A positive feedback loop between Flower and PI(4,5)P2 at periactive zones controls bulk endocytosis in Drosophila

Synaptic vesicle (SV) endocytosis is coupled to exocytosis to maintain SV pool size and thus neurotransmitter release. Intense stimulation induces activity-dependent bulk endocytosis (ADBE) to recapture large quantities of SV constituents in large endosomes from which SVs reform. How these consecutive processes are spatiotemporally coordinated remains unknown. Here, we show that Flower Ca2+ channel-dependent phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) compartmentalization governs control of these processes in Drosophila. Strong stimuli trigger PI(4,5)P2 microdomain formation at periactive zones. Upon exocytosis, Flower translocates from SVs to periactive zones, where it increases PI(4,5)P2 levels via Ca2+ influxes. Remarkably, PI(4,5)P2 directly enhances Flower channel activity, thereby establishing a positive feedback loop for PI(4,5)P2 microdomain compartmentalization. PI(4,5)P2 microdomains drive ADBE and SV reformation from bulk endosomes. PI(4,5)P2 further retrieves Flower to bulk endosomes, terminating endocytosis. We propose that the interplay between Flower and PI(4,5)P2 is the crucial spatiotemporal cue that couples exocytosis to ADBE and subsequent SV reformation.


Introduction
Proper synaptic vesicle (SV) exocytosis dictates the robustness of brain activity. Coupling SV exocytosis with proper endocytosis is crucial for maintaining a balance of SV proteins at the release site, plasma membrane equilibrium, SV identity, and SV pool size (Chanaday et al., 2019;Haucke et al., 2011;Lou, 2018;Wu et al., 2014a). Currently, four modes of SV endocytosis are proposed. These differ in terms of stimulation intensity for their induction, formation, and molecular components (Haucke et al., 2011;Kononenko and Haucke, 2015;Wu et al., 2014a). Under mild neuronal stimulation, the SV partially fuses with the plasma membrane and reforms at the active zone, the so called 'kiss and run' mode. During clathrin-mediated endocytosis (CME), the SV fully collapses into the plasma membrane, followed by reformation of a single SV. Ultrafast endocytosis was also shown to recycle SVs at a sub-second timescale by forming a~80 nm-sized bulk endosome predominantly at the edge of the active zone. SVs subsequently regenerate from this bulk endosome (Granseth et al., 2006;Watanabe et al., 2013a;Watanabe et al., 2013b;Zhu et al., 2009). High frequency stimulation and thus exocytosis could easily surpass the capacity of the three abovedescribed endocytic mechanisms. It has therefore been proposed that activity-dependent bulk endocytosis (ADBE) has the necessary recapture capacity upon intense stimulation (Clayton et al., 2008;Soykan et al., 2017;Wu and Wu, 2007). This retrieval mode is elicited at the periactive zone to recapture large quantities of SV constituents via bulk endosome (~100-500 nm) formation, from which SVs subsequently reform. Hence, specific routes of SV recycling may fit the specific demands of a wide range of neuronal activities at the synapse. However, how stimulation intensity dictates the choice between these different endocytic modes is not well understood.
Coordinated SV protein and membrane retrieval plays an important role in maintaining the identity of newly formed SVs during SV recycling (Kaempf and Maritzen, 2017;McMahon and Boucrot, 2011;Saheki and De Camilli, 2012;Traub and Bonifacino, 2013). It has been well documented that proper sorting of SV proteins to the nascent SV is achieved in CME by the cooperative action of PI(4,5)P 2 and adaptor protein complexes (Saheki and De Camilli, 2012;Traub and Bonifacino, 2013). Recent studies have also revealed a distinct sorting mechanism that retrieves selective SV cargoes to the bulk endosome via ADBE (Kokotos et al., 2018;Nicholson-Fish et al., 2015). Several lines of evidence further suggest that clathrin and adaptor protein complexes are required for reforming SVs from the bulk endosome (Cheung and Cousin, 2012;Glyvuk et al., 2010;Kokotos et al., 2018;Kononenko et al., 2014;Park et al., 2016). A dynaminI/dynaminIII/clathrinindependent mechanism has also been reported as being involved in this process (Wu et al., 2014c). Thus, during SV regeneration via ADBE, multiple protein sorting steps may be required to ensure that the SVs harbor the proper compositions of lipids and proteins, thereby endowing specific release probabilities in relation to other modes of endocytosis (Cheung et al., 2010;Hoopmann et al., 2010;Nicholson-Fish et al., 2015;Silm et al., 2019). The mechanism by which protein sorting and membrane retrieval are coordinated in this process remains to be explored.
