You are on page 1of 10

Curvature-induced microemulsion

Israel Barragan
February 11, 2013
Abstract
1 Model
In this work we consider a coarse-grained, solvent-free model to describe the
dynamics of lipid chains that shape the membrane. This chapter introduces
physical and computational aspects of the model.
1.1 Physical description
Our coarse-grained discretization for the amphiphiles consists of a chain with N
A
beads for the tail (hydrophobic) and N
B
beads for the head group (hydrophilic),
under the constrain N
A
+ N
B
= N = const. The physical length of this chains
can be characterized by the mean-root-square end-to-end distance (R
eo
), which
represents the distance between terminal atoms of the polymer chain and that
can be used to describe its conguration [1].
We assume that beads in a single chain are joint by spring-like forces dened
through rest length l
0
and spring constant k
s
. In order to control the stiness
and congurational uctuations of the chain we also include a bond-angle bend-
ing force with constant k
b
. So the bonded Hamiltonian for a single chain can
be written as
H
b
k

T
=
k
s
2
N1

i=1
(r
i,i+1
l
0
)
2
+ k
b
N1

i=2
(1 cos
i
). (1)
where r
i,i+1
= |r
i+1
r
i
|,
i
is the angle between vectors r
i1,i
and r
i,i+1
. This
equation has been written in a convenient way, setting k

T as the unit of energy,


and in what follows well keep this assumption, provided that all our simulations
take place at constant and nite temperature.
Up to now, we have described only the interaction of beads in a single chain.
However, the inter-chain interactions and those between chains and the solvent,
also must be described. A natural way to achieve this would be an explicit
inclusion of solvent molecules and setup of their corresponding interactions.
However, in order to access the length and time scales relevant for biological
membranes, it is necessary to provide a reduced description in which those
interactions are treated in an eective way [2]. Following this approach, non-
bonded interactions will be incorporated through a third order expansion for
1
the free energy in terms of the local densities of beads of each type, so that the
non-bonded Hamiltonian can be expressed as [3]:
H
nb
(
A
,
B
)
k

T
=
_
d
3
r
R
3
eo

(r)
_
v

(r) +
w

(r)

(r)
_
, (2)
where has been used Einsteins convention over repeated Greek indexes (,
(A, B)). The virial coecients v

and w

set the interaction between dif-


ferent beads and well see that some of them can be approximated form the
systems equation of state (EOS). The local densities appearing in this expres-
sion are dened by:

(r) =
R
3
eo
N
nN

i=1
(r
i
r)
,t(i)
, (3)
where t(i) (A, B) is the type of bead r
i
.
In order to estimate the virial coecients, let us consider a collection of
single-type, oil-soluble chains immersed in water. In that case, chains will col-
lapse in dense and almost incompressible regions, in order to reduce the un-
favorable contact with water. At thermal equilibrium, the normal pressure at
both sides of the oil-water interface has to be the same. In a solvent-free model,
this corresponds to a zero pressure coexistence of the liquid with its low density
vapor (
coex
). Under this assumption, the third-order virial equation of state
for a single component can be written as:
PR
3
eo
k

T

coex
+
v
AA
2

2
coex
+
2w
AAA
3

3
coex
= 0. (4)
A second equation follows from the isothermal compressibility
1

T
= V
P
V

N,T
=
coex
+ v
AA

2
coex
+ 2w
AAA

3
coex
, (5)
which can be expressed in terms of the dimensionless compressibility N by:
1

T
=
coex
k

T(1 + N). (6)


