PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of elifeeLifeRecent contentAbout eLifeFor authorsSign up for alerts
 
eLife. 2017; 6: e22964.
PMCID: PMC5404929

Dilation of fusion pores by crowding of SNARE proteins

Axel T Brunger, Reviewing editor
Axel T Brunger, Stanford University Medical Center, United States;

Abstract

Hormones and neurotransmitters are released through fluctuating exocytotic fusion pores that can flicker open and shut multiple times. Cargo release and vesicle recycling depend on the fate of the pore, which may reseal or dilate irreversibly. Pore nucleation requires zippering between vesicle-associated v-SNAREs and target membrane t-SNAREs, but the mechanisms governing the subsequent pore dilation are not understood. Here, we probed the dilation of single fusion pores using v-SNARE-reconstituted ~23-nm-diameter discoidal nanolipoprotein particles (vNLPs) as fusion partners with cells ectopically expressing cognate, 'flipped' t-SNAREs. Pore nucleation required a minimum of two v-SNAREs per NLP face, and further increases in v-SNARE copy numbers did not affect nucleation rate. By contrast, the probability of pore dilation increased with increasing v-SNARE copies and was far from saturating at 15 v-SNARE copies per face, the NLP capacity. Our experimental and computational results suggest that SNARE availability may be pivotal in determining whether neurotransmitters or hormones are released through a transient ('kiss and run') or an irreversibly dilating pore (full fusion).

DOI: http://dx.doi.org/10.7554/eLife.22964.001

Research Organism: E. coli

Introduction

Fusion pores are nanoscale connections between membrane-enclosed compartments that are key intermediates during membrane fusion reactions such as the exocytotic release of neurotransmitters and hormones (Lindau et al., 2003). Following nucleation by specialized proteins (Chernomordik and Kozlov, 2008), fusion pores flicker repeatedly and then dilate or reseal during the release of hormones (Lindau et al., 2003) or neurotransmitters (Alabi and Tsien, 2013; He and Wu, 2007; Staal et al., 2004), or during fusion mediated by viral proteins (Cohen and Melikyan, 2004). The mechanisms that govern these behaviors are poorly understood, despite the availability of sensitive electrical and electrochemical methods to detect single fusion pores during protein-free fusion (Chanturiya et al., 1997; Mellander et al., 2014), viral-protein-induced fusion (Cohen and Melikyan, 2004), and exocytosis (Lindau, 2012). Even the very nature of the fusion pore intermediate (whether lipid- or protein-lined) is debated (Bao et al., 2016).

During exocytotic neurotransmitter or hormone release, a fusion pore opens as vesicle-associated soluble N-ethylmaleimide-sensitive factor attachment protein receptors (v-SNAREs) pair with cognate t-SNAREs on the target plasma membrane (Südhof and Rothman, 2009). This is a tightly regulated process that requires the coordinated actions of several proteins, including Munc18, Munc13, and others (Rizo and Xu, 2015). Complex formation between the v- and t-SNAREs is likely to start at the membrane distal N-termini and may proceed in stages toward the membrane-proximal regions (Gao et al., 2012). Assembly of the SNARE domains results in a four-helix bundle (SNAREpin) that brings the two bilayers into close proximity, but assembly is thought to be halted at some stage to poise vesicles for fast release. Calcium influx in response to depolarization is thought to lead to further SNARE assembly that promotes pore nucleation. This last step — coupling calcium entry to fusion — also requires Synaptotagmin and Complexin, which may actively contribute to pore opening. The initial fusion pore is a metastable structure that may reseal without ever dilating beyond ~1–2 nm in size. This results in transient ‘kiss and run’ exocytosis, a well-established mode of fusion for hormone-secreting cells (Alabi and Tsien, 2013; Hanna et al., 2009; Fulop et al., 2005). By contrast, whether transient fusion is a relevant mode of release for synaptic vesicle fusion has been a subject of debate (Alabi and Tsien, 2013; He and Wu, 2007; Staal et al., 2004; Pawlu et al., 2004; Chapochnikov et al., 2014), mainly because of technical challenges in probing fusion pores directly during synaptic release. For both neuronal and endocrine release, little is known about the molecular mechanisms that govern pore dilation and that set the balance between transient and full fusion (Alabi and Tsien, 2013), in large part due to a lack of biochemically defined assays that are sensitive to single-pores.

Here, using a recently developed nanodisc-cell fusion system (Wu et al., 2016), we found that the presence of just a few SNARE complexes can nucleate a pore, but that reliable pore dilation necessitates many more.

Results

Fusion between v-SNARE reconstituted nanolipoprotein particles and flipped t-SNARE cells

We used 21–27 nm diameter nanolipoprotein particles (NLPs) (Bello et al., 2016) to determine whether SNAREs alone can catalyze pore dilation. By contrast, most previous studies employed much smaller, 6–18-nm diameter nanodiscs (NDs) (Bao et al., 2016; Wu et al., 2016; Shi et al., 2012), whose dimensions restricted pore diameters to 4 nm (Wu et al., 2016) and SNARE copy numbers to 9, precluding studies of pore dilation (Bello et al., 2016). We incorporated v-SNAREs into NLPs that were stabilized by a recombinant apolipoprotein E variant consisting of the N-terminal 22-kDa fragment (ApoE422k). We varied the lipid-to-ApoE422k ratio to control the NLP size and the VAMP2-to-ApoE422k ratio to tune the number of v-SNARE copies per NLP (Bello et al., 2016) (Figure 1). We confirmed that vNLPs fused with liposomes that were reconstituted with t-SNAREs in a SNARE-dependent manner using a previously described bulk assay that monitors calcium release through pores connecting v-SNARE nanodiscs with t-SNARE liposomes (Bello et al., 2016; Shi et al., 2012) (Figure 1—figure supplement 1a). Although NLP pores could in principle grow to 10 nm in diameter (Figure 4b), much larger than the ~4 nm allowed by the membrane scaffold protein (MSP) based small ND geometry (Wu et al., 2016; Shi et al., 2012), bulk calcium release rates were comparable between vNLP and vMSP NDs loaded with similar v-SNARE copy numbers (Figure 1—figure supplement 1b), confirming that the bulk assay is largely insensitive to pore properties under these conditions (Bello et al., 2016).

Figure 1.
Size separation and characterization of NLPs.

We then confirmed lipid mixing between the membranes of vNLPs and flipped t-SNARE cells (tCells) using a previously described protocol (Wu et al., 2016) (Figure 2). NLPs co-labeled with one mole % each of DiI (donor) and DiD (acceptor) were incubated with tCells for 30 min at 4°C, a temperature that allows docking but not fusion. Cells were then rinsed to remove free NLPs and mounted onto a confocal microscope stage held at 37°C to initiate fusion and imaging of DiI and (directly excited) DiD fluorescence. At the concentrations used, the DiI fluorescence is quenched by DiD when the dyes are initially in the NLP membrane. Upon fusion, the dyes become diluted in the plasma membrane and the DiI fluorescence increases. Directly excitated DiD fluorescence provides a measure of the amount of docked NLPs. The ratio of the DiI to DiD intensity normalizes the lipid-mixing signal to the amount of docked NLPs. Normalized lipid-mixing signals increased when tCells were incubated with vNLPs carrying eight v-SNAREs total (vNLP8), but not with empty NLPs (eNLP) or with NLPs loaded with eight copies of a v-SNARE construct, VAMP2-4X, carrying mutations in the C-terminal hydrophobic layers (L70D, A74R, A81D, and L84D) (Figure 2b). These mutations prevent zippering of the C-terminal half of the SNARE domains (Wu et al., 2016; Krishnakumar et al., 2011, 2013), a perturbation that does not affect docking (Figure 2c) but prevents fusion (Figure 2b,d) (Wu et al., 2016; Krishnakumar et al., 2013). A variation of the assay that avoided the 4°C incubation but prevented live imaging in the presence of the labeled NLPs confirmed these results (Figure 2d). We estimate that 4–5% of the docked vNLPs undergo fusion with the flipped t-SNARE cells over the course of ~20 min (Figure 2—figure supplement 1). In comparison, fusion between v-SNARE NLPs and t-SNARE liposomes yields a similar extent of lipid mixing over the same period (Bello et al., 2016).

Figure 2.
vNLPs induce lipid mixing when incubated with flipped t-SNARE cells (tCells).

Lipid mixing could result from the merging of only the proximal lipid bilayer leaflets of the vNLPs and the tCells. To test whether full fusion occurred, we loaded the cells with Fluo-4, a fluorescent calcium probe, and monitored calcium signals. If full fusion occurred, then calcium influx through the fusion pores connecting vNLP and tCell membranes should increase cytosolic Fluo-4 signals (Wu et al., 2016). This was indeed the case for vNLP8 samples, but not for empty NLPs or for NLPs loaded with VAMP2-4X (Figure 3). Using this calcium-influx assay, we also assessed whether pores eventually resealed by washing away the free vNLPs after 5 min of incubation. Cellular calcium levels returned to baseline within a few minutes, suggesting that pores eventually resealed (Figure 3—figure supplement 1).

Figure 3.
Calcium-influx assay.

Dynamics of single fusion pores

Next, we probed single pores connecting vNLPs to tCell membranes (Wu et al., 2016) (Figure 4). We voltage-clamped a tCell, in the cell-attached configuration, that was ectopically expressing ‘flipped’ neuronal/exocytotic t-SNAREs syntaxin1 and SNAP25 (Hu et al., 2003) (Figure 4). NLPs that were reconstituted with eight copies of the complementary neuronal v-SNARE VAMP2/synaptobrevin (vNLPs) were included in the pipette solution (100 nM vNLPs, 120 µM lipid); they diffused to the pipette tip and fused with the patch. Because an NLP is not a closed structure like a vesicle, its fusion with the voltage-clamped membrane patch establishes a direct conduction pathway between the cytosol and the pipette solution, leading to direct-currents whose magnitude reflects pore geometry (Wu et al., 2016).

Figure 4.
Detection of single-pores between vNLP nanodiscs and tCells.

Currents appeared in bursts with an average frequency of ~0.2 bursts per min, or ~2.5 per patch (Figure 4c). The very low burst frequency, together with small unitary conductances (see below), strongly suggest that each burst represents currents passing through a single pore (Wu et al., 2016). Currents fluctuated and returned to baseline multiple times, as if the fusion pore fluctuated in size and opened and closed repeatedly, i.e. flickered (Figure 4d and Materials and methods). Nucleation was blocked when the cytoplasmic domain of VAMP2 (CDV) was included in the pipette solution, vNLPs were treated with the tetanus neurotoxin light chain (TeNT), or when empty NLPs (eNLPs) were used (Figure 4c). CDV competes with the NLP v-SNAREs for binding to the flipped t-SNAREs on the patch surface, and TeNT cleaves VAMP2. When we used VAMP2-4X, pore nucleation rate was not significantly different than for any of the other negative controls (Figure 4c). Because this construct allows efficient docking (Figure 2c) but is fusion incompetent (Figure 2b,d) (Wu et al., 2016; Krishnakumar et al., 2013), this result indicates that ApoE does not induce pores even when kept in close proximity to the target membrane. Collectively, these observations indicate that, similar to their smaller ND courterparts (Wu et al., 2016), NLPs fuse with liposomes or cell membranes in a strictly SNARE-dependent manner.

Combining data from 64 current bursts, we obtained distributions for vNLP8-tCell fusion pores as shown in Figure 4e–f and Figure 4—figure supplement 1. The number of pore flickers and burst durations were well described by geometric and exponential distributions, respectively, with Nflickers=16±2.7 flickers per burst and To=10.3±2.2 s (mean ± S.E.M.), as would be expected for discrete transitions between open, transiently blocked, and closed states (Sakmann and Neher, 2009) (Figure 4—figure supplement 1).

Conductances in the open-state and corresponding radii were broadly distributed (Materials and methods and Figure 4e,f), with mean Gpo=450 pS (S.E.M. = 169 pS), and rpo=0.84 nm (S.E.M =0.09 nm), respectively. Surprisingly, these values were significantly less than the maximum possible value predicted from NLP dimensions (Figure 4b). This suggested a substantial inherent resistance to pore expansion, independent of the constraints imposed by the NLP dimensions. To quantify the resistance, we computed the apparent pore free energy U(rpo) from the distribution of pore radii, P(rpo) ~ eU(rpo)kT. This suggested that ~2 kT energy was required for every 1 nm increase in pore radius above the most likely value rpo0.5 nm (Figure 4f,g).

A few SNARE complexes are sufficient to create a fusion pore, but many more are needed to dilate it

We then varied the number of SNAREs, and found that just a few SNARE complexes are sufficient to create a fusion pore, but many more are needed to dilate it. We repeated the measurements shown in Figure 4 using NLPs loaded with total v-SNARE copy numbers ranging from 1 (vNLP1) to ~30 (vNLP30, ~15 copies per face) (Figure 5). Pore nucleation required at least two v-SNAREs per NLP face and maximal nucleation rates were reached at around the same value (Figure 5a). By contrast, when four v-SNAREs per NLP face were present, the pore conductances (Figure 5b) and radii (Figure 5—figure supplement 1b) were significantly larger than the SNARE-free values and increased dramatically as the copy number per NLP face reached 15. Conductance fluctuations about the mean increased even more sharply (Figure 5—figure supplement 1a), while burst lifetimes and pore open probability showed a more gradual increase (Figure 5—figure supplement 1c,d). Thus, different numbers of SNARE complexes cooperate at the distinct stages of fusion pore nucleation and pore dilation.

Figure 5.
Only a few SNARE complexes are required to nucleate a pore, but more than ~15 are required to dilate it.

Is the increase in the mean pore conductance as the SNARE copy numbers are increased (Figure 5b) due to the appearance of multiple small pores per NLP or due to an increase in the mean size of a single pore? The latter is much more likely, for the following reasons. First, a probe that cannot pass through small pores becomes permeant to pores when large copy number vNLPs are employed (Figure 5—figure supplement 4). If multiple small pores were present when vNLP30 are used, then the probe should be equally impermeant. The probe employed was N-methyl-D-glucamine (NMDG+), a large ion of ~1.1×0.5 nm in size without its hydration shell (Melikov et al., 2001), which replaced sodium in the pipette solution. Conductance was low when ~15 nm MSP nanodiscs with eight copies of v-SNAREs (vMSP8) were used (Wu et al., 2016), but not affected when ~23 nm NLPs bearing 30 v-SNAREs were employed (vNLP30). These results are consistent with those of Bello et al. (2016), who showed that progressively larger cargo could be released from t-SNARE liposomes during fusion with vNLPs as the v-SNARE copies per NLP was increased. Second, conductance of n small pores in a single NLP would be additive, giving total conductance equal to Gpo=n × gpo, where gpo is the mean open-pore conductance of a small pore. Doubling the SNARE copies would presumably at most double n, and by consequence, total conductance. The fact that we find faster than linear increase in mean pore conductance (Figure 5b) is consistent with each NLP bearing a single pore whose size increases with increasing SNARE copies. Third, if multiple small pores occurred per NLP, this should be evident in the distribution of point-by-point conductance values, with peaks at n × gpo, where n=1, 2, 3. Instead, for the distribution of mean Gpo for vNLP30 we find a peak at ~300 pS, and a broad peak at ~3–14 nS (Figure 6b). If the typical small pore has 300 pS conductance, then to have ~6 nS (typical large conductance), there would have to be ~20 small pores per NLP. It is hard to imagine that this many pores could coexist in this small area. Finally, unless the multiple pores occurred simultaneously, we would also find that the fusion rate increases with copy number. Instead, the rate saturates at around two copies (Figure 5a). In conclusion, although we cannot rule out that, very occasionally, a small number of pores may simultaneously appear in a single NLP, all the evidence suggests that this cannot be very common.

