PHYSICAL REVIEW E 93, 012409 (2016)

Swing motion as a diffusion mechanism of lipid bilayers in a gel phase

Younghoon Oh,1 Jeongmin Kim,1 Arun Yethiraj,2 and Bong June Sung1,*
1
Department of Chemistry and Research Institute for Basic Science, Sogang University, Seoul 121-742, Republic of Korea
2
Department of Chemistry, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA
(Received 27 April 2015; revised manuscript received 23 September 2015; published 19 January 2016)
Lipid bilayers are a model system for studying the properties of cell membranes. For lipid bilayers of a single
lipid component, there is a phase transition from a fluid phase to a gel phase as the temperature is decreased. The
dynamic behavior of lipids in the gel phase is interesting: some models show dynamic heterogeneity with a large
disparity in timescales between fast and slow molecules, and a spatial segregation of the slow molecules. In this
paper we study the dynamics of coarse-grained models of lipid bilayers using the dry Martini, Lennard-Jones
Martini, polarizable Martini, and BMW models. All four models show similar dynamical behaviors in the gel
phase although the transition temperature is model-dependent. We find that the primary mode of transport in the
gel phase is a hopping of the lipid molecules. Hopping is seen in both the translational and rotational dynamics,
which are correlated, i.e., the lipid molecules display a swing-like motion in the gel phase.

DOI: 10.1103/PhysRevE.93.012409

I. INTRODUCTION has recently been reported in experiments [41,43] and simula-

tions of a simple model of lipid bilayers [31].
Cell membranes are composed of many lipid components,
In this work, we investigate the translational and rotational
as well as proteins and cholesterols [1–4]. They are structurally
diffusion of DPPC lipids in single component lipid bilayers by
complex [5–7] and display interesting phase [8–14] and
performing molecular dynamics simulations. We employ four
dynamic [15–20] behaviors that could be of physiological different models (big multipole water (BMW) [44], polarizable
importance [21–24]. In the raft hypothesis, for example, Martini [45], Lennard-Jones (LJ) Martini models [46,47], and
it is proposed that a spatial segregation of certain lipids dry Martini models [48]). We find that the fluid-to-gel phase
would be important for the amplification of signaling in transition occurs for all models and the area-per-lipid changes
cells [25–30]. Recently it has been suggested that many of sharply at the transition. But the transition temperature (Tt )
these characteristics could be observed in a single component depends strongly on the model employed. Tt is smaller by
bilayer, namely a one-component phospholipid bilayer near about 40 K with BMW model than polarizable and LJ Martini
the transition from fluid to gel phase [31]. In this paper we models. Tt is higher for the dry Martini model by 25 K than
study the dynamics of the lipid bilayers using coarse-grained for LJ and polarizable Martini models. The lipid diffusion
models for the lipids. constant is much lower than in fluid phase, and the dynamics of
Falk et al. investigated trajectories for dipalmitoylphos- lipids becomes strongly heterogeneous. We also find that lipids
phatidylcholine (DPPC) lipids in the fluid phase and illustrated undergo both translational and rotational hopping motions
that lipids underwent concerted motion along with a clear in gel phase. These hops are correlated, suggesting a swing
streaming pattern [32]. Marrink et al. investigated the phase motion for the transport of lipid molecules.
transition, structures, and dynamics of lipids systematically The rest of the article is organized as follows. The model
and extensively [14]. They showed that there was no evidence and methods are described in Sec. II, results are presented and
for a hexatic phase upon melting, and that the translation and discussed in the Sec. III, and a summary and conclusions are
rotation of lipids slowed down dramatically in the gel phase. presented in Sec. IV.
However, the detailed mechanism of transport of lipids remains
elusive in the gel phase. In this work, we perform molecular
II. SIMULATION MODEL AND METHODS
dynamics simulations and investigate how lipids diffuse in the
gel phase. We show that the lipids undergo swing motion in the We investigate a single component lipid (DPPC) bilayer
gel phase, i.e., lipids hop both translationally and rotationally with both explicit and implicit water models. We employ four
simultaneously. different coarse-grained force-fields: dry Martini, Lennard-
Dynamic heterogeneity is present in a number of complex Jones Martini, polarizable Martini, and big multipole water
fluids including super-cooled liquids [33–37], glasses [38–40], Martini models [44–48]. In LJ Martini, polarizable Martini,
and cells [19,22,41,42]. As the dynamics becomes slower, for and BMW models, four water molecules are coarse-grained
example, when a liquid is cooled toward the glass transition, into a single water bead. In the case of the dry Martini model,
the distribution of displacement of single molecules becomes the effects of solvents are implicitly implemented [48]. LJ
bimodal, with a subset of slower molecules [36]. The slow Martini water model has no electrostatic interaction between
molecules are often spatially correlated into domains that are water beads but just the van der Waals interaction that is
a few molecular diameters in size. Such dynamic heterogeneity described by Lennard-Jones potential. A polarizable Martini
water bead consists of three interaction sites (Fig. 1). The
polarizable Martini water bead interacts with other beads
via the first interaction site by means of Lennard-Jones
*
Corresponding author: [email protected] potential. The other two interaction sites of the polarizable

