SUMOylation of synapsin Ia maintains synaptic vesicle availability and is reduced in an autism mutation

Synapsins are key components of the presynaptic neurotransmitter release machinery. Their main role is to cluster synaptic vesicles (SVs) to each other and anchor them to the actin cytoskeleton to establish the reserve vesicle pool, and then release them in response to appropriate membrane depolarization. Here we demonstrate that SUMOylation of synapsin Ia (SynIa) at K687 is necessary for SynIa function. Replacement of endogenous SynIa with a non-SUMOylatable mutant decreases the size of the releasable vesicle pool and impairs stimulated SV exocytosis. SUMOylation enhances SynIa association with SVs to promote the efficient reclustering of SynIa following neuronal stimulation and maintain its presynaptic localization. The A548T mutation in SynIa is strongly associated with autism and epilepsy and we show that it leads to defective SynIa SUMOylation. These results identify SUMOylation as a fundamental regulator of SynIa function and reveal a novel link between reduced SUMOylation of SynIa and neurological disorders.

T he coordinated release of neurotransmitter from synaptic vesicles (SVs) is fundamental to synaptic function and plasticity. It is generally accepted that SVs are organized into distinct pools at the presynapse including the readily releasable pool that is synchronously released immediately following stimulation and the reserve pool (RP), that is mobilized during prolonged stimulations. Vesicles that are refractory to release are often referred to as the resting pool 1 . SynIa maintains the RP of SVs by tethering them to each other and to the presynaptic actin cytoskeleton near to presynaptic release sites. Presynaptic depolarization causes SynIa phosphorylation through the CamKII and PKA pathways, which decreases binding affinity and releases the tethered SVs allowing them to move to the active zone [2][3][4] . On the other hand, phosphorylation by other kinases such as c-Src and cdk5 alters the distribution of SVs between different pools 5,6 .
SUMOylation is the covalent attachment of the small ubiquitin-like modifier (SUMO) to lysine residues on target proteins. In neurons, SUMOylation a key regulator of transcription and extranuclear SUMOylation plays fundamentally important roles in synaptic function [7][8][9][10] and neuroprotective responses to severe stressors 11 . Recently, we reported that SUMOylation of RIM1a is required for normal SV exocytosis, demonstrating that SUMOylation also has a role in presynaptic regulation 12 .
Here we show that SUMOylation of SynIa at K687 enhances binding to SVs and facilitates synapsin-mediated SV clustering and anchoring. Replacement of endogenous SynIa with a non-SUMOylatable K687R mutant in neurons reduces the number of releasable SVs and impairs exocytosis. Furthermore, we demonstrate that an A548T mutation in SynIa, which has been linked to autism spectrum disorder (ASD) and epilepsy, reduces SynIa SUMOylation and mirrors the functional defect of non-SUMOyatable SynIa, suggesting a possible causal link between dysfunctional SynIa SUMOylation and neurological disorders.

Results and Discussion
Synapsin Ia is a SUMO substrate. To validate SynIa as a SUMO-1 substrate, we used a SUMOylation assay system in which we exogenously expressed HA-SynIa, Flag-Ubc9 and either YFP-SUMO or non-conjugatable YFP-SUMO-DGG in N2A cells. A SUMOylated SynIa band was detected in the cells expressing YFP-SUMO but not YFP-SUMO-DGG (Fig. 1a, upper panel). This band was also present in green fluorescent protein (GFP)trap pull downs of YFP-SUMO from the N2A lysate, further confirming it corresponds to SUMOylated SynIa (Fig. 1a, lower panel). We attribute the presence of an equal density unmodified SynIa band to dimerization of SynIa and SynIa-SUMO.
Ubc9 binding to the target protein is a critical step in SUMOylation. Consistent with SynIa being a robust SUMO substrate, endogenous SynIa in cultured cortical neuronal lysates is strongly retained by GST-Ubc9 pull downs (Fig. 1b).
