated into a continuous
three-dimensional macromolecular protein network, the
unfolding process affecting its constituent fibrils generates
a plateau in the stressstrain behavior. Therefore, the
unfolding of the globular domain permits us to explain the
nature of the yield point that characterizes the stress
strain behavior of the distal segment (Fig. 1) and the
resulting highly compliant and elastic unloading/reloading
behavior.
Network model: The fibrous network, in a simplified
mathematical representation, is described by:
an average fiber end-to-end distance, r0;
an average fully extended end-to-end distance, L;
globular domains
folded domain
unfolding
unfolded domain
Fig. 9 Byssal preCOL flanked by globular domains that unfold when
stretched

123
J Mater Sci (2007) 42:89438956 8949

an idealized network connectivity; q q

i i i i
an average initial orientation of the fibers, h0. kf v0
Cv0 I4 : 17

To capture the initial orthotropy of the network with a
preferred fiber orientation, an eight-chain orthotropic unit
cell is employed, as proposed by Bischoff et al. [15, 16].
The eight-chain network considers an idealized structure of
8 fibers located along the diagonals of a prismatic cell (with
dimensions a, b and c along the material axes singled out
by the unit vectors ^ea , ^eb and ^ec , respectively) (Fig. 10).
However, in this study the material axes will be aligned
with the coordinate axes singled out by the unit vectors ^e1 ,
^e2 and ^e3 and we will assume a = c, so that the initial cell
can be completely defined by the undeformed end-to-end
distance r0 and the average fiber orientation h0,
p ponents following a KronerLee decomposition [28, 29]
ac 2r0 sin h0 ; b 2r0 cos h0 : 14
Fiber strain energy: The three-dimensional stressstrain
behavior of the fibrous network can be determined
employing the orthotropic eight-chain network together
with a representation of the axial force-stretch behavior
of a bent protein fiber containing folded domains. In
determining the axial force-stretch behavior of such fi-
bers, we need to derive an expression for the force
displacement relation when axially stretching the fiber by
displacing the two end points and thus increasing the
end-to-end distance from r0 to r. The fiber stretching is
thus given by kf = r/r0 and may be multiplicatively
decomposed into unbending and axial stretching com-
(Fig. 11),

Therefore, the initial direction of each fiber in the cell is kf ksf kuf ; 18
described by the unit vectors
where kuf and ksf denote the stretch accommodated by
1 5
v0 v0 v01^e1 v02^e2 v01^e3 ; unbending and stretching, respectively. The force-stretch
2 6 behavior will consist of an initially compliant region where
v0 v0 v01^e1 v02^e2  v01^e3 ;
3 7
15 the axial stretch is primarily accommodated by the
v0 v0 v01^e1  v02^e2 v01^e3 ; unbending of the fiber; as the fiber straightens, the force-
4 8
v0 v0 v01^e1  v02^e2  v01^e3 ; stretch behavior stiffens, since the fiber will now be
directly axially stretched.
where
Idealizing the bent fiber to consist of two long, straight,
stiff segments of length L1 and L2 joined by a bend
1 as shown in Fig. 12, an expression for the fiber strain
v01 p sin h0 ; v02 cos h0 : 16
2 energy contribution due to unbending is obtained (see
The long, continuous nature of the fibers permits the Appendix A)
assumption of affine deformation, giving the fiber stretch
EI
k(i)
f of the i-th fiber to be wuf a  a0 2 19
2q0 a0
where EI is the bending stiffness of the fiber, q0 the initial
bend radius and a0 = a10 + a20 and a = a1 + a2 the initial
banana and current bend angle, respectively, given by

b

u
f

x2 unbending
s
f

stretching
x1 initial configuration

a f= f f
s u
x3
current configuration
a
Fig. 11 Schematic of decomposition of kf into unbending and
Fig. 10 Eight-chain, three-dimensional orthotropic unit cell stretching parts

123
8950 J Mater Sci (2007) 42:89438956

Fig. 12 Single bent fiber (left) Single fiber Fiber idealization

and its approximation into two
rigid segments of length L1 and L1 a L2
L2 connected by a circular bend r
a1 a2

Ar

2q3  
4L22 kuf 2 r02 kuf 2 r02 L22 L21 
2
_ f c xu
4 5; Lu au Lmax  Lt exp ; 25
a1 arcsin 2ku r0 L1
kB T
f

  20
where
kuf 2 r02 L22 L21
a2 arccos 2kuf r0 L2 :
 
_ DGu
Models treating the chain as an entropic wormlike chain au L0u exp  26
kB T
(WLC) and incorporating additional stretching terms
appropriate for enthalpic contributions have been is a lumped parameter. In Eqs. 25 and 26 L_0u denoted the
developed for polymers [30, 31]. However, here, the pre-exponential factor, DGu the energy barrier to unfolding,
increase of the fiber contour length during unfolding is xu the width of the activation barrier, Lt the current fibril
taken to be gradual and slow, so that its thermal contour length, Lmax the maximum fibril contour length, T
fluctuations may be neglected. Thus, the fiber strain the absolute temperature and kB the Boltzmann constant.
energy contribution due to axial stretching is simply Therefore, the fiber contour length is updated as
taken as
LtDt Lt L_u Dt: 27
EA 2 s
wsf ks r k  12 ; 21 The limited stressstrain data available at different rates
2L 0 f
makes it difficult to more fully assess the approximation of
where EA denotes the fiber axial stiffness and L the fiber
an Eyring model; more data is needed to either confirm this
contour length. The axial force versus extension
approach or to point to other process models.
relationship is obtained via f ow
or , yielding Note that refolding is an extremely low probability
during stretching and thus is neglected. This will be dis-
2EI kuf r0 a  a0 cussed later.
fu r
h i2 ; 22
a0 q0 Fibrous network stressstrain behavior: The strain en-
4L22 kuf 2 r02  kuf 2 r02 L22  L21
ergy density of the fibrous network is determined using
both the expression for the fibril strain energy and the or-
and
thotropic eight-chain network structure.
When the cell is stretched, its strain energy is simply
EA given by the sum of the strain energy of the eight con-
fs r0 ksf  1; 23
L stituent fibers
and fiber equilibrium requires that
X
4
i
Wcell 2 wkf ; 28
f f u f s: 24 i1
14
Therefore, employing Eqs. 18 and 2224, kuf and ksf can be (the fact that the stretch of fibers v0 is the same as in
58
solved. fibers v0 has been used), so that the strain energy
Unfolding criterion: The unfolding is a thermally acti- density is given by
vated process whereby stress on the folded domains lowers
the energy barrier of unfolding and hence increases its Wcell m X 4
i
probability. It is introduced to the fibrous network by Wani wkf ; 29
abc 4 i1
specifying the evolution of the contour length of its con-
stituents. Here, following Qi et al. [32], the rate depen- where m denotes the fiber density (number of fibers per unit
dence of unfolding is captured by an Eyring model [33, 34] reference volume). At this point, the first Piola-Kirchhoff

123
J Mater Sci (2007) 42:89438956 8951

stress tensor Sani can be obtained by differentiation of the Model implementation

strain energy density
! The constitutive model outlined above has been imple-
i
oWani mX 4
owf oI4 mented into the commercial finite element code ABAQUS/
Sani 2F 2F ; 30
oC 4 i1 oI i oC Standard, Version 6.6-1, using the user subroutine UMAT.
4
oI4
i
i i
The application of this subroutine requires 10 material
and since oC v0  v0 , Eq. 30 can be rewritten as constants
Isotropic matrix
mX 4
1 owf i i
Sani v  v0 ; 31
4 i1 ki oki l Initial shear modulus
f f 36
K Bulk modulus
i owf
where vi and Fv0 i f r0 . Substitution of Eqs. 23
okf
into Eq. 31 yields Fiber network

  h0 Initial fiber orientation

mX 4
1 EA 2 si i
Sani r k  1 vi  v0 ; 32 E Fiber Youngs modulus
4 i1 ki L 0 f
f
L10 Initial length of one of the two stiff fiber segments
or alternatively (substituting Eqs. 22 into 31) L20 Initial length of the other stiff fiber segment
H0 Initial fiber center-to-center distance
mX 4
1 d Fiber diameter
Sani
4 i1 ki 37
f
2 3
Therefore, the bending and axial fiber stiffnesses are given
6 2EI kuf r02 a  a0 7 i i
6 r 7 by
4 a0 q h i 2 5 v  v0 :
0
4L22 kuf 2 r02  kuf 2 r02 L22  L21
pd4 pd2
33 EI E ; EA E ; 38
256 4
The corresponding Cauchy stress is given by and the initial end-to-end distance is

  q q
m X4
1 EA 2 si r0 H02 L210 H02 L220 : 39
rani r k  1 vi  vi : 34
4J i1 ki L 0 f
f
Unfolding

Composite constitutive model

Lmax Maximum fiber contour length
au Lumped parameter 40
As outlined in Material constitutive behavior section, the xu Width of the activation barrier
strain energy function and Cauchy stress for the composite
material are simply given by the sum of the matrix and
fiber contributions, yielding The microstructural geometry parameters have all
been characterized by microscopy and documented in the
rriso rani literature; the remaining parameters are obtained by fit-
 
l mX 4
1 EA 2 si ting to available mechanical stressstrain data. The val-
BIKJ1I r k 1 vi vi : ues of the material constants used in the simulations
J 4J i1 ki L 0 f
f presented in Results section are listed in Table 1.
35 Appendix B details the procedure for obtaining these

Table 1 Material parameters

l [MPa] 7.5 K [MPa] 372.5 h0 25
used in Results section
E [GPa] 45 L10 [nm] 200 L20 [nm] 100
H0 [nm] 22 d [nm] 9 Lmax[nm] 385
au [s1] 2.5
104 xu [nm] 1
107

123
8952 J Mater Sci (2007) 42:89438956

parameters and also describes additional tests that are the banana shaped protein fiber; as the fiber straightens and
needed to better facilitate the identification of these it is stretched directly, the response becomes much stiffer.
material parameters. Finally, it is important to note that the model also captures
the rate dependence of the unfolding process.

Results Simple shear

Constitutive model Figure 14 shows the material behavior for the case of
simple shear loading, where
In this section, the response of the constitutive model for
the distal tissue described in Material constitutive X X0 tan c x02 ^e1 ; 41
behavior section is investigated.
where c denotes the shear angle. In this case four fibers of
Uniaxial tension the eight-fiber network are observed to extend (denoted
with Fiber A in Fig. 14) and four are observed to com-
Figure 13 demonstrates the ability of the constitutive press (denoted with Fiber B in Fig. 14), where the four
model to capture the characteristic features of material compressed fibers begin to elongate at the largest strains
response to large uniaxial tensile deformations: initial lin- since they have also rotated. When a fiber is compressed,
ear response, yielding, stretch-induced softening and the unfolding process is not activated and therefore the rate
strain hardening. The evolution of the contour length is dependent aspect of the response of that fiber is not acti-
also reported in the plot, showing the link between the vated. All fibers are observed to rotate towards the shear
material response and the unfolding process. The transition direction further demonstrating how macroscopic defor-
from the initially stiff to a very compliant response is due mation is accommodated by both rotation and extension/
to the start of the unfolding process, whereas strain hard- compression of the fibers.
ening begins when the unfolding process terminates. Fig- In the case of simple shear, it is not just the azimuthal
ure 13 also shows an initially (at strain levels lower than orientation (i.e., the azimuth angle with respect to the fiber
0.01) compliant response generated by the unbending of axis given by the angle h0 in Fig. 15left) of the fiber that

Fig. 13 Model prediction of 100

380
80
nominal stress-engineering Fibers contour 0.0133/s
Fiber contour length [nm]

90 70 1.33/s
Nominal Stress [Mpa]

Nominal Stress [Mpa]

strain behavior in uniaxial 80

length 340

tension for the composite 300 60

70
material (left) together with the 50
60
evolution of the fibers contour
50 40 Fibrous network
length (right axis). The
40 Nominal stress
separated contribution of the 30
fibrous network and of the 30
20
isotropic matrix are reported in 20
Isotropic matrix
the plot on the right 10 10

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Eng. Strain Eng. Strain

Fig. 14 Model prediction of 45

Fiber contour length [nm]

380
Fiber A
Nominal Shear Stress [Mpa]

nominal shear stress-nominal 40 340

shear strain tan c in simple 35 300
shear for the composite material Fiber B
0.5
Fiber B
30
at shear strain rates of 0.01/s Fiber A
0.4
and 1/s (left). On the right, the 25
f [N]

corresponding evolution of the 20

0.3
Fibrous network
axial force and the contour 0.2
15
length for both the fibers in Fiber A
10
tension and compression in the 0.01/s
0.1
Fiber B
eight-chain network are 5 1/ s
0
reported 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Nominal Shear Strain (tan ) Nominal shear strain (tan )

123
J Mater Sci (2007) 42:89438956 8953

Fig. 15 Model prediction of 50

x2 Top view - Fibers location
nominal shear stress-nominal 45
0.5 C

Nominal Shear Stress [Mpa]

shear strain tan c in simple 40 D B
shear for the composite material q0 x1
35 0.4 E A
at shear strain rates of 0.01/s x1
30 x3 F H

fc [N]
and 1/s in the case of a 16-chain 0.3 G
x3
network (left). On the right, the 25

corresponding evolution of the 20 0.2

Fiber A
axial force for all the fibers in 15
the 16-chain network is reported 0.1
10 B-H C-G D-F
0.01/s
5 1/s E
0

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Nominal Shear Strain (tan ) Nominal shear strain (tan )

determines its stretch. Therefore, in addition to an eight- Figure 16 shows the simulation (for both the constitu-
chain arrangement of fibers with initial azimuthal orienta- tive and the micromechanical model) and experimental
tion of 25, a case considering a cone of 16 fibers oriented results for uniaxial tension at a strain rate of 0.0133/s and
at the azimuthal angle of 25 with the same overall fiber 1.33/s. The simulations results agree very well with the
density is also subjected to simple shear. Figure 15 shows experimental data and correctly capture the rate dependent
the material behavior for this case of 16-fiber cone oriented behavior of the stressstrain behavior. Using the micro-
at h0 = 25 and the fibers equally spaced around the cone graph of Fig. 2 as a guide, the cavity is taken to be ellip-
(see detail in Fig. 15). In the uniaxial tension case the soidal with major and minor semi-axes a and b equal to 425
material response obtained employing the 16-chain net- and 75 nm, respectively (see detail in Fig. 16), whereas the
work is exactly the same as the one obtained employing the dimensions of the RVE are taken as 575 and 150 nm, so
eight-chain network. In contrast, when simple shear load- that the volume fraction of cavities f is equal to 0.12. In the
ing is considered, the responses given by the two different left part of Fig. 16 the cavities are taken to be water filled,
networks is slightly different, since the constituents fibers whereas the right part of Fig. 16 shows the material
are loaded differently (Fig. 15right). behavior comparing the cases when taking the cavities to
be empty versus water-filled.
For the case of uniaxial tension, the water-filled and
Micromechanical model empty cavities representations yield very similar stress
strain behavior for the thread. This nearly identical behavior
A two-dimensional axisymmetric finite element model of is a result of the cavity not providing any shear stiffness
the RVE of the distal segment of the thread has been (whether water filled or empty), for this case of uniaxial
constructed in the commercial code ABAQUS with a mesh tension, noting that uniaxial tension primarily samples the
constructed of 4-node, bilinear, hybrid, axisymmetric ele- shear stiffness of a material for materials possessing a low
ments (CAX4H). The fluid-filled cavity has been modeled shear modulus relative to the bulk modulus. In contrast, in
employing hydrostatic fluid elements (FAX2) and the fluid the case of a hydrostatic loading situation, the two cases
has been assumed to be incompressible with a density of would give very different results, since we would be sam-
q = 1,000 kg/m3. pling the volumetric stiffness of the thread.

Fig. 16 Nominal stress versus 70 70

engineering strain for uniaxial

A2b

60 1.33/s 60
tension at
_ 0:0133 /s and

_ 1:33 /s. Experimental and
Nominal Stress [Mpa]

Nominal Stress [Mpa]

50 50 1.33/s
Aa

simulation results (for both

constitutive and 40 40

micromechanical model with Ab b

water filled cavities) are 30 0.0133/s 30 0.0133/s
reported (left), whereas
20 20
micromechanical model results
Experiments
obtained with empty cavities 10 Micromechanical model 10 Water filled cavities
and water-filled cavities are Constitutive model Empty cavities

shown on the right 0 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Eng. Strain Eng. Strain

123
8954 J Mater Sci (2007) 42:89438956

In Figs. 17 and 18, the contours of the axial stress and
axial strain, respectively, are plotted in the RVE for both

_ 0:0133 /s and
_ 1:33 /s at strain level equal to 0.2,
0.4 and 0.6.
Simulations on cyclic loading-unloading tests were
conducted at a strain rate
_ 0:0133 /s and
_ 1:33 /s and
Fig. 19 shows cyclic loading to different maximum strains.
The unloading-reloading behavior shown in Fig. 19 pro-
vides the key that the yield is not a classic plasticity type
of yield, but it is a structural breakdown that changes the
hyperelastic behavior due to unfolding.
Conclusions
A constitutive model for the hysteretic large strain behavior
of the distal portion of mussel byssus threads has been
presented, providing new insight into its rate dependent
behavior. The constitutive model results demonstrate that
the yielding and the following plateau region in the
stressstrain curve correspond to simultaneous fiber
stretching and module unfolding taking place in the dif-
ferent constituents fibers. The rate effects on the stress
strain behavior due to the rate-dependence of unfolding are

Fig. 17 Progression of axial

stress during a test for the RVE
with water-filled cavities loaded
in uniaxial tension at

_ 0:0133 /s (left) and

_ 1:33 /s (right)

Fig. 18 Progression of axial

strain during a test for the RVE
with water-filled cavities loaded
in uniaxial tension at

_ 0:0133 /s (left) and

_ 1:33 /s (right)
Nom. Stress [MPa]

Fig. 19 Numerical simulations

of cyclic (loadunloadreload) 60 Max. strain = 0.08 0.0133/s Max. strain = 0.16
tests: Nominal stress- 1.33/s
40
engineering strain behavior at at
a strain rate
_ 0:0133 /s and 20

_ 1:33 /s. Four different strain (a) (b)
0
levels are considered: (a) 0.08,
(b) 0.16, (c) 0.35, (d) 0.65
Nom. Stress [MPa]

60
Max. strain = 0.35 Max. strain = 0.65
40

20
(c) (d)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Eng. Strain Eng. Strain

123
J Mater Sci (2007) 42:89438956 8955

also captured by the model. However the model does not 2q3
4L22 k2 r02  k2 r02 L22  L21 2
capture the difference between the unloading and reloading a1 arcsin4 5;
path seen in the cyclic tests. During unloading the unfolded 2kr0 L1
A:4
domains partially refolds, giving a reloading path some-  2 2 
k r0 L22  L21
what stiffer than the unloading path (depending on the hold a2 arccos :
time between unloading and reloading). This important 2kr0 L2
aspect of material behavior shall be incorporated into fu- The strain energy of the fiber due to the unbending of the
ture extensions of the model. The model shall also be ex- bend is given by
tended to the proximal region and adhesive plaque, so that
it will be possible to use model-based FEM simulations to 1
systematically investigate the behavior of multiple threads wu EIDj2 q0 a0 ; A:5
2
connected to the animal and their load sharing mecha-
nisms, which will ultimately provide insights into the where EI is the bending stiffness and Dj denotes the
adhesive mechanisms of this biologically engineered sys- change in curvature of the bend
tem.  
Finally, we would like to remark the fact that the pro- 1 1 a  a0
Dj  : A:6
posed constitutive model for mussel byssus threads is a q q0 q0 a0
starting point to understand the role of modular protein in
Substitution of Eq. A.6 into Eq. A.5 yields
many biological materials, including muscles [35, 36], red
blood cell membranes [37, 38] (M Arslan submitted),
EI
spider silk [39] and organic matrix in nacre [27, 40]. wu a  a0 2 ; A:7
2q0 a0
Acknowledgments This research was supported by the Dupont
so that the fiber force-stretch behavior is given by
MIT Alliance.

owu 2EI kr0 a  a0

fu q : A:8
or a0 q 0
4L22 k r02  k2 r02 L22  L21 2
2
Appendix A: Derivation of the strain energy
contribution due to unbending Note that in case of a symmetrically bent fiber (i.e.
L1 = L2), Eq. A.8 reduces to
Approximating the bent fiber to consist of two straight
segments of length L1 and L2 connected by a circular bend  
of length q a (with a = a1 + a2), as shown in Fig. 12, a u 2EI a0  2 arccos k cos a20 a0
f q
  cos : A:9
a0 q0 r0 2 2
form for the force-stretch behavior of the unbending of a 1  k cos a20
single fiber is derived. Both segments are assumed to be
rigid, so that all the deformation is accommodated by
unbending of the kinked region. In addition, the contour
length of the kink is assumed to be fixed (i.e. stretching of Appendix B: Parameter identification for the constitu-
the kink region is neglected), so that tive model
qa q0 a0 : A:1
The mussel byssus is a biological material and thus its
Figure 12 clearly shows that the angles a1 and a2 satisfy the mechanical properties are observed to vary, mostly influ-
condition enced by the season, the temperature, the water salinity
[41, 42]. Thus the identification of material parameters
L1 sin a1 L2 sin a2 ; A:2 entering in the constitutive model is aleatoric and related to
all these effects. However, Figs. 1 and 4 clearly show that
and, taking L > > q a, the fiber stretching k = r/r0 may be the stressstrain behavior of the distal segment consists of
approximated as three different regions (Fig. 20):

L1 cos a1 L2 cos a2 an initial linear response: it is used to identify the value

k ; A:3 of the fiber Youngs modulus E;
r0
yielding followed by stretch-induced softening: the
so that from Eqs. A.2 and A.3 a1 and a2 may be expressed yielding points at different strain rates are used to
as a function of k as identify the unfolding parameters xu and au;

123
8956 J Mater Sci (2007) 42:89438956

80 8. Hassenkam T, Gutsmann T, Hansma P, Sagert J, Waite JH (2004)

linear stretch-induced Biomacromolecules 5:1351
70 response softening strain hardening 9. Yi J, Boyce MC, Lee GF, Balizer E (2006) Polymer 47:319
10. Deschanel S, Boyce MC, Cohen RE (2007) Rate dependent
60 mechanical performance of Ethylene Methacrylic Acid (EMAA)
Nominal Stress [MPa]

E xu, au Lmax copolymers and POSS-enhanced EMAA nanocomposites. Pro-

50
40
yield point
30
20
1000 mm/min
10 10 mm/min
0 16. Bischoff JE, Arruda EM, Grosh K (2002) J Appl Mech 69:198
0 0.1 0.2 0.3 0.4 0.5 0.6
Eng. Strain
Fig. 20 Nominal stress-engineering strain behavior in uniaxial
tension of the distal region at extension rates of 10 mm/min and
1,000/mm. It consists of three different regions which used to identify
the material parameters
strain hardening: it permits us to identity the length of
the fully unfolded fiber Lmax.
The geometrical parameters h0, L10, L20, H0 and d have
all been be measured with transmission electron micros-
copy and the values reported in Hassenkam et al. [8] are
used.
To better facilitate the identification of the material
parameters entering in the constitutive model additional
tests are needed:
additional tensile tests at multiple strain rates to better
understand the rate-dependence of yield;
Relaxation tests to better understand the sources of the
rate-dependence;
Unloadingreloading tests at higher strain rates to
better understand the unloading/reloading process;
Force test of a single macromolecule to validate the
choice of material parameters.
References
1. Bairati A, Vitellaro-Zuccarello L (1976) Cell Tiss Res 166:219
2. Benedict CV, Waite JH (1986) J Morphol 189:261
3. Bairati A (1991) In: Lanzavecchia G, Valvassori R (eds) Form
and function in zoology. Mucchi, Modena, pp 163177
4. Bell E, Gosline JM (1996) J Exp Biol 199:1005
5. Carrington E, Gosline JM (2004) Am Malacol Bull 18:135
6. Coyne KJ, Qin XX, Waite JH (1997) Science 277:1830
7. Waite JH, Qin XX, Coyne KJ (1998) Matrix Biol 17:93
ceedings of ACS Spring MeetingSymposyum on segmental
block copolymers, Chicago
11. Qi HJ, Boyce MC (2005) Mech Mater 37:817
12. Socrate S, Boyce MC (2000) J Mech Phys Solids 48:233
13. Gurtin ME (1972) In: Flugge S (ed) Handbuch der Physik,
vol 6a(2). Springer, Berlin
14. Spencer AJM (ed) (1984) In: Continuum theory of the mechanics
of fibre-reinforced composites. Springer-Verlag, New-York, pp
132
15. Bischoff JE, Arruda EM, Grosh K (2002) J Appl Mech 69:570
0.7 0.8
17. Holzapfel GA (2000) Nonlinear solid mechanics, a continuum
approach for engineering. John Wiley and Sons, Chichester
18. Holzapfel GA (2003) In: Holzapfel GA, Ogden RW (eds) Bio-
mechanics of soft tissue in cardiovascular systems CISM lectures,
vol 441. Springer, Wien, pp 109184
19. Horgan CO, Saccomandi G (2005) J Mech Phys Solids 53:1985
20. Cantournet S, Boyce MC, Tsou AH (2007) J Mech Phys Solids
55:1321
21. Sun C, Vaccaro E, Waite JH (2001) Biophys J 81:3590
22. Ogden RW (1984) Non-linear elastic deformations. Ellis
Horwood, Chichester
23. Boyce MC, Arruda EM (2000) Rubber Chem Technol 73:504
24. Rief M, Gautel M, Oesterhelt F, Fernandez JM, Gaub HE (1997)
Science 276:1109
25. Oberhauser AF, Marszalek PE, Erickson HP, Fernandez JM
(1998) Nature 393:181
26. Fisher TE, Oberhauser AF, Carrion-Vazquez M, Marszalek PE,
Fernandez JM (1999) TIBS 24:379
27. Smith BL, Schaffer TE, Viani M, Thompson JB, Frederick NA,
Kindt J, Belcher A, Stucky GD, Morse DE, Hansma PK (1999)
Nature 399:761
28. Kroner E (1960) Arch Ration Mech Anal 4:273
29. Lee EH (1969) ASME J Appl Mech 36:1
30. Odijk T (1995) Macromolecules 28:7016
31. Wang MD, Yin H, Landick R, Gelles J, Block SM (1997)
Biophys J 72:1335
32. Qi HJ, Ortiz C, Boyce MC (2006) J Eng Mater Technol 128:509
33. Eyring H (1936) J Chem Phys 4:283
34. Bell GI (1978) Science 200:18
35. Minajeva A, Kulke M, Fernandez JM, Linke WA (2001) Biophys
J 80:1442
36. Dorfmann A, Trimmer BA, Woods WA Jr (2006) J R Soc
Interface (on line)
37. Mohandas N, Evans E (1994) Annu Rev Biophys Biomol Struct
23:787
38. Arslan M, Boyce MC (2006) J Appl Mech 73:536
39. Gosline JM, Guerette PA, Ortlepp CS, Savage KN (1999) J Exp
Biol 202:3295
40. Qi HJ, Bruet BJF, Palmer JS, Ortiz C, Boyce MC (2005) In:
Holzapfel GA, Ogden RW (eds) Mechanics of biological tissues.
Proceedings IUTAM. Springer-Verlag, Graz, Austria, pp 180
205
41. Carrington E (2002) Limnol Oceanogr 47:1723
42. Moeser GM, Carrington E (2006) J Exp Biol 209:1996

123
J Mater Sci (2007) 42:89578965
DOI 10.1007/s10853-007-1653-3
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
The micro-structural strain response of tendon
Vinton W. T. Cheng Hazel R. C. Screen
Received: 13 December 2006 / Accepted: 1 March 2007 / Published online: 19 July 2007

Springer Science+Business Media, LLC 2007
Abstract Tendons are multi-level fibre-reinforced com-
posites, designed to transmit muscle forces to the skeleton.
During physiological loading, tendons experience tensile
loads, which are transmitted through the structure to the
cells, where they may initiate mechanotransduction path-
ways. The current study examines the structural reorgani-
sation and resulting local strain fields within the tendon
matrix under tensile load. It uses confocal microscopy to
photobleached a grid onto the collagen and image its
deformation under the application of incremental tensile
strain. Six parameters are used to quantify fibril and fibre
movement and examine the mechanisms of extension
employed by fascicles.
Results demonstrated an inhomogeneous strain response
throughout the matrix and large variability between sam-
ples. Local strains in the loading axis were significantly
smaller than the applied values. However, large compres-
sive strains, perpendicular to the loading axis, were re-
corded. The average Poissons ratio (0.8) suggested cells
may experience significant compression during loading.
Deflection of the grid lines, indicating sliding between
collagen fibres, and rotation of the grid were also recorded.
These data highlight the non-homogenous strain environ-
ment of fascicles and provide further evidence for fibre
sliding under tensile load. They also suggested a rotary
V. W. T. Cheng
H. R. C. Screen (&)
Medical Engineering Division, Department of Engineering,
Queen Mary, University of London, Mile End Road,
London E1 4NS, UK
e-mail: H.R.C.Screen@qmul.ac.uk
V. W. T. Cheng
H. R. C. Screen
IRC in Biomedical Materials, Department of Engineering,
Queen Mary, University of London, Mile End Road,
London E1 4NS, UK
component to tendon response, which may indicate a
helical organisation to the tendon matrix.
Introduction
Tendons are fibrous connective tissues, specialised in the
function of uniaxial, tensile force transmission [1]. Struc-
turally, tendons are natural, multi-level fibre-reinforced
composites, built from the basic collagen molecule into
units of increasing diameter [2, 3]. The fibres are pre-
dominantly collagen type I, comprising between 70% and
90% of the tendon dry mass, depending on the location and
functional role of the tendon within the body and the
species of test subject [4, 5]. The collagen units are gen-
erally aligned in the direction of loading and are the
principle load-bearing components of tendon, conferring
much of its mechanical strength [6]. At the molecular level,
individual tropocollagen molecules, approximately 280 nm
in length, are packed at repeating 64 nm intervals to form
microfibrils, which aggregate into fibrils [7, 8]. Structural
integrity of the fibril is ensured by inter-molecular and
inter-microfibrillar cross-links, which aid strain transfer
between the discontinuous collagen molecules [9, 10].
The fibrils aggregate within the proteoglycan rich non-
collagenous matrix to build fibres and fascicles, the higher
order structural levels of the tendon. However, the detailed
organisation of these levels and the mechanisms by which
they facilitate the transfer of tensile load are less clearly
characterised. A number of studies have attempted to
determine the length and continuity of the collagen fibrils,
although their large aspect ratio and imaging limitations
have precluded a conclusive answer. Scanning electron
123
8958
microscopy images, showing very few fibril ends in mature
tissue, have led some authors to conclude that the fibrils are
continuous throughout the length of the tendon [11].
However, analysis of the composition and arrangement of
the non-collagenous matrix between fibrils, and an exam-
ination local mechanical properties have led other authors
to hypothesise a model with discontinuous fibrils embed-
ded within a functional non-collagenous matrix that di-
rectly facilitates strain transfer [1214].
Electron microscopy studies have located the proteo-
glycan decorin, bridging adjacent collagen fibrils within
tendon [15]. This has led to the development of a shape
molecule hypothesis, in which decorin is responsible for
controlling shear resistance and maintaining structural
integrity between adjacent discontinuous fibrils [16]. De-
corin has a horseshoe shaped core protein that binds around
collagen fibrils and a single, orthogonally aligned dermatan
sulphate glycosaminoglycan (GAG) chain which bridges
adjacent fibrils [13, 17]. It is hypothesised that the
recruitment of a large number of weak van der Waals
interactions between dermatan sulphate chains can provide
sufficient shear resistance to control strain transfer between
fibrils [16, 18].
X-ray diffraction studies also support the discontinuous
fibril hypothesis, reporting that only 40% of the total strain
applied to a tendon fascicle can be accounted for by fibril
extension, indicating the remaining 60% must occur
through relative movement between the higher structural
levels [19, 20]. In agreement, confocal microscopy studies
implicate fibre sliding as a dominant extension mechanism
for tendon and other hierarchical collagenous materials [14,
21]. Using the tenocyte cell nuclei as markers of fibre
motion, relative sliding of adjacent fibres has been visual-
ised under the application of tensile load. In addition, fur-
ther studies have demonstrated that altering the composition
of the non-collagenous proteoglycan matrix influences the
relationships between fibril and fibre sliding within tendon
fascicles, supporting a possible role for proteoglycans in
regulating strain transfer mechanisms [22]. These studies
provide important data relating to cellular deformation and
mechanotransduction; however the relationships between
the cells and the collagen matrix will influence the observed
response. In order to examine the matrix response directly, a
recent study has utilised a fluorescent photobleaching
technique to image the movement of the collagen fibrils
themselves and examine the strain response of the tissue
[23]. The authors photobleached lines through the collagen
matrix prior to the application of tensile strain. From images
of the lines under strain, they have suggested that fibril
sliding dominates extension, shielding the tenocytes from
excessive shear strains.
Fluorescent microscopy exploits the ability to visualise
fluorophores by exciting them with laser light. By
123
J Mater Sci (2007) 42:89578965
controlling the binding of fluorophores to particular mol-
ecules within a sample, specific structures can be visual-
ised. Photobleaching techniques, which further utilise this
principle, have gained considerable popularity in recent
years, with the development of techniques such as fluo-
rescent recovery after photobleaching (FRAP) and fluo-
rescent loss in photobleaching (FLIP) for the analysis of
cell dynamics [24]. In typical experiments a fluorescent
labelled protein is incorporated into a sample and small
target regions of the dye then photobleached by repeated
exposure to high intensity light from a laser beam. At high
intensities, the fluorophores are over excited causing them
to permanently lose their ability to fluoresce. If the sample
is then excited by a lower intensity laser, the undamaged
fluorophores in the non-photobleached areas, will still
fluoresce clearly demarcating the two regions. The mobility
of cellular components can be examined through moni-
toring the movement of the remaining fluorescing fluoro-
phores [25]. Although most commonly utilised to image
cellular structures, it is possible to use photobleached lines
as markers of strain within tissue samples, by staining the
complete tissue matrix with a fluorescent dye and then
selectively photobleaching regions to providing well con-
trolled areas of the sample that will not fluoresce when the
fluorophores are excited. The current study utilises this
technique to examine strain transfer through tendon fasci-
cles, testing the hypothesis that sliding will occur at both
the fibril and fibre levels of the tendon hierarchy.
Materials and methods
For all experiments, fascicles were taken from the proximal
end of the tails of male Wistar rats aged between 4 and
8 months. Fascicles, approximately 300 lm in diameter,
were teased from the tails within five hours of sacrifice, cut
to lengths of around 45 mm, and maintained in a moistened
state, in paper towel soaked with phosphate buffered saline
(PBS; Sigma, Poole, UK).
Straining system
Testing utilised a custom designed tensile straining rig
(Fig. 1), able to load fascicles while in situ on the stage of a
confocal microscope (TCS SP2, Leica Microsystems
GmbH, Wetzlar, Germany). A series of preliminary tests
were carried out to validate the system for the testing of
tendon fascicles and to determine the loading environment.
The rig and the grips are made from stainless steel, to
minimise corrosion and facilitate ease of cleaning and
sterilisation. The grips are shaped as triangular wedges
(Fig. 1b) in order to hold the fascicle adjacent to the glass
coverslip against the base of the rig, enabling the objective
J Mater Sci (2007) 42:89578965
Fig. 1 Photographs of (a) the confocal rig on the microscope stage
and (b) the grips for straining the sample
lens to focus internally within the tissue. Grips are main-
tained in horizontal alignment by a ball slider attached to
the lid of the rig, and stepping linear actuators control the
movement of the grips, allowing accurate and reproducible
strains to be applied to the specimen. Each step is
25 5 lm, corresponding to a strain of less than 0.02% on
a sample of length 15 mm. The maximal uniaxial dis-
placement is 15 mm, and the motors can withstand loads of
up to 50 N.
Fifteen incremental strain to failure tests were carried out
within the rig. Fascicles were secured in one grip and
allowed to hang under their own weight while the second
grip was positioned and then tightened, creating a test
length of 15 mm and defining the 0% strain position. The
grips were then loaded into the rig, immersed in PBS, and
the rig placed on the stage of the confocal microscope,
enabling the sample to be imaged under standard brightfield
light microscopy using a 4 magnification objective lens
(Plan Apo x4, Nikon, Kingston-Upon-Thames, UK). The
fascicles were loaded to failure in 1% strain increments at a
strain rate of 10% min1, taking an immediate snap shot of
the grips and sample at each increment. From these images,
the displacement of the grips at each increment was
examined and compared with the specified applied dis-
placement. In addition, the fascicles were examined after
failure, to assess the location and mechanism of failure.
8959
A further ten fascicles were subjected to a second series
of validation experiments. Samples were secured in the rig
in the same manner, and imaged under brightfield settings
using a 10 magnification air immersion objective lens
(Plan Apo x10, Nikon, Kingston-Upon-Thames, UK).
Fascicles were once again loaded to failure in 1% strain
increments at a rate of 10%/min1, with a hold time of
approximately 20 s at each increment, while an image of
the sample was recorded from which the gross sample and
collagen fibre orientation was examined. Once again, the
mechanism and location of failure was determined for each
fascicle.
Microstrain analysis
In order to examine the microstrain environment, twenty
fascicles were subjected to a sequential incubation in viable
fluorescent dyes to stain the collagen and the cell nuclei.
Each sample was incubated at room temperature (24 C)
following the procedure outlined below:
1. Twenty minutes (20 min) in 5-dicholorotriazynl fluo-
rescein (5-DTAF; Molecular Probes, Oregon, USA) at
a concentration of 2 mg/mL 0.1 M sodium bicarbonate
buffer (pH 9; Sigma, Poole, UK) to stain the collagen
components.
2. Twenty minutes (20 min) post-staining wash in PBS.
3. Forty minutes (40 min) in 5 mM Acridine orange
(Sigma, Poole, UK) in PBS to stain the tenocyte
nuclei.
Specimens were then briefly rinsed in PBS, prior to
loading into the confocal straining rig at a test length of
15 mm, following the procedure described in the Strain-
ing system section. The orientation of each sample was
checked under brightfield settings, with a 20 magnifica-
tion air immersion objective lens (HC PL Fluotar, Nikon,
Kingston-Upon-Thames, UK) and the fascicle then photo-
bleached, using a 488 nm Krypton-Argon laser. A series of
1 lm thick lines were bleached in the central region,
through the full width of the 5-DTAF stain, to create a grid
of four identical squares, 50 lm 50 lm. The laser
intensity was then reduced to the imaging range, and the
sample imaged with the same objective lens at a resolution
of 2,048 2,048 pixels.
An image of the photobleached grid was taken in a focal
plane approximately 2025 lm from the sample surface,
identified by the cell nuclei. The fascicle was then strained
to 1% at a rate of 10%/min1 and the cell nuclei used to
refocus the grid at the same focal depth, before imaging
again. Each fascicle was loaded to 2%, 4%, 6% and 8%
strain, and the grid imaged at each strain increment, with a
hold period of approximately 45 s before imaging at each
increment, whilst the focal plane was located. In addition, a
123
8960
series of ten images were taken at 15 lm increments
through the thickness of the sample, at the 8% applied
strain increment.
Image analysis
Images were processed using the analysis software Image J
(1.34s, National Institute of Health, USA) following a
similar procedure to that previously described by Bruehl-
mann and co-workers [23]. Two identical copies of each
image were filtered using Gaussian blur filters with radii of
10 and 20 pixels. The images were subtracted before
thresholding and skeletonising, to produce a single pixel
trace of the photobleached grid following the midline of the
original conformation (Fig. 2a). Any dirt or noise not related
to the photobleached region was removed, and the pixel
co-ordinates for the grid converted to data points for analysis
in Excel (Excel 2003, Microsoft Corp., Redmond, WA).
Six parameters, shown in Fig. 2b, were qualitatively
analysed to describe the in situ micromechanical environ-
ment.
x strain (100Dx/x) Percentage strain in x direction (with
axis of loading)
y strain (100Dy/y) Percentage strain in y direction
(perpendicular to loading axis)
y angle (hy) Angle of perpendicular lines relative to the y
direction
x angle (hx) Angle of parallel lines relative to the x
direction
Deflection (d) Step deflection along the length of the y line
Displacement (D) Displacement between ends of y line
along the x axis
Statistical analysis
Each parameter was analysed for the effect of increasing
strain application using StatsDirect (2005, StatsDirect Ltd.,
Cheshire). Analysis of Variance (ANOVA) statistical tests
Fig. 2 (a) Typical confocal (a)
image of a photobleached grid
at a depth of 30 lm into the
fascicle and at 4% applied
strain, showing the image
processing to provide a single
pixel trace of the grid. (b)
Schematic of the grid after
deformation, showing the six
deformation measurements
123
J Mater Sci (2007) 42:89578965
and Tukey post-hoc tests for analysing multiple compari-
sons of mean were applied. Statistical significance was set
at p < 0.05.
Results
Straining system
The grip displacement at each strain increment, measured
from the 4 magnification brightfield images, closely fol-
lowed that applied to the stepper motors, with a variability
between predicted and measured displacement of 5 lm
(3.3%). The grips successfully held the individual fascicles
and no slippage occurred in any of the samples. Twelve of
the samples (80%) failed in the central region at a distance
from each grip, and failure occurred through a slow
unravelling of the collagen fibres.
Examination of the fascicles under 10 magnification
brightfield light microscopy demonstrated that the samples
were aligned in the grips, and that the collagen fibres were
fully aligned in the direction of loading (Fig. 3). At the
gross level that fascicle was loaded uniformally and main-
tained the same orientation throughout testing. The crimped
fibres were seen to straighten and extend and no rotation
was visible. Sample failure was visualised as a pull out of
adjacent groups of fibres. Commencing at a strain increment
between 9% and 13%, visible relaxation of the matrix be-
came apparent, followed by sliding between groups of fibres
at random locations across the sample width. The same
sliding behaviour was visible along the length of the sam-
ple, and failure was finally seen to occur through complete
pull out of adjacent fibre groups, occurring at the 14%
increment in the majority (70%) of samples.
Microstrain results
A series of images from an example fascicle at 0%, 4% and
8% strain are shown in Fig. 4. Distortion of the photo-
(b) D
x
y +y
d
y
x + x
Applied Load
J Mater Sci (2007) 42:89578965 8961

(a) 1

0.8

Mean x strain (%)

0.6

0.4

0.2

0
0 2 4 6 8
Applied Strain (%)
Fig. 3 Typical image of an unstrained fascicle under brightfield
imaging, as taken at the start of every test. Image shows the alignment (b) 0
of the collagen fibres at the start of loading 0 2 4 6 8
-2

Mean y strain (%)

bleached grid was recorded in every sample, indicating an
inhomogeneous response to strain within the fascicle.
However, the nature of the strain response was highly
variable between samples. Figure 5a plots the mean strain
across the grid in the direction of loading (x strain) against
the applied strain for each increment. The local x strain
demonstrated a statistically significant increase during the
initial 2% of applied load, equalling a quarter of the applied
strain. With increasing applied strain, the rate of increase
decreased, leading to a local strain approximately 10% of
the applied value at 8%. The maximum local x strain
recorded in any sample was 2.5%.
By contrast, the local y strains perpendicular to the
loading axis (Fig. 5b) were highly variable, but indicated
large compressive strains. There was a linear relationship
between applied strain and mean y strain with a mean
Poissons ratio of 0.8 taken over all data points. Data from
some fascicles indicated a Poissons ratio in excess of 1,
with compressive y strains reaching values greater than
10% at 8% applied strain.
The other four strain parameters analysed the degree and
nature of distortion within the grid (Figs. 6, 7). Once again,
-4
-6
-8
-10
-12
Applied Strain (%)
Fig. 5 Mean local strain curves showing how (a) the x strain (n = 20)
and (b) the y strain (n = 11), in the grid change with applied strain.
Data depicts mean and standard error
the contribution of each parameter varied between samples,
as summarised in Table 1. A point of deflection (d) was
seen along the length of perpendicular grid lines, which
increased with applied strain in a statistically significant
manner. The deflection increased most rapidly during the
initial 2% of straining (Fig. 6a) and was clearly visible to
the eye (over 5 lm) in 55% of the samples, reaching a
mean value of 5.8 lm at 8% applied strain. The deflection
reached over 16 lm in two samples, accounting for
approximately 2% of the applied displacement.

Fig. 4 Images of the strain grid

in a typical sample at 0%, 4%
and 8% applied strain

123
8962 J Mater Sci (2007) 42:89578965

(a) 8 with applied strain, to reach a mean value of 1.9 at 8%

7 applied strain (Fig. 6b). However, the parallel lines became
Mean Deflection d (um)
6
distorted in each sample as the fibres became fully aligned
5
in the direction of loading, limiting the resolution capa-
4
3
bility of this measurement.
2 Displacement D provides a single measure to determine
1 the total grid deformation in the parallel plane. As reported
0 for other strain parameters in the perpendicular plane, the
0 2 4 6 8
most rapid increase was seen during the initial 2% of ap-
Applied Strain (%)
plied strain, after which a continual linear increase was
(b) 8 recorded to reach a mean maximum of 12.3 lm at 8%
7
Mean Angular Orientation ()

strain (Fig. 7). The 100 lm square grid covered approxi-

6
(y) mately one third of the total fascicle width (~300 lm). As
5
such, total displacement across the fascicle width may be in
4
the range of 3040 lm, accounting for approximately 3%
3
2
(x) of the applied displacement.
1
The analysis of the z-series images, taken at 8% applied
0
strain, demonstrated a non-homogenous strain response
0 2 4 6 8 through the thickness of the sample. The variation in local
Applied Strain (%) x and y strain with depth for three typical fascicles are
Fig. 6 Analysis of the mean local strain response of the grid with shown in Fig. 8. The x strain was seen to vary by a mean
applied strain depicting (a) the size of step deflections (d) along the amount of 0.3% through the thickness of each sample,
length of a line and (b) the angular reorientation of the lines. hy compared with a mean of 2% for the y strain. Both varia-
denotes the angle of the perpendicular lines, relative to the initial y tions are around a quarter of the mean local strain value for
direction and hx the angle of the parallel lines relative to the initial x
direction. Data depicts mean and standard error. (d) n = 20; (hy) the parameter, reported in Fig. 6.
n = 20; (hx) n = 11

16
Discussion
Mean Total x displacement (um)

14
12
10
8
6
4
2
0
0 2
Applied Strain (%)
Fig. 7 Analysis of the total displacement (D) enabled across the
length of the grid. Data depicts mean and standard error (n = 20)
The gridlines perpendicular to the loading axis demon-
strated a rotation, relative to their original orientation (hy).
Rotational direction varied between samples, hence
absolute values are reported, in order to assess the extent of
the reorientation. The reorientation hy demonstrated a sta-
tistically significant trend of increasing with applied strain
to reach a mean absolute value of 6.2 (Fig. 6b).
The angular reorientation of the parallel grid lines (hx)
followed the direction of hy for each individual fascicle but
was consistently statistically smaller. The mean absolute
angle followed a statistically significant trend of increasing
4 6
The confocal rig used for the current study provides a
mechanism of incrementally loading tendon fascicles in a
controlled manner, whilst visualising their in situ micro-
mechancial environment. Using the rat tail tendon model,
direct staining and photobleaching of the collagen matrix
was employed to provide a clear mechanism of monitoring
collagen behaviour directly. Validation experiments con-
8 firmed a controlled loading environment, in which the
fascicles were securely gripped and subjected to uniform
loading at the gross level, enabling the sub-structural
response to be examined.
The use of the two viable fluorescent dyes, 5-DTAF and
Acridine Orange, enabled both the cells and the photo-
bleached grid to be viewed simultaneously. The cells
provided an important marker to ensure all images were
taken at the same focal depth, and the ability to image the
cells and grid at the same wavelength ensured a minimal
imaging period at each increment. The suitability of both
stains has been demonstrated in previous publications, with
5-DTAF successfully staining collagen fibres in vitro and
in vivo [23, 26, 27], and Acridine Orange staining tenocyte
nuclei [14, 28]. However, the two dyes are rarely used
together, as they share similar absorption/emission max-
ima. Results demonstrated competitive staining between

123
J Mater Sci (2007) 42:89578965 8963

Table 1 Summary of mean data and range of values recorded for each deformation parameter at the 8% applied load increment
Values at 8% applied strain x strain (%) y strain (%) Deflection, d (lm) Angle, hy () Angle hx () Displacement, D (lm)

Sample number 20 11 20 11 20 20
Minimum value 0.17 0.48 0.73 1.05 0.84 1.7
Maximum 2.49 11.16 16.05 15.3 3.35 35.6
Mean value 0.83 6.84 6.22 6.2 1.9 12.3
Standard error 0.16 2.8 0.83 1.3 0.37 2.7

(a)
y studies by the authors have demonstrated a swelling of
x fascicles in response to incubation in different solutions,
and a subsequent change in the local strain parameters [22].
z
It is not possible to carry out these analyses without an
incubation step, and the fascicles in the current study have
(b) y strain (%) x strain (%) been treated in a similar fashion to the control (unswollen)
0
-12 -10 -8 -6 -4 -2 0 2 4
fascicles in the previous work. The differences recorded
-1 between fascicles in the earlier work were attributed to the
-2
longer twenty-four hour incubation period, however, it
z i n c r e me n t ( i n c r e a s i n g d e pt h )

should be noted that some swelling effects may be seen in

-3
all of the untreated fascicles, including those in the current
-4 study.
-5 Monitoring deformation of the grid enabled the analysis
of a series of local strain parameters, related directly to
-6
changes in the fibre or fibrillar structure. As in previous
-7 studies, there was a delay of approximately 60 s between
-8 straining the sample and taking the image, whilst the grid
was located and the image focused in the correct focal
-9
plane. Recent studies by the authors have examined the
Fig. 8 (a) Image of a fascicle, denoting the orientation of the x, y and stress relaxation response of tendon fascicles during the
z planes. (b) The variation in x strain and y strain with depth through initial 60 s after loading (not currently published). These
three typical fascicles, at 8% applied strain. Owing to the limited
z-resolution, depth is given in arbitrary units, where 1 unit is
data have demonstrated that the majority of relaxation
approximately 15 lm occurs during the initial 15 s of the relaxation period, by
sliding between adjacent fibres. About 45 s after the load
had been applied, the relaxation rate had significantly
the two dyes, evident from the limited number of visible slowed, suggesting that in the current study, the fascicle
cells. As the cell density was insufficient to confidently response was analysed after stress relaxation effects.
ascertain the fibre boundaries, there was no attempt to In good agreement with previous studies, the local strain
establish where line deflections occurred in relation to the in the direction of loading was found to be significantly
fibrillar structures. However, enough cells were visible to smaller than the applied strain at the fascicular level and
provide location markers for depth and ensure the grid was the strain response non-homogenous across the fascicle
continually imaged in the same focal plane. The use of a width [14, 23]. The development of differential strain fields
20 microscope objective lens greatly simplified tracking with the application of strain would be expected in such a
of the grid and provided a clear overview of the fascicle non-heterogeneous structure, and has previously been
response. However, its low numerical aperture limited its attributed to the sequential straightening and loading of
depth correction capabilities in the z-plane to approxi- individual fibrils with different crimp periodicity [30]. This
mately 1.28 lm, compared with 0.68 lm in the xy plane may also explain the variability seen in each parameter,
[29]. Data from z-series images were subsequently pre- with the deformation seen dependent on the crimp angle,
sented using arbitrary units and confined to providing an the original orientation of the collagen fibres and the order
indication of the change in xy parameters at each depth in which they straighten, all of which will vary with
increment. imaging location.
The staining procedures necessitate a period of incuba- Although the local x strain remained small, a progres-
tion in solution to enable the uptake of dye. Previous sive increase in the compressive y strain was recorded,

123
8964
indicative of a reduction in fibre diameter as the fascicle
was subjected to tensile loading (Fig. 5b). The mean
Poissons ratio, calculated to be 0.8, suggests that there
may be significant compressive forces generated within the
matrix under physiological loading conditions, that could
be transmitted to the cells, influencing the mechanotrans-
duction pathways. One possible explanation for the large
Poissons ratio is the exudation of fluid from the extra-
cellular matrix during stretching, as previously reported
[31, 32]. Electron microscopy has demonstrated inter-
fibrillar space within the structure, which may well
facilitate this response [22]. A visible improvement in the
definition of the collagen fibrils with increasing strain was
also noted, and images demonstrated fibrils falling in and
out of the focal plane. Both of these factors may be asso-
ciated with fibre alignment to the axis of loading and result
in a reduction of interfibrillar space.
The compressive strains in the fascicle matrix are larger
than those previously reported for compressive nucleus
strains under tensile load [28]. However, previous cell
mechanics studies have demonstrated that the nucleus of
connective tissue cells is generally stiffer than the sur-
rounding cytoplasm [33], hence strains may be represen-
tative of those experienced by the cell body. Compression
of the tendon ultrastructure under tensile load has also been
described by Wang and co-workers [34] using laser Raman
spectroscopy to examine the response of rat tail tendon
fascicles to uniaxial tension. The authors reported a rise in
wavelength shift after the toe region of the loading curve,
indicating lateral compression of the collagen fibre, pos-
sibly associated with the reorganization of the chemical
groups (particularly carbonyls) aligned orthogonally to the
axis of applied strain.
Deflection (d) was employed as a measure of relative
sliding between collagen structures. Most samples dem-
onstrated a single deflection across the width of the 100 lm
grid, indicative of sliding behaviour at the fibre level.
However, the limited capability to image cells meant it was
not possible to confirm that this point of deflection was at a
fibre edge. Deflections of up to 16 lm were seen in sam-
ples, with a mean value of 6.2 lm at 8% applied strain.
This accounts for less than 1% of the applied displacement,
however the grid covered approximately one third of the
fascicle width, so further deflection sites may be present
across the width. These values were smaller than those
previously reported for sliding behaviour between rows of
tenocytes [14, 22]. However, it is notable that the sliding
values previously reported by Screen and co-workers clo-
sely match the maximum displacement measurement D
taken in the current study. The displacement D incorpo-
rated both rotation and displacement of the grid and
reached a mean maximum value of 12.2 lm at 8% applied
strain. In previous studies, the analysis of nuclei position
123
J Mater Sci (2007) 42:89578965
would not have been able to differentiate between these
two parameters, hence the previous data may well have
been ascribed to rotation as well as fibre sliding and may
provide further explanation of why cells are seen to fall in
and out of focus.
In the current study, rotation was measured indepen-
dently, through an analysis of angular reorientation of grid
lines, both perpendicular and parallel to the loading axis.
Reorientation of perpendicular lines, reached a mean angle
of 6.2 and coupled with reorientation and distortion of the
parallel grid lines, indicates there may be a rotary element
to the sample strain response, supporting the hypothesis of
a helical tendon structure. Helical crimp patterns have
previously been reported in tendon [35, 36]. Using inter-
ference and polarised light microscopy, de Campos Vidal
[36] demonstrated wide variations in geometric features
and birefringence in the crimp structure of bovine and rat
tail tendon sections, which was attributed to a spiralling
fibril arrangement and a helical supra-organisation. In
addition, Kannus [37] described how fascicles, or tertiary
tendon bundles, frequently showed a spiral formation along
the course of the tendon. The chirality of the collagen
molecule has led a number of authors to hypothesise a
superstructure showing elements of handedness, as a
helical organisation would improve stability of higher
order structures through reciprocity in winding patterns
[3840]. This has caused some controversy, with other
authors suggesting that the reduced stiffness associated
with a helical structure would be too inefficient for the
tendon to effectively transfer force [41]. Although such a
structure would imply a less efficient means of transferring
muscle action to the skeleton, a helical organisation would
be better suited to resist flexion, lateral compression or
multidirectional deformation, all of which have been
reported to occur within tendon [4].
This study has concentrated on extension mechanisms
enabled by fibril and fibre movement within a single focal
plane. However, as a three dimensional structure, it would
be anticipated that similar extension mechanisms should be
seen in the z-plane. The z-series of images taken at 8%
applied strain confirmed that strain dispersal through the
matrix incorporated relative sliding between fibres in the
z-plane also. In agreement with data relating to individual
xy planes (Fig. 5), greater variation was recorded in the y
strains. However, it was notable that the extent of the
variability equated to approximately a quarter of the mean
strain value for that parameter.
Sliding has been implicated in a number of microme-
chanical studies as a mechanism for strain transfer and the
subsequent extension of tendon fascicles [14, 20, 21]. The
current study provides further evidence of sliding behav-
iour within the tendon ultrastructure, but also indicates
that rotation occurs in undamaged fascicular samples. It
J Mater Sci (2007) 42:89578965 8965

indicated that tendon extension could be consistent with a 19. Sasaki N, Odajima S (1996) J. Biomech 29(5):655
helical or spiral geometry in the fascicular structure, 20. Puxkandl R, Zizak I, Paris O, Keckes J, Tesch W, Bernstorff S,
Purslow P, Fratzl P (2002) Philos Trans R Soc Lond B Biol Sci
whereby extension is facilitated by a straightening of the 357:191
coil in addition to fibre sliding. 21. Bruehlmann SB, Matyas JR, Duncan NA (2004) Spine 29:2612
22. Screen HR, Shelton JC, Chhaya VH, Kayser MV, Bader DL, Lee
Acknowledgement Many thanks to Dr. Martin Knight, for his DA (2005) Ann Biomed Eng 33(8):1090
expert advice and assistance with the confocal microscopy. 23. Bruehlmann SB, Kelly EJ, Duncan NA (2005) Trans Orthop Res
Soc 30:389
24. Goodwin JS, Kenworthy AK (2005) Methods 37:154
25. Koster M, Frahm T, Hauser H (2005) Curr Opin Biotech 16:28
References 26. Woo HM, Kim MS, Kweon OK, Kim DY, Nam TC, Kim JH
(2001) Br J Ophthalmol 85:345
1. Woo SLY (1982) Biorheology 19:385 27. Davison PF, Galbavy EJ (1985) Invest Ophthalmol Vis Sci
2. Harris B (1980) Symp Soc Exp Biol 34:37 26:1202
3. Hiltner A, Cassidy JJ, Baer E (1985) Ann Rev Mater Sci 15:455 28. Arnoczky SP, Lavagnino M, Whallon JH, Hoonjan A (2002) J
4. Benjamin M, Ralphs JR (1997) Histol Histopathol 12:1135 Orthop Res 20:29
5. Ker RF (2002) CBPA 133:987 29. Petran M, Boyde A, Hadravsky M (1990) In: Confocal micros-
6. Elliott DM, Robinson PS, Gimbel JA, Sarver JJ, Abboud JA, copy. Academic Press, London, vol 9, p 262
Iozzo RV, Soslowsky LJ (2003) Ann Biomed Eng 31:599 30. Hansen KA, Weiss JA, Barton JK (2002) J Biomech Eng 124:72
7. Wess TJ, Hammersley AP, Wess L, Miller A (1998) J Struct Biol 31. Lanir Y, Salant EL, Foux A (1988) Biorheology 25:591
122:92 32. Hannafin JA, Arnoczky SP (1994) J Orthop Res 12:350
8. Buehler MJ (2006) PNAS 103(33):12285 33. Knight MM, van de Breevaart Bravenboor J, Lee DA, van Osch
9. Avery NC, Bailey AJ (2005) Scan J Med Sci Sports 15:231 GJVM, Weinans H, Bader DL (2002) Biochim Biophys Acta
10. Wess TJ, Cairns DE (2005) J Synchrotron Rad 12:751 1570:1
11. Provenzano PP, Vanderby R Jr (2006) Matrix Biol 25:271 34. Wang YN, Galiotis C, Bader DL (2000) J Biomech 33:483
12. Derwin KA, Soslowsky LJ, Kimura JH, Plaas AH (2001) J 35. Yahia LH, Drouin G (1989) J Orthop Res 7:2243
Orthop Res 19:269 36. de Campos Vidal B (2003) Micron 34:423
13. Redaelli A, Vesentini S, Soncini M, Vena P, Mantero S, 37. Kannus P (2000) Scand J Med Sci Sports 10:312
Montevecchi FM (2003) J Biomech 36:1555 38. Ottani V, Martini D, Franchi M, Ruggeri A, Raspanti M (2002)
14. Screen HR, Lee DA, Bader DL, Shelton JC (2004) J Eng Med Micron 33:587
218:109 39. Wess TJ, Hammersley AP, Wess L, Miller A (1998) J Mol Biol
15. Scott JE, Orford R (1981) Biochem J 197:573 275:255
16. Scott JE (2003) J Physiol 55:2335 40. de Campos Vidal B (2006) Matrix Biol 25:132
17. Weber IT, Harrison RW, Iozzo RV (1996) J Biol Chem 41. Raspanti M, Manelli A, Franchi M, Ruggeri A (2005) Matrix Biol
271:31767 24:503
18. Vesentini S, Redaelli A, Montevecchi FM (2005) J Biomech
38:433

123
J Mater Sci (2007) 42:89668973
DOI 10.1007/s10853-007-1586-x
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Towards an atomistic understanding of apatitecollagen
biomaterials: linking molecular simulation studies of complex-,
crystal- and composite-formation to experimental findings
Dirk Zahn Oliver Hochrein Agnieszka Kawska
Jurgen Brickmann Rudiger Kniep
Received: 7 December 2006 / Accepted: 6 February 2007 / Published online: 19 July 2007

Springer Science+Business Media, LLC 2007
Abstract The investigation of the atomistic mechanisms
of crystal nucleation constitutes a major challenge to both
experiment and theory. Understanding the underlying
principles of composite materials formation represents an
even harder task. For the investigation of the mechanisms
of crystal nucleation a profound knowledge of the ion
solvent and the ionion interactions in solution is required.
Studying biocomposites like fluorapatitecollagen materi-
als, we must furthermore account for the biomolecules and
their effect on the growth process. Molecular simulation
approaches directly offer atomistic resolution and hence
appear particularly suited for detailed mechanistic analy-
ses. However, the computational effort is typically
immense and for a long time the investigation of crystal
nucleation from atomistic simulations was considered as
impossible. We therefore developed special simulation
strategies, which allowed to considerably extent the limi-
tations of computational studies in this field. In combina-
tion with advanced experimental investigations this
provided new insights into the nucleation of biomimetic
apatitegelatin composites and the mechanisms of hierar-
chical growth at the micro- and mesoscopic scale. Along
this line, molecular simulation studies reflect a powerful
tool to achieve a profound understanding of the complex
growth processes of apatite/collagen composites. Apart
D. Zahn (&)
O. Hochrein
A. Kawska

R. Kniep
Max-Planck-Institut fur Chemische Physik fester Stoffe,
Nothnitzer Str. 40, 01187 Dresden, Germany
e-mail: zahn@cpfs.mpg.de
J. Brickmann
Eduard-Zintl-Institut fur Anorganische und Physikalische
Chemie, Technische Universitat Darmstadt, Petersenstrae 20,
64287 Darmstadt, Germany
e-mail: brickmann@molcad.de
123
from reviewing related work we outline future directions
and discuss the perspectives of simulation studies for the
investigation of biomineralization processes in general.
Introduction
The nucleation and growth of crystals is a phenomenon of
fundamental interest. This does not only apply to physics,
chemistry and materials science, but also to a specific field
of bio-sciences and medicine: biominerals reflect a fasci-
nating blend of organic soft matter and inorganic nano-
crystals. The resulting composites exhibit properties of
both hard and soft matter. Indeed, biominerals combine the
toughness of an inorganic material and the flexibility of
biological tissues. A profound understanding of these
properties requires investigating the interplay of the
organic molecules with the ions of the inorganic compo-
nent. This task is far from trivial. Exploring the atomistic
structure of a composite is complicated by non-periodic
local constellations of the organic molecules. Crystallo-
graphic methods may hence only provide insights in the
inorganic, crystalline component of biominerals. Apart
from identifying the atomistic structure, understanding the
mechanisms leading to these structures reflects an even
harder challenge. It requires an atomistic in-situ investi-
gation of ion aggregation, mineralization of the biomole-
cules and self-organization of the composite material.
Molecular simulation approaches may easily achieve the
microscopic resolution needed for such detailed mecha-
nistic analyses. In particular molecular dynamics (MD)
simulations appear to be a promising tool for studying
biomineral formation. However, the computational effort is
immense and special MD techniques are mandatory to
J Mater Sci (2007) 42:89668973
make simulations reasonable. The present review cannot
account for all of such approaches. We instead focus on
particularly promising methods, which have proven suc-
cessful in a series of apatite/gelatin aggregation studies.
Therein, the computational investigations are intended to
complement experimental work and cross-links are of vital
importance. Indeed, we feel that combined efforts from
both experiment and theory are needed to tackle the chal-
lenging complexity of biomineral formation [1, 2].
Apatite/collagen composites belong to the most abun-
dant biominerals in both humans and animal life forms. The
importance of this material as the predominant component
of bones and teeth motivated a large number of both
experimental and theoretical studies [1]. Despite these
efforts, our understanding of biomineral formation is still
limited. This particularly applies to phenomena taking place
at the atomistic scale. Indeed, only little is known about the
interplay of the apatite ions and the collagen proteins. The
latter issue reflects a key aspect of the composite as it
accounts for crucial characteristics of the biomineral such as
composition, morphogenesis and mechanic properties.
Atomistic simulation approaches
Atomistic simulations approaches may be divided in two
categories depending on the use of quantum or classical
mechanics. The latter concept relies on empirical potential
energy terms, which may account for hydrogen-bonding
and non-bonding interactions with reasonable accuracy.
However, the appropriate theoretical treatment of bond
breaking and formation usually requires quantum
approaches. In contrast to classical force fields, quantum
chemical calculations are considerably more demanding
with respect to computational resources. The desire to
combine the advantages of both classes of methods has
motivated the formulation of mixed quantum/classical
schemes. These approaches are based on reducing the
computationally very demanding quantum calculations on
the most relevant degrees of freedom, while describing the
remaining modes at much lower costs by classical
mechanics. Implemented in an appropriate way, quantum/
classical schemes may preserve much of the accuracy
obtained from quantum chemistry. On the other hand, the
benefits from transferring the efficiency of classical force
fields to large parts of the simulation model may easily
save orders of magnitudes of the computational costs.
Apatite simulation studies
The importance of apatite has motivated a manifold of
atomistic simulation studies. Early theoretical works on
bulk apatite crystals date back up to about 10 years ago and
8967
are based on quantum mechanical calculations, exclu-
sively. Lacking reliable empirical force-fields, ab-initio
approaches were chosen to investigate bulk crystal struc-
tures and the local arrangement of defects in apatites.
Many of the variety of apatite species may be written as
an isomorphous Ca10(PO4)6X2 series with X = OH, F,
Cl, or Br. Therein the ions X are embedded in channels
formed by staggered triangles of calcium ions. While suf-
ficiently small ionic species X = F are located in the
center of the Ca-triangles, larger ions were found off-center
along the [001] direction [3]. In hydroxyapatite, the most
relevant apatite for human hard materials, the OH ions
may be located either above or below the Ca-triangles.
From both crystallographic structure refinements and
ab-initio structure relaxation studies, the distance between
the hydroxide oxygen atom and the center of the Ca-tri-
angle was found as about 0.2 A [4, 5]. In the stable con-
figurations, the positively charged hydrogen atom of the
hydroxide ion is pointing away from the nearest Ca-trian-
gle. Transformation of the OH ordering therefore requires
not only crossing the embedding Ca-triangle by dislocation
of the oxygen atom by 0.4 A , but also an orientation
inversion of the complex anion.
Depending on the preparation conditions pure
hydroxyapatite exhibits either monoclinic (space group
P21/b; a = 9.4214(8) A , b = 2a, c = 6.8814(7) A ,
b = 120.00(8), Z = 4) [6] (Fig. 1, top) or hexagonal
(space group P63/m; a = b = 9.432 A , c = 6.881 A
, Z = 2)
[5] symmetry (Fig. 1, bottom). The latter modification
corresponds to the high temperature phase, which can be
obtained from the monoclinic modification via an
order disorder transition affecting the orientation of
the hydroxide ions, only. By means of optical birefringence
experiments [7], X-ray and difference scanning calorimetry
measurements this transformation was observed at tem-
peratures above about 200 C [8]. In principle, two dif-
ferent models may account for the (average) hydroxide ion
orientation (occupation of 0.5 for each crystallographic
position) in the high temperature phase: the mirror plane
perpendicular to the c-axis of the space group P63/m may
be associated to (i) a disordered column model in which
the orientation of the hydroxide ions is inverted at random
sites within the channels or (ii) an ordered column
model in which all OH ions in a single row point in the
same direction, while the orientation of each specific col-
umn with respect to the collective of all rows is randomly
distributed [5].
Both models are indistinguishable by X-ray or neutron
scattering experiments [9] and so far the only evidence
which picture accounts for the hexagonal modification of
hydroxyapatite was obtained from theory [4, 10, 11]. De
Leeuw performed a series of quantum mechanical structure
optimization studies based on various predefined OH
123
8968
Fig. 1 Monoclinic and
hexagonal structure of
hydroxyapatite. While the
monoclinic hexagonal
transition occurs at 200 C, the
hexagonal modification is meta-
stable at lower temperature
arrangements and energy minimization to local minima [4].
This approach is typically used for low-temperature
structure determinations from ab-initio calculations.
Indeed, the energy minimization corresponds to zero
Kelvin. Simulation setups of non-zero temperature require
Monte-Carlo or molecular dynamics approaches which are
computationally much more demanding. This motivated
the development of empirical potential energy functions for
modeling apatites at favorable costs [12, 13]. On the
basis of such force-fields we performed molecular
dynamics simulations of the temperature-induced mono-
clinic hexagonal transition [10]. The results of the
latter study and the structure optimization calculations of
de Leeuw are quite controversial. While de Leeuw
predicted concerted inversions of ordered columns [4],
independent flipping events of single hydroxide ions were
observed from the molecular dynamics simulations [10].
Quantum chemical approaches allow very precise
treatment of the atomic interactions. At first sight one
might therefore tend to favor the predictions made on the
basis of ab-initio calculations. However, in quantum cal-
culations limited computational resources typically imply
considerable model simplifications whose effect may out-
balance the benefits compared to empirical force-fields.
Indeed, in the present case modeling a temperature-induced
phase transition at zero Kelvin reflects such an oversim-
plification. Exploring inversion events of single hydroxide
ions on the basis of free energy calculations at room tem-
perature, Hauptmann observed a barrier of 87 kJ mol1
(S. Hauptmann, Private communication). At higher tem-
perature, the lattice expands and the edges of the Ca-triangles
are more susceptible to elongations. This reduces the
hydroxide-flipping barrier to 50 kJ mol1 at 200 C, i.e.,
the temperature at which the monoclinic hexagonal
transition takes place [10]. Moreover, the changes in
potential energy related to a single OH inversion event in
monoclinic hydroxyapatite reduces from 42, 23 kJ mol1 to
123
J Mater Sci (2007) 42:89668973
less than 10 kJ mol1 at 0, 300, and 573 K, respectively
[4, 10] (S. Hauptmann, Private communication).
Apatites grown in biologic systems typically contain a
manifold of defects (carbonates, water, etc.) [9]. Atomistic
simulations offer very detailed insights into the arrange-
ment of defects and local structure changes arising from
impurity incorporation. Most of the defect studies related to
apatites were dedicated to hydroxyapatite. A central
motivation for this is given by the fact that the inorganic
component of bones is represented by non-stoichiometric
(Ca-deficient) hydroxyapatite [9] whose impurities are
believed to be relevant for biological functions [14].
The general perspectives of atomistic simulations to
non-stoichiometric apatite studies may be very nicely
illustrated by the work of Astala and Stott [15]. Focusing
on carbonate incorporation in hydroxyapatite, these authors
investigated both hydroxide and phosphate substitution. To
avoid net charges, replacing OH or PO43 by CO32 ions
requires further changes in the system and Schottky defects
and protons (associated to hydroxide or phosphate ions)
accompany carbonate incorporation. Astala and Stott pre-
pared a series of model systems corresponding to a variety
of possible defect arrangements [15]. Each of the simula-
tion models was then optimized by relaxation to a local
minimum in the potential energy landscape. From this an
energetic scoring was obtained which may be related to the
occurrence of specific defect arrangements.
Apart from carbonates, the most abundant impurities in
biologic hydroxyapatite are H+ (associated to phosphate or
hydroxide ions), fluoride and chloride ions [16]. The
incorporation of halide ions is related to hydroxide sub-
stitution and directly preserves the total charge of the
crystal. From a dentists point of view, the most important
of these defect species is represented by fluoride ions,
which are known to protect human tooth enamel from
caries. OH F ion replacement in bulk hydroxyapatite
was investigated in a series of simulation studies employing
J Mater Sci (2007) 42:89668973
similar strategies as described above [4, 11]. On this basis,
fluoride defects were found to promote the mono-
clinic hexagonal transition. Indeed, the preferred
fluoride incorporation corresponds to OH


F


HO


OH
constellations. This fully compensates the energetic costs
for inverting the orientation of a hydroxide ion within the
aligned OH


OH


OH


OH columns of monoclinic 5 ps, and 0.5 ps/125 ps as thermostat and barostat relaxa-
hydroxyapatite [11].
Ion/water and apatite/water simulation studies
Interface studies
In biologic systems, apatites are formed from aqueous
solutions and remain in contact with water during their use
as bone or dental materials. The latter issue, i.e., the
structure of apatite/water interfaces may be investigated
from relatively straightforward molecular simulation
approaches. In 2003, we presented a molecular dynamics
simulation study of the hydration of (001) surfaces of
hydroxyapatite [17]. The investigated surface types were
obtained from different (001) cuttings from the ideal
crystal and should be considered as limiting cases locally
covering the features of slightly rough crystal faces. Unless
acidic conditions are applied, the solubility of hydroxyap-
atite in aqueous solution is extremely low. Indeed, our
0.5 ns molecular dynamics runs reflect a stable apatite/
water interface for which a series of structural and
dynamical properties could be sampled. Soon after this
work a similar molecular dynamics simulation was pre-
sented by de Leeuw from which hydroxide ion dissociation
was observed [18]. While no comparison of the two studies
was given in [18], we feel the striking differences must be
discussed before continuing the use of either simulation
model.
The two pictures may be easily discriminated from
experiment by adding dry apatite and pure water. Even
pulverized apatite samples with large surface to volume
fractions do not alter the pH of the system. This is in clear
contradiction to the observation of hydroxide ion release in
the simulations of de Leeuw [18] giving rise to the
assumption that the underlying empirical interaction model
might be inappropriate. However, comparing the de Leeuw
force field and the Hauptmann force field [12] used in our
simulations [17] we found both models to be very similar.
Instead, it seems the different setups of the molecular
dynamics simulations caused the lack of concordance. De
Leeuw applied identical relaxation times of 0.5 ps to the
isothermalisobaric algorithm. This choice is rather unu-
sual as in most constant pressureconstant temperature
molecular dynamics simulations artificial couplings of the
respective variables are avoided by assigning different
8969
relaxation times. Decorrelation is typically achieved by
setting the barostat relaxation time about ten times larger
than the thermostat relaxation time.
To check the effect of the isothermalisobaric algorithm
on the apatite/water interface reported in [17] we rerun 5 ns
sketches of our model system using 0.5 ps/0.5 ps, 0.5 ps/
tion times, respectively. The larger the thermostat/barostat
decoupling, the more reasonably small cell fluctuations
were observed. This particularly applies to the cell edge
parallel to the c-axis of the apatite slab, which varies from
0.5 to +10% for the 0.5 ps/0.5 ps setup. Applying 0.5 ps/
5 ps already reduces the corresponding fluctuations to less
than 1% (0.1% for the 0.5 ps/125 ps setup). Increased
flexibility, orientation inversion, and release of interfacial
hydroxide ions are strongly promoted by simulation cell
deformation. In fact, we expect the anomalously large
compression/expansion fluctuations of the apatite/water
interface study of de Leeuw to account for the observed
hydroxide dissociation.
Aggregation studies
For the investigation of apatite nucleation a profound
knowledge of both the ionwater and the ionion interac-
tions in solution is required. While there are well-estab-
lished empirical force-fields for studying calcium,
phosphate, hydroxide and fluoride ions in aqueous solution,
proton transfer events cannot be assessed without quantum
mechanics so far. Apatite formation is known to dramati-
cally depend on the pH [1]. A full account of the ionion/
ionsolvent interactions hence must include the consider-
ation of proton transfer reactions. At ambient conditions, it
is reasonable to assume that these processes only occur in
solution and at the apatite/water interfaces. Proton transfer
reactions may hence be expected during ion aggregation
and crystal growth, but not inside the bulk material.
The crystallization of calciumphosphates from aqueous
solution is a process of considerable complexity and the
degree of protonation of the phosphate ion can play an
important role. Depending on the pH, phosphoric acid
exhibits several stages of deprotonation (pKa1 = 2.16,
pKa2 = 7.21, pKa3 = 12.32; at 25 C [19]). At physiologic
conditions the fraction of completely deprotonated phos-
phate ions may hence be expected to be extremely small.
Nevertheless, crystallization may yield calciumphosphates
(including hydroxyapatite) consisting of PO43 ions even
under acid conditions. In such cases, the depronation of the
(di)hydrogenphosphate ion has to occur during crystal
growth. We investigated the initial steps of calcium and
hydrogenphosphate ion aggregation in aqueous solution by
means of quantum/classical molecular mechanics simula-
tions [20]. By combination of studying calcium ion
123
8970
aggregation to (hydrogen)phosphate ions and investigating
hydrogenphosphate deprotonation, we identified the
[Ca2+


PO43


Ca2+] ion triple as the smallest stable
aggregate, which may be expected to contain an entirely
deprotonated phosphate ion. This example demonstrates
the deprotonation of the hydrogenphosphate ions to be
promoted by its local environment, i.e., by positively
charged calcium ions:
During the initial steps of calciumphosphate nucleation
in aqueous solution the acidicity of HPO42 increases
dramatically. By association of one Ca2+ ion, the anionic
acid HPO42 is converted into a [Ca2+


HPO42]0 complex,
which may be interpreted in terms of a neutral acid. While
more likely to be deprotonated, the acidicity of
[Ca2+


HPO42]0 is still rather weak. Proton transfer
becomes likely only after aggregation of a second calcium
ion leading to a [Ca2+


