Control of membrane fusion mechanism by lipid composition: predictions from ensemble molecular dynamics.

Abstract

Membrane fusion is critical to biological processes such as viral infection, endocrine hormone secretion, and neurotransmission, yet the precise mechanistic details of the fusion process remain unknown. Current experimental and computational model systems approximate the complex physiological membrane environment for fusion using one or a few protein and lipid species. Here, we report results of a computational model system for fusion in which the ratio of lipid components was systematically varied, using thousands of simulations of up to a microsecond in length to predict the effects of lipid composition on both fusion kinetics and mechanism. In our simulations, increased phosphatidylcholine content in vesicles causes increased activation energies for formation of the initial stalk-like intermediate for fusion and of hemifusion intermediates, in accordance with previous continuum-mechanics theoretical treatments. We also use our large simulation dataset to quantitatively compare the mechanism by which vesicles fuse at different lipid compositions, showing a significant difference in fusion kinetics and mechanism at different compositions simulated. As physiological membranes have different compositions in the inner and outer leaflets, we examine the effect of such asymmetry, as well as the effect of membrane curvature on fusion. These predicted effects of lipid composition on fusion mechanism both underscore the way in which experimental model system construction may affect the observed mechanism of fusion and illustrate a potential mechanism for cellular regulation of the fusion process by altering membrane composition.

Full text

Page 1

Control of Membrane Fusion Mechanism by
Lipid Composition: Predictions from Ensemble
Molecular Dynamics
Peter M. Kasson*, Vijay S. Pande*
Department of Chemistry, Stanford University, Stanford, California, United States of America
Membrane fusion is critical to biological processes such as viral infection, endocrine hormone secretion, and
neurotransmission, yet the precise mechanistic details of the fusion process remain unknown. Current experimental
and computational model systems approximate the complex physiological membrane environment for fusion using
one or a few protein and lipid species. Here, we report results of a computational model system for fusion in which the
ratio of lipid components was systematically varied, using thousands of simulations of up to a microsecond in length to
predict the effects of lipid composition on both fusion kinetics and mechanism. In our simulations, increased
phosphatidylcholine content in vesicles causes increased activation energies for formation of the initial stalk-like
intermediate for fusion and of hemifusion intermediates, in accordance with previous continuum-mechanics
theoretical treatments. We also use our large simulation dataset to quantitatively compare the mechanism by which
vesicles fuse at different lipid compositions, showing a significant difference in fusion kinetics and mechanism at
different compositions simulated. As physiological membranes have different compositions in the inner and outer
leaflets, we examine the effect of such asymmetry, as well as the effect of membrane curvature on fusion. These
predicted effects of lipid composition on fusion mechanism both underscore the way in which experimental model
system construction may affect the observed mechanism of fusion and illustrate a potential mechanism for cellular
regulation of the fusion process by altering membrane composition.
Citation: Kasson PM, Pande VS (2007) Control of membrane fusion mechanism by lipid composition: Predictions from ensemble molecular dynamics. PLoS Comput Biol 3(11):
e220. doi:10.1371/journal.pcbi.0030220
Introduction
Membrane fusion plays a key role in cellular function,
allowing processes as diverse as neurotransmitter release,
secretion of peptide hormones, and infection by enveloped
viruses. Fusion is also critical in allowing intracellular
transport. The cellular membranes participating in these
fusion reactions contain complex mixtures of lipids and
proteins. The composition and potentially the organization
of these mixtures are both variable and closely regulated by
the cell [1–4].
A variety of experimental and computational model
systems [5–14] have been used to gain insight into the basic
physical properties underlying membrane fusion. Such model
systems are of necessity much simpler in nature than the
physiologic context for fusion, containing one or a few lipid
species and omitting or simplifying the protein environment.
In designing and interpreting model systems for fusion, it is
important to consider how variations in the lipid mixtures
chosen may affect the results. Investigating how changes in
lipid composition affect the kinetics and mechanism of fusion
also provides fundamental insight by illuminating which
aspects of fusion are robust to small perturbations to the
model system used and which are highly model-dependent.
Experimental data on lipid vesicle fusion indicate that
variation in lipid composition plays an important role in
determining fusogenicity. Much initial work concentrated on
the effects on fusion by lipids that prefer either positive or
negative membrane curvature [5,15]. Perhaps most exten-
sively characterized is the association between increased
fusogenicity and increased fraction of phosphatidylethanol-
amine (PE) [7], a lipid headgroup thought to promote
negative membrane curvature. Plasma membranes in fuso-
genic contexts (synaptic vesicles, synaptic membranes, viral
membranes) have a ratio of phosphatidylcholine (PC) to PE
headgroups measured at between 0.9 and 2.0 [16–18]. The
presence of PE has been identified as a critical factor for
fusion [19–21], and many systems used to develop structural
models for fusion intermediates contain a substantially
higher proportion of PE than physiologic levels [3,8,22].
Similarly, theoretical treatments of proposed fusion inter-
mediates suggest that PE is important to the energetic
favorability of the stalk-like state that has been proposed as
an early fusion intermediate [8,10,23–25]. Other lipid species
likely play an additional modulatory role in fusion, and the
PC:PE ratio in physiological membranes also varies inde-
Editor: Philip E. Bourne, University of California San Diego, United States of
America
Received April 13, 2007; Accepted September 26, 2007; Published November 16,
2007
A previous version of this article appeared as an Early Online Release on September
26, 2007 (doi:10.1371/journal.pcbi.0030220.eor).
Copyright: ? 2007 Kasson and Pande. This is an open-access article distributed
under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the
original author and source are credited.
Abbreviations: MSM, Markovian state model; PC, phosphatidylcholine; PE,
phosphatidylethanolamine; POPC, palmitoyloleoyl-PC; POPE, palmitoyloleoyl-PE
* To whom correspondence should be addressed. E-mail: kasson@stanford.edu
(PMK); pande@stanford.edu (VSP)
PLoS Computational Biology | www.ploscompbiol.org November 2007 | Volume 3 | Issue 11 | e2202228

