While the importance of actin in synaptic endocytosis is clear (Kononenko et al., 2014; Wang et al., 2010; Wu et al., 2016), until now there has been no quantitative analysis of individual actin-dependent membrane-remodeling events at synapses. To better understand presynaptic F-actin dynamics and to identify sites where the cytoskeleton- and membrane-remodeling machinery is active, we quantified individual F-actin assembly events by spinning disc confocal microscopy of NMJs presynaptically expressing fluorescent actin probes. To control for developmental variation, all experiments were performed on late third-instar larvae (~96–120 hr after egg laying) on muscle 6/7 NMJs at abdominal segments 3–4, since the development and physiology of these synapses are well characterized (Harris and Littleton, 2015). To control for variation in size between neurons, we normalized patch frequencies by the synapse area measured and presented data per 10 µm², which is approximately the size of a synaptic bouton in this system. We performed these experiments under resting conditions, where vesicle release is spontaneous at a rate of ~5–6 vesicles/10 µm²/min (Akbergenova et al., 2018; Melom et al., 2013), presumably requiring a similar rate of compensatory endocytosis (Sabeva et al., 2017).

We first compared the dynamics of three actin markers: GFP::actin, GFP-tagged moesin F-actin-binding domain (GMA), and Lifeact::Ruby. The predominant structures labeled by these markers were transient patches at the presynaptic membrane (Video 1, Figure 1A, Figure 1—figure supplement 1A), as has been previously observed (Nunes et al., 2006; Pawson et al., 2008; Piccioli and Littleton, 2014). We then quantified individual actin patch dynamics using automated particle tracking and quantification (Berro and Pollard, 2014; Tinevez et al., 2017), which captured on the order of 30–50% of visible actin structures (see 'Materials and methods', Figure 1—figure supplement 2, and Figure 6—figure supplement 1 for more details on optimization and validation of actin particle analysis). We first imaged at 0.25 Hz and measured an average of 1.2 GMA patches/10 μm²/min, exhibiting a mean duration of 48.0 s ± 45.6 s, with an average relative amplitude of 68 ± 32% ((I_max-I_min)/I_mean) (Figure 1B–D). Quantification of GFP::actin and Lifeact::Ruby showed very similar dynamics to GMA, suggesting that these measurements robustly reflect the underlying actin dynamics and not the specific properties of a particular probe. We did note a high percentage of patches in the minimum duration bin, suggesting the existence of even briefer patches. To address this, we also performed imaging at 1 Hz, which could not capture the entire lifetime distribution due to photobleaching but was able to identify a larger population of short-duration patches (Figure 1—figure supplement 1B) with an average duration of ~16 +/- 20 s. Given this range of measurements at different sampling frequencies and the efficiency of our automated detection, we estimate that patch frequency is between 2.8 and 10.3 events/10 µm²/min (see 'Materials and methods' for calculations), on par with the expected frequency of endocytic events, and with a similar albeit broader distribution of durations compared to yeast (15 s; Berro and Pollard, 2014) and mammalian cells (~40 s; Taylor et al., 2011).

We next examined the molecular determinants of synaptic actin patch assembly. Patches strongly co-labeled with Arp3::GFP (Pearson coefficient 0.70), significantly higher than the active zone marker Bruchpilot (BRP), which served as a punctate and membrane-localized negative control (Figure 1E–F). These data suggest that actin patches are predominantly composed of branched F-actin, similar to sites of endocytosis in other cell types (Akamatsu et al., 2020; Collins et al., 2011). To test whether synaptic actin patches require Arp2/3 activation, we analyzed patch dynamics in larvae lacking the Arp2/3 activator WASp. We compared a genomic mutant (Figure 1G–I), likely hypomorphic due to maternal contribution (Ben-Yaacov et al., 2001), to presynaptic depletion in neurons expressing WASp RNAi (Figure 1—figure supplement 1 and C–F). Using both approaches allows us to distinguish neuron-autonomous from non-autonomous effects of WASP, which is present both pre- and postsynaptically (Coyle et al., 2004). Both genomic and RNAi manipulations significantly reduced the number of actin patches, while genomic mutants also skewed the distribution of patch durations toward both shorter and longer events (Figure 1I). These differences could reflect variable loss of function between the RNAi and mutant, or identify separable presynaptic autonomous (patch frequency) vs non-autonomous (patch duration) effects of WASp. Overall, these data clearly indicate that WASp is autonomously required in neurons to initiate assembly of presynaptic actin patches, similar to its involvement in endocytosis in yeast, mammalian non-neuronal cells, and in the NMJ (Hussain et al., 2001; Kessels and Qualmann, 2004; Khuong et al., 2010; Madania et al., 1999).