Importantly, we also detected endogenous SUMOylated SynIa in neurons by anti-SUMO co-immunoprecipitation from whole-brain lysate. As expected, anti-SynI antibody immunoblots of the immunoprecipitated protein revealed a B110-kDa band corresponding to SUMOylated SynIa. Critically, this band was removed by treatment with the deSUMOylating enzyme SENP1 (Fig. 1c). Again, we attribute the presence of unmodified SynIa in this co-immunoprecipitation to dimerization.
SynIa is a multidomain protein; therefore, to define the SUMOylation site(s), we first generated and expressed SynIa mutants lacking the E domain (DE) or both D and E domains (DDE) in HEK293T cells and performed GST-Ubc9 pull-down assays (Fig. 1d). Removal of the E domain of SynIa greatly decreased the interaction with GST-Ubc9, and additional deletion of the D domain completely abolished the interaction (Fig. 1d). These data indicate that Ubc9 binds mainly to a region within the E domain, but may also have some interaction within the D domain.
Consistent with this, the E domain contains the SUMO consensus sequence VKAE at residues 686-689. Mutation of the Val686 or Glu689 residues flanking the target lysine to Ala disrupted the SynIa-GST-Ubc9 interaction (Fig. 1e) and mutating Lys687 to Arg completely prevented SynIa SUMOylation (Fig. 1f).
The E domain is involved in dimerization, vesicle clustering, presynaptic targeting, cytoskeleton interaction and vesicle release 13,14 . All synapsin a-isoforms (SynIa, SynIIa and SynIIIa) contain highly conserved E domains 15,16 and SynIIa could also be SUMOylated in N2A cells, even though the consensus sequence is AKAE in this protein ( Supplementary Fig. 1a). As shown in Fig. 1e, V686A inhibited Ubc9 binding to SynIa in HEK293T cells. However, in N2A cells, this mutation could still be SUMOylated ( Supplementary Fig. 1b), indicating that an additional factor present in this neuronal cell line, presumably an E3 ligase, is required to enhance SUMOylation. Sequence alignment shows that K687 is also conserved across a-isoforms of synapsins in vertebrates ( Supplementary Fig. 1c). However, despite its importance, the molecular mechanisms underlying the function of the E domain is not well characterized.
Preventing SynIa SUMOylation causes defective SV exocytosis. The SUMO consensus sequence VKAE is present in SynIa across most vertebrate species consistent with SUMOylation playing an important role in SynIa function throughout evolution. We therefore tested the effect of acutely reducing endogenous SynIa by short hairpin RNA (shRNA) knockdown and replacement with shRNA-insensitive wild-type SynIa or the non-SUMOylatable mutant. Transfection of hippocampal neurons with SynI shRNA decreased in SynI levels to 50% of control levels. Transfection with shRNA-resistant wild-type or K687R SynIa constructs restored SynI to near-endogenous levels ( Supplementary Fig. 2).
We investigated the dynamics of SV exocytosis using the genetically encoded reporter Synaptophysin-pHluorin (SypHy) 17 . Hippocampal neurons transfected with both SypHy and knockdown-rescue constructs were subjected to 600 action potentials at 20 Hz using electrical field stimulation to evoke SV exocytosis of all releasable vesicles (that is, readily releasable pool and RP). These experiments were performed in the presence of the proton pump inhibitor bafilomycin1A to prevent reacidification of endocytosed SVs so that each SV is counted only once on first-time release 18 . The SypHy fluorescence was expressed as the change in fluorescence normalized against the basal signal (DF/F 0 ), presented in Fig. 2a,b. To assess the average size of the SV pool in the presynapse, the mean level of DF/F 0 at the end of the experiment (during 50 mM NH 4 Cl, F max ) was calculated and is presented in Fig. 2c. No significant difference was observed between any conditions, indicating that the knockdown of Syn1 or replacement with a non-SUMOylatable mutant did not change the total SV pool. To selectively assess the size and dynamics of the releasable pool, we normalized the change in fluorescence against F max (DF/F max ). Consistent with previous reports using knockout mice, acute shRNA-mediated knockdown of SynI significantly decreased the total level of SV exocytosis, measured by the average DF/F max after stimulation, indicating a reduction in releasable vesicle pool size [19][20][21] . Importantly, a scrambled shRNA control had no effect ( Supplementary Fig. 3a,b). Replacement of the knockdown with wild-type SynIa returned SV exocytosis to control levels. However, replacement with the SynI-K687R mutant did not rescue the normal release profile with levels of exocytosis indistinguishable from the knockdown neurons (Fig. 2d,e). This result cannot be attributed to a dominant-negative effect of K867R Syn1a, as expression of the K687R mutant in nonknockdown neurons did not cause any effect on SV exocytosis ( Supplementary Fig. 3c,d). These data demonstrate that SynIa SUMOylation is required for normal levels of SV exocytosis.
