“Kiss-shrink-run” unifies mechanisms for synaptic vesicle exocytosis and hyperfast recycling

Abstract

Synaptic vesicle (SV) exocytosis underpins neuronal communication, yet its nanoscale dynamics remain poorly understood owing to limitations in visualizing rapid events in situ. Here, we used optogenetics-coupled, time-resolved cryo–electron tomography to capture SV exocytosis in rat hippocampal synapses. Within 4 milliseconds of synaptic activation, SVs transiently “kiss” the plasma membrane, forming a ~4-nanometer lipidic fusion pore flanked by putative soluble NSF-attachment protein receptor (SNARE) complexes and then rapidly “shrink” to approximately half of their original surface area. By 70 milliseconds, most shrunken SVs recycle via a “run-away” pathway, whereas others collapse into the presynaptic membrane. Ultrafast endocytosis retrieves the expanded presynaptic membrane after 100 milliseconds. These findings reveal a “kiss-shrink-run” mechanism of SV exocytosis and hyperfast recycling, reconciling conflicting models and elucidating the efficiency and fidelity of synaptic transmission.

INTRODUCTION:

Synaptic vesicle (SV) exocytosis is triggered by an action potential and leads to neurotransmitter release. As such, SV exocytosis is fundamental to neuronal communication. However, the structural and biophysical mechanisms underlying SV exocytosis remain incompletely understood. In particular, questions remain regarding the dynamic interactions between the SV membrane, the presynaptic membrane, and the protein complexes involved. This gap has fueled a long-standing debate over the existence of transient “kiss-and-run” fusion versus irreversible “full-collapse” fusion in central synapses.

RATIONALE:

Resolving this debate requires techniques that can achieve nanometer spatial resolution and millisecond temporal resolution. To address this need, we developed a time-resolved, cellular cryo–electron tomography (cryo-ET) method to image intact synapses in cultured rat hippocampal neurons. This approach integrated optogenetic stimulation for synaptic activation and plunge-freezing at millisecond intervals. It also provided high three-dimensional spatial resolution, enabling accurate vesicle size measurements and detailed structural analysis of vesicle–plasma membrane interactions. Using this technique, we acquired more than 1000 tomograms of entire excitatory synapses, frozen at various time points from 0 to 300 ms post–action potential. This large dataset facilitated rigorous statistical analysis of vesicle states and subtomogram averaging to visualize fusion-pore structures.

RESULTS:

Near the presynaptic active zones, we observed two distinct SV populations with diameters centered at ~29 and ~41 nm. These SVs were mainly categorized into seven distinct structural states: tethered (large and small), semifused (large and small), pore-opened (large and small), and Ω-shaped (small). Notably, the population of small SVs decreased markedly after neural network inactivity with 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and was completely absent when vesicle exocytosis was blocked by tetanus toxin.

Time-resolved cryo-ET revealed a sequence of transitions among these vesicle states. In the resting condition, large SVs were primarily tethered to the plasma membrane (docking). Within 4 ms post–action potential, docked SVs transitioned to large semifused SVs (priming) that transiently “kiss” the plasma membrane. These primed SVs then formed pore-opened SVs with a ~4-nm lipidic fusion pore flanked by putative soluble NSF attachment protein receptor (SNARE) complexes. These fused SVs rapidly shrank to small pore-opened SVs with approximately half the surface area of the original large SVs. Most shrunken SVs subsequently closed their fusion pores and converted into small semifused SVs.

By 70 ms, small semifused SVs detached from the presynaptic membrane (“run-away”), whereas the remaining shrunken, pore-opened SVs fully collapsed into the plasma membrane. After 100 ms, the run-away SVs began to migrate to the periphery of the SV cluster, and the resulting expanded presynaptic membrane began to be retrieved through ultrafast endocytosis.

CONCLUSION:

Our study identifies a kiss-shrink-run pathway as the dominant biophysical mechanism for SV exocytosis and rapid recycling in hippocampal synapses. This kiss-shrink-run mechanism reconciles the kiss-and-run and full-collapse models of neurotransmission and provides a unified explanation for the high efficiency and fidelity of synaptic transmission. Our integrative methodology also establishes a general framework for probing membrane dynamics and molecular interactions in situ with high spatiotemporal precision. ⬜

Editor’s summary

Neurons transmit signals through synaptic vesicle release, but the nanoscale dynamics of this process have been unclear. Tao et al. revealed these dynamics in hippocampal synapses using optogenetics and time-resolved cryo–electron tomography (see the Perspective by Lichter). Their innovative approach captures synaptic vesicle exocytosis at millisecond resolution, identifying the “kiss-shrink-run” pathway in which synaptic vesicles briefly contact the presynaptic membrane, shrink significantly, and then detach for rapid recycling. This mechanism unifies competing neurotransmitter release models and elucidates the underpinnings of synaptic efficiency and reliability. —Stella M.Hurtley

Graphical Abstract

Kiss-shrink-run mechanism of SV release revealed by time-resolved cryo-ET.