Making use of (4) and (6) we can express the single component virial coecients
as
v
AA
= 2
N + 3

coex
and w
AAA
=
2
3
_
N + 2

2
coex
_
. (7)
Notice that in a way these results resemble the essential behavior of Lennard-
Jones-like potentials, in which attractive interactions dominate at the limit of
large intermolecular distances (low densities).
The mixed coecient v
AB
can be related with the Flory-Huggins parameter
in the following way [4]
v
AB
=
N

coex
+
1
2
(v
AA
+ v
BB
). (8)
is a measure of the anity between beads of dierent spices.
2
In this work, the virial coecients for the single component hydrophilic
beads are chosen such that the interactions between them are purely repulsive
(v
BB
, w
BBB
0) and, for simplicity, the rest of the third order mixing coef-
cients are set equal to w
AAA
. This way, we have described the non-bonded
interactions in terms of four experimentally accessible quantities:
coex
, N,
N and R
eo
.
Now lets have a look at the densities dened in (3). The local character of
this denition doesnt account for the short-range correlations that take place
at high densities [5]. A way to handle this situation is through the denition of
coarse-grained densities ( ), which smooth the peaks of
(r) =
_
dr

(r

)w(|r r

|) (9)
where w(r) is a normalized weight function (
_
drw(r) = 1).
As noted when approximating virial coecients, second order terms usually
correspond to attractive interactions, whereas third order ones stand for repul-
sion. So, their contribution to packing eects will be dierent. This fact can be
incorporated into our model, setting up dierent weighting functions for second
and third order coecients (w
m
with m = 2, 3). Taking this into account, the
weighting densities for our model will be given by

m
=
R
3
eo
N
nN

i=1
w
m
(|r
i
r|)
t(i)
. (10)
Following [3] we adopt the next splines as the denition of our weighting
functions:
w
2
(r) = A
_
_
_
(r
c
a)
3
, 0 r < a,
2r
3
3(a + r
c
)r
2
+ 6arr
c
3ar
2
c
+ r
3
c
, a r < r
c
,
0, r, r
c
,
(11)
and
w
3
(r) =
15
2
_
(r
c
r)
2
, 0 r < r
c
,
0, r, r
c
,
(12)
where r
c
is the cuto radius for non-bonded interactions, A = 15/2(2a
6

3a
5
r
c
+ 3ar
5
c
2r
6
c
) is a normalization constant, and a is a parameter that
controls beads hardness, i.e., for large a values, beads resemble hard spheres
whereas for small values they exhibit a soft volume and no packing eects are
observed [7].
Finally, making use of the denition for the weighted densities (10), we can
rewrite the non-bonded Hamiltonian as:
H
nb
k

T
=

t(i)
_
v

2N

2
(r
i
) +
w

3N

3
(r
i
)
3
(r
i
)
_
(13)
3
1.2 Computational part
In this section we give an overview of the computational aspects involved in the
simulation of a collection of n chains described by our model. The rst point to
address is the correct setup of control parameters. We have already discussed
how virial coecients are related through some experimentally accessible quan-
tities. However, as pointed out in [3], there isnt an analytical expression that
tells us the existing relationship between parameters in the bonded Hamilto-
nian (1). In the referred work, the congurational space of a single chain was
sampled in order to achieve values of (r
N
r
1
)
2
consistent with the desired
value of R
eo
, for given values of k
s
, l
0
, k
b
and N. We will use those results to
set the rest of our parameters.
The equations of motion are obtained by dierentiating (1) and (13) with
respect to bead coordinates (r
i
), yielding
F
spr
i
= k
s
_
1
l
0
r
i,i+1
_
r
i,i+1
(14)
F
ang
i1
=
k
b
r
i1,i
[r
i,i+1
r
i1,i
cos
i
] (15a)
F
ang
i
=
k
b
r
i1,i
[r
i,i+1
r
i1,i
cos
i
]
k
b
r
i,i+1
[r
i1,i
r
i,i+1
cos
i
] (15b)
F
ang
i+1
=
k
b
r
i,i+1
[r
i1,i
r
i,i+1
cos
i
] (15c)
F
nbd
i
=
k

