All-Atom Molecular Dynamics Elucidating Molecular Mechanisms of Single-Transmembrane Model Peptide Dimerization in a Lipid Bilayer

Abstract

Protein–protein interactions between transmembrane helices are essential elements for membrane protein structures and functions. To understand the effects of peptide sequences and lipid compositions on these interactions, single-molecule experiments using model systems comprising artificial peptides and membranes have been extensively performed. However, their dynamic behavior at the atomic level remains largely unclear. In this study, we applied the all-atom molecular dynamics (MD) method to simulate the interactions of single-transmembrane helical peptide dimers in membrane environments, which has previously been analyzed by single-molecule experiments. The simulations were performed with two peptides (Ala- and Leu-based artificially designed peptides, termed “host peptide”, and the host peptide added with the GXXXG motif, termed “GXXXG peptide”), two membranes (pure-POPC and POPC mixed with 30% cholesterols), and two dimer directions (parallel and antiparallel), consistent with those in the previous experiment. As a result, the MD simulations with parallel dimers reproduced the experimentally observed tendency that introducing cholesterols weakened the interactions in the GXXXG dimer and facilitated those in the host dimer. Our simulation suggested that the host dimer formed hydrogen bonds but the GXXXG dimer did not. However, some discrepancies were also observed between the experiments and simulations. Limitations in the space and time scales of simulations restrict the large-scale undulation and peristaltic motions of the membranes, resulting in differences in lateral pressure profiles. This effect could cause a discrepancy in the rotation angles of helices against the membrane normal.

Full text

All-Atom Molecular Dynamics Elucidating Molecular Mechanisms of
Single-Transmembrane Model Peptide Dimerization in a Lipid
Bilayer
Hayato Itaya, Kota Kasahara,*Qilin Xie, Yoshiaki Yano, Katsumi Matsuzaki, and Takuya Takahashi
Cite This: ACS Omega 2021, 6, 11458−11465
Read Online
ACCESS Metrics & More Article Recommendations *
sıSupporting Information
ABSTRACT: Protein−protein interactions between transmembrane helices are essential
elements for membrane protein structures and functions. To understand the effects of
peptide sequences and lipid compositions on these interactions, single-molecule
experiments using model systems comprising artificial peptides and membranes have
been extensively performed. However, their dynamic behavior at the atomic level remains
largely unclear. In this study, we applied the all-atom molecular dynamics (MD) method
to simulate the interactions of single-transmembrane helical peptide dimers in membrane
environments, which has previously been analyzed by single-molecule experiments. The
simulations were performed with two peptides (Ala- and Leu-based artificially designed peptides, termed “host peptide”, and the host
peptide added with the GXXXG motif, termed “GXXXG peptide”), two membranes (pure-POPC and POPC mixed with 30%
cholesterols), and two dimer directions (parallel and antiparallel), consistent with those in the previous experiment. As a result, the
MD simulations with parallel dimers reproduced the experimentally observed tendency that introducing cholesterols weakened the
interactions in the GXXXG dimer and facilitated those in the host dimer. Our simulation suggested that the host dimer formed
hydrogen bonds but the GXXXG dimer did not. However, some discrepancies were also observed between the experiments and
simulations. Limitations in the space and time scales of simulations restrict the large-scale undulation and peristaltic motions of the
membranes, resulting in differences in lateral pressure profiles. This effect could cause a discrepancy in the rotation angles of helices
against the membrane normal.
1. INTRODUCTION
Many proteins embedded in biological membranes perform
essential functions, such as mediating communications across
the membrane and energy synthesis. Elucidating the molecular
principles of membrane proteins has gained much attention in
the context of molecular biology. In particular, the formation of
multimeric complexes in the membrane environment plays an
important role to establish the molecular functions of
membrane proteins. The characteristics of the protein−protein
interactions of membrane proteins are distinct from those of
soluble proteins because of the differences in their environ-
ments.
1−3
For example, the low permittivity of the membrane
environment significantly strengthens the electrostatic inter-
actions compared to the solution environment. Therefore,
both the two factors (i) the sequence of the membrane
proteins and (ii) the lipid composition of the membrane are
essential for protein−protein interactions.
Typically, the protein−protein interactions between mem-
brane proteins are established by the packing of trans-
membrane helices. The sequences of the binding interface in
transmembrane helices have been well characterized; small
residues, i.e., Gly, Ala, and Ser, are enriched in the interfaces
and tend to be distributed at intervals of two or three
residues.
4,5
These kinds of sequence elements are known as
motifs, e.g., GXXXG, SXXXG, and GXXXGXXG. The small
steric hindrance of these side chains brings the helices closer to
each other. In the GXXXG motif, an important example of the
transmembrane helix binding motifs, the weak hydrogen bond
between the Cα−H and O moieties of Gly residues is
considered to be a key player in the binding.
6−10
Another key feature is the lipid composition of the
membrane. The structures and functions of membrane
proteins are influenced by lipid composition. In particular,
the effects of cholesterol have been well studied. There is much
evidence for cholesterol-induced modulations of conforma-
tional states, multimer formation, ligand-binding activity, and
ion channel activity.
11,12
To investigate the molecular mechanisms of transmembrane
helix binding and the effects of the sequence motif and the
membrane composition on their binding, a dimer of single-
transmembrane helices can be considered one of the simplest
models. Dimerization mechanisms have been extensively
studied by targeting stereotypical single-transmembrane
Received: January 26, 2021
Accepted: April 8, 2021
Published: April 22, 2021
Article
http://pubs.acs.org/journal/acsodf
© 2021 The Authors. Published by
American Chemical Society 11458
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
Downloaded via 179.61.177.49 on May 5, 2021 at 08:49:13 (UTC).