Here, we show that, upon intense stimulation, PI(4,5)P 2 is compartmentalized into microdomains at periactive zones in the synaptic boutons of Drosophila larval neuromuscular junctions (NMJs). Blockade of PI(4,5)P 2 microdomain formation diminishes ADBE and SV reformation from the bulk endosome. Increased intracellular Ca 2+ and SV exocytosis are prerequisites for initiating ADBE (Morton et al., 2015;Wu and Wu, 2007). We have previously shown that Flower (Fwe), a SV-associated Ca 2+ channel, regulates both CME and ADBE, and that its channel activity is strongly activated upon intense stimulation to elicit ADBE (Yao et al., 2017). We show that Fwe initiates a positive feedback loop upon PI(4,5)P 2 increase to ensure the formation of PI(4,5)P 2 microdomains and thus trigger ADBE and subsequent SV reformation. Intriguingly, PI(4,5)P 2 also participates in retrieval of Fwe to the bulk endosome, thereby stopping membrane recycling. Hence, spatiotemporal interplays between Flower and PI(4,5)P 2 coordinate the retrieval of SV cargos and membranes, coupling exocytosis to ADBE and subsequent SV reformation.

Results
Intense neuronal activity induces formation of PI(4,5)P 2 microdomains at the presynaptic periactive zone of Drosophila synapses To investigate the dynamics of PI(4,5)P 2 in the presynaptic compartment, we expressed a GFP fusion protein of the pleckstrin homology (PH) domain of PLC d1 (PLC d1 -PH-EGFP) in synaptic boutons of Drosophila larval NMJs using nSyb-GAL4, a pan-neuronal driver. PLC d1 -PH-EGFP binds to PI(4,5)P 2 with high affinity and is widely used to label subcellular compartments in which PI(4,5)P 2 is enriched Khuong et al., 2010). We delivered 20 Hz stimuli for three minutes to synaptic boutons in a 2 mM extracellular Ca 2+ solution and performed live imaging in the third minute. Consecutive snapshot images were taken before stimulation, during the third minute of stimulation, and after stimulation. As shown in Figure 1a-b, we observed a very subtle increase in PLC d1 -PH-EGFP fluorescence in individual boutons (white arrows), similar to findings of a previous study (Verstreken et al., 2009). However, when we raised the stimulus intensity to 40 Hz, we recorded a robust increase in fluorescence relative to a GFP fusion protein of the plasma membrane-integrated mCD8 domain (UAS-mCD8-GFP). Fluorescence signals rapidly returned to basal levels within tens of seconds when the stimuli were removed. We have previously documented that treatment with 40 Hz electric pulses or 90 mM high KCl solution can cause comparable stimulation intensities in Drosophila NMJ boutons (Yao et al., 2017). High K + treatment also increased the fluorescence signal of PLC d1 -PH-EGFP. No increase in the presynaptic protein level of PLC d1 -PH-EGFP was found under this condition (Figure 1-figure supplement 1a-b), arguing that this stimulation does not induce protein synthesis. These results suggest that, in response to intense stimulation, PLC d1 -PH-EGFP is redistributed and concentrated to PI(4,5)P 2 -enriched subdomains of the plasma membrane, thereby enhancing the overall fluorescence.
To characterize the subcellular distribution of the stimulus-dependent PI(4,5)P 2 induction, we conducted a chemical fixation protocol whereby the NMJ boutons were fixed at rest or immediately after high K + stimulation and then immunostained for PLC d1 -PH-EGFP using an a-GFP antibody to enhance the signal for high-resolution microscopic imaging. Confocal images revealed that, for neurons at rest, native PLC d1 -PH-EGFP fluorescence was weakly detected on the presynaptic plasma membrane labeled by a-Hrp staining, with some additional fluorescent signal being dispersed in the cytosol (Figure 1-figure supplement 1c). We then elevated PI(4,5)P 2 on the plasma membrane by removing a copy of synaptojanin (synj), which encodes the major neuronal PI(4,5)P 2 phosphatase (Tsujishita et al., 2001;Verstreken et al., 2003). Consistent with previous studies Verstreken et al., 2009), this reduction in Synj levels enhanced PLC d1 -PH-EGFP fluorescence (Figure 1-figure supplement 1c-d). In this context, we did not observe a significant change in protein expression of PLC d1 -PH-EGFP in the presynaptic compartment (Figure 1-figure supplement 1e-f). By using a-GFP immunostaining, these PLC d1 -PH-EGFP signals could be faithfully amplified (Figure 1-figure supplement 1c-d). Therefore, this immunostaining approach was subsequently used to monitor presynaptic PI(4,5)P 2 levels.
Next, we stimulated the boutons expressing PLC d1 -PH-EGFP with 90 mM K + and 2 mM Ca 2+ for 10 min. Similar to our live-imaging results, PLC d1 -PH-EGFP signals were significantly increased on the presynaptic plasma membrane (Figure 1-figure supplement 1g-h). In particular, we observed high-level induction of PLC d1 -PH-EGFP puncta (Figure 1c-d).
