The endosomal Q-SNARE, Syntaxin 7, defines a rapidly replenishing synaptic vesicle recycling pool in hippocampal neurons

Upon the arrival of repetitive stimulation at the presynaptic terminals of neurons, replenishment of readily releasable synaptic vesicles (SVs) with vesicles in the recycling pool is important for sustained neurotransmitter release. Kinetics of replenishment and the available pool size define synaptic performance. However, whether all SVs in the recycling pool are recruited for release with equal probability and speed is unknown. Here, based on comprehensive optical imaging of various presynaptic endosomal SNARE proteins in cultured hippocampal neurons, all of which are implicated in organellar membrane fusion in non-neuronal cells, we show that part of the recycling pool bearing the endosomal Q-SNARE, syntaxin 7 (Stx7), is preferentially mobilized for release during high-frequency repetitive stimulation. Recruitment of the SV pool marked with an Stx7-reporter requires actin polymerization, as well as activation of the Ca2+/calmodulin signaling pathway, reminiscent of rapidly replenishing SVs characterized previously in calyx of Held synapses. Furthermore, disruption of Stx7 function by overexpressing its N-terminal domain selectively abolished this pool. Thus, our data indicate that endosomal membrane fusion involving Stx7 forms rapidly replenishing vesicles essential for synaptic responses to high-frequency repetitive stimulation, and also highlight functional diversities of endosomal SNAREs in generating distinct exocytic vesicles in the presynaptic terminals. Yasunori Mori et al. find that a subset of neurotransmitter-bearing synaptic vesicles are marked for release by the endosomal Q-SNARE protein Stx7. They show that Stx7 function is necessary for the rapid replenishment of synaptic vesicles that is needed to sustain synaptic transmission during high-frequency stimulation.

N eurons translate electrical stimuli to chemical signals in the form of neurotransmitters, which are transmitted to target cells by the exocytic fusion of neurotransmitterfilled vesicles termed synaptic vesicles (SVs). Hundreds to tens of thousands of morphologically homogeneous neurotransmitterfilled SVs are clustered at presynaptic terminals, potentially capable of releasing their contents upon the arrival of electrical stimuli 1 . Yet, only a fraction of all SVs is usually available for stimulus-dependent release 2,3 (but see ref. 4 ). During the last few decades, the three-pool model has emerged, in which neurotransmitter-filled synaptic vesicles are classified into three functionally distinct pools: the readily releasable pool (RRP), the recycling pool that replenishes the RRP during sustained activity, and the reserve pool (RP, also referred to as the resting pool) that rarely participates in recycling 2,3 . Despite intensive research, however, the molecular underpinnings of the three pools have remained elusive.
Exocytic fusion of SVs to the plasma membrane is mediated by the formation of so-called SNAP-receptor (SNARE) complexes, consisting of vesicular SNARE synaptobrevin (Syb)/VAMP2 and two target SNAREs Syntaxin 1 (Stx1) and SNAP-25 5 . Considering the extraordinary speed of SV exocytosis in response to action potentials (typically within several ms), the RRP is believed to dock physically to the plasma membrane prior to fusion and the canonical neuronal SNAREs described above are responsible for their docked status 2,3 . Aside from the necessity of the dominant SNAREs for the final step of SV exocytosis, non-canonical SNAREs that mediate fusion reactions between intracellular organelles in non-neuronal cells are not only present but also enriched in SVs purified from rat brains 6 (Supplementary Fig. 1), albeit at lower expression levels than the canonical SNAREs 7 , making them likely candidate protein constituents in functionally distinct subsets of SVs, such as the recycling and reserve pools. Indeed, in addition to synapsins, which are reportedly associated with RP 8 , endosomal SNAREs are also present in distinct vesicle pools, namely in RP or spontaneously releasing vesicles (VAMP7 or vti1a), both of which are somewhat refractory to evoked release 9,10 . Similarly, several endosomal SNAREs including vti1a, Stx6, and Stx12/13, also engage in SV recycling and may supply RRP 11 , and VAMP4 is directed to the asynchronous SV pool 12 . However, whether and/or to what extent these presynaptic endosomal SNAREs are present within the recycling pool and consequently confer unique properties upon it, has not been determined.
Aside from the three functionally distinct SV pools described above, accumulating evidence suggests that even within the same presynaptic terminals, distinct types of SVs coexist. For instance, vesicular glutamate transporters (VGLUTs) and vesicular monoamine transporter 2, which are responsible for uptake of glutamate and biogenic monoamines, respectively, into SVs, are present on different populations of SVs 13,14 . In addition to their distinct exocytosis kinetics at a fixed stimulation frequency, exocytotic kinetics of respective SVs during repetitive stimulation appeared to differ depending on stimulation frequencies 14 , suggesting the existence of various SV recycling pools with distinct release probabilities. Although SV biogenesis mediated by adaptor protein (AP)-3 is a key step in the formation of functionally distinct SV pools 14 , it remains unknown whether differences in their molecular compositions are due to minor SV proteins, such as presynaptic endosomal SNAREs.
To gain deeper insights into the contribution of endosomal SNAREs in SV recycling, we monitored behaviors of presynaptic endosomal SNAREs conjugated C-terminally with pHluorin (also referred to as SEP) 15,16 in a comprehensive manner in response to various stimulation protocols. We found that among endosomal SNAREs, Stx7 is exclusively sorted to a subpopulation of the SV recycling pool that responds preferentially to high-frequency repetitive stimulation. Rapid recruitment of this pool is activated by a Ca 2+ /calmodulin-mediated pathway, and also depends on actin dynamics, resembling a previously reported SV population that rapidly replenishes RRP after synaptic depression 2 . Notably, blockade of Stx7 function by overexpressing its N-terminal domain selectively abolishes this pool. Thus, our results reveal a hitherto unrecognized mechanism by which the rapidly replenishing SV recycling pool that sustains synaptic transmission during high-frequency stimulation (HFS) is generated by endosomal fusion events involving Stx7 during SV recycling.