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

peptides, e.g., glycophorin A
6
and growth hormone recep-
tors.
13,14
Yano et al.
15−18
reported in vitro binding experiments using
single-pair Förster resonance energy transfer (FRET) measure-
ments. These studies analyzed the behavior of model peptides
composing a single-transmembrane helix, the sequence of
which is a triple repeat of AALALAA and its variant with a
GXXXG motif. The single-pair FRET experiments compared
the binding kinetics of these peptides in the two types of
membrane environments: pure-POPC membrane and a POPC
membrane mixed with 30%-cholesterol. Although cholesterols
enhanced the binding of the “host”peptide without the
GXXXG motif, they inhibited the binding of the GXXXG
peptide.
16,17
The molecular mechanisms of this phenomenon are not fully
understood at the atomic level. The atomic details of
interactions, such as those in highly complicated, heteroge-
neous, and dynamical molecular systems, cannot be easily
observed through current experimental techniques. A promis-
ing method to observe such details is the molecular dynamics
(MD) method, which simulates atomic motions of molecular
systems based on Newtonian mechanics. Although MD
simulations have been extensively applied to investigate the
dynamic behavior of membranes and membrane proteins,
19−27
it remains challenging to treat membrane systems. In
particular, the success of MD simulations is based on the
data from the X-ray crystal structure analyses of proteins.
Current MD methods can be efficiently employed for proteins
with well-packed, stable folds; however, there are many
challenges for their use with unstable proteins with structures
that are difficult to be analyzed by experiments, e.g., dynamic
features of dimerized transmembrane helices
19
and intrinsically
disordered proteins.
20
Further validations of this methodology
with direct comparison with experimental observations are
required.
Here, we applied the all-atom MD method to analyze
molecular systems emulating the in vitro single-pair FRET
experiments provided by Yano et al.
16,17
to elucidate the
atomic details of transmembrane helices binding with and
without the GXXXG motif in the pure-POPC and cholesterol-
mixed POPC environments. In addition, we directly compared
the results of MD simulations with those of single-pair FRET
experiments and discussed the current limitations of MD
techniques to provide implications that may improve the
simulation method.
2. RESULTS
2.1. Simulation Overview. We analyzed molecular
systems comprising a dimer of a single-transmembrane peptide
embedded in a membrane surrounded by explicit water
molecules. For the single-transmembrane peptide, two types
of sequences were used: AALALAA-AALALAA-AALALAA
and AALALAA-AGLALGA-AALALAA. The first sequence is
the triple repeat of AALALAA. The second introduces the
GXXXG motif in the middle of the first sequence to enhance
the dimerization. In this paper, we termed these model
peptides host peptide and GXXXG peptide, respectively,
hereinafter. To eliminate the charges at the termini, the model
peptides were capped with standard capping groups in the
following manner: an acetyl group at the N-terminus and a
methyl group at the C-terminus. For the membrane, two types
of lipid bilayers were used: a pure-POPC bilayer and a bilayer
with 30% cholesterol and 70% POPC (the percentage indicates
the ratio of the number of lipid molecules). The 30%
cholesterol mimics the cholesterol composition of the plasma
membrane.
21
Additionally, the pure-POPC membrane mimics
intracellular membranes, such as the endoplasmic reticulum.
For simplicity, we termed these membrane models pure-POPC
membrane and cholesterol-mixed membrane, respectively,
hereinafter.
In total, eight systems were analyzed: the combination of
two peptide sequences (host and GXXXG peptides), two
dimer orientations (parallel and antiparallel), and two
membranes (pure-POPC and cholesterol mixed). For each
system, four runs of 500 ns NPT simulations were performed
with different initial atomic velocities. The trajectories of the
last 300 ns in each run were analyzed. The simulation model is
shown in Figure 1. The simulation conditions are summarized
in Table S1 in the Supporting Information.
2.2. Conformational Stability during the Simulation.
In all of the simulation conditions, transmembrane peptides
retained their helical conformation, except for few residues at
the termini, and no irreversible rupture of the membrane was
observed (Figures S1 and S2 in the Supporting Information).
The root-mean-square fluctuation (RMSF) showed that there
were no significant conformational changes in each peptide in
the simulations, whereas their termini tended to fluctuate
(Figure S3 in the Supporting Information). These results imply
that the initial conformations were reasonably stable during the
entire simulations.
Comparing the host and GXXXG peptides, replacing Ala
with Gly at residues 9 and 13 increased the flexibility of the
Figure 1. Initial structure of production simulations with the cholesterol-mixed membrane. The gray and cyan lipids are POPCs and cholesterols,
respectively. The blue and red ribbons indicate the peptides. The red dots are water molecules. (A) A side view. (B) A top view.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11459

peptide backbone (Figure S4 in the Supporting Information),
but the average values of the dihedral angles did not change
significantly. In addition, the N-terminal region of the host
peptide was unstable compared with that of the GXXXG
peptide (Figure S3 in the Supporting Information), although
these two peptides have the same initial structures.
Note that each trajectory may not reach equilibrium in the
500 ns time course. There are detectable differences among the
four replicates of simulations with different initial velocities
(and different initial positions of cholesterols). We primarily
focus on the differences in averaged trends over four repeated
runs rather than the details of each trajectory.