HPO42


Ca2+]2+ complex which
may be considered as a cationic acid.
For a realistic molecular dynamics study of crystalliza-
tion processes from solution simulation models corre-
sponding to slightly super-saturated salt solutions are
required. In the case of low salt solubility, this implies
immense numbers of solvent molecules. The major obsta-
cle to straightforward molecular dynamics simulation of
crystal formation from strongly diluted solutions is related
to the mainly diffusion controlled character of such sys-
tems. In the extreme limit the ion concentration is so low
that most of the solutes does not interact with each other
and behave as if being in an infinite dilution. In this picture
the ions diffuse freely, until some of them happen to get at
close distance to each other. Attractive ionion interactions
may then lead to the formation of aggregates.
The formation of ion pairs and triples as described above
was investigated by ignoring ion diffusion and focusing on
the association of nearby ions. The related free energy
landscape is scanned as a function of a few degrees of
freedom and the minimum energy path is used as a model
reaction coordinate. In complex systems like ion solutions
the real reaction coordinate depends on all degrees of
freedom, including the solvent molecules. However, for
studying the association of an ion pair, in many cases the
ionion distance reflects a very suitable model of the
123
J Mater Sci (2007) 42:89668973
reaction coordinate. Adding further ions, we already have
to consider at least two degrees of freedom. Apart from the
distance coordinate also the orientation of the aggregate
with respect to the newly attached ion must be considered.
For solute/solvent combinations of very low solubility
such as calciumphosphate/water or CaF2/water systems we
found very stable ion complexes which did not dissolve
even during high-temperature annealing runs [20, 21].
Aggregates larger than ion pairs or triples exhibit several
adsorption sites. While the energy score of one ion docking
process might be more favorable than another, a variety of
adsorption sites may lead to stable aggregates. In such
systems, crystal growth does not necessarily follow the
minimum energy path, but is also controlled by the local
availability of ions which is directly connected to ion dif-
fusion in the dilute solution.
As computational costs force us to avoid direct molec-
ular dynamics simulations of slow diffusion processes we
developed a simulation scheme which instead mimics ion
distribution in solution originating from diffusion [21].
This is accomplished by a Monte-Carlo step in which an
ion is placed at a random position in the vicinity of the
aggregate. The nearest adsorption site is then guessed from
steepest descent energy optimizations. Finally, full relax-
ation of the total simulation system provides a new
aggregate to which further ions may be attached. During
this iterative procedure the ions are added one-by-one to
the aggregate being formed. The initial aggregate consists
of a single ion and each growth step represents the asso-
ciation of a further ion. Despite its appealing simplicity, we
wish to point out that this ion-by-ion attachment procedure
relies on a series of approximations which should be
carefully checked for each specific study of crystal
formation from solution.
The example of CaF2 aggregation from aqueous solu-
tions nicely demonstrates the perspectives of our aggregate
growth model. Starting from single ions, we can track
crystal nucleation from the very beginning. The early stage
of CaF2 formation is of particular interest because the
evolution of small complexes to aggregates counting
50100 ions reflects a disorder order transition. Indeed,
our simulations offer insights into the self-organization
process leading to the CaF2 crystal structure. We identified
the formation of regular motifs to nucleate in the inner core
of the aggregate (Fig. 2). Adding further ions to the aggre-
gate surface implies structural rearrangements which cause
the growth of the ordered domain and eventually leads to
crystallite formation.
The limitations of the aggregate growth model may be
nicely demonstrated by the example of NaCl aggregation
from aqueous solution, i.e., a solute/solvent combination of
high solubility. Transition path sampling molecular
dynamics studies of this process revealed a competition of
J Mater Sci (2007) 42:89668973
Fig. 2 Snapshot of a [Ca28F54]4+ aggregate as obtained from our
aggregation model reported in [21]. Calcium and Fluoride ions are
illustrated in green and blue, respectively. Only two water molecules
of the dilute aqueous solution are shown (upper part). The dashed
lines indicate ion water (yellow) and waterwater (cyan) interactions
which stabilize the aggregate surface and its solvent structure
association and dissociation events, as well as the agglom-
eration of small aggregates [22]. Crystal growth is hence
controlled by the net dynamics of growth/dissolution
events. In the aggregate growth scheme described above this
competition is expected to be fully biased in favor of
association steps only. Dissociation events are assumed to
be negligible. In terms of classical nucleation theory this
means that even the smallest possible aggregate is consid-
ered as a post-critical nucleus. This approximation may only
hold for solute/solvent combinations of very low saturation
concentration exhibiting purely diffusion controlled aggre-
gate growth. As NaCl solubility in water is rather large, it is
not surprising that dissociation events were observed
during the aggregate relaxation runs [21]. This is a clear
indicator of the invalidity of our aggregate growth scheme
for the study of this specific solute/solvent combination.
Our aggregate growth model should be considered as
complementary to other approaches to crystal formation
such as transition path sampling and free energy methods
combined with classical nucleation theory. The latter
methods are suited for systems of large ion concentration
(most applications are actually related to crystallization
from the melt, i.e., at 100% ion concentration). While these
approaches become increasingly inefficient for solutions of
low ion concentration, our method is dedicated to the study
of the aggregation of compounds exhibiting very low sol-
ubility constants. As nature chose particularly insoluble (in
water) compounds as biominerals, we consider our
approach as well-suited for aggregation studies in biomi-
metic model systems.
8971
Apatite/collagen/water systems
Setting our focus to apatite/collagen composites we can
clearly identify which interactions may be modeled by
empirical force fields and for which a quantum description
is inevitable. In bulk apatite the arrangement of the ions is
dominated by the Coulomb interactions and repulsive for-
ces, which prevent atomic overlaps. This interplay can be
described by empirical force fields at reasonable accuracy
[12, 13, 23]. A similar situation is represented by the soft
matter part of the composite, i.e., the collagen proteins. The
latter may be modeled by force field packages designed for
biopolymers like Amber, Charmm, etc. Along this line,
Buehler performed comprehensive model studies of the
mechanical properties of collagen triple helices and
assemblies thereof [24, 25]. Apart from quantitative
agreement with experimental data this work provides new
insights to collagen deformation and fracture behavior and
reflects a nice example how model simulations may com-
plement experimental findings to extend our mechanistic
understanding.
However, this only holds for the bulk phases. In the
previous section we discussed proton transfer reactions
occurring during phosphate aggregation from aqueous
solution. Similarly, also the collagen protein could be
subject to (de)protonation events promoted by ion associ-
ation. While the most appropriate account of such pro-
cesses is surely represented by quantum chemistry
methods, the complexity of collagen/ion/water systems
motivates the choice of computationally less expensive
approaches. In the two case studies described in the fol-
lowing ion aggregation is explored by considering different
protonation states in parallel classical mechanics calcula-
tions. Within the mainframe of the force fields, each spe-
cific protonation state is fixed and bond breaking/formation
reactions cannot be assessed. However, by comparison of
several possibilitieslike HPO42 and PO43 ion aggre-
gation to collagenone can at least obtain some insights
into the importance of proton transfer reactions.
The association of single calcium, phosphate and fluo-
ride ions to a collagen protein reflects the very initial step
of fluorapatite nucleation in a collagen matrix. As an entire
matrix of fiber proteins cannot be tackled within atomistic
simulations, it is reasonable to focus on specific aspects of
the collagen matrix. The tails of the collagen fibers are
expected to be relatively flexible (compared to the inner
part of the triple-helices) and were therefore suggested as
particularly suitable for associating ions and accommo-
dating apatite aggregates [1]. Indeed, the charged groups
located in the ends of the peptide strands were found to be
practically freely accessible to ion attachment from aque-
ous solution [26]. This applies to several side chains and
the termini of the peptide backbone. The variety of putative
123
8972
adsorption sites in the tails of collagen fibers was system-
atically investigated from free energy calculations. Each
candidate for calcium, phosphate or fluoride ion association
was ranked on the basis of the resulting energy profiles.
The analysis of proteinion bonds demonstrated the ability
of the collagen tails to trap specific ions from the aqueous
solution and hence promote local aggregate formation. The
most stable complexes were found for calcium association,
which is favored by 2025 kJ mol1 over the dissolved
state [26]. The tendency to phosphate and fluoride ion
adsorption is less pronounced, however these processes are
still exothermic exhibiting formation energies of about
10 kJ mol1. Other species, like sodium or chloride ions,
were found to be unable to form stable bonds to either the
teleopeptide model (tails of collagen) [26] or the bulk triple
helix [27]. These findings are in excellent agreement with
experimental findings. Indeed, sodium or chloride ions are
often used as inactive counterparts for biomimetic apatite
crystallization from phosphate and calcium solutions,
respectively [1]. On the other hand, from atomic force
microscopy experiments calcium ions are known to
strongly interact with collagen and even account for a
stiffening of the biomolecules [2830].
The hardening of the collagen fibers by pretreatment
with calcium ions was recently used to control the mor-
phogenesis of biomimetic fluorapatitegelatin composites
[27]. In order to explain the relation of the different growth
mechanisms observed from electron microscopy and the
ion impregnation prior to the crystallization experiments,
we performed a series of Ca2+, HPO42, and PO43 docking
simulations to a model triple helix [27]. Therein the col-
lagen fibers were mimicked by (Hyp-Pro-Gly)n polypetides
(Fig. 3) and the aggregate growth scheme [21] described in
the previous section was used for efficient sampling of
possible ionprotein complexes. Along this line, calcium
ions were found to be incorporated inside the triple helix.
Though the local structure of the protein must be changed,
calcium association does not alter the overall shape of the
triple helix which may be described as a linear coil. On the
other hand, phosphate ions (irrespective of the protonation
state) were found to bind laterally to the collagen model.
The proteinphosphate ion interactions may involve 13
hydrogen bonds. Binding via a single proteinphosphate
contact does not affect the shape of the triple helix.
However, this picture changes dramatically when more
than one hydrogen bond are involved. In the latter case we
observed biomolecule deformation induced by partial
wrapping of the protein around the phosphate ion. By
association of multiple phosphate ions, the fiber is bent
several times with no preference for either direction.
Impregnation of the collagen molecules by phosphate ions
hence implies a considerable structural manifold, whereas
calcium impregnation fixates the linear alignment of the
123
J Mater Sci (2007) 42:89668973
Fig. 3 3 (Hyp-Pro-Gly)n polypeptide model mimicking a collagen
triple helix. The backbone of each fiber strand is illustrated by a
yellow ribbon. Left: association of a phosphate ion with binds laterally
and causes a deformation of the biomolecule. Right: incorporation of
a calcium ion leading to a stiffening of the polypeptide. The water
molecules of the solvent are omitted for better visibility
triple helices. On the basis of atomistic simulations we can
hence identify different collagen pre-structuring effects
induced by calcium and phosphate ion association and
rationalized the different composite morphogeneses
observed from experiment [27].
Conclusion
It is clearly a long way to go from single ion association to
apatite/collagen composite formation, and our work should
be considered as a first step on this journey. In yet ongoing
efforts, we explore the formation of larger aggregates using
the same methods as described in this review. However,
with the computational resources available today, atomistic
simulations can only focus on fundamental aspects of
biomineralization. Our studies of ion aggregation in water
and in collagen/water systems may illustrate the perspec-
tives of such simulations.
While special methods of atomistic simulations allow
the investigation of nanocrystal aggregation from solution
[21], insights at the meso- or macroscopic scale require
different approaches. Atomistic models mimicking a
mesoscale system would comprise of an overwhelming
number of particles. To avoid this computational collapse,
we need strong simplifications whose basis is hard to get.
In a recent study of the electric field intrinsic to apatite/
collagen composites, we approximated the fiber proteins by
a simple dipole model [31]. In this specific case, the
required model parameterization could be based on com-
prehensive experimental data [1]. Nevertheless, we are still
J Mater Sci (2007) 42:89668973 8973

far from a generally applicable mesoscopic model of apa- 13. de Leeuw NH (2004) Phys Chem Chem Phys 6:1860
tite/collagen materials and complexity reduction by coarse 14. Posner PA (1969) Physiol Rev 49:760
15. Astala R, Stott MJ (2005) Chem Mater 17:4125
graining of biominerals remains a major task for the future. 16. Dorozhkin SV, Epple M (2002) Angew Chem 114:3260
17. Zahn D, Hochrein O (2003) Phys Chem Chem Phys 5:4004
Acknowledgment Financial support was provided by the Deutsche 18. de Leeuw NH (2004) J Phys Chem B 108:1809
Forschungsgemeinschaft. 19. Lide DR (19971998) CRC handbook of chemistry and physics,
78th edn. CRC Press, Boca Raton, New York
20. Zahn D (2004) Z Anorg Allgem Chem 630:1507
References 21. Kawska A, Brickmann J, Kniep R, Hochrein O, Zahn D (2006)
J Chem Phys 124:24513
22. Zahn D (2004) Phys Rev Lett 92:40801
1. Kniep R, Simon P (2007) Top Curr Chem 270:73 23. Bhowmik R, Katti SK, Katti D (2007) Polymer 48:664
2. Harding JH, Duffy DM (2005) J Mater Chem 16:1105 24. Buehler M (2006) J Mater Res 21:1947
3. Rulis P, Ouyang L, Ching WY (2004) Phys Rev B 70:155104 25. Buehler M (2006) Proc Nat acad Sci 103:12285
4. de Leeuw NH (2002) Phys Chem Chem Phys 4:3865 26. Schepers T, Brickmann J, Hochrein O, Zahn D (2007) Z Anorg
5. Kay MI, Young RA, Posner RA (1964) Nature 204:1050 Allgem Chem 633:411
6. Elliot JC, Mackie PE, Young RA (1973) Science 180:1055 27. Tlatlik H, Simon P, Kawska A, Zahn D, Kniep R (2006) Angew
7. van Rees HB, Menngeot M, Kostiner E (1973) Mater Res Bull Chem 118:1939
8:1307 28. Thompson JB, Kindt JH, Drake B, Hansma HG, Morse DE,
8. Suda H, Yashima M. Kakihana M, Yoshimura M (1995) J Phys Hansma PK (2001) Nature 414:773
Chem 99:6752 29. Currey J (2001) Nature 414:699
9. Elliot JC (1994) Structure and chemistry of the apatites and other 30. Fantner GE, Hassenkam T, Kindt JH, Weaver JC, Birkedal H,
calcium orthophosphates in studies in Inorganic Chemistry 18. Pechenik L, Cutroni JA, Cidade GAG, Stucky GD, Morse DE,
Elsevier, Amsterdam Hansma PK (2005) Nat Mater 4:612
10. Hochrein O, Kniep R, Zahn D (2005) Chem Mater 7:1978 31. Simon P, Zahn D, Lichte H, Kniep R (2006) Angew Chem Int Ed
11. Zahn D, Hochrein O (2006) Z Anorg Allgem Chem 632:79 45:1911
12. Hauptmann S, Duffner H, Brickmann J, Kast SM, Berry RS
(2003) Phys Chem Chem Phys 5:635

123
J Mater Sci (2007) 42:89748985
DOI 10.1007/s10853-007-1642-6
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Modifications of spider silk sequences in an attempt to control
the mechanical properties of the synthetic fibers
Florence Teule William A. Furin Alyssa R. Cooper
Joshua R. Duncan Randolph V. Lewis
Received: 5 December 2006 / Accepted: 26 February 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract Bacteria were genetically engineered to pro-
duce two spider silk protein variants composed of basic
repeat units combining a flagelliform elastic motif
([GPGGX]4) and a major ampullate silk strength motif
([linker/poly-alanine]. The secondary structures of the pure
recombinant proteins in solution were determined by cir-
cular dichroism. The data presented suggest that the nature
of the 5th and 10th amino acid (X) in the [GPGGX]2 elastic
motif and temperature have an impact on the amount of b-
sheet structures present in the proteins. More specifically,
increasing temperatures seem to be positively correlated
with b-sheet formation for both proteins and this state is
irreversible or reversible when both X (5th and 10th) in the
elastic motif are hydrophilic or hydrophobic respectively.
Moreover, each pure silk-like protein was able to sponta-
neously self-assemble into films from aqueous solutions.
Two kinds of synthetic fibers were made by pulling fibers
from these preassembled films as well as spinning fibers
from each protein resolubilized in HFIP. The mechanical
data show that the pulled fibers are far tougher than the
spun fibers suggesting a better fiber organization.
Electronic supplementary material The online version of this
article (doi:10.1007/s10853-007-1642-6) contains supplementary
material, which is available to authorized users.
F. Teule (&)
W. A. Furin
A. R. Cooper

J. R. Duncan
R. V. Lewis
Department of Molecular Biology (Dept. 3944),
University of Wyoming, 1000 E. University Avenue,
Laramie, WY 82071, USA
e-mail: fteule@uwyo.edu
123
Introduction
Spider silks have great potential to provide a new genera-
tion of bio-based materials for applications ranging from
medical (micro-sutures, artificial ligaments, tendons, and
drug-delivery coatings) to military (body armor, light
weight gear) to civilian (textiles). In order to custom design
and produce biomaterials, a critical step is to relate struc-
ture to function in these fiber proteins. For this purpose,
spider silks are the model of choice because of the wide
variety of silks spun, their different uses, and thus their
differences in mechanical properties.
Of all the orb weaver-spun silks, the major ampullate
silk, or dragline silk, from Nephila clavipes has been most
extensively studied because of its extreme strength (4 GPa)
[1, 2] and moderate elasticity (35%) [2, 3] resulting in an
impressive toughness (160 MJ m3) [3, 4]. Overall, this
performance is believed to be dictated by the silk proteins
primary structures, more specifically, by the presence and
combination of key amino acid motifs that have previously
been identified. Crystalline forming motifs such as poly-
alanine (A)n (n 9) present in the highly repetitive MaSp2
protein, one of the two proteins forming the major ampul-
late silk [5, 6], are thought to confer strength to the fiber.
Structural studies have shown that in the fiber, these poly-
alanine motifs form crystalline b-sheets aligned parallel to
the fiber axis [7, 8]. Another silk of interest is the extremely
elastic flagelliform silk, also called viscid silk, which is
composed of a single highly repetitive protein containing no
poly-alanine motifs [9, 10]. The basic flagelliform protein
(Flag) repeat unit has been characterized, and it is mainly
composed of a (GPGGX)n motif (43 n 63, and usually
X = A/V, or Y/S) juxtaposed to a (GGX)n motif (n = 12)
and a spacer sequence containing non-traditional silk amino
acids [9, 10]. These (GPGGX)n motifs are believed to adopt
J Mater Sci (2007) 42:89748985
b-turn structures, and a series of these motifs would form
spring-like spirals, or nanosprings, conferring elasticity to
the fiber [1113]. Structural studies of the recombinant
polypeptide 1 showed that its sequence, [(GPGGS
GPGGY)2 GPGGK]11, is composed of type II b-turns [13].
The central (PG) of the (GPGGX) consensus motif, also
present in several elastomeric proteins such as elastin
[1416], and glutenin [17], has been found to be favorable
for the formation of type II b-turns [18, 19]. Thus, the
mechanism of silk elasticity is thought to resemble the one
described for these b-turn forming polypeptides and seems
to be entropically driven [3, 13]. Therefore, the extremely
high number of (GPGGX)n motifs in the Flag protein is
believed to be responsible for the extraordinary elasticity
measured for the flagelliform silk (200%) [1, 2]. This
argument is reinforced by the fact that a lower amount of
this motif in the dragline protein MaSp2, and its absence in
the minor ampullate silk proteins Misp1 and Misp2, seem to
result in much lowered elasticity of these two silk fibers [6].
We have engineered and produced recombinant repeti-
tive silk-like proteins with basic repeats containing variants
of the elastic (GPGGX) motifs found in the Flag protein,
either (GPGGA)4 (=A1 motif) or [(GPGGY) (GPGGS)]2
(=Y1 motif) combined with a strength motif [Linker-(A)8]
(=S8 motif) found in the dragline silk MaSp2 protein. We
have produced and purified the two proteins (A1S820 and
Y1S820), confirmed their primary structures by amino acid
analyses, characterized their secondary structures using
circular dichroism, and generated two types of fibers from
both proteins either from aqueous solutions or organic
solvents. The synthetic fibers were mechanically tested to
give us an evaluation of the performance resulting from the
combination of either of the Flag protein elastic motifs, and
the dragline silk protein strength motif.
Materials and methods
Gene construction and cloning in pBluescriptII SK+
in E. coli
Three sets of complementary synthetic oligonucleotides
(roughly 60 bp each, Electronic Supplementary data or
ESM, Fig. 1) were designed (Dr M. B. Hinman): (1) flagel-
liform putative elastic encoding motifs A1 and Y1 (coding
respectively for (GPGGA)4 and (GPGGSGPGGY)2), (2) the
dragline silk putative strength encoding motif S8 (coding for
a linker-polyalanine: (GGPSGPGS(A)8)). These three types
of synthetic oligonucleotides (Y1, A1 and S8) were synthe-
sized and assembled as double stranded cassettes each
cloned in pBluescriptII SK+ (Stratagene) at the Hind III/
Sma I sites (Midland Certified Reagents Inc, Texas). These
double stranded sequences were engineered with a 5Xma I,
8975
and a 3BspE I restriction sites allowing the use of a previ-
ously described multimerization or doubling strategy based
on compatible but non-regenerable restriction sites [20]. The
two synthetic genes constructed by successive doubling of
each different basic repeat unit were called (A1S8)20 and
(Y1S8)20. These sequences were cloned in a pBluescriptII
SK+ plasmid and used to transform electrocompetent XL1-
Blue cells (Stratagene). The recombinant plasmids contain-
ing the final (A1S8)20 and (Y1S8)20 silk-like sequences
(2,300 bp each) were partially sequenced on both strands
using vector specific primers.
Cloning of the silk-like sequences in the pET19b
(KanR) expression vector
The pET19b (KanR) expression vector was previously
obtained by replacing the Bpu1102/AlwNI fragment con-
taining the ampicillin resistance gene in pET19b (AmpR)
(Novagen) by the Bpu1102/AlwNI fragment containing the
kanamycin resistance gene (KanR) from pET26b ()
(Novagen) (Dr M. B. Hinman, unpublished data).
The plasmid DNAs of the bacterial pBluescript clones
containing the confirmed (A1S8)20 and (Y1S8)20 silk-like
sequences, as well as the pET19b (KanR) expression vec-
tor, were isolated using an alkaline lysis protocol [21].
The expression vector and each recombinant silk plas-
mid were subjected to a Bam HI restriction enzyme
digestion (New England Biolabs Inc.). The totality of the
restriction enzyme digestion sample was subjected to
electrophoresis in a 0.8% agarose gel. The silk inserts and
the pET19b (KanR) vector were purified by electroelution
following standard protocols [22]. The DNA fragments
were then recovered by addition of ammonium acetate to a
final 1.5 M concentration, followed by the addition of two
volumes of cold ethanol. The samples were placed at
80 C for 45 min and the purified DNA fragments were
recovered by centrifugation at 18,000g for 25 min at room
temperature. The supernatant was discarded and each DNA
pellet was dried and resuspended in 30 lL of TE buffer
(10 mM TrisHCl/1 mM EDTA, pH 8).
Each purified Bam HI silk fragment was ligated to the
purified Bam HI pET19b (KanR) expression vector using
T4 DNA ligase following the recommendations of the
Manufacturer (Promega). Each ligation reaction was used
to transform electrocompetent E. coli XL1-Blue cells
(Stratagene). The plasmid DNA from the recombinant
clones were extracted by alkaline lysis and characterized
by restriction enzyme digestion with Bam HI (New Eng-
land Biolabs Inc.). These plasmids were also subjected to a
restriction enzyme digestion with Xma I (New England
Biolabs Inc.) to verify the inserts orientation. The two
kinds of recombinant pET clones displaying the (A1S8)20
or (Y1S8)20) silk insert in the right orientation were
123
8976
sequenced partially on both strands. The selected clones
were named pE(A1S8)20x and pE(Y1S8)20x.
Once the silk insert sequence was confirmed, these two
selected recombinant plasmid DNAs were introduced by
electroporation in the E. coli cell line BL21 (DE3) (Nov-
agen) for expression purposes. The recombinant clones
were once again characterized by restriction enzyme
digestion with Bam HI (New England Biolabs Inc.) to
confirm the presence of the silk inserts. The selected
recombinant plasmids containing the silk insert in the right
orientation were named pET(A1S8)20x and pET(Y1S8)20x.
These plasmids were partially sequenced on both strands.
Gene expression studies
The pET recombinant clones in BL21 (DE3) cells
(pET(A1S8)20x and pET(Y1S8)20x) were cultured overnight
in 10 mL of LB containing 50 lg/mL kanamycin in a
shaking incubator at 37 C. The overnight culture was used
to inoculate 1 L of LB containing 50 lg/mL of kanamycin.
The cultures were induced by addition of 1 mM IPTG
(Isopropyl-b-D-thiogalactopyranoside; Biochemica &
Synthetica, Switzerland), when the optical density at
600 nm (OD600) of the culture had reached 0.60.8. The
cells were then harvested by centrifugation at 5,300g/25 min
about 34 h after induction. The media were discarded, and
the cell pellets were washed once in sterile distilled water.
After a second centrifugation at 3,300g for 15 min, and
removal of the supernatant, the masses of the cell pellets
were calculated and the samples were stored at 80 C.
E. coli cell lysis
The cell pellets were resuspended in 1 lysis buffer
(50 mM TrisHCl pH 8, 10 mM MgCl2, 10 mM NaCl) at
3 mL of buffer per every gram of cell. Lysozyme (Sigma)
was added to each sample to a final concentration of
0.2 mg/mL and the samples were incubated on ice for
30 min swirling periodically. At this stage, 1 mM PMSF
(phenylmethylsulphonylfluoride; Sigma) was added to the
lysates to prevent protein degradation. Then 1.5 g of
deoxycholic acid (MP Biomedicals LLC) were added per
gram of cells and the lysates were incubated 20 min at
37 C. At this point, 0.02 mg of DNAse I (Sigma) was
added to the cell lysates per gram of cell and the samples
were incubated at room temperature for 30 min on a
platform shaker. The lysates were then subjected to cen-
trifugation at 3,300g for 15 min to pellet the cellular
debris. After a heat-treatment at 80 C for 10 min, the
recovered lysates, or supernatants, were again subjected to
centrifugation at 3,300g for 15 min to pellet the denatured
proteins. The cleared heat-treated lysates were stored at
80 C until use for protein purification.
123
J Mater Sci (2007) 42:89748985
Protein purification
Sample preparation
The heat-treated cell lysates containing the recombinant
His-tagged silk proteins were diluted in 1 binding buffer
(20 mM TrisHCl pH 8/0.5 M NaCl/5 mM imidazole) at a
1:4 ratio (v/v).
IMAC (immobilized metal ion affinity chromatography)
Each sample was loaded on a 10 mL polyprep chroma-
tography column (Biorad) containing 2 mL (resin bed
volume, BV) of Ni-NTA His
Bind resin (Novagen) previ-
ously washed once with 10 BV of water, and equilibrated
with 5 BV of 1 binding buffer. The proteins were eluted
using an imidazole step gradient. The concentrations of
imidazole of the 1 binding buffer used in the successive
washes were adjusted to 20 mM, then 40 and 50 mM. The
recombinant His-tagged silk proteins were eluted when
using imidazole concentrations equal or greater to 60 mM
(80 and 250 mM were best). The resin was stripped of the
nickel ions using a 1 strip buffer (100 mM EDTA, 0.5 M
NaCl, 20 mM TrisHCl pH 8). This strip fraction along
with the other fractions were collected and saved for
analyses.
Protein characterization
SDS-PAGE analyses
The heat-treated protein extracts and the purified protein
fractions were analyzed by SDS-PAGE. For all SDS-PAGE
analyses, 4% stacking and 10% separating polyacrylamide
gels containing 0.1% SDS were made in TrisHCl buffers
(Mini PROTEAN3 Cell protocol for SDS-PAGE buffer
system, Biorad). The composition of the sample buffer and
the 5 electrode buffer used are the ones recommended by
the manufacturer (Biorad). Typically, 10 lL of protein
sample were subjected to SDS-PAGE analysis. A Kalei-
doscopeTM Prestained Standards (Biorad, 10 lL) or a
Precision Plus Protein Dual Color Standard (Biorad, 8 lL)
were used as a molecular weight marker. All electrophor-
eses were performed using the Mini PROTEAN3 Cell
(Biorad) at a constant voltage of 80 V.
Staining of polyacrylamide gels
After SDS-PAGE analysis, the gels were stained with
Coomassie Brillant Blue (R-250) dye according to the
method published [23]. The stained gels were placed in a
10% glycerol solution for 1 h before being dried between
two sheets of Ultra Clear Cellophane (Research Products
J Mater Sci (2007) 42:89748985
International Corp., Mount Prospect, Illinois) using a
plexiglass frame.
Western blot analyses
The proteins samples separated by SDS-PAGE were
transferred to a PVDF/ImmobilonTM P transfer membrane
(Millipore) by electroblotting using the Mini Trans-Blot
Electrophoretic Transfer Cell (Biorad). Blots were set up as
specified by the manufacturer (Biorad). All transfers were
performed overnight at room temperature under a constant
current of 25 mA. After fixation of the proteins, the
membranes were subjected to Western blot analyses using
the His
Tag antibody (Novagen) directed against the
(histidine)10 tag ((His)10) present in the amino terminus of
the fusion proteins. The secondary antibody used was the
Goat Anti-Mouse IgG HRP conjugate (H + L) (Novagen).
All Western blot analyses were performed according to the
protocols described by the manufacturers. Chemilumines-
cent detections were performed using the ECLTM Western
Blotting Detection Reagents (Amersham Biosciences). The
membranes were then wrapped in plastic wrap and placed
in a light proof cassette against a high performance
chemiluminescence film (HyperfilmTM ECL, Amersham
Biosciences) to visualize the proteins of interest.
Amino acid analyses
The purified protein samples were dialyzed against 5 mM
ammonium bicarbonate and lyophilized. The samples were
treated using the Waters AccQ
TagTMMethod for hydro-
lysate amino acid analysis (Millipore) and then subjected to
HPLC using a Hitachi LC 6500. The amino acid compo-
sitions were expressed in mole % (ESM, Table 1).
Circular dichroism (CD)
For CD, the purified proteins samples were dialyzed
extensively against 5 mM TrisHCl pH 8 or 0.1 PBS
(Phosphate Buffered Saline: 13.7 mM NaCl/0.27 mM KCl/
1 mM Na2HPO4/0.18 mM KH2PO4) five times and sub-
sequently concentrated by dialysis against a 20% PEG
(polyethylene glycol)/5 mM TrisHCl pH 8% or 20%
PEG/0.1 PBS using a dialysis tubing with a smaller
molecular weight cut-off (MWCO: 3,500 Daltons). The
final CD samples concentrations were estimated by BCA
assay (Pierce) using a BSA standard as well as home
made standards of A1S820 and Y1S820 proteins (2 mg of
each pure lyophilized A1S820 or Y1S820 protein were
resolubilized in 1 mL of 5 mM TrisHCl pH 8 or 0.1
PBS). The concentrations of the pure soluble A1S820
dilutions used for CD were 0.98 mg/mL in TrisHCl pH 8
and 0.35 mg/mL in 0.1 PBS. The concentrations of the
8977
pure soluble Y1S820 dilutions used for CD were 0.75 mg/mL
in TrisHCl pH 8 and 0.54 mg/mL in 0.1 PBS. The CD
spectra were recorded on a JASCO-810 spectropolarimeter
using the Spectra Manager for Windows 95/NT software
(Version 1.18.00). A total of 8 spectra accumulations were
recorded from 185 nm to 260 nm with a resolution of
0.1 nm and a path length of 0.01 cm. Melt and anneal
experiments were conducted successively from 0 C to
85 C and 85 C to 0 C at 5 C intervals. The data
obtained were submitted to the following online site for
analyses: www.cryst.bbk.ac.uk/cdweb/html/home.htwl [24].
We used the SELCON 3, CONTIN and CDSSTR methods
for protein secondary structure predictions (see [24] for full
references).
Production of synthetic fibers
Two methods of fiber production were used for both types
of synthetic proteins (Fig. 1). The first one relied on the
natural ability of these purified synthetic fibers to sponta-
neously form fibers in aqueous solutions, and the second
relied on the wet spinning/extrusion of a silk dope made in
an organic solvent.
Hand pulled fibers
The Y1S820 pure protein spontaneously formed an oily
looking film at the surface of the elution or strip solutions
collected in a glass dish, thus naturally separating from the
solution. We used forceps to pull the edge of the Y1S820
film that lifted into single fibers (Fig. 1a). While A1S820
did not spontaneously form a film at the surface of the
elution or strip fractions collected, we noticed a film-like
structure at the bottom of the dish. Thus shaking vigorously
Fig. 1 Synthetic fiber production. (a) Hand pulled fibers from the
protein pure fraction: the fibers were hand pulled from a self-
assembled film using forceps. (b) Wet spinning of fibers using an
extruder/spinneret: the spinning dope loaded in a syringe was placed
in a computer controlled spinneret and extruded at a constant speed
into a coagulation bath. The extruded fibers were recovered using
forceps
123
8978
the dish to mix the contents resulted in the surfacing of
preformed A1S820 films that could then be pulled into
fibers using forceps (Fig. 1a).
Silk dope preparation and fiber wet spinning
The pure lyophilized proteins were resolubilized in
1,1,1,3,3,3-hexafluoro-2-propanol (HFIP; TCI America,
Portland OR) to make a 2530% (w/v) spinning dope.
The fibers were extruded using a DACA Instruments
SpinLine system (DACA Instruments, Santa Barbara CA)
provided by Nexia Biotechnologies (Montreal, Canada).
The spinneret was composed of a 1 mL Hamilton Gas-
tight syringe (Hamilton company) mounted with a
10 cm long red PEEK tubing (Upchurch Scientific)
with an internal diameter of 0.127 mm. The syringe was
loaded with the silk dope and placed in the barrel of the
extruder. Prior to spinning, we added about 15% of H2O
to the dope and slowly mixed it with the dope in the
syringe. The silk fibers were extruded at a plunger speed
of 0.6 mm/minute into a 90% isopropyl alcohol coagu-
lation bath (Fig. 1b). The extruded synthetic fibers were
collected from the bath with forceps, cut and mounted on
testing cards.
Mechanical testing and data analyses
All fibers (pulled and spun) were mounted on testing
cards. The initial length of the fiber to be tested was
15 mm. Three to five pictures of each synthetic fiber were
taken along the fiber axis under a light microscope using
the 20 and 40 magnifications. The average diameters of
the fibers were calculated after the pictures were imported
into the ImageJ software prior to testing. We tested all the
fibers produced using a custom made 10 g load cell
(Transducer Techniques, Temecula CA) mounted on a
MTS Synergie 100 system (MTS Systems Corporation,
Eden Prairie MN). The mechanical data was recorded at a
strain rate of 5 mm/min using a frequency of 250 MHz.
Parameters such as load and elongation were recorded and
allowed the successive calculation of the engineered
stresses and strains. The stress/strain curves as well as the
polynomial regressions of the curves necessary to calcu-
late the toughness (or energy to break/work of rupture),
Youngs Modulus (stiffness = initial slope of the curve),
and determination of parameters such as maximum stress
and maximum extension (=maximum % strain), were
performed using MATLAB (Version 7.1). Any mechani-
cal data obtained having a polynomial fit for the data
points with an R2 0.9 was eliminated from the study.
The best fits for the data points recorded were obtained
using polynomial regressions of 8th or 9th order.
123
J Mater Sci (2007) 42:89748985
Results and discussion
Gene expression and protein production
We used sequences encoding flagelliform silk-like putative
elastic motifs (A1 or Y1) and a dragline silk putative
strength motif (S8) to build two spider silk-like genes.
These three synthetic spider silk-like sequences were used
to build two types of basic repeat units: (A1S8) and
(Y1S8). The only difference between the (A1S8) and
(Y1S8) basic repeat units is the nature of the amino acids
present in the 5th and 10th positions in the [GPGGX]4
elastic motif. More specifically, the elastic motif encoded
by the A1 sequence is [(GPGGA) (GPGGA)]2 whereas the
one encoded by the Y1 sequence is [(GPGGY) (GPGGS)]2.
Such motifs naturally occur in the native flagelliform
sequence [10]. However, in the putative elastic motif of the
native flagelliform silk protein, when there is an alanine
(A) in 5th position, there is usually a valine (V) in the 10th
position. We chose to alternate alanine (A) in the 5th
position with alanine (A) in the 10th position, as this motif
is also present in the native flagelliform putative elastic
motifs, and since this change does preserve the hydro-
phobic nature of the amino acid in the 10th position.
We used the doubling strategy described earlier [20] to
rapidly increase the sizes of these two kinds of silk basic
units. The spider silk-like constructs were sequenced after
each doubling experiment to avoid the selection of muta-
tions that could occur during cloning. The final sizes of the
synthetic silk repeats were 2,300 bp for both (A1S8)20 and
(Y1S8)20. These two spider silk-like constructs, initially
built in the pBluescriptII SK vector, were successfully
cloned in the pET19b (KanR) expression vector and intro-
duced in the E. coli cell line BL21 (DE3) for expression:
clones pET(A1S8)20x and pET(Y1S8)20x. The cell lysates,
or protein extracts, were heat-treated at 80 C for 10 min as
a first purification step. Spider silk-like proteins are usually
fairly heat-stable, and such heat-treatments were first
investigated as initial steps (95 C/5 min) in the purifica-
tion of spider silk-like proteins produced in plants [25].
This heat-treatment was also used successfully as a first step
during the purification of MaSp 1-like homopolymer and
copolymers produced in yeast (Pichia pastoris) [26]. In this
case, during heat-treatment, most of the E. coli native
proteins were denatured and could be eliminated by cen-
trifugation. All of the recombinant spider silk-like proteins
produced in E. coli are fusion proteins possessing a histi-
dine tag ((His)10) in their carboxyl terminus facilitating
purification by ion metal affinity chromatography. The
primary structures, or amino acid sequences, of the two
spider silk-like recombinant proteins A1S820 (57.64 kDa)
and Y1S820 (61.96 kDa) that were produced in E. coli are
shown in Fig. 2a.
J Mater Sci (2007) 42:89748985
(a)
MG(H)10SSGHIDDDDKHMLEDPP-[A1S820 or Y1S820 silk repeat]20-EISGSGC
Silk repeats:
A1S820: [(GGAGPGGAGPGGAGPGGAGP)1 (GGPSGPGSAAAAAAAAGP)]20 57.64 kDa
Y1S820: [(GGYGPGGSGPGGYGPGGSGP)1 (GGPSGPGSAAAAAAAAGP)]20 61.96 kDa
(b)
1 2 3 4 5 6 7 8
250
150
100
76
50
37
25
Fig. 2 Primary structures and Western Blot analyses of the Y1S820
and A1S820 recombinant proteins. (a) Sequences and molecular
weights of the spider silk-like proteins produced in E. coli. The
molecular weights of the fusion proteins are indicated next to each
silk repeat. (b) SDS-PAGE analysis of the purification of A1S820 and
Y1S820 from heat-treated extracts (stained with Coomassie Brillant
Blue). (c) Western blot analysis of (b) using the His
Tag antibody as
primary antibody and the Goat Anti-Mouse IgG HRP conjugate as
Protein purification and characterization
The A1S820 and Y1S820 proteins were easily purified with
classic metal affinity chromatography and were recovered
using as low as 60 mM imidazole in the elution buffer. The
SDS-PAGE and Western Blot analyses of the purification
of Y1S820 and A1S820 are shown in Fig. 2 (b and c). The
two pure full-size proteins were visible when stained with
Coomassie blue (Fig. 2b). Note that on the Western blot
analyses, we can see both full-size proteins, as well as
truncation products only for Y1S820 (Fig. 2c). Truncated
proteins are likely the direct result of translation from a
prematurely terminated transcript and are commonly
observed for native [3, 10] or recombinant spider silk
proteins [20, 2730]. After expression, heat-treatment and
affinity chromatography purification, we were able to
recover of 710 mg/L of pure A1S820 or Y1S820 proteins.
The results of the amino acid analyses of the purified
A1S820 and Y1S820 proteins confirmed their identities and
their high purity (ESM, Table 1).
While purifying Y1S820 and A1S820, we observed the
formation of fibers at the bottom of the affinity chroma-
tography columns as the elution or strip fractions were
being collected (Fig. 3a and b). Samples of these in situ
formed fibers were collected on a slide and observed under
a light microscope (Fig. 3, c through e). It looked as though
the proteins were forming a film at the periphery of an
extremely viscous droplet. These fibers were drawn out of
the column by the weight of the droplet that dangled at the
8979
(c)
9 1 2 3 4 5 6 7 8 9
secondary antibody. In (b) and (c): Lane 1 is a Molecular weight
marker; Lanes 25 show the A1S820 purification; Lanes 69 show the
Y1S820 purification; Lanes 2 and 6 show the unbound protein
fractions; Lanes 3 and 7 show the 40 mM imidazole washes; Lanes 4
and 8 show the 50 mM imidazole washes; Lanes 5 and 9 show the
eluted/strip fractions. In both (b) and (c), the blue and green arrows
point at the purified A1S820 and Y1S820 proteins respectively
bottom of the forming fiber thus allowing the film at the
periphery of the droplet to fuse into a single fiber (Fig. 3,
d and e). Moreover, we determined that the aggregation of
these proteins was concentration dependent as dilution of
the original heat-treated protein extract loaded on the col-
umn resulted in no such fiber formation. We believe that
fiber formation is also purity dependant as these fractions
were extremely pure. It is important to stress the fact that
these fibers were able to form in an aqueous solution close
to physiological pH rather than in harsh solvents. More
importantly, these aqueous solutions might help preserve
the original secondary structures of the recombinant silk
proteins that might be critical for the resulting mechanical
properties of the fiber.
After further observation, we noticed that the eluted
Y1S820 protein could self-assemble to form a very uniform
film with an oily appearance that floated at the surface of
the purified protein fraction. A major difference in behavior
between the two pure proteins was that A1S820 did not
form an apparent film at the surface of the pure protein
fraction like the one observed for Y1S820 but would rather
form a viscous layer at the bottom of the dish covered by a
more fluid layer. However, by vigorously shaking the dish
containing the pure A1S820 fraction, we were able bring
the viscous layer to the surface as a somewhat broken up
film. We used forceps to pull each type of films made by
these two proteins into fibers (Fig. 1a) and mounted the
pulled fibers on testing cards for mechanical testing. The
film formed by Y1S820 could be reeled in and be wound
123
8980
Fig. 3 Pictures of the in situ
fibers formed on the column and
their observation of under a
light microscope. (a) The
different stages of Y1S820 fiber
formations are shown (a through
d) while recovering the
extremely pure fraction (strip
fraction) from the affinity
chromatography columns.
(b) Close up on the fiber formed
(40). (c), (d), and (e) are
pictures of the observations
under a light microscope of the
types of fibers collected from
the column and mounted on a
slide. (c) shows a Y1S820 fiber
(40), (d) shows an A1S820
fiber (20), (e) is an
enlargement (40) of the part of
the fiber in (d) indicated by the
square box
around a tube very easily without breaking. The fibers
pulled from the A1S820 small films were shorter, however,
limited by the size of the broken film pieces.
Structural analyses and hypotheses correlating
secondary structure and self assembly
Circular dichroism was used to determine the secondary
structures of the A1S820 and Y1S820 silk-like proteins in
aqueous solutions. The CD spectra of both melting (from
0 C to 85 C) and successive annealing (from 85 C to
0 C) for both proteins in the 5 mM TrisHCl pH 8 are
shown in Fig. 4. We chose to use a TrisHCl buffer pH 8
for CD analyses since we observed protein self-assembly in
the elution and strip buffers containing TrisHCl pH 8.
Even though these two proteins only differ by two amino
acids (in the 5th and 10th positions in the [GPGGX]4
repeat), there were substantial differences in the secondary
structures observed for both proteins.
For A1S820, the CD analyses show that at 0 C, the
protein is mostly unordered (about 76.30% and 64.26% of
random coils observed respectively in TrisHCl pH 8 and
PBS). Random coils are noticeable in the melt profile of
A1S820 in TrisHCl (Fig. 4, top left) by the presence of
a lower maximum near 200 nm. As the temperature
increases, the SELCON 3, CONTIN and CDSSTR analyses
of the CD spectra for the A1S820 protein in TrisHCl show
a slight increase in helices (to a maximum of 10%).
Although there seems to be a slight increase in sheets and
turns, the percentages of these two species at higher tem-
peratures indicated by the SELCON 3 and CONTIN
methods are substantially lower than indicated by the
CDSSTR method. Indeed, for both SELCON3 and CON-
TIN, the results of the analyses for A1S820 in TrisHCl
123
J Mater Sci (2007) 42:89748985
from 65 C to 85 C show an average of 6.33% sheets,
9.10% turns, 8.30% helices, and 77.68% random coils
while the ones obtained for the same temperature bracket
with CDSSTR show 21.20% sheets, 17.80% turns, 8.60%
helices, and 53.00% random coils. From the CD spectra of
the A1S820 melt presented in Fig. 4, we can confirm that
there is indeed an increase in sheets and turns (see the
increase at 200 nm and the decrease at 220 nm for the
sheets, and the increase at 210 nm for the turns). However,
there is still a high content of random coils (lower maxi-
mum before 200 nm) thus we feel that the results given by
SELCON 3 and CONTIN seem most probable for the
A1S820 protein. The same trends were observed for this
protein in PBS (ESM, Fig. 2). The successive annealing of
the A1S820 protein from 85 C to 0 C clearly shows that
the heat induced changes in secondary structure observed
during the melting are almost totally reversible (Fig. 4, top
right).
For Y1S820, the CD analyses show that at 0 C in Tris
HCl, the protein is mostly unordered (about 55% in both
buffers) but it contains less random coil than was observed
for A1S820 (Fig. 4, bottom left). According to the melt
spectra and the results of the SELCON 3, CONTIN, and
CDSSTR analyses however, there are substantial changes
in the secondary structures when the temperature increases
to 85 C. For Y1S820, in the melt profiles shown in Fig. 4,
the maximum peak observed at 200 nm accompanied by a
minimum at 220 nm are indicative of a noticeable increase
in sheets (up to 31.32%), and the maximum peak observed
at 210 nm (broadened by the sheet peak occurring at
200 nm) is indicative of an increase in turns (about 20%)
that is visible as soon as the temperature reaches 15 C.
The same trend was observed for this protein in PBS (ESM,
Fig. 2). The successive anneal experiment for Y1S820
J Mater Sci (2007) 42:89748985 8981

10
A1S820- Melt 0 A1S8 20- Anneal
5
0 -5
-5 -10

CD (m deg)
CD (m deg)

-10
-15
-15
-20 -20
-25 -25
-30
-30
-35
-40 -35
190 200 210 220 230 240 250 260 190 200 210 220 230 240 250 260
Wavelength (nm) Wavelength (nm)

Y1S820- Melt Y1S820- Anneal

5 10
0
5
CD (mdeg)

CD (m deg)
-5
0
-10
-5
-15
-10
-20
-15
190 200 210 220 230 240 250 260 190 200 210 220 230 240 250 260
Wavelength (nm) Wavelength (nm)

Fig. 4 Circular dichroism spectra for A1S820 and Y1S820. Melt the scale from one graph to another is not the same for the CD [mdeg]
(from 0 C to 85 C) and successive anneal (from 85 C to 0 C) units (vertical axes)
spectra obtained for both proteins in 5mM TrisHCl pH 8. Note that

(Fig. 4, bottom left) shows that once the secondary struc-
tures have formed during the melt, there is no reversion to
the initial structure of the protein.
According to this data, for Y1S820, higher temperatures
promote irreversible sheet and turn formation through
hydrogen bonding that remain stable. Such irreversible
temperature-induced b-sheet formation has also been
observed in melt experiments of major and minor ampul-
late silk gland contents [31]. These structures for A1S820,
though less pronounced, are reversible. Moreover, the fact
that the amount of turns increase and remain stable for
Y1S820 may indicate that the [(GPGGY)(GPGGS)]2 motif
is able to adopt the b-turns structures proposed earlier for
this motif [6] and may be stabilized by intra-molecular
hydrogen bonds involving tyrosine (Y) and serine (S). We
hypothesize that by achieving a better organization or
folding of this latter motif, the protein might be able to
better self organize, thus allowing the poly-alanine seg-
ments to interact with one another and lock into b-sheets by
a nucleation process. This nucleation process may initiate
the self-assembly of the Y1S820 molecules into a supra-
molecular structure. The fact that we do get spontaneous
self-assembly of the pure Y1S820 proteins that were
subjected to a heat-treatment prior to purification reinforces
this argument (Fig. 3). Moreover, changing the hydrophilic
5th (Y) and 10th (S) residues and replacing them by
hydrophobic residues such as alanine (A) in A1S820
deprives the protein of the ability to retain large amounts of
stable turn and sheet structures even when tested at a higher
concentration than Y1S820. This may be the reason shear
forces are needed to force film formation from a liquid
crystalline phase from the pure A1S820 fractions.
Regarding the native Flag protein in the flagelliform
gland, by having both [GPGGY GPGGS] alternating with
[GPGGA GPGGV/A], the protein may have parts of its
elastic segments in a stable b-turn conformation, possibly
under salt and pH control as suggested for other silk glands
[32, 33], while the rest of the elastic components are
unfolded, or more disordered, thus allowing the molecules
to remain in a liquid crystalline phase avoiding premature
aggregation in the gland. Although in the Flag protein there
is no poly-alanine segments [6, 9, 10], the GGX repeats are
known to promote self-assembly in proteins such as lam-
prin [34] and might play the same role as the poly-alanine
segments, or poly-(glycine/alanine) in other silks, in initi-
ating the self-assembly of the Flag molecules. A film

123
8982
similar to the ones seen in the A1S820 and Y1S820 pure
fractions may self-assemble in the spiders silk glands.
Then, shear forces resulting from a combination of the
spiders pull on the fiber and the extremely small size of the
spinning duct the proteins travel through might be enough
to induce the proper folding of the [GPGGA GPGGV]
motifs. By locking all its [GPGGX]n motifs into the proper
secondary structures, the rest of the molecules would then
be able to self organize and assemble into a fiber. Moreover
in our case, by pulling the structure together, and having
properly folded motifs, the (GPS)2 linker sequence directly
preceding the poly-alanine segments might be able to form
stable intermolecular hydrogen bonds that would stabilize
the supramolecular structure. In the native Flag protein, we
can hypothesize that this stabilizing role of the supramo-
lecular structure may be fulfilled by the spacer region,
positioned next to the (GGX)12 sequence, containing both
highly hydrophobic and highly hydrophilic amino acids.
Recent molecular dynamic simulation studies on the
(GVPGV)7 elastin pentapeptide suggest that such a
sequence has a higher propensity for PPII (poly-proline II)
structures since they cannot stabilize any turns through
hydrogen bonding [35]. It also retains a high backbone
hydration level that would constrain these peptides in dis-
ordered conformations that are unable to exclude water
when aggregated, thus providing elastomeric properties to
the matrix formed [35]. Such a model might be applicable
to the more hydrophobic elastic motifs found in flagelli-
form silk as well as the elastic motif of the A1S820 protein.
Moreover in elastins, glycine-rich repeats deprived of
Fig. 5 Stress/strain curves of
the synthetic fibers. P1-
P12 = Pulled fibers 1 through
12; S1-S12 = Spun fibers 1
through 12. Note that the scales
for both stress and strain differ
from one graph to another
123
J Mater Sci (2007) 42:89748985
proline seem to promote self aggregation in an amyloido-
genic fashion [35], thus reinforcing the argument stated
above about the role of GGX sequences in the self-
assembly of the native Flag protein.
Mechanical performances of the synthetic pulled and
spun fibers
Using an extruder, we were able to spin fibers out of both
the A1S820 and Y1S820 proteins that were resolubilized in
100% HFIP. However, we determined that adding 15%
water to the dope prior to spinning dramatically improved
the mechanical properties as well as the appearance of the
fibers. This can be explained by the fact that although HFIP
might promote intramolecular hydrogen bonding [36, 37],
water was necessary to force the hydrophobic poly-alanine
segments away from the water phase and into proper sheet
structures, thus initiating self-assembly. The stress/strain
curves for A1S820 and Y1S820, both pulled (P) and spun
(S) are shown in Fig. 5 while fiber specifics (number of
fibers used and diameter) and mechanical performance data
(maximum stress, maximum extension, Youngs modulus,
and toughness) are shown in Table 1. Before tensile test,
pictures of each type of fibers used in this study were taken,
and the pictures of the fibers displaying the best extension,
best maximum stress, and average maximum stress are
shown in Fig. 6. According to our data, the pulled fibers,
which had a much smaller diameter on average than the
spun ones, outperformed the spun fibers in average maxi-
mum extension, maximum stress and toughness (Table 1
J Mater Sci (2007) 42:89748985 8983

Table 1 Mechanical testing data

Fiber type Total of fibers Diameters (lm) Youngs modulus (MPa) Maximum stress (MPa) Maximum extension (%) Toughness (MJ/m3)

A1S820 P 15 12.20 4.99 1,706.8 791.87 28.64 8.41 18.99 12.88 3.41 2.61
A1S820 S 19 32.15 16.24 759.68 540.27 28.58 17.18 3.72 1.24 0.464 0.30
Y1S820 P 31 15.79 6.05 1,081.49 1,000 49.64 19.35 34.06 25.30 10.6 10.2
Y1S820 S 18 28.4 11.32 933.62 727.14 10.21 7.32 1.59 1.03 0.089 0.11
Average values measured for all types and kinds of synthetic fibers. The standard deviation of each value is indicated ( STD). P = pulled;
S = spun

and Fig. 5). The appearance of these pulled fibers was also
strikingly different than the spun ones (Fig. 6). Not only
were their surfaces smoother, but they were also more even
and had greater sheen than any of the spun fibers. This can
be explained by the fact that the action of pulling the
preassembled film into a fiber was probably close to post
spin drawing that would improve the overall organization
of the fiber, hence its performance.
Within the pulled group, the Y1S820 fibers were the
toughest (average toughness = 10.6 MJ/m3). They were
also twice as extensible (average maximum exten-
sion = 34.06%), and could withstand almost twice as much
maximum stress (average maximum stress = 49.64 MPa)
as the A1S820 fibers which as a result displayed a higher
stiffness (average Youngs modulus = 1.706 GPa). This is
understandable if we consider the secondary structures
present in these Y1S820 proteins in an aqueous buffer at pH
8 (more b-sheets and more turns than for A1S820), and the
fact that the elastic motifs in Y1S820 can be stabilized by
hydrogen bonds due to the presence of Y and S in the
elastic motif.
Of the two synthetic constructs studied here, the primary
structure of the Y1S820 protein consensus repeat more
closely resembles the native N. clavipes dragline MaSp2
consensus repeat ([(GPGQQ GPGGY)2 GPSGPS (A)9]n)
[6], and it is interesting to note that the Y1S820 pulled
fibers exhibit the same elasticity as the native dragline silk
(34% vs. 35%), thus confirming the role of the MaSp 2
protein in the overall elasticity of the dragline. The Y1S820
pulled fibers, however, cannot withstand as much stress as
the native dragline silk (0.050 GPa vs. 4 GPa), or flagel-
liform silk (0.050 GPa vs. 0.5 GPa). Consequently, their
toughness (10.6 MJ/m3, Table 1) is lower than the ones
reported for both dragline silk (160 MJ/m3) and flagelli-
form silk (150 MJ/m3) [3, 4]. We can speculate that this
difference in stress threshold is the result of two major
factors. The first one is that the synthetic proteins are much
smaller than the native silk proteins forming the fiber thus
limiting the number of intermolecular chain interactions
necessary to stabilize the overall fiber structure and
allowing weaker spots by lack of molecular overlap.
Indeed, the sizes of both dragline silk proteins are between
300 kDa and 350 kDa [38], and although the size of Flag
has not been determined, the fact that the Flag mRNA is
bigger (15 kb) [6] than both MaSp mRNAs (11 kb and
12 kb) [39, 5] suggests that this protein is at least bigger
than 350 kDa. Additionally, it is possible that a difference
in sizes of the crystals formed and their lack of orientation
affect the strength of the synthetic fiber. The second may
be attributed to the fact that the native dragline contains a

Fig. 6 Pictures of pulled and

spun A1S820 and Y1S820 fibers.
For each fiber type, the pictures
taken before tensile tests show
the fibers that achieved the best
extension, best maximum stress
and average maximum stress.
The fibers shown here
correspond to the ones plotted in
the stress/strain curves (Fig. 5):
fiber type (P = pulled fiber;
S = spun fiber) and identity
(number) of each individual
fiber is indicated in bold in the
bottom left corner of each
picture

123
8984
second protein, MaSp1 [39], rich in GGX repeats and
containing both (GA) and (A)n crystalline forming motifs
that can impart additional strength to the fiber.
The comparison between the structures and perfor-
mances of both the Y1S820 pulled fibers and the dragline
silk might give some insight as to the necessity of two
proteins in the native dragline silk (MaSp 1 and MaSp2).
The amount of (GPGXX)4 motifs adjacent to crystalline
forming motifs in the MaSp 2 consensus repeat, as well as
their spacing, seem to be optimal for retaining reasonable
elasticity (35%) while maintaining high strength. There-
fore, a second protein is necessary for additional strength
possibly because adding extra crystalline forming motifs to
the MaSp2 consensus repeat might decrease the overall
elasticity of fiber. Moreover, water is a known plasticizer
of silks. It has been shown that native dragline silk
extruded in water is stronger thus tougher than those
extruded in air [40]. In the case of flagelliform silks, the
presence of a natural aqueous glue coating is critical in
promoting extreme elasticity (200%) [1]. It is possible that
the mechanical performances of these pulled fibers may
be modified, even improved, when wet although no such
mechanical data is available to support this statement.
Regarding the performances of the spun fibers (Fig. 5
and Table 1), we found, not surprisingly, that the A1S820
fibers on average outperformed the Y1S820 fibers for
maximum stress (28.58 MPa vs. 10.21 MPa), maximum
extension (3.72% vs. 1.59%), and toughness (0.464 MJ/m3
vs. 0.089 MJ/m3). In HFIP, we think that the elastic motifs
of the Y1S820 protein may not be able to properly form
stabilized b-turns. Indeed, in this solvent, having bulky
residues present in the 5th and 10th positions (Y and S) of
the elastic motif may impair its folding, while the A1S820
protein has the advantage of small residues (A) in these two
positions in the elastic motif. This may better accommo-
date the proper folding of the molecule providing better
overall organization allowing the poly-alanine to be locked
together into sheets and promote self assembly. However,
the extension of the spun A1S820 fibers is poor compared to
the pulled A1S820 fibers reinforcing the idea that additional
shear forces, such as post-spin draw, might be necessary for
this protein to achieve better mechanical properties.
From our results, it appears that the Y1S820 protein may
be more suitable to generate fibers with reasonable
mechanical properties when using aqueous solutions while
it is not a good candidate to generate fibers from organic
solvents.
During these studies, we noticed a high variability in the
performance of the pulled fibers as well as the spun ones
within each type of fiber (Fig. 5). Such variability, al-
though not as substantial, exists in native silk fibers [41,
42] as well as synthetic polymeric fibers. In our experi-
ments, no discernible factors such as the use of different
123
J Mater Sci (2007) 42:89748985
protein batches to make the dopes, the age of the dope,
the diameters of the fibers for instance, could account for
the distinct mechanical behaviors observed within a given
fiber group. This variability is important to notice as it
defines the upper and lower limits for the mechanical
performance of any fiber. Refining these limits, that is
lowering the variability in the mechanical performance of
any synthetic fibers, should be a goal when making such
synthetic fibers and may be achieved by mastering the
spinning conditions. Hence, we want to stress the impor-
tance of repetitions when testing synthetic materials to
know the full spectra of their performance.
Conclusion
We demonstrated a temperature dependent b-sheet induc-
tion in aqueous solution irreversible for Y1S820 and
reversible for A1S820, possibly explaining the difference in
spontaneous film/fiber formation observed for the two
proteins in aqueous solutions.
The mechanical data obtained from both types of syn-
thetic fibers made from the recombinant silk protein ana-
logs (A1S820 and Y1S820) clearly demonstrates that the
[GPGGX]n motif is indeed responsible for the elasticity
displayed by both native dragline and flagelliform silks.
Moreover, the nature of the [GPGGX]n motif, that is the
hydrophobicity or hydrophilicity of the 5th and 10th amino
acid residues, has a large impact in the level of elasticity
probably due to stability issues (presence or absence of
internal hydrogen bonding in the motif) as well as in the
self organization process of the proteins into films or fibers.
Acknowledgements We would like to thank Dr Michael B Hinman
(Department of Molecular Biology, University of Wyoming) for
providing the original A1, Y1 and S8 pBluescript clones, and for
performing the amino acid analyses of the recombinant proteins. We
also want to thank Professor Michael S Ellison (School of Materials
Science and Engineering, Clemson University) for his valuable advice
on the mechanical testing methods and analyses. This research was
supported by grants from NIH, NSF and AFoSR.
References
1. Gosline J, Denny M, DeMont M (1984) Nature 309:551
2. Stauffer S, Coguill S, Lewis R (1994) J Arachnol 22:5
3. Gosline J, Guerette P, Ortlepp C, Savage K (1999) J Exp Biol
202:3295
4. Denny M (1976) J Exp Biol 65:483
5. Hinman M, Lewis R (1992) J Biol Chem 267:19320
6. Hayashi C, Shipley N, Lewis R (1999) Int J Biol Macromol
24:271
7. Simmons A, Michal C, Jelinski L (1996) Science 271:84
8. Termonia Y (1994) Macromolecules 27:7378
9. Hayashi C, Lewis R (1998) Mol Biol 275:773
10. Hayashi C, Lewis R (2000) Science 287:1477
J Mater Sci (2007) 42:89748985
11. Becker N, Oroudjev E, Mutz S, Cleveland J, Hansma P, Hayashi
C, Makarov D, Hansma H (2003) Nat Mater 2:278
12. Hayashi C, Lewis R (2001) BioEssays 23:750
13. Zhou Y, Wu S, Conticello V (2001) Biomacromolecules 2:111
14. Chang D, Ventachalam C, Prasad K, Urry D (1989) J Biomol
Struct Dynamics 6(i5):851
15. Urry D, Shaw R, Prasad K (1985) Biochem Biophys Res Com-
mun 130(i1):50
16. Urry D, Hugel T, Seitz M, Gaub H, Sheiba L, Dea J, Xu J, Parker
T (2002) Phylos Trans R Soc London, Ser B 357(i1418):169
17. Tatham A, Hayes L, Shewry P, Urry D (2001) Biochim Biophys
Acta 1548:187
18. Hutchinson E, Thornton J (1994) Protein Sci 3:2207
19. Wilmot C, Thornton J (1988) J Mol Biol 5:221
20. Lewis R, Hinman M, Kothakota S, Fournier M (1996) Protein
Expr Purif 7:400
21. Birnboim H, Doly J (1979) Nucleic Acids Res 7:1513
22. Sambrook J, Fritsch E, Maniatis T (1989) Molecular cloning. A
laboratory manual. Second Edn. Cold Spring Harbor Laboratory
23. Wong C, Sridhara S, Bardwell J, Jakob U (2000) BioTechniques
28:426
24. Whithmore L, Wallace B (2004) Nucleic Acids Res, Web Server
issue 24:W668
25. Scheller J, Guhrs K, Grosse F, Conrad U (2001) Nature 19:573
26. Teule F, Jung S, Wood J, Marcotte W, Ellison M, Abbott A
(2002) In: Brebbia C, Sucharov L, Pascolo P (eds) Design and
nature, Udine, Italy. WIT press, p 379
8985
27. Prince J, McGrath K, DiGirolamo C, Kaplan D (1995) Bio-
chemistry 34:10879
28. Fahnestock S, Irvin S (1997) Appl Microbiol Biotechnol 47:23
29. Fahnestock S, Bedzyk L (1997) Appl Microbiol Biotechnol 47:33
30. Arcidiacono S, Mello C, Kaplan D, Cheley S, Bayley H (1998)
Appl Microbiol Biotechnol 49:31
31. Dicko C, Knight D, Kenney J, Vollrath F (2004) Biomacromol-
ecules 5:2105
32. Vollrath F, Knight D (2001) Nature 410:541
33. Knight D, Vollrath F (2001) Naturwissenschaften 88:179
34. Robson P, Wright G, Sitarz E, Maiti A, Rawa M, Youson J,
Keeley F (1993) J Biol Chem 268(2):1440
35. Rauscher S, Baud S, Miao M, Keeley F, Pome`s R (2006)
Structure 14:1667
36. Creighton T (1993) Proteins: structures and molecular properties,
Second Edn. Freeman, New York
37. Haq S, Khan R (2005) Int J Biol Biomacromolecules 36:47
38. Sponner A, Schlott B, Vollrath F, Unger E, Grosse F, Weisshart
K (2005) Biochemistry 44:4727
39. Xu M, Lewis R (1990) Proc Natl Acad Sci USA 87:7120
40. Chen X, Shao Z, Vollrath F (2006) Soft Matter 2:448
41. Blackledge T, Hayashi C (2006) J Exp Biol 209:2452
42. Blackledge T, Hayashi C (2006) J Exp Biol 209:3131
123
J Mater Sci (2007) 42:89868994
DOI 10.1007/s10853-007-1831-3
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Time-dependant intercellular delamination of human stratum
corneum
Kenneth S. Wu James Li K. P. Ananthapadmanabhan
Reinhold H. Dauskardt
Received: 30 March 2007 / Accepted: 8 May 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract Time-dependent intercellular delamination
behavior of human stratum corneum (SC) is explored. We
demonstrate that SC is susceptible to such time-dependent
subcritical delamination at loads below the critical
delamination energy. Hydration of the SC together with
chemical treatments using selected surfactants were shown
to affect both critical and subcritical delamination behav-
ior. Increased tissue hydration resulted in accelerated
delamination growth rates consistent with behavior antici-
pated from critical delamination testing. Chemical treat-
ments including 10% wt/wt sodium dodecyl sulfate (SDS)
and chloroform-methanol (2:1 v:v) soaks that remove lipid
were shown to suppress growth rates compared to untreated
controls. Possible mechanisms for subcritical delamination
involving extrinsic mechanisms of moisture-assisted reac-
tion process at the intercellular delamination tip or intrinsic
mechanisms involving kinetic relaxation processes in the
SC are discussed.
Abbreviations
SC Stratum corneum
DCB Double-cantilever beam
K. S. Wu
Department of Mechanical Engineering, Stanford University,
Stanford, CA 94305, USA
J. Li
R. H. Dauskardt (&)
Department of Materials Science and Engineering, Stanford
University, 416 Escondido Mall, Bldg 550, Rm 550G, Stanford,
CA 94305-2205, USA
e-mail: dauskardt@stanford.edu
K. P. Ananthapadmanabhan
Unilever Research and Development, Trumbull, CT 06611, USA
123
RH Relative humidity
AAS Amidosulfosuccinate
APG Alkyl polyglucoside
CMT Chloroform-methanol treated
SDS Sodium dodecyl sulfate
Introduction
The stratum corneum (SC) serves as a mechanical and
semipermeable chemical barrier between the body and the
surrounding environment. We recently reported on the
intercellular delamination behavior of SC to quantify the
cohesive properties that maintain the mechanical integrity
of this epidermal layer [1, 2]. The critical delamination
energy, Gc [J/m2], measured in terms of the applied strain
energy release rate, G, was shown to vary significantly with
SC hydration, temperature, and chemical treatment. Such
conditioning and treatment affects the intercellular lipids
and cellular proteins (e.g. keratinkeratin interactions) and
their effects on intercellular delamination energies were
quantified. The progressive degradation of corneosomes, or
structural protein linkages, between SC cells toward the
outer skin surface was also shown to result in reduced
intercellular delamination energy using a modified version
of the delamination technique [2]. The degraded corneo-
somes are associated with the continual renewal of SC in
which exterior cells are replaced by less mature interior
cells [38]. These studies, however, did not address the
possibility of time-dependent delamination behavior.
Time-dependent delamination or cracking can provide
useful insights into the kinetic bond rupture or molecular
separation process. In some materials, such cracking is
J Mater Sci (2007) 42:89868994
associated with a chemical reaction that occurs between
strained crack tip bonds and an environmental species such
as H2O that leads to bond rupture and crack advance [9]. In
other materials, inherent creep or relaxation processes at
the crack tip may lead to time-dependent material separa-
tion and crack advance [10, 11]. The phenomena is often
referred to as subcritical cracking or delamination since it
occurs at G < Gc, that is, at applied loads lower than those
required for fracture in the absence of an environmental
species or inherent relaxation process. Some materials
exhibit no detectable subcritical cracking. In the case of SC
no studies have reported on time-dependent intercellular
delamination.
In the present study we demonstrate that SC is suscep-
tible to subcritical delamination for applied G < Gc.
Intercellular delamination growth rates, v, were character-
ized in selected moist air environments. We employed
methods recently applied to measure subcritical delami-
nation behavior in thin-film and layered structures for
nanoscience and biomedical applications [9, 1216].
Delamination growth rates were shown to be a strong
function of the applied G resulting in a characteristic vG
curve. Conditioning of the SC tissue by changing hydration
content and chemical treatments using surfactants were
demonstrated to affect both critical and subcritical delam-
ination behavior. Increasing SC hydration is shown to
accelerate delamination growth rates consistent with
behavior anticipated from critical delamination testing.
Chemical exposure including 10% wt/wt sodium dodecyl
sulfate (SDS) and chloroform-methanol (2:1 v:v) soaks
that remove lipid suppressed growth rates compared to
untreated controls. These treatments have been shown
previously to affect lipid content and fluidity [4, 17].
Measured subcritical delamination behavior as a function
of treatment is interpreted in terms of changes to SC
microstructure and constituents.
Materials and methods
Tissue preparation
Human cadaver SC tissue was isolated from two female
Caucasian donors, 57 and 101 years of age, from the thigh.
Epidermal tissue was separated from dermis by immersion
of donor tissue cleared of adipose tissue in a 35 C water
bath for 10 min followed by a 1 min soak at 60 C and
subsequent mechanical separation from the dermis using a
flat-tipped spatula. Stratum corneum was isolated from
underlying epidermis by soaking in a trypsin enzymatic
digest solution (0.1% wt/wt in 0.05 M, pH 7.9 Tris buffer)
at 35 C for 120 min. During separation, the orientation of
the outer SC surface was recorded. The isolated SC was
8987
rinsed with room temperature water and allowed to dry on
filter paper (Grade 595 General-Purpose Filter Paper,
Schleicher & Schuell MicroScience GmbH, Dassel, Ger-
many) then removed and stored in ambient conditions of
1823 0.8 C and 3555 2% relative humidity (RH)
(Dickson TM325 Temperature & Humidity Data Logger,
Addison, Illinois). Comparative tests were performed on
single donor tissue specimens to reduce variability within
test sequences. During separation, orientation of the outer
SC surface was recorded.
To vary SC hydration, isolated tissue was exposed to
environments with different RH values at ambient tem-
perature (1823 C). A portion of the SC was placed in an
enclosure held at 40% RH while additional SC was placed
in another enclosure along with an open container of water
to create a 100% RH environment to increase SC moisture
content. The tissue was allowed to equilibrate for at least
24 h prior to fabrication of test specimens.
For comparison, SC, 60 60 mm2, was chloroform-
methanol treated (CMT) to delipidize the structure with a
120 min 30 mL chloroform:methanol (2:1 by volume) soak
followed by two 30 min 30 mL water rinses. Additional
tissue was treated in selected surfactant solutions to examine
their effects on SC mechanical integrity. These specimens of
SC 30 50 mm2 were immersed in 100 mL of 10% wt/wt
H2O solutions of alkyl amidosulfosuccinate (AAS), alkyl
polyglucoside (APG), and SDS for 18 h, then rinsed with
distilled water and dried on filter paper. During treatment of
SC, the solution pH values were pH 5.2, 8.18.7, and 7 0.1
(Accumet Research AR25 Dual Channel pH/Ion Meter,
Accumet Research, Fisher Scientific, Fair Lawn, NJ) for
each treatment, respectively. These surfactant solutions have
been shown to cause varying levels of lipid and protein
damage as indicated in a comparative manner in Table 1.
Note that our previous studies have revealed that donor
age and body location has an effect on measured delami-
nation behavior [1, 2]. In the present study we compare
results from single donors to negate these variations.
Delamination energy measurements
The fracture mechanics technique developed to measure
the delamination energy of SC tissue has been described
elsewhere [1]. Briefly, specimens were prepared by
adhering SC tissue between two elastic substrates of
polycarbonate (Hyzod GP, Sheffield Plastics Inc., Shef-
field, Massachusetts) with cyanoacrylate adhesive (Instant
Krazy Glue
Gel, Elmers Products Inc., Columbus, Ohio)
to form a double-cantilever beam (DCB) fracture
mechanics specimen as shown in Fig. 1. The substrate
dimensions of 40 10 2.88 mm3 were selected to ensure
elastic deformation and the valid application of linear
elastic fracture mechanics.
123
8988 J Mater Sci (2007) 42:89868994

Table 1 Table of showing SC tissue treatments, their relative protein and lipid damaging properties, and critical and subcritical delamination
failure parameters
Treatment Protein damage Lipid damage Gc (J/m2) Slope, m (da/dt/J/m2) GTH (J/m2)

Control (25 C, 45% RH) 3.66.0 2.7 0.3 1.92.9

Hydrated (25 C, 100% RH) 1.12.4 9.1 4.1
CMT Low High 6.98.0 2.8 4.85.7
SDS High Medium 7.7 2.9 4.7
APG Low Medium 1.45.9 5.0 4.2 1.0
AAS Low Low 1.62.7 5.7 3.6 0.71.5
Values accurate within 5%

P of each polycarbonate beam. For the present specimens, the

Delamination
Stratum
h Corneum
Polycarbonate
h
Polycarbonate
a b
P
Fig. 1 Fracture mechanics DCB specimen geometry illustrating
relevant loading parameters (P, D), delamination extension, a, as
measured from loading axis points and relevant specimen dimensions
(b = 10 mm, h = 2.88 mm, length = 40 mm)
The specimens were mounted at the SC-free end via
loading tabs in an adhesion test system (Delaminator Test
System v4.0, DTS Company, Menlo Park, CA) with a
computer controlled DC servoelectric actuator operated in
displacement control. The adhesion test apparatus and
DCB specimens were placed in an environmental chamber
(Model LH-6, Associated Environmental Systems, Ayer,
Massachusetts or Model ZH-16-2-H/WC, Cincinnati Sub-
Zero [CSZ], Cincinnati, Ohio) to control test environment.
Specimens were loaded in Mode I tension. Several critical
delamination energy (Gc) values were measured for each
DCB specimen and tested using either 2 or 5 lm/s dis-
placement rates during delamination extension with cor-
responding loads measured with a 222 N load cell. The
delamination length, a, was measured from recorded load-
displacement curves, PD, and their elastic compliance
relationship:
values of E and m for the polycarbonate were 2.379 and
0.38 GPa, respectively. By measuring the critical load, PC,
and the delamination length, a, at incipient delamination
extension, the delamination resistance, Gc, was determined
from critical values of the strain energy release rate,
G [18, 19]:
p   !
12P2 a2 5h 1 h 2
G 2 3 0 1 2
bh E 2 a 2 a
More detailed methods and results for critical delami-
nation energies as a function of treatment and conditioning
are described elsewhere [1].
Subsequently, after creation of a uniform delamination
front during critical delamination testing, the DCB speci-
mens were unloaded. The beam ends were then displaced
at a constant rate of 5 lm/s while monitoring the load such
that delamination growth was not observed to occur. Once
a load below the anticipated critical load, PC, to initiate
delamination growth was obtained, the displacement of the
beams was stopped, and the resulting subcritical load was
monitored as a function of time. The measured loads were
observed to decrease with time as the delamination prop-
agated through the specimen, increasing the compliance of
the DCB structure. From the compliance measured during
initial displacement of the DCB substrates prior to holding
constant displacement, the initial delamination length, ai,
can be determined by rearranging Eq. 1 to yield the
following expression for delamination length:

D 2 a 0:64h3  1=3
C 1 D E0 bh3
P 3 E0 I a
 0:64h 3
P 8
where C is the DCB specimen compliance, P is the load,
D/2 is the corresponding displacement of each beam from During the load-relaxation component of the subcritical
its original position at the loading point, E = E/(1V2) is test, the displacement, D, is held fixed with time and
the plane strain Youngs modulus for the polycarbonate, m knowledge of the initial debond length, ai, and peak load,
is Poissons ratio, I = bh3/12 is the area moment of inertia, Pi, from the initial loading of the specimen, allows calcu-
b is the polycarbonate substrate width, and h is the height lation of the delamination length as a function of time:

123
J Mater Sci (2007) 42:89868994 8989

  
h Pi 1=3 (a) 2.4
at ai 1 0:64
 0:64h 4 Pi o
25 C 40% RH
ai Pt 2.0

The delamination growth velocity can be calculated by 1.6

Load, P (N)
taking the time derivative of the above expression:
1.2
  !
1=3
da ai h Pi dP 0.8
 1 0:64
4=3
5
dt 3 ai Pt dt
0.4

Subcritical values of G are calculated from the measured

0.0
loads corresponding calculated delamination lengths (b)
(Eq. 2). Knowledge of these two parameters allows one to

Delamin. Length, a (mm)

plot delamination growth rate (v = da/dt) as a function of 19
the applied strain energy release rate, G, also known as vG
curves. A typical vG curve is shown in Fig. 3.
18
For the above calculations, the resultant error for the
calculated crack length, a, is <1% and for delamination
ai
resistance values, G, and delamination growth velocity, 17
da/dt, <5%. In particular, instrument error associated with
compliance, C, and geometric dimension measurements, b
16
and h were <0.2% for the values measured. 0 400 800 1200 1600 2000
. . Time, t (s)
=C =0

Results Fig. 2 Load-relaxation data illustrating for SC conditioned at

Load-relaxation data for SC conditioned at 1823 C,
40% RH and tested in a 25 C, 45% RH environment is
shown in Fig. 2. The data clearly show that significant
load-relaxation has occurred, and corresponding calcula-
tion of delamination length revealed subcritical delami-
nation growth with time. The initial delamination length,
ai, and peak load, Pi, used to calculate delamination
length and growth rate with time are indicated. Analysis
of the load-relaxation data using Eqs. 2 and 5 allowed
determination of delamination growth rates as a function
1823 C, 40% RH and tested in a 25 C, 45% RH environment
showing (a) initially increasing load at constant displacement rate
D_ C followed by load decrease with time once displacement is
fixed D_ 0; and (b) delamination length as a function of time
calculated from specimen compliance. Corresponding subcritical
curve shown in Fig. 3
inorganic interface [20]. For both hydration and surfactant
treated SC, a summary of salient treatments, cohesive
delamination energy, Gc, values and parameters for
-4
10 o GC
D e la m . G r o w t h R a t e , d a / d t ( m / s )

25 C 40% RH
of applied G as shown in Fig. 3. The corresponding
-5
critical delamination energy, Gc, is included for compar- 10
ison. Delamination growth rates were characterized over
-6
nearly five orders of magnitude, and were clearly sensi- 10
tive to the applied loads. Intermediate growth rates in the
-7
range of 108105 m/s exhibited an exponential depen- 10
dence on G. Alternatively, a power-law relationship of the -8
form da/dt = C
Gm can be employed to characterize the 10
intermediate delamination growth rates, where C and the -9 G TH
10
exponent m are constants dependent on tissue condition
and testing environment. Threshold behavior for applied -10
10
loads approaching a threshold applied strain energy 1 2 3 4
2
release rate, GTH, was clearly apparent at low growth Applied Strain Energy Release Rate, G (J/m )
rates approaching 109 m/s. Delamination is presumed
Fig. 3 Typical delamination growth rate (da/dt) versus applied strain
dormant at applied G less than GTH. Not all materials
energy release rate (G) for SC conditioned at 1823 C, 40% RH and
exhibit a clear threshold load for crack or debond growth tested in a 25 C, 45% RH environment showing threshold behavior
as reported in a recent study of debonding of a polymer/ (GTH) and critical delamination energy values (Gc)

123
8990 J Mater Sci (2007) 42:89868994

characterization of subcritical delamination behavior are 10

-4
25C GC 40% RH GC GC
presented in Table 1. 100% RH (45% RH)

Delam. Growth Rate, da/dt (m/s)

-5 (92% RH)
10
Subcritical delamination as a function of tissue
hydration 10
-6
100% RH
(45% RH) 40% RH
The effects of preconditioning the SC on delamination 10
-7 (92% RH)
growth rate behavior are shown in Figs. 4 and 5. The
curves in Fig. 4 show behavior for specimens from the -8
model prediction
10
same tissue sample conditioned in selected RH air envi- Eq. 6
ronments at ambient temperatures 1823 C for at least -9
10 precond. RH
24 h followed by testing in similar RH environments at
(chamber RH)
25 C. Data presented in Fig. 5 includes similarly prepared -10
10
specimens from another tissue sample as well as specimens 0 1 2 3 4 5
2
preconditioned at one RH and tested in a different RH. Applied Strain Energy Release Rate, G (J/m )
Debond growth rate curves for tissue conditioned and
Fig. 5 Delamination growth rate (da/dt) as a function of applied
tested at either 4045% RH or 90100% RH exhibited well
strain energy release rate (G) showing SC preconditioned and tested
defined debond growth curves with evidence of debond in similar environments (filled and open points) and for specimens
growth thresholds, GTH, at growth rates approaching preconditioned at one RH and tested in an environment with different
109 m/s. For each RH range examined, some scatter in the RH (half filled, half open points). Subcritical curves for 100% RH
conditioned specimens are observed to plateau at lower strain energy
crack growth curves was apparent and is indicated by the
release rates. Labels show the preconditioning RH and environmental
shaded area in the figures. Such scatter is not uncommon test chamber RH in parentheses
for crack growth behavior and certainly not excessive for
natural tissues. Increasing SC hydration with higher RH
conditioning led to accelerated growth rates and steeper RH and tested in a 92% RH environment exhibited inter-
slopes of the delamination growth curves. Critical Gc val- mediate growth rates in the same range as tissue condi-
ues were also found to decrease with increasing hydration tioned and tested in a 4045% RH environment. However,
as previously reported [1]. as the growth rates decreased, a plateau in the growth rate
The effects of testing a specimen preconditioned in one at approximately 6 109 m/s was apparent with little
RH environment and tested in a different RH environment dependence on applied G. With decreasing loads, the
are shown in Fig. 5. The specimen preconditioned at 40% growth rate remained relatively constant until G reached
values typical for the 100% RH conditioned tissue. Growth
rates then appeared to be consistent with those for the
-4
10 o
25 C GC GC GC GC 100% RH conditioned tissue and the onset of a threshold
was also apparent. The time the tissue was exposed to the
Delam. Growth Rate, da/dt (m/s)