Page 2

pendently in the inner and outer leaflets [26,27]. In this work,
we examine the effect of PC:PE ratio as a first-order
approximation of lipid composition effects; we first treat
uniform composition in the inner and outer membrane
leaflets and then examine asymmetric compositions.
Membrane curvature has long been believed an important
modulator of the fusion process. Simply conceived, the
curvature strain of a highly curved membrane can provide
a driving force that reduces the energetic barrier for the
fusion process. Most physiological fusion occurs between
membranes that have a relatively large radius of curvature
(e.g., synaptic vesicles, virions, cells) or are even concave
relative to the fusion site (presynaptic terminals, endosomes).
However, the local curvature at the site of fusion has been a
subject of much discussion, with several theories positing a
membrane protrusion with high local curvature at the site of
fusion [28–32]. Recent experimental evidence suggests that
synaptotagmin, one of the neuronal fusion proteins, prefer-
entially binds to curved membranes, induces tubulation of
model membranes with a diameter of 17.5 nm, and
accelerates vesicle fusion rates [33].
This important finding provides more direct evidence that
highly curved membrane structures such as we simulate here
may play an important role in physiological fusion and that
induction of curvature may be a major function of the fusion
protein apparatus. The possibility that physiological fusion
may involve highly curved membranes raises the questions of
to what degree this curvature is required for fusion and how
curvature may differentially affect the formation and stability
of fusion intermediates. Here, we report initial results in that
regard; we refer the reader to the forthcoming work by Lee
and Schick for a more extensive investigation of this
phenomenon from a field-theoretic perspective (Lee and
Schick, unpublished data).
Because the structures and detailed kinetics of fusion
intermediates are difficult to observe experimentally, com-
putational simulations of fusion play an important role in
generating detailed structural and kinetic models for fusion
that are consistent with available experimental data. Coarse-
grained simulations (such as the early rigid-amphiphile work
by Noguchi and Takasu [34]) allow a substantial increase in
computational tractability at the expense of some fidelity to
experiment; the degree of fidelity varies with the particular
approximations made in constructing the model. The lipid
model developed by Marrink and Mark [35] provides a
quantitative approximation of experimentally observable
lipid parameters such as area per head group, diffusion,
and bilayer width, and represents a good compromise
between computational tractability and level of atomistic
detail. Initial results reported for membrane fusion with this
model [11] demonstrate the fusogenicity of 15 nm vesicles
composed of either pure palmitoyloleoyl-PE (POPE) or a
dipalmitoyl-PC:dipalmitoyl-PE (DPPC:DPPE) mixture, also
observing faster fusion pore formation in the DPPC:DPPE
mixture than in pure POPE (n ¼ 6 and n ¼ 4 simulations,
respectively). Simulated pure DPPC vesicles were not ob-
served to form fusion pores.
In subsequent work, we have used this Marrink-Mark
coarse-grained model to systematically predict intermediates
and kinetics for the fusion of small POPE vesicles [36]. Our
simulations suggested a branching fusion pathway in which
vesicles first react to form a stalk-like intermediate but then