TR
3
eo
N
2

j
r
ij
_
v
t(i)t(j)
w

2
(r
ij
)
+
2w
t(i)t(j)
R
3
eo
3N
w

3
(r
ij
) [
3
(r
i
) +
3
(r
j
)]
_
(16)
where superscripts spr, ang and nbd stand for spring-like, bond angle and non-
bonded interactions respectively. These equations will be integrated with the
aid of velocity Verlet algorithm. Some of the features of this second order, two-
step method are its easy implementation, time reversibility, angular momentum
conservation and symplecticity [8]. In the rst step (prior to forces evalua-
tion), velocities are updated by half-time step (h/2) using old accelerations,
then positions are updated by a full time step using the intermediate velocities
v(t + h/2) = v(t) +
h
2
a(t), (17)
r(t + h) = r(t) + hv(t + h/2). (18)
In the second step forces are evaluated and then velocities are updated by an-
other half-time step
v(t + h) = v(t + h/2) +
h
2
a(t + h). (19)
In order to set the desired average temperature (T), the system has to be
coupled to a thermal reservoir. To achieve this, we will use Lowe-Andersen
4
thermostat [9, 10], where, in every time-step the relative velocities of all in-
teracting pairs will be randomly reassigned with probability t (where is
the frequency of collision with reservoirs particles), according to the Maxwell-
Boltzmann distribution for T. The replacement is done such that the total
momentum of the pair is conserved, i.e., the velocity component along the line
passing their centers will be taken from the distribution
ij
_
2k

T/m, where m
is the mass of beads and is a Gaussian distribution with zero mean and unit
variance. So the new velocities will be
v
i
= v
i

1
2
_

ij
_
2k

T
m
v
ij
r
ij
_
r
ij
, (20)
v
j
= v
j
+
1
2
_

ij
_
2k

T
m
v
ij
r
ij
_
r
ij
. (21)
The drawback of this method is its stochastic nature, since the random
reassignment of velocities generates non-continuous trajectories in the phase
space. So, the interpretation of dynamical properties has to be done with special
care. However, as with the standard Andersen thermostat, this generalization
reproduces the canonical distribution, and also has the advantage of momentum
conservation. Hence, this approach will be useful as long as we focus on static
and structural properties.
Some of our simulations will be concerned with mixtures of dierent lipids.
For example, binary mixtures in which N = N
A
B
+N
B
= N
A
C
+N
C
, where N
is the total number of beads in one chain and N
A
B
and N
B
stand, respectively,
for the number of tail and head group beads for one kind of lipid, which will
be denoted by B-type lipid. Of course, the same happens with beads of C-type
lipids. A statistical ensemble particularly useful to study this kind of systems
is the Semi-Grand Canonical [11, 12], in which the total number of chains
(n = n
B
+n
C
, where n