Figure 6.
Increasing v-SNARE copy numbers increases the occurrence of large pores.

Previous reports suggested that pore nucleation is promoted by the assembly of the v- and t-SNARE transmembrane domains (TMDs) (Wu et al., 2016; Shi et al., 2012; Fdez et al., 2010). To test whether pore dilation also required the TMDs, we replaced the v-SNARE TMDs with lipid anchors (Shi et al., 2012). We used long-chain anchors that span both leaflets of the bilayer because previous work suggested that lipid anchors spanning a single leaflet are not efficient in inducing full fusion (McNew et al., 2000; Chang et al., 2016). Lipid-anchoring VAMP2 into NLPs significantly reduced pore nucleation frequency and increased the mean burst duration (Figure 5—figure supplement 2a,b,e), consistent with previous work using smaller MSP NDs (Wu et al., 2016) and with reduced overall fusion efficiency reported for lipid-anchored VAMP2 (Shi et al., 2012; Chang et al., 2016). Importantly however, fusion pores induced by lipid-anchored v-SNAREs displayed the same trends as their intact counterparts: as VAMP2-C45 copy number increased, so did mean conductance, fluctuations, burst lifetimes, and pore radii, but the pore open probability during a burst varied little (Figure 5—figure supplement 2). These results suggest that specific interactions between v-SNARE and t-SNARE TMDs are not critical for cooperative pore dilation by SNARE proteins.

The target membrane during exocytosis (where the t-SNAREs reside) is the inner leaflet of the plasma membrane, which is rich in acidic phospholipids. By contrast, in our vNLP-tCell fusion assay, the target membrane is the outer leaflet of the plasma membrane, which is largely devoid of negatively charged lipids. In general, a limitation of our system is that the lipid composition of the outer leaflet of the t-SNARE-presenting cell differs substantially from that of the plasma membrane inner leaflet, and lipid composition can play a key role in fusion. To test whether the target membrane composition affected fusion, we swapped the locations of the v- and t-SNAREs and fused flipped v-SNARE cells with t-SNARE NLPs. This allowed us to have a better mimic of the inner plasma membrane leaflet composition on the target membrane (now the tNLP membrane). This swap resulted in similar fusion rates and pore properties for two different SNARE copy numbers per NLP (Figure 5—figure supplement 3), suggesting that fusion mediated by SNAREs alone may not be very sensitive to target membrane composition within a certain range (Stratton et al., 2016).

Interestingly, the increase in mean pore conductance as v-SNARE copy numbers are increased does not occur homogeneously across all pores. For vNLPs bearing four or more total copies of v-SNAREs, we found two types of fusion pores. Most had small mean conductance 1 nS, but with increasing SNARE load, an increasing fraction of pores had much larger conductances of a few nS (Figure 6a). By contrast, for NLPs that contained no copies or just one copy of VAMP2, the pores that occasionally occurred all had small mean conductance 1 nS. The distribution of mean conductances for individual pores revealed two components for vNLP30 (Figure 6b), with a sharp peak at ~300 pS (71% of total mass) and a much broader population centered at 7.21 nS, separated at ~1 nS. Conductance fluctuations increased sharply for larger pores with Gpo1 nS (Figure 6—figure supplement 1), indicating a change in behavior above this threshold. Thus, multiple criteria indicated ~1 nS as a cut-off that separated small and large pores. We applied this cut-off to all NLPs tested, and clustered pore conductances in each NLP group into low (Gpo<1 nS) and high conductance (Gpo>1 nS) states (Figure 6c). The occurrence of high conductance pores increased with increasing SNARE copy number, suggesting that dilation of the pores to >1 nS (corresponding to rpo1.7 nm) is facilitated by SNARE crowding.

We defined the pore dilation probability, Pdilation, as the fraction of pores in the high-conductance state for a given SNARE copy number (Figure 6c), and plotted Pdilation as a function of v-SNARE copies per face (Figure 6d). Even at the maximum SNARE load of ~15 copies per face, Pdilation was ~0.30, far from saturating. To estimate how many SNARE complexes would be required to reach saturation, we assumed that the ratio between the probabilities of high and low conductance states, Pdilation/(1Pdilation) is equal to a Boltzmann factor eΔE/kT, with ΔE the difference between the energy levels of the two conductance states, and kT thermal energy. Making the simplest assumption that ΔENSNAREN0, where NSNARE is the number of SNARE complexes involved and N0 is the copy number that would make Pdilation=0.5, we found N0=19.3 (Figure 6d). Thus, pore dilation with Pdilation=0.90 would require N30 complexes (Figure 6d, open black circle), about 10-fold more than is required for nucleation (Figure 5a).

SNARE crowding generates entropic forces that drive pore expansion

The characteristic fusion pore free-energy function U(rpo) progressively softened as the SNARE copy number increased (Figure 7a). The minimum at rpo0.5 nm did not shift, but for larger pore radii, the slope decreased and the profile broadened. These free-energy profiles quantify how fusion-pore dilation is driven by SNARE proteins. For example, an energy ~6 kT is required to expand the one-SNARE fusion pore from its preferred radius of ~0.5 nm to a 3-fold larger pore, showing that such an expansion is unlikely to occur spontaneously. On the other hand, with four SNAREs per face, the same expansion requires only ~3 kT, and only ~2 kT with 15 SNAREs (Figure 7a), bringing the expansion within reach of spontaneous fluctuations. The broad and shallow profile suggests that a fully developed pore would then result, if the NLP were replaced by a vesicle as in the physiological setting (Figure 7a). These effects can equivalently be phrased in terms of force: ~24 pN opposes pore expansion with one SNARE, but this is lowered to ~8 pN and ~5 pN by 4 and 15 SNAREs per face, respectively.

Figure 7.
Free energy profiles for pore dilation, experimental results and model predictions.

To help to elucidate the molecular mechanism underlying SNARE-mediated pore dilation, we developed a coarse-grained mathematical model that assumed that the bilayer-SNARE system is equilibrated, consistent with the long-lived current bursts, so that U(rpo) is then the true thermodynamic free energy (see Materials and methods and Table 1 for model parameters). In our model, for a given pore size, the free energy represents an average over possible fusion pore heights, pore shapes and SNARE complex configurations. SNAREs can be fully zippered at the fusion pore waist, or they may unzipper and roam (Figure 7—figure supplement 1).

Table 1.
Parameters used in the analytical model of fusion pores. (A) Measured by fitting a cylinder to the part of the fully zippered SNARE protein structure without the TMDs, produced using PyMOL software with PDB code 3HD7 (http://www.pymol.org ...

The model free energies reproduced experimental profiles with physiologically realistic parameters (Figure 7b). Protein-free pores resisted expansion because a bigger pore has greater area: a 1 nm increase in pore radius required ~3.0 kT work against membrane tension and increased membrane bending energy by ~2.4 kT. Thus, a net force ~22 pN resists pore expansion, close to the experimental value of ~24 pN (Figure 7a). When SNAREs were present, zippering of SNARE linker domains (Gao et al., 2012) and TMDs drove several SNAREs to fully assemble at the fusion pore waist, where crowding generated an entropic pore expansion force (Figure 7c). Bigger pores were associated with more zippered SNAREs at the waist (Figure 7—figure supplement 1). With 15 SNAREs per face, the entropic expansion force of 17 pN was within 5 pN of the 22 pN resistance.

Consistent with our experiments using lipid-anchored v-SNAREs, when we ran the model with lowered excluded volume among zippered SNAREs to mimic the lipid anchor, the best fit total zippering energy was reduced by <1 kT (see Materials and methods). This suggests that the driving force for SNARE zippering that underlies pore expansion does not rely on putative v- and t-SNARE TMD interactions, but has a significant contribution from non-specific interactions that favor the alignment of membrane anchors.

Discussion

In summary, we find that a few SNAREs can nucleate a fusion pore, consistent with previous findings (Bao et al., 2016; Shi et al., 2012; Sinha et al., 2011; Mohrmann et al., 2010; van den Bogaart et al., 2010; Karatekin et al., 2010), but the pore is highly unlikely to expand significantly without additional forces. Pore dilation is resisted by intrinsic bilayer properties (Chanturiya et al., 1997; Chizmadzhev et al., 1995; Jackson, 2009), but promoted by the action of many SNAREs that cooperatively exert expansion forces of entropic origin (Figure 7c).

In our study, pores fluctuated in size, and closed and opened (flickered) multiple times before resealing, as do exocytotic fusion pores that have been recorded from neuroendocrine cells or neurons (Lindau et al., 2003; Staal et al., 2004; Fulop et al., 2005; He et al., 2006; Klyachko and Jackson, 2002). Each such burst lasted several seconds on average. Confinement of the pore radius by the NLP scaffold to 7–8 nm probably contributed to this long lifetime. However, we suggest the lifetime also reflects the natural tendency of small pores to resist dilation, for the following reasons. First, the most likely pore radius, ~0.5 nm (Figures 4f and and7a),7a), is much smaller than the maximum allowed by the NLP geometry, ~7–8 nm. Second, fusion pores connecting protein-free bilayers flicker for seconds, and do not dilate unless increased membrane tension is applied (Chanturiya et al., 1997). Third, pores lasting of the order of a second or longer have been documented during exocytosis using capacitance recordings made, for example, in beta cells secreting insulin (Hanna et al., 2009; MacDonald et al., 2006) or during synaptic vesicle fusion (He et al., 2006). Amperometry often reports shorter pore lifetimes when compared to capacitance measurements (Chang et al., 2015); it may underestimate pore lifetimes because no signal can be observed once all cargo is released. It is also possible that a pore can reseal after partial dilation giving rise to an amperometric spike, leading to an underestimation of pore lifetime based on the pre-spike foot feature alone (Mellander et al., 2012). Fourth, a recent FRET-based study suggested the existence of long-lived, narrow fusion pores during neuronal SNARE-driven fusion between surface-tethered liposomes, dilation of which was promoted by Synaptotagmin-Ca2+ and Complexin (Lai et al., 2013). Finally, various theoretical models suggested that small pores are metastable (Chizmadzhev et al., 1995; Jackson, 2009; Nanavati et al., 1992).

We measured a low fusion efficiency in our assay (4–5% of docked NLPs undergo lipid mixing within ~20 min (Figure 2—figure supplement 1), possibly due to the absence of factors known to be essential for exocytosis in our assay. Indeed, the low fusion efficiency in SNARE-only reconstitutions that lack other factors that are required in vivo for exocytosis, such as Munc13, is well documented (Bao et al., 2016; Bello et al., 2016; Weber et al., 1998; Liu et al., 2016; Hernandez et al., 2014; Diao et al., 2013). Another possibility is that the NLPs that actually fuse may be biased toward higher copy numbers of proteins, if higher copy number corresponded to higher fusion rates. While we cannot categorically exclude this possibility, we think it unlikely. First, the fusion rates that we measured were statistically indistinguishable for copy numbers four (vNLP4, vNLP8, vNLP15 and vNLP30, Figure 5a). Thus, there is no evidence of a bias due to differential fusion rates. Second, even if there were such a bias, our results would still correctly report the general trend of pore properties versus copy number, as can be seen from the following argument. Assume a Poisson distribution for the copy number up to a maximum of 30, the maximum attainable value in our experiments (presumably a packing constraint). Then, for large mean copy numbers, this distribution has a small width, so that even if the ~5% fused fraction corresponds to the tail of this distribution, the copy numbers involved will not be much greater than the mean value. Thus, the typical copy number of the NLPs whose pore properties are measured would still be an increasing function of the mean value.

A wide range of SNARE copy number requirements for fusion have been reported in the literature (Bao et al., 2016; Shi et al., 2012; Sinha et al., 2011; Mohrmann et al., 2010; van den Bogaart et al., 2010; Karatekin et al., 2010; Hernandez et al., 2014; Domanska et al., 2010; Montecucco et al., 2005), depending on the system studied and the read-out used for fusion. Most studies concluded that only a few copies of neuronal SNAREs are sufficient for calcium-triggered exocytosis and fusion of small liposomes (Bao et al., 2016; Shi et al., 2012; Sinha et al., 2011; Mohrmann et al., 2010; van den Bogaart et al., 2010; Karatekin et al., 2010; Domanska et al., 2010). Despite this, the average synaptic vesicle carries 70 v-SNARE copies (Takamori et al., 2006) and at least as many t-SNAREs are clustered at plasma membrane docking and fusion sites in neuroendocrine cells (Knowles et al., 2010). Our results provide a rationalization for this situation, as they suggest that reliable pore dilation may require the engagement of many SNARE complexes. As the demands for SNARE cooperativity may be different at different stages of the fusion reaction, interpretation of copy number requirements should be made with caution. Methods that rely on lipid mixing or on the exchange of small ions through pores (e.g. capacitance or pH sensing) may measure the requirement for the opening of small fusion pores, which may differ substantially from the requirements for pore dilation.

The action of SNAREs is highly regulated by other proteins during neurotransmitter or hormone release (Südhof and Rothman, 2009). In addition to manipulations of SNAREs (Han et al., 2004; Fang et al., 2008; Kesavan et al., 2007), mutations in Munc18 (Jorgacevski et al., 2011), Synaptotagmin (Wang et al., 2001, 2003a, 2003b), and complexin (Dhara et al., 2014) affect fusion pore properties, linking these proteins to pore dynamics. Thus, one must be cautious when extending our SNARE-only results to physiological neurotransmitter and hormone release. Nevertheless, the concept of the promotion of pore dilation by protein crowding is a general principle that may hold qualitatively in the presence of additional components of the physiological fusion machinery. Indeed, a previous study indicated that the availability of SNAREs affects neurotransmitter release kinetics in neurons (Acuna et al., 2014), while another suggested release occurred faster at sites with more t-SNAREs (Zhao et al., 2013). Thus, we tentatively suggest that some proteins may exert their exocytotic regulatory function by organizing SNARE complexes around the fusion site and thereby controlling the number that participate, or by sequestering SNAREs to limit that number. Given the steep dilation probability curve (Figure 6d), our results suggest a high sensitivity in the balance between transient versus full fusion.

Materials and methods

Stable flipped SNARE and wild-type HeLa cell culture

HeLa cell lines stably co-expressing flipped v-SNAREs (flipped VAMP2 and cytosolic DsRed2-nes, ‘vCells’) and t-SNAREs (flipped Syntaxin-1 and flipped SNAP-25 and the nuclear fluorescent marker CFP-nls, ‘tCells’) were generated in the Rothman laboratory as described (Giraudo et al., 2006). The cells were a generous gift from the Rothman laboratory. The cells were maintained in DMEM (4500 mg/L glucose, L-glutamine, sodium pyruvate and sodium bicarbonate) and 10% (v/v) fetal calf serum at 37°C. A new aliquot of cryopreserved cells was thawed after at most three weeks of cell culture and cultured at least five days before data acquisition.