2470-0045/2016/93(1)/012409(8) 012409-1 ©2016 American Physical Society

OH, KIM, YETHIRAJ, AND SUNG PHYSICAL REVIEW E 93, 012409 (2016)

Dipalmitoylphosphatidylcholine LJ MARTINI Water semi-isotropic NpT ensemble using Berendsen thermostat and
Berendsen barostat. Production runs are also conducted under
1 semi-isotropic NpT ensemble with V-rescale thermostat and
2 Parrinello-Rahman barostat [51]. Semi-isotropic NpT is an
ensemble where the pressures of lateral directions (x and
3
y directions) and normal direction are controlled separately.
Polarizable MARTINI Water The values of pressures for both lateral directions and normal
4
+q direction are, however, set to 1 atm. In our simulations we apply
5
5’ thermostats to three different groups separately: one group for
-q
water molecules and two groups for two leaflets of a lipid
6 bilayer. We use the time-steps, interaction cutoff, and treatment
6’
Big Multipole Water (BMW) of electrostatics as recommended by the developers of the force
7 fields [47,52,53]. Some of the systems are also simulated under
7’ +q
NVE ensemble to check whether there is any unexpected effect
8
-2q of either thermostat or barostat. All of our data produced
8’ +q from both ensembles agree qualitatively with no significant
difference observed. When we conduct semi-isotropic NpT
ensemble MD simulations, the integration time step is 20 fs
FIG. 1. A schematic for three different coarse-grained models for
a DPPC lipid and a water: Big Multipole Water (BMW), Polarizable
and the neighbor list is updated every integration step.
Martini and LJ Martini models.
III. RESULTS AND DISCUSSIONS
Martini water bead contain +q and −q charges, respectively, A. The fluid to gel phase transition
which interact with other beads via Coulomb interaction. The The transition from fluid phase to gel phase occurs at
distance between two interaction sites inside a water bead a temperature Tt , along with a sharp decrease in the area-
is fixed and the harmonic angle potential is employed to per-lipid Al [Fig. 2(a)]. Al changes only gradually with
adjust the polarizability of water. The diameter and mass of temperature T in both gel and fluid phases but decreases
both Lennard-Jones and polarizable Martini water beads are sharply from 0.6 to 0.45 at the transition. In case of DPPC
0.47 nm and 72 amu, respectively. A BMW water bead also lipid bilayers, the experimental value of Tt is around 314 K at
consists of three interaction sites. One of three interaction sites 1 atm [48].
owns a negative charge (−2q) and interacts with other water Considering that the packing efficiency should increase by
beads via a soft Born-Mayer-Huggins (BMH) potential and up to 30% across the transition, it is not surprising that the
coulomb interaction. The other two interaction sites contain lipid-lipid intermolecular interaction should play a critical role
identical positive charges (+q) and interact with other water in the phase transition. Tt , therefore, should depend on the force
beads only via electrostatic interactions. Unlike the polarizable field. The LJ and polarizable Martini models predict Tt ≈
Martini model, both the bond lengths and the angle of three 295 ± 5 K, which is somewhat lower than an experimental
interaction sites are fixed. value of 314 K. On the other hand, for dry Martini, Tt ≈
A DPPC lipid molecule is modeled as a chain of 12 beads 325 ± 5 K. BMW Martini predicts Tt ≈ 263K.
(4 beads for a head group and 8 beads for hydrocarbon tail For a fixed area fraction the surface tension of systems with
groups as depicted in Fig. 1). The chemical bond between BMW water is larger than that with Martini water. Such a large
two neighbor beads is described by a harmonic potential. In surface tension of lipids bilayer with BMW water results in
case of tail groups, the chemical bond length varies from 0.47 larger values of Al than with Martini water for temperatures
to 0.37 nm. The angle between three consecutive beads is between 260 K and 300 K. This may be attributed to the
also described by a harmonic potential. The spring constant
of harmonic angle potential for BMW lipid tail is weaker than
those of both of polarizable Martini and LJ Martini models. 0.75
LJ MARTINI
4
(a) (b) BMW
The simulation system consists of up to 100 000 beads 0.70 Polarizable MARTINI
Polarizable
BMW 3
depending on the simulation cell dimension. In case of explicit
Al (nm )

0.65 Dry MARTINI

2

solvent models, the number of water molecules per lipid is

ρ/ρz

0.60 2
fixed to about 20. When calculating the area per lipid (Al ) for 0.55
different models as a function of temperature, we consider the 0.50 1
systems with 128 lipids. On the other hand, when estimating 0.45
the mobility map for dry Martini models, we consider the 0
240 280 320 -4 -2 0 2 4
systems with 2048 lipids. The initial configurations of DPPC T (K) z (nm)
lipid bilayers are obtained using insane, the versatile python
script provided by Marrink group [49]. The potential energy FIG. 2. (a) Simulation results for Al as a function of T for four
is minimized using steep algorithm embedded in Gromacs. different models, dry Martini, BMW, polarizable Martini, and LJ
We propagate our systems of DPPC lipid bilayers by Martini. (b) The normalized density profiles of lipids (dashed) and
using the simulator GROMACS 4.6.5 [50]. Every system is water molecules (solid) for BMW and polarizable Martini models at
equilibrated at least for several hundred nanoseconds under T = 290 K.