This decrease in SV exocytosis could be mediated by reduced vesicle fusion or by a decrease in the number of releasable vesicles. We therefore assessed the rate of exocytosis by measuring the time constant of a single exponential growth curve fit to DF/F post-stim . As shown in Fig. 2f,g, there was no difference in the rates of vesicle fusion in neurons in which SynIa was knocked down, or knocked down and replaced with either wild-type SynIa or with the K687R mutant. Therefore, we attribute the decreased exocytosis in neurons expressing the non-SUMOylatable SynIa-K687R to a reduction in the size of the releasable vesicle pool, that is, fewer SVs available for release, rather than a defect in vesicle exocytosis.
We also tested if SV exocytosis is affected by a global decrease in SUMOylation by overexpressing the catalytic domain of SENP. Consistent with our results using non-SUMOylatable SynIa, overexpression of constitutively active SENP induced a decrease in the overall amount of SV exocytosis, whereas overexpression of the protease inactive mutant SENP C603S had no effect ( Supplementary Fig. 4). This decrease was considerably more pronounced than molecular replacement with SynIa-K687R, suggesting that other proteins involved in SV regulation are also SUMO targets. While we cannot definitively exclude the possibility that downstream effects induced by decreased SUMOylation of non-synaptic proteins, for example, transcription factors, could contribute to this effect, our data are consistent with previous studies that identify other presynaptic SUMO substrates 12,22,23 .
SUMOylation of SynIa enhances binding to SVs. It is well known that changes in the binding affinity of SynIa to SVs plays a major role in SV dynamics 3 . Therefore, we investigated whether SUMOylation of Syn1a affects its binding to SVs, using an  with an enriched SV suspension from rat brain homogenate. The SVs were pelleted by ultracentrifugation and subjected to quantitative western blot analysis to determine the amount of bound SynIa-DE in SV fraction. SUMOylated SynIa-DE displayed significantly higher affinity binding to SVs than non-SUMOylated or deSUMOylated SynIa-DE (Fig. 3b,c). These results demonstrate that, opposite to CamKII phosphorylation of SynIa, SUMOylation increases the affinity of SynIa binding to SVs. Thus, these data support the hypothesis that SUMOylation of SynIa promotes SV binding and plays key roles in establishing and maintaining the reserve vesicle pool. The generally accepted model of synapsin structure and function is that the D and E domains interact with SVs by binding to specific vesicle proteins, whereas the A and C domains directly bind to the vesicle phospholipid bilayer 14,26,27 . SUMOylation can alter substrate protein conformation 28 and/or promote protein-protein interactions via SUMO-interacting motifs 29 . Enhanced association of SUMOylated SynIa to SVs may be mediated by (i) SUMOylation-dependent changes in the conformation of SynI, causing an increase in SV affinity and/or by (ii) SUMOylated SynIa binding to a SUMO-interacting motifcontaining SV protein. Moreover, since CamKII phosphorylation promotes the dissociation of SynIa and the release of SVs from the actin cytoskeleton 30 , it is also possible that SUMOylation of SynIa could reduce or prevent CamKII phosphorylation.
SUMOylation is crucial in SynIa targeting and clustering. Because the association of SynIa with SVs is critical for correct localization at the presynaptic terminal 31-33 , we next investigated the effect of SUMOylation on SynIa targeting. We overexpressed the SENP1 catalytic domain fragment in hippocampal neurons to decrease the overall cellular SUMOylation level. Endogenous SynI was then stained together with the presynaptic marker Bassoon. SynI puncta that co-localized with Bassoon were counted and normalized to the total number of SynI puncta 34 . When compared with the overexpression of a catalytically inactive mutant (C603S) of SENP1, there were significantly more SynI puncta that did not co-localize with Bassoon, suggesting that SUMOylation influences the targeting activity of SynI (Fig. 4a,b).