The flipped t-SNARE HeLa cells were tested by PCR Mycoplasma Test Kit I/C (cat. No. PK-CA91-1048, Promo Kine, Heidelberg, Germany), which showed contamination (Figure 4—figure supplement 3A). Because our assay fuses discs to the surface of these cells, no effect of mycoplasma contamination is expected. Indeed the controls in Figures 24 indicate that fusion is SNARE-driven. Nevertheless, we tested any possible impact of mycoplasma contamination on our results by repeating some of our single-pore measurements with cells treated with an antimycoplasma reagent (Plasmocin, cat. code ant-mpt, InvivoGen, California, USA). Fusion rates and pore properties were indistinguishable when untreated or treated cells were used (Figure 4—figure supplement 3, B–E), suggesting that mycoplasma contamination does not affect fusion with NLPs.

Plasmids, protein expression and purification

Expression and purification of the t-SNARE complex used in vND-tSUV fusion experiments is described in Parlati et al. (1999). The cytoplasmic domain of VAMP2 (CDV) was produced using a method that is similar to an earlier protocol (Weber et al., 1998), except that a SUMO vector was used. We followed the methods of Giraudo et al. (2006) for expression and purification of the tetanus neurotoxin light chain, TeNT. VAMP2 proteins were expressed and purified as described earlier by Shi et al. (2012). To produce full-length WT VAMP2, we used the plasmid pET-SUMO-VAMP2 (Shi et al., 2012). To produce lipid-anchored VAMP2, we followed the methods of McNew et al. (2000) and Shi et al. (2012). We first used a previously described construct (McNew et al., 2000) to produce VAMP95Cys containing the entire cytoplasmic domain of VAMP2 (residues 1–95) with a C-terminal cysteine residue. We then coupled this protein to maleimidopropionic acid solanesyl ester (maleimide-C45), produced as previously described (Shi et al., 2012). For producing MSP NDs, we used the vector pET28-MSP1E3D1 (Addgene, Cambridge, MA) to express and purify MSP1E3D1 as described previously (Ritchie et al., 2009), except that we cleaved the MSP proteins directly off the column by TEV protease overnight at 4 °C (Wu et al., 2016).

Plasmid pET32a-Trx-His6X-ApoE422K, which we used to express the N-terminal 22 kDa fragment of apolipoprotein E4 (residues 1–199), was kindly provided by Dr Nicholas Fischer, Lawrence Livermore National Laboratory, CA (Morrow et al., 1999; Blanchette et al., 2008). ApoE422K was expressed and purified as previously described (Morrow et al., 1999) with the following modifications. The His6-ApoE422K was cleaved off the Ni-NTA beads (Qiagen, Germantown, MD) using 100U of Thrombin at 4°C overnight. The protein was eluted in 25 mM HEPES, 140 KCl, pH 7.4 buffer containing 1% octylglucoside (OG), and was functional for up to 4 weeks when stored at 4°C. Protein concentrations were determined using the Bradford assay (Bio-Rad, Hercules, CA) with bovine serum albumin as standard.

Characterization of nanolipoprotein particles (NLP)

Details are provided by Bello et al. (2016). Briefly, nanolipoprotein particles containing VAMP2 (vNLP) were produced using a modified version of the established protocol to generate SNARE-nanodiscs (Shi et al., 2012, 2013). A palmitoyl-2-oleoylphosphatidylcholine (POPC): 1,2-dioleoyl phosphatidylserine (DOPS) = 85:15 molar ratio lipid mixture (Avanti Polar Lipids, Alabaster, AL) was dried under nitrogen flow, followed by vacuum for 1 hr. The lipid film was re-suspended in 25 mM HEPES, pH 7.4, 140 mM KCl, buffer with 1% OG supplemented by the desired amount of VAMP2. The mixture was vortexed at room temperature (RT) for 1 hr followed by the addition of ApoE422K and vortexed for another hour at RT. The ApoE422K:VAMP2:lipid ratio was varied to tune the v-SNARE copy number per NLP as 1:0.2:180 (1 VAMP2, ‘vNLP1’); 1:1:180 (four copies, ‘vNLP4’); 1:2:180 (eight copies, ‘vNLP8’); 1:4:180 (15 copies, ‘vNLP15’) and 1:8:180 (30 copies, ‘vNLP30’). NLPs containing 1, 4 and 15 copies of VAMP-C45 were obtained using a similar approach. Excess detergent was removed using SM-2 bio-beads (Bio-Rad) overnight at 4°C with constant mixing. The assembled v-NLPs were separated from free proteins and lipids via gel filtration on a Superose six column (Figure 1a). Samples were concentrated using Amicon Ultra (50 KDa cutoff) centrifugal filter units, and analyzed by SDS-PAGE with Coomassie staining (Figure 1b). The number of VAMP2 copies per disc was determined by the VAMP2-to-ApoE ratio by densitometry using ImageJ (NIH). The number of ApoE copies per disc was estimated (Bello et al., 2016) using the calibration of disc size vs the number of ApoE copies previously reported (Blanchette et al., 2008). The size distribution of the v-NLPs was determined for every batch of production using transmission electron microscopy. To do this, the NLP discs were diluted (1:50), mounted onto carbon-coated 400 mesh copper electron microscopy grids, negatively stained with 2% uranyl acetate (w/v) solution, and subsequently examined in an FEI Tecnai-12 electron microscope operated at 120 kV. Micrographs of the specimen were taken on a Gatan Ultrascan4000 CCD camera at a magnification of 42,000. Typical micrographs and a size distribution are shown in Figure 1c,d. The size of the NLP discs with 1:180 ApoE422K: lipid ratio was typically 24 ± 2 nm (100–200 NLP discs were analysed for every production batch). Representative size distributions are shown as box plots for the conditions tested in Figure 1e. At least three independent batches of NLPs were used per condition. tNLPs were produced in a similar fashion, using a t-SNARE:ApoE:lipid ratio of 0.8:1:180 and 3:1:180 for tNLP4 (four copies of t-SNARE complex Stx/SN25 total per NLP) and tNLP 15 (15 total copies of t-SNAREs per NLP) samples, respectively.

Bulk fusion of NLPs with t-SNARE liposomes

We used a previously established assay (Bello et al., 2016; Shi et al., 2012) to monitor the release of calcium from t-SNARE-reconstituted small unilamellar vesicles (t-SUVs) as they fused with discs loaded with v-SNAREs. 40 μl t-SUVs entrapping 50 mM calcium were mixed with 5 μl vMSP NDs (prepared as described previously [Shi et al., 2012]) or with 10 μl of vNLP-discs in a buffer containing 2 μM of the calcium-sensitive dye mag-fluo-4 (Invitrogen, Carlsbad, CA). The mixture was loaded into a 96-well plate, and the mag-fluo-4 fluorescence (λex= 480 nm, λem=520 nm, 515 cutoff) was recorded by a SpectraMax M5 plate reader (Molecular Devices, Sunnyvale, CA). After 60 min, 15 μl of 5% dodecylmaltoside was added and the mixture was incubated for an additional 20 min to release all remaining entrapped calcium and thus to establish the maximum mag-fluo4 signal. Fusion is reported as percent of maximum fluorescence in Figure 1—figure supplement 1.

Single-cell lipid mixing and calcium-influx assays

These assays were carried out as described in Wu et al. (2016).Briefly, for lipid mixing, tCells were plated in 35 mm poly-D-lysine-coated glass bottom dishes (MatTek Corporation, MA, USA) and vNLP8s were prepared as described above, except that one mole % each of 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (DiI, cat. no. D282, Molecular Probes, Eugene, OR) and 1,1'-dioctadecyl-3,3,3',3'-tetramethylindodicarbocyanine perchlorate (DiD, cat. no. D307, Molecular Probes) fluorescent lipid labels were included in the lipid composition. For each reaction, 15 µl of vNLP8 was added onto tCells (final NLP lipid concentration was ~54 µM) and incubated for 30 min at 4°C, a temperature at which SNARE complexes assemble but cannot drive fusion (Weber et al., 1998). Fusion was started by raising the temperature to 37°C and monitored by the dequenching of the DiI fluorescence using confocal microscopy. As controls, empty nanodiscs (eNLP) or NLPs bearing VAMP2-4X (harboring the mutations L70D, A74R, A81D and L84D) were used (Figure 2b). To estimate the extent of lipid mixing, at the end of some experiments, DiD fluorescence was completely bleached using direct excitation at 647 nm with 100% laser power. This resulted in the maximum possible donor (DiI) intensity, Fmax. We then rescaled the donor fluorescence values F(t) to obtain the fraction of maximum DiI fluorescence: F~(t)=(FF0)/(FmaxF0), where F0 is the minimum at the beginning of acquision.

To assess lipid mixing, we also used an alternative protocol that avoided the cold incubation step; this protocol could not be used for time-course measurements because of the high background resulting from excess NLPs. The same amount of vNLP8 as above was added to tCells at 37°C. After 15 min incubation, excess NLPs were washed, and DiI, DiD and CFP fluorescence levels were acquired using confocal microscopy (Figure 2c,d).

To measure the influx of calcium through fusion pores, tCells were loaded with 5 μM of Fluo-4 AM (Life Technologies, Carlsbad, CA), a cell-permeant calcium-sensitive fluorescent dye, as previously described (Wu et al., 2016). After washing to remove dye that was not taken in by cells, vNLP (15 µl) were added to tCells at 37°C and the influx of calcium was tracked by imaging of Fluo-4 fluorescence using a confocal microscope (Wu et al., 2016).

Electrophysiology

Details are given in Wu et al. (2016). Briefly, flipped t-SNARE HeLa cells (tCells) were cultured in 3 cm dishes. For recordings, a dish was placed in a temperature-controlled holder (TC-202A by Harvard Apparatus (Holliston, MA), or Thermo Plate by Tokai Hit (Shizuoka-ken, Japan)) set at 37°C. Cells were visualized with an inverted microscope (Olympus IX71, Olympus Corp., Waltham, MA) using an Andor DU-885K EMCCD camera controlled by Solis software (Andor, South Windsor, CT). Recording pipettes (borosilicate glass, BF 150-86-10, Sutter Instruments, Novato, CA) were pulled using a model P-1000 pipette puller (Sutter Instruments) and polished using a micro-forge (MF-830, Narishige, Tokyo, Japan). Pipette resistances were 5–10 MΩ in NaCl-based solution. The bathing medium contained: 125 mM NaCl, 4 mM KCl, 2 mM CaCl2, 1 mM MgCl2, and 10 mM HEPES, (pH adjusted to 7.2 using NaOH) for the cell-attached recordings. 10 mM glucose was added to the medium before use. All voltage- and current-clamp recordings were made using a HEKA EPC10 Double USB amplifier (HEKA Elektronik Dr. Schulze GmbH, Lambrecht/Pfalz, Germany), controlled by Patchmaster software (HEKA). Currents were digitized at 20 kHz and filtered at 3 kHz.

To measure SNARE-mediated single fusion pore currents in the cell-attached mode (Yang and Sigworth, 1998), electrodes were filled with the pipette solution composed of 125 mM NaCl, 4 mM KCl, 10mM HEPES, 13 mM or 26 mM tetraethylammonium-Cl (TEA-Cl, K+-channel antagonist), adjusted to pH 7.2 using NaOH. This solution had resistivity of 0.60 Ohm.m, measured using a conductivity cell (DuraProbe, Orion Versa Star, Thermo Scientific). For experiments designed to test the presence of multiple pores connecting large copy number vNLPs to tCells (Figure 5—figure supplement 4), electrodes were filled with a solution containing: 129 mM N-methyl-d-glucamine (NMDG), 10 mM HEPES, 26 mM TEA-Cl, pH adjusted to 7.2 using HCl, resistivity 0.88 Ohm.m, 305 mOsm. The pipette tip was initially filled with 1 μl of NLP-free buffer and back-filled with vNLPs suspended in the same buffer (final [vNLP] = 100 nM, 120 μM lipids). This allowed the establishment of a tight seal (Rseal>10 GOhm) with high success rate, as well as the recording of a stable baseline before the vNLPs diffused to the membrane patch and started fusing with it 2–18 min later. Such a back-filling strategy is typically used in perforated patch measurements (Sakmann and Neher, 2009). All cell-attached recordings were performed using a holding potential of 40 mV relative to bath. With a cell resting membrane potential of 56±7 mV (mean ± S.D., n = 36), this provided 16 mV driving force across the patch membrane.

Analysis of fusion pore data

The analysis of fusion pores is described in detail in Wu et al. (2016).Briefly, we developed an interactive graphical user interface in Matlab to help to identify, crop and process single fusion pore currents. Traces were exported from Patchmaster (HEKA Electronik) to Matlab (Mathworks) and low-pass filtered (280 Hz cutoff); frequencies that were the result of line voltage were removed using notch filtering. Zero phase shift digital filtering algorithms (Matlab Signal Processing Toolbox function filtfilt) were employed to prevent signal distortion. Filtered traces were averaged in blocks of 80 points (125 Hz final bandwidth) to achieve rms baseline noise 0.2 pA. Currents I for which |I|>2.0 pA for at least 250 ms were accepted as fusion pore current bursts. During a burst, rapidly fluctuating currents often returned to baseline multiple times, i.e. pores flickered. To quantify pore flickering, we defined currents <0.25 pA and lasting  60 ms (15 points) as open pores and currents not meeting these criteria as closed. For a given burst, the number of open periods was equal to the number of flickers, Nflickers. To estimate the fusion rate for each recording (i.e. the rate at which current bursts appeared), we counted the number of current bursts that fit the set criteria (current amplitude >2 pA for at least 250 ms) and divided this number by the duration of the recording. Rates from different records (patches) were averaged for each condition. We also refer to this rate as the pore nucleation rate. Periods during which the baseline was not stable were excluded from this analysis. Many recordings ended with what seemed to be currents from overlapping fusion pores. Such end-of-record currents were also excluded because they could also be attributed to a loose seal. Thus, the fusion rates that we report may underestimate the true rates, especially for conditions in which fusion activity was high. For distributions of conductances and radii, we used pore open-state values, denoted by the subscript ‘po’. For Figure 4e,f, we first computed the probability density functions (PDFs) for individual pores using a fixed bin width for all, then averaged these to give equal weight to all pores. All distribution fitting was performed using Matlab Statistics Toolbox functions fitdist or mle, using maximum likelihood estimation. Open-pore conductance values were used point-by-point to estimate the open-pore radii, by approximating the pore as a cylinder and using the expression (Hille, 2001rpo=(ρλGpo/π)1/2, where ρ is the resistivity of the solution, λ=15 nm is the length of the cylinder, and Gpo is the open-pore conductance. For assessing statistical significance when comparing sample means, we used the two-sample t-test or the nonparametric two-sample Kolmogorov-Smirnov test (ttest2 or kstest2, Matlab Statistics Toolbox), as indicated in the figure legends. We considered each single-pore measurement to be a biological replicate.