012409-2
SWING MOTION AS A DIFFUSION MECHANISM OF . . . PHYSICAL REVIEW E 93, 012409 (2016)

1
10 Liquid phase
(a) (b) LJ
0 Polarizable

〈(Δr) (t)〉 (nm )

10

2
BMW
Dry
-1
10

-2
10

2
Gel phase
LJ Polarizable
-3
10 BMW Dry
1 ps 1 ns 1 μs
-4
10
-4 -2 0 2 4
10 10 10 10 10

FIG. 3. The second carbon beads (6 and 6 ) of lipid tails are t (ns)
projected onto XY plane. The snapshots are obtained using the dry
Martini model at (a) T = 320 K and (b) T = 330 K for N = 512. FIG. 4. The lateral mean-squared displacements (r)2 (t) of
Vacancies in the snapshots are filled with beads other than 6 and 6 lipids of various models in both liquid and gel phases. Lipid bilayers
carbon beads. at T = 330 K corresponds to the liquid phase regardless of the model
employed. For the gel phase, T = 320 K, 290 K, 290 K, and 250 K for
dry Martini, LJ Martini, polarizable Martini, and BMW, respectively.

fact that the intermolecular interaction between BMW water

and lipid head groups is softer than that between Martini
water and lipid head groups. Softer intermolecular interaction Martini, 290 K for LJ and polarizable Martini, and 250 K for
would allow water molecules to reside more easily at the BMW, respectively). This is because Tt depends on the model
water-lipid interface, thus increasing the effective density of employed to describe lipids. (r)2 (t) of lipids in the gel
the interface and the surface tension. Figure 2(b) depicts the phase is much smaller than in the liquid phase as expected. And
density distribution functions of both DPPC lipid head groups (r)2 (t) of all models show a strong subdiffusive behavior
(dotted) and water molecules (solid) at 290 K. BMW water at intermediate time scales.
molecules penetrate more deeply into the lipid bilayer, thus Figure 5 depicts the mobility map of lipid tail groups (6
increasing the effective density of the water-lipid interface and and 6 carbon beads) in both gel and fluid phases. We estimate
the effective surface tension of the lipid bilayer. This suggests the mobility of a lipid by calculating the distance that the lipid
that the hydration of lipid head groups and the intermolecular travels during a characteristic time τ . The mobility map is
interaction between water and lipids are also important factors estimated from a single trajectory of the simulation system but
for the phase transition. Unless otherwise stated, all simulation is not averaged over leaflets. τ is the average time that a lipid
results presented below are obtained by employing the dry diffuse by its own size, i.e., (r)2 (t = τ ) = Al , where Al
Martini model under a semi-isotropic NpT ensemble. Dynamic is the area per lipid. Al = 0.63 in the liquid phase and 0.48
and structural behaviors of lipids in gel phase are qualitatively in the gel phase. An arrow in the mobility map indicates the
identical for all models. direction that each lipid travels during τ . And the length of the
In the gel phase the lipid molecules show hexagonal arrow represents the magnitude of the lipid lateral mobility. In
orientational ordering as in Fig. 3(a). In Fig. 3 two carbon the fluid phase there is no domain of different dynamics, thus
beads (6 and 6 beads in Fig. 1) of each lipid are projected the lipid dynamics being spatially homogeneous. However, we
onto a xy plane (parallel to the lipid bilayer surface) as a observe the streaming pattern of lipids, i.e., a collective motion
dimer. The hexagonal arrangement of carbon beads (or lipid of neighbor lipids, which is consistent with previous simulation
tails) exhibits a long-range correlation. On the other hand, such studies [32]. For example, in the circle of Fig. 5(b), the
hexagonal arrangement of lipid tails is not observed in fluid directions of arrows of lipids mobility are strongly correlated
phase [Fig. 3(b)]. during the characteristic time τ .

