Increased flexibility may promote membrane fusion by allowing TMDs to sample conformations compatible with membrane shape changes that accompany fusion (Langosch et al., 2007; Neumann and Langosch, 2011). Consistent with this view, in reconstituted bulk fusion assays fusion activity correlated with β-branched residue content in the TMD sequence (Hofmann et al., 2004; Langosch et al., 2007), although some of these experiments used small sonication-generated liposomes that can be prone to fusion even without fusogens, and leakiness during fusion was not always tested.

TMD flexibility was also identified as a key factor in secretion kinetics in recent studies of exocytosis from mouse chromaffin cells. Dhara et al. (2016) studied cells lacking Syb2 and Cellubrevin (Syb3), that expressed exogenous Syb2 with the entire TMD replaced with a sequence containing various combinations of only Ile, Leu or Val. The results ranged from severe impairment to normal exocytosis. The degree of restoration of secretion correlated well with the fraction of β-branched residues that the Syb2 TMD sequence contained in its N-terminal half (that portion embedded in the cytoplasmic leaflet of the vesicular/plasma membrane). Replacing the wild-type TMD with a polyL stretch reduced the amplitudes of the rapid phases of exocytosis ~5 fold and the slow phase <2 fold as measured by whole-cell capacitance. Interestingly, however, the kinetics were unaffected. In amperometric measurements of single-vesicle release events, replacing the native TMD by polyV or polyI led to shorter, faster rising spikes (∝ amount of catecholamine flux reaching the detector), and shorter, higher amplitude pre-spike features (related to release through the initial pore). Further, pre-spike fluctuations increased in frequency and amplitude. Replacing the TMD by an α-helix-stabilizing polyL sequence produced the opposite effects. These findings suggest that β-branched residues destabilize the pore, facilitating its nucleation (increasing event frequency), and accelerating pore dilation (shorter pre-spike duration and sharper spikes).

In the VAMP2 TMD, a highly conserved Gly100 is followed by another tiny residue three positions later (Cys, Gly, or Ala, Figure 1H), which may be important in allowing a kink in the TMD toward the middle of the bilayer (Han et al., 2016b). Hastoy et al. (2017) studied the role of VAMP2 G100 and C103 by substituting them with Val. These substitutions should impair the ability of the TMD to kink. Using both short peptides encompassing the TMD and full-length purified VAMP2 reconstituted in artificial lipid membranes, the authors found that substitution of C103 and especially of G100 with Val rendered the membranes less fluid and impaired a transition from alpha-helical to beta-sheet conformation of the TMD as the protein concentration increased. These effects correlated with reduced exocytosis in PC12 and INS-1 cells when the knocked-down endogenous VAMP2 expression was rescued with the mutants. Fusion pores expanded faster for the Val mutants, but they also resealed faster after discharging less cargo—a result compatible with transient, kiss and run fusion.

These effects of TMD flexibility on fusion pores are further discussed in the “Computer Simulations of SNARE TMDs and Their Influence on the Fusion Pore” section.

Interactions among v- and t-SNARE TMDs can take three forms: homotypic (vTMD-vTMD or tTMD-tTMD), heterotypic (vTMD-tTMD) or interactions of SNARE TMDs with TMDs of other proteins, e.g., synaptophysin (Adams et al., 2015).

Homodimerization of v-SNARE TMDs has long been known (Washbourne et al., 1995; Roy et al., 2004; Langosch et al., 2007), but it has been argued that these weak interactions are of little consequence for exocytosis (Bowen et al., 2002; Fdez et al., 2010; Dhara et al., 2016). Neuronal t-SNAREs form clusters in artificial membranes (Bacia et al., 2004; Murray and Tamm, 2009), neuroendocrine cells (Lang et al., 2001; Barg et al., 2010; van den Bogaart et al., 2011; Honigmann et al., 2013, Honigmann-NSMB13; Gandasi and Barg, 2014) and at the neuromuscular junction (Khuong et al., 2013). However, these clusters seem to arise from electrostatic interactions between phospholipids and the JMR of Stx, or from recruitment by vesicle docking (Gandasi and Barg, 2014) rather than specific TMD-TMD interactions.