To assess the possibility that chemical fixation may have altered membrane properties and consequently plasma membrane PLC d1 -PH-EGFP clustering upon intense stimulation, we further examined the distributions of mCD8-GFP and PLC d1 -PHS39R-EGFP, a PI(4,5)P 2 -binding mutant (Khuong et al., 2010;Verstreken et al., 2009). There was no obvious change in the plasma membrane pattern of mCD8-GFP in fixed boutons stimulated with high K + (Figure 1-figure supplement 2a-b). Unlike PLC d1 -PH-EGFP, PLC d1 -PHS39R-EGFP was found mainly in the cytosol at rest. After intense stimulation, immunostaining signals of PLC d1 -PHS39R-EGFP did not accumulate on the plasma membrane ( Figure 1-figure supplement 2c-d) and were not increased within the boutons (Figure 1-figure supplement 2e). Thus, clustering of PLCd1-PH-EGFP on the plasma membrane upon stimulation results from an increase in local PI(4,5)P 2 concentration, although a potential effect of chemical fixation, if any, on PLC d1 -PH-EGFP clustering cannot be excluded. To assess potential dominant-negative effects of PI(4,5)P 2 binding by PLC d1 -PH-EGFP on the stimulation-induced accumulation of PI(4,5)P 2 , we further examined the recruitment of the AP-2 complex by PI(4,5)P 2 . Similar to PLC d1 -PH-EGFP, levels of the a subunit of a  Figure 1. PI(4,5)P 2 forms microdomains at periactive zones under conditions of intense stimulation. (a-b) Increased fluorescence of PLC d1 -PH-EGFP but not mCD8-GFP in NMJ boutons upon intense stimulation. (a) (Top) Live images of the boutons (arrows) expressing UAS-PLC d1 -PH-EGFP or UAS-mCD8-GFP. The larvae were reared at 25˚C. Electrical (20 or 40 Hz) or chemical (90 mM K + ) stimulation was conducted in a 2 mM-Ca 2+ solution for 3 min (electrical) or 5 min (chemical) and then rested in 0 mM Ca 2+ and 5 mM K + . Snapshot images taken before stimulation, at the third (electrical) and fifth Figure 1 continued on next page the AP-2 complex (AP-2a) were also increased on the plasma membrane upon high K + stimulation (Figure 1-figure supplement 3a-b). Together, these data indicate that intense stimulation can promote the formation of PI(4,5)P 2 microdomains.
Next, to determine the Ca 2+ dependence of the PI(4,5)P 2 microdomains, we stimulated the boutons in a solution of 90 mM K + and 0 mM Ca 2+ , which resulted in failure to induce PI(4,5)P 2 microdomain formation (Figure 1d). We obtained a similar result using 1 min stimulation of 90 mM K + and 0.5 mM Ca 2+ (Figure 1d), which was previously shown to primarily elicit CME but not ADBE (Yao et al., 2017). Hence, these results suggest that intense stimulation can elicit Ca 2+ -driven compartmentalization of PI(4,5)P 2 at the periactive zone in synaptic boutons of Drosophila NMJs.
SVs regenerate from bulk endosomes within minutes of their formation (Cheung and Cousin, 2012;Glyvuk et al., 2010;Kononenko et al., 2014;Stevens et al., 2012;Wu et al., 2014c). Using an approach employed previously (Stevens et al., 2012), we treated the boutons with high K + followed by an incubation in 5 mM K + and 0 mM Ca 2+ solution for 10 or 20 min, allowing the SVs to reform from the bulk endosomes. In controls (nSyb-GAL4), the number and area of the induced bulk endosomes reverted to almost basal levels within 10 min ( . Hence, the formation of PI(4,5)P 2 microdomains is essential for both ADBE and SV reformation from the bulk endosome, with the latter event being elicited by relatively high levels of PI(4,5)P 2 . Note that perturbation of PI(4,5)P 2 microdomain formation did not cause SV accumulation on the bulk endosome during recovery periods, indicating that PI(4,5)P 2 microdomains may play an early role in SV reformation from bulk endosomes. PI(4,5)P 2 microdomains are established via a positive feedback loop of fwe and PI(4,5)P 2 We have previously shown that the SV-associated Ca 2+ channel Flower (Fwe) elevates presynaptic Ca 2+ levels in response to strong stimuli to trigger ADBE (Yao et al., 2009;Yao et al., 2017). Given the Ca 2+ dependence of PI(4,5)P 2 microdomains (Figure 1d), we hypothesized that exocytosis evoked by intense stimulation promotes Fwe clustering at periactive zones, where it may provide the Ca 2+ influx to induce PI(4,5)P 2 microdomain formation. To test this hypothesis, we first conducted a proximity ligation assay (PLA) (Sö derberg et al., 2008) to investigate if there is a close association between Fwe and PI(4,5)P 2 in response to stimulation. In our PLA (Figure 3a), UAS-Flag- , mild Synj expression (nSyb-GAL4/UAS-synj at 25˚C), or high Synj expression (nSyb-GAL4/UAS-synj at 29˚C). At rest (10 min incubation of 5 mM K + /0 mM Ca 2+ ). High K + (10 min stimulation of 90 mM K + /2 mM Ca 2+ ). 10 min recovery (10 min stimulation of 90 mM K + /2 mM Ca 2+ , followed by 10 min incubation of 5 mM K + /0 mM Ca 2+ ). 20 min recovery (10 min stimulation of 90 mM K + /2 mM Ca 2+ , followed by 20 min incubation of 5 mM K + /0 mM Ca 2+ ). Bulk endosomes (>80 nm in diameter, red asterisks). Mitochondria (mt). Quantification data for total number of bulk endosomes per bouton area (d).