For clustering average single-pore conductances Gpo for vNLP30 measurements, we used a two-component Gaussian mixture model (Figure 6b) that indicated a boundary between the two components at ~1 nS. Applying this cutoff to all vNLP samples, we produced the boxplot in Figure 6c, where the central red line on each box marks the median, the edges of the box are the 25th and 75th percentiles, and the whiskers extend from q3 +1.5(q3 – q1) to q1 – 1.5(q3 – q1), where q1 and q3 are the 25th and 75th percentiles, respectively. For a given v-SNARE copy number per NLP face, NSNARE, we defined the probability Pdilation of achieving a high-conductance pore (Gpo>1 nS) as the fraction of high conductance pores observed for that copy number. For example, only 3 out of 64 pores were large conductance for vNLP8, which had four copies per face (Figure 6c), hence Pdilation(NSNARE=4)=3/64. We plotted Pdilation as a function of v-SNARE copy number per NLP face in Figure 6d. In Figure 7a, to estimate the energy profiles of fusion pores for a given v-SNARE copy number, we first calculated the probability density function for open-pore radii as in Figure 4f. The probability Pr that the radius is between r and r+Δr is the density at that bin ×Δr, where Δr is the bin width. We estimated the energy U(r) of a pore with radius r by U/kT=ln(Pr)+A, where A is an arbitrary constant.

Mathematical model of the fusion pore between a nanodisc and planar membrane in the presence of SNAREs

Membrane free energy

We modeled the fusion pore as having a toroidal shape formed between a nanolipoprotein particle (NLP) modeled as a planar bilayer of diameter D and the tCell membrane modeled as an infinite planar bilayer, both of which are at a constant membrane tension (Figure 7c). This toroidal assumption is similar to that in previous theoretical studies that assumed a toroidal shape of the fusion pore (Chizmadzhev et al., 2000; Jackson, 2010; Kozlov et al., 1989). Experimental studies also observed an hourglass-shaped fusion pore that could be considered approximately toroidal (Curran et al., 1993; Haluska et al., 2006).

The fusion pore is parametrized by the toroidal shape parameters: the radius of the toroid rpo, which corresponds to the fusion pore radius rpo, and the separation of the membranes at the edge of the NLP h. The fusion pore is completely toroidal at small pore sizes. The free energy of the fusion pore is calculated using the Helfrich energy form, as was used in previous studies (Chizmadzhev et al., 2000; Jackson, 2010; Kozlov et al., 1989)

Umb(rpo,h)=Ubend(rpo,h)+γΔA(rpo,h)
(1)

Here, the energy due to bending is given by

Ubend(rpo,h)=κ2M(2C)2dA
(2)

where κ and C are the bending modulus and the mean curvature of the membrane, respectively. The energy expended to add membrane area due to membrane tension γ is the second term, where ΔA is the change in total membrane area due to pore formation, given by ΔA(rpo,h)=Apo(rpo,h)Arim(rpo,h). Here, Apo is the area of the fusion pore. Arim is the area of both rims of the fusion pore, which is the area that has to be removed from the infinite tCell membrane and the NLP membrane to form the pore. We evaluated all integrals and all areas over the midplane M of the membrane forming the pore to give:

Ubend(rpo,h)=πκ{2(R+H)2HR(R+2H)tan1(R+2HR)4}
(3)
ΔA(rpo,h)= πH((2π4)H+ 2πR)2π(H+R)2
(4)

where H=h/2+δ/2 and R=rpo+δ/2. Values of κ in previous studies range from 10–50 kT (Cohen and Melikyan, 2004; Brochard and Lennon, 1975; Khelashvili et al., 2013; Marsh, 2006). A commonly used value is κ=20kT (Jackson, 2009), which we used here (Table 1). γ was obtained as a best-fit parameter (Table 1).

Free energy due to twisting of the ApoE proteins that line the boundary of the ND

Owing to the finite size of the NLP, toroidal states are not possible for large pores with rpo and h that constitute a sizeable fraction of the NLP diameter, D. These shapes are partially toroidal and come into existence when rpo+h/2+δD/2, where δ is the membrane thickness. The ApoE proteins that line the NLP boundary need to be rotated through an angle ϕ to form these shapes. An example of one such shape is the right-hand side of Figure 7c.

We assumed these proteins exert a constant torque τ per unit length of the NLP boundary to resist this rotation. The ApoE proteins exert no torque in the completely toroidal states as ϕ vanishes for these states. The free energy of these proteins is

UApoE(rpo,h)=τπDϕ
(5)
ϕ(rpo,h)=sin1(rpo+h2+δD2h2), rpo+h2+δD2
(6)

Thus, the ApoE proteins resist pore expansion as ϕ increases with the size of the pore. We obtained τ as a fitting parameter and D was measured in this study (values in Table 1). For further details about the partially toroidal states, please see the penultimate subheading ‘Description of partially toroidal states and calculation of membrane free energy’.

Free energy contribution from short-ranged steric-hydration forces

Steric-hydration repulsion between membranes is prominent at small membrane separation. Experimentally measured steric-hydration pressures between planar membranes of separation d are of the form P0exp(d/λ) (Rand and Parsegian, 1989). Values of P0 and λ have been measured before for several membrane compositions; λ is within 0.1–0.3 nm (Rand and Parsegian, 1989). As the pore sizes over which these effects are appreciable (λ) are very small compared with the NLP diameter D, only toroidal states are considered for this calculation. We obtained P0 from previous studies and λ as a best-fit parameter (Table 1).

The steric-hydration forces act in two orthogonal directions on the membranes comprising the fusion pore: to increase the pore radius (rpo) and the separation between membranes (h). The sum of the work done by these two forces gives the free energy of the steric-hydration interaction

Uhyd(rpo,h)=P0λ(πD24)exp(hλ)+P0(2πl)exp(2rpoλ) (λ2rpo+ (λ2)2)
(7)

where l is the effective pore height, i.e. the height of the section of the pore that contributes substantially to the steric-hydration interaction. To obtain the free energy contribution of the steric-hydration forces, we now calculate the work done by these forces to assemble the fusion pore. We first observe that the work done (W) to bring two patches of membranes of area δA to a separation h from a large distance apart is

W=P0hδAexp(yλ)dy
(8)

The first term in Equation 7 is the work done to separate the planar part of the membranes to a distance h. As the pore area where these forces are relevant (λ2) is very small compared with the NLP area πD2/4, we set the area of the planar region δA=πD2/4 in Equation 8 to obtain the first term. The second term in Equation 7 is the work done to separate the membranes to form a pore of diameter 2rpo. To calculate this, we can imagine that the pore is a cylinder of diameter 2rpo and height l, where l=2λ(h+2δ), since the change in pore diameter over the height l is negligible. Here, δ is the thickness of the bilayer and l is the height over which the cross-sectional diameter of the toroidal shape increases from 2rpo to 2rpo+λ.

To obtain l, we consider the inner surface of the toroidal pore. This shape is formed by revolving the semicircle given by x=(rpo+R)Rcosθ,z=Rsinθ where 900θ900 in the XZ plane about the Z axis. Here, rpo is the radius of the pore and R' is the radius of the semicircle, which is also equal to half of the maximum separation between the heads of the monolayers that line the inner surface of the pore, R'=h/2+δ as can be seen from Figure 7c. l/2 is that value of z at which the cross-sectional radius of the pore x increases to rpo+λ/2. Thus, l is obtained by solving the equation of the semicircle (rpo+h/2+δr)2+(l/2)2=(h/2+δ)2 to first order in λ where r=rpo+λ/2.

To calculate the steric-hydration contribution from a pore of size rpo and height h, one need only consider the pore over that height l at which the cross-sectional diameter of the toroidal shape increases from 2rpo to 2rpo+λ, since λ is the range of the steric-hydration force. Thus, the area of the cylinder is δA(rpo)=2πrpol, and using Equation 8, the work done to set up the pore is P02rpo(2πly)exp(y/λ)dy, giving the second term of Equation 7.

Free energy contribution from SNARE proteins

In this section, we calculate the free energy due to the SNARE proteins. We fix the total number of v-SNAREs. Out of these N v-SNAREs, Nz are fully zippered and Nu are partially zippered. Only the TMDs and the linker regions of these partially zippered SNAREs are unzippered. For each N, we allowed Nz and Nu to vary from 0 to N to obtain an equilibrium distribution for both. We calculated the free energy of the SNAREs for = 0,2,4,8,15, which correspond to the total number of v-SNAREs per NLP face used experimentally (Figure 7a). To match with experiment, we used the assumption that only half of the total number of SNAREs present in the NLP would be present on the side of the NLP that faces the tCell, and that all of these SNAREs would be available to participate in fusion pore expansion.

We assumed that the fully zippered SNAREs form a ring at the waist of the fusion pore (Figure 7c). Their free energy is

Uz(rpo,Nz)=NzkT (ln2πrpoNzbb+1)NzεzipNzkTlnΩz
(9)

The first term is the positional entropy of the zippered TMDs whose diameter is b, which we measured by fitting a cylinder to the measured crystal structure (Stein et al., 2009). The second term is the energy released when a partially zippered SNAREpin completes its zippering. The third term is the orientational entropy associated with the zippered SNAREs. We assume that these are very stiff cylindrical rods. Due to their high stiffness, these rods can only explore a small solid angle Ωz=0.05sr (Table 1). We calculated this angle based on an atomistic molecular dynamics study of the t-SNARE TMD that shows that these domains explore angles of ~100 around their equilibrium position in a bilayer (Knecht and Grubmüller, 2003). The SNAREs fluctuate about their equilibrium orientation, which we assume is the local normal to the membrane.

The free energy of the partially unzippered SNAREs is

Uuz(Nuz)=NuzkT (ln2πDb)NuzkTlnπ
(10)

The first term is the positional entropy of the TMDs. These partially zippered SNAREs are in a Y-shape with both unzippered TMD domains on the same side of the pore, either on the vNLP or the tCell membrane. The linker domains are also unzippered and this imparts flexibility to these SNAREs (Jahn and Scheller, 2006). This orientational freedom is given by the second term. These SNAREs can adopt all orientations in which they do not intersect with the membrane; this corresponds to a solid angle of π steradian. We restricted these SNAREs to a circle of radius equal to that of the NLP, as this considerable orientational freedom is only available when the SNAREpin body is away from the fusion pore lumen.

Calculation of the total free energy U as a function of pore size and number of SNAREs

The probability that a fusion pore accesses a radius rpo in the presence of N SNAREs is proportional to exp(U(rpo,N)/kT) in equilibrium, as this is the Boltzmann distribution where U is the total free energy. We assumed that the bilayer-SNARE system is equilibrated as the current bursts measured experimentally are long-lived. To calculate this free energy, we summed the Boltzmann factor of all states that comprise such a system:

exp(U(rpo,N)kT)=Nz=0N bexp(Utot(rpo,h,N,Nz)kT)dh
(11)

Here, Utot is the total free energy of one fusion pore state with Nz zippered SNAREs and of membrane separation h, given by

Utot=Umb+Uz+Uuz+Uhyd+UApoE
(12)

We performed the integration and the sum over all states in Equation 11 numerically in MATLAB.

Derivation of best-fit model parameters by fitting model-predicted free energy to experiment

We performed a numerical calculation using Equation 11 to obtain free -energy curves as a function of SNARE copy numbers and pore size. For the membrane parameters, it is best to fit to data from membranes with no SNAREs. We fit the fusion pore free energy predicted by the model with no SNAREs (setting = 0 in Equation 11) to the experimentally measured curve for one SNARE, assuming that such a pore behaves similarly to a protein-free pore. We first fit the model-predicted pore size at the minimum in the free energy to experiment by using the steric-hydration force scale λ as a best-fit parameter. We obtained the bending modulus κ from Jackson (2009) and fit the slope following the minimum in the region 0.5 nmrpo1.0 nm using the membrane tension γ as a best-fit parameter, as γ largely determines the slope beyond this minimum.

Using these parameters, we calculated the free energy versus pore radius in the presence of SNAREpins (Figure 7b). As SNAREpins are introduced, the model predicts that the minimum barely shifts, while the slope beyond the minimum decreases with increasing numbers of SNAREs by an amount depending on the zippering energy parameter εzip. We selected a typical experimental curve (vNLP30) and fit the slope of the free energy from simulation to that measured from experiment for 0.5 nmrpo2.5 nm and N=15 SNAREs (as vNLP30 corresponds to 15 SNAREs per face) using εzip as a fitting parameter, and obtained εzip=9.6kT. This is higher than the zippering energy of the linker domains alone, which was measured to be ~5 kT by Gao et al. (2012).

Further increases in pore size cause increases in free energy, as the fusion pore shapes are partially toroidal and the twisting torque from the ApoE proteins at the NLP boundary resists further expansion. Thus, we fit the slope of the free energy curve in this region (4 nmrpo4.5 nm) measured from simulation to that measured experimentally for = 4 SNAREs to obtain the torque per unit length τ=8.43 pN as a fitting parameter.

To understand how lipid-anchored VAMP2 affects pore dilation, we reduced the size of the zippered TMDs by 50% as the zippered SNARE complex lacks the v-SNARE TMD. This is the maximum possible reduction in the excluded volume of the zippered SNAREs. We then varied the zippering energy to ensure that the model-predicted mean pore size was invariant with respect to this reduction, consistent with the invariance in mean pore size observed in experiments (Figure 5—figure supplement 2f). The best-fit zippering energy is 0.43 kT lower than the best-fit value obtained when both TMDs are present. This is an upper bound on the reduction in zippering energy given that we used the largest possible reduction in the excluded volume.

Description of partially toroidal states and calculation of membrane free energy

Due to the finite size of the NLP, toroidal states are not possible for large pores. We instead assumed that these shapes are partially toroidal. These come into existence when rpo+h/2+δD/2, where δ is the membrane thickness. We set the shape of these states as follows. We constructed a toroidal pore with the shape parameters rpo,h, and truncated the top half of the toroid at the plane where the cross-sectional diameter of the toroid is equal to the NLP diameter (Figure 7c, right panel). In these partially toroidal states, the ApoE proteins at the edge of the NLP are rotated through an angle ϕ compared with the fully toroidal states (Equation 6).

Evaluating the integral from Equation 2

Ubend(rpo,h,ϕ)=Ubend(rpo,h,0)+2πκ(1cosϕ)πκ(R+H)2tan1(α)HR(R+2H)
(13)

where H=h/2+δ/2,R=rpo+δ/2,tan1α=tan1(RR+2Hcot(π2ϕ4))tan1(RR+2H), and Ubend(rpo,h,0) is the bending energy in a fully toroidal state, given by Equation 3. The change in total membrane area due to pore formation in these partially toroidal states is given by

ΔApo(rpo,Rpo,ϕ)=πH((R+H)(2πϕ) 2H(1+cosϕ))+π(R+H)2π(D2)2
(14)

In the partially toroidal states, h does not correspond to the membrane separation at the edge of the pore, although rpo is still the pore radius. Thus, we indicate the membrane separation at the NLP boundary by hpo given by

hpo={h2(1+cosϕ),rpo+h2+δD2h,rpo+h2+δ<D2
(15)

Acknowledgements

We thank F Sigworth and D Zenisek for expert advice and discussions, and J E Rothman, M Caplan, and members of the Karatekin and Rothman laboratories for helpful discussions. We are grateful to James E Rothman, Frederick Sigworth, David Zenisek, Yongli Zhang, and Thomas Melia for critically reading the manuscript. Jing Wang kindly provided C45 lipids. This work was supported by NIH grant R01GM108954 and a Kavli Neuroscience Scholar Award to EK. SMA, OB, and SK are members of the Rothman laboratory, supported by NIH R01DK027044. WV was supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), fellowship VE760/1-1.