B. Dynamic heterogeneity of lipids in gel phase (nm2)

2 (a) (b)
The lateral mean-squared displacements ((r) (t)) of
lipids and the lateral diffusion coefficient (D) are calculated us-
ing various models with N = 512 lipids. D changes drastically
by more than 3 orders of magnitude as the phase of membrane
changes. Figure 4 depicts (r)2 (t) for different models in
both liquid and gel phases. Lipid bilayers at T = 330 K
corresponds to the liquid phase regardless of models employed.
At T = 330 K in the liquid phase, (r)2 (t) of different 5 nm 5 nm
models are close to one another. In the Fig. 4, different
temperatures are used for lipids of different models in the FIG. 5. Mobility maps of lipids of (a) the gel phase and (b) the
gel phase. The highest temperature in our simulations at liquid phase. N = 2048 lipids of dry Martini models are used to
which lipids can form the gel phase is used (320 K for dry obtain the mobility maps.

012409-3
OH, KIM, YETHIRAJ, AND SUNG PHYSICAL REVIEW E 93, 012409 (2016)

(nm)
(a) LJ Martini (b) Dry Martini (c) Polarizable Martini

5nm 5nm 5nm

(d) LJ Martini (e) Dry Martini (f) Polarizable Martini

5nm 5nm 5nm

FIG. 6. Mobility maps of different models and phases of lipid bilayers that consist of 2048 DPPC molecules. (a)–(c) represent liquid phase
bilayers and (d)–(f) represent gel phase bilayers that are constructed using dry Martini, LJ Martini, and polarizable Martini models.

The mobility maps obtained using different models are depicts the translational displacement probability distribution
qualitatively similar to one another. Regardless of models function (P (r,t) = 2π rGs (r,t)) of lipids in both fluid and gel
(dry Martini, LJ Martini, polarizable Martini, and BMW), phases, where r(t) is the position vector of two second tail
the translational diffusion is homogeneous in the liquid phase carbons (6 and 6 ) of a lipid at time t. In the fluid phase at
while fast and slow domains coexist in the gel phase. The T = 330 K, P (r,t) is singly peaked as expected. A solid line
−r 2 /(r)2 (t)
mobility maps obtained from other models are depicted in is a fit to the simulation results using Gs (r,t) = eπ(r)2 (t) .
Fig. 6. All models in Fig. 6 show homogeneous dynamics An agreement between the fit and simulation results indicates
and collective motion of neighbor lipids. The results are that the translational motion of lipids at liquid phase should
consistent with previous simulation studies [32]. Figures 6(d)– show Gaussian statistics. In the gel phase, on the other hand,
6(f) represent the mobility map of gel phase. In the gel phase, P (r,t) is multiply peaked, indicating that lipid lateral diffusion
the dynamic heterogeneity appears for the lipid diffusion, is spatially heterogeneous, i.e., some lipids are sedentary and
i.e., the fast and slow dynamic regions are separated clearly. confined in a cage while other lipids are quite mobile. More
Previous studies suggested that the collective motion in the interesting is that the peak positions correspond to the integral
liquid phase would result from the hydrodynamic interactions multiples of λ = 0.5 nm, which is the first peak position of the
of water molecules [54,55]. In our simulations with different radial distribution functions between the second tail carbons.
models, however, we observe such a collective motion in both This suggests that lipids would undergo translational hopping
implicit (dry Martini) and explicit water models. motions. Figure 8 depicts values of x coordinates of 6 and 6
In the gel phase the domains of mobile lipids are observed
clearly in the background of sedentary lipids. And the lipid
diffusion is spatially heterogeneous in gel phase. In gel phase 1
5 T = 320 K 10 T = 320 K
[Fig. 5(a)] domains of mobile lipids (with long red arrows) (a) T = 330 K
(b) T = 330 K
appear and become surrounded by slow lipids (with short 4 0
Ga(θ,t = τ)

10
P(r, t = τ)

purple arrows). Slow lipids in the gel phase hardly diffuse 3

-1
10
during τ while mobile lipids diffuse more than their own size. 2
This shows that the dynamic heterogeneity exists even in single 1 10
-2

component lipid bilayers in the gel phase.

0 -3
10
0.0 0.5 1.0 1.5 2.0 2.5 -4 -2 0 2 4

C. Swing motion of lipids in the gel phase r (nm) θ (rad)

The lipid molecules undergo both translational and rota- FIG. 7. (a) P (r,t)(=2π rGs (r,t)) and (b) Ga (θ,t) of lipid tails at
tional hopping motions in the gel phase. When the diffusion of T = 320 K and 330 K. T = 330 K and 320 K correspond to the liquid
a molecule is Fickian, i.e., (r)2 (t) ∼ t, the self-part of the and gel phases of lipids of dry Martini model, respectively. The black
van Hove correlation function Gs (r,t) ≡ δ{r − [r (t) − r(t = solid lines are fits of Gaussian functions to the simulation results (red
0)]} is usually expected to be Gaussian [56]. Figure 7(a) symbols) at 330 K.

012409-4
SWING MOTION AS A DIFFUSION MECHANISM OF . . . PHYSICAL REVIEW E 93, 012409 (2016)

12 30 be attributed to the hexagonal orientational arrangement of

Δx6(t)
lipid tails in the gel phase. Figure 8 depicts θ (t) of a lipid as
Δx6'(t) 25 a function of time and corroborates the existence of rotational
8
θ(t) hopping motions.

