Symmetrical organization of proteins under docked synaptic vesicles

Abbreviations

CLEM, correlative light‐electron microscopy

Cpx, complexin

EM, electron microscopy

ET, electron tomography

fLM, fluorescence light microscopy

MBIR, model‐based iterative reconstruction

NGF, nerve growth factor

PM, plasma membrane

SNAP, SNARE protein

SV, synaptic vesicles

Syt, synaptotagmins

VPP, volta phase plate

The release of neurotransmitters stored within synaptic vesicles (SV) occurs in a remarkably synchronous fashion, taking place between 200 μs and 1 ms after the action potential triggered influx of calcium ions (Ca²⁺) 1, 2. The central components involved are the synaptic SNARE (soluble N‐ethylmaleimide–sensitive factor attachment protein receptor) proteins that catalyze SV fusion and fusion regulators, complexin (Cpx) and Synaptotagmins (Syt) 1, 2, 3. During this process, partially‐zippered SNARE complexes (SNAREpins) tightly clamped by Cpx and Syt1 are synchronously released to complete fusion 2, 3. Because each individual SNAREpin requires ~ 1 s to drive fusion 4, 5, 6, either very large numbers of SNAREpins would be required, or alternatively a more modest number of SNAREpins would need to be organized in a highly co‐operative framework.

In principle, there could be up to ~ 70 SNAREpins formed because each SV contains about this number of copies of the v‐SNARE, VAMP2/Synaptobrevin 7. But this hypothesis seems untenable because several studies have suggested that synchronous release can still take place with as few as 2–3 copies of SNARE proteins (SNAP) 8, 9, 10. Evidence favoring the main alternative, a well‐organized co‐operative structure, has come from the recent discovery that the Ca²⁺ sensor/clamp Syt1 self‐assembles in vitro into ring‐like oligomers based on polymerization of its C2B domains 11, 12, 13. Indeed, a targeted mutation (F349A) in the Syt1 C2B domain that selectively destabilizes the Syt1 oligomers was found to release the fusion clamp and increase spontaneous exocytosis in a model neuroendocrine lineage cell line 14. However, the proposed ring‐like oligomer of Syt1 has never been visualized, directly or indirectly, in cells.

X‐ray crystallography has revealed that two Syt1 molecules simultaneously bind the SNAREpin on opposite surfaces, each interaction mediated by its respective C2B domain 15. One Syt1 C2B domain (‘primary’) binds to SNAP‐25 within the SNAREpin in a manner compatible with its further assembly into a Syt1 ring‐oligomer and likely contacts the plasma membrane (PM). The other Syt1 C2B domain (‘tripartite’) is bound in an incompatible manner, and likely contacts the SV surface16. Both the primary and tripartite‐binding sites are required for clamping and for synchronous release 15.

We recently showed how these structural features, namely the Syt1 ring oligomers and the dual‐SNARE binding module, can be easily and elegantly combined with the curved structure of the SNARE‐chaperone Munc13 to form a symmetrical assembly between the SV and PM 16. This proposed organization (‘buttressed ring’) will naturally template a defined number of SNAREpins, sterically block their full‐assembly, and co‐operatively release them upon Ca²⁺‐influx 16.

If this or another supra‐molecular organization of SNAREpins were to exist under a docked vesicle, it could potentially be visualized by cryo‐electron tomography (cryo‐ET), which has been successfully used to reveal symmetrical protein structures in situ in cells at resolution sufficient to delineate their overall shape and locate protein constituents within them based on high‐resolution structures of isolated components 17, 18, 19.

Here, we report the initial results from cryo‐ET concerning the organization of the clamped exocytic machinery under SVs in neurites that develop from nerve growth factor (NGF)‐differentiated pheochromocytoma (PC12) cells. Consistent with the buttressed ring hypothesis, we observed six distinct masses arranged symmetrically at the SV–PM interface. Each of these rod‐like densities could accommodate the full complement of the minimal release machinery, including a SNAREpin with its two Syt1, a Cpx, and a copy of the other assembly chaperone Munc18. Even though the Syt1 ring‐like structure could not be visualized at this limited resolution, the symmetrical arrangement was entirely disrupted with the Syt1 F349A mutation. In sum, our data strongly suggests that a small number of SNAREpins prearranged by Syt1 ring‐like oligomers represents a central organizing principle of Ca²⁺‐regulated exocytosis.