Individual data values are shown in graphs. p values: ns, not significant; *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001. Mean ± SEM. Scale bar: 500 nm. Statistics: one-way ANOVA with Tukey's post hoc test. The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Source data for Figure 2.     Fwe-HA was expressed in a fwe mutant background to replace endogenous Fwe protein with a tagged protein, and expression of UAS-PLC d1 -PH-EGFP reported the localization of PI(4,5)P 2 . Primary antibodies against the HA tag and EGFP protein were used to detect interactions between Flag-Fwe-HA and PLC d1 -PH-EGFP. The PLA signal was low in resting boutons, whereas high K + treatment significantly increased PLA signal intensity (Figure 3a-b). Therefore, these results suggest that Fwe and PI(4,5)P 2 are closely colocalized when intense stimulation triggers PI(4,5)P 2 microdomain formation.
Next, we investigated the effect of loss of Fwe on PI(4,5)P 2 microdomain formation. Basal levels of PI(4,5)P 2 were not affected in the fwe mutant relative to control (Figure 3c-d). However, intense stimulation failed to elicit PI(4,5)P 2 microdomain formation in the fwe mutant (Figure 3c-d), revealing a crucial role for Fwe in this process. To determine if the Ca 2+ channel activity of Fwe is responsible for this activity, we conducted rescue experiments based on our previous report in which fwe mutant boutons exhibited expression of the wild-type Fwe transgene or the FweE79Q mutant transgene that has reduced Ca 2+ conductance (Yao et al., 2017). We found that the wild-type Fwe transgene promoted PI(4,5)P 2 microdomain formation, whereas the FweE79Q mutant transgene lost that ability (Figure 3e-f). Thus, Fwe triggers the formation of PI(4,5)P 2 microdomains in a Ca 2+ channeldependent manner.
increase in intracellular Ca 2+ upon high K + stimulation, fwe mutant boutons exhibit an impaired Ca 2+ response (Figure 3i-j). Unexpectedly, canA1 knockdown also elicited the same deficient Ca 2+ response (Figure 3i-j). Similar suppressive effects were obtained upon overexpressing Synj or the PLC d1 -PH domain (Figure 3i-j). Hence, Calcineurin activation may increase PI(4,5)P 2 activity, which may in turn promote the Ca 2+ channel activity of Fwe to further increase intracellular Ca 2+ levels. To investigate the potential for such feedback regulation, we expressed the PLC d1 -PH domain in a loss of Fwe background. Expression of the PLC d1 -PH domain did not rescue the low Ca 2+ concentration caused by the fwe mutation (Figure 3i-j), showing that the Ca 2+ suppression exerted by the PLC d1 -PH domain indeed depends on Fwe. These findings support that a positive feedback loop involving Fwe and PI(4,5)P 2 is responsible for the formation of PI(4,5)P 2 microdomains. PI(4,5)P 2 gates fwe A well-known function of PI(4,5)P 2 is to modulate ion channel activity through its electrostatic binding to clustered positively-charged amino acids adjacent to the transmembrane domains of ion channels (Hille et al., 2015;Suh and Hille, 2008). Through protein alignment analysis, we found tandem positively-charged amino acids, including lysine (K) and arginine (R), in the intracellular juxta-transmembrane regions of Fwe. These residues are evolutionarily conserved in mice and humans (Figure 4a), whereas other cytosolic residues show poor conservation. This feature inspired us to test the potential impact of PI(4,5)P 2 on the channel function of Fwe. To determine direct interaction between Fwe and PI(4,5)P 2 , we conducted a nanoluciferase (Nluc)-based bioluminance resonance energy transfer (BRET) assay (Cabanos et al., 2017). In our BRET assay (Figure 4a), upon ion channel binding of BODIPY-TMR-conjugated PI(4,5)P 2 , illumination of Nluc-fused ion channels wrapped in detergent-formed micelles can excite BODIPY-TMR-conjugated PI(4,5)P 2 to emit a BRET signal. As shown in Figure 4b, we reconstituted the micelles containing purified Nluc-Fwe-1D4 fusion proteins and, after adding BODIPY-TMR-PI(4,5)P 2 and furimazine (a Nluc substrate), we observed a remarkable increase in BRET signal. Excess cold PI(4,5)P 2 reduced the signal to~50% of the BRET signals by competing for the PI(4,5)P 2 binding sites in Fwe, suggesting direct PI(4,5)P 2 binding to Fwe. To assess the involvement of the positively-charged amino acids of Fwe in PI(4,5)P 2 binding, we mutated all of the clustered positively-charged amino acids to non-charged alanine to eliminate the electrostatic interactions. Residue substitution resulted in a significant reduction in BRET signal, comparable to the competitive effect attributable to provision of excess cold PI(4,5)P 2 . Therefore, the majority of the PI(4,5)P 2 binding activity of Fwe is mediated by these positively-charged amino acids. Specifically, alanine substitution of residues in both the middle (K95, K100, R105) and C-terminal (K146, K147, R150) regions of Fwe reduced PI(4,5)P 2 -specific binding (Figure 4b-c). Moreover, when N-terminal residues (K29, R33) of Fwe were further substituted with alanines, binding of PI(4,5)P 2 was also reduced (Figure 4b-c). Hence, these in vitro assays reveal that Fwe directly binds PI(4,5)P 2 through multiple regions.