A millisecond-precision time-resolved cryo-ET system, involving plunge-freezing coupled with optogenetic stimulation (top left), captured distinct intermediate states of SV exocytosis in intact hippocampal synapses (top right). With population analysis and subtomogram averaging, a kiss-shrink-run sequence was revealed as the prominent SV exocytosis and recycling pathway in hippocampal synapses (bottom). AP, action potential.

The brain’s function relies on information transmission between neurons. This task is accomplished by synaptic vesicle (SV) exocytosis, which releases neurotransmitters in response to a presynaptic action potential (AP) (1–5). Over the past several decades, electrophysiological studies have uncovered a series of rapid events involved in SV exocytosis, including docking, priming, and fusion-pore formation, all of which occur within milliseconds (5–8). Biochemical and molecular biological analyses have identified key synaptic proteins, particularly the soluble NSF-attachment protein receptor (SNARE) complex, as critical mediators of these biophysical events (5, 9–12). However, the fundamental biophysical and structural underpinnings of synaptic transmission—particularly how vesicular and presynaptic membranes and protein complexes interact with one another during this highly dynamic process—have remained elusive (13).

Two main mechanisms of SV exocytosis and recycling have been proposed, the “kiss-and-run” model and the “full-collapse” model. In the kiss-and-run model, the fusion pore opens briefly—allowing neuro-transmitter release—then closes, and the vesicle is retrieved for reuse (14–21). By contrast, the full-collapse fusion model involves the complete merging of the vesicle with the plasma membrane (3, 16). The kiss-and-run model is considered more energy efficient and allows for higher rates of synaptic transmission (15, 18, 22–24).

Classic studies using electron microscopy (EM) have visualized vesicle exocytosis and subsequent endocytosis in the neuromuscular junction at milliseconds and minutes after electrical stimulation, providing direct evidence for the occurrence of full-collapse fusion and clathrin-mediated endocytosis (3, 25). In hippocampal synapses, studies combining optical stimulation with high-pressure freezing EM (flash-and-freeze) revealed the collapse of fused vesicles followed by clathrin-independent endocytosis of large (~80-nm diameter) endocytic vesicles that occur ~100 ms after stimulation (26). This process, known as ultrafast endocytosis, has been confirmed through optical imaging and electrophysiological recording experiments (27, 28). Conversely, super-resolution optical imaging has captured the dynamic process of kiss-and-run exocytosis of large dense-core vesicles in chromaffin cells (29–32). However, the occurrence of kiss-and-run of SVs in central synapses has remained controversial, partly because tools with sufficient temporal and spatial resolution to resolve key features, such as fusion-pore dynamics, are lacking (13–20, 26).

In recent years, cryo–electron tomography (cryo-ET) with nanometer spatial resolution has been used to study synaptic ultrastructures, revealing detailed structural features of vesicle interactions with the plasma membrane (33–39). Particularly, molecular bridges, referred to as “tethers” that connect vesicles with the presynaptic membrane, have been shown to depend on proteins such as SNAP25 or Munc13 (35). With high-concentration K⁺ spray stimulation followed by plunge-freezing, instances of vesicles directly contacting the plasma membrane have been observed in isolated synaptosomes and interpreted as exocytosis intermediates (34). However, owing to limited sample size and possible complications associated with synaptosome preparation, the precise temporal sequence of these intermediates—and whether they represent stages of exocytosis or endocytosis—remains unclear. To directly visualize the dynamic process of SV exocytosis in situ, we developed a time-resolved, cellular cryo-ET method coupled with millisecond-precision optogenetic stimulation.

Distinct subpopulations of SVs near the presynaptic active zone

We imaged intact synapses of rat hippocampal neurons grown on holey carbon-coated gold cryo-EM grids using cellular cryo-ET and obtained their three-dimensional (3D) reconstructions at molecular resolution (Fig. 1, A and B, figs. S1 and S2, and movie S1), following previously established procedures (38–43). Synapses within the cryo-tomograms were identified based on characteristic features, including closely apposed compartments, one of which was populated by vesicles, and a relatively uniform-width synaptic cleft containing transcleft protein densities (Fig. 1, A to D, figs. S1A and S2A, and movie S1). Synapses were further classified as excitatory or inhibitory based on criteria established in previous correlative imaging studies (38, 39, 41–43). Inhibitory synapses were distinguished by their uniform, thin, sheet-like postsynaptic densities (PSDs) and the presence of γ-aminobutyric acid type A (GABA_A) receptor–like densities on postsynaptic membranes (fig. S2A), whereas the remaining synapses were considered excitatory, exhibiting thicker PSDs or occasionally lacking clear PSDs (Fig. 1, A and C). Here, we focused exclusively on excitatory synapses.