To ensure that the mistargetting effect was specifically due to SynIa SUMOylation, endogenous SynIa was knocked down in rat hippocampal neurons and replaced with either GFP-tagged WT or K687R SynIa, then fixed and the co-localization with bassoon and the SV marker VAMP2 assessed. As shown in Fig. 4c,d, while both WT and K687R SynIa showed punctate distributions, there was significantly lower co-localization of the K687R mutant with Bassoon or VAMP2 compared with WT SynIa. There was no difference in the intensity of VAMP2 staining caused by Syn1a K687R expression ( Supplementary Fig. 5a), indicating that SVs themselves are still correctly localized at the active zone when Syn1a SUMOylation is inhibited. These results indicate that preventing SUMOylation causes a defect in presynaptic vesicle targeting of SynIa, resulting in a functional defect in the vesicle pool.
To further explore the role(s) of SUMOylation in SynIa association with SVs, we performed SynIa dispersion and reclustering assays, which measure changes in fluorescence of GFP-tagged SynI at synaptic boutons 35 . Intense synaptic activity causes SynI to diffuse away from synaptic boutons due to dissociation from the SVs. Following stimulation, SynI then reclusters at the synapse. Using this approach, we were able to   determine the contribution of SUMOylation to both association and disassociation of SynIa with SVs.
We expressed GFP-tagged SynIa WT and K687R mutant in hippocampal neurons. At rest, both proteins displayed a clear punctate distribution. Electrical field stimulation of 900 AP at 10 Hz produced a significant decrease in punctate fluorescence as GFP-SynIa diffused away from synaptic boutons (Fig. 4e,f). Furthermore, WT and K687R SynIa showed a similar magnitude and time constant of dispersion (Fig. 4g,h), as well as the number of responsive puncta (Supplementary Fig. 5b). Importantly, however, the K687R mutant exhibited markedly different reclustering dynamics. Punctate fluorescence for WT SynIa completely recovered to the baseline levels 3 min after stimulation, whereas the non-SUMOylatable K687R mutant only recovered B50% at 3 min and had a significantly higher time constant for reclustering (Fig. 4i). These results indicate that SUMOylation does not participate in SynIa dissociation from SVs but that it is required for reclustering after synaptic activation. Furthermore, because CamKII phosphorylation mediates SynIa dissociation from SVs, these data argue against a direct interplay between phosphorylation and SUMOylation of SynIa in regulating SynIa-SV interactions. Rather, our results suggest that phosphorylation and SUMOylation control SynIa function via separately coordinated pathways.
Taken together, our data indicate that SUMOylation of SynIa promotes binding to SVs and that this is required for correct presynaptic targeting of SynIa and reclustering of SVs following stimulation. This, in turn, replenishes and maintains the reserve vesicle pool to ensure presynaptic integrity and correct dynamics of neurotransmitter release. We attribute the decrease in SV exocytosis when SynIa SUMOylation is inhibited (Fig. 2) to the defects in the establishment and maintenance of the vesicle pools, resulting in fewer vesicles being available for release during prolonged stimulation. In particular, this phenotype bears striking resemblance to that observed in sypHy exocytosis assay under cdk5 inhibition 5 , in which a shift in SV ratio between releasable and non-releasable pool was observed. One attractive possibility is that SUMOylation and cdk5 are part of a pathway that fine-tunes the ratio of releasable and non-releasable SVs.
As reported for most SUMO substrates 36 , only a small proportion of SynIa is SUMOylated at any given time. Consistent with this, our proposed mechanism of action only requires SynIa to be SUMOylated for the initiation of SV association. Once the binding is stabilized, SUMO can be removed as SynIa itself has a high affinity for SV. (Graphical representation in Supplementary Fig. 6).
