Molecular dynamics simulation for membrane fusion

Abstract

The soluble N-ethylmaleimide sensitive factor attachment protein receptor (SNARE) protein complex drives membrane fusion, and this process is further aided by accessory proteins, including complexin and alpha-synuclein. To understand the molecular mechanism underlying membrane fusion, we introduce an all-atom molecular dynamics (MD) simulation method. This method tracks the conformations of protein and lipids, membrane geometry, and their interaction at femtosecond precision. Simulation results reveal information on distinct membrane fusion stages, including docking, hemifusion, and kiss-and-run fusion. Here, we introduce the simulation workflow, consisting of pre-MD construction, pre-MD setup in GROMACS, MD in GROMACS, and analysis.

Introduction

In the neuron, synaptic vesicles undergo membrane fusion with the plasma membrane, and this process is mediated by SNARE (soluble N-ethylmaleimide sensitive factor attachment protein receptor) proteins.^1,2 The cytoplasmic domains of SNAREs can form a four-α-helix bundle to act as a zipper for drawing two membranes together.^3,4 Moreover, SNARE proteins can accomplish membrane fusion in neurotransmitter release with the assistance of complexin, Munc18, synaptotagmin, and alpha-synuclein.^5–11 Other steps in the synaptic vesicle (SV) recycling process are modulated by these proteins, including vesicle docking and vesicle clustering.^12–19 Molecular dynamics (MD) simulation has been widely used to study protein-protein and membrane-protein interactions. However, the tradeoff in precision is between computational cost and the choice of timescale, as more complex systems at longer timescales require further approximation, and so different methods have been used to study intermediate fusion stages, including hemifusion structures and kiss-and-run fusion.^20–30

Mostafavi et al. used coarse-grained (CG) MD methods to show that membrane fusion does not require a critical number of SNARE complex constituents, but in fact, fusion rates increase with the number of SNAREpins.²⁵ Sharma and Lindau used CGMD simulation of nanodisc and bilayer membrane system bridged by SNARE proteins to reveal zipping of SNAREs pull C-terminal residues of synaptobrevin 2 and syntaxin 1A to form hydrophilic core between membrane leaflets which induces fusion pore formation.²⁶ However, another fusion pathway has been proposed, which is that a pore may form outside the hemifusion diaphragm, called leaky fusion. Previously, we performed all-atom molecular dynamics simulations to study the evolution of the hemifusion diaphragm structure with various lipid compositions. It was found that the lipid cholesterol decreased water penetrability to inhibit leakage pore formation.²⁷ There are other non-trivial details underlying pore formation, and mechanisms driving the evolution pathways of the hemifusion structure are of particular importance.

Another study combined all-atom MD and metadynamics simulation to study mechanisms of water pore formation in membranes in an applied electric field, which decreases the energy barrier for water molecules to pass through the membrane.²⁸ Furthermore, we used MD simulations of two lipid membranes to show that close proximity (<5nm) alone is enough to initiate fusion pore formation, inducing a transmembrane voltage that allows ion transport across the membrane and eventual pore healing (on the scale of a few nanoseconds), consistent with kiss-and-run fusion.²⁹ Moreover, external forces on the membrane from native proteins regulate the stability and lifetime of fusion pores and modulate the fusion pathway, whether kiss and run or full fusion.^10,29

Kasson et al. showed a branched pathway for fusion, in which a common stalk-like intermediate can either rapidly form a fusion pore or remain in a metastable hemifused state that slowly forms fully fused vesicles.³⁰ Using CGMD, Sharma et al showed the transmembrane domains (TMDs) of t-SNARE complexes form aggregates, leading to the formation of lipid nanodomains enriched in cholesterol and other lipids that induce membrane curvature that promotes closer contact between the vesicle and plasma membrane.²¹ McDargh et al. used CGMD to reveal a two-stage kinetic pathway for fusion, whereby zippering energy is dissipated, and entropic forces assemble SNAREpins into a ring, followed by expansion of the ring via entropic forces which initiate close membrane contact and catalyzes fusion. Moreover, they predict that any number of SNAREs can fuse membranes, but the timescale of fusion decreases with more SNAREs.²² Using all-atom MD, Rizo et al showed that SNAREs alone induce formation of extended membrane-membrane contact interfaces that slowly fuse, and suggest a mechanism in which trans-SNARE complexes prime the system for fast fusion upon Ca²⁺ influx.²³ Furthermore, in a later study they used all-atom MD to show that fast (microsecond scale) membrane fusion occurs when the SNARE helices zipper into juxtamembrane linkers which promote interactions of acyl chains between bilayers at the polar interface. This hydrophobic nucleus rapidly expands into stalk-like structures that eventually form a fusion pore.²⁴

In the following section, we describe how to setup MD simulation of a single protein and membrane system. Once this is achieved, it is straightforward to extend/modify these steps to prepare a two-membrane system with or without multiple SNARE and accessory proteins, to study the dynamics and molecular mechanism underlying membrane fusion.