100% RH
-5
10 (92% RH) higher humidity testing environment was approximately
2.6 104 s.
-6
10 increasing Tissue conditioned at 100% RH and tested in the drier
hydration 45% RH environment exhibited behavior essentially the
-7
10 40% RH same as that for specimens treated and tested at the higher
(45% RH)
RH. However after 4 103 s, the tissue was affected by the
-8
10 drier testing environment and exhibited some contraction
that was evidenced by the measured load beginning to
-9
10 precond. RH increase with time. This behavior invalidated the load
(chamber RH) relaxation analyses and the test was terminated at the last
valid growth rate in the vicinity of 108 m/s.
-10
10
1 2 3 4 5 6
2
Applied Strain Energy Release Rate, G (J/m )
Effects of tissue surfactant treatments
Fig. 4 Delamination growth rate (da/dt) versus applied strain energy
release rate (G) for SC specimens conditioned in environments of Delamination was performed on CMT, AAS, APG, and
different RH at ambient temperatures 1823 C and tested in
SDS treated tissue to examine the effects of these solutions
environments with RH similar to their preconditioning values. Labels
show the preconditioning RH and environmental test chamber RH in on SC delamination growth rate behavior compared to that
parentheses of untreated controls. These specimens were conditioned

123
J Mater Sci (2007) 42:89868994
and tested at 45% RH at ambient temperatures 1823 C
and 25 C, respectively. These data are presented in
Fig. 6 with parameters listed in Table 1. The shapes of
the delamination growth rate curves were similar to the
untreated SC although the slope, and particularly the
position of the vG curves along the G-axis, exhibited a
marked sensitivity to the tissue treatment. The curves all
exhibited threshold behavior at low delamination growth
rates. The harsh CMT and SDS surfactant treated tissue
exhibited delamination growth rate curves at significantly
higher G values compared to the untreated control, and the
tissue treated in the milder surfactants, AAS and APG,
exhibited curves at lower applied G values indicative of
decreased resistance to delamination. Critical delamination
energy values, Gc, exhibited similar trends with treatment
to the time-dependent delamination curves.
Discussion
Susceptibility to subcritical delamination
The data presented in the present study demonstrates that
SC tissue is susceptible to time-dependent intercellular
delamination growth under load in moist environments.
Not all materials or interfaces exhibit such time dependent
delamination or cracking at applied loads less than the
fracture energy Gc.
The observed behavior may be associated with a
chemical reaction that occurs between strained crack tip
10
-4
GC
GC GC
Delam. Growth Rate, da/dt (m/s)
-5 APG AAS Ctrl
10
-6
10
-7
10
-8
10
-9
10
-10
10
0 1 2 3 4 5
Applied Strain Energy Release Rate, G (J/m )
Fig. 6 Effect of chemical treatments on critical and subcritical
delamination behavior. Critical and subcritical delamination energy
measurement results for various pH and surfactant treatments (SDS:
sodium dodecyl sulfate (10% wt/wt); CMT: chloroform-methanol
(2:1 v:v) treatment; APG: alkyl polyglucoside (10% wt/wt); AAS:
alkyl amidosulfosuccinate (10% wt/wt))
8991
bonds and an environmental species such as H2O which
leads to bond rupture and crack advance [9, 1416]. This
type of environmentally assisted cracking generally has
characteristic features of the resulting vG curve. Inter-
mediate growth rates in the range of 108105 m/s are
strongly sensitive to applied G, the activity of the envi-
ronmental species and temperature, indicative of the
environmentally assisted reaction kinetics associated with
the crack tip bond rupture process. Further decreases in G
often result in threshold behavior below approximately
109 m/s where crack growth rates decrease below
detectable limits. At higher growth rates, a transport-lim-
ited region largely independent of G is characterized by a
plateau in the vG curve [9, 1416]. We did not observe a
transport limited region in the present study, but such
behavior is often observed at higher crack growth rates,
which we did not characterize. With still further increases
in applied loads where G approaches Gc fracture occurs at
high crack growth rates and is largely independent of
environmental test conditions.
Time dependent fracture may also be related to inherent
creep or relaxation processes that occur in a process zone
surrounding the crack tip that lead to material separation
and crack advance. Such behavior has been reported for
metals, inorganic, and organic materials [10, 11]. In the
case of polymers, the time dependence of crack growth
results from viscoelastic deformation in the crack tip pro-
cess zone which is dependent on both testing temperature
and crack growth rate [11]. While we were not able to
unambiguously establish the precise time dependent deb-
onding mechanism in the present study, the intercellular
separation process that occurs during delamination of SC
GC GC
involving separation of intercellular lipids and corneodes-
mosomes suggests that the latter process involving an
SDS CMT inherent time dependent molecular separation process is
more likely. We note also that the SC does not contain
atomic bonds like silanol bonds in inorganic glasses that
are particularly sensitive to moisture adsorption and asso-
ciated bond cleavage. Organic molecules containing carbon
backbones are generally insensitive to cleavage by H2O or
OH molecules in moist environments and therefore typi-
cally insensitive to environmentally assisted cracking.
Hydration effects
o
25 C 45% RH
The reduction in critical delamination energies, Gc, with
6 7 8
2 SC hydration (Figs. 4, 5) is consistent with prior examin-
ations of critical delamination energy values [1, 2]. The
effects of hydration on SC components are numerous
including corneocytes swelling and the presence of water
pooling between cells [21, 22], disruption of intercellular
lipid lamellae [23, 24], and even the degradation of cor-
neodesmosomes [7, 21]. We discussed the decreased
123
8992
intercellular delamination energy Gc values with increased
SC hydration in terms of increased lipid mobility and
possible intercellular boundary separation by the presence
of water which would lower the measured Gc [1, 2].
The same mechanisms leading to decreased Gc values
with increasing SC hydration likely affect the observed
time-dependent delamination behavior and lead to vG
curves shifted to lower values of applied G. Alternatively,
for a given value of applied G, delamination growth rates
would be significantly faster with increased SC hydration.
The steeper slopes of the vG curves with increased SC
hydration shown in Fig. 5 (crack growth exponents m in
Table 1) further highlight the structurally degrading effect
of tissue hydration. Delamination growth rates were more
sensitive to changes in applied G for highly hydrated SC
compared to the less hydrated specimens.
The effects of plastic energy dissipation in the SC layer
are not likely to contribute significantly to delamination
energy which has been linked more strongly to the inter-
cellular separation process [1, 2]. The observation that in-
creased hydration accelerates delamination growth rates
further corroborates this notion. If plasticity played an
important role in delamination behavior, growth rates
would be expected to decrease with increasing hydration
(i.e. vG curves would shifted to higher values of applied G
compared to drier specimens) due to increased tissue
plastic deformation associated with the lower yield stress
reported for hydrated SC [1].
To elucidate the possible mechanisms associated with
time dependent delamination and to probe the effects of
changing the H2O activity in the testing environment, vG
curves were measured for several specimens precondi-
tioned at either a low or high humidity and then tested in
the opposite humidity environment. The results shown in
Fig. 5 clearly reveal that changes to the testing environ-
ment RH do not cause changes in the vG curves for
delamination growth rates greater than 108 m/s. The
insensitivity to environmental RH suggests that time
dependent delamination behavior is determined by the SC
hydration rather than that of the testing environment. This
suggests that the mechanism of time-dependent intercel-
lular delamination is not controlled by a chemical reaction
process similar to silanol bond cleavage in moist environ-
ments reported in silica glasses. This mechanism would
lead to significant sensitivity to the moisture content of the
test environment.
Only at low growth rates less than 108 m/s when the
low humidity conditioned tissue was exposed to the higher
humidity testing environment for longer times did the
growth rate behavior begin to approach that of SC condi-
tioned in the higher humidity environment. This is clearly
apparent for the SC preconditioned at 40% RH and tested
in a 92% RH environment (Fig. 5). At low growth rates the
123
J Mater Sci (2007) 42:89868994
G applied
Polycarbonate
moisture diffusion SC
delamination velocity
Fig. 7 Schematic illustration of intercellular delamination in a DCB
specimen showing diffusion of moisture from the test environment
into a region surrounding the delamination
SC near to the crack tip absorbs water from the testing
environment as shown schematically in Fig. 7 and growth
rates eventually approach that of the higher humidity
conditioned SC. When hydrated tissue is tested in a dry
environment, the opposite behavior involving diffusion of
water out of a zone surrounding the delamination would
occur. This phenomenon represents a coupled rate depen-
dent process where the rate of moisture diffusion competes
with the delamination growth rate. When the delamination
growth rate is sufficiently high (for the present SC and
temperature >108 m/s) the delamination propagates faster
than moisture diffusion can equilibrate the tissue sur-
rounding the delamination tip to the testing environment.
At lower growth rates the SC at the delamination tip
becomes equilibrated to the moisture content of the testing
environment.
The above observations suggest that the intrinsic SC
hydration dominates the time-dependent intercellular sepa-
ration process and delamination growth rates. In a recent
study, the debonding behavior of weakly bound interfaces
between polymer layers and an under laying elastic substrate
were characterized in moist environments [20]. At low
growth rates, a similar plateau was observed where debond
growth rates were characterized by a weak dependence on
applied G, a strong dependence on moisture activity, and the
absence of a measurable threshold. The mechanism of deb-
onding was associated with the diffusion of moisture in the
polymer layer to the interface where it weakens the interface
and gives rise to time dependent delamination. The weak
dependence on applied G is related to the stress dependent
diffusion of water that is well known in polymers. Incorpo-
rating the stress dependent diffusion constant of water in the
polymer later into a debond growth model allowed the
following prediction for the debond growth rate:
 p
da Ed a G
C aH2 O D0 exp 6
dt RT
J Mater Sci (2007) 42:89868994
where C a constant that includes the concentration of
reactant molecules, the interfacial bond length and areal
bond density, aH2 O is the moisture activity, D0 is the pre-
exponential diffusion coefficient of water in the polymer,
Ed is the activation energy for diffusion, a is a constant
involving the activation volume associated with the diffu-
sion process, R the universal gas constant, and T the tem-
perature in Kelvin. The constant a/RT in the exponential
was found to be approximately 1 kJ1/2 m/mol. While is it
premature to attempt a similar detailed fit of the above
model to the present study where the structure of the
intercellular delamination path including details of the lipid
and corneodesmosome bond length and areal density are
not known, it is nevertheless instructive to note that the
observed plateau in growth rates is likely related to a very
similar mechanism involving moisture diffusion through
the SC layer. If that is the case, then it is also possible that
the moisture diffusion will exhibit the same stress depen-
dence observed in other organic polymer materials. We
should then expect to observe a similar weak dependence
of growth rates in the plateau region on the applied G as
predicted by Eq. 6. Using a simplified form of Eq. 6 we
find that a best fit to the data yields a value of
a/RT = 1.5 kJ1/2 m/mol which is in close agreement with
the previous value of a for weakly bound polymer inter-
faces, and further that the model captures the plateau
growth rate dependence on G very well. In fact, the
successful application of the stress-dependent transport
model suggests that like other organic polymer materials,
water diffusion in SC may well be stress dependent.
Chloroform-methanol and surfactant effects
When exposed to chemical treatments, SC exhibits sig-
nificant changes in delamination behavior as seen in Fig. 6
and quantified in Table 1 for both critical and subcritical
measurements. The treatments used to condition SC in
these subcritical tests have been shown to damage protein
and lipids structures to varying degrees as compared in
Table 1. Chloroform-methanol extraction has been shown
previously to remove the majority of intercellular lipids
from SC tissue facilitating the direct apposition of unex-
tracted lipids covalently bound to the cornified envelopes
of adjacent corneocytes [4, 2529]. The increased critical
delamination energies of delipidized SC have been attrib-
uted to the interdigitation of opposing corneocyte envelope
lipids [28, 30]. Only with extraction of the lamellae and
envelope lipids is dissociation of SC tissue into individual
cells observed, supporting this explanation [31]. Through
thickness multiple delamination tests on CMT SC has
indicated that the extraction is pervasive throughout the SC
8993
layer leading to overall increases in delamination energies
compared to untreated controls [2]. The strong surfactant
quality of SDS likely plays a similar role in extracting
intercellular lipids leading to increases in critical delami-
nation energies.
Extending our knowledge of surfactant behavior on
critical delamination energies to subcritical behavior, we
find delamination growth rate curves shifted to higher
G-values with strong surfactant treatments (Fig. 6) similar
to trends observed between critical delamination energy
values. The similar shapes of the vG curves between un-
treated and treated specimens indicate that the mechanisms
driving delamination growth are similar but that higher
applied G values are required to facilitate activation of the
kinetic processes leading to subcritical delamination
growth. The similar behavior to reaction-controlled curves
for treated delamination growth velocities as a function
of G suggests that similar mechanisms to those seen
in untreated SC may be occurring. The differences in
subcritical curve positions along the G axis may be
attributed to modifications of the intercellular material
through which the debonding occurs. For the CMT and
SDS treated SC, depletion of intercellular lipid bonds with
strongly corneocyte envelope lipid interactions provides an
obvious explanation of the increased G-values; however,
the accelerated delamination rates that occur with exposure
to milder surfactants (AAS and APG) cannot be explained
in the same manner.
Examining the milder surfactant treatments, their critical
delamination energies fall toward lower values than those
of the untreated controls. These results suggest that these
milder surfactants do not affect the SC in the same manner
as the harsher surfactants. The results are surprising
because any lipid extraction would be expected to cause
increases in SC delamination energies while exposure to
these solutions led to decreases. One competing mecha-
nism that may explain the incongruous behavior is that of
SC hydration content. While the described tests control
external factors such as environmental RH, the actual water
content of the SC may vary with treatment as water-hold-
ing capacity is modulated if hygroscopic natural moistur-
izing factor (NMF) content is changed in the system.
Alternatively, the treatments may increase water holding
capacity of the skin as is the case with glycerol which has
been shown to absorb into SC and whose large water-
holding capacity is well known [32, 33]. This effect of
hydration may explain the decreases in delamination
energies with AAS and APG treatment but require further
examination to provide evidence of these effects. The
subcritical data, however, faithfully track with the critical
values and exhibit similar behavior to that of untreated and
harsher surfactant-treated SC.
123
8994
Conclusions
We demonstrate that SC is susceptible to time-dependent
subcritical delamination at loads below the critical
delamination energy. Increased hydration has been shown
to accelerate delamination growth rates for a given applied
strain energy release rate; however, the SC is sensitive only
to its hydration state and not to that of the external envi-
ronment as shown from tests in which specimens are pre-
conditioned in one environment with set RH and tested in
another with different RH. Chemical treatments were
demonstrated to shift the subcritical growth rate curves
along the applied strain energy release rate axis compared
to untreated controls. The slope of the chemically treated
SC growth rate curve was similar to the untreated speci-
mens preconditioned and tested at similar temperature and
RH values. The delamination growth rate behavior was
characteristic of classical chemical reaction rate or creep
crack growth controlled behavior without evidence of a
transport controlled region. At low growth rates, specimens
preconditioned in a relatively dry environment (45% RH)
and tested in a humid environment (92100% RH) exhib-
ited a plateau like growth rate behavior as a function of
applied loads that appears related to a stress dependent
moisture diffusion process in the SC. Based on the sensi-
tivity of intercellular delamination growth rates to tissue
conditioning and insensitivity to the moisture content of the
external environment, a mechanism involving inherent
creep or relaxation processes appeared more likely. We
believe that the research demonstrates that the time
dependent delamination of human SC provides a poten-
tially sensitive indication of the effects of tissue condition
and treatment. It may therefore have important application
in the screening of treatments.
Acknowledgments This work was supported by a grant from
Unilever Research and Development (Edgewater, NJ; Trumbull, CT).
References
1. Wu KS, van Osdol WW, Dauskardt RH (2006) Biomaterials
27(5):785
2. Wu KS, Stefik MM, Ananthapadmanabhan KP, Dauskardt RH
(2006) Biomaterials 27(34):5861
123
J Mater Sci (2007) 42:89868994
3. Chapman SJ, Walsh A (1990) Arch Dermatol Res 282(5):304
4. Chapman SJ, Walsh A, Jackson SM, Friedmann PS (1991) Arch
Dermatol Res 283(3):167
5. Lundstrom A, Serre G, Haftek M, Egelrud T (1994) Arch Der-
matol Res 286(7):369
6. Walsh A, Chapman SJ (1991) Arch Dermatol Res 283(3):174
7. Rawlings A, Harding C, Watkinson A, Banks J, Ackerman C,
Sabin R (1995) Arch Dermatol Res 287(5):457
8. Simon M, Bernard D, Minondo AM, Camus C, Fiat F, Corcuff P
et al (2001) J Invest Dermatol 116(1):23
9. Wiederhorn SM (1967) J Am Ceram Soc 50(8):407
10. Riedel H (1987) Fracture at high temperatures. Springer-Verlag,
Berlin, New York
11. Bradley W, Cantwell WJ, Kausch HH (1997) Mech Time-
dependent Mater 1(3):241
12. Ohashi KL, Dauskardt RH (2000) J Biomed Mater Res 51(2):172
13. Ohashi KL, Yerby SA, Dauskardt RH (2001) J Biomed Mater
Res 54(3):419
14. Guyer EP, Dauskardt RH (2005) J Mater Res 20(3):680
15. Guyer EP, Dauskardt RH (2004) Nat Mater 3(1):53
16. Lane MW, Snodgrass JM, Dauskardt RH (2001) Microelectron
Reliab 41(910):1615
17. Ananthapadmanabhan KP, Lips A, Vincent C, Meyer F, Caso S,
Johnson A et al (2003) Int J Cosmet Sci 24:103
18. Kanninen MF (1973) Int J Frac 9(1):83
19. Ohashi KL, Romero AC, McGowan PD, Maloney WJ, Dauskardt
RH (1998) J Orthop Res 16(6):705
20. Sharratt BM, Wang LC, Dauskardt RH (2007) Acta Mater
55(10):3601
21. Bouwstra JA, de Graaff A, Gooris GS, Nijsse J, Wiechers JW,
van Aelst AC (2003) J Invest Dermatol 120(5):750
22. van Hal DA, Jeremiasse E, Junginger HE, Spies F, Bouwstra JA
(1996) J Invest Dermatol 106(1):89
23. Warner RR, Stone KJ, Boissy YL (2003) J Invest Dermatol
120(2):275
24. Warner RR, Boissy YL, Lilly NA, Spears MJ, McKillop K,
Marshall JL et al (1999) J Invest Dermatol 113(6):960
25. Sznitowska M, Janicki S, Williams A, Lau S, Stolyhwo A (2003)
J Pharm Sci 92(1):173
26. Swartzendruber DC, Wertz PW, Madison KC, Downing DT
(1987) J Invest Dermatol 88(6):709
27. Wertz PW, Madison KC, Downing DT (1989) J Invest Dermatol
92(1):109
28. Wertz PW, Swartzendruber DC, Kitko DJ, Madison KC,
Downing DT (1989) J Invest Dermatol 93(1):169
29. Wertz PW, Swartzendruber DC, Madison KC, Downing DT
(1987) J Invest Dermatol 89(4):419
30. Stewart ME, Downing DT (2001) J Lipid Res 42(7):1105
31. Smith WP, Christensen MS, Nacht S, Gans EH (1982) J Invest
Dermatol 78(1):7
32. Reiger MM, Deem DS, Skin Moisturizers II (1974) J Soc Cosmet
Chem 25:253
33. Froebe CL, Simion FA, Ohlmeyer H, Rhein LD, Mattai J, Cagan
RH et al (1990) J Soc Cosmet Chem 41:51
J Mater Sci (2007) 42:89959004
DOI 10.1007/s10853-007-1741-4
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
Mechanical modeling of viral capsids
Melissa M. Gibbons William S. Klug
Received: 26 February 2007 / Accepted: 2 April 2007 / Published online: 17 July 2007

Springer Science+Business Media, LLC 2007
Abstract As revealed by techniques of structural biology
and single-molecule experimentation, the protein shells of
viruses (capsids) are some of natures best examples of
highly symmetric multiscale self-assembled structures, with
impressive mechanical properties of strength and elasticity.
Mechanical models of viral capsids built from the bottom
up, i.e., from all-atom models in the context of molecular
dynamics and normal mode analysis, have chiefly focused
on unforced vibrational capsid dynamics. Due to the size of
viral capsids, which can contain several hundred thousand
atoms, the computer power needed for these types of
methods has only recently reached the level required for all-
atom simulations of entire viral capsids. Coarse-grained
normal mode analysis has provided a simplified means of
studying the unforced vibrational dynamics of viral capsids.
Recent focus on top-down mechanical models of viral
capsids based on two- and three-dimensional continuum
elasticity have provided a theoretical complement to single
molecule experiments such as atomic force microscopy, and
have advanced the fundamental understanding of the forced
mechanics. This review serves to assess the current state of
modeling techniques for the study of the mechanics of viral
capsids, and to highlight some of the key insights gained
from such modeling. In particular, a theme is established
of a link between shapeor geometryand the global
mechanical properties of these hierarchical multiscale
biological structures.
M. M. Gibbons (&)
W. S. Klug
Department of Mechanical and Aerospace Engineering,
University of California, Los Angeles, CA 90095, USA
e-mail: gibbons@ucla.edu
W. S. Klug
e-mail: klug@ucla.edu
Introduction
The use of biological materials in modern materials science
is a rapidly growing field. The nanometer-sized protein
shells, or capsids, of viruses are attractive candidates for
the design of new materials, primarily because they
assemble from a small set of components to form simple
structures with precisely determined geometries and sizes.
In recent years, several novel uses for viruses in nanoscale
materials have been proposed. One idea is to use soft
materials, as found in the biological world, to organize
inorganic materials, creating a class of new hybrid mate-
rials. Accordingly, Lee et al. [1] engineered a highly ori-
ented, self-supporting composite material assembled from
a bacteriophagea bacterial virus(M13) and ZnS solu-
tion. In another study, Blum et al. [2] engineered cysteine
residue placement in the subunit structure of cowpea mo-
saic virus, such that the capsid is utilized as a scaffold for
the binding of gold nanoparticles in specific, alterable, and
reproducible configurations. Similarly, viruses themselves
have been organized by binding them to nucleotide chains
[3]. Viral capsids have also been used as nanocages for
constrained nanomaterials synthesis [4, 5]. By slightly
altering the charges on the interior of the capsid proteins,
the chemical reactivity can be tuned to match the reactivity
of the material in question. These examples serve as a first
look at ways in which viruses may be employed outside of
the biological world.
To fully embrace viral capsid structures as components
of new materials, the mechanical properties and behavior
must be fully elucidated. The major theme of this review is
that mechanical models of viral capsids will serve as a
centerpiece to that end. Yet, apart from theoretical mod-
eling, much is known about the mechanical properties of
viral capsids simply from observing them experimentally.
123
8996
To begin with, viral capsids are a particularly impressive
result of two-dimensional protein assembly, which serve to
contain and protect viral genome molecules (either DNA or
RNA). These protein shells assemble from multiple copies
of only a few capsid proteins (in some cases only one) held
together by non-covalent forces into very regular structures
that are, in most cases, either cylindrical or spherical in
shape. In rare cases, capsid self-assembly is achievable
in vitro from purified protein components [6]. Because of
their regularity, the structures of many capsids have been
determined to very high resolution (better than 3 A ) molecular spatial scales as well. For example, as revealed in
through combined X-ray crystallography and cryo-electron
microscopy (cryo-EM) studies. Figure 1 shows represen-
tations of two viruses that have been extensively studied
both structurally and mechanically: the plant virus, cowpea
chlorotic mottle virus (CCMV), and the bacteriophage
/29. The structure of CCMV is known at a resolution of
about 3 A [7], and so the atomic coordinates are known
with a high degree of certainty, as shown in Fig. 1a. The
low-resolution representation of the CCMV capsid is
shown in Fig. 1b, enabling the icosahedral nature of the
capsid to be clearly seen. The structure of the /29 capsid is
not yet known at atomic resolution, thus the only data
available is a density map created from low-resolution
cryo-EM data. The density map as determined by Morais
et al. [8] is shown in Fig. 1c.
The structure of spherical capsids is ordered according to
the symmetry of an icosahedron, with capsid proteins
grouped locally into fivefold and sixfold capsomers. Five-
fold capsomers (pentamers) adopt the role of the 12 icosa-
hedral vertex positions, whereas the sixfold capsomers
(hexamers) fill in the 20 icosahedral face regions. The
pentamerhexamer count can vary according to the
T-number classification proposed by Caspar and Klug stimulus. A notable exception is the maturation of FHV
[12] in the 1960s, which groups the total number of 60T
capsid proteins into 12 pentamers and 10(T1) hexamers.
The T-number index can only adopt certain integer values,
1, 3, 4, 7, 9, 13, . Among the different spherical viruses,
Fig. 1 Representations of the CCMV (a) and (b) and /29 (c) viral
capsids. The structure of CCMV has been determined to atomic
resolution, and therefore the atomic positions are known with a high
degree of precision, as shown in the atomic representation (a). The
low-resolution representation of CCMV in (b) shows the icosahedral
structure. The capsid structure of /29 has not beed determined to
123
J Mater Sci (2007) 42:89959004
capsid diameters vary roughly from about 20 nm to a few
hundred nm. Constituent proteins are roughly similar in size
for most viruses, such that capsid size tends to increase
directly with T number. For most viruses, capsid size is
uniformly and precisely determined and is consistently
reproduced under physiological conditions. One notable
exception is the human immunodeficiency virus, HIV-1,
the capsid of which is polymorphic, marked by variation
in size and shape [13]. Interestingly, similarities in the
multiscale structural ordering extend to sub-protein,
structures determined by X-ray crystallography and cryo-
EM, the coat proteins of many plant, insect, and animal
viruses having icosahedral structure are formed around the
same core structural motif, that of the so called jelly roll
b barrel [14].
Despite their precise ordering, significant configura-
tional changes have been observed at various stages
throughout the viral life cycle for some capsids. For
instance, many viruses first assemble as procapsids
which then mature into infectious capsids. Maturation can
involve structural transitions where proteins are rearranged
(e.g., HIV-1 [15]), existing covalent bonds are broken (e.g.,
flock house virus (FHV) [16]), or new covalent bonds are
formed (e.g., HK97 [17]). Structural transitions in capsids
can also be triggered by environmental changes, as is the
case for the pH controlled swelling of CCMV [7, 18] and
the pH triggered release of a pentamer by tymovirus [19].
Similarly, calcium-mediated deformation of simian virus
40 (SV40) is thought to play a role in its infection process
[20]. The mechanics of these structural transitions is not
well understood nor is it generally known how these con-
formational changes can be controlled by mechanical force
which is influenced by pressure on the capsid [16].
Recently, technological advances in single-molecule
experimentation have enabled direct measurement of the
strength and elasticity of viral capsids. Measurement of the
atomic resolution, thus only low-resolution representations such as
density maps are available, and shown in (c). Renderings of CCMV
and /29 were generated with the molecular imaging software,
Chimera [9]. The CCMV atomic coordinates were downloaded from
the RCSB Protein Data Bank [10], and the /29 density map was
downloaded from EMBL-EBI [11]
J Mater Sci (2007) 42:89959004
forces required to package DNA into bacteriophage /29
using optical tweezers has shown that its capsid is strong
enough to sustain effective pressures estimated on the
order of 60 atm [2124]. Traditional experimental methods
for determining elastic properties of materials involve
stretching and compressing them. Atomic force micros-
copy (AFM) allows for this to be done with single mole-
cules and their assemblies. AFM has been used to study the
mechanical properties of the capsids of several viruses:
/29 [25], CCMV [26, 27], parvovirus minute virus [28],
and murine leukemia virus [29]. In fact the technique is
useful for a general class of protein assemblies and has
also been used recently to study buckling of microtubules
[30]. In these experiments, capsids are indented with the
AFM cantilever tip, producing a force-deflection curve.
Figure 2 shows force-indentation results from experiments
performed on CCMV at two different pH levels [26, 27].
At pH 5 a jump in force is observed at a compression of
about 30% and accompanied by a permanent decrease in
stiffness (Fig. 2b) indicating an irreversible mechanical
failure of the capsid. In contrast, the pH 6 force response
remains linear without significant hysteresis throughout the
complete indentation range, even to the point where the
capsid is compressed such that the top and bottom inner
surfaces of the capsid shell are pressed into contact. Thus,
at pH 6 CCMV is almost perfectly elastic, and effectively
indestructible under nanoindentation.
The collection of experiments on viral capsids reveal a
number of mechanical properties that contribute to their
allure for materials applications. Namely, their surprising
strength and flexibility as found by AFM experiments, even
under large deformations, sets them apart from most tra-
ditional materials. Additionally, the ability to control and
alter the material properties of some viral capsids as a
function of adjustable environmental conditions, such as
pH, makes them versatile. And yet, viral capsids are not
weak; as noted above, some viral capsids are able to sustain
extremely high pressures [2124]. Finally, viral capsids can
Fig. 2 Indentation force plotted versus substrate displacement for
CCMV capsids at pH 6 (left) and pH 5 (right) [26, 27]. For a
corresponding force, the indentation is given by the distance between
the black curve (cantilever alone) and the colored curves (cantilever
8997
exhibit some degree of structural flexibility, as they may
undergo conformational changes in which their size can
expand or contract, and may self-assemble from pure
components. Effective utilization of these properties re-
quires a full understanding of the underlying mechanisms.
Complementary theories are needed to bring about the
understanding of these mechanisms, and mechanical
models provide an excellent foundation for determining the
properties that control these behaviors.
To date, the use of bottom-up mechanical models such
as molecular dynamics (MD) and normal mode analysis
(NMA) have been the most widely used methods to study
the unforced vibrational dynamics of viral capsids. These
methods tend to involve an all-atom description of capsid
structure, although coarse-graining has gained in use. In
fact, viral capsids, which are large macromolecules with
many thousands of atoms, are nearly impossible to study
with these methods without some form of coarse-graining.
Because these methods are meant to study the unforced
vibrational dynamics of a system, they give an indirect
look at the mechanics of viral capsids. One of the important
results to come out of the coarse-grained studies of viral
capsids is that the overall structure, and not the atomic
detail, is the main factor in controlling the unforced
mechanics. With the development of new direct experi-
mental methods of probing viral capsids, such as the AFM
experiments described above, mechanical models are nee-
ded to describe the forced mechanical behavior. In the last
few years, top-down approaches such as continuum mod-
eling have been developed and used as a theoretical com-
plement to describe the response of viral capsids under
forced conditions, with the results suggesting that the
overall shape of the capsid also plays a dominant role in
governing the forced mechanical behavior. Ultimately, it is
likely that novel multiscale methods will be required to
capture both the global structural response of the capsid in
addition to the detailed atomic response that will govern
complex behaviors such as failure.
and capsid in series). Loading is described by the red curves,
unloading by the blue curves. The large drop in force and presence of
significant hysteresis in the pH 5 case is indicative of capsid failure.
(Figure adapted from [27].)
123
8998
Modeling viral capsids from the bottom up
Experimental methods such as X-ray crystallography and
cryo-EM used to determine the atomic structures of viral
capsids show that they are not entirely static. At the atomic
scale, thermal motions are detectable and the equilibrium
structure describes the average position of all atoms over
time. These structural methods are also able to detail changes
in the atomic positions in response to the binding of external
molecules, such as antiviral complexes, and the structural
rearrangements seen during conformational changes. In or-
der to understand both the equilibrium fluctuations and
structural changes, mechanical models of viral capsids were
first developed to study their unforced vibrational dynamic
behavior almost twenty years ago. Methods such as molec-
ular dynamics (MD) and normal mode analysis (NMA)
represent a bottom-up approach to modeling viral capsids;
the models were initially used with atomic level detail, and
more recently, coarse-grained models have been introduced.
Molecular dynamics
Beyond studying the equilibrium fluctuations of viral caps-
ids, MD has been used to model several specific types of
problems: assembly [31, 32], stability [33, 34], and antiviral
activity [3538], to name a few. MD is useful for the mod-
eling of very short time scale dynamics (on the order of pico
and nanoseconds), and captures both anharmonic and har-
monic motions. The main challenge in the application of MD
to viral capsids is the number of atoms in the system. There is
no general rule for estimating the number of atoms per pro-
tein subunit, and thus the number of atoms for different
viruses with the same T-number may widely vary. However,
among several representative T = 3 viruses, the total number
of atoms varies from about 150,000 to 250,000. Conse-
quently, most MD simulations performed on viral capsids are
limited to single subunits, with symmetry conditions
enforced so as to capture the icosahedral environment.
Viral capsids have been studied using MD for almost
20 years, with the first simulation, performed in 1988,
consisting of a single human T = 1 rhinovirus subunit in
vacuo, with an antiviral compound bound to the subunit in
order to determine the dynamic effects of the binding [39].
This simulation was performed on ~8000 atoms, and the
results showed large edge effects as no boundary condi-
tions were applied, and thus the icosahedral environment of
the entire capsid was not modeled. In 1991, a method was
developed to model the complete capsid environement by
taking advantage of the rotational symmetry boundary
conditions present in icosahedral viruses, and MD was
again applied to the human rhinovirus subunit in vacuo
[35], and fully solvated [36]. In this way, the physical
123
J Mater Sci (2007) 42:89959004
environment of the entire capsid is modeled, while the
smaller number of atoms in a single subunit is utilized. The
use of the rotational symmetry boundary conditions
showed a significant reduction in the edge effects as atoms
near the boundaries remained close to the crystal coordi-
nates. However, the total simulation time only varied from
15 ps [35] to 60 ps [36].
Until very recently, MD simulations performed on viral
capsids were limited to the asymmetric unit (the collection
of subunits that comprise one icosahedral unit, and in the
case of T = 1 capsids, a single subunit) alone, as detailed
above. The first molecular dynamics simulation on an en-
tire virus was completed in 2006 [40]. The simulation was
performed on the T = 1 satellite tobacco mosaic virus
(STMV). The T = 1 capsid structure is the smallest among
spherical viruses. The simulation was performed with both
the capsid proteins and a highly simplified model of the
nucleic acid, which was built from identical RNA segments
to match the icoshedrally averaged electron density of
RNA, in a completely solvated environment, with
~1,060,000 total atoms, ~900,000 of which were water.
The time scale of the simulation was 10 ns. The aim of the
simulation was to investigate the stability of the viral
capsid with and without the RNA present, and it was found
that the capsid without the RNA was unstable after 10 ns.
The capsid was, however, extremely stable with the RNA
present in the simulation. Unlike some viruses that have
been observed to self-assemble in vitro into empty capsids
(no nucleic acid packaged), empty STMV has not been
observed [40].
In order to perform MD on viral capsids, an all-atom
structure needs to be available for the system of interest.
These are not available for all viral capsids. All-atom
descriptions necessarily limit the size of the system that can
be studied, and at the time scales sampled by MD, large
deformations of a capsid, such as conformational changes,
cannot be modeled. Even the studies that use the rotational
symmetry boundary condition method to model only a
small fraction of the capsid are limited in the types of
motions that are output; any type of asymmetric or global
motion will not be captured. Bottom-up descriptions such
as MD are advancing very slowly in terms of their capa-
bility. The use of MD to model externally applied forces,
such as would be present in experimental methods such as
AFM, is probably not feasible. Experimental methods such
as AFM incur large deformations and occur on a time scale
of milliseconds to seconds. These deformation and time
scales are simply inaccessible by all-atom methods, unless
some form of coarse-graining is applied. Aware of these
limitations, a new coarse-grained MD method has been
recently developed to study the dynamic stability of larger
viral capsids, extending the timescale to the microsecond
range [65].
J Mater Sci (2007) 42:89959004
Normal mode analysis
Normal mode analysis (NMA) was developed as a more
computationally efficient alternative to MD. Initially, NMA
simulations were all-atom and involved the same interatomic
potential field as MD, but by using a quadratic approximation
for the potential energy, the problem is reduced to a normal
mode analysis on a structure and only outputs harmonic
motions. For an excellent reference on both the theory and
applications of NMA, see Cui and Bahar [41]. The potential
energy, or dynamical, matrix is formulated from the second
derivatives of the potential function. In this way, the time
propagation of MD is eliminated, and the computationally
expensive step is diagonalizing the dynamical matrix, after
which the output eigenvalues (frequencies) and eigenvectors
(normal modes) are easily extracted. However, for larger
systems, the diagonalization becomes a major bottleneck as
the size of the dynamical matrix grows proportionally to the
size of the system being studied. Thus, the use of all-atom
NMA was limited for many years to small proteins due to
computational limitations.
It has been shown that the lowest modes tend to repre-
sent coordinated global motions of the structure being
studied, and are therefore most relevant for large motions
such as conformational changes [42]. Generally, NMA is
used to find normal modes that capture conformational
changes of proteins [43], as an aid in the refinement of low-
resolution cryo-EM density data [4446], and to describe
the general unforced dynamics of macromolecules [47, 48].
As applied to viral capsid structures, NMA is used to study
maturation dynamics and conformational changes such as
capsid swelling [4951]. The time scales associated with
the NMA frequencies are longer than that of MD, but are
still limited to the microsecond range.
An early all-atom NMA study was performed on the rod
shaped tobacco mosaic virus (TMV), in which one full unit
that is necessary for symmetryseventeen subunits that
make up on turnwas modeled for the analysis [52]. The
standard CHARMM force field was used [53]. In 2005, van
Vlijmen and Karplus [54] reported the first atomic-level
NMA calculations on a full icosahedral virus capsid, also
using a full interatomic potential. However, all-atom NMA
calculations using standard interatomic potentials remain
quite rare due to the computational demands, the devel-
opment of simplified potentials, and the use of coarse-
grained models.
In 1996, Tirion [55] introduced a simplified pairwise
Hookean potential to replace the complex interatomic
potential, and the method became known as the elastic
network (EN) model. The potential involves a single
parameter that is equal for all interactions, thus the need for
all atoms is avoided. This allows for NMA to be applied to
coarse-grained models, and consequently for larger systems
8999
such as viral capsids to be studied. In addition, the intital
energy minimization required when using the complex in-
teratomic potential is also avoided. The EN model is useful
for the lowest modes of vibration as the high frequency
modes cannot be modeled well with simplified potentials.
In two studies, Tama and Brooks [50, 51] used coarse-
grained NMA with the EN model to study the conforma-
tional changes observed in icosahedral virus capsids. In
2002, they reported that a coarse-grained NMA of the
CCMV capsid, which undergoes a pH induced swelling
with ~10% size increase in the radial direction, showed that
just a few of the lowest symmetric normal modes con-
tributed to the swelling displacement [50]. The goal was to
model several pathway intermediate structures, to get a
better idea of the structural changes during swelling. The
CCMV capsid was coarse-grained using the Ca atoms, the
simplified Hookean potential was used, and the rotations-
translations of blocks (RTB) method [56, 57] was used to
simplify the problem by assuming each individual subunit
was rigid. The normal modes are computed as a linear
combination of the rotations and translations of the rigid
blocks. In 2005, the same methods were applied to study
conformational changes of other viruses [51]. It was shown
that for viruses with known conformational changes
(CCMV, HK97, and NxV) one symmetric mode captures
the bulk of the motion. The next few lowest symmetric
modes are needed to more accurately describe the known
conformational pathway. They argue that this shows the
method is robust, and can be predictively applied to other
viruses for which both conformational states are not
known, or for viruses for which it is not known if con-
formational changes exist. Also in 2005, Rader et al. [49]
applied coarse-grained NMA to study the conformational
changes that the HK97 virus undergoes during maturation.
It was shown that the first eleven modes capture over 98%
of the observed maturation pathway. The low-frequency
modes also showed regions of relative flexibility, at sub-
strate recognition sites, and relative stiffness, at anchors or
hinge sites. This study proposed a maturation pathway
ultimately leading to a cross-linked chain-mail complex.
If EN models are used in a normal mode analysis, at any
level of coarse-graining, the need for the initial minimi-
zation of the potential energy is eliminated, but the results
are less physical. As the leading constant in the Hookean
potential is the only tunable parameter in the model, the
value is found by fitting the normal mode results to
experimental vibrational data. There is then a loss of some
predictive power. Similarly, using NMA to predict the
pathways of large conformational changes is only possible
if the initial and final states are known, as it is not obvious
which of the lowest modes capture the relevant pathway.
The only exception to this might be viral capsid swelling,
which appears to involve only symmetric motion, and thus
123
9000
comprised of only the symmetric modes. However, van
Vlijmen and Karplus [54] argue that symmetric modes,
which Tama and Brooks [50, 51] use in their coarse-
grained NMA analyses of viral capsids, are not necessarily
the true pathway of a conformational change, and non-
symmetric modes may need to be studied to fully under-
stand the pathway.
It is fair to say that NMA is currently the best method
available to study unforced vibrational macromolecular
dynamics. Currently, all-atom NMA simulations are very
complex, so coarse-grained models are the preferred
method for large systems such as viral capsids. Not only
has NMA established coarse-graining as a useful step in
studying large systems, but as a necessary one as well.
Coarse-graining allows for systems without full atomic
descriptions to be studied, in addition to highlighting the
fact that the atomic details are not a main factor in deter-
mining the global motions. One of the main conclusions
that has been drawn from the numerous coarse-grained
NMA studies is that the overall shape of a structure gov-
erns its unforced vibrational dynamics [42]. This may have
far-reaching implications for modeling viral capsid systems
under forced conditions. In fact, it is reasonable to assume
that shape governs forced mechanics, and based on recent
computational studies of viral capsids where only the shape
is preserved, this appears to be precisely the case.
Modeling viral capsids from the top down
Recent single molecule experiments such as atomic force
microscopy have given rise to a number of new top-down
two- and three-dimensional continuum elasticity models
involving the mechanical response of viral capsids. As
discussed above, a major result of previous coarse-grained
NMA studies was that the most important factor deter-
mining a structures unforced vibrational response was its
shape, not atomic detail. Similarly, here it is argued that the
same is true for forced mechanical response, as the results
of several recent continuum studies of viral capsid
mechanics seem to support.
Two-dimensional continuum modeling
From the number of viral capsid structures determined
experimentally by X-ray crystallography and cryo-EM, it
has been observed that icosahedral capsids, which come in
a variety of sizes, have what appears to be a size dependent
shape character. Smaller viruses tend to appear more
spherical and smooth, whereas larger viruses tend to show
faceting and visible ridges. Recent computational models
of icosahedral capsids have been created on the basis of the
123
J Mater Sci (2007) 42:89959004
theory of thin elastic shells to study the shapes and shape
transitions of the icosahedral viruses. These models treat
the total energy of the system as a sum of both bending and
stretching contributions. Building off of previous work on
the stability of planar elastic sheets [58], Lidmar, Mirny
and Nelson (LMN) [59] hypothesized that the pentamers of
an icosahedral capsid act as fivefold disclinations, which
are necessary to form a closed shell from an otherwise
hexameric lattice. They showed that the degree of faceting
is dependent on the relative stiffness of bending and
stretching and the size of the shell. In particular, the energy
minimizing shape of the capsid shell transitions from
roughly spherical to icosahedrally faceted with an increase
in value of a dimensionless parameter, the Foppl-
vonKarman (FVK) number c = YR2/j, where Y is the two-
dimensional Youngs modulus, j is the bending rigidity,
and R is the average radius of the capsid. There is a critical
value of the FVK number at which a transition occurs that
can be described as buckling. For sub-critical FVK num-
bers bending stiffness dominates over stretching and the
capsid shape is roughly spherical. Above this transition,
the structure has a lower energy if the surface is bent near
the disclination, thereby relaxing the otherwise dominant
stretching energy. This effect is argued to be a cause for the
visible faceting. As the subunits of different viruses have
similar molecular structure, they should be expected to
have similar mechanical properties. Thus, the spherical
or faceted shape of a virus is predicted to scale with its size.
Following the LMN study, Nguyen et al. [60, 61] built on
the thin elastic shell theory to include spontaneous curva-
ture in an attempt to discover non-spherical equilibrium
capsid shapes, such as the conical capsid shape often
seen in HIV. The two-dimensional icosahedral models of
Zandi and Reguera [62] showed that in addition to the size
dependent shape character of viral capsids, there is a size
dependent stress character; the degree of stress concentra-
tion at the pentamer sites grows with increasing T-number.
Icosahedral models of viruses have very recently been
extended to model the mechanical response of capsids
under applied loads. Simulations of AFM experiments have
been performed to study the effect of the shape transitions
implied by the FVK number. Vliegenthart and Gompper
[63] simulated indentation of T-number lattice triangula-
tions, which for sufficiently large T-numbers are consistent
with the Foppl-vonKarman continuum elastic description.
Using a molecular dynamics approach to solve for equi-
librium shapes, they showed that the buckling instabilities
of fivefold disclinations can be excited during indentation,
leading to discontinuous drops in indentation force. Fur-
thermore Vliegenthart and Gommper argue based on their
results that size effects may be important for smaller
capsids, causing their force-indentation character to differ
from that predicted by continuum mechanics. Using a
J Mater Sci (2007) 42:89959004
thin-shell continuum finite element approach, Klug et al.
[27] recently modeled AFM experiments on the empty
CCMV viral capsid modeled as an icosahedron with
varying values of the FVK number. The shells with FVK
numbers below the buckling threshold (more round) show
linear behavior consistent with pH 6 experiments, while
the shells with FVK numbers above the buckling threshold
(more faceted) show sharp discontinuities in the force due
to inversion of buckled fivefold disclinations, reminiscent
of the failure of capsids seen experimentally at pH 5. In
particular it is argued that the pH indued swelling transition
of CCMV detailed earlier has the effect of softening the
mechanical response of the capsid by lowering the effec-
tive Youngs modulus of the capsid, and thus lowering the
FVK number. This is proposed as a mechanism responsible
for the presense and absense of failure in AFM indentation
experiments at pH 5 and 6 respectively.
The results of the LMN study support the notion estab-
lished by NMA studies [42] of a strong link between shape
and mechanical properties for macromolecular assemblies.
Indeed, whereas NMA studies have suggested that shape
governs mechanics, the LMN model represents an example
of the reverse, i.e., how mechanics can govern shape. The
results of Klug et al., and Vliegenthart and Gompper go
further by examining the link between shape and mechan-
ical response of icosahedral capsids under the influence of
externally applied mechanical forces. The one common
feature of these two-dimensional elastic shell models is the
assumption that because of their geometry the coat proteins
most naturally fit together in a planar hexagonal lattice,
which represents a stress-free reference configuration. Thus
the presence of fivefold disclinations necessarily introduces
pre-stresses in the closed capsid shell. This assumption re-
sides at a midpoint within the multiscale hierarchy of capsid
structure, at a length scale above that of primary and sec-
ondary molecular structure, but below that of the global
capsid assembly. It may perhaps be surprising then, that
other recent continuum models working purely at the global
structural scale also seem to do well in explaining features
of capsid mechanics observed in experiments.
Three-dimensional continuum modeling
As a key theoretical complement to the recent experimental
advances in single molecule experiments such as atomic
force microscopy (AFM), three-dimensional continuum
modeling has emerged as a simple but powerful tool in
understanding viral capsid mechanics. Because the problem
has complex boundary conditions and geometry, analytical
solutions are not feasible without generalizations that would
result in a loss of precision, and so finite element methods
have been used to simulate the AFM experiments.
9001
In 2004, the first AFM study on a viral capsid along
with continuum model elasticity studies was reported by
Ivanovska et al. [25]. The virus studied was the bacterio-
phage /29, which has icosahedral endcaps and a cylindrical
center section made from a ring of subunits arranged into
hexameric groups. It was shown that the empty capsids had a
linear response up to displacements of ~30% of the capsid
height, and was completely reversible unless the capsid was
displaced to the point of failure, at which drops in the contact
force were seen. In order to extract material parameters for
the capsid, a three-dimensional continuum elastic finite
element model was created, on which the AFM experiment
was simulated. The Youngs modulus was varied until the
slopes of the experimental and simulated contact force
curves matched. Similar studies were performed on the plant
virus CCMV and subsequently modeled using three-
dimensional continuum elastic finite elements by Michel
et al. [26], and Gibbons and Klug [64]. It was shown by
Gibbons and Klug in a series of parametric studies that the
observed linear behavior may be understood as a combina-
tion of several geometric effects; mainly that the thickness
of the capsid and size of the AFM tip (which is on the order
of the capsid size itself) largely determine the mechanical
response to applied loads [64]. Interestingly, the response
was not sensitive to the constitutive law chosen, as long as
large strain measures were used. (The use of linearized small
strains was shown to be inadequate for such analyses.) From
these studies, it appears that the geometry of the capsid and
AFM tip, rather than local material response, have the most
impact on the resulting force curves.
Three-dimensional continuum models were also used to
model immature and mature murine leukemia virus (MLV)
capsids [29]. AFM studies on the immature and mature
capsid showed quantitative and qualitative differences in
the contact force response curves; the immature capsid is
20 nm thick, and showed a response similar to Hertzian
contact behavior, while the mature virus, which is only
4 nm thick, showed a linear contact force response. The
nominal external radius in both cases is the same, at
~50 nm. These results agree very well with the contact
force curves of the AFM experiments.
Similar AFM studies were performed on the minute virus
of mice (MVM) with and without the DNA genome enclosed
[28]. The studies showed that the capsid without the genome
showed force responses that were equal for all three sym-
metry faces, but showed an anisotropic increase in stiffness
when the DNA was enclosed. A corresponding continuum
model was created in which the capsid was modeled as an
icosahedral shell with the average thickness of the capsid,
with the DNA modeled as circular disks that add thickness to
the capsid at different points on a given face of the icosa-
hedron. They were able to show the anisotropic increase in
stiffness with the capsid and DNA model.
123
9002
Continuum finite element models for both empty and full
capsids have served as a sufficient means of capturing some
of the most important mechanical behaviors observed
experimentally. Although icosahedral viruses can have very
complex topographies, these features do not appear to play a
main role in the observed elastic response of the capsids;
rather, the details will more likely play an important role in
failure mechanisms, which may depend more heavily on the
specific regions that are weak, and are likely to be directly
related to the strength of individual capsomercapsomer
interactions. The studies cited above have shown that con-
tinuum models that take into account the overall shape of
viral capsids do quite well at reproducing experimentally
observed contact force curves. This serves as an inspiration
for further continuum and coarse-grained approaches,
specifically to investigate mechanisms of failure.
Conclusions
Molecular dynamics and normal mode analysis have
revealed important mechanical properties of viral capsids
by studying the unforced vibrational dynamics. One of the
main conclusions of molecular modeling is that the overall
shape of viral capsids, and not the atomic detail, governs
the unforced mechanical response. In particular, coarse-
grained NMA models are able to capture the dynamic
behavior of viral capsids, such as conformational changes,
without the fine scale detail used in MD simulations.
Despite the success of these bottom-up methods, there are
limitations to both. Computational expense is a large
problem that has only recently been addressed with coarse-
grained methods, the time scales on which both MD and
NMA are valid are still quite small, and perhaps most
importantly, the methods are limited to unforced dynamic
problems. The latter limitation excludes the study of a
range of problems for which experimental data is now
available, such as the newly developed AFM experiments
that were detailed above. Additionally, the trend towards
coarse-grained models has moved the models closer to
continuum descriptions, and while the ability to describe
movements at atomistic detail is lost, the bulk dynamic
behavior is retained, and this suggests that continuum
models may provide insights into the forced mechanical
behavior of viral capsids without a loss of physicality.
It has been demonstrated that continuum models provide
a wealth of information about the response of viral capsids
to applied loads. By modeling the global shape of viral
capsids, without atomic level structural information, both
thin and thick shell mechanics seem to consistently predict
what is observed experimentally. Results of several con-
tinuum finite element studies have demonstrated that the
shell-like geometry of the capsid is the most important
123
J Mater Sci (2007) 42:89959004
factor in determining the mechanical response to external
forces, while the models are not highly dependent on the
constitutive modeling details. These results are comple-
mentary to the results of the molecular mechanical models,
which have shown that the details of the atomic structure
and interatomic potentials are not essential to capture the
unforced vibrational behavior. In particular it is important
to recognize the consistency between the insensitivity to
constitutive law for continuum models and the insensitivity
to interatomic potentials for the molecular models. This
provides some justification for the modeling assumption of
constitutive homogeneity. It may seem incredible that
continuum elasticity theory would be meaningful for a
nanometer-scale molecular structure such as a viral capsid.
Yet, the results of recent modeling studies establish that
this is precisely the case.
Yet, caution should be exercised in celebrating the
sucesses of continuum theory. The established continuum
models of capsid mechanics are not without limitations, the
largest of which is the exclusion of dynamic behavior. In
particular, thermal forces and fluctuations are not consid-
ered, and although these details may not be important to the
quasi-static global mechanical response of the viral capsids,
time sensitive behaviors are not captured. Dynamic behav-
iors are extremely important when studying assembly and
disassembly, and viscoelastic/rheological behavior. Fur-
thermore, in the quest for deeper understanding of the
material properties of capsid shells, some questions that
simple continuum elasticity models have difficulty answer-
ing will need to be addressed. For instance, elasticity isby
definitionincapable of explaining the irreversible damage
observed in the experiments on CCMV [26] and /29 [29];
what are the mechanisms that lead to irreversible failure at
one pH, and nigh-invulnerability at another? More generally,
how are the microscopic conformational changes of
individual proteins manifested in the macroscopic struc-
tural mechanics of a large protein aggregate? Perhaps con-
tinuum methods can be stretched and extended via multiscale
modeling to provide answers to these questions.
As informative as bottom-up methods such as MD and
NMA are on short time scales and at atomic levels, top-
down continuum models hold the most promise for
understanding viral capsid behavior under the variety of
applications they now serve in. Yet to overcome the limi-
tations of the current continuum methods, incorporating
atomic level detail when needed may be necessary. Thus,
the development of multiscale methods will most likely be
needed to understand complex dynamic behavior of viral
capsids, such as conformational changes. NMA has been
used to describe the pathway of such deformations, but
perhaps continuum and multiscale methods will deepen the
understanding about the molecular mechanisms behind
these behaviors. For example, continuum models have
J Mater Sci (2007) 42:89959004
been used to study the conformational changes of the
CCMV capsid, and have shown that relevant modes may be
exictable when forces are applied, and this exictation will
manifest as a change in material properties [27]. The exact
molecular mechanisms responsible are as of yet unknown.
Additionally, multiscale models will be needed to model
the forced mechanical response of viral capsids beyond the
elastic regime, to study the failure of viral capsids, as
failure will be initiated at the atomic level due to the
breaking of bonds.
Note: Just before this article went to print, an article
appeared detailing AFM experiments on the bacteriophage
k, studying the effects of packed genome on the strength of
the capsid [66]. We would also like to report that this same
group is preparing an article on the subject of viral capsid
failure.
References
1. Lee SW, Mao C, Flynn CE, Belcher AM (2002) Science 296:892
2. Blum AS, Soto CD, Wilson JD, Cole M, Kim, Gnade B, Chatterji
A, Ochoa WF, Lin T, Johnson JE, Ratna BR (2004) Nano Lett
4(5):867
3. Strable E, Johnson JE, Finn G (2004) Nano Lett 4(8):1385
4. Douglas T, Young M (1998) Nature 393(6681):152
5. Douglas T, Strable E, Willits D, Aitouchen A, Libera M, Young
M (2002) Adv Mater 14:415
6. Bancroft JB, Hiebert E (1967) Virology 32:354
7. Speir JA, Munshi S, Wand GJ, Baker TS, Johnson JE (1995)
Structure 3(1):63
8. Morais MC, Choi KH, Koti JS, Chipman PR, Anderson DL,
Rossmann MG (2005) Mol Cell 18:149
9. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt
DM, Meng EC, Ferrin TE (2004). J Comput Chem 25(13):1605
10. Berman HM, Henrick K, Nakamura H (2003) Nat Struct Biol
10(12):980, http://www.pdb.org/
11. Brooksbank C, Camon E, Harris MA, Magrane M, Martin MJ,
Mulder N, ODonovan C, Parkinson H, Tuli MA, Apweiler R,
Birney E, Brazma A, Henrick K, Lopez R, Stoesser G, Stoehr P,
Cameron G (2003) Nucl Acids Res 31:43, http://www.ebi.ac.uk.
12. Caspar DL, Klug A (1962) Cold Spring Harb Symp Quant Biol 27: 1
13. Ganser-Pornillos BK, von Schwedler UK, Stray KM, Aiken C,
Sundquist WI (2004) J Virol 78:2545
14. Branden C, Tooze J (1999) Introduction to protein structure.
Garland Publications, New York
15. Wilk T, Gross I, Gowen BE, Rutten T, de Haas F, Welker R,
Krausslich H-G, Boulanger P, Fuller SD (2001) J Virol 75:759
16. Schwarcz WD, Barroso SP, Gomes AM, Johnson JE, Schnee-
mann A, Oliveira AC, Silva JL (2004) Cell Mol Biol 50:419
17. Wikoff WR, Liljas L, Duda RL, Tsuruta H, Hendrix RW, John-
son JE (2000) Science 289:2129
18. Liu H, Qu C, Johnson JE, Case DA (2003) J Struct Biol 142:356
19. Michels B, Leimkhler M, Lechner MD, Adrian M, Lorber B,
Witz J (1999) Eur J Biochem 264:965
20. Li PP, Naknanishi A, Tran MA, Ishizu K-I, Kawano M, Phillips
M, Handa H, Liddington RC, Kasamatsu H (2003) J Virol
77:7527
21. Smith DE, Tans SJ, Smith SB, Grimes S, Anderson DL, Busta-
mante C (2001) Nature 413(6857):748
9003
22. Kindt J, Tzlil S, Ben-Shaul A, Gelbart WM (2001) PNAS
98(24):13671
23. Tzlil S, Kindt J, Gelbart WM, Ben-Shaul A (2003) Biophys J
84:1616
24. Purohit PK, Kondev J, Phillips R (2003) PNAS 100(6):3173
25. Ivanovska IL, de Pablo PJ, Ibarra B, Sgalari G, MacKintosh FC,
Carrascosa JL, Schmidt CF, Wuite GJL (2004) Proc Natl Acad
Sci USA 101:6700
26. Michel J-P, Ivanovska IL, Gibbons MM, Klug WS, Knobler CM,
Schmidt CF, Wuite GJL (2006) Proc Natl Acad Sci USA
103(16):6184
27. Klug WS, Bruinsma RF, Michel J-P, Knobler CM, Ivanovska IL,
Schmidt CF, Wuite GJL (2006) Phys Rev Lett 97:228101
28. Carrasco C, Carreira A, Schaap IAT, Serena PA, Gomez-Herrero
J, Mateu MG, de Pablo PJ (2006). Proc Natl Acad Sci USA
103(37):13706
29. Kol N, Gladnikoff M, Barlam D, Shneck RZ, Rein A, Rousso I
(2006) Biophys J 91(2):767
30. Schaap IAT, Carrasco C, de Pablo PJ, MacKintosh FC, Schmidt
CF (2006) Biophys J 91(4):1521
31. Reddy VS, Giesing HA, Morton RT, Kumar A, Post CB, Brooks
CL, Johnson JE (1998) Biophys J 74(1):546
32. Zlotnick A, Johnson JM, Wingfield PW, Stahl SJ, Endres D
(1999) Biochemistry 38:14644
33. Warwicker J (1992) J Mol Biol 223:247
34. van Vlijmen HWT, Curry S, Schaefer M, Karplus M (1998) J Mol
Biol 275:295
35. Cagin T, Holder M, Pettitt BM (1991) J Comput Chem 12(5):627
36. Yoneda S, Kitazawa M, Umeyama H (1996) J Comput Chem
17(2):191
37. Phelps DK, Post CB (1995) J Mol Biol 254:544
38. Phelps DK, Rossky PJ, Post CB (1998) J Mol Biol 276:331
39. Lau WF, Pettitt BM, Lybrand TP (1988) Mol Simul 1:385
40. Freddolino PL, Arkhipov AS, Larson SB, McPherson A, Schulten
K (2006) Structure 14:437
41. Cui Q, Bahar I (eds) (2005) Normal mode analysis: theory and
applications to biological and chemical systems. Chapman &
Hall/CRC, Boca Raton
42. Tama F, Brooks CL III (2006) Annu Rev Biophys Biomol Struct
35:115
43. Tama F, Sanejouand Y-H (2001) Protein Eng 14:1
44. Ming D, Kong Y, Lambert MA, Huang Z, Ma J (2002) PNAS
99(13):8620
45. Tama F, Wriggers W, Brooks CL III (2002) J Mol Biol
321(2):297
46. Delarue M, Dumas P (2004) PNAS 101:6957
47. Go N, Noguti T, Nishikawa T (1983) Proc Natl Acad Sci USA
80:3696
48. Brooks BR, Karplus M (1983) Proc Natl Acad Sci USA 80:6571
49. Rader AJ, Vlad DH, Bahar I (2005) Structure 13:413
50. Tama F, Brooks CL III (2002) J Mol Biol 318:733
51. Tama F, Brooks CL III (2005) J Mol Biol 345(2):299
52. Simonson T, Perahia D (1992) Biophys J 61:410
53. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan
S, Karplus M (1983) J Comput Chem 4(2):187
54. van Vlijmen HWT, Karplus M (2005) J Mol Biol 350:528
55. Tirion MM (1996) Phys Rev Lett 77(9):1905
56. Durand P, Trinquier G, Sanejouand Y-H (1994) Biopolymers
34:759
57. Tama F, Gadea FX, Marques O, Sanejouand Y-H (2000) Pro-
teins: Struct, Funct, Genet 41:1
58. Seung HS, Nelson DR (1988) Phys Rev A 38(2):1005
59. Lidmar J, Mirny L, Nelson DR (2003) Phys Rev E 68:051910
60. Nguyen TT, Bruinsma RF, Gelbart WM (2005) Phys Rev E
72:051923
123
9004 J Mater Sci (2007) 42:89959004

61. Nguyen TT, Bruinsma RF, Gelbart WM (2006) Phys Rev Lett 64. Gibbons MM, Klug WS (2007) Phys Rev E 75:031901.
96:078102 65. Arkhipov A, Freddolino PL, Schulten K (2006) Structure 14:1767
62. Zandi R, Reguera D (2005) Phys Rev E 72(2):021917 66. Ivanovska I, Wuite G, Jonsson B, Evilevitch A (2007) Proc Natl
63. Vliegenthart GA, Gompper G (2006) Biophys J 91(3):834 Acad Sci USA 104:9603

123
J Mater Sci (2007) 42:90059014
DOI 10.1007/s10853-007-1936-8
NANO- AND MICROMECHANICAL PROPERTIES OF HIERARCHICAL BIOLOGICAL MATERIALS
The structural efficiency of orthotropic stalks, stems and tubes
Ulrike G. K. Wegst Michael F. Ashby
Received: 31 January 2007 / Accepted: 12 June 2007 / Published online: 7 August 2007

Springer Science+Business Media, LLC 2007
Abstract An optimised structure is one which uses the
smallest quantity of the best material to perform its
function, with adequate safety factor or margin for error.
Structural optimisation occurs not only in mechanical
engineering, but also in nature: plants with hollow stems or
stalks gain a height advantage, and are thus more efficient,
by approaching the optimum shape. Here we consider the
optimisation of orthotropic tubes, typifying, in a mechan-
ical sense, stalk and stem. The stiffness and strength of
orthotropic tubes of initially circular section are reviewed,
and diagrams are proposed which allow the optimum
section shape to be selected.
Introduction
In creating structures to carry loads, hollow, thin-walled
sections can be more efficient than thick-walled or solid
ones. They use less material and are therefore lighter (more
economical) while resisting the same bending or tor-
sional load; and this is true whether the design is based on
stiffness or on strength. When the mode of loading is
bending and the direction of loading is unknown, circular
tubes are better than other shapes. And if nature is to act as
a guide, circular tubes which are orthotropic are better than
those that are isotropic using the same amount of material.
An orthotropic tube is one in which the modulus and
U. G. K. Wegst (&)
Department of Materials Science and Engineering, Drexel
University, 3141 Chestnut Street, Philadelphia, PA 19104, USA
e-mail: uwegst@coe.drexel.edu
M. F. Ashby
Department of Engineering, Cambridge University,
Trumpington Street, Cambridge CB2 1PZ, UK
strength parallel to the tube axis differ from those in the
circumferential direction; if this difference is chosen
properly, the orthotropic tube is both stiffer and stronger in
bending than the equivalent isotropic one. Orthotropic
tubes are exploited both by engineers (composites, highly-
drawn metals) and by nature (stalks, stems, bamboo culms)
and are almost invariably structured so that the stiff, strong
direction lies parallel to the axis of the tube. The question
we address here is this: how can the tube shape (that is, the
ratio of wall-thickness to tube radius) and the anisotropy
ratio (ratio of axial to radial modulus and strength) be
optimised to maximise the performance?
To answer this requires a study of the potential failure
modes of the tube. When a tube is bent, its section tends to
ovalise, loosing stiffness. Bent far enough, it fails in mode
1: ovalising and kinking with catastrophic stiffness
losslike a plastic drinking straw, bent until it collapses.
But bending also creates tensile and compressive stresses
in the tube wall; if either of these exceeds the uniaxial
strength of the tube wall, the tube fails in mode 2: tensile
yield or fracture, or compressive collapselike a stick of
celery, bent until it snaps. Finally, ovalisation has another,
subtler, consequence: it creates circumferential stresses in
the tube wall which, if they exceed the circumferential
strength (almost always the lower one), cause mode 3
failure: longitudinal splittinglike a stick of celery,
pinched between the fingers.
Observations and analyses of the bending response of
thin-walled orthotropic tubes appear in two quite separate
bodies of literature. That relating to plant stems and stalks
is largely experimental [114]. That focusing on light-
weight engineering structures, and particularly on poly-
mer-composite tubes are predominantly analytical, and
generally treat only one aspect of what is a multi-faceted
problem [1526]. Only one paper [27] attempts, as we do
123
9006 J Mater Sci (2007) 42:90059014

here, a comprehensive survey of competing failure modes mode, but we shall ignore this and identify the elastic
and explores how well plantsparticularly bambooare collapse moment with the Brazier moment.
structured to combat them. We now review the stiffness To find it, we follow the method and notation of [29].
and strength of thin-walled orthotropic tubes in more Ovalisation is measured by f, the cross-sectional shape-
detail, allowing for ovalisation during bending and for change parameter (Fig. 1):
four distinct failure modes. Much of the analysis closely
r
a
parallels that of [27] and will therefore be kept brief. The 1 5
r
recasting of all of the results in terms of section area and
shape, and the novel construction and optimisation, both where r and a are the radius of the original circular shape
of shape and of anisotropy this allows, are new. and the minor axis of the ellipse respectively.
Then the total strain energy per unit length, U, of an
Bending of thin-walled orthotropic tubes isotropic tube which has been deformed into an arc of
curvature C, replacing Eq. 4, becomes
In designing a thin-walled tube to carry bending moments, 

1 3 2 3 5 2 3 pt3 2
two sets of consideration arise. The first set relates to U pr tEC 1
1 1 1E 6
2 2 8 8 r
stiffness, the second to the collapse moment (or strength).
The first term of the right hand side of this equation
Stiffness, ovalisation and the Brazier moment for describes the strain energy due to longitudinal stretching,
orthotropic tubes the second that due to ovalisation (circumferential bending).
For an orthotropic material this expression is modified
The curvature, C, of an elastically isotropic beam is related by replacing E with the longitudinal modulus, Ejj , in
to the bending moment, M, which it carries by the longitudinal stretching term and with the transverse
modulus, E? , in the ovalisation term.
M IEC 1


where E is the Youngs modulus of the material of which 1 3 2 3 5 2 3 pt3 2
U pr tEjj C 1
1 1 1 E? 7
the beam is made and I is its second moment of area. We 2 2 8 8 r
focus on a beam which is a thin-walled tube of initial Omitting the f2 term enclosed in square brackets (justified
radius, r, and wall-thickness, t, with cross-section A 2prt. in a moment), we seek the ovalisation, f, for a given
If the bending is slight, the cross-section remains circular, curvature, C, which minimises the total energy, U. Setting
when dU=df 0 leads to

I pr 3 t 2  
r 4 Ejj 4
1 2 C 2 c2 8
Then the moment is related to the curvature by t E? 3
where the dimensionless curvature, c, is defined by
M pr 3 tEC 3
 
and the elastic strain energy per unit length of the tube is 3 r 4 Ejj
c2 C2 9
4 t 2 E?
1
U pr 3 tEC 2 4 Substituting Eqs. 8 and 9 (inverted) for f and C2 in Eq. 7,
2 gives the strain energy, U, as a function of c alone
However, if the bending is substantial, the tensile and
compressive stresses in the tube walls due to its longitu- 2p t3  2 
U E? c
c 4 10
dinal curvature cause the cross-section to ovalise as in 3 r
Fig. 1, and this has several consequences. The ovalisation From this we calculate the bending moment, M Cc dU
dc , as
reduces the second moment of area, I; further bending
causes the ovalisation to increase and the stiffness to 2p  1=2  
decrease, until a maximum bending moment, the Brazier M p rt2 Ejj E? c
2c3 11
3
moment, is reached and catastrophic failure follows [28].
In practice tubes fail at bending moments which are a little Equation 11 is the generalisation of the simple Eq. 3 for
less than this because local defects trigger a local buckling orthotropic tubes, including the effect of ovalisation, and

123
J Mater Sci (2007) 42:90059014
Fig. 1 Schematic representation of the ovalisation of a tube loaded in
bending. The radius of the original circular cross-section is r0 and the
minor axis of the ovalised section is a
reduces to it when Ejj E? and c  1 . The maximum
Mthe Brazier moment MBis found by setting the
differential of Eq. 11 with respect to C equal to zero
dM dM c
0 12
dC dc C
p  2 1=2  1=4  2 1=2
giving the critical value c 1= 6 which, when reinserted
into Eq. 11 gives:
p with the same limit (W < 0.5) required to suppress ovali-
2 2 2 1=2
MB prt Ejj E? 13
9
Strategies exist for suppressing ovalisation. Bamboo culm
(to take an example from nature) is an orthotropic tube
which is divided into chambers by stiff diaphragms which
are positioned at the nodes from which the leaves grow. It
has been considered [1, 4, 9, 11, 27, 3033] that these
might act as ring stiffeners, suppressing ovalisation. The
isotropic tube with periodic stiffeners is well treated in the
mechanics literature (see for example [29, 34]).
Consider a tube of finite length, L, thickness, t, and
radius, r, which is fitted with end-pieces which preserve the
original circular shape (thus L becomes the stiffener
spacing or internode length, as in Fig. 2a). Calladine [29]
derives the dimensionless group
 2 1=2
tL
X 14
r3
which characterises the problem: if W < 0.5, ovalisation is
largely suppressed, stiffening the tube and local buckling
occurs when the critical stress is reached to cause
axisymmetric buckling under uniaxial compression on the
compressive face of the tube [35]; for 0:5\X\2 , ovali-
sation at the point of local buckling increases with an
increase in W and the critical moment decreases until at
W 2 it is approximately equal to that at failure of an
9007
infinitely long tube. Thus for W > 0.5 the end-pieces or
diaphragms have little influence.
Orthotropy changes this. A high value of the circum-
ferential modulus E? makes ovalisation difficult; a low
value makes it easy. A tube for which Ejj =E? > 1 ovalises
more, for the same curvature C, than one for which the
opposite is true; to prevent the ovalisation becoming crit-
ical, the stiffeners must be placed closer together. Suo [36,
37] introduces an orthotropic scaling factor, the anisotropy
ratio, n, which takes account of this phenomenon
Ejj
n 15
E?
With it, the effective length, k, of the internode can be
calculated
 1=4
1=4 Ejj
kn L L 16
E?
and the expression for W for an orthotropic tube becomes
tk Ejj tL
X 17
r3 E? r3
sation. In plants a typical anisotropy ratio n is 10 [38].
Taking this value, the effective length of an internode is
increased by a factor of 1.8 so that the ovalisation even of
relatively short internodes with either L/r = 1.7 and r/t = 6
or with L/r = 1 and r/t = 2 cannot be prevented.
A second schemeone seen in many plant stemsis
that of filling the tube with a foam (Fig. 2b). A low-density
foam is effective in suppressing local buckling [39, 40],
which is a bifurcation, but less so in suppressing ovalisa-
tion, which is not. But since local buckling occurs before
the Brazier moment is reached, foam-filling provides a way
of raising the failure moment without greatly increasing the
mass.
A third strategyone seen in bone, particularly that of
birdsis to insert a network of transverse struts, forcing
the tube to remain circular, until the struts fail. Provided
the struts are separated by no more than a few tube-radii
they can, like the ring-stiffeners, successfully suppress
ovalisation.
Suppressing ovalisation is important: not only does it
increase the bending stiffness of the tube, but it also
inhibits two of the three failure modes discussed next.
Failure of thin-walled orthotropic tubes
The tube can fail in one of three ways.
123
9008 J Mater Sci (2007) 42:90059014

M2 pr 2 trcjj 20

Failure by longitudinal splitting

The last failure mode is a little harder to analyse. A con-

sequence of a fibre lay-up which increases longitudinal
stiffness and strength is that the transverse properties
suffer. In particular, the transverse tensile strength (which
we shall call rt? ) of near-uniaxial fibre composites can be
so low that the tube, when bent, splits along its length. The
tensile stress which causes this arises from the ovalisation,
as sketched in Fig. 3: it is greatest on the inner surface of
the tube at the point A and on the outer one at B. It is
calculated by imagining that ovalisation is caused by two
point forces, W, which generate a moment M* at B [41],
Fig. 2 Schematic showing suppression of ovalisation (a) by dia-
phragms and (b) by a foam core
where

Failure by ovalisation, instability and local kinking M  0:18  Wr 21

The point loads also lead to an ovalisation of the cross-
This, as described above, occurs at a moment M1 which is
section [41] with a change in vertical diameter Dv, which
sufficiently close to the Brazier moment MB (Eq. 13) curve
we describe by the dimensionless cross-sectional shape-
that it can be approximated by this value:
change parameter as before
M1 MB 18
Dv 0:149  Wr 3 Wr 2
1 0:89  22
If the tube is loaded with a moment larger than this, a kink 2r 2rEI E? t 3
forms in the tube wall which leads to catastrophic failure
Neglecting the small additional compressive stresses, W/2t,
by buckling.
which are due to the point loads, we find that the maximum
circumferential stress is related to the bending moment by
Failure by tensile fracture or compressive collapse

The maximum stress, rmax, in the wall of a thin-walled tube 1

M  r?max t2 23
of radius r and thickness t with a second moment of area 6
I pr 3 t , loaded by a bending moment, M, is: Rearranging and substituting Eq. 22 for Wr in Eq. 21 and
equating the result with Eq. 23 yields a correlation between
Mr M the maximum transverse stress in the tube wall and the
rmax   2 19
I pr t ovalisation
It is tensile on the side of the tube furthest from the centre
of curvature, and compressive on the other side. If the
tensile stress exceeds the strength in tension of the material
of which the wall is made, the tube will fracture. If, instead,
the compressive stress exceeds the compressive collapse
load of the tube wall, failure again follows. Both are treated
by equating Eq. 19 to rf, where this is the tensile failure
stress or the compressive collapse stress, whichever is
smaller. Both natural and man-made orthotropic tubes are,
very frequently, composites with fibres which lie
predominantly parallel to the tube axis. Such structures
fail in compression by fibre buckling at a stress which is
lower than the tensile strength, and for this reason we will
for the moment identify rf with the compressive failure Fig. 3 Schematic of a bamboo cross-section, illustrating the way in
strength parallel to the tube axis, rcjj , giving: which the tensile stress which causes splitting is calculated

123
J Mater Sci (2007) 42:90059014 9009

r?max r Rearranging Eq. 18 (with (13)) give an expression for

1 0:82  24
E? t failure by buckling:
The final step is to link the ovalisation to the curvature  
M1 1 Ejj E? 1=2
which is its real origin. Equating this expression for the p 29
ovalisation with the one we obtained in the buckling A3=2 9 p /
analysis above (Eq. 8) we find a correlation between the That for tensile failure/compressive collapse (Eq. 20) now
maximum transverse stress, rmax, and the dimensionless reads:
curvature
M2 1
 1=2 3=2
p rcjj /1=2 30
r?max r A 8p
c 0:62  25
E? t
and that for splitting (Eq. 26) becomes:
Equating r?max with the transverse tensile strength of the

 
tube wall, rt? , and substituting Eq. 25 for c in Eq. 11 we M3  1=2 rt?
find a new general expression for the bending moment for 0:18  rt? Ejj 1
1:24 / 31
A3=2 E?
longitudinal splitting of the tube:

 
 1=2 r rt? Failure maps
M3 2:86  rt3=2 Ejj rt? 1
1:24 26
t E?
Figure 3 shows the construction of a failure map for
As expected, M3 decreases as the transverse strength, rt? , orthotropic tubes without stiffeners, loaded in bending.
decreases. Three curves are plotted in Fig. 4a, corresponding to each
of the three failure modes: buckling (Eq. 29), fracture/yield
(Eq. 30), and splitting (Eq. 31).
Optimisation of shape and anisotropy: failure maps The overall failure moment of the tube is defined by the
lower envelope of these curves, with one qualification,
Failure mechanisms compete. The dominant mechanism is detailed below. A change of mechanism takes place where
the one which occurs first as the bending moment M on the the curves intersectthat is, at the points A, B and C. The
tube is progressively increased; this usually means that it is first of these occurs at the value of / at which failure by
the one with the lowest value of the failure moment fracture/yieldEq. 30and by splittingEq. 31have
M1 ; M2 ; M3 ; etc: , though there is one exception. We now the same value of M/A3/2. The failure map of Fig. 4c
explore this competition, seeking first the optimum shape illustrates that the point A lies well to the left of the
for the tube, then the optimum anisotropy ratio. To do so, downward-curving segment of the M3 curve, which is
we express the geometric variables (contained in I) in such caused by the term in square brackets in Eq. 31. Neglecting
a way as to separate section-area this term, equating the two and solving for / gives:
!
A 2prt 27 rt? Ejj
/A 0:81  32
from section-shape characterised by the shape factor, /: r2cjj

(a full solution of /A is given in Appendix A). The second

4pI r change of mechanism occurs at the value of / at which
/ 2 28
A t bucklingEq. 29and fracture/yieldEq. 30have the
same value of M/A3/2. Equating these and solving for / gives:
The shape factor / is dimensionless; beams with the same
/ have cross-sections which may differ in size, but have  1=2
the same proportions: their cross-sections look like pho- Ejj E?
/B 0:31  33
tographic reductions or enlargements of each other. The rcjj
form of the expressions for the failure moments M1 ; M2
and M3 suggests a plot of M/A3/2 (with units of GPa) However, since the splitting curve (Eq. 31) has the lowest
against the shape factor, / = r/t, to examine the compe- values of M/A3/2 for /A \/\/C in Domain (2), failure
tition in failure modes. (Dividing the moment by A3/2 en- occurs by splitting.
sures that we are comparing tubes of similar cross-sectional The situation in Domain (3) requires a little more
area and same mass per unit length.) explanation. At a first glance one might think that, as in

123
9010 J Mater Sci (2007) 42:90059014

by splittingalthough splitting may subsequently occur.