2.3. Peptide−Peptide Contacts. To assess the inter-
actions between two transmembrane peptides, the intermo-
lecular, inter-residue contacts were detected based on the
criterion that the Cα−Cαdistance is within 8.0 Å. For the
dimers with a parallel orientation, the average numbers (and
the standard errors) of the inter-residue contacts between the
GXXXG peptides were 23.5 (7.66) and 4.64 (2.53) for pure-
POPC and cholesterol-mixed membranes, respectively, and for
the host peptide, these values were 10.44 (3.96) and 8.30
(2.59), respectively (Table S1 in the Supporting Information).
The GXXXG peptide was in tighter contact than the host
peptides in the pure-POPC environment (Figure 2), reflecting
the fact that the GXXXG motif facilitates the dimer formation.
The host peptide in the pure-POPC membrane yielded a
bimodal distribution (Figure 2B, red). The first peak at zero
number of contacts indicates that the dimer was dissociated.
For the GXXXG peptide in the pure-POPC membrane, the
dimer was not dissociated, and there were peaks around 11, 26,
and 50 contacts (Figure 2D, red). The snapshots of the third
peak show tightly contacted dimer conformations that are not
observed for the host peptides (Figure 2D). In addition, it was
confirmed that adding cholesterols to the membrane clearly
reduced the interactions of the GXXXG dimers. The contact
distribution for the GXXXG peptide in the cholesterol-mixed
membrane has a strong peak at zero contact. These results
were qualitatively consistent with those of the single-pair
FRET experiments reported by Yano et al.
17
In the antiparallel configuration, cholesterol facilitated
interactions of the host peptide dimer (Figure 2A), in
agreement with the findings of Yano et al.
16
In contrast, the
results of the antiparallel GXXXG dimer disagreed with the
experimental results. Although the experiment reported that
the GXXXG helix dimerization was inhibited by addition of
cholesterols in both parallel and antiparallel configurations,
17
our simulation with the antiparallel GXXXG dimer yielded
more frequent contact in the cholesterol-mixed membrane
than in the pure-POPC membrane.
2.4. Hydrogen Bonds. The binding of the two trans-
membrane helices included several hydrogen bonds. The
hydrogen bonds were assessed using the criteria that the
acceptor−donor distance was less than or equal to 3.5 Å, and
the donor−hydrogen−acceptor angle was greater than or equal
to 120°. We analyzed the two types of hydrogen bonds:
standard backbone hydrogen bonds (N−H···O) and weak
hydrogen bonds (Cα−H···O). As a result, in the parallel
configuration, the dimer in the cholesterol-mixed membrane
exhibited fewer hydrogen bonds than those in the pure-POPC
membrane for both the peptides (Figure 3C,D). Interestingly,
the antiparallel configuration showed the opposite trend
(Figure 3A,B). This is consistent with the differences in the
number of inter-residue contacts shown in Figure 2.
Comparing the host and GXXXG peptides, a lower frequency
of hydrogen bond formation was observed in the GXXXG
peptide than in the host peptide, regardless of the frequency of
inter-residue contacts. In the parallel direction, although the
GXXXG peptide had more frequent inter-residue contacts, it
had fewer hydrogen bonds than the host peptide. This
indicates that the GXXXG motif facilitates the peptide−
peptide binding but it decreases hydrogen bonding and these
binding modes are distinct between the host and GXXXG
peptide dimers. Hydrogen bonds were not major driving forces
for the dimerization of the GXXXG peptides. The hydrogen
bonds in the host peptides were formed at the termini (Figure
S5 in the Supporting Information).
2.5. Helix Tilt Angles and Membrane Thickness. The
binding modes of peptide dimers can also be analyzed in terms
of the tilt angles and crossing angles of the two helices (Figures
Figure 2. Distributions of the number of contacts in each condition.
The horizontal axis indicates the number of inter-residue contacts
between two peptides. The vertical axis is the relative frequency in
each ensemble. The upper row (A and B) and the lower row (C and
D) show the host and GXXXG peptides, respectively. The left column
(A and C) and right column (B and D) show the antiparallel and
parallel directions of the dimer, respectively. The red and blue curves
correspond to pure-POPC and cholesterol-mixed membrane,
respectively. The bin width of the histogram is 1.
Figure 3. Average number of hydrogen bonds between the two
peptides under each condition. (A and C) and (B and D) show the
data for cholesterol-mixed and pure-POPC membranes, respectively.
(A and B) and (C and D) present the antiparallel and parallel
directions of the dimer, respectively. Cyan and orange indicate Cα−
H··· O and N−H···O hydrogen bonds, respectively. The error bars
shown as the dashed lines indicate the standard errors calculated over
the four trajectories. Those shown in solid lines represent the standard
errors calculated by performing bootstrap analysis. In this analysis, a
random sampling of a trajectory with replacement from the four
trajectories was repeated four times, and an ensemble with the same
number of snapshots as the original ensemble was generated. The
standard errors were calculated over 100 regenerated ensembles.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11460

4and S6 in the Supporting Information). The tilt angle is
defined as the angle between the membrane normal and the
vector sum of the helical axes of the two peptides. The crossing
angle is the angle between the two helical axes. In general, the
host peptides had larger angles than the GXXXG peptide, and
the pure-POPC membrane exhibited larger angles than the
cholesterol-mixed membrane.
To some extent, the effects of the membrane environment
on the angles can be explained in terms of membrane
thickness. Embedding the cholesterols into the POPC
membrane thickened the membrane, but the sequence and
direction of the peptide dimer (parallel or antiparallel) did not
affect the membrane thickness (Figure 5), which is consistent
with the findings of the previous experiment.
16
Insertion of stiff
lipids, that is, cholesterol, caused ordering of the acyl chains of
POPC and thickening of the membrane (Figure S7 in the
Supporting Information). Thinner membranes tended to have
larger transmembrane helix angles because of the hydrophobic
mismatch.
