All-atom molecular dynamics simulations of Synaptotagmin-SNARE-complexin complexes bridging a vesicle and a flat lipid bilayer

  1. Josep Rizo  Is a corresponding author
  2. Levent Sari
  3. Yife Qi
  4. Wonpil Im
  5. Milo M Lin
  1. Department of Biophysics, The University of Texas Southwestern Medical Center, United States
  2. Department of Biochemistry, University of Texas Southwestern Medical Center, United States
  3. Department of Pharmacology, University of Texas Southwestern Medical Center, United States
  4. Green Center for Systems Biology, The University of Texas Southwestern Medical Center, United States
  5. Department of Medicinal Chemistry, School of Pharmacy, Fudan University, China
  6. Department of Biological Sciences, Lehigh University, United States
  7. Department of Chemistry, Lehigh University, United States
  8. Department of Bioengineering, Lehigh University, United States
  9. Department of Computer Science and Engineering, Lehigh University, United States

Abstract

Synaptic vesicles are primed into a state that is ready for fast neurotransmitter release upon Ca2+-binding to Synaptotagmin-1. This state likely includes trans-SNARE complexes between the vesicle and plasma membranes that are bound to Synaptotagmin-1 and complexins. However, the nature of this state and the steps leading to membrane fusion are unclear, in part because of the difficulty of studying this dynamic process experimentally. To shed light into these questions, we performed all-atom molecular dynamics simulations of systems containing trans-SNARE complexes between two flat bilayers or a vesicle and a flat bilayer with or without fragments of Synaptotagmin-1 and/or complexin-1. Our results need to be interpreted with caution because of the limited simulation times and the absence of key components, but suggest mechanistic features that may control release and help visualize potential states of the primed Synaptotagmin-1-SNARE-complexin-1 complex. The simulations suggest that SNAREs alone induce formation of extended membrane-membrane contact interfaces that may fuse slowly, and that the primed state contains macromolecular assemblies of trans-SNARE complexes bound to the Synaptotagmin-1 C2B domain and complexin-1 in a spring-loaded configuration that prevents premature membrane merger and formation of extended interfaces, but keeps the system ready for fast fusion upon Ca2+ influx.

Editor's evaluation

Using all-atom molecular dynamics simulations to visualize the pre-fusion primed state during synaptic vesicle fusion is very original and this approach will certainly be used by others in the future. This work provides new insights into the protein organization prior to vesicle fusion that will help better understand the mechanisms of vesicle priming and evoked-release.

https://doi.org/10.7554/eLife.76356.sa0

Introduction

The release of neurotransmitters by Ca2+-triggered synaptic vesicle exocytosis is key for communication between neurons. The high speed of this process arises in part because synaptic vesicles are first tethered at the plasma membrane and undergo priming processes that leave them ready for fast fusion when an action potential induces Ca2+ influx (Südhof, 2013). Research for over three decades has led to extensive characterization of the neurotransmitter release machinery (Brunger et al., 2018; Rizo, 2022), allowing reconstitution of basic features of synaptic vesicle fusion with the central components of this machinery (Lai et al., 2017; Liu et al., 2016; Ma et al., 2013) and uncovering key aspects of the underlying mechanisms. The SNARE proteins syntaxin-1, SNAP-25 and synaptobrevin form a tight four-helix bundle called the SNARE complex that assembles (zippers) from the N- to the C-terminus to bring the vesicle and plasma membranes together, and is key for membrane fusion (Hanson et al., 1997b; Poirier et al., 1998; Söllner et al., 1993; Sutton et al., 1998). This complex is disassembled by NSF and SNAPs (no relation to SNAP-25) to recycle the SNAREs for another round of fusion (Mayer et al., 1996; Söllner et al., 1993). Munc18-1 and Munc13-1 play central roles in synaptic vesicle priming by organizing assembly of the SNARE complex via an NSF-SNAP-resistant pathway (Ma et al., 2013; Prinslow et al., 2019) whereby Munc18-1 first binds to a self-inhibited ‘closed’ conformation of syntaxin-1 (Dulubova et al., 1999; Misura et al., 2000). Munc13-1 later opens syntaxin-1 (Ma et al., 2011; Yang et al., 2015) while bridging the vesicle and plasma membranes (Quade et al., 2019; Xu et al., 2017), and Munc18-1 forms a template for SNARE complex assembly (Baker et al., 2015; Jiao et al., 2018; Parisotto et al., 2014; Sitarska et al., 2017).

Synaptotagmin-1 (Syt1) acts as the Ca2+ sensor that triggers fast release (Fernández-Chacón et al., 2001) in a tight interplay with complexins whereby both Syt1 and complexins play inhibitory and active roles (Giraudo et al., 2006; Reim et al., 2001; Schaub et al., 2006; Tang et al., 2006). The Ca2+ sensor function is performed by the two C2 domains of Syt1 (C2A and C2B), which form most of its cytoplasmic region and bind multiple Ca2+ ions via loops at the tip of β-sandwich structures (Fernandez et al., 2001; Sutton et al., 1995; Ubach et al., 1998; Figure 1—figure supplement 1A). These loops also mediate Ca2+-dependent binding of both C2 domains to phospholipids, which is critical for release (Fernández-Chacón et al., 2001; Rhee et al., 2005). The C2B domain also binds to PIP2 through a polybasic region on the side of the β-sandwich (Bai et al., 2003), which induces binding to the plasma membrane. Moreover, the C2B domain can bind to the SNARE complex through three different surfaces (Brewer et al., 2015; Zhou et al., 2015; Zhou et al., 2017), although only binding through a so-called primary interface (Zhou et al., 2015) is firmly established as physiologically relevant (Guan et al., 2017; Voleti et al., 2020). Such binding is disrupted upon Ca2+-dependent binding of Syt1 to PIP2-containing membranes (Voleti et al., 2020).

Complexin-1 binds tightly to the SNARE complex through a central α-helix that is preceded by an accessory helix (Chen et al., 2002) and may play a stimulatory role in release by promoting formation of a primed state with enhanced release probability (Chen et al., 2002), by protecting trans-SNARE complexes from disassembly by NSF and αSNAP (Prinslow et al., 2019) and/or by synchronizing Ca2+-triggered fusion mediated by the SNAREs and Syt1 (Diao et al., 2012). The complexin-1 accessory helix inhibits release (Xue et al., 2007), possibly because it causes steric clashes with the vesicle, hindering C-terminal assembly of the SNARE complex (Radoff et al., 2014; Trimbuch et al., 2014). These and other findings suggested that complexin-1 and Syt1 bind simultaneously to the SNARE complex in the primed state of synaptic vesicles, stabilizing this state and preventing premature fusion; in this model, Ca2+ influx relieves the inhibition by inducing dissociation of Syt1 from the SNAREs and enabling cooperation between Syt1 and the SNAREs in promoting fusion (Voleti et al., 2020).

Despite this wealth of knowledge, fundamental questions remain about the mechanism of neurotransmitter release, particularly regarding how the SNAREs and Syt1 trigger fast, Ca2+-dependent membrane fusion. Major hurdles to address this question are the dynamic nature of this process and the fact that the protein complexes that trigger fusion are assembled between two membranes. Although important clues on the nature of the primed macromolecular assembly have been obtained with structural studies of soluble proteins or complexes anchored on one membrane (Chen et al., 2002; Grushin et al., 2019; Voleti et al., 2020; Zhou et al., 2015), this assembly is most likely affected by its location between two membranes. This feature strongly hinders the possibility of crystallization, while application of NMR spectroscopy for structure elucidation is hampered by the large size of any reconstituted two-membrane system (Voleti et al., 2021). Conversely, the small size of the SNAREs, Syt1 and complexin-1 hinders visualization by cryo-EM. Moreover, it is extremely challenging to capture transient states formed during the pathway to Ca2+-triggered membrane fusion experimentally.

Molecular dynamics (MD) simulations offer a powerful tool to analyze dynamic biomolecular processes and model cellular membranes (Marrink et al., 2019). Simulations using continuum and/or coarse-grained representations have provided important insights into SNARE-mediated membrane fusion (Fortoul et al., 2015; Kasson et al., 2006; Manca et al., 2019; McDargh et al., 2018; Mostafavi et al., 2017; Risselada et al., 2011; Sharma and Lindau, 2018). Continuum models can access the longest timescales, but require experimental data or atomistic simulations to parameterize the material properties, and often need to constrain geometries or material properties due to lack of context-dependent parameters (Fortoul et al., 2015). Coarse-grained molecular simulation approaches are freed from some of these constraints but at the expense of reduced simulation speed, and are limited in their ability to capture certain entropic effects and protein conformational changes (see below). To date, coarse-grained models of SNARE-mediated fusion have accessed the low microsecond timescale (Kasson et al., 2006; Risselada et al., 2011; Sharma and Lindau, 2018). All-atom simulations are better suited to reproduce the finely-balanced network of interactions between proteins, Ca2+, and lipids that are expected to lead to membrane fusion but, because of the large size of the systems involved (millions of atoms), the low microsecond time scale has only recently become accessible through the most powerful available high-performance computing resources. In this context, it is worth noting that the delay from Ca2+ influx into the presynaptic terminal to observation of postsynaptic currents in rat cerebellar synapses at 38 °C is 60 μs (Sabatini and Regehr, 1996), and that multiple events occur within this time frame, including Ca2+ binding to the sensor, release of inhibitory interactions that hinder premature fusion, Ca2+-evoked synaptic vesicle fusion, opening of the fusion pore, diffusion of neurotransmitters through the synaptic cleft, binding of the neurotransmitters to their postsynaptic receptors and opening of the channels that underlie the postsynaptic currents. These observations suggest that the fusion step may occur in just a few microseconds and hence that it may be possible to recapitulate the initiation of Ca2+-dependent synaptic vesicle fusions in all-atom MD simulations starting with a properly designed initial configuration.

Here, we present all-atom MD simulations with explicit water molecules of systems containing four trans-SNARE complexes bridging two flat bilayers or a vesicle and a flat bilayer, without or with fragments of Syt1 and/or complexin-1. Because of the limited simulation times and the absence of key components, our results cannot lead to definitive conclusions but they help visualize potential trajectories and intermediates along the pathway to fusion and reveal intriguing features, leading to predictions or hypotheses that can be tested experimentally and with additional simulations. Our data indicate that trans-SNARE complexes strongly pull two membranes together, as expected, but have a tendency to induce extended membrane-membrane adhesion interfaces that have been observed experimentally but fuse slowly (Hernandez et al., 2012; Witkowska et al., 2021). Our results also suggest that, in the primed state of synaptic vesicles, Syt1 and complexin-1 form a spring-loaded macromolecular assembly with trans-SNARE complexes that hinders formation of such extended contact interfactes and premature bilayer merger, but is ready for fast membrane fusion upon Ca2+ influx.

Results

Four trans-SNARE complexes between two flat lipid bilayers

The possibility of observing membrane fusion in the low microsecond time scale in all-atom MD simulations depends critically on the choice of the starting configuration, but the exact nature of the primed state of synaptic vesicles is unknown. Hence, we used the structural and functional information available on this system to generate potential starting configurations. The MD simulations presented here involved systems ranging from 1.7 to 5.9 million atoms. While multiple replicas of each simulation should ideally be carried out to verify the consistency of the results, performing replicated simulations would have limited the number of systems that we could study. Moreover, each simulation included four-trans SNARE complexes bridging two lipid bilayers and the variability in the behavior of the complexes in each simulation already provided insights into the consistency of the observed behaviors. Hence, we chose to use the available high performance computing time to investigate systems with different components and/or distinct geometry, designing each new starting configuration according to what we had learned from the previous simulations. The simulations generated a large amount of data and it is impossible to describe a thorough analysis within the constraints of a single paper. Here, we present the main observations from the analyses that we have performed, and key files from the simulations are available in Dryad for further analyses.

In all the systems that we built, the composition of the bilayer containing anchored synaptobrevin approximated the lipid composition of synaptic vesicles (Takamori et al., 2006) and that of the bilayer with anchored syntaxin-1 was based on the lipid composition of the plasma membrane (Chan et al., 2012; Table 1). Both bilayers had asymmetric lipid distributions in the two leaflets to mimic those present in vivo (Kobayashi and Menon, 2018). The first system that we built was designed to examine whether SNARE complexes alone can bend two flat lipid bilayers and initiate bilayer fusion. The system contained four trans-SNARE complexes between two square lipid bilayers. The number of SNARE complexes was based on symmetry considerations and the finding that fast vesicle fusion typically observed in synapses requires at least three SNARE complexes (Mohrmann et al., 2010). For simplicity, the SNARE complexes contained the four SNARE motifs, the transmembrane (TM) sequences of syntaxin-1 and synaptobrevin, and the juxtamembrane linkers between their respective SNARE motif and TM region, but did not include the syntaxin-1 N-terminal region or the long linker between the two SNAP-25 SNARE motifs.

Table 1
Size in atoms, length of productions MD simulations, temperature, speed of the simulations on Frontera at TACC and lipid composition of the flat bilayers and the vesicle of the different systems.
qscff1699436 atoms750 ns310 K24 ns/day with 16 nodes
Upper bilayerCHL1POPCPOPSSAPESDPESDPStotal
Upper leaflet54046807212001200
%453906100
Lower leaflet54014684120240961226
%4411.96.99.819.67.8
Lower bilayerCHL1POPCPOPSSAPI2DSAPESDPESDPStotal
Upper leaflet54013412060841681201226
%44119.84.96.913.79.8
Lower leaflet54053800488401200
%454400470
sqscff1700475 atoms270 ns310 K 24 ns/day with 16 nodes
qscv3222393 atoms520 ns310 K23 ns/day with 32 nodes
454 ns325 K
VesicleCHL1POPCSAGLSAPESAPI2DSDPESDPSSOPStotal
Outer leaflet1258296053402822101992779
%45.310.6019.2010.17.67.2
Inner leaflet8146680183099111766
%46.137.8010.405.60.10.1
Flat bilayerCHL1POPCSAGLSAPESAPI2DSDPESDPSSOPStotal
Upper leaflet540961284601561201201188
%45.48.117.15.113.110.110.1
Lower leaflet5405161236084001188
%45.443.41307.100
prsg5056443 atoms336 ns310 K13 ns/day with 32 nodes
Flat bilayerCHL1POPCSAGLSAPESAPI2DSDPESDPSSOPStotal
Upper leaflet83015118132932621821811849
%44.98.217.1514.29.89.8
Lower leaflet81077418720126001800
%4543140700
prs25870280 atoms310 ns310 K16 ns/day 48 nodes
Flat bilayerCHL1POPCSAGLSAPESAPI2DSDPESDPSSOPStotal
Upper leaflet1035184231611153222302302300
%458175141010
Lower leaflet100996422880154002237
%45.143.113.906.900
prsncpxca5870246 atoms439 ns310 K16 ns/day 48 nodes
  1. Vesicle: same as qscv system.

  2. Flat bilayer: same as prs2 system.

A key aspect in the design of realistic potential states of trans-SNARE complexes is the conformation of the juxtamembrane linkers of syntaxin-1 and synaptobrevin. Popular models of SNARE-mediated membrane fusion depicted continuous helices spanning the SNARE motifs, juxtamembrane linkers and TM regions for both synaptobrevin and syntaxin-1, envisioning that these helices can bend to accommodate the geometry of trans-SNARE complexes (Hanson et al., 1997a; Sutton et al., 1998; Weber et al., 1998). These models were supported by coarse-grained MD simulations that used the MARTINI force field and modeled the SNAREs in continuous helical conformations (Risselada et al., 2011). However, the intrinsic helical restraints enforced by the force field might bias the results and/or obscure the potential role of conformational changes in the dynamical coupling of the SNAREs to membrane fusion. Moreover, the bending of the helices required to form trans-SNARE complexes leads to unrealistic conformations that are expected to be unfavorable energetically because of their distorted geometry and are not commonly observed in protein structures. Thus, the helical restraints might have played a key role in membrane fusion in these simulations. Although continuous helices were observed in the crystal structure of a cis-SNARE complex that represents the configuration occurring after membrane fusion (Stein et al., 2009), the natural expectation is that the helical structure must break somewhere to accommodate the geometry of a trans-SNARE complex, most likely at the juxtamembrane linker. This expectation has been supported experimentally (Kim et al., 2002) and with all-atom MD simulations (Bykhovskaia, 2021). Moreover, helix continuity in the linkers is not required for neurotransmitter release (Kesavan et al., 2007; Zhou et al., 2013).

Thus, to generate trans-SNARE complexes for our simulations we started with the crystal structure of the cis-SNARE complex but we did not impose restraints on the conformation of the juxtamembrane linkers. Since the N-terminal half of the SNARE four-helix bundle is more stable than the C-terminal half (Chen et al., 2002; Gao et al., 2012) and is more distal from the membrane, we imposed position restraints for only the N-terminal half. In addition, we used position restraints to force the TM regions of synaptobrevin and syntaxin-1 to designed locations for insertion in their corresponding bilayers.

A short (1 ns) restrained MD simulation in water was sufficient for this purpose and led to unstructured conformations for the juxtamembrane linkers without substantially altering the four-helix bundle even at the C-terminal half, which was not restrained (Figure 1—figure supplement 1B). Four copies of the resulting trans-SNARE complex were generated by translations and rotations (Figure 1—figure supplement 1C), and were merged with the two bilayers separated by 5 nm to generate the initial configuration of this system (Figure 1A). We then carried out an unrestrained production simulation of this system for 750 ns at 310 K. As expected, the two membranes became almost circular to minimize tension and were gradually drawn together by the SNAREs, although the minimal distance between the bilayers reached a plateau (Figure 1—figure supplement 1D, E). The two bilayers were actually drawn to each other on one side first (at about 110 ns, Figure 1B) and later on the other side (Figure 1C), leading to close packing of the lipids against the SNARE four-helix bundles (Figure 1D). It is noteworthy that the four SNARE complexes were zippered at the C-terminus to the same extent as in the initial configuration and that extensive interactions were established between the juxtamembrane linkers and the membranes during the simulations. Such interactions were not unexpected, as both linkers contain abundant basic residues, the synaptobrevin linker in addition contains hydrophobic residues, and both linkers were shown to interact with the adjacent membrane (Brewer et al., 2011; Kim et al., 2002). Since much of the SNARE four-helix bundle is negatively charged, these findings suggest that any electrostatic repulsion existing between the SNARE four-helix bundle and the membranes can be readily overcome by the high stability of the SNARE four-helix bundle and perhaps some contribution from the linker-bilayer interactions. During the 750 ns of the simulation we occasionally observed mild buckling of the syntaxin-1 membrane, but the buckling was reversible and there was no progress toward fusion. These findings suggest that four trans-SNARE complexes are unable to fuse two flat bilayers in the 1 μs time scale.