Results
Endosomal SNARE-SEPs localize in distinct types of presynaptic vesicle compartments. To explore the recycling properties of SVs carrying presynaptic endosomal SNAREs under various stimulation protocols, we utilized a pHluorin reporter (super-ecliptic pHluorin: SEP) fused to the luminal C-terminal portions of individual SNARE proteins (Fig. 1a). These include five Syntaxin (Stx) family members (Stx6, 7, 8, 12/13, and 16), all of which were highly enriched in the pure SV fraction 6 ( Supplementary Fig. 1). We also included two additional endosomal SNAREs, vti1a and VAMP7, which reportedly mediate spontaneous release rather than stimulusdependent evoked release 9,10 . As a control for authentic SV residents, two widely used SEP constructs, SypHy (Synaptophysin (Syp) fused with SEP) (Fig. 1a) and VGLUT1-SEP were used 17,18 . These SEP reporters have been widely used to monitor SV recycling, owing to their pH dependence, by which SEP fluorescence is quenched in the acidic lumen of SVs, and de-quenched upon exocytosis by exposure to the neutral extracellular pH (Fig. 1b−d). When SypHy was lentivirally transduced into cultured hippocampal neurons 19,20 , it was sorted to presynaptic compartments, evidenced by immunostaining with anti-Syb2 antibody (Fig. 1b). When neurons were electrically stimulated repetitively at 10 or 40 Hz (300 or 200 action potentials (APs)), SypHy fluorescence robustly increased in response to the stimuli, and >30−40% of all SypHy molecules, estimated by application of 50 mM NH 4 Cl at the end of the recordings, was engaged in exocytosis by the end of stimulation ( Fig. 1d−f). On the other hand, although presynaptic endosomal SNARE-SEPs were similarly colocalized with Syb2 ( Fig. 1e), the vast majority of endosomal SNARE-SEPs showed relatively smaller responses compared to SypHy (Fig. 1e, f; <25%) and in some cases, responses upon 40 Hz stimulation were significantly smaller than those elicited at 10 Hz (e.g., Stx8-SEP, Stx16-SEP) (Fig. 1e, f). Strikingly though, Stx7-SEP exhibited unique behavior, i.e., stimulation at 10 Hz failed to elicit a clear fluorescence increase, while 40 Hz stimulation caused a robust increase up to~40% (Fig. 1e, f). Quantitative comparisons of all SEPreporters in response to 10 and 40 Hz stimulation (with 200 APs) revealed that Stx7-SEP exhibited an~8-fold fluorescence increase at 40 Hz compared to 10 Hz (Fig. 1e, f). Overall, these results indicate that different presynaptic endosomal SNARE-SEPs localize to distinct membrane compartments that are capable of recycling in an activitydependent manner, to different degrees, at presynaptic terminals.
Only Stx7-SEP exclusively localizes to Syb2-positive SVs. Given distinct recycling behaviors of presynaptic endosomal SNARE-SEPs compared to SypHy, we wondered to what extent these endosomal SNARE-SEPs are sorted into functionally fusioncompetent SVs. To test this, SEP responses were monitored after pretreatment with tetanus toxin (TeNT), which enzymatically cleaves Syb2/VAMP2 21 (also cleaves VAMP1 or VAMP3/ Cellubrevin) 22,23 , thereby terminating SV exocytosis 24 . The robust response of Stx7-SEP at 40 Hz was completely blocked by TeNT treatment, as was observed for SypHy and VGLUT1-SEP, indicating that Stx7-SEP was sorted into Syb2/VAMP2-positive, functionally competent SVs ( Fig. 1g; Supplementary Fig. 2). By contrast, responses of other endosomal SNARE-SEPs were not completely abolished by TeNT treatment, but were only diminished to varying extents ( Fig. 1g and Supplementary Fig. 2). We also examined luminal pH of intracellular compartments bearing SEP-reporters by sequential application of an acidic solution (pH 5.5) and a 50 mM NH 4 Cl solution 25 , and subsequent electrical stimulation to ensure active synapses ( Fig. 1h and Supplementary  Fig. 3). These analyses revealed that the vesicular pH of Stx7laden vesicles (mean ± standard error of the mean (s.e.m.), 6.09 ± 0.05) was not significantly different from those of SypHy (mean ± s.e.m., 5.96 ± 0.04) and VGLUT1-SEP (mean ± s.e.m., 5.98 ± 0.05), whereas those of other endosomal SNARE-SEPladen vesicles were significantly higher (>6.40) (Fig. 1h). We also examined additional putative SNARE pairs of Stx7 suggested in non-neuronal cells, including VAMP8 and vti1b 26,27 . When VAMP8-SEP and vti1b-SEP were lentivirally transduced into neurons, they rarely co-localized with Syb2/VAMP2-positive puncta, and only minor portions of punctate structures labeled with these SNAREs responded to repetitive stimulation at 40 Fig. 4). Thus, these results collectively demonstrate that, among the presynaptic endosomal SNAREs examined here, Stx7 is peculiar in that it exclusively localizes to a subpopulation of the recycling pool containing genuine, fusion-competent SVs that respond preferentially to HFS.