System assembly and structure file generation

1. Go to CHARMM-GUI, at www.charmm-gui.org.^31,32

2. Go to Input Generator -> Membrane Builder.

3. Select Protein/Membrane System.

4. Select the PDB file for the target protein; see Figure 1. Here, monomeric WT α-synuclein is used as an example. Download PDB file: 1XQ8, download Source: RCSB, or leave it blank for the membrane-only system.

5. [Optional] If CHARM-GUI is not used to produce the structure files, extra caution needs to be used to correct the atom names, chain ID, and other structural information-based on interatomic distances.

6. Select the residue chain of the protein, as shown in Figure 2 above. Select residues 1 to 140 for full-length α-synuclein or residues 1–60 for its N-terminal domain.

7. Modify protein if necessary, by specifying residue-specific mutations, phosphorylation, etc., as shown in Figure 3 above.

8. Orient protein relative to the membrane, as shown in Figure 4 above. Rotate the protein around an axis by 90 degrees if needed, depending on the initial orientation of interest. For the data shown below, the initial protein orientation was aligned parallel to the membrane surface

9. Confirm protein cross-sectional area, choose membrane lipid composition (either by number or ratio) from the lipid library, which also contains fatty acids, sterols, and more; see Figure 5 above.

10. Select ions to be added to the solvent in the next step, as shown in Figure 6.

11. Assemble and use energy minimization to solvate geometry and add ions (this can later be modified in GROMACS). Use TIP3 as the default water model.³³

12. Choose the force field and specify output file types, including optional (PME FFT) grid, minimization, and equilibration parameter files; see Figure 7 above. Force Field Options: CHARMM36m, Input Generation Options: GROMACS.^34,35

13. Select the download.tgz button to download output files, which will be in a compressed TAR file called charmm-gui.tgz.

14. To unzip .tgz file in Linux, open the folder in the terminal and use the following:

    tar -xzvf file.tgz

Figure 1.

The First step of system assembly from Membrane Builder in CHARMM-GUI, protein selection.

Figure 2.

The second step of system assembly from Membrane Builder in CHARMM-GUI, residue specification.

Figure 3.

The third step of system assembly from Membrane Builder in CHARMM-GUI, protein modification.

Figure 4.

The fourth step of system assembly from Membrane Builder in CHARMM-GUI, protein orientation.

Figure 5.

The fifth step of system assembly from Membrane Builder in CHARMM-GUI, membrane composition specification.

Figure 6.

The sixth step of system assembly from the membrane builder in CHARMM-GUI is ion specification.

Figure 7.

The seventh step of system assembly from Membrane Builder in CHARMM-GUI, forcefield and equilibration file output specification.

Preparation of MD models in GROMACS

1. Import structure files output from CHARMM-GUI.

2. Modify box geometry if necessary using the code below.

   gmx editconf -f protein.gro -o protein_newbox.gro -box (membrane box vectors) -center x y z

3. Combine structure files if necessary, for instance, to make a structure file with 2 parallel membranes using one membrane structure file, using the code below.

   cat system_1.gro system_2.gro > system_combined.gro

4. Perform energy minimization with restrained protein and membrane.

5. Perform NVT ensemble equilibration to achieve stable temperature, with protein and membrane restrained.*

6. Perform NPT equilibration to achieve stable densities and appropriate lipid packing density, with restrained protein and membrane.*

7. Perform MD simulation.*

   Remark: Total simulation runtime depends on several factors, including system size (total number of atoms N), and length of simulation (total number of frames; 50,000,000 timesteps for 100 ns at a timestep of 2 fs). To solve at most 6N constrained equations of motion 50,000,000 times in sequence, or 300,000,000*N floating point operations, a single 100 ns run typically takes 10–12 hours, not counting queue time or additional runs for statistical ensemble averages, using the Ohio Supercomputer Center (OSC).

* All simulations use periodic boundary conditions, with temperature set to 310 K using the V-rescale algorithm.³⁶ The pressure was set to $p_{zz}=1\mathit{\text{bar}}$, $p_{xx}=p_{yy}=-28\mathit{\text{bar}}$ to achieve appropriate lipid packing density, using the Parrinello–Rahman method.³⁷ The LINCS algorithm is used to constrain bonds to hydrogen atoms.³⁸ The time step used was 2.0 fs. Long-range electrostatic interactions were calculated via PME, i.e., particle-mesh Ewald summation method.³⁹ The distance used for cutoff of non-bonded interactions was set to 12 Å.

Analysis of MD Trajectories in GROMACS

1. Calculate interaction energies by specifying index groups from index.ndx.

   gmx energy -f system.edr -n index.ndx -o output_file.xvg

   One can use the rerun feature in GROMACS to redefine energy groups for quick calculation or interaction energies between different groups of atoms, residues, or molecules without having to wait for a full-time run.

2. Calculate hydrogen bonds by specifying index groups from index.ndx.

   gmx hbond -f system.xtc -s system.tpr -n index.ndx -num output_file.xvg

   Specify the atomic or molecular groups (from index.ndx) required to count hydrogen bonds between each frame of the trajectory file (system.xtc).