Figure 1 with 1 supplement see all
MD simulations of four trans-SNARE complexes bridging two flat bilayers.

(A–C) Initial configuration of the system with SNARE complexes only (A), and snapshots of the MD simulation after 110 and 270 ns (B, C). The SNARE complexes are illustrated by ribbon diagrams in salmon. The lipids are shown as thin stick models. The scale bar in (A) equals 10 nm, which is a little shorter than the length of the SNARE four-helix bundle. (D) Snapshot of the same MD simulation at 270 ns showing all non-solvent atoms as spheres. (E–F) Initial configuration of the system containing four Ca2+-bound Syt1 C2AB molecules in addition to the four trans-SNARE complexes (E) and snapshot of the simulation at 270 ns (F). SNARE complexes are illustrated by ribbon diagrams in salmon and the C2AB molecules are shown as ribbon diagrams with C2A in cyan and C2B in violet. The lipids are shown as thin stick models. The atom color code for the lipids is: carbon lime, oxygen red, nitrogen blue, phosphorous orange. Ca2+ ions are shown as yellow spheres.

To explore whether Syt1 might cooperate with the SNAREs in bending two flat bilayers to initiate membrane fusion, we performed another simulation with an analogous system where we included a fragment spanning the two C2 domains of Syt1 (C2AB) bound to five Ca2+ ions (Fernandez et al., 2001; Ubach et al., 1998; Figure 1E). During a 270 ns production MD simulation of this system, we observed that the C2AB molecules hindered the action of the trans-SNARE complexes in bringing the two bilayers closer (Figure 1—figure supplement 1F), particularly when the C2 domains bind to one bilayer through the Ca2+-binding loops and to the other bilayer via the opposite side of the β-sandwich, which is basic (Figure 1F). Although such bilayer-bilayer bridging might help in fusion (Araç et al., 2006) in a different configuration, it appeared that such potential action would require a much longer time scale in this system and we did not continue this simulation.

Four trans-SNARE complexes bridging a vesicle and a flat bilayer

Based on the results from the simulations with four trans-SNARE complexes between two flat bilayers, we reasoned that, if synaptic vesicle fusion indeed occurs in the low microsecond time scale, this speed might require the geometry occurring at synapses, where small synaptic vesicles (ca. 40 nm diameter) fuse with the plasma membrane. To test this notion, we built a system with four trans-SNARE complexes (Figure 2—figure supplement 1A) bridging a vesicle and a flat bilayer (Figure 2A). The initial diameter of the vesicle (26 nm) was chosen as a compromise between making the system realistic and minimizing the overall size of the system to limit the time required for MD simulations. The vesicle was practically in molecular contact with the flat bilayer so that the system was poised for fusion. Since the lipid density of the vesicle was close to but not optimal, holes appeared in an initial production MD simulation. The holes were filled manually in an iterative process until the vesicle was stable (see Methods). During this procedure, the flat bilayer became circular and the vesicle became slightly smaller (24 nm diameter) (Figure 2—figure supplement 1B, C), but the diameter remained stable in subsequent production runs.

Figure 2 with 1 supplement see all
MD simulation of four trans-SNARE complexes bridging a vesicle and a flat bilayer.

(A) Overall view of the initial system. (B) Close-up view of one of the trans-SNARE complexes in the initial system. (C) Snapshot of the system after a 520 ns MD simulation at 310 K and a 454 ns simulation at 325 K. (D–G) Close-up views of the four trans-SNARE complexes (named SC1-SC4) after the 520 ns MD simulation at 310 K and the 454 ns simulation at 325 K. In (A–G), the SNARE complexes are illustrated by ribbon diagrams in salmon. The lipids are shown as thin stick models (carbon lime, oxygen red, nitrogen blue, phosphorous orange). (H–K) Thin slices of the system in its initial configuration (H), after the equilibration steps (I), after 280 ns at 310 K (J) and after 520 ns at 310 K and 454 ns at 325 K (K). In (H–K) Phosphorous atoms of phospholipids and the oxygen atoms of cholesterol molecules are shown as spheres to illustrate the approximate locations of lipid head groups.

With the system equilibrated, we performed a production run of 520 ns at 310 K. Although we observed occasional flips of cholesterol molecules, there were no persistent perturbations of the bilayers that might signal the initiation of fusion. We raised the temperature to 325 K and carried out a production run of 454 ns in an attempt to accelerate fusion, but observed similar results. The final configuration illustrates that the vesicle diffused to some extent to one side with respect to the flat bilayer during the simulations (Figure 2C). The four-helix bundles of the four trans-SNARE complexes remained fully assembled up to the last hydrophobic layer [referred to as layer +8 (Sutton et al., 1998)] at the end of the simulations (Figure 2D–G), as in the initial configuration of the system (Figure 2B). One of the SNARE complexes became parallel to the flat bilayer (Figure 2F), whereas the other three had similar orientations as in the starting configuration (Figure 2B, D, E and G). The juxtamembrane linkers of synaptobrevin and syntaxin-1 established extensive interactions with the lipids early in the simulations. After 280 ns of the simulation at 310 K, we observed that the bottom of the vesicle was flattened, resulting in an extended contact interface with the flat bilayer (compare the slice view of Figure 2J with those of the initial configurations in Figure 2H, I). To corroborate these findings quantitatively, we calculated the number of contacts between oxygen atoms of the vesicle and the flat bilayer as a function of time. We assigned a contact to each oxygen-oxygen distance below 1 nm, which is a common cutoff used to calculate van der Waals and electrostatic interactions between atoms. The results showed that the number of contacts increased rapidly up to about 300 ns and then leveled off (Figure 2—figure supplement 1D). The extended contact interface persisted until the end of the simulation at 325 K, and during this simulation the flat bilayer became slightly curved to adapt to the shape of the vesicle (Figure 2K).

These finding correlates with results obtained in cryo-EM analyses of liposome fusion reactions mediated by the neuronal SNAREs, which revealed extended contact interfaces between the liposomes that are referred to as tight docking intermediates (Hernandez et al., 2012). These intermediates eventually evolve to yield membrane fusion, but fusion occurs in the second-minute time scale (Hernandez et al., 2012; Witkowska et al., 2021). Hence, it is unlikely that such extended interfaces occur in the pathway that leads to fast Ca2+-triggered synaptic vesicle fusion. Note also that the energy required to initiate membrane fusion is expected to increase with the area of the interface between the two membranes, as larger areas require more lipid molecules to be rearranged. Interestingly, cryo-EM images of reconstitution reactions including additional components of the release machinery suggested that these additional components prevent formation of extended contact interfaces, favoring interfaces with smaller contact area between the bilayers that are referred to as point-of-contact interfaces (Gipson et al., 2017).

Simulations of the primed Synaptotagmin-1-SNARE-complexin-1 complex

Overall, our simulations do not rule out the possibility that SNAREs alone might be able to induce membrane fusion in the low microsecond time scale, as it is plausible that other geometries might be more efficient in inducing fusion. However, the correlation of our results with the cryo-EM images of reconstitution experiments suggests that fast fusion requires additional proteins. Formation of a primed state of synaptic vesicles that is ready for fast release is the key to achieve fast Ca2+-triggered fusion in synapses. The exact nature of this state is unclear, but Syt1 and complexin are most likely bound to trans-SNARE complexes in this state, as both proteins bind to the SNARE complex and are critical for fast, Ca2+-triggered neurotransmitter release (Fernández-Chacón et al., 2001; Reim et al., 2001). A model of this state (Voleti et al., 2020) was proposed based on crystal structures of the SNARE complex bound to a complexin-1 fragment (Chen et al., 2002) or to the Syt1 C2B domain through the primary interface (Zhou et al., 2015), as well as a cryo-EM structure of Syt1 bound to lipid nanotube-anchored SNARE complex (Grushin et al., 2019). In this model, complexin-1 and the Syt 1 C2B domain bind to opposite sides of the SNARE four-helix bundle, and the C2B domain binds to the plasma membrane through a polybasic region on the side of the C2B domain opposite to the primary interface. However, the orientation of this macromolecular assembly with respect to the vesicle and plasma membranes, and the extent to which the SNARE complex is zippered, are unclear.

To gain insights into the nature of the primed Syt1-SNARE-complexin complex that is ready for fast Ca2+-triggered membrane fusion, we built a system with four trans-SNARE complexes bridging a vesicle and a flat bilayer, each bound to a complexin-1 fragment and the Syt1 C2AB fragment as observed by crystallography (Chen et al., 2002; Zhou et al., 2015; below referred to as primed complexes). The complexin-1 fragment spanned residues 27–72 [Cpx1(27-72)], which include the central helix that binds to the SNARE complex and the preceding accessory helix that is believed to underlie the inhibitory activity of the accessory helix (Trimbuch et al., 2014; Xue et al., 2007). The system was designed to resemble a potential arrangement of the primed state, but implementing some flexibility such that the system could progress towards a preferred configuration of the proteins with respect to the two membranes. A restrained MD simulation performed to generate the initial protein arrangement led to partial unfolding of the C-terminal halves of the SNARE four-helix bundles but to distinct extents (Figure 3A, B, D, F, H and J, Figure 3—figure supplement 1), thus yielding a variety of starting configurations of the complexes.

Figure 3 with 2 supplements see all
First MD simulation of primed complexes bridging a vesicle and a flat bilayer.

(A) Overall view of the initial system after equilibration. (B) Close-up view of the four primed complexes in the initial system after equilibration. (C) Snapshot of the system after a 336 ns MD simulation. (D–K) Close-up views of the individual primed complexes (named PC1-PC4) in the initial configuration (D,F,H,J) and after the 336 ns MD simulation (E,G,I,K). In all panels, the primed complexes are illustrated by ribbon diagrams, with the SNAREs in salmon, Cpx1(27-72) in yellow and the Syt1 C2AB fragment in cyan (C2A domain) and violet (C2B domain). The lipids are shown as thin stick models (carbon lime, oxygen red, nitrogen blue, phosphorous orange).

After equilibration, we carried out a production simulation of 336 ns that resulted in the state shown in Figure 3C. Most substantial changes in the system occurred early in the simulation and each of the primed complexes appeared to reach a stable or metastable configuration by the end. The primed complexes exhibited some common behaviors and also distinct features. The SNARE four-helix bundle of one of the primed complexes (PC1) was almost fully assembled at the start of the simulation (up to layer +7, with a break in a SNAP-25 helix) and remained equally assembled at the end (Figure 3D, E). The C-terminal halves of the SNARE four-helix bundles of the other three complexes were considerably more disrupted and, although they exhibited substantial changes during the simulation, they did not progress toward full assembly (Figure 3F–K). Interestingly, a few of the most C-terminal layers (+5 to+7) still formed a four-helix bundle in two complexes (PC2 and PC3, Figure 3F-I) and hence they may be particularly stable, but this feature did not seem to facilitate reassembly of the section of the four-helix bundle that was disrupted. Hence, although coil-to-helix transitions are known to occur very fast, in the 100 ns time scale (Muñoz and Cerminara, 2016), it appears that the constraints placed on the motions of the SNAREs in this complex system hinder the evolution toward a fully formed SNARE four-helix bundle.

Interestingly, the Cpx1(27-72) accessory helix exhibited clear steric clashes with the vesicle in all complexes. To avoid such clashes, the continuity between the central and accessory helices was broken in some cases, with the helix bending to one side or another (Figure 3G, I). In PC4, the entire helix changed orientation (Figure 3K), whereas in PC1, where the four-helix bundle is almost fully assembled, the helix was distorted into a snake shape (Figure 3E). The ‘struggle’ of the accessory helix to avoid bumps with the vesicle is particularly well illustrated by distinct bends of the Cpx1(27-72) helix occurring in PC3 during the simulations (Figure 3—figure supplement 2A–E). It is also noteworthy that Cpx1(27-72) remained bound to the SNAREs throughout the simulations due to interactions of the C-terminal end of the Cpx1(27-72) helix, particularly the Y70 aromatic ring, with a hydrophobic pocket of the SNARE complex, which persisted even when the overall direction of the helix changed in PC4 (Figure 3—figure supplement 2F). Overall, these observations provide a vivid visual illustration of the steric clashes between the complexin-1 accessory helix that may occur in the primed state, which were proposed to underlie the inhibition of neurotransmitter release caused by this helix (Trimbuch et al., 2014).

A common feature of the four primed complexes at the end of the simulation was the arrangement of the Syt1 C2B domain, which was initially placed between the SNARE four-helix bundle and the flat bilayer but in all primed complexes changed orientation, establishing extensive interactions between its polybasic face and the flat bilayer, and bringing the SNARE four-helix bundle close to the flat bilayer (Figure 4A–D). This arrangement dictates that the Cpx1(27-72) helix points toward the vesicle membrane, in agreement with the proposal that binding of Syt1 to the SNARE complex through the primary interface supports the inhibitory activity of Syt1 and complexin-1 (Guan et al., 2017; Voleti et al., 2020). The Syt1 C2A domain adopted distinct orientations in the different primed complexes, consistent with the fact that no stable Ca2+-independent interactions of this domain with membranes or the SNARE complex have been identified. The C2B domain remained bound to the SNARE four-helix bundle via the primary interface in all four primed complexes throughout the simulation. The binding modes in the primed complexes resembled those observed in various crystal structures containing the primary interface (Zhou et al., 2015; Zhou et al., 2017), particularly in the so-called region I of this interface that includes Y338 among other side chains of C2B (e.g. Figure 4E). However, there were differences in the other region of this interface (region II), which includes R281, R398, and R399 of the C2B domain. In the crystal structures, there was variability in the contacts made by these side chains and R398 did not interact with acidic residues or was at moderate proximity with E238 of syntaxin-1 (Figure 4—figure supplement 1A, B). However, the R398 side chain interacted with a negative pocket formed by E55, D58 and E62 of SNAP-25 in the four primed complexes of our simulation (Figure 4F, Figure 4—figure supplement 1C–E). The findings that an R398Q mutation impairs binding of the C2B domain to the SNARE complex in vitro (Voleti et al., 2020) and disrupts neurotransmitter release in neurons (Xue et al., 2008) support the relevance of the interactions of R398 uncovered by our simulation and suggest that crystal packing might have slightly distorted the binding mode, but it is also plausible that the binding mode is dynamic in this area.

Figure 4 with 4 supplements see all
Additional views of the first MD simulation of primed complexes bridging a vesicle and a flat bilayer.

(A–D) Close-up views of the four primed complexes after 336 ns showing how the Syt1 C2B domain binds to the SNARE complex through the primary interface and to the flat bilayer with the polybasic face, which dictates that the Cpx1(27-72) helix is oriented toward the vesicle and bends in different ways and directions to avoid steric clashes. This arrangement forces the SNARE four-helix bundle to be close to the flat bilayer. The primed complexes are illustrated by ribbon diagrams, with the SNAREs in salmon, Cpx1(27-72) in yellow and the Syt1 C2AB fragment in cyan (C2A domain) and violet (C2B domain). The lipids are shown as thin stick models (carbon lime, oxygen red, nitrogen blue, phosphorous orange). (E–F) Two different close-up views of the primary interface between the C2B domain and the SNARE complex in PC1 after 336 ns showing site I of the interface (E) or site II where R398,R399 of the C2B domain are located (F). The C2B domain and the SNARE complex are illustrated by ribbon diagrams and stick models with oxygen atoms in red, nitrogen atoms in blue, sulfur atoms in light orange and carbon atoms in violet [for the C2B domain] or salmon (for the SNAREs). The positions of selected side chains are indicated. (G) Close-up view of the interaction of the C2B domain of PC1 with the flat bilayer after 336 ns. The positions of PIP2 headgroups, basic side chains involved in interactions with the lipids, and the hydrophobic side chain of I367 at the tip of a Ca2+-binding loop that inserts into the bilayer, are indicated. (H) Thin slice of the system showing a point-of-contact interface between the vesicle and the flat bilayer at 336 ns. Phosphorous atoms of phospholipids and the oxygen atoms of cholesterol molecules are shown as spheres to illustrate the approximate locations of lipid head groups. The positions of PC1 and PC3 are indicated.

In all four primed complexes, the extensive interactions of the C2B domain with the flat bilayer involved not only a polybasic sequence (residues 321–327) known to bind to PIP2 (Bai et al., 2003) but also other basic residues on this face of the β-sandwich that are also important for neurotransmitter release [e.g. K313, (Brewer et al., 2015; Figure 4G, Figure 4—figure supplement 1F-H)]. PIP2 molecules of the flat bilayer were often involved in these interactions. In addition, for all primed complexes, one of the C2B domain Ca2+-binding loops interacted extensively with the flat bilayer, inserting the hydrophobic residue at its tip (I367) into the acyl region. We also observed some interactions of the flat bilayer with basic residues of the SNARE four-helix bundle (e.g. R30, R31 from synaptobrevin and R176 from SNAP-25), which appeared to be favored because the clashes between the Cpx1(27-72) helix and the vesicle push the SNARE four-helix bundle and C2AB toward the flat bilayer. There was some variability in the four-helix bundle-flat bilayer interactions observed in the different primed complexes, but the overall arrangement of the C2B domain with respect to the flat bilayer and the SNARE four-helix bundle was very similar in all complexes, regardless of the orientation of the Cpx1(27-72) helix and the state of assembly of the SNARE four-helix bundle at the C-terminus (Figure 4A–D).

To further test the consistency of our results with respect to the configuration of the primed Syt1-SNARE-complexin-1 complex, we built a similar system but using different configurations of the four initial primed complexes and a slightly larger square bilayer to provide more space for protein-membrane interactions. The final system (Figure 4—figure supplement 2A, B) was used to run a production simulation of 310 ns. Figure 4—figure supplement 2C shows the final configuration. The behaviors of the primed complexes were similar to those of the previous simulation. The four initial four-helix bundles again had different levels of assembly at the C-terminus, with PC1 being the only one that was almost completely assembled, and there was not much progress toward full assembly in the other three complexes (Figure 4—figure supplement 2D-K). The Cpx1(27-72) helix again exhibited strong clashes with the vesicle and distinct ways to overcome such clashes, whereas the C2B domain changed orientation to establish extensive interactions with the flat bilayer while remaining bound to the SNARE four-helix bundle via the primary interface (Figure 4—figure supplement 2D, Figure 4—figure supplement 3A–D). It is also noteworthy that, in the two simulations of primed complexes, contacts between the vesicle and the plasma membrane were established at about 210–230 ns and the contacts increased gradually afterwards, but appeared to be leveling off at the end of the simulations (Figure 4—figure supplement 4), resulting in point-of-contact interfaces between the vesicle and the flat bilayer, without flattening of the vesicle (Figure 4H, Figure 4—figure supplement 3E).