22
Tilted conformations can minimize the hydro-
phobic mismatch when the membrane is thinner than the
length of the helices.
In contrast, larger angles of the host dimer than the GXXXG
dimer were not caused by the hydrophobic mismatch because
the membrane thickness did not depend on the peptide species
or direction. As mentioned above, these two types of
transmembrane helices had different mechanisms of dimeriza-
tion; the GXXXG dimer had tight inter-residue contacts,
whereas the host dimer had some hydrogen bonds. This
difference can cause differences in the angles. To form a tight
packing of the dimer interface, there is a steric requirement to
maximize inter-residue contacts between the two helices.
Crossing angles that are too large decrease the contacts
between the terminal sides of the helices, and crossing angles
that are too small may involve Leu−Leu side chain crashes,
creating a distance between the helices. Therefore, the host
peptides that have some intermolecular hydrogen bonds allow
larger angles compared with the GXXXG peptides.
2.6. Peptide−Lipid Interactions. The effects of lipid
composition on the peptide−peptide interactions can be
categorized into two classes: direct and indirect effects. As an
example of indirect effects, the cholesterol-mixed membrane
was thicker and affected the peptide−peptide interactions by
adjusting the hydrophobic thickness. In contrast, the direct
effects are mediated by direct contact or interaction between
peptide and lipid molecules. To assess the direct effects, we
analyzed the duration times for a peptide−lipid interaction,
which is defined by the following criterion: the minimum
distance between the Cαof peptides and the representative
head group atom of a lipid molecule (P for POPC and O for
cholesterol) is less than or equal to 10 Å. The time from
peptide−lipid association to their dissociation was measured
for every association/dissociation event. As a result, although
there were detectable differences in the duration times between
POPC and cholesterol, the difference was not large (Figure 6).
3. DISCUSSIONS
3.1. Convergence of Simulation Trajectories. In this
study, four repeats of 500 ns simulations were performed for
each condition. We confirmed that the peptide conformation
was stably maintained for 500 ns in all of the simulations
(Figure S1 in the Supporting Information). However, the
interactions between the two peptides diverged during the
simulation (Figure S8 in the Supporting Information). The
distribution of the number of interpeptide contacts signifi-
cantly differed among the four trajectories under the same
condition (Figure S9 in the Supporting Information). In
general, because of the complexity of the membrane protein
system, an equilibrium state cannot be easily achieved in a
single run of canonical MD simulations.
23
However, a comparison of ensemble averages over the last
200 ns of the four trajectories yielded remarkable differences
among the different conditions. While the distribution of
interpeptide contacts in each 200 ns time window shows
detectable differences between 200−400 and 300−500 ns
(Figure S10 in the Supporting Information), the differences
among the conditions were qualitatively conserved. We
Figure 4. Distributions of tilt (A−D) and crossing (E−H) angles. Red
and blue lines indicate results of pure-POPC and cholesterol-mixed
membranes, respectively. The bin width of the histogram is 0.1. The
panels (A, B, E, and F) and (C, D, G, and H) show the host and
GXXXG peptides, respectively. The left column (A, C, E, and G) and
the right column (B, D, F, and H) show the antiparallel and parallel
directions of the dimer, respectively.
Figure 5. Distributions of the membrane thickness. Red and blue lines
indicate the results of pure-POPC and cholesterol-mixed membranes,
respectively. The membrane thickness was measured as the distance
between the centroids of carbonyl oxygen atoms in each leaflet. The
upper row (A and B) and the lower row (C and D) show the host and
GXXXG peptides, respectively. The left column (A and C) and the
right column (B and D) show the antiparallel and parallel directions
of the dimer, respectively. The bin width of the histogram is 0.1.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11461

assessed errors for the average number of interpeptide contacts
(Figure S11 in the Supporting Information) and hydrogen
bonds (Figure 3) in the following two ways: the standard
errors calculated over the four samples of trajectory and those
calculated with bootstrap analysis. The values showed marked
differences among the conditions of the lipid composition,
sequence motif, and dimer orientation.
3.2. Facilitation of Dimer Formation by the GXXXG
Motif. The GXXXG motif is a well-known sequence element
that enhances the homodimerization of transmembrane
helices.
24
As a prominent example, glycophorin A contains
the GXXXG motif, and it has been reported that the three-
dimensional structure of its dimer forms an hourglass-shaped
dimer (a helix−helix crossing angle of ∼40°) with Cα−H weak
hydrogen bonds between Gly residues.
6−8
In contrast, our
simulation with the GXXXG motif revealed a tightly packed
dimer with a small crossing angle (Figure 4), in agreement with
the findings of the single-pair FRET experiments.
17
In
addition, only a few hydrogen bonds were observed in the
GXXXG dimer (Figure 3). This difference in the mechanisms
of dimerization by the GXXXG motif is likely caused by the
peptide sequences. Statistically, β-branched residues such as
Val and Ile are enriched at the neighboring position of Gly of
the GXXXG motif, like glycophorin A
25
and ErbB growth
factor receptors. Our model peptides have two Leu residues,
which have a γ-branched side chain with a large variety of
rotamers, between two Gly residues. Entropy loss involved in
the binding of these Leu residues can impact the binding
mechanism of the GXXXG dimer. In addition, our model
peptides consist of only three types of residues, Gly, Ala, and
Leu, and the simple pattern of this sequence makes it possible
to form tight packing of the entire length of the two helices.
Introducing the Gly residue into the host peptide increases the
flexibility of the helical backbone (Figure S4)
26
and allows
dimer packing to be optimized.