Figure 3b helps to explain this: it is a graph showing the
bending moment, M, plotted against dimensionless curva-
ture, c, and illustrates the sequence of events. On bending, M
increases with c until the Brazier moment, MB, is reached,
beyond which it decreases. For values of / less than /C, the
condition for splitting is met before the Brazier moment is
reached, and splitting therefore determines the failure
moment in Domain (2). But for values of / greater than /C,
the tube is so thin-walled that the Brazier moment is reached
first, before the stresses in the tube wall are large enough to
cause splitting. Thus the tube then buckles, and splitting, if it
occurs at all, is a secondary event. The transition occurs at the
value of / at which the M3 is tangent topM 1, when, as shown
in the development of Eq. 13, c 1= 6 and, from Eq. 25
(with r/t = / and r?max rt? )
 
E?
/C 0:27  34
rt?

These boundary values of / allow the failure map to be

completed. The result is shown in Fig. 4c. It displays the
fields of dominance of each failure mechanism. Failure
maps are material-specific, that is, they are calculated
for given sets of the four key material properties,
Ejj ; E? ; rcjj and rt? ; the values used for Fig. 4 are listed
in Table 1.

Optimisation of shape

The best shape for the tube is that whichfor a given

cross-section Acarries the greatest moment M. We
therefore seek the value of / at which M/A3/2 is a maxi-
mum. For the conditions of Fig. 4c the answer is simple:
the best shape is that corresponding to /A, and since this
lies well to the left of the downward-curving part of the M3
curve, it is well approximated by Eq. 27.
There is another possibility. As rt? increases, splitting
becomes more difficult, the M3 curve moves upwards and
the splitting domain shrinks in size and finally disap-
pears. At the point at which this happens

/A /B 35
Fig. 4 (a) A failure map of loading coefficient M/A3/2 plotted against
the shape factor / for the three failure curves M1 (fracture/yield), M2 giving (by Eqs. 32 and 33) the condition for the transition
(buckling) and M3 (splitting). The intersections at I, II and III indicate
the transitions from one failure mechanism to another. (b) A graph to no-splitting as:
showing bending moment, M, plotted against dimensionless curva-
ture, c. (c) A failure map showing the three failure domains fracture/  1=2
yield (1), buckling (2) and splitting (3)
rt? E?
0:38  36
rcjj Ejj
Domain (2), splitting determines the failure moment (since
the lowermost curve is that of M3), but this is not so. The When this condition is met, the shape which gives the
peak moment in Domain (3) is set by the Brazier moment, not largest value of M/A3/2 is that corresponding to /B.

123
J Mater Sci (2007) 42:90059014 9011

Table 1 Data used in plotting

Youngs Modulus Compressive Strength Tensile Strength
Figs. 3 and 5: Youngs modulus
parallel and perpendicular to the (parallel) Ejj (perpendicular) E? (parallel) rcjj (perpendicular) rt?
tube axis, compressive strength [GPa] [GPa] [MPa] [MPa]
parallel to the tube axis and
tensile strength perpendicular to Green
the tube axis Low 5 0.5 45 4
Medium 10 1 50 5
High 15 1.5 55 6
Dry
Low 10 1 100 8
Medium 15 1.5 115 10
High 20 2 130 12

In the practice of engineering design, gradual or be-
nign failure modes, such as yielding, are viewed with less
alarm than those which are sudden or catastrophic,
among which are buckling and splitting. Responding to
this, the designer applies a greater safety factor to cata-
strophic modes than to those which are benign, and there is
some evidence (see below) that nature may do the same.
The practical consequence of this is to shift the optimum
shape a little to the left on Fig. 4c, that is, to slightly
smaller values of /.
anisotropy is possible by adjusting the lay-up. It is then
reasonable to ask: what is the best value for the
anisotropy? To answer this, consider an idealised
example: that of a tube made of fibres of modulus Ef
and strength rf, in a matrix of modulus Em and strength
rm. Let a fraction c of the total volume fraction of fibres,
f, lie parallel to the tube axis; the remaining fraction
(1 c) is oriented circumferentially. Then, to an adequate
approximation for illustrative purposes, the moduli of the
material are:

Optimisation of anisotropy
Ejj 1
f  Em c  f  Ef and
39
The last section identified two expressions for the optimum E? 1
f  Em 1
c  f  Ef
shape, /, for orthotropic tubes, expressed in terms of
material properties. By substituting these back in the fail- Defining E
as the value of E for which c = 1/2 (when the
ure equations, the failure moment of a tube of optimum material is quasi-isotropic)
shape can be written as a function of material properties
alone. If splitting is possible, (that is, Eq. 36 is not satis- 1
E
1
f  Em  f  Ef 40
fied), the ideal shape is that corresponding to /A. Substi- 2
tuting this into Eq. 30 gives the optimum moment that can
be achieved with a given set of material properties: we find
   
Mopt  1=2 1 Ef
0:18  rt? Ejj Ejj E
1 c
f and
A 3=2 2 E

!     41
rt? Ejj 37 1 Ef
with /opt 0:81  E? E
1
c f
r2cjj 2 E

with similar expressions for the strengths in terms of the
If, on the other hand, splitting is suppressed (that is, Eq. 36
strength r at u = 1/2:
is satisfied), the ideal shape is that corresponding to /B.
Substituting this into Eq. 30 gives
   
1 rf
  rjj r

1 c
f and
Mopt 1=2 1=2 2 r


0:11  rcjj E jj E ?     42
A3=2 1 rf
 1=2 38 r? r
1
c f
Ejj E? 2 r


with /opt 0:31 
rcjj
We take, as an example, the case in which the optimum
Many orthotropic materials are fibre-reinforced shape is limited by /B, when Eq. 38 applies. Substituting
composites, and for these a degree of control of the for Ejj ; E? , and rcjj gives:

123
9012 J Mater Sci (2007) 42:90059014

Mopt 1=2

 E

0:11r
A3=2 (   1=2
1
 1 c
b
2

      1=4 )
1 1
 1 c
a 1
c a
2 2
43
E r
with a f E
f and b f r
f .
The limits for a are determined for the extreme cases
that Em  Ef , when
Ef f  Ef f  Ef
af
1
1 2 44
E 1
f  Em 2  f  Ef 2  f  Ef
Fig. 5 The graph shows the expression in curly brackets ({}) of
and that Em Ef =2 , when Eq. 43 plotted against c for a number of different values for a and b.
Note that a and b will vary independently but in parallel. In the
Ef f  Ef extreme case that the material is truly homogeneous (a = b = 0), the
af

 2f 45 failure moment is, correctly, independent of the tubes shape (black
E 1=2  1
f Ef f  Ef line)

With typical values for the fibre-volume fraction

0  f  0:6 and an analogous analysis for b, we find that
0 a < 2 and 0  b\2 .
The term in the curly brackets of Eq. 43 describes the
way in which the efficiency of material usage, measured
by M/A3/2, varies with the degree of anisotropy. When
c = 1/2, the material is isotropic and the bracketed term
takes the value 1. When c = 1 all the fibres are aligned
parallel to the tube axis and anisotropy is maximised.
Figure 5 shows this expression plotted against
c 0:5\c\1 for different values for a and b. Note that a
and b will vary independently but in parallel. In the ex-
treme case that the material contains no fibres and is truly
homogeneous (a = b = 0), the failure moment is, cor-
rectly, independent of the parameter c. There is a broad
maximum between c = 0.75 and c = 1.0, depending on
the values of a and b, meaning that there is an ideal factors, in the range / = 26. The shape factors and
anisotropy that maximises the efficiency. To illustrate this
with an example, we seek to optimise the fibre distribution
of a composite tube with a 48% volume fraction of fibres
with a Youngs modulus ten times higher than that of the
matrix. Assuming that also the fibre strength is ten times
higher than that of the matrix, we calculate that
a b 1:648 . With these values we find that the
expression in the curly brackets in Eq. 43 reaches a
maximum at c = 0.803, thus when about 80% of the
fibres are aligned with the longitudinal axis of the tube;
giving an efficiency that is 14% larger than that of an
isotropic tube.
With a more detailed model of how stiffness and strength
depend on structure (here described by a, b and c)
a precise optimisation becomes possible, this is, however,
beyond the scope of this study.
An application: bamboo culm
In Fig. 6a and b, failure curves are plotted for a typical
range of green and dry bamboo culms. The three sets of
lines on each chart describe culms with high, medium
and low properties as listed in Table 1. In Fig. 6c the
failure modes of a green bamboo culm (green lines) and a
dry bamboo culm (red lines) both with medium properties
are compared. We may now evaluate the stiffness
requirement required by M4 in Eq. A4, using typical shape
stiffness requirement circumscribe the region of the chart
on which bamboo culms lie; it is plotted as a box. In all
cases the failure moment increases with increasing / up to
the boundary between the fracture/yield region and the
splitting regime; beyond this it slowly falls. The optimum
shape is that at the peak.
From an engineering point of view, the most efficient
shape for safe design is that which falls into the region
just below the highest part of the fracture-domain (the
peak of the fracture envelope). This ensures that cata-
strophic failure by local buckling and splitting is
prevented while the material is used most efficiently.
This is what we find for both the green and the dry
bamboo.

123
J Mater Sci (2007) 42:90059014 9013

Acknowledgements The ideas presented here have been helped by

discussions with numerous colleagues and associates. We particularly
wish to acknowledge the inputs and suggestions of Prof. C.R. Call-
adine, Dr. H. R. Shercliff, Dr. P. M. Weaver and of an anonymous
reviewer. We also wish to acknowledge the support of the Royal
Society of London and the US Advance Research Project Agency
through the University Research Initiative under Office of Naval
Research Contract No. N-00014092-J-1808.

Appendix 1 full solution for /A

The transition from fracture/yield to splitting is found by

equating the respective moments of failure at A:
 2 1=2
A A
/A 
B A1
2 4
with
2   2 3
Ejj rt?  2  2
61 2:037  E? rcjj 7 1 E?
6
A4  2   7 and B
Ejj rt?
5 1:236 r t?
1:259  rEt?? r2 cjj

Appendix 2 failure through inadequate stiffness

The final criterion for the macroscopical mechanical per-

formance is that of stiffness. Mosbrugger [42] classifies
plants according to their structural behaviour: either the
plant is a flexibility strategist and reduces external loads
by bending or it is a stability strategist and has a
structure which is stiff and strong enough to withstand the
loads without much bending. As the principal load is that
due to wind and the velocity of wind increases with height
above ground, a flexible tree which bends in a strong wind
reduces the moment arm of the net wind force, especially
if elastic deformation of its crown reduces its down-wind
profile. Tree trunks are frequently stability strategists,
whereas their branches must be capable of bending to a
quarter-circle. The curvature of the bamboo culm, C, can
then be expressed as a function of the length, l, of the
stem

1 p
Fig. 6 (a) and (b) Failure maps plotting the loading coefficient C  A2
M/A3/2 against shape factor / for typical green (a) and dry (b) R 21
bamboo culms. The green, blue and red sets of lines indicate low, Substituting this expression for C in Eq. 8 and inserting the
medium and high property values as listed in Table 1. The boxes
circumscribe the region in which bamboo typically lies according result for c in Eq. 13b gives the bending moment, M, which
to its shape and the stiffness criterion. The numbers (1), (2) and (3) bends the stem into a quarter-circle
label the three failure domains fracture/yield, splitting and
p 
 
3p2 r 2 2 Ejj
buckling, respectively. (c) A failure map comparing green (green
lines) and dry (red lines) bamboo culms of medium material p r 3=2 1=2
M p A / Ejj 1
/ A3
properties 4 2 l 8 l E?

123
9014
The second expression in the square bracket is very small
due to the high slenderness ratios of bamboo (/ = l/r =
5501000, [43]) and may therefore be neglected. The
moment which bends a stem to a quarter-circle may there-
fore be rewritten as
p  23. NASA (1968) Space vehicles design criteria (structures)buck-
p r 3=2 1=2
M4 p A / Ejj A4
4 2 l
If the plant is to function as a flexibility strategist, it must
be able to do this without failing by any of the other three
mechanisms analysed above.
References
1. Amada S, Munekata T, Nagase Y, Ichikawa Y, Kirigai A, Yang
ZF (1996) J Comp Mat 30:800
2. Arce-Villalobos OA (1993) Fundamentals of the design of
bamboo structures. PhD Thesis, Eindhoven University of Tech-
nology, Netherlands
3. Niklas KJ (1989) Am J Bot 76:521
4. Niklas KJ (1992) Plant biomechanics: an engineering approach to
plant form and function. University of Chicago Press, Chicago
5. Niklas KJ (1995) Ann Bot 75:133
6. Niklas KJ (1997a) Ann Bot 80:275
7. Niklas KJ (1997b) Ann Bot 80:437
8. Niklas KJ (1998) Ann Bot 81:11
9. Spatz HC, Speck T, Vogellehner D (1990) Bot Acta 103:123
10. Spatz HC, Boomgaarden C, Speck T (1993) Bot Acta 106:254
11. Spatz HC, Speck T (1994) Biomimetics 2:149
12. Spatz HC, Kohler L, Speck T (1998) Am J Bot 85:305
13. Speck T, Speck O, Emanns A, Spatz HC (1998) Bot Acta
111:366
14. Schulgasser K, Witztum A (1997) Ann Bot 80:35
15. Cecchini LS, Weaver PM (2002) AIAA J 40:2136
16. Corona E, Rodrigues A (1995) Comp Eng 5:163
17. Fabian O (1977) Int J Sol Struct 13:1257
18. Gerard G (1968) Minimum weight design of compressive struc-
tures. New York University Press/Interscience, New York
123
J Mater Sci (2007) 42:90059014
19. Harursampath D, Hodges DH (1999) Int J Non-Linear Mech
34:1003
20. Kedward KT (1978) In: Proceedings of the 2nd international
conference on composite materials (ICCM/2). Met Soc of AIME
p 353
21. Libai A, Bert CW (1994a) Int J Sol Struct 31:1003
22. Libai A, Bert CW (1994b) Int J Sol Struct 31:1019
ling of thin-walled circular cylinders. Technical Report NASA
SP-8007, NASA
24. Tatting BF, Gurdal Z, Vasiliev VV (1996) AIAA J 34:1934
25. Tatting BF, Gurdal Z, Vasiliev VV (1997) Int J Solids Struct
34:1419
26. Tsai SW, Wu EM (1971) J Comp Mater 5:58
27. Schulgasser K, Witztum A (1992) J Theor Biol 155:497
28. Brazier LG (1927) Proc Roy Soc Lond A 116:104
29. Calladine CR (1983) Theory of shell structures. Cambridge
University Press, Cambridge
30. Ghavami K (1990) In: Ramanuja Rao I, Gnanaharan R, Sastry C
(eds) Proceedings of the international bamboo workshop held in
Cochin, India, 1418 November. The Kerala Forest Research
Institute, India and International Development Research Centre,
Canada, p 235
31. Li S, Fu S, Zeng Q, Zhao X, Zhou B (1994) Biomimetics 2:15
32. Oda J (1980) Trans Japan Soc Mech Eng Ser A 46:997
33. Ueda K (1980) Res Bull Coll Exp For Hokkaido University
37:817
34. Akselrad EL (1965) Izv Akad Nauk USSR, Otdelenie Tek-
nicheskikh Nauk Mech No 4:123
35. Seide P, Weingarten V (1961) J Appl Mech 28:112
36. Suo Z (1990a) J Appl Mech 57:627
37. Suo Z (1990b) Proc R Soc Lond A 427:331
38. Wegst UGK, Ashby MF (2004) Phil Mag 84:2167
39. Karam GN, Gibson JL (1995a) Int J Sol Struct 32:1259
40. Karam GN, Gibson JL (1995b) Int J Sol Struct 32:1285
41. Young WC (1989) Roarks formulas for stress and strain, 6th edn.
McGraw-Hill, London
42. Mosbrugger V (1990) The tree habit in land plants. Lecture notes
in Earth Sciences: 28. Springer, Berlin
43. Liese W (1985) Bamboos: biology, silvics, properties, utilization
(Schriftenreihe der GTZ, Nr. 180, Deutsche Gesellschaft fur
Technische Zusammenarbeit (GTZ) GmbH, Dag-Hammerskjold-
Weg 1+2, D-65760 Eschborn, Germany)
J Mater Sci (2007) 42:90159020
DOI 10.1007/s10853-007-1834-0
A novel method to prepare magnetic nanoparticles: precipitation
in bicontinuous microemulsions
Jesus Esquivel Isabel A. Facundo M. Esther Trevino
Raul G. Lopez
Received: 18 September 2006 / Accepted: 8 May 2007 / Published online: 20 July 2007
Springer Science+Business Media, LLC 2007
Abstract A novel method to prepare iron oxide nano-
particles by precipitation in bicontinuous microemulsions
is reported. Precipitation reactions were carried out in
microemulsions stabilized with a mixture of dodecyltrim-
ethylammonium bromide/didodecyldimethylammonium
bromide (3/2, w/w). Nanoparticles were characterized by
X-ray diffraction, transmission electronic microscopy
(TEM) and vibrating sample magnetometry (VSM), and
consisted of magnetite or a mixture of magnetite and
maghemite. The particles have an average diameter of
8 nm with a relatively narrow particle size distribution and
show possible superparamagnetic behavior. Noticeably
higher yields of precipitate are observed for this new
approach in comparison with those typicall obtained using
reverse microemulsions.
Introduction
Nanoscale superparamagnetic iron oxide particles, such
as magnetite (Fe3O4) and maghemite (c-Fe2O3), offer
very attractive possibilities in biomedical applications
because of their magnetic properties and size specifically
in chemotherapy, radiotherapy and hyperthermia for
J. Esquivel
I. A. Facundo
M. E. Trevino

R. G. Lopez (&)
Centro de Investigacion en Qumica Aplicada, Boul. Ing.
Enrique Reyna No. 140, Saltillo, Coahuila 25253, Mexico
e-mail: glopez@ciqa.mx
J. Esquivel
Facultad de Ciencias Qumicas, Universidad Autonoma de
Coahuila, Saltillo, Coahuila 25240, Mexico
treatment of solid tumors [14]. For these applications
the magnetic particles should be coated with such
materials as natural or synthetic polymers to allow for
the attachment of drugs and other molecules capable of
specific recognition of, and binding to, malignant cells.
Although many synthesis methods are available for
preparing superparamagnetic iron oxide particles [3],
many papers have demonstrated that precipitation in re-
verse microemulsions is the most adequate method for
preparing magnetic particles with diameters smaller than
10 nm and narrow particle size distribution in a direct
way [514]. In the other known methods the obtained
particles must be subjected to a size-selection process in
order to narrow the particle size distribution [15, 16].
Within the reports on precipitation in reverse micro-
emulsions are noticeable the works of: Gobe et al. [5],
Bandow et al. [6], Dresco et al. [10], Santra et al. [12]
and, more recently, Lee et al. [14], who prepared mag-
netic particles with mean diameters of 3, 3.6, 7.3, 1 and
3 nm, respectively, and with relatively narrow particle
size distribution. In spite of all this work, which deals
mainly with the synthesis and characterization of the
particles, little attention has been paid to the mechanistic
of such systems [810, 12]. However, it is now recog-
nized that the nucleation and growth of the nanoparticles
in these systems occur inside of the swollen micelles (or
droplets) of the reverse microemulsions, which usually
have diameters smaller than 10 nm. In this way, swollen
micelles function as nanoreactors for the preparation of
the particles [17].
Despite the size control and distribution of the magnetic
particles, the precipitation method using the reverse
microemulsions has the drawback of a relatively low
yield. Calculations based on the reported recipes for pre-
paring magnetic nanoparticles using this technique, show
123
9016
yields from 0.1 g [7] to 0.4 g [5] of product per 100 g of
total mixture. Surprisingly, and as far as we know, the
precipitation in bicontinuous microemulsions has not been
considered as an option that could increase this yield. In
contrast to reverse microemulsions, which consist of
aqueous droplets dispersed in an oleic continuous phase,
bicontinuous microemulsions are formed from intercon-
nected aqueous channels with diameters usually smaller
than 10 nm immersed in an oleic continuous phase [18].
This structure allows bicontinuous microemulsions to
accept contents of the aqueous phase usually between 50
and 100% higher than those accepted by reverse micro-
emulsions [18, 19]. We speculate that the reason why
bicontinuous microemulsions have not been considered
earlier as media for the precipitation of magnetic particles
is because of their microstructure. On the first hand one
would not expect to obtain spherical or equiaxial particles
from the cylindrical channels of a bicontinuous micro-
emulsion in contrast to what can be obtained or expected
from the spherical droplets of reverse microemulsions.
Nowadays, it is accepted that the mechanism through
which particles are obtained in reverse microemulsions
includes nucleation in those droplets that contain a mini-
mum number of precipitate molecules and further growth
through recruiting precursor ions from those droplets
containing no particles inside either by diffusion or droplet
collision [17]. Although particle growth in bicontinuous
microemulsions cannot proceed by droplet collision, there
is no obvious reason why nucleation inside the channels
and further growth of nuclei through recruiting precursors
ions from surroundings could not occur.
In this paper, we report a new method that uses the
channels of bicontinuous microemulsions as media to
prepare nanoscale magnetic particles. The results of the
characterization of these particles by X-ray diffraction,
transmission electron microscopy and vibrating sample
magnetometry were compared with those of the particles
prepared in a reverse microemulsion.
Experimental
Materials
All reagents were high-purity grades from Aldrich used
without further purification: ferric chloride (FeCl3
4H2O,
99%), ferrous chloride (FeCl2
6H2O, 98%), methyl meth-
acrylate (MMA, 99%), aqueous ammonia (NH4OH,
57.6 wt.%), dodecyltrimethylammonium bromide (DTAB,
99%) and didodecyldimethylammonium bromide (DDAB,
98%). Water was triple-distilled deionized grade.
123
J Mater Sci (2007) 42:90159020
Phase diagram determinations
Microemulsion regions at 80 C were determined by titra-
tion of solutions of the DTAB/DDAB mixture (3/2, w/w) in
MMA with 0.75 M aqueous solution of a FeCl3
6H2O/
FeCl2
4H2O mixture (3/2, mol/mol). Phase boundary was
detected visually at each one of the constant (DTAB/
DDAB)/MMA lines studied. Then, samples with composi-
tions slightly below and above that of the visually deter-
mined phase boundary were prepared by weighting each
component and allowing to reach equilibrium in a water
bath at 80 C to determine more precisely the phase
boundary.
Synthesis of magnetic particles
Precipitation reactions were carried out at 80 C in a
100 mL jacketed glass reactor equipped with a reflux
condenser and inlets for argon, mechanical agitation and
feed of the microemulsion and ammonium hydroxide
aqueous solution. To help stabilize the system during the
reaction, the reactor was immersed in an ultrasonic bath.
Two precipitation reactions in bicontinuous microemul-
sions were carried out along with one precipitation reac-
tion in a reverse microemulsion with the same surfactant
content (33 wt.%) for comparison. When the reaction
media was the reverse microemulsion a stirrer speed of
300 rpm was used; mechanical agitation was not provided
when the reactions were carried out in the bicontinuous
microemulsions. Reverse microemulsion was prepared
with 57 wt.% MMA and 10 wt.% aqueous solution of a
mixture of ferric and ferrous chlorides in a molar ratio of
3/2 with an overall concentration of 0.75 M. The bicon-
tinuous microemulsions were composed by 30 and
40 wt.% of the same aqueous solution of the mixture of
ferric and ferrous chlorides and 37 and 27 wt.% MMA,
respectively. In all cases, an excess of ammonium
hydroxide (twice the stoichiometric requirements) was
used. The typical procedure for the precipitation reaction
started with charging the previously deoxygenated mi-
croemulsion in the reactor and then raising the tempera-
ture to 80 C. To initiate the reaction aqueous solution of
ammonium hydroxide was added to the reactor. The
reacting system was continuously purged with argon
during the entire process. The precipitation reaction was
allowed to proceed for 30 min whereupon a mixture of
acetone and water was added to cause the final precipitate
recovered by magnetic separation. Noticeably the super-
natant contained no visible particles in suspension. After
several washes with acetone, the precipitate was dried
yielding a black powder.
J Mater Sci (2007) 42:90159020
Characterization
Electrical conductivities were measured at 80 C and
1 KHz with an Orion 115 conductivity meter. Magnetic
measurements were determined in fields up to 12.5 kOe
using Lake Shore vibrating sample magnetometer model
VSM 735. X-ray determinations were done on a Siemens
D-5000 X-ray diffractometer. Particle size was determined
from micrographs obtained using a JEOL 1200 EXII
transmission electron microscope. TEM samples were
prepared by dispersing the powders in ethanol using an
ultrasonicator, and depositing the dispersion on a copper
grid.
Results and discussion
The partial phase diagram at 80 C of the system composed
by DTAB/DDAB (3/2, w/w) mixture, MMA and an
aqueous solution of FeCl3
6H2O and FeCl2
4H2O in 3/2
molar ratio with an overall concentration of 0.75 M is
shown in Fig. 1. We note the one-phase microemulsion
region initiating at the MMA-(DTAB/DDAB) side which
extends to the central zone of the phase diagram. The form
of this region suggests a transition from reverse to bicon-
tinuous microemulsions when the system goes from the
MMA-(DTAB/DDAB) side to the central zone of the
diagram.
In order to determine where the transition occurs, the
electrical conductivities of the microemulsions samples
with the same surfactant concentration (33 wt.%), but
Fig. 1 Partial phase diagram at 80 C of system DTAB/DDAB,
MMA and a 0.75 M aqueous solution of the mixture of iron chlorides
showing microemulsion regions (1/). Circles on the horizontal line
crossing the diagram (33 wt.% surfactant concentration) indicate the
composition of microemulsions in which precipitation reactions were
carried out
9017
Fig. 2 Electrical conductivity at 80 C of microemulsions containing
33 wt.% surfactant and different concentrations of 0.75 M aqueous
solution of the mixture of iron chlorides
different contents of aqueous solutions of the mixture of
iron chlorides, were measured at 80 C. The results are
shown in Fig. 2 where it is possible to identify an interval
of low conductivity (up to 15 wt.% aqueous solution). This
agrees with the typical behavior of reverse microemulsions
where low conductivities are observed due to the discon-
tinuous nature of reverse microemulsions. On the other
hand, there is a second interval of higher conductivity from
the 20 wt.% aqueous solution in which electrical conduc-
tivity increases linearly. The higher conductivity agrees
with the increased capacity of the bicontinuous micro-
emulsions for conducting electrical current as a result of
their continuous long aqueous phase. Accordingly, the
transition from reverse to bicontinuous microemulsions
prepared with 33 wt.% surfactant should occur at some
point between 15 and 20 wt.% aqueous solution. Circles on
the horizontal line crossing the diagram (33 wt.% surfac-
tant concentration) in Fig. 1 indicate the composition of
microemulsions in which precipitation reactions were car-
ried out.
All of the precipitation reactions turned black upon the
addition of aqueous ammonium hydroxide. This is typical
of precipitation reactions where magnetite has been
obtained [10]. Once the products of the reactions were
recovered, yield calculations indicated 0.51 and 1.16 g of
magnetic particles per 100 g of total mixture for bicon-
tinuous microemulsions with 30 and 40 wt.% aqueous
solution of chlorides, respectively, compared to 0.29 g of
magnetic particles per 100 g of total mixture for the reverse
microemulsion. Under the conditions studied, these data
demonstrate the yield advantage of bicontinuous over
reverse microemulsions in the synthesis of nanoscale
123
9018 J Mater Sci (2007) 42:90159020

Fig. 3 TEM micrographs of

particles prepared in: reverse
microemulsion (a) and
bicontinuous microemulsion
with 30 (b) and 40 (c) wt.%
aqueous solution of the mixture
of iron chlorides. Insets show
the corresponding high-
resolution images. Histograms
of the size distributions are also
included

particles. It should be noted that we found it difficult to (8 nm) with relatively narrow particle size distributions.
compare our results with those reported in other studies on The results demonstrate that nucleation and growth of
precipitation of magnetic particles in reverse microemul- nanoparticles can be carried out in the channels of bicon-
sions. The reason is that the authors do not report the yields tinuous microemulsions. This observation opens the pos-
obtained in their studies but only a few of them provide the sibility that the channel diameters may play a role in
sufficient data for calculating the theoretical yields. Using determining the final particle size.
the data reported was possible to estimate the yields in X-ray diffraction patterns of the samples are shown in
those studies carried out by Gobe et al. [5], Lee et al. [7] Fig. 4. Peak broadening observed is consistent with the
and Selim [20], as 0.4, 0.1 and 0.2 g of magnetic particles small particle size [12, 21] with patterns matching those of
per 100 g of total mixture, respectively. In comparison, the
yields obtained in our study resulted in similar values for
the reverse microemulsion; however, those obtained from
precipitation in bicontinuous microemulsion were higher.
Low and high magnification TEM images are shown in
Fig. 3 along with histograms. Table 1 shows the number-
average diameter (Dn) and the polydispersity index
(Dw/Dn) calculated from the TEM photomicrographs.
Data in Table 1 indicate that the precipitation reaction
carried out in the reverse microemulsion produces small
diameter particles (5.6 nm) with a relatively narrow parti-
cle size distribution. For particles prepared in the bicon-
tinuous microemulsions, small diameters are also observed

Table 1 Average diameter and polydispersity index of particles

prepared by precipitation in reverse and bicontinuous microemulsions
as determined from TEM micrographs
Precipitation reaction media Dn (nm) Dw/Dn

Reverse microemulsion 5.6 1.20

Bicontinuous microemulsion
(30 wt.% aq. solution)
Bicontinuous microemulsion
(40 wt.% aq. solution)
8.1 1.22
Fig. 4 X-ray diffraction patterns of particles prepared in: reverse
7.9 1.25 microemulsion (a) and bicontinuous microemulsion with 30 (b) and
40 (c) wt.% aqueous solution of the mixture of iron chlorides.
Standard patterns of magnetite (d) and maghemite (e) are included

123
J Mater Sci (2007) 42:90159020
Fig. 5 Magnetization curves for particles prepared in reverse
microemulsion (h) and bicontinuous microemulsion with 30 (s)
and 40 (n) wt.% aqueous solution of the mixture of iron chlorides
magnetite and maguemite. Although, differentiation
between the species is not possible from the present X-ray
data, Fe (II) is present as indicated by the black color of the
samples. Thus, the particles produced by precipitation
reactions carried out in the bicontinuous microemulsions
may consist of magnetite or a mixture of the two species.
Differentiation is left for future studies using bicontinuous
microemulsions.
Figure 5 shows the room temperature magnetization
curves of the particles. None of the samples are saturated
up to 12.5 kOe. This behavior, typical of magnetite parti-
cles with diameters smaller than 15 nm, has been consis-
tently reported in the literature [8, 1214, 22] and arises
from the difficulty for aligning the magnetic moments of
the surface atoms, which constitute a large fraction of the
total volume of a particle, in the direction of applied
magnetic field [23]. Superparamagnetic behavior of all the
samples is suggested by the low remnant magnetization
(1.7 emu/g) and coercitive field (45 Oe). This behavior was
expected because of the very small sizes of the particles in
our samples. The explanation is that superparamagnetism
implies that the magnetic particles have essentially single
domains, i.e., all magnetic moments in the particle are
aligned in one direction [10], and as is well known, mag-
netic particles below 10 nm become single domains [24].
The magnetization values of the samples at 12.5 kOe
obtained by precipitation in the bicontinuous microemul-
sions (6365 emu/g) are higher than that prepared in the
reverse microemulsion (45 emu/g). This agrees with the
known fact that magnetization of small particles (diameters
smaller than ca. 15 nm) decreases as particle size decreases
[8, 12]. In general, the magnetization values obtained in
9019
our study are higher than those reported for magnetic
particles prepared by precipitation in reverse microemul-
sions at relatively low temperatures (2550 C) [20, 25]
but they are close to those values obtained from the par-
ticles prepared at higher temperatures [14, 25]. Our
explanation for this behavior is that an increase in precip-
itation temperature enhances the crystallinity of the iron
oxide nanoparticles [14] and, as is well known, magneti-
zation of magnetic nanoparticles depends in a direct way
on their crystallinity.
Conclusions
Magnetic nanoparticles were obtained by precipitation
reactions in bicontinuous microemulsions. X-ray diffrac-
tion analysis demonstrated that the particles consist of
magnetite or a mixture of magnetitemaghemite. TEM
analyses showed that the particles have average diame-
ters close to 8 nm and a relatively narrow particle size
distribution. VSM data suggest superparamagnetic
behavior. The bicontinuous microemulsion synthesis for
the preparation of the present nanoparticles shows yields
appreciably higher than those obtained using reverse
microemulsions technique. We propose the bicontinuous
microemulsion technique as a general method for the
preparation of nanoscale particles.
Acknowledgment CONACyT supported this research through
grant SEP 2003-CO2-45436. Two of us (IAF and JE) acknowledge
scholarships from CONACyT to pursue her Doctoral and MSc works,
respectively. We are grateful to Selene Guzman for her support in
TEM measurements and Janet Valdes and Patricia Siller for their
assistance in acquiring various specialized documents.
References
1. Shinkai M (2002) J Biosci Bioeng 94:606
2. Pankhurst QA, Connolly J, Jones SK, Dobson J (2003) J Phys D:
Appl Phys 36:R167
3. Tartaj P, Morales MP, Veintemillas-Verdaguer S, Gonzalez-
Carreno T, Serna CJ (2003) J Phys D: Appl Phys 36:R182
4. Berry CC, Curtis ASG (2003) J Phys D: Appl Phys 36:R198
5. Gobe M, Kon-no K, Kandori K, Kitahara A (1983) J Colloid
Interf Sci 93:293
6. Bandow S, Kimura K, Kon-no K, Kitahara A (1987) Jpn J Appl
Phys 26:713
7. Lee KM, Sorensen CM, Klabunde KJ, Hadjipanayis GC (1992)
IEEE Trans Magn 28:3180
8. Liz L, Lopez-Quintela MA, Mira J, Rivas J (1994) J Mater Sci
29:3797
9. Pillai VK (1995) PhD Thesis, University of Florida
10. Dresco PA, Zaitsev VS, Gambino RJ, Chu B (1999) Langmuir
15:1945
11. Lee HS, Lee WC, Furubayashi T (1999) J Appl Phys 85:5231
12. Santra S, Tapec R, Theodoropoulou N, Dobson J, Hebard A, Tan
W (2001) Langmuir 17:1900
123
9020
13. Liu ZL, Wang X, Yao KL, Du GH, Lu QH, Ding ZH, Tao J, Ning
Q, Luo XP, Tian DY, Xi D (2004) J Mater Sci 39:2633
14. Lee Y, Lee J, Bae CJ, Park JG, Noh HJ, Park JH, Hyeon T (2005)
Adv Funct Mater 15:503
15. Lopez-Lopez MT, Duran JDG, Delgado AV, Gonzalez-Caballero
F (2005) J Colloid Interf Sci 291:144
16. Mikhaylova M, Kim DK, Bobrysheva N, Osmolowsky M,
Semenov V, Tsakalakos T, Muhammed M (2004) Langmuir
20:2472
17. Osseo-Asare K (1999) Handbook of microemulsion science and
technology, chap. 18. Marcel Dekker, Inc., New York
18. Ezrahi S, Aserin A, Garti N (1999) Handbook of microemulsion
science and technology, chap. 7. Marcel Dekker, Inc., New York
123
J Mater Sci (2007) 42:90159020
19. Salager JL, Anton RE (1999) Handbook of microemulsion sci-
ence and technology, chap. 8. Marcel Dekker, Inc., New York
20. Selim S (1997) U.S. Patent 5695901
21. Lopez Perez JA, Lopez Quintela MA, Mira J, Rivas J, Charles
SW (1997) J Phys Chem B 101:8045
22. Lopez-Perez JA, Lopez-Quintela MA, Mira J, Rivas J (1997)
IEEE Trans Magn 33:4359
23. Kodama RH (1999) J Magn Magn Mater 200:359
24. Cornell RM, Schwertmann U (1996) The iron oxides. structure,
properties, reactions, occurrence and uses. VCH, Weinheim,
Germany
25. Lee Y, Lee J, Bae CJ, Park JG, Noh HJ, Park JH, Hyeon T (2005)
Adv Funct Mater 15:2036
J Mater Sci (2007) 42:90219029
DOI 10.1007/s10853-007-1881-6
The relationship between atomic partitioning and corrosion
resistance in the weld-heat affected zone microstructures of UNS
S32304 duplex stainless steel
Carlos Mario Garzon Carlos A. Serna
Sergio D. Brandi Antonio J. Ramirez
Received: 7 March 2006 / Accepted: 25 May 2007 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007
Abstract Wrought material as well as physically simu-
lated welding heat affected zone (HAZ) samples of an UNS
S32304 duplex stainless steel were subjected to electro-
chemical corrosion tests and electron microscopy charac-
terization. An impaired corrosion resistance of the HAZ
microstructures compared to the wrought material micro-
structure was observed. Calphad-based numerical simula-
tion of phase transformations and solute redistribution
taking place during welding provided an explanation of the
observed corrosion behavior. The poor corrosion resistance
of the HAZ microstructures studied was mainly attributed
to a decrease in corrosion resistance of ferritic grains after
welding, which exhibited lower chromium content than
ferritic grains in the wrought material.
Introduction
The welding of duplex stainless steels (DSSs) strongly
affects their fine-grained and duplex-balanced ferrite plus
C. M. Garzon
A. J. Ramirez (&)
Brazilian Synchrotron Light Laboratory, Caixa Postal
6192 - CEP 13083-970 Campinas, SP, Brazil
e-mail: ramirez@lnls.br
Present Address:
C. M. Garzon
Physics Department, National University of Colombia,
Avenida Cra. 30 No 45-03, Ciudad Universitaria, Edificio 404,
Bogota, Colombia
C. A. Serna
S. D. Brandi
Metallurgical and Materials Engineering Department,
University of Sao Paulo, Av. Prof. Mello Moraes 2463,
CEP 05508-900 Sao Paulo, SP, Brazil
austenite microstructure [16]. In particular, in the high
temperature heat affected zone (HAZ), a coarse micro-
structure is formed by a high fraction of ferrite, allotri-
morphic and Widmanstatten austenite, and extensive
chromium nitride precipitation within ferritic grains [16].
When this as-welded microstructure is reheated during
multipass welding or during post-weld heat treatment, new
austenite, called secondary austenite (c2), is formed. This
c2 may grow from preexisting austenite particles (so called
primary austenite, c1) and is called intergranular c2.
Another form of c2 can be formed by the nucleation and
growth of new intragranular c2 particles, mostly within the
nitride colonies. Nitride distribution will also be modified,
with some intragranular nitrides dissolving and/or new
nitride precipitation occurring mainly along ferrite/
austenite interfaces [27].
It has been frequently reported that the DSS HAZ
microstructure exhibits impaired corrosion resistance when
compared to the wrought material [14]. On the other hand,
there is no consensus on the origin of this reduction in
corrosion resistance or on the role of c2 precipitation.
The detrimental effect of intragranular c2 precipitation
on the localized corrosion resistance of DSSs welded joints
has been reported, and this reduced corrosion resistance has
been associated with the lower N, Cr, and Mo content of c2
when compared to c1 [2, 3, 8, 9]. It has also been suggested
that the poor corrosion resistance of the HAZ microstruc-
tures is mainly associated with chromium depletion around
chromium nitrides within ferritic grains [1].
The objective of this investigation was to study the role of
alloy element distribution among ferrite, c1, and intergran-
ular c2 on corrosion resistance of UNS S32304 DSS HAZ
microstructures. Thus, both electron microscopy character-
ization and electrochemical corrosion tests were performed
on simulated HAZ microstructures. The as-welded HAZ and
123
9022
reheated-HAZ microstructures were obtained in a Gleeble
thermo-mechanical simulator. In addition, the analysis of the
experimental results was supported by thermodynamic and
kinetic simulation of the phase transformations taking place
during the thermal cycles.
Experimental
A commercially available UNS S32304 duplex stainless
steel was subjected to physical simulation to obtain
samples with both as-welded HAZ microstructures
(ferritization treatments) and reheated (multipass) HAZ
microstructures (reheating treatments). The chemical
composition of the DSS is shown in Table 1. The ferriti-
zation treatments consisted of an isothermal anneal at
1,350 C for 5 s followed by rapid cooling; the average
heating rate was 300 C
s1 and the cooling rate at 800 C
was 46 C
s1. The reheating treatments were applied to
prior ferritized samples, and consisted of isothermal heat
treatments at 900 or 1,200 C for 10 s. The heating and
cooling conditions for the reheating and ferritization
treatments were similar. The physical simulations were
performed in a thermo-mechanical simulator Gleeble
1500, using prismatic samples of 6 6 90 mm. Detailed
information about these treatments was presented
elsewhere [6, 7].
Microstructural characterization was performed using
optical microscopy (OM), scanning electron microscopy
(SEM) and transmission electron microscopy (TEM).
Detailed microstructural characterization of the simulated
samples is presented elsewhere [6, 7]. Metallographic
preparation of samples for OM and SEM consisted of
abrasive paper grinding followed by diamond paste pol-
ishing and a final stage of 0.05 lm colloidal silica pol-
ishing. The specimens were electrolytically etched using a
40%-vol HNO3 aqueous solution at 25 C, and applying
1.1 V during ~70 s. The SEM analyses were performed
using a JEOL JSM 5900 LV microscope as well as a
CAMBRIDGE Steroscan-440 microscope coupled with a
wavelength-dispersive X-ray spectrometer (XWDS). TEM
samples were prepared by jet polishing with a 70%-vol
ethyl alcohol, 20%-vol glycerin and 10%-vol perchloric
acid (HClO4) solution at 20 C and 30 V. The TEM
analyses were performed using a JEOL JEM 3010 URP
microscope coupled with an energy dispersive X-ray
spectrometer (XEDS) system.
Table 1 Chemical composition of the UNS S32304 duplex stainless steel (wt-%)
Fe Cr Ni Mn
Bal. 22.64 4.81 1.53
123
J Mater Sci (2007) 42:90219029
The pitting corrosion resistance was evaluated by cyclic
potentiodynamic polarization experiments carried out at
25 1 C in a 3.5 wt-% NaCl naturally aerated solution
using a potential scan rate of 1 mV
s1.
The degree of sensitization (DOS) was evaluated
through double loop electrochemical potentiokinetic
reactivation (DL-EPR) tests. The DOS was measured as
the Ir/Ia ratio, where Ir is the maximum value of the
anodic current in the reactivation peak and Ia is the
maximum value of the anodic current in the activation
peak [10]. The DL-EPR tests were performed in a 0.5 M
H2SO4 + 0.01 M KSCN naturally aerated solution at
30 1 C, using a potential scan rate of 1 mV
s1. The
scanned potential range was from the corrosion potential
up to 500 mV-SCE.
The cyclic polarization and DL-EPR tests were
performed using a Princeton Applied Research Potentio-
statGalvanostat 273A. Saturated calomel and platinum
electrodes were used as reference and counter electrodes,
respectively. Prior to the corrosion tests, the samples were
mechanically ground using abrasive paper to 600 mesh
final finish. Five tests were performed for each studied
condition.
Numerical simulation
The thermodynamics and kinetics assessment of both
austenite growth and chromium nitride dissolution and/or
precipitation in the HAZ microstructures was performed
using the Calphad-based Thermocalc and Dictra soft-
ware [11] along with the Fe-data and Mob2 databases. A
detailed analysis of this numerical modeling is presented
elsewhere [12] and only the most important features are
presented in this work.
The diffusion-controlled growth of a c (austenite) region
into a a + M2N (ferrite) region (where M2N is a nitride
with two metallic (M) atoms per each nitrogen (N) atom)
was simulated. A closed one-dimensional multi-component
system and a planar sharp c/a interface were assumed.
The atomic diffusion in both ferritic matrix and
austenitic grains was computed by solving the multicom-
ponent non-steady state atomic diffusion equation 1 [13].
The interface velocity was calculated using the mobile
interface model [11], which is implemented in Dictra
software. The establishment of the local equilibrium (LE)
condition at the migrating c/a interface was assumed.
Si Mo Cu C N
0.41 0.30 0.33 0.014 0.10
J Mater Sci (2007) 42:90219029
X n
0 ol
JK  Lki i 1 (NC) of the main elements in the ferritic and austenitic grains of a
i1
oz
0 Phase
In Eq. 1, JK is the flux of the atomic species k; Lki is a
proportionality factor that depends on the mobility of the
atomic species; li is the chemical potential for the species
in the system; and z is the spatial coordinate.
Nitrides can precipitate or dissolve within the ferritic
matrix as function of temperature and local alloying con-
tent. This aspect was taken into consideration by using the
diffusion in dispersed systems model [11], which is also
implemented in Dictra software.
To compute the growth of austenitic grains, the so-
called primary austenite (c1), during ferritization treat-
ments, it was assumed the prior existence of 1 nm thick
gamma nucleus that begins to grow during the cooling
stage at 1,200, 1,100, 1,000, or 900 C. The chemical
composition of this gamma nucleus was set at the equi-
librium value corresponding to the temperature at which
this nucleus begins to grow. However, to compute the
growth of secondary austenite (c2) during reheating treat-
ments, it was assumed that c2 grows over the previously
precipitated c1 grains. The average c1 width was deter-
mined by metallographic measurements and its average
chemical composition taken from the previously described
precipitation simulations. The size and chemical compo-
sition of the remnant ferritic regions were established
through mass balance computation by calculating the
amount of a + M2N necessary to reach the steel nominal
composition.
Results
Microstructure and chemical composition
The previously reported ferrite grain size and c1 fraction on
ferritized samples were 240 lm and 16 2%-vol, respec-
tively [6]. In addition, the weighted average width of c1
grains was ~7 lm.
After numerical simulation of c1 precipitation, it was
predicted that c1 grains grow with negligible substitutional
solute partitioning with the matrix. Furthermore, it was
observed that the average chemical composition of c1 did
not change appreciably as the precipitation temperature
was varied between 1,200 and 900 C. Therefore, the
average chemical compositions of c1 and remnant ferritic
regions were obtained from these calculations. The calcu-
lated chemical compositions as well as SEM-XWDS
measurements are presented in Table 2. One can see that
the calculated values are close to the measured ones. The
only important difference is the nitrogen content of c1. It
9023
Table 2 Chemical analysis (XWDS) and numerical calculations
ferritized sample (wt-%)
Fe Cr Ni C N
a XWDS Bal. 24.0 4.6 N.M N.M
NC Bal. 22.7 4.8 0.009 0.06
c1 XWDS Bal. 23.3 4.9 N.M 0.48
NC Bal. 22.4 4.8 0.039 0.31
N.M: Non measurable value (content was lower than the detection
limit)
must be noted that large experimental errors are expected
with XWDS measurements of light element such as
nitrogen.
Table 3 presents the Fe, Cr, and Ni weight fraction of
nitrides and the ferritic matrix after the ferritization treat-
ments as measured by TEM-XEDS and as determined via
computational modeling. Because the chemical composi-
tion of the ferrite measured via XWDS-SEM was used as a
standard for the XEDS measurement in the TEM, both
compositions are forced to be identical. It is not possible to
measure the C and N contents using XEDS in the TEM.
Therefore, only the composition of the primary solute
elements (metal) is reported. The XEDS analyses of the
nitrides confirmed their expected high Cr content. The
experimental and calculated values presented in this table
show that the average chromium depletion induced in the
matrix during the ferritization treatment and as a conse-
quence of nitride precipitation is low; the average chro-
mium content of the as-welded a + M2N (ferritic) regions
is ~22.7 wt-% and the computed average chromium
metallic fraction of the ferritic matrix is 22.3 wt-%.
However, according to literature [11], narrow chromium
depleted zones are established around chromium nitride
precipitates, which can be strongly detrimental to localized
corrosion resistance.
Figure 1 shows the calculated chromium, nickel and
nitrogen composition profiles established in the overall
microstructure after reheating treatments. As presented in
Fig. 1a, intergranular c2 formed during the 1,200 C iso-
thermal anneal exhibited a heterogeneous chemical com-
position profile, which has been experimentally verified by
the authors using XEDS microanalysis in the TEM [12].
This c2 is composed of an outer shell (next to the c/a
interface) enriched in Cr and N, and a core region (next to
c1) depleted in these alloying elements. These unusual
solute profiles are established due to the intergranular c2
growth during the high temperature isothermal anneal. The
high diffusivity of the substitutional alloying elements at
high temperatures makes this partitioning possible. How-
ever, during the cooling stage the limited diffusivity of the
123
9024 J Mater Sci (2007) 42:90219029

Table 3 TEM measured chemical analysis (XEDS) and numerical In contrast to the solute partitioning observed as a result
calculations (NC) of the metallic fraction of the main elements in the of the isothermal anneal at 1,200 C, the c2 formed at
ferritic matrix and nitrides in a ferritized sample (wt-%)
900 C exhibits a nearly homogeneous chemical compo-
Phase Fe Cr Ni sition, as presented in Fig. 1b. The c2 formed at the lower
temperature inherits the substitutional alloying concentra-
a XEDS 70.4 24.0 4.6
tion of the ferritic matrix due to low diffusivity of substi-
NC 72.85 22.3 4.84
tutional elements.
M2N XEDS 8.8 89.9 0.18
SEM and TEM analyses of nitride precipitation were in
NC 2.4 97.6 0.05
good agreement with numerical simulations. A consider-
able amount of chromium nitride precipitates were ob-
served in ferritized as well as 900 C reheated samples,
whereas virtually no nitrides were observed in the 1,200 C
reheated samples. Figure 2 shows TEM micrographs of the
general view of intragranular nitrides observed in a sample
reheated to 900 C. The intragranular nitrides varied in
width between approximately 40 and 120 nm. Figure 2a
shows a general view of a ferritic region containing Cr2N
nitrides (marked with arrows in the figure) and also some
dislocations. In Fig. 2b, a selected area diffraction (SAD)
pattern from the ferritic matrix, at the axis zone [011] a is
shown. Figure 2c presents an enlarged detail of one of the
nitrides in Fig. 2a; wider darkbright contrast fringes
inside nitrides can be seen, which correspond to Moire
patterns. A detailed crystallographic study of this nitride
precipitation is presented in a previous work [6], where
these nitrides were identified, by nano diffraction analysis,
as trigonal Cr2N.

Electrochemical experiments

Figure 3 shows the relationship between potential and

Fig. 1 Calculated alloy content across the system for samples
reheated at 1,200 and 900 C
substitutional elements leads to the occurrence of c2 growth
with negligible substitutional solute partitioning with the
parent ferrite.
current density established during the electrochemical pit-
ting and DL-EPR experiments of HAZ microstructures.
One representative curve of five tests is presented. In
addition, Table 4 gives the average pitting potential and
degree of sensitization (DOS) from the potentialcurrent
density curves. The lower corrosion resistance of HAZ
microstructures compared to the wrought material is
clearly evident. The pitting potential of the wrought
material is 535 mV-SCE whereas the pitting potential of
the HAZ microstructures is between 332 and 443 mV-
SCE. Intense sensitization occurrence in the as-welded and
in the 900 C reheated HAZ microstructures was observed.
On the other hand, the wrought and 1,200 C reheated
HAZ microstructures were not sensitized. One can addi-
tionally see in Table 4 a moderate increase in corrosion
resistance after reheating treatments when compared to the
ferritized samples. The occurrence of sensitization is
explained by the SEM and TEM observations of Cr nitride
precipitation and by the modeling results of nitrogen
supersaturation in ferritic regions: the larger the nitrogen
supersaturation in the ferritic regions, the larger the

123
J Mater Sci (2007) 42:90219029 9025

Fig. 2 TEM micrographs

showing general view of nitride
precipitation in a sample
reheated at 900 C. Wider dark
bright contrast fringes inside
nitrides correspond to Moire
patterns

sensitization due to Cr nitride precipitation. Detailed

discussion is presented in Analysis and discussion.
Figure 4 shows the surface appearance of the sample
reheated at 900 C after it has been subjected to the pitting
corrosion tests. The sample surface appearance for all the
other heat treatment conditions is similar to the one shown
in Fig. 4. The largest part of the sample surface (almost
95%-area) was only slightly affected by the pitting

Fig. 3 Current density vs. Potential during pitting (a) and DL-EPR
(b) electrochemical corrosion tests

Table 4 Pitting potential and degree of sensitization of the DSS UNS

S32304 in the wrought condition, after ferritization and after reheating
at 1,200 and 900 C
Condition Pitting potential Degree of sensitization
(mV-SCE) (dimensionless)

Wrought material 535 52 0.0

Ferritized 332 30 8.4 1 104 Fig. 4 SEM micrographs showing the surface appearance of sample
(1,350 C) reheated to 900 C after being subjected to pitting corrosion
Reheated 1,200 C 443 30 0.0 electrochemical tests. (a) General appearance of the surfacewhich
900 C 401 27 1.5 0.5 104 represents ~95% of the test surface-; (b) appearance of deeply
damaged regions

123
9026
corrosion test; in these regions (Fig. 4a) a low density,
between approximately 50 and 200 pit
mm2, of small pits,
between 1 and 3 lm in diameter, was observed. However,
a few regions, with a number of regions per unit area
ranging between approximately 0.15 and 0.35 region
mm
2
,were observed that were severely corroded (Fig 4b).
These regions, which varied in size from approximately
200450 lm, consisted of a network of connected and
unconnected pits of diverse sizes. Qualitatively, it was
observed that the extent of damage in these regions fol-
lowed the same order as the pitting potential: wrought
material < 1,200 C reheated < 900 C reheated < ferri-
tized.
Although it is not unambiguously clear, large side plates
(possibly austenite) are evident at the sample surface of the
corroded samples, Fig. 4b, that may possibly act as barriers
to additional pitting damage. Aiming to further clarify the
actual localization of pit-corrode regions among ferrite or
austenite regions in microstructure, a sample reheated at
1,200 C was electrolytic-etched after being submitted to
pitting corrosion test. Figure 5 shows the appearance of
that sample reheated at 1,200 C after the pitting corrosion
test followed by electrolytic etching. Although the elec-
trolytic etching significantly modifies the prior appearance
of the surface subjected to pitting test, it can be seen that
pit-damaged regions were inside prior ferritic grains. As
can be seen in this figure, the coarse intergranular allotri-
omorphic austenite and intragranular austenite regions
acted as barriers for the growth of pitting damage. The
detail presented on the top right of Fig. 5 shows the barrier
effect of these austenite particles.
Figure 6 shows the appearance of the sample surface
after DL-EPR tests. Corrosion attack of the sensitized areas
in the ferritized sample as well as the one reheated at
900 C occurred, but no such attach was observed in the
Fig. 5 SEM micrographs showing the appearance of the sample
surface after pitting electrochemical corrosion test pursued by
electrolytic etching. Sample reheated to 1,200 C
123
J Mater Sci (2007) 42:90219029
1,200 C reheated samples. Corrosion attack during DL-
EPR tests was mainly observed in two locations: (i) the
middle region of ferritic grains where intense chromium
nitride precipitation is expected and (ii) the near region of
some c/a interfaces, where chromium nitride precipitation
is also expected, as previously discussed. The superior
localized corrosion resistance of both c1 and c2 grains was
observed.
The effect of alloying redistribution on the computed
PRE index profiles
Figure 7 shows the calculated pitting resistance equivalent
index (PRE = wt-% Cr + 3.3 wt-% Mo + 16wt-%N) pro-
files, established in the overall microstructure after
reheating treatments. Aiming to simplify the kinetic com-
puting simulations, which computing time strongly in-
creases as a consequence of increasing the number of
alloying elements included in the system, the molybdenum
content was not used in the PRE computations. This sim-
plification should exert a minor effect on the overall results
because, (i) the Mo-content of the alloy is low (0.3 wt-%)
and (ii) prior thermodynamic simulations (not presented)
shown atomic partitioning of this element in microstructure
contributes to PRE differences in a minor extent when
compared to the effect of Cr- and N-content. In addition,
only the elements in solid solution within the ferrite and
austenite were taken into consideration to compute the PRE
index profiles, i.e. the elements present in the form of M2N
where not taken into account.
The most significant aspect regarding the PRE index
profiles (Fig. 7) is the low PRE index (22.3) of ferritic
regions in the as-welded high temperature HAZ micro-
structures (ferritized samples) compared to the wrought
material. In the wrought material the PRE index of a and c
are 25.3 and 23.1, respectively. On the other hand, the PRE
index of reheated high temperature HAZ microstructures
varies between 21.8 and 22.6 for a and 23.5 and 32.7 for c.
This reduction in PRE index suffered by ferritic regions, as
a consequence of the welding thermal cycles, is mainly due
to the almost absence of chromium partitioning between
the austenitic (c1 + c2) and ferritic grains in the HAZ mi-
crostructures. This behavior is quite different to the ob-
served in the wrought material, where the carefully
selected combination of chemical composition and solu-
bilization heat treatment results on an almost balanced PRE
index between the two phases. Care should be taken be-
cause the reported PRE reduction within the ferritic regions
(a + M2N) has not taken into account the localized Cr
depletions around M2N precipitates, which will have a
strong influence on the localized corrosion resistance of
such microstructures.
J Mater Sci (2007) 42:90219029
Fig. 6 SEM micrographs
showing the appearance of the
sample surface after DL-EPR
electrochemical corrosion tests
of (a) ferritized, (b) reheated to
900 C, and (c) reheated to
1,200 C samples
Fig. 7 Calculated PRE index profiles across the system for samples
reheated at 1,200 and 900 C. In the general PRE index equation
(PRE = wt-% Cr + 3.3 wt-% Mo + 16wt-%N) the Mo-content effect
was disregarded
The PRE index was developed to predict the overall
pitting corrosion resistance of stainless steels in chloride
solutions and this does not necessary predict the corrosion
resistance in neither other solutions nor the differences of
corrosion resistance inside microstructure among different
constituents. Despite of this, in the next Analysis and
discussion, it is observed that the computed PRE-index
profiles are a helpful tool to qualitatively describe the
results of SEM analysis of the appearance of the sample
surface after electrochemical experiments.
Analysis and discussion
The composition profiles calculated in this work for alloy
UNS S32304, which technically is a Mo-lean alloy, should
9027
not be extrapolated to other duplex stainless steels as UNS
S32205 or UNS S32750 because the effect of Mo on the
thermodynamics and kinetics of those systems have not
been evaluated in the present research. However, the
present results can be used with care to evaluate previously
reported results on the intragranular c2 precipitation and its
effect on the corrosion resistance of DSS. In the case of
intragranular c2, the nucleation of new austenite particles
precedes austenite growth. Previous work has indicated
several possible nucleation mechanisms for such new
particles [2, 6, 7]. Among these mechanisms is heteroge-
neous austenite nucleation on dissolving nitrides [6, 7] and
inclusions [2], which is more important in weld metal. In
addition, sympathetic nucleation has been suggested [2].
Disregarding the nucleation mechanism, the growth of in-
tragranular c2 will have some differences with respect to
the growth of intergranular c2. One of the most important
differences can be the abundant supply of nitrogen and
chromium within the dissolving nitride colony where the
intragranular c2 particles are growing. However, the much
larger diffusivity of nitrogen compared to chromium
suggest that Cr diffusion can be the controlling factor on
intragranular c2 growth. Therefore, the high chromium
content within this changing colony should play an
important role on the acicular morphology of these
particles.
In addition to the previously described differences
between intergranular and intragranular c2 growth, there
are some relevant similarities that should be mentioned.
These similarities include the apparent controlling effect of
chromium diffusion and the inhomogeneous composition
profile along c1/c2/a interfaces. It has been previously
reported experimental evidence of such chemical inho-
mogeneity along c1/c2/a interfaces on very small prior-
intragranular c1 that precipitated within nitrides colonies,
123
9028
which grows as c2 within these changing nitride colonies
during reheating treatments [7]. Therefore, these similari-
ties strongly suggest that the results here presented for
intergranular c2 growth can be a reasonable approximation
to intragranular c2 growth and of course to its effects on the
corrosion resistance of UNS S32304 DSS. Further
research, including numerical simulation is necessary to
clarify these similarities. However, the simulation of
intragranular c2 nucleation and growth requires more
elaborated models and one dimension approximations as
the ones used here to model intergranular c2 growth may
not be considered satisfactory.
The calculated chemical profiles presented in Fig. 1 do
not fully agree with previously reported results on the
chemical composition of intragranular c2 on super-duplex
stainless steels with additions of Cu and W [8, 9]. On the
other hand, the predicted and measured [12] composition
profiles support corrosion resistance results for DSS UNS
S13803 [14], which exhibits a similar chemical composi-
tion to the studied alloy UNS S32304. However, in addi-
tion to the differences in chemical composition of the
Superduplex alloys with Cu and W additions, the calcula-
tions presented here are based on more elaborate kinetics
models, which are expected to better represent the system.
However, the simplifications here used include the reduc-
tion of the system to one dimension and the isothermal
treatments. Thus, important differences with actual welds
may be expected, especially for Superduplex alloys. Nev-
ertheless, this new contribution helps to better understand
the chemical partitioning among ferrite and austenite in
duplex stainless steels and particularly its effects on cor-
rosion resistance.
The PRE index of ferritic regions among the different
HAZ microstructures, namely as-welded HAZ (ferritized
samples) and reheated-HAZ (1,200 and 900 C reheated
samples), suffered only minor variations. However, the
nitrogen content within the ferrite, and thus the fraction of
chromium nitrides precipitated in this phase, underwent large
changes. These changes have an important effect on the
corrosion resistance of such microstructures. The predicted
supersaturation of ferritic regions with nitrogen in the as-
welded HAZ microstructure, which results in abundant in-
tragranular M2N precipitation, was reduced after reheating
treatments. The 900 C reheating treatment induced a mod-
erate nitrogen content reduction in the ferritic regions near to
the growing austenite, as shown in Fig. 1b. However, the
ferritic region immediately adjacent to the a/c2 interface
underwent an important nitrogen enrichment, which explains
the previously reported nitride (M2N) precipitation at the a/c
interface [6]. On the other hand, the 1,200 C reheating
treatment induced an almost complete exhaustion of nitrogen
from ferritic regions. In both cases, a large amount or almost
all the M2N nitrides that formed within nitrogen
123
J Mater Sci (2007) 42:90219029
supersaturated ferrite during the fast cooling following the
ferritization treatment, dissolved during the reheating treat-
ments and the nitrogen is incorporated by the growing aus-
tenite. The long diffusion paths during reheating at high
temperatures allow for the nitrogen to be mostly incorporated
by preexisting austenite particles that are growing, resulting
in nitride-free ferrite, as verified by the presented calculation.
On the other hand, the shorter diffusion paths experienced at
lower temperatures favor a modest growth of preexisting
austenite particles and the formation of acicular-type intra-
granular c2 colonies within the preexisting nitrides colonies
that were far apart from austenite particles [7].
The superior localized corrosion resistance of both c1
and c2 grains was observed, even for the 1,200 C reheat-
ing treatments in which relatively low PRE index value in
the middle of c2 grains was predicted (Fig. 7). This result
agrees with reported research on DSS UNS S31803 [14].
Besides localized corrosion occurrence in sensitized areas,
DL-EPR test leads to generalized corrosion [15]. The
extent of localized corrosion at phase interfaces was dif-
ferent in the diverse HAZ microstructures (Fig. 6). In the
ferritized samples as well as the 900 C reheated samples
the corroded regions adjacent to the phase interfaces were
narrow, contrary to the 1,200 C reheated samples in which
this corrosion damage occurred along a broader region
around to the interface. This is probably due to occurrence
of galvanic effects between a and c grains, which have
different Cr- and Ni-content in the near interface region
(Fig. 1). The largest difference in Cr- and Ni-content in the
near interface region between a and c was in the 1,200 C
reheated samples (see Fig. 1), which were the samples
displaying the broadest corrosion at phase interfaces. One
can also see in Fig. 6 that some intragranular austenitic
grains can be detached from the ferritic matrix as a con-
sequence of phase interface corrosion, which is due to the
above pointed out galvanic effect, as well as the large
surface/volume ratio and small size of these c particles.
Thus, this small intragranular austenite particles detach-
ment from the sample will worsen the corrosion damage in
the material. This can be the mechanism behind the pre-
viously reported damaging effect of intragranular c2 on the
corrosion resistance of duplex stainless steels. However,
more research in different duplex alloys will be necessary
to generalize this mechanism.
Summarizing, the numerical simulation and experi-
mental results revealed that the moderately low PRE index
established in ferritic regions of welded DSS UNS S32304
is the main cause for lower localized corrosion resistance
of HAZ microstructures when compared with the wrought
material. In addition, it can be also inferred that chromium
nitride precipitation within ferritic regions further impairs
the corrosion resistance of HAZ microstructures due to
sensitization occurrence.
J Mater Sci (2007) 42:90219029
Thus, selective corrosion inside gamma grains, c1 and
c2, is not the main cause for the poor corrosion resistance of
DSS UNS S32304 HAZ microstructures. However, the
fraction of chromium nitrides within the ferritic regions
indirectly depends on the total fraction of austenite
(c1 + c2), as austenite grains exhaust the nitrogen from the
ferritic regions. In addition, at high reheating temperatures,
the nitrides modify the substitutional alloy content profiles
in the region near to the c/a interfaces.
Precipitation of austenite displaying superior PRE index
in the region adjacent to the c/a interfaces when compared
with ferrite can deteriorate the generalized corrosion
resistance of HAZ microstructures due to galvanic effects.
This effect should be particularly important in regions
where profuse precipitation of small intragranular austenite
particles occurs. In this case ferrite around these c particles
can be easily damaged due to both the large surface/volume
ratio of c particles and their small size.
Conclusions
The results and analyses from the calculations and exper-
imental measurements on heat-treated DSS UNS S32304
have led to the following conclusions:
The electrochemical pitting and DL-EPR tests showed
an impaired corrosion resistance of HAZ microstruc-
tures with respect to the wrought material. This
reduction of corrosion resistance after welding was
mainly attributed to a decrease in the corrosion
resistance of the HAZ ferritic grains. This was attrib-
uted to (i) the mean chromium content of these ferritic
grains is the same chromium content of the alloy (i.e.
ferritic grains in the HAZ exhibited lower chromium
content than ferritic grains in the wrought material) and
(ii) the presence of chromium depleted regions around
nitrides, which were often observed.
As-welded HAZ microstructures displayed lower local-
ized corrosion resistance and higher degree of sensiti-
zation than reheated HAZ microstructures. The extent
of corrosion resistance improvement after reheating
treatment was mainly caused by the removal of
9029
nitrogen from ferritic regions, which occurred as a
consequence of c2 growth.
The observed corrosion resistance improvement was
more pronounced after isothermal reheating to
1,200 C, compared to the 900 C reheating treatment,
which appears to be directly related to the removal of
nitrogen within the ferrite grains.
In contrast with what has been reported in literature for
superduplex stainless steels [2, 3], a direct detrimental
effect of intergranular c2 precipitation on corrosion
resistance was not observed. Selective corrosion inside
gamma grains, c1 and c2, was not the main cause for the
impaired corrosion resistance of DSS UNS S32304
HAZ microstructures. However, the observed ferritic
matrix dissolution around intragranular c2 suggest that
small gamma particles detachment from the material is
the probable cause for a high corrosion rate.
Acknowledgments The authors would like to acknowledge to
LNLS and FAPESP, for funding the actual and predecessor projects.
References
1. Ogawa TO, Koseki T (1989) Weld J 68:181
2. Atamert S, King JE (1991) Z Metallkd 82:230
3. Nilsson JO, Karlsson L, Andersson JO (1995) Mat Sci Techn
11:276
4. Gooch TG (1996) Weld J 75:135
5. Gregory A, Nilsson J-O (2002) Met Mat Trans A 33A:1009
6. Ramirez AJ, Lippold JC, Brandi SD (2003) Met Mat Trans A
34A:1575
7. Ramirez AJ, Ph.D. Thesis, University of Sao Paulo, Sao Paulo,
Brazil, 2001, p. 241. Full text available at http://www.te-
ses.usp.br/teses/disponiveis/3/3133/tde02012002175418/
8. Hertman S, Huhtala T, Karlsson L, Nilsson J-O, Jargelius-
Pettersson R, Wilson A (1997) Mater Sci Tech 13:604
9. Nilsson J-O, Huhtala T, Jonsson P, Karlsson L, Wilson A (1996)
Met Mat Trans A 27A:2196
10. Majidi AP, Streicher MA (1984) Corrosion 40:393
11. Andersson JO, Helander T, Hoglund LH, Shi PF, Sundman B
(2002) CALPHAD 26:273
12. Garzon CM, Ramirez AJ (2006) Acta Mater 54:3321
13. Andersson JO, A gren J (1992) J Appl Phys 72:1350
14. Muthupandi V, Bala Srinivasan P, Seshadri SK, Sundaresan S
(2004) Sci Technol Weld Joi 9:47
15. Menendez H, Devine TM (1990) Corrosion 46:410
123
J Mater Sci (2007) 42:90309036
DOI 10.1007/s10853-007-1860-y
The effect of Sm substitution on properties of Bi1.6Pb0.4Sr2Ca2x
SmxCu3Oy superconductors
Mustafa Yilmazlar Huseyin Aydin
Ahmet Varilci Cabir Terzioglu
Received: 21 February 2007 / Accepted: 17 May 2007 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007
Abstract The effect of the partial substitution of Ca by
Sm in the Bi-2223 superconducting samples have been
investigated in terms of X-ray diffraction (XRD), EDXRF
(Energy Dispersive X-ray Fluorescent), magnetoresistivity,
critical temperature, transport critical current density, and
ac susceptibility measurements. The samples were pre-
pared by the conventional solid-state reaction method.
XRD patterns are used to calculate lattice parameters and
phase ratio of the Bi-2223 samples. The volume fraction
was determined from the intensities of Bi-2223 and
Bi-2212 peaks. The room temperature XRD patterns of the
samples showed the presence of Bi-2223 phase decreases
with increasing the Sm content. We estimated the transition
temperature of the samples from the resistivity versus
temperature measurements in dc magnetic fields up to
0.6 T. We observed that transition temperature, Tc, and
transport critical current density, Jctrans , depend on the Sm
substitution. They both decrease with increasing the Sm
substitution. We extracted the peak temperature, Tp, and
the pinning force density from our previous ac suscepti-
bility measurements. The pinning force density decreased
with increasing the Sm content. The possible reasons for
the observed decreases in critical temperature and critical
current density due to Sm substitution were discussed.
M. Yilmazlar
Faculty of Education, Sakarya University, Sakarya,
Hendek 54300, Turkey
H. Aydin
A. Varilci
C. Terzioglu (&)
Department of Physics, Faculty of Arts and Sciences,
Abant Izzet Baysal University, Bolu 14280, Turkey
e-mail: terzioglu_c@ibu.edu.tr
123
Introduction
It is understood that the high temperature superconductors
(HTSCs) show unusual behaviors in the mixed state,
smearing from their layered structure, high transition
temperature, short coherence length and etc. Among these
unusual behaviors, giant flux creep is suggested to be an
important one, which has become an interesting subject for
both theory and experiment in the past several years [1].
The results are used to discuss the flux pinning and ther-
mally activated flux motion in the mixed state of HTSCs.
These are related to the magnetic and electric properties of
HTSCs in the magnetic field. The broadening of resistive
transition in the magnetic field is also attributed to the
thermally activated flux motion [2, 3]. Some groups [47]
have done studies of flux pinning and flux activation, and
several models have been proposed. Among these, both
vortex-glass model [8] and collective-pinning model [9]
predict the same power-law form for the dependence of the
activation energy Ueff (J) on current density. According to
the thermally activated flux motion model, the essential
problem is to study the nonlinear Ueff (J) dependence of the
HTSCs.
Magnetic relaxation and flux creep exists in type II
superconductors causing decay of the irreversible magne-
tization. The irreversibility line can be determined from
various measurements such as dc resistivity, ac suscepti-
bility, and etc. [10]. Muller et al. [11] first reported the
irreversibility line in HTSCs. The measurements of ac
susceptibility are commonly used to determine magnetic
and superconducting properties of materials. In particular,
the ac susceptibility measurement is useful in determining
the specimen and for distinguishing between inter- and
intragrain properties. For sintered high-Tc superconductors
the temperature dependence of the imaginary part of the
J Mater Sci (2007) 42:90309036
complex ac susceptibility shows two peaks, Tp and Tg, at
high ac fields [12, 13]. The first peak (at high field) appears
at a temperature Tg slightly bellow Tc due to intragranular
supercurrents, and a second loss peak (at low-field) appears
at a temperature Tp lower than Tg due to sample-circulating
intergranular supercurrents.
Doping with different elements at various amounts and
adjusting preparation condition affect phase formation and
physical properties in Bi-based system. Many studies of
doping into superconductor oxide ceramics have been
made in order to improve their superconducting, magnetic
and mechanical properties [1421]. The objectives of
doping studies are to optimize the hole concentration, to
introduce pinning centers and to enhance the formation of
Bi-2223 phase. Li, Ti, W and Mo have been doped into
BiPbSrCaCuO bulk samples, and it was reported that Li
doping led to an increase in critical current density (Jc),
while Ti doping led to a decrease, and W and Mo doping
did not contribute a substantial change in Jc [23, 24].
Substitution of Eu, Dy and Tm in the Bi-2212 system has
caused a transition from superconductor to insulator
[2426]. The partial replacements of Ca by Y, Gd, and Sm
in the Bi-2212 and Bi-2223 systems decreased carrier
concentration due to structural modulations and changes in
the valency of Cu [2528]. On the other hand, it was
observed that the substitution of divalent Ca by trivalent
rare-earth elements in the Bi-2212 system causes a sharp
change in transition from a superconductor to an insulator
and a decrease in the lattice parameter c because of a
drastic change in the carrier concentration [22, 27]. A
comprehensive study of Sm addition to Bi-2212 and
Bi-2223 was reported in the literature [1619, 29]. In our
previous study, we investigated the effect of the partial
substitution of Ca by Sm in the Bi-2223 superconducting
samples on its structural and physical and magnetic prop-
erties [1619]. It was obtained that superconducting (the
values of Tc and Jc) and mechanical properties degrade and
the lattice parameter c and the Bi-2223 phase decreased
with increasing Sm content. It was also reported that for
x = 1.0 and 1.5 the compounds showed semiconducting
behavior. We investigated effect of the partial substitution
of Ca by Sm in Bi(Pb)SrCaCuO system by using the low
field ac magnetic susceptibility method [18]. The investi-
gations consisted of ac susceptibility, the temperature
dependence of intergranular critical current density,
Jcinter Tp , and scanning electron microscopy (SEM) mea-
surements. It was observed that the grain size of the pure
sample is considerably larger than the Sm-doped samples,
the overall susceptibility curves of the samples are shifted
to lower temperature and the value of Jcinter Tp decreases
by increasing the Sm concentration.
In this work, we report measurements of electrical
resistivity as a function of temperature under magnetic field
9031
up to 0.6 T and transport critical current density in order to
investigate the effect of the partial substitution of Ca by Sm
on superconducting properties of Bi-2223 samples. We
also reported XRD measurements to extract the lattice
parameters and the relative portion of Bi-2223 and that of
Bi-2212 phases and EDXRF measurements to estimate
stoichiometric ratios of the considered samples.
Experimental details
The Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy samples with
0 < x < 0.005 Sm were prepared by using the standard
solid-state reaction method [16]. Rectangular bars were cut
from the sintered samples for electrical resistivity, critical
current and ac susceptibility measurements. The typical
sample size was 2.9 2.1 12.1 mm3.
XRD data were taken using a Rigaku D/Max-IIIC dif-
fractometer with CuKa radiation in the range 2h = 460
with a scan speed of 3/min and a step increment of 0.02
at room temperature. Phase purity and the lattice parame-
ters were obtained from these XRD patterns. The accuracy
in determining the lattice parameter c was 0.001 A . The
mean values of lattice parameter c for Bi-2223 phase are
determined from (002), (0010), (0012), (115) and (0014)
peaks of the XRD measurements for pure and Sm substi-
tuted Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy samples. The relative
proportions of the Bi-2223 phase were determined from
(002), (008), (0010), (115), (0012), (200), (1111) and
(0212) peaks and the relative proportions of Bi-2212 phase
were determined from (115), (220), (315), (317), (1115)
and (1112) peaks. The relative proportions of the Bi-2223
phase were determined from (002), (008), (0010), (115),
(0012), (200), (1111) and (0212) peaks and the relative
proportions of Bi-2212 phase were determined from (115),
(220), (315), (317), (1115) and (1112) peaks, using the
following well-known expressions [30, 31];
P
IH hkl
f2223 P P 1
IH hkl ILhkl
P
ILhkl
f2212 P P : 2
IH hkl ILhkl
Here IH(hkl) and IL(hkl) are the intensities of the (hkl)
diffraction lines for Bi-2223 and Bi-2212 phases, respec-
tively (Fig. 2).
The transport critical current (Ic) (77 K, self field,
1 lV cm1) and dc resistivity (5 mA dc current) were
measured by a conventional four-probe method. The con-
tacts were made with silver paint. The outside pin-points
feed the current and the intermediate ones (l = 7 mm)
allow the voltage drop measurement. A Keithley 220
123
9032
programmable current source and a Keithley 2182A nano-
voltmeter were used for the resistivity and IV measure-
ments. The Ic was converted to Jctrans by dividing by the
area of cross section of the samples. Actual dc currents
were run through the samples up to 5 A. Resistive contacts
caused heating and the test was completed in possible short
time, in order to avoid this effect.
We measured temperature (40130 K) and dc magnetic
field (00.6 T) dependent of resistivity of the samples with
5 mA dc current in a cryostat. The magnetic field was
applied parallel to the current direction. The external dc
magnetic fields for resistivity measurements were provided
by an electromagnet. The transition temperature Tc was
determined as the temperature at zero resistivity.
Results and discussion
In our previous work, we investigated effect of the partial
substitution of Ca by Sm in Bi(Pb)SrCaCuO high tem-
perature superconductors by using the low field ac mag-
netic susceptibility method [18]. The investigations
consisted of ac susceptibility as a function of temperature
(77120 K) under different ac field amplitudes, the tem-
perature dependence of intergranular critical current den-
sity, Jcinter Tp , and SEM measurements. It was observed
that the overall susceptibility curves of the samples are
shifted to lower temperature by increasing the ac field
amplitude. The intensity of the imaginary parts increased
with increasing the ac field amplitude. The susceptibility
temperature curves shifted to lower temperatures and
considerably increased the transition width in v T
curves, and also decreased the shielding fraction of the
superconducting phase in the samples with increasing the
Sm concentration. In the calculation of Jcinter Tp as a
function of the peak temperature from our previous work
[18], it was found that the value of Jcinter Tp increases with
decreasing the Sm content and temperature. From the
previous SEM analysis [16], it was observed that the grain
size of the pure sample is considerably larger than the
Sm-doped samples.
In the present study, we have studied the peak temper-
ature dependence as a function of ac magnetic field, Hac, in
order to investigate the effect of partial substitution of Ca
by Sm on the intergranular pinning force. The values of Tp
were estimated from our previous ac susceptibility versus
temperature measurements for the ac field amplitudes of
111, 222, 333, 444 and 555 A/m at 1 kHz [18]. It was also
observed that peak temperatures, Tp, shift to lower tem-
perature with increasing ac field amplitudes. The shift of Tp
with field can be explained in terms of the strength of the
pinning force. As can be seen from Fig. 1 a linear depen-
dence of Tp as a function of Hac is observed for all the
123
J Mater Sci (2007) 42:90309036
Fig. 1 Intergranular peak temperature as a function of ac magnetic
field amplitude for x = 0.0000, 0.0005, 0.0010, and 0.0050 Sm
samples. This is consistent with the previous works
[3234]. Muller critical state model assumes a magnetic flux
independent pinning force densities, and aJ and ag for inter-
and intragranular vortices described by the relation [33]:

1=2
l0 leff 0
Tp Tp0  Tp0 Hac 3
2aaJ 0
where a is the height of the cylinder samples, leff (0) is the
effective permeability of the ceramic and aJ (0) is the
intergranular pinning force density. From a least squares fit
of this expression (Eq. (3)) to the data Tp0 and
l0 leff 0=2aaJ 0 were extracted for all samples. The
values
 of Tp0  decreased while the values of
l0 leff 0=2aaJ 0 increased with increasing the Sm
content. This means that the values of aJ (0) decrease with
increasing the Sm concentration. A decreasing trend in
pinning force of the samples with increasing the Sm con-
tent is attributable to greater voids and defects as revealed
by our previous SEM images [16]. This result is also
consistent with the temperature dependence of Jcinter(Tp) in
the previous study [18]. The effective pinning force density
decreases with increasing ac magnetic field causing Tp to
shift to lower temperature. The rapid decrease of the
activation energy induces the steep dropping of the critical
current density with increasing magnetic field.
The XRD patterns of pure and Sm substituted samples
(x = 0, 0.0005, 0.001 and 0.005) are shown in Fig. 2. XRD
pattern was taken by use of CuKa radiation and measure-
ments were taken at room temperature. All peaks in the
XRD pattern are indexed. There is a significant increase in
the intensities of particular peaks between the pure and Sm
substituted samples. As can be seen from the figure, the
intensities of the Bi-2223 peaks decrease and the intensities
J Mater Sci (2007) 42:90309036 9033

Fig. 2 The XRD patterns for

a) x=0.0

(0012)H
(0010) H
(a) pure sample and Sm
substituted samples (b) x =

(0014)H
0.0005, (c) 0.001, and (d) 0.005.