Materials and methods

The overall workflow of experimental procedures is outlined in Fig. S1. To generate the VAMP2‐4X‐pHluorin construct, we used a QuickChange mutagenesis kit (Agilent Technologies, Santa Clara, CA, USA) to introduce L70D, A74R, A81D & L84D mutations into the previously described VAMP2‐pHluorin construct 20, 21. The Syt1 F349A used has been described previously 14. The rat adrenal PC12 cell line used was a gift from B. Ehrlich, Yale University.

Results and Discussion

In this study, we used NGF‐differentiated PC12 cells combined with cryo‐ET to investigate the protein organization under docked SV (Fig. S1). Before differentiation, the PC12 cells contain a mixture of secretory vesicles, mainly consisting of ‘dense‐core’ vesicles (~ 100–150 nm in diameter) but also a second population of smaller ‘clear’ vesicles (~ 40–60 nm in diameter), which are similar to SV in composition and function 34, 35. NGF treatment (3–7 days) dramatically increased the relative abundance of the small, synaptic‐like vesicles (~ 45 nm in diameter), with a majority localizing to the neurite‐like varicosities that develop during differentiation (Fig. 1A,B). Importantly, these neurite varicosities were sufficiently thin and ideal for cryo‐ET analysis (Fig. 1C).

To preserve these neurite varicosities in their native state, we carried out an in situ vitrification, using NGF‐differentiated PC12 cells grown directly on gold EM grids. The NGF‐differentiated PC12 cells lack mature active zones and do not maintain a pool of release‐ready vesicles 34, 35. Thus, to increase the population of such vesicles under these conditions, we overexpressed a VAMP2 protein with mutations in the C‐terminal half (L70D, A74R, A81D & L84D; termed VAMP2‐4X), which allows SNAREpins to zipper approximately halfway but not completely, and as such allows vesicles to dock but not fuse 20, 21. EM images confirmed that the VAMP2‐4X overexpression does not alter the ultrastructure of the neurite varicosities as they were indistinguishable from the nontransfected neurites (data not shown). However, the proportion of SVs proximal to PM in VAMP2‐4X varicosities was nearly double that of nontransfected boutons (49.1 ± 0.03% vs 25.3 ± 0.03%, respectively, P < 0.05). Specifically, we used a version of VAMP2‐4X that was tagged with pHluorin on the luminal side of the SV and employed a correlative light‐EM (CLEM) workflow to identify the mutant neurite varicosities for cryo‐ET data collection (Fig. 1C).

We acquired a tilt‐series using a 300‐kV Titan Krios; which along with the combination of direct electron detector and Volta‐phase plate and electron energy filter, resulted in tomograms with good contrast. To delineate the molecular architecture at the vesicle–membrane interface, we focused on the vesicles that locate close/proximal to the top or bottom PM (Fig. 2). We used the XZ slice of the tomographic 3D reconstruction to initially screen and identify PM‐proximal vesicles. For this subset of vesicles, we further carried out Z‐stack analysis on the XY slices and the absence of any organelles between the periphery of the vesicle and the boundary of the cell was used as the criteria to define ‘docked’ vesicles.

From the 574 high‐quality tomograms of neurites thus obtained, we generated reconstructions of 4757 docked vesicles. These vesicles were first sorted into twenty 3D‐classes based on the size and shape of the vesicles (Fig. S3). The 3D classes featuring spherical vesicles with diameter of 45 ± 5 nm were then selected to generate a homogeneous dataset. These were further subjected to several cycles of alignment and global averaging (using vesicle center as reference) until there was no further improvement in the homogeneity in size and shape (45 ± 2 nm). These vesicles (2434 in total) were pooled together for local alignment and averaging with the center of alignment now at the site of docking (i.e., SV–PM interface).