We next examined the synaptic distribution of two likely WASp regulators, Nwk and Dap160. By conventional and super-resolution microscopy of neurons in diverse organisms, these and other presynaptic membrane-remodeling proteins localize to a broad membrane domain surrounding active zones, termed the periactive zone (PAZ) (Coyle et al., 2004; Denker et al., 2011; Gerth et al., 2017; Koh et al., 2004; Marie et al., 2004; Sone et al., 2000). Consistent with these prior descriptions, we observed by structured illumination microscopy (SIM) that the PAZ proteins Nwk and Dap160 localize to a membrane-proximal mesh that surrounds active zones, which were labeled with BRP (Figure 2A). We observed similar results by live imaging of an endogenously tagged Nwk protein by SIM, which revealed most proteins to be close to the plasma membrane (Figure 2B). We then compared the localization of PAZ proteins to F-actin patches at the NMJ. As expected, actin patches were much sparser than the endocytic machinery, and GMA-labeled patches only partially overlapped with each of the presynaptic WASp, Nwk, and Dap160 (Figure 2C–F; Pearson’s coefficients of 0.38, 0.38, and 0.36, respectively). These data confirm that, in sharp contrast to the actin regulatory machinery, which localizes broadly across the PAZ, actin assembly itself is much sparser both spatially and temporally at the NMJ. This raises the question of how PAZ machinery might itself be locally regulated to promote the formation of productive synaptic actin assemblies.

The hypothesis that PAZ protein-mediated actin assembly might be locally activated is particularly interesting given that we and others have previously shown that autoinhibition of both Nwk and its mammalian homolog FCHSD2 suppresses both WASp activation and membrane binding (see Figure 3A for summary model; Almeida-Souza et al., 2018; Rodal et al., 2008; Stanishneva-Konovalova et al., 2016). These results suggest that transient or localized relief of autoinhibition could explain how the PAZ controls actin assembly. To determine if and how the candidate activator Dap160 might relieve Nwk autoinhibition, we first mapped their specific interaction domains using glutathione-S-transferase (GST) pulldown assays and found that purified Dap160 SH3C-containing protein fragments (SH3C, SH3CD, or SH3ABCD) directly interact with Nwk^SH3b, while SH3D alone does not (Figure 3B, Figure 3—figure supplement 1; see Figure 3—figure supplement 2A for details of constructs used). Unexpectedly, Dap160 SH3C, SH3D, and SH3CD domain fragments also, each, interact with the isolated Nwk F-BAR domain (Figure 3—figure supplement 1B). We next determined how Dap160 interacts with Nwk^F-BAR compared to a Nwk fragment containing the F-BAR and both SH3 domains. Dap160-Nwk^F-BAR interactions were progressively eliminated by increasing salt, suggesting they are mediated by electrostatic interactions. By contrast, Dap160^SH3CD-Nwk interactions were maintained (Figure 3B, Figure 3—figure supplement 1C), suggesting that the SH3-SH3 interaction is mediated primarily by hydrophobic interactions, consistent with their mammalian homologs (Almeida-Souza et al., 2018; see summary of interactions in Figure 3C). Finally, we found that truncation of Dap160^SH3CD decreased the levels of Nwk in synaptic boutons similarly to Dap160 knockdown (Figure 3D, Figure 3—figure supplement 2B–C). Dap160^ΔSH3CD also exhibited reduced colocalization with Nwk compared to wild-type Dap160 (Figure 3E, Figure 3—figure supplement 2C), further supporting an in vivo requirement for this interaction. Notably, truncation of Dap160^SH3D did not exhibit a phenotype in these assays despite lower levels of expression (Figure 3—figure supplement 2B), suggesting that additional factors absent from our in vitro assays may collaborate to regulate Nwk in vivo.

We have previously shown that Nwk only weakly activates WASp-dependent actin assembly in vitro, due to Nwk autoinhibition (Stanishneva-Konovalova et al., 2016). To test whether Dap160^SH3CD might relieve Nwk autoinhibition, we performed pyrene-actin assembly assays (Figure 4). At moderate Nwk-Dap160 concentrations (500 nM and 2 µm, respectively), Nwk and Dap160^SH3CD significantly enhanced the rate of WASp-Arp2/3-mediated actin assembly compared to Nwk plus WASp alone (Figure 4A). This effect is through Nwk, as Dap160^SH3CD had no effect on WASp-Arp2/3 in the absence of Nwk. Further, Dap160 enhancement of Nwk-WASp actin assembly required the Dap160^SH3D domain, further showing that the specific Dap160^SH3D-Nwk^F-BAR interaction relieves functional Nwk autoinhibition in vitro. Thus, multiple Nwk-Dap160 interactions work together to relieve autoinhibition of Nwk.