Fig. 1. In situ cryo-ET imaging revealed various types of SVs near the AZs under basal conditions.

(A) A tomographic slice showing the ultrastructure of an excitatory synapse. Subcellular components—including SVs (green circles), actin filaments (orange arrows), endoplasmic reticulum (ER), mitochondria (Mito), ribosomes (cyan circles), and a multivesicular body (MVB) in the presynaptic bouton and/or postsynaptic spine—are clearly visible. The borders of the presynaptic AZ and PSD are indicated by a magenta double-arrowhead line. (B) 3D rendering of the synaptic structures in the tomogram shown in (A). SVs are color-coded by diameter from small (red) to medium (green) to large (blue). (C) Zoomed-in view of the AZ from (A) showing tethered vesicles (blue arrows) and semifused vesicles (red arrowheads). (D) 3D rendering of the presynaptic profile, including an orthogonal view (inset) of the entire AZ. (E) Scatter plot showing vesicle diameter versus distance from the presynaptic membrane for vesicles within 50 nm of the membrane. The red dashed line marks the size threshold distinguishing small from large vesicle clusters, and the blue dashed line denotes the distance threshold separating the AZ band from the reserve vesicle pool. (F) Number of SVs as a function of their distances from the presynaptic membrane. The AZ band was defined as the presynaptic region within 20 nm of the AZ (blue dashed line). The red, green, and blue lines represent the three Gaussian components, and the gray dashed line represents their sum. (G) Diameter distribution of SVs in the AZ bands, which was fitted with a double-Gaussian function (μ1 = 28.99, σ1 = 2.73; μ2 = 41.46, σ2 = 3.86). The intersection point of the two individual Gaussian curves is at 34 nm (red dashed line). The red and black lines indicate the two Gaussian components, and the blue dashed line represents their sum. (H) Examples of various types of vesicles in the AZ bands, including tethered, semifused, pore-opened, and Ω-shaped SVs. Colors are the same as in (I). (I) Size distributions of different types of SVs in the AZ bands. Statistical analysis combined data from 139 synapses under basal conditions (without CNQX treatment). The number of experiments and data points used for analysis are detailed in table S1.

Opposite the synaptic cleft and PSD, the presynaptic membrane region known as the active zone (AZ) is where SVs dock and fuse (44, 45). Consistent with the absence of heavy-metal staining in cryo-ET, the dense projection at the AZ that is prominent in conventional EM imaging was less apparent in our datasets (39, 44, 46) (Fig. 1A). We therefore delineated the boundary of the AZ by the region underlying the uniform-width synaptic cleft. We further referred to the ~20-nm cytoplasmic layer immediately beneath the AZ as the submembrane AZ band, which accommodated most membrane-interacting SVs and may correspond to the readily releasable pool (7, 23) (Fig. 1, A to F, and fig. S1A). By quantifying the size of the SVs, we found that many SVs within the AZ bands were substantially smaller than those away from them in the presynaptic boutons (Fig. 1, C to I, and fig. S2). Overall, the size distribution of SVs in the AZ bands exhibited two distinct Gaussian peaks, one centered at an average diameter of 29 nm and the other at 41 nm; the latter was typical for most of the vesicles located outside the AZ bands (Fig. 1, E and G). We defined 34 nm, the intersection point of the two individual Gaussian curves, as the threshold to categorize SVs into two distinct size groups (Fig. 1G).

Within the AZ band, we identified four major types of SVs: (i) tethered vesicles, linked to the presynaptic membrane via protein filaments; (ii) semifused vesicles, partially fused with the presynaptic membrane; (iii) pore-opened vesicles, which form a narrow fusion pore with the plasma membrane (neck width <22 nm); and (iv) Ω-shaped vesicles, which form a wide fusion pore with plasma membrane (neck width >22 nm) (Fig. 1H and fig. S1D; see methods for details). We also occasionally observed small pits, clathrin-coated vesicles (47), and large vesicles resembling those involved in ultrafast endocytosis (26) at the presynaptic membrane (fig. S1, F to H). Additionally, a number of tethered SVs were observed at distances up to 50 nm from the presynaptic membrane (fig. S1, I and J). The size of these vesicles strongly correlated with their type: Most tethered SVs had larger diameters, whereas most of the semifused, pore-opened, and Ω-shaped SVs had smaller diameters (Fig. 1I and fig. S1E), indicating that their direct interaction with the plasma membrane may relate to the loss of vesicular membrane.