Exocytosis of Stx7-SEP vesicles requires HFS. One of the most striking features revealed by our comprehensive SEP imaging analysis (Fig. 1e, f) was that Stx7-SEP responded preferentially to HFS at 40 Hz, but rarely responded to 10 Hz stimulation. To confirm this phenomenon and to avoid any bias possibly due to functional heterogeneity intrinsic to each bouton or culture preparation, we continuously monitored changes in SEP fluorescence at individual boutons with various stimulation frequencies ranging from 5 to 40 Hz at 5 min intervals (Fig. 2a). Unlike SypHy, which reliably exhibited a robust exocytotic fluorescence increase irrespective of stimulation frequency (Fig. 2a, top), Stx7-SEP hardly responded during 5 Hz or 10 Hz stimulation, while it showed robust fluorescence increases at 20 and 40 Hz in the same boutons (Fig. 2a, bottom). We ruled out the possibility that stimulus-dependent increases in Stx7-SEP fluorescence resulted from vesicle fusion to neutral intracellular compartments during HFS, since the application of an acidic solution (pH 5.5) right after cessation of stimulation largely quenched the fluorescence Stx7-SEP ( Supplementary Fig. 5), as was observed in the case of SypHy 19 . We then wondered if overexpression of Stx7-SEP simply attenuated SV exocytosis in general, for instance by inactivating Ca 2+ channels, or if Stx7-SEP localized at non-synaptic areas where non-SV-type secretory vesicles that responded only to HFS were preferentially recorded. To explore these possibilities, we coexpressed Stx7-SEP and Syp-mOr, in which a pH-sensitive orange fluorescent protein, mOrange2, was fused to the luminal region of synaptophysin, instead of SEP 19,20 , and restricted the analysis to Syp-mOr responding areas ( Supplementary Fig. 6). Yet, Stx7-SEP rarely responded to 10 Hz stimulation, but exhibited a drastic increase in response to 40 Hz stimulation with this analysis (Supplementary Fig. 6).
We next wondered if more prolonged stimulation at low frequencies (5 and 10 Hz) could induce exocytosis of Stx7-SEPladen vesicles. Since endocytosis and subsequent re-acidification of SVs during prolonged stimulation might mask the SEP fluorescence increase due to SV exocytosis 28 , we blocked re-acidification of endocytosed vesicles with 2 µM bafilomycin A1 (Baf), a potent inhibitor of vacuolar-type H + -ATPases (V-ATPases) (Fig. 2b). We adopted 600 action potentials (APs) because that was sufficient to induce exocytosis of the entire recycling pool in cultured hippocampal neurons 29 . In the case of SypHy, fluorescence signals plateaued at the same levels (about 40−50%) irrespective of stimulation frequencies (5−40 Hz) (Fig. 2c). In stark contrast, Stx7-SEP exhibited similar exocytic properties as SypHy at 20 and 40 Hz, whereas it showed much lower and slower responses during 5 and 10 Hz stimulation (Fig. 2c). Comparisons between SypHy and Stx7-SEP by replotting as a function of stimulus numbers clearly showed that the rise kinetics of Stx7-SEP were identical to those of SypHy during 20 and 40 Hz stimulation, whereas those of Stx7-SEP were much slower than SypHy at 5 and 10 Hz stimulation (Fig. 2d).
The reluctant, and incomplete recruitment of Stx7-SEP vesicles during prolonged stimulation at low frequencies (5 or 10 Hz) allows us to examine whether Stx7-SEP vesicles represent a subgroup of the total recycling pool labeled with SypHy, or whether they are separate groups, although TeNT treatment clearly abolished Stx7-SEP responses within a restricted time frame (Fig. 1g). To this end, neurons expressing either SypHy or Stx7-SEP were subjected to low-frequency stimulation at 5 Hz, for 500 APs, in order to deplete the total recycling pool of SypHy, and then subsequently subjected to HFS at 40 Hz, 600 APs ( Supplementary Fig. 7). The second stimulation at 40 Hz produced a scant increase in SypHy fluorescence, whereas the same stimulation produced a drastic increase in Stx7-SEP, indicating that Stx7-SEP vesicles represent a negligible portion of the total recycling pool. However, pretreatment of neurons with TeNT did not completely abolish the Stx7-SEP response to the second stimulation at 40 Hz, incompatible with results in the absence of prolonged pre-stimulation at 5 Hz ( Fig. 1g and Supplementary  Fig. 2), strongly indicating that Stx7-SEP vesicles recruited for the release under this condition were not typical SVs. These unexpected observations can be explained if a substantial shift of Stx7-SEP from the SV pool to the non-SV pool occurs, perhaps mediated by fusion events via Stx7 and other endosomal SNAREs in non-SV compartments during the prolonged stimulation at 5 Hz. The unsolved question of whether Stx7-SEP vesicles comprise part of the total recycling pool will be addressed by other approaches, as described below (see Figs. 6 and 7).