3.3. Effects of Cholesterols in the Membrane on the
Dimerization of the Host Peptides. For the host peptide,
the single-pair FRET experiments by Yano et al.
16
showed that
the dimer formation was observed in the antiparallel dimer in
the cholesterol-mixed membrane, and no FRET signals were
observed in the other three conditions, namely, the antiparallel
dimer in the pure-POPC and the parallel dimer in both the
membrane conditions. This indicates that cholesterol enhances
the dimerization of the host helix, but the dimer cannot be
formed in the parallel configuration. Our previous in vitro
studies showed that the helix macrodipole is an essential factor
for antiparallel dimerization of the host peptides.
15,16,18
The
association enthalpy of the host helix antiparallel dimer
coincided with the estimated energy of the helix macrodi-
pole.
18
In general, whereas the helix macrodipole effect is
strongly shielded in high permittivity environments, burying
the helix termini into the membrane environment reduces the
shielding effects of the solvent. A theoretical study by Sengupta
et al.
27
reported that shorter helices yield stronger helix
macrodipole effects. Our simulations reproduced an increase in
peptide−peptide interactions in the antiparallel dimers by
introducing cholesterol (Figure 2A). Although a certain
amount of interpeptide contacts were observed in the parallel
dimers (Figure 2B), they were clearly weaker than those of the
antiparallel dimer in the cholesterol-mixed membrane.
However, there is a discrepancy between the experiments
and simulations regarding the tendencies of the helix tilt and
crossing angles. The experiments indicated that the helix
orientation angles were almost vertical (∼0°) in the pure-
POPC and ∼28°in the cholesterol-mixed membrane.
16
Note
that the orientation angles measured in the experiments are
calculated from the amide I bonds, and the helix tilt angles
against the membrane normal and crossing angles cannot be
distinguished. Our simulations yielded more tilted dimer
conformations in the pure-POPC membrane (crossing angles
of ∼20°in the cholesterol-mixed membrane and ∼29°in the
pure-POPC membrane for the crossing angles). According to
the results of helix orientation angles, Yano et al. concluded
that the host helices form an hourglass-shaped dimer with the
penetration of water molecules into the membrane regions to
release the lateral pressure in the cholesterol-mixed membrane.
Our MD simulation may not precisely reproduce the macro
properties of the membrane environment, including the lateral
pressure, owing to the limited size of the simulation cell, and
thus the dimer did not form the hourglass shape. Instead, the
dimer was close to a vertical orientation to reduce the
hydrophobic mismatch, given that the membrane was
thickened by the addition of cholesterols. An increase in the
interpeptide contacts was driven by strengthening helix
macrodipole interactions by thickening the membrane.
As discussed in the study by Yano et al.,
16
there are three
driving forces for the dimerization in the cholesterol-mixed
membrane: (i) release of the lateral pressure by dimerization,
(ii) lipophobic interactions, and (iii) strengthening of the helix
macrodipole interactions. Our simulations may not adequately
account for effect (i). To assess effect (ii), we analyzed
duration times for the events in which a lipid molecule
interacts with the peptide dimer (Figure 6) and found that
there was no notable difference in the kinetics of lipid−peptide
interactions between the pure-POPC and cholesterol-mixed
membranes. This result is not paradoxical to that of the
experiments by Yano et al. Effect (iii) facilitated the dimer
formation in the simulations.
3.4. Effects of Cholesterols in the Membrane on the
Dimerization of the GXXXG Peptides. In the models of the
GXXXG parallel dimers, our simulations showed that
introducing cholesterols reduced the interpeptide contacts,
Figure 6. Distribution of duration times for the event beginning with
a lipid molecule encountering the peptide and ending with the
dissociation. The horizontal axis means the duration time, and the
vertical axis means the observed number of events. Solid and dashed
lines indicate the results from the simulations with pure-POPC and
cholesterol-mixed membranes, respectively. Color specifies species of
lipid molecules: orange for POPC and cyan for cholesterol. The upper
row (A and B) and the lower row (C and D) show the host and
GXXXG peptides, respectively. The left column (A and C) and the
right column (B and D) show the antiparallel and parallel directions
of the dimer, respectively.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11462

which agrees with the results of the single-pair FRET
experiments.
17
Because the cholesterols thickened the
membrane (Figure 5), the tilt and crossing angles decreased
to minimize the hydrophobic mismatch (Figure 4). Tighter
packing of the two helices sterically requires a certain crossing
angle, and thus a decrease in the crossing angle weakens the
peptide−peptide packing.
In contrast, the behavior of the GXXXG antiparallel dimer
did not agree with the experimental results, which reported
that cholesterols inhibit the dimer formation of the GXXXG
peptides. Our simulations showed that the GXXXG helices had
a certain amount of interpeptide contacts, even in the
cholesterol-mixed membrane (Figure 2C). The introduction
of cholesterols increased the membrane thickness (Figure 4)
and decreased the angles (Figure 3), but interactions were
retained, in contrast to the parallel dimer. This discrepancy
may originate from the differences in the treatment of lateral
pressure. Yano et al.
17
discussed that it is difficult to release
lateral pressure for the GXXXG peptides because of the
flexibility of the Gly backbone and this effect can inhibit
dimerization by cholesterol. Our simulation may not
appropriately reproduce the macro properties of the
membrane.
4. CONCLUSIONS
We investigated the molecular mechanisms of dimer formation
in single-transmembrane model peptides using all-atom MD
simulations. To compare our simulations with the in vitro
experiments reported by Yano et al.