Action potential–triggered SV exocytosis captured by time-resolved cryo-ET

To investigate whether these different types of SVs in the AZ band reflect different stages of exocytosis, and to determine the chronological sequence of this process, we developed a time-resolved cryo-ET method (Fig. 2 and figs. S3 to S5). In this approach, the action potential (AP) was triggered by optogenetic stimulation, followed by plunge-freezing with millisecond precision (Fig. 2A and fig. S5). We transfected neurons with ChIEF, a variant of channelrhodopsin (48), and confirmed that AP firing occurred in these neurons upon light stimulation (Fig. 2, B and C, and fig. S3). To suppress spontaneous recurrent activation of these neurons during sample handling, we applied CNQX (6-cyano-7-nitroquinoxaline-2,3-dione) to block excitatory transmission (49).

Fig. 2. AP-evoked vesicle exocytosis revealed by millisecond-precision time-resolved cryo-ET.

(A) Schematic overview of the home-built flash-and-plunge-freezing device. Immediately after blotting with filter papers, the sample grid is plunged into the cryogen [liquid ethane (LE) surrounded by liquid nitrogen (LN₂)]. A 470-nm laser was mounted on the side of the cryo-plunger at a fixed height atop the surface of the liquid ethane to illuminate the dropping grid. (B) Bright field (left) and fluorescence (right) microscopy images of hippocampal neurons cultured on an EM grid. The neurons expressed ChIEF-tdTomato proteins (white arrows). (C) Patch-clamp recordings showing that APs can be induced in the neuron with ChIEF protein expression (ChIEF⁺), shown in (B) by the white arrow, using a 470-nm laser stimulation. Under the same light conditions, APs were not evoked in a nearby neuron lacking ChIEF protein expression (ChIEF⁻). The blue line indicates the duration of light stimulation. (D to I) Example tomographic slices showing synapses frozen without light stimulation (D); frozen at 4 ms (E), 8 ms (F), 30 ms (G), or 70 ms (H) post-AP; or frozen at 8 ms post-AP under TeNT treatment (I). Arrowheads indicate tethered (blue), semifused (red), pore-opened (green), and Ω-shaped (brown) SVs in the AZ bands. (D1) to (I1) show 3D renderings of the presynaptic profiles in (D) to (I), respectively, including their orthogonal views of the entire AZ (magenta, presynaptic membrane; blue, tethered SVs; red, semifused SVs; green, pore-opened SVs; brown, Ω-shaped SVs; gray, SVs not within the AZ band). (J to L) Normalized number of tethered (J), semifused (K), and pore-opened (L) SVs in the AZ band of each synapse under different conditions. Data are presented as mean ± SEM. Statistical significance was evaluated using the Mann-Whitney test; *p < 0.05, **p < 0.01, and ***p < 0.001. The number of experiments and data points used for analysis are provided in table S1. Specific p values for the significance comparisons are detailed in table S2.

Optogenetic stimulation was delivered using a fixed-position laser beam intersecting the grid’s falling path (Fig. 2A and fig. S5). Using this method, we acquired cryo-tomograms of synapses frozen at approximately 4, 8, 30, and 70 ms after AP firing (post-AP), based on four different fixed positions of the laser beam (Fig. 2, D to I, and fig. S5). The plunge-freezing chamber was set to 34°C so that the ambient temperatures for the 30- and 70-ms positions were 34°C. The ambient temperatures near the cryogen surface were measured at ~24° and ~29°C for the 4- and 8-ms positions, respectively, owing to the proximity of the laser beam (fig. S5, B, C and G; see methods for details). The temporal variability of our time-resolved approach stemmed from three aspects, which were calibrated as follows: (i) The grid’s falling speed was consistent, with time variation at any given position of the grid’s falling trajectory being less than 0.6 ms, as measured using a fast-speed video recorder (fig. S5, D and E); (ii) the delay of AP induction by optogenetic stimulation was ~4.5 ms and remained temperature independent across 20° to 34°C, calibrated by electrophysiological measurements (fig. S3, D to G); and (iii) synaptic transmission was robustly evoked by optogenetic stimulation across 20° to 34°C, with the kinetics of excitatory postsynaptic current decay slowed at low temperatures (with a variability of 3 ms), calibrated by patch-clamp recordings (fig. S3, H to K). Thus, the minimal temporal variability of our approach ensured confidence in the temporal assignment of the data.

Using this time-resolved cryo-ET approach, we observed substantial increases in the numbers of both the semifused and pore-opened vesicles 4 ms post-AP (Fig. 2, K and L, and fig. S6, A to C), indicating that the generation of these vesicles was triggered by the AP. To investigate whether the formation of semifused and pore-opened SVs involves membrane fusion machinery, we treated the neurons with tetanus neurotoxin (TeNT), which blocks vesicle exocytosis through cleavage of synaptobrevin, a major component of the SNARE complex (50) (fig. S4). TeNT completely blocked the formation of semifused and pore-opened vesicles (Fig. 2, I, K, and L) and caused an accumulation of tethered vesicles in both resting and light-stimulated synapses (Fig. 2J). Thus, the AP-triggered formation of semifused and pore-opened vesicles requires functional SNARE complexes.