Funding Statement

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

Funding Information

This paper was supported by the following grants:

  • http://dx.doi.org/10.13039/100000057National Institute of General Medical Sciences R01GM108954 to Erdem Karatekin.
  • http://dx.doi.org/10.13039/100001201Kavli Foundation Neuroscience Scholar Award to Erdem Karatekin.
  • http://dx.doi.org/10.13039/501100001659Deutsche Forschungsgemeinschaft VE760/1-1 to Wensi Vennekate.
  • http://dx.doi.org/10.13039/100000062National Institute of Diabetes and Digestive and Kidney Diseases R01DK027044 to Shyam S Krishnakumar.

Additional information

Competing interests

The authors declare that no competing interests exist.

Author contributions

ZW, Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—review and editing.

ODB, Data curation, Formal analysis, Investigation, Methodology, Writing—review and editing.

ST, Software, Formal analysis, Investigation, Visualization, Methodology, Writing—review and editing.

SMA, Data curation, Formal analysis, Investigation, Visualization, Writing—review and editing.

WV, Formal analysis, Writing—review and editing.

SSK, Supervision, Investigation, Methodology, Writing—review and editing.

BO, Software, Formal analysis, Supervision, Validation, Investigation, Methodology, Writing—review and editing.

EK, Conceptualization, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing.

References

  • Acuna C, Guo Q, Burré J, Sharma M, Sun J, Südhof TC. Microsecond dissection of neurotransmitter release: snare-complex assembly dictates speed and Ca²+ sensitivity. Neuron. 2014;82:1088–1100. doi: 10.1016/j.neuron.2014.04.020. [PMC free article] [PubMed] [Cross Ref]
  • Alabi AA, Tsien RW. Perspectives on kiss-and-run: role in exocytosis, endocytosis, and neurotransmission. Annual Review of Physiology. 2013;75:393–422. doi: 10.1146/annurev-physiol-020911-153305. [PubMed] [Cross Ref]
  • Bao H, Goldschen-Ohm M, Jeggle P, Chanda B, Edwardson JM, Chapman ER. Exocytotic fusion pores are composed of both lipids and proteins. Nature Structural and Molecular Biology. 2016;23:67–73. doi: 10.1038/nsmb.3141. [PMC free article] [PubMed] [Cross Ref]
  • Bello OD, Auclair SM, Rothman JE, Krishnakumar SS. Using ApoE nanolipoprotein Particles to analyze SNARE-Induced fusion pores. Langmuir. 2016;32:3015–3023. doi: 10.1021/acs.langmuir.6b00245. [PMC free article] [PubMed] [Cross Ref]
  • Blanchette CD, Law R, Benner WH, Pesavento JB, Cappuccio JA, Walsworth V, Kuhn EA, Corzett M, Chromy BA, Segelke BW, Coleman MA, Bench G, Hoeprich PD, Sulchek TA. Quantifying size distributions of nanolipoprotein particles with single-particle analysis and molecular dynamic simulations. The Journal of Lipid Research. 2008;49:1420–1430. doi: 10.1194/jlr.M700586-JLR200. [PubMed] [Cross Ref]
  • Brochard F, Lennon JF. Frequency spectrum of the flicker phenomenon in erythrocytes. Journal De Physique. 1975;36:1035–1047. doi: 10.1051/jphys:0197500360110103500. [Cross Ref]
  • Chang CW, Chiang CW, Gaffaney JD, Chapman ER, Jackson MB. Lipid-anchored Synaptobrevin provides little or no support for exocytosis or Liposome fusion. Journal of Biological Chemistry. 2016;291:2848–2857. doi: 10.1074/jbc.M115.701169. [PMC free article] [PubMed] [Cross Ref]
  • Chang CW, Hui E, Bai J, Bruns D, Chapman ER, Jackson MB. A structural role for the synaptobrevin 2 transmembrane domain in dense-core vesicle fusion pores. Journal of Neuroscience. 2015;35:5772–5780. doi: 10.1523/JNEUROSCI.3983-14.2015. [PMC free article] [PubMed] [Cross Ref]
  • Chanturiya A, Chernomordik LV, Zimmerberg J. Flickering fusion pores comparable with initial exocytotic pores occur in protein-free phospholipid bilayers. PNAS. 1997;94:14423–14428. doi: 10.1073/pnas.94.26.14423. [PubMed] [Cross Ref]
  • Chapochnikov NM, Takago H, Huang CH, Pangršič T, Khimich D, Neef J, Auge E, Göttfert F, Hell SW, Wichmann C, Wolf F, Moser T. Uniquantal release through a dynamic fusion pore is a candidate mechanism of hair cell exocytosis. Neuron. 2014;83:1389–1403. doi: 10.1016/j.neuron.2014.08.003. [PubMed] [Cross Ref]
  • Chernomordik LV, Kozlov MM. Mechanics of membrane fusion. Nature Structural and Molecular Biology. 2008;15:675–683. doi: 10.1038/nsmb.1455. [PMC free article] [PubMed] [Cross Ref]
  • Chizmadzhev YA, Cohen FS, Shcherbakov A, Zimmerberg J. Membrane mechanics can account for fusion pore dilation in stages. Biophysical Journal. 1995;69:2489–2500. doi: 10.1016/S0006-3495(95)80119-0. [PubMed] [Cross Ref]
  • Chizmadzhev YA, Kuzmin PI, Kumenko DA, Zimmerberg J, Cohen FS. Dynamics of fusion pores connecting membranes of different tensions. Biophysical Journal. 2000;78:2241–2256. doi: 10.1016/S0006-3495(00)76771-3. [PubMed] [Cross Ref]
  • Cohen FS, Melikyan GB. The energetics of membrane fusion from binding, through hemifusion, pore formation, and pore enlargement. Journal of Membrane Biology. 2004;199:1–14. doi: 10.1007/s00232-004-0669-8. [PubMed] [Cross Ref]
  • Curran MJ, Cohen FS, Chandler DE, Munson PJ, Zimmerberg J. Exocytotic fusion pores exhibit semi-stable states. The Journal of Membrane Biology. 1993;133:61–75. doi: 10.1007/BF00231878. [PubMed] [Cross Ref]
  • Dhara M, Yarzagaray A, Schwarz Y, Dutta S, Grabner C, Moghadam PK, Bost A, Schirra C, Rettig J, Reim K, Brose N, Mohrmann R, Bruns D. Complexin synchronizes primed vesicle exocytosis and regulates fusion pore dynamics. The Journal of Cell Biology. 2014;204:1123–1140. doi: 10.1083/jcb.201311085. [PMC free article] [PubMed] [Cross Ref]
  • Diao J, Grob P, Cipriano DJ, Kyoung M, Zhang Y, Shah S, Nguyen A, Padolina M, Srivastava A, Vrljic M, Shah A, Nogales E, Chu S, Brunger AT. Synaptic proteins promote calcium-triggered fast transition from point contact to full fusion. eLife. 2013;1:e00109 doi: 10.7554/elife.00109. [PMC free article] [PubMed] [Cross Ref]
  • Domanska MK, Kiessling V, Stein A, Fasshauer D, Tamm LK. Single vesicle millisecond fusion kinetics reveals number of SNARE complexes optimal for fast SNARE-mediated membrane fusion. Journal of Biological Chemistry. 2010;285:32158–32166. doi: 10.1074/jbc.A109.047381. [PMC free article] [PubMed] [Cross Ref]
  • Fang Q, Berberian K, Gong LW, Hafez I, Sørensen JB, Lindau M. The role of the C terminus of the SNARE protein SNAP-25 in fusion pore opening and a model for fusion pore mechanics. PNAS. 2008;105:15388–15392. doi: 10.1073/pnas.0805377105. [PubMed] [Cross Ref]
  • Fdez E, Martínez-Salvador M, Beard M, Woodman P, Hilfiker S. Transmembrane-domain determinants for SNARE-mediated membrane fusion. Journal of Cell Science. 2010;123:2473–2480. doi: 10.1242/jcs.061325. [PubMed] [Cross Ref]
  • Fulop T, Radabaugh S, Smith C. Activity-dependent differential transmitter release in mouse adrenal chromaffin cells. Journal of Neuroscience. 2005;25:7324–7332. doi: 10.1523/JNEUROSCI.2042-05.2005. [PubMed] [Cross Ref]
  • Gao Y, Zorman S, Gundersen G, Xi Z, Ma L, Sirinakis G, Rothman JE, Zhang Y. Single reconstituted neuronal SNARE complexes zipper in three distinct stages. Science. 2012;337:1340–1343. doi: 10.1126/science.1224492. [PMC free article] [PubMed] [Cross Ref]
  • Giraudo CG, Eng WS, Melia TJ, Rothman JE. A clamping mechanism involved in SNARE-dependent exocytosis. Science. 2006;313:676–680. doi: 10.1126/science.1129450. [PubMed] [Cross Ref]
  • Haluska CK, Riske KA, Marchi-Artzner V, Lehn JM, Lipowsky R, Dimova R. Time scales of membrane fusion revealed by direct imaging of vesicle fusion with high temporal resolution. PNAS. 2006;103:15841–15846. doi: 10.1073/pnas.0602766103. [PubMed] [Cross Ref]
  • Han X, Wang CT, Bai J, Chapman ER, Jackson MB. Transmembrane segments of syntaxin line the fusion pore of Ca2+-triggered exocytosis. Science. 2004;304:289–292. doi: 10.1126/science.1095801. [PubMed] [Cross Ref]
  • Hanna ST, Pigeau GM, Galvanovskis J, Clark A, Rorsman P, MacDonald PE. Kiss-and-run exocytosis and fusion pores of secretory vesicles in human beta-cells. Pflügers Archiv - European Journal of Physiology. 2009;457:1343–1350. doi: 10.1007/s00424-008-0588-0. [PubMed] [Cross Ref]
  • He L, Wu LG. The debate on the kiss-and-run fusion at synapses. Trends in Neurosciences. 2007;30:447–455. doi: 10.1016/j.tins.2007.06.012. [PubMed] [Cross Ref]
  • He L, Wu XS, Mohan R, Wu LG. Two modes of fusion pore opening revealed by cell-attached recordings at a synapse. Nature. 2006;444:102–105. doi: 10.1038/nature05250. [PubMed] [Cross Ref]
  • Hernandez JM, Kreutzberger AJ, Kiessling V, Tamm LK, Jahn R. Variable cooperativity in SNARE-mediated membrane fusion. PNAS. 2014;111:12037–12042. doi: 10.1073/pnas.1407435111. [PubMed] [Cross Ref]
  • Hille B. Ion Channels of Excitable Membranes. 3rd edn. Sinauer; 2001.
  • Hu C, Ahmed M, Melia TJ, Söllner TH, Mayer T, Rothman JE. Fusion of cells by flipped SNAREs. Science. 2003;300:1745–1749. doi: 10.1126/science.1084909. [PubMed] [Cross Ref]
  • Jackson MB. Minimum membrane bending energies of fusion pores. Journal of Membrane Biology. 2009;231:101–115. doi: 10.1007/s00232-009-9209-x. [PMC free article] [PubMed] [Cross Ref]
  • Jackson MB. SNARE complex zipping as a driving force in the dilation of proteinaceous fusion pores. Journal of Membrane Biology. 2010;235:89–100. doi: 10.1007/s00232-010-9258-1. [PMC free article] [PubMed] [Cross Ref]
  • Jahn R, Scheller RH. SNAREs--engines for membrane fusion. Nature Reviews Molecular Cell Biology. 2006;7:631–643. doi: 10.1038/nrm2002. [PubMed] [Cross Ref]
  • Jorgacevski J, Potokar M, Grilc S, Kreft M, Liu W, Barclay JW, Bückers J, Medda R, Hell SW, Parpura V, Burgoyne RD, Zorec R. Munc18-1 tuning of vesicle merger and fusion pore properties. Journal of Neuroscience. 2011;31:9055–9066. doi: 10.1523/JNEUROSCI.0185-11.2011. [PubMed] [Cross Ref]
  • Karatekin E, Di Giovanni J, Iborra C, Coleman J, O'Shaughnessy B, Seagar M, Rothman JE. A fast, single-vesicle fusion assay mimics physiological SNARE requirements. PNAS. 2010;107:3517–3521. doi: 10.1073/pnas.0914723107. [PubMed] [Cross Ref]
  • Kesavan J, Borisovska M, Bruns D. v-SNARE actions during ca(2+)-triggered exocytosis. Cell. 2007;131:351–363. doi: 10.1016/j.cell.2007.09.025. [PubMed] [Cross Ref]
  • Khelashvili G, Kollmitzer B, Heftberger P, Pabst G, Harries D. Calculating the bending Modulus for multicomponent lipid membranes in different thermodynamic phases. Journal of Chemical Theory and Computation. 2013;9:3866–3871. doi: 10.1021/ct400492e. [PMC free article] [PubMed] [Cross Ref]
  • Klyachko VA, Jackson MB. Capacitance steps and fusion pores of small and large-dense-core vesicles in nerve terminals. Nature. 2002;418:89–92. doi: 10.1038/nature00852. [PubMed] [Cross Ref]
  • Knecht V, Grubmüller H. Mechanical coupling via the membrane fusion SNARE protein syntaxin 1A: a molecular dynamics study. Biophysical Journal. 2003;84:1527–1547. doi: 10.1016/S0006-3495(03)74965-0. [PubMed] [Cross Ref]
  • Knowles MK, Barg S, Wan L, Midorikawa M, Chen X, Almers W. Single secretory granules of live cells recruit syntaxin-1 and synaptosomal associated protein 25 (SNAP-25) in large copy numbers. PNAS. 2010;107:20810–20815. doi: 10.1073/pnas.1014840107. [PubMed] [Cross Ref]
  • Kozlov MM, Leikin SL, Chernomordik LV, Markin VS, Chizmadzhev YA. Stalk mechanism of vesicle fusion. intermixing of aqueous contents. European Biophysics Journal : EBJ. 1989;17:121–129. doi: 10.1007/BF00254765. [PubMed] [Cross Ref]
  • Krishnakumar SS, Kümmel D, Jones SJ, Radoff DT, Reinisch KM, Rothman JE. Conformational dynamics of calcium-triggered activation of fusion by synaptotagmin. Biophysical Journal. 2013;105:2507–2516. doi: 10.1016/j.bpj.2013.10.029. [PubMed] [Cross Ref]
  • Krishnakumar SS, Radoff DT, Kümmel D, Giraudo CG, Li F, Khandan L, Baguley SW, Coleman J, Reinisch KM, Pincet F, Rothman JE. A conformational switch in complexin is required for synaptotagmin to trigger synaptic fusion. Nature Structural and Molecular Biology. 2011;18:934–940. doi: 10.1038/nsmb.2103. [PMC free article] [PubMed] [Cross Ref]
  • Lai Y, Diao J, Liu Y, Ishitsuka Y, Su Z, Schulten K, Ha T, Shin YK. Fusion pore formation and expansion induced by Ca2+ and synaptotagmin 1. PNAS. 2013;110:1333–1338. doi: 10.1073/pnas.1218818110. [PubMed] [Cross Ref]
  • Lindau M, Alvarez de Toledo G, de Toledo GA. The fusion pore. Biochimica Et Biophysica Acta (BBA) - Molecular Cell Research. 2003;1641:167–173. doi: 10.1016/S0167-4889(03)00085-5. [PubMed] [Cross Ref]
  • Lindau M. High resolution electrophysiological techniques for the study of calcium-activated exocytosis. Biochimica Et Biophysica Acta (BBA) - General Subjects. 2012;1820:1234–1242. doi: 10.1016/j.bbagen.2011.12.011. [PMC free article] [PubMed] [Cross Ref]
  • Liu X, Seven AB, Camacho M, Esser V, Xu J, Trimbuch T, Quade B, Su L, Ma C, Rosenmund C, Rizo J. Functional synergy between the Munc13 C-terminal C1 and C2 domains. eLife. 2016;5:e13696 doi: 10.7554/eLife.13696. [PMC free article] [PubMed] [Cross Ref]
  • MacDonald PE, Braun M, Galvanovskis J, Rorsman P. Release of small transmitters through kiss-and-run fusion pores in rat pancreatic beta cells. Cell Metabolism. 2006;4:283–290. doi: 10.1016/j.cmet.2006.08.011. [PubMed] [Cross Ref]
  • Marsh D. Elastic curvature constants of lipid monolayers and bilayers. Chemistry and Physics of Lipids. 2006;144:146–159. doi: 10.1016/j.chemphyslip.2006.08.004. [PubMed] [Cross Ref]
  • McNew JA, Weber T, Parlati F, Johnston RJ, Melia TJ, Söllner TH, Rothman JE. Close is not enough: snare-dependent membrane fusion requires an active mechanism that transduces force to membrane anchors. The Journal of Cell Biology. 2000;150:105–117. doi: 10.1083/jcb.150.1.105. [PMC free article] [PubMed] [Cross Ref]
  • Melikov KC, Frolov VA, Shcherbakov A, Samsonov AV, Chizmadzhev YA, Chernomordik LV. Voltage-induced nonconductive pre-pores and metastable single pores in unmodified planar lipid bilayer. Biophysical Journal. 2001;80:1829–1836. doi: 10.1016/S0006-3495(01)76153-X. [PubMed] [Cross Ref]
  • Mellander LJ, Kurczy ME, Najafinobar N, Dunevall J, Ewing AG, Cans AS. Two modes of exocytosis in an artificial cell. Scientific Reports. 2014;4:3847 doi: 10.1038/srep03847. [PMC free article] [PubMed] [Cross Ref]
  • Mellander LJ, Trouillon R, Svensson MI, Ewing AG. Amperometric post spike feet reveal most exocytosis is via extended kiss-and-run fusion. Scientific Reports. 2012;2:907 doi: 10.1038/srep00907. [PMC free article] [PubMed] [Cross Ref]
  • Mitra K, Ubarretxena-Belandia I, Taguchi T, Warren G, Engelman DM. Modulation of the bilayer thickness of exocytic pathway membranes by membrane proteins rather than cholesterol. PNAS. 2004;101:4083–4088. doi: 10.1073/pnas.0307332101. [PubMed] [Cross Ref]
  • Mohrmann R, de Wit H, Verhage M, Neher E, Sørensen JB. Fast vesicle fusion in living cells requires at least three SNARE complexes. Science. 2010;330:502–505. doi: 10.1126/science.1193134. [PubMed] [Cross Ref]
  • Montecucco C, Schiavo G, Pantano S. SNARE complexes and neuroexocytosis: how many, how close? Trends in Biochemical Sciences. 2005;30:367–372. doi: 10.1016/j.tibs.2005.05.002. [PubMed] [Cross Ref]
  • Morrow JA, Arnold KS, Weisgraber KH. Functional characterization of apolipoprotein E isoforms overexpressed in Escherichia coli. Protein Expression and Purification. 1999;16:224–230. doi: 10.1006/prep.1999.1069. [PubMed] [Cross Ref]
  • Nanavati C, Markin VS, Oberhauser AF, Fernandez JM. The exocytotic fusion pore modeled as a lipidic pore. Biophysical Journal. 1992;63:1118–1132. doi: 10.1016/S0006-3495(92)81679-X. [PubMed] [Cross Ref]
  • Parlati F, Weber T, McNew JA, Westermann B, Söllner TH, Rothman JE. Rapid and efficient fusion of phospholipid vesicles by the alpha-helical core of a SNARE complex in the absence of an N-terminal regulatory domain. PNAS. 1999;96:12565–12570. doi: 10.1073/pnas.96.22.12565. [PubMed] [Cross Ref]
  • Pawlu C, DiAntonio A, Heckmann M. Postfusional control of quantal current shape. Neuron. 2004;42:607–618. doi: 10.1016/S0896-6273(04)00269-7. [PubMed] [Cross Ref]
  • Rand RP, Parsegian VA. Hydration forces between phospholipid bilayers. Biochimica Et Biophysica Acta (BBA) - Reviews on Biomembranes. 1989;988:351–376. doi: 10.1016/0304-4157(89)90010-5. [Cross Ref]
  • Ritchie TK, Grinkova YV, Bayburt TH, Denisov IG, Zolnerciks JK, Atkins WM, Sligar SG. Chapter 11 - Reconstitution of membrane proteins in phospholipid bilayer nanodiscs. Methods in Enzymology. 2009;464:211–231. doi: 10.1016/S0076-6879(09)64011-8. [PMC free article] [PubMed] [Cross Ref]
  • Rizo J, Xu J. The Synaptic Vesicle Release Machinery. Annual Review of Biophysics. 2015;44:339–367. doi: 10.1146/annurev-biophys-060414-034057. [PubMed] [Cross Ref]
  • Sakmann B. Neher E. Single-Channel Recording. 2nd edn. Springer; 2009.
  • Shi L, Howan K, Shen QT, Wang YJ, Rothman JE, Pincet F. Preparation and characterization of SNARE-containing nanodiscs and direct study of cargo release through fusion pores. Nature Protocols. 2013;8:935–948. doi: 10.1038/nprot.2013.048. [PubMed] [Cross Ref]
  • Shi L, Shen QT, Kiel A, Wang J, Wang HW, Melia TJ, Rothman JE, Pincet F. SNARE proteins: one to fuse and three to keep the nascent fusion pore open. Science. 2012;335:1355–1359. doi: 10.1126/science.1214984. [PMC free article] [PubMed] [Cross Ref]
  • Sinha R, Ahmed S, Jahn R, Klingauf J. Two synaptobrevin molecules are sufficient for vesicle fusion in central nervous system synapses. PNAS. 2011;108:14318–14323. doi: 10.1073/pnas.1101818108. [PubMed] [Cross Ref]
  • Staal RG, Mosharov EV, Sulzer D. Dopamine neurons release transmitter via a flickering fusion pore. Nature Neuroscience. 2004;7:341–346. doi: 10.1038/nn1205. [PubMed] [Cross Ref]
  • Stein A, Weber G, Wahl MC, Jahn R. Helical extension of the neuronal SNARE complex into the membrane. Nature. 2009;460:525–U105. doi: 10.1038/nature08156. [PMC free article] [PubMed] [Cross Ref]
  • Stratton BS, Warner JM, Wu Z, Nikolaus J, Wei G, Wagnon E, Baddeley D, Karatekin E, O'Shaughnessy B. Cholesterol increases the openness of SNARE-Mediated flickering fusion pores. Biophysical Journal. 2016;110:1538–1550. doi: 10.1016/j.bpj.2016.02.019. [PubMed] [Cross Ref]
  • Südhof TC, Rothman JE. Membrane fusion: grappling with SNARE and SM proteins. Science. 2009;323:474–477. doi: 10.1126/science.1161748. [PMC free article] [PubMed] [Cross Ref]
  • Takamori S, Holt M, Stenius K, Lemke EA, Grønborg M, Riedel D, Urlaub H, Schenck S, Brügger B, Ringler P, Müller SA, Rammner B, Gräter F, Hub JS, De Groot BL, Mieskes G, Moriyama Y, Klingauf J, Grubmüller H, Heuser J, Wieland F, Jahn R. Molecular anatomy of a trafficking organelle. Cell. 2006;127:831–846. doi: 10.1016/j.cell.2006.10.030. [PubMed] [Cross Ref]
  • van den Bogaart G, Holt MG, Bunt G, Riedel D, Wouters FS, Jahn R. One SNARE complex is sufficient for membrane fusion. Nature Structural and Molecular Biology. 2010;17:358–364. doi: 10.1038/nsmb.1748. [PMC free article] [PubMed] [Cross Ref]
  • Wang CT, Grishanin R, Earles CA, Chang PY, Martin TF, Chapman ER, Jackson MB. Synaptotagmin modulation of fusion pore kinetics in regulated exocytosis of dense-core vesicles. Science. 2001;294:1111–1115. doi: 10.1126/science.1064002. [PubMed] [Cross Ref]
  • Wang CT, Lu JC, Bai J, Chang PY, Martin TF, Chapman ER, Jackson MB. Different domains of synaptotagmin control the choice between kiss-and-run and full fusion. Nature. 2003a;424:943–947. doi: 10.1038/nature01857. [PubMed] [Cross Ref]
  • Wang CT, Lu JC, Chapman ER, Martin TFJ, Jackson MB. Synaptotagmin IV induces long-duration kiss-and-run exocytosis through small fusion pores. Biophysical Journal. 2003b;84:209a–225a.
  • Weber T, Zemelman BV, McNew JA, Westermann B, Gmachl M, Parlati F, Söllner TH, Rothman JE. SNAREpins: minimal machinery for membrane fusion. Cell. 1998;92:759–772. doi: 10.1016/S0092-8674(00)81404-X. [PubMed] [Cross Ref]
  • Wu Z, Auclair SM, Bello O, Vennekate W, Dudzinski NR, Krishnakumar SS, Karatekin E. Nanodisc-cell fusion: control of fusion pore nucleation and lifetimes by SNARE protein transmembrane domains. Scientific Reports. 2016;6:27287 doi: 10.1038/srep27287. [PMC free article] [PubMed] [Cross Ref]
  • Yang Y, Sigworth FJ. Single-channel properties of IKs potassium channels. The Journal of General Physiology. 1998;112:665–678. doi: 10.1085/jgp.112.6.665. [PMC free article] [PubMed] [Cross Ref]
  • Zhao Y, Fang Q, Herbst AD, Berberian KN, Almers W, Lindau M. Rapid structural change in synaptosomal-associated protein 25 (SNAP25) precedes the fusion of single vesicles with the plasma membrane in live chromaffin cells. PNAS. 2013;110:14249–14254. doi: 10.1073/pnas.1306699110. [PubMed] [Cross Ref]
2017; 6: e22964.