either rapidly form a fusion pore from the stalk-like state (the
‘‘direct’’ pathway) or fuse via an intermediate with an
expanded hemifusion diaphragm that is metastable on the
microsecond timescale (the ‘‘indirect’’ pathway; see Figure 1
for a structural illustration). Previous theoretical work
(reviewed in [37]) has suggested these two pathways (without
kinetic detail) as alternative hypotheses; we have predicted
that these pathways may coexist within a single system. In our
initial report, we predicted pure POPE vesicles to react via
both pathways but to primarily undergo fusion via the late-
hemifused intermediate using the indirect pathway. We now
extend this model to address how more complex mixtures of
lipids may affect the reaction mechanism and kinetics of
fusion.
Physiologic membrane fusion takes place in membranes
consisting of a complex [3,16–18] and likely inhomogeneous
[38–42] mixture of multiple lipid species and proteins.
Experimental and computational model systems for fusion
are far simpler, containing well-defined lipid mixtures; in the
case of computational studies, single-component membranes
are routinely used [10,12,43]. Given this gap in complexity
between model systems used for mechanistic studies of fusion
and the physiologic situation they are meant to reproduce, we
wish to understand how fusion mechanisms and properties
depend on membrane composition. We have therefore
performed thousands of simulations of vesicle fusion at
different lipid compositions to systematically predict how
membrane fusion kinetics and intermediates depend on lipid
composition. This molecular-dynamics approach yields sim-
ilar results to those recently obtained using self-consistent
field theory methods [44]. Our molecular-dynamics analyses,
the field-theoretic results of Schick and coworkers, and prior
continuum-mechanics work by Kozlov expand computational
models to more complex membrane compositions to better
approximate experimental models and physiologic scenarios.
Results
To predict the effect of lipid composition on membrane
fusion mechanisms, we have performed massively distributed
PLoS Computational Biology | www.ploscompbiol.org November 2007 | Volume 3 | Issue 11 | e2202229
Author Summary
Membrane fusion is the transport process used for neurotransmitter
release, insulin secretion, and infection by enveloped viruses. The
precise mechanism of fusion is not yet understood, nor is the means
by which membrane properties such as composition and curvature
affect the fusion process. Here, we use molecular-dynamics
simulations of lipid vesicle fusion under different lipid compositions
to generate a more detailed explanation for how composition
controls membrane fusion. We predict that lipid composition affects
both the initial process of forming a contact ‘‘stalk’’ between two
vesicles and the formation of a metastable ‘‘hemifused’’ intermedi-
ate. These two roles act in concert to change both the rate of fusion
and the level of detectable fusion intermediates. We also present
initial results on fusion of vesicles at different membrane curvatures.
Recent experimental results suggest that the creation of highly
curved membranes is important to fusion of synaptic vesicles. Our
simulations cover a curvature regime similar to these experimental
systems. In combination with previous results, we predict that the
effect of lipid composition on fusion is general across different
membrane curvatures, but that the rate of fusion is controlled by
both composition and curvature.
Composition Effects on Membrane Fusion