Overall, the arrangements of the Syt1 C2B domain with respect to the flat bilayer and the SNARE four-helix bundle in the eight primed complexes from the two simulations were very similar, and in all cases dictated that the Cpx1(27-72) helix was oriented toward the vesicle (Figure 4A–D, Figure 4—figure supplement 3A–D). The consistency of these results, together with the abundant data available on the functional importance of the C2B-membrane, C2B-SNARE and Cpx1(27-72)-SNARE interfaces present in these complexes [e.g. Chen et al., 2002; Li et al., 2006; Zhou et al., 2015] suggest that these complexes resemble those present in the primed state of synaptic vesicles.

Simulation of the primed Synaptotagmin-1-SNARE-complexin-1 complex in the presence of Ca2+

Ca2+ binding to Syt1 is believed to induce a tight, PIP2-dependent interaction of the C2B domain with the plasma membrane and dissociation from the SNARE complex to relieve the inhibition of release caused by Syt1 and complexin-1 (Voleti et al., 2020). Based on the estimated kD of the interaction between the C2B domain and the SNARE complex [ca. 20 μM Voleti et al., 2020], the off rate for dissociation is expected to be at most 2000 Hz and hence too slow for the time scales reachable in our simulations. However, it is plausible that dissociation might be strongly accelerated by changes in the orientation of the C2B domain with respect to the membrane induced by Ca2+ (Voleti et al., 2020). To examine whether we could observe the dissociation step and investigate how the system evolves afterwards through MD simulations, we generated a system analogous to that used for our first simulation of primed complexes, but with the larger flat bilayer used for the second simulation of primed complexes to provide sufficient room for Ca2+-dependent binding to the C2 domains. We added five Ca2+ ions to the corresponding binding sites of C2A and C2B, and we removed the Cpx1(27-72) molecules to facilitate potential eventual fusion and to study at the same time how the system evolves without complexin-1 (Figure 5A).

Figure 5 with 2 supplements see all
MD simulation of C2AB bound to Ca2+ and to trans-SNARE complexes bridging a vesicle and a flat bilayer.

(A) Close-up view of the four C2AB-SNARE complexes in the initial system. (B) Close-up view of the system after a 439 ns MD simulation. (C–J) Close-up views of the assemblies between C2AB molecules (named C2AB1-4) and SNARE complexes (SC1-SC4) in the initial configuration (C,E,G,I) and after the 439 ns MD simulation (D,F,H,J). In all panels, the SNAREs are represented by ribbon diagrams in salmon and the Syt1 C2AB fragment by ribbon diagrams in cyan (C2A domain) and violet (C2B domain). Ca2+ ions are shown as yellow spheres. The lipids are shown as thin stick models (carbon lime, oxygen red, nitrogen blue, phosphorous orange).

We performed a production MD simulation of 439 ns, which led to the final configuration shown in Figure 5B. The C2AB-SNARE complexes generally behaved similarly to the primed complexes in the previous simulations, but with some differences. The SNARE four-helix bundle that was almost fully assembled in the starting configuration remained assembled almost completely (up to layer +7), whereas the other three complexes did not make much progress toward C-terminal assembly (Figure 5C–J). We note again that 439 ns are expected to provide ample time for helix formation and large conformational rearrangements, which is exemplified by the behavior of one of the SNARE four-helix bundles (SC4) during the simulation. Thus, the helix corresponding to the SNAP-25 C-terminal SNARE motif was almost fully formed after 5 ns, even though there was a substantial break in the helix in the beginning, and there were considerable structural changes at 75 ns, but only limited changes from 75 to 439 ns (Figure 5—figure supplement 1A–D). These findings again show that the constraints imposed by the system hinder fast assembly of the C-terminus of the SNARE four-helix bundle. Interestingly, the SNARE four-helix bundles exhibited less interactions with the flat bilayer than in the simulations of primed complexes including Cpx1(27-72), consistent with the notion that the steric clashes of the complexin-1 accessory helix with the vesicle push the SNARE four-helix bundle toward the flat bilayer. As observed in previous simulations, the C2B domains of the four complexes established extensive interactions with the flat bilayer and remained bound to the SNARE complex through the primary interface (Figure 5D, F, H and J). We did observe that SC4 became detached from R398,R399 of the C2B domain early in the simulation and there were additional interactions in region I of the primary interface that remained at the end of the simulation (Figure 5—figure supplement 1E, F). However, it is unclear whether this change was caused by Ca2+ binding to the C2B domain. These findings suggest that dissociation of the C2B domain from the SNAREs requires longer time scales and may be a rate limiting step in release, which is supported by the finding that an E295A/Y338W in the C2B domain primary interface enhances SNARE complex binding (Voleti et al., 2020) but disrupts Ca2+-evoked neurotransmitter release (Zhou et al., 2015). In this simulation, the vesicle came into contact with the flat bilayer at about 400 ns and the number contacts increased gradually afterwards but without reaching a plateau at the end (Figure 5—figure supplement 2).

Discussion

Enormous advances have been made to elucidate the molecular mechanisms underlying neurotransmitter release and have suggested that Syt1 and complexin are bound to trans-SNARE complexes in the primed state that renders synaptic vesicles ready for fast fusion upon Ca2+ influx. However, the configuration of the resulting macromolecular assembly is still unclear. Our MD simulations, together with previously available data, suggest that trans-SNARE complexes alone induce extended vesicle-plasma membrane contact interfaces that fuse slowly. Our results also indicate that binding of Syt1 and complexin-1 to trans-SNARE complexes in primed vesicles leads to a spring-loaded arrangement that hinders formation of such extended contact interfaces, keeping the system ready for fast fusion but at the same time hindering premature fusion before Ca2+ influx.

Our results need to be interpreted with caution because of the limited simulation times, the dependence of the results on the initial configurations and the absence of key elements of the release machinery. Nevertheless, multiple features observed in the simulations make sense from structural and energetic points of view, and are consistent with abundant experimental data available on this system. The SNARE four-helix bundle strongly drew the membranes together in our simulations with SNAREs alone (Figure 1A–D), in agreement with the high stability of the four-helix bundle [e.g. Gao et al., 2012]. However, assembly of the four-helix bundle brought the flat bilayers within a few nm from each other, and it is unclear how the SNAREs exert additional force on the membranes to induce fusion. A major problem with the widespread notion that the SNARE motif, juxtamembrane linker and TM region of synaptobrevin and syntaxin-1 form continuous helices that force fusion as they zipper from the N- to the C-terminus (Hanson et al., 1997a; Sutton et al., 1998; Weber et al., 1998) is that the bent conformations of the linkers envisioned in these models are unrealistic from an energetic point of view. Although optical tweezer data suggested that interactions between the juxtamembrane linkers contribute to exerting force on the membranes to induce fusion (Gao et al., 2012) helix continuity in the linkers is not required for neurotransmitter release (Kesavan et al., 2007; Zhou et al., 2013). Substantial release was observed even upon insertion of a five-residue sequence into the synaptobrevin linker (Kesavan et al., 2007) despite the fact that this sequence should break the register of linker-linker interactions and contained two (helix disrupting) glycine residues. Note also that the optical tweezer data were obtained in the absence of membranes and that, in vivo, the linkers are likely to interact with the lipids given the proximity of each linker to the adjacent membrane and the abundance of basic residues in the linker sequence (and aromatic residues in the case of synaptobrevin). The extensive interactions of the linkers with the membranes observed in all our simulations support this prediction.

Based on conformational grounds, it is not surprising that the juxtamembrane linkers became unstructured during the simulation that we performed to generate a trans-SNARE complex starting from the crystal structure of the cis-SNARE complex. Note however that we did not perform a systematic analysis to examine the range of linker structures that are compatible with the geometry of a trans-SNARE complex. In any case, the linkers are expected to be unstructured before SNARE complex assembly, which is supported by EPR data (Kim et al., 2002). Therefore, configurations with unstructured linkers were natural, unbiased starting points for the simulations, and helical conformation could be adopted by the linkers during the simulations if they were preferred. However, the linkers remained unstructured during all our simulations, which facilitated the extensive linker-membrane interactions observed. These interactions may have contributed to pull the two flat bilayers together in our first simulation (Figure 1A–D) and to induce the formation of an extended contact interface in the simulation with a vesicle and a flat bilayer (Figure 2J). Coarse-grained simulations have suggested that formation of such interfaces can also arise from entropic forces that favor outward movement of the SNARE complexes, away from the center of the interface (Mostafavi et al., 2017). These extended interfaces have been observed by cryo-EM and by fluorescence microscopy, and evolve to fusion in long time scales (seconds-minutes) (Diao et al., 2012; Hernandez et al., 2012; Witkowska et al., 2021), in agreement with data showing that liposome fusion occurs minutes after liposome docking (Cypionka et al., 2009). SNARE-mediated fusion was also observed at faster time scales (Domanska et al., 2009; Heo et al., 2021). Thus, it is plausible that fusion is slower under conditions that favor formation of extended interfaces. Interestingly, cryo-EM studies indicated that formation of such extended interfaces is hindered by other proteins involved in Ca2+-evoked release, which favor point-of-contact interfaces that fuse faster (Diao et al., 2012; Gipson et al., 2017).

Our two simulations including Syt1 C2AB and the complexin-1 (27–72) fragment are consistent with this proposal, as these proteins appeared to hinder formation of extended interfaces (Figure 4—figure supplement 4), but longer simulations will be required to further test this notion. The initial configurations used in these simulations (Figure 3A, Figure 4—figure supplement 2A) were built to mimic and investigate potential states of the Syt1-SNARE-complexin-1 macromolecular assemblies that are likely central components of the primed state of synaptic vesicles. Overwhelming evidence supports the physiological relevance of the binding modes of the SNARE complex to complexin-1 and to the primary interface of the Syt1 C2B domain used to build these initial configurations (Chen et al., 2002; Guan et al., 2017; Xue et al., 2007; Zhou et al., 2015). These binding modes were largely preserved in the eight primed complexes during the two simulations, although there were slight rearrangements in the primary interface that allow closer interactions of the critical R398 side chain of the C2B domain (Xue et al., 2008) with acidic residues from the SNARE complex (Figure 4F).

In the initial configurations, we placed the C2B domain between the SNARE complex and the flat bilayer to facilitate interactions of a polybasic β-strand (residues 321–327) with PIP2 that are believed to mediate binding of the C2B domain to the plasma membrane (Bai et al., 2003). Interestingly, each of the eight primed complexes was re-oriented during the two simulations to enable extensive interactions of the C2B domain with the flat bilayer that involve not only the polybasic β-strand but also other basic residues from the same side of the β-sandwich that are also known to play a key role in release [e.g. K313; Brewer et al., 2015; Figure 4G, Figure 4—figure supplement 1F–H, Figure 4A-D]. Thus, the resulting C2B domain-flat bilayer binding mode is consistent with the physiological importance of multiple residues of the polybasic face (Brewer et al., 2015). This binding mode resembles that observed in previous all-atom simulations of a C2AB-SNARE-complexin-1 assembly on a flat bilayer (Bykhovskaia, 2021) and makes sense from a physicochemical point of view given the numerous electrostatic interactions involved. It is also worth noting that the extent of C-terminal assembly of the SNARE four-helix bundle was distinct in each of the eight complexes. Thus, the orientation of the C2B domain-SNARE-Cpx1(27-72) assembly with respect to the membranes consistently observed in the eight complexes can be accommodated by different extents of C-terminal assembly. All these observations suggest that the C2B domain binds to the plasma membrane in a similar mode in the primed state of synaptic vesicles and that some of the configurations of the complexes visited in the simulations resemble those present in the primed state.

Due to the C2B-flat bilayer interactions, the SNARE four-helix bundle came near the flat bilayer, with the C2B domain on one side rather than between the bundle and the flat bilayer, and with the helix of Cpx1(27-72) oriented toward the vesicle (Figure 4A–D, Figure 4—figure supplement Figure 4—figure supplements 2 and 3A–D). This orientation dictates that formation of a continuous, straight helix by Cpx1(27-72) leads to steric clashes with the vesicle if the SNARE four-helix bundle is fully or close to fully zippered, consistent with the proposal that such steric clashes underlie the inhibitory role of the accessory helix and with physiological data supporting this notion (Trimbuch et al., 2014). These steric clashes most likely cause, at least to some extent, the distinct ways by which the Cpx1(27-72) helix was bent in the different complexes (Figures 3D–K4A–D, Figure 4—figure supplement 3A-D). These variations may also arise in part from random motions and are consistent with data suggesting that the accessory helix can adopt additional orientations that depend on the extent of C-terminal zippering (Choi et al., 2016; Kümmel et al., 2011). However, NMR data showed that the accessory helix sequence has a high propensity to form α-helical conformation that nucleates the central helix and that structures with a continuous, straight helix as observed by X-ray crystallography are substantially populated when there is no steric hindrance (Chen et al., 2002; Pabst et al., 2000; Radoff et al., 2014). Overall, these observations show that, even if there are natural motions in the complexin-1 helix, transient formation of a continuous straight helix results in steric clashes with the vesicle if the four-helix bundle zippers fully.

It is difficult to estimate the energy barrier imposed by this behavior on SNARE zippering, but it is known that the energy cost of breaking just one hydrogen bond to distort the complexin-1 helix can be about 1.8 kBT (Nick Pace et al., 2014), which could translate to 7.2 kBT by cooperative action of four primed complexes. Such an energy barrier would slow down SNARE zippering by a factor of 1340 (=e^7.2). For comparison, the enhancements of spontaneous neurotransmitter release observed upon deletion of complexin range from 3 to >20 (Huntwork and Littleton, 2007; Martin et al., 2011). Thus, even a small energy barrier caused by steric hindrance can account for these enhancements. Note also that we observed occasional interactions of the Cpx1(27-72) accessory helix with C-terminal residues of the synaptobrevin and SNAP-25 SNARE motifs that are favored by proximity (e.g. Figure 3G, Figure 3—figure supplement 2G), and recent cross-linking experiments suggested that very weak interactions of the complexin accessory helix with synaptobrevin and SNAP-25 hinder C-terminal zippering and release (Malsam et al., 2020). It is unlikely that specific interactions underlie the complexin inhibitory function, as this function was retained when the accessory helix was replaced with an unrelated helical sequence, and helix propensity appears to be the key determinant for the inhibitory function (Radoff et al., 2014). However, very weak complexin-SNARE interactions that do not need to be specific may slow down C-terminal SNARE zippering and thus contribute also to inhibition of release by the complex accessory helix.

While all these observations support the proposal that the consistent overall configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly observed for the eight primed complexes in the two simulations resembles that present in primed vesicles, there are clear uncertainties with regard to the extent of C-terminal SNARE zippering and the vesicle-plasma membrane distance, which are closely related. In our two simulations, the vesicle was drawn into contact with the flat bilayer. However, other factors in addition to complexin are likely to hinder zippering in vivo, most notably Munc13-1 because in the absence of Ca2+ it bridges membranes in approximately perpendicular orientations that keep the membranes apart (Camacho et al., 2021; Grushin et al., 2022; Quade et al., 2019). Analyses by high-pressure freezing electron tomography (ET) (Imig et al., 2014) and cryo-ET (Radhakrishnan et al., 2021) showed that docked vesicles exhibit a distribution of distances from the plasma membrane that range from 0 to several nm, and a density map built from a subset of subtomograms revealed a distance of 3.5 nm. However, a recent model that can explain a large amount of available presynaptic plasticity data invoked two primed states, one that involves partially assembled SNARE complexes and has low release probability (loose state), and another with more fully assembled SNARE complexes that has a much higher release probability (tight state; Neher and Brose, 2018). Hence, it is plausible that the vesicles that are closest to the plasma membrane account for much of the neurotransmitter release observed.

Based on these observations and our simulations, we envision two scenarios. In one scenario, vesicles in the tight primed state are very close to or in contact with the plasma membrane and include primed complexes that resemble PC1 at the end of each of our two simulations (Figure 3E, Figure 4—figure supplement 2E). In both of these poses, the four-helix bundle was almost fully zippered, but the vesicle was not as close to the flat bilayer around the C-terminus of the four-helix bundle as in the simulation with only SNAREs (Figure 6A–C), which led to an extended vesicle-flat bilayer interface (Figure 2J). Since complexin-1 binding stabilizes the C-terminus of the SNARE complex (Chen et al., 2002), it is tempting to speculate that, to maximize the speed of release, complexin-1 favors formation of an almost fully zippered, tight primed state that is ready for fast Ca2+-triggered fusion, but hinders the ‘final pull’ of the SNARE four-helix bundle to bring the membranes together, preventing premature fusion as well as formation of an extended interface. In this spring-loaded model, dissociation of the C2B domain from the SNARE complex upon Ca2+ influx (Voleti et al., 2020) allows rotations of the four-helix bundle that eliminate the steric clashes and facilitate cooperation between the Syt1 C2 domains and the SNAREs in triggering fusion by a mechanism that remains unclear. Hence, this model can explain the finding that complexins are required for the dominant negative effect of mutations in the Ca2+-binding site of the Syt1 C2B domain (Zhou et al., 2017). In the second scenario, the distance from the vesicle to the plasma membrane in the tight primed state is a few nm. Manually moving the vesicle in the pose of Figure 6B to a distance of 3 nm from the flat bilayer (Figure 6D) shows that the same spring-loaded configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly could be kept by stretching the synaptobrevin juxtamembrane linker, and the Cpx1(27-72) helix would still hinder progress toward final zippering and fusion. This configuration is also largely compatible with longer vesicle-flat bilayer distances if the linker is stretched further and/or there is partial SNARE unzippering. In this case, steric clashes of the Cpx1(27-72) helix with the vesicle might be alleviated or eliminated, but they would occur as soon there is full zippering and hence would still hinder vesicle fusion.

Complexin-1 may hinder the final action of trans-SNARE complexes to bring membranes together.

(A–C) Close-up views of one of the SNARE complexes bridging a vesicle and a flat bilayer after simulation for 280 ns at 310 K (A) (shown in Figure 2D after 520 ns at 310 K and 454 ns at 325 K), of PC1 in the first MD simulation of primed complexes after 336 ns (B) (also shown in Figure 3E) and of PC1 in the second MD simulation of primed complexes after 310 ns (C) (also shown in Figure 4—figure supplement 2E). The complexes are illustrated by ribbon diagrams, with the SNAREs in salmon, Cpx1(27-72) in yellow and the Syt1 C2AB fragment in cyan (C2A domain) and violet (C2B domain). The positions of the syntaxin-1 juxtamembrane linkers (sxl) are indicated. The comparison shows how the SNARE complex with the fully assembled four-helix bundle in (A) drew the two membranes closer than the two primed complexes of (B,C). (D) Close up view of the pose shown in (B) after moving the vesicle upwards so that it’s minimal distance from the flat bilayer is 3 nm.