To directly test how PI(4,5)P 2 affects Flower Ca 2+ channel function, we generated UAS transgenes for the Fwe variants in which all or subsets of positively-charged amino acids were mutated to alanine and performed a mutant rescue experiment using nSyb-GAL4. Mutations of all nine residues or only those in the middle region (K95/K100/R105) led to very low protein expression levels, preventing further study. However, alanine substitution of C-terminal residues K146/K147/R150 did not affect SV localization of Fwe or its ability to regulate presynaptic Ca 2+ concentration and induce PI (4,5)P 2 microdomain formation (Figure 4-figure supplement 1a-f), suggesting that these residues do not play a regulatory role in Fwe channel activity. We also mutated the N-terminal residues K29/ R33. The resulting K29A/R33A variant was still able to properly localize to presynaptic terminals (Figure 4-figure supplement 2a-b). However, upon high K + stimulation, the K29A/R33A variant lost that ability to maintain proper intracellular Ca 2+ levels (Figure 4d-e). Moreover, that variant failed to promote PI(4,5)P 2 microdomain formation upon high K + stimulation (Figure 4f-g). These results reveal that the positive feedback loop involving Fwe and PI(4,5)P 2 relies on PI(4,5)P 2 -dependent gating control of Fwe.     Blockade of the positive feedback loop reduces ADBE and SV reformation from bulk endosomes Next, we assessed the impact of the Fwe and PI(4,5)P 2 regulatory feedback loop on ADBE. Loss of Fwe severely impaired formation of bulk endosomes induced by ADBE under high K + conditions compared to the fwe mutant rescue control (Figure 5a-b), consistent with our previous findings (Yao et al., 2017). Next, we found that expression of wild-type Fwe protein or the K146A/K147A/ R150A mutant variant restored proper ADBE in the fwe mutant background (Figure 5a-b; Figure 4figure supplement 1g-h), whereas expression of the K29A/R33A variant failed to rescue ADBE deficiency (Figure 5a-b). Consistent with the suppressive effects caused by expression of the PLCd1-PH domain or Synj (Figure 2), all of the bulk endosomes that remained in boutons lacking fwe could not generate new SVs during a 10 min or even 20 min recovery period (Figure 5a Neuronal canA1 knockdown by expressing canA1 RNAi construct also impaired ADBE relative to the nSyb-GAL4 control (Figure 5a-b). To verify if CanA1 regulates ADBE in an Fwe-dependent manner, we overexpressed Fwe to increase the intracellular Ca 2+ levels under canA1 RNAi knockdown conditions. Fwe overexpression significantly reversed the ADBE defect (Figure 5a-b), suggesting that increasing Fwe-dependent Ca 2+ influx can augment activation of the residual CanA1 enzymes and thus normalize downstream ADBE. Taken together with our results reported in previous sections, we propose that the positive feedback loop involving Fwe, CanA1 and PI(4,5)P 2 compartmentalizes PI(4,5)P 2 microdomains at the periactive zone of boutons to dictate and coordinate ADBE and subsequent SV reformation.