It has also been proposed that both t- and v-SNARE TMDs may homo-oligomerize into channel-like structures (Han et al., 2004; Jackson and Chapman, 2008; Chang et al., 2015). Remarkably, systematic mutagenesis showed that residues that affected fusion pore currents all fall on one side of the t-SNARE TMD helix, possibly facing the pore’s lumen (Han et al., 2004). In order to release cargo from the vesicular lumen to the extracellular space, a pore lined with t-SNARE TMDs would require a complementary pore formed on the vesicular side by oligomerization of v-SNARE TMDs. Although some evidence supports this idea (Chang et al., 2015; Bao et al., 2016), it is less compelling than that for t-SNARE TMDs. It is also possible that the vesicular hemi-channel includes TMDs from another vesicular protein such as synaptophysin (Chang et al., 2017). A channel-like pore might constitute only the initial structure, yielding a lipid-lined pore once the pore expands (Chang et al., 2017). Since the initial pore lasts only a few milliseconds, it is difficult to confirm or refute such a highly transient, channel-like structure. The notion of a channel-like structure would gain credibility if the contacts between channel-forming units could be identified and manipulated, e.g., to stabilize the pores.

Interactions between v- and t-SNARE TMDs were reported nearly two decades ago (Poirier et al., 1998; Margittai et al., 1999). More recently, Stein et al. (2009) solved the crystal structure of the neuronal SNARE complex including the Syb2 and Stx TMDs in the presence of detergent. The α-helices of Syb2 and Stx1 continued beyond the SNARE domain to the C-termini, spanning the linker region and the TMDs. Contacts between certain Syb2 and Stx1 residues were identified in both linker and TMD domains (Figures 1F,G).

This raises an interesting question: do the TMD-TMD contacts represent specific interactions promoting SNARE complex zippering through the bilayer and affecting the fusion process, or are they artifacts of crystal packing constraints? Wu et al. (2016) sought to answer this question using a novel approach in which single fusion pores can be probed in a biochemically defined system. Three isoleucines in the Syb2 TMD that contact Stx1 TMD residues in the crystal structure were mutated to alanines, and the fusion rate and individual pore properties were monitored. The manipulation reduced the fusion rate moderately, but increased pore lifetimes 10-fold, from ~6 s to ~60 s. Replacing the entire TMD with that of a non-exocytotic v-SNARE or a lipid anchor spanning the entire bilayer resulted in qualitatively similar outcomes. These results suggest that specific interactions between Syb2 and Stx1 TMDs are not essential, but may help fine-tune the fusion reaction.

In a later study, Wu et al. (2017) found that pore dilation does not rely on putative v- and t-SNARE TMD interactions, but rather their results support a dilation mechanism from entropic forces generated by crowding of SNARE complexes at the fusion pore.

Clusters formed by neuronal t-SNAREs are cholesterol-dependent in artificial membranes (Bacia et al., 2004; Murray and Tamm, 2009; Milovanovic et al., 2015) and in live neuroendocrine cells (Lang et al., 2001). Unlike raft-associated proteins, t-SNAREs were found to be enriched in cholesterol-poor membrane regions. Milovanovic et al. (2015) showed that the TMD of Stx1 alone is sufficient for cholesterol-dependent clustering and argued that this effect originated in hydrophobic mismatch. Cholesterol-rich membrane regions tend to form thicker liquid-ordered L_o domains, whereas cholesterol-poor regions tend to form thinner, liquid-disordered (L_d) domains. Remarkably, the TMD length of most membrane proteins matches the thickness of the membrane in which they normally reside (Mitra et al., 2004; Sharpe et al., 2010), while neuronal/exocytotic SNAREs appear to be exceptions in that their TMDs are considerably shorter than the thickness of the plasma membrane (Sharpe et al., 2010). This length mismatch may introduce lipid-packing defects that can be minimized if the offending TMDs are clustered (Milovanovic et al., 2015).

Addition of even a single charged residue to the lumenal C-terminus of Syb2 inhibits fusion (Ngatchou et al., 2010). In contrast, Syb2 fused to a pH-sensitive GFP via a flexible linker sequence composed of S and G supports exocytosis (Miesenböck et al., 1998). Ngatchou et al. (2010) interpreted this as evidence that a few of the lumenal Syb2 residues adjacent to the TMD may need to move toward the hydrophobic core of the membrane. This would destabilize the vesicular membrane and help open a fusion pore. D’Agostino et al. (2016) argued that at least during yeast vacuolar fusion this “penetration model” is unlikely to hold.

Conformational flexibility of the TMDs has been proposed to play a role in fusion (Langosch et al., 2007; Stelzer et al., 2008). Atomistic simulations of a v-SNARE C-terminal fragment in a membrane identified three types of flexibility possessed by the α-helical linker-TMD regions: (i) tilt relative to the membrane normal; (ii) a kink feature at the Gly100 residue; and (iii) conformational flexibility of the entire backbone (Blanchard et al., 2014; Han et al., 2016b). The tilt and kink angles were uncorrelated, as expected for a flexible TMD, yet confined to a narrow range ~10⁰ (Blanchard et al., 2014). Similar kinked conformations were seen in simulations with CG representations of TMDs and lipids (Durrieu et al., 2009; Lindau et al., 2012), and in an atomistic study of a t-SNARE C-terminal fragment (Knecht and Grubmuller, 2003).