Clearly, these models will need to be tested with further MD simulations and experimentation, and there are multiple factors of the primed state of synaptic vesicles that were not included in the simulations. As mentioned above, particularly important for this state is Munc13-1, but other missing elements include Munc18-1, the N-terminal region of syntaxin-1, the SNAP-25 linker joining its SNARE motifs, the N- and C-terminal regions of complexin-1 and the TM region plus linker sequence of Syt1 [reviewed in Rizo, 2022]. Moreover, the Syt1 C2B domain can also bind to the SNARE complex through a so-called tripartite interface whereby an α-helix of the C2B domain is adjacent to the complexin-1 helix (Zhou et al., 2017), which could stabilize the complexin-1-SNARE interface to hinder rotations that might lead to dissociation (see Figure 3K, Figure 3—figure supplement 2F). There is also evidence that Syt1 forms oligomeric rings that hinder spontaneous neurotransmitter release (Tagliatti et al., 2020), and it is plausible that the SNAREs alone can induce fast fusion in different configurations that we did not study. Furthermore, some evidence suggested many years ago that direct binding of Ca2+ to phospholipids could trigger synaptic vesicle fusion (Papahadjopoulos et al., 1976) and, although it is now generally believed that Ca2+ triggers neurotransmitter release by binding to Syt1, it is plausible that Ca2+-phospholipid interactions might also contribute to trigger membrane fusion. Addressing all these issues with all-atom MD simulations will be challenging because of the limited simulation times that are currently reachable. Continuum and coarse-grained simulations, which have already provided important insights into SNARE-mediated membrane fusion (Fortoul et al., 2015; Manca et al., 2019; McDargh et al., 2018; Mostafavi et al., 2017; Risselada et al., 2011; Sharma and Lindau, 2018), offer the opportunity to explore much longer time scales with similar systems. Hence, a marriage of approaches whereby the most interesting results obtained by such simulations are investigated in further detail by all-atom simulations will likely provide a powerful strategy to unravel the intricate mechanisms that govern fast Ca2+-triggered membrane fusion. The systems that we present here provide a framework to pursue these studies and gradually incorporate additional elements of the neurotransmitter release machinery.

Methods

High-performance computing

Most high-performance computing, including all production MD simulations, were performed using Gromacs (Pronk et al., 2013; Van Der Spoel et al., 2005) with the CHARMM36 force field (Huang et al., 2017; Klauda et al., 2010; Lee et al., 2019; Wu et al., 2014a; Wu et al., 2014b) on Frontera at TACC. Some of the initial setup tests, solvation, ion addition, minimizations, and equilibration steps were performed using Gromacs with the CHARMM36 force field at the BioHPC supercomputing facility of UT Southwestern, or on Lonestar5 or Stampede2 at TACC. System visualization and manual manipulation were performed with Pymol (Schrödinger, LLC).

System setup

All systems were built by manually combining coordinates of the protein components with coordinates of the membranes, solvating the system with explicit water molecules (TIP3P model) and adding potassium and chloride ions as needed to reach a concentration of 145 mM and make the system neutral. Flat lipid bilayers were built with the Membrane Builder module (Jo et al., 2007; Jo et al., 2009) in the CHARMM-GUI (Jo et al., 2008) website (https://charmm-gui.org/), providing the coordinates of the TM region of synaptobrevin or syntaxin-1 in their desired positions as input. The bilayers contained mixtures of cholesterol (CHL1), 16:0-18:1 phosphatidylcholine (POPC), 18:0-22:6 phosphatidyltethanolamine (SDPE), 18:0-22:4 phosphatidyltethanolamine (SAPE), 18:0-18:1 phosphatidylserine (SOPS), 18:0-22:6 phosphatidylserine (SDPS), 18:0-20:4 phosphatidylinositol 4,5-bisphosphate (SAPI2D) and/or 18:0-20:4 glycerol (SAGL). Table 1 list the number of atoms and lipid compositions of the membranes; each entry corresponds to a system denoted by the abbreviations described below, which were used as roots for the filenames of the corresponding simulations.

Four trans-SNARE complexes bridging two flat bilayers (qscff system). The starting point to generate a trans-SNARE complex was the crystal structure of the neuronal SNARE complex that included the TM regions of synaptobrevin and syntaxin-1 (Stein et al., 2009). Two residues at the C-terminus of syntaxin-1 (residues 287–288) were added in Pymol, and four residues of the C-terminus of SNAP-25 (residues 201–204) were added manually based on a crystal structure of soluble SNARE complex (PDB accession code 1NS7). The resulting complex included residues 30–116 of synaptobrevin, residues 189–288 of syntaxin-1, and residues 8–82 and 141–204 of SNAP-25. To move the TM regions of synaptobrevin and syntaxin-1 to designed positions where they were later inserted into the flat lipid bilayers (Figure 1—figure supplement 1B), the cis-SNARE complex was solvated with explicit water molecules and energy minimized. Then a 1 ns production MD simulation at 310 K was performed imposing position restraints to keep of all heavy atoms of the N-terminal half of the SNARE complex, up to the polar layer (residues 30–56 of synaptobrevin, residues 189–226 of syntaxin-1, and residues 8–53 and 141–174 of SNAP-25), in their original coordinates (force constant 1,000 kJ/mol/nm2), and to force the backbone atoms of the TM regions (residues 95–116 of synaptobrevin and residues 266–288 of syntaxin-1) to move to the designed positions (force constant 300 kJ/mol/nm2). Four copies of the final structure were rotated and translated to desired positions (Figure 1—figure supplement 1C), and merged with two square flat bilayers of 26 × 26 nm2 each, separated by 5 nm. The size of the bilayers was designed to provide space for the SNAREs to bend the membranes and induce fusion while limiting the overall size of the system.

Four trans-SNARE complexes bridging two flat bilayers including four Ca2+-bound C2AB molecules (Sqscff system). To generate C2AB molecules for this simulation, the C2AB molecule from a complex with the SNAREs (PDB accession number 5CCH) (Zhou et al., 2015) was used as a starting point. Ca2+ ions were added at the corresponding sites as observed in the solution NMR structures of the C2A and C2B domains (PDB accession codes 1BYN and 1K5W) (Fernandez et al., 2001; Shao et al., 1998). After solvation with explicit water molecules and energy minimization, a 10 ns production MD simulation was carried out using position restraints to keep the initial coordinates of the C2B domain and move the C2A domain to a designed location so that the Ca2+-binding loops of both C2 domains point in similar directions (Figure 1—figure supplement 1A) and hence can bind in a Ca2+-dependent manner to the same membrane (force constant 1,000 kJ/mol/nm2). The final C2AB structure was energy minimized and four copies of it were incorporated into the system containing four trans-SNARE complexes bridging two flat bilayers (qscff), interspersed between the SNARE complexes but without contacting them.

Four trans-SNARE complexes bridging a vesicle and a flat bilayer (qscv system). The vesicle was built by adaptation of the scripts for building coarse-grained vesicle systems from CHARMM-GUI Martini Maker (Qi et al., 2015). The radius of the vesicle was set to 11 nm and the number of lipids in the inner and outer layer of the vesicle, given the specific vesicle radius and lipid ratio in each layer, were calculated using the same scheme as in Martini Maker. As the final system was too large for long-time equilibration of the lipids in the inner and outer layer using water pores along the x, y and z axis, no water pore was created in the vesicle (i.e. water pore radius was set to 0 nm). The four trans-SNARE complexes were slightly modified with respect to those used for the qscff system to tilt the synaptobrevin TM regions such that they were perpendicular to the vesicle surface, and to tilt the SNARE four-helix bundles such that their long axis had similar angles with respect to the vesicle and the flat bilayer (Figure 2B, Figure 2—figure supplement 1A). The trans-SNARE complex built for the qscff system was used, after solvation, minimization and equilibration, as a starting point for a 2 ns production MD simulation imposing position restraints to keep all heavy atoms of the N-terminal half of the SNARE complex in their original designed coordinates and to force the backbone atoms of the TM regions of synaptobrevin and syntaxin-1 to move to their designed positions (force constant 1,000 kJ/mol/nm2). Four copies of the final structure were rotated and translated to designed positions (Figure 2—figure supplement 1A), and merged with the vesicle and a square flat bilayer of 26 × 26 nm2 each. After solvation and equilibration, we performed a 7 ns production MD simulation and observed the appearance of holes in the vesicle that arose because the lipid density was not optimal. The holes were filled manually with lipid patches from the original vesicle and a 5 ns production MD simulation was carried out with position restraints to keep the SNAREs in their initial locations. New holes appeared and were filled manually again. After another 10 ns production MD simulation with position restraints on the SNARE coordinates, no additional holes appeared. A final 80 ns production MD simulation with position restraints on the SNAREs was performed to equilibrate the vesicle lipids, which yielded the initial equilibrated system (Figure 2—figure supplement 1B,C) that was used to initiate an unrestrained production MD simulation.

First simulation of primed complexes bridging a vesicle and a flat bilayer (prsg system). The complexin-1 fragment [Cpx1(27-72)] was built starting from the crystal structure of Cpx1(26-83) bound to the SNARE complex (PDB accession code 1KIL) (Chen et al., 2002), which contained electron density for residues 32–73 of complexin-1. Five additional N-terminal residues that may be important for the steric clashes of complexin-1 with the vesicle, which were proposed to underlie the inhibitory activity of the accessory helix (Trimbuch et al., 2014), were added in a random conformation with Pymol. The initial conformation of the C2AB fragment was generated starting from the coordinates of C2AB used for the Sqscff system and, after solvation, minimization and equilibration, a 5 ns production MD simulation was performed with position restraints to keep the C2B domain at its original position (force constant 1,000 kJ/mol/nm2) and additional position restraints to move the C2A domain so that its Ca2+-binding loops can readily interact with the flat bilayer while the C2B domain binds to the flat bilayer through the polybasic face (force constant 100 kJ/mol/nm2).

To create the initial configurations of the primed complexes, we designated four positions to place the SNARE four-helix bundles bound to Cpx1(27-72) and to the Syt1 C2AB through the primary interface as determined by crystallography (Chen et al., 2002; Zhou et al., 2015). The position of the C2B domain was designed such that the polybasic region was placed close to but not contacting the flat bilayer to limit the bias introduced by the initial configurations of the primed complexes. The four trans-SNARE complexes generated for the qscv system were used as starting point to generate the four primed complexes. Four Cpx1(27-72) molecules were rotated and translated to interact each with a corresponding trans-SNARE complex based on the crystal structure (Chen et al., 2002), while four C2AB molecules were placed at their final designed places. The system was solvated, minimized and equilibrated, and a 1 ns MD simulation was carried out with the following position restraints: (i) strong restraints (force constant 4000 kJ/mol/nm2) to keep the synaptobrevin TM regions in their initial positions, as they were intended to remain inserted in the same positions in the vesicle; (ii) mild position restraints (force constant 100 kJ/mol/nm2) to keep the C2AB molecules at their initial (designed) positions; (iii) mild position restraints (force constant 100 kJ/mol/nm2) on the syntaxin-1 TM regions to move them to their intended designed positions in the flat bilayer, which we planned to place further from the vesicle that in the qscv system; and (iv) mild position restraints (force constant 100 kJ/mol/nm2) on the N-terminal half of the SNARE four-helix bundle and on Cpx1(27-72) to move them to their designed positions so that each SNARE complex interacted with a corresponding C2AB molecule through the primary interface as in the crystal structure (Zhou et al., 2015). Note that we did not include any position restraints on the C-terminal half of the SNARE four-helix bundle and the juxtamembrane regions to allow them to adapt to the imposed restraints. We built a square flat bilayer of 31.5 × 31.5 nm2 to provide sufficient space for interactions with the proteins and placed it 2.3 nm below the equilibrated vesicle from the qscv system. The flat bilayer and the vesicle were then merged with the four primed complexes to provide the starting point for the simulation.

Second simulation of primed complexes bridging a vesicle and a flat bilayer (prs2 system). The four primed complexes were generated using the initial primed complexes from the prsg system (Figure 3—figure supplement 1) and running a 50 ns MD simulation with the same restraints used to create the initial complexes. The four resulting primed complexes were merged with the equilibrated vesicle from the qscv system and with a flat bilayer of 35 × 35 nm2 that was built slightly larger than that used for the prsg system to provide more space for protein-membrane interactions.

Four C2AB molecules bound to Ca2+ and to four trans-SNARE complexes bridging a vesicle and a flat bilayer (prsncpxca system). The Cpx1(27-72) molecules were removed from the four initial primed complexes of the prsg system and five Ca2+ ions were added to the corresponding binding sites of each C2AB molecule. The resulting complexes were merged with the equilibrated vesicle from the qscv system and the same flat bilayer of 35 × 35 nm2 used for the prs2 system.

MD simulations

After energy minimization, all systems were heated to 310 K over the course of a 1 ns MD simulation in the NVT ensemble and equilibrated for 1 ns in the NPT ensemble using isotropic Parrinello-Rahman pressure coupling (Parrinello and Rahman, 1981). NPT production MD simulations were performed for the times indicated in for each system using 2 fs steps, isotropic Parrinello-Rahman pressure coupling and a 1.1 nm cutoff for non-bonding interactions. All simulations were performed at 310 K except one simulation with the qscff system, which was performed at 325 K after a 310 K simulation. Nose-Hoover temperature coupling (Hoover, 1985) was used separately for three groups: (i) protein atoms plus Ca2+ ions if present; (ii) lipid atoms; and (ii) water and KCL. Periodic boundary conditions were imposed with Particle Mesh Ewald (PME) (Darden et al., 1993) summation for long-range electrostatics. The speeds of the production simulations ran on Frontera at TACC are indicated in Table 1.

Analysis of vesicle-flat bilayer contacts

Because the vesicle and the flat bilayers of the different systems contain large numbers of atoms, it was impractical to analyze vesicle-flat bilayer contacts through measurement of the distances between all atoms of the vesicle and all the atoms of the flat bilayer in multiple time frames of a trajectory. To limit the calculations, we selected only oxygen atoms from frames taken at 1 ns steps of each trajectory and measured the distances between all the oxygen atoms of the vesicle and all the oxygen atoms of the flat bilayer in each frame. The number of vesicle-flat bilayer contacts in each frame was defined as the number of oxygen-oxygen distances below 1 nm.

Data availability

Most files corresponding to our molecular dynamics simulations are available in the dryad database (doi:10.5061/dryad.ns1rn8pw6). Because of the very large size of trajectory files, it was not practical to deposit them in this database, but these files are available from the corresponding author upon reasonable request.

The following data sets were generated
    1. Rizo J
    (2022) Dryad Digital Repository
    Data from: All-atom molecular dynamics simulations of Synaptotagmin-SNARE-complexin complexes bridging a vesicle and a flat lipid bilayer.
    https://doi.org/10.5061/dryad.ns1rn8pw6

References

    1. Darden T
    2. York D
    3. Pedersen L
    (1993)
    Particle mesh Ewald: an N·log(N) method for Ewald sums in large systems
    The Journal of Chemical Physics 98:10089–10092.

Decision letter

  1. Frederic Pincet
    Reviewing Editor; Ecole Normal Superieure, France
  2. Vivek Malhotra
    Senior Editor; The Barcelona Institute of Science and Technology, Spain
  3. Ben O'Shaughnessey
    Reviewer

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "All-atom molecular dynamics simulations of synaptic vesicle fusion I: a glimpse at the primed state" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Vivek Malhotra as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Ben O'Shaughnessey (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) Manuscript readability

a. the manuscript should be shortened. This notably applies to the Results section.

b. the manuscript should better focus on the relevant conclusions relative to the SNAREs (degree of zippering, juxtamembrane linker – lipid interactions), synaptotagmin and complexin. These modifications will be an opportunity to discuss the results in view of other simulations, i.e. coarse grain approaches accessing fusion timescales.

2) Limitations of all atom MD simulations

a. At the molecular level, the assumptions regarding the initial arrangement of the proteins and the missing aspects (e.g. Munc13, Munc18, Syt linkers, Cpx-SNARE interactions) must be explicitly stated and discussed in view of the current knowledge in the field.

b. Realistic statements about likely fusion times need to be compared to the all-atom simulation times.

3) "Primed" state

a. The term "primed", used in the title and in the manuscript is misleading because other core synaptic proteins are not included in the simulations.

b. It is difficult to assess whether the vesicle in the proposed molecular arrangement is actually primed. On the contrary, given the narrow intermembrane distance with molecular contacts, it is very likely the membranes will ultimately fuse. All-atom simulations cannot reach the relevant time scales to be conclusive.

c. The Cpx accessory helix looks like a wobbly finger unlikely to support much force.

4) Calcium addition

a. Calcium-phospholipid may play an important part in the molecular arrangement and the fusion process. This is ignored here and should be addressed.

b. How does one reconcile that the aliphatic loops on Synaptotagmin C2B domain do not insert into the membrane upon calcium binding as observed in previous structural/functional studies?

Reviewer #1 (Recommendations for the authors):

1) It is difficult to assess if the results from simulations are a true representative of the biological process or an outcome of the initial condition/constraints chosen. For example, it is puzzling that there is no intra-molecular assembly of the SNAREpins during the simulation even though the coil-coil interactions are expected to occur in the simulation time scales. It appears, in almost all cases, that the SNARE zippering is unaltered at the end of the simulation. Also, it might be possible that the authors' choice to model the juxtamembrane region as a fully unstructured region prevents membrane fusion under the current simulation conditions. While it is not clear if the SNAREs zippering extends through the juxtamembrane (JM) region into the transmembrane region as observed in the crystal structure of full-length SNARE (Stein et al. Nature 2009), it stands to reason the JM region needed to be at least partially structured for effective force transfer to catalyze merging of the bilayers.

2) A major conclusion of the report is that the steric clash between Complexin accessory helix and vesicle serves as the fusion clamp and indeed drives the positioning of the SNARE and Synaptotagmin on the planar bilayer. However, there are a couple of factors that might alleviate or even mitigate this steric clash: (i) the vesicle and bilayer are positioned at ~2.3 nm apart at the beginning of the simulation. However, high-resolution cyroEM analysis in synaptosomes/cultured neurons (Fernandez-Busnadiego, R. J Cell Biol 2013; Radhakrishnan et al. PNAS 2021) show that the inter-bilayer distance of docked/primed vesicle is ~4.5 nm. Thus, it might be imperative to carry out the simulation with the physiological accurate inter-bilayer distance (ii) Complexin molecule has been positioned on SNAREs assuming a fully-zippered SNARE complex. However, there is sufficient evidence that SNAREs are likely only partially-assembled in an RPP vesicle (Hua & Charlton, Nat Neurosci 1999, Prashad & Charlton, PLoS One, 2014), and the positioning of the CPX, esp. the accessory helix is correlated to the extent of SNARE assembly (Choi et al. ELife 2016, Kummel et al. Nat Struct Biol 2011; Zhou et al. Nature 2017). Furthermore, accessory helix has been shown to interact with c-terminal ends of t- and v-SNARE molecules (Kummel et al. Nat Struct Mol Biol 2011; Malsam et al., Cell Reports 2020). Thus, it is possible that the alternate positioning of the accessory helix and other interactions might reduce the observed steric clash.