“Kiss-shrink” process of SV exocytosis

By separating the semifused and pore-opened SVs into large and small subpopulations (Fig. 3A), we observed that the number of large semifused SVs peaked at 4 ms post-AP and then decreased at later time points. By contrast, the numbers of large pore-opened and small pore-opened vesicles peaked at 8 ms, and the number of small semifused SVs continued to increase until 70 ms post-AP (Fig. 3B). These trends indicated a sequence of SV state transition: from large semifused SVs (peak at 4 ms), to large pore-opened SVs (peak at 8 ms), and then quickly to small pore-opened SVs (peak at 8 ms), followed by a slower transition to small semifused SVs (increase until 70 ms) with fusion pore closure. Notably, the large semifused SVs lacked fusion pores and were also observed in resting synapses (Fig. 3, A and B), consistent with the electrophysiologically defined priming state (6). In the large semifused SVs, the vesicular membrane made direct contact with a protruded tip of the plasma membrane, forming a metastable intermediate that we refer to as “kissing.” After the kissing step, the fusion pore opened, as indicated by an increase in large pore-opened SVs. These vesicles then quickly shrunk into small pore-opened SVs with approximately half of their original surface area, reflected by a diameter reduction from 41 to 29 nm. Thus, we propose a “kiss-shrink” mechanism for vesicle exocytosis: the kissed large semifused vesicles first open the fusion pore and then shrink as neurotransmitter is released (Fig. 3G).

Fig. 3. Kiss-shrink-pause pathway of SV exocytosis.

(A) Example tomographic slices showing semifused (left) and pore-opened (right) SVs of large (top row) and small (bottom row) sizes. (B) Normalized number of the four subtypes of SVs per synapse at different time points post-AP. (C to F) Central slices of the subtomogram averages for the four subtypes of SVs. Red arrows indicate the putative SNARE complex. (C1) to (F1) present 3D renderings of the subtomogram averages shown in (C) to (F), respectively. Atomic models of the SNARE protein coiled coils [Protein Data Bank (PDB) ID 3HD7 in (D1) and (E1); PDB ID 1SFC in (F1)] are presented as cylinder diagrams, fitted into the cryo-ET density. (G) Schematic diagram illustrating the kiss-shrink model of vesicle exocytosis. Data are presented as mean ± SEM. Statistical significance was evaluated using the Mann-Whitney test; *p < 0.05, **p < 0.01, and ***p < 0.001. The number of synapses, vesicles, and specific p values for these comparisons are provided in table S3.

The extensive cryo-ET dataset of SVs interacting with the plasma membrane enabled further analysis of the molecular machinery mediating membrane fusion. We performed subtomogram averaging of both large and small semifused and pore-opened vesicles (Fig. 3, C to F, and fig. S7). This revealed a clear fusion pore, with a defined bilayer wall and a direct tunnel connecting the vesicular lumen to the extracellular space (Fig. 3, D and E). The fusion pore had an approximate diameter of 4 nm, large enough to allow for the release of neurotransmitters, represented here by glutamate, which is less than 1 nm in size. We estimated that neurotransmitter release from these pore-opened vesicles occurs within a time course of ~50 μs, whereas fusion pores likely remain open for about 2 ms (see methods), sufficient for near-complete neurotransmitter release from the vesicle.

In all subtomogram averages of semifused and pore-opened SVs, we identified weak but persistent linear densities flanking the contacting sites (Fig. 3, C to F, and fig. S7). These densities likely correspond to coiled-coil structures of the tripartite SNARE complex, as previously resolved in atomic models, supporting our observation that the formation of semifused and pore-opened SVs was SNARE dependent. Indeed, multiple distinct putative SNARE densities were observed in individual vesicles (fig. S7, A and B). However, reconstructions of these vesicles using different rotational symmetries produced inconsistent results (fig. S7, I and J), suggesting that SNARE protein number and localization may be heterogeneous or that the contact sites are intrinsically asymmetric. Nevertheless, in some subtomogram reconstructions, the structures of synaptotagmin-1 and the SNARE complex could be cofitted (fig. S7J).

Overall, the distinct temporal dynamics of the various SV subpopulations supported a plausible SNARE-dependent kiss-shrink scenario of exocytic events. In the initial kiss step, large semifused vesicles are formed, likely transitioned from tethered vesicles. During the following shrink step, large pore-opened vesicles rapidly transform into small ones. Subsequently, they “pause” at the small semifusion state, as their number continues to increase (Fig. 3G).