Stx7 localizes to a subset of SVs with low abundance at the presynaptic terminals. In order to gain further evidence for the localization of Stx7 in a subpopulation of SVs, we adopted morphological and biochemical approaches. First, we asked whether endogenous Stx7 localizes to a specific part of the SV cluster. To this end, we co-stained cultured hippocampal neurons with antibodies against Stx7 and Syp, the latter of which should illuminate entire SV clusters. Although Stx7 immunoreactivity was apparent in cell somas and dendrites, it was also observed along axons ( Supplementary Fig. 8a). Images at higher magnification revealed that Stx7 fluorescence partially overlapped with Syp signals (Fig. 3a). We then performed triple staining, including an active zone (AZ) marker bassoon (BSN). Whereas BSN signals often appeared as small, compact structures located at the center of Syp-positive puncta, Stx7 immunoreactivity only partially overlapped with BSN signals and located surrounding BSN signals ( Fig. 3b and Supplementary Fig. 8b), indicating that intrinsic Stx7 localizes at the distal side of the SV cluster from the AZs. To substantiate these observations, we next performed immunoelectron microscopy. Since our initial attempts to detect endogenous Stx7 with the same antibody did not produce reliable signals, we expressed Stx7-SEP as performed for SEP imaging, and proceeded to immunostaining using anti-GFP antibody. In comparison, we adopted cultured cells transfected with SypHy, which could be detected with the same antibody. Consistent with immunofluorescence data, immunoparticles for SypHy were widely spread all over SV clusters in presynaptic structures, whereas Stx7-SEP immunoparticles were sparsely distributed within SV clusters ( Fig. 3c and Supplementary Fig. 9). Quantification of densities of immunoparticles revealed that SypHy was expressed at a significantly higher level than Stx7-SEP (approximately 6-fold; mean ± s.e.m., 294 ± 33 particles/µm 2 for SypHy vs. 47 ± 5 particles/µm 2 for Stx7-SEP), while the numbers of both immunoparticles showed a positive correlation with areas of synaptic varicosity (Fig. 3d, e and Supplementary Fig. 9). Notably, distances of Stx7-SEP immunoparticles to the nearest AZ membranes (defined by electron-dense postsynaptic density (PSD) structures) were significantly longer than those of SypHy, which was evident from a right shift of a cumulative plot for Stx7-SEP compared to that for SypHy (Fig. 3f). These observations were fully compatible with our immunofluorescence data, as well as with a previous observation under STED microscopy 7 . We next attempted to determine the copy number of Stx7 in an SV. Previously, Stx7 content in the synaptosomal fraction isolated from rat brains was reported to be 78.6 copies per synapse 7 . However, the synaptosomal fraction also contains postsynaptic compartments in which Stx7 might be present ( Supplementary  Fig. 8a). To estimate a fractional contribution of presynaptic Stx7 to the synaptosomal fraction, we quantified the Stx7 content in purified SVs in which Stx7 was shown by western blot analysis to be highly enriched 6 . We estimated the copy number of Stx7 molecules per SV to be 0.14 ± 0.03 (mean ± standard deviation) (Fig. 3g, h and Supplementary Figs. 10, 11). Assuming that an average synaptosome contains~380 SVs 7 , the total number of Stx7 molecules located on SVs is calculated to be~53, which roughly coincides with the number of Stx7 molecules in the synaptosomal fraction 7 , suggesting that more Stx7 molecules are present in presynaptic terminals than in postsynaptic compartments.
Recruitment of Stx7-vesicles for exocytosis is triggered by the Ca 2+ /calmodulin pathway and requires actin polymerization. The fact that Stx7-SEP responded preferentially to HFS (>20 Hz) suggested that exocytosis of Stx7-SEP-bearing SVs would require a high concentration of Ca 2+ . To test this hypothesis, we raised external Ca 2+ from 2 to 8 mM, and examined whether even 10 Hz stimulation would cause robust responses. The response of SypHy was facilitated in the presence of 8 mM Ca 2+ (Fig. 4a, left), as observed previously 30 . Notably, Stx7-SEP also exhibited a robust exocytic response to 10 Hz stimulation when external Ca 2+ was raised to 8 mM (Fig. 4a, right). To further elucidate signaling pathways downstream of Ca 2+ , we focused on calmodulin, since it is a Ca 2+ -sensor protein that mediates fast SV replenishment after RRP depletion at the calyx of Held synapse and at hippocampal synapses 31,32 . To this end, we co-expressed the SEP reporters and a calmodulin inhibitory peptide (CIP), and monitored fluorescence changes in the presence of Baf for prolonged periods (either at 10 Hz for 60 s or at 20 Hz for 30 s). Consistent with previous reports, CIP expression slowed SV recruitment, measured with SypHy, during 20 Hz stimulation (Fig. 4b), whereas no changes were observed during 10 Hz stimulation (Fig. 4b), indicating that the activation of the Ca 2+ / calmodulin pathway promotes rapid SV replenishment, preferentially during intensive stimulation. No changes in total pool sizes were observed at either stimulation frequency (Fig. 4b).
GST-Stx7-N (fmol) 15  Notably, recruitment of Stx7-SEP vesicles for exocytosis was also significantly slowed by CIP at 20 Hz stimulation (Fig. 4c). These results indicate that a calmodulin-dependent SV replenishment process preferentially operates at HSF (>20 Hz) in cultured hippocampal neurons, and that Stx7-SEP-laden vesicles may well constitute (at least a part of) such an SV pool.
We next focused on the actin cytoskeleton, because Ca 2+ / calmodulin-dependent fast SV replenishment after RRP depletion reportedly requires actin polymerization in calyx of Held synapses 33,34 . In accordance with a previous report 9 (Fig. 5a). However, Lat-A significantly retarded the SypHy response upon 40 Hz stimulation (Fig. 5a, upper right), indicating that there exists a subset of SVs for which mobilization for release preferentially during the intense stimulation is actin-dependent. This result is consistent with previous observations obtained using neurons in which actin isoforms were conditionally delepted 33 . Intriguingly, more pronounced reduction by Lat-A was observed for Stx7-SEP responses upon 40 Hz stimulation (Fig. 5a), indicating that Stx7-SEP localizes to a subpopulation of recycling SVs for which recruitment for exocytosis is actindependent. Lat-B, another actin polymerization inhibitor, also decreased the SypHy and Stx7-SEP responses to a similar extent at 40 Hz, but not at 10 Hz ( Supplementary Fig. 12). Finally, when Stx7-SEP and SypHy were monitored during 10 Hz stimulation in the presence of 8 mM Ca 2+ , Lat-A significantly reduced the responses of both (Fig. 5b).