(200)H
(002)H

(119)H

(1111)H
(115)H
The peaks indexed (hkl)L and

(0212)H

(0024)H
(220)H
(220)L

(1119)H
(hkl)H represent the Bi-2212

(115)L
(008)H

(317)L
(315)L
(004)H
and Bi-2223 phases,
respectively
b) x=0.0005

(0012)H
(0010)H

(0014)H
(200)H
(002)H

(1111)H
(119)H
I nt e ns i t y (a . u. )

(115)H

(0024)H
(0212)H

(1119)H
(220)H
(008)H

(2012)L
(220)L

(315)L
(004)H

(317)L
(115)L
c) x=0.001

(0010) H

(0012)H

(0014)H
(002)H

(200)H
(119)H

(1111)H
(115)H

(0024)H
(220)H
(0212) H

(1119)H
(220)L

(2012)L
(008)H

(315)L
(317)L
(004)H

(115)L
d) x=0.005

(0012)H
(0010)H

(0014)H
(002)H

(119)H
(200)H
(115)H

(1111) H

(0212) H

(0024)H
(220)L

(1119)H
(220)H
(008)L

(2012)L
(115)L

(1115)L
(008)H
(004)H

5 10 15 20 25 30 35 40 45 50 55 60
2 ( Degree)

of the Bi-2212 peaks increase with increasing substitution lattice parameter c is due to incorporation of Sm ions into
of Sm for Ca. The determined lattice parameters from the the interstitial sites in the unit cell rather than occupation of
(00l) peaks of the XRD data are given in Table 1. It is the Ca sites. The relative volume fractions of the Bi-2223
observed that the lattice parameter c decreases significantly and Bi-2212 phases were determined from the peak
with increasing Sm content as listed in Table 1. Similar intensities of the same particular reflections using Eqs. (1)
result has been reported by Zandbergen et al. [35]. As and (2). As listed in Table 1, the volume fraction of
pointed out by Zandbergen et al., the behavior of lattice Bi-2223 phase decreased and that of Bi-2212 phase
parameter can be explained by the increase of the oxygen increased with increasing the Sm content.
content in the unit cell by the replacement of Ca2+ by Sm3+ The stoichiometry of Sm substituted annealed
in the structure. It was speculated that the excess of oxygen Bi1.6Pb0.4Sr2Ca2xSmxCu3Oy samples were investigated by
goes into the bismuth oxide layers causing a decrease in EDXRF technique. Figure 3 shows the EDXRF spectrum
lattice parameter c. It is also believed that the decrease of of Sm substituted Bi1.6Pb0.4Sr2Ca2xSmxCu3Oy samples
with x = 0.0005, 0.001 and 0.005. Stoichiometric ratios of
the analysed samples are given in Table 2 [19]. In this
analysis, Bi atomic ratio is given as 1.6. As seen from the
Table 1 Some characteristics of superconducting samples
table, Sm amount in the sample up to x = 0.001 is not
Samples Lattice Volume Tconset Tp (K) (for enough to examine by means of the EDXRF technique.
parameter fraction (%) (K) 111 A/m)
)
c (A
The electrical resistivity as a function of temperature
Bi-2223 Bi-2212 was measured in an external magnetic field up to 0.6 T for
Pure 37.236 94 06 105 97
the Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy samples with
0 < x < 0.005 Sm. Figure 4 shows temperature depen-
x = 0.0005 37.206 92 08 102 88
dence of resistivity for Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy with
x = 0.001 37.197 83 17 99 86
x = 0, 0.0005, 0.0010 and 0.0050 at zero fields. The
x = 0.005 37.190 79 21 97 85
resistivity versus temperature plots show that the resistivity

123
9034
Fig. 3 EDXRF spectra of
Bi1.6Pb0.4Sr2Ca2xSmxCu3Oy
samples with (a) x = 0.0005,
(b) x = 0.001 and (c) x = 0.005
decreases linearly with temperature in the normal state. As
can be seen from the figure the transition temperature,
Tcoffset , decreases from 90 K to 75 K with increasing the Sm
concentration at zero magnetic field. Room temperature
resistivity increases while critical transition temperature
decreases with increasing x from zero to 0.005 Sm. The
behavior of the resistivity is the same as our previous
123
J Mater Sci (2007) 42:90309036
measurements [16] except the values of transition tem-
perature and room temperature resistivity. In the present
work, dc resistivity measurements on the same samples
were made approximately 2 years later than the previous
work. It is observed that the values of Tc decrease while the
values of room temperature resistivity increase (10 times)
in the present study. The transition curves from normal to
J Mater Sci (2007) 42:90309036 9035

Table 2 Calculated
Samples Bi Sr Ca Sm Cu
stoichiometry of Sm substituted
Bi1.6Pb0.4Sr2Ca2xSmxCu3Oy Pure 1.600 0.950 0.320 0.003 1.790
samples from EDXRF results
[19] x = 0.0005 1.600 0.860 0.240 0.000 1.730
x = 0.0010 1.600 0.850 0.240 0.000 1.760
x = 0.0050 1.600 0.820 0.2300 0.005 1.680
x = 0.0100 1.600 0.800 0.220 0.007 1.580
x = 0.1000 1.600 0.820 0.200 0.063 1.650

Fig. 4 Temperature dependence of resistivity for Bi1.6Pb0.4Sr2

(Ca2xSmx)Cu3Oy with x = 0, 0.0005, 0.001 and 0.005 at zero fields

superconducting state indicate two-step transitions and it is

believed that it is related to a structural phase transfor-
mation (from 2223 to 2212) and weak links between the
grains in the samples. The structural phase transformation
is supported by XRD measurements. It is observed that the
broadening of the resistivity transition width increases with
increasing the Sm concentration.
Figure 5 shows temperature dependence of resistivity
for Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy with x = 0.0005 and
x = 0.001 at various magnetic fields (0, 0.3 and 0.6 T). It is
observed that, with increasing dc magnetic field up to
0.6 T, Tcoffset ranges from 90 K to 70 K for the pure sample,
85 K to 54 K for x = 0.0005 Sm, 80 K to 49 K for Fig. 5 Temperature dependence of resistivity for Bi1.6Pb0.4Sr2
x = 0.001 Sm, and 75 K to 35 K for x = 0.005 Sm. (Ca2xSmx)Cu3Oy with (a) x = 0.0005 and (b) x = 0.0010 at various
The transport critical current densities as a function of magnetic fields
the Sm concentrations were measured in liquid nitrogen at
zero magnetic fields as shown in Fig. 6. It observed that the Conclusions
Jctrans decreases with increasing the Sm concentrations. This
result agrees with our aJ (0) calculations. Granularity of the Superconducting samples with nominal composition
samples was the reason for low critical current densities. Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy were prepared the solid-
The relatively low Jctrans for Sm substituted sample with state reaction method. The XRD measurements were done
respect to the undoped sample may be attributed to have at room temperature and the lattice parameters of the
smaller grain size. samples were obtained by indexing all the peaks. The

123
9036
45
40
35
30
J (A/cm )
2
25
c
20
15
10
5 178
0 0.001 0.002 0.003 0.004 0.005 0.006
Sm concentrations
Fig. 6 Transport critical current density versus Sm concentration at
77 K
measurements show that there exist two phases for all the
samples in which Bi-2223 phase decreases while Bi-2212
phase increases with increasing the Sm content. The dc
electrical resistivity and critical current density measure-
ments were carried out on the samples. The Sm Ca
substitution in Bi1.6Pb0.4Sr2(Ca2xSmx)Cu3Oy system
decreased the Jctrans and Tc of the samples. Tconset ranged
from 102 K to 98 K for the pure and the Sm-doped
samples, respectively. The temperature dependence of the
electrical resistivity is linear in the normal state. Room
temperature resistivity increases with increasing x from
zero to 0.005 Sm. The width of intragrain transition
increases with increasing the Sm content and external dc
magnetic field. It is observed that Tc, and Jctrans depend on
both the Sm content of samples and the applied magnetic
field. The calculated aJ (0) from our previous ac suscep-
tibility measurements, we found that the values of aJ (0)
decrease with increasing the Sm concentration.
Acknowledgments This work is supported partly by Research
Foundation of Kirikkale Univerity (Project No: 2005/39), partly by
The Scientific and Technological Council of Turkey (Project No:
104T325) and partly by SRP unit of Abant Izzet Baysal University
(Project No: 2003.03.02.194).
References
1. Sun JJ, Zhao BR, Li L, Xu B, Li JV, Guo SQ, Yin B (1997)
Physica C 291:257
2. Tinkham M (1988) Phys Rev Lett 61:1658
123
J Mater Sci (2007) 42:90309036
3. Palstra TTM, Batlogg B, Schneemeyer LF, Waszczak JV (1988)
Phys Rev Lett 61:1662
4. Anderson PW (1962) Phys Rev Lett 9:309
5. Beasley MR, Labusch R, Webb WW (1969) Phys Rev 181:682
6. Maley MP, Wills JO, Lessure H, McHenry ME (1990) Phys Rev
B 42:2639
7. Zeldov E, Amer NM, Koren G, Gupta A, McElfresh MW,
Kambino RJ (1990) Appl Phys Lett 56:680
8. Fisher MPA (1989) Phys Rev Lett 62:1415
9. Feigelman MV, Geshkenbein VB, Larkin AI, Vinokur VM
(1989) Phys Rev Lett 63:2303
10. Nezir S, Celebi S, Altunbas M (2000) J Alloys Compd 302:235
11. Muller KH, Takashige M, Bednorz JG (1987) Phys Rev Lett
58:1143
12. Clim JR (1991) Physica C 50:153
13. Muller KH, Macfarlane JC, Driver R (1988) Physica C 203:
14. Pan SH, Hudson EW, Lang KM, Eisaki H, Uehida S, Davis JC
(2000) Nature 403:746
15. Eisaki H, Kaneko N, Feng DL, Damascelli A, Mank PK, Shen
KM, Shen ZX, Greven M (2004) Phys Rev B 69:064512
16. Terzioglu C, Yilmazlar M, Ozturk O, Yanmaz E (2005) Physica
C 423:119
17. Yilmazlar M, Cetinkara HA, Nursoy M, Ozturk O, Terzioglu C
(2006) Physica C 442:101
18. Terzioglu C, Yegen D, Yilmazlar M, Gorur O, Akdogan M,
Varilci A (in press) J Mater Sci, doi: 10.1007/s10853-006-0541-6
19. Yegen D, Varilci A, Yilmazlar M, Terzioglu C, Belenli I (in
press) Physica C
20. Prabitha VG, Biju A, Abhilash Kumar RG, Sarun PM, Aloysius
RP, Syamaprasad U (2005) Physica C 433:28
21. Nanda KK, Muralidhar M, Hari Babu V, Pena O, Sergent M,
Beniere F (1993) Physica C 229304:204
22. Awana VPS, Agarwal SK, Ray R, Gupta S, Narlikar AV (1992)
Physica C 191:43
23. Kambe S, Guo YC, Dou SX, Liu HK, Wakahara Y, Maeda H,
Kakimoto K, Yavuz M (1998) Supercond Sci Technol 11:1061
24. Kambe S, Guo YC, Dou SX, Liu HK (1996) Proc R Soc London
96:454
25. Kishore KN, Satyavathi S, Muralidhar M, Babu VH, Pena O,
Sergent M, Beniere F (1994) Phys Status Solidi A 143:101
26. Kishore KN, Satyavathi S, Babu VH, Pena O (1996) Mater Sci
Eng B38:267
27. Kishore KN, Muralidhar M, Babu VH (1993) Physica C 204:299
28. Satyavathi S, \Muralidhar M, Kishore KN, Babu VH (1995) Appl
Superconductivity 3:187
29. Prabitha VG, Biju A, Abhilash Kumar RG, Sarun PM, Aloysius
RP, Syamaprasad U (2005) Physica C 433:28
30. Chiu CW, Meng RL, Gao L, Huang ZJ, Chen F, Xue YY (1993)
Nature 365:323
31. Halim SA, Khawaldeh SA, Mohammed SB, Azhan H (1999)
Mater Chem Phys 61:251
32. Ilonca G, Pop AV, Yang T-R, Jurcut T, Lung C, Stiufiuc G,
Stiufiuc R, Panfilescu IA (2001) Int J Inorg Mater 3:763
33. Babic E, Drobac D, Horvat J, Marohnic Z, Prester M (1989)
J Less Comm Met 151:89
34. Ozturk O, Yegen D, Yilmazlar M, Varilci A, Terzioglu C (2007)
Physica C 451:113
35. Zandbergen HW, Greon WA, Smit A, Tendeloo GV (1990)
Physica C 168:426
J Mater Sci (2007) 42:90379044
DOI 10.1007/s10853-007-1819-z
Study of air oxidation of amorphous Zr65Cu17.5Ni10Al7.5 by X-ray
photoelectron spectroscopy (XPS)
A. Dhawan V. Zaporojtchenko F. Faupel
S. K. Sharma
Received: 2 April 2007 / Accepted: 2 May 2007 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007
Abstract The oxidation of the bulk amorphous alloy
Zr65Cu17.5Ni10Al7.5 in air in its amorphous and the super-
cooled liquid states was studied in the temperature range
573663 K using X-ray photoelectron spectroscopy (XPS).
The oxide film mainly consisted of the oxides of Zr (as
ZrO2) and Al (as Al2O3). No Cu or Ni was found in the oxide
film formed on the amorphous state of the alloy while sig-
nificant Cu (as CuO) was present in the oxide film formed on
the alloy in its supercooled liquid state. The role of the
various alloying elements during oxidation at high temper-
atures in air is discussed in the paper. The XPS data from
oxide film support the previously suggested mechanism for
oxidation of this alloy, i.e. the rate controlling process
during oxidation of the alloy at low temperatures (in the
amorphous state) is the back-diffusion of Ni and Cu, while
the oxidation at high temperatures (in the supercooled liquid
state) is dominated by the inward diffusion of oxygen.
Introduction
Bulk amorphous alloys show a fairly wide temperature
range between glass transition and the crystallization
A. Dhawan
S. K. Sharma (&)
Department of Physics, Malaviya National Institute of
Technology, Jaipur 302 017, India
e-mail: sksh@datainfosys.net
V. Zaporojtchenko
F. Faupel
Faculty of Engineering, University of Kiel, Kaiserstr. 2, 24143
Kiel, Germany
Present Address:
A. Dhawan
Department of Physics, Jagannath Gupta Institute of Engineering
& Technology, Jaipur 302 022, India
temperature and a remarkable resistance to crystallization.
The bulk metallic glasses require cooling rates of about
1100 K/s or less unlike conventional melt-spun metallic
glasses which require very high cooling rates of about
106 K/s for processing [1]. The interest in bulk amorphous
glasses has considerably grown in recent times due to their
ability to offer the scope for carrying out research inves-
tigations both in the amorphous and the supercooled liquid
states of a glassy alloy. Novel multicomponent Zr-based
alloys like ZrTiCuNiBe and ZrCuNiAl form an
important class of bulk metallic glasses with some inter-
esting applications [2].
In contrast to several studies done on the oxidation
behaviour of conventional melt-spun amorphous alloys
[37 references therein], not many studies are available on
oxidation of bulk metallic glasses [816]. In one of the
early investigations on high temperature oxidation of the
bulk amorphous alloy Zr60Al15Ni25 in air it was suggested
that the growth of the oxide is controlled by Ni back-dif-
fusion in the alloy [11]. Contrary to these observations
Triwikantoro et al. [12] investigated the oxidation of sev-
eral Zr-based amorphous alloys (ZrCuNiAl) in air and
reported the presence of Cu and Ni in the oxide scale. In
another investigation by the present authors on oxidation of
the bulk amorphous alloy Zr65Cu17.5Ni10Al7.5 [14], a
mechanism for its oxidation in air was proposed suggesting
that the back diffusion of Ni, and possibly Cu also, is the
rate limiting process during the oxidation of the alloy in the
amorphous or the glassy state, while oxidation process in
the supercooled liquid state is more likely dominated by the
diffusion of oxygen through the oxide layer [14]. In order
to verify this suggested mechanism, it is necessary to
obtain information about the presence of various alloy
constituents, especially Ni and Cu, in the oxide film using
some depth profiling technique. This was a motivating
123
9038
factor for carrying out the present investigations. The
technique of X-ray photoelectron spectroscopy (XPS) was
employed in conjunction with Ar+ ion sputtering for
obtaining the desired information from the oxide film
formed on amorphous and supercooled liquid states of the
bulk alloy Zr65Cu17.5Ni10Al7.5 during its oxidation in air in
the temperature range 573663 K.
Experimental
Specimens of size (10 mm 15 mm) were cut from the
as-cast amorphous ribbon (10 mm wide 30 lm thick) of
the bulk amorphous alloys Zr65Cu17.5Ni10Al7.5. The
samples were first cleaned ultrasonically in acetone and
ethanol and later dried under a jet of pressurised air.
These samples were put in an open-ended quartz tube
which was inserted in a tubular furnace and annealings
were performed at 573 K for 20 h, 603 K for 8 h, 633 K
for 3 h and 663 K for 1 h in dry air environment inside
the open-ended quartz tube. As the oxidation was carried
out ex-situ in a furnace it was not possible to remove the
native oxide layer from the specimens by carrying out ion
beam sputter cleaning or scribing in a vacuum chamber.
The study of the native oxide formed on this alloy has
earlier been reported in another investigation [9]. The
choice of timetemperature combination for annealing
was made on the basis of the results obtained in
our previous investigation [14] and the known TTT
diagram for this alloy [17, 18] so that the alloy
Zr65Cu17.5Ni10Al7.5 remained after oxidation treatment
either in the amorphous (573 and 603 K) or in the
supercooled liquid state (633 and 663 K) and did not
crystallize during annealing. Then the annealed samples
were taken out of the furnace and the shiny side (free
surface during melt- spinning of the alloy) was analysed
by X-ray photoelectron spectroscopy (XPS). The XPS
measurements were done using an electron spectrometer
(VG MK II) equipped with a non-monochromatic Mg Ka
source energy (hm = 1253.6 eV) in a base pressure better
than 1.0 1010 m-bar in the analysis chamber and a
hemispherical electron analyser at a pass energy of 50 eV.
XPS peaks for constituent elements: Zr 3d, Al 2s, Cu 2p
and Ni 2p were recorded along with the XPS peaks for O
1s and C 1s. The sub-surface layers in oxide films were
analysed by performing sequential sputtering using argon
ions (Ar+) of 5.0 keV energy. A typical mean sputtering
rate of 0.71 nm/min was obtained at this energy of argon
ions (Ar+) on the basis of sputter-rate versus depth scale
calibration measurements. The surface morphology of alloy
specimens after oxidation was observed with a scanning
electron microscope (SEM, FEI Quanta) operating at
30 kV.
123
J Mater Sci (2007) 42:90379044
Results and discussion
The oxide films formed on the alloy surface as a result of
high temperature (in the temperature range 573663 K)
oxidation in air were characterized by XPS. The main
features of XPS peaks from specimens annealed at 573 and
603 K (corresponding to the oxidation of specimens in the
amorphous state) were quite similar. Similar observations
were made from XPS peaks obtained from specimens
annealed at 633 and 663 K (corresponding to the oxidation
of specimens in the supercooled liquid state). Therefore, the
figures at two typical temperatures 573 and 633 K which
correspond to oxidation of specimens in the amorphous and
the supercooled liquid states of the alloy, respectively, are
shown here and discussed in the following.
Figure 1 depicts the Zr 3d spectra recorded from the
oxide films formed on specimens of amorphous
Zr65Cu17.5Ni10Al7.5 at temperatures 573 and 633 K for the
as-received specimens and after sputtering with 5 keV Ar+
ions. The XPS peaks for Zr 3d are reported to occur at
178.7 and 182.2 eV for Zr0 (in metallic form) and Zr4+ (in
oxide form), respectively [19]. Therefore, the Zr 3d XPS
peaks in Fig. 1a and b occurring at 181.9 eV and 182.4 eV
for the specimen oxidized at 573 K from the as-received
and the sputtered specimen surface, respectively and also
the corresponding XPS peaks (c) and (d) in the same figure
occurring at 182.0 and 182.5 eV for oxidation at 633 K, are
suggestive of the presence of Zr4+ species on the top sur-
face [9, 1921]. A small shift of about 0.5 eV in Zr 3d peak
towards higher binding energy side after sputtering , i.e.
after removal of the adsorbed hydrocarbon impurity layer
after argon ion sputtering, may be attributed to chemical
Zr 3d peaks at 573 K
d a as-received
b spt. 5 min
c Zr 3d peaks at 633 K
c
d
as-received
spt. 4 min
b
Intensity [a.u.]
a
190.0 187.5 185.0 182.5 180.0 177.5
Binding Energy [eV]
Fig. 1 XPS peaks for Zr 3d from specimens of amorphous
Zr65Cu17.5Ni10Al7.5 after oxidation at 573 K: (a) as-received, (b)
after 5 min sputtering; and after oxidation at 633 K: (c) as-received,
(d) after 4 min sputtering
J Mater Sci (2007) 42:90379044 9039

binding effects in the multicomponent amorphous alloy [9, Ni 2p peaks at 573 K

a as-received
20, 22]. b spt 5 min
a Ni 2p peaks at 633 K
Figures 2 and 3 represent Cu 2p and Ni 2p XPS peaks, c
d
as-received
spt. 4 min

respectively. The oxide film formed at 633 K seems to

contain significant amount of Cu as is evident from Fig. 2c b
and d. On the other hand the poor signals of Cu 2p peaks

recorded from the specimen surface (Fig. 2a, b) oxidized at
573 K indicate hardly any presence of Cu. The Cu 2p peak
in Fig. 2c occurs at 933.6 eV from the as-received amor-
phous specimen Zr65Cu17.5Ni10Al7.5 oxidized at 633 K
indicating the presence of Cu2+, possibly as CuO [19].
After sputtering for 4 min with Ar+ ions, the Cu 2p3/2 peak
(Fig. 2d) shifts to the lower binding energy side, i.e.
932.5 eV indicating the reduction of oxidized copper
(Cu2+) to lower valent states (Cu+, Cu0) [22, 23]. It is
noteworthy here that the possibility of preferential sput-
tering and the reduction of the oxides to lower valent states
during argon ion sputtering of the oxidised alloy surface is
not ruled out [22, 23]. However, the qualitative nature of
conclusions derived from observations made in the present
study is not changed by these effects. It is further inter-
esting to note from Fig. 3 that hardly any Ni is present as
can be inferred from the extremely poor signals in the
observed Ni 2p peaks from specimens oxidized at 573 and
633 K. It is worth mentioning here that features for Cu and
Ni peaks at 603 and 663 K (figures not shown here) were
similar to those at 573 and 633 K, respectively. Figure 4
shows Al 2s XPS peaks recorded from the specimen
oxidized at 573 and 633 K. The Al 2s peak (Fig. 4) was
recorded instead of the more intense Al 2p peak due to the
overlap of the latter with the Cu 3p peak [19]. The Al 2s
peaks for oxidic Al3+ and metallic Al0 are known to occur
at 119.0 and 117.0 eV, respectively [9, 19]. The Al 2s
Intensity [a.u.]
c
d
870 865 860 855 850 845
Binding Energy [eV]
Fig. 3 XPS peaks for Ni 2p from specimens of amorphous
Zr65Cu17.5Ni10Al7.5 oxidised at 573 K: (a) as-received, (b) after
5 min sputtering; and oxidised at 633 K: (c) as-received, (d) after
4 min sputtering
peaks in Fig. 4 occur about at 119.0 eV in these spectra
and thus point to the presence of Al3+, possibly as Al2O3 on
the surface along with ZrO2. However, it is interesting to
note from this figure that the peak of Al is seen after oxi-
dation at 573 K (Fig. 4a, b) while no significant Al peak is
observed after oxidation at 633 K (Fig. 4c, d). A broad
peak seen at about 122.0 eV in Fig. 4d stems from Cu 3s
[19]. These observations can be understood in conjunction
with Cu 2p peaks shown in Fig 2c and d which indicate the
presence of significant amount of Cu in the oxide film
formed at 633 K and thus the relative concentration of Al
in the oxide layer becomes too low to give a significant

Al 2s peaks at 573 K
a as-received
b spt. 5 min
d Cu 2p peaks at 573 K
a as-received Al 2s peaks at 633 K
b spt 5 min c as-received
b d spt. 4 min
Cu 2p peaks at 633 K
c as-received
d spt. 4 min a
Intensity [a.u.]
In t e n s i t y [ a . u . ]

c
d
b
c
a

950 945 940 935 930 925 920 124 122 120 118 116 114 112
Binding Energy [eV] Binding Energy [eV]

Fig. 2 XPS peaks for Cu 2p from specimens of amorphous Fig. 4 XPS peaks for Al 2s from specimens of amorphous
Zr65Cu17.5Ni10Al7.5 oxidised at 573 K: (a) as-received, (b) after Zr65Cu17.5Ni10Al7.5 oxidised at 573 K: (a) as-received, (b) after
5 min sputtering; and oxidised at 633 K: (c) as-received, (d) after 5 min sputtering; and oxidised at 633 K: (c) as-received, (d) after
4 min sputtering 4 min sputtering

123
9040
XPS peak of Al from this sample (Fig. 4c, d). It is further
interesting to note here that no significant peak due to Cu is
seen in the oxide film on the sample oxidised at 573 K
(Fig. 2a, b) while Al is present in the oxide film of this
sample (Fig. 4a, b). Further, the peak features of Cu and Al
peaks after oxidation at 603 and 663 K (figures not shown)
were similar to those observed at 573 and 633 K, respec-
tively. Thus from an analysis of Al (Fig. 4) and Cu (Fig. 2)
peaks it may be concluded that due to strong segregation of
Cu at higher oxidation temperatures (633 and 663 K) the
relative amount of Al in the oxide film formed at these
temperatures becomes too small to yield significant XPS
peaks due to Al. (Fig. 4c, d). This is also evident from
Al/Zr depth profile shown in Fig. 7 later.
The formation of various oxides (ZrO2, CuO, Al2O3) is
supported by the peak positions of the O 1s peak along with
those of the respective alloy constituents (Figs. 1, 2 and 4).
Figure 5 depicts the O 1s XPS peaks from oxidized spec-
imens. The O 1s peaks from the as-received surfaces
appear at 530.0 eV (Fig. 5a, c) for oxidation at 573 and
633 K, respectively and are shifted to higher binding
energy side after sputtering (530.4 eV and 530.6 eV)
(Fig. 2b, d). The O 1s peak for oxide formation (O2 2 ) has
been reported to occur at 530 0.4 eV [9, 19, 20, 24, 25].
The observed asymmetry at high binding energy side in the
O 1s curves in Fig. 5a and c before sputtering most likely
corresponds to the presence of hydroxyl ions on the surface
[9, 19, 20, 25]. This feature disappears after sputtering as is
evident from Fig. 5b and d. Therefore, the O 1s spectra
shown in Fig. 5 correspond to the formation of oxides of
various alloy constituents (ZrO2, Al2O3, CuO) on the top
surface at these high temperatures [9, 19, 20, 24, 25].
An understanding about the formation of various oxides on
the alloy surface during oxidation in air at various
d
O1s peaks at 573 K
a as-received
b spt 5 min
c O1s peaks at 633 K
c as-received
d spt 4 min
b 633 K
Intensity [a.u]
a 603 K
538 536 534 532 530 528 526
Binding Energy [eV]
Fig. 5 XPS peaks for O 1s from specimens of amorphous
Zr65Cu17.5Ni10Al7.5 oxidised at 573 K: (a) as-received, (b) after
5 min sputtering; and oxidised at 633 K: (c) as-received (d) after
4 min sputtering
123
J Mater Sci (2007) 42:90379044
temperatures (in the temperature range 573663 K) can be
obtained from the available values for heat of oxide
formation. The heats of formation for ZrO2, Al2O3, NiO
and CuO are 1101.1, 1117.6, 497.7 and 314.8 KJ/mol
O2, respectively [26]. This suggests that oxides of Zr and
Al in the alloy Zr65Cu17.5Ni10Al7.5 are likely to be formed
first because of their strong affinity for oxygen, followed by
Ni and Cu. However, the back-diffusion or segregation of
alloy constituents during oxidation controls the oxidation
kinetics and thus the formation of oxidation products in the
oxide film.
In the following discussion the role of various alloying
elements during oxidation is analysed. It is interesting to
note here (Figs. 2 and 3) that hardly any Cu or Ni is seen in
the oxide film after oxidation of the alloy at 573 and 603 K
(which correspond to the oxidation of the alloy in the
amorphous state). However, significant amount of Cu (as
CuO) is observed in the oxide film after oxidation at higher
temperatures 633 and 663 K (which correspond to the
oxidation of the alloy in the supercooled liquid state). On
the other hand Al peaks (Fig. 4) show contrasting behav-
iour with Cu peaks. In order to clearly highlight these
observations depth profiles for the concentration ratio of
these elements with respect to Zr in the alloy were obtained
and are shown in Figs. 6 and 7. An interesting feature of
Fig. 6 pertains to the absence of both Cu and Ni in the
oxide film formed at 573 and 603 K which correspond to
the oxidation of the alloy sample in the amorphous state. In
contrast, appreciable amount of Cu is seen in the oxide film
formed at 633 and 663 K which correspond to the forma-
tion of oxide film in the supercooled liquid state of the
alloy. It is interesting to mention here that depth profiles for
Al shown in Fig. 7 are in sharp contrast with those for Cu
(Fig. 6). Al is present on the surface after oxidation at all
temperatures, but hardly any Al is seen deeper in the oxide
1.2
Cu/Zr
573 K
1.0 603 K
663 K
Concentration Ratio
Ni/Zr
0.8
573 K
633 K
0.6 663 K
0.4
0.2
0.0
0 4 8 12 16 20 24 28 32
Sputter Depth [nm]
Fig. 6 Plots of atomic concentration ratio [Cu/Zr] and [Ni/Zr] versus
sputter depth for the alloy Zr65Cu17.5Ni10Al7.5 oxidised at different
oxidation temperatures
J Mater Sci (2007) 42:90379044
Al/Zr
0.12
0.10
Concentration Ratio
0.08
0.06
0.04
0.02
0.00
0 4 8 12 16 20 24 28
Sputter Depth [nm]
Fig. 7 Plots of atomic concentration ratio [Al/Zr] versus sputter
depth for the alloy Zr65Cu17.5Ni10Al7.5 oxidised at different oxidation
temperatures
film formed at higher temperatures, i.e. at 633 and 663 K
(Fig. 7). It is very likely that due to strong segregation of
Cu at these temperatures (Fig. 6) the relative concentarion
of Al in the oxide film containing Zr and Cu becomes too
low to yield a significant XPS peak of Al (Figs. 4 and 7). In
fact, Al seen on the surface just at the beginning of the
oxidation (see Fig. 7) pertains to the initial amount of Al
on the surface, which undergoes oxidation, but its amount
is drastically reduced as a result of strong segregation of Cu
in the oxide film formed at higher temperatures (Fig. 6).
This is supported by the fact that no segregation of Cu is
observed during oxidation at 573 and 603 K (Fig. 6) and
thus significant amount of Al is seen in the oxide film at
these temperatures (Fig. 7). In view of these observations it
is suggested that alloy constituents, especially Cu and Ni
seem to play an important role during oxidation of the alloy
Zr65Cu17.5Ni10Al7.5. The role of Cu and Ni is explained by
the controlling oxidation mechanisms and is discussed in
the following paragraphs.
It was suggested [14] in our previous investigation on
oxidation of this alloy that the oxidation in the amorphous
state is controlled by the back-diffusion of Ni and Cu while
the inward diffusion of oxygen becomes dominant during
oxidation of the supercooled liquid state of the alloy. The
fact that no Ni and Cu are seen during oxidation of the
alloy in its amorphous state clearly supports this mecha-
nism for oxidation of the alloy [14] that the back diffusion
of Ni and Cu seems to be the rate controlling process
during oxidation of the alloy in its amorphous state. In
order to analyse the depth profile in Fig. 6, it would be
interesting to get an estimate of the diffusion lengths of
these elements using the known diffusivity data [2729] in
this and other similar types of amorphous alloys.
Diffusion rates of Ni in amorphous Zr65Cu17.5Ni10Al7.5
are available in the literature [27, 28] while no diffusion
data for Cu diffusion in this alloy have so far been reported.
9041
p
An estimate shows that the diffusion length 4Dt of Ni at
573 K 573 K is about ~34 nm (DNi at 573 K = 3.9 1021 m2/s
603 K
633 K and t = 20 h, being the time corresponding to the oxidation
663 K
of the alloy in its amorphous state). A similar calculation at
603 K yields the value of diffusion length as 56 nm (DNi at
603 K = 2.7 1020 m2/s and t = 8 h). These estimates are
indicative of the fact that due to back diffusion of Ni during
oxidation of the alloy specimen in the amorphous state Ni
has moved by about 3456 nm deeper into the alloy
beyond the oxide-alloy interface. Though no data for dif-
fusion of Cu in this alloy have been reported in the liter-
32 ature, diffusion of both Cu and Ni have been investigated
in another conventional type amorphous alloy Zr50Cu50
[29]. The reported data for diffusion of Ni and Cu in
amorphous Zr50Ni50 [29] point to the fact that diffusion of
Ni is faster than that of Cu by about an order of magnitude
at these temperatures. This is supported by the observed
size effects on diffusion in amorphous alloys [28, 30]
suggesting that smaller atoms diffuse faster than bigger
atoms in amorphous alloys. It thus appears from the above
analysis that both Cu and Ni have moved beyond the oxide-
alloy interface by several tens of nanometers after oxida-
tion at 573 and 603 K. Figures 6 and 7 show depth profiles
up to about 2329 nm only and the oxide-alloy interface is
not yet reached as evident from the plots. Due to practical
difficulties encountered in experimentation, especially in
increasing the sputtering rate, it was not possible to carry
out the sputtering of the entire oxide film and thus to obtain
an estimate of the oxide film thickness. Sun et al. reported
oxide film thickness of about 60 nm formed on amorphous
Zr60Al15Ni25 after its oxidation in air at 603 K for 3 h and
at 623 K for about 5 h [11]. It is thus likely that the oxide
film thickness on the specimens of Zr65Cu17.5Ni10Al7.5 in
the present study might be of this order, i. e. about several
tens of nanometers and the depth profiles in Fig. 6 origi-
nated from a part of the oxide film. However, the above
analysis and interpretation of these depth profiles do pro-
vide valuable information about possible oxidation mech-
anisms for this alloy.
It is, thus, suggested from the above discussion that the
depletion of Ni and Cu in the oxide film, as seen in Fig. 6,
possibly arises due to their back diffusion into the alloy
during oxidation of the alloy and the absence of these
elements (Ni and Cu) in depth profiles after oxidation of
the alloy specimen in the amorphous state can be under-
stood. As the alloy reaches the supercooled liquid state
during oxidation (at temperatures 633 and 663 K) strong
segregation of Cu is noticed in the oxide film (Fig. 6)
which point to a change in the mechanism of oxidation of
the alloy in its supercooled liquid state, i.e. the oxidation
mechanism is dominated by the inward diffusion of oxygen
than the back diffusion of Ni and Cu into the alloy [11, 14].
During oxidation of the alloy the rate controlling process is
123
9042
the back diffusion of Cu and Ni so long as the alloy
remains in its amorphous state (which can be ascertained
from the known TTT diagram of the alloy [17, 18] and it
is most likely that Cu, being a slower diffuser than Ni, is
present closer to the oxide-alloy interface than Ni and thus
gets easily oxidized at 633 and 663 K when the alloy
attains the supercooled liquid state after initial oxidation.
This is evident from the plots shown in Figs. 2 and 6. In the
supercooled liquid state the oxide film formation is domi-
nated by the inward diffusion of oxygen anions and out-
ward diffusion of cations of zirconium and copper. The
diffusion rates of smaller atom (Cu here) are, in fact, about
two orders of magnitude larger at temperatures of interest
in the supercooled liquid state (633 and 663 K) than that in
the amorphous state (573 and 603 K) [2730]. This facil-
itates the outward diffusion of Cu and its consequent oxi-
dation by inwardly moving oxygen atoms. Moreover, the
larger amount of Cu (17.5 at.%) in the alloy than Ni (10
at.%) and the proximity of the former to the oxide-alloy
interface after the initial oxidation till the alloy remained in
the amorphous state are the favourable factors leading to
the oxidation of Cu and the corresponding suppression of
the oxidation of Ni in the supercooled liquid state of the
alloy specimen. This explains the absence of Ni in depth
profiles for oxidation of the alloy in the supercooled liquid
state (Fig. 6). Some traces of Cu and Ni are seen (Figs. 2
and 3) on the as-received surface at 573 and 603 K (cor-
responding to the amorphous state) which possibly arise
due to oxidation of some initial amount of Cu and Ni which
was available just at the beginning of the oxidation process
and these elements get subsequently depleted in the oxide
layer and are possibly enriched near the oxide-alloy inter-
face as the dominant rate controlling process for oxidation
of the alloy in its amorphous state is their back-diffusion
into the alloy. As a result no further presence of Cu or Ni is
seen in the depth profiles of these elements in the oxide
film formed on the alloy surface in its amorphous state (i.e.
at 573 and 603 K).
Therefore, the observed trends of the depth profiles in
Fig. 6 point to the role of Cu and Ni during the oxidation of
the amorphous Zr65Cu17.5Ni10Al7.5 in air and support our
previously proposed mechanism for oxidation of this alloy
[14]. The change in diffusion mechanism at the glass
transition temperature may likely be associated with the
differences in the diffusion behaviour of smaller atoms (Cu
and Ni) in the amorphous and the supercooled liquid states
of the alloy [27, 28]. It is suggested from the above analysis
that during oxidation in the amorphous state both Zr and Al
having the most negative values of the heat of formation
get preferentially oxidised with the rejection of Ni and Cu
from the oxide film. Oxidation mechanisms suggested for
this alloy are indeed consistent with the previously pro-
posed mechanisms for air oxidation of similar type of
123
J Mater Sci (2007) 42:90379044
alloys in the literature [1113]. Sun et al. [11] during
oxidation of amorphous Zr60Ni25Al15 observed the absence
of Ni in the oxide film formed on amorphous Zr65Ni25Al15
after oxidation in air at temperatures below the glass
transition temperature of the alloy. It was proposed [11]
that Ni was rejected from the oxide film and its back dif-
fusion into the alloy was the rate controlling process. The
observations made in the present study also point to the fact
that a similar mechanism appears to hold for oxidation of
the amorphous Zr65Cu17.5Ni10Al7.5 in the amorphous state
as suggested by Sun et al. [11] for the oxidation of the
amorphous Zr60Ni25Al15. It is noteworthy here that the
oxidation of the alloy in amorphous and the supercooled
liquid states was found to obey parabolic rate law with
different activation energies in the respective states [14].
The observations of significant amount of Cu in the oxide
films formed after oxidation in the supercooled liquid state
clearly points to a different oxidation process. It is sug-
gested that during oxidation in the supercooled liquid state
the inward diffusion of oxygen becomes a dominant
and the rate controlling process for the oxidation of the
alloy in the supercooled liquid state. Triwikantoro et al.
[12] and Koester and Trwikantoro [13] have suggested a
similar mechanism for the oxidation of amorphous
Zr69.5Cu12Ni11Al7.5 in air at 633 K for 4 h. It was men-
tioned in their investigation [13] that the alloy had a glass
transition temperature of 635 K and the formation of some
quasicrystals was noticed in the oxidised alloy specimen. It
is very likely that the alloy oxidation did not correspond to
the oxidation in the fully amorphous state and perhaps due
to this reason no rejection of Ni and Cu was observed in
their investigation and based on the results of their study it
was suggested that the inward diffusion of oxygen was the
dominant process for the growth of the oxide film on
Zr69.5Cu12Ni11Al7.5. Therefore, the oxidation studies by
Sun et al. Zr60Ni25Al15 [11] and by Triwikantoro et al. [13]
on Zr69.5Cu12Ni11Al7.5 are available evidences in the lit-
erature showing different oxidation mechanisms for high
temperature air oxidation of a ZrCuNiAl amorphous
alloy depending on whether the alloy is fully amorphous
or not. The results of our oxidation investigations carried
out in air in the temperature range 573663 K on
Zr65Cu17.5Ni10Al7.5 also point to different type of oxidation
processes for the alloy being in the amorphous or the
supercooled liquid state during oxidation treatment.
In order to get some insight into the nature of oxide
films formed on specimens oxidised in the amorphous and
the supercooled liquid states, surface morphology of the
oxide film was examined by SEM. Figures 8 and 9 repre-
sent surface morphologies obtained from specimens oxi-
dised at two typical temperatures 603 K (corresponding to
the amorphous state) and 663 K (corresponding to the
supercooled liquid state). Figure 8 shows a relatively
J Mater Sci (2007) 42:90379044
Fig. 8 Scanning electron micrograph of the oxide film formed on the
amorphous Zr65Cu17.5Ni10Al7.5 after oxidation at 603 K for 3 h (alloy
specimen in the amorphous state)
granular homogeneous distribution of oxidation products
on the sample oxidised in its amorphous state at 603 K. On
the other hand sample oxidised at 663 K (in its supercooled
liquid state) showed a different morphology with some
noticeable cracks in the oxide film. Such cracks have ear-
lier been reported in the oxide films formed during oxi-
dation of ZrNi amorphous alloys in air [7]. In fact, the
presence of cracks in the oxide film formed at higher
temperatures points to the presence of metallic species in
oxidised forms. Cracks may have developed by expansion
due to oxidation of metallic components into oxides,
especially by the oxidation of Cu in the alloy under Zr-rich
top layer. In fact, during the beginning of oxidation an
initial Zr-rich oxide layer is already formed on the alloy
specimen before it is transformed into its supercooled
liquid state after short annealing time which can be
ascertained from the known TTT diagram of the alloy
Fig. 9 Scanning electron micrograph of the oxide film formed on the
amorphous Zr65Cu17.5Ni10Al7.5 after oxidation at 663 K for 3 h (alloy
specimen in the supercooled liquid state)
9043
[17, 18]. The high ratio between the volume of oxide and
metal leads to a compressive stress in the oxide film
resulting in a partial cracking of the oxide film. This is
supported by the fact that the Pilling-Bedworth ratio of
CuO, Cu2O and ZrO2 have values 1.73, 1.67 and 1.57,
respectively [31]. These observations support our propo-
sition that the rate kinetics for oxidation at higher tem-
peratures (in supercooled liquid state) is dominated by
inward diffusion of oxygen anions and outward diffusion of
metal cations, especially of Zr and Cu [14].
Therefore, the observations reported in this study
support the view that different oxidation mechanisms are
operative for oxidation of the alloy Zr65Cu17.5Ni10Al7.5 at
low temperatures (corresponding to the amorphous state
of the alloy) and at high temperatures (corresponding to
the supercooled liquid state of the alloy). It may, how-
ever, be argued that the change in mechanism could
purely be a kinetic effect not being related to the change
of structural state of the alloy. Though this possibility
seems to be rather unlikely as the proposition for change
in the oxidation mechanism was made on the basis of
results obtained by study of oxidation kinetics of the alloy
using the TGA [14] and the XPS analysis of the oxide
films (present study), it may not be totally ruled out on
the basis of the observations made in the present study.
Further careful investigations to study the segregation/
diffusion behaviour of Cu and Ni in the formation of
oxide films at higher temperatures with much smaller
oxidation times (e.g. permissible oxidation time is only
20 s at 633 K from the TTT diagram for the alloy
Zr65Cu17.5Ni10Al7.5 [17, 18] so that the sample remains
fully in the amorphous state in contrast to the oxidation
time of 3 h used in the present study, which corresponds
to the supercooled liquid state of the alloy) are required in
order to unambiguously prove this point.
Conclusions
The results of oxidation of the bulk amorphous alloy
Zr65Cu17.5Ni10Al7.5 carried out in air in the temperature
range 573663 K revealed the following:
(1) The oxide film contained ZrO2 as the major constit-
uent along with some Al2O3 and CuO. Al2O3 was
found in specimens oxidized at 573 and 603 K while
CuO was present in specimens oxidized at 633 and
663 K. The observed depth profiles for Al and Cu
suggest that the relative concentration of Al in the
oxide films formed at 633 and 663 K becomes too
low to yield a significant XPS peak due to Al as a
result of strong segregation of Cu during oxidation at
these temperatures.
123
9044
(2) No Ni and Cu were detected in the oxide film formed
at 573 and 603 K (corresponding to the amorphous
state of the specimen) while appreciable amount of
Cu (as CuO) was found in the oxide film formed at
633 and 663 K (corresponding to the supercooled
liquid state). In the later case no Ni was detected due
to strong segregation of Cu (Cu being present in much
higher concentration in the alloy than Ni) and the
proximity of Cu to the oxidealloy interface.
(3) The above results are consistent with the previously
proposed mechanism for the air oxidation of this alloy
suggesting that the rate controlling process during the
oxidation of the alloy at low temperatures (in the
amorphous state) is the back diffusion of Ni and Cu
while the oxidation at higher temperatures (in the
supercooled liquid state) is dominated by the inward
diffusion of oxygen and the outward diffusion of
zirconium and copper.
Acknowledgements The financial support for this work from the
Board of research in Nuclear Sciences (BRNS), Department of
Atomic Energy (DAE), Government of India under the DAE/BRNS
research Project (Grants No. 99/37/25/BRNS) is gratefully acknowl-
edged. Thanks are due to Dr. K Raetzke for a critical reading of the
manuscript and for many useful suggestions. The help provided by
Ms Jugrita Zekonyte and Mr. Stefan Rehders during XPS measure-
ment is gratefully acknowledged. Thanks are due to Dr. M. Sunda-
raraman, Dr. J. G. Shah and Ms. M. S. Tawade from BARC, Mumbai
for their help in recording SEM micrographs from oxidised alloy
specimens.
References
1. Zhang T, Inoue A, Masumoto T (1991) Mater Trans JIM 32:1005
2. Johnson WL (1996) Curr Opin Solid State Mater Sci 1:383
3. Peker A, Johnson WL (1993) Appl Phys Lett 63:2342
4. Hashimoto K (1983) Amorphous metallic alloys. In: Luborsky FE
(ed) Butterworths, London, p 471
5. Naka M, Hashimoto K, Masumoto T (1976) Corrosion 32:146
6. Dey GK, Savalia RT, Sharma SK, Kulkarni SK (1989) Corros Sci
29:823
123
J Mater Sci (2007) 42:90379044
7. Asami K, Kimura HM, Hashimoto K, Masumoto T (1995) Mat
Trans JIM 36:988
8. Kiene M, Strunskus T, Hasse G, Faupel F (1999) Mater Res Soc
Symp Proc. Materials Research Society, Pittsburgh, PA, USA,
554:167
9. Sharma SK, Strunskus T, Ladebusch H, Faupel F (2001) Mater
Sci Eng A 304306:747
10. Dhawan A, Raetzke K, Faupel F, Sharma SK (2001) Bull Mater
Sci 24:281
11. Sun X, Schneider S, Geyer U, Johnson WL, Nicolet M-A (1996)
J Mater Res 11:2738
12. Triwikantoro, Toma D, Meuris M, Koster U (1999) J Non-Cryst
Solids 250252:719
13. Koester U, Triwikantoro D (2001) Mater Sci Forum 360362:29
14. Dhawan A, Raetzke K, Faupel F, Sharma SK, Phys Stat Sol (a)
199 No. 3 (2003) 431438
15. Tam CY, Shek CH (2005) J Mater Res 20:1396
16. Kai W, Hsieh HH, Nieh TG, Kawamura Y (2002) Intermetallics
10:1265
17. Busch R (Private Communication)
18. Knorr K (1999) In: Selbstdiffusion in Metallischen Massivg-
laesern. Ph. D Thesis, University of Muenster, Muenster,
Germany
19. Wanger CD, Riggs WM, Davies LE, Moulder JF, Muilenberg GE
(1979) Handbook of X-ray photoelectron spectroscopy. Perkin-
Elmer, Eden Prairie, MN
20. Walz B, Oelhafen P, Guentherodt H-J, Baiker A (1989) Appl Surf
Sci 37:337
21. Satoh H, Nakane H, Adachi H (1996) Appl Surf Sci 94/95:247
22. Betz G, Wehner GK (1983) Sputtering by particle bombardment
II, vol 52. In: Behriech R (ed) Springer Verlag, Berlin, p 11
23. Malherbe JB, Hofmann S, Sanz JM (1986) Appl Surf Sci 27:355
24. Brundle CR, Chuang TJ, Wandelt K (1977) Surf Sci 68:459
25. Wandelt K (1982) Surf Sci Rep 2:75
26. CRC handbook of chemistry and physics (199596) 76th edn.
Lide DR (ed) CRC Press, Boca Raton, pp 3447
27. Knorr K, Macht M-P, Mehrer H (1999) Bulk metallic glasses. In:
Johnson WL, Liu CT, Inoue A (ed) MRS symposia proceedings
No. 554. Materials Research Society, Pittsburgh, p 269
28. Faupel F, Frank W, Macht M-P, Mehrer H, Naundorf V, Raetzke
K, Schober HR, Sharma SK, Teichler H (2003) Revi Modern
Phys 75:237
29. Hahn H, Averback RS, Shyu H-M (1988) J Less-Common Met
140:345
30. Sharma SK, Banerjee S, Kuldeep, Jain AK (1989) J Mater Res
4:603
31. Bradford SA (1990) In: ASM handbook, vol 13. ASM Interna-
tional, Materials Park, OH, p 64
J Mater Sci (2007) 42:90459056
DOI 10.1007/s10853-007-1824-2
Structureproperty relationship of shape memory polyurethane
cross-linked by a polyethyleneglycol spacer between polyurethane
chains
Byoung Chul Chun Tae Keun Cho Mi Hwa Chong
Yong-Chan Chung
Received: 25 February 2007 / Accepted: 4 May 2007 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007
Abstract Shape memory polyurethane (SMPU) cross-
linked by a polyethyleneglycol (PEG) spacer at its side
was compared with a linear SMPU and the one randomly
cross-linked by glycerol. The SMPU was composed of
4,4-methylenebis(phenylisocyanate) (MDI), polytetram-
ethyleneglycol-2000 (PTMG-2000), 1,4-butanediol (BD),
glycerol, and PEG-200 as a spacer. PEG-200 connected
the glycerol hydroxyl groups of two polyurethane chains
via the connecting agent, MDI. Tm of the soft segment
increased with the increase of cross-linking content.
Cross-link density measured by the swelling experiment
increased in proportion to the glycerol content. A sur-
prising increase in maximum stress compared to the linear
SMPU was attained without any sacrifice in strain.
Remarkably, the stressstrain curve revealed that the PEG
cross-linked SMPU exhibited a similar behavior and
superior tensile mechanical properties to natural rubber.
Storage modulus also increased and loss tangent became
broad in distribution over the temperature range after PEG
cross-linking. Shape recovery rate went up to 96.8%, and
shape recovery speed was three times faster than that of
linear SMPU.
B. C. Chun (&)
T. K. Cho
M. H. Chong
Department of Polymer Engineering, The University of Suwon,
Hwasung-shi, Kyonggi-do 445-743, Korea
e-mail: bcchun@suwon.ac.kr
B. C. Chun
T. K. Cho
M. H. Chong
Y.-C. Chung
Intelligent Textile System Research Center, Seoul, Korea
Y.-C. Chung
Department of Chemistry, The University of Suwon,
Hwasung-shi, Kyonggi-do 445-743, Korea
Introduction
Shape memory polymer has drawn widespread attention
because of the features like low density (10001300 kg/m3),
high shape recovery, low manufacturing cost, and ease
of processing [14]. The first shape memory polymer,
polynorborene, was introduced by Nippon Zeon company
in 1984, and various forms of shape memory polymers such
as poly(isoprene-butadiene-styrene), polyurethane, and
polystyrene series have followed [58]. Among these can-
didates, shape memory polyurethane (SMPU), first devel-
oped by Mitsubishi Chemical in 1988, was most favored
due to its excellent properties which include a wide shape
recovery temperature range (30 to 70 C), a high shape
recovery rate, easy processing conditions, and a possible
biocompatibility [7, 911]. Like other shape memory
polymers, SMPU also has a phase-separated structure where
hard and soft domains are formed due to the difference in
intermolecular attraction of hard and soft segments.
Hydrogen bonding together with dipole-dipole interaction
binds hard segments to form a hard domain which plays an
important role in recovering the original shape above Tm.
The flexible soft segment absorbs any external stress or
impact and keeps the polymer resilient at low temperature.
Because the previous linear SMPUs are not able to
withstand the cyclic distortion and recovery test, cross-
linking of polyurethane has been tried by us and other
researchers [4, 12, 13]. Although the addition of a cross-
linking agent into the polyurethane chain improved the
mechanical and shape memory properties, retention of
high shape recovery rate after repetitive test cycles and
attainment of higher mechanical properties remain to be
resolved. Therefore, a different cross-linking method
needs to be devised to improve both mechanical and
shape memory properties. In our design, some of the chain
123
9046 J Mater Sci (2007) 42:90459056

extenders, 1,4-butanediol, are replaced with glycerol that is Table 1 Composition of linear and cross-linked SMPU
used as a connecting spot with other polyurethane chains. Sample Reactant, mol ratio wt.% of
The target polyurethane has a network-structure compared code glycerol
to the randomly cross-linked one, and a substantial MDI PTMG BD Glycerol PEG-200
improvement in mechanical and shape memory properties L-0 5.0 2.0 3.0
is achieved. In this investigation, PEG-200 is selected as the C1-5 5.0 2.0 2.775 0.15 5
spacer because of its flexibility and short chain length, and C1-10 5.0 2.0 2.55 0.3 10
the mechanical and shape memory properties are compared C1-15 5.0 2.0 2.325 0.45 15
with linear or randomly cross-linked ones. C2-5 5.0 2.0 2.85 0.15 0.15 5
C2-10 5.0 2.0 2.7 0.3 0.3 10
C2-15 5.0 2.0 2.55 0.45 0.45 15
Experimental
C2-20 5.0 2.0 2.4 0.6 0.6 20
C2-25 5.0 2.0 2.25 0.75 0.75 25
Materials

4,4-methylenebis(phenylisocyanate) (MDI, Tokyo Kasei),

poly(tetramethyleneglycol) (PTMG, Aldrich Chemical, by the solvent-casting method: DMF in the PU solution
Mw = 2 kg/mol), and polyethyleneglycol (PEG-200, was slowly evaporated at 70 C for 216,000 s to yield a
Mw = 0.2 kg/mol, Duksan Chemical) were dried under 7 1039 103 m thick PU sheet and the specimen was
high vacuum (0.1 mmHg) for 432,000 s before use. prepared from that sheet according to ASTM D638.
1,4-butanediol (BD, Duksan Chemical) and glycerol
Number of hydroxyl groups
(Duksan Chemical) were also dried under high vacuum.
Dimethylformamide and phthalic anhydride were also ob-
The number of free hydroxyl group in the PU chain after
tained from Duksan Chemical.
glycerol insertion was found by following the titration
method [14]. Phthalic anhydride solution was prepared by
Polymer synthesis
dissolving 0.014 kg of phthalic anhydride in 9.15 105 m3
of DMF. Phthalic anhydride solution (5 106 m3) was
In a 5 104 m3 four neck cylindrical flask, a mixture of
added to 2 106 m3 of the PU reaction mixture, and the
MDI and PTMG was allowed to react by mechanical stir-
mixture was sonicated for 600 s at 45 C, incubated at 98 C
ring under nitrogen purge at 50 C for 7200 s to prepare
for 900 s, and allowed to cool to room temperature. After the
the prepolymer. After the addition of BD and glycerol
addition of 1 105 m3 of DMF and 2 106 m3 of distilled
dissolved in 3 105 m3 of dry DMF to the prepolymer,
water, the mixture was titrated with 0.1 N NaOH using a
the reactants were stirred under the same conditions for
drop of 1% phenolphthalein indicator. Blank titration of the
another 3600 s. The final PU product was dried at 60 C
phthalic anhydride solution was carried out without the
for 144,000 s to remove the DMF. The amount of each
reaction mixture by the same procedure. An average of five
reactant is shown in Table 1, where the hard segment
titrations was used for the hydroxyl number calculation. The
content is fixed at 30 wt%. Synthesis of the C1 series
exact mass of PU in a sampled reaction mixture was found by
finished at this step, but the C2 series continued for PEG
weighing the residue after complete evaporation of DMF in a
cross-linking. In the C1 series, glycerol can randomly
vacuum oven for 108,000 s. The hydroxyl number was
cross-link PU chains, but PEG-200 in the C2 series cross-
calculated from the following equations.
links PU chains through glycerol hydroxyl groups. The
hydroxyl number was measured to determine the molarity 0:1B  A=PU mass Hydroxyl number (mol/kg) 1
of free hydroxyl group on the prepolymer by a titration
method shown in the following section. An exact amount (A = titration volume of PU, and B = blank volume)
of MDI based on the hydroxyl number was added to the
prepolymer solution and the reaction mixture was stirred Hydroxyl number  Total mass = Total hydroxyl group
under the same conditions for 3600 s. Subsequently, PEG- 2
200 equivalent to half mole of MDI was added, and the
mixture was stirred under the same conditions for 3600 s.
The C2 series product was dried at 60 C for 144,000 s to Characterization
remove DMF. As a control, linear PU without glycerol was
prepared for comparison. The SMPU structures are shown A FT-IR spectrometer (JASCO 300E) equipped with ATR
in Fig. 1. A specimen was prepared for mechanical testing was used to take an IR spectrum of the PU sample with

123
J Mater Sci (2007) 42:90459056 9047

Fig. 1 Structures of (a) L-0, (a)

(b) C1 series, and (c) C2 series
O O O O
CNH CH2 NHCO CH2CH2CH2CH2O CNH CH2 NHCOCH2CH2CH2CH2CH2O
n

(b)
O O O O
CNH CH2 NHCO CH2CH2CH2CH2O CNH CH2 NHCOCH2CHCH2O
n
O

PU chain

(c)
O O O O
CNH CH2 NHCO CH2CH2CH2CH2O CNH CH2 NHCOCH2CHCH2O
n
O

MDI + PEG-200

O O O O O
CNH CH2 NHCO CH2CH2CH2CH2O CNH CH2 NHCOCH2CHCH2O
n

400 m1 resolution, 25 scans, and a scan speed of 2
103 m/s. X-ray diffraction spectra were scanned under the
conditions of 2h = 540 and 0.016/second by a wide-
angle X-ray diffractormeter (WAXD, Rigaku Rint 2000,
CuKa, 40,000 V, and 0.03 A). A differential scanning
calorimeter (DSC-2010, TA instrument) was used for the
heating and cooling scan at a rate of 0.16 C/s between
100 C and 250 C. The tensile mechanical properties
were measured by a Universal Testing Machine (LR50K,
Lloyd instrument, UK) according to ASTM D638 using
0.025 m gauge length, 1.6 105 m/s crosshead speed, and
a 2500 N load cell. The dynamic mechanical properties
were measured by a dynamic mechanical analyzer (DMA-
2980, TA instrument). The storage modulus and loss tan-
gent (tan d) were scanned between 80 C and 65 C at a
heating rate of 0.05 C/s, and 1 Hz. The cross-link density
was calculated according to the FloryRehner equation for
a specimen (0.025 0.025 0.001 m) swelled in toluene,
a process for which further explanation will appear in the
results section.
Shape memory test
The UTM was also used for the cyclic shape memory test.
Lo was the original length of the specimen. The shape
retention rate was calculated by Eq. (3). L1 was measured
after the specimen had been drawn 100% at 20 C above
the Tm of the soft segment for 300 s, and then allowed to
shrink at 20 C below the Tm for 1800 s. The shape
recovery rate was calculated by Eq. (4). L2 was measured
after the stretched specimen had been incubated at 20 C
above the Tm for 600 s and cooled at 20 C below the Tm.
The shape memory test was repeated three times.
Shape retention rate L1  Lo  100=Lo % 3
Shape recovery rate 2Lo  L2  100=Lo % 4
Results and discussion
Synthesis
Previously, the cross-linking method was dependent on the
three-way linker that connected PU chains in a random
manner. Connection of the nearby PU chains by a flexible
cross-linker and control of the cross-link density may
improve the mechanical and shape memory properties due
to the resultant network structure. In this investigation,
some of BD, a chain-extender, is replaced with glycerol
and the extra hydroxyl group of glycerol is used as the
cross-linking point. The extra hydroxyl group is activated
by MDI, and PEG-200 reacts with the MDI moiety on each
chain. Because a PU chain is connected side by side at the
various sites along the PU chain, the overall strength can be

123
9048 J Mater Sci (2007) 42:90459056

increased in comparison to that of the randomly cross- Crosslink density

linked one. The free hydroxyl groups on a PU chain are
found by titration method and used for the calculation of Cross-link density is calculated from the polymer swelling
the required quantities of MDI and PEG-200 for cross- experiment. The interaction parameter (v) between the
linking. solvent and the polymer can be found from the following
The free hydroxyl groups attack phthalic anhydride to Eq. (5) [15].
form a polymer-phthalic acid intermediate and the excess
phthalic anhydride is hydrolyzed with water to give v d1  d2 2 V1 =RT 5
phthalic acid (Fig. 2). The carboxyl groups from phthalic
acid are titrated with the NaOH solution and the difference d1 and d2 = solubility parameter of solvent and polymer;
between the blank and the PU titration is used for the V1 = molar volume of solvent; R = gas constant;
calculation of free hydroxyl groups [14]. The hydrolysis of T = absolute temperature.
phthalic anhydride finishes in 120 s, judging from the fact Solubility parameters of toluene (d1) and PU (d2) are
that the volume of NaOH was almost constant after 120 s 18.2 and 20.5 (MPa)1/2, respectively. The degree of cross-
(Fig. 3(a)). An aliquot of the reaction mixture after glyc- linking is calculated from the FloryRehner equation (6).
erol addition is taken from the reactor at 600 s interval, and
1=3
the amount of free hydroxyl groups remaining after the ln1  v2 v2 vv22  V1 nv2  1=2v2  6
glycerol addition is followed by the above titration method.
As more glycerol reacts, the number of free hydroxyl v2 = volume fraction of polymer in the swollen mass;
groups decreases. Also, the number of carboxyl groups v = interaction parameter; n = cross-link density.
from the phthalic anhydride hydrolysis during titration The cross-link density of the C1 and C2 series is sum-
increases due to the decrease in free hydroxyl groups. The marized in Table 2. Cross-link density increases with
glycerol reaction is over in 3600 s, as determined from the increasing glycerol contents, and the C2 series have
result that NaOH volume reaches its maximum in approximately two times higher cross-link density than the
Fig. 3(b). The free hydroxyl group of the reaction mixture C1 series at the same glycerol content. Linear PU does not
is calculated from the titration result, and the equivalent have the cross-link density. The organized cross-linking
amount of MDI is added to activate the free hydroxyl between PU chains in the C2 series has raised the cross-
groups. Lastly, PEG-200 is used to link the MDI-activated linking density compared to the random cross-linking of
PU chains. the C1 series.

IR analysis
O O

OH + O P O IR peak of the hydrogen-bonded urethane carbonyl group

P
HO appears at 16991706 cm1, a little lower than that of the
O
O
free urethane carbonyl group (17311733 cm1) [16]. The
stretching vibration of the carbonyl group is also affected
Fig. 2 Reaction between prepolymer and phthalic anhydride by the dipole-dipole interactions between PU chains [13].

(a) 131 (b) 130

NaOH Volume x 106 (m3)

NaOH Volume x 106 (m3)

128
130

126

129
124

128 122
0 100 200 300 400 500 600 700 0 1000 2000 3000 4000
Time (s) Time (s)

Fig. 3 (a) Time course of the phthalic anhydride hydrolysis. (b) Titration of the periodically sampled reaction mixture after glycerol addition

123
J Mater Sci (2007) 42:90459056 9049

Table 2 Cross-link density of SMPU by swelling experiment The ratio, ebonded/efree, is close to 1.0 [18]. The degree of
a 3 b phase separation (DPS) and the degree of phase mixing
Sample code q (kg/m ) Q vc2 3 d
10 n
(DPM) are calculated from the following equations using
L-0 932 the R value of Eq. (7) [18]. The combined result is
C1-5 999 2.620 0.276 0.547 summarized in Table 3.
C1-10 895 2.516 0.284 0.581
C1-15 975 2.364 0.297 0.632 Cbonded R
DPS 8
C2-5 893 2.072 0.326 0.784 Cbonded Cfree R 1
C2-10 934 1.832 0.353 0.954
C2-15 968 1.637 0.379 1.139 DPM 1  DPS 9
C2-20 987 1.380 0.420 1.477
As the glycerol content increases, R decreases from 1.15
C2-25 965 1.176 0.460 1.866
(L-0) to 1.02 (C1-15) and 0.97 (C2-25). In the C1 series, R
a
Density; bDegree of swelling; cVolume fraction of polymer in decreases from 1.08 (C1-5) to 1.02 (C1-15). In the C2
swollen mass; dCross-link density
series, R decreases from 1.13 (C2-5) to 0.97 (C2-25).
Hydrogen bonding becomes weaker as the glycerol content
The peak around 1730 cm1 increases slightly with the increases. The change in phase separation is more obvious if
increase of glycerol content for both C1 and C2 series, the DPS is compared. The DPS decreases from 53.4% (L-0)
while the peak around 1700 cm1 decreases (Fig. 4). Based to 50.5% (C1-15) and 49.4% (C2-25). When the DPS at the
on IR results, the shift from bonded carbonyl to free car- same glycerol content is compared, no difference is found:
bonyl goes on as the glycerol content increases for both C1 52.1% (C1-5) vs. 52.9% (C2-5), and 50.5% (C1-15) vs.
and C2 series. Therefore, it seems that glycerol cross- 50.9% (C2-15). Now, it is clear from the IR experiment that
linking reduces the hydrogen bonding, and disrupts hard phase separation for both C1 and C2 series is decreased by
segment interaction. glycerol cross-linking and the difference in phase separa-
The peak shift can be used in quantifying the phase tion between them at the same glycerol content is minimal.
separation of hard and soft segments. Phase separation can
be also calculated from the IR spectra by comparing the XRD analysis
peak intensity ratio of carbonyl groups at 1703 cm1 and
1733 cm1 [17]. The parameter indicating phase separa- The X-ray deflection peak of PU, irrespective of glycerol
tion, hydrogen bonding index (R), is calculated by Eq. (7). content, is observed at 2Q = 19.5 [7]. Linear PU (L-0)
Cbonded ebonded A1703 shows a very high peak intensity compared to cross-linked
R 7 PUs, and the peak intensity decreases with any increase in
Cfree efree A1733
glycerol content (Fig. 5). In line with the IR data, hydrogen
In Eq. (7), A is absorbance, C is concentration, ebonded is bonding between hard segments is unaffected by cross-
the extinction coefficient of the peak at 1703 cm1, and linking in L-0, and the ordered hard segment alignment
efree is the extinction coefficient of the peak at 1733 cm1. increases the peak intensity. However, glycerol cross-linking

(a)
(b)
Transmittance (%)

Transmittance (%)

C2-5
L-0 C2-10
C1-5 C2-15
C1-10 C2-20
C1-15 C2-25

1800 1750 1700 1650 1600 1800 1750 1700 1650 1600
Wave number (cm-1) Wave number (cm-1)

Fig. 4 Infrared spectra of (a) L-0 and C1 series, and (b) C2 series

123
9050 J Mater Sci (2007) 42:90459056

Table 3 Comparison of thermal properties of SMPU Table 4 Degree of phase separation (DPS) and degree of phase
a mixing (DPM) of SMPU
Sample code R (A1703/A1733) DPS
a
Sample code Tg (C) Tm (C) DHm (J/kg) DHc (J/kg)
L-0 1.15 53.4
C1-5 1.08 52.1 L-0 52.47 10.00 5050 3550
C1-10 1.04 51.1 C1-5 50.72 11.35 1190 520
C1-15 1.02 50.5 C1-10 50.68 12.35 1180 110
C2-5 1.13 52.9 C1-15 49.96 11.57 1060 500
C2-10 1.07 51.6 C2-5 46.69 18.08 4040 420
C2-15 1.04 50.9 C2-10 42.99 20.35 3110 370
C2-20 1.01 50.4 C2-15 41.99 18.65 860 330
C2-25 0.97 49.4 C2-20 40.04 18.85 840 420
a C2-25 38.70 20.21 280 180
A1703: absorbance at 1703 cm1, and A1733: absorbance at 1733 cm1
a
DHc: heat of crystallization

disrupts hydrogen bonding and hard segment alignment,
and reduces the peak intensity in proportion to the glycerol
content. In the extreme case of C2-25, the peak intensity is
too reduced to be considered as a peak. In line with the
DPS data, it is also very difficult to differentiate C1 and C2
series simply by XRD spectra.
Thermal analysis
The soft segment Tg is detected by the more sensitive DMA
analysis, because it is very hard to detect Tg by DSC due to
the very low temperature range and weak signal. However,
it is not impossible to detect Tm of the soft segment by DSC.
The combined thermal analysis data are summarized in
Table 4. Tg is very low for L-0 (52.47 C) and increases
with cross-linking for both C1 and C2 series. C2 series
shows a higher Tg for a given glycerol content: 50.72 C
(C1-5) vs. 46.69 C (C2-5) and 49.96 C (C1-15) vs.
41.99 C (C2-15). Tg increases up to 38.70 C in
the case of C2-25. Tm of L-0 (10.00 C) is quite high
compared to Tg and also increases with cross-linking for
both C1 and C2 series. C2 series also exhibit higher Tm at
the same glycerol content: 11.35 C (C1-5) vs. 18.08 C
(C2-5) and 11.57 C (C1-15) vs. 18.6 C (C2-15). Similar
to the IR and XRD results, the network structure of C2
series is more influential in raising both Tg and Tm than the
randomly cross-linked C1 series. It is very interesting that
Tm did not increase after cross-linking although the glyc-
erol content is raised for both C1 and C2 series. It seems
that the cross-linking density of hard segment does not
affect the soft segment melting because soft segment gets
enough energy for melting at Tm, a relatively high tem-
perature compared to Tg. In DSC heating scans, L-0 shows
a large endothermic peak around Tm, and the endothermic
peak shrinks as the glycerol content increases for both C1
and C2 series (Fig. 6). The peak area can be quantitatively
compared by heat of melting (D Hm) in Table 4. In
accordance with the peak shrinkage with cross-linking in
Fig. 6, DHm decreases significantly from 5050 J/kg to
1060 J/kg (C1-15) and 280 J/kg (C2-25). The restricted
soft segment melting by glycerol cross-linking is respon-
sible for the decrease of DHm. The heat of crystallization

7000
(a) (b) 7000

6000
L-0 6000 C2-5
C1 -5
C2-10
5000 C1 -10 5000
C1 -15 C2-15
C2-20
Intensity

Intensity

4000 4000 C2-25

3000 3000

2000 2000

1000 1000

0 0
10 15 20 25 30 35 10 15 20 25 30 35
2 (deg ree) 2 (degree)

Fig. 5 X-ray diffraction spectra of (a) L-0 and C1 series, and (b) C2 series

123
J Mater Sci (2007) 42:90459056 9051

(a) (b)

Endothermic
E n d othermic

L-0 C2-5
C2-10
C1-5
C2-15
C1-10
C2-20
C1-15
C2-25

-50 0 50 100 150 200 250 -50 0 50 100 150 200

Temperature (C) Temperature (C)

Fig. 6 DSC heating scans of (a) L-0 and C1 series, and (b) C2 series

Fig. 7 Comparison of C1 and (a) 30 (c)15

C2 series in the points of (a)

Tensile modulus (MPa)

Maximum stress (MPa)

C1 series C1 series
maximum stress, (b) strain at 25
C2 series 12 C2 series
break, (c) tensile modulus, and 20
(d) yield stress 9
15
6
10
3
5

0 0
0 10 20 30 0 5 10 15 20 25 30
Glycerol contents (wt%) Glycerol contents (wt%)
(b) 1500 (d) 5
Strain at break (% )

Yield stress (MPa)

1200 4

900 3

600 2

300 C1 series 1 C1 series

C2 series C2 series
0 0
0 5 10 15 20 25 30 0 10 20 30
Glycerol contents (wt%) Glycerol contents (wt%)

(DHc) obtained from the cooling scan also supports that
cross-linking restricts soft segment movement (Table 4).
Tensile mechanical properties
Tensile mechanical properties averaged from five tests are
compared in Fig. 7. Maximum stress of cross-linked PU
(C2-10) steeply increases up to 800% compared to L-0,
although further increase of cross-linked content reduces
the maximum stress. Maximum stress of the C2 series is
300% higher than that of C1 series at 10 wt% cross-
linking. The strain at break is similar for both C1 and C2
series, and increases with the increase of cross-linking
content. Tensile modulus is maximized at 5 wt% glycerol
for both C1 and C2 series, and decreases with further
increase of glycerol content. Yield stress of C2 series is
about 1.5 times higher than that of C1 series at 5 wt%
glycerol, and decreases at higher wt% of glycerol. It is
remarkable that changing the cross-linking method makes
a big difference in maximum stress while maintaining a
similar modulus, and the strain at break goes up to 1200%
even after cross-linking. The tensile mechanical behavior
is also compared in a stressstrain curve (Fig. 8). C1 and
C2 series show similar slopes up to 600% strain, but a
sudden increase in slope is observed for C2 series, espe-
cially C2-5, C2-10, and C2-15. Because the behavior of
C2 series is similar to that of natural rubber, the four
polymers, L-0, C1-10, C2-10, and natural rubber (NR),
are plotted in Fig. 8(c). Notably, C2-10 is different from
other PUs and is close to natural rubber in overall shape

123
9052 J Mater Sci (2007) 42:90459056

(a) 25
L-0
(b) 25
C2-5
C1-5 C2-10
C1-10 C2-15
20 C1-15
20
C2-20
C2-25

Stress (MPa)
Stress (MPa)

15 15

10 10

5 5

0 0
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
Strain (%) Strain (%)

(c) 25
L-0
C1-10
20 C2-10
NR
Stress (MPa)

15

10

0
0 200 400 600 800 1000 1200 1400
Strain (% )

Fig. 8 Stressstrain curve of (a) C1 series, (b) C2 series, and (c) SMPU and natural rubber (NR)

except the higher stress and strain. The steep rise of stress
in natural rubber originates from the strain-induced crys-
tallization, and the same reasoning can be applied to the
explanation of C2 series behavior. C2 series polymer
aligns along the chain length because each polymer chain
is connected by a PEG-200 spacer at its side and becomes
stronger when stretched due to the strain-induced
crystallization. But C1 series, connected in a random
manner, is unable to form the aligned structure and shows
no sign of strain-induced crystallization. Previous exper-
iments on SMPU have demonstrated that cross-linking of
hard segment can increase maximum stress but strain is
sacrificed instead. But the adoption of a flexible spacer
and cross-linking at its side in C2 series elevates the

(b) 2500
(a) 2000 L-0 C2 -5
C1-5 C2 -1 0
2000
C1-10 C2 -1 5
1500
C2 -2 0
C1-15
E' (MPa)

1500 C2 -2 5
E' (MPa)

1000
1000

500 500

0
0 -80 -60 -40 -20 0 20
-80 -60 -40 -20 0 20
Temperature (C)
Temperature (C)

Fig. 9 Storage modulus vs. temperature profile of (a) L-0 and C1 series, and (b) C2 series

123
J Mater Sci (2007) 42:90459056 9053

(a) 0.6 (b) 0.6

L-0 C2-5
0.5 C1 -5 C2-10
0.5
C1 -10 C2-15
0.4 C1 -15 C2-20
0.4
C2-25
Tan

T an
0.3 0.3

0.2 0.2

0.1 0.1

0 0
-80 -60 -40 -20 0 20 40 60 -80 -30 20 70
Temperature (C) Temperature (C)

Fig. 10 Loss tangents vs. temperature profile of (a) L-0 and C1 series, and (b) C2 series

polymer, than C1 series, which is due to the high cross-link

maximum stress without any sacrifice of strain. As noted
in the above analyses and mechanical tests, C2 series is
different and special because of the flexible spacer and
sideway cross-linking.
Dynamic mechanical properties
Storage modulus (E) and loss tangent (tan d) results are
shown in Figs. 9 and 10. It is already known that the cross-
linked structure generally shows a higher storage modulus
than the linear one. C2 series has a higher storage modulus
than C1 series at the same glycerol content, which is due to
the network structure of C2 series. The sudden decrease in
storage modulus starting at 70 C and ending at 20 C is
related to the phase transition of the soft segment. C2 series
is lower in loss tangent, a measure of the resilience of
density of C2 series. Tg of the cross-linked SMPU in
Table 4 is detected from the peak loss tangent value. As
shown in shape memory effect section, the higher storage
modulus of C2 series than that of C1 series results in the
better shape recovery of C2 series over C1 series, but shape
retention of C2 series compared to C1 series is sacrificed
due to the higher storage modulus.
Shape memory effect
Cyclic shape memory test is used to compare the shape
recovery rate and shape retention rate for the Tm 20C
range (Figs. 11, 12). Shape memory test is generally car-
ried out around Tg of soft segment for most of the ther-
mally-induced shape memory polymers. In some cases, if
Tg is too low, Tm of soft segment is used instead of Tg;

Fig. 11 Cyclic shape memory (a) 3 (c) 4

Cycle 1 Cycle 1
test of (a) L-0, (b) C1-5, Cycle 2 Cycle 2
Cycle 3 Cycle 3
(c) C1-10, and (d) C1-15 Cycle 4 3 Cycle 4
Strees (MPa)

Strees (MPa)

2
2

1
1

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Strain (%) Strain (%)
(b) 4 (d) 4
Cycle 1 Cycle 1
Cycle 2 Cycle 2
3 Cycle 3 3 Cycle 3
Strees (MPa)

Strees (MPa)

Cycle 4 Cycle 4

2 2

1 1

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Strain (%) Strain (%)

123
9054 J Mater Sci (2007) 42:90459056

Fig. 12 Cyclic shape memory (a) 4 (b) 4

Cycle 1 Cycle 1
test of (a) C2-5, (b) C2-10, (c) Cycle 2 Cycle 2
C2-15, (d) C2-20, and (e) C2-25 3 Cycle 3
3 Cycle 3
Cycle 4 Cycle 4

Stress (MPa)

Stress (MPa)
2 2

1 1

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Strain (%) Strain (%)

(c) 4 Cycle 1 (d) 4

Cycle 1
Cycle 2 Cycle 2
Cycle 3 Cycle 3
3 Cycle 4
3 Cycle 4

Stress (MPa)

Stress (MPa)
2 2

1 1

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Strain (%) Strain (%)
(e) 3 Cycle 1
Cycle 2
Cycle 3
Stress (MPa)

Cycle 4
2

0
0 20 40 60 80 100
Strain (%)

such type of shape memory polymer is called Tm-type- analytical and mechanical results, the shape memory test
shape memory polymer to differentiate it from Tg-type one. also demonstrates that the alignment of polymer chains and
This SMPU belongs to Tm-type, because Tg is too low and the cross-linking at the side with a flexible spacer play a
it is hard to control the temperature around Tg [2, 12, 13]. decisive role in improving the overall properties compared
As compared in Table 5, shape recovery process is over in to the randomly cross-linked C1 series and linear L-0.
900 s at Tm + 20 C. Shape recovery rate increases from Therefore, the finding of C2 series is very encouraging and
83.7% for L-0 to 89.3% for C1-5 and 96.8% for C2-10. C2 argues the importance of polymer chain alignment and
series exhibits a higher shape recovery rate than C1 series cross-linking method. More variations of SMPU based on
at the same glycerol content: 89.3% for C1-5 vs. 94.8% for C2 series can be produced in near future.
C2-5, and 85.1% for C1-15 vs. 94.0% for C2-15 (Table 5).
The shape recovery rate of C2-10, the best one of C2 series, Table 5 Comparison of shape recovery speed of SMPU
does not fall below 93% even after cycle 4. Although many
Sample code Shape recovery rate (%)
shape memory polymers, either polyester or polyurethane,
have been tested by us, shape recovery rate is usually less 300 s 600 s Maximum
than 95% and decreases to below 90% after cycle 3. L-0 80.2 83.1 83.7
Because the shape recovering force of C2 series is so C1-5 80.0 82.7 89.3
strong, shape retention rate decreases compared to C1 C1-10 81.0 85.8 88.7
series (Fig. 13). Shape recovery pictures of L-0, C1-10, and C1-15 80.4 82.3 85.1
C2-10 are shown in Fig. 14. After each specimen is rolled, C2-5 88.8 90.5 94.8
the time required for unfolding is compared. C1-10 reduces
C2-10 90.7 93.7 96.8
the unfolding time from 200 second of L-0 to 121 second,
C2-15 91.2 92.5 94.0
and C2-10 achieves the unfolding in just 66 second.
C2-20 88.0 90.8 90.8
Therefore, C2-10 recovers the original shape in one third of
C2-25 89.9 91.3 91.5
the time necessary for L-0. Along with all the above

123
J Mater Sci (2007) 42:90459056 9055

(a) 100 (b) 100

95 95
Shape retention (% )

Shape recovery (% )
90 90

85 85

80 80

75 75
C1 series C1 series
70 70 C2 series
C2 series
65 65
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Glycerol content (wt%) Glycerol content (wt%)

Fig. 13 (a) Shape retention and (b) shape recovery vs. glycerol content profile of SMPU

Conclusion The behavior of C2-10 in stressstrain curve is very similar

C2 series cross-linked with a PEG-200 spacer is compared
with the linear L-0 and the randomly cross-linked C1 ser-
ies. The cross-link density increases and the phase sepa-
ration decreases with the increase of glycerol content for
both C1 and C2 series. The ordered structure of hard seg-
ment is disrupted by cross-linking based on XRD analysis.
The cross-linking of PU raises Tg and Tm for both C1 and
C2 series. Maximum stress of C2 series increases up to
800% compared to L-0, while maintaining a similar strain.
to natural rubber, but both stress and strain are 300% higher
than natural rubber. Shape recovery rate of C2-10 goes up
to an amazing 96.8% and does not decrease below 93%
after cycle 4, all without showing any drop in the shape
recovery speed. All of the special and outstanding shape
memory and mechanical properties of the C2 series origi-
nate from the parallel networking of the PU chains with the
flexible PEG-200 spacer, and there remains much work to
be done to attain the goal of a higher shape recovery rate
and an upkeep of the high shape recovery rate after use.

