1607500450

BIOLOGICAL AND BIOMEDICAL INFRARED SPECTROSCOPY
Advances in Biomedical Spectroscopy

Spectroscopic methods play an increasingly important role in studying the molecular details of complex biological systems in health and disease. However, no single spectroscopic method can provide all the desired information on aspects of molecular structure and function in a biological system. Choice of technique will depend on circumstance; some techniques can be carried out both in vivo and in vitro, others not, some have timescales of seconds and others of picoseconds, whilst some require use of a perturbing probe molecule while others do not. Each volume in this series will provide a state of the art account of an individual spectroscopic technique in detail. Theoretical and practical aspects of each technique, as applied to the characterisation of biological and biomedical systems, will be comprehensively covered so as to highlight advantages, disadvantages, practical limitations and future potential. The volumes will be intended for use by research workers in both academic and in applied research, and by graduate students working on biological or biomedical problems. Series Editor: Dr. Parvez I. Haris De Montfort University, Leicester, United Kingdom
Volume 2
Recently published in this series Vol. 1. B.A. Wallace and R.W. Janes (Eds.), Modern Techniques for Circular Dichroism and Synchrotron Radiation Circular Dichroism Spectroscopy
ISSN 1875-0656
Biological and Biomedical Infrared Spectroscopy
Edited by
Andreas Barth
Stockholm University, Stockholm, Sweden
and
Parvez I. Haris
De Montfort University, Leicester, UK
Amsterdam Berlin Tokyo Washington, DC
2009 The authors and IOS Press. All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without prior written permission from the publisher. ISBN 978-1-60750-045-2 Library of Congress Control Number: 2009932637 Publisher IOS Press BV Nieuwe Hemweg 6B 1013 BG Amsterdam Netherlands fax: +31 20 687 0019 e-mail: order@iospress.nl Distributor in the UK and Ireland Gazelle Books Services Ltd. White Cross Mills Hightown Lancaster LA1 4XS United Kingdom fax: +44 1524 63232 e-mail: sales@gazellebooks.co.uk Distributor in the USA and Canada IOS Press, Inc. 4502 Rachael Manor Drive Fairfax, VA 22032 USA fax: +1 703 323 3668 e-mail: iosbooks@iospress.com
LEGAL NOTICE The publisher is not responsible for the use which might be made of the following information. PRINTED IN THE NETHERLANDS
Biological and Biomedical Infrared Spectroscopy A. Barth and P.I. Haris (Eds.) IOS Press, 2009 2009 The authors and IOS Press. All rights reserved.
Series Preface
In the post-genomic era there is a great need to understand the structure and dynamics of macromolecules, not just single molecules but also their multiple interactions as part of a systems biology approach. It is therefore not surprising that in recent years several Nobel Prizes have been awarded to scientists who have developed well established analytical techniques to the study of biological and medical systems, these includes mass spectrometry, NMR spectroscopy, magnetic resonance imaging. There is no doubt that the development of new analytical techniques and the effective utilisation of existing methods is vital for obtaining a better picture of the molecular details of complex biological systems in both health and disease. Such progress is important for disease diagnosis and drug discovery processes. However, the complexity of biological systems is such that no single experimental method can provide information on all aspects of molecular structure and function. There are a large number of spectroscopic methods that can be used in the analysis of biological systems. Some can be used to carry out analysis in both in vivo and in vitro settings whereas others are restricted, at least currently, to one particular environment. The timescales of many of these techniques can be very different. Some require the use of potentially perturbing probe molecules, whereas others do not. Clearly, no single technique is perfect and each has its respective advantages and disadvantages. Consequently, a serious scientist would not be fully satisfied with the analysis of particular system based on results from a single technique. Ideally, one should use a battery of techniques before drawing a final conclusion. Considering the wide array of techniques available for analysis of biological systems, producing a single book on one particular spectroscopic technique would not be sufficient to meet the needs of scientists engaged in understanding biological molecules and their interactions. Therefore, I decided to produce a series of books on emerging and established spectroscopic methods to serve the needs of academics, industrial scientists as well as graduate students who are currently using or seeking to use a particular spectroscopic method in their research work. The books are intended to provide advances in theoretical and practical aspects of each technique, as applied to the characterisation of biological and biomedical systems, highlighting advantages, disadvantages and potential pitfalls. The first volume of the series provides a comprehensive discussion of the state-ofthe-art methods in Circular Dichroism spectroscopic analysis of biological systems. The volume was edited by Bonnie Ann Wallace (Birkbeck College, University of London) and Robert William Janes (Queen Mary & Westfield College, University of London). Bonnie Wallace has been awarded the 2010 AstraZeneca Award by the Biochemical Society, UK and the 2010 Interdisciplinary Award by the Royal Society of Chemistry, UK for her work on the development of Synchrotron Radiation-based Circular Dichroism Spectroscopy for biological studies. The current volume is devoted to the application of infrared spectroscopy in biological and biomedical studies. It is edited by Andreas Barth (Stockholm University) and myself and brings together contributions from leading experts in infrared spectroscopy.
vi
Finally, I would like to thank all the editors for their hard work in bringing together leading experts in their field to make contributions that ultimately result in the production of each volume in the series. Parvez I. Haris Leicester, United Kingdom
vii
Preface
This book aims to provide an insight into some of the key areas where infrared spectroscopy has been successfully applied to understand important biological and biomedical processes. It highlights the latest advances and the directions for the future. The book provides a historical framework for the development of biological infrared spectroscopy. Key methodologies that are in current use and latest advances, in both theoretical and practical aspects, are discussed. Examples of applications, ranging from characterisation of individual macromolecules (DNA, RNA, lipids, proteins) to complex systems such as human tissues, cells and whole organisms are covered. The main focus is in the mid-infrared region as the vast majority of studies are conducted in this region. However, there is increasing use of the near-infrared region for biomedical application and hence a chapter is devoted to this part of the infrared spectrum. Biological spectroscopy is a highly interdisciplinary field of research requiring involvement of life scientists and analytical chemists. Advances in instrumentation technology and methods for analysis and interpretation of the spectroscopic data require input from multiple disciplines including Chemistry, Physics, Mathematics, Computer Science and Engineering. It is this co-operation between scientists from diverse disciplines that ultimately results in the utilisation of a physical technique for understanding the molecular details of biological processes and systems. Such co-operation is vital if spectroscopists are to play a significant role in the analysis of the vast number of genes and proteins that are being identified by the various genome sequencing projects. Currently, it is not impossible for a gene sequencing laboratory to produce as much data in less than a week as was produced by Shakespeare in his entire life-time. However, an understanding of the molecular details of the genes and proteins identified, and their diverse interactions, require application of biophysical techniques such as infrared spectroscopy. Continued technological development in spectroscopic methods is vital to keep pace with the breathtaking advances in the field of molecular biology. Nearly 400 years ago Shakespeare described the seven ages of life in the following manner: All the worlds a stage, And all the men and women merely players: They have their exits and their entrances; And one man in his time plays many parts, His acts being seven ages. Using this as an analogy, Laitinen in 1973 wrote an editorial in Analytical Chemistry describing the seven ages of an analytical method (H.A. Laitinen, Anal. Chem. 45 (1973) 2305). He used infrared spectroscopy as an example to illustrate how it has reached its seventh age. His description of this seventh age is as follows: Seventh, a period of senescence occurs as other methods of greater speed, economy, convenience, sensitivity, selectivity, etc., surpass the method under consideration. It is surprising that Laitinen chose infrared spectroscopy as his example, since at that time the first commercial Fourier transform infrared spectrometers were being delivered to laboratories around the world. As such it was a very exciting time for infra-
viii
red spectroscopy. Indeed, a year earlier, in 1972, Peter Griffiths published a letter in the same journal entitled Trading rules in infrared Fourier transform spectroscopy (P.R. Griffiths, Anal. Chem., 44 (1972), 1909). As an editor of the journal, Laitinen must have been aware of the revolution taking place in infrared spectroscopy. The widespread availability of FT instruments and the use of computers for recording and analysis of infrared spectra, heralded a new era in infrared spectroscopy. Now it was possible to analyse biological molecules, in aqueous media, at fast speeds and at high resolutions that was virtually impossible with dispersive instruments. Far from reaching its seventh age infrared spectroscopy is a vibrant methodology playing a central role in some of the latest discoveries in biology and medicine, including some recent Nobel Prize winning work. For example, Stanley Prusiner was awarded the Nobel Prize for Physiology or Medicine in 1997 and infrared spectroscopy played an important role in his work. In a section of his Nobel lecture (S.B. Prusiner, Proc. Natl. Acad. Sci. USA, 95 (1998), 1336313383) he states the following: For more than 25 years, it had been widely accepted that the amino acid sequence specifies one biologically active conformation of a protein. Yet in scrapie we were faced with the possibility that one primary structure for PrP might adopt at least two different conformations to explain the existence of both PrPC and PrPSc. When the secondary structures of the PrP isoforms were compared by optical spectroscopy, they were found to be markedly different. Fourier-transform infrared (FTIR) and circular dichroism (CD) studies showed that PrPC contains about 40% -helix and little sheet, whereas PrPSc is composed of about 30% -helix and 45% -sheet). Nevertheless, these two proteins have the same amino acid sequence! It is noteworthy that the abnormal form of the prion protein (PrPSc) misfolds and forms aggregates that are virtually impossible for characterisation using X-ray crystallography, NMR and CD spectroscopy. In order to overcome this problem, Prusiner and co-workers used infrared spectroscopy to obtain direct evidence for an increase in betasheet structure in the PrPSc aggregates. In recent years infrared spectroscopy is going through a renaissance catalysed by some exciting developments in technology. This includes the use of the bright synchrotron radiation for recording infrared spectra. Latest breakthroughs also include the development of two-dimensional infrared spectroscopy and the ability to record infrared spectra at ultrafast speeds. There are also some major advances in theoretical analysis that is enabling a better interpretation of the infrared spectra of biological molecules. Considering these advances, we felt it would be timely to produce a book that brings together some of the key developments in the field. The book is intended for both experts and those who are new to the field of biological infrared spectroscopy. It would be particularly beneficial for graduate students and research scientists in both industry and academia. Finally, we would like to thank all the authors who have contributed in this volume. Without their co-operation it would not have been possible to accomplish this task. Andreas Barth (Stockholm University, Stockholm, Sweden) Parvez I. Haris (De Montfort University, Leicester, UK)
ix
List of Contributors
Tsutomu ARAKAWA Andreas BARTH Alliance Protein Laboratory, Inc., 3957 Corte Cancion, Thousand Oaks, CA 91360, USA Department of Biochemistry and Biophysics, The Arrhenius Laboratories for Natural Sciences, Stockholm University, S-10691, Stockholm, Sweden Institute of Organic Chemistry and Biochemistry, Academy of Sciences, Czech Republic SOLEIL Synchrotron, Saint-Aubin BP 48, 91192 Gif-sur-Yvette Cedex, France Department of Chemistry and Center for Multidimensional Spectroscopy, Korea University, Seoul 136-701, Korea Department of Chemistry and Center for Multidimensional Spectroscopy, Korea University, Seoul 136-701, Korea SOLEIL Synchrotron, Saint-Aubin BP 48, 91192 Gif-sur-Yvette Cedex, France Robert Koch-Institute, Nordufer 20, D-13353 Berlin Department of Chemistry, Stanford University, Stanford, CA 94305-5080, USA Department of Chemistry, Stanford University, Stanford, CA 94305-5080, USA Laboratory for the Structure and Function of Biological Membranes, Center for Structural Biology and Bioinformatics, Universit Libre de Bruxelles, CP 206/2, Boulevard du Triomphe, B-1050 Brussels, Belgium Faculty of Health & Life Sciences, De Montfort University, Leicester, UK Department of Computer Science, University of Applied Sciences Ulm, Prittwitzstrae 10, 89075 Ulm, Germany Department of Chemistry, Stanford University, Stanford, CA 94305-5080, USA Department of Chemistry-Biology, University of Qubec at Trois-Rivires, C.P. 500, Trois-Rivires (Qubec) Canada
Petr BOU
Sirinart CHIO-SRICHAN Jun-Ho CHO
Minhaeng CHO
Paul DUMAS Heinz FABIAN Michael D. FAYER Ilya J. FINKELSTEIN Erik GOORMAGHTIGH
Parvez I. HARIS Joachim A. HERING
Haruto ISHIKAWA David JOLY
Biomacromolecular Drug Delivery, Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark Timothy A. KEIDERLING Department of Chemistry, University of Illinois at Chicago, USA Seongheun KIM Department of Chemistry, Stanford University, Stanford, CA 94305-5080, USA Jan KUBELKA Department of Chemistry, University of Wyoming, USA Peter LASCH Robert Koch-Institute, Nordufer 20, D-13353 Berlin, Germany Tiansheng LI PacificBio Inc., 1152 Tourmalin Drive, Newbury Park, CA 91320, USA Andrew MACNAB Faculty of Medicine, Departments of Pediatrics and Urologic Sciences, Director Near-Infrared Study Group, University of British Columbia, Bladder Care Centre, Canada Lisa M. MILLER National Synchrotron Light Source, Brookhaven National Laboratory, Upton, NY 11973 USA Dieter NAUMANN Robert Koch-Institute, Nordufer 20, D-13353 Berlin, Germany Christophe N. NSOUKPO-KOSSI Department of Chemistry-Biology, University of Qubec at Trois-Rivires, C.P. 500, Trois-Rivires (Qubec) Canada G9A 5H7 Heidar-Ali TAJMIR-RIAHI Department of Chemistry-Biology, University of Qubec at Trois-Rivires, C.P. 500, Trois-Rivires (Qubec) Canada G9A 5H7 Mark J. TOBIN Australian Synchrotron, 800 Blackburn Road, Clayton, Victoria, 3168, Australia Marco VAN DE WEERT Biomacromolecular Drug Delivery, Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark Willem F. WOLKERS Institute of Multiphase Processes, Leibniz Universitt Hannover, Hannover, Germany
Lene JORGENSEN
xi
Contents
Series Preface Parvez I. Haris Preface Andreas Barth and Parvez I. Haris List of Contributors Infrared Spectroscopy Past and Present Andreas Barth and Parvez Haris The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy Andreas Barth Ultrafast 2D-IR Vibration Echo Spectroscopy of Proteins Haruto Ishikawa, Seongheun Kim, Ilya J. Finkelstein and Michael D. Fayer FTIR Data Processing and Analysis Tools Erik Goormaghtigh FTIR Spectroscopy for Analysis of Protein Secondary Structure Joachim A. Hering and Parvez I. Haris Infrared Spectroscopy of Protein Pharmaceuticals Marco van de Weert and Lene Jorgensen Quantum Mechanical Calculations of Peptide Vibrational Force Fields and Spectral Intensities Jan Kubelka, Petr Bou and Timothy A. Keiderling Computational Linear and Nonlinear IR Spectroscopy of Amide I Vibrations in Proteins Jun-Ho Choi and Minhaeng Cho Application of Isotope-Edited FTIR Spectroscopy to the Study of Protein-Protein Interactions Tiansheng Li and Tsutomu Arakawa Biomedical FTIR Spectroscopy of Lipids Willem F. Wolkers Structural Analysis of Protein-DNA and Protein-RNA Interactions by FTIR Spectroscopy H.A. Tajmir-Riahi, C.N. Nsoukpo-Kossi and D. Joly FTIR Spectroscopy of Cells, Tissues and Body Fluids Dieter Naumann, Heinz Fabian and Peter Lasch v vii ix 1
53 79 104 129 168
178
224
261 272
288 312
xii
Biomedical Applications of Near Infrared Spectroscopy Andrew Macnab The Use of Synchrotron Radiation for Biomedical Applications of Infrared Microscopy Lisa M. Miller, Mark J. Tobin, Sirinart Chio-Srichan and Paul Dumas Author Index
355
403
429
Biological and Biomedical Infrared Spectroscopy A. Barth and P.I. Haris (Eds.) IOS Press, 2009 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-045-2-1
Infrared Spectroscopy Past and Present

a
Andreas BARTH a,1 and Parvez HARIS b,2 Department of Biochemistry and Biophysics, Stockholm University, Stockholm, Sweden b Faculty of Health and Life Sciences, De Montfort University, Leicester, United Kingdom
Abstract. History of infrared spectroscopy as well as current technology and applications are reviewed. Keywords. FTIR, infrared, spectroscopy, history, Herschel, Melloni, Langley
1. Early Days 1.1. The Discovery It is sometimes of great use in natural philosophy, to doubt of things that are commonly taken for granted; especially as the means of resolving any doubt, when once it is entertained, are often within our reach [1]. This timeless expression of critical thinking by William Herschel (17381822) introduces his four publications in 1800 dealing with the rays that occasion heat [2]. For the second article [3], he is generally accredited with the discovery of infrared radiation, although some ascribe this to Carl Wilhelm Scheele (17421786) somewhat before 1777 or Marc-Auguste Pictet (17521825) in 1790 [4]. By 1800, Herschel was a respected astronomer, famous for his discovery of the planet Uranus in 1781. He was born 1738 in Hanover/Germany [5,6] as one of ten children of a musician in the Hanoverian Guard Band. 1757 he emigrated to England, where he later changed his name from Friedrich Wilhelm to William, which became Sir William in 1816 when he was knighted. Until 1782 he earned his living as a professional musician but became increasingly interested in astronomy, which was initially a hobby that he pursued together with his sister Caroline Lucretia Herschel (17501848). Not satisfied with the available and affordable telescopes [7], he constructed his own and became a professional telescope maker [6,8]. In 1782, the English king George III appointed him to his court astronomer, following Herschels discovery of Uranus and his suggestion to name the newly discovered planet after the king. This position was
Corresponding Author: Andreas Barth, Department of Biochemistry and Biophysics, The Arrhenius Laboratories for Natural Sciences, Stockholm University, S-10691 Stockholm, Sweden; E-mail: barth@dbb.su.se. 2 Corresponding Author: Parvez I. Haris, Faculty of Health and Life Sciences, De Montfort University, Leicester, United Kingdom, E-mail: pharis@dmu.ac.uk.
1
A. Barth and P. Haris / Infrared Spectroscopy Past and Present
associated with an income [5,6], which enabled him to pursue his astronomical studies without the distraction of earning his living with music. William Herschel was a dedicated researcher who is known to have worked once for three days and nights in a row without break [6]. In 1787 also Caroline received a salary from the king as Williams assistant [7], which made her the first woman in England on a paid scientific position [5]. She also pursued independent studies and received several distinctions for her scientific work in the later stage of her long life [5,7]. In Herschels observations of the sun, the heat generated by reflecting telescopes was a problem. In other words, infrared radiation made itself aware as a nuisance. This caused its entry into science, since Herschels aim to reduce the heat [6,8,9] led to the discovery of invisible radiation beyond the red light. In his systematic study of the heat effect, he dispersed the solar spectrum with a prism and measured the heat with thermometers. The first article [1] explored the visible spectral range, as others did before him [10], solved the heat problem by introducing blue and green coloured glasses into the telescope and speculated on the possibility that the maximum of heat radiation is found beyond the red light. The second article, dated March 17, 1800, reported the detection of infrared radiation with the apparatus shown in Fig. 1: the four last experiments prove, that the maximum of the heating power is vested among the invisible rays [3]. After measuring the infrared spectrum at three different wavelengths (later he added a fourth data point), he then probed the ultraviolet region where no radiation had been detected before: so fine a day, with regard to clearness of sky and perfect calmness, was not to be expected often, at this time of the year; I therefore hastened to make a trial of the other extreme of the prismatic spectrum. However, he found no effect. Ultraviolett radiation was eventually discovered one year later by Johann Wilhelm Ritter (17761810) [10]. Herschel continued in his third article to prove that heat radiation obeys the optical laws of reflection and refraction and introduced a candle and a chimney fire as new sources of radiation [2]. In the fourth article [11], he measured spectral distributions of light and heat, observed that near-infrared radiation is less scattered than visible light, designed the first double beam instruments and used them to study the near-infrared absorption of several substances, including water and some alcoholic beverages. The double beam apparatus for the candle experiments is shown in Fig. 2. Nowadays it is well appreciated that light sensation and heating effect are two aspects of the same kind of radiation. Light caused the temperature rise observed by Herschel in the visible spectral range. In the invisible spectral region beyond the red light, heating was due to infrared radiation, which differs from visible light only by its longer wavelength but not by its nature. It is invisible because our eyes are not sensitive to it, but it can be detected by its heating effect. All this was not obvious around 1800. Nevertheless, Herschel considered infrared radiation first as a spectral extension of visible light and used for it the expression invisible light [1] that differs only in momentum [1] from visible light: radiant heat will at least partly, if not chiefly, consist, if I may be permitted the expression, of invisible light; that is to say, of rays coming form the sun, that have such a momentum as to be unfit for vision [1]. The reason for being invisible was correctly attributed to the properties of the eye: it is highly probable, that the organs of sight are only adapted to receive impressions from [light] particles of a certain momentum [1]. This argument is repeated in the second article where the question whether light be essentially different from radiant heat is explicitly posed [3].
Figure 1. The experiment that discovered infrared radiation in 1800. A prism dispersed sunlight, the spectrum fell on a table and a moveable stand with mounted thermometers. Thermometers 1 and 2 were exposed to the radiation, whereas thermometer 3 served as a control [3].
However, the experiments described in his fourth article [11] made him change his mind [10,12] and he regarded light and heat radiation as different phenomena thereafter. One argument for this arose from his measurement of the spectral distributions of brightness and heating effect, shown in Fig. 3, which seemed to indicate maxima at different spectral positions. Discussing the spectra from larger to smaller wavelengths he concluded that those who would have the rays of heat also to do the office of light must be obliged to maintain the following arbitrary and revolting positions; namely, that a set of rays conveying heat, should all at once, in a certain part of the spectrum, begin to give a small degree of light; that this newly acquired power of illumination
Figure 2. The first double beam instrument. A candle served as light source, two thermometers as detectors. The sample was placed between one of the thermometers and the candle while the second thermometer served as control. The apparatus on the table (labelled Fig. 2) was used to cover the two holes between candle and thermometers simultaneously [11].
Figure 3. The spectra of heat radiation (shaded, labelled with S) and of light (labelled with R) as measured by Herschel [11]. The spectra are dispersed horizontally with wavelengths decreasing from left to right.
should increase, while the power of heating is on the decline; that when the illuminating principle is come to a maximum, it should, in its turn, also decline very rapidly, and vanish at the same time with the power of heating. How can effects that are so opposite be ascribed to the same cause? first of all, heat without light; next to this, decreasing heat, but increasing light; then again, decreasing heat and decreasing light [11]. We know now why the maximum of the heat radiation lay in the infrared spectral region in Herschels experiment. His thermometer bulb sampled a larger spectral range in the infrared than in the visible spectral range because the dispersion of glass prisms decreases with increasing wavelength. If this is corrected for, the maximum of the heat effect lies near 600 nm in the orange region of the spectrum [9,13]. The second argument for seeing light and heat radiation as independent phenomena stemmed from their different absorptions by glass filters and other substances. This
is due to the wavelength dependence of absorption, a factor not considered in Herschels long quantitative discussion of light and heat absorption. So he concluded that heat and light seem to be entirely unconnected. However, he was close to doing the decisive experiment, stating that the answer to the question whether heat and light can be occasioned by the same rays or not [11] lies in the visible spectral range: it can only become a subject of inquiry, whether some of these heat-making rays may not have a power of rendering objects visible, superadded to their now already established power of heating bodies. So he asked is the heat which has the refrangibility of the red rays occasioned by the light of these rays? and studied the absorption of the heat of red light by various substances. Surprisingly in light of the preceding meticulous experiments, he did not relate this effect quantitatively to the perceived brightness of the transmitted red light. Instead he started handwaving when discussing a dark-red glass that absorbed close to 70% of the heat generated by red light: I am assured that red glass does not stop red rays. Indeed the appearance of objects seen through such coloured glasses will be a sufficient proof to every one that they transmit red light in abundance [11]. Here he was probably deceived by the complicated non-linear relationship between light intensity and perceived brightness [14], bringing with it that a relatively large loss of light intensity might remain unnoticed. Ignorant of this possible source of error he concluded here we have a direct and simple proof, in the case of the red glass, that the rays of light are transmitted, while those of heat are stopped, and that thus they have nothing in common but a certain equal degree of refrangibility [11]. The separation of radiation into three different kinds visible light, heat radiation and radiation producing chemical effects (UV light) was widely accepted during the first half of the 19th century [12,15] and later abandoned in favour of a unified theory of radiation. Early advocates of the latter were Andr Marie Ampre (17751836) between 1832 and 1835 [10,15,16], Macedonio Melloni (17981854) around 1842 [10,12,15,16] after a U-turn in his interpretation, and Sir John Frederick William Herschel (17921871), the only child of William, between 1835 and 1845 [16]. More detailed accounts of the arguments for and against the unified theory of radiation have been published [10,12,15,17]. When Herschel discovered infrared radiation, most scientists did not accept the wave theory of light and the concept that light of a certain colour corresponds to radiation of a certain wavelength. This did not change for more than a decade, even though the wavelength of visible light was measured by Thomas Young (17731829) shortly afterwards [10,18]. Since light was dispersed by prisms, its wavelength was not directly accessible in most of the infrared experiments until the 1950s. By observing interference fringes or by using gratings, wavelength information could be obtained. In this way, the calibrated spectral range was extended slowly throughout the 19th century to 1.445 m in 1847 by Jean Bernard Leon Foucault (18191868) and Armand Hippolyte Louis Fizeau (18191896) [10], to 1.9 m in 1859 by J. Mller [19], to 7 m in the mid-infrared range (2.5 m to 50 m) by Paul Quentin Desains (18171885) and Pierre Curie (18591906) in 1880 [10,19], and to 150 m in the far-infrared range (50 m to 1000 m) in 1897 by Heinrich Rubens [10,19]. The mid-infrared spectral range attracted most attention thereafter, whereas only a handful of articles were published in the far-infrared range up to 1938 [20] and only 91 publications in the near-infrared range until 1950 [21].
1.2. The Long Way to Modern Instruments Sir John Frederick William Herschel (17921871) followed in the footsteps of his father and became himself a renown astronomer. He devised an ingenious way to render infrared radiation visible and reported 1840 probably the first multichannel infrared spectrometer [22]. The publication deals primarily with photographic experiments and foresees already naturally coloured photographic images. Note III of this work is concerned with a process for rendering visible the calorific spectrum by its effect on paper properly prepared. The process is described as follows: It is well known to artists in water colours, that their tints, when freshly laid on and wet, are deeper and darker than they ultimately become on drying, a change which must be allowed for in the colouring, or the effect will be spoiled. If a paper so over-coloured be dried unequally, those parts which are dry first appear lighter than the rest. He used this evaporation effect to afford a visible picture of the thermic spectrum. In his experiment, dispersed sunlight fell on a specially prepared paper: one side of this paper is to be smoked in the flame of oil of turpentine, or over a candle burning with a smoky flame, by drawing it often and quickly through the flame, giving it time to cool between each exposure, till it is coated on the under side with a film of deposited black, as nearly uniform as possible. The paper presents its white side to the incident spectrum. Then a flat brush, equal in breadth to the paper, dipped in good rectified spirit of wine, is to be passed over the white surface till the paper is completely saturated, which will be indicated by its acquiring a uniform blackness in place of the white it at first exhibited. After a few moments exposure, a whitish spot begins to appear considerably below the extreme red end of the luminous spectrum, (supposing the violet end uppermost). As shown in Fig. 4, five white spots were produced on the paper, indicating regions of transmission for solar radiation. Herschel discussed already that the atmospheres of the sun and the earth might cause the observed pattern: The gaseous media through which the rays have reached their point of action, are the atmospheres of the sun and earth. The effect of the former is beyond our control, unless we could carry our experiments to such a point of delicacy as to operate separately on rays emanating from the centre and borders of the suns disc. That of the earths, though it cannot be eliminated any more than in the case of the suns, may yet be varied to a considerable extent by experiments made at great elevations and under a vertical sun, and compared with others where the sun is more oblique, the situation lower, and the atmospheric pressure of a temporarily high amount. Should it be found that this cause is in reality concerned in the production of the spots, we should see reason to believe that a large portion of solar heat never reaches the earths surface, and that what is incident on the summits of lofty mountains differs not only in quantity, but also in quality, from what the plains receive [22]. As this example also demonstrates, exploration of the infrared spectral range was impeded by the lack of sensitive detectors. This situation improved slowly during the 19th century. The first two important innovations were made by Macedonio Melloni (17981854) in 1830 [24,25] and by Samuel Pierpont Langley (18341906) in 1880 [10,19,26]. Melloni developed a thermophile detector for heat radiation which was based on the discovery of the thermoelectric effect by Thomas Johann Seebeck (17701831) in 1821 [10] and which was inspired by a thermophile thermometer developed by his friend Leopoldo Nobili (17841835) by 1830 [24]. Mellonis detector could detect the radiation from a person 610 m away and was 40 times more sensitive than
Figure 4. The near infrared spectrum of solar radiation recorded with an early multichannel infrared spectrometer in 1840 by Sir John Herschel [22]. Top (labelled Fig. 2): light intensity made visible by evaporation of alcohol from a soaked paper. The light is dispersed horizontally with longer wavelengths on the left hand side. The y-axis indicates yellow light (~580 nm). Spots to are transmission maxima of solar radiation in the near-infrared spectral region. The dark spaces between them are regions of absorption of atmospheric gases [23] (if not the glass prism contributes), in particular of water vapour and CO2. Bottom (labelled Fig. 3): visible spectrum as seen with the naked eye [22]. A reasonable interpretation of the top spectrum seems to be that the waist between spots and is due to the several atmospheric absorptions around 750 nm, and that the dark regions between spots and , spots and and spots and are due to the water vapour absorptions around 930, 1130 and 1400 nm, respectively. The recorded spectrum therefore seems to extend out to 16001700 nm into the near-infrared spectral region. This interpretation is in line with the fact that smaller wavelengths produce a larger spread of a fixed wavelength interval on the recording paper than larger wavelengths. Between 400 and 580 nm the spread is estimated to be ~2-fold larger than between 580 and 750 nm and ~4-fold larger than between 750 and 930 nm in accordance with known values [9,13].
thermometers [24,25]. In contrast to William Herschels passing interest, Melloni studied heat radiation throughout his scientific career. Amongst his other achievements were the discovery of the transparency of rock salt (NaCl) for infrared radiation in 1833 [2729], which was of eminent importance to expand the accessible spectral range, usage of the first [8] rock salt prism in the first mid-infrared spectrometer in 1833 [2729], detection of infrared radiation from the moon in 1846 [25] and of variations in the earths atmosphere in 1852 [10,25]. The next milestone in the improvement of infrared detectors took place in 1880 [10,19,26] when Samuel Pierpont Langley (18341906), who later became also an aeronautic pioneer, developed his first bolometer. This detector measured radiation intensity via a resistance change of a small metal strip in a Wheatstone bridge, a measuring principle which had been demonstrated before by Adolf Ferdinand Svanberg (18061857) in 1851 [30]. Langleys bolometer could detect temperature differences of 105 C [10,26]. Langley and his assistant Charles Greeley Abbot further improved the sensitivity 400-fold by 1898 and enabled detection of a cow at a distance of 400 m in 1901 or of a temperature difference of 108 C [26,31]. Further technical development of dispersive infrared spectrometers is beyond the scope of this overview. However, a remark on the effort needed to obtain an infrared spectrum might be of interest. Langley measured more than two weeks for some of the data points of his solar spectrum in the early 1880s [10,26]. Coblentz around 1905 needed only 1.5 min per data point or four hours for a spectrum from 5000 to
670 cm1 [8,32], a situation which did not improve for nearly half of a century [33,34] until the second generation of commercial instruments became available after the second world war. They reduced the measuring time to 20 min [35]. This is still 1000times slower than modern rapid scanning Fourier transform spectrometers, which can acquire a spectrum in as little as 10 ms or high quality spectra within seconds. The obstacles for the early investigators implied that only a few spectrometers world-wide were operated by specialised physicists for much of the first half of the 20th century. The instruments and their accessories were mostly custom-built. This is very different to the present situation where routine measurements can be run by personal with little training and spectrometers as well as ample accessories are commercially available for the entire field of bioanalytics. The first commercial infrared spectrometer was produced in 1913 [3537] by the English company Adam Hilger Ltd which continued manufacturing infrared spectrometers until 1974 [37]. In 1936 American Cyanamid Co. started a small scale production of infrared spectrometers for industry [8,38]. During World War II, the USA government commissioned infrared spectrometers from National Technical Laboratories (later Beckman Instruments, model IR-1 from 1942 see Fig. 5) and Perkin-Elmer (model 12A from 1944). Figure 6 shows the successor model 12C. These early models were dc instruments that would operate only in humidity- and temperature-controlled atmospheres. Because central air conditioning was still a few years away, the first industrial spectroscopists tended to barricade themselves in their specially controlled rooms, letting no one else in lest the spectrum being recorded be ruined. As a result, spectroscopists earned the reputation of being recluses who spoke only to other spectroscopists [38]. What it was like to operate these early commercial instruments is vividly described by F.A. Miller [35] at the example of the Perkin-Elmer instrument 12B, which was upgraded from model 12A by an automatic spectrum recorder. This simplified the measurements enormously compared to the previous manual recording of every data point. The instrument was a single beam spectrometer, which meant that the spectra with and without sample had to be recorded one after the other. The instrument was not linear in anything useful m, cm1, %T, or absorbance. Extensive replotting of the raw data was therefore necessary to obtain a real spectrum [35]. Marks that were automatically drawn on the abscissa of the chart paper had to be calibrated against gas reference spectra in order to translate them into wavelength or wavenumber. The ordinate recorded the signal of the thermocouple detector, which was determined only partly by sample absorption but also by lamp and spectrometer characteristics, atmospheric absorption and any other heat change close to the spectrometer. If one lit a match and held it near the instrument housing, the pen moved [35] or in a laboratory that was heated by steam radiators: when the steam came on, the baseline climbed upward and went off the paper if zero was not reset [35]. Therefore, the signal without measuring light had to be checked frequently and the measuring light intensity manually evaluated as the difference between the signal with and without measuring light. In consequence one had to work hard to run and plot two spectra a day [35]. An anecdote says that students of MIT, who had to evaluate the original recordings of such a spectrometer, complained to Arthur C. Cope, chairmen of the department, but Cope said that it was good experience for them. Finally one day Cope said that he would replot a spectrum to show the students that he was willing to do what he asked of them. He took the material home that weekend, and the story is that when he came in on
Figure 5. Photograph of Beckman IR-1, one of the first commercial infrared spectrometers. Photograph courtesy of Heritage Exhibitions.
Monday he authorized the purchase of a Baird double-beam instrument [35]. This instrument avoided these troubles. The introduction of double-beam spectrometers with chopped infrared radiation shortly after World War II was a vast improvement in comfort and velocity [35]. Detecting the signal with and without measuring light in rapid succession avoided the problem of thermal drift of the older, dc-operated instruments and increased the sensitivity. This and other improvements were greeted in 1948 by V.Z. Williams stating the infrared spectrometer is no longer a capricious instrument which must be housed in the sub-basement for mechanical and thermal stability. It has changed in appearance from the bulky, wax-sealed box and accessory equipment which required a fair sized room to [a] compact unit which [occupies] the space of a standard desk [16]. The double beam spectrometers of the 1950s were expensive instruments. A review in 1963 [39] gives a price of 15 000 $ for a good spectrometer a couple of years earlier, a price that corresponds to 110 000 $ in May 2008 taking into account 640% inflation from January 1960 [40]. An example is the first widely used commercial infrared spectrometer, the PerkinElmer 21 (see Fig. 7), which was officially introduced in 1950. The design and flexibility of this double-beam spectrometer made infrared spectroscopy readily accessible to non-specialists. This led to a rapid growth in the application of infrared spectroscopy for analysis of diverse chemical and biological systems. It would not be incorrect
10
Figure 6. Photograph of RDB Fraser next to a infrared spectrometer in 1958 at the CSIRO Division of Protein Chemistry in Melbourne, Australia. This spectrometer is a Perkin-Elmer Model 12c modified for use with a selenium film polarizer, and a 0.8 NA reflecting microscope (the latter shown detached on the top of the pen recorder). Photograph courtesy of RDB Fraser.
Figure 7. Photograph of Juana Bellanato next to a Perkin-Elmer model 21 spectrometer. The photograph was taken in 1956 when she was working in the Physical Chemistry Institute of the University of Freiburg (Germany) as a postdoctoral researcher with Prof. Mecke and Dr. E. Schmid. She was one of the first scientists to apply infrared spectroscopy for characterisation of lipids. The photograph is reproduced with the permission of Juana Bellanato.
11
to state that this provided the basis for laying the early foundations of biological infrared spectroscopy. During late 1940s and 1950s, there were a large number of publications that used dispersive infrared instruments to analysis nucleic acids, lipids, peptides, proteins, carbohydrates, bacterium, viruses and human tissues (see later). Indeed, most of the systems that are currently being studied using the sophisticated FT instruments had already been investigated using dispersive instruments in late 1940s and 1950s. 1.3. Discovery of the Usefulness of Infrared Spectroscopy for Chemical Analysis Infrared spectroscopy is one of the classical methods for structure determination of small molecules. This standing is due to its sensitivity to the chemical composition and architecture of molecules. Bond lengths and bond angles can be measured and bond distortions detected with picometer precision. Apart from that, redox state, interactions with the environment like hydrogen bonding and electric fields as well as conformational freedom reflect in the spectra. This usefulness of infrared spectroscopy for chemical analysis was discovered mainly at the end of the 19th and the beginning of the 20th century. The first indication came however already in 1833. Melloni studied the transmission of heat radiation through substances. He used light sources at different temperatures, which provided radiation with different spectral compositions according to Wiens displacement law. He found that each substance had characteristic transmission properties [2729]. With this it may be said that analytical infrared spectroscopy was born [25]. 50 years later, Sir William de Wiveleslie Abney (18431920) also known for his fundamental work in photography pioneered the use of infrared spectroscopy for chemical analysis [8]. In 1881, Abney and Edward Robert Festing published the first [21] near infrared spectra of organic compounds, reporting amongst others the absorption of 48 organic liquids [41] up to about 1.3 m [10]. In this spectral region predominantly the absorption due to overtones of C-H stretching vibrations were observed [8]. Bands specific for aromatic and ethyl groups were found which indicated the analytical potential of the method as pointed out by the authors: We may say, however, it seems highly probable by this delicate mode of analysis that the hypothetical position of any hydrogen which is replaced may be identified, a point which is of prime importance in organic chemistry and It seems to us that the spectra leave as definite characters to read as are to be found in hieroglyphics, and we venture to think that we have given a clue to enable them to be deciphered [41]. The characteristic absorption spectra provided even a tentative interpretation of the solar spectrum: in two instances at least, a study of the absorption spectra of organic bodies has to some extent thrown a glimmering of meaning on some of the absorption lines of the solar spectrum . One of the authors of this chapter (AB) was pleased to find out that Knut Johann ngstrm (18571910) recorded one of the first [8] mid-infrared spectra of organic liquids and gases in 1890 [42] while being employed by the predecessor of ABs university, Stockholms Hgskola. Three years later, also William Henry Julius (1860 1925) published mid-infrared spectra of organic compounds [8,32] and suggested that the absorption is due to internal motions in the molecule and that the internal structure of the molecule determines the spectrum [8]. The concept that many chemical groups absorb in narrow frequency bands was finally established by the work of William Weber Coblentz (18731962). Not satisfied with the available data he wrote in
12
1904 [43]: The investigation of absorption spectra far into the infra-red has never been made in a thoroughly systematic manner. This is no doubt due to the enormous difficulties to be encountered and the slowness with which observational data can be obtained, so that usually after investigating half a dozen compounds, the results have been given to the public. As a consequence the agreement in the location of certain absorption bands are not always convincing. Accordingly he measured spectra of more than hundred compounds around 1904 [43] and published a list of group frequencies [44]. Surprisingly, it took about 30 years until the usefulness of infrared spectroscopy began to be realised in industry [8,16,36,45] and in medical research [8] during the 1930s. The same time period saw the onset of biological work as described in the next section. The breakthrough of infrared spectroscopy came in World War II, where it was employed in three programs of the USA and UK governments: for quality control in the production of synthetic rubber [4,8,35], for the analysis of petroleum [4,8,35], for example to trace the origin of gasoline used by the German airforce [4,46], and to resolve the structure of penicillin [8,35,36]. As a consequence, the number of instruments rocketed from around 10 to several hundred in the USA [4,47,48]. After World War II, with commercial availability of infrared spectrometers, there was rapid growth in the use of infrared spectroscopy for analysis of organic molecules. The work from this period has been described in the following manner by Williams in 1951 [49]: The period since World War II has resulted in a steady growth throughout the infrared field. This growth has been mainly one of utilization and extension of wartime developments and techniques rather than one of fundamentally new discoveries. Even so, the process of utilization and extension has been so broad that the field has altered radically in the last five years. An example of this alteration is the acceptance of the infrared spectrum by the organic chemist. [The] issue of the Journal of the American Chemical Society, January, 1950, shows a large number of infrared spectra of organic materials . The interesting point is that, in many cases, no description at all is given of the spectrometer or the sampling conditions used. At some sites, evidently, the infrared spectrometer has been so successful as to reach a stage of oblivion, and now ranks with the distillation column or gravimetric balance as standard laboratory furniture. Actually, this is to be expected since there are probably over 1000 infrared spectrometers in use today. 1.4. From Chemistry to Biology The high information content in an infrared spectrum is of use also for biological systems. This makes infrared spectroscopy a valuable tool for the investigation of structure and function of biomolecules and of cells and tissue. One of the first infrared studies of biological systems was performed by Nobili and Melloni with their new, sensitive thermopile detector for heat radiation. They investigated more than 400 insects, discovered that caterpillars have a higher temperature than the butterflies and chrysalides which proceed from them [50] and related this to their higher metabolic activity: the insect, in the first period of its life, where its nourishment is abundant and its growth rapid, converts into carbonic acid a much greater quantity of oxygen, than at subsequent periods. the heat of the animal will vary, so to speak, proportionally to the quantity of oxygen employed in the act of respiration. Nobili and Melloni pointed out that their method was non-invasive and thus superior to previous experiments which
13
did not give the temperature of the animal in the natural state, but in a maimed and suffering condition whereas their thermo-multiplier offers the means of repeating these experiments without incurring any of the inconveniences above alluded to [50]. Still nowadays, the non-invasiveness of infrared spectroscopy is one of its striking advantages, since no artificial spectroscopic probes have to be introduced into the sample that might alter its properties. The most abundant biological molecule water was exposed to infrared radiation already by William Herschel. Melloni studied amongst other substances cowhorn, citric acid, sugar and ice [2729]. Ernest Fox Nichols (18691924) recorded nearinfrared spectra of chlorophyll and hemoglobin in 1892 [51], which were, however, dominated by solvent absorption. Pioneering work was also done by Coblentz, who studied fatty acids in 1904 [43]. Between 1930 and 1950, at the same time when industry became interested in infrared spectroscopy, research in the mid-infrared spectral range started on all kinds of biomolecules and on larger biological systems: tissue in 1933 [52,53]; polysaccharides [54], amino acids (some of them in aqueous solution) [55,56], polypeptides [55], and proteins (some in water) in 1935 [54,57,58]; the simple fatty acid acetic acid in 1936 [59], longer fatty acids in 1940 [60], and steroids in 1946 [8]; vitamin C in aqueous solution in 1937 [61]; and nucleic acids in 1948 [62]. The penicillin program during World War II led to the elucidation of its structure by infrared spectroscopy [8,36]. Otherwise, the war years had a negative impact on biological infrared spectroscopy as some of the scientists had to focus on war related projects and this is evident from a survey of the literature published during that period. For example, in one paper published in 1946 by Furchgott et al. [63], the authors wrote the following as a footnote: Infra-red analysis of the steroids was begun in this laboratory by Carl Herget and Ephraim Shorr, and was the subject of a brief report in 1941. At that time Dr. Herget left this work in order to engage in war research at the Underwater Sound laboratory, Harvard University. Over the last 30 years the application of infrared spectroscopy for biological applications has been accelerated by the following key factors that will be discussed in more detail below. 1. Advances in instrumentation, especially the commercial availability of FTIR spectrometers, has been the single most important factor in the rapid growth in the application of infrared spectroscopy for biological analysis. 2. The use of computers has also been an important development since it made possible the digital spectral subtraction, especially the absorption of water, for obtaining difference spectra. However, it is important to stress that use of computers for analysis of infrared spectra began before the advent of FT instruments. 3. Advances in chemical and molecular biological methods that enabled production of samples with targeted alterations for structure-activity and spectral interpretation studies. Ability to carry out site directed mutagenesis, chemically synthesise peptides and introducing isotopically labelled groups probably had the greatest impact. 1.5. Emergence of FTIR Instruments and Computer Controlled Spectrometers Triggers Renaissance in Biological Infrared Spectroscopy In the 1960s, Jones [64] and Savitzky [65] were pioneers in the development of mathematical and computational approaches for analysis of infrared spectral data. The effort of these scientists, and many others, led to the development of computer aided
14
dispersive instruments in the 1970s. This was happening at virtually the same time when Fourier transform infrared (FTIR) spectrometers were also entering the market. It is interesting to note, however, that despite the appearance of FT instruments, some in the scientific community dismissed the idea that dispersive instruments were confined to history. As an example, James Mattson stated the following in one of his articles [66]: Infrared spectrophotometry has long had the reputation of being semiquantitative. For this reason, some of the attention being lavished on Fourier transform infrared instrumentation is derived from misconceptions regarding conventional dispersive instrumentation. He then goes on to predict a bright future for dispersive instruments: The dispersive instrument manufacturers are moving slowly into the world of today, an era of sophisticated, computer-based data acquisition and reduction. As they begin redesigning their 10-year old dispersive instrumentation with low-cost mini- or microcomputers, the field will enjoy a second childhood. However, the predictions of Mattson [66] did not come to fruition and virtually all spectrometer manufactures are currently producing FT-instruments instead of dispersive instruments. Mattson is no longer engaged in scientific research but in a recent communication with him, he stated the following about dispersive instruments: About 10 years ago, I visited the Rosenstiel School of Marine & Atmospheric Science campus and saw my beloved Perkin-Elmer 180 sitting in a hallway because nobody knew what to do with it. That was a great machine. NSF (National Science Foundation) paid a lot of money for it. There was a world of things I could have done and would have done if that machine had been under my control. So many problems; so much to be done. I assume there are many research-quality spectrophotometers sitting around university laboratories doing nothing. Although dispersive instruments are no longer routinely used for recording infrared spectra, they are no means obsolete as they can be valuable for specialised applications such as time-resolved studies. The first commercial FTIR spectrometer (Model FTS-14), totally computer controlled, was introduced by Digilab in 1969. This led to a surge of application similar to what was observed when the first commercial dispersive infrared spectrometers gained widespread use in the late 1940s and through 1950s. However, the difference this time was that the advance in instrumentation caught the attraction of scientists who have been waiting to make use of infrared spectroscopy for analysis of biological systems in aqueous media. As such, it would not be incorrect to state that the advent of FTIR spectrometers had a greater impact for life science research than chemical science. Indeed, the first FTIR paper to appear in the literature database (ISI) was a biological study reported in 1972 [67]. Figure 8 shows a Digilab FTIR spectrometer in James Albens laboratory that has been used for studying protein-ligand interactions by infrared difference spectroscopy. Recent communication with James Alben, provided some insight about his experience of using both dispersive instruments and FT-instruments. Firstly, he described his early studies with dispersive instruments in the following manner: The need to understand the roles of iron, copper and oxygen in respiration led to studies with Winslow Caughey at Johns Hopkins, on the coordination chemistry of transition metal porphyrin complexes, and the effects of ligand field strength. We characterized a range of porphyrin derivatives by infrared spectroscopy by use of a sodium chloride prism instrument (Perkin-Elmer Model 21) which allowed a 5x expansion of %Transmission. Inadequacies of measurement led to use of a Perkin-Elmer 400 grat-
15
Figure 8. A photograph of a Digilab FTIR instrument with modified data system displaying CO bound to hemoglobin minus its photoproduct. The photograph, courtesy of James Alben, was taken in Albens laboratory sometime in the early 1980s.
ing instrument, courtesy of Ellis Lippincot at the University of Maryland, where we first observed carbon monoxide coordinated to iron in hemoglobin. An example of a Perkin-Elmer 21 instrument that Alben refers to is shown in Fig. 7 from a photograph taken in 1956. In early 1970s, Alben had access to FTinstrumentation [67]. He states the following about his experience with FT-instruments: A major breakthrough in spectral resolution, signal/noise, and baseline stability, came with incorporation of a Block Engineering interferometer into a bench-top instrument by Tom Dunn Associates (later to become Digilab Division of Block Engineering). This instrument was originally delivered with a Data General Nova minicomputer that contained 16 kilobytes of core memory (3 microsecond cycle time) but no disk drive. Three-fourths of that memory was eventually exchanged for a 128 kilobyte head-per-track disk drive that permitted data collection at 0.5 cm-1 resolution and 32 bit fourier transforms. The signal/noise obtained was ten-fold better than that of the best grating instrument. With the commercial availability of FT-instruments, increasing number of scientists were using the technique for studying different types of biomolecules. The first application of FTIR spectroscopy for analysis of lipids in aqueous media was reported by Jack Koenigs and Henry Mantschs group [68,69]. However, not everyone had the necessary funds to purchase a new FTIR spectrometer, and hence many leading groups continued to use their dispersive instruments well into the mid-1980s. Following is a quote from Michael Byler, who was working with Heino Susi, regarding his desire to purchase a FT-instrument (see photograph of Michael Byler in front of his new FTIR spectrometer, Fig. 9): At this time, commercial mid-IR Fourier-transform spectrometers (based on the principle of the Michelson interferometer) had been on the market for just a few years.
16
Figure 9. Photograph of Michael Byler, colleague of Heino Susi, proudly sitting in front of his newly acquired FTIR spectrometer at the U.S. Department of Agriculture, the Eastern Regional Research Center in Wyndmoor, PA in August 1981. Photograph courtesy of Michael Byler.
According to the literature, they offered numerous advantages over even the best dispersive instrument. But they were expensive: more than $80000, probably equivalent to at least $200000 in todays funds. At the time, USDA management had other spending priorities and Susi was rather sanguine about the prospect of them granting us such a sum of money. Even our second choice, a Perkin-Elmer 180, perhaps the best all round dispersive instrument then available, was priced at over $50000. Nonetheless each new budget year we added our request for a new IR to managements instrumentation wish list. Suddenly in 1980, sufficient funds became available to purchase a new Nicolet 7199 FTIR. Ironically, at the time both of us were so deeply involved with other assigned research, that we did not attempt our first protein spectrum until more than two years later. Interestingly, standard FTIR spectrometers are currently sold at the same nominal price as the dispersive spectrometers of the 1960s implying that the better performance of the present instruments can be obtained at ~7-times lower cost. It would be plausible to state that 1980s is the starting point for the new era of biological infrared spectroscopy with highly active research groups emerging in different parts of the world. Rapid data acquisition and greater access to instruments made infrared spectroscopy accessible to a wider community of scientists and not only chemists and physicists. It was now much more common for infrared spectrometers to be found in biochemistry and life science laboratories in universities and research institutes around the world.
17
1.6. Use of Computers in Infrared Spectroscopy As already mentioned earlier, the use of computers in biological infrared spectroscopy has been a major contributor in the growth of biological applications. Even before the advent of FT-instruments, computer aided infrared spectroscopy demonstrated that difference spectra can be obtained by digitally subtracting one spectrum from another and that the broad infrared bands can be analysed using mathematical and computational tools to reveal subtle details of molecular structure. The following examples illustrate biomolecular studies with dispersive instruments coupled to microprocessors. The first highlights also the importance of collaboration between academia and industry which played a significant role in advancing biological infrared spectroscopy. For example, Dennis Chapman at London University collaborated with Perkin-Elmer in England in his first studies of proteins and membranes in H2O [70]. Juan Gomez-Fernandez, a co-author of this paper, in recent communication with one of us, made the following remarks about the background to this study whilst working with Dennis Chapman in London (references removed): In 1978, Mantsch and his group began the application of FT-IR spectroscopy to the study of aqueous lipids and soon after they reported the possibility of water subtraction from aqueous lipid samples. When early in the summer of 1979 I returned to spend the summer period working with Dennis, I have seen these papers and I showed them to Dennis. He rapidly reacted realizing the wealth of possibilities that this technical advance could permit. At that moment Dennis did not know of any FT-IR spectrometer available to us, but he has a very good knowledge of Perkin-Elmer innovations, among other reasons because he lived in Beaconsfield at a very short distance of a Perkin-Elmer factory. Dennis knew that a grating infrared spectrometer controlled by a data station has been just introduced. It should be commented, at this point, that computers were in its infancy in 1979, and its use was still very rare. The data station was a very primitive computer, but it was sufficient to permit the digital acquisition of spectra and to work with them, performing, for example, subtractions. Dennis quickly called Mary Barnard, a scientist working for Perkin-Elmer, and he concerted a visit to the factory. Rapidly my colleague Felix Goi and I prepared samples containing just lipids and also protein-lipids samples. In a few days we were taking our first spectrum. The study by Chapman and co-workers described above [70] was conducted using thin pathlength transmission cells. In contrast, Mattson et al. [71] used internal reflection spectrometry to record infrared spectra of proteins in aqueous solution. They also used a minicomputer, interfaced to a dispersive Perkin-Elmer 180 spectrometer, to obtain the protein spectra in H2O and carry out digital subtraction of the overlapping water absorbance. In spite of the first encounters between computers and infrared spectrometers in the 1960s, it was not until the mid-1970s that the marriage between computer and infrared machines was initiated with the advent of Fourier transform instruments. It was now possible to record spectra very easily and what was needed were mathematical tools that could be used to process and analyse the data. One of the fundamental advantages of computerisation has been the ability to digitally subtract the absorbance of H2O from aqueous samples and thereafter analyse the broad infrared spectra of biomolecules with computational and statistical tools. The other major problem in biological infrared spectroscopy is the overlap of peaks arising from different structures and
18
groups in complex macromolecules. Here again it is the development of computer programmes for unmasking the complex band envelope that played a pivotal role in proliferating the use of the technique for analysis of biological molecules. In the 1980s the Fourier Self-Deconvolution method for analysis of infrared spectra was introduced by Kauppinen et al. [72]. This is a mathematical procedure for resolving overlapping component bands in a complex spectrum. It is also referred to as resolution enhancement although the experimental spectral resolution remains unchanged. At virtually the same time, the FTIR spectrometer manufacturers were busy developing their own suite of software programmes to aid the analysis of infrared spectral data. Virtually, all the FTIR manufacturers provided programmes for spectral subtraction, smoothing, derivative and deconvolution analysis. The availability of such software played an important role in broadening the application of the technique to the analysis of subtle changes in infrared spectra of biomolecules. Below is a quote from Michael Byler (with references removed) indicating the helpful role played by spectrometer manufacturers in providing software programmes for data analysis: After checking with the manufacturer, I learned that a modified version of the deconvolution algorithm of Kauppinen et al. was available. We ordered the software and I soon learned how to apply it to a variety of spectra. About the same time, I had learned that comparable band narrowing could also be achieved by means of calculating the second derivative of a spectrum. Each method presented its own advantages and disadvantages, but for initial exploration, the theory and application of differentiation proved to be simpler. Unlike deconvolution, analytical calculation of a derivative of a spectrum requires no a priori knowledge of any band parameters. One of the first infrared spectroscopic application of resolution enhancement methods for protein analysis was made by Susi and Byler in 1983 [73]. They obtained second derivative Fourier transform infrared spectra of the native and denatured soluble proteins in deuterium oxide between 1350 to 1800 cm1. They state in their paper In the second derivative spectra, clearly resolved peaks are observed which can be associated with the alpha-helix, beta-strands, and turns. No protein spectra with such resolution have heretofore been reported. The data appear to present the first direct spectroscopic evidence of turns in a native protein. This first paper was important in showing the usefulness of second-derivative analysis for protein analysis using infrared spectroscopy which prompted a plethora of studies to be reported in the literature during the late 1980s. Infrared analysis by Byler and Susi were conducted for proteins dissolved in 2H2O. The resolution enhancement techniques enabled a detailed analysis of bands making it possible to compare spectra of proteins recorded in both H2O and 2H2O. As a consequence, for the first time, the complications associated with the interpretation of spectral data for samples in 2H2O was highlighted by Olinger et al. [74]. The authors state This paper represents the first example of the use of deconvoluted Fourier transform infrared spectra in conjunction with hydrogen-deuterium exchange in order to aid in the assignment of a proteins infrared bands. The paper by Haris et al. [75] states The results show that it is necessary to be cautious in making band assignments based on exchange methods unless the extent of exchange is known. Furthermore, it is seen that the combination of Fourier transform infrared spectroscopy and hydrogen-deuterium exchange is a powerful technique for revealing small differences in protein secondary structure. Before ending this section, it is important to state that long before the use of the so called resolution enhancement procedures, difference spectroscopy (see section on
19
difference spectroscopy and Chapter by Barth) was already being used to probe subtle changes in biomolecular structure using infrared spectroscopy. However, these studies were mainly restricted to systems which can be triggered from one state to another without having to re-assemble the infrared cell which may lead to changes in pathlength, resulting in artefacts. Difference spectroscopy became more popular as a result of development of FT instruments since high signal-to-noise ratio spectra could be obtained rapidly.
2. Dealing with H2O in Infrared Analysis of Biological Systems This section summarises the historical context of some of the key issue of dealing with H2O absorbance that has been a major hindrance for biological infrared studies. The major challenge for recording infrared spectra of biological molecules was the strong absorbance of H2O over much of the mid-infrared region. Early infrared measurements in H2O were restricted to analysis in the near-infrared region and to selected window regions in the mid-infrared. In order to illustrate the problems encountered in biological infrared spectroscopy in the 1950s, we have included below quotes from Barer, working at Oxford University, in his bid to use infrared microspectroscopy for biological studies. In a discussion of the Faraday Society, Barer [76], wrote the following with respect to the problem of water absorbance and possible ways to overcome this problem: The next question which is of considerable interest to the biologist is whether it will ever be possible to apply the method to the study of living cells. The difficulties here are formidable. With few exceptions, living cells must be examined in an aqueous medium, and all cells contain water. Water possesses a number of strong absorption bands in the infra-red region, and, indeed, workers in this field know all too well that the effect of even the small amount of water vapour normally present in the atmosphere can be very disturbing. It might conceivably be possible to work with extremely thin films of water and at wavelengths at which the absorption due to such films would not overshadow everything else. Another possibility is to use heavy water, which has a rather different absorption spectrum, provided that it did not affect the structure and viability of the cell. In this way, by the use of two different media it might be possible to derive the absorption spectrum of the object itself. Another factor to be considered is the possible action of the absorbed radiation on the cell. I have carried out preliminary observations on the action of short-wave infra-red radiations on living cells and the results suggest that they do not tolerate this treatment very well. For all these reasons it must be admitted that the prospect of applying the method to living material is extremely remote. Clearly, measurement in aqueous media has been a major hindrance in the application of infrared spectroscopy for biological applications a fact that has been repeatedly highlighted in many publications from 1930s until the mid 1970s. The first reference to recording the infrared spectrum was spectrum of H 2O was reported in 1895 [77]. In 1911, Coblentz reported a study investigating the interaction between gelatine and water and investigated bands associated with both of these molecules [78]. After Coblentz, a number of scientists were very active in the analysis of the interaction between water and biomolecules. They published a series of papers between late 1930s and mid 1940s studying proteins, carbohydrates and amino acids. For example, Buswell and Rodebush analysed biomolecules in water in late 1930s [79]. Ellis &
20
Bath [80] obtained near infrared spectra of oven-dried gelatin and gelatin saturated with H2O and 2H2O vapour respectively. These authors continued their infrared studies on proteins and carbohydrates during the 1940s. Towards the late 1940s there was greater confidence in the ability to carry out analysis in H 2O and 2H2O. 2H2O provides further window regions for assessing bands arising from biological molecules which was appreciated in late 1940s [81]. Gore et al. state the following in one of their articles [81]: One of the widespread misconceptions concerning the application of infrared spectrometry in organic analysis is that it is impossible or at least difficult to obtain spectra of aqueous solutions. On the contrary, it is often of extreme value to observe spectra in aqueous solution, at various hydrogen ion concentrations, especially in the case of carboxylic acids and amino acids. One of the key persons to pioneer the application of infrared spectroscopy for analysis of biomolecules in aqueous solution (2H2O) is Henri Lenormant [82] who was one of the first to record spectra of biomolecules in solution by dissolving them in 2 H2O. Subsequently, he collaborated with Blout which led to the publication of number of papers on analysis of biomolecules in aqueous system [83]. Lenormant and Blout were very active in infrared measurements in 2H2O during the 1950s. In late 1950s, Parker and co-workers [84,85] took advantage of advances in instrumentation and availability of barium fluoride windows (for containing aqueous solutions), to obtain spectra of biomolecules in H2O. Parker reported studies on infrared spectra in H2O solutions saturated with particular biomolecules such as amino acids and proteins so that the peaks in the window regions (approx. 1550950 cm 1) where H2O does not absorb strongly can be monitored. In the 1960s, Susi and co-workers [86] were one of the first to attempt analysing the amide I band which occurs at virtually the same frequency as the O-H bending vibration of the H2O molecule. Up to that point, the vast majority of studies of biomolecules in H2O avoided regions containing strong absorption bands arising from the water molecule. The following text from the Susi et al. paper in 1967 [86] highlights the difficulty they had to go through in order to visualise the amide I band in H2O: Measurements of amide I and amide II frequencies in H20 solution are possible only by extremely careful differential procedures; these are not easily adopted for routine investigations. Absorption by the solvent was cancelled by repeatedly adjusting the path length of the reference cell and measuring the differential absorption at various wave length settings until a spectrum was obtained which showed only characteristic polypeptide bands. To prevent aggregation of the solute, it was necessary to use dilute solutions, while the intense H2O absorption precluded the use of cell thicknesses larger than 0.01 mm. An ordinate scale expansion device, which forms an integral part of the employed spectrophotometer, was therefore used. Although data obtained in this manner are not as precise as corresponding data obtained in D2O solution, repeated experiments led to reproducible and internally consistent results. The tedious and timeconsuming experiments were carried out in order to obtain information concerning the effect of dissolution of proteins in water without simultaneously introducing additional variables associated with deuteration. To the best of our knowledge this was the only paper where Susi reported the analysis of the amide I band for proteins dissolved in H2O. This despite the fact that from 1980 (personal communication with his colleague Michael Byler), he had access to a Fourier transform instrument. This is not surprising if one reads the following
21
statements taken from a book chapter [87] he wrote with Michael Byler (reference removed): When FTIR first became available, it was thought that the increased sensitivity would render infrared spectroscopy of proteins feasible even in H2O solution. Instead of the earlier differential methods, FTIR permitted one to obtain the solution spectrum and the solvent spectrum separately, and then to subtract the latter from the former, to obtain the spectrum of the pure solute. Unfortunately, neither the subtraction nor differential procedures are straightforward as they first appear. He then goes on to comment on the first attempt at using the FTIR instrument for analysis of the amide I band of proteins in H2O by Koenig and Tabb (1980) [88]. Susi and Byler noted that the amide I frequencies observed by Koenig and Tabb, appear to be uncertain and are actually in error for ribonuclease, evidently because of the difficulties inherent in subtracting the strong H2O band which absorbs close to 1640 cm1. On this basis, they go on to make the following recommendation: We strongly suggest that protein structure studies based on the important amide I band which absorbs at 16201680 cm1 be carried out in D2O solution wherever possible. Certainly, at low protein concentrations water subtraction is difficult and can be rather subjective. However, increasing number of protein spectra were being recorded in H2O and the similarity between these spectra and those recorded in 2H2O proved that H2O subtraction can be done accurately. As an example, Chapman and co-workers digitally subtracted H2O absorbance from spectra of proteins and lipids in 1980 [70]. They were excited about their new found ability to use computer programs for subtracting water absorbance and wrote the following in their paper [70]: A Perkin-Elmer infrared data station associated with a simple IR spectrometer (model 298) is shown to give excellent results with aqueous model and biomembrane systems. FTIR spectrometers were in the market place at the time when Chapman reported the above study, unfortunately he did not a get grant to purchase a FT instrument until 1985. Although he took advantage of the microprocessor for subtraction of the water absorbance using a dispersive instrument, he was looking forward to using the FT instruments as he was not entirely happy with the performance of the dispersive instrument. This is quite evident from the following statement taken from a book Chapter [89] by Chapman and Goni (reference removed): Perkin-Elmer introduced in 1975 the first microprocessor-controlled commercial dispersion infrared spectrometer, and advantage was taken of this facility for lipid studies a few years later. However, the main problems arising from conventional dispersive spectrometry were only overcome by the use of interferometric methods. In contrast to Chapman, Koenigs group had access to FT instruments almost as soon as they became commercially available. They took advantage of this advance in technology and were one of the main pioneers in demonstrating the practicality of digital removal of overlapping water absorption bands from spectra of biopolymers in H 2O solutions. By 1980, Koenig and co-workers published FTIR spectra of several globular proteins in aqueous solution after digital subtraction of water [88]. The FTIR spectra of proteins in solution were reported in a PhD thesis by Tabb [90]. During this period, the number of studies in H2O was still rather limited and vast majority of the FTIR studies were conducted in 2H2O although computer aided digital subtraction of 2H2O from protein solutions were carried out. Koenig wrote the following in one article (references removed) [91]: One of the spectral processing operations most widely used, beyond the simple computation of transmission and absorbance spectra, is the digital subtraction of ab-
22
sorbance spectra in order to reveal or emphasize subtle differences between two materials. Numerous applications of this procedure to polymer systems have been presented previously, and it is this digital subtraction capability more than any other single factor that has inspired subsequent investigation of polymeric materials by Fourier transform infrared spectroscopy. In a recent personal communication with Koenig, he said the following regarding his work on water subtraction: Three things came together which allowed spectral subtraction of water from aqueous solutions of proteins. First, FTIR produced an infrared signal that was stronger and digital compared to the dispersion. Secondly, there were (and we wrote some of them) computer programs written which could scale the signal linearly and allowed digital subtraction. Finally, researchers were convinced that the IR spectra of an aqueous solution of a protein was useful in understanding the structure of proteins. In 1986, Byler and Susi published a paper on quantification of protein secondary structure [92]. Few years later quantitative analysis on secondary structure of proteins from infrared spectra recorded in H2O were successfully achieved. As an example, three groups of scientists, working independently, published papers at a virtually identical period demonstrating the potential of quantifying secondary structure of proteins recorded in H2O. These were studies by Dong et al. 1989 [93, received by journal on July 11, 1989), Lee et al. 1990 [94, received by Journal on 5 Oct. 1989] and Dousseau and Pezolet 1990 [95, received by journal on 14 March 1990]. Dong et al. [93] quantified the secondary structure from the intensities of the peaks in the second-derivative spectra of proteins. This is similar to the method first developed by Susi & Byler in 1986 [92 also see section on quantitative analysis] which used the intensities of peaks in deconvolved infrared spectra of proteins in 2H2O using the curve-fitting method. In contrast, others used factor analysis [94] and partial least squares method [95] for their analysis. All these methods demonstrated good agreement between the X-ray data and the infrared data for the proteins recorded in H2O. After these pioneering studies, other methods for improving the quantification of protein secondary structure from infrared spectra of proteins have been reported. These have included methods that take into consideration the overlap of amino acid absorbance in the amide I region. Other developments include use of artificial intelligence techniques such as neural networks and genetic algorithms for quantification of protein secondary structure [96,97].
3. Infrared Sampling Techniques Infrared spectra of biomolecules can be recorded by transmission, reflection, emission and photoacoustic modes. Of these different methods, transmission and reflection modes are most widely used for biological applications. A brief discussion on the historical background to sampling methods is given below. 3.1. Analysis of Biological Samples by Transmission Infrared Spectroscopy Transmission studies have been the most established, and widely used, method for recording spectra of biological samples. Much of the early infrared studies on biomolecules were conducted in the solid state and involved producing thin films on infrared transmitting substrates. Samples were mulled in various mineral oils (for example, Nujol) and then spread on infrared transmitting disks. High vacuum sublimation onto
23
infrared transmitting disks and casting of aqueous solutions to give films on silver chloride disks were also used. The problems encountered by scientists wanting to analyse biological molecules in solution is evident from the statements made in various publications between 1940s to early 1970s. Following is a quote from a paper by Sutherland and co-workers [98]: For accurate analytical work using infrared spectra, the sample should be either in the gaseous or liquid state, although solids which can be obtained as thin plates or films of which the thicknesses can be accurately determined are acceptable. Solvents having no intense absorption over the spectral regions investigated are very scarce in infrared work, and none of these was suitable for this problem. Estimations were accordingly made on the solid material suspended as a paste in liquid paraffin (Nujol). The acetamido acids were finely ground in a mortar and well mixed with an approximately equal quantity of Nujol until a smooth homogeneous paste was obtained Another method of recording infrared spectra of molecules in the solid state, that did not involve suspending samples in Nujol or producing a thin film, was introduced in 1952 [99]. Stimson & ODonnell (1952) introduced the concept of recording infrared spectra in the solid state using KBr. Sister Miriam Michael Stimson and Sister Marie Joannes ODonnell were catholic nuns working at the The Research Laboratories of the Institutum Divi Thomae, Cincinnati, Ohio. These two ladies are the first to develop the KBr disk approach for recording infrared spectra in the solid state and reported their findings at a meeting in 1951 [cited in ref 99]. The authors wrote the following in their article: A method has been developed whereby the usual nujol mull employed for the study of solid organic compounds in the infrared region of the spectrum may be obviated. They chose two biologically relevant molecules for their analysis, namely cytosine and isocytosine. The compounds were mixed in KBr and transparent disks were produced after application of high pressure. This method of recording infrared spectra in the solid state gained increasing popularity and continues to be used today. It is interesting that the work of Stimson and ODonnell in analysing the infrared spectra of DNA bases, using KBr disks, played an important role in the structural elucidation of DNA bases and the double helix. It is important to also note that Scheidt from the Max-Planck Institute for Biochemistry in Tbingen Germany also independently contributed in the development of the KBr method for recording infrared spectra [100]. 3.2. Emergence of ATR Method of Recording Infrared Spectra In early 1960s, internal reflection spectroscopy (IRS) also knows as attenuated total reflection (ATR) spectroscopy came to the scene as a serious alternative to transmission measurements. Two people played a major role in the development of this technique. They are Fahrenfort [101] and Harrick [102], both of whom were working independently in the industrial sector during the late 1950s. Fahrenfort was working on organic compounds at the Royal Dutch Shell laboratories in Amsterdam, whereas Harrick was working on semiconductors at the Philips Laboratories in New York. Fahrenfort was first to report an article entitled Attenuated Total ReflectionA New Principle for the Production of Useful Infrared Reflection Spectra of Organic Compounds at a spectroscopy Meeting in Bologna in 1959 [101]. The technique was used for analysis of aqueous solutions in 1963 [103]. One of the first biomedical appli-
24
cations using the ATR method was also reported in 1964 when Scheuplein [104] investigated the quantitative determination of skin reflectance using ATR-infrared spectroscopy. Increasing number of scientists were attracted to this new method of recording infrared spectra. Parker and Ans [105] recorded infrared spectra of human and animal tissues and the authors noted the advantages of the technique in the following manner: Little preparation of the samples is necessary. Single-reflection spectra have shown the application of the method for examining tissues, both normal and diseased. The changes in tissue chemistry produced by the diseased state are evident. Multiple reflection has afforded more intense bands and increased the resolution of the spectra. The tissues examined were human fat, spleen, and aorta, and rat heart, chicken heart, and veal heart (endocardium and atrium). Currently, a wide array of biological and biomedical studies are being reported that utilise the ATR method for recording infrared spectra. The infrared spectroscopy community is divided into two main groups those who mainly use the ATR method and those who favour the transmission method. Below is a quote from a paper by Khurana and Fink [106] who justify their reasons for favouring the ATR technique over the transmission method (references removed): Hydrated thin-film ATR-FTIR was chosen over transmission FTIR because of its technical superiority. For example, the ATR mode has a much higher signal-to-noise ratio, data acquisition is much faster and easier, the samples can be in H2O rather than 2 H2O (2H2O may affect protein conformation in some cases), and analysis of the spectra is facilitated by the absence of liquid water. Thin-film ATR-FTIR spectra of proteins are comparable to those obtained by transmission FTIR, and the secondary structure analysis by both methods gives equivalent results. Proteins in the thin films are fully hydrated and hence would be assumed to exist in their native conformation. Others hold different views regarding this and below is a quote from Arbely et al. [107] who consider the transmission method to be easier then the ATR method: In general it is easier to undertake transmission measurements versus ATR studies. This is accompanied by the nontrivial theoretical considerations that one has to undertake when performing ATR measurements. As an example, one can note the debate that exists over the use of thin-film versus thickfilm approximations. The use of the ATR method for protein analysis has been strongly criticised by the group of Henry Mantsch at the National Research Council in Canada who were concerned that protein adsorption on the ATR crystal could lead to serious errors in the interpretation of protein infrared spectra [108]. However, others have suggested that, with adequate care, the ATR method can be highly effective in the analysis of proteins [109]. These authors have been actively engaged using the ATR method for analysis of biomolecules, especially membrane proteins for many years. The ATR method is certainly an attractive method for determining orientation of proteins in membrane although such analysis can be equally done using the transmission method [107,110]. There are few studies in the literature that have carried out a systematic analysis, comparing the transmission method and the ATR method for recording infrared spectra of biomolecules [e.g. 111]. Van Weert et al. [111] recorded FTIR spectra of five wellknown proteins, (bovine serum albumin, lysozyme, ribonuclease A, -lactalbumin, immunoglobulin G) using several different sampling methods. Transmission spectra were recorded for solid proteins in KBr pellets, protein solutions (in both 2H2O and H2O) in a liquid cell. ATR spectra of the proteins solutions (in both 2H2O and H2O) were also recorded. Results obtained showed that that environmental effects and physi-
25
cal nature of the sampling technique can significantly influence the infrared spectra of proteins. Hence, it was suggested that comparison of spectral data, obtained using different sampling methods, should be done with extreme care. Ideally, infrared spectra of a biomolecule should only be compared when they have been recorded using the same sampling technique. Differences in the spectrum of a protein recorded in solution and in the solid states may arise from changes in the phase of the sample rather than alteration in molecular structure. This fact has been recognised over sixty years ago when HW Thompson, working in the Physical Chemistry Laboratory at Oxford University, discussed how the infrared spectrum of a sample can alter due to a change in its physical state [112]. Thomson wrote the following in his article (reference removed): As pointed out in an earlier paper, the passage from one physical state to another will involve a change in both the potential energy function of the system and of selection rules, and with a long branched paraffin chain the frequencies and intensities of the bands may be expected to change. The ATR versus transmission debate has been ongoing in the literature for some time. It appears that some consensus has been reached on this issue and that both methods can give consistent results. The ATR method can provide some advantages such as stopped-flow measurements and for protein-ligand interactions as it provides the opportunity for one of the reactants to be immobilised on the surface of the ATR cell. 3.3. Infrared Microspectroscopy Barer et al. [113] were the first to apply infrared microspectroscopy for biological applications in 1949 and used a reflecting microscope. Regarding the use of infrared microspectroscopy for biological application it is appropriate to refer back to Barer [76], who noted that interpretation of spectra of biological tissues is not easy. He expressed this concern by firstly discussing about how difficult it is to interpret the spectrum of a single molecule let alone a complex mixture of macromolecules. Regarding this he states the following: the infra-red absorption spectrum gives what is essentially information concerning the presence or absence of certain specific chemical groups such as OH, CH, NH, CO, etc. In a complex molecule such as a protein, many such groups will be present and it may be wondered whether it is possible to distinguish different proteins by their infra-red spectra. This subject is still in its infancy. Noting the complexity of interpreting protein infrared spectra, Barer [76] goes on to state the following: If this is the case with single proteins, how much more complicated will be the position in the living cell in which we have a mixture of proteins, lipoids, nucleic acids and other substances? Any observations on the infra-red spectroscopy of cells or tissues will almost certainly be largely empirical until the fundamental data on these classes of substances have been obtained and interpreted. He [76] then went on to point out why the situation is even more difficult when carrying out infrared microspectroscopy compared to traditional transmission measurements. He was very much aware of the limited technology accessible to him and noted the following: The difficulties of examining such complex material are made even greater by using a microscope, for the amount of energy available is less and there is a consequent loss of spectral resolving power due to having to increase slit widths, and if the amplification of the small signals available is pushed to its limit the noise level and base-line instability may make quantitative measurements of transmission unreliable. If the rec-
26
ognition of different parts of the cell or different types of cells is to depend on small differences of frequency and height of absorption bands, some improvement in the performance of infra-red spectrophotometers, particularly in the detecting devices, is highly desirable, and until such improvements have taken place, the full value of the technique as applied to cytological material will not be realized. However, Barer [76] concludes on a more optimistic note: These conclusions may seem to be pessimistic but nevertheless there are a number of ways in which infra-red microspectrophotometry could be applied to cytological material. In the first place, we can attempt to alter the chemical constitution of the cell by various procedures. One method of approach is to extract materials from the cell and to follow the structural and spectroscopic changes which occur. Preliminary work on the extraction of nerve tissue with lipid solvents has already been reported and comparisons between the spectra from unmyelinated nerves from various species have been made with those from extracted myelinated nerves. Barers article [76] clearly points out the major difficulties that one encounters when using infrared spectroscopy for analysis of complex biological systems. The first commercial infrared microscope was produced in 1953 by Perkin-Elmer. Thanks to continued developments in technology, infrared miscrospectroscopy is a highly successful and productive area of research. Powerful FTIR spectrometers, with highly sensitive detectors, in conjunction with sophisticated computational and mathematical techniques, have made infrared microspectroscopy a formidable technique in the characterisation of complex biological systems. Bio-Rad Digilab Division, in Cambridge, MA was the first to produce FTIR spectrometers coupled to a microscope which entered the market in 1983. In one of the first articles on FTIR microspectroscopy, Krishnan [114] from Bio-Rad, described the accessory in the following manner: The accessory consists of an all-reflecting infrared microscope coupled with a high sensitivity, small area mercury-cadmium-telluride (MCT) detector. Since the mid-1980s nearly all the major infrared manufacturers provide necessary accessories for carrying out microspectroscopic measurements using their FTIR instruments. As a consequence, there has been a surge in the application of FTIR microspectroscopy for biological studies. Some of the latest applications are discussed by Naumann, Fabian and Lasch, in a later Chapter of this book. The need to use large quantities of samples for obtaining good quality infrared spectra has been a major hurdle for some biological applications due to the lack of availability of sufficient quantities of sample. Hence, the devolvement of FTIR microspectroscopy offered the possibility of overcoming this problem. Vogel and coworkers [115] took advantage of this new advance and published a paper entitled Downscaling Fourier Transform Infrared Spectroscopy to the Micrometer and Nanogram Scale: Secondary Structure of Serotonin and Acetylcholine Receptors. In this study they used FTIR microspectroscopy to analyse the receptors using micrometer-sized, fully hydrated protein films. They described the advantage of this method in the following manner: Because this novel procedure requires only nanogram quantities of membrane proteins, which is 45 orders of magnitude less than the amount of protein typically used for conventional FTIR spectroscopy, it opens the possibility to access the structure and dynamics of many important mammalian receptor proteins. We are currently witnessing a renaissance in infrared microspectroscopy with the use of the bright synchrotron radiation for recording spectra (see Chapter by Dumas,
27
Miller and co-workers). This is likely to make the technique amenable to the analysis of the most difficult biological samples at high resolution. 3.4. Development of Specialised Accessories for Recording Infrared Spectra of Biomolecules Accessories for recording infrared spectra of biomolecules have been developed by both industrial and academic laboratories. For example, diverse range of ATR accessories has been developed that are commercially available to users. ATR accessories can either be vertical or horizontal. The vertical accessories are restricted to samples in the solid state whereas the horizontal-ATR accessories can be used to measure samplers in both solution and solid states. One of the popular ATR devices known as the cylindrical infrared reflection cell for liquid evaluation has been widely used for many years. Subsequently, other devices have gained popularity including the Golden Gate ATR System. Accessories for carrying out measurements using the transmission mode have also been developed. Commercial cells designed to simplify analysis of biomolecules in H2O have been produced such as Confocheck (Bruker Optics) and BioCell (BioTools). These cells contain infrared transmitting windows with fixed pathlength (67 m) that are suitable for recording infrared spectra of molecules in H2O without having to worry about pathlength changes. Protein samples can be injected into the cells directly instead of having to take the infrared windows apart and then loading the sample and reassembling the cell. They are particularly useful for those who are new to analysis of biomolecules in H2O. Disadvantage of these devices is that they are more difficult to clean and often require a large volume of sample. Different types of cells designed for specialised applications includes the reactioninduced cells that allow certain chemical reaction to occur in situ, within the cell, without having to take the infrared cell windows apart. For example, transmission cells developed for recording spectra of a protein in the reduced and oxidised states to enable in situ monitoring of products and intermediates in biological redox processes. The availability of optically transparent electrodes was the important development that made possible the recording of infrared spectra directly through the electrode [116]. Mark & Pons (1966) [116] were one of the first to record infrared spectra of molecules at the electrode surface during electrolysis. Some two decades later, Mntele and coworkers extended this approach for biological studies [117]. They constructed an infrared spectroelectrochemical cell that enabled the coupling of biological electrochemical reactions with infrared spectroscopy. The cell that is suitable for studies of proteins in aqueous solution [118] and is particularly useful for infrared difference spectroscopic analysis of redox events (See Chapter by Barth). Fourier-transformed infrared photoacoustic spectroscopy (PAS) has been a more recent addition to the sampling techniques available to infrared spectroscopists. The method has been demonstrated to be useful for characterisation of biological samples in the solid state with the advantage that it involves no sample preparation [119]. More recently, surface enhanced infrared spectroscopy [120] has been developed that can be used to obtain spectra from very small amounts of sample. The principle of this technique is analogous to the better known surface enhanced Raman spectroscopy. This new approach can be used for analysis of protein monolayers.
28
Attempts to develop stopped-flow technology for investigating kinetics of biochemical reactions, using spectroscopic techniques in general, started in the 1950s [121]. Infrared spectroscopy has lagged behind UV, fluorescence and CD spectroscopy for stopped-flow studies of biological systems. There have been few infrared studies of biological systems which require rapid mixing of aqueous solutions. This is mainly due to the fact high pressure is required to flow a solution through the short path length infrared transmission cells (550 um) and the viscosity of the concentrated solution. Wharton and co-workers [122] were one of the first to attempt to produce a stopped-flow device for rapid mixing of solutions for analysis of enzymatic reactions using FTIR spectroscopy. The authors did their measurements in 2H2O and predicted further application of this approach for monitoring enzymatic reactions. Stopped-flow study using the ATR method is more widely used compared to the transmission mode [e.g. 123].
4. Spectral Acquisition & Processing Methods In order to obtain specific types of information on molecular structure using infrared spectroscopy, different methods have been developed for recording infrared spectra of biomolecules. Here, some examples of commonly used methods will be provided. 4.1. Polarised Infrared Spectroscopy One of the important modes of recording infrared spectra that have been extremely valuable for biological studies is polarised infrared spectroscopy. One of the first reports on use of polarised infrared spectroscopy for determining the orientation of polymers and proteins was reported in 1947 by Mann and Thomson [124] from Oxford University. Elliott and Ambrose [125] were some of the first scientists to use polarised infrared spectroscopy for monitoring protein folding. The work of Bruce Fraser [126] using polarised infrared spectroscopy is also important, especially with respect to polarised infrared spectroscopic analysis of DNA and fibrous proteins. The method was productively used to determine orientation of fibrous proteins [127,128] and lipids [129]. In 1966, Wallach and Zahler [130] attempted to determine the orientation of bacteriorhodospin in purple membranes using polarised infrared spectroscopy. However, the finding of this study was inconclusive and Rothschild and Clark in 1979 [110] carried out a more detailed polarised infrared study to determine the orientation of bacteriorhodoposin in purple membranes. Unlike the previous studies, this was [110] the first study to use a FT instrument. Rothschild and Clark used a dual beam FTIR spectrometer (Digilab FTS-14). In a recent communication with Rothschild, he made the following remarks about his early work (references removed): in collaboration with Wim DeGrip, we demonstrated that opsin, the form of rhodopsin which lacks the retinal chromophore, had very little beta-structure but is predominantly alpha-helical . An important goal was then to determine how these alpha-helices were arranged in the membrane. For this purpose, we worked with Noel Clark, who was on the faculty at Harvard University, to develop a technique for orienting photoreceptor and other membranes using an ultracentrifuge which we termed isopotential spin-drying . This method enabled us to demonstrate using polarized FTIR that both bacteriorhodopsin (BR) in purple membrane, the light-driven proton
29
pump, and rhodopsin in the photoreceptor membrane contained a bundle of alphahelices oriented predominantly perpendicular to the membrane plane. These early studies laid the foundation for subsequent application of polarised infrared spectroscopy to determine orientation of biological molecules. The determination of orientation of peptides and proteins in lipid membranes has been the most active and fruitful area of polarised infrared spectroscopy. Few techniques are readily amenable for structural determination of proteins in membranes and a survey of the literature clearly shows that many studies on membrane proteins had relied on infrared spectroscopy to give an idea about the orientation of the protein and its secondary structural elements with respect to the lipid bilayer. The latest advances in polarised infrared spectroscopy of biomolecules have been the effective use of isotopically labelled residues for determining the orientation of specific residues in peptides and proteins and probing if they undergo any conformational changes due to interaction with other molecules or alterations in the functional state of a protein [for a review see 131]. 4.2. Biological Infrared Difference Spectroscopy Infrared difference spectroscopy has been used to analyse biomolecules long before the advent of FT instruments. However, the problem due to absorbance of water is a recurrent theme in biological infrared applications that continued to be reported until the 1970s. Here is a statement from one study [132], published in the late 1960s, which highlights the authors attempt to overcome the problem of water absorbance and their hopes for the future: The application of infrared spectroscopy to the study of enzyme action has been severely limited by the very strong infrared absorption of aqueous protein solutions. However, all aqueous protein solutions have relatively high transmittance in the range, 2000 to 2800 cml. As technological progress continually brings us more intense light sources and better radiation detectors we have a better chance of seeing how enzymes work by looking through this infrared window. These authors measured the difference infrared spectrum of CO2-equilibrated bovine carbonic anhydrase against the ethoxzolamide- or azide-inhibited enzyme, They obtained a band at 2341 cml due to the antisymmetric stretching of the CO2 molecule bound to a hydrophobic surface at the active site of the enzyme. They were able to monitor this band in aqueous solution since H2O does not absorb in this region. Thus during the 1960s and early 1970s, before FT instruments became widely accessible, also biological infrared difference spectroscopy in aqueous media was restricted to the window regions where H2O does not absorb strongly. In early 1960s, infrared difference spectroscopy has been used to monitor changes protein structure [133]. For example, infrared difference spectra of protein samples recorded at low and high pD values were obtained which showed differences in the Amide I and Amide II bands. The spectra were recorded in 2H2O and the changes in both of these bands were attributed to exchange of hydrogens for deuteriums on the peptide amide groups. Proteins that have been most extensively studied by FTIR difference spectroscopy are those that are available in large quantities and even more importantly happen to contain a chromophore that can be switched from one state to the other without having to disturb the sample in the infrared cell. This approach of triggering a protein reaction directly in the infrared cell is called reaction-induced infrared difference spectroscopy.
30
The most common approach has been to use light to switch the protein from one functional state to another (see Chapter by Barth). Hence the most well studied proteins using FTIR spectroscopy are light driven proteins such as bacteriorhodopsin, rhodopsin and the photosynthetic reaction centres. Several groups have been working on these proteins for three to four decades and obtained valuable information that have played a pivotal role in our understanding of their mechanism of action. Thus for example, the 3D structures of bacteriorhodopsin and rhodopsin have given the overall molecular structure but information from infrared spectroscopic analysis of these proteins at different stages of its functional cycle have been vital for understanding the conformational changes of individual amino acid residues and the overall protein backbone. One of the infrared spectroscopists who contributed significantly in understanding the structure of bacteriorhodopsin and rhodopsin is Kenneth Rothschild who in a recent communication stated the following regarding his early infrared difference spectroscopic measurements (references removed): Around 1975 I realized that the increased sensitivity of FTIR spectroscopy might enable conformational changes of membrane proteins to be detected even down to the level of individual amino acid side chains. Bacteriorhodopsin and rhodopsin provided an ideal system to test this idea since they could be activated by light. Attempts by us in this direction began in 1976 but it wasnt until we acquired an MX-1 Nicolet FTIR Spectrometer in the summer of 1981 in my lab at Boston University that we were able to successfully detect for the first time small bands which we were able to assign using isotope labeling and resonance Raman spectroscopy to changes in individual molecular groups. Cytochrome c oxidase is another protein that has been extensively studied by FTIR spectroscopy due to the fact it can be obtained in large quantities. Furthermore, it can be converted to different states, bound to different ligands, and the structure of the ligands and the protein analysed simultaneously by infrared spectroscopy. Since 1960s Alben and Caughey [134] have been studying ligand binding to heme containing proteins and one of their earliest studies investigated carbon monoxide binding to human red blood cells as well as to isolated haemoglobin and the heme carbonyl [134]. They have contributed significantly to our understanding of several respiratory proteins and their interaction with different ligands. 4.3. Two-Dimensional Infrared Spectroscopy Unfortunately, FTIR has been lagging behind NMR spectroscopy with respect to development of methodologies for spectral recording that would enable greater simplification of the spectra for ease of band assignment and obtaining additional information. Historically, the development of two-dimensional infrared spectroscopy has been influenced by the much earlier work on two-dimensional NMR spectroscopy. The first report on two-dimensional infrared spectroscopy was published by Noda in 1986 [135] who was working at the Procter and Gamble Company in the USA. However, due to the specialised nature of the technique it did not gain wide-spread application until the 1990s when it became increasingly apparent that two-dimensional infrared spectroscopy can help unmask the overlapping signals that often congest infrared spectra of biological molecules. The efforts of Noda and Ozaki have been pivotal in raising the awareness of the potential offered by 2D-IR spectroscopy [e.g. 136].
31
Few years after the work of Noda (1986) [135], another important breakthrough in 2D-IR spectroscopy was reported when Tanimura and Mukamel [137] reported nonlinear optical 2D-IR spectroscopy based on ultrafast laser pulses. This approach is much similar to the concept used for obtaining 2D-NMR spectra which involves application of radiofrequency pulses. Due to the non-trivial nature of conducting 2D-IR spectroscopy using ultrafast pulses, only a few groups of workers are currently using this approach compared to the much widespread use of the generalised 2D correlation spectroscopy developed by Noda (1986) [135]. Nevertheless, several important studies based on pulse photon echo approach have been reported [137]. Applications include studies of structure and dynamics of peptides and fingerprinting of peptides and proteins [e.g. 138, 139 and Chapter by Ishikawa, Kim, Finkelstein and Fayer]. Although the development of 2D infrared spectroscopy is a major development, we are still a very long way before we can determine the complete 3D structure of a protein in a way that can be achieved using NMR spectroscopy or X-ray crystallography. However, the use of powerful computers to calculate vibrational modes of peptides and some small proteins is promising and in conjunction with 2D-IR techniques it may be possible one day to elucidate the complete 3D structure of a protein. 4.4. Time-Resolved Infrared Spectroscopy Before the advent of FT-instruments, time-resolved infrared measurements on biomolecules have been conducted using dispersive spectrometers. In the study of bacteriorhodopsin and rhodopsin this involved the use of flash photolysis in conjunction with infrared measurements at a specific wavelength. This approach provided submillisecond time resolution [140]. The disadvantage of this approach is the need for substantial signal averaging at a single wavelength in order to access a significant infrared spectral region. However, with FT instruments it is possible to simultaneously record a large infrared spectral region. Taking advantage of this, a method based on sweeping the interferometer moving mirror rapidlyenough to obtain a FTIR spectrum from 0 to 2000 cm1 with 8 cm1 resolution, within 5 milliseconds was first reported by Rothschild and co-workers in 1985 [141]. The step scan technique improved the time resolution of FTIR spectrometers to the sub-microsecond range. This technique records time-resolved intensity changes from a kinetic experiment at fixed mirror positions of the FTIR interferometer. The measurement is repeated for all mirror positions needed to construct time-resolved spectra of a desired spectral resolution. A requirement for the step scan technique is that the experiment can be accurately reproduced at least several hundred times, since kinetic traces at typically about 600 mirror positions have to be sampled at 4 cm 1 optical resolution [142]. However, the time resolution of the step scan technique was still not sufficient to access some of the very fast biological processes. Hochstrasser and co-workers have taken advantage of advances in laser technology to carry out biological studies on proteins using infrared spectroscopy at femtosecond resolution [143]. More recently ultrafast infrared spectroscopy is being used to probe protein folding of peptides and proteins. The Chapter by Ishikawa, Kim, Finkelstein and Fayer, later in this book, discusses advances in ultrafast infrared spectroscopy as applied to biological systems.
32
5. Application of Infrared Spectroscopy for Characterisation of Biological Systems A historical perspective of the application of infrared spectroscopy of biological molecules, cells, tissues and organisms is provided. Furthermore, obtaining quantitative information from infrared spectra of biological molecules is discussed. 5.1. Protein Studies Stair and Coblentz in 1935 [144] used infrared spectroscopy to characterise plant and animal tissues this included analysis of carbohydrates and proteins. Later, Klotz, Gruen and co-workers recorded infrared spectra of purified crystallised, proteins using a Beckman IR-2 instrument in late 1940s [145]. They were concerned that some of the earlier infrared studies were conducted on samples of questionable homogeneity and was restricted to a narrow region of the infrared spectrum. They produced films of the proteins and peptides on appropriate supporting plate, such as silver chloride, after evaporation in a vacuum desiccator. They analysed both the amide A, amide I, amide II as well as some side absorbance bands such as those arising from tyrosine. Besides soluble proteins, they analysed small peptides such gramicidin. Films of the soluble proteins were made by drying aqueous solution onto silver chloride films. Films of gramicidin were produced by dissolving the peptide in 95% ethanol and then producing a dry film on silver chloride plates. Dieter Gruen [145] is one of the early pioneers of biological infrared spectroscopy whose main area of scientific interest was disrupted by the war. After graduation, Gruen worked on the Manhattan Project at Oak Ridge, Tennessee helping to separate uranium isotopes using calutrons. After the war, he returned to Northwestern to do a Masters thesis with Irving Klotz before continuing with his Ph.D studies at the University of Chicago. In a personal communication, Dieter Gruen (currently Argonne Distinguished Fellow at Argonne National Laboratory) said the following about his early infrared studies of proteins, at Northwestern University, more than sixty years after the work was published: The chemistry department had just acquired a Beckman IR2 spectrometer and I was one of the first to make use of it. Most of the data were obtained by doing manual scans, an arduous procedure considering todays FTIR instrumentation. Those were exciting times since there was virtually no data of this kind available at the time and we were helping to lay the foundations on which future generations of chemists and biochemists were able to build. The enormous advances in instrumentation including IR have helped to make biology into the magnificent edifice it is today and we sometimes forget the role played by the pioneers in a field. In the UK, the pioneering work of Sutherland [e.g. 146], Elliott and Ambrose [e.g. 125,147] and Fraser [148] was important for laying the foundation of biomolecular infrared spectroscopy. The early work of Elliott and Ambrose in 1950s [125,147] was important in demonstrating that the two main types of protein secondary structure can be distinguished by infrared spectroscopy. The work was performed with proteins in the solid state and for the next 15 years there were not many studies that exploited this potential. The major advance came some 15 years later when Heino Susi measured proteins in 2H2O and analysed model polypeptide (poly-L-lysine) in different conformations [86]. Susis group persisted in using infrared spectroscopy of proteins between
33
1966 to 1986. They continued to publish papers that motivated others to enter the field although some had been using infrared spectroscopy for analysis of biomolecules long before Susi. As an example, Dennis Chapman, Henry Mantsch and others, who had been using infrared spectroscopy for analysis of lipids for many years [149], also started to work on proteins using infrared spectroscopy. Susis work on proteins in 2 H2O made the technique more attractive to biologists as it represented a significant step forward from looking at biomolecules in the dry solid state. However, measurement in 2H2O had its limitations, as mentioned earlier, since 1H-2H exchange of amide proteins can cause complications in the assignment of peaks [74,75]. A review article written by Hvidt and Nielson [150] discusses in detail the use of infrared spectroscopy for monitoring 1H-2H exchange in proteins, peptides and small molecules. During the 1960s stopped-flow methods have been used to monitor 1H-2H exchange in N-methylacetamide with half-lives of up to few seconds [150]. With modern FT-instruments it is now possible to carry out hydrogen-deuterium exchange at faster speeds and monitor complex macromolecular interactions [e.g. 151]. As one would expect mistakes are clearly evident in some of the early infrared studies of biological molecules. For example, it was not uncommon to see infrared papers on protein spectra that contained peaks from water vapour and some were erroneously assigned to protein structure. In due course, it became increasingly clear that purging the sample compartment with dry air or nitrogen was vital to reduce overlap of peaks from water vapour. Some suggested subtraction of water vapour spectrum from the protein spectrum. Others were hesitant to do this since under- or over-subtraction can result in either appearance or elimination of peaks in the final spectrum. Spectrometer manufacturers also contributed towards resolving the problem of water vapour overlap in protein studies. In the field of modern biological infrared spectroscopy spectra of only a single or few proteins are recorded and published, due to various difficulties. Therefore the field lacks the type of systematic study that Coblentz felt was necessary when he was analysing organic molecules in early 1900s. It is true that attempts are being made to systematically record the infrared spectra of a large number of proteins by various authors. Indeed, a suggestion has been made to produce an infrared protein spectra database similar to the Protein Data Bank that contains the X-ray & NMR structure of proteins. Unfortunately, the progress in this field is still slow due to the much smaller number of scientists engaged in protein infrared spectroscopy compared to NMR or X-ray crystallography. 5.2. Lipid Studies Some of the first studies on lipid like molecules using infrared spectroscopy were reported by Coblentz [43] and Lecomte [152]. Several workers reported studies employing infrared spectroscopy for analysis of fats during the late 1940s [153,154]. However, some were not satisfied by the quality of the early studies and this is evident from an article by OConnor et al. [155] who begin their paper article with the following sentences: Before many successful applications of infrared spectroscopy to fatty acid and vegetable oil chemistry can be made, extensive spectral data on a large number of pure reference compounds will be required. Heretofore the scant spectral data available on fatty acids, esters, and triglycerides have consisted either of measurements made over
34
Figure 10. Photograph of Dennis Chapman describing the structure of a phospholipid molecule to HRH the Prince of Wales (photograph taken few years before his death in 1999 [see 157]).
a very limited range of the infrared spectrum or have been obtained with compounds of undescribed purity. However, the authors also noted in the article some comprehensive studies [e.g. 154] which started to appear at the same time as they were working. This is described in the following manner (reference removed): After the work to be described in this communication was well under way, Shreve, Heether, Knight, and Swern presented infrared absorption data on a number of long chain saturated and mono-unsaturated fatty acids, methyl esters, and alcohols. Their data constitute the most complete study of the infrared spectral properties of fatty acids and esters which has yet been described. The number of studies on lipids, fats and oils using infrared spectroscopy were still rather limited. This is evident from a review published by Binkerd and Harwood in 1950 [156], who state the following regarding the lack of infrared studies on fats and oils: The literature contains many references to the use of infrared spectroscopy, but little is to be found in regard to its application to fat and oil chemistry. They concluded their review [156] by adding the following statements: In fattyacid chemistry, as in all organic chemistry, infrared spectroscopy has already become an indispensable tool. It is being applied to both theoretical and practical problems and has only begun to demonstrate its real value. Almost at the same time, as the above articles were published, several detailed infrared studies on lipids and lipoproteins were reported in the literature [e.g. 158,159]. Norman Jones based at the National Research Council in Canada made important contributions in the analysis of lipids using infrared spectroscopy [160]. From late 1950s, Dennis Chapman (see Fig. 10), based at Unilever in the UK, was one of the most active scientists applying infrared spectroscopy for analysis of lipids especially for monitoring lipid polymorphism. In one of his first publications [149], he wrote the following which indicates the limited number of studies on lipids and absence of studies on lipid polymorphism at that time (references removed): Whilst infrared spectra of some monoglycerides have been reported, none of this work was concerned with the various polymorphic forms, and the spectra given there are mainly solution spectra or those of the stable p-form. Up to the present, spectra of 2-monoglycerides have not been reported.
35
Chapman successfully demonstrated in this study [149] the usefulness of infrared spectroscopy for monitoring lipid polymorphism and concluded the paper in the following manner: In conclusion, the investigation has shown that infrared spectra can provide a good standard for observing and following polymorphic transitions. This method for studying polymorphic transitions has some advantages over the X-ray method, notably the speed of obtaining the spectra and the information from them about the bonding of the groups. As a supplement to the X-ray technique it is of obviously great potential value. Chapman continued his work on infrared studies of lipids well into 1990s. During late 1970s Henry Mantsch at the National Research Council in Canada published many papers on lipids and lipid-protein interactions mainly using FTIR instruments. He published his first infrared analysis of lipids using a dispersive instrument in 1978 [161]. In the same year he published a paper on the analysis of lipids using a FT instrument [69]. Infrared spectroscopy continues to be a powerful tool for analysis of lipids and interaction of lipids with diverse molecules [for a review see 162 and also see Chapter by Wolkers]. 5.3. Studies of Nucleic Acids and Carbohydrates The vast majority of infrared spectroscopic studies on biomolecules have focused on characterisation of proteins and lipids. Much fewer studies have been reported on nucleic acids and carbohydrates. Nevertheless, infrared spectroscopy offers a number of advantages for studying these latter macromolecules. The first studies on nucleic acids can be traced back to late 1940s and 1950s [62,164]. The spectra of the nucleic acids were recorded in the solid state. Blout and Fields [164] noted the following in their article regarding their inability to measure the nucleic acids in aqueous systems: It should be noted that because of the very slight solubility of the nucleic acids, nucleotides, and nucleosides in any but aqueous solvents and the rather strong absorption of infra-red by water in the region 2 to 15 except in very thin layers it is necessary to measure these materials in the solid state. We have used the following techniques: (a) casting of a concentrated aqueous solution on silver chloride disks, followed by removal of the water, leaving a continuous film; (b) evaporation of the material in high vacuum upon sodium chloride disks; (c) finely divided powders on sodium chloride disks; and (d) powders mulled into mineral oil. Few years later the situation changed and Blout in conjunction with Lenormant published infrared spectra of nucleic acids, proteins and even bacteria in aqueous media (H2O and 2H2O) [165]. Others who contributed in the early infrared studies of nucleic acids includes Fraser who was the first to study oriented films of DNA using infrared spectroscopy [126,166]. Subsequently, Sutherland, and Tsuboi in 1957 also studied DNA using polarised infrared spectroscopy [167]. In the latter study, infrared spectra of oriented films of sodium deoxyribonucleate using polarised radiation and under varying degrees of relative humidity. The authors also recorded spectra of films that have been deuterated by vapour-phase exchange with 2H2O. In the 1960s, Lord, Falk [168] and Thomas [169] made important contributions in the analysis of DNA using infrared spectroscopy. Thomas reported an infrared spectroscopic method [169] that can be applied to determine the fractions of Watson-Crick base pairs at a given temperature in any RNA containing the four common bases in known ratios.
36
In the 1970s a number of scientists have been highly active applying infrared spectroscopy to the study of DNA [170]. Tailandier and co-workers were one of the first to apply FTIR instruments for analysis of DNA [171]. The advent of FT-instruments led to a renaissance in infrared spectroscopy of nucleic acids [see for e.g. 172174]. With respect to carbohydrates, one of the first studies using infrared spectroscopy was reported in 1933 [54]. Work on this class of biomolecules continued after World War II [175]. Goulden studied interaction between proteins and carbohydrates in 1956 [176]. FTIR spectroscopy has been applied to the characterisation of diverse carbohydrates [see for e.g. reference 177]. 5.4. Infrared Spectroscopy of Cells, Tissues and Intact Organisms The appearance of commercial infrared instruments in late 1940s attracted many scientists to apply infrared spectroscopy to the analysis of complex systems ranging from the analysis of bacteria to human fingers. Some examples of the pioneering studies are described below. Hardy and Muschenheim in 1934 [178], used infrared spectroscopy to characterise human skin by recording emission, reflection and transmission spectra. For emission and reflectance measurements, spectra were recorded for human finger. The authors placed the subjects finger immediately in front of the spectrometer slit. Using this approach they also compared the reflectance spectra for subjects belong to two different racial groups. From the spectral analysis, the authors [178] concluded the following: It is also significant that the amount of reflection in the infra-red is about the same for negro skin as it is for white skin. For recording transmission spectra of human skin it is obviously not possible to use human fingers, therefore the authors used fragments of human skin from surgical amputations. The authors were careful to produce a very thin sample in order to reduce absorption from the skin and water. Regarding water absorbance, the authors noted the following: The small amount of water on the tissue during the time of measurement could not have represented a film of greater thickness than 0.05 mm. Such a thickness transmits the infra-red readily. The authors [178] were optimistic about the potential of infrared spectroscopy for analysis of human tissue and make the following statement in their conclusions: The infra-red transmission spectrum of skin has a characteristic fine structure which may prove to be of physiological interest. However, despite the optimism the problem of studying biological tissues containing water was evident from their work. Indeed, this was clearly shown by these authors in a later paper [179] where they state: The absorption spectrum of normally wet skin is essentially that of liquid water. Upon drying, other absorption bands not due to water become evident The study by Hardy and Muschenheim (1934) [178] is an example of one of the detailed infrared studies on human tissues reported in the literature although other brief studies have been reported earlier. For example, as far as back as 1927, infrared spectroscopy has been used for biomedical applications [180]. Cartwright in 1930 [181] reported infra-red transmission of the flesh. Pearson and Norris in 1933 [182], used infrared spectroscopy to analyse horny layers of human skin. When commercial instruments became available a large number of scientists were engaged in biomedical studies including characterisation of bacteria, viruses, human
37
tissues and studies on metabolic biochemistry. Taking advantage of the Perkin-Elmer 12A instrument Norman Jones [183] and colleagues wrote the following in their paper: It is to be expected that the technique will soon be used extensively in medical and biological research, since adequate instruments are now available commercially. In this paper is described the use of infra-red spectrometry as an analytical tool for the detection and characterization of small quantities of steroid metabolites in urinary fractions. To aid identification of the steroids in complex systems, the same group of authors published a paper in 1949 where they reported the use of deuterium labelled steroids in infrared studies metabolism [184]. This is one of the first examples of isotope-edited biological infrared spectroscopy. These series of studies are very good examples of collaboration between scientists from different disciplines, with academic and industrial backgrounds, coming together to solve a problem. Norman Jones was an infrared spectroscopist from the Chemistry department at the National Research Council in Canada. In contrast, Dobriner and Libermen were from the Sloan-Kettering Institute for Cancer Research in New York and were well known for their work on steroid metabolism in health and disease. Williams and Barnes were based at American Cyanamid Company and engaged in industrial applications of infrared spectroscopy. In the field of infrared spectroscopy the close co-operation between academia and industry played an important role in advancing the application of this technique to a diverse range of systems. No doubt the financial strength of the industrial partners played a useful role in obtaining access to the latest instrument and expensive chemicals which were often out of reach for many in University departments. Infrared spectra of tissues and cells have been recorded by a number of workers [e.g. 185187] and the technique has been applied to the study of viruses [188] and bacteria [e.g. 189] in the 1950s. The advent of FT-instruments and associated technology has led to a renaissance in the application of infrared spectroscopy for analysis of cells and tissues [for reviews see 19091 see also the Chapter by Naumann, Fabian and Lasch later in this book]. 5.5. Quantitative Information from Infrared Spectra of Biological Systems Edsall in 1938 showed the possibility of using Raman spectroscopy for distinguishing between different amino acids by monitoring differences in their vibrational frequencies [192]. This concept was subsequently extended to the analysis of amino acid mixtures by Buswell and Gore (1942) [193] and also by Sutherland and co-workers in 1948 [98]. Buswell and Gore were one of the first to attempt to obtain quantitative information on proteins using infrared spectroscopy [193]. Their aim was to see if it was possible to separate the bands from the different amino acids and determine their extinction coefficients which will enable them to estimate the content of specific amino acids. They carried out analysis on salmine which is a water soluble protein. The authors noted that they were not able to record the spectra of the protein in the dissolved state since the protein was not sufficiently soluble in infrared-transparent solvents. Therefore, they recorded the spectra by producing solid films on thin microscope cover glasses [193]. Some seven years later, Sutherland and co-workers [98] working in Cambridge tried to estimate the isoleucine/leucine ratio for protein hydrolysates as well as for control samples. These pioneering studies were some of the first attempts towards obtaining quantitative information from protein infrared spectra. The first quanti-
38
fication of protein secondary structure content from infrared spectra can be attributed to the work of led by Susi and co-workers in the 1980s [92]. Below is a quote (with references omitted) from Michael Byler regarding their early attempts at obtaining quantitative information on protein secondary structure: Because of his experience in working with Fourier deconvolution of infrared spectra of a variety of substances, we sought the collaboration of Professor Peter R. Griffiths, then at the University of California at Riverside. He kindly agreed. Griffiths and his graduate student W.-J. Yang wished to focus on the deconvolution of diffuse reflectance spectra of solid proteins; we continued our work with D2O solutions. Because deconvolution requires a user to make subjective choices regarding the two parameters to be employed to obtain optimum band narrowing (band shape, band width, and the resolution-enhancement factor K) many spectroscopists had serious misgivings about how trustworthy any data obtained by this technique would ultimately prove to be. Our joint investigation indicated that spectra measured and manipulated independently at two different laboratories on samples of different origin [gave] similar results. Nevertheless, care must be taken to ensure that band intensities in the original are not saturated or distorted. In addition, because in principle deconvolution does not alter the area under a peak, we felt that this new method should enable quantitative estimates of the proportion of each conformation in a protein to be calculated. Byler and Susi pursued a more detailed study on secondary structure quantification from protein infrared spectra which was published in 1986 [92]. This can be considered as one of the land mark papers in modern protein FTIR spectroscopy in relation to protein secondary structure analysis. In this study, the authors estimated the secondary structure of 21 proteins [92]. They deconvolved the spectra of the proteins, that were recorded in 2H2O, and used curve-fitting method to estimate the content of helical and beta-structure. The accuracy of the prediction was very good and this encouraged others to use the approach for their studies. It is not surprising that the paper has been cited nearly 1,000 times and is probably the most highly cited biological FTIR spectroscopy paper in the scientific literature. As discussed earlier in this Chapter, after this study by Susi and Byler, a number of authors have shown that quantification can also be done for proteins in H2O [93,94,96,97]. Some quantitative studies on lipids have also been reported in the 1950s. For example, Freeman et al. (1953) [158] reported how the intensity of the lipid ester carbonyl band and certain strong bands of proteins can be used to obtain quantitative information on the lipid-protein ratio in lipoproteins. With respect to carbohydrates, the potential of using infrared spectroscopy for quantitative determination of nitrate groups in nitrocellulose has been reported by Kuhn in 1950 [175]. Infrared spectroscopy based methods have been developed for quantitative analysis of proteins adsorbed on surfaces such as biomaterials. In one early study, ATR spectra of surface-adsorbed proteins were correlated with measurements determined by 125Ilabeled proteins. The authors demonstrated a linear correlations between the intensity of the major infrared bands of proteins and the quantity of proteins [194] Early attempts have also been made at obtaining quantitative information on analysis of bacteria using infrared spectroscopy [e.g. 187]. Currently, quantitative analysis of changes in complex systems such as cells and tissues are also being developed. Sophisticated statistical and computational tools are being developed to distinguish between normal and diseased tissues, identification & classification of bacteria etc. [e.g. 190,191 and Chapter by Naumann, Fabian and Lasch]. Quantitative information
39
on the strength of molecular interactions and of covalent bonds can also be obtained and is discussed in the Chapter by Barth. 6. Impact of Advances in Chemical & Molecular Biological Methods for Biological Infrared Spectroscopy Developments in synthetic chemistry have played major roles in extending the potential of infrared spectroscopy. The effect of amino acid sequence on protein structure and stability using infrared spectroscopy has been made amenable due to the possibility to readily synthesise peptides using automated peptide synthesisers. Woodward and Schramm in 1947 were the first to successfully synthesise peptides [195]. It is interesting that the peptides synthesised by the latter workers were subjected to infrared spectroscopic analysis by Sutherland in the same year [146]. From a historical context, this is probably the first study to be reported in the literature that analysed the structure of a synthetic polypeptide and also make comparison with a naturally occurring protein. The potential offered by peptide synthesis for structure characterisation and interpretation of infrared spectral data is nicely summarised by Sutherland and co-workers [196] in the following manner (references removed): The recent synthesis of polypeptides from given amino-acids by Woodward and Schramm, following the neglected Leuchs polymerization, has reopened a promising line of attack on the problem of protein structure. Comparison of the properties of such synthetic polypeptides of known composition with those of proteins built of the same amino-acids should prove of great value, and in particular should help considerably in the interpretation of X-ray and infra-red data, which are still so imperfectly understood owing to the complexity of even the simplest protein. The peptide synthesis work developed by Woodward and Schramm and the subsequent spectroscopic work pioneered by Sutherland in late 1940s has now developed into a highly productive area of research [197]. In recent years synthetic peptides have been extensively investigated using infrared spectroscopy, especially in the study of complex proteins such, amyloids, prions and membrane proteins. Isotopically labelled molecules have been used in biological infrared spectroscopy for many years. Use of 2H2O, instead of H2O, to obtain a window in the amide I region and for monitoring 1H-2H exchange of amide proteins are the most well known application of isotope-edited biological infrared spectroscopy dating back to late 1930s. One of the first studies recording infrared spectra of a mixture of H 2O and 2H2O was reported in 1934 [e.g. 198] In lipid work, deuterated and 13C labelled lipids have been used for probing specific regions of lipid membranes since the 1970s, especially using NMR spectroscopy [199]. However, the use of labelled peptides and proteins have been more recent and became possible after availability of isotopically labelled amino acids for peptide synthesis. It is for this reason that there have been very few studies with isotopically labelled peptides and proteins until the early 1990s other than studies of proteins in 2H2O to monitor 1H-2H exchange. Advances in synthetic peptide chemistry, development of automated peptide synthesisers, recombinant DNA technology and bacterial expression of proteins all have been instrumental in making it possible to obtain isotopically, labelled peptides and proteins. The first FTIR study on 13C labelled peptides was reported back in 1991 [200]. Far fewer studies have been possible with isotopically labelled proteins. The first study investigating protein-protein interac-
40
tion, where one of the proteins is fully labelled using 13C and 15N was reported in 1992 [201]. Isotopically labelled proteins provide immense potential for interpretation of infrared spectra and also for obtaining detailed information at the single residue level. Unfortunately, due to the expense associated with producing labelled proteins the number of infrared studies employing labelled proteins continues to be rather limited. Nevertheless, the method is being successfully employed for characterisation of a number of complex systems such as membrane proteins and insoluble aggregates of proteins. In one study polarised infrared spectroscopy was used to determine the local amide orientation in ordered insoluble protein. The work involved labeling of individual amide carbonyl carbons with 13C which enabled the systematic assignment of amide I modes of specific amino acid residues [202]. The other major help in advancing biological applications of infrared spectroscopy has been the ability to carry out site-directed mutagenesis studies on proteins using recombinant DNA technology. This ability has been instrumental in our understanding of structure-function relationship of several proteins. A good example, of highly successful research in this area is the large number of studies on bacteriorhodopsin. Other advances that have been helpful for biological infrared spectroscopy are the synthesis of photolabile caged compounds that releases a specific ion (for example Calcium) or molecule (e.g. ATP) in response to light trigger, enabling the spectra of a protein to be recorded in more than one state so that a difference spectrum can be obtained (see Chapter by Barth).
7. Theoretical Analysis of Protein Infrared Spectra Theoretical analysis of infrared spectra has been vital for understanding the relationship between molecular structure and infrared spectra. This approach generally involved calculating spectral frequencies of a molecule, based on its molecular structure, and correlating this to experimental data. Although this approach has been successful with small molecules, it has been more challenging with complex molecules such as peptides, proteins, nucleic acids etc. Herzberg (see Fig. 11) was one of the first to use theoretical methods for analysis of infrared vibrations [203]. He was awarded a Nobel Prize for Chemistry in 1971 and had a profound effect on a future generation of infrared spectroscopists, especially those at the National Research Council, Ottawa, Canada, including Henry Mantsch (see Fig. 11). In a recent communication with Henry Mantsch, he stated the following regarding the views of Herzberg, compared to others, regarding the application of infrared spectroscopy for analysis of biological systems: Until the end of his long life Herzberg was often wondering how far we had taken his molecular spectroscopy, unlike some of his purist colleagues who accused me of having sacrificed infrared spectroscopy on the altar of biology. Scientists who pioneered the theoretical analysis of infrared spectra of complex polymers, with relevance to biological systems, include Sutherland [204], Miyazawa [205] and Krimm [206]. Miyazawas work on normal vibrations of N-methylacetamide have been important for understanding the infrared spectra of peptides and proteins [207]. In an article published in 1950 [204], Sutherland summarised the key problems encountered in theoretical analysis of large molecules:
41
Figure 11. Photograph of Henry Mantsch discussing energy levels with Herzberg on an old fashioned blackboard. Photograph courtesy of Henry Mantsch.
During the past ten years or so infra-red spectroscopy has changed from a branch of molecular physics dealing with the structural features of the smaller, and generally inorganic, molecules to a subject of considerable interest to the organic chemist and the biochemist, and of growing interest to the biologist. Whereas it is possible to give a complete analysis of the spectrum of the ammonia molecule, from which the bond lengths and angles can be deduced with an accuracy of 1 part in 1000 and the height of the potential barrier inhibiting inversion is known to within a few per cent, the most that can be stated at present about a large organic molecule or polymer is that it does or does not contain certain chemical groups, and even this statement has frequently to be hedged about with qualifications. He concluded his article [204] by identifying some key areas where much work needs to be carried out: (i) Intermolecular forces (including hydrogen bonding) encountered in solution, in the liquid state and in solids, both amorphous and crystalline. (ii) Weak intramolecular forces such as occur in internal hydrogen bonding and in restricted rotation about a single bond. (iii) Changes of charge distribution in characteristic groups caused by the presence of strongly electronegative or electropositive groups in certain positions relative to the group under consideration. (iv) Changes in group force constants due to changes in bond hybridization, arising from different environments of a particular group in different molecules. The effects of all these factors on the intensities as well as on the positions of characteristic frequencies must be studied. Finally the need for more theoretical work on the determination of the force constants governing the frequencies of characteristic groups must be stressed. In the last analysis it is the force constants and not the frequencies which are the fundamental physical constants of groups of atoms in large molecules.
42
Figure 12. Photograph showing Yuri Chirgadze with visiting guests in a laboratory at the Institute of Protein Research during the IV International Congress of Biophysics, Moscow, 1972. From left to right: post graduate student of Harvard Medical School Ms B. Doyle and Prof. E. Blout (Boston, USA), Dr. Y. Chirgadze, wife of Prof. E. Blout, and Prof. S. Krimm (USA). Photograph courtesy of Yuri Chirgadze.
These recommendations of Sutherland [204] are as important today as they were nearly sixty years ago when he wrote the above statements. Samuel Krimm (see Fig. 12) who continues to be active in the field to this day, worked with Sutherland at Michigan on theoretical analysis of infrared spectra of high polymers since the early 1950s. In a recent communication with him, he summarised some of his key contributions in this field (references removed): My own efforts in biological IR spectroscopy began during my spectroscopic studies on synthetic polymers, initiated during a postdoc with Gordon Sutherland at Michigan starting in 1950. My early efforts to relate spectra to conformation dealt with extensions of the Miyazawa and Blout interaction treatment of amide modes. I also presented spectral evidence for C-HO hydrogen bonding in polyglycine II. But I then began to feel that the only solid way to correlate spectra and structure was through normal mode calculations, and since this required a good force field we began to prepare the way with an analysis of a model system, N-methylacetamide, choosing to develop a valence force field rather than the Urey-Bradley field being used by Shimanouchi and collaborators. From this emerged our first polypeptide calculations, the concept of transition dipole coupling to account for amide I splittings, and the beginning of a series of some 50 detailed papers on peptides and polypeptides, part of which is summarized in the review article in Adv. Protein Chem. that tried to emphasize the power of vibrational spectroscopy in studying protein structure. Despite my recent work on developing spectroscopically accurate molecular mechanics energy functions, I am very pleased that we have come up with a new idea for using spectra to determine conformation: the sensitivity of the C(alpha)-D(alpha) stretch frequency to the phi,psi angles at the C(alpha) atom, which led to confirmation of the presence of CHO(water) hydrogen bonding in aqueous solution.
43
Advances in computational approaches for calculating vibrational spectra of biomolecules will not only aid a better understanding of spectra-structure relationship but will also enable the determination of macromolecular 3D structures at high resolutions. Chapters by Kubelka, Bour and Keiderling and by Choi and Cho in this book discuss the latest advances in the theoretical analysis of protein spectra.
8. Growth of Biological Infrared Spectroscopy 8.1. Biological Infrared Spectroscopy from an International Perspective This section provides a brief historical context of key developments in the application and spread of biological infrared spectroscopy at the international level. The main countries that were involved in laying the foundation of biological infrared spectroscopy were from the USA, UK, France and Japan with major contributions occurring between 1940s through to 1960s. Some of the contributions made by scientists from these countries have been highlighted earlier in this Chapter. From 1960s other countries also entered the field including Germany and USSR, and scientists from these countries made significant contributions in the development of biological infrared spectroscopy. One of the key pioneers in protein infrared spectroscopy from the USSR was Yuri Chirgadze who had a significant impact in infrared studies of proteins and published his first paper in this area in 1961 [208]. In a recent communication with him, he stated the following regarding how he first started applying infrared spectroscopy for biological studies (references removed): Last half of 1958 I was a graduate student and started to work in the laboratory of Dr Natalia S. Andreeva, who was a pioneer in solving X-ray crystal structure of small biological molecules (dipeptides), and then she was the first Soviet scientist who began to solve X-ray structure of a globular protein, pepsin. That time she offered me to apply IR spectroscopy to analyze the structure of model peptides related to structural genesis of a collagen molecule, i.e. oligopeptides containing imino acids proline and hydroxiproline. This was a theme of my University Diploma Thesis. As far as I know, it was the first attempt to apply method of IR spectroscopy for studying peptides in the USSR. A result of this work was published in my first scientific paper in 1961 in Russian journals: Physical Chemistry and Biophysics. The instrumental apparatus for this work was a single beam prism IR spectrometer IKS-1 manufactured serially by the optical and mechanical corporation LOMO in Leningrad. Figure 12 shows a photograph of Chirgadze with some visiting scientists from the USA including leading pioneers in infrared spectroscopy of proteins, Krimm and Blout. Chirgadze also commented on how the support of science in former USSR, the ability to purchase foreign instruments, and interactions with scientists from other countries played a valuable role in their research activities: infrared spectroscopy in the Soviet Union was intensively developed both experimentally and theoretically. At that time the biological science was strongly supported by the government. And the striking example of this was the foundation of a number of specific small scientific towns around Moscow, Leningrad (today SaintPetersburg), Novosibirsk and other cities. During all that time, we had an essential financial support for purchasing modern commercial spectrophotometers and other special equipment from the Soviet Union, Germany (DDR), USA, Japan etc. It is
44
very important that we could present our results at many conferences and congresses in the Soviet Union and abroad. I was personally introduced to famous scientists in the field Prof L. Pauling (USA), Prof. J. Watson and Prof. F. Crick (USA), Prof. E. Blout (USA), when they visited the Soviet Union and Prof. T. Miyazawa (Japan) when I visited shortly Osaka University in 1970. As one would expect, the use of Fourier transform instruments was largely determined by the accessibility and affordability of these spectrometers. It is not a coincidence that the first use of FT instruments by the academic community involved close interaction with instrument manufacturers. The manufactures were keen to sell their machines by highlighting the power of the new FT instruments and in this they found a natural ally in research scientists who were keen to analyse their difficult samples using these modern machines. As already mentioned, the first commercial FT instruments were produced by Digilab. Hence, most of the first FTIR studies on biological molecules were conducted using Digilab machines. Furthermore, the company being based at Boston (USA), meant that the first infrared studies on biological systems using FT instruments were carried out by scientists based in the USA. For example, Alben, who published the first FTIR paper on a protein molecule, used a Digilab FTS14 instrument (See Fig. 8). This is also the case for Rothschild who did the first FTIR study on a membrane protein used a Digilab instrument. After the scientists from USA, it was the Canadians who were quick to use FT instruments for biological studies. The European scientific community entered the field infrared spectroscopy of biological systems using FT instruments, some 68 years after the North Americans. The first biological application is reported by Siebert, Mntele and Kreutz in 1980 [209] from the University of Freiburg in West Germany . The biological FTIR spectroscopy community based in Freiburg in Germany were the leaders in Europe at that time. In the UK, Chapman and Belton were one of the first to use FT instruments for analysis of biological molecules [75,210,211]. Although, papers dealing with infrared studies on biological systems were being published by scientists in the former USSR, there are no reports on the use of FT instruments for biological studies until well after the end of communism. One of the first applications of FTIR instruments in Japan, for biological studies, was reported in 1978 [212]. The authors of that study used FTIR spectroscopy to analyse plasma proteins adsorbed on polymer surfaces. Japan continues to be the leading country in Asia in the field of biological spectroscopy. Indeed, Japanese scientists have played significant role in advancing the theoretical and practical aspects of biological infrared spectroscopy since late 1940s. Amongst the Asian countries, biological infrared spectroscopy has become more common in China. 8.2. Growth of Conferences Focusing on Biological Infrared Spectroscopy The late 1980s and the 1990s was a particularly exciting time to be involved in infrared spectroscopy of biomolecules. Several international conferences started in the 1980s which became the venue for dialogue, debate and dissemination of the latest advances in the field. One such meeting that attracted many infrared spectroscopists, engaged in biological studies, was the European Conference on the Spectroscopy of Biological Molecules (ECSBM). The first ECSBM was held in Reims in 1985 and the latest one (the XIII in the series) was held in Paris in 2007. Selected contributions from the last conference were published in Spectroscopy Biomedical Applications. Cur-
45
rently, there are many other conferences that have proliferated and some focus exclusively on infrared spectroscopy. 8.3. The Infrared Spectroscopy Literature The fact that infrared spectroscopy had already entered the realm of life sciences is clearly evident from the number of publications that were appearing between 1930s to 1950s in journals devoted to advances in subjects such as biochemistry, physiology, microbiology etc. A significant number of early biological infrared papers were published in journals such as Proceedings of the Royal Society of London, Discussions of the Faraday Society, Biochemical Journal, Nature, Science, Applied Spectroscopy, Journal of Biological Chemistry, Bulletin of the Chemical Society of Japan, Journal of the American Oil Chemists Society, Journal of Bacteriology, The Journal of Chemical Physics. At present, many of the biological applications of infrared spectroscopy were published in journals such as Biochemistry, Biochimica et biophysica acta, Applied Spectroscopy, Biophysical Journal, European Journal of Biochemistry, FEBS Letters etc. With further growth of biological infrared spectroscopy, specialised journals devoted to this area started to appear. For example, Vibrational Spectroscopy appeared in 1990, followed by Biospectroscopy in 1995. Existing journals were also changing their focus to meet the growth in biological applications and keep pace with the changing times. Thus for example, one of the authors (PIH) of this article took over the editor-inchief role of Spectroscopy An International Journal and changed its focus towards publication of biological and biomedical applications. Continuing with this development, the name of this journal has been recently modified to Spectroscopy Biomedical Applications. Modern biological infrared spectroscopy is still a relatively young and highly specialised field. Nevertheless, biological infrared spectroscopy publications are some of the highly cited papers in the literature. The most cited infrared spectroscopy paper dealing with analysis of a biological molecule is related to the Nobel Prize winning work by Prusiner on the prion protein [213], although this paper is not exclusively on infrared spectroscopy. The most highly cited paper, dealing exclusively on biological infrared spectroscopy, is the Byler and Susi paper [92] where they estimated the secondary structure of 21 proteins from their deconvolved spectra. There are two other papers in the literature, exclusively on biological infrared spectroscopy, that are highly cited. Whilst being an excellent source of information for the infrared spectroscopist, both of these papers are critical about the misuse of infrared spectroscopy. The most cited (570) is by Surewicz, Mantsch and Chapman [214] and deals with the need for caution in protein secondary structure analysis by FTIR spectroscopy. The second most cited (525), published two years later by Mantsch and Jackson [215], is a more detailed article discussing the use and misuse of infrared spectroscopy in the determination of protein structure. The citation of the papers by Mantschs group is far greater than the citations of their other FTIR papers with the closest one being a review published earlier [216] which focused on the potentials offered by infrared spectroscopy for biological studies (over 470 citations). The critical assessment articles by Mantsch, Chapman and co-workers were particularly important at a time when increasing number of people were entering the field and some of the new and unwary users needed guidance regarding not only potentials, but also about problems and pitfalls that one should be aware of.
46
9. Further Reading The occupation with the history of infrared spectroscopy is truly fascinating and it is easy to submerge in past arguments and the ingenious experiments of our scientific ancestors. For those interested in this experience, some of the original articles from the 19th century are available via ABs web site (http://www.dbb.su.se/Faculty/ Andreas_Barth). Besides the cited original literature, this overview relies on previous compilations. We found particularly valuable refs. [4,8,10,19]. The book chapter by R.N. Jones has been republished in a series of three articles [217219]. Accounts of the work by Herschel [6], Melloni [15,25], Langley [26] and Coblentz [32,220] are also available. Jones labels the time period from 1960 to 1985 the age of the acronym. Acronyms are still used extensively and Jones criticism well worth consideration: extreme specialisation has encouraged the use of new jargons which make it increasingly difficult for analytical spectroscopists to communicate effectively across these selferected barriers. The custom has also developed to identify these narrow fields by acronyms or lettered abbreviations, often to the extent of using them exclusively and dropping the descriptive name, even in the titles of publications [8]. He demonstrates his point by listing 64 acronyms, some of which have several meanings (e.g. MIR stands for multiple internal reflection or mid infrared) and others are confusingly similar (e.g. FT-IR for Fourier transform infrared and FTIR for frustrated total internal reflection). Infrared spectroscopic work on biological systems has been reviewed already in 1940 by Loofbourow [221] in an article titled borderland problems in biology and physics. Interestingly, the author decided not to use the title biophysics instead, because this term did not self-evidently include the use of physical methods at that time. This has changed and the expression biophysics nowadays means physical methods and physical principles applied to biology and biochemistry as suggested by Loofbourow [221]. Early work on proteins and their constituents has been reviewed in 1952 by Sutherland [222]. The history of biological applications has been summarised by During and Gerson in 1979 [223], that of lipids by Mantsch in 1998 [46] and that of near-infrared spectroscopy by McClure in 2003 [21].
10. Future Directions H.A. Laitinen, wrote an editorial in 1973 in Analytical Chemistry [224] where he made an analogy between Shakespeares seven ages of man and developments in analytical techniques. Using infrared spectroscopy as an example, he stated how this technique had reached its final (seventh age) stage (see preface). This judgement is far from the reality as is evident from some of the very recent advances in technology that enables the use of infrared spectroscopy to probe complex biological systems at a single residue level that was previously unimaginable. Future advances will not be restricted to new technological improvements in instrumentation and associated accessories but advances in chemistry and molecular biology will be harnessed to enable more challenging problems to be addressed. This will include site-specific isotopic labelling of proteins and peptides for detailed structure elucidation and macromolecular interactions. There are exciting advances in synchrotron radiation infrared spectroscopy which is likely to herald a new renaissance in bio-
47
logical infrared spectroscopy. In conjunction with computational and statistical tools, infrared spectroscopy will extend its power in the detection, identification, quantification and classification of molecular changes in complex systems and environments at high resolution. It will enable highly detailed fingerprinting of diverse chemical components in complex systems such as cells, tissues and whole organisms. Infrared spectroscopy is likely to play a central role in the rapidly emerging fields of systems biology, metabolomics, proteomics etc. It may not be too long before we see infrared spectroscopy as a diagnostic tool from clinics to bedside. Finally, we would like to end our article by quoting R.D.B. Fraser (See Fig. 6) who is one of the early pioneers in biological infrared spectroscopy but who also used X-ray diffraction. Fraser made a significant contribution in the elucidation of DNA structure whilst he was based at Kings College, London which is now widely recognised [225]. One of us (PIH) asked him if he considered infrared spectroscopy as a second-fiddle to X-ray diffraction. His response to the question is given below: Infrared spectroscopy has always been a powerful tool in the analysis of materials of unknown structure and was used extensively in the study of synthetic polypeptides and fibrous proteins to asses the proportions of alpha helix, beta structure and coil. Ambrose and Elliott were the pioneers in this application and also in the use of polarized radiation to determine the orientation of the chain axes in the alpha and beta sections. About the same time I found, using a high resolution spectrometer that Bill Price and I had developed, that the NH stretching frequency in collagen was about 30 wave numbers higher than in other fibrous proteins confirming that the conformation of the polypeptide chain was substantially different from that of the alpha or beta forms. At that time the precise conformation of the polypeptide chains in any protein were unknown and infrared studies were a valuable tool in the construction of plausible models. Remember that it is still only possible to obtain electron density maps at atomic resolution from X-ray diffraction patterns for crystalline proteins. With the fibre-type X-ray patterns obtained from fibrous proteins such as muscle, tendon and hair trial and error methods still have to be used. The search for the structure of feather keratin started in 1932 with the observation by Astbury and Street that the X-ray pattern resembled that of the stretched wool (beta) rather than wool (alpha) and evidence from polarized ir studies played a vital part in the development of a model that accounted for the broad features of the X-ray diffraction pattern. Even today our latest model still relies on the IR study that showed that feather keratin is based on the anti parallel chain pleated sheet. So I would say complimentary rather than second-fiddle.
11. Acknowledgements and Apologia Due to lack of time and space limitations, it has not been possible to cover the work of all the scientists who have made important contributions in the field of biological infrared spectroscopy. We consider this article as work in progress and hope to publish more in this area in the future and include key contributions that we may have omitted. In this context, we welcome any advice, comments, information and corrections from our readers. In order to avoid confusion, we have deleted references within quotes from various scientists and sections taken from the published literature. AB would like to thank the staff of the Chemistry library of Stockholm University for their professional help with the endless requests for external loans. PIH would like
48
to thank RDB Fraser, Dieter Gruen, Juana Bellanato, Samuel Krimm, James Alben, Jack Koenig, Henry Mantsch, Yu Chrigadze, Michael D. Byler, Kenneth Rothschild, Juan Gomez-Fernandez and James Mattson for taking their valuable time to answer many questions about their early contributions in biological infrared spectroscopy and some cases provide photographs etc. We would also like to thank Pat Ashton, Senior Communications Specialist at the Heritage Exhibit for providing the photograph of the Beckman IR-1 spectrometer. References
[1] [2] [3] [4] [5] [6] [7] [8] W. Herschel, Philos. Trans. Roy. Soc. London 90 (1800), 255-283. W. Herschel, Philos. Trans. Roy. Soc. London 90 (1800), 293-326. W. Herschel, Philos. Trans. Roy. Soc. London 90 (1800), 284-292. Y.M. Rabkin, Isis 78 (1987), 31-54. Hutchinson dictionary of scientific biography. Helicon Publishing, Abingdon, 2005. E.S. Barr, Infrared Phys. Technol. 1 (1961), 1-4. Encyclopaedia Britannica Online Version. R.N. Jones, Analytical applications of vibrational spectroscopy a historical review, in: Chemical, biological and industrial applications of infrared spectroscopy, ed. J.R. During, John Wiley & Sons, Chichester, 1985, 1-50. A.M.C. Davies, Spectrosc. Eur. 12 (2000), 10-16. E.S. Barr, Am. J. Phys. 28 (1960), 42-54. W. Herschel, Philos. Trans. Roy. Soc. London 90 (1800), 437-538. H. Chang and S. Leonelli, Stud. Hist. Phil. Sci. 36 (2005), 477-508. T. Chester, Reconciling the Herschel experiment, http://home.znet.com/schester/calculations/herschel/ index.html. D. Purves, S.M. Williams, S. Nundy, and R.B. Lotto, Physiol. Rev. 111 (2004), 142-158. K. Hentschel, NTM 13 (2005), 216-237. V.Z. Williams, Rev. Sci. Instrum. 19 (1948), 135-178. H. Chang and S. Leonelli, Stud. Hist. Phil. Sci. 36 (2005), 686-705. E.S. Barr, Phys. Teach. 5 (1967), 53-60 (reprint of Appl. Opt. 2 (1963), 639). R.A. Smith, F.E. Jones, and R.P. Chasmar, The detection and measurement of infra-red radiation. Clarendon Press, Oxford, 1968. H.M. Randall, Rev. Mod. Phys. 10 (1938), 72-85. W.F. McClure, J. Near Infrared Spectrosc. 11 (2003), 487-518. J.F.W. Herschel, Philos. Trans. Roy. Soc. London 130 (1840), 1-59. C.A. Gueymard, D. Myers, and K. Emery, Sol. Energy 73 (2002), 443-467. L. Nobili, Ann. Phys. Chem. 20 (Vol. 96 of the whole series) (1830), 245-252 (German translation from Biobliothque universelle 44 (1830), 225). E.S. Barr, Infrared phys. 2 (1962), 67-73. E.S. Barr, Infrared phys. 3 (1963), 195-206. M. Melloni, Ann. Chim. Phys. 55 (1833), 337-397. M. Melloni, Ann. Phys. Chem. 35 (1835), 385-413 and 529-578 (German translation of Ann. Chim. Phys. 55 (1833), 337-397). M. Melloni, Taylors Sci. Mem. 1 (1837), 39-74 (English translation of Ann. Chim. Phys. 55 (1833), 337-397). A.F. Svanberg, Ann. Phys. 160 (1851), 411-418. A. Rogalski, Infrared Phys. Technol. 43 (2002), 187-210. R.N. Jones, Appl. Opt. 2 (1963), 1090-1097. M. Davies, Infra-red spectroscopy and molecular structure. Elsevier, Amsterdam, 1963. H. Gershinowitz and E.B. Wilson Jr., J. Chem. Phys. 6 (1938), 197-200. F.A. Miller, Anal. Chem. 64 (1992), 824A-831A. B. Schrader, Nachr. Chem. Tech. Lab. 47 (1999), 1019-1022. N. Sheppard, Anal. Chem. 64 (1992), 877A-883A. P.A. Wilks Jr., Anal. Chem. 64 (1992), 833A-838A. W.P. Jencks, Methods Enzymol. 6 (1963), 914-928. Inflation calculator, http://inflationdata.com/Inflation/Inflation_Calculators/Inflation_Rate_Calculator. asp.
[9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40]
49
[41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93]
W. de W. Abney and E.R. Festing, Philos. Trans. Roy. Soc. London 172 (1881), 887-918. K. ngstrm, fv. Kongl. Vet. Akad. Frh. 47 (1890), 331-352. W.W. Coblentz, Astrophys. J. 20 (1904), 207-223. W.W. Coblentz, Phys. Rev. (Series 1) 20 (1905), 273-291 and 337-363. K.F. Luft, Angew. Chem. 19 (1947), 2-12. H.H. Mantsch, Chem. Phys. Lipids 96 (1998), 3-7. J. Lecomte, Le rayonnement infrarouge. Tome II. La spectromtrie infrarouge et ses applications physico-chimiques. Gauthier-Villars, Paris, 1949. R.B. Barnes, R. Perkin, J.A. Sanderson, and M.E. Warga, Physics Today June (1966), 115-117. V.Z. Williams, Appl. Spectrosc., 6 (1951), 3-29. L. Nobili and M. Melloni, Am. J. Sci. 23 (1833), 185-190 (English translation of Ann. Chim. Phys. 48, 198). E.F. Nichols, Phys. Rev. 1 (1893), 1-18. F. Rcker, Pflgers Arch. Eur. J. Physiol. 231 (1933), 742-749. E.R. Blout and R.C. Mellors, Science 110 (1949), 137-138. R. Stair and W.W. Coblentz, J. Res. Natl. Bur. Stand. 15 (1935), 295-316. E. Heintz, C. R. Acad. Sci. 201 (1935), 1478-1480. N. Wright, J. Biol. Chem. 120 (1937), 641-646. F. Vls and E. Heintz, Comptes Rendus 200 (1935), 1927-1929. S.E. Darmon and G.B.B.M. Sutherland, J. Am. Chem. Soc. 69 (1947), 2074. R.H. Gillette and F. Daniels, J. Am. Chem. Soc. 58 (1936), 1139-1142. R.C. Herman, J. Chem. Phys. 8 (1940), 252-258. D. Williams and L.H. Rogers, J. Am. Chem. Soc. 59 (1937), 1422-1423. E.R. Blout and M. Fields, Science 107 (1948), 252. R.F. Furchgott, H. Rosenkrantz and E. Shorr, J. Biol. Chem., 163 (1946), 375-386 R.N. Jones (Editor). Computer programs in infrared spectrophotometry. NRC Bull. 11, 12 (1968). A. Savitzky and M.J.E. Golay, Anal. Chem., 36 (1964), 1627-1639. J.S. Mattson, Anal. Chem., 49 (1977), 470-478. L.Y. Fager and J. O. Alben, Biochemistry, 11 (1972), 4786-4792. D.G. Cameron and H.H. Mantsch, Biochem. Biophys. Res. Commun. 83 (1978), 886-92. S.M. Greenwald, A.J. Hancock, H.Z. Sable, L. DEsposito and J.L. Koenig, Chem. Phys. Lipids. 18 (1977), 154-69. D. Chapman, J.C.Gomez-Fernandez, F.M. Goni and M.J. Barnard, Biochem. Biophys. Methods, 2 (1980), 315-323. J.S. Mattson, C.A. Smith and K.E. Paulsen, Anal. Chem., 47 (1975), 736738. J.K. Kauppinen, D.J. Moffatt, H. H. Mantsch and D. G. Cameron, Appl. Spectrosc. 35 (1981), 271276. H. Susi and D.M. Byler, Biochem. Biophys. Res. Commun. 115 (1983), 391-397. J.M. Olinger, D.M. Hill, R.J. Jakobsen and R.S. Brody, Biochim. Biophys. Acta. 869 (1986), 89-98. P.I. Haris, D.C. Lee and D. Chapman, Biochim. Biophys. Acta. 874 (1986), 255-265. R. Barer, Discussions of the Faraday Society, (1950), 369-378. Aschkinass, Ann. der Phys. 55 (1895), 401-431. W.W. Coblentz, J. Franklin Inst. 172 (1911), 309 A.M. Buswell, K. Krebs and W.H. Rodebush, J. Am. Chem. Soc., 59 (1937), 2603-2605. J.W. Ellis and J. Bath, J. Chem. Phys. 6 (1938), 723-729. R.C. Gore, R.B. Barnes, and E. Petersen, Anal. Chem. 21 (1949), 382-386. H.J. Lenormant, J. Physiol. (Paris), 42 (1950), 639-640. E.R. Blout and H.J. Lenormant, J. Opt. Soc. Am., 43 (1953), 1093-1095. F.S. Parker, Appl. Spectrosc. 12 (1958), 163-166. F.S. Prker and D.M. Kirschenbaum, Nature 187 (1960), 386-388. H. Susi, S.N. Timasheff and L. Stevens, J. Biol. Chem. 242 (1967), 6460-5466. H. Susi and D.M. Byler, in Methods for protein analysis By John P. Cherry, Robert A. Barford, American Oil Chemists Society, 1988, 235-255. J.L. Koenig and D.L. Tabb, in Analytical Applications of FT-IR to Molecular and Biological Systems (Durig, J.R., Ed.) D. Reidel, Boston. 1980, 241-255. D. Chapman and F.M. Goni in The Lipid Handbook By F. D. Gunstone, John L. Harwood, Fred B. Padley. Published by CRC Press, 1994, 487-504. D.L. Tabb, Ph.D. Thesis, Case Western Reserve University, 1974. J.L. Koenig and M. K. Antoon, Appl. Opt. 17 (1978), 1374-1385. D.M. Byler and H. Susi Biopolymers, 25 (1986), 469-487. A. Dong, P. Huang and W.S. Caughey, Biochemistry 29 (1989), 3303-3308.
50
[94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [120] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146]
D.C. Lee, P.I. Haris and D. Chapman and R.C. Mitchell, Biochemistry, 29 (1990), 9185-9193. F. Dousseau and M. Pezolet, Biochemistry 29 (1990), 8771-8779. M. Severcan, F. Severcan and P.I. Haris, J. Mol. Struc. 565-566 (2001), 383-387. J.A. Hering, P.R. Innocent and P.I. Haris, Spectroscopy An Int. J., 16 (2002), 53-69. S.E. Darmon, G.B.B.M. Sutherland, and G.R. Tristram, Biochem. J. 42 (1948), 508-516. M.M. Stimson, and M.J. ODonnell J. Am. Chem. Soc., 74 (1952), 1805-1808. U. Scheidt and H.Z. Rheinwein, Naturforsch. 7b (1952), 270. J. Fahrenfort, in Molecular Spectroscopy, Proceedings IV Int. Meeting Bologna, 1959, A. Mangini, Ed. (Pergamon, London), 2 (1962), 437. N.J. Harrick, Ann. N.Y. Acad. Sci. 101 (1963), 928-959. B. Katlafsky and R.E. Keller. Anal. Chem. 35 (1963 ), 1665-1670. R.J. Scheuplein, J. Soc. Cosm. Chem., Vol. 15 (1964), 111-122. F.S. Parker and R. Ans, Ana.l Biochem. 18 (1967), 414-422. R. Khurana and A.L. Fink, Biophys. J. 78 (2000), 994-1000. E. Arbely, I. Kass, and I.T. Arkin, Biophys. J. 85 (2003), 2476-2483. M. Jackson and H.H. Mantsch, Appl. Spectrosc., 46 (1992), 699-701. E. Goormaghtigh, V. Raussens, J.M. Ruysschaert, Biochem. Biophys. Acta., 1422 (1999), 105-185. K.J. Rothschild and N. A. Clark, Biophys. J. 25 (1979), 473-487. M. van de Weert, P.I. Haris, W.E. Hennink, D.J.A. Crommelin, Anal. Biochem. 297 (2001), 160-169. H.W. Thompson, Nature 158 (1946), 234. R. Barer, A.R.H. Cole and H.W. Thompson, Nature, 163 (1949), 198-201. K. Krishnan, American Chemical Society, Polymer Preprints, Division of Polymer Chemistry, 25 (1984), 182-184. P. Rigler, W.P. Ulrich, R. Hovius, E. Ilegems, H. Pick and H. Vogel, Biochemistry, 42 (2003), 1401714022. H.B. Mark, Jr. and B.S. Pons, Anal Chem., 38 (1966), 119-121. W. Mntele, A. Wollenweber, F Rashwan, J. Heinze, E. Nabedryk, G. Berger and J. Breton, Photochem. and Photobiol. 47 (1988), 451-455. D. Moss, E. Nabedryk, J. Breton and W. Mntele. Eur. J. Biochem. 187 (1990), 565-572. M.G. Rockley, D.M. Davis and H.H. Richardson, Science 21 (1980), 918-920. A. Hartstein, J.R. Kirtley and J.C. Tsang, Phys. Rev. Lett. 45 (1980), 201-204. B. Chance, Rev. Sci. Instrum. 22 (1951), 619-627 A.J. White, K. Drabble and C.W. Wharton, Biochem. J. 306 (1995), 843-849. B.C. Dunn, J.R. Marda, and E.M. Eyring, Appl. Spectrosc. 56 (2002), 751-755. J. Mann and H.W. Thompson, Proc. Royal Soc. London. Series A., 192 (1948), 489-497. A. Elliott and E.J. Ambrose, Discuss. Faraday Soc., 9 (1950), 246 251. M.J. Fraser and R.D.B. Fraser, Nature 167 (1951), 761-762. R.D.B. Fraser and T.P. MacRae in Conformation in Fibrous Proteins and Related Synthetic Polypeptides. Academic Press, Inc., New York. 1973. R.D.B. Fraser, J. Chem. Phys. 21 (1953), 1511-1515. H. Akutsu and H., Y. Kyogoku, H. Nakahara and K. Fukuda, Chem. Phys. Lipids. 15 (1975), 222-242. D.F.H. Wallach and P.H. Zahler Proc. Natl. Acad. Sci. 56 (1966), 1552-1559. A. Kukol, Spectroscopy An Int. J. 19 (2005), 1-16. M.E. Riepe and J. H. Wang, J. Biol. Chem. 243 (1968), 2779-2787. W.J. Leonard, K.K. Vijai and J.F. Foster, J. Biol. Chem. 238 (1963), 1984-1988. J.O. Alben and W.S. Caughey, Biochemistry, 7 (1968), 175-183. I. Noda. Bull. Am. Phys. Soc. 31 (1986), 31, 520. I. Noda, Y. Liu and Y. Ozaki, J. Phys. Chem, 100 (1996), 8674-8680. Y. Tanimura, S.J. Mukamel, Chem. Phys. 99 (1993), 9496-9511. F. Fournier, E.M Gardner, R. Guo, P.M. Donaldson, L.M.C. Barter, D.J. Palmer, C.J. Barnett, K.R. Willison, I.R. Gould and D.R. Klug, Anal. Biochem. 374 (2008), 358-365. C. Kolano, J. Helbing, M. Kozinski, W. Sander, P. Hamm, Nature 444 (2006), 469-472. F. Siebert and W. Mntele, Biophys. Struct. Mech. 6 (1980), 147-164. M.S. Braiman, P.L. Ahl and K.J. Rothschild, in Spectroscopy of Biological Molecules, eds. Alix, A.J., Bernard, L. & Manfait, M. (Wiley-Interscience, New York), 1985, 57-59. W. Uhmann, A. Becker, C. Taran, F. Siebert, Appl. Spectrosc. 45 (1991), 390-397. P.A. Anfinrud, C. Han, J.N. Moore, P.A. Hansen, and R.M. Hochstrasser. In Ultrafast Phenomena VI. T. Yajima, K. Yoshihara, C.B. Harris, S. Shionoya, editors. Springer Verlag, Berlin, 1988, 442-446. R. Stair and W.W. Coblentz, J. Research Nat. Bur. Standards, 15 (1935), 295-316. I.M. Klotz, P. Griswold and D.M. Gruen, J. Am. Chem. Soc., 71 (1949), 1615-1620. S.E. Darmon and G.B.B.M. Sutherland, J. Amer. Chem. Soc., 69 (1947), 2074.
51
[147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] [201] [202]
A. Elliott and E.J. Ambrose, Nature, 165 1950), 921-922. R.D.B. Fraser, W.C. Price, Nature, 170 (1952), 490-491. D. Chapman, J. Chem. Soc., (1956), 55-60. A. Hvidt, and S.O. Nielsen, Adv. Protein Chem., 21 (1966), 287-386. S. Meskers, J.M. Ruysschaert, and E. Goormaghtigh, J. Am. Chem. Soc. 121 (1999), 5115-5122. J. Lecomte, Le Spectre Infrarouge Paris, 1928. P.C. Rao, B.F. Daubert, J. Am. Chem. Soc., 70 (1948), 1102-1104; O.D. Shreve, M.R. Heether, H.B. Knight and D. Swern, Anal. Chem., 22 (1950), 1498-1501. R.T. OConnor, E.T. Field, W.S. Singleton, J. Am. Oil Chem. Soc. 28 (1951), 154-160. E.F. Binkerd and H.J. Harwood, J. Am. Oil Chem. Soc. 27 (1950), 60-62. P.I. Haris, Trends Biochem. 25 (2000), 104-105. N.K. Freeman, F.T. Lindgren, Y.C. Ng, A.V. Nichols, J. Biol. Chem., 203 (1953), 293-304. M.J. Barcelo and J. Bellanato, Anal.RS Esp. Fis. Quim. (Madrid). 49 (1953), 557. R.G. Sinclair, A.F. McKay, R.N. Jones, J. Am. Chem. Soc., 74 (1952), 2570-2575. S. Sunder, D. Caemeron, H.H. Mantsch and H.J. Bernstein, Can. J. Chem. 56 (1978), 2121. H.H. Mantsch and R.N. McElhaney, Chem. Phys Lipids., 57 (1991), 213-226. C. Clark, Appl. Spectroscp. 6 (1951), 14-17. E.R. Blout and M. Fields, J. Biol. Chem., 178 (1949), 335-43. E.R. Blout, H. Lenormant, J. Opt. Soc. Am. 43 (1953), 1093-1095. R.D.B. Fraser, Nature 170 (1952), 491. G.B.B.M. Sutherland and M. Tsuboi, Proc. Roy. Soc. (London). Series A, 239 (1957), 446-463. M. Falk, K.A. Hartman and R.C. Lord, J. Am. Chem. Soc., 85 (1963), 387-391. G.J. Thomas Jr., Biopolymers, 7 (1969), 325-334. J. Liquier, M. Pinot-Lafaix, E. Taillandier, J. Brahms, Biochemistry, 14 (1975), 4191-4197. J. Liquier, A. Mchami, E. Taillandier, J. Biomol. Struct. Dyn., 7 (1989), 119-26. E. Taillandier and J. Liquier, Methods Enzymol., 211 (1992), 307-335. D.C. Malins, N.L. Polissar, S.J. Gunselman Proc. Natl. Acad. Sci. U. S. A., 94 (1997), 3611-3615. J.F. Arakawa, J.F. Neault and H.A. Tajmir-Riahi, Biophys. J. 81 (2001), 1580-1587. L.P. Kuhn Anal. Chem., 22 (1950), 276-283. J.D.S. Goulden, Nature 177 (1956), 85-86. M. Kacurkov and R.H. Wilson, Carbohyd. Poly., 44, (2001), 291-303. J.D. Hardy and C.J. Muschenheim, J. Clin. Invest. 5 (1934), 817-831. J.D. Hardy and C.J. Muschenheim, J. Clin Invest. 1 (1936), 1-9. W.E. Pauli, and I. Ivancevic, Strahlentherapie, 25 (1927), 532. C.H. Cartwright, J. Opt. Soc. Am., 20 (1930), 81. A.R. Pearson and R.E. Norris, Brit. J. Radiol., 6 (1933), 480. K. Dobriner, S. Lieberman, C.P. Rhoads, R.N. Jones, V.Z. Williams, RB. J. Biol. Chem. 172 (1948), 297-311. K. Dobriner, T.H. Kritchevsky, D.K. Fukushima, S. Lieberman, T.F. Gallagher, J.D. Hardy, R.N. Jones and G. Cilento, Science, 109 (1949), 260-261. E.R. Blout, R.C. Mellors, Science, 110 (1949), 137-138. H.P. Schwarz, H.E. Riggs, C. Glick, W. Cameron, E Beyer, Jaffe and L. Trombetta,, Proc.Soc. Exp. Biol. Med. 76 (1951), 267-72. H.M. Randall, D.W. Smith, A.C. Colm, W.J. Nungester, Am. Rev. Tuberc. 63 (1951), 372-380. M. Pollard, F.B. Engley Jr., R.F. Redmond, H.I. Chinn and R.B. Mitchell, Proc. Soc. Exptl. Biol. Med., 81 (1952), 10-11. S. Levine, H.J.R. Stevenson, L.A. Chambers, B.A. Kenner, J. Bacteriol. 65 (1953), 10-15. D. Naumann, Appl. Spectroscopy Rev.36 (2001), 239-298. M. Jackson, M.G. Sowa and H.H. Mantsch, Biophys. Chem. 68 (1997), 109-125. J.T. Edsall, Cold Spr. Harb. Symp. quant. Biol. 6 (1938), 40. A.M. Buswell and R.C. Gore J. Phys. Chem., 46 (1942), 575-581. D.J. Fink, T.B. Hutson, K.K. Chittur, R.M. Gendreau, Anal. Biochem. 165 (1987), 147-54. R.B. Woodward, C.H. Schramm J. Am. Chem. Soc., 69 (1947), 1551-1552. W.T. Astbury, C.E. Dalgliesh, S.E. Darmon and G.B.B.M. Sutherland, Nature 164 (1949), 440-441. P.I. Haris and D. Chapman, Biopolymers, 37 (1995), 251-263. J.W. Ellis, and B.W. Sorge, J. Chem. Phys. 2 (1934), 559. E. Oldfield, D. Chapman, and W. Derbyshire, FEBS Lett. 16 (1971), 102-104. L. Tadesse, R. Nazarbaghi and L. Walters, J. Am. Chem. Soc., 113 (1991), 7036-7037. P.I. Haris, G.T. Robillard, A.A. Van Dijk, D. Chapman, Biochemistry, 31 (1992), 6279-6284. T.S. Anderson, J. Hellgeth and P.T. Lansbury Jr., J. Am. Chem. Soc., 118 (1996), 6540-6546.
52
[203] G. Herzberg, Molecular Spectra and Molecular Structure: II Infrared and Raman Spectra of Polyatomic Molecules, D. Van Nostrand Co. Inc., New York, 1945. [204] G.B.B.M. Sutherland, Discuss. Faraday Soc., 9 (1950), 274-281. [205] T. Miyazawa, T. Shimanouchi, S. Mizushima, J. Chem. Phys. 24 (1956), 408. [206] S. Krimm, J. Mol. Biol. 4 (1962), 528-40 [207] T. Miyazawa, J. Chem. Phys, 32 (1960), 1647-1652. [208] Y.N. Chirgadze, L.A. Gribov, N.S. Andreeva, N.E. Shutzkever, Zhurnal Fizicheskoy Khemie (Moscow), 35 (1961), 755-760. [209] F. Siebert and W. Mntele, Eur. J. Biochem., 130 (1983), 565-73. [210] D.C Lee and D. Chapman, Biosci. Rep. 6 (1986), 235-256. [211] P.S. Belton, R.H. Wilson and D.H. Chenery, Int. J. Biol. Macro. 8 (1986), 247-251. [212] T. Matsui, S. Tanaka and T. Akaike, J. Bioeng., 2 (1978), 539-541. [213] K.-M., Pan, M. Baldwin, J. Nguyen, M. Gasset, A. Serban, D. Groth, I. Mehlhorn, Z. Huang, R.J. Fletterick, F.E. Cohen and S.B. Prusiner, Proc. the Natl. Acad. Sci. USA, 90 (1993), 10962-10966. [214] W.K. Surewicz, H.H. Mantsch and D. Chapman. Biochemistry 32 (1993), 389-394. [215] M. Jackson and H.H. Mantsch, Crit. Rev. Biochem. Mol. Biol. 30 (1995), 95-120. [216] W.K. Surewicz and H.H. Mantsch, Biochimica et Biophysica Acta, 952 (1988), 115-130. [217] R.N. Jones, Eur. Spec. News 70 (1987), 10-20. [218] R.N. Jones, Eur. Spec. News 72 (1987), 10-20. [219] R.N. Jones, Eur. Spec. News 74 (1987), 20-34. [220] E.K. Plyler, Appl. Spectrosc. 16 (1962), 73-77. [221] J.R. Loofbourow, Rev. Mod. Phys. 12 (1940), 267-358. [222] G.B.B.M. Sutherland, Adv. Prot. Chem. 7 (1952), 291-318. [223] J.R. Durig and D.J. Gerson, Historical survey of the infrared and Raman spectroscopic study of biological molecules, in: Infrared and Raman spectroscopy of biological molecules, ed. T.M. Theophanides, D. Reidel Publishing Company, Dordrecht, 1979, 35-43. [224] H.A. Laitinen, Anal. Chem. 45 (1973), 2305. [225] A.C. Steven and W. Baumeister, J. Struc. Biol. 145 (2004), 181-183.
53
The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy

Andreas BARTH1 Department of Biochemistry and Biophysics, Stockholm University
Abstract. Reaction-induced infrared difference spectroscopy of proteins is reviewed. This technique enables detailed characterization of enzyme function on the level of single bonds of proteins, cofactors or substrates. Discussed are methods to initiate protein reactions in the infrared samples, general aspects of spectra interpretation, measurements of enzyme activity and studies of protein function at the example of the Ca2+ pump. Keywords. FTIR, protein structure, protein function, SERCA, enzyme activity, ligand binding.
1. Difference Spectroscopy - the Technique of Choice to Monitor Single Bonds in Large Proteins Elucidating the molecular mechanism of protein reactions is a major challenge for the life science community. Applying infrared spectroscopy to the study of protein reactions combines several of its advantages: (i) high time resolution (< 1 s), (ii) universal applicability from small soluble proteins to large membrane proteins, (iii) the high molecular information content, and (iv) a sensitivity high enough to detect a change in bond strength of a single bond in a large protein. In favourable cases, a protein reaction can be observed in the infrared absorption spectrum [1-4]. In most cases however, the effects are too small to be obvious. In addition, the information provided by an absorption spectrum is limited because it is composed of many overlapping bands. The key to obtain detailed structural information is to reduce the number of groups that contribute to a spectrum. This can be done by difference techniques which are particularly suited for protein reactions. Infrared difference spectroscopy monitors the changes in infrared absorption associated with a reaction, or in other words, it records the infrared difference spectrum of the reaction. A difference spectrum can be obtained by carefully subtracting the spectrum of a sample where the protein is in a particular state A from a spectrum where it is in a state B, see for example refs. [1, 5-12]. However, the absorbance changes usually observed
1 Corresponding author: Andreas Barth, Department of Biochemistry and Biophysics, The Arrhenius Laboratories for Natural Sciences, Stockholm University, S-10691 Stockholm, Sweden; E-mail: barth@dbb.su.se
54
A. Barth / The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy
for protein reactions are very small, on the order of 0.1% of the maximum absorbance. This is illustrated in Fig. 1. In consequence, subtracting spectra obtained from different protein samples does not generally allow the sensitive detection of the small absorbance changes between two protein states. Instead, the protein reaction of interest has to be initiated directly in the cuvette. This technique is termed reaction-induced infrared difference spectroscopy. For reviews see refs. [13-24]. In a typical experiment that employs reaction-induced difference spectroscopy, the protein is prepared in the stable state A and the absorbance of this state is measured. Then the reaction is triggered, the protein proceeds to state B and again the absorbance is recorded. Instead of only one particular state B, also a sequence of transient states may be adopted in the course of the reaction. In that case the interconversion between the product states B1, B2, etc. can be followed by time-resolved methods. From the spectrum recorded before the start of the reaction - state A - and the spectra recorded during and after the reaction - state(s) B - difference spectra are calculated. They originate only from those groups that are affected by the reaction. All passive residues are invisible in the difference spectrum which means that the number of observed groups is dramatically reduced compared to the absorption spectrum. Therefore, a difference spectrum exhibits details of the reaction mechanism on the molecular level despite a large background absorption.
Figure 1. The need for reaction-induced difference spectroscopy. Panel (a) compares absorption (full line) and difference spectrum (dashed line) of a membrane protein (Ca2+-ATPase) in 2H2O. The absorption spectrum exhibits prominent bands due to the C=O stretching vibration of lipids (~1730 cm-1) and the amide I vibration of proteins (~1650 cm-1). The difference spectrum appears as flat line when viewed on the same scale as the absorption spectrum, although it has been recorded for a reaction with relatively large absorbance changes. Panel (b) shows the same difference spectrum on a 100 times larger scale. Positive and negative bands in the difference spectrum are due to absorbance changes associated with the protein reaction. There is no obvious noise in the spectrum, demonstrating the sensitivity of infrared difference spectroscopy.
55
This chapter discusses first technical aspects of reaction-induced difference spectroscopy, then general aspects of spectra interpretation and finally two exemplary applications: enzyme activity measurements and the study of protein function. 2. Bottleneck 1 of Infrared Difference Spectroscopy: Concentrated Protein Samples Are Needed One drawback of infrared spectroscopy of aqueous solutions is the strong absorption of water in the mid-infrared spectral region (near 1645 cm -1) [25] which overlaps the important amide I band of proteins and some side chain bands (see section 4.3). When these protein bands are of interest, the strong water absorption demands a short path length for aqueous samples in transmission experiments, which is typically around 5 m. This implies a relatively high protein concentration for studies of protein reactions in order to be able to detect individual molecular groups in the spectrum. A desirable protein concentration is 1 mM or ~100 mg/ml which is nearly as high as the protein concentration found in cells [26]. Using 2H2O, the pathlength can be increased to 50 m and the concentration lowered because the water band is downshifted to ~1210 cm-1. Nevertheless, proteins often have to be concentrated for infrared spectroscopy, for example by partially drying the protein sample on an infrared window in a stream of nitrogen or in vacuum, by centrifugation and subsequent transfer of the pellet onto infrared windows, by direct centrifugation onto an infrared window, by incubation in an atmosphere of constant humidity defined by a saturated salt solution, or by micro-concentrators in case of soluble proteins or solubilised membrane proteins. Cooling of the sample might help to maintain protein functionality during concentration. An alternative to transmission experiments is the attenuated total reflectance (ATR) technique [19, 27-29]. Compared to transmission experiments, it avoids the handling problems which are caused by the required short pathlength. In an ATR experiment a sample is placed on a crystal. The total reflectance process at the crystal surface senses the absorption of the sample in a layer that extends about one wavelength away from the crystal surface, which means that the optical thickness of the sample is small enough for measurements of aqueous solutions. The sample is usually a protein film [28, 30-32], often prepared by drying, with a buffer solution above it. The thickness of the buffer layer does not influence the measured spectrum. The advantage of ATR spectroscopy is, that the buffer can be exchanged, which makes sample manipulations relatively simple and the method very flexible. The disadvantage may be that it is difficult to prepare a stable film. An unstable protein layer is easily disturbed upon buffer exchange which makes the recording of small absorbance changes impossible. 3. Bottleneck 2 of Infrared Difference Spectroscopy: Triggering Protein Reactions A crucial problem in reaction-induced difference spectroscopy is how to trigger the protein reaction of interest. The number of methods for this has constantly increased in the last decade and the main approaches are summarised in the following. Light-induced infrared difference spectroscopy is the technique that has first been applied to proteins from the early 1980s [33-35]. Here, continuous illumination or a light flash induces a reaction in photosensitive proteins, like bacteriorhodopsin [14, 15,
56
36, 37] or photosynthetic reaction centres [38-41]. From spectra recorded before and during/after illumination, light-induced difference spectra can be calculated. Concentration jump techniques are required to study the effects of ions or molecules on proteins. Three different approaches have been applied to generate a concentration jump in an infrared sample: (i) the ATR technique, (ii) the infrared variants of the stopped-flow and continuous-flow techniques, and (iii) the photolytical release of effector substances from biologically silent precursors (termed caged compounds). ATR has the advantage that manipulation of medium composition is straightforward when a stable protein film can be prepared. For example ligands can be added or the pH can be changed. Examples are studies of ligand binding to the nicotinic acetylcholine receptor [42, 43], to the gastric H+/K+-ATPase [44], to transhydrogenase [45], and of protein-protein interaction between transducin and rhodopsin [46]. A recent extension of the ATR technique separates a sample compartment close to the ATR crystal from a reservoir by a dialysis membrane. In this way the medium in the reservoir can be altered without disturbing the protein sample in the sample compartment [46-49]. A protein film is not required, making the technique also applicable for soluble proteins. Rapid mixing techniques are difficult to apply in infrared spectroscopy because of the viscous consistency of a concentrated protein solution and the small pathlength of less than 10 m for measurements in 1H2O. Nevertheless, successful applications of mixing devices have been reported [50-53], as reviewed recently [54]. A photolytically induced concentration jump can be achieved with photosensitive molecules that release a compound of interest upon illumination in the UV spectral range (300-350 nm). These molecules are termed caged compounds and have been used for 30 years to study biological reactions [55, 56]. In its caged form, the effector compound is modified such that it does not react with the protein of interest (see Fig. 2 for caged ATP). Photolysis of the caged compound leads to a sudden concentration jump of the free effector substance on the s to ms timescale (< 10 ms for caged ATP of Fig. 2) which initiates the protein reaction. A recent extension of the approach uses helper enzymes to induce a series of consecutive reactions [57]. Binding of the effector substance to a protein and subsequent conformational changes alter the infrared spectrum [58]. In addition to protein and effector molecule bands, the photolysis reaction is reflected in the difference spectra. In infrared studies of proteins, caged nucleotides, caged Ca2+ (Nitr-5 or DMNitrophen) and caged electrons (described in the photoreduction section) [59] have been used most often. The studies have dealt with two main aspects, molecule-protein recognition and the molecular basis of enzyme function. Most of these studies have been done on the sarcoplasmic reticulum Ca2+-ATPase (see section 6), which has also been the first enzyme to be studied with this technique [58]. Other proteins studied are alkaline phosphatase [60], annexin VI [61], DNaK [62], glutamate receptor [63], GroEL [64], kinases [65-67], Ras [68-70], and RecA [71]. Temperature- and pressure-jumps can be used to study folding and unfolding of proteins. The temperature jump is generated either by injecting a protein solution into a cuvette that is held at a different temperature (time resolution 100 ms) [72, 73], or by a laser pulse near 2 m that is absorbed by the sample solvent 2H2O [74-77] (temperature maximum reached after 20ns [74]), or by a visible laser pulse that excites a heat transducing dye [78, 79]. Pressure-jump experiments [80, 81] require a cell that allows
57
to set the sample under pressures of several kbar and have an experimental dead time of about 20 s.
Figure 2. Infrared difference spectroscopy with caged compounds at the example of caged ATP. (a) Photolysis of caged ATP. Caged ATP is modified at the -phosphate so that it does not react with the ATPase. Upon an UV flash (300-350 nm), the caged molecule photolyses which leads to a sudden concentration jump of free ATP (< 10 ms). (b) Infrared difference spectra upon release of ATP from caged ATP in the presence and absence of protein. From a spectrum recorded before ATP release and spectra recorded after ATP release, difference spectra are calculated (absorbance after ATP release minus absorbance before ATP release). Bands in the difference spectra originate only from those groups that are affected by ATP release with positive bands being characteristic of the state after ATP release and negative bands characteristic of the initial state. All "passive" residues are invisible in the difference spectrum which, therefore, highlights what happens to the "active" residues and exhibits details of the molecular reaction mechanism despite a large background absorption. Negative bands in difference spectra are characteristic of the initial state before release of ATP, while positive bands reflect the state(s) after ATP release. In addition to protein and ATP bands, the photolysis reaction is reflected in the difference spectra. Bold line: spectrum of caged ATP photolysis in the absence of protein. The main bands at 1524 and 1342 cm-1 have been assigned to the antisymmetric and symmetric stretching vibrations of the nitro group of caged ATP, respectively, and below 1270 cm-1 to a diminution of electron density in the phosphate P-O bonds upon photolysis [58]. Further information can be found in refs. [82-87]. Thin line: ATP release in a Ca2+-ATPase sample. Two reactions contribute to the signals: (i) caged ATP photolysis and (ii) the transition of the Ca2+ loaded ATPase Ca2E1 to the E2P phosphoenzyme where Ca2+ has been pumped and released. Tentative assignments of selected infrared difference bands are given. Prot: bands of protonated carboxyl groups due to protonation of acidic Ca2+ ligands upon Ca2+ release. The right and the left bands are characteristic of carbonyl groups with and without hydrogen bonding, respectively. Ca2+ release: two pairs of bands, each consisting of a positive and a negative component, originating from a change in absorption of carboxylate groups due to Ca2+ release; phosphate: band of the E2P phosphate group; conf: bands predominantly due to conformational changes of the protein backbone. Reprinted in modified form from [23]. 2002 American Chemical Society.
58
Equilibrium electrochemistry can be used to initiate redox reactions of proteins. Infrared investigations became available with the development of an ultra-thin-layer spectroelectrochemical cell suitable for protein investigations in aqueous solution [88]. The spectroelectrochemical cell permits control of the redox state of proteins in the infrared cuvette by applying a potential to a working electrode which changes the potential of the sample volume that is probed by the infrared beam. By applying a potential step at the working electrode, a redox reaction can be triggered in the electrochemical cell. From the infrared absorption spectrum before and after the potential step, a difference spectrum can be calculated that reflects only the redox reaction of the protein. The electrochemical cell is particularly useful when a protein contains several redox active cofactors with different midpoint potentials as it is often the case for proteins involved in photosynthesis and respiration. Since the method allows the precise control of the sample redox state, it is possible to selectively induce the redox reaction of only one particular cofactor, i.e. to dial-a-cofactor [21]. In a study of photosystem I for example, careful experimentation has avoided the oxidation of the abundant antenna chlorophylls and has allowed the exclusive titration of signals of the primary electron donor P700 (a chlorophyll dimer) [89]. While much of the initial work has been on photosynthetic proteins [21, 89-91], recent work has concentrated on cytochrome c oxidases [92-100]. More information on this method can be found in reviews [23, 24, 40, 100]. Photoreduction is a second way to induce redox reactions. It is based on photoexcitable electron donors like riboflavin [101, 102] or Ru2+ complexes [103] for which the expression caged electron has been coined [59]. Light absorption converts these systems directly or indirectly to a highly reducing state which can transfer one electron to a protein. The principle is similar to that of photosynthetic reaction centres where photoexcitation of the primary donor produces a highly reducing excited state and initiates electron transfer reactions within the protein. Photoreduction has been used in studies of aa3 cytochrome c oxidase of Rhodobacter sphaeroides [59], bo3 [59, 104, 105] and bd [106] ubiquinol oxidases of Escherichia coli and cytochrome P450cam [103]. 4. How to Make Sense of Infrared Difference Spectra 4.1. The Origin of Difference Bands Bands appear in difference spectra of protein reactions for several reasons: the chemical structure might change (protonation/deprotonation, catalytic reactions) or the three-dimensional structure of protein or cofactor. The former gives rise to a different absorption spectrum before and after the reaction. The latter changes vibrational coupling between neighbouring groups or the environment around particular functional groups causing band shifts and changes in the absorption index. In many cases, negative bands in a difference spectrum are characteristic of the state before the reaction and positive bands of the state(s) after the reaction. This interpretation might be misleading in the case of a change in absorption index since only the strength of absorption is affected but not the band position.
59
4.2. The Difference Spectrum Seen as a Fingerprint of Conformational Change Infrared difference spectra usually contain many difference bands which indicate the wealth of information that is encoded in the spectrum. However, extracting this information is often difficult. A simple first approach is to regard the spectra as a characteristic fingerprint of the conformational change. The spectral signature of a change in structure can then be used to detect and define transient states of a protein. Similar approaches have a long tradition in fluorescence and absorption spectroscopy. The approach can be used to study reaction intermediates and to classify and quantify conformational changes. It has even provided molecular information in a study that mapped substrate protein interactions [107] (see section 6.3 in this chapter). Intermediates. From the time course it is possible to evaluate the number of intermediates in the reaction. Here, time-resolved vibrational spectroscopy has the advantage that the observation is not restricted to a limited number of chromophores (i.e. Trp residues) or to an extrinsic fluorescence label which will largely reflect local changes in the vicinity of the chromophore(s) and may miss conformational changes occurring in distant regions of the protein. Instead, in vibrational spectroscopy all carbonyl chromophores of the backbone amide groups are monitored, and this will reveal any change in backbone conformation even if very small. Additionally, it is possible to follow the fate of individual catalytically "acting" groups in the same experiment. Thus, infrared spectroscopy simultaneously looks, on the one hand locally at the catalytic site, and on the other hand at the protein as a whole. This approach has been used in studies of the Ca2+-ATPase pump mechanism [108] of the photoactive yellow protein photocycle [109], and of protein folding studies [53, 74, 110, 111]. Similar conformational changes. From the shape of the spectra, conformational changes can be classified according to their similarity. This can be used to compare different preparations of a protein or related partial reactions [42, 89, 95, 112]. The extent of conformational change. From the magnitude of the difference signals, the extent of conformational change in a protein reaction may be estimated [108, 113-115] using the amide I region of the spectrum (1700 to 1610 cm -1) which is due to the absorption of the backbone carbonyl groups. Their absorption maximum depends on secondary structure due to transition dipole coupling [116] and throughbond coupling [117, 118], as well as on the strength of hydrogen bonding [119-121]. On this basis, the amplitude of the infrared difference signals in the amide I region can be used to estimate the change of secondary structure. Proceeding along this line, one has to consider the following: Signals of conformational changes may overlap in a way that they cancel each other leading to an underestimation of the extent of structural change. Therefore, the infrared difference spectrum reveals only the net change of secondary structure. A worst case scenario is shown in Fig. 3a, where nearly all residues change their secondary structure, but the net change is zero. Movements of rigid domains are not visible, only the working portion that changes its backbone geometry reflects in the difference spectra. Thus, it may be misleading to use terms such as "large" and "small conformational change" since considerable movements of rigid domains may originate from very small flexible parts of a protein like hinge regions that comprise only a few residues. An example is shown in Fig. 3b. Movement of the rigid domains (shown in grey) does not lead to signals in the infrared difference spectrum. Only the flexible part (shown in black), where the conformational change alters the
60
relative orientation of neighbouring amide groups, gives rise to infrared difference signals. Since transition dipole coupling leads to delocalised amide I modes, a simple linear relationship between signal magnitude and secondary structure change is not expected when individual residues change their secondary structure. The sensitivity towards conformational changes however, seems to be very high. For example, if an -helix shortens, this affects not only the amide modes of the backbone portion that unwinds, but also those of the remaining helix [122]. In addition to a secondary structure change, more subtle changes within a persisting secondary structure will also manifest in the spectrum. Examples are a change in hydrogen bonding to the C=O oxygens, a change in the twist of a -sheet or bending of an -helix. Signals due to amino acid side chains may overlap, although the amide I mode has a strong extinction coefficient [123, 124] which is generally larger than that of amino acid side chains in the amide I region [125, 126]. Several approaches have been proposed to quantify the number of residues participating in a secondary structure change [108, 113-115]. In spite of the implications and limitations discussed above, the approaches nevertheless seem to provide realistic estimates of the net secondary structure change [22, 127, 128].
Figure 3. Quantifying the extent of conformational change with infrared difference spectroscopy. (a) Worst case scenario: the protein undergoes a large conformational change, but the net change of secondary structure is zero since the N-terminal -sheet converts into an -helix and the C-terminal -helix into a -sheet. Infrared difference spectroscopy would not detect that conformational change - only the net change is detected. (b) Movement of rigid domains is invisible for infrared difference spectroscopy. When they move relative to each other, only the working part of the protein that causes the movement (shown in black) shows up in the spectrum. A large change in shape of a protein may therefore be accompanied only by small infrared absorbance changes.
61
Usually, signals of protein backbone perturbations have been found to be rather small, as shown for the electron transfer reactions of the photosynthetic reaction centre [39, 129], cytochrome c oxidase [92], cytochrome c [88, 130], bacterial cytochrome c3 [131], and myoglobin [132]. This indicates that the protein often provides an optimised solvent [129] rather than to act via a considerable reorganisation of secondary structure. On the other hand, small net secondary structure changes have also been observed for the Ca2+-ATPase [108], which alters the relative arrangement of its domains during catalysis. Therefore, small infrared signals in the amide I region do not rule out a considerable change in protein shape. 4.3. The Absorption of Amino Acid Side Chains Amino acid side chains are often at the heart of the molecular mechanism of proteins. Thus, side chain absorption can provide very valuable information, in particular when it is possible to follow the fate of the participating groups in a single time-resolved experiment. The aim of this kind of research is to identify the catalytically important side chains and to deduce their environmental and structural changes from the spectrum in order to understand the molecular reaction mechanism. For example, information may be obtained on the protonation state, coordination of cations and hydrogen bonding. Table 1 in the appendix gives an overview of the infrared absorption of amino acid side chains in 1H2O and 2H2O [24, 133]. Only the strongest bands are listed in table 1, or those in a spectral window free of overlap by bands from other groups. The absorption of a side chain in a protein may deviate significantly from its absorption in solution or in a crystal. The special environment provided by a protein is able to modulate strength and polarity of bonds, thus changing the vibrational frequency and the absorption coefficient. Therefore, the band positions given in the table should be regarded only as guidelines for the interpretation of spectra. It may be mentioned here that also the pKa of acidic residues in proteins may differ significantly from solution values. An example is Asp-96 of bacteriorhodopsin for which a pKa > 12 has been found [134]. 4.4. Molecular Interpretation: Band Assignment To fully exploit the information in an infrared difference spectrum, the spectroscopist needs to know which molecular group causes a given feature in the spectrum. Assignment of infrared bands to specific molecular groups is possible by studying model compounds, by chemical modifications of cofactors or ligands, by site-directed mutagenesis and by isotopic labeling of ligands, cofactors and amino acids. Model spectra. contributions of cofactors or substrate molecules to the infrared spectrum can be identified by normal mode calculations or by comparison with the spectra of the isolated molecules or model compounds in an appropriate environment. An example are chlorophyll studies [135-137]. Site-directed mutagenesis is a very powerful approach. Ideally, an infrared signal due to a specific amino acid is missing when this amino acid has been selectively replaced. The missing signal can then be assigned to the mutated amino acid. However, mutagenesis cannot be applied to crucial amino acids because their mutation abolishes protein function. Also, a mutation may exert wide-spread conformational effects on the protein which extensively modify the structural changes and thus the infrared
62
difference spectrum. Therefore, the effect of mutation on infrared difference spectra has to be evaluated very carefully. Isotopic labeling has been used as a powerful tool since the early days of infrared spectroscopy on proteins in order to observe a specific group in a large protein [5, 7]. It avoids perturbations of protein structure that might be introduced by mutagenesis and allows to label crucial amino acids that cannot be mutated without loss of function. Due to the mass effect on vibrational frequencies, infrared absorption bands of a labeled group are shifted with respect to those of the unlabeled groups and can be identified in the spectrum. Ligands, cofactors and protein side chains as well as backbone groups can be labeled. Labeling ligands is very informative when the interactions between ligands and proteins are investigated. Such studies include the binding of CO [5] and O 2 [138] to hemoglobin, of O2 [138] to myoglobin, of carbonyl groups to triosephosphate isomerase [7] and phospholipase A2 [139], and of phosphate groups to Ras [68] and Ca2+-ATPase [140]. Protein cofactors have also been labeled for example for bacteriorhodopsin [36, 141], photosynthetic reaction centres [38, 40, 41] and cytochrome c oxidase [94]. In favourable cases the substrate can transfer a labeled group to the protein which can then be studied in its protein environment. This approach has been used to study the acyl enzyme of serine proteases [20] and the phosphate group of the phosphoenzyme intermediates of the Ca2+-ATPase [142, 143]. Amino acids in proteins can be labeled in various ways. 1H/2H exchange is simply done by replacing 1H2O by 2H2O which exchanges the protons of accessible acidic groups, like OH, NH and SH groups, by deuteriums [144]. The observed characteristic band shifts often allow the assignment of these bands to peptide groups or to specific amino acid side chains. An additional advantage is the shift of the strong water absorption away from the amide I region (1610-1700 cm-1) which is sensitive to protein structure. Recombinant proteins can be labeled uniformly with for example 13C or 15N [145], all amino acids of one type can be labeled, or a label can be placed specifically on one particular amino acid [146, 147]. This site-directed labeling is the most powerful interpretation tool, unfortunately, it requires great effort and is usually not feasible. 4.5. Molecular Interpretation: Quantification Once a difference band is assigned to a specific molecular group, it is evident that this group participates in the studied protein reaction. Furthermore its frequency or the wavenumber of the corresponding infrared band provides precise information on a number of bond parameters and other molecular properties [148]. The quantitative interpretation becomes even more powerful with the increased ease of quantumchemical calculations of the vibrational spectrum [149]. For enzymes, the information obtained from calculations is a telltale of how the environment shapes the molecular properties of catalytically active groups. This is the information needed to unveil the secrets behind the amazing catalytic power of enzymes.
63
5. Enzyme Activity Measurements Enzymes are not only fundamental to life, they are also used in many biotechnological processes. Therefore, enzymatic activity is an important parameter. It is usually measured indirectly because substrate and product cannot be distinguished in the UV or visible range of the spectrum. Therefore coloured or fluorescent substrate analogues have been developed or the enzymatic reaction of interest is coupled to auxiliary enzymatic reactions that can be followed in the UV or visible spectral range. In contrast, infrared spectroscopy can provide a direct, "on-line" monitor of enzymatic reactions, because the infrared spectra of educts and products of an enzymatic reaction are often different. Measurements of enzyme activity with infrared spectroscopy are relatively straightforward and of interest even for researchers whose focus is to study the conformational changes associated with a protein reaction. This is because the infrared signals of the conversion of substrate to product allow an activity control in the same experiment that is used to monitor conformational changes. Examples for infrared spectroscopic measurements of enzyme activity are urea hydrolysis by urease [193], cefoxitin hydrolysis by -lactamase [150], deacylation of cinnamoyl-chymotrypsin [151], ATP hydrolysis by the Ca2+-ATPase [152], oxidation of D-glucose by glucose oxidase [153], hydrolysis of amides [154, 155], synthesis of hydroxamic acid derivatives [155] by amidase and the reaction of -ketoglutarate and Ala to Glu and pyruvate by glutamic-pyruvic transaminase [156]. In the example discussed here, ATP hydrolysis by the Ca2+-ATPase has been followed with infrared spectroscopy [152]. The different infrared absorption of substrates and products could be exploited to measure enzyme activity with only 7.5 g enzyme needed. Hydrolysis of ATP involves the net transformation of one PO2- group into a PO32group as shown in Fig. 4a. This is accompanied by a decrease in electron density in the terminal P-O bonds leading to a decrease in the vibrational frequency and thus a change in the vibrational spectrum. Fig. 4b shows the spectrum of the substrate ATP and that of the products ADP and Pi in a 1:1 mixture. Near 1230 cm-1 the antisymmetric stretching vibration of the PO2- group absorbs [157] and the - and the -PO2- group of ATP give rise to a prominent ATP band. This band is significantly reduced for the ADP and Pi mixture, since only the -PO2- group of ADP contributes. In contrast, near 1080 cm-1 the products absorb more strongly. This is the spectral region of the asymmetric stretching vibration of the PO32- group [157] and there are two PO32groups in the products ADP and Pi but only one in the substrate ATP. When an enzymatic reaction is followed by infrared spectroscopy, first a reference spectrum is recorded that represents the sample with substrate, in this case ATP. Then the reaction is started and successive spectra are recorded until the reaction is complete. From these spectra, the reference spectrum is subtracted to obtain difference spectra that only reflect the changes of absorption that occur in the course of the reaction. From the absorption spectra of ATP and the products ADP and Pi, a difference spectrum can be calculated that models ATP hydrolysis: absorbance of ADP and Pi minus the absorbance of ATP. This difference spectrum is shown in Fig. 4b. The following difference bands are observed: (i) a negative band near 1230 cm-1 reflecting the disappearing ATP band at 1230 cm-1 and (ii) a positive band near 1080 cm-1 reflecting
64
the increased absorbance of the products formed. This shows that both, the substrate concentration of ATP and the product concentration of ADP and Pi can be followed when ATP hydrolysis is monitored with infrared spectroscopy. The difference bands discussed above are expected for ATP hydrolysis catalysed by an enzyme. An example gives Fig. 4c: difference spectra after the release of ATP from caged ATP in the presence of the Ca2+-ATPase. They show bands due to two reactions: hydrolysis of ATP and photolysis of caged ATP. The first spectrum is dominated by photolysis bands, subsequent changes are due to the hydrolysis of ATP. As expected for ATP hydrolysis, the negative PO2- band evolves near 1230 cm-1 and a positive PO32- band near 1080 cm-1.
Figure 4. Infrared spectra of ATP hydrolysis showing that it is possible to measure ATPase activity with infrared spectroscopy. (a) Hydrolysis of ATP. (b) Model spectra: infrared absorption spectrum of 100 mM ATP (bold line), of 100 mM ADP plus 100 mM Pi (thin line) and difference spectrum (dotted line): absorbance of ADP and Pi minus absorbance of ATP. (c) ATPase activity measurement: infrared difference spectra induced by the release of ATP from caged ATP in the presence of Ca2+-ATPase. Five subsequent spectra of a typical sample are shown that monitor the progress of the hydrolysis reaction. The average time of recording is indicated. Reprinted from [152] with permission from Sage, Inc.
65
The PO2- band was used to measure enzyme activity with infrared spectroscopy [152]. The resulting specific activity determined from 6 measurements was 4.1 0.5 moles mg-1min-1. The major source of error seems to be the handling of minute protein and substrate volumes (1 l) and this can be minimised when automated mixing devices are used. In spite of this limitation, the specific activity obtained by infrared spectroscopy is in excellent agreement with the results of an independent activity measurement which gave 4.6 0.3 moles mg-1min-1 (5 measurements). This example shows that infrared spectroscopy can be used to measure enzyme activity with good accuracy. The amount of enzyme needed (here less than 10 g) is comparable to current methods and considerably less than needed for infrared studies of the molecular function of proteins. The advantages of infrared spectroscopy are: (i) no activity assay is required since substrates and products are monitored directly, (ii) the experimental conditions are not limited to those necessary for an activity assay and (iii) the infrared method has wide applicability since many enzymatic reactions lead to changes in the infrared spectrum. These advantages are not limited to the specific approach used here to start the reaction, but also apply to the more general mixing techniques. The ongoing development of mixing cuvettes will facilitate infrared activity measurements considerably - measuring enzyme activity with infrared spectroscopy will therefore become a wide-spread method in the immediate future. 6. Protein Function the Ca2+ Pump 6.1. The Ca2+ Pump P-type ATPase are major players in primary active transport of ions across biological membranes. Their name derives from the fact that these enzymes become phosphorylated by ATP during the transport cycle. One of the best characterized members of this family is the Ca2+-ATPase of the sarcoplasmic reticulum (SR) membrane from skeletal muscle (SERCA1a) [158] which serves as a model for the whole family of P-type ATPases. For reviews see [159-163]. The SR Ca2+-ATPase transports Ca2+ from the cytoplasm of muscle cells into the SR lumen which relaxes a flexed muscle. Protons are countertransported in exchange for Ca2+. Active transport of two Ca2+ is fuelled by the free energy from the hydrolysis of one molecule of ATP which is used with up to 100% efficiency [164]. A simplified version of the reaction sequence is given in Fig. 5. The Ca 2+-free ATPase exists in a pH dependent equilibrium between an E2 and an E1 form. It binds two cytosolic Ca2+ to the two high affinity Ca2+ binding sites in exchange for protons. Ca2+ binding enables ATP to phosphorylate Asp-351 of the ATPase which occludes the bound Ca2+. At least two phosphoenzyme intermediates (Ca2E1P and E2P) with different properties are formed consecutively. The first phosphoenzyme intermediate Ca2E1P is ADP-sensitive, i.e. dephosphorylates with ADP to form ATP, the second phosphoenzyme intermediate E2P is ADP-insensitive (E2P) and dephosphorylates by reaction with water. Phosphoenzyme conversion from Ca2E1P to E2P is accompanied by release of Ca2+ into the SR lumen and uptake of protons from the SR lumen. Hydrolysis of E2P and regeneration of the high affinity Ca2+ binding sites complete the reaction cycle. The Ca2+-ATPase serves as example to illustrate the study of protein function with infrared difference spectroscopy. The ATPase was the first protein to be studied by a
66
photolytically induced concentration jump [58]. Caged ATP [58, 108, 165, 166] and caged Ca2+ [114, 167, 168] have been predominantly used. The rapid scan technique with a time resolution of 65 ms is sufficient to kinetically resolve the main intermediates in the pump cycle after ATP release [108]. The functionality of the ATPase in the infrared samples has been demonstrated by a number of control experiments, for example Ca2+ uptake [165] and intrinsic fluorescence measurements [168], and by control experiments with inhibitors [114, 165-167, 169]. 6.2. Making Use of the Fingerprint Approach Regarding the infrared signals in the amide I region as fingerprint of the conformational change has enabled several conclusions on the transport mechanism. From the infrared data, it does not seem possible to distinguish between minor and major secondary structure changes in the catalytic cycle of the Ca2+-ATPase. This is in contrast to what would be expected from the classical model of the ATPase reaction cycle by de Meis and Vianna [170] that is based on only two main protein conformations E1 and E2, but is in line with the recent structural data [160, 162]. Instead, all main reaction steps studied are associated with secondary structure changes of comparable magnitude with that of phosphorylation somewhat smaller [108]. The clear detection of a conformational change upon phosphorylation is particularly interesting, because it is missed by X-ray crystallography [171]. In addition, evidence has also been obtained for a pH dependent conformational change of the protein that affects the Ca2E1 E2P transition [172].
Figure 5. Simplified reaction cycle of the Ca2+-ATPase. The cartoons illustrate accessibility and protonation state of the Ca2+ binding sites. For clarity only one Ca2+ and one H+ is indicated. Cyt stands for cytoplasm and L for lumen.
67
For all partial reactions investigated, the overall backbone conformational changes proceed at the same time as the local perturbations of side chains [108]. An analysis of the kinetics has not detected one of the postulated intermediates in the reaction cycle [108]. It was therefore concluded that it is either short-lived or does not exist. Difference spectra of the two Ca2+ release reactions from the phosphorylated (Ca2E1P E2P) and the unphosphorylated enzyme (Ca2E1 E1/E2) show striking similarity [112] and similar conformational changes have been concluded from this observation. Since difference spectra of a reaction contain information on protein structure, side chain protonation, hydrogen bonding and Ca2+ binding mode of the initial and the final state, the observed similarity suggests that the occupied and unoccupied Ca2+ binding sites are most likely the same in the two reactions. Thus, a model with only one pair of binding sites for Ca2+ is favoured from the infrared spectra. It is in contrast to the model by Jencks [173] that proposes two different pairs of sites for cytoplasmic high affinity and lumenal low affinity binding sites, respectively. However, it is in agreement with mutagenesis studies [174] and with crystal structures in different ATPase states [175, 176]. 6.3. Nucleotide Binding Depends on Individual Interactions One line of studies has been to map interactions between protein and substrate ATP. Fig. 6a shows infrared absorbance changes induced by nucleotide (NTP) binding to the Ca2+-ATPase. The spectra reveal the difference in absorbance between the initial nucleotide-free state Ca2E1 and the nucleotide-ATPase complexes Ca2E1NTP. The difference spectra reflect conformational changes of the protein backbone in the amide I (1700-1610 cm-1) region. The signals near 1693, 1641 and 1628 cm-1 are characteristic of -sheets, those near 1665 cm-1 are suggestive of turns, and those near 1653 cm-1 are indicative of -helical structures. The spectra indicate that -helices, sheets and turns are affected by nucleotide binding [107]. Close ATP analogs (Fig. 6b) produce nucleotide binding spectra that are different from that obtained with ATP (Fig. 6a). Therefore, the conformational change upon nucleotide binding depends to a surprising degree on individual interactions between ATPase and nucleotide [107, 177]. The lack of individual interactions produces more than just local adjustments, it affects the entire conformation of the nucleotide-ATPase complex. Surprisingly, modification at opposite ends of the ATP molecule, interacting with different domains, produces similar effects. In particular, omission of the phosphate [177] and modification of the amino [107] group both reduce the conformational change, with the latter modification having a more dramatic effect. This suggests a concerted conformational change upon ATP binding for which all interactions need to be in place [107]. As a consequence, the (average) structure of the nucleotide-ATPase complex is characteristic of the nucleotide bound. From the sensitivity of the conformational change on individual interactions it has been concluded that the ATPase interacts with the -phosphate [177], the ribose hydroxyls and the amino function [107] of ATP. The interactions identified by infrared spectroscopy have later been confirmed by X-ray crystallography [171, 178].
68
Figure 6. Infrared absorbance changes induced by nucleotide binding to the ATPase. (a) Difference spectra of nucleotide binding to the Ca2+-ATPase (Ca2E1 Ca2E1NTP) obtained with ATP, 2'-deoxyATP, 3'deoxyATP and ITP [107]. Results for ADP [177] and the close ATP analog AMPPNP (,-imidoadenosine 5'-triphosphate) [107] are not shown. Labels indicate the band positions of the ATP binding spectrum. (b) Structures of ATP and ATP analogues highlighting the modified functional groups of ATP. The functional groups interact with different domains of the ATPase. Light and dark grey indicate interaction with the N and P domain, respectively. Each modification of ATP affects the binding induced conformational change. Thus all modified groups are involved in important interactions with the ATPase. Reprinted in modified form from [127]. 2006 Nova Science Publishers.
Binding of ATP closes a cleft between nucleotide-binding (N) and phosphorylation (P) domain of the ATPase [160, 179]. This movement delivers the -phosphate to the phosphorylation site Asp-351 in the P domain [171, 178]. The two domains are bridged by ATP in the ATP-ATPase complex (Ca2E1ATP), as sketched in Fig. 6b. This bridging function of ATP provides an explanation for the drastic structural effects of modifying the ribose 3'-OH and the adenine amino function. Interactions of both groups stabilise the closed conformation of the complex, 3'-OH directly via an interaction with Arg-678 in the P domain, and the amino group indirectly because its interaction with Glu-442 of the N domain seems to position the ribose hydroxyls such that interaction with the P domain is possible. When inosine triphosphate (ITP) is used
69
instead of ATP, Glu-442 will repel the negative partial charge on the inosine oxygen, reorient the inosine moiety and thereby sacrifice interactions of the ribose hydroxyls with the P domain. This explains the weaker binding of ITP and the smaller extent of conformational change upon ITP binding [107]. The results discussed so far in this chapter have been obtained by monitoring the conformational change of the peptide backbone. However, the absorption of the bound nucleotide can be observed directly if isotopic labeling is used to identify difference bands of specific nucleotide groups. Since the vibrational frequency depends on the masses of the vibrating atoms, isotopic labeling shifts bands of the labeled group which can then be identified in the spectrum. In studies of nucleotide binding to the Ca 2+ATPase, - and -phosphates have been labeled. The spectral band positions of the and -phosphate bands indicate that P-O bond strengths of bound - and -phosphate are similar to those of ATP in aqueous solution, i.e. that ATP's hydrogen bonds of ATP to water are largely replaced by interactions with the protein for bound ATP [140]. Compared to GTP binding to Ras [68, 69, 180, 181], the phosphate bands of ATP bound to the ATPase are generally found at similar positions. However, the extent of vibrational coupling between different phosphate groups seems to be different: they are largely decoupled in the GTP-Ras complex, whereas they are coupled for ATP bound to the ATPase. Transfer of the nucleotidic -phosphate to Asp351 follows nucleotide binding and depends also on the type of nucleotide [128]. An interesting feature has been observed with ITP: its phosphorylation spectrum shows additional signals in the amide I region as compared to ATP, which are similar to nucleotide binding signals [128]. Thus it seems that upon phosphorylation with ITP the enzyme catches up on a conformational change that cannot be achieved by ITP binding because the interactions between protein and base moiety are impaired. ADP dissociation from the phosphoenzyme Ca2E1P results in conformational changes which are the reverse of those induced by ADP binding to the unphosphorylated ATPase Ca2E1 [57]. This is indicated by infrared experiments that accumulate Ca2E1P and in which ADP is removed by the helper enzyme apyrase. Upon dissociation of ADP from the phosphoenzyme, the conformation relaxes partially back to that of the unphosphorylated state Ca2E1. Thus, ADP plays an important role in stabilizing the closed conformation of Ca2E1P [57, 128]. ADP dissociation from Ca2E1P does not trigger the transition to E2P [57] as proposed previously, since the spectral characteristics of E2P [108, 142, 182] are not observed upon ADP dissociation. 6.4. Protonation of the Empty Ca2+ Binding Sites Has Been Observed Directly Amongst the side chain groups, protonated carboxyl groups are often the infrared spectroscopist's favourite. They absorb in a region (1700 - 1800 cm-1) that is usually free from the absorption of other groups and thus can readily be assigned. The ability of infrared spectroscopy to detect protonation states of amino acids is important, since protons are invisible in X-ray crystallography but proton transfer steps are often an essential part of the enzymatic reaction mechanism. For the Ca2+-ATPase, infrared bands of protonated Asp and Glu residues of E2P [142, 165, 182] and of Ca2+ free ATPase (E1 or E2) [112, 114, 167, 168] have been observed upon Ca2+ release from the phosphoenzyme and Ca2+ binding to the unphosphorylated ATPase, respectively. The E2P bands are marked with "prot" in Fig. 2. They are composed of at least five overlapping bands. Three of the E2P bands are
70
pH dependent and titrate with a pKa value near 8.3 [172] which is similar to the apparent pKa value of residues binding lumenal H+ for proton countertransport [183185]. This similarity of the pKa values supports the earlier interpretation that the bands originate from the protonation of carboxyl groups in the Ca2+ binding sites [114, 168], which are involved in H+ transport [112]. The spectral position of the bands indicates that some of the carboxyl carbonyl oxygens are hydrogen bonded while others are not. This observation has enabled a tentative assignment of the signals to Glu-771 (H-bonded), Asp-800 (some conformers H-bonded, some not) and Glu-908 (H-bonded) by multiconformation continuum electrostatics calculations [172]. A definitive assignment has to await experiments with site-directed mutants. 6.5. The Environment of the Phosphate Group Facilitates Phosphate Transfer The ATPase is one of the many examples in which phosphorylation controls biochemical reactions. The ATPase phosphoenyzymes have different properties, which is essential for coupling ATP hydrolysis to Ca2+ transport [186]. For example, E2P dephosphorylates faster with water than Ca2E1P and the model compound acetyl phosphate. This is required for the fast progression of the pump cycle and therefore for the efficient removal of Ca2+ from the cytoplasm of muscle cells. Obviously, the environment of the phosphate group is important in controlling the dephosphorylation properties. Bond properties and interactions of the phosphate group have been characterised by infrared spectroscopy. The essential step here is to identify the phosphate absorption in the difference spectra with help of isotopic substitution. Work on Ca 2E1P [140, 187] and an initial study on E2P [142] have compared infrared difference spectra obtained with labeled and unlabeled -phosphate of ATP. This approach has identified isotopesensitive bands of Ca2E1P and E2P which can be assigned to the phosphate group. The band positions are different for the two phosphoenzymes, indicating a conformational change that directly affects geometry and electron density of the phosphate group and makes the environment in E2P more hydrophobic [142]. The complete set of three E2P P-O stretching vibrations has been determined in an isotope exchange experiment which is more sensitive than the comparison of spectra obtained with different isotopes of ATP [188]. The experiment has observed an oxygen isotope exchange at the phosphate group that is catalysed by the ATPase [189]. It provides an infrared spectrum at atomic resolution in a crowded spectral region [143, 188] which reveals the three stretching vibrations of the transiently bound phosphate group in spite of a background absorption of 50 000 protein vibrations. The spectrum is shown in Fig. 7a. Bands of the terminal P-O stretching vibrations of the unlabeled phosphate group are found at 1194, 1137, and 1115 cm-1.
71
The information on the E2P phosphate vibrations has been evaluated [143] using a correlation between P-O frequency and P-O bond valence, the bond valence model and empirical correlations to calculate P-O bond strengths and lengths, and the dissociation energy of the bridging P-O bond [143]. Compared to the model compound acetyl phosphate, structure and charge distribution of the E2P aspartyl phosphate resemble somewhat the transition state in a dissociative phosphate transfer reaction: the aspartyl phosphate of E2P has 0.02 shorter terminal P-O bonds and a 0.09 longer bridging P-O bond, which is ~20% weaker and has 64 - 90 kJ/mol less bond energy [143]. These findings are summarised in Fig. 7b. Similar effects have been concluded for Ca 2E1P [140, 190], the values of which are between those of acetyl phosphate and E2P, but closer to those of E2P. Interestingly, the differences between acetyl phosphate and the phosphoenzymes in the bridging P-O equilibrium bond length (max. 0.09 ) are comparable to the bond length fluctuations in the vibrational ground state (0.06 ) [190].
Figure 7. Determination of phosphate bond properties by infrared spectroscopy. (a) Infrared difference spectrum of E2P16O3 E2P18O3 isotope exchange at the phosphate group [143] calculated by subtracting the spectrum before exchange from the spectrum after exchange. Reprinted from [127]. (b) Phosphate bond parameters for the model compound acetyl phosphate and E2P [143]. Italic numbers above bonds give bond lengths in , normal print numbers below bonds give bond valences in vu [192]. The values have been rounded to two decimal digits. Reprinted in modified form from [127]. 2006 Nova Science Publishers.
72
The destabilisation of the bridging P-O bond is a ground state property of Ca2E1P and E2P. This finding is consistent with the view that part of the catalytic power of enzymes derives from ground state properties, in particular from favouring near-attack conformations in which the arrangement of reacting atoms is similar to that in the transition state [191]. The weaker bridging P-O bond of E2P accounts for a 1011- to 1015-fold hydrolysis rate enhancement implying that P-O bond destabilization facilitates phosphoenzyme hydrolysis. P-O bond destabilization is caused by a shift of non-covalent interactions from the phosphate oxygens to the aspartyl oxygens. Therefore it has been proposed [143] that the relative strength of non-covalent bonding to the phosphate and aspartyl oxygens is one of the key factors that tunes the hydrolysis rate of the ATPase phosphoenzymes and related phosphoproteins. Weaker bonding to the phosphate oxygens and stronger bonding to the aspartyl oxygens weakens the bridging P-O bond, which increases dramatically the catalytic power of the enzyme. Weakening and elongation of the bridging P-O bond is not accomplished by external mechanical forces that pull the bond apart. Instead it is an in-built response of aspartyl phosphate to a shift of interactions from phosphate to aspartyl oxygens, with only subtle changes in distances are required. This provides an elegant "handle" for the enzyme to control hydrolysis. 7. Appendix
Table 1. Overview of amino acid side chain infrared bands. Extended and corrected version of a previous compilation [133]. For more information and references see [24, 133]. If available, parameters of infrared spectra of amino acid side chains are given. If not, data are taken from infrared spectra of model compounds or from Raman spectra. Band positions are given for 1H2O and 2H2O, the value in brackets is the absorption coefficient or extinction coefficient . The shift upon 1H/2H exchange is given when a compound in both solvents is compared in the original work. The listing of internal coordinate contributions to a normal mode is according to their contribution to the potential energy of the normal mode (if specified in the literature): if the contribution of an internal coordinate to the potential energy of a normal vibration is 70% only that coordinate is listed. Two coordinates are listed if their contribution together is 70%. In all other cases those three coordinates that contribute strongest to the potential energy are listed. If no assignment is listed, no or multiple assignments are given in the original publications. Vibrations dominated by amide group motions are not included. : stretching vibration, s: symmetric stretching vibration, as: antisymmetric stretching vibration, : in plane bending vibration, as: asymmetric in plane bending vibration, w: wagging vibration, t: twisting vibration, r: rocking vibration. Assignment Cys, (SH) Asp, (C=O) Glu, (C=O) Asn, (C=O) Arg, as(CN3H5 ) Gln, (C=O) Arg, s(CN3H5+) HisH2 , (C=C)
+ +
Band position / cm-1, ( / M-1cm-1) in 1H2O 2551 1716 (280) 1712 (220) 1677-1678 (310-330) 1652-1695 (420-490) 1668-1687 (360-380) 1614-1663 (300-340) 1631 (250)
Band position / cm-1, ( / M-1cm-1) in 2H2O 1849 1713 (290) 1706 (280) 1648 (570) 1605-1608 (460) 1635-1654 (550) 1581-1586 (500) 1600 (35), 1623 (16)
73
Lys, as(NH3+) Tyr-OH, (CC) (CH) Asn, (NH2) Trp, (CC), (C=C) Tyr-O , (CC) Tyr-OH, (CC) Gln, (NH2) HisH, (C=C) Asp, as(COO ) Glu, as(COO ) Lys, s(NH3 ) Tyr-OH, (CC), (CH) Trp, (CN), (CH), (NH) Tyr-O , (CC), (CH) Trp, (CC), (CH) Phe, (CCring) as(CH3) (CH2) Pro, (CN) Trp, (CH), (CC), (CN) His , (CH3), (CN) Trp, (NH), (CC), (CH) Gln, (CN) Glu, s(COO ) Asp, s(COO ) s(CH3) Trp Trp (CH) Trp, (NH), (CN), (CH) Tyr-O-, (C-O), (CC) Asp, Glu, (COH) Trp, (CH), (CC) Tyr-OH (C-O), (CC) His, (CH), (CN), (NH) Trp, (CC)
+ -
1626-1629 (60-130) 1614-1621 (85-150) 1612-1622 (140-160) 1622 1599-1602 (160) 1594-1602 (70-100) 1586-1610 (220-240) 1575,1594 (70) 1574-1579 (290-380) 1556-1560 (450-470) 1526-1527 (70-100) 1516-1518 (340-430) 1509 1498-1500 (700) 1496 1494 (80) 1445-1480 1425-1475 1400-1465 1462 1439 1412-1435 1410 1404 (316) 1402 (256) 1375 or 1368, 1385 1352-1361 1334-1342 1315-1350 1276 1269-1273 (580) 1264-1450 1245 1235-1270 (200) 1217, 1229, 1199 1203
1201 1612-1618 (160)
1618 1603 (350) 1590-1591 (<50) 1163 1569, 1575 1584 (820) 1567 (830) 1170 1513-1517 (500)
1498-1500 (650)
1455 (200) 1439 1382 1409 1407 1404
1334 (100)
955-1058
1248-1265 (150) 1217, 1223, 1239
74
Ser, (COH) or (CO2H), (CO) w(CH2) Tyr-OH, (COH) Asp,Glu, (C-O) His, (CN), (CH) Trp, (CH), (NC) Trp, (NC), (CH), (CC) t(CH2) Thr, (C-O) Ser, (C-O) Trp, (CC), (CH) Ser, (CO) or (CC) Ser, (CO), (CO H) Thr, (CO H) r(CH2)
2 2
1181-1420 1170-1382 1169-1260 (200) 1160-1253 1104,1090,1106,1094 1092 1064 1063-1295 1075-1150 1030 1012-1016 983
875-985
913 1250-1300 1104,1096,1107,1110
1023 1012
940 865-942 724-1174
References
[1] J. Trewhalla, W.K. Liddle, D.B. Heidorn, and N. Strynadka, Biochemistry 28 (1989), 1294-1301. [2] M. Jackson, P.I. Haris, and D. Chapman, Biochemistry 30 (1991), 9681-9686. [3] M. Nara, M. Tasumi, M. Tanokura, T. Hiraoki, M. Yazawa, and A. Tsutsumi, FEBS Lett. 349 (1994), 8488. [4] H. Fabian, T. Yuan, H.J. Vogel, and H.H. Mantsch, Eur. Biophys. J. 24 (1996), 195-201. [5] J.O. Alben and W.S. Caughey, Biochemistry 7 (1968), 175-183. [6] M.E. Riepe and J.H. Wang, J. Biol. Chem. 243 (1968), 2779-2787. [7] J.G. Belasco and J.R. Knowles, Biochemistry 19 (1980), 472-477. [8] P. Tonge, G.R. Moore, and C.W. Wharton, Biochem. J. 258 (1989), 599-605. [9] P.J. Tonge, M. Pusztai, A.J. White, C.W. Wharton, and P.R. Carey, Biochemistry 30 (1991), 4790-4795. [10] G. Zundel, Adv. Chem. Phys. 111 (2000), 1-217. [11] G. Iliadis, G. Zundel, and B. Brzezinski, Biospectroscopy 3 (1997), 291-297. [12] F. Bartl, D. Palm, R. Schinzel, and G. Zundel, Eur. Biophys. J. 28 (1999), 200-207. [13] M.S. Braiman and K.J. Rothschild, Annu. Rev. Biophys. Biophys. Chem. 17 (1988), 541-570. [14] K. Gerwert, Biol. Chem. 380 (1999), 931-935. [15] J. Heberle, Recent Res. Devel. Applied Spectroscopy 2 (1999), 147-159. [16] C. Jung, J. Mol. Recognit. 13 (2000), 325-351. [17] R. Vogel and F. Siebert, Curr. Opin. Chem. Biol. 4 (2000), 518-523. [18] S. Kim and B.A. Barry, J. Phys. Chem. 105 (2001), 4072-4083. [19] K. Fahmy, Recent Res. Devel. Chem. 2 (2001), 1-17. [20] C.W. Wharton, Nat. Prod. Rep. 17 (2000), 447-453. [21] W. Mntele, TIBS 18 (1993), 197-202. [22] A. Barth and C. Zscherp, Quart. Rev. Biophys. 35 (2002), 369-430. [23] C. Zscherp and A. Barth, Biochemistry 40 (2001), 1875-1883. [24] A. Barth, Biochim. Biophys. Acta 1767 (2007), 1073-1101. [25] S.Y. Venyaminov and F.G. Prendergast, Anal. Biochem. 248 (1997), 234-245. [26] R.J. Ellis, Curr. Opin. Struct. Biol. 11 (2001), 114-119. [27] U.P. Fringeli, In situ infrared attenuated total reflection membrane spectroscopy, in: Internal reflection spectroscopy, ed. F.M.J. Mirabella, Marcel Dekker, Inc., New York, 1992, 255-324.
75
[28] E. Goormaghtigh, V. Raussens, and J.-M. Ruysschaert, Biochim. Biophys. Acta 1422 (1999), 105-185. [29] L.K. Tamm and S.A. Tatulian, Quart. Rev. Biophys. 30 (1997), 365-429. [30] K.A. Oberg and A.L. Fink, Anal. Biochem. 256 (1998), 92-106. [31] P. Rigler, W.P. Ulrich, P. Hoffmann, M. Mayer, and H. Vogel, CHEMPHYSCHEM 4 (2003), 268-275. [32] P. Rigler, W.P. Ulrich, and H. Vogel, Langmuir 20 (2004), 7901-7903. [33] F. Siebert, W. Mntele, and W. Kreutz, Biophys. Struct. Mech. 6 (1980), 139-146. [34] J.O. Alben, D. Beece, S.F. Bowne, L. Eisenstein, H. Frauenfelder, D. Good, M.C. Marden, P.P. Moh, L. Reinisch, A.H. Reynolds, and K.T. Yue, Phys. Rev. Lett. 44 (1980), 1157-1160. [35] K.J. Rothschild, M. Zagaeski, and W.A. Cantore, Biochem. Biophys. Res. Commun. 103 (1981), 483489. [36] K.J. Rothschild, J. Bioenerg. Biomembr. 24 (1992), 147-167. [37] A. Maeda, Israel J. Chem. 35 (1995), 387-400. [38] J. Breton, Biochim. Biophys. Acta 1507 (2001), 180-193. [39] W. Mntele, Infrared vibrational spectroscopy of the photosynthetic reaction center, in: The photosynthetic reaction center, eds. J. Deisenhofer and J.R. Norris, Vol. 2, Academic Press, San Diego, 1993, 239-283. [40] W. Mntele, Infrared vibrational spectroscopy of reaction centers, in: Anoxygenic Photosynthetic Bacteria, eds. E. Blankenship, M.T. Madigan, and C.E. Bauer, Kluwer Academic Publishers, Dordrecht, 1995, 627-647. [41] E. Nabedryk, Light-induced Fourier transform infrared difference spectroscopy of the primary electron donor in photosynthetic reaction centers, in: Infrared spectroscopy of biomolecules, eds. H.H. Mantsch and D. Chapman, Wiley-Liss, New York, 1996, 39-82. [42] J.E. Baenziger, K.W. Miller, and K.J. Rothschild, Biochemistry 32 (1993), 5448. [43] J.E. Baenziger, K.W. Miller, and K.J. Rothschild, Biophys. J. 61 (1992), 983-992. [44] F. Scheirlinckx, V. Raussens, J.-M. Ruysschaert, and E. Goormaghtigh, Biochem. J. 382 (2004), 121129. [45] M. Iwaki, N.P.J. Cotton, P.G. Quirk, P.R. Rich, and J.B. Jackson, J. Am. Chem. Soc. 128 (2006), 26212629. [46] K. Fahmy, Biophys. J. 75 (1998), 1306-1318. [47] E. Agic, O. Klein, and W. Mntele, Binding and interaction of effector molecules to proteins studied with an attenuated total reflection infrared (ATR-IR) microdialysis cell, in: Book of abstracts: 10th European conference on the spectroscopy of biological molecules, eds. B. Szalontai and Z. Kta, JATEPress, Szeged, 2003, 93. [48] S. Gourion-Arsiquaud, S. Chevance, P. Bouyer, L. Garnier, J.-L. Montillet, A. Bondon, and C. Berthomieu, Biochemistry 44 (2005), 8652-8663. [49] M. Krasteva, S. Kumar, and A. Barth, Spectroscopy 20 (2006), 89-94. [50] A.J. White, K. Drabble, and C.W. Wharton, Biochem. J. 306 (1995), 843-849. [51] P. Hinsmann, M. Haberkorn, J. Frank, P. Svasek, M. Harasek, and B. Lendl, Appl. Spectrosc. 55 (2001), 241-251. [52] R. Masuch and D.A. Moss, Stopped flow system for FTIR difference spectroscopy of biological macromolecules, in: Spectroscopy of biological molecules: new directions, eds. J. Greve, G.J. Puppels, and C. Otto, Kluwer Academic Publishers, Dordrecht, 1999, 689-690. [53] E. Kauffmann, N.C. Darnton, R.H. Austin, C. Batt, and K. Gerwert, Proc. Natl. Acad. Sci. USA 98 (2001), 6646-6649. [54] H. Fabian and D. Naumann, Methods 34 (2004), 28-40. [55] J.H. Kaplan, B. Forbush, and J.F. Hoffman, Biochemistry 17 (1978), 1929-1935. [56] M. Goeldner and R. Givens, eds., Dynamic studies in biology. Wiley-VCH, Weinheim, 2005. [57] M. Liu, E.-L. Karjalainen, and A. Barth, Biophys. J. 88 (2005), 3615-3624. [58] A. Barth, W. Mntele, and W. Kreutz, FEBS Lett. 277 (1990), 147-150. [59] M. Lbben and K. Gerwert, FEBS Lett. 397 (1996), 303-307. [60] L. Zhang, R. Buchet, and G. Azzar, Biophys. J. 86 (2004), 3873-3881. [61] J. Bandorowicz-Pikula, A. Wrzosek, M. Danieluk, S. Pikula, and R. Buchet, Biochem. Biophys. Res. Commun. 263 (1999), 775-779. [62] F. Moro, V. Fernandez-Saiz, and A. Muga, Protein Sci. 15 (2006), 223-233. [63] H. Fabian, D. Chapman, and H.H. Mantsch, New trends in isotope-edited infrared spectroscopy, in: Infrared spectroscopy of biomolecules, eds. H.H. Mantsch and D. Chapman, Wiley-Liss, New York, 1996, 341-352. [64] F. Von Germar, A. Galn, O. Llorca, J.L. Carrascosa, J.M. Valpuesta, W. Mntele, and A. Muga, J. Biol. Chem. 274 (1999), 5508-5513. [65] C. Raimbault, R. Buchet, and C. Vial, Eur. J. Biochem. 240 (1996), 134-142. [66] C. Raimbault, F. Besson, and R. Buchet, Eur. J. Biochem. 244 (1997), 343-351.
76
[67] E.M. White, A.R. Holland, and G. MacDonald, Biochemistry 47 (2008), 84-91. [68] V. Cepus, A.J. Scheidig, R.S. Goody, and K. Gerwert, Biochemistry 37 (1998), 10263-10271. [69] X. Du, H. Frei, and S.-H. Kim, J. Biol. Chem. 275 (2000), 8492-8500. [70] H. Cheng, S. Sukal, H. Deng, T.S. Leyh, and R. Callender, Biochemistry 40 (2001), 4035-4043. [71] B.C. Butler, R.H. Hanchett, H. Rafailov, and G. MacDonald, Biophys. J. 82 (2002), 2198-2210. [72] J. Backmann, H. Fabian, and D. Naumann, FEBS Lett. 364 (1995), 175-178. [73] D. Reinstdler, H. Fabian, J. Backmann, and D. Naumann, Biochemistry 35 (1996), 15822-15830. [74] R.B. Dyer, F. Gai, W.H. Woodruff, R. Gilmanshin, and R.H. Callender, Acc. Chem. Res. 31 (1998), 709-716. [75] R. Gilmanshin, S. Williams, R.H. Callender, W.H. Woodruff, and R.B. Dyer, Biochemistry 36 (1997), 15006-15012. [76] S. Williams, T.P. Causgrove, R. Gilmanshin, K.S. Fang, R.H. Callender, W.H. Woodruff, and R.B. Dyer, Biochemistry 35 (1996), 691-697. [77] J. Wang and M.A. El-Sayed, Biophys. J. 76 (1999), 2777-2783. [78] C.M. Phillips, Y. Mizutani, and R.M. Hochstrasser, Proc. Natl. Acad. Sci. USA 92 (1995), 7292-7296. [79] A.P. Ramajo, S.A. Petty, and M. Volk, Chem. Phys. 323 (2006), 11-20. [80] G. Panick, R. Malessa, R. Winter, G. Rapp, K.J. Frye, and C.A. Royer, J. Mol. Biol. 275 (1998), 389402. [81] G. Panick and R. Winter, Biochemistry 39 (2000), 1862-1869. [82] A. Barth, K. Hauser, W. Mntele, J.E.T. Corrie, and D.R. Trentham, J. Am. Chem. Soc. 117 (1995), 10311-10316. [83] A. Barth, J.E.T. Corrie, M.J. Gradwell, Y. Maeda, W. Mntele, T. Meier, and D.R. Trentham, J. Am. Chem. Soc. 119 (1997), 4149-4159. [84] V. Cepus, C. Ulbrich, C. Allin, A. Troullier, and K. Gerwert, Methods Enzymol. 291 (1998), 223-245. [85] A. Barth, Time-resolved IR spectroscopy with caged compounds: An introduction, in: Dynamic studies in biology: Phototriggers, photoswitches and caged biomolecules, eds. M. Goeldner and R.S. Givens, Wiley-VCH, Weinheim, 2005, 369-399. [86] V. Jayaraman, IR spectroscopy with caged compounds: Selected applications, in: Dynamic studies in biology: Phototriggers, photoswitches and caged biomolecules, eds. M. Goeldner and R.S. Givens, Wiley-VCH, Weinheim, 2005, 400-410. [87] A. Barth and C. Zscherp, FEBS Lett. 477 (2000), 151-156. [88] D. Moss, E. Nabedryk, J. Breton, and W. Mntele, Eur. J. Biochem. 187 (1990), 565-572. [89] E. Hamacher, J. Kruip, M. Rgner, and W. Mntele, Spectrochim. Acta A 52 (1996), 107-121. [90] M. Leonhard and W. Mntele, Biochemistry 32 (1993), 4532-4538. [91] M. Bauscher, M. Leonhard, D.A. Moss, and W. Mntele, Biochim. Biophys. Acta 1183 (1993), 59-71. [92] P. Hellwig, B. Rost, U. Kaiser, C. Ostermeier, H. Michel, and W. Mntele, FEBS Lett. 385 (1996), 5357. [93] P. Hellwig, J. Behr, C. Ostermeier, O.M. Richter, U. Pfitzner, A. Odenwald, B. Ludwig, H. Michel, and W. Mntele, Biochemistry 37 (1998), 7390-7399. [94] J. Behr, P. Hellwig, W. Mntele, and H. Michel, Biochemistry 37 (1998), 7400-7406. [95] P. Hellwig, C. Ostermeier, H. Michel, B. Ludwig, and W. Mntele, Biochim. Biophys. Acta 1409 (1998), 107-112. [96] P. Hellwig, T. Soulimane, G. Buse, and W. Mntele, FEBS Lett. 458 (1999), 83-86. [97] B. Rost, J. Behr, P. Hellwig, O.M. Richter, B. Ludwig, H. Michel, and W. Mntele, Biochemistry 38 (1999), 7565-7571. [98] J. Behr, H. Michel, W. Mntele, and P. Hellwig, Biochemistry 39 (2000), 1356-1363. [99] E.D. Dodson, X.J. Zhao, W.S. Caughey, and C.M. Elliott, Biochemistry 35 (1996), 444-452. [100] R.B. Gennis, FEBS Lett. 555 (2003), 2-7. [101] G. Tollin, J. Bioenerg. Biomembr. 27 (1995), 303-309. [102] R. Traber, H.E.A. Kramer, and P. Hemmerich, Biochemistry 21 (1982), 1687-1693. [103] J. Contzen and C. Jung, Biochemistry 38 (1999), 16253-16260. [104] M. Lbben, A. Prutsch, B. Mamat, and K. Gerwert, Biochemistry 38 (1999), 2048-2056. [105] Y. Yamazaki, H. Kandori, and T. Mogi, J. Biochem. 126 (1999), 194-199. [106] Y. Yamazaki, H. Kandori, and T. Mogi, J. Biochem. 125 (1999), 1131-1136. [107] M. Liu and A. Barth, J. Biol. Chem. 278 (2003), 10112-10118. [108] A. Barth, F. von Germar, W. Kreutz, and W. Mntele, J. Biol. Chem. 271 (1996), 30637-30646. [109] R. Brudler, R. Rammelsberg, T.T. Woo, E.D. Getzoff, and K. Gerwert, Nature Struct. Biol. 8 (2001), 265-270. [110] D. Reinstdler, H. Fabian, and D. Naumann, Proteins 34 (1999), 303-316. [111] A. Troullier, D. Reinstdler, Y. Dupont, D. Naumann, and V. Forge, Nature Struct. Biol. 7 (2000), 7886.
77
[112] A. Barth, W. Mntele, and W. Kreutz, J. Biol. Chem. 272 (1997), 25507-25510. [113] R.S. Chittock, S. Ward, A.-S. Wilkinson, P. Caspers, B. Mensch, M.G.P. Page, and C.W. Wharton, Biochem. J. 338 (1999), 153-159. [114] A. Troullier, K. Gerwert, and Y. Dupont, Biophys. J. 71 (1996), 2970-2983. [115] F. Scheirlinckx, R. Buchet, J.-M. Ruysschaert, and E. Goormaghtigh, Eur. J. Biochem. 268 (2001), 3644-3653. [116] S. Krimm and Y. Abe, Proc. Natl. Acad. Sci. USA 69 (1972), 2788-2792. [117] P. Hamm, M. Lim, W.F. DeGrado, and R.M. Hochstrasser, Proc. Natl. Acad. Sci. USA 96 (1999), 2036-2041. [118] H. Torii and M. Tasumi, J. Raman Spectrosc. 229 (1998), 81-86. [119] H. Torii, T. Tatsumi, T. Kanazawa, and M. Tasumi, J. Phys. Chem. B 102 (1998), 309-314. [120] J.R. Parrish and E.R. Blout, Biopolymers 11 (1972), 1001-1020. [121] S.T.R. Walsh, R.P. Cheng, W.W. Wright, D.O.V. Alonso, V. Daggett, J.M. Vanderkooi, and W.F. DeGrado, Protein Sci. 12 (2003), 520-531. [122] N.A. Nevskaya and Y.N. Chirgadze, Biopolymers 15 (1976), 637-648. [123] S.Y. Venyaminov and N.N. Kalnin, Biopolymers 30 (1990), 1259-1271. [124] Y.N. Chirgadze, B.V. Shestopalov, and S.Y. Venyaminov, Biopolymers 12 (1973), 1337-1351. [125] S.Y. Venyaminov and N.N. Kalnin, Biopolymers 30 (1990), 1243-1257. [126] Y.N. Chirgadze, O.V. Fedorov, and N.P. Trushina, Biopolymers 14 (1975), 679-694. [127] A. Barth, Infrared spectroscopy, in: Methods in protein structure and stability analysis: vibrational spectroscopy, eds. V.N. Uversky and E.A. Permyakov, Nova Science Publishers, New York, 2007, 69151. [128] M. Liu and A. Barth, J. Biol. Chem. 279 (2004), 49902-49909. [129] W. Mntele, Infrared and Fourier-transform infrared spectroscopy, in: Biophysical techniques in photosynthesis, eds. J. Amesz and A.J. Hoff, Kluwer Academic Publishers, Dordrecht, 1996, 137-160. [130] D.D. Schlereth and W. Mntele, Biochemistry 32 (1993), 1118-1126. [131] D.D. Schlereth, V.M. Fernandez, and W. Mntele, Biochemistry 32 (1993), 9199-9208. [132] D.D. Schlereth and W. Mntele, Biochemistry 31 (1992), 7494-7502. [133] A. Barth, Prog. Biophys. Mol. Biol. 74 (2000), 141-173. [134] C. Zscherp, R. Schlesinger, J. Tittor, D. Oesterhelt, and J. Heberle, Proc. Natl. Acad. Sci. USA 96 (1999), 5498-5503. [135] J.J. Katz, L.L. Shipman, T.M. Cotton, and T.R. Janson, Chlorophyll aggregation: coordination interactions in chlorophyll monomers, dimers, and oligomers, in: The porphins, ed. D. Dolphin, Vol. 5, Academic Press, New York, 1978, 401-458. [136] J.J. Katz, R.C. Dougherty, and L.J. Boucher, Infrared and nuclear magnetic resonance spectroscopy of chlorophyll, in: The chlorophylls, eds. L.P. Vernon and G.R. Seely, Academic Press, New York, 1966, 185-251. [137] M. Lutz and W. Mntele, Vibrational spectroscopy of chlorophylls, in: Chlorophylls, ed. H. Scheer, CRC Press, Boca Raton, Florida, 1991, 855-902. [138] W.T. Potter, M.P. Tucker, R.A. Houtchens, and W.S. Caughey, Biochemistry 26 (1987), 4699-4707. [139] P.K. Slaich, W.U. Primrose, D.H. Robinson, C.W. Wharton, A.J. White, K. Drabble, and G.C.K. Roberts, Biochem. J. 288 (1992), 167-173. [140] M. Liu, M. Krasteva, and A. Barth, Biophys. J. 89 (2005), 4352-4363. [141] K. Gerwert and F. Siebert, EMBO J. 5 (1986), 805-811. [142] A. Barth, J. Biol. Chem. 274 (1999), 22170-22175. [143] A. Barth and N. Bezlyepkina, J. Biol. Chem. 279 (2004), 51888-51896. [144] S.W. Englander and N.R. Kallenbach, Quart. Rev. Biophys. 4 (1984), 521-655. [145] P.I. Haris, G.T. Robillard, A.A. Van Dijk, and D. Chapman, Biochemistry 31 (1992), 6279-6284. [146] S. Sonar, C.P. Lee, M. Coleman, N. Patel, X. Liu, T. Marti, H.G. Khorana, U.L. RajBhandary, and K.J. Rothschild, Struct.Biol. 1 (1994), 512-517. [147] J.L. Spudich, Struct.Biol. 1 (1994), 495-496. [148] M.A. Palafox, Trends Appl. Spectrosc. 2 (1998), 37-57. [149] R.J. Meier, Vibrational Spectroscopy 43 (2007), 26-37. [150] J. Fisher, J.G. Belasco, S. Khosla, and J.R. Knowles, Biochemistry 19 (1980), 2895-2901. [151] A.J. White, K. Drabble, S. Ward, and C.W. Wharton, Biochem. J. 287 (1992), 317-323. [152] D. Thoenges and A. Barth, J. Biomol. Screen. 7 (2002), 353-357. [153] K. Karmali, A. Karmali, A. Teixeira, and M.J.M. Curto, Anal. Biochem. 333 (2004), 320-327. [154] R. Pacheco, M.L.M. Serralheiro, A. Karmali, and P.I. Haris, Anal. Biochem. 322 (2003), 208-214. [155] R. Pacheco, A. Karmali, M.L.M. Serralheiro, and P.I. Haris, Anal. Biochem. 346 (2005), 49-58. [156] W. Wright and J.M. Vanderkooi, Biospectroscopy 3 (1997), 457-467. [157] H. Takeuchi, H. Murata, and I. Harada, J. Am. Chem. Soc. 110 (1988), 392-397.
78
[158] W. Hasselbach and M. Makinose, Biochem. Z. 333 (1961), 518-528. [159] J.P. Andersen, Biochim. Biophys. Acta 988 (1989), 47-72. [160] C. Toyoshima and G. Inesi, Annu. Rev. Biochem. 73 (2004), 269-292. [161] H.-J. Apell, Bioelectrochemistry 63 (2004), 149-156. [162] J.V. Mller, P. Nissen, T.L.-M. Srensen, and M. le Maire, Curr. Opin. Struct. Biol. 15 (2005), 387393. [163] A. Barth, Spectroscopy (2008), in press. [164] W. Hasselbach and W. Waas, Ann. N. Y. Acad. Sci. 402 (1982), 459-469. [165] A. Barth, W. Mntele, and W. Kreutz, Biochim. Biophys. Acta 1057 (1991), 115-123. [166] R. Buchet, I. Jona, and A. Martonosi, Biochim. Biophys. Acta 1104 (1992), 207-214. [167] R. Buchet, I. Jona, and A. Martonosi, Biochim. Biophys. Acta 1069 (1991), 209-217. [168] H. Georg, A. Barth, W. Kreutz, F. Siebert, and W. Mntele, Biochim. Biophys. Acta 1188 (1994), 139150. [169] M. Liu and A. Barth, Biophys. J. 85 (2003), 3262-3270. [170] L. De Meis and A. Vianna, Annu. Rev. Biochem. 48 (1979), 275-292. [171] T.L.-M. Srensen, J.V. Mller, and P. Nissen, Science 304 (2004), 1672-1675. [172] J. Andersson, K. Hauser, E.-L. Karjalainen, and A. Barth, Biophys. J. in press (2008). [173] W.P. Jencks, Biosci. Rep. 15 (1995), 283-287. [174] J.P. Andersen, Biosci. Rep. 15 (1995), 243-261. [175] C. Toyoshima and H. Nomura, Nature 418 (2002), 605-611. [176] C. Toyoshima, H. Nomura, and T. Tsuda, Nature 432 (2004), 361-368. [177] M. Liu and A. Barth, Biopolymers (Biospectroscopy) 67 (2002), 267-270. [178] C. Toyoshima and T. Mizutani, Nature 430 (2004), 529-535. [179] D.L. Stokes and N.M. Green, Annu. Rev. Biophys. Biomol. Struct. 32 (2003), 445-468. [180] C. Allin and K. Gerwert, Biochemistry 40 (2001), 3037-3046. [181] H. Cheng, S. Sukal, R. Callender, and T.S. Leyh, J. Biol. Chem. 276 (2001), 9931-9935. [182] A. Barth, W. Kreutz, and W. Mntele, Biochim. Biophys. Acta 1194 (1994), 75-91. [183] X. Yu, L.N. Hao, and G. Inesi, J. Biol. Chem. 269 (1994), 16656-16661. [184] C. Peinelt and H.J. Apell, Biophys. J. 82 (2002), 170-181. [185] F. Tadini-Buoninsegni, G. Bartolommei, M.R. Moncelli, R. Guidelli, and G. Inesi, J. Biol. Chem. 281 (2006), 37720-37727. [186] W.P. Jencks, Methods Enzymol. 171 (1989), 145-164. [187] A. Barth and W. Mntele, Biophys. J. 75 (1998), 538-544. [188] A. Barth, Biopolymers (Biospectroscopy) 67 (2002), 237-241. [189] T. Kanazawa and P.D. Boyer, J. Biol. Chem. 248 (1973), 3163-3172. [190] J. Andersson and A. Barth, Biopolymers 82 (2006), 353-357. [191] T.C. Bruice and F.C. Lightstone, Acc. Chem. Res. 32 (1999), 127-136. [192] I.D. Brown, The chemical bond in inorganic chemistry. The bond valence model. Oxford University Press, Oxford, 2002. [193] W.P. Jencks, Methods Enzymol. 6 (1963), 914-928.
79
Ultrafast 2D-IR Vibration Echo Spectroscopy of Proteins

Haruto ISHIKAWA, Seongheun KIM, Ilya J. FINKELSTEIN, and Michael. D. FAYER1 Department of Chemistry Stanford University, Stanford, CA 94305-5080, USA
Abstract. The investigation of protein structural dynamics on short time scales (100 fs to 100 ps) using ultrafast two dimensional (2D-IR) vibrational echo spectroscopy is presented. Under thermal equilibrium conditions, a proteins structure is constantly fluctuating among conformations associated with different positions on the broad rough minimum on the free energy landscape. Although different conformational substates may be apparent in a vibrational absorption spectrum, linear IR absorption spectra cannot provide information on structural dynamics because dynamical information is masked by inhomogeneous broadening of the lineshapes. 2D-IR vibrational echo spectroscopy makes structural fluctuations a direct experimental observable. Changes in structure manifest themselves through the time evolution of the 2D-IR line shape (spectral diffusion). Here details of the experimental method including the pulse sequence, heterodyne detection to provide full phase information, and the extraction of the molecular dynamics from 2D-IR spectra are outlined. The method and the nature of the information that can be obtained are illustrated with four examples: the influence of mutations on myoglobin dynamics, differences between the dynamics of neuroglobin and myoglobin, the effect of the disulfide bond in neuroglobin on its structural dynamics, and how substrate binding to the enzyme horseradish peroxidase influences its structural fluctuations. Keywords. Two-dimensional infrared; vibrational echo; protein dynamics; heme proteins; neuroglobin; horseradish peroxidase; myoglobin
Introduction Proteins and other biological molecules are dynamic structures, and their dynamics are intimately related to their function. For example, the diffusion of a ligand through a protein like myoglobin or hemoglobin to the active site is made possible by structural fluctuations[1,2]. Even under thermal equilibrium conditions, proteins are never at rest. A folded protein in a particular structure occupies a minimum on the free energy landscape[3,4]. There may be more than one minimum. Each minimum corresponds to a different conformation called a conformational substate (see Figure
1
Corresponding Author: E-mail: fayer@stanford.edu
80
H. Ishikawa et al. / Ultrafast 2D-IR Vibration Echo Spectroscopy of Proteins
1)[3,4]. Within the minimum for a particular conformational substate, the free energy landscape is broad and rough with many local minima separated by low barriers with a wide range of barrier heights. Cryochemical experiments confirm that a protein in a substate has a variety of structures that are represented in Figure 1 by the shallow local minima in a particular substate valley[5,6]. At normal biological temperatures under thermal equilibrium conditions there is sufficient thermal energy to produce transitions among these minima that are responsible for protein structural fluctuations. As indicated in Figure 1, there can be hierarchies of barrier heights giving rise to structural fluctuations on different time scales. The linear IR absorption spectrum of a protein can provide information on aspects of protein structure. For example, CO bound to the active site of myoglobin (MbCO) displays three CO peaks in the IR absorption spectrum of the CO stretch region (~1950 cm-1)[7]. These correspond to three substates of the protein that produce distinct configurations of the distal histidine (His64)[8-12]. However, the IR absorption spectrum does not provide information on protein dynamics. The absorption bands in MbCO and other proteins are inhomogeneously broadened because of the large number of structural configurations associated with the energy landscape surrounding the minimum for each substates (see Figure 1). The width of the absorption band reflects the distribution of protein configurations but does not give information on interconversion among configurations. Ultrafast two dimensional infrared (2D-IR) vibrational echo spectroscopy can obtain information on protein dynamics and structure that cannot be obtained from the IR absorption spectrum alone. The ultrafast vibrational echo method was first applied as a one dimension experiment in 1993[13] and first applied to proteins in 1996[14-16]. Since these early experiments[17,18], vibrational echo spectroscopy has made major advances in both the nature of the technique and the range of applications. It has become a full two dimensional spectroscopy akin to 2D NMR[19] but operating on time scales many orders of magnitude faster and directly examining the structural degrees of freedom of complex molecular systems[20,21].
folded protein
substates
Energy landscape, low barriers, fast fluctuations. Energy landscape, moderate barriers, moderately fast fluctuations.
Figure 1. Schematic illustration of a conformational energy landscape showing the folded protein well with two substate minima and the energy landscape structure about the minima. The hierarchies of barrier height give rise to structural fluctuations on different time scales.
81
Ultrafast 2D-IR vibrational echo spectroscopy has been applied to a wide variety of problems. Because of the nature of the method, it a useful tool for the study of problems involving rapid dynamics under thermal equilibrium conditions in condensed phases. Such problems are ubiquitous in nature and difficult to study by other means. 2D-IR experiments can have temporal resolution of < 50 fs, which is sufficiently fast to study the fastest chemical processes. The vibrational excitation associated with 2D-IR experiments produce negligible perturbations of molecular systems. Vibrational excitation does not change the chemical properties of the samples, in contrast to electronic excitation that may produce substantial perturbations. 2D-IR experiments can also be useful as a tool for chemical structural analysis by revealing the relationship among different mechanical degrees of freedom of a molecular or biomolecular system. 2D-IR vibrational echo experiments have been applied to study fast chemical exchange reactions and solution dynamics[22-24], water dynamics[25-30], hydrogen network evolution[31], intramolecular vibrational energy relaxations[32], and of particular interest here protein structure and dynamics[12,20,33-53]. The pulse sequence in 2D-IR vibrational echo experiment induces and then probes the coherent evolution of excitations (vibrations) of a molecular system. The signal is generated by a sequence of three ultrashort IR pulses tuned to the vibrational transitions of interest. The first pulse in the sequence causes vibrational modes of an ensemble of molecules to oscillate initially all with the identical phase. The second pulse in some sense labels the frequencies of the molecules initially excited by the first pulse. During the period between the second and third pulses, structural changes in the system cause the frequencies of the labeled molecules to change. The third pulse begins the read out of the molecular frequencies and generates the observable signal, a fourth pulse, the vibrational echo. The characteristic spectrum obtained from observing the frequencies, intensities, and phases of the vibrational echo and Fourier transformation into the frequency domain, is sensitive to changes in environments of individual molecules during the experiment, even if the aggregate populations in distinct environments do not change. For example, the structural fluctuations of a protein or the formation and dissociation of molecular complexes under thermal equilibrium conditions can be observed. In this chapter, the theoretical background, methodology, and recent progress of the 2D-IR vibrational echo measurements for heme proteins are described. For heme proteins, 2D-IR vibrational echo experiments use the heme-ligated CO vibration as a direct sensor of protein dynamics[11,12,20,36,47-49,54-57]. The linear IR absorption spectrum of the CO stretching mode of heme protein generally displays several bands that reflect structural differences, i.e., distinct structural substates[7,58]. While the linear IR absorption spectrum can not provide information on a proteins structural dynamics, the time evolution of the 2D-IR spectra (spectral diffusion) of the CO bands reveals the fast protein structural fluctuations of the substates. The heme-ligated CO absorption bands are observed in 1,900cm-1 to 2,000cm-1 range, which is separate from other absorbing group. Three protein systems will be discussed. Myoglobin and myoglobin mutants will be used to demonstrate how small changes in the amino acid sequence can have significant effects on protein dynamics as sensed by CO bound at the active site[51]. Neuroglobin, a recently discovered heme protein found in the brain and nerve tissue[59], is compared to myoglobin, and it is used to study the influence that removing a disulfide bond has on protein dynamics[51,53]. Finally, the enzyme horseradish peroxidase is studied with and without a bound substrate. It is
82
demonstrated that substrate binding makes substantial changes in the enzymes dynamics[36].
1. Methodology The ultrafast 2D-IR vibrational echo measurement involves three femtosecond IR pulses that are tuned to the frequency of the vibrational modes of interest[60-62]. Because of the very short pulses, they have a broad bandwidth that can simultaneously excite a number of vibrational modes or a broad spectral feature. Femtosecond IR pulses employed in the experiments are generated using a Ti:Sapphire regeneratively amplified laser/optical parametric amplifier (OPA) system[63]. The spectral widths of the heme protein-CO transition discussed below are relatively narrow (10 20 cm-1) and the spread in the transition frequencies that occur in the various systems is ~100 cm-1. Therefore, the pulse durations in the IR are tailored to produce the appropriate band width for the experiments. The output of the regenerative amplifier is ~100 fs and produces transform limited ~0.5 mJ pulses centered at ~800 nm at 1 kHz repetition rate. These are used to pump an IR OPA. The bandwidth and pulse duration of IR pulses are 150 cm-1 and 100 fs, respectively. A 2D-IR vibrational echo spectroscopy setup is illustrated schematically in Figure 2. The three pulses have wave vectors k1, k2, and k3 with variable delay time W between the pulses 1 and 2 with wave vectors k1 and k2 and with variable delay time Tw between pulses 2 and 3 with wave vectors k2 and k3. The vibrational echo pulse is detected in the ke = k2 + k3 k1 phase-matched direction. When the vibrational echo pulse is sent directly into an IR detector, its intensity is measured but phase information is lost. To perform Fourier transforms from the time domain into the frequency domain, phase relationships are necessary. To obtain the vibrational echo signal E-field rather than detecting the intensity, the echo pulse is overlapped with a fifth pulse called the local oscillator (LO). The function of the LO is to phase resolve and optically heterodyne amplify the vibrational echo signal. In the heterodyne detected vibrational echo experiments, the vibrational echo pulse, which is overlapped with LO pulse, is passed through a monochromator and then detected with a 32-element HgCdTe (MCT) IR array detector. The following is a qualitative description of the vibrational echo experiment. The first pulse excites the molecules to a coherent superposition of the ground state (0) and first excited state (1) with all of the vibrational oscillators initially oscillating in phase at their initial frequencies. The phase relationships among the oscillators decay quickly because of inhomogeneous broadening of the spectral line (the range of transition frequencies across the spectroscopic line) with additional contributions from fast fluctuations of the transition frequencies caused by structural dynamics of the system. This initial loss of phase relationships is called the free induction decay (FID). The second pulse transfers the initial coherent superposition states of each molecule into a population state in either the 0 or 1 states. Because of structural evolution of the system during the population period, Tw, the molecular oscillators (the CO stretch for the experiments discussed below) undergo frequency shifts, called spectral diffusion, which cause molecules to lose memory of their initial frequencies. The third pulse again generates coherent superposition states of the oscillators. Initially, the oscillators are not in phase, but the pulse sequence initiates a rephasing process. If some memory of the initial frequency is retained, the vibrational echo
83
pulse is generated after the third pulse at a time W, because the vibrational oscillators are again oscillating in phase. Decay of memory of the initial frequencies of the molecular oscillators caused by structural evolution of the system (spectral diffusion) causes the 2D spectrum to change shape. Even if all of the memory of the initial frequencies has been lost because of rapid structural evolution, there is still a vibrational echo signal, but the 2D spectrum is symmetrical, and its width and shape reflect the width and shape of the absorption spectrum. Observation of the vibrational echo is limited by the vibrational life time (T1). Decay to the ground state of the excited vibrations causes the amplitude of the signal to decay. If the vibrational lifetime causes the vibrational echo signal to decay to zero before spectral diffusion is complete (all structures have been sampled), the 2D spectrum will not reach its symmetrical shape. In short, the first pulse labels the initial structures of the species in the sample and initiates the first coherence period The second pulse ends the first coherence period W and starts clocking the waiting time during which the frequency-labeled molecular oscillators experience structural dynamics that can cause them to evolve to different frequencies. The third pulse ends the waiting period Tw and begins a third period of length W, which ends with emission of the vibrational echo pulse. The echo signal reads out information about the final structures experienced by the labeled oscillators. A 2D vibrational echo spectrum is obtained with the initial labeled frequencies as one axis (ZW) and the final frequencies of the sample as the other axis (Zm).
Figure 2. Schematic layout of the 2D-IR vibrational echo experiment. The three mid-IR pulses have wave vectors k1, k2, and k3 with variable delay time W between the pulses 1 and 2 (k1 and k2) and with variable delay time Tw between pulses 2 and 3 (k2 and k3). The vibrational echo signal is detected in the phase-matched direction. The emitted vibrational echo pulse and the local oscillator pulse are combined to enable the measurement of both intensity and phase information.
84
An example is given in Figure 3 for the CO adduct of L29F mutant Mb[51]. Figure 3A shows the linear FTIR absorption spectra of the CO stretching mode bound to the ferrous heme iron in the L29F mutant. The single peak indicates that the L29FCO mutant exists in a single conformational substate. In the 2D vibrational echo spectrum, there are two frequency axes, which require two Fourier transforms to the 2D frequency spectrum. The vibrational echo signal is measured as a function of one frequency variable, Zm, and two time variables, W and Tw. One of the two Fourier transform is done by the monochromator. The spectrum of the combined vibrational echo-local oscillator wave packet resolves it into its frequency components. This Fourier transform provides the vertical axis, Zm (m for monochromator), of the spectrum, corresponding to the time between the third pulse and the vibrational echo pulse. The Zm axis is the axis corresponding to the frequencies of the vibrational echo emission. The other axis is obtained by scanning W. As W is scanned, the vibrational echo pulses moves in time relative to the fixed local oscillator, producing an interferogram. There is one such interferogram for each frequency Zm. These interferograms are numerically Fourier transformed to provide the data along the ZW axis. The ZW axis corresponds to the frequency of the first interaction of the radiation field (first pulse) with the molecular oscillators. Two-dimensional vibration echo spectra are recorded as a function of Tw. The amplitude is depicted as a function of both ZW and Zm that correspond to the Z1 and Z3 axes, respectively in 2D-NMR. Details of the method including phase error corrections have been presented[63,64]. The Tw dependent spectral changes in 2D-IR provide direct information on the time evolution of the protein through the influence of the structural changes on the frequency of the CO vibrational mode.
A
absorbance (norm)
1.0 0.8 0.6 0.4 0.2 0.0
B
1940
Tw = 0.25 ps
Zm (cm-1)
1930 1920 1910 1900

1900 1910 1920 1930 1940
1900 1910 1920 1930 1940 1950 1960
frequency (cm-1)
ZW (cm-1)
Figure 3. 2D-IR measurement on the CO stretch of CO bound to the L29F Mb mutant. (A) The linear FT-IR spectrum of L29FCO. (B) 2D-IR vibrational echo spectrum of L29FCO at Tw = 0.25 ps. There is a positive going peak (labeled +) on the diagonal and a negative going peak (labeled ) off diagonal that correspond vibrational echo emission at the 0-1 and 1-2 vibrational transitions, respectively.
85
The 2D-IR vibrational echo spectrum of L29FCO at Tw = 0.5ps is shown in Figure 3B. There is a positive going peak on the diagonal and a negative going peak off diagonal, which correspond to vibrational echo emission at the 0-1 and 1-2 vibrational transitions, respectively. The off-diagonal 1-2 band is shifted along the Zm axis by the vibrational anharmonicity of the CO stretching mode. The band on the diagonal has ZW = Zm. The frequency of the initial excitation (first pulse) is equal to frequency of the echo emission. The off diagonal 1-2 peak arises as follows. The first pulse excites a coherent superposition state of the 0-1 transition. The second pulse produces a population in the 1 state. The third pulse couples the 1 state to the 2 state, which produces a coherent superposition of 1 and 2 vibrational states, and leads to vibrational echo emission at the 1-2 transition frequency. The band is off-diagonal because the first interaction is the ZW frequency of the 0-1 transition, but the echo emission is at the Zm frequency of the 1-2 transition resulting in the shift to lower frequency along Zm by the vibrational anharmonicity, ~25 cm-1. The 1-2 band is negative going, while the 0-1 band is positive going. Because the dynamical information obtained from the off-diagonal 1-2 bands is the same as that obtained from the 0-1 bands, only the 0-1 vibrational transition peaks are analyzed below.
2. Data Analysis Figure 4A shows an example of 2D-IR spectra for CO bound to the L29F mutant of Mb at several values of Tw. Only the 0-1 transition region is shown. The peak position is located on the diagonal at (ZW Zm) = (1,933 cm-1, 1933 cm-1). The position of the peak is time-independent. However, the band shape changes. As Tw increases, the band go from highly elongated along the diagonal to less elongated and increasingly broad along the ZW axis. In the long time limit, the band would become round. The change in band shape is caused by spectral diffusion arising from protein structural fluctuations. To analyze the time evolution of the 2D-IR vibrational echo spectrum, the inverse of the center line slope (CLS) is used[21,51,53,65]. A slice through the 2D spectrum at a particular Zm value is projected onto the ZW axis to give a spectrum. The peak position on the ZW axis is determined. The result is a point with coordinates (Zm, ZW). Many such slices are taken and the set of (Zm, ZW) points forms a line, the center line. Examples of center lines are shown in Figure 4A as the white dotted lines in the panels with Tw = 0.5 and 32 ps. The slope of the center line changes as Tw increases and spectral diffusion makes the band more symmetrical. It has been demonstrated theoretically, that the change in the inverse of the slope of the center line (referred to as the CLS) is directly related to the underlying dynamics of the system[21,65]. In the absence of a homogeneous broadening component (see below), the initial slope would be that of a line at 45q, that is, a slope of 1. At sufficiently long time, the shape of the 2D spectrum is symmetrical, and the center line is vertical with an infinite slope. Therefore, the inverses of the center line slopes range from 1 to 0. In Figure 4A, it can be seen that for Tw = 32 ps the slope of the center line is steeper than at Tw = 0.5 ps. The Tw dependent CLS for CO bound to L29F mutant Mb are shown in Figure. 4B. A quantitative description of the amplitudes and time scales of frequency fluctuations of a vibrational oscillator is provided by the frequency-frequency correlation function (FFCF)[26,61,66,67]. The FFCF connects the experimental
86
observables to the underlying dynamics in the sample. The FFCF is a joint probability distribution that the frequency has a certain initial value at t = 0 and another value at t. The CLS method is used for extracting the FFCF from the Tw dependence of the 2D-IR spectra[21,65]. Once the FFCF is known, all linear and non-linear optical experimental observables can be calculated using time-dependent diagrammatic perturbation theory. Conversely, the FFCF can be extracted from 2D-IR vibrational echo spectra with additional input from the linear FTIR absorption spectrum, including the Tw-independent homogeneous component. A multiexponential form of the FFCF, C(t), is used to model the multi-time scale dynamics of the protein structural fluctuations and has been found to reproduce the influence of structural dynamics on the CO frequency in heme proteins[11,51,55,56,68]. The FFCF has the form
C (t )
' e
i 1
2 t /W i i
'2 s .
(1)
The 'i and Wi terms are the amplitudes and correlation times, respectively, of the frequency fluctuations induced by protein structural dynamics. Wi reflects the time scale of a set of structural fluctuations and the 'i is the range of CO frequencies sampled due to the structural fluctuations. The experimental time window is several times the vibrational life time, T1, because lifetime decay of the excited vibrational state reduces the signal to zero. The 2D-IR vibrational echo experiment is sensitive to fluctuations a few times longer than this window, i.e., a hundred picoseconds or more depending on the sample, because some portion of slower fluctuations will occur in the experimental window if their time scale is not too slow[69]. Protein structural dynamics that are sufficiently slow will appear as static inhomogeneous broadening, which is reflected in C(t) by 's, a static term. In obtaining the FFCF from the data, the 'i and the Wi are determined. However, for ultrafast dynamics in the motionally narrowed limit ('W < 1), only the product, '2W = 1/T2* or ** = (ST2*)-1, can be obtained[56,70]. T2* is called the pure dephasing time, which gives rise to the pure homogeneous linewidth, **. The total dephasing time T2 is
1 T2
1 1 1 . T2 2T1 3Tor
(2)
where ** = (ST2*)-1 is the Lorentzian homogeneous line width. T1 and Tor are the vibrational life time and orientational relaxation time, respectively. Because the rotational diffusion of the protein is very slow relative to the vibrational life time, the orientational relaxation contribution can be neglected. T2 is determined from the CLS with use of the linear absorption spectrum, and T1 is obtained from the independent IR pump-probe experiments. Therefore, the pure dephasing time, T2*, is obtained using 1/T2 = 1/T2* + 1/2T1. The homogeneous contribution is not dependent on Tw. It manifests itself as a deviation from the CLS being equal to 1 at Tw = 0. The CLS data points in Figure 4B are fit to a biexponential function. A biexponential without a static term, 's, is sufficient to describe the Tw dependent protein dynamics sensed by CO bound at the active site of the L29F mutant. The homogeneous contribution is obtained from the Tw = 0 value of the CLS and the linear absorption spectrum as described in detail previously[21,65]. The fit is shown as the line in Figure 4B. The FFCF parameters obtained from 2D-IR and linear FT-IR spectra, T2*, and T1 values are given in Table 1. The FFCF parameters delineate that
87
A
Zm (cm-1)
1950 1940 1930
Tw = 0.5
Tw = 4
0.7 0.6 0.5
CLS
1920
Tw = 16
Zm (cm-1)
1940 1930 1920 1920 1930 1940
Tw = 32
0.4 0.3 0.2
ZW (cm-1)
1920 1930 1940 1950
0.1
ZW (cm-1)
10
Tw (ps)
20
30
Figure 4. Time dependent spectral diffusion of L29F Mb mutant. (A) A series of 2D-IR spectra for L29FCO at several Tw values. Only the 0-1 transition is shown, and the white dashed lines are the center lines (see text). (B) Tw dependent CLS data (circles) for L29FCO. The solid curve is calculated with the FFCF obtained from the CLS data.
the L29FCO band has a fast (1.7 ps) decay followed by slower decay of 66 ps. Because L29FCO does not have a static term, all possible structures that give rise to the inhomogeneous broadened CO absorption band are sampled in ~200 ps. This is an interesting and important result. The equilibrium structural fluctuations of L29F have sampled all structural configurations about the folding minimum in ~200 ps. The results demonstrate that there are no slower components so long as the protein remains in the folded substates. As illustrated in Figure 1, the very fast fluctuation arise from transitions between minima with very small barrier heights while the slower fluctuations involve transitions between a set of minima separated by higher barriers.
Table 1. Experimental parameters for the proteins studied

protein Mb L29F Mb H64V wild-type Ngb N3 wild-type Ngb N0 reduced Ngb N3 reduced Ngb N0 HRP free red HRP BHA T2* (ps) 4.8 7.7 5.0 11.9 6.0 11.0 14.0 7.6 1 (cm-1) 2.8 2.1 1.9 1.8 3.0 3.0 3.1 2.3
W1 (ps)
1.7 5.2 2.0 11.5 1.3 3.7 1.5 4.4
2 (cm-1) 2.2 2.7 3.3 5.6 -
W2 (ps)
66 14 22 21 -
s (cm-1) 2.7 3.1 3.5 2.9 4.3 1.9
T1 (ps) 15.8 24.1 19.3 18.4 16.0 16.1 8.0 8.0
88
3. Applications 3.1. Comparison of the protein dynamics for the different conformational substates Proteins do not necessarily fold into a single tertiary structure. As discussed in connection with Figure 1, these somewhat different structures are called conformational substates[3,4]. In ambient temperature, a protein can convert from one substate conformation to another. The different conformational substates usually show the different reaction rates for the protein function. Differences in protein structural dynamics for different substates have been revealed by several kinds of experiments, such as NMR, flash photolysis, kinetic hole burning, fluorescence and resonance Raman spectroscopy[71-75]. Here using 2D-IR vibrational echo spectroscopy we show that the differences in fast dynamics can be measured and quantified. The vibrational echo experiment is sensitive to both the local heme pocket dynamics and the global protein dynamics[11,12]. The CO transition frequencies are exquisitely sensitive to electric fields[7,10,68,76-78]. The protein is composed of charged, polar, and relatively non-polar groups. Structural fluctuations produce fluctuating electric fields at the CO and couple to the CO transition frequency through the Stark effect[11,12,68]. Therefore, the vibrational fluctuations measured by the 2D-IR vibrational echo experiment and quantified through the determination of the FFCF are determined by motions throughout the protein[11,12].
Closed
Leu29 His64 CO Val68 Ile107
Open
His93
B 1.0 absorbance (norm)

0.8 0.6 0.4 0.2 0.0 1900 1920 1940 1960 1980 2000
L29F (A3)
H64V (A0)
frequency
(cm-1)
Figure 5. Substate dependent spectral changes in MbCO. (A) 3D structure of the active site of CO-bound to wild-type Mb for open (right) and closed (left) conformations (Protein Data Bank). The heme and some selected amino acid residues are shown. (B) The linear FT-IR spectrum of L29FCO and H64VCO mutants.
89
Myoglobin is a small globular heme protein having 153 amino acids, which is found in mammalian muscle tissue. Heme can reversibly bind a number of small gaseous molecules such as O2, CO, and NO. The heme bound CO shows a number of stretching bands in the mid-IR region between 1,900 and 2,000 cm-1[7]. In the linear FT-IR absorption spectrum of the CO stretch of wild-type MbCO, there are three well known bands, denoted A0, A1, and A3. Mutant studies have shown that the distal histidine, His64, plays a prominent role in determining the CO stretching bands in Mb[7,77]. The physiological importance of the substates has been attributed to either steric hindrance of CO or stabilization of O2 binding by a hydrogen bond with the distal histidine[79-81]. The A1 and A3 conformations have the distal histidine localized in the heme pocket (closed), resulting in the distal histidine interacting significantly with the heme bound CO (Figure 5A left panel)[11]. The A0 substate has the distal histidine rotated out of the pocket (open), and there is little interaction between the distal histidine and the ligand (Figure 5A right panel)[76]. The A1 substate of wild-type MbCO is predominately populated and dominates the FT-IR absorption spectrum. The structural dynamics for the A1 and A3 conformations of wild-type MbCO have been measured previously using vibrational echo spectroscopy[11,12,20]. However, the A3 band overlaps substantially with the much larger A1 band and the A0 band is too small in wild-type MbCO to be useful for quantitative 2D-IR vibrational echo experiments. To compare the dynamics of conformational substates in MbCO, the mutant Mb proteins L29F and H64V were used. As shown in the absorption spectrum (Figure 5B), the L29F mutant has one major band at 1,932 cm-1 corresponding to the A3 conformation. Because the position 29 is close to the distal histidine, the replacement of amino acid residue from leucine to phenylalanine changes the distribution of the conformational substates. The H64V mutant protein, with the distal histidine replaced by a valine, mimics the situation in which the distal histidine is not in the heme pocket. The linear FT-IR absorption band for H64V at 1,968 cm-1 corresponds to the A0 conformation in wild-type Mb. Figure 4A and 6A show the 2D-IR vibrational echo spectra of CO bound to L29F and H64V Mb at several Tws, respectively[51]. The diagonal 0-1 transition bands and representative center lines are shown. It is clear that the Tw dependent elongation for CO stretching band of H64V (spectral diffusion) is slower than that of L29F. It also can be seen that for Tw = 32 ps the slope of the center line for L29F is steeper than that of H64V. Therefore, the inverse of the slope of the center line, which is related to the FFCF, is closer to zero for L29F. These results indicate that the different conformational substates in Mb have different rates of structural fluctuation induced spectral diffusion. The Tw dependent CLS data (points) and the fits (solid curves) for the L29F and H64V proteins are presented in Figure 6B[51]. (Data are acquired and fit to 100 ps but are not shown so that the fast component can be seen more readily.) The fits combined with the linear absorption spectrum yield the FFCF. It is clear from inspection of the data that the dynamics of the two Mb substates are very different. The spectral diffusion (structural fluctuations) of L29F is significantly faster than that of H64V. The differences in the dynamics of the two proteins can be quantified through their FFCFs. The FFCF parameters are given in Table 1[51]. The motionally narrowed component, characterized by T2* is slower for H64V than L29F. However, because this contribution to the 2D line shape is motionally narrowed, it is not possible to
90
separate the magnitude of the frequency fluctuations (') and the decay rate (W). The other components of the FFCF are more informative. The fast decay component of the FFCF ('1 and W1) are very different for L29F and H64V. The decay constant for L29F is a factor of three faster than that of H64V and the amplitudes of the fast components are similar. L29F has a W2 decay component of 66 ps while H64Vs slower component is static. A static component means that the dynamics are too slow for the 2D-IR experiment to measure within the time window of that experiment that is limited by the vibrational lifetime, T1. T1 = 24.1 ps for H64V, and typically we can obtain useful data for times as long as ~5T1. Therefore, the static component is significantly slower than several hundred ps. The amplitudes '2 for L29F and 's for H64V are about the same. In L29F, the structural fluctuations sample all accessible configurations about the protein folding minimum in several hundred ps. That is, all configurations of the protein that give rise to the L29 F CO absorption line (Figure 5) are sampled rapidly. In contrast, only about half of the H64V configurations are sampled on the same time scale. The FFCFs of the two proteins demonstrate quantitatively what can clearly be seen in Figure 6 B; the structural fluctuation dynamics of H64V are significantly slower than those of L29F. The structural fluctuations sensed by the CO bound at the active site of heme proteins are both global and local[11,12]. The major difference between L29F and H64V is the removal of the distal histidine and its replacement with valine, which has a small non-polar side group, in contrast to the histidine. Previous vibrational echo experiments and molecular dynamics (MD) simulations show that the A3 substate of MbCO (the equivalent of L29F) has the protonated epsilon nitrogen (NH) of the distal histidines imidazole side group closely associated, probably hydrogen bonded, to the CO ligand[11,12]. The results presented here indicate that removal of this interaction enables the protein to obtain a different conformation that is significantly less flexible and has greatly reduced fast (1 100 ps) structural fluctuations. Because of the great similarities between Mbs A3 substate and L29F, and Mbs A0 substate and H64V, it is reasonable to assume that Mbs A3 and A0 substates behave in the same manner as that observed for L29F and H64V, respectively.
A
Zm (cm-1)
1980 1975 1970 1965
Tw = 0.5
Tw = 4
1.0 0.8 0.6
CLS
H64V (A0 )
0.4 0.2 0.0 0
Zm (cm-1)
1975 1970 1965 1960
Tw = 16
Tw = 32
L29F (A3 )
1965 1970 1975
ZW (cm-1)
196519701975
ZW (cm-1)
10
20
Tw (ps)
30
40
50
Figure 6. Comparison of the protein dynamics of L29FCO and H64VCO. (A) 2D-IR spectra of H64VCO Mb at various times, Tw. Only the 0-1 transition is shown and white dashed lines are the center lines. (B) Tw dependent CLS for L29FCO (circles) and H64VCO (triangles). The calculated solid curves are from the FFCF obtained from the CLS data.
91
3.2. Neuroglobin
Neuroglobin (Ngb) is a recently discovered family of vertebrate globin proteins, which is expressed predominantly in the nervous system[59]. Comparison of Ngb with vertebrate Mb and Hb sequences show only minor similarities at the amino acid level (<25% identity), but Ngb features the conserved globin fold and contains heme[59]. Ngb has been hypothesized to facilitate O2 diffusion to the mitochondria. However, the concentration of Ngb in the brain is too low to play the role that Mb plays in the muscles. Since the expression level of Ngb is increased under hypoxic condition, Ngb may be involved in neuronal response to ischemia[82,83]. Another possibility is that Ngb detoxifies reactive oxygen species or is involved in signal transduction in the brain[84-86]. The structural feature of Ngb is that the heme iron is hexacoordinated both in the ferrous and ferric forms of Ngb[87]. An external gaseous ligand must compete with the sixth ligand, the distal histidine, for binding. While the ligand binding process in Ngb is distinct, several key residues in the heme pocket of Mb are conserved in Ngb (Figure 7A). However, leucine at position 29 in Mb is replaced by phenylalanine at position 28 in Ngb. The role of phenylalanine at position 28 in Ngb has not been elucidated, although many nonsymbiotic plants Hbs that have hexacoordinated binding to the heme iron also have phenylalanine at the same position[88]. Another feature of Ngb is that the structural analysis revealed that human Ngb has an intramolecular disulfide bond which affects its oxygen affinity[89]. Thus, the human Ngb sample discussed here contains an intramolecular disulfide bond. As shown in Figure 7B, the CO adduct of wild-type Ngb that contains a disulfide bond has two CO absorption bands at 1,933 cm-1 and 1,968 cm-1 that correspond to A3 and A0 conformations in Mb, respectively[51,90]. These bands have been called the N3 and N0[51,90]. The N3 band closely corresponds to the absorption at band 1,932 cm-1 of the L29F mutant of Mb. Because the leucine at position 29 in Mb is replaced by phenylalanine in Ngb, the L29F mutant Mb mimics the heme pocket structure in Ngb. The N0 band arises for the structure in which the distal histidine is rotated out of the heme pocket[90,91]. The H64V mutant of Mb mimics the situation and has the CO absorption band at 1,968 cm-1.
Phe28
B
absorbance (norm)
1.0 0.8 0.6 0.4 0.2 0.0 1900
L29F
N3
H64V
Phe42
His64 Val68
His96
N0
1920 1940 1960 1980 2000
frequency (cm-1)
Figure 7. Structural and spectral comparison of Ngb and Mb. (A) 3D structure of the active site of NgbCO (light gray) and L29FCO Mb (dark gray) proteins (Protein Data Bank). The heme and some selected amino acid residues are shown. The amino acid residue numbers are based on Ngb. (B) The linear FT-IR spectra of wild-type NgbCO (solid curve), L29FCO (dashed curve) and H64VCO (dot-dashed curve).
92
A
Zm (cm-1)
1980 1960 1940 1920 1980 1960 1940 1920
Tw = 0.5 +
Tw = 4
1.0 0.8 0.6 0.4 0.2 0.0
N0 N3
Tw = 16
Tw = 32
Zm (cm-1)
1920 1940 19601980 192019401960 1980
CLS
ZW
(cm-1)
ZW
(cm-1)
10
20
Tw (ps)
30
40
50
Figure 8. Structural dynamics of N3 and N0 substates of wild-type NgbCO. (A) 2D-IR spectra of wild-type NgbCO as a function of Tw. The positive going peaks (labeled +) on the diagonal and negative going peaks (labeled ) off diagonal correspond to vibrational echo emission at 0-1 and 1-2 vibrational transition frequencies, respectively. (B) Tw dependent CLS for N3 (circles) and N0 (triangles) conformations of wild-type NgbCO. The solid curves are calculated with the FFCF obtained from the CLS data.
To compare the structural dynamics of the wild-type Ngb and Mb mutants, the 2D-IR vibration echo spectrum of CO bound wild-type Ngb was measured[51]. Figure 8A presents 2D-IR spectrum of CO bound to Ngb at several values of Tw. There are two positive gong bands on the diagonal (N3 and N0), which correspond to the 0-1 vibrational transitions. The negative going bands arise from vibrational echo emission at the 1-2 vibrational transition. In Figure 8A, the peaks are normalized to the largest peak in each panel. As discussed above for the 2D-IR vibrational spectra of the mutant Mbs, the bands go from highly elongated along the diagonal to less elongated and increasingly broad along the ZW axis at long Tw. The Tw dependent CLS for the N3 and N0 bands for wild-type Ngb are presented in Figure 8B[51]. The difference between the CLS decays of N3 and N0 bands is qualitatively similar to the difference between the L29F and H64V mutant Mb (Figure 6B). These results indicate that the configuration of the distal histidine has a large effect on the structural dynamics and therefore the spectral diffusion of both of Ngb and Mb. When the distal histidine is out of the heme pocket, a strong interaction between the distal histidine and the ligated CO is eliminated. In both of Ngb and Mb, the configuration of the distal histidine is closely related to the fast protein fluctuation. The change in structure associated with the presence or absence of the distal histidine side group in the pocket influences the local and global protein structural dynamics. Although the qualitative nature of the relationship of the dynamics of N3/N0 to L29F/H64V is the same, the FFCF parameters are quite different (Table 1) [51]. The W1 value for the N3 band is similar to that of L29F mutant Mb, but the N3 longer time scale dynamics are quite different. The longer time scale equilibrium structural fluctuations of L29F show a complete sampling of all structures in several hundred ps, with a time constant of 66 ps. In contrast, N3 has a transition (14 ps) to a static component of the FFCF. Therefore, N3 has very slow dynamics that are not present in L29F. A substantial fraction of the structures in the N3 substate are sampled on times long compared to 100 ps. Therefore, in spite of the near identity of the heme pocket region of the protein, the structural fluctuations of N3 have a very slow component compared to L29F. It is possible that the fast component of both L29F and N3 arise
93
from the motions of the distal histidine and the differences in the FFCFs arise from differences in the global structural dynamics of the protein. The dynamics of the N0 substate in wild-type Ngb is also different from those of H64V although neither have the distal histidine side group in the heme pocket. Although both have static components of similar amplitude, the W1 values for N0 and H64V are 11.5 ps and 5.2 ps, respectively. Because the distal histidine is absent in both proteins, these measurements also support the idea that the observed structural fluctuations of wild-type Ngb and Mb have contributions from global motions rather than only very local interactions with the side group of the distal histidine at the active site. Ultrafast 2D-IR vibrational echo experiments on wild-type Ngb reveal that the protein fluctuations in the globin family proteins differ in the picosecond time region. Four types of globins have been discovered in humans and other vertebrates. They have a globin fold with several conserved key residues around the heme. However, they have distinct roles in the human body. Hb transports O2 in red blood cells; Mb in muscle provides O2 to mitochondria; Ngb and Cytoglobin (Cgb) are two newly discovered globin family proteins[59,92,93], which may provide O2 for mitochondria and may be involved in a signal transduction and/or scavenging reactive oxygen species. The differences in Ngb and Mb dynamics may be related to their distinct functions.
3.3. The Influence of the Disulfide B on Fast Protein Dynamics
Intramolecular covalent disulfide bonds take part in the regulation of protein folding, stability, and activity[94-96]. Because disulfide bonds are rigid structural elements in proteins, local and global protein structural fluctuations might be expected to be regulated by such bonds. Disulfide bond dependent structural configurations have been investigated by NMR and molecular dynamics (MD) simulation studies[97-101]. However, the effect of the bonds on the fast protein dynamics has not been well characterized experimentally.
A
Cys55 CO
Cys120
1.0 0.8
N0
CLS
Cys46
0.6 0.4 0.2
N3
10
20
30
40
50
Tw (ps)
Figure 9. Structure and disulfide bond dependent dynamics of NgbCO. (A) 3D structure of human Ngb (the position Cys55 and Cys120 are mutated by Ser; the position Cys46 is mutated by Gly) (Protein Data Bank). Cys46 and Cys55 of human Ngb are form a disulfide bond. (B) Tw dependent CLS data for wild-type (filled symbols) and reduced (open symbols) NgbCO. The curves through the data points (wild-type Ngb solid; reduced Ngb dashed) are calculated with the FFCF obtained from the CLS data.
94
The structural analysis of human Ngb has shown the presence of an intramolecular disulfide bond that regulates the O2 affinity in vitro (see Figure 9A)[89]. Although the physiological role of the disulfide bond in human Ngb is not clear, hypoxia may induce the reduction of the disulfide bond and result in a subsequent release of O2[89]. These results suggest that the formation of the intramolecular disulfide bond stresses the protein, and breaking the disulfide bond provides additional structural degrees of freedom of the protein, resulting in an increased affinity for O2 binding. Because Ngb contains a disulfide bond and heme, 2D-IR vibrational echo spectroscopy can investigate the effect of an intramolecular disulfide bond on dynamics via the spectral diffusion of the heme-ligated CO stretching band. As discussed above, the 2D-IR vibrational echo spectroscopy is sensitive to not only very local interactions at the active site but also global motions of proteins. Comparison between human Ngb with and without the disulfide bond provides insights into disulfide dependent structural modulation. To reveal the structural regulation of a disulfide bond in human Ngb, the disulfide bond is eliminated by reduction. The linear FT-IR spectrum of reduced Ngb is almost identical to that of wild-type Ngb, which has two CO bands at 1933 cm-1 and 1967 cm-1 (see Figure 7 B)[53]. Figure 9B shows a comparison of the 2D-IR vibrational echo CLS data for wild-type Ngb (filled symbols) and reduced Ngb (open symbols)[53]. It is clear that reduction of the disulfide bond has increased the rate of structural fluctuations on fast time scales for both N3 and N0 conformations compared to those of the wild-type Ngb protein. It is important to note that the disulfide bond is ~20 from the CO bound to the heme[102-104]. Therefore, the observed changes in the dynamics with the elimination of the disulfide bond are most likely global modifications of the protein fluctuations rather alterations that occur very locally in the heme pocket. The quantitative description of the time scales of the N3 and N0 bands are provided by the FFCFs (Table 1)[53]. The curves through the data points (wild-type Ngb solid curves; reduced Ngb dashed curves) are calculated from the FFCF obtained using the CLS method. The N3 substates of the wild-type and reduced Ngb have a homogeneously broadened component (T2*), two dynamic components (W1 and W2) that give rise to spectral diffusion within the time window of the experiment, and a static component ('s). For both the N3 and N0, reduction of the disulfide bond leaves T2* unchanged. Therefore, the ultrafast small very local motions of the protein are unchanged by elimination of the disulfide bond. The major change upon reduction of the disulfide bond is in the time constant W1 of both N3 and N0. Elimination of the disulfide bond reduces the fast N3 dynamics by ~1/3 but reduces the N0 dynamics by a factor of ~3. The fast dynamics of the N0 substate of the wild-type protein is much slower than that of the N3 substate probably because of there is not a contribution from to the spectral diffusion from the imidazole side group of the distal histidine. Without the contribution from the distal histidine, the N0 band is more sensitive to the global structural fluctuations of the protein. These results show that the disulfide band significantly reduces fast structural fluctuations of the protein. The N0 band does not display an intermediate time scale decay. Both the wild-type and reduced N0 bands have significant static components. The N3 band of both the wild-type and reduced proteins have an intermediate decay, which is actually faster for the wild-type. This may occur because some of the intermediate time scale fluctuations of the reduced protein have become part of the fast component and some fluctuations that were too slow to be observed and part of the wild-type static component have become faster
95
with the elimination of the disulfide bond and now contribute on the intermediate time scale. While the rate of fast structural fluctuations is increased in reduced Ngb, the reduction of the disulfide bond does not change the linear FT-IR absorption spectrum of the CO bound at the active site. The lack of change in the linear FT-IR spectrum indicates that the distribution of structures in the vicinity of the folded protein free energy minimum that are sampled under thermal equilibrium conditions are not altered significantly. These results suggest that the disulfide bond in Ngb regulates the heights of the relatively low energy landscape barriers that determine the rate of the fast structural fluctuations. Disruption of the disulfide bond in Ngb lowers the O2 affinity in vitro[89]. The disulfide bonds influence on the protein dynamics of Ngb may play a physiological.
3.4. The Influence of Substrate Binding on Enzyme Dynamics
Protein fluctuations are intimately coupled to the binding of a substrate to an enzyme[105,106]. An enzyme has many local structural minima on the free energy landscape and is constantly fluctuating among them[3,4]. Substrate binding to an enzyme induces conformational changes in the protein that may produce either a significant shift of the energy landscape minimum or the generation of what should be considered a new substate. These structural changes may be necessary for substrate binding and the subsequent chemical reaction. Insights into the influence of substrate binding on an enzyme can be gained by observing the affect of binding on the proteins dynamics. Horseradish peroxidase (HRP) is a heme containing enzyme that catalyzes a variety of organic molecules in the presence of hydrogen peroxide as the oxidizing agent[107]. HRP is widely used in analytical biochemistry and biotechnology[108-110]. To characterize the influence of substrate binding on enzyme dynamics, 2D-IR vibrational echo experiments were conducted on the substrate-free HRP and the substrate-bound HRP[36]. Since the heme in both substrate-free and substrate-bound HRP can bind CO, the CO spectral diffusion can be used as a probe of protein dynamics in the same manner as discussed for Mb and Ngb above. Five substrates, which are benzhydroxamic acid (BHA) analogs, have been studied with the 2D-IR vibrational spectroscopy[36]. Figure 10A shows the protein structure in the region of the active site with BHA bound.
A
His42 BHA Arg38
B
absorbance (norm)
1.0 0.8 0.6 0.4 0.2 0.0 1880 1900 1920
O N H OH
BHA
His170
1940
1960
frequency
(cm-1)
Figure 10. The influence of substrate binding to HRP. (A) 3D structure of the active site of HRP with the substrate BHA bound (Protein Data Bank). (B) The linear FT-IR spectra of HRPCO in the substrate-free (solid curve) and BHA-bound (dashed curve) forms. The structure of BHA is also shown.
96
A
Zm(cm-1)
Tw= 0.5 ps + 1930

1910 1890 1930 1910
Tw= 2 ps
B 0.8
0.7 0.6
Tw= 8 ps
Tw= 32 ps
CLS
0.5 0.4 0.3 0.2 0 5
No substrate (red)
Zm(cm-1)
BHA substrate
1890 1890 1910 1930 1890 1910 1930
ZW (cm-1)
ZW (cm-1)
10
Tw (ps)
15
20
25
30
Figure 11. The influence of substrate binding on HRP dynamics. (A) 2D-IR spectra of substrate-free HRPCO as a function of Tw. The positive going peaks (labeled +) on the diagonal and negative going peaks (labeled ) off diagonal correspond to vibrational echo emission at 0-1 and 1-2 vibrational transition frequencies, respectively. (B) Tw dependent CLS data for HRPCO without substrate (squares) and HRPCO with the substrate BHA bound (triangles). The curves through the data points (no substrate solid; with substrate dashed) are calculated with the FFCF obtained from the CLS data.
Figure 10 B shows the linear FT-IR spectra of the substrate-free and BHA-bound HRPCO[36]. Free HRP has two spectroscopically distinct substates, with CO absorption bands at 1,903 cm-1 and 1,934 cm-1. Previous studies indicated that the heme-ligated CO in the red state (absorbing at 1,903 cm-1) is nearly normal to the heme plane and has a strong interaction with the histidine residue and a weaker one with the arginine residue in the heme pocket[111,112]. In the blue state (absorbing at 1,934 cm-1), the CO ligand has a strong interaction with the arginine and a weaker one with the histidine[112]. When HRP binds BHA, the CO stretching band becomes a single at 1,909 cm-1[111,113]. Binding of all five substrates studied results in a single band. These bands have slightly different center frequencies that all strongly overlap the red band of the free HRP. Therefore, the single bound substate is mostly more closely related to the red state of the free HRP than to the blue state. Here for brevity only the red state of the free HRP will be discussed. Both free substates and all five substrates have been discussed previously[36]. Figure 11 A shows 2D-IR spectrum of CO bound to substrate-free HRP at several values of Tw. There are two positive gong bands on the diagonal (red and blue substates), which correspond to the 0-1 vibrational transitions. The negative going bands arise from the 1-2 vibrational transition. The Tw dependent CLS data for the red state of the substrate-free HRPCO (squares) and for HRP with BHA bound are presented in Figure 11 B[65]. The change in the dynamics upon substrate binding is dramatic. The solid curves are calculated with the FFCF determined for the CLS data. In addition to a motionally narrowed component, the red substate of substrate-free HRP has two time scales, W1 =1.5 ps and W2 = 21 ps, and no static term (see Table 1). All possible structures in the red substate of HRP without a bound substrate are sampled in less than 100 ps. However, in addition to a motionally narrowed component, the HRP with BHA bound has a single decay (W1 = 4.4 ps) and a static term. Therefore, substate binding causes
97
the fast component to slow by a factor of ~3 and the slow component to become static within the experimental time window, which indicates that the dynamics are slower than 100 ps. The other HRP-substrates also exhibit single time constants of ~2 ps to ~5 ps and a static term[36]. Detailed comparisons of the 2D-IR vibrational echo data for substrate bound an substrate free HRP with Mb mutants (see Figure 6 B) suggests that a major effect of substrate binding is lock up the distal arginine and distal histidine. The decay of the free HRP data is qualitatively similar to that of L29F while the decay of the BHA bound HRP is qualitatively similar to H64V[36]. While Mb only has a distal histidine, the difference in the dynamics of L29F and H64V, which does not have distal histidine, suggests a significant change is caused by the lack of the distal histidine side group motion in H64V. HRP with a substrate bound still has both distal ligands, but if their motions are greatly impeded on a fast time scale, the effect on the vibrational echo data will be similar. The substrate binding produces a new protein conformation (substate) that has greatly reduced dynamics. The changes will be both local in the pocket and global. Since the distal residues play an important role in every step of the enzymatic cycle of HRP, the conformational restrictions induced by the substrate binding may be important for the subsequent enzymatic reaction.
4. Concluding Remarks
2D-IR vibrational echo spectroscopy has made it possible to investigate fast protein dynamics under thermal equilibrium condition. Here we have briefly described the applications of 2D-IR vibrational echo methods to the study of heme protein dynamics. The experiments explicated the influences on dynamics of (1) different protein conformations, (2) different families of globin proteins, (3) the elimination of an intramolecular disulfide bond, and (4) substrate binding to an enzyme[20,36,51,53]. In these experiments, CO bound to the iron-heme was used as a reporter of protein dynamics. The CO provides a well defined vibrational chromophore that reports on both global and local protein structural fluctuations. Many other topics have been studied by a variety of research groups using 2D-IR spectroscopy[37,38,50,114-116]. Comparison of 2D-IR vibrational echo spectroscopy with other spectroscopic methods for the study of protein dynamics makes clear the utility of 2D-IR vibrational echo technique. NMR spectroscopy provides extremely high resolution structural information. However, the time resolution of NMR is much slower than that of 2D-IR vibrational echo spectroscopy. While the other optical methods, such as time resolved UV/vis and resonance Raman spectroscopy, can operate on ultrafast time scales, these are generally limited to photo induced processes rather than measuring dynamics under thermal equilibrium conditions. To understand the relationship between dynamics and function of biological molecules, thermal equilibrium condition dynamics are important. Therefore, the 2D-IR vibrational echo spectroscopy is a useful method for the understanding of structural dynamics and the nature of molecular interactions in biological molecules. In the experiments presented here, a single CO at a known location was used as the vibrational probe. Experiments can be conducted on, for example, the amide I band, but these lack location specificity for investigating dynamics. The types of experiments conducted here can be expanded beyond CO as a single site vibrational probe using artificially added vibrational dynamics probes much in the manner that
98
spin labels have been used in ESR. Recent progress of molecular biology makes it possible to generate proteins having unnatural amino acid residues[117]. For example, cyano-phenylalanine, azido-tyrosine, azido-phenylalanine, or azido-alanine containing protein have been synthesized[118-121]. All of these are IR active. The heme-ligated azide has been used for vibrational echo measurements[122]. The advantage of the introduction of an unnatural amino acid residue with an IR probe is that the position of probe is variable in protein sample. Another approach is the introduction of a substrate with an IR probe into enzyme. Such substrate probes will provide dynamical data the vantage point of the bound substrate. Over the past decade 2D-IR spectroscopy has developed rapidly. Currently, a vibrational echo spectrometer must be assembled. An instrument can be built from commercially available laser equipment, a great deal of optics, and commercially available very specialized detection equipment. In addition, a great deal of in house generated software is required to take and process the data. The early days of NMR experiments were very similar. Now multidimensional NMR instruments can be purchased as commercial packages. But it is important to recall that any sophisticated NMR instrument has a dedicated Ph. D. level operator that makes its use by a non-specialist community possible. One day soon, like NMR, a vibrational echo spectrometer will come as a package. However, right now, a Ph. D. 2D-IR vibrational echo spectroscopist could put together and operate a vibrational echo spectrometer for non-experts at a fraction of the cost of a top of the line NMR instrument. The availability of such instruments to the wider community would advance the fields of 2D-IR and biological research.
5. Acknowledgment
We thank Prof. R. Kopito (Stanford University) for the use of protein expression and purification equipment; Dr. K. Wakasugi (The Tokyo University) for kindly provide the plasmid neuroglobin; and Professor John S. Olson (Rice University) for providing the myoglobin mutant proteins. This work was supported by the NIH (2 R01 GM-061137-05). H. I. was supported by the Human Frontier Science Program. S. K. was supported by a fellowship from the Korea Research Foundation Grant funded by the Korean Government (KRF-2006-214-C00038).
References
[1] [2] D. Beece, L. Eisenstein, H. Frauenfelder, D. Good, M.C. Marden, L. Reinisch, A.H. Reynolds, L.B. Sorensen, K.T. Yue, Solvent viscosity and protein dynamics, Biochemistry 19 (1980), 5147-5157. P.J. Steinbach, A. Ansari, J. Berendzen, D. Braunstein, K. Chu, B.R. Cowen, D. Ehrenstein, H. Frauenfelder, J.B. Johnson, D.C. Lamb, Ligand binding to heme proteins: Connection between dynamics and function, Biochemistry 30 (1991), 3988-4001. H. Frauenfelder, F. Parak, R.D. Young, Conformational substates in proteins, Ann Rev Biophys Biophys Chem 17 (1988), 451-479. H. Frauenfelder, S.G. Sligar, P.G. Wolynes, The energy landscapes and motions of proteins, Science 254 (1991), 1598-1603. R.H. Austin, K. Beeson, L. Eisenstein, H. Frauenfelder, I.C. Gunsalus, V.P. Marshal, Activation-energy spectrum of a biomolecule: Photodissociation of carbonmonoxy myoglobin at low-temperatues, Phys Rev Lett 32 (1974), 403-405. R.H. Austin, K.W. Beeson, L. Eisenstein, H. Frauenfelder, I.C. Gunsalus, Dynamics of ligand binding
[3] [4] [5]
[6]
99
[7] [8]
[9] [10]
[11]
[12]
[13]
[14] [15] [16] [17] [18] [19] [20]
[21] [22] [23] [24]
[25]
[26] [27] [28] [29] [30]
[31] [32]
to myoglobin, Biochemistry 14 (1975), 5355-5373. G.N. Phillips, Jr., M.L. Teodoro, T. Li, B. Smith, J.S. Olson, Bound CO is a molecular probe of electrostatic potential in the distal pocket of myoglobin, J Phys Chem B 103 (1999), 8817-8829. A. Ansari, J. Berendzen, D. Braunstein, B.R. Cowen, H. Frauenfelder, M.K. Hong, I.E.T. Iben, J.B. Johnson, P. Ormos, T.B. Sauke, et al., Rebinding and relaxation in the myoglobin pocket, Biophys Chem 26 (1987), 337-355. T.S. Li, M.L. Quillin, G.N. Phillips, Jr., J.S. Olson, Structural determinants of the stretching frequency of CO bound to myoglobin, Biochemistry 33 (1994), 1433-1446. D. Morikis, P.M. Champion, B.A. Springer, S.G. Sligar, Resonance Raman investigations of site-directed mutants of myoglobin - effects of distal histidine replacement, Biochemistry 28 (1989), 4791-4800. K.A. Merchant, W.G. Noid, R. Akiyama, I. Finkelstein, A. Goun, B.L. McClain, R.F. Loring, M.D. Fayer, Myoglobin-CO substate structures and dynamics: Multidimensional vibrational echoes and molecular dynamics simulations, J Am Chem Soc 125 (2003), 13804-13818. K.A. Merchant, W.G. Noid, D.E. Thompson, R. Akiyama, R.F. Loring, M.D. Fayer, Structural assignments and dynamics of the a substates of MbCO: Spectrally resolved vibrational echo experiments and molecular dynamics simulations, J Phys Chem B 107 (2003), 4-7. D. Zimdars, A. Tokmakoff, S. Chen, S.R. Greenfield, M.D. Fayer, T.I. Smith, H.A. Schwettman, Picosecond infrared vibrational echoes in a liquid and glass using a free electron laser, Phys Rev Lett 70 (1993), 2718-2721. C.W. Rella, A. Kwok, K.D. Rector, J.R. Hill, H.A. Schwettmann, D.D. Dlott, M.D. Fayer, Vibrational echo studies of protein dynamics, Phys Rev Lett 77 (1996), 1648-1651. C.W. Rella, K.D. Rector, A.S. Kwok, J.R. Hill, H.A. Schwettman, D.D. Dlott, M.D. Fayer, Vibrational echo studies of myoglobin-CO, J Phys Chem 100 (1996), 15620-15629. M.D. Fayer, Fast protein dynamics probed with infrared vibrational echo experiments, Ann Rev Phys Chem 52 (2001), 315-356. A. Tokmakoff, M.D. Fayer, Infrared photon echo experiments: Exploring vibrational dynamics in liquids and glasses, Acc Chem Res 28 (1995), 437-445. M.D. Fayer (Ed), Ultrafast infrared and Raman spectroscopy Marcel Dekker, Inc, New York, Basel, 2001. M.A. Brown, R.C. Semelka, Mri: Basic principles and applications, Wiley-Liss, 1999. I.J. Finkelstein, J. Zheng, H. Ishikawa, S. Kim, K. Kwak, M.D. Fayer, Probing dynamics of complex molecular systems with ultrafast 2D IR vibrational echo spectroscopy, Phys Chem Chem Phys 9 (2007), 1533-1549. S. Park, K. Kwak, M.D. Fayer, Ultrafast 2D-IR vibrational echo spectroscopy: A probe of molecular dynamics, Laser Phys Lett 4 (2007), 704-718. J. Zheng, K. Kwak, J.B. Asbury, X. Chen, I. Piletic, M.D. Fayer, Ultrafast dynamics of solute-solvent complexation observed at thermal equilibrium in real time, Science 309 (2005), 1338-1343. Y.S. Kim, R.M. Hochstrasser, Chemical exchange 2D IR of hydrogen-bond making and breaking, Proc Natl Acad Sci USA 102 (2005), 11185-11190. J. Zheng, K. Kwak, X. Chen, J.B. Asbury, M.D. Fayer, Formation and dissociation of intra-intermolecular hydrogen bonded solute-solvent complexes: Chemical exchange 2D IR vibrational echo spectroscopy, J Am Chem Soc 128 (2006), 2977-2987. J.B. Asbury, T. Steinel, C. Stromberg, S.A. Corcelli, C.P. Lawrence, J.L. Skinner, M.D. Fayer, Water dynamics: Vibrational echo correlation spectroscopy and comparison to molecular dynamics simulations, J PhysChem A 108 (2004), 1107-1119. J.B. Asbury, T. Steinel, K. Kwak, S.A. Corcelli, C.P. Lawrence, J.L. Skinner, M.D. Fayer, Dynamics of water probed with vibrational echo correlation spectroscopy, J Chem Phys 121 (2004), 12431-12446. C.J. Fecko, J.D. Eaves, J.J. Loparo, A. Tokmakoff, P.L. Geissler, Local and collective hydrogen bond dynamics in the ultrafast vibrational spectroscopy of liquid water, Science 301 (2003), 1698-1702. S. Park, M.D. Fayer, Hydrogen bond dynamics in aqueous NaBr solutions, Proc Natl Acad Sci USA 104 (2007), 16731-16738. S.T. Roberts, J.J. Loparo, A. Tokmakoff, Characterization of spectral diffusion from two-dimensional line shapes, J Chem Phys 125 (2006), 084502. C.J. Fecko, J.J. Loparo, S.T. Roberts, A. Tokmakoff, Local hydrogen bonding dynamics and collective reorganization in water: Ultrafast infrared spectroscopy of HOD/D2O, J Chem Phys 122 (2005), 054506. J.B. Asbury, T. Steinel, M.D. Fayer, Hydrogen bond networks: Structure and evolution after hydrogen bond breaking, J Phys Chem B 108 (2004), 6544-6554. M. Khalil, N. Demirdoven, A. Tokmakoff, Vibrational coherence transfer characterized with Fourier-transform 2D IR spectroscopy, J Chem Phys 121 (2004), 362-373.
100
[33] K.D. Rector, C.W. Rella, A.S. Kwok, J.R. Hill, S.G. Sligar, E.Y.P. Chien, D.D. Dlott, M.D. Fayer, Mutant and wild type myoglobin-CO protein dynamics: Vibrational echo experiments, J Phys Chem B 101 (1997), 1468-1475. [34] M.T. Zanni, R.M. Hochstrasser, Two-dimensional infrared spectroscopy: A promising new method for the time resolution of structures, Curr Opin Struct Biol 11 (2001), 516-522. [35] P. Mukherjee, I. Kass, I.T. Arkin, M.T. Zanni, Picosecond dynamics of a membrane protein revealed by 2D IR, Proc Natl Acad Sci USA 103 (2006), 3528-3533. [36] I.J. Finkelstein, H. Ishikawa, S. Kim, A.M. Massari, M.D. Fayer, Substrate binding and protein conformational dynamics measured via 2D-IR vibrational echo spectroscopy, Proc Natl Acad Sci USA 104 (2007), 2637-2642. [37] J. Wang, J. Chen, R.M. Hochstrasser, Local structure of beta-hairpin isotopomers by FTIR, 2D IR, and ab initio theory, J Phys Chem B 110 (2006), 7545-7555. [38] H.S. Chung, M. Khalil, A.W. Smith, Z. Ganim, A. Tokmakoff, Conformational changes during the nanosecond-to-millisecond unfolding of ubiquitin, Proc Natl Acad Sci USA 102 (2005), 612-617. [39] S. Woutersen, R. Pfister, P. Hamm, Y. Mu, D.S. Kosov, G. Stock, Peptide conformational heteogeneity revealed from nonlinear vibrational spectroscopy and molecular-dynamics simulations, J Chem Phys 117 (2002), 6833-6840. [40] Y. Kim, R.M. Hochstrasser, Dynamics of amide-I modes of the alanine dipeptide in D2O, J Phys Chem B 109 (2005), 6884-6891. [41] C. Fang, R.M. Hochstrasser, Two-dimensional infrared spectra of the 13C=18O isotopomers of alanine residues in an alpha-helix, J Phys Chem B 109 (2005), 18652-18663. [42] P. Mukherjee, A.T. Krummel, E.C. Fulmer, I. Kass, I.T. Arkin, M.T. Zanni, Site-specific vibrational dynamics of the CD3zeta membrane peptide using heterodyned two-dimensional infrared photon echo spectroscopy, J Chem Phys 120 (2004), 10215-10224. [43] M.F. DeCamp, L. DeFlores, J.M. McCracken, A. Tokmakoff, K. Kwac, M. Cho, Amide I vibrational dynamics of N-methylacetamide in polar solvents: The role of electrostatic interactions, J Phys Chem B 109 (2005), 11016-11026. [44] N. Demirdoven, C.M. Cheatum, H.S. Chung, M. Khalil, J. Knoester, A. Tokmakoff, Two-dimensional infrared spectroscopy of antiparallel beta-sheet secondary structure, J Am Chem Soc 126 (2004), 7981-7990. [45] S. Mukamel, D. Abramavicius, Many-body approaches for simulating coherent nonlinear spectroscopies of electronic and vibrational excitons, Chem Rev 104 (2004), 2073-2098. [46] J.H. Choi, H. Lee, K.K. Lee, S. Hahn, M. Cho, Computational spectroscopy of ubiquitin: Comparison between theory and experiments, J Chem Phys 126 (2007), 045102. [47] A.M. Massari, I.J. Finkelstein, M.D. Fayer, Dynamics of proteins encapsulated in silica sol-gel glasses studied with IR vibrational echo spectroscopy, J Am Chem Soc 128 (2006), 3990-3997. [48] I.J. Finkelstein, A.M. Massari, M.D. Fayer, Viscosity dependent protein dynamics, Biophys J 92 (2006), 3652-3662. [49] A.M. Massari, B.L. McClain, I.J. Finkelstein, A.P. Lee, H.L. Reynolds, K.L. Bren, M.D. Fayer, Cytochrome c mutants: Structure and dynamics at the active site probed by multidimensional NMR and vibration echo spectroscopy, J Phys Chem B 110 (2006), 18803-18810. [50] C. Fang, A. Senes, L. Cristian, W.F. DeGrado, R.M. Hochstrasser, Amide vibrations are delocalized across the hydrophobic interface of a transmembrane helix dimer, Proc Natl Acad Sci USA 103 (2006), 16740-16745. [51] H. Ishikawa, I.J. Finkelstein, S. Kim, K. Kwak, J.K. Chung, K. Wakasugi, A.M. Massari, M.D. Fayer, Neuroglobin dynamics observed with ultrafast 2D-IR vibrational echo spectroscopy, Proc Natl Acad Sci USA 104 (2007), 16116-16121. [52] C. Kolano, J. Helbing, M. Kozinski, W. Sander, P. Hamm, Watching hydrogen-bond dynamics in a beta-turn by transient two-dimensional infrared spectroscopy., Nature 444 (2006), 469-472. [53] H. Ishikawa, S. Kim, K. Kwak, K. Wakasugi, M.D. Fayer, Disulfide bond influence on protein structural dynamics probed with 2D-IR vibrational echo spectroscopy, Proc Natl Acad Sci USA 104 (2007), 19309-19314. [54] I.J. Finkelstein, B.L. McClain, M.D. Fayer, Fifth-order contributions to ultrafast spectrally resolved vibrational echoes: Heme-CO proteins, J Chem Phys 121 (2004), 877-885. [55] I.J. Finkelstein, A. Goj, B.L. McClain, A.M. Massari, K.A. Merchant, R.F. Loring, M.D. Fayer, Ultrafast dynamics of myoglobin without the distal histidine: Stimulated vibrational echo experiments and molecular dynamics simulations, J Phys Chem B 109 (2005), 16959-16966. [56] A.M. Massari, I.J. Finkelstein, B.L. McClain, A. Goj, X. Wen, K.L. Bren, R.F. Loring, M.D. Fayer, The influence of aqueous vs. Glassy solvents on protein dynamics: Vibrational echo experiments and molecular dynamics simulations, J Am Chem Soc 127 (2005), 14279-14289. [57] K.A. Merchant, D.E. Thompson, Q.-H. Xu, R.B. Williams, R.F. Loring, M.D. Fayer, Myoglobin-CO
101
[58]
[59] [60] [61] [62] [63] [64] [65]
[66] [67]
[68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80]
[81] [82] [83] [84] [85] [86]
conformational substate dynamics: 2D vibrational echoes and md simulations, Biophys J 82 (2002), 3277-3288. H. Hartmann, F. Parak, W. Steigemann, G.A. Petsko, D.R. Ponzi, H. Frauenfelder, Conformational substates in a protein: Structure and dynamics of metmyoglobin at 80k, Proc Natl Acad Sci USA 79 (1982), 4967-4971. T. Burmester, B. Weich, S. Reinhardt, T. Hankeln, A vertebrate globin expressed in the brain, Nature 407 (2000), 520-523. M. Khalil, N. Demirdoven, A. Tokmakoff, Coherent 2D IR spectroscopy: Molecular structure and dynamics in solution, J Phys Chem A 107 (2003), 5258-5279. S. Mukamel, Principles of nonlinear optical spectroscopy, Oxford University Press, New York, 1995. D.M. Jonas, Two-dimensional femtosecond spectroscopy, Annu Rev Phys Chem 54 (2003), 425-463. J.B. Asbury, T. Steinel, M.D. Fayer, Vibrational echo correlation spectroscopy probes hydrogen bond dynamics in water and methanol, J Lumin 107 (2004), 271-286. M. Khalil, N. Demirdoven, A. Tokmakoff, Obtaining absorptive line shapes in two-dimensional infrared vibrational correlation spectra, Phys Rev Lett 90 (2003), 047401. K. Kwak, S. Park, I.J. Finkelstein, M.D. Fayer, Frequency-frequency correlation functions and apodization in two-dimensional infrared vibrational echo spectroscopy: A new approach, J Chem Phys 127 (2007), 124503. J.D. Eaves, J.J. Loparo, C.J. Fecko, S.T. Roberts, A. Tokmakoff, P.L. Geissler, Hydrogen bonds in liquid water are broken only fleetingly, Proc Natl Acad Sci USA 102 (2005), 13019-13022. P. Hamm, R.M. Hochstrasser, Structure and dynamics of proteins and peptides: Femtosecond two-dimensional infrared spectroscopy. In Ultrafast infrared and Raman spectroscopy. Edited by M.D. Fayer, Marcel Dekker, Inc., 2001,273-347. R.B. Williams, R.F. Loring, M.D. Fayer, Vibrational dephasing of carbonmonoxy myoglobin, J Phys Chem B 105 (2001), 4068-4071. Y.S. Bai, M.D. Fayer, Time scales and optical dephasing measurements: Investigation of dynamics in complex systems, Phys Rev B 39 (1989), 11066-11084. J. Schmidt, N. Sundlass, J. Skinner, Line shapes and photon echoes within a generalized kubo model., Chem Phys Lett 378 (2003), 559-566. J.S. Olson, G.N. Phillips, Jr., Kinetic pathways and barriers for ligand binding to myoglobin, J Biol Chem 271 (1996), 17593-17596. H.J. Dyson, P.E. Wright, Insights into protein folding from NMR, Annu Rev Phys Chem 47 (1996), 369-395. Y. Mizutani, T. Kitagawa, Ultrafast dynamics of myoglobin probed by time-resolved resonance Raman spectroscopy, Chem Rec 1 (2001), 258-275. R.M. Hochstrasser, D.K. Negus, Picosecond fluorescence decay of tryptophans in myoglobin, Proc Natl Acad Sci USA 81 (1984), 4399-4403. V. Srajer, P.M. Champion, Investigations of optical line shapes and kinetic hole burning in myoglobin, Biochem 30 (1991), 7390-7402. E. Oldfield, K. Guo, J.D. Augspurger, C.E. Dykstra, A molecular model for the major conformational substates in heme proteins, J Am Chem Soc 113 (1991), 7537-7541. T.G. Spiro, I.H. Wasbotten, CO as a vibrational probe of heme protein active sites, J Inorg Biochem 99 (2005), 34-44. E.S. Park, S.G. Boxer, Origins of the sensitivity of molecular vibrations to electric fields: Carbonyl and nitrosyl stretches in model compounds and proteins, J Phys Chem B 106 (2002), 5800-5806. E. Antonini, M. Brunori, Hemoglobin and myoglobin in their reactions with ligands, North-Holland, Amsterdam, 1971. J.B. Johnson, D.C. Lamb, H. Frauenfelder, J.D. Muller, B. McMahon, G.U. Nienhaus, R.D. Young, Ligand binding to heme proteins. 6. Interconversion of taxonomic substates in carbonmonoxymyoglobin, Biophys J 71 (1996), 1563-1573. S.E. Phillips, B.P. Schoenborn, Neutron diffraction reveals oxygen-histidine hydrogen bond in oxymyoglobin, Nature 292 (1981), 81-82. Y. Sun, K. Jin, X.O. Mao, Y. Zhu, D.A. Greenberg, Neuroglobin is up-regulated by and protects neurons from hypoxic-ischemic injury, Proc Natl Acad Sci USA 98 (2001), 15306-15311. Y. Sun, K. Jin, A. Peel, X.O. Mao, L. Xie, D.A. Greenberg, Neuroglobin protects the brain from experimental stroke in vivo, Proc Natl Acad Sci USA 100 (2003), 3497-3500. K. Wakasugi, T. Nakano, I. Morishima, Oxidized human neuroglobin acts as a heterotrimeric Galpha protein guanine nucleotide dissociation inhibitor, J Biol Chem 278 (2003), 36505-36512. K. Wakasugi, I. Morishima, Identification of residues in human neuroglobin crucial for guanine nucleotide dissociation inhibitor activity, Biochemistry 44 (2005), 2943-2948. S. Herold, A. Fago, R.E. Weber, S. Dewilde, L. Moens, Reactivity studies of the Fe(III) and Fe(II)NO
102
forms of human neuroglobin reveal a potential role against oxidative stress, J Biol Chem 279 (2004), 22841-22847. [87] S. Dewilde, L. Kiger, T. Burmester, T. Hankeln, V. Baudin-Creuza, T. Aerts, M.C. Marden, R. Caubergs, L. Moens, Biochemical characterization and ligand binding properties of neuroglobin, a novel member of the globin family, J Biol Chem 276 (2001), 38949-38955. [88] S. Kundu, J.T. Trent, 3rd, M.S. Hargrove, Plants, humans and hemoglobins, Trends Plant Sci 8 (2003), 387-393. [89] D. Hamdane, L. Kiger, S. Dewilde, B.N. Green, A. Pesce, J. Uzan, T. Burmester, T. Hankeln, M. Bolognesi, L. Moens, et al., The redox state of the cell regulates the ligand binding affinity of human neuroglobin and cytoglobin, J Biol Chem 278 (2003), 51713-51721. [90] H. Sawai, M. Makino, Y. Mizutani, T. Ohta, H. Sugimoto, T. Uno, N. Kawada, K. Yoshizato, T. Kitagawa, Y. Shiro, Structural characterization of the proximal and distal histidine environment of cytoglobin and neuroglobin, Biochemistry 44 (2005), 13257-13265. [91] T. Uno, D. Ryu, H. Tsutsumi, Y. Tomisugi, Y. Ishikawa, A.J. Wilkinson, H. Sato, T. Hayashi, Residues in the distal heme pocket of neuroglobin. Implications for the multiple ligand binding steps, J Biol Chem 279 (2004), 5886-5893. [92] T. Burmester, B. Ebner, B. Weich, T. Hankeln, Cytoglobin: A novel globin type ubiquitously expressed in vertebrate tissues, Mol Biol Evol 19 (2002), 416-421. [93] J.T. Trent, 3rd, M.S. Hargrove, A ubiquitously expressed human hexacoordinate hemoglobin, J Biol Chem 277 (2002), 19538-19545. [94] D. Barford, The role of cysteine residues as redox-sensitive regulatory switches, Curr Opin Struct Biol 14 (2004), 679-686. [95] H. Liu, R. Colavitti, Rovira, II, T. Finkel, Redox-dependent transcriptional regulation, Circ Res 97 (2005), 967-974. [96] P.J. Hogg, Disulfide bonds as switches for protein function, Trends Biochem Sci 28 (2003), 210-214. [97] S.F. Betz, J.L. Marmorino, A.J. Saunders, D.F. Doyle, G.B. Young, G.J. Pielak, Unusual effects of an engineered disulfide on global and local protein stability, Biochemistry 35 (1996), 7422-7428. [98] S.A. Beeser, T.G. Oas, D.P. Goldenberg, Determinants of backbone dynamics in native BPTI: Cooperative influence of the 14-38 disulfide and the Tyr35 side-chain, J Mol Biol 284 (1998), 1581-1596. [99] J.J. Kelley, III, T.M. Caputo, S.F. Eaton, T.M. Laue, J.H. Bushweller, Comparison of backbone dynamics of reduced and oxidized Escherichia coli glutaredoxin-1 using 15N NMR relaxation measurements, Biochemistry 36 (1997), 5029-5044. [100] B. Tidor, M. Karplus, The contribution of cross-links to protein stability: A normal mode analysis of the configurational entropy of the native state, Proteins 15 (1993), 71-79. [101] M.E. Moghaddam, H. Naderi-Manesh, Role of disulfide bonds in modulating internal motions of proteins to tune their function: Molecular dynamics simulation of scorpion toxin Lqh III, Proteins 63 (2006), 188-196. [102] A. Pesce, S. Dewilde, M. Nardini, L. Moens, P. Ascenzi, T. Hankeln, T. Burmester, M. Bolognesi, Human brain neuroglobin structure reveals a distinct mode of controlling oxygen affinity, Structure 11 (2003), 1087-1095. [103] B. Vallone, K. Nienhaus, M. Brunori, G.U. Nienhaus, The structure of murine neuroglobin: Novel pathways for ligand migration and binding, Proteins 56 (2004), 85-92. [104] B. Vallone, K. Nienhaus, A. Matthes, M. Brunori, G.U. Nienhaus, The structure of carbonmonoxy neuroglobin reveals a heme-sliding mechanism for control of ligand affinity, Proc Natl Acad Sci USA 101 (2004), 17351-17356. [105] R. Jimenez, G. Salazar, J. Yin, T. Joo, F.E. Romesberg, Protein dynamics and the immunological evolution of molecular recognition, Proc Natl Acad Sci USA 101 (2004), 3803-3808. [106] B. Ma, M. Shatsky, H.J. Wolfson, R. Nussinov, Multiple diverse ligands binding at a single protein site: A matter of pre-existing populations, Protein Sci 11 (2002), 184-197. [107] N.C. Veitch, Horseradish peroxidase: A modern view of a classic enzyme, Phytochemistry 65 (2004), 249-259. [108] N.C. Veitch, A.T. Smith, Horseradish peroxidase, Adv Inorg Chem 51 (2000), 107-162. [109] S.M. Aitken, J.L. Turnbull, M.D. Percival, A.M. English, Thermodynamic analysis of the binding of aromatic hydroxamic acid analogues to ferric horseradish peroxidase, Biochemistry 40 (2001), 13980-13989. [110] A.T. Smith, N.C. Veitch, Substrate binding and catalysis in heme peroxidases, Curr Opin Chem Biol 2 (1998), 269-278. [111] W.J. Ingledew, P.R. Rich, A study of the horseradish peroxidase catalytic site by FTIR spectroscopy, Biochem Soc Trans 33 (2005), 886-889. [112] S. Hashimoto, H. Takeuchi, Protonation and hydrogen-bonding state of the distal histidine in the CO
103
complex of horseradish peroxidase as studied by ultraviolet resonance Raman spectroscopy, Biochemistry 45 (2006), 9660-9667. [113] I.E. Holzbaur, A.M. English, A.A. Ismail, Infrared spectra of carbonyl horseradish peroxidase and its substrate complexes: Characterization of pH-dependent conformers, J Am Chem Soc 118 (1996), 3354-3359. [114] H.S. Chung, Z. Ganim, K.C. Jones, A. Tokmakoff, Transient 2D IR spectroscopy of ubiquitin unfolding dynamics, Proc Nat Acad Sci USA 104 (2007), 14237-14242. [115] P. Mukherjee, I. Kass, I.T. Arkin, M.T. Zanni, Structural disorder of the CD3zeta transmembrane domain studied with 2D IR spectroscopy and molecular dynamics simulations, J Phys Chem B 110 (2006), 24740-24749. [116] H. Maekawa, C. Toniolo, Q.B. Broxterman, N. Ge, Two-dimensional infrared spectral signatures of 3 10- and alpha-helical peptides, J Phys Chem B 111 (2007), 3222-3235. [117] J. Xie, P.G. Schultz, Adding amino acids to the genetic repertoire, Curr Opin Chem Biol 9 (2005), 548-554. [118] K.C. Schultz, L. Supekova, Y. Ryu, J. Xie, R. Perera, P.G. Schultz, A genetically encoded infrared probe, J Am Chem Soc 128 (2006), 13984-13985. [119] S. Ohno, M. Matsui, T. Yokogawa, M. Nakamura, T. Hosoya, T. Hiramatsu, M. Suzuki, N. Hayashi, K. Nishikawa, Site-selective post-translational modification of proteins using an unnatural amino acid, 3-azidotyrosine, J Biochem 141 (2007), 335-343. [120] J.W. Chin, T.A. Cropp, J.C. Anderson, M. Mukherji, Z. Zhang, P.G. Schultz, An expanded eukaryotic genetic code, Science 301 (2003), 964-967. [121] K.L. Kiick, E. Saxon, D.A. Tirrell, C.R. Bertozzi, Incorporation of azides into recombinant proteins for chemoselective modification by the staudinger ligation, Proc Natl Acad Sci USA 99 (2002), 19-24. [122] M. Lim, P. Hamm, R.M. Hochstrasser, Protein fluctuations are sensed by stimulated infrared echoes of the vibrations of carbon monoxide and azide probes., Proc Natl Acad Sci USA 95 (1998), 15315-15320.
104
FTIR Data Processing and Analysis Tools