3) How does one reconcile that the aliphatic loops on Synaptotagmin C2B domain do not insert into the membrane upon calcium binding as observed in previous structural/functional studies (Grushin et al. Nat Comms 2019; Kuo et al. J Mol Biol 2009) even though synaptotagmin interacts with the membrane, including partial insertion of the C2B aliphatic loop, under calcium-free conditions. This is a rather crucial and missing piece considering that calcium-triggered membrane insertion is predicted to be the driving force for triggered fusion.

Reviewer #2 (Recommendations for the authors):

1. The authors' major conclusion is that the AA simulations support the model of Voleti et al. for the organization that clamps fusion in the pre-ca primed state. However, from Figures 3, 4 (and associated figure supplements) fusion seems very likely not to be clamped, given the vesicle contacts the planar membrane (the degree of contact is still growing at the end of the simulation, Figure 4 supp 4). As stated in (lines 420-425) the vesicle membrane is not flattened. This indicates a lower force than with SNAREs alone, but seems unlikely to block fusion. Due to running time limitations, AA simulations cannot test if fusion would occur in a physiological time. The structure does not keep the membranes apart, as it rotates and permits contact. The authors are clear about this – indeed, to predict the orientation is stated as a major objective. But the conclusions of lines 365-367 and the final sentence of the abstract, suggesting these results demonstrate a fusion clamp, seems unjustified as far as I can see. The emphasis on the cpx accessory helix role also appears somewhat exaggerated, as if on its own it provides a mini-buttress that separates vesicle and planar membrane. It's hard for me to imagine it supports much force in this configuration.

2. The simulations with bound calcium (final section of Results) seem inconclusive. The number of contacts is still growing at the end of the simulation, and we cannot know if the C2B will ever dissociate from the SNAREs. It's very reasonable to try this simulation but given the outcome I'm not sure a long section is merited, particularly with the tentative title "Potential effects of Ca2plus binding to synaptotagmin-1." This negative, albeit interesting finding, might be briefly summarized in the main text.

3. The manuscript would be strengthened by a more balanced presentation acknowledging the limitations of AA simulations (while of course still extolling their merits) and connecting to some degree with analysis on other scales, including coarse-grained approaches beyond MARTINI. SNARE-mediated fusion was studied using ultra coarse-grained (Mostafavi et al., 2017; McDargh et al., 2018) and even continuum (Manca et al., 2019) representations. Every approach has strengths and weaknesses. AA approaches scrutinize local issues as no others can, but presently they are remote from being able to demonstrate hemifusion, fusion, unclamping and ca-evoked fusion. Making matters worse, NT release is clearly stochastic, so multiple runs are needed for each condition. These limitations are apparent in this study: almost every conclusion comes with a caveat related to running time. In previous seminal MARTINI studies that achieved fusion (Risselada, Sharma and Lindau) the conditions were intentionally biased for fusion (vesicle size, lipid composition, temperature, helical LDs) or nanodiscs were used. In (Risselada, 2011) no fusion was observed when the LD was made unstructured.

In the Introduction the authors assert that experiments suggest "..the fusion step occurs in just a few microseconds," which timescales AA simulations may be able to access. They quote the 60 microsec delay times (ca influx to first sign of the excitatory post synaptic current, EPSC) reported by Sabatini and Regehr at 38 degrees. However, 0.5 -2 ms is much more typical in the literature (admittedly, the 38 degrees study is distinguished by the temperature being physiological.) Related, long-ago Katz argued other processes (e.g. NT diffusion across the synaptic cleft) are much faster than NT release (Katz and Miledi, 1965).

4. In simulations with SNAREs only, the SNARE complexes are cleared laterally, and the membranes are squashed together, generating an ECZ (extended contact zone, a flat portion of vesicle), Figures 1C, 1F. This is precisely the behavior seen in highly coarse-grained simulations (Mostafavi et al., 2017, Mcdargh et al., 2018), where entropic SNARE-SNARE and SNARE-membrane forces cleared the fusion site and pressed the vesicles together (those studies used undeformable membrane surfaces, so no vesicle flattening occurred). The entropic forces were predicted to provoke fusion after a time of order msec, with faster fusion for more SNAREs. These coarse-grained simulations and their relation to the present findings should be discussed.

The authors suggest the pressing together of the membranes is caused by binding of the LDs to the vesicle membrane (lines 262-266). This does not seem a plausible alternative to the proposed entropic forces, as LD-membrane adhesion would not favor the SNAREs being pushed outwards as far as I can see.

The authors argue that the ECZ in the SNARE-only simulations suggests SNAREs alone cannot fuse membranes rapidly, since fusion was slow in Hernandez et al., 2012 and Witkowska et al., 2021 where ECZs were seen. However, in those in vitro studies many other processes preceded fusion (SNARE assembly, docking etc) and micron scale GUVs were used by Witkowska et al.

5. A concern is the presentation, whose clarity would benefit from a more concise text. It is laudable to convey the details (computational papers where readers cannot tell what was done are frustrating), but many passages are long repeats of previous passages. For example, opening paragraphs of sections in Results often repeat descriptions of simulations in previous sections at great length, then specifying what was different in the current section. These could be massively shortcut. Short summaries in the main text, with details left to Methods or Supplementary materials, would be more digestible for readers.

6. This paper describes many current hot issues in the field, a great service. The figures are very nice but would be helped by a simple visual key to identify β sandwiches, the polybasic face, ca-binding loops, etc. For an uninitiated reader, it is tough staring at these protein structures trying to figure out which features are where? Also, I suggest adding a length bar to one or more Figures

7. It is stated that the Cpx accessory helix inhibits release "likely" because it causes steric clashes with the vesicle (line 85). I think this is a powerful and very reasonable suggestion, but perhaps "possibly" would better reflect current uncertainty about the mechanism.

8. Their 26 nm diameter vesicles are ~ 2-fold smaller than synaptic vesicles. I do understand why this measure is taken (and the authors mention why), but the synaptic vesicle size should be stated.

Reviewer #3 (Recommendations for the authors):

My overall sense from this study is that the simulation efforts are preliminary and sufficiently incomplete to cause concern about the validity of the conclusions. I am concerned about several omissions and their potential impact on conclusions about the prefusion complex and the possible trajectories leading to fusion:

1. SNARE/Syt/Cpx omissions – What is the potential impact of removing the Habc region of syntaxin 1 given its significant excluded volume and potential interactions with membrane PIP2? Similarly, excluding the palmitoylated linker regions of SNAP-25 may play important and interesting roles affecting SNARE orientation, the distribution of forces between SNAREs and membranes, and membrane behavior. The lack of a Syt1 juxtamembrane region (as well as its transmembrane anchor) seems like a real missed opportunity given past work suggesting several interesting hypotheses for intramolecular and membrane interactions of this region. Finally, omitting the C-terminal domain of Cpx1 with its known membrane-interacting region may have significant implications for the detailed behavior of Cpx1 and the forces acting on its SNARE-binding region. While no realistic simulation could currently hope to capture all of this, I would have preferred fewer simulations with more assessment of whether or not some of these omissions would cause major changes to the behavior of the simulated system.

2. Calcium-phospholipid interactions – When the authors included 5 calcium ions per Syt1 to assess the impact of elevated local calcium on the simulation dynamics, I was struck by a lack of corresponding calcium interactions with PS and PIP2. 20 calcium ions in the simulated volume would roughly correspond to 1 mM calcium, and even that wouldn't necessarily lead to all 20 potential binding sites on Syt1 being occupied. At the same time, one would expect divalent interactions with PS and PIP2, which could neutralize membrane repulsion and significantly lower at least one aspect of the complex membrane fusion energy barrier. Work by chemists such as Feigenson have indicated strong calcium-mediated interactions between even PS and PC at concentrations much lower than 1 mM (Biochemistry 1989). Some of these chemical details may not be capable of proper simulation in the MD formalism deployed in the current study, but this should be addressable in some fashion.

3. I was not convinced by the authors' reasoning regarding one microsecond being a relevant timescale for synaptic vesicle fusion. And given that even some initial phase of membrane fusion was not observed in these simulations, I find it impossible to access wherein the process of priming/fusion these current simulations reside. The fastest reported latency between presynaptic calcium entry and fusion is around 60 microseconds as the authors point with the Sabatini/Regehr study. Importantly, that was not a single-synapse measurement but instead, a population measure involving 1000s of synapses. So the first latency likely represents a small population from the fast tail of a distribution of fusion times. And given the 1-2 microsecond delay for cleft glutamate diffusion and the 10-20 microsecond activation time of a stellate cell AMPA receptor, it is likely that the calcium-fusion delay at this synapse resides in the 50-100 microsecond time window. Thus, a 400-nanosecond simulation would seem far too brief to do this process justice.

4. Since the simulations are certainly not trying to capture relevant roles and impacts of other core synaptic proteins such as Munc13 and Munc18, I thought that the use of 'primed' state was a bit oversold and misleading in this manuscript. These simulations seem most appropriate for interpreting in vitro liposome fusion experiments utilizing just SNAREs or SNAREs plus Cpx/Syt1. I am not sure what it would mean to describe a primed state for the SNAREs and synaptic vesicle without also having Munc13 present and bound at least to the two membranes if not also to the SNAREs. I appreciate that the authors are modeling something that represents our best guess for the SNARE assembly on a tightly docked and primed vesicle, but this simulation clearly lacks crucial elements that go into what the field usually refers to as a primed synaptic vesicle. I would want the language used to reflect this as much as possible.

5. I do not have a sense for how worrisome it is from a technical perspective to forgo replicate simulations. For instance, is it better to have two replicates each of three simulations rather than six slightly different simulations each done once? It would be useful to have some discussion of the uncertainty/reliability attached to these conclusions given the absence of replicates.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "A glimpse at the primed Synaptotagmin-SNARE-complexin complex from all-atom molecular dynamics simulations" for further consideration by eLife. Your revised article has been evaluated by Vivek Malhotra (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below.

Please address the points brought up by Reviewer #3 on the primed state and on calcium/lipid interactions, at least at the writing level. To avoid ambiguity, it would be better to remove the word 'primed' from the title, lines 34 and 37 in the abstract, and line 129 in the introduction, and change the corresponding sentences when needed. This is probably not sufficient because there are so many mentions of 'primed states', primed complexes', or 'vesicle priming'.

Regarding the interactions of calcium with lipids, we realize that getting computational time is a limiting and costly resource currently. Asking to perform additional simulations involving lipid/calcium interactions may be difficult but the authors can certainly mention that it is a limitation of their simulations that may affect the outcome and should be tested in the future.

Reviewer #3 (Recommendations for the authors):

Rizo and colleagues have shortened and edited their manuscript as requested in the first review. I wasn't strongly enthusiastic about this MD study during the first round and remain somewhat dissatisfied after reading the authors' response to our concerns. Two of my concerns were largely ignored by the authors in their rebuttal but remain worrisome to me nonetheless.

One concern is the authors' continued declaration that their simulations are synonymous with the primed fusion complex. Vesicle priming is already a somewhat muddled concept in the field and this manuscript doesn't help the confusion. I appreciate that they edited their title a bit but anyone glancing at the paper or searching for it on PubMed would very likely interpret this as the primed state prior to fusion. In addition, they conclude in the abstract (line 37) that 'the primed state contains macromolecular assemblies …' whereas I don't believe the simulations warrant this conclusion. This is reiterated in the last sentence of the introduction (line 129-132) but at least they soften the conclusion with 'suggest that'. I am not sure the authors got that much more out of the model than they put in to begin with since they chose starting points that they were already convinced represented their best guess at the primed state of the fusion complex. Perhaps some of the observations regarding the juxtamembrane linkers of the SNAREs are moderately unexpected, but given that no fusion was witnessed, the reader doesn't know which details of the current model truly correspond to relevant prefusion scenarios.

My other concern is that the authors continue to ignore the very real possibility that calcium interactions directly with the phospholipids (independent of Syt1 C2 domains) are a critical aspect of membrane fusion. This has been studied chemically and using in vitro membrane fusion assays for 50 years but wasn't even discussed as a possible explanation for the lack of fusion in the simulation where calcium was included. Just to be explicit, I am thinking of papers such as Papahadjopoulos BBA 1976, Feigenson Biochem 1986,1987, and 1989 studies, Kachar Biophys J 1986, and modern studies such as Churchward Biophys J 2008. While I don't know what the technical limitations of implementing calcium-phospholipid interactions are in all-atom MD, I can find examples in the literature such as Allolio and Harries ACS Nano 2021 and Allolio et al. PNAS 2018 where calcium ion interactions with phospholipids during membrane fusion are explicitly incorporated, so I assume there isn't a fundamental reason this cannot be explored or acknowledged. I don't think it would be surprising if some of the key results here such as the juxtamembrane linker electrostatic interactions with the membrane would be strongly affected in addition to the possibility that the fusion energy barrier would be lowered sufficiently to witness the beginning of a fusion event on a microsecond time scale.

Overall, this was a nice first effort at an ambitious simulation scale and could serve as an introductory template for future attempts at modeling SNARE-mediated fusion. The preliminary and underdeveloped feel of the manuscript and notably, the lack of some sort of fusion-like transition captured in the simulations diminish my enthusiasm a bit.

https://doi.org/10.7554/eLife.76356.sa1

Author response

General comments:

We thank the reviewers for their very careful review of our manuscript and their constructive criticisms, which have helped us to substantially improve the quality of the manuscript. We particularly appreciate the positive overall evaluation despite the concerns regarding the limitations of our simulations.

On a personal note, I would like to point out that I was very aware of these limitations when I initiated this project. Nevertheless, I decided to devote a very large amount of my own time to perform these simulations with the help from my co-authors because, after 29 years working intensely in this field, I felt that we really did not understand membrane fusion and it was very difficult to obtain a description of the dynamic events that lead to membrane fusion at atomic detail through experiments only. I believed that there was a chance that we might observe initiation of membrane fusion if we came up with a proper initial configuration. Just seeing some lipids emerging from the bilayers to initiate formation of non-bilayer intermediates that are necessary for membrane merger would be tremendously informative and would lead to a hypothesis(es) that could be tested experimentally. And, even if we did not observe initiation of fusion, we still could learn important insights. Moreover, the three-dimensional models that we built to perform the simulations would be very valuable tools to visualize this system and evaluate mechanistic models. I believe that the results described in the paper support these expectations.

The simulations were performed at TACC on Frontera, one of the fastest supercomputers available for academic research, with a Pathways allocation that I was awarded through XSEDE. The allocation provided 246,940 node hours and the simulations of the larger systems were performed using 32-48 nodes. To perform additional simulations to address some of the reviewer concerns would require application for another large allocation specifically targeted for this purpose and, if the allocation was awarded, it would take several months to perform the simulations. Hence, performance of these simulations would unduly delay publication of our results, which we believe will already be of interest to the research community. Moreover, this paper provides a framework not only for my lab but also for others to perform additional all-atom simulations, which provide an important and timely complement to experimental and coarse-grained simulation data. We hope that the systems that we describe will serve as a basis to apply for high performance computing time and continue this research in labs around the world.

The reviewers will see that we have different opinions on some of the responses described below, which often arise from different appreciation of the available experimental data. We hope the reviewers realize that some concepts are much less well established than is often stated in the literature and will keep an open mind about some of the specific issues that we debate.

Essential revisions:

1) Manuscript readability

a. the manuscript should be shortened. This notably applies to the Results section.

The manuscript has been considerably shortened and in particular the Results section, which is now five pages shorter.

b. the manuscript should better focus on the relevant conclusions relative to the SNAREs (degree of zippering, juxtamembrane linker – lipid interactions), synaptotagmin and complexin. These modifications will be an opportunity to discuss the results in view of other simulations, i.e. coarse grain approaches accessing fusion timescales.

We have removed the last simulation presented in the original manuscript, which was designed to investigate how the synaptotagmin-1 C2 domains cooperate with the SNAREs in triggering fusion of a vesicle and a flat bilayer, because it did not offer conclusive insights about fusion. We have also rewritten most of the discussion to focus on the most relevant conclusions as well as to discuss several key issues raised by the reviewers. We realize that, although we were cautious in presenting our conclusions, we did not discuss alternative views sufficiently. The reviewer comments have been very helpful to provide a more balanced perspective. Moreover, in the introduction and the discussion we now emphasize the importance of continuum and coarse-grained approaches as a complementary strategies to all-atom simulations, while also providing a more concrete justification for the need for atomistic simulations:

(lines 98-112): ‘Simulations using continuum and/or coarse-grained representations have provided important insights into SNARE-mediated membrane fusion (Fortoul et al., 2015; Kasson et al., 2006; Manca et al., 2019; McDargh et al., 2018; Mostafavi et al., 2017; Risselada et al., 2011; Sharma and Lindau, 2018). Continuum models can access the longest timescales, but require experimental data or atomistic simulations to parameterize the material properties, and often need to constrain geometries or material properties due to lack of context-dependent parameters (Fortoul et al., 2015). Coarsegrained molecular simulation approaches are freed from some of these constraints but at the expense of reduced simulation speed, and are limited in their ability to capture certain entropic effects and protein conformational changes (see below). To date, coarse-grained models of SNARE-mediated fusion have accessed the low microsecond timescale (Kasson et al., 2006; Risselada et al., 2011; Sharma and Lindau, 2018). All-atom simulations are better suited to reproduce the finely-balanced network of interactions between proteins, ca2+ and lipids that are expected to lead to membrane fusion but, because of the large size of the systems involved (millions of atoms), the low microsecond time scale has only recently become accessible to the most powerful available high-performance computing resources.’

(lines 593-599): ‘Continuum and coarse-grained simulations, which have already provided important insights into SNARE-mediated membrane fusion (Fortoul et al., 2015; Manca et al., 2019; McDargh et al., 2018; Mostafavi et al., 2017; Risselada et al., 2011; Sharma and Lindau, 2018), offer the opportunity to explore much longer time scales with similar systems. Hence, a marriage of approaches whereby the most interesting results obtained by such simulations are investigated in further detail by all-atom simulations will likely provide a powerful strategy to unravel the intricate mechanisms that govern fast ca2+-triggered membrane fusion.’

2) Limitations of all atom MD simulations

a. At the molecular level, the assumptions regarding the initial arrangement of the proteins and the missing aspects (e.g. Munc13, Munc18, Syt linkers, Cpx-SNARE interactions) must be explicitly stated and discussed in view of the current knowledge in the field.