Fig. 14 Photographic comparison of shape recovery speed of (a) L-0, (b) C1-10, and C2-10

123
9056
Acknowledgments The authors of this paper would like to thank
the Korea Science and Engineering Foundation (KOSEF) for spon-
soring this research through the SRC/ERC Program of MOST/KOSEF
(R11-2005-065).
References
1. Yang JH, Chun BC, Chung YC, Cho JH (2003) Polymer 44:3251
2. Li F, Chen Y, Zhu W, Zhang X, Xu M (1998) Polymer 39:6929
3. Liang C, Rogers CA, Malafeew E (1997) J Intell Mater Syst
Struct 8:380
4. Chun BC, Cha SH, Chung YC, Cho JW (2002) J Appl Polym Sci
83:27
5. Kim BK, Lee SY, Xu M (1996) Polymer 37:5781
6. Tobushi H, Hara H, Yamada E, Hayashi S (1996) Smart Mater
Struct 5:483
123
J Mater Sci (2007) 42:90459056
7. Takahashi T, Hayashi N, Hayashi S (1996) J Appl Polym Sci
60:1061
8. Lee BS, Chun BC, Chung YC, Sul KI, Cho JW (2001) Macro-
molecules 34:6431
9. Ota S (1981) Radia Phys Chem 18:81
10. Chen LW, Lin JR (1998) J Appl Polym Sci 69:1563
11. Chen LW, Lin JR (1998) J Appl Polym Sci 69:1575
12. Hu J, Yang Z, Yeung L, Ji F, Liu Y (2005) Polym Int 54:854
13. Yang Z, Hu J, Liu Y, Yeung L (2006) Mater Chem Phys 98:368
14. Carey MA, Wellons SL, Elder DK (1984) J Cellular Plastic 20:42
15. Petrovic ZS, Javni I, Divjakovic V (1998) J Polym Sci B: Polym
Phys 36:221
16. Hur MK, Kwak JM, Hur T (1996) Polymer (Korea) 20:392
17. Coleman MM, Lee KH, Skrovanek DJ, Painter PC (1986) Mac-
romolecules 19:2149
18. Seymour RW, Estes GM, Cooper SL (1970) Macromolecules
3:579
J Mater Sci (2007) 42:90579062
DOI 10.1007/s10853-007-1833-1
Formation of surface magnetite nanoparticles
from iron-exchanged zeolite using microwave radiation
J. A. Heuser W. U. Spendel A. N. Pisarenko
C. Yu M. J. Pechan G. E. Pacey
Received: 16 January 2007 / Accepted: 8 May 2007 / Published online: 20 July 2007
Springer Science+Business Media, LLC 2007
Abstract Microwave radiation simplifies synthesis
methods by reducing reaction times, requiring fewer
materials, and also controlling reaction processes. We have
successfully synthesized nanoparticles of iron oxide and
zinc oxide coated on zeolite A using microwaves. The
radiation assisted in displacing either ferrous or zinc ions
from the pre-loaded zeolite network and increasing
reaction speed with solution at the interface. Products were
characterized by TEM, XRD, VSM, ICP-AES, and
fluorescence. We demonstrate the ability of using cation-
exchanged zeolites as microreactors to bias reactions onto
the zeolite surface. Efficient structure-directed surface
reactions are a potential route to making unique supported
nanomaterials for applications such as sensors, environ-
mental remediation, and chemical catalysis.
Introduction
Many of the unique phenomena that nanomaterials exhibit
are highly dependent not only on the chemical constituents,
but also on the size and shape of the nanomaterial. Given
that controlling the rate of nucleation is the key to
J. A. Heuser (&)
W. U. Spendel
A. N. Pisarenko

G. E. Pacey
Miami University Center for Nanotechnology, Department
of Chemistry and Biochemistry, Miami University, Oxford,
OH 45056, USA
e-mail: heuserja@muohio.edu
C. Yu
M. J. Pechan
Department of Physics, Miami University, Oxford, OH 45056,
USA
controlling size, and in some cases shape, it would be
desirable to quickly and uniformly synthesize nanomate-
rials. One approach to obtaining this control is the use of
microwave systems in synthesis. Besides the obvious
advantage of reducing reaction times from days to minutes,
the nucleation processes are better controlled under
microwave conditions since the reaction is driven by the
quick heat and pressure produced by the microwave system
[13]. The resulting physical characteristics of the nano-
particles are controllable and more reproducible [4]. For
example, microwave methods now exist to synthesize
nanoparticles with high degrees of monodispersity [5]. An
added advantage is that the amounts of reactants and sol-
vents used in the synthesis are reduced, providing a
greener synthesis route with fewer cleanups required.
Of particular importance to this report is that microwave
synthesis of submicron-sized magnetite powders has been
shown [4]. They reacted both ferrous and ferric salts
together and the ferrous salt alone with sodium hydroxide
in teflon digestion vessels in a high-powered (1,200 W)
microwave. By using only two chemicals in the reaction,
the conversion has few byproducts, resulting in a product
with little impurity. The complement of microwave power
facilitates rapid synthesis and relatively consistent heating
within the vessel.
Solution-based nanoparticles are not always the optimal
path of synthesis since the functionality of the particles can
be realized only after removal of the solvent and placement
of the particles in another media. For instance, some
nanoparticle metal oxides are important in the development
of industrial catalysts [6]. It would be ideal to synthesize
the desired nanoparticles within the proper medium,
thereby eliminating the transfer step. One widely-used
catalyst is zeolite A, a white powder of aluminosilicate
networks. Oliveria et al. used this zeolite not as a catalyst,
123
9058
but rather as an ion-extracting agent in water [7]. The
group used solution-based synthesis to produce a magnetic
zeolite by incorporating magnetite (Fe3O4) about the zeo-
lite particles. The zeolite framework remained open to
extract cations from the solution and the attached magnetite
allowed powder withdrawal by using a simple magnet.
As an extension of prior work, we are reporting a
microwave synthesis method to yield submicron-sized
magnetite and zinc oxide particles coated on zeolite A.
Microwave radiation served to both rapidly displace ions
from the zeolite matrix and position them at the zeolite/
solution interface. After preloading a zeolite with selected
cations, they were displaced from the framework to the
solution interface where the appropriate chemical reactions
took place. The significance is our method of employing
microwaves to a chemical system yielding nanoparticles
coated directly on the support matrix with little solution
reaction competition. Minimal preparation and reaction
time enhance the success of this method.
Experimental procedure
Chemicals were used without additional purification unless
otherwise noted. Ferrous sulfate, zinc nitrate, and sodium
hydroxide were purchased from Aldrich. Valfor 100
zeolite was donated by The PQ Corporation.
Zeolite was first preloaded with the desired cation by
stirring 100 mL of 1 M ferrous sulfate or zinc nitrate
solution with 8 g of zeolite. The suspension was allowed to
equilibrate for at least 24 h to ion-exchange either iron or
zinc, respectively. After sufficient mixing the slurry was
filtered and thoroughly washed with deionized water, then
diluted to 30 mL. Once the zeolite was loaded and washed
it was ready for reaction in the microwave vessel.
Sodium hydroxide solution was used for both syntheses
as the other reactant. A 4.8 mL aliquot of 0.25 M NaOH
(reaction concentration of 0.2 M) was combined with
1.2 mL of loaded zeolite suspension in a glass reaction
vessel and capped with a pressure-sensitive lid. Identical
samples were also left to sit overnight without being
exposed to radiation to observe unassisted reactions.
Solution reactions without zeolite present were synthesized
by combining 1.2 mL of either 0.2 M ferrous sulfate or
zinc nitrate solutions with 4.8 mL of 0.25 M NaOH. The
reaction tubes were exposed to microwave irradiation in a
CEM Discover unit using the following parameters:
Power300 W; Ramp time10 s; Hold time10 min; reaction mixture as the typical ferrous ion/zeolite mix. A
Max temperature150 C; Max pressure200 PSI; TEM image (not shown due to rapid decomposition under the
Stirring and powerMaxOn.
Products were investigated with transmission electron
microscopy (TEM), X-ray diffraction (XRD), vibrating
sample magnetometry (VSM), fluorescence spectropho-
123
J Mater Sci (2007) 42:90579062
tometry (zinc synthesis only), and inductively coupled
plasma (ICP) emission spectroscopy. The specifications are
as follow. TEM pictures were obtained using a Zeiss 10C
operated at 100 KeV with the images being captured on
Kodak 4489 film and scanned in on an Agfa Duoscan at
1000PPI. XRD patterns were obtained using a Scintag X1
powder X-ray diffractometer. VSM measurements were
made on a home-assembled apparatus utilizing the Faraday
Law to detect induced voltage in the pick-up coils by the
vibration of the measured magnetic sample in the magnetic
field. The frequency used for the vibrator is about 48 Hz
and the system was calibrated using a standard Pd sample
with known moment. This VSM system has a moment of
sensitivity of about 10 micro-emu. Fluorescence data was
obtained on a Horiba Jobin Yvon Fluorolog-3 FL3-22
spectrophotometer. ICP emission data was obtained on a
Varian Liberty 150 spectrometer.
Glycerin (60% v/v) was added to a ferrous ion/zeolite/
hydroxide solution to increase the solution viscosity and
affect magnetite particle nucleation and growth. It was
hypothesized that different solvent properties would affect
formation of the iron oxide particles. Microwave condi-
tions and ion concentrations remained unchanged from the
previous experiments.
Since zeolite is often used as a cation-exchange med-
ium, the exchange capability was tested using calcium ions.
After the products were obtained, filtered, and rinsed, sat-
urated calcium solutions were added. They were allowed to
equilibrate for at least 24 h. The filtered solids were dried
under ambient conditions and digested with hydrochloric
acid. The resulting solution was analyzed using ICP
emission spectroscopy.
Results and discussion
The results obtained clearly demonstrate the hypothesized
interface-directed nanoparticle syntheses. Both iron oxide
and zinc oxide nanoparticles were synthesized about the
zeolite surface after exposure to microwave radiation.
Samples not exposed to microwave radiation took hours to
react and the particles were of micrometer size. The reac-
tion did not appear to go to completion after 24 h. How-
ever, under microwave radiation conditions, small, more
desirable nanoparticles were synthesized rapidly.
A representative TEM picture of the iron oxide/zeolite
product is given by Fig. 1. Glycerin was added to the same
electron beam) of the glycerin samples shows a significantly
narrower size distribution relative to the water-only syn-
theses. Particles about the zeolite were between one to three
nanometers and showed a high degree of uniformity.
J Mater Sci (2007) 42:90579062
Fig. 1 TEM image of the magnetite/zeolite product
The same product shown in Fig. 1 was investigated
using XRD to determine the form of iron oxide present. As
can be seen by Fig. 2, the zeolite peaks dominate the
spectrum, obscuring iron oxide peaks. This indicates the
zeolite remained intact after microwave exposure but gives
no data regarding the iron oxide species formed.
In order to determine what form of iron oxide was likely
synthesized, we obtained an XRD diffractogram for the
solution reaction product in the absence of zeolite. The
XRD pattern of our iron oxide is shown in Fig. 3; black
arrows indicate experimental peaks that align with litera-
ture peaks for magnetite. Although the peaks are weak in
Fig. 2 XRD pattern of ferrous
ion-loaded zeolite A after
microwave exposure
9059
intensity, they align well to the literature diffraction pattern
of magnetite. Small particle sizes or the presence of
amorphous material may be responsible for these weak
signals. This result indicates the crystalline product is
magnetite, as opposed to other ferrous/ferric oxides/
hydroxides, was the dominant product. Since we did not
obtain an XRD pattern containing peaks for both zeolite
and magnetite species, the iron oxide in the zeolite reaction
may not be entirely magnetite. However, the magnetization
results support magnetite formation.
A further investigation into the magnetism of the iron
oxide products gave an interesting result. VSM loops were
obtained for four samples: magnetite from ferrous ions and
hydroxide; ferrous ions, hydroxide, and glycerin; ferrous
ions, zeolite, and hydroxide; and ferrous ions, zeolite,
hydroxide, and glycerin. Figure 4 shows the normalized
magnetization loops, indicating that solution conditions
influence the magnetic properties of the synthesized prod-
ucts. The samples with the presence of glycerin have lower
remanence and higher saturation field, compared to the
water-only products, regardless of the zeolite presence.
However, zeolite presence slightly increases the coercivity
of the products from about 270300 Oe. All samples do not
saturate within 8 kOe available for these measurements.
The shape of the hysteresis loops suggests a coexistence of
superparamagnetism and ferromagnetism of the magnetic
particles due to a size distribution.
To confirm our synthesis method was general and not
unique to iron oxide, we carried out the synthesis with zinc
ions in place of iron ions. As in the iron reaction, we again
observed nanoparticles, here as zinc oxide draped about the
zeolite surface. The TEM image of the product, Fig. 5,
shows this; the zinc oxide coverage, however, is not as
123
9060 J Mater Sci (2007) 42:90579062

Fig. 3 XRD pattern of iron

oxide powder

Fig. 4 VSM loops of magnetite/zeolite products

extensive as the coverage by the magnetite about the

zeolite. Explanations for the observed coverage differences
are being pursued.
Zinc oxide growth composition about the zeolite was
more varied in comparison with the iron oxide. Cones,
large clusters, and flat sheets of zinc oxide were observed.
The implication is the zinc oxide nucleated and grew via
multiple mechanistic pathways.
Because zinc oxide is a known fluorophore, fluorescence
spectra were obtained to compare the zinc oxide solution
and zeolite interface-directed synthesis products. Spectra
were normalized at the 416 nm peak for comparison. The
spectra given in Fig. 6, collected with an excitation
wavelength of 365 nm, show highly resolved vibrational
structure for zinc oxide on zeolite. Comparing the zinc-
exchanged and washed zeolite reaction spectrum (labeled
ZnO/Zeolite washed) with the zinc-exchanged zeolite reaction is highly weighted toward zinc oxide nucleation
Fig. 5 TEM image of the zinc oxide/zeolite product
reaction in the presence of solution zinc ions (labeled
ZnO/Zeolite) spectrum indicates the surface reaction is
favored. The spectrum of zinc oxide solution reaction with
no zeolite (labeled ZnO no zeolite) has a strong emis-
sion peak at 416 nm and a broad emission at 572 nm. The
competitive solution product formation of the ZnO/Zeolite
product is most clearly seen in the higher fluorescence
intensity at 572 nm versus the ZnO/Zeolite washed spec-
trum. The observed fine structure and limited elevated
fluorescence between 510 nm and 620 nm indicates the

123
J Mater Sci (2007) 42:90579062
Fig. 6 Fluorescence spectra of
zinc oxide and zinc oxide/
zeolite products
on the zeolite. Had the solution-synthesized zinc oxide
reaction dominated we would not have seen the predomi-
nant spectral fine structure between 480 nm and 550 nm;
the broad fluorescence between 510 nm and 620 nm would
have been more prominent.
The exchange capacities of the following solids were
investigated: magnetite on zeolite, zinc oxide on zeolite,
microwaved zeolite, and zeolite. Results from the acid
digestion and subsequent ICP analysis indicate that no
exchange capacity was lost when compared to the reagent
zeolite sample, given in Table 1. An interesting point is
that the microwaved zeolite control had a somewhat higher
degree of exchange capacity, possibly attributed to the
radiation increasing cavity activity compared with the
untreated commercial product. The magnetite also shows a
higher level of exchange, but this value may be enhanced
by calcium adsorption on the magnetite. Magnetic zeolite
has previously been shown to be a more efficient cation
extractor than zeolite alone [7].
Also listed in Table 1 are the theoretical maximum
percentages of product expected based on the ICP values
obtained and the 5.5 milliequivalent per gram exchange
capacity of zeolite A [8]. The data we obtained are slightly
lower than the maximum values mainly due to inefficient
exchange and incomplete cation reaction at the zeolite
interface. Some metal oxide product formed may not have
Table 1 Calcium exchange data and oxide percent values, both
theoretical and experimental
meq/g Ca2+ % Oxide Theoret. %
Magnetite/zeolite 4.7 0.3 17.2 1.6 17.5
Zinc oxide/zeolite 4.1 0.3 14.5 0.6 18.3
Microwaved zeolite 5.3 0.1 1. Komarneni S, Hussein MZ, Liu C, Breval E, Malla PB (1995) Eur
Zeolite 4.1 0.2
9061
bound to the zeolite. Despite this, the numbers confirm the
majority of the metal oxide reacts at the zeolite/solution
interface.
Conclusions
We have demonstrated a zeolite interface-directed metal
oxide nanoparticle synthesis method potentially useful in
making catalysts, sensors, and a new approach to making
nanostructured interfaces. These experiments show the
ability of using zeolite ion-exchange as a means of deliv-
ering cations in reactions where the zeolite serves as a
microreactor. Microwave radiation increases the ion ex-
change rate in a controlled manner to displace ions from
the zeolite network to form magnetite or zinc oxide by
reactions of simple ferrous or zinc salts with hydroxide.
Cation delivery from the zeolite adds the ability to control
nanoparticle growth and nucleation at the zeolite interface.
Nucleation control at the interface and solvent effects
provide a pathway to synthesize new nanostructured
interfaces.
Future work in our laboratory is directed towards
establishing microwave solution reaction conditions to
control nanoparticle shape and size. Nanoparticle micro-
wave synthesis provides a significant opportunity for time-
efficient (less than 10 min) and green synthesis of unique
interfaces with potential applications in catalysis, chemical
analysis, electronic devices, and sensors.
References
J Solid State Inorg Chem 32:837
2. Komarneni S, Katsuki H (2002) Pure Appl Chem 74:1537
123
9062 J Mater Sci (2007) 42:90579062

3. He R, Qian X-F, Yin J, Xi H-A, Bian L-J, Zhu Z-K (2003) Colloid
Surf A 220:151
4. Khollam YB, Dhage SR, Potdar HS, Deshpande SB, Bakare PP,
Kulkarni SD, Date SK (2002) Mater Lett 56:571
5. Gerbec JA, Magana D, Washington A, Strouse GF (2005) J Am
Chem Soc 127:15791
6. Glaspell G, Fuoco L, El-Shall MS (2005) J Phys Chem B
109:17350
7. Oliveira LCA, Petkowicz DI, Smaniotto A, Pergher SBC (2004)
Water Res 38:3699
8. Breck DW (1974) Zeolite molecular sieves, structure, chemistry,
and use. John Wiley & Sons

123
J Mater Sci (2007) 42:90639069
DOI 10.1007/s10853-007-1692-9

Effect of chromium addition on microstructure, tensile properties

and creep resistance of as-cast Ti-48Al alloy
E. Hamzah M. Kanniah M. Harun

Received: 24 November 2006 / Accepted: 13 March 2007 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007

Abstract The effect of Cr on the microstructure, tensile regard, alloying chemistry may be used to achieve
properties and creep resistance of as-cast Ti48AlxCr balanced mechanical properties.
(x = 0, 2, 4 at.%) alloys were studied. The dependence of The alloying elements can be classified to two broad
the tensile properties and creep resistance of as-cast TiAl groups. The first group of elements is the solid solution
on the solid solution strengthening and formation of b strengthening elements or beta stabilizers such as Cr, Mn,
phase due to addition of Cr was investigated. V, Nb, Fe, Mo, W, etc., and the second group of elements
consists of B, Si and C which provide precipitation or
particle strengthening [6]. Among the various alloying
elements, Cr is unique in several ways [7]. Firstly, TiAlCr
Introduction
alloys containing the ternary Laves phase Ti(Cr,Al)2 not
only have outstanding oxidation resistance [8], but more
TiAl-based alloys are being considered as potential light-
importantly, have very similar thermal expansion coeffi-
weight materials for number of high temperature applica-
cients to those of c-alloys. Secondly, theoretical calcula-
tions [1, 2]. These alloys consist of high volume fraction of
tions based on the total energy LMTO method have shown
c phase together with small volume fraction of a2 phase.
that interaction between Cr and Al does not stabilize x
One of the major challenges encountered in the develop-
phase formation in the b phase [9, 10]. This makes it
ment and potential application of two-phase TiAl is how to
possible to ductilise the c-alloys by compositing the c
optimize the mechanical properties. With respect to opti-
phase with a ductile b phase of A2 or B2 structure [7].
mization of the mechanical properties of TiAl, alloying and
This investigation is to study the effect of Cr addition on
microstructure control must be considered [3]. Engineering
the microstructure, tensile properties and creep resistance
TiAl based alloys should exhibit balanced room tempera-
of as-cast TiAl alloy. The microstructure and mechanical
ture ductility, fracture toughness, high temperature
properties of as-cast gamma titanium aluminide containing
strength, creep and oxidation resistance [4]. So far, much
beta phase, Ti48Al4Cr are described and a comparison is
effort world-wide has been devoted to studying the effects
made with typical two-phase, first generation aluminides
of thermomechanical processing and alloying in TiAl. The
exemplified by binary Ti48Al and solid solution
general consensus is, however, that where termomechani-
strengthen, Ti48Al2Cr. The dependence of the tensile
cal processing results in the improvement of a specific
properties and creep resistance of as-cast TiAl on the solid
property, it is often at the expense of another [5]. In this
solution strengthening and formation of b phase due to
addition of Cr was investigated.
E. Hamzah (&)
M. Kanniah
Faculty of Mechanical Engineering, Universiti Teknologi
Malaysia, Skudai, Johor 81310, Malaysia
e-mail: esah@fkm.utm.my Materials and experimental procedures
M. Harun
Industrial Technology Division, Malaysian Institute for Nuclear The materials investigated in this study have the nominal
Technology Research, Bangi, Kajang, Selangor 43000, Malaysia composition of Ti48 at.%Al, Ti48 at.%Al2 at.%Cr and

123
9064
Ti48 at.%Al4 at.%Cr. The alloys were produced by IRC
University of Birmingham, United Kingdom by plasma
melting casting into 2 kg buttons. All the alloys will be
referred in atomic percentage hereafter. The samples for
metallography analysis were cut using Manho EDM wire
cut to 10 mm 10 mm 10 mm cube. The samples were
ground and polished according to standard procedures. The
polished surface was etched using Krolls reagent (94 ml
H2O + 4 ml HNO3 + 2 ml HF). The microstructures were
examined using Phillips XL40 scanning electron micro-
scope. The grain size and volume fraction is measured
using the intercept and point counting methods respec-
tively. The lattice parameters of the alloys were measured
by Siemens D 5000 X-ray diffractometer.
Flat samples with dimension, 3 mm 2 mm with
15 mm gauge length were prepared for the tensile and
creep test. Prior to testing, the test pieces were ground
along the gauge length in the longitudinal direction to
prevent premature crack initiation at surface defects during
tension or creep. Instron tensile testing machine was used
to perform the tensile testing at room temperature. Con-
stant load tensile creep test was performed at 600800 C
with initial stresses of 150180 MPa. The constant load
tensile creep test was conducted using Mayes TC 20 High
Temperature Creep Testing Machine, consisting three zone
temperature furnace and a loading lever arm ratio of 10:1
in air. The test temperatures were maintained constant with
a precision of 1 C and monitored for 2 h before the test.
The temperature was monitored during the creep test by
using a thermocouple in contact with the gauge section of
the sample. The sample was allowed to soak at test tem-
perature for 1 h before the load was applied. The test was
interrupted after 20 h and the samples were left to cool in
air to room temperature.
Experimental results
Microstructural characterisation
Ti48Al and Ti48Al2Cr exhibits nearly lamellae lamellae, c grains and b phase, were measured using EDX
microstructure consisting high volume fraction of c phase
(tetragonal, L10 structure) and a small amount of a2 phase
(hexagonal, D019). Ti48Al4Cr too exhibits nearly
lamellae microstructure but consist of lamellae, fine c and
b phase. Addition of Cr up to 4 at.%, introduces b phase
(bcc structure). The lamellae structure consists of alter-
nating laths of the c-TiAl and a2-Ti3Al phases. Such
lamellae structure results from the solid-state phase trans-
formation of the primary disordered a dendrites. The single
fine c regions surrounding the lamellae grains result from
the transformation of the aluminum rich interdendritic melt
due to incompleteness of the peritectic reactions. The two
123
J Mater Sci (2007) 42:90639069
peritectic reactions, L + b a and L + a c, is
hardly to be completed due to limited diffusion caused by
the formation of a solid envelope of the peritectic phase,
avoiding the physical contact between the reactants [11].
As far b phase is concerned, even though it is found to exist
together with a(a2) and c phases in b stabilizer added TiAl,
attention has been paid mainly to the role of b phase in the
mechanical properties but not to the formation mechanism
of the microstructure as well as the phase equilibria among
b, a and c phases. Therefore, the microstructural formation
in the alloys under development or study has in most case
been interpreted based on the binary phase diagram, since
the amount of the additive elements is not large [12].
However, little attempt has been made to establish the
phase relationship among the b, a and c at elevated tem-
peratures [7, 13]. Figure 1a, b and c shows the micro-
structure of Ti48Al, Ti48Al2Cr, and Ti48Al4Cr
respectively.
Figure 1 clearly shows that Ti48Al and Ti48Al2Cr
exhibits a nearly lamellae microstructure which consists
mainly the lamellae grain with a grain size of 500800 lm
and 25% of fine gamma grains at the grain boundaries.
Even though the lamellae grain size of Ti48Al and
Ti48Al2Cr is in a similar range, the volume fraction of
c phase in Ti48Al2Cr is significantly higher than in
Ti48Al. The microstructure of Ti48Al4Cr consists
three phase which are the lamellae (300600 lm), fine
gamma (510%) and beta (25%) phase. The beta grains
are observed to be irregular in shape together with fine c
grains along grain boundaries. Therefore, it is very hard to
distinguish between c and b grains of Ti48Al4Cr at low
magnifications. SEM image in Fig. 1c clearly reveals the
presence of b phase (bright contrast) along lamellae grain
boundaries. This type microstructure appears to be com-
monly observed in b containing TiAl alloy, e.g., Ti46Al
(1.52)Cr2Mo0.25Si0.3B [14, 15], Ti46.5Al2Cr
2Nb0.8Mo0.2W0.2Si [16] and Ti-4510Nb [17].
Table 1 provides the lamellae spacing and grain size of the
alloys. Generally, these features are in the similar range.
The composition of all the phases, such as a2 + c
method. The solution of Cr in c phase increases with
addition of Cr (04 at.%) in Ti48Al. The solution of Cr in
the b phase of Ti48Al4Cr is 11.15 at.%, indicating that
Cr is a strong b stabilizer. This results shows that Cr sta-
bilizes b and slightly c phase. Table 2 shows the compo-
sitions of all the phases of the alloys studied.
Table 3 shows the variation of lattice parameter a and c
of the c phase and increment of Cr solution in c phase with
addition of Cr in Ti48Al. The variation of lattice param-
eter in the c phase is complex. The variation of lattice
parameter and increment of Cr in c phase shows the solid
solution strengthening of Cr in c phase.
J Mater Sci (2007) 42:90639069 9065

Table 2 Compositions of a2 + c lamellae, fine c and b phase in

TiAl and TiAlCr
Alloy Phase Ti (at.%) Al (at.%) Cr (at.%)

Ti48Al a2 + c lamellae 58.32 41.68

Fine c 52.88 47.12
Ti48Al2Cr a2 + c lamellae 54.40 43.73 1.87
Fine c 53.55 44.19 2.27
Ti48Al4Cr a2 + c lamellae 52.41 42.86 4.73
Fine c 52.56 43.65 3.79
b 51.62 37.23 11.15

Table 3 Lattice parameter and Cr solution in c phase of the alloys

Alloys c (nm) a (nm) Cr (at.%)

Ti48Al 4.049 3.976

Ti48Al2Cr 24.32 3.971 2.27
Ti48Al4Cr 24.32 3.971 3.79

alloys at room temperature are illustrated in Fig. 2a and b

respectively. Figure 2a and b shows that fracture strength
and ductility significantly increases with Cr addition of up
to 2.1 at.%, where the peak strengths and ductility are
540 MPa and 2.15% respectively, and then sharply
decrease with further Cr additions of up to 4 at.%.
Ti48Al2Cr exhibited highest fracture strength and duc-
tility compared to the other two alloys. The increase of
fracture strength at room temperature is chiefly attributed
by the solid solution strengthening of chromium whereas
the increase of ductility is due to contribution of super-
dislocation besides the ordinary dislocation in c phase [6].
The presence of b phase in TiAl decreases the fracture
strength and ductility at room temperature. This is because
the b phase, with a high content of Cr, is brittle and hard at
Fig. 1 Microstructure of as-cast (a) Ti48Al, (b) Ti48Al2Cr and room temperature. Therefore the sharp decrease in fracture
(c) Ti48Al4Cr strength and ductility with increasing Cr content is
undoubtedly due to the onset of b phase in TiAl. This
Table 1 Lamellae spacing and grain size of the alloys
agrees to the previous research that b reduces tensile
strength and ductility [6]. It is reported that the composi-
Alloys Lamellae Lamellae grain tion where fracture strength starts to decrease after reach-
spacing (lm) size (lm)
ing the peak, are the point where b phase starts to form [6].
Ti48Al 1.7 500800
Ti48Al2Cr 1.7 500800 Room temperature fracture
Ti48Al4Cr 1.7 300800
Figure 3ac shows representative fractographs of Ti48Al,
Ti48Al2Cr and Ti48Al4Cr following tensile tests at
Room temperature tensile properties room temperature respectively. Scanning electron micros-
copy examination indicates a mixed mode of failure for all
The effect of chromium on fracture strength and ductility of the alloys. As shown in Fig. 3a, the fracture surface of
binary, Ti48Al and ternary Ti48Al2Cr and Ti48Al4Cr Ti48Al exhibits large cleavage facets which are due to

123
9066 J Mater Sci (2007) 42:90639069

Fig. 2 (a) Fracture strength and 600 2.5

(b) ductility at room
500
(a) (b)

Tensile Strength (MPa)

temperature as function of Cr 2
content

Ductility (%)
400
1.5
300
1
200

0.5
100

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Amount of Chromium (%) Amount of Chromium (%)

transgranular fracture across the c grains. In the lamellae

phases, crack propagation appears to progress along the
a2/c interfaces (interlamellae separation). For Ti48Al
2Cr, cleavage faceted transgranular fracture of the c grains
is still evident, but a more dimple-like fracture appear-
ance is more generally present (Fig. 3b). This reflects a
higher ductility. The fracture surface of Ti48Al4Cr
(Fig. 3c) are characteristic of flat, featureless facets with a
transgranular fracture mode, in which typical elongated
cleavage facets corresponding to c and b phase are ob-
served. This indicates that the b phase in TiAl fractured by
a cleavage mode is brittle at room temperature.

Creep resistance

Creep curves of all the three compositions at 600800 C

with initial stresses 150180 MPa are illustrated in
Fig. 4ad, with the primary creep strain and secondary
creep rate are summarized in Table 4. Figure 4a show
creep curves of the alloys at 600 C with initial stress of
180 MPa. All the three alloys showed steady state creep
behaviour. The creep curve of Ti48Al and Ti48Al2Cr is
almost identical. The total creep strains accumulated for
20 h, instantaneous creep strain and steady state creep rate
of both that alloys is in a similar range. It appears that the
solid solution strengthening does not have significant effect
on the creep resistance. On the other hand, beta phase
contained Ti48Al4Cr showed excellent creep resistance
compared to Ti48Al and Ti48Al2Cr. The primary creep
strain and steady state creep rate is significantly low
compared to first generation Ti48Al and Ti48Al2Cr. At
600 C, the steady state creep rate of Ti48Al is reduced
69.5% with presence of beta phase (comparison between
Ti48Al and Ti48Al4Cr). This figure illustrates the
significant effect of beta phase on creep resistance.
Analysis on Fig. 4b shows that the creep response of the
alloys at 700 C with initial stress of 180 MPa is similar as
at 600 C. However, there is slight improvement in
creep resistance of Ti48Al2Cr compared to Ti48Al.
Ti48Al4Cr showed excellent creep resistance. However,
the steady state creep rate is only reduced 56% (compari- Fig. 3 Fractographs of (a) Ti48Al, (b) Ti48Al2Cr, (c) Ti48Al4Cr
son between Ti48Al4Cr and Ti48Al). It is evident from after tension at room temperature

123
J Mater Sci (2007) 42:90639069 9067

Fig. 4 Typical e vs. t creep (a) 1.2 (b) 1.4

Ti-48Al Ti-48Al
curves of the alloys at (a) interrupted
1.2 interrupted
1
600 C, (b) 700 C, (c) 800 C

Creep strain (%)

Creep strain (%)

Ti-48Al-2Cr Ti-48Al-2Cr
with initial stress of 180 MPa interrupted 1 interrupted
0.8
and (d) 800 C with initial 0.8
0.6
stress of 150 MPa 0.6
Ti-48Al-4Cr
0.4 interrupted
Ti-48Al-4Cr 0.4
interrupted
0.2 0.2

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (hours) Time (hours)
2.5 10
(c) (d) Ti-48Al-4Cr
Ti-48Al 9 ruptured
2 interrupted 8
Creep strain (%)

Creep strain (%)

Ti-48Al-2Cr 7
interrupted
1.5 6
5
1 4
3
Ti-48Al
0.5 2 interrupted

1 Ti-48Al-2Cr
interrupted
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (hours) Time (hours)

Table 4 Primary creep strain

Alloy Temperature (C) Initial stress Primary creep Secondary
and secondary creep rate of the
(MPa) strain (%) creep rate (s1)
alloys
Ti48Al 600 180 0.320 2.78 105
Ti48Al2Cr 600 180 0.310 2.69 105
Ti48Al4Cr 600 180 0.045 8.49 106
Ti48Al 700 180 0.690 3.23 105
Ti48Al2Cr 700 180 0.330 2.87 105
Ti48Al4Cr 700 180 0.060 1.42 105
Ti48Al 800 180 1.340 4.74 105
Ti48Al2Cr 800 180 0.980 4.06 105
Ti48Al4Cr 800 180
Ti48Al 800 150 0.710 3.13 105
Ti48Al2Cr 800 150 0.590 2.69 105
Ti48Al4Cr 800 150 0.680 8.50 105

these results that the effect of beta phase on creep resis-
tance depends on the temperature.
Figure 4c shows the creep curves at 800 C with initial
stress of 180 MPa. Ti48Al and Ti48Al2Cr exhibits
0.53% and 0.6% instantaneous creep strain respectively
when loaded with initial stress of 180 MPa. Both alloys
showed steady state behaviour. However, Ti48Al4Cr
failed upon loading. It is quite interesting to note that
Ti48Al4Cr which exhibits low primary creep strains and
steady state creep rate compared to Ti48Al at 600 and
700 C failed upon loading at 800 C whereas Ti48Al and
Ti48Al2Cr exhibits steady state value until interruption.
Therefore a creep test is attempted with initial stress of
150 MPa to study the behaviour of Ti48Al4Cr at 800 C
and possible cause for the sudden failure during loading of
180 MPa.
Typical creep curve at 800 C with initial stress of
150 MPa are shown in Fig. 4d. Ti48Al and Ti48Al2Cr
exhibits steady state creep behaviour whereas Ti48Al4Cr
exhibits minimum creep behaviour. After the initial
instantaneous strain, Ti48Al and Ti48Al2Cr undergoes
a period of primary creep where the strain rate (de/dt)
decreases with time to an apparent steady state value that
persists for a long period of time until interruption. On the
other hand, Ti48Al4Cr which exhibits low instantaneous
creep strain compared to Ti48Al and Ti48Al2Cr,
undergoes a short primary creep and reaches a minimum
value before the onset of tertiary creep. This type of creep
behaviour is also referred as inverse creep where the
creep curve can be divided into two regimes, before and
after the minimum creep rate, instead of the common three.
Ti48Al4Cr sample failed after 16.5 h. Ti48Al and

123
9068
25
Ti-48Al
Ti-48Al-2Cr
20
Ti-48Al-4Cr
Time (hours)
15
10
5 treatment. However, controlled alloying chemistry, e.g., in
0
0.8 1 chemistry on microstructural features. The result of
Creep strain (%)
Fig. 5 Time needed to reach 0.8 and 1% of Creep Strain for the
Binary and Ternary Alloys
Ti48Al2Cr exhibits 0.53% and 0.47% of instantaneous
creep strain respectively, whereas Ti48Al4Cr exhibits
0.32%. The instantaneous creep strain of Ti48Al and
Ti48Al2Cr is in a similar range. Note also that the creep
strain of Ti48Al2Cr is just slightly less than of Ti48Al.
On the other hand, Ti48Al4Cr exhibits greater creep
strain before rupture. The period of succeeding accelerating
creep rates in the tertiary regime was found to be large and
corresponds to roughly 8590% of the total creep strain to
fracture. The time in the tertiary creep regime was about
8085% of the total time to rupture. The secondary creep
rate of Ti48Al and Ti48Al2Cr is almost identical,
however, Ti48Al4Cr exhibit a greater secondary creep
rate. Although, it is common to assume minimum creep
rate is equal to steady state creep rate for creep analysis or
comparison, however, it does not really accurate. To fur-
ther compare the alloys, the time necessary to reach 0.8 and
1% creep strains is recorded. Figure 5 shows that Ti48
Al2Cr took the longest time to reach 0.8 and 1% creep
strains followed by Ti48Al and Ti48Al4Cr respec-
tively. This indicates that solid solution strengthening by
chromium in fine c phase, slightly improves the creep
strength. On the other hand, Ti48Al4Cr had reached 0.8
and 1% creep strains in a short period which indicates low
creep strength. Beta phase deteriorates the creep resistance
at high temperature.
Discussion
Role of chromium addition on strengthening
mechanisms and microstructural modification
The variation of lattice parameters of c phase and incre-
ment of Cr in c phase with addition of Cr in Ti48Al shows
123
J Mater Sci (2007) 42:90639069
the solid solution strengthening of Cr in c phase. Each
alloying element has it own strengthening factor, therefore,
it is not necessary that solid solution strengthening by
singular alloying element would able to improve all the
mechanical properties. It might have strong effect on one
and none on the other.
Microstructural features such as lamellae spacing and
grain size has significant effect on the mechanical proper-
ties of TiAl. These features are usually controlled by heat
this present study; systematic addition of Cr (04 at.% Cr)
provides a path to study the minor effect of alloying
microstructural characterization shows that the lamellae
spacing and grain size is in a similar range for all the alloys
investigated. Therefore, it is concluded that addition of Cr
in the present study does not have significant effect on
microstructural features. However addition of 4 at.% Cr
introduces b phase. Therefore, the variation in mechanical
properties between the alloys studied is solely on the solid
solution strengthening and b phase.
Effect of chromium addition on tensile and creep
properties
The increase of the fracture strength at room temperature
with the Cr content is chiefly attributed to the solid-solution
strengthening of Cr in c phase. However, presence of b
phase in TiAl does not improve, but deteriorates the frac-
ture strength and ductility at room temperature. It is
because the b phase, with a high content of Al and Cr, is
brittle at room temperature. Tensile tests show that the
Ti48Al4Cr has intermediate ductility between Ti48Al
and Ti48Al2Cr. Although the fracture strength and
ductility of Ti48Al4Cr is higher than of Ti48Al, how-
ever it is doubted whether it is due to beta phase or not. The
cleavage like fracture surface of Ti48Al4Cr after room
temperature tension indicates the brittle characteristic of
beta phase. Therefore, b phase is undoubtedly detrimental
at room temperature due to its extreme brittleness.
Although it had been reported that solid solution
strengthening by addition of Cr increases the creep life of
TiAl at high stresses [6], however it has no significant effect
to the applications. This is due the creep strain exhibited by
solid solution strengthening TiAlCr (Ti48Al2Cr) is
only slightly lesser than binary Ti48Al at the intermediate
to high temperature. Aerospace and automotive applica-
tions not only concerns creep life but creep strain too.
Steady state creep rate and the time necessary to reach 1%
creep strain of Ti48Al2Cr are not significantly improved.
On the other hand, b phase becomes very ductile at high
temperature due to its open bcc structure [16]. It is evident
from the high creep strain exhibited by Ti48Al4Cr before
J Mater Sci (2007) 42:90639069
rupture during creep test at 800 C with initial stress of
150 MPa as shown in Fig. 4d. High creep strain would
exceed the tolerance limit of a component and thus prevents
the applicability of the alloy. Therefore, b phase is detri-
mental at both room and high temperature due to its extreme
brittle-ductile character. However, this study shows that b
phase might be useful at intermediate temperatures. This is
due to the excellent creep resistance of Ti48Al4Cr
compared to Ti48Al and Ti48Al2Cr at temperatures 5.
from 600700 C. The brittle-ductile character of b phase is
suggested to be moderate and therefore have good signifi-
cant effect on mechanical properties at the intermediate
temperatures. From the present study, it is concluded that
solid solution strengthening by Cr in c phase does not have
significant effect on creep strength of TiAl and introducing
b phase with addition of Cr is found to be detrimental for
creep strength with respect to the applications.
Modifications to the alloy chemistry have been
attempted in order to maintain a balance between the room
temperature ductility and high temperature creep resis-
tance. It is clear that as with all alloys, the high and low
temperature properties cannot be generally be optimized
simultaneously and the aim of the present program is to
assess the optimum alloying routes which allow an
acceptable compromise between these properties. It is of
course widely accepted limiting factor of the applications
of these alloys is their cost and development of cost-
effective processing is not been addressed in this paper.
Conclusions
1. As-cast Ti48Al and Ti48Al2Cr exhibited a nearly
lamellae microstructure consisting of a high volume
fraction of c phase and a small amount of a2 phase.
Ti48Al4Cr too exhibits a nearly lamellae micro-
structure but the lamellae grain boundaries is occupied
by fine c and body centered cubic (bcc) structured b
phase.
2. Addition of Cr stabilizes c (TiAl) and b phase. The
solubility of Cr in c and b increases with the addition
of Cr from 2 at.% to 4 at.%.
3. Ti48Al2Cr exhibited the highest fracture
strength and ductility followed by Ti48Al4Cr and
Ti48Al. The solid solution strengthening by Cr in
single c phase is responsible for this improvement.
However, further addition of Cr stabilizes the b phase
(Ti48Al4Cr) which decreases the tensile properties.
9069
4. All the alloys showed a mixed mode fracture surface
after tensile test. Generally, transgranular brittle frac-
ture is more evident due to large cleavage facet cor-
responding to c and b phase. Interlamellae separation
observed are due to crack along the a2/c interface.
Slight dimple-like appearance of fracture surface of
Ti48Al2Cr reflects higher ductility compared to
Ti48Al and Ti48Al4Cr.
Solid solution strengthening by chromium in c phase
has no significant effect on creep resistance with
respect to applications.
6. b phase deteriorates the creep resistance at high tem-
perature. However, presence of b phase has good
effect on creep resistance at intermediate temperatures,
600700 C.
7. Beta phase is very sensitive to temperature and detri-
mental for mechanical properties at both room and
high temperature.
Acknowledgements The authors would like to thank the Malaysian
Ministry of Science, Technology and Innovation (MOSTI) for the
research fund under the IRPA program (Project No. 09-02-06-0002
EA-002).
References
1. Yamaguchi M, Inui H (1993) In: Darolia R, Liu CT, Martin PL,
Miracle DB, Wagner R (eds) Structural intermetallics. TMS,
Warrendale, PA, USA, p 127
2. Kim YW, Dimiduk DM (1991) J Metals 43:40
3. Crofts PD, Bowen P, Jones IP (1996) Scripta Mater 35:1391
4. Marketz WT, Fischer FD, Clemens H (2003) Plasticity 19:281
5. Cheng TT, Wills MR, Jones IP (1999) Intermetallics 7:89
6. Sun FS, Cao CX, Kim SE, Lee YT, Yan MG (2001) Metall Mater
Trans A 32:1573
7. Shao G, Tsakiropoulos P (1999) Intermetallics 7:579
8. Brady MP, Smialek JL, Smith J, Humphrey DL (1997) Acta
Mater 45:2357
9. Shao G, Tsakiropoulos P, Nguyen-Manh D, Pettifor DG (1997)
Phil Mag Lett 76:207
10. Shao G, Passa E, Tsakiropoulos P (1997) In: Nathal MV, Darolia
R, Liu CT, Martin PL, Miracle DB, Wagner R, Yamaguchi M
(eds) Structural Materials. TMS, Warrendale, PA, USA. p 825
11. Duarte A, Viana F, Henrique M, Santos CM (1999) Mater Res 2:1
12. Takeyama M, Ohmura Y, Kikuchi M, Matsuo T (1998) Inter-
metallics 6:643
13. Jewett TJ, Ahrens B, Dahms M (1997) Mater Sci Eng A 225:29
14. Schillinger W, Clemens H, Dehm G, Bartels A (2002) Inter-
metallics 10:459
15. Zhang D, Arzt E, Clemens H (1999) Intermetallics 7:1081
16. Wang JG, Nieh TG (2000) Intermetallics 8:737
17. Zhang WJ, Chen GL, Appel F, Nieh TG, Deevi SC (2001) Mater
Sci Eng A 315:250
123
J Mater Sci (2007) 42:90709074
DOI 10.1007/s10853-007-1640-8
Phosphonium iodine as nickel corrosion inhibitor in 1 M sulfuric
acid medium
Fatima Said Nebil Souissi Karima Es-Salah
Najat Hajjaji Ezzeddine Triki Abdullah Srhiri
Received: 9 February 2006 / Accepted: 26 February 2007 / Published online: 18 July 2007

Springer Science+Business Media, LLC 2007
Abstract The inhibiting action of alkyltriphenylphos-
phonium iodine salt ((C8H17)Ph3P+,I) towards the corro-
sion behaviour of nickel in 1 M H2SO4 solution has been
studied. This compound was found to retard both anodic
and cathodic reactions of nickel corrosion. At constant
temperature, the corrosion rate decreases with increasing
inhibitor concentration. On the other hand, the increase in
temperature leads to an increase in the corrosion rate. The
activation energy, DEa, were calculated. They were found
19.3 kJ mol1 and 71.1 kJ mol1, respectively for the
uninhibited solution and in the presence of 103 M of
phosphonium salt. The inhibitor adsorption was identified
to occur according to Langmuir isotherm. The equilibrium
constant, k, as well as the free energy of adsorption,
DadsG, for inhibitor process were then calculated. Phos-
phonium iodine exhibited a singular behaviour for T
318 K where inhibitor desorption increases.
F. Said
K. Es-Salah
A. Srhiri
Laboratoire dElectrochimie, des Etudes de Corrosion &
dEnvironnement, Faculte des Sciences de Kenitra, BP 133,
Kenitra 1400, Marocco
N. Souissi (&)
E. Triki
Unite de Recherche Corrosion & Protection des Metalliques,
Ecole Nationale dIngenieurs de Tunis, BP 37, Le Belvedere
1002, Tunisie
e-mail: Nebil.Souissi@fsb.rnu.tn
N. Hajjaji
Laboratoire de Synthe`se Organique et de Reactivite
Electrochimique, Faculte des Sciences de Kenitra, BP 133,
Kenitra 1400, Marocco
123
Introduction
Phosphonium salts are considered as excellent corrosion
inhibitor, especially in acidic media. In fact, Morad [1]
used six organic phosphonium compounds of the structure
YTP+,X, (T= phenyl, X = Br or Cl and Y = propyl,
propargyl, cyclopropyl, allyl, 1,3- dioxolanyl and cinnam-
yl) to inhibit the corrosion of mild steel in aerated 0.5 M
H2SO4 solution. Khaled [2] evaluated the inhibiting action
of (chloromethyl) triphenyl phosphonium chloride, triphe-
nyl (phenylmethyl) phosphonium chloride and tetraphenyl
phosphonium chloride on the corrosion of iron in 1 M HCl
solution. In previous works [3], we tested tetraphenyl
phosphonium bromide as nickel corrosion inhibitor in
sulfuric acid medium. Recently, we have evaluated the
effect of R+,X (R+ = (C8H17)Ph3P+ or K+, X = I or Br or
Cl) salts addition on the corrosion behaviour of nickel in
1 M H2SO4 medium [4]. The results achieved allowed us
considering that phosphonium iodine addition modifies the
interface behaviour due to the interaction between the
molecule and the material surface.
Thus, in this research program we studied phosphonium
iodine salt adsorption on nickel when immersed in 1 M
sulphuric acid medium.
Experimental
Electrochemical experiments were conducted in a classical
three-electrode cell. Saturated calomel electrode was used
as reference and a platinum wire as counter electrode.
Working electrodes were prepared from high purity
nickel (>99.99 wt%). Before use, they were embedded in a
chemically inert resin and mechanically polishing up to
2500 SiC grade. Their active area is equal to 1 cm2.
J Mater Sci (2007) 42:90709074
An aerated 1 M H2SO4 aqueous solution, prepared from
ultra pure reagent, was used as corrosive electrolyte. The
halide salt was C8H17Ph3P+,I and it was prepared in the
laboratory by adding triphenylphosphine to heptane iodine.
The electrochemical instrumentation consisted of a
Taccussel potentiostat-galvanostat PGP 201. Voltamaster 1
software was employed for instrumentation control and
data treatment.
Results and discussion
Figure 1 shows anodic and cathodic polarization curves for
nickel electrode immersed for 1 h at open circuit potential
in 1 M sulphuric acid aqueous media, in the absence and
presence of 103 M of the iodine alkyl triphenyl phos-
phonium. The scan rate has been fixed at 8.33 mV/s and
the temperature at 298 K.
In absence of the halide salt and independently on the
temperature, one notes the presence of four regions on
the polarization curves. The region I represent the
cathodic domain characterized by a charge transfer
process. It can be assigned to the strong concentration of
acid (1 M) and the mobility of the H+ ion that conceals
the contribution of the oxygen in the global kinetics of
the cathodic reaction. In the region II the electrochemical
process at the NickelH2SO4 interface is controlled by
activation kinetic. The extrapolation of this part to the
potential of corrosion permits us calculating the material
corrosion rate. The region III is characterized by the
presence of two oxidation peaks (1a) and (2a) corre-
sponding to the electrochemical generation of nickel
oxides and/or sulfates. Then, the current falls (region IV).
A surface steady state appeared indicating that the
3 I II III
10
2
10 1a
2a
1 be made:
10
2
j/mA/cm
0 (1)
10
-1
10
-2
10
-3 10-3 M (inh)
10
blank
-1000 -500 0 500
E/mV/SCE
Fig. 1 Polarization curves of nickel in 1 M H2SO4 with and without
103 M phosphonium iodine salt
9071
corrosion products formed at the electrode surface al-
lowed a pseudo-passivation of the material.
When the inhibitor is added, we observed an increase of
the cathodic current density especially for region I. Then,
the current density tends to a plateau. Thus, the cathodic
reaction is limited by oxygen diffusion. Finally, the current
density increases corresponding to H+ reduction.
One can conclude that the organic halide salt denotes an
inhibition activity for the cathodic reaction taking place at
the surface of nickel in 1 M H2SO4.
For the anodic domains (regions II, III and IV), one can
notes that the corrosion potential tends to more anodic
value. Indeed, only the second activation peak is observed.
Furthermore, the steady state current decreased for the
entire over potential interval considered.
Hence, one can conclude that organic iodine salt acts as
mixed inhibitor. The observed inhibitive action could be
due to an adsorption phenomenon at the surface of nickel
making a barrier for charge and mass transfer between the
metal and its environment.
To characterize the nickel interfacial behaviour, we
carried out the voltammetric investigation at various tem-
peratures ranging from 308 K to 328 K. The polarization
curves obtained are reported in Fig. 2.
We have measured the corrosion potential (Ecorr) and the
corrosion current densities (jcorr) from the points of inter-
section of the extrapolated anodic and cathodic Tafel
regions. Then, the Inhibition Efficiency percentage (IE%)
and the degree of surface coverage (h) were calculated as
follow:
j0  j1
IE% 100 1
j0
 
j0  j1
h 2
j0
where j0 and j1 are the corrosion current densities obtained
IV in the absence and the presence of the inhibitor.
The results obtained are gathered in Table 1.
From the data of Table 1, the following conclusions can
jss
at constant temperature, IE% increases with the
increase in the concentration of the studied inhibitor.
Such behaviour could be explained on the basis of
increased adsorption of the inhibitor on the metal
surface;
(2) in the presence of the same concentration of each
inhibitor, IE% decreases with the rise in temperature
1000
due to the enhanced effect of temperature on the
dissolution process of nickel in H2SO4 solution and;
or the partial desorption of the inhibitor from the
metal surface.
123
9072 J Mater Sci (2007) 42:90709074

Fig. 2 Polarization curves of 10

3
10
3

298K 308 K
nickel in 1 M H2SO4 at different 2
10 2
phosphonium iodine 10

Log j /mA/cm2
concentrations for various 10
1
1
10
temperatures raging from 298 K

2
j/mA/cm
0
10
to 328 K 0
10
-1
10
-1
-2 10
10
-2
10
-3 10

-4 -3
10 10
-1000 -500 0 500 1000 -1000 -500 0 500 1000
E/mV/SCE E/mV/SCE
3 3
10 10
318 K 328 K
2 2
10 10

1 1
10 10

2
2

j/mA/cm
j/mA/cm

0 0
10 10

-1 -1
10 10

-2 -2
10 10

-3 -3
10 10
-1000 -500 0 500 1000 -1000 -500 0 500 1000
E/mV/SCE E/mV/SCE
-6 -5 -4 -3
0 M (inh); 10 M (inh); 10 M (inh); 10 M (inh); 10 M (inh);

Table 1 Corrosion potentials

T/K [inh]/mol/L Ecorr/mV/SCE jcorr/lA/cm2 %g h
Ecorr, corrosion current densities
jcorr, percentage inhibition 298 0 296 28
efficiencies (% g), and the
degree of surface coverage h at 106 243 13 54 0.54
different temperatures for nickel 105 235 11 61 0.61
electrode immersed in 1 M 10 4
224 5.4 81 0.81
sulphuric acid medium with and
103 56 0.8 97 0.97
without 103 M phosphonium
iodine salt 308 0 82 35
106 11 7 80 0.80
105 1 6 83 0.83
104 1 4 89 0.83
3
10 14 1.5 96 0.96
318 0 74 41
106 3 12 71 0.71
105 3 10 76 0.76
104 3 5 87 0.87
3
10 6 3 93 0.93
328 0 153 59
106 96 47 20 0.20
105 96 46 22 0.22
104 0 25 57 0.57
3
10 7 12 80 0.80

123
J Mater Sci (2007) 42:90709074 9073

The dependence of the corrosion rate on temperature can 1,5x10

-3

be expressed by the Arrhenius equation: 298K Y = 1,22X + 310

-5

2
308K R = 0.998
 
DEa 318K
Logjcorr constant  3 328K
2:303RT 1,0x10
-3

-6
Y = 1.07X + 310
2
where: R =1

C /
-6
Y = 1.04X + 610
DEa: the activation energy of the corrosion reaction; -4
2
R = 0.999
5,0x10
T: the absolute temperature;
R: the universal gas constant. -6
Y = 1.02X + 910
2
Figure 3 shows the Arrhenius plots for the corrosion rate R = 0.999
of nickel in 1 M H2SO4 with and without the presence of 0,
-4 -4 -4 -3
00
103 M of phosphonium iodine slat. 2,50x10 5,00x10 7,50x10 1,00x10
-1
The results give rise to satisfactory straight lines from the C (mol.L )
slopes of which one can calculate the activation energy. DEa
Fig. 4 Langmuir adsorption isotherms on nickel in 1 M H2SO4
values were found to be 19.3 kJ mol1 for the uninhibited solution with and without the presence of 103 M of the phosphonium
solution and 71.1 kJ mol1 in the presence of 103 M of iodine slat
phosphonium salt. The value of DEa for the inhibited solution
is higher than that for uninhibited one, indicating the greater considered that the surface coverage (h) of the inhibitor on
tendency of such compound to react at the surface of nickel, the nickel specimen is related to the concentration (C) of
in 1 M H2SO4 electrolyte [57]. the inhibitor in the bulk of the solution according to
From the above mentioned results it is concluded that Langmuir adsorption isotherm:
the inhibition of nickel corrosion in 1 M H2SO4 solution  
kC
occurs by adsorption of the additive used [8, 9]. h 4
The most commonly used isotherms for studying the 1 kC
adsorption mechanism of an inhibitor on a metal electrode
where k is the equilibrium constant for the adsorption
surface are listed in the literature [1019].
process.
In order to choose the appropriate isotherm model to
Rearranging Eq. 4:
describe the adsorption mechanism of phosphonium iodine
at nickel surface, we plotted C/h versus concentration C  
1
(Fig. 4). C=h C 5
k
A linear relationship is obtained, independently on
the temperature considered. To explain such result, we One can conclude that the absorption of phosphonium
iodine salt on nickel surface in 1 M sulphuric acid medium
occurs according to the Langmuir adsorption isotherm.
2,0
From the intercepts of the straight lines on the C/h axis,
k values could be calculated (Table 2).
1,5 Furthermore, the equilibrium constant (k) for the
adsorption process of the inhibitor could also be related to
the free energy of adsorption DadsG by the relationship:
1,0 Without inhibitor  
log jcorr

1 Dads G
With inhibitor k exp  6
55:5 RT
0,5
The calculated values are gathered in Table 2. Then, we
plotted DadsG against temperature (Fig. 5).
0,0
Independently on the temperature considered, negative
values are obtained for DadsG indicating the spontaneous
3,0 3,1 3,2 3,3 3,4
3 -1 -1
interaction between phosphonium iodine and nickel sur-
10 T (K ) face.
Fig. 3 Arrhenius plots for nickel in 1 M H2SO4 with and without the However, the free energy decreases from 28.8 kJ mol1
presence of 103 M of the phosphonium iodine slat at 298 K to 33.6 kJ mol1 at 318 K. A linear relationship

123
9074 J Mater Sci (2007) 42:90709074

Table 2 Values of k and DadsG for the adsorption of the phospho- could be attributed to the competition adsorption/desorp-
nium iodine salt on the nickel surface in the temperature domain 298 tion of the additive on nickel surface for T 318 K.
328 K
T k (kJ mol1) log k DadsG = RT log k
(kJ mol1) Conclusion
298 111 11.62 28.8
308 167 12.02 30.8 The results of these experiments revealed that:
318 333 12.72 33.6 (1) phosphonium iodine decreases the corrosion rate of
328 33.3 10.41 28.4 nickel in 1 M H2SO4 electrolyte;
(2) at constant temperature, IE% increases with increas-
ing inhibitor concentration;
(3) at the same concentration, IE% decreases with
-28 increasing temperature;
(3) inhibition occurs by adsorption according to the
-29 Y = -0.24X + 43.42 Langmuir isotherm;
2
R = 0.999
(4) inhibitor desorption increases for T 318 K.
-30
adsG (kJ.mol )
-1

-31 Acknowledgements The authors would like to thank AUF (Agence

Universitaire de la Francophonie) for financial support.
-32

-33 References
-34 1. Morad MS (2000) Corros Sci 42:1307
2. Khaled KF (2004) Appl Surf Sci 230:307
295 300 305 310 315 320 325 330 3. Niass SO, Touhami ME, Hajjaji N, Srhiri A, Takenouti H (2001)
() J Appl Electrochem 31:85
4. Said F, Souissi N, Dermaj A, Hajjaji N, Triki E, Srhiri A (2005)
Fig. 5 Adsorption standard free energy (DadsG) versus temperature Mater Corrosion 56(9):619
5. Saleh JM, Al Haidari YK (1989) Bull Chem Soc Jpn 62:1237
could be used to describe DadsG variation versus T in this 6. Abd El Nabey BA, Kamis E, Ramadan MS, El Gindy A (1996)
interval. Then, other thermodynamic functions can be Corrosion 52:671
calculated through the following equation: 7. Abd El Aal EE, Zakria W, Diab A, Abd El Haleem SM (1999)
J Chem Technol Biotechnol 74:1061
8. Dinnappa RK, Mayanna SM (1982) Corrosion 38:525
Dads G Dads H  TDads S 7 9. Rudresh HB, Mayanna SM (1977) Surf Technol 6:139
10. Langmuir I (1918) J Am Chem Soc 40:1361
where DadsH and DadsS the enthalpy and the entropy of 11. Frumkin AN (1925) Z Phys Chem 116:466
adsorption. They were equal to +43.4 kJ mol1 and 12. Hill TL (1952) J Chem Phys 20:141
13. de Boer JH (1953) In: The dynamical character of adsorption.
+240 J K1 mol1, respectively.
Oxford University Press, Oxford
In the temperature domain considered, both thermody- 14. Parsons R (1964) J Electroanal Chem 8:93
namic functions values are positive reflecting for phos- 15. Damaskin BB, Petrii OA, Batrakov VV (1971) In: Adsorption of
phonium iodine adsorption on nickel surface: organic compounds on electrodes. Plenum Press, New York,
p 86, 94 and 247
(1) an endothermic behaviour; 16. Kastening B, Holleck L (1965) Talanta 12:1259
(2) an increase of the molecular disorder. 17. Bockris JO_M, Swinkels DAJ (1964) J Electrochem Soc 111:736
18. El-Awady AA, Abd-El-Nabey BA, Aziz SG (1992) J Electro-
Then, the free energy exhibited a singular behaviour. In chem Soc 139:2149
fact, it increases reaching 28.4 kJ mol1. Such result 19. Khamis E, Hosny A, EL-Hadary S. (1995) Affinidad 95:456

123
J Mater Sci (2007) 42:90759082
DOI 10.1007/s10853-007-1894-1
Laser-induced anisotropy in Ag+-doped glasses
Arashmid Nahal Fariba Moslehirad
Received: 22 January 2007 / Accepted: 29 May 2007 / Published online: 27 July 2007

Springer Science+Business Media, LLC 2007
Abstract An induced optical anisotropy is observed as a
result of interaction of a high-power CW Ar+ laser beam,
with silver-ion-exchanged glasses. We have shown that the
absorption of the beam by the thin layer of Ag+ produces a
temperature gradient resulting in a radial stress on the
surface of the sample. The induced anisotropy makes the
sample behave as a thin uniaxial optical medium with axis
along the direction of the beam propagation. For the
polarized light, the induced anisotropy restricts the appli-
cation of micro-lenses, which are made by this method.
The average refraction index of the interaction area is
measured.
Introduction
It is well-known that placing an anisotropic absorbing
crystal between two crossed polarizers and illuminating it
with a white light, results in the appearance of a charac-
teristic image with isogyres and isochromates, called the
conoscopic pattern [1].
Induced anisotropy as a thermal effect of the interaction
of light (laser beam, flash lamp,...) with crystals and some
other amorphous materials, leading to formation of cono-
scopic patterns, has already been observed and studied by
many researchers [214]. In all above-mentioned cases,
especially the laser materials, the light induced anisotropy
A. Nahal (&)
Department of Physics, University of Tehran, Karegar Shomali
Ave., Tehran 14399-55961, Iran
e-mail: nahal@khayam.ut.ac.ir
F. Moslehirad
Qazvin Branch, Azad University, Daneshgah Ave.,
P.O. Box 34185-1416, Qazvin, Iran
causes the interacted medium to behave as an optically
uniaxial one, with the optical axis along the direction of the
beam propagation. It is possible to induce similar effects by
unpolarized light [15]. Interaction of Ag-containing thin
layers with polarized laser beam could make the material
anisotropic, leading to a dichroism [1618], which is out of
the scope of the present article.
In laser materials such as Nd+: YAG rod, the tempera-
ture gradient created by the flash lamp generates a
mechanical stress in the material, which leads to a thermal
strain resulting in the appearance of a birefringence [68,
14]. Irradiating crystals such as Tb3Ga5O12 or semicon-
ductor-doped glasses by a laser beam also results in gen-
eration of stress birefringence (Dn) [4, 5, 10, 12, 13, 19],
due to the thermal gradient induced by the Gaussian laser
beam. In above-mentioned cases if one places the inter-
acting medium between two crossed polarizers and illu-
minate it with a white light, a characteristic conoscopic
pattern could be observed. This pattern means that a dis-
tinguished direction is induced (z-direction, along the
propagation of the laser beam), which coincides with the
dielectric principal axis ez i.e. ex ey 6 ez . Thus, the
medium becomes optically uniaxial similar to the case of a
natural crystal [1]. The magnitude of Dn increases with the
radius, by quadratic law [5]. Therefore, a linearly polarized
light passing through the sample will experience a phase
difference between the two components along nr and n/
(the radial and tangential photoelastic changes in the index
of refraction, respectively). Thus, except for points located
along the directions of the polarizer and the analyzer (x and
y) for which the polarization remains linear, at each point
of the transverse section, a polarization change yields a
spatial elliptical polarization pattern.
We have observed the same phenomenon during the
interaction of a CW Ar+ laser beam (Pmax = 8 W), working
123
9076
in the multiline regime with Ag+-doped glasses. Absorption
of the laser beam energy by the thin layer of the embedded
silver ions in the glass matrix increases the temperature of the
incident point of the laser beam on the sample. The Gaussian
intensity distribution of the beam results in establishing a
temperature gradient across the incident point leading to the
appearance of a stress birefringence in the sample. We have
found that if one put the irradiated sample between two
crossed polarizers, using white light, then a characteristic
conoscopic pattern would be observed. That is, the irradiated
ion-exchanged glass behaves as uniaxial optical medium,
with axis along the beam propagation. This effect has not
been studied for silver-ion-exchanged glasses neither
reported for thin layer of silver ions in a glass matrix.
Doping a glass matrix with metal clusters produces a
nonlinear optical medium, which is promising for opto-
electronics applications like optical switches and slab
waveguides. One of the most investigated doped glasses is
the Na+Ag+ ion-exchanged one. Studying the interaction
of a high-power laser beam with Ag+-doped glass develops
our knowledge about the limits of the above-mentioned
optical material, and also the temporary and permanent
physical changes, which happen during and after the
interaction. Here we report one of the effects (the induced
optical anisotropy), which we have observed. There are
other effects, as the results of the interaction of a high-
power laser beam with Ag+-doped glass, which we have
reported elsewhere in detail [17, 20, 21].
Our experimental results show clearly that the micro-
lenses produced by irradiating the ion-exchanged glass
using a high-power laser beam [22, 23] are not suitable for
use with polarized white light, due to the induced optical
anisotropy in it. This important point should be considered
during the production of above-mentioned micro-lenses.
Local heating of doped glasses or polymers by a laser beam
could produce a micro-lense. The formation of the micro-
lens occurs, due to the local melting of subsurface layer of
the glass. The glass melting causes by an increase of the
light absorption with temperature. Thus, when laser radi-
ation heats the glass the light penetration depth decreases,
and in spite of the thermal diffusion the process of heat
deposition becomes increasingly localized at the surface of
the glass. This thermal runaway causes the host glass to
melt locally under laser radiation. Because of the lower
density of the molten glass compared with that of the solid
glass, the material wells out and solidifies forming a micro-
lense [22, 23]. Micro-lenses have many applications such
as microlense array for telescope application [24], micro-
lense-based projection displays [25] and etc.
The organization of the paper is as follows. In
Sect. Experiment, the experimental setup would be ex-
plained in detail the results of the experiments are reported
in this section. The Sect. Discussion includes a
123
J Mater Sci (2007) 42:90759082
preliminary explanation of the obtained results. The con-
clusion of report is given in the last section.
Experiment
Sample preparation
The samples are prepared using the well-known ion-
exchange method [26]. For this, slides of commercial soda-
lime glasses (20 30 0.8 mm) were placed in mixed
molten salts of NaNO3/AgNO3 (96/4 wt.%) in a tempera-
ture of 400 C for 4 h, in the atmospheric pressure. In the
ion-exchange process some of the sodium ions of the glass
matrix would be exchanged by some of the silver ions of
the molten mixed salt, due to the diffusion process. The
temperature and duration time of the ion-exchange process
determines the concentration, diffusion depth in the matrix,
and even the size of the formed silver ionic clusters in the
glass matrix. The composition of the glass also affects the
above-mentioned parameters. As a result, a thin layer of
ion-exchanged glass would be formed in the glass
substrate. In our experiments the average thickness of the
ion-exchanged layer was about 180 lm, measured by a
microscope equipped with a micrometer. Then the prepared
samples were irradiated by a CW high-power Ar+ laser
beam (Pmax = 8 W), working in multiline regime.
The ion-exchanged glass has a brownish color with a
continuous absorption spectra in visible region with a peak
in the UV region at about wavelength ka = 310 nm
(Fig. 1), which is related to very small ionic silver clusters
[27, 28]. The color of the incident point of the laser on the
sample becomes yellowish after the interaction, with an
orange colored periphery, which is made out of silver
neutral clusters.
Interaction with the laser beam
The setup shown in Fig. 2 was used for studying the
thermal effect of the laser beam on the optical properties of
the ion-exchanged glass in real time.
The sample was placed in front of the Ar+ laser with a
little tilt to prevent back-reflection of the beam from the
sample towards the laser cavity. An expanded linearly
polarized HeNe laser beam and a polaroid plate, placed
after the sample, were used as a probe (Pprobe = 1 mW) to
study the induced-optical anisotropy and any optical path
changes during the interaction, in real time. The transmit-
ted probe beam was used to study the induced anisotropy,
and the reflected beams from both sides of the sample were
used to study the optical path changes (Fig.2).
Our experiments show that by increasing the power of
the Ar+ laser beam, the temperature of the incident point of
J Mater Sci (2007) 42:90759082
Fig. 1 The absorption spectra of an ion-exchanged glass: (1) before
the interaction; (2) after the interaction. The peak around the
wavelength 310 nm is related to existence of very small ionic silver
clusters. The peak around the wavelength 430 nm is related to the
surface palsmon resonance of the produced neutral silver clusters
the beam on the sample rises. Greater than a power
threshold (Pth
4 W, for ion-exchanged samples at
400 C for 4 h) a conoscopic pattern appears on the screen
behind the polaroid for the transmitted polarized probe
beam (Fig. 2, down-left), and at the same time the inter-
ference pattern of the reflected probe beam becomes
twisted at the place of the incident point (Fig. 2, down-
right). Simultaneously, the reflected argon laser beam
expands rapidly and at the end of the interaction forms
co-centered rings on the screen behind the laser (Fig. 2,
top-left). The same co-centered rings have been observed
by other researchers during the interaction of laser beams
with photorefractive crystals [29] and liquid crystals [30],
as a thermal effect of the interaction. This pattern means
that a thermal lens is formed. All the real-time observations
of this experiment (conoscopic pattern, twisted interference
pattern, and the co-centered rings pattern) indicate that, as
a result of the thermal effect due to the interaction of the
CW Ar+-laser beam with the thin layer of the silver-ion-
exchanged glass, a thermal lens and an optical anisotropy
are induced into the sample.
The same conoscopic patterns, as those obtained in the
real-time study, are observed when we put the sample
(after the interaction) between two crossed polarizers
which are illuminated with a white light (Figs. 3, 4a). The
conoscopic pattern is also observed, for the case, when the
polarizer and analyzer are parallel, but the dark and bright
parts are complementary relative to the case when the
polarizers were crossed (Fig. 4b). As it can be seen from
the Fig. 4b, the isochromates for the blue light is placed
closer to the center, relative to the isochromates for the red
9077
light. An orange circle also is seen (Fig. 4b), which is
related to the silver nanoparticles on the surface of the
sample [17, 20]. The produced micro-lens can be seen in
Fig. 4b. From Fig. 4a, also it can be seen that, when the
polarizer and the analyzer are crossed the first and third
quadrants of the conoscopic pattern have a dominant blue
color and the second and forth quadrants of the conoscopic
pattern have a dominant yellow color. According to the
Refs. [1, 31] the above-mentioned blue and yellow quad-
rants in the characteristic pattern are observed for optically
positive uniaxial medium. Because of the remained thermal
stress, the produced micro-lenses by this method are very
unstable mechanically and would be cracked even with
very little pressure on the sample. Thus, one should be
careful with the samples after the interaction.
Measurement of average index of refraction of the
interaction area
It would be useful if we could have some idea about the
index of refraction of the interaction area. As the glass
matrix melts during the interaction, and at the same time
the silver ions reduce to the neutral ones, and also simul-
taneously the produced silver nanoparticles move toward
the periphery, so that the index of refraction of the inter-
action area changes. On the other hand, the thermal effect
of the beam changes the surface of the sample and pro-
duces a lens-like surface profile. Therefore, twisting of the
fringes of the interference pattern, shown in down-right of
Fig. 2, includes information about both the index changes
and also modification of the profile of the surface. In order
to measure the average index of refraction of the interac-
tion area we use Fig. 2 (down-right). A schematic draw is
shown in Fig. 5 to explain how the average index of
refraction is measured. From Figs. 2, 5 and knowing the
condition for constructive interference for the lights
(HeNe laser: probe) which have reflected from both side
of the sample [32] we can write:
2ni x k1 constructive interference 1
where x is the thickness equivalent to a difference of height
for neighboring fringes, k1 = 632.8 nm is the wavelength
of the probe laser beam and ni is the index of refraction of
the sample within the interacted area. From Fig. 5 if we
take the average slope of the surface of the interacted area
equal to b, and the inter-fringes for neighboring fringes as
Dl, then we can write:
x
tan b 2
Dl
for the small angle: tan b
b, then from Eqs. 1 and 2 we
have:
123
9078 J Mater Sci (2007) 42:90759082

Fig. 2 The real-time study

setup. After the interaction the
reflection of the Argon laser
beam from the interaction area
forms the co-centered circular
interference fringes (top-left).
The conoscopic pattern appears
on the screen for the transmitted
probe beam (down-left), and for
the reflected probe beam the
fringes of the interference
pattern would be twisted after
the interaction (down-right).
The conoscopic patterns are
shown for three cases when (a)
~ A;
P? ~ (b) \P; ~ 45 ; and
~ A
~ ~
(c) PjjA, respectively

Fig. 3 The setup for studying

the conoscopic patterns of the
irradiated sample with white
light

k1 on a test plate (roughness: k/20), and illuminated it with a

b 3
2ni Dl
Now if we rewrite the relation of Eq. 3 for p fringes,
which occupy length l on the sample, we obtain:
pk1
b 4
2ni l
On the other hand we can measure the height of the
produced microlens, using the interferometric methods. In
order to do this, we placed the curved surface of the sample
standard sodium lamp (k2 = 589.3 nm). The same twisted
interference pattern, as in Fig. 2, would be formed, but this
time for the air layer between the sample and the test plate.
In this case we have the same relation as Eq. 4, but with q
fringes in the length l for the wavelength k2, and nair = 1:
qk2
b 5
2l
Then from Eqs. 4 and 5, the average index of refraction
of the interaction is given by:

123
J Mater Sci (2007) 42:90759082 9079

Fig. 4 The conoscopic patterns

of the irradiated sample, using
the setup shown in Fig. 3. (a)
The conoscopic pattern for the
case whenP?~ A,~ and (b) for the
~ It is seen that
~ A.
case whenPjj
the isochromates for the blue
color have smaller radius in
comparison with the red one

Fig. 5 A schematic drawing for

showing the parameters, used to
measure the average index of
refraction of the interacted area
on the sample. b is the slope of
the modified surface, Dl is the
distance between two
neighboring fringes, l is the
distance occupied by p number
of fringes, Dh is the height of
the produced thermal lens, and x
is the height difference between
two neighboring fringes

pk1 As shown in Fig. 1, the Ag+-doped glass has an

ni 6 absorption in whole visible spectrum, with a peak in
qk2
In our measurements we obtained p = 13, q = 9, with
k1 = 632.8 nm, and k2 = 589.3 nm, respectively. Then
relation of Eq. 6 gives: ni = 1.551. Using the Brewsters
angle measurement for HeNe laser beam for
k = 632.8 nm the index of refraction of the non-interacted
area on the sample is: nn = 1.549. Thus, dn = ni
nn
0.002. Our experimental results show that the average
index of refraction of the interaction area has a little
increase in its value relative to those parts which are not
irradiated. This increase should be related to the reduction
of silver ions to the neutral ones, and also to the concen-
tration of silver nanoparticles near the periphery of the
interaction area (orange ring in Fig. 4b) [20].
Using the interfrometric method the height of the gen-
erated thermal lens on the sample (4% AgNO3, 400 C,
4 h) was measured equal to Dh
6 lm for Pmax = 8 W.
Discussion
In this section we give a preliminary explanation for our
experimental observations, which are reported in this article.
wavelength ka = 310 nm, leading to the brownish color for
the ion-exchanged glass. This spectrum allows the sample
to absorb the Ar+ laser radiation, which results in heating
the sample, locally, at the incident point of the laser beam
and generation of neutral silver nanoparticles, with an
absorption band peak about kb = 410 nm [17], resulting in
the yellowish color of the interaction area. At the same
time, because of the increase in the temperature, the
interaction area melts and a micro-lens forms. Simulta-
neously, the produced silver nanoparticles move toward the
periphery of the interaction area [17, 20, 22, 23]. The
pushed silver nanoparticles make an orange colored ring on
the sample (Fig. 4b). The melting of the glass matrix at the
incident point of the laser generates a radial thermal gra-
dient from center toward the periphery of the interaction
area, which results in inducing a stress birefringence.
Hence, in the interaction area both the index of refraction
and the thickness of the medium will be changed. The
whole thermal effect makes the irradiated sample act as a
thin uniaxial anisotropic optical medium, with optical axis
along the direction of beam propagation. Therefore, plac-
ing the interacted sample between two crossed polarizers
results in appearance of the conoscopic pattern.