To generate salient physiological force, actin assembly must be coupled to membranes, and negatively charged lipids are an important ligand for both Nwk and WASp. Thus, we next tested whether addition of PI(4,5)P₂-rich liposomes modified actin assembly by Nwk, Dap160, and WASp (Figure 4B). Indeed, PI(4,5)P₂-containing liposomes synergistically activated WASp-mediated actin assembly in concert with Dap160 and Nwk. By contrast, neither Nwk, PI(4,5)P₂, nor Nwk + PI(4,5)P₂ on their own were sufficient to activate WASp above baseline (Figure 4B). Since PI(4,5)P₂ is also insufficient to robustly activate either WASp or Nwk under these conditions (Stanishneva-Konovalova et al., 2016), our data suggest that WASp activation reflects coordinated relief of Nwk autoinhibition by both Dap160 and membranes. To further explore the coupling between lipid association and actin assembly, we conducted F-actin assembly assays in a droplet assay, in which protein-containing aqueous droplets are surrounded by a lipid interface, with lipid head groups facing the aqueous phase (Figure 4C). In this assay, we found that coordinated interactions among Nwk, Dap160, and WASp directed actin assembly to the lipid interface. By contrast, substitution of Nwk lacking its autoinhibitory and Dap160-interacting SH3b domain (Nwk^ΔSH3b) caused actin to assemble as free-floating asters (Figure 4C). We have previously found that expression of a similarly deregulated fragment (Nwk^1-631) at the NMJ led to diffuse actin filament assembly throughout the synapse (Stanishneva-Konovalova et al., 2016). Together, these data suggest that Nwk^SH3b has a dual role in maintaining autoinhibition via Nwk-F-BAR interactions and permitting actin assembly at specific synaptic locations via Dap160-mediated activation.

Our actin assembly data suggest that membrane recruitment is a critical regulator of the Nwk-Dap160-WASp complex (Figure 4B–C). To test whether Nwk-Dap160 interactions directly regulate membrane recruitment, we performed liposome cosedimentation assays. We found that Dap160^SH3CD enhanced Nwk membrane binding in a dose-dependent fashion (Figure 5A). This effect depended on membrane charge, as Dap160^SH3CD significantly enhanced Nwk membrane binding at both 5 and 10%, but not at 2.5% PI(4,5)P₂ (Figure 5B). Only at 10% PI(4,5)P₂ did Dap160^SH3CD promote Nwk membrane binding to the same extent as the completely uninhibited Nwk^FBAR domain alone, suggesting that membrane charge and intermolecular interactions with Dap160 together tune Nwk membrane recruitment. Indeed, this effect required the full Dap160^SH3CD-Nwk^SH3b interaction: Dap160^SH3C alone was unable to promote membrane binding by Nwk, and Dap160^SH3CD did not enhance membrane binding of Nwk lacking its Dap160-interacting SH3b domain (Figure 5—figure supplement 1A). These data further support the hypothesis that Dap160^SH3CD relieves Nwk^SH3b-mediated autoinhibition.

As we found that Dap160^SH3CD is insufficient to fully activate membrane binding by Nwk at intermediate phosphoinositide concentrations (Figure 5A), we asked whether WASp could further enhance Nwk membrane recruitment. Indeed, the addition of Dap160^SH3CD and WASp together enhanced Nwk membrane association to the level of the isolated F-BAR domain (Figure 5C). Moreover, coordinated binding of all three components resulted in significantly enriched membrane association of both WASp and Dap160 (Figure 5C). We directly observed the coordinated recruitment of Nwk and Dap160 in the presence of WASp using fluorescently labeled proteins on GUVs (Figure 5D). Consistent with the direct Dap160-Nwk^SH3b interaction, we found that deletion of the Nwk^SH3b domain abolished both the Dap160^SH3CD-dependent increase and the coordinated recruitment of WASp and Dap160 (Figure 5—figure supplement 1A). Notably, addition of Dap160 and WASp did not change the nature of membrane deformations generated by Nwk (scalloped and pinched membranes; Becalska et al., 2013), suggesting that Dap160 and WASp together potentiate rather than alter the inherent activity of Nwk (Figure 5—figure supplement 1D). These data indicate that Dap160-Nwk SH3-mediated interactions potentiate Nwk association with membranes in vitro.

Finally, to test whether Dap160 promotes Nwk membrane association in vivo, we examined the dynamics of Nwk at the synapse in the presence and absence of Dap160. Knockdown of Dap160 by RNAi (Figure 5E, Figure 3—figure supplement 2D) led to a striking loss of endogenously tagged Nwk::GFP from synaptic membranes (note strong peripheral labeling in control bouton cross-sections; Figure 5E). Further, Dap160 knockdown significantly increased the rate of recovery of Nwk::GFP after photobleaching, consistent with a shift in localization from membrane-bound to cytosolic (Figure 5F). These data suggest that the Dap160^SH3CD-Nwk interaction promotes Nwk membrane association in vivo. Taken together, our data indicate that multiple coordinated interactions between Nwk, WASp, Dap160^SH3CD, and membranes are required to relieve Nwk autoinhibition, allowing for tight control of membrane-coupled actin assembly in the PAZ.