SynIa SUMOylation and disease. Mutations in SynIa have been associated with epilepsy and ASD 37 . These include two nonsense mutations that abolish the E domain and other less severe mutations that do not significantly affect SV binding or phosphorylation by CamKII, but do result in defective SynIa targeting and SV exocytosis 34 . We therefore investigated the possible role of SUMOylation in the pathogenesis of these mutations. Of the mutants we investigated, the A548T mutation (analogous to A550T in human SynIa 34 ) significantly decreased the levels of SynIa SUMOylation in N2A cells (Fig. 5a,b), and the defect in SV exocytosis caused by the A548T mutation is strikingly similar to that observed with K687R mutation 34 , suggesting a link between SUMOylation and ASD/epilepsy. We next compared the presynaptic localization of GFP-SynIa WT, K687R and A548T in neurons in which endogenous SynIa was knocked down (Fig. 5c). Both non-SUMOylatable K687R and the autism mutant A548T exhibited similar SynIa targeting defects compared with WT SynIa (Fig. 5d). While we cannot formally exclude the possibility that targeting defects result from disruption of distinct pathways, these results are consistent with a common mechanism of action. To explore this in more detail, we used the SynIa dispersion assay to compare the effects of the A548T with a double A548T/K687R mutant. As expected, there were no differences in the dynamics of dispersion between the WT and either mutant, but both mutants were significantly slower in reclustering compared with WT. Moreover, the time constants for reclustering were identical for the A548T and the A548T/K687R double mutant (Fig. 5e-i) indicating that there is no additive effect of the A548T mutation in non-SUMOylatable SynIa. Again, these data are consistent with both A548T and K687R disrupting SynIa reclustering via the same pathway, and strongly suggest that the defects in SV exocytosis and SynIa targeting/clustering in this ASD mutation can be attributed to defective SUMOylation.
How the A548T mutation affects SUMOylation at the molecular level remains to be determined. Residues A548 and K687 are relatively distant from each other and in silico analysis of the A548T mutation does not predict any marked alteration in SynIa structure 34 . Interestingly, an in vitro SUMOylation assay showed no decrease in SUMOylation efficiency for A548T mutant ( Supplementary Fig. 7). Since SUMOylation in vitro requires only Ubc9 interaction, it is unlikely that A548T mutation directly influences Ubc9 binding. Nonetheless, one possible explanation is that A548T changes SynIa conformation in such a way to occlude the SUMO consensus sequence. Alternatively, A548 may be required for SynIa interactions with a SUMO E3 ligase. Overall, our data clearly support the hypothesis that the A548T mutation-mediated defects in SynIa SUMOylation result in the dysfunction of SV dynamics. These changes, in turn, likely cause imbalances between excitatory and inhibitory stimulation that can underlie ASD and epilepsy [38][39][40] . Thus, we propose that SynIa SUMOylation plays a critical role in presynaptic function and dysfunction and that modulation of this pathway represents a potentially powerful strategy for the development of new approaches to treating synaptopathies including ASD and epilepsy.

Conclusion
Here we show that SUMOylation of SynIa enhances its SV reclustering at the RP, which is required to maintain presynaptic responsiveness and normal synaptic transmission. We further demonstrate that the A548T mutation in SynIa reduces SUMOylation and that this may account for the epilepsy and ASD phenotype associated with this mutation. These results expand the repertoire of synaptic proteins that are regulated by SUMOylation and provide further evidence for the fundamentally important roles of SUMO modification of synaptic proteins in health and disease.

Methods
Molecular biology. For knockdown-rescue experiments, shRNA-resistant SynIa constructs were expressed as mCherry/GFP-IRES-SynIa cassette on a pFIV plasmid, which also expressed SynI shRNA (vector made in house, shRNA target sequence 5 0 -CCATGGAGAAATTGACATTAATA-3 0 against SynI). For all knockdown studies, 'Control' cells were transfected with pFIV (expressing mCherry or GFP) without shRNA. The Flag-tagged SUMOylation vector used in bacterial SUMOylation of SynIaDE are engineered by inserting Flag tag at the N terminus of SUMO in pE1E2S1 vector. All subcloning and mutagenesis reactions were performed according to standard protocols.