is the total number of -type chains) and their chemical


potentials (
A
and
B
) are xed, while their concentrations

= n

/(n

+n

)
can uctuate. This ensemble is implemented through a Monte Carlo move which
randomly swaps the type of one lipid with an acceptance
P
acc
= min
_
1, e
(U)
_
, (22)
where stands for the dierence between new and old congurations.
5
2 Membrane heterogeneity
During last decades a lot of research as been done to understand the physical
mechanisms behind biological processes, such as the complex biochemical re-
actions taking place in cellular membranes. In early stages, the lipid bilayer
of cellular membranes was thought as a complex mixture of lipids free to dif-
fuse along the membrane and whose only function was to isolate the cell from
its environment [13]. However, further observations suggested that beside its
isolating character, lipid bilayers may play an active roll in the regulation of
cellular functions, since the presence of local inhomogeneities in its composi-
tion has been claimed to aect the functionality and distribution of proteins
embedded in the cellular membrane [14], serve as an entry point for pathogens,
increase the processing speed of cellular signaling [15], etc. Beside the experi-
ments showing their existence, there is a lot of debate regarding the nature and
driving forces for the formation of these rafts. In order to have a better under-
standing of these structures, in the last years, research has been focused on the
study of simpler systems known as giant unilamellar vesicles (GUVs), in which
membranes composition can be controlled. In particular, one of the simplest
structures that exhibit the formation of this domains, are bilayers composed by
binary mixtures of saturated and unsaturated lipids together with cholesterol.
The presence of rafts in these articial membranes has been associated with
the existence of a miscibility critical point [15] near which strong composition
uctuations take place. However, the question remains whether these uctua-
tions last long enough to accomplish the roles they are supposed to. Also, this
theory is incapable of explaining the observed composition asymmetry between
inner and outer leaets [16]. A promising theory to handle these problems de-
picts rafts as characteristic droplets of a curvature-induced microemulsion [17].
This theory relays on the dierent spontaneous curvatures for the lipids con-
forming the bilayer. According to it, lipids with larger headgroups and smaller
tails will be attracted to regions of the outer leaf curved in the inward direction,
i.e., regions where the area per lipid is increased, while lipids with smaller heads
and larger tails will be attracted to the inner leaf, where the area per lipid has
been reduced.
In order to test the validity of this hypothesis we will study the formation
of rafts in membranes built from binary mixtures of single-tailed lipids, all of
them with the same number of chains (N = 16), but dierent types and lengths
for their head groups. The rst group will be conformed by 12 A-type beads for
the tails and 4 B-type beads for the heads, while the second group will be made
of 11 A-type tail beads and 5 C-type head beads. Virial coecients will be set
by expressions (7) and (8). However, we have to include two additional anity
coecient (
AC
and
BC
), one for the interaction between tails and heads of
the new chain (v
AC
) and the other to set the interaction between headgroups of
dierent types (v
BC
), but apart from this the equation (8) remains the same.
For simplicity, third order coecients will be set equal to w
AAA
, with excep-
tion of w
BBB
, w
CBB
, w
CCB
and w
CCC
which will be zero, while second order
coecients for single headgroup beads will be v
BB
= v
CC
= 0.1. The rest of
virial coecients will be obtained from the mentioned equations together with
the following set of parameters:
coex
= 18.0, N = 100.0,
AC
=
AB
= 30.0
and
BC
= 28.8. Finally, parameters associated with bonded interactions will
be: R
eo
= 3.5r
c
, k
s
= 19.0k

T/r
2
c
, l
0
= 0 and k
b
= 5.0k

T.
6
In order to speedup the diusion of lipids in the bilayer, their type will be
randomly swapped as described in the previous section, i.e., simulations will be
carried out in the semi-grand canonical ensemble, setting
B
= 0 for the rst
group (short headgroup) and
C
= for the second one (large headgroup).
The rst thing to do, is characterize the response in the concentration of lipids
of each species (

) as function of . In particular, it is important to identify


the coexistence region where concentration of both spices is around 50%. Also,
we would like to reproduce some qualitative results found by Hoemberg and
Mueller, who observed hysteresis loops in the

vs curves for systems


under dierent tensions: tensionless systems exhibited greater hysteric eects
than those under tension.
Two dierent systems were studied, the rst one consisting of 2901 lipids
enclosed in a simulation box of dimensions 30r
c
30r
c
50r
c
, with the mem-
branes normal oriented along the z axis (the larger one) and lateral tension
set equal to 2.45k

T/r
2
c
. Initially, setting = 0, the system was driven to
a state with high concentration of short headgroup chains, specically 2895 B
chains and 6 C ones.
Figure 1: Initial conguration for the system with 2901 chains. Short head-
groups are depicted in blue and large ones in red.
Then, was slightly increased and we let the system evolve to a new equilib-
rium conguration with a higher concentration of C chains. Once in equilibrium,
we started sampling until we got condent values for the new average concen-
trations,

. Then was incremented again and the equilibration-averaging


process was repeated. This procedure continued until system saturated in the
opposite direction, that is, until we found congurations with high concentra-
tion of C chains (roughly 99.88% for = 8k

T). From that point we started


the way back decreasing . However, as shown in Fig:2 and in accordance
with the results of Hoemberg and Mueller, no hysteric eects were found for
these simulations. Also, at that moment we didnt pay too much attention to
sample the transition region with much detail, since it is characterized by strong
uctuations that had to be averaged over larger times.
Now we present typical snapshots for the cases = 4.1k