PI(4,5)P 2 facilitates retrieval of fwe to bulk endosomes
It was reported recently that a SV protein sorting process occurs during ADBE (Kokotos et al., 2018;Nicholson-Fish et al., 2015). VAMP4, a v-SNARE protein, is essential for ADBE to proceed, and it is selectively retrieved by ADBE (Kokotos et al., 2018;Nicholson-Fish et al., 2015). Given the important role of Fwe in triggering ADBE, we wondered if Fwe is sorted to bulk endosomes during ADBE. To visualize the vesicular localization of Fwe, we rescued the fwe mutant by expressing a UAS transgene of the APEX2 fusion protein of Fwe (UAS-HA-Fwe-APEX2), and then conducted diaminobenzidine (DAB) labeling and TEM. By means of confocal microscopy, we observed that HA-Fwe-APEX2 immunostaining signals were properly present on the SVs marked by Syt and Csp immunostaining ( Figure 6-figure supplement 1a). Furthermore, expression of this fusion protein was able to rescue the endocytic defects (Figures 4 and 5) and early animal lethality (not shown) caused by loss of fwe. Therefore, HA-Fwe-APEX2 is functionally equivalent to endogenous Fwe. APEX2 is an engineered peroxidase that is capable of catalyzing DAB polymerization and proximal deposition, with the DAB polymers binding electron-dense osmium to enhance electron microscopy contrast (Lam et al., 2015). Whereas no DAB staining signals were observed in SVs in the Flag-Fwe-HA rescue control boutons stimulated by high K + (yellow arrows, To minimize staining variability across boutons, we compared DAB staining intensities on bulk endosomes and SVs in the same boutons. Compared to Flag-Fwe-HA rescue boutons, in HA-Fwe-APEX2 rescue boutons, the staining intensities of bulk endosomes were more abundant compared to those of the surrounding SVs (Figure 6f), revealing a mechanism by which Fwe is recycled to bulk endosomes after it initiates ADBE.
PI(4,5)P 2 is known to recruit adaptor protein complexes to sort SV proteins to the nascent SV during CME (Saheki and De Camilli, 2012). Next, we assessed if PI(4,5)P 2 microdomain formation may be involved in sorting of Fwe to bulk endosomes. When PI(4,5)P 2 microdomains were perturbed by Synj overexpression (Figure 2b), the bulk endosome localization of HA-Fwe-APEX2 was significantly reduced (Figure 6d-f), whereas there was only a mild reduction in its SV localization (Figure 6d-e). Therefore, in addition to initiating ADBE, PI(4,5)P 2 microdomains play a role in facilitating the retrieval of Fwe to the bulk endosome, enabling ADBE to remove its trigger via a negative feedback regulatory mechanism and reducing endocytosis to prevent excess membrane uptake. ). TEM processing was performed after the following treatments: at rest (10 min incubation of 5 mM K + /0 mM Ca 2+ ); high K + stimulation (10 min stimulation of 90 mM K + /2 mM Ca 2+ ); 10 min recovery (10 min stimulation of 90 mM K + /2 mM Ca 2+ , followed by 10 min incubation of 5 mM K + /0 mM Ca 2+ ); or 20 min recovery (10 min stimulation of 90 mM K + /2 mM Ca 2+ , followed by 20 min incubation of 5 mM K + / 0 mM Ca 2+ ). Bulk endosomes (>80 nm in diameter, red asterisks). Mitochondria (mt). (b) Quantification data of total numbers of bulk endosomes per bouton area. Individual data values are shown in graphs. p values: ns, not significant; *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001. Mean ± SEM. Scale bar: 500 nm. Statistics: one-way ANOVA with Tukey's post hoc test. The online version of this article includes the following source data and figure supplement(s) for figure 5: Source data 1. Source data for Figure 5.

Discussion
ADBE occurs immediately after exocytosis to retrieve required SV protein and lipid constituents to further regenerate SVs under conditions of high-frequency stimulations. Here, we show that the Fwe Ca 2+ channel-dependent compartmentalization of PI(4,5)P 2 orchestrates coupling of exocytosis to ADBE and subsequent SV reformation. Based on our findings, we propose a model for this interplay (depicted in Figure 7). Under conditions of strong stimulation, SV exocytosis transfers Fwe from SVs to the periactive zone, where some of the activated Fwe provides the low Ca 2+ levels that initiate Calcineurin activation to upregulate PI(4,5)P 2 (Step 1). Increased PI(4,5)P 2 enhances Fwe Ca 2+ channel activity, thereby establishing a positive feedback loop that induces PI(4,5)P 2 microdomain formation (Step 2). High levels of PI(4,5)P 2 within these microdomains elicit bulk membrane invagination by triggering actin polymerization (Step 3). In parallel, PI(4,5)P 2 facilitates proper retrieval of Fwe to the bulk endosome (Step 4), thereby terminating the ADBE process. Finally, PI(4,5)P 2 microdomains dictate SV reformation from the bulk endosomes (Step 5), coordinating ADBE and subsequent SV reformation. Source data 1. Source data for Figure 6.