Taken together, simulations and experiments suggest that TMD conformational flexibility (in particular kinking and/or backbone flexibility—types (ii) and (iii) above) promotes exocytosis. In chromaffin cell experiments with the Syb2 native TMD replaced by sequences containing only Val, Ile, or Leu, β-branched residue content correlated with restoration of secretion (Dhara et al., 2016). Simulations appear to identify flexibility as the relevant property, because polyI and polyV substitutions increased simulated backbone TMD flexibility, while polyL substitution decreased the flexibility (Han et al., 2016b). By contrast, in these simulations all three substitutions reduced the tilt and straightened the TMD compared to wildtype, while fluctuations in the associated angles were reduced. These results suggest tilt and kink flexibilities are of minor importance to pore dilation. However, mutations of the Syb2 residues Gly100 and Cys103 to Val that would be expected to reduce the N-terminal kinking resulted in a decrease in exocytosis and pores that expanded and closed faster following partial release (Hastoy et al., 2017). Therefore, both types of flexibility (kinking and backbone fluctuations) appear to be relevant.

Thus, increased TMD flexibility may favor pore nucleation and expansion. What is the underlying mechanism? Greater TMD flexibility in the N-terminal portion of the TMD might splay lipids and so relieve the high negative curvature in the outer phospholipid leaflet in a small fusion pore, thereby promoting pore nucleation (Dhara et al., 2016). Dhara et al. (2016) also proposed that this mechanism would promote pore expansion, but this is less clear since pore geometry is complex. Consistent with the ability of TMDs to moderate lipid ordering, TMDs disturbed lipid ordering in their vicinity in atomistic and Martini simulations (Risselada et al., 2011; Han et al., 2016b). TMD flexibility may also promote fusion more directly, by helping membranes to assume configurations required to navigate pathways to fusion. Further, several known or potential TMD-TMD interactions may be affected by Syb2 TMD mutations. One might expect that any specific interactions would be disrupted by replacing the TMD with a simple I, L, or V repeat sequence (Dhara et al., 2016). However, specificity in TMD-TMD interactions may be encoded in just a single TMD residue in a cellular context, likely due to packing interactions (Heim et al., 2015).

Many questions remain unanswered regarding the role of TMD flexibility in fusion. For example, if indeed greater flexibility translates to more frequent and faster exocytosis, one might anticipate that lipid-anchored Syb2 would provide the most efficient fusion. However, no such enhancement is seen when Syb2 is anchored by palmitoylated cysteine string protein (CSP) in chromaffin cells (Chang et al., 2016; Dhara et al., 2016), neurons (Chang et al., 2016), or when a lipid anchor that spans the entire bilayer is used (McNew et al., 2000; Wu et al., 2016, 2017).

A mathematical model suggested that another possible mechanism whereby TMD flexibility could promote fusion is by enhancing entropic forces tending to expand the pore. In the model, increased TMD flexibility would increase orientational fluctuations of and mutual steric interactions among the cis-SNARE complexes, increasing the entropic pore expansion force (Wu et al., 2017).

The mathematical model of Wu et al. (2017) identified a critical role for specific and non-specific TMD-TMD interactions in fusion. These interactions drive zippering of cis-SNARE complexes to the fusion pore waist, forcing the SNAREs to interact sterically and thus generating entropic forces that drive pore expansion.

TMD interactions may be important at other stages of fusion. Simulations show v-SNARE TMDs interact and can homodimerize or form higher order oligomers (Fleming and Engelman, 2001; Han et al., 2015, 2016a), with an interaction energy of ~10 k_BT between v-SNARE C-terminal fragments measured in a hybrid atomistic-MARTINI approach (Han et al., 2015). A MARTINI study suggested that the fusion pathway passes through a hemifused state with a HD, and that homodimerization of SNARE TMDs restricts the HD to remain small and therefore to transit more readily to a fusion pore (Risselada et al., 2011).

Interactions between lipids and SNARE TMDs or JMRs may assist fusion. In Martini simulations, post fusion SNARE complexes surrounding the fusion pore were constrained to retain their Y shape by the energy penalty associated with moving the C-terminal polar residues through the hydrophobic membrane core (Risselada et al., 2011). Thus, the bending energy stored in the C-terminal portion of the complexes could be released only by pore expansion. Other MARTINI and hybrid atomistic/CG studies have shown that PI(4,5)P₂ concentrates at t-SNARE JMRs due to interactions with the charged Lys and Arg residues (Khelashvili et al., 2012; Sharma et al., 2015). These effects are thought to help cluster neuronal t-SNAREs (van den Bogaart et al., 2011).