123
9080 J Mater Sci (2007) 42:90759082

p
Let us have a brief review for propagation of light in an I I1 I2 2 I1 I2 cos d 10
optical anisotropic medium, which results in forming the    
 !2  !2
conoscopic patterns. When a light beam under an incident where I1  OF  , I2  OG  . Using Eqs. 8, 9, and 10 the
angle h1 enters into an anisotropic medium with thickness intensity distribution right after the analyzer would be
d, the beam is broken up into two beams with different equal to:
refractive indices n, n, where k0 nk0 , k00 nk0 ; with k the
wavelength in the vacuum. Hence, they propagate in the I d
cos2 v  sin2 /sin2/  vsin2 11
medium with different velocities, which results in retar- I 2
dation between them. That is, the exiting beams from the  2
 !
medium according to the Fig. 6a, have a phase difference where I  OE  .
~ A
For the case when P? ~ (v p we have:
equal to (see the Appendix A): 2

0 s1 I d
q  2
2pd @no sin h 1
sin2 2/sin2 12
d n2e  sin2 h1  no 1  A 7 I 2
k ne no
~A
and for the case when Pjj ~ (v = 0) we have:

where no and ne are the ordinary and extraordinary indices

I d
of refraction, respectively. 1  sin2 2/sin2 13
I 2
Leaving the medium, the two light beams with the phase
difference k = 632.8 nm could interfere. Now, if one pla-
It is clear that, in the two above-mentioned cases the
ces the anisotropic medium between a crossed polarizer
patterns are complementary to each other. Thus, the rela-
and an analyzer (P ~ respectively), with general angle
~ and A,
tion of Eq. 11 gives the intensity distribution of the
v between them, the characteristic intensity distribution
conoscopic patterns for the case when the anisotropic
could be observed. Let us take D~o for vibration direction of
optical medium is placed between the polarizer and the
~
the ordinary beam and De for vibration direction of the
analyzer. In the Ref. [4], authors give a relation from
extraordinary beam, with D ~o (Fig. 6b). Taking !
~e ?D OE for
which, using the parameters of the conoscopic pattern and
amplitude of the incident beam, from Fig. 6b:
the sample, one can estimate the induced birefringence Dn:
! !
OB OE cos / Nk
! ! 8 Dn 14
OC OE sin / Leff
~ and D
where / is the angle between P ~o . The analyzer just
! where, k is the wavelength of the light, N is the number of
passes those components that are parallel to OA : the fringes in the conoscopic pattern and Leff is the thick-
ness of the anisotropic medium. In our case:
! !
OF OB cos/  v Leff = 180 lm, k = 550 nm (average wavelength for the
! ! 9
OG OC sin/  v white light) and N = 1 (Fig. 4). Thus, from Eq. 14 we
obtain Dn = 0.003. It should be mentioned that the diffu-
The two output beams could interfere, resulting in the sion of silver ions into the glass substrate produces a thin
intensity distribution: layer, for which the concentration of silver ions and con-
sequently the refractive index follow, in general, an
exponential dependence [33, 34]. Thus, it is not easy to
specify a sharp boundary between the doped layer and the
substrate. Because of that, an effective thickness is intro-
duced.
On the basis of the relations between Eqs. 12 and 13,
using parameters of our samples we calculated the cono-
scopic patterns for both cases when P? ~ A ~ and Pjj ~ The
~ A.
results are shown in Fig. 7. It can be seen that the exper-
Fig. 6 (a) Refraction of a light beam entering into a birefringent imental results (Fig. 4) and the calculation (Fig. 7) are
medium. h1 is the incident angle, d is the thickness of the medium. (b) consistent.
The diagram used to calculate the intensity distribution for the sample
From Eq. 11 one can see that there is a dependence of
placed between the polarizer and the analyzer. v is the angle between
the polarizer and the analyzer, / is the angle between the polarizer intensity distribution of the conoscopic pattern on the
and the direction of Do wavelength of the light. We have drawn the interference

123
J Mater Sci (2007) 42:90759082 9081

of interaction with a high-power Ar+ laser beam. Our

Fig. 7 The simulated conoscopic patterns for an irradiated sample,
for two cases when the polarizer and the analyzer were perpendicular
and parallel to each other, respectively. For the calculations the
experimental results, measured in our samples, were used
experiments show that the thermal effect of the interaction
makes the glass matrix, with a thin layer of silver ions,
behave as an uniaxial optical medium, with axis along the
direction of beam propagation. The average index of
refraction of the interaction area was measured, and it is
shown that the index is increased. Our experimental results
show that the induced anisotropy restricts the application of
the microlenses, which are made by the interaction of
pulsed laser beams with the ion-exchanged glasses [22, 23],
for the polarized white light.
Acknowledgments Authors are grateful to Prof. M.T. Tavassoly
from University of Tehran, for valuable discussions. We also appre-
ciate assistance of Mr. J. Mostafavi-Amjad in sample preparation,
from Institute for Advanced Studies in Basic Sciences (IASBS). Some
parts of Experiments are done in IASBS. A. Nahal thanks Interna-
tional Center for Theoretical Physics (ICTP, Italy) for its support. We
also are grateful to Dr. M. Nouri from University of Tehran for
correcting the English language of this paper.

Appendix A

According to the Fig. 6a the phase difference equals to:

2p 00 2p 00 2p
d AB B C  0 AB0 a
k00 k k
with AB00 cosdh00 , AB0 cosd h0 , B00 C dsinh1
0 0
tan h  tan h and using the Snells law with n1 = 1 for
air, we have:

2pd 00
d n cos h00  n0 cos h0 b
k

Fig. 8 The calculated isochromates for our samples using the relation If we take the AB beam as the ordinary beam, then:
of Eq. 11. The blue isochromate (bold circle) has smaller radius in
comparison to the red one (the dashed circle). Their values are n0 : no ; h0 : ho
normalized to one, for simplicity. This calculation is in good
s
 
accordance to the experimental results (Fig. 4b)
0 0 sin h1 2
sin h1 no sin ho ) n cos h no 1 
isochromates for both blue and red lights (Fig. 8) using the no
parameters of our samples. It can be seen that the iso- c
chromates of blue light have smaller radius relative to the
isochromates of red light for the same order of interference. For the extraordinary beam we can write:
Our experimental results (Fig. 4b) show the same result for
the irradiated Ag+-doped glass by the Ar+ laser beam. As it h00 : he ) n00 sin he sin h1 d
can be seen from Fig. 4b the blue fringes are closer to the
center compared to the red ones. If the optical axis is normal to the surface, then the
relation between n and ne can be obtained from the Fres-
nels formula:
Conclusion
no ne
n00 q e
In this article, we have reported the observation of an 2
n2o sin he n2e cos2 he
induced optical anisotropy in Ag+-doped glasses as a result

123
9082 J Mater Sci (2007) 42:90759082

Therefore, using Eqs. d, e and Fig. 6a we obtain: 12. Chen X, Calemczuk R, Salce B, Lavorel B, Akir C, Rajanoha L
(1999) Solid State Commun 110:431
13. Chen X, Lavorel B, Dreier T, Genetier N, Misserey H, Michaut X
n2o n2e  n2o sin2 h1
cot2 he f (1998) Opt Commun 153:301
n2e sin2 h1 14. Koechner W, Rice DK (1970) IEEE J Quant Electron QE-
6(9):557
After some simple trigonometric calculation and using 15. Tikhomirov VK, Elliott SR (1994) Phys Rev B 49:17476
16. Kolobov AV, Lyubin VM, Tikhomirov VK (1992) Philos Mag
Eqs. d, e, and f the relation for n00 cos h00 could be written Lett 65:67
as: 17. Nahal A, Khalesifard HRM, Mostafavi-Amjad J (2004) Appl
q Phys B 79:513
no 18. Nahal A, Miloslavsky VK, Ageev LA (1998) Opt Commun
n00 cos h00 n2e  sin2 h1 g 154:234
ne
19. Chen X, Lavorel B, Boquillon JP, Saint-Loup R, Jannin M (1998)
Solid-State Electron 42:1765
Substituting Eqs. c and g in b we end up with: 20. Nahal A, Mostafavi-Amjad J, Ghods A, Khajehpour MRH, Re-
0 s1 ihani SNS, Kolahchi MR (2006) J Appl Phys 100:053503
q  2 21. Nahal A, Khalesifard HRM (2007) Opt Mater 29:987
2pd @no sin h 1 A
d n2e  sin2 h1  no 1  h 22. Kaganovski Yu, Antonov I, Bass F, Rosenbluh M (2001) J Appl
k ne no Phys 89:8273
23. Antonov I, Bass F, Kaganovski Yu, Rosenbluh M (2003) J Appl
Phys 93:2343
24. http://www.aqua.epfl.ch./projects.html
25. http://www.electronicimaging.org/program/04/
References 26. Najafi SI (1992) Introduction to glass integrated optics. Artech
House, Boston
1. Born M, Wolf E (1964) Principle of optics. Pergamon, Oxford 27. Nahal A, Rezae N (2005) In: Proc. 12th Conf. on Optics and
2. Chen X, Berger H (1999) J Phys: Condens Matter 11:7377 Photonics, Shiraz Univ., Iran, pp 537539 (In Persian)
3. Froehly L, Verrier I, Froehly C, Burn G, Veillas C (1999) Opt 28. Kreibig U, Vollmer M (1995) Optical properties of metal clusters.
Commun 167:27 Springer, Berlin
4. Petrov DV, Gomes ASL, De Araujo CB (1994) Phys Rev B 29. Horowitz M, Daisy R, Werner O, Fischer B (1992) Opt Lett
50:9092 17(7):475
5. Chen X, Gonzalez S (1998) Appl Phys B 67:611 30. Durbin SD, Arakelian SM, Shen YR (1981) Opt Lett 6(9):411
6. Foster JD, Osternik LM (1970) J Appl Phys 41(9):3656 31. http://www.tulane.edu/~sanelson/geo1211/interference_of_light.
7. Sims SD, Stein A, Roth C (1967) Appl Opt 6(3):579 htm
8. Eichler HJ, Haase A, Menzel R, Siemoneit A (1993) J Phys D: 32. Pedrotti FL, Pedrotti LS (1993) Introduction to optics. Prentice
Appl Phys 26:1884 Hall, New Jersey
9. Ayras PH, Friberg AT, Kaivola MAJ, Salomaa MM (1999) Appl 33. Takeda S, Yamamoto K, Matsumoto K (2000) Non-Crystalline
Opt 38(25):5399 Solids 265:133
10. Chen X, Chaux R (1999) Opt Commun 171:119 34. Martin M, Videau JJ, Canioni L, Adamietz F, Sarger L, Le Flem
11. Sims SD, Stein A, Roth C (1966) Appl Opt 5(4):621 G (2000) Appl Opt 39:435

123
J Mater Sci (2007) 42:90839091
DOI 10.1007/s10853-006-0957-z
Study on the non-isothermal crystallization behaviors
of PA6/silica nanocomposites prepared by the solgel process
Caixian Zhao Ping Zhang Shaorong Lu
Jiangping He Xiayu Wang
Received: 26 May 2005 / Accepted: 8 September 2006 / Published online: 20 July 2007

Springer Science+Business Media, LLC 2007
Abstract A organicinorganic PA6/partially functional-
ized silica hybrid nanocomposites was prepared by capro-
lactam polymerized and silica nanoparticles in situ
gernerated through condensation of partially functionalized
silicic acid, which was prepared through hydrolysis of
tetraethoxysilane (TEOS) and 3-glycidoxypropyltrimeth-
oxysilane (GPTES), in one process. The non-isothermal
crystallization behaviors of pure PA6, PA6/unfunctional-
ized silica (PA6S) and PA6/functionalized silica (PA6FS)
hybrid materials were investigated by differential scanning
calorimetry (DSC). Two methods, namely, the Avrami and
the Mo, were applied to describe the crystallization process
of these three materials. DSC results showed that the
crystallization rates of PA6FS is faster than that of pure
PA6 and PA6S at the same cooling rate, the nano-silica
particles functionalized by GPS in hybrid material acted as
more effective nucleation agents, and the crystallization
rates of all these samples increased with the cooling rate
increasing. XRD results indicated that the addition of
functionalized nano-silica particles favored the formation
of the c crystalline form.
C. Zhao
S. Lu
J. He
X. Wang (&)
Institute of Polymer Science & Engineering, Xiangtan
University, Xiangtan 411105 Hunan Province, P.R. China
e-mail: wxy@xtu.edu.cn
C. Zhao
P. Zhang
S. Lu
J. He
X. Wang
Key Laboratory of Advance Materials and Rheological
Properties, Ministry of Education, Xiangtan 411105 Hunan
Province, P.R. China
C. Zhao
P. Zhang
Insitute of Fundamental Mechanics and Material Engineering,
Xiangtan University, Xiangtan 411105 Hunan Province,
P.R. China
Introduction
In recent years, polymer-based organicinorganic hybrid
material, which considered as innovative advanced mate-
rials, has gained increasing attention in the field of material
science [19]. In order to prepare hybrid materials with the
size of inorganic particles in the nanoscale, one of the most
promising composite systems is the hybrid based on the
solgel process, which is a convenient method for prepa-
ration of hybrid materials from alkoxysilyl group contain-
ing materials via continuous reaction steps of hydrolysis
and condensation catalyzed by acidic or basic catalysts [1,
2, 5, 9]. From our laboratory, we reported a novel method
[9], which nanoparticles in situ generated though hydro-
lysis and condensation of inorganic acid and polymeriza-
tion of caprolactam in one process, of prepared PA6/
inorganic nanoparticles. In our previous work [10], we
prepared PA6/functionalized silica hybrid nanocomposites
(PA6SF) by caprolactam and partially functionalized silicic
acid, which prepared by hydrolysis of TEOS and GPTES,
in one process, which showed dramatic increases in
mechanical properties and heat resistance. In this article,
we would like to focus on the crystallization behaviors of
PA6FS hybrid nanocomposites.
It is well known that the crystallization process of a
polymer plays an important role in the production of
composites, and it also influences directly the properties of
the material. The quality of a crystal involves its mor-
phology, crystalline structure and crystallization degree,
which influence greatly the physical properties and chem-
ical properties of the nanocomposites. The behavior of
thermoplastic semi-crystalline polymers during non-iso-
thermal crystallization from the melt is of increasing
technological importance, because these conditions are the
closest to real industrial processing conditions.
123
9084
There are many of papers on studying the crystallization
kinetics of PA6 nanocomposites [1113], but no publica-
tion has been found concerning the crystallization kinetics
of the PA6/silica hybrid nanocomposites made by func-
tionalized silica nanoparticles in situ generated and cap-
rolactam polymerized in one process. This study
investigated the crystallization behavior of these PA6/
functionalized silica hybrid nanocomposites under non-
isothermal conditions. The Avrami method and the Mo
method were used for the analysis. XRD was also used to
clarify the crystalline structure of PA6FS hybrid nano-
composites obtained under different cooling conditions.
Experimental
Materials
e-caprolactam was purchased from Yueyang Chemical
Plant (China) without further purification. 3-glycidoxy-
propyltrimethoxysilane (GPTES) was purchased from
Chemical Reagent Co. of wuhan University (China) and
was purified by distillation before use. Tetraethoxysilane
(TEOS) and ethanol (both chemical reagent grade), were
ordered from Xilong Chemical Fractory, Guangdong
(China).
Preparation of PA6FS hybrid nanocomposites
The PA6FS hybrid nanocomposites used in this study was
prepared in three steps as follows: (1) Into a 500 mL three-
necked round-bottomed flask, which was equipped with a
Scheme 1
Si(OR)4 +
CO(CH2)3Si(OR)3
O O
Silica HO
CH CH2
O OH
123
J Mater Sci (2007) 42:90839091
N2 inletoutlet and a cooler, 117 mL TEOS, 190 mL
ethanol and 13 g GPTES were added. After vigorous stir-
ring with a magnetic at room temperature for 15 min,
36 mL 0.05 M hydrochloric acid was added. The stirring
was carried out for about 30 min under ambient condition
and a transparent solution was obtained; (2) 3,000 g e-
caprolactam, the solution obtained by step (1) and 1 g
antioxidant were added into a reactor equipped with a
mechanical stirrer. The mixture was heated to and held at
383 K for 12 h under vacuum; (3) 80 g distilled water as
the initiator was added into the reactor, the mixture was
heated to and held at 453493 K for 1 h and then heated at
538 K for 5 h at 0.5~0.8 MPa in a N2 atmosphere, then at
0~0.06 MPa for about 1 h. The material was extruded to
produce cylindrical extrudates by N2, followed by im-
mersed immediately in a cold-water (about 293 K) and
pelletized with an adjustable rotating knife, located after
water bath, into 5 mm length. Then these granules were
washed with water at 357 K for 10 h. The chemical reac-
tions of the synthesis of PA6FS hybrid nanocomposites are
shown in Scheme 1. Pure PA6 and PA6S (prepared by
reported [9]) hybrid nanocomposites were also prepared
under the same condition.
Differential scanning calorimetry measurement
The non-isothermal analyses were carried out using a TA
DSC-Q10 differential scanning calorimeter thermal ana-
lyzer. The samples were heated to 523 K at a rate of
20 K/min under nitrogen atmosphere, then kept for 5 min
at this temperature to eliminate the heat history before
cooling at a specified cooling rate. Constant cooling rate
O Si
ethanol,H2O O Si O
O
CO(CH2)3 Si O Si Si
H+ O O O
Si
Si O
( Silica CH CH2 )
O
O Silica CHCH2NH
O
OH
C(CH2)5NH C(CH2)5NH2
Silica CHCOCO
H
Silica C CH2
OH N
O C
J Mater Sci (2007) 42:90839091
2.5, 5, 10, 15, 20 and 40 K/min were applied. The ther-
mograms corresponding to the heating and cooling cycles
were recorded and analyzed to estimate the non-isothermal
crystallization kinetics and crystallinity degree.
Transmission electron microscope (TEM) observation
Thin plates were cut by ultramicrotomy from the systems
and observed under a TECNAI G2 20 Transmission Elec-
tron Microscope (TEM) at a high voltage of 120 kV.
X-ray diffraction (XRD)
To eliminate the influence of the specimen thickness, about
100 lm films of PA6 and PA6FS were prepared as reported
[8]: The granules of PA6 and PA6FS were put between two
glass slides, and were heated in oil-bath up to 523 K. After
the granules entirely melted, a load was applied on the
surfaces of the slides to press the melted materials into thin
films. These films were kept on 523 K for 5 min to elim-
inate the heat history, then cooled under three kinds of
conditions: (1) The films were cooled down in the oil-bath
from 523 K to 293 K naturally; (2) The films were cooled
down in the air at 293 K; (3) The films were quenched in
the water at 293 K.
XRD patterns were recorded by a D/max 2550 X-ray
diffractometer. The CuKa radiation source was operated at
40 kV and 300 mA. Patterns were recorded by monitoring
those diffractions from 5 to 40. The scan speed was
4/min.
Non-isothermal crystallization kinetics model
It is well known that isothermal crystallization kinetics of
polymers is commonly studied by the Avrami method [14]
1  Xc t expZt tn 1
where Xc(t) is the relative crystallinity degree, n is the
Avrami crystallization exponent dependent on the mecha-
nism of nucleation, t is the time taken during the crystal-
lization process, and Zt is a crystallization growth rate
constant. Both Zt and n are constants which are denoted as
a given crystalline morphology and type of nucleation at a
particular crystallization condition. Therefore, we have
log ln1  Xc t log Zt n log t 2 for this is possible that GPTES used as a coupling agent has
Plotting log ln1  Xc t versus log t for a certain These results mean that the fine silica particles are com-
cooling rate, the Avrami exponent n and the crystallization
rate Zt can be calculated. Equation (1) is suitable for an
9085
isothermal crystallization system. Just like isothermal
analysis, non-isothermal crystallization can also be ana-
lyzed by the Avrami equation, but, considering the non-
isothermal characterization, Jeziorny [15] presented the
final form of the parameter characterizing the kinetics of
non-isothermal crystallization as follows:
log Zc log Zt =/ 3
where / is the cooling rate.
The non-isothermal crystallization can also be analyzed
using the Ozawa method [16]. The Ozawa equation is as
follows
1  Xc expkT=/m  4
where k(T) is the crystallization rate constant, Xc is the
relative crystallinity, / is the cooling rate, and m is the
Ozawa exponent depending on the crystal growth and
nucleation mechanism.
Mo and Liu [17, 18] have proposed a new kinetic
equation of non-isothermal crystallization by combing the
Avrami and Ozawa equations:
log / log FT  a log t 5
where a refers to the ratio of the Avrami exponent n to the
Ozawa exponent m(a = n/m); the parameter
FT KT=Zt 1=m represents the value of cooling rate,
which has to be chosen at unit crystallization time when the
measured system amounts to a certain degree of crystal-
linity. According to Eq. 5, the plot of ln/ versus ln t at a
given crystallinity will be a straight line. Parameters a and
F(T) can be obtained from the slope and the intercept of the
line.
Results and discussion
Morphologies of SiO2
TEM photographs of PA6S and PA6FS are shown in
Fig. 1. Figure 1a shows that the silica nanoparticles were
approximately 50~70 nm in diameter, the nanoparticles
formed a slightly aggregated and became larger in size, and
dispersion of the particles in PA 6 was poor. However, the
size silica in the PA6FS is 30 nm, this indicates that the
silica particles exhibited a homogenous matrix. The reason
an epoxy group and this group reacted with PA6 matrix.
posed of silica and a part of epoxy network which should
form an interpenetrating network structure between the
123
9086 J Mater Sci (2007) 42:90839091

16

14
40K/min
12 20K/min
10K/min
10 5K/min
2.5K/min
8

exo
6

0
380 400 420 440 460 480 500 520
Fig. 1 TEM graphs of PA6S (a) and PA6FS (b) Temperature (K)

Fig. 3 DSC thermograms of non-isothermal crystallization for PA6S

organic and inorganic components with a covalent bond. It
suggested that the GPTES coupling agent resulted in the
morphological change of the hybrid materials.
Non-isothermal crystallization behavior
The DSC thermograms of non-isothermal crystallization
for pure PA6, PA6S and PA6FS hybrid nanocomposites at
different cooling rates are presented in Figs. 24, respec-
tively. Figure 2 reveals a small crystallization exotherm at
455.5 K(Tp) for pure PA6 cooled at 2.5 K/min, Tp peak
became more intense and shifts to lower temperature as the
cooling rate increases. A similar trend was observed in
PA6S (Fig. 3) and PA6FS (Fig. 4) hybrid nanocomposites.
The values of Tp, Ti (initial crystallization temperature) and
DH (the crystallization enthalpy) of non-isothermal crys-
tallization exotherms of these three materials were listed in
Table 1. It is clear that all the samples have nearly same Ti,
which means the addition of silica nanoparticles has little
nanocomposites at different cooling rates
effect on the onset temperature of the crystallization. As for
the DH, which is proportional to the degree of crystallinity
Xc, it can be seen that the value change in the following
order: pure PA6 > PA6S > PA6FS.
Table 1 listed the value of the crystallization peak time
tp, which is denoted as the time the sample spends during
the temperature drop from Ti to Tp. The tp decreases
roughly with the increase of the cooling rate, meaning the
increase in the rate of crystallization. It can also be seen
that the value of tp of PA6FS is smaller than that of pure
PA6, indicating that the crystallization of pure PA6 is more
difficult than that of PA6FS hybrid nanocomposites. For
comparison, the reciprocal values of the time tp versus
cooling rates for all the samples were plotted in Fig. 5. It
was found that for a tp certain sample, the rate of non-
isothermal crystallization increases with increasing the

22
14
20
a. 40 K/min
b. 20 K/min 12
18 40K/min
c. 10 K/min
16 20K/min
d. 5 K/min 10
10K/min
14 e. 2.5 K/min
5K/min
12 8 2.5K/min
exo

exo

10
6
8

6 4

4
2
2

0 0
380 400 420 440 460 480 500 360 380 400 420 440 460 480 500
Temperature (K) Temperature(K)

Fig. 2 DSC thermograms of non-isothermal crystallization for pure Fig. 4 DSC thermograms of non-isothermal crystallization for
PA6 at different cooling rates PA6FS nanocomposites at different cooling rates

123
J Mater Sci (2007) 42:90839091 9087

Table 1 Values of DH, Ti, Tp and tp at various cooling rates for pure PA6, PA6S and PA6FS
R (K/min) DH (J/g) Ti (K) Tp (K) tp (min)

Pure PA6
2.5 60.62 472.7 455.5 6.9
5 59.76 469.5 449.6 4.0
10 57.15 465.8 443.1 2.3
20 54.40 461.5 436.0 1.3
40 53.7 455.7 425.6 0.8
PA6S
2.5 48.36 469.5 457.0 5
5 45.80 465.1 452.0 2.7
10 44.29 460.1 445.8 1.4
20 42.14 455.3 437.9 0.9
40 40.50 448.4 427.0 0.6
PA6FS
2.5 50.71 473.1 467.0 2.4
5 50.85 468.9 462.2 1.3
10 51.32 465.1 456.9 0.8
20 52.10 459.7 450.4 0.5
40 53.39 454.0 442.8 0.3

cooling rate. The rate of crystallization of PA6FS is the
highest one at all cooling rates, and the rate of PA6S is
located between in the rates of PA6FS and pure PA6,
which means the addition of silica nanoparticles in situ
generated and functionalized by GPTES accelerates crys-
tallization greatly.
Figures 6, 7 and 8 present the relative crystallinity de-
gree as a function of temperature for these three materials
crystallized at various cooling rates. The higher the cooling
rate, the lower the temperature range at which the crys-
tallization occurs, therefore, the transformation is con-
trolled by nucleation. In addition, all curves have
approximately the same shape, indicating that only the
retardation effect of cooling rate on the crystallization is
observed in these curves.
Non-isothermal crystallization kinetics analyses
Figures 9, 10 and 11 present plots of log[ln(1 Xc(t))] as
a function of logt for pure PA6, PA6S and PA6FS,
respectively. The parameters n, Zt, Zc and t1/2 showed in
Table 2 were obtained from Figs. 911. n values of pure
PA6 vary from 2.56 to 3.11, and those of PA6S range from
2.72 to 3.56, which means the addition of silica influences

3.5
100
pure PA6
3.0 PA6S
PA6FS
80
Relative crystallinit y(%)

2.5
e
60
2.0
1/tp (min )

d
-1

a. 2.5K/min c
1.5 40
b. 5K/min b
c. 10K/min
1.0 a
20 d. 20K/min
e. 40K/min
0.5
0

0.0
0 5 10 15 20 25 30 35 40 45 400 420 440 460 480
Cooling rate (K/min) Temperature (K)

Fig. 5 Plots of 1/tp as a function of cooling rate for pure PA6, PA6S Fig. 6 Plot of relative crystallinity as a function of temperature for
and PA6FS nanocomposites pure PA6 crystallized non-isothermally at various cooling rates

123
9088 J Mater Sci (2007) 42:90839091

100 1.0

0.5
Relative crystallinity (%)

80

0.0
e

log(-ln(1-Xc))
60
-0.5
d
40
c -1.0
a.2.5K/min 2.5 K/min
b.5K/min b 5 K/min
20 c.10K/min -1.5 10 K/min
d.20K/min a 20 K/min
e.40K/min -2.0 40 K/min
0

400 420 440 460 480 -2.5

-0.5 0.0 0.5 1.0
Temperature (K)
log t (min)
Fig. 7 Plot of relative crystallinity as a function of temperature for
Fig. 10 Plot of log[ln(1 Xc)] as a function of logt for PA6S
PA6S crystallized non-isothermally at various cooling rates

the mechanism of nucleation and the growth of PA6

crystallites. Those of PA6FS range from 2.95 to 3.91,
100 which means the addition of silica nanoparticles in situ
functionalized is more effective than that of unfunctional-
80 ized silica. The data listed in Table 2 also show that for all
Relative crystallinity (%)

e
samples Zc increases and t1/2 decreases as the cooling rate
60 increases, which means an increase of the rate of crystal-
d
lization. At the same cooling rate, the higher Zc of PA6FS
40
a.2.5K/min c than that of pure PA6 and PA6S indicates that the silica
b.5K/min
c.10K/min modified by GPTES prompted crystallization effectively.
b
20 d.20K/min
a
Figures 12, 13 and 14 present plots of log/ as a function
e.40K/min
of logt for pure PA6, PA6S and PA6FS, respectively. The
0 slope a and intercept log F(T) showed in Table 3 were
obtained from the fitted straight line. It can be seen from
400 420 440 460 480
Table 3 that the values of F(T) systematically increase with
Temperature (K)
an increase in the relative degree of crystallinity. At a given
Fig. 8 Plot of relative crystallinity as a function of temperature for degree of crystallinity, the higher the F(T) value, the higher
PA6FS crystallized non-isothermally at various cooling rates cooling rate is needed within unit crystallization time,
indicating the difficulty of polymer crystallization. By
comparing the values of F(T) of different samples, we have

1.0
1.0

0.5
0.5

0.0
0.0
log(-ln(1-Xc))

log(-ln(1-Xc))

-0.5
-0.5
-1.0

-1.0 2.5 K/min

-1.5 2.5K/min 5 K/min
5K/min 10 K/min
-2.0 10K/min -1.5 20 K/min
20K/min 40 K/min
40K/min
-2.5 -2.0

-3.0 -2.5
-1.0 -0.5 0.0 0.5 1.0 -0.5 0.0 0.5 1.0
log t(min) log t (min)

Fig. 9 Plot of log[ln(1 Xc)] as a function of logt pure PA6 Fig. 11 Plot of log[ln(1 Xc)] as a function of logt for PA6FS

123
J Mater Sci (2007) 42:90839091 9089

Table 2 Values of n, Zt, Zc and t1/2 at various cooling rates for pure PA6, PA6S and PA6FS nanocomposites
/ (K/min) n Zt Zc (minn)a t1/2 (s)

Pure PA6
2.5 2.56 0.01 0.13 392
5 2.87 0.02 0.44 230
10 2.99 0.07 0.76 134
20 3.02 0.30 0.94 79
40 3.11 1.33 1.01 47
PA6S
2.5 2.72 0.01 0.13 291
5 2.98 0.04 0.51 154
10 3.21 0.14 0.82 85
20 3.33 1.41 1.02 53
40 3.56 6.83 1.05 31
PA6FS
2.5 2.95 0.07 0.35 129
5 3.23 0.17 0.70 84
10 3.68 1.54 1.04 52
20 3.91 7.52 1.11 31
40 3.86 85.22 1.12 21
a
Is calculated from Eq. 3

found that the value of PA6FS is lower than those of pure the influence of silica on the crystalline fraction. It was
PA6 and PA6S, meaning that the crystallization rate of previously reported [20] that the a form exhibited two
PA6FS is faster than those of pure PA6 and PA6S. This is reflection peaks at 2h  20 [a1 (200), d = 0.44 nm] and
in accordance with the result obtained from the Avrami 24 [a2 (002), d = 0.37 nm], where only one reflection
approach. peak at 2h  21 [c (200), d = 0.42 nm] was observed for
the c form. The a1 peak is sensitive to the distance between
Crystalline form hydrogen-bonded chains inside the sheet-like structures
whereas the a2 peak is sensitive to the separation distance
PA6 is known to crystallize into various phases: the a between the sheets [21].
stable crystal and among others the c-phase [8, 19]. In the Figures 15 and 16 show the XRD patterns of pure PA6
a-phase the hydrogen bonds are formed between antipar- and PA6FS hybrid nanocomposites under different cooling
allel chains whereas in the c-phase they are formed conditions, respectively. The results indicate that in the
between parallel chains. XRD was used to gain insight into

1.4 1.4 Xc=0.3

Xc=0.3
Xc=0.4
Xc=0.4
Xc=0.5
Xc=0.5 1.2
1.2 Xc=0.6
Xc=0.6
Xc=0.7
Xc=0.7
1.0 1.0
log (K/min)

log (K/min)

0.8 0.8

0.6 0.6

0.4 0.4

-0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6
log t (min) log t (min)

Fig. 12 Plot of log/ as a function of logt for pure PA6 Fig. 13 Plot of log/ as a function of logt for PA6S nanocomposites

123
9090 J Mater Sci (2007) 42:90839091

1(200)
1.4
Xc=0.3
2(002)
Xc=0.4
1.2 Xc=0.5
Xc=0.6

Intensity(*1000 cps)
Xc=0.7
1.0
log (K/min)

0.8

1
0.6
2
0.4 3 2(200)

-0.4 -0.2 0.0 0.2 0.4 0.6 5 10 15 20 25 30 35 40

log t (min) 2(Deg)

Fig. 14 Plot of log/ as a function of logt for PA6FS nanocomposites Fig. 15 XRD patterns of pure PA6 under various cooling conditions:
(1) oil-bath cooling, (2) air cooling and (3) water cooling; the curves
Table 3 Values of F(T) and a at various relative crystallinity for were vertically offset for clarity
pure PA6, PA6S and PA6FS
2(002)
Xc F(T) a
1(200)
Pure PA6
0.3 9.68 1.30
Intensity(*1000 cps)

0.4 11.21 1.32

0.5 12.29 1.32 1
0.6 13.53 1.34
0.7 15.75 1.36
2(200)
PA6S 2
0.3 9.75 1.32
0.4 10.43 1.34
0.5 11.85 1.35 3
0.6 12.77 1.35
0.7 14.36 1.37 5 10 15 20 25 30 35 40

PA6FS
0.3 8.44
0.4 9.23
0.5 10.34
0.6 10.96
0.7 11.01
slow and the medium cooling rates (cooling in oil-bath and
in air), there exists only the a form, while cooled rapidly
(quenched in water), the a and c form co-existed in pure
PA6 (Fig. 15). The a and c form coexist in PA6/modified
silica hybrid nanocomposites films under air cooling rate.
But the c form becomes dominant in higher cooling rate
(Fig. 16). Under cooling in oil-bath, in pure PA6, the a1
peak is higher than the a2 peak (Fig. 15, curve 1). How-
ever, in PA6/modified silica hybrid nanocomposites film,
the a1 peak is lower than the a2 peak (Fig. 16, curve 1).
This means that the extent of hydrogen-bonded NH asso-
ciated with the a-phase in PA6FS hybrid nanocomposites is
2(Deg)
1.19
Fig. 16 XRD patterns of PA6FS nanocomposites under various
1.21 cooling conditions: (1) oil-bath cooling, (2) air cooling and (3) water
1.23 cooling; the curves were vertically offset for clarity
1.24
1.26 reduced [8]. Vaia et al. [22] suggested that the addition of
silicate clay layers forces the amide groups of PA6 out of
the plane formed by the chains. This results in conforma-
tional changes of the chains, which limits the formation of
H-bonded sheets and the c-phase is favored. We think that
the same reason result in the c-phase is favored in PA6FS
hybrid nanocomposites.
Conclusion
A systematic study of the non-isothermal crystallization
kinetics of pure PA6, PA6S and PA6FS has been per-
formed by the DSC technique. The crystallization kinetics
of each sample was investigate according to two different

123
J Mater Sci (2007) 42:90839091 9091

kinetic models, namely, the Avrami and the Mo. Both 3. Yang F, Ou YC, Yu ZZ (1998) J Appl Polym Sci 69:355
models can describe the experimental data very well. 4. Balazs AC, Singh C, Zhulina E (1998) Macromolecules 31:8370
5. Hajji P, David L, Gerard JF, Paul C (1999) J Polym Sci B37:3172
The DSC results imply that the nano-sized SiO2 in situ 6. Reynaud E, Jouen T, Gauthier C, Vigier G, Varlet J (2001)
generated and functionlized by GPTES in PA6FS hybrid Polymer 42:8759
nanocomposites act as nucleation agents and accelerates 7. Li Y, Yu J, Guo ZX (2002) J Appl Polym Sci 84:827
the crystallization during the cooling process from the 8. Wu QJ, Liu XH, Lars AB (2002) Polymer 43:2445
9. Zhang P, Wang XY, Wei SS (2004) A direct method for synthesize
melt. But the introduction of silica nanoparticles does not PA6/inorganic nanocomposites. China Patent: ZL021398369.4
have great effect on the onset temperature of crystalliza- 10. Zhao CX, Zhang P, He JP, Lu SR, Wang XY (2007) Poly Mater
tion. The addition of silica influences the mechanism of Sci Eng (Chinese) 23(1):218
nucleation and the growth of polyamide crystallites. 11. Liu XH, Wu QJ (2002) Eur Poly J 38:1383
12. Fornes TD, Paul DR (2003) Polymer 44:3945
XRD results indicated that the addition of functionalized 13. Tol RT, Mathot VBF, Groeninckx G (2005) Ploymer 46:2955
nano-silica particles favored the formation of the c crys- 14. Avrami M (1940) J Chem Phys 8:212
talline form. 15. Celina M, George GA (1995) Degrad Stab 48:297
16. Ozawa T (1971) Polymer 12:150
Acknowledgements The financial support from the National Nat- 17. Liu JP, Mo ZS (1991) China Polym Bull (in Chinese) 4:199
ural Science Foundation of China (No. 10372087), Scientific Re- 18. Liu TX, Mo ZS, Wang SG et al (1997) Polym Eng Sci 37:568
search Fund of Hunan Provincial Education Department (No. 19. Men YF, Rieger J (2004) Euro Poly J 40:2629
05B004) and the College Students Innovation Foundation of 20. Campoy I, Gomez MA, Marco C (1998) Polymer 39:6279
Xiangtan University is greatly acknowledged. 21. Murthy NS, Curran SA, Aharoni SM, Minor H (1991) Macro-
molecules 24:3215
22. Lincoln DM, Vaia RA, Wang ZG, Hsiao BS (2001) Polymer
42:1621
References

1. Schmidt H (1985) Non-cryst Solids 73:681

2. Wen JY, Wilkes GL (1996) Chem Mater 8:1667

123
J Mater Sci (2007) 42:90929096
DOI 10.1007/s10853-006-1260-8
Modelling microstructure evolution towards ultrafine crystallinity
produced by severe plastic deformation
Yuri Estrin Hyoung Seop Kim
Received: 26 May 2006 / Accepted: 28 August 2006 / Published online: 4 August 2007

Springer Science+Business Media, LLC 2007
Abstract The significance of severe plastic deformation
by equal channel angular pressing (ECAP) as a promising
technique for extreme grain refinement down to sub-
micrometer and sometimes nanometer scale is generally
recognised. We present a modelling frame for describing
ECAP based on microstructure evolution. Following a
particular scenario of grain refinement, in which a dislo-
cation cell structure is considered as a precursor of the
developing grain structure, the variation of the cell size with
the number of ECAP passes is traced. Finite element sim-
ulations based on the model compare favourably with the
experimental data. Further features of the model such as a
provision for modelling the variation of the misorientation
angle distribution and texture evolution are also discussed.
Introduction
Among the severe plastic deformation techniques used as a
means of extreme grain refinement in bulk metallic mate-
rials [1, 2], equal channel angular pressing (ECAP) has
Y. Estrin (&)
ARC Centre of Excellence for Design in Light Alloys,
Department of Materials Engineering, Monash University,
Clayton, Vic. 3800, Australia
e-mail: yuri.estrin@eng.monash.edu.au
Y. Estrin
CSIRO Division of Manufacturing and Materials Technology,
Clayton, Vic. 3168, Australia
H. S. Kim
Department of Metallurgical Engineering,
Chungnam National University, Daejeon 305-764, Korea
e-mail: hskim@cnu.ac.kr
123
emerged as arguably the most promising one. While a large
amount of experimental data have been collected over the
past decade, often supported by finite element calculations
based on simplified constitutive models, microstructure
based modelling has been relatively scarce. The present
authors, together with colleagues, have been working in the
field of ECAP modelling for some five years. This article
outlines the general frame of their modelling approach and
gives a summary of results, along with an outlook for
outstanding problems.
Among the possible scenarios of grain structure evolu-
tion under ECAP processing, we select a gradual
transformation of the dislocation cell structure to a new,
refined grain structure. This involves accumulation of
misorientation between neighbouring dislocation cells with
strain turning low angle boundaries to large angle ones
[35]. Prangnell et al. [6] proposed a picture in which a
decrease of lateral dimensions of elongated grains termi-
nates in their subdivision into smaller grains once the grain
thickness drops down to the dislocation cell size. Basically,
in their model, too, it is the dislocation cell size that
determines the final grain size in the material undergoing
large strain deformation by ECAP.
The ECAP model providing a description of the dislo-
cation cell size evolution goes back to the strain hardening
model for large strains [7]. A brief outline of the model, as
well as a description of some recent developments, are
given in the subsequent sections. Experimental data vali-
dating numerical simulation results are also presented there.
Outline of the microstructure evolution model
The model of large strain deformation [7] underlying our
approach applies to dislocation cell-forming materials. The
J Mater Sci (2007) 42:90929096 9093

initial stage of the process when a primordial cell struc-
ture is formed is not considered. The cells of average size d
are assumed to have a cubic shape. The volume fraction of
the cell walls is given by
d 3
d
w3
f ; 1
d3
where w denotes the wall thickness. The evolution of the
dislocation densities in the cell walls and the cell interiors,
denoted respectively qw and qc, is described by the set of
coupled differential equations
p  p  
1=n represents the inverse of the rate of this variation.
dqw 6b 1
f 2=3 3b 1
f qw c_

k0 qw ;
dc bdf fb c_ 0
p  
1=n
dqc 1 qw 6 c_
a p
b
k 0 qc :
dc 3 b bd1
f 1=3 c_ 0
In the above equations it was assumed that the shear
strain c in the cell walls and the cell interiors is the same.
The quantity c_ o denotes a reference shear rate and b the
magnitude of the Burgers vector. The quantities a* and b*
are numerical constants. The physical origin of the various
terms in the evolution equations representing possible
dislocation reactions involved (such as loss of cell interior
dislocations to become entrapped in the walls given by the
first term of Eq. 1 and the second term of Eq. 2) has been
explained elsewhere [35, 7]. The exponent n in the last,
dynamic recovery, terms in both equations can be taken to
be inversely proportional to the absolute temperature, given
the fact that ECAP processing is typically conducted below
half the melting temperature, while the coefficient ko can
be considered constant.
Whereas Eqs. 2 and 3 describe the detail of the evolution
of the dislocation density, that of the other microstruc-
tural variables, d and f, involves certain assumptions.
Following Ref. [7] we assume that the average grain size d
scales with the inverse square root of the total dislocation
density
q f qw 1
f qc ; 4 order of 250 nma value found experimentally [4].
p Equation 5 can roughly be re-written as
d K= q; 5
where K is a constant. With the variation of the average
grain size thus defined, a description of the variation of f
requires the knowledge of the evolution of the wall
thickness w. The latter quantity undergoes a gradual
decrease with straining, as the cell walls, initially fuzzy
due to the abundance of geometrically unnecessary
statistical dislocations, become sharper. This effect
outstrips the concurrent increase of the total grain
boundary area, so that the volume fraction of the wall
material turns out to be a decreasing function of strain,
which can be represented [2, 3] by
f f1 fo
f1 exp
c=~c: 6
This function was chosen to fit available experimental
data for Cu and describes the variation with strain c of the
volume fraction f from an initial value fo to an asymptotic
value f?, which is smaller than fo. The parameter ~c
Strain hardening is considered by relating the equivalent
resolved shear stresses src and srw in the cell interiors and the
2 cell walls, respectively, to the equivalent resolved plastic
shear rates in these phases (both assumed equal to c_ ) and
the respective dislocation densities:
 1=m
p c_
3 src aGb qc ; 7
c_ 0
 1=m
p c_
srw aGb qw : 8
c_ 0
Here G is the shear modulus, 1/m is the strain rate
sensitivity parameter and a is a numerical constant. The
strain hardening behaviour of the composite is defined by
a scalar quantity, sr, that is obtained by applying the rule of
mixtures:
sr f srw 1
f src : 9
The model outlined was shown to describe strain
hardening behaviour of dislocation-cell forming metallic
materials, including late stages of hardening, very well [7].
It is particularly suited for simulating ECAP processing, in
which very large strains are involved.
In the scenario of grain refinement adopted, the smallest
dislocation cell size achievable for a given material pre-
determines the eventual grain size. With the highest dis-
location density not exceeding 2.5 1015 m
2, cf. [8], and
K of the order of 10, the smallest grain size in copper
ECAP can produce would be, as predicted by Eq. 5, of the
d G
KMa  ; 10
b r
where r is the flow stress and M is the Taylor factor. From
this relation it is easily seen that for the grain size to
become smaller than a 100-fold of the lattice parameter,
say, the stress should become higher than one tenth of the

123
9094 J Mater Sci (2007) 42:90929096

theoretical strength, which is hardly possible with common

metallic materials.
A typical example of an ultrafine grained structure
produced in copper by ECAP is shown in Fig. 1. The grain
size distribution evolving during the ECAP (but basically
fairly stable already after the first ECAP pass) is seen in
Fig. 2.
While the evolution of the dislocation cell size leading
to the average grain size as seen in Fig. 1 is described by
the above model, an extra model was developed in Ref.
[10] to account for the grain size distributions.

Modelling of texture evolution

To describe texture evolution, the above microstructure

evolution module can be combined with a crystal plas- ECAP passes (Route C [10])
ticity model [7]. A first variant of the model used so far is
admittedly oversimplified. Thus, misorientations between
the dislocation cells within an individual grain are disre-
garded, so that the grain can be characterized by a single
equivalent resolved shear strain rate. This approach was
developed at a time when no model for the evolution of
misorientations was available. Still as illustrated below,
reasonably good results with regards to texture prediction
were obtained [3]. The equivalent resolved shear strain rate
was expressed in terms of the resolved shear strain rates c_ rs
on the active slip systems s:
Fig. 2 Grain size distribution in copper for different numbers of
" #1m
m
X
N
rm1
r
c_ c_ s m ; 11
s1
where N is the number of active slip planes. Using the full-
constraint Taylor approach, and employing a technique of
random selection of five slip systems from among the 12
potentially active ones in each grain, finite element calcu-
lations with ABAQUS were performed. With 300 grains
per node (initially randomly oriented), pole figures were
determined by aggregating the grain orientations. A typical
result for copper is presented in Fig. 3. Similarly good

Fig. 3 Simulated (right) and experimental (left) (111) pole figures for
Fig. 1 TEM micrograph of the grain structure in copper after 8 the uniform part of a copper billet deformed by 2 and 4 ECAP passes
ECAP passes (Route Bc [9]) (Route A [11])

123
J Mater Sci (2007) 42:90929096
predictive capability of the model was also proven for Al
and IF steel.
Now that models for the evolution of the misorientation
angle distributions for the dislocation cell structure within a
grain are available (see next section), further development
of the texture module has become possible. One can
envisage a procedure for updating the cell/grain population
at each simulation step that would account for the fact that
part of the cells develop into new grains by acquiring
sufficiently large misorientations with the adjacent cells.
Modelling of the evolution of dislocation cell
misorientations
Recently, a modification of the model presented in section
Outline of the microstructure evolution model was
proposed [12] with the aim of tracing the average misori-
entation angle. To that end, the dislocation density in the
cell walls was divided in two distinct groups. Dislocations
of the first group contribute to an imbalance in the Burgers
vector in a dislocation cell wall and thus produce a mis-
orientation across the wall. Accordingly, they are referred
to as geometrically necessary ones, their density being qgw :
The other group are statistical dislocations, which are
redundant in the sense that they do not produce misorien-
tation across the boundary. (Their density is denoted qsw :)
The absolute value of the misorientation angle between the
neighbouring cells is then given by
 p before possible modifications are considered is whether the
p
h arctan b qgw b qgw : 12
The main assumption made in Ref. [12] was that a certain
fraction n of the dislocations incoming into cell walls from
the cell interiors contribute to local imbalance in the
Burgers vector by build-up of the density of geometrically
necessary dislocations and thus misorientation. The rate of
growth of qgw was written in the form
6b c_ c 1
f 2=3
q_ gw n : 13
bdf
It should be noted that this evolution equation does not
include any recovery terms, as geometrically necessary
dislocations are assumed to be unrecoverable. Adjustments
of the evolution equations for the statistical dislocation
density in the walls and the total dislocation density in cell
interiors are straightforward [12]. A very simple equation,
dh v 1
; 14
dc 2 h
that follows from Eqs. 12 and 13 can easily be solved. Here
the notation
9095
b 1
f 2=3
v 6b n ; 15
d f
was used. For sufficiently large strains when d and f
approach their saturation values, v tends to a constant, and
the solution of Eq. 14 yields a parabolic law of the kind
observed experimentally [13]. Using reasonable parameter
values, the model equations can be solved numerically,
yielding the dependence of the average misorientation
angle for copper. It was recognised, however, that the
predicted misorientation angles were significantly smaller
than the observed ones. This discrepancy can be eliminated
by a slight modification of the model [12]. We consider
that with the progress of misorientation accumulation the
efficiency of cell walls as places for storage of incoming
cell interior dislocations increases. This is taken into
account by making b* and increasing function of h, e.g. by
replacing the old b* with the function b b
b
exp
12h=p: This function describes the evolution of this
efficiency parameter from an initial value of b* to a
saturation value of b**, the characteristic rate of this
evolution being determined by a misorientation angle of
15 (p /12)a value that is commonly accepted to signify
transition from small angle to large angle case. Figure 4
shows the predicted variation of h as well as the mea-
surement data. While reasonable agreement was found for
a particular choice of the parameter values (b 6b ; the
validity of the above ansatz needs to be investigated fur-
ther. One issue that requires experimental clarification
average misorientation angle saturates with the number of
ECAP passes. The current version of the model does not
predict saturation within a practicable number of passes.
8
Experimental
Average Misorientation Angle, degree
7
Calculation 7
6
6
5 5
4 4
3
3
2
2 1
1
0
0 1 2 3 4 5 6 7 8
Accumulated Equivalent Strain
Fig. 4 Evolution of the average misorientation angle with strain (or
the number of ECAP passes in a 90 ECAP die) for copper:
calculation results for b 6b vis-a`-vis experiment [13]
123
9096 J Mater Sci (2007) 42:90929096

Of great interest for modelling the effects of ECAP on
the material microstructure is the misorientation angle
distribution. A provision for a probabilistic description
involving distribution functions can be made in terms of a
Fokker-Planck equation [12, 14]. A particular form of the
Fokker-Planck equation, derived using the Langevin
approach [12], reads
 
oPh; c kT o2 Ph; c o 1 74/122) and Korea Research Foundation Grant (KRF-2005-202-

FPh; c ; 16 D00205) is gratefully acknowledged.
oc g oh2 oh g
where Ph; c represents the misorientation angle
distribution function for a given value of the shear strain
c, while the quantities g and Fare given by
v Sci 45:103
g 2 17
2h
and
1 dv
F : 18
2h dc
A discussion of this probabilistic approach and an
analysis of the ECAP history on the misorientation angle
distribution will be given elsewhere.
Conclusion
In a snapshot presentation of our research on the evolu-
tion of the microstructure and texture during severe plastic
deformation by equal channel angular pressing, we
attempted to show the recent successes of modelling and
also indicate some outstanding problems. These involve
particularly the understanding and modelling of the evo-
lution of misorientation angles in the dislocation cell
structure and the effect of this evolution on texture devel-
opment. It is the belief of the authors that these aspects of
modelling of ECAP deformation should be among the main
targets of research in this area in a near future.
Acknowledgements The authors have greatly benefited from col-
laboration and innumerable discussions with Y. Brechet, A. Molinari,
L. Toth, R. Hellmig and many other colleagues to whom they wish to
express their sincere gratitude. Financial support from DFG (grant ES
References
1. Valiev RZ, Islamgaliev RK, Alexandrov IV (2000) Prog Mater
2. Valiev RZ, Estrin Y, Horita Z, Langdon TG, Zehetbauer MJ, Zhu
YT (2006) JOM 58:33
3. Baik SC, Estrin Y, Kim HS, Hellmig RJ (2003) Mater Sci Eng A
351:86
4. Baik SC, Hellmig RJ, Estrin Y, Kim HS (2003) Z Metallkd
94:754
5. Kim HS, Estrin Y (2005) Mater Sci Eng A 410411:285
6. Prangnell PB, Bowen JR, Apps PJ (2004) Mater Sci Eng A 375
377:178
7. Toth LS, Molinari A, Estrin Y (2002) J Eng Mater Technol
124:71
8. Gubicza J, Balogh L, Hellmig RJ, Estrin Y, Ungar T (2005)
Mater Sci Eng A 400401:334
9. Hadzima B, Janecek M, Hellmig RJ, Kutnyakova Y, Estrin Y
(2006) Mater Sci Forum 503504:883
10. Gendelman OV, Shapiro M, Estrin Y, Hellmig RJ, Lekhtmakher S
(2006) Mater Sci Eng A 438:88
11. Baik SC, Estrin Y, Hellmig RJ, Jeong H-T, Brokmeier HG, Kim
HS (2003) Z Metallkd 94:1189
12. Estrin Y, Toth LS, Brechet Y, Kim HS (2006) Mater Sci Forum
503504:675
13. Hellmig RJ, Kim HS, Estrin Y (2005) Trans Indian Inst Met
58:1107
14. Pantleon W (2005) Mater Sci Eng A 400401:118

123
J Mater Sci (2007) 42:90979111
DOI 10.1007/s10853-006-1261-7
Recent progress on the study of the microstructure
and mechanical properties of ECAE copper
Florian H. Dalla Torre Azdiar A. Gazder
Elena V. Pereloma Christopher H. J. Davies
Received: 2 June 2006 / Accepted: 9 November 2006 / Published online: 4 August 2007

Springer Science+Business Media, LLC 2007
Abstract Results on the microstructure and the tensile
properties of equal channel angular extruded (ECAE)
copper processed for one to 16 passes are presented and
compared with the available literature data. With increas-
ing number of passes (N), the microstructure changes from
a strongly elongated shear band structure after N = 1 and 2,
towards a more equiaxed subgrain and grain structure. This
is accompanied by a decrease in the cell wall or subgrain-
boundary widths and an increase in recovered or even
recrystallised grain structures with low dislocation densi-
ties. Electron backscatter diffraction measurements have
indicated that for lower N, the location of R3 boundaries is
restricted to shear bands, while at greater N, R3 boundaries
were found to be more widely distributed. Texture mea-
surements indicate close similarity with simple shear
texture components and a spread of the orientation com-
ponents with greater N. Upon comparing the tensile
behaviour of as-ECAE Cu with the surveyed literature,
broad agreement on the strength of the material is
achieved. However, a strong variation in the percentage
elongation to failure is also noted. Strain hardening and
deformation kinetic analysis via strain rate jump tests
indicate an evolution from stage III to V hardening during
F. H. Dalla Torre (&)
Laboratory of Metal Physics and Technology, Department of
Materials, ETH Zurich, Wolfgang-Pauli-Str. 10, Zurich 8093,
Switzerland
e-mail: Florian.DallaTorre@mat.ethz.ch
A. A. Gazder
E. V. Pereloma
C. H. J. Davies
Department of Materials Engineering, Monash University,
Melbourne, VIC 3800, Australia
A. A. Gazder
E. V. Pereloma
C. H. J. Davies
Victorian Centre for Advanced Materials Manufacturing,
Belmont, VIC, Australia
post-ECAE compression and a saturation in the strain rate
sensitivity after N = 4 resulting in maximum values of
*0.02. Our results suggest that rather than a change in
deformation mechanism, the increase in ductility with
increasing N is associated with an increase in the mean free
path of dislocationswith the grain boundaries remaining
actively involved as the transmitter of plastic strain and
their interaction with dislocations being the rate controlling
deformation mechanism.
Introduction
Copper (Cu) represents an ideal model material to study the
process of deformation and recrystallisation due to its
intrinsic softness and formability, low cost, fcc crystal
structure and medium stacking fault energy. In early
studies on large plastic deformation, Cu was often chosen
for deformation experiments which were conducted using
conventional methods such as rolling, torsion, extrusion or
compression, all of which are capable of imparting large
strains to the workpiece ([13] and the references therein).
These and other studies conducted between 1950 and 1980
represent the basis of our understanding of the materials
properties and its associated microstructure resulting from
severe plastic deformation (SPD), a term coined by Russian
researchers in the 1980s (see [4] among others). Compared
to classical deformation processes, the big advantage of
SPD techniques (represented in particular by equal channel
angular extrusion (ECAE) [5, 6] and high pressure torsion
(HPT) [79]) is the lack of shape-change post-deformation
and the consequent possibility to impart extremely large
strains. Thanks to Gleiters research on nanocrystalline
materials [10, 11], SPD has received enormous interest
123
9098 J Mater Sci (2007) 42:90979111

over the past two decades as a method capable of

producing fully dense and bulk nanostructured materials. In
this context, Cu has remained the best-studied material and
has only recently been outnumbered by publications on
steels mainly due to the large demand for the latter in Asia
[12]. Knowledge of the amount of accumulated strain
required to achieve a stable end micro- or nano-structure is
important should the so-produced materials enter future
industrial applications. Towards this end, a comparison of
experimental results with published literature using similar
deformation procedures is necessary to evaluate the
sensitivity of the material to the type of apparatus used.
Thus, this study aims to summarise microstructural
results on ECAE processed Cu. A survey of microstructural
evolution with emphasis on the grain size and twin for-
mation is presented. Based on previous work, we focus on
deformation at room temperature (RT) with a die angle of
U = 90, a sharp outer die corner (W = 0) and a clockwise
90 rotation of the billet along its longitudinal direction
between successive passes. This so called route BC has
proven to show the most efficient procedure by which an
equiaxed subgrain or grain structure of minimum dimen-
sions can be achieved [1315]. Moreover this paper
discusses and reviews the post-ECAE deformation behav-
iour in compression and tension. The tensile elongation and
the hardening behaviour during compression testing are
also emphasised.
Experimental details
Commercially available, high purity (>99.95%) Cu rods
(20 20 80 mm3) were annealed at 873 K for 2 h in an
inert atmosphere resulting in a mean grain size of
*20 16 lm. Three die inserts were arranged to form an
L-shaped channel within the ECAE die cradle (Fig. 1).
Room temperature (RT), U = 90 and W = 0 ECAE was
performed with a constant forward speed (2 mm s1) and
back-pressure (60 MPa) for up to N = 1, 2, 4, 8, 12 and 16
passes via route BC. While larger magnitudes of back
pressure have been shown to promote billet homogeneity
[16, 17], the back pressure used in the present study serves
mainly to allow for efficient and complete filling of the die
channel during ECAE and also enables the billet to roughly
keep its original rectangular shape. Moreover, due to the
nature of the die set-up and applied hydrostatic pressures, a
slight swelling of the billet across its thickness also occurs,
which effectively locks the billet in between the die walls
of the exit channel. Consequently, removing the three die
inserts allows for easy retrieval of the billet after each pass.
Between successive passes, the billets were ground back to
their original dimensions using a water soluble oil (Shell
TM
Dromus-B in a 40:1 water and oil ratio) which served as
Fig. 1 Schematic of ECAE tooling and U = 90 die set-up
both, lubricant and cooling medium. Considering the
machining undertaken to fit the billet back into the entry
channel for successive passes, slightly shorter billets with
increasing N are expected. Additionally, billets and speci-
mens were stored at RT between extrusions and before and
during microstructural analysis, respectively.
Specimens for microstructural characterisation via
transmission electron microscopy (TEM), X-ray diffraction
(XRD) and Electron Back-Scatter Diffraction (EBSD) were
cut perpendicular to the extrusion direction (ED) from the
center of the uniformly sheared billet length and are
referred to as the x-plane samples.
Thin foils of the x-plane samples for transmission
electron microscopy (TEM) were prepared by electrolytic
thinning (for more details, please refer to [18]). EBSD
measurements were performed on a LEO-1530 FEG-SEM
TM
fitted with a Nordlys-II EBSD detector. Mostly, a scan
area of 10 18 lm2 with a 40 or 80 nm step size was used.
In some specific cases, larger scans of up to 25 50 lm2
were also performed. Post-processing of the raw data was
undertaken using VMAP [19, 20] and HKL-Channel 5. For
grain and subgrain size measurements, the equivalent circle
diameter (ECD) method was used, wherein the area is
measured and the grain/subgrain size is defined as the
diameter of a circle with the area of the grain. In the
256-colour EBSD maps, the various colours represent
the different Euler angles of each volume element with a 5
angular variation. Note that these Euler angles refer to
specimen directions when inserted in the microscope and
therefore, do not generally correspond to the Euler angles
conventionally used to describe a deformed material. In

123
J Mater Sci (2007) 42:90979111
many cases, regions of similar orientation have similar
colours; however, this is not always the case due to the
discontinuities in Euler space [19].
Texture analysis was performed on 20 20 3 mm3
ED-samples using a GBC-MMA texture goniometer
working at 40 kV and 25 mA and equipped with a CuKa
anode and a polycapillary beam enhancer, resulting in a
collimated beam of 10 10 mm2. Raw data in the form of
incomplete pole figures (PFs) (v = 080) from the (111),
(200), and (220) crystallographic planes were used to cal-
culate the orientation distribution functions (ODFs) f(g)
using the spherical harmonic method without imposing any
sample symmetry conditions. g = (/1, /, /2) refers to
grain orientation denoted by the three Euler angles in
Bunges notation [21]. All end-texture PFs and ODFs were
subsequently rotated to TD-view and will be discussed in
the xyz laboratory system (Fig. 1). Correspondingly, the
TD, ED and ND-axes are denoted along the /1, / and /2
directions, respectively. Further details regarding the ideal
texture components developed during ECAE are included
in Table 2.
In order to produce tension and compression specimens,
the billets were first sectioned into quarters. Three such
sections were used to produce the circular tensile speci-
mens with a diameter of 4 mm and a gage length of
25 mm. Correspondingly, the center of the fourth section
was used to produce cylindrical compression test samples
of 4 mm diameter and 8 mm length. The tensile tests were
performed at RT on a screw-driven Instron operating at a
cross head speed of 2.5 mm min1, resulting in a starting
strain rate of e_ 1:6  103 s1 : The strain was measured
with a clip-gage mounted directly onto the tensile speci-
mens, yielding a Youngs modulus of about 130 GPa, in
close agreement with the theoretical value of Cu. It should
be noted that while the gage length is situated well within
the uniformly sheared stable billet length for N  8, the
gage lengths of specimens machined after N = 12 and 16
could contain non-uniformly sheared volumes. For the
compression tests, a cross head speed of 0.5 mm min1
was used on the same screw-driven Instron resulting in an
initial strain rate of e_ 1  103 s1 : Cyclic strain rate
jump tests in compression were performed on a hydraulic
Instron to evaluate the influence on strain rate sensitivity of
N. The strain rate jumps were performed between 1 104
and 1 102 s1. In both type of compression tests, the
strain was measured by the cross-head displacement only,
which due to the machine compliance leads to a signifi-
cantly lower apparent Youngs modulus in the uncorrected
stressstrain curves shown in section Mechanical prop-
erties of ECAE-Cu. Beyond 30% strain a steady increase
in the stress with strain has been associated with the
gradual increase in frictional forces. Hence only data below
30% strain were considered as valid for the analysis.
9099
Microstructure of ECAE processed Cu
In this section, a survey on ECAE Cu is presented with the
idea of imparting a general description of the microstruc-
ture and to highlight the typical features found by several
research groups. For comparison of subgrain and grain size
measurements, the available literature data has been
compiled in Table 1.
In the early 1990s, studies on the microstructure of SPD
metals produced via HPT or ECAE were conducted by the
Russian group led by Prof. R.Z. Valiev. The research was
mainly focussed on characterising the end-physical prop-
erties of SPD materials after a minimal (preferably, nano-
sized), homogeneous grain structure was reached. The
differences between these ultra-fine grained (UFG) met-
als and their coarse-grained counterparts [2224] in terms
of their elastic moduli, diffusion coefficients, internal
friction, Curie and Debye temperatures and superplasticity
was highlighted and attributed to their small grain sizes and
the development of non-equilibrium grain boundaries [23,
2527].
One of the first mentioned studies on ECAE Cu pro-
cessed at RT up to N = 4 passes using route BC led to an
equiaxed grain structure and an average grain size of
210 nm [27]. The authors observed that smaller grains
(<100 nm) with a volume fraction of *15% were absent of
lattice dislocations, while larger grains (400500 nm) with
a volume fraction of *48% indicate the presence of a cell
structure with subgrain boundaries and the presence of
lattice dislocations having a mean density inside grains of
5 1014 m2 [27]. About 37% of the volume is reported to
be constituting of intermediate grains of 100200 nm in
size with a chaotic dislocation structure [27].
Similarly, Ferasse et al. [14] also investigated the effect
of the number of passes on pure Cu and concluded that
after N = 4, grain refinement reaches a minimal size of
200300 nm, concurrent with a stabilisation of strength
and hardness. Their results on the grain size measured by
TEM were later confirmed by Mishin and Gottstein [28]
who, apart from dislocation density estimates of
5 1014 m2 (in agreement with the value given by [25,
27]), reported the occurrence of twins formed during
ECAE and a major fraction of random high-angle grain
boundaries (HAGBs). However, the latter observation has
been proven incorrect in a subsequent work by Mishin
et al. [29, 30] which concentrated on detailed misorienta-
tion measurements of the ECAE substructure deformed up
to N = 8, at U = 90 via route BC. Misorientation mea-
surements on 219 misorientations showed that 63% are
low-angle grain boundaries (LAGBs) and that the intra-
granular volumes are never surrounded solely by HAGBs.
Statistical measurements revealed a HAGB spacing per-
pendicular to the dominant band structure between 0.2 lm
123
9100 J Mater Sci (2007) 42:90979111

Table 1 Grain and subgrain size measurements of ECAE-Cu for varying number of passes and processing routes taken from various references
and own data
Processing parameters Method Avg. grain Avg. subgrain / % volume Ref.
a
size (nm) cell size (nm) HAGB
ECAE die angle Passes (N) Route Sample plane
Inner angle (U) Outer angle (W)

90 4, 12 BC x TEM 210 [4, 27]

90 4 BC TEM BF 200300 [14]
90 14 C x TEM, EBSD 200 *70 [28]
90 45 8 BC y TEM 2002000 200400 37 [29, 30]
90 0 8 BC y TEM 2002000 200300 29 [30]
90 20 18 C TEM, BF 200 [36]
90 112 BC x TEM, DF 250 : with N [38]
90 8 BC x TEM, BF 300400 (ignored) [39]
90 8 BC TEM, BF 300 [40]
90 20 8 BC x TEM, BF 500 150 30 [41]
90 20 8 C x TEM, BF 950 190 20 [41]
90 10 BC x TEM, BF 200250 [42]
90 4 BC - TEM 300 [43]
90 16, 20 A y EBSD 420, 370 68, 72 [37, 44]
90 16, 20 BC y EBSD 490, 390 67, 71 [37, 44]
92 8 BC TEM 400 [46]
90 0 4  N  16 BC x EBSD 200 120 45 own data
90 0 4  N  16 BC x TEM 170 own data
a
The various sample planes are represented perpendicular to the directions x, y, z as denoted in the coordinate system as defined by Fig. 1.
EBSD measurements have been performed with a step size of 40 nm

and 2 lm. In agreement with the comprehensive investi-
gation of Agnew [31], Mishin et al. [29, 30] concluded
that: (i) the reported grains size of 0.20.3 lm measured on
ECAE Cu included boundaries of high- and low- angle and,
(ii) ECAE deformed metals should rather be characterised
in terms of deformation structures rather than be classified
as a UFG material. These authors [29, 31] further
recognised that ECAE Cu exhibits elongated subgrain
structures, which are subdivided transversely by low-angle
boundaries similar to what has been observed as incident
dislocation boundaries in cold-rolling deformed structures
[32]. An important point highlighted in [33] concerns the
technique of measuring misorientations by means of TEM
selected area diffraction (SAD) patterns. Measurements
based on SAD patterns as performed in [13, 34, 35] do not
account for misorientations adjacent to a grain, leading to
an overestimation of the volume of HAGBs in ECAE
metals [33]. This was explained on the basis of the spatial
limitation in conventional TEM studies with large dia-
phragms (in the micrometer-range) yielded diffraction
patterns from multiple boundaries (which in bright-field
(BF) mode were shown to be spaced only few hundreds of
nanometers apart).
Several other TEM studies have quantitatively repro-
duced microstructures similar to those characterised above
[3644]. However, in most of these studies the grain size
has been defined by the shape of the boundaries rather than
by an accurate statistical misorientation measurement.
These results indicate an average size between 200 nm and
400 nm with a typical log-normal distribution function
spread from 100 nm to 600 nm (see Table 1). In order to
differentiate between the high and low misorientation
angles of the boundaries, dark-field (DF) images were
proposed for grain size measurements. However, these
studies also yielded similar grain sizes to previous calcu-
lations based on BF images [38].
The microstructural evolution of Cu processed via route
C for N = 18 passes has been studied by Baik et al. [36]
in order to deduce the mean cell size as a function of N.
They observed a minimum value of *200 nm peak cell
sizes already after the first pass, but indicated a decrease in
the spread of the cell size distribution with increasing N.
On the other hand, using a U = 90 die Huang et al. [41]
studied the influence of the strain path via routes BC and C
for up to N = 8 by comparing the microstructural differ-
ences with respect to sample orientations x and z (see
Fig. 1). They showed that while the x-plane of the micro-
structure of route BC and C samples exhibited a nearly
equiaxed cell/subgrain structure, the z-plane indicated a
higher fraction of elongated cells/subgrains for both routes

123
J Mater Sci (2007) 42:90979111
by approximately the same amount. In agreement with
previous results [1315], they observed a smaller grain and
cell/subgrain size and a higher fraction of HAGBs for route
BC compared to route C. Analysis of 100 boundaries
indicated that LAGBs always showed a diffused appear-
ance due to the dense dislocation forests at the boundaries.
Alternatively, HAGBs comprised sharp contrast and
mainly dislocation-free adjacent areas [41]. Their obser-
vation and the statistical TEM misorientation
measurements of Mishin et al. [29] now serve as a guide
for qualitatively measuring the misorientation from the
boundary morphology in order to achieve higher statistics.
Microstructural evolution of as-ECAE Cu
Upon adopting the methodology of Huang et al. [41], our
microstructure investigations [18, 45] on ECAE Cu for up
to N = 16 via route BC have yielded similar results.
Figure 2 shows that cell boundaries of low-angle configu-
ration (i.e. LAGBs) are identified depending on the
orientation of the adjacent cells either by clearly distin-
guishable high densities of dislocations and dislocation-
debris or by fuzzy diffraction contrasts (), while the
sharper boundaries (c) (often also confined to grains where
thickness fringes are visible), are considered to be HAGBs
[41]. As seen in the coupled TEM images and SAD pat-
terns after N = 1 and 2 (Fig. 3a, b), the microstructure is
dominated by lamellar boundaries (LBs) oriented in the
h110i direction (along the trace of the {111} plane),
resulting in a microstructure similar to rolling [32]. Dis-
location cell boundaries are evident and as indicated by the
SAD pattern mainly LAGBs are present. This observation
Fig. 2 TEM bright field (BF) image showing cell wall boundaries
and/or subgrain boundaries () and high angle grain boundaries (c)
in a specimen deformed up to N = 16
9101
has been reproduced more quantitatively by EBSD mis-
orientation measurements (Fig. 4a).
After N = 4 and 8, the microstructure is inhomogeneous
with respect to the various grain shapes. In some areas
grains and subgrains are still elongated, while other areas
are representative of equiaxed grain and substructures
(Fig. 3c, d). With greater N, the volume fraction of equi-
axed microstructural features increases and the distribution
of the diffraction spots is more uniform. This change
in diffraction infers an increase in the misorientation
(Fig. 4bd), and subsequently recovery within the grains
and their grain boundaries (Fig. 4e, f). An enhancement of
recovered grain structures showing low dislocation densities
in the interior and sharp grain boundaries with increasing N
is related to the higher effectiveness of dislocation annihi-
lation due to an increase in the boundary misorientations.
This in turn, is manifested by a decrease in the average
boundary width from 52 nm to 34 nm and a decrease in
the dislocation density (6.9 1013 m2) at N = 16 after
reaching a maximum (3.2 1014 m2) for N = 4 (see
Ref. [18] for details). Furthermore, in agreement with [41]
for route C,1 some larger (up to 500 nm) grains character-
ised by sharp and straight boundaries were observed for
samples with N  8 and have been said to be precursors to
recrystallised grains [18]. Also, stacking fault tetrahedra and
dislocation loops observed in larger grains containing only
few lattice dislocations, lend further support to the occur-
rence of recovery processes (Fig. 4c, d in Ref. [18]).
To improve the statistics on the variation of grain size
due to recovery and/or recrystallisation structures occur-
ring during ECAE we performed EBSD measurements
over a 40 30 lm area. Figure 5 shows a section taken
from this larger EBSD map of the N = 16 specimen. The
black lines represent HAGBs, while thicker red lines cor-
respond to R3 boundaries. The few larger grains (up to
1 lm in size) with a high abundance of R3 boundaries are
surrounded by smaller grains *170 nm in size. Over the
entire map of 40 30 lm about 12% of the grains can be
classed as recrystallised grains with a mean size of 1.0 lm,
based upon the qualifying assumptions that: (i) the
recrystallised grain size is six times larger than the mean
subgrain size, (ii) a recrystallised grain is bounded by at
least 10% HAGB and, (iii) that the image quality remains
constant.2 A similar observation of such grains of 1 lm
size with a volume fraction of 1.5% have recently been
reported for a Cu specimen subjected to N = 8 passes,
route BC [46]. From the current standpoint, the processes
that lead to the formation of such large grains requires
1
They measured a volume fraction of 1.5% represented by dynam-
ically recrystallised grains.
2
With increasing fraction of recrystallised grains, the image quality
is expected to improve.
123
9102 J Mater Sci (2007) 42:90979111

Fig. 3 TEM BF images and

SAD patterns of Cu samples
subjected to N = 1 (a), 2 (b), 4
(c), 8 (d), 12 (e) and 16 (f)

Fig. 4 Misorientation (a) 50 N=1 (b) 50 N=4

measurements of Cu after
N = (a) 1, (b) 4, (c) 8 and, 40 40
V o lu m e f r a c ti o n [ % ]

V o lu m e f r a c ti o n [ % ]

(d) 16 passes (route BC)

30 30

20 20

10 10

0 0
0-4
4-8
8-12
12-16
16-20
20-28
28-32
32-36
36-40
40-44
44-48
48-52
52-56
56-60
60-64

0-4
4-8
8-12
12-16
16-20
20-28
28-32
32-36
36-40
40-44
44-48
48-52
52-56
56-60
60-64

Misorientation angle [] Misorientation angle []

(c) 50 N=8 (d) 50 N=16

40 40
V o l u m e f r a c ti o n [ % ]

V o l u m e f r a c ti o n [ % ]

30 30

20 20

10 10

0 0
0-4
4-8
8-12
12-16
16-20
20-28
28-32
32-36
36-40
40-44
44-48
48-52
52-56
56-60
60-64
0-4
4-8
8-12
12-16
16-20
20-28
28-32
32-36
36-40
40-44
44-48
48-52
52-56
56-60
60-64

Misorientation angle [] Misorientation angle []

123
J Mater Sci (2007) 42:90979111
Fig. 5 EBSD map of a specimen subjected to N = 16 passes. Black
lines indicate high angle grain boundaries, red thicker lines
correspond to R3 boundaries
further investigation. Thus, the nature of these recrystal-
lised grainswhether they are of dynamic (i.e. during
ECAE deformation) or static (i.e. either upon removal of
hydrostatic pressure conditions in the ECAE set-up at the
end of a pass or during storage of specimens at RT) origin
requires clarification. With specific regard to latter, it
should be noted that the process of relaxation is assumed to
follow an Arrhenius-type law. Consequently, beyond a
critical time period, no microstructural differences should
be detectable after billet removal from the ECAE appara-
tus. To this end, TEM investigations on samples produced
at varying time intervals spread out over a few weeks have
also returned no discernable substructural differences.
Statistical TEM cell size measurements on our samples
are also in qualitative agreement with results published on
route C [36]. The cell size perpendicular to the dominant
LBs reaches a minimum value of *170 nm after the first
pass and remains roughly constant. However, our mea-
surements also showed that the cell sizes parallel to the LBs
decreases more gradually. It is important to note that this
behaviour is quite different to ECAE Al, which shows an
overall gradual decrease in grain size, a much larger mean
grain size of *1 lm and the appearance of a strongly
recovered microstructure [47, 48]. This behaviour can be
attributed to either the significantly higher homologous
processing temperature of Al (the melting temperature (Tm)
of Cu is 1357 K compared to 933 K for Al) or explained by
the larger stacking fault energy of Al, which favours cross
slip and dislocation climb, both effects instigating higher
rates of recovery at a constant temperature.
Twinning in ECAE Cu
Several observations on heavily deformed Cu via ECAE or
HPT (the latter yielding a grain size of about *100 nm due
9103
to the suppression of vacancy annihilation under hydro-
static pressure conditions [49]) have shown the presence of
stacking faults and twins [18, 45, 5052]. This is somewhat
surprising since deformation twinning is not regarded as
energetically favourable at low strain rates and ambient
temperatures due to the many slip systems still able to
operate in coarse-grained Cu [53, 54]. However, specific to
coarse-grained Cu, deformation twinning occurred after
substantial strain hardening [55] and has been shown to
become favourable once a critical dislocation density is
reached [56]. Huang et al. [50] indicated that the twinning
mechanism does change as a function of strain and/or grain
size. After N = 1, twins have been found in or at shear
bands indicating that in areas where locally high strains are
reached, deformation twins can be favoured over disloca-
tion slip.
In apparent agreement with these observations, our
EBSD experiments also revealed a higher density of R3
boundaries at shear bands (see white lines in Fig. 6a,
indicated by the arrow). Note that R3 boundaries can have
different geometries and comprise, apart from symmetric
tilt boundaries (i.e. coherent twin boundaries), incoherent
P
twins 3n and stacking faults in fcc materials [57, 58].
Huang et al. [50] associated the occurrence of twins
observed after N = 1 with the pole mechanism, which
nucleates from the dissociation of prismatic dislocations
into partials and grows helically via climb of the Shockley
partial dislocations [59]. At larger strains (N  16 route
BC), twins associated with the emission of partial dislo-
cations from grain boundaries (as suggested by computer
simulations [60] and experiments [6164] on nanocrystal-
line (nc) fcc metals) were also observed by Huang et al.
[50]. However, our results indicate that when compared
with the total amount of boundaries, the fraction of R3
boundaries remains low: which suggests that discussing
twinning as a deformation mechanism in ECAE Cu is of
comparatively minor importance. Figure 6b and c shows
EBSD maps of samples subjected to N = 4 and 16 passes,
where R3 boundaries are rather widely distributed over the
sample area. As shown in the close-up of the specimen
subjected to N = 16 (Fig. 6d), R3 boundaries extend over
entire grains, are sometimes interacting with subgrain
boundaries and can be found in grains of smaller than
200 nm in size. TEM observations also reveal a similar
picture. In Fig. 7a and b, the TEM image of a large grain is
shown wherein dislocations are moving from the left-hand
grain boundary towards a boundary which forms ledges
(probably due to stress incompatibility). These ledges
associated with stacking faults and a high density of grain
boundary dislocations resemble those typically related to
the initiation of annealing twins at the ledges as suggested
by the pop-out mechanism, and where growth occurs via
migration of the incoherent and more mobile R3
123
9104 J Mater Sci (2007) 42:90979111

Fig. 6 EBSD maps of N = 1

(a), N = 4 (b) and N = 16
(c + d). Grey, black and white
lines correspond to subgrain,
grain and R3 boundaries,
respectively, (d) is an enlarged
area of (c)

Fig. 7 Deformation structures

of Cu specimens subjected to
N = 8 passes. (b) shows an
enlarged area of (a); (c) stacking
faults crossing entire grain

boundaries [65]. In their model, Meyers and Murr [65]
explain the driving force for the nucleation of twins by the
reduction in dislocation density, which is again in agree-
ment with our observation of lower dislocation densities in
larger grains as those shown in Fig. 7. Here, a clearer
identification of the nucleation mode (i.e. annealing or
deformation) is not possible. In Fig. 7c, two stacking faults
are shown crossing an entire grain. Considering the
observations made by TEM and EBSD, it is important to
emphasise the difference in appearance between these
twins and those associated with the large grain shown in
Fig. 5. It is however interesting to note that R3 boundaries
also occur within the fine grained matrix (Fig. 5), which
might suggest that these serve as nuclei for the growth of
annealing twins. However, this caveat requires the defor-
mation and annealing twins to be crystallographically
identical as they consist of stacking faults on consecutive
{111} planes [66].
Texture evolution in ECAE-Cu
Figure 8 shows the (111) experimental pole figure of the
0-pass billet prior to ECAE. Similar to a previous study
[67], this initial texture is weak and consists mainly of a
h111i h001i duplex fiber with the fiber axis parallel to
the billet longitudinal axis.