16,17
directly, we performed
simulations using a combination of the various conditions: two
types of peptide sequences (the host and GXXXG peptides),
two types of dimer topology (parallel and antiparallel), and
two types of membrane composition (the pure-POPC and
cholesterol-mixed membranes). In the same conditions, the
simulations reproduced the trends observed in the experi-
ments. Introducing cholesterols facilitated the dimer formation
of the host peptides and inhibited the formation of the
GXXXG dimer in the parallel configuration.
In contrast, there are some discrepancies between the MD
simulations and the in vitro experiments by Yano et al.
16,17
Several features that are difficult to include adequately in MD
simulations. (i) Owing to the limitation of the simulation cell
size and applying the periodic boundary condition, the
macroscopic mechanical properties of the membrane may
not be reproduced by MD simulations. The wavelength of the
undulation and peristaltic motion of the membrane is limited
up to the cell dimension along the lateral direction.
28,29
To
discuss the impacts of such effects, further evaluations using
molecular models with different cell sizes should be performed
elsewhere. (ii) The simulations may not capture equilibrium
properties because of the limitation of the time scale.
23
To
reduce the bias from the initial condition, we repeated four
simulation runs with different initial atomic velocities (and
different positions of cholesterols) for each system. To
improve the reliability of sampling for equilibrium states,
generalized ensemble methods represent a promising ap-
proach. Sampling the canonical ensemble by exploring the
conformational space and by conducting analyses of the free-
energy landscape provide insights into the binding mode of
dimer formation. Additionally, information on the interaction
energy for dimerization can be dissected based on the
conformational ensemble including both the associated and
dissociated states. However, the application of generalized
ensemble approaches to membrane systems is not necessarily
straightforward.
19,30,31
Further developments of the method
are necessary.
The simulation study dissects the physical features that are
intrinsically inseparable in experiments, namely, local molec-
ular interactions and macroscopic mechanical features of the
membrane. For example, the result of the dimer formation of
the antiparallel GXXXG helices being inhibited by cholesterols
could not be reproduced by our simulations, implying that the
effects omitted in the simulations would be important for this
phenomenon. In contrast, the local interactions treated in the
simulations may be sufficient to explain the cholesterol-
induced inhibition of the parallel dimers. There are essential
differences between parallel and antiparallel dimers.
5. METHODS
5.1. Simulation System. For the initial structures of MD
simulations, two types of dimer conformations, which are
parallel and antiparallel dimers, were built by the template-
based modeling based on the bacteriorhodopsin structure
(PDB ID: 2ntu) using MODELLER software.
32
The template
was arbitrarily selected as a typical structure of the trans-
membrane helix bundle without any specific motif. The
template of the parallel dimer was the first helix (residues 8−
28) and the seventh helix (residues 202−222), whereas that of
the antiparallel dimer was the sixth helix (residues 168−188)
and the seventh helix (residues 202−222). The dimer models
were embedded in a lipid bilayer and bathed in a 0.15 M NaCl
solution using CHARMM-GUI.
33
The membrane comprised
256 lipid molecules (128 lipids for each leaflet). The cell
dimensions were approximately 90 Å ×90 Å ×70 Å in the
initial structure.
In total, eight systems were analyzed: the combination of
two peptide sequences, two dimer orientations, and two
membranes. These models were prepared to emulate the
experiments by Yano et al.
16,17
For each of the eight models,
we ran four series of simulations independently. The initial
configuration of the cholesterol molecules and initial atomic
velocities were generated by different random seeds for each
run. The simulation conditions are summarized in Table S1.
5.2. Molecular Simulation Methods. The equilibration
protocol used the default setting of CHARMM-GUI. First, the
energy minimization was performed using the steepest descent
method. Second, the NPT simulations with a Berendsen
barostat were performed with the positional restraint; the force
constant of which was gradually relaxed. Subsequently, a 500
ps NPT simulation was performed without restraint. The
integration time step for the equilibration was 2.0 fs, and the
covalent bonds with hydrogen atoms were constrained by the
LINCS method.
34,35
Finally, 500 ns NPT simulations were
performed at 298 K and 1.0 atm using the Nosé
−Hoover
thermostat and the Parrinello−Rahman barostat with a 2.0 fs
integration time step. In total, 16 μs simulations were
performed. The trajectory for the last 300 ns of each run
was analyzed. The CHARMM36m force field
36
and the TIP3P
water model
37
were applied. Electrostatic potentials were
calculated using the smooth particle mesh Ewald method.
38
The real-space cutofflength was 1.0 nm. All of the simulations
were performed using GROMACS software.
39
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11463

■ASSOCIATED CONTENT
*
sıSupporting Information
The Supporting Information is available free of charge at
https://pubs.acs.org/doi/10.1021/acsomega.1c00482.
Time evolution of helix content of each run (Figure S1);
examples of snapshots at the final steps (Figure S2);
RMSF and helix content of each residue (Figure S3);
backbone dihedral angles of the 9th and 13th residues
(Figure S4); residue-wise frequency for hydrogen bond
formation (Figure S5); distributions of helix tilt angles
(Figure S6); order parameters of POPC (Figure S7);
time evolution of the number of interpeptide contacts
(Figure S8); distribution of the number of interpeptide
contacts in each replicate of simulations (Figure S9);
distribution of the number of interpeptide contacts in
each 200 ns time window (Figure S10); average and
errors for the number of inter-residue contacts (Figure
S11); and summary of simulation systems (Table S1)
(PDF)
■AUTHOR INFORMATION
Corresponding Author
Kota Kasahara −College of Life Sciences, Ritsumeikan
University, Kusatsu, Shiga 525-8577, Japan; orcid.org/
0000-0003-0207-6271; Email: ktkshr@fc.ritsumei.ac.jp
Authors
Hayato Itaya −Graduate School of Life Sciences, Ritsumeikan
University, Kusatsu, Shiga 525-8577, Japan
Qilin Xie −College of Pharmaceutical Sciences, Ritsumeikan
University, Kusatsu, Shiga 525-8577, Japan
Yoshiaki Yano −Graduate School of Pharmaceutical Sciences,
Kyoto University, Kyoto 606-8501, Japan
Katsumi Matsuzaki −Graduate School of Pharmaceutical
Sciences, Kyoto University, Kyoto 606-8501, Japan;
orcid.org/0000-0002-0182-1690
Takuya Takahashi −College of Life Sciences, Ritsumeikan
University, Kusatsu, Shiga 525-8577, Japan
Complete contact information is available at:
https://pubs.acs.org/10.1021/acsomega.1c00482
Author Contributions
The manuscript was written through contributions of all
authors. All authors have given approval to the final version of
the manuscript.