T (Fig:3) and
= 4.2k

T (Fig:4). The common remark between these congurations is


7
Figure 2: Concentration of B-type chains as function of the dierence in chem-
ical potentials (). Black curve represent the results obtained by increasing
starting from the conguration with = 0k

T and the red curve cor-


respond to the results obtained by decreasing from the conguration with
= 8k

T.
that lipids with the lower concentration (red ones for the picture at the left and
blue ones for the picture at the right) arent randomly distributed but packed
into small domains all over the membrane. However, we havent detected any
peculiar deformation or asymmetry between upper and lower leaets, related
with these domains. Of course, a more detailed analysis has to be done in order
to determine their typical sizes and spacial distribution. Also, itll be important
to study their stability to make sure they are not driven by the presence of a
miscibility critical point.
Figure 3: Typical conguration
for = 4.1k

T.
Figure 4: Typical conguration
for = 4.2k

T.
8
The second system studied in this work consisted of a tensionless bilayer
made of 4680 lipids with the same chain structures and parameters than those
used in the previous case. In contradiction with the results obtained by Hoem-
berg and Mueller for tensionless systems, no hysteric eect were found in these
simulations (as shown in Fig:5) and in fact, concentration curves look pretty
much the same for both systems with and without tension. This suggests that
something is wrong with our simulations. To verify it, we ran Hoembergs pro-
gram for the equilibrated conguration at = 4.16k

T, nding a dierent
concentration, as shown in Fig:5 with the black arrow. This means that the
results obtained by our program are not equivalent to Hoembergs.
Figure 5: Concentration of B-type chains as function of the dierence in chemi-
cal potentials (). Black curve represent the results obtained by increasing
starting from the conguration with = 0k

T and the red curve correspond to


the results obtained by decreasing from the conguration with = 8k

T.
The arrow represents the dierence in concentration calculated with Hoembergs
program for the same equilibrated conguration at = 4.16.
Next step will be to determine what causes this dierence in calculated
concentrations. There are two probable sources of discrepancy: either the in-
tegration of motion equations or the Monte Carlo move associated with the
semi-grand canonical ensemble.
References
[1] C. Rivetti, C. Walker and C. Bustamante, J. Mol. Biol. 280, 41 (1998).
[2] G. Brannigan, L. Lin and F. Brown, Eur. Biophys. J. 35, 104 (2006).
[3] M. Hoemberg and M. Mueller, J. Chem. Phys. 132, 155104 (2010).
9
[4] M. Mueller, Macromol. Theory Simul. 8, 343 (1999).
[5] P. Tarazona, U. Marini and R. Evans, Mol. Phy. 60, 573 (1987).
[6] P. Tarazona, J. Cuesta and Y. Martnez-Raton, Lect. Notes Phys. 753, 247
(2008).
[7] G. Marelli, Minimal models for lipid membranes: local modications around
fusion objects, PhD thesis.
[8] The leapfrog method and other symplectic algorithms for integration of
Newtons laws of motion.
[9] C. Lowe, Europhys. Lett. 47, 145 (1999).
[10] C. Lowe and E. Koopman, J. Chem. Phys. 124, 204103 (2006).
[11] A. Werner, F. Schmid, M. Mueller and K. Binder, J. Chem. Phys. 107,
8175 (1997).
[12] P. Bolhuis and D. Frenkel, Physica A, 244, 45 (1997).
[13] S. Singer and G. Nicolson, Nature, 175, 720 (1972).
[14] S. Mishra and P. Joshi, J. Neurochem, 103, 135 (2007).
[15] S. Semrau and T. Schmidt, Soft Matter, 5, 3174 (2009).
[16] P. Janmey and P. Kinnunen, Trends. Cell. Biol. 16, 538 (2006).
[17] M. Schick, Phys. Rev. E, 85, 31902 (2012).
10

You might also like