To further determine whether Lat-A treatment reduced the size of the total recycling pool or whether it simply slowed SV recruitment for release, we monitored SypHy fluorescence in the presence of Baf (Fig. 5c). Whereas release kinetics (τ exo ) and the total recycling pool size of SypHy during 10-Hz stimulation were not affected by Lat-A, rise kinetics of SypHy during 20 Hz stimulation were significantly slowed by Lat-A with no changes in the total recycling pool size of SypHy (Fig. 5c).
Taken together, these results indicate that a portion of the SV recycling pool is recruited through a mechanism requiring activation of the Ca 2+ /calmodulin pathway and actin polymerization, and that Stx7-SEP preferentially localizes to that pool, resembling a rapidly replenishing pool observed in calyx of Held synapses 31,33,[35][36][37] .
The N-terminal domain of Stx7 is responsible for its characteristic behavior. Like other Syntaxin family members, Stx7 comprises an N-terminal domain (NTD), the SNARE domain, and a transmembrane domain (TMD) 5 (Fig. 6a). To understand the molecular mechanism by which Stx7 is selectively sorted to a subset of the SV recycling pool, we constructed two deletion mutants either lacking the NTD (Stx7-ΔNTD-SEP) or the SNARE motif (Stx7-ΔSNARE-SEP) (Fig. 6a). When these mutants were expressed in cultured neurons, Stx7-ΔNTD-SEP was properly sorted to Syb2-positive bouton-like puncta (Fig. 6b) in which it is preferentially localized to the intracellular acidic compartments, with albeit higher luminal pH compared to genuine SVs (Fig. 6c, d and Supplementary Fig. 12). However, unlike Stx7-SEP, Stx7-ΔNTD-SEP responded to 10 Hz stimulation as it did to 40 Hz stimulation (Fig. 6e) and TeNT treatment did not completely abolish the responses at 10 Hz (Fig. 6f). Furthermore, responses of Stx7-ΔNTD-SEP at 40 Hz were no longer sensitive to Lat-A treatment (Fig. 6g). Thus, the NTD is essential for proper sorting of Stx7 to the actin-dependent SV subpool that is preferentially recruited during HFS. By contrast, Stx7-ΔSNARE-SEP did not accumulate at Syb2-positive bouton-like puncta, and was distributed evenly on cell surfaces (Fig. 6b-d and Supplementary  Fig. 13), suggesting that the SNARE domain is indispensable for its presynaptic, as well as vesicular, localization.
Overexpression of Stx7-NTD selectively reduces SV recruitment during HFS. Although the results described above demonstrate that Stx7-SEP is directed to a subset of the SV recycling pool, whether Stx7 is necessary for this subpopulation of SVs remains unclear, especially given that only a small population of SVs carries Stx7 (Fig. 3). To address this question, we first examined whether silencing of Stx7 with specific shRNAs affects the SV recycling subpool. However, chronic knockdown of Stx7 expression severely reduced SypHy responses at 10 and 40 Hz, indicating that Stx7 is necessary to establish this SV pool per se during synapse development and maturation processes (Supplementary Fig. 14). As an alternative approach to inactivate Stx7, we explored specific dominant-negative effects by overexpressing the N-terminal domain of Stx7. The rationale for this is that the NTD of Stx7 interacts with its own SNARE motif, thereby inhibiting SNARE complex formation with cognate SNAREs 37 . In addition, the results above (Fig. 6) clearly demonstrate that the NTD of Stx7 is essential for its proper sorting to the actindependent SV recycling pool. To this end, we placed a P2A selfcleaving peptide between the SypHy and Stx7-NTD sequences (a.a. 13-149) to ensure co-expression of both proteins 20 , and monitored SV recycling reported by SypHy (Fig. 7a). Expression of Stx7-NTD did not alter the total recycling pool size or the kinetics of vesicle release elicited at 10 Hz (Fig. 7b). In stark contrast, expression of the Stx7-NTD significantly slowed the increase of SypHy fluorescence during 20 Hz stimulation, while total recycling pool size was affected only slightly (Fig. 7b). Treatment with Lat-A in addition to Stx7-NTD overexpression did not further slow release kinetics (Fig. 7c), indicating direct involvement of Stx7 in the actin-dependent SV recycling pool that rapidly and preferentially replenishes RRP upon HFS.

Discussion
Departing from comprehensive analyses of optical reporters for multiple presynaptic endosomal SNAREs, we report that an endosomal Q-SNARE, called Stx7, specifically targets a unique SV pool that replenishes RRP only during intense activity. Recruitment of this SV pool, enabled by Stx7, is activated by the Ca 2+ / calmodulin pathway and requires actin polymerization, characteristics resembling a population of the recycling SV pool that rapidly replenishes RRP in calyx of Held synapses 33,[36][37][38] . Our results collectively indicate that endosomal fusion supported by Stx7, presumably together with its cognate SNAREs, is responsible for generating a functionally distinct recycling pool of SVs.