   Remark: Trajectory analysis runtime depends on several factors, including system size, length of simulation, and the number of energy groups defined, but typically, this is relatively short (< 1–2 min) for energy analysis. Larger energy group analysis (e.g., per residue interaction energies) can be done using the -rerun feature in GROMACS, which allows for quick energy analysis with newly defined groups (given in the index) without full simulation runtime. However, hydrogen bond analysis takes significantly longer because the script checks the distances and angles between two groups of atoms (not just bonded atoms!), proceeds to do this check for every possible group of atoms, and defines a hydrogen bond if the distance and angle are below (default) cutoff values, given by $r\le 0.35\mathit{\text{nm}}$ and $\theta \le 30^{\circ }$; see figure below. Typical runtimes are approximately 5–10 min.

Summary

MD simulation sheds insights into the dynamics of interacting biophysical structures at scales that cannot be probed experimentally using modern optical methods, including spectroscopy and FRET-based assays. Furthermore, in a multiple-lipid vesicle system, it is difficult to describe the dynamics of the interaction of a protein and a single class of lipids in the vesicle from the experiment. MD simulation, however, addresses this issue by providing the ability to specify the lipid group of interest from an index file and calculate interaction energies from the output trajectory file, revealing dynamics of protein-lipid interaction, even at the level of individual residues. Result interpretation requires at least two simulation systems, one as a control and the other as the system of interest, e.g., “0% DHA” compared to “20% DHA” in Figure 10. Furthermore, this also allows the comparison of ensemble interaction energies among different systems.

Figure 10.

Comparison of ensemble average interaction energies for protein-membrane interaction in a membrane system with and without the fatty acid DHA, where the defined groups are the entire protein and a group consisting of all membrane lipids. (A) The ensemble average number of hydrogen bonds for protein-membrane interaction, where the defined groups are the entire protein and a group consisting of all membrane lipids. (B) The condition “0% DHA” corresponds to a composition of 12%DOPS, 20%DOPE, and 68%DOPC, while “20% DHA” corresponds to a composition of 20% DHA, 12%DOPS, 20%DOPE, and 48%DOPC. Averaging was performed over all simulations of a given composition. The graph corresponds to an average of 4 individual simulations of the same composition and MD parameters.

Figure adapted from Tyoe et al,¹⁷ Docosahexaenoic acid promotes vesicle clustering mediated by alpha-Synuclein via electrostatic interaction, The European Physical Journal E - Soft Matter, Springer Nature, Oct 12, 2023, Copyright © 2023, The Author(s), under exclusive licence to EDP Sciences, SIF and Springer-Verlag GmbH Germany, part of Springer Nature.

For a single protein-membrane embedding system, the interaction energy (at any time after embedding) is a negative number associated with binding energy induced by a combination of electrostatic and hydrophobic interaction energies. Thus, for the protein-membrane system example above, DHA addition results in a greater (absolute value of) interaction energy, indicating stronger interaction. The study of α-Synuclein binding to synaptic vesicles and plasma membrane is particularly interesting because even neutral molecules, such as DHA and cholesterol, modulate α-Synuclein’s interaction with the membrane by affecting physical membrane properties, including changes in thickness, packing defects, curvature, etc. This observation is consistent with recent work, which suggested that DHA enhances α-Synuclein-membrane interaction energy and reduces membrane perturbation induced by α-Synuclein insertion.^17,40 Furthermore, other membrane components may regulate this interaction via different mechanisms. For instance, cholesterol affects the curvature and distribution of packing defects in the membrane, regulating α-Synuclein-membrane interaction in a concentration-dependent way.⁴¹

Vesicle-associated membrane protein 2 (VAMP2) combined with syntaxin-1A and synaptosome-associated protein 25 (SNAP-25) induces enhanced formation of fusion pores, which regulates the fusion of synaptic vesicles and the release of neurotransmitters. SNAREs also regulate membrane fusion by interacting with other accessory proteins, such as synaptotagmin and complexin. Triggered by calcium flux, these accessory proteins can modulate membrane association and functions.^24,42 Moreover, α-Synuclein has been shown to cross-bridge VAMP2 and anionic phospholipids to mediate SNARE-dependent vesicle docking.⁴³ While these processes have been probed experimentally, all-atom molecular dynamics simulation could provide nanoscale dynamics of vesicle docking and pinpoint mechanistic insights at the level of individual lipids or residues. This information is crucial in delineating the physiological and pathological role of accessory proteins in membrane fusion.

Figure 8.

Snapshot of a pre-MD simulation setup to simulate the interaction bewteen the N-terminal domain of alpha-synuclein (residues 1–60) and an anionic lipid bilayer. In the membrane, cyan represents C-C bonds, red represents O- bonds, orange P- bonds, blue N- bonds, and white H-bonds. This particular membrane is composed of 12%DOPS, 20%DOPE, and 68%DOPC.

Figure 9.

Snapshot of MD simulation of N-terminal region of alpha-synuclein (residues 1–60) binding to anionic lipid bilayer after energy minimization, equilibration, and 20 ns MD.

Figure 11.

Geometric criterion to determine the status of hydrogen bond.