Cell culture. Embryonic cortical and hippocampal neurons were isolated from E18 Wistar Rattus Norvergicus embryos. Brain areas were dissected and trypsinized, before being plated on either PLL-coated 25-mm glass coverslips (for hippocampal cells) or PLL-coated six-well plates (for cortical cells). Cells were initially plated in plating media (Neurobasal media þ 10% horse serum, B27 supplement, 2 mM Glutamax, 1 Â penicillin/streptomycin), which after 24 h was replaced with feeding media (Neurobasal media, B27 supplement, 1.2 mM Glutamax, 1 Â penicillin/ streptomycin). For more details, see ref. 41. For biochemistry, cortical neurons were used at DIV18. For SypHy experiments, hippocampal neurons are transfected at DIV9-10 and imaged 5 days later. For dispersion/reclustering assay and localization experiments, hippocampal neurons are transfected at DIV9-10 and imaged 7 days later. All neuronal transfections were performed using Lipofectamine 2000 (Invitrogen) according to manufacturer's instructions.
Where cropping of western blots was performed, the full blots are supplied ( Supplementary Fig. 8), with boxed areas indicating the cropped areas used.
Bacterial expression and purification of SynIa constructs. Bacterial expression of His-HA-SynIaDE was carried out in BL21(DE3) E. coli using a standard isopropylthiogalactoside overnight induction protocol. The harvested bacteria were lysed by sonication in Nickel-Binding Buffer (25 mM HEPES, 500 mM NaCl, 20 mM Imidazole, protease inhibitors, pH 7.4). Precleared bacterial lysate was then load onto Ni-NTA resin. After extensive washes, the protein was eluted with elution buffer (25 mM HEPES, 200 mM imidazole, pH 7.4), then concentrated to 2 mg ml À 1 before snap freezing for storage at À 80°C. SUMOylated SynIaDE was produced by co-expression of His-HA-SynIaDE and Flag-SUMOylation plasmid in BL21(DE3) E. coli and metal-affinity purification described above and the eluate loaded onto a Flag-affinity gel (SIGMA). Retained proteins were eluted using 50 mg ml À 1 Flag peptide (SIGMA) in 25 mM HEPES, pH 7.4. Eluates were then snap-frozen for storage at À 80°C.
SV-binding assay. Whole rat brain was homogenized in 10 ml ice-cold buffered sucrose (4 mM HEPES, 320 mM Sucrose, pH 7.3) using a glass-Teflon homogenizer. The homogenate was then centrifuged in a fixed angle rotor at 800g for 10 min. The supernatant was centrifuged at 9,200g for 15 min and then the subsequent supernatant was discarded. The pellet was resuspended with 10 ml of buffered sucrose and then centrifuged at 10,200g for 15 min. The supernatant was discarded and the pellet was resuspended in 1 ml of buffered sucrose. Ice-cold water (9 ml) was then added and the suspension was homogenized with a glass-Teflon homogenizer briefly at maximum speed. HEPES-NaOH (1 M, pH 7.4) was added to the suspension to a final concentration of 3 mM, and was incubated on ice for 30 min, then centrifuged at 25,000g for 20 min. The pellet, containing crude SVs, was retained. For more details, see ref. 43. SVs were resuspended in 10 ml buffered sucrose (5 mM HEPES, pH 7.4, 300 mM Sucrose) and 2.5 ml of SV suspension was mixed with either 3 mg of SUMOylated or non-SUMOylated/ SENP-deSUMOylated His-HA-SynIa-DE and incubated at 4°C with gentle agitation for 1 h. The SVs were then collected by ultracentrifugation at 55,000g for 2 h, resuspended in 200 ml buffered sucrose and the extent of His-HA-SynIa-DE binding was assessed by western blot analysis using an anti-HA antibody. SENP treatment of SUMO-SynIaDE was performed by adding purified recombinant SENP (catalytic domain only) at 1:100 mass ratio to SUMOylated His-HA-SynIaDE at incubate at room temperature for 30 min.