123
J Mater Sci (2007) 42:90979111 9105

ND fiber running from Ch to Ah =A
h (/ symbolises or)

ED
Max = 1.59
Fig. 8 (111) pole figure of the 0-pass starting texture for Cu. The
longitudinal axis of the billet is horizontal. Contour levels: 0.5
In order to evaluate textures developed during ECAE, a
translation from the negative simple shear (SS) reference
system (or x0 y0 z system) to the sample coordinate system
(or xyz system) is accomplished by a clockwise (CW)-
rotation of h = U/2 = 45 about the z-axis. Li et al. [68]
showed that ECAE specific orientations can be derived by
increasing the /1 of ideal SS orientations by h = 45 while
the other two Euler angles (/ and /2) remain constant. The
Euler angles are listed in Table 2. The symbols marked in
the key (111) PF (Fig. 9a) represent these ideal ECAE
orientations for fcc metals [69] that distribute themselves
along two partial fibers running along: (i) {111} planes
with huvwi directions or, (ii) {hkl} planes with h110i
directions, counter-clockwise (CCW)-rotated by h = 45
around the TD from the ND-plane and ED, respectively
and are labelled using subscript (h).
They are designated
as: (i) f111g huvwih and, (ii) fhklg 110ih -type partial
fibers, respectively.
As seen in Fig. 9b, the 1-pass experimental pole figure
contains dyadic symmetry and depicts an inversion center
projected on to the z-plane. It consists of a partial h110ih
through Bh =B
h orientations and a partial h111ih fiber
extending from A*1h to A*2h through Ah =A
h : Relatively high
orientation densities can be found near the Ch and A*1h
components. In [68] it is shown that the A*1h component is
stronger than the A*2h component as a result of negative SS.
Generally speaking, the present texture components after a
single pass are closely aligned to the ideal ECAE orienta-
tions, which also correspond to the ideal simple shear
textures obtained via torsion [7072]. N = 2 pass PFs
(Fig. 10a) indicate that while the orientations remain dis-
tributed along the partials described above, there are some
significant differences in terms of orientation distribution
and intensity when compared to their 1-pass counterparts
(Fig. 9b). The near-monoclinic symmetry of 1-pass ECAE
is lost, the Ch, Ah and A
h orientations have weakened, and
the components near the Bh and B
h orientations have
strengthened. Overall, this leads to incomplete and non-
uniform orientation densities along the fibers.
After N = 4 (Fig. 10b), the incompleteness and non-
uniformity of the fibers remain as orientation densities
differ and also result in an overall reduction in the maxi-
mum texture intensity (f(g)). As a direct consequence of
further processing, minor peaks belonging to retained
components from previous passes are also observed.
Compared to the 1 pass specimen, the end of N = 4 passes
results in significant weakening of intensities around the
A*2h and Ch components accompanied by the moderately
stronger A*1h orientation. When comparing the relative
strengths of orientation densities among the various com-
ponents after N = 8 (Fig. 10c), the results are similar to
N = 4 with comparatively lesser f(g) recorded for intensi-
ties around A*2h and Ch However, despite the
reproducibility in texture components, larger disparities do
exist. In direct comparison to the N = 4 condition, N = 8
resulted in a further strengthening of A*1h and B
h orienta-
tions. The most notable difference in terms of orientation

Table 2 Ideal orientations and fibers for fcc materials after a single pass 90 ECAE
Notation Euler angles () for /2 = 0 and 45 sections Equivalent TD-rotation Fiber family
by h = 45
/1 / /2

A*1h 80.3, 260.3 45 0 111

1
12h {111}h
170.3, 350.3 90 45
A*2h 9.7, 189.7 45 0 11111

2h {111}h
99.7, 279.7 90 45
Ah 45 35.3 45 (111)[110]h f111gh , h110ih

h
A 225 35.3 45 (1 1 1) [1 1 0]h f111gh , h110ih


Bh 45, 165, 285 54.7 45 1

12110 h 110ih



h
B 105, 225, 345 54.7 45

112

1
10h 110ih



Ch 135, 315 45 0 001 110ih 110ih
45, 225 90 45

123
9106 J Mater Sci (2007) 42:90979111

Fig. 9 (a) (111) pole figure A / A A *1 A*2 B / B C

showing ideal orientations
(symbols) and partial fibers y, ND
(thick solid lines) after a single y ND
A (a) (b)
pass of ECAE deformation for *
A A 1
fcc materials using a U = 90 45o
A*2
die [67]. (b) (111) pole figure of
the experimental textures after
N = 1. Contour levels: 0.5 ED
x, ED
<110>

{111} A*2
*
A 1 A
Max = 3.02
A

Fig. 10 (111) pole figures of (a) ND

(b) ND
(c) ND
the experimental textures after
N = (a) 2, (b) 4, (c) 8, (d) 12
and, (e) 16 passes, route BC
ECAE. Contour levels: 0.5 ED ED ED

Max = 2.78 Max = 2.22 Max = 3.88

(d) ND (e) ND

ED ED

Max = 1.95 Max = 2.21

densities between N = 4 compared to N = 12 and 16
(Fig. 10d, e) is that greater strain not only produces a net
spread of orientation components but also results in a
weakening of all ideal shear-type orientations in agreement
with previous work [67, 7274]. Interestingly and common
to N = 4, 8, 12 and 16 pass conditions, the re-appearance
of weak texture components of the initial N = 0 texture
reflects glide-reversal effects that return part of the
substructure into an equiaxed configuration.
It must be noted that the weakening trend of f(g) is in
contradiction with other investigations [75, 76] that pos-
tulate an increase in texture strengthening with greater
imparted strain. Mainly, these disparities allude to the
heterogeneity of the as-deformed billet due to varying
friction conditions at the billet-die interface [69, 73, 77]
and also the degree of difficulty in locating the exact center
of the billet stable length (from where the ED-plane sample
is cut) for equivalent texture scans with increasing N. In
addition, differences such as: (i) the starting texture of the
material at N = 0 [74] and, (ii) the area of the sample over
which the texture scan is performed [67, 75] also mean that
direct comparisons of our texture results with those found
in literature must be made with care.
Mechanical properties of ECAE-Cu
During ECAE processing, accurate stress and strain levels
cannot be quantified due the operation of friction forces at
the billet-die interface. Consequently, variations in material
properties as a function of the employed strain-path chan-
ges between successive passes cannot be properly
evaluated. From strain reversal experiments, it is well
known that a change in the strain path produces a change in
the internal stressgenerally described as the Bauschinger
effect. In cases, where the reloading stress exceeds the pre-
deformation stress level, a dissolution of the previous cell
structures and a untangling of formerly strong dislocation
arrangements within the walls is accompanied by a stag-
nation of the strain hardeningbefore a steady state is

123
J Mater Sci (2007) 42:90979111 9107

re-established after few percentage of reloading strains ([3] 450 50

and references therein). Therefore, if the deformation 45

Total strain [%]

400
behaviour of pre-strained materials is being evaluated, it is

UTS [MPa]
40
350
important to account for such effects. 35
300 30
250 25
Tensile testing 20

YS,
200

Uniform,
15
150
Figure 11 shows the true strain true stress curves in (a) 10
100
tension and, (b) compression derived from engineering 5
50 0
stressstrain curves using standard equations. Alterna-
tively, Fig. 12 shows a quantitative summary of tensile test 0 2 4 6 8 10 12 14 16

data after N = 0, 1, 4, 8, 12 and 16 passes represented in Number of passes

terms of yield and ultimate tensile strength (YS and UTS),
Fig. 12 Tensile test data of Cu processed up to N = 16 passes for the
uniform and total strain before failure. The annealed N = 0 yield strength (YS), ultimate tensile strength (UTS) and uniform and
specimen shows a low yield strength, followed by a long total elongation
period of uniform strain hardening and a short post-necking
elongation. On the other hand, ECAE-Cu specimens show
significantly higher yield strengths, followed by a short true strain curves cannot capture the effective stress and
amount of uniform elongation with steep hardening and a strain in the necking part of the sample after having
larger post-necking elongation. Note that the true stress reached the UTS. The YS and UTS increase up to N = 4
and decrease moderately thereafter. A similar observation
has been made on Cu deformed via route C up to N = 16
(a) 500 using a U = 120 die angle [78]. Generally, tensile tests
N=0
N ==1
indicate only a very small increase in uniform plastic
400
N=2 elongation and a total elongation (etot) below 10% for all
N=4
N=8 passes. These values are within the range prescribed by
N = 12
True stress [M Pa]

300 N = 16
[79].
Slightly longer post-necking elongation values, but still
much larger than the uniform strain, have been recently
200
reported for specimens deformed for N = 16 and 20 via
route BC (etot * 15%) and for N = 2 to 4 via route C
100 (etot * 1719%) [37, 8082]. Yet, significantly larger total
elongation has been reported for higher numbers of passes
0 using route BC [83]. Recently, a total elongation value of
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
etot * 35% (which exceeds that of the annealed N = 0
True strain sample) was presented for Cu samples subjected to routes
(b) 500 C and E for N = 8 and 16 passes processed in a U = 120
angle die [84]. Since the tensile sample dimensions in these
400 results are not indicative of any consistent size effect, other
reasons have to be considered to explain the measured
True stress [MPa]

300
differences in total and/or uniform elongation. Note that
the UTS for the tensile strains given in [37, 8084] are all
within the same range as measured here, indicating that the
200 N=0
N=1 enhanced strain is not due to a larger grain size.
N=2
N=4
Wang and Ma [40] suggested that an increase in the
100 N=8 strain rate sensitivity (SRS) would be one way to improve
N = 12
N = 16 the tensile ductility as superplastically deformed materials
0 have typical SRS values that range between 0.3 and 0.8.
0.00 0.05 0.10 0.15 0.20 0.25
Earlier, such an explanation was also given by Valiev et al.
True strain [83] for explaining the high tensile strength of their ECAE
Fig. 11 Stressstrain curves for 016 passes in tension (a) and Cu samples deformed at RT [83]. However, our data
compression (b) extracted from strain rate jump tests in compression shown

123
9108
Table 3 Compressive yield strength and strain rate sensitivity (m) of Cu specimens subjected to N = 116 passes ECAE (route BC)
Mechanical property Number of passes (N)
0 1
ry (0.2%) (MPa) 60 324
M 0.007 0.001 0.009 0.003 0.015 0.005
in Table 3 (also plotted in [85]), indicate that the SRS
value saturates at *0.02 and is in agreement with the
values given in [79] where smaller strain rate steps in-
between the jumps had been used. It should be noted that
such low values of SRS are not sufficient to delay defor-
mation instability (and consequent necking) according to
Harts criterion [86]. As pointed out by Wang and Ma, only
if the strain rate is sufficiently low (*106 s1) and
enough time is given for the operation of diffusional
mechanismswill an increase in the SRS occur and pro-
duce an observable increase in the uniform elongation [40].
It has also been suggested that increasing the percentage
volume fraction of HAGBs proves beneficial to the uniform
elongation [83, 87]. A recent study on a Cu0.36Cr alloy
deformed by route A up to N = 16, exhibits uniform and
total elongations of *10% and *23% respectively, and a
UTS of 410 MPa. On the other hand, a sample deformed
up to N = 16 via an alternated route C and A type pro-
cessing schedule, showed uniform and total elongations of
*4% and *20%, respectively and a UTS of 435 MPa.
EBSD studies on these samples indicate the prevalence of a
greater fraction of HAGBs (fHAGB *6775%) for the
sample with greater uniform strain compared to the sample
(fHAGB *5263%) exhibiting less uniform strain [88].
Qualitatively, this is in agreement with our observations
wherein a lower fraction of HAGB and a lower uniform
elongation was measured for samples at N  4 compared
to those deformed for N > 4. It had been suggested [83, 87]
that a microstructure comprising mainly HAGBs would
promote deformation mechanisms other than those medi-
ated by dislocations. But taking the low magnitude of
measured SRS values into consideration (see Table 3),
mechanisms which are solely dependent on diffusion pro-
cess are of minor importance. However, diffusion
phenomena are indirectly connected to the major operative
mechanismwhich is characterised by dislocation glide
through the crystals and their accommodation and evolu-
tion within the cell wall structure. The possibility of texture
effects (whereby an increase in ductility via dislocation slip
along more preferential orientations can be induced) is also
negated as the present texture results show a generally
route-specific and reproducible trend of preferred orienta-
tions with increasing N.
In order to further clarify this question, we have also
carried out detailed TEM and XRD analysis on the
123
J Mater Sci (2007) 42:90979111
4 8 12 16
369 332 346 340
0.018 0.003 0.021 0.002 0.022 0.002
evolution of the deformation microstructure with respect to
N [18]. With increasing N (up to N = 16), a decrease in the
wall thickness and dislocation density within the subgrains
occurs. All the while, the cell or subgrain size remains
roughly constant. These findings are in agreement with a
corresponding decrease in the microstrain measured by
XRD (see Ref. [18] for more details). This would strongly
suggest that the mean free path of dislocations increases in
samples processed beyond N  4. It can be therefore
suggested that an increased volume fraction of HAGBs
(which themselves are known to be effective sinks for
incoming lattice dislocations), and a lower dislocation
density within subgrains (and hence, a greater mean free
path of dislocations) help reduce sources of plastic insta-
bility such as stress concentrations and thereby makes
possible an increase in percent elongation. Apropos to the
agreement between microstructural analysis of the ECAE
material performed before mechanical testing and conse-
quent mechanical behaviour during testing, it can be
assumed that microstructural changes occurring during the
few percentage of strain reached during tensile testing are
of minor importance for the interpretation of the stress
strain curves.
It should also be noted that while the above mechanisms
could explain the mechanical behaviour of SPD-UFG
materials to a very large extent, another source of incon-
sistency in the available literature data could originate from
a higher material-specific sensitivity to external flaws, a
phenomena which is also known to affect nc metals [89].
Thus, further experimental results are still needed to
resolve these differences and/or ambiguities.
Strain hardening stages III, IV and V measured in
compression
Unlike tensile tests, uniaxial compression tests do not
suffer from necking instability and can be used to describe
the strain hardening behaviour. In the following passages,
the hardening behaviour of the ECAE deformed samples is
described in the classical context of strain hardening theory
reviewed in [13]. In doing so, the effects of strain path
changes from simple shear (or, ideal ECAE) to pure shear
(compression) are neglected. Nevertheless, this description
showed usefulness in describing the hardening behaviour
J Mater Sci (2007) 42:90979111 9109

(a) 1000 Table 4 Calculated shear stress values for the different hardening
N=1 stages of Cu specimens subjected to N = 116 passes ECAE, via route
N=2 BC processing
800 N=4
N=8 Shear stress Number of passes (N)
N = 12
600 N = 16 1 2 4 8 12 16
[MPa]

sIIIs 116 142 144 145 139 139

400
sIV 118 144 148 148 141 143
sV n.a. n.a. 158 156 149 154
200
n.a. = not achieved

0
100 110 120 130 140 150 160 170
III S During stage III, a net increase in the dislocation density
[MPa] III S
is still prevalent, which when contrasted with stage II is
(b) 120 less dominant and bears a dislocation annihilation com-
N=1
100 N=2 ponent [13]. As compared to stage II, stage III is also
N=4 commonly described by a higher strain rate and tempera-
N = 16
80 ture sensitivity which increases with increasing stress and
is associated with processes of dynamic recovery [13, 91].
[M P a]

60 During stage IV, the annihilation rate of dislocations

40
IV 20
III IV III
0 increase in the SRS with increasing stress in the one pass
120 130 140
IV IV
[MPa]
Fig. 13 Strain hardening behaviour of Cu processed by ECAE for
N = 116 passes. In (b) a close-up of (a) is shown
as a function of the number of passes and is in agreement
with the general microstructural and hardening response
for stages IIV.
In Fig. 13, the strain hardening versus the shear stress of
samples up to N = 116 is shown. The shear stress has
been calculated by dividing the true stress by a constant
Taylor factor (MT) of 3.06, which corresponds to the value
of a non-textured sample.3 The shear strain hardening rate
results from the conversion of the true stress and strain via
or=oe
1=MT2 os=oc; where s is the shear stress
and c the shear strain. By linear extrapolation of the steep
negative slope, a saturation stress (sIIIs) for stage III
hardening can be estimated. This value reaches a maximum
between N = 48 (Fig. 13a; Table 4). It is also important
to note that despite reaching stages IV and V hardening
during ECAE, a clear indication of the re-occurrence of
stage III during post-ECAE compression straining is visible
in the hardening curves [85]. This observation has also
been recently recognised for ECAE Ni [90].
3
Possible effects on the shear stress due to changes in the Taylor
factor during straining have been neglected.
IV
V becomes as strong as the storage rate so the hardening
capacity declines and leads to a steady SRS state for a
corresponding change in stress [3, 91]. In agreement with
IV V
the description of stages III and IV, we also observe an
150 160 170 specimen followed by a plateau of the SRS level (see
V
S Fig. 6 in Ref. [85]) measured in the two pass specimen. In
Fig. 13b, a further change in the slope after stage IV can be
observed for the specimens subjected to N  4 and is
denoted as stage V. This stage shows a decrease in HV
values with stress and leads to a final saturation stress (ss)
before friction and barrelling effects negate it at large
compression strains. However, in the case of multi-pass
ECAE, the dependency of the SRS on the stress deviates
from the classically described mechanisms ascribed to
stage V, where an increase in SRS values with increasing
stress and strain should be observed [91]. Conversely, our
results for N  4 show a decrease in the SRS with
increasing stress and strain (Figs. 5, 6 in Ref. [85]). This
behaviour linked to the saturation limit reached in terms of
accumulated stress and dislocation density for N  4 can
be associated with the change in the strain path from ECAE
route BC, to compression which causes a destruction and
reorganisation of the microstructure: 2030% larger cell
wall sizes have been measured for the compressed N = 4
samples as compared to non-compressed N = 4 samples,
while for N = 16 pass samples no clear differences in cell
size have been observed between compressed and un-
compressed samples. Recovered grains with sharp grain
boundaries and dislocation-free interiors often observed in
the un-compressed N = 16 samples are not present after
compression testing to 40%, but exchanged by grains
which are decorated with a new dislocation structure (see

123
9110
Fig. 4 in Ref. [85]). Similar microstructural observations in
strain reversal experiments are also associated with the
Bauschinger effect. According to these observations, it is
currently not clear how far the stage V hardening observed
in Fig. 13 is influenced by the change in the strain path.
Conversely, as a change in the stress causes a corre-
sponding change in the SRS, the apparent activation
volume Vap kT D ln c_ Ds is also transformed for a given
change in the hardening stage (see Fig. 7 in [85]). For
N > 4, activation volumes between 48 and 110 b3, where
b is the Burgers vector, have been measured with lowest
values reached at very low strain rates, which are achieved
in relaxation tests (for further details, see [85]). These
values are also in agreement with those presented recently
by Wei et al. [92]. Separating the effective activation
volume into its constituents Veff b
x
l; (with x being
roughly equal b), a length scale (l) of *1530 nm results.
This length scale is in the same range as deduced from our
TEM analysis of the dislocation density within cell walls
(qwall *1.9 1015 m2 [18]) if it is assumed that
p References
lTEM 1= qwall : This gives credence to the assumption that
the rate controlling mechanism in ECAE Cu at higher N
(  8) is the interaction of lattice dislocations with cell wall
or subgrain boundaries as also stated by [92, 93].
Summary
This work summarises our and current understanding of the
microstructural evolution and the mechanical behaviour of
ECAE pre-deformed Cu via route BC. While good agree-
ment on the sub-grain/grain size and shape as a function of
number of passes (N) results from a comparison with
available literature values, a less consistent picture can be
deduced from misorientation measurements. Our results
performed by electron backscatter diffraction (EBSD)
indicate an increase in the amount of HAGBs with
increasing N, which saturates at about 50%. In agreement
with literature an increase in the amount of R3 boundaries,
stacking faults, recovered or even partially recrystallised
grains and decrease in boundary width with increasing N is
observed. Consistent with previous texture work a spread
in the texture components away from the ideal simple shear
components is observed with greater N.
Upon comparing with available literature on the
mechanical behaviour of ECAE Cu good agreement has
been found in terms of maximum yield and tensile stress
reached. In this study a softening after N = 8, in agreement
with microstructural evolution is observed. With respect to
the maximum tensile strain reached before failure in ECAE
Cu larger disparities were found, which might be associ-
ated with testing conditions and external flaws. Here, total
elongations do not exceed 10% strain for all passes, but do
123
J Mater Sci (2007) 42:90979111
show a slight increase in uniform elongation for greater N.
Despite the large amount of pre-strain imparted via ECAE,
the strain hardening measured during compression tests
indicates a recovery of stage III, which is followed by stage
IV on further deformation. For samples processed to higher
N, a hardening behaviour associated with stage V is also
observed. Strain rate sensitivity values of the order of 0.02
and activation volume estimates, which yield a length scale
of the same order as deduced from dislocation spacings in
the cell wall boundaries suggest that the deformation
kinetics are dictated by the interaction of lattice disloca-
tions with the boundaries.
Acknowledgements Mrs. C. Gu is kindly acknowledged for assis-
tance in texture measurements. This work was supported by the Swiss
National Science Foundation, Grant # 200021-105647 (FHDT), Mo-
nash International Post-Graduate Research Scholarship (AAG) and by
the Australian Research Council Discovery Project, DP0557255
(CHJD, EVP).
1. Sevillano JG, Van Houtte P, Aernoudt E (1981) Prog Mater Sci
25:69
2. Nes E (1998) Prog Mater Sci 41:129
3. Kocks UF, Mecking H (2003) Prog Mater Sci 48:171
4. Valiev RZ, Islamgaliev RK, Alexandrov IV (2000) Prog Mater
Sci 45:103
5. Segal VM, Reznikov VI, Drobyshevskiy AE and Kopylov VI
(1981) Russian Metall (Engl. Transl.) 1:115
6. Segal VM (1995) Mater Sci Eng A 197:157
7. Zhorin VA, Shashkin DP, Yenikoponyan NS (1984) DAN, SSSR
278:144
8. Kuznetsov RI, Bykov VI, Chernyshov VP, Pilyugin VP, Yefre-
mov NA, Posheyev VV, Sverdlovsk (1985) IFM UNTS RAN,
Preprint 4/85 (in Russian)
9. Bridgman PW (1952) Studies in large plastic flow and fracture.
McGraw-Hill, New York and London, p 174 (Reprinted by 870
Harvard University Press, Cambridge, Mass., USA, 1964)
10. Birringer R, Gleiter H, Klein H-P, Marquardt P (1984) Phys Lett
A 102:365
11. Gleiter H (1989) Prog Mater Sci 33:223
12. Lowe TC (2006) Mater Sci Eng 503504:355
13. Iwahashi Y, Horita Z, Nemoto M, Langdon TG (1998) Acta
Mater 46:3317
14. Ferrasse S, Segal VM, Hartwig KT, Goforth RE (1997) Metall
Mater Trans A 28:1047
15. Furukawa M, Iwahashi Y, Horita Z, Nemoto M, Langdon TG
(1998) Mater Sci Eng A 257:328
16. Lapovok RYe (2005) J Mater Sci 40:341 DOI:10.1007/
s10853-005-6088-0
17. Bowen JR, Gholinia A, Roberts SM, Prangnell PB (2000) Mater
Sci Eng A 287:87
18. Dalla Torre FH, Lapovok R, Sandlin J, Thompson PF, Davies
CHJ, Pereloma EV (2004) Acta Mater 52:4819
19. Humphreys FJ (2001) J Mater Sci 36:3833 DOI:10.1023/
A:1017973432592
20. Bowen JR, Mishin OV, Prangnell PB, Juul Jensen D (2002)
Scripta Mater 47:289
21. Bunge H-J (1982) Texture analysis in materials science: mathe-
matical methods, 1st edn. Butterworth & Co., Berlin
J Mater Sci (2007) 42:90979111
22. Valiev RZ, Krasilnikov NA, Tzenev NK (1991) Mater Sci Eng A
137:35
23. Valiev RZ, Mulyukov RR, Ovchinnikov VV (1990) Phil Mag
Lett 62:253
24. Valiev RZ, Kozlov EV, Mulyukov RR (1993) Mater Sci Eng A
168:141
25. Valiev RZ, Mulyukov RR, Ovchinnikov VV, Shabashov VA
(1991) Scripta Metall Mater 25:2717
26. Valiev RZ, Korznikova GF, Mulyukov KY, Mishra RS, Muk-
herjee AK (1997) Phil Mag B 75:803
27. Valiev RZ, Kozlov EV, Ivanov YuF, Lian J, Nazarov AA, Bau-
delet B (1994) Acta Metall Mater 42:2467
28. Mishin OV, Gottstein G (1998) Philos Mag A 78:373
29. Mishin OV, Juul Jensen D, Hansen N (2000) In: Hansen N,
Huang X, Juul Jensen D (eds) Proc. 21nd Ris international symp
mater. sci.: recrystallisation-fundamental aspects and relations to
deformation microstructure. National Laboratory, Roskilde,
Denmark, p 445
30. Mishin OV, Juul Jensen D, Hansen N (2003) Mater Sci Eng A
342:320
31. Agnew SR (1998) Ph.D. thesis, Northwestern University
32. Hughes DA, Hansen N (2000) Acta Mater 48:2985
33. Mishin OV, Huang X, Brown JR, Juul Jensen D (2001) In:
Hansen N, Huang X, Juul Jensen D (eds) Proc. 22nd Ris
internat. symp. mater. sci.: recrystallisation-fundamental aspects
and relations to deformation microstructure. National Laboratory,
Roskilde, Denmark, p 335
34. Iwahashi Y, Horita Z, Nemoto M, Langdon TG (1997) Acta
Mater 45:4733
35. Komura S, Horita Z, Nemoto M, Langdon TG (1999) J Mater Res
14:4044
36. Baik SC, Hellmig RJ, Estrin Y, Kim HS (2003) Z Metallkd 94:759
37. Vinogradov A, Hashimoto S, Patlan V, Kitagawa K (2001) Mater
Sci Eng A 319321:862
38. Mingler B, Karnthaler HP, Zehetbauer M, Valiev RZ (2001)
Mater Sci Eng A 319321:242
39. Hoppel HW, Zhou ZM, Mughrabi H, Valiev RZ (2002) Philos
Mag A 82:1781
40. Wang YM, Ma E (2004) Acta Mater 52:1699
41. Huang WH, Yu CY, Kao PW, Chang CP (2004) Mater Sci Eng A
366:221
42. Wu SD, Wang ZG, Jiang CB, Li GY, Alexandrov IV, Valiev RZ
(2004) Mater Sci Eng A 387389:560
43. Han S, Lim C, Kim C, Kim S (2005) Metall Mater Trans A
36:467
44. Vinogradov A, Suzuki T, Hashimoto S, Kitagawa K, Kuznetsov
A, Dobatkin S (2006) Mater Sci For 503504:971
45. Dalla Torre FH, Gazder AA, Gu CF, Davies CHJ, Pereloma EV,
Met Mater Trans A (submitted)
46. Etter AL, Solas D, Baudin T, Penelle R (2005) Mater Sci For
495497:845
47. Baik SC, Estrin Y, Kim HS, Hellmig RJ (2003) Mater Sci Eng A
351:86
48. Chang CP, Sun PL, Kao PW (2000) Acta Mater 48:3377
49. Zehetbauer MJ, Steiner G, Schafler E, Korznikov A, Korznikova
E (2006) Mater Sci For 503504:57
50. Huang CX, Wang K, Wu SD, Zhang ZF, Li GY, Li SX (2006)
Acta Mater 54:655
51. Liao XZ, Zhao YH, Srinivasan SG, Zhu YT, Valiev RZ, Gun-
derov DV (2004) Appl Phys Lett 84:592
52. Liao XZ, Zhao YH, Zhu YT, Valiev RZ, Gunderov DV (2004)
J Appl Phys 96:636
53. Meyers MA, Vohringer O, Lubarda VA (2001) Acta Mater
49:4025
54. Meyers MA, Andrade RU, Chokshi HA (1995) Metall Mater
Trans A 26:2881
9111
55. Christian JW, Mahajan S (1995) Prog Mater Sci 39:1
56. El-Danaf E, Kalidindi SR, Doherty RD (1999) Metall Mater
Trans A 30:1223
57. Randle V (2004) Acta Mater 52:4067
58. Sutton AP, Balluffi RW (1996) Interfaces in crystalline materials.
Oxford Science Publications, Clarendon Press, p 305
59. Venables JA (1961) Philos Mag 6:379
60. Yamakov V, Wolf D, Phillpot SR, Mukherjee AK, Gleiter H
(2002) Nat Mater 1:45
61. Chen MW, Ma E, Hemker KJ, Sheng HW, Wang YM, Cheng
XM (2003) Science 300:1275
62. Liao XZ, Zhou F, Lavernia EJ, Srinivasan SG, Baskes MI, He
DW, Zhu YT (2003) Appl Phys Lett 83:632
63. Liao XZ, Zhao YH, Srinivasan SG, Zhu YT, Valiev RZ, Gun-
derov DV (2004) Appl Phys Lett 84:592
64. Rosner H, Markmann J, Weissmuller J (2004) Phil Mag Lett
84:321
65. Meyers MA, Murr LE (1978) Acta Metall 26:951
66. Mahajan S, Pande CS, Imam MA, Rath BB (1997) Acta Mater
45:2633
67. Li S, Beyerlein IJ, Alexander DJ, Vogel SC (2005) Scripta Mater
52:1099
68. Li S, Beyerlein IJ, Bourke MAM (2005) Mater Sci Eng A 394:66
69. Li S, Beyerlein IJ, Necker CT, Alexander DJ, Bourke MA (2004)
Acta Mater 52:4859
70. Agnew SR, Weertman JR (1998) Mater Sci Eng A 242:174
71. Huang WH, Chang L, Kao PW, Chang CP (2001) Mater Sci Eng
A 307:113
72. Gholinia A, Bate P, Prangnell PB (2002) Acta Mater 50:2121
73. Gazder AA, Dalla Torre FH, Gu CF, Davies CHJ, Pereloma EV
(2006) Mater Sci Eng A 415:126
74. Ferrasse S, Segal VM, Kalidindi SR, Alford F (2004) Mater Sci
Eng A 368:28
75. Gubicza J, Balogh L, Hellmig RJ, Estrin Y, Ungar T (2005)
Mater Sci Eng A 400401:334
76. Cao WQ, Godfrey A, Liu W, Liu Q (2003) Mater Sci Eng A
360:420
77. Li S, Bourke MAM, Beyerlein IJ, Alexander DJ, Clausen B
(2004) Mater Sci Eng A 382:217
78. Shih MH, Yu CY, Kao PW, Chang CP (2001) Scripta Mater
45:793
79. Wang YM, Ma E (2003) Appl Phys Lett 83:3165
80. Haouaoui M, Karaman I, Maier HJ, Hartwig KT (2004) Metall
Mater Trans A 35:2935
81. Wang JT, Du ZZ, Kang F, Chen G (2006) Mater Sci For 503
504:663
82. Krishanaiah A, Chakkingal U, Venugopal P (2006) Mater Sci For
503504:733
83. Valiev RZ, Alexandrov IV, Zhu YT, Lowe TC (2002) J Mater
Res 17:5
84. Maier HJ, Gabor P, Gupta N, Karaman I, Haouaoui M (2006) Int
J Fatigue 28:243
85. Dalla Torre FH, Pereloma EV, Davies CHJ (2006) Acta Mater
54:1135
86. Hart EW (1967) Acta Metall 15:351
87. Valiev RZ (2003) Adv Eng Mater 5:296
88. Vinogradov A, Ishida T, Kitagawa K, Kopylov VI (2005) Acta
Mater 53:2181
89. Dalla Torre FH, Van Swygenhoven H, Victoria M (2002) Acta
Mater 50:3957
90. Hollang L, Thiele E, Holste C, Brunner D (2006) Mater Sci Eng
A 424:138
91. Zehetbauer M, Seumer V (1993) Acta Metall Mater 41:577
92. Wei Q, Cheng S, Ramesh KT, Ma E (2004) Mater Sci Eng 381:71
93. Wang YM, Ma E (2004) Appl Phys Lett 85:2750
123
J Mater Sci (2007) 42:91129115
DOI 10.1007/s10853-007-2065-0
LETTER
Accelerated bainitic transformation during cyclic austempering
Vivekanand Sista Philip Nash Satyam S. Sahay
Received: 21 May 2007 / Accepted: 26 July 2007 / Published online: 13 August 2007
Springer Science+Business Media, LLC 2007
Austempering is an important thermal processing opera-
tion, where strong and tough bainitic steel is produced in a
single heat treatment [1, 2]. During the austempering pro-
cess, the steel is first austenitized and then cooled rapidly
just above the martensite start temperature until bainite
nucleates and grows, usually until the transformation stops
and then it is cooled to room temperature. Due to the
sluggish solid-state transformation kinetics, industrial
austempering necessitates isothermal holds of 224 h,
depending on the size and composition of steel. In contrast
to conventional isothermal processing, cyclic thermal
processing has been shown to accelerate the kinetics of
several phase transformations [35], with a significant
beneficial impact on productivity and energy consumption
of these energy intensive operations. This was attributed to
the non-isothermal effects resulting from cyclic treatment.
The non-isothermal effect on phase transformations has
also been utilized to enhance the productivity of a modern
batch annealing operation [6]. In the present work, the
effect of cyclic processing on austempering kinetics has
been compared with conventional isothermal processing
for 1080 steel. Furthermore, the effect of cyclic frequency
on the austempering kinetics has been studied.
Austempering kinetics experiments were performed on
6 mm diameter cylindrical samples of a 1080 steel using
a GleebleTM 3500 thermo-mechanical simulator (DSI
V. Sista
P. Nash (&)
Thermal Processing Technology Center, IIT, 10 W 32nd St.,
Chicago, IL 60616, USA
e-mail: philip.nash@iit.edu
S. S. Sahay
Division of Tata Consultancy Services Ltd., Tata Research
Development and Design Centre, 54B, Hadapsar Industrial
Estate, Pune 411 013, India
123
Poestenkill, NY). A diametrical dilatometer was mounted
on the specimen to measure the diameter change during the
thermal processing. The austempering experiments were
performed in two cycles, where the first cycle provides the
same initial microstructure prior to each experiment. After
completing the first cycle, the cylindrical specimens were
heated to the austenitizing temperature (850 C), held for
5 min and then cooled to different austempering tempera-
tures, where the bainite transformation was monitored for
the desired period of time, followed by cooling to room
temperature. The cooling rate was sufficiently fast to avoid
any transformation occurring before reaching the austem-
pering temperature (Fig. 1). The isothermal experiments
were carried out at austempering temperatures of 260 and
300 C, whereas the cyclic experiments were carried out
between 260 C and 300 C at two different heating/
cooling rates of 1 and 5 C/min. The percentage of bainitic
transformation as a function of time was computed from
the dilation curve [2] in conjunction with the microstruc-
tural examination by optical microscopy and SEM.
The normalized dilatation curves for isothermal aus-
tempering at 260, 300 C, and cyclic austempering
between 260 C and 300 C for 1 C/min and 5 C/min are
shown in Fig. 2. In this figure, only the portion of the curve
corresponding to austempering is shown and the dilatation
curve has been offset to start at zero. The plateau of the
dilatation curve with respect to the time represents the end
of the transformation to bainite. It must be noted that the
end of the bainite transformation does not necessarily
correspond to 100% volume fraction of bainite [1]. Previ-
ously, using the same type of Gleeble experiments, it was
shown that the end of the transformation corresponds
closely to the austenite carbon content corresponding to a
T0 equal to the austempering temperature [2, 7]. Further-
more the transformation dilatation at the end of the
J Mater Sci (2007) 42:91129115 9113

1000
o
850 C, 5 min
Temperature,oC

750

500
o
Austempering: 260-300 C
250
o
Ms ~231.2 C
0
Time

Fig. 1 Austempering experiment temperature-time profiles for 1080

steel

transformation was less the higher the austempering tem-

perature as expected with less bainite fraction [8]. In this
work the total transformation dilatation calculated at 260
and 300 C from Fig. 2 are 0.0069 and 0.0065, respec-
tively. The dilatation at 260 C is greater at 300C
corresponding to a higher fraction transformed, as expected
at lower austempering temperatures. The percentage bai-
nite at the end of the bainitic transformation was estimated
from optical and SEM microscopy. The dilatation versus
time profiles for long times (250300 min) were used to
determine the actual time for complete transformation for
both the isothermal and cyclic processes. It can also be
observed from this figure that the cyclic austempering
kinetics is faster than the isothermal austempering. As
shown in Fig. 3, the isothermal austempering time for the
complete bainite transformation requires 160 and 140 min
at 260 and 300 C, respectively. In contrast, for cyclic
austempering at 1 and 5C/min the transformation is
completed in much shorter times of about 80 20 and
32 4 min, respectively. The error limit is due to the
estimation of the completion time from the plateau of the
dilatation curve, at both 260 and 300 C and allowing for
the difference in dilatation caused by thermal expansion/
contraction this can be estimated to within the time for half
a thermal cycle. This conclusion is also evident from the
SEM microstructures shown in Fig. 4, where islands of
retained austenite are visible in the isothermally trans-
formed samples, whereas the cyclically austempered
samples reveal completely transformed bainitic sheaths
after 32 min. It must be noted that the morphology of the
isothermally austempered sample at 260 C is slightly
different at 300 C, whereas the cyclic austempering
sample exhibits mixed morphology. The exact nature of the
Fig. 2 Dilation as a function of time; (a) isothermal austempering at
accelerated transformation kinetics will be the subject of 260 C for 240 min (b) austempered at 300 C for 240 min (c) cyclic
further work. austempering between 260 C and 300 C at 1 C/min for 330 min
In the present work, it has been shown that cyclic aus- (d) cyclic austempering between 260 C and 300 C at 5 C/min for
tempering can result in up to about an 80% reduction in 330 min

123
9114 J Mater Sci (2007) 42:91129115

Austempering Time, min 200

Isothermal
150
Cyclic

100

50

0
260 C 300 C 1 C/min 5 C/min
Austempering Condition

Fig. 3 The time to complete the bainitic transformation during

isothermal and cyclic austempering

time for complete bainitic transformation as compared to

conventional isothermal austempering. This has direct
implications for the productivity of industrial austemper-
ing operations. Furthermore, increase in cyclic frequency
from 1 C/min to 5 C/min resulted in about a 50%
reduction in transformation time. As has been explained
earlier, [3] these results cannot be explained on the basis
of the prevalent quasi-isothermal approach as that does
not capture non-isothermal effects, such as heating rate
and temperature reversal effects imposed during cyclic
annealing. Although these results are in-line with the
cyclic recrystallization and grain growth kinetics in cold
rolled steel samples, the magnitude of accelerated kinetics
(80% increase for austempering) observed in the present
work is much higher than the earlier [3] observations of
around 20% enhancement during recrystallization and
grain growth kinetics. This is likely due to additional
effects for bainite transformation, such as enhanced
nucleation and relief of transformation stresses. Further
work is also underway to model and validate the cyclic
austempering behavior, as well as to optimize the cyclic
frequency and amplitude for maximizing the austemper-
ing kinetics and examining its feasibility for industrial
implementation.
In summary, we have shown that the bainitic transfor-
mation kinetics during austempering is significantly
accelerated (up to 80% reduction in time) during cyclic
processing as compared to the conventional isothermal
processing. These results could have a significant impact Fig. 4 SEM Microstructure after (a) isothermal austempering at
on industrial austempering productivity. Further work is 260 C for 32 min (b) isothermal austempering at 300 C for 32 min
(c) cyclic austempering between 260 C and 300 C at 5 C/min for
underway to optimize the cyclic austempering profile, the 32 min
mechanism of accelerated transformation and its feasibility
for industrial implementation.
The authors wish to thank Professor Mathai Joseph, Services Ltd. The authors are also grateful to Mr. Karthik
Executive Director, TRDDC for providing financial sup- Krishnan and Dr. Srinivasan Raghavan of TRDDC for
port during the summer of 2006 to the first author under the helpful discussions and to the TPTC at IIT for the provision
Global Student Internship Program of Tata Consultancy of the material and financial support of the project.

123
J Mater Sci (2007) 42:91129115
References
1. Bhadeshia HKDH (ed) (2001) Bainite in steels. Cambridge
University Press, London, UK
2. Chupatanakul S, Nash P (2006) J Mater Sci Lett 41:4965
3. Sahay SS, Malhotra CP, Kolkhede AM (2003) Acta Mater 51:339
9115
4. Schuh CA, Dunand DC (2002) Acta Mater 50:1349
5. Geng H, He S, Lei T (1997) Mater Trans 28A:1809
6. Sahay SS, Krishnan K, Kulthe M, Chodha A, Bhattacharya B, Das
AK (2006) Ironmaking Steelmaking 33:306
7. Chupatanakul S, Nash P, Chen D (2006) Metals Mater Int 12:453
8. Chupatanakul S (2006) Ph.D thesis, IIT, Chicago, IL, USA
123
J Mater Sci (2007) 42:91169120
DOI 10.1007/s10853-007-2063-2
LETTER
Cracking of densely coated layer adhesively bonded to porous
substrates under Hertzian stress
Kee Sung Lee Sang Kyum Kim Chul Kim
Tae Woo Kim Do Kyung Kim
Received: 27 January 2007 / Accepted: 26 July 2007 / Published online: 12 August 2007

Springer Science+Business Media, LLC 2007
Structural components under contact stresses are designed
so as to have high damage tolerance. Unexpected degra-
dation of initial strength under the components actual
environmental conditions leads to economic loss, and even
potential loss of life, through unanticipated permanent
deformation and fractures. Therefore, material design to
suppress contact stress is of practical importance in engi-
neering applications, such as for the components utilized in
machinery, and bio-, thermal or energy applications [14].
Layered composite design has been investigated with
consideration of damage tolerance in previous studies
[5, 6]. Hard outer layers provide surface resistance against
damage from contact loading and soft underlayers con-
tribute to stress redistribution and delayed fracture of the
coating layer. An adhesive is sometimes used to bond a
hard layer with a soft material in making dental-layered
composites. The coating thickness and the elastic modulus
mismatch are crucial parameters in geometrical and
material designing to suppress Hertzian contact stresses
[7]. The design of such layered composites is one of the
important issues, because a softer support causes the
coating to flex under the constrained contacts.
Recently, we have investigated the cracking of brittle
coatings adhesively bonded to substrates with different
modulus [8]. A trilayer system consisting of an upper soda-
lime glass layer (coating) bonded to a thick glass layer
K. S. Lee (&)
S. K. Kim
C. Kim
T. W. Kim
School of Mechanical and Automotive Engineering, Kookmin
University, Seoul 136-702, Korea
e-mail: keeslee@kookmin.ac.kr
D. K. Kim
Department of Materials Science and Engineering,
Korea Advanced Institute of Science and Technology,
Yuseong, Daejeon 305-701, Korea
123
(substrate) with a thin layer of epoxy adhesive facilitated
in situ observation of cracks during contact loading. We
have investigated the role of substrates of different elastic
modulus with respect to crack initiation by in situ obser-
vations of cone cracking from the surface and radial
cracking from damage at the interface between two layers.
In the present study, we investigate the role of porosity
in the substrate on the cracking of a hard-coating layer. To
this end, we prepared porous substrates with different
porosity and the same epoxy and glass plate. We extended
the system to a hard substrate with a high elastic modulus.
Two starting powders were used to prepare the porous
substrate, Al2O3 (AKP-50, AES-11, ALCOA, Japan), and
SiO2 (SE-8, SO-R31, Tokuyama, Japan). In order to vary
the porosity in the porous substrate, we used different
powder sizes and also varied the sintering temperature. In
Al2O3, the starting powders were 1.22 lm, 5.5 lm, and
9.8 lm in size. We varied the sintering temperatures from
1,000 C to 1,400 C at an increasing interval of 100 C.
Two types of SiO2 powders were used, (SE-8, and
SO-R31), 10 lm and 1 lm, and the ratio of staring powders
was varied, 10 lm: 1 lm = 6:4, 7:3, 8:2, and 9:1, respec-
tively. The powders were sintered at 1,300 C or 1,350 C.
Therefore, we could vary the porosity and the material. The
porosity was controlled in the range of 0.4242.6% and
8.225.4% for Al2O3 and SiO2 substrates, respectively,
through the control of sintering temperature and/or starting
powder size.
The layered composite systems were fabricated from
cover glass with dimensions of 4 10 0.16 mm bonded
to a 4 4 36 mm substrate with a 10 lm thick layer
of epoxy adhesive. The surface was polished to 1-lm
diamond finish before bonding, and both the top and
bottom surfaces of the glass layers were abraded with a
slurry of SiC grit (#600), in order to induce uniform flaw
J Mater Sci (2007) 42:91169120 9117

distribution. The upper glass and the lower substrate were
bonded with epoxy adhesive (Harcos Chemicals, Belles-
ville, NJ, USA) under light pressure.
Spherical (Hertzian) indentation was used to induce
contact stress on the layered composite systems, as shown
in Fig. 1 [914]. Indentations were made with a universal
static testing machine (Microload test system, Unitec.,
Korea, or Instron 5567, Instron, USA) with a tungsten
carbide (WC, J & L Industrial Supply Co., Livonia, MI,
USA) ball of radius r = 3.18 mm at P = 150 N, at a con-
stant crosshead speed of 0.09 mm/min. We have applied
the spherical WC indenter (r = 3.18 mm) up to 150 N
contact load on the layered composites, where a glass
coating with both surfaces abraded to ensure uniform flaw
density was utilized.
The elastic modulus of each layered material is given in
Fig. 2. The graphs in Fig. 2 show the change in the elastic
modulus of the substrate monolith according to the porosity
in the substrate. The elastic modulus of each layer monolith
is measured by acoustic impulse excitation apparatus
(Tektronics, 5800PR, USA) and plotted as a function of
porosity. The elastic modulus of the coating material
monolith and epoxy adhesive are also measured using the
same method and included in the graphs. The elastic
modulus of porous ceramics such as Al2O3 and SiO2
experimentally determined constant [16]. Solid curves are
fits from Eq. 1 and the data fall on a theoretical curve.
In Fig. 2, it is noteworthy that the elastic modulus of
Al2O3 is higher than that of the glass-coating layer, and that
of SiO2 is lower than the coating layer. The epoxy adhesive
has a low modulus, as compared to the Al2O3 and SiO2.
Conversely, Al2O3 ceramics with higher porosity present a
small elastic modulus mismatch after the layered com-
posite is constructed, and SiO2 ceramics with higher
porosity have a large elastic modulus mismatch. A soft
support by epoxy adhesive between the coating and the
substrate layer allows the coating to flex beneath the con-
tact. This leads to changes in the fracture patterns in the
coating layer, from the mode of surface cracking to sub-
surface radial cracking, as found in a previous study [8].
The previous study shows that the initiation of radial
cracking is related with elastic mismatch, the following
equation is applied:
(a) 400
Elastic modulus, E(GPa)

300
decreases as the porosity increases, according to the Rice
equation [1517].
200
E Eo expbP 1
where Eo is the elastic modulus of a material with no
porosity (0% of porosity), P is porosity, and b an 100 glass

adhesive
0
0 10 20 30 40 50
Porosity(%)

(b) 80
glass
Elastic modulus, E(GPa)

60

40
Ring Crack

Dense Glass
20

Radial Crack adhesive

0
Porous Ceramic 0 10 20 30
Porosity(%)
Fig. 1 Schematic diagram of Hertzian indentation test on densely
coating/porous-substrate layered composite bonded with thin Fig. 2 Plot of elastic modulus of each layer material as a function of
adhesive porosity for (a) Al2O3 and (b) SiO2

123
9118 J Mater Sci (2007) 42:91169120

Prad BrF d 2 = logCEc =Es 2 (a)

where B and C are dimensionless constants, rF is the

strength of materials, d coating thickness, and Ec/Es elastic
modulus mismatch. The Eq. 2 indicates the coating fracture
1.0
depends on the elastic modulus of substrate layer material
in the layered composite. On the other hand, the initiation
of ring cracks, Pcone, can be modeled [6, 17] by the 0.8

following equation for cone cracking:

Crack size (mm)

0.6
Pcone A r Gc 3
where Gc is the crack resistance, r the radius of indenter, A 0.4
a dimensionless constant. This critical load does not
depend on the elastic modulus mismatch of layered mate- 0.2
rials. Notably, the above equations are modeled on the
results of all dense layered composites. I II
0.0
A plot of the maximum diameter of ring cracks in the 0 10 20 30 40 50
coating layers using a WC sphere r = 3.18 mm at a load Porosity(%)
P = 150 N is plotted in Fig. 3 as a function of porosity for (b)
two porous ceramic substrates. Data are only plotted for
ring cracks, and the results of crack propagation at
P = 150 N of the contact load are shown. Each data point
represents the mean and standard deviation of 10 indenta-
1.0
tions. The solid curves are best fits. The data shows two
stages for the porous Al2O3 substrate only. In Fig. 3a, the
size of contact cracks initially decreases as the porosity 0.8

increases to a maximum of 20% as shown in stage I,

Crack size (mm)

thereafter, the size of cracks increases as the amount of the 0.6

porosity in the substrate increases to a maximum of 42.6%
as in stage II. The optical micrographs of contact damages
0.4
on the surface for selected conditions are included in the
top of the graph. In these micrographs, it is notable that no
damages are found for the 25% porosity Al2O3 substrate. 0.2

Extensive damage including multiple cracks is apparent for

the 40% porosity Al2O3 substrate. Apparent cracking of the
layered composite with high porosity in the substrate is
found for the SiO2 substrate, as shown in Fig. 3b. The trend
corresponds with that of Al2O3 in II.
The result in Fig. 3b indicates that the contact cracking
of the coating layer crucially depends on the elastic mod-
ulus mismatch, Ec/Es. We have measured the modulus of
the substrate monolith with that of the glass monolith,
while the mismatch increases according to the porosity for
the SiO2 substrate as shown in Fig. 2b. The cracking
damage in Fig. 3b can be understandable, if we assume that
the damage is more apparent at lower Pc, which is pre-
dictable from Eq. 2. It is also general that brittle ceramics
consist of higher porosity exhibits lower toughness, that is,
lower value of Gc in Eq. 3.
Here we can compare the Al2O3 substrate monolith in
Fig. 3a. Notably, the modulus mismatch generally dimin-
ishes at higher porosity for the Al2O3 substrate shown in
Fig. 2a. Therefore the initial decrease of crack size as the
0.0
0 10 20 30
Porosity(%)
Fig. 3 Plot of the maximum sizes of crack diameter using WC sphere
radius r = 3.18 mm at load P = 150 N as a function of porosity for
layered composites consisting of glass coatings on porous (a) Al2O3
and (b) SiO2. The contact damage on the surface for each condition is
included in figure. The size of scale bar = 250 lm
porosity increases to a maximum of 20% as shown in stage
I of Fig. 3a is explained by diminishing effect of elastic
modulus mismatch. However, the graph in Fig. 3a suggests
some range of porosity in the substrate layer can suppress
the initiation of ring cracks in the coating layer of layered
composite. Moreover, it is noteworthy that the suppression
phenomenon of the coating fracture is found at layered
composite with stiffer substrate material (Note the included
optical micrographs of the contact damage on the surface
of the coating for Al2O3 substrate is less apparent for

123
J Mater Sci (2007) 42:91169120
cracking than for SiO2 substrate). The results in Fig. 3a
indicate that the contact cracking crucially depends on the
porosity, as well as the elastic modulus mismatch.
Therefore, the rigidity of substrate for the layered
composite bonded with the substrate controls the initiation
and propagation of contact cracks in the coating layer.
Strong and stiffer substrates protect coating layers bonded
with very-soft adhesive layers (d = 10 lm) from cracking
caused by flexure stress (a high modulus of substrate
delays initiation of subsurface radial cracking according to
Eq. 2). This is because the controlling material quantity in
Eq. 2 is strength, rF, relating with flexure. Plots in Fig. 3a
illustrate the beneficial role of well-controlled pores in the
ceramic substrate with respect to contact fracture of the
coating layer. For small porosity in the substrate (stage I),
the pores may suppress cracking by damage re-distribu-
tion. On the other hand, radial cracking by flexural stress
due to elastic modulus mismatch, especially apparent in
the layered composite with SiO2 substrate as shown in
Fig. 3b, causes severe damage to the coating layer. Contact
cracks including radial cracking are apparent in the porous
SiO2 support at any range of porosity, because the hard/
soft-layered composite has a high elastic modulus mis-
match, which strongly influences cracking in the coating
layer.
SEM micrographs of section views showing subsurface
damages in the brittle substrate monolith as well as crack
profiles produced in the coating layers are shown in Fig. 4.
Fig. 4 Section views of SEM micrographs of the glass/Al2O3 layered composite from Hertzian indentation at load P = 150 N. The porosity of
the substrate layer is (a) 13.3%, (b) 34.3%, and (c) 41.2%
9119
Although cracking appears in the coating layer of the glass/
Al2O3 layered composite with small porosity, it is notable
that the damage is not critical in the substrate layer of
Fig. 4a and b relative to Fig. 4c. On the other hand, the
result suggests that if yielding of the substrate occurs
during contact with the higher porosity substrate as shown
in Fig. 4c, extensive cracking in the coating layer is pro-
duced. An extensive yield in the substrate of the layered
composite is observed with more extensive cracking in the
coating layer, as indicated in Fig. 4c. The extensive
cracking in the coating layer can be appeared if the
extensive yielding occurs in the substrate even though
the elastic modulus mismatch is diminished. Therefore, the
increase of size of cracks as the amount of the porosity
in the Al2O3 substrate increases to a maximum of 42.6% in
stage II of Fig. 3a is closely due to yield of subsurface in
the substrate. The determining factor is whether yield
occurs in the porous substrate during the contact stress. The
result indicates that the substrate should not yield before
the coating layer fractures.
In designing the coating/substrate-layered composite,
bonding with adhesive requires a higher substrate modulus
and hardness with a matching of the substrate, therefore
higher damage tolerance. The smaller volumes of pores
under the critical porosity range can contribute to delayed
cracking if they are not yielded or deformed before coating
fracture. More detailed and systematic investigations in this
regard should be conducted in future study.
123
9120
Acknowledgements This work was partly supported by Korea
Research Foundation Grant funded by the Korean Government
(MOEHRD) (KRF-2004-003-D00002) and partly by a grant from
Electric Power Industry R&D Project funded by the Ministry of
Commerce, Industry and Energy.
References
1. Lawn BR (1998) J Am Ceram Soc 81:1977
2. Lee KS, Seo DW, Yu JH, Woo SK (2003) J Kor Ceram Soc
40:241
3. Jung HH, Kim DJ, Han JS, Lee DY (2004) J Kor Ceram Soc
41:508
4. Beele W, Marinjiniseen G, van Lieshout A (1999) Surf Coat Tech
120121:61
5. Lee KS, Lee SK, Lawn BR, Kim DK (1998) J Am Ceram Soc
81:2394
6. Lawn BR, Lee KS, Chal H, Pajares A, Kim DK, Wuttiphan S,
Peterson IM, Hu X (2000) Adv Eng Mater 11:745
123
J Mater Sci (2007) 42:91169120
7. Lee KS, Wuttiphan S, Hu XZ, Lee SK, Lawn BR (1998) J Am
Ceram Soc 81:571
8. Lee KS, Lee YW, Blackburn DH, Chai H, Lawn BR (2000)
J Mater Res 15:1653
9. Hertz H (1896) Hertzs miscellaneous papers, Ch. 5, 6. Mac-
millan, London
10. Lawn BR (1993) Fracture of brittle solids, 2 edn. Cambridge
Univ. Press, Cambridge
11. Pajares A, Wei L, Lawn BR, Padture NP, Berndt CC (1996)
Mater Sci Eng A208:158
12. Lee KS, Wuttiphan S, Hu XZ, Lee SK, Lawn BR (1998) J Am
Ceram Soc 81:571
13. Lee KS, Lee SK, Lawn BR, Kim DK (1998) J Am Ceram Soc
81:2394
14. Rhee YW, Kim HW, Deng Y, Lawn BR (2001) J Am Ceram Soc
84:1066
15. Rice RW (1993) J Mater Sci 28:2187
16. Rice RW (1989) Mater Sci Eng A112:215
17. Ha JH, Kim JH, Kim DK (2004) J Jap Ceram Soc 112:1084
J Mater Sci (2007) 42:91219124
DOI 10.1007/s10853-007-2056-1
LETTER
Effect of long-range ordering in a NiMo alloy on its mechanical
properties and corrosion resistance
H. M. Tawancy F. K. Alyousef
Received: 18 March 2007 / Accepted: 25 July 2007 / Published online: 11 August 2007

Springer Science+Business Media, LLC 2007
Ordered alloys and intermetallic compounds are playing an
increasingly important role in advanced corrosion applica-
tions, e.g. [1]. As design schemes increase in sophistication,
the search continues for alloys, which offer higher strength
and can handle a larger variety of complex aggressive
media in comparison with the conventional alloys. How-
ever, the issue of low ductility remains to be an important
one and, therefore, efforts are focused on improving duc-
tility and/or innovations in manufacturing technology.
Among the most important alloys in this class are the those
based upon the NiMo and NiMoCr [2, 3, 4]. A common
feature of these alloys is that they undergo a series of long-
range ordering reactions upon exposure at elevated tem-
peratures [5, 6]. This involves a slight atoms rearrangement
on {420}fcc planes resulting in closely-related superalat-
tices: D1a (Ni4Mo), Pt2Mo-type: {(Ni2Cr, and Ni2(Cr,Mo)},
and DO22 (basis for the Ni3Mo structure) [7].
It was the objective of this study to determine the effect
of long-range ordering in a NiMo alloy of commercial
grade on its mechanical properties and corrosion resistance.
The alloy was processed into sheets about 2 mm in thick-
ness. Its chemical composition is listed in Table 1.
Specimens were annealed at 1,065 C for 20 min and then
water quenched. Inorder to induce long-range ordering,
annealed specimens were exposed up to 1,000 h at 700 C
and then air-cooled. Mechanical properties were deter-
mined from room temperature tensile tests (50.8 mm gage
length). Aqueous corrosion rates were calculated from
weight loss measurements made on 25.4 mm 25.4
H. M. Tawancy (&)
F. K. Alyousef
Center for Engineering Research, Research Institute,
King Fahd University of Petroleum and Minerals,
Dhahran 31261, Saudi Arabia
e-mail: tawancy@kfupm.edu.sa
mm 2 mm specimens after 24 h immersion in 20%
boiling hydrochloric acid. Transmission electron micros-
copy and diffraction as well as x-ray diffraction were used
to characterize the microstructure.
Figure 1 shows the effect of up to 1,000 h of exposure at
700 C on the room temperature tensile properties and
corrosion rate. It is observed that after about 1 h of expo-
sure, the 0.2% yield strength was nearly doubled relative to
the annealed condition; however, about 70% of the initial
tensile ductility in 50.8 mm gage length was still main-
tained. Corresponding to these changes there was no
significant change in corrosion rate. With continued ther-
mal exposure, the yield strength remained nearly
unchanged after up to 1,000 h. In contrast, there was a
significant loss of tensile ductility as well as a substantial
increase in corrosion rate. Such changes in properties were
correlated with long-range ordering to Ni4Mo as demon-
strated below.
Figure 2 shows x-ray diffraction patterns illustrating the
effect of exposure time at 700 C on the average structure.
In the annealed condition, only characteristic face-centered
cubic (fcc) reflections are observed indicative of a disor-
dered solidsolution. After thermal exposure at 700 C,
characteristic superlattice reflections of Ni4Mo were
observed as expected (tetragonal D1a superlattice:
a = 0.5720 nm, c = 0.3564 nm). However, as can be seen
the relative intensities of the superlattice reflections
remained nearly unchanged as the exposure time was
increased from 1 h to 1,000 h. This suggested that ordering
was completed after about 1 h of exposure at 700 C.
An example illustrating the evolution of the micro-
structure of Ni4Mo within the alloy matrix is given in
Fig. 3. Arrays of discrete particles were observed during
the early stages of exposure (Fig. 3a). This was followed
by the formation and growth of a mosaic assembly of
123
9122 J Mater Sci (2007) 42:91219124

Table 1 Chemical composition (weight %) demonstrated earlier (Fig. 1). Corresponding deformation
Ni Mo Cr Fe Mn Si Co Al C
substructures are shown in Fig. 4. As expected, in the
annealed condition, deformation occurred by slip as shown
Balance 26.92 0.64 0.93 0.24 <0.02 <0.1 0.19 0.002 in Fig. 4a. Slip lines could be clearly distinguished in the
deformation substructure with evidence of cross-slip sug-
twin-related variants along {100}fcc planes (Fig. 3b, c). gesting medium stacking fault energy. However, after
Corresponding to these microstructural changes, the 0.2% long-range ordering to Ni4Mo, the predominant deforma-
yield strength was nearly doubled relative to the annealed tion mode became twinning on the {111}fcc planes as
condition, however, about 70% of the initial tensile duc- shown in the example of Fig. 4b. This observation indi-
tility in 50.8 mm gage length was still maintained as cated that was a favorable deformation mode in the ordered

Fig. 1 Effect of exposure time 5

Corrosion Rate (mm/year)

up to 1,000 h at 700 C on the
4
room temperature tensile
properties and corrosion rate in 3
boiling 20% HCl
2

80
Tensile Elonagation
in 50. 8 mm ( % )

60

40

20

955
0. 2% Y iel d Str e ngth (MP a )

820

685

550

415

280
0 1/4 1/2 1 2 4 8 24 100 1000
Exposure Time (Hours)

Fig. 2 Comparative X-ray

diffraction patterns derived
from an annealed sample and
samples exposed for different
times at 700 C

123
J Mater Sci (2007) 42:91219124 9123

Fig. 3 Dark-field TEM images

formed with the 1/5 <420>
superlattice reflection to
illustrate the effect of exposure
time at 700 C on the
microstructure of Ni4Mo. (a)
1 h. (b) 24 h. (c) 1,000 h

state consistent with the crystallographic features of the fcc
(disordered) ? D1a superlattice (Ni4Mo). It can readily be
shown that all {111} <110> slip systems are suppressed in
the ordered state, however, eight of the {111} <112>
twinning systems remain to be energetically favorable [8].
Corresponding to these changes in tensile properties, there
was only a slight increase in corrosion rate, which could be
interpreted in terms of Mo-depleted zones in the vicinity of
Ni4Mo domains, as reported in an earlier study [2].
After extended exposure corresponding to almost com-
plete loss of tensile ductility (1,000 h at 700 C), only
{111}fcc stacking faults were observed in the deformation
substructure as shown in Fig. 4c. Noting that a stacking
fault is a thin layer of a twin, this observation indicated that
fracture occurred prior to the development of deformation
twins. Figure 5 illustrates characteristic morphologies of
tensile fracture surfaces as functions of exposure time at
700 C. As expected, in the annealed condition (Fig. 5a),
the fracture mode was transgranular (dimple-type rupture).
Also, after 24 h of exposure the fracture mode remained to
be transgranular, however, the dimples became rather
shallower indicating less strain prior to fracture. Although
after 1,000 h of exposure, the fracture mode became
interngranular, dimples were observed at separated grain
facets (Fig. 5c) indicating highly localized deformation
alongside grain boundaries. This behavior is usually asso-
ciated with alloy depletion in the vicinity of grain
boundaries as demonstrated in Fig. 6. The respective val-
ues of Mo concentrations were obtained from
concentration profiles derived by in-situ fracture and
sputtering of aged specimens in an Auger spectrometer.
The value of 4.7 atomic % corresponds to a freshly exposed
grain boundary, which is to be compared with the bulk
value of 19.1 atomic %. It is observed that ordering in the
vicinity of grain boundaries occurred by a discontinuous
mechanism resulting in alternating lamellae of Ni4Mo and

Fig. 4 Bright-field TEM

images illustrating the effect of
long-range ordering to Ni4Mo
on the deformation substructure
(a) Annealed (3% strain). (b)
Exposed for 24 h at 700 C
(15% strain) (c) Exposed for
1,000 h at 700 C (2% strain;
fractured specimen)

Fig. 5 Secondary electron

SEM images illustrating the
effect of long-range ordering to
Ni4Mo on the morphology of
tensile fracture surfaces. (a)
Annealed. (b) Exposed for 24 h
at 700 C. (c) Exposed for
1,000 h at 700 C

123
9124
Fig. 6 Bright-field STEM image showing lamellar grain boundary
structure of Ni4Mo after 1,000 h of exposure at 700 C; also shown is
the Mo concentration alternating lamellae as derived from micro-
chemical analysis
a Mo-depleted solidsolution. This could also explain the
substantial increase in corrosion rate as demonstrated in
Fig. 7. In the annealed condition and after matrix ordering,
corrosion was predominantly of the uniform or general
type (Fig. 7a). However, after the onset of the discontinu-
ous grain boundary reaction, most of the metal wastage
appeared to be due to intergranular attack as shown in
Fig. 7b.
In conclusion, the present study had demonstrated that
long-range atomic order may not necessarily degrade the
mechanical strength and corrosion resistance of certain
alloy systems. Conversely, it may even have some bene-
ficial effects on properties particularly mechanical strength.
However, property degradation may result from the
123
J Mater Sci (2007) 42:91219124
Fig. 7 Light optical micrographs along cross-sections of corrosion
tested specimens. (a) annealed. (b) exposed for 1,000 h at 700 C
specific morphology produced by localized ordering in the
vicinity of grain boundaries as shown in this study.
Therefore, it may prove viable to use single crystals of
ordered alloys in certain applications and/or identify means
for modifying the structure of grain boundaries to suppress
the undesirable discontinuous ordering reaction.
References
1. Nguyen-Manh D, Vitek V, Horsfield AP (2007) Prog Mater Sci
52(2/3):255
2. Tawancy HM (2006) J Mater Sci 41(24):8359
3. Svistunova TV (2005) Met Sci Heat Treat 47(7/8):383
4. Friend WZ (1980) Corrosion of Nickel and Nickel-base alloys.
John Wiley and Sons, New York, New York, p 248, 292
5. Tawancy HM (1993) Structure and properties of high-temperature
alloys: applications of analytical electron microscopy. KFUPM
Press, Dhahran, p 165, 175
6. Tawancy HM, Asphahani AI (1984) In: Koch CC, Liu CT, Stoloff
NS (eds) High temperature ordered intermetallic alloys, vol 39.
Materials Research Society, Pittsburgh, Pennsylvania, p 485
7. Tawancy HM (1995) J Mater Sci 30:522
8. Tawancy HM (1995) Scripta Met et Materialia 32(12):2055
J Mater Sci (2007) 42:91259129
DOI 10.1007/s10853-007-2010-2
LETTER
Three-dimensional (3D) microstructure-based modeling of crack
growth in particle reinforced composites
A. Ayyar N. Chawla
Received: 28 June 2007 / Accepted: 20 July 2007 / Published online: 4 August 2007
Springer Science+Business Media, LLC 2007
The mechanical properties of particle reinforced compos-
ites are largely dependent on the reinforcement particle
distribution and volume fraction [1, 2]. Crack growth in
particle reinforced metal matrix composites (MMCs), such
as SiC particle reinforced aluminum (Al/SiCp), is very
much dependent on the reinforcement microstructure [1, 3]
as well as the matrix characteristics [4, 5]. A robust
numerical model to simulate crack growth must incorpo-
rate the true particle geometry, orientation, size
distribution, and spatial distribution [6]. Ayyar and Chawla
[6] modeled the crack growth behavior in an Al/SiCp
composite, using 2D microstructures obtained from optical
microscopy. The results of these models showed that the
incorporation of the actual microstructural attributes
significantly affected the simulated crack growth response.
Of course, 2D models must be conducted under sim-
plified stress states, such as plane stress or plane strain
conditions. Thus, 2D simulations do not provide as com-
plete a picture of the crack growth processes as 3D
simulations. In reality, the stress state in 3D is quite
complex. This is particularly true of composite materials,
where the geometry of the SiC particle in the third
dimension and the matrix around the particle play an
important role. Because the geometry of the SiC particle is
very complex, a 2D representation of the microstructure
cannot adequately capture the behavior of the composite.
Chawla et al. [79] have shown that predictions from 3D
A. Ayyar
Department of Mechanical and Aerospace Engineering, Arizona
State University, Tempe, AZ 85287-8706, USA
N. Chawla (&)
School of Materials and Department of Mechanical Engineering,
Arizona State University, Tempe, AZ 85287-8706, USA
e-mail: nchawla@asu.edu
microstructure-based models simulating the tensile behav-
ior of heterogeneous materials correlated well with
experimental results. Most of the 3D microstructure-based
modeling efforts have focused on predicting the elastic
modulus and simulating the elastic-plastic behavior of the
composite in tension. The onset and evolution of damage
by particle fracture has been studied by a few researchers
[1014], although the reinforcement particles are modeled
as simple spheres. Crack growth in 3D has not been
modeled in particle reinforced metal matrix composites. In
this paper, crack growth in a SiC particle reinforced Al
composite was simulated, using the actual 3D microstruc-
ture of the composite.
A serial sectioning process was used to capture the
complex geometry of the SiC particles from a series of
micrographs of a SiC particle reinforced 2080 Al matrix
composite [7, 8]. The SiC particles had a volume fraction
of approximately 20%, average particle size of about 8 lm,
and an average aspect ratio of 2. The serial sectioning
process allows one to capture the realistic microstructure of
the particles, including the geometry, orientation, and dis-
tribution of the particles. A typical flow chart of the serial
sectioning and 3D reconstruction process is shown in
Fig. 1. Details of this process can be found in reference [7].
The sample was cut and mounted for polishing and a
representative region of the microstructure was selected.
The term representative is somewhat subjective as it is
dependent on the size of the SiC particles, spatial distri-
bution, etc. As the number of particles increases, however,
the computational demands also increase. For the Al/SiCp
composites modeled in this paper, a volume of about 32
particles was included in the model (which is about half of
the total particles reconstructed after serial sectioning),
because of reasonable limits on computational efficiency.
Fiducial marks were made by Vickers indentation. These
123
9126
marks were used to measure the material thickness loss
during the polishing process. The thickness removed is a
function of the microstructural feature size. In this case,
since the particles were approximately 68 lm in size, a
removal rate of about 1 lm/section was selected. Using the
cyclic process of polishing and imaging, it was possible to
generate a series of 2D micrographs. These 2D micro-
graphs were segmented to differentiate between the
particles and the matrix. The 3D morphology of the par-
ticles was reconstructed from the series of 2D segmented
images using commercially available software (SURF-
driver, Kailua, Hawaii).
The 3D reconstructed microstructure was then meshed
using HyperMesh1 (Altair, Los Angeles, CA), Fig. 2. A
2D surface mesh was first created on the SiC particle and
the Al matrix surfaces. This 2D mesh consisted of 3-node
triangular elements. The 2D mesh was then used to create a
3D mesh consisting of 4-node tetrahedral elements (C3D4).
The tetrahedral elements conformed well to the irregular
shape of the SiC particles. A variable mesh density
approach was used to reduce the overall size of the model
without sacrificing the accuracy of the model. The mesh in
the SiC particles and the Al matrix near the particles were
finer compared to the mesh on the surfaces of the Al
matrix, Fig. 2c. An embedded cell, with average com- matrix, and the average composite were modeled as
posite properties, was used to minimize the effects of free
edges. There were approximately 45,000 nodes and
240,400 tetrahedral elements (C3D4) in the model. The
discretized model was then imported into a commercial
finite element analysis software (ABAQUS, Pawtucket,
RI). Boundary conditions were applied to the model as
shown in Fig. 2d. The element elimination method was
used to simulate crack growth in the particle reinforced
composite. In this method an element is eliminated from
Fig. 1 A typical flow chart of
the serial sectioning and 3D
reconstruction process [7]
Material
preparation
3D Microstructure
of the composite
123
J Mater Sci (2007) 42:91259129
the calculations when a certain parameter in the element
exceeds a predefined value. The mesh was refined, since
the crack profile is highly dependent on the mesh size and a
fine mesh is required to obtain accurate results. Although
Al is highly ductile, for the sake of simplicity, and to
illustrate the feasibility of crack growth modeling using
actual 3D microstructures, the Rankine criterion was used
for element elimination (failure occurs when the maximum
principal stress in the element is exceeded). The element
elimination process, using the Rankine criterion, was
simulated in the Dynamic-Explicit module in ABAQUS.
This module requires that a mass be assigned to all the
materials. Therefore, it is extremely important that the load
be applied very slowly to the system to avoid any inertia
effects. A velocity of 0.01 unit/s was applied to the nodes
on the upper surface, as shown in Fig. 2d, until the crack
had propagated through the matrix material. The value of
the velocity was obtained from an iterative procedure to
quantify any inertia effects. The elements in the Al matrix
were eliminated when the stress reached the yield stress for
the Al alloy matrix (approximately 320 MPa). Fracture of
the SiC particles was not incorporated in this model,
although the authors have used it in modeling 2D micro-
structures [15]. All three components, SiC particles, Al
purely elastic. The elastic properties of SiC and Al alloy
are shown in Table 1. The properties for the Average
composite were computed using a rule-of-mixtures
approach. The density was 2.81 g/cm3, the Poissons ratio
was 0.31 and the Youngs modulus was approximately
120 GPa. Particles were assumed to be bonded perfectly to
the matrix. The simulation time for this model, on a DELL
Precision workstation with single processor (3.0 GHz
speed) and 2 GB of RAM, was approximately 6 hours.
Indentation of
Polishing
fiducial marks
Imaging
3D Visualization Regeneration of Image
of SiC particles 3D particles segmentation
J Mater Sci (2007) 42:91259129 9127

Fig. 2 (a) Embedded cell (a) Embedded Cell

approach showing the 32 SiC 106 m
C 35 m G
particle composite model. The
notch provides a stress F
B
concentration point for crack
initiation, (b) variable mesh
density approach, and (c)

Plane 1 (ABCD)

Plane 2 (EFGH)
boundary conditions applied to
the model

127 m
55 m
X
D H

20 m E
A

Al Matrix Y Z
Average composite

G
(b) (c) Vx=0.01 unit/s
SiC

C
F

D E

Y Z
Uy,z=0
Al Matrix
A Ux=0

Table 1 Material properties (Elastic) used in the model increases, as has been observed experimentally [1]. When
Material Density Youngs Poissons particle fracture is modeled, the crack goes through the SiC
(g/cc) modulus (GPa) ratio particles, and the crack path is quite linear [15]. Crack
growth in a homogeneous model with average composite
Aluminum alloy 2080-T6 2.75 74 0.33
properties, i.e., where the microstructure of the SiC parti-
SiC particles 3.2 410 0.19
cles was not explicitly considered was also modeled. The
distance between plane 1 and 2, in the homogeneous
Figure 3 shows the simulated 3D crack profile for the model, is the same as that in the microstructure-based
model of Al/SiCp composite. The model clearly shows the model. As expected, here the crack profile was planar and
complex crack growth phenomenon in particle reinforced not very different between the outer and inner surfaces.
composites. The crack profile on the outer surface was very When one incorporates the explicit microstructure, it is
different from the crack profile on the inner surface quite clear that the crack profile through the SiC particles is
because of the nature of the local microstructure. The crack captured. The tortuous nature of crack growth, as well as
initiates in the Al matrix along plane 1 (ABCD) and ter- the deflection of the crack due to particles can be seen.
minates in the matrix at plane 2 (EFGH). On the inner Crack deflection, of course, reduces the crack-tip driving
surface it was observed that the crack did not reach plane 2 force of the material. The applied load is plotted versus the
(EFGH) as it was arrested by a SiC particle. Note that the nodal displacements for both models (microstructure-based
crack branches around one of the SiC particles, Fig. 3b, model and homogeneous model) in Fig. 4. The crack ini-
and is forced to grow around it, Fig. 3c, as the load tiates at plane 1 (ABCD) and ends at plane 2 (EFGH).

123
9128 J Mater Sci (2007) 42:91259129

Fig. 3 3D microstructure-based
(a) (a)
simulation of crack growth
showing crack arrest, crack
deflection, and tortuous 3D
crack surface
Plane 1 Plane 1

d = 0.053 d = 0.053

(b) (b)

Plane 1 Plane 1

d = 0.066 d = 0.066

(c) (c)

Plane 1 Plane 1

d = 0.069 d = 0.069

Crack growth Crack growth

direction direction

Fig. 4 Load versus Crack initiates at Plane 1 and terminates at Plane 2 Displacement of this node
displacement plots. In the 100
microstructure-based model Plane 1
90
more energy is required to
80
propagate the crack through the
Applied load (GN)

70
same distance (from Plane 1 to Plane 1 Plane 2
60
Plane 2)
50 Plane 2
40
30
Plane 2
Plane 1

20
10
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 Homogeneous model
Displacement of node (units)
Microstructure-based
model

Clearly, more energy was required for the crack to prop-
agate in the microstructure-based model. This is a result of
crack deflection and crack arrest by the particles in the
microstructure-based model.
In summary, a 3D microstructure-based model was
developed to study crack growth in Al/SiCp composites.
The complex geometry of the SiC particles, in all three
dimensions, was captured using a serial sectioning
approach. Three-dimensional crack growth is a complex
phenomenon, which involved crack deflection and crack
arrest. A complex and tortuous 3D crack profile was sim-
ulated (assuming no particle fracture), which cannot be
adequately represented using conventional 2D representa-
tions. This crack deflection due to reinforcement particles
makes the crack path tortuous and improves the crack
growth resistance of the composite.

123
J Mater Sci (2007) 42:91259129 9129

Acknowledgements The authors acknowledge useful discussions 5. Sugimura Y, Suresh S (1992) Metall Trans A 23:2231
with Prof. K. K. Chawla, University of Alabama at Birmingham, and 6. Ayyar A, Chawla N (2006) Compos Sci Technol 66:1980
Prof. Y.-L. She, University of New Mexico. 7. Chawla N, Ganesh VV, Wunsch B (2004) Scr Mater 51:161
8. Chawla N, Sidhu RS, Ganesh VV (2006) Acta Mater 54:1541
9. Chawla N, Sidhu RS (2007) J Mater Sci-Mater Electron 18:175
10. Han W, Eckschlager A, Bohm HJ (2001) Compos Sci Technol
References
61:1581
11. Bohm HJ, Eckschlager A, Han W (2002) Comput Mater Sci
1. Chawla N, Chawla KK (2006) Metal matrix composites. 25:42
Springer, New York 12. Eckschlager A, Han W, Bohm HJ (2002) Comput Mater sci
2. Divecha AP, Fishman SG, Karmakar SD (1981) J Metals 9:12 25:8591
3. Shang JK, Yu WK, Ritchie RO (1988) Mater Sci Eng A102:181 13. Segurado J, Gonzalez C, Llorca J (2003) Acta Mater 51:2355
4. Bonnen JJ, You CP, Allison JE, Jones JW (1990) In: Proceedings 14. Segurado J, Llorca J (2006) Mech Mater 38:873
of the international conference on fatigue. Pergamon Press, New 15. Ayyar A, Chawla N (2007) Acta Mater (in press)
York, p 887

123