Decision letter

Axel T Brunger, Reviewing editor
Axel T Brunger, Stanford University Medical Center, United States;

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Dilation of fusion pores by crowding of SNARE proteins" for consideration by eLife. Your article has been favorably evaluated by Randy Schekman (Senior Editor) and three reviewers, one of whom is a member of our Board of Reviewing Editors. The following individuals involved in review of your submission have agreed to reveal their identity: Josep Rizo (Reviewer #2); Jiajie Diao (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor concluded that major revisions are required before a final decision can be made.

Summary:

This study shows how the number of SNARE complexes affects the formation of fusion pores between discoidal lipid nanoparticles containing VAMP2 (vNPLs) and cells expressing flipped t-SNAREs at the plasma membrane. The approach used in this work is an extension of previous work (Wu et al., 2016) but now with larger (21-27 nm diameter) nanolipoprotein particles (nlp) instead of the smaller nanodiscs used in the previous work (6-18 nm diameter). The larger nlp discs can accommodate more v-SNARE proteins. The electrophysiological approach used by the authors offers better time resolution than optical microscopy approaches used to date to study fusion in reconstituted systems. Their data show how pores form and flicker back and forth for long periods of time. Even relatively large numbers of SNAREs yield pores of limited conductance. However, there are a number of potential concerns that limit the insights into SNARE mediated fusion. Overall, this appears to be more of a methods paper and the conclusions regarding SNARE mediated fusion should be toned down.

Essential revisions:

1) From the example shown in Figure 5D, it appears all fusion pores eventually reseal. Why is that? Could this be a consequence of cellular resealing mechanisms or constraints imposed by the patch clamp? Please test this possibility by using different size clamps.

2) These experiments use flipped t-SNAREs which exhibit considerably slower kinetics than wild type SNAREs (Giraudo et al., 2006): fusion in the presence of synaptotagmin and Ca2+ occurs in several minutes, vs. msec to sec. in reconstituted systems with wildtype SNAREs and synaptotagmin. Thus, use of the flipped t-SNAREs may profoundly alter the kinetics of the fusion pore opening and dilation as well as the number of SNARE complexes required to promote fusion. Moreover, the lipid composition of the outer leaflet of the cell membrane may be quite different from that of the inner leaflet of the plasma membrane (there is little PS in the outer leaflet of the cell membrane). Ideally, the authors should consider experiments to alleviate these concerns, but the minimum, these limitations need to be discussed in detail.

3) Another limitation of this study is that it is focuses on SNAREs only. Prior experiments have been done with a combination of large and small fluorescent probes to measure fusion pore dilation, e.g.: (Lai et al., 2013), and concluded that factors such as synaptotagmin are required for efficient pore expansion. Moreover, with the help of these other proteins, the number of SNARE complexes needed for fusion could be much smaller. The authors do mention accessory proteins at the end of the manuscript, but only in the context of organizing SNARE complexes, as indicated by the term 'accessory' itself, and without considering a direct role in membrane fusion. At the minimum, the authors should tone down every conclusion regarding the number of SNAREs required for neurotransmitter release.

4) Please indicate the percentage of vNLPs that fuse with the cells for the different vNLPs. This is a critical issue when rationalizing the data in terms of how the number of SNAREs influence fusion pore properties because the vNPLs are expected to have a distribution of VAMP copies rather than a single number. If the percentage of vNLPs that is low, the fusion may arise from the population that has much higher VAMP copies than assumed from the average. Describing the percentage of vNPLs that fuse is also important to evaluate this overall approach.

5) The number of fusion pores/min observed with empty NLPs (is what eNLP means?) is low compared to that observed with vNLP8, but not negligible; it appears to be about 8 times smaller in Figure 4C. Does this mean that eNLPs fuse spontaneously with the cells? If this is the case, an acceleration by a factor of 8 suggests that 8 SNARE complexes only provide about 1 kcal/mol to facilitate membrane fusion. If there is no flaw in this argument, the authors should emphasize this point and compare this energetic estimate will all the energetic arguments they make later in the manuscript.

6) Conclusions regarding fusion pore expansion should be taken with caution because the scaffolding protein may impose constraints that would not be present in vivo.

7) A concern is about the possibility of formation of multiple pores. In the third paragraph of the Results, the authors provided an explanation for single pore formation. However, they apparently excluded the possibility of pore formation involving multiple v-SNARE nanolipoprotein particles at the same time. Due to the large size of nanolipoprotein particles (23 nm in diameter), it is possible that there is more than one fusion pore for individual nanolipoprotein particle in the presence of many SNAREs. For example, compared to the large jump from 15 to 30 v-SNAREs, the difference between 8 and 15 v-SNAREs is negligible (Figure 5B). Could this imply the formation of another fusion pore?