Funding
K.K. was supported by JSPS KAKENHI Grant Number
JP20K12069.
Notes
The authors declare no competing financial interest.
■ACKNOWLEDGMENTS
The computational resources were provided by the HPCI
System Research Project (Project IDs: hp190017, hp190018,
hp200063, and hp200090), the NIG supercomputer at ROIS
National Institute of Genetics, Human Genome Center (the
University of Tokyo), and Research Center for Computational
Science, Okazaki, Japan.
■REFERENCES
(1) Lee, A. G. How lipids affect the activities of integral membrane
proteins. Biochim. Biophys. Acta, Biomembr. 2004,1666,62−87.
(2) Marsh, D. Protein modulation of lipids, and vice-versa, in
membranes. Biochim. Biophys. Acta, Biomembr. 2008,1778, 1545−
1575.
(3) Phillips, R.; Ursell, T.; Wiggins, P.; Sens, P. Emerging roles for
lipids in shaping membrane-protein function. Nature 2009,459, 379−
385.
(4) Russ, W. P.; Engelman, D. M. The GxxxG motif: A framework
for transmembrane helix-helix association. J. Mol. Biol. 2000,296,
911−919.
(5) Feng, X.; Barth, P. A topological and conformational stability
alphabet for multipass membrane proteins. Nat. Chem. Biol. 2016,12,
167−173.
(6) Smith, S. O.; Song, D.; Shekar, S.; Groesbeek, M.; Ziliox, M.;
Aimoto, S. Structure of the Transmembrane Dimer Interface of
Glycophorin A in Membrane Bilayers†.Biochemistry 2001,40, 6553−
6558.
(7) Senes, A.; Ubarretxena-Belandia, I.; Engelman, D. M. The Cα
H···O hydrogen bond: A determinant of stability and specificity in
transmembrane helix interactions. Proc. Natl. Acad. Sci. U.S.A. 2001,
98, 9056−9061.
(8) Arbely, E.; Arkin, I. T. Experimental Measurement of the
Strength of a Cα−H···O Bond in a Lipid Bilayer. J. Am. Chem. Soc.
2004,126, 5362−5363.
(9) Mueller, B. K.; Subramaniam, S.; Senes, A. A frequent, GxxxG-
mediated, transmembrane association motif is optimized for the
formation of interhelical Cα−H hydrogen bonds. Proc. Natl. Acad. Sci.
U.S.A. 2014,111, E888−E895.
(10) Anderson, S. M.; Mueller, B. K.; Lange, E. J.; Senes, A.
Combination of Cα−H Hydrogen Bonds and van der Waals Packing
Modulates the Stability of GxxxG-Mediated Dimers in Membranes. J.
Am. Chem. Soc. 2017,139, 15774−15783.
(11) Baenziger, J. E.; Morris, M. L.; Darsaut, T. E.; Ryan, S. E. Effect
of membrane lipid composition on the conformational equilibria of
the nicotinic acetylcholine receptor. J. Biol. Chem. 2000,275, 777−
784.
(12) Corradi, V.; Sejdiu, B. I.; Mesa-Galloso, H.; Abdizadeh, H.;
Noskov, S. Y.; Marrink, S. J.; Tieleman, D. P. Emerging Diversity in
Lipid−Protein Interactions. Chem. Rev. 2019,119, 5775−5848.
(13) Prakash, A.; Janosi, L.; Doxastakis, M. GxxxG Motifs,
Phenylalanine, and Cholesterol Guide the Self-Association of
Transmembrane Domains of ErbB2 Receptors. Biophys. J. 2011,
101, 1949−1958.
(14) Brooks, A. J.; Dai, W.; O’Mara, M. L.; Abankwa, D.; Chhabra,
Y.; Pelekanos, R. A.; Gardon, O.; Tunny, K. A.; Blucher, K. M.;
Morton, C. J.; et al. Mechanism of Activation of Protein Kinase JAK2
by the Growth Hormone Receptor. Science 2014,344, No. 1249783.
(15) Yano, Y.; Matsuzaki, K. Measurement of Thermodynamic
Parameters for Hydrophobic Mismatch 1: Self-Association of a
Transmembrane Helix. Biochemistry 2006,45, 3370−3378.
(16) Yano, Y.; Kondo, K.; Kitani, R.; Yamamoto, A.; Matsuzaki, K.
Cholesterol-Induced Lipophobic Interaction between Transmem-
brane Helices Using Ensemble and Single-Molecule Fluorescence
Resonance Energy Transfer. Biochemistry 2015,54, 1371−1379.