SVs predominantly “run away” after the kiss-shrink step

The accumulation of small semifused SVs after AP stimulation raised the question of their subsequent fate. Notably, in synapses under basal conditions (i.e., without network inactivity by CNQX), the size distribution of SVs that lacked direct contact with the plasma membrane (i.e., tethered and free vesicles) in the AZ band exhibited two distinct peaks at ~29 and ~41 nm, corresponding to small and large SVs, respectively (fig. S6D). By contrast, the small SV peak was substantially reduced after CNQX treatment and completely abolished by TeNT treatment, suggesting that the formation of small tethered and small free SVs depends on synaptic activity and is likely a result of kiss-shrink exocytosis (fig. S6, E, G, and H).

More directly, the number of small tethered SVs in the AZ band tightly correlated with the time elapsed after the AP (Fig. 4, A to C, and fig. S6F). The number of small tethered vesicles began to increase ~30 ms post-AP and continued to rise beyond 70 ms, indicating that these vesicles appear after the small semifused SVs (Fig. 4C). These observations suggested a “run-away” pathway after kiss-shrink fusion, thus completing a “kiss-shrink-run” route of exocytosis and endocytosis.

Fig. 4. Vesicle recycling predominantly occurs through the run-away pathway after kiss-shrink exocytosis.

(A) Tomographic slice (left) of a synapse frozen at 70 ms post-AP, showing small tethered SVs (purple arrowheads) and small semifused SVs (red arrowheads) within the AZ band. Shown on the right are zoomed-in views of individual SVs in the AZ band, categorized as small semifused (sSF) and small tethered (sT) SVs. (A1) presents a 3D rendering of the presynaptic profile, including the orthogonal view of the entire AZ band (magenta, presynaptic membrane; red, semifused SVs; purple, small tethered SVs; gray, SVs not in the AZ band). (B) Size distribution of tethered SVs in the AZ bands of synapses frozen at 70 ms post-AP, revealing two peaks (black arrows). The diameter distribution was fitted with a double-Gaussian function. The red and blue lines represent two Gaussian components, and the black dashed line represents their sum. (C) Normalized number of small tethered SVs in the AZ bands of synapses frozen at different time points post-AP. (D) Example tomographic slices of collapsing SVs, including Ω-shaped SVs (pore neck width >22 nm; top) and pits (bottom). (E) Normalized number of collapsing SVs per synapse at different time points post-AP. (F) Diameter distribution of collapsing SVs, fitted with a double-Gaussian function (μ1 = 29.06, σ1 = 5.959; μ2 = 36.57, σ2 = 0.79). The red and green lines represent two Gaussian components, and the blue line represents their sum. (G) Normalized number of different types of SVs in the AZ bands of synapses frozen at different time points post-AP. (H) Schematic diagram depicting the two concurrent routes of vesicle exocytosis in hippocampal synapses: the kiss-shrink-run and the kiss-shrink-collapse. Arrows indicate transitions between vesicle states, with the arrow thickness roughly corresponding to the relative rate of transition. The blue lightning bolts indicate synaptic activations triggered with optogenetic stimulation. Data are presented as mean ± SEM. Statistical significance was determined using the Mann-Whitney test; **p < 0.01, and ***p < 0.001. The number of synapses and vesicles, along with p values for significance comparisons, are provided in table S4.

In addition, we observed vesicles collapsing into the presynaptic membrane, exhibiting Ω-shaped or pitlike profiles, during later stages of AP-triggered exocytosis (Fig. 4, D and E). Most of these collapsing SVs were small, comparable to the small semifused or pore-opened SVs in size (Fig. 4F). Chronologically, most of the collapsing SVs appeared around 70 ms post-AP, after the increase in small pore-opened vesicles, suggesting a “kiss-shrink-collapse” sequence. At all time points after the AP, the number of collapsing SVs remained a small fraction of the total vesicles compared with those following the run-away path (Fig. 4G). Thus, the kiss-shrink-run pathway appears to be the dominant mechanism of SV exocytosis and recycling in hippocampal synapses.

Capturing the entire process of kiss-shrink-run and kiss-shrink-collapse allowed us to correlate morphologically defined large and small tethered, semifused, and pore-opened vesicles with the previously defined docking and priming vesicle states (10, 35, 51) (Fig. 4H). A subpopulation of large tethered vesicles with very short tethers may correspond to the docking state. Following this, the distinct large semifused vesicles can be considered as being in a priming state, ready to open fusion pores. Notably, the small vesicles (including small pore-opened, small semifused, small tethered, and small free SVs), which accounted for about ~20% of the SVs in the AZ band and up to 70% of the SVs directly contacting the plasma membrane, are in the postrelease state and no longer in the docking or priming states.

Run-away vesicle migration and ultrafast endocytosis after 100 ms post-AP

To further investigate the fate of run-away small vesicles in the AZ band (i.e., small tethered and small free SVs), we performed a time-resolved cryo-ET study on synapses frozen at >100 ms post-AP (fig. S8, A and B). At this time point, we observed small vesicles located outside of the AZ band (Fig. 5, A and B). We suspected that these vesicles migrated from the AZ band after the kiss-shrink-run pathway. These recycled SVs were primarily localized at the periphery of the vesicle cluster within the presynaptic bouton (Fig. 5C and fig. S8C), suggesting that run-away vesicles do not integrate directly into the reserve pool of regular-sized vesicles.