To determine how these mechanisms direct WASp-mediated actin assembly at the synapse, we measured actin dynamics in nwk (Figure 6A–C, Video 2) and dap160 domain (Figure 6D–F) mutant NMJs. We predicted two possible but non-exclusive functions based on the dual roles that we found for the Nwk-Dap160-WASp module in vitro: if Nwk and Dap160 are primary activators of WASp, then loss-of-function mutants are likely to diminish patch frequency, duration, or intensity. Importantly, multiple WASp activators exist in the synaptic endocytic machinery (e.g., Cip4 and Snx9; Almeida-Souza et al., 2018; Gallop et al., 2013; Nahm et al., 2010; Ukken et al., 2016), and therefore, these could make significant contributions to WASp activation in addition to Nwk. Conversely, if an important function of autoinhibition is to ‘clamp’ actin assembly at the synapse, we expected that loss of Nwk and/or Dap160 would lead to spurious actin assembly events by these other WASp regulators. We found that both nwk and Dap160^ΔSH3CD mutants significantly increased patch frequency (Figure 6B,E, Figure 6—figure supplement 1), supporting a clamp function for these proteins. We did not detect a difference in the distribution of patch lifetimes, suggesting that it is the frequency of events, and not their duration per se, that changes (Figure 6C,F).

We also analyzed actin dynamics using a complementary approach in which we measured the normalized intensity variation (coefficient of variation, CoV) over time across the entire NMJ. Interestingly, the magnitude of variation was significantly higher in nwk mutants (Figure 6—figure supplement 2A–B), but the area of the NMJ that was highly variant was similar between genotypes, suggesting that actin assembly is more dynamic in time in these mutants, rather than more extensive in space (Figure 6—figure supplement 2C). We validated this analysis for its sensitivity in detecting changes in event frequency by analyzing synthetic data (Figure 6—figure supplement 2D, see 'Materials and methods' for details). The modeled data suggest that the difference in CoV that we measured between Control and nwk is consistent with a 43% increase in patch frequency, which is slightly higher than our measurement by particle tracking (28%; Figure 6A). This complementary analysis does not rely on particle tracking and makes no assumptions about the nature of actin dynamics, and is consistent with our particle-based metrics. Thus, we conclude that these phenotypes are robust to the method of analysis used.

We next investigated the physiological function of actin patches in vivo. Considering that patch morphology, frequency, and duration resembled endocytic dynamics, we first compared actin patches with the endocytic adaptors Clc and AP2α. Like other endocytic proteins, both presynaptically expressed Clc::GFP and endogenously tagged AP2α::GFP were primarily enriched at the plasma membrane relative to the cytoplasm (Figure 7—figure supplement 1A) and covered a large area fraction of the membrane, similar to other endocytic proteins (Figure 2). In addition to diffuse signal, both probes localized to short- and long-lived puncta, a subset of which dynamically colocalized with actin patches (Figure 7A–C, Figure 7—figure supplement 1B, Video 3). A significant proportion of endogenously labeled AP2 at the NMJ is likely associated with the closely apposed postsynaptic membrane, which accounts for its slightly lower correlation coefficient with Lifeact::Ruby. Considering that the rates of exo/endocytosis at this synapse at rest are relatively low (see above), these observations suggest that like other PAZ endocytic proteins, a large pool of membrane-localized clathrin coat and adaptor proteins are not actively engaged in endocytosis. Despite these caveats, we found that actin significantly colocalized with both Clc (Figure 7C) and AP2 (Figure 7—figure supplement 1C), consistent with a role in endocytosis for these actin-enriched sites. To more rigorously and functionally test the hypothesis that actin patches are endocytic, we acutely disrupted endocytic dynamics using the temperature-sensitive dominant-negative dynamin/shi^TS1 allele. When imaged under restrictive conditions, shi disruption decreased the frequency of actin patch dynamics (Figure 7D–E). Together, these data suggest that a significant fraction of presynaptic actin patches are associated with endocytosis.