Erik GOORMAGHTIGH Laboratory for the Structure and Function of Biological Membranes, Center for Structural Biology and Bioinformatics, Universit Libre de Bruxelles, Belgium
Abstract: The information retrieved from FTIR spectra largely depends on both the quality of the original spectra and on the correction and processing methods. This contribution reviews the entire process driving to a fine and reliable interpretation of the data. Keywords: Fourier transform infrared, Attenuated total reflection, water vapor, side chain, noise, correlations, principal component analysis, deuterium exchange secondary structure, orientation
1. Introduction In the course of the last 20 years we have used FTIR spectroscopy for the study of membrane and membrane proteins. In turn we essentially used attenuated total reflection (ATR) instead of transmission spectroscopy as the former method allows the recording of the orientation of specific chemical groups in orientated membranes. The potential differences between transmission and ATR spectra as well as the specific advantages of ATR have been described thoroughly in a previous review [1]. The relevant conclusion of this previous review for the present chapter is that, in some instances, ATR spectra can present particularities not encountered in transmission spectra. Significant distortions occur when the spectra are recorded close to the critical angle, which happen to be the case at 45 incidence with common materials such as KRS-5 or ZnSe. All the spectra presented in this review have been recorded on germanium crystals, at 45 incidence. Because of the high refractive index of germanium, the critical angle is far from 45 and such distortions do not occur. The reader is referred to our previous review [1] for a detailed discussion. The purpose of this chapter is to present some practical information about the recording and processing of the FTIR spectra.
____________________________
1 Corresponding Author: Laboratory for the Structure and Function of Biological Membranes, Center for Structural Biology and Bioinformatics, Universit Libre de Bruxelles, CP 206/2, Boulevard du Triomphe, B1050 Brussels, Belgium; E-mail: egoor@ulb.ac.be.
E. Goormaghtigh / FTIR Data Processing and Analysis Tools
105
2. Methodology: Recording of Spectra The recording of spectra is the most important step prior to analysis. If the data do not contain the information or contain too much noise, there is no way to overcome the problem. We present below the typical procedure we follow for the recording of the spectra. Attenuated total reflection infrared (ATR-FTIR) spectra presented in this chapter were obtained on one of our five Bruker FTIR spectrophotometers of the IFS55/equinox family (Ettlingen, Germany) all equipped with a MCT detector (broad band 12000-420 cm-1, liquid N2 cooled, 24h hold time). Spectra were recorded at a resolution of 2 cm-1 with an aperture of 3.5 mm and acquired in the double-sided, forward-backward mode. The spectrometer was placed on vibration-absorbing sorbothane mounts (Edmund Industrial Optics, Barrington, NJ, USA). Two levels of zero filling of the interferogram were applied prior to Fourier transform. Spectra were finally saved in the ASCII JCAMP format, then they were conveniently encoded with one data point every wavenumber for subsequent manipulations. The spectrometer was continuously purged with dry air (Whatman 75-62, Haverhill, MA, USA or K-MT8 air dryer from Zander, Essen, Germany). For a better stability, the purging of the spectrometer optic compartment (5 l/min) and of the sample compartment (10-20 l/min) were controlled independently by flowmeters (Fisher Bioblock Scientific, Illkirch, France). Room temperature was maintained constant at 22C with an air conditioning system. The germanium crystals were washed in Superdecontamine (Intersciences, AS, Brussels, Belgium), a lab detergent solution at pH 13, rinsed with distilled water, washed with methanol, then with chloroform and finally placed for 2 min in a plasma cleaner PDC23G (Harrick, Ossining, NY, USA) working under reduced air pressure. Thin films were obtained by slowly evaporating a sample containing a total 10-100 g of protein or lipid on one side of the ATR plate under a stream of nitrogen. This hold for large internal reflection element, 52x20x2 mm trapezoidal germanium ATR plate (ACM, Villiers St Frdric, France) with an aperture angle of 45 yielding 25 internal reflections. In some experiments a Golden Gate diamond ATR unit (Specac, Orpington, UK) was used with 0.1 to 1 g sample. It must be stressed that, as a rule, the dry weight of buffer, salt and other molecules from the solution must be kept smaller than the dry weight of membranes (lipids + proteins). If this condition is not respected, the spectrum intensity is weak because of the exponential decay of the evanescent wave. In fact, diluting the molecules of interest in others, even non-absorbing molecules, results in parallel decrease of the spectral intensity. We found that an excellent way to improve the quality of the spectra is to overlay the membrane film with ca 300 l of a buffer, then remove the liquid as much as possible by tilting the crystal and soaking the liquid with a filter paper, and finally dry the film again under N2. This procedure also allows the composition of the buffer, or pH to be modified at will [2]. The germanium crystal was placed in an ATR holder for liquid sample with an in- and out-let (Specac, Orpington, UK). The liquid cell was placed at 45 incidence on a Specac vertical ATR setup. Two such setups, the second mounted as the mirror image of the first one were fitted on the sample shuttle provided by Bruker, allowing the recording of two samples or two H/D exchange kinetic experiments almost simultaneously. This is an important feature since the two samples can be compared under identical conditions (temperature, gas flow rate). Furthermore, an elevator under computer control made it possible to move the whole setup along a vertical axis (built by WOW Company SA, Nannine,
106
Belgium). This allows the crystal to be separated in different lanes, one of them being used for the background. This elevator is absolutely required when the temperature of the sample is to be changed in the course of the experiment, for instance when monitoring phase transitions in lipid membranes. For hydrogen/deuterium exchange experiments, nitrogen gas was saturated with 2 H2O by bubbling through a series of four vials containing 2H2O. The flow rate of 50 ml/min was controlled by a flowtube (Fisher Bioblock Scientific, Illkirch, France). Bubbling was started at least one hour before starting the experiments. At zero time, the tubing was connected to the cavity of the liquid cell chamber surrounding the film. 20 scans were recorded and averaged for each time point. The time interval was increased exponentially (see Figure 6). After 27 minutes, the interval between the scans was large enough to allow the interdigitation of a second kinetics measurement. The second sample was then analyzed with the same time sampling but with a 27 minute offset. Deuteration started by connecting it to the 2H2O-saturated N2 flow from the output of the first sample chamber. Sample shuttle movements and spectrum recording were under control of a macro program written for OPUS (Bruker, Ettlingen, Germany). For the analysis of 1H/2H experiments, the areas of the lipid Q(C=O), amide I, I', and II were obtained by automatic integration. For each spectrum, the area of amide II was divided by the area of amide I in order to take into account the swelling of the sample layer due to the presence of 2H2O. All kinetic curves were analyzed as multiexponential decays of populations Di of amide protons with the same timeconstants Tj, using a nonlinear-least-squares procedure. It is usual to fit the proportion of unexchanged amide proton curve H(t) by a small number M of exponential (typically 3) representing each a class Aj of amide groups:
H (t )
A
j 1
exp(
t ) Tj
(1)
2D correlation spectra were calculated according to Noda [3-5] and as described by Nabet and Pzolet [6]. Computation was carried out using the Hilbert transform as recently reported [7]. The 2D correlation spectra can be interpreted using the rules described by Noda [8] and more recently by Ekgasit et al. [9]. Fourier self-deconvolution, when required, was performed according to [10-12]. Deconvolution was performed with a Lorentzian line (FWHH = 30 cm-1) and apodization with a Gaussian line (FWHH = 15 cm-1) resulting in a so-called linenarrowing factor (K) of 2.0. It must be noted that while the narrowing effect of Fourier self-deconvolution has been widely used in the past, the shape and width of the deconvoluting line shape are usually unknown, resulting in less efficient band narrowing as clearly illustrated in the past [13, 14] and more recently by LorenzFonfria et al [15]. Smoothing was simply obtained by apodization of the spectrum Fourier transform, typically by the Fourier transform of a 4 cm-1 wide Gaussian. All the software used for data processing was written under MatLab (Mathworks Inc, Natick, Ma, USA).
107
3. Primary Spectral Processing We present below the processing of the spectra which is almost automatically applied to recorded data. 3.1. Spectral Corrections for Atmospheric Water The presence of sharp atmospheric water absorption lines superimposed on the sample spectra cannot be avoided in long-term experiments or when recording very small absorbances. Atmospheric water absorbance is better corrected when spectra are recorded at relatively high resolution for taking advantage of the linewidth difference existing between the atmospheric water vapor and the solid sample bands [16]. This is illustrated on Figure 1.
0.4 0.35
120 100
Resolution (cm-1) 0.5 1.0 2.0 4.0 8.0

1700 1600 cm-1 1500 14
0.3 0.25 0.2 0.15 0.1 0.05 0 1800 1700 1600 cm-1 1500 14
Absorbance
Absorbance
80 60 40 20 0 1800
Figure 1. Water vapor spectra recorded at different nominal resolution. The same amount of water vapor is present in all the spectra and the spectra are presented on the same scale, with an offset. On the left panel, water vapor spectra have been recorded after decreasing the purging of the sample compartment. Spectra were recorded at 0.5, 1, 2, 4 and 8 cm-1 as indicated. On the right panel, the spectra of a tumor cell line are overlaid with the spectra from the left panel. It can be observed that the same amount of water is practically undetectable for the lower resolutions.
Obviously no smoothing should be carried out before this step. Simple subtraction of a reference water vapor spectrum is sufficient in most cases but derivative-shaped lines are usually left over. The reason is obviously related to a difference between the reference water vapor used for the correction and the one present on the sample spectrum. The problem is further complicated when one realizes that there is not one but up to 4 water vapor spectra involved. These contributions comes from 1) the water vapor present at the time of the recording of the background, 2) the water vapor present at the time of the recording of the sample, 3) the water vapor present at the time of the recording of the reference of the water vapor spectrum and 4) the water vapor present
108
at the time of the recording of the background used for the latter reference spectrum. The physical reasons for the different spectra are not completely clear. Temperature and atmospheric pressure play an important role. In addition we have noticed that the precise positioning of the germanium crystal contributes to the problem. The problem of water vapor subtraction has attracted the attention of several groups in the past [16, 17]. We present below different approaches that, in our hands, best help solve the problem. 1) purging thoroughly the sample compartment until "equilibrium". This appears obvious but in practice equilibrium is never reached. Our software can plot the subtraction coefficients used for the correction for overnight experiments. Fluctuations keep appearing even after 12 hours demonstrating that the stability of the purge system is not sufficient. 2) collecting and subtracting a water reference spectrum. It can be hypothesized that the best reference spectrum is the one collected on top of the sample. A spectrum (SpA) is collected say after 10 min purging and another one (SpB) after 12 min purging. Simply computing SpA-SpB yields a water vapor spectrum collected exactly in the same conditions as the sample and both backgrounds (for the sample and reference water vapor) are identical. In turn the 4th contribution mentioned above is not relevant anymore. The program then computes the subtraction coefficient as the ratio of the atmospheric water band area between 1562 and 1555 cm-1 (a straight line is draw between the spectrum points at these two wavenumbers) on the sample spectrum and on the reference atmospheric water spectrum. This level of processing is generally sufficient for experiments such as the monitoring of H/D kinetics. If an acceptable result is not achieved at this point a further step must be undertaken. 3) In order to account for the small frequency shifts of the water vapor bands we wrote software that accurately evaluates the position of the 1573-1579 cm-1 band by curve fitting a Gaussian lineshape on the reference and sample spectra. Once the shift is determined, the reference water vapor spectrum is shifted (typically by 0.1-0.2 cm-1) by linear interpolation before subtraction as above. This approach improves the results in most cases but may also fail to yield a correct result because of the multiplicity of the water contributions as described above. The most difficult cases are processed with either one of the two last methods. Figure 2 presents the reference water vapor spectrum (curve A), the original sample spectrum (curve B), the corrected spectrum obtained without shift of the reference spectrum (curve C) and the corrected spectrum obtained after shift of the reference spectrum (curve D). The smoothed curve D appears as curve E. This Figure demonstrates clearly the need for the wavenumber adjustment of the reference water vapor spectrum. This approach also demonstrates that the internal shift to be applied is fluctuating in an apparent random manner during the course of an overnight experiment (not shown). 4) In order to account for several contributions with several bandshifts, we use a combination of a reference spectrum with itself shifted by several values (typically the starting values are -0.1 and + 0.1 cm-1). In turn, 3 reference spectra are generated and 5 parameters must be determined to optimize the subtraction: the three subtraction coefficients and the two shifts. We found it convenient to use a least square procedure to minimize the "length" of the
109
resulting spectrum obtained after water vapor subtraction in a defined spectral range. This idea comes from the observation that at rather "high" resolution, the sample spectrum overlaid by the spiky water vapor spectrum is much "longer" than the intrinsically smooth spectrum of the sample. The "length" of the spectrum is obtained as the sum of the distance between all the datapoints in the selected spectral range. Practically it is obtained by subtracting the resulting spectrum from itself after shifting the datapoints by one point. The sum of the difference is minimized in order to adjust the 5 different coefficients. Unexpectedly we found that the recording of the reference spectrum is not critical. Good results are obtained when using a standard vapor spectrum recorded several years before.
250
3 Absorbance x10 Absorbance
200 150 100 50 0 1800
E. D. C. B. A.
1700 1600 cm-1 1500 14
Figure 2. Illustration of water vapor band removal on a spectrum of triose phosphate isomerase. Reference water vapor spectrum (curve A), original sample spectrum (curve B), corrected spectrum obtained without shifting the reference spectrum (curve C), corrected spectrum obtained after shifting the reference spectrum (curve D) and smoothing of curve D by apodization of its Fourier transform by a Gaussian lineshape in order to obtain a resolution of 4 cm-1 (curve E). The subtraction coefficient is computed as the ratio of the area of the band present between 1562 and 1555 cm-1 as indicated by the arrows.
5)
The alternative procedure does not use shifts but many spectra that represent the variety of the water vapor contributions. Principal components were used to reconstruct the water vapor contribution. The principal components were obtained from initially 1360 water vapor spectra collected in recent year by different researchers in the lab. After rescaling, analysis of the spectra for noise, rejection of outliers, the series was subjected to principal component decomposition (Figure 3). As before the "length" of
110
6)
the curve between 1700 and 1600 cm-1 is minimized in order to determine the subtraction coefficient for the different principal components. The best results are usually obtained with 2-5 principal components. Using too many principal components can result in a degradation of the result. Finally, whatever the procedure followed, the corrected spectrum is smoothed by apodization of its Fourier transform by the Fourier transform of a 4 cm-1 wide Gaussian lineshape.
0.7 0.6
Absorbance x103
Absorbance
0.5 0.4 0.3 0.2 0.1 1800 1750 1700 1650 1600
-1 cm cm
-1
1550
1500
1450
14
Figure 3. 4 first principal components describing water vapor contribution in the 1800-1400 cm-1 range.
3.2. Spectral Corrections for Amino Acid Side Chain Contribution In the course of the 1H/2H exchange experiments, once the atmospheric water bands have been removed, the spectra display a distinct feature located near 1580 cm-1 whose intensity rapidly increases as a function of the deuteration time (Figure 4, left arrow). Another shoulder is present near 1515 cm-1 throughout the experiment (Figure 4, right arrow) but becomes more and more visible as the exchange proceeds. These are due to absorbances of the protein amino acid residue side chains. The 1515 cm-1 can easily been assigned to tyrosine ring vibration and the broad shoulder at 1580 cm-1 is an overlap of Qas(COO-) from Asp and Glu and of Arg Qs(CN32H5+). Figure 4 demonstrates that the area under amide II as limited by the baselines drawn on the Figure is underestimated because of the side chain contributions. In the undeuterated spectrum, the sum of the side chain contributions represents about 10% of the amide I intensity (Figure 5). Because of its overall broad shape, it is usually not taken into account when the amide I shape is analyzed for protein secondary structure determination. In the deuterated spectrum, a distinct maximum appears near 1580 cm-1 (Figure 5) and makes it impossible to establish a reliable baseline to evaluate the amide II area. We therefore developed software that computes the contribution of the protein side-chains as a function of the extent of the deuteration. The deuterated and undeuterated contributions of the side chains (computed from the amino acid composition in the protein and from
111
the data reported by Venyaminov and Kalnin [18], Chirgadze and Brazhnikov [19], summarized in [20-22]) appear in Figure 5. The contributions of individual side chains reported in the literature had to be slightly adjusted in order to fit the amino acid side chain contributions observed here according to [23]. More recent reports on side chain contributions could help improve the quality of the correction [21, 24-26] but a precise modelisation of the complex environment found in real proteins remains out of reach. Subtracting the side chain contribution from every spectrum of the series recorded in the time course of the kinetics pre-supposes the knowledge of 1) the pH which governs the ionization state of the carboxylic amino acids, 2) the fraction of deuterated and undeuterated amino acid side chain for every spectrum of the kinetic and, 3) the subtraction coefficient. In the present study, 1) the pH was supposed to be the pH of the solution from which the film was prepared since we showed previously that ionization of carboxylic acids was identical in films and solutions [27] and that pH dependency of the exchange rate [23] and of peptide orientation [2] are maintained in film samples. 2) for every spectrum of the kinetic a side chain deuteration index was computed from the intensity decay at 1673 cm-1 during the first 10 minutes of the experiment, which monitors mainly arginine and asparagine deuteration, i.e. the main side-chain contributions sensitive to deuteration in the present case. In agreement with previous observations [28], the exchange of the amino acid side chains was fast. 3) The scaling factor for the subtraction of the total amino acid side chain contribution was based on the integrated intensity of the side chain contributions (see above) on the one hand and on the amide I extinction coefficients computed according to the protein secondary structure from data reported in the literature ([29-31], summarized in [20]) on the other hand. In the course of this work, it appeared that the so-determined subtraction coefficients are not completely adequate (accuracy of the reported extinction coefficients, difference in experimental conditions) and should be multiplied by 1.5 to obtain a satisfactory subtraction of the Tyr contributions.
400
3 Absorbance x10 Absorbance x10 Absorbance
3
300 200 100 0
1750 1700 1650 1600 1550 1500 1450 -1 cm cm-1
Figure 4. Spectra of the gastric ATPase obtained before (bottom) and after 80 minutes of exposure to 2H2O. The baselines drawn are used to delimitate the area of the ester Q(C=O), amide I and amide II. Absorbance reading refers to scaled spectra.
112
400
A.
Absorbance 300 200 100 0
1750 1700 1650 1600 1550 1500 1450 -1 cm
Asn Gln Tyr Arg Asp Glu Tyr Phe
B.
80 60 40 20 0
Absorbance
1750 1700 1650 1600 1550 1500 1450 -1 cm
Figure 5. Illustration of side chain contribution removal from an undeuterated spectrum (bottom) and from a spectrum deuterated for 80 min (top). The original (thin line) and corrected (thick line) spectra obtained after subtraction of the side chain contribution (dotted line) appear in panel. The sum of the side chain contributions as well as their individual contributions appear in panel B for the spectra presented on panel A. Absorbance scale refer to scaled spectra.
113
3.2.1. Critical Appraisal of Side Chain Contribution Subtraction While useful when amide II area has to be evaluated, limitations of the method need to be stressed. First, the reconstituted side chain contributions presented here are based on model compounds whose characteristics have been measured a long time ago. The type of spectrometer, resolution and hydration levels are not considered. Second, the large variety of environments present in real protein cannot be represented by the bands used. Third, the scaling for the subtraction is at best an approximation. In turn, when we tried to improve secondary structure prediction as explained in [32] by getting rid of side chain contributions in the amide I and amide II range, we rather observed a slight degradation of the predictions. 3.3. Spectral Rescaling In ATR experiments the intensity of the spectra depends on both the amount of material and on the way the material is spread on the ATR crystal. The latter parameter is never fully under control. In turn a scaling of the spectra is often required. This scaling is necessary to compare spectra, to average spectra, to run PCA or correlations analyses. Typically, the area under a band or several bands will be set to a given value. An alternative method is to use the entire spectrum for scaling. In 1H/2H exchange experiments, the additional hydration of the film induced upon contact with 2H2O gas results in a swelling of the membrane film. In turn, the intensity of the whole spectrum decreases because of the exponential decay of the evanescent wave outside the germanium crystal. In order to account for this intensity decrease that is not related to the deuteration process, the amide II areas (Figure 4) are rescaled with respect to either the lipid Q(C=O) band or the amide I band. Results are comparable for both bands (not shown). These two bands are equally affected by the swelling effect and are conveniently located close to the amide II band. This is important since the effect of the swelling on the absorbance is expected to depend on the wavelength because of the wavelength dependency of the penetration depth of the evanescent wave (see equation 12 and Figure 5 in [1]). 3.4. Noise Level Check Noise is defined here as the standard deviation in a segment of the spectrum after subtraction of a linear baseline. For noise estimation, a spectral region (preferentially without absorbance) is defined, typically 2200-2100 cm-1. A segment of defined length is moved on the spectrum. For every position 1) a linear baseline is fitted to this segment and subtracted to account for any general tilt of the baseline and 2) the standard deviation around the mean is computed. This can be repeated by steps of short segments (e.g. 12 cm-1) to minimize the contribution of broad bands when present. A rejection threshold should be set to eliminate automatically spectra of bad quality when studying large series of spectra. Scaling of the spectra, if used, should be applied before this step for the sake of the comparison of the standard deviations. Decent spectra should have a standard deviation below 1/1000 of the amide I intensity.
114
4. Tools for the Analysis of FTIR Spectra Numerous approaches are currently used for extracting the relevant information from FTIR spectra. Most often series of spectra are produced, requiring some kind of analysis able to take into account the small variations which describe the dynamics of the system. Among the method used, difference spectroscopy is one of the most powerful. As it will be described in more detail in the present book, the reader is referred to that chapter or to previous publications on this topic [22, 24, 33-35]. 4.1. 1H/2H Exchange Kinetics: Hydrogen isotope exchange has long been used for the analysis of protein structure and dynamics [36-38] (for a review see [39, 40]). When compared to mass spectrometry or 1 H/3H exchange, the great advantage of monitoring the exchange by FTIR is that the measure is focused on the amide protons only, yielding data proportional to the number of residues in the protein. The interest of a correct and accurate analysis of the 1H/2H exchange kinetics is twofold. First, the exchange rate contains important information on the structure and structure stability of a protein at a submolecular level. Second, the as clearly illustrated in the past [13, 14] and more recently by Lorenz-Fonfria et al [41] exchange acts as a perturbation that can help reveal otherwise hidden components in the spectra. Simultaneous use of polarized and 1H/2H exchange experiments adds a further dimension to the analysis [42, 43]. Series of spectra recorded in the course of 1H/2H exchange experiments are reported on Figure 6 before and after correction for the contributions of the side chains. Figure 7 reports H(t), the evolution of the number of 1H amide remaining as a function of the time. It is usual to fit the curve H(t) by a small number M of exponential (typically 3) representing each a class Aj of amide groups (equation 1). The result of a curve fitting with three exponentials also appears on Figure 7. The curve fitting indicates that 384 residues belong to the fast exchanging class T=1.2 min) and 804 to the slowly exchanging amide protons. Further experiments described in the literature report that hundreds of residues change their accessibility to exchange [44, 45] in the presence of ligand inducing an E1 or E2 conformation. Simultaneously, attenuated total refection Fourier transform infrared experiments under a flowing buffer were carried out to modulate the environment of the protein inside the measurement cell . The high accuracy of the results allows to demonstrate that the E1 to E2 transition induces a net change in secondary structure that concerns 10 15 amino acid residues over a total of 1324 in the proteins [44]. Decomposition of the exchange curves as presented in Figure 7 supposes 3 distinct -1 classes of amide protons. Alternatively, the inverse Laplace transform L immediately yields the distribution shape without hypothesis on the number of classes (2) f ( k ) L-1 ^H (t ) ` Knox and Rosenberg [46] suggested a dimensionless presentation of the distribution function obtained after rewriting of the integral expression H(t)
H (t )
f
f
k f ( k ) exp( kt ) d ln( k )
(3)
Solving the inverse Laplace transform is subject to several artifacts if not carefully treated [47]. We used here the CONTIN program kindly provided by Dr. Provencher
115
[38, 48]. Figure 8 shows the time constant distribution for the experiment described in Figure 7.
A.
Absorbance x103
Absorbance x103
B.
cm-1
Figure 6. Evolution with time of exposure to 2H2O of the spectrum of the gastric H+,K+-ATPase in native tubulovesicle membranes. The first 10 spectra were recorded before the beginning of the deuteration. The spectra have been recorded after 0; .25; .50; .75; 1.00; 1.25; 1.50; 1.75; 2.00; 2.50; 3.00; 3.50; 4.00; 5; 6; 7; 8; 10; 12; 14; 16; 20; 24; 28; 32; 40; 48; 56; 64; 80; 96; 112; 128; 160; 192; 224; 256; 320; 384; 448 and 512 min. Spectra have been corrected for water vapor contribution. They are shown before (A) and after correction for side chain contribution (B). The direction of the main time evolutions is indicated by arrows.
116
H(t) (%)
Proportion Residue number
Fast (T=1.2 min) 28.9% 384
Intermediate (T=8.3 min) 12.9% 171
Slow (T=2275 min) 60.5% 804
Time (min)
Figure 7. Evolution of the area of amide II/amide I ratio (in %) for the gastric ATPase exposed to 2H2O vapor. A curve fitting with 3 exponentials (equation 1) resulted in the line drawn through the experimental points (circles). The characteristics obtained for the 3 exponentials are tabulated on the Figure. Residue number come from the conversion of the percentage into amino acid residues taking into account the 1324 residues present in the entire protein
Figure 8. Distribution of the 1H/2H exchange time constants k=1/T for the gastric ATPase obtained after inverse Laplace transform (equation 3) of the curve presented in Figure 7.
117
So far, 1H/2H exchange demonstrated a unique sensitivity to structural changes. The question of which secondary structure are involved in the different classes of amide protons can be addressed in different ways. Amide I can be analyzed in details to obtain the exchange rates of the different secondary structure types as demonstrated earlier [28, 49]. Alternatively, since amide II is as sensitive as amide I to secondary structure ([32] and see below), amide II decay at different wavenumbers will yield information on different secondary structures. An inverse Laplace transform was performed on the decay observed every other wavenumber between 1580 and 1520 cm1 for the data displayed on Figure 6. Results are reported in Figure 9. It can be observed in Figure 9 that the intermediate component exchanging in ca 10 min is largely due to helices whose maximum appears near 1545-1550 cm-1 in the amide II while the slower component near 1000 min has a maximum at lower wavenumbers, i.e. closer to the Esheet contribution (near 1540 cm-1 and below). Asynchronous correlations computed in the same spectral range confirm the presence of different secondary structure exchanging at different rates. Figure 10 reports the asynchronous map for the series of spectra plotted on Figure 6B. A clear maximum (indicated by a circle on the Figure) at 1540/1555 cm-1 confirms the inverse Laplace analysis results. It must be noted that the synchronous map does not reveal the feature. Once the time constant present in the exchange process are known, it is relatively easy to express the entire series of spectra as a linear combination of three spectra weighted each by an exponential decay as described in equation 1. A simple matrix inversion allows the extraction of the three spectra representing the exchange for the three time constants [50] (Figure 11).
1580
Wavenumber (cm-1)
1570 1560 1550 1540 1530 1520 0 2 4 -log(k) 6
Figure 9. Contour plot of the inverse Laplace transform computed every other wavenumber between 1580 and 1520 cm-1 for the spectra reported in Figure 6B.
118
1520
Wavenumber (cm-1)
1530 1540 1550 1560 1570 1580 1580 1560 1540

-1
1520
Wavenumber (cm )
Figure 10. Asynchronous map for the series of spectra plotted on Figure 6B. The circle indicates the 1540/1555 cm-1 peak.
20 15 Absorbance 10
-1660
-1619
x 10
-3
T=2275 min
-1628
-1648
-1523
T=8.3 min
-1629 -1534
5
-1661
0 -5 1800
T=1.2 min
-1536
1700
1600 -1 cm
1500
14
Figure 11. Decomposition of the spectra from Figure 6B into 3 spectra representing the spectral variations occurring for each time constant revealed in Figure 7. The decomposition is obtained by linear regression as explained in [50]
119
The continuous shift of the negative and positive peaks in the amide I towards smaller wavenumbers is in line with helical structure exchanging first, then sheet structure exchanging more slowly. The similar shift in the amide II region confirms this interpretation. It must be kept in mind that the position of the original bands is different from the extrema found in the difference-shaped spectra. In conclusion, evaluation of 1H/2H exchange kinetics requires a thorough preparation of the data. We have presented here some of the key issues that should be taken into account and the main approaches used for the analysis of the data. 4.2. Secondary Structure Determination Since the eighties a large number of methods to estimate protein secondary structure content via the analysis of FTIR spectra have been reported. Curve fitting was originally used and an example of such an analysis is that published by Byler and Susi [51] in which protein amide I bands was analyzed by fitting with a series of Gaussian curves. The success reported in the original paper was spectacular: the RMS errors for D-helix and E-sheet were on the order of ca 2.5%. The curve fitting method compensates for band position variation among a same secondary structure assignment by assigning all component bands found in a given regions of the spectrum to a particular structure. Used with Fourier self-deconvolution, this method can be highly effective when applied by one experienced in its use [13, 51-54]. Yet, curve fitting requires a series of subjective decisions that can dramatically affect both the results and the interpretation [13, 55, 56]. Furthermore, curve fitting has a tendency to overestimate the E-sheet content of primarily helical proteins, and routinely finds 15-20% E-sheet for proteins that actually have none [51, 54, 57-59]. Multivariate statistical analysis methods have proven to be an alternative powerful tool for the analysis of protein spectra e.g. factor analysis [60-64], singular value decomposition [65], more sophisticated approaches such as the holistic approach developed in [66], multiple neural network in [67-70] or the enhanced prediction of secondary structure obtained by combining curve analysis and hydrogen/deuterium exchange [71], curve analysis and isotope editing [72] or curve analysis and temperature [73]. Both transmission [74, 75] and ATR-FTIR spectroscopy [1] have been reviewed extensively. A critical parameter that has not been taken into account systematically, except for genetic algorithms [76, 77] and the local regression method interval partial leastsquares (iPLS) [78] is the selection of the wavenumbers used for building models. Including large wavenumber ranges involves wavenumbers that are not correlated with the particular secondary structure to be estimated and in turn results in a degradation of the prediction accuracy. Discussion is still current about the interest of the various regions of the spectrum for the determination of the secondary structure content [66] and computation of spectra might shed some new light on this problem in a near future [79]. Today, the high quality of the FTIR spectrometers makes the absorbance at every single wavenumber almost noiseless. We address here the question of wavenumber information content and of the redundancy of the information present at different wavenumbers for secondary structure prediction. It appears that at most the absorbances at 3 distinct wavenumbers contain all the non-redundant information that can be related to one secondary structure content. Addition of more spectral data points is useless or even degrades the prediction quality. Interestingly; wavenumber by wavenumber analyses identify the relevance of every wavenumber in the IR spectrum for the prediction of a given secondary structure and yields a particularly simple
120
method for computing the secondary structure content since a linear equation that contains the absorbance at only a few wavenumber yields the best predictions. We have previously built a protein database that covers as well as possible the / secondary structure space, the fold space as described by CATH (Class, Architecture, Topology and homology classification of proteins [80] as well as other structural features such as helix length, and number of chains in a sheet. We identified 50 commercially available proteins that can be obtained with sufficient purity and for which we assessed the quality of the crystal-derived structure [81]. From this database we can address questions about the secondary structure information contained in the spectra. Figure 12 reports the error on the secondary structure prediction for the D-helix and E-sheet when building simple linear model such as Struct(%)=c1 + c2 . absorbance at wavenumber i
Error on structure prediction (standard deviation)
20 Absorbance
D-helix 1 wavenumber
-1224
-1515
15
-1694
-1624 -1302
-1655
10 1800 1700
D-helix 2 wavenumbers
1200 1100 10
-1545
1600
1500
1400 cm-1
1300
16 Absorbance 14 12 10 8 1800 1700

-1696 -1628 -1656
E-sheet 1 wavenumber
-1423
-1514 -1543
-1221
E-sheet 2 wavenumbers
1100 10
-1304
1600
1500
1400 cm-1
1300
1200
Figure 12. Evolution of the standard deviation on the predicted D-helix and E-sheet content (%) in the protein database when building a simple linear model based on the absorbance at one wavenumber or including two wavenumbers. Such models have been built for every wavenumber of the spectrum and reported in the Figure. A baseline has been subtracted at 1720, 1485, 1426, 1355, 1211 and 1010 cm-1 and the spectra were scaled between 1720 and 1485 cm-1.
121
The constants c1 and c2 were determined by linear regression for all the wavenumbers i and the models were evaluated by the error on the structure prediction. The profiles for the errors reported in Figure 12 present a view of the information contained in the spectra for every secondary structure. Surprisingly, the best wavenumber identified for D-helix prediction is 1545 cm-1 in the amide II. Amide III also contains very valuable information on the helix content. Yet, because amide III is most often overlapped with buffer or lipid contributions it remains practically less interesting. For the E-sheet structure, the best prediction is obtained using the absorbance at 1628 cm-1. The model can be complicated and improved by adding a second wavenumber. In the example of Figure 12 we have retained the best wavenumber identified as described above and screened for the additional gain in prediction accuracy when a second one is added. The curves presented for both structures indicate that most of the information was contained in the first wavenumber; the residual information presented is significant but rather small. These profiles emphasize the redundancy of the information content as, for instance for the D-helix structure, once the absorbance at 1545 cm-1 is included in the model the large predicting power of the other bands largely disappear. It must be emphasized that the profiles presented in Figure 12 do not allow drawing conclusions on the wavenumbers assigned to each structure. This is due to the fact that D-helix and E-sheet structures are complementary in any database. Predicting one already yields most of the information on the other. Obviously the profile reported for the D-helix structure has minima at wavenumbers associated with helix bands (1655 and 1545 cm-1) but also with bands associated with E-sheet (1694 and 1624 cm-1). The first principal component (Figure 13) obtained by principal component analysis (PCA) contains describes 63% of the variance between 1800 and 1400 cm-1 and is highly correlated to the D-helix and E-sheet content. It displays the D-helix feature positive at 1655 and 1545 cm-1 on the one hand and the E-sheet feature negative at 1694, 1629 and 1515 cm-1. The value of 1515 cm-1 is lower than expected, probably because of a mixing with side chain contributions.
-1655
Absorbance
0.1
-1545
0
-1694
-1466
-1586
-1567
-1515
-0.1 -0.2 1800
-1629
1700
1600 -1 cm cm-1
1500
14
Figure 13. First principal component obtained from the protein database used for building Figure 12.
122
E
1500
E E D -
Synchronous correlation
1550
1600
E
1650
D E
1700 1700 1500
1650
1600
1550
1500
E D x
Asynchronous correlation
1550
1600
E
1650
D E
1650 1600 1550 1500
1700 1700 1500
E D x
Coorelation coefficient
1550
1600
E
1650
D E
1700 1700
1650
1600
1550
1500
cm-1
Figure 14. Synchronous correlation, asynchronous correlation and correlation coefficient in the 1700-1500 cm-1 region computed on the spectra of the 50 protein database described above. D and E refer to a potential assignment to D-helix and E-sheet respectively. Negative values are plotted as thin lines.
123
Synchronous and asynchronous maps shed more light on the correlation features present in the 50 protein spectra. Figure 14 displays some features of the correlation coefficient, synchronous and asynchronous maps. Following the dotted line at 1545 cm1 from the bottom we can read that this band is negatively correlated with 1695, 1625 and 1520 cm-1, all characteristics of E-sheet structures and positively correlated with 1655 cm-1, a wavenumber characteristic of the helix structure. The synchronous map gives the same reading but misses the 1695 cm-1 band. The asynchronous map reveals additional cross-peak at 1640 and 1580 cm-1 belonging more likely to another structure (turn, random) and/or to side chain contributions. The 6 bands crossed when moving vertically along the dotted line drawn at 1545 cm-1 are the six major features that can be distinguished on the maps. A vertical dotted line has been draw at these wavenumbers on Figure 14. This short discussion indicates that both the correlation coefficients and the asynchronous maps bring complementary information. They allowed the identification of 6 major distinct contributions in the amide I/amide II region of the spectrum. The synchronous and the correlation coefficient maps are very similar in nature but the sensitivity of the correlation coefficient is much better as variations at every wavenumber are rescaled by the variance at the same wavenumber. In turn, smallamplitude variations are as significant as the largest ones. The 1695 cm-1 E-sheet band for instance appears strongly on the correlation coefficient but not on the synchronous map. The unassigned band at 1585 cm-1 only appears on the correlation coefficient map. The latter contribution is likely related to side chain contributions. 4.3. Orientation Determination of molecular orientations from polarized ATR-FTIR has been reviewed previously [1, 14, 54] and a deep insight about the dipole orientation and geometry of the secondary structures has been provided [82-86]. Polarized ATR-FTIR approach is extremely powerful as it allows the determination of the orientation of membrane molecules including the lipids, protein [43, 87-100, 100, 101], peptides [91, 102-110] or drugs [111, 112] simultaneously, on the same sample, without labeling. The only question we will address here is that of the quality of the orientation of the membranes. The measured order parameter S, denoted Sexperimental, obtained from RATR as explained elsewhere [1, 14, 113, 114] can be generally expressed as the product of several order parameters related to a set of nested, uniaxial symmetric distributions [115]. In this condition
Sexperimental = Smembrane . Shelix . Sdipole
(4)
where Smembrane describes the distribution of the lipid membrane patches (smallest planar membrane unit) with respect to the internal reflection element, Shelix describes the orientation of the helices within the membrane plane and Sdipole describes the dipole orientation of either amide I or amide II with respect to the helix axis. A schematic representation of these nested distributions appears in Figure 15. Each contribution to the product can be seen itself as the product of the mean tilt contribution by the contribution of the disorder characterized by the distribution of the angular values about their mean.
124
Figure 15. Set of nested axially symmetric distributions. The membrane normal is distributed about the germanium crystal normal (angle J), the secondary structure axis about the membrane normal (angle E) and the transition dipole moment about the secondary structure axis (angle D).
Remarkably, because of the symmetry of the experiment, it can be demonstrated that only these two Legendre polynomials contribute to the IR dichroism [14, 115]. The coefficient <Pn> can be evaluated as
S
Pn !
2 0
D(J ).Pn (cos J ). sin J .dJ

S
(5)
2 0
D(J ). sin J .dJ
It means that whatever the shape D(J) of the distribution of the tilts about the mean value J0, the <P0> and <P2> coefficients fully describe the IR dichroism even if they may poorly describe the angular distribution. The value of the coefficient <P2> is usually called order parameter, S, as it describes the disordering around the mean value for P-ATR experiments. In recent paper [116], D(J) for the membranes was directly measured from AFM images and its projection on P2, i.e. <P2>=Smembrane, evaluated according to equation 5. The membrane used were the intracytoplasmic tubulovesicles bearing the H+,K+-ATPase directly extracted from pig stomachs. These membranes represent an example of native membrane by opposition to better ordered systems built with synthetic lipids. Details about the preparation and characterization of the tubulovesicles can be found elsewhere [93, 95, 117-120]. The spectral intensity and linear dichroism were measured for average thicknesses ranging between 0 and 100
125
bilayers. Height profiles were obtained by atomic force microscopy (AFM) along lines randomly through the image (Figure 16). Orientation distribution function were obtained from the slopes and decomposed into Legendre polynomials. It was found that the second Legendre polynomials coefficient characterizing the membrane orientation was always larger than 0.9 [116]. Remarkably, addition of tubulovesicle membranes in small amount (about just enough to cover the area) smears out the roughness of the germanium surface resulting from the polishing. Further addition of membrane materials fully hides the polishing grooves. In conclusion, it appears that even for natural membrane, the disordering of the membrane on a clean germanium crystal is quite small and can be ignored. Thermally induced bending fluctuations as described by Marsh, Shanmugavadivu and Kleinschmidt [121] will have little impact in stacks of membranes prepared in the absence of an excess of water.
Figure 16. Slope distributions obtained on a 5x5 m2 image for a tubulovesicle multilayer stack.
5. Conclusions Modern recording techniques allow the recording of hundreds or thousands of spectra every day. The strength of FTIR precisely relies on its unique capability to detect small differences in spectra, either recorded on different objects (imaging produces thousands of spectra in a matter of minutes) or in the course of a reaction. The present challenge is to handle these spectra. Corrections for water vapor, smoothing, detection of outliers must be automated to be of practical interest. Similarly, spectral investigations rely upon correlation analyses, decomposition into components that have a special meaning or advanced statistics. The questions to be answered vary from one problem to the next. In turn, a great flexibility is required. Hypotheses must put forward and tested. In this review we have presented a few examples of specific analyses related to specific problems. Other problems will raise other questions, other hypotheses and other testing. We found it most useful in recent year to build the ability to handle the spectra in a flexible programming environment.
126
6. References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] Goormaghtigh,E., V.Raussens, and J.M.Ruysschaert. Biochim.Biophys.Acta 1422 (1999), 105-185. Bechinger,B., J.M.Ruysschaert, and E.Goormaghtigh. Biophys J 76 (1999), 552-563. Noda,I. Appl.Spectrosc. 44 (1990), 550-561. Noda,I. Appl. Spectrosc. 47 (1993), 1329-1336. Noda,I., A.E.Dowrey, C.Marcott, G.M.Story, and Y.Ozaki. Appl. Spectrosc. 54 (2000), 236A-248A. Nabet,A. and M.Pzolet. Appl. Spectrosc. 51 (1997), 466-469. Sasic,S., A.Muszynski, and Y.Ozaki. Appl. Spectrosc. 55 (2001), 343-349. Noda,I., A.E.Dowrey, and C.Marcott. Appl.Spectrosc. 47 (1993), 1317-1323. Ekgasit,S. and H.Ishida. Appl. Spectrosc. 49 (1995), 1243-1253. Kauppinen,J.K., D.J.Moffat, H.H.Mantsch, and D.G.Cameron. Anal.Chem. 53 (1981), 1454-1457. Kauppinen,J.K., D.J.Moffat, H.H.Mantsch, and D.G.Cameron. Appl.Spectrosc. 35 (1981), 271-276. Kauppinen,J.K., D.J.Moffat, and H.H.Mantsch. Appl.Opt. 20 (1981), 1866-1880. Goormaghtigh,E., V.Cabiaux, and J.M.Ruysschaert. Subcell.Biochem. 23 (1994), 405-450. Goormaghtigh,E. and J.M.Ruysschaert. 1990. In Molecular description of biological membrane components by computer-aided conformational analysis. R.Brasseur, editor. CRC Press, Boca Raton FL. 285-329. Lorenz-Fonfria,V.A., J.Villaverde, and E.Padros. Appl. Spectrosc. 56 (2002), 232-242. Goormaghtigh,E. and J.M.Ruysschaert. Spectrochim.Acta 50A (1994), 2137-2144. Bruun,S.W., A.Kohler, I.Adt, G.D.Sockalingum, M.Manfait, and H.Martens. Appl. Spectrosc. 60 (2006), 1029-1039. Venyaminov,S.Y. and N.N.Kalnin. Biopolymers 30 (1991), 1243-1257. Chirgadze,Y.N., O.V.Fedorov, and N.P.Trushina. Biopolymers 14 (1975), 679-694. Goormaghtigh,E., V.Cabiaux, and J.M.Ruysschaert. Subcell.Biochem. 23 (1994), 329-362. Barth,A. Progress in Biophysics & Molecular Biology 74 (2000), 141-173. Barth,A. and C.Zscherp. Quaterly Rev.Biophys. 35 (2002), 369-430. Goormaghtigh,E., H.H.de-Jongh, and J.M.Ruysschaert. Appl.Spectrosc. 50 (1996), 1519-1527. Barth,A. Biochim.Biophys.Acta 1767 (2007), 1073-1101. Bush,M.F., M.W.Forbes, R.A.Jockusch, J.Oomens, N.C.Polfer, R.J.Saykally, and E.R.Williams. J. Phys.Chem. A 111 (2007), 7753-7760. Forbes,M.W., M.F.Bush, N.C.Polfer, J.Oomens, R.C.Dunbar, E.R.Williams, and R.A.Jockusch. J. Phys.Chem. A 111 (2007), 11759-11770. de-Jongh,H.H., E.Goormaghtigh, and J.M.Ruysschaert. Biochemistry 34 (1995), 172-179. de-Jongh,H.H., E.Goormaghtigh, and J.M.Ruysschaert. Biochemistry 36 (1997), 13593-13602. Chirgadze,Y.N., B.V.Shestopalov, and S.Y.Venyaminov. Biopolymers 12 (1973), 1337-1351. Chirgadze,Y.N. and E.V.Brazhnikov. Biopolymers 13 (1974), 1701-1712. Venyaminov,S.Y. and N.N.Kalnin. Biopolymers 30 (1990), 1259-1271. Goormaghtigh,E., J.M.Ruysschaert, and V.Raussens. Biophys.J. 90 (2006), 2946-2957. Liu,M., M.Krasteva, and A.Barth. Biophys. J.89 (2005), 4352-4363. Stolz,M., E.Lewitzki, D.Thoenges, W.Mantele, A.Barth, and E.Grell. J. Gen. Physiol. 126 (2005), 32A. Ritter,M., O.Anderka, B.Ludwig, W.Mantele, and P.Hellwig. Biochemistry 42 (2003), 12391-12399. Gregory,R.B. and R.Lumry. Biopolymers 24 (1985), 301-326. Knox,D.G. and A.Rosenberg. Biopolymers 19 (1980), 1049-1068. Provencher,S.W. and V.G.Dovi. Can J.Biochem. 1 (1979), 313-318. Englander,S.W. and N.R.Kallenbach. Q.Rev.Biophys. 16 (1984), 521-655. Kim,P.S. Methods in Enzymology 131 (1986), 136-156. Lorenz-Fonfria,V.A., J.Villaverde, and E.Padros. Appl. Spectrosc. 56 (2002), 232-242. Garczarek,F. and K.Gerwert. Journal of the American Chemical Society 128 (2006), 28-29. Grimard,V., C.Vigano, A.Margolles, R.Wattiez, H.W.van-Veen, W.N.Konings, J.M.Ruysschaert, and E.Goormaghtigh. Biochemistry 40 (2001), 11876-11886. Scheirlinckx,F., R.Buchet, J.M.Ruysschaert, and E.Goormaghtigh. Eur.J.Biochem. 268 (2001), 36443653. Vigano,C., M.Smeyers, V.Raussens, F.Scheirlinckx, J.M.Ruysschaert, and E.Goormaghtigh. Biopolymers 74 (2004), 19-26. Knox,D.G. and A.Rosenberg. Biopolymers 19 (1980), 1049-1068.
[15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46]
127
[47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92]
Provencher,S.W. and V.G.Dovi. J.Biochem.Biophys.Methods 1 (1979), 313-318. Provencher,S.W. Computer Physics Communications 27 (1982), 213-227. de-Jongh,H.H., E.Goormaghtigh, and J.M.Ruysschaert. Biochemistry 36 (1997), 13603-13610. Raussens,V., J.M.Ruysschaert, and E.Goormaghtigh. Appl.Spectrosc. 58 (2004), 68-82. Byler,D.M. and H.Susi. Biopolymers 25 (1986), 469-487. Cabiaux,V., R.Brasseur, R.Wattiez, P.Falmagne, J.M.Ruysschaert, and E.Goormaghtigh. J.Biol.Chem. 264 (1989), 4928-4938. Prestrelski,S.J., K.A.Pikal, and T.Arakawa. Pharm.Res. 12 (1995), 1250-1259. Goormaghtigh,E., V.Cabiaux, and J.M.Ruysschaert. Eur.J.Biochem. 193 (1990), 409-420. Jackson,M. and H.H.Mantsch. Crit.Rev.Biochem.Mol.Biol. 30 (1995), 95-120. Surewicz,W.K. and H.H.Mantsch. Biochim.Biophys.Acta 952 (1988), 115-130. Haris,P.I., D.Chapman, and G.Benga. Eur.J.Biochem. 233 (1995), 659-664. Jap,B.K., M.F.Maestre, S.B.Hayward, and R.M.Glaeser. Biophys J 43 (1983), 81-89. Van Hoek,A.N., M.Wiener, S.Bicknese, L.Miercke, J.Biwersi, and A.S.Verkman. Biochemistry 32 (1993), 11847-11856. Baumruk,V., P.Pancoska, and T.A.Keiderling. J Mol Biol 259 (1996), 774-791. Lee,D.C., P.I.Haris, D.Chapman, and R.C.Mitchell. Biochemistry 29 (1990), 9185-9193. Pribic,R., I.H.van Stokkum, D.Chapman, P.I.Haris, and M.Bloemendal. Anal.Biochem. 214 (1993), 366-378. Sarver,R.W., Jr. and W.C.Krueger. Anal.Biochem. 199 (1991), 61-67. Dousseau,F. and M.Pezolet. Biochemistry 29 (1990), 8771-8779. Rahmelow,K. and W.Hubner. Anal.Biochem 241 (1996), 5-13. Vedantham,G., H.G.Sparks, S.U.Sane, S.Tzannis, and T.M.Przybycien. Anal.Biochem. 285 (2000), 3349. Hering,J.A., P.R.Innocent, and P.I.Haris. Proteomics 2 (2002), 839-849. Hering,J.A., P.R.Innocent, and P.I.Haris. Spectroscopy-An International Journal 16 (2002), 53-69. Severcan,M., P.I.Haris, and F.Severcan. Anal. Biochem. 332 (2004), 238-244. Hering,J.A., P.R.Innocent, and P.I.Haris. Proteomics 4 (2004), 2310-2319. Baello,B.I., P.Pancoska, and T.A.Keiderling. Anal.Biochem. 280 (2000), 46-57. Venyaminov,S.Y., J.F.Hedstrom, and F.G.Prendergast. Proteins 45 (2001), 81-89. Arrondo,J.L., J.Castresana, J.M.Valpuesta, and F.M.Goni. Biochemistry 33 (1994), 11650-11655. Arrondo,J.L.R., I.Etxabe, U.Dornberger, and F.M.Goni. Biochem.Soc.Trans. 22 (1994), S380. Arrondo,J.L.R. and F.M.Goni. Prog.Biophys.Mol.Biol. 72 (1999), 367-405. Smith,B.M. and S.Franzen. Anal.Chem 74 (2002), 4076-4080. Smith,B.M., L.Oswald, and S.Franzen. Anal.Chem 74 (2002), 3386-3391. Navea,S., R.Tauler, and A.de Juan. Anal.Biochem. 336 (2005), 231-242. Brauner,J.W., C.R.Flach, and R.Mendelsohn. J.Am.Chem.Soc. 127 (2005), 100-109. Orengo,C.A., A.D.Michie, S.Jones, D.T.Jones, M.B.Swindells, and J.M.Thornton. Structure 5 (1997), 1093-1108. Oberg,K.A., J.M.Ruysschaert, and E.Goormaghtigh. Protein Science 12 (2003), 2015-2031. Marsh,D. Biophys. J.72 (1997), 2710-2718. Marsh,D., M.Muller, and F.J.Schmitt. Biophys.J. 78 (2000), 2499-2510. Marsh,D., M.Muller, and F.J.Schmitt. Biophys. J.78 (2000), 2499-2510. Marsh,D. and T.Pali. Biophys J 80 (2001), 305-312. Marsh,D. J.Mol. Biol. 338 (2004), 353-367. Alegre-Cebollada,J., A.M.del Pozo, J.G.Gavilanes, and E.Goormaghtigh. Biophys. J.93 (2007), 31913201. Challou,N., E.Goormaghtigh, V.Cabiaux, K.Conrath, and J.M.Ruysschaert. Biochemistry 33 (1994), 6902-6910. Goormaghtigh,E., L.Vigneron, M.Knibiehler, C.Lazdunski, and J.M.Ruysschaert. Eur.J.Biochem. 202 (1991), 1299-1305. Le-Saux,A., J.M.Ruysschaert, and E.Goormaghtigh. Biophys J 80 (2001), 324-330. Lopes,S.C.D.N., E.Goormaghtigh, B.J.C.Cabral, and M.A.R.B.Castanho. J.Am.Chem.Soc. 126 (2004), 5396-5402. Raussens,V., V.Narayanaswami, E.Goormaghtigh, R.O.Ryan, and J.M.Ruysschaert. J.Biol.Chem. 270 (1995), 12542-12547.
128
[93] Raussens,V., J.M.Ruysschaert, and E.Goormaghtigh. J.Biol.Chem. 272 (1997), 262-270. [94] Raussens,V., C.A.Fisher, E.Goormaghtigh, R.O.Ryan, and J.M.Ruysschaert. J Biol Chem 273 (1998), 25825-25830. [95] Raussens,V., H.de-Jongh, M.Pezolet, J.M.Ruysschaert, and E.Goormaghtigh. Eur.J.Biochem. 252 (1998), 261-267. [96] Raussens,V., J.Drury, T.M.Forte, N.Choy, E.Goormaghtigh, J.M.Ruysschaert, and V.Narayanaswami. Biochem. J. 387 (2005), 747-754. [97] Sonveaux,N., A.B.Shapiro, E.Goormaghtigh, V.Ling, and J.M.Ruysschaert. J.Biol.Chem. 271 (1996), 24617-24624. [98] Sturgis,J., B.Robert, and E.Goormaghtigh. Biophys J 74 (1998), 988-994. [99] Vigneron,L., J.M.Ruysschaert, and E.Goormaghtigh. J.Biol.Chem. 270 (1995), 17685-17696. [100]Wald,J.H., E.Goormaghtigh, J.De-Meutter, J.M.Ruysschaert, and A.Jonas. J.Biol.Chem. 265 (1990), 20044-20050. [101]Abrecht,H., E.Goormaghtigh, J.M.Ruysschaert, and F.Homble. J Biol Chem 275 (2000), 40992-40999. [102]Aisenbrey,C., E.Goormaghtigh, J.M.Ruysschaert, and B.Bechinger. Molecular Membrane Biology 23 (2006), 363-374. [103]Aisenbrey,C., R.Kinder, E.Goormaghtigh, J.M.Ruysschaert, and B.Bechinger. J.Biol.Chem. 281 (2006), 7708-7716. [104]Demel,R.A., E.Goormaghtigh, and B.de-Kruijff. Biochim.Biophys.Acta 1027 (1990), 155-162. [105]Haro,A., M.Velez, E.Goormaghtigh, S.Lago, J.Vazquez, D.Andreu, and M.Gasset. J.Biol.Chem. 278 (2003), 3929-3936. [106]Houbiers,M.C., C.J.Wolfs, R.B.Spruijt, Y.J.Bollen, M.A.Hemminga, and E.Goormaghtigh. Biochim Biophys Acta 1511 (2001), 224-235. [107]Houbrechts,A., B.Moreau, R.Abagyan, V.Mainfroid, G.Preaux, A.Lamproye, A.Poncin, E.Goormaghtigh, J.M.Ruysschaert and J.A.Martial Protein Eng. 8 (1995), 249-259. [108]Leenhouts,J.M., Z.Torok, V.Mandieau, E.Goormaghtigh, and B.de-Kruijff. FEBS Lett. 388 (1996), 3438. [109]Lopes,S.C.D.N., C.M.Soares, A.M.Baptista, E.Goormaghtigh, B.J.C.Cabral, and M.A.R.B.Castanho. J.Phys.Chem. B 110 (2006), 3385-3394. [110]Martin,I., E.Goormaghtigh, and J.M.Ruysschaert. Biochim.Biophys.Acta (Biomembranes) 1614 (2003), 97-103. [111]Fa,N., S.Ronkart, A.Schanck, M.Deleu, A.Gaigneaux, E.Goormaghtigh, and M.P.Mingeot-Leclercq. Chem.Phys.Lipids 144 (2006), 108-116. [112]Goormaghtigh,E., R.Brasseur, P.Huart, and J.M.Ruysschaert. Biochemistry 26 (1987), 1789-1794. [113]Fringeli,U.P. and H.H.Gnthard. Mol.Biol.Biochem.Biophys. 31 (1981), 270-332. [114]Harrick,N.J. 1967. Interscience Publischers, New York. [115]Rothschild,K.J. and N.A.Clark. Biophys.J. 25 (1979), 473-487. [116]Ivanov,D., N.Dubreuil, V.Raussens, J.M.Ruysschaert, and E.Goormaghtigh. Biophys.J. 87 (2004), 1307-1315. [117]Scheirlinckx,F., V.Raussens, J.M.Ruysschaert, and E.Goormaghtigh. Biochem.J. 382 (2004), 121-129. [118]Raussens,V., M.Pezolet, J.M.Ruysschaert, and E.Goormaghtigh. Eur J Biochem 262 (1999), 176-183. [119]Raussens,V., M.le Maire, J.M.Ruysschaert, and E.Goormaghtigh. FEBS Lett 437 (1998), 187-192. [120]Raussens,V., V.Narayanaswami, E.Goormaghtigh, R.O.Ryan, and J.M.Ruysschaert. J.Biol.Chem. 271 (1996), 23089-23095. [121]Marsh,D., B.Shanmugavadivu, and J.H.Kleinschmidt. Biophys. J.91 (2006), 227-232.
129
FTIR Spectroscopy for Analysis of Protein Secondary Structure

a
Joachim A. HERING b and Parvez I. HARIS a,1 Faculty of Health and Life Sciences, De Montfort University, The Gateway, Leicester, LE1 9BH, UK b Department of Computer Science, University of Applied Sciences Ulm, Prittwitzstrae 10, 89075 Ulm, Germany
Abstract. One of the major challenges of the post-genomic era is the rapid characterisation of protein structure. High-throughput structural genomic projects involving X-ray crystallography and NMR spectroscopy are in progress to solve the three-dimensional structures of a large of number of proteins. These techniques have their advantages and disadvantages and cannot be applied to study all proteins, giving sufficient opportunity for other techniques to also a play significant role in proteomics research. Fourier transform infrared (FTIR) spectroscopy is one of the techniques that has gained popularity in this area since measurements on small quantities of proteins can be carried out very rapidly in various environments. However, there is a need for improvements in the interpretation of protein FTIR infrared spectra and development of methods for accurately quantifying protein secondary structure from infrared spectra of proteins. Over the years, much progress has been made in this area and here we provide an overview of the major progress made so far, along with their strengths and weaknesses. The particular focus of the Chapter is on methods used for quantitative prediction of secondary structure from infrared spectra. Keywords. Secondary structure, FTIR spectroscopy, Neural networks, Multivariate analysis, Curve-fitting
1. Introduction FTIR spectroscopy has been applied to study the secondary structure of proteins in aqueous solution for about 60 years [13]. Elliot and Ambrose were the first to demonstrate that infrared spectral data may be used to obtain information on protein secondary structure [1,4,5]. They showed that an empirical correlation between the amide I and amide II absorption bands of a protein and the secondary structure ( -helix and -sheet) contents of proteins as determined by X-ray crystallography exists. Since then, FTIR spectroscopy as a technique to determine the secondary structure of proteins has become increasingly popular especially in situations where methods such as X-ray diffraction [6] and nuclear magnetic resonance (NMR) spectroscopy cannot be readily applied [7]. Although X-ray crystallography and NMR are very precise and capable of determining the protein secondary and tertiary structure at atomic
1 Correspondence should be addressed to Parvez I. Haris, School of Molecular Sciences, De Montfort University, The Gateway, Leicester, LE1 9BH, United Kingdom, E-mail: pharis@dmu.ac.uk.
130
J.A. Hering and P.I. Haris / FTIR Spectroscopy for Analysis of Protein Secondary Structure
resolution, they are not free of limitations. Before X-ray crystallography studies can be performed to determine the three-dimensional structure of a protein at high resolution, a well-ordered crystal of the molecule is required. However, this is not possible for all proteins (e.g., the vast majority of membrane proteins). Even if it is possible, obtaining suitable protein crystals is still a difficult, time-consuming task and there is a potential risk that the condition used for crystallisation of a protein may change its structure. Although multidimensional NMR spectroscopy as an alternative to X-ray crystallography allows protein structure determination in solution, it is currently limited to the examination of small proteins with approximately 200 amino acid residues. Additionally, high concentrations of proteins are required for NMR analysis. These limitations have led to the development of alternative methods not working at atomic resolution but still capable of providing protein secondary structural information. These methods include vibrational (FTIR, Raman) and circular dichroism (CD) spectroscopy. CD spectroscopy as an alternative to FTIR spectroscopy, has been widely applied for determination of protein secondary structure [885]. In contrast to FTIR spectroscopy, CD spectroscopy can not be applied over a wide range of concentrations and is limited to only optically clear solutions. Additionally, -sheet and random coil structures have relatively small CD signals making their interpretation more prone to errors. In contrast, FTIR spectroscopy is not limited by protein size and the physical state of the sample allowing the examination of protein secondary structure in a variety of environments. This makes FTIR spectroscopy one of the few techniques that can be used to study the role of the surrounding environment on protein conformation. Additionally, FTIR spectra with high quality can be obtained relatively easily without problems of background fluorescence and light scattering. FTIR spectroscopy may also be used as a tool to distinguish between native and aggregated (unfolded) proteins [86]. All these advantages of FTIR spectroscopy, and the fact that costs of the required equipment are relatively low, have lead to the popularity of FTIR spectroscopy for protein secondary structure quantification [2,8691]. This Chapter focuses on various methods that are currently used for the prediction of secondary structure from protein infrared spectra. An excellent review on infrared spectroscopy of proteins has been published by Barth (2007) which covers theoretical and practical aspects of protein infrared spectroscopy [92]. See Chapter 1 for discussion on the historical development of infrared spectroscopy. Chapter 4 also discusses in some detail the methods used for analysis of protein infrared spectra including data processing techniques. 1.1. Amide Vibrations By far most studies related to protein FTIR spectroscopy for the analysis of protein secondary structure are using the amide bands. Altogether, up to 9 characteristic bands named amide A, B, I, II, , VII have been identified [93]. However, due to technical and theoretical limitations, only the amide I, II, and III bands are used to investigate the secondary structure of proteins. A description of amide modes has been given by Miyazawa et al. as well as by Krimm et al. [94103]. It is assumed that the exact frequencies of the amide I and II absorption is influenced by the strength of any hydrogen bonds involving amide C=O and N-H groups. Since each individual secondary structural conformation is associated with a characteristic hydrogen bonding pattern between these groups, each type of secondary structure
131
Table 1. Empirically determined, structure-sensitive regions within the amide I region. Note that the ranges given here are the union of ranges taken from a number of references -helix -sheet turns unordered H 2O 16481660 cm1 16201640 cm1 16701695 cm1 16201640 cm1 16501695 cm1 16401657 cm1 16601670 cm1 D2O 16421660 cm1 16151640 cm1 16701694 cm1 16531694 cm1 16391654 cm1 References [1,2,4,5,8688,106,111,113,140 142,227,229,230] [1,2,4,5,8689,106,111,113,140 142,227,229,230] [28,8688,103,106,111,113,227,229231] [1,4,5,8688,106,111,227,229,230]
gives rise to different frequencies at which amide bond vibrations occur resulting in characteristic amide I and II absorption. It is this separation of the amide absorption, which forms the basis of protein structure quantification from FTIR spectra of proteins. Correlation between amide frequency and protein secondary structure has been demonstrated using both normal mode calculations [103,104] and experimental studies on peptides and proteins [2,88,105,106]. However, because of the complexity of naturally occurring proteins, most of this data has been obtained from studies on model compounds with only a single secondary structure. These model compounds ranged from simple amino acid derivatives to large synthetic polypeptides. 1.1.1. Amide I Band Absorption The most widely used of these amide bands is the amide I band [2,12,87,88,106111]. Amide I absorption is directly related to the backbone conformation with major contribution from C=O stretching vibration and minor contribution from the C-N stretching vibration. Absorption for this band occurs in the region 16001700 cm1. Empirical studies have identified regions within the amide I band, which are sensitive to particular secondary structural conformation (see Table 1). In most cases, good relationship exists between protein secondary structure and respective amide I frequencies. However, this correlation does not generally apply to all proteins and peptides. For example, deviations may occur with proteins containing unusual or less common structures. Examples for -helix structure in H2O include poly-L-lysine where absorption around 1638 cm1 has been reported [112,113] and bacteriorhodopsin absorbing at 1662 cm1 [113]. Possibility of significant difference of absorbing band frequency for -helix structure between short, solvent exposed peptides and helix structure buried within a solvent inaccessible region of a highly folded globular protein has been pointed out [113]. For example, absorption for small peptides has been shown to occur at 1632 cm1 [114]. Additionally, absorption from amino acid side chains, steric situations, and dielectric properties of the solvent are known to influence the frequency of amide vibrations [115118]. Table shows how a variety of amide I band frequencies has been attributed to helical structure in the literature. Figures 13 shows the absorbance and second-derivative spectra of three proteins with differing secondary structures. Spectra were obtained for proteins dissolved in H2O. The absorbance and second-derivative spectra of a predominantly -helical protein (cytochrome c), a predominantly -sheet protein (prealbumin) and a protein with a mixture of -helical and sheet structure are presented.
132
A
B
1655
Absorbance/Wavenumber
Absorbance
1679
1700
1680
1660
1640
1620
1600
1700
1680
1660
1656
1640
1632
1620
1615
1600
Wavenum ber
Wavenum ber
Figure 1. FTIR spectrum of Cytochrome C from horse heart (A) and its second derivative spectrum (B). Cytochrome C is a predominantly helical protein as is evident from the main amide I band maximum at 1656 cm1. Parameters used for derivation: Savitzky-Golay, 2nd, Points 13.
A
1632
Absorbance
1688
1673
1652
1700
1680
1660
1640
1620
1600
1700
1680
1660
1640
1631 1620
1600
Wavenum ber
Wavenum ber
Figure 2. FTIR spectrum of Prealbumin from human plasma (A) and its second derivative spectrum (B). Prealabumin is a predominantly beta-sheet proten which is conistent with the amide I maximum at 1631 cm1. Parameters used for derivation: Savtizky-Golay, 2nd, Points 13.
A
1650
B
2
Absorbance
1685
1672
1700
1680
1660
1640
1620
1600
1700
1680
1660
1650
1640
1633
1623
1620
1615
1600
Wavenum ber
Wavenumber
Figure 3. FTIR spectrum of Papain from papaya latex (A) and its second derivative spectrum (B). Papain is a protein with a mixture of -helical and -sheet structure evident from the strong bands at 1650 cm and 1633 cm1, respectively. Parameters used for derivation: Savitzky-Golay, 2nd, Points 13.
The relatively high intensity of the amide I band, compared to other amide vibrations, has been an important factor behind its use for protein secondary structure analysis. However, it also happens to occur in a region most prone to errors due to the overlap with the O-H stretching band at 1640 cm1. Furthermore, interference from absorp-
133
tion of water and water vapour mainly within the amide I region used to be a significant problem, since H2O strongly absorbs in that region due to OH vibrations. As a result, in the past most infrared studies were restricted to measurements in the solid state or in 2H2O. However, since it is now possible to digitally subtract overlapping H2O absorption from the spectrum of the protein solution, this is not a problem anymore provided that a relatively high protein concentration of about 10 mg/ml and short path length cells of 6m are used for recording spectra of proteins in H2O. Additionally, purging the sample compartment with dry air or nitrogen is generally performed to eliminate water vapour inside the sample compartment allowing good spectra with high signal-to-noise ratio to be recorded. Over or under subtraction of liquid water and water vapour absorbance has been a very common problem encountered by protein infrared spectroscopists, especially when working with low peptide and protein concentrations. Complications arising from absorbance from non-protein moieties has also led to erroneous assignments of peaks in the amide I region. This is mainly due to presence of molecules that are linked with various steps in the isolation and purification of proteins (for example detergents, buffers) and peptides (e.g. acids). For example, several studies in the literature attributed a band at 1674 cm1 to secondary structural elements when in reality it arises from carboxyl group of trifluoroacetic acid (TFA) that remains strongly bound to peptides after their synthesis & purification. Most experienced protein infrared spectroscopists are now aware of this problem and remove the bound TFA by acid treatment prior to infrared analysis. 1.1.2. Amide II Band Absorption Amide II absorption results both from N-H bending vibration and from C-N stretching vibration. The absorption for this band occurs in the region 15001600 cm1. Empirically, the amide I absorption has been found to be more useful for protein secondary structure determination than the amide II absorption [2,12,87,88,106111]. However, its inclusion with the amide I band has been reported to provide improved prediction accuracy by some workers using multivariate data analysis techniques [117,119121]. 1.1.3. Amide III Band Absorption Determination of protein secondary structure has also been demonstrated based on the amide III region [122126]. Absorption for this band occurs in the region 12201330 cm1. Amide III absorption mainly arises from C-N stretching vibrations as well as N-H in-plane bending vibrations, with weak contributions from C-C stretching and C=O in-plane bending vibrations [93]. The amide III region has been characterised to be a less well defined vibrational mode, with contributions from different vibrations varying between proteins [103]. Additionally, signal contribution arising from amide vibration within the amide III region has been found to be very weak and extensively mixed with CH vibration of amino acid side chains [21]. However, despite relatively weak intensity in the amide III region, there are no interfering OH vibrations from water. Generally, -helix structure occurs in the region 12931328 cm1, -sheet in the region 12251250 cm1, and unordered structures in the region 12571288 cm1 [127].
134
1.1.4. Interference from Amino Acid Side Chain Absorption Absorption from a number of amino acid residues has been found to occur in the amide I and amide II regions which may influence protein secondary structure prediction from FTIR spectra of proteins [115,116,128]. Effects of amino acid side chain absorption on secondary structure quantification have been investigated based on proteins in 2 H2O and proteins in H2O [115,116,128]. Chirgadze et al. report amino acid side chain absorption of asparagine, glutamine, aspartic acid, glutamic acid, arginine, and tyrosine in the region 15001800 cm1 in 2H2O [115]. Venyaminov and Kalnin report amino acid side chain absorption of asparagine, glutamine, aspartic acid, glutamic acid, arginine, lysine, tyrosine, histidine, and phenylalanine in the region 14001800 cm 1 in H2O [116]. In both studies, band assignments and intensities were established based on curve fitting procedures assuming Gaussian/Lorentzian lineshapes. Rahmelow et al. have investigated amino acid side chain absorption in the region 14401800 cm1 based on infrared spectra of 9 amino acids, 23 dipeptides, 7 tripeptides, 7 tetrapeptides, and three polypeptides in aqueous solution [128]. These samples were chosen such that for each absorbing amino acid residue at least three different compounds were measured. They used an inverse matrix method to show that protein secondary structure prediction accuracy may be improved by subtracting amino acid side chain absorption of the residues asparagine, glutamine, aspartic acid, glutamic acid, arginine, tyrosine, and lysine from the amide I and amide II regions. Side chain contribution has been subtracted based on spectra from model compounds in aqueous solution. However, as Barth and Zscherp state, this may be problematic for side chains not exposed to the surrounding aqueous environment, since the influence of the protein on the spectral characteristics of these side chains is unknown [86]. Although spectral characteristics of amino acid side chain absorption was obtained by a linear matrix model as opposed to a curve fitting procedure used by Venyaminov and Kalnins study [116], good agreement of results has been reported. Simonetti and Di Bello propose a method based on a traditional curve fitting approach, utilising FTIR isotopic exchange techniques in the presence of organic solvents incapable of donating hydrogens [129,130]. Based on a synthetic fragment of proocytocin [129] as well as a series of synthetic fragments corresponding to the processing site of the proocytocin-neurophysin precursor [130], they demonstrated that interference from amino acid side chain absorption could be reduced after H-D exchange in the presence of dimethylsulfoxide. Additionally, quantification of -turn structure is facilitated. 1.1.5. Developments in Amide Band Assignments Apart from empirical studies for identifying secondary structure sensitive regions, computational approaches have been suggested. Pancoska, Kubelka et al. suggest a modification of Nodas algorithm [131] for calculating two-dimensional correlation maps to identify spectral regions associated with a specific secondary structure [23,26]. In their approach, the two-dimensional maps have been generated by fitting the intensity variance at each frequency by a polynomial. Their method has been applied to protein spectral data from FTIR, CD, and Raman spectroscopy. Recently, we introduced an automatic amide I frequency selection procedure based on a hybrid between genetic algorithms and neural networks [132]. Based on a reference set of 18 spectra from proteins in H2O, this procedure identified frequen-
135
cies within the amide I band of protein FTIR spectra, which could be best related to secondary structure contents by subsequent neural network analysis. Similar approaches have been suggested based on the combination of Genetic Algorithms with multivariate data analysis techniques for a variety of prediction problems in chemistry [133139].
2. Quantitative Estimation of FTIR Spectroscopy For recent articles on quantitative estimation of protein secondary structure based on FTIR spectra, see [29,86,92,113,140146]. To date, existing methods for protein secondary structure quantification from FTIR spectral data fall into two main categories: Those based on band narrowing and decomposition of mainly the amide I band shape into its underlying components often referred to as frequency based or curve fitting approaches and those based on the principle of pattern recognition. Of the pattern recognition based approaches, most work has been done using multivariate data analysis techniques [12,16,109,110,117,120,121,146]. Alternative pattern recognition approaches used are based on the principle of artificial neural networks. All of the techniques suggested for protein secondary structure quantification from FTIR spectra of proteins have their advantages and disadvantages [113]. There is clearly a great need for further work to improve current methods or to develop additional approaches to achieve better quality of secondary structure prediction from infrared spectral data. Here, the most widely used methods for protein secondary structure quantification will be critically reviewed and compared. Additionally, a summary of prediction accuracy achieved by different methods in predicting protein secondary structure contents from their FTIR spectra in terms of the standard error of prediction (SEP) for each secondary structure under investigation is given (see Table 2). The same definition as that given in Lee et al.s paper [109] is used:
SEP =
( p
j =1
cj
pxj )
Equation 1.
where pcj = the proportion of structure predicted for left-out protein j by the respective method, pxj = the proportion of structure calculated from the original X-ray data for protein j, and n = the number of proteins. 2.1. Curve Fitting Curve fitting, the most widely used method for protein secondary structure quantification, mainly involves curve fitting of the amide I band, e.g., [2,87,88,147154] as well as occasionally the amide III band, e.g., [124,126,152]. This procedure has been reported to provide good estimation of protein secondary structure (see Table 2). The basic principle of the curve fitting procedure is to resolve the original protein spectrum into individual bands that fit the spectrum. However,
Table 2. Comparison of various secondary structure prediction methods from FTIR spectra in terms of the SEP Data set size 11 6 14 12 14 14 18 13 17 21 39 23 13 8 23 18 18 18 50 FTIR spectra Spectral region used combined (FTIR) for best with CD resultsa data No No No No No No No No No Yes No No No No No No No No No I I I I I + II I + II I I + II I I I + II I + II I (18001600 cm1) I I + II + I + II I I I Selected Method used for calculating target secondary structure fractionsb LG LG LG LG R R LG LG/KS KS KS KS KS LG/KS ST KS LG KS LG KS Prediction SEP for SEP for method helixd (%) sheetd (%) usedc C C C C C M M M M M M M M M M N N N M 2.17 2.24 10.31 5.76 5.95 5.57 7.8 5.11 9.8 7 12.14 8.6 8.32 2.1 5.34 7.7 4.58 4.47 5.5 6.34 2.76 2.55 6.87 6.82 2.56 2.4 9.7 3.71 11.22 9.5 9.08 7.34 8.79 2.9 6.33 6.4 5.72 6.16 6.6 6.18 SEP for turns (%) NPe NP NP 8.07 3.99 3.51 4.3 NP 6.61 7 NP 1.39 6.48 3.7 3.39 4.8 4.42 4.61 3.4 4.69 SEP for bends (%) NP NP NP NP NP NP NP NP NP NP NP 3.55 NP NP 4.19 NP 3.95 NP NP 3.9 SEP for otherf (%) NP NP NP 6.02 3.27 3.73 NP 5.14 9.18 10 NP 3.79 8.77 2.3 5.37 NP 6.12 NP 8 5.97 Average of SEPs (%) 2.47 2.4 8.59 6.67 3.94 3.80 7.27 4.65 9.20 8.38 10.61 4.93 8.09 2.75 4.92 6.3 4.96 5.08 5.88 5.84
136
Ref. [2] [87] [168] [108] [119] [169] [109] [117] [110] [12] [120] [121] [180] [184] [16] [194] [132] [196] [146] Average
a b
Year 1986 1986 1990 1990 1994 1996 1990 1990 1991 1993 1996 1997 1998 2000 2000 2001 2001 2002 2006
I: amide I; II: amide II; I + II: Both amide I and amide II region; I + II + I + II: amide I, amide II, amide I, and amide II regions were used. LG: Levitt and Greer [170]; KS: Kabsch & Sanders DSSP [177]; R: Ramachandran plots, ST: STRIDE [228]. c C: Curve fitting; M: Multivariate data analysis; N: Neural network analysis. d If the secondary structure has been further divided (e.g., parallel, anti-parallel -sheet), the average is taken. e NP: This type of secondary structure class has not been predicted. f This structural class is also often referred to as unordered, random coil, random, irregular, and undefined.
137
before the curve fitting procedure can be performed, band narrowing techniques need to be applied to obtain estimates of the number and positions of discrete absorption that make up the complex amide I band profile. Since the outcome of the curve fitting procedure is highly dependent on the outcome of these estimates, the most widely used band narrowing techniques, namely Fourier self-deconvolution (FSD) and derivation, are briefly discussed first. 2.1.1. Band Narrowing Techniques Since most proteins usually contain more than one secondary structure, they give rise to several amide absorption bands. A considerable amount of overlap in terms of width and separation of these absorption bands often results in featureless absorption profiles. This makes it difficult to analyse the frequency of the composite amide maximum and any visible shoulders, which may lead to misinterpretation of spectral shifts [155]. Jackson & Mantsch therefore argue, that deduction of structural parameters from the relatively featureless amide I band alone is of limited use with regards to the curve fitting method [142]. In an attempt to tackle this problem, a number of mathematical data processing techniques have been developed to visualise those overlapping bands allowing to extract detailed information from infrared spectra of proteins [156161]. These techniques are often referred to as resolution enhancement techniques. However, since resolution is an instrumental parameter that cannot be increased after a spectrum is recorded, these techniques are more correctly referred to as band narrowing techniques, since they mainly involve narrowing the widths of infrared bands, allowing increased separation of the overlapping components. The two most popular of those techniques are second-derivative (see Figs 13) and Fourier self-deconvolution (see Figs 4, 5). Fourier self-deconvolution (FSD) [156,157,162,163] is based on the assumption that in the liquid or solid state, the absorption bands are broadened (or convoluted) by a function such that bands overlap and cannot be distinguished in the amide envelope. The self-deconvolution procedure uses this function to narrow (or deconvolute) the spectrum. Although the exact shape of the convolution function is still not determined, most authors have assumed Lorentzian or Gaussian functions [2,87,108,119]. The basic underlying principle of FSD is described elsewhere [88,156,157,162,163]. However, for our discussion it is important to note, that FSD is controlled by two parameters: The full-width at half-height (FWHH) and the resolution enhancement factor (K). Both have to be determined manually, mainly by trial and error. The exact number and frequency of the resulting components is highly dependent on the choice of deconvolution parameters. Different combinations of values for the FWHH and for K yield different shapes of the resulting deconvoluted spectrum. E.g., if the FWHH is too small, little narrowing is obtained. If it is too large, negative side-lobes appear. Since the choice of the deconvolution parameters is subjective, varying results of subsequent curve fitting may be expected for the same protein FTIR spectrum under investigation. Spectral derivation an operation similar to deconvolution has been originally realised by the method of Savitzky & Golay [164]. In the area of spectral analysis, second order derivative is most widely applied. The Savitzky-Golay method uses information from a localised segment of the spectrum to calculate the derivative at a particular wavelength rather than the difference between adjacent data points. In most cases, this reduces the problem of noise enhancement and may actually apply some smoothing to
138
Ovalbumin
Original
AU
FSD
SD
1800
1700
1600
1500
-1
1400
1300
WAVENUMBER (cm )
Figure 4. The original FTIR spectrum of ovalbumin in H2O solution (top), the Fourier Self-Deconvolution spectrum (middle), and the inverted Second Derivative spectrum (bottom). Fourier Self-Deconvolution was performed with a full-width at half-height of 24.0 cm1 and a resolution enhancement factor of 2.6. Second Derivative spectrum was obtained by applying the Savitsky-Golay method with 7 convolution points.
the data. However, large side lobes on either side of intense absorption bands are produced making it difficult to determine the true limits of these absorption bands. A significant problem of spectral derivation is that it does not preserve relative intensities of absorption bands. Relative absorption bands in derivative spectra are strongly dependent on the width of the absorption in the original spectrum where narrow absorption bands will be enhanced at the expense of broader bands. For example, a broad band in the original spectrum will be reduced to hardly any feature at all after derivation. Due to these problems, derivative spectra are often merely used to confirm the initial identification of the band positions by deconvolution [143]. However, successful application of curve fitting directly based on derivative spectra has been demonstrated recently, e.g., [22,108,165,166]. Since both spectral derivation and FSD are very sensitive to changes in the spectra, it is possible that noise may be amplified using these methods [167]. Therefore, Singh suggests that curve fitting should not be performed on a second derivative and/or deconvolved spectrum directly [143]. They should merely be used to determine the parameters of the curve fitting procedure (i.e., the number of component bands and their positions). 2.1.2. Curve Fitting Procedure A complete description of the curve fitting procedure can be found elsewhere, e.g., [143]. Basically, it involves an iterative process, where a set of Gaussian, Lorentzian, or a mixture of Gaussian/Lorentzian-shaped curves is determined, that best fits the original protein spectrum. The best fit is obtained by a root mean square analy-
139
d A/d
T S
1700
1680
1660
1640
1620
1600
WAVENUMBER (cm-1)
Figure 5. Curve-fitting of the inverted second derivative spectrum from ovalbumin in H2O and assignment to protein secondary structure. Table 3. Frequencies, relative areas and assignments of second-derivative infrared amide I components of ovalbumin in H2O solutiona Assignment -sheet turns turns -helix -sheet -sheet a The amide I band assignments were made on the basis of previous FT-IR spectroscopic studies of other globular proteins in H2O solutions [232]. Frequency (cm1) 1697.2 1686.0 1675.5 1657.4 1638.2 1624.7 Band Area (%) 1.5 12.2 6.6 36.6 30.8 12.3
sis determining the optimal set of curve fitting parameters (band height, band width, band position, and baseline). The band area or intensity of each individual band is used to calculate its relative contribution to a particular protein secondary structure in relation to the overall area of the original spectrum. It should be noted, that variations of the just described curve fitting process have also been applied [2,87,108,119,124,143,168,169]. An example of curve-fitting applied to the inverted second derivative spectrum from ovalbumin in H2O along with assignments to protein secondary structure are provided in Fig. 4 and Table 3, respectively. Susi & Byler [2,87] have investigated the quantitative estimation of helix and sheet structure from second derivative and deconvolved FTIR spectra using curve fitting applied to the deconvolved FTIR spectra directly. They have reported good agreement with secondary structure calculations from X-ray crystallography by Levitt & Greer [170]. For six proteins, their reported predictions have resulted in an SEP of 2.24% for helix and an SEP of 2.55% for extended chain structure [87]. For 11 of the
140
proteins covered by Levitt and Greer [170], their reported estimations have resulted in an SEP of 2.17% for helix structure and an SEP of 2.76% for beta structure [2]. The resulting averages of SEPs were 2.4% and 2.47% respectively. In contrast to Susi et al.s procedure where the curve fitting parameters were chosen manually, Goormaghtigh et al. [168] have employed an automatic procedure for choosing the curve fitting parameters in an attempt to make the curve fitting procedure more objective and more accessible to other investigators. Curve fitting was performed on spectra obtained with very little deconvolution. Goormaghtigh et al. were interested in secondary structure prediction based on thin hydrated films of 17 proteins by attenuated total reflection (ATR) spectroscopy. The structural properties of interest were helix, -sheet, -turn, and random structure. The target fractions of secondary structure were calculated based on data from Levitt & Greer [170]. However, since these target fractions of secondary structure have only been given for 14 proteins as well as only for -helix and -sheet structure, the SEPs will be reported for these only. The application of their more objective automatic procedure resulted in worse SEPs as compared to Susi & Bylers work. Using their method, an SEP of 10.31% for -helix and an SEP of 6.87% for -sheet was achieved. The average of SEPs is 8.59%. Dong et al. [108] have investigated the quantitative estimation of -helix, -sheet, turn structure, and unordered for twelve globular proteins in aqueous solution. They have performed the estimation directly from the band areas (integrated intensities) of the second derivative spectra and have subsequently compared them with the amounts obtained by Levitt & Greer from X-ray crystallography [170]. For the twelve proteins investigated, their reported predictions have resulted in an SEP of 5.76% for -helix, an SEP of 6.82% for -sheet, an SEP of 8.07% for turn, and an SEP of 6.02% for unordered. The average of SEPs is 6.67%. Note, that for hemoglobin and myoglobin they have reported that the band due to unordered structure appeared as a shoulder on the -helix band and was therefore too small (less than 5%) to be separated from the helix structure. Hence, it was included in the -helix value. For the calculation of the SEP values, we have therefore set the fraction of the unordered structure to 0. Also, for major histocompatibility complex antigen A2 and 2-microglobulin, no X-ray data was given for turn and unordered structure. Therefore, these values have been excluded from the calculation of the SEPs in Table 2. In another approach Singh et al. have used the deconvolved and second derivative spectra for peak assignment, followed by curve fitting on the original spectrum [124,143]. Unfortunately, we were unable to calculate the SEPs from the data provided. Kumosinski & Unruh [119] have estimated the secondary structure proportions of 14 globular proteins by curve fitting both amide I and amide II bands. They have used second derivative spectra for identifying individual peak positions. The original spectra have then been subjected to Fourier deconvolution with subsequent curve fitting. The secondary structural properties of interest were helix, sheet, turn, and irregular structure. The target secondary structure proportions were based on traditional Ramachandran plots calculated from the X-ray crystallographic structure of their proteins in conjunction with data from the Protein Data Bank (PDB). Note that their analysis involved manual assignment of up to 28 component peaks to secondary structure, which is a good example of the large amount of subjectivity inherent in the curve fitting method.
141
Applying their method to the 14 proteins under investigation resulted in an SEP of 5.95% for -helix, an SEP of 2.56% for -sheet, an SEP of 3.99% for turn structure, and an SEP of 3.27% for irregular. The average of SEPs is 3.94%. In a similar study, Kumosinski and Unruh have performed a Gauss-Newton nonlinear iterative curve-fitting analysis based on 14 globular proteins and two synthetic polypeptides with known secondary structure from Ramachandran analysis of the X-ray crystallographic structure [169]. They have fitted both the amide I and amide II bands. Again, FSD spectra, second-derivative spectra, and the original spectra were used for their analysis resulting in an SEP of 5.57% for helix, an SEP of 2.4% for sheet structure, an SEP of 3.51% for turn structure, and an SEP of 3.73% for irregular. The average of SEPs is 3.8%. Summarising, it can be said that there are a number of assumptions implicit in the curve fitting approach that are not necessarily justified [88]. The real number and position of individual amide I band contours present in a protein is not guaranteed to be accurately reflected in the number obtained from deconvolution or derivation. Hence, there is a potential source of error in the assignment of each absorption. Additionally, it is often assumed that unique infrared frequencies can be uniquely correlated to specific structural types [10,167]. However, the assignments of particular amide I absorption to specific secondary structures is not always clear and often subjective, i.e., all the bands within the amide I region need to be assigned to secondary structure types manually. Based on a survey of the current literature on protein infrared spectroscopy, it has been shown that there are in fact conflicting amide I assignments for the different types of protein secondary structure [113]. The curve fitting method usually assumes Lorentzian or Gaussian band shapes (or a mixture of those). This assumption, however, may not be true for complex molecules such as proteins. It is also not clear to what extent environmental effects such as the hydrogen bond strength are important in determining band shape. 2.2. Pattern Recognition Based Methods Alternative methods for quantitative analysis of protein secondary structure have been introduced that require fewer assumptions and remove much of the subjectivity inherent in the curve fitting method described above. The basic underlying principle is to extract common features or patterns associated with fractions of secondary structures from a reference set of spectra with known secondary structure mostly from X-ray crystallography studies. Hence, these techniques are often referred to as pattern recognition based techniques, a category, which encompasses both multivariate data analysis and neural network methods. 2.2.1. Multivariate Data Analysis Techniques The most popular methods of the pattern recognition based approaches are the multivariate data analysis methods. Multivariate data analysis methods basically involve relating two sets of data, commonly expressed in two matrices X and Y, by regression. Typically, one set of data consists of target secondary structure fractions for proteins in the reference set (Y) and the other set consists of the absorption values for a range of wavenumbers characterising these proteins (X). Typically, rows in X are the n proteins of the reference set and each column in X represents one variable characterising the respective protein. Generally, when dealing with data from FTIR spectroscopy, X con-
142
tains p variables, one for each wavenumber whose values are absorption values for the respective wavenumber of the reference protein spectra. Following multivariate data analysis terminology, one protein infrared spectrum in X is specified by a point in pdimensional space, where p is the number of wavenumbers recorded. X and the corresponding Y together are often referred to as the calibration set or training set. Based on this calibration set, a multivariate regression model is derived which can then be used to predict the secondary structure fractions of a new protein based on its infrared spectral data, i.e. the built model is used on new X to predict the new Y. These methods have been pioneered by applications based on spectral data from other techniques like CD, e.g., [21,31,3346] and Raman spectroscopy, e.g., [171176]. The classical regression method, multiple linear regression (MLR), may lead to serious misinterpretations which may even remain undiscovered, when the X-variables are intercorrelated (linearly dependent) and when there is noise, i.e., errors in X. This lead to the development of more advanced projection based regression methods like Principal Component Regression (PCR) and the Partial Least Squares (PLS) methods. These methods avoid the problem of intercorrelation and are more capable of dealing with significant errors in the spectral data (X). As already mentioned, protein infrared spectra of the reference set can be thought of as a swarm of points in p-dimensional space, where p is the number of variables in the original spectral data matrix X. The projection based methods involve transformation of these original p dimensions into another coordinate system with fewer dimensions still describing the largest variation in the original data. This is achieved through projection. The basic assumption is that this new object space still corresponds to the most useful part of the data, i.e. the main multivariate trends. The dimensions of the original data set that contribute the least to the variation in the data set are eliminated during the process. They are assumed to represent the noise part of the data. Hence, this process is often referred to as decomposition of the original data matrix into a structure part and a noise part. The probably most important property of the variables comprising the new object space describing the protein spectra is that they are orthogonal, i.e., linearly independent. Most widely used methods employed for achieving the just described transformation include Principal Component Analysis (PCA), Factor Analysis (FA), and Singular Value Decomposition (SVD) methods that are closely related. As a result, in projection based regression methods, intercorrelated data sets may be modelled without difficulty. This makes spectroscopic data ideal applications of these projection methods since it is usually highly intercorrelated. Lee et al. [109] have applied a regression method based on a factor analysis method to a set of 18 FTIR spectra from proteins in H2O. The target fractions of secondary structure were calculated based on data from Levitt & Greer [170]. The structural properties of interest were -helix, -sheet, and turn structure. The quality of prediction was evaluated using the leave-one-out method. Using the normalised region of the amide I band for the 18 proteins, their method achieved an SEP of 7.8% for -helix, an SEP of 9.7% for -sheet, and an SEP of 4.3% for turn. The average of SEPs is 7.27%. Dousseau & Pezolet have investigated the application of both a classical least squares method (MLR) and a partial least squares method (PLS) for the estimation of protein secondary structure for -helix, -sheet, and undefined structure based on FTIR spectra from 13 proteins in H2O and 2H2O [117]. They were interested in the performance of the employed methods based merely on the amide I band as well as the
143
combination of the amide I and amide II band. The target fractions of secondary structure were calculated based on data from Levitt & Greer [170], except for myoglobin, for which the results of Kabsch and Sander [177] were used. Based on the results published in their paper (Table II) [117], we have calculated the performance of their presented methods in terms of the SEP. PLS proved to be more appropriate than the classical linear regression method (MLR). The best results were achieved by employing PLS, breaking down the -helix structure into ordered and disordered -helix structure, and by the inclusion of the data points both from the amide I and amide II region. This resulted in an overall SEP of 5.11% for -helix, an SEP of 3.71% for -sheet, and an SEP of 5.14% for unordered. The average of SEPs is 4.65%. Sarver and Krueger [110] employed a method based on single value decomposition to predict the secondary structure fractions for helix, -sheet, -turn, and other from the amide I region of FTIR spectra from 17 proteins. The target fractions of secondary structure were calculated using Kabsch & Sanders DSSP program [177]. For the 17 proteins, their reported secondary structure estimations using the leave-oneout method resulted in an SEP of 9.80% for helix, an SEP of 11.22% for -sheet, an SEP of 6.61% for -turn, and an SEP of 9.18% for other. The average of SEPs is 9.2%. Pribic et al. suggested an approach, where both FTIR and CD spectra of 21 reference proteins of known structure were used to generate spectra characteristic of helix, anti-parallel -sheets, parallel -sheets, -turns, and other [12]. Quantification was performed by applying a multivariate linear model (Gauss-Markoff model) in combination with singular value decomposition. The underlying assumption was that a protein spectrum to be analysed is a linear superposition of the characteristic spectra. This approach allowed for the combination of different spectral regions in a single analysis. The target fractions of secondary structure were calculated using Kabsch & Sanders DSSP program [177]. The secondary structure fractions were predicted with varying spectral regions from separate spectra as well as from combined FTIR and CD spectra. Their predictions were evaluated by employing the leave-one-out method. Pribic et al. [12] demonstrated a slightly improved quantification by combining the 21 FTIR spectra (amide I region) with circular dichroism CD spectra of the same proteins and using this extended spectrum for the analysis. This resulted in an SEP of 7% for helix, an SEP of 9.5% for sheet structure, an SEP of 7% for turn structure, and an SEP of 10% for other. The average of SEPs is 8.38%. Rahmelow and Hbner have investigated the accuracy of secondary structure prediction based on 39 FTIR spectra from proteins with known structure from X-ray crystallography by applying different multivariate data analysis techniques, namely, classical and inverse least squares, singular value decomposition, partial least squares, and ridge regression [120]. The structural properties of interest were helix and -sheet. The structural proportions were calculated from X-ray crystallography data using the DSSP program based on the work from Kabsch & Sander [177]. The quality of the investigated methods was expressed in terms of a slightly modified version of the standard error of prediction as used here (see Eq. 1).
SEPRahmelow / Hbner =
( p
j =1
cj
pxj )
n 1
Equation 2. SEP as defined by Rahmelow and Hbner.
144
where pcj = the proportion of structure predicted for left-out protein j by the respective method, pxj = the proportion of structure calculated from the original X-ray data for protein j, and n = the number of proteins. However, knowing that n in their investigation was 39, the SEP using our definition could be calculated using the following equation:
SEPOur _ Definition = 38 ( SEPRahmelow / Hbner )2 39
Equation 3. Calculation of the SEP as defined by us.
This enabled us to express their results in terms of the SEP as defined here. Applying singular value decomposition with seven primary factors, the best prediction of secondary structures was obtained when both the amide I and amide II bands were included in the calculations. An SEP of 12.14% for helix and an SEP of 9.08% for -sheet were achieved. The average of SEPs is 10.61%. Based on the results presented in their paper, they have concluded, that the methods singular value decomposition, partial least squares, and ride regression are very similar in terms of their prediction capabilities and should be given priority over the less robust classical and inverse least squares methods. They have also acknowledged that the prediction accuracy is not limited by the applied procedure, but by the quality of the data set. Finally, a decomposition of the data set into subgroups with similar spectra as performed by a cluster analysis did not further improve prediction accuracy. Wi et al. [121] have studied the effect of an increased band shape variation of the amide I and amide II regions of 23 FTIR protein spectra in H2O on the prediction accuracy of their factor analysis based restricted multiple regression method. The increased band shape variation was achieved by applying Fourier self-deconvolution (FSD) to the spectral data. The structural properties of interest were helix, sheet, turn, bend, and other. The structural proportions were calculated using the DSSP program based on work from Kabsch & Sander [177]. Their prediction method was based on a principal component method of factor analysis (PC/FA) where the spectral data of a training set is decomposed it into linear combinations of a set of orthogonal component spectra. They claim that the loadings of these component spectra, for each protein, provide a compact numerical description of spectral bandshape variability, and form a pool of statistically significant spectral descriptors. From these descriptors the optimal subset that can be correlated to the fraction of each type of secondary structure was selected using a restricted multiple regression (RMR) analysis where details can be found in [13]. They recognised the fact that the choice of optimal deconvolution parameters is a key and quite subjective step of the type of spectral analysis they used. They have therefore varied both deconvolution parameters in a series of steps with subsequent application of their FA/RMR analysis. Since the prediction quality varied depending on the deconvolution parameters used, they have defined the optimally deconvolved set as that one which gives the best improvement in the prediction of protein secondary structure. They have acknowledged, however, that in practice this optimally deconvolved set may be different for different secondary structure fractions (helix, sheet, etc.) as well as for different spectral data. The quality of prediction was evaluated using the leave-one-out method. In their article, they report that after optimisation, FSD has only little impact on the prediction
145
of helix and sheet fractions. However, using a moderately deconvoluted set in combination with a number of component spectra of 16 retained in the pool to select which loadings to be used in the selective multiple linear regression (RMR) optimisation, best overall results were obtained. This resulted in an SEP of 8.6% for helix, an SEP of 7.34% for sheet, an SEP of 1.39% for turn, an SEP of 3.55% for bend, and an SEP of 3.79% for other. The average of SEPs is 4.93%. Note that since only the detailed results for 22 of the 23 proteins were given in their paper, the SEPs calculated here were based on these 22 proteins. Baello et al. [16] have presented an equilibrium hydrogen exchange FTIR spectroscopy method for the prediction of secondary structure proportions. The structural properties of interest were helix, sheet, turns, bends, and other. The structural proportions were calculated from X-ray crystallography data using the DSSP program based on work from Kabsch & Sander [177]. In their studies, the amide I, I, II, and II regions were used resulting in a range from 18001350 cm1. In carrying out their method, four basic steps were performed: First, for each protein, six spectra were measured with a systematic variation of the solvent H-D ratio. Second, this set of spectra was subjected to factor analysis to determine the most significant component spectra for each protein. Basically, this step aimed at extracting independent aspects of the spectral response to deuteration. Third, these component spectra were subjected to a second factor analysis over the entire training set to determine components of the bandshape based on their commonality over the training set of component spectra provided by the third step. The loadings of each resulting component spectrum are normally used to fit to the secondary structural proportions as determined by X-ray crystallography. The resulting fitting relationship can subsequently be used to predict the secondary structural fraction of an unknown protein. Hence, in a last step, Baello et al. have used restricted multiple regression analysis to selectively choose those loadings of the resulting component spectra from step three resulting in the most reliable predictions for each structural type independently. The quality of prediction was evaluated using the leave-one-out method, where one protein was systematically removed before developing a regression relation between the loadings and the fractions of the secondary structural components under investigation (i.e., helix, sheet, turns, bends, other). The predicted secondary structure fractions of the protein left out were then calculated. The leave-one-out method was restricted to only 19 proteins, since only for those proteins good X-ray crystal structure analysis had been available. The optimal predictions were then determined for the set of loadings that resulted in the lowest prediction error for the entire set. The resulting differences of secondary structure predictions from the secondary structure calculated based on X-ray crystal structure analysis for each of the 19 proteins was given in their paper, enabling us to calculate the corresponding SEPs. Their method achieved an SEP of 5.34% for helix, an SEP of 6.33% for sheet, an SEP of 3.39% for turns, an SEP of 4.19% for bends, and an SEP of 5.37% for other. The average of SEPs is 4.92%. Recently, Wu et al. have used two-dimensional IR correlation spectroscopy (using time-dependent spectral variations) in combination with PCA to investigate the secondary structure and the kinetics of H-D exchange of human serum albumin in D2O [178]. In their study, they made use of the fact that the amide protons of each secondary structural conformation are not exchanged at the same time. Hence, contributions of secondary structures to amide bands can be separated using H-D exchange studies. They, they report that for human serum albumin, H-D took place in the follow-
146
ing order: Extended chain and -turns first (after a few minutes), the water accessible parts of -helices (after about 6 minutes). H-D exchange for -turns continued until 13.4 minutes after initiation and for 25% of the -helix structure, no H-D exchange took place even after 4 hours after initiation at 25C. PCA was used both to separate the acquired spectra into different groups and to deconvolute amide I and amide II bands (using loadings plots). Filosa et al. have used resolution-enhanced two-dimensional infrared correlation spectroscopy to study structural changes in response to change in temperature of structurally homologous proteins, namely horse, cow, and tuna ferricytochromes c [179]. Despite high similarity of the respective sequences, Filosa et al. found that ferricytochrome c from horse and cow had different thermal unfolding pathways. Forato et al. applied singular value decomposition (SVD) to FTIR spectra of 13 globular proteins in KBr pellet [180]. Based on their results, they claim, that protein secondary structure is preserved in solid state. Target secondary structure has been calculated based both on Levitt and Greers method [170] and Kabsch and Sanders algorithm [177]. Unfortunately, the authors did not specify which target secondary structure had been calculated with which method. Since the amide I band in solid state reaches nearly 1800 cm1, the authors have included a range from 1800 cm1 to 1600 cm1 in their analysis. Using their approach, they report an SEP of 8.32% for helix, an SEP of 8.79% for sheet, an SEP of 6.48% for turns, and an SEP of 8.77% for other. The average of SEPs is 8.09%. Vedantham et al. claim that the impact of solutes (i.e., proteins) on O-H bending and stretching vibrations of solvent (i.e., water) spectra as shown by Raman spectroscopy [176,181,182] as well as possibly varying molar extinction coefficients for different absorbing secondary structures [112,142,183] has not been accounted for sufficiently by current methods for secondary structure prediction from infrared spectra [184]. They also claim that normalisation of the amide region should be performed before background subtractions to be able to correctly account for overlapping regions between peaks that correlate with protein structure and those that do not. To account for these problems, Vedantham et al. employ a method for generating idealised reference spectra in the amide I and amide III regions. This method, which has been originally suggested by Sane et al. for data based on Raman spectroscopy [176], all subtractions, normalisation, and amide band deconvolution steps are performed simultaneously in a single mathematical function for protein spectra. Additionally, varying molar extinction coefficients are allowed for peaks in the infrared spectra correlating with protein secondary structure. The underlying assumption is that all underlying components of protein spectra are additive so that the overall spectral intensity may be described by a single function. Reference spectra were generated using singular value decomposition on the isolated amide I and III bands based on eight proteins in the reference set. Subsequently, their procedure permits the estimation of protein secondary structure of samples outside the reference set. They claim, that based on a reference set of eight protein infrared spectra in H2O, their method provides good secondary structure estimates for proteins comparing well with other established methods for protein secondary structure prediction. Best results were reported based on amide I data. They report an SEP of 2.1% for helix, an SEP of 2.9% for sheet, an SEP of 3.7% for turns, and an SEP of 2.3% for random. The average of SEPs is 2.75%. Note, that results reported for predictions made for proteins within the reference set are not given here. It should also be noted that no details are given about the method for prediction accuracy
147
validation (i.e., it is not clear if the leave-one-out method has been used or if any other method has been employed) and which proteins have been used for prediction accuracy evaluation. Hence, the results reported should be viewed with caution and may not be directly compared with the results obtained by the other methods reported here. Vedantham et al. have calculated secondary structure assignments based on the STRIDE algorithm developed by Frishman and Argos [185]. They claim that the use of the STRIDE algorithm results in significantly improved prediction accuracies compared to results obtained based on calculations from Kabsch and Sanders DSSP method [177] and Levitt and Greers algorithm [170]. In a previous study, Sparks [186] has used the same algorithm for secondary structure prediction approach based on a set of five protein spectra. Here, target secondary structure fractions were calculated based on Levitt and Greers method [170]. The results reported were significantly worse with an SEP of 4.1% for helix, an SEP of 10.1% for sheet, an SEP of 4.4% for turns, and an SEP of 5.6% for random. The average of SEPs is 6.05%. Note, that results reported for predictions made for proteins within the reference set are not given here. Some recent advances include the attempt to select specific wavenumbers for prediction of protein secondary structure [132,146]. This approach is based on the fact that the advent of FT instruments makes it possible to obtain absorbance values for individual wavenumbers at high signal-to-noise ratio. These studies have shown for the first time that few distinct frequencies are sufficient to provide information on protein secondary structure content. In their study, Goormagtigh et al. used an ascending stepwise method that identifies the relevance of every wavenumber in the FTIR spectrum for the prediction of a particular secondary structure [146]. For the analysis they developed a 50-protein database (with minimal fold redundancy). The standard error of prediction in cross-validation was found to be 3.4%, 5.5% and 6.6% for -turn, -helix and -sheet, respectively. 2.2.2. Neural Network Analysis Neither multivariate data analysis methods nor curve fitting methods are free of problems and further improvements in this field are necessary [167]. An alternative pattern recognition approach emerged, which is based on the principle of artificial neural networks. Neural networks have their roots in artificial intelligence. Their basic underlying principles are described in many text books, e.g. [187]. Figure 6 shows an example of a typical neural network architecture for protein secondary structure prediction from FTIR spectral data of proteins with one input layer, one hidden layer, and one output layer. The nodes (often referred to as neurons) of each layer are fully connected with weights. Input values (FTIR spectral data) are simply fed into the input layer nodes. The output value of each following node is calculated as the weighted sum of values of each preceding node. This value is then fed into an activation (sigmoidal function). This way, input values are fed through the neural network to produce outputs (fractions of protein secondary structure content). In the past, neural networks have been most widely used for secondary structure prediction from amino acid sequences, e.g., [188191] as well as from CD spectral data, e.g., [14,17,48,192]. In the present review, our main focus is on protein secondary structure prediction techniques based on FTIR spectra. Hence, we will focus our discussion on work from that domain. Only very few neural network approaches have
148
Figure 6. Example of a neural network architecture for protein secondary structure prediction from FTIR spectral data of proteins.
been suggested particularly based on FTIR spectra of proteins [17,24,193198]. The results in Table 2 show that neural networks bear great potential in making good predictions about the secondary structure of proteins from their FTIR spectra. However, as with all other methods discussed here, the success of neural networks depends very much on the right configuration. It has been shown that the choice of training scheme, neural network topology as well as the choice of data pre-processing methods are important design choices in achieving good prediction accuracy [17,24,193196]: For example, techniques have been successfully employed to reduce the number of weight connections in the neural networks to a number appropriate for the limited number of training patterns available which in fact had a favourable effect on their generalisation capabilities and hence their prediction accuracy [194196]. Obviously, the reduction of neural network weight connections has the additional side-effect of faster neural network training, which might become an issue as the size of the training set increases. Most of the work in the field of applying neural networks to problems in analytical biochemistry has been done using feed-forward multi-layer perceptrons trained with the conventional backpropagation algorithm [199]. However, there are a number of potential problems and pitfalls of the basic backpropagation algorithm, which have lead to the development of improved neural network training algorithms like for instance the locally adaptive learning scheme, resilient backpropagation [200]. Although the basic backpropagation learning rule described in [199] is relatively simple, it is often a difficult task to choose the learning-rate appropriately, since it is strongly dependent on the shape of the error function. The shape of the error function, however, is usually not known and changes with the learning task. A small learning-
149
rate results in long convergence time on a flat error function. A large learning rate, on the other hand, may lead to oscillations, preventing the error to fall below a certain value. Additionally, there is no guarantee that the algorithm will find a global minimum of the error function at all. Another problem with the standard backpropagation algorithm employing gradient-descent is the contra intuitive influence of the partial derivative on the size of the weight step. If the error function is shallow, the derivative is relatively small, resulting in small weight steps. In the presence of steep ravines in the error surface where cautious steps should be taken, large derivatives lead to large weight steps, which could possibly take the algorithm to a completely different region in weight space. Since the introduction of the backpropagation algorithm [199] there have been several suggestions to improve weight training in feed-forward neural networks based on the concept of supervised learning in multi-layer perceptrons using the technique of gradient-descent. Recent studies [132,194198] have successfully employed the resilient backpropagation learning algorithm [200] which makes use of a local adaptation strategy to improve the conventional backpropagation learning technique. We believe that this neural network learning technique is particularly suitable for protein secondary structure prediction from spectral data where mostly only a limited amount of training data is available. Therefore, there is no easy detection of possible overfitting of the neural networks due to the very limited number of training patterns available. Overfitting refers to the case where the neural network has begun to memorise each individual training pattern rather than settling for weights that generally describe the mapping for all cases. Resilient backpropagation has a number of properties explaining the superiority over the conventional backpropagation algorithm in the domain of secondary structure prediction from FTIR spectral data. The harmful influence of the size of the partial derivative on the weight step in the standard backpropagation algorithm is eliminated in resilient backpropagation by considering only the sign of the derivative to indicate the direction of the weight update. This is done by modifying the size of the weight step directly by introducing the concept of resilient update values, resulting in an adaptation effort, which is not prone to unpredictable gradient behaviour. Another advantage of the resilient backpropagation algorithm over the conventional backpropagation algorithm is the speed of convergence. In a study comparing backpropagation to resilient backpropagation and two other adaptive learning methods it was demonstrated on a couple of representative benchmark problems that local adaptive algorithms, in particular resilient backpropagation, converge considerably faster than the ordinary backpropagation (gradient-descent) algorithm [201]. Additionally, robustness of resilient backpropagation with respect to the choice of the initial parameters has been dramatically improved. Consequently, the choice of initial parameters does not have such an important impact on the outcome of the neural network training as with the conventional backpropagation algorithm. This certainly is an additional factor of introducing reliability of the neural network learning algorithm employed. One of the most notable advantages of resilient backpropagation over standard backpropagation is its improved generalisation capability: Riedmiller showed in one of his studies [202] that the introduction of a weight-decay parameter in combination with a relatively low maximum step size did in fact lead to improved generalisation: He demonstrated that overfitting did not occur even with long training times (large number of epochs). Improved generalisation capability is an important feature in protein secondary structure prediction from FTIR spectra of proteins, where there is generally only a limited amount of reference spectra with known X-ray struc-
150
ture available. In that case, usual splitting up of the original pattern set into a training-, validation-, and test set would not be sensible. Consequently, the leave-one-out method is often employed, e.g., [12,16,18,109,110,121], since there is no easy detection of possible overfitting of the neural network. Severcan et al. have used a neural network approach employing the standard resilient backpropagation learning algorithm [194]. They have recognised the importance of good generalisation capabilities of neural networks particularly with respect to the limited set of spectral data available. Hence, they have employed a discrete cosine transform retaining 13 coefficients to reduce the number of neural network inputs to 13 for each spectrum. Instead of having one neural network with three outputs for each secondary structural property, they have employed specialised neural networks with one neural network for each secondary structural feature to be predicted. Their analysis was based on the same set of 18 FTIR spectra of proteins as in Lee et al.s factor analysis approach [109]. The quality of prediction was evaluated using the leave-one-out method. By using a generated set of 10 networks for each prediction and averaging the predicted values, they achieved an SEP of 7.7% for -helix, an SEP of 6.4% for -sheet, and an SEP of 4.8% for turn. The average of SEPs is 6.3%. Based on the same set of 18 FTIR spectra of proteins as in Lee et al.s factor analysis approach [109], we have recently suggested neural network approaches with improved prediction accuracy [132,196,197]. Those studies are based on multi-layer feedforward neural networks using an enhanced version the resilient backpropagation training algorithm, where a weight decay parameter has been added to the error function for improved generalisation [200]. A first approach showed that providing the neural network analysis with only part of the amide I region from empirically determined structure sensitive regions in combination with appropriate pre-processing of the spectral data produced better results than providing all data of the amide I region. This lead to a standard error of prediction (SEP) of 4.47% for -helix, an SEP of 6.16% for -sheet, and an SEP of 4.61% for turns. The average of SEPs is 5.08%. In a further study we employed a genetic algorithm to automatically identify an optimal set of amide I frequencies most suited for our neural network analysis. This resulted in an SEP of 4.58% for helix, an SEP of 5.72% for sheet, an SEP of 4.42% for turn, an SEP of 3.95% for bend, and an SEP of 6.12% for other. The average of SEPs is 4.96%. Pancoska et al. claim that the traditional description of protein secondary structure in terms of overall fractions of helix, sheet and other components is only part of the protein structural information that can be derived from spectroscopic data [17,24,25]. They argued that in addition to overall fractional secondary structure composition, secondary structure segment length and perturbations of regular secondary structures can lead to observable spectral effects. Hence, they introduced a matrix descriptor with integer entries representing the number of secondary structure segments of each type on the diagonal and the number of their interconnections as the off-diagonal elements. Subsequently, neural networks were employed to study the correlation of spectral data with their matrix descriptor. A three-layer backpropagation network with a hyperbolic tangent transfer function trained by a normalised cumulative delta rule was employed. The neural network topology used had one hidden layer where the number of neurons in the hidden layer was determined by a network topology optimisation scheme implementing principal component analysis of the synaptic weights [18]. The neural network had eight output values reflecting their matrix descriptor for helix, sheet, and other segments. The number of inputs to the neural network was determined by the
151
number of spectral points used, i.e., 100 equidistantly digitised intensities of ECD spectra (180260 nm), 103 equidistantly recorded intensities of VCD (2H2O) spectra (15701720 cm1) and 126 points of FTIR (H2O) spectra (14781720 cm1) were used after normalisation. The mapping between the spectral data and the matrix descriptor was evaluated using the leave-one-out method with 23 protein spectra from electronic circular dichroism (ECD) and vibrational circular dichroism (VCD). Pancoska et al. demonstrated that their matrix descriptor could be predicted to an accuracy comparable to that of conventionally predicted average fractional secondary structures. Furthermore, they report that the ECD predictions were more accurate than the VCD ones, which may have resulted from the longer range length dependence of the ECD bandshape and intensity. Results for a parallel analysis using FTIR spectra have indicated a lower reliability than that for VCD. A recent study employed both neural networks trained with resilient backpropagation and adaptive neuro-fuzzy inference systems (ANFIS) to predict helix/sheet segment information based on an extended reference set of 41 FTIR spectra of proteins [203]. Overall, better predictions were achieved using ANFIS. Although information on the number of helix/sheet segments and information on the average helix/sheet segment length are closely related, more accurate predictions were made for the latter. Finally, it was observed that predictions for average helix/sheet length merely based on the amide I band maximum position and the full-width at half-height were comparable to those when individual absorbance values were provided highlighting the importance of that information. 2.2.3. Secondary Structure Prediction of Unknown Proteins (Generalisation) Most pattern recognition based methods are supervised learning techniques, i.e. they base their predictions on a calibration or training set of reference spectra. Based on this set of spectra, a model is built, which allows secondary structure predictions to be made for proteins not seen during the analysis. Probably the most important feature the resulting model must have is the ability to generalise. Generalisation in pattern recognition based approaches is referred to as the capability of producing reasonable outputs for patterns presented, that have not been seen during the analysis. Here, we want the pattern recognition based methods to extract common features from data based on our database of FTIR spectra with known secondary structure from X-ray crystallography studies building a model that correctly maps the spectral data presented to it to its corresponding secondary structural fractions. If the pattern recognition based methods are capable of generalising, they are able to make good predictions for spectral data from new proteins presented to it, which have not been seen during the analysis. 2.2.4. Finding a Representative Training Set Clearly, for the pattern recognition based approaches to be able to come up with a model with good generalisation, the reference set of infrared spectra used for training needs to be representative. The more representative those spectral samples, the more accurate and reliable predictions about the secondary structure of unknown proteins will be. However, finding an optimal composition of representative spectra will be a difficult task since merely increasing the size of the spectral training set may not have the effect of improved prediction accuracy. In a recent study, Pancoska et al. [13] were interested in the effects of increasing the number of proteins with known crystal struc-
152
tures included in the training set. Based on their results, they reported that inclusion of more spectral data in the analysis lead to a deterioration of the quality of prediction. In order to find an optimal set of representative spectra, a joint goal of all researchers in that area should be to build and constantly increase a common database of FTIR spectra for all proteins with known secondary structure. It would certainly be ideal, if access to FTIR spectral data could be added to the existing PDB, a repository for the processing and distribution of 3-D macromolecular structure data primarily determined experimentally [204]. Obviously, standards would need to be defined and additional functionality would need to be implemented which is outside the scope of the present paper. However, work in that direction is a vital step to be able to build systems capable of making good and reliable predictions of new proteins based on their FTIR spectra. Once sufficiently large sets of protein infrared spectra with known secondary structure are available, we are also in a position to separate protein spectra according to structural properties taken from classification databases like e.g. the Structural Classification of Proteins (SCOP) database [205,206]. Individual pattern recognition based systems could then be trained one for each structural class and predictions could be made based on those structure specialised systems. Recently, we introduced a SCOP class specialised neural networks architecture combining an adaptive neuro-fuzzy inference system (ANFIS) with SCOP class specialised backpropagation neural networks [198]. Here, proteins were accurately classified into two main classes all alpha proteins and all beta proteins merely based on the amide I band maximum position of their FTIR spectra. This allowed structure specialised predictions to be made. Our study showed improved predictions using structure specialised neural networks compared to a conventional neural network approach, where one neural network was trained with spectra of both structural classes. 2.2.5. Input Data Reduction for Improved Generalisation Substantial variation exists in the literature suggesting the optimal neural network topology to achieve good generalisation [207209]. However, one factor critical for achieving good generalisation, which all authors seem to agree on is to keep the number of weight connections in the neural network relatively low. This has been achieved by employing various input reduction techniques [132,194196,210,211]. These studies showed that a reduction in input data and hence in the number of neural network weight connections did in fact have a favourable effect on the neural networks generalisation capabilities. Since multivariate data analysis techniques include techniques such as principle component analysis, factor analysis, or singular value decomposition, the input data, i.e., the X data matrix dimensions, generally gets significantly reduced prior to regression. In other words, multivariate data analysis techniques have an efficient data reduction technique already embedded in the overall procedure. Hence, Esbensen et al. claim, that if the X-variables (i.e., the variables describing the absorbances for each wavenumber) are correlated (which is generally the case regarding spectroscopic data), 20 to 60 samples and thousands of spectral wavelengths (i.e., wavenumbers) may not be a problem at all [212]. However, the results in Table 2 suggest, that other data compression techniques like the boxcar averaging method may lead to improved results. Since neural networks generally do not rely on the input data to be linearly independent, they are more flexible in the choice of input data technique to be employed.
153
2.3. Pattern Recognition Approaches Versus Curve Fitting Approaches Although multivariate data analysis approaches and neural network approaches are different in the actual implementation, at higher level, they both belong to the same class, i.e. they are both pattern recognition based methods. Hence, most of the strengths and weaknesses can be generally attributed to both approaches. Here, we will therefore focus on the strengths and weaknesses of pattern recognition based approaches in general and compare them to those of the curve fitting methods. For the pattern recognition based methods to be performed, deconvolution and/or derivation of the spectral band envelope as well as manual assignment of resulting bands to different structures is not required. This removes a significant amount of subjectivity compared to curve fitting. However, the reduction of subjectivity is at the expense of the need for a representative reference set of spectra with known secondary structure. That is, pattern recognition based techniques are dependent on the composition of the training set used to establish good correlation with structure. Therefore, for these approaches to produce good results, the database of reference proteins must be representative, i.e., it must consist of sufficient proteins composed of the types of secondary structure likely to be encountered. E.g., if the reference set is used to build a model for quantifying secondary structure of proteins with only little -helix contents, it would not be appropriate to include only data from proteins with high -helix contents in the reference set. If the composition of the spectra in the reference set is not carefully selected, we may not expect good predictions to be made for a new protein, which has not been used during the analysis. Additionally, Simonetti and Di Bello claim, that since the pattern recognition methods applied thus far are mainly based on a set of reference spectra composed of proteins, prediction of short peptides is likely to be imprecise [129]. They argue that proteins do not represent suitable models for short peptides. The size of data sets used for the pattern recognition based approaches presented here are summarised in Table 2. They range from 8 to 50 which can be said to be a representative data set size for the majority of work done in that area. However, even with a set of 50 spectra, it should not be concluded too easily, that these reference spectra represent all features common to all the thousands of existing proteins. Clearly, the larger and more representative the set of protein spectra used as reference for the analysis, the more reliable the resulting system will be to make good predictions about spectra from proteins with unknown secondary structure. For such increased reference sets to be built, however, we have to allow reference sets to be composed of spectral data recorded in different laboratories under varying conditions. Possible effects on prediction accuracy achieved by a neural network analysis when using reference sets composed of FTIR spectra from different laboratories were investigated as a first step towards developing a common protein infrared spectral database composed of spectral data from different laboratories [213]. However, with the limited size of training set data currently used to date, the potential of protein secondary structure prediction techniques can still be demonstrated. At the end of the curve fitting method, the area of each fitted band is expressed as a percentage of the total area of the amide envelope under investigation. However, this procedure assumes that molar absorptivities of C=O groups responsible for each of the fitted bands are equal commonly, no weighting function is applied to each band. However, studies on poly-L-lysine have shown that the molar absorptivities of differ-
154
ent secondary structures can in fact vary slightly [112,116]. De Jongh et al. suggest, that sheet structure displays greatest molar absorptivity [183]. In constrast, others suggest that molar absorptivities are essentially independent of secondary structure [116,214,215]. Such problems are of less significance for pattern recognition based approaches, since predictions are mainly based on band shape variation. Hence, pattern recognition approaches are much less dependent on intensity measures. As a result, they should be at least partially immune to problems of different molar absorptivities, water vapour and solvent absorbance interference when they are similar for all protein spectra of the reference set. Another potential problem for curve fitting approaches are contributions from amino acid side chains, which may also be less problematic with pattern recognition based approaches. Rahmelow et al. suggested that protein secondary structure prediction based on multivariate data analysis methods may not be disturbed by amino acid side chain absorbances provided that only little variation of absorbing amino acid residues exists amongst the proteins of the reference set [128]. Hence, it is important to calculate the respective standard deviation for each absorbing amino acid across the reference set of proteins. If high variation occurs, amino acid side chain absorbance will have to be taken into account to avoid deterioration in prediction accuracy. We believe that this is generally true for all pattern recognition based approaches including neural network analysis. In their multivariate data analysis study, Rahmelow et al. reported that subtraction of amino acid side chain absorbance from the protein infrared spectra of concanavalin A and erabutoxin lead to an improvement in prediction accuracy. 2.4. Multivariate Data Analysis Approaches Versus Neural Network Methods Multivariate data analysis methods generally involve compressing the input data using techniques like principal component analysis (PCA). Because of the transformation of the original data set into a linearly independent object space with generally significant reduction in dimensions, full spectra instead of only few selected wavelengths may be used for the analysis. Hence, data compression techniques or selection of particular regions within the protein spectra prior to analysis are usually not required as opposed to neural network analysis. However, one critical parameter in these projection methods is the choice of the correct number of dimensions of the new object space, i.e. the decision about what is considered structure and what is considered noise in the original data. If there are too few dimensions, some of the important information of the original data may have been lost. This may become an issue particularly with nonlinearity in the data, since PCA used for data compression is a linear projection method, it imposes a linear structure on the spectral data set, which may not be appropriate if non-linear features are present. As a result, this non-linearity may not be described by the first PCs as in the case of linear data. If there are too many dimensions, however, significant parts of noise may be included in the resulting model to predict protein secondary structure possibly leading to misinterpretations. Additionally, when plotting the reduced orthogonal data set resulting from PCA against the data of the original spectrum, hardly any resemblance is apparent [210]. What is often shown are reconstructed spectra. Compression techniques like boxcar averaging on the other hand compress the spectral data in a way that the overall bandshape is well preserved.
155
With respect to prediction accuracy, PLS may achieve better results than PCR since the Y-variables to be predicted guide the decomposition of X. A comparison of various types of these methods can be found in [9,120]. A good comparison between neural networks and multivariate data analysis techniques has been given by Despagne and Massart [216]. One of the useful features of neural networks in protein secondary structure prediction is that they are universal approximators, i.e., they are capable of fitting any continuous function (including linear and non-linear functions) to a predefined degree of accuracy [217]. As with multivariate data analysis methods, neural networks may also be used to build models of the form Y = F(X) + . However, they are best used on non-linear data sets, e.g., where random noise is present in the data. Neural networks have been shown to be particularly well suited for dealing with noisy data where good results could be obtained despite the presence of noise in the data [218221]. This robustness may be explained by the highly distributed way, information is stored across the neural network weights avoiding single units of the neural network to fail [222,223]. In contrast, multivariate data analysis techniques like PLS or PCR require the inclusion of higher order components to model non-linear data. These higher order components, however, are more likely to be distorted by noise [224]. One of the main advantages of multivariate data analysis methods like MLR, PCR, and PLS is their ease of model interpretation. PLS and PCR, for example, use linear combinations of the original variables (absorbances at wavenumbers) for modelling. Through the projection of the samples into a space with significantly reduced dimensionality, outliers or possible clusters in the data may be easily visualised. Neural networks on the other hand are often criticised as being black box models, where a model is built merely on the basis of input-output data pairs and where interpretation of the resulting model is far more complex than for techniques like PLS and PCR. This is mainly due to the highly distributed way, information is stored across weight connections. However, we have recently introduced a SCOP class specialised neural networks architecture, where an adaptive neuro-fuzzy inference system (ANFIS) is used to classify proteins into structural classes [198]. ANFIS is an adaptive network, which is functionally equivalent to a fuzzy inference system consisting of a set of rules and hence allows inspection and fine-tuning of the rules generated provided that there are not too many inputs. What multivariate data analysis techniques and neural networks have in common is that the underlying training methods are driven by minimising a least squares criterion. Despagne and Massart claim that if only linear transfer functions are used for hidden and output units in a neural network, similarity with multivariate data analysis techniques exists [216]. In this case, the two linear combinations performed between neural network layers is equivalent to a single MLR regression. The only difference lies in the way, the model parameters are optimised. Neural networks make use of an iterative optimisation process (i.e., neural network training) whereas MLR employs matrix inversion. Additionally, if linear transfer functions are used for the neural network, similarity also exists with PCR and PLS [225]. Here, weights between input and hidden layers may be compared to the X-data loadings and the hidden layer units activation may be compared with scores. However, where in neural networks adjustable parameters are fitted to minimise a least squares criterion regardless of any restrictions, additional constraints are taken into account for PCR and PLS. E.g., orthogonality of score values, maximum of variance in the X-data (PCR) and maximum of covariance
156
in the X-Y-data (PLS). Hence, parameters obtained by PCR and PLS can be expected to be different to those obtained by neural networks [216].
3. Applications of Protein Secondary Structure Prediction Table 2 summarises representative applications of protein secondary structure prediction methods along with reported prediction accuracy. Here, prediction accuracy is expressed in terms of the standard error of prediction (SEP). Only those studies are listed, where SEPs could be calculated from the information provided. Best results achieved are only shown when the entire infrared spectral data set available was included in the analysis. Although all of these methods have demonstrated to be very useful techniques to relate infrared spectral data to protein secondary structure (see Table 2), there is still a high degree of freedom in the application of these methods. With respect to curve fitting, we believe that it would be beneficial to standardise its application by developing automated procedures to determine the secondary structure without manual determination of FSD or derivation parameters, no manual peak assignments to secondary structure, no manual choice about the shape of the convolution function, no manual choice of which exact implementation of the curve fitting procedure to use, and no choice whether to apply the curve fitting to the original spectrum, the FSD spectrum or the second derivative spectrum. An attempt in that direction has been made by Rahmelow and Hbner [226]. They have suggested a procedure, which allows the fully automated determination of the parameters required for deconvolution, the subsequent execution of deconvolution using these parameters, and the determination of favourable starting values for a subsequent curve fitting procedure. However, the resulting bands still need to be assigned to the respective types of secondary structure. In their study, they have performed this assignment based on both empirical results taken from Susi and Byler [87] and by applying an algorithmic procedure suggested by Goormaghtigh et al. [168]. Unfortunately, they had to conclude that the results using their automated procedure (SEPs around 14% to 16%) are not satisfying. Inferiority of an automated procedure is probably due to the fact that, as pointed out by Goormaghtigh et al. [227], there is a gap between the explanations most often given for the success of the curve fitting method and its true basis. If a better understanding of the relationship between the curve fitting procedure as a mathematical technique and its underlying physical principles could be established, it would be easier to identify a set of general rules to build a more reliable and accurate automatic procedure. We would then be in a position to give a better and more consistent explanation of how to apply the curve fitting procedure. In contrast, pattern recognition based techniques as an alternative approach to curve fitting may be applied with less subjectivity. However, to date, these techniques suffer from the problem that they are based on only a very limited number of protein spectra in the calibration/training set. Hence, the reliability of these approaches for making good predictions about all existing proteins not seen during the analysis may be questioned. Clearly, further work needs to be done, always with the goal in mind to constantly increase and enhance the set of reference protein spectra with known secondary structure as a basis for further analysis. Additionally, pattern based approaches do also require certain parameters (e.g., training set size, data pre-processing method, number of inputs used, number of principal components) to be determined. Like with
157
curve fitting, the choice of those parameters is mostly rather subjective. However, with respect to neural networks, using the resilient backpropagation algorithm, the impact of choice of initial training parameters could be considerably reduced [132,194198].
4. Conclusion About 60 years ago, the fact that information on protein secondary structure can be derived from FTIR spectra has been established [1,4,5]. Our interpretation of protein infrared spectra has progressed significantly since then but unfortunately sufficiently enough to eliminate doubts about assignment of certain peaks without reliance on additional data. Nevertheless, FTIR spectroscopy is firmly established as a tool for protein structure analysis and in many cases is the only technique that can be used since other techniques are simply incapable of analysis of such samples. One of the major advance since the days of Elliott & Ambrose has been the utilisation of statistical and computational methods for quantifying secondary structure from FTIR spectra of proteins. Not having to rely on subjective interpretation of individual bands for quantitative analysis has been an important advance. The prediction accuracy of the different methods in use is as good, if not better, compared to other techniques such as Circular Dichroism & Raman Spectroscopy. However, much progress still needs to be made in this area. Although both the curve fitting and the pattern recognition based approaches report good prediction accuracy, at present, for none of them it can be claimed that secondary structure proportions of any new protein with unknown structure can be predicted reliably based on its FTIR spectrum. Due to the complexity and multitude of proteins existing as well as due to the inherent experimental variation in recording FTIR spectra, a complete, all-encompassing set of rules describing the relationship between FTIR spectral bandshapes and secondary structural fractions for all proteins will be very hard to establish. There is still great need for further work to develop new methods or improve the existing methods to arrive at the best possible approximation of this mapping. Even if we were unable to sufficiently express such a set of underlying rules in terms of physical underlying phenomena, we can still employ techniques with the potential to automatically discover a mapping reflecting the overall bandshape-structure relationship. A first step in that direction has been made using pattern recognition based approaches.
Acknowledgements We would like to thank Dr. Aichun Dong (University of Northern Colorado) for providing us with Figs 4, 5, and Table 3.
References
[1] A. Elliot, E.J. Ambrose, Structure of synthetic polypeptides, Nature 165 (1950), 921-922. [2] D.M. Byler, H. Susi, Examination of the Secondary Structure of Proteins by Deconvolved FTIR Spectra, Biopolymers 25(3) (1986), 469-487. [3] U. Goerne-Tschelnokow, D. Naumann, C. Weise, and F. Hucho, Secondary structure and temperature behaviour of acetylcholinesterase Studies by Fourier-transform infrared spectroscopy, 213(3) (1993), 1235.
158
[4] E.J. Ambrose, A. Elliot, Infrared spectroscopic studies of globular protein structure, Proc. R. Soc. London Ser. A 208 (1951), 75-90. [5] A. Elliot, Infrared spectra of polypeptides with small side chains, Proc. R. Soc. London Ser. A 226 (1954), 408-421. [6] A. Wlodawer, R. Bott, and L. Sjolin, The refined crystal structure of ribonuclease A at 2.0 resolution, J. Biol. Chem. 257 (1982), 1325-1332. [7] W. Braun, Distance geometry and related methods for protein structure determination from NMR data, Quarterly Reviews of Biophysics 19 (1987), 115-157. [8] B. Dalmas, W.H. Bannister, Prediction of protein secondary structure from circular dichroism spectra: An attempt to solve the problem of the best-fitting reference protein subsets, Analytical Biochemistry 225(1) (1995), 39-48. [9] V. Baumruk, P. Pancoska, and T.A. Keiderling, Predictions of secondary structure using statistical analyses of electronic and vibrational circular dichroism and fourier transform infrared spectra of proteins in H-2O, Journal of Molecular Biology 259(4) (1996), 774-791. [10] P. Pancosca, L. Wang, and T.A. Keiderling, Comparison of protein FT-IR absorption and vibrational circular dichroism frequency analysis in terms of secondary structure, Protein Science 2 (1993), 411419. [11] R. Pribic, Principal Component Analysis of Fourier-Transform Infrared and/or Circular-Dichroism Spectra of Proteins Applied in a Calibration of Protein Secondary Structure, Analytical Biochemistry 223(1) (1994), 26-34. [12] R. Pribic, I.H.M. Van Stokkum, D. Chapman, P.I. Haris, and M. Bloemendal, Protein secondary structure from Fourier transform infrared and/or circular dichroism spectra, Analytical Biochemistry 214(2) (1993), 366-378. [13] P. Pancoska, E. Bitto, V. Janota, M. Urbanova, V.P. Gupta, and T.A. Keiderling, Comparison of and limits of accuracy for statistical analyses of vibrational and electronic circular dichroism spectra in terms of correlations to and predictions of protein secondary structure, Protein Science 4(7) (1995), 1384-1401. [14] M.A. Andrade, P. Chacon, J.J. Merelo, and F. Moran, Evaluation of Secondary Structure of Proteins From UV Circular- Dichroism Spectra Using an Unsupervised Learning Neural-Network, Protein Engineering 6(4) (1993), 383-390. [15] N.J. Greenfield, Methods to Estimate the Conformation of Proteins and Polypeptides from Circular Dichroism Data, Analytical Biochemistry 253(1) (1996), 1-10. [16] B.I. Baello, P. Pancoska, and T.A. Keiderling, Enhanced prediction accuracy of protein secondary structure using hydrogen exchange Fourier transform infrared spectroscopy, Analytical Biochemistry 280(1) (2000), 46-57. [17] P. Pancoska, V. Janota, and T.A. Keiderling, Novel Matrix Descriptor for Secondary Structure Segments in Proteins: Demonstration of Predictability From Circular Dichroism Spectra, Analytical Biochemistry 267(1) (1999), 72-83. [18] P. Pancoska, V. Janota, and T.A. Keiderling, Interconvertibility of Electronic and Vibrational Circular Dichroism Spectra of Proteins: a Test of Principle Using Neural Network Mapping, Applied Spectroscopy 50(5) (1996), 658-668. [19] N. Sreerama, R.W. Woody, Protein Secondary Structure From Circular-Dichroism Spectroscopy Combining Variable Selection Principle and Cluster-Analysis With Neural-Network, RidgeRegression and Self-Consistent Methods, Journal of Molecular Biology 242(4) (1994), 497-507. [20] P. Pancoska, V. Janota, and T.A. Keiderling, Novel approaches to protein structural analyses using combinations of optical spectroscopic methods (electronic and vibrational circular dichroism and FTIR studies), Spectroscopy of biological molecules European conference; 7th, (P. Carmona, R. Navarro, and A. Hernanz, eds.), Kluwer Academic Publishers, 1997, pp. 13-14. [21] G. Bhm, R. Muhr, and R. Jaenicke, Quantitative analysis of protein far UV circular dichroism spectra by neural networks, Protein Engineering 5(3) (1992), 191-5. [22] A. Dong, J.D. Meyer, J.L. Brown, M.C. Manning, and J.F. Carpenter, Comparative Fourier Transform Infrared and Circular Dichroism Spectroscopic Analysis of alpha~1-Proteinase Inhibitor and Ovalbumin in Aqueous Solution, Archives of Biochemistry and Biophysics 383(1) (2000), 148-155. [23] P. Pancoska, J. Kubelka, and T.A. Keiderling, Novel Use of a Static Modification of TwoDimensional Correlation Analysis. Part I: Comparison of the Secondary Structure Sensitivity of Electronic Circular Dichroism, FT-IR, and Raman Spectra of Proteins, 53(6) (1999), 655-665. [24] P. Pancoska, H. Fabian, G. Yoder, V. Baumruk, and T.A. Keiderling, Protein structural segments and their interconnections derived from optical spectra. Thermal unfolding of ribonuclease T-1 as an example, Biochemistry 35(40) (1996), 13094-13106.
159
[25] P. Pancoska, V. Janota, and J. Nesetril, Novel matrix descriptor for determination of the connectivity of secondary structure segments in proteins. Analysis of general properties using graph theory, Discrete Mathematics 235(1-3) (2001), 399-423. [26] J. Kubelka, P. Pancoska, and T.A. Keiderling, Novel Use of a Static Modification of TwoDimensional Correlation Analysis. Part II: Hetero-Spectral Correlations of Protein Raman, FT-IR, and Circular Dichroism Spectra, Applied Spectroscopy 53(6) (1999), 666-671. [27] N. Sreerama, S.Y. Venyaminov, and R.W. Woody, Estimation of the number of alpha-helical and beta-strand segments in proteins using circular dichroism spectroscopy, Protein Science 8(2) (1999), 370-80. [28] E. Vass, M. Kurz, R.K. Konat, and M. Hollosi, Ftir and Cd Spectroscopic Studies on Cyclic Pentaand Hexa- Peptides. Detailed Examination of Hydrogen Bonding in Beta- and Gamma-Turns Determined by Nmr, Spectrochimica Acta Part a-Molecular and Biomolecular Spectroscopy 54(5) (1998), 773-786. [29] J.T. Pelton, L.R. Mclean, Spectroscopic Methods for Analysis of Protein Secondary Structure, Analytical Biochemistry 277(2) (2000), 167-176. [30] R.W. Sarver, W.C. Krueger, An Infrared and Circular-Dichroism Combined Approach to the Analysis of Protein Secondary Structure, Analytical Biochemistry 199(1) (1991), 61-67. [31] M. Bloemendal, W.C. Johnson, Physical Methods to Characterise Pharmaceutical Proteins, (J.N. Herron, W. Jiskoot, and D.J.A. Crommelin, eds.), Plenum, New York, 1995, pp. 65-100. [32] S.Y. Venyaminov, J.T. Yang, Determination of protein secondary structure, Circular dichroism and the conformational analysis of biomolecules, (G.D. Fasman, ed.), Plenum, 1996, pp. 69-108. [33] N. Sreerama, R.W. Woody, Poly(Pro)Ii Helices in Globular-Proteins Identification and Circular Dichroic Analysis, Biochemistry 33(33) (1994), 10022-10025. [34] N. Sreerama, R.W. Woody, A Self-Consistent Method for the Analysis of Protein Secondary Structure From Circular-Dichroism, Analytical Biochemistry 209(1) (1993), 32-44. [35] A. Perczel, M. Hollosi, G. Tusnady, and G.D. Fasman, Convex Constraint Analysis a Natural Deconvolution of Circular-Dichroism Curves of Proteins, Protein Engineering 4(6) (1991), 669-679. [36] P. Pancoska, S.C. Yasui, and T.A. Keiderling, Statistical-Analyses of the Vibrational CircularDichroism of Selected Proteins and Relationship to Secondary Structures, Biochemistry 30(20) (1991), 5089-5103. [37] V.V. Shubin, M.L. Khazin, and T.B. Efimovskaya, Prediction of Secondary Structure of GlobularProteins Using Circular-Dichroism Spectra, Molecular Biology 24(1) (1990), 165-176. [38] P. Manavalan, W.C. Johnson, Variable Selection Method Improves the Prediction of Protein Secondary Structure From Circular-Dichroism Spectra, Analytical Biochemistry 167(1) (1987), 76-85. [39] I.A. Bolotina, V.O. Chekhov, V.Y. Lugauskas, and O.B. Ptitsyn, Determination of the Secondary Structure of Proteins From Circular-Dichroism Spectra. 3. Protein-Derived Reference Spectra for Antiparallel and Parallel Beta-Structures, Molecular Biology 15(1) (1981), 130-137. [40] I.A. Bolotina, V.O. Chekhov, V.Y. Lugauskas, A.V. Finkelshtein, and O.B. Ptitsyn, Determination of the Secondary Structure of Proteins From the Circular-Dichroism Spectra .1. Protein Reference Spectra for Alpha Structure, Beta Structure, and Irregular Structure, Molecular Biology 14(4) (1980), 701-709. [41] J.P. Hennessey, W.C. Johnson, Information-Content in the Circular-Dichroism of Proteins, Biochemistry 20(5) (1981), 1085-1094. [42] S.W. Provencher, J. Glockner, Estimation of Globular Protein Secondary Structure From CircularDichroism, Biochemistry 20(1) (1981), 33-37. [43] S. Brahms, J. Brahms, Determination of protein secondary structure in solution by vacuum ultraviolet circular dichroism, Journal of Molecular Biology 138 (1980), 149-178. [44] Y.H. Chen, J.T. Yang, A new approach to the calculation of secondary structures of globular proteins by optical rotary dispersion and circular dichroism, Biochem Biophys Res Commun 44 (1971), 1285-1291. [45] N.J. Greenfield, G.D. Fasman, Computed circular dichroism spectra for the evaluation of protein conformation, Biochemistry 8(10) (1996), 4108-4116. [46] I.H.M. van Stokkum, H.J.W. Spoelder, M. Bloemendal, R. van Grondelle, and F.C.A. Groen, Estimation of protein secondary structure and error analysis form circular dichroism spectra, Analytical Biochemistry 191 (1990), 110-118. [47] K.A. Oberg, V.N. Uversky, Secondary structure of the homologous proteins, alpha-fetoprotein and serum albumin, from their circular dichroism and infrared spectra, Protein and Peptide Letters 8(4) (2001), 297-302.
160
[48] P. Unneberg, J.J. Merelo, P. Chacon, and F. Moran, SOMCD: Method for Evaluating Protein Secondary Structure From UV Circular Dichroism Spectra, Proteins-Structure Function and Genetics 42(4) (2001), 460-470. [49] B.A. Wallace, J. Lees, and R.W. Janes, Fold Recognition by Synchrotron Radiation Circular Dichroism (SRCD) Spectroscopy: a New Tool for Structural Genomics, Biophysical Journal 82(1) (2002), 1752. [50] B.A. Wallace, R.W. Janes, Synchrotron Radiation Circular Dichroism Spectroscopy of Proteins: Secondary Structure, Fold Recognition and Structural Genomics, Current Opinion in Chemical Biology 5(5) (2001), 567-571. [51] B.A. Wallace, Synchrotron Radiation Circular-Dichroism Spectroscopy as a Tool for Investigating Protein Structures, Journal of Synchrotron Radiation 7 (2000), 289-295. [52] J.G. Lees, B.A. Wallace, Synchrotron radiation circular dichroism and conventional circular dichroism spectroscopy: A comparison, Spectroscopy 16(3/4) (2002), 121-126. [53] J. Reed, T.A. Reed, A Set of Constructed Type Spectra for the Practical Estimation of Peptide Secondary Structure From Circular Dichroism, Analytical Biochemistry 254(1) (1997), 36-40. [54] P. Pancoska, E. Bitto, V. Janota, and T.A. Keiderling, Quantitative-Analysis of Vibrational CircularDichroism Spectra of Proteins Problems and Perspectives, Faraday Discussions (99) (1994), 287-310. [55] G. Deleage, C. Geourjon, An Interactive Graphic Program for Calculating the Secondary StructureContent of Proteins From Circular-Dichroism Spectrum, Computer Applications in the Biosciences 9(2) (1993), 197-199. [56] A. Perczel, K. Park, and G.D. Fasman, Analysis of the Circular-Dichroism Spectrum of Proteins Using the Convex Constraint Algorithm a Practical Guide, Analytical Biochemistry 203(1) (1992), 83-93. [57] A. Perczel, K. Park, and G.D. Fasman, Deconvolution of the Circular-Dichroism Spectra of Proteins the Circular-Dichroism Spectra of the Antiparallel Beta-Sheet in Proteins, Proteins 13(1) (1992), 57-69. [58] E.A. Carrara, C. Gavotti, P. Catasti, F. Nozza, L.L.B. Bergotto, and C.A. Nicolini, Improvement of Protein Secondary Structure Prediction by Combination of Statistical Algorithms and CircularDichroism, Archives of Biochemistry and Biophysics 294(1) (1992), 107-114. [59] W.C. Johnson, Protein Secondary Structure and Circular-Dichroism a Practical Guide, Proteins 7(3) (1990), 205-214. [60] M.C. Manning, Underlying Assumptions in the Estimation of Secondary Structure-Content in Proteins by Circular-Dichroism Spectroscopy a Critical-Review, Journal of Pharmaceutical and Biomedical Analysis 7(10) (1989), 1103-1119. [61] L. Menendezarias, J. Gomezgutierrez, M. Garciaferrandez, A. Garciatejedor, and F. Moran, A Basic Microcomputer Program to Calculate the Secondary Structure of Proteins From Their CircularDichroism Spectrum, Computer Applications in the Biosciences 4(4) (1988), 479-482. [62] L.A. Compton, W.C. Johnson, Analysis of Protein Circular-Dichroism Spectra for Secondary Structure Using a Simple Matrix Multiplication, Analytical Biochemistry 155(1) (1986), 155-167. [63] P. Manavalan, W.C. Johnson, and P.D. Johnston, Prediction Structure Type for Human-Leukocyte Interferon Subtype-a From Circular-Dichroism, Febs Letters 175(2) (1984), 227-230. [64] A. Lobley, L. Whitmore, and B.A. Wallace, Dichroweb: an Interactive Website for the Analysis of Protein Secondary Structure From Circular Dichroism Spectra, Bioinformatics 18(1) (2002), 211-212. [65] N. Sreerama, S.Y. Venyaminov, and R.W. Woody, Analysis of Protein Cd Spectra With a Reference Protein Set Based on Tertiary Structure Class, Biophysical Journal 80(1) (2001), 1342. [66] A. Lobley, B.A. Wallace, Dichroweb: a Website for the Analysis of Protein Secondary Structure From Circular Dichroism Spectra, Biophysical Journal 80(1) (2001), 1570. [67] N. Sreerama, S.Y. Venyaminov, and R.W. Woody, Estimation of Protein Secondary Structure From Circular Dichroism Spectra: Inclusion of Denatured Proteins With Native Proteins in the Analysis, Analytical Biochemistry 287(2) (2000), 243-251. [68] N. Sreerama, R.W. Woody, Estimation of Protein Secondary Structure From Circular Dichroism Spectra: Comparison of Contin, Selcon, and Cdsstr Methods With an Expanded Reference Set, Analytical Biochemistry 287(2) (2000), 252-260. [69] W.C. Johnson, Analyzing Protein Circular Dichroism Spectra for Accurate Secondary Structures, Proteins 35(3) (1999), 307-312. [70] V. Sieber, F. Jurnak, and G.R. Moe, Circular-Dichroism of the Parallel Beta-Helical Proteins Pectate Lyase-C and Lyase-E, Proteins 23(1) (1995), 32-37. [71] J.C. Sutherland, A. Emrick, L.L. France, D.C. Monteleone, and J. Trunk, Circular-Dichroism User Facility at the National Synchrotron Light-Source Estimation of Protein Secondary Structure, Biotechniques 13(4) (1992), 588-590.
161
[72] A. Toumadje, S.W. Alcorn, and W.C. Johnson, Extending Cd Spectra of Proteins to 168 Nm Improves the Analysis for Secondary Structures, Analytical Biochemistry 200(2) (1992), 321-331. [73] S.Y. Venyaminov, I.A. Baikalov, C.S.C. Wu, and J.T. Yang, Some Problems of Cd Analyses of Protein Conformation, Analytical Biochemistry 198(2) (1991), 250-255. [74] P. Pancoska, T.A. Keiderling, Systematic Comparison of Statistical-Analyses of Electronic and Vibrational Circular-Dichroism for Secondary Structure Prediction of Selected Proteins, Biochemistry 30(28) (1991), 6885-6895. [75] W.C. Johnson, Secondary Structure of Proteins Through Circular-Dichroism Spectroscopy, Annual Review of Biophysics and Biophysical Chemistry 17 (1988), 145-166. [76] J.T. Yang, C.S.C. Wu, and H.M. Martinez, Calculation of Protein Conformation From CircularDichroism, Methods in Enzymology 130 (1986), 208-269. [77] M. Schnarr, J.C. Maurizot, Secondary Structure of the Lac Repressor Headpiece Possibilities and Limitations of a Joint Infrared and Circular- Dichroism Study, European Journal of Biochemistry 128(2-3) (1982), 515-520. [78] A. Perczel, Deconvolution of the circular dichroism spectra of proteins: the circular dichroism spectra of the antiparallel beta-sheet in proteins, Proteins 13(1) (1992), 57-69. [79] C.C. Baker, On the analysis of circular dichroic spectra of proteins, Biochemistry 15(3) (1976), 629634. [80] A. Bobba, Estimation of protein secondary structure from circular dichroism spectra: a critical examination of the CONTIN program, Protein Seq Data Anal 3(1) (1990), 7-10. [81] G. Willick, Equivalency of linear least squares curve fitting and reciprocal functions in protein circular dichroic spectra analysis, Biophys Chem 7(3) (1977), 223-227. [82] R.G. Hammonds, Least-squares analysis of circular dichroic spectra of proteins, Eur J Biochem 74(2) (1977), 421-424. [83] M. Bloemendal, Structural information on proteins from circular dichroism spectroscopy possibilities and limitations, Pharm Biotechnol 7 (1995), 65-100. [84] Chen et al., Determination of the helix and b-form of proteins in aqueous solution by circular dichroism, Biochemistry 13(16) (1974), 3350-3359. [85] T.A. Keiderling, Protein and Peptide Secondary Structure and Conformational Determination with Vibrational Circular Dichroism, Current Opinion in Chemical Biology 6(5) (2002), 682-688. [86] A. Barth, C. Zscherp, What Vibrations Tell Us About Proteins, Quarterly Reviews of Biophysics 35(4) (2002), 369-430. [87] H. Susi, D.M. Byler, Resolution-Enhanced Fourier-Transform Infrared-Spectroscopy of Enzymes, Methods in Enzymology 130 (1986), 290-311. [88] W.K. Surewicz, H.H. Mantsch, New insight into protein secondary structure from resolution-enhanced infrared spectra, Biochimica et Biophysica Acta 952 (1988), 115-130. [89] M. Jackson, P.I. Haris, and D. Chapman, Fourier transform infrared spectroscopic studies of lipids polypeptides and proteins, Journal of Molecular Structure 214 (1989), 329-355. [90] J.L.R. Arrondo, A. Muga, J. Castresana, and F.M. Goni, Quantitative Studies of the Structure of Proteins in Solution by Fourier-Transform Infrared-Spectroscopy, Progress in Biophysics & Molecular Biology 59(1) (1993), 23-56. [91] M. Jackson, H.H. Mantsch, Biomembrane structure from FTIR spectroscopy, Spectrochim. Acta Rev. 15 (1993), 53-69. [92] A. Barth, Infrared spectroscopy of proteins, Biochimica et Biophysica Acta 1767 (2007), 1073-1101. [93] J. Bandekar, Amide modes and protein conformation, Biochimica et Biophysica Acta 1120 (1992), 123-143. [94] T. Miyazawa, T. Shimanouchi, and T. Mizushima, Characteristic infrared bands of mono-substituted amides, J. Chem. Phys. 24 (1956), 408-418. [95] T. Miyazawa, T. Shimanouchi, and T. Mizushima, Normal vibrations of N-methylactamide, J. Chem. Phys. 29 (1958), 611-616. [96] T. Miyazawa, The characteristic band of secondary amides at 3100 cm1, J. Mol. Spectrosc. 4 (1960), 168-172. [97] T. Miyazawa, Internal rotation and low frequency spectra of ester, mono-substituted amides and polyglycine, Bull. Chem. Soc. Jpn. 34 (1961), 691-696. [98] T. Miyazawa, Characteristic amide bands and conformations of polypeptides, Polyamino Acids, Polypeptides and Proteins, (M.A. Stahmann, ed.), University of Wisconsin Press, 1962, pp. 201-217. [99] J. Jakes, S. Krimm, Valence force field for the amide group, Spectrochim. Acta Part A 27 (1971), 19-34. [100] J. Jakes, S. Krimm, Normal coordinate analysis of molecules with the amide group, Spectrochim. Acta Part A 27 (1971), 35-63.
162
[101] Y. Abe, S. Krimm, Normal vibrations of crystalline polyglycine I, Biopolymers 11 (1972), 1817-1840. [102] Y. Abe, S. Krimm, Normal vibrations of polyglycine II, Biopolymers 11 (1972), 1841-1853. [103] S. Krimm, J. Bandekar, Vibrational spectroscopy and conformation of peptides, polypeptides and proteins, Adv. Protein Chem. 38 (1986), 181-364. [104] T. Miyazawa, E.R. Blout, The infrared spectra of various polypeptides in various conformations: Amide II bands, J. Am. Chem. Soc. 83 (1961), 712-719. [105] H. Susi, S.N. Timasheff, and L. Stevens, Infrared spectra and protein conformations in aqueous solutions: I The amide I band in H2O and D2O solution, J. Biol. Chem. 242 (1967), 5460-5466. [106] P.I. Haris, D. Chapman, Does Fourier-Transform Infrared-Spectroscopy Provide Useful Information on Protein Structures, Trends in Biochemical Sciences 17(9) (1992), 328-333. [107] M.S. Braiman, K.J. Rothschild, Fourier-Transform Infrared Techniques for Probing Membrane- Protein Structure, Annual Review of Biophysics and Biophysical Chemistry 17 (1988), 541-570. [108] A. Dong, P. Huang, and W.S. Caughey, Protein secondary structure from second derivative amide I infrared spectra, Biochemistry 29 (1990), 3303-3308. [109] D.C. Lee, P.I. Haris, D. Chapman, and R.C. Mitchell, Determination of Protein Secondary Structure Using Factor- Analysis of Infrared-Spectra, Biochemistry 29(39) (1990), 9185-9193. [110] R.W. Sarver, W.C. Krueger, Protein secondary structure from Fourier transform infrared spectroscopy: A database analysis, Analytical Biochemistry 194 (1991), 89-100. [111] L.K. Tamm, S.A. Tatulian, Infrared Spectroscopy of Proteins and Peptides in Lipid Bilayers, Quarterly Reviews of Biophysics 30(4) (1997), 365-429. [112] M. Jackson, P.I. Haris, and D. Chapman, Conformational Transitions in Poly(L-Lysine) Studies Using Fourier-Transform Infrared-Spectroscopy, Biochimica Et Biophysica Acta 998(1) (1989), 75-79. [113] P.I. Haris, Fourier Transform Infrared Spectroscopic Studies of Peptides: Potentials and Pitfalls, ACS Symposium series, (B.R. Singh, ed.), American Chemical Society, 2000, pp. 54-95. [114] R. Gilmanshin, S. Williams, R.H. Callender, W.H. Woodruff, and R.B. Dyer, Fast events in protein folding: Relaxation dynamics of secondary and tertiary structure in native apomyoglobin, Proceedings National Academy of Sciences USA, (Anonymous), USA: National Academy Of Sciences, 1997, pp. 3709-3713. [115] Y.N. Chirgadze, O.V. Fedorov, and N.P. Trushina, Estimation of amino acid residue side chain absorptions in infrared spectra of protein solutions in heavy water, Biopolymers 14 (1975), 679-694. [116] S.Y. Venyaminov, N.N. Kalnin, Quantitative IR spectrophotometry of peptide compounds in water (H2O) solutions. I. Spectral parameters of amino acid residue absorption bands, Biopolymers 30 (1990), 1243-1257. [117] F. Dousseau, M. Pezolet, Determination of the secondary structure content of proteins in aqueous solutions from their amide I and amide II infrared bands. Comparison between classical and partial leastsquares methods, Biochemistry 29 (1990), 8771-8779. [118] H. Torii, T. Tatsumi, T. Kanazawa, and M. Tasumi, Effects of Intermolecular Hydrogen-Bonding Interactions on the Amide I Mode of N-Methylacetamide: Matrix-Isolation Infrared Studies and ab Initio Molecular Orbital Calculations, 102(1) (1998), 309-314. [119] T.F. Kumosinski, J.J. Unruh, Global-secondary-structure analysis of proteins in solution Resolutionenhanced deconvolution Fourier transform infrared spectroscopy in water, Molecular Modeling From virtual tools to real problems, in ACS Symposium Series, (T.F. Kumosinski, M.N. Liebman, eds.), American Chemical Society, 1994, pp. 71-98. [120] K. Rahmelow, W. Huebner, Secondary structure determination of proteins in aqueous solution by infrared spectroscopy: A comparison of multivariate data analysis methods, Analytical Biochemistry 241(1) (1996), 5-13. [121] S. Wi, P. Pancoska, and T.A. Keiderling, Predictions of protein secondary structures using factor analysis on Fourier transform infrared spectra: Effect of Fourier self-deconvolution of the amide I and amide II bands, Biospectroscopy 4(2) (1997), 93-106. [122] K. Kaiden, T. Matsui, and S. Tanaka, A study of amide III band by FT-IR spectrometry of the secondary structure of albumin, myoglobin and -globulin, Applied Spectroscopy 41 (1981), 180-184. [123] G. Anderle, R. Mendelsohn, Thermal denaturation of globular proteins: Fourier transform infrared studies of the amide III spectral region, Biophysical Journal 52 (1987), 69-74. [124] B.R. Singh, D.B. DeOliveira, F. Fu, and M.P. Fuller, Fourier transform infrared analysis of amide III bands of proteins for the secondary-structure estimation [1890-11], Biomolecular spectroscopy III, in Proceedings Spie the international society for optical engineering, (L.A. Nafie, H.H. Mantsch, eds.), SPIE, 1993, pp. 47-55. [125] F.-N. Fu, D.B. DeOliveira, W.R. Trumble, and H.K. Sarkar, Secondary Structure Estimation of Proteins Using the Amide III Region of Fourier Transform Infrared Spectroscopy: Application to Analyze Calcium-Binding-Induced Structural Changes in Calsequestrin, 48(11) (1994), 1432.
163
[126] B.R. Singh, M.P. Fuller, and B.R. DasGupta, Botulinum neurotoxin type A: structure and interaction with the micellar concentration of SDS determined by FT-IR spectroscopy, J. Protein Chem. 10 (1991), 637-649. [127] K. Griebenow, A.M. Klibanov, Lyophilization-induced reversible changes in the secondary structure of proteins, Proceedings National Academy of Sciences USA, (Anonymous), USA: National Academy of Sciences, 1995, pp. 10969-10976. [128] K. Rahmelow, W. Huebner, and T. Ackermann, Infrared Absorbances of Protein Side Chains, Analytical Biochemistry 257(1) (1998), 1-11. [129] Simonetti M., C. Di Bello, New Fourier transform infrared based computational method for peptide secondary structure determination. I. Description of method, Biopolymers 62(2) (2001), 95-108. [130] M. Simonetti, C. Di Bello, New Fourier transform infrared based computational method for peptide secondary structure determination. II. Application to study of peptide fragments reproducing processing site of ocytocin-neurophysin precursor, Biopolymers 62(2) (2001), 109-121. [131] I. Noda, Generalized Two-Dimensional Correlation Method Applicable to Infrared, Raman, and Other Types of Spectroscopy, 47(9) (1993), 1329. [132] J.A. Hering, P.R. Innocent, and P.I. Haris, Automatic Amide I frequency selection for rapid quantification of protein secondary structure from FTIR spectra of proteins, Proteomics 2(7) (2002), 839-849. [133] C.B. Lucasius, M.L.M. Beckers, and G. Kateman, Genetic Algorithms in Wavelength Selection a Comparative- Study, Analytica Chimica Acta 286(2) (1994), 135-153. [134] R. Leardi, Application of a Genetic Algorithm to Feature-Selection Under Full Validation Conditions and to Outlier Detection, Journal of Chemometrics 8(1) (1994), 65-79. [135] D. Jouan-Rimbaud, D.L. Massart, R. Leardi, and O.E. Denoord, Genetic Algorithms as a Tool for Wavelength Selection in Multivariate Calibration, Analytical Chemistry 67(23) (1995), 4295-4301. [136] J.M. Roger, V. Bellon-Maurel, Using Genetic Algorithms to Select Wavelengths in Near-Infrared Spectra: Application to Sugar Content Prediction in Cherries, Applied Spectroscopy 54(9) (2000), 1313-1320. [137] A.S. Bangalore, R.E. Shaffer, G.W. Small, and M.A. Arnold, Genetic Algorithm-Based Method for Selecting Wavelengths and Model Size for Use With Partial Least-Squares Regression: Application to Near-Infrared Spectroscopy, Analytical Chemistry 68(23) (1996), 4200-4212. [138] M.J. Arcos, M.C. Ortiz, B. Villahoz, and L.A. Sarabia, Genetic-Algorithm-Based Wavelength Selection in Multicomponent Spectrometric Determinations by Pls: Application on Indomethacin and Acemethacin Mixture, Analytica Chimica Acta 339(1-2) (1997), 63-77. [139] B.M. Smith, L. Oswald, and S. Franzen, Single-Pass Attenuated Total Reflection Fourier Transform Infrared Spectroscopy for the Prediction of Protein Secondary Structure, Analytical Chemistry 74(14) (2002), 3386-3391. [140] P.I. Haris, Characterization of protein structure and stability using Fourier transform infrared spectroscopy, Pharmacy and Pharmacology Communications 5(1) (1999), 15-25. [141] P.I. Haris, D. Chapman, Analysis of polypeptide and protein structures using Fourier transform infrared spectroscopy, Microscopy, optical spectroscopy, and macroscopic techniques, in Methods in Molecular Biology, (C. Jones, B. Mulloy, and A.H. Thomas, eds.), Humana Press Inc., 1994, pp. 183-202. [142] M. Jackson, H.H. Mantsch, The Use and Misuse of FTIR Spectroscopy in the Determination of Protein-Structure, Critical Reviews in Biochemistry and Molecular Biology 30(2) (1995), 95-120. [143] B.R. Singh, Basic Aspects of the Technique and Applications of Infrared Spectroscopy of Peptides and Proteins, ACS Symposium series, (B.R. Singh, ed.), American Chemical Society, 2000, pp. 2-37. [144] S. Krimm, Interpreting Infrared Spectra of Peptides and Proteins, in ACS Symposium series, (B.R. Singh, ed.), USA: Washington, DC; American Chemical Society, 2000, pp. 38-53. [145] C. Jung, Insight into protein structure and protein-ligand recognition by Fourier transform infrared spectroscopy, Journal of Molecular Recognition 13(6) (2000), 325-351. [146] E. Goormaghtigh, J.-M. Ruysschaert, and V. Raussens, Evaluation of the Information Content in Infrared Spectra for Protein Secondary Structure Determination, Biophysical Journal 90 (2006), 2946-2957. [147] M. Ruegg, V. Metzger, and H. Susi, Computer analysis of characteristic infrared bands of globular proteins, Biopolymers 14 (1975), 1465-1471. [148] J. Villalain, J.C. Gomez-Fernandez, M. Jackson, and D. Chapman, Fourier transform infrared spectroscopic studies on the secondary structure of the Ca2+-ATPase of sarcoplasmic reticulum, Biochimica et Biophysica Acta 978 (1989), 305-312. [149] A. Dong, B. Kendrick, L. Kreilgard, J. Matsuura, M.C. Manning, and J.F. Carpender, Spectroscopic Study of Secondary Structure and Thermal Denaturation of Recombinant Human Factor XIII in Aqueous Solution, Archives of Biochemistry and Biophysics 347(2) (1997), 213-220.
164
[150] A. Dong, J. Matsuura, M.C. Manning, and J.F. Carpenter, Intermolecular Beta-Sheet Results from Trifluoroethanol-Induced Nonnative Alpha-Helical Structure in Beta-Sheet Predominant Proteins: Infrared and Circular Dichroism Spectroscopic Study, Archives of Biochemistry and Biophysics 355(2) (1998), 275-281. [151] J.L.R. Arrondo, F.M. Goni, Structure and dynamics of membrane proteins as studied by infrared spectroscopy, Progress in Biophysics & Molecular Biology 72 (1999), 367-405. [152] K.G. Carrasquillo, C. Sanchez, and K. Griebenow, Relationship between conformational stability and lyophilization-induced structural changes in chymotrypsin, Biotechnol. App. Biochem. 31 (2000), 41-53. [153] A. Troullier, D. Reinstdler, Y. Dupont, D. Naumann, and V. Forge, Transient non-native secondary structures during the refolding of alpha-lactalbumin detected by infrared spectroscopy, Nature Structural Biology 7(1) (2000), 78-86. [154] M. Balsera, J.B. Arellano, J.R. Gutierrez, P. Heredia, J.L. Revuelta, and J. De Las Rivas, Structural Analysis of the PSBQ Protein of Photosystem Ii by Fourier Transform Infrared and Circular Dichroic Spectroscopy and by Bioinformatic Methods, Biochemistry 42(4) (2003), 1000-1007. [155] W. Hbner, H.H. Mantsch, and H.L. Casal, Beware of frequency shifts, Applied Spectroscopy 44 (1990), 732-734. [156] J.K. Kauppinen, D.J. Moffatt, H.H. Mantsch, and D.G. Cameron, Fourier transforms in the computation of self-deconvoluted and first-order derivative spectra of overlapped band contours, Analytical Chemistry 53 (1981), 1454-1457. [157] J.K. Kauppinen, D.J. Moffatt, H.H. Mantsch, and D.G. Cameron, Fourier self-deconvolution: a method for resolving intrinsically overlapped bands, Applied Spectroscopy 35 (1981), 271-276. [158] D.G. Cameron, D.J. Moffatt, Deconvolution, derivation and smoothing of spectra using Fourier transforms, Journal of testing and evaluation 12 (1984), 78-85. [159] D.G. Cameron, D.J. Moffatt, A generalized approach to derivative spectroscopy, Applied Spectroscopy 41 (1987), 539-544. [160] H.H. Mantsch, D.J. Moffatt, Computer-aided methods for the resolution enhancement of spectral data with special emphasis on infrared spectra, NATO ASI Series, Mathematical and Physical Sciences, (R. Fausto, ed.), USA: Kluwer Academic Publishers, 1993, pp. 113-124. [161] P.R. Griffiths, G.L. Pariente, Introduction to spectral deconvolution, Trends Anal. Chem. 5 (1986), 209-215. [162] H. Stone, Mathematical Resolution of Overlapping Spectral Lines, Journal of the Optical Society of America 52(9), 998-1003. [163] J.K. Kauppinen, D.J. Moffatt, H.H. Mantsch, and D.G. Cameron, Smoothing of Spectral Data in the Fourier Domain, Applied Optics 21(10) (1982), 1866-1872. [164] A. Savitsky, M.J.E. Golay, Smoothing and differentiation of data by simplified least squares procedures, Analytical Chemistry 36 (1964), 1627-1639. [165] B.W. Caughey, A. Dong, K.S. Bhat, D. Ernst, S.F. Hayes, and W.S. Caughey, Secondary structure analysis of the scrapie-associated protein PrP 27-30 in water by infrared spectroscopy, Biochemistry 30 (1991), 7672-7678. [166] A. Dong, W.S. Caughey, and T.W. Du Closs, Effects of calcium, magnesium and phosphorylcholine on secondary structures of human C-reactive protein and serum amyloid P component observed by infrared spectroscopy, Journal of Biological Chemistry 269 (1994), 6424-6430. [167] W.K. Surewicz, H.H. Mantsch, and D. Chapman, Determination of Protein Secondary Structure by Fourier Transform Infrared Spectroscopy: A Critical Assessment, 32(2) (1993), 389-394. [168] E. Goormaghtigh, V. Cabiaux, and J.M. Ruysschaert, Secondary structure and dosage of soluble and membrane proteins by attenuated total reflection Fourier-transform infrared spectroscopy on hydrated films, Eur. J. Biochem. 193 (1990), 409-420. [169] T.F. Kumosinski, J.J. Unruh, Quantitation of the global secondary structure of globular proteins by FTIR spectroscopy: comparison with X-ray crystallographic structure, 43(2) (1996), 199-219. [170] M. Levitt, J. Greer, Automatic Identification of Secondary Structure in Globular Proteins, Journal of Molecular Biology 114 (1977), 181-293. [171] J.L. Lippert, D. Tyminski, and P.J. Desmeules, J. Am. Chem. Soc. 101 (1976), 5111-5121. [172] B.M. Bussian, C. Sander, How to Determine Protein Secondary Structure in Solution by RamanSpectroscopy Practical Guide and Test Case Dnase-I, Biochemistry 28(10) (1989), 4271-4277. [173] R.W. Williams, A.K. Dunker, Determination of the Secondary Structure of Proteins From the Amide-I Band of the Laser Raman-Spectrum, Journal of Molecular Biology 152(4) (1981), 783-813. [174] R.W. Williams, Estimation of Protein Secondary Structure From the Laser Raman Amide-I Spectrum, Journal of Molecular Biology 166(4) (1983), 581-603.
165
[175] M. Berjot, J. Marx, and A.J.P. Alix, Determination of the Secondary Structure of Proteins From the Raman Amide-I Band the Reference Intensity Profiles Method, Journal of Raman Spectroscopy 18(4) (1987), 289-300. [176] S.U. Sane, S.M. Cramer, and T.M. Przybycien, A Holistic Approach to Protein Secondary Structure Characterization Using Amide I Band Raman Spectroscopy, Analytical Biochemistry 269(2) (1999), 255-272. [177] W. Kabsch, C. Sander, Dictionary of Protein Secondary Structure Pattern-Recognition of HydrogenBonded and Geometrical Features, Biopolymers 22(12) (1983), 2577-2637. [178] Y.Q. Wu, K. Murayama, and Y. Ozaki, Two-dimensional infrared spectroscopy and principle component analysis studies of the secondary structure and kinetics of hydrogen-deuterium exchange of human serum albumin, Journal of Physical Chemistry B 105(26) (2001), 6251-6259. [179] A. Filosa, Y. Wang, A. Ismail, and A.M. English, Two-dimensional infrared correlation spectroscopy as a probe of sequential events in the thermal unfolding of cytochromes c, Biochemistry 40(28) (2001), 8256-8263. [180] L.A. Forato, R. Bernardes-Filho, and L.A. Colnago, Protein Structure in KBr Pellets by Infrared Spectroscopy, Analytical Biochemistry 259(1) (1998), 136-141. [181] N. Colloch, C. Etchebest, E. Thoreau, and B. Henrissat, Comparison of three algorithms for the assignment of secondary structure in proteins: the advantages of a consensus assignment, Protein Engineering 6(4) (1993), 377-382. [182] S. Leikin, V.A. Parsegian, W.-H. Yang, and G.E. Walrafen, Raman spectral evidence for hydration forces between collagen triple helices, in Proceedings National Academy of Sciences USA, (B. Mazur, K. Rubin, eds.), USA: National Academy of Sciences, 1997, pp. 11312-11317. [183] H.H.J. De Jongh, E. Goormaghtigh, and J.-M. Ruysschaert, The Different Molar Absorptivities of the Secondary Structure Types in the Amide I Region: An Attenuated Total Reflection Infrared Study on Globular Proteins, Analytical Biochemistry 242(1) (1996), 95-103. [184] G. Vedantham, H.G. Sparks, S.U. Sane, S. Tzannis, and T.M. Przybycien, A Holistic Approach for Protein Secondary Structure Estimation from Infrared Spectra in H~2O Solutions, Analytical Biochemistry 285(1) (2000), 33-49. [185] D. Frishman, P. Argos, Seventy-five percent accuracy in protein secondary structure prediction, Proteins Structure Function and Genetics 27(3) (1997), 329-335. [186] H.G. Sparks, PhD Thesis, Rensselaer Polytechnic Institute, (1999). [187] Kevin Swingler, Applying Neural Networks A practical guide, Academic Press, 96. [188] J. Moult, The Current State of the Art in Protein Structure Prediction, Current Opinion in Biotechnology 7(4) (1996), 422-427. [189] B. Rost, S. ODonoghue, Sisyphus and prediction of protein structure, CABIOS 13(4) (1997), 345356. [190] P. Baldi, S. Brunak, P. Frasconi, G. Soda, and G. Pollastri, Exploiting the past and the future in protein secondary structure prediction, Bioinformatics (Oxford) 15(11) (1999), 937-946. [191] P. Baldi, S. Brunak, Y. Chauvin, C.A.F. Andersen, and H. Nielsen, Assessing the accuracy of prediction algorithms for classification: An overview, Bioinformatics (Oxford) 16(5) (2000), 412-424. [192] K.A. Oberg, J.-M. Ruysschaert, and G. Goormaghtigh, The optimization of protein secondary structure determination with infrared and circular dichroism spectra, Eur. J. Biochemistry 271 (2004), 2937-2948. [193] V. Cabiaux, K.A. Oberg, P. Pancoska, T. Walz, P. Agre, and A. Engel, Secondary structures comparison of aquaporin-1 and bacteriorhodopsin: A Fourier transform infrared spectroscopy study of twodimensional membrane crystals, Biophysical Journal 73(1) (1997), 406-417. [194] M. Severcan, F. Severcan, and P.I. Haris, Estimation of Protein Secondary Structure From FTIR Spectra Using Neural Networks, Journal of Molecular Structure 565 (2001), 383-387. [195] J.A. Hering, P.R. Innocent, and P.I. Haris, An improved method for rapid quantification of protein secondary structure from FTIR spectra of proteins, 2001 Congress Functional Proteomics, in Proceedings of the Swiss Proteomics Society, (P.M. Palagi, J.-C. Sanchez, and R. Stcklin, eds.), Fontis Media, 2001, pp. 128-132. [196] J.A. Hering, P.R. Innocent, and P.I. Haris, An alternative method for rapid quantification of protein secondary structure from FTIR spectra using neural networks, Spectroscopy 16(2) (2002), 53-69. [197] J.A. Hering, P.R. Innocent, and P.I. Haris, New approaches for quantification of protein secondary structure from FTIR spectra of proteins, 2002 Congress Applied Proteomics, in Proceedings of the Swiss Proteomics Society, (P.M. Palagi, M. Quadroni, J.S. Rossier, J.-C. Sanchez, and R. Stcklin, eds.), Fontis Media, 2002, pp. 163-165. [198] J.A. Hering, P.R. Innocent, and P.I. Haris, Neuro-Fuzzy SCOP classification for improved protein secondary structure prediction, Proteomics 3(8) (2003), 1464-1475.
166
[199] J.L. McClelland, and D.E. Rumelhart, Parallel Distributed Processing: Explorations in the Microstructure of Cognition, in Volume 2: Psychological and Biological Models, MIT Press, 87. [200] M. Riedmiller, H. Braun, A direct adaptive method for faster backpropagation learning: The RPROP algorithm, IEEE International Conference on Neural Networks (ICNN-93), (H. Ruspini, ed.), IEEE, San Francisco, USA, 1993, pp. 586-591. [201] M. Riedmiller, Advanced Supervised Learning in Multi-layer Perceptrons From Backpropagation to Adaptive Learning Algorithms, International Journal of Computer Standards and Interfaces, Special Issue on Neural Networks 5 (1994), 8. [202] M. Riedmiller, Untersuchungen zur Konvergenz und Generalisierungsfhigkeit berwachter Lernverfahren mit dem SNNS, Workshop SNNS-93: Simulation Neuronaler Netze mit SNNS, (Anonymous), Universitt Stuttgart, Fakultt Informatik, 1993, pp. 107-116. [203] J.A. Hering, P.R. Innocent, and P.I. Haris, Beyond average protein secondary structure content prediction using FTIR spectroscopy, Accepted for publication in Applied Bioinformatics . [204] H.M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T.N. Bhat, H. Weissig, I.N. Shindyalov, and P.E. Bourne, The Protein Data Bank, Nucleic Acids Research 28(1) (2000), 235-242. [205] A.G. Murzin, S.E. Brenner, T. Hubbard, and C. Chothia, SCOP: A Structural Classification of Proteins Database for the Investigation of Sequences and Structures, Journal of Molecular Biology 247(4) (1995), 536-540. [206] L. Lo Conte, B. Ailey, T.J. Hubbard, S.E. Brenner, A.G. Murzin, and C. Chothia, SCOP: A Structural Classification of Proteins database, Nucleic Acids Research 28(1) (2000), 257-259. [207] E.B. Baum, D. Haussler, What size net gives valid generalisation?, Neural computing 1 (1989), 151-160. [208] S. Haykin, Neural Networks, A comprehensive Foundation, Macmillan College Publishing Company, Inc., 94, 176-177. [209] R. Lange, R. Mnner, Quantifying a critical training set size for generalisation and overfitting using teacher neural networks, Proceedings of the International Conference on Artificial Neural Networks (ICANN), (P. Morasso, M. Marinaro, eds.), 1994, pp. 497-500. [210] C. Klawun, and C.L. Wilkins, Optimization of functional group prediction from infrared spectra using neural networks, Journal of Chemical Information and Computer Sciences 36(1) (1996), 69-81. [211] V. Tchistiakov, C. Ruckebusch, L. Duponchel, J.P. Huvenne, and P. Legrand, Neural network modelling for very small spectral data sets: reduction of the spectra and hierarchical approach, Chemometrics and intelligent laboratory systems 54(2) (2000), 93-106. [212] K. Esbensen, S. Schnkopf, and T. Midtgaard, Multivariate Analysis in Practice, Computer-Aided Modelling AS, 94. [213] J.A. Hering, P.R. Innocent, and P.I. Haris, Towards developing a protein infrared spectra databank (PISD) for Proteomics research, Accepted for publication in Proteomics . [214] Y.N. Chirgadze, B.V. Shestopalov, and S.Y. Venyaminov, Biopolymers 12 (1973), 1337-1351. [215] Y.N. Chirgadze, E.V. Brazhnikov, Biopolymers 13 (1974), 1701-1712. [216] F. Despagne, D.L. Massart, Neural Networks in Multivariate Calibration, Analyst 123(11) (1998), 157-178. [217] K. Hornik, M. Stinchcombe, and H. White, Multilayer Feedforward Networks Are Universal Approximators, Neural Networks 2(5) (1989), 359-366. [218] S.R. Amendolia, A. Doppiu, M.L. Ganadu, and G. Lubinu, Classification and Quantitation of H-1 Nmr Spectra of Alditols Binary Mixtures Using Artificial Neural Networks, Analytical Chemistry 70(7) (1998), 1249-1254. [219] R. Goodacre, Use of Pyrolysis Mass Spectrometry With Supervised Learning for the Assessment of the Adulteration of Milk of Different Species, Applied Spectroscopy 51(8) (1997), 1144-1153. [220] R. Goodacre, M.J. Neal, and D.B. Kell, Rapid and Quantitative-Analysis of the Pyrolysis MassSpectra of Complex Binary and Tertiary Mixtures Using Multivariate Calibration and Artificial Neural Networks, Analytical Chemistry 66(7) ( 1994), 1070-1085. [221] J.R. Long, V.G. Gregoriou, and P.J. Gemperline, Spectroscopic Calibration and Quantitation Using Artificial Neural Networks, Analytical Chemistry 62(17) (1990), 1791-1797. [222] S. Biswas, S. Venkatesh, The devil and the network: what sparsity implies to robustness and memory, Advances in Neural Information Processing Systems, (R.P. Lippman, J.E. Moody, and D.S. Touretzky, eds.), 1991, pp. 883-889. [223] B. Hitzmann, A. Ritzka, R. Ulber, T. Scheper, and K. Schugerl, Computational Neural Networks for the Evaluation of Biosensor Fia Measurements , Analytica Chimica Acta 348(1-3) (1997), 135-141. [224] C. Borggaard, H.H. Thodberg, Optimal Minimal Neural Interpretation of Spectra, Analytical Chemistry 64(5) (1992), 545-551.
167
[225] J. Naes, K. Kvaal, T. Isaksson, and C. Miller, Artificial neural networks in multivariate calibration, Journal of Near Infrared Spectroscopy 1 (1993), 1-12 . [226] K. Rahmelow, W. Huebner, Fourier Self-Deconvolution: Parameter Determination and Analytical Band Shapes, 50(6) (1996), 795-804. [227] E. Goormaghtigh, V. Cabiaux, and J.-M. Ruysschaert, Determination of Soluble and Membrane Protein Structure by Fourier Transform Infrared Spectroscopy: III. Secondary Structures, in Subcellular Biochemistry, (H.J. Hilderson, G.B. Ralston, eds.), USA: Plenum Publishing Corporation, 1994, pp. 405-450. [228] D. Frishman, P. Argos, Knowledge-Based Protein Secondary Structure Assignment, Proteins 23(4) (1995), 566-579. [229] E. Goormaghtigh, J.M. Ruysschaert, Molecular Description of Biological Components by Computer Aided Conformational Analysis, (R. Brasseur, ed.), CRC Press, 1998, pp. 285-329. [230] J.L.R. Arrondo, F.M. Goni, Protein-Lipid Interactions, (A. Watts, ed.), Elsevier, 1993, pp. 321-349. [231] H.H. Mantsch, A. Perczel, M. Hollosi, and G.D. Fasman, Characterization of a-Turns in Cyclic Hexapeptides in Solution by Fourier Transform IR Spectroscopy, Biopolymers 33(2) (1993), 201-207. [232] A. Dong, W.S. Caughey, Infrared Methods for Study of Hemoglobin Reactions and Structures, in Methods in Enzymology, (J. Everse, K.D. Vandegriff, and R.M. Winslow, eds.), USA: Academic Press Inc Ltd, 1994, pp. 139-175.
168
Infrared Spectroscopy of Protein Pharmaceuticals