θ(t) (rad)
Δx(t)/λ

20 More interesting is that rotational hopping motions often

4
accompany translational hopping motions. As depicted in
15
0
Figs. 8(a) and Fig. 8(b), the translational hopping motions
at t = 15 and 18 μs occur together with rotational hopping
10
motions. When both translational and rotational hopping
-4
motions occur simultaneously, a lipid swings as illustrated
0 10 20 30 40 in Fig. 8 (see also Supplemental Material [59]). We find
t (μs) that lipids swing quite frequently in the gel phase, which
should be an important transport mechanism in the gel
FIG. 8. A trajectory of a single lipid obtained by using the dry phase.
Martini model. The x coordinates and θ of 6 or 6 carbon beads of In order to quantify and investigate the correlation be-
the lipid are depicted as a function of time t. A schematic figure for tween translational and rotational hopping motions, we es-
the swing motion is also included. timate the two-dimensional probability distribution functions
(P (r6 ,r6 )) of the displacement vectors (r6 ,r6 ) of the
second-tail carbon beads (6 and 6 ) of a lipid at a characteristic
beads of a single lipid as a function of time t. As clearly shown time t = τ . In the liquid phase, P (r6 ,r6 ) is singly peaked
in Fig. 8, lipids do hop translationally to diffuse. with a peak at (r6 ,r6 ) = (λ, λ) [Fig. 9(b)]. λ is the
We also investigate the van Hove angular correlation first peak position of the radial distribution function of the
function, Ga (θ,t) ≡ δ{θ − [θ (t) − θ (t = 0)]}, where θ (t) is second carbon beads. In the gel phase [Fig. 9(a)], on the other
the unbounded orientational angle of the vector that connects hand, P (r6 ,r6 ) shows multiple peaks, which indicates
the bead 6 to the bead 6 at time t [57,58]. Note that the clearly that lipids in gel phase should undergo the translational
Brownian rotational diffusion of a two-dimensional rotor is hopping motions. We assign each peak to a corresponding
equivalent with the one-dimensional random walk. Therefore, region as indicated in Fig. 9(a). In region I, both the second
if the rotational diffusion of the dimer of the 6 and 6 carbon carbon beads are sedentary and do not translate much within
beads of a lipid is Brownian (or Fickian), Ga (θ,t) should be t = τ . In regions II and II , one of the two second carbon
Gaussian. In case of liquid phase at T = 330 K, our simulation beads undergoes a translational hopping motion while the
results for Ga (θ,t) are Gaussian for relatively small angles. A other stays sedentary. In region III, both the second carbon
black solid line in Fig. 7(b) is the fit to the simulation result beads undergo a translational hopping motion. In region IV,
2 2
e−θ /2θ (t) the second carbon beads seem to undergo multiple hopping
using Ga (θ,t) = √ 2
. For −2π  θ  2π , simulation motions.
2πθ (t)
results and the Gaussian function overlap well with each other. As depicted in Fig. 10(a), we also estimate the correlation
On the other hand, for large angles (|θ (t)| > 2π ), Ga (θ,t) function of cos(θ (t = τ )) for a given set of (r6 ,r6 ). θ (t) is
deviates from the Gaussian statistics, which indicates that a the angle displacement of the vector that connects the beads 6
small fraction of lipids may rarely undergo long-angle hopping and 6 during time t. Regions I, II, II, and III in Fig. 10(a)
motions. Interesting is that in the gel phase Ga (θ,t) is multiply are determined based on the two-dimensional probability
peaked with peaks at multiple integers of π/3, which indicates distribution functions (P (r6 ,r6 )) in Fig. 9(a). In region
that lipids may undergo rotational hopping motions, too. The I, cos(θ (t = τ )) ≈ 1 (red), thus implying that the second
rotational hopping motions by multiple integers of π/3 should carbon beads of this region hardly rotate and their rotational

FIG. 9. The two-dimensional probability distribution functions (P (r6 ,r6 )) of the displacement vectors (r6 ,r6 ) of the second-tail
carbon beads (6 and 6 ) of a lipid at a characteristic time t = τ in (a) the gel phase and (b) the liquid phase.