Fwe-dependent PI(4,5)P 2 microdomains trigger ADBE
The role of actin polymerization in ADBE has been investigated in mammals (Kononenko et al., 2014;Soykan et al., 2017;Wu et al., 2016), as well as in Drosophila (Akbergenova and Bykhovskaia, 2009). PI(4,5)P 2 is known to control a range of actin regulators, thereby modulating the dynamics of actin polymerization and branching (Janmey et al., 2018). It has been shown previously that, in response to nicotine stimulation, PI(4,5)P 2 forms clustered microdomains of sub-micrometer scale prior to the appearance of an actin-based ring structure in bovine chromaffin cells (Gormal et al., 2015). In agreement with this observation, we show that intense activity stimulation drives the formation of PI(4,5)P 2 microdomains at the periactive zone of Drosophila NMJ synaptic boutons. Perturbations of the formation of these microdomains reduces ADBE activity very significantly, demonstrating that rapid accumulation of PI(4,5)P 2 in microdomains is needed to trigger extensive actin polymerization, which likely generates sufficient mechanical force to produce the large endosomes. Furthermore, loss of fwe or RNAi-mediated calcineurin knockdown effectively inhibited PI(4,5)P 2 microdomain formation and, as a consequence, ADBE. Those results are consistent with previous data supporting that Ca 2+ promotes ADBE by activating its sensor Calcineurin (Cousin and Robinson, 2001;Jin et al., 2019;Sun et al., 2010;Wu et al., 2009;Wu et al., 2014b;Xue et al., 2011).
Our data show that direct binding of PI(4,5)P 2 is required for the Ca 2+ channel activity of Fwe. Perturbation of PI(4,5)P 2 -Fwe binding further impaired the formation of PI(4,5)P 2 microdomains as well as ADBE initiation. Hence, PI(4,5)P 2 controls Fwe gating, so that Fwe can promote PI(4,5)P 2 compartmentalization through positive feedback regulation. Furthermore, loss of Fwe impaired the intracellular Ca 2+ increase that was evoked upon strong activity stimulation (Figure 3g; Yao et al., 2017). These results support that, in addition to PI(4,5)P 2 , the channel function of Fwe may be gated by a significant change in membrane potential. Expanding on that notion, it is therefore possible that both factors may gate Fwe, thereby only allowing channel opening when exocytosis directs Fwe to periactive zones. Future studies should explore the details of this channel gating mechanism. PI(4,5)P 2 microdomains coordinate retrieval of SV membranes and proteins for SV reformation via ADBE Since ADBE is triggered very rapidly by intense stimuli, it was thought that this type of recycling randomly retrieves SV proteins and that the sorting process takes place when SVs regenerate from bulk endosomes. However, recent work has highlighted a distinct retrieval route for SV proteins during ADBE (Kokotos et al., 2018;Nicholson-Fish et al., 2015). Very little is known about the mechanisms underlying that retrieval route. Interestingly, removing VAMP4 or mutating its di-leucine motif was shown to impair ADBE (Nicholson-Fish et al., 2015). The di-leucine motif of transmembrane proteins is known to mediate binding with the AP-2 adaptor complex (Traub and Bonifacino, 2013). Given that the AP-2 adaptor complex works closely with PI(4,5)P 2 (McMahon and Boucrot, 2011), these findings imply a role for PI(4,5)P 2 and the AP-2 adaptor complex in SV protein sorting via ADBE. Indeed, our data show that bulk endosomes recycle few in a PI(4,5)P 2 microdomain-dependent manner. Hence, in addition to initiating ADBE, PI(4,5)P 2 may participate in SV protein sorting to bulk endosomes.
SV regeneration occurs following formation of the bulk endosome. Our results also show that either removing Fwe-derived Ca 2+ or perturbing PI(4,5)P 2 activity impaired the ability of SVs to reform from the bulk endosome, highlighting the essential role of PI(4,5)P 2 microdomains in this process. How could PI(4,5)P 2 of the plasma membrane regulate subsequent SV reformation? It has been shown that PI(4,5)P 2 is rapidly downregulated on bulk endosomes once formed by ADBE (Chang-Ileto et al., 2011;Cremona et al., 1999;Milosevic et al., 2011). It is conceivable that the high concentrations of PI(4,5)P 2 in microdomains may compensate for rapid turnover, thereby ensuring appropriate concentrations of PI(4,5)P 2 or PI(4)P for further recruitment of clathrin and adaptor protein complexes, such as AP-1 and AP-2 (Blumstein et al., 2001;Cheung and Cousin, 2012;Faúndez et al., 1998;Glyvuk et al., 2010;Kokotos et al., 2018;Kononenko et al., 2014;Park et al., 2016). Alternatively, PI(4,5)P 2 may facilitate SV protein sorting prior to ADBE, meaning proper SV protein compositions on bulk endosomes could control recruitment of adaptor protein complexes. Both of these potential mechanisms are not mutually exclusive and may operate in parallel. Therefore, we propose that the Fwe-dependent formation of PI(4,5)P 2 microdomains is potentially important in coordinating retrieval of SV membranes and cargos when SVs are recycled via ADBE. Notably, the Fwe channel is evolutionarily conserved from yeast to human (Yao et al., 2009). We have also previously demonstrated conserved functions of Fwe in CME and ADBE at the mammalian central synapse (Yao et al., 2017). A recent study has also identified Fwe as a key protein mediating Ca 2+ -dependent granule endocytosis in mouse cytotoxic T lymphocytes (Chang et al., 2018). Hence, we hypothesize that the mechanism of ADBE we report here may be generally deployed across synapses and species, even in other non-neuronal cells.