Page 3

molecular-dynamics simulations of vesicle fusion at different
lipid molar ratios of palmitoyloleoyl-PC (POPC) and POPE.
Starting conformations for the 15 nm vesicles used are
displayed in Figure 2. The resulting dataset contains ?1,000
separate simulations of vesicle fusion, up to 1 ms in length, at
each lipid composition. We have used this dataset to
construct kinetic models for fusion and to assess how the
kinetics, mechanisms, and intermediates of fusion vary with
increasing PC:PE ratio. We show that the same basic pathways
underlie fusion reactions in our simulations at all composi-
tions studied. However, our simulation data also suggest that
the relative frequency of reaction pathways and relative
stability of reaction intermediates vary substantially. In-
creased PC content in vesicles is associated with increased
utilization of the direct fusion pathway, with the result that
different lipid compositions have different preferred reac-
tion pathways for fusion. We also report results on how
membrane curvature affects rates of fusion stalk formation in
our simulations; while fusion proceeds more slowly at lesser
curvatures, the effect of lipid composition appears to operate
independently of vesicle curvature.
Kinetics and Mechanism of Fusion at Different Lipid
Compositions
We use Markovian state models (MSMs) to analyze the
kinetics and mechanism of fusion in a systematic manner.
Described in detail in previous work on both protein folding
and membrane fusion [36,45–48], MSMs provide a means for
systematically analyzing large numbers of molecular-dynam-
ics trajectories to yield quantitative predictions of membrane
fusion mechanisms. By constructing a series of MSMs at
different lipid compositions, we are able to predict the long-
timescale kinetics for vesicle fusion at each lipid composition
(Figure 3). We use measurements of lipid and contents mixing
between vesicles to characterize each structural snapshot as
unfused, stalk-like (outer leaflets fusing with a small contact
area), hemifused (outer leaflets fusing with a large contact
area), or fully fused (Figure 1A; for details, see [36]). We have
recently introduced a topological metric for assessing fusion
intermediate structures: persistent voids [49]. For the fusion
of 15 nm vesicles, the width of the hemifusion ‘‘neck’’
between the outer leaflets of each vesicle correlates relatively
well with the extent of lipid mixing. Model-building and
analysis of molecular-dynamics data were performed both
independently for each lipid composition and in a conjoint
manner (see Methods), yielding similar rates and mechanisms
via each method (see Figure S1 for a detailed comparison).
The kinetic profiles for fusion predicted via MSM analysis
(Figure 3) show a number of differences in the kinetics of
fusion intermediates as vesicle composition varies from pure
POPE through increasing concentrations of POPC. When the
vesicle composition is changed from pure POPE to 1:1
POPE:POPC, more closely approximating the average phys-
iologic ratio in cell membranes, the unfused state decays
approximately 5-fold more slowly (t1/2 of 36 ns for POPE
versus 184 ns for POPC). However, the fused state is formed
almost an order of magnitude more quickly (t1/2 of 4,212
versus 348 ns, respectively). This difference results in part
from altered usage of the stable late-hemifused intermediate.
Figure 1. Reaction Rates and Mechanisms for Vesicle Fusion
(A) Representative structures from unfused vesicles and each major fusion intermediate: unfused vesicles, a stalk-like state, a hemifused intermediate,
and fully fused vesicles. Vesicles fuse either via a direct pathway from stalk-like to fully fused (i) or via an indirect pathway (ii) using a metastable
hemifused intermediate. Surfaces rendered in gray represent solvent-accessible area; red and green spheres denote the inner-leaflet phosphate groups
of each vesicle respectively.
(B–D) Reaction rates in s?1determined for each reaction pathway at each lipid composition. (B) Rate for pure POPE. (C) Rate for for 1:1 POPC:POPE. (D)
Rate for for 2:1 POPC:POPE. Hemifused 1 and Hemifused 2 denote structurally similar late-hemifused states; kinetic differentiation of these states was
determined by kinetic consistency analysis (see Methods). Hemifusion occupies a sufficiently large area of state space that inverconversion between
some hemifused microstates is slow, resulting in kinetic differentiation.
doi:10.1371/journal.pcbi.0030220.g001
PLoS Computational Biology | www.ploscompbiol.org November 2007 | Volume 3 | Issue 11 | e220 2230
Composition Effects on Membrane Fusion