Recycling vesicle pool heterogeneity conferred by endosomal SNAREs. Previous studies using SEP constructs have suggested that release properties of a subpopulation of SVs are dependent upon endosomal SNAREs 9-12 . For instance, the Qc-SNARE, vti1a, and the R-SNARE, VAMP7, reside mainly on the reserve pool that is reluctantly mobilized during activity. Instead, SVs bearing these SNAREs are prone to fuse spontaneously in the absence of activity 9,10 . Likewise, other endosomal SNAREs, Stx6, Stx12/13, and VAMP4, are capable of recycling, albeit to a lesser extent than authentic SV proteins, such as Syb2 11,12 . However, our comprehensive SEP imaging analyses to examine effects of TeNT on stimulus-dependent fluorescence changes, as well as vesicular pHs of vesicle compartments on which they reside, reveal that presynaptic endosomal SNAREs can be categorized into two distinct classes; i.e., Stx7 and other endosomal SNAREs. First, Stx7-SEP resides on a fraction of the recycling vesicle pool that is preferentially mobilized during HFS (e.g., >20 Hz). Mobilization of this pool depends on an increase in cytosolic [Ca 2+ ] exceeding a certain threshold to activate calmodulin, and also requires actin dynamics (Figs. 4 and 5). Since Stx7-laden vesicles share some characteristic features of genuine SVs, e.g., the same vesicular pH as SypHy-or VGLUT1-carrying vesicles, and complete silencing of activity-dependent exocytosis by TeNT treatment, we conclude that Stx7 resides on a subpopulation of the recycling SV pool.
In great contrast to Stx7, other presynaptic endosomal SNAREs exhibit different characteristics and behaviors compared to genuine SV residents. While they responded to a wide range of stimulation, their exocytotic responses were only partially attenuated by TeNT treatment, indicating that these endosomal SNAREs are present, albeit to various extents, on Syb2/VGLUT1negative vesicle compartments. This is further supported by the fact that the average vesicular pH of vesicles containing these endosomal SNAREs appeared to be significantly higher (~6.4) than the pH of authentic SV residents. Thus, although it was proposed that Stx6, Stx12/13, and vti1a are involved in the replenishment of RRPs 11 , they are only minimally present in the RRP and recycling SV pool. Specifically, they are present in non-SV compartments, which nevertheless recycle in an activitydependent manner. Additionally, although a previous study proposed that vti1a selectively resides on Syb2-free vesicles, which are preferentially utilized for spontaneous release rather than stimulus-dependent evoked release 10 , our results reveal that the vast majority of presynaptic endosomal SNAREs are present on Syb2-free vesicles. The identity and function of such non-SV recycling vesicles at presynaptic terminals conferred by these endosomal SNAREs are enigmatic, but they may utilize noncanonical neuronal SNARE proteins and a Ca 2+ sensor for exocytotic fusion with the plasma membrane, such as SNAP-29 or -47, Stx3 or 4, and Synaptotagmin-7 32,38,39 . Notably, recent studies have proposed the existence of non-SV-type secretory organelles at synaptic sites or in close proximity to presynaptic active zones, which secrete neuropeptides from lysosomes or dense-core vesicles 38,40,41 . Unraveling SNAREs or Ca 2+ sensors involved in these non-SV vesicles will enable us to understand the physiological significance of these non-SV secretory organelles for synaptic performance and signaling.

Stx7
Stx7-ΔN Actin and SV mobilization. The role of the actin cytoskeleton in regulating synaptic vesicle mobilization has long been controversial. While Lat-A, an actin polymerization inhibitor, had little impact on SV mobility or recycling in hippocampal neurons 9,42 , similar treatment resulted in a deceleration of SV replenishment of the RRP in calyx of Held synapses [35][36][37] , indicating the presence of actin-dependent, rapidly replenishing SVs. Here, using SypHy imaging, we provide evidence compatible for both views regarding the role of actin dynamics in SV exocytosis and recruitment. Specifically, SVs are recruited for release in an actin-independent manner during mild stimulation, while the exocytotic release of a significant fraction of SVs is accelerated by an actin-dependent process during high-frequency stimulation (Fig. 5). The former is consistent with the previous observations 9,42 and the latter is compatible with reduced SV replenishment into the RRP proposed in calyx of Held synapses, reportedly mediated by Ca 2+ /Calmodulin-dependent processes 31 . Interestingly, SVs carrying Stx7-SEP are preferentially utilized for release during intense activity and are more dependent on actin dynamics than the remainder of the recycling pool revealed by SypHy. Thus, our results collectively indicate that SVs characterized by Stx7 may represent a fraction of the recycling SV pool rapid mobilization of which for release requires actin polymerization, presumably mediated by the Ca 2+ /calmodulin pathway, to replenish the RRP during high activity.
How does Stx7 define rapidly replenishing SVs? How does Stx7 contribute to rapidly replenishing SVs? In non-neuronal cells, Stx7 is required for homotypic/heteromeric fusion of late-endosomes and lysosomes 26,27,43 or lysosomes and autophagosomes 44 , and these processes may contribute to protein sorting and organelle maturation 45 . In accordance with the role of Stx7 in non-neuronal cells, we propose that Stx7 may contribute to the formation of the rapidly replenishing SV subpool by mediating homotypic, as well as heterotypic fusion with existing endosomal structures. Recent evidence suggests that the generation of clathrin-independent, large, endosome-like vacuoles or bulk endocytosis is the predominant form of SV endocytosis, especially after repetitive stimulation [46][47][48] . Although current models predict clathrin-mediated SV reformation directly from these endosomal membranes, it is also evident that homotypic or heterotypic fusion of endosomes mediated by endosomal SNAREs can occur before SV reformation, even with multiple rounds during high activity 49,50 . Although a molecular link between Stx7 and actin is enigmatic, Stx7 may help generate the rapidly replenishing SV pool, as a result of endosomal fusion and molecular sorting. This model seems to work if Stx7, but not other cognate SNAREs, are destined to SVs, and if exogenously expressed Stx7-SEP follows the trafficking route of endogenous Stx7.