012409-5
OH, KIM, YETHIRAJ, AND SUNG PHYSICAL REVIEW E 93, 012409 (2016)

1.5
(a) 〈cos(θ(t=τ))〉 (b) 3 Region I
1.0 Region II
Region II'

P(cos(θ(t=τ)))
II III
Δr6'(t=τ)/λ
1.0 2 Region III
0.5

0.0
0.5 1
-0.5
I II’
0.0 -1.0 0
0.0 0.5 1.0 1.5
-1.0 -0.5 0.0 0.5 1.0
Δr6(t=τ)/λ cos(θ(t=τ))

FIG. 10. (a) Simulation results for cos(θ(t = τ )) as a function of r6 and r6 and (b) its probability distribution functions
P (cos(θ (t = τ ))) for different regions in the gel phase.

motions are sedentary. Therefore, in region I, the lipids neither of r6 and r6 (Fig. 11). rcross is defined as rcross ≡
translate nor rotate. In regions II and II , cos(θ (t = τ )) ≈ 0.5 min{|r6 (t) − r6 (t = 0)|,|r6 (t) − r6 (t = 0)|}. rcross is a mea-
(yellow), which indicates that θ (t) ≈ 60◦ and the bond vector sure of whether one of two 6 and 6 carbon beads takes over
between the beads 6 and 6 rotates by about 60◦ during the position of other bead at time t = 0. If rcross ≈ 0, the
t = τ . Considering that in regions II and II only one of carbon bead 6 (or 6 ) is located at the initial position of
the two second carbon beads undergo a translational hopping the carbon bead 6 (or 6). Interestingly, as shown in Fig. 11,
motion, the beads 6 and 6 in regions II and II undergo swing rcross ≈ 0 for the region III. This implies that in the region
motions by hopping rotationally and translationally at the same III lipid should diffuse with constrains that rcross ≈ 0 and
time. θ (t = τ ) ≈ 0◦ , 180◦ , 60◦ , and −60◦ . We realize that there
In region III, cos(θ (t = τ )) ≈ 0 (green), which seemingly could be several paths to satisfy those constraints as shown
suggests that the lipids would rotate by 90◦ . But if we take in the following schematic of Fig. 12. One of the paths
a look at the distribution P (cos(θ (t = τ ))) of the values of would be that a single lipid may glide such that one of two
cos(θ (t = τ )) in region III [Fig. 10(b)], cos(θ (t = τ )) can carbon atoms can take over the initial position of the other
have roughly four different values: cos(θ (t = τ )) = −1, −0.5, carbon atom, and rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ 1
0.5, and 1. This suggests that the lipids might either glide or [Fig. 12(a)]. A lipid may undergo several swing motions
rotate by one of the angles, 180◦ , 60◦ , and −60◦ . However, in within the characteristic time τ . For example, as shown in
regions I, II, and II , P (cos(θ (t = τ )))’s have only one peak at Fig. 12(b), a lipid may undergo a two-step swing motion where
cos(θ (t = τ )) = 1, 0.5, and −0.5, respectively. This implies the A particle would be placed at the initial position of the
that there should be only one dominant rotational motion in B particle, thus rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ −0.5.
regions I, II, and II . Similarly, a lipid may undergo more complicated motions
In order to scrutinize the dynamics of lipids in region III, we such that rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ 0.5 or −1.
estimate the cross-correlation function (rcross ) as a function In Figs. 12(c) and Fig. 12(d), a lipid may swing three
times within the characteristic time. In case of the type I
three-step swing motion, rcross (t = τ ) ≈ 0 and cos(θ (t =
τ ) ≈ 0.5. And for the type II three-step swing motion, the
positions of A and B particles are switched, resulting in
rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ −1. In regions I, II,
II , and III of the gel phase, lipids undergo both rotational
and translational hopping motions that are strongly correlated
with each other, thus making the lipids swing in the gel
phase.

IV. SUMMARY AND CONCLUSIONS

We investigate both translational and rotational diffusion
of lipids in single component lipid bilayers. We find that not
only the lipid-lipid interaction but also lipid-water interaction
play a critical role in the phase transition from the fluid
phase to the gel phase. How effectively water molecules
FIG. 11. The correlation function (rcross ) as a function of r6 hydrate the lipid head groups influences the surface tension,
and r6 . thus affecting the phase transition. In the gel phase, the

012409-6
SWING MOTION AS A DIFFUSION MECHANISM OF . . . PHYSICAL REVIEW E 93, 012409 (2016)

(a) hydrocarbon tails of lipids align normal to the bilayer surface.

And when we project two carbon beads of each lipid (6 and
6 carbon beads in Fig. 1) as a dimer onto a xy plane, we can
observe long-range hexagonal arrangement of those dimers
in the gel phase. These features have been observed before
for some CG models and we confirm that other than the
(b) transition temperature, the features are similar in all models
tested.
In the gel phase, a single lipid hops either translation-
ally or rotationally. Trajectories of single lipids and the
multi-peaked van Hove correlation functions (Gs (r,t) and
Ga (θ,t)) show that lipids hop in the gel phase. More inter-
esting is that the rotational hopping motion is accompanied
by the translational hopping motion, thus resulting in the
swing motion of lipids. Such swing motion of lipids are
(c) observed quite frequently in our simulations. We inves-
tigated three different correlation functions, P (r6 ,r6 ),
cos(θ (t = τ )), and rcross , and confirm that the transla-
tional and rotational motions are highly correlated with each
other.
Cell membranes usually consist of more than one type of
lipids. In many cases, cell membranes are mixtures of nega-
tively charged lipids (such as phosphatidylserine) and zwitte-
rioninc lipids. Scientific questions may arise accordingly: how
(d) the composition of negatively charged lipids would affect the
phase transition and dynamic heterogeneity and what size the
domains of mobile lipids would be, etc. We plan to perform
molecular simulations in a future study in order to tackle such
questions.