Immunocytochemistry. Immunocytochemistry was performed following paraformaldehyde fixation of cells and permeabilization with 0.1% Triton X-100. For presynaptic co-localization assays, anti-bassoon at 1:500 (Abcam, #ab82958) and anti-VAMP2 at 1:1,000 (Synaptic Systems, #104202) was used to stain the presynaptic terminal. The number of GFP puncta positive for bassoon or VAMP2 were counted and reported as a percentage of total GFP puncta. Each field is counted as one n and consists of at least 20 puncta and at least 3 independent neuronal culture preparations were assessed. For quantification in the knockdownrescue experiments, anti-SynI at 1:500 (Novua # NB300-104) was used to assess SynI levels in mCherry-expressing neurons. All quantification was performed using ImageJ software.
SypHy experiments. Neurons co-transfected with SynIa knockdown, or knockdown-rescue constructs (see Molecular Biology section for details) were mounted in a Warner stimulation chamber (Harvard Apparatus), connected to a Digitimer Constant Voltage stimulator and a Master-8 pulse train generator. Exocytosis was triggered using electrical field stimulation, using 1 ms pulses (APs) of 50 V with a paradigm of 600 APs at 20 Hz. All experiments were performed in HBS with 50 mM D-AP5, 25 mM CNQX and 1 mM bafilomycin A. Perfusion with NH 4 Cl (pH 7.4), was used to reveal total levels of SypHy loading at the end of each experiment. Images were taken at 0.5 Hz using a CCD camera with a GFP filter cube. For each cell, 10 regions of interest (ROIs) with area of 2.13 Â 2.13 mm were manually placed covering single SypHy puncta chosen blindly. Average fluorescence signal were measured with non-responsive ROIs discounted. Data were first normalized to background levels (that is, DF/F 0 ), and then expressed as a percentage of fluorescence after NH 4 Cl perfusion (F max ) or after stimulation (F post-stim ). Extrasynaptic ROIs were used to correct for background fluorescence. For more details, please see ref. 17.
SynIa dispersion and reclustering assay. GFP-SynIa-transfected neurons were mounted in the electrical field stimulation chamber, filled with HBS supplemented with 25 mM APV and 50 mM CNQX. Confocal images were recorded at 0.2 Hz for 10 min using 488 nm argon laser excitation, with a 500-530-nm band-pass filter. Images were scanned at 512 Â 512 resolution with suitable gain and offset for the best contrast and avoiding saturation of signals. This recording was to establish the effect of photobleaching, and therefore no stimulation was performed. After the recording was finished, a nearby region of the same neuron was imaged with identical settings. For this recording, 900 AP at 10 Hz (90 s) electrical stimulation (50 V, 1 ms pulses) were delivered 1 min after the start of image recording. For each cell, seven ROIs with area of 0.83 Â 0.83 were manually placed covering single GFP puncta, chosen blindly at t ¼ 0. Average fluorescence signals were measured with non-responsive ROIs discounted. Changes in fluorescence from initial values were normalized to initial levels (that is, DF/F 0 ). The same analysis was performed on the unstimulated recording to correct for photobleaching. To calculate the dispersion rate, the DF/F 0 at t ¼ 60-150 s of each neuron were fitted with a single-fall exponential decay as follows where A is the plateau of the curve and t is the time constant that reflects the rate of increase. The t from neurons transfected with the same construct were averaged for comparison between constructs with statistical testing.
To calculate the reclustering rate, the DF/F 0 at t ¼ 150-600 s of each neuron were fitted with a single-fall exponential decay as follows where A is the plateau of the curve and t is the time constant that reflect the rate of increase. The t from neurons transfected with the same construct were averaged and then compared between different constructs with statistical testing.
For further details, see ref. 35.
Statistical analysis. Graphpad Prism software (Graphpad Inc.) was used for statistical analysis. Data were analysed using one-way analysis of variance (ANOVA) with Bonferroni's post hoc test (for comparisons between multiple data sets) or two-tailed Student's t-test (comparison between two data sets). Curve fitting in Figs 2, 4 and 5 was performed using Graphpad Prism, fitting to a single exponential rise function y ¼ a(1 À e À t/t ) where t is the time constant.
Sequence alignment. Sequence alignment is done in UGENE using ClustalX algorithm. Synapsins sequences were obtained from NCBI.