Some other scenarios may explain the contribution of Stx7 to the observed results. First, it is possible that Stx7 directly or indirectly associates with actin filaments and that rapid recruitment of Stx7-vesicles is only possible when the Ca 2+ /calmodulin cascade is activated. Alternatively, Stx7 may contribute to compound membrane fusion with its cognate SNAREs, involving the fusion of multiple vesicles prior to a final fusion 51 . However, considering the limited number of copies of Stx7 in a subpopulation of SVs (only 14% of SVs can harbor a single copy of Stx7) and the relatively high impact of Stx7-NTD overexpression on the SypHy response, as well as the highly specific behavior of Stx7 among presynaptic endosomal SNAREs, these are not likely possibilities. Further studies will be necessary to determine how Stx7, presumably with its cognate SNAREs, functionally differentiates this SV pool, thereby affecting activitydependent synaptic performance. Fig. 7 Overexpression of Stx7-NTD slowed SypHy responses only during HFS. a Schematic diagram of SypHy and SypHy co-expressed with Stx7-NTD. Stx7-NTD was placed after a P2A sequence so that all SypHy-positive cells co-expressed Stx7-NTD. b The total recycling pool monitored by SypHy responses in the absence (black) and presence (red) of Stx7-NTD at 10 Hz, 600APs (left) and 20 Hz, 600APs (right) after bafilomycin treatment. Bottom box-whisker plots show quantification of total recycling pool sizes (left) and time constants of rise time (right, τ exo ). c Pretreatment with Lat-A did not further reduce release kinetics upon Stx7-NTD overexpression. Responses of SypHy with Stx7-NTD in the presence of Lat-A (blue) were compared with control (SypHy only, black) and SypHy with Stx7-NTD (red) in response to at 20 Hz, 600APs. A box-whisker plot shows time constants of the rise time (τ exo ). SypHy responses with Stx7-NTD (red) and Stx7-NTD + Lat-A (blue) did not differ significantly (p > 0.05). d Schematic summary of this study, depicting that Stx7 is preferentially present in the rapidly replenishing SV pool during intense, repetitive stimulation.

Methods
Molecular biology. SypHy, Syp-mOr, VGLUT1-SEP, and endosomal SNARE-SEPs were expressed in a neuron-specific manner using lentivirus-based vectors, in combination with the Tet-Off system 19,20 . Two vectors were used, a 'regulator' vector expressing an advanced tetracycline transactivator (tTAad) under the control of the human synapsin 1 promoter (pLenti6PW-STB), and a 'response' vector that expressed SypHy, Syp-mOr, VGLUT1-SEP or endosomal SNARE-SEPs under the control of a modified tetracycline-response element (TRE) composite promoter (pLenti6PW-TRE). To construct endosomal SNARE-SEPs and deletion mutants, mouse coding sequences for Stx6 (accession no. NM_021433. 3 .1-167, 228-261) without stop codons were amplified by PCR using mouse adult brain complementary DNA as a template, and fused with SEP. Thereafter, they were cloned into a SmaI site located downstream of the TRE sequence in pLenti6PW-TRE using an In-Fusion Cloning Kit (Clontech) 19,20 .
VGLUT1-SEP was constructed as described previously 18 and cloned into the SmaI site of pLenti6PW-TRE. SypHy and Syp-mOr plasmids were generated as described previously 19,20 .
Neuronal culture and viral expression. Primary hippocampal cultures were prepared from embryonic day 16 ICR mice, as described previously 19,20 , with slight modifications. Briefly, hippocampi were dissected and incubated with papain (90 units/mL, Worthington) for 20 min at 37˚C. After digestion, hippocampal cells were plated onto poly-D-lysine-coated coverslips framed either in 24 well plates, 12 well plates (Falcon), or a Nunc four-well dish (Thermo Fisher) at a cell density of 20,000−30,000 cells/cm 2 , and kept in a 5% CO 2 humidified incubator. At 2−4 days in vitro (DIV), 40 μM FUDR (Sigma) and 100 μM uridine (Sigma) were added to inhibit the growth of glial cells. One-fifth of the culture medium was routinely replaced with fresh medium every 2-4 days until recordings. Cultures were transduced with two lentiviral vectors, the regulator vector and response vectors that encode SEP reporters or shRNAs for Stx7 at between 2 to 7 DIV, and were subjected to experiments at 12−14 DIV. To measure the kinetics of exocytosis and total recycling pool size, neurons were treated with a standard extracellular solution containing 2 µM bafilomycin A1 (Baf) (BioViotica) for 1 min, and placed in the imaging chamber. Baf-containing solution was continuously applied to neurons throughout the recordings. For HFS, we adopted a stimulation frequency of 20 Hz instead of 40 Hz, since prolonged stimulation at 40 Hz often causes unavoidable air bubbles around stimulation electrodes, which hindered reliable imaging. For TeNT experiments, neurons were incubated with 10 nM recombinant TeNT (Sigma-Aldrich) for 16 h in a CO 2 incubator. For Lat-A and Lat-B experiments, neurons were treated with either 5 µM Lat-A (Wako) or 10 µM Lat-B (Cayman Chemical) for 1 min at RT, and subjected to imaging experiments. Lat-Aor Lat-B-containing buffer was continuously supplied to neurons throughout the recordings.
To estimate luminal pH and the surface fraction of SEP probes, a low pH solution containing (in mM): 140 NaCl, 2.4 KCl, 10 MES, 4 MgCl 2 (pH 5.5), and 50 mM NH 4 Cl solution described above were successively applied to cultured neurons as described previously 19,20 . To restrict analyses at active synaptic boutons, electrical stimulation at 40 Hz was applied for 10 s at the end of recordings.
Fluorescence images (1024 × 1024 pixels) were acquired with an ORCA-Flash 4.0 sCMOS camera (Hamamatsu Photonics) in time-lapse mode either at 1 Hz (for imaging in response to stimulation) or 5.7 Hz (for estimating vesicular pH and surface fraction) under control of MetaMorph software (Molecular Devices). Fluorescence of SypHy, VGLUT1-SEP, or endosomal SNARE-SEPs was imaged with 470/22 nm excitation and 514/30 nm emission filters, and Syp-mOr was imaged with 546-to 556 nm excitation and 575-to 625 nm emission filters.