We next tested the physiological requirement of the Nwk and Dap160^SH3CD interaction. As both Nwk and Dap160 are implicated in the endocytic trafficking of synaptic growth-promoting bone morphogenetic protein (BMP) receptors (O'Connor-Giles et al., 2008; Rodal et al., 2008), we tested whether the Dap160-Nwk interaction was required for normal synaptic growth, which we assayed by counting satellite boutons, a hallmark phenotype of both null mutants. Surprisingly, we found that both Dap160^ΔSH3D and Dap160^ΔSH3CD truncations rescued satellite bouton numbers to wild-type levels (Figure 7—figure supplement 2). These data indicate that synaptic vesicle and growth factor endocytosis are mechanistically separable, and suggest that actin dynamics phenotypes in the Dap160^ΔSH3CD mutant are not associated with synaptic growth regulation. We next examined synaptic vesicle endocytosis and recycling by FM dye uptake. nwk^1/2 null mutants led to a 34% decrease in FM4-64fx uptake compared to controls (Figure 7F–G), an intermediate phenotype compared to dominant negative dynamin in shi^TS1 mutants (72% decrease). dap160 null mutants have been previously shown to exhibit an endocytosis defect (Koh et al., 2004; Marie et al., 2004), so we next tested whether the interaction between Dap160 and Nwk is required to support normal endocytosis. Indeed, we found that expression of Dap160^ΔSH3CD in dap160 null mutants also significantly diminished FM dye uptake to a similar extent as loss of nwk (27% reduction; Figure 7H–I). By contrast, loss of the Dap160^SH3D domain alone caused no defects in FM uptake, consistent with the lack of effect of this mutation on Nwk accumulation and localization (Figure 3D–E), suggesting that this interaction, though required in vitro, may be compensated by additional factors in vivo. Both nwk and Dap160^ΔSH3CD mutants unloaded FM dye to the same extent as controls, suggesting that diminished endocytosis is a direct phenotype, and not secondary to exocytic deficits (Figure 7—figure supplement 3). Importantly, these data indicate that spurious actin assembly events in nwk and dap160 mutants are likely to be unproductive for normal endocytosis. Overall, our data support the hypothesis that normal synaptic actin patches represent active endocytic events and indicate that Dap160-Nwk regulation of actin patch dynamics is functionally required for synaptic vesicle endocytosis.

Here we provide the first quantitative analysis of the composition and dynamics of individual presynaptic F-actin structures. Numerous studies have examined actin dynamics at the level of entire synapses or qualitatively described dynamics of discrete actin structures (Bloom et al., 2003; Colicos et al., 2001; Nunes et al., 2006; Piccioli and Littleton, 2014; Sankaranarayanan et al., 2003; Zhao et al., 2013), and identified diverse roles for actin, including synaptic vesicle endocytosis (Holt et al., 2003; Kononenko et al., 2014; Soykan et al., 2017; Watanabe et al., 2013; Wu et al., 2016; Zhao et al., 2013), synaptic vesicle organization and mobilization (Guzman et al., 2019; Lee et al., 2012; Marra et al., 2012; Owe et al., 2009; Sakaba and Neher, 2003; Wolf et al., 2015), active zone organization and function (Pilo Boyl et al., 2007; Morales et al., 2000; Wagh et al., 2015; Waites et al., 2011; Wang et al., 1999), and receptor-mediated endocytosis (Kim et al., 2019; Rodal et al., 2008). Bulk analyses, which do not separate individual dynamic actin structures in space and time, are limited in their ability to discern how the regulation and dynamics of actin contribute to these distinct functions. We leveraged our ability to extract data describing individual structures to find that synaptic actin predominantly assembled into discrete Arp2/3-associated patches, and identified points of control over their dynamics. Specifically, we found that loss of endocytic proteins differentially affected the frequency and kinetics of individual actin patches, which correlate with functional deficits in endocytosis.

The link between the actin structures that we observed and endocytic events is supported by several lines of evidence: the morphology and duration of synaptic actin patches are similar to WASp/Arp2/3-dependent endocytic actin dynamics in yeast (16 s; Berro and Pollard, 2014) and somewhat briefer than in cultured mammalian cells (~40 s; Grassart et al., 2014; Merrifield et al., 2004; Taylor et al., 2011). Given the measured time constant for endocytosis (~14 s; Poskanzer et al., 2006) and clathrin dependence of vesicle cycling in this synapse (Heerssen et al., 2008), these values support the hypothesis that synaptic actin patches are likely sites of clathrin-mediated endocytosis. Further, the frequency of patch assembly, which we measured in resting synapses, approaches the rate of spontaneous synaptic vesicle release at this synapse (Figure 1—figure supplement 1B, Figure 1—figure supplement 2A) (~5–6/10 μm²/min; Melom et al., 2013; Akbergenova et al., 2018). Further, actin patches colocalize partially with endocytic adaptors, and their assembly is sensitive to disruption of endocytosis (Figure 7). Finally, we found that the same endocytic proteins and protein interactions that regulate endocytosis at this synapse also alter the dynamics of actin patches.