We now point out the limitations of the simulations, including the absence of key components, in the abstract, at the end of the introduction and in the discussion:

(lines 32-33): ‘Our results need to be interpreted with caution because of the limited simulation times and the absence of key components’.

(lines 123-124): ‘Because of the limited simulation times and the absence of key components, our results cannot lead to definitive conclusions’

(lines 436-438): ‘Our results need to be interpreted with caution because of the limited simulation times, the dependence of the results on the initial configurations and the absence of key elements of the release machinery’

(lines 581-591): ‘Clearly, these models will need to be tested with further MD simulations and experimentation, and there are multiple factors of the primed state of synaptic vesicles that were not included in the simulations. As mentioned above, particularly important for this state is Munc13-1, but other missing elements include Munc18-1, the N-terminal region of syntaxin-1, the SNAP-25 linker joining its SNARE motifs, the N- and C-terminal regions of complexin-1 and the TM region plus linker sequence of Syt1 [reviewed in (Rizo, 2022)]. Moreover, the Syt1 C2B domain can also bind to the SNARE complex through a so-called tripartite interface whereby an α-helix of the C2B domain is adjacent to the complexin-1 helix (Zhou et al., 2017), which could stabilize the complexin1-SNARE interface to hinder rotations that might lead to dissociation (see Figure 3K, Figure 3—figure supplement 2F). Moreover, there is evidence that Syt1 forms oligomeric rings that hinder spontaneous neurotransmitter release (Tagliatti et al., 2020). It is also plausible that the SNAREs alone can induce fast fusion in different configurations that we did not study.’

We hope the reviewers realize that many important results in our field were obtained with experiments using minimal systems that included even fewer components than our simulations, and those contributions are now widely considered as seminal. Moreover, the systems that we built provide a framework to add additional components in the future.

b. Realistic statements about likely fusion times need to be compared to the all-atom simulation times.

We hope that the reviewers can keep an open mind about this issue. We have tried to improve the rationale behind the statement that the fusion step may occur in a few microseconds in the introduction:

(lines 112-120): ‘In this context, it is worth noting that the delay from ca2+ influx into the presynaptic terminal to observation of postsynaptic currents in rat cerebellar synapses at 38°C is 60 µs (Sabatini and Regehr, 1996), and that multiple events occur within this time frame, including ca2+ binding to the sensor, release of inhibitory interactions that hinder premature fusion, ca2+-evoked synaptic vesicle fusion, opening of the fusion pore, diffusion of neurotransmitters through the synaptic cleft, binding of the neurotransmitters to their postsynaptic receptors and opening of the channels that underlie the postsynaptic currents. These observations suggest that the fusion step may occur in just a few microseconds and hence that it may be possible to recapitulate the initiation of ca2+-dependent synaptic vesicle fusions in all-atom MD simulations starting with a properly designed initial configuration.’

Please note that we do not state definitively that the fusion step occurs in just a few microseconds; we present this as a possibility. And we now do not suggest the possibility of recapitulating the entire process of fusion with all-atom MD simulations, but we do suggest that it may be possible to observe the initiation of fusion. It is also worth noting that folding of small proteins can occur in the low microsecond time scale, so this time scale might allow substantial rearrangements necessary to initiate fusion, and that the available coarse-grained simulations of SNARE-mediated fusion were performed in the 1 µs time scale (Kasson et al., 2006; Risselada et al., 2011; Sharma and Lindau, 2018).

We also would like to emphasize that the rate limiting step may not be fusion itself but another of the multiple events that occur from ca2+ influx to postsynaptic currents, particularly the release of inhibitory interactions that hinder premature fusion. For instance, the interaction of the synaptotagmin-1 C2B domain with the SNARE complex is believed to hinder neurotransmitter release, which is supported by screens for mutations that remove dominant negative phenotypes of synaptotagmin-1 mutants [Guan et al. (2017) ELife 6, e28409] and by the finding that a mutation that enhances this interaction (E295A/Y338W) impairs release [Zhou et al. (2015) Nature 525, 62; Voleti et al. (2020) eLife 9, e57154]. The dissociation constant of this interaction is estimated to be about 20 µM, which translates to an off rate of 2,000 Hz (0.5 ms time scale) assuming that the on rate is diffusion limited. Dissociation is likely accelerated by a ca2+-induced change in the orientation of the C2B domain with respect to the membrane, but we do not know to what extent. Hence, we really do not know what is the time scale of the fusion step itself. We now discuss these issues in the following sentences:

(lines 388-392): ‘Based on the estimated kD of the interaction between the C2B domain and the SNARE complex [ca. 20 µM (Voleti et al., 2020)], the off rate for dissociation is expected to be at most 2,000 Hz and hence too slow for the time scales reachable in our simulations. However, it is plausible that dissociation might be strongly accelerated by changes in the orientation of the C2B domain with respect to the membrane induced by ca2+ (Voleti et al., 2020)’.

3) "Primed" state

a. The term "primed", used in the title and in the manuscript is misleading because other core synaptic proteins are not included in the simulations.

We agree that the term ‘primed state’ in the title may be misleading and have replaced this term in the title with ‘primed Synaptotagmin-SNARE-complexin complex’. We hope that this term will be acceptable to the reviewers, as the notion that a complex of synaptotagmin, SNAREs and complexin is a central element of the primed state is widely accepted in the literature [e.g. Zhou et al. (2017). The primed SNARE-complexin-synaptotagmin complex for neuronal exocytosis. Nature 548, 420]. Please note that we use the word ‘glimpse’ in the title to soften the message and not give the impression that we have elucidated the structure of the primed complex. Furthermore, as stated above, we now point out in prominent parts of the manuscript that key components are absent in our simulations.

b. It is difficult to assess whether the vesicle in the proposed molecular arrangement is actually primed. On the contrary, given the narrow intermembrane distance with molecular contacts, it is very likely the membranes will ultimately fuse. All-atom simulations cannot reach the relevant time scales to be conclusive.

We agree that we cannot be sure that the proposed molecular arrangement reflects that present in the primed state. This is why we are careful throughout the text to use terms like ‘suggest’ or ‘indicate’ rather than ‘show’ or ‘demonstrate’. We hope that these terms are acceptable, as we must be able to somehow describe what the simulations are telling. We also note that we place particular emphasis on results from our simulations that have clear correlations with experimental data. For instance, the physiological relevance of the interactions of the SNARE complex with complexin and with the primary interface of the synaptotagmin-1 C2B domain are supported by overwhelming experimental evidence [e.g. Xue et al. (2007) Nat Struct Mol Biol 14, 949; Maximov et al. (2009) Science 323, 516; Zhou et al. (2015) Nature 525, 62; Guan et al. (2017) ELife 6, e28409], and these interactions are generally believed to occur in the primed state given the role of synaptotagmin-1 and complexins in the ca2+-triggered step of neurotransmitter release.

We also would like to emphasize that there was a total of eight synaptotagmin-1-SNAREcomplexin-1 assemblies in the two simulations of the primed complex, and the arrangements observed at the end of the simulations were very similar in all of them. The observed arrangements make a lot of sense from the chemical-biophysical point of view, as they allow very extensive interactions of the polybasic face of the C2B domain with the lipids while maintaining the C2B domain-SNARE and complexin-1-SNARE interfaces observed by crystallography. The physiological importance of some residues of the polybasic face has also been well established [e.g. Brewer et al. (2015) Nat Struct Mol Biol 22, 555]. We now discuss these issues in lines 484-499.

We agree that membranes that are in contact will eventually fuse, but the key point is: at what rate? We now discuss this issue in the following sentences:

(lines 471-478): ‘These extended interfaces have been observed by cryo-EM and by fluorescence microscopy, and evolve to fusion in longer time scales (seconds-minutes) (Diao et al., 2012; Hernandez et al., 2012; Witkowska et al., 2021), in agreement with data showing that liposome fusion occurs minutes after liposome docking (Cypionka et al., 2009). SNARE-mediated fusion was also observed at faster time scales (Domanska et al., 2009; Heo et al., 2021). Thus, it is plausible that fusion is slower under conditions that favor formation of extended interfaces. Interestingly, cryo-EM studies indicated that formation of such extended interfaces is hindered by other proteins involved in ca2+evoked release, which favor point-of-contact interfaces that fuse faster (Diao et al., 2012; Gipson et al., 2017).’

Note that in the cryo-EM analyses of Hernandez et al. the extended contact interfaces between liposomes occur early in the reactions and gradually disappear, and that Diao et al. also conclude that extended interfaces fuse slowly. Moreover, spontaneous release of synaptic vesicles occurs in the minute time scale at hippocampal synapses [e.g. Rhee et al. (2005) PNAS 102, 18664; Xue et al. (2008) PNAS 105, 7875]. Hence, the notion that primed synaptic vesicles may be in contact with the plasma membrane is not inconsistent with the observed spontaneous release rates.

Nevertheless, we agree that the membranes may be further apart in the primed state and in the discussion now explain that the arrangement of the primed complexes that we observed can be consistent with a range of distances:

(lines 572-580): ‘In the second scenario, the distance from the vesicle to the plasma membrane in the tight primed state is a few nm. Manually moving the vesicle in the pose of Figure 7B to a distance of 3 nm from the flat bilayer (Figure 7D) shows that the same spring-loaded configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly could be kept by stretching the synaptobrevin juxtamembrane linker, and the Cpx1(27-72) helix would still hinder progress toward final zippering and fusion. This configuration is also largely compatible with longer vesicle-flat bilayer distances if the linker is stretched further and/or there is partial SNARE unzippering. In this case, steric clashes of the Cpx1(27-72) helix with the vesicle might be alleviated or eliminated, but they would occur as soon there is full zippering and hence would still hinder vesicle fusion.'

With regard to the comment that ‘All-atom simulations cannot reach the relevant time scales to be conclusive’, please see the response to point 2b above.

c. The Cpx accessory helix looks like a wobbly finger unlikely to support much force.

We invite the reviewers to think about these mechanistic issues in terms of energy barriers that slow down reactions. We now discuss this issue in energetic terms:

(lines 525-531): ‘It is difficult to estimate the energy barrier imposed by this behavior on SNARE zippering, but it is known that the energy cost of breaking just one hydrogen bond to distort the complexin-1 helix can be about 1.8 kBT (Nick Pace et al., 2014), which could translate to 7.2 kBT by cooperative action of four primed complexes. Such an energy barrier would slow down SNARE zippering by a factor of 1,340 (=e^7.2). For comparison, the enhancements of spontaneous neurotransmitter release observed upon deletion of complexin range from 3 to > 20 (Huntwork and Littleton, 2007; Martin et al., 2011). Thus, even a small energy barrier caused by steric hindrance can account for these enhancements.’

It is also important to consider that there may be other energy barriers, and the acceleration of release induced by ca2+ is expected to arise in substantial part from ca2+dependent interactions of the synaptotagmin-1 ca2+-binding loops with the lipids.

4) Calcium addition

a. Calcium-phospholipid may play an important part in the molecular arrangement and the fusion process. This is ignored here and should be addressed.

This issue was not ignored. In the introduction we clearly stated that ca2+-dependent binding to phospholipids is critical to trigger release (now in lines 66-68 of the revised manuscript). Ca2+ was not included in most simulations. In the simulation with two flat bilayers that included ca2+, we frequently observed membrane binding through the ca2+binding loops of the C2 domains (e.g. Figure 1F), and we also observed such interactions in the simulation of a vesicle and a flat bilayer with ca2+-bound C2AB dissociated from the SNAREs, which we removed in the revised manuscript (see answer to point 1b above). In the simulation of a vesicle and a flat bilayer with C2AB bound to ca2+ and to the SNAREs, we positioned the C2A domain so that its ca2+-binding loops can readily interact with the flat bilayer (Figure 5A), and such interactions were often observed during the simulation (e.g. Figure 5B). One of the ca2+-binding loops of each C2B domain was inserted into the flat bilayer soon after the start of the simulation and remained inserted (Figure 5D, F, H, J), but the other loop could not insert into the membrane because the C2B domain did not dissociate from the SNAREs, remaining approximately parallel to the flat bilayer.

b. How does one reconcile that the aliphatic loops on Synaptotagmin C2B domain do not insert into the membrane upon calcium binding as observed in previous structural/functional studies?

See answer to point 4a above.

Reviewer #1 (Recommendations for the authors):

1) It is difficult to assess if the results from simulations are a true representative of the biological process or an outcome of the initial condition/constraints chosen. For example, it is puzzling that there is no intra-molecular assembly of the SNAREpins during the simulation even though the coil-coil interactions are expected to occur in the simulation time scales. It appears, in almost all cases, that the SNARE zippering is unaltered at the end of the simulation. Also, it might be possible that the authors' choice to model the juxtamembrane region as a fully unstructured region prevents membrane fusion under the current simulation conditions. While it is not clear if the SNAREs zippering extends through the juxtamembrane (JM) region into the transmembrane region as observed in the crystal structure of full-length SNARE (Stein et al. Nature 2009), it stands to reason the JM region needed to be at least partially structured for effective force transfer to catalyze merging of the bilayers.

We would like to ask the reviewer to keep an open mind about the various issues raised here:

Yes, one might hypothesize that zippering of the C-terminus of the four-helix bundle would be fast once the N-terminal half is assembled, but this prediction is not supported by the simulations. In retrospect, this is not so surprising when one looks at these systems and realizes their complexity, with many interactions that can impose energy barriers to conformational transitions necessary for full zippering (we invite the reviewer to look at some of the snapshots that are now available on Dryad at https://datadryad.org/stash/share/BvqW82678udTxtZ3ifYsx9H2LXy_ttCwm271QmVXvoI). In our opinion, these results are teaching us features that we did not consider previously because we did not have a three-dimensional model of a plausible configuration of this system to look at. This is one of the reasons why we believe that the simulations will be of interest to researchers in the field, even if we did not observe the initiation of fusion in our simulations.

We want to clarify that we did not choose to model the juxtamembrane (JM) regions as fully unstructured sequences. We simply did not impose constraints to keep a helical structure in the JM regions. The notions that the energy of formation of SNARE complex needs to be transferred to the membranes to induce membrane fusion and that the JMs form continuous helices with the SNARE motifs and TM regions to exert this force transfer are hypotheses that have not been demonstrated even though they are widely accepted in the literature. In fact, there are very strong arguments against models invoking continuous helices in the JM regions, as we now discuss:

(lines 442-457): ‘However, assembly of the four-helix bundle brings the membranes within a few nm from each other, and it is unclear how the SNAREs exert additional force on the membranes to induce fusion. A major problem with the widespread notion that the SNARE motif, juxtamembrane linker and TM region of synaptobrevin and syntaxin-1 form continuous helices that force fusion as they zipper from the N- to the C-terminus (Hanson et al., 1997a; Sutton et al., 1998; Weber et al., 1998) is that the bent conformations of the linkers envisioned in these models are unrealistic from an energetic point of view. Although optical tweezer data suggested that interactions between the juxtamembrane linkers contribute to exerting force on the membranes to induce fusion (Gao et al., 2012) helix continuity in the linkers is not required for neurotransmitter release (Kesavan et al., 2007; Zhou et al., 2013). Substantial release was observed even upon insertion of a five-residue sequence into the synaptobrevin linker (Kesavan et al., 2007) despite the fact that this sequence should break the register of linker-linker interactions and contained two (helix disrupting) glycine residues. Note also that the optical tweezer data were obtained in the absence of membranes and that, in vivo, the linkers are likely to interact with the lipids given the proximity of each linker to the adjacent membrane and the abundance of basic residues in the linker sequence (and aromatic residues in the case of synaptobrevin). The extensive interactions of the linkers with the membranes observed in all our simulations support this prediction.’

We want to emphasize that we do not really conclude that the linkers are unstructured and we now make it clear that there are alternative possibilities, but unstructured linkers provide a natural, unbiased starting point for the simulations. These issues are presented in the following sentences:

(lines 458-465): ‘Based on conformational grounds, it is not surprising that the juxtamembrane linkers became unstructured during the simulation that we performed to generate a trans-SNARE complex starting from the crystal structure of the cis-SNARE complex. Note however that we did not perform a systematic analysis to examine the range of linker structures that are compatible with the geometry of a trans-SNARE complex. In any case, the linkers are expected to be unstructured before SNARE complex assembly, which is supported by EPR data (Kim et al., 2002). Therefore, configurations with unstructured linkers were natural, unbiased starting points for the simulations, and helical conformation could be adopted by the linkers during the simulations if they were preferred.’

It is also worth noting that coarse-grained simulations using the MARTINI force field appeared to support the notion that continuous helical conformations of the SNAREs are critical to induce membrane fusion (Risselada et al., 2011), but this result could be a major potential artifact of the approach used. Thus, in the MARTINI force field the secondary structure of proteins is constrained to be mostly fixed due to the lack of explicit hydrogen bonding. Therefore, any reaction that involves a partial (un)folding mechanism would be missed, and the fusion observed in the simulations may have been caused by the helical restraints intrinsic to the force field. We now explain better these issues in the revised manuscript:

(lines 168-181): These models were supported by coarse-grained MD simulations that used the MARTINI force field and modeled the SNAREs in continuous helical conformations (Risselada et al., 2011). However, the intrinsic helical restraints enforced by the force field might bias the results and/or obscure the potential role of conformational changes in the dynamical coupling of the SNAREs to membrane fusion. Moreover, the bending of the helices required to form trans-SNARE complexes leads to unrealistic conformations that are expected to be unfavorable energetically because of their distorted geometry and are not commonly observed in protein structures. Thus, the helical restraints might have played a key role in membrane fusion in these simulations. Although continuous helices were observed in the crystal structure of a cis-SNARE complex that represents the configuration occurring after membrane fusion (Stein et al., 2009), the natural expectation is that the helical structure must break somewhere to accommodate the geometry of a trans-SNARE complex, most likely at the juxtamembrane linker. This expectation has been supported experimentally (Kim et al., 2002) and with all-atom MD simulations (Bykhovskaia, 2021). Moreover, helix continuity in the linkers is not required for neurotransmitter release (Kesavan et al., 2007; Zhou et al., 2013).

Finally, it is also worth noting that there are alternative models that do not envisage a force transfer by the SNAREs on the membranes but rather propose that the SNAREs bring the membranes together and then membrane fusion is catalyzed by bilayer perturbations caused by insertion of hydrophobic sequences, for instance from Sec17 in yeast vacuolar fusion or from the synaptotagmin-1 C2 domains in neurotransmitter release [Wickner and Rizo (2017) Mol Biol Cell 28, 707]. Indeed, a recent paper showed that liposome fusion can be observed upon crippling three SNAREs such that C-terminal zippering of the four-helix bundle is impossible when the reactions included Sec17 and Sec18 (the yeast homologues of SNAPs and NSF), and fusion required a hydrophobic loop of Sec17 that inserts into membranes. While it is unclear to what extent these results are relevant to ca2+-triggered neurotransmitter release, they do support this alternative model, and it is also known that ca2+-dependent insertion of the ca2+-binding loops into membrane is critical to trigger release [e.g. Fernandez-Chacon et al. (2001) Nature 410, 41; Rhee et al. (2005) PNAS 102, 18664].