Fig. 5. Recycling of shrunken vesicles and presynaptic membrane after 100 ms post-AP.

(A) Tomographic slice of a synapse frozen at 200 ms post-AP, showing one semifused SV at the AZ (red arrowhead) and one large vesicle undergoing ultrafast endocytosis at a peri-AZ site (yellow dashed circle). (A1) and (A2) show zoomed-in views of the boxed areas in (A) taken at different Z heights through the top (A1) and bottom (A2) of the presynaptic bouton. Small vesicles (brown arrows) were enriched at the top and bottom of the vesicle pool. (B) 3D rendering of all vesicles in the entire presynaptic bouton shown in (A). (C) The density of large (left) and small (right) vesicles across three layers of the entire vesicle pool, excluding vesicles in the AZ band. (D) Tomographic slice of a synapse frozen at 200 ms post-AP, showing a large vesicle under ultrafast endocytosis at the edge of the AZ (yellow dashed circle). (D1) presents a 3D rendering of the presynaptic profile, including the orthogonal view of the entire AZ band (magenta, presynaptic membrane; red, semifused SVs; blue, tethered SVs; yellow, collapsing SVs; gray, SVs not in the AZ band). (E) Number of vesicles undergoing ultrafast endocytosis (UFE) and clathrin-mediated endocytosis (CME) in synapses under different conditions. (F) Schematic diagram depicting the organization and dynamic of SVs in the presynaptic bouton during AP-induced synaptic release. In the resting condition, SVs are either tethered to the presynaptic membrane or form large semifused vesicles, ready for release in response to AP firing. Within ~4 ms post-AP, approximately one vesicle per synapse rapidly releases its transmitter through the kiss-shrink pathway. By 70 ms post-AP, most shrunken SVs detach from the presynaptic membrane (run-away), whereas others collapse into the presynaptic membrane. After 100 ms post-AP, the run-away shrunken vesicles further migrate to the periphery of the vesicle pool, and the expanded presynaptic membrane begins recycling through ultrafast endocytosis. Data are shown as mean ± SEM. Statistical significance was determined using the Mann-Whitney test; *p < 0.05, and ns indicates p > 0.05. The number of vesicles, along with p values for significance comparisons, are provided in table S5.

An alternative explanation for the appearance of small vesicles outside the AZ band is that they may result from peri-AZ release. To evaluate this possibility, we analyzed synapses frozen at 0 to 300 ms post-AP and found that exocytic events outside the AZ were fewer than 1 of 10 of those occurring within the AZ (fig. S8, D, E, and G). By contrast, the number of small free vesicles outside the AZ band was found to be five times higher (fig. S8F), implying that most of these vesicles had migrated from the AZ band. Although we cannot entirely exclude minor peri-AZ release, its contribution appears minimal.

We also observed an increase in large endocytic structures (~80 nm in diameter) in the peri-AZ region at 100 to 300 ms post-AP (Fig. 5, A, D, and E), likely corresponding to previously reported ultrafast endocytosis (26). By contrast, clathrin-mediated endocytosis events remained unchanged across the 0- to 300-ms time span (Fig. 5E), likely because clathrin-mediated endocytosis occurs over a much longer timescale (seconds to minutes). These observations suggested that ultrafast endocytosis could compensate for membrane loss resulting from both the kiss-shrink-run and kiss-shrink-collapse modes of vesicle exocytosis (Fig. 5F).

Discussion

By capturing transient intermediate states of SV exocytosis with millisecond precision, we have demonstrated that SVs undergo a rapid kiss-shrink-run process upon AP firing (Figs. 4H and 5F and movie S2). In the sequence, a SV first makes direct contact with the presynaptic membrane (kiss), then rapidly shrinks to approximately one-third of their original volume after forming a narrow fusion pore, releasing neurotransmitters through both diffusion and pressure-driven expulsion (shrink). The SV then typically detaches (run) within tens of milliseconds, enabling hyperfast recycling (Fig. 4H). This kiss-shrink-run pathway accounts for more than 80% of releasing events, whereas a smaller fraction of vesicles undergoes a kiss-shrink-collapse type of exocytosis. Conceivably, the lost vesicular membrane during both kiss-shrink-run and kiss-shrink-collapse events must be retrieved into the presynaptic bouton, likely through subsequent endocytosis over different timescales, including the well-documented ultrafast endocytosis within a few hundred milliseconds post-AP and perhaps slower clathrin-mediated endocytosis (19, 25–28, 52, 53). Thus, our findings reconcile the historically debated kiss-and-run and full-collapse models and provide a unified mechanistic framework for the process of vesicle exocytosis and recycling (Fig. 5F and movie S2).