Technical challenges due to the high density of endocytic proteins and synaptic vesicle cargoes, and the difficulty of conducting sparse single vesicle measurements at this synapse (compared to neurons in culture; Chanaday and Kavalali, 2018; Peng et al., 2012), make it difficult to directly link the dynamics of actin structures to specific membrane or cargo internalization events. However, the frequency of the events captured by our approach makes it unlikely that they represent rare F-actin-dependent events at this synapse, such as those that control macropinocytosis or new bouton growth (Khuong et al., 2010; Kim et al., 2019; Piccioli and Littleton, 2014), and more likely that they represent bona fide endocytic events. Thus, while we do not rule out other biological functions for a subset of patches, together our data indicate that a significant and measurable fraction of synaptic actin patches are associated with endocytosis.

Many endocytic proteins accumulate across broad membrane domains at the Drosophila NMJ and other synapses (Gerth et al., 2017; Guichet et al., 2002; Koh et al., 2007; Roos and Kelly, 1998; Verstreken et al., 2008, Verstreken et al., 2003). Our data indicate that much of this membrane-remodeling machinery is likely held in an inactive state at the presynaptic membrane: Nwk and Dap160 accumulate in a micron-scale membrane domain (Figure 2), and their loss increases the frequency of short-lived actin patches (Figure 6). These data suggest that these PAZ proteins are held in a partially autoinhibited state at the membrane in vivo, consistent with our prior in vitro observations (Stanishneva-Konovalova et al., 2016). This is further consistent with the broad distribution of Clc and AP2 to the membrane (Figure 7, Figure 7—figure supplement 1). Given the comparatively low rate of endocytosis expected at rest at this synapse, this suggests that most Clc and AP2 puncta at the synapse are either not stabilized to form productive endocytic sites (Aguet et al., 2013) or associated with some non-endocytic functions (Gimber et al., 2015).

The fact that loss of Nwk increases the frequency of patches while decreasing FM uptake suggests that the actin structures assembled in the nwk mutant are unproductive for synaptic vesicle endocytosis. These spurious patches could reflect non-specific actin assembly, perhaps due to unmasking of the Nwk ligand PI(4,5)P₂ at the membrane and/or inappropriate activation of alternative WASp-dependent actin assembly pathways. Indeed, additional WASp activators such as Snx9 and Cip4/Toca-1 may play accessory roles in endocytic actin assembly (Almeida-Souza et al., 2018; Gallop et al., 2013), consistent with our finding that loss of presynaptic WASp leads to a decrease in the total number of patches (Figure 1G–I, Figure 1—figure supplement 1C–E). Our data indicate that at the synapse, where endocytic machinery accumulates at high concentrations (Wilhelm et al., 2014) and recruitment appears uncoupled from activation, these layers of autoregulation are critical to constrain actin assembly generally.

Our findings on autoinhibition and clamping connect two prevailing models of the organization and function of the synaptic endocytosis machinery—preassembly and delivery. In the first model, preassembly of clathrin and accessory proteins is hypothesized to ensure fast endocytosis in response to synaptic vesicle fusion (Hua et al., 2011; Mueller et al., 2004; Wienisch and Klingauf, 2006). Here, Nwk autoinhibition provides a mechanism to assemble an inactive, yet poised endocytic apparatus. In the second model, endocytic machinery associates with the synaptic vesicle pool, providing a ready source or buffer of proteins that can be released to the plasma membrane upon calcium signaling or vesicle fusion (Bai et al., 2010; Denker et al., 2011; Gerth et al., 2017; Winther et al., 2015). Because Dap160/intersectin can shuttle between the synaptic vesicle pool and the plasma membrane, is itself subject to autoregulation (Gerth et al., 2017), and can regulate other endocytic proteins (e.g., dynamin, Nwk), it could serve as a single activator that couples the preassembly and delivery models.

Our in vitro data show that beyond functioning as a clamp, Nwk and Dap160 collaboratively activate WASp to promote robust actin assembly. Together with the defects we observed in vivo for actin dynamics and FM dye uptake, these data suggest that Dap160-Nwk-WASp interactions could serve as a coincidence detection mechanism to relieve autoinhibition of Nwk and promote productive actin assembly with other WASp activators. Coincidence detection has been demonstrated in several systems to control membrane-associated actin assembly (Case et al., 2019; Sun et al., 2017), suggesting that amplification of WASp membrane binding could drive robust actin patch assembly at synapses. Similarly, in human cells, the interaction between FCHSD2, intersectin, and WASp promotes actin assembly and endocytic maturation (Almeida-Souza et al., 2018) or initiation (Xiao et al., 2018). The Dap160-Nwk module could act by directing and/or organizing actin assembly specifically at endocytic events, akin to the membrane-directed actin assembly we observed in vitro (Figure 5D), and/or ensure that it is sufficiently robust for productive membrane remodeling (Akamatsu et al., 2020). Direct support for these models will require new analytical or imaging approaches to directly visualize the coupling of membranes and actin to the endocytic machinery, in order to distinguish spurious (due to unclamping) vs bona fide but underpowered endocytic actin assembly events.