After five cycles of alignment and classification, the tomograms were sorted into ten 3D‐classes (Fig. S3). In each and every 3D‐class, we observed six rod‐like densities in the XY plane corresponding to the vesicle–cell membrane interface. These protein densities were symmetrically organized in a clear majority (~ 80%) of the 3D‐classes. To improve the signal‐to‐noise ratio, all subtomograms in these final ten classes were pooled together and averaged, both without (Fig. 3A) and with imposing a sixfold rotational symmetry (Fig. 3B). The resulting density map (for both symmetrized and nonsymmetrized reconstructions at resolution of 5 nm and 5.4 nm respectively) showed six pronounced protein densities symmetrically organized (~ 60° apart) as if equally spaced on a common circle of ~ 35 nm diameter. There was also a consistent, but weaker protein density in the center.

The surface representation of the resulting 3D‐map revealed that each of the six radially‐organized protein masses likely corresponds to highly structured, large protein complex as it is prominent even at the highest threshold level (σ = 2). In contrast, the central protein mass was weak and visible (Fig. 3A,B) only at a lower threshold (σ = 1), signifying unstructured regions. Critically, no strong electron density was observed on the vesicle outside the PM contact regions, indicating that the observed protein organization was specific to the vesicle–membrane interface. Note that the six‐sided arrangement in the center of Fig. 3B is present only at the lower threshold and most probably, an artifact due to the imposed symmetry.

The synaptic‐like vesicles that we have reconstructed are expected to be analogous to the synaptic vesicles at neuronal synapses 34, 35. In these docked vesicles, SNAREs are ?clamped’ in an approximately half‐zippered state, such that they can release rapidly following Ca²⁺‐influx. Consistent with this, (a) these vesicles are located in close proximity to the PM; (b) they contain the overexpressed VAMP2‐4X‐pHluorin, which prevent fusion as shown by CLEM; (c) the lumen of these SV is inaccessible to protons from the extracellular medium (Fig. S4).

From established biochemistry and structural biology, each SNAREpin is known to be bound within an ‘exocytosis module’ consisting minimally of two copies of Syt1, one copy of Cpx, and one copy of Munc18. The combined molecular weight of this module is ~ 250 kDa. It is likely that each of the six symmetrical masses we have discovered consists of one such module. Modeling shows that the shape and size of each such mass is sufficient to accommodate one exocytosis module (Fig. 3C) but not two. At the present resolution, largely limited by the geometry of docked vesicle (only top/bottom and no side view) and the missing‐wedge effect, it is not possible to rigorously assign specific protein domains to features, even consistent features, within the masses. As a result of the missing wedge, the resolution in the XY plane is better than it is along the Z axis (perpendicular to the PM plane), which has the effect of exaggerating the shape of objects along Z. This likely explains why there appears to be significant unfilled space in XZ views (Fig. 3C, middle) but much less so in XY views (Fig. 3C, left). Altogether it is evident that the shapes are well occupied by one module and clearly cannot accommodate two modules. Therefore, we can clearly conclude that exactly six SNAREpins are present and likely co‐operate to release synaptic‐like vesicles (Fig. 3C).

How they co‐operate is not apparent from these images, but their exactly symmetrical positioning every 60° along a circle of ~ 35 nm outer diameter strongly suggests they are attached at regular intervals to an underlying framework. Considering that Syt1 is known to polymerize into ring‐like oligomers of similar diameters 11, 12, 13, and also to bind SNAREpins 15, the simplest hypothesis would be that each mass consists of a SNAREpin with its associated proteins bound to a Syt1 ring at regular intervals. We would not expect to directly visualize the proposed Syt1 ring at the present resolution if it were present due to its small thickness (~ 4 nm). The same is true for the even thinner MUN domain‐based outer rings predicted by the buttressed ring hypothesis.

To test this hypothesis, we tested the effect of a Syt1 mutant (F349A), which is designed to specifically destabilize the Syt1 oligomers 14. As reported earlier, this F349A mutation supports vesicle docking and fusion in PC12 cells, but failed to clamp the vesicles 14. We over‐expressed the Syt1 F349A mutant, along with VAMP2‐4X‐pHluorin and generated 3105 reconstructions of docked SVs from 141 tomograms. We carried out subtomogram averaging and 3D classification exactly as before. We observed protein densities at the vesicle–membrane interface in the subtomogram class averages (Fig. 4A) but there was no evidence of symmetrical organization. In stark contrast, symmetrical organization was clearly evident in each of the subtomogram class averages for the wild‐type Syt1 (Fig. S3). We suspect that the observed densities with the F349A mutant are weak as the electron density of the dis‐organized exocytosis modules is smeared out during the averaging process. At the very least, the results establish that the same surface of the C2B domain that is utilized for ring‐like oligomer assembly in vitro is also needed to circularly organize the SNAREpin‐containing exocytosis modules, thus providing compelling albeit indirect evidence that the organizing framework consists of ring‐like oligomers of Syt1 (Fig. 4B).