Page 4

When the fraction of POPC is increased to 67% (2:1
POPC:POPE), this trend continues further. The unfused state
decays with a t1/2of 4.8 ls, and the timescale for fusion is
dominated by the slow reaction of the unfused vesicles. We
predict a decrease in fusogenicity from a 1:1 PC:PE ratio to a
2:1 PC:PE ratio (rates given below; see Table S2 for an
energetic analysis) that is in qualitative agreement with
experimental studies of PEG-induced vesicle fusion [7]. The
overall rate of formation for fused vesicles containing 2:1
POPC:POPE approximates that of pure POPE (t1/24.2 versus
4.8 ls using a single-exponential approximation), although
the mechanism of fusion differs substantially between these
compositions in our simulations.
Much of this mechanistic difference derives from energetic
changes in the late-hemifused pathway. Considered as an
ensemble, vesicle fusion reactions at higher POPC:POPE
ratios have a lower concentration of late-hemifused inter-
mediates than do fusion reactions of pure POPE vesicles. This
results from both a decreased usage of the hemifused pathway
(Figure 1A, pathway ii; Figure 4A), leading to fewer hemifused
intermediates formed, and decreased stability of the hemi-
fused state (Figure 4B). Although 95% of pure POPE vesicle
fusion trajectories form hemifused intermediates, only 19%
of trajectories containing a 1:1 POPC:POPE mixture and 11%
of vesicles containing a 2:1 POPC:POPE mixture form
hemifused intermediates. The remainder of fusion trajecto-
ries transition rapidly from stalk-like states to fused states
without isolated expansion of a hemifusion diaphragm.
Our previous analyses have shown that such rapid fusion
from the stalk-like state involves concerted conformational
change in both inner and outer vesicle leaflets. Using our
MSM analysis of fusion trajectories, we predict t1/2values of
5,320, 424, and 123 ns for the decay of the hemifused state in
pure POPE, 1:1 POPC:POPE, and 2:1 POPC:POPE vesicles,
respectively (decay plots shown in Figure 4B). We postulate
that the decreased usage of the hemifusion pathway and
decreased stability of hemifused vesicles results from in-
creased free energy of the hemifused state relative to the
fused state.
We have also computed detailed predictions of rates and
mechanism at each lipid composition using our MSM analysis
(Figure 1 and Table 1). These data show that rates of stalk
formation decrease with increasing POPC concentration in
the simulation (p , 0.001, pairwise Kolmogorov-Smirnov
tests), in accordance with previous theoretical predictions
Figure 2. Starting Structures for Vesicles at Three Different Lipid
Compositions
(A) Pure POPE vesicles.
(B) Vesicles composed of a 1:1 POPC:POPE mixture.
(C) Vesicles composed of a 2:1 POPC:POPE mixture.
POPE lipids are rendered in pink, POPC in purple. Rendered surfaces
represent the solvent-accessible surface of the vesicle; spheres represent
the phosphate groups of the vesicle inner leaflet lipids. Vesicles are
linked by a single crosslinker molecule.
doi:10.1371/journal.pcbi.0030220.g002
Figure 3. Kinetics of Vesicle Fusion Simulated at Different Lipid Compositions
The relative concentrations of fusion intermediates and fused vesicles are plotted for each time point of the simulation for (A) pure POPE, (B) a 1:1
POPC:POPE mixture, and (C) a 2:1 POPC:POPE mixture. Predictions are derived from MSM analysis of all fusion trajectories. Dotted lines represent 90%
CIs.
doi:10.1371/journal.pcbi.0030220.g003
PLoS Computational Biology | www.ploscompbiol.org November 2007 | Volume 3 | Issue 11 | e2202231
Composition Effects on Membrane Fusion