Our data suggest that the endocytic machinery can be deployed as clamped, primed, or activated complexes at different locations at the synapse. The next critical step will be to determine the mechanisms that control switching between these states. Many potential mechanisms that link calcium-dependent exocytosis and endocytosis could activate actin assembly. These include direct effects of calcium on the endocytic machinery (Maritzen and Haucke, 2018), the accumulation of synaptic vesicle cargoes (Cousin, 2017), stoichiometry-dependent changes in protein interactions or activities (Case et al., 2019), changes in membrane mechanics (Anantharam et al., 2010; Dai et al., 1997; Roux et al., 2010), and changes in membrane charge/mode of membrane binding (Kelley et al., 2015). One intriguing possibility is that these mechanisms might enable an endocytic PAZ to rapidly switch between different modes of endocytosis (e.g., ultrafast, conventional, or bulk) in response to a wide range of synaptic activity patterns (Gan and Watanabe, 2018). These endocytic regulatory mechanisms could also be locally poised to regulate, respond, or adapt to the specific release properties of nearby active zones (Akbergenova et al., 2018; Melom et al., 2013; Dickman et al., 2006), and serve as novel points of control over synaptic plasticity and homeostasis.

Spinning disc confocal time series were acquired at 15 stacks/min (Figure 1), 60 stacks/min (Figure 6, Figure 7E), or 2.2 stacks/min (Figure 7A). A maximum intensity projection was made of each time point, videos were registered using the FIJI plugin StackReg, and analyzed by Patchtracker, based on Trackmate (Berro and Pollard, 2014) as follows. First, we qualitatively evaluated the optimal intensity threshold for patch detection by identifying the maximum threshold intensity at which all obvious patch structures were detected in the first frame of videos. This process was performed independently by three independent observers over multiple datasets. The threshold for patch detection was normalized to the mean probe intensity in the presynaptic area (threshold = Probe Mean * 0.32). All other settings for patch detection and tracking were default: estimated patch diameter = 0.6 μm, median filter = false, subpixel detection = true, linking max distance = 0.5 μm, gap-closing distance = 0.5 μm, gap-closing frame gap = 0. For 0.25 Hz imaging experiments, patches between 16 and 356 s could be detected. For 1 Hz imaging experiments, patches between 4 and 139 s could be detected. Because this analysis rejects a significant number of detected patches due to tracking defects or tracking path overlap, we estimated the true patch frequency as follows. We combined detections from 0.25 Hz and 1 Hz imaging experiments by averaging the frequencies over the shared detection range (20–150 s) and adding the lower and higher duration patches that were specific to each imaging regime (4–16 s for 1 Hz and 150–360 s for 0.25 Hz). Then we ‘corrected’ for rejected tracks and considered the lower bound of the estimate to be the actual, corrected merged frequency of detection (2.8 patches/10 μm/min) and the upper bound to include every rejected track (10.3 patches/10 μm/min).

We further validated our patch dynamics analysis by measuring patch frequencies at a wide range of patch intensity thresholds and track linking distances. For both 0.25 Hz WASp (Figure 1—figure supplement 2) and 1 Hz Nwk (Figure 6—figure supplement 2) datasets, we found measurements of control patch frequencies to be robust to these parameters and in strong agreement with the estimates described above across the entire parameter space tested (1.1–8.4 for 0.25 Hz imaging, 1.2–7.9 for 1 Hz imaging). Further, our phenotypic analyses (decreased patch frequency in WASp mutants and increased frequency in Nwk mutants) were also both highly robust to tracking parameters.

Actin dynamics were also analyzed by measuring intensity variation over time over the entire NMJ, that is, without thresholding or particle tracking. We measured this by extracting the intensity value for each pixel over time and calculating the CoV (Std Dev/Mean) for each pixel. We estimated the percentage of ‘highly variant’ pixels by thresholding these values using Li (Li and Tam, 1998) and Moments (Tsai, 1995) algorithms. While these two algorithms gave different estimates of the fraction of NMJs covered by highly variant pixels, both indicated the same relationship between genotypes. To validate this approach, we created synthetic data using a custom FIJI script, with a spatial and temporal scale that matched our in vivo imaging, and in which we varied parameters expected to impact this metric (signal intensity, noise level, fraction of dynamic pixels, dynamics frequency, dynamics duration, dynamics amplitude), and subjected the synthetic data to the same CoV over time analysis.