In summary, there are strong reasons to keep an open mind about the mechanism underlying SNARE-dependent membrane fusion.

2) A major conclusion of the report is that the steric clash between Complexin accessory helix and vesicle serves as the fusion clamp and indeed drives the positioning of the SNARE and Synaptotagmin on the planar bilayer. However, there are a couple of factors that might alleviate or even mitigate this steric clash: (i) the vesicle and bilayer are positioned at ~2.3 nm apart at the beginning of the simulation. However, high-resolution cyroEM analysis in synaptosomes/cultured neurons (Fernandez-Busnadiego, R. J Cell Biol 2013; Radhakrishnan et al. PNAS 2021) show that the inter-bilayer distance of docked/primed vesicle is ~4.5 nm. Thus, it might be imperative to carry out the simulation with the physiological accurate inter-bilayer distance.

The initial distance between the vesicle and the bilayer that we used (~2.3 nm) was based on extensive EM studies of synapses performed with high-pressure freezing that observed many vesicles close to being in contact with the plasma membrane and assumed that docked/primed vesicles were within 0-5 nm from the plasma membrane [Imig et al. (2014) Neuron 84, 416]. The Radhakrishnan et al. PNAS 2021 paper that described a detailed analysis of vesicle-plasma membrane distances in synapses analyzed by cryoelectron tomography was published during the course of this work. The data in this paper show that there is a distribution of distances and vesicles are almost in contact with the plasma membrane in some of the images of the paper (e.g. top left of Figure 2A and top right of Figure 3A in that paper). The 3D map constructed from a subset of subtomograms yielded a distance of 3.5 nm between the vesicle and the plasma membrane. Considering the limited resolution of the map (4.4 nm) and the overall distribution of distances observed, these data are not inconsistent with the Imig et al. data. Overall, it seems clear that there is a considerable uncertainty in the distance between a primed vesicle and the plasma membrane.

We agree that it would be desirable to perform additional simulations with different distances between the vesicle and the flat bilayer. However, without including a protein that can bridge the membranes and keep them apart, such as Munc13-1, it is most likely that the SNAREs would still bring the membranes into contact. We do plan to investigate this issue in the future but, as explained above, it will take a long time to apply for and obtain computing time on Frontera to perform these additional studies. To address the concern expressed by the reviewer, we now discuss the uncertainty in the vesicle-plasma membrane distance:

(lines 541-556): ‘While all these observations support the proposal that the consistent overall configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly observed for the eight primed complexes in the two simulations resembles that present in primed vesicles, there are clear uncertainties with regard to the extent of C-terminal SNARE zippering and the vesicle-flat membrane distance, which are closely related. In our two simulations, the vesicle was drawn into contact with the flat bilayer. However, other factors in addition to complexin are likely to hinder zippering in vivo, most notably Munc13-1 because in the absence of ca2+ it bridges membranes in approximately perpendicular orientations that keep the membranes apart (Camacho et al., 2021; Grushin et al., 2022; Quade et al., 2019). Analyses by high-pressure freezing electron tomography (ET) (Imig et al., 2014) and cryo-ET (Radhakrishnan et al., 2021) showed that docked vesicles exhibit a distribution of distances from the plasma membrane that range from 0 to several nm, and a density map built from a subset of subtomograms revealed a distance of 3.5 nm. However, a recent model that can explain a large amount of available presynaptic plasticity data invoked two primed states, one that involves partially assembled SNARE complexes and has low release probability (loose state), and another with more fully assembled SNARE complexes that has a much higher release probability (tight state) (Neher and Brose, 2018). Hence, it is plausible that the vesicles that are closest to the plasma membrane account for much of the neurotransmitter release observed.’

Please note also that, as explained in our response to point 3b of essential revisions, the overall configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly is compatible with vesicle-plasma membrane distances ranges from 0 to a few nm, which is now illustrated in Figure 6D.

(ii) Complexin molecule has been positioned on SNAREs assuming a fully-zippered SNARE complex. However, there is sufficient evidence that SNAREs are likely only partially-assembled in an RPP vesicle (Hua & Charlton, Nat Neurosci 1999, Prashad & Charlton, PLoS One, 2014), and the positioning of the CPX, esp. the accessory helix is correlated to the extent of SNARE assembly (Choi et al. ELife 2016, Kummel et al. Nat Struct Biol 2011; Zhou et al. Nature 2017). Furthermore, accessory helix has been shown to interact with c-terminal ends of t- and v-SNARE molecules (Kummel et al. Nat Struct Mol Biol 2011; Malsam et al., Cell Reports 2020). Thus, it is possible that the alternate positioning of the accessory helix and other interactions might reduce the observed steric clash.

As explained above, the analysis from Neher and Brose 2018 indicates that a large amount of physiological data can be explained by a model with two states that have different extents of C-terminal zippering, and most ca2+-triggered release occurs from the state that is more zippered. We do not know the extent of zippering in the tight state, but it is plausible that zippering is similar to that observed in our simulations. We agree that the position of the complexin-1 accessory helix may depend on the extent of C-terminal SNARE complex assembly, but please note that we observed fast motions of the accessory helix during the simulations, which is particularly well illustrated by the changes observed during 336 ns in Figure 3—figure supplement 2. Hence, there was ample freedom for the accessory helix to sample alternative positions and the behavior that we observed is consistent with the data suggesting alternative positions of the accessory helix presented in the papers cited by the reviewer. It is also important to note that, even if the SNARE complex is less zippered in the primed state than proposed in our model, as is the case for some of the complexes in our simulations, there still can be steric clashes of the helix with the vesicle. Even if there are less steric clashes, there would be more clashes as soon as the SNARE complex zippers further to induce fusion, so steric clashes would still hinder release. We now discuss this issue in the following sentences:

(lines 573-580): ‘Manually moving the vesicle in the pose of Figure 6B to a distance of 3 nm from the flat bilayer (Figure 6D) shows that the same spring-loaded configuration of the Syt1 C2B domain-SNARE-Cpx1(27-72)-flat bilayer assembly could be kept by stretching the synaptobrevin juxtamembrane linker, and the Cpx1(27-72) helix would still hinder progress toward final zippering and fusion. This configuration is also largely compatible with longer vesicle-flat bilayer distances if the linker is stretched further and/or there is partial SNARE unzippering. In this case, steric clashes of the Cpx1(27-72) helix with the vesicle might be alleviated or eliminated, but they would occur as soon there is full zippering and hence would still hinder vesicle fusion.’

The Malsam et al., Cell Reports 2020 did provide some evidence of interactions of the accessory helix with SNAP-25, but note that replacement of the accessory helix with an unrelated helical sequence still retained its inhibitory activity fully [Radoff et al. (2014) eLife 3, e04553] (see answer to point 3c of Essential Revisions above). This observation argues against an important role for specific complexin-SNARE interactions in inhibiting release. In any case, non-specific interactions could play a role, and our simulations did not preclude interactions of the complexin-1 accessory helix. In fact we did observe such interactions occasionally, as we now show in a new figure panel (Figure 3—figure supplement 2G). We now address these issues in the discussion:

(lines 531-540): ‘Note also that we observed occasional interactions of the Cpx1(27-72) accessory helix with C-terminal residues of the synaptobrevin and SNAP-25 SNARE motifs that are favored by proximity (e.g. Figure 3G, Figure 3—figure supplement 2G), and recent cross-linking experiments suggested that very weak interactions of the complexin accessory helix with synaptobrevin and SNAP-25 hinder C-terminal zippering and release (Malsam et al., 2020). It is unlikely that specific interactions underlie the complexin inhibitory function, as this function was retained when the accessory helix was replaced with an unrelated helical sequence, and helix propensity appears to be the key determinant for the inhibitory function (Radoff et al., 2014). However, very weak complexin-SNARE interactions that do not need to be specific may slow down C-terminal SNARE zippering and thus contribute also to inhibition of release by the complex accessory helix.’

3) How does one reconcile that the aliphatic loops on Synaptotagmin C2B domain do not insert into the membrane upon calcium binding as observed in previous structural/functional studies (Grushin et al. Nat Comms 2019; Kuo et al. J Mol Biol 2009) even though synaptotagmin interacts with the membrane, including partial insertion of the C2B aliphatic loop, under calcium-free conditions. This is a rather crucial and missing piece considering that calcium-triggered membrane insertion is predicted to be the driving force for triggered fusion.

Please see also answer to point 4a of Essential revisions above.

Reviewer #2 (Recommendations for the authors):

1. The authors' major conclusion is that the AA simulations support the model of Voleti et al. for the organization that clamps fusion in the pre-ca primed state. However, from Figures 3, 4 (and associated figure supplements) fusion seems very likely not to be clamped, given the vesicle contacts the planar membrane (the degree of contact is still growing at the end of the simulation, Figure 4 supp 4). As stated in (lines 420-425) the vesicle membrane is not flattened. This indicates a lower force than with SNAREs alone, but seems unlikely to block fusion. Due to running time limitations, AA simulations cannot test if fusion would occur in a physiological time. The structure does not keep the membranes apart, as it rotates and permits contact. The authors are clear about this – indeed, to predict the orientation is stated as a major objective. But the conclusions of lines 365-367 and the final sentence of the abstract, suggesting these results demonstrate a fusion clamp, seems unjustified as far as I can see. The emphasis on the cpx accessory helix role also appears somewhat exaggerated, as if on its own it provides a mini-buttress that separates vesicle and planar membrane. It's hard for me to imagine it supports much force in this configuration.

Please see the responses to point 2b, 3b and 3c of Essential revisions, as well as to point 2 from reviewer 1.

2. The simulations with bound calcium (final section of Results) seem inconclusive. The number of contacts is still growing at the end of the simulation, and we cannot know if the C2B will ever dissociate from the SNAREs. It's very reasonable to try this simulation but given the outcome I'm not sure a long section is merited, particularly with the tentative title "Potential effects of Ca2plus binding to synaptotagmin-1." This negative, albeit interesting finding, might be briefly summarized in the main text.

We agree that the simulations with ca2+ did not yield very conclusive data and have removed the last simulation in which the synaptotagmin-1 C2AB fragment was dissociated from the SNAREs, shortening this section considerably. We prefer to keep the simulation started with C2AB bound to ca2+ and to the SNARE complex because it illustrates two points: i) that no ca2+-induced dissociation of C2B from the SNAREs occurred in the time scale of this simulation for any of the four complexes; and ii) that there was lower tendency of the SNARE complex to interact with the flat bilayer because of the absence of clashes between complexin-1 and the vesicle. We have kept this description as a small section (with a different title) because the previous section is very long. We hope that this is acceptable and would be willing to remove this section if the reviewer feels strongly about it. As we do throughout the paper, we do not draw firm conclusions from these results but point out what the data suggest.

3. The manuscript would be strengthened by a more balanced presentation acknowledging the limitations of AA simulations (while of course still extolling their merits) and connecting to some degree with analysis on other scales, including coarse-grained approaches beyond MARTINI. SNARE-mediated fusion was studied using ultra coarse-grained (Mostafavi et al., 2017; McDargh et al., 2018) and even continuum (Manca et al., 2019) representations. Every approach has strengths and weaknesses. AA approaches scrutinize local issues as no others can, but presently they are remote from being able to demonstrate hemifusion, fusion, unclamping and ca-evoked fusion. Making matters worse, NT release is clearly stochastic, so multiple runs are needed for each condition. These limitations are apparent in this study: almost every conclusion comes with a caveat related to running time. In previous seminal MARTINI studies that achieved fusion (Risselada, Sharma and Lindau) the conditions were intentionally biased for fusion (vesicle size, lipid composition, temperature, helical LDs) or nanodiscs were used. In (Risselada, 2011) no fusion was observed when the LD was made unstructured.

We fully agree with this criticism and now present a more balanced perspective. Please see our responses to point 1b of Essential Revisions and the introductory comments from reviewer 2 above.

In the Introduction the authors assert that experiments suggest "..the fusion step occurs in just a few microseconds," which timescales AA simulations may be able to access. They quote the 60 microsec delay times (ca influx to first sign of the excitatory post synaptic current, EPSC) reported by Sabatini and Regehr at 38 degrees. However, 0.5 -2 ms is much more typical in the literature (admittedly, the 38 degrees study is distinguished by the temperature being physiological.) Related, long-ago Katz argued other processes (e.g. NT diffusion across the synaptic cleft) are much faster than NT release (Katz and Miledi, 1965).

Please see our response to point 2b of Essential revisions above.

4. In simulations with SNAREs only, the SNARE complexes are cleared laterally, and the membranes are squashed together, generating an ECZ (extended contact zone, a flat portion of vesicle), Figures 1C, 1F. This is precisely the behavior seen in highly coarse-grained simulations (Mostafavi et al., 2017, Mcdargh et al., 2018), where entropic SNARE-SNARE and SNARE-membrane forces cleared the fusion site and pressed the vesicles together (those studies used undeformable membrane surfaces, so no vesicle flattening occurred). The entropic forces were predicted to provoke fusion after a time of order msec, with faster fusion for more SNAREs. These coarse-grained simulations and their relation to the present findings should be discussed.

We now mention that extended contact interfaces can arise from the entropic forces mentioned by the reviewer in the discussion:

(lines 469-471): ‘Coarse-grained simulations have suggested that formation of such interfaces can also arise from entropic forces that favor outward movement of the SNARE complexes, away from the center of the interface (Mostafavi et al., 2017).’

With regard to the issue of the time scale of fusion of these interfaces, please see response to point 2b of Essential revisions.

The authors suggest the pressing together of the membranes is caused by binding of the LDs to the vesicle membrane (lines 262-266). This does not seem a plausible alternative to the proposed entropic forces, as LD-membrane adhesion would not favor the SNAREs being pushed outwards as far as I can see.

We believe that the linker-membrane interactions may contribute to bring the membranes together but we agree that they would not favor pushing the SNAREs outwards. We have softened the statement regarding the potential role of these interactions in bringing membranes together in the Results section:

(lines 204-206): ‘these findings suggest that any electrostatic repulsion existing between the SNARE four-helix bundle and the membranes can be readily overcome by the high stability of the SNARE four-helix bundle and perhaps some contribution from the linkerbilayer interactions’.

The authors argue that the ECZ in the SNARE-only simulations suggests SNAREs alone cannot fuse membranes rapidly, since fusion was slow in Hernandez et al., 2012 and Witkowska et al., 2021 where ECZs were seen. However, in those in vitro studies many other processes preceded fusion (SNARE assembly, docking etc) and micron scale GUVs were used by Witkowska et al.

As we explain in the response to point 3b of Essential Revisions, the contact interfaces observed by cryo-EM in Hernandez et al. 2012 occurred early in the liposome fusion reactions and gradually disappeared over the time course of the reactions (minute time scale). Hence, at least a fraction of liposomes that had extended contacts took minutes to fuse. This conclusion is consistent not only with the GUV data from Witkowska et al. 2021 but also with fluorescence correlation spectroscopy data showing that liposome docking preceded fusion and that a majority of liposomes fused minutes after docking [Cypionka et al. (2009) PNAS 106, 18575], which we now mention in the paper.

5. A concern is the presentation, whose clarity would benefit from a more concise text. It is laudable to convey the details (computational papers where readers cannot tell what was done are frustrating), but many passages are long repeats of previous passages. For example, opening paragraphs of sections in Results often repeat descriptions of simulations in previous sections at great length, then specifying what was different in the current section. These could be massively shortcut. Short summaries in the main text, with details left to Methods or Supplementary materials, would be more digestible for readers.

We have shortened the Results section by five pages, merging the descriptions that overlapped with the methods section into this section.

6. This paper describes many current hot issues in the field, a great service. The figures are very nice but would be helped by a simple visual key to identify β sandwiches, the polybasic face, ca-binding loops, etc. For an uninitiated reader, it is tough staring at these protein structures trying to figure out which features are where? Also, I suggest adding a length bar to one or more Figures

We have redesigned Figure 1—figure supplement 1 to increase the size of the ribbon diagram of the synaptotagmin-1 C2AB fragment that was in panel E and now is in panel A so that the reader knows where its key elements are from the beginning. In the diagram we have included stick models of the side chains that form the polybasic face and the primary interface, and we have labeled the side chains so that the reader has a guide for their location. The SNARE complex provides a good reference for sizes, as its length is 11 nm. We have put a 10 nm length bar in Figure 1A for the benefit of readers who are less familiar with the SNARE complex.

7. It is stated that the Cpx accessory helix inhibits release "likely" because it causes steric clashes with the vesicle (line 85). I think this is a powerful and very reasonable suggestion, but perhaps "possibly" would better reflect current uncertainty about the mechanism.

We have replaced ‘likely’ with ‘possibly’ (line 79 now).

8. Their 26 nm diameter vesicles are ~ 2-fold smaller than synaptic vesicles. I do understand why this measure is taken (and the authors mention why), but the synaptic vesicle size should be stated.

The vesicle size is stated in the same paragraph describing the 26 nm vesicle that we built (lines 222-233).

Reviewer #3 (Recommendations for the authors):

My overall sense from this study is that the simulation efforts are preliminary and sufficiently incomplete to cause concern about the validity of the conclusions. I am concerned about several omissions and their potential impact on conclusions about the prefusion complex and the possible trajectories leading to fusion:

1. SNARE/Syt/Cpx omissions – What is the potential impact of removing the Habc region of syntaxin 1 given its significant excluded volume and potential interactions with membrane PIP2? Similarly, excluding the palmitoylated linker regions of SNAP-25 may play important and interesting roles affecting SNARE orientation, the distribution of forces between SNAREs and membranes, and membrane behavior. The lack of a Syt1 juxtamembrane region (as well as its transmembrane anchor) seems like a real missed opportunity given past work suggesting several interesting hypotheses for intramolecular and membrane interactions of this region. Finally, omitting the C-terminal domain of Cpx1 with its known membrane-interacting region may have significant implications for the detailed behavior of Cpx1 and the forces acting on its SNARE-binding region. While no realistic simulation could currently hope to capture all of this, I would have preferred fewer simulations with more assessment of whether or not some of these omissions would cause major changes to the behavior of the simulated system.