(17) Yano, Y.; Kondo, K.; Watanabe, Y.; Zhang, T. O.; Ho, J.-J.;
Oishi, S.; Fujii, N.; Zanni, M. T.; Matsuzaki, K. GXXXG-Mediated
Parallel and Antiparallel Dimerization of Transmembrane Helices and
Its Inhibition by Cholesterol: Single-Pair FRET and 2D IR Studies.
Angew. Chem. 2017,129, 1782−1785.
(18) Yano, Y.; Watanabe, Y.; Matsuzaki, K. Thermodynamic and
kinetic stabilities of transmembrane helix bundles as revealed by
single-pair FRET analysis: Effects of the number of membrane-
spanning segments and cholesterol. Biochim. Biophys. Acta, Biomembr.
2021,1863, No. 183532.
(19) Kharche, S. A.; Sengupta, D. Dynamic protein interfaces and
conformational landscapes of membrane protein complexes. Curr.
Opin. Struct. Biol. 2020,61, 191−197.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11464

(20) Kasahara, K.; Terazawa, H.; Takahashi, T.; Higo, J. Studies on
Molecular Dynamics of Intrinsically Disordered Proteins and Their
Fuzzy Complexes: A Mini-Review. Comput. Struct. Biotechnol. J. 2019,
17, 712−720.
(21) Song, Y.; Kenworthy, A. K.; Sanders, C. R. Cholesterol as a co-
solvent and a ligand for membrane proteins. Protein Sci. 2014,23,1−
22.
(22) Ren, J.; Lew, S.; Wang, Z.; London, E. Transmembrane
orientation of hydrophobic alpha-helices is regulated both by the
relationship of helix length to bilayer thickness and by the cholesterol
concentration. Biochemistry 1997,36, 10213−10220.
(23) Dynamics of Lipids, Cholesterol, and Transmembrane α-
Helices from Microsecond Molecular Dynamics Simulations. J. Phys.
Chem. B 2014,118 13590 13600. DOI: 10.1021/jp507027t.
(24) Teese, M. G.; Langosch, D. Role of GxxxG Motifs in
Transmembrane Domain Interactions. Biochemistry 2015,54,
5125−5135.
(25) Senes, A.; Gerstein, M.; Engelman, D. M. Statistical analysis of
amino acid patterns in transmembrane helices: the GxxxG motif
occurs frequently and in association with β-branched residues at
neighboring positions. J. Mol. Biol. 2000,296, 921−936.
(26) Högel, P.; Götz, A.; Kuhne, F.; Ebert, M.; Stelzer, W.; Rand, K.
D.; Scharnagl, C.; Langosch, D. Glycine Perturbs Local and Global
Conformational Flexibility of a Transmembrane Helix. Biochemistry
2018,57, 1326−1337.
(27) Sengupta, D.; Behera, R. N.; Smith, J. C.; Ullmann, G. M. The
alpha helix dipole: screened out? Structure 2005,13, 849−855.
(28) Lindahl, E.; Edholm, O. Mesoscopic Undulations and
Thickness Fluctuations in Lipid Bilayers from Molecular Dynamics
Simulations. Biophys. J. 2000,79, 426−433.
(29) Braun, A. R.; Brandt, E. G.; Edholm, O.; Nagle, J. F.; Sachs, J.
N. Determination of Electron Density Profiles and Area from
Simulations of Undulating Membranes. Biophys. J. 2011,100,
2112−2120.
(30) Mori, T.; Miyashita, N.; Im, W.; Feig, M.; Sugita, Y. Molecular
dynamics simulations of biological membranes and membrane
proteins using enhanced conformational sampling algorithms.
Biochim. Biophys. Acta, Biomembr. 2016,1858, 1635−1651.
(31) Rajagopal, N.; Nangia, S. Obtaining Protein Association Energy
Landscape for Integral Membrane Proteins. J. Chem. Theory Comput.
2019,15, 6444−6455.
(32) Webb, B.; Sali, A. Comparative Protein Structure Modeling
Using MODELLER. Curr. Protoc. Bioinf. 2016,54, 5.6.1−5.6.37.
(33) Jo, S.; Cheng, X.; Lee, J.; Kim, S.; Park, S. J.; Patel, D. S.;
Beaven, A. H.; Lee, K. I.; Rui, H.; Park, S.; et al. CHARMM-GUI 10
years for biomolecular modeling and simulation. J. Comput. Chem.
2017,38, 1114−1124.
(34) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M.
LINCS: a linear constraint solver for molecular simulations. J.
Comput. Chem. 1997,18, 1463−1472.
(35) Hess, B. P-LINCS: A Parallel Linear Constraint Solver for
Molecular Simulation. J. Chem. Theory Comput. 2008,4, 116−122.
(36) Huang, J.; Rauscher, S.; Nawrocki, G.; Ran, T.; Feig, M.; de
Groot, B. L.; Grubmüller, H.; Mackerell, A. D. CHARMM36m: an
improved force field for folded and intrinsically disordered proteins.
Nat. Methods 2017,14,71−73.
(37) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R.
W.; Klein, M. L. Comparison of simple potential functions for
simulating liquid water. J. Chem. Phys. 1983,79, 926−935.
(38) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.;
Pedersen, L. G. A smooth particle mesh Ewald method. J. Chem. Phys.
1995,103, 8577−8593.
(39) Abraham, M. J.; Murtola, T.; Schulz, R.; Páll, S.; Smith, J. C.;
Hess, B.; Lindahl, E. GROMACS: High performance molecular
simulations through multi-level parallelism from laptops to super-
computers. SoftwareX 2015,1−2,19−25.
ACS Omega http://pubs.acs.org/journal/acsodf Article
https://doi.org/10.1021/acsomega.1c00482
ACS Omega 2021, 6, 11458−11465
11465