For intensity and colocalization, the presynaptic region was masked in 3D using a presynaptically enriched label: either HRP (Figure 3—figure supplement 2D), Nwk (Figure 2A, Figure 2C), Dap160 (Figure 2E, Figure 3D–E, Figure 3—figure supplement 2B–C), or Lifeact::Ruby (Figure 1E, Figure 7C. Figure 7—figure supplement 1C). For mask generation, images were subjected to a gaussian blur filter and thresholded by intensity. Blur radius and the specific threshold algorithms used were empirically optimized for each experiment to consistently and accurately reflect the presynaptic area in control and mutant groups (and the same settings were used for all NMJs within any given experiment). Signal intensities were measured in 3D using a FIJI script, and colocalization analysis was performed in 3D on Airyscan or SIM reconstructed image stacks using the Coloc2 plugin for ImageJ (https://imagej.net/Coloc_2). For all images, background was subtracted using the rolling ball method with a radius of 50 pixels.

Coprecipitation with GST-tagged proteins was conducted as described previously (Kelley et al., 2015). Concentrations of GST fusions on beads were normalized using empty beads and bead volume was restricted to two-thirds of the total reaction volume. GST fusions were incubated by agitation with His-tagged target proteins at room temperature for 1 hr in binding buffer (20 mM Tris, pH 8.0, 20 mM KCl, 0.5 mM DTT). For salt sensitivity experiments, the indicated concentrations of NaCl were used in place of KCl in the binding buffer. Beads were then pelleted and washed once with buffer after removing the supernatant. Pellets and supernatants were then boiled in Laemmli sample buffer and fractionated by SDS-PAGE, followed by Coomassie staining or immunoblotting as noted in figure legends, followed by imaging and analysis on a LICOR Odyssey device.

Lipid cosedimentation assays were conducted as described previously (Becalska et al., 2013). In brief, liposomes were swelled from dried lipid films in 20 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), pH 7.5, and 100 mM NaCl. Specific lipid compositions are indicated in the figure legends. Proteins were then mixed with 1 mg/ml liposomes, incubated for 30 min at room temperature, and then pelleted for 20 min at 18,000 ×g at 4°C. Pellets and supernatants were then denatured in Laemmli sample buffer and fractionated by SDS-PAGE, followed by Coomassie staining, and imaging and analysis conducted on a LICOR Odyssey device.

GUVs were generated by gentle hydration. Briefly, 10 µl of 10 mg/ml lipids dissolved in 19:1 chloroform:methanol were dried under vacuum, and then swelled in 300 µl of 5 mM HEPES 300 mM sucrose, pH 7.5, overnight at 70°C. GUVs were imaged on a Marianas spinning disc confocal system (see above). 3 µl GUVs were diluted into 5 mM HEPES 150 mM KCl, pH7.5, incubated with protein as noted in figures, and imaged using a x100/NA 1.4 objective in multiwell slides (Lab-Tek) precoated with 1 mg/ml bovine serum albumin. After 30 min of incubation, 1% agarose in 5 mM HEPES 150 mM KCl, pH 7.5, was added (final agarose concentration, 0.5%) to limit GUV mobility. Images were analyzed by line tracing intensity profiles across a medial optical section of GUVs.

Lipids (97.5% DPHPC [1,2-diphytanoyl-sn-glycero-3-phosphocholine] [Avanti Polar Lipids] and 2.5% DPHPC:PI(4,5)P₂) were mixed in chloroform, dried under vacuum, and rehydrated to 23 mM (20 mg/ml) in decane. The indicated proteins were added to the lipid mix at 1:50 vol ratio and pipetted vigorously until cloudy before imaging by spinning disc confocal microscopy.

Rabbit muscle actin [5% (mol/mol) pyrene-labeled] was gel-filtered, prespun at 90,000 xg, exchanged from Ca²⁺ to Mg²⁺, and assembled at a final concentration of 2.5 µM as described previously (Moseley et al., 2006). Proteins were preincubated with 74 µg/ml liposomes or control buffer for 30 min before actin assembly reactions. Assembly was monitored with a spectrofluorometer (Photon Technology International) using an excitation wavelength of 365 nm and an emission wavelength of 407 nm. Rates were calculated from slopes of curves in the linear range, and curves were plotted using GraphPad Prism software.

Graphs were prepared and statistical analyses performed using Graphpad Prism software. For normally distributed data, comparisons were made using either t-test or analysis of variance with posthoc Bonferroni’s multiple comparisons test. For non-normally distributed data, comparisons were made using either Mann-Whitney U test or Kruskal-Wallis test with posthoc Dunn’s test. No specific power analyses were performed; sample sizes were chosen based on established protocols and statistical analyses for significance, as detailed for all experiments here and in Supplementary file 1. Comparison of patch-duration distributions was performed using a Kolmogorov-Smirnoff test. Please see Supplementary file 1 for each statistical test performed for each experiment presented in this study. All data are shown as the mean ± sem. Statistical significance denoted in all graphs *p<0.05, **p<0.01, ***p<0.001.