Page 5

[10,25]. The reaction rate from stalk-like to metastable
hemifused states decreases with increasing POPC concen-
tration (p , 0.001), and the rate of fusion pore formation by
hemifused states increases significantly with the introduction
of POPC (p , 0.001). The rate of direct fusion pore formation
from the stalk-like state also increases significantly (p ,
0.001).
Analysis of Transition State Energies
Using classical transition state theory, one can interpret
these rate changes in terms of free energy of activation using a
single-barrier approximation. Such a model oversimplifies the
large coordinated rearrangements involved in fusion; none-
theless, it allows gross prediction of how transition state
energies vary with lipid composition. Relative free energy
values for each transition state at each lipid composition
are listed in Table S1. Analysis of the difference in free
energies for each transition state (Figure 5) shows significant
differences in activation free energies as a function of lipid
composition. As these data are derived from the rate
calculations, the trends are similar, but analysis in terms of
activation free energies eases comparison between reactions
and between lipid compositions.
We predict large (?1 kcal/mol) and significant DDGzvalues
for the following reactions when pure POPE vesicles are
exchanged for a 1:1 POPC:POPE mixture: formation of the
stalk-like state from unfused vesicles (Figure 5, reaction i),
formation of the fused state from the stalk-like state (Figure 5,
reaction ii), and formation of the fused state from the
hemifused state (Figure 5, reaction v). The activation energy
for formation of the hemifused state from the stalk-like state
also increases significantly (p , 0.001, Kolmogorov-Smirnov
Figure 4. POPC Content Associated with Decreased Formation and Decreased Stability of the Hemifused State
(A) The fraction of all trajectories that form a hemifused state is plotted for each vesicle composition. Vesicles with greater POPC content are less likely
to form hemifused states.
(B) The time-evolution of the hemifused state is plotted for each vesicle composition. In vesicles with greater POPC content, the hemifused state is less
stable and decays more quickly to form fully fused vesicles. Dashed lines represent 90% CIs. Because vesicles containing a 2:1 ratio of POPC:POPE rarely
form hemifused intermediates, the sampling uncertainty for the reaction of hemifused vesicles is proportionately greater.
doi:10.1371/journal.pcbi.0030220.g004
Table 1. Predicted Reaction Rates for Fusion
Reaction TypeLipid Composition
POPE
s?1
1:1 POPC:POPE
s?1
6.7 3 104
2:1 POPC:POPE
s?1
3.0 3 103
90% CI
3.2 3 105– 3.5 3 105
90% CI
6.4 3 104– 7.1 3 104
90% CI
2.0 3 103– 3.0 3 103
Forward
reactions
Unfused to stalk3.4 3 105
Stalk to fused
Stalk to Hemifused 1
Hemifused 1
to Hemifused 2
Hemifused 2 to fused
Stalk to unfused
1.9 3 104
1.6 3 105
1.8 3 104– 2.0 3 104
1.6 3 105– 1.7 3 105
1.6 3 105
5 3 104
1.5 3 105– 1.7 3 105
6.7 3 104– 8.4 3 104
2.6 3 105
3.1 3 104
2.1 3 105– 3.0 3 105
1.5 3 104– 5.4 3 104
1.5 3 105
2.0 3 103
,1.4 3 102
1.4 3 105– 1.5 3 105
2.0 3 103– 3.0 3 103
1.8 3 105
2.5 3 104
,1.6 3 10?10
1.6 3 105– 2.0 3 105
2.0 3 104– 2.9 3 104
2.4 3 105
4.9 3 104
,5.7 3 10?11
1.2 3 105– 4.3 3 105
1.4 3 104– 1.1 3 105
Reverse
reactions (s?1)
Fused to stalk
Hemifused 1 to stalk
Hemifused 2
to Hemifused 1
Fused to Hemifused 2
,3.2 3 102
2.8 3 104
3.5 3 104
,1.6 3 10?12
2.5 3 104
1.8 3 104
,1.1 3 10?10
7.2 3 104
6.6 3 104
2.6 3 104– 2.9 3 104
3.3 3 104– 3.6 3 104
1.8 3 104– 3.4 3 104
1.4 3 104– 2.3 3 104
1.3 3 104– 2.0 3 105
1.9 3 104– 1.5 3 105
,8.7 3 100
,1.1 3 10?10
,1.2 3 10?12
Predicted reaction rates for each step in the fusion pathway are listed for each vesicle composition tested. 90% confidence intervals represent sampling error. The five macrostates shown
were determined via grouping analysis of k-means clusters. For transitions that are sparsely sampled, a 95% confidence upper bound is reported instead of a calculated rate and 90%
confidence interval.
doi:10.1371/journal.pcbi.0030220.t001
PLoS Computational Biology | www.ploscompbiol.orgNovember 2007 | Volume 3 | Issue 11 | e220 2232
Composition Effects on Membrane Fusion