How can precisely six SNAREpins be assembled from each SV when there is a very much larger supply of both v‐ and t‐SNAREs locally available? Evidently the limiting component must be a required chaperone. Of the two chaperones (Munc18 and Munc13) required for SNAREpin assembly 2, 3, 36, Munc18 enters the reaction bound to Syntaxin in a 1 : 1 stoichiometry 3, 36, 37 and thus, unlikely to be a limiting component. By elimination, this suggests that Munc13, which co‐operates with Munc18 to assemble half‐zippered SNAREpins, but enters separately is most likely the limiting component.

In formulating the buttressed ring hypothesis 16, we pointed out that the conserved MUN domain of each Munc13 forms an arc that subtends an angle of ~ 60° and thus can be naturally arranged to form a planar, closed circle formed of exactly six copies of these chaperones. This proposed outer ring of Munc13 can snugly enclose an inner ring of Syt1 oligomers and was predicted to assemble exactly six SNAREpins under each vesicle positioned every 60°, which would explain what we have now observed. We note that numerous related chaperones containing the conserved MUN domain of Munc13 play an essential role in the assembly of most if not all SNARE complexes, including those facilitating the majority of intracellular transport processes which occur constitutively, not involving a Synaptotagmin clamp 16, 38, 39. Perhaps these MUN domains also form rings and template six SNAREpins, which in the absence of an inner Synaptotagmin ring can proceed to fuse spontaneously without an imposed delay.

It may seem surprising that a handful of SNAREpins – if they are synchronously released – could achieve sub‐millisecond fusion pore opening. However, recent modeling that includes the concept of mechanical coupling predicts exactly this 40. Membranes are rigid on the length scale of a fusion pore, and as a result, SNAREpins will necessarily be mechanically coupled. Each time any one of the SNAREpin zippers up, it pulls the membranes closer together, triggering its neighbors to zipper at nearly the same time 40. At this stage, each SNAREpin still has ~ 5 k_BT of mechanical work to deliver toward overcoming the bilayer fusion barrier (~ 25 k_BT for physiologically relevant lipid compositions 41). For a single SNAREpin, this barrier is overcome in about 1 s 4, 5, 6 Therefore, each added SNAREpin reduces this waiting time for fusion by a factor of about 10² (calculated as e ^(−5 kBT)), so as the extraordinary consequence of mechanical coupling just three SNAREpins will reduce the time for vesicle release to ~ 0.1 ms 40.

Another feature of mechanical coupling is that, together with the measured energy landscape of the SNAREpin zippering 42, it also predicts that there is an optimum number of SNAREpins that maximizes fusion rate 40. The reason is that fusion cannot occur until the last SNAREpin fully zippers, and when there are a large number of them this becomes rate limiting. Each SNAREpin is metastable in the half‐zippered conformation 42, stabilized by a small energy well resulting from the polar ‘zero’ layer of the four‐helix bundle that interrupts the otherwise continuous hydrophobic heptad repeat 43. Though small for any one of the SNAREpin, the probability that one or more SNAREpins are ‘left behind’ increases exponentially with the number of SNAREpins 40. The tradeoff between these opposing processes predicts an optimum of 4–6 SNAREpins is required to achieve submillisecond release as needed for synaptic transmission, depending on the exact choice of various parameters 40. It is remarkable that this optimum, predicted from first principles, should align so closely with what molecular imaging of synaptic‐like vesicles in situ suggests.

Author contributions

XL, AR designed experiments, collected, analyzed and interpreted cryoET data. KG analyzed/interpreted cryoET data and built molecular models. RK, AC and SG optimized cell cultures, over‐expression and collected data. SSK, JL and JER designed the experiments, provided supervision, analyzed/interpreted the data and wrote the manuscript. All authors read and revised the manuscript.

Supporting information