Image data analysis. Acquired fluorescence images were analyzed using Meta-Morph software. Active synapses were identified manually by changes in live measurements of fluorescence. Circular regions of interest (ROIs; 2.26 µm diameter and 4 µm 2 area) were positioned manually at the center of the highlighted fluorescence spots. Non-responsive puncta, especially for presynaptic endosomal SNARE-SEPs, which became apparent only upon NH 4 Cl application, were excluded from analysis, since they may represent endosomal SNARE-SEPs expressed outside of presynaptic terminals. To extract fluorescence changes associated with the stimulation at each bouton, an average of five consecutive fluorescence values was taken as F 0 , and this value was subtracted from subsequent fluorescence values (ΔF). To estimate relative fluorescence over total SEP molecules, the peak value during NH 4 Cl application was taken as ΔF NH4Cl (F NH4Cl − F 0 ) and used to normalize the data (ΔF/ΔF NH4Cl ). For each construct, fluorescence changes of at least 50 boutons were analyzed. To calculate rise kinetics of the fluorescence increase (τ exo ), an average trace from one image (containing 5-10 active boutons) was collected, and taken as n = 1. For each averaged trace, τ exo was calculated by fitting the trace with a single exponential function using a Solver function from Excel software. Time constants obtained from at least ten images were statistically evaluated.
Acquired images for vesicular pH and surface fraction measurements were also processed as described above. Vesicular pH and surface fraction were calculated as described previously 19,20 , except that fluorescence during NH 4 Cl application was approximated as fluorescence at pH 7.4. Values of pK a (7.1) and a Hill co-efficient (1.0) for SEP reported previously 16,25 were used for calculation.
Immunoelectron microscopy. Hippocampal neurons grown on coverslips were transduced with lentiviruses carrying either SypHy or Stx7-SEP, and were fixed for 2 h at RT with 4% paraformaldehyde and 0.05% glutaraldehyde in 0.1M phosphate buffer (PB) (pH7.4) at 16 DIV. These neurons were processed for subcellular localization of SypHy or Stx7-SEP with an anti-GFP antiserum by employing preembedding immunoelectron microscopy. Briefly, plasma membranes of fixed neurons were permeabilized with 0.1% Triton X-100 in phosphate-buffered saline (PBS) for 15 min followed by blocking with 10% fetal bovine serum in PBS for 30 min and then incubated with rabbit anti-GFP antiserum (1:1,000 dilution in PBS containing 10% fetal bovine serum) for 24 h at 4°C. After several washes with PBS, they were incubated with 1.4 nm immunogold-conjugated anti-rabbit IgG nanoprobes (1:100 dilution in 2% normal goat serum in PBS) overnight at 4°C and then subjected to post-fixation with 1% glutaraldehyde in PBS for 10 min. Immunogold particles were enhanced for 7 min using the HQ silver enhancement kit (Nanoprobe) according to the manufacturer's instructions. After post-fixation with 1% osmium tetroxide in PB for 40 min on ice, these samples were stained with 1% uranyl acetate solution for 35 min, dehydrated with an ethanol series and propylene oxide for 10 min each, and then embedded in Durcupan resin. After polymerization of the resin at 60°C for 48 h, coverslips were removed and serial ultrathin sections (70 nm thick) were prepared. These sections were observed under a transmission electron microscope at 100 kV (H-7650, Hitachi Co., Japan) and serial images of individual presynaptic varicosities, which were identified by thickenings of axons with SV clusters and by postsynaptic counterparts with electron-dense thickenings of the plasma membrane (postsynaptic density: PSD) with a constant gap between the two profiles, were taken with a CCD camera (Quemessa; Olympus-SIS, Germany). In quantification of immunolabeling, one image, in which PSD was clearly seen, from serial images of individual presynaptic varicosities was used and only immunoparticles located within the presynaptic profiles were considered as having originated from SEPs expressed in SVs or early endosomes to exclude signals from SEPs expressed on the plasma membrane. The area of presynaptic varicosities, the number of immunoparticles within the individual varicosities were measured. In order to measure the nearest distance to the AZ membrane for individual immunoparticles, the AZ membrane was defined as a portion of the presynaptic plasma membrane in apposition to the PSD, and vertical lines were manually drawn from the edges of the AZ and then only immunoparticles located within the enclosed areas were used for measurements ( Supplementary Fig. 9). All of the above measurements were carried out using iTEM software (Olympus-SIS).
To estimate the copy number of Stx7 per SV, the CPG-purified SV fraction (a gift from Reinhard Jahn) with known protein concentration and vesicle number, was subjected to western blotting together with various amounts of standard proteins for Stx7 and Syb2. Band intensities in acquired images were quantified using Quantity One software (BioRad).
Animals. Pregnant ICR mice were purchased from SLC, Japan. All mice were given food and water ad libitum. Animals were kept in a local animal facility with a 12 h light/12 h dark cycle. Ambient temperature was maintained around at 21°C with a relative humidity of 50%. All animal experiments were approved by the Institutional Animal Care and Use Committee of Doshisha University.
Statistics. For imaging experiments, unpaired t-tests were applied to compare the means of two experimental groups. All statistical tests were two-tailed, and the level of statistical significance is indicated by asterisks: *p < 0.05, **p < 0.01, ***p < 0.001. n.s., not significant. To evaluate the significance of distances from immunoparticles to the nearest AZ membrane, a Kolmogorov−Smirnov test was performed.