We agree that some components that we omitted may affect the mechanism of membrane fusion and we now emphasize this in the manuscript, as explained above. However, we note that several of the regions mentioned by the reviewer (SNAP-25 linker, Syt1 juxtamembrane, complexin-1 N- and C-terminal regions) are expected to be largely unstructured even if they form small elements of secondary structure, and there is very limited information on how they are arranged with respect to the rest of the corresponding protein, if there is any specific arrangement. Modeling these regions in arbitrary conformations could yield energy barriers that would further compound the problem with time scale that is a key challenge in these simulations. We also note that the syntaxin-1 Habc domain is crucial for synaptic vesicle priming because it is central for the Munc18-1-Munc13-1-dependent pathway that leads to SNARE complex assembly [Ma et al. (2013) Science 339, 421], but the release probability is not substantially affected by deletion of the Habc domain [Vardar et (2021) ELife 10, e69498], suggesting that this domain is not involved in the ca2+-triggered fusion step or in the primed complexes that are ready for fast release.

As stated above, we hope the reviewer realize that many important contributions in our field were made with experiments using minimal systems that included even fewer components than our simulations.

2. Calcium-phospholipid interactions – When the authors included 5 calcium ions per Syt1 to assess the impact of elevated local calcium on the simulation dynamics, I was struck by a lack of corresponding calcium interactions with PS and PIP2. 20 calcium ions in the simulated volume would roughly correspond to 1 mM calcium, and even that wouldn't necessarily lead to all 20 potential binding sites on Syt1 being occupied. At the same time, one would expect divalent interactions with PS and PIP2, which could neutralize membrane repulsion and significantly lower at least one aspect of the complex membrane fusion energy barrier. Work by chemists such as Feigenson have indicated strong calcium-mediated interactions between even PS and PC at concentrations much lower than 1 mM (Biochemistry 1989). Some of these chemical details may not be capable of proper simulation in the MD formalism deployed in the current study, but this should be addressable in some fashion.

Much of the work was performed without ca2+ and all main results that we stress were obtained without ca2+. Please see also the response to point 4a of Essential Revisions.

3. I was not convinced by the authors' reasoning regarding one microsecond being a relevant timescale for synaptic vesicle fusion. And given that even some initial phase of membrane fusion was not observed in these simulations, I find it impossible to access wherein the process of priming/fusion these current simulations reside. The fastest reported latency between presynaptic calcium entry and fusion is around 60 microseconds as the authors point with the Sabatini/Regehr study. Importantly, that was not a single-synapse measurement but instead, a population measure involving 1000s of synapses. So the first latency likely represents a small population from the fast tail of a distribution of fusion times. And given the 1-2 microsecond delay for cleft glutamate diffusion and the 10-20 microsecond activation time of a stellate cell AMPA receptor, it is likely that the calcium-fusion delay at this synapse resides in the 50-100 microsecond time window. Thus, a 400-nanosecond simulation would seem far too brief to do this process justice.

Please see the response to point 2b of Essential revisions.

4. Since the simulations are certainly not trying to capture relevant roles and impacts of other core synaptic proteins such as Munc13 and Munc18, I thought that the use of 'primed' state was a bit oversold and misleading in this manuscript. These simulations seem most appropriate for interpreting in vitro liposome fusion experiments utilizing just SNAREs or SNAREs plus Cpx/Syt1. I am not sure what it would mean to describe a primed state for the SNAREs and synaptic vesicle without also having Munc13 present and bound at least to the two membranes if not also to the SNAREs. I appreciate that the authors are modeling something that represents our best guess for the SNARE assembly on a tightly docked and primed vesicle, but this simulation clearly lacks crucial elements that go into what the field usually refers to as a primed synaptic vesicle. I would want the language used to reflect this as much as possible.

Please see the response to point 3a of Essential revisions.

5. I do not have a sense for how worrisome it is from a technical perspective to forgo replicate simulations. For instance, is it better to have two replicates each of three simulations rather than six slightly different simulations each done once? It would be useful to have some discussion of the uncertainty/reliability attached to these conclusions given the absence of replicates.

Please see General comments above.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below.

Please address the points brought up by Reviewer #3 on the primed state and on calcium/lipid interactions, at least at the writing level. To avoid ambiguity, it would be better to remove the word 'primed' from the title, lines 34 and 37 in the abstract, and line 129 in the introduction, and change the corresponding sentences when needed. This is probably not sufficient because there are so many mentions of 'primed states', primed complexes', or 'vesicle priming'.

Please see response to the first point from reviewer #3.

Regarding the interactions of calcium with lipids, we realize that getting computational time is a limiting and costly resource currently. Asking to perform additional simulations involving lipid/calcium interactions may be difficult but the authors can certainly mention that it is a limitation of their simulations that may affect the outcome and should be tested in the future.

We mentioned and still mention the limitations arising from the limited time of our simulations in the abstract, the introduction and the discussion. See also the response to the second point from reviewer #3.

Reviewer #3 (Recommendations for the authors):

Rizo and colleagues have shortened and edited their manuscript as requested in the first review. I wasn't strongly enthusiastic about this MD study during the first round and remain somewhat dissatisfied after reading the authors' response to our concerns. Two of my concerns were largely ignored by the authors in their rebuttal but remain worrisome to me nonetheless.

One concern is the authors' continued declaration that their simulations are synonymous with the primed fusion complex. Vesicle priming is already a somewhat muddled concept in the field and this manuscript doesn't help the confusion. I appreciate that they edited their title a bit but anyone glancing at the paper or searching for it on PubMed would very likely interpret this as the primed state prior to fusion. In addition, they conclude in the abstract (line 37) that 'the primed state contains macromolecular assemblies …' whereas I don't believe the simulations warrant this conclusion. This is reiterated in the last sentence of the introduction (line 129-132) but at least they soften the conclusion with 'suggest that'. I am not sure the authors got that much more out of the model than they put in to begin with since they chose starting points that they were already convinced represented their best guess at the primed state of the fusion complex. Perhaps some of the observations regarding the juxtamembrane linkers of the SNAREs are moderately unexpected, but given that no fusion was witnessed, the reader doesn't know which details of the current model truly correspond to relevant prefusion scenarios.

Both in the paper and the rebuttal letter, we provided multiple arguments supporting the notion that the simulations with synaptotagmin-1-SNARE-complexin-1 complexes are relevant to the primed state of synaptic vesicles. The reviewer is reluctant to accept this relevance (and even the concept of a primed state), but does not provide specific reasons against any of our arguments. It is not only us that are convinced that the primed state of synaptic vesicles includes trans-SNARE complexes bound to synaptotagmin-1 and complexin-1 through the same modes as the initial state that we built; this view is based on extensive evidence and is shared by many people in the field, including leaders such as James Rothman, Axel Brunger, Erwin Neher and Nils Brose among others [e.g. Grushin et al. (2019) Nat Commun 10, 2413; Radhakrishnan et al. (2021) PNAS 118, e2024029118; Brunger et al. (2018) Annu Rev Biophys 47, 469; Neher and Brose (2018) Neuron 100, 1283]. Note also that the reviewer trivializes our results, as the main purpose of the simulations with synaptotagmin-1-SNARE-complexin-1 complexes was to investigate how they are oriented with respect to the membranes, and our simulations yielded consistent results for eight out of eight complexes in this respect.

We are aware of the limitations of our study and we emphasize them prominently in the abstract, introduction and discussion, normally using verbs likes suggest rather than demonstrate. In line 37 of the abstract we did not use ‘conclude’, as indicated by the reviewer; the corresponding sentence started in line 35 and used the expression ‘suggest that’. We need to somehow relate the simulations to the existing knowledge in the field; otherwise, readers will be completely confused, will not know why we performed these simulations and will not understand the potential implications of our results. To partially address the concern from the reviewer, we have checked the manuscript carefully to make sure that we do not use strong words like demonstrate or show when referring to the relevance of the results to vesicle priming, and we have changed the title to:

‘All-atom molecular dynamics simulations of Synaptotagmin-SNARE-complexin complexes bridging a vesicle and a flat lipid bilayer’

We hope that this is acceptable.

My other concern is that the authors continue to ignore the very real possibility that calcium interactions directly with the phospholipids (independent of Syt1 C2 domains) are a critical aspect of membrane fusion. This has been studied chemically and using in vitro membrane fusion assays for 50 years but wasn't even discussed as a possible explanation for the lack of fusion in the simulation where calcium was included. Just to be explicit, I am thinking of papers such as Papahadjopoulos BBA 1976, Feigenson Biochem 1986,1987, and 1989 studies, Kachar Biophys J 1986, and modern studies such as Churchward Biophys J 2008. While I don't know what the technical limitations of implementing calcium-phospholipid interactions are in all-atom MD, I can find examples in the literature such as Allolio and Harries ACS Nano 2021 and Allolio et al. PNAS 2018 where calcium ion interactions with phospholipids during membrane fusion are explicitly incorporated, so I assume there isn't a fundamental reason this cannot be explored or acknowledged. I don't think it would be surprising if some of the key results here such as the juxtamembrane linker electrostatic interactions with the membrane would be strongly affected in addition to the possibility that the fusion energy barrier would be lowered sufficiently to witness the beginning of a fusion event on a microsecond time scale.

We apologize because in the previous round of review we thought that the reviewer was referring to interactions between lipids and ca2+ ions bound to the synaptotagmin-1 C2 domains. Such interactions are known to dramatically increase the affinity of synaptotagmin-1 for ca2+ and hence are intrinsic part of the role of synaptotagmin-1 as a ca2+ sensor for release.

We did not realize that the reviewer was referring to interactions of lipids with free ca2+. Some scientists indeed considered the possibility that such interactions mediate synaptic vesicle fusion many years ago. However, measurements at the calyx of Held showed that 10 µM ca2+ triggered release of 80% of available vesicles [Schneggenburger and Neher (2000) Nature 406, 889]. Moreover, it was established that synaptotagmin-1 acts as the major ca2+ sensor that triggers release and the apparent ca2+ affinity of synaptotagmin-1 in phospholipid binding correlated well with the ca2+-dependence of release [FernandezChacon et al. (2001) Nature 410, 41]. These findings argue against a role for interactions of phospholipids with free ca2+ in release, as the affinity of these interactions is much weaker than that of synaptotagmin-1 in the presence of phospholipids and does not explain the ca2+ dependence of neurotransmitter release.

Lamellar phases of phosphatididyl serine (PS) can bind to ca2+ at low µM concentrations due to cooperativity effects, but membranes with a lipid composition more similar to synaptic membranes bind ca2+ with much lower affinity. For instance, 5 mM ca2+ was used to observe ca2+-induced clustering of liposomes containing 25% PS by electron microscopy in the Kachar et al. (1986) reference mentioned by the reviewer (Biophys. J. 50, 779), but no clustering of liposomes containing 30% PS was observed by dynamic light scattering in the presence of 1 mM ca2+ [Arac et al. (2006) Nat. Struct. Mol. Biol 13, 209]. The reviewer mentions papers describing simulations that incorporated ca2+, and these papers indeed reported ca2+-induced membrane fusion. However, Allolio et al. (2018) [Proc Natl Acad Sci U S A 115, 11923] included 450 mM ca2+ in the simulations, and Allolio et al. (2021) [ACS Nano, 15, 12880] used ca2+ to lipid ratios of 1:10 or higher, which in our qscv system with SNAREs alone would translate to 14.4 mM ca2+ or higher. We believe it is highly questionable that results obtained with such high ca2+ concentrations are physiologically relevant. We also note that, if we wanted to use the 10 µM ca2+ concentration that triggers release at the calyx of Held in our qscv system, we would need to add 0.2 ca2+ ions in the simulation box. One could claim that binding of ca2+ to synaptotagmin-1 could considerably raise the local ca2+ concentration, but this ca2+ would not be free, and local enrichment of ca2+ at the membrane surface is much less favorable because of the much lower ca2+ affinity.

We agree with the reviewer that the possibility that binding of free ca2+ to membranes might facilitate synaptic vesicle fusion was an interesting idea and we believe that the idea cannot be completely ruled out. However, in our opinion this is not likely and we cannot agree with the reviewer that the absence of free ca2+ in our simulations casts serious doubts on our results. We also believe that a substantial discussion of this subject as described above would unnecessarily lengthen the already long discussion. To partially address the comments from the reviewer, we added the following sentence in the final paragraph of the discussion of the revised manuscript (lines 592-595):

‘Furthermore, some evidence suggested many years ago that direct binding of ca2+ to phospholipids could trigger synaptic vesicle fusion (Papahadjopoulos et al., 1976) and, although it is now generally believed that ca2+ triggers neurotransmitter release by binding to Syt1, it is plausible that ca2+-phospholipid interactions might also contribute to trigger membrane fusion.’

Overall, this was a nice first effort at an ambitious simulation scale and could serve as an introductory template for future attempts at modeling SNARE-mediated fusion. The preliminary and underdeveloped feel of the manuscript and notably, the lack of some sort of fusion-like transition captured in the simulations diminish my enthusiasm a bit.

We are glad the reviewer considers our work ‘a nice first effort’.

https://doi.org/10.7554/eLife.76356.sa2

Article and author information

Author details

  1. Josep Rizo

    1. Department of Biophysics, The University of Texas Southwestern Medical Center, Dallas, United States
    2. Department of Biochemistry, University of Texas Southwestern Medical Center, Dallas, United States
    3. Department of Pharmacology, University of Texas Southwestern Medical Center, Dallas, United States
    Contribution
    Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Resources, Validation, Visualization, Writing – original draft
    For correspondence
    Jose.Rizo-Rey@UTSouthwestern.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1773-8311
  2. Levent Sari

    1. Department of Biophysics, The University of Texas Southwestern Medical Center, Dallas, United States
    2. Green Center for Systems Biology, The University of Texas Southwestern Medical Center, Dallas, United States
    Contribution
    Conceptualization, Formal analysis, Methodology, Validation, Visualization, Writing – review and editing
    Competing interests
    No competing interests declared
  3. Yife Qi

    Department of Medicinal Chemistry, School of Pharmacy, Fudan University, Shanghai, China
    Contribution
    Funding acquisition, Investigation, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  4. Wonpil Im

    1. Department of Biological Sciences, Lehigh University, Bethlehem, United States
    2. Department of Chemistry, Lehigh University, Bethlehem, United States
    3. Department of Bioengineering, Lehigh University, Bethlehem, United States
    4. Department of Computer Science and Engineering, Lehigh University, Bethlehem, United States
    Contribution
    Funding acquisition, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5642-6041
  5. Milo M Lin

    1. Department of Biophysics, The University of Texas Southwestern Medical Center, Dallas, United States
    2. Green Center for Systems Biology, The University of Texas Southwestern Medical Center, Dallas, United States
    Contribution
    Conceptualization, Methodology, Validation, Visualization, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8680-2685

Funding

National Institute of Neurological Disorders and Stroke (R35 NS097333)

  • Josep Rizo

Welch Foundation (I-1304)

  • Josep Rizo

National Science Foundation (MCB-2111728)

  • Wonpil Im

Natural Science Foundation of Shanghai (19ZR1473600)

  • Yife Qi

University of Texas at Austin

  • Josep Rizo

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

Most of the work presented in this paper was performed through a Pathways allocation for high performance computing using Frontera (project MCB20033) at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin (URL: http://www.tacc.utexas.edu). This research also used computational resources provided by the BioHPC supercomputing facility located in the Lyda Hill Department of Bioinformatics, UT Southwestern Medical Center, TX (URL: https://portal.biohpc.swmed.edu). This work was supported by grant I-1304 from the Welch Foundation (to JR), by NIH Research Project Award R35 NS097333 (to JR), by NSF Research Project Award MCB-2111728 (to WI), and by the Natural Science Foundation of Shanghai Grant 19ZR1473600 (to YQ).

Senior Editor

  1. Vivek Malhotra, The Barcelona Institute of Science and Technology, Spain

Reviewing Editor

  1. Frederic Pincet, Ecole Normal Superieure, France

Reviewer

  1. Ben O'Shaughnessey

Publication history

  1. Received: December 14, 2021
  2. Preprint posted: December 29, 2021 (view preprint)
  3. Accepted: June 15, 2022
  4. Accepted Manuscript published: June 16, 2022 (version 1)
  5. Version of Record published: June 28, 2022 (version 2)

Copyright

© 2022, Rizo et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 868
    Page views
  • 552
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Josep Rizo
  2. Levent Sari
  3. Yife Qi
  4. Wonpil Im
  5. Milo M Lin
(2022)
All-atom molecular dynamics simulations of Synaptotagmin-SNARE-complexin complexes bridging a vesicle and a flat lipid bilayer
eLife 11:e76356.
https://doi.org/10.7554/eLife.76356

Further reading

    1. Immunology and Inflammation
    2. Structural Biology and Molecular Biophysics
    Toshiyuki Oda et al.
    Research Article

    Langerhans cells are specialized antigen-presenting cells localized within the epidermis and mucosal epithelium. Upon contact with Langerhans cells, pathogens are captured by the C-type lectin langerin and internalized into a structurally unique vesicle known as a Birbeck granule. Although the immunological role of Langerhans cells and Birbeck granules have been extensively studied, the mechanism by which the characteristic zippered membrane structure of Birbeck granules is formed remains elusive. In this study, we observed isolated Birbeck granules using cryo-electron tomography and reconstructed the 3D structure of the repeating unit of the honeycomb lattice of langerin at 6.4 Å resolution. We found that the interaction between the two langerin trimers was mediated by docking the flexible loop at residues 258-263 into the secondary carbohydrate-binding cleft. Mutations within the loop inhibited Birbeck granule formation and the internalization of HIV pseudovirus. These findings suggest a molecular mechanism for membrane zippering during Birbeck granule biogenesis and provide insight into the role of langerin in the defense against viral infection.

    1. Structural Biology and Molecular Biophysics
    Huan-Huan Hu et al.
    Research Article Updated

    Phosphoribosyl pyrophosphate (PRPP) is a key intermediate in the biosynthesis of purine and pyrimidine nucleotides, histidine, tryptophan, and cofactors NAD and NADP. Abnormal regulation of PRPP synthase (PRPS) is associated with human disorders, including Arts syndrome, retinal dystrophy, and gouty arthritis. Recent studies have demonstrated that PRPS can form filamentous cytoophidia in eukaryotes. Here, we show that PRPS forms cytoophidia in prokaryotes both in vitro and in vivo. Moreover, we solve two distinct filament structures of E. coli PRPS at near-atomic resolution using Cryo-EM. The formation of the two types of filaments is controlled by the binding of different ligands. One filament type is resistant to allosteric inhibition. The structural comparison reveals conformational changes of a regulatory flexible loop, which may regulate the binding of the allosteric inhibitor and the substrate ATP. A noncanonical allosteric AMP/ADP binding site is identified to stabilize the conformation of the regulatory flexible loop. Our findings not only explore a new mechanism of PRPS regulation with structural basis, but also propose an additional layer of cell metabolism through PRPS filamentation.