At the beginning of the kiss-shrink-run process, AP stimulation apparently facilitates the transition of a large tethered vesicle into a metastable, large semifused state, which may correspond to the previously described priming (5, 7, 10). This semifused vesicle is then further triggered by AP-induced Ca²⁺ influx, leading to the opening of the fusion pore that bridges the vesicular and plasma membrane and facilitates two concurrent processes. First, neurotransmitters within the vesicle’s lumen diffuse out through the fusion pore. Second, the vesicle undergoes rapid shrinkage to roughly one-third of its original volume, which further accelerates the diffusion of its contents. Overall, the kiss-shrink process facilitates neurotransmitter release from SVs.

Mechanistically, the driving force for vesicle shrinkage may arise from membrane tension imbalance after fusion-pore formation (29, 54–59), which typically promotes pore dilation and presumably causes the collapse of the vesicle into the plasma membrane. However, our data showed that the fusion pore can be rather stably maintained. This suggested the existence of a constricting barrier surrounding the fusion pore, likely formed by SNARE complexes and/or their regulatory proteins, which mediates the interactions between vesicular and plasma membranes (9, 36, 60, 61). Moreover, the size of a shrunken vesicle also stabilized at about half of its original membrane area, probably reflecting the balance of factors such as membrane curvature, molecular composition, protein interaction, and restricted diffusion.

After the kiss-shrink phase, the run process retrieves the shrunken vesicle with its vesicular proteins within tens of milliseconds. This not only provides a hyperfast route for membrane recycling but also leaves a “clean” site for a new vesicle to dock and fuse. Thus, the kiss-shrink-run mechanism enables the hyperfast turnover of vesicles at the active zone, supporting high rates of synaptic transmission (24). The shrunken SV likely contains a more concentrated set of proteins and possesses a lipid composition that favors higher membrane curvature (55, 62). Consequently, upon eventual detachment from the presynaptic membrane, the shrunken vesicle may be unable to dock and fuse again. Indeed, these recycled vesicles typically did not integrate into the reserve SV pool in the presynaptic bouton. In this way, the kiss-shrink-run process may act as a natural quality control mechanism, preventing neurotransmitter-free vesicles from occupying limited number of docking sites and responding to APs (Fig. 5F). Thus, in addition to improving the efficiency of vesicle recycling and recovery of release sites, the kiss-shrink-run route ensures the fidelity of synaptic transmission.

The uncovering of the kiss-shrink-run pathway relied on our new strategy—time-resolved cryo-ET of intact hippocampal synapses—which offered three key advantages: (i) sufficient temporal resolution of optogenetic stimulation and plunge-freezing, enabling synchronization with synaptic activation; (ii) high spatial resolution in 3D, allowing precise size measurements of individual vesicles and detailed analysis of structural features of their interaction with the plasma membrane, for example, the narrow fusion pores; and (iii) a sufficiently large sample size of more than 1000 intact synapses, ensuring statistically rigorous analyses. By combining these advantages, we were able to quantitatively analyze a series of exocytic events and the hyperfast recycling of small vesicles. More generally, this integrative strategy allows in situ investigation of cellular membrane dynamics and molecular interactions with high spatiotemporal precision.

Is kiss-shrink-run a general mechanism? In theory, it is not impossible that vesicle exocytosis and recycling in various types of synapses may occur along different biophysical or cellular pathways. Thus, whereas cultures of dissociated hippocampal neurons provide valuable assays for studying fundamental biophysical mechanisms such as SV dynamics, our findings should be interpreted with caution. Notably, datasets from previous studies using high-pressure freezing, freeze substitution, and room-temperature ET of acute brain slices or slice culture (10, 63, 64), as well as studies using cryo-ET imaging of isolated synaptosomes undergoing high-concentration K⁺ spray-mixing plunge-freezing (34), suggest that docked or primed vesicles may exist in different sizes. It would be informative to quantitatively analyze whether they also belong to distinct vesicle states. Future studies with more physiological preparations, for example, by combing flash-freezing of brain slices with cryo-ET, will provide insight into the broader physiological relevance of the kiss-shrink-run mechanism and its role in synaptic communication across various neuronal types.

Materials and methods are available in the supplementary materials.

Supplementary Material

science.org/doi/10.1126/science.ads7954

Materials and Methods; Figs. S1 to S8; Tables S1 to S8; References (66–82); MDAR Reproducibility Checklist; Movies S1 and S2

Data and materials availability:

Represented subtomogram average maps of the semifused and pore-opened vesicles have been deposited in the Electron Microscopy Data Bank (EMDB) under accession nos. EMD-65182, EMD-65183, EMD-65184, and EMD-65186. Custom-written codes can be found at Zenodo (65).