ACKNOWLEDGMENTS
FIG. 12. Possible mechanisms for the lipid dynamics in the region
III where rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ 1, 0.5, −0.5, and This research was supported by the EDISON (EDucation-
−1. The dotted circles represent the initial positions of 6 and 6 research Integration through Simulation On the Net) Program
carbon beads. Red, orange, and yellow circles represent the position through the National Research Foundation of Korea (NRF)
of 6 and 6 carbon beads after the first, second, and third swing funded by the Ministry of Science, ICT & Future Planning
motions, respectively. (a) A lipid glides and the A particle takes (Grant No. NRF-2012M3C1A6035363). This work was also
over the initial position of the B particle such that rcross (t = τ ) ≈ 0 supported by the National Institute of Supercomputing and
and cos(θ(t = τ ) ≈ 1. (b) A lipid swings two times by about 60◦ , Network/Korea Institute of Science and Technology Infor-
thus rcross (t = τ ) ≈ 0 and cos(θ (t = τ ) ≈ −0.5. (c) and (d) A lipid mation with supercomputing resources including technical
swings three times such that one of two particles takes over the initial support (Grant No. KSC-2013-C2-017).
position of the other particle such that rcross (t = τ ) ≈ 0.

[1] H. I. Ingólfsson, M. N. Melo, F. J. van Eerden, C. Arnarez, [7] K. Jacobson, O. G. Mouritsen, and R. G. W. Anderson, Nat. Cell
C. A. López, T. A. Wassenaar, X. Periole, A. H. de Vries, Biol. 9, 7 (2007).
D. P. Tieleman, and S. J. Marrink, J. Am. Chem. Soc. 136, [8] S. L. Duncan, I. S. Dalal, and R. G. Larson, Biochim. Biophys.
14554 (2014). Acta 1808, 2450 (2011).
[2] G. Van Meer, D. R. Voelker, and G. W. Feigenson, Nat. Rev. [9] J. M. Kosterlitz and D. J. Thouless, J. Phys. C: Solid State Phys.
Mol. Cell Biol. 9, 112 (2008). 6, 1181 (1973).
[3] J. L. Sampaio, M. J. Gerl, and C. Klose, Proc. Natl. Acad. Sci. [10] S. L. Veatch and S. L. Keller, Phys. Rev. Lett. 89, 268101 (2002).
USA 108, 1903 (2011). [11] J. C. Stachowiak, C. C. Hayden, M. A. A. Sanchez, J. Wang,
[4] C. Klose, M. A. Surma, and K. Simons, Curr. Opin. Cell Biol. B. C. Bunker, J. A. Voigt, and D. Y. Sasaki, Langmuir 27, 1457
25, 406 (2013). (2011).
[5] G. van Meer, EMBO J. 24, 3159 (2005). [12] A. T. Hammond, F. A. Heberle, T. Baumgart, D. Holowka,
[6] K. Simons and G. Van Meer, Biochemistry 27, 6197 B. Baird, and G. W. Feigenson, Proc. Natl. Acad. Sci. USA
(1988). 102, 6320 (2005).

012409-7
OH, KIM, YETHIRAJ, AND SUNG PHYSICAL REVIEW E 93, 012409 (2016)