8) In the section 'Derivation of best-fit model parameters…', the authors showed how they obtain the parameters in the model by fitting the calculated results of the model (equation 11) to the experimental data. For example, they fit the results in the range of 0.2 nm ≤ rpo ≤ 1.5 nm to obtain the tension of membrane γ and the steric-hydration force length scale λ (0.5 nm ≤ rpo ≤ 2.5nm) to determine a range for epsilon (4 nm ≤ rpo ≤ 4.5nm) to obtain tau. However, in equation 11, the total free energy depends on all the parameters in the full range of rpo, so how can the authors obtain an individual parameter in a different range of rpo?

9) The parameters in the mathematical model (i.e., rpo, h, δ, D, and the twisting angle phi of ApoE proteins) should be explicitly shown in the model in Figure 7.

10) To calculate the second term in equation 7, the authors approximated the pore as a cylinder of radius rpo and height λ = 0.13 nm. It seems that the height of the cylinder is too small, as the neck of the fusion pore can be on the order of ~5 nm in height. Please explain why choosing such a small height for the cylinder is a reasonable approximation.

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

Thank you for resubmitting your work entitled "Dilation of fusion pores by crowding of SNARE proteins" for further consideration at eLife. Your revised article has been favorably evaluated by Randy Schekman (Senior editor) and three reviewers, one of whom is a member of our Board of Reviewing Editors.

The manuscript has been improved but there are two remaining issues that need to be addressed before acceptance, as outlined below, using the same bullet numbers used for the points of the decision letter.

2) Some of the arguments offered with regard to lipid composition are reasonable, but saying that 'short PI(4,5)P2 does not have an effect (data not shown)' is not convincing, and the problem still remains that it is difficult to control lipid composition in this system and make it similar to physiological. This is particularly important considering the low percentage of fusion that they now report (point 4). The authors cite Giraudo et al., 2006, but the work in that paper suffers from the same problem. This concern does not invalidate the results from the authors, but they should acknowledge the problem in the manuscript. The fact is that lipids could play key roles in the fusion mechanism and the tendency of SNARE-centric models of membrane fusion to ignore this fact is deleterious to the scientific discussion in the field.

4) The authors again make some good arguments, but they really do not address the heart of the problem. With such low percentage of fusion, it seems very likely that fusion occurs for low populations of nanodiscs with higher copy number of SNAREs than the mean. The authors claim that this is unlikely because 'the fusion rate does not increase for mean copy number greater than 2-4 per NLP side (Figure 5A)'. However, in Figure 5A one can see a tendency for the fusion rate to increase up to vNLP15. Even though the differences may not be statistically significant, one just cannot draw the conclusion written by the authors. Hence, the authors should explicitly acknowledge that the actual numbers of SNAREs underlying fusion may be (in fact that are very likely to be) higher than the mean values described. The same issue applies to measurements of conductance.

Essential revisions:

1) From the example shown in Figure 5D, it appears all fusion pores eventually reseal. Why is that? Could this be a consequence of cellular resealing mechanisms or constraints imposed by the patch clamp? Please test this possibility by using different size clamps.

The simplest explanation for pore closure is that the pore does not have an alternative. It cannot expand beyond a maximum size due to constraints imposed by the scaffold protein. It also does not remain open indefinitely, because presumably the pore is a higher energy structure than a resealed membrane (either a hemi-fission structure or two separate bilayers). This is consistent with the fact that it is difficult to capture fusion pores in various experiments (e.g. fusion pores were not detected in EM studies by Hernandez..Jahn Science, 2012, 336:1552, Diao…Brunger eLife, 2012, 1:e00109, or Shi…Pincet, Science 2012, 335:1355). So we expect the pore to reseal at some point given enough time, independent of any cellular response of contraints imposed by the cell-attached patch clamp method.

Indeed, previous work that studied fusion between v-SNARE nanodiscs to t-SNARE liposomes also found pores eventually reseal. Both the Rothman (Shi et al., Science, 2012; Bello et al.,) and the Chapman labs (Bao et al., 2016) showed that addition of dithionite after the fusion reaction quenched only a fraction of the NBD-PE lipid labels. Had the pores remained open, the quencher (0.2-0.3 nm Stokes radius) would have had access to the liposome lumen and quenched all the signal. In addition, both laboratories found at least some of the nanodiscs attached to the liposomes, perhaps in a hemi-fission state.

To exclude artifacts due to the cell-attached recording configuration, in previous work (Sci. Rep. 6:27287, 2016) we patched flipped t-SNARE cells in the whole cell configuration, puffed nanodiscs onto the cell using a pressure-driven perfusion system, and monitored whole-cell currents. We observed large currents upon perfusion of discs, consistent with disc-cell fusion.

The currents returned to baseline upon cessation of disc application, indicating pores resealed.

To test if pores also reseal in our calcium influx assay, we have performed new experiments in which we loaded cells with a calcium indicator, Fluo-4, then added nanodiscs. The Fluo-4 fluorescence increased, consistent with fusion allowing calcium entry into the cells, as in Figure 3. We then washed away unbound discs, and continued to monitor Fluo-4 signals which returned to the baseline after a few minutes. These results are consistent with our whole-cell recordings mentioned earlier and suggest that fusion pores connecting nanodiscs to cells do close, even in the absence of any patch-clamp recordings. We have added these new results as Figure 3—figure supplement 1.

Although it is clear that the pore must reseal, it is still possible that the pore lifetime is affected by the patch method or cellular processes. In our cell-attached recordings of single-pore currents, we do not expect cellular wound healing mechanisms to contribute significantly. Such repair mechanisms have been studied in response to much larger pores (~ 1 micron or larger). In addition, a large extracellular calcium concentration (~1 mM or larger) is an essential requirement for cell wound healing (Moe, Golding, and Bement, Sem. Cell. Dev. Biol., 2015, 45:1823), but our pipette solutions did not contain any calcium.

Finally, an indication that cellular processes do not have an appreciable effect on pore lifetimes is that the largest effect on pore lifetimes so far was obtained by three point mutations in the v- SNARE transmembrane domain (Wu et al., 2016), prolonging pore lifetimes 10-fold. It is hard to imagine that point mutations on the v-SNARE TMD would affect any cellular processes.

2) These experiments use flipped t-SNAREs which exhibit considerably slower kinetics than wild type SNAREs (Giraudo et al., 2006): fusion in the presence of synaptotagmin and Ca2+ occurs in several minutes, vs. msec to sec. in reconstituted systems with wildtype SNAREs and synaptotagmin. Thus, use of the flipped t-SNAREs may profoundly alter the kinetics of the fusion pore opening and dilation as well as the number of SNARE complexes required to promote fusion. Moreover, the lipid composition of the outer leaflet of the cell membrane may be quite different from that of the inner leaflet of the plasma membrane (there is little PS in the outer leaflet of the cell membrane). Ideally, the authors should consider experiments to alleviate these concerns, but the minimum, these limitations need to be discussed in detail.

The slow fusion kinetics in the cell-cell fusion experiments in Giraudo et al., which included flipped synaptotagmin and GPI-anchored complexin in addition to flipped SNAREs, was almost certainly limited by the detection method used. First, the classification of cells as fused required complete mixing of cytoplasms and the uniform appearance of dsRed (cytosolic), CFP (nuclear) and YFP (nuclear) signals, which typically requires minutes after fusion. Second, rather than performing live-cell imaging, the authors fixed the cells after a given incubation period, mounted them, and collected confocal images at a later time. This allowed several experiments and controls to be run in parallel in multiwell plates, at the expense of high-resolution kinetic measurements. The assay was designed to acquire images of large numbers of cells, using a small magnification objective (20x) for most experiments. This approach precluded analysis of the true kinetics of the fusion reaction, which was presumably much faster than the minutes time scale that could be probed in the assay.

In sharp contrast to cell-cell fusion experiments, much faster time scales can be probed in liposome-liposome fusion experiments, because the volumes and areas involved for content and lipid mixing are much smaller in the liposomes than in cells and diffusion is not hindered by cytosolic or membrane proteins.

It is true that the lipid composition of our target membrane (outer leaflet of the plasma membrane) is very different from that of the physiological target membrane (the inner leaflet of the plasma membrane). This is a potential issue we tested as follows. First, we flipped the configuration and fused flipped v-SNARE cells with t-SNARE NLPs. This allowed us to have a better mimic of the inner plasma membrane leaflet composition on the target membrane (now the tNLP membrane). This resulted in similar fusion rates and pore properties for two different SNARE copy numbers per NLP. This new data is now added as Figure 5—figure supplement 3 and mentioned in the main text.

Second, we added short-chain PI(4,5)P2 to the outer leaflet of the flipped v-SNARE cells. This manipulation did not affect the fusion rate or pore properties (data not shown), consistent with the lack of effects in Giraudo et al., 2006) in the absence of synaptotagmin. Thus, SNARE-only fusion can tolerate a range of target membrane compositions.

3) Another limitation of this study is that it is focuses on SNAREs only. Prior experiments have been done with a combination of large and small fluorescent probes to measure fusion pore dilation, e.g.: (Lai et al., 2013.), and concluded that factors such as synaptotagmin are required for efficient pore expansion. Moreover, with the help of these other proteins, the number of SNARE complexes needed for fusion could be much smaller. The authors do mention accessory proteins at the end of the manuscript, but only in the context of organizing SNARE complexes, as indicated by the term 'accessory' itself, and without considering a direct role in membrane fusion. At the minimum, the authors should tone down every conclusion regarding the number of SNAREs required for neurotransmitter release.

The question of copy number requirements for SNAREs during fusion has been an ongoing debate for some time, with essentially all in vitro studies having been based on experiments with SNAREs alone. in vivo studies suggested 2-16 SNARE complexes may be required for exocytosis, whereas SNARE-alone in vitro experiments suggested 1 to more than 20 SNAREs may be needed. The fact that similar ranges have been found for in vivo and in vitro studies suggests that the SNARE-alone in vitro studies may indeed have some relevance to exocytosis, but the broad range of values reported precludes any firm conclusions. Only one study offered a possible explanation for the broad range of SNARE requirements; smaller liposomes required fewer SNAREs to fuse than larger ones, suggesting curvature may be an important parameter (Hernandez…Jahn, PNAS 2014, 111:12037-12042). In this respect, our studies of SNARE-alone fusion are highly relevant, as they suggest i) pore expansion requires many more SNAREs than does pore opening, a point that was not suspected previously, ii) when copy number requirements are studied using small soluble probes (or lipid labels) that can permeate small pores, lower number requirements are are likely to be found, iii) curvature effects alone cannot explain copy number requirements (as both the disc and plasma membranes are rather flat in our studies).

The term “accessory proteins” is not intended in any way to diminsh the importance of the other proteins involved in exocytosis. Some of these, such as Munc18, are as essential as SNAREs for exocytosis. However, the SNAREs hold the core position in the fusion machinery because of their ability to induce fusion (albeit inefficient and slow) when reconstituted alone. The other constituents of the fusion machinery, regardless of whether they are required in vivo, cannot hold the same claim.

We changed the word “accessory” to “other”, and have modified the Introduction (see below) and Discussion extensively to clarify that mutations in Munc18, Synaptotagmin, and Complexin affect fusion pore properties. Therefore, all these, and likely other components of the exocytotic machinery are important regulators of fusion pores. Because we have not yet tested the roles of these components, one should be cautious about extrapolating our results to physiological neurotransmitter or hormone release. Nevertheless, the concept of protein crowding driving pore dilation is such a general principle that is likely to apply to the physiological situation qualitatively. Indeed, there are some suggestions in the literature that SNARE availability affects release kinetics (Acuna…Sudhof, Neuron, 2014, and Zhao[…]Lindau, PNAS 2013).

We now cite Lai et al. along with other relevant work in the Discussion. Lai et al. found pore dilation to be surprisingly slow in SNARE-only fusion in a tethered single-liposome fusion assay. Inclusion of Synaptotagmin, Ca2+ and complexin increased the speed (to ~10 s) and efficiency of content mixing significantly (to ~5% of docked liposomes). Increasing SNARE copy numbers increased content mixing with the small probe sulforhodamine B, but the effect on the permeability to the large DNA cargo was not investigated. Our results are consistent with the Lai et al. study. We also expect Synaptotagmin and other components of the fusion machinery to modify pore properties in our assay, as their mutations do during exocytosis. Just like the reviewers, we expect Synaptotagmin may reduce the SNARE copy requirement for pore dilation to smaller numbers. We also expect the more copies of Syt-SNARE complexes are involved, the easier will pore dilation be. Incorporating these other important exocytotic proteins into the assay will be an exciting and important future direction.

4) Please indicate the percentage of vNLPs that fuse with the cells for the different vNLPs. This is a critical issue when rationalizing the data in terms of how the number of SNAREs influence fusion pore properties because the vNPLs are expected to have a distribution of VAMP copies rather than a single number. If the percentage of vNLPs that is low, the fusion may arise from the population that has much higher VAMP copies than assumed from the average. Describing the percentage of vNPLs that fuse is also important to evaluate this overall approach.

In response to this excellent point made by the reviewer, we performed a new set of experiments, which showed that the fraction of docked NLPs that fuse is of order 4-5% over 20 min. These new results are presented in Figure 2—figure supplement 1. These new experiments are similar to our single-cell fluorescence measurements that probed lipid mixing (Figure 2), in which we include a pair of fluorophores, DiI and DiD, in the NLPs. We incubated these fluorophores with flipped t-SNARE cells and monitored DiI (donor) fluorescence over time, as before. In the new experiments, however, at the end of the experiment, we bleached the acceptor DiD completely to obtain the maximum possible donor (DiI) intensity, Fmax We then rescaled the donor fluorescence values F˜(t)=(FFo)/(FmaxFo), where Fo is the minimum at the beginning of acquision. This analysis revealed that 4-5% of maximal lipid mixing was reached at the end of 20 minutes of acquisition. (This is likely an underestimate of the true fraction that fuses, as some fusion inevitably occurs during the few minutes between the time the 4°C fusion block is removed and acquisition starts on the 37°C stage-top incubator of the microscope). By comparison, fusion between v-SNARE NLPs and t-SNARE liposomes yields a similar extent of lipid mixing over ~20 min (Bello et al., 2016).

The fact that the fusion rate does not increase for mean copy number greater than 2-4 per NLP side (Figure 5A) strongly suggests that there is no bias for higher copy numbers (assuming docking is comparable for all samples). Were there such a bias, i.e. higher copy numbers fused more readily, we would then expect the fusion rate and hence the fused faction to increase with mean copy number, since higher mean copy number means that more NLPs are present with large copy numbers.

Even if there were such a bias for high copy numbers, our results would still correctly report the general trend of pore properties versus copy number. This can be seen from the following argument. A reasonable guess for the distribution of VAMP copy numbers per NLP, n, is a Poisson distribution, p(n)=en¯(n¯)n/n!, provided the mean copy number n¯ is well below 30, the maximum attainable value in our experiments (presumably a packing constraint). In our experiments, we fix n¯ to have various values. For example, consider experiments with a mean of 4 copies per NLP, n¯=4; then even if the 5% of NLPs that fuse correspond to only the largest copy numbers, this would mean that most fused NLPs would have n = 8, 9 or 10 SNAREs (because n=830p(n) ≈ 5%, and the biggest contributions are from 8 ≤ n ≤ 10). Similarly, for n¯ = 8 most fused NLPs would have 13 ≤ n ≤ 15. Thus, even if there is the bias that the reviewer fears, the typical copy number of the NLPs whose pore properties are measured is an increasing function of the mean value n¯. It’s just that n¯ underestimates somewhat the actual copy numbers involved.