Marco VAN DE WEERT 11 and Lene JORGENSEN Biomacromolecular Drug Delivery, Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark
1. Introduction In the last few decades, the pharmaceutical industry has seen a rapid rise in the number of drug products containing a protein as the active compound. At present, about one third of new products contains a protein as the active compound [1]. These pharmaceutical proteins are mainly used for life threatening and/or chronic diseases, such as hepatitis, diabetes, cancer, growth disorders, and anemia. Protein pharmaceuticals introduce several challenges for the pharmaceutical scientist [2]. First of all, they are expensive to manufacture at the required high purity. Second, they are very poorly absorbed through biological membranes, making injection or infusion the most practical method of administration. This is not only inconvenient for the patient, but also puts severe restrictions on the additives in the formulation. Finally, proteins are highly complex molecules, and prone to a variety of physicochemical degradation processes. The regulatory agencies do not accept the presence of a significant amount of degradation products, and require that the protein itself, as well as most important degradation products, are thoroughly characterised. The characterisation of a protein requires the use of several techniques that give complementary information. The analytical toolbox for protein characterisation is still expanding, and consists of a wide variety of techniques. The interested reader is referred to the book Methods for structural analysis of protein pharmaceuticals for a description of several methods used for protein pharmaceuticals [3]. One of the methods in this set of techniques is infrared spectroscopy (FTIR). Until relatively recently, FTIR was only sparsely used to characterise protein pharmaceuticals. Most of the early products were formulated as aqueous solutions with low protein concentrations (<1 mg/ml), which precludes the use of FTIR. However, recent years have seen an increasing focus on developing high concentration protein pharmaceuticals, especially monoclonal antibodies, where concentrations well above 100 mg/ml may be the intended target [4]. Many analytical techniques are not capable of handling such high concentrations, while the quality of FTIR spectra actually improves at high concentrations. In addition, in the 1990s a correlation between protein FTIR spectra in the solid state and their storage stability was found [5], as will be discussed further below.
1 Corresponding Author: Marco Van De Weert, Biomacromolecular Drug Delivery, Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark.
M. van de Weert and L. Jorgensen / Infrared Spectroscopy of Protein Pharmaceuticals
169
1700
1675
1650
1625
1600
Wavenumber (cm -1)

Figure 1. Example of an area overlap calculation using normalised and inverted second derivative spectra of native lysozyme (solid line) and heat-denatured lysozyme (dashed line). The grey area indicates the area overlap between the two spectra, which amounts to 55% of the normalised area of the two spectra.
Finally, there is an increasing focus on developing sustained release systems for protein pharmaceuticals. Many of these systems scatter radiation to such an extent that most spectroscopic techniques are not able to characterise the protein inside these systems. Also in this case FTIR has shown its value [69]. As a result, FTIR is now a well-accepted analytical technique in the pharmaceutical industry for analysis of protein pharmaceuticals. In this chapter, we will only discuss the analysis of pharmaceutical proteins in the solid state, in sustained release systems, and in a more complex mixture of additives. Analysis of protein pharmaceuticals in solution does not differ from analysis of any protein in solution, and has been discussed in detail in chapter 5.
2. Methodology For an in-depth description of the methodology used to analyse protein structure, the interested reader is referred to chapters 4, 5 and 6. These methods are highly useful to obtain global structural information about the protein. Often, however, the pharmaceutical scientist is less interested in the exact secondary structure of the protein, but rather wishes to monitor deviations from the native state. It is these deviations from the native state that often indicate physicochemical degradation of the protein, which usually is undesirable when developing a protein pharmaceutical. Thus, the pharmaceutical scientist is mostly interested in spectral similarity between the native protein and the protein in its formulation. Various methods have been developed to determine spectral similarity. Perhaps the most useful and practical method is the area overlap method [10]. In this method, the extent of area overlap between two normalised spectra is calculated (Fig. 1). Both raw and resolution-enhanced spectra can be used to this purpose. However, a main advantage of using second derivative spectra is the elimination of spectral distortions, such as baseline slope and band broadening due to scattering. Theoretically, the area overlap may range from 0 to 1 (or 0% to 100%), but in practice the lower limit is around 0.5 (50%) due to band overlap of the various secondary structural elements.
170
The example in Fig. 1 shows the area overlap between the area-normalised (inverted) second derivative spectra of native lysozyme and a heat-denatured sample. The area overlap, or similarity, between the two spectra is 55%. Visual inspection of the spectrum shows a major increase in intermolecular -sheets mainly at the expense of -helical structure, as is common upon heat denaturation [11]. It should be noted that the changes in the spectra upon heat denaturation will be smaller for -sheet proteins. This is due to the smaller spectral change from intramolecular to intermolecular -sheet, as compared to the change from -helix to intermolecular -sheet. Thus, the values obtained from the area overlap method should only be compared for a single protein, and not between proteins.
3. Applications 3.1. Solid Protein Products Freeze-Drying Many protein pharmaceuticals are insufficiently stable in solution to be stored for prolonged periods of time. Generally, a protein pharmaceutical should be stable for at least 2 years, and only very limited physicochemical degradation is allowed. Since many of the degradation processes involve or are catalysed by water, the pharmaceutical industry is required to dry these formulations to slow these processes. For example, aggregation is significantly inhibited in the solid state due to the limited mobility, while deamidation requires the presence of water molecules, which are in abundance in solution [12]. An important aspect of this drying is the removal of water molecules, including those that are tightly bound to the protein. The most common drying procedure is freeze-drying, but alternative methods such as spray-freeze-drying and spray-drying are receiving increased attention. These drying procedures do impose stress forces on the protein, such as potential cold and heat denaturation, introduction of various interfaces (ice-water, water-air), up-concentration of the protein and other components in the formulation, and the removal of hydration water [13]. Thus, the drying process must be designed carefully to assure long-term storage stability of the dried formulation. A significant body of research is available on the freeze-drying of protein pharmaceuticals. It has been shown that a careful design of the process is an absolute requirement [13,14]. That is, a simple freeze-and-vacuum dry process often results in significant protein degradation, a solid material that is difficult to redisperse, and/or an aesthetically unacceptable solid material. A proper freeze-drying process usually involves a freezing phase, sometimes including an annealing step to increase the size of the ice crystals, a primary drying phase under vacuum to remove most of the ice crystals, and a secondary drying phase, in which the temperature of the sample is increased to rapidly remove the remaining ice as well as most of the tightly-bound water molecules. This type of freeze-drying process may take days, and much time can be saved by prior knowledge of fundamental physicochemical parameters, in particular the glass transition (Tg) or eutectic temperature of the liquid formulation [14]. Apart from a carefully designed process, most proteins also require the presence of so-called lyoprotectants, such as a non-reducing carbohydrate, to obtain a stable product. Initially it was thought that these lyoprotectants protected the protein by forming a glassy state with high glass transition temperature (Tg) upon drying, thus reducing po-
171
A
K
B
K
55
60
65
70
75
80
Similarity (%)
85
87
Similarity (%)
89
91
93
95
Figure 2. Correlation between spectral similarity of proteins in the native state and in a freeze-dried formulation, and the observed degradation rate upon storage in the solid state. (A) Rate constant of loss of monomeric growth hormone versus spectral similarity [16]. (B) Aggregation rate constant of a monoclonal antibody versus spectral similarity [17].
tential mobility of the protein. However, highly dried proteins themselves can have very high Tgs (> 100 C) [15], and several compounds with a high Tg do not stabilise proteins as efficiently as those with lower Tg [13]. A potential answer to the function of lyoprotectants has come from FTIR analysis. When comparing the protein structure before and after freeze-drying, many have found a negative correlation between the extent of the changes and the (storage) stability of the protein in the freeze-dried formulation. That is, a lyoprotectant that is capable of reducing the extent of the spectral changes, usually increases the stability of the protein in the formulation. Figure 2 contains a few literature examples, where we have plotted the level of spectral similarity of native and freeze-dried protein against the degradation rate of the freeze-dried protein. Unfortunately, such literature examples are very limited, and many articles merely mention there is a better stability for freeze-dried proteins of which the FTIR spectrum more closely resembles that of the native protein. In Fig. 2A, we have plotted the spectral similarity in the spectra versus the observed rate constant of loss of monomeric human growth hormone, as reported by Costantino et al. [16]. The authors reported only the percentage -helix, -sheet and unordered structure, and we calculated the average percentage similarity to the native state of these three structures. Generally, there is a trend that a higher similarity to the native state yields a better stability. However, these data were obtained using different stabilisers, each with their own glass transition and tendency to crystallise. Thus, this data does not rule out other explanations for the increased stability at higher spectral similarity. For example, some of the samples with the lowest spectral similarity contained polyalcohols with very low glass transition temperatures, or polyalcohols that tend to crystallise. Figure 2B shows a plot of spectral similarity versus the observed degradation (aggregation) rate of a freeze-dried IgG1 antibody [17]. In this case, all samples contain the same stabiliser (sucrose). It is clearly visible that the degradation rate is smaller if the structure more closely resembles that of the native protein. However, the increased retention of native structure also correlates with the relative amount of sucrose in the sample. That is, the reduced aggregation rate may also be caused by a more pronounced dilution of the protein within the freeze-dried sample. A correlation between spectral change and (storage) stability can be explained by assuming that the spectral change is mainly due to a structural change. Generally, any
172
A
1700
1675
Wav enumber (cm-1 )
1650
1625
1600
Figure 3. Normalised inverted second derivative spectra of glucagon and PEGylated glucagon. Glucagon in solution (solid line), glucagon freeze-dried from a water/acetonitril mixture (dashed line), and freeze-dried PEGylated glucagon (5 kDa PEG chain) from a water/acetonitril mixture (dotted line) [18].
structural change of a protein may expose otherwise buried hydrophobic patches. Even within a solid with high Tg molecular motions are still possible, and partially unfolded proteins in the freeze-dried formulation may thus interact and slowly aggregate. Moreover, the protein may be more extended and expose more sites susceptible to chemical degradation. Over time, the formulation containing partly unfolded protein is then more prone to physicochemical degradation. Conversely, a freeze-dried formulation containing the protein in its (near-)native state is less likely to degrade. The apparent correlation means that FTIR can be used to decrease the development time of a stable freeze-dried product. Those formulations in which the protein spectrum in the solid state differs significantly from that in the native state are more likely to have a poor long-term stability, compared to those where the spectra are more similar. Thus, those formulations and freeze-drying protocols that are least likely to give a stable product can be rapidly identified and eliminated, rather than waiting long periods of time, sometimes months, until degradation may be apparent after rehydrating the freeze-dried product. Ultimately, however, long-term stability studies will be required to show that the chosen formulations are good enough to yield a two-year stable product. FTIR has also been used by Stigsnaes et al. [18] to study the effect of PEGylation on the processing stability of the model peptide glucagon. PEGylation is a common process for many peptide and protein pharmaceuticals, and involves covalent attachment of polyethyleneglycol to selected amino acids in the protein chain. This chemical modification increases circulation times with factors up to 100, increases solubility, and decreases potential immune responses to the protein [19]. The study by Stigsnaes et al. [18] showed that PEGylation of glucagon also results in a peptide that is less sensitive to the stresses incurred by the freeze-drying process, as evidenced by the reduction in spectral changes (Fig. 3). 3.2. Solid Protein Pharmaceuticals Spray-Drying FTIR is not only used in the analysis of freeze-dried proteins, but also dried formulations obtained through other means, such as spray-drying. Although research within
173
A
1700
1675
Wav enumber (cm-1 )
1650
1625
1600
Figure 4. FTIR spectra of spray-dried calcitonin formulations with varying weight ratios of calcitonin:chitosan:mannitol. 1:0:18 (solid squares), 1:1:18 (open squares), 1:3:16 (solid circles), 1:4:15 (open circles). The arrow shows the decrease in -helical structure upon increasing the amount of chitosan [28].
100 95 90 85 80
Recovery (%)
85
90
Similarity (%)
95
100
Figure 5. Correlation between spectral similarity of the spectra in Fig. 4 with the sample without chitosan and recovery upon rehydration in phosphate buffer pH 7.4 (squares) or acetate buffer pH 4.4 (triangles) [28].
this area is still limited, the available literature indicates a similar correlation between spectral changes and (storage) stability [2027]. In studies on spray-dried calcitonin formulations containing chitosan, increasing protein spectral changes were observed upon increasing the amount of chitosan in the formulation (Fig. 4) [28]. Subsequent rehydration of these chitosan-containing formulations indicated a reduced recovery of the protein, correlating with the reduction in area overlap compared to the sample spray-dried in the absence of chitosan (Fig. 5). The recovery was higher in the acetate buffer, since calcitonin fibrils will dissolve in this buffer, but not in phosphate buffer [28]. The observed negative influence of chitosan is surprising, since no interaction between calcitonin and chitosan was observed in solution. Currently, no satisfying explanation is available to explain the reduced recovery [28]. These studies thus call for a
174
A
1710
1690
1670
1650
1630
1610
1590
Wav enumber (cm-1 )

Figure 6. Normalised inverted second derivative spectra of insulin entrapped in alginate-chitosan nanoparticles. Different weight ratios of alginate-chitosan were used, resulting in very similar spectra (black lines). A control spectrum of insulin in solution is also shown (dashed line) [29].
more critical approach towards the use of chitosan in formulations containing proteins, as many of such formulations will ultimately be dried. Similar unwanted interactions may occur in formulations containing other proteins. 3.3. Solid Protein Pharmaceuticals Sustained Release Systems Almost all protein formulations are administered through injection or by infusion, which introduces a significant problem in terms of patient compliance. Over the last few decades an enormous effort has been aimed at developing so-called sustained, or controlled, release systems for proteins. Slow, sustained release of the protein from such a system significantly prolongs therapeutic concentrations in the bloodstream, minimising the number of injections required to sustain these concentrations. Moreover, a properly designed system would decrease the fluctuations in blood concentration, which may increase the therapeutic efficacy of the protein drug. The major challenges in designing a proper sustained release system are to achieve appropriate release kinetics, and to assure protein stability within the system. The latter imposes a significant analytical challenge, since these systems are designed to entrap the protein. Any method to break down the system may have a pronounced effect on the protein structure, making it impossible to identify whether the protein was already altered in the system itself, or affected by the extraction procedure. Once again, FTIR allows analysis of the protein structure while it is still entrapped in the system, and thus (partly) resolves this question [69]. Figure 6 shows an example of spectral analysis of an entrapped protein [29], and contains FTIR spectra of insulin entrapped in nanoparticles prepared from mixtures of alginate and chitosan. These oppositely charged polymers interact through electrostatic interactions, forming complex coacervates of about 800 to 1700 nm in diameter. Only limited structural changes of the protein, insulin, are observed upon entrapment, suggesting that the interaction and entrapment procedure do not have a major detrimental effect on protein structure. Moreover, the released protein also has the same structure as the native protein (not shown). Whether this also means that the protein is stable for prolonged periods of time within this matrix needs to be investigated further. Espe-
175
cially when these particles need to be (freeze-)dried, there is some concern as to the possible negative effects of chitosan, as described above for calcitonin. A number of studies have focused on entrapment into poly(lactide-co-glycolide) (PLGA) microspheres. Here, various spectral changes have been observed, including the formation of intermolecular -sheets [8]. The latter structure typically results in (partially) non-reversible protein aggregates. Thus, this reduces the content of active protein, and may ultimately result in the release of aggregated protein into the circulation. Aggregates are thought to be one of the main causes of the unwanted immune response to protein pharmaceuticals [30], and their level must be kept as low as possible. Interestingly, the presence of a large amount of structurally altered lysozyme inside a polymeric matrix composed of a poly(ether ester) has also been observed using FTIR [31]. The spectra indicated that the protein had formed non-covalent aggregates inside the matrix, as evidenced by large absorption bands around 1625 and 1695 cm 1. However, the protein was released in its native and fully active form. A possible explanation for this surprising observation is the formation of rather loose aggregates, which easily dissociate upon hydration. Moreover, the matrix may act as a template for refolding. However, in the pharmaceutical industry and the regulatory agencies, in particular, the presence of these aggregates inside the matrix would raise significant concerns. It is likely that these aggregates become less reversible upon long-term storage, and they may be more prone to other types of degradation. 3.4. Controlled Release Systems Emulsions Proteins may also be entrapped in emulsions. Such emulsions scatter light, and are, by design, difficult to break into the original two phases. Thus, structural analysis of protein structure in such emulsions is very difficult, although not impossible [32]. Here also, FTIR plays an important role. The structure of proteins entrapped inside these emulsions is often different from that in solution. For some proteins a decrease in the -helical content is observed, for others formation of (intermolecular) -sheet is apparent, as is the case for growth hormone (Fig. 7A) [33]. For yet other proteins only slight alterations may be observed, exemplified using glucagon in Fig. 7B. Formation of intermolecular -sheets, at the expense of -helix, has also been observed for -lactoglobulin adsorbed to the oilwater interface [3436]. Thus, this structural rearrangement appears to be common for several proteins, and is a potential concern when using emulsions for drug delivery purposes. Their appearance suggests the formation of aggregate formation, which is highly undesirable for protein pharmaceuticals. Establishment of the structure of the released protein, its activity upon release, and whether all protein is released will have great influence on the choice of delivery system. Protein release from the water-in-oil emulsions is possible, as has been shown by in vitro release of aprotinin (up to 2% released) and insulin aspart (up to 30% released) [3739]. The amount of insulin aspart released can be modified by alterations of the osmotic gradient from the internal aqueous phase to the release media. Nevertheless, only ~30% insulin aspart is released from the emulsion and no certainty of retention of the activity is present [39]. The activity of aprotinin after extraction from the water-inoil emulsion was about 87%, indicating that the structure of aprotinin is not irreversibly altered by the incorporation into the emulsion [37]. It would thus appear that for some proteins an emulsion may be a suitable drug delivery system.
176
A
A
B
A
1700
1700
1650
1600
1650
1600
Wav enumber (cm-1 )
Wav enumber (cm-1 )
Figure 7. Normalised inverted second derivative spectra of native protein (solid line) and protein entrapped in emulsions (dotted line). (A) Growth hormone; (B) Glucagon.
4. Concluding Remarks and Future Directions FTIR is slowly becoming an accepted technique in the pharmaceutical industry for the analysis of protein pharmaceuticals. Its application in detecting drying-induced instability of protein formulations is a standard method in many companies, although the number of publications is still limited. The same applies to structural analysis of proteins entrapped in sustained release systems. Proof-of-Concept is available, but wellestablished procedures and methods are still missing. A more concerted effort is required to firmly establish FTIR as an important and invaluable technique for analysis of protein pharmaceuticals. For example, structural analysis of a wide variety of proteins in a wide variety of sustained release systems is required to thoroughly establish any correlation between FTIR spectra and quality of the release system. Other vibrational techniques are also slowly being implemented in the pharmaceutical industry. Raman and NIR spectroscopy are suitable for on-line analysis, and may further shape the rapid development of protein pharmaceuticals. Although application of these techniques to protein pharmaceuticals is still limited, we are convinced that the next decade will see an exponential growth within this area. Process analytical technology (PAT), aimed at on-line analysis, is one of the key areas that regulatory agencies would like to see developed further. Proper use of these techniques would, however, require a better insight into the correlation between protein structure and protein Raman or NIR spectra. Moreover, improved algorithms will be required to handle the large amount of data coming from such on-line analysis. This is likely to be the main challenge for the next decade.
Acknowledgements The authors would like to acknowledge Drs Pernille Stigsnaes, Mingshi Yang, and Bruno Sarmento for providing the data shown in Figs 3, 4+5, and 6, respectively.
References
[1] S. Lawrence, Nature Biotech. 24 (2007), 1466. [2] M. van de Weert, L. Jorgensen, E.H. Moeller, and S. Frokjaer, Expert Opin. Drug Deliv. 2 (2005), 1029-1037.
177
[3] Methods for structural analysis of protein pharmaceuticals, W. Jiskoot and D.J.A. Crommelin, eds., AAPS Press, Arlington, 2005. [4] S. Matheus, W. Friess, and H.-C. Mahler, Pharm. Res. 23 (2006), 1350-1363. [5] J.F. Carpenter, S.J. Prestrelski, and A. Dong, Eur. J. Pharm. Biopharm. 45 (1998), 231-238. [6] K. Fu, K. Griebenow, L. Hsieh, A.M. Klibanov, and R. Langer, J. Control. Release 58 (1999), 357-366. [7] K. Griebenow, I.J. Castellanos, and K.G. Carrasquillo, Int. J. Vib. Spec. 3, 5 (1999), 2. [8] M. van de Weert, R. van t Hof, J. van der Weerd, R.M.A. Heeren, G. Posthuma, W.E. Hennink, and D.J.A. Crommelin, J. Control. Release 68 (2000), 31-40. [9] T.-H. Yang, A. Dong, J. Meyer, O.L. Johnson, J.L. Cleland, and J.F. Carpenter, J. Pharm. Sci. 88 (1999), 161-165. [10] B.S. Kendrick, A. Dong, S.D. Allison, M.C. Manning, and J.F. Carpenter, J. Pharm. Sci. 85 (1996), 155-158. [11] M. van de Weert, P.I. Haris, W.E. Hennink, and D.J.A. Crommelin, Anal. Biochem. 297 (2001), 160-169. [12] Lyophilization of biopharmaceuticals, H.R. Costantino and M.J. Pikal, eds., AAPS Press, Arlington, 2005. [13] J.F. Carpenter, M.J. Pikal, B.S. Chang, and T.W. Randolph, Pharm. Res. 14 (1997), 969-975. [14] X. Tang and M.J. Pikal, Pharm. Res. 21 (2004), 191-200. [15] D.S. Katayama, J.F. Carpenter, M.C. Manning, T.W. Randolph, P. Setlow, and K.P. Menard, J. Pharm. Sci. 97 (2008), 1011-1022. [16] H.R. Costantino, K.G. Carrasquillo, R.A. Cordero, M. Mumenthaler, C.C. Hsu, and K. Griebenow, J. Pharm. Sci. 87 (1998), 1412-1420. [17] L. Chang, D. Shepherd, J. Sun, D. Ouellette, K.L. Grant, X. Tang, and M.J. Pikal, J. Pharm. Sci. 94 (2005), 1427-1444. [18] P. Stigsnaes, S. Frokjaer, S. Bjerregaard, M. van de Weert, P. Kingshott, and E.H. Moeller, Int. J. Pharm. 330 (2007), 89-98. [19] M. Morpurgo and F.M. Veronese, Methods Mol. Biol. 283 (2004), 45-70. [20] A.M. Abdul-Fattah, V. Truong-Le, L. Yee, L. Nguyen, D.S. Kalonia, M.T. Cicerone, and M.J. Pikal, J. Pharm. Sci. 96 (2007), 1983-2008. [21] H.-K. Chan, A.R. Clark, J.C. Feeley, M.-C. Kuo, S.R. Lehrman, K. Pikal-Cleland, D.P. Miller, R. Vehring, and D. Lechuga-Ballesteros, J. Pharm. Sci. 93 (2004), 792-804. [22] D.L. French, T. Arakawa, and T. Li, Biopolymers 73 (2004), 524-531. [23] Y.-H. Liao, M.C. Brown, T. Nazir, A. Quader, and G.P. Martin, Pharm. Res. 19 (2002), 1847-1853. [24] A. Mauerer and G. Lee, Eur. J. Pharm. Biopharm. 62 (2006), 131-142. [25] M. Maury, K. Murphy, S. Kumar, A. Mauerer, and G. Lee, Eur. J. Pharm. Biopharm. 59 (2005), 251-261. [26] S.U. Sane, R. Wong, and C.C. Hsu, J. Pharm. Sci. 93 (2004), 1005-1018. [27] S. Schle, W. Friess, K. Bechtold-Peters, and P. Garidel, Eur. J. Pharm. Biopharm. 65 (2007), 1-9. [28] M. Yang, S. Velaga, H. Yamamoto, H. Takeuchi, Y. Kawashima, L. Hovgaard, M. van de Weert, and S. Frokjaer, Int. J. Pharm. 331 (2007), 176-181. [29] B. Sarmento, D.C. Ferreira, L. Jorgensen, and M. van de Weert, Eur. J. Pharm. Biopharm. 65 (2007), 10-17. [30] A.S. Rosenberg, AAPS J. 8 (3) (2006), 59. [31] M. van de Weert, R. van Dijkhuizen-Radersma, J.M. Bezemer, W.E. Hennink, and D.J.A. Crommelin, Eur. J. Pharm. Biopharm. 54 (2002), 89-93. [32] L. Jorgensen, M. van de Weert, C. Vermehren, S. Bjerregaard, and S. Frokjaer, J. Pharm. Sci. 93 (2004), 1847-1859. [33] L. Jrgensen, C. Vermehren, S. Bjerregaard, and S. Froekjaer, Int. J. Pharm. 254 (2003), 7-10. [34] Y. Fang and D.G. Dalgleish, J. Colloid Interface Sci. 196 (1997), 292-298. [35] F.A. Husband, M.J. Garrood, A.R. Mackie, G.R. Burnett, and P.J. Wilde, J. Agric. Food Chem. 49 (2001), 859-866. [36] T. Lefvre and M. Subirade, J. Colloid Interface Sci. 263 (2003), 59-67. [37] S. Bjerregaard, L. Wulf-Andersen, R.W. Stephens, L. Rge Lund, C. Vermehren, I. Sderberg, and S. Frokjaer, J. Control. Release 71 (2001), 87-98. [38] S. Bjerregaard, H. Pedersen, H. Vedstesen, C. Vermehren, I. Sderberg, and S. Frokjaer, Int. J. Pharm. 215 (2001), 13-27. [39] L. Jorgensen, C. Vermehren, S. Bjerregaard, and S. Frokjaer, J. Drug Del. Sci. Tech. 14 (2004), 455-459.
178
Quantum Mechanical Calculations of Peptide Vibrational Force Fields and Spectral Intensities
Jan KUBELKAa* , Petr BOUb , Timothy A. KEIDERLINGc a Department of Chemistry, University of Wyoming, USA b Institute of Organic Chemistry and Biochemistry, Academy of Sciences, Czech Republic c Department of Chemistry, University of Illinois at Chicago, USA
Abstract. Vibrational spectra are frequently used for studies of the structure and dynamics of peptides and proteins. Structural interpretation of the experimental data, however, requires theoretical simulation of the spectra for model peptide geometries. Quantum mechanical, in particular density functional theory (DFT), methods have proven exceptionally valuable for calculations of the vibrational force fields and both IR and Raman intensities. A brief review of some recent trends in computation of molecular force fields and spectral intensities is presented. Particular attention is paid to experiments involving circularly polarized light, as these provide enhanced structural information for chiral molecules. Following a historical overview of common approaches, the fundamental theoretical aspects of the calculations of the molecular vibrational spectra are summarized. Special emphasis is given to the problem of simulating spectra for biological molecules (oligo-peptides and nucleotides, proteins and nucleic acids) using DFT methods. The methodology for simulations of large biopolymers with DFT level force fields and intensity parameters abstracted from smaller molecules is reviewed. Several examples with the discussion of successes and difficulties of the vibrational spectra simulations for model peptides are presented. Finally, methods for incorporating the solvent in the spectral simulations are reviewed and discussed.
Keywords. Vibrational spectra, Infrared, Raman, vibrational circular dichroism, peptide force fields, density functional theory, secondary structure, solvent effects.
* Corresponding Author: Department of Chemistry, University of Wyoming, Laramie, WY 82071 USA; Email: jkubelka@uwyo.edu. Corresponding Author: Institute of Organic Chemistry and Biochemistry, Academy of Sciences, Flemingovo nm. 2, 16610 Prague 6, Czech Republic; E-mail: bour@uochb.cas.cz. Corresponding Author: Department of Chemistry, University of Illinois at Chicago, Chicago IL 606077061 USA; E-mail: tak@uic.edu.
J. Kubelka et al. / QM Calculations of Peptide Vibrational Force Fields and Spectral Intensities
179
1. Introduction Vibrational spectroscopic methods are important experimental tools for studies of biological molecules: peptides, proteins and nucleic acids. In particular, various Fourier Transform Infrared (FTIR) and high sensitivity Raman techniques are among the most frequently used methods in protein structure, folding and function analyses [1-4]. The vibrational modes of the amide group (the most useful ones are illustrated in Figure 1) are sensitive probes for the secondary structure, which can be utilized in both infrared (IR) and Raman studies. Historically, such analyses were carried out by empirical correlations of IR or Raman amide band frequencies with secondary structures for model polypeptides. Theoretical approaches to understanding the characteristic polypeptide spectral features were based on simplified vibrational calculations through coupled oscillator models or parameterized force fields (FF) obtained by fitting empirical force constants for internal coordinates to observed spectra. Within the last decade, quantum mechanical (QM) computations of complete FF offered a powerful means for determination of peptide vibrational properties independent of problemspecific empirical parameters [5].
Amide I 1600-1700 cm-1
Amide II 1480-1580 cm-1
Amide III 1210-1350 cm-1
Figure 1. Schematic representation of three normal modes of vibration of the amide group which are most important for peptide and protein secondary structure analyses: (a) amide I, (b) amide II, (c) amide III.
While most early IR and Raman studies used frequencies of component bands as structural markers, the intensities and polarizations of individual modes provide additional, valuable structural insights. In particular they are necessary for modeling band shapes of overlapping transitions, as found in biopolymer systems. Vibrational circular dichroism (VCD) and Raman optical activity (ROA) are polarization methods whose conformationally determined sign patterns add to the vibrational frequency resolution and enhance structural sensitivity through characteristic bandshapes. VCD and ROA intensities arise from the differential response (absorption or scattering, respectively) to left- and right-circularly polarized light by chiral structures, and therefore are very sensitive to the conformation of peptides (or nucleic acids) [6-8]. A major aspect of the renewed interest in vibrational spectroscopy studies of biopolymeric molecules arises from the ability of IR and Raman to sense fast time scale motions. Various rapid mixing schemes have been proposed for millisecond resolution [9], laser phototriggering [10-12] and temperature-jump [13, 14] designs for specific protein schemes have been utilized to follow nanosecond scale processes. In addition recent 2-D IR studies have probed bio-molecules with femtosecond resolution [15, 16]. Finally, utilization of selective isotopic labeling has introduced site-specific
180
conformational sensitivity into IR, VCD and Raman techniques, which normally provide only average structural information [17]. Interpretation of the additional spectral details and associated extended applications for vibrational spectra is dependent on their accurate theoretical description [18, 19]. With development of fast, inexpensive computers and approximations allowing solution of the Schrdinger equation for larger systems, quantum mechanical (QM) force fields (FF) as well as intensities for IR, Raman, VCD [20-23] and ROA can be determined [22-25]. Initial biomolecule-oriented studies focused on small model systems, but QM vibrational spectral calculations have become feasible for moderate oligopeptides [26, 27], and by use of transfer methods [28], QM level FFs are now realistic for large peptides [29-31] and even proteins [32]. Parallel applications for nucleic acids have also developed [33]. The development of such ab initio FFs and intensity parameters for peptides and their application to vibrational spectra are the topic of this review. As for all theoretical models, the ab initio simulations of vibrational spectra must be critically evaluated by comparison with experiment, for which we supply several examples, as do other chapters in this book. After a brief survey of empirical methods, a more detailed description of ab initio spectral calculations with an emphasis on density functional theory (DFT) approaches will be presented. Extension to large biomolecules utilizing the transfer of DFT vibrational parameters will be discussed along with specific examples focused on peptide systems. Finally, approaches to correction of the simulated spectra for solvation and other structural perturbations will be discussed. This review focuses primarily on equilibrium applications and linear vibrational spectra analyses; in a separate chapter Choi and Cho discuss dynamics and non-linear, 2-D IR simulations [34].
2. Observed Peptide and Protein Vibrational Spectra Infrared (IR) absorption frequencies have for a long time characterized peptide and protein secondary structure [1, 35-39]. The most studied band in the IR of peptides is the amide I, which is primarily amide C=O stretch (Figure 1) and generally appears between 1600 and 1700 cm-1, but typically at 1650-1660 cm-1 for D-helices and 16201640 cm-1 with a weaker component at ~1680-90 cm-1 for E-sheets (see Figure 2 for example peptide IR and VCD spectra). Due to the C=O group orientation, the amide I is polarized with respect to the helix axis in D-helices and E-sheets, which provides an added diagnostic for oriented samples [40]. In VCD, the D-helix and coil amide I bands have oppositely signed couplet patterns (Figure 2), while for E-sheets they are weak and predominantly negative in polypeptides (aggregates), but stronger in globular proteins where such sheets are limited in extent and quite twisted.
181
Figure 2. Examples of IR (left) and VCD (right) spectra (amide I and II regions) for model peptides with characteristic secondary structures: (top) D-helix (middle) E-sheet, (bottom) unordered, often termed: random coil but locally left-handed 31-helical (which is sometimes denoted PPII).
Figure 3: Examples of experimental Raman spectra in the amide I III region for an alanine-rich peptide, Ac-(AAAAK)3-AAAAY-NH2 (left), in an D-helical conformation (A) at 5oC and in an unordered conformation (B) at 55 oC, plus (right) a highly D-helical protein, bovine serum albumin (C), and a mostly E-sheet protein, concanavalin A (D).
The amide II (primarily in-plane NH deformation mixed with C-N stretch, ~15001580 cm-1) and the amide A (N-H stretch, ~3300 cm-1 but quite broad) bands have less pronounced frequency shifts with change in secondary structure (Figure 2) although they are polarized and highly sensitive to deuteration effects, which can act as a measure of solvent exposure [41-43]. The amide III (opposite-phase combination of NH bend plus C-N stretch) is mixed with other local modes, particularly the CD-H deformation, and very weak in the IR, [36, 44-46, 47 , 48], but is much more important in Raman analyses, where both the amide I and III are used for peptide structural studies [3, 4, 49]. Some examples of Raman spectra in amide I III regions are in Figure 3. Raman is also sensitive to some side-chain modes, particularly of aromatics and disulfide (-S-S-) linkages [50]. Resonance Raman studies of aromatic residues and ligands (such as heme groups) have proven very useful for detailed protein studies of
182
local changes [51-53] while UV Resonance Raman studies of the amide modes have been applied to folding analyses [54-57].
3. Computations of Peptide Vibrational Spectra 3.1. Empirical Calculations of Peptide FF and Vibrational Frequencies Miyazawa proposed a scheme for calculation of vibrational frequencies of the amide I mode for regular helical chains, using a coupled oscillator treatment, whereby identical, local amide oscillators were weakly coupled to nearest neighbors through covalent and hydrogen bonds [58, 59]. Assuming an infinitely long chain in an exciton-like model with helical or sheet-like symmetry reduces the problem to a few allowed distributed modes [60]. This theory provided a basic explanation for the IR spectral differences observed for different conformations. Chirgadze and Nevskaya carried out coupled oscillator calculations of amide I IR spectra for antiparallel and parallel E-sheet structures [61, 62], and of the amide I and II for D-helices [63]. Their perturbation model was based on coupling of the single amide oscillators via a transition dipole coupling (TDC) mechanism, which was able to account for the experimentally observed bandshapes for different structures and establish segment size dependence, predicting the effect of D-helix or a E-strand length and the number of strands in a E-sheet. Torii and Tasumi used essentially the same methodology to simulate the amide I IR spectral bandshapes of complete globular proteins and their segments [64-66]. This simplification of the problem to a single oscillator for each amide group made it possible to diagonalize a full protein interaction matrix, since proteins do not have translational symmetry of regular polypeptides. The coupled oscillator approach has been extended by combining various vibrational coupling mechanisms (through-bond, hydrogen bond and TDC), and is often used for simulations of spectra of various structures, including isotopically labeled peptides and locally distorted structures with resolved component bands [6770]. An alternate formulation of the coupled oscillator scheme, termed exciton coupling, considers vibrational coupling between excited states of local, otherwise independent, quantum mechanical oscillators. In practice, this is equivalent to the coupled oscillator approach with alternate interactions [15, 31, 71-73]. All atom empirical force fields (FF) are derived by adjusting internal coordinate force constants (bond stretches and bends) based on the detailed structure of the molecule to best fit observed frequencies. Systematic development of the force fields for polypeptides was undertaken by Krimm and coworkers by refinement of the force field parameters to fit the experimental data [74]. Normal mode calculations and vibrational frequency assignments were developed for a number of polypeptide conformations. TDC was also incorporated into their atomic FF [74, 75], which provided better agreement of the calculated amide I vibrational frequencies with experiment. Others have taken a similar approach by transferring ab initio FF parameters from N-methylacetamide (NMA) or small peptides [76-79].
183
3.2. Empirical Calculations of Spectral Intensities Approximate calculations of the IR spectral intensities are straightforward within the empirical approaches discussed above, and can be derived from the unperturbed amide transition dipoles (coupled oscillator approaches) [58, 61, 62, 64-66, 69, 70, 80-82], bond dipoles or atomic partial charges (atomic FF methods) [18, 75, 83, 84]. These calculations are sufficient to account for the overall appearance of the IR spectra, such as locating the strongest and weakest bands and even approximate the exciton distribution in the bands. While extensive formulations of the Raman polarizability tensors have appeared for many years and have been used for qualitative analyses of Raman intensities, until recently, less work was done on systematic simulation of spectra, particularly for peptides. Some models like the bond and atom polarizability approaches did appear [22, 85-87] and variants are used in materials applications [88]. However, for polarization spectra, such as linear dichroism (LD) and especially VCD or ROA, the fundamental spectral bandshapes are critically dependent on the relative intensities and the sign patterns of the differential spectral bands. These tend to arise from detailed impact of molecular distortions on the electronic distributions and cannot be accurately modeled with fixed dipole or charge models. The first theoretical VCD was an empirical coupled oscillator model, applied to a cyclic dipeptide, where the TDC causes the frequency splitting but also gives rise to the circular dichroism signal [89, 90]. The first polypeptide applications used parameters from Miyazawa and Krimm perturbation theories for coupling of amide oscillators and experimental transition electric dipole values [91, 92]. While this model provided correct qualitative predictions in some cases, especially for the D-helical amide I, it was unsuccessful for other modes and conformations. Diem extended the model to non-degenerate oscillators and used it to calculate IR absorption and VCD of polypeptides and nucleic acids [68, 93-96]. The classical DeVoe polarizability theory approach [97, 98] has been applied to IR and VCD of nucleic acids [99, 100]. A similar model, formulated in terms of the linear response polarizability tensors has been applied to polypeptides [101] and small cyclic peptides [102]. In general, while these models often give reasonable predictions for the VCD of nucleic acids, where through-space coupling as modeled by TDC dominates, and are sometimes adequate for peptide IR, particularly for sheet or extended conformations, they are typically inaccurate when applied to peptide VCD. Accurate simulation of peptide VCD requires quantum mechanical treatment to properly model through-bond effects (coupling) on the wave function.
4. Quantum Mechanical Calculations of Vibrational Spectra Quantum mechanical calculations provide complete molecular FFs (including all the vibrational interactions) without need for empirical parameterization. In addition, QM models, now developed mostly at the Density Functional Theory (DFT) level, are critical for obtaining accurate IR, Raman, ROA and VCD spectral intensities and are implemented in most quantum mechanical computational packages. Although VCD computation has some fundamental issues requiring a certain degree of care, various theoretical schemes for VCD have been implemented, as has been extensively reviewed [20-23, 103-105]. Finally, methods for determining the magnetic and quadrupolar derivatives needed for ROA have become increasingly available [25, 106].
184
4.1. Vibrational Spectral Frequencies In the Born-Oppenheimer (BO) approximation [107], central to most molecular QM calculations, the total molecular wavefunction is separated into electronic and nuclear components, represented as <(r,R) = Iel(r,R)FX(R) where the electronic function depends on the nuclear positions, R, but not velocities. Solving the electronic Schrdinger equation yields an electronic energy for the ground state, H(Ri). Combining H(Ri) with the nuclear repulsion, Vnn(Ri), forms an effective (average) potential energy term for the motion of the nuclei. Most molecular vibrational analyses are done in the harmonic approximation, where the potential is expanded around the equilibrium position with respect to the nuclear displacements, 'Ri, and only the leading (quadratic) term is retained. The nuclear Hamiltonian then becomes: 2 3N 3N 1 t 1 3 N Pi t (1) Fij 'Ri 'R j H 2 p .p q .f.q 2 i 1 mi i 1 j 1 where qi
mi 'Ri and f ij
Fij / mi m j are the mass-weighted coordinates and force
field (q and f are vector and matrix representations), respectively, and Pi are the nuclear momenta (correspondingly, pi are mass-weighted). The Cartesian force constant matrix, w2H , (2) Fij wRiwR j is referred to as the (harmonic) force field (FF, also referred to as the Hessian). Anharmonic effects (higher order energy derivatives) are small for the most important peptide vibrations (amide I III), but can be included [22] at much greater computational cost [108, 109]. In the harmonic approximation, the multidimensional Hamiltonian can be elegantly reduced to a sum of independent one dimensional operators by a unitary transformation of coordinates from mass-weighted Cartesian displacements and momenta (qi and pi) to normal mode coordinates Qk and corresponding momenta 3k,
Qk
s
j 1
3N
1 kj
qj ,
3k
s
j 1
3N
1 kj
pj
(3)
By substitution of (3) into (1)
1 t t (4) 3 .s .s.3 Q t .s t .f.s.Q . 2 When the transformation is unitary (st.s = E, where E is the identity matrix) and the transformed force field, st.f.s = /, is diagonal (i.e. /ij=0 for izj and /ii=Zi2, which can be always done for a quadratic potential form), the Hamiltonian becomes a sum of onedimensional harmonic oscillator Hamiltonians hi H
3N 1 3N 2 2 2 (5) 3 Z Q i i i hi (Qi ) . 2i1 i 1 Consequently, the equation of motion for the nuclei (substituting the QM operator for momentum, 3 = -i w/wQ) reduces to a set of 3N uncoupled 1-D Schrdinger equations
Note, in molecular mechanics force field denotes a more general, empirical dependence of the energy on internal coordinates than represented by these derivatives.
185
1 2 w2 ! Zi2Qi2 2 F i (Qi ) 2 w Q i
Ei F i (Qi ) ,
i = 1 , 2, , 3N
(6)
The harmonic oscillator energies, Ei,v, and wavefunctions, Fi, can be obtained analytically, as 1 (7a) E i ,X !Z i X i 2 (7b) AXi H Xi ([ )e 2 i where i is the quantum number, [ Z !Qi and AH are the normalized Hermite polynomials. Instead of s, a direct Cartesian-normal mode transformation matrix, S, can be defined as w'R j (8) S s / m .
kj
F X ([ )
[2
wQk
kj
Once the molecular FF is known, calculation of the vibrational frequencies and normal modes of vibration is straightforward: the Cartesian force constant matrix is mass weighted and diagonalized; the resulting diagonal constants correspond to squares of the quantum energies for the fundamental transitions (i = 0 for all i o i = 1 for a normal mode i). The accuracy of vibrational frequencies and normal modes is determined by the accuracy of the FF. Obtaining an accurate FF (equation 2) is therefore the main and most difficult task of vibrational frequency calculations. Coordinates and geometries. QM vibrational calculations are generally done using force constants (Fij) directly computed in Cartesian coordinates for which the solution contains 3N coordinates Qk. The rotational and translational degrees of freedom, corresponding to external motion of the molecule as a whole, will correspond to six (five if the molecule is linear) coordinates whose eigenvalues are zero and can be separated from the internal nuclear (vibrational) coordinates [22, 110, 111]. This is in contrast to more traditional approaches to molecular FF determination, which are based on more chemically intuitive and transferable internal coordinate representations (i.e. bond stretches and bends). While the force field could be transformed to internal, nonredundant symmetry-adapted coordinates, where the zero-energy modes do not appear, this has a disadvantage in that the kinetic energy in (1) becomes more complicated in internal coordinates, no longer being separable. Since harmonic vibrational frequency calculations are applicable only to molecules whose geometry corresponds to the minimum of the potential energy surface, geometry optimizations should be carried out prior to calculation of vibrational frequencies. However, to explore the vibrational characteristics of specific biopolymer conformations, it is typically necessary to constrain the geometry optimizations to represent realistic, large biomolecular structures. Smaller model conformers used for computation are not stable under full, unconstrained energy minimization due to the limited sizes and solvation approximations required for QM FF calculations. However, in our work, only backbone torsional motions are constrained, since they do not significantly affect the high frequency, stretching and bending modes of interest for spectral analysis, whose coordinates were fully optimized. Alternately, low-frequency normal mode coordinates can be fixed directly, as they approximately correspond to the torsional and other large-amplitude motions [112-114]. Other groups have carried out full minimizations of larger structures, by limiting basis sets or coupling with semiempirical models [27, 233, 234].
186
4.2. Spectral Intensities The spectral intensities of molecular vibrational transitions can be formulated using quantum mechanical time-dependent perturbation theory [22, 87]. IR and Raman intensities can be obtained at the first level of approximation, where the electromagnetic field interacts only with the molecular dipole moment. For VCD and ROA the magnetic dipolar and electric quadrupolar interactions must be included. An important difference between the absorption (IR, VCD) and scattering (Raman, ROA) processes is that the former are one-photon and the latter two-photon processes. IR intensity: atomic polar tensor. In the electric dipole approximation, the IR intensities are commonly expressed as the dipolar strength, which is the square of the transition electric dipole moment:
DX 0
P0X PX 0
PX 0
FX PD F 0
(9)
where PD denotes the electric dipole moment operator and the index refers to the vibrational quantum number. Due to selection rules for the harmonic oscillator ' = r1, only = 1 needs to be considered in (9). The dipole moment can be expanded around the equilibrium nuclear positions as a function of the normal modes, Qj. Keeping only the linear term and using linear harmonic oscillator wavefunctions, the D-th component of transition dipole moment for the normal mode j becomes:
F 1 PD F 0
j
! 2Z j
wPD wQ j
! 2Z j
PED S E
A A,
(10)
AE , j
where S is defined in (8) and the atomic polar tensor (APT, "dipole derivatives") is defined as: wP A (11) P A D = E ED Z A eG ED
ED
wR AE 0
A where E ED is the electronic contribution, and the last term in (11) is the nuclear contribution to the APT. Evaluation of the nuclear contribution is simple and can be related to empirical modeling of intensities based on effective charges fixed to the nuclei [91, 115]. In that model, initially partial charges were guessed and later taken from quantum chemical calculations of charge distributions (Mulliken populations). Later extensions allowed for charge flow and for localized molecular orbital motion [23, 83, 105, 116]. These methods are not very accurate since electron charge responds virtually instantaneously to nuclear motion, the basis of the BO approximation. However, they may be still useful in simplified QM/MM models to obtain a first approximation of the spectral intensities [117]. A less trivial task is to obtain the electronic contribution, which is done using perturbation theory [22, 118, 119]: wM , w w 2H wH (12) A G
EED
wR A, E
MG PD MG
wR wF A, E D 0 , 0
2 MG wFD 0 wRA, E 0
0 where M G is the electronic ground state wavefunction and F the electric field, the
electric field derivative of the Hamiltonian is the dipole operator: (H/FD)0 = -Pel,D The derivatives can be obtained either numerically or, more accurately and efficiently, by use of analytical derivative techniques, which will be discussed later.
187
Raman intensity: polarizability derivatives. Raman intensities have similar dependencies on the properties of the BO wavefunction and the molecular polarizability tensor, D, once expanded in the normal coordinates. This approach is known as the Plazcek approximation [22, 87, 120]. The polarizability depends on the light frequency Z as: Z gn 2 (13) D Re P gn P ng , 2 ! n Zng Z 2 where g and n denote the ground and excited electronic state and Zgn is the energy difference of ground and excited states expressed as a frequency, 'E/!. Within the BO (Plazcek) and harmonic approximations, each vibrational normal mode contributes independently to the spectral intensity, via isotropic invariants of the polarizability. These polarizability derivatives can be computed ab initio, as the second derivative of the energy H with respect to the electric field, F, and the normal mode derivatives can be obtained by further differentiation of D by Q. In practice, Cartesian derivatives are computed and transformed into normal modes utilizing the S-matrix, much as for the APT. Cartesian derivatives of D related to individual atomic coordinates are analogs of the APT for the scattering process. Taking the linear term in the expansion of D, the allowed transitions in the harmonic approximation are from i =0 to i=1. Observed Raman transition intensities are not just proportional to the square of the polarizability change, [Di]2 = [D/Qi]2 , but additional tensor components contribute depending on the experimental setup. For example, back-scattering Stokes Raman intensity for an isotropic sample within the harmonic approximation would be given by
1 , (14) Z k T D x, y,z E x, y,z i B where K is a constant (since absolute scattered intensities are rarely measured). VCD intensity: atomic axial tensor. To evaluate VCD intensity, it is necessary to calculate the rotational strength, which can be expressed as: (15) F0 , FX FX m R0X Im 0X m 0X Im F 0 Si 6K
7D
(i ) (i ) (i ) DE D DE D EE D DD 1 exp (i )
Zi
1
where P and m are, respectively, electric and magnetic transition dipole moments. The electric moment is evaluated as before (eqn. (10)-(12)). For the magnetic moment it is necessary to consider the dependence of the electronic wavefunction on the nuclear velocities (i.e. momenta) [20-23, 103, 104]. Expansion of the magnetic moment in the nuclear momenta, P, and retaining only the linear term leads to the expression:
1 !Z j 2 wmD A 3 A 2 (16) D F0 F1 m i 2 M A wP A M ED S AE , j 2! Z j M ED S AE , j A, E A, E E where SAE,i is defined in (8) and iM A A (17) wmD I ADE ieZ A H DJE R 0 AJ M ED A 2! wPE 0 4! is the atomic axial tensor (AAT). As for the APT (eqn. 11), MADE is separated into the j 1
electronic ( I DE ) and nuclear part, the last term in (17), where ZA, MA and R0A are the nuclear charge, mass, and equilibrium position, respectively, and HDJE is the antisymmeric (Levi-Civita) tensor. The nuclear contribution is again straightforward to evaluate, and a similar procedure as for the APT can be used to obtain the AAT
188
electronic contribution, except a magnetic field is now the perturbation. Formal derivation of the electronic part requires dependence the wavefunction on nuclear momenta, which is equivalent to inclusion of non-Born-Oppenheimer corrections. The final expression for the electronic part of the AAT contains only the nuclear position and magnetic field derivatives of the wave function [20, 103, 121, 122]:
A I ED
wMG wR A, E
wMG wB D 0
(18)
where
wM G wB D 0
M K M K mD M G
EG E K
(19)
K zG
The expression (18) corresponds to the most commonly used magnetic field perturbation (MFP) formalism for VCD, first derived by Stephens [20, 103] and independently by Buckingham and coworkers [123]. Stephens showed that a sum over states formula describing the response of MG to the nuclear motion (needed in (18)) can be avoided, without any approximations, by introducing the magnetic field perturbation [124]. The magnetic term in (18), (19) containing the gradient operator suggests that accurate wave functions might be needed to obtain useful AAT, a condition that has led to much testing of this theory for its basis set sensitivities. There are several other theoretical models of VCD [23, 105, 125-131] but these are not commonly used. ROA intensity: optical activity tensors. For ROA terms beyond the (electric) dipolar approximation must also be considered, in particular, A, the electric dipoleelectric quadrupole (4) and G, the electric dipole-magnetic dipole (the optical rotation tensor) polarizabilities: Z gn 2 (20) AD ,EJ Re PD , gn 4 EJ ,ng , 2 ! n Zng Z 2
Z 2 (21) Im PD , gn mE ,ng . 2 ! n Zng Z 2 As for ordinary Raman, the expressions for the experimentally observed ROA intensities are complex, depending on the experimental design. For example, the simple Stokes backscattering ROA intensity is given by
G 'DE
S
48K Z 1 i) (i ) (i ) (i ) (i ) (i ) DDD ZH DEJ D DG G 'DE G ' (EE AEJG / 31 exp i 3DDE c D ,E x, y , z kT Zi
1
(22)
where c is the velocity of light. The derivatives of A and G can be thought of as scattering analogues of AAT. These are normally computed using Cartesian derivatives [22, 24, 25], and a fully analytical implementation has recently appeared [25, 106]. 4.3. Computational Aspects The first non-empirical computations of molecular FF and intensities used semiempirical QM methods (e.g. CNDO or MNDO) [132, 133], and later ab initio HartreeFock (HF) approaches were developed [134-136]. Initially these used finite derivative approaches, where the energy would be recalculated for a set of small deviations from the equilibrium geometry, but now most use more efficient and accurate analytical methods, as originated by Pulay [5, 118, 119, 137, 138]. While HF methods are more accurate than semi-empirical FF, they require greater computer resources, limiting the
189
accessible molecular size. HF-level calculations are also systematically in error (frequencies too high). Higher level correlated calculations (e.g. MP2) can provide improvement of the results, but require a much greater computational cost. Fortunately, more efficient density functional theory (DFT) methods [139143] were adapted to IR, Raman and VCD calculations enabling more accurate modeling even for bigger molecules [5, 104, 122, 130, 144-147]. Accuracy approaching that of the correlated calculations could be obtained with some DFT functionals using CPU times comparable to that needed for HF calculations. However, DFT incorporates the functional as an unknown, since there is not a systematic method like the variation principle for its choice. Experience now provides guidance whereby the results obtained with some standard functionals and basis sets have been shown to be reliable, although systematic testing is still desirable [144, 145, 147-150]. DFT is an ideal method for the simulations of spectra for large biological molecules [151], where it is necessary to find a compromise between accuracy and computer resources. The following sections provide a brief overview of DFT and analytical derivative DFT theory. DFT has been reviewed extensively [140-142, 152-154], including focuses on biological systems [151], comparison of practical DFT and HF methods [143] as well as various density functional methods [144] and analytical derivative methods [5, 137, 144, 155]. Several different QM codes have been developed that incorporate DFT calculations. Perhaps the most commonly used is the Gaussian suite of programs [156], which has a long history in chemistry labs and is fairly accessible. CADPAC [157] and the Dalton suite of programs [158], are more specialized for calculations of molecular electromagnetic properties, the latter being freely available. The DFT-only Amsterdam density functional package (ADF), based on Slater atomic orbitals, computes IR and VCD intensities as well [159]. DFT Energy Computations. QM methods that solve the electronic Schrdinger equation with no empirical parameters are referred to as ab initio. Until recently these mostly used the HF approximation, where the wavefunction is replaced by a Slater determinant (antisymmetric product) of molecular orbitals. The coupled one-electron HF equations !2 2 (23) \ i V\ i H i\ i 2me were solved to self-consistency (SCF). DFT encompasses the various attempts to replace the wavefunction by the electronic density, which would also allow molecular properties to be determined "ab initio. Molecular DFT methods are based on the Kohn and Sham (KS) approximation, where a HF-type wavefunction is introduced and resultant equations resemble the HF one-particle equations (23). The only difference is in the electron-electron potential term, V, which can be a parameterized function ("functional") of the density in DFT. In Kohn-Sham (KS) DFT [139] the molecule is approximated as a system of noninteracting electrons, with independent orbitals \i for which a one body potential adjusts the independent particle density 2 (24) U \i
i
to be the same as for the real, fully interacting system. The energy of the real system is
190
T0 Vext Udr J U E XC U
(25)
where T0 is the kinetic energy of the non-interacting system, !2 * (26) T0 \ i 2\ i dr , 2me i Vext represents the external field (e.g. electron-nuclear attraction), J(U) is the Coulombic electron repulsion 1 U r1 U r2 (27) J U dr1dr2 r12 8SH 0 and EXC(U) is called exchange-correlation (XC) energy, which includes everything not contained in the first three-terms. The KS equations are obtained by variation of the energy (25) using relation (24) between the KS orbitals \i and U, yielding !2 2 !2 2 (28) \ i Vext V J V XC \i \ i V\ i H i\ i 2 me 2 me where VJ is the Coulomb potential and VXC is the exchange-correlation potential, a functional derivative of the XC energy. This yields essentially the same equation as (23), but the potential is different and contains, at least in principle, full electron correlation. Thus the KS method retains the HF simplicity, but is more powerful. Since the quality of DFT calculations depend on EXC, which is unknown, development of approximations of exchange and correlation density functionals is a key for success of DFT methods as reviewed by Becke [160]. There are three main classes of density functionals. The first, termed local density approximation (LDA, also known as LSDA for Linear Spin-Density Approximation) has the form LDA (29) E XC (r ) f ( U (r ))dr where f depends on the "local" electron density. LDA calculations often improve on HF [140, 151], but overestimate correlation and binding energies. Generalized gradient approximation (GGA) functionals correct the overbinding by addition of terms dependent on the gradient of the density, yielding functionals of the form GGA (30) E XC (r ) f ( U (r ), U (r )) dr . which are also termed gradient-corrected or non-local functionals. Combined GGA functionals have been proposed including the exchange of Becke (B) [161] and the correlation of Lee Yang and Parr (LYP) [162] or Perdew and Wang (PW91) [163, 164], which combine to yield BLYP and BPW91 functionals. PW91 [165, 166] uses no empirical parameters, therefore it is rigorously ab initio, while the Becke exchange functional contains a parameter fit to the exact exchange energies of noble gas atoms. Aside from these two "pure" DFT methods, a third class, the hybrid functionals, incorporate the Fock exchange integral
E
F XC
F X
1 8SH 0

i, j
\ i * r1 \ j * r1 \ j r2 \ i r2
r12
dr1dr2 ,
(31)
where the sum runs over all spin orbitals. The most popular of these is the Becke three-parameter hybrid functional B3LYP [167]
B 3 LYP E XC S 1 c1 E X F B S VWN LYP c1 E X c2 ( E X EX ) 1 c3 EC c3 EC
(32)
191
in which c1, c2 and c3 are constants determined by fitting experimental data, giving these methods an empirical flavor. In addition, the Fock exchange (31) is a truly nonlocal term, which significantly increases computational complexity and scaling of the hybrid methods. Nevertheless, B3LYP is widely used for vibrational analysis, due to the very good agreement obtained with it between calculated and experimental data. Basis sets. Both the HF and KS (DFT) methodologies depend on the concept of use of linear combination of atomic orbitals (LCAO) to form molecular orbitals (MO)
\ i ( MO )
c M
i k k
( AO ) .
(33)
Most quantum chemical programs use Gaussian type AOs (GTOs). Slater type AOs (STOs) allow a decrease of the number of AOs needed to attain a given precision, but are less efficient. Less commonly used basis sets are formed from purely numerical representations or plane waves [168]. In general large basis sets provide better results by imposing fewer constraints on the electron distribution at the cost of longer computational times. Incomplete AO sets may cause a significant error in computed molecular properties, therefore selection of an efficient basis set is important. Basis sets are divided into several types: Minimal basis sets contain the minimum number of basis functions for each atom, an example of which is STO-3G [169] where three Gaussian functions are used to approximate a Slater-type atomic orbital. Split valence basis sets have two or more basis functions for each valence (not core) orbital, examples of which are double zeta basis sets, such as 6-31G [170] with one orbital function made of six contracted Gaussians for heavy atom core electrons and three for H as well as three plus an extra non-contracted function for valence electrons on second row atoms. Triple zeta basis sets like 6-311G [171], use three separate functions for valence electrons. Polarized basis sets contain added functions with higher angular momentum beyond that required for the simple ground configuration of each atom. For example, 6-31G(d) (also denoted 6-31G*), adds a d-type function for heavy atoms and 6-31G(d,p) (6-31G**) has a p-type function for hydrogen. Diffuse basis sets [172] add radially larger s- and p-type functions, such as 6-31++G which contains extra 2s orbitals for hydrogen and 3p orbitals for second row atoms. Diffuse functions allow electrons more overlap with other atoms and are important for anions, atoms with lone pairs, excited states and some properties, in particular the electronic polarizability. In principle, basis sets can be expanded infinitely, eventually yielding exact results, but in practice, this is severely limited by available computational power. Generally, molecular properties computed with DFT methods show fairly rapid convergence with basis set size. Typically, valence double-zeta (or split-valence) polarized quality basis sets as, for example 6-31G(d), produce acceptable results for geometry optimization and vibrational frequencies, with some notable exceptions [151]. Nevertheless, evaluation of the influence of the basis set size on calculated properties is important. Computing analytical harmonic FF, APT and AAT. The most computationally demanding task is the DFT computation of the harmonic FF, i.e. the second derivatives of the total energy with respect to the nuclear displacements. Analytical derivative calculations require solving coupled perturbed Kohn-Sham (CPKS) equations, in analogy to the coupled perturbed HF (CPHF) theory [118, 119, 137, 173, 174]. The XC integrals cannot be computed analytically and in general must be evaluated by numerical quadrature on a grid. However, the use of cutoffs, efficient weighing schemes and grid compression can significantly reduce the computational cost [175]. In addition, more efficient evaluation of Coulombic term derivatives (27) in CPKS can be
192
obtained by use of the fast multipole method (FMM) [155, 176, 177]. As a result, the DFT analytical second derivatives scale much better with the size of the system than do the HF level analytical derivatives (CPHF equations), since the latter require evaluation of the Fock exchange terms, which cannot be improved by FMM [155]. The more favorable scaling makes DFT spectral calculations more attractive, especially for large peptides. By contrast, hybrid DFT methods, such as B3LYP (32), are computationally more expensive, since they require evaluations of both the HF exchange as well as the DFT XC quadrature. Advanced implementations of DFT promise near linear scaling for large molecules [178-180]. Improvements of the efficiency of DFT geometry optimizations and frequency calculations have been described in detail [175]. The APTs are normally calculated together with vibrational frequencies using (12) and do not add a significant overhead to the computational cost [181]. Additional considerations, however, are necessary for computations of the AAT (18) to obtain VCD intensities, since with conventional basis sets they are dependent on the choice of the coordinate origin [121]. The gauge origin problem is a direct consequence of incomplete basis sets, but can be eliminated by using magnetic field dependent atomic orbitals, known as gauge including atomic orbitals (GIAO) [182, 183]. Bak and coworkers [184-186] first used GIAO for VCD calculations, which reduces the need for large basis sets; in fact, GIAOs accelerate basis set convergence of magnetic properties [184]. Similarly, ROA intensities become origin-independent with GIAOs.
5. QM Simulations of Peptide Vibrational Spectra
5.1. Model IR and VCD Simulations Due to computational limits, ab initio calculations of IR spectra for peptides initially focused on small model amides, such as NMA [77, 187-192], and then expanded to two and three amide containing peptides [193-196]. Some recent studies still focus on this level of molecular complexity while pursuing models of ever more complex interactions. HF force fields normally overestimate the vibrational frequencies, which seems to correlate to inaccurate optimized geometries (the bond lengths are typically too short) [5]. As a remedy, scaling of ab initio FF, termed scaled quantum mechanical (SQM) FF, was proposed to best match experiment, but in turn makes it partially empirical [5, 135]. Development and implementation of DFT methods with GGA or hybrid functionals, allowed for efficient computation of much more accurate force fields [147, 197-199] as well as infrared intensities [148, 149, 200]. NMA, having a single peptide bond, has become a benchmark for amide vibrational frequency calculations [168, 187-191, 201-216]. Figure 4 shows example results from our laboratories, comparing experimental frequencies for NMA in vapor with computed values using HF, correlated and various DFT methods (Fig 4A-left) and various sized basis sets with one DFT functional (Fig 4B-right). As can be seen the pattern of improvement with regard to fitting these two experimental frequencies is not smooth, but the HF and higher correlated level results (with the same basis set) are all >100 cm-1 too high, while the DFT results with various basis sets agree better although varying non-monotonically. From comparison of various DFT methods, we and others [217] have found the BPW91 functional to perform best for simulations of the amide I and II spectra, but B3LYP and other hybrid functionals may be better for other modes. Generally, our DFT vibrational spectra simulations use BPW91 density functional.
193
Figure 4. Quantum chemical calculations of amide I (dots), amide II (squares) and amide III (triangles) vibrational frequencies contrasted with experimentally determined gas phase values (dashed lines). Left (A) Comparison of HF, higher correlated methods (CASSCF-complete active space SCF, MP2-second order Moeller-Plesset perturbation theory, QCISD quadratic configuration interaction singles, doubles, CCD-coupled clusters doubles) and several common DFT methods (LDSA linear spin-density approximation, GGA-generalized gradient approximation and hybrid HF+GGA density functionals). The calculations all use a 6-31G(d) basis set. Right: (B) Basis set dependence of density functional calculations using BPW91. Note: 6-31G(d0.3) is a 6-31G(d) basis set with a stretched d-basis function using a lower Gaussian exponent of 0.3 (default is 0.8).
Comparison of calculated spectra for small peptides with experimental data for larger peptides with defined secondary structure requires constraining the peptide backbone geometries (I\ torsions) to specific values. Determining an experimental system whose spectra can sensibly be compared with fully optimized small peptide structures is a major impediment, since real peptides fluctuate and short sequences do not usually have stable, well-defined secondary structures, computationally or experimentally, unless stabilized by specific interactions. Although small peptides have been shown to have a significant component of polyproline II (PPII or 31-helical) type structure [218-222], which stems from the local left-handed turn or extended helix nature of the random coil [219, 223], this has little impact on understanding of spectra for well-defined conformations. Recent interest in unstructured proteins and peptides [224-226] as well as development of structural tools based on coupling of inequivalent modes have increased studies of small peptide structures, but realistic modeling of solvent effects remains a challenge (vide infra). Already at the dipeptide level, the computed spectra for constrained helical or sheet conformations have characteristic features of the specific secondary structures due to the dominance of near-neighbor coupling [227]. Tripeptides give similar results, and show a clear discrimination between helix and sheet, as well as helix and coil, see
194
Figure 5 [228]; however, small peptide models lack the build-up of intensity in single delocalized modes characteristic of exciton coupling in polymers. Similar calculations for many small peptides have been carried out by other groups, several using full sidechains to relate to experimental conditions (see below), and/or full minimization to find a QM structure [26, 27, 228-234].
Figure 5: Simulated IR (top), VCD (middle) and Raman (bottom) spectra for triamides Ac-Ala2-NHCH3 in (a) D-helical, (b) 310-helical and (c) left-handed 31-helical (PolyProII-like) conformations, calculated at the DFT: BPW91/6-31G* level. Vertical lines indicate the positions and relative intensities of the normal modes. Envelopes represent Lorentzian broadened bandshapes centered on those positions which were summed to give a representative spectrum in molar units (for IR and VCD).
Development of improved computer systems, codes and DFT methods, made it possible to move beyond these small oligopeptides and directly address computations of moderately large peptide structures. For modeling D-helices, we first computed 7amide oligo-Ala peptides so that the center amide group had H-bonds forward and back, then these were extended to 10- and 11-amide structures [26]. In addition to D-helices, we have also modeled 310- and 31-helical as well as E-sheet geometries [26, 29, 112, 227, 228, 235-241]. A set of 10-mer structures were allowed to fully minimize from initial D-helix, 310-helix and ProII-helix geometries. Without solvent (vacuum) the AcAla9-NH-Me D-helical structures reverted to a 310-helical geometry, but the Ac-Pro9NH-Me did minimize to both Pro I and ProII conformations (cis and trans amide structures with right- (103) and left- (31) handed helical structures) [26]. With a solvent correction (see below), all three structures were stable. This minimization study enabled comparison of spectra calculated for different oligopeptide lengths, and also for idealized D-helices with those for fully minimized structures. As shown in Figure 6, with increasing length of the peptide, the IR intensity builds in one component and the amide I becomes dominant, which is characteristic for extended repeating structures.
195
The differences between the fully minimized and an ideal oligomer (Fig. 6c,d) are very minor. Subsequently Cheeseman and Stephens [234] have done similar full minimizations on peptides as long as 17, 20 and 25 residues, and Dannenberg up to 18 residues [233], using DFT/semiempirical ONIOM methodology [27, 242]. Raman simulations have been done for similar constrained structures (Roy et al. unpublished).
Figure 6. Simulated IR and VCD spectra for various length D-helical model oligopeptides at the DFT BPW91/6-31G(d) level: (a) idealized D-helical triamide Ac-(Ala)2-NH-CH3, (b) idealized D-helical heptaamide Ac-(Ala)6-NH-CH3, (c) idealized D-helical undecaaamide Ac-(Ala)10-NH-CH3, and (d) fully optimized decaamide Ac-(Ala)9-NH-CH3. Intensity units as in Figure 5.
Important insights into the characteristics of the vibrational spectra of the E-sheet structures were obtained from the DFT simulations of the model peptide E-sheets. In Figure 7 the simulated IR and VCD are compared for a single tripeptide E-strand, and two- and three-stranded antiparallel peptide E-sheets. While even the single strand bears the characteristic qualitative features of the Esheet IR and VCD spectra [228] (Figure 2), the cross-strand interactions and H-bonding enhance the amide I band dispersion, make the amide II region complex and normalize the relative amide I-II intensity ratio [30, 239]. The frequency shifts reflect the increase in H-bonding, since the edge residues point into vacuum but the interior ones (multistrand models) are Hbonded; this difference, however, would be less dramatic if the edges were solvated with water [239]. The VCD decreases with added strands, reflecting the delocalization of the underlying modes over the effectively more planar (and therefore less chiral), larger sheet structure. These results are even more evident in larger Esheets, whose spectra were simulated using the parameter transfer method (next section). Assuming that the polypeptide can be treated as having identical residues is computationally useful and reflects the empirical observation that all peptides and proteins have similar frequency patterns. However, when trying to match the more detailed experimental features, particularly for heterogenous structures (see below [235, 237, 238, 241, 243, 244]), the sequence becomes more important. For a series of tripeptides, AlaXxxAla, all constrained to the same geometry, we varied Xxx to see the effect on the computed frequencies (diagonal FF). Site specific spectral shifts due to the side chain can best be recognized by isotopic labeling the Xxx amide C=O group with 13C=18O to decouple it from the other C=O groups (Table 1). Beyond the variations seen here, a Pro in the sequence makes the Ala-Pro tertiary amide chemically different, shifting the frequency significantly down (while losing the amide II contribution) which can have a significant impact on the interpretation, especially of E-
196
sheet like modes [235]. Others have also investigated the effects of different sidechains on the spectra for small peptides [245, 246].
Figure 7. Simulated IR and VCD spectra for Esheet model oligopeptides at the DFT BPW91/6-31G(d) level: (a) a triamide Ac-(Ala)2-NH-CH3 single Estrand, (b) antiparallel two-stranded E-sheet (triamide strands) and (c) antiparallel three-stranded E-sheet (triamide strands). Intensity units as in Figure 5.
Table 1. Uncorrected amide I frequencies (cm-1) simulated at the BPW91/6-31G** level for tripeptides labeled on the center (*) amide with 13C=18O. AXAa PPII 310 D-helix AG*A 1667 1658 1658 AA*A 1662 1655 1651 AV*A 1655 1657 1641 * AL A 1650 1655 1637 AF*Ab 1663(F), 1616(ring) 1647(F), 1612(ring) 1649(F), 1616(ring) a Constraints for (I\) are: D-helix (-57q, -47q), PPII (-78q, +149q), 310 (-60q, -30q) [247] b Phe ring modes were separated enough to avoid coupling to the amide.
5.2 Model Raman and ROA Simulations Similar to VCD and IR, accurate force field modeling is vital for interpreting the structural information from Raman and ROA spectra, which has been exemplified by published simulations of these spectra for the polyproline II conformation [248]. Due to the variation in sensitivity to modes other than the large electric dipole transitions seen in IR and VCD, the Raman/ROA provide a complementary probe of structure. In general, this implies that sidechain modes should also be simulated. The precision of ROA simulations is rather limited for polypeptides. Two reasons for such difficulties were found in small molecule studies. For the alanine zwitterion the experimental signal can be simulated faithfully only by accounting for internal rotation of NH3+, CO2-, and CH3 groups [249]. Sensitivity to rotamer populations (effectively coupling low- and high-frequency modes, primarily an issue for side chains) requires Boltzmann averaging of an ensemble of conformations. Second, for the lower-frequency modes, the spectrum of the solvent becomes inseparable from that of the solute [250].
197
Simulations of Raman for amide modes (using Ala oligomer models) yield frequency patterns shifted from the IR bands as expected from empirical results.
5.3. Cartesian Coordinate Transfer (CCT) Method In the examples above, we have shown selected computational results using DFT methods to simulate spectra for relatively small conformationally constrained peptides as well as fully minimized structures. For very large molecules this approach is not practical with available computational resources. Consequently, we developed a method of transfer of force constants, APT, AAT and the Raman tensor derivatives from computations on smaller molecules to much larger ones [28]. This approach is based on the basic assumption of transferability of the force constants, which harks back to empirical methods, in which force constants expressed in terms of internal coordinates were transferred from various molecules onto more complex systems and optimized by fitting experimental spectra. The same approach is also widely used in developing molecular mechanics (MM) FFs in wide use today. The differences are first that we calculate not only local force constants but complete inter-residue interactions, albeit limited in range, and second that our FFs are not empirical but represent DFT results transferred to larger structures. The CCT is based on the property of regular conformations that amide units will have the same local environment in a large oligopeptide, L, with N amides, as in a smaller oligopeptide, s, with n amides, constrained to have the same conformation. FF and intensity parameters from s (e.g. those calculated at the DFT level) can be successively assigned, unit by unit, to the corresponding parts of L [28]. This is illustrated in Figure 8 where L is an extended D-helix, N = 20, and s is a blocked hexapeptide, n = 7. In this case, the nitrogen-terminal atom parameters of s can be transferred to the corresponding N-terminus of L and the C-terminal parameters of s to the L C-terminal atoms. The center residue in s is H-bonded forward and back and provides a good model for the repeating residues in the longer helix. A single calculation on s then suffices to provide data to create a nearly ab initio (DFT) quality FF for L, with the only loss being interaction constants for residues more than n positions apart in the chain, assuming the longer peptide is fully regular. The applicability of our original transfer method for oligopeptide FF, APT and AAT stems from the approximation that oligopeptides with defined secondary structure are composed of repeating connected amide units, if the differences in the side-chains are ignored. The approximation of identical side-chains is a minor limitation for most systems, but for some sequences, especially those containing Pro, it may not be sufficient, as described above [235]. For a coupled helical sequence (or coupled strands to form a sheet), the inter-residue coupling is larger than or similar to typical diagonal FF variations between residues, and these generally do not greatly disturb the overall spectral appearance and dispersion. A more restrictive assumption in this CCT model is that of repeating geometry, or translational symmetry. For systems with an irregular structure we have developed an analogous, although less efficient procedure utilizing DFT computations for a greater number of fragments [235, 238, 239, 251, 252].
198
Figure 8. Schematic of the CCT transfer idea from a helical 7-amide small peptide, Ac-(Ala)6-NHMe, to a helical 20-amide large oligopeptide, Ac-(Ala)19-NHMe. The N-terminal residue parameters transfer to the N-terminus, and C-terminal to the C-terminus, while the parameters from the center residue, H-bonded both forward and back, transfer to all the central residues that have the same properties in the large peptide.
The CCT algorithm. In practice, the CCT method uses a molecular graphics routine [253] for fitting s onto the corresponding region of L to specify the atomic overlaps for transfer of the appropriate FF, APT and AAT parameters. The FF, APT and AAT all transform as second rank Cartesian tensors under rotations, while derivatives of D and G have three and A has four indices to transform. To assign the proper parameters to the target atoms (A, B, . . .), the corresponding atoms from the best fit on the small fragment, (a, b, . . .), must be transformed to the same orientation, by minimizing 2 (34) G U R Ur where R is a column vector of coordinates of (A, B, , M) on L, r is a column vector of coordinates of (a, b, , m) on s and U is a unitary (UtU=E), 3u3 rotation matrix dependent on the Euler angles (TIF). The matrix U is then used to transform the FF: fAB=UtfabU (35) The tensors dependent on one atom only, such as APT, AAT, D, are transformed in the same manner [28]: P' A Ut Pa U (36) M' A Ut M a U
' A Ut a UU In transfers from smaller segments the transferred FF inevitably lacks the interactions between some distant atoms in the target molecule, but for sufficiently large fragments, the impact is negligible (as shown below). Other approaches that just use NMA or simple dipeptides to get diagonal and possible near-neighbor interaction constants (all normally identical) make a much bigger assumption about transferability,
199
but of course gain flexibility to apply their parameters to a variety of structures [31, 254, 255]. Torii [77] and others [73, 215, 254, 256] have developed maps of interaction terms by doing FF calculations for small peptides whose conformations were varied over a range of I\ angles. The appropriate conformational correction values for each local element can be abstracted from the maps and added to the diagonal terms (typically degenerate) to develop a more conformationally sensitive FF for larger hetero-structured molecules. In our model, the missing long-range interaction constants can be approximated, for example, by semiempirical calculation of the FF of the whole peptide onto which the more accurate ab initio FF can be subsequently transferred, or by TDC calculations of the missing interaction terms [112, 257]. Practically, we have not found this to be very useful, assuming sufficiently large fragments, e.g. about n t 5, were used for s. The combination of different FFs, APT and other parameters obtained from different sources is also enabled by the CCT. This is useful where implementation of some parameters was not available with DFT or with solvent (PCM) correction, and is useful for transfer of Raman tensors, i.e., the polarizability derivatives as well as G and A tensors for ROA. For example, even the local and non-local part of the G tensor, computed in different ways, can be combined using the same algorithm [24, 25, 250].
Figure 9. Raman (top), IR (middle) and VCD (bottom) spectra of an alanine 21-amide peptide, Ac-Ala20NH-CH3, calculated by CCT transfer of FF, APT and AAT parameters from shorter fragments having (I\) appropriate for an (a) D-helix, (b) 310-helix, (c) 31-helix. Comparison to Figures 5 and 6 shows the effect of lengthening the strand and developing extended exciton coupled transitions in these repeating structures. The large intensity in the Raman at ~1500 cm-1 is due to the CH3 groups, which are computed too high and overlap the amide II position.
Example peptide IR and VCD spectra calculated by CCT. The main strength of the CCT methodology is allowing non-empirical simulations of vibrational spectra for oligopeptides, which correspond to experimental peptide models with stable secondary structures. Such peptide sequences, typically around twenty or more amino acids, are generally much larger than can be efficiently calculated with DFT methods alone.
200
Figure 9 shows an example of the IR and VCD spectral simulations by CCT for 21mers in three different helical conformations [257]. In general transfer onto longer structures preserves the basic characteristics of the DFT simulated spectra of the smaller oligomers. However, longer peptides tend to have more intense and narrower bands, due to diminished end effects and to exciton coupling of repeating modes into polymer-like coupled modes. The amide I also gains intensity, both absolute and relative to the amide II, due to the H-bonds in the D- and 310-helices, but not in the 31helix (no H-bonds). The 31-helix results have the amide I IR weaker than the amide II, the opposite of what is typically observed for coil conformations, which are normally hydrated in solution and expected to be locally 31-helical [219, 223, 225, 258]. Proline peptides, which form regular 31-helices (poly-Pro II structure) do not have an amide II for comparison. However, when solvent effects are added to the 31 calculation, the amide I/II relative intensities are computed correctly. VCD patterns for the D-helix and 31-helix amide I closely resemble the experimental results, as do D- and 310-helical amide II predictions, but for E-sheets, while the IR is very good, the VCD agreement is more qualitative. Transfers of Raman tensors have also been carried out to model multiple transitions with some success (A. Roy et al., unpublished) [248]. The simulated 310-helix spectra in Figure 9 differ in detail from the experimental amide I VCD, which is less intense and has a conservative couplet bandshape. However, since Aib (D-amino isobutyric acid) is often used to promote formation of the 310-helix, the effects of Aib on the spectral signatures of 310-helices were tested by comparing CCT simulations for Aib2n and (Aib-Ala)n 310-helical peptide models [259]. The results showed that Aib vs. Aib-Ala peptide sequences differ in the amide I VCD bandshapes in a systematic manner that highlighted the potential accuracy of CCT DFT vibrational spectra simulation. The positive bias in the 310-helical amide I VCD for Aib2n and its amide II being sharper than for (Aib-Ala)n are both predicted in detailed agreement with the experimental spectra for 310-helices.
Figure 10. Comparison of simulated IR and VCD spectra for a three-stranded E-sheet with different lengths. The spectra were simulated by transfer of parameters from DFT calculations on three-stranded fragments with triamide strands and corresponding geometries. Anti-parallel planar E-sheet composed of three octa-amide strands, Ac-Ala7-NH-CH3, (3x8) (a) exhibits the greatest amide I mode dispersion resulting in a characteristic IR with low frequency intense maximum and a high frequency secondary maximum, and very weak VCD signal. A protein like (b) 3x8 anti-parallel E-sheet (model from fatty acid binding 1IFC) and parallel E-sheets, both (c) a planar 3x8 model and (d) a twisted 3x5 variant (Ac-Ala4NH-CH3 strands) from a protein (pectate lyase 1PEC) have qualitatively similar, but less split and more broadened amide I IR, and more intense predominantly negative VCD.
201
The CCT technique has also been used to simulate vibrational spectra for very large, regular E-sheet structures, which can model peptide aggregates or even multiplestranded E-sheets found in proteins [30, 257]. While the basic qualitative features of the E-sheet spectra are apparent even from the DFT calculations for small peptides, the dispersion of the amide I IR bands, which is characteristic of large, extended E-sheets, can be simulated only by CCT. Our calculations [30] in Figure 10, show that planar anti-parallel sheets exhibit the most distinct amide I IR bandshapes with extremely weak VCD ('A/A ~ 5x10-6), since the sheets are almost planar and therefore the C=O interactions are nearly achiral. By contrast, twisted E-sheets, such as those found in globular proteins have less pronounced amide I IR splitting and stronger, but still weak VCD ('A/A ~ 2-3x10-5). The IR amide I of parallel E-sheets, both planar and proteinlike, are similar, due to more pleating and greater deviations from planarity. An enhancement of the structural resolving power of vibrational spectroscopy in biomolecular applications is enabled by site-specific isotopic labeling [260-262]. The coupling between isotopically labeled sites, which is dependent on local structure, can be identified by vibrational spectra, as has been explored by several groups [17, 69, 263-265]. We have done a succession of studies with this technique on both helix and sheet models [29, 112, 228, 236-239, 241, 251, 266-268]. An example of direct experimental measurement of the vibrational coupling between specific sites was demonstrated using a 25 residue D-helical peptide, with two central residues labeled by 13 C substitution on the amide C=O [236]. These were compared by varying their relative separations in the sequence. These data show a reversal in the sign of the coupling constant (which shifts the IR component frequencies and flips the sign of the 13 C VCD) between the case of sequential (neighboring) labels and those separated by one residue, as shown in Figure 11, This trend is perfectly predicted by the theoretical model. Additional studies showed that the coupling drops off with separation, and that larger signals are detectable with added numbers of labels, as would be expected.
Figure 11. Experimental amide I (a) IR and (b) VCD spectra for a helical (low temperature), Ala-rich 25-residue peptide unlabeled (thin line) and with two 13C=O labels placed either adjacent (solid line) or separated (by one residue, dashed line) in the center of the sequence, and compared with the result for calculated amide I (c) IR and (d) VCD obtained using CCT to transfer parameters calculated at DFT:BPW91/6-31G** level for helical Ac-Ala10-NHMe to a 25 residue helical Ala peptide [236].
202
Theoretical modeling of the effects of isotopic labeling on E-sheet systems brought out two interesting amide I IR spectral enhancements for the intensity for the 13C=O band. First, if two (or more) labels are introduced to a strand either in sequence (nearneighbor) or separated by one residue (alternate) the intensities for the 13C=O band are drastically different [69]. With CCT DFT methods, this 13C absorbance enhancement can be seen to arise from formation of multiple-stranded antiparallel E-sheet aggregates [266]. The anomalously large intensity for the alternately labeled case arises from their being in-phase dipole oscillators in the lowest frequency, highest intensity mode. If on the other hand, single labels are on each strand, they can cross-strand couple to give unique patterns, provided they are located relatively near to each other [239, 266]. This provides a potential method of determining strand alignment and distinguishing parallel from anti-parallel structures, much as done with solid state NMR methods [269, 270]. We have compared the coupling determined with TDC model to that obtained from a full DFT calculation on 2 strands of 6 amides each. Coupling drops off with distance so that DFT results are required for close interactions, while TDC works relatively well for larger separations [239, 240]. A similar conclusion was found when a comparison was made of TDC and DFT results for helices, suggesting a hybrid approach can be useful to encompass long range interactions [236].
1690 1628 1740 1064
0.4
(rA)8*(rU)8 (Calc.)
1107 1073 1087
1705 1677 1635
0.0
1106 1084 1066 1053 1020 989 977 924 959 1121 1095 1075 1021 986 952 913
1740 1696
1124 1100
979
998
0.2
1697 1665 1636 1623
'H
960
974
poly(rA)*poly(rU) (Exp.)
1564 1523
1689 1669 1631
863
poly(rA)*poly(rU) (Exp.)
813
0 1800
1600
1569
1000
-1
860
1000
(rA)8*(rU)8 (Calc.)
1633
800
Wavenumber/cm
Figure 12. Calculated and experimental RNA spectra: (Left) VCD and IR for the (rA)8* (rU)8 duplex [271]. For the calculations, the octanucleotide duplex was simulated by use of parameters transferred from fragments including a base pair (A-U) and a sugar-phosphate dimer at the DFT(BPW91/6-31G**) level and longer range interactions represented in a duplexed pair of dinucleotides computed at the P3 level.
Nucleic acids. The methods described in this chapter are not limited to peptides and are principle applicable to other biopolymers, such as nucleic acids (NA). Computationally, NAs differ from peptides mainly by the size and nature of the chromophores (e.g. the basic chiral unit contains two base pairs, sugars and phosphate residues, which includes many more atoms than a dipeptide). Nevertheless, such systems are accessible to computational methods, as illustrated in Figure 12, where experimental spectra for an RNA double strand (duplex) octa-nucleotide are compared to simulations using DFT-level parameters from smaller NA segments [271]. Clearly, the simulations predict most of the characteristic features of the spectra, including
203
relative intensities and VCD signs. As for peptides, the C=O stretching mode in the bases dominates the IR intensities, which suggests that the TDC model would work relatively well for C=O coupling, and it does even better for PO2- (sym) coupling [33, 272-274]. Similar studies enabled interpretation of spectra for various RNA structures [271, 275, 276], non-periodic NA structures [274, 275, 277] and DNA-platinum complexes [278]. Limitations and extensions. Provided that the structures of the small fragments, whose spectra are calculated at the DFT level, and the target large peptide are identical, the CCT effectively provides DFT-level FF and intensity parameters for the target large molecule. The only approximation is that the effects of interactions between the atoms that are separated beyond the span of the small fragment are neglected. We have explored the effects of the size of the small fragment that produces converged results, and the effect of neglected long-range interactions on the simulated spectra. In D-helical studies, transfer from a tripeptide gives qualitatively correct amide I II band shapes, but misses the internal H-bond effects [228]. Use of a longer peptide fragment (or more strands for E-sheets) makes it possible to model internal H-bonds for all relevant residues and improve the amide I-II frequency separation and intensity distribution [26, 29, 112, 235]. Approximating the long range effects by semiempirical (AM1 or PM3) QM FF or TDC [112, 257] leads to discernable effects for calculations based on transfer of a triamide FF, but virtually no change in the D-helical spectral bandshapes computed by transfer from a 7-mer. For the D-helical conformation, the 7amide fragment can thus be considered to produce qualitatively converged results for use in the CCT method [257]. Furthermore, there is very little effect of increasing the size of the small fragment from a 7-mer to an 11-mer for computing the spectra of a 25mer D-helix [236, 257]. Similar behavior was found for E-sheets by comparing transfer with 2 strands of 3 residues vs. 2 of 6 residues, except that the shorter strands retain end effects on the cross-strand H-bonded rings [239]. Additional extensions are necessary for peptides of non-uniform structures. For example, in order to relate the structures of E-sheet models used in computations to experimental peptide sample conditions, a means of experimentally controlling the degree of aggregation was necessary. Consequently we have been preparing and studying various hairpin models [235, 237, 239, 241, 243 , 267], which are monomers and have a well-defined number of strands but are also non-uniform, forcing us to use an alternative to the CCT model, since the turn cannot be modeled with the strand segments. The simplest method is to assume an ideal hairpin, use two strand segments to model the strand part and a turn model for the turn, allowing enough overlap to eliminate the end (truncation) effects arising from the small oligomer [235]. If such hairpins are 13C labeled, they generate cross strand coupling which gives rise to the same sorts of spectral patterns noted above for just simple 2-strand anti-parallel sheet models. If the labels are part of a 10-member H-bonded ring, the cross-strand coupling is strong, but the lower frequency component is the more intense giving the spectra a large 13C-12C splitting in the amide I. By contrast, formation of a labeled 14-member ring gives rise to opposite sign coupling, which results in a more intense but higher frequency 13C=O band, less separate from 12C=O, as shown in Figure 14 [239, 241, 251]. For these ideal hairpins, the theoretical predictions fit the observed 13C=O effects well, but they did not reproduce the experimental 12C=O spectra since the termini are disordered in real, solvated molecules and the turn is not well described.
204
Figure 13. Comparison of (a) computed and (b) experimental amide I IR for the Gellman A hairpin, a 12-residue hairpin peptide stabilized by an Aib-Gly turn, with a RYVEVBGKKILN sequence. 13C shifted bands are labeled in both figures with L for cross-strand 13C=O H-bonds forming a 14-atom ring, labeled on positions V3-K8 (dashed line), and S for those forming a 10-atom ring, positions I10-V3 (thick solid line), compared to the unlabeled result (thin solid line) [241].
Figure 14. Amide I IR spectra of 12-amide E-hairpin models. Snapshot schematics of the backbone conformation for highly ordered, partially unfolded and unfolded structures as shown in parts a, b, and c, respectively. The corresponding simulated spectra are in parts a, b, and c, respectively. Increase in disorder is predicted to cause a shift to increased wave number and a broadening of features, much as seen experimentally with increase in temperature [251].
To overcome the limitations of ideal structures, we developed a fragmentation method which allows us to compute spectra for each segment of the target peptide, based for example on structures derived from NMR analyses or MD calculations, and to transfer their parameters onto the full peptide [238, 239, 251]. This approach is limited by the need for the fragments to overlap so that effects of the truncation can be eliminated, but makes it possible to explore the variations of the spectra that
205
accompany dynamic fluctuations of the peptide, such as in the ensemble of structures that can be represented by an MD trajectory. In Figure 14 are shown three example structures (sampled from a 450K, ~10 ns, MD trajectory) for a 12-residue hairpin fully folded (a), partially unfolded (b, frayed ends) and fully unfolded (c) [241, 251]. The spectra show a gradual broadening and shift of the main amide I absorbance. This is consistent with what happens when the molecule is heated and undergoes a phase transition to the unfolded state, but it is also representative of the low temperature ensemble. The differences suggest that following a phase transition with IR will not give a two state behavior since the various intermediates will have temperature dependent populations and thus the spectrum will shift continuously, a behavior characteristic of a number of hairpins [235, 237, 241, 267, 268, 279-282]. 5.4. Incorporating Solvent into Spectral Simulations For biological relevance, peptide and protein spectra are normally measured in aqueous solutions. Solvent has a profound effect on the amide frequencies, in particular on the amide I, as illustrated in Figure 15 (left) for experimental NMA IR spectra in the gas phase and in acetonitrile and aqueous solutions [207, 209, 215, 216, 283-285]. Accurate theoretical modeling of biomolecular systems thus cannot ignore the effect of solvent. The most rigorous, but also computationally expensive, correction is to include added solvent molecules explicitly with the peptide in the computational model. Continuum solvent models, including the Onsager and Polarized Continuum Models (PCM) are less expensive. Both are implemented in DFT-based methods in QM programs and are referred to as Self Consistent Reaction Fields (SCRF). An alternate approach combines MD simulations of the fluctuating solvent geometry and empirical electrostatic correction of the spectra [210, 212, 213, 215, 246, 285, 286].
Figure 15. Comparison of (left) experimental IR spectra for NMA in (a) gas phase and in (b) acetonitrile and (c) aqueous solution (dash line is D2O) with (right) simulated amide I and II vibrational frequencies in water as obtained with explicit and implicit solvent corrections. The amide I frequencies shift lower in going from (a) the gas phase spectrum to (c) water by ~ 100 cm-1,while the amide II (only in H2O) shifts ~ 85 cm-1 higher. As seen in the comparisons (right) of theoretical models, use of explicit hydrogen-bonded water has the greatest effect on the amide I and II frequencies; however, additionally including a continuum solvent model in the NMA-water cluster significantly improves the predictions.
206
Examples using explicit solvent. To achieve the best correction of the amide I and II frequencies, solvent molecules should be explicitly included at some level in the calculation. This has been used for modeling spectra of NMA, [188, 206, 212, 287290] amino acids and dipeptides [229, 230, 291-295]. Results of DFT frequency calculations for NMA [209] with explicit as well as implicit solvent models Figure 15 (right) demonstrate that much better amide I-II frequencies can be calculated for aqueous solutions if hydrogen bonded molecules are represented explicitly (using two waters on C=O and one on N-H) [204, 296]. Of course such models must assume a structure for the solvent molecules, which actually have a high degree of dynamic variation. Thus while frequencies improve, other aspects, such as chiral properties, can be distorted unless averaging over the ensemble or other corrections are used. While added improvement can be obtained by inclusion of added layers of water [213], their structures are even more fluxional, and a simpler approximation is obtained by implicitly including their solvent effect by means of PCM models [209]. A similar level of explicit solvent correction was applied to an alanine tripeptide to investigate the effects of deuteration on the Amide I VCD bandshape, [112] as well as for calculations of fully solvated D- (7-mer), 310- and 31- (5-mers) helices [297]. The resulting FF parameters were subsequently transferred to 21-mer structures with the same helical conformation using the CCT approach [297]. The smaller oligomers show increased dispersion in the amide I and II bands due to interaction with the water molecules and have dramatic amide I frequency shifts, due to H-bonding, as expected. For the longer peptides, exciton coupling dominates the local dispersion and similar patterns, as seen for the vacuum calculations, result at the shifted frequencies, as can be seen in the selected examples in Figure 16.
Figure 16. Simulations of IR (top) and VCD (bottom) spectra for helical peptides with explicit solvent. DFT calculations were made for an D-helical hepta-amide, Ac-Ala6-NH-CH3, and 310 helical and 31 helical pentaamides, Ac-Ala4-NH-CH3 with solvent modeled by hydrogen-bonded water molecules on each intrapeptide hydrogen bonded amide C=O and a non-hydrogen bonded N-H group, and two water molecules forming hydrogen bonds to the terminal, non-intrapeptide hydrogen bonded C=O groups. These parameters were transferred onto helical 21-amide models, Ac-Ala20-NH-CH3, with the same (I\) angles, and the results directly compare to the vacuum calculations in Figure 9. The predominant effect of the solvent is shifts in the amide I and II vibrational frequencies. An increase in the IR intensity is also predicted, especially for the amide I, whose relative intensity with respect to amide II now more closely mimics experiment. The VCD bandshapes remain unchanged with solvent as compared to vacuum, with the exception of the amide II for 31 helix and small variations in the amide I of the 310 helix.
207
Figure 17. Comparison of simulated D-helical 15-amide, Ac-Ala14-NH-CH3, IR and VCD amide I for unsolvated (gas phase), partially solvated (mimicking an D-helix in a protein) and fully solvated (mimicking solvated model D-helical) peptide models. The gas phase (top) and fully solvated (bottom) peptides exhibit the same spectral bandshapes, both for IR and VCD, with the solvated amide I shifted to lower frequency. The partially solvated peptide (middle) has an intermediate amide I frequency, but also exhibits additional broadening, the higher ones representing an extreme of what might correspond to the interior (desolvated) residues in a protein. The amide I VCD preserves the D-helical shape and is dominated by the signal from the higher frequency modes, with a second D-helix like couplet predicted, corresponding to the solvated groups. The IR simulations are in qualitative agreement with the solvated and buried amide components often observed in protein IR. However, the simulated results overestimate the effects of hydrogen-bonded water and better correspond to the cryogenic experiments [298] rather than room temperature spectra, where the dynamic nature of the solvent must be taken into account.
While the overall spectral bandshape patterns do not change significantly, inclusion of solvent in the calculations leads to better quantitative agreement with experiment in the calculated spectral intensities, and even in some qualitative details of the spectral dispersion. In the IR, the amide I for the solvated 31-helix is predicted to be more intense than amide II, in agreement with condensed phase results, while in the gas phase (Fig. 9) the relative amide I and II intensities are reversed, which has not been seen experimentally. In VCD, the amide I band shape change upon N-deuteration from a (-,+) couplet to a (-,+,-) shape for an D-helix is correctly reproduced only when solvent is included [213, 297]. Much better quantitative agreement of the calculated 13C IR and VCD intensities with experiments for isotopically labeled alanine-rich peptides [29] is obtained by recalculating the spectra using CCT to transfer hydrated representations of the local helical sequence [297]. The solvent correction is therefore useful not only for predicting correct frequencies, but also for more subtle features of the normal mode distributions and spectral intensities. On the other hand, the overall qualitative pattern seen in the vacuum calculations persists, so the need to correct for solvent depends on the detail one wishes to derive from the spectral data. The corrections are often fairly predictable, so empirical methods might provide useful substitutes for computationally intense DFT level calculations of these solvent effects. The effect of partial solvation on the D-helices has also been studied [252] to mimic the environment in proteins. Partially solvated D-helices were constructed using
208
a fragmentation method, similar to that discussed above, employing parameter transfer from several short helical fragments solvated by explicit water at different positions. A comparison of simulated gas phase, partially- and fully-solvated D-helix IR amide I spectra is shown in Figure 17. These computations predicted significant broadening and splitting of the amide I into the solvated and unsolvated parts, in agreement with cryogenic experiments [298], but showed that explicit hydrogen-bonded water overcorrects for the solvent effects, when compared to vacuum, but this does not fully account for the effect of the protein environment on the non-solvated side of the helix. In order to reproduce the spectral bandshapes at room temperature, the dynamical nature of the solvent might be taken into account, for example by the electrostatic solvent correction [210, 211, 213, 246]. Continuum solvent models. In continuum models, the electrostatic effects of bulk solvent are represented by a continuous dielectric medium outside a cavity occupied by a solute molecule [299]. This ignores much molecular detail, while assuming a linear response, but it provides a computationally tractable approximation of the electrostatic effects of the solvent, as has been extensively reviewed [300, 301]. The Onsager SCRF model [302] places the solute molecule at the center of a spherical cavity of an appropriate radius, and the solute charge distribution is reduced to just the dipole moment. This allows the solute-solvent interaction to be expressed in a simple closed form, which is robust and computationally efficient [303]. However, the shape restriction of the Onsager cavity to a sphere introduces artifacts, and only the resulting dipole moment contributes to the reaction field, reflecting, for example, the helix macrodipole but not those of individual polar groups. Polarized continuum models (PCM), developed following Tomasi and coworkers [304], provide more realistic cavity shapes. Another, very efficient approach called "COSMO" was pioneered by Klamt [305], in which the polarized continuum outside the solute cavity is described using electrostatic field boundary conditions of a conductor. Later, this model was adapted so that it allowed for an arbitrary solvent permittivity [306-308] and has been implemented within the PCM framework [309]. In the Gaussian programs this model is referred to as CPCM. The PCM cavity is realized as a sum of interlocking spheres centered on each nucleus with the appropriate atomic radii [300]. For practical calculations the cavity surface is partitioned into small, planar, triangular domains, called tesserae, whose contributions summed give the reaction field energy. CPCM adds only a small computational cost to both energies and analytical gradients [309] and turns out to be somewhat faster than dielectric-based PCM models. CPCM is thus in principle more efficient for geometry optimizations, particularly for large molecules. Implementation of analytical second derivatives within (C)PCM opened new possibilities for obtaining more realistic peptide force fields corrected for the environment [310]. However, practical calculations with currently available PCM and CPCM methods do have some stability problems [300]. In particular, geometry optimizations can converge very slowly and sometimes do not converge at all within the default convergence criteria. Electrostatic solvent correction. A computationally cheaper alternative for explicit solvent correction was first proposed by Cho and coworkers [211] where an empirical frequency correction, due to the solvent induced electric field, is applied to the ab initio amide I frequencies (Z0) obtained from a vacuum calculation. In the original
209
formulation [211, 311] the amide I stretching frequency Z (in solvent) linearly relates to the electrostatic solvent potential Mi, as
Z Z0 biM i
k 1
(37)
measured at the amide group atoms (CNHCOC). (In another chapter, Choi and Cho provide more detail [34].) The correction can be generalized to any chromophore (any number of atoms) and applied to the intensity tensors [210, 287]. The coefficients bi can be obtained by a fit to ab initio computations involving explicit solvent molecules. For intensity parameters, it is more convenient to express the fitting coefficients in a local coordinate system, yielding the corrected atomic polar tensor, for example, as
O PDE O PDE
( 0)
bi ,OD , E M i ,
i 1
(38)
where indices follow those used in (11). We have developed a set of bi parameters for NMA in water clusters by referencing to DFT computed FF results [210, 287]. The approach is especially suitable to combined QM/MM modeling, as the correction is fast and allows for averaging of large ensembles of geometries. The electrostatic correction can reproduce ab initio results; for example the changes observed in D-helical peptide spectra under deuteration are correctly predicted [213]. Furthermore, the complicated IR and VCD spectral patterns of E-hairpins can be better understood in terms of the electrostatic interactions between the solvent and the amide groups, or in terms of shielding of some groups by the side chains [238].
6. Concluding Remarks
The examples given above demonstrate that it is now fully possible to compute vibrational spectra for moderately sized peptides to an experimentally useful degree of accuracy using a method free of empirical parameterization, except for the fairly universal DFT functionals. These developments reflect continuing advances in quantum chemical computational methodology, particularly with DFT based methods, and in widely available computer capabilities. The analyses have gone beyond frequency correlation to address apparent band shapes due to the dispersion of intensity in the exciton coupled modes characteristic of repeating molecular structural units in such polymeric species. Modern developments are permitting computation to address the effects of solvation and structural fluctuation in ways previously viewed as unachievable. These developments have made computational simulation an integral part of experimental vibrational spectroscopy. The ab initio approach to simulations of peptide vibrational spectra has been criticized [19, 70] for not providing physical insight into the nature of amide vibrational coupling. However, reproducing the experimental spectra using an assumed coupling mechanism, e.g. through-bond and through space, such as TDC, with adjustable parameters can provide a rather misleading picture. The through-bond and through-space schemes in coupled oscillator approaches are inseparable. The amide vibrations are coupled through the electronic structure, which is only approximately separated into such conceptually attractive components. On the contrary, by virtue of not relying on empirically (spectrally) derived parameters, quantum mechanical simulations of the spectra provide important physical insight. For example consider the
210
vibrational frequencies of peptides in solution: while in coupled oscillator schemes the unperturbed frequency of the amide modes is adjusted to fit the data, the DFT calculations clearly show that the solvent is necessary to reproduce the vibrational frequencies measured in solution, while the vacuum, isolated molecule calculations, not surprisingly, can reproduce gas phase frequency values [209, 216]. Quantum mechanics in fact provides the means to test rigorously the validity of approximations for inter-amide coupling. Since the DFT calculations yield transition dipole moments, it is always possible to test, for example, to what extent the transition dipole coupling contributes to observed spectra. As we have demonstrated, TDC alone is not sufficient to account for observed bandshapes in helical or E-sheet peptides [240]. TDC becomes reasonable approximation for long-range coupling between residues that are far separated in sequence. The coupled oscillator must make up for this deficiency by using much larger transition dipoles than the more accurate values calculated by DFT (which actually replicate the overall vibrational intensity). The advantages of DFT calculations as compared to TDC were also verified experimentally, via the doubly 13C labeled D-helical peptide results, which allow direct measures of the vibrational coupling between two amides. Varying the distance in the sequence between the two labels demonstrates the variation of the magnitude and also the sign of the vibrational coupling. DFT calculations provide quantitative explanations of both IR and VCD patterns, while TDC is a good approximation only for distant oscillators, where the coupling essentially approaches zero [236]. These summary comments are meant to illuminate the importance of accurate, non-empirical calculations of the polypeptide force fields for correct physical understanding of the interactions that govern the peptide and protein spectral features. Certainly there is a role for theory at multiple levels, but now that the DFT approaches have become accessible and that our CCT and related transfer or localized methods permit high level modeling of larger structures, it is important to do systematic comparisons with the simpler methods before relying on them to determine the finer details of spectra and structure that are now being derived from careful peptide and protein vibrational spectroscopic studies.
Acknowledgement. The work at UIC is currently sponsored by the National Science Foundation (CHE03-16014 and 07-18543 to TAK), that at the CAS by the Czech Science Foundation (grants Nos. 203/06/0420, 202/07/0732 to PB) and the Grant Agency (A4005507020) and at UW by the Faculty Grant-in-Aid and Basic Research Grant programs of the University of Wyoming (to JK). We thank Ahmed Lakahani, George Papadantonakis, Anjan Roy, Rong Huang, Heng Chi and Ling Wu for unpublished results and help with assembling references and figures.
References
[1] P.I. Haris, Fourier Transform Infrared Spectroscopic Studies of Peptides: Potentials and Pitfalls. In: Infrared Analysis of Peptides and Proteins: Principles and Applications. ACS Symposium Series. B. Ram Singh (Ed.), ACS, Washington DC, 2000, 54-95. [2] A. Barth, Infrared spectroscopy of proteins, Biochim. Biophys. Acta. 1767 (2007) 1073-1101. [3] A.T. Tu, Raman Spectroscopy in Biology, Wiley, New York, 1982. [4] R.W. Williams, Protein secondary structure analysis using Raman amide I and amide III spectra, Methods Enzymol. 130 (1986) 311-331.
211
[5] P. Pulay, Analytical derivative techniques and the calculation of vibrational spectra. In: Modern electronic structure theory D.R. Yarkony (Ed.), Vol. 2, World Scientific, Singapore, 1995, 1191-1240. [6] T.A. Keiderling, Peptide and protein conformational studies with vibrational circular dichroism and related spectroscopies. In: Circular Dichroism: Principles and Applications, 2nd Ed., N. Berova, K. Nakanishi and R.W. Woody (Eds.), Wiley-VCH, New York, 2000, 621-666. [7] T.A. Keiderling, Protein and peptide secondary structure and conformational determination with vibrational circular dichroism, Curr. Opin. Chem. Biol. 6 (2002) 682-688. [8] L.D. Barron and L. Hecht, Vibrational Raman optical activity: From fundamentals to biochemical applications. In: Circular dichroism, principles and applications K. Nakanishi, N. Berova and R.W. Woody (Eds.), Wiley-VCH, New York, 2000, 667-701. [9] H. Fabian and D. Naumann, Methods to study protein folding by stopped-flow FT-IR, Methods. 34 (2004) 28-40. [10] R. Vogel and F. Siebert, Vibrational spectroscopy as a tool for probing protein function, Curr. Opin. Chem. Biol. 4 (2000) 518-523. [11] M.S. Braiman and Y.W. Xiao, Step-scan time-resolved FT-IR spectroscopy of biopolymers. In: Vibrational Spectroscopy of Biological and Polymeric Materials V.G. Gregoriou and M.S. Braiman (Eds.), CRC Press, 2005, 353-419. [12] C. Kotting and K. Gerwert, Proteins in action monitored by time-resolved FTIR spectroscopy, ChemPhysChem. 6 (2005) 881-888. [13]R.B. Dyer, F. Gai and W.H. Woodruff, Infrared studies of fast events in protein folding, Acc. Chem. Res. 31 (1998) 709-716. [14] R. Callender and R.B. Dyer, Advances in time-resolved approaches to characterize the dynamical nature of enzymatic catalysis, Chem. Rev. 106 (2006) 3031-3042. [15] S. Woutersen and P. Hamm, Nonlinear two-dimensional vibrational spectroscopy of peptides, J. Phys. Condens. Matter. 14 (2002) R1035-1062. [16] H.S. Chung, Z. Ganim, K.C. Jones and A. Tokmakoff, Transient 2D IR spectroscopy of ubiquitin unfolding dynamics, Proc. Natl. Acad. Sci. U. S. A. 104 (2007) 14237-14242. [17] S.M. Decatur, Elucidation of Residue-Level Structure and Dynamics of Polypeptides via Isotope-Edited Infrared Spectroscopy, Acc. Chem. Res. 39 (2006) 169-175. [18] S. Krimm, Interpreting Infrared Spectra of Peptides and Proteins. In: Infrared Analysis of Peptides and Proteins: Principles and Applications. ACS Symposium Series. B.R. Singh (Ed.), ACS, Washington DC, 2000, 38-53. [19] R. Schweitzer-Stenner, Advances in vibrational spectroscopy as a sensitive probe of peptide and protein structure - A critical review, Vibr. Spectrosc. 42 (2006) 98-117. [20] P.J. Stephens, Theory of vibrational circular dichroism, J. Phys. Chem. 89 (1985) 748-752. [21] D. Yang and A.J. Rauk, The a priori calculation of vibrational circular dichroism intensities. In: Reviews in Computational Chemistry K.B. Lipkowitz and D.B. Boyd (Eds.), Vol. 7, VCH Publishers, Inc., New York, 1996, 261-301. [22] P.L. Polavarapu, Vibrational spectra: principles and applications with emphasis on optical activity, Vol. 85, Elsevier, Amsterdam, 1998. [23] L.A. Nafie and T.B. Freedman, Vibrational optical activity theory. In: Circular Dichroism. Principles and Applications. 2nd edition. N. Berova, K. Nakanishi and R.W. Woody (Eds.), Wiley-VCH, New York, 2000. [24]P. Bou, Computations of the Raman optical activity via the sum-over-states expansions, J. Comp. Chem. 22 (2001) 426-435. [25] K. Ruud, T. Helgaker and P. Bou, Gauge-origin independent density-functional theory calculations of vibrational Raman optical activity, J. Phys. Chem. A. 106 (2002) 7448-7455. [26] P. Bou, J. Kubelka and T.A. Keiderling, Quantum Mechanical Models of Peptide Helices and Their Vibrational Spectra, Biopolymers. 65 (2002) 45-69. [27] R. Wieczorek and J.J. Dannenberg, Amide I vibrational frequencies of alpha-helical peptides based upon ONIOM and density functional theory (DFT) studies, J. Phys. Chem. B. 112 (2008) 1320-1328. [28] P. Bou, J. Sopkov, L. Bednrov, P. Malo and T.A. Keiderling, Transfer of molecular property tensors in Cartesian coordinates: A new algorithm for simulation of vibrational spectra, J. Comput. Chem. 18 (1997) 646-659.
212
[29] R.A.G.D. Silva, J. Kubelka, S.M. Decatur, P. Bou and T.A. Keiderling, Site-specific conformational determination in thermal unfolding studies of helical peptides using vibrational circular dichroism with isotopic substitution., Proc. Natl. Acad. Sci. U. S. A. 97 (2000) 8318-8323. [30] J. Kubelka and T.A. Keiderling, Differentiation of E-sheet forming structures: ab initio based simulations of IR absorption and vibrational CD for model peptide and protein E-sheets., J. Am. Chem. Soc. 123 (2001) 12048-12058. [31] J.-H. Choi, J.-S. Kim and M. Cho, Amide I vibrational circular dichroism of polypeptides: Generalized fragmentation approximation method, J. Chem. Phys. 122 (2005) 174903-174913. [32] J.H. Choi, H. Lee, K.K. Lee, S. Hahn and M. Cho, Computational spectroscopy of ubiquitin: Comparison between theory and experiments, J. Chem. Phys. 126 (2007). [33] V. Andrushchenko, H. Wieser and P. Bou, B-Z conformational transition of nucleic acids monitored by vibrational circular dichroism. Ab Initio Interpretation of the Experiment, J. Phys. Chem. B. 106 (2002) 12623-12634. [34] J.H. Choi and M. Cho. In: FTIR Spectroscopy in Biomedical Applications A. Barth and P.I. Haris (Eds.), 2008. [35] A. Elliott and E.J. Ambrose, Structure of synthetic polypeptides, Nature. 165 (1950) 921-922. [36] T. Miyazawa, Infrared Spectra and Helical Conformations. In: Poly-D-Amino Acids: Proteins Models for Conformational Analysis G.D. Fasman (Ed.), Dekker, New York, 1967, 69-103. [37] R.D.B. Fraser and T.P. MacRae, Infrared Spectroscopy. In: Conformation in Fibrous Proteins and related synthetic polypeptides B. Horecker, N.O. Kaplan, J. Marmur and H.A. Scheraga (Eds.), Academic Press, New York, 1973, 94-123. [38] D.M. Byler and H. Susi, Examination of the secondary structure of proteins by deconvolved FTIR spectra, Biopolymers. 25 (1986) 469-487. [39] P.I. Haris and D. Chapman, The conformational analysis of peptides using Fourier Transform IR spectroscopy, Biopolymers. 37 (1995) 251-263. [40] E.J. Ambrose and A. Elliott, The structure of synthetic polypeptides. II. Investigation with polarized infra-red spectroscopy, Proc. Royal Soc. A205 (1951) 47-60. [41] E.R. Blout, C. De Loze and A. Asadourian, The deuterium exchange of water-soluble polypeptides and proteins as measured by infrared spectroscopy, J. Am. Chem. Soc. 83 (1961) 1895-1900. [42] A.C. Sen and T.A. Keiderling, Vibrational circular dichroism of polypeptides. III. Film studies of several alpha-helical and beta-sheet polypeptides, Biopolymers. 23 (1984) 1533-1545. [43] V.P. Gupta and T.A. Keiderling, Vibrational CD of the amide II band in some model polypeptides and proteins, Biopolymers. 32 (1992) 239-248. [44] K. Kaiden, T. Matsui and S. Tanaka, A Study of the amide III band by FT-IR spectrometry of the secondary structure of albumin, myoglobin and gamma-globulin, Appl. Spectrosc. 41 (1987) 180-184. [45] B.R. Singh, M.P. Fuller and G. Schiavo, Molecular structure of tetanus neurotoxin as revealed by FT-IR and CD spectroscopy, Biophys. Chem. 46 (1990) 155-166. [46] F. Fu, D.B. DeOliveira, W.R. Trumble, H.K. Sarkar and B.R. Singh, Secondary structure estimation of proteins using the amide III region of Fourier transform infrared spectroscopy: Application to analyze calcium-binding-induced structural changes in calsequestrin, Appl. Spectrosc. 48 (1994) 1432-1440. [47] B.I. Baello, P. Panoka and T.A. Keiderling, Vibrational circular dichroism spectra of proteins in the amide III region. Measurement and correlation of bandshape to secondary structure, Anal. Biochem. 250 (1997) 212-221. [48] S. Cai and B.R. Singh, Determination of the secondary structrue of proteins from amide I and amide III infrared bands using partial least square method. In: Infrared Analysis of Peptides and Proteins: Principles and Applications. ACS Symposium Series. B.R. Singh (Ed.), ACS, Washington DC, 2000, 117-129. [49] R. Tuma, Raman spectroscopy of proteins: from peptides to large assemblies, J. Raman Spectrosc. 36 (2005) 307-319. [50] W. Qian and S. Krimm, Vibrational studies of the disulfide group in proteins. VI. General correlations of SS and CS stretch frequencies with disulfide bridge geometry, Biopolymers. 32 (1992) 1025-1033. [51] R.P. Rava and T.G. Spiro, Resonance enhancement in the ultraviolet Raman spectra of aromatic amino acids, J. Phy. Chem. 89 (1985) 1856-1861.
213
[52] T.G. Spiro (Ed.), Resonance Raman Spectra of Polyenes and Aromatics, Vol. 2, Wiley, New York, 1987. [53] X.J. Zhao and T.G. Spiro, Ultraviolet resonance raman spectroscopy of hemoglobin with 200 and 212nm excitation: H-bonds of tyrosines and prolines, J. Raman Spectrosc. 29 (1998) 49-55. [54] S. Song, S.A. Asher and S. Krimm, Assignment of a new conformation-sensitive UV resonance Raman band in peptides and proteins, J. Am. Chem. Soc. 110 (1988) 8547-8548. [55] Z.H. Chi and S.A. Asher, UV Resonance Raman determination of protein acid denaturation - selective unfolding of helical segments of horse myoglobin, Biochemistry. 37 (1998) 2865-2872. [56] I.K. Lednev, A.S. Karnoup, M.C. Sparrow and S.A. Asher, Alpha-helix peptide folding and unfolding activation barriers: A nanosecond UV resonance raman study . J. Am. Chem. Soc. 121 (1999) 8074-8086. [57] S.A. Asher, A. Ianoul, G. Mix, M.N. Boyden, A. Karnoup, M. Diem and R. Schweitzer-Stenner, Dihedral psi angle dependence of the amide III vibration: a uniquely sensitive UV resonance Raman secondary structural probe., J. Am. Chem. Soc. 123 (2001) 11775-11781. [58] T. Miyazawa, Perturbation treatment of the characteristic vibrations of polypeptide chains in various conformations, J. Chem. Phys. 32 (1960) 1647-1652. [59] T. Miyazawa and E.R. Blout, The infrared spectra of polypeptides in various conformations: amide I and II bands, J. Am. Chem. Soc. 83 (1961) 712-719. [60] P.W. Higgs, The vibration spectra of helical molecules: infra-red and Raman selection rules, intensities and approximate frequencies, Proc. R. Soc. London. A220 (1953) 472-485. [61] Y.N. Chirgadze and N.A. Nevskaya, Infrared Spectra and Resonance Interaction of Amide I Vibration of the Antiparallel-Chain Pleated Sheet, Biopolymers. 15 (1976) 607-625. [62] Y.N. Chirgadze and N.A. Nevskaya, Infrared spectra and resonance interaction of amide-I vibration of the parallel-chain pleated sheet, Biopolymers. 15 (1976) 627-636. [63] N.A. Nevskaya and Y.N. Chirgadze, Infrared spectra and resonance interactions of Amide-I and II vibrations of D-helix, Biopolymers. 15 (1976) 637-648. [64] H. Torii and M. Tasumi, Application of the three-dimensional doorway-state theory to analyses of the amide I infrared bands of globular proteins, J. Chem. Phys. 97 (1992) 92-98. [65] H. Torii and M. Tasumi, Model calculations on the amide-I infrared bands of globular proteins, J. Chem. Phys. 96 (1992) 3379-3387. [66] H. Torii and M. Tasumi, Theoretical analyses of the amide I infrared bands of globular proteins. In: Infrared spectroscopy of biomolecules H.H. Mantsch and D. Chapman (Eds.), Wiley-Liss, Chichester UK, 1996, 1-17. [67] S.S. Birke, I. Agbaje and M. Diem, Experimental and Computational Infrared CD Studies of Prototypical Peptide Conformations, Biochemistry. 31 (1992) 450-455. [68] T. Xiang, D.J. Goss and M. Diem, Strategies for the computation of infrared CD and absorption spectra of biological molecules: ribonucleic acids, Biophys. J. 65 (1993) 1255-1261. [69] J.W. Brauner, C. Dugan and R. Mendelsohn, 13C Isotope labeling of hydrophobic peptides. Origin of the anomalous intensity distribution in the infrared amide I spectral region of beta-sheet., J. Am. Chem. Soc. 122 (2000) 677-683. [70] J.W. Brauner, C.R. Flach and R. Mendelsohn, Quantitative reconstruction of the amide I contour in the IR spectra of globular proteins: From structure to spectrum, J. Am. Chem. Soc. 127 (2005) 100-109. [71] S. Woutersen and P. Hamm, Structure determination of trialanine in water using polarization sensitive two-dimensional vibrational spectroscopy, J. Phys. Chem. B. 104 (2000) 11316-11320. [72] R. Schweitzer-Stenner, Secondary structure analysis of polypeptides based on an excitonic coupling model to describe the band profile of amide I ' of IR, Raman, and vibrational circular dichroism spectra J. Phys. Chem. B. 108 (2004) 16965-16975. [73] J. Wang and R.M. Hochstrasser, Characteristics of the two-dimensional infrared spectroscopy of helices from approximate simulations and analytic models, Chem. Phys. 297 (2004) 195-219. [74] S. Krimm and J. Bandekar, Vibrational spectroscopy and conformation of peptides, polypeptides and proteins, Adv. Protein Chem. 38 (1986) 181-364. [75] W.H. Moore and S. Krimm, Transition dipole coupling in amide I modes of E polypeptides., Proc. Natl. Acad. Sci. U.S.A. 72 (1975) 4933-4935. [76] S.H. Lee and S. Krimm, Ab initio-based vibrational analysis of alpha-poly(L-alanine), Biopolymers. 46 (1998) 283-317.
214
[77] H. Torii and M. Tasumi, Ab initio molecular orbital study of the amide I vibrational interactions between the peptide groups in di- and tripeptides and considerations on the conformation of the extended helix, J. Raman Spectrosc. 29 (1998) 81-86. [78] C. Fang, J. Wang, A.K. Charnley, W. Barber-Armstrong, A.B. Smith III, S.M. Decatur and R.M. Hochstrasser, Two-dimensional infrared measurements of the coupling between amide modes of an alpha-helix, Chem. Phys. Lett. 382 (2003) 586-592. [79] R. Schweitzer-Stenner, F. Eker, K. Griebenow, C. Xiaolin and L.A. Nafie, The conformation of tetraalanine in water determined by polarized Raman, FT-IR, and VCD spectroscopy, J. Am. Chem. Soc. 126 (2004) 2768 - 2776 [80] T. Miyazawa, Characteristic amide bands and conformations of polypeptides. In: M.A. Stahmann (Ed.), Polyamino Acids, Polypeptides and Proteins: International Symposium, University of Wisconsin, Madison, WI, 1962, pp. 201-217. [81] Y.N. Chirgadze, B.V. Shetopalov and S.Y. Venyaminov, Intensities and other spectral parameters of infrared amide bands of polypeptides in the beta and random forms, Biopolymers. 12 (1973) 1337-1351. [82] H. Torii and M. Tasumi, Three-dimensional doorway-state theory for analyses of absorption bands of many-oscillator systems, J. Chem. Phys. 97 (1992) 86-91. [83] H. Torii and M. Tasumi, Infrared intensities of vibrational modes of an alpha-helical polypeptide. Calculations based on the equilibrium charge-charge flux (ECCF) model, J. Mol. Struct. 300 (1993) 171179. [84] K. Palmo and S. Krimm, Electrostatic model for infrared intensities in a spectroscopically determined molecular mechanics force field., J. Comp. Chem. 19 (1998) 754-768. [85] D.A. Long, Intensities in Raman spectra . 1.a Bond polarizability theory, Proceedings of the Royal Society of London Series a-Mathematical and Physical Sciences. 217 (1953) 203-221. [86] L.D. Barron, J.R. Escribano and J.F. Torrance, Polarized Raman optical-activity and the bond polarizability model, Mol. Phys. 57 (1986) 653-660. [87] L.D. Barron, Molecular Light Scattering and Optical Activity, Cambridge University Press, Cambridge UK, 2004. [88] L. Wirtz, M. Lazzeri, F. Mauri and A. Rubio, Raman spectra of BN nanotubes: Ab initio and bondpolarizability model calculations, Physical Review B. 71 (2005). [89] G. Holzwarth and I. Chabay, Optical activity of vibrational transitions: A coupled oscillator model, J. Chem. Phys. 57 (1972) 1632. [90] I. Tinoco, Radiation Res. 20 (1963) 133. [91] J.A. Schellman, Vibrational optical activity, J. Chem. Phys. 58 (1973) 2882-2886. [92] J. Snir, R.A. Frankel and J.A. Schellman, Optical activity of polypeptides in the infrared. Predicted CD of the amide I and amide II bands, Biopolymers. 14 (1975) 173-196. [93] M. Gulotta, D.J. Goss and M. Diem, IR vibrational CD in model deoxyoligonucleotides: Observation of the B - > Z phase transition and extended coupled oscillator intensity calculations., Biopolymers. 28 (1989) 2047-2058. [94] W. Zhong, M. Gulotta, D.J. Goss and M. Diem, DNA solution conformation via infrared circular dichroism: Experimental and Theoretical Results for B-Family Polymers, Biochemistry. 29 (1990) 74857491. [95] M. Diem, O. Lee and G.M. Roberts, Vibrational studies, normal-coordinate analysis, and infrared VCD of alanylalanine in the amide-III spectral region, J. Phys. Chem. 96 (1992) 548-554. [96] S.S. Birke, I. Agbaje and M. Diem, Experimental and computational infrared CD studies of prototypical peptide conformations, Biochemistry. 31 (1992) 450. [97] H. DeVoe, Optical properties of molecular aggregates. II. Classical model of electronic absorption and refraction, J. Chem. Phys. 41 (1964) 393-400. [98] H. DeVoe, Optical properties of molecular aggregates. II. Classical theory of refraction, absorption and optical activity in solutions and crystals, J. Chem. Phys. 43 (1965) 3199-3208. [99] B.D. Self and D.S. Moore, Nucleic acid vibrational circular dichroism, absorption and linear dichroism spectra. 1. A DeVoe theory approach, Biophys. J. 73 (1997) 339-347. [100] B.D. Self and D.S. Moore, Nucleic acid vibrational circular dichroism, absorption and linear dichroism spectra. 2. A DeVoe theory approach, Biophys. J. 74 (1998) 2249-2258.
215
[101] H. Ito and Y.J. I'Haya, Linear response polarizability theory for vibrational circular dichroism: VCD and IR bandshape calculations of D-helical and E-pleated polypeptides, Bull. Chem. Soc. Jpn. 67 (1994) 1238-1245. [102] H. Ito, Linear response polarizability bandshape calculations of vibrational circular dichroism, vibrational absorption, and electronic circular dichroism of cyclo(Gly-Pro-Gly-D-Ala-Pro): a small cyclic peptide having E- and J-turns, Biospectroscopy. 2 (1996) 17-37. [103] P.J. Stephens and M.A. Lowe, Vibrational circular dichroism, Ann. Rev. Phys. Chem. 36 (1985) 213241. [104] P.J. Stephens, F.J. Devlin, C.S. Ashvar, C.F. Chabalowski and M.J. Frisch, Theoretical calculation of vibrational circular dichroism spectra, Faraday Discuss. 99 (1994) 103-119. [105] T.B. Freedman and L.A. Nafie, Theoretical formalism and models for vibrational circular dichroism intensity. In: Modern Nonlinear Optics M. Evans and S. Kielich (Eds.), Vol. 3, Wiley, New York, 1994, 207-263. [106] V. Liegeois, K. Ruud and B. Champagne, An analytical derivative procedure for the calculation of vibrational Raman optical activity spectra, J. Chem. Phys. 127 (2007). [107] M. Born and R. Oppenheimer, Zur Quantentheorie der Molekeln, Annalen der Physik. 84 (1927) 457284. [108] P. Bou and L. Bednrov, Anharmonic force field of formamide - a computational study, J. Phys. Chem. 99 (1995) 5961-5966. [109] P. Danek, J. Kapitn, V. Baumruk, L. Bednrova, V. Kopeck and P. Bou, Anharmonic effects in IR, Raman, and Raman optical activity spectra of alanine and proline zwitterions, J. Chem. Phys. 126 (2007). [110] S. Califano, Vibrational states, John Wiley & Sons, London, 1976. [111] E.B. Wilson, J.C. Decius and P.C. Cross, Molecular vibrations, Dover, New York, 1980. [112] J. Kubelka, R.A.G.D. Silva, P. Bou, S.M. Decatur and T.A. Keiderling, Chirality in peptide vibrations. Ab Initio computational studies of length, solvation, hydrogen bond, dipole coupling and isotope effects on vibrational CD. In: Chirality: Physical Chemistry. ACS Symposium Series J.M. Hicks (Ed.), Vol. 810, American Chemical Society, Washington DC, 2002, 50-64. [113] P. Bou and T.A. Keiderling, Partial optimization of molecular geometry in normal coordinates and use as a tool for simulation of vibrational spectra, J. Chem. Phys. 117 (2002) 4126-4132. [114] P. Bou, Convergence properties of the normal mode optimization and its combination with molecular geometry constraints, Collect. Czech. Chem. Commun. 70 (2005) 1315-1340. [115] W.T. King, Effective Atomic Charge. In: Vibrational Intensities in Infrared and Raman Spectroscopy W.B. Person and G. Zerbi (Eds.), Elsevier, Amsterdam, 1982, 122-142. [116] P.L. Polavarapu, Recent advances in model calculations of vibrational optical activity, Vib. Spect. Struct. 13 (1984) 103-160. [117] J. Hornek, P. Kaprlov and P. Bou, Simulations of vibrational spectra from classical trajectories: Calibration with ab initio force fields, J. Chem. Phys. 127 (2007). [118] J. Gerratt and I.M. Mills, Force constants and dipole-moment derivatives of molecules from perturbed hartree-fock calculations I., J. Chem. Phys. 49 (1968) 1719-1729. [119] M.J. Frisch, M. Head-Gordon and J.A. Pople, Direct analytic second derivatives and electric field properties, Chem. Phys. 141 (1990) 189-196. [120] L.A. Nafie, Raman Optical Activity. In: Modern Nonlinear Optics, Part 3 M. Evans and S. Kielich (Eds.), Vol. 85, Wiley, New York, 1994, 105-206. [121]P.J. Stephens, Gauge dependence of vibrational magnetic transition moments and rotational strengths, J. Phys. Chem. 91 (1987) 1712-1715. [122] P.J. Stephens, C.S. Ashvar, F.J. Devlin, J.R. Cheeseman and M.J. Frisch, Ab Initio calculation of atomic axial tensors and vibrational rotational strengths using density functional theory, Mol. Phys. 89 (1996) 579-594. [123] A.D. Buckingham, P.W. Fowler and P.A. Galwas, Velocity-dependent property surfaces and the theory of vibrational circular dichroism, Chem. Phys. 112 (1987) 1-14. [124] M.A. Lowe, G.A. Segal and P.J. Stephens, The theory of vibrational circular dichroism: trans 1,2dideuteriocyclopropane, J. Am. Chem. Soc. 108 (1986) 248-256.
216
[125] L.A. Nafie and T.B. Freedman, Vibronic coupling theory of infrared vibrational transitions., J. Chem. Phys. 78 (1983) 7108-7116. [126] P. Lazaretti, R. Zanasi and P.J. Stephens, Magnetic dipole transition moments of vibrational transitions. An alternative formalism., J. Phys. Chem. 90 (1986) 6761-6763. [127] K.J. Jalkanen, P.J. Stephens, P. Lazaretti and R. Zanasi, Nuclear shielding tensors, atomic polar and axial tensors, and vibrational dipole and rotational strengths of NHDT, J. Chem. Phys. 90 (1989) 32043213. [128] K.J. Jalkanen, P.J. Stephens, P. Lazaretti and R. Zanasi, Random phase approximation calculations of vibrational circular dichroism: trans-2,3-dideuteriooxirane, J. Phys. Chem. 93 (1989) 6583-6584. [129] D. Yang and A.J. Rauk, Vibrational circular dichroism intensities. Ab initio vibronic coupling theory using the distributed origin gauge., J. Chem. Phys. 97 (1992) 6517-6534. [130] P.J. Stephens, F.J. Devlin, C.F. Chabalowski and M.J. Frisch, Ab Initio calculation of vibrational absorption and circular dichroism spectra using density functional force fields, J. Phys. Chem. 98 (1994) 11623-11627. [131] P. Bou, J. McCann and H. Wieser, The excitation scheme: a new method for calculation of vibrational circular dichrosim, J. Chem. Phys. 108 (1998) 8782-8789. [132] R.E. Bruns, Approximate quantum mechanical calculations of infrared and Raman intensities. In: Vibrational Intensities in Infrared and Raman Spectroscopy W.B. Person and G. Zerbi (Eds.), Elsevier, Amsterdam, 1982, 143-158. [133] J.P.P. Stewart, Semiempirical vibrational frequencies (including scaling). In: The Encyclopedia of Computational Chemistry P.R. Schleyer, N.L. Allinger, T. Clark, J. Gasteiger, P.A. Kollman, H.F. Schaefer III and P.R. Schreiner (Eds.), Vol. 3, John Wiley & Sons, Chichester, 1998, 2579-2582. [134]P. Pulay, Ab initio calculaion of force constants and equilibrium geometries in polyatomic molecules. I. Theory, Mol. Phys. 17 (1969) 187-304. [135] P. Pulay, G. Fogarasi, G. Pongor, J.E. Boggs and A. Vargha, Combination of theoretical ab initio and experimental information to obtain reliable harmonic force constants. Scaled Quantum Mechanical (SQM) force fields for glyoxal, acrolein, butadiene, formaldehyde and ethylene, J. Am. Chem. Soc. 105 (1983) 7037-7047. [136]G. Fogarasi and P. Pulay, Ab initio vibrational force-fields, Annu. Rev. Phys. Chem. 35 (1984) 191-213. [137] N.C. Handy, D.J. Tozer, G.J. Laming, C.W. Murray and R.D. Amos, Analytic second derivatives of the potential energy surface, Isr. J. Chem. 33 (1993) 331-344. [138] Y. Yamaguchi, J.D. Goddard, Y. Osamura and H. Schaefer, A New Dimension to Quantum Chemistry : Analytic Derivative Methods in Ab Initio Molecular Electronic Structure Theory, Oxford University Press, New York, 1994. [139] W. Kohn and L.J. Sham, Self-consistent equations including exchange and correlation effects, Phys. Rev. A. 140 (1965) 1133-1138. [140] T. Ziegler, Approximate density functional theory as a practical tool in molecular energetics and dynamics, Chem. Rev. 91 (1991) 651-667. [141] R.G. Parr and W. Yang, Density functional theory of the electronic structure of molecules, Annu. Rev. Phys. Chem. 46 (1995) 701-728. [142] W. Kohn, A.D. Becke and R.G. Parr, Density functional theory of electronic structure, J. Phys. Chem. 100 (1996) 12974-12980. [143] P.M.W. Gill, Density functional theory, Hartree-Fock, and the self-consistent field. In: Encyclopedia of computational chemistry P.R. Schleyer, N.L. Allinger, T. Clark, J. Gasteiger, P.A. Kollman, H.F. Schaefer III and P.R. Schreiner (Eds.), Vol. 1, Wiley, Chichester, 1998, 678-689. [144] B.G. Johnson and M.J. Frisch, An implementation of analytic second derivatives ofthe gradientcorrected density functional energy, J. Chem. Phys. 100 (1994) 7429-7442. [145] F.J. Devlin, J.W. Finley, P.J. Stephens and M.J. Frisch, Ab Initio calculation of cibrational absorption and circular dichroism spectra using density functional force fields - A comparison of local, nonlocal and hybrid density functionals, J. Phys. Chem. 99 (1995) 16883-16902. [146] J.R. Cheeseman, M.J. Frisch, F.J. Devlin and P.J. Stephens, Ab Initio calculation of atomic axial tensors and vibrational rotational strengths using density functional theory, Chem. Phys. Lett. 252 (1996) 211-220.
217
[147] A.A. El-Azhary and H.U. Suter, Comparison between optimized geometries and vibrational frequencies calculated by the DFT methods., J. Phys. Chem. 100 (1996) 15056-15063. [148] G. Keresztury and B. Paizs, The quality of ab initio quantum mechanical prediction of vibrational transition dipole directions and infrared intensities, J. Mol. Struct. 410 (1997) 331-337. [149] M.D. Halls and H.B. Schlegel, Comparison of the performance of local, gradient corrected, and hybrid density functional models in predicting infrared intensities., J. Chem. Phys. 109 (1998) 10587-10593. [150] B. Galabov, Y. Yamaguchi, R.B. Remington and H.F. Schaefer, High level ab initio quantum mechanical predictions of infrared intensities, J. Phys. Chem. A. 106 (2002) 819-832. [151] A. St-Amant, Density functional methods in biomolecular modeling. In: Reviews in Computational Chemistry K.B. Lipkowitz and D.B. Boyd (Eds.), Vol. 7, VCH Publishers, New York, 1996, 217-259. [152] E.S. Kryachko and E.V. Ludena, Energy density functional theory of many-electron systems, Kluwer, Dordrecht, 1990. [153] R.G. Parr and W. Yang, Density-functional theory of atoms and molecules, Oxford university press, New York, 1994. [154] L.J. Bartolotti and K. Flurchick, An introduction to density functional theory. In: Reviews in computational chemisry K.B. Lipkowitz and D.B. Boyd (Eds.), Vol. 7, VCH Publishers, New York, 1996. [155] J.C. Burant, M.C. Strain, G.E. Scuseria and M.J. Frisch, Kohn-Sham analytic energy second derivatives with the Gaussian ver fast multipole method (GvFMM), Chem. Phys. Lett. 258 (1996) 4555. [156] GAUSSIAN, http://www.gaussian.com. [157] R.D. Amos, CADPAC: The Cambridge Analytic Derivative Package, SERC Laboratory, Daresbury, UK, 1995. [158] DALTON, http://www.kjemi.uio.no/software/dalton/dalton.html. [159] ADF, http://www.scm.com/. [160] A.D. Becke, Exchange-correlation approximations in density-functional theory. In: Modern electronic structure theory D.R. Yarkony (Ed.), Vol. 2, World Scientific, Singapore, 1995, 1022-1046. [161] A. Becke, Density-functional exchange-energy approximation with correct asymptotic behavior., Phys. Rev. A. 38 (1988) 30983100. [162] C. Lee, W. Yang and R.G. Parr, Development of the Colle-Salvetti correlation-energy formula into a functional for the electron density, Phys. Rev. B. 37 (1988) 785-789. [163] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh and C. Fiolhais, Atoms, molecules, solids, and surfaces: Applications of the generalized gradient approximation for exchange and correlation, Phys. Rev. B. 46 (1992) 6671-6687. [164] J.P. Perdew and Y. Wang, Accurate and simple analytic representation of the electron-gas correlation energy, Phys. Rev. B. 45 (1992) 13244-13249. [165] J.P. Perdew, Accurate density functional for the energy: Real-space cutoff of the gradient expansion for the exchange hole, Phys. Rev. Lett. 55 (1985) 1665-1668. [166] J.P. Perdew and Y. Wang, Accurate and simple density functional for the electronic exchange energy: Generalized gradient approximation, Phys. Rev. B. 33 (1988) 8800-8802. [167] A.D. Becke, density-functional thermochemistry: the role of exact exchange, J. Chem. Phys. 98 (1993) 5648-5652. [168] M.P. Gaigeot, M. Martinez and R. Vuilleumier, Infrared spectroscopy in the gas and liquid phase from first principle molecular dynamics simulations: application to small peptides, Mol. Phys. 105 (2007) 2857-2878. [169] W.J. Hehre, R.F. Stewart and J.A. Pople, Self-consistent molecular-orbital methods .I. Use of Gaussian expansions of Slater-type atomic orbitals, J. Chem. Phys. 51 (1969) 2657-2662. [170] R. Ditchfield, W.J. Hehre and J.A. Pople, Self-consistent molecular-orbital methods .9. Extended Gaussian-type basis for molecular-orbital studies of organic molecules, J. Chem. Phys. 54 (1971) 724729. [171] R. Krishnan, J.S. Binkley, R. Seeger and J.A. Pople, Self-consistent molecular-orbital methods .20. Basis set for correlated wave-functions, J. Chem. Phys. 72 (1980) 650-654.
218
[172] M.J. Frisch, J.A. Pople and J.S. Binkley, Self-consistent molecular-orbital methods .25. Supplementary functions for Gaussian-basis sets, J. Chem. Phys. 80 (1984) 3265-3269. [173] R. McWeeny, Perturbation theory for the Fock-Dirac density matrix, Phys. Rev. 126 (1962) 1028-1034. [174] R.M. Stevens, R.M. Pitzer and W.N. Lipscomb, Perturbed Hartree-Fock calculations. I. magnetic susceptibility and shielding in the LiH molecule, J. Chem. Phys. 38 (1963) 550-560. [175] R.E. Strattmann, J.C. Burant, G.E. Scuseria and M.J. Frisch, Improving harmonic vibrational frequency calculations in density functional theory, J. Chem. Phys. 106 (1997) 10175-10183. [176] L. Greengard and V. Rokhlin, A fast algorithm for particle simulations, J. Comput. Phys. 73 (1987) 325-348. [177] M.C. Strain, G.E. Scuseria and M.J. Frisch, Achieving linear scaling for the electronic quantum Coulomb problem, Science 271 (1996) 51-53. [178] M. Challacombe and E. Schwegler, Linear Scaling Computation of the Fock matrix, J. Chem. Phys. 106 (1996) 5526-5535. [179] R.E. Strattmann, G.E. Scuseria and M.J. Frisch, Achieving linear scaling for exchange-correlation density functional quadratures, Chem. Phys. Lett. 257 (1996) 213-223. [180] G.E. Scuseria, Linear scaling density functional calculations with Gaussian orbitals, J. Phys. Chem. A. 103 (1999) 4782-4790. [181] Y. Yamaguchi, M.J. Frisch, J. Gaw, H.F. Schaeffer and J.S. Binkley, Analytic evaluation and basis set dependence of intensities of infrared spectra, J. Chem. Phys. 84 (1986) 2262-2278. [182] F. London, The quantic theory of inter-atomic currents in aromatic combinations, J. Phys. Radium. 8 (1937) 397. [183] R. Ditchfield, Self-consistent perturbation theory of diamagnetism I. A gauge-invariant LCAO method for N. M. R. chemical shifts., Mol. Phys. 27 (1974) 789-807. [184] K.L. Bak, F.J. Devlin, C.S. Ashvar, P.R. Taylor, M.J. Frisch and P.J. Stephens, Ab initio calculations of vibrational circular dichroism spectra using gauge-invariant atomic orbitals, J. Phys. Chem. 99 (1995) 14918-14922. [185] K.L. Bak, P. Jorgensen, T. Helgaker and K. Ruud, Basis set convergence and correlation effects in vibrational circular dichroism calculations using London orbitals, Faraday Disc. 99 (1994) 121-129. [186] K.L. Bak, P. Jorgensen, T. Helgaker, K. Ruud and H.J. Jensen, Gauge-origin independent multiconfigurational self-consistent-field theory for vibrational circular dichroism, J. Chem. Phys. 98 (1993) 8873-8887. [187] N.G. Mirkin and S. Krimm, Ab initio vibrational analysis of hydrogen-bonded trans- and cis-Nmethylacetamide, J. Am. Chem. Soc. 113 (1991) 9742-9747. [188] N.G. Mirkin and S. Krimm, Ab initio vibrational analysis of isotopic derivatives of aqueous hydrogenbonded trans-N-methylacetamide, J. Mol. Struct. 377 (1996) 219-234. [189] L.M. Markham and B.S. Hudson, Ab initio analysis of the effects of aqueous solvation on the resonance Raman intensities of N-methylacetamide, J. Phys. Chem. 100 (1996) 2731-2737. [190]S. Samdal, Acetamide, a challenge to theory and experiment? On the molecular structure, conformation, potential to internal rotation of the methyl group and force fields of free acetamide as studied by quantum chemical calculations., J. Mol. Struct. 440 (1998) 165-174. [191] T. Kupka, I.P. Gerothanassis and I.N. Demetropoulos, Density functional study of a model amide. Prediction of formamide geometry, dipole moment, IR harmonic vibrational QC=O and GIAO NMR shieldings., J. Mol. Struct. (THEOCHEM). 531 (2000) 143-157. [192] C. Bruyneel, A.K. Chandra, T. Uchimaru and T. Zeegers-Hyuskens, Theoretical and experimental study of the vibrational spectrum of N-acetyl-L-alanine., Spectrochem. Acta A. 56 (2000) 591-602. [193] T.C. Cheam and S. Krimm, Ab initio force fields for alanine dipeptide in C5 ad C7 conformations., J. Mol. Struct. 188 (1989) 15-43. [194] T.C. Cheam and S. Krimm, Ab initio force fields of alanine dipeptide in four non-hydrogen bonded conformations, J. Mol. Struct. 206 (1990) 173-203. [195] T.C. Cheam, Normal mode analysis of alanine dipeptide in the crystal conformation using a scaled ab initio force field, J. Mol . Struct. 295 (1993) 259-271. [196] A.F. Weir, A.H. Lowrey and R.W. Williams, Scaled quantum mechanical force field for alanyl-alanine peptide in solution, Biopolymers. 58 (2001) 577-591.
219
[197] B.G. Johnson, P.M.W. Gill and J.A. Pople, The performance of a family of density functional methods, J. Chem. Phys. 98 (1992) 5612-5626. [198] M.W. Wong, Vibrational frequency prediction using density functional theory, Chem. Phys. Lett. 256 (1996) 391-399. [199] A.P. Scott and L. Radom, Harmonic vibrational frequencies: An evaluation of Hartree-Fock, MllerPlesset, quadratic configuration interaction, density functional theory and semiempirical scale factors., J. Phys. Chem. 100 (1996) 16502-16513. [200] L.Y. Fan and T. Ziegler, Application of density functional theory to infrared-absorption Intensity calculations on main group molecules, J. Chem. Phys. 96 (1992) 9005-9012. [201] T. Miyazawa, T. Shimanouchi and S.-I. Mizushima, Normal vibrations of N-methyl acetamide, J. Chem. Phys. 29 (1958) 611-616. [202] X.G. Chen, R. Schweitzer-Stenner, S. Krimm, N.G. Mirkin and S.A. Asher, N-methylacetamide and its hydrogen-bonded water molecules are vibrationally coupled, J. Am. Chem. Soc. 116 (1994) 1114111142. [203] X.G. Chen, R. Schweitzer-Stenner, S.A. Asher, N.G. Mirkin and S. Krimm, Vibrational assignments of trans-N-methylacetamide and some of its deuterated isotopomers from band decomposition of IR, visible, and resonance Raman spectra, J. Phys. Chem. 99 (1995) 3074-3083. [204] W.-G. Han and S. Suhai, Density functional studies on N-Methylacetamide-water complexes., J. Phys. Chem. 100 (1996) 3942-3949. [205] S.A. Asher, P. Li, Z.H. Chi, X.G. Chen, R. Schweitzer-Stenner, N.G. Mirkin and S. Krimm, Vibrational assignments of trans-N-methylacetamide and some of its deuterated isotopomers from band decomposition of IR, visible, and resonance Raman spectra., J. Phys. Chem. 99 (1997) 3074-3083. [206] H. Torii, T. Tatsumi, T. Kanazawa and M. Tasumi, Effects of intermolecular hydrogen-bonding interactions on the amide I mode of N-methylacetamide: matrix-isolation infrared studies and ab initio molecular orbital calculations, J. Phys. Chem. B. 102 (1998) 309-314. [207] R. Schweitzer-Stenner, G. Sieler, N.G. Mirkin and S. Krimm, Intermolecular coupling in liquid and crystalline states of trans-N-methylacetamide investigated by polarized Raman and FT-IR spectroscopies, J. Phys. Chem. 102 (1998) 118-127. [208] V. Bakken, T. Helgaker, W. Klopped and K. Ruud, The calculation of molecular geometrical properties in Hellman-Feynman approximation, Mol. Phys. 96 (1999) 653-671. [209] J. Kubelka and T.A. Keiderling, Ab initio calculation of amide carbonyl stretch vibrational frequencies in solution with modified basis sets, J. Phys. Chem. A. 105 (2001) 10922-10928. [210] P. Bou and T.A. Keiderling, Empirical modeling of the peptide amide I band IR intensity in water solution, J. Chem. Phys. 119 (2003) 11253-11262. [211] S. Ham, J. Kim, H. Lee and M. Cho, Correlation between electronic and molecular structure distortions and vibrational properties. II. Amide I modes of NMA-nD2O complexes, J. Chem. Phys. 118 (2003) 3491-3498. [212] N.A. Besley, Ab initio modeling of amide vibrational bands in solution, J. Phys. Chem. A. 108 (2004) 10794-10800. [213] P. Bou, D. Michalk and J. Kapitn, Empirical solvent correction for multiple amide group vibrational modes, J. Chem. Phys. 122 (2005) 144501. [214] T. Hayashi, W. Zhuang and S. Mukamel, Electrostatic DFT map for the complete vibrational amide band of NMA, J. Phys. Chem. A. 109 (2005) 9747-9759. [215] T.L. Jansen and J. Knoester, A transferable electrostatic map for solvation effects on amide I vibrations and its application to linear and two-dimensional spectroscopy, J. Chem. Phys. 124 (2006). [216] K.E. Amunson and J. Kubelka, On the temperature dependence of amide I frequencies of peptides in solution, J. Phys. Chem. B. 111 (2007) 9993-9998. [217] G.V. Papamokos and I.N. Demetropoulos, Vibrational frequencies of amides and amide dimers: The assessment of PW91(xc) functional, J. Phys. Chem. A. 108 (2004) 7291-7300. [218] R.K. Dukor, Vibrational Circular Dichroism of Selected Peptides, Polypeptides and Proteins, PhD Thesis, University of Illinois at Chicago, Chicago, Illinois, 1991. [219] R.K. Dukor and T.A. Keiderling, Reassessment of the random coil conformation: Vibrational CD study of proline oligopeptides and related polypeptides, Biopolymers. 31 (1991) 1747-1761.
220
[220] N.R. Kallenbach, Breathing life into the folding pathway of cytochrome c, Nat. Struct. Biol. 2 (1995) 813-816. [221] T.P. Creamer and G.D. Rose, Interactions between hydrophobic side-chains within alpha-helices, Protein Sci. 4 (1995) 1305-1314. [222] R.K. Dukor and T.A. Keiderling, Mutarotation studies of poly-L-proline using FTIR, electronic and vibrational circular dichroism, Biospectroscopy. 2 (1996) 83-100. [223] M.L. Tiffany and S. Krimm, Effect of temperature on the circular dichroism spectra of polypeptides in the extended state, Biopolymers. 11 (1972) 2309-2316. [224] G.D. Rose, Unfolded Proteins, Adv. Prot. Chem. (2002). [225] Z.S. Shi, C.A. Olson, G.D. Rose, R.L. Baldwin and N.R. Kallenbach, Polyproline II structure in a sequence of seven alanine residues, Proc. Natl. Acad. Sci. USA. 99 (2002) 9190 - 9195. [226] T.A. Keiderling and Q. Xu, Unfolded peptides and proteins studied with infrared absorption and vibrational circular dicrhoism spectra, Adv. Prot. Chem. (2002) 111-161. [227] P. Bou and T.A. Keiderling, Ab initio simulation of the vibrational circular dichroism of coupled peptides, J. Am. Chem. Soc. 115 (1993) 9602-9607. [228] P. Bou, J. Kubelka and T.A. Keiderling, Simulations of oligopeptide vibrational circular dichroism. Effects of isotopic labeling, Biopolymers. 53 (2000) 380-395. [229] K.J. Jalkanen and S. Suhai, N-acetyl-L-alanine-N'-methylamide: a density functional analysis of the vibrational absorption and vibrational circular dichroism spectra., Chem. Phys. 208 (1996) 81-116. [230] W.-G. Han, K.J. Jalkanen, M. Elstner and S. Suhai, Theoretical study of aqueous N-acetyl-L-AlanineN'-Methylamide: Structures and Raman, VCD, and ROA spectra., J. Phys. Chem. B. 102 (1998) 25872602. [231] M. Knapp-Mohammady, K.J. Jalkanen, F. Nardi, R.C. Wade and S. Suhai, L-Alanyl-L-alanine in the zwitterionic state: structures determined in the presence of explicit water molecules and with continuum models using density functional theory, Chem. Phys. 240 (1999) 63-77. [232] F. Eker, X. Cao, L. Nafie and R. Schweitzer-Stenner, Tripeptides Adopt Stable Structures in Water. A Combined Polarized Visible Raman, FTIR, and VCD Spectroscopy Study, J. Am. Chem. Soc. 124 (2002) 14330-14341. [233] R. Wieczorek and J.J. Dannenberg, Comparison of fully optimized alpha- and 3(10)-helices with extended beta-strands. An ONIOM density functional theory study, J. Am. Chem. Soc. 126 (2004) 14198-14205. [234] P.J. Stephens, J.R. Cheeseman and F.J. Devlin, Ab initio vibrational spectra of D-helices preparation). (in
[235] J. Hilario, J. Kubelka and T.A. Keiderling, Optical spectroscopic investigations of model beta-sheet hairpins in aqueous solution J. Am. Chem. Soc. 125 (2003) 7562-7574. [236] R. Huang, J. Kubelka, W. Barber-Armstrong, R.A.G.D. Silva, S.M. Decatur and T.A. Keiderling, The nature of vibrational coupling in helical peptides: An isotopic labeling study, J. Am. Chem. Soc. 126 (2004) 2346-2354. [237] V. Setnika, R. Huang, C.L. Thomas, M.A. Etienne, J. Kubelka, R.P. Hammer and T.A. Keiderling, IR study of cross-strand coupling in a beta-hairpin peptide using isotopic labels J. Am. Chem. Soc. 127 (2005) 4992-4993. [238] P. Bou and T.A. Keiderling, Vibrational spectral simulation for peptides of mixed secondary structure: Method comparisons with the Trpzip model hairpin, J. Phys. Chem. B. 109 (2005) 23687-23697. [239] P. Bou and T.A. Keiderling, Ab initio modeling of amide I coupling in antiparallel beta-sheets and the effect of 13C isotopic labeling on infrared spectra, J. Phys. Chem. B. 109 (2005) 5348-5357. [240] J. Kubelka, J. Kim, P. Bou and T.A. Keiderling, Contribution of transition dipole coupling to amide coupling in IR spectra of peptide secondary structures, Vib. Spectrosc. 42 (2006) 63-73. [241] R. Huang, V. Setnika, M.A. Etienne, J. Kim, J. Kubelka, R.P. Hammer and T.A. Keiderling, Crossstrand coupling of a b-hairpin peptide stabilized with an Aib-Gly turn using isotope-edited IR spectroscopy, J. Am. Chem. Soc. (2007). [242] S. Dapprich, I. Komaromi, K.S. Byun, K. Morokuma and M.J. Frisch, A new ONIOM implementation in Gaussian98. Part I. The calculation of energies, gradients, vibrational frequencies and electric field derivatives, J. Molec. Struct. (THEOCHEM) 462 (1999) 1-21.
221
[243] J. Hilario, J. Kubelka, F.A. Syud, S.H. Gellman and T.A. Keiderling, Spectroscopic characterization of selected beta-sheet hairpin models, Biospectroscopy. 67 (2002) 233-236. [244] J. Kim, J. Kapitn, A. Lakhani, P. Bou and T.A. Keiderling, Tight beta-turns in peptides. DFT-based study of infrared absorption and vibrational circular dichroism for various conformers including solvent effects, Theor. Chem. Acc.. 119 (2008) 81-97. [245] T. Measey, A. Hagarman, F. Eker, K. Griebenow and R. Schweitzer-Stenner, Side chain dependence of intensity and wavenumber position of amide I' in IR and visible Raman spectra of XA and AX dipeptides, J. Phys. Chem. B. 109 (2005) 8195-8205. [246] H. Lee, S.S. Kim, J.H. Choi and M. Cho, Theoretical study of internal field effects on peptide amide I modes, J. Phys. Chem. B. 109 (2005) 5331-5340. [247] G. Papadonankis and A. Lakhani, Tripeptide model DFT calculated frequencies, (unpublished). [248] J. Kapitn, V. Baumruk and P. Bou, Demonstration of the ring conformation in polyproline by the Raman optical activity, J. Am. Chem. Soc. 128 (2006) 2438-2443. [249] J. Kapitn, V. Baumruk, V. Kopeck and P. Bou, Conformational flexibility of L-alanine zwitterion determines shapes of Raman and Raman optical activity spectral bands, J. Phys. Chem. A. 110 (2006) 4689-4696. [250] J. Kapitn, V. Baumruk, J. Kopeck, R. Pohl and P. Bou, Proline zwitterion dynamics in solution, glass, and crystalline state, J. Am. Chem. Soc. 128 (2006) 13451-13462. [251] J. Kim, R. Huang, J. Kubelka, P. Bou and T.A. Keiderling, Simulation of infrared spectra for betahairpin peptides stabilized by an Aib-Gly turn sequence: Correlation between conformational fluctuation and vibrational coupling, J. Phys. Chem. B. 110 (2006) 23590-23602. [252] D.R. Turner and J. Kubelka, Infrared and vibrational CD spectra of partially solvated alpha-helices: DFT-based simulations with explicit solvent, J. Phys. Chem. B. 111 (2007) 1834-1845. [253] P. Bou and P. Malo, MCM95 molecular graphics, Academy of Sciences, Prague (1995-2009). [254] J.H. Choi, S.Y. Ham and M. Cho, Local amide I mode frequencies and coupling constants in polypeptides, J. Phys. Chem. B. 107 (2003) 9132-9138. [255] J.-H. Choi and M. Cho, Amide I vibrational circular dichroism of dipeptide: Conformation dependence and fragment analysis, J. Chem. Phys. 120 (2004) 4383-4392. [256] C. Lee and M.H. Cho, Local amide I mode frequencies and coupling constants in multiple-stranded antiparallel beta-sheet polypeptides, J. Phys. Chem. B. 108 (2004) 20397-20407. [257] J. Kubelka, IR and VCD spectroscopy of model peptides. Theory and Experiment., PhD thesis, University of Illinois at Chicago, Chicago, 2002. [258] I.H. McColl, E.W. Blanch, L. Hecht, N.R. Kallenbach and L.D. Barron, Vibrational Raman optical activity characterization of poly(L-proline) II helix in alanine oligopeptides J. Am. Chem. Soc. 126 (2004) 5076-5077 [259] J. Kubelka and T.A. Keiderling, Discrimination between peptide 310- and D-helices. Theoretical analysis of the impact of D-methyl substitution on experimental spectra, J. Am. Chem. Soc. 104 (2002) 5325-5332. [260] L. Tadesse, R. Nazarbaghi and L. Walters, Isotopically enhanced infrared spectroscopy: a novel method for examining secondary structure at specific sites in conformationally heterogeneous peptides, J. Am. Chem. Soc. 113 (1991) 7036-7037. [261] K. Halverson, I. Sucholeiki, T.T. Ashburn and R.T. Lansbury, Location of E-sheet-forming sequences in amyloid proteins by FTIR, J. Am. Chem. Soc. 113 (1991) 6701-6703. [262] S.M. Decatur and J. Antonic, Isotope-edited FTIR spectroscopy of helical peptides, J. Am. Chem. Soc. 121 (1999) 11914-11915. [263] T.D. Anderson, J. Hellgeth and P.T. Lansbury Jr., Isotope-edited infrared linear dichroism: Determination of amide orientational relationships, J. Am. Chem. Soc. 118 (1996) 6540-6546. [264] S.Y. Venyaminov, J.F. Hedstrom and F.G. Prendergast, Analysis of the segmental stability of helical peptides by isotope-edited infrared spectroscopy, Proteins-Structure Function and Genetics. 45 (2001) 81-89. [265] C.Y. Huang, Z. Getahun, T. Wang, W.F. DeGrado and F. Gai, Time-resolved infrared study of the helix-coil transition using 13C-labeled helical peptides, J. Am. Chem. Soc. 123 (2001) 12111-12112.
222
[266] J. Kubelka and T.A. Keiderling, The anomalous infrared amide I intensity distribution in 13C isotopically labeled peptide E-sheets comes from extended, multiple-stranded structurres. An ab initio study., J. Am. Chem. Soc. 123 (2001) 6142-6150. [267] K. Hauser, C. Krejtschi, R. Huang, L. Wu and T.A. Keiderling, Site-specific relaxation kinetics of a tryptophan zipper hairpin peptide using temperature-jump IR spectroscopy and isotopic labeling., J. Am. Chem. Soc. 130 (2008) 2984-2992. [268] R. Huang, L. Wu, D. McElheny, P. Bour, A. Roy, T. A. Keiderling, Cross-strand coupling and sitespecific unfolding thermodynamics of a trpzip beta-hairpin peptide using 13C isotopic labeling and IR spectroscopy, J. Phys. Chem. B 113 (2009), 56615674 [269]A.T. Petkova, Y. Ishii, J.J. Balbach, O.N. Antzutkin, R.D. Leapman, F. Delaglio and R. Tycko, A structural model for Alzheimer's -amyloid fibrils based on experimental constraints from solid state NMR, Proc. Natl. Acad. Sci. USA. 99 (2002) 16742 - 16747. [270] S. Chimon and Y. Ishii, Capturing intermediate structures of Alzheimer's beta-amyloid, Abeta(1040) by solid-state NMR spectroscopy, J. Am. Chem. Soc. 127 (2005) 13472 - 13473. [271] V. Andrushchenko, H. Wieser and P. Bou, RNA structural forms studied by vibrational circular dichroism: Ab initio interpretation of the spectra, J. Phys. Chem. B. 108 (2004) 3899-3911. [272] L. Wang and T.A. Keiderling, Vibrational circular dichroism studies of the A-to-B conformational transition in DNA, Biochemistry. 31 (1992) 10265-10271. [273] L. Wang and T.A. Keiderling, Helical nature of poly(dI-dC)poly(dI-dC). Vibrational circular dichroism results, Nucleic Acids Res. 21 (1993) 4127-4132. [274] L.J. Wang, L.G. Yang and T.A. Keiderling, Vibrational circular dichroism of A-, B-, and Z-form nucleic acids in the PO2- stretching region, Biophys. J. 67 (1994) 2460-2467. [275] A. Annamalai and T.A. Keiderling, Vibrational circular dichroism of poly(ribonucleic acids). A comparative study in aqueous solution, J. Am. Chem. Soc. 109 (1987) 3125-3132. [276] L. Yang and T.A. Keiderling, Vibrational CD study of the thermal denaturation of poly(rA)poly(rU), Biopolymers. 33 (1993) 315-327. [277] P. Bou, V. Andrushchenko, M. Kabel and V.W. Maharaj, H, simulations of structure and Vibrational Spectra of Deoxyoctanucleotides, J. Phys. Chem. B. 109 (2005) 20579-20587. [278] V. Andrushchenko, H. Wieser and P. Bou, DNA Oligonucleotide-cis-platin binding: Ab initio interpretation of the vibrational spectra, J. Phys. Chem. A. 111 (2007) 9714-9723. [279] W.-Y. Yang, J.W. Pitera, W.C. Swope and M. Gruebele, Heterogeneous folding of the trpzip hairpin: Full atom simulation and experiment, J. Mol. Biol. 336 (2004) 241-251. [280] C.D. Snow, L.L. Qiu, D.G. Du, F. Gai, S.J. Hagen and V.S. Pande, Trp zipper folding kinetics by molecular dynamics and temperature-jump spectroscopy, Proc. Natl. Acad. Sci. U. S. A. 101 (2004) 4077-4082. [281] D.G. Du, Y.J. Zhu, C.Y. Huang and F. Gai, Understanding the key factors that control the rate of betahairpin folding, Proc. Natl. Acad. Sci. U. S. A. 101 (2004) 15915-15920. [282] D.G. Du, M.J. Tucker and F. Gai, Understanding the mechanism of beta-hairpin folding via phi-value analysis, Biochemistry. 45 (2006) 2668-2678. [283] G. Eaton, M.R.C. Symons and P.P. Rastogi, Spectroscopic studies of the solvation of amides with N-H groups, J. Chem. Soc. Faraday Trans. 1. 85 (1989) 3257-3271. [284] H. Torii, T. Tatsumi and M. Tasumi, Effects of hydration on the structure, vibrational wavenumbers, vibrational force field and resonance Raman intensities of N-methylacetamide, J. Raman Spectrosc. 29 (1998) 537-546. [285] M.F. DeCamp, L. DeFlores, J.M. McCracken, A. Tokmakoff, K. Kwac and M. Cho, Amide I vibrational dynamics of N-methylacetamide in polar solvents: The role of electrostatic interactions, J. Phys. Chem. B. 109 (2005) 11016-11026. [286] J.L. Gao and M. Freindorf, Hybrid ab initio QM/MM simulation of N-methylacetamide in aqueous solution, J. Phys. Chem. A. 101 (1997) 3182-3188. [287] P. Bou, On the influence of the water electrostatic field on the amide group vibrational frequencies, J. Chem. Phys. 121 (2004) 7545-7548. [288] R. Zhang, H.R. Li, Y. Lei and S.J. Han, Structures and interactions in N-methylacetamide-water mixtures studied by IR spectra and density functional theory, J. Mol. Struct. 693 (2004) 17-25.
223
[289] B. Mennucci and J.M. Martinez, How to model solvation of peptides? Insights from a quantummechanical and molecular dynamics study of N-methylacetamide. 1. Geometries, infrared, and ultraviolet spectra in water, J. Phys. Chem. B. 109 (2005) 9818-9829. [290] J.J. Dannenberg, Enthalpies of hydration of N-methylacetamide by one, two, and three waters and the effect upon the C=O stretching frequency. An ab initio DFT study, J. Phys. Chem. A. 110 (2006) 57985802. [291] K.J. Jalkanen, R.M. Nieminen, K. Frimand, J. Bohr, H. Bohr, R.C. Wade, E. Tajkhorshid and S. Suhai, A comparison of aqueous solvent models used in the calculation of the Raman and ROA spectra of Lalanine Chem. Phys. 265 (2001) 125-151 [292] S. Abdali, K.J. Jalkanen, H. Bohr, S. Suhai and R.M. Nieminen, The VA and VCD spectra of various isotopomers of L-alanine in aqueous solution, Chem. Phys. 282 (2002) 219-235. [293] V.W. Jurgensen and K. Jalkanen, The VA, VCD, Raman and ROA spectra of tri-L-serine in aqueous solution, Physical Biology. 3 (2006) S63-S79. [294] K.J. Jalkanen, M. Elstner and S. Suhai, Amino acids and small peptides as building blocks for proteins: comparative theoretical and spectroscopic studies, J. Mol. Struct.-Theochem. 675 (2004) 61-77. [295] E. Deplazes, W. van Bronswijk, F. Zhu, L.D. Barron, S. Ma, L.A. Nafie and K.J. Jalkanen, A combined theoretical and experimental study of the structure and vibrational absorption, vibrational circular dichroism, Raman and Raman optical activity spectra of the L-histidine zwitterion, Theor. Chem. Acc. 119 (2008) 155-176. [296] H. Guo and M. Karplus, Solvent influence on the stability of the peptide hydrogen bond: a supramolecula cooperative effect, J. Phys. Chem. 98 (1994) 7104-7105. [297] J. Kubelka, R. Huang and T.A. Keiderling, Solvent effects on IR and VCD spectra of helical peptides: DFT-based static spectral simulations with explicit water, J. Phys. Chem. B. 109 (2005) 8231-8243. [298] E.S. Manas, Z. Getahun, W.W. Wright, W.F. Degrado and J.M. Vanderkooi, Infrared spectra of amide groups in alpha-helical proteins: evidence for hydrogen bonding between helices and water, J. Am. Chem. Soc. 122 (2000) 9883-9890. [299] B. Roux and T. Simonson, Implicit solvent models, Biophys. Chem. 78 (1999) 1-20. [300] J. Tomasi and M. Persico, Molecular interactions in solution: an overview of methods based on continuum distribution of the solvent, Chem. Rev. 94 (1994) 2027-2094. [301] C.J. Cramer and D.G. Truhlar, Implicit solvation models: equilibria, structure, spectra and dynamics., Chem. Rev. 99 (1999) 2161-2200. [302] L. Onsager, Electric moments of molecules in liquids, J. Am. Chem. Soc. 58 (1936) 1486-1493. [303] J.B. Foresman, T.A. Keith, K.B. Wiberg, J. Snoonian and M.J. Frisch, Solvent effects. 5. Influence of cavity shape, truncatin of electrostatics, and electron correlation on ab initio reaction field calculations, J. Phys. Chem. 100 (1996) 16098-16104. [304] J. Tomasi, R. Bonaccorsi, R. Cammi and F.J. Olivares del Valle, Theoretical chemistry in solution. Some results and perspectives of the continuum methods and in particular of the polarizable continuum model, J. Mol. Struct. 234 (1991) 401-424. [305] A. Klamt and G. Schuurmann, COSMO: A new approach to dielectric screening in solvent with explicit expression for the screening energy and its gradient., J. Chem. Soc. Perkin Trans. 2 (1993) 799-805. [306] A. Klamt, Conductor-like screening models for real solvents: a new approach to the quantitative calculation of solvation phenomena, J. Phys. Chem. 99 (1995) 2224-2235. [307] A. Klamt, COSMO and COSMO-RS. In: The Encyclopedia of Computational Chemistry P.R. Schleyer, N.L. Allinger, T. Clark, J. Gasteiger, P.A. Kollman, H.F. Schaefer III and P.R. Schreiner (Eds.), Vol. 1, John Wiley & Sons, Chichester, 1998, 604-615. [308]A. Klamt, V. Jonas, T. Burger and J.C.W. Lohrentz, Refinement and parametrization of COSMO-RS, J. Phys. Chem. A. 102 (1998) 5074-5085. [309] V. Barone and M. Cossi, Quantum calculations of molecular energies and energy gradients in solution by a conductor solvent model, J. Phys. Chem. A. 102 (1998) 1995-2001. [310] M. Cossi and V. Barone, Analytical second derivatives of the free energy in solution by polarizable continuum models, J. Chem. Phys. 109 (1998) 6246-6254. [311]M. Cho, Correlation between electronic and molecular structure distortions and vibrational properties. I. Adiabatic approximations, J. Chem. Phys. 118 (2003) 3480.
224
Computational Linear and Nonlinear IR Spectroscopy of Amide I Vibrations in Proteins

Jun-Ho CHOI and Minhaeng CHO 1 Department of Chemistry and Center for Multidimensional Spectroscopy, Korea University, Seoul 136-701, Korea
Abstract. Amide I vibrational spectra of proteins can provide critical information on secondary structure elements and structural inhomogeneity. Despite that there exist a number of linear and nonlinear vibrational spectroscopic studies reported, it has been quite difficult to quantitatively simulate the amide I vibrational spectra of polypeptides and proteins. To achieve this goal, we have developed theoretical and computational methods such as constrained molecular dynamics simulation, Hessian matrix reconstruction method, and fragmentation approximation to describe delocalized amide I normal modes of proteins as linear combinations of amide I local modes. By using the computational scheme, amide I IR, vibrational circular dichroism, and two-dimensional IR photon echo spectra of polypeptides and protein in solution were simulated and compared with experimental results. The structure-spectrum relationships established are discussed. It is believed that the present computational method will be of use in shedding light on the underlying vibrational dynamics of protein as well as in interpreting experimentally measured linear and nonlinear amide I spectra of proteins in the future. Keywords. IR, 2D-IR, amide I vibration, peptide, protein
Introduction A variety of spectroscopic methods have been used to study structures and dynamics of proteins in solution. In particular, vibrational spectroscopy such as IR absorption and Raman scattering has served as the central tool for quantitatively determining the extent of secondary structure element in a given protein but also for investigating structural evolution of protein in a non-equilibrium state, i.e., protein folding and unfolding [13]. As a result, a number of characteristic marker bands that are particularly sensitive to peptide backbone structure have been found [410]. Among them, the amide I vibrations that are typically delocalized over several peptide bonds have been paid special attention, because the amide I IR band and its peak frequency are known to be highly informative [3,1122]. Successful empirical rules relating the measured amide I spectrum to proteins secondary structure have been reported and shown to be of use [1,13,2327]. For instance, even though the amide I IR band typically appears at
Corresponding Authors: Department of Chemistry and Center for Multidimensional Spectroscopy, Korea University, Seoul 136-701, Korea.
1
J.-H. Choi and M. Cho / Computational Linear and Nonlinear IR Spectroscopy
225
16101680 cm1, the peak frequency of right-handed -helix is about 16501660 cm1, whereas the amide I band of anti-parallel -sheet consists of one intense peak at 1620 1630 cm1 and a relatively weak peak at about 1680 cm1 [3,11,13]. Also, it was shown that the amide I IR band of -turn has a peak frequency in the range from 1660 to 1695 cm1. Despite that these empirical IR band analysis methods have been proven to be useful, a number of exceptions have also been observed because the amide I frequency and line shape are strongly dependent on the distribution of inhomogeneous local environment around peptide bonds and of vibrational couplings [2830]. Furthermore, due to spectral congestion such empirical rules are not always valid nor useful to quantitatively assess the relative populations of different secondary structure elements in a given protein. A more detailed information about protein structure could be obtained with vibrational circular dichroism (VCD), which measures the rotational strength [31]. In comparison to the IR absorption spectroscopy, VCD spectrum reveals additional information on the angle between transition electric dipole vector and transition magnetic dipole vector of a given normal mode note that if the angle is smaller (larger) than 90 degrees the corresponding VCD rotational strength and peak amplitude are positive (negative). That angle turned out to be strongly dependent on the local structure around the target peptide as well as on the inter-peptide amide I vibrational coupling strengths [3133]. Since the number of observables (peak frequencies, peak magnitudes, and signs) extracted from VCD spectrum is larger than that from IR spectrum (peak frequencies and magnitudes), VCD spectroscopy has a certain advantage over the IR spectroscopy. Recently, 2D vibrational spectroscopic techniques utilizing femtosecond IR laser pulses and multichannel array detectors have been developed and used to extract structural and dynamical properties of polypeptides [3441], nucleic acids [42,43], and organic molecular complexes [4446]. Due to the additional dimension in the frequency domain, the measured 2D IR spectrum contains far more detailed information on intermolecular and intramolecular vibrational interactions of complex molecular systems such as proteins. Note that the amide I vibrational coupling strengths and frequencies strongly depend on peptide local structure and surrounding environment such as hydrogen-bonding solvent and other peptides around the target peptide group [29,30,4749]. Consequently, those weak (secondary vibrational spectroscopic) properties and interactions become manifest in the two-dimensionally displayed spectrum as the corresponding cross peaks [50]. The 2D vibrational spectroscopy also has a dramatic gain in timeresolution achievable since it is only limited by the laser pulse width. Over the last decade, the 2D vibrational spectroscopy has been found to be of intense use to follow the dynamical evolution of protein structures and other complex systems in subpicosecond time scale [51]. As an attempt to theoretically analyze the spectral band shape and peak frequency of the amide I vibrational spectrum, the transferring force field approach has been used to calculate the Hessian matrix element such as diagonal and off-diagonal force constants. There are several approaches to compute the force field for large polypeptides. As a first approach, Krimm and co-workers used the classical normal coordinate analysis or the Wilson GF matrix method to obtain the F matrices for large polypeptides [11,52]. To obtain the off-diagonal elements of the Hessian matrix, they proposed the
226
transition dipole coupling (TDC) model where the inter-peptide interaction was approximated by a dipole-dipole interaction potential between neighboring peptides [53].
j Fjk (4 ) 1 q j
2 I 3r jk r jk rjk k T , T = jk jk 5 rjk qk 0 0
Fjk (in cm1) =
848619 Fjk (in mdyn 1 amu1)
(1)
Here, the first derivative of the dipole moment with respect to the amide I vibrational coordinate, ( j q j )0 , is the local transition dipole moment and T jk is the dipole dipole interaction tensor. They showed that the TDC model could provide a reasonable description on the frequency splitting of the C=O stretching modes in the hydrogenbonded carboxylic acid dimers and -sheet polypeptide [52]. Since the TDC model captures an essential part of the inter-peptide interaction-induced couplings and has been considered to be one of the simplest theories for such calculations, it has been widely used to interpret the amide I IR bands of various polypeptides [11]. Particularly, Torii and Tasumi showed that the TDC model can provide quantitatively reasonable prediction of the amide I IR band shapes for a few globular proteins in water [54]. Recently, the same model has been used to calculate the two-dimensional (2D) IR pumpprobe and photon echo spectra of polypeptides [34,5557]. Mendelshon and coworkers later used various empirical formulas with numerical parameters to take into consideration of the valence bond interaction, hydrogen bonding interaction, and through space electrostatic interaction such as transition dipole coupling, and they showed that the empirical approach could provide fairly accurate results on the amide I IR band when the three-dimensional structure of protein is known [14,58,59]. In order to describe the delocalized nature of peptide vibrations, Miyazawa [60,61] and Chirgadze [62] regarded the amide I vibrations of polypeptide as a collection of interacting oscillators where each oscillator is assumed to be localized on each peptide bond. They showed that the large frequency splitting of two peaks in the amide I IR spectrum of anti-parallel -sheet is caused by inter-strand hydrogen-bonding interaction. Extending the theoretical model developed by Miyazawa and Chirgadze, Cheatum et al. were able to numerically simulate the linear and nonlinear IR spectra of antiparallel -sheet polypeptides and showed that the observed cross peaks in their 2D IR photon echo spectra can be a direct signature of anti-parallel -sheet structure [63]. Later, Demirdven et al. found the Z-form 2D IR spectrum in the amide I vibrations of proteins is the characteristic feature of anti-parallel -sheet, where the Z shape 2D spectrum results from combined contributions from two diagonal peaks and two offdiagonal cross peaks [37,64]. The TDC model for simulating amide I IR spectra, which is based on the point dipole-dipole interaction approximation to the inter-peptide interaction, however requires assumptions such as: (i) a given amide I local mode transition is fully represented by a point dipole with properly adjusted location and direction, (ii) inter-peptide interaction is solely dictated by a dipole-dipole interaction without including other effects such as mechanical or high-order multipolar interactions, (iii) the coupling constant is therefore
227
given by the dipolar interaction between the two transition dipole moments. However, as found by the authors and coworkers, the TDC model has a certain limitation in quantitatively describing the coupling between two nearest neighboring amide I local modes, even though the long-range amide I vibrational couplings were successfully reproduced by the TDC model [29,6567]. Also, the intramolecular hydrogen bond and other types of inter-peptide interactions can induce diagonal force-constant shift [68,69], which were not properly taken into account in the previous works until recently. We found that the effects from the diagonal frequency shifts on the linear and nonlinear amide I vibrational spectra are significant. Particularly, noting that the local environment around each peptide in a given protein is fairly inhomogeneous due to the surrounding side chains and varying local dielectric constant inside the protein [30,70], In relation to these efforts and works, it is worth mentioning a few other approaches developed by others over the years. Lee and Krimm [71] obtained an empirical force field by performing ab initio calculations of N-methylacetamide and L-alanylL-alanine and improved the force field previously estimated to calculate the amide I band for -helical poly (L-alanine) [72]. Similarly, Keiderling and coworkers carried out systematic ab initio calculations for small oligopeptide fragment containing two to three peptide bonds and determined the force field of larger secondary structure polypeptides such as -helices and -sheets by transferring the force field of thus obtained fragment peptides [7375]. The simulated amide I IR and VCD spectra for the helices, -sheets, and -hairpins were found to be in good agreement with the experimental results [7476]. Nevertheless, this approach cannot be applied to globular proteins because there are no repeating regular structural motifs and furthermore it cannot be easily generalized to the polypeptide solutions where the polypeptide-solvent interaction induces significant solvatochromic amide I mode frequency shifts. In order to develop a more reliable computational algorithm for simulating amide I IR, VCD, and 2D IR spectra of polypeptides and proteins, we had to resolve a number of challenging issues: (i) the peptide-water interactions induce solvatochromic frequency shifts of amide I mode frequencies, (ii) they also modulate the amide I frequencies and such frequency fluctuations play an important role in vibrational dephasing, (iii) the intramolecular peptide-peptide interactions also induce amide I frequency shifts and fluctuations and are responsible for non-zero vibrational couplings, (iv) one should have a quantitatively reliable model for calculating transition magnetic dipoles that are key factors determining the amide I VCD intensities, (v) for any arbitrary polypeptides and proteins having no repeating structural motifs, one should have transferable force fields to construct the amide I vibrational Hamiltonian, and finally (vi) one needs to have a systematic theory and computational method for simulating 1D and 2D vibrational spectra of amide I vibrations in proteins. Over the years, we have developed a series of computational methods that can be used to overcome and resolve most of the issues raised above [29,32,33,47,77,78]. The basic strategy is to adopt the coupled-oscillator model, where both the diagonal and off-diagonal Hessian matrix elements in the amide I mode subspace are properly calculated by accounting the effects from the peptide-solvent and peptide-peptide interactions and from local environments. To properly identify the basis modes, that are amide I local modes in the present case, we developed the so-called Hessian matrix reconstruction (HMR) procedure [28,29]. Secondly, it was shown that the peptide-solvent and inter-peptide interaction-induced amide I local mode frequency shift and fluctua-
228
tion can be described by using a multivariate linear equation for amide I frequency [30,4749,79,80]. Thirdly, a fragmentation approximation method has been developed to calculate both the transition electric and magnetic dipoles of all amide I normal modes [32,33]. This method is based on the assumption that the transition electric and magnetic dipoles of a given normal mode can be written as linear combinations of a unit peptide (fragment) and the coefficients are eigenvector elements of a given amide I normal mode. Combining these methods together with classical MD simulations, we were able to numerically simulate amide I IR, VCD, and 2D IR spectra of small oligopeptides [81,82], -helix [4749], -sheets [49], -hairpins [78], and even a globular protein ubiquitin [30] in water. The agreements of thus calculated spectra with experimental results were found to be qualitatively good and for some cases they are in quantitative agreement. In the present chapter, we shall provide accounts on these computational methods.
1. Amide I local modes In order to quantitatively determine the vibrational coupling constant between two nearest neighboring amide I local modes as a function of the two dihedral angles and , we have developed the so-called Hessian matrix reconstruction (HMR) method utilizing ab initio geometry optimization and vibrational analysis of polypeptide under investigation [28,29]. Once the ab initio calculation of a given polypeptide is performed, the amide I normal mode frequencies and eigenvectors are the direct outputs. Here, we provide a brief outline of the HMR procedure. From the ab initio vibrational analysis, one can obtain the original Hessian matrix in the mass-weighted Cartesian coordinates, which is a 3Na by 3Na matrix, where Na is the number of constituent atoms, i.e.,
F11 F 21 = F3 Na ,1 F12 F22 F1,3 Na F3 Na ,3 Na
F (0)
(2)
Then, there exists the U-matrix that diagonalizes the above Hessian matrix, that is to say, U 1 F (0)U = . Note that the diagonal matrix elements of are the normal mode frequencies and the corresponding eigenvectors are the columns of the U-matrix. Since we are only interested in the amide I vibrations, we shall not pay attention to the eigenvectors and force constants of other types of peptide vibrations in the U and matrices. Now, let us assume that the eigenvector of the jth amide I local mode is fully determined by the eigenvector elements, in the Cartesian coordinate system, associated with the six atoms, O(=C), C(=O), N(-H), H(-N), C, and H(-C) that constitute the jth peptide bond. By denoting Um,j as the corresponding eigenvector elements, where m is the index of the mass-weighted Cartesian coordinates of the six atoms constituting the jth peptide bond, the Hessian matrix elements in the amide I local mode subspace can be obtained as [48]
229
1 (0) Fij = U i , n Fnm U m , j . n,m
(3)
The diagonal and off-diagonal matrix elements of F correspond to the amide I local mode force constants and coupling force constants, respectively. Depending on the ab initio calculation method chosen, one should properly multiply a scaling factor to correct the harmonic frequency thus obtained. However, we found that a special care has to be taken when the AM1 method was chosen. We found that, instead of simply multiplying a scaling factor, one should use a proper scaling equation to correct the amide I local mode frequencies. The validity of such a scaling method was tested by comparing the AM1 results with high-level (HF/6-311++G**) ab initio calculated amide I frequencies for a wide range of peptides. An appropriate scaling equation converting the AM1 amide I frequency into HF/6-311++G** frequency was found to be [83]
scaled = a + b( AM 1 1994.5) ,
(4)
where the two constants a and b are 1707.1 cm1 and 1.8, respectively. Note that AM 1 is the un-scaled AM1 amide I local mode frequency. This scaling equation was further tested by directly comparing the numerically simulated IR, VCD, and 2D IR spectra of -helix and its isotopomers [48] in water as well as of -hairpin [78] in water with experimental results. In the following subsections, we will present the HMR calculation results for small oligopeptides, alanine-based polypeptides, and ubiquitin. 1.1. Oligopeptides with Different Secondary Structures The ab initio geometry optimizations and vibrational analyses of polypeptides containing two to five peptide bonds were performed with basis set of RHF/6-311++G** level. The seven representative conformations, such as RHH (right-handed -helix; = 57 and = 47), 310H (310-helix; = 49 and = 26), H (-helix; = 57 and = 70), LHH (left-handed -helix; = 60 and = 60), PB (parallel -strand; = 119 and = 113), APB (anti-parallel -strand; = 139 and = 135), and FEB (fully extended -strand; = 180 and = 180) have been considered in detail. By using the Hessian matrix reconstruction method described above, the diagonal force constants and coupling constants of amide I modes were obtained. The amide I normal modes of a given polypeptide with N peptide bonds are assumed to be described as linear combinations of amide I local mode of N peptide units (see Fig. 1). A consequence of this critical assumption, that is the key aspect of the coupled oscillator model, is that all vibrational properties of amide I normal modes can be determined by properly combining those of each individual peptide unit. In Fig. 2(a), the average vibrational coupling constants between two nearest neighboring amide I local modes, which were obtained from the ab initio calculations in combination with the HMR analyses for the seven representative secondary structure oligopeptides, are directly compared with the TDC constants, to test the validity of the TDC theory in these cases. The magnitude of the amide I local mode transition dipole is as usual as-
230

O CH3 H CH3 H H C C N C C H3C N C C N C CH3 H H O H O O pep1 pep2 pep3 O C N H pepN CH3
zj
O H2N H H N H C CD3 CH3
xj
CH3
fragment unit peptide
Figure 1. Molecular structures of alanine-based polypeptides and the unit peptide.
sumed to be 2.73 D 1 amu1/2, its orientation is 10 away from the peptides C=O bond, and the origin of the transition dipole is located at 0.868 from the carbonyl (N ) carbon atom in the peptide bond. In Fig. 2(a), < V jk > denotes the average coupling constant between the jth and kth amide I local modes in polypeptide containing N peptide bonds. It was found that the coupling constants between two nearest neighboring (N ) amide I local modes, which are denoted as V j , j +1 for j = 1 to N1, are strongly dependent on the conformation (secondary structure) of the oligopeptides but not on the number of the peptide bonds in the oligopeptide. This indicates that the vibrational coupling between any two amide I local modes is highly local property determined by short-range interaction. As can be seen in Fig. 2(a), the TDC model works very well (N ) only for long-distance inter-peptide amide I vibrational couplings such as < V j , j + 2 > (N ) and < V j , j + 3 > . However, as expected, the TDC model significantly underestimates the coupling constants in PB, APB, and FEB oligopeptides, whereas it overestimates ) < V j(,N j +1 > value of 310H. Thus, the TDC approach should be used with care, if one is interested in quantitative (not just qualitative) description of vibrationally delocalized amide I normal modes and in simulating the linear and nonlinear vibrational spectra. We next consider diagonal force constants obtained by using the Hessian matrix reconstruction method. The diagonal force constants of the Hessian matrix represent the intrinsic vibrational frequencies of the amide I local modes. Torri and Tasumi in their numerical simulations of amide I IR spectra of a few globular proteins, used 1.605 mdyn 1 amu1 for the diagonal force constants of all peptide groups in several peptides, which corresponds to 1650 cm1 [54,84].
Fjj (in cm1) = 1302.79 F jj (in mdyn 1 amu1)
(5)
231
Ab-initio calculated coupling constants
12 8 4 0 -4
Local amide I mode frequency (cm )
-1
(a)
FEB APB PB 310H
LHH RHH
1760
(b)
1740
) V j(,N j +1
1720
RHH 310 H LHH H PB APB FEB
(N ) j, j +2
) V j(,N j +3
1700
dCO(NMA)
1680
-4 0 4 8 12 -1 Transition dipole coupling constants (cm )
1.185
1.190
dCO ()
1.195
1.200
1.205
Figure 2. (a) Ab initio calculated coupling constants are directly compared with those obtained from the TDC model. (b) Amide I local mode frequencies (diagonal Hessian matrix elements in cm1) are plotted with respect to the C=O bond lengths. For the sake of comparison, the C=O bond length of NMA is also shown in this figure. Adapted from Ref. [29].
Mendelshon and coworkers tried to re-adjust the diagonal force constants of amide I local modes when the corresponding peptide bond forms a hydrogen bond with other peptides, by subtracting an ad hoc value proportional to the hydrogen bonding interaction strengths [14,58]. There are however various factors affecting the diagonal force constants (amide I local mode frequencies) such as hydrogen-bonding interactions with either other peptide groups or solvent molecules. Due to the lack of a proper correction method for calculating such diagonal force constants, it was difficult to predict the amide I local mode frequencies in polypeptides and proteins in solution, even in the case when their structures are known. However, using the Hessian matrix reconstruction method combined with quantum chemistry calculations, we could obtain a systematic procedure to calculate the amide I local mode frequencies for any arbitrary polypeptides. In Fig. 2(b), a hint on this method can be found, where the carbonyl C=O bond lengths are plotted with the amide I local mode frequencies. The linear relationship between the two groups of data suggests that the amide I local mode frequencies can be estimated from the quantum chemistry calculations by merely examining the C=O bond lengths. Clearly, the amide I local mode frequencies are strongly dependent on the 3D structure of the polypeptide and furthermore the assumption that the diagonal Hessian matrix elements are constant is not valid at all. In addition, unlike the vibrational coupling constants that are more or less constant throughout the polypeptide chain when the secondary structure is fixed [49], the amide I local mode frequency is strongly dependent on the location of the corresponding peptide bond in a chain [29,30]. This indicates that the diagonal Hessian matrix elements are influenced by neighboring peptide groups (more specifically the backbone dihedral angles). Even when a given polypeptide is dissolved in solution, once we have a correction formula for predicting amide I local mode frequency of the target peptide bond, it becomes possible to quantitatively determine the amide I local mode frequency [30,4749,78]. 1.2. Alanine-Based Model Polypeptides The semiempirical (AM1) geometry optimizations and vibrational analyses of polypeptides were performed to study the amide I local mode frequency shifts, frequency-
232
(a)
amide I local mode frequency (cm )
1740
-1
(b)
Isolated N-methylacetamide (NMA)

1720 1700 1680 1660 1640 1620 1600
RHH 310H 9 13
APB PB PII 17 21
Ensemble averaged solvated NMA

1 5 9 13 17 21
Peptide number
Peptide number
Figure 3. (a) Amide I local mode frequencies of isolated polypeptides. Six different conformations, RHH (right-handed -helix), H (-helix), 310H (310 helix), APB (anti-parallel strand), PB (parallel strand), and PII (polyproline II), are considered. During the geometry optimization of polypeptide, the backbone structure is deliberately fixed. (b) Ensemble-averaged amide I local mode frequencies of polypeptides in water. Note that the amide I local mode frequencies are significantly red-shifted in comparison to the gasphase values in Fig. 3(a). Adapted from Ref. [49].
frequency correlations of all amide I local modes, vibrational dephasings, and average vibrational coupling constants and their fluctuation amplitudes. The six representative secondary structures, i.e., RHH (right-handed -helix), H (-helix), 310H (310 helix), APB (anti-parallel strand), PB (parallel strand), and PII (polyproline II; = 78 and = 149), were considered. In this subsection and throughout this chapter, we will consider the alanine-based model polypeptide with 21 alanine residues and the aminoand carboxy-terminals are capped with acetyl and N-methylamino groups, respectively. Consequently, the total number of peptide bonds is 22. First of all, the Hessian matrix reconstruction method and the scaling Eq. (4) were used to extract the amide I local mode frequencies and coupling constants from the AM1 results for the six cases when the backbone structure (dihedral angles) of the alanine-based polypeptide is fixed but all other degrees of freedom are relaxed. In Fig. 3(a), the amide I local mode frequencies for an isolated polypeptide are plotted to emphasize that they are strongly dependent on the position. Except for the residues near the two ends, the amide I local mode frequencies of inner peptide bonds are more or less constant, as expected. From the HMR analyses, we obtained the average vibrational coupling constants, <Vi,i+1>, <Vi,i+2>, <Vi,i+3>, and <Vi,i+4>, that are summarized in Table 1. Note that the vibrational coupling constant between two nearest neighboring amide I local modes is large for RHH and H polypeptides, which results in strong mode mixings and delocalizations of the amide I normal modes. A notable feature is that the vibrational coupling constant between the jth and (j+2)th amide I local modes is large for 310H, which can be understood by noting that the the jth amino acid forms an hydrogen bond with the (j+3)th residues in the 310H polypeptide. In Fig. 3(b) and Table 1, we also presented the data, such as solvatochromic frewater and average number of hydrogen-bonded water molecules quency shifts v N Hyd , for alanine-based polypeptides dissolved in water. However, detailed discussions on these results will be presented later in Section 2, after the description of the
233
Table 1. Solvatochromic local amide I mode frequency shifts and coupling constants in cm1 are presented. water and Vi , j are average values that are sampled from the 8 amide I local modes in the Here, v middle of the polypeptide chain. and angles for each conformation are given in the parenthesis of the first column. Adapted from Ref. [49]
v water
N Hyd
<Vi,i+1> 6.5 8.8 0.3 0.6 0.9 1.2
<Vi,i+2> 3.4 0.0 4.5 0.7 1.2 -0.9
<Vi,i+3> 4.4 3.2 0.2 0.2 0.2 0.2
<Vi,i+4> 0.1 4.4 0.5 0.0 0.1 0.1
RHH (57, 47) H (57, 70) 310H (49, 26) APB (139,135) PB (119,113) PII (78, 149)
19 18 23 46 48 51
0.8 0.6 1.0 1.6 1.7 1.7
amide I frequency correction method used to calculate the amide I frequencies of peptides in solution (not in a gas phase) is presented. 1.3. Ubiquitin In order to apply the computational methods mentioned above, the ubiquitin protein was chosen for detailed studies. It is a small globular protein with 76 peptides and contains various secondary structure elements such as an -helix, five-stranded mixed sheet, and reverse turns [85] (see Fig. 4). To calculate various vibrational properties such as amide I local mode frequencies, coupling constants, and dipole and rotational strengths of the amide I normal modes, AM1 semiempirical quantum chemistry calculation of an isolated ubiquitin was first performed. All amide I local mode frequencies and coupling constants are plotted in Figs 5(a) and 5(b), respectively. The average amide I local mode frequency was found to be 1667 cm1 and the standard deviation is about 23 cm1. The average amide I local mode frequencies of peptides in the , 1, 2, 3, 4 and 5 segments are estimated to be 1663, 1671, 1652, 1664, 1687, and 1680 cm1, respectively. It should be noted that the amide I local mode frequencies vary greatly so that the heterogeneity of diagonal elements of the vibrational Hamiltonian matrix in the amide I local mode basis should not be ignored. Furthermore, the amide I local mode frequencies of peptides even in a welldefined and relatively rigid segment of -helical or -sheet regions, have a significant distribution, which was found to be caused by inhomogeneous local electrostatic environments around peptide bonds. The vibrational coupling constants, Vj,j+1, between two nearest neighboring peptides in the -helix are, as expected, comparatively large. The other coupling constants, Vj,j+2, Vj,j+3, and Vj,j+4, are also large, which is consistent with the results found for model -helix with 21 alanine residues (see Table 1). The average values <Vj,j+1>,
234
[ Z \ X Y
Figure 4. Ribbon structure of ubiquitin. There are one -helical segment and five strands.
Figure 5. (a) Amide I local mode frequencies (closed squares) of an isolated ubiquitin and ensembleaveraged amide I local mode frequencies (open squares) of ubiquitin in water. (b) Vibrational coupling constants in cm1. Coupling constants between peptides in the -sheet strands can be found in the circles, whereas those between peptides in the -helix is shown in the black circle. Adapted from Ref. [30].
<Vj,j+2>, and <Vj,j+3> in this -helix are found to be 6.5, 3.4, and 4.4 cm1 (see the data points in the black circle of Fig. 5(b)). Also, as can be seen in the off-diagonal region of Fig. 5(b), the inter-strand amide I vibrational coupling constants between 3 and 5, between 1 and 5, and between 1 and 2 are notably large due to the -sheetforming hydrogen-bonds (see the data in the grey circles). Consequently, one can expect that there are amide I normal modes delocalized over the five -sheet strands. Using these results, we were able to simulate a variety of linear and nonlinear amide I IR spectra, which will be discussed later in this chapter. We next consider how to incorporate the solvation effects on amide I frequency shifts and fluctuations of polypeptides in solution.
2. Solvation Effects on Amide I Modes If proteins and polypeptides are dissolved in water, the peptide-water interactions induce a number of observable changes in the amide I vibrational spectra. First of all, the amide I IR peak frequency is red-shifted and its shifting magnitude increases as the solvent polarity increases. Secondly, the amide I band becomes broad, indicating peptide-solvent interaction-induced frequency fluctuation. Thirdly, the delocalized natures of amide I normal modes change upon solvation so that the vibrational lineshape be-
235
comes different from that of the isolated peptide. Also, due to the distributions of peptide conformation and of local environment around different peptides, the amide I frequency might have a significant inhomogeneity. As expected, one cannot use ab initio calculation methods to resolve these complicating issues originating from the solvation. Thus, a number of attempts and efforts have been made to resolve one or all of these issues. One of the most straightforward approaches is to use classical molecular dynamics simulation methods [86,87]. However, as well-known, the classical force fields are not accurate enough to provide quantitatively reliable results. We have therefore tried to develop a computational method combining the classical MD simulation with theory that takes into account the effects from the surrounding environment (including solvent molecules and other neighboring peptides and side chains) on the amide I local mode frequencies [77,79]. Note that such amide I local mode frequencies are the diagonal elements of the amide I vibrational Hamiltonian matrix. The off-diagonal Hamiltonian matrix elements, which are coupling constants, can be estimated from the vibrational coupling constant map obtained from detailed quantum chemistry calculations of model alanine dipeptide [32]. Thus, once the solvation effects on the amide I local mode frequencies of a given polypeptide or protein are understood, one can construct the amide I vibrational Hamiltonian matrix, which is in turn used to determined amide I normal mode frequencies, eigenvectors, transition electric and magnetic dipoles, and all other spectroscopic properties. Molecular dynamics simulation method has been extensively used to investigate conformational heterogeneity of polypeptide in solution. In particular, it was necessary to develop a proper theoretical method that can be used to predict amide I local mode frequencies. There have been numerous theoretical attempts to achieve this goal. One of the simplest approaches is to use the linear relationship between the amide I mode frequency shift I and the hydrogen-bond length r(OH) in , i.e.,
I = Hyd {2.6 r (O H )} ,
(6)
where the slope Hyd was typically assumed to be 30 cm1 [34,55]. Although this simple model is valid in a limited range of r(OH) near the optimum hydrogen-bond distance, it is of limited use due to the following reasons: (i) the amide I mode frequency shift observed when the N-H group of a given peptide is hydrogen-bonded to a water molecule cannot be described by this model; (ii) the fitting constant Hyd could vary with respect to hydrogen-bonding pair, (iii) it does not properly take into account the dependency of I on the orientation of the hydrogen-bonded molecule, and (iv) the amide I frequency is in fact non-linearly dependent on r(OH) [77,79]. Because of these limitations of the above empirical relationship (6), one cannot use it to accurately describe the amide I mode frequency shift in general for polypeptides dissolved in water. In order to calculate the solvation-induced amide I mode frequency shift and fluctuation, we have developed an electrostatic potential model, which is based on the theoretical consideration of inter-molecular interaction potential between solute and solvent molecules [68,77,88]. The amide I frequency shift was found to be related to the electrostatic potential at different sites of the peptide bond. Considering four atomic sites such as O(=C), C(=O), N(-H), and H(-N), we found that the amide I frequency shift
236
induced by the partial charge distribution of surrounding water molecules is given as [69,77]
water m (t ) = l j jwater ( m ) (t ) j =1 4
(7)
where j ( m ) is the Coulomb electrostatic potential at the jth site of the mth peptide bond at time t. Here, the partial charges of the deuterium and oxygen atoms of D2O are 0.412e and 0.824e, respectively, where e is the electronic charge. The parameters, l j , of the four sites, i.e., O(=C), C(=O), N(-H), and H(-N), were obtained before and they are lO = 0.00554e, lC = 0.0016e, lN = 0.00479e, and lH = 0.00086e [69]. These parameters were determined by carrying out systematic ab initio calculation studies on the amide I mode frequency shifts for a number of NMA-water clusters. The dimension of the electrostatic potential should be converted into cm1/e. Now, the amide I local mode frequency of the mth peptide bond is therefore given as a sum of two terms,
isolated water m (t ) = m + m (t )
water
(8)
isolated is the amide I local mode frequency of the mth peptide in an isolated where m polypeptide. Note that the local conformation-dependency of the mth amide I local isolated . The remaining contribution to mode frequency is fully taken into account by m the amide I mode frequency shift is thus solely from the solvation by water molecules. Once the snapshot configurations of the composite system with a polypeptide solute and water molecules are sampled from the MD trajectories, the solvatochromic amide I local mode frequency trajectories can be obtained with Eqs (7) and (8). The validity of the above method has been tested for various NMA-water clusters [77], aqueous NMA solution [79,89], NMA dissolved in methanol [80], and polypeptide [4749] and ubiquitin in water [30] by directly comparing simulated amide I IR absorption spectra and 2D IR spectra with experimental results. In relation to this approach, there exist a few different methods proposed and tested over the years. Instead of electrostatic potential calculations [9092], it was found that the NMA amide I frequency is in good correlation with electric field vectors, which is based on the vibrational Stark theory [93]. In the present section, the electrostatic potential approach designed to describe the solvatochromic amide I frequency shift is discussed for a few different systems.
2.1. NMA in Water The NMA has been served as a prototype molecule for peptide. By carrying out ab initio calculations of a variety of NMA-nD2O complexes, the l j parameters for six sites, which are O, C, N, H, CH3(N), and CH3(C), were determined by performing multivariate least square fitting analysis (see Fig. 6). In Fig. 6, the amide I frequency shift of the NMA, which is defined as the difference between the amide I frequency of an isolated NMA and that of NMA-water cluster, is directly compared with the leastsquare fitting result, and the correlation was found to be excellent for these cases.
237
Amide I mode frequency (cm-1)
1700 1680 1660 1640 1620

NMA in D2O solution
NMA-D2O(A) NMA-D2O(B) NMA-D2O(C) NMA-2D2O NMA-3D2O NMA-4D2O NMA-5D2O
1600 -110-100 -90 -80 -70 -60 -50 -40 -30 -20 -10
-1 Frequency shift (theory) (cm )
Figure 6. Multivariate least square regression analysis result for 96 NMA-D2O complexes. In this figure, the range of the amide I mode frequency shift of the NMA in liquid water is also drawn by extrapolating the multidimensional linear line. Since NMA has three hydrogen-bonding sites, NMA-D2O(m) for m = A, B, and C represents the case when the mth site is hydrogen-bonded to a D2O molecule. The other NMA-nD2O clusters were geometry optimized to calculate the amide I frequencies. Adapted from Ref. [77].
To obtain the amide I frequency trajectory for NMA in water, classical MD simulations of NMA in H2O and D2O were performed. For a given snapshot configuration of NMA-water solution, which is sampled from the MD trajectory, the instantaneous amide I frequency at time t can be calculated with the six-site model described above. The ensemble averaged amide I mode frequency shift was found to be 78 cm1 in comparison to that of the gas-phase NMA molecule [79]. This red-shift was found to be in excellent agreement with the experimental value of 81 cm1 [94]. From the amide I frequency trajectory, the frequency-frequency correlation function, I (t ) I (0) , was calculated to examine the underlying frequency component of coupled solvent water modes. I (t ) I (0) exhibited a bimodal decaying pattern, which is quite similar to the solvation correlation function of solute in water [79,95]. It was found that the librational motion, which is a hindered rotation of water molecule within the hydrogen-bond network, as well as the hindered translational motion play critical role in modulating the amide I frequency. The solvent isotope effect was found to be less important. Noting that the frequency-frequency correlation function is directly related to the vibrational pure dephasing, it was possible to estimate the pure dephasing constant of the amide I mode for the aqueous NMA solution, which is 11 cm1. Furthermore, from the lineshape analysis of the simulated amide I IR band, the vibrational broadening mechanism is largely determined by the motional narrowing process [79,96]. By using the semiclassical theory, the amide I IR spectrum was successfully calculated without any adjustable parameters except for the lifetime and its full width at half maximum was estimated to be about 27 cm 1 [79], which agrees to the experimental value [94,96]. Motivated by the success of the electrostatic potential model, numerical simulations of 2D IR pump-probe spectrum of the NMA-water solution was performed and directly compared with the corresponding experimental result by Hamm and coworkers [87]. The same model was also used to quantitatively describe amide I IR absorption and 2D IR photon echo spectra of NMA in other polar solvents [80,88,89].
238
2.2. Alanine-Based Model Polypeptides Upon hydration of peptide, the amide local mode frequencies and vibrational coupling constants fluctuate in time due to the peptide-water interaction and conformational fluctuation of the polypeptide. However, it has been shown that the rigid polypeptide simulation can provide reliable results for simulating the IR, VCD, and 2D IR photon echo spectra of various different secondary-structure polypeptides as well as ubiquitin [30,48,49,78]. This suggests that the fluctuation amplitudes of amide I local mode frequencies and coupling constants induced by the conformational fluctuation are relatively small and that the dominant factor determined amide I coherence decay is from the peptide-water interactions. First, to examine the absolute magnitude of solvatochromic amide I frequency redshift induced by solvation, for six representative secondary-structure polyalanines, water = we calculated the locally-averaged frequency shifts defined as 15 water (1 8) m =8 m (see Table 1). Note that only the inner eight amide I local modes among 22 amide I local modes are taken into consideration for calculating the average water value. In the cases of the three helical peptides such as RHH, 3 10H, and H, values are about 20 cm1, whereas those of three extended structure polypeptides such as APB, PB, and PII are about 50 cm1 (compare Figs 3(a) and 3(b)). In the cases of the helical polypeptides, the peptide groups are already in the hydrogen-bond interactions with other peptide groups so that the additional frequency red-shift induced by hydration is relatively small. On the other hand, the three extended-structure polypeptides are significantly exposed to surrounding water molecules so that the frequency red-shift induced by hydration should be large. An evidence for this interpretation can be found by examining the average number of hydrogen-bonded water molecules < N hyd > around each peptide group (see Table 1). Indeed, < N hyd > of helical polypeptides is less than or equal to 1, whereas that of extended-structure polypeptide is significantly larger than 1. To quantitatively determine m (t ) values of all amide I local modes, we therefore used Eq. (8) to describe the solvation-induced amide I frequency shift as well as to include the inter-peptide hydrogen-bonding effects on the amide I local mode frequency shift. 2.3. |biquitin Unlike polypeptide with a well-defined secondary structure, a real protein consists of a number of structure elements and each peptide bond has different local environment due to solvation and interactions with other peptides and amino acid side chains. In order to study such inhomogeneity effects on amide I frequencies and coupling constants, constrained MD simulations of ubiquitin in water were carried out with the AMBER suite of program with ff99 force field parameters. Here, the backbone structure of ubiquitin in water, which was obtained from the semiempirical quantum chemistry calculation, was fixed, whereas all other degrees of freedom such as amino acid side chain and water molecules were allowed to move during the MD simulation. In Fig. 5(a), the ensemble-averaged amide I local mode frequencies, < m > for m = 1 ~ 75, are plotted (open squares in this figure). As expected, the amide I local mode frequencies become strongly red-shifted due to the electrostatic interaction of
239
solvent-exposed peptides with surrounding water molecules. The average amide I frewater , is 18 cm1. For the -helical peptides in ubiquitin, quency red-shift, water 1 , is 7 cm , and it is about 10 cm1 for peptides in the five strands forming water for those in the seven turns (710, 1821, the -sheet. On the other hand, 3740, 4647, 5154, 5760, 6264 residues) is 28 cm1. This relatively large frequency shift of the amide I local modes in the turn regions of ubiquitin is because the corresponding peptides are more exposed to solvent water than those in the inner part of ubiquitin. The results and data discussed in this section were directly used to construct the amide I vibrational Hamiltonian matrix, i.e.,
1 (t ) V1,2 V 2 (t ) H (t ) = 1,2 V2, N V1, N V1, N V2, N N (t )
(9)
The Hamiltonian matrix is then divided into two parts as

H (t ) = H 0 + H (t ) ,
(10)
where H 0 is the time-averaged matrix and H (t ) is the fluctuating part. The amide I normal mode frequencies and eigenvectors are obtained by diagonalizing H 0 . The fluctuating part H (t ) is then responsible for amide I vibrational dephasing processes. 3. Fragmentation Approximation Once the amide I vibrational Hamiltonian matrix is determined and the vibrational transition frequencies and fluctuating frequency-frequency correlation functions are calculated, the next step toward numerical simulations of vibrational spectra is to calculate the electric and magnetic dipole moments associated with the amide I vibrational transitions. In this section, we present a discussion on the essential part of the fragmentation approximation method [32,33], and show that the amide I transition electric and magnetic dipole moments can be written as linear combinations of those of properly chosen unit peptides (fragments). The first step is to obtain atomic axial tensor (AAT) and atomic polar tensor (APT) elements of unit peptide, which are needed in the calculations of transition electric and magnetic dipole matrix elements. Taking into account the origin- and coordinatedependencies of electric and magnetic dipoles of each fragment peptide, we developed a theoretical procedure to convert the AAT and APT tensor elements of the N fragment peptides into those in the global molecular coordinate frame. Now, the stepwise procedure is summarized below: Step 1. Although the global molecular (X, Y, Z) Cartesian coordinate frame can be chosen arbitrarily, it was assumed that its origin is located on the carbonyl carbon atom of the first peptide near the amino-terminal. The X-axis is on the C=O bond of the first
240
peptide and the direction from oxygen atom to carbon atom is defined as the positive direction, the three atoms O=C-N in the first peptide bond are placed on the X-Y plane and the Y-axis points toward the nitrogen atom from the carbonyl carbon atom, and a right-handed coordinate system is used. Step 2. The origin of the Cartesian coordinate frame defining the orientation of the jth fragment peptide is also assumed to be on the carbonyl carbon of the jth peptide. Step 3. Since the origin of the jth local coordinate frame is displaced from that of the global coordinate frame, the transition magnetic dipole of the jth amide I local mode in the global coordinate frame should be treated with care. Thus, we have
m q j
glb
m = q j
loc
i 4 c
( j)
q j
loc
(11)
q j
glb
= q j
loc
(12)
where Y ( j ) is the displacement vector from the origin of the global coordinate frame to that of the jth local coordinate frame. m and are the transition magnetic and electric dipole moments, respectively, and qj is the amide I local mode coordinate of the jth peptide. Step 4. Within the fragmentation approximation scheme, the transition electric and magnetic dipole vectors of the kth amide I normal mode are then given as
/ Q = U j , ( / q j )glb
j
m / Q = U j , (m / q j )glb
j
(13)
where U j , is the jth eigenvector element of the th amide I normal mode, which was obtained by diagonalizing the amide I vibrational Hamiltonian matrix. The th amide I normal (not local) mode coordinate was denoted as Q. The results in Eq. (13) have been used to calculate the simulated amide I IR, VCD, and 2D IR spectra. The validity of the fragmentation approximation method was tested for dipeptide with various conformations [32] and polypeptides [33] by directly comparing the magnitudes of transition electric and magnetic dipoles of amide I normal modes with ab initio calculated values. A few comparative investigation results are discussed below.

Alanine dipeptide analog (ADA)
O H 3C C N H H C CH 3 C O H N CH 3
241
O H 3C C N H
H C
CH 3 CD 3 H 2N
H C
CH 3 C O
H N CH 3
Fragment 1
Fragment 2
Figure 7. Molecular structures of alanine dipeptide analog (ADA) and fragments 1 and 2.
3.1. Dipeptide In order to apply the fragment approximation method to the calculations of dipole and rotational strengths of amide I vibrations in alanine dipeptide analog (ADA) in Fig. 7, we assumed that the ADA can be dissected into two unit peptides, fragments 1 and 2. The fragment 1 has a single chiral group and the dihedral angle determines its structure. The fragment 2 also contains a chiral carbon and its structure is determined by the angle. By considering these two monopeptides separately and by carrying out ab initio vibrational analyses, it was shown that a variety of vibrational spectroscopic properties such as amide I local and normal mode frequencies, coupling constant, and dipole and rotational strengths of the ADA can be quantitatively calculated by combining those quantities of the two fragments. In Fig. 8, the dipole and rotational strengths of the symmetric and asymmetric amide I normal modes of the ADA for varying and angles, which were obtained by using the fragmentation approximation method, are directly compared with the DFTcalculated values. There appear reasonably good correlations. 3.2. Oligopeptides Although the ADA is a good model dipeptide system, the amino- and carboxyterminals are capped with acetyl and N-methylamino groups. Therefore, the two fragments in Fig. 7 are not useful for simulating amino acid residue in a polypeptide. Consequently, a more generalized fragment shown in Fig. 1 was considered in detail. Note that the structure of the generalized fragment (with two chiral carbons) in Fig. 1 is determined by both and angles. Therefore, we needed to have the transition electric and magnetic dipole maps with respect to and . Then, invoking the fragmentation approximation, one can calculate the transition dipoles of amide I normal modes in any arbitrary polypeptide by using the transition dipoles of the fragment unit peptide. Note that the polypeptide is viewed as specifically aligned N fragment unit peptides. The fragmentation approximated transition dipoles of 28 different alanine oligopeptides with two to five peptide bonds were directly compared with ab initio calculation results. The agreement was found to be acceptable [33].
242
1000
(a)
Dipole strength
DFT result (10 esu cm )
2
600 400 200 0 -200 -400 -600
(b) Rotational strength
DFT result (10 esu cm )
800 600 400 200 0 0 200 400
-40
-44
Symmetric Normal Mode Asymmetric Normal Mode

600 800
Symmetric Normal mode Asymmetric Normal mode
1000
-600
-400
-200
200
400
600
Theory (fragment analysis)
Theory (fragment analysis)
Figure 8. The DFT-calculated dipole and rotational strengths are compared with fragment-approximated values in (a) and (b), respectively. Changing the Ramachandran angles and of the ADA, we calculated the dipole and rotational strengths of the ADA. To obtain the same quantities by using the fragmentation approximation method, we calculated the transition electric and magnetic dipoles of the two fragments in Fig. 7 for varying and angles. Then, Eq. (13) was used to calculate the transition electric and magnetic dipoles of the symmetric and asymmetric amide I normal modes of the ADA that are linear combinations of fragments transition electric and magnetic dipoles. Adapted from Ref. [32].
However, most of the oligopeptides considered assumed to be in one particular conformation. That is to say, the and angles throughout a given oligopeptides are constant. However, a real protein usually does not have any repeating structural motifs. All s and s in a real protein are likely to be different from one another. Thus, it was necessary to examine the validity of the fragmentation approximation method for more general polypeptides. In this regard, three different segments taken from the globular protein ubiquitin, i.e., right-handed -helix, -sheet, and -turn segments, were considered. First of all, for fixed backbone configurations, density functional theory calculations of the three segments were performed. Vibrational analyses at the same level of calculation provided amide I normal mode frequencies, eigenvectors, and transition dipoles. These quantities were used to numerically simulate the amide I IR and VCD spectra (see solid lines in Fig. 9). The corresponding spectra obtained by using the fragmentation approximation method are also plotted in Fig. 9 (see dashed lines). Even for these three segments having no repeating backbone structures, we could confirm that the fragmentation approximation method for calculating transition electric and magnetic dipoles works well. By combining the theory and models outlined so far, such as Hessian matrix reconstruction, electrostatic potential model for describing the solvatochromic amide I frequency shift, fragmentation approximation method for calculating transition electric and magnetic dipoles of amide I normal modes, it is now ready to simulate the amide I IR, VCD, and 2D IR photon echo spectra of any arbitrary polypeptide and protein in solution. The general computational procedure is presented as a flow-chart in Fig. 10.
243
IR
DFT calculation Fragment approximation
VCD
-helical segment
-sheet segment
-turn segment
1620
1670
1720
Frequency (cm )
-1
1770
1620
1670
1720
1770
-1
Frequency (cm )
Figure 9. The density functional theory (DFT) calculated IR and VCD spectra (solid lines) for the three segments representing the -helical, -sheet, and -turn structures in ubiquitin protein are compared with fragmentation-approximated spectra (dashed lines). The DFT-calculated optimized structures of three segments sampled from ubiquitin are also shown in each figure. Adapted from Ref. [33].
4. Simulations of Amide I IR and VCD Spectra The IR and VCD spectra of polypeptide having N amide I normal modes can be obtained by taking Fourier transformations of the weighted linear response functions as [97]
I IR ( )

dt eit D e i t J (t )
=1
(14)
244
Vj,j+1 from HF/DFT of model dipeptides Vj,j+n (n>1) with TDC model
HF/DFT + HMR (oligopeptides)
Theoretical model (electrostatic potential model)
Vibrational coupling constants

Off-diagonal Hessian matrix elements
Amide I local mode frequencies

Diagonal Hessian matrix elements
Amide I Hessian matrix

Matrix diagonalization
Eigenvectors (Uj,k) and frequencies of amide I normal modes
Dephasing models
Transition electric and magnetic dipoles of amide I normal modes
IR and VCD spectra
Transition electric and magnetic dipoles in the global coordinate frame

Origin and coordinate transformations
Transition electric and magnetic dipoles in a local coordinate frame
Ab initio calculations of the unit peptide Fragmentation Approximation Method
Figure 10. Schematic diagrams used to numerically simulate the IR and VCD spectra of oligopeptides and proteins. Amide I vibrational coupling constants can be calculated directly from the HMR analysis of ab initio calculated polypeptide or from the ab initio calculated coupling constant map for model dipeptide with TDC model for long range coupling constant. Adapted from Ref. [33].
IVCD ( )
dt eit R e i t J (t )
=1
(15)
where is the ensemble-averaged th amide I normal mode transition frequency. The dipole and rotational strengths, D and R, are defined as
D ( Q )
2
(16)
R Im ( Q ) ( m Q ) .
(17)
The linear response function J(t) describing vibrational coherence decay of the th amide I normal mode is given as
245
t J (t ) exp + i d ( ) 0
(18)
where ( ) is the fluctuating part of the th amide I normal mode frequency. denotes the average over the bath degrees of freedom. The second-order cumulant approximation was used to rewrite the linear response function J(t) in terms of the frequency-frequency correlation function as
J (t ) = exp d 2
0
2 0
d 1 ( 1 ) (0)
}.
(19)
To calculate < (t ) (0) > , let us consider the amide I vibrational Hamiltonian matrix written as a sum of ensemble (time) averaged part and fluctuating part, as H (t ) = H 0 + H (t ) . Diagonalization of the reference Hamiltonian matrix, H 0 , i.e., U 1 H 0U = , provides eigenvectors. Then, the fluctuating part of the th amide I 2 normal mode frequency can be written as (t ) = 2 c m (t )U m , , where m (t ) is the fluctuating mth amide I local mode frequency,m i.e., m (t ) = m (t ) < m > . Combining the classical MD simulation with the electrostatic potential model, one can obtain the frequency trajectories for all amide I local modes in a given polypeptide. Then, we have [78]
2 2 (t ) (0) = 4 2 U m , U n , m (t ) n (0) m,n
4 4 2 U m , m (t ) m (0) m
(20)
where the cross-correlation between the mth and nth (for m n) amide I local mode frequency fluctuations was assumed to be negligibly small [47,78]. The lifetime broadening was included in an ad hoc manner by multiplying exp [ t / 2T1 ] function to the linear response function, where the lifetime of all amide I normal mode excited states was assumed to be 0.33 ps. We have calculated all frequency-frequency correlation functions, m (t ) m (0) for m = 1~22 of alanine polypeptides and for m = 1~75 of a ubiquitin protein, to eventually obtain 22 and 75 correlation functions of (t ) (0) , respectively. 4.1. Alanine-Based Model Polypeptides In the previous sections, we considered alanine-based model polypeptides with 22 peptide bonds. Six different conformers were specifically studied. From the constrained MD simulations, it was possible to calculate all 22 frequency-frequency correlation functions, m (t ) m (0) (for m = 1~22). They are inserted into Eq. (20) to obtain (t ) (0) s. By using Eqs (14) and (15) with dipole and rotational strengths for
246
(a)
IR
RHH 310H APB PB PII
(b)
VCD
1620
Frequency(cm )
1650
1680
-1
1710
1620
Frequency(cm )
1650
1680
-1
1710
Figure 11. Simulated amide I IR (a) and VCD (b) spectra of six different secondary structure polypeptides in water. Adapted from Ref. [49].
all amide I normal mode transitions, it was possible to simulate amide I IR and VCD spectra of the six model polypeptides in water (see Figs 11(a) and 11(b)). The amide I IR spectra of RHH, APB, PB, and PII peptides appear to be broad and featureless, whereas the IR spectra of H and 310H exhibit a doublet pattern. As discussed before, the two peaks in the H and 310H spectra represent the A- and E-modes [49]. The A-symmetric amide I normal mode is strongly IR-active and its transition dipole is parallel to the helix axis so that it was denoted as -mode. On the other hand, the two-fold degenerate E-symmetric amide I normal modes have transition dipole vectors perpendicular to the helix axis so that they are -modes. We shall use these notations, and , later in Section 5. Although an ideal RHH has such Aand E-symmetric amide I normal modes, if the two mode frequencies are close to each other, the corresponding amide I IR spectrum could appear as a singlet instead of a doublet. Nevertheless, it is clear that the amide I IR spectroscopy itself is not enough to precisely and uniquely determine secondary structure of polypeptide in general. In contrast, the VCD spectra in Fig. 11(b) appear to be distinctively different from one another. Experimentally, it was found that the VCD spectra of RHH and PII polypeptides exhibit (,+,) and (,+) peak patterns, respectively. Indeed, our simulation results reproduce such spectral patterns quite well. Interestingly, the three extended structure polypeptides such as PII, APB, and PB, show the same (,+) pattern, when the spectrum is read from low to high frequency region. Therefore, one can use the VCD measurement technique to distinguish the helical conformations from the extended ones. Secondly, the VCD intensity of the PII polypeptide appears to be much larger than those of the other two extended conformations, i.e., APB and PB. These quantitative difference in the VCD intensity distribution can be understood by noting that, as the polypeptide and angles approach 180 degrees, (i) the polypeptide backbone structure becomes planar, (ii) vibrational coupling constants become small, (iii) mode mixing strengths are weak, (iv) electric dipole coupling-induced magnetic dipoles become small, and thus (v) the overall molecular chirality associated with the amide I VCD transition becomes small [49].

t h e o r y ( w it h h e t e r o g e n e it y ) t h e o r y ( w it h o u t h e te r o g e n e it y ) e x p e r im e n t
247
(b )
(a )
1639
1673
c a
1570 1620 1670 1720 1570 1620
-1
b )
1670 1720
F re q u e n c y (c m
Figure 12. Amide I IR (a) and VCD (b) spectra of ubiquitin in water. The numerically simulated and experimentally spectra are solid and dashed lines, respectively. The dotted lines correspond to the numerically simulated amide I IR and VCD spectra, when the amide I local mode frequencies are all assumed to be 1650 cm1. Adapted from Ref. [30].
4.2. Ubiquitin Using the same procedure, we calculated the amide I IR spectrum of ubiquitin (solid line in Fig. 12(a)). In this figure, we also plot the experimentally measured amide I spectrum (dashed line) [30]. In the experimental spectrum, the peak frequency is 1639 cm1 and a shoulder band appears at 1673 cm1. The presence of these two bands is the characteristic feature that the protein contains significant portion of anti-parallel -sheets. In our simulated spectrum, though the overall band is slightly blue-shifted, one can find a similar asymmetric line shape though the high-frequency shoulder band is less distinctive. In the same Fig. 12(a), we also plot the simulated IR spectrum (dotted line) when the diagonal Hessian matrix elements, amide I local mode frequencies, are all assumed to be 1650 cm1. This is the case when one completely ignores the sitefrequency distribution intrinsically present in a real protein. The amide I band in this zero diagonal heterogeneity limit is much narrower than the experimental spectrum. This indicates that the amide I local mode frequency inhomogeneity has to be correctly taken into consideration to obtain quantitatively reliable amide I IR spectrum. This becomes quite clear when the experimentally measured VCD spectrum of ubiquitin is compared with numerically simulated VCD spectra with and without including the diagonal heterogeneity. In Fig. 12(b), we compare the numerically simulated VCD spectrum (solid line) with experiment (dashed line). As shown in Fig. 4, ubiquitin consists of an -helix and significant portion of -sheets. Usually such kind of + protein having both -helix and -sheet segments, e.g., ovalbumin and trypsin, exhibits W-form (,+,) peak pattern in the VCD spectra, where the low-frequency negative peak at ~1630 cm1 was typically attributed to the amide I modes of constituent -sheet peptides, the weak positive peak at ~1640 cm1 and the second negative peak at ~1660 cm1 were to the amide I modes of -helix [98,99]. However, such mode assignments are not always valid. The ubiquitin VCD spectrum in Fig. 12(b) appears to be different from those proteins having approximately 50:50 -helix and -sheet peptides. The overall VCD lineshape is rather close to (,+)
248
pattern, and it is similar to those of twisted -sheets, -hairpin, and -turns [75,76]. This is likely because the relative percentage of -helical peptides is comparatively small. When the experimentally measured VCD spectrum (dashed line) is compared with the simulated one (solid line), there are some differences. Although the experimentally measured spectrum has a large positive peak at 1665 cm1, our simulated spectrum does not. Secondly, the simulated spectrum shows three negative peaks, but the experimental one has only two. Although we could approximately assign the three notable features, denoted as a, b, and c bands in Fig. 12(b) to corresponding peaks in the experimental spectrum, it is necessary to develop far more accurate computational method for simulation of VCD spectrum of real protein. Finally, we performed numerical simulation of the ubiquitin VCD spectrum with the assumption that the diagonal heterogeneity of the amide I local mode frequencies is zero. We deliberately put 1650 cm1 for all amide I local mode frequencies, but used the same vibrational coupling constants and computational procedure. The dotted line in Fig. 12(b) corresponds to the ubiquitin VCD spectrum in this limit of zero diagonal heterogeneity. It appears to be completely featureless with a single negative peak. Furthermore, the overall VCD band width is far narrower than that of the experimentally measured spectrum. This suggests that the diagonal heterogeneity or site-dependency of the amide I local mode frequency should be properly taken into consideration to successfully simulate the amide I VCD spectrum of protein. 5. Two-Dimensional Vibrational Spectra A brief account of the time-correlation function formalism for 2D IR spectroscopy, which have been developed recently, will be given in this section note that far more detailed discussion on theoretical aspects of coherent 2D IR spectroscopy of peptides can be found in the review article [51]. Then, numerical simulation results for a few cases of polypeptides and protein will be presented to show the advantages of 2D vibrational spectroscopic techniques [30,49]. The 2D IR photon echo spectroscopy employing spectral interferometric heterodyne-detection method requires three temporally separated IR pulses to generate two-dimensional vibrational coherences [100,101]. Yet another IR pulse, which is a reference pulse, is used to detect the photon echo field via spectral interferometric detection. The delay times between the first two pulses and between the second and third pulses are controlled to be and T, respectively (see Fig. 13). The generated echo field at time t after the third pulse can be indirectly detected via inverse Fourier transformation of the measured spectral interferogram. Then, the double Fourier transformations of the echo signal with respect to and t provides the 2D IR photon echo spectrum. Here, the conjugate Fourier frequencies are denoted as and t, respectively. If the polarization directions of the three injected pulses used to generate IR photon echo field are all parallel to the laboratory Z axis and if the Z-component of the photon echo signal field vector is detected, the rephasing 2D photon echo spectrum is given as [102]
1 (1) diag S ( , T , t ) Re S + (1) S cross + (2) S 5
(21)
249
solution sample k3 ks=-k1+k2+k3 photon echo k2 k1
Figure 13. Experimental beam configuration for photon echo spectroscopy. The wavevector of the jth IR pulse is kj. The phase-matched echo field propagates with wavevector k1 + k2 + k3. The echo field is then combined with a reference pulse to generate the interference signal. The interfering field is injected into a dispersive device and the spectral interferogram is recorded. Note that the spectral interferogram contains information on the spectral components of the echo signal field.
where
(1)
S diag = e j g
j
(R
SE ejej
( , T , t ) + ReGB e ( , T , t ) )
j j
(1)
1 S cross = e j g 3 j k j
ek g
( 2 cos
2
ek , e j
SE GB +1 Rek e j ( , T , t ) + Rek e j ( , T , t )
(2)
1 S = fl e j fl ek l j k 3
{(
)(
ek g
e j g + fl e j ek g
) (
)(
fl ek
e j g
)
(22)
+ fl e j e j g
)(
fl ek
ek g
)} R
EA fl ek e j
( , T , t ) .
Here, the detailed expressions of the 2D frequency-domain response function compo(n) nents, Re e ( , T , t ) , can be found in reference [78]. Here, the superscripts SE, GB, and EA are the stimulated emission, ground-state bleaching, and excited-state absorption, respectively. In the case of the photon echo measurement, during the first delay time , the system is on a vibrational quantum coherence state, i.e., a superposition state of the ground state and a singly-excited state. Therefore, the spectrum along the -axis in a given 2D spectrum provides information on the resonant vibrational frequencies associated with the transitions from the ground state to one of the singlyexcited states. During the waiting time T, either vibrational coherences between two different singly-excited states or populations on singly-excited states evolve in time. Finally, during t, the system is again on a vibrational quantum coherence state. Not only the transitions between the ground and a singly-excited state but also those bej j
250
tween singly-excited states and doubly-excited (overtone or combination bands) states are allowed. Thus, the spectrum along the t-axis in a given 2D spectrum provides useful additional information on the resonant vibrational frequencies between the singlyand doubly-excited states. The singly-excited states are first excited states of the amide I normal modes. For the polypeptide with N amide I local modes (peptide bonds), there are N singly-excited states. Unlike the linear spectroscopy such as IR and VCD, the 2D IR spectroscopy involves transitions from singly-excited states to doubly-excited states, where the latter states are given as linear combinations of overtone states of amide I local modes and the product states of two different amide I local mode excited states. Hereafter, the singly-excited states are denoted as ej and the doubly-excited states are as fk. Thus, the summations over j and k in Eqs (22) mean that the singly- and doublyexcited states are to be included in the calculations of the 2D IR spectrum. In the case of the IR spectroscopy, the peak intensity is determined by the dipole 2 strength such as | e j g | . In contrast, the VCD peak intensity is determined by not only the transition electric and magnetic dipoles but also the angle between the two dipole vectors of the same jth singly-excited state. However, the 2D IR spectrum is additionally dependent on the angle between the two different transition dipole vectors, e j g and ek g , which is denoted as ek , e j . Therefore, these angle-dependencies of 2D IR spectrum provide additional information on the detailed 3D molecular structure. A notable application was that the polarization-controlled 2D IR spectroscopy can be used to distinguish anti-parallel -sheet structure from parallel -sheet structure because ek , e j for two strongly IR-active amide I normal modes do strongly depend on the detailed polypeptide conformation [102]. As briefly mentioned above, during the experimentally controllable delay time T, the molecular system evolves on the population or coherence states within the singly(1) diag excited state manifold. In Eq. (21), the first term S ( , T , t ) produces positive peaks along the diagonal line in the 2D spectrum the superscript (1) emphasizes (1) diag that the nonlinear transition pathways contributing to S (1 , T , 3 ) involve transitions between the ground state and singly-excited states. The amplitude of the positive diagonal peak at ( = e j g , t = e j g ) is determined by the square of the dipole (1) cross ( , T , t ) is associated with the vibrational strength, i.e., | e j g |4 . Secondly, S transition pathways that produce cross peaks in the off-diagonal region of the 2D spectrum. As can be inferred from Eq. (22), the magnitude of the cross peak at ( = e j g , t = ek g ) is proportional to | e g |2 | e g |2 (2 cos 2 e ,e + 1) . Therefore, in j k j k order for the cross peak to be measurably large, the corresponding two dipole strengths should be large. Comparing the cross peak amplitude at ( = e j g , t = ek g ) and the ( = ek g , t = ek g ) , one can extwo diagonal peaks at ( = e j g , t = e j g ) and 2 (2 cos e j , ek + 1) in the ideal case when the tract information on the magnitude of cross and diagonal peaks are frequency-resolved. Then, the angle e j ,ek thus determined can be of use to determine the peptide backbone structure. Finally, the third term (2) S ( , , t ) in Eq. (21) is associated with pathways involving transitions from singly-excited states to doubly-excited states. This contributes to the 2D spectrum negatively, because it involves an absorptive field-matter interaction, i.e., excited-state absorption. Due to the intrinsic vibrational anharmonicities in the multidimensional potential energy surface of the amide I vibrations, the third term does not completely cancel out the first and second terms. Therefore, the diagonal and cross peaks naturally exhibit positive-negative features when the signal is measured at the amplitude-level.
251
In addition to the parallel polarization (ZZZZ-component) signal, the perpendicular polarization detection scheme has also been used, where the [ZXXZ] tensor element of the photon echo response function is detected note that the beam propagation directions are assumed to be along the Y-axis in the space-fixed frame. The perpendicular polarization rephasing photon echo spectrum was found to be[102]
S ( , T , t ) 1 diag cross Re (1) S + (1) S + 15
(2)
S ,
(23)
where
(1) diag S = (1)
S diag
2 2
(1)
cross S =
1 e jg 2 j k j
ek g
(3cos
2
ek , e j
SE GB 1 R ek e j ( , T , t ) + R ek e j ( , T , t )
(2)
1 S = 4 fl e j e j g l j k j 2
{(
)(
ek g
fl ek fl e j ek g
) (
)(
fl ek
e j g
)
(24)
fl e j fl ek
)(
ej g
ek g
)} R
EA fl ek e j
( , T , t ) ) .
Note that, due to the differences in the rotational averages of the fourth-rank tensorial nonlinear response function for randomly orientated molecules in solution, the two signals, S ( , T , t ) and S ( , T , t ) , slightly differ from each other. First of all, the diagonal components are identical 2except for a constant factor, i.e., 1/5 vs. 1/15. Secondly, the terms involving cos ek ,e j are different for S ( , T , t ) and S ( , T , t ) . Also, the third terms in Eqs (21) and (23) are different from each other. S ( , T , t ) = Therefore, the difference 2D spectrum defined as S ( , T , t ) 3S ( , T , t ) does not have contributions from the diagonal peaks. Experimentally one can directly measure the difference 2D spectrum by controlling the polarization directions to be [0, /3, /3, 0] or [/4, /4, /2, 0] note that the latter polarization scheme can remove the diagonal peaks even when there is a rotational motion-induced relaxation process as shown in reference [103]. The difference 2D IR spectrum S ( , , t ) is then given as
S ( , T , t ) = 1 e g 6 j k j j
2 (2) (3) ek g sin 2 ek ,e j Rek e j ( , T , t ) + Rek e j ( , T , t ) 2
252
1 fl e j fl ek 6 l j k
{(
)(
ek g
e j g + fl e j ek g
) (
)(
fl ek
e j g
)
(25)
2 fl e j e j g
)(
fl ek
ek g
)} R
(5) fl ek e j
( , T , t ) .
It should be emphasized that the cross peak amplitude in a given difference 2D spectrum is mainly determined by the factor, jk , defined as
jk e j g
2
ek g sin 2 ek , e j ,
(26)
which is product of the corresponding two dipole strengths and angle-dependent sin2term. The partial cancellation (destructive interference) between the two terms in Eq. (25) reduces the actual cross peak amplitudes. However, as demonstrated recently for distinguishing anti-parallel and parallel -sheet polypeptides [102], the dominant features in a given difference 2D spectrum are still determined by jk values. This is quite important in quantitatively describing the relative amplitudes of the cross peaks that are critically dependent on the secondary structure of alanine-based polypeptide. 5.1. Alanine-Based Model Polypeptides Numerically simulated 2D IR spectra S ( , t ) of the six polypeptides are shown in Fig. 14. One can clearly see the differences in the overall 2D line shapes of these polypeptides. Let us consider the three helical structures (RHH, H, and 310H) and the corresponding 2D spectra in the first row of Fig. 14. The S ( , t ) spectrum of RHH exhibits a single diagonal peak consisting of positive and negative peaks. Note that the frequency difference between the positive and negative peaks is caused by the overtone anharmonicity (~16 cm1). In contrast, the S ( , t ) spectra of H and 310H show two frequency-resolved diagonal peaks. These two peaks are associated with the A- and E-symmetric amide I normal modes. Due to the enhanced frequency resolution of the 2D IR spectroscopy, the peak-to-peak frequency difference can be accurately measured. Recently, Ge and coworkers demonstrated that the doublet feature in the amide I band of homo-octapeptide, Z-[L(Me)Val]8-OtBu, which has been known to form 310-helix conformation, can be observed with the 2D IR spectroscopic method [39]. Although the three 2D spectra of RHH, H, and 310H in Fig. 14 appear to be distinctively different, the 2D spectra of APB, PB, and PII are strikingly similar to one another. Furthermore, the relative amplitudes are also indistinguishably similar too. Therefore, the parallel-polarization 2D IR spectroscopic technique is not quite useful for distinguishing one extended structure from others.
253
Figure 14. Parallel-polarization 2D IR photon echo spectra S (1 , 3 ) of six different secondary structure polypeptides. The maximum peak amplitude of the 310H spectrum is assumed to be one, and all other spectra are rescaled with this maximum value for the sake of quantitative comparisons. Adapted from Ref. [49].
The origin of the different structure-spectrum relationships observed in Fig. 14 can be understood from detailed analyses of delocalized amide I normal modes. In the cases of the three extended structures, the vibrational coupling constants are relatively small so that the amide I normal modes are not as delocalized as those of three helical polypeptides (see Table 1). As the inter-peptide amide I vibrational coupling constants increase, the amide I normal modes are increasingly delocalized. Then, the couplinginduced frequency splitting is large for RHH, H, and 310H, which makes the 2D spectrum comparatively complicated and sensitive to detailed conformation. Despite that the S ( , t ) spectra show some dependency on the polypeptide secondary structure, the difference 2D spectra S ( , t ) should in principle provide far more revealing information on the conformation because the cross peaks without diagonal peak contributions are directly dependent on the structure-sensitive e j ,ek angles. In Fig. 15, the S ( , t ) spectra are plotted and the amplitudes of these spectra are rescaled by assuming that the maximum amplitude of the 310H S ( , t ) spectrum equals 1. There are a few notable features in the S ( , t ) spectra of different polypeptides. First of all, the amplitudes of the S ( , t ) spectra of RHH and H are much smaller than that of the 310H. Secondly, the two cross peaks are clearly visible in the 310H spectrum. A key to understand these two differences is the factor, jk , defined in Eq. (21). 2 2 and of the two most strongly IRIn Table 2, the dipole strengths active modes, i.e., and modes, respectively, and the angles are given.
254
Figure 15. 2D IR difference spectra S ( , t ) of six different secondary structure polypeptides. The maximum peak amplitude of the 310H S ( , t ) spectrum is assumed to be one, and all other spectra are rescaled with this maximum value for the sake of quantitative comparisons. Adapted from Ref. [49].
Due to delicate differences in these values, the of 310H is the largest among the three right-handed helices. In addition to this factor, another important aspect is that the A-E frequency splitting amplitude ( - ) is very large when the polypeptide adopts the 310H structure. Consequently, the cancellation induced by destructive interferences between two different sets of nonlinear optical transition pathways, i.e., the two different terms in Eq. (25), is less complete for the 310H than those of the other two helices. Therefore, the strong cross peak amplitudes of the 3 10H S (1 , 3 ) spectrum, in addition to the notable 2D line shape difference, is a characteristic feature of 3 10H, which is distinguished from the other two helices. We next consider the S ( , t ) spectra of the three extended structures. Note that the APB S ( , t ) spectrum is almost identical to that of PB. On the other hand, the cross peak amplitudes in the PII S ( , t ) spectrum are a few times stronger than those of the APB and PB. To describe this quantitative difference in the S ( , t ) spectra of extended structure peptides, it is necessary to carefully examine the dipole strengths and angles of the two strongly IR-active modes in these peptides. In Table 2, the mode has relatively large dipole strength for all three cases of extended structures note that the C=O bond directions in these extended structures are approximately perpendicular to the strand. The dipole strength of mode in the cases of APB and PB are very small even though it is the second most strongly IRactive mode. In contrast, for the PII polypeptide the dipole strength of the -mode is quantitatively similar to that of the -mode and they are relatively large. Therefore, 2 2 value is large for the PII. Since the angles between and modes for APB, PB, and PII are all close to 90 degrees, does not play a key role in determin-
255
Table 2. Two strongly IR-active mode frequencies (in cm1), angles (in degree) between the two transition 2 2 dipole vectors, dipole strengths (in au), and (= sin 2 ) values. Adapted from Ref. [49]
RHH H 1656.8
1659.4
87.0
7.6
300.5
39.6
1678.7
1673.6
44.6
31.7
17.9
281.3
310H
1620.4
1639.3
85.3
51.7
14.6
752.1
APB
1654.1
1653.1
87.8
4.5
40.2
179.0
PB
1657.5
1658.4
114.7
3.6
75.1
222.7
PII
1659.5
1658.8
80.0
22.1
30.9
663.0
ing value in these three cases. Then, what causes this difference in the dipole 2 of PII an order of magnitude larger than strengths? More specifically, why is those of APB and PB? An answer to this question can be found by examining the molecular structures of PII, APB, and PB (see Fig. 14 or 15). The two extended structures, APB and PB, are rather close to planar, whereas the PII is, as well-known, a left-handed helical conformation with an approximate C3 symmetry. Therefore, the -mode in the PII polypeptide is delocalized over more than one peptide bond note that the amide I vibrational coupling constant is large in comparison to those of APB and PB. Then, the transition dipole moment of the -mode is given as an additive (in-phase) summation of the amide I local mode transition dipoles pointing to the same direction. On the other hand, due to the small vibrational coupling constants in the cases of APB and PB, the -mode is rather localized so that its transition dipole is close to that of an amide I local mode. Thus, it becomes clear that the strong vibrational coupling depending on the polypeptide secondary structure is the key factor making the cross peak amplitudes large in a given 2D spectrum. Overall, it was shown that the small difference of structural alignment of peptide groups in a given polypeptide makes a big difference in the S ( , t ) spectrum, which clearly demonstrates the sensitivity of 2D vibrational spectroscopic method for distinguishing different secondary structures of polypeptides in solution. 5.2. Ubiquitin The 2D IR spectra, S ( , t ) and S ( , t ) , of ubiquitin are shown in Fig. 16 note that the maximum peak amplitude of S ( , t ) is set to 1 and the S ( , t ) spectrum is thus re-scaled for the sake of direct quantitative comparison between S ( , t ) and S ( , t ) . Tokmakoff and coworkers recently performed 2D IR
256
Figure 16. (a) Parallel-polarization 2D IR photon echo rephasing spectra S (1 , 3 ) of a ubiquitin. (b) 2D IR difference spectra S (1 , 3 ) of a ubiquitin. Adapted from Ref. [30].
spectroscopic studies of various proteins containing significant amount of anti-parallel -sheet segments and observed Z-shape 2D spectrum, which was suggested to be a characteristic feature of anti-parallel -sheet polypeptide [37]. The more relevant work on ubiquitin was reported by Chung et al. [64] and they found that the 2D spectrum of ubiquitin exhibits a similar Z-shape 2D spectral pattern that reflects the contribution from the short and twisted five-stranded -sheet in the ubiquitin. As shown in the numerically simulated S ( , t ) spectrum in Fig. 16(a), indeed the 2D spectrum is diagonally elongated and a hint of the Z-shape can be observed, even though the cross peak is not as notable as the experiment. We found that the cross peaks between the mode at ~1670 cm1 are strongly coupled to the other low-frequency IR-active modes delocalized on the -sheet segment peptides. The parallel-polarization 2D IR photon echo spectrum in Fig. 16(a) certainly provides detailed information on the couplings between different amide I modes, the difference 2D IR spectrum can be far more revealing and exhibits a few notable cross peaks. Figure 16(b) depicts the ubiquitin difference 2D spectrum S ( , t ) . Note that the diagonal peak intensities are significantly diminished. However, due to couplings between two neighboring modes in the frequency-domain, there still appear strong features close to the diagonal line note that the -sheet structure in the ubiquitin is not an ideal anti-parallel nor parallel -sheet so that the elimination of the diagonal peaks cannot be complete. In order to quantitatively describe the origins of cross peaks in Fig. 16(b), it is necessary to pay attention to the factor, jk , defined in Eq. (25). Among different features revealed by the difference 2D spectrum in Fig. 16(b), we focus on three cross peaks , , and . We found that the spectral amplitudes in the region around the cross peaks and are quantitatively similar to those in the paraland lel-polarization 2D spectrum S ( , t ) . This indicates that the cross peaks are genuine. In the difference 2D spectrum, the most intense cross peak is produced by the two modes that are the -helix-mode and the low-frequency -hairpin (consisting of 1 and 2 strands) mode. Therefore, the cross peak is a signature of amide I mode mix-
257
ing due to the spatial proximity of the -helix and the two 1 and 2 strands in ubiquitin. The second cross peak , on the basis of its location in the difference 2D spectrum, reveals coupling between the 35 mode and the -helix-mode. Here, the 35 mode is the amide I normal mode delocalized on the -helix, 3, and 5 strands. Again, due to the direct contact between the -helix and the two 3 and 5 strands, their couplings can be large and produces the cross peak . The cross peak is less distinguishable in comparison to the other two cross peaks and . However, the cross peak is the one that was experimentally observed by Chung et al. [64] On the basis of our mode assignments in reference [30], it originates from coupling the 35 mode and the lowfrequency -hairpin (consisting of 1 and 2 strands) mode. Since the two 1 and 2 strands with the other 3 and 5 strands form a distorted -sheet structure via intramolecular hydrogen bonding interactions, the coupling constant can be large enough to make the two (35 and -hairpin) modes mixing. However, since the angle between the two transition dipoles can be small because they are parts of the -sheet, the cross peak amplitude is relatively small. It is believed that a detailed analysis of temperature-dependent different 2D IR spectra via measuring the cross peak amplitude changes will provide a picture on the unfolding mechanism.
6. Summary and a Few Concluding Remarks We have developed a systematic computational method for numerically simulating amide I IR, VCD, and 2D IR spectra of arbitrary polypeptides in solutions. Combining quantum chemistry calculation, molecular dynamic simulation, Hessian matrix reconstruction, and fragmentation approximation method, we have theoretically studied the vibrational properties of amide I vibrations in proteins and discussed characteristic features of various spectra of secondary structure polypeptides and a protein ubiquitin. Although, the amide I local mode frequencies of six representative model polypeptides in water do not exhibit a strongly site-dependent behavior, the amide I local mode frequencies of ubiquitin in water were found to be highly dependent on where the corresponding peptide bond is located. Taking into account the water-peptide interactioninduced amide I frequency shifts and fluctuations, we performed numerical simulations of the amide I IR spectra and directly compared them with experimentally measured spectra. For the amide I IR and VCD spectra of RHH and PII polypeptides, theoretical results reproduce the experimental results quite well. In the case of ubiquitin, even though the -sheet strands in this protein are rather short and its structure is twisted, we could reproduce the characteristic amide I vibrational spectrum of -sheet. In addition to the linear spectroscopy, we have shown that the 2D IR spectroscopy can provide far more detailed information on the protein secondary structure. The 2D IR photon echo spectra of model polypeptides and ubiquitin were numerically simulated and directly compared with experimental results. Analyzing components contributing to the difference 2D IR spectrum and quantitatively estimating the angles between amide I normal mode transition dipoles and jk values, we were able to provide explanations on the origins of cross peaks. The present investigation shows that, even though there are some quantitative or qualitative agreements between theoretically simulated spectra and experimental results, the current computational simulation methods developed recently need to be improved
258
as long as the polypeptide of interest is a real protein having (i) significant diagonal heterogeneity, (ii) no repeating regular secondary structures, and (iii) sizable amount of peptide segments that cannot be classified as one of the well-known secondary structures. Nevertheless, the lessons we learned from the present comparative investigations are as follows: (i) the amide I local mode frequencies (diagonal vibrational Hamiltonian matrix elements) are strongly site-dependent and sensitive to local environments, (ii) a more elaborate model for computing amide I vibrational coupling constants is however required to accurately simulate the delocalized natures of amide I normal modes in a real protein, (iii) the cross peaks in the 2D IR spectrum can provide additional information on mode-to-mode coupling patterns. Despite that we have focused on amide I vibrations in proteins, one can extend the present coupled-oscillator model to quantitatively describe other types of peptide vibrations in the future.
Acknowledgment This work was supported by the CRIP of KOSEF (MOST, Korea).
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] W.K. Surewicz and H.H. Mantsch, Biochim. Biophys. Acta 952 (1988), 115-130. J. Bandekar, Biochim. Biophys. Acta 1120 (1992), 123-143. H. Susi and D.M. Byler, Methods Enzymol. 130 (1986), 290-311. R.C. Lord, Appl. Spectrosc. 31 (1977), 187-194. M.R. Oboodi, C. Alva and M. Diem, J. Phys. Chem. 88 (1984), 501-505. G.M. Roberts, O. Lee, J. Calienni and M. Diem, J. Am. Chem. Soc. 110 (1988), 1749-1752. S.A. Asher, A. Ianoul, G. Mix, M.N. Boyden, A. Karnoup, M. Diem and R. Schweitzer-Stenner, J. Am. Chem. Soc. 123 (2001), 11775-11781. R. Schweitzer-Stenner, F. Eker, Q. Huang, K. Griebenow, P.A. Mroz and P.M. Kozlowski, J. Phys. Chem. B 106 (2002), 4294-4304. A.V. Mikhonin, S.V. Bykov, N.S. Myshakina and S.A. Asher, J. Phys. Chem. B 110 (2006), 19281943. A.V. Mikhonin, N.S. Myshakina, S.V. Bykov and S.A. Asher, J. Am. Chem. Soc. 127 (2005), 77127720. S. Krimm and J. Bandekar, Adv. Protein Chem. 38 (1986), 181-364. A. Barth and C. Zscherp, Q. Rev. Biophys. 35 (2002), 369-430. W.K. Surewicz, H.H. Mantsch and D. Chapman, Biochemistry 32 (1993), 389-394. J.W. Brauner, C.R. Flach and R. Mendelsohn, J. Am. Chem. Soc. 127 (2005), 100-109. J. Kubelka, J. Kim, P. Bour and T.A. Keiderling, Vib. Spectrosc. 42 (2006), 63-73. R. Schweitzer-Stenner, Vib. Spectrosc. 42 (2006), 98-117. A. Dong, P. Huang and W.S. Caughey, Biochemistry 29 (1990), 3303-3308. J.L.R. Arrondo and F.M. Goni, Prog. Biophys. Mol. Biol. 72 (1999), 367-405. M. Simonetti and C.D. Bello, Biopolymers 62 (2001), 95-108. R. Moritz, H. Fabian, U. Hahn, M. Diem and D. Naumann, J. Am. Chem. Soc. 124 (2002), 6259-6264. S.M. Decatur and J. Antonic, J. Am. Chem. Soc. 121 (1999), 11914-11915. S.M. Decatur, Acc. Chem. Res. 39 (2006), 169-175. D.M. Byler and H. Susi, Biopolymers 25 (1986), 469-487. H. Susi and D.M. Byler, Biochem. Biophys. Res. Commun. 115 (1983), 391-397. M. Jackson, P.I. Haris and D. Chapman, J. Mol. Struct. 214 (1989), 329-355. S.J. Prestrelski, D.M. Byler and M.P. Thompson, Biochemistry 30 (1991), 8797-8804. D.C. Lee, P.I. Haris, D. Chapman and R.C. Mitchell, Biochemistry 29 (1990), 9185-9193.
259
[28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75]
S. Ham, S. Cha, J.-H. Choi and M. Cho, J. Chem. Phys. 119 (2003), 1451-1461. J.-H. Choi, S. Ham and M. Cho, J. Phys. Chem. B 107 (2003), 9132-9138. J.-H. Choi, H. Lee, K.-K. Lee, S. Hahn and M. Cho, J. Chem. Phys. 126 (2007), 045102. Circular Dichroism: Principles and Applications, N. Berova, K. Nakanishi and R.W. Woody, ed., Wiley-VCH, New York, 2000. J.-H. Choi and M. Cho, J. Chem. Phys. 120 (2004), 4383-4392. J.-H. Choi, J.-S. Kim and M. Cho, J. Chem. Phys. 122 (2005), 174903. P. Hamm, M. Lim and R.M. Hochstrasser, J. Phys. Chem. B 102 (1998), 6123-6138. C. Fang, J. Wang, Y.S. Kim, A.K. Charnley, W. Barber-Armstrong, A.B. Smith III, S.M. Decatur and R.M. Hochstrasser, J. Phys. Chem. B 108 (2004), 10415-10427. P. Hamm, M. Lim, W.F. DeGrado and R.M. Hochstrasser, Proc. Natl. Acad. Sci. U. S. A. 96 (1999), 2036-2041. N. Demirdoven, C.M. Cheatum, H.S. Chung, M. Khalil, J. Knoester and A. Tokmakoff, J. Am. Chem. Soc. 126 (2004), 7981-7990. C. Kolano, J. Helbing, M. Kozinski, W. Sander and P. Hamm, Nature 444 (2006), 469-472. H. Maekawa, C. Toniolo, A. Moretto, Q.B. Broxterman and N.-H. Ge, J. Phys. Chem. B 110 (2006), 5834-5837. P. Mukherjee, I. Kass, I.T. Arkin and M.T. Zanni, J. Phys. Chem. B 110 (2006), 24740-24749. S. Woutersen and P. Hamm, J. Chem. Phys. 115 (2001), 7737-7743. A.T. Krummel, P. Mukherjee and M.T. Zanni, J. Phys. Chem. B 107 (2003), 9165-9169. A.T. Krummel and M.T. Zanni, J. Phys. Chem. B 110 (2006), 13991-14000. Y.S. Kim and R.M. Hochstrasser, Proc. Natl. Acad. Sci. U. S. A. 102 (2005), 11185-11190. J. Zheng, K. Kwak, J.B. Asbury, X. Chen, I. Piletic and M.D. Fayer, Science 309 (2005), 1338-1343. J. Zheng, K. Kwak, X. Chen, J.B. Asbury and M.D. Fayer, J. Am. Chem. Soc. 128 (2006), 2977-2987. S. Ham, S. Hahn, C. Lee, T.K. Kim, K. Kwak and M. Cho, J. Phys. Chem. B 108 (2004), 9333-9345. J.-H. Choi, S. Hahn and M. Cho, Int. J. Quan. Chem. 104 (2005), 616-634. J.-H. Choi, Hahn, S., Cho, M., Biopolymers 83 (2006), 519-536. M. Cho, Nature 444 (2006), 431. M. Cho, Bull. Kor. Chem. Soc. 27 (2006), 1940-1960. S. Krimm and Y. Abe, Proc. Natl. Acad. Sci. U. S. A. 69 (1972), 2788-2792. T.C. Cheam and S. Krimm, Chem. Phys. Lett. 107 (1984), 613-616. H. Torii and M. Tasumi, J. Chem. Phys. 96 (1992), 3379-3387. C. Scheurer, A. Piryatinski and S. Mukamel, J. Am. Chem. Soc. 123 (2001), 3114-3124. S. Woutersen and P. Hamm, J. Phys. Chem. B 104 (2000), 11316-11320. P. Hamm, M. Lim, W.F. DeGrado and R.M. Hochstrasser, J. Phys. Chem. A 103 (1999), 10049-10053. J.W. Brauner, C. Dugan and R. Mendelsohn, J. Am. Chem. Soc. 122 (2000), 677-683. M.A. Bryan, J.W. Brauner, G. Anderle, C.R. Flach, B. Brodsky and R. Mendelsohn, J. Am. Chem. Soc. 129 (2007), 7877-7884. T. Miyazawa, J. Chem. Phys. 32 (1960), 1647-1652. T. Miyazawa and E.R. Blout, J. Am. Chem. Soc. 83 (1961), 712-719. Y.N. Chirgadze and N.A. Nevskaya, Biopolymers 15 (1976), 607-625. C.M. Cheatum, A. Tokmakoff and J. Knoester, J. Chem. Phys. 120 (2004), 8201-8215. H.S. Chung, M. Khalil, A.W. Smith, Z. Ganim and A. Tokmakoff, Proc. Natl. Acad. Sci. U. S. A. 102 (2005), 612-617. S. Cha, S. Ham and M. Cho, J. Chem. Phys. 117 (2002), 740-750. J.-H. Choi, S. Ham and M. Cho, J. Chem. Phys. 117 (2002), 6821-6832. H. Torii and M. Tasumi, J. Raman Spectrosc. 29 (1998), 81-86. M. Cho, J. Chem. Phys. 118 (2003), 3480-3490. S. Ham and M. Cho, J. Chem. Phys. 118 (2003), 6915-6922. H. Lee, S.-S. Kim, J.-H. Choi and M. Cho, J. Phys. Chem. B 109 (2005), 5331-5340. S.-H. Lee and S. Krimm, Biopolymers 46 (1998), 283-317. A.M. Dwivedi and S. Krimm, Biopolymers 23 (1984), 923-943. P. Bour, J. Sopkova, L. Bednarova, P. Malon and T.A. Keiderling, J. Comp. Chem. 18 (1997), 646659. P. Bour, J. Kubelka and T.A. Keiderling, Biopolymers 65 (2002), 45-59. J. Kubelka and T.A. Keiderling, J. Am. Chem. Soc. 123 (2001), 12048-12058.
260
[76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103]
J. Hilario, J. Kubelka and T.A. Keiderling, J. Am. Chem. Soc. 125 (2003), 7562-7574. S. Ham, J.-H. Kim, H. Lee and M. Cho, J. Chem. Phys. 118 (2003), 3491-3498. S. Hahn, S. Ham and M. Cho, J. Phys. Chem. B 109 (2005), 11789-11801. K. Kwac and M. Cho, J. Chem. Phys. 119 (2003), 2247-2255. K. Kwac, H. Lee and M. Cho, J. Chem. Phys. 120 (2004), 1477-1490. K.-K. Lee, K.-I. Oh, H. Lee, C. Joo, H. Han and M. Cho, ChemPhysChem. 8 (2007), 2218-2226. S. Hahn, H. Lee and M. Cho, J. Chem. Phys. 121 (2004), 1849-1865. C. Lee and M. Cho, J. Phys. Chem. B 108 (2004), 20397-20407. H. Torii and M. Tasumi, J. Chem. Phys. 97 (1992), 92-98. M. Rechsteiner, Ubiquitin, Plenum Press, New York, 1984. Y. Mu and G. Stock, J. Phys. Chem. B 106 (2002), 5294-5301. S. Woutersen, R. Pfister, P. Hamm, Y. Mu, D.S. Kosov and G. Stock, J. Chem. Phys. 117 (2002), 6833-6840. M.F. DeCamp, L. DeFlores, J.M. McCracken, A. Tokmakoff, K. Kwac and M. Cho, J. Phys. Chem. B 109 (2005), 11016-11026. K. Kwac and M. Cho, J. Chem. Phys. 119 (2003), 2256-2263. T.M. Watson and J.D. Hirst, Mol. Phys. 103 (2005), 1531 - 1546. P. Bour and T.A. Keiderling, J. Chem. Phys. 119 (2003), 11253-11262. P. Bour, D. Michalik and J. Kapitan, J. Chem. Phys. 122 (2005), 144501. J.R. Schmidt, S.A. Corcelli and J.L. Skinner, J. Chem. Phys. 121 (2004), 8887-8896. G. Eaton, M.C.R. Symons and P.P. Pastogi, J. Chem. Soc., Faraday Trans. 1 85 (1989), 3257-3271. M. Maroncelli and G.R. Fleming, J. Chem. Phys. 86 (1987), 6221-6239. M.T. Zanni, M.C. Asplund and R.M. Hochstrasser, J. Chem. Phys. 114 (2001), 4579-4590. D.A. McQuarrie, Statistical Mechanics, Harper & Row, New York, 1976. V. Baumruk and T.A. Keiderling, J. Am. Chem. Soc. 115 (1993), 6939-6942. G. Shanmugam and P.L. Polavarapu, J. Am. Chem. Soc. 126 (2004), 10292-10295. T. Brixner, I.V. Stiopkin and G.R. Fleming, Opt. Lett. 29 (2004), 884-886. D.M. Jonas, Annu. Rev. Phys. Chem. 54 (2003), 425-463. S. Hahn, S.-S. Kim, C. Lee and M. Cho, J. Chem. Phys. 123 (2005), 084905. M.T. Zanni, N.-H. Ge, Y.S. Kim and R.M. Hochstrasser, Proc. Natl. Acad. Sci. U. S. A. 98 (2001), 11265-11270.
261
Application of Isotope-Edited FTIR Spectroscopy to the Study of Protein-Protein Interactions

Tiansheng LI a,1 and Tsutomu ARAKAWA b PacificBio Inc., 1152 Tourmalin Drive, Newbury Park, CA 91320 b Alliance Protein Laboratory, Inc., 3957 Corte Cancion, Thousand Oaks, CA 91360
a
Abstract. Detailed analysis of molecular interactions is the key to understanding the mechanism of signal transduction and its intervention by inhibitory compounds. In this chapter, we shall review and discuss the application of isotope edited FTIR spectroscopy to the investigation of protein-protein interactions. Recently we have employed this technique to investigate the molecular interactions of granulocyte colony stimulating factor (G-CSF) with the isolated immunoglobular domain (Ig) of its receptor [1,2]. To resolve the amide I band overlap of G-CSF with that of the receptor in the FTIR spectrum of the complex, 13C/15N uniformly labeled GCSF was prepared for this study. By comparing the FTIR spectra of the isotopelabeled G-CSF and the isolated receptor with that of the complex, we have provided spectral evidence that the AB loop region involving the unique 310 helix segment of G-CSF likely undergoes a conformational change to a regular -helix upon binding to the receptor domain. The IR data also indicate significant conformational changes involving -turns and irregular structures in the Ig domain of the receptor in the complex. Furthermore, FTIR spectra of G-CSF, the receptor, and their complex demonstrate clearly that protein conformations of both G-CSF and the receptor are dramatically stabilized by complex formation. Together, the current data strongly suggest that the AB loop region including the 310 helix interacts specifically with the immunoglobulin-like domain of the receptor, which may play a role in receptor dimerization. This conclusion supports the structural model recently proposed by Layton and co-workers [3]. In summary, this work demonstrates that specific structural information of protein-protein complexes can be obtained by employing isotope-edited FTIR spectroscopy. Keywords. Isotope-edited, FTIR, spectroscopy, G-CSF, receptor, immunoglobular domain, IgG
Introduction Detailed analysis of molecular interactions is the key to understand the mechanism of signal transduction and its intervention by inhibitory molecules, which can be developed as therapeutic drugs. Among the spectroscopic techniques, FT-IR spectroscopy has been used extensively to study secondary structures of proteins in various physical states [412]. However, band overlaps in the amide I/I region in FTIR spectra of protein-protein or protein-peptide complex hinder the usefulness of conventional FTIR
1 Corresponding Author: Tiansheng Li, PacificBio Inc., 1152 Tourmalin Street, Newbury Park, CA 91320. E-mail: tli@pacificbioinc.com.
262
T. Li and T. Arakawa / Application of Isotope-Edited FTIR Spectroscopy
spectroscopy in detailed structural analysis of protein complexes. In 1992, Haris and coworkers were first to demonstrate that by 13C uniform labeling of amide C=O groups the amide I band of protein can be red-shifted to approximately 4045 cm1 [13], and this makes it possible to determine secondary structures of individual subunit in a protein/protein or protein/peptide complex. Since then, this method has been applied to the structural study of several protein/protein and protein/peptide systems [12,14,15]. Because the frequency of 13C labeled amide group is approximately 40 cm1 lower relative to that of unlabeled one [13,1618], it is possible to resolve amide I/I bands of two subunits in a protein/protein complex if only one of the subunits is 13C uniformly labeled. Until 1997, there has few reports on the application of FTIR spectroscopy to the investigation of cytokine/receptor complex because of spectral overlap in amide I/I region. With the exception of a few ligand/receptor complexes whose crystal structures have been determined, there is limited structural information available with respect to the conformational changes in cytokine/receptor complexes. Granulocyte colony stimulating factor (G-CSF) stimulates the growth of neutrophils upon binding to its receptor [1924]. It is widely accepted that the first step in signal transduction pathway is receptor dimerization induced by the binding of cognate ligand. Conformational changes may also occur in the receptor that can modulate signal transductions and in the ligand that can bring about cooperative ligand/receptor interactions [23,24]. In 1997, we have applied isotope-edited FTIR spectroscopy for the first time to the structural study of granulocyte colony stimulating factor (G-CSF)/receptor interactions and demonstrated its usefulness in determining unambiguously the protein secondary structures of both ligand and receptor in the signal transduction complex [12]. In the previous study, we investigated the structural changes in G-CSF and its receptor upon complex formation, and showed that both G-CSF and the receptor likely undergo conformational changes in the complex [12]. Structure of G-CSF has been determined by X-ray crystallography [25], which reveals a classical four-helical bundle structure (designated A, B, C and D helices) with up-up-down-down topology. In contrast to the structure of G-CSF, the structure of its receptor is more complex and remains to be fully illustrated. The entire extracellular domain of G-CSF receptor consists of six distinct structural domains with each one consisted of approximately 100 amino acid residues. Those six distinct domains include the immunoglobular (Ig-like) domain, the CRH (cytokine receptor homologous domain consisted of BN and BC domains) and the FN type III (fibronectin III) domain (containing three separate domains). Structure of G-CSF/receptor complex formed between G-CSF and the truncated CRH domain of its receptor has been determined by Xray crystallography [26]. The crystal structure revealed direct contacts between helices A and C of G-CSF and the hinge region of the CRH domain of the receptor [26]. The entire extracellular domain of G-CSF receptor consists of the immunoglobular-like (Ig) domain, the cytokine receptor homology (CRH) domain and the fibronectin III-like domain. Both Ig and CRH domains have been shown to be critical in the interaction with the ligand [3]. Molecular mutegenesis data have shown that the Ig domain plays a pivotal role in the formation of active signal transduction complex with G-CSF [3]. However, the molecular mechanism of interaction between G-CSF and the Ig domain of its receptor remains to be illustrated. So far, there is little structural information available on the complex formed between G-CSF and the Ig domain of its receptor. Previous report has shown that interaction between G-CSF and the Ig domain of its receptor is critical in forming a biologically active signal transduction complex [3,27].
263
We have reported previously that structural information of the complex formed between G-CSF and its full-length receptor can be obtained by isotope edited FTIR spectroscopy [12]. Recently, we have investigated the interaction between G-CSF and the Ig domain of its receptor, and we have purified the Ig domain of G-CSF receptor from E. coli expression system [1,2]. Here, we shall review our previous studies shown that G-CSF can form complex with the Ig domain of its receptor and that the complex formation appears to stabilize structures of both G-CSF and the Ig domain. FTIR spectrum of the complex has also revealed that conformational changes in both G-CSF and the Ig-like domain of the receptor occur upon complex formation [1,2].
1. Experimental Methods 1.1. Purification of G-CSF Receptor Ig Domain and Isotope Labeled G-CSF The Ig domain has been expressed in E. coli cells and the inclusion body containing the Ig domain was purified by using repeated wash/centrifugation procedures using 3 M urea, 10 mM lysine, 10 mM DTT, pH 7.5. The final inclusion body was dissolved in 6 M guanidine-HCl, 10 mM DTT, pH 8.0, and was then dialysed at 4C against refolding buffer containing 10 mM Tris, 1 mM EDTA, 2 mM reduced glutathione, 10 M oxidized glutathione, pH 8.0. After 4 days of refolding at 4C, the protein solution was passed through a G-CSF-sepharose column equilibrated with PBS, pH 7.3. The bound Ig domain receptor was eluted off the column by using a low pH buffer, and subsequently was dialyzed against a neutral (pH 7) phosphate buffer. The G-CSF/Ig domain complex was prepared by mixing excess amount of purified G-CSF with the isolated Ig domain in the neutral phosphate buffer and the mixture was loaded onto a Superex 75 column equipped with a FPLC system (Pharmacia Inc.), with a running buffer at neutral pH. The G-CSF/Ig complex was purified by collecting fractions from the Superex 75 column. Uniform 13C/15N isotope labeling of G-CSF was achieved by following the same procedure as described previously [12]. Both 13C/15N uniformly labeled G-CSF and the labeled G-CSF/receptor complex were purified as described previously [12]. The purified G-CSF, the receptor and the G-CSF/receptor complex were dissolved in 10 mM sodium phosphate, pH 7.0 and buffer exchanged into D2O solutions for FTIR data collection. 1.2. FTIR Spectroscopy Stock solutions of G-CSF in NaAc, pH 4.5 buffer was concentrated and hydrogendeuterium exchanged into 10 mM sodium phosphate, 50 mM NaCl, pD 7.0 buffer using a Centricon (Amicon) cartridge with a membrane of 10 kD molecular weight cutoff. Diafiltration of G-CSF in D2O buffer was repeated several times to ensure complete HD exchange. The final protein concentrations were approximately 10 mg/mL before IR data collection, and the protein solution was injected into a sample holder consisted of a pair of CaF2 windows with a 25 m spacer to form an uniformly thin film. Sample temperature was maintained at 20C during IR data collection using a thermal jacket controlled by an electronic thermal controller (Boulder Nonlinear Inc.). Typically, 1024 interferograms were co-added and Fourier-transformed to generate an absorbance spectrum at 4 cm1 resolution by employing a Mattrson Research Series Model 1000 spec-
264
Table 1. Assignments of Amide I Bands
trometer with a MCT detector cooled by liquid nitrogen. The spectrometer was continuously purged with a dry air system (Model 7560, Balston Filter System, Whatman) to eliminate the spectral interference from atmospheric water vapor. Residual contributions from atmospheric vapor were digitally subtracted from the protein spectra. The broad buffer background of D2O was subtracted from protein spectra to ensure a flat baseline in the spectral region 14001800 cm 1. FTIR spectral analysis including spectral deconvolution and calculation of second derivative spectra, has been carried out by employing Grams/386 software (Galactic Industries Corp.).
2. Results and Discussion Table 1 lists the frequencies of amide I bands of natural and 13C uniformly labeled proteins, and the corresponding secondary structures. This summary cites the FTIR data from a number of published literatures [1318]. In general, a 4050 cm 1 isotopic shift (red-shift) of amide I band is expected when a protein is 13C uniformly labeled. Though not listed in Table 1, 15N uniform labeling of proteins does not lead to significant frequency shift of amide I/I band. This is due to the fact that carbonyl (C=O) stretching vibration is the major contributor to amide I/I band. Crystal structure of G-CSF has been determined by Hill and co-workers in 1992, and it reveals a classical four helical bundle structure with up-up-down-down topology [25]. In 1999, the crystal structure of G-CSF complexed with its receptor CRH (cytokine receptor homology) domain has been determined by Aritomi and coworkers [26]. The CRH domain of G-CSF receptor does not include the immunoglobular (Ig) domain. In the crystal structure of G-CSF/CRH complex, there is no direct contact between the AB loop region of G-CSF and the CRH domain of its receptor. However, further studies of the receptor Ig domain have suggested that mutations in the AB loop region of G-CSF dramatically reduce its activity and receptor binding capability [3,27]. Both the AB loop region and the D-helix of G-CSF are now believed to be involved in the specific interactions with the receptor. Though biochemical and mutagenesis work indicate that specific interactions may occur between G-CSF and the Ig domain of its receptor, the molecular mechanism and structural information of the complex formed between G-CSF and the receptor Ig-domain remains unknown at this point. Binding to CRH receptor domain leads to conformational changes in G-CSF mostly involving the AB loop and the N-terminal portion of the A-helix [26].
265
Figure 1. The elution profile of the receptor Ig domain and its complex with G-CSF from Superex 75 column (Pharmacia Inc.) under native conditions.
To further investigate the molecular interactions between G-CSF and the Ig domain of its receptor, we have carried out a systematic FTIR analysis of G-CSF/Ig complex [1]. As reported previously, our recent study show that G-CSF can form specific complex with the Ig domain of its receptor [1]. Figure 1 shows the elution profile of the receptor Ig domain and its complex with G-CSF from a Superex 75 column (Pharmacia Inc.). The purified Ig domain of G-CSF receptor elutes at about 13.5 minute while free G-CSF elutes at about 11 minute. In the presence of excess of G-CSF, the purified Ig domain can be completely titrated and forms complex with the ligand. Interestingly, the complex shows one main peak near 9 minute and one minor peak near 8 minute, which indicates the presence of relatively large complex. The native sizeexclusion chromatographic data suggests that the refolded and purified Ig domain receptor is active and capable of binding to its ligand, G-CSF. The G-CSF/Ig complex has been purified by collecting the fractions from the Superex 75 column. Figure 2 compares FTIR spectra of non-labeled and 13C/15N uniformly labeled G-CSF, its clear that these two spectra have nearly identical spectral features with the exception of frequency shift. These two spectra were published in our previous study [12]. The major amide I bands at 1653 cm1 in the spectrum of the unlabeled G-CSF (Fig. 2, top panel) is characteristic of proteins containing -helix as the predominant secondary structure. Minor amide I band at 1677 cm1 is normally associated with loops and/or -turns. Though amide I band at 1640 cm1 is normally associated with -sheet structure, this band in the spectrum of G-CSF is unlikely due to -sheet since the crystal structure of G-CSF reveals virtually no presence of -sheet [25]. In literature, the amide I band at 1641 cm1 has also been assigned to 310 helix in some proteins [2831]. The amide I band at 1641 cm1 in the spectrum of the unlabeled G-CSF is shifted to 1598 cm1 in the spectrum of the labeled protein. The pattern of the amide I bands in the spectrum of the unlabeled G-CSF is consistent with the reported FTIR spectra of other -helical proteins. In the spectrum of the 13C/15N uniformly labeled G-CSF (Fig. 2, bottom panel), there is no significant amide I bands above
266
Figure 2. Second derivative FTIR spectra of unlabeled (top, panel A) and 13C/15N isotope labeled (bottom, panel B) G-CSF proteins.
Figure 3. Second derivative FTIR spectra of 13C/15N uniformly labeled G-CSF at temperatures from 20 to 90C.
1635 cm1. The complete shift of amide I bands suggests that 13C/15N labeling of GCSF is close to 100%. Figure 3 shows the second derivative FTIR spectra of 13C/15N uniformly labeled G-CSF at temperatures from 20 to 90C. As temperature is increased to 90C, the in-
267
Figure 4. Second derivative FTIR spectra of the isolated receptor Ig domain (top panel), G-CSF/Ig complex (middle panel) and the isotope labeled G-CSF/Ig complex (bottom panel).
tensity of the major amide I band at 1610 cm1 diminishes dramatically, and the intensity of amide I band at 1573 cm1 increases correspondingly. In addition, the amide I band at 1635 cm1 is shifted to 1639 cm1 when the labeled protein is fully denatured. Considering the frequency shift due to isotope labeling, the changes in the FTIR spectrum of the labeled G-CSF are consistent with the thermal penetration of -helical protein into aggregates containing mostly anti-parallel -sheet. Figure 4 compares the second derivative FTIR spectra of the receptor Ig domain, G-CSF/Ig complex and the isotope labeled G-CSF/Ig complex. The second derivative spectrum of the isolated Ig domain of the receptor exhibits two major amide I bands at 1630 and 1682 cm1 and minor bands at 1660 and 1675 cm1. Those spectral features are indicative of -strand as the dominant protein secondary structures in the isolated Ig domain. In the spectrum of the unlabeled G-CSF/Ig complex (Fig. 4, middle panel), there is significant spectral overlap between the major amide I band at 1639 cm 1 (due to Ig domain) and that at 1653 cm1 (due to unlabeled G-CSF). Due to this spectral overlap, it is difficult to obtain information about secondary structures of the unlabeled G-CSF and the complexed Ig domain. In the spectrum of the labeled G-CSF/Ig complex, the amide I bands of the labeled G-CSF and the complexed Ig domain are nearly completely resolved. The major amide I band at 1610 cm1 in the spectrum of the labeled G-CSF/Ig complex is due to the -helical content in G-CSF. It is noted that the shoulder component at 1598 cm1 is absent in the spectrum of the labeled complex, suggesting significant conformational change in the labeled G-CSF within the complex. In our previous study, the amide I band at 1598 cm1 is attributed to 310 helix present in the AB loop region of the labeled G-CSF [12], thus indicating that the absence f this band in the complex is due to the conversion of 310 helix into the regular helix. Similar spectral change has been observed in the labeled G-CSF complexed with the full-length receptor [12]. However, it was unclear in our previous study which part of the receptor is involved in the interaction with the 310 helix region of G-CSF [12]. In combination
268
Figure 5. Second derivative FTIR spectra of the isolated receptor Ig domain (top panel) and CSF/Ig complex (bottom panel).
13
C/15N G-
with the recent data [1,2], it is therefore concluded that the 3 10 helix region of G-CSF is involved in the interaction with the Ig domain of its receptor. Figure 5 compares the second derivative FTIR spectrum of the 13C/15N G-CSF/Ig complex with that of free 13C/15N G-CSF. There are significant differences in both band intensities and frequencies between these two spectra. Figure 5 reveals significant conformational changes in the Ig domain upon ligand binding, as indicated by amide I bands at 1646, 1664, 1674, and 1686 cm1. The conformational changes appear to involve mostly irregular structures and -turns in the Ig domain. These structural changes in both the Ig domain and the G-CSF structure may confer enhanced affinity for their bindings. From Fig. 2, it is clear that 13C/15N G-CSF does not contribute to any amide I bands above 1635 cm1. Therefore, amide I bands at 1646, 1664, 1674 and 1686 cm1 in the spectrum of the 13C/15N G-CSF/Ig complex are due to the receptor Ig domain. For the isolated Ig domain, the major amide I band at 1630 cm1 appears broad, and other minor amide I bands above 1630 cm1 are not well defined. Overall, the spectral features of the isolated Ig domain indicate that though it contains -sheet as the predominant secondary structure significant amount of irregular structures are also present in the isolated Ig domain. In contrast, the complexed Ig domain exhibits a sharp amide I band at 1630 cm1 and other well defined minor amide I bands at 1646, 1664, 1674 and 1686 cm1, which suggests less amount of irregular structures in the ligand-bound Ig domain than in the isolated one. The amide I band at 1686 cm1 in the spectrum of 13C/15N G-CSF/Ig complex is shifted to 1682 cm1 in that of the isolated Ig receptor domain. The most likely explanation for this observation is that G-CSF/Ig interactions prevent H-D exchange of -turns in Ig domain. An alternative explanation would be that the frequency shift is due to conformational changes in the -turn regions of the Ig domain interacting with G-CSF in the G-CSF/Ig complex.
269
Figure 6. Second derivative FTIR spectra of the isolated receptor Ig domain at temperatures from 20 to 90C.
Figure 6 shows the second derivative FTIR spectra of the isolated receptor Ig domain at temperatures from 20 to 90C. It is clear from Fig. 6 that the isolated Ig domain of G-CSF receptor appears fairly unstable and shows a melting transition from 30 to 40C, as indicated by the shift of the major amide I band from 1630 to 1619 cm 1, and the frequency shift of the amide I band from 1682 to 1687 cm1. The spectral features of the denatured Ig domain are characteristically associated with anti-parallel sheet present in most of the thermally denatured proteins. Figure 7 shows the second derivative FTIR spectra of the 13C/15N G-CSF/Ig complex at temperatures 20 to 90C. At 20C, the major amide I band at 1610 cm 1 is due to -helix in 13C/15N labeled G-CSF, and the amide I band at 1632 cm1 is due to sheet structure in the Ig domain. As temperature is raised from 20 to 90C, the intensity of amide I band at 1610 cm1 decreases and that of amide I band at 1574 cm1 increases. As discussed previously (Fig. 3), the spectral changes associated with amide I bands at 1610 and 1574 cm1 is due to the thermal denaturation of 13C/15N labeled GCSF at higher temperatures. The melting transition of G-CSF in the complex is around 50C, and that of the Ig domain in the complex occurs at approximately the same temperature. This suggests that thermal denaturation of the complex occurs at near 50C. Thermal denaturation of the labeled G-CSF in the complex is marked by the decrease in intensity of amide I band at 1610 cm1, and the appearance of amide I band at 1574 cm1. Thermal denaturation of the Ig domain in the complex is marked by the frequency shift of amide I band from 1681 to 1685 cm1, and the appearance of amide I band at 1619 cm1. Specifically, the melting transition (Tm) of -helix in G-CSF is increased by nearly 10C and that of -strand in the receptor Ig-domain by nearly 15C in the G-CSF/receptor complex. However, the increased stabilization of G-CSF is not as high as that in the complex with the full-length receptor [12]. It is likely that G-CSF
270
Figure 7. Second derivative FTIR spectra of the 90C.
13
C/15N G-CSF/Ig complex at temperatures from 20 to
interaction with the Ig domain only stabilizes part of G-CSF, while the interaction with the full-length receptor stabilizes multiple sites of G-CSF and thereby confers greater stability to the ligand.
3. Conclusions In this chapter, we discussed the usefulness of FTIR spectroscopy in the study of protein-protein interactions. Specifically, we illustrated the unique aspect of using isotope edited FTIR spectroscopy to study the molecular interactions of G-CSF with the immunoglobular domain (Ig) of its receptor. It was shown that 13C/15N uniform isotope labeling of G-CSF can lead to significant frequency separation between the amide I bands of the isotope labeled protein and those of the unlabeled receptor in the FTIR spectrum of their complex. Consequently, definitive spectral assignments can be achieved and meaningful spectral interpretations regarding the secondary structures of individual subunits in a protein-protein complex are made possible. It is also noted that thermal denaturation processes of both the isotope labeled subunit and the unlabeled one in a protein/protein complex can be simultaneously monitored by FTIR spectroscopy because of the large frequency separation between amide I/I bands of the labeled subunit and those of the unlabeled one. It is therefore possible to access unambiguously the effect of complex formation on the structural integrity of protein subunits in complexes by employing isotope edited FTIR spectroscopy. Spectral shifts upon complex formation may be used to identify small molecule inhibitorsm which intervene he formation of the complex, and may also be compared with the mode of interaction between the inhibitors and the proteins, to which they bind.
271
References
[1] Li, T., Horan, T., Arakawa, T., Chang, B., Biophys. J. 80(2), 2002. [2] Li, T. Conformational changes and structural stabilization of G-CCF upon complex formation with the immunoglobulin-like (Ig) domain of its receptor: An investigation by isotope edited FTIR spectroscopy. Invited Speaker at the 227th ACS National Meeting, Anaheim, CA, March 28-April 1, 2004. [3] Layton, J.E., Hall, N.E., Connell, F., Venhorst, J. & Treutlein, H.R., J. Biol. Chem. 276 (2001), 36779-36787. [4] Byler, D.M. & Susi, H., Biopolymers 25 (1986), 469-487. [5] Byler, D.M. & Susi, H., J. Industrial Microbiol. 3 (1988), 73-84. [6] Arrondo, J.L.R., Muga, A., Castresana, J., & Goni, F.M., Prog. Biophys. Mol. Biol. 59 (1993), 23-56. [7] Prestrelski, S.J., Tedeschi, N. Arakawa T. and Carpenter, J.F., Biophys. J. 65 (1993), 661-671. [8] Dong, A., Prestrelski, S.J., Allison, S.D. and Carpenter, J.F., J. Pharm. Sci. 84 (1995), 415-424. [9] Jackson, M. & Mantsch, H.H., Critical Rev. Biochem. Mol. Biol. 30 (1995), 95-120. [10] Haris, P.I. & Chapman, D., Biopolymers 37 (1995), 251-263. [11] Prestrelski, S.J., Pikal, K.A. and Arakawa, T., Pharm. Res. 12 (1995), 1250-1259. [12] Li, T., Horan, T., Osslund, T., Stearns, G. & Arakawa, T., Biochemistry 36 (1997), 8849-8857. [13] Haris, P.I., Robillard, G.T., van Dijk, A.A., & Chapman, D., Biochemistry 31 (1992), 6279-6284. [14] Zhang, M., Fabian, H., Mantsch, H.H., & Vogel, H.J., Biochemistry 33 (1994), 10883-10888. [15] Ludlam, C.F.C., Sonar, S., Lee, C.-P., Coleman, M., Herzfeld, J., RajBhandary, U.L., & Rothschild, K.J., Biochemistry 34 (1995), 2-6. [16] Hubner, W., Mantsch, H.H., & Casal, H.L., Appl. Spectrosc. 44 (1990), 732-734. [17] Tadesse, L., Nazarbaghi, R., & Walters, L., J. Am. Chem. Soc. 113 (1991), 7036-7037. [18] Martinez, G.V., Fiori, W.R., & Millhauser, G., Biophys. J. 66 (1994), A65. [19] Nicola, N.A., Metcalf, D., Matsumoto, M., & Johnson, G.R., J. Biol. Chem. 258 (1983), 9017-9023. [20] Souza, L.M., Boone, T.C., Gabriloe, J., Lai, P.H., Zaebo, K.M., Murdock, D.C., Chazin, V.R., Bruszewski, J., Lu, H., Chen, K.K., Barendt, J., Platzer, E., Moore, M.A.S., Mertelsmann, R. & Welte, K., Science 232 (1986), 61-63. [21] Clark, S.C. & Kamen, R., Science 236 (1987), 1229-1237. [22] Nagata, S., & Fukunaga, R., Prog. Growth Factor Res. 3 (1991), 131-141. [23] Stahl, N. & Yancopoulos, G.D., Cell 74 (1993), 587-590. [24] Heldin, C.-H., Cell 80 (1995), 213-223. [25] Hill, C.P., Osslund, T.D., Eisenberg, D. Proc. Natl. Acad. Sci. USA 90 (1993), 5167. [26] Aritomi, M., Kunishima, N., Okamoto, T., Kuroki, R., Ota, Y., Morikawa, K., Nature 401 (1999), 713. [27] Layton, J.E., Shimamoto, G., Osslund, T., Hammacher, A., Smith, D.K. Treutlein, H.R., Boone, T., J. Biol. Chem. 274 (1999), 17445-17451. [28] Dwivedi, A.M., Krimm, S., & Malcolm, B.R., Biopolymers 23 (1984), 2026-2065. [29] Holloway, P. & Mantsch, H.H., Biochemistry 28 (1989), 931-935. [30] Prestrelski, S.J., Byler, D.M., & Thompson, M.P., Int. J. Peptide Protein Res. 37 (1991), 508-512. [31] Miick, S.M., Martinez, G.V., Fiori, W.R., Todd, A.P. & Millhauser, G.L., Nature 359 (1992), 653-655.
272
Biomedical FTIR Spectroscopy of Lipids

a
Willem F. WOLKERS a,1 Institute of Multiphase Processes, Leibniz Universitt Hannover, Hannover, Germany
Abstract. This essay shows how Fourier Transform infrared spectroscopy (FTIR) can be applied to study membrane phase behavior of cells that are relevant for biomedical applications such as red blood cells and platelets. FTIR studies are minimally invasive and do not require labeling. FTIR can give unique information on conformation and stability of membranes in cells that are exposed to heating, freezing or dehydration stress. By combining in situ FTIR techniques with cell viability studies, cell damage can be correlated with membrane phase changes. Understanding the complex behavior of biomembranes during heating, freezing and drying is directly relevant for thermal processing of cells such as is done in cryopreservation and cryosurgery. Keywords. FTIR, membrane phase behavior, liposomes, erythrocytes, platelets, cryopreservation, cryosurgery
1. Biological Membranes 1.1. Phase Behavior of Lipids in Biological Membranes The lipid fraction of cellular natural membranes consists of different lipid species each with their own unique thermotropic, lyotropic and miscibility properties [6]. In particular, the nature of the acyl chain structure: the closer and more regular the packing of the chains, the more viscous and less fluid the bilayer will be at physiological temperature. There are two factors that affect how tightly the bilayer can pack; length and degree of saturation of the acyl chains. A shorter chain length reduces the tendency of the acyl chains to interact with one another through van der Waals forces, increasing the fluidity and conformational disorder of the bilayer [7,33]. In addition to length, kinks in the chains produced by the double bonds make it more difficult for chains to pack against one another. Lipids in bilayers undergo radical changes in physical state over narrow temperature ranges at a characteristic phase transition temperature [7]. The bilayer below the phase transition temperature exists in a closely packed gel state, with the acyl chains relatively immobilized in a tightly packed array. In the more fluid, liquid crystalline state, there is relatively more conformational disorder [38]. When exposed to excess water, most membrane phospholipids form the lamellar liquid crystalline phase. Phosphatidylcholine (PC), the most abundant phospholipid in animal cell membranes, is such a lipid. However other membrane phospholipids, including phosphatidylethanolamine (PE), the second most abundant animal cell phospholipid, spontaneously form nonlamellar structures like the reversed hexagonal phase [28]. The cone-shaped geometry of the latter lipids, with their relatively small head group size coupled with
1 Corresponding Author: Institute of Multiphase Processes, Leibniz Universitt Hannover, Callinstrasse 36, D-30167 Hannover, Germany, wolkers@imp.uni-hannover.de
W.F. Wolkers / Biomedical FTIR Spectroscopy of Lipids
273
Figure 1. Schematic representation of typical lipid phases. (A) Liquid crystalline phase, (B) gel phase, and (C) inverted hexagonal phase (HII).hort caption.
the relatively large area occupied by the acyl chains, promotes the formation of these structures [31,26]. Figure 1 illustrates the different lipid phases of natural lipids. In cellular membranes, the situation is more complex, because the mixture of lipids, sterols and proteins causes a complex phase transition behavior. Biological membranes are thought to be in the fluid phase under physiological conditions. According to the fluid mosaic model of Singer and Nicolson, biological membranes can be considered as a two-dimensional liquid where all lipid and protein molecules diffuse more or less freely [54]. The Fluid-mosaic model for the structure of biological membranes has been revisited, with the discovery that such membranes consist of fluid domains in which more ordered domains are embedded. These liquid-ordered microdomains, known as rafts are enriched in cholesterol and sphingolipids [53,4,5]. It has been suggested that the high acyl chain melting temperature of sphingolipids promotes phase separation of rafts from the more fluid domains of the membrane. In addition, cholesterol preferentially interacts with sphingomyelin by intercalating with its long unsaturated acyl chains and by hydrogen bonding with its headgroup [48,3,53,2].
274
2. Use of FTIR to Study Biomolecules There are several spectroscopic techniques available that can be applied to study conformation and molecular interactions of biomolecules, i.e. NMR, ESR, CD, FTIR, Raman, Absorption, Neutron diffraction, and X-ray crystallography. However, in complex systems (cells and tissues) only a few techniques can be applied. One of the few suitable techniques for cell and tissue analysis which does not require labeling is Fourier Transform infrared spectroscopy (FTIR). With this technique, molecular vibrations can be measured irrespective of the hydration state of the tissue. Molecular vibrations are sensitive to intra- and intermolecular interactions. FTIR can be used to monitor molecular vibrations in cells as they undergo phase changes during changes in temperature or during drying [60]. This application of FTIR is relevant to study cells during thermal treatments such as ablation [29], loading cells with extracellular molecules [61] or freeze-drying [62]. On account of characteristic molecular vibrations, information can be derived on the molecular conformation and the intermolecular interactions of biomolecules in their native environment. The temperature dependence of several characteristic molecular vibrations can be used to study membrane phase behavior and heatinduced protein unfolding. In the following section, the application of FTIR to study the molecular conformation and intermolecular interactions of biomolecules, particularly lipids and proteins, in model and cellular systems is described. 2.1. Infrared Spectroscopy The mid infrared region of the electromagnetic spectrum ranges from 4004000 cm 1. The energy of most molecular vibrations corresponds to the infrared region. A vibration is only infrared-active if it is associated with a change in the electric dipole moment. If a diatomic molecule is considered as a simple harmonic oscillator, the vibrational frequency (f) is:
f = 1 2 k
(1)
where is the reduced mass of the two atoms ( = m1.m2/(m1+m2) with m1 and m2 as the atomic masses, and k is the force constant determined by the strength of the bond. The potential energy (V) of the oscillator is given by:
V= 1 k (r re) 2 2
(2)
where r is the distance between the two vibrating atoms and re, the equilibrium distance. In infrared spectroscopy the frequency is expressed as wavenumbers (f = c(1/)); this transforms equation 1 into:G
1 1 = 2 .c k
(3)
275
Figure 2. Infrared absorption spectrum of dried POPC vesicles. The symmetric and asymmetric CH stretching bands from the acyl chains, and the C=O and PO4 stretching bands from the headgroups are indicated.
where 1/ represents wavenumber () usually expressed in cm1.G The IR-technique is of great value in the study of intra- and intermolecular hydrogen bonding. Formation of a hydrogen bond will cause any stretching vibration to broaden and shift to lower wavenumber [57]. For a macromolecule, there are very many vibrational transitions. However, many of the vibrations can be assigned to particular bonds or groupings. This forms the basis of characteristic group frequencies. The main experimental parameter is the position of the maximum of the absorption band, in cm1. 2.2. FTIR to Study Lipids FTIR is a powerful, non-perturbing technique that has been used for the detection and characterization of lipid phase transitions in model membranes. The technique is quite versatile, covering a wide range of applications, from which detailed information about the structure and organization of biomembranes and lipid assemblies can be obtained [35]. FTIR has been widely used to study lipid phase and orientation in liposomal [8,47] and monolayer [40] systems. In a recent study, FTIR was used to probe the effect of cholesterol on membrane permeance [8]. Cholesterol increases the headgroup hydration and decreases interchain order within the liposome bilayer. Deuterated lipids can be used to reveal the melting of individual lipid components in binary lipid mixtures [20]. Two-dimensional infrared spectroscopy has also been used to probe structure and dynamics in binary lipid systems [58]. The technique was used to visualize the segregation of lipid components in sphingomyelin/phospholipid liposomes. Figure 2 shows an FTIR spectrum of palmitoyl-oleoyl-phosphatidylcholine (POPC), with the main characteristic bands indicated in the figure and in Table 1. The symmetric and asymmetric C-H stretching vibrations of the lipid acyl chains provide an
276
Table 1. Band positions of characteristic molecular groups commonly found in phospholipids Characteristic group vibration terminal CH3 antisymmetric stretch chain CH2 antisymmetric stretch terminal CH3 symmetric stretch chain CH2 symmetric stretch ester C=O stretch CH2 scissor CH2 bend
+
Wavenumber (cm1) 2957 2921 2873 2851 1743 1468 1420 1396 1380 1223 1173 1088 1068 970
NCH3 symmetric deformation
terminal CH3 symmetric bend PO2 antisymmetric stretch ester COC antisymmetric stretch PO2 symmetric stretch ester COC symmetric stretch
+
NCH3 antisymmetric stretch
evaluation of the conformational order in lipid bilayers [38]. The packing arrangement of lipid acyl chains can also be studied based on correlation field splittings of the CH 2 scissoring mode [52]. The phosphate stretching band and the ester carbonyl C=O band of PCs exhibits hydration-dependent wavenumber shifts during thermotropic or lyotropic phase transitions [50,10]. 2.3. FTIR to Study Lipid Protein Interactions FTIR is also widely used for conformational analysis of peptides in a range of environments. Measurements can be performed in aqueous solution, organic solvents, detergent micelles as well as in phospholipid membranes [27]. Information on the secondary structure of peptides can be derived from the analysis of the strong amide-I band [32]. Orientation of secondary structural elements within a lipid bilayer matrix can be determined by means of polarized attenuated total reflectance-FTIR spectroscopy [15,39]. Hydrogen-deuterium exchange of the protein amide-II band in D2O provides an additional tool to study structure and solvent accessibility of membrane proteins [19]. The advantage of FTIR is that it can be used to simultaneously study lipid phase and protein conformation. Fore example, the effect of antimicrobial peptides (AMPs), produced by certain amphibians, on biomembranes can be studied [51]. These AMPs generally kill their microbial target cells primarily by destabilizing the lipid bilayers of their cell membranes. The conformation of these antimicrobial peptides when bound to lipid bilayers has been identified, as well as the effect of these peptides on the organization of the host phospholipid bilayer.
277
2.4. FTIR Sampling Techniques Transmission FTIR is widely used to study biomolecules in water. The strong absorbance of water in the amide-I region (16001700 cm1), however, complicates the analysis of proteins in water. In order to record spectra in H2O, short pathlength cells need to be employed. Consequently, a relatively high sample concentration is needed. However, D2O and many organic solvents do not have strong absorption bands in the amide-I region and in these cases much longer pathlength cells can be used. It is necessary to accurately subtract not only the overlapping absorption of the solvent, but also of any absorption from buffers, from the spectrum of the sample. This is especially critical for measurements with samples in H2O [27]. Attenuated total reflectance (ATR) FTIR is another widely used method. ATRFTIR is a very powerful tool to gain information on the structure of biological molecules [21,22]. This technique can be extended to study the orientation of components of thin films, such as lipids and peptides in biomembranes. An advantage of ATR-FTIR to study the structure of biomembranes is that the membrane can be deposited on the surface of the internal reflection element as a thin film of highly oriented membranes by evaporation of the water. Using parallel and perpendicular polarized incident light, it is possible to detect changes in the orientation of a number of chemical bonds belonging to lipids and proteins [23]. Changes in the secondary structure of proteins can be evaluated from the amide-I band shape. A broad range of FTIR accessories is available to overcome the problems of studying biological samples including diffuse reflection, grazing angle specular reflection and FTIR microscopes. Coupling of optical microscopes to an FTIR spectrometer enables the study of very small quantities of biological samples and can be used for biochemical mapping of tissues. FTIR microspectroscopy has been increasingly applied to study human tissues and tissue pathologies [34]. New technical developments such as multichannel detectors or synchrotron IR microspectroscopy permit high-quality infrared microspectroscopic imaging of biomedical samples [34].
3. Biomembranes Studied by FTIR FTIR not only has been used to study lipid phase behavior of model membranes, but also to study membrane phase behavior of biomembranes in intact cells. Biomembranes can be studied by performing temperature scanning FTIR measurements on whole cells. The basis for FTIR studies on cells and tissues are the characteristic group frequencies of endogenous biomolecules. The main experimental parameter is the position of the maximum of the absorption band. The temperature dependence of a molecular vibration can be used to verify further the assignment of the type of biomolecule that is observed. IR spectra of biological cells and tissues as a function of temperature show shifts of bands, associated with the melting of lipids in membranes (CH-stretching vibrations), with protein denaturation (C=O stretching vibration) and with the melting of cytoplasmic glasses (OH-stretching vibration), which can be measured simultaneously [60,63]. FTIR has an advantage over other methods such as DSC that, besides thermodynamic parameters, information can be derived on molecular conformation and intra- and intermolecular interactions. In the next section, an overview is given of how FTIR spectroscopy can be applied to study biothermodynamic properties of cells that have important biomedical applications such as red blood cells and platelets.
278
We focus on the following subjects: FTIR spectra of mammalian cells (red blood cells, platelets and tumor cells), thermotropic phase behavior of cellular membranes studied by FTIR; and membrane phase behavior during freezing and drying studied by FTIR. The latter two subjects involve cooling and heating scans on cell pellets, from which lipid phase changes and phase changes of water into ice during freezing can be derived simultaneously. FTIR spectra were recorded using an IR-spectrometer equipped with a liquid N2-cooled MCT detector, as previously described (see [62,61,59] for details). Samples were assayed in transmission mode in a temperature controlled sample holder. Samples of hydrated cell pellets were loaded between two CaF2 windows and sealed to avoid dehydration of the sample. As the experimental materials, red blood cells, platelets, and tumor cells were used. 3.1. Thermotropic Membrane Phase Behavior of Mammalian Cells FTIR has been applied to the qualitative and quantitative study of molecular conformational order in models for biological membranes [38]. Much of our current understanding of biological membranes is derived from studies on human erythrocytes. The asymmetric distribution of lipid classes across the human erythrocyte membrane is well established [17,49]. The preservation of lipid asymmetry is important in the maintenance of the biconcave shape of the erythrocyte [16] as well as other cellular functions [65,9]. Studies on intact human erythrocytes have demonstrated the utility of FTIR spectroscopy as a unique biophysical technique for obtaining molecular conformation information from live cells [43]. These studies have utilized the C-H stretching modes of the erythrocyte membrane acyl chains to evaluate conformational order in cells. Such studies provide an overview of the conformational status of the entire membrane. Experiments with deuterated lipids provide more detailed information on the conformational status of specific lipid classes in erythrocyte membranes [42,44,45]. Figure 3 depicts IR absorbance spectra of porcine platelets and erythrocytes. Overall, the IR spectrum of the cells in media is dominated by the signal from water. Water exhibits strong vibrational bands at around 3300 cm1, 2200 cm1, and 1650 cm1 arising from stretching, libration and bending combination, and scissoring vibrational modes, respectively. In the 30002800 cm1 region, the symmetric and asymmetric CH2 stretching vibrations of lipid acyl chains are visible. Characteristic protein bands are visible at 1655 cm1 (amide-I band) and at 1550 cm1 (amide-II band). The amide-I band overlaps with the scissoring vibrational mode of water. In the amide-III region of the spectrum (1330 and 1200 cm1), several weak bands are visible. The 3000 2800 cm1 region (Fig. 3 B and C) can be used for lipid analysis, particularly the symmetric CH2 stretching band. The symmetric CH2 stretching band arising from lipid acyl chains is clearly visible in platelets, but is relatively weak in erythrocytes. In erythrocytes the protein CH3 band is relatively strong, due to the extremely high protein content of these cells. In order to study membrane phase behavior of intact cells, FTIR spectra have to be recorded as a function of temperature. Figure 4 shows an example of spectra that were recorded of platelets during heating from 5 to 90 C. Not much detail can be seen in the whole spectra, other than the strong contribution of the water bands at 3300, 2200, and 1650 cm1. Enlargement of the C-H stretching region between 3000 and 2800 cm1 reveals the relevant information on membrane conformational disorder
279
Figure 3. Infrared absorption spectra of porcine erythrocytes (upper traces) and platetelets (lower traces). (A) Whole spectra in the 4000900 cm1 region. (B) Enlargement of the 30002800 cm1 region. (C) Second derivative spectra in the 30002800 cm1 region. Data are adapted from [61].
(Fig. 5A). Second derivative analysis is a useful tool to analyze hydrated cell pellets (Fig. 5B), because it increases the resolution and removes the base line effects that are due to the contribution of the strong OH stretching band of water (at ~3300 cm1). The band position of the symmetric CH2 stretching band, plotted as a function of temperature reveals the thermotropic response of the platelet membranes (Fig. 5C). The band position of the symmetric CH2 stretching vibration reveals that the average conformational state of the membrane is somewhere between a fully ordered state (~2849 2850 cm1) and a fully disordered state (~28532854 cm1). The gradually increasing wavenumber as the temperature is increased from 0 to 90 C represents the progressive occurrence of conformational disorder. First derivative analysis of the wavenumber versus temperature plot shows inflections more clearly and reveals multiple transitions, likely reflecting the sequential melting of different lipid classes in the membrane. The generally non cooperative phase transitions in mammalian cells are due to the cholesterol in the membrane, which has been demonstrated in a variety of cases [25,61,37]. When cholesterol is depleted from the membranes of human platelets
280
Figure 4. Infrared absorption spectra of porcine platelets during a temperature scan from 5 to 90 C. Details are described in [61].
by methyl--cyclodextrin (MCD), a clear cooperative phase transition becomes visible (Fig. 6A). This is also seen in the cholesterol enriched raft fraction. The lower CH2 of isolated rafts suggests that they are more ordered compared to the overall conformational order of the membranes in the intact platelets. Rafts also exhibit a monotonous increase in conformational disorder with increasing temperature. When cholesterol is removed from the rafts, a clear transition at around 35 C becomes visible, which reflects the melting of sphingolipids in the rafts. FTIR studies on platelets have been used to correlate membrane phase behavior with cold-induced activation of platelets at 4 C [56]. Because of this cold-induced activation, platelets can only be stored for a limited time at 22 C (5 days), which is a major problem in platelet banking and transport. FTIR studies have also been used to study membrane phase behavior in freeze-dried platelets [64,62]. 3.2. Lyotropic Membrane Phase Transitions Not only temperature, but also dehydration can initiate lipid phase changes [14]. This has been studied in detail in liposomal systems. Palmitoyl-oleoyl-phosphatidylcholine (POPC), for example exhibits a Tm of 0 C in the hydrated state, but dehydration causes the Tm to increase by more than 70 C [47]. This implies that when POPC lipids are dried at room temperature, they undergo a lyotropic liquid crystalline to gel phase transition. Lyotropic effects on lipid phase behavior are to a certain extent dependent on the polarity and functional groups of the phospholipid head group. Palmitoyl-oleoylphosphatidylglycerol (POPG), which has the same acyl chains compared to POPC, is less affected by dehydration, because the phosphate and glycerol moiety of the head group form intramolecular hydrogen bonds upon dehydration [36].
281
Figure 5. Data-processing of temperature-dependent FTIR spectra from porcine platelets. (A) Infrared absorption spectra in the 30002800 cm1 region as a function of temperature. (B) Second derivative spectra in the 28652835 cm1 region. (C) CH2 versus temperature plot (filled circles) and first derivative (solid line). Data are adapted from [61].
Figure 6. Membrane phase transitions of control and MCD treated platelets (A) and isolated rafts (B). The data points reflect wavenumber (CH2) vs. temperature plots of intact control (filled circles) or MCD treated samples (open circles). Data are adapted from [25].
282
Figure 7. Membrane phase behavior of hydrated and air-dried multi lamellar POPC vesicles. The data points reflect those of hydrated vesicles (filled circles), and vesicles that were dried in the absence (open circles) or presence (filled squares) of sucrose. Data are adapted from [46].
3.3. Stabilization of Liposomes and Water Replacement Hypothesis Liposomes that are used for drug delivery can be stabilized in the dried state by drying them in the presence of protectants. Non-reducing disaccharides such as sucrose and trehalose are particularly effective in stabilization of liposomes in the dried state [13]. This yields a shelf stable dried product of liposomes that can be encapsulated with a drug. Drying liposomes in the absence of protectants can be very damaging due to lyotropic phase transitions. Lyotropic phase transitions cause leakage of intraliposomal content [14], which can be circumvented when certain sugars are added to protect the liposomes during drying. The effects of drying on lipid phase behavior is illustrated in Fig. 7. While hydrated liposomes have a Tm at approximately 0 C and the air-dried liposomes at 60 C, the Tm of liposomes dried in the presence of sucrose is 21 C, far below that of the hydrated control. By interacting with the polar headgroups of phosholipids, sucrose and certain other sugars can depress Tm of model membranes [11,12]. Figure 8 illustrates how sugars (particularly sucrose and trehalose) can prevent the dehydration-induced increase of Tm in liposomes. When liposomes are dried in the presence of protective sugars, the lipid bilayer remains in the liquid crystalline phase during dehydration, which is one of the prerequisites for the protection of liposomes in the dry state [13]. These studies have important pharmaceutical applications for longterm storage of liposomes that are used for drug delivery. 3.4. Effects of Freezing on Biomembranes: Cryopreservation and Cryosurgery Lyotropic membrane phase transitions also play a role during cryopreservation and cryosurgery [59]. Cryopreservation relies on cryogenic temperatures and cryoprotective agents to preserve cells or tissues for prolonged periods of time in the frozen state [1], whereas cryosurgery relies on the damaging effects of freezing and cryoadjuvants to destroy i.e. tumors [30]. A two factor hypothesis of freeze injury solution effects injury due to exposure to increasing solute concentrations during slow freezing, and intracellular ice formation (IIF) due to cytoplasmic supercooling and consequent ice nucleation during rapid freezing [41].
283
Figure 8. Schematic representation illustrating the effects of drying on lipid conformation in the dried state. The grey filled circles represent water and the grey filled squares represent sucrose. Drying in the absence of a protectant (sucrose) results in a liquid crystalline to gel phase transition during dehydration and vice versa during rehydration. The liquid crystalline phase is maintained during drying (and rehydration) in presence of a protectant. This idea has been postulated as the water replacement hypothesis [11].
Disruption of the plasma membrane is one of the primary causes of freezing injury. The mechanisms of injury resulting from freezing include destabilization of the plasma membrane resulting from freeze-induced cell dehydration and lyotropic phase transitions in the plasma membrane lipids [24]. The latter could trigger irreversible, lateral phase separations, which destabilize the membrane bilayer structure, leading to loss of membrane integrity and eventually to cell death [55,18]. Freezing-induced membrane injury is due to the reorganization of lipids during freezing and changes in membrane conformational disorder due to phase transitions from liquid crystalline to gel phase [59] or from liquid crystalline to hexagonal phase [55]. In a recent FTIR study on prostate tumor cells, we have shown that ice formation can induce strong lyotropic membrane phase transitions in mammalian cells [59]. The effect of ice formation on membrane phase behavior primarily depends on the cooling rate and the ice nucleation temperature, which determine the extent of cellular dehydration and the incidence of intracellular ice formation (Fig. 10). Under cooling and freezing conditions that cause cellular dehydration, cellular membranes display a highly cooperative liquid crystalline to gel phase transition coinciding with ice formation in the system (Fig. 9). The thermotropic response of the symmetric CH2 stretching vibration shows that the membrane phase behavior of LNCaP tumor cells during cooling is affected by the nucleation temperature (Fig. 9). When the sample is nucleated at 3 C, the membranes undergo a highly co-operative phase transition with an onset
284
Figure 9. Membrane phase behavior of LNCaP tumor cells during cooling and nucleation at various subzero temperatures. The data points reflect CH2 during freezing the cells down to 80 C. Samples were nucleated at 3 C (filled circles), 6 C (filled triangles) or at 10 C (open circles). Data are adapted from [59].
Figure 10. Schematic representation illustrating the effects of freezing on cells and cellular membranes. The small grey filled circles represent water and the star shaped asterisks represent ice crystals. Slow freezing and high sub zero nucleation result in cellular dehydration and gel phase formation in membranes. Fast freezing and low ice nucleation temperatures (supercooling) result in intracellular ice formation. Membranes remain relatively hydrated and the residual conformational disorder of membranes is greater compared to dehydrated cells.
temperature that coincides with the nucleation temperature of ice in the system. Nucleation at 10 C, however, results in a much less co-operative phase transition. In addition, CH2 is considerably higher at 80 C when the sample is nucleated at 10 C, indicating greater residual conformational disorder compared to the sample nucleated at 3 C. Intermediate membrane phase behavior is observed when the sample is nucleated at 6 C. The effect of the nucleation temperature on membrane phase behavior
285
during freezing can directly be correlated with cell survival after thawing [59]. Optimal survival is found at intermediate nucleation temperatures in between cellular dehydration and intracellular ice formation. FTIR studies thus provide a powerful tool to study cell behavior during freezing, which can be used either to optimize cryopreservation protocols for cells or to maximize damage during cryosurgery.
4. Concluding Remarks Understanding the complex behavior of biomembranes during heating, freezing and drying is directly relevant for thermal processing of cells such as is done in cryopreservation and cryosurgery. FTIR is a unique technique to study membrane phase behavior of cells over a wide temperature and hydration regime, including cells that are exposed to heating, freezing or drying. In situ FTIR studies can give information on conformational disorder and stability of membranes in cells. By combining FTIR studies with viability assays, cell stability can be correlated with membrane phase behavior.
References
[1] J.P. Acker, Biopreservation of cells and engineered tissues, Advances in Biochemical Engineering/ Biotechnology 103 (2007) 157-187. [2] S.N. Ahmed, D.A. Brown, E. London, On the origin of sphingolipid/cholesterol-rich detergentinsoluble cell membranes: physiological concentrations of cholesterol and sphingolipid induce formation of a detergent-insoluble, liquid-ordered lipid phase in model membranes, Biochemistry 36 (1997) 10944-10953. [3] R. Bittman, C.R. Kasireddy, P. Mattjus, J.P. Slotte, Interaction of cholesterol with sphingomyelin in monolayers and vesicles, Biochemistry 33 (1994) 11776-11781. [4] D.A. Brown, E. London, Functions of lipid rafts in biological membranes, Annual Review of Cell and Developmental Biology 14 (1998) 111-136. [5] D.A. Brown, E. London, Structure and function of sphingolipid- and cholesterol-rich membrane rafts, Journal of Biological Chemistry 275 (2000) 17221-17224. [6] M. Caffrey, The combined and separate effects of low temperature and freezing on membrane lipid mesomorphic phase behavior: relevance to cryobiology, Biochimica et Biophysica Acta 896 (1987) 123-127. [7] M. Caffrey, J. Hogan, LIPIDAT: A database of lipid phase transition temperatures and enthalpy changes. DMPC data subset analysis, Chemistry and Physics of Lipids 61 (1992) 1-109. [8] C. Chen, C.P. Tripp, An infrared spectroscopic based method to measure membrane permeance in liposomes. Biochimica et Biophysica Acta 1778 (2008) 2266-2272. [9] W.E. Conner, D.S. Lin, G. Thomas, F. Ey, T. DeLoughery, N. Zhu, Abnormal, Abnormal phospholipids molecular species of erythrocytes in sickle cell anemia, Journal of Lipid research 38 (1997) 2516-2528. [10] L.M. Crowe, J.H. Crowe, D. Chapman, Interaction of carbohydrates with dry dipalmitoylphosphatidylcholine, Archives of Biochemistry and Biophysics 236 (1985) 289-296. [11] J.H. Crowe, F.A. Hoekstra, L.M. Crowe, Anhydrobiosis, Annual Review of Physiology 54 (1992) 579-599. [12] J.H. Crowe, F.A. Hoekstra, K.H. Nguyen, L.M. Crowe, Is vitrification involved in depression of the phase transition temperature in dry phospholipids?, Biochimica et Biophysica Acta 1280 (1996) 187-196. [13] J.H. Crowe, S.B. Leslie, L.M. Crowe, Is vitrification sufficient to preserve liposomes during freezedrying?, Cryobiology 31 (1994) 355-366. [14] J.H. Crowe, A.E. Oliver, F.A. Hoekstra, L.M. Crowe, Stabilization of dry membranes by mixtures of hydroxyethyl starch and glucose: the role of vitrification, Cryobiology 35 (1997) 20-30. [15] N. Dave, V.A. Lrenz-Fonfra, G Leblanc, E. Padrs, FTIR spectroscopy of secondary-structure reorientation of melibiose permease modulated by substrate binding, Biophysical Journal 94 (2008) 3659-3670.
286
[16] P.F. Devaux, Static and dynamic lipid asymmetry in cell membranes, Biochemistry 30 (1991) 1163-1173. [17] P.F. Devaux, A. Zachowski, Maintenance of membrane phospholipids asymmetry, Chemistry and Physics of Lipids 73 (1994) 107-120. [18] E.Z. Drobnis, L.M. Crowe, T. Berger, T.J. Anchordoguy, J.W. Overstreet, J.H. Crowe, Cold shock damage is due to lipid phase transitions in cell membranes: a demonstration using sperm as a model, Journal of Experimental Zoology 265 (1993) 432-437. [19] E. Dzafi, O. Klein, E. Screpanti, C. Hunte, W. Mntele, Flexibility and dynamics of NhaA Na(+)/H(+)-antiporter of Escherichia coli studied by Fourier transform infrared spectroscopy, Spectrochimica Acta. Part A, Molecular and Biomolecular Spectroscopy 72 (2009) 102-109. [20] M. Fidorra, T. Heimburg, H.M. Seeger, Melting of individual lipid components in binary lipid mixtures studied by FTIR spectroscopy, DSC and Monte Carlo simulations. Biochimica et Biophysica Acta (2009), doi:10.1016/j.bbamem.2008.12.003. [21] E. Goormaghtigh, V. Cabiaux, J.M. Ruysschaert, Determination of soluble and membrane protein structure by Fourier transform infrared spectroscopy. I. Assignments and model compounds, Subcellular Biochemistry 23 (1994) 329-362. [22] E. Goormaghtigh, V. Cabiaux, J.M. Ruysschaert, Determination of soluble and membrane protein structure by Fourier transform infrared spectroscopy. II. Experimental aspects, side chain structure, and H/D exchange, Sub-cellular Biochemistry 23 (1994) 363-403. [23] E. Goormaghtigh, V. Raussens, J.M. Ruysschaert, Attenuated total reflection infrared spectroscopy of proteins and lipids in biological membranes, Biochimica et Biophysica Acta 1422 (1999) 105-185. [24] W.J. Gordon-Kamm, P.L. Steponkus, Lamellar-to-hexagonalII phase transitions in the plasma membrane of isolated protoplasts after freeze-induced dehydration, Proceedings of the National Academy of Sciences of the United States of America 81 (1984) 6373-6377. [25] K. Gousset, W.F. Wolkers, N.M. Tsvetkova, A.E. Oliver, C.L. Field, N.J. Walker, J.H. Crowe, F. Tablin, Evidence for a physiological role for membrane rafts in human platelets, Journal of Cellular Physiology 190 (2002) 117-128. [26] S.M. Gruner, Intrinsic curvature hypothesis for biomembrane lipid composition: a role for nonbilayer lipids, Proceedings of the National Academy of Sciences of the United States of America 82 (1985) 3665-3669. [27] P.I. Haris, D. Chapman, The conformational analysis of peptides using Fourier transform IR spectroscopy. Biopolymers 37 (1995) 251-263. [28] K. Harlos, H. Eibl, Hexagonal phases in phospholipids with saturated chains: phosphotidylethanolamines and phosphatidic acids, Biochemistry 20 (1981) 2888-2892. [29] X. He, W.F. Wolkers, J.H. Crowe, D.J. Swanlund, J.C. Bischof, In situ thermal denaturation of proteins in Dunning AT-1 prostate cancer cells: implications for hyperthermic cell injury, Annals of Biomedical Engineering 32 (2004) 1384-1398. [30] N.E. Hoffmann, J.C. Bischof, The cryobiology of cryosurgical injury, Urology 60 (2002) 40-49. [31] J.N. Israelachvili, S. Marcelja, R.G. Horn, Physical principles of membrane organization, Quaterly Reviews of Biophysics 13 (1980) 121-200. [32] E.M. Jones, K. Surewicz, W.K. Surewicz, Fibrillization and prion amyloid propagation in vitro, The Journal of Biological Chemistry 281 (2006) 8190-8196. [33] R. Koynova, M. Caffrey, Phases and phase transitions of the phosphatidylcholines, Biochimica et Biophysica Acta 1376 (2001) 91-145. [34] P. Lasch, D. Naumann, Spatial resolution in infrared microspectroscopic imaging of tissues Biochimica et Biophysica Acta 1758 (2006) 814-829. [35] R.N. Lewis, R.N. McElhaney, Fourier transform infrared spectroscopy in the study of lipid phase transitions in model and biological membranes: practical considerations. Methods Molecular Biology 400 (2007) 207-226. [36] L.J. Linders, W.F. Wolkers, F.A. Hoekstra, K. van t Riet, Effect of added carbohydrates on membrane phase behavior and survival of dried Lactobacillus plantarum, Cryobiology 35 (1997) 31-40. [37] A. Luria, V. Vegelyte-Avery, B. Stith, N.M. Tsvetkova, W.F. Wolkers, J.H. Crowe, F. Tablin, R. Nuccitelli, Detergent-free domain isolated from Xenopus egg plasma membrane with properties similar to those of detergent-resistant membranes, Biochemistry 41 (2002) 13189-13197. [38] H.H. Mantsch, R.N. McElhaney, Phospholipid phase transitions in model and biological membranes as studied by infrared spectroscopy, Chemistry and Physics of Lipids 57 (1991) 213-226. [39] D. Marsh, Orientation and peptide-lipid interactions of alamethicin incorporated in phospholipids membranes: polarized infrared and spin-label EPR spectroscopy, Biochemistry 48 (2009) 729-737. [40] J. Mascetti, S. Castano, D. Cavagnat, B. Desbat, Organization of -Cyclodextrin under pure cholesterol, DMPC, or DMPG and mixed cholesterol/phospholipid monolayers. Langmuir 24 (2008) 9616-9622.
287
[41] P. Mazur, S.P. Leibo, E.H. Chu, A two-factor hypothesis of freezing injury. Evidence from Chinese hamster tissue-culture cells, Experimental Cell Research 71 (1972) 345-355. [42] D.J. Moore, S. Gioioso, R.H. Sills, R. Mendelsohn, Some relationships between membrane phospholipids domains, conformational order, and cell shape in intact human erythrocytes, Biochimica et Biophysica Acta 1415 (1999) 342-348. [43] D.J. Moore, R.H. Sills, R. Mendelsohn, Peroxidation of erythrocytes FTIR spectroscopy studies of extracted lipids, isolated membranes, and intact cells, Biospectroscopy 1 (1995) 133-140. [44] D.J. Moore, R.H. Sills, R. Mendelsohn, Conformational order of specific phospholipids in human erythrocytes: correlations with changes in cell shape, Biochemistry 36 (1997) 660-664. [45] D.J. Moore, R.H. Sills, N. Patel, R. Mendelsohn, Conformational order of phospholipids incorporated into human erythrocytes: an FTIR spectroscopy study, Biochemistry 35 (1996) 229-235. [46] H. Oldenhof, W.F. Wolkers, F.Fonseca, S. Passot, M. Marin, Effect of sucrose and maltodextrin on the physical properties and survival of air-dried Lactobacillus bulgaricus: an in situ Fourier transform infrared spectroscopy study, Biotechnology Progress 21 (2005) 885-892. [47] A.V. Popova, D.K. Hincha, Effects of the sugar headgroup of a glycoglycerolipid on the phase behavior of phospholipids model membranes in the dry state, Glycobiology 15 (2005) 1150-1155. [48] M.B. Sankaram, T.E. Thompson, Modulation of phospholipid acyl chain order by cholesterol. A solidstate 2H nuclear magnetic resonance study, Biochemistry 29 (1990) 10676-10684. [49] A.J. Schroit, R.F.A. Zwaal, Transbilayer movement of phospholipids in red cells and platelet membranes, Biochimica et Biophysica Acta 1071 (1991) 313-329. [50] C. Selle, W. Pohle, Fourier transform infrared spectroscopy as a probe for the study of the hydration of lipid self-assemblies. II. Water binding versus phase transitions, Biospectroscopy 4 (1998) 281-294. [51] G.W. Seto, S. Marwaha, D.M. Kobewka, R.N. Lewis, F. Separovic, R.N. McElhaney, Interactions of the Australian tree frog antimicrobial peptides aurein 1.2, citropin 1.1 and maculatin 1.1 with lipid model membranes: differential scanning calorimetric and Fourier transform infrared spectroscopic studies, Biochimica et Biophysica Acta 1768 (2007) 2787-2800. [52] D.J. Siminovitch P.T. Wong, H.H. Mantsch, Effects of cis and trans unsaturation on the structure of phospholipids bilayers: a high-pressure infrared spectroscopic study, Biochemistry 26 (1987) 32773287. [53] K. Simons, E. Ikonen, Functional rafts in cell membranes, Nature 387 (1997) 569-72. [54] S.J. Singer, G.L Nicolson, The fluid mosaic model of the structure of cell membranes, Science 175 (1972) 720-731. [55] P.L. Steponkus, D.V. Lynch, Freeze/thaw-induced destabilization of the plasma membrane and the effect of cold acclimation, Journal of Bioenergetics and Biomembranes 21 (1989) 21-41. [56] F. Tablin, A.E. Oliver, N.J. Walker, L.M. Crowe, J.H. Crowe, Membrane phase transition of intact human platelets: correlation with cold-induced activation, Journal of Cellular Physiology 168 (1996) 305313. [57] S.N. Vinogradov, R.N. Linnell, Hydrogen Bonding, Van Nostrand Reinhold Co, New York (1971). [58] V.V. Volkov, R. Chelli, R. Righini, Domain formation in lipid bilayers probed by two-dimensional infrared spectroscopy. Journal of Physical Chemistry B 110 (2006) 1499-1501. [59] W.F. Wolkers, S.K. Balasubramanian, E.L. Ongstad, H.C. Zec, J.C. Bischof, Effects of freezing on membranes and proteins in LNCaP prostate tumor cells, Biochimica et Biophysica Acta 1768 (2007) 728-736. [60] W.F. Wolkers, F.A. Hoekstra, In situ FTIR assessment of desiccation-tolerant tissues, Spectroscopy 17 (2003) 297-313. [61] W.F. Wolkers, S.A. Looper, R.A. Fontanilla, N.M. Tsvetkova, F. Tablin, J.H. Crowe, Temperature dependence of fluid phase endocytosis coincides with membrane properties of pig platelets, Biochimica et Biophysica Acta 1612 (2003) 154-163. [62] W.F. Wolkers, S.A. Looper, A.E. McKieman, N.M. Tsvetkova, F. Tablin, J.H. Crowe, Membrane and protein properties of freeze-dried mouse platelets, Molecular Membrane Biology 19 (2002) 201-210. [63] W.F. Wolkers, A.E. Oliver, F. Tablin, J.H. Crowe, A Fourier-transform infrared spectroscopy study of sugar glasses, Carbohydrate research 339 (2004) 1077-1085. [64] W.F. Wolkers, N.J. Walker, F. Tablin, J.H. Crowe, Human platelets loaded with trehalose survive freeze-drying, Cryobiology 42 (2001) 79-87. [65] A. Zachowski, Phospholipids in animal eukaryotic membranes: transverse asymmetry and movement, Biochemical Journal 294 (1993) 1-14.
288
Structural Analysis of Protein-DNA and Protein-RNA Interactions by FTIR Spectroscopy

H.A. TAJMIR-RIAHI *, C.N. NSOUKPO-KOSSI and D. JOLY Department of Chemistry-Biology, University of Qubec at Trois-Rivires, C.P. 500, Trois-Rivires (Qubec) Canada G9A 5H7
Abstract. In this chapter the fundamental question of how does protein-DNA or protein-RNA interaction affect the structures and dynamics of DNA, RNA and protein is addressed. Models for calf-thymus DNA and transfer RNA interactions with human serum albumin (HSA), ribonuclease A (RNase A) and deoxyribonuclease I (DNase I) are presented here, using Fourier Transform Infrared (FTIR) spectroscopy in conjunction with UV-visible and CD spectroscopic methods. In the models considered, the binding sites, stability and structural aspects of proteinDNA and protein-RNA are discussed and the effects of protein interaction on the secondary structures of DNA and protein were determined. Keywords. DNA, protein, binding mode, binding constant, secondary structure, FTIR spectroscopy Abbreviations. HSA, human serum albumin; RNase A, ribonuclease A; DNase I, deoxyribonuclease I; G, guanine; A, adenine; T, thymine; U, uracil; FTIR, Fourier Transform infrared; CD, circular dichroism
Introduction Fourier Transform Infrared (FTIR) spectroscopy has widespread application to qualitative and quantitative analyses in Chemistry, Biochemistry, Biology, Medicinal Chemistry and Environmental Science. Its single most important use has been for the identification of organic compounds, drugs, and pollutants. But now, FTIR spectroscopy is an established method for the structural characterization of proteins, DNA and RNA. For instance, FTIR spectroscopy with its secondary derivatives were used to determine protein conformation. Furthermore, changes in the secondary structure of proteins as a function of changes in pH, solvent composition, temperature, ligand binding, and exposure to DNA, RNA and lipids or other compounds in solution (e.g. drugs) have also been investigated. Similarly, DNA and RNA conformational transitions induced by ligand, drug or protein interactions were studied by FTIR spectroscopy. In addition, the application of FTIR in conjunction with Mass Spectrometry (MS), Ultra-Violet Spectrophotometer (UV), Nuclear Magnetic Resonance (NMR) and CD spectroscopy has been the most powerful
*
Corresponding Author: Tel: 819-376-5052 (ext. 3310); fax: 819-376-5084, E-mail: tajmirri@uqtr.ca.
H.A. Tajmir-Riahi et al. / Structural Analysis of Protein-DNA and Protein-RNA Interactions
289
method of identifying the chemical structures or subtle structural variations in biomacromolecules induced by ligand or drug interaction. DNA-protein and RNA-protein interactions play important roles in a variety of biomolecular functions. Gene expression, transcription, replication, recombination, packaging and repairs all are controlled by DNA-protein interactions. Although the physical basis for these recognition processes is not fully understood, x-ray crystallography, NMR spectroscopy and molecular modeling provide us with a wealth of information on DNA recognitions [15]. The quantitative assessment of DNA-protein interaction is essential to understanding transcription, the beginning of biological processes including normal cellular function, development and many diseases [6]. RNA plays major role in diverse functions within the cell. Protein-RNA complexation is essential in many of these biological functions. Transfer RNAs bind to aminoacyl-tRNA synthetases for the translation of the genetic code during protein synthesis [7,8], while ribonucleoproteins bind RNA in post-transcriptional regulation of gene expression [9]. Although the biological significance of protein complexation with RNA has been well recognized, the specific mechanism of protein-RNA interaction is not fully understood [10]. Measurement of sequencespecific DNA-protein and RNA-protein interactions is a key experimental procedure in molecular biology of gene regulation. The most commonly used method is the electrophoretic gel mobility shift assay (EMSA), in which a radioactively labeled DNA probe is mixed with a solution of protein of interest and after a short reaction period, loaded on an electrophoretic gel [11]. Since there are some limitations using EMSA method, affinity capillary electrophoresis has been widely used to measure the mobility shift or zone electrophoresis to estimate the affinity of DNA-protein complexation [1214]. Human serum albumin (Fig. 1A) is a principal extracellular protein with a high concentration in blood plasma (40 mg/ml) [1518]. HSA is a globular protein composed of three structurally similar domains (I, II and III), each containing two subdomains (A and B) and stabilized by 17 disulphide bridges [1922]. Aromatic and heterocyclic ligands were found to bind within two hydrophobic pockets in sub-domains IIA and IIIA, namely site I and site II [1922]. Seven binding sites are localized for fatty acids in sub-domains IB, IIIA, IIIB and on the sub-domains interfaces [16]. HSA has also a high affinity metal binding site at the N-terminus [15]. The multiple binding sites underlie the exceptional ability of HSA to interact with many organic and inorganic molecules and make this protein an important regulator of intercellular fluxes, as well as the pharmacokinetic behavior of many drugs [1522]. RNase A (Fig. 1B) catalyzes the cleavage of P-O5 bonds in RNA on the 3 side of pyrimidine to form cyclic 2,5-phosphates under specific conditions [2325]. Oligomers of RNase A exhibit antitumor activity [26]. In recent years, RNase A has been the molecular target for the development of small inhibitors. These inhibitors are developed to restrain biological activity of different RNase A homologs in a variety of pathological conditions [2732]. DNase I (Fig. 1C) hydrolyses double stranded DNA predominantly by a singlestranded nicking mechanism under physiological conditions in the presence of Mg 2+ and Ca2+ [3335]. X-ray structural analysis of DNase I-d(GCGATCGC)2 and Dnase I-d(GGTATACC)2 complexes showed that protein binds to the minor groove and the backbone phosphate group with no contact with the major groove of the right-handed DNA duplexes, altering groove geometry as well as causing distortion of B-DNA conformation [3639].
290
HSA
RNase A
DNase I Figure 1. Ribbon structure of human serum albumin (HSA) (A), RNase A (B), and DNase I (C) derived from their crystal structure determined by X-ray diffraction at 2.5 (ref.17), 2.32 (ref. 58), and 2.8 (ref. 59), respectively.
Structural analysis of calf-thymus DNA and transfer RNA interactions with HSA, RNase A and DNase I are presented here, using Fourier Transform Infrared (FTIR) in conjunction with UV-visible and CD spectroscopic methods. The binding sites, stability and structural aspects of protein-DNA and protein-RNA complexes are discussed and the effects of protein interaction on the secondary structures of polynucleotides and protein are reported here.
Experimental Section Materials Highly polymerized type I calf-thymus DNA sodium salt (7% Na content) and Baker Yeast tRNA were purchased from Sigma Chemical Co., and deproteinated by the addition of CHCl3 and isoamyl alcohol in NaCl solution. In order to check the protein content of DNA and RNA solutions, the absorbance at 260 and 280 nm was recorded. The A260/A280 ratio was 1.85 for DNA and 2.1 for RNA showing that the polynucleotides
291
were sufficiently free from protein. Human serum albumin (fatty acid free) fraction V and bovine pancreatic ribonuclease A (type XII-A) were from Sigma Chemical Company. Bovine pancreatic deoxyribonuclease I (type XII-A) was purchased from MP Biochemical Inc. Preparation of Protein-DNA and Protein-RNA Complexes Sodium-DNA or sodium-tRNA was dissolved to 2% w/w (0.05 M DNA (phosphate)) in Tris-HCl buffer (pH 7.30 0.1) at 5 C for 24 h with occasional stirring to ensure the formation of a homogeneous solution. The final concentration of the stock polynucleotide solution was determined spectrophotometrically at 260 nm by using molar extinction coefficient of 6600 cm1 M1 for DNA and 9250 cm1 M1 for tRNA (expressed as molarity of phosphate group) [40,41]. The UV absorbance at 260 nm of a diluted solution (1/250) of calf-thymus DNA used in our experiments was measured to be 0.661 (path length was 1 cm) and the final concentration of the stock DNA solution was calculated to be 25 mM in DNA phosphate. The average length of the DNA molecules, estimated by gel electrophoresis was 9000 base pairs (molecular weight ~ 6 106 Da). The appropriate amount of HSA was prepared in phosphate buffer (pH 7.3 0.1). The protein solution then was added dropwise to DNA solution to attain desired protein concentration of 0.125 0.250, 0.500 and 1.000 mM with a final DNA or RNA concentration of 12.5 mM polynucleotide (phosphate) for infrared measurements. For capillary electrophoresis, mixtures contained various concentrations of HSA and constant concentration of DNA (2.5 mM phosphate) with proteinpolynucleotide molar ratios of 1/250 to 1/55. The pH solution was adjusted to 6.807.30, using NaOH solution. The infrared spectra were recorded 1 h after incubation of protein with DNA or RNA solution. FTIR Spectra For IR measurements, hydrated films containing various protein concentrations (34 to 300 M) and constant nucleotide concentration of 12.5 mM were used. Infrared spectra were recorded on a Bomem DA3-0.02 FTIR spectrometer equipped with a nitrogen cooled HgCdTe detector and KBr beam splitter or on Nicolet FTIR spectrometer (Impact 420 model), equipped with DTGS (deuterated triglycine sulfate) detector and KBr beam splitter, using AgBr windows. The solution spectra are taken using AgBr windows with resolution of 2 to 4 cm1 and 100500 scans. The water subtraction was carried out with 0.1 M NaCl solution used as a reference at pH 6.57.5 [42]. A good water subtraction is achieved as shown by a flat baseline around 2200 cm1, where the water combination mode is located. This method is a rough estimate, but removes the water content in a satisfactory way [42]. The difference spectra [(polynucleotide solution + protein solution) (polynucleotide solution)] are produced, using a sharp DNA band at 968 cm1 and RNA band at 864 cm1 as internal references. These bands, which are due to the sugar C-C and C-O stretching vibrations, exhibit no spectral changes (shifting or intensity variations) on protein-polynucleotide complexation and were cancelled upon spectral subtraction. The spectra are smoothed with a Savitzky-Golay procedure (25). The intensity ratios of several DNA in-plane vibrations related to A-T and G-C and A-U base pairs and the PO2 stretching are measured with respect to the reference band at 968 cm1 or 864 cm1 as a function of protein concentration with an error of 3%.
292
These intensity ratio measurements are used to quantify the amounts of protein binding to the backbone PO2 group and DNA bases [43]. Analysis of Protein Secondary Structure Analysis of the secondary structure of DNase I and its tRNA complexes was carried out on the basis of the procedure reported [44]. The protein secondary structure is determined from the shape of the amide I band, located at 16501660 cm 1. Fourier selfdeconvolution and second derivative resolution enhancement were applied to increase the spectral resolution in the region of 17001600 cm1. The second derivatives were produced using a point convolution 11 or 13. The resolution enhancement resulting from self-deconvolution and the second derivative is such that the number and the position of the bands to be fitted are determined. In order to quantify the area of the different components of amide I contour, revealed by self-deconvolution and second derivative, a least-square iterative curve fitting was used to fit the Gaussian line shapes to the spectra between 17001600 cm1. CD Spectroscopy CD spectra were recorded at pH 7.4 with a Jasco J-720 spectropolarimeter. For measurements in the Far-UV region (200320 nm), a quartz cell with a path length of 0.01 cm was used. Three scans were accumulated at a scan speed of 50 nm per minute, with data being collected at every nm from 200 to 320 nm. Sample temperature was maintained at 25 C using a Neslab RTE-111 circulating water bath connected to the water-jacketed quartz cuvettes. Spectra were corrected for buffer signal and conversion to the Mol CD () (for proteins) was performed with the Jasco Standard Analysis software. The protein concentrations used in our experiment were 0.35 to 35 M, while tRNA and DNA concentration was fixed at 2.5 mM. The protein secondary structure was calculated using CONTINLL software [45] which predicts the different assignments of secondary structures by comparison with different ranges of proteins from high quality X-ray diffraction data [46]. The program CONTINLL is provided in CDPro software package which is available at the website: http://lamar.colostate.edu/ ~sreeram/CDPro. Absorption Spectroscopy The absorption spectra were recorded on a Perkin Elmer Lambda 40 Spectrophotometer. Quartz cuvettes of 1 cm were used. The absorbance measurements were performed at pH 7.4 by keeping the concentration of polynucleotide (40 M) constant, while using different protein contents (1 M to 20 M). The values of the binding constants K were obtained according to the methods reported [47,48]. By assuming that there is only one type of interaction between protein and polynucleotides in aqueous solution, the Equations 1 and 2 can be established: Polynucleotide + protein (polynucleotide: protein) complex K = [polynucleotide: protein] / [polynucleotide] [protein] Polynucleotide = DNA or RNA and protein = HSA, RNase A or DNase I (1) (2)
293
The values of the binding constants K were obtained from the DNA or RNA absorption at 260 nm according to the method described by published methods [47,48] where the bindings of various ligands to biomacromolecule were described. For weak binding affinities the data were treated using linear reciprocal plots based on
1 1 1 1 = + A - A 0 A - A 0 K(A - A 0 ) Cligand
(3)
where, A0 is the absorbance of DNA or RNA at 260 nm in the absence of protein, A is the final absorbance of the protein-polynucleotide complex and A is the recorded absorbance of complexes at different protein concentrations. Thus, the double reciprocal plot of 1/(AA0) vs. 1/Cprotein is linear and the binding constant (K) can be estimated from the ratio of the intercept to the slope [47,48].
Results and Discussion FTIR Spectra of HSA-DNA and HSA-RNA Adducts In order to characterize the protein-DNA bindings the infrared spectra of HSA-DNA complexes were recorded using constant amount of DNA with various concentrations of protein and the results are presented in Figs 2A and 2B. The DNA in-plane vibrations at 17501500 cm1 related to the G-C and A-T base pairs and the back bone phosphate group at 12501000 cm1 [42,43,4952] were perturbed upon protein interaction. The mainly guanine carbonyl vibration at 1710 cm1 of the free DNA gained intensity and shifted towards a higher frequency at 1713 cm 1 upon HSA complexation (Fig. 2A). Similarly, an increase in the intensity of the backbone PO 2 asymmetric stretching band at 1225 cm1 was observed, which shifted towards a lower frequency at 1210 cm1, in the spectra of HSA-DNA complexes (Fig. 2A). The positive features at 1707 and 12171218 cm1 in the difference spectra of HSA-DNA complexes are coming from an increase in the intensity of the guanine band at 1710 and the PO 2 band at 1225 cm1, respectively (Fig. 2B). In addition to a major spectral shifting of the PO 2 stretching at 1225 cm1, the relative intensities of the asymmetric (as) and symmetric (s) stretching vibrations of the backbone phosphate group were altered upon HSA interaction [42]. The s PO2 (1088 cm1) and as PO2 (1225 cm1) have changed, with the ratio s/as going from 1.75 (free DNA) to 1.55 (protein-DNA complexes). The observed spectral changes are due to the participation of the G-C bases (mainly guanine) and the backbone PO2 group in the HSA-DNA interactions. Similar alterations of the phosphate group vibrational frequencies were observed in the infrared spectra of the calf-thymus DNA-polypeptide complexes, where the protein-nucleic acid interaction was mainly through the backbone phosphate group and the positively charged amino acids on the surface of protein [53,54]. Therefore, it can be assumed that the protein bindings are mainly through G-C bases and the backbone phosphate groups in the HSA-DNA adducts. No major protein-RNA interaction occurs with HSA at low concentration (40 M). Evidence for this comes from lack of spectral changes for several RNA in-plane vibrations at 1698 (mainly guanine), 1654 (mainly uracil), 1608 (adenine), 1488 (mainly
294
Figure 2. FTIR spectra (A) and difference spectra [(DNA solution + protein solution) (DNA solution)] (B) in the region of 18001500 cm1 for the free DNA and human serum albumin (HSA) and their complexes in aqueous solution at physiological pH with various protein concentrations.
cytosine) and 1244 cm1 (asymmetric PO2 stretch) [42,4952] upon HSA interaction. As protein concentration increased to 80 M, minor intensity increases were observed for the PO2 bands at 1244 and 1087 cm1, with positive features at 1244 and 1088 cm1 in the difference spectra of protein-RNA complexes (Fig. 3A and 3B). The observed spectral changes are indicative of some degree of protein-PO2 interaction. As
295
Figure 3. FTIR spectra (A) and difference spectra [(tRNA solution + protein solution) (tRNA solution)] (B) in the region of 18001500 cm1 for the free tRNA and human serum albumin (HSA) and their complexes in aqueous solution at physiological pH with various protein concentrations.
HSA content increased to 300 M, the guanine band at 1698 cm1 shifted to a lower frequency at 1695 cm1 in the spectra of protein-RNA adducts (Fig. 3A). The shifting of the guanine band at 1698 cm1 was also accompanied by major intensity variations of this band upon HSA interaction. The changes observed are due to protein interaction
296
with guanine bases in HSA-RNA complexes. Further evidence regarding HSA-PO2 interaction is also coming from the intensity ratio alterations of symmetric and asymmetric PO2 bands at 1086/1240 cm1 [42]. The ratio of s/as was changed from 1.55 (free RNA) to 1.40 (complexed RNA) upon HSA interaction (Fig. 3A, 300 M). It has been suggested that many proteins bind DNA or RNA through positively charged amino acids on their surfaces [54]. Such positive charge can bind the negatively charged backbone PO2 group through electrostatic interactions. Additional evidence regarding HSA-DNA and HSA-RNA interactions comes from a major shifting of the protein amide I at 1656 and amide II band at 1541 cm 1 [55]. The amide I band at 1656 cm1 shifted towards lower frequencies at 1651 (HSA-DNA), whereas the amide II band at 1541 cm1 shifted towards higher frequencies at 1549 (HSA-DNA) upon DNA-protein interaction (Figs 2A and 3A). Similarly, the protein amide A band at 3300 cm1 (NH stretching) [37] shifted towards a lower frequency at 3290 cm1 upon HSA-DNA and HSA-RNA adduct formation (spectra not shown). The observed spectral changes are indicative of protein-polynucleotide interaction via polypeptide C=O, C-N and NH groups (H-bonding). HSA Conformation DNA and RNA interactions with HSA induced no major alterations of the protein secondary structure. Conformational analysis of the free HSA in H2O solution shows -helix 55% (1656 cm1), -sheet 22% (16161635 cm1), -anti 12% (1684 cm1) and 11% turn 11% (1673 cm1) (Fig. 4 A and Table 1) [56]. The random structure was estimated 12% from D2O solution (16401645 cm1). The second derivative and curvefitting procedures [37,38] were applied to the protein amide I band (at 1651 cm 1 of the difference spectra of HSA-DNA and RNA adducts). The results showed no major alterations of the HSA secondary structure, on DNA and RNA complexation (Fig. 4B and 3C and Table 1). This is indicative of some degree of stabilization of the protein secondary structure upon polynucleotide interaction. FTIR Spectra of RNase-DNA and RNase-RNA Adducts The IR spectral features of RNase-DNA interaction are presented in Figure 5. Evidence for protein-DNA binding comes from spectral changes observed for both free DNA and free RNase upon complexation. The band at 1710 cm1 of the free DNA spectrum assigned mainly to guanine bases [42,43,4952] gained intensity in the spectra of RNaseDNA complexes (Fig. 5). The increase in intensity is clearly shown in difference spectra of RNase-DNA complexes (Fig. 5, diff., 34 M and 270 M). The presence of positive features at 1709 (DNA), 1660 (protein) and 1646 cm1 (protein), in the difference spectra are due to increase in intensities of DNA and RNase vibrations. The positive peak at 1709 cm1 is due to increase in intensity of DNA band at 1710 cm1 (guanine), while the positive bands at 1660 and 1646 cm1 are due to RNase amide I band at 1659 and 1644 cm1 (Fig. 5, diff., 34 M and 270 M). The spectral changes observed are due to the participation of DNA bases (G) in the RNase complexation. The protein-PO2 interaction is also evident from increase in intensity and shifting of the PO 2 asymmetric band at 1225 cm1, which appeared at 1228 cm1 in the spectra of the RNase-DNA complexes (Fig. 5, 270 M). The positive features at 1212 and 1218 cm1 (Fig. 5, diff. 34 M and 270 M) are due to increase in intensity of the asymmetric PO 2 band at
297
Figure 4. Curve-fitted amide I region (17001612 cm1) and secondary structure determination of the free human serum albumin (A) and its HSA-DNA (B) and HSA-tRNA (C) in aqueous solution with 300 M protein and 12.5 mM polynucleotide concentrations.
1225 cm1 upon protein interaction (Fig. 5, diff. 34 M and 270 M). Further evidence regarding RNase-PO2 interaction is also coming from the intensity ratio variations of symmetric and asymmetric PO2 bands at 1086/1225 [42]. The ratio of s/as was changed from 1.50 (free DNA) to 1.80 (complexed DNA) upon RNase interaction (Fig. 5, 270 M). Evidence for RNase-tRNA interaction comes from infrared spectral analysis of both tRNA and RNase A presented in Fig. 6. The band at 1698 cm1 of tRNA spectrum
298
Table 1. Secondary structure analysis of the free HSA, RNase A, and DNase I and their DNA and RNA complexes in H2O at physiological pH at 25 oC Amide I components (cm1) 16921680 -anti 16801660 turn 16601649 -helix 16481641 random 16401615 -sheet Free HSAHSA DNA H2O (%) 12.0 1 12.0 HSARNA Free RNase H2O % 9.0 Free DNase H2O (%) 2.0
RNaseDNA
RNaseRNA
DNaseDNA
DNaseRNA
7.0
12.0
4.0
7.0
8.0
11.0 1
13.0
15.0
16.0
12.0
31.0
12.0
110
15.0
55.0 3
56.0
58.0
28.0
16.0
18.0
40.0
27.0
27.0
11 1.0
9.0
17.0
17.0
12
20.0
15.0
22.0 2
21.0
20.0
38.0
43.0
30.0
34.0
35.0
34.0
assigned mainly to guanine bases [42,43,4952] gained intensity and shifted towards a lower frequency at 1689 cm1 upon RNase interaction (Fig. 6). The increase in intensity and shifting are clearly observed in difference spectra of RNase-tRNA complexes with protein concentrations of 34 M and 270 M (Fig. 6, diff.). The presence of positive features at 1691 cm1 (RNA), 1661 cm1 (RNA), and 1643 and 1545 cm1 (RNase) in the difference spectra are coming from both tRNA and RNase due to increase in intensities of these vibrations in the complexes. The positive peak at 1691 cm 1 is due to increase in intensity of tRNA band at 1698 cm1 (guanine), while the positive band at 1661 cm1 can arise from uracil band at 1657 cm1 and RNase amide I band at 1657 cm1 (Fig. 6, diff., 34 M). Because amide I band at 1657 cm1 did not exhibit major shifting upon RNA complexation (the feature at 1661 should result from the uracil vibration at 1657 cm1) (Fig. 6, diff., 34 M)), the spectral changes observed are due to the participation of RNA bases (G and U) in RNase complexation. Major protein-PO2 interaction is also evident from the remarkable reduction in intensity of the PO2 asymmetric band at 1240 cm1 and increase in intensity of symmetric stretching at 1086 cm1 with a characteristic positive feature at 1072 cm 1 (Fig. 6, diff., 34 M). Further evidence regarding RNase-PO2 interaction also comes from the intensity ratio variations of symmetric and asymmetric PO2 bands at 1086/1240 (Alex and Dupuis, 1989). The ratio of s/as was changed from 1.30 (free RNA) to 1.90 (complexed RNA) upon RNase interaction (Fig. 6, 270 M). At high protein concentration (270 M), the guanine band showed increase in intensity and appeared at 1691 cm1 in the difference spectra (Fig. 6, diff., 270 M), while protein amide I and amide II were shifted at 1660, 1642 and 1545 cm1. These spectral changes are the result of continued interaction of protein with RNA at high RNase content. The presence of positive features at 1232 and 1074 cm1 in the difference spectra come from the spectral changes of asymmetric and symmetric PO 2 vibrations due to a major protein-phosphate interaction (Fig. 6, diff., 270 M).
299
Figure 5. FTIR spectra and difference spectra [(DNA solution + RNase A solution) (DNA solution)] in the region of 1800600 cm1for the free calf-thymus DNA (12.5 mM) and free RNase (270 M) and their complexes in aqueous solution at pH 7.3 with various protein concentrations (34 and 270 M) and constant DNA concentration (12.5 mM).
RNase A Conformation The results of conformational analysis (IR spectroscopy) of the free RNase A and its DNA and RNA adduct is shown in Fig. 7 and Table 1). The free RNase A in H2O solution shows -helix 28% (1658 cm1), -sheet 38% (16181638 cm1), -anti 9% (1692 cm1), turn 16% (1679 cm1) and random coil 9% (1648 cm1) (Fig. 7A) consistent with crystal structure of RNase A, which contains mainly -sheet structure [58]. No major alterations of protein conformation were observed upon DNA interaction at low RNase concentration (34 M). However, at high protein content (270 M), major
300
Figure 6. FTIR spectra and difference spectra [(tRNA solution + RNase A solution) (tRNA solution)] in the region of 1800600 cm1 for the free tRNA (12.5 mM) and free RNase (270 M) and their complexes in aqueous solution at physiological pH (7.4) with various protein concentrations (34 to 270 M).
reduction of -helix from 28 to 16% and increase of -sheet from 38 to 43% and random coil from 9 to 17% were observed (Fig. 7B and Table 1). The results of conformational analysis RNase-tRNA complexes are shown in Fig. 7C. No major alterations of protein conformation were observed upon RNA interaction at low RNase concentration (3.5 M). However, at high protein content (270 M), a major reduction of -helix from 28 to 18% and of -sheet from 38 to 30% was observed (Fig. 7C and Table 1). The reduction in -helix (from 28% to 18%) and -sheet (from 38% to 30%) structure was associated with increase in turn structure from 16% to 31% and in random coil structure from 9% to 17% in the protein-tRNA
301
Figure 7. Curve-fitted amide I region (17001600 cm1) obtained from IR spectra with secondary structure determination of the free RNase A (A) and its DNA (B) and tRNA (C) complexes in aqueous solution with 270 M protein concentration and final polynucleotide content 12.5 mM at pH 7.4.
complexes (Fig. 7C, Table 1). The reduction of -helix and -sheet structure in favor of random coil can be attributed to a partial unfolding of protein upon tRNA complexation. The reduction of -helix was also observed during heat denaturation of RNase A shown by infrared spectroscopy [57]. FTIR Spectra of DNase-DNA and DNase-RNA Adducts The IR spectral features of DNase-DNA interaction are presented in Fig. 8. Evidence for protein-DNA binding comes from spectral changes observed when comparing both free DNA and free DNase spectra and their complexes. To begin, the band located at 1710 cm1 of the free DNA spectrum assigned mainly to guanine bases [42,43,4952] exhibited no major intensity variations in the spectra of DNase-DNA complexes (Fig. 8). But, the thymine band at 1661 cm1 showed major intensity increase with a
302
Figure 8. FTIR spectra and difference spectra [(DNA solution + DNase I solution) (DNA solution)] in the region of 1800600 cm1 for the free calf-thymus DNA (12.5 mM) and free DNase (250 M) and their complexes in aqueous solution at pH 7.4 with various protein concentrations (10, 62 and 250 M) and constant DNA concentration (12.5 mM).
positive peak at 1660 in the difference spectra, upon DNase interaction (Fig. 8, diffs, 10 and 62 M). Similarly the adenine band at 1610 cm1 of the free DNA spectrum gained intensity with a positive feature at 1591 cm1, in the difference spectra of pro-
303
tein-DNA complexes (Fig. 8, diff., 10 M). Since thymine band at 1660 cm 1 and protein amide I band at 1656 cm1 are overlapped at high protein concentration, it is difficult to attribute the presence of the strong positive feature found at 1658 cm1 to thymine or protein vibrations separately (Fig. 8, diff., 250 M). The overall spectral changes observed for the thymine and adenine bands are due to major interaction of protein with A-T bases in the minor groove of DNA duplex. However, the other protein binding site is located at the backbone phosphate group. The major shifting of the PO 2 asymmetric vibration at 1225 cm1 to 1234 cm1 and the symmetric vibration at 1088 to 1067 cm1 is due to protein-PO2 interaction (Fig. 8). Additional DNase-DNA interaction is evidenced by shifting of the protein amide I band at 1642 cm1 to 1645 cm1, upon protein-DNA complexation. Similarly, the protein amide II band at 1536 cm1 appeared with a new component band at 1545 cm1 in the spectrum of protein-DNA complexes (Fig. 8, 250 M). The observed spectral changes for amide I and amide II bands are due to protein-DNA interaction via polypeptide C=O, C-N and N-H groups (H-bonding). Results presented in Fig. 9 and the analysis of infrared spectra of both tRNA and DNase I showed major DNase I-tRNA interaction. The band at 1698 cm1 of tRNA spectrum assigned mainly to guanine bases [42,43,4951] gained intensity and shifted towards a lower frequency at 1690 cm1 upon DNase I interaction (Fig. 9). The increase in intensity and shifting is clearly evident in difference spectra of DNase I-tRNA complexes with protein concentrations of 10 M (Fig. 9, diff. 10 M). Indeed, positive features at 1681 and 1648 cm1 in the difference spectra which come from tRNA are the result of increase in intensities of guanine band at 1698 and uracil band at 1660 cm 1 (Fig. 9, diff., 10 M). However, the positive peaks at 1648 cm1 and 1536 cm1 in the difference spectra with 10 M and 62 M) protein concentration come from protein amide I and amide II vibrations (Fig. 9, diff., 10 M and 62 M). The spectral changes observed for the guanine band at 1690 and uracil band at 1660 cm1 arise from the participation of RNA bases (G and U) in protein complexation. Major protein-PO2 interaction is also evident from the increase in intensity and the shifting of the backbone PO 2 asymmetric band at 1240 cm1 and symmetric stretching at 1085 cm1 (Fig. 9, 250 M). Further evidence regarding protien-PO2 interaction is also coming from the intensity ratio variations of symmetric and asymmetric PO2 bands at 1085/1240 (Alex and Dupuis, 1989). The ratio of s/as was changed from 1.65 (free tRNA) to 2 (complexed RNA) upon DNase interaction (Fig. 9, 250 M). At high protein concentration (250 M), due to overlapping of tRNA in-plane vibrations with protein amide I vibrations in the region of 16591650 cm1, it is difficult to draw a firm conclusion about the nature of the protein-tRNA binding (Fig. 9, 250 M). However, additional evidence regarding DNase I-tRNA interaction comes from shifting of the protein amide I bands at 1656 cm1, 1642 cm1 and amide II band at 1536 cm1. The amide I band at 1656 cm1 shifted towards a higher frequency at 1657 and the band at 1645 observed at 1638 cm1, while amide II band at 1536 shifted towards a higher frequency at 1542 cm1 upon protein-tRNA complexation (Fig. 9, 250 M). In the difference spectra of DNase I-tRNA, the bands at 1655 and 1638 cm 1 are coming from protein amide I bands, while the band at 1542 cm1 is related to amide II bands (Fig. 9, diff., 250 M). The observed spectral changes for amide I and amide II bands are indicative of protein-tRNA interaction via polypeptide C=O, C-N and NH groups (H-bonding).
304
Figure 9. FTIR spectra and difference spectra [(tRNA solution + DNase I solution) (tRNA solution)] in the region of 1800600 cm1 for the free RNA (12.5 mM) and free DNase (250 M) and their complexes in aqueous solution at pH 7.4 with various protein concentrations (10, 62 and 250 M) and constant tRNA concentration (12.5 mM).
305
Figure 10. Curve-fitted amide I region (1700-1600 cm1) with secondary structure determination of the free DNase I (A) and its DNA (B) and tRNA (C) complexes in aqueous solution with 250 M protein concentration and final polynucleotide content 12.5 mM at pH 7.4.
DNase I Conformation The results of conformational analysis (IR spectroscopy) of the free DNase I and its DNA complex are shown in Fig. 10A. The free DNase I in buffer solution shows extensive -helix 40% (1659 cm1), central -sheet 34% (16121630 cm1), -anti 2% (1692 cm1), turn 12% (1680 cm1) and random coil 12% (1642 cm1) (Fig. 10A and Table 1) that are consistent with structural analysis of DNase I [59]. At high protein content (250 M), major reduction of -helix from 40 to 27% and increase of random coil from 12 to 20% and -anti from 2% to 7% were observed (Fig. 10B and Table 1). The reduction of -helix and increase in -anti and random coil structures are attributed to a partial unfolding of protein upon DNA complexation. These results are consistent with the reduction of -helix observed during protein unfolding in protein polyamine complexes [56]. The conformational changes of DNase I-tRNA complexes are shown in Fig. 10C. The free DNase I in buffer solution shows extensive -helix 40% (1659 cm 1), central -sheet 34% (16121630 cm1), -anti 2% (1692 cm1), turn 12% (1680 cm1) and random coil 12% (1642 cm1) (Fig. 10A) that are consistent with structural analysis of DNase I [59]. At high protein content (250 M), major reduction of -helix from 40 to 27% and increase of random coil from 12 to 16%, -anti from 2% to 8% and turn from 12 to 15% were observed (Fig. 10C and Table 1). The central -sheet remained un-
306
changed. The reduction of -helix and increase in -anti and random coil structures are attributed to a partial unfolding of protein upon tRNA complexation. DNA and RNA Conformations Minor alterations of B-DNA structure were observed upon HSA complexation. Evidence for this comes from the spectral change for B-DNA marker bands at 1222 cm1 (PO2 stretch) and 1717 cm1 (mainly guanine) [49,50,60], upon protein complexation (Fig. 2A). Other B-DNA indicator at 836 cm1 (phosphodiester mode), was overlapped by the several protein bands centered at about 800 cm1 (Fig. 2A). In a B to A transition, the marker band at 837 cm1 shifts towards a lower frequency at about 810 cm1 and the guanine band at 1717 cm1 appears at 1700 cm1, while the phosphate band at 1222 cm1 shifts towards a higher frequency at 1240 cm1 [50,60]. In a B to Z conformational changes, the sugar-phosphate band at 836 cm1 appears at 800780 cm1, and the guanine band displaces to 1690 cm1, while the phosphate band shift to 1216 cm1 [50,60]. The shifting of the bands at 1717 (G) to 1713 and 1222 (PO2 stretch) to 1217 cm1 are due to some degree of DNA conformational changes upon protein interaction. The changes are due to minor perturbations of B-DNA structure towards Z-conformation. However, it is not a complete Z-formation, while the band at 1717 cm1 appeared at 1713 cm1 (in a complete Z structure this band appeared at 1690 cm1), while the shift of the PO2 band at 1222 to 1217 cm1 is consistent with Zconformation (Fig. 2A). However, no conformational changes occurred for tRNA upon HSA adduct formation. Evidence for this is given by the absence of major alterations of A-RNA marker bands at 16981695 cm1 (guanine), 12441243 cm1 (phosphate) and 810 cm1 (ribose-phosphate) (Fig. 3A) in both free tRNA and its protein complexes [61,62]. The DNase I-DNA interaction induced a partial DNA conformational transition. The CD spectrum of the free DNA is composed of four major peaks at 214 (negative), 225 (positive), 245 (negative) and 280 nm (positive) (not shown). This is consistent with the CD spectrum of double helical DNA in B conformation [6365]. Upon addition of DNase I, (25 and 62 M), major shifting of CD bands were observed (spectra not shown). The band at 214 nm shifted to 208210 nm and the one at 280 appeared at 271 nm upon protein complexation [65]. In a complete B to A transition, the CD marker bands at 214 nm lost intensity, the positive band at 225 nm shifted towards higher wavelength, while the positive band at 280 nm gained intensity and shifted to 267 nm [63]. The spectral changes reported for A-DNA were consistent with the double helical RNA in A conformation [64]. Since the major band at 280 nm shifted to 271 nm, the alterations of DNA structure is consistent with a partial B to A transition upon DNase interaction [66]. This is consistent with the infrared results on the DNaseDNA complexes that showed a partial B to A-DNA transition with shifting of the marker B-DNA bands at 1710 cm1 (G) to 1708 cm1, 1225 cm1 (PO2) to 1234 cm1, and 834 cm1 (phosphodiester modes) to 831 cm1 (Fig. 8). In a complete B to A transition, the B-DNA marker IR bands are observed at 1700 cm1 (G), 1240 cm1 (PO2) and 810 cm1 (phosphodiester) [49,50,60]. However, the major intensity increase of the bands at 214 and 280 nm in the CD spectra of protein-DNA complexes are attributed to the base destacking interaction upon DNase complexation. The CD spectrum of free tRNA is composed of three major peaks at 207 (negative), 220 and 269 nm (positive) (not shown). This is consistent with the CD spectrum of double helical RNA in A conformation [6365]. Upon addition of DNase I
307
Table 2. Binding constants for protein DNA and protein-RNA complexes DNA HAS RNase A DNase I Binding constant (M1) 4.5 105 and 6.1 104 6.1 104 5.7 105
RNA HAS RNase A DNase I 1.45 104 4.0 105 2.1 104
(25.0 M), no major shifting of the bands were observed, while an increase of the molar ellipticity of the band at 207 nm was observed [67]. However, as protein concentration increased (62.5 M), major increase in molar ellipticity of the band at 207 nm and minor increase in intensity of the band at 269 nm were observed [67]. Since there was no major shifting of the bands at 207 and 269 nm, tRNA remains in A-conformation. However, the major alterations of the intensity of the band at 207 nm can be attributed to the reduction of the base stacking interaction and tRNA aggregation upon protein complexation. This is consistent with the infrared data on the DNase I-tRNA adducts that showed tRNA in A-conformation with marker IR bands at 1698 (G), 1240 (PO2) and 865 and 815 cm1 (phosphodiester modes) [49,50] in both free tRNA and in DNase I-RNA complexes (Fig. 9). It is important to note that there has been no digestion of tRNA by DNase I because no major structural changes occurred for tRNA under our experimental conditions (Fig. 9, infrared spectra). RNase-DNA complexation did not alter DNA conformation. Evidence for this comes from no major alterations of the IR and CD marker bands of B-DNA upon RNase interaction. The CD spectrum of the free B-DNA is composed of four major peaks at 214 (negative), 225 (positive), 245 (negative) and 280 nm (positive) exhibit no major changes upon RNase A complexation. Similarly, the IR marker bands for BDNA at 1710, 1225, 834 cm1 showed no major alterations in the spectra of RNaseDNA complexes (Fig. 5). These are consistent with DNA remaining in B-family structure upon RNase interaction. Similarly, the CD spectrum of free tRNA, which is composed of three major peaks at 207 (negative), 220 and 269 nm (positive) and the IR marker bands at 1698, 1240, 865 and 810 cm1 (Fig. 6) showed no major spectral changes on RNase complexation. This is indicative of tRNA remaining in A-conformation upon protein interaction. Stability of Protein-DNA and Protein-RNA Adducts The protein-DNA and protein-tRNA binding constants estimated by UV-visible spectroscopy are presented in Table 2. The UV absorption spectra of protein-polynucleotide complexes are calculated using the double reciprocal plot of 1/(AA0) vs 1/(protein concentration), which is linear and the binding constant (K) can be estimated from the ratio of the intercept to the slope. A0 is the initial absorbance of the free DNA or RNA at 260 nm and A is the recorded absorbance of polynucleotides in the presence of different protein concentrations. The overall binding constants for protein-DNA and protein-RNA complexes are given in Table 2. HSA-DNA complexes showed two bindings
308
with K1 = 4.5 105 M1 and K2 = 6.1 104 M1, while one binding with K = 1.45 104 M1 was observed for HSA-tRNA adducts (Table 2). Similarly, one binding was observed for RNase-DNA with K = 6.1 104 M1, RNase-tRNA with K= 4.0 105 M1, DNase-DNA with K = 5.7 105 M1 and DNase-tRNA with K = 2.1 104 M1 (Table 2). The stability of the protein-nucleic acids complexes are consitent with the binding constants of other protein-DNA and protein-RNA complexes [6871]. It is worth mentioning that there has been no digestion of DNA by DNase I or tRNA by RNase A because no major structural changes occurred for DNA or RNA under these experimental conditions (Figs 5, 6, 8, 9).
Summary and Future Directions Structural analysis of protein-DNA and protein-RNA complexes are important for better understanding of molecular interactions that govern transcriptional regulation. Even though the driving forces moving protein into the major and minor grooves of DNA duplex are reviewed [72], each protein binding to DNA grooves induces specific structural alterations of DNA conformation. HSA binding occurred with major and minor groove of DNA and RNA duplexes as well as with the backbone PO2 group. RNase A interactions with G-C bases and the PO2 were predominant, while no DNA and RNA digestion occurred at our experimental condition. Similarly, DNase I bindings were mainly with the backbone phosphate group and the A-T bases in the minor groove of DNA and RNA duplexes with no polynucleotide hydrolysis. Protein-polynucleotide interaction altered protein secondary structure for RNase A and DNase I and induced a partial B to A-DNA transition, while RNA remained in A-conformation. Even though FTIR spectroscopy proved to be a useful tool in determining the protein binding sites in the major or minor groove of nucleic acid duplexes and the structural variations of protein and DNA and RNA complexes, further study is needed to apply molecular modeling based on energy minimization to present proper structural models for protein-DNA and protein-RNA interactions.
Acknowledgments This work is supported by grants from Natural Sciences and Engineering Research Council of Canada (NSERC) and FCAR (Qubec).
References
[1] S.C. Harrison, A structural taxonomy of DNA-binding domains, Nature 353 (1991), 715-719. [2] B.F. Luisi, DNA-protein interaction at high resolution. In DNA-protein structural interactions. Lilley, D.M.J. (ed.) New York, Oxford University Press, 1-48, 1995. [3] N.M. Luscombe, S.E. Austin, H.M. Berman, and J.M. Thornton, An overview of the structures of protein-DNA complexes, Genome Biol. 1 (2000), 1-37. [4] N.M. Luscombe, A. Laskowski, and J.M. Thornton, Amino acid-base interactions: a three-dimensional analysis of protein-DNA interactions at an atomic level. Nucl. Acids Res. 29 (2001), 2860-2874. [5] M. Yonezawa, N. Doi, Y. Kawahashi, T. Higashinakagawa, and H. Yanagawa, DNA display for in vitro selection of diverse peptide libraries, Nucl. Acids Res. 31 (2003), 1-5. [6] I.V. Smolina, V.V. Demidov, and M.D. Frank-Kamenetskii, Paulsing of DNA polymerases on duplex DNA templates due to ligand binding in vitro, J. Mol. Biol. 326 (2003), 1113-1125.
309
[7] T. Namanbhoy, A.J. Morales, A.T. Abraham, C.S. Vortler, R. Giege, and P. Schimmel, Simultaneous binding of two proteins to opposite sides of a single transfer RNA, Nature Struct. Biol. 8 (2001), 344-348. [8] D. Moras, Aminoacyl-tRNA synthetase, Curr. Opin. Struct. Biol. 2 (1998), 138-142. [9] G. Varani and K. Nagai, RNA recongnition by NPR protein during RNA processing, Annu. Rev. Biophys. Biomol. Struct. 27 (1998), 407-445. [10] S. Jones, D.T.A. Daley, N.M. Luscombe, H.M. Berman, and J.T. Thornton, Protein-RNA interactions: a structural analysis, Nucl. Acids Res. 29 (2001), 943-954. [11] G.C. Foulds and H. Etzkom, A capillary electrophoresis mobility shift assay for protein-DNA binding affinities free in solution, Nucleic Acids Res. 26 (1998), 4304- 4305. [12] C. Li and L.M. Martin, A robust method for determining DNA binding constants using capillary zone electrophoresis, Anal. Biochem. 263 (1998), 72-78. [13] J. Xian, M.G. Harrington, and E.H. Davidson, DNA-protein binding assays from a single sea urchin egg: A high-sensitive capillary electrophoresis method, Proc. Natl. Acad. Sci. USA, 93 (1996), 86-90. [14] T. Guszcynski, and T.D. Copeland, A binding shift assay for the zinc-bound and zinc-free HIV-1 nucleocapsid protein by capillary electrophoresis, Anal. Biochem. 260 (1998), 212-217. [15] T. Peters, All about albumin. Biochemistry, genetics and medical application; Academic Press, San Diego, 1996. [16] D.C. Carter and J.X. Ho, Structure of serum albumin, Adv. Protein Chem. 45 (1994), 153-203. [17] S. Sugio, A. Kashima, S. Mochizuki, M. Noda, and K. Kobayashi, Crystal structure of human serum albumin at 2.5 resolution Protein Eng. 12 (1999), 439-446. [18] H.M. He, and D.C. Carter, Atomic structure and chemistry of human serum albumin, Nature 358 (1992), 209-215. [19] T. Peters, Serum albumin, Adv. Protein Chem. 37 (1985), 161-245. [20] S. Curry, P. Brick, and N.P. Frank, Fatty acid binding to human serum albumin: New insights from crystallographic studies, Biochim. Biophys. Acta 1441 (1999), 131-140. [21] I. Petitpas, T. Grune, A.A. Battacharya, and S. Curry, Crystal structure of human serum albumin complexed with monounsaturated and polyunsaturated fatty acids, J. Mol. Biol. 314 (2001), 955-960. [22] L. Painter, M.M. Harding, and P.J. Beeby, Synthesis and interaction with human serum albumin of the first 3,18-disubstituted derivative of bilirubin, J. Chem. Soc., Perkin Trans 18 (1998), 3041-3044. [23] J.F. Riordan, Ribonucleases: Structure and Functions. eds. G. DAlessio and J.F. Riordan, Academic Press, New York, pp. 445-489, 1997. [24] R.T. Raines. Ribonuclease A, Chem. Rev. 98 (1999), 1045-1065. [25] J.L. Neira, P. Sevilla, M.A. Margarita, B. Marta, and R. Manuel, Hydrogen exchange in ribonuclease A and ribonuclease S: Evidence for residual structure in the unfolded state under native condition, J. Mol. Biol. 285 (1999), 627-643. [26] J. Matousek, G. Gotte, P. Pouckova, J. Soucek, T. Slavik, F. Vottariello, and M. Libonatti, Antitumor activity and other biological actions of oligomers of ribonuclease A, J. Biol. Chem. 278 (2003), 23817-23822. [27] R. Berisio, F. Sica, V.S. Lamzin, K.S. Wilson, A. Zagari, and L. Mazzarella, Atomic resolution structures of ribonuclease A at six pH values, Acta Crystallogr. D 58 (2002), 441-450. [28] P. Fu, J. Chen, Y. Tian, T. Watkins, X. Cui and B. Zhao, Anti-tumor effect of hematopoietic cells carring the gene of ribonucelase inhibitor, Cancer Gene Ther. 12 (2005), 268-275. [29] M.C. Haigis, E.L. Kurten, R.L. Abel, and R.T. Raines, KFERQ sequence in ribonuclease A-mediated cytotoxicity, J. Biol. Chem. 277 (2002), 11576-11582. [30] K.A. Dickson, C.L. Dahlberg, and T.R. Raines, Compensating effects of the cytotoxicity of ribonuclease A variants, Arch. Biochem. Biophys. 415 (2003), 172-177. [31] P.A. Leland, L.W. Schultz, B.M. Kim, and R.T. Raines, Ribonuclease A variants with potent cytotoxic activity, Proc. Natl. Acad. Sci. USA 95 (1998), 10407-10412. [32] T. Soucek, R.T. Raines, M. Haugg, S.A. Raillard-Yonn, and S.A. Benner, Structural changes of ribonuclease A and their effects on biological activity, Comp. Biochem. Physiol. Part C 123 (1999), 103-111. [33] C.G. Dos Remedios, D. Chhabra, M. Kekic, I.V. Dedova, M. Tsubakihara, D.A. Berry, and N.J. Nosworthy, Actin binding proteins:Regulation of cytoskeletal microfilaments, Physiol. Rev. 83 (2003), 433-473. [34] V.W. Campbell and D.A. Jackson, The effect of divalent cations on the mode of action of DNase I, J. Biol. Chem. 255 (1980), 37263735. [35] P.A. Price, Characterization of Ca2+ and Mg2+ binding to bovine pancreatic deoxyribonuclease, J. Biol. Chem. 247 (1972), 2895-2899. [36] S. Cal, K.L. Tan, A. McGregor, and B.A. Connolly, Conversion of bovin pancreatic DNase I to repair endonuclease with a high selectivity for abasic sites. EMBO J. 17 (1998), 7128- 7138.
310
[37] C.Q. Pan and R.A. Lazarus, Hyperactivity of human DNase I variants, J. Biol. Chem. 273 (1998), 11701-11708. [38] A. Lahm and D. Suck, DNase I-induced DNA conformation: 2 structure of a DNase I-octamer complex, J. Mol. Biol. 221 (1991), 645-667. [39] S.A. Weston, A. Lahm, and D. Suck, X-ray structure of the DNase I-d(GGTATACC)2 complex at 2.3 resolution, J. Mol. Biol. 226 (1992), 1237- 1256. [40] M.E. Reichmann, S.A. Rice, C.A. Thomas, and P. Doty, A further examination of the molecular weight and size of deoxypentose nucleic acid, J. Am. Chem. Soc.76 (1954), 3047-3053. [41] R. Vijayalakshmi, M. Kanthimathi, and V. Subramanian, DNA cleavage by a chromium(III) complex, Res. Commun. 271 (2000), 731-734. [42] S. Alex, and P. Dupuis, FTIR and Raman investigation of cadmium binding by DNA, Inorg. Chim. Acta 157(1986), 271-281. [43] A. Ahmed Ouameur and H.A. Tajmir-Riahi, Structural analysis of DNA interactions with biogenic polyamines and cobalt(III)hexamine studied by Fourier transform infrared and capillary electrophoresis, J. Biol. Chem. 279 (2004), 42041-42053. [44] D.M. Byler and H. Susi, Examination of the secondary structure of protein by deconvoluted FTIR spectra, Biopolymers 25 (1986), 469-487. [45] W.C. Johnson, Analyzing protein circular dichroism spectra for accurate secondary structure, Protein. Struct. Funct. Genet. 35 (1999), 307-312. [46] N. Sreerama and R.W. Woddy, Estimation of protein secondary structure from circular dichroism spectra: Comparison of CONTIN, SELCON and CDSSTR methods with an expanded reference set, Anal. Biochem. 287 (2000), 252-260. [47] J.J. Stephanos, Drug-protein interactions. Two-site binding of heterocylclic ligands to a monomeric haemoglobin, J. Inorg. Biochem. 62 (1996), 155-169. [48] J.J. Stephanos, S.A. Farina, and A.W. Addison, Iron ligand recognition by monomeric hemoglobins, Biochim. Biophys. Acta 1295 (1996), 209-221. [49] M. Loprete, and K.A. Hartman, Conditions for the stability of the B, C, and Z structural forms of poly(dG-dC) in the presence of lithium, potassium, magnesium, calcium and zinc cations, Biochemistry 32 (1993), 4077-4082. [50] E. Taillandier, and J. Liquier, Infrared spectroscopy of DNA, Methods Enzymol. 211 (1992), 307-335. [51] E.B. Starikov, M.A. Semenov, V.Y. Maleeve, and A.I. Gasan, Evidential study of correlated events in biochemistry: physicochemical mechanisms of nucleic acid hydration as revealed by factor analysis, Biopolymers 31 (1991), 255-273. [52] R. Ahmad, H. Arakawa, and H.A. Tajmir-Riahi, A comparative study of DNA binding to Mg(II) and Ca(II) in aqueous solution: Major and minor grooves bindings, Biophys. J. 84 (2003) 2460-2466. [53] S.B. Dev and L. Walters, Fourier transform infrared spectroscopy of the characterization of a model peptide-DNA interaction, Biopolymers 29 (1990), 289-299. [54] A. Podesta, M. Indrieri, D. Brogioli, G.S. Manning, P. Milani, R. Guerra, L. Finzi, and D. Dunlap, Positively charged surfaces increase the flexibility of DNA, Biophys. J. 89 (2005), 2558-2563. [55] S. Krimm, and J. Bandekar, Vibrational spectroscopy and conformation of peptides, polypeptides, and proteins, Adv. Protein Chem. 38 (1986), 181-364. [56] A. Ahmed Ouameur, E. Mangier, R. Rouillon, R. Carpentier, and H.A. Tajmir Riahi, Effects of organic and inorganic polyamine cations on the structure of human serum albumin, Biopolymers 73 (2004), 503-509. [57] S. Seshardi, K.A. Oberg, and A.L. Fink, Thermally denatured ribonuclease A retains secondary structure as shown by FTIR, Biochemistry 33 (1994), 1351-1355. [58] J.C. Fontecilla-Camps, R. de Llorens, M.H. le Du, and C.M. Cuchillo, Crystal structure of ribonuclease A.d(ApTpApApG) complex, J. Biol. Chem. 269 (1994), 21526-21531. [59] W. Kabsch, H.G. Mannherz, D. Suck, E.F. Pai, and K.C. Holmes, Atomic structure of the actin: DNase I complex, Nature 347 (1990), 37-44. [60] H.A. Tajmir-Riahi, J.F. Neault, and M. Naoui, Does DNA acid fixation produce left-handed Z structure? FEBS Lett. 370 (1995), 105-108. [61] H. Malonga, H. Arakawa, J.F. Neault, and H.A. Tajmir-Riahi, DNA interaction with human serum albumin studied by affinity capillary electrophoresis and FTIR spectroscopy, DNA&Cell Biology 25 (2006), 63-68. [62] H. Malonga, H. Arakawa, J.F. Neault, and H.A. Tajmir-Riahi, Transfer RNA binding to human serum albumin: A model for protein-RNA interaction, DNA&Cell Biology 25 (2006), 393-398 [63] M. Vorlickova, Conformational transitions of alternating purin-pyrimidin DNAs in perchlorate ethanol solutions, Biophys. J. 69 (1995), 2033-2043. [64] J. Kypr and M. Vorlickova, Circular dichroism spectroscopy reveals invariant conformation of guanine runs in DNA, Biopolymers 67 (2002), 275-277.
311
[65] B.I. Kankia, V. Bukin, and V.A. Bloomfiled, Hexamine cobalt(III)-induced condensation of calfthymus DNA: circular dichroism and hydration measurements, Nucl. Acids Res. 29 (2001), 2795-2801. [66] C.N. Nsoukpo-Kossi, C. Ragi, and H.A. Tajmir-Riahi, DNA interaction with RNase A alters protein conformation, DNA&Cell Biology 26 (2007), 28-35. [67] C.N. Nsoukpo-Kossi, C. Ragi, and H.A. Tajmir-Riahi, RNase A-tRNA binding alters protein conformation, Biochem. Cell Biol. 85 (2007), 311-318. [68] S.J. Koch, A. Shundrovsky, B.J.C. Jantzen, and M.D. Wang, Probing protein-DNA interactions by unzipping of a single DNA double helix, Biophys. J. 83 (2002), 1098-1105. [69] G. Wieland, P. Hemerich, M. Koch, T. Stoyan, J. Hegemann, and S. Diekmann, Determination of the binding constants of the centromere protein Cbf1 to all 16 centeromer DNAs of Saccharomyces cereviase, Nucl. Acids. Res. 29 (2001), 1054-1060. [70] M. Lutsky and L.A. Mimy, Kinetics of protein-DNA interaction: Facilitated target location in sequence-dependent potential, Biophys. J. 87 (2004), 4021-4035. [71] X. Xheng, and C. Bevilacqua, Straightening of bulged RNA by double-stranded RNA-binding domain from the protein kinase PKR, Proc. Natl. Acad. Sci. USA 97 (2000), 14162- 14167. [72] P.L. Privalov, A.I. Dragan, C. Crane-Robinson, K.J. Brreslauer, D.P. Remeta, and C.A.S.A. Minetti, What derives proteins into the major or minor grooves of DNA? J. Mol. Biol. 365 (2007), 1-9.
312
FTIR Spectroscopy of Cells, Tissues and Body Fluids

Dieter NAUMANN1, Heinz FABIAN, Peter LASCH Robert Koch-Institute, Berlin, Germany
Abstract. The last 15 years have witnessed the development of modern infrared spectroscopy into a useful biodiagnostic tool for the analysis of cells, tissues, and body fluids. Dedicated technologies have evolved for rapidly discriminating between diverse microorganisms, testing single cells, and identifying various disease states in humans and animals. Particularly interesting applications arose by means of light microscopes coupled to infrared spectrometers. Infrared microscopes equipped with focal plane array detectors allow routinely the parallel collection of thousands of pixel spectra across microscopic areas of biological samples. This imaging technology can be used for automatic histological segmentation and imaging of tissue structures without dyes or molecular probes. Recent biomedical studies have proven that FT-IR imaging can be used to objectively differentiate benign from malignant histopathological structures in various tissues. Basically the same experimental set-up is also well suited to integrate the fundamental tasks of microbiological analysis, namely detection, enumeration, and differentiation of micro-organisms in one single apparatus. Due to its high brilliance, IR-synchrotron light coupled into high-quality FT-IR microscopes has been used for spectral mapping of single cells at a spatial resolution near the diffraction limit of mid-infrared light. The problem of extracting the characteristic information from the typically very complex, fingerprint-like infrared signatures of biological samples is generally addressed by applying bioinformatic techniques such as factor-, cluster-, linear discriminant analysis, and artificial neural networks together with so-called feature extraction algorithms. Examples are given on the characterization of micro-organisms, analysis of single eukaryotic cells, imaging of diseased human tissues, and disease recognition from body fluids that highlight the new possibilities of modern biomedical infrared spectroscopy. Keywords. Infrared spectroscopy, microspectroscopy, single cells, tissues, tissue segmentation, microorganisms, identification of microorganisms, principal component analysis, cluster analysis, artificial neural net analysis, body fluids, serum, infrared imaging, biomedical spectroscopy, storage compounds.
Introduction The last 15 years have witnessed the emergence of sensitive, rapid and increasingly precise physical techniques for biomedical analysis and medical diagnosis. These new techniques range from various spectroscopies such as nuclear magnetic resonance spectroscopy, mass spectroscopy, and molecular spectroscopy (including fluorescence, Raman and Fourier transform infrared spectroscopy) to different separation techniques like gas chromatography, high performance liquid chromatography and the use of
_______________________________________ 1 Corresponding Author: Robert Koch-Institute, Nordufer 20, D-13353 Berlin, Germany; E-mail: naumannd@rki.de
D. Naumann et al. / FTIR Spectroscopy of Cells, Tissues and Body Fluids
313
various imaging techniques such as positron emission tomography, computer-aided tomography, and magnetic resonance imaging. This development is paralleled by the dynamic advances of molecular genetic techniques in many fields of biology, microbiology, biochemistry and in the medical sciences, which presumably will develop into most sensitive and specific tools for biomedical characterizations in the future. In 1911 W.W. Coblentz was probably the first scientist suggesting that biological materials can profitably be analysed by means of infrared spectroscopy. Already in the 50s and 60s spectroscopists have demonstrated the feasibility to characterize microorganisms, higher organized cells and even tissues by infrared spectroscopy. Unfortunately, due to the lack of efficient computers and the weak instrumental specifications at that time, reports on microorganisms and tissue characterizations by infrared spectroscopy became less frequent in the 60s and ceased in the 70s [1, 2]. It is the development of modern interferometric infrared spectroscopy, the availability of low-cost mini-computers and powerful new algorithms of multivariate statistical analysis that contributed greatly to the revival of infrared spectroscopy as a means for characterizing biomedically relevant microbial and human cells, tissues, and body fluids in the last twenty years. This chapter is not intended to exhaustively review the details of technical developments and the FT-IR procedures developed for biomedical characterizations by several groups. This contribution will rather highlight some recent, and from the author's subject perspective, important applications and possible future developments in the field. The great progress made in the field of hard tissue (bone and cartilage) and efforts for in vivo applications of infrared spectroscopy [3-8] will not be reviewed here. Furthermore, only data will be considered that have concern with FT-IR on intact or even live cells, tissues, and body fluids. No attempt will be made to comment on the potentials of near infrared spectroscopy for biomedical diagnostic purposes or on infrared spectroscopy for the analysis of isolated cell fragments or compounds, macromolecules, and cell metabolites.
1. Composition and Structure of Complex Biological Material and the Nature of Spectral Information Understanding the nature of infrared spectra of microbial, plant, animal cells, tissues and body fluids requires at least a general understanding of its composition, major cell types and chemical structures present and some knowledge on the differentiation of cells and tissues. At the simplest level all biological systems are composed of water, lipids, proteins, nucleic acids, and carbohydrates. The gross composition of bacterial (prokaryotic), yeast, and mammalian (eukaryotic) cells has been described in the literature, and it is well-known that the composition and structure of these biological entities may be considerably different. 1.1. Microbial Cells In contrast to yeast, fungal, plant or e.g. mammalian cells (eukaryotic organisms), bacteria (prokaryotes) exist in only a limited number of morphological forms (e. g. rods, cocci, chains and spirals), their chemical composition and structures, however,
314
vary considerably. The cytoplasmic structures of bacteria are less organized (compartmentalized) and much simpler than those of animals, plants, yeasts and fungi, and have complex and very diverse macromolecular structures outside the cytoplasm. These include the cell wall, outer membrane, capsules and, sometimes, specific layers like e.g. the S-layers. Many chemical structures used to differentiate between bacteria reside in the cellenvelope, which is generally defined as the cytoplasmic membrane plus the cell wall. Most cell-envelopes fall in two categories, the so-called Gram-positive bacteria consisting only of the cytoplasm, the plasma membrane, and the cell wall, and the more complex Gram-negative which additional to the cytoplasmic membrane contain the socalled outer membrane. Some bacterial species, the mycoplasms, lack any cell wall at all, but express a rather rigid plasma membrane. The shape-giving function of bacteria is mediated by a rigid, high-molecular network made up primarily of the peptidoglycan, which protects the cells from osmotic disruption. Its primary structure consists basically of disaccharide-pentapeptide subunits with unusual features such as the occurrence of alternating D- and L-amino acids and a -bonded D-glutamic acid residue. Its structural variants are found to be different for various groups of bacteria. Many Gram-positive bacteria have additional polymers such as the teichoic and teichuronic acids covalently bound to the peptidoglycan. The teichoic acids are ribitol or glycerol containing macromolecules, built up by a phosphate carrying backbone with side-chains of variable composition. The mycobacteria, nocardia, corynebacteria and some related groups have very unusual cell-envelopes which form thick, wax-like layers around the outside of the cell wall. Major compounds present in this impermeable and rigid layer are complex, longchain fatty acids, the mycolic acids. Gram-negative bacteria exhibit an additional membrane, the so-called outer membrane. This outer membrane is the major permeability barrier that protects these cells against bile-salts, degradation by digesting enzymes and which prevails the cells from hydrophobic drugs and antibiotics diffusing through this particular layer. The inner leaflet of the outer membrane contains phospholipids as membrane forming lipids with nearly the same composition as found in the cytoplasmic membrane, while the outer leaflet embodies exclusively one particular type of amphiphilic molecules, the lipopolysaccharides. The various pore-forming proteins, the porins, are spanning both leaflets. Some bacteria form capsules (sometimes referred to also as "slime-layers") surrounding the cell-envelope. These are not essential structures and are frequently built up of unusual, negatively charged polysaccharide compounds. Some bacilli exhibit capsules composed of negatively charged homooligopolypepdides such as polyD-glutamic acids. The expression of such capsules in pathogenic bacteria can inhibit ingestion by phagocytes and may play an important role as "pathogenicity factors" in infectious diseases. Many bacilli and clostridia may form endospores which are modified cellstructures that can survive under unfavourable environmental conditions. These endospores have two membrane-like layers. Between these two layers, a spore-specific peptidoglycan (the so-called cortex) is found which differs in primary and three dimensional structure from the peptidoglycan of vegetative cells in that the muramic acid of peptidoglycan is modified to a lactam derivative and is less cross-linked. A keratin coat is located on the cell exterior and it has been reported that large quantities of Ca+2-dipicolinate are related to heat resistance of endospores.
315
1.2. Eukaryotic Cells, Tissues and Body Fluids Eukaryotic cells are organized into complex structures enclosed within membranes and are generally much larger than prokaryotic cells. They have a variety of internal membranes, organelles, such as mitochondria, and a cytoskeleton. Spatial and temporal variability of human cells or cells in tissues and the complex composition of body fluids is an inherent property of these structures. Temporal changes of cells is a consequence of alterations within tissues, e.g. as a function of cell cycle and age, or of cyclic processes as in the female reproductive system. As for microbial cells, the human body is primarily composed of water, different lipids, proteins, carbohydrates, and nucleic acids. These building blocks are organized into about 200 distinct types of cells which assemble to form various tissues. Only four basic types of tissues are found e.g. in mammalians namely, epithelial, connective, muscular, and nervous tissue. Epithelial tissues are built up by specialized and densely packed cells which stick to each other very strongly to form extended sheets that cover the inside and outside surfaces of the body, i.e. the skin, inner surface of blood vessels, and the gastrointestinal tract. A major constituent of connective tissues is an extracellular matrix secreted by cells which is mainly composed of protein fibers such as collagen and elastin and a matrix substance made up by various complex mixtures of glycoporteins and proteoglycans. Bone, adipose tissues and cartilage are specialized forms of connective tissue. Nervous tissues consist of two basic cell type neurons and neuroglial cells which support and nourish the neurons. The central nervous system (CNS) can visually be divided in grey matter, containing mainly neuronal cell bodies and glial cells, and white matter, which contain mainly the myelinated axons of the neurons. Muscle tissue can be divided into three main classes skeletal, cardiac and smooth muscle, according to different functions, visual appearance, and structural characteristics. Body fluids are found in the bodies of man and animals. They include fluids that are excreted or secreted from the body as well as fluids that normally are not. The most important fluids used in clinical chemistry and diagnosis include: Blood (frequently also defined as a tissue), blood plasma and serum, cerebrospinal fluid, synovial fluid, amniotic fluid, saliva, semen, tears, and urine. 1.3. Assignment of Infrared Absorption Bands Fig. 1 displays the infrared spectra of four main building blocks in biological samples (lipid, protein, nucleic acid, and carbohydrate), which are constantly present in complex biological materials. Some of these bands are numbered and tentative assignments given in Table 1. For practical purposes, rough band assignments may be obtained from group frequency charts published in several bibliographies that may help to obtain correlations between partial structures and band frequencies. Several spectra descriptions and excellent structure-spectra correlations for important biological macromolecules present in cells, tissues or body fluids can be obtained from the literature [9-12].
316
Figure 1. Mid-infrared spectra of macromolecular building blocks in biological systems. 1) Nucleic acids: RNA from yeasts. 2) Protein: Ribonuclease A. 3) Carbohydrate: Glycogen. 4) Lipids: Dimyristoylphosphatidylcholin (DMPC, synthetic). Spectra have been measured from dry film samples on ZnSe substrates. Bold numbers in brackets refer to the assignments given in Table 1.
Figure 2. Survey spectra obtained from different biological samples. 1) Body fluid (synovial fluid aspirated from a patient suffering from rheumatoid arthritis). 2) Microbial sample (Staphylococcus aureus). 3) Tissue section (central nervous system material from Scrapie-infected hamster brain). All samples have been measured as dried films deposited on a ZnSe optical plate.
317
Table 1. Tentative assignment of some bands frequently found in biological FT-IR spectra Band numbering (cf. Fig. 1 ) Frequency (cm-1) ~ 3500 ~ 3200-3300 1 2 3 ~ 2955 ~ 2930 ~ 2918 ~ 2898 ~ 2870 4 5 ~ 2850 ~ 1740 ~ 1715 6 ~ 1680-1715 1690 7 1685 1675 8 9 10 11 12 13 14 15 16 17 18 ~ 1655 ~ 1637 ~ 1550-1520 ~ 1515 ~ 1468 ~ 1400 ~ 1310-1240 ~ 1250-1220 ~ 1200-900 ~ 1090-1085 ~ 720 ~ 900-600 amide I band components resulting from antiparallel pleated sheets and -turns of proteins amide I of -helical structures amide I of -pleated sheet structures amide II tyrosine band C-H def of >CH2 (scissoring) C=O str (sym) of COOamide III band components of proteins P=O str (asym) of >PO2- (phosphodiesters) C-O, C-C str, C-O-H, C-O-C def (of carbohydrates) P=O str (sym) of >PO2C-H rocking of >CH2 (in fatty acid chains) "fingerprint region" Assignment O-H str of hydroxyl groups N-H str (amide A) of proteins C-H str (asym) of -CH3 in fatty acid chains C-H str (asym) of >CH2 C-H str (asym) of >CH2 in fatty acid chains C-H str of C-H in methine groups C-H str (sym) of -CH3 C-H str (sym) of >CH2 in fatty acid chains >C=O str of esters >C=O str of carbonic acid >C=O in nucleic acids
str = stretching; def = deformation; sym = symmetric; asym = antisymmetric
318
Figure 3. Tentative assignments for some major functional group frequencies found in a typical biological sample (Staphylococcus aureus). Top: Original absorbance spectrum as measured from a dried film sample of Staphylococcus aureus and its second derivative spectrum. Bottom: FT-NIR-Raman spectrum of the same dried sample for comparison.
Fig. 2 shows typical spectra of a microbial, body fluid, and tissue sample, each dehydrated to a thin film disc on a ZnSe optical substrate. While exact assignments to discrete molecular compounds are certainly not possible, many spectral features can at least be made visible for the eyes applying resolution enhancement techniques. In general 40 to 60 spectral features are resolved applying e.g. second derivative calculation (see Fig. 3, top spectrum). Tentative assignments based essentially on the comparison of resolution enhanced spectra with those of the main building blocks in whole cells can be derived from Table 1, some are given in Fig. 3. 1.4. Infrared Spectra Provide Fingerprint-Like Signatures of Biological Structures Biomedical infrared spectroscopy is testing biological samples in a way that the infrared active vibrational modes of all constituents present in the mixture are observed in a single experiment, resulting into very complex spectra with broad and superimposed spectral features throughout the whole spectral range. Thus, infrared spectra of intact cells, tissues or body fluids cannot provide information on a single or even a few specific compounds present. The spectra rather provide spectroscopic fingerprints of the total chemical and biochemical composition of the material under study. This situation inevitably results from the fact, that the complex superposition of the characteristic absorption signals of all constituents in biomaterials (nucleic acids, proteins, carbohydrates, lipids, and other low molecular compounds etc.) are observed simultaneously producing spectral features that "encode" an enormous amount of information potentially useful for biodiagnostic purposes.
319
One very attractive peculiarity of vibrational spectroscopy is that it does not only provide information on the total composition of complex biological material but also on structural states of the molecules under study, since certain bands are sensitive e.g. to the secondary structure in proteins, while others report on the state of order of the membranes or the conformation of the nucleic acid structures. In this sense the total information content in vibrational spectra of biological materials is enormous. One can possibly say that, with the exception of Raman spectroscopy, there are presently no other techniques available that can provide such a huge amount of information in one single experiment. Infrared spectra of cells, tissues, and biofluids are the expression of the sum of cellular chemistry/biochemistry and give a snapshot on cell division, differentiation, growth and metabolism. Thus, infrared spectra may report on the chemistry, biochemistry and physiology or the response to environmental stress in biological entities. In this view, biological infrared spectroscopy is a "phenotypic" and explorative analysis modality par excellence that can be used to diagnose disease or dysfunction in cells, tissues and bio fluids via "spectral fingerprints" and may change as an indicator of the presence of a particular disease or in response to drug intervention, environmental stress or genetic modification. The fundamental fingerprinting nature of infrared spectra of complex biological samples is thus a big advantage, but also a drawback, since comprehensive understanding of these spectra is desirable but not achievable in most cases despite some pretentious papers in the literature.
2. Experimental Techniques and Methodologies A major advantage of IR spectroscopy is that nearly any kind of material can be measured and that it is not limited to the physical state of the sample (samples may be solutions, viscous liquids, suspensions, inhomogeneous solids or powders). Additionally, there are no principal restrictions to record infrared spectra of a given sample under different physicochemical condition concerning temperature, pressure, state of dispersion, pH etc.. This is of direct relevance for biomedical analyses, since it is pertinent to test biological specimens under conditions that leave the sample's structures "as they are", preferentially hydrated, unperturbed and non disintegrated. 2.1. Infrared Techniques Suitable for Biomedical Analysis In general, a biological sample does not behave ideally in that it is uniform in thickness and refractive index or optimal in concentration etc.. It has been suggested by scientists since long that biological specimens can best be compared when: (i) infrared absorbances of the samples to be compared are as similar as possible, (ii) IR-bands are not too intensive in order to avoid detector non-linearities, (iii) the problem of varying baseline shifts due to diffuse scattering at the sample surface and/or due to inhomogeneity within the sample itself is minimized, and (iv) signal-to-noise ratio is sufficiently high. These requirements can still be best fulfilled by the traditional absorbance/transmission and the attenuated total reflection techniques.
320
2.1.1. Absorbance/Transmission Measurements (A/T) Running a traditional absorbance /transmission (A/T) spectrum is a straightforward way of "looking" at a sample. A/T spectra can be obtained from liquid solutions, dispersions or suspensions, viscous solutions or from dried solid films cast on suitable IR transparent substrates. Since water is the ubiquitous medium of all biological samples, water-insoluble and IR transparent optical materials have to be used. These are CaF2, BaF2, ZnSe, ZnS, KRS-5, or germanium which differ in refractive index, spectral transparencies, and water-solubility. Different cuvette systems have been designed for IR measurements of biological samples. Two commercially available technical solutions for the analysis of many biological specimens as dried or fully hydrated film samples are shown in Figs. 4A and B, respectively. In the literature the notion is often found that infrared spectroscopy is generally not suitable for biological materials in aqueous surrounding. Recently several papers have been published describing straightforward and relatively simple methodologies to perform transmission/absorbance mid-infrared spectroscopic measurements on fully hydrated proteins, cells or biofluids in solution or suspensions [13-15]. For analysing intact bacterial cells the authors chose CaF2 as the optical material because of its low refractive index and low solubility in water and a special cuvette system as shown in Fig. 4B. Provided the cell density is high enough as is the case for microbial cell smears taken directly from solid culture plates or cell pellets after centrifugation the
Figure 4. (A) FT-IR instrument IFS 28/B from Bruker Optics Inc. and a cuvette useful for measuring dry film samples. (B) Layout of a cuvette for fully hydrated samples.
321
subtraction of the water or buffer background turned out to be not necessary, when the amide I spectral region was excluded from analysis and first or second derivatives were used for data analysis. Using a commercially available, high-pressure stable flowthrough cell [16] together with sensitive broad band MCT detectors, the quality of spectral information obtained in this way was sufficient or even better than obtained with the same samples dried to film discs [14]. Thus, drying of the samples on infrared transparent optical substrates was not required and the necessity to rigidly control the relative humidity of the dried samples and unfavourable scattering at the sample surface due to heterogeneity of the dried film discs could be avoided. The possibility to measure body fluids like blood serum as they are, that is in aqueous solution, will be demonstrated in section 5.1.1. 2.1.2. Attenuated Total Reflectance Measurements (ATR) Theory and experimental principles of ATR-spectroscopy are described elsewhere [17, and literature cited therein]. Briefly, when performing an ATR experiment, the sample is no longer placed in the path of the propagating IR-beam as for A/T-experiments, but is rather brought into contact with the surface of an "internal reflection" plate or prism, where it can interact with the IR radiation "evanescing" from the optically denser (ATR plate) to the rarer (sample) medium. The so-called penetration depth of the electromagnetic wave into the rarer medium is defined by the ratio n2/n1 of the refractive indices of the denser (n2) and the rarer (n1) media, the wavelength , and by the angle of incidence and is in the order of a few micrometers. For a defined ATRelement (n1) and angle of incidence the "effective" optical path length is similar for different samples, provided they have comparable refractive indices and the optical contact between ATR element and samples is the same. This is of great practical advantage when comparing different samples of complex biological materials, since the above mentioned experimental prerequisite can be fulfilled at least approximately for a large number of biological samples. Specially designed ATR-cuvettes suitable for the measurement of biological samples as dried or fully hydrated film discs on ATR crystals have been used by several groups [17-20]. 2.1.3. Infrared Microspectrometry The light microscope is a standard analytical instrument in many research and routine laboratories. Its broad acceptance is certainly due to its capability of providing diagnostic information on different morphologies, colours etc. of nearly any kind of material. The performance of the microscope can be greatly enhanced by various illumination techniques and the use of polarized light and allows for rapid discrimination between particulate matters in complex mixtures. While shape, colour, contrast etc. provide information on the optical properties of a given sample, vibrational spectroscopy may give spatially resolved information at the molecular level and on the identity of structures within a given complex mixture. Thus, the combination of a light microscope with the specificity of FT-IR spectroscopy may provide magnitudes of additional information. Already as early as in 1949, it was suggested to combine an infrared spectrometer with a traditional light microscope. However, because of the limitations of classical dispersive IR spectrometers, these first trials remained restricted to a few research laboratories. With the development of modern FT-IR spectrometers and the availability and of small and sensitive MCT detector elements, the performance of microscopes coupled to FTIR spectrometers
322
could be significantly improved and high quality FT-IR microscopes appeared on the market. This combination pushed the detection limit of infrared spectroscopy down to the subnanogram level and opened the field of spatial resolution and imaging to infrared spectroscopy [21, 22]. Since the conventional light microscope uses condensors and objective materials that have limited transparencies for IR radiation, the "FT-IR microscope" generally requires all-reflecting optical devices, so-called cassegrain objectives. An assortment of apertures placed in the plane of the real intermediate image of the sample is used to mask the sample areas of interest down to 10 m lateral resolution, close to the diffraction limit of infrared light used. Using such a FT-IR microscope and collecting pixel spectra across a sample with a computer controlled x,y stage, a complete image of the spatial distribution of different histological structures in a microtomed tissue slice can be obtained [23, 24]. Recently so-called focal plane array (FPA) detectors have been implemented in research grade infrared microscopes that substituted the single element detectors by a matrix of individual detector elements. There are two basic types of focal plane arrays available: linear and area. Linear focal plane arrays consist of a single line of detector elements (pixels). Area focal plane arrays consist of rows and columns of pixel elements. These FPA detectors made it possible for the first time to measure many microscopically small spots in heterogeneous biological samples simultaneously without the necessity to map step by step over the sample and thus paved the way for true infrared imaging of biological specimen [21, 25]. 2.2. Sampling of Biological Material, Data Acquisition Sampling biological specimen, sample preparation and the selected FT-IR spectroscopic technique used to acquire spectral information are key features in biomedical infrared spectroscopy, which directly have impact on the reproducibility and repeatability of measurements. There is certainly not a single answer to these questions. However, for cell suspensions, body fluids, and tissue samples a number of reasonable suggestions have been published in the literature [26-33]. For microbial cells, to give an example, standardized sampling, sample preparation, data acquisition, and evaluation techniques have been described in detail [26, 34-36]. These standardization efforts have been stimulated by the necessity to exchange spectral data between different laboratories and construct reference data bases for routine analysis and identification of microorganisms. FT-IR spectra with sufficient reproducibility can e.g. be obtained from microorganisms grown either in liquid cultures or on standard solid agar media, provided the microbiological parameters influencing cell growth (composition of growth media, incubation time, temperature of growth etc.) can be controlled and standardized rigidly. Sample collection, sample preparation, and spectroscopic data acquisition parameters (spectral resolution, scanning time etc.) are then of secondary importance. Microbial samples suitable for IR-measurements can be obtained from liquid cultures or directly harvested from solid nutrient agar plates. These samples can be measured as hydrated pellets or dried films applying either the A/T or ATR techniques. A typical, very simple protocol, used to test microorganisms with the devices shown in Fig 4 can be taken from the literature [26, 34-36]. To achieve data compatibility, the physical parameters have to be kept constant for all measurements to be compared. A suitable set of data parameters for measurements with the device of Fig. 4, to give an example, is as follows: 6 cm -1 nominal spectral
323
resolution, Blackman-Harris 3 term apodization function, one point per wavenumber data point resolution and a sufficient number of scans to reach a signal-to-noise ratio better than 3000:1. It is advised to take the single beam reference spectrum through an empty place of the multisample cuvette system (see Fig. 4) directly before the single beam sample spectrum is obtained. This eliminates virtually all contributions from impurities on the optical materials and minimizes problems arising from water vapour and CO2 bands due to possible instabilities of dry air purging of the instruments. 2.3. Variability of Biological Specimen, the Problem of Reproducibility and Standardization The most serious problem when analyzing complex biological materials is the enormous diversity and complexity of cell types, tissue structures, or body fluids that makes it an absolute necessity to test statistically significant numbers of samples in independent measurements. Repeatability and reproducibility of sampling, sample preparation, and measurement significantly influences data structure of the primary data base. Especially when reference data bases are to be established for use in different laboratories one has to establish reproducibility and repeatability of the measurement technique in quantitative numbers. This is only possible when standard operation procedures are specified that are accepted by the scientific community. For measurements of microbiological specimens, to give an example, it is useful to define different levels of reproducibility (RLi) when repetitive IR measurements on microbiologically identical samples are to be compared (physical parameters must be kept constant): 1. RL 0 for measurements of identical samples of a given strain on different instruments in different laboratories. This gives the possibility to quantify the instrument to instrument variation when measuring identical biological samples. 2. RL 1 for measurements of independent sample preparations obtained from identical bacterial suspensions (see 4.1) of a given strain measured with the same instrument. This gives the possibility to quantify the repeatability of the sample preparation procedure. 3. RL 2 for measurements of independent sample preparations from different bacterial suspensions of a given strain grown on the same type of agar plates produced from the same batch of culture medium. This gives the possibility to quantify the influence of "biological variance" on experimental repeatability. 4. RL 3 for measurements of independent sample preparations from different bacterial suspensions of a given strain grown on the same type of agar plates produced over a long period of time from different batches of culture medium. This gives the possibility to quantify the repeatability of measurements achieved under practical working conditions To calculate quantitative numbers for reproducibility, an objective measure for comparison of independent measurements obtained in the same or in different laboratories is pertinent. One possibility is the cross-wise calculation of the correlation coefficient (Pearson's product-momentum correlation coefficient) between pairs of measured spectra [18, 26, 36]. From the Pearson's correlation coefficient , a so-called distance value or differentiation index "D" can be used as defined in the literature [18, 26, 34-36].
324
In series of measurements on 13 "test microorganisms", which had been defined for a joint pilot study [35], the D-value distributions (in general Poisson-like distributions) for the various RL as defined above were determined. Using, for example, the first derivatives calculated in the spectral range between 1200 and 900 cm-1 typically yielded D-values of D < 0.2 0.1 for RL 0, D < 0.4 0.4 for RL 1, D < 2 1 for of RL 2, and D < 10 8 for RL 3, (D: mean D-value standard deviation). Thus when comparing D-values between different laboratories one has to define precisely the spectral range of interest. The "practical" reproducibility level RL 3, although still low, is more than an order of a magnitude higher than that of the sample preparation technique itself (see RL 1 above) and the instrument to instrument variation (see RL 0 above). Hence, the conclusion may be drawn that microbiological parameters (e.g., the quality of the growth medium and batch to batch variation) are essentially defining the limitations of the FT-IR technique. Considering that the use of different batches of a given medium seems to be the minimum prerequisite for working on a practical level, a mean D-value of less than about 10 (spectral range 1200900 cm-1) limits the actual differentiation capacity of the FT-IR approach. Similar calculations have been performed from repetitive measurements of body fluids [14]. The effect of spatial variability within eukaryotic cells and tissues is self-evident. Quite different spectra will be obtained depending on the sample spot where the spectrum is taken. Furthermore, the spectra will depend on the lateral resolution (different apertures) or the particular FPA detector used with the FT-IR microscope (see also sections 2.1.3, 5.2, and 5.3).
3. Data Treatment and Evaluation Techniques In biomedical infrared spectroscopy not only a single spectrum has to be analysed at a given time but thousands and in case of spectroscopic imaging even hundred thousands of spectra. Thus, the availability of sophisticated data evaluation concepts is a virtual necessity. These concepts should ideally include efficient data pre-treatment algorithms such as quality testing, normalization, filtering, and adequate techniques for data reduction. The necessity to use multivariate pattern recognition methodologies when dealing with spectral data of complex biomedical materials has been realized by the infrared spectroscopic community more than 20 years ago. One of the first who recognized this problem had been scientists working with FT-IR spectroscopic data of intact microorganisms [37]. The arsenal of multivariate pattern recognition techniques provides methodologies for the pre-treatment, evaluation, and representation of huge and complex data sets. These techniques give at hand not only the possibility to obtain a "survey" over the data, but also enable the direct analysis and the interpretation of structures and interrelationships between these. While univariate statistical analysis considers only a single property of a given object, multivariate statistics evaluate several properties of the objects at the same time. In this way also the interrelationships between the properties can be taken into account. Out of a large number of statistical techniques available, three have been frequently and successfully used when dealing with infrared spectra as the basis of biomedical diagnosis. These are principal component analysis (PCA), hierarchical cluster analysis (HCA), and artificial neural network analysis (ANN).
325
Figure 5. Principal component analysis map obtained from a data set of approximately 190 FT-IR spectra measured from Gram-positive and Gram-negative bacteria and yeasts. For two-dimensional projection of the data, the factorial coordinates (factor loadings) F2 and F3 are used. As input data for factor analysis, the normalized first derivatives in the spectral ranges of 2800 to 3000 cm-1 and 1400 to 1500 cm-1 have been used. For each strain 10 independently measured spectra have been analysed.
The difference between these methods is that PCA is used primarily to achieve data reduction and the classification of patterns, while hierarchical clustering, a so-called unsupervised classification method, attempts to find intrinsic group structure within a complex data set without the need of any a priori class assignment or partitioning of the data into training and test data sets. In contrast to hierarchical clustering, artificial neural network analysis is a supervised classification approach by which the class assignment of each individual (sample, spectra etc.) is needed from the beginning. Partitioning of the whole data into a training and test data set is needed to recognize overfitting and ensure reliability of results. Robustness of classification is further enhanced by cross-validation using the stringent leave-one-out method or better by external validation. Details can be taken from the literature [38]. Fig. 5 shows a so-called PCA-map obtained for a data set of approx. 190 spectra collected from independent measurements on strains from different species comprising Gram-positive, Gram-negative bacteria and yeasts. Microbial samples have been prepared for measurement and the spectra collected according to the standard experimental protocol as described in [18, 26, 34-36] using the multicuvette system and the FT-IR spectrometer shown in Fig. 4A. The map has been constructed by two-dimensional projection of factor spectra 2 and 3 such, that the intrinsic group or class structure of the data set
326
Figure 6. Minimal spanning trees, also called "dendrograms", graphically showing the fusion steps of cluster analysis process on the same spectra of Fig. 5. (A) Dendrogram obtained by calculating the distance matrix on the basis of the distance values in spectra space. (B) Dendrogram obtained calculating the distances in factor space using the first 3 factorial coordinates (factor loadings). Three spectral ranges were used: 30002800, 1500-1400, and 1200-900 cm-1. Spectra were pre-treated by calculating the first derivatives. The Ward's algorithm [J.H. Ward, Journal of the American Statistican Association, 58, 236-244, 1963] was used to calculate the dendrogram.
can already be seen by visual inspection. Each point in the map represents a spectrum, the factorial coordinates ("factor loadings") 2 and 3 are used for projection of data.
327
PCA was applied using the first derivatives in the spectral ranges between 2800 - 3000 and 1400 - 1500 wavenumbers as input data. Three diffuse data clouds can be distinguished comprising exclusively the Gram-positive and the Gram-negative bacteria and yeast strains, respectively. The essential information for data pre-treatment is given in the legend of Fig. 5. The history of hierarchical classification analysis is represented by a minimal spanning tree, also called a "dendrogram" in which the merging process of classes can be visually followed as is exemplary shown by the two dendrograms of Fig. 6. These dendrograms have been calculated with the same spectra used for constructing the PCA-map of Fig. 5. Fig. 6A gives the dendrogram after calculating the distance matrix directly from the spectra, while the dendrogram of Fig. 6B has been obtained calculating the distance matrix from the spectra in factor space after PCA-analysis. The classification schemes of Fig. 6 do not provide an objective criterion of "best partitioning". In some way or other, the number of classes has to be predetermined by the user, who needs at least some a priori knowledge about the inherent class structure of the data cloud. The final decision whether the partition is useful or not is a subjective decision. Meanwhile nearly the whole arsenal of multivariate bioinformatic techniques has been used and multivariate statistical analysis of spectroscopic data constitutes an own discipline within the scientific area of biomedical spectroscopy. As for any other scientific discipline, these methods can not only be used to evaluate given data sets, but allow completely new problem solutions to be addressed. New applications arose e.g. when it has been realized that determining the covariance between different large data matrices obtained from the same sample populations with fundamentally different techniques is not only a challenge per se, but also provides insight into the interlink between biological structures. One of these new applications recently published was the use of genetic algorithms in combination with partial least square regression (PLSR) analysis to correlate genes selected from gene expression profiles (obtained by micro-array technologies) to metabolic markers from spectral data sets measured from the same samples by IR spectroscopy [39-41]. The analysis of covariance patterns in these very complex mixed data sets helped to rapidly recognize and visualize the interrelationships and trends in a developing and changing biological system that is not easily achieved by any other means.
4. Characterization of Microorganisms
4.1. Detection and characterization of specific cell components The detection of a single or even a few specific components in a complex mixture of chemical/biochemical species in cells, tissues and biofluids is intrinsically problematic. In some cases however, microbial spectra were obtained which showed some extra bands of variable intensities that could not be considered as spectral variation due to experimental conditions or changes of microbiological parameters. A detailed analysis of these spectral features revealed the presence of cell constituents such as intracellularly accumulated storage materials, cell surface structures like proteineacous or polysaccharidal capsules, and endospores [42, 43].
328
Poly--hydroxy fatty acids (PHF) are energy and carbon reserves found in prokaryotes which also turned out to be biotechnologically interesting macromolecules. In many cases PHF compounds are accumulated under limitation of nutrients when supply of energy and carbon is in excess. Under conditions of starvation, PHF can be utilized and degraded by the microorganisms helping the cells to survive under severe starvation conditions. It is known that the survival rate is related to the amount of PHF, which is intracellular accumulated as small granules. These granules can easily be detected light-microscopically or by electron microscopy. Poly--hydroxybutyrate (PHB), to give an example, is frequently found in bacteria (e.g. Bacilli, Acetobacter, Pseudomonas). In most microbial IR-spectra the ester carbonyl band of lipids at 1738 cm -1 is only a small, weakly expressed shoulder. This band is caused predominantely by the >C=O stretching vibration of ester bound fatty acids, such as present in the membrane
Figure 7. FT-IR spectra of Legionella bozemanii strain L2165 after different cultivation times. Spectra obtained after 72, 96, 120, 168 and 192 h, respectively and the difference spectrum between 48 and 240 h are shown. Some absorption bands typical for PHB are annoted. The inset gives a graph showing the relative amount of PHB as a function of growth time. The intensity of ester carbonyl band at 1738 cm-1 (I1) and amide II band near 1550 cm-1 (I2) has been used to determine relative PHB-amount from the equation = I1/I2 (see text). Strains were grown on charcoal yeast extract (CYE) agar plates at 37 oC.
329
forming phospholipids. Gram-negative bacteria usually show a stronger ester carbonyl band than the Gram-positive organisms due to the presence of an additional membrane layer, the outer membrane, which is built up by an asymmetric bilayer composed of a phospholipid layer at the inner and the so-called lipopolysaccharides at the outer layer leaflet. In spectra of some microorganisms, however, a rather prominent ester >C=O stretching band, accompanied by a number of additional bands between 900 and 1500 cm-1, is observed. Interestingly, these bands turned out to be not permanently present throughout cultivation [42, 43]. In spectra of some Legionella strains, for instance, this band reached a maximum and then again a minimum during cultivation. Fig. 7 shows a series of FT-IR spectra collected of Legionella bozemanii samples taken directly from cultures growing on solid nutrient agar plates. For sample preparation the experimental protocol described in [26, 36] was used. The difference spectrum at the bottom of Fig. 7 has been calculated from the two spectra taken at 48 and 240 hours, respectively. This difference spectrum closely resembles the FT-IR spectra recorded for isolated and purified PHB (spectrum not shown). At least ten bands were identified and assigned to a typical polyester compound. The relative content of PHB can be semi-quantitatively estimated by calculating the ratio = I1/I (see Fig. 7), where I1 is the intensity of the ester carbonyl peak at 1738 cm-1 used as a measure of PHB content, and I2 the intensity of the amide II peak at 1550 cm-1 used as a measure for approximate total cell mass. The inset at the top of Fig. 7 shows a curve obtained in this way from the spectra shown in Fig. 7. 4.2. Differentiation, Classification and Identification of Microorganisms The microbiological characterization of unknown clinical, food, water, or air born microbiological specimens include the fundamental steps of detection, enumeration, and differentiation of microbial cells. In the practice of microbiological analysis quite different techniques are used to detect, count, and differentiate microorganisms. These techniques are e.g. the counting of colony forming units, the measurement of optical density, the use of cell counters or cell sorters and the application of the whole arsenal of techniques by which microbial cells can be differentiated including modern molecular biology methodologies. Bacterial diversity is always structural and biochemical diversity. As already stated, FT-IR spectroscopy on intact microorganisms provides information on the structure and composition of the whole cells. It is therefore a potentially valuable tool for characterizing and differentiating microorganisms at very different taxonomic levels. In the last 15 years FT-IR spectroscopy has been used by many groups to differentiate, classify, and identify microorganisms [26, 36, 37, 44-65]. In order to appreciate these efforts, two examples of typical bacterial classification trials based on whole cell FT-IR spectra are given which also show the flexibility of the technique. Fig. 8A shows a dendrogram obtained by hierarchical cluster analysis performed on various different species and strains of the genus Candida. To obtain these results, sub-cultured microbial strains were cultivated on solid nutrient agar plates (Sabouraud, 2% glucose). Time and temperature of growth was 24h and 370C, respectively. Small amounts of the bacteria were carefully removed from the agar plates with a standard platinum loop and directly transferred as fully hydrated smears between two CaF2-plates of the multi-cuvette shown in Fig. 4B and the
330
A
2500 3500 3000
Heterogeneity 2000 1500 1000 500
0
Candida albicans Candida tropicalis Candida kefyr Candida parapsilosis Candida glabrata Candida krusei
Wards Algorithm 2nd Derivative
Frequency Ranges =-1 1070 - 1014 cm -1 1171 - 1099 cm -1 1420 - 1339 cm -1 1662 - 1670 cm
B
3000 2500
Heterogeneity 2000 1500 1000 500
0
Control C 1580/95 Control A 23 I/96 1810/95 3142/94 317/95 723/95 2200/94 1890/95 1887/95 2147/95 828/95
Ward's Algorithm 2nd Derivative
Frequency Ranges = -1 891 - 859 cm -1 761 - 735 cm -1 991 - 889 cm -1 1401 - 1369 cm
Figure 8. (A) Dendrogram of a hierarchical cluster analysis performed on strains belonging to 6 different species of the genus Candida. (B) Spectral typing of closely related strains and isolates of the species Candida albicans. In (A) and (B) cluster analysis was performed using the second derivatives and considering the spectral ranges shown at the bottom of the dendrogram. Ward's algorithm was used for clustering the spectra and all spectral ranges were equally weighted [26]. For calculating the distance matrix the so-called D-value [26] was used as distance measure. All spectra have been measured from fully hydrated samples directly taken from the nutrient agar plates using the multisample cuvette shown in Fig. 4B.
spectra measured with the FT-IR spectrometer shown in Fig. 4A. The physical parameters were kept fixed for all measurements. A suitable set of acquisition parameters was as follows: 6 cm-1 nominal spectral resolution, Blackman-Harris 3 term apodization function, one point per wavenumber data point spacing and 64 scans to
331
reach a signal-to-noise ratio better than 3000:1. The single beam reference spectrum was taken through an empty place of the multi-sample cuvette system (see Fig. 4) directly before the single beam sample spectrum was measured. This eliminated virtually all contributions from water vapour and CO2 bands due to possible instabilities of dry air purging of the instruments. For the classification schemes shown in Fig. 8 hydrated samples directly taken from the solid nutrient agar plates have been prepared and measured with the multisample cuvette system shown in Fig. 4. From the dendrogram of Fig. 8A it is evident that the different strains could be successfully classified at the species level. Additionaly these spectra could also be used to differentiate the different strains of Candida albicans at the sub-species level as is documented by Fig. 8B. To obtain the dendrogram of Fig. 8B a different combination of spectral ranges was used as documented in the legend. While the identification of the different species may be relatively easy to achieve by conventional techniques, sub-species differentiation for epidemiological purposes is labour intensive, time consuming and only practicable by trained personnel. There is presently no other technique available that can achieve species and strain differentiation similarly quick and easy. It should also be mentioned that such results can be achieved on the basis of a single spectrum obtained from individual yeast strains. Many examples of species and strain differentiation by FT-IR spectroscopy have been published and one can probably say that FT-IR spectroscopy of microorganisms is presently the most frequent and best developed application of biomedical vibrational spectroscopy (see recent review [66] and the literature cited therein). 4.3. Infrared Microspectroscopy of Microorganisms In clinical microbiology time needed for completion of analysis is important. The earlier identification results are obtained, the sooner appropriate antibiotic therapy can be started. This will contribute to decrease mortality rates and variable costs for the hospitals. Ideally, one would expect to get identification results on the same day the patient material is obtained. Since spatial resolution of FT-IR microscopes is limited by the diffraction of infrared light used for analysis (in this case in the order of micrometers), single microbial cell analysis is not feasible. However, the analysis of small microcolonies in the order of 20 to 100 m growing on solid nutrient agar plates after short cultivation times is feasible. In order to sample microbial cells for FT-IR microscopic measurements, the following standard operation procedure has been established [18, 34, 43, 67]: Aliquots of the microbial cell suspension, sufficiently diluted to guarantee single colony growth on solid agar plates, are plated and incubated over a period of 6-8 hours. After these growing times, colony formation is generally not yet visible by eyes. A round IR transparent plate made e.g. of ZnSe, CaF 2 or BaF2 is then gently pressed to the agar surface using a special stamping device [67]. This imprinting technique transfers spatially accurate small amounts of the microcolonies (two to three microbial layers) to the optical plate and provides replica of dried microbial film spots that can be measured by the FT-IR microscope. With the aid of a computer controlled x,y-stage these spots are measured automatically or operator controlled using digital cameras and imaging techniques. Additionally, the number of colony spots can be counted, and size classification of these spots is also possible.
332
Figure 9. Detection and classification of bacteria by FT-IR microspectroscopy based on microcolonies. (A) Micrograph of different colony spots deposited on a ZnSe substrate by the stamping technique described in [18, 34, 43, 67]. Optical properties and size of the microcolony imprints suggest the presence of 3 different colony types. (B) Dendrogram of hierarchical cluster analysis performed on spectra of 60 different colony spots. The clusters prove the presence of three different bacterial specimen: Staphylococcus aureus strain DSM 20231; Streptococcus faecalis strain DSM 20371; Escherichia coli strain RKI A139.
333
Fig. 9 gives the example of measurements performed on a mixed culture containing three different microorganisms, namely Staphylococcus aureus, Streptococcus faecalis and Escherichia coli. Fig. 9A gives the micrograph of a selected area of the colony imprint, which shows a number of different microcolonies that can already be morphologically differentiated by eyes using criteria like colony size, colony form, surface roughness, etc.. Single spectra obtained from these colony spots exhibit typical spectral features (not shown), which immediately suggest that the microcolony imprints have been derived from different microorganisms. In a second step of analysis, a representative sampling of FT-IR microscopic spectra of bacterial colony spots is obtained and the spectral data subjected to multivariate statistical analysis. Fig. 9B shows results when using hierarchical clustering of measurements on 60 microcolonies, which unequivocally prove that the spectra obtained from these microcolonies are indeed characteristic of three different microorganisms present in mixed culture. During the last few years several reports have been published on the microcolony based identification and typing of bacteria and the analysis of mixed populations without the need to isolate pure cultures [44, 45, 60, 67]. In a recent study published by Maquelin et al. [44], to give an example, a spectral data base of reference spectra and an optimized ANN model was elaborated to identify medically relevant organisms from blood cultures in a typical clinical setting. This data base included FT-IR spectra of 89 strains of Gram-negative bacteria from the genera Enterobacter, Escherichia and Pseudomonas as well as Gram-positive cocci from the three genera Staphylococcus, Enterococcus, and Streptococcus. Using this data base and ANN model, the authors were able to correctly identify 112 from 114 (98.2 % accuracy) isolates from clinical origin in a prospective clinical study. Rebuffo-Scheer et al. [60] have analyzed mycobacteria, which are slow growers, and could prove that species identification and subspecies typing could be achieved after significantly reduced cultivation times from 10-15 days to 40-50 hrs. A novel application of FT-IR microspectroscopy in microbiology was recently published by Kirkwood et al. [68, 69], who used a FPA detector equipped FT-IR microscope together with microarray printing of microbial samples carried out by highly automated robots. In this way hundreds of spotted samples could be placed on optical substrates for automatic FT-IR measurement. 4.4. The Problem of Identification and Establishment of Validated Spectral Reference Libraries Once reproducibility and repeatability of the technique are defined and quality control measures established, important prerequisites for elaborating a spectral reference date base are: (i) a representative number of different species per genus and reference strains per species, (ii) a well balanced composition of the data base with no species dominating the others, and (iii) all reference strains should be unequivocally identified by independent methods to serve as real references. Several ways of establishing such spectral data bases have been described in the literature. One simple possibility to identify unknown strains is to use a univariate distance measure, the so-called D-values as defined above. This possibility has originally been proposed by Helm et al. [26, 36] and turned out to be quite useful [1, 10, 14, 16, 20, 22]. Subsequently this approach was implemented under the Identity Test module of the OPUS software provided by Bruker Optics Inc. and is used by several groups worldwide.
334
Much more flexibility and efficiency is however provided by artificial neural networks (ANNs). Applying ANNs, complex identification problems could be solved by combining several optimized ANN modules in a hierarchical architecture [44, 59, 70]. Recently the superiority of ANNs over other methods was shown for the identification of Listeria by Rebuffo et al. [59]. The power of ANNs is now recognized by many groups and ANNs are increasingly used for differentiation problems in microbiology [71-74]. It has been emphasized in the literature, that the validity of all identification models has to be checked objectively in order to avoid artefacts such as overfitting produced by the techniques used. For this purpose internal validation and external validation has to be documented. For internal validation only new spectra of strains contained in the reference dataset are used and identification rates should be high. Otherwise, the robustness of the model is insufficient and needs modification. Only by external validation the performance of the model can be tested objectively, since spectra of unknown strains not included in the reference dataset are used as independent test cases. Only from this validation process the identification accuracy expected under practical conditions can be predicted. Several applications for FT-IR based reference data bases have bee published: FTIR based characterization of the Gram behaviour or further differentiation into families like Enterobacteriaceae or Pseudomonadaceae [36] can substitute Gram-staining and test for the presence of cytochrome oxidase, recognition of spore forming bacilli by the detection of specific spectral markers for endospores [42, 74], or the production of storage materials like e.g. polyhydroxybutyric acid (PHB) [42, 43, 45]. Databases for the identification of micro-organisms are already available. Bruker Optics Inc. (Germany) offers datasets for gram testing and for species identification of water-borne micro-organisms. Furthermore, a number of libraries for species identification of Listeria, bacilli, staphylococci, actinomycetes, enterobacteria, members of the genus Pseudomonas, lactic acid bacteria, and yeasts comprising more than 700 different species have been compiled by the Department of Microbiology, Technical University of Munich (www.wzw.tum.de/micbio) and can be purchased via Bruker Optics Inc.
5. FT-IR Spectroscopy Based Diagnosis of Disease States in Humans or Animals A rational behind the believe that infrared spectroscopy may be useful for disease diagnosis is that disease processes must, generally spoken, be accompanied by changes in the chemistry/biochemistry of cells, tissues, or body fluids, and that vibrational spectroscopy is suited as a diagnostic technique for sensitive detection of disease specific changes. It has been anticipated that these changes should be detectable also before morphological and systemic manifestation allow clinical diagnosis by current traditional methods. Given the fact that sample preparation and measurement are generally very simple and collection times in the range of seconds or minutes, FT-IR spectroscopy should be an ideal modality to establish rapid non-subjective and costeffective tools for early diagnosis of disease processes in biological individuals. As infrared spectroscopy can be used to identify microorganisms using their whole cell infrared spectra, it is also applicable to human cells, tissues and biofluids. An overview on the most recent developments can be found in a series of excellent reviews (see [75, 76] and the literature cited therein). They cover a wide range of applications as: (i) the analysis of various biofluids like synovial fluids (to diagnose arthritic
335
disorders in joints), amniotic fluids (to detect abnormalities in fetal development), blood and serum (for clinical chemistry based diagnosis and disease pattern analysis), (ii) the characterization of various disease states in tissues, such as a particular type of cancer (skin, breast or colon cancer), abnormalities within the central nervous system (transmissible spongiform encephalopathies, multiple sclerosis, Alzheimer disease), or (iii) the analysis of cells of exfoliated cells and cells of the immune system. In the following, some examples will be given that outline the possibilities and the progress achieved to date. 5.1 Diagnosis of Disorders from Body Fluids, FT-IR Clinical Chemistry Biological fluids (blood, serum, synovial fluid, amniotic fluid) are routinely analysed in clinical chemistry to obtain information on pathological processes because biochemical changes associated with different diseases influence the composition of body fluids and since these fluids can easily be isolated from the body. For many pathological conditions, however, there is no single analyte which alone is predictive for the presence or the stage of a disease. Instead, complex changes in the composition of the fluids, involving several chemical species, are the more general case. Diagnostic approaches that are able to detect several or even all biochemical species at the same time are infrared and Raman spectroscopy. When analyzing FT-IR spectra of biofluids, one possibility is to extract quantitative sample information, typically the concentration of analytes such as cholesterol, glucose, urea, or albumin in human serum, or of glucose, lactate and lipids in amniotic fluid [19, 77-80]. This type of spectral analysis needs a minimal understanding of the vibrational spectra. Quantitative analysis of spectra from bodily fluids can be achieved e.g. by multivariate analysis techniques such as partial least squares regression (PLSR). For analysis, the spectral data set has to be split into a training and an independent validation set for PLSR modeling, and a data set for external blinded validation of the PLSR model. Many publications appeared in the literature on the quantification of glucose, triglycerides, total protein, cholesterol and high density lipoprotein (HDL), low density lipoprotein (LDL), urea and uric acid in vitro (see e.g. Shaw et al. and literature cited therein [81]). Another possibility relies on more qualitative data analysis concepts. Here, the infrared spectra are taken as fingerprints which can be analyzed and classified by pattern recognition techniques. This approach assumes, that characteristic disease related compositional and/or structural changes in the body fluids cause a multiplicity of reproducible spectral alterations. It is suggested that the sum of these changes constitutes another spectral fingerprint which is characteristic of the disease. The classification of disease-related spectral pattern has also been called "diagnostic pattern recognition" (DPR) [82]. DPR studies should therefore cover samples from specific "diseased" and "healthy" individuals, or preferentially the complete differential diagnoses of the disease under study. This analysis concept has been successfully applied in the last decade to establish diagnostic mid-IR based blood, or serum assays for the diagnosis of diabetes mellitus, metabolic syndrome, and rheumatoid arthritis and transmissible spongiform encephalopathies (TSEs) [27, 82-93].
336
Figure 10. FT-IR cell and experimental design used for measuring liquid serum samples. (A) Picture of the pressure-stable cell from Microbiolytics GmbH, Freiburg, Germany. The FT-IR cell had a pathlength of approx. 7 m, the optical material was CaF2.. (B) Schematic diagram of the apparatus used for measurement of the serum samples. Top: Injection port in fill position and stop-flow valve in flow position. Bottom, left: Injection port in injection position, sample is injected by the HPLC-pump into the flow-through cell. Bottom, right: Stop-flow valve in stop position, sample is measured.
Here the example of a study on serum from bovine spongiform encephalopathy (BSE) infected cattle is given. 5.1.1. A New Technique for Measurement of Liquid Serum Samples Serum samples represent rather concentrated solutions of biomolecules in water (approximately 60-80 mg/ml with clotting factors removed), which are well suited for FT-IR transmission spectroscopy. Thus, a technique, which allows carrying out absorbance/transmission spectroscopy of serum in aqueous environment was designed and tested [14].
337
The device used to study different liquid sample (see Fig. 10) combines a commercially available HPLC pump, a manual injector with a 7-port sample injection valve a cheminert valve model, and a pressure stable flow-through cell. In front of the FT-IR cell a bulk-head filter with a replaceable 2 m screen is mounted, which allows to remove possible residual particulate matter from the samples and, thus, to minimize the danger of damage of the hardware. The FT-IR cell (see Fig. 10A) is a specially designed microfabricated flow-through cell from Microbiolytics Inc, Germany with a spacer from a sealant material of defined elasticity, which ensures high-pressure stability and fast relaxation for a constant effective sample thickness [13, 16]. Valve switching, triggering of sample injection, and control of HPLC pump were synchronized by means of a microcomputer. To start a measurement (see Fig. 10B), rinsing of the flow cell with distilled water is stopped, and a background spectrum of the cell filled with water is recorded. At the same time, the sample loop of the injection port is filled manually with the sample (serum). To bring the sample solution into the flow, the injection port is switched manually into the injection position. After a defined time, a trigger signal is send to the stop/flow valve to stop the flow through the IR cell after proper filling of the IR cell with sample (Fig. 10B). After discharge of pressure, the spectrum of the sample solution is measured. During the measurement, the sample injection port can be switched back to its initial position and refilled for the next measurement. In order to rinse the system with water, the stop/flow valve is switched back to its initial position after the measurement. Overlaid mid-IR spectra from 3 independent measurements of a serum sample measured with the device of Fig. 10 are shown in Fig. 11A. These FT-IR spectra are dominated by the amide I and II absorption features of the protein bands between 1500 and 1700 cm-1, with spectral features of other molecular species superimposed on the protein absorption profile (for detailed assignments of protein IR bands see [9, 14]). The range 1000-1200 cm-1 contains infrared bands associated with stretching vibrations of carbon-carbon and carbon-oxygen single bonds, such as present in sugars or carbohydrates. Lipids also give rise to a number of absorptions in IR spectra. The most intense of these absorptions are found in the range 2700-3100 cm -1, attributed to asymmetric and symmetric stretching vibrations of CH3 and CH2 groups of the acyl chains. The latter region also contains the corresponding stretching vibrations from amino acid side chains of the proteins. Obviously, the liquid technique allowed to measure highly reproducible IR spectra of serum protein solutions. The negative features around 1700 cm-1 above the amide I region of the serum spectra indicate a slight overcompensation of the water absorption bands. This is due to the identical pathlength of the IR cell used to obtain the spectra of water and serum, which prevents a perfect compensation of the water absorption bands. To get quantitative numbers for the reproducibility, the distance values ("Dvalues", see also chapter 2.3 [18, 26, 34, 36]) between pairs of spectra obtained from series of independent measurements of a given serum were calculated. The D-values for selected spectral regions of the IR serum spectra are shown in Table 2, which illustrate that the reproducibility varies as a function of the spectral region selected. The analysis of the amide I region reveals very low D-values of 0.3 (i.e., almost identical spectra) for the spectra of repetitively measured samples of the same serum specimen.
338
Table 2. Reproducibility of data acquisition of the liquid samples technique Spectral window (cm-1) 1600-1700 10501500 2800-3100 Liquid samples (D-values)* 0.33 0.09 1.31 0.16 10.00 2.14
*D = (1-) x 1000; = Person's product-momentum correlation coefficient
5.1.2. Spectral Diagnostic Pattern Analysis of Bovine Serum from Cattle Infected with BSE Very briefly, BSE is a disease belonging to the group of transmissible spongiform encephalopathies (TSEs). TSEs such as Scrapie in sheep, bovine spongiform encephalopathy (BSE) in cattle, chronic wasting disease (CWD) in deer and elk, or the new variant of Creutzfeldt-Jakob disease (vCJD) in humans belong to a group of fatal neurodegenerative disorders which are caused by a misfolded and aggregated isoform of the prion protein accumulating in the central nervous system (CNS) [94-96]. With the device of Fig. 10 infrared spectra of liquid bovine serum samples from a cohort of bovine spongiform encephalopathy (BSE) positive and BSE-negative control cattle were measured and the infrared data analyzed by ANN methodologies in order to differentiate between samples originating from BSE-positive and BSE-negative cattle. The study design and sample division for ANN analysis is summarized by Table 3. Sera from 115 BSE-positive cattle and 109 from negative control animals were investigated. The BSE-positive cattle were in the clinical stage of the disease and BSEdisease was post mortem approved by histopathological examination. Spectra were converted to second derivatives, vector normalized and specific spectral regions and combination of regions used for analysis by artificial neural nets. Second derivatives were calculated to compensate for small baseline drifts in the spectra and for minor differences in the solute-to-water ratio of the samples. Mean IR spectra (second derivatives) obtained from the liquid sera of 115 BSE-positive animals and of 108 BSE-negative controls, respectively, are shown in Fig. 11B (top spectra). Only small spectral differences were observed, most pronounced in the amide I region (see difference spectrum at the bottom of Fig. 11B [14]). Beside the amide I/II regions, minor spectral differences were also observed between 1000 and 1500 cm -1. The small spectral differences between the mean IR spectra of sera of BSE-positive and BSEnegative in the Amide I region are certainly not directly associated with the presence of the prion protein, since the sensitivity of FT-IR for detection of the misfolded prion aggregates eventually present in serum is far too low. Rather it seems that a plurality of serum constituents is slightly changed as suggested by the complex spectral difference pattern over the whole spectral range (see in Fig. 11B).
339
Table 3. Design of the serum study, number of samples and data pre-treatment for ANN analysis Cattle were in the clinical stage, BSE was post mortem approved by histopathological examination Total No. of BSE-negative samples No. of BSE-positive samples 109 115 Teaching 77 76 int. Validation 11 13 ext. Validation 21 26
Data pre-treatment for ANN-analysis: Spectra converted to second derivatives, vector normalized (28202985 cm-1); data reduction by averaging three adjacent data points.
A
Absorbance
0.12 0.08 0.04 0.00 1900 1800 1700 1600 1500 1400 Wavenumber / cm-1 1300 1200
A x 1 A x 5
BSE-negative minus BSE- positive
1639
1655
3100
2900
1600 1400 2700 1800 Wavenumber / cm-1
1200
1000
Figure 11. FT-IR spectra obtained with the device of Fig. 10. (A), top panel: Overlaid spectra from 3 independent "shots" obtained with a liquid serum sample. (B), top spectra: Two overlaid mean FT-IR spectra (second derivatives) obtained from 115 BSE-positive and 108 BSE-negative serum samples, respectively. Each serum was measured in triplicate. (B), bottom spectra: Difference spectra (BSE-positive minus BSEnegative) of the mean spectra shown on top. For the bottom difference spectra the scale was expanded by a factor of 1 and 5, respectively.
340
Table 4. Sensitivity and specificity values from external validation Liquid technique Without amide I SENS % ANN average 92.3 SPEC % 84.1 With amide I SENS % 89.7 SPEC % 92.4
SENS: correctly classified BSE-positive samples SPEC: correctly classified BSE-negative samples
To analyse the complex pattern of differential features over the whole spectral range, ANN analysis together with spectral feature extraction methodologies was used to differentiate between samples originating from BSE-positive and BSE-negative cattle. Extraction of spectral features optimal for discrimination between the samples was absolutely demandatory to achieve satisfactory results. Clinical information available for each sample spectrum was used to teach the classifier in order to predict the class identity of unknown spectra afterwards. The quality of the teaching process was assessed by reclassifying the internal validation subset of spectra. Finally the trained and optimized ANN classifier was challenged by an independent data set used for external validation. This subset of spectra ("unknown spectra") was kept totally separate from the teaching and internal validation procedure (for details of ANN analysis, including data pre-processing and spectral feature selection, see [14]). As can be seen from Table 3, the external validation data set consisted of spectra from 26 BSE-positive and 21 BSE negative animals. The corresponding sensitivity and the specificity values are given in Table 4. Obviously more than 90% of the BSEpositive samples and approx. 90% of the BSE-negative samples could be identified correctly (see Table 4). It is interesting to note that the best sensitivity was obtained by analyzing the spectral information contained in the wavenumber range between 900 and 1750 cm-1, in combination with the CH stretching region (2800 and 3100 cm-1). These results correlated well with data obtained by analysis of much larger data sets of infrared spectra collected from dried film samples of bovine sera from BSE infected cattle published recently by two different groups [27, 91]. 5.2. Characterization of Single Eukaryotic Cells, IR-Cytology FT-IR spectral variations in single eukaryotic cells within the cell cycle have been investigated in depth by M. Diem and co-workers [28-31, 105]. The same group also published many stimulating FT-IR microspectroscopic data on cultured and exfoliated human cells including cancer, oral mucosa, canine, cervical, urothelial, giant sarcoma and fibroblast cells and detailed descriptions of technical details such as cultivation and preparation for FT-IR measurement, data acquisition and data evaluation [28, 100-102]. Several papers have been published on FT-IR spectroscopic results on single normal and cancerous cells using single detector or FPA-equipped FT-IR microscopes [103105]. Also FT-IR measurements at infrared synchrotron sources have been published [106-108]. Another interesting paper on the use of FT-IR microspectroscopy in cytology is the FT-IR microscopic imaging of the differentiation of individual human mesenchymal stem cells [109]. A recently published review nicely verifies that FT-IR spectroscopy is getting to become an extremely valuable tool in cytology [28]. Here an
341
example will be given on the subcellular mapping of Scrapie infected single neuronal cells by FT-IR microspectroscopy at an infrared synchrotron source. 5.2.1. Infrared Mapping of Scrapie Diseased Single Neuronal Cells Using a Synchrotron Source An interesting example of FT-IR microspectroscopic investigations on diseased single cells is the in situ characterization of misfolded prion protein in single Scrapie-infected neuronal cells of hamster nervous tissue. In this particular case, the brilliant infrared source of a synchrotron storage ring was used to achieve sufficiently high spatial resolution near the diffraction limit [110, 111]. Scrapie is a fatal neurodegenerative disease in sheep and goats which belongs to the family of prion diseases. This disease shows the specific feature of amyloid deposition of the pathological prion protein (PrPSc) accumulating as -sheet rich aggregates in certain regions of the central and peripheral nervous system [96]. These pathological protein deposits can be visualized by immunostaining of the prion protein in nervous tissue, and also by FT-IR imaging via characteristic changes in the secondary structure sensitive amide I band [110, 111]. It has been argued that the concentration of prion protein in neuronal cells is too low and the size of the aggregate structure too small to be detectable by FT-IR spectroscopy. The amyloid aggregates should, however, be detectable when PrPSc is preferentially deposited in certain regions of the cell and the lateral resolution of the FT-IR microspectroscopic set-up is high enough to focus into these small areas. Indeed, spectral changes in the amide I region could not be observed in Scrapie-infected tissue using a standard mid-infrared light source and applying a lateral resolution of only 50 m or larger [112, 113]. Therefore, the synchrotron light source of the NSLS beamline U10B at Brookhaven, USA was used to investigate nervous tissue sections from hamsters infected with the Scrapie strain 263K in comparison to control samples from healthy hamsters. For these experiments, dorsal root ganglia were prepared and cryo-sectioned for microspectroscopic analysis. Two adjacent sections, one stained with the prion-specific antibody mAb 3F4 and the other unstained, were mounted onto CaF2 optical substrates for histopathological inspection and FT-IR microspectrometry, respectively. Single neuronal cells showing PrPSc positive staining in the adjacent section as well as single neurons of uninfected tissue were then tested by FT-IR microspectroscopic mapping. As an example, Fig. 12A, right panel, top shows micrographs of neurons from a stained and unstained infected neuronal cell. The latter was mapped at a lateral resolution of approximately 8 m (at 1600 cm-1) using a rectangular aperture of 10x10 m and applying a step-width of 4 m for testing the sample in a grid-like manner. To evaluate possible secondary structure changes, spectra obtained by these experiments were first subjected to hierarchical cluster analysis using the structure sensitive amide I band between 1600 and 1700 cm-1 as input for data analysis. Fig. 12A, left panel shows such a dendrogram, which suggests the presence of two main large clusters. From this dendrogram the lateral distribution of these spectra could be reconstructed as cluster analysis maps. The map reconstructed from cluster C1 (see Fig. 12A, right panel, bottom) clearly suggests that these pixel spectra are located in a narrow region close to the cell membrane. The same panel also shows another map re-assembled using the frequency shift of the amide I band. Obviously, the spectra showing distinct differences by cluster analysis co-localize with those characterized by a systematic low wavenumber shift of the amide I band.
342
Figure 12. FT-IR microspectroscopic mapping of a single neuronal cell at an infrared synchrotron radiation source. (A), right panel: The upper row shows micrographs of the stained and unstained neuron (thin sections). The mapped area is marked by a rectangle. Immunohistochemical staining was performed as described in [110]. (A), left panel: Dendrogram of a cluster analysis performed on the mapping data set. The spectral information between 1700 and 1600 cm-1 (protein amide I band) was used as input for cluster analysis (original spectra). The dendrogram displays two major classes C1 and C2. From the pixel spectra in class C1 the cluster analysis map shown in the bottom row was obtained. These spectra co-localize with those which exhibit a distinct low-wavenumber shift of the amide I band as can be seen from the map on the right side. (B), right panel: Representative pixel spectra (second derivatives) in the different regions of the neuron. The maps on the left side panel were constructed plotting the intensity ratio of the amide I band component at 1637 cm-1 (stands for -sheet) and at 1657 cm-1 (stands for -helix) as a function of pixel location. Spectra found in cluster C1 of Fig. 12A co-localize with those showing dramatic changes in the amide I band contour.
343
Fig. 12B shows FT-IR maps constructed on the basis of the second derivative spectra using the intensity ratio of the peaks at 1656 cm-1 (stands for -helix) and at 1637 cm-1 (stands for -sheet). Indeed, those spectra located in a narrow region near the cell membrane exhibited distinct spectral differences that suggested a higher -sheet content while the -helix band was shifted to lower values and decreased in intensity at the same time. These experiments proved for the first time that pathological secondary structure changes related to the -helix / -sheet transition of the prion protein could be observed in situ at a sub-cellular level using FT-IR microspectrometry. This work was later extended to more animal samples testing animals at different stages of the disease development [111-113]. The importance of lateral resolution in these experiments was also indicated, since spectral changes characteristic of protein structure changes could only be detected when a lateral resolution better than 10 m was used, which was very likely a consequence of local PrPSc concentration, size of the PrPSc aggregates, and the detection limits of FT-IR microspectroscopy. For example, when the size of the aperture was computationally increased to an aperture size of 30x30 m2 by pixel binning, the PrPSc spectral features at 1637 cm-1 were optically diluted and became undetectable. Thus, the possibility of characterizing the secondary structure of small protein aggregates in a complex environment strongly depends on the spatial resolution and on whether these aggregates assemble in certain areas of the cells or not. 5.3. Characterization of Tissues, FT-IR-Histopathology FT-IR microspectrometry is particularly useful for the examination of small particles in a complex and heterogeneous biological environment (e.g. crystal deposits in tissues, silicone gel in human breast tissue, white deposits in human kidney and calcium oxalate in bladder tissues) [114, 115]. However, FT-IR microspectroscopic mapping over a tissue section to attain segmentation of histological structures is another option. FT-IR imaging using infrared microscopes equipped with single detection elements or focal plane array detector (FPA) systems has been successfully applied to various tissues [116, 117] and a considerable number of applications have been published in the last 10 years which address the problem of how to achieve segmentation of histological and particularly diseased structures in tissues by pattern recognition methodologies [32, 33, 118-125]. 5.3.1. Characterization of Histological Structures by Chemical Mapping Infrared microspectrometry can be used to map the relative absorbances of particular functional groups across a sample. In this way chemical information is obtained about the molecular structure for which the functional group stands. That is why this approach was given the name "chemical mapping" or "chemical imaging". Ratios of absorbances from pairs of functional groups can also be used for mapping experiments (e.g. from amide I-band of the protein and lipid >CH2 stretching bands). This is particularly useful since one of the bands is treated as an internal standard in order to account for unavoidable variations in film thickness across the tissue section.
344
Figure 13. FT-IR microspectroscopic mapping of a hamster brain tissue thin section. (A) Cartoon of a section through a brain sample (left panel) and a micrograph of the stained tissue sample after magnification. The tissue slice was stained with "methylene blue". The histological structures are as follows: (1) stratum moleculare (dendrites of Purkinje cells, basket cells, stellate cells, astrocytes; (2) stratum ganglionare (Purkinje cells, Bergmann's cells); (3) stratum granulosum (granular cells, Golgi epithelial cells, astrocytes, oligodendrocytes); (4) substantia alba (axons of Purkinje cells, oligodendrocytes). (B) FT-IR image reconstructed by the so-called functional group or chemical mapping approach. The ratio of absorbance measured at 1655 cm-1 (amide I, stands for protein) and at 2921 cm-1 (asymmetric stretching of >CH2, stands for lipids) was used to obtain segmentation of tissue structures. The number of spectra was approx. 900 using an aperture of 100 m and a step width of 50 m. (C) Typical FT-IR spectra obtained from white and grey matter, respectively.
345
Fig. 13A gives an example on a cerebellar thin section obtained from a piece of Syrian hamster brain. In this case, micro-cryotomed tissue slices were mounted as 8 m thin slices on CaF2 microscopic slides (1 mm thickness). No fixation or embedding or staining procedures were applied. The standard histopathological way of testing such a sample is to apply various staining techniques to visualize the histological structures. Thus, after FT-IR microscopic analysis, these thin sections were stained with methylene blue and analysed light microscopically for histopathological examination. Fig. 13, right shows such a cryosection after methylene blue staining, which enables one to visualize the various cerebellar structures such as Stratum moleculare, Stratum ganglionare, Stratum granulosum (grey matter), and white matter of the cerebellum. By simply plotting the ratio of absorbance of the asymmetric >CH2 stretching band at 2921 cm-1 and the amide band at 1655 cm-1, which stand for lipidrich and protein-rich structures in the tissue, respectively, as a function of lateral location, many of the histological structures can already be reconstructed in falsecolours as is exampled in Fig. 13B. The single detector FT-IR microscope used for this experiment was equipped with a computer controlled x,y-stage, 1500 1500 m2 rectangular regions of the samples were mapped with 50 m steps in x and y direction using a 100 m aperture and a nominal resolution of 6 cm-1 with 50 scans averaged per pixel spectrum. The first step of data evaluation was a "quality test" of raw data. This test included check for water vapour bands, search for spectra which were too low or too high in absorbance (as an indication of preparation artefacts like micro-rips due to microtoming and/or drying of the sample). All spectra which had passed this filter were subsequently baseline corrected and finally the absorbance ratios calculated. 5.3.2. Segmentation of Histological Structures by Multivariate Imaging The functional group or chemical mapping approach works nicely to get a quick overview of histological structures in a tissue slice without the necessity to stain the tissue sample. The lateral resolution and identification of diseased tissue structures in a complex environment of different other histological structures (which in itself are composed of hundreds of biological macromolecules) is, however, a more complicated problem. Indeed, the investigation of various tissue samples by the chemical mapping approach did not lead to satisfactory results, and a precise segmentation of histological structures and particularly the differentiation between healthy and malignant regions of the tissues from different patients generally failed. To improve discrimination capacity and to enhance image contrast, multivariate data analysis techniques turned out to be more efficient [32, 33, 118-126]. Starting from the postulation that disease processes in tissues or cells will induce changes in intrinsic biochemical composition and structure, any disease state should produce an unique infrared spectroscopic pattern. However, these signatures are very complex and generally superimposed by spectral features of the various different molecular structures present. Thus, not a single spectral property (e.g. a functional group frequency or absorbance value), but rather a multiplicity of defined spectral features or traits will be necessary to obtain sufficient information for the characterization of what is called a "normal" or "diseased" tissue structure in a given sample. To this end, the necessary amount of IR spectral traits that are diagnostic for
346
Figure 14. FT-IR image reconstruction using multivariate clustering as described in [126]. For FT-IR microspectroscopic mapping, the same spectra of Fig. 12 have been used. (A), left panel: PCA plot of the spectra using the factor loadings F2 and F3 for projection of data. (A), right panel: Dendrogram of a hierarchical cluster analysis with spectral distances calculated in factor space and using the first three factor loadings. Nine clusters can be defined which contain the spectra corresponding to the histological structures explained in (B). (B), left panel: PCA-cluster analysis based FT-IR maps (1-9), each reconstructed from the pixel spectra of the 9 clusters in Fig. 13A. (B), right panel: FT-IR image obtained by superimposing the nine single maps of Fig. 13B, left panel. The nine cluster images correspond to the following histological structures: 1 to stratum ganglionare; 2 and 3 to stratum moleculare; 4, 5 and 6 to stratum granulosum; 7, 8 and 9 to substantia alba.
347
the diseased state(s) should be extractable by pattern recognition and optimal spectral selection techniques. Several approaches to this problem have been published in the literature [32, 33, 118-126]. One possibility is to use cluster analysis methodologies to achieve objective segmentation of histological structures. To make this approach intelligible, an example on the characterization of cerebellar thin sections from Syrian hamster brain samples by multivariate imaging using hierarchical clustering is given in Fig. 14 [126]. Here, the same cryotomed tissue specimen shown in Fig. 13 was used and the spectral data subjected to PCA and PCAbased hierarchical clustering (see PCA map and subsequent cluster analysis in Fig. 14A). Since each spectrum in the clusters 1 to 9 of the dendrogram in Fig. 14A is a pixel spectrum with known spatial coordinates and belongs to a specific histological structure, segmentation of histological structures could nicely be achieved without any tissue staining by simply assigning false colours to these pixel spectra. Fig. 14B, left panel shows single images based on the various different clusters of the dendrogram of Fig. 14A. Interestingly, the FT-IR image is now based on the spatial distribution of spectral patterns, each of which is being characteristic of a particular tissue structure. Fig. 14B, right panel presents an image obtained by simply superimposing all 9 single images. The results available from the literature convincingly demonstrate that FT-IR microspectrometry may have great potential for histological and histopathological analysis in the future. Obviously, the information contained in mid-infrared spectra of histological samples such as tissue slices is sufficient to distinguish between various tissue structures and pathologies. However, for reliable tissue segmentation pattern recognition methodologies are definitely required to effectively take advantage of the huge amount of spectral information available. 5.3.3. Objective Identification of Histological Structures Based on Spectral Reference Data Bases and Artificial Neural Networks Multivariate techniques such as PCA and HCA can be used to achieve segmentation of histological structures as is exampled by the false colour images in Fig. 14B. Ultimately, however, the assignment of the spectra to the various histological structures is only achieved by comparing the FT-IR image with the same or adjacent tissue slice stained by conventional techniques and the help of a trained histopathologist. Thus, to identify histological structures objectively in tissue samples of unknown patient material, image reconstruction on the basis of representative reference spectra from ideally all relevant tissue structures is needed. The following is a brief description of the experimental and evaluation strategy on the example of colon tissue and colon adenocarcinoma recently published as the proof of principle of this approach [122, 125]: Spectra were collected from cryo-sectioned 8m tissue slices mounted onto standard sized microscope slides made of CaF2 as the optical substrate. Data acquisition was performed with a PerkinElmer spectrum spotlight 300-system equipped with a linear MCT-array (16 x 1 detector elements) at a spatial resolution of about 10m, a step size of 6.25 m, a spectral resolution of 8 cm -1 and a data point spacing of two data points per wavenumber. A test data base of colon reference spectra was created by selecting 4120 spectra from more than 1,5 million pixel spectra obtained from 28 individual patient samples. The spectra were then assigned to individual histological structures with the help of a
348
Table 5. Spectral data base of colon patient samples used to establish a reference data base for objective identification of histological structures in tissues Tissue structure Adenocarcinoma Fibrovascular stock Mucin Fat Crypts Lymphocytes Lamina muscularis mucosae Tunica muscularis Necrosis Lamina propria mucosae Submucosa Vessel (blood, lymph) # of spectra selected 1805 878 385 26 153 41 170 138 169 128 34 193
trained histopathologist (see Table 5). These spectra were first selected from individual patient samples and then compiled in a reference data base for subsequent objective identification of histological structures in tissue slices of unknown patient samples. The spectral reference data were used to train modular ANNs for objective identification of tissue structures in general and diseased structures in particular as has recently been shown by Peter Lasch and co-workers [125]. Fig. 15A shows the micrograph of the H&E stained cryosection from an unknown patient sample. Fig. 15B gives an FT-IR image using a modular ANN classifier for tissue structure identification as briefly outlined above (for details, see [125]). Fig. 15B suggests that the main histological structures could be objectively identified using the reference spectra of Table 5 together with an optimized modular ANN classifier. Obviously, the predictions obtained correlated quite well with histopathology and, interestingly, also the regions of the adenocarcinoma could be identified. From these results it was concluded [125], that FT-IR imaging of histological specimen in general and patient samples in particular could be an interesting alternative or compliment to the more subjective traditional ways used in histopathology nowadays. To this end, however, comprehensive reference data libraries of tissue structures and optimized classification models have to be established by expert groups that allow objective segmentation of histological structures and prediction of disease processes in patient samples.
349
Figure 15. Objective segmentation of histological structures based on a validated spectral reference data base. Image reconstruction for a patient sample B2847/97 with adenocarcinoma of the rectum G1 (grading TNM 3-3-0). (A) H&E stained cryo-section (staining performed after FT-IR imaging). (B) Image reconstructed using an optimized molecular ANN (see text and [125]). The image was re-assembled using an ANN with 12 classes as described in [125]. Assignment of false colours to histological structures: Blue = tumor; green yellow = muscularis mucosae; salmon = tunica muscularis; light yellow = necrosis; deep pink = lymphocytes; aqua = crypts; corn flower blue = propria; red = fibrovascular tissue; beige = vessel. Pixel spectra for mucin, fat, and submucosa are not present in this tissue sample.
6. Conclusions 6.1. Characterization of Microorganisms The main advantages of FT-IR spectroscopy constituting its attractiveness, are extreme rapidity compared to conventional techniques, uniform applicability to very diverse microorganisms, and a high specificity that allows differentiations even down to subspecies level. Drawing upon the experience obtained to date, the serial type of a dedicated instrument for FT-IR characterizations of microorganisms and other biological samples is already available from an FT-IR spectrometer producing company. The combination of IR focal plane array detectors and micro-array printing technologies may contribute to render microbiological FT-IR analysis a rapid, costeffective and unprecedented high-throughput technology for microbiological analyses. This technology may not only help to scale down the number of cells needed for analysis, to investigate mixed cultures, and to perform population analyses, but also to detect light microscopic and spectroscopic features simultaneously, with the prospect of a fully automated IR microscopic system combining detection, enumeration, and identification of microorganisms in one single instrument. 6.2. Diagnostic Pattern Recognition from Biofluids The results obtained so far on various different biofluids demonstrate that mid-infrared spectroscopy in conjunction with sophisticated multivariate data analysis could become a novel practical application in clinical laboratory analysis. This approach will benefit from its regent-free way to analyse the samples, its inherent speed and the option to obtain information in a variety of different solutes from one single spectrum. While the DPR approach may be of profit where the biochemical/chemical origin of the disease is unknown, the simultaneous quantification of several analytes from one single spectrum
350
could be an alternative for the more conventional standard tests used in clinical laboratories. 6.3. Characterization of Eukaryotic Cells and Tissues The most important step forward is certainly the coupling of FT-IR spectrometers to light microscopes to obtain spectral information from single cells and to achieve spatial resolution in tissue analysis in a way that is familiar to cytologists or histopathologists. The technological progress has been enormous and high-quality FT-IR microscopes are available on the market, which can be used to image tissues, single cells, and even analyse sub-cellular compartments. Today, bench-top instrumentation for routine FTIR imaging of diseased tissue sections is available. Equipped with focal plane array detectors for mid-infrared imaging rapid segmentation of histological structures without any tissue staining and imaging of larger cells is possible also under conditions of a routine laboratory. However, to push biomedical infrared spectroscopy forward, multi-center clinical trials focussing on selected clinical indications are needed to attract the attention of the clinicians and to establish objective sensitivity and specificity parameters under practical constraints. Yet, it remains undefined who could conduct such trials, on which relevant cancer types or clinical samples or whether fresh patient biopsies or archive materials, and which technological platforms should be used.
Abbreviations ANN A/T ATR BSE FPA FT FT-IR HCA IR PCA PHB PHF artificial neural network absorbance/transmission attenuated total reflectance bovine spongiform encephalopathy Focal plane array Fourier transform Fourier transform-infrared Hierarchical cluster analysis infrared principal component analysis poly--hydroxybutyric acid poly--hydroxybutyric fatty acid
Acknowledgements The excellent technical assistance of Angelika Brauer in preparing this manuscript is gratefully acknowledged. The assistance of Maren Stmmler and Angelika Brauer in sample preparations, measurements, and evaluation is much appreciated. The technical drawings in Fig. 4 have been provided by Bruker Optics Inc, Ettlingen, Germany. This contribution is mainly based on the published work obtained in the author's group at the Robert Koch-Institute and are cited in the original literature. The authors would like to
351
particularly acknowledge the contributions from Dieter Helm, Janina Kneipp, and Ariane Kretlow.
References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] B.A. Kenner, J.W. Riddle, S.W. Rockwood, R.H. Bordner, J. Bacteriol. 75 (1958), 16-20. J.W. Riddle, P.W. Kabler, B.A. Kenner, R.H. Bordner, S.W. Rockwood, H.J.R. Stevenson, J. Bacteriol. 72 (1956), 593-603. R. Mendelsohn, C.R. Flach, D.J. Moore, Biochim. Biophys. Acta 1758 (2006), 923-933. G. Zhang, K.L.A. Chan, C.R. Flach, R. Mendelsohn, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 357-378. H. Ou Yang, E.P. Paschalis, A.L. Mayo, A.L. Boskey, R. Mendelsohn, J. Bone Miner. Res. 16 (2001), 893-900. X. Bi, X. Yang, M.P.G. Bostrom, D. Bartusik, S. Ramaswamy, K.W. Fishbein, R.G. Spencer, N.P. Camacho, Anal. Bioanal. Chem. 387 (2007), 1601-1612. M. Kim, X. Bi, W.E. Horton, R.G. Spencer, N.P. Camacho, J. Biomed. Opt. 10(3) (2005), 031105. X. Bi, G. Li, S.B. Doty, N.P. Camacho, Osteoarth. Cartil. 13 (2005), 1050-1058. H. Fabian, W. Mntele, Handbook of Vibrational Spectroscopy Vol 5, J.M. Chalmers, P.R. Griffiths, Editors, John Wiley & Sons Ltd, (2002), 3399-3425. R.N.A.H. Lewis, R.N. McElhaney, Handbook of Vibrational Spectroscopy Vol 5, J.M. Chalmers, P.R. Griffiths, Editors, John Wiley & Sons Ltd, (2002), 3447-3464. E. Taillandier, J. Liquier, Handbook of Vibrational Spectroscopy Vol 5, J.M. Chalmers, P.R. Griffiths, Editors, John Wiley & Sons Ltd, (2002), 3465-34-80. K. Brandenburg, U. Seydel, Handbook of Vibrational Spectroscopy Vol 5, J.M. Chalmers, P.R. Griffiths, Editors, John Wiley & Sons Ltd, (2002), 3481-3507. F. Sokolowski, A.J. Modler, R. Masuch, D. Zirwer, M. Baier, G. Lutsch, D.A. Moss, K. Gast, D. Naumann, J. Biol. Chem. 278 (2003), 40481-40492. H. Fabian, P. Lasch, D. Naumann, J. Biomed. Opt. 10(3) (2005), 03110301-03110310. D. Naumann, P. Lasch, H. Fabian, Proc. of SPIE Vol. 6093 (2006), 609301-1-609301-12. R. Masuch, D.A. Moss, Appl. Spectrosc. 57 (2003), 1407-1418. H.M. Heise, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 9-37. D. Naumann, Encyclopedia of Analytical Chemistry, R.A. Meyers, Editor, John Wiley & Sons, Ltd. (2000), 102-131. G. Hosafi, O. Klein, G. Oremek, W. Mntele, Anal. Bioanal. Chem. 387 (2007), 1815-1822. G. Mazarevica, J. Diewok, J.R.Baena, E. Rosenberg, B. Lendl, Appl. Spectrosc. 58 (2004), 804-810. D.L. Wetzel, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 39-78. D.L. Wetzel, S.M. LeVine, Science 285 (1999), 1224-1225. J.A. Reffner, Cell. Mol. Biol. 44 (1998), 1-9. J.A. Reffner, Am. Lab. 9 (2000), 36-40. E.N. Lewis, P.J. Treado, R.C. Reeder, G.M. Story, A.E. Dowrey, C. Marcott, I.W. Levin, Anal. Chem. 67 (1995), 3377. D. Helm, H. Labischinski, D. Naumann, J. Microbiol. Meth. 14 (1991), 127-142. P. Lasch, J. Schmitt, M. Beekes, Th. Udelhoven, M. Eiden, H. Fabian, W. Petrich, D. Naumann, Anal. Chem. 75 (2003), 6673-6678. M.J. Romeo, S. Boydston-White, C. Matthus, M. Miljkovic, B. Bird, T. Chernenko, P. Lasch, M. Diem, Vibrational Spectroscopy for Medical Diagnosis, M. Diem, P.R. Griffiths, J.M. Chalmers, Editors, John Wiley & Sons, Ltd (2008), 27-70. M. Diem, M.J. Romeo, S. Boydston-White, M. Miljkovic, C. Matthus, Analyst 129 (2004), 880-885. M.J. Romeo, B. Mohlenhoff, M. Jennings, M. Diem, Biochim. Biophys. Acta 1758 (2006), 915-922. S. Boydston-White, M.J. Romeo, T. Chernenko, A. Regina, M. Miljkovic, M. Diem, Biochim. Biophys. Acta 1758 (2006) 908-914. R. Bhargava, S.M. Hewitt, I.W. Levin, Nat. Biotechnol. 25 (2007), 31-33. R. Bhargava, I.W. Levin, Vibrational Spectroscopy for Medical Diagnosis, M. Diem, P.R. Griffiths, J.M. Chalmers, Editors, John Wiley & Sons, Ltd (2008), 155-185. D. Naumann, H. Labischinski, P. Giesbrecht, Modern Techniques for Rapid Microbiological Analysis, W.H. Nelson, Editor, VCH-Publisher (1991), 43-96. D. Naumann, Proc. of SPIE Vol. 6853 (2008), 68530G-1-68530G-12.
[29] [30] [31] [32] [33] [34] [35]
352
[36] D. Helm, H. Labischinski, G. Schallehn, D. Naumann, J. Gen. Microbiol. 137 (1991), 69-79. [37] D. Naumann, D. Helm, H. Labischinski, Nature 351 (1991), 81-82. [38] W. Petrich, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 315-332. [39] A. Kohler, M. Hanafi, D. Bertrand, E.M. Qannari, A. Oust Janbu, T. Mretr, K. Naterstad, H. Martens, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 333-356. [40] A. Oust, B. Moen, H. Martens, K. Rudi, T. Naes, C. Kirschner, A. Kohler, J. Microbiol. Meth. 65 (2006), 573-584. [41] B. Moen, A. Oust, O. Langsrud, N. Dorrell, G.L. Marsden, J. Hinds, A. Kohler, B.W. Wren, K. Rudi, Appl. Environ. Microbiol. 71 (2005), 2086-2094. [42] D. Helm, D. Naumann, FEMS Microbiol. Lett. 126 (1995), 75-80. [43] N.A. Ngo Thi, D. Naumann, Anal. Bioanal. Chem. 387 (2007), 1767-1777. [44] K. Maquelin, C. Kirschner, L.P. Choo-Smith, N.A. Ngo-Thi, T. van Vreeswijk, M. Stammler, H.P. Endtz, H.A. Bruining, D. Naumann, G.J. Puppels, J. Clin. Microbiol. 41 (2003), 324-329. [45] M. Wenning, V. Theilmann, S. Scherer, Environ. Microbiol. 8 (2006), 848-857. [46] B.J. Tindall, E. Brambilla, M. Steffen, R. Neumann, R. Pukall, R.M. Kroppenstedt, E. Stackebrandt, Environ. Microbiol. 2 (2000), 310-318. [47] I. Adt, D. Toubas, J.M. Pinon, M. Manfait, G.D. Sockalingum, Arch. Microbiol. 185 (2006), 277-285. [48] A. Oust, T. Mretr, K. Naterstad, G.D. Sockalingum, I. Adt, M. Manfait, A. Kohler, Appl. Environ. Microbiol. 72 (2006), 228-232. [49] N.M. Amiali, M.R. Mulvey, J. Sedman, M. Louie, A.E. Simor, A.A. Ismail, J. Microbiol. Meth. 68 (2007), 236-242. [50] S.H. Beattie, C. Holt, D. Hirst, A.G. Williams, FEMS Microbiol. Lett. 164 (1998), 201-206. [51] R. Goodacre, E.M. Timmins, R. Burton, N. Kaderbhai, A.M. Woodward, D.B. Kell, P.J. Rooney, Microbiology 144 (1998), 1157-1170. [52] H. Haag, H.U. Gremlich, R. Bergmann, J.J. Sanglier, J. Microbiol. Meth. 27 (1996), 157-163. [53] C. Kirschner, K. Maquelin, P. Pina, N.A. Ngo Thi, L.P. Choo-Smith, G.D. Sockalingum, C. Sandt, D. Ami, F. Orsini, S.M. Doglia, P. Allouch, M. Mainfait, G.J. Puppels, D. Naumann, J. Clin. Microbiol. 39 (2001), 1763-1770. [54] M. Kmmerle, S. Scherer, H. Seiler, Appl. Environ. Microbiol. 64 (1998), 2207-2214. [55] D. Toubas, M. Essendoubi, I. Adt, J.M. Pinon, M. Mainfait, G.D. Sockalingum, Anal. Bioanal. Chem. 387 (2007), 1729-1737. [56] M.A. Miguel Gomez, M.A. Bratos Perez, F.J. Martin Gil, A. Duenas Diez, J.F. Martin Rodriguez, P. Gutierrez Rodriguez, A. Orduna Domingo, A. Rodriguez Torres, J. Microbiol. Meth. 55 (2003), 121131. [57] H. Oberreuter, H. Seiler, S. Scherer, Int. J. Syst. Evol. Microbiol. 52 (2002), 91-100. [58] A. Oust, T. Mretr, C. Kirschner, J.A. Narvhus, A. Kohler, J. Microbiol. Meth. 59 (2004), 149-162. [59] C.A. Rebuffo, J. Schmitt, M. Wenning, F. von Stetten, S. Scherer, Appl. Environ. Microbiol. 72 (2006), 994-1000. [60] C.A. Rebuffo-Scheer, C. Kirschner, M. Stmmler, D. Naumann, J. Microbiol. Meth. 68 (2007), 282290. [61] C. Sandt, C. Madoulet, A. Kohler, P. Allouch, C. De Champs, M. Manfait, G.D. Sockalingum, J. Appl. Microbiol. 101 (2006), 785-797. [62] M. Wenning, H. Seiler, S. Scherer, Appl. Environ. Microbiol. 68 (2002), 4717-4721. [63] S. Lai, R. Goodacre, L.N. Manchester, Syst. Appl. Microbiol. 27 (2004), 186-191. [64] G. Fischer, S. Braun, R. Thissen, W. Dott, J. Microbiol. Meth. 64 (2006), 63-77. [65] M. Kansiz, P. Heraud, B. Wood, F. Burden, J. Beardall, D. McNaughton, Phytochemistry 52 (1999), 407-417. [66] M. Wenning, S. Scherer, D. Naumann, Vibrational Spectroscopy for Medical Diagnosis, M. Diem, P.R. Griffiths, J.M. Chalmers, Editors, John Wiley & Sons, Ltd (2008), 71-96. [67] N.A. Ngo Thi, C. Kirschner, D. Naumann, J. Mol. Struct. 661-662 (2003), 371-380. [68] J. Kirkwood, A. Ghetler, J. Sedman, D. Leclair, F. Pagotto, J.W. Austin, A.A. Ismail, J Food Prot 69 (2006), 2377-2383. [69] J. Kirkwood, S.F. Al-Khaldi, M.M. Mossoba, J. Sedman, A.A. Ismail, Appl. Spectrosc. 58 (2004), 1364-1368. [70] T. Udelhoven, D. Naumann, J. Schmitt, Appl. Spectrosc. 54 (2000), 1471-1479. [71] R. Goodacre, E.M. Timmins, P.J. Rooney, J.J. Rowland, D.B. Kell, FEMS Microbiol. Lett. 140 (1996), 233-239. [72] D.J. Mouwen, R. Capita, C. Alonso-Calleja, J. Prieto-Gomez, M. Prieto, J. Microbiol. Meth. 67 (2006), 131-140.
353
[73] K. Tintelnot, G. Haase, M. Seibold, F. Bergmann, M. Stmmler, T. Franz, D. Naumann, J. Clin. Microbiol. 38 (2000), 1599-1608. [74] R. Goodacre, B. Shann, R.J. Gilbert, E.M. Timmins, A.C. McGovern, B.K. Alsberg, D.B. Kell, N.A. Logan, Anal. Chem. 72 (2000), 119-127. [75] Vibrational Spectroscopy for Medical Diagnosis, M. Diem, P.R. Griffiths, J.M. Chalmers, Editors, John Wiley & Sons, Ltd, (2008) [76] Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Ltd, (2008). [77] R.A. Shaw, S. Kotowich, M. Leroux, H.H.Mantsch, Ann. Clin. Biochem. 35 (1998), 624-632. [78] R.A. Shaw, H.H. Eysel, K.Z. Liu, H.H. Mantsch, Anal. Biochem. 259 (1998), 181-186. [79] K.Z. Liu, R.A. Shaw, A. Man, T.C. Dembinski, H.H. Mantsch, Clin. Chem. 48 (2002), 499-506. [80] K.Z. Liu, H.H. Mantsch, Am. J. Obstet. Gynecol. 180 (1999), 696-702. [81] R.A. Shaw, S. Low-Ying, A. Man, K.Z. Liu, C. Mansfield, C.B. Rileg, M. Vijarnsorn, Biomedical Vibrational Spectroscopy, P. Lasch, J. Kneipp, Editors, John Wiley & Sons, Inc. (2008), 79-104. [82] W. Petrich, B. Dolenko, J. Frh, M. Ganz, H. Greger, S. Jacob, F. Keller, A.E. Nikulin, M. Otto, O. Quarder, R.L. Somorjai, A. Staib, G. Werner, H. Weilinger, Appl. Opt. 39 (2000), 3372-3379. [83] G. Werner, J. Frh, F. Keller, H. Greger, R. Somorjai, B. Dolenko, M. Otto, D. Bcker, Proc. of SPIE Vol. 3257 (1998), 35-41. [84] W. Petrich, B. Dolenko, D.J. Fink, J. Frh, H. Greger, S. Jacob, F. Keller, A. Nikulin, M. Otto, M.S. Pessin-Minsely, O. Quarder, R. Somorjai, A. Staib, U. Thienel, G. Werner, H. Wielinger, Proc. of SPIE Vol. 4252 (2001), 72-80. [85] W. Petrich, B. Dolenko, J. Frh, H. Greger, S. Jacob, F. Keller, A.E. Nikulin, M. Otto, O. Quarder, R.L. Somorjai, A. Staib, G. Werner, H. Wielinger, Proc. of SPIE Vol. 3918 (2000), 91-96. [86] W. Petrich, A. Staib, M. Otto, R. Somorjai, Vib. Spectrosc. 28 (2002), 117-129. [87] J. Frh, S. Jacob, B. Dolenko, H.-U. Hring, R. Mischler, O. Quarder, W. Renn, R. Somorjai, A. Staib, G. Werner, W. Petrich, Proc. of SPIE Vol. 4614 (2002), 63-69. [88] A. Staib, B. Dolenko, D.J. Fink, J. Frh, A.E. Nikulin, M. Otto, M.S. Pessin-Minsley, O. Quarder, R. Somorjai, U. Thienel, G. Werner, W. Petrich, Clin. Chim. Acta 308 (2001), 79-89. [89] J. Schmitt, M. Beekes, A. Brauer, Th. Udelhoven, P. Lasch, D. Naumann, Anal. Chem. 74 (2002), 3865-3868. [90] J. Moecks, G. Kocherscheidt, W. Koehler, W. Petrich, Proc, of SPIE Vol. 5321 (2004), 117-123. [91] T.C. Martin, J. Moecks, A. Belooussov, S. Cawthraw, B. Dolenko, M. Eiden, J. von Frese, W. Kohler, J. Schmitt, R. Somorjai, T. Udelhoven, S. Verzakov, W. Petrich, Analyst 129 (2004), 897-901. [92] D. Rohleder, G. Kocherscheidt, K. Gerber, W. Kiefer, W. Kohler, J. Mocks, W. Petrich, J. Biomed Opt. 10 (2005), 031108. [93] P. Lasch, M. Beekes, J. Schmitt, D. Naumann, Anal. Bioanal. Chem. 387 (2007), 1791-1800. [94] S.B. Prusiner, Science 278 (1997), 245-251. [95] K.M. Pan, M. Baldwin, J. Nguyen, M. Gasset, A. Serban, D. Groth, I. Mehlhorn, Z. Huang, R.J. Fletterick, F.E. Cohen, S.B. Prusiner, Proc. Natl. Acad. Sci. USA 90 (1993), 10962-10966. [96] M. Beekes, P.A. McBride, FEBS J. 274 (2007), 588-605. [97] M. Diem, S. Boydston-White, L. Chiriboga, Appl. Spectrosc. 53 (1999), 148A-161A. [98] L. Chiriboga, P. Xie, H. Yee, V. Vigorita, D. Zarou, D. Zakim, M. Diem, Biospectroscopy 4 (1998), 47-53. [99] L. Chiriboga, P. Xie, H. Yee, V. Vigorita, D. Zarou, D. Zakim, M. Diem, Biospectroscopy 4 (1998), 55-59. [100] M. Diem, M. Romeo, S. Boydston-White, M. Miljkovic, C. Matthus, Analyst 129 (2004), 880-885. [101] S. Boydston-White, M.J. Romeo, T. Chernenko, A. Regina, M. Miljkovic, M. Diem, Biochim. Biophys. Acta 1758 (2006), 908-914. [102] S. Boydston-White, T. Chernenko, A. Regina, M. Miljkovic, C. Matthus, M. Diem, Vib. Spectrosc. 38 (2005), 169-177. [103] M. Diem, L. Chiriboga, P. Lasch, A. Pacifico, Biopolymers (Biospectroscopy) 67 (2002), 349-353. [104] P. Lasch, A. Pacifico, M. Diem, Biopolymers (Biospectroscopy) 67 (2002), 335-338. [105] P. Lasch, M. Boese, A. Pacifico, M. Diem, Vib. Spectrosc. 28 (2002), 147-157. [106] N. Jamin, P. Dumas, J. Moncuit, W.H. Fridman, J.L. Teillaud, G.L. Carr, G.P. Williams, Proc. Natl. Acad. Sci. USA 95 (1998), 4837-4840. [107] N. Jamin, P. Dumas, J. Moncuit, W.H. Fridman, J.L. Teillaud, G.L. Carr, G.P. Williams, Cell. Mol. Biol. 44 (1998), 9-13. [108] H.N. Holman, M.C. Martin, E.A. Blakely, K. Bjornstad, W.R. McKinney, Biopolymers (Biospectroscopy) 57 (2000), 329-335. [109] C. Krafft, R. Salzer, S. Seitz, C. Ern, M. Schieker, Analyst 132 (2007), 647-653. [110] J. Kneipp, L.M. Miller, M. Joncic, M. Kittel, P. Lasch, M. Beekes, D. Naumann, Biochim. Biophys. Acta 1639 (2003), 152-158.
354
[111] Q. Wang, A. Kretlow, M. Beekes, D. Naumann, L.M. Miller, Vib. Spectrosc. 38 (2005), 61-69. [112] J. Kneipp, L.M. Miller, S. Spassov, F. Sokolowski, P. Lasch, M. Beekes, D. Naumann, Biopolymers 74 (2004), 163-167. [113] J. Kneipp, L.M. Miller, S. Spassov, F. Sokolowski, P. Lasch, M. Beekes, D. Naumann, Proc. of SPIE Vol. 5321 (2004), 17-25. [114] L.H. Kidder, V.F. Kalasinsky, J.L. Luke, I.W. Levin, E.N. Lewis, Nature Med. 3 (1997), 235-237. [115] E.N. Lewis, A.M. Gorbach, C. Marcott, I.W. Levin, Appl. Spectrosc. 50 (1996), 263-269. [116] S.M. LeVine, D.L.B. Wetzel, Appl. Spectrosc. Rev. 28 (1993), 385-412. [117] S.M. LeVine, D.L.B. Wetzel, A.J. Eilert, Int. J. Devl. Neuroscience 12 (1994), 275-288. [118] P. Lasch, W. Wsche, W.J. McCarthy, G. Mller, D. Naumann, Proc. of SPIE Vol. 3257 (1998), 187-198. [119] P. Lasch, D. Naumann, Cell. Mol. Biol. 44 (1998), 189-202. [120] P. Lasch, E.N. Lewis, L.H. Kidder, D. Naumann, Proc. of SPIE Vol. 3920 (2000), 129-139. [121] P. Lasch, J. Schmitt, D. Naumann, Proc. of SPIE Vol. 3918 (2000), 45-56. [122] P. Lasch, W. Haensch, E.N. Lewis, L.H. Kidder, D. Naumann, Appl. Spectrosc. 56 (2002), 1-9. [123] P. Lasch, W. Haensch, D. Naumann, M. Diem, Biochim. Biophys. Acta 1688 (2004), 176-186. [124] P. Lasch, M. Diem, D. Naumann, Proc. of SPIE Vol. 5321(2004), 1-9. [125] P. Lasch, M. Diem, W. Hnsch, D. Naumann, J. of Chemometrics 20 (2006), 209-220. [126] J. Kneipp, P. Lasch, E. Baldauf, M. Beekes, D. Naumann, Biochim. Biophys. Acta 1501 (2000), 189-199.
355
Biomedical Applications of Near Infrared Spectroscopy

Andrew MACNAB 1 Faculty of Medicine, University of British Columbia, Departments of Pediatrics and Urologic Sciences. Director Near-infrared Study Group, University of British Columbia, Bladder Care Centre, Canada
Abstract. This Chapter discusses the latest advances in the application of Near Infrared Spectroscopy (NIRS) in biomedical science.
Introduction Near infrared spectroscopy (NIRS) is a non-invasive technology which uses energy from light in the near-infrared spectrum to monitor changes in local blood flow and hemodynamics and detect differences in tissue oxygen delivery, consumption and utilization. Norris [216] described the application of NIRS to the study of in situ human tissues in 1977, and in the same year Jobsis [2] published ground-breaking work applying NIRS to study cerebral oxygenation in vivo. Since then this technology has been widely applied as a biomedical research, diagnostic and clinical monitoring tool. Comprehensive reviews exist in the literature. This chapter provides an overview of the basic principles of NIRS spectroscopy, an introduction to the instrumentation and methods of measurement, a review of the inherent limitations and considerations relevant to biomedical applications of NIRS and a summary of such applications emphasizing more recent developments.
Methodology A Med-Line, Pub-Med, Biological abstracts, EMBASE/Exerpta Medica search of the literature from 1970 to 2008 was performed using combinations of spectroscopy, near infrared spectroscopy, NIRS, and terms related to medicine and biomedical applications. The search was not restricted to language, animal studies or types of article. Further publications were identified through review of the references from the initial set of articles. The principal materials which have been cited and summarized are published reviews of the basic science of NIRS, NIRS instrumentation, limitations of the technology and major biomedical applications of NIRS. Literature relating to studies published since relevant major reviews has been added, as have publications describing unique contributions to the evolution of biomedical application of NIRS, inCorresponding Author: Andrew MACNAB, Faculty of Medicine, University of British Columbia, Departments of Pediatrics and Urologic Sciences, Director Near-infrared Study Group, University of British Columbia, Bladder Care Centre, Canada.
1
356
A. Macnab / Biomedical Applications of Near Infrared Spectroscopy
cluding some citations describing novel techniques or innovative technology with future potential. Much has had to be omitted including the contribution of NIRS principles to pulse oximetry, most applications of specialized instrumentation beyond continuous wave and spatially resolved spectrometers, and basic science research remote from biomedical applications relevant to clinical practice. Scientific Principles of NIRS NIRS employs many of the fundamental principles of physics relating to the transmission of light through tissue. However, it is the unique combination of the transparency of tissue to near infrared light and the specific absorption spectra of individual chromophores that forms the basis of biomedical applications of near infrared spectroscopy. At most wavelengths of the visible spectrum, light is absorbed by skin and tissue, but photons generated in the near infrared spectrum 7001300 nanometers (nm) pass through skin and scatter in soft tissue and bone. In addition, naturally occurring chromophores absorb these photons in varying amounts determined by the chemical structure, color and concentration of each chromophore, and the wavelength of the light transmitted [337]. The principal chromophore of interest in biomedical applications using NIRS is hemoglobin (Hb) which has a different pattern of absorption when oxygenated (O2Hb) and deoxygenated (HHb). Cytochrome-c-oxidase (CCO) is another chromophore measured by NIRS. CCO catalyzes the metabolism of molecular oxygen in the mitochondria and absorbs light differently across the near-infrared spectrum depending on its redox status. The CCO signal is only about one tenth of the amplitude of the Hb signal [59]. In muscle the chromophore myoglobin (Mb) accounts for approximately 10% of the NIRS light absorption signal [177] however because the absorption spectra for Mb and Hb overlap NIRS is unable to distinguish between the two chromophores. In prior publications alternative abbreviations for chromophores include HbO 2 for O2Hb and Cyt, Cytaa3, or CytOx for CC0. Figure 1 shows the varying absorption of O2Hb and HHb and CCO at different near-infrared wavelengths and the extinction coefficients of adult hemoglobin. NIRS instruments use lasers to transmit pulses of multiple wavelengths of light into the tissues, and sensors to detect the photons returning that are not absorbed. Using lasers provides high spectral resolution, coherence the ability to detect very small changes in the propagation medium, and directivity, and high sensitivity with ultra fast photo detectors [93]. The technology uses modifications of the conventional Beer Lambert principles [71,165,248] (which relate media thickness and concentration in purified samples) to convert the ratio of emitted to detected light into changes in the concentration of oxygenated hemoglobin (O2Hb), deoxygenated hemoglobin (HHB) and the net redox status of cytochrome-c-oxidase (CCO). Mathematical software algorithms are used to convert the ratio between the amount of light transmitted and the amount detected returning into a derived concentration for each chromophore. The algorithms also accommodate for essential aspects of the basic physics of NIR light transmission influenced by the use of NIRS to monitor human tissue, e.g. light undetected because of scattering beyond the field of view and the pathlength of light through tissues [71,93,248]. Jobsis original study [137] using manipulation of inspired oxygen concentration and changes in the attenuation of NIR light across the head of a cat highlighted the principal limitations of NIRS. Namely that quantification of the changes observed in vivo is not possible where the pathlength is unknown, and that light attenuation occurs
357
Figure 1. The extinction coefficients of adult Hb and the varying absorption of O2Hb (HbO2) and HHb (Hb) and CytOx (CC0) across the NIR spectrum. (Reprinted with permission of The Royal Society from Delpy D. and Cope M. Quantification in tissue near-infrared spectroscopy. Phil Trans R Soc Lond B 352 (1997), 649659, [72].)
because of photon scattering as well as absorption. Hence, the light lost cannot be quantified, and the relationship between the changes in absorption that occur and the attenuation that results is non-linear [72]. The major limitation in clinical applications of NIRS continues to be the fact that the initial concentration of each chromophore is unknown, and that only absolute changes in concentration relative to the initial baseline concentration can be derived. Another limitation is that NIRS cannot distinguish between arterial and venous hemoglobin. However, with real time sampling and graphic conversion of data, patterns of change in chromophore concentration and magnitudes of change are derived which provide valuable information, as they can be used to infer physiologic change occurring within the tissue interrogated [30,72,119,248]. Such changes include: an increase or decrease in O2Hb (an indirect measure of oxygen content); an increase or decrease in the total hemoglobin tHb (change in blood volume); an abrupt decrease in O2Hb with simultaneous increase in HHb (ischemia); and a gradual decrease in O2Hb and increase in HHb (hypoxia). Also, as cytochrome-c-oxidase is the terminal enzyme in the intra-cellular mitochondrial respiratory chain and drives >95% of oxygen consumption and the synthesis of adenosine triphosphate (ATP), changes in CCO redox status provide information relating to electron transport and oxidative phosphorylation at the cellular level. This is a unique ability provided by NIRS monitoring [55,288] that has been studied in animal
358
models [302,271,272], and in humans [56,57,106,263,277,287,288]. Thus, interpretation of NIRS data that includes changes in O2Hb, HHb and CCO signals can offer insights into the basic physiology and potential clinical relevance of oxygen utilization and energy dynamics at a cellular level. Algorithms and assumptions: [52,72,107,248]. The majority of NIRS systems used in biomedical applications incorporate in their basic underlying formula the same equation as other scientific applications use for constant relative rate of growth/decay e.g. Newtons law of cooling, the estimation of bacteria populations, and carbon-14 dating. For NIRS this equation is written as: I = Ioe rt [I is the decreased intensity of the emitted light: Io is the original intensity of the emitted light: e is Eulers e (i.e. 2.718282): r is a constant known as the absorption coefficient specific to the wavelength of light and chromophore being sampled: and t is the linear thickness of the sampling medium. The foregoing relates to Lamberts findings of 1760, but a century later Beer determined that the same formula can be applied to chromophore concentration rather than thickness. Clinical NIRS measurements of light transport depend upon combined thickness and concentration which can be resolved mathematically as: I = Ioe rct and alternatively I/Io = 10ct or log (Io/I) = ct [c is the chromophore concentration: is the reciprocal value of the thickness in centimeters of a one molar solution when the light emerging from the solution is one tenth of the original intensity]. These formulae assume that there is only one chromophore and also that it is uniformly distributed throughout the sampling medium. However, in biomedical applications there are multiple chromophores residing in chaotic fashion within a medium of undetermined thickness. This is resolved using multiple wavelengths of emitted light, at least one per chromophore of interest, and modifying the Beer/Lambert formula given above to replace linear thickness with a correction factor for the separation between emitter and detector in conjunction with a scattering co-efficient to account for non-absorbing undetected photons. Thus a matrix solution method is required of the form: log (Io/I) = [1c1 + 2 c 2 + 3c3 + .ncn]dP + S identifies the equation within the matrix consisting of one equation for each wavelength: d is the linear distance between the light emitter and detector: P is a path length correction factor to account for non-linearity: and S is the scattering coefficient [17,102,103,311]. However, with application of such formulae to NIRS and living tissue it must be assumed that each of the chromophores present, but not accounted for in the matrix solutions, are either too weakly absorbing or do not change in concentration [179]. It must also be assumed that the number of scattered photons remains constant, the wavelength dependent path length factor is valid for the given tissue in the given subject, and that the separation between emitter and detector is also constant [75,151]. A pri-
359
mary assumption is that Eulers e realistically mimics the continuous compounding of the loss of photons as biological tissue thickness increases. While the use of e is a logical assumption to gain improved accuracy, there is no experimental validation for its use in clinical NIRS. e is a standard component of bench-top analytical spectroscopy, but in that setting the chromophore is sampled at a standardized concentration in a standardized volume and is homogenously distributed; therefore, e can be validated to represent the missing photons that cannot be accounted for as being either absorbed by the chromophore or collected by detectors surrounding the sample chamber [320]. In bench-top spectroscopy, a somewhat imperfect e can be used as a substitute for the true lost photon count because it is used identically for both the known purified-sample and the comparison unknown-sample. However, the same is not true in biomedical applications of NIRS where the known purified-sample data is derived from bench-top non-reactive purified chromophores in a standardized volume, temperature, concentration, and distribution, but, in contrast, the in vivo tissue sample to which it is compared is of unknown volume (as it cannot be surrounded entirely by photon detectors), its temperature is uneven, and it is biologically active and heterogeneous. The inherent assumptions of biomedical NIRS where a modified Beer/Lambert equation is employed dictate that it can only be used to determine absolute changes in concentration relative to the initial unknown concentration at the start of data collection. Consequently, NIRS can only be used for intra-subject trials where the data collection is not interrupted and cannot be used for inter-subject trials or interrupted data collections except by comparison of the patterns of change, and their magnitude and rate of change. Furthermore, such data collections require that during study the tissue or organ monitored undergoes a physiologic change generated by a clinical intervention (e.g. induction on of hypoxia or ischemia) or a physiologic activity (e.g. isometric muscle contraction or bladder contraction during voiding). Individual algorithms differ between NIRS instruments. Errors sometimes contained in early versions of such software and have compromised the validity of related research [171]. Significant efforts have been made to refine and improve the algorithms incorporated in modern equipment. Such improvements include algorithms to discriminate better between changes due to CCO and Hb in order to provide a more discriminant level of resolution [56,233]. Doubts have been expressed regarding the validity of measurements of CCO concentration change. Such doubts were based on concerns that the CCO signal might either be a ghost generated when large changes the hemoglobin signal occur or an artifact derived via faulty algorithms [181]. However recent studies support the validity of CCO measurement. Cooper et al used perfluorocarbon blood exchange in a neonatal animal model to identify that the CCO signal was stable despite large changes in O 2Hb and HHb [57]; and Tachtsidis et al induced orthostatic hypotension in humans and from a range of relationships derived inferred that artefactual changes in CCO from crosstalk between hemoglobin and CCO were not present [288]. Instrumentation Recent reviews describe the development of NIRS instrumentation, the technical specifications of individual units and recent advances in the technology [61,93,169,248,314,334]. The majority of commercially available instruments are continuous wave (CW) spectrophotometers. These instruments have proven reliability in the measurement of
360
changes in O2Hb and HHb. Multiple wavelength instruments can also monitor change in redox status of CCO. Instruments modified for spatially resolved spectroscopy incorporate multiple sensors at different distances from the emitter. Multi-channel CW instruments can monitor more than one site simultaneously and when configured with the appropriate optode holder and software are used for mapping (fNIRS). Phase modulated (PM) and time resolved (TR) spectrometers differ from CW instruments, and were developed principally for the measurement of mean optical pathlength [71,76,248]. PM spectroscopy calculates the phase shift and phase modulation of a continuous frequency- modulated source of light [43]. TR instruments input light into tissue as a picosecond pulse and the emerging intensity is detected as a function of time (the temporal point spread function) [248,334]. Absolute concentrations of O 2Hb and HHb can be measured using PM and TR spectrometers [72,93]. However, this is expensive technology and instruments are large, two issues making clinical use problematic. CW instruments either use a white light source combined with a CCD array and discriminate for wavelength with a grating [59], or use light sources with discrete wavelengths (e.g. laser diodes) and a photodiode, photo multiplier tube or CCD as a detector. With CW technology the assumption is made that photon scatter in tissue is constant and a tissue specific differential path length factor (DPF) is used to calculate the optical path length from the inter optode distance [71]. Studies using PMS indicate that in brain in a particular subject photon scatter and path length are probably relatively constant [156]. Basic commercial CW NIRS instruments contain the following: a) at least one pulsed laser diode for each chromophore being sampled. Typically the lasers emit light in 1, 2 or 4 wavelengths in the 729 to 920 nm near infrared wavelength range with a 5 nm spectral width and pulse duration of 100 nanoseconds at 2 kHz cycle frequency; b) Fibreoptic bundles that transmit light from the source to a tissue interface (probe or patch) and back to the instruments photo counting hardware; c) Optodes in the tissue interface that emit light into the tissue and receive the photons returning; d) Photon counting hardware (photodiode, photomultiplier or CCD); d) Computer with software containing algorithms for converting raw optical data into chromophore concentrations, storing and displaying data; e) A visual display where NIRS data are displayed numerically and/or graphically against time. Some machines provide the option to select wavelengths from multiple options; the ability to use more than one data channel to allow comparison between monitoring sites and/or tissue; a signal weighted towards brain or muscle tissue (the former is achieved by subtracting a superficial signal from a deeper signal) [185,314]; or use spatially resolved spectroscopy (SRS), which enables the ratio of oxygenated to total tissue hemoglobin to be measured and a quantitative measure of tissue oxygenation to be derived [30,93,248]. Each single channel optode is comprised of a paired emitter and receiver. Optodes may be positioned in one of two modes for conventional NIRS monitoring, transmission or reflectance. In transmission mode the emitter and receiver are positioned on either side of the tissue to be interrogated approximately opposite each other. This mode has been mainly used for brain studies in small newborn infants but can be used for muscle and other studies. The interoptode spacing must be kept constant to avoid alteration of optical pathlength. The advantage of transmission mode is that relatively global sampling of the tissue of interest occurs. For effective NIRS monitoring in the newborn brain in this mode the upper limit for interoptode separation is approximately 8 cm. In reflectance mode the emitter and receiver are positioned in the same plane on
361
Figure 2. The configuration of a NIRS system for transcutaneous interrogation of the bladder detrusor; the banana shape of the photon path through tissue and an approximation of the depth of penetration. (Reproduced with permission from: Stothers L, Shadgan B, Macnab AJ. Urologic applications of near infrared spectroscopy. Can J Urol. 15(6) (2008), 43994409 [282].)
the tissue, usually angled slightly towards each other. The distance between the emitter and receiver, or inter-optode distance (IOD), determines the depth of penetration of photons into the tissue. Penetration depth is approximately half the IOD [66,122]. Reflectance mode is used for the majority of NIRS studies. Interoptode distances vary, and are commonly between 48 cm. Closer distances result in a greater proportion of the signal coming from subcutaneous tissue and at larger distances signal quality can become problematic because of noise. Optodes can be positioned in more than one location to simultaneously sample and compare data e.g. opposite sides of the brain or opposite limbs, or to monitor two tissues simultaneously. Some instruments with multiple channels can be used for mapping/imaging by using combinations of multiple emitters and sensors mounted in a grid configuration. The fNIRS technique does not require strict motion restriction so is well suited for monitoring during normal activities e.g. exercise [132]. Whatever the mode of optode placement it is essential that they are attached in a manner that prevents any change in interoptode separation from occurring during monitoring. Even very small changes will introduce significant artifact. There are many different techniques for optode attachment including incorporation in an adhesive patch, location in a holder attached by double-sided adhesive discs, and a variety of individual methods combining straps, tape and adhesive. Figure 2 illustrates: a typical set up of a CW NIRS instrument for transcutaneous monitoring in reflectance mode (in this case for interrogation of the bladder detrusor muscle [282]); a two-dimensional projection of the most probable photon path between emitter and receiver [34]; and the depth of penetration into tissue which is approximately half the inter-optode distance [66,122].
362
Noise and sensitivity vary between instruments. Units with lower noise level can usually sample at faster rates. Manufacturers supply an appropriate phantom to test the basic performance of their system. Function can also be confirmed by performing a simple physiologic measure such as forearm ischemia with a sphygmomanometer cuff occluding the upper arm. The optical cables are made from bundles of individual glass fibres which can fracture over time reducing light transmission. Some instruments have a default warning in this event, cables can also be checked by eye with a handheld light source. While NIRS data is typically displayed graphically against time, recent advances also allow fNIRS data to be viewed as a colour video of dynamic change of O2Hb and HHb respectively, and to interrogate larger areas of tissue or the surface of an organ. Multiple channel systems (224) provide enhanced spatial resolution; used with multiple optode grids such systems are relatively inert to movement artifact because of the multidistance geometries of the sensor interface [234] and are increasingly used for brain mapping [8,37,132,218,314,334]. Spatially resolved spectroscopy (SRS) uses CW light emission with a multiple segment photodiode chip to allow light detection at two or more different distances from the emitter. This enables the absolute ratio of O2Hb to tHb to be determined and hence a calculation of tissue oxygen saturation to be made. Assumptions made in this calculation include that the tissue interrogated is homogenous. As NIR light is fully absorbed in large vessels this parameter represents the average saturation of the Hb present within the small vessels within the photon path (venules, arterioles, and capillaries). Also, as only the minority of this blood is in capillaries and arterioles tissue oxygen saturation predominantly measures venous oxygen saturation. On Hamamatsu (NIRO) instruments this parameter is referred to as the tissue oxygenation index (TOI) and is calculated using photon diffusion theory [285]. A NIRS system from Somanetics (INVOS) that also uses multi-distance measurements of light attenuation from spaced optical detectors provides a calculation of mean tissue Hb saturation referred to as rSO2 by measuring the ratio of light absorption by O2Hb and tHb [222,297]. When used on the brain this instrument is described as a cerebral oximeter. While both instruments use SRS, photodiodes as detectors and photon diffusion theory to estimate absolute path length [243], comparisons between TOI and rSO2 in animal and human studies emphasize important differences between the devices and that the absolute values of TOI and rSO2 are not identical [105]. Differences between the NIRO 300 and INVOS 5100 include: The computational algorithms used to convert light attenuation values to chromophore concentration [102,103,171]; The number of laser diodes and photodiodes used; The separation between the emitter and detector(s) which influence sample volume and depth of penetration [149]; The separations between the multiple photodiode receivers which influences path length; data sampling e.g. NIRO 300 records at precise intervals and INVO 5100 records at irregular intervals [105]; Recognition that hemoglobin concentration and the thickness of the cerebrospinal fluid layer and skull have a greater effect on rSO 2 than TOI [345].
363
Although individual instruments have been shown to demonstrate changes that correlate with alteration of important physiologic variables, and some studies show agreement between the instruments, many investigators have concerns that technical limitations currently compromise the applicability of TOI and rSO2 measurement [74,189,290]. In a review of the quantification, bias and precision of these measurements, Greisen [116] concluded that the precision of the instruments is insufficient for their clinical use at present. There are also conflicting views that advocate both for [79,127] and against [67,200] the use of such monitoring during cardiac surgery. It appears that both instruments require further development and validation, and the technology related to SRS will continue to advance as the performance and construct validity of these instruments is studied further [205]. Concerns also exist regarding the inability of CW NIRS studies to provide quantitative data, the limited number of studies with substantial patient populations, the variations reported in inter-subject measurements, and the lack of endpoints for clinical decision making. Consequently, clinicians are reserved regarding the contributions CW NIRS instruments can play in the context of care. Such reservations combined with concerns over the relative complexity of using many NIRS instruments while providing clinical care is reflected by the very limited number of commercial devices incorporating this technology. However, there is consistency and reproducibility in the patterns of change in O2Hb and HHb concentration observed using conventional CW NIRS in different tissues, in studies conducted by different investigators, and when using NIRS equipment from different manufacturers. Such data is important and of relevance to many basic science and clinical research applications. Also, it is increasingly recognized that variations in NIRS data reflect in part important individual differences between subjects (e.g. body mass index or small anatomical or physiological variations in the tissue interrogated) rather than flaws in measurement. Also, that many of the parameters measured by NIRS are unique and cannot always be directly compared to other measurements. In addition, it is always important when conducting research using NIRS or interpreting published studies that the characteristics and limitations of the hardware, algorithms and software of the NIRS instruments employed are recognized and factored into comparisons between studies. NIRS hardware and software are continuously evolving [93,125,135,169,248,254,261,314,334,340,341]. Novel fiber optic probe design has advanced applications of reflectance, polarized reflectance, fluorescence, and Raman spectroscopy [306]. In the context of instrumentation, although instruments using PM and TR spectroscopy have important advantages their size, expense, complexity and inherent technical limitations render them unsuitable for use at present in all but a few specialized applications. Also, published validation studies are few in number.
Data Analysis Parameters Measured by NIRS Various authors summarize the parameters measured by NIRS and address the validation of the related methods of measurement [8,93,61,176,202,325,334,336]. Parameters used in muscle at rest and during exercise are also reviewed [23,30,9,204,229,241,334].
364
Table 1. Parameters measured directly and indirectly by near-infrared spectroscopy and imaging instrumentation PARAMETER O2Hb, HHb, tHb oxCCO OI UNITS MODALITY AUTHOR [Reference] Delpy [72] Tisdall [296] Grassi [112] Matcher[180], De Blasi [68,69], Quaresima [242], Cuccia [65] Fantini [89] Oda [219] Benni [20] Matcher [179], Cooper [54] Yoxall [349] Franceschini [100] Franceschini [99] Durduran [77] De Blasi [69] Boushel [28] Wolf [335] De Blasi [68, 69]
a.u., M cm, M
D D (by SRS)
D (by PMS) D (by TRS) D (by callibration) Second differential Muscle SvO2 % I (by VOM) D Muscle tHb M D (by PMS) a.u. D (by DWS) Muscle BF mL/100 mL/ min I (by VOM) I (by ICG) Muscle Hb flow M / min I (by VOM) Muscle VO2 Ml/ 100g/ min I (by VOM) I (by AOM) Muscle recovery time s D Chance [44] Muscle compliance mL/L/mmHg I Binzoni [24] Cerebral SvO2 % I (by VOM) Yoxall [348] D Wolf [332] Cerebral tHb M D (by PMS) Choi [48] I (by O2 swing) Wolf [333] I (by O2 swing) Wyatt [339], Wolf [333] Cerebral BV mL/100 mL SRS and second Leung [161] differential I (by ICG) Hopton [129] a.u. D (by DWS) Durduran [78], Li [162] Cerebral BF mL/100 mL/ min I (by O2 swing) Edwards [81] I (by ICG) Roberts [246], Keller [146] Cerebral VO2 mL/ 100g /min Combination cerebral Elwell [87] SvO2 and BF = Relative chances from arbitrary baseline, AOM = arterial occlusion Method, a.u. = arbitrary units, BF = blood flow, BV = blood volume, DWS = diffusing-wave spectroscopy, D = directly, I = indirectly, ICG = indocyanine green, OI = oxygenation index (O2Hb-HHb), oxCCO = cytochrome c oxidase redox state, PMS = phase modulation spectroscopy, SRS = spatially resolved spectroscopy, SvO2 = venous O2 saturation, tHb = O2Hb+HHb, TRS = time resolved spectroscopy, VO2 = oxygen consumption, VOM = venous occlusion method. (Reproduced with permission from: M. Wolf, M. Ferrari, V.Quaresima, Progress of near-infrared spectroscopy and topography for brain and muscle clinical applications, J Biomed Optics, 12(6) (2007), 062104 [334].)
Tissue O2 saturation
Greisen provides a concise overview including the relevant formulae and an appraisal of the studies in neonates evaluating NIRS methods of measurement, and in the case of TOI explains the problems with this measure and conflicting research related to it [116]. Table 1 summarizes the parameters that are directly or indirectly measured by NIRS spectroscopy.
365
Interpretation, Relevance, Reproducibility and Validation While many NIRS measurement parameters have been compared with other standard measures some constitute unique entities where no comparable gold standard measure is available for validation, e.g. changes in redox status of cytochrome. Other measures, e.g. tissue oxygenation index, reflect discrete regional measurements that often change appropriately in response to systemic physiologic change but do not compare consistently with the values obtained from standard lab values that reflect systemic rather than regional change. The lack of absolute values for O2Hb, HHb and CCO concentration, reports of difficulty with reproducibility of data, and concerns re TOI and comparable measures of brain oximetry have understandably compromised widespread adoption of NIRS, but should not lead to all applications and data collected being regarded as invalid. Some of the most significant data have come from studies using techniques for measuring CBF, CBV and venous O2 saturation with manipulation of FiO2 or impeding venous outflow [116]. Good correlation is reported between NIRS values and cerebral blood flow velocity by transcranial Doppler in the evaluation of cerebral vasomotor reactivity [266,294,318]. NIRS parameters also follow cerebral oxygenation during changes in ventilation, and relate to PaCo2 and SaO2 [176]. Over all it is clear that thoughtful and rational application of NIRS can provide important insights into the complex interrelationships among physiologic and pathologic conditions [336], and much valuable and unique and valuable in vivo metabolic information [265]. Reviews of methodologies used to validate NIRS [52,61,93,116,119,176,248,265] describe doppler ultrasonography, magnetic resonance (MR) plethysmography, positron emission tomography (PET), pulse oximetry, jugular bulb co-oximetry, and 131 Xenon Clearance. Indirect clinical validation comes from observation of the changes in chromophore concentration that occur during recognized physiologic events. These are reported during cardiac surgery, carotid artery surgery, exercise, pharmacological therapies and bladder emptying and most consistently with ischemia, hypoxia and changes in blood volume [52,107,172,173,176,243,309]. Importantly such patterns of change are seen consistently by different investigators using a range of instruments in a variety of human and animal studies. Figure 3 illustrates the effects of ischemia in muscle, brain and bladder measured by CW instruments, and Ferrari has published data from brachioradialis muscle in response to arterial occlusion measured by single distance continuous wave, phase modulated and spatially resolved spectrometers respectively [93]. Refinement of the methodology for measurement has been required for some parameters. For example, the promise of NIRS as a method for quantifying cerebral blood flow (CBF) and cerebral blood volume (CBV) did not hold up well initially. The original description of the method proved problematic for others to duplicate. It relied on a brief reduction in arterial oxygen saturation (SaO2) followed by an abrupt increase to generate a marker that could be followed through brain tissue via NIRS, and the validity of the calculations from which CBF was derived was soon called into question [181,208]. One major difficulty was that the point at which SaO2 begins to rise must be defined precisely, as temporal movement of this point by the smallest amount in either direction has a major influence on the CBF value obtained. It is essential that an oximeter with enhanced capability re sampling time is used [96] conventional commercial unit sampling rates and averaging software are not adequate. Another concern was
366
Figure 3. (Reproduced with permission from: Stothers L, Shadgan B, Macnab AJ. Urologic applications of near infrared spectroscopy. Can J Urol. 15(6) (2008), 43994409 [282]). Three graphs showing the pattern of change in chromophore concentration in response to ischemia in different tissues. NIRS values for (oxygenated hemoglobin [O2Hb], deoxygenated hemoglobin [HHb]) are plotted against time and illustrate the consistency of the pattern of change seen in response to ischemia in A) muscle (human forearm during arterial occlusion following rhythmic isometric exercise) Oxymon NIRS prototype (University of Nijmegen NL) transcutaneous mode (reprinted with permission of Blackwell Pub. Ltd. from: van Beekvelt M et al. Clin Physiol & Fun Im. 22(3) (2002), 18 [308]) [stable total hemoglobin [tHb] indicates no change in blood volume]; B) brain (human during circulatory arrest on cardiac bypass) Hamamatsu NIRO 300 transcutaneous mode (reprinted with permission of IOS Press from: Macnab A et al. Spectroscopy. 17 (2003), 483490 [173]) [the equal and opposite change in cerebral O2Hb and HHb indicates no change in total blood volume]; and C) bladder (rabbit in response to aortic occlusion Hamamatsu NIRO 300 with emitter and receiver directly opposed across the surgically exposed bladder [the rise in HHb is unequal to the fall in O2Hb due to volume loss via the unobstructed venous outflow in this model].
367
the evidence that manipulation of SaO2 in itself can influence CBF [118,120]. Refinements to the CBF measurement methodology followed, and later a modification using the dye indocyanine green as a tracer was developed. Initial work during cardiopulmonary bypass surgery began a series of improvements involving both the experimental technique and the required algorithm. Ultimately Brown et al. [33] described the necessary methodology for quantitative measurement of CBF, CBV and mean transit time (MTT) in an animal model over a range of physiologic situations (including where there were variations in PCO2 and/or hemodynamics). Subsequently the accuracy of their measurement method has been validated against computerized tomography. This is an important example of how methodical attention to detail in research has advanced the validity and relevance of NIRS spectroscopy. Quantitative measurement of CBF using indocyanine green dye dilution [146] was later validated by data showing agreement with CBF values obtained by perfusion weighted MRI [147]. Recent literature summarizes the techniques of measurement, reproducibility and correlation of NIRS measurements in exercise physiology, trauma and neuroscience [93,119,176,204,248,325] with gold standards where they exist. In studies involving muscle, myoglobin (Mb) is an additional chromophore of relevance. Mb accounts for approximately 10% of NIRS light absorption [177], but because the absorption spectra for Mb and Hb overlap NIRS is unable to distinguish between the two chromophores. However, absorption by myoglobin is essentially constant as the concentration of this chromophore remains unchanged within the NIRS filed of view during the period of study, but the presence of Mb and its potential impact on NIRS data remains a factor in interpretation of muscle studies [23,30,93,204]. Several NIRS parameters measured in muscle have sufficient reproducibility to be used in sports medicine and exercise science studies. Van Beekvelt studied the calf muscles of 11 healthy adult volunteers during an exercise protocol which included a period of arterial occlusion and found similar responses for O 2Hb, HHb and tHb in the overall group although individual tracings varied (Fig. 4) [309]. Muscle venous saturation (SvO2) values also have good agreement when measured using the venous occlusion method [349], and via a newer method based on respiratory induced oscillations of the absorption of NIR light [100]. High within subject reproducibility of temporal information from NIR topography of the brain (sensorimotor cortex) is also reported in healthy adults. The maximum signal amplitudes and location of activation centers were compared for each subject between two sessions over about 6 months. While signal amplitudes varied and no consistent tendency in the changes was found among subjects, the distance between the activation centers was relatively small (<20 mm) across subjects, and within-subject comparisons of signal time courses showed high correlation coefficients (>0.8) between sessions [255]. Huppert [134] reviews studies examining the relationship between NIRS and fMRI, discusses the conflicting results, and describes greater correlation of fMRI with NIRS measurement of HHb than with O2Hb or tHb. NIRS and BOLD (blood oxygen level dependent) and ASL (arterial spin labeling) based fMRI were done simultaneously during a shortduration, event-related motor task in human subjects, and the temporal dynamics of the hemodynamic responses compared. A high correlation was found between the NIRS derived tHb and O2Hb and fMRI ASL measured cerebral blood flow. Co-variance was observed in these parameters between subjects but as this was in agreement with known hemodynamic models it was deemed to support fMRI and NIRS having similar vascular sensitivity. Brain oximetry measurements (NIRO 300 TOI and Invos
368
Figure 4. This shows the changes observed in gastrocnemius muscle O2Hb, HHb and tHb (+/ SD) n = 11 (SD presented one-sided for clarity) during an isometric plantarflexion exercise protocol of 1 minute (min) rest; 3 min 30% maximum voluntary contraction (MVC) force; 8 min rest; 3 min rest + arterial occlusion, 3 min MVC + arterial occlusion; 10 min rest. The grey bars = periods of sustained isometric exercise and the black bar = period of arterial occlusion. Rapid deoxygenation occurred during exercise with rapid reoxygenation after cessation. Blood volume was constant during arterial occlusion and both exercise phases apart from a small drop initially at the start of exercise. At the end of 3 min occlusion at rest oxygen stores were not completely depleted since HbO2 declined further during the exercise phase that followed. Even during ischemic exercise 36% of subjects showed a slow persistent de-oxygenation and no plateau in O2Hb indicating lack of complete depletion of O2Hb, while in the remaining 64% a plateau was only reached after an average of 2.5 min. of ischemic exercise. Reproduced with permission from M. van Beekvelt 2002 [309].
rSo2) have been evaluated in a variety of studies and reviews. Although conflicting, the evidence suggests poor reproducibility [74,116]. However, there is widespread agreement of the validity of monitoring the rate and pattern of change in chromophore concentration (O2Hb, HHb, tHb, and CCO) in a given individual and equating the alterations observed with physiologic effects related to tissue oxygenation and hemodynamics [30,93,119,248,325,336]; particularly as a means of recognizing physiologic changes of clinical relevance in real time, such as the onset of ischemia or evolution of hypoxia. The relevance of such data is emphasized by the patterns of such changes being comparable between individuals, in different tissue and across studies. The magnitude and rate of chromophore change do often differ to some degree, but in carefully conducted studies this is probably mostly due to intersubject differences influencing photon scatter and pathlength. However, many studies do report a wide variation in chromophore concentration values. This issue added to the inability of NIRS to provide an absolute value makes comparison of NIRS chromophore concentration data between patients problematic, and also makes it difficult to define clinical endpoints for NIRS values. The basic science and clinical relevance of changes in CCO concentration are the subject of ongoing study. Studies in animal models [42,57,137,271,302,303] have predominantly examined the brain, but also the spinal cord [172] and more recently muscle. A strong association has been shown between changes in CCO and the status of cellular energy [15,47,183,228,272,302]. Human studies have included newborn infants [58,82], children [106,263] and adults [56,187,287,288]. Amongst surgical patients changes in CCO correlate with neurological outcome following cardiovascular procedures [141,213,215]. Thus, interpretation of NIRS data that includes changes in
369
O2Hb, HHb and CCO signals can offer insights into the basic physiology and potential clinical relevance of oxygen utilization and energy dynamics at a cellular level [180,187,287,288]. Cells progress through a series of changes in response to a fall in oxygen tension [15]. Initially oxygen uptake is maintained which implies that redox state increases before oxygen consumption decreases. When oxygen delivery (DO2) decreases to a critical level energy production (ATP) occurs via anaerobic glycolysis, and lactate accumulates. Then, with a further decrease in DO2 cell function becomes sufficiently compromised that oxygen consumption (VO2) decreases, and ATP production fails. The relationship of tissue oxygen saturation, venous oxygen saturation, oxygen delivery and oxygen consumption is dynamic and multidirectional [240,325], and impacted by any degree of mitochondrial failure. Consequently changes in CCO must be interpreted in the broad context of the physiologic state of the patient and the tissue monitored; as substrate and energy depletion and any cytokine mediated mitochondrial dysfunction due to sepsis [95] will impact CCO redox status in addition to changes in oxygen delivery and consumption. Limitations of NIRS Limitations relevant to biomedical applications include issues related to studies involving human tissue posed by the basic science principles underlying NIRS, and constraints imposed by current hardware and software. NIRS only provides measurement of change in chromophore concentration from baseline rather than an absolute quantitative measurement, as the exact concentration of the chromophores within the tissue being studied is not known. Consequently NIRS data are most informative where a temporary change in the physiologic state of the tissue can be induced or is anticipated e.g. ischemia; a change in oxygen saturation or blood volume; or where physiologic changes occur as a result of organ function e.g. isometric muscle contraction, respiration, or voiding. Transcutaneous sampling requires photons to travel through skin, subcutaneous adipose tissue, underlying muscle and sometimes bone. Biological tissues influence photon transmission and the majority of photons are scattered. Scattering is widespread, causes photons to travel further than the direct path between a NIRS emitter and sensor, and is influenced by tissue planes and cellular boundaries. The scattering of photons in biological tissue and their unknown pathlength through tissue between emitter and receiver are the principal factors confounding the ability of NIRS to provide absolute chromophore quantification [72,93,119,176,248]. In in-vitro studies the effect of human skin is fairly small and bone is relatively translucent, but in-vivo skull thickness and the depth of the cerebrospinal fluid layer can negatively impact the intensity of light reaching and returning from brain tissue [220,345,346]. In humans measurement of cerebral blood volume before and after scalp ischemia indicate that provided the distance between light entry and exit on the skin surface is sufficiently large changes in scalp blood flow have no effect on NIRS measurement of cerebral hemodynamics [221]. In studies with adequate interoptode separation approximately 5% of the signal is absorbed by fat in lean individuals [120,128]. However individual differences in adipose tissue thickness do effect NIRS measurements, and in subjects with significant subcutaneous fat and high body mass index (BMI) the NIRS signal is blunted because the optical characteristics [184] and oxygen consumption [307] of adipose tissue are different to those of muscle [186]. The precise in vivo influence of adipose tissue on NIRS measurement remains uncertain however as only a few studies have investi-
370
gated these effects [307]. Consequently in studies comparing individuals with large differences in body fat the effect of adipose tissue on the NIRS signal should be considered, and data between patients evaluated accordingly. An equation for correction of NIRS changes due to the fat layer has been reported based on the correction curve of measurement sensitivity against fat layer thickness from simulation and in vivo [211]. Pathlength measurement during real time clinical studies is viewed as impractical at present by most investigators, consequently a differential pathlength factor (DPF), usually obtained from direct time of flight reference data obtained in vitro, is applied to the algorithm solving for chromophore concentration. DPF values exist for various tissues including cranium (newborn and adult), and forearm and calf (adult male and female) [5,71,75,312]. Where DPF values are unknown it is the convention for a stated approximation to be incorporated into the relevant algorithm. Measurements of pathlength in vivo are possible using phase modulated instruments and other forms of spectroscopy used experimentally. The depth of penetration of NIR light into tissue is limited and with CW NIRS studies is approximated as half the distance between the emitter and sensor [66,122]. This restricts the transcutaneous application of NIRS to the interrogation of superficial tissues and organs, and means the effective depth of study to 3040 mm, with perhaps a maximum limit of 60 mm [30]. However, a wide range of custom optical probes have been developed that extend the scope of spectroscopic studies to a range of applications [306]. Peebles [226] developed a probe that could be inserted though the human cervix during labour to monitor changes in fetal cerebral haemoglobin and oxygenation. This concept has most recently enabled the urethral sphincter and pelvic floor to be monitored via a transvaginal approach [284]. Other investigators have developed probes that can be introduced invasively to various locations where tissue apposition allows sufficient penetration of photons for the organ of interest to be studied, including versions that provide NIRS monitoring during surgical procedures [7,172,195]. An endoscopic NIR optical tomography technique has been described [232], and an in vivo method for optical characterization of human prostate tissue [286]. Surgical resection of colorectal tissue using NIRS analysis via a fibre-optic probe to classify tissue as cancerous or normal with high accuracy based on NIR vibrational spectra [152]. In animal models probes have been developed for transesophageal evaluation of changes in myocardial oxygen supply, intra-gastric study of the effects of hemorrhage [325] and intra-abdominal monitoring of liver blood flow during induced septic shock [203]. There is the potential for ambient light to interfere with transcutaneous measurement. To address this optode patches generally incorporate light screening material. This approach is usually adequate except when monitoring in bright sunlight or under lights of high intensity such as those found in an operating room. If necessary an opaque cover can be used to shield the emitter/receiver further. A default that warns when excess extraneous light is detected is available on some systems, and others incorporate a daylight filter into the receiver optical cable [314]. On rare occasions data has to be collected in darkness to overcome the transparency of the tissues, e.g. during studies requiring a small experimental animal such as a rabbit. Other limitations and extraneous variables that also need to be considered include: Absorption of light by water occurs, but only to a small degree when using the NIR wavelengths commonly chosen for NIRS lasers, as these have the lowest absorption for water.
371
Attenuation by structures other than the tissue of interest will influence the signal. This is of particular concern when monitoring the brain where skin and subcutaneous tissues exert a contributory effect. This can usually be accommodated and the signal obtained utilized, but major degrees of light absorption that can invalidate the signal can occur if either the temporalis muscle or the sagittal venous sinus are within the NIRS field. Attenuation by blood accumulated within the field because of bleeding or the presence of a hematoma. Elevation of serum bilirubin can also have an effect on measures of cerebral oxygenation [176]. Reduced photon penetration occurs to some degree in patients with dark skin pigmentation [327]. Acute hemodilution alters optical path length [159]; Electromagnetic interference (EMI) adds noise [170], EMI effects can be resolved by grounding the patient via a wrist strap or a contact incorporated within the optode patch. Fibreoptic cable failure compromises light transmission. Individual fibres are subject to fatigue and as more and more fracture photon transmission can become inadequate for measurement. Some instruments include a default to alarm in this event, or another means of indicating the adequacy of light transmission.
The effects of movement require special consideration. Movement that changes the interoptode spacing to any degree during measurement alters the photon path length and invalidates the data. However, spontaneous movement that occurs with the interoptode spacing kept intact does not result in artefactual data. Movement of a limb or the whole patient is easily recognizable, usually as a deviation from baseline involving all NIRS parameters. Rhythmic isometric handgrip exercise generates regular fluctuation in chromophore concentration (O2Hb and HHb) of low amplitude during studies on the forearm [308]. Strict motion restriction is also not necessary with fNIRS studies using grids of emitters and sensors, hence the applicability of this methodology to brain mapping and neuroimaging techniques conducted during a range of normal activities [132]. Optode pressure should be kept constant during monitoring. Variations in pressure will have an effect on path length and can alter the hemodynamics within the tissue field. The high temporal resolution of NIRS makes it possible to visualize the effect of a variety of physiologic events in the data stream, and normal respiratory effort is evident when NIRS optodes are placed on the soft tissue of the intact chest. Breathing results in an undulating pattern of change in O2Hb and HHb concentration that corresponds with the cycle of inhalation and expiration. However, this pattern of change may not be entirely attributable to actual change in chromophore concentration because some alteration in the linear path length between the emitting and receiving optodes likely occurs as the chest expands and contracts. Similar oscillations seen in the absence of respiratory effort have been attributed to Mayer waves from vasomotor activity [53]. Consequently, while NIRS in this application may be unsuitable as a means of estimating changes in O2Hb during normal breathing it can be used to determine rate of respiration, or monitor for significant physiologic change during breath holding or hypoxic/ischemic episodes. In brain studies small sinusoidal changes are seen in cerebral hemodynamics during expiration there is an increase in tHb and O2Hb and a smaller rise in HHb, these changes are reversed in inspiration [86].
372
NIRS relies on software algorithms to enable the basic principles of NIRS physics to be applied in a biomedical context and to facilitate data interpretation. In this regard NIRS is no different to other technologies already in the medical mainstream such as oximetry, the most widely used application of photonics technology, and CT and MRI imaging. Similarly, it is important to recognize when comparing a novel technology such as NIRS to a current gold standard that the virtually all extant technologies also make similar concessions through their software algorithms to a range of basic science principles in order for them to have clinical applicability. Adequate reproducibility of any measure is essential for scientific and clinical use. The NIRS literature has been problematic in this regard, and many publications are limited to single cases or at best small series. While important to demonstrate the feasibility of NIRS measurement, especially in new applications, most of these studies are unable to provide sufficient robust data for conclusions to be made regarding reproducibility. Many excellent scientists have concerns in regard to the reproducibility of NIRS data and the potential for research studies to reflect differences due to the methodology or technology used rather than true differences in the parameters measured [116]. Hence the appropriate calls for both caution and more research and development. There is however, also recognition of the value of data demonstrating consistency in the patterns of change in chromophore concentration seen in response to specific physiologic stimuli across a range of tissue. The body of work in this regard, particularly through studies involving muscle [23,29,93,119,308,309,] is important. Once again, as stressed above in the context of instrumentation, it is essential when reviewing literature or study data relating to NIRS that the algorithms relied on and the methods of measurement employed are known and their limitations considered. The data available do indicate that for wider application of NIRS in biomedical research and for clinical monitoring further studies are required that are specifically designed to address the consistency of measurement and reproducibility of data. Confidence in NIRS Measurements In any biomedical application of NIRS a number of practical questions need to be considered in addition to the inherent limitations of the technology before the data obtained can be accepted with confidence. The principal questions are: How do we know we are detecting changes in chromophore concentration that reflect physiologic change? Is the tissue or organ of interest within the field interrogated by the NIRS photons?
Confidence that NIRS signals represent physiologic change in an organ of interest rather than happenstance, interference from subcutaneous tissue or artifact is based on a variety of scientific data. Firstly, the consistency with which characteristic patterns of change in the concentration of chromophores occurs with specific physiologic interventions (ischemia, hypoxia and changes in blood volume). This makes it most probable that when recognized and reproducible patterns are observed, even in studies involving unfamiliar organs or tissue, that they do represent physiologic change. Secondly, that when NIRS sensors are placed on the skin over an organ or directly on the organ surface comparable patterns of change occur. Thirdly, in the context of studies during a physiologic event (e.g. rhythmic isometric handgrip, or spontaneous voiding of urine),
373
that changes in chromophore concentration of significant magnitude and duration only occur in temporal association with the physiologic event, and fourthly, that simultaneous recording of NIRS parameters at a remote site yields contrasting patterns that are clearly random. Ensuring that the organ of interest is within the field of view and that a significant amount of tissue is being interrogated can be achieved by confirming the anatomy of the tissue(s) include in the field interrogated, e.g. by surface marking of the known landmarks or via ultrasound, and having adequate interoptode separation for the depth of penetration required [66,122]. Furthermore, studies using NIRS and fNIRS in conjunction with other imaging modalities confirm that comparable tissue regions are being monitored, and that with fNIRS it is possible to study a larger area than is currently interrogated using a single NIRS channel with one emitter and sensor.
Biomedical Applications of NIRS There are comprehensive reviews in the literature of biomedical applications of NIRS for basic science and clinical studies [2,8,10,23,30,93,107,116,119,124,132,176,218, 248,325,334,336], and applications of mainstream NIRS and variants of the technology to specific entities such as cancer [152,340]. This section will highlight some of these and outline recent applications and newer studies published since 2003. The Evolution of NIRS The German physiologist Karl von Vierordt published the first report describing spectroscopy of tissue in 1876 following studies of blood in the human hand [323]. Matthes hypothesized in 1939 that near infrared light could be used for measuring blood volume, and the first clinical application of NIRS took place circa 1940 at the SloanKettering Institute for Cancer Research to study the metabolism of steroids [138]. In 1941 Millikan described the first experimental oximeter, the precursor of pulse oximetry [193] following earlier studies of the absorption changes by myoglobin and hemoglobin [194]. In 1977, Norris [216] described the use of NIRS in the study of in situ human tissues and Jobsis published his application of infrared monitoring to study cerebral and myocardial oxygen sufficiency [137]. Over the following two decades, many studies focused on the assessment of changes in O2Hb and HHb concentration within the brain [22,31,32,54,73,85,86,121,156,165,166,167,246,262,321,338,339, 347,350] especially in neonates; research continued on the changes in redox status of CCO [55,58,263,331]; new research quantified CBF [36,81,88]; NIRS was used during surgery [83,88,115,215,291]; fetal studies were conducted [4,225,226,259]; there was a growing interest in the study of muscle [68,92], and exploration of the use of multichannel optical detectors [89] and imaging technology [76]. During the last decade the number of applications of spectroscopic techniques in health science research has grown steadily [93,107,119,132,176,222,325,248,356]. NIRS has continued to be used to monitor brain hemodynamics and oxygenation [190,297,241,326,333,350], measure skeletal muscle oxygenation [23,29,52,128,242, 308,309,335] and for studies in exercise sports science [204,229]. Applications have expanded to include continuous monitoring of tissue hemodynamics during cardiac and general surgical operations [80,189,176,249,343], and at the bedside in critical care units following trauma during management of shock, trauma and compartment syn-
374
drome [2,25,63,109,260,298,313,325]. At the cellular level, NIRS studies have advanced the understanding of metabolic and mitochondrial myopathies [113]. There has also been a resurgence of interest in fetal monitoring, applications of novel NIRS technology to derive spatial and spectral information [60,90,341], and simultaneous monitoring with other technologies [119]. Amongst the newest applications are: Extensive interest in brain mapping and topography [8,37,125,130,132,191, 218,289] identified by event-related increases in O2Hb in the cortex, and functional optical imaging [10,16,136,292]. New opportunities in the field of urology [175,261,282,283]. Optical imaging spectroscopy and spectral diagnostic techniques for characterization of cancerous versus normal tissue, and early detection/ diagnosis [91,153,210,217,237,273,340]. Spectroscopic characterization and analysis of atherosclerotic plaques in cardiac disease [40,324,329]. Use of novel variants of NIRS, derived spectra, autofluorescence, fiber optic probes, and analysis for basic science, pharmaceutical and medical applications, and diagnosis [9,27,49,253306], to derive venous pH from tissue spectra [267,342], and non-invasive glucose monitoring via diffuse reflectance [178], and drug related effects on hemodynamics [61].
Brain: The brain is the principal organ where NIRS monitoring has been applied in the context of basic research and measurement of parameters of potential clinical relevance. This is not surprising because of the immediacy of the effect of significant changes in oxygenation and blood flow, the clinical relevance of detecting such changes in real time, the value of being able to assess the efficacy of treatment intervention, and the long term consequences of delayed recognition or ineffective intervention. In addition there are many clinical situations where greater knowledge of the physiology underlying the acute changes that occur in the brain during illness and injury would allow logical and much needed extension of current treatment options. Neurological complications during critical illness remain a frequent cause of morbidity and mortality. To date, clinical measures to monitor cerebral function include electroencephalography, jugular bulb mixed venous oxygen saturation, and transcranial Doppler, but all are limited by either requiring an invasive procedure and/or not being sensitive enough to effectively identify patients at risk for cerebral hypoxia. Regional changes in brain oxygenation can be measured by PET-scanning and functional magnetic resonance imaging (fMRI), and brain blood flow determined via 131 Xenon enhanced CT scan. However these techniques have limitations in their applicability to clinical care at the bedside. They almost always require the patient to be moved from the intensive care unit and involve isotopes and radiation, and the fact that they all only provide data as a snapshot at a given moment in time is a major limitation. NIRS provides data that is unavailable at the bedside by other means emphasizing the longstanding recognition of the potential value of this form of monitoring [45]. In theory real time, continuous monitoring of physiological perturbations of cerebral perfusion and oxygen availability, and information on changes in cerebral blood volume should be possible. NIRS has high temporal resolution and via localized monitoring of the cerebral cortex reflects changes in cerebral oxygenation during arterial hypotension, hypoxic hypoxia and hyper and hypocapnea [176]. Most of the signal attributed to the
375
brain comes from the upper 12 mm of the cortical surface during measurement with an inter-optode separation of 3 cm [131]. Instruments providing a running estimate of vascular haemoglobin oxygen saturation offer the prospect of a surrogate measure of cerebro-venous saturation, an important variable in neuro-intensive care. However, reviews indicate that the level of precision of such measurement is currently a concern [74,116,297], and both the instrumentation and the level of understanding of the physiology involved has been described as immature [111]. The objectivity of studies reporting negative findings may be questioned [46], none the less studies with the latest instruments continue to demonstrate significant bias of their measurements to both venous oxygenation in the jugular bulb (SjO2) and central SvO2. Nagdyman [202] found TOI values were significantly lower than those for rSO2, a potential explanation being their non-linearity [105]. TOI is lower at high saturations and rSO2 is lower at low saturations. Also these cerebral oxygenation indices reflect measurement of only a small region of the cerebral microcirculation and are hence obviously different to more global measures. In spite of the significant correlation with co-oximetric measurement of SjO2 and SvO2 the substantial bias of the measurements of these NIRS instruments prohibits these methods as being regarded as interchangeable [202]. Comprehensive reviews have been published that outline early brain studies in animals [248]; the application of NIRS in pediatric neurology [176,268]; and methods for monitoring cerebral oxygenation with an evaluation of the current evidence that NIRS can identify deficits in cerebral oxygenation and counterpoint explaining the limitations and caveats that apply currently to NIRS use [2,61,93,116,176,297,325]. NIRS has yielded much credible and some important clinical research data [265]. The continuous nature of NIRS has been combined with monitoring of arterial pressure to provide measures of cerebrovascular regulation, and the most important results have been obtained using quantitative techniques for measuring cerebral blood flow, cerebral blood volume, or venous oxygen saturation with manipulation of FiO2 or impeding venous outflow from the brain [116]. Brain monitoring will undoubtedly continue in a research context [2,176] but the complexity of NIRS methodology of measurement and the lack of user friendliness of many instruments continues to hamper wider clinical interest along with the established concerns of reproducibility and defined intervention endpoints. There are many reports of novel applications of NIRS to evaluate conditions such as hydrocephalus and acute cerebral infarction [30], orthostatic hypotension [133], seizures, brain cell activation and psychiatry [176]. NIRS monitoring of the brain and spinal cord makes it possible to identify the onset of potentially remedial adverse events (e.g. the onset of ischemia) in real time [173]. A threshold for cerebral ischemia has been defined using SRS [2], and the majority of the work related to the redox status of cytochrome has been done on brain [288]. Brain mapping of functional activity of the human cortex and imaging techniques are recent extensions of the use of NIRS. Two-dimensional brain imaging was described by van Houten [315]. Reviews speak to the feasibility and scope of mapping, the potential for effective imaging using NIRS, and the instrumentation required [124,218]. Technology and software advances [6,16,136] are increasing the quality and scope of mapping and offering three-dimensional imaging [124,125]. fNIRS is seen as means of providing a more complete temporal picture of brain hemodynamics than fMRI because the temporal resolution of Hb detection is not acquisition limited and can be much faster than the hemodynamic response itself, although limited by spatial sensi-
376
tivity and depth penetration [134]. Reviews from 20058 include the consensus of a workshop on the potential utility of fNIRS for studies of brain activity underlying cognitive processing in human infants [8]; the evolution of the technology; and advantages and limitations of fNIRS and optical tomography [10,37,130,132]. Spatial localization of NIRS signals will never achieve the precision of fMRI-BOLD, [134], but simultaneous NIRS and fMRI-BOLD measurement of cortical oxygenation showed good correlation between the methods in young and elderly subjects in a study that demonstrated an effect of aging [191], and used in conjunction with EEG/ERP techniques NIRS is emerging as a third window into the infant brain [8]. A wealth of information is available from the application of this technology to study infant perception, speech and cognition [26], and responses are described to a range of visual, olfactory and auditory stimuli and passive movements in infants [12,13,132,289]. In adults multiple fNIRS applications are described including novel uses in psychiatric evaluation [37], signal acquisition for support of a brain-computer interface [60], and growing numbers of multimodal studies [37,134]. Muscle: One of the principal roles of NIRS is now the study of muscle. Basic research in animals [248] and humans [92] has led to significant applications in exercise and sports science where tissue metabolism can be evaluated during exercise [30,93,119,204,229,242,243,334]. The majority of studies have been non-invasive and the parameters measured include those conventionally accessible via NIRS, with the addition of muscle SvO2, muscle tHb, muscle blood flow, muscle Hb, muscle VO2, muscle recovery time, muscle compliance and tissue oxygen saturation [334]. Many studies assess the balance between oxygen delivery and tissue oxygen consumption using the ratio of O2Hb to tHb as an index of oxygen saturation, or evaluate how well this measure relates to the oxygen saturation of venous blood. Review of studies during dynamic exercise [23] indicates that NIRS can be used to provide data on muscle oxygen saturation and blood volume, and define the thresholds for lactate (ventilator) and respiratory compensation during incremental exercise. During constant work-rate exercise, changes in whole body oxygen uptake kinetics can be detected above the lactate/ventilator threshold. Significant age-related effects are seen on the degree of muscle deoxygenation at the same absolute oxygen uptake. But no significant difference is evident in the rate of deoxygenation in the arm (biceps) and leg (vastus lateralis) during exercise studies between males and females. NIRS studies demonstrate substantial aerobic metabolism with such exercise, and are able to evaluate the effects of training, as well as sustained exercise. There is general agreement that NIRS is an appropriate measure for local muscle VO2 and can be used to quantify blood flow using non-invasive and invasive (ICG dye) techniques [30,307]. When using ICG dye as a tracer, blood flow values are derived from the ratio of ICG in the tissue compared to arterial blood. Muscle VO 2 can be calculated from the rate of conversion of O2Hb to HHb during induced ischemia in the arm or leg. It can also be measured both at rest and during maximum voluntary contraction, either with or without vascular occlusion [93,252]. Yoxall [349] described a method to estimate muscle venous saturation (SvO2) by applying venous occlusion and measuring changes in O2Hb versus tHb. Relying on the principal that the blood volume increase during occlusion represents the accumulation of venous blood, plots of oxygen saturation as a function of tHb allow derivation of venous oxygen saturation. Good correlation was demonstrated between this method and SvO2 values derived by a more recent methodology based on respiratory induced oscillations of the absorption of NIR
377
light [100]. The effects on muscle of ischemia and hyperemia have been extensively researched. The rate of oxygen desaturation reflects oxygen consumption by the tissue, provided oxygen delivery is interrupted. Variations in the technique allow the effects of various aspects of exercise on metabolic rate to be determined including the effects of fitness level, training regiments, and the ability of muscles to recover. Key aspects of these NIRS measurements correlate with NMR studies of muscle metabolism [30]. While the majority of studies support the validity of these measures [23,30,204] it is important to recognize that there are differences between regional and systemic measurement and also that there is heterogeneity within and between specific muscles [30,55,243]. In addition, correlation evident in studies at rest may not be evident during exercise. At rest SaO2 is usually stable and the distribution of blood between arteries, capillaries and veins is constant. But during exercise oxygen consumption increases and muscle contraction and capillary recruitment alter the distribution of blood. With an increase in arterial blood supply tissue oxygen utilization may be exceeded in the early stages of exercise so oxygen saturation may increase, and higher blood flow can result in tissue oxygen saturation remaining unchanged during exercise. Studies of blood flow and oxygen saturation during exercise are not limited to muscle; a small series relate to connective tissue such as the Achilles tendon [28,30]. There is also clinical evidence from NIRS assessment of pathologic states involving muscle blood supply and metabolism of the beneficial effects of surgery in peripheral vascular disease [83]; the negative effects of heart failure on muscle oxygenation, and new insights into mitochondrial dysfunction. In chronic fatigue syndrome an impairment of oxygen delivery has been demonstrated using NIRS [186]. Findings from recent studies in health and disease are reviewed [30, 93, 119] and individual studies include: Lower muscle VO2 and blood flow in the forearms of individuals with repetitive strain injury compared to controls at similar working intensities, indicating that the underlying vasculature may be impaired [35]; Muscle blood volume and oxygenation changes in the erector spinae muscle of healthy subjects and those with low back pain (active and sedentary) during BieringSorensen muscle endurance testing suggest that factors other than erector spinae aerobic capacity influence performance [145]; Reduced capillary volume expansion in diabetics during a plantar-flexion and treadmill-walking exercise regimen even in the absence of peripheral vascular disease, likely due to impaired vasodilation secondary to endothelial dysfunction [197]; Better aerobic fitness and respiratory muscle oxygenation in children with congenital heart disease following general physical training at sub-maximal intensity [196]; Diagnostic potential for NIRS in acute [84,260] and chronic compartment syndrome [313]; Confirmation of central (cardiovascular) and peripheral (skeletal muscle) impairment of oxidative metabolism in heart transplant recipients and the value of serial evaluation with NIRS during rehabilitation [158].
The ability to make comparison between sites of measurement and conduct studies in real time during exercise is an advantage of the NIRS technique valued by investigators. Reviews [204, 243] summarize the broad range of activities studied in sports sci-
378
ence that have utilized NIRS. These include: alpine skiing, speed skating, cycling, weight lifting, treadmill, running, and simulated cycling, rowing, and skiing. It appears probable that NIRS applications in such studies will continue to contribute valuable knowledge regarding muscle physiology in health and disease. Neonates: The relevance and benefit of real time detection of significant hypoxic or ischemic events in newborn infants are universally recognized [126], as is the vulnerability of this population [336]. Not surprisingly NIRS has been studied extensively in animal and human models as a means of monitoring cerebral hemodynamics in premature infants [1,31,32,81,248,328,338,339]. The evolution and scope of NIRS research involving neonates has been the subject of recent reviews [22,116,209,248,336], and others [8,10] describe applications of fNIRS to brain mapping. Brazy reported the first observations in human neonates [31]. Early studies primarily aimed at establishing normative values produced very different results [248,336], but it became clear later that in neonates hemodynamic variables alter significantly with gestational age, patient activity and treatment variables [1,167,322]. Wyatt and his group contributed early methods to quantify NIRS indices [338]. Methodologies to determine cerebral blood flow (oxygen marker method) [81] and cerebral blood volume [339] (oxyhemoglobin indicator dilution technique) increased the scope of NIRS studies, but there were difficulties in applying these methods in practice [116,248] and wide variation in the values reported. While some of this variation can be explained on the basis of different physiologic effects influencing individual populations of the neonates studied (e.g. cyclical fluctuations in blood pressure and heart rate [322], and CBV [167,322]), concern was expressed over cerebral blood flow measurement in particular. Later, methodological refinements resulted in improvement. Changes in CBV generated by jugular venous occlusion correlate well with strain gauge plethysmography [330], and CBF estimation using indocyanine green injection as the required marker resolved most concerns [33,246]. Acceptable agreement between NIRS generated CBF and 133Xe clearance values was demonstrated [36,262], but the technique has found scarce clinical application [93]. NIRS peripheral venous oxyhemoglobin saturation can be calculated in newborns using a venous occlusion method [347]. Peripheral SvO2 values (venous occlusion method) have a good correlation after bias adjusting (r = 0.96, p > 0.05) with measurements of central SvO2 by co-oximetry. Reproducibility of NIRS was high with a test-retest variation of 2.51 +/ 1.41%, suggesting that NIRS is practical for monitoring relative changes in central venous saturation [14]. Cerebral venous oxygen saturation (cSvO2) measured with partial jugular venous occlusion [348] produces similar values to invasive measurements of SvO2 (jugular bulb blood cooximetry). Cerebral fractional oxygen extraction represents the balance between cerebral oxygen delivery and consumption and can be calculated from SaO2 (pulse oximetry) and cSvO2 measured via NIRS with partial jugular venous occlusion [326]. The physiologic effects of a range of clinical issues and pharmacological agents have been investigated using NIRS. In asphyxiated infants important effects on CBF, CBV, and cerebral tissue oxygen extraction have been detected during the first 24 hours of life, and specific patterns related to more severe brain injury [190,299]. Intrapartum hypoxic ischemic injury is followed by significant disruption of oxidative metabolism in the hours after birth [11], CBV decreases [310], as does the effective extraction of oxygen. NIRS data point to the importance of therapeutic measures to minimize secondary injury from inadequate brain oxygenation and blood flow, and the probability of an effective window for therapeutic intervention prior to the secondary
379
energy failure that occurs at 812 hours post injury [117,310]. NIRS data reflecting impaired cerebral metabolism are paralleled by MRI and H-MRS data indicating the relationship between the extent of such derangement and severity of neurological outcome [250] and the contributory effect of seizures to the eventual degree of brain injury after HI [192]. In animal models NIRS studies confirm the evolving loss of CCO after severe HI [47] and also demonstrate a close correlation between CCO and depletion of high energy metabolites on MRS [47,228,302]. Apnea does not have an effect on cerebral blood volume in itself, but when combined with others stressors such as bradycardia CBV [224], CBV and oxygen supply [234], and tHb [166] tend to be adversely affected. In preterm infants CBV decreases with continuous negative extra-thoracic pressure and intermittent positive pressure ventilation [223]. Aminophyline and indomethacin have negative effects on CBV and CBF, caffeine and ibruprophen do not [336]. NIRS has been used as a screening tool for patent ductus arteriosus in extremely low birth weight infants [305], and a NIRS derived cerebro-splanchnic oxygenation ratio evaluated for detection of gut ischemia [98]. Studies evaluating cerebral auto-regulation demonstrate an adverse effect on CBF when mean arterial pressure is less than 30mmHg [201] and in the context of respiratory distress syndrome, impairment of cerebral autoregulation may make sicker infants more vulnerable to cerebral damage [160]. Studies on the impact of interventions such as suctioning, surfactant administration, lumbar puncture and heel prick blood sampling report a range of effects on CBV, O2Hb and CCO [104,248,336]. Brain TOI and rSO2 measurements have been reported in a variety of circumstances, current reviews question the level of precision of these measures [116]. In 20 stable infants studied with simultaneous measures of right and left optode placement, and sensor placement exchange TOI values were not reproducible under clinical conditions. Bland-Altman bias analysis was particularly poor for sensor exchange experiments but poor agreement was also seen though to a lesser extent during simultaneous left and right measurements [74]. Functional brain imaging with a multichannel NIR optical topography techniques (e.g. Hitachi Medical, and Artinis Medical Systems BV) can detect event-related changes in O2Hb in localized areas of the cerebral cortex in response to brain activation [132,289]. The improved spatial resolution and higher temporal resolution achieved has been used to study spontaneous changes in sleeping infants, and various effects in awake infants including those in response to visual [289], olfactory [12] and noxious [13] stimuli. One area where NIRS and fNIRS would likely add important physiologic data is in the study of pain in infants, particularly those who are extremely premature. NIRS monitoring would also likely contribute data that would aid in the management of infants and children requiring periods of support on extra-corporeal membrane oxygenation (ECHMO) [20]. The effects of asphyxia and the impact of therapeutic interventions causing noxious stimuli are important to clinicians, as a better understanding of their physiologic consequences would allow logical intervention, with perhaps a better long-term neurological outcome. For this reason NIRS remains an important research tool for investigating the complex interrelationships amongst physiologic and pathologic conditions that contribute to brain injury in sick newborn infants [1,19,117,299,336], and particularly the latent period between injury and secondary energy failure that offers a window for therapeutic intervention [117, 310], even though for the clinician NIRS is not able at present to contribute effectively to decision making or the management of individual infants [116, 209].
380
Obstetrics; fetus and placenta: Fetal surveillance during labour is the cornerstone of obstetric assessment of fetal well-being. Methods include fetal heart rate analysis, ECG, scalp pH, scalp PO2 and pulse oximetry [164]. Technical challenges for using NIRS for fetal surveillance include transmission of sufficient light through the maternal abdomen to monitor placental oxygenation, and optode design and placement able to accommodate movement of the fetal head down the birth canal. Following experiments by Schmidt [248,259] Peebles achieved intrapartum measurement of changes in human fetal cerebral hemoglobin in 8 fetuses during uterine contractions [225]. A specialized fetal probe was developed [226]; this maintained two glass prisms at a constant interoptode separation in a black silicone rubber molding which also shielded the optodes from reflected light from the myometrium, and could be apposed to the fetal scalp within the birth canal. A series of studies followed [226], and other investigators [4] found subsequently that the falls in tHb and O2Hb described during contractions were more profound where there were associated late decelerations in fetal heart rate. Although small changes in the probes position during measurement do not lead to major artifacts [226], the potential for movement artifact remains a technical challenge [248]; intensity-modulated optical spectrometry has been explored as a means of addressing this via measurement of the phase shift between light entering and exiting the head and provision of optical pathlength and light attenuation data [226]. Electronic fetal heart rate (FHR) monitoring is the gold standard method for detecting changes indicative of non-reassuring fetal oxygenation status, although the specificity of pathological alterations is poor [226]. Consequently, FHR monitoring and has now been trialed successfully in combination with supplemental fetal pulse oximetry using transabdominal CW NIRS technology, and the NIRS device obtained measurements comparable to transvaginal fetal pulse oximetry [319]. The feasibility of trans-abdominal monitoring of placental oxygenation using a NIRO 300 with a modified probe [142], and with TOI [144] has also been reported recently. Animal work with NIRS has also added important knowledge re fetal and newborn cerebral physiology and hemodynamics. Studies in fetal sheep suggest that fetal cerebral metabolism slows during hypoxia [207], and Bennet et al. [19] recently added to their earlier work with NIRS which showed that hypo-perfusion and reduced oxygen delivery occur post asphyxia [18], by showing that there is progressive mitochondrial failure and evolving impairment of cerebral oxygen metabolism following profound asphyxia. Also that this effect is preceded by a latent period. Translated to humans this latent period prior to secondary energy failure likely represents an important window of therapeutic opportunity. Peebles has observed deleterious effects of E coli lipopolysaccharide on cerebral oxygenation in fetal sheep, likely due to endotoxin mediated vasodilatation [227]. Maternal studies during elective caesarian section under low dose spinal anaesthesia show a relationship between cerebral oxygenation (ScO2) and impending hypotension, and the authors identify an intervention end point for immediate stabilization of blood pressure as a decrease of >5% in NIRS determined cerebral oxygenation. [21]. Trauma: NIRS applications to monitor oxygen transport in trauma patients have evolved because clinicians recognize the limitations of physical examination when assessing the evolution of shock or the adequacy of resuscitation [2,51,238,240,325]. Shock is a complex process involving simultaneous neurovascular, metabolic and inflammatory responses [240]. The potential for NIRS to aid in shock diagnosis and management has been evaluated in several organ systems in animal and human studies
381
[163,325] Ward, principally via assessment of hemoglobin oxygen saturation within tissue (StO2) as an indicator of shock severity [62,63,51,236,238]. In studies in animal models of hemorrhagic shock and resuscitation NIRS parameters showed sensitivity in detecting skeletal muscle and visceral ischemia [15,62,239,244]. Invasive NIRS identified that gastric tissue oxygenation fell with hemorrhage but recovered during resuscitation [50,316], and that NIRS derived smallbowel pH may be a means of monitoring the adequacy of resuscitation [239]. When CCO redox status is also measured in organs not containing myoglobin, changes in StO2 occur first in response to reduction in oxygen delivery, and are restored last when oxygenation improves [244]. In humans, NIRS derived StO2 of the skin and deltoid muscle was compared to other endpoints during resuscitation from shock, and showed a good relationship to systemic O2 delivery and lactate during and after resuscitation [188]. Crookes [63] first obtained normal values in 707 healthy adult volunteers; the average St0 2 value was 87 +/ 6 %, although the range was broad (5097%). 145 patients with shock classified as nil, moderate or severe were then evaluated using Sto2 from the thenar eminence. Based on receiver operator curve (ROC) comparison of StO2, maximum heart rate, base deficit and minimum systolic blood pressure, the latter outperformed all other indictors, followed by StO2. After severe torso trauma NIRS-derived thenar muscle StO2 measurements (inSpectra, Hutchinson Tech. Inc) performed similarly to blood base deficit (acidity) values as a measure of poor perfusion and potential to develop multiple organ dysfunction syndrome or death [57]. Using the same NIRS technique to monitor patients undergoing surgery on cardiopulmonary bypass Sto2 was found to respond significantly to regional changes in oxygen delivery, and identify perfusion deficits earlier than lactate or base deficit [238]. In patients with severe left heart failure StO 2 values tracked SVO2 except in cases where there was associated severe sepsis or septic shock [236]. Clinicians using NIRS in this context of shock have identified the non-invasive nature and continuous measurement capability of the technique as advantageous. These studies provide an important foundation for further research as StO 2 is as good a predictor of multiple organ dysfunction syndrome in trauma patients as base deficit [240] as it reflects the effect of the vasoconstriction which follows decreased cardiac output in hypovolemic shock [240]. However, the literature to date also raises important questions. The relationship of tissue oxygen saturation, venous oxygen saturation, oxygen delivery and oxygen consumption is dynamic and multidirectional [240,325], and can also be disrupted by any degree of mitochondrial failure. Consequently, although in many situations StO2 can mirror important change in other parameters, in a clinical context is unlikely to prove a specific marker alone, and should not be relied on for early monitoring of tissue hypoperfusion [63]. Other NIRS parameters, alone or in combination, can provide valuable assessment of a patients physiologic reserve and the metabolic activity of muscle e.g. the response to brief forearm ischemia from which the rate of reoxygenation can be measured [240], changes in CO redox status, and in the context of hypotension, the observation that TOI and tHb fall in patients with autonomic dysfunction when a fall in blood pressure is generated by head up tilt is also of interest [133]. Abdominal and limb compartment syndrome [84,260] are other major clinical problems studied. NIRS data from a gastric probe detected changes in mesenteric oxygen saturation in swine before the onset of anaerobic respiration [316]. In a comparison of affected limbs and uninjured limbs in humans pre and post fasciotomy an obvious
382
treatment effect was often seen, but was not evident in all cases, and standard deviations were large [109]. NIRS derived StO2 correlated with increases in compartment pressure in volunteers [108]. In NIRS studies of many conditions it more likely that physiologic variation due the causal etiology will be recognized if comparison data are collected simultaneously from a reference data channel monitoring tissue unaffected in the disease process. Similarly, following multiple NIRS parameters e.g. including changes in CCO redox status can be of value. The stability of the redox status of CCO most probably correlates with the adequacy tissue perfusion and the balance of oxygen supply and demand. Evidence for this comes from various studies [15,39,47,228,244,302] and data from human and animal studies are in keeping with there being an endpoint beyond which ongoing deterioration of CCO redox status is a predictor of mortality if left uncorrected. However, factors other than an imbalance in oxygen supply and demand undoubtedly influence CCO redox status, so that changes must be interpreted with care. Such factors include substrate and energy depletion that impacts the respiratory chain, and global cytopathic hypoxia [95] where mitochondrial dysfunction probably occurs due to systemic cytokine or endotoxin mediated damage [97,256,264]. The ability to measure CBF following trauma has relevance but in trauma cases NIRS is rarely used clinically. As with other technology methods need to be developed that allow either identification of thresholds for critically low or high CBF in individual patients, allow monitoring of oxygen extraction, or provide a measure of the volume of ischemic or hyperemic brain [274]. Trauma and critical care are areas where clinicians are making definite progress with NIRS, particularly via muscle monitoring of the adequacy of perfusion and oxygenation. As with most applications more research and a better understanding of the physiology underlying the changes in the NIRS signals measured are required. Values obtained from baseline measurements remain broad and specific resuscitation endpoint targets are elusive. The ability to detect the onset of ischemia in the spinal cord in real time via NIRS monitoring [172] has yet to be translated into clinical use. Potential future applications of NIRS and related spectroscopic techniques to critical care include non-invasive measurement of tissue pH [267,342], and glucose (diffuse reflectance) [178], and early detection of severe pancreatitis (scanning spectroscopy of serum) [230]. Surgery: Many surgical procedures and the associated anaesthesia required have been monitored using NIRS with the premise that morbidity and mortality from various complications could be reduced. Monitoring cerebral oxygenation status during prolonged anaesthesia is widely practiced [80,168,222]. Major cardiac surgery and aortic aneurysm repair requiring extracorporeal circulation (bypass) are associated with high rates of neurologic damage [94,110,206], and a compelling need has been demonstrated in multi-center studies for improved cerebral surveillance during such surgery [245]. Much of this damage occurs because of hypoxic ischemic injury, which results in an imbalance between neuronal energy supply and demand [58]. For cardiac surgery on bypass, pump flow rates are calculated from a formula that approximates cerebral blood flow to body surface area, and it is recognized that patients may be underperfused over-perfused as a result. Both situations can contribute to morbidity. Early studies used CW instruments to monitor cerebral oxygen supply and utilization via O2Hb, Hb and tHb in both adults and children during a range of cardiac surgeries, and explore measurement of cerebral blood flow [88] and the redox status of CCO.
383
NIRS use in cardiovascular surgery has been reviewed [115,212,213]. NIRS, including phase-resolved spectroscopy [344], has been used in humans and animal models to study the effects of circulatory intervention including retrograde perfusion [70,107], hypothermia and pH strategy [251,293], aortic aneurysm repair [252] complications of surgery [173,215], neuronal death after hypothermic circulatory arrest [157], tracheal extubation [198], and monitor O2Hb and HHb, and sometimes CCO, during repair of cyanotic and non-cyanotic cardiac defects, the aortic arch and vena cava [73,107,150, 173,199,247,291,344]. Although there were important differences in these early observations, as intraoperative monitoring evolved its relevance and potential were always emphasized. Intraoperative monitoring during cardiac surgery has yielded important data using conventional NIRS data and the cytochrome signal in particular. During cardiac bypass stability of the redox status of the CCO signal most probably correlates with adequacy of pump flow and the maintenance of cerebral blood flow, and a progressive fall, if uncorrected, correlates with mortality. Because of the relevance of stability of this parameter as an indicator of the balance between cerebral oxygen supply and demand changes in CCO redox status have been measured in a variety of human and animal studies [73,106,140,263]. Although concern has been raised about the validity of this measure, recent studies support the genuine nature of the cytochrome signal [288]. Animal studies also indicate that NIRS can detect the onset of ischemia in the spinal cord in real time which has relevance in spinal cord surgery and repair of aortic aneurism [172]. Concealed hemorrhage and pump flow problems during cardiac surgery can be identified via ischemic patterns of change in chromophore concentration, and other recognizable patterns are seen with defined physiologic events such as sub optimal anaesthesia and atrial fibrillation (when cardiac output falls by 40%) [173]. Special challenges exist when using NIRS during cardiac surgery. In addition to well recognized questions such as whether or not brain tissue is being selectively monitored, and the meaning of measures obtained, challenges include: the effects of the fluctuation in Hb concentration that occurs due to hemodilution; blood loss and transfusion; and the effects of deep hypothermia on NIRS [159,293]. It was suggested that NIRS monitoring of O2Hb, HHb and CCO should become a standard of care for complex procedures on bypass to avoid over or under perfusion of the brain during circulatory arrest [115,173,247]. Many would still agree with this concept. Some advocate rSo2 [79,127] others are opposed [67,200]. The majority do support further refinement of instrumentation and more prospective evaluative research. To this end investigators continue to conduct cerebral monitoring with NIRS during cardiac surgery [129,189,199,251,290,301]. Monitoring is often done in conjunction with other technologies [129,189]. A novel depth-resolved technique has been described using a frequency domain NIRS system (ISS Inc.) and intravenous indocyanine green boluses to evaluate cerebral hemodynamics during bypass. The advantage is greater discrimination between superficial and deep tissue layers [275]. Recently the majority of intraoperative monitoring has been done with spatially resolved instruments with the objective of quantifying TOI or rSO2 data, but the extensive literature now available demonstrates genuine concern with the accuracy and validity of this measurement. Animal and human studies have also shown the dependence of rSO2 values on hemoglobin concentration and mean arterial pressure [343]. In recent reviews studies predominate that monitor rSO2 derived by the USA FDA approved INVOS system or TOI from NIRO (Hamamatsu) instruments. Several such studies are considered together in a systematic review of 48 papers describing 5931 patients (coro-
384
nary artery bypass graft >83%) which also addresses the theoretical capability of cerebral oximetry to monitor changes in brain tissue rather than overlying structures [290]. Some bias was identified in the citations selected [214], and the studies compared used instruments recognized to employ different methods for both data collection and data management [200]. The majority were prospective but major methodological limitations were identified, and although NIRS appeared to detect brain desaturation episodes encountered during surgery the level of evidence was low [290]. Opinion is divided regarding the pros [79,127] and cons [67,200] of this form of NIRS monitoring during complex cardiac surgery. A 2007 study of 25 newborns monitored during major surgery (NIRO 300 TOI and chromophore concentration) concluded that the changes in cerebral oxygenation and hemodynamics observed allowed real time evaluation of intraoperative effects [249]. However, brain TOI measurements (NIRO 300 oximeter) made in 20 stable infants with simultaneous measures of right and left optode placement, and with sensor exchange placement were not reproducible under clinical conditions, and poor agreement was seen in Bland-Altman bias analysis. This was particularly poor for sensor exchange experiments but was also seen to a lesser extent during simultaneous left and right measurements [74]. NIRS use in neurosurgery has been reviewed [176]; and monitoring reported during temporal lobectomy, resection of anaplastic astrocytomas and glioblastomas [7], and stereotactic surgery for movement disorders [176]. NIRS use has been reported in vascular and plastic surgery during vascular grafting [83]; carotid endarterectomy [2,154]; and abdominal aortic aneurysmectomy [252,101]. Also, to evaluate burn patients [64,269,270]; the viability of skin flaps [114,258]; the viability of tissue in transplant organs [295]; and during orthoptic liver transplantation [235]. NIRS has also been used to evaluate intra-operative blood transfusion [300], laparoscopic cholecystectomy [107], and compartment syndrome [108,109,260,298,313]. In chronic compartment syndrome the sensitivity of noninvasive NIRS was clinically equivalent to that of invasive intra-compartmental pressure measurements [313]. NIRS is not reliable for postoperative hematoma detection or monitoring intracranial status in patients after craniotomy [139]. In the work-up of patients with leg ischemia NIRS measures of tissue oxygenation were very reproducible but had no correlation with other micro or macro-circulatory parameters [304]. Kidney, testis and Bladder: NIRS has only begun to be used to evaluate these organs and clinical urologic conditions recently. Animal and human studies include: testicular ischemic conditions [53,41], erectile dysfunction [38], and renal dysfunction [155,231]. In addition, NIRS has been used to evaluate skeletal muscle metabolism in patients with end stage renal disease [148,182,317]. Urinary incontinence and bladder dysfunction are conditions that negatively affect the quality of life of millions of people [276]; their evaluation currently relies on urodynamic assessment of the bladder (uroflowmetry, filling cystometry and pressure-flow studies) [257] which involves invasive urethral and rectal catheterization and provides no direct physiologic measurement. Recent non-invasive NIRS studies have begun to asses the physiology and pathology of bladder dysfunction [174]. NIRS was first applied in urology by Colier et al. [53] in combination with pulse oximetry to measure the blood supply to intra abdominal testes in an animal model of cryptorchidism. The results could quantify active testicular blood volume which equated to testicular viability. Later, the feasibility of sensitive detection of acute tes-
385
ticular hypoxia following testicular torsion was shown using NIRS in a sheep model [41], and reperfusion of the hypoxic testis was demonstrated after the torsion was reduced. Penile blood volume changes and their time course were studied in men with erectile dysfunction and normal volunteers using a customized NIRS probe (wavelength selectivity of 805 nm for hemoglobin absorption) and simultaneous color duplex ultrasonography [38]. The findings suggest that NIRS can produce diagnostic ranges that identify non-vasculogenic to severe vasculogenic causes of erectile dysfunction, which may predict benefit from pharmacological treatments. When trans-abdominal NIRS was first done simultaneously during invasive urodynamic evaluation of the human bladder [174] filling and voiding studies demonstrated a significant correlation between changes in O2Hb and HHb concentration in the detrusor muscle with bladder contraction during voiding. Subsequent studies have explored using the patterns of chromophore change generated by NIRS during natural filling and/or voiding to distinguish between specific urinary pathologies [277281]. A prospective study evaluated men with lower urinary tract symptoms using simultaneous NIRS and conventional urodynamic testing. Data from 57 subjects showed that NIRS patterns of chromophore change plus measurements of post voiding residual volume (PVR) and peak uroflow rate (Qmax) could be used to distinguish between those with and without obstruction [175], and that this could be done with good sensitivity and specificity. This finding was validated in an independent study. Subsequently classification and regression tree analysis (CART) was used to develop an algorithm that achieved comparable discriminant ability when only the non-invasive NIRS data during voiding were used. These studies introduce the prospect of non-invasive screening of patients with lower urinary tract symptoms [282]. It is also feasible using a multichannel array (fNIRS) to transcutaneously map dynamic change of detrusor hemodynamics during voiding [283]. Also to use a vaginal probe incorporating NIRS optodes to interrogate the posterior wall of the bladder, the urethral sphincter and evaluate the muscle of the pelvic floor [261]. As with other applications of NIRS confidence is required that transcutaneous bladder monitoring does interrogate the tissue of interest and reflect physiologic change, also that movement does not negatively impact the ability of the data derived to contribute valid information. Such confidence is accumulating. Review of animal and human data indicates that: the principles of photon penetration into tissue make interrogation of the anterior bladder wall via transcutaneous sensors feasible without significant attenuation in subcutaneous tissue [29,93,334]; changes in chromophore concentration are only detected in strict temporal relationship to physiologic events during voiding, and by sensors over the bladder [282]; NIRS of the bladder detects changes comparable to those seen in other tissues in response to altered oxygen demand in the microcirculation [93,229,304,334]; non-invasive NIRS data have comparable discriminant ability to other invasive bladder diagnostic methodology [175]; and significant changes in NIRS parameters occur when bladder volume remains constant. Thus changes in detrusor chromophore concentration do offer the potential for new knowledge and non-invasive evaluation of patients with bladder pathology. The effect of hypoxic episodes on cerebral and renal tissue oxygenation was evaluated in ventilated preterm neonates [231]. Episodes of decreased cerebral and renal tissue oxygenation (SaO2 7080%) were not associated with compromise of cerebral oxygen utilization but increases in oxygen extraction in renal tissue did occur. NIRS monitoring of calf muscle oxygenation and perfusion studied in patients with chronic renal failure (CRF) on hemodialysis (HD) [148] showed an effect of muscle
386
mass on dialysis efficiency and muscle dysfunction likely related to a mitochondrial defect. NIRS monitoring of O2Hb and HHb and Mb (myoglobin) concentration in calf muscle plus MRI was used to demonstrate a lack of effect from L-carnitine supplementation on muscle bioenergetics and function in CRF patients [317]. Skeletal muscle oxidative metabolism in the forearm of children with end stage renal disease (ESRD) was measured before and after renal transplantation [182]. NIRS was used to monitor alternations in Hb/Mb deoxygenation during arterial occlusion as an indicator of the rate of oxygen consumption in mitochondria, and recovery time as an indicator of muscle aerobic capacity following a hand-grip exercise. Oxidative metabolism in skeletal muscle during exercise was impaired and improved remarkably after renal transplantation. The investigators recommended NIRS as a useful method to monitor muscle metabolism in children with ERSD. Renal tolerance to contrast agents was assessed via a small NIRS probe placed on the renal cortex of rats. The alteration in tissue oxygen saturation following injection of iodinated contrast media with and without the addition of a prostacyclin analog (iloprost) was measured [155]. Development of less toxic contrast media has long been a subject of research in uroradiology [143]. The latest extensions of time-resolved NIRS technology are being explored experimentally to examine the optical properties of prostatic tissue in vivo in the context of developing photodynamic therapy as a modality for treatment of prostatic cancer [286]. Summary NIRS is a unique and versatile technology that makes non invasive monitoring of chromophore change in a variety of tissue possible in health and disease. Continuous wave spectrometers have good temporal resolution, and are widely employed in research to provide a semi-quantitative measurement of oxygenation and changes in blood volume in tissue via change from baseline in chromophore concentration derive from oxygenated and deoxygenated haemoglobin. NIRS instruments of different types have a variety of proven advantages. The limitations of applying NIRS technology to study human tissue in vivo are well recognized. By using measures that have been validated through basic science and human experiments investigators continue to learn and report new findings, and explore new avenues for the application of this technology to answer physiologic questions of relevance that are not readily addressed by other means. The translation of NIRS from a research tool to a clinical monitoring or diagnostic entity has been widely predicted but in this regard NIRS has not met its potential. This is due in part to the recognized limitations in the technique, particularly the unknown degree of photon scattering and the absence of absolute quantification of haemoglobin concentration. However there has been the tendency to report clinical feasibility and describe the potential for NIRS to contribute clinically and then silence with regard to the necessary follow up with prospective trials to validate the new applications and confirm the reproducibility of the data derived. In a clinical context this translates into a dilemma for clinicians, as although trend monitoring in an individual patient may be of value, the absence of quantification and defined NIRS end points that equate with defined physiologic parameters and outcome leaves too much unknown for critical decisions to be made routinely based on NIRS data. In addition clinicians rightly want reproducible and robust data when evaluating patient status and are rightly concerned
387
that a number of the NIRS measures that have been described are currently unable to deliver in this regard. This applies particularly to attempted measurement of venous oxygen saturation via TOI/rSo2. There are however analogies to be drawn between the evolution of oximetry as a measure of SaO2 and NIRS. Oximetry languished for years as a cumbersome or imprecise measure until key breakthroughs were made that resulted in improvements in the hardware available and in how the software was written so that data of relevance to clinicians was presented numerically in real time. So well did these improvements meet clinical need that oximetry is now the fifth vital sign in spite of the fact that the instruments are only accurate and reflective of real physiologic change over a very narrow range of values. NIRS hardware and software continue to be refined with many basic scientists and investigators making important advances. But funding for this research, especially the essential clinical components is not easy to come by as NIRS falls outside most conventional funding channels. This limitation exists in spite of the continuing interest by clinicians in the potential of NIRS and the huge impact advances in this form of monitoring could make in critical areas of clinical care. Importantly many established investigators retain their interest in NIRS and demand for NIRS instruments continues, particularly for applications by basic scientists. Interest continues to grow within the field of sports medicine and amongst muscle physiologists, and those interested in brain mapping. An increasing number of multiple (824) channels instruments designed for mapping are now available. The prospect of increasingly detailed spatial information on cerebral oxygenation becoming available at the bedside is real. New NIRS applications continue in familiar fields like neonatology, surgery and critical care, and a NIRS module has been incorporated into an established commercial clinical diagnostic system (urodynamics) used widely in by urologists in order for simultaneous NIRS measurements to be made. Novel approaches to instrumentation and incorporation of sophisticated mathematical modeling into software are occurring which improve data display and data interpretation for diagnosis. In the quest for absolute quantification the value of recognizing reproducible patterns of chromophore change in real time has been eclipsed. The ability to detect the onset of physiologic changes such as ischemia, hypoxia and alteration of blood volume in real time already provides diagnostic opportunities, and makes intervention possible during monitoring prior to significant morbidity occurring. In addition, with the challenges that exist clinically in monitoring, imaging, diagnosis and management across the biomedical spectrum there have to be many clinically relevant applications that have not yet been explored adequately or at all. Consequently if scientists and clinicians go further, and collaborate to define the critical end points relevant to patient care, and do the due diligence required by conducting the basic science studies required and the prospective trials necessary to validate the measures examined, effective translation of NIRS into defined areas of clinical care will ultimately occur.
References
[1] L.M. Adcock, L.S. Wafelman, S. Hegemier, A.A. Moise, M.E. Speer, C.F. Contant and J. GoddardFeingold. Neonatal intensive care applications of near-infrared spectroscopy. Clin Perinatol. 26 (4) (1999), 893-903. [2] P.G. Al-Rawi, Near infrared spectroscopy in brain injury: todays perspective, Acta Neurochir Suppl 95 (2005) 453-7.
388
[3] P.G. Al-Rawi and P.J. Kirkpatrick. Tissue oxygen index: thresholds for cerebral ischemia using nearinfrared spectroscopy. Stroke. 37 (11) (2006), 2720-5. [4] C.J. Aldrich, D. DAntona, J.A.D. Spencer, D.T. Delpy, E.O.R. Reynolds and J.S. Wyatt. Fetal heart rate changes and cerebral oxygenation measured by near-infrared spectroscopy during the first stage of labour. Eur J Obs Gynecol and Rep Biol. 64 (1996), 189-195. [5] S.R. Arridge, M. Cope and D.T. Delpy. The theoretical basis for the determination of optical pathlengths in tissue: temporal and frequency analysis. Phys Med Biol 37 (7) (1992), 1531-60. [6] S.R. Arridge, M. Schweiger. Image reconstruction in optical tomography. Phil Trans R Soc Lond B. 352 (1354) (1997), 717-26. [7] S. Asgari, H.J.Rohrborn, T. Engelhorn, and D. Stolke, Intra-operative characterization of gliomas by near-infrared spectroscopy: possible association with prognosis, Acta Neurochir (Wien) 145 (6) (2003), 453-59. [8] R.N. Aslin, and J. Mehler, Near-infrared spectroscopy for functional studies of brain activity in human infants: promise, prospects, and challenges. J Biomed Opt. 10 (1) (2005), 11009. [9] M. Attas, M. Hewko, J. Payette, T. Posthumus, M. Sowa, and H. Mantsch, Visualization of cutaneous hemoglobin oxygenation and skin hydration using near-infrared spectroscopic imaging, Skin Res Technol 1 (4) (2001), 238-45. [10] T. Austin. Optical imaging of the neonatal brain. Arch Dis Child Fetal Neonatal Ed. 92 (2007), F238-F241. [11] D. Azzopardi, J.S. Wyatt, E.B. Cady, D.T. Delpy, J.Baudin, A.L. Stewart, P.L. Hope, P.A. Hamilton, and E.O. Reynolds, Prognosis of newborn infants with hypoxic-ischemic injury assessed by phosphorus magnetic resonance spectroscopy. Pediatr Res. 25 (1989), 445-451. [12] M. Bartocci, J. Winberg, C. Ruggiero, L.L. Bergqvist, G. Serra and H. Largercrantz, Activation of the olfactory cortex in newborn infants after odor stimulation: a functional near-infrared spectroscopy study, Pediatr Res 48 (2000), 18-23. [13] M. Bartocci, L.L. Bergqvist, H. Lagercrantz and K.J. Anand, Pain activates cortical areas in the preterm newborn brain. Pain. 122 (1-2) (2006), 109-17. [14] R. Bay-Hansen, B. Elfving and G. Greisen. Use of near infrared spectroscopy for estimation of peripheral venous saturation in newborns: comparison with co-oximetry of central venous blood. Biol Neonate. 82 (1) (2002), 1-8. [15] G.J. Beilman, D. Myers, R.F.B. Cerra, V. Lazaron, R.A. Dahms, M.J. Conroy, and B.E. Hammer. Near-infrared and nuclear magnetic resonance spectroscopic assessment of tissue energetics in an isolated, perfused canine hind limb model of dysoxia. Shock. 15 (5) (2001), 392-7. [16] D.A. Benaron, S.R. Hintz, A. Villringer, D. Boas, A. Kleinchmidt, J. Frahm, C. Hirth, H. Obrig, J.C. van Houten, E.L. Kermit, W.F. Cheong and D.K. Stevenson. Non-invasive functional imaging of the human brain using light. J Cereb Blood Flow Metab. 20 (2000), 469-77. [17] R.E. Benesch, R. Benesch and S. Yung, Equations for the spectrophotometric analysis of hemoglobin mixtures, Anal Biochem 55 (1973), 245-248. [18] L. Bennet, S. Rossenrode, M.I. Gunning, P.D. Gluckman and A.J. Gunn. The cardiovascular and cerebrovascular responses of the immature fetal sheep to acute umbilical cord occlusion. J Physiol. 517 (1) (1999), 247-57. [19] L. Bennet, V. Roelfsema, P. Pathipati, J.S. Quaedackers and A.J. Gunn. Relationship between evolving epileptiform activity and delayed loss of mitochondrial activity after asphyxia measured by nearinfrared spectroscopy in preterm fetal sheep. J Physiol. 572 (1) (2006), 141-54. [20] P.B. Benni, B. Chen, F.D. Dykes, S.F. Wagoner, M. Heard, A.J. Tanner, T.L. Young, K. RaisBahrami, O. Rivera, and B.L. Short, Validation of the CAS neonatal NIRS system by monitoring vvECMO patients: Preliminary results, Adv. Exp. Med. Biol. 566 (2005), 195-201. [21] P.A. Berlac and Y.H. Rasmussen. Per-operative cerebral near-infrared spectroscopy (NIRS) predicts maternal hypotension during elective caesarean delivery in spinal anaesthesia. Int J Obst Anesth. 14 (1) (2005), 6-31. [22] G. Bernert, K. von Siebenthal, C. Kohlhauser and P. Casaer, near infrared spectroscopy: methodological principles and clinical application in preterm infants. Wiener Klinische Wochenschrift. 107 (19) (1995), 569-73. [23] Y.N. Bhambhani, Muscle oxygenation trends during dynamic exercise measured by near-infrared spectroscopy. Can J Appl Physiol. 29 (2004), 504-23. [24] T. Binzoni, V. Quaresima, M. Ferrari, E. Hiltbrand, and P. Cerretelli, Human calf microvascular compliance measured by near-infrared spectroscopy, J. Appl. Physiol. 88 (2) (2000), 369-372. [25] J. Boldt, Clinical review: hemodynamic monitoring in the intensive care unit. Crit Care. 6 (2002), 52-9. [26] H. Bortfeld, E. Wruck and D.A. Boas. Assessing infants cortical response to speech using near-infrared spectroscopy. Neuroimage. 34 (2007), 407-415
389
[27] G. Bottiroli and A.C. Croce, Autofluorescence spectroscopy of cells and tissues as a tool for biomedical diagnosis, Photochem Photobiol Sci 3 (11-12) (2004), 189-210. [28] R. Boushel, H. Langberg, J. Olesen, M. Nowak, L. Simonsen, J. Bulow, and M. Kjaer, Regional blood flow during exercise in humans measured by near-infrared spectroscopy and indocyanine green, J. Appl. Physiol. 89 (5) (2000), 1868-1878. [29] R. Boushel and C.A. Piantadosi, Near-infrared spectroscopy for monitoring muscle oxygenation, Acta Physiologica Scandinavica 168 (2000), 615-22. [30] R. Boushel, H. Langberg, J. Olesen, J.Gonzales-Alonzo, and J. Bulow, M. Kjaer. Monitoring tissue oxygen availability with near infrared spectroscopy (NIRS) in health and disease. Scand J Med Sci Sports 11 (2001), 213-22. [31] J.E. Brazy, D.V. Lewis, M.H. Mitnisk, and F.F. Jobsis-VanderVliet, Non-invasive monitoring of cerebral oxygenation in preterm infants: preliminary observation, Pediatrics 175 (1985), 217-25. [32] J.E. Brazy, Cerebral oxygen monitoring with near-infrared spectroscopy: clinical application to neonates. J Clin Mon. 7 (1991), 325-344. [33] D.W. Brown, P.A. Picot, J.G. Naeini, R. Springett, D.T. Delpy and T.Y. Lee. Quantitative near infrared spectroscopy measurement of cerebral hemodynamics in newborn piglets. Pediatr Res. 51 (2002), 564-570. [34] N. Bruce. Experimental study of the effect of absorbing and transmitting inclusions in highly light scattering media. Appl Optics. 33 (28) (1994), 6692-98. [35] J.J. Brunnekreef, J. Oosterhof, D.H. Thijssen, W.J.N. Colier, C.J. van Uden, and J.T. Carlo, Forearm blood flow and oxygen consumption in patients with bilateral repetitive strain injury measured by near infrared spectroscopy, Clin Physiol & Fun Imag. 26 (3) (2006), 178-184. [36] H.U. Bucher, A.M.Weindling, N.H.Dawani, and I.Peart, Comparison between near infrared spectroscopy and 131Xenon clearance for estimation of cerebral blood flow in critically ill preterm infants, Pediatr Res 33 (1993), 56-60. [37] S.C. Bunce, M. Izzetoglu, K. Izzetoglu, B. Onaral, and K. Pourrezaei, Functional near-infrared spectroscopy, IEEE Eng. Med. Biol. 25 (2006) 54-62. [38] A.L. Burnett, R.P. Allen, D.M. Davis, D.C.Wright, I.N.Trueheart and B.Chance, Near infrared spectrophotometry for the diagnosis of vasculogenic erectile dysfunction, Int J Impot Res 12 (2000), 247-54. [39] C.B. Cairns, F.A. Moore, J.B. Haenel, B.L. Gallea, J.P. Ortner, S.J. Rose and E.E. Moore. Evidence for early supply independent mitochondrial dysfunction in patients developing multiple organ failure after trauma, J Trauma. 42 (3) (1997), 532-6. [40] J.P. Caplan, S.Waxman, R.W.Nesto and J.E.Muller, Near-infrared spectroscopy for the detection of vulnerable coronary artery plaques, J Am Coll Cardiol 47 (2006), 92-96. [41] G.A. Capraro, T.J. Mader, B.F. Coughlin, C. Lovewell, M.R. St. Louis, M. Tirabassi and G. Wadie, Feasibility of using near-infrared spectroscopy to diagnose testicular torsion: an experimental study in sheep, Ann Emerg Med. 49 (2007), 520-5. [42] B. Chance The Biochemistry of Copper in, G. Peisach, P. Aiisen, W.E. Blumberg, eds. Academic Press, New York, 1966, pp. 293-301. [43] B. Chance, M. Maris, J. Sorge, and M.Z. Zhang, A phase modulation system for dual wavelength difference spectroscopy of hemoglobin deoxygenation in tissues in: Time Resolved Laser Spectroscopy in Biochemistry II, J. R. Lakowicz, ed., Proc. Soc. Photo-Opt. Instrum. Eng. (1990) 1204, 481-491. [44] B. Chance, M.T. Dait, C. Zhang, T. Hamaoka, and F. Hagerman, Recovery from exercise-induced desaturation in the quadriceps muscles of elite competitive rowers, Am. J. Physiol. 262 (3 Pt. 1) (1992), C766-C775. [45] B. Chance, Non-invasive approaches to tissue bioenergetics. Biochem Soc Trans. Nov 22 (4) (1994), 983-7. [46] B. Chance. Quantification of brain oxygenation in humans in vivo. Pediatr Crit Care Med 7 (1) (2006) 93. [47] Y.S. Chang, W.S. Park, M. Lee. K.S. Kim, S.M. Shin and J.H. Choi. Near infrared monitoring of secondary energy failure after transient global hypoxia-ischemia in the newborn piglet. Neurol Res 21 (1999), 216-224. [48] J. Choi, M. Wolf, V. Toronov, U. Wolf, C. Polzonetti, D. Hueber, L.P. Safonova, R. Gupta, A. Michalos, W. Mantulin, and E. Gratton, Non-invasive determination of the optical properties of adult brain: Near-infrared spectroscopy approach, J. Biomed Opt. 9 (1) (2004), 221-229. [49] E.W. Ciurczak and J.K. Drennen, Pharmaceutical and Medical Applications of Near-Infrared Applications (Practical Spectroscopy), Marcel Dekker Inc., New York, 2002. [50] S.M. Cohn, J.E. Varela, G. Giannotti, M.O. Dolich, M. Brown, A. Feinstein, M.G. McKenney and P. Spalding, Splanchnic perfusion evaluation during hemorrhage and resuscitation with gastric nearinfrared spectroscopy. J. Trauma 50 (4) (2001), 629-635.
390
[51] S.M. Cohn, A.B. Nathens, F.A. Moore, P. Rhee, J.C. Puyana, E.E. Moore and G.J. Beilman; (the StO2 in trauma patients trial investigators), Tissue oxygen saturation predicts the development of organ dysfunction during traumatic shock resuscitation, J Trauma 62 (1) (2007) 44-54. [52] W.M. Colier, Near infrared spectroscopy: toy or tool, and investigation on the clinical ability of near infrared spectroscopy. Ph.D. Dissertation, University of Nijmegen, 1995, ISBN 90-9008670-6. [53] W.M. Colier, F.M. Froeling, J.D. de Vries and B. Oesburg, Measurement of the blood supply to the abdominal testis by means of near infrared spectroscopy, Eur Urol 27 (1995), 160-6. [54] C.E. Cooper, C.E. Elwell, J.H. Meek, S.J. Matcher, J.S. Wyatt, M. Cope, and D.T. Delpy, The noninvasive measurement of absolute cerebral deoxyhemoglobin concentration and mean optical path length in the neonatal brain by second derivative near infrared spectroscopy, Pediatr Res. 39 (1) (1996), 32-38. [55] C.E. Cooper and R. Springett. Measurement of cytochrome oxidase and mitochondrial energetics by near-infrared spectroscopy. Phil Trans R Soc Lond B Biol Sci 352 (1354) (1997), 669-676. [56] C.E. Cooper, D.T. Delpy and E.M. Nemoto, The relationship of oxygen delivery to absolute haemoglobin oxygenation and mitochondrial cytochrome oxidase redox state in the adult brain: a nearinfrared spectroscopy study. Biochem J 332 (3) (1998), 627-32. [57] C.E. Cooper, M. Cope, R. Springett, J. Penrice, L. Tyszczuk, S. Punwani, R. Ordidge, J.S. Wyatt and D.T. Delpy, Use of mitochondrial inhibitors to demonstrate that cytochrome oxidase near-infrared spectroscopy can measure mitochondrial dysfunction non-invasively in the brain, J Cereb Blood Flow Metab 19 (1999) 27-38. [58] C.E. Cooper, In vivo measurements of mitochondrial function and cell death following hypoxic/ ischaemic damage to the new-born brain, Biochem. Soc. Symp 66 (1999), 123-140. [59] M. Cope, D.T. Delpy, S. Wray, J.S. Wyatt and E.O.R. Reynolds. A CCD spectrometer to quantitate the concentration of chromophores in living tissue utilizing the absorption peak of water at 975 nm, Adv Exp Med Biol. 248 (1989), 33-40. [60] S.M. Coyle, T.E. Ward and C.M. Markham, Brain-computer interface using a simplified functional near-infrared spectroscopy system, J Neural Eng 4 (2007), 219-226. [61] F. Crespi, Near-infrared spectroscopy (NIRS): a non-invasive in vivo methodology for analysis of brain vascular and metabolic activities in real time in rodents, Curr Vasc Pharmacol. 5 (4) (2007), 305-21. [62] B.A. Crookes, S.M. Cohn, E.A. Burton, J. Nelson and K.G. Proctor, Non-invasive muscle oxygenation to guide fluid resuscitation after traumatic shock, Surgery 135 (2004), 662-670. [63] B.A. Crookes, S.M. Cohn, S. Bloch, J. Amortegui, R. Manning, P. Li, M.S. Proctor, A. Hallal, L.H. Blackbourne, R. Benjamin, D. Soffer, F. Habib, C.I. Schulman, R. Duncan and K.G. Proctor. Can near-infrared spectroscopy identify the severity of shock in trauma patients? Surgery 58 (4) (2005), 806-16. [64] K.M. Cross, L. Leonardi, J.R. Payette, M. Gomez, M.A. Levasseur, B.J. Schattka, M.G. Sowa and J.S. Fish, Clinical utilization of near-infrared spectroscopy devices for burn depth assessment, Wound Repair Regen 15 (3) (2007), 332-40. [65] D.J. Cuccia, F. Bevilacqua, A.J. Durkin, and B.J. Tromberg, Modulated imaging: Quantitative analysis and tomography of turbid media in the spatial-frequency domain, Opt Lett 30 (11) (2005), 1354-1356. [66] W. Cui, C. Kumar, and B. Chance, Experimental study of the migration depth for the photons measured at sample surface. Time resolved spectroscopy and imaging, Proc Int Soc Opt Eng 1431 (1991), 180-91. [67] L.K. Davies and G.M. Janelle, Con: all cardiac surgical patients should not have intraoperative cerebral oxygenation monitoring, J Cardiothorac Vasc Anesth 20 (3), (2006), 450-5. [68] R.A. De Blasi, M. Cope, C. Elwell, F. Safoue, and M. Ferrari, Non-invasive measurement of human forearm oxygen consumption by near infrared spectroscopy, Eur J Appl Physiol 67 (1) (1993), 20-25. [69] R.A. De Blasi, M. Ferrari, A. Natali, G. Conti, A. Mega and A. Gasparetto, Non-invasive measurement of forearm blood flow and oxygen consumption by near-infrared spectroscopy, J Appl Physiol 76 (3) (1994), 1388-1393. [70] R.A. De Blasi, N.Almenrader and M.Ferrari, Brain oxygenation monitoring during cardiopulmonary bypass by near infrared spectroscopy, Adv Exp Med Biol 413 (1997), 97-104. [71] D.T. Delpy, M. Cope, P. van der Zee, S. Arridge, S. Wray and J.S. Wyatt. Estimation of optical path length through tissue from direct time of flight measurements, Phys Med Biol 33 (1988), 1433-1442. [72] D.T. Delpy and M. Cope. Quantification in tissue near-infrared spectroscopy, Phil Tans R Soc Lond B Biol Sci 352 (1354) (1997), 649-659. [73] A.J. Du Plessis, J. Newburger, R.A. Jonas, P. Hickey, H. Naruse, M. Tsuji, A. Walsh, G. Walter, D. Wypij and J.J. Volpe. Cerebral oxygen supply and utilization during infant cardiac surgery, Ann Neurol 37 (4) (1995), 488-497.
391
[74] A. Dullenkopf, B. Frey, O. Baezinger, A. Gerber and M. Weiss, Measurement of cerebral oxygenation state in anaesthetized children using the INVOS 5100 cerebral oximeter, Paediatr Anaesth 13 (2003), 384-391. [75] A. Duncan, J.H. Meek, M. Clemence, C.E. Elwell, P. Fallon, L. Tyszczuk, M. Cope and D.T. Delpy, Measurement of cranial optical path length as a function of age using phase resolved near infrared spectroscopy, Pediatr Res 39 (5) (1996), 889-894. [76] A. Duncan, T.L. Whitlock, M. Cope and D.T. Delpy, A multiwavelength, wideband, intensity modulated optical spectrometer for near infrared spectroscopy and imaging. SPIE Proc, 1888 (1993), 248-257. [77] T. Durduran, Y. Guoqiang, Z. Chao, G. Lech, B. Chance, and A.G. Yodh, Quantification of muscle oxygenation and flow of healthy volunteers during cuff occlusion of arm and leg flexor muscles and plantar flexion exercise, Proc. SPIE 4955 (1) (2003), 447-453. [78] T. Durduran, G. Yu, M.G. Burnett, J.A. Detre, J.H. Greenberg, J. Wang, C. Zhou and A.G. Yodh, Diffuse optical measurement of blood flow, blood oxygenation, and metabolism in a human brain during sensorimotor cortex activation, Opt Lett 29 (15) (2004), 1766-1768. [79] H.L. Edmonds Jr., Pro: all cardiac surgical patients should have intraoperative cerebral oxygenation monitoring, J Cardiothorac Vasc Anesth 20 (3), (2006), 445-9. [80] H.L. Edmonds, B.L. Ganzel and E.H. Austin. Cerebral oximetry for cardiac and vascular surgery. Semin Cardiothorac Vasc Anesth 8 (2) (2004), 147-66. [81] A.D. Edwards, C.E. Richardson, D.T. Delpy, M. Cope and E.O.R. Reynolds, Cotside measurement of cerebral blood flow in ill newborn infants by near infrared spectroscopy. Lancet. 2 (1988), 770-8 [82] A.D. Edwards, G.C. Brown, M. Cope, J.S. Wyatt, D.C. McCormick, S.C. Roth, D.T. Delpy and E.O.R. Reynolds, Quantification of concentration changes in neonatal human cerebral oxidised cytochrome oxidase, J Appl Physol 71 (1991), 1907-1913. [83] J.P. Eiberg, T.V. Schroeder. K.C. Vogt and N.H. Secher. Near infra-red spectroscopy during peripheral vascular surgery. Cardiovasc Surg 5 (3) (1997), 304-308. [84] K.G.B. Elliott and A.J. Johnstone, Diagnosing acute compartment syndrome, J Bone Joint Surg 85 (5) (2003), 625-32. [85] C.E. Elwell, H. Owen-Reece, M. Cope, J.S. Wyatt, A.D. Edwards, D.T. Delpy and E.O.R. Reynolds, Measurement of adult cerebral haemodynamics using near infrared spectroscopy, Acta Neurochir Suppl (Wien) 59 (1993), 74-80. [86] E. Elwell, H. Owen-Reese, J.S. Wyatt, M. Cope, E.O.R. Reynolds and D.T. Delpy, Influence of respiration and changes in expiratory pressure on cerebral haemoglobin concentration measured by near infrared spectroscopy, Cereb Blood Flow Metab. 16 (2) (1996), 353-357. [87] E. Elwell, J.R. Henty, T.S. Leung, T. Austin, J.H. Meek, D.T. Delpy, and J.S. Wyatt, Measurement of CMRO2 in neonates undergoing intensive care using near infrared spectroscopy, Adv. Exp. Med. Biol. (566) (2005), 263-268. [88] P. Fallon, I. Roberts, F.J. Kirkham, M.J. Elliott, A. Lloyd-Thomas, R. Maynard and A.D. Edwards, Cerebral hemodynamics during cardiopulmonary bypass in children using near-infrared spectroscopy, Ann Thorac Surg 56 (6) (1993), 1473-7. [89] S. Fantini, M.A. Franceschini, J.S. Maier, S.A. Walker, B. Barbieri, and E. Gratton, Frequencydomain multichannel optical detector for noninvasive tissue spectroscopy and oximetry, Opt. Eng. 34 (1995), 32-42. [90] S. Fantini, E.L. Heffer, V.E. Pera, A. Sassaroli and N. Liu. Spatial and spectral information in optical mammography, Technol Cancer Res Treat 4 (5) (2005), 471-482. [91] S. Fantini and P. Taroni, Optical mammography, in: Cancer imaging: lung and breast carcinomas, M.A. Hayat, Ed., Elsevier New York, 2007, pp. 449-58. [92] M. Ferrari, T. Binzoni and V. Quaresima. Oxidative metabolism in muscle, Philos Trans R Soc Lond B Biol Sci 352 (1354) (1997), 677-683. [93] M. Ferrari, L. Mottola and V. Quaresima. Principles, techniques and limitations of near infrared spectroscopy, Can J Appl Physiol 29 (4) (2004), 463-487. [94] P.C. Ferry, Neurologic sequellae of open-heart surgery in children; an irritating question, Am. J. Dis. Child. 144 (1990), 369-373. [95] M.P. Fink, Cytopathic hypoxia. Crit Care Clin 18 (1) (2002), 165-178. [96] M. Firbank. C.E. Elwell, C. Cooper and D.T. Delpy, Experimental and theoretical comparison of NIR spectroscopy measurements of cerebral hemoglobin changes, J Appl Physiol 85 (5) (1998), 1915-1921. [97] A.P. Forget, J. Mangalaboyi, S. Mordon, B. Guery, B. Vallet, F. Fourrier and C. Chopin. Escherichia coli endotoxin reduces cytochrome aa3 redox status in pig skeletal muscle. Crit Care Med 28 (10) (2000), 3491-7.
392
[98] P.M. Fortune, M. Wagstaff and A.J. Petros, Cerebro-splanchnic oxygenation ratio (CSOR) using near infrared spectroscopy may be able to predict splanchnic ischaemia in neonates, Intensive Care Med 27 (8) (2001), 1401-07. [99] M.A. Franceschini, D. Wallace, B. Barbieri, S. Fantini, W.W. Mantulin, S. Pratesi, G.P. Donzelli, and E. Gratton, Optical study of the skeletal muscle during exercise with a second generation frequency domain tissue oximeter, Proc. SPIE 2979 (1997), 807-814. [100] M.A. Franceschini, D.A. Boas, A. Zourabian, S.G. Diamond, S. Nadgir, D.W. Lin, J.B. Moore, and S. Fantini, Near-infrared spiroximetry: Noninvasive measurements of venous saturation in piglets and human subjects, J Appl Physiol 92 (1) (2002), 372-384. [101] D. Fukui, H. Urayama, K. Tanaka and S. Kawasaki. Use of near-infrared spectroscopic measurement at the buttocks during abdominal aortic surgery. Circ J. 66 (12) (2002), 1128-1131. [102] R.E. Gagnon, F.A. Gagnon and A.J. Macnab, Comparison of 13 published cytochrome c oxidase nearinfrared spectroscopy algorithms, Eur J Appl Physiol 74 (6) (1996), 487-495. [103] R.E. Gagnon and A.J. Macnab, C/C++ Coding for matrix pseudo inverses in clinical near infrared spectroscopy, Comput Methods Biomech Biomed Engin 1 (2) (1998), 69-86. [104] R.E. Gagnon, A. Leung and A.J. Macnab. Variations in regional cerebral blood volume in neonates associated with nursery care events. Am J Perinatol 16 (1) (1999), 7-11. [105] R.E. Gagnon, A.J. Macnab, F.A. Gagnon, D. Blackstock and J.G. LeBlanc, Comparison of two spatially resolved NIRS oxygenation indices, J Clin Monit Comput 17 (7-8) (2002), 385-391. [106] R.E. Gagnon, A.J. Macnab and J.G. LeBlanc, Patterns of change in cytochrome c oxidase redox status, Spectroscopy 18 (2004), 161-166. [107] R.E. Gagnon and A.J. Macnab, Near Infrared spectroscopy (NIRS) in the clinical setting an adjunct to monitoring during diagnosis and treatment. Spectroscopy 19 (2005), 221-233. [108] L.M. Gentilello, A. Sanzone, L. Wang, P.Y. Liu and L. Robinson, Near-infrared spectroscopy versus compartment pressure for the diagnosis of lower extremity compartmental syndrome using electromyography-determined measurements of neuromuscular function, J Trauma 51 (1) (2001), 1-8. [109] G. Giannotti, S.M. Cohn, M. Brown, J.E. Varela, M.G. McKenney and J.A. Wiseberg, Utility of nearinfrared spectroscopy in the diagnosis of lower extremity compartment syndrome, J Trauma 48 (3) (2000), 396-9. [110] S. Gilman, Neurological complications of open heart surgery, Ann Neurol 28 (4) (1990), 475-476. [111] B. Goldstein, New technologies in the intensive care unit: A cautionary tale. Pediatr Crit Care Med 6 (3) (2005), 378-379. [112] B. Grassi, V. Quaresima, C. Marconi, M. Ferrari, and P. Cerretelli, Blood lactate accumulation and muscle deoxygenation during incremental exercise, J Appl Physiol 87 (1) (1999), 348-355. [113] B. Grassi, M. Marzorati, F. Lanfranconi, A. Ferri, M. Longaretti, A. Stucchi, P. Vago, C. Marconi and L. Mordani, Impaired oxygen extraction in metabolic myopathies: detection and quantification by near-infrared spectroscopy, Muscle Nerve 35 (2007), 510-20. [114] A.I. Gravvanis, D.A. Tsoutsos, D. Karakitsos, P. Panavotou, T. Iconomou, G. Zografos, A. Karabinis, O. Papadopoulos, Application of the pedicled anterolateral thigh flap to defects from the pelvis to the knee, Microsurgery 26 (2006), 432-8. [115] W.J. Greeley, F.H. Kern, J. Meliones and R.M. Ungerleider, Monitoring the brain during cardiac surgery in children (Editorial), Can J Anaesth 40 (4) (1993), 291-297. [116] G. Greisen, Is near-infrared spectroscopy living up to its promises? Semin Fetal Neonatal Med. 11 (6) (2006), 498-502. Epub 2006 Sep 7. [117] A.J. Gunn, M. Battin, P.D. Gluckman, T.R.Gunn and L. Bennet, Therapeutic hypothermia: from lab to NICU. J Perinat Med. 33 (2005), 340-346. [118] A.K. Gupta, D.K. Menon, M. Czosnyka, P. Smielewski and J.G. Jones, Thresholds for hypoxic cerebral vasodilation in volunteers. Anesth Analg 85 (1997), 817-826. [119] T. Hamaoka, K.K. McCully, V. Quaresima, K. Yamamoto and B. Chance, Near-infrared spectroscopy/imaging for monitoring muscle oxygenation and oxidative metabolism in healthy and diseased humans, J Biomed Optics 12 (6) (2007), 062105 1-12. [120] N.B. Hampson and C.A. Piantadosi, Near infrared monitoring of human skeletal muscle oxygenation during forearm ischemia, J Appl Physiol 64 (1988), 2449-2547. [121] N.B. Hampson, E.M. Camporesi, B.W. Stolp, R.E. Moon, J.E. Shook, J.A. Griebel and C.A. Piantadosi. Cerebral oxygen availability by NIR spectroscopy during transient hypoxia in humans, J Appl Physiol 69 (1990), 907-913. [122] D.N. Harris, F.M. Cowans, D.A. Wertheim and S. Hamid, NIRS in adults effects of increasing optode separation, Adv Exp Med Biol. 354 (1994), 837-841. [123] M. Hayashida, N. Kin, T. Tomioka, R. Orii, H. Sekiyama, H. Usui, M. Chinzei and K. Hanaoka, Cerebral ischaemia during cardiac surgery in children detected by combined monitoring of BIS and near infra-red spectroscopy, Br J Anaesth 92 (5) (2004), 662-9.
393
[124] J.C. Hebden Advances in optical imaging of the newborn infant brain, Psychophysiology 40 (2003), 501-10. [125] J.C. Hebden, A. Gibson, T. Austin, R.M. Yusof, N. Everdell, D.T. Delpy, S.R. Arridge, J.H. Meek and J.C. Wyatt, Imaging changes in blood volume and oxygenation in the newborn infant brain using three-dimensional optical tomography, Phys Med Biol. 49 (7) (2004), 1117-130. [126] D.G. Hirtz, Report of the national institute of neurological disorders and stroke workshop on near infrared spectroscopy. Pediatr 91 (2) (1993), 414-7. [127] G.M. Hoffman, Pro: near-infrared spectroscopy should be used for all cardiopulmonary bypass, J Cardiothorac Vasc Anesth. 20 (4) (2006), 606-12. [128] S. Homma, H. Eda, S. Ogasawara and A. Kagaya, Near-infrared estimation of O2 supply and consumption in forearm muscles working at varying intensity, J Appl Physiol 80 (4) (1996), 1279-84. [129] P. Hopton, T.S. Walsh, and A. Lee, Measurement of cerebral blood volume using near-infrared spectroscopy and indocyanine green elimination, J Appl Physiol 87 (5) (1999), 1981-87. [130] Y. Hoshi, Functional near-infrared optical imaging: utility and limitations in human brain mapping. Psychophysiology 40 (2003), 511-20. [131] Y. Hoshi, M. Shimada, C. Sato and Y. Iguchi, Re-evaluation of near-infrared propagation in the adult human head: implications for functional near-infrared spectroscopy. J Biomed Opt 10 (6) (2005), 064032. [132] Y. Hoshi, Functional near-infrared spectroscopy: current status and future prospects. J Biomed Optics 12 (6) (2007), 062106. [133] K. Hunt, I. Tachtsidis, K. Bleasdale-Barr, C. Elwell, C. Mathias and M. Smith, Changes in cerebral oxygenation and hemodynamics during postural blood pressure changes in patients with autonomic failure. Physiol Meas 27 (9) (2006), 777-85. [134] T.J. Huppert, R.D. Hoge, S.G. Diamond, M.A. Franceschini and D.A. Boas, A temporal comparison of BOLD, ASL, and NIRS hemodynamic responses to motor stimuli in adult humans, Neuroimage 29 (2) (2006), 368-82. Epub 2005 Nov 21. [135] S. Ijichi, T. Kushaka, K. Isobe, F. Islam, K. Okubo, H. Okada, M. Namba, K. Kawada, T. Imai and S. Itoh, Quantification of cerebral hemoglobin as a function of oxygenation using near-infrared timeresolved spectroscopy in a piglet model of hypoxia, J Biomed Opt 10 (2) (2005), 024026. [136] F. Irani, S.M. Platek, S. Bunce, A.C. Ruocco and D. Chute, Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging technology with important applications for the study of brain disorders, Clin Neuropsychol. 21 (1) (2007), 9-37. [137] F.F. Jobsis, Non-invasive infrared monitoring of cerebral and myocardial oxygen sufficiency and circulatory parameters, Science 198 (1977), 1264-7. [138] R.N. Jones, Analytical applications of vibrational spectroscopy, a historical review, in: Chemical, Biological, and Industrial Applications of Infrared Spectroscopy, John Wiley and Sons, New York, 1985, pp. 1-43. [139] S. Kahraman, H. Kayali, C. Atabey, F. Acar and S. Gocmen, The accuracy of near-infrared spectroscopy in detection of subdural and epidural hematomas, J Trauma 61 (6) (2006), 1480-3. [140] Y. Kakihana, A. Matsunaga, K. Tobo, S. Isowaki, M. Kawakami, I. Tsuneyoshi, Y. Kanmura and M. Tamura, Redox behavior of cytochrome oxidase and neurological prognosis in 66 patients who underwent thoracic aortic surgery, Eur J Cardiothoracic Surg 12 (2002), 434-9. [141] Y. Kakihana, T.Kuniyoshi, S. Isowaki, K.Tobo, E. Nagata, N. Okayama, K. Kitahara, T. Moriyama, T. Omae, M. Kawakami, Y. Kanmura and M. Tamura, Relationship between redox behavior of brain cytochrome oxidase and neurological prognosis, Adv Exp Med Biol 530 (2003), 413-419. [142] J. Kakogawa, K. Sumimoto, E. Ho and N. Kanayama, Transabdominal measurement of oxygenation of the placenta by near-infrared spectroscopy, Semin Thromb Hemost. 31 (3) (2005), 297-301. [143] R.W. Katzberg, Urography in the 21st century: New contrast media, renal handling, imaging characteristics, and nephrotoxicity, Radiology 204 (1997), 297-312. [144] T. Kawamura, J. Kakogawa, Y. Takeuchi, S. Takani, S. Kimura, T. Nishiguchi, M. Sugimura, K. Sumimoto and N. Kanayama, Measurement of placental oxygenation by transabdominal nearinfrared spectroscopy. Am J Perinatol. 24 (3) (2007) 161-6. Epub 2007 Feb 15. [145] R.T. Kell and Y. Bhambhani, Relationship between erector spinae static endurance and muscle oxygenation-blood volume changes in healthy and low back pain subjects, Eur J Appl Physiol 96 (3) (2006), 241-248. [146] E. Keller, G. Wietasch, P. Ringleb, M. Scholz, S. Schwarz, R. Stingele, S. Schwab, D. Hanley, and W. Hacke, Bedside monitoring of cerebral blood flow in patients with acute hemispheric stroke, Crit Care Med 28 (2) (2000), 511-516. [147] E. Keller, A. Nadler, H. Alkadhi. S.S. Kollias, Y. Yonekawa and P. Niederer, Noninvasive measurement of cerebral blood flow and regional cerebral blood volume by near-infrared spectroscopy and indocyanine green dye dilution, Neuroimage 20 (2) (2003), 828-39.
394
[148] G.J. Kemp, A.V. Crowe, H.K. Anijeet, Q.Y. Gong, W.E. Bimson, S.P. Frostick, J.M. Bone, G.M. Bell and J.N. Roberts, Abnormal mitochondrial function and muscle wasting, but normal contractile efficiency, in haemodialysed patients studied non-invasively in vivo, Nephrol Dialys Transplan 19 (2004), 1520-7. [149] P.J. Kirkpatrick, Use of near-infrared spectroscopy in the adult, Philos Trans R Soc Lond B Biol Sci 352 (1354) (1997), 701-05. [150] T. Kita, T. Mammoto and Y. Kishi, Monitoring of cerebral oxygenation status with near-infrared spectroscopy during inferior vena caval reconstruction under extracorporeal circulation, Jap J Anesth 50 (2001), 890-4. [151] M. Kohl, C. Nolte, H.R. Heekeren, S. Horst, U. Scholz, H. Obrig and A. Villringer, Determination of the wavelength dependence of the differential pathlength factor from near-infrared pulse signals, Phys Med Biol 43 (6) (1998), 1771-1782. [152] V.R. Kondepati, M. Keese, R. Mueller, B.C. Manegold and J. Backhaus, Application of near-infrared spectroscopy for the diagnosis of colorectal cancer in resected human tissue specimens, Vib Spectrosc 44 (2) (2007), 236-242. [153] V.R. Kondepati, H.M. Heise and J. Backhaus, Recent applications of near-infrared spectroscopy in cancer diagnosis and therapy, Anal Bioanal Chem 390 (1) (2008), 125-139. [154] B. Kragsterman, H. Parsson and D. Bergqvist, Local haemodynamic changes during carotid endarterectomy the influence on cerebral oxygenation. Eur J Vasc Endovasc Surg 27 (4) (2004), 398-402. [155] W. Krause, P. Muschick and U. Kruger, Use of near-infrared reflection spectroscopy to study the effects of x-ray contrast media on renal tolerance in rats: effects of a prostacyclin analogue and of phosphodiesterase inhibitors, Invest Radiol 37 (2002), 698-705. [156] C.D. Kurth, J.M. Steven, D. Benaron and B. Chance, Near-infrared monitoring of the cerebral circulation, J Clin Monit 9 (1993), 163-70. [157] C.D. Kurth, M. Priestly, J. Golden, J. McCann and R. Raghupathi, Regional patterns of neuronal death after deep hypothermic circulatory arrest in newborn pigs, J Thorac Cardiovasc Surg 118 (6) (1999), 1068-1077. [158] F. Lanfranconi, E. Borrelli, A. Ferri, S. Porcelli, M. Maccherini, M. Chiavarelli and B. Grassi, Noninvasive evaluation of skeletal muscle oxidative metabolism after heart transplant. Med Sci Sports Exerc 38 (8) (2006), 1374-83. [159] A. Lassnigg, M. Hiesmayr, P. Keznickl, T. Mullner, M. Ehrlich and G. Grubhofer, Cerebral oxygenation during cardiopulmonary bypass measured by near-infrared spectroscopy: effects of hemodilution, temperature, and flow, J Cardiothorac Vasc Anesth 13 (5) (1999), 544-548. [160] P.M. Lemmers, M. Toet, L.V. van Schelven and F van Bel, Cerebral oxygenation and cerebral oxygen extraction in the preterm infant: the impact of respiratory distress syndrome, Exp Brain Res. 173 (3) (2006), 458-67. [161] T.S. Leung, I. Tachtsidis, M. Smith, D.T. Delpy and C.E. Elwell, Measurement of the absolute optical properties and cerebral blood volume of the adult human head with hybrid differential and spatially resolved spectroscopy, Phys. Med. Biol. 51 (3) (2006), 703-717. [162] J. Li, G. Dietsche, D. Iftime, S. E. Skipetrov, G. Maret, T. Elbert, B.Rockstroh, and T. Gisler, Noninvasive detection of functional brain activity with near-infrared diffusing-wave spectroscopy, J Biomed Opt 10 (4) (2005), 44002. [163] A. Lima, J. Bakker, Noninvasive monitoring of peripheral perfusion, Int Care Med 31 (2005), 1316-26. [164] R. Liston, J. Crane, O. Hughes, S. Kuling, C. MacKinnon, K. Milne, B. Richardson and M.J. Trepanier, Fetal health surveillance in labour, J Obstet Gynaecol Can 24 (4) (2002), 342-55. [165] G. Litscher and G. Schwartz, eds., Transcranial cerebral oximetry, Lengerich Berlin Dusseldorf Riga Scottsdale Wien Zagreb: Pabst Science Publishers, 1997. [166] L.N. Livera, S.A. Spencer, M.S. Thorniley, Y.A. Wickramasinghe and P Rolfe, Effects of hypoxemia and bradycardia on neonatal cerebral hemodynamics, Arch Dis Child 66 (4) (1991), 37680. [167] L.N. Livera, Y.A. Wickramasinghe, S.A. Spencer, P Rolfe and M.S. Thorniley, Cyclical fluctuations in cerebral blood volume, Arch Dis Child 67 (1) (1992), 62-63. [168] A.T. Lovell, H. Owen-Reece, C.E. Elwell, M. Smith and J.C. Goldstone, Continuous measurement of cerebral oxygenation by near-infrared spectroscopy during induction of anesthesia, Anesth Analg 88 (3) (1999), 554-558. [169] A.J. Macnab and L. Stothers, Development of a near-infrared spectroscopy instrument for applications in urology, Can J Urol 15 (5) (2008), 4233-4240. [170] A.J. Macnab and R.E. Gagnon, Electromagnetic interference testing of somatosensory and nearinfrared devices used in the microgravity and hospital environments, Biomed Instrum Technol 30 (3) (1996), 236-42.
395
[171] A.J. Macnab and R.E. Gagnon. Potential sources of discrepancy between living tissue near infrared spectroscopy algorithms. Anal Biochem 236 (2) (1996), 375-377. [172] A.J. Macnab, R.E. Gagnon and F.A. Gagnon. Near infrared spectroscopy for intraoperative monitoring of the spinal cord, Spine 27 (1) (2002), 17-20. [173] A.J. Macnab, R.E. Gagnon, F.A. Gagnon and J. LeBlanc, NIRS monitoring of brain and spinal cord: Detection of adverse intraoperative events, Spectroscopy. 17 (2003), 483-490. [174] A.J. Macnab, R.E. Gagnon and L. Stothers, Clinical NIRS of the urinary bladder a demonstration case report, Spectroscopy 19 (2005), 207-212. [175] A.J. Macnab, L. Stothers, Near-infrared spectroscopy: validation of bladder outlet-outlet obstruction assessment using non-invasive parameters, Can J Urol 1595) (2008), 4241-4240. [176] P.L. Madsen and N.H. Secher, Near-infrared oximetry of the brain. Prog Neurobiol 58 (1999), 541-560. [177] D.M. Mancini, L. Bolinger, H. Li, K. Kendrick, B. Chance and J.R. Wilson, Validation of near infrared spectroscopy in humans, J Appl Physiol 77 (6) (1994), 2740-47. [178] K. Maruo, T. Oota, M. Tsurugi, T. Nakagawa, H. Arimoto, M. Tamura, Y. Ozaki and Y. Yamada, New methodology to obtain a calibration model for non-invasive near-infrared blood glucose monitoring, Appl Spectrosc 60 (4) (2006), 441-9. [179] S.J. Matcher, M. Cope and D.T. Delpy, Use of the water absorption spectrum to quantify tissue chromophore concentration changes in near-infrared spectroscopy, Phys Med Biol 39 (1) (1994), 177-196. [180] S.J. Matcher, P. Kirkpatrick, K. Nahid, M. Cope and D.T. Delpy, Absolute quantification methods in tissue near-infrared spectroscopy, Proc SPIE 2389 (1995), 486-495. [181] S.J. Matcher, C.E. Elwell, C.E. Cooper, M. Cope and D.T. Delpy, Performance comparison of several published tissue near-infrared spectroscopy algorithms, Annal Biochem 27 (1995), 54-68. [182] N. Matsumoto, S. Ichimura, T. Hamaoka, T. Osada and M. Hattori, Impaired muscle oxygen metabolism in uremic children: improved after renal transplantation, Am J Kidney Dis. 48 (2006), 473-80. [183] A. Matsunaga, Y. Nomura, S. Kuroda, M. Tamura, J. Nishihira and Yoshimura, Energy-dependent redox state of heme a + a3 and copper cytochrome oxidase in perfused rat brain in situ, Am J Physiol 275 (1998), C1022-30. [184] K. Matsushita, S. Homma and E. Okada, Influence of adipose tissue on muscle oxygenation measurement with NIRS instrument, Proc SPIE 3194 (1998), 151-165. [185] P.W. McCormick, M. Stuart, M.G. Goetting, G. Balakrishnan, Regional cerebrovascular oxygen saturation measured by optical spectroscopy in humans, Stroke 22 (1991), 596-602. [186] K.K. McCully, C. Halber and J.D. Posner, Exercise-induced changes in oxygen saturation in the calf muscles of elderly subjects with peripheral vascular disease, J Gerontol 49 (1994), B128-134. [187] A.D. McGown, H. Makker, C. Elwell, P.G. Al-Rawi, A. Valipour, and S.G. Spiro, Measurement changes in cytochrome oxidase redox state during obstructive sleep apnea using near-infrared spectroscopy, Sleep 26 (2003), 710-16. [188] B.A. McKinley and B.D. Butler, Comparison of skeletal muscle po2, pco2 and ph with gastric tonometric p(co2), and ph in hemorrhagic shock, Crit Care Med 27 (9) (1999), 1869-77. [189] P.S. McQuillen, M.S. Nishimoto, C.L. Bottrell, L.D. Fineman, S.E. Hamrick, D.V. Glidden, A. Azakie, I. Adatia and S.P. Miller, Regional and central venous oxygen saturation monitoring following pediatric cardiac surgery: concordance and association with clinical variables, Pediatr Crit Care Med. 8 (2) (2007), 154-60. [190] J.H. Meek, C.E. Elwell, D. McCormick, A.D. Edwards, J.P. Townsend, A.L. Stewart and J.S. Wyatt. Abnormal cerebral haemodynamics in perinatally asphyxiated neonates related to outcome. Arch Dis Child Fetal Neonatal Ed 81 (1999) F110-F115. [191] D.J. Mehagnoul-Schipper, B.F. van der Kallen, W.N. Colier, M.C. Van der Sluijs, L.J. van Erning, H.O. Thijssen, B. Oeseburg, W.H. Hoefnagels, and R.W. Jansen, Simultaneous measurements of cerebral oxygenation changes during brain activation by near-infrared spectroscopy and functional magnetic resonance imaging in healthy young and elderly subjects, Hum Brain Map 16 (2002), 14-23. [192] S.P. Miller, J. Weiss, A. Barnwell, D.M. Ferriero, B. Latal-Hajnal, A. Ferrer-Rogers, N. Newton, J.C. Partridge, D.V. Glidden, D.B. Vigneron and A.J. Barkovich, Seizure-associated brain injury in term newborns with perinatal asphyxia. Neurology. 58 (4) (2002), 542-548. [193] G.A. Millikan, Experiments on muscle haemoglobin, Proc R Soc Lond A 123 (1937), 218-224. [194] G.A. Millikan, The oximeter, an instrument for measuring continuously the oxygen saturation of arterial blood in man, Rev Sci Instrum 13 (1941), 431-44. [195] H. Mitsuta, H. Ohdan, Y. Fudaba, T. Irei, H. Tashiro, T. Itamoto and T. Asahara, Near-infrared spectroscopic analysis of hemodynamics and mitochondrial redox in right lobe grafts in living-donor liver transplantation, Am J Transplant. 6 (4) (2006), 797-805.
396
[196] W. Moalla, Y. Maingourd, R. Gauthier, L.P. Cahalin, Z. Tabka and S. Ahmaidi, Effect of exercise training on respiratory muscle oxygenation in children with congenital heart disease, Eur J Cardiovasc Prev Rehabil 13 (4) (2006), 604-611. [197] E.R. Mohler, G. Lech, G.E. Supple, H. Wang and B. Chance, Noninvasive evaluation of skeletal muscle oxidative metabolism after heart transplant, Diabetes Care. 29 (8) (2006), 1856-59. [198] Y. Morimoto, O. Kemmotsu, S. Gando, T. Shibano and H. Shikama, The effect of calcium channel blockers on cerebral oxygenation during tracheal extubation, Anesth Analg 91 (2) (2000), 347-352. [199] Y. Morimoto, Y. Niida, K. Hisano, Y. Hua, O. Kemmotsu, T. Murashita and K. Yasuda, Changes in cerebral oxygenation in children undergoing surgical repair of ventricular septal defects, Anaesthesia 58 (1) (2003), 77-83. [200] S. Muehlschlegel and E. Lobato, Con: All surgical patients should not have intraoperative cerebral oxygen monitoring, J Cardiothorac Vasc Anesth 20 (4) (2006), 613-615. [201] M.J. Munro. A.M. Walker and C.P. Barfield, Hypotensive extremely low birth weight infants have reduced cerebral blood flow, Pediatrics 114 (6) (2004), 1591-6. [202] N. Nagdyman, P. Ewert, B. Peters, O. Miera, T. Fleck, and F. Berger, Comparison of different nearinfrared spectroscopic cerebral oxygenation indices with central venous and jugular venous oxygenation saturation in children, Paediatr Anaesth 18 (2008) 160-66. [203] Nahum, P. Skippen, E. Skarsgard, R. Gagnon and A.J. Macnab, Correlation of transcutaneous hepatic near-infrared spectroscopy readings with liver surface readings and perfusion parameters in a piglet endotoxemic shock model. Liver Int 26 (10) (2006), 277-82. [204] J.P. Neary, Application of near infrared spectroscopy to exercise sports science, Can J Appl Physiol 29 (2004), 488-503. [205] L.A. Nelson, J.C. McCann, A.W. Lopeke, J. Wu, B. Ben Dor and C.D. Kurth, Development and validation of a multiwavelength spatial domain near-infrared oximeter to detect cerebral hypoxiaischemia, J Biomed Optics 11 (6) (2006), 064022. [206] J.W. Newburger, R.A. Jonas, G. Wernovsky, D. Wypij, P.R. Hickey, K.C.K. Kuban, D.M. Farrell, G.L. Homes, S.L. Helmers, J. Constantinou, E. Carrazana, J.K. Barlow, A.Z. Walsh, K.C. Lucius, J.C. Share, D.L. Wessel, F.L. Hanley, J.E. Mayer, A.R. Castaneda, and J.H. Ware, A comparison of the perioperative neurologic effects of hypothermic circulatory arrest vs. low-flow cardiopulmonary bypass in infant heart surgery, New Engl J Med 329 (1993), 1057-1064. [207] J.P. Newman, D.M. Peebles and M.A. Hanson, Adenosine produces changes in cerebral hemodynamics and metabolism as assessed by near-infrared spectroscopy in late-gestation fetal sheep in utero, Pediatr Res 50 (2) (2001), 217-21. [208] C.R. Newton, D.A. Wilson, E. Gunnoe, B. Wagner, M. Cope and R.J. Traystman, Measurement of cerebral blood flow in dogs with near infrared spectroscopy in the reflectance mode is invalid, J Cereb Blood Flow Metab 17 (1997), 695-703. [209] S.F. Nicklin, I.A. Hassan, Y.A. Wickramsinghe and S.A. Spencer, The light still shines, but not that brightly? The current status of perinatal near infrared spectroscopy. Arch Dis Child Fetal Neonatal Ed 88 (4) (2003), F263-8. [210] S. Nioka and B. Chance, NIR spectroscopic detection of breast cancer, Technol Cancer Res Treat 4 (5) (2005), 497-512. [211] M. Niwayama, L. Ling, J. Shao, N. Kudo and K. Yamamoto, Quantitative measurement of muscle hemoglobin oxygenation using near-infrared spectroscopy with correction for the influence of a subcutaneous fat layer, Rev Scientific Instruments, 71 (12) (2000), 4571-75. [212] G. Nollert, T. Shinoka and R.A. Jonas, Near-infrared spectrophotometry of the brain in cardiovascular surgery, Thorac Cardiovasc Surg 46 (1998), 67-75. [213] G. Nollert, R.A. Jonas and B. Reichart, Optimizing cerebral oxygenation during cardiac surgery: a review of experimental and clinical investigations with near infrared spectrophotometry, Thorac Cardivasc Surg 48 (4) (2000), 247-253. [214] G. Nollert, Clinical evaluation of near-infrared spectroscopy, Can J Anaesth 53 (3) (2006), 323. [215] G. Nollert, P. Mohnle, P. Tassani-Prell, I. Uttner, G.D. Borasio, M. Schmoeckel and B. Reichart, Postoperative neuropsychological dysfunction and cerebral oxygenation during cardiac surgery, Thorac Cardiovasc Surg 43 (5) (1995), 260-244. [216] K.H. Norris. Light transmitted through human tissues, in: the science of photobiology, K.C. Smith, ed., Plenum Press, New York, 1977, pp. 400-409. [217] V. Ntziachristos and B. Chance, Probing physiology and molecular function using optical imaging: applications to breast cancer. Breast Cancer Res 3 (2001), 41-46. [218] H. Obrig and A. Villringer, Beyond the visibleimaging the human brain with light, J Cereb Blood Flow Metab 23 (1) (2003), 1-18. [219] M. Oda, Y. Yamashita, G. Nishimura, and M. Tamura, A simple and novel algorithm for timeresolved multiwavelength oximetry, Phys. Med. Biol. 41 (3) (1996), 551-562.
397
[220] E Okada and D.T Delpy, Near-infrared light propagation in an adult head model; II effect of superficial tissue thickness on sensitivity of the near-infrared spectroscopy signal, Appl Opt 42 (2003), 2915-22. [221] H. Owen-Reece, C.E. Elwell, J.S. Wyatt and D.T. Delpy. The effect of scalp ischemia on measurement of cerebral blood volume by near-infrared spectroscopy. Physiol Meas 17 (4) (1996), 279-86. [222] H. Owen-Reece, M. Smith, C.E. Elwell, J.C. Goldstone, Near infrared spectroscopy, Br J Anaesth 82 (1999), 418-26. [223] K.S. Palmer, S.A. Spencer, Y.A. Wick, T. Wright, D.P. Southall and P Rolfe, Effects of positive and negative pressure ventilation on cerebral blood volume in newborn infants, Acta Paediatr 84 (2) (1995), 132-39. [224] C. Payer, B. Urlesberger, M. Pauger and W. Muller, Apnea associated with hypoxia in preterm infants: impact on cerebral blood volume. Brain Dev. 25 (1) (2003), 25-31. [225] D.M. Peebles, A.D. Edwards, J.S. Wyatt, A.P. Bishop, M. Cope, D.T. Delpy and E.O.R. Reynolds, Changes in human fetal cerebral hemoglobin concentration and oxygenation during labor measured by near-infrared spectroscopy, Am J Obstet Gynecol 166 (1992), 1369-73. [226] D.M. Peebles, Cerebral hemodynamics and oxygenation in the fetus: The role of intrapartum nearinfrared spectroscopy, Clinics in Perinatology 24 (3) (1997), 547-65 [227] D.M. Peebles, S. Miller, J.P. Newman, R. Scott and M.A. Hanson. The effect of systemic administration of lipopolysaccharide on cerebral haemodynamics and oxygenation in the 0.65 gestation ovine fetus in utero. BJOG 110 (8) (2003), 735-43. [228] C. Peeters-Scholte, E. van den Tweel, F. Groenendaal and F. van Bel, Redox status of near infrared spectroscopy-measured cytochrome aa3 correlates with periventricular leucomalacia, J Pediatr 143 (2004), 20-26. [229] M.I.R. Pereira, P.S.C. Gomes and Y.N. Bhambhani, A brief review of the use of near infrared spectroscopy with particular interest in resistance exercise, Sports Med 37 (7) (2007), 615-624. [230] M.S. Petrov, A.S. Gordetzov, and M.V. Kukosh, Early prediction of severity in acute pancreatitis using infrared spectroscopy of serum, Pancreatology 7 (5-6), 451-58. [231] A. Petrova and R. Mehta, Near-infrared spectroscopy in the detection of regional tissue oxygenation during hypoxic events in preterm infants undergoing critical care, Ped Crit Care Med 7 (2006), 449-54. [232] D. Piao, H. Xie, W. Zhang and J.S. Krasinski, Endoscopic, rapid near-infrared optical tomography, Opt Lett 31 (19) (2006), 2876-78. [233] C.A. Piantadosi, M. Hall, and B.J. Comfort, Algorithms for in vivo near-infrared spectroscopy, Anal Biochem 253 (1997) 277-79. [234] G. Pichler, B. Urlesberger and W. Muller, Impact of bradycardia on cerebral oxygenation and cerebral blood volume during apnoea in preterm infants, Physiol Meas 24 (3) (2003), 671-80. [235] J. Plachky, S. Hofer, M. Volkmann, E. Martin, H.J. Bardenheuer and M.A. Weigand, Regional cerebral oxygen saturation is a sensitive marker of cerebral hypoperfusion during orthoptic liver transplantation, Anesth Analg 99 (2) (2004), 344-9. [236] M. Podbregar and H. Mozina, Skeletal muscle oxygen saturation does not estimate mixed venous oxygen saturation in patients with severe left heart failure and additional severe sepsis or septic shock, Crit Care 11 (1) (2007), 1-8. [237] B.W. Pogue, S.D. Jiang, H. Dehghani, C. Kogel, S. Soho, S. Srinivasan, X.M. Song, T.D. Tosteson, S.P. Poplack and K.D. Paulsen, Characterization of hemoglobin, water and NIR scattering in breast tissue: analysis of intersubject variability and menstrual cycle changes, J Biomed Opt 9 (3) (2004), 541-552. [238] B. Putnam, S. Bricker, P. Fedora, J Zelada, S. Shebrain, B. Omari and F. Bongard, The correlation of near-infrared spectroscopy with changes in oxygen delivery in a controlled model of altered perfusion, Am Surg 73 (10) (2007), 1017-22. [239] J.C. Puyana, B.R. Soller, S. Zhang and S.O. Heard, Continuous measurement of gut ph with nearinfrared spectroscopy during hemorrhagic shock, J Trauma 46 (1) (1999), 9-15. [240] J.C. Puyana and M.R. Pinsky. Searching for non-invasive markers of tissue hypoxia, Crit Care 11 (1) (2007), 116-117. [241] V. Quaresima, S. Sacco, R. Totaro, and M. Ferrari, Noninvasive measurement of cerebral haemoglobin oxygenation saturation using two near infrared spectroscopy approaches, J Bio-med Opt 5 (2) (2000), 201-205. [242] V. Quaresima, T. Komiyama, and M. Ferrari, Differences in oxygen re-saturation of thigh and calf muscles after two treadmill stress tests, Zentralbl Bakteriol Mikrobiol Hyg Abt 1 Orig B 132 (1) (2002), 67-73. [243] V. Quaresima, R. Lepanto and M. Ferrari, The use of near infrared spectroscopy in sports medicine, J Sports Med Phys Fitness 43 (1) (2003), 1-13.
398
[244] P. Rhee, L. Langdale, C. Mock and L.M. Gentilello, Near infra-red spectroscopy: continuous measurement of cytochrome oxidation during hemorrhagic shock, Crit Care Med 25 (1) (1997), 166-70. [245] G.W. Roach, M. Kanchuger, C.M. Mangano, N. Newman, N. Nussmeier, R. Woolman, A. Aggarwal, K. Marshall, S.H. Graham and C. Lev, Adverse cerebral outcomes after coronary bypass surgery. Multicenter Study of Perioperative Ischemia Research Group and the Ischemia Research and Education Foundation Investigators. N Engl J Med 335 (25) (1996), 1857-1863. [246] I. Roberts, P. Fallon, F.J. Kirkham, A. Lloyd-Thomas, C. Cooper, R. Maynard, M. Elliot, and A.D. Edwards, Estimation of cerebral blood flow with near infrared spectroscopy and indocyanine green, Lancet 342 (8884) (1993), 1425. [247] R.A. Rodrigues, G. Cornel, W.M. Splinter, N.A. Weerasena and C.W. Reed, Cerebral vascular effects of aortovenous cannulations for pediatric cardiopulmonary bypass, Ann Thorac Surg 64 (9) (2000), 1229-1235. [248] P. Rolfe, In vivo near-infrared spectroscopy, An Rev Biomed Eng 2 (2000), 715-714 [249] D.M. Romeo, P. Betta, G. Sanges, V. Di Benedetto, M .Astuto and M.G. Romeo, Cerebral hemodynamics and major surgery, Minerva Pediatr. 59 (3) (2007), 233-7. [250] S.C. Roth, J. Baudin, E. Cady, K. Johal, J.P. Townsend, J.S. Wyatt, E.O. Reynolds and A.L. Stewart, Relation of deranged neonatal cerebral oxidative metabolism with neurodevelopmental outcome and head circumference at 4 years, Dev Med Child Neurol 39 (1997), 718-725. [251] T. Sakamoto, H. Kurosawa, T. Shinoka, M. Aoki and Y. Isomatsu, The influence of ph strategy on cerebral and collateral circulation during hypothermic cardiopulmonary bypass in cyanotic patients with heart disease: results of a randomized trial and real-time monitoring, J Thorac Cardiovasc Surg 127 (1) (2004), 12-19. [252] H. Sako, T. Hadama, S. Miyamoto, H. Anai, T. Wada, E. Iwata, H. Hamamoto, H. Tanaka and M. Morita, Limb ischemia and reperfusion during abdominal aortic aneurysm surgery. Surg Today 34 (10) (2004), 832-6. [253] A. Sakudo, Y. Taniuchi, T. Kobayashi, T. Onodera and K. Ikuta, Normal cytochrome c oxidase activity in prion protein gene-deficient mice, Protein Pept Lett 15 (3) (2008), 250-54. [254] H. Sato, M. Kiguchi, F. Kawaguchi and A. Maki, Practicality of wavelength selection to improve signal-to-noise ratio in near-infrared spectroscopy, Neuroimage 21 (4) (2004), 1554-62. [255] H. Sato, M. Kiguchi, A. Maki, Y. Fuchino, A. Obata, T. Yoro and H. Koizumi, Within-subject reproducibility of near-infrared spectroscopy signals in sensorimotor activation after 6 months, J Biomed Opt, 11 (1) (2006), 014021. [256] C.F. Schaefer, M.R. Lerner and B. Biber, Dose-related reduction of cytochrome a,a3 induced by endotoxin in rats, Circ Shock, 33 (1) (1991), 17-25. [257] W. Schafer, P. Abrams, L. Liao, A. Mattiasson, F. Pesce, A. Spangberg, A.M. Sterling, N.R. Zinner and P. van Kerrebroeck; international Continence Society, Good urodynamic practices: uroflowmetry, filling cystometry, and pressure-flow studies, Neurourol Urodyn 21 (3) (2002), 261-274. [258] O. Scheufler, K. Exner and R. Andresen, Investigation of TRAM flap oxygenation and perfusion by near-infrared reflection spectroscopy and color-coded duplex sonography, Plast Reconstr Surg 113 (1) (2004), 141-52. [259] S. Schmidt, S. Gorinssen, H. Eilers, H. Fahnenstich, A. Dorer and D. Krebs, Animal experiments for the evaluation of laser spectroscopy in the fetus during labor, J Perinat Med 19 (1-2) (1991), 107-13. [260] B. Shadgan, M. Menon, P.J. OBrian and W.D. Reid, Diagnostic techniques in acute compartment syndrome of the leg, J Orthop Trauma 22 (2008), 581-567. [261] B. Shadgan, L. Stothers, A. Macnab, A transvaginal probe for near infrared spectroscopic monitoring of the bladder detrusor muscle and urethral sphincter, Spectroscopy 22 (6) (2008), 429-436. [262] L. Skov, O. Pryds and G. Greisen, Estimation of cerebral blood flow in newborn infants: comparison of near infrared spectroscopy and 133Xe clearance, Pediatr Res 30 (1991), 570-573 [263] L. Skov and G. Greisen, Apparent cerebral cytochrome aa3 reduction during cardiopulmonary bypass in hypoxaemic children with congenital heart disease. A critical analysis of in vivo near-infrared spectrophotometric data, Physiol Meas 15 (4) (2004), 447-457. [264] S.G. Simonson, K. Welty-Wolf, Y.T. Huang, J.A. Griebel, M.S. Caplan, P.J. Fracica and C.A. Piantadosi, Altered mitochondrial redox responses in gram negative septic shock in primates, Circ Shock. 43 (1) (1994), 34-43. [265] S.G. Simonson and C.A. Piantadosi, Near-infrared spectroscopy, clinical applications, Crit Care Clin 12 (4) (1996), 1019-1029. [266] P. Smielewski, P. Kirkpatrick, P. Minhas, J.D. Pickard and M.Czosnyka, Can cerebrovascular reactivity be measured with near-infrared spectroscopy? Stroke, 26 (12) (1995), 2285-92. [267] B.R. Soller, P.O. Idwasi, J. Balaguer, S. Levin, S.A. Simsir, T.J. Vander, H. Collette, and S.O. Heard, Non-invasive near infrared spectroscopic measured muscle ph and po2 indicate tissue perfusion for cardiac surgical patients undergoing cardiopulmonary bypass, Crit Care Med 31 (9) (2003), 2324-31.
399
[268] J.S. Soul and J.S. du Plessis, New technologies in pediatric neurology, Near-infrared spectroscopy. Semin Pediatr Neurol 6 (2) (1999), 101-10. [269] M.G. Sowa, L. Leonardi, J.R. Payette, K.M. Cross, M. Gomez and J.S. Fish. Classification of burn injuries using near-infrared spectroscopy, J Biomed Opt 11 (5) (2006), 054002. [270] M.G. Sowa, L. Leonardi, J.R. Payette, K.M. Cross, J.S. Fish and H.H. Mantsch, Near infrared spectroscopic assessment of hemodynamic changes in the early post-burn period. Burns 27 (3) (2001), 241-9. [271] R. Springett, J. Newman, M. Cope and D.T. Delpy, Oxygen dependency and precision of cytochrome oxidase signal from full spectral NIRS of the piglet brain, Am J Physiol Heart Circ Physiol 279 (2000), H2202-H2209. [272] R. Springett, M. Wylezinska, E.B. Cady, V. Hollis, M. Cope and D.T. Delpy, The oxygen dependency of cerebral oxidative metabolism in the newborn piglet studied with 31P NMRS and NIRS, Adv Exp Med Biol 530 (2003), 555-63. [273] S. Srinivasan, B.W. Pogue, B. Brooksby, S.D. Jiang, H. Dehghani, C. Kogel, W.A. Wells, S.P. Poplack and K.D. Paulsen, Near-infrared characterization of breast tumors in vivo using spectrally-constrained reconstruction, Technol Cancer Res Treat 4 (5) (2005), 513-26. [274] L.A. Steiner and M. Czosnyka, Should we measure cerebral blood flow in head-injured patients? Br J Neurosurg 16 (5) (2002), 429-39. [275] J.Steinbrink, T. Fischer, H. Kuppe, R. Hetzer, K. Uludag, H. Obrig and W.M. Kuebler, Relevance of depth resolution for cerebral blood flow monitoring by near-infrared spectroscopic bolus tracking during cardiopulmonary bypass, J Thorac Cardiovasc Surg 132 (5) (2006), 1172-8. [276] L. Stothers, Reliability, validity, and gender differences in the quality of life index of the SEAPIQMM incontinence classification system, Neurourol Urodyn 23 (3) (2004), 223-8. [277] L. Stothers, A. Macnab and R. Gagnon, Changes in cytochrome C levels in the human bladder during the filling and emptying cycle, J Urol (Suppl) 173 (4) (2005), 353-354. [278] L. Stothers, A. Macnab and R. Gagnon, Non invasive urodynamics using near infrared spectroscopy in the human, J Urol (Suppl) 173 (4) (2005), 354. [279] L. Stothers, A. Macnab and R. Gagnon, Patterns in detrusor oxygenation during flow and pressure flow studies in men using near infrared spectroscopy (NIRS), J Urol (Suppl) 175 (4) (2006), 444. [280] L. Stothers, A. Macnab, and R. Gagnon, A description of detrusor cellular respiration during urodynamics in humans using non invasive near infrared spectroscopy, J Urol (Suppl) 175 (4) (2006), 446. [281] L. Stothers and A. Macnab, Near-infrared Spectroscopy (NIRS) changes in oxy and deoxyhemoglobin during natural bladder filling and voiding in normal volunteers, J Urol (Suppl) 177 (4) (2007), 506. [282] L. Stothers, B. Shadgan and A.J. Macnab, Urologic applications of infrared spectroscopy (NIRS), Can J Urol 15 (6) (2008), 4399-4409. [283] L. Stothers, B. Shadgan and A.J. Macnab, Functional near infrared spectroscopy (fNIRS); dynamic topography of the human bladder during voiding, J Urol (Suppl) 179 (4) (2008), 517. [284] L. Stothers and A. Macnab, Transvaginal near-infrared spectroscopy (NIRS) of the female mid urethra during voiding and pelvic floor exercises, J Urol (Suppl) 179 (4) (2008) 517-18. [285] S. Suzuki, S. Takasaki, T. Ozaki and Y. Kobayashi, A tissue oxygenation monitor using NIR spatially resolved spectroscopy, Proc SPIE 3597 (1999), 582-592. [286] T. Svensson, S. Andersson-Engels, M. Einarsdottir and K. Svanberg, In vivo optical characterization of human prostate tissue using near-infrared time-resolved spectroscopy, J Biomed Optics 129 (1) (2007), 014022. [287] I. Tachtsidis, C.E. Cooper, A.D. McGown, H. Makker, D.T. Delpy and C.E. Elwell, Changes in cerebral total haemoglobin volume and cytochrome oxidase redox state during deep apnoeas in patients with obstructive sleep apnoea, Optical Society of America Technical Digest, (2004) WF6. [288] I. Tachtsidis, M. Tisdall, T.S. Leung, C.E. Cooper, D.T. Delpy, M. Smith, C.E. Elwell, Investigation of in vivo measurement of cerebral cytochrome-c-oxidase redox changes using near-infrared spectroscopy in patients with orthostatic hypotension, Physiol Meas (28) 2007, 199-211. [289] G. Taga, K. Asakawa, A. Maki, Y. Konoshi and H. Koizumi, Brain imaging in awake infants by nearinfrared optical tomography, PNAS 100 (19) (2003), 10722-10727. [290] M.C. Taillefer and A.Y. Denault, Cerebral near-infrared spectroscopy in adult heart surgery: systematic review of its clinical efficacy, Can J Anaesth. 52 (1) (2005), 79-87. [291] M. Tamura, Non-invasive monitoring of brain oxygenation and metabolism during cardiopulmonary bypass by near infrared spectrophotometry, Jpn Circ J 55 (1991), 330-335. [292] M. Tamura, Y. Hoshi and F. Okada, Localised near-infrared spectroscopy and functional optical imaging of brain activity, Philos Trans R Soc Lon B Biol Sci, 352 (1354) (1997), 737-742 [293] Y. Teng, H. Ding, Q. Gong, Z. Jia, and L Huang, Monitoring cerebral oxygen saturation during cardiopulmonary bypass using near infrared spectroscopy: the relationships with body temperature and perfusion rate, J Biomed Optics, 11 (2) Mar-Apr (2006), 024016.
400
[294] C. Terborg, F. Gora and J. Rother, Reduced vasomotor reactivity in cerebral microangiopathy: a study with near-infrared spectroscopy and transcranial Doppler sonography, Stroke 31 (4) (2000), 924-9. [295] M.S.Thorniley, S. Simpkin, E. Balogun, K. Khaw, C. Shurey, K. Burton and C.J. Green, Measurements of tissue viability in transplantation, Philos Trans R Soc Lond B Biol Sci 352 (1354) (1997), 685-696. [296] M.M. Tisdall, I. Tachtsidis, T.S. Leung, C.E. Elwell, M. Smith, Near-infrared spectroscopic quantification of changes in the concentration of oxidized cytochrome c oxidase in the healthy human brain during hypoxemia, J.Biomed Opt 12 (2) (2007), 024002. [297] J.D. Tobias, Cerebral oxygenation monitoring: near-infrared spectroscopy, Expert Rev Med Devices. 3 (2) (2006), 235-43. [298] J.D. Tobias and D.G. Hoernschemeyer, Near-infrared spectroscopy identifies compartment syndrome in an infant, J Pediatr Orthop 27 (3) (2007), 311-3. [299] M.C. Toet, P.M. Lemmers, L.J. van Schelven and F. van Bel, Cerebral oxygenation and electrical activity after birth asphyxia: their relation to outcome, Pediatr 117 (2) (2006), 333-339. [300] F. Torella, S.L. Haynes and C.N. McCollum, Cerebral and peripheral oxygen saturation during red cell transfusion, J Surg Res 110 (1) (2003), 217-221. [301] T.A. Tortoriello, S.A. Stayer, A.R. Mott, E.D. McKenzi, C.D. Fraser, D.B. Andropoulos and A.C. Chang, A noninvasive estimation of mixed venous oxygen saturation using near-infrared spectroscopy by cerebral oximetry in pediatric cardiac surgery patients. Paediatr Anaesth 15 (2005), 495-503. [302] M. Tsuji, H. Naruse, J. Volpe and D. Holzman, Reduction in cytochrome aa3 measured by nearinfrared spectroscopy predicts cerebral energy loss in hypoxic piglets, Pediatr Res 37 (3) (1995), 253-9. [303] T. Tsukihara, H. Aoyama, E. Yamashita, T. Tomizaki, H. Yamaguchi, K. Shinzawa-itoh, R. Nakashima, R. Yaono and S.Yoshikawa, Structures of metal sites of oxidized bovine heart cytochrome c oxidase at 2.8 A, Science 269 (1995), 1069-1074. [304] D.T. Ubbink and B. Koopman, Near-infrared spectroscopy in the routine diagnostic work-up of patients with leg ischaemia, Eur J Vasc Endovasc Surg 31 (4) (2006), 394-400. [305] M.A. Underwood, J.M. Milstein and M.P. Sherman, Near-infrared spectroscopy as a screening tool for patent ductus arteriosus in extremely low birth weight infants. Neonatology 91 (2) (2007), 134-9. [306] U. Utzinger and R.R. Richards-Kortum, Fiber optic probes for biomedical optical spectroscopy, J Biomed Optics 8 (1) (2003), 121-147. [307] M.C.P. van Beekvelt, M.S. Borghuis, B.G.M. van Engelen, R.A. Wevers and W.N.Colier, Adipose tissue thickness affects in vivo quantitative near-IR spectroscopy in human skeletal muscle. Clin Sci (London) 101 (1) (2001), 21-28. [308] M.C.P. van Beekvelt, B.G.M. van Engelen, R.A. Wevers and W.J.M. Colier, In vivo quantitative nearinfrared spectroscopy in skeletal muscle during incremental isometric handgrip exercise, Clin Physiol & Fun Im 22 (2002), 1-8. [309] M.C.P. van Beekvelt, Quantitative near-infrared spectroscopy in human muscle. Methodological issues and clinical application, Thesis. University of Nijmegen. ISBN 90-9015688-7. (2002). 1-170. [310] F. van Bel, C.A. Dorrepaal, M.J. Benders, P.E. Zeeuwe, M van de Bor and H.M. Berger, Changes in hemodynamics and oxygenation in the first 24 hours after birth asphyxia, Pediatr 92 (3) (1993), 365-72. [311] H.C. van de Hulst, Light scattering by small particles, Dover Publications, New York, 1981, pp. 28-128. [312] P. van de Zee, M. Cope, S.R. Arridge, M. Essenpreis, L.A. Potter, A.D. Edwards, J.S. Wyatt, D.C. McCormick, S.C. Roth, E.O.R. Reynolds and D.T. Delpy, Experimentally measured optical pathlengths for the adult head, calf, and forearm, and the head of the newborn infant as a function of interoptode spacing, Adv Exper Med Biol 316 (1992), 143-153. [313] J.G. van den Brand, E.J. Verleisdonk and C. van der Werken, The diagnostic value of intracompartmental pressure measurement, magnetic resonance imaging, and near- infrared spectroscopy in chronic exertional compartment syndrome: a prospective study in 50 patients, Am J Sports Med 33 (5) (2005), 699-704. [314] M.C. van der Sluijs, N.J.M. Colier, R.J.F. Houston and B. Oesburg, A new and highly sensitive continuous wave near infrared spectrophotometer with multiple detectors, Proc SPIE. 3194 (1997), 63-72. [315] J.P. van Houten, D.A. Benaron, S. Spillman and D.K. Stevenson, Imaging brain injury using timeresolved near infrared light scanning, Pediat Res 39 (3) (1996), 470-76. [316] J.E. Varela, S.M. Cohn, G.D. Giannotti, M.O. Dolich, H. Ramon, J.A. Wiseberg, and M. McKenney, Near-infrared spectroscopy reflects changes in mesenteric and systemic perfusion during abdominal compartment syndrome, Surgery 29 (3) (2001), 363-70. [317] E.C. Vaux, D.J. Taylor, P. Altmann, B. Rajagopalan, K, Graham, R. Cooper, Y. Bonomo and P. Styles, Effects of carnitine supplementation on muscle metabolism by the use of magnetic reso-
401
[318]
[319] [320] [321] [322]
[323] [324]
[325]
[326] [327] [328]
[329]
[330]
[331] [332]
[333]
[334] [335]
[336] [337]
[338] [339]
[340]
nance spectroscopy and near-infrared spectroscopy in end-stage renal disease, Nephron Clin Pract (97) (2004) 41-8. F. Vernieri, F. Tibuzzi, P. Pasqualetti, N. Rosato, F. Pasarelli, P.M. Rossini and M. Silvestrini, Transcranial Doppler and near-infrared spectroscopy can evaluate the hemodynamic effect of carotid artery occlusion, Stroke 35 (1) (2004), 64-70. A.M. Vintzileos, S. Nioka, M. Lake, P. Li, Q. Luo and B. Chance, Transabdominal fetal pulse oximetry with near-infrared spectroscopy, Am J Obs Gynecol 192 (2005), 129-33. A.I. Vogel, A text-book of quantitative inorganic analysis including elementary instrumental analysis, 3rd Edition, Longmans Green & Co., London, 1961, pp. 740-742. K. von Siebenthal, G. Bernert and P. Casaer, Near-infrared spectroscopy in newborn infants, Brain & Development 14 (1992), 135-43. K. von Siebenthal, J. Beran, M. Wolf, M. Keel, V. Dietz, S. Kundu and H.U. Bucher. Cyclical fluctuations in blood pressure, heart rate and cerebral blood volume in preterm infants, Brain Dev, 21 (8) (1999), 529-34 H. von Vierordt, Die quantitative spectralanalyse in ihrer Anwendung auf Physiologie, Physik, Chemie und Technologie. Tubingen H. Laupp, ed. (1876). J. Wang, Y.J. Geng, B. Guo, T. Klima, B.N. Lal, J.T. Willerson, and W. Casscells, Near-infrared spectroscopic characterization of human advanced atherosclerotic plaques, J Am Coll Cardiol 39 (2002) 1305-1313. K.R. Ward, R.R. Ivatury , R.W. Barbee, J. Terner, R. Pittman, I.P. Filho and B. Spiess, Near infrared spectroscopy for evaluation of the trauma patient: a technology review, Resuscitation 68 (1) (2006), 27-44. Epub 2005 Dec 1. S.P. Wardle, C.W. Yoxall and A.M. Weindling, Determinants of cerebral fractional oxygen extraction using near infrared spectroscopy in preterm infants, J Cereb Blood Flow Metab 20 (2000), 272-79. E.B. Wassenaar and J.G. Van den Brand, Reliability of near-infrared spectroscopy in people with dark skin pigmentation, J Clin Monit Comput 19 (2005), 195-199. S.L. Watkin, S.A. Spencer, P.W. Dimmock, Y.A. Wickramasinghe and P. Rolfe, A comparison of pulse oximetry and near infrared spectroscopy (NIRS) in the detection of hypoxaemia occurring with pauses in nasal airflow in neonates, J Clin Monit Comput 15 (1999), 441-447. D.L.Wetzel, G.R. Post and R.A. Lodder, Sychrotron infrared micro spectroscopic analysis of collagens I, III, and elastin on the shoulders of human thin-cap fibroatheromas, Vib Spectrosc 38 (1-2) (2005), 53-9. Y.A. Wickramasinghe, L.N. Livera, S.A. Spencer, P. Rolfe and M.S. Thornilly, Plethysmographic validation of near infrared spectroscopic monitoring of cerebral blood volume, Arch Dis Child 67 (1992), 407-11. Y.A. Wickramasinghe, P. Rolfe, K. Palmer and S.A. Spencer, Investigation of neonatal cytochrome redox state, Dev Brain Res 89 (2) (1995), 307-8. M. Wolf, G. Duc, M. Keel, P. Niederer, K. von Siebenthal, and H.U. Bucher, Continuous noninvasive measurement of cerebral arterial and venous oxygen saturation at the bedside in mechanically ventilated neonates, Crit Care Med 25 (9) (1997), 1579-1582. M. Wolf, K. von Siebenthal, M. Keel, V. Dietz, O. Baenziger, and H.U. Bucher, Comparison of three methods to measure absolute cerebral hemoglobin concentration in neonates by near-infrared spectrophotometry, J Biomed Opt 7 (2) (2002), 221-227. M. Wolf, M. Ferrari and V.Quaresima, Progress of near-infrared spectroscopy and topography for brain and muscle clinical applications, J Biomed Optics 12 (6) (2007), 062104. U. Wolf, M. Wolf, J.H. Choi, M. Levi, D. Choudhury, S. Hull, D. Coussirat, L.A. Paunescu, L.P. Safonova, A. Michalos, W.W. Mantulin, and E. Gratton, Localized irregularities in hemoglobin flow and oxygenation in calf muscle in patients with peripheral vascular disease detected with nearinfrared spectrophotometry, JVasc Surg 37 (5) (2003), 1017-1026. A.J. Wolfberg and A.J. du Plessis, Near-infrared spectroscopy in the fetus and neonate, Clin Perinatol 33 (2006), 707-728. S. Wray, M. Cope, D.T. Delpy, J.S. Wyatt and E.O.R. Reynolds, Characterization of the near infrared absorption spectra of cytochrome aa3 and haemoglobin for the non-invasive monitoring of cerebral oxygenation. Biochim Biophys Acta 933 (1988), 184-192. J.S. Wyatt, Cerebral oxygenation and haemodynamics in the foetus and newborn infant, Phil Trans R Soc Lond B 352 (1997), 697-700. J.S. Wyatt, M. Cope, D.T. Delpy, P. van der Zee, S. Arridge, A.D. Edwards, S. Wray and E.O.R. Reynolds, Quantitation of cerebral blood volume in newborn infants by near infrared spectroscopy, J Appl Physiol 68 (1990), 1086-91. R.X. Xu and S.P. Povoski. Diffuse optical imaging and spectroscopy for cancer. Expert Rev Med Devices 4(1) (2007), 83-95.
402
[341] R.X. Xu, B. Qiang, J.J. Mao and S.P. Povoski, Development of a hand-held near-infrared imager for dynamic characterization of in vivo biological tissue systems, Appl Opt 46 (30) (2007), 7442-7451. [342] Y. Yang, O.O. Soyemi, M.R. Landry and B.R. Soller, Non-invasive in vivo measurement of venous blood pH during exercise using near-infrared reflectance spectroscopy, Appl Spectrosc 61 (2) (2007), 223-229. [343] K. Yoshitani, M. Kawaguchi, M. Iwata, N. Sasaoka, S. Inoue, N. Kurumatani and H. Furuya. Comparison of changes in jugular venous bulb oxygen saturation and cerebral oxygen saturation during variations of hemoglobin concentration under propofol and servofluorane anaesthesia, Br J Anaesth 994 (2005), 341-346. [344] K. Yoshitani, M. Kawaguchi, T. Okuno, T. Kanoda, Y. Ohnishi, M. Kuro, and M. Nishizawa, Measurements of optical pathlength using phase-resolved spectroscopy in patients undergoing cardiopulmonary bypass, Anesth Analg 104 (2) (2007), 341-6. [345] K. Yoshitani, M. Kawaguchi, N. Miura, T. Okuno, T.Kanoda, Y. Ohnishi and M. Kuro, Effects of hemoglobin concentration, skull thickness, and the area of the cerebrospinal fluid layer on nearinfrared spectroscopy measurements, Anesthesiology, 106 (3) (2007), 458-462. [346] A.E. Young, T.J.Germon, N.J.Barnett, A.R. Manara and R.J.Nelson, Behaviour of near-infrared light in the adult human head: Implications for clinical near-infrared spectroscopy, Br J Anaesth 84 (2000), 38-42. [347] C.W.Yoxall and A.M.Weindling, The measurement of peripheral venous oxyhemoglobin saturation in newborn infants by near infrared spectroscopy with venous occlusion, Pediatr Res 39 (6) (1995), 1103-08. [348] C.W. Yoxall, A.M. Weindling, N.H. Dawani, and I. Peart, Measurement of cerebral venous oxyhemoglobin saturation in children by near-infrared spectroscopy and partial jugular venous occlusion, Pediatr Res 38 (3) (1995), 319-323. [349] C.W. Yoxall and A.M. Weindling, Measurement of venous oxyhaemoglobin saturation in the adult human forearm by near infrared spectroscopy with venous occlusion, Med Biol Eng Comput 35 (4) (1997), 331-336. [350] C.W. Yoxall and A.M. Weindling, Measurement of cerebral oxygen consumption in the human neonate using near infrared spectroscopy: Cerebral oxygen consumption increases with advancing gestational age, Ped Res 44 (3) (1998), 283-90
403
The Use of Synchrotron Radiation for Biomedical Applications of Infrared Microscopy

Lisa M. MILLER a, Mark J. TOBIN b, Sirinart CHIO-SRICHAN c and Paul DUMAS 1,c
National Synchrotron Light Source, Brookhaven National Laboratory, Upton, NY 11973 USA b Australian Synchrotron, 800 Blackburn Road, Clayton, Victoria, 3168, Australia c SOLEIL Synchrotron, Saint-Aubin BP 48, 91192 Gif-sur-Yvette Cedex, France
a
Abstract. Synchrotron infrared microscopy exploits the so-called brightness (or brilliance) advantage of synchrotron radiation. The main characteristics of such a source, coupled with an infrared microscope, are reviewed. Several important applications in the biomedical field are summarized including drug effects on cancer cells, growth factor signaling in single cells, compositional changes in microdamaged bone, infectious prion protein structure in nervous tissues, human substantia nigra in Parkinsons disease and -amyloid deposition in Alzheimers disease. Keywords. Synchrotron radiation; infrared microscopy, vibrational spectroscopy
1. Introduction Vibrational spectroscopy has become a well-accepted analytical tool for biological and biomedical applications [1]. As explained in more detail in several chapters of this book, the primary reason is that many common biomolecules, such as nucleic acids, proteins, lipids and carbohydrates, have characteristic and well-known vibrational fingerprints, which has led to many important and extensive investigations of biological samples by IR spectroscopy [2,3]. Nevertheless, the intrinsic heterogeneity and complexity of biological cells and tissues have demonstrated the need for microscopic characterization. Microscopic analysis with IR spectroscopy, i.e. IR microspectroscopy, is achieved by coupling a modified light microscope with an associated IR spectrometer [4]. Over the past 20+ years, infrared microspectroscopy has been used to reveal a wide range of information on the biochemical makeup of numerous cells and tissues [5,6]. Today, it is commonly accepted that the chemical pathology of samples can be characterized by their infrared signatures, especially when the IR spectra are spatially resolved [7]. The applicability of IR microspectroscopy to biological and pathological problems depends on the achievable level of spectral and spatial detail. IR microspectroscopy has been more widely employed than Raman microspectroscopy, due to the superior signal-to-noise (S/N) obtained; however the latter technique is increasing in popularity [8].
1 Corresponding Author: Paul DUMAS, SOLEIL Synchrotron, Saint-Aubin BP 48, 91192 Gif-sur-Yvette Cedex, France.
404
L.M. Miller et al. / The Use of SR for Biomedical Applications of Infrared Microscopy
Higher brightness infrared sources are necessary and they do exist (e.g. lasers), but the infrared beam emitted from a synchrotron light source is the only white source that delivers the entire infrared spectral range, i.e. from far-IR to near-IR, with an exceptionally high brightness compared to conventional laboratory sources [9]. The specific aims of this chapter are to familiarize readers with the properties of synchrotron radiation, the design and use of synchrotron IR beamlines, the key issues that pertain specifically to synchrotron-based IR microspectroscopy (SR-FTIRM), and the important benefits of SR-FTIRM for biomedical applications. Rather than presenting a comprehensive review of all up-to-date biomedical applications of SR-FTIR spectromicroscopy, which have recently been reviewed [6], we shall focus on contents that illustrate the requirements and utility of SR-FTIRM as a chemical probe for tracking biomolecular changes relevant in biomedicine. Moreover, we will describe how this technique will complement the large scale systematic studies ongoing with conventional instruments for the development of future laboratory-based diagnostic tools [10].
2. The Synchrotron Infrared emission from a synchrotron light source was proposed many years ago as a beneficial source for throughput-limited IR experiments [1115]. The initial applications of synchrotron IR light focused on long wavelength applications (i.e. often called the far-IR or THz range) and were carried out at Tantalus, Daresbury and BESSY I, starting in the 1970s [16]. The real boost in infrared activities at synchrotron facilities started shortly thereafter, at UVSOR [17], NSLS [1820], and later LURE SuperACO [21]. All beamlines were designed with reasonably large apertures in order to efficiently extract the infrared photons (see below). However at this time, electron beam motions in the synchrotron ring generated very noisy photon beams. In addition, the beam current, which is the primary factor that determines IR intensity, was quite limited (see next section). Most science programs utilizing such infrared beamlines concentrated on surface science and solid state physics applications [22,23]. However, synchrotron-based infrared applications exploded with the marriage of synchrotron IR light and the infrared microscope, where the brightness advantage of the synchrotron source was demonstrated for microanalysis [24,25]. The high brightness of synchrotron IR radiation has impacted the field of infrared microspectroscopy in a variety of scientific disciplines, including hard and soft condensed matter [24,26], geology [27], and biology [2832].The principal motivation for synchrotron-based IR microspectroscopy is to achieve significantly greater lateral resolution (typically at the diffraction limit) while recording data of superior signal-to-noise characteristics without resorting to prohibitively long acquisition times. Accordingly, facilities for performing IR spectroscopy have expanded throughout the world in response to an increasing demand for beamtime from the scientific community. In North America, the NSLS, New York presently operates five IR microscopes, while IR microscopes can also be found at the ALS, Berkeley, CAMD, Baton Rouge and SRC, Stoughton (USA); the CLS, Saskatoon (Canada). In Asia, UVSOR, Okasaki and SPring8, Nishi-Harima (Japan) and NSRRC, Hsinchu (Taiwan) operate IR microscopes. In Europe, IR activities exist at the SRS, Daresbury (UK); ESRF, Grenoble (France); MAXLAB, Lund (Sweden); DANE, Frascati (Italy); ELETTRA, Trieste (Italy); ANKA, Karlsruhe (Germany) BESSY II, Berlin (Germany), and SOLEIL, Paris
405
Figure 1. Schematic drawing of the infrared emission from edge radiation (ER) and bending magnet radiation (BM). L is the length of the straight section, and this influences the interference effects with the upstream ER dipole emission.
(France). The Australian Synchrotron, Melbourne and the Swiss Light Source (Villigen-Switzerland) also have a new IR programs opened to users. Other facilities that are planning IR spectroscopy programs include Diamond, Didcot (UK); DELTA, Dortmund (Germany); Metrology Light Source (MLS) Berlin (Germany); ALBA (Barcelona-Spain); NSRC (Thailand); the Pohang Accelerator Laboratory (Korea); NSRC (Thailand); INDUS1 (India); and NSRL, Hefei (China).
3. The Making of Synchrotron Light Electron-based synchrotron light sources use magnetic fields to bend the electrons into a closed orbit. Synchrotron radiation (SR) is produced at each of these bending magnets. The emitted radiation spans an extremely broad spectral domain, extending from the X-ray regime to the very far infrared region. Infrared radiation is generated by electrons traveling at relativistic velocities, either in a curved path through a constant magnetic field (i.e. bending magnet radiation [11]) or when their trajectories encounter variable magnetic fields, e.g. at the edges of bending magnets (i.e. edge radiation) (Fig. 1). In the latter, less-exploited case, the edge radiation (ER) is generated by a highenergy charged particle when it passes the entrance or exit through the region of a rapid change of magnetic field amplitude of a bending magnet. The photons emitted at two adjacent bending magnets bounding a straight section, appear in the same cone and are subsequently synchronized by the electron itself. This leads to an interference effect, which manifests itself in oscillations of radiation intensity [33,34]. Flux and brightness for the two types of infrared emission are almost equivalent, but the opening angle of the edge radiation is narrower than that of the SR from constant field of a bending magnet. The intensity profiles of the two types of emission are also different, and also depend upon wavelength. In Fig. 2, the distribution profiles at two wavelengths (10 m and 100 m) have been calculated for a medium energy storage ring (3.0 GeV) for a prototypical opening angle of the dipole chamber of 20 mrad (v) 40 mrad (h) for the collection of bending magnet infrared emission, and 20 mrad (v) 20 mrad (h) for the collection of the edge radiation infrared emission. These profiles illustrate that, for bending magnet emission, the angle of emission is larger than that of the edge radiation.
406
= 1000 cm-1
20mm Vertical Position
= 100 cm-1
20 Vertical Position 10 0 -10
10 0 -10 -20
BM
-20mm
-40mm
-20
0 Horizontal Position
20
40
-40mm
-20
0 Horizontal Position
20
40
20
20mm
10
10
Vertical Position
Vertical Position
ER
-10
-10
-20mm -20mm -10 0 Horizontal Position 10 20
-20 -20mm -10 0 Horizontal Position 10 20
Figure 2. Intensity distribution of edge radiation, for two wavelengths, = 100 m (left column) and = 10 m (right column). The distribution has been calculated for a bending magnet (BM) emission, and an edge radiation (ER) emission. This calculation has been carried out for the case of a storage ring of 3 GeV, and a vertical aperture of 20 mrad.
/LQHDU +RUL]RQWDO
20
/LQHDU 9HUWLFDO
20
20
&LUFXODU 5LJKW
20
&LUFXODU OHIW
10
10
10
10
ER
Vertical Position
Vertical Position
Vertical Position
Vertical Position
-10
-10
-10
-10
-20mm -20mm -10 0 Horizon t al Position 10 20
15 10
15 10 Vertical Position
Vertical Position
15 10
15 10 Vertical Position
-20mm -10 0 10 Horizontal Position 20
Vertical Position
5 0 -5 -10
5 0 -5 -10
5 0 -5 -10
5 0 -5 -10
BM
-15mm -20mm -10 0 10 Horizontal Position 20
-15mm -20mm -10 0 10 Horizo ntal Position 20
-15mm
-15mm -20mm -10 0 10 Horizontal Position 20
Figure 3. Intensity distribution for various polarization state for edge radiation (ER) and bending magnet radiation (BM).
Edge radiation has some advantages for engineering purposes, since the dipole chamber exit port in a synchrotron storage ring has to be enlarged for bending magnet radiation. Thus today, both sources are used for extraction, depending upon availability of a straight section and/or bending magnet. A marked difference between the two sources is their polarization properties (Fig. 3). Compared to bending magnet emission, which is strictly horizontally polarized in the plane of the electron trajectory, edge radiation has a radial polarization. 3.1. Flux vs. Brightness The brightness of synchrotron light is defined as the photon flux or power emitted per source area and solid angle. Most of the experiments in synchrotron IR spectroscopy exploit the brightness advantage of the source, while the flux advantage is also exploited in the far infrared region. It is important to note that synchrotron radiation does not provide a much higher photon flux than a conventional IR source (such as a Globar), unlike the traditional case in the X-ray region. In fact, the total flux can be much
407
1E-3
Flux ( in Watts/cm-1)
1E-4
T=2000K
1E-5 Synchrotron E=2.75 GeV
1E-6
Synchrotron E= 0.8 GeV
1E-7 1 10 100
Wavenumbers Wavelength ( m)
Figure 4. Comparison of the infrared flux, expressed in watts/cm1, between a blackbody heated up to 2000 K, and emitted into the 2 solid angle, and a synchrotron emission (emitted in a much narrow solid angle, as explained in the text), such as NSLS (0.85 GeV nominal energy) or SOLEIL (2.75 GeV nominal energy).
less for low synchrotron ring currents (about one to two orders of magnitude) in the mid infrared region, although it becomes superior at long wavelengths (Fig. 4). The crucial factor is that the effective synchrotron source size is quite small, i.e. on the order of ~100 m or less, for newer synchrotron storage rings. Thus, the light is emitted into a narrow range of angles, and the resulting brightness is vastly increased. More specifically, one can estimate the value of the brightness (also called brilliance or spectral radiance) to be:
B ( ) = Fsrc . x . y . x y
where Fsrc is the photon flux, x and y are the RMS dimensions of the photon beam in x and y directions respectively, and x and y are the emission angles, taken at the photon beam waist. Following this equation, one can calculate the source brightness ratio between the synchrotron source and the blackbody source.
BSynchrotron ( ) Bblackbody ( ) =
( Fsrc )Synchrotron ( x . y . x . y ) Blackbody . ( Fsrc )Blackbody ( x . y . x . y ) Synchrotron
2.102
Thus, the apparent brightness of synchrotron radiation can be 23 orders of magnitude higher than that of a Globar source. However, it is interesting to note that sample heating (and therefore sample damage) is negligible, permitting analysis of single cells for time scales from hours to days [35]. 3.2. Lateral Resolution When considering the available spatial resolution, two issues are important to take into account. The first is the acceptable signal-to-noise (S/N) of the data. The flux and thus S/N decreases drastically as apertures in the IR microscope are reduced to confine the IR beam to smaller areas. The second issue is diffraction (residual optical aber-
408
Figure 5. Simplified schematic of an infrared beamline. The emitted photons are first deviated by a flat mirror, and the beam is then focused onto a diamond window. After passing through the window, which separates the ultra high vacuum of the storage ring from the rough vacuum or purged atmosphere of the beamline, the beam is transported to the spectrometer.
rations of the focusing optical elements are neglected). The resolution issue has been evaluated both theoretically and experimentally by G.L. Carr [36]. The diffraction limit is achieved when the microscopes apertures define a region with dimensions approximately equal to the wavelength of interest, using the commonly available microscope objectives. However, the use of a confocal optical arrangement leads to a 30% improvement in the spatial resolution, in agreement with diffraction theory [36].
4. Infrared Beamlines and Microscopes 4.1. Beamlines Extraction of IR light from a synchrotron storage ring is generally accomplished with a combination of gold- or aluminum-coated plane and toroid/ellipsoid or spherical mirrors (Fig. 5). In most of the existing infrared beamlines, the first mirror is flat and deviates the beam either vertically or horizontally to the first focusing mirror. This extraction mirror must be able to handle the heat load of higher energy photons (i.e. x-rays); so water-cooling, water-cooled masks, and/or slotted mirrors are often employed. Combinations of focusing and flat mirrors (mostly metallic and often gold-coated) are then used to transport the beam outside the shield wall of the storage ring. The IR light is then focused through an IR-transparent window (usually diamond, but in a few cases IR transparent windows such as KBr, ZnSe and KRS5 have been used), which separates the ultrahigh vacuum (UHV) conditions of the storage ring (1091010 mbar) and the rough vacuum of the IR beamline (103104 mbar). This diamond window is usually slightly wedged to avoid interference fringing, and is small in diameter (10 to 40 mm clear aperture). Some sophisticated wheels containing windows of several materials (e.g. calcium fluoride, diamond, cesium iodide) have been recently implemented at the Swiss Light Source for the same purpose [37]. The rough vacuum of an IR beamline is generally terminated with an IRtransparent window (KBr, CsI, polyethylene) either after a vacuum-based IR spectrometer or prior to a purged instrument. This window isolates the beamline vacuum
409
from the ambient pressure of the spectrometer and/or microscope. To date, all commercial IR microscopes operate at ambient pressure and are typically purged with dry N 2 or dry air to remove any water vapor and carbon dioxide (CO2) from the IR spectrum. 4.2. Infrared Microscopes and Spatial Resolution Infrared microscopes are commercially available from a number of companies worldwide. In recent years, these microscopes have been improved so that they now include many of the features of research-grade optical microscopes such as polarization, Differential Interference Contrast (DIC), and epifluorescence. They are also equipped with sophisticated software for generating and analyzing chemical images. Typically, IR microscopes are configured in one of two ways for FTIR microspectroscopy (FTIRM) or FTIR Imaging (FTIRI). In a few cases, single IR microscopes can operate in both configurations. For an FTIRM instrument, a small area (a point) is spectroscopically sampled by the instrument, and an image is built-up by raster-scanning the specimen through the focused beam. Since only a single point is sampled at a time, these instruments use a single-element detector. The microscope uses reflecting Schwarzschild-type objectives to avoid absorption and chromatic aberrations over the large mid-IR spectral range. One objective serves to focus the light onto the specimen, while the other collects the light and sends it to the detector. An aperture is used to constrain the illuminated and/or detected area on the specimen. When examining the chemical makeup of biological cells and tissues with an IR microscope, it is important to achieve sub-cellular spatial resolution. The spatial resolution is limited by the wavelengths of IR light, which are longer than visible light wavelengths used for conventional optical microscopy. The diffraction-limited spatial resolution is dependent upon the wavelength of light and the numerical aperture (NA) of the focusing optic [38]. Typical IR microscopes utilize Schwarzschild objectives with a NA of ~0.6. Some FTIRM microscopes utilize a single aperture before the sample, which controls the region illuminated. With a single aperture, the diffraction-limited spatial resolution is approximately 2/3 [36]. Thus for the mid-IR range, the diffraction-limited spatial resolution is approximately 1.7 m (at 4000 cm1) 13 m (at 500 cm1). Other microscopes operate in a confocal arrangement, where a second aperture is used after the sample to define the region being sensed by the IR detector. For such a confocal microscope, where objectives and apertures are placed both before and after the sample, the spatial resolution is improved to ~/2 [36]. In addition, the confocal arrangement also reduces the Schwarzschilds first- and higher-order diffraction rings, dramatically improving image contrast [36]. To date, most synchrotron-based infrared microscopes operate in a confocal (i.e. dual-aperture) arrangement and utilize a single-element IR detector. For these microscopes, very few modifications are needed in order to adapt the commercial instrument to a synchrotron infrared source. In most cases the synchrotron beam is collimated as it enters the interferometer, which modulates the wavelength of light, and then is directed towards the IR microscope. 4.3. Confocal Geometry and Contrast Fidelity Due to their complexity and heterogeneity, analysis of biomedical samples benefits greatly from high quality data collected at high spatial resolution. As such, domain sizes close to the diffraction limit are important and blurring effects can distort the
410
(a)
(b)
(c)
Figure 6. Calculated images for a 14 m diameter ring-shaped object at = 6 m. (a): actual object; (b): image for a non-confocal (single aperture) 32 Schwarzschild objective with NA = 0.65; (c): image for a confocal (dual aperture) 32 Schwarzschild objective with NA = 0.65. (From [39].)
(a)
(b)
( c)
Figure 7. (a) Light microscopy and (b, c) IR images of the lipid distribution throughout a cross-section of a human hair. The analysis was performed on the same section with (b) a microscope employing an internal source and an array detector, and (c) with a synchrotron-powered infrared microscope with a single detector element.
spectroscopic information. Since this aspect is a fundamental advantage of the infrared synchrotron source the following illustrates the importance of preserving the contrast fidelity. As an example of the importance of a confocal geometry, Fig. 6 shows a ringshaped test pattern with a 14 m ring diameter and a thickness of 2 m. Figures 6b and 6c represent the calculated chemical image with a 6 m wavelength for a single aperture (non-confocal geometry) and two apertures (confocal geometry), respectively. It is very clear that, for the non-confocal geometry, one may erroneously conclude that there is a chemical component with a characteristic frequency at 6 m. This shows that an artifact can be introduced for high spatial resolution. To further illustrate this point, we have recorded the chemical image of lipids (at 2920 cm1) for a cross-section of human hair using either a confocal synchrotron IR microscope or a thermal source-based microscope equipped with an array detector. In both case, the projected size of the aperture was approximately ~6 6 m 2. Marked differences between the two images can be observed in Fig. 7. First, the lipids content at the center of the hair section appears to be much higher with the conventional microscope equipped with array detector (Fig. 7b). Second, and more importantly, the presence of lipids in the outside layer of the hair section, the cuticle, is not revealed with a non-confocal arrangement. The two above examples illustrate the main advantage of using the synchrotron source in biological specimens: one can preserve the contrast fidelity at very high spatial resolution, while maintaining a high spectral quality (i.e. a high signal-to-noise
411
(a) (b)
3500 3000 2500 2000 1500
Wavenumbers (cm-1)
Figure 8. Infrared spectra recorded with a dual aperture of 6 6 m2, at 4 cm1 resolution, coadding 128 scans with (a) the synchrotron source and (b) with the internal (Globar) source illustrating the improved S/N with the synchrotron IR source.
ratio), which are both essential for detecting subtle biochemical changes in cells and tissues. 4.4. Spectral Quality Improvements Using Synchrotron Infrared Microscopy In order to illustrate the superior spectral quality (signal-to-noise), while importantly preserving the contrast fidelity (afforded by a confocal geometry), we have compared the spectra obtained on the same individual cell (HELA cell), using either the synchrotron beam, or the internal source, with an aperture of 6 6 m2, corresponding roughly to the size of the nucleus of this type of cell (Fig. 8).
5. Applications It is well accepted that infrared microspectroscopy can provide a direct indication of a samples biochemistry [6]. Armed with information on sample histology and pathology, variations in nucleic acid, protein, and lipid content or structure can provide important details about the chemistry of diseased states. Specifically, the dominant absorption features in lipid spectra are found in the region from 28003000 cm1, and are assigned to the asymmetric and symmetric C-H stretching vibrations of CH3 (2956 and 2874 cm1) and CH2 (2922 and 2852 cm1). In addition, strong bands at 1736 cm1 arise from ester C=O groups in the lipid. Protein spectra have two primary features, the Amide I (16001700 cm1) and Amide II (15001560 cm1) bands, which arise primarily from the C=O and C-N stretching vibrations of the peptide backbone, respectively. The frequency of the Amide I band is particularly sensitive to protein secondary structure [4042]. In nucleic acids, the region between 10001500 cm1 contains contributions from asymmetric (1224 cm1) and symmetric (1087 cm1) PO2 stretching vibrations. The assignments of various spectral features in biological samples have been the subject of numerous publications as are detailed in several chapters of this book.
412
Aperture dimension
3x3 m2 6x6 m2 9x9 m2 12x12 m2

Wavenumbers (cm-1)

Figure 9. Infrared spectra recorded on the same cell, at different size of the projected aperture centered on the single cell. A marked change in the Amide I band shape is evident for apertures below 6 6 microns. This corresponds to the average size of the nucleus. This shows that the use of confocal geometry, combined with the high brightness of the synchrotron source, reveals the specific secondary structure of proteins inside the cell nucleus.
5.1. Sub Cellular Analysis of Cancerous Cells Cancer screening of tissues using conventional infrared microscopy is an active area of ongoing research [43]. Complementarily, synchrotron infrared studies provide unique information for diagnosis of the states of single cells. This is clearly exemplified in the following study of a drug interaction with individual cancer cells [44]. Photodynamic therapy (PDT) is a technique employing a photosentitizer that, after penetration into cells, exhibits anticancer activity when irradiated with light of wavelength matching its main absorption band [44]. In this example, HeLa cells were studied interacting with Hypocrellin A (HA), a lipid-soluble peryloquinone derivative isolated from natural fungus sacs of Hypocrella bambusae. HA exhibits an important absorption band at around 460 nm [44]. HeLa cells incubated with various concentration of HA, and irradiated at 460 nm during variable exposure time, were studied using MTT assays (rapid colorimetric assay for cellular growth and survival). The lowest rate of survival was found for an incubation in 10 M of HA solution, and subsequent illumination with a light dose of 48 J cm2 [44]. Several individual cells were probed with synchrotron infrared microscopy, while varying the probed area from 3 3 m2 up to 12 12 m2. The experiments were performed at the infrared beamline at ESRF (ID21-IR). Figure 9 displays the spectra recorded from the same HeLa cell at various aperture sizes. It is evident that the Amide I band exhibits an upward frequency shift when the aperture is larger than 6 6 m2. This is roughly the average size of the HeLa cell nucleus. This behavior can be interpreted as a higher contrast provided by probing subcellular regions containing mainly the nucleus, which requires the confocal geometry and synchrotron IR source. At larger apertures, an averaging of the infrared signal modifies the band position of the Amide I. The marked change of the Amide I band inside the nucleus is related to the specific secondary structure of proteins inside the nucleus and this is consistent with the fact that such a change is often correlated with the cell apoptosis, i.e. programmed cell death [45,46]. Such behavior is further confirmed by recording spectra across one diameter of one cell, with an aperture of 3 3 m2 (Fig. 10).
413
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
Wavenumbers (cm-1)
Figure 10. Linear scan recorded across an individual HELA cell with a 3 3 m2 aperture and a step of 3 m. Spectra were acquired at 4 cm1 resolution and with 128 accumulations per spectrum.
Figure 11. (A): Light micrograph showing the area probed inside HeLa cells using synchrotron infrared microscopy. The dimension of the projected aperture on the sample was set to 6 6 m2 (represented as a dashed square), allowing analysis of either the nucleus or the cytoplasm of each cell. (B): PCA analysis of the whole set of spectra taken inside the nucleus and the cytoplasm of about 100 cells. There is a clear separation of the spectra into two clusters, where the discrimination is 92% along the PC1 axis.
The diagnostic capability, for biomedical application, is readily strengthened by using synchrotron infrared microscopy. For HeLa cells manifesting a very low rate of survival after HA interaction and light exposure, the biochemical signature of the cell death is revealed, when a large number of analysis are made either in the cytoplasm or the nucleus (such a differentiation is due to the sub-cellular resolution achievable with the synchrotron source). A large set of infrared spectra, recorded either inside the cytoplasm or the nucleus of about 100 individuals cells (Fig. 11A), has been analyzed using Principal Components Analysis (PCA). The scores plot shown in Fig. 11B exhibits a clear separation of two classes: spectra recorded inside the cytoplasm and those recorded inside the nucleus. Correspondingly, the loadings plot (Fig. 12) show that the spectral variation derives mainly from the secondary structure difference between these two sub-regions of the cells, with a more pronounced -type secondary structure inside the nucleus. Other studies have shown that spectral differences exist between normal and cancerous oral epithelial cells [47,48] healthy and nutrient-repleted Micrasterias hardyi
414
0.4
0.3
0.2
PC1 Loadings
0.1 0.0
-0.1 -0.2
1525 cm-1
-0.3 1300 1400 1500
1630 cm-1
1600 1700 1800
Wavenumbers ( cm-1)
Figure 12. Loadings plot of PC1, showing that, for spectra lying along the negative values of PC1 axis, -sheet vibrational bands are responsible for the marked loading difference, as manifested by the peaks at 1630 and 1520 cm1, and are predominantly located inside the nucleus.
algal cells [49] and HepG2 cells exposed to low doses of 2,3,7,8-tetrachlorodibenzo-pdioxin [35]. Variations in DNA/RNA content and packing have also been demonstrated during the cell cycle of human lung epithelial cells [45] as well as in prostate cancer cells [50]. Individual mouse hybridoma B cells have been examined during necrosis and the end phases of mitosis [51], and also during the process of apoptosis [52]. 5.2. Growth Factor-Dependent Signaling and Downregulation in Single Cells The Epidermal Growth Factor receptor (EGFR) mediates the biological signal from several polypeptide mitogens, including Epidermal Growth Factor (EGF) and Transforming Growth Factor- (TGF-), and EGFR consequently plays a crucial role in the regulation of cell proliferation, cell morphology and other cell functions such as endocytosis and exocytosis. To develop a more fundamental understanding of the cellular processes involved in the transition of cells from normal to neoplastic growth, A431 carcinoma cells have been studied using synchrotron IR microspectroscopy under the stimulus of the key cell growth regulatory hormone EGF and key changes in the infrared spectrum that relate to the activation of the growth factor signaling mechanism have been detected. Many research groups have applied FTIR microspectroscopy in the study of isolated or cultured cells, which can be prepared or grown in controlled conditions, including examples reported elsewhere in this chapter. These studies have included those of chronic lymphocytic leukemia [53] where differences were detected in DNA, protein and lipid content between normal and leukemic cells. In a study closely related to the study of growth factor signaling reported here, fibroblasts have been studied following transfection with the H-ras oncogene [54]. In normal cells, the H-ras gene product is an important intermediate in the signaling pathway of several growth hormones, but it is often found to be over-expressed in cancer cells. This results in a disruption of the regulatory pathway controlling DNA replication and cell division. Comparison of fibroblasts expressing the H-ras protein with those not transfected showed differences in the levels of phosphate and carbohydrate absorption bands, as well as a difference in the RNA/DNA ratio. The growth factor signaling pathway, in which H-ras plays a role, leads ultimately to DNA replication and cell division, and both growth factors and Hras have been implicated in regulation of the progression of oral epithelial cells from proliferation to apoptosis. Result have shown that the spectral differences observed for different stages of the cell cycle for myeloid leukemia (ML-1) cells, are similar to the
415
differences observed between normal and abnormal exfoliated cells implying that the stage of the cell cycle is an important contributing factor to the variability of the IR absorption spectra of cells within apparently similar tissues [55,56]. The IR microscopy beamline at Daresbury Laboratory was used to collect absorbance spectra from individual cervical epithelial cells (A431, European Collection of Animal Cell Cultures, Porton Down, UK) grown under a range of culture conditions. A431 cells are derived from a cervical epithelial carcinoma, and are known to overexpress EGFR. Cells were routinely cultured on infrared reflecting slides at EGF concentrations between 109 M and 107 M, plus controls of PBS without EGF. Spectra were collected at 8 cm1 spectral resolution with a sampling area of 15 15 m and coadded for 128 scans for samples with time intervals from between 1 min and 3 h after addition of the EGF. In order to assess spectral differences between cells, a measurement was made of the following spectral features for each cell examined: (i) Amide I peak position (ii) Amide I peak area (corrected for baseline between 790 cm 1 and 1800 cm1) (iii) Amide II peak position (iv) Amide II peak area (corrected for baseline between 790 cm1 and 1800 cm1) (v) 1735 cm1 C=O stretch peak area (baseline corrected from 1723 cm1 to 1760 cm1, and normalized against amide I area) (vi) 970 cm1 peak area (baseline corrected from 940 cm1 to 980 cm1, and normalized against amide I area) (vii) 1170/1155 cm1 peak area ratio (baseline corrected between 1142 cm1 and 1182 cm1). Cross correlation of pairs of spectral features showed EGFdependent changes in the spectra of the cells. Most prominent was the variation in the correlation between the 970 cm1 absorption peak area and the peak position of the amide I protein band with time after EGF stimulation. Figure 13 shows the correlation between these spectral parameters for cells exposed to 1 108 M and 1 109 M EGF 1 min, 10 min, 45 min and 3 h after incubation (each point represents one cell). After 1 min exposure at 1 108 M (Fig. 13a), the cells examined showed a correlation distribution similar to that observed in untreated cells. After 10 min approximately half of the cells showed both increased 970 cm1 peak area and a shift in the amide I peak position to higher wavenumbers (b), with a smaller proportion of cells continuing to show this relationship after 45 min (c). This change in correlation could still be detected after 3 h (d). At 1 109 M EGF (a concentration close to physiological levels of EGF) a spectral distribution similar to that observed at 10 min with 1 10 8 M was observed after only 1 minute (Fig. 13e). This correlation distribution was still in evidence after 10 min and 45 min (f, g), but had largely returned to the un-stimulated distribution after 3 h (h). Exposure of cells to high concentrations of EGF (1 10 7 M) resulted in a slight change in the correlation distribution after 45 min (data not shown), but the effect was weaker than that observed at 1 108 and 1 109 M. The shift observed in the EGF stimulated cells from around 1648 cm1 towards 1664 cm1 is characteristic of a shift from highly -helical structure towards a higher level of turns and bends. This may be spectroscopic evidence of increased protein translation in response to the mitogenic signal of the EGF, resulting in elevated levels of incompletely folded protein within the cells. The detection of a correlated increase in the weak absorption peak at 970 cm1 can be interpreted in one of two ways. Firstly, it may be as a result of increased protein phosphorylation following stimulation of the EGF signaling cascade, since Sanchez-Ruiz and Martinez-Carrion [57] have shown increases in a similar absorption in phosvitin and ovalbumin produced following phosphorylation of the proteins at specific residues. The phosphorylation of signaling peptides, including a domain of the EGF receptor itself, by protein kinases, is a crucial step
416
Figure 13. Correlation between 970 cm1 absorption peak area and amide I peak position for A431 cells at time after addition of 1 108 M and 1 109 M EGF.
in the translation of the EGF stimulus into the stimulus for DNA replication and cell division. It is therefore possible that the increased absorption at 970 cm 1 observed after EGF stimulation is a result of the triggering of this phosphorylation cascade. An alternative interpretation for this absorption has been suggested, pointing to bulk changes in the DNA absorption properties of the cell, with the DNA phosphate backbone being responsible for the absorption. Diem and coworkers [48,56,58] suggest that the decondensation of the nuclear DNA in advance of replication leads to a change in the DNA absorption, including a band near 970 cm1. The evidence from our data of a slower signaling event at higher EGF concentrations may be as a result of receptor down regulation within the first minute after addition of EGF. A431 cells overexpress the EGF receptor to a very high level (3 106 receptors per cell) and the cells, in response to large stimuli, employ signal down regulation through the removal of large numbers of receptors from the cell membrane. Stimulation of the cells with EGF concentrations above physiological levels can therefore lead to a block in further propagation of the EGF signal and eventually to cell death. Figure 14 shows two cells from a group of A431 cells mapped at a step size of 1 micron giving a diffraction limited spatial resolution of the order of 3 microns at 1650 cm1 and is included to illustrate the capability of studying the cytoplasm, nucleus and cell membrane of individual cells by synchrotron FTIR using a confocal apertured microscope system. In each of the two cells, a CHrich crescent can be seen around the nucleus and probably indicates the location of endoplasmic reticulum or the Golgi apparatus of the cells. There is evidence too that one cell (upper cell) is undergoing cell division as a clear protein signal is observed from the de-condensed nucleus, and the ester carbonyl map suggests reorganization of the nuclear membrane across the dividing nucleus. This work has shown growth factor dependent changes in the DNA and protein IR absorption in cultured cells, with evidence for down regulation of the EGF signaling
417
Figure 14. FTIR maps of two A431 cells showing subcellular structure. Bar = 20 microns.
mechanism at higher growth factor concentrations. In particular, the ability of synchrotron IR microscopy to make such measurements at the single cell and sub-cellular level has been clearly established. 5.3. Chemical Makeup of Microdamaged Bone Bone is made up of organic and mineral components, mainly type I collagen which is mineralized with nanocrystalline biological apatite. Cationic calcium and anionic phosphate substitutions make it difficult to determine the exact structure and properties of bone apatite, because they can affect bones solubility, density, hardness, and growth morphology [59,60]. Additionally bone composition has been shown to change with age [6164] and is also affected by disease [6568] and treatment for disease [69]. During normal daily activities, tiny cracks can develop in bone called microdamage. These cracks act to absorb and disperse energy as bones are loaded. In healthy individuals, areas of microdamage are targeted for repair during the bone turnover process. But if bone remodeling is suppressed via bisphosphonate treatment a common therapy for osteoporosis the microdamage burden increases and bone quality and mechanics can be negatively affected [69]. It is unclear whether certain areas of bone are more susceptible to microdamage than others due to compositional differences, such as increased tissue mineralization that makes them more brittle. However to date, little is known about the microscopic tissue composition where microdamage occurs and whether compositional differences play a role in microdamage location. Synchrotron infrared microscopy has been an essential tool to examine whether areas of microdamaged bone are chemically different than undamaged areas, because the technique is sensitive to the protein and mineral content of the tissue [70,71]. Additionally the spatial resolution of the technique allows the examination of the narrow areas of microdamage, which are typically 15 m wide [30]. In this study bone samples (L3 vertebrae) harvested from 15 dogs were examined [72]. Samples were prepared for FTIRM examination by embedding the bone samples in poly-methylmethacrylate, and subsequently cutting them into 5-m-thick sections. Regions containing microdamage were identified by fuchsin staining. Synchrotron infrared microspectroscopy was used to determine the local bone composition within areas of microdamage and the surrounding undamaged tissue. Microdamaged tissue, indicated by positive (i.e. purple) fuchsin staining is visible in Fig. 15. FTIRM chemical images demonstrate that microdamaged areas of bone are chemically different from the surrounding undamaged tissue. Specifically, these results
418
Figure 15. (Top) Light microscope image of a microcrack indicating the area imaged with the IR microscope. The fuchsin stain (magenta) highlights the microcrack. Data were collected within the crack and from the surrounding area. Panels AF represent the corresponding IR images of the same area. The individual images illustrate the (A) carbonate/phosphate ratio, (B) carbonate/protein ratio, (C) acid phosphate/total phosphate ratio, (D) phosphate/protein ratio, (E) crystallinity, and (F) collagen cross-linking. Scale bar is 25 m. From [73].
shows that the mineral stoichiometry is altered in microdamaged bone, where the carbonate/protein ratio and carbonate/phosphate ratio were significantly lower in areas of microdamage, and the acid phosphate content was higher. No differences were observed in tissue mineralization (phosphate/protein ratio) or crystallinity between the microdamaged and undamaged bone, indicating that the microdamaged regions of bone were not over-mineralized. The collagen cross-linking structure was also significantly different in microdamaged areas of bone, consistent with ruptured cross-links and reduced fracture resistance. All differences in composition had well-defined boundaries in the microcrack region, strongly suggesting that they occurred after microcrack formation and implying that bisphosphonate treatment does not alter bone composition, but the existence of microdamage does. Thus, since long-term bisphosphonate treatment can lead to the build-up of microdamage, the accompanying chemical changes to bone tissue could eventually lead to increased bone fragility. 5.4. Prion-Infected Nervous Tissue Transmissible spongiform encephalopathies (TSEs), also termed prion diseases, are a group of fatal, neurodegenerative disorders that occur sporadically but may also have genetic and infectious origins. The hallmark of this type of disease is the misfolding and subsequent accumulation of the normal, -helical rich prion protein (cellular prion protein, or PrPC), into its pathological isoform, PrPSc (also called PrPTSE), which has an abnormally high content of -sheet. Scrapie, first described in the 18th century in sheep and goats, is the oldest known prion disease, and has since been established as a model in rodents to study the pathogenesis and pathology of TSE. Since a multitude of molecular parameters change in the tissue over the course of the disease, FTIR microspectroscopy has been proposed as a valuable method to study and identify prionaffected tissues due to its ability to detect a variety of changes in molecular structure and composition simultaneously. We refer the readers to an exhaustive recent review on this subject [74].
419
Figure 16. Absorbance spectra and amide I frequency maps from scrapie-infected and control ganglia. (A) Absorbance spectra (vector-normalized) representing different regions in a neuron from an infected ganglion (solid lines, red and blue) and from a control ganglion (red dashed line). A shift of the amide I band can be observed in some of the spectra from the scrapie neuron (blue spectra). (B and C) Frequency maps from scrapie-infected (B) and control (C) ganglia. The maps were constructed by plotting the characteristic frequency of the amide I band as a function of pixel location. The black pixels indicate spectra of insufficient quality, which were excluded from the analysis. Each map comprises a single neuron The uppermost right map in (B) originates from a measurement that had to be terminated early for technical reasons. (D) Second derivatives of spectra from different regions in a nerve cell from an infected ganglion. The colors of the spectra correspond to the regions shown in the intensity ratio map on the left. A photomicrograph of the unstained section with an inset depicting the area of measurement is shown below. From [81].
Among several studies related to prion diseases, it was found that not all neurons are affected by PrPSc, even at the terminal stage of the disease. Therefore, the spatial and temporal distribution of PrPSc in the nervous system is of significance to understanding the pathological mechanism of scrapie and subsequently other TSEs. Indeed, conventional infrared microspectroscopy has been very useful in determining the molecular composition of certain areas of the central nervous system, revealing diseaseassociated spectral differences compared to uninfected control samples [7578]. However, although TSEs are characterized by the accumulation of a -sheet rich protein, no spectral differences in the amide I band could be detected. Since PrPSc aggregates in scrapie are microdisperse, conventional infrared microspectroscopy does not provide sufficient spatial resolution to detect those focalized regions of high -sheet with acceptable signal to noise ratios. Recently, synchrotron-assisted infrared microspectroscopy was used to reveal the protein structural changes in scrapie-infected hamsters [79, Kretlow, submitted to BBA] [80]. Specifically, dorsal root ganglia (DRG) of the spinal cord contain neurons that are ~3050 m in diameter, so that the distribution of the misfolded prion protein can be examined on a subcellular level. Typical IR absorbance spectra of DRG from different regions of both scrapie-infected and control neurons are shown in Fig. 16. Here, a shift of the amide I band to lower frequencies can be seen in some of the scrapie-infected ganglia, suggesting an increased content of -sheet in the studied area. Spectra from infected ganglia displaying the band shift are shown in blue; those showing no different peak position as compared to spectra from healthy ganglia are shown in red (solid lines in Fig. 16A). Importantly, the shift was not observed in any of the spectra from the control ganglia (see dashed red lines in Fig. 16A).
420
At each pixel in the image, the band position of the amide I absorption maximum was determined and a frequency map of a neuron was generated. Figures 16B and 16C display such maps for scrapie-infected and control ganglia, respectively. While the amide I frequency varied between 1658 and 1654 cm1 in normal ganglia (Fig. 16C), approximately 15% of the spectra obtained from scrapie-infected hamsters displayed a downshift of the amide I band of approximately 10 cm1, resulting in an absorption maximum at around 1645 cm1 (Fig. 16A). To estimate the contributions of the different secondary structure components, the second derivatives were calculated (Fig. 16D). Figure 16A shows that the shift of the amide I band maximum to lower frequency is caused by two factors, namely an increased intensity of the -sheet band at 1637 cm1, together with the appearance of a new component at 1631 cm1 in some spectra, and a frequency shift of the -helix band at 1657 to 1650 cm 1 (blue spectra in Fig. 16D). The ratio of the bands representing -sheet and -helical structures (1637 cm1/1657 cm1) was calculated for all spectra (Fig. 16D) and the resulting images correlated well with the amide I frequency maps (Fig. 16B,C), demonstrating that many, but not all, scrapie-infected neurons contain regions of elevated -sheet structure. Subsequent immunostaining of the investigated samples with the prion specific antibody, 3F4, revealed that the detected increase in -sheet is at least partly due to the accumulation of disease associated prion protein. This example illustrates on how the application of a brilliant IR synchrotron light source enables the in situ investigation of localized changes in protein structure and composition in nervous cells due to PrPSc deposition, and a demonstration on how the IR spectral information can be correlated with results of complementary studies such as immunocytochemistry. Synchrotron FTIR microspectroscopy is readily capable of detecting the misfolded prion protein in situ without the necessity of immunostaining or purification procedures. 5.5. Biomolecular Investigation of Human Substantia Nigra (SN) in Parkinsons Disease In a recent study, SR-FTIR microscopy were used to distinguish chemical composition and morphology of the human substantia nigra (SN) of brain between normal and Parkinsons diseased tissues [82]. Parkinsons disease (PD) is a chronic, progressive neurodegenerative movement disorder. It results from the degeneration and loss of dopamine-producing nerve cells in the brain, mainly in the SN [83]. It is well accepted that PD is a multietiological disease. Most researchers believe that oxidative stress and mitochondrial dysfunction play a crucial role in degeneration and death of nerve cells in the SN. Moreover, dyshomeostasis of Fe ions, protein aggregation, increased peroxidation of lipids, changes in the activity of enzymes of the antioxidant system and some other factors are considered at investigations of PD pathogenesis. The aforementioned processes should induce changes in main biological molecules. Therefore, this region was investigated with a high spatial resolution, thanks to the synchrotron source, in order to detect any variation in the biochemical composition of the human substantia nigra of brain between normal and Parkinsons diseased tissues. The studies were carried out for thin tissue sections, focusing more particularly on nerve cell bodies that are affected in Parkinsons disease (PD). The major spectral differences between normal (control) and PD tissues appear at the following vibrational frequencies: 2930, 2850, 1655, 1380, 1236, 1173 and 1086 cm1, as well as in the Amide I region (Fig. 17).
421
Figure 17. The comparison of FTIR absorption spectra acquired in the neurons of substantia nigra for the control and Parkinsons diseased (PD) cases. Spectra were collected with a 6x6 micron square aperture.
Chemical Image Proteins
Chemical Image DNA/RNA
Neurons Control
Neurons PD infected
Figure 18. Chemicals images of protein and DNA/RNA obtained with a 6 6 micron square aperture, using a synchrotron infrared microscope. Spectra were been collected Parkinsons diseased (PD) and control SN neurons [82].
The infrared imaging of the protein and nucleic acids functional groups indicates a higher concentration of these two components inside the neuron cell body, while this is markedly not the case for PD samples (Fig. 18). These results strengthen the hypothesis that PD is a multietiological disorder. Moreover, the reported results clearly indicate that, in addition to a distinct visual observation, the diseased nerve cells exhibits change of their biochemical composition. It suggests that disturbances of normal functioning of SN neurons appear before their morphological atrophy. 5.6. SR-FTIR Study of -Amyloid Deposits in Alzheimers Disease Alzheimers disease (AD) is characterized by the misfolding and plaque-like accumulation of a naturally occurring peptide in the brain called amyloid beta (A). This process has been associated with the binding of metal ions such as iron (Fe), copper (Cu), and zinc (Zn) [84]. It is thought that metal dyshomeostasis is involved in protein misfolding and may lead to oxidative stress and neuronal damage.
422
1800
1700
1600
1500
1400
Wavenumbers (cm-1)
(A)
(B)
Absorbance
(a) (b)
4000
3500
3000
2500
2000
1500
1000
500
Wavenumbers (cm-1)
(D)
10 m
(E)
(C)
Figure 19. (A) Light microscope image of Alzheimers diseased brain tissue. (B) Infrared spectra recorded inside and outside the plaque, showing the shoulder at 1630 cm1, assigned to -amyloid. (C) Infrared spectra from inside the plaque. One of them displays additional absorption features, which are assigned to creatine. (D) Infrared image of amyloid plaque distribution, calculated as a peak height ratio of 1630 / 1655 cm1. (E) Chemical image of creatine distribution inside the plaque.
However, the exact role of the misfolded proteins and metal ions in the degenerative process of AD is not yet clear. Synchrotron infrared micro-spectroscopy has been used to image the in situ secondary structure of amyloid plaques in brain tissue of AD patients with high spatial resolution [8587]. Figure 19B illustrates the FTIR spectra obtained inside a plaque (Fig. 19A). Some brain samples have been shown to exhibit additional absorption features in the FTIRM spectra, assigned to creatine (Fig. 19C) [87]. Chemical images of A and creatine distribution from a hippocampal section of Alzheimer diseased human brain, obtained with an aperture of 3 3 m2 and 1 m step size, are displayed in Fig. 19D and E, respectively. Synchrotron X-ray fluorescence (SXRF) microprobe is a complementary technique used to probe trace elements with sensitivities in the sub-mg kg1 range and a spatial resolution similar to FTIRM (210 m) [88]. Because of the low power deposition that x-rays provide and the ability to conduct the measurements in air, these analyses can be done non-destructively on a much wider array of sample types, especially relatively fragile biological samples. For example, the alterations in trace metals such as Fe, Cu, and Zn have been observed in neurological diseases such as Parkinsons disease, amylotrophic lateral sclerosis, Alzheimers disease, and prion diseases [86,89,90]. It would be very beneficial if both FTIRM and SXRF information from a single sample can be combined. However, this requires precise overlap of the IR and X-ray images, which has recently been achieved [91]. FTIRM and SXRF microprobe have been used to correlate protein structure and metal ion accumulation, respectively, in amyloid plaques from brain tissue of AD patients [86,92] and in scrapie-infected nervous tissue [86,92]. SXRF microprobe was used to image trace elements, specifically Ca, Fe, Cu, and Zn, in identical areas of the same tissue sample as imaged with synchrotron FTIRM. In AD, SXRF microprobe was used to reveal hot spots of accumulated metal
423
Figure 20. Infrared, X-ray fluorescence, and UV-epifluorescence images of Alzheimers tissue. Singlechannel color images of (A) Zn (SXRF, red), (B) Thioflavin staining (epifluorescence, green), and (C) -sheet protein (infrared, blue). (DF) Dual-channel correlation images and (G) the RGB correlation image. Scale bar is 100 m.
ions, specifically Cu and Zn, with a strong spatial correlation between these two ions. The hot spots of accumulated Zn and Cu were also co-localized with -amyloid plaques (Fig. 20). 6. Future Directions of Synchrotron Infrared Spectroscopy and Microscopy Many synchrotrons worldwide have active programs involving biomedical applications of synchrotron IR microscopy. For scientists that are not familiar with the operation of these centers, access to these facilities may seem restrictive. However, almost all synchrotron IR beamlines worldwide offer beam time free of charge based on a peerreviewed proposal system. In many cases, researchers have active programs using a conventional infrared microscope, equipped either with a single detector or an array detector. The synchrotron source provides a complementary capability for high spectral quality at higher spatial resolution. Since biochemical changes can be very subtle, statistical analyses of spectra or images greatly benefit from high spectral quality. One of the pitfalls of FTIRM remains the diffraction limited spot size and slow data collection rates. Nowadays, focal plane array detectors are being implemented in microscopes that use a globar source, and the performance of these detectors has not yet been exploited with a synchrotron source. Clearly, the size of the detector array has to be adapted to match the projected size of the synchrotron beam, in order to keep the brightness advantage of this source. These small detector arrays might well be available soon, and will allow faster acquisition of data, and a slightly improved lateral resolution using point-spread function (PSF) deconvolution. Readers interested in the details of future improvements with such a combination of bright infrared sources with up-todate bidimensional detectors can find them in [93]. This is a direction that can be complemented by the collection of many more infrared photons from a bending magnet with much large horizontal opening angles.
424
Several efforts are devoted today to near field FTIR microscopy (with lateral resolution between /10 to /100) using the synchrotron source. Results are still preliminary, but it is clear that such an approach is promising for the near future. Another direction for the future, which is starting to become more popular at synchrotron facilities, is the opportunity to combine IR microscopy with other synchrotron based techniques. In the preceding section, we illustrated the complementary use of Xray fluorescence microscopy with IR microscopy. The strategy for efficient combination relies upon the choice of adequate sample substrates, good positioning registration between instruments, and careful coordination between beamtime allocations at different endstations. Other combinatory approaches are underway using synchrotron-based (auto-) fluorescence microscopy and infrared microscopy at SOLEIL synchrotron and efforts to combined scanning transmission x-ray microscopy (STXM), x-ray microdiffraction, and x-ray tomography. Raman microscopy is also an excellent approach to vibrational spectroscopy, with an equivalent, or often better, spatial resolution than synchrotron infrared microscopy. The data are complementary, but often Raman does not provide adequate spectral quality (i.e. signal to noise), data can be complicated by intrinsic fluorescence emission, and samples can suffer from laser damage. Thus, it is valuable to combine the two micro-spectroscopic approaches, and this should be made available for users at synchrotron facilities. Another upcoming area in biomedical research is the use of the far-IR, or terahertz (THz) energy range. Very intense THz beams can be generated with so-called coherent emission. Coherent synchrotron radiation is produced when the relativistic electron bunches have longitudinal density variations on a scale comparable to or smaller than the wavelength. Synchrotron and transition radiation emitted from a bunched electron beam becomes coherent and highly intense at wavelengths about or longer than the electron bunch length. The radiation has a continuous spectrum in the submillimeter to millimeter wavelength range. One of the main factors determining the intensity of the coherent radiation is the electron bunch shape. The intensity of the coherent radiation emitted from an electron bunch is approximately given [94] by
p( ).N 2 . f ( )
where p( ) is the intensity of radiation emitted from an electron, N the number of electrons in the bunch, the wavelength of radiation, f ( ) the longitudinal bunch form factor given by
S ( z) exp(i 2 z / )dz
where S(z) is the normalized distribution of electrons in the bunch on the beam axis. The bunch form factor has a value below 1 and is nearly equal to 1 at sufficiently long wavelengths compared with the bunch length. In this case, the intensity of the multiparticle coherent radiation is N times that of the incoherent radiation [9597]. Over the years, the efforts devoted by accelerator physicists to the stability of the electron beam in storage rings, as well as for the isolation of vibrational noise in beamline design, have resulted in a very stable synchrotron IR source. But very recently, significant progress has been made towards stable operation of storage rings with a small momentum compaction factor, which results in a short (sub-millimeter) longitudinal size of the electron beam, and as a consequence, a high flux of coherent synchrotron emission in the far-IR spectral range [98]. It is also possible that ultra short pulses
425
can also be generated with the assistance of laser slicing techniques [99]. One can expect that if this progress continues, generation of coherent SR may be possible also in the mid-IR range. If this happens, the synchrotron sources will be much more advantageous over black-body sources not only in terms of brightness, but also in terms of flux. Other exciting new directions in synchrotron infrared spectroscopy are emerging. The pulsed nature of the synchrotron source has been exploited in few cases [100], but synchronization with external lasers will certainly see more applications in the future. Free electron lasers are also being developed in many countries. Production of intense infrared pulses, with very short duration, is currently possible. There are new projects combining free electron lasers and continuous emission (synchrotron) from a storage ring. Therefore, synchrotron infrared spectroscopy and microscopy are very attractive for many scientific disciplines, in a wavelength energy domain extending from the very far- to the near-infrared. The activity in these and related areas are expected to continue to grow, thanks to the unique properties of synchrotron IR light and a continuously high demand from the user community.
Acknowledgments The authors are grateful to O. Chubar, F. Jamme, M. Refregiers and F. Polack (SOLEIL), Larry Carr, Randy Smith, Ariane Kretlow, Meghan Ruppel, Andreana Leskovjan (NSLS), Gwyn Williams (Jefferson Lab), John Chalmers (VS Consulting), Sheila Fisher (University of Leeds) and Sanjika Dias-Gunasekara (University of Durham) for their very active collaboration and interesting discussions.
References
[1] J.M. Chalmers and P.R. Griffiths, Applications of Vibrational Spectroscopy in Life, Pharmaceutical and Natural Sciences, Vol. 5, John Wiley & Sons, Ltd, New York, 2001. [2] M. Jackson and H.H. Mantsch. in (Wiley-Liss, ed.) Infrared Spectroscopy of Biomolecules, New York 1996, pp. 311. [3] M. Jackson and H.H. Mantsch. Encyclopedia of Analytical Chemistry, Wiley & Sons, Chichester, Sussex, UK 2000. [4] J.M. Chalmers and P.R. Griffiths, Theory and Instrumentation, Vol. 1, John Wiley & Sons, Ltd 2001. [5] D.L. Wetzel and S.M. LeVine, Science 285 (1999) 1224-1225. [6] L.M. Miller and P. Dumas, Biochim. Biophys. Acta 1758 (2006) 846-857. [7] M. Diem, S. Boydston-White and L. Chiriboga, Appl. Spectrosc. 53 (1999) 148A-161A. [8] G.J. Puppels, T.C.B. Schut, P.J. Caspers, R. Wolthuis, M. Van Aken, A. Van der Laarse, H.A. Bruining, P.J. Buschman, M.G. Shim and W. B.C., Marcel Dekker, New York, 2002. [9] W. Duncan and G.P. Williams, Applied Optics 22 (1983) 2914. [10] P. Dumas, G.D. Sockalingum and J. Sul-Suso, Trends in Biotechnology 25 (2007 ) 40-44. [11] W. Duncan and G.P. Williams, Applied Optics 22 (1983) 2914-2922. [12] D. Einfeld and J. Nagel, BESSY Report N TB34 (1981). [13] P. Lagarde, Infrared Physics 18 (1979) 395-460. [14] J.R. Stevenson, H. Ellis and R. Bartlett, Appl. Optics 12 (1973) 2884-2889. [15] J.R. Stevenson and J.J. Lathcart, Nucl. Instr. Meth. Phys. Res. B 172 (1980) 367. [16] E. Schweizer, J. Nagel, W. Braun, E. Lippert and A.M. Bradshaw, Nucl. Instr. Meth. Phys. Res. A 239 (1985) 630-634. [17] T. Nanba, Int J Millimeter Waves 7 (1986) 759-770. [18] G.P. Williams, Nuclear Instruments and Methods in Physics Research A291 (1990) 8-12. [19] G.P. Williams, C. J. Hirschmugl, E.M. Kneedler, E.A. Sullivan, D.P. Siddons, Y.J. Chabal, F. Hoffmann and K.D. Moeller, Review of Scientific Instruments 60 (1989) 2176.
426
[20] G.P. Williams, P.Z. Takacs, R.W. Klaffky and M. Schleifer, Nucl. Instr. Meth. Phys. Res. A 246 (1986) 165. [21] P. Roy, Y.-L. Mathis, S. Lupi, A. Nucara, B. Tremblay and A. Gerschel, Synchrotron Radiation News 8 (1995) 10. [22] L. Forro, G.L. Carr, G.P. Williams, D. Mandrus and L. Mihaly, Phys. Rev. Lett. 65 (1990) 1941. [23] C.J. Hirschmugl, G.P. Williams, F.M. Hoffmann and Y.J. Chabal, Phys. Rev. Lett. 65 (1990) 408. [24] G.L. Carr, M. Hanfland and G.P. Williams, Rev. Sci. Instr. 66 (1995) 1643-1645. [25] G.L. Carr, J. Reffner and G.P. Williams, Rev. Sci. Instr. 66 (1995) 1490. [26] G. Ellis, M.A. Gmez and C. Marco, J. of Macromolecular Science, part B-Physics 43 (2004) 191-206. [27] N. Guilhaumou, P. Dumas, G.L. Carr and G.P. Williams, Applied Spectroscopy 52 (1998) 1029-1034. [28] P. Dumas and L. Miller, Journal of Biological Physics 29 (2003) 201-218. [29] N. Jamin, P. Dumas, J. Moncuit, W.H. Fridman, J.L. Teillaud, G.L. Carr and G.P. Williams, Proc. Natl. Acad. Sci. 95 (1998) 4837-4840. [30] L.M. Miller, C.S. Carlson, G.L. Carr and M.R. Chance, Cell. Mol. Biol. 44 (1998) 117. [31] L.M. Miller, G.L. Carr, M. Jackson, P. Dumas and G.P. Williams, Synchr. Rad. News 13 (2000) 31-37. [32] L.M. Miller, P. Dumas, N. Jamin, J.L. Teillaud, J. Miklossy and L. Forro, Review of Scientific Instruments 73 (2002) 1357-1360. [33] R.A. Bosch, Nucl. Instrum. Meth. in Phys. Research. A 386 (1997) 525. [34] Y.L. Matthis, P. Roy, B. Tremblay, A. Nucara, S. Lupi, P. Calvani and A. Gerschel, Phys. Rev. Lett. 80 (1998) 1220. [35] H.Y.N. Holman, R. Goth-Goldstein, M.C. Martin, M.L. Russel and W.R. McKinney, Environ. Sci. Technol. 34 (2000) 2513-2517. [36] G.L. Carr, Review of Scientific Instruments 72 (2001) 1613-1619. [37] P. Lerch, (Private communication). [38] R.G. Messerschmidt. in (Humecki, H., ed.) Practical Guide to Infrared Microspectroscopy, Marcel Dekker, Inc., New York 1995, pp. 3-40. [39] G.L. Carr, O. Chubar and P. Dumas. in (Levin, R.B.I., ed.) Spectrochemical Analysis Using Multichannel Infrared Detectors, Blackwell Publishing 2005. [40] D.M. Byler and H. Susi, Biopolymers 25 (1986) 469-87. [41] H. Susi and D.M. Byler, Biochem Biophys Res Commun 115 (1983) 391-7. [42] H. Susi, D.M. Byler and J.M. Purcell, J. Biochem. Biophys. Methods 11 (1985) 235-40. [43] M.J. Walsh, M.J. German, M. Singh, H.M. Pollock, A. Hammiche, M. Kyrgiou, H.F. Stringfellow, E. Paraskevaidis, P.L. Martin-Hirsch and F.L. Martin, Cancer Letters 246 (2007) 1-11. [44] S. Chio-Srichana, M. Refregiers, F. Jamme, S. Kascakovac, S. Deshayes, V. Rouam and P. Dumas, Biochim. Biophys. Acta Accepted Feb. 2008 (2008). [45] H.Y.N. Holman, M.C. Martin, E.A. Blakeley, K. Bjornstad and W.R. McKinney, Biopolymers 57 (2000) 329-335. [46] J. Sul-Suso, D. Skingsley, G.D. Sockalinghum, A. Kohler, G. Kegelaer, M. Manfait and A. El Haj, Vibrational Spectrosocopy 38 (2005) 179-184. [47] M.J. Tobin, M.A. Chesters, J.M. Chalmers, F.J.M. Rutten, S.E. Fisher, I.M. Symonds, A. Hitchcock, R. Allibone and S. Dias-Gunasekara, Faraday Discuss. 126 (2004) 27-39. [48] M. Diem, P. Lasch, L. Chiriboga and A. Pacifico, Biopolymers: Biospectrosocopy 67 (2002) 349-353. [49] P. Heraud, B.R. Wood, M.J. Tobin, J. Beardall and D. McNaughton, FEMS Microbiol. Lett. 249 (2005) 219-225. [50] E. Gazi, J. Dwyer, N.P. Lockyer, J. Miyan, P. Gardner, C.A. Hart, M.D. Brown and N.W. Clarke, Vibrational Spectroscopy 38 (2005) 193-201. [51] N. Jamin, P. Dumas, J. Moncuit, W.H. Fridman, J.L. Teillaud, G.L. Carr and G.P. Williams, Cell Mol Biol (Noisy-le-grand) 44 (1998) 9-13. [52] N. Jamin, L.M. Miller, F. Montcuit, P. Dumas and J.L. Teillaud, Biopolymers 72 (2003) 366-73. [53] C.P. Schultz, K.Z. Liua, J.B. Johnston and H.H. Mantsch, Journal of Molecular Structure 408-409 (1997) 253-256. [54] A. Salman, J. Ramesh, V. Erukhimovitch, M. Talyshinsky, S. Mordechai and M. Huleihel, Journal of Biochemical and Biophysical Methods 55 (2003) 141-153. [55] S. Boydston-White, T. Gopen, S. Houser, J. Bargonetti and M. Diem, Biospectroscopy 5 (1999) 219- 227. [56] M. Diem, M. Romeo, S. Boydston-White, M. Miljovic and C. Matthus, Analyst 129 (2004) 880-885. [57] J.M. Sanchez-Ruiz and M. Martinez-Carrion, Biochemistry 27 (1988) 3338-3342. [58] M. Diem, L. Chiriboga, P. Lasch, A. Pacifico and C.U.o.N.Y.H.C.P.A.N.Y.N.Y.U.S.A.m.a.c. Department of Chemistry, Biopolymers. 67(4-5) (2002).
427
[59] W.F. Neuman and M.W. Neuman, The Chemical Dynamics of Bone Mineral, The University of Chicago Press, Chicago, 1958. [60] H. Ouyang, P.J. Sherman, E.P. Paschalis, A.L. Boskey and R. Mendelsohn, Appl Spectrosc 58 (2004) 1-9. [61] O. Akkus, A. Polyakova-Akkus, F. Adar and M.B. Schaffler, J Bone Miner Res 18 (2003) 1012-9. [62] R. Legros, N. Balmain and G. Bonel, Calcif Tissue Int 41 (1987) 137-44. [63] C. Rey, H.M. Kim, L. Gerstenfeld and M.J. Glimcher, J Bone Miner Res 10 (1995) 1577-88. [64] R.K. Nalla, J.J. Kruzic, J.H. Kinney and R.O. Ritchie, Bone 35 (2004) 1240-6. [65] N.P. Camacho, W.J. Landis and A.L. Boskey, Connect Tissue Res 35 (1996) 259-65. [66] S.J. Gadeleta, A.L. Boskey, E. Paschalis, C. Carlson, F. Menschik, T. Baldini, M. Peterson and C.M. Rimnac, Bone 27 (2000) 541-50. [67] R.Y. Huang, L.M. Miller, C.S. Carlson and M.R. Chance, Bone 33 (2003) 514-21. [68] L.M. Miller, J.T. Novatt, D. Hamerman and C.S. Carlson, Bone 35 (2004) 498-506. [69] T. Mashiba, T. Hirano, C.H. Turner, M.R. Forwood, C.C. Johnston and D.B. Burr, J Bone Miner Res 15 (2000) 613-20. [70] C. Rey, V. Renugopalakrishnan, B. Collins and M.J. Glimcher, Calcif Tissue Int 49 (1991) 251-8. [71] C. Rey, M. Shimizu, B. Collins and M.J. Glimcher, Calc. Tissue Int. 49 (1991) 383-388. [72] M.E. Ruppel, D.B. Burr and L.M. Miller, Bone 39 (2006) 318-324. [73] M.E. Ruppel, D.B. Burr and L.M. Miller, Bone 39 (2006) 318-324. [74] A. Kretlow, Q. Wang, J. Kneipp, P. Lasch, M. Beekes, L. Miller and D. Naumann, Biochimica et Biophysica Acta 1758 (2006) 948-959. [75] J. Kneipp, P. Lasch, E. Baldauf, M. Beekes and D. Naumann, Biochim Biophys Acta 1501 (2000) 189-99. [76] D.S. Lester, L.H. Kidder, I.W. Levin and E.N. Lewis, Cell Mol Biol (Noisy-le-grand) 44 (1998) 29-38. [77] J. Kneipp, M. Beekes, P. Lasch and D. Naumann, J. Neurosci. 22 (2002) 2989-2997. [78] M.J. Walsh, M.J. German, M. Singh, H.M. Pollock, A. Hammiche, M. Kyrgiou, H.F. Stringfellow, E. Paraskevaidis, P.L. Martin-Hirsch and F.L. Martin, Cancer Letters xx (2007) 1-11. [79] J. Kneipp, L.M. Miller, M. Joncic, M. Kittel, P. Lasch, M. Beekes and D. Naumann, Biochim. Biophys. Acta 1639 (2003) 152-158. [80] A. Kretlow and e. al., (submitted to BBA) (2008). [81] J. Kneipp, L.M. Miller, S. Spassov, F. Sokolowski, P. Lasch, M. Beekes and D. Naumann, Biopolymers 74 (2004) 163-167. [82] M. Szczerbowska-Boruchowska, P. Dumas, M.Z. Kastyak, J. Chwiej, M. Lankosz, D. Adamek and A. Krygowska-Wajs, Archives of Biochemistry and Biophysics 459 (2007) 241248. [83] E.E. Kolesnikova and T.V. Serebrovskaya, Neurophysiology 35 (2003) 56-68. [84] M.A. Lovell, J.D. Robertson, W.J. Teesdale, J.L. Campbell and W.R. Markesbery, J Neurol Sci 158 (1998) 47-52. [85] L.P. Choo, D.L. Wetzel, W.C. Halliday, M. Jackson, S.M. LeVine and H.H. Mantsch, Biophys J 71 (1996) 1672-9. [86] L.M. Miller, Q. Wang, T.P. Telivala, R.J. Smith, A. Lanzirotti and J. Miklossy, J Struct Biol 155, (2006) 30-37. [87] M. Rak, M.R. Del Bigio, S. Mai, D. Westaway and K.M. Gough, Biopolymers 4 (2007) 201-217. [88] S.R. Sutton, P.M. Bertsch, M. Newville, M. Rivers, A. Lanzirotti and P. Eng, Reviews in Mineralogy & Geochemistry: Applications of Synchrotron Radiation in Low-Temperature & Environmental Science Vol 49, (2002) 429-483. [89] B. Tomik, J. Chwiej, M. Szczerbowska-Boruchowska, M. Lankosz, S. Wojcik, D. Adamek, G. Falkenberg, S. Bohic, A. Simionovici, Z. Stegowski and A. Szczudlik, Neurochem Res 31 (2006) 321-31. [90] J.F. Collingwood, A. Mikhaylova, M. Davidson, C. Batich, W.J. Streit, J. Terry and J. Dobson, J Alzheimers Dis 7 (2005) 267-72. [91] L.M. Miller, Q. Wang, R.J. Smith, H. Zhong, D. Elliott and J. Warren, Analytical & Bioanalytical Chemistry 387 (2007.) 1705-15. [92] Q. Wang, A. Kretlow, M. Beekes, D. Naumann and L.M. Miller, Vibrational Spectroscopy 38 (2005) 61-69. [93] G.L. Carr, O. Chubar and P. Dumas. in (R. Bahrgava, I.W.L., ed.) Spectrochemical Analysis Using Infrared Detectors, Blackwell Publishing 2006. [94] S. Okudaa, M. Takanakab, M. Nakamurab, R. Katob, T. Takahashic, S.-K. Namd, R. Taniguchia and T. Kojimaa, Radiation Physics and Chemistry 75 (2006) 903-907. [95] G.P. Williams, Rep. Prog. Phys. 68 (2005) 1-26.
428
[96] G.R. Neil, G.L. Carr, J.F. Gubeli, K. Jordan, M.C. Martin, W.R. McKinney, M. Shinn, M. Tani, G.P. Williams and X.C. Zhang, Nuclear Instruments and Methods in Physics Research A 507 (2003) 537-540. [97] G.R. Neil and G.P. Williams, Infrared Physics & Technology 45 (2004) 389-391. [98] M. Abo-Bakr, K. Feikes, P. Holldack, P. Mushke, W.B. Peatman, U. Schade, H.W. Hubers and G. Wustefled, Phys. Rev. Lett. 90 (2003) 094801. [99] R.W. Schoenlein, S. Chattopadhyay, H.H.W. Chong, T.E. Glover, P.A. Heimann, C.V. Shank, A.A. Zholents and M.S. Zolotorev, Science 287 (2000) 2237. [100] G.L. Carr, R. Lobo, J. LaVeigne and D. Reitze, Phys. Rev. Lett.. 85 (2000) 3001-3004.
Biological and Biomedical Infrared Spectroscopy A. Barth and P.I. Haris (Eds.) IOS Press, 2009 2009 The authors and IOS Press. All rights reserved.
429
Author Index
Arakawa, T. Barth, A. Bou, P. Chio-Srichan, S. Cho, M. Choi, J.-H. Dumas, P. Fabian, H. Fayer, M.D. Finkelstein, I.J. Goormaghtigh, E. Haris, P.I. Hering, J.A. Ishikawa, H. Joly, D. 261 vii, 1, 53 178 403 224 224 403 312 79 79 104 v, vii, 1, 129 129 79 288 Jorgensen, L. Keiderling, T.A. Kim, S. Kubelka, J. Lasch, P. Li, T. Macnab, A. Miller, L.M. Nsoukpo-Kossi, C.N. Naumann, D. Tajmir-Riahi, H.A. Tobin, M.J. van de Weert, M. Wolkers, W.F. 168 178 79 178 312 261 355 403 288 312 288 403 168 272
This page intentionally left blank

1607500450

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1607500450

Uploaded by

Copyright:

Available Formats

BIOLOGICAL AND BIOMEDICAL INFRARED SPECTROSCOPY

Advances in Biomedical Spectroscopy

Biological and Biomedical Infrared Spectroscopy

Amsterdam Berlin Tokyo Washington, DC

Sirinart CHIO-SRICHAN Jun-Ho CHO

Parvez I. HARIS Joachim A. HERING

Haruto ISHIKAWA David JOLY

53 79 104 129 168

Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

A. Barth and P. Haris / Infrared Spectroscopy Past and Present

The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy

A. Barth / The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy

A. Barth / The Study of Protein Reactions by Reaction-Induced Infrared Difference Spectroscopy