[13] S. Baoukina and E. Mendez-Villuendas, J. Am. Chem. Soc. 134, [38] M. D. Ediger and P. Harrowell, J. Chem. Phys. 137, 080901
17543 (2012). (2012).
[14] S. J. Marrink, J. Risselada, and A. E. Mark, Chem. Phys. Lipids [39] K. Paeng and L. J. Kaufman, Chem. Soc. Rev. 43, 977 (2014).
135, 223 (2005). [40] S. Karmakar, E. Lerner, and I. Procaccia, Physica A 391, 1001
[15] P. G. Saffman and M. Delbruck, Proc. Natl. Acad. Sci. USA 72, (2012).
3111 (1975). [41] L. Zhang and S. Granick, MRS Bull. 31, 527 (2006).
[16] M. P. Clausen and B. C. Lagerholm, Nano Lett. 13, 2332 (2013). [42] S. Pronk, E. Lindahl, and P. M. Kasson, Nat. Commun. 5, 1
[17] Y. Gambin and R. Lopez-Esparza, Proc. Natl. Acad. Sci. USA (2013).
103, 2098 (2006). [43] D. Lingwood, K. HJ, I. Levental, and K. Simons, Biochem. Soc.
[18] F. Quemeneur, J. K. Sigurdsson, M. Renner, P. J. Atzberger, Trans. 37, 955 (2009).
P. Bassereau, and D. Lacoste, Proc. Natl. Acad. Sci. USA 111, [44] Z. Wu, Q. Cui, and A. Yethiraj, J. Phys. Chem. B 114, 10524
5083 (2014). (2010).
[19] J.-H. Jeon, Hector Martinez-Seara Monne, M. Javanainen, and [45] S. O. Yesylevskyy, L. V. Schäfer, and D. Sengupta, PLoS Comp.
R. Metzler, Phys. Rev. Lett. 109, 188103 (2012). Biol. 6, e1000810 (2010).
[20] R. S. Cantor, J. Phys. Chem. B 101, 1723 (1997). [46] S. J. Marrink, A. H. de Vries, and A. E. Mark, J. Phys. Chem. B
[21] M. Pinot, S. Vanni, S. Pagnotta, and S. Lacas-Gervais, Science 108, 750 (2004).
345, 693 (2014). [47] S. J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tieleman, and
[22] B. R. Parry, I. V. Surovtsev, M. T. Cabeen, and C. S. O’Hern, A. H. de Vries, J. Phys. Chem. B 111, 7812 (2007).
Cell 156, 183 (2014). [48] C. Arnarez, J. J. Uusitalo, and M. F. Masman, J. Chem. Theory
[23] T. B. Pedersen, M. C. Sabra, and S. Frokjaer, Chem. Phys. Lipids Comput. 11, 260 (2015).
113, 83 (2001). [49] T. A. Wassenaar, H. I. Ingólfsson, R. A. Böckmann, D. P.
[24] E. Oldfield and D. Chapman, FEBS Lett. 23, 285 (1972). Tieleman, and S. J. Marrink, J. Chem. Theory Comput. 11,
[25] D. A. Brown and E. London, Annu. Rev. Cell Dev. Biol. 14, 111 2144 (2015).
(1998). [50] S. Pronk, S. Pall, R. Schulz, P. Larsson, P. Bjelkmar, R.
[26] K. Simons and E. Ikonen, Nature 387, 569 (1997). Apostolov, M. R. Shirts, J. C. Smith, P. M. Kasson, D. van der
[27] M. Edidin, Annu. Rev. Biophys. Biomol. Struct. 32, 257 (2003). Spoel, B. Hess, and E. Lindahl, Bioinformatics 29, 845 (2013).
[28] K. Jacobson and C. Dietrich, Trends Cell Biol. 9, 87 (1999). [51] G. Bussi, D. Donadio, and M. Parrinello, J. Chem. Phys. 126,
[29] E. J. Smart, G. A. Graf, M. A. McNiven, W. C. Sessa, J. A. 014101 (2007).
Engelman, P. E. Scherer, T. Okamoto, and M. P. Lisanti, Mol. [52] S. J. Marrink, X. Periole, D. P. Tieleman, and A. H. de Vries,
Cell Biol. 19, 7289 (1999). Phys. Chem. Chem. Phys. 12, 2254 (2010).
[30] A. Hryniewicz-Jankowska, K. Augoff, A. Biernatowska, J. [53] M. Winger, D. Trzesniak, R. Baron, and W. F. van Gunsteren,
Podkalicka, and A. F. Sikorski, Biochim. Biophys. Acta 1845, Phys. Chem. Chem. Phys. 11, 1934 (2009).
155 (2014). [54] Y. Wang, J. K. Sigurdsson, E. Brandt, and P. J. Atzberger, Phys.
[31] F. W. Starr, B. Hartmann, and J. F. Douglas, Soft Matter 10, Rev. E 88, 023301 (2013).
3036 (2014). [55] T. Ando and J. Skolnick, Biophys. J. 104, 96 (2013).
[32] E. Falck, T. Róg, and M. Karttunen, J. Am. Chem. Soc. 130, 44 [56] M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids
(2008). (Oxford University Press, Oxford, 1989).
[33] E. Flenner, H. Staley, and G. Szamel, Phys. Rev. Lett. 112, [57] J.-P. Hansen and I. R. McDonald, Theory of Simple Liquids
097801 (2014). (Academic Press, New York, 2013).
[34] P. G. Debenedetti and F. H. Stillinger, Nature 410, 259 (2001). [58] K. V. Edmond, M. T. Elsesser, G. L. Hunter, D. J. Pine, and
[35] K. Kim and R. Yamamoto, Phys. Rev. E 61, R41 (2000). E. R. Weeks, Proc. Natl. Acad. Sci. USA 109, 17891 (2012).
[36] C. Donati, J. F. Douglas, W. Kob, S. J. Plimpton, P. H. Poole, [59] See Supplemental Material at http://link.aps.org/supplemental/
and S. C. Glotzer, Phys. Rev. Lett. 80, 2338 (1998). 10.1103/PhysRevE.93.012409 for the supplementary animation
[37] M. D. Ediger, Annu. Rev. Phys. Chem. 51, 99 (2000). and a brief description for the animation.

012409-8