Let us also consider the extreme limit where fusion might arise from NLPs bearing the maximum possible copy number of v-SNAREs, ~30, due to packing constraints. If this were the case, then as the mean copy number increased, the fusion rate should also increase (inconsistent with observations of Figure 5A), but every pore would have similar properties (since every pore would be induced by ~30 SNARE copies). This is also inconsistent with data (Figure 5B and its supplement 1).

5) The number of fusion pores/min observed with empty NLPs (is what eNLP means?) is low compared to that observed with vNLP8, but not negligible; it appears to be about 8 times smaller in Figure 4C. Does this mean that eNLPs fuse spontaneously with the cells? If this is the case, an acceleration by a factor of 8 suggests that 8 SNARE complexes only provide about 1 kcal/mol to facilitate membrane fusion. If there is no flaw in this argument, the authors should emphasize this point and compare this energetic estimate will all the energetic arguments they make later in the manuscript.

Yes, eNLP means empty NLP, as defined in Results, paragraph two.

The level of activity that we see with eNLPs is about the same as the “baseline” or “background” level of activity that we measure when NLPs are omitted altogether. (This is true of all the negative controls we test, eNLP, TeNT, etc., see Wu et al., 2016). Now the pores in the absence of NLPs cannot be fusion pores, which form from two apposed bilayers, but are instead likely to be simple bilayer pores in the plasma membrane, leaks around the patch, or any other artifact that results in a small transient current. Thus, it is likely that eNLPs do not produce fusion pores.

For this reason, we believe that the reviewer’s argument leading to ~ 1 kcal/mol is actually not applicable: whatever their origin, because these background currents are not fusion pores, SNARE-reconstituted NLPs do not just accelerate the rate of occurrence of fusion pores. SNAREs induce fusion, a qualitatively different event.

Regarding the energetic arguments made later in the manuscript, we wish to emphasize that these refer to an already nucleated fusion pore, and describe the relative energies of pores of different sizes. These relative energies are quite different to the energy governing the rate of fusion, namely the energy to nucleate a fusion pore in the first place.

6) Conclusions regarding fusion pore expansion should be taken with caution because the scaffolding protein may impose constraints that would not be present in vivo.

We agree, and we now allude to a potential effect of this constraint on pore lifetimes in the modified Discussion. However, we note that we chose to use large NLPs rather than smaller, MSP-based discs so that the scaffold would not hinder pore dilation up to large pore sizes (at least up to ~5 nm radius). Indeed, most pores do not reach large enough sizes to experience constraints imposed by the scaffold proteins.

7) A concern is about the possibility of formation of multiple pores. In the third paragraph of the Results, the authors provided an explanation for single pore formation. However, they apparently excluded the possibility of pore formation involving multiple v-SNARE nanolipoprotein particles at the same time. Due to the large size of nanolipoprotein particles (23 nm in diameter), it is possible that there is more than one fusion pore for individual nanolipoprotein particle in the presence of many SNAREs. For example, compared to the large jump from 15 to 30 v-SNAREs, the difference between 8 and 15 v-SNAREs is negligible (Figure 5B). Could this imply the formation of another fusion pore?

We believe that the occurrence of multiple pores per NLP is unlikely for the following reasons.

i) In previous work we studied pores in MSP based nanodiscs, which can accommodate only a single pore due to their limited size. We showed that pores connecting MSP discs (~15 nm diameter, bearing 8-9 total copies of v-SNAREs) and flipped t-SNARE cells are not permeable to a large ion, NMDG+ (Wu et al., 2016), likely because the maximal size of such pores is constrained by the scaffold protein (see Figure 7A, energy profile labeled “vMSP”).

If a larger NLP bearing the maximally allowed SNARE copies had multiple small pores each similar to the pore seen in MSP discs, these pores would similarly be impermeable to NMDG+. If instead there were a single large pore per NLP, then NMDG+ could permeate the pore. To address this question, we performed new experiments in which we measured the permeability of vNLP30 pores to NMDG+. In contrast to MSP nanodiscs the large ion was permeant through these vNLP30 pores. These new results are presented in Figure 5—figure supplement 4.

These results are consistent with those of Bello et al., 2016 who showed that progressively larger cargo could be released from t-SNARE liposomes during fusion with vNLPs as the v-SNARE copies per NLP was increased.

ii) Conductance of n small pores in a single NLP would be roughly additive, giving total conductance equal to gpo=n x gpo, where gpo is the mean open-pore conductance of a small pore. Doubling the SNARE copies would presumably at most double n, and by consequence, total conductance. The fact that we find faster than linear increase in mean pore conductance as a function of copy number (Figure 5B) is consistent with each NLP bearing a single pore whose size increases with increasing SNARE copies (recall that conductance is proportional to the cross- sectional area of the pore, [proportional, variant] rpore2 i.e. very sensitive to changes in pore size).

iii) If multiple small pores occurred per NLP, this should be evident in the distribution of point- by-point conductance values, with peaks at n × gpo, where n = 1, 2, 3 …. Instead, for the distribution of mean gpo for vNLP30 we find a peak at ~300 pS, and a broad peak at ~3-14 nS (Figure 6B). If the typical small pore has 300 pS conductance, then to have ~6 nS (typical large conductance), there would have to be ~20 small pores per NLP. It is hard to imagine this many pores coexisting in this small area.

iv) If the observed conductance increase with increasing SNARE copy number were due to an increasing number of small pores in each NLP, we might expect the fusion rate we measure (the number of bursts per unit time) to increase with copy number. On the contrary, we find the rate saturates above ~ 4 copies. While we cannot exclude that the multiple pores could be occurring simultaneously, this saturation strongly suggests the number of pores is not increasing.

In conclusion, although we cannot rule out that occasionally a small number of pores simultaneously appear in a single NLP, all the evidence suggests this is an infrequent event.

We added these arguments in the manuscript.

8) In the section 'Derivation of best-fit model parameters[…]', the authors showed how they obtain the parameters in the model by fitting the calculated results of the model (equation 11) to the experimental data. For example, they fit the results in the range of 0.2 nm ≤ rpo ≤ 1.5 nm to obtain the tension of membrane γ and the steric-hydration force length scale λ (0.5 nm ≤ rpo ≤ 2.5nm) to determine a range for epsilon (4 nm ≤ rpo ≤ 4.5nm) to obtain tau. However, in equation 11, the total free energy depends on all the parameters in the full range of rpo, so how can the authors obtain an individual parameter in a different range of rpo?

The different parameters in the model exert their strongest influence over different ranges of pore sizes and SNARE copy number. For each parameter, we fix its value to produce a best fit of model predictions to experiment in its dominant range. Once parameters are set in this way, we compare model predictions and experimental data for all values of pore size, with good agreement over the entire range (Figure 7B). This scheme tests our model more stringently, we believe, than simultaneously choosing all parameters to globally optimize the fit to all data.

For example for the membrane parameters (tension γ, hydration length λ and bending modulus κ) it is best to fit to pure membrane data. The closest we can get to this is 1 SNARE data (vNLP1, Figure 7A). Thus, we choose λ to reproduce the location of the minimum of the experimental vNLP1 curve, as these smallest scales are dominated by hydration effects. As the tension γ largely determines the slope beyond this minimum, we chose γ to fix this slope by fitting the model in the region 0.5 nm ≤ rpo ≤ 1 nm (i.e., from the minimum to the limit of the vNLP1 data). Finally, the bending modulus κ is a fixed material parameter whose value we use from previous experimental measurements in the literature.

As SNAREs are introduced, the model predicts the minimum barely shifts, while the slope beyond the minimum decreases with increasing numbers of SNAREs by an amount depending on the zippering energy parameter ϵzip. Thus we selected a typical experimental curve (vNLP30) and chose ϵzip to reproduce this slope beyond the minimum up to the limit of the data (0.5 nm ≤ rpo≤ 2.5 nm).

Finally, for pore sizes beyond ~2.5 nm the ApoE proteins are twisted and resist further expansion in our model (Figure 7C and Figure 7—figure supplement 1A). Only in this region is the free energy profile affected by τ, the torque per unit length exerted by these proteins. Thus, we obtained τ as a best-fit parameter by comparing slopes for large pore sizes (> 4 nm) where twisting is the dominant effect resisting expansion.

To help clarify this overall approach, we modified “Derivation of best-fit model parameters by fitting model-predicted free energy to experiment” in Materials and methods to include a better description of this procedure.

Please note. In the originally submitted manuscript we wrongly used the vMSP curve of Figure 7a to fix the membrane parameters γ, λ. In the revised manuscript we used the correct curve, vNLP1 (the closest to membrane-free). Thus, the best fit parameter values were updated; they differ typically by only ~20% from those presented in the original manuscript.

9) The parameters in the mathematical model (i.e., rpo, h, δ, D, and the twisting angle phi of ApoE proteins) should be explicitly shown in the model in Figure 7.

Thank you for this excellent suggestion. We updated the schematic of Figure 7C (and caption) so that these quantities are now explicitly indicated (this is our understanding of the reviewer’s suggestion). To further convey the nature of ϕ, the twist angle of the ApoE proteins, we also updated Figure 7—figure supplement 1A to include the dependence of ϕ on pore size rpo.

10) To calculate the second term in equation 7, the authors approximated the pore as a cylinder of radius rpo and height λ = 0.13 nm. It seems that the height of the cylinder is too small, as the neck of the fusion pore can be on the order of ~5 nm in height. Please explain why choosing such a small height for the cylinder is a reasonable approximation.

We updated “Free energy contribution from short-ranged steric-hydration forces” of Materials and methods to explain this point, as follows.

The steric hydration interaction has a very short spatial range, of order the hydration scale λ. The biggest contribution to the energy comes from the narrowest part of the pore, the waist of diameter 2rpo. In addition, a height l of pore such that the pore diameter increases to 2rpo + λ will contribute; parts of the pore further away than l contribute negligibly because the pore diameter is much greater than 2rpo + λ at those locations, so the hydration energy will have decayed to negligible values relative to its value at the waist. The value of l is very small (much smaller than the true pore height h) because λ is very small.

Please note: on re-examining this point in response to the reviewer’s query, we discovered that we had erroneously concluded that the height of the pore that contributes is equal to the hydration length, l = λ. In fact, the correct expression is l = l=2λ(H+2δ where δ is the thickness of the membrane. We updated “Free energy contribution from short-ranged steric- hydration forces” to reflect this correction and redid our procedure to obtain λ as the best-fit parameter. The best-fit value is now λ = 0.10 nm (cf. 0.13 nm previously), Figure 7—figure supplement 2. The figures presenting our model results were accordingly updated; most are almost identical to the original figures (Figure 7B, Figure 7—figure supplement 1).

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

The manuscript has been improved but there are two remaining issues that need to be addressed before acceptance, as outlined below, using the same bullet numbers used for the points of the decision letter.

2) Some of the arguments offered with regard to lipid composition are reasonable, but saying that 'short PI(4,5)P2 does not have an effect (data not shown)' is not convincing, and the problem still remains that it is difficult to control lipid composition in this system and make it similar to physiological. This is particularly important considering the low percentage of fusion that they now report (point 4). The authors cite Giraudo et al., 2006, but the work in that paper suffers from the same problem. This concern does not invalidate the results from the authors, but they should acknowledge the problem in the manuscript. The fact is that lipids could play key roles in the fusion mechanism and the tendency of SNARE-centric models of membrane fusion to ignore this fact is deleterious to the scientific discussion in the field.

We agree with the reviewer that our assay does have the limitation that the lipid compositions involved do not match physiological lipid compositions. Accordingly, we reworded the paragraph that discusses this issue (subsection “A few SNARE complexes are sufficient to create a fusion pore, but many more are needed to dilate it”). This paragraph now explicitly states this limitation:

“In contrast, in our vNLP-tCell fusion assay, the target membrane is the outer leaflet of the plasma membrane which is largely devoid of negatively charged lipids. In general, a limitation of our system is that the lipid composition of the outer leaflet of the t-SNARE-presenting cell differs substantially from that of the plasma membrane inner leaflet, and lipid composition can play a key role in fusion.”

In addition, the final sentence of the paragraph regarding the tolerance of SNARE- only fusion to various lipid compositions was softened. It now reads:

“This swap resulted in similar fusion rates and pore properties for two different SNARE copy numbers per NLP (Figure 5—figure supplement 3), suggesting that fusion mediated by SNAREs alone may not be very sensitive to target membrane composition within a certain range.”

4) The authors again make some good arguments, but they really do not address the heart of the problem. With such low percentage of fusion, it seems very likely that fusion occurs for low populations of nanodiscs with higher copy number of SNAREs than the mean. The authors claim that this is unlikely because 'the fusion rate does not increase for mean copy number greater than 2-4 per NLP side (Figure 5A)'. However, in Figure 5A one can see a tendency for the fusion rate to increase up to vNLP15. Even though the differences may not be statistically significant, one just cannot draw the conclusion written by the authors. Hence, the authors should explicitly acknowledge that the actual numbers of SNAREs underlying fusion may be (in fact that are very likely to be) higher than the mean values described. The same issue applies to measurements of conductance.

In response to the reviewer’s comment, we statistically compared the fusion rates for copy numbers equal to and greater than 4. (In the previous version of the manuscript we compared these fusion rates to the value for zero copy number, i.e. SNARE-free discs, but we did not compare them to one another). This analysis shows definitively that there are no statistically significant differences between the fusion rates for vNLP4, vNLP8, vNLP15 and vNLP30. Thus, there is no evidence for a bias to copy numbers higher than the mean value arising as a result of higher fusion rates for higher copy numbers.

We made the following changes to the manuscript. (1) This statistical analysis is now depicted in Figure 5A, mentioned in its caption, and the source data and statistical analyses are uploaded as a zipped source data file for the figure. (Chernomordik and Kozlov, 2008) We added a new paragraph (Discussion, paragraph three) mentioning and interpreting this statistical result. (3) The new paragraph also presents a condensed version of the simple argument we presented in our previous rebuttal, based on an assumed Poisson distribution of copy numbers. This argument shows that, even if such a bias were present, the copy numbers that fuse would still increase with the mean copy number so that the general trends we report would remain valid. (4) This new paragraph also includes an acknowledgement that the bias feared by the reviewer remains a possibility: “While we cannot categorically exclude this possibility, we think it unlikely.”

While carrying out the new analysis, we discovered an error in the calculation of the fusion rate for the vNLP1 samples (Figure 5A). The error resulted in an estimate twice the actual rate for this category only. This error did not change any of our conclusions (the fusion rates are not significantly different among the vNLP4, vNLP8, vNLP15 and vNLP30 samples, at least two SNAREs per face are needed for significant fusion, etc.). We updated the figure and our analysis using the correct rate. We also uploaded the analysis files (which include raw data) as source data, as mentioned above.


Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd