Compartmentalization of soluble endocytic proteins in synaptic vesicle clusters by phase separation

Summary Synaptic vesicle (SV) clusters, which reportedly result from synapsin’s capacity to undergo liquid-liquid phase separation (LLPS), constitute the structural basis for neurotransmission. Although these clusters contain various endocytic accessory proteins, how endocytic proteins accumulate in SV clusters remains unknown. Here, we report that endophilin A1 (EndoA1), the endocytic scaffold protein, undergoes LLPS under physiologically relevant concentrations at presynaptic terminals. On heterologous expression, EndoA1 facilitates the formation of synapsin condensates and accumulates in SV-like vesicle clusters via synapsin. Moreover, EndoA1 condensates recruit endocytic proteins such as dynamin 1, amphiphysin, and intersectin 1, none of which are recruited in vesicle clusters by synapsin. In cultured neurons, like synapsin, EndoA1 is compartmentalized in SV clusters through LLPS, exhibiting activity-dependent dispersion/reassembly cycles. Thus, beyond its essential function in SV endocytosis, EndoA1 serves an additional structural function by undergoing LLPS, thereby accumulating various endocytic proteins in dynamic SV clusters in concert with synapsin.


INTRODUCTION
Interneuronal communication occurs primarily at specialized cellular junctions called synapses. On the presynaptic side, synaptic vesicles (SVs), tiny membranous sacs $40 nm in diameter that store neurotransmitter molecules, form clusters in the vicinity of release sites called active zones (AZs) and undergo activity-dependent exocytosis, thereby releasing neurotransmitter molecules, which are then recognized by receptors on neighboring postsynaptic cells. 1 The number of SVs in a presynaptic terminal varies from several hundreds to tens of thousands, depending on the synapse type, but SVs are categorized into three functionally distinct groups: a readily releasable pool (RRP) that undergoes exocytosis immediately on the arrival of an action potential, a recycling pool that replenishes the RRP during sustained stimulation, and a resting pool that participates in exocytosis only on prolonged stimulation. 2,3 Because strength and plasticity of synaptic transmission critically depends on the size and availability of SV pools for release, maintenance and faithful reformation of SV clusters, especially during and after intensive stimulation, is essential for proper synaptic transmission.
Recently, the liquid-like nature of synapsin, the abundant soluble protein family specifically expressed in neurons, has been demonstrated in vitro, and may explain how this protein helps to form SV clusters. 4 Synapsin undergoes liquid-liquid phase separation (LLPS) to form micron-sized, membrane-less organelles, and also recruits liposomes mimicking the lipid composition of SVs, which contain negatively charged lipids. 5 Such properties have been recapitulated in heterologous expression in COS7 cells, in which coexpression of synapsin with synaptophysin, which promotes formation of small vesicles, resulted in formation of SV-like clusters in the cytoplasm. 6 However, formation of synapsin condensates in vitro takes $1 h, and mere expression of synapsin in COS7 cells does not result in droplet-like structures. 4,6 Although dispersion of SV clusters by injection of an antibody that impedes formation of synapsin LLPS into lamprey synapses suggests that synapsin LLPS predominantly contributes to maintenance of SV clusters also in living synapses, 7 it remains controversial whether additional components are necessary to form SV clusters at rest, and also for reassembly of clusters during and after intensive stimulation. Of interest, a-synuclein, another SV-associated soluble protein, also has potential to undergo LLPS. It also aids formation of In this study, we demonstrate that endophilin A1 (EndoA1), known as an endocytic scaffold protein, forms highly condensed, self-organized assemblies via LLPS at physiologically plausible concentrations in presynaptic terminals. Although EndoA1 alone cannot promote SV-like vesicle clusters in COS7 cells, EndoA1 condensates drastically facilitate formation of synapsin droplets and exhibit potential to recruit various endocytic proteins, such as intersectin and dynamin. Furthermore, EndoA1 accumulates in SVlike clusters in a synapsin-dependent manner, suggesting that EndoA1 condensates are responsible for accumulation of other endocytic proteins in SV clusters. Consistent with liquid-like behaviors in vitro and in heterologous systems, EndoA1, positioned in SV clusters by LLPS, is dispersed from presynaptic terminals simultaneously with synapsin during repetitive stimulation, and is re-assembled after stimulation with kinetics similar to those of synapsin in cultured hippocampal neurons, indicating that cooperative behavior of EndoA1-and synapsin 1-condensates may serve an essential function in formation and reassembly of SV clusters in response to stimulation.

Endophilin A1 undergoes LLPS in vitro
To examine whether soluble endocytic proteins have potential to undergo LLPS in vitro, we set up a series of expression vectors, either pGEX or pET, encoding FCHo2, FCHSD2, a full-length or a (SH3) 5 -domain of intersectin 1 (ITSN1-(SH3) 5 ), endophilin A1 (EndoA1), amphiphysin, synaptojanin and dynamin 1, and expressed them in E. coli. These proteins potentially form a multivalent protein network, 21 which represents the central characteristics of protein phase separation via distinctive domain structures (SH3 domain, proline-rich domain and mHD domain) ( Figure S1). 13,21,22 After single-step affinity column purification, we obtained GST-fused EndoA1 fulllength and 63 His-tagged ITSN1-(SH3) 5 with a high degree of purity and sufficient yields, whereas others were either poorly expressed, contained multiple degraded bands, or were partitioned into the insoluble fraction ( Figure 1A). An initial test to examine their potential to form droplets under differential interference contrast (DIC) microscopy revealed that GST-EndoA1 formed micron-sized droplet-like structures in the presence of 10% polyethylene glycol (PEG), a widely used crowding reagent, at a final protein concentration of $20 mM, whereas 63 His-tag ITSN1-(SH3) 5 (also at a final concentration of $20 mM) formed droplets irrespective of PEG ( Figure 1B). To exclude undesired effects from affinity tags or minor contamination from host cells, we cleaved off affinity tags using proteases and removed contamination through further purification steps using an ion exchange column and/or a gel filtration column, resulting in much purer preparations ( Figure 1C). Using these purified proteins, we then tested their potential to form droplet-like structures in various protein and PEG concentrations. Notably, in the presence of >7.5% PEG, EndoA1 formed clear droplets at protein concentrations above 10 mM, which was the physiologically relevant concentration in synaptosomes derived from rat cortex (16 mM) 23 ( Figure 1D, left). On the other hand, ITSN1-(SH3) 5 lacking 63 His-tag did not form clear droplets at its physiological concentration (20 mM) even in the presence of 10% PEG, and formed clear droplets only at 60 mM, far beyond its concentration in synaptosomes 23 (Figure 1D, right). These droplets spontaneously merged ( Figure 1E for EndoA1), progressively increasing in size up to 10 mm during incubation up to $60 min at room temperature ( Figure 1F).
To determine which part of EndoA1 accounts for its capacity to undergo LLPS, we prepared its N-BAR domain (aa 1-247) and tested whether it formed droplets ( Figure 1G). At  iScience Article condensates. Notably, EndoA1-BAR protein formed much larger amorphous structures on the surface of a coverslip which were not seen in the case of full-length EndoA1, indicating a contribution of the C-terminus of EndoA1 to define biophysical properties of EndoA1 droplets (see also Figure 2).
To further validate the potential of EndoA1 to form droplets via LLPS, and its ability to recruit ITSN1, which reportedly interacts with EndoA1 through SH3-SH3 interactions, 24 we covalently labeled EndoA1 and  5 ) and GST-endophilin A1 (EndoA1) proteins with single-step affinity chromatography. 13.5 mg of recombinant proteins were loaded onto a 12.5% SDS-PAGE gel. Molecular weights are indicated on the left (kDa). (B) Phase separation assay for 63 His-ITSN1-(SH3) 5 and GST-EndoA1 proteins. 63 His-ITSN1-(SH3) 5 (20 mM) forms droplet-like structures irrespective of the presence of 10% PEG8000 (upper panels). On the other hand, GST-EndoA1 (20 mM) forms droplet-like structures only in the presence of PEG8000 (lower panels). All images were taken under differential interference contrast microscopy (DIC) after 60 min incubation at room temperature either in the absence (0%) or presence (10%) of PEG8000.
(C) Coomassie blue staining of purified recombinant ITSN1-(SH3) 5 and EndoA1 proteins after removal of affinity tags and multi-step column chromatography. 13.5 mg of recombinant proteins were loaded onto a 12.5% SDS-PAGE gel.
(D) Phase diagrams of EndoA1 and ITSN1-(SH3) 5 Figure 2A). In contrast, FRAP of Cy5-labelled EndoA1-BAR domain exhibited much slower recovery than that for full-length EndoA1 (Figure 2A), indicating the rest of the EndoA1 protein, e.g., the SH3 domain, may help to determine the biophysical properties of EndoA1 condensates. When Cy5-labelled EndoA1 was mixed with Cy3-labelled ITSN1-(SH3) 5 (both at 20 mM in the presence of 10% PEG, in which ITSN1-(SH3) 5 by itself did not undergo phase separation, Figure S2), ITSN1-(SH3) 5 formed spherical droplets that perfectly co-localized with EndoA1 (Figure 2B). In contrast, the BAR domain of EndoA1, lacking the SH3 domain, did not recruit ITSN1-(SH3) 5 into the droplets ( Figure 2B), suggesting that the SH3 domain of EndoA1 is responsible for recruiting ITSN1 into the condensed phase. It should be noted that the characteristics of EndoA1 and its BAR domain that enable it to undergo LLPS are fully compatible with recent observations by Mondal et al., in which formation of EndoA1 condensates with its binding partners via multivalent interactions are implicated in fast EndoA1-mediated, clathrin-independent endocytosis operating at plasma membranes of non-neuronal cell types. 25 However, the endogenous expression level of EndoA1 in non-neuronal cells, which might be one of the key parameters to define the ability of proteins to phase separate in a physiological context, were not carefully taken into account in their study (see also discussion).

EndoA1 forms liquid-like droplets in COS7 cells
Consistent with results in vitro with purified proteins, EndoA1 C-terminally tagged either with EGFP or TagRFP formed droplet-like structures in cytoplasm when expressed in COS7 cells, whereas EGFP or TagRFP alone did not ( Figure 3A). Co-expression of EndoA1 tagged with different fluorescent proteins resulted in condensates that contained both ( Figure 3B), ruling out a potential artifact that weak homophilic interactions between the respective tags facilitated droplet formation. These droplets exhibited FRAP, albeit more slowly and to a lesser extent than those observed in vitro ( Figure S3). Although inspection of EndoA1 levels in COS7 cells by immunoblotting suggested that EndoA1-EGFP expression was $8-fold higher than that observed in lysates extracted from cultured hippocampal neurons ( Figures 3C-3E), the expression level of EndoA1-EGFP in COS7 cells might not be much above its endogenous EndoA1 levels in presynaptic terminals, because EndoA1 is substantially enriched in presynaptic terminals, 26 (also see below). It should be emphasized here that endogenous EndoA1 in COS7 cells eluded detection, suggesting that the endogenous EndoA1 level is not sufficiently high to undergo LLPS in the cytoplasm of fibroblast cells. iScience Article EndoA1 is recruited into SV-like vesicle clusters in a synapsin-dependent manner in COS7 cells The ability of EndoA1 to form condensates in COS7 cells enabled us to examine possible contributions of EndoA1 in SV cluster formation, using those cells as a model. A previous study demonstrated that co-expression of synapsin and tag-free synaptophysin (Syph) induces formation of SV-like vesicle clusters in the cytoplasm, whereas Syph C-terminally tagged with EGFP alone also induced similar clusters, presumably owing to the property of EGFP to form weak dimers ( Figure S4). 6 Of interest, despite the ability of synapsin to undergo LLPS in vitro, 4 expression of synapsin alone in COS7 cells does not show droplet-like appearance, but rather shows a coarse cytoplasmic distribution. 6 Thus, this system is well suited for testing possible contributions of EndoA1 in modulation of synapsin-LLPS, as well as in formation of SV-like vesicle clusters, either dependent on synapsin, or not.
First, based on biochemical data showing a direct interaction between EndoA1 and synapsin, 19 we examined whether co-expression of EndoA1 had any impact on synapsin in COS7 cells ( Figure 4A). Consistent with a previous study, 6 TagRFP-tagged synapsin 1 exhibited a coarse cytoplasmic distribution without clear signs of droplet-like puncta. However, co-expression of EndoA1-EGFP drastically promotes synapsin droplets in the cytoplasm that perfectly colocalize with EndoA1 ( Figure 4A). iScience Article Next, we examined whether EndoA1 can promote accumulation of SV-like vesicles into clusters. To this end, either EndoA1-EGFP or TagRFP-synapsin 1 was co-expressed with tag-free Syph ( Figure 4B). Although synapsin facilitated formation of SV clusters marked by Syph antibody as previously reported, 6 EndoA1 condensates did not accumulate Syph-carrying vesicles into the droplets ( Figure 4B). Moreover, when EndoA1 was co-expressed with Syph-EGFP, EndoA1 formed droplet-like structures independent of Syph-EGFP positive vesicle clusters ( Figures 4C and 4E), whereas synapsin was co-localized with Syph-EGFP-positive puncta as previously reported. 6 Finally, when EndoA1 was co-expressed with synapsin and tag-free Syph, EndoA1 was effectively recruited into synapsin/Syph double-positive puncta ( Figures 4D and 4E). These results demonstrate that although EndoA1 alone does not suffice to form SV clusters, it is recruited into SV clusters in a synapsin-dependent manner and can cause synapsin to undergo LLPS.

EndoA1 and synapsin form cytoplasmic condensates independent of SV-like vesicle clusters
BAR-domain containing proteins, including EndoA1, bind and tubulate lipid membranes in vitro and in heterologous cells. [27][28][29] When present at high concentrations, e.g., 40 mM, however, the BAR domains can produce small vesicles, e.g., $50 nm in diameter, from larger liposomes. 30 Moreover, endophilin has been implicated in one of the clathrin-independent endocytosis modes, i.e., the Fast Endophilin Mediated Endocytosis (FEME). 31 Given the potential of synapsin to recruit negatively charged vesicles 4 and to coalesce with EndoA1 ( Figure 4A), it is therefore possible that overexpression of EndoA1-EGFP in COS7 cells could facilitate the intrinsic FEME, resulting in hyper-production of small vesicles that could then be recruited into cytoplasmic EndoA1 condensates, in particular, with the aid of synapsin. To examine this possibility, a correlative light and electron microscopy (CLEM) analysis was performed to visualize membrane structures within the droplets that were fluorescently labeled with proteins of interest. For comparisons, (lower panels) Co-expression of TagRFP-synapsin 1 (magenta) and tag-free synaptophysin (Syph, green) forms SV-like clusters structures in COS7 cells (see also Figure 5). Tag-free Syph was visualized by immunostaining after fixation. (C) Synapsin, but not EndoA1, is recruited into SV-like clusters induced by expression of Syph-EGFP in COS7 cells. (upper panels) EndoA1-TagRFP (magenta) forms independent droplets from Syph-EGFP puncta (green). (lower panels) TagRFP-synapsin 1 (magenta) colocalizes with Syph-EGFP puncta (green).
(E) Quantification of EndoA1 droplets co-localized either with Syph-EGFP or with tag-free Syph co-expressed with synapsin 1 (Syn1). Colocalization analysis was performed using the Manders coefficient. Values are mean G s.e.m. with n = 9 images and n = 11 images for Syph-EGFP/EndoA1-TagRFP and tag-free Syph/TagRFP-Syn1/EndoA1-EGFP, respectively. ***p < 0.001, unpaired t-test. iScience Article two conditions were examined, i.e., co-expression of EndoA1 and synapsin 1 either in the absence or presence of tag-free Syph, corresponding to the conditions shown in Figures 4A and 4D, respectively. When EndoA1 and synapsin were co-expressed, we did not detect any membranous structures that correlated well with positions of fluorescence-positive puncta throughout Z-stacks ( Figure 5A and VideoS1). In stark contrast, when tag-free Syph was transfected in addition, the fluorescent puncta that were positive both for synapsin and EndoA1 contained highly electron-dense structures, which appeared under higher magnification to be clusters of small SV-like vesicles, consistent with a previous observation 6 ( Figure 5B and VideoS2). These results not only confirm that EndoA1 indeed co-assembles with SV-like vesicle clusters induced by synapsin and Syph, but also indicate that EndoA1 can form condensates that are devoid of small vesicle membranes even in the presence of synapsin in a cellular context.
EndoA1 condensates recruit ITSN1, dynamin, and amphiphysin that are not accumulated into SV-like vesicle clusters by themselves As indicated by in vitro experiments with EndoA1 and ITSN1-(SH3) 5 ( Figure 2) as well as the ability of EndoA1 to recruit synapsin into condensates in COS7 cells ( Figure 4A), it is plausible that EndoA1 condensates also recruit its binding partners, including some endocytic proteins of which we failed to obtain pure protein preparations with the E. coli expression system. When dynamin 1 or amphiphysin (Amph) tagged with a fluorescent protein was transfected in COS7 cells, both proteins showed dispersed cytoplasmic distributions without visible droplet-like structures ( Figure 6A). In contrast, EGFP-ITSN1, a full-length longsplicing variant, formed spherical droplets in the cytoplasm ( Figure 6A), likely reflecting the potential of ITSN1-(SH3) 5 to undergo LLPS at high concentrations ( Figure 1D). When co-expressed with EndoA1, both dynamin 1 and ITSN1 formed droplets together with EndoA1, whereas Amph did not ( Figures 6A  and 6B). The latter was unexpected, but an additional transfection of dynamin robustly promoted formation of droplet-like structures that contained Amph, dynamin and EndoA1 ( Figure 6C). These results demonstrate that EndoA1 condensates serve as a reservoir for other endocytic proteins through direct or indirect interactions that could form multivalent molecular networks among endocytic proteins. 22 As suggested by the above observations, we then examined whether ITSN1, dynamin, or Amph can be recruited into SV-like vesicle clusters like EndoA1. Despite the potential of ITSN1 and Amph to interact with synapsin, 18,20 however, none of them were recruited into SV-like vesicle clusters induced by synapsin 1 and tag-free Syph ( Figure 6D). Notably, despite in vitro observations that a recombinant ITSN1-(SH3) 5 fragment was recruited into synapsin droplets, 4 full-length ITSN1 formed cytoplasmic droplets that were independent of SV-like clusters marked by synapsin/Syph. Thus, these results collectively indicate that localization of these endocytic proteins in SV clusters depends critically on the presence of EndoA1. iScience Article EndoA1 localizes within SV clusters at hippocampal synapses The above results in vitro and in COS7 cells support the notion that multiple endocytic proteins are clustered in SV clusters, aided by phase separation of synapsin and EndoA1 through multivalent interactions.
To gain further insights into EndoA1 condensates in mammalian central synapses, which exhibit more complex membrane organization than lamprey synapses, we attempted to map EndoA1 proteins related to SV clusters in fixed cultured hippocampal neurons in greater detail by employing a three-dimensional stochastic optical reconstruction microscope (3D-STORM) 32 combined with a quantitative framework to evaluate colocalization of two synaptic proteins in the same preparation ( Figure 7A). To validate EndoA1 localization in presynaptic terminals, synapsin 1, VGLUT1, complexin and bassoon were chosen as specific markers for SV clusters, a presynaptic soluble protein and active zones. 4,33- 35 We compared spatial relationships of 5 pairs of synaptic proteins listed in Figure 7C by evaluating three parameters; i.e., Manders coefficients, 36 volume ratios, and the center-of-mass distance between the two synaptic protein clusters ( Figure 7C). Such analyses revealed that (1) protein clusters of EndoA1, synapsin1, and VGLUT1 were highly co-localized (>80%), (2) volumes occupied by EndoA1, VGLUT1, and synapsin 1 were identical, as evidenced by respective cluster volume ratios close to 1.0 (1.15 G 0.04 for EndoA1/synapsin 1, n = 25 synapses; 1.02 G 0.03 for . Values indicate means G s.e.m. with n = 9 images, n = 7 images, and n = 8 images for Dynamin, Amph, and ITSN1, respectively. ***p < 0.001, ns, not significant, one-way ANOVA followed by the Tukey-Kramer post-hoc test. (C) Amph co-assembles with EndoA1 droplets through dynamin 1. Triple-transfection of EndoA1 (green), dynamin 1 (magenta), and Amph (red) in COS7 cells results in complete colocalization of the three proteins.
(D) ITSN1, dynamin and Amph are not recruited into SV-like vesicle clusters induced by Syn1 + Syph. COS7 cells were triple-transfected with each endocytic protein plus synapsin + tag-free Syph (Syn1 + Syph). Insets are magnified views of areas indicated by arrowheads. Note that, unlike EndoA1, none of these proteins co-assemble with Syn1 + Syph.  (3) center-ofmass distances were not statistically different ( Figures 7C and S5), indicating that spatial distributions of the three proteins are almost identical. In stark contrast, comparisons of complexin 1/2 and synapsin 1 revealed that the volume ratio of complexin 1/2 clusters to synapsin 1 clusters was significantly higher than 1 (1.68 G 0.08 for complexin1/2 / synapsin1, n = 32 synapses). Considering the short center-of-mass distance ($50 nm) and the high colocalization ratio ($90%) between them, it is conceivable that synapsin 1, and therefore also EndoA1 and VGLUT1, partially occupy complexin 1/2 clusters. Given the volume ratio between complexin 1/2 to synapsin (1.68), and assuming that complexin 1/2 is evenly distributed in presynaptic terminals, SV clusters containing synapsin, as well as VGLUT1 and EndoA1, would occupy $60% of total volume of presynaptic terminals. As expected, quantification of bassoon clusters relative to EndoA1 with STORM showed much less overlapping distribution of these proteins with $200 nm of center-of-mass distance and only $30% cluster overlap ( Figures 7C and S5).

EndoA1 exhibits liquid-like properties in cultured hippocampal neurons
Evidence for EndoA1 and synapsin to form condensates in living synapses is still lacking. To examine this, cultured hippocampal neurons were exposed to a solution containing 3% 1,6-hexanediol (1,6-HD), which has often been used to disrupt biomolecular condensates, 12 for 1 min and were immediately fixed for immunolabeling. After this treatment, the punctate appearance of EndoA1, as well as that of synapsin, disappeared, as evident from significant decreases in the coefficient of variation of the fluorescent signals ( Figures 8A and S6). Immunolabeling of VGLUT1, an SV marker in glutamatergic synapses was also dispersed, albeit to lesser extent, supporting the role of synapsin 1 to maintain SV clusters. On the other hand, immunolabeling of bassoon (an AZ marker) remained unaffected. Although it has been proposed  4) The center-of-mass and the volume of each protein cluster were determined by calculating mean voxel coordinates and total voxels of extracted major clusters for each synaptic protein, respectively. Manders coefficients were quantified by calculating the ratio of overlapping voxels to total voxels of the reference protein cluster. (B) (left panels) A representative 2D projected two-color STORM image of complexin 1/2 (green) and synapsin 1 (magenta) in a single synaptic bouton. Complexin 1/2 immunoreactivity is distributed more widely than synapsin 1 signals and wraps around the synapsin 1 cluster. (right panels) 3D visualization of the complexin 1/2 (green) and synapsin 1 (magenta) major clusters in a single synaptic bouton.
(B) EndoA1 and synapsin 1 dispersed from presynaptic terminals upon high K + stimulation, and subsequently recovered after 10 min. Images show representative immunolabeling of synapsin 1 (green), EndoA1 (magenta), bassoon (green) and VGLUT1 (magenta) without stimulation (Ctrl, top), fixed immediately after 1 min high K + (45 mM) treatment (High K + , middle) and fixed at 10 min after treatment, when high K + solution was changed to ACSF (Recovery, bottom). Values are means G s.e.m. for synapsin 1 and EndoA1 (n Ctrl = 6 images, n High K + = 8 images, n Recovery = 8 images, n Ca 2+ free = 8 images), for bassoon and VGLUT1 (n Ctrl = 6 images, n High K + = 8 images, n Recovery = 8 images, n Ca 2+ free = 8 images) normalized to the control. *p < 0.05, ***p < 0.001, ns, not significant. one-way ANOVA followed by the Tukey-Kramer post-hoc test. iScience Article that the AZ proteins, presumably including bassoon, are assembled with liquid-like condensates composed of various AZ proteins such as RIM and RIM-BP2, 35,37 the resistance to 1,6-HD exposure in our conditions suggests that the molecular assembly of AZ proteins seems to be more rigid than that of SV clusters composed of SVs, synapsin and EndoA1.
As an alternative approach, we expressed EndoA1-EGFP or EGFP-synapsin 1 in cultured neurons, both of which showed punctate presynaptic localization, and monitored their behaviors in response to 3% 1,6-HD treatment for 1 min and thereafter. We confirmed that both proteins were dispersed rapidly upon addition of 1,6-HD, and recovered to their initial states within 5 min ( Figure S6), indicating that both proteins indeed form condensates through LLPS in living synapses.

Activity-dependent dispersion and reassembly of EndoA1 and its synchronicity with synapsin in cultured hippocampal neurons
Finally, we asked whether EndoA1 condensates would dynamically transform during neural activities, in which robust SV exocytosis, subsequent endocytosis, and SV reformation would be occurring continuously. It has been demonstrated that synapsin condensates formed in vitro as well as cytoplasmic droplets consisting of synapsin and Syph-laden vesicles in COS7 cells, dissociate on activation of CaMKII that phosphorylates synapsin, suggesting that activity-dependent Ca 2+ influx into presynaptic terminals may trigger dissociation of synapsin condensates, and thereby, SV clusters. 4,6 Indeed, synapsin was previously shown to exhibit activity-dependent dispersion from presynaptic terminals in cultured living neurons. 38 To test whether EndoA1 also undergoes activity-dependent dispersion, we first exposed cultured neurons on coverslips to a high potassium solution (high K + ) for 1 min, fixed them immediately or after 10 min, and subjected them to immunolabeling. Comparisons of the coefficient of variation of immunolabeling for synapsin and EndoA1 clearly indicate their activity-dependent dispersion from boutons and their subsequent recovery, whereas those of bassoon and VGLUT1 stayed within boutons ( Figure 8B). Dispersion of synapsin and EndoA1 is clearly activity-dependent, because it was not observed in the absence of extracellular Ca 2+ .
To examine further whether the activity-dependent dispersion and the subsequent recovery of EndoA1 are kinetically synchronized with those of synapsin, we performed live imaging of EndoA1-EGFP or EGFP-synapsin 1 expressed in cultured neurons ( Figure 8B). Although the extent of dispersion from puncta was significantly larger for synapsin 1 than for EndoA1 during 600 action potentials (APs) at 10 Hz ( Figures 8C and 8D), kinetics of dispersion during repetitive stimulation and of recovery after stimulation were indistinguishable (Figures 8C, 8E, and 8F).

DISCUSSION
Recent evidence has emerged to support a novel concept that various sub-compartments in presynaptic boutons are organized by phase separation of distinct protein components, such as synapsin for SV clusters and the major AZ proteins, such as RIM/RIM-BP2, liprin, and ELKS for AZ assembly. [39][40][41] Our current results provide another example of biomolecular condensates in presynaptic terminals that enable condensation of various endocytic proteins within SV clusters. Our findings not only provide strong support for the series of electron microscopic observations in lamprey reticulospinal synapses that various endocytic proteins including endophilin reside in the SV clusters, 9,42 but are also compatible with a proposal by Denker et al. that the resting pool of SVs functions as a 'buffer' for various soluble proteins associated with SV recycling. 43 More importantly, our results indicate that there is a certain degree of hierarchy among endocytic proteins in terms of their properties to accumulate in SV clusters. Among proteins tested in this study, EndoA1 is the only protein that has potential to undergo phase separation at its physiologically relevant concentration at presynaptic terminals, to recruit various endocytic proteins into the condensates, and to accumulate in SV-like clusters via synapsin. In contrast, other endocytic proteins examined (dynamin 1, ITSN1 and amphiphysin) do not form condensates by themselves, but can be recruited into SV clusters aided by synapsin-EndoA1 condensates. Intriguingly, EndoA1 drastically drives synapsin to form condensates in COS7 cells, implicating that EndoA1 can modulate SV cluster formation through synapsin. These features of EndoA1 suggest that, in addition to its essential roles during SV endocytosis and reformation reported previously, 10,11,26,27,44,45 EndoA1 serves a structural function in maintenance of dynamic SV clusters that commingle various endocytic proteins.
Although a recent study demonstrated the potential for EndoA1 to form condensates, which is fully compatible with our in vitro results, 25  iScience Article place locally at the inner surface of plasma membrane in the size range of tens of nanometers, 25 our results suggest that micron-sized EndoA1 condensates assemble various endocytic proteins in SV clusters, which fits well with the size of SV clusters in various types of synapses. 3 Second, whereas our results revealed that EndoA1 by itself forms condensates at physiologically relevant cytoplasmic concentrations in presynaptic terminals ($20 mM), much lower expression of all endophilin isoforms in fibroblast cells, e.g., <1 mM in total in HeLa cells 46 (Figure 3) strongly suggest that it cannot form cytoplasmic condensates, at least by itself in fibroblast cells. Furthermore, synapsin that underlies formation of SV clusters is not expressed in fibroblast cells. 46 Thus, the molecular assembly of synapsin and EndoA1 mediated by LLPS and described in this study may represent a structural specification of presynaptic terminals that is required for their peculiar functioning.
Because protein phase separation is highly dependent on concentrations of components, we speculate, according to our results with ITSN1, that EndoA1 may not be the only endocytic protein that functions as a 'scaffold' to recruit 'client' proteins into the condensates. 47 Our results in vitro as well as in COS7 cells strongly suggest that ITSN1 can behave as another 'scaffold' protein especially when it is once recruited and condensed into EndoA1 droplets, in which the concentration of ITSN1 becomes much higher than when it is in diluted phase and may therefore exceed a critical concentration for ITSN1 to undergo LLPS. Likewise, other endocytic proteins not tested in this study also have potential to interact with EndoA1 and ITSN1 through multivalent interactions among SH3 domains, PRD and mHD domains. 21 It is, therefore, conceivable that many, if not all, endocytic proteins at presynaptic terminals accumulate in SV clusters through phase separation of synapsin and various endocytic proteins, making a dynamic sub-cellular compartment that involves SV membranes and various protein condensates within presynaptic boutons.
Having said that, the peculiar ability of EndoA1 to facilitate formation of synapsin condensates in COS7 cells suggests that EndoA1 has the potential to modulate SV cluster formation and maintenance, primarily regulated by synapsin condensates. At lamprey synapses, however, injection of an antibody that recognizes the IDR of synapsin causes dispersal of distal SV clusters, whereas individual inhibition of interactions between synapsin and the SH3 domain of ITSN, amphiphysin, endophilin and syndapin does not cause vesicle dispersion at resting state, 7,44,48 arguing against an essential role of those SH3 containing proteins in maintaining SV clusters under resting conditions. Nevertheless, because the synapsin IDR, self-interaction of which would be perturbed by the antibody, contains two PRDs that are responsible for the interactions with various SH3-containing proteins including EndoA1, a possibility cannot be completely ruled out that the antibody binding to the synapsin IDR would have interfered possible interactions with other proteins (e.g., endophilin) that would consequently affect liquid phase organization of SV clusters. With such technical difficulties to specifically interfere with individual protein-protein interactions in mind, it should be noted that injection of antibody that binds a-synuclein, which indeed promotes synapsin droplets on co-expression in COS7 cells as EndoA1 does, causes a severe depletion of SV clusters in lamprey synapses under resting conditions, 49,50 indicating that other protein factors may be involved in formation and maintenance of SV clusters, presumably in a cooperative manner with synapsin.
Although it is difficult to test experimentally at this point, we envision the significance of EndoA1 condensates in synaptic physiology. First, by maintaining endocytic proteins at high concentrations in SV clusters near release sites, condensates might enable a rapid supply of those endocytic proteins to peri-active zones (endocytic hot spots) on demand. Consistent with this, EndoA1 condensates are dispersed, in parallel with synapsin condensates in response to stimulation (Figure 8), leading to an increase in concentrations of their constituents at peri-active zones. Subsequent reassembly of EndoA1 condensates would then lead to a reduction in cytoplasmic concentrations of EndoA1 and its 'client' endocytic proteins as endocytosis is being terminated. Such activity-dependent dispersion and reassembly of EndoA1 condensates involving other endocytic proteins may therefore constitute an elastic supramolecular assembly that enables activation of endocytic proteins in a manner that is coupled spatio-temporally with SV exocytosis. 51 Furthermore, the ability of synapsin and EndoA1 to form membrane-free condensates in COS7 cells (Figure 5) suggests that these condensates, if they can form condensates outside of SV clusters, may serve as a molecular glue to catch newly regenerated SVs and transfer them back to existing SV clusters. The clathrin uncoating defects reported in endophilin triple KO mice, 11 and the requirement of Syph in SVs to form vesicle clusters with synapsin 6 collectively indicate that exposure of Syph on newly formed vesicles to the cytoplasm by shedding clathrin-coats may trigger the capture and subsequent recruitment of reformed SVs into SV clusters by synapsin-EndoA1 condensates. Fine manipulation of EndoA1 and synapsin ll OPEN ACCESS iScience 26, 106826, June 16,2023 iScience Article condensates in living neurons, i.e., optogenetic tools for manipulating condensates, will certainly help to decipher the role of these condensates in SV recycling.
In summary, we propose that liquid-like protein assembly mediated by EndoA1 functions as a dynamic reservoir for multiple endocytic proteins to direct them into SV clusters formed by synapsin. During neural activity, EndoA1 is dispersed concomitantly with synapsin from SV clusters, and re-assembled as endocytosis is completed. Although it has long been proposed that such migration cycles of endocytic proteins between SV clusters and peri-AZs during synaptic activity may contribute to tight coupling between SV exo-endocytosis in synapses, 42 our results provide a plausible explanation for the underlying biophysical principle, that is, the liquid-like nature of synapsin and various endocytic proteins assembled by EndoA1.

Limitations of the study
The ability of EndoA1 full-length, as well as its isolated N-BAR domain, to undergo LLPS is compatible with a recent study by Mondal et al., 25 but is somewhat surprising, because it does not contain any intrinsically disordered regions (IDRs) which are believed to be prone to phase separate. Although mechanistic understanding of how the BAR domain undergoes LLPS is still lacking, the amphipathic nature of BAR-domains, commonly seen in BAR-domain-containing proteins, may provide the force to self-assemble, which eventually triggers condensate formation. Of interest, other N-BAR domains of Amph1 and BIN1 also phase separate in vitro. 25 It will be interesting to see whether the capacity to undergo LLPS is a common feature of BAR domains.
Another limitation of this study is the lack of direct evidence to support a role of EndoA1 in formation and reassembly of SV clusters in presynaptic terminals. Further experiments involving EndoA1 loss of function and rescue with an EndoA1 mutant that specifically disables its capacity to undergo LLPS but retains full activity in SV endocytosis might help. However, as stated above, it is likely that the amphipathic property of the EndoA1-BAR domain may be responsible for its droplet formation, and if so, candidate mutations in the BAR domain will likely disrupt its membrane interaction and membrane deformation activity that are essential for SV endocytosis. 10 Dissecting these two distinct roles included in the same domain, i.e., the potential to undergo LLPS and to mediate SV endocytosis, with rigorous biochemical as well as functional assays will be the key to determine the function of EndoA1 condensates in synaptic physiology.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:  We appreciate the support of The IRCN Imaging Core at The University of Tokyo for the use of confocal microscopy. We thank Drs. Volker Haucke and Reinhard Jahn for kindly providing plasmids and antibodies used in this study. Finally, we are indebted to Dr. Hiroaki Misonou for critically reading the manuscript and to Dr. Steven D. Aird for English editing.  Figures 7 and S5, hippocampi were dissected from embryonic-day-21 Sprague-Dawley rats and incubated with 1% trypsin (Sigma) and 1 mg/mL DNase I (Sigma) for 5 min. Dissociated hippocampal cells were plated onto the glial cell layer precultured on poly-L-lysine-and laminin-(Thermo Fisher Scientific) coated coverslips at a cell density of 2,000 cells/cm 2 . Neural cells were grown in neurobasal medium (Thermo Fisher Scientific) supplemented with 2% B27 (Thermo Fisher Scientific), and 0.5 mM Glutamax (Thermo Fisher Scientific), 1 mM sodium pyruvate (Wako), and 1% penicillin/streptomycin (Thermo Fisher Scientific) at 37 C, 5% CO 2 . 2.5 mM Cytosine b-Darabinofuranoside (Sigma) was added to the culture medium at 2 DIV to limit glial proliferation. About half of the growth medium was replaced with fresh medium once a week.

Study approval
All animal experiments in the S.T. laboratory at Doshisha University, and in the K.H. laboratory at the University of Tokyo, were carried out in accordance with respective institutional regulations for animal experiments based on the governmental Guideline for Proper Conduct of Animal Experiment and Related Activities, and were approved by the institutional committees of Doshisha University and the University of Tokyo.

Molecular cloning
To construct each plasmid, corresponding coding sequences were amplified by polymerase chain reaction (PCR) using complementary DNAs reverse-transcribed from mouse brain total RNA or plasmids containing the corresponding cDNA as templates. For PCR, PrimeSTAR Max DNA Polymerase (Takara) was routinely used to amplify the desired DNA fragments, and the resulting PCR products were subcloned into appropriate vectors using In-Fusion (Takara) or using available restriction sites. Sequences of all plasmids were verified by DNA sequencing. Plasmids constructed in this paper are listed in the key resources

Protein labeling with fluorophores
Purified proteins were labeled with Cy3 or Cy5 maleimide mono-reactive dyes (Cytiva) in protein solution buffer at a concentration ratio of 1 mg protein/0.05 vial by incubating for 30 min at RT with gentle agitation every 10 min. Fluorophores and other small molecules were removed from proteins by passing them over a PD-10 desalting column (Cytiva) and protein concentrations were adjusted with an Amicon Ultra cartridge (0.5 mL, 10 kDa cut-off, Merck Millipore For evaluation of nanoscale co-localization of synaptic proteins, synapse structures in two-color 3D-STORM images were manually and randomly selected. Selected images were smoothed with a 3D Gaussian filter with a radius of 2 voxels (20 nm), and then binarized with the Otsu algorithm. Positive voxels were divided into connected components, and each component was then labeled as an individual cluster. The largest cluster in each synapse for each synaptic protein was extracted for further analysis while all smaller clusters were excluded from analysis. The center-of-mass of each protein cluster was quantified by calculating mean voxel coordinates. The volume of each cluster was quantified by counting total voxels of the extracted major cluster for each synaptic protein. Manders coefficients were quantified by calculating the ratio of voxel overlap (both positive) to total voxels of the reference protein cluster. Image data analysis was performed using Mathematica (Wolfram).

Live cell imaging of neural cultures
To monitor behaviors of EndoA1 and synapsin in response to electrical stimulation in neurons, neuronal cultures were transduced with lentiviral vectors encoding either EndoA1-EGFP or EGFP-synapsin 1 at 7 DIV. Lentiviral vectors were produced using tsA201 cells as hosts, as described previously. 54 Briefly, tsA201 cells were plated on 100-mm TC-treated cell culture dish (CORNING). After 2-3 h, cells were transfected with 17 mg of lentiviral backbone vector based on pFUGW and helper plasmids (pCAG-kGP1 10 mg, pCAG4-RTR2 5 mg and pCAG-VSVG 5 mg) by calcium phosphate transfection. 59 After 16 h, cultured medium was replaced with fresh neural culture medium. After another 48 h, supernatants were collected, centrifuged at 500 3 g for 15 min, and were filtered through a 0.45 mm filter (Millipore). Viral aliquots were flash frozen in liquid nitrogen and stored at À80 C until use. Image analysis was performed with Fiji software 55 using the Time Series Analyzer plugin. Acquired images were background subtracted using a rolling ball algorithm with a radius of 50 pixels. Image drift was corrected using the Fiji plugin (correct 3D drift tool). Circular regions of interest (ROIs, diameter = 1.08 mm) were manually positioned at the center of punctate fluorescence signals. EGFP fluorescence in ROIs was calculated as DF (F -F t = 15 )/F0. DF (F -F t = 15 )/F0 of $10 boutons from a single experiment (512 3 512 pixels) were averaged and expressed as n = 1. The mobile fraction is the absolute value of the DF (F -F t = 15 )/F0 at t = 75 of each experiment. To calculate the time constant of dispersion, the DF (F -F t = 15 )/F0 at t = 15-75 of each experiment was fitted with monoexponentially decay curve [y = A (1 -e Àt/t )] using Solver, Excel add-in program. To calculate the time constant of recovery, the DF (F -F t = 75 )/F0 at t = 75-320 of each experiment was fitted with a monoexponential decay curve [y = A (1 -e Àt/t )] using Solver, an Excel add-in program. Note that a time constant of dispersion or recovery that did not fit within the imaging time was excluded from all data analysis.

Study design
We confirmed that all experimental findings were reliably reproduced. Randomization and blinding were not used in this study. Data exclusions were clearly indicated with the exclusion criteria of the data in the relevant section. Sample size was determined based on the previous publications that included similar research methods (examples; Milovanovic et al., 4 Park et al., 6 and Lopez-Hernandez et al. 54 )

QUANTIFICATION AND STATISTICAL ANALYSIS
Data were analyzed with Excel 2016 and Python. All data are given as means G standard errors of the means (s.e.m). All statistical analyses were two-tailed, and levels of statistical significance are indicated by asterisks: *p < 0.05, **p < 0.01, ***p < 0.001. n.s.; not significant.