Immunohistochemistry
Third instar larvae were fixed with 4% paraformaldehyde for 20 min. We used 1xPBS buffer containing 0.1% Tween-20 to stain the HA-tagged Fwe proteins. We used 1xPBS buffer containing 0.1% Triton X 100 to stain PLC d1 -PH-EGFP or AP-2a. We used 1xPBS buffer containing 0.2% Triton X 100 to stain GCaMP6f. Primary antibodies were used as follows: guinea pig a-Fwe B isoform (1:400) (Yao et al., 2017), chicken a-GFP (Invitrogen, 1:500); mouse a-HA (Sigma, 1:400), mouse a-Bruchpilot (Developmental Studies Hybridoma Bank nc82, 1:100); rabbit a-AP-2a (1:3000) (González-Gaitán and Jäckle, 1997) guinea pig a-Eps15 (1:3000) (Koh et al., 2007); rabbit a-HRP conjugated with Alexa Fluor 488, Cy3 or Cy5 (Jackson ImmunoResearch Laboratories, 1:250). Secondary antibodies conjugated to Alexa Fluor 488, Alexa Fluor 555, or Alexa Fluor 647 (Invitrogen and Jackson ImmunoResearch) were used at 1:500. The NMJ boutons were derived from muscles 6 and 7 of abdominal segments 2/3. To detect PI(4,5)P 2 microdomains, the NMJ boutons were fixed immediately after high K + stimulation. To quantitatively compare PLC d1 -PH-EGFP or AP-2a immunostaining signal among different sets of experiments, fixed larval fillets derived from different conditions were collected into the same Eppendorf tube. The NMJ boutons were stained for PLC d1 -PH-EGFP using a-GFP and for the neuronal membranes using fluorescein-conjugated a-HRP. Consecutive singleplane images of the boutons of muscles 6 and 7 in abdominal segments 2 or 3 of all different experimental sets were taken using a Zeiss LSM 780 confocal microscope with a Plan-Apochromat 63x/1.4 Oil DIC M27 objective under a 1 mm interval setup and equal laser power and laser exposure time. For data quantification, single-plane images of five different individual boutons from each NMJ bouton image were used. The presynaptic plasma membrane regions of the type Ib boutons defined by a-HRP immunostaining were outlined manually, and native fluorescence or a-GFP immunostaining signal intensities of PLC d1 -PH-EGFP or AP-2a on the plasma membrane were quantified using ImageJ and averaged to serve as one individual data-point. For data quantification of SIM images, we chose single-plane SIM images focused on a central section of an individual bouton and outlined plasma membrane-associated PLC d1 -PH-EGFP-immunostained clusters using ImageJ. The area of these clusters of each individual bouton was measured using ImageJ, and the areas of clusters measuring over 0.032 mm 2 (equal to a~200 nm diameter circle, i.e. the resolution limit of SIM) were averaged to serve as one data-point. We assessed 37 boutons derived from five NMJs of three different larvae. Image processing was achieved using LSM Zen.

Western blot
For western blotting, the brain and ventral nerve chord of larval fillets were removed and subjected to different stimulation conditions. Afterwards, the fillets were crushed in 1xSDS sample buffer and boiled for 5 min. Dilutions for primary antibodies were as follows: mouse anti-a-actin, 1:20000 (Sigma); chicken anti-GFP, 1:5000 (Invitrogen).

PLA
Third-instar larvae were fixed with 4% paraformaldehyde for 20 min and permeabilized with 1xPBS buffer containing 0.1% Tween-20. Larval fillets were incubated with mouse a-HA (Sigma, 1:200) and rabbit a-GFP (Invitrogen, 1:500) in 1xPBS buffer containing 0.1% Tween-20 at 4˚C for 12 hr. Excess antibodies were washed out using 1xPBS buffer containing 0.1% Tween-20. The samples were mixed with the PLA probe (Sigma, 1:5) for 2 hr at 37˚C. After washing with 1x buffer A, the samples were incubated with ligation solution (1:40) for 1.5 hr at 37˚C. After again washing with 1x buffer A, the samples were incubated with amplification solution (1:80) for 2 hr at 37˚C. Next, the samples were washed with 1x buffer B and then 0.01x buffer B. The samples were stained with anti-chicken Alexa Fluor 488-conjugated IgG and anti-mouse Alexa Fluor 647-conjugated IgG, followed by a wash of 1x PBS buffer containing 0.1% Tween-20. To quantitatively compare PLA signal, fixed larval fillets derived from different experimental conditions were collected into the same Eppendorf tube and processed. Consecutive single-plane images of the boutons of muscles 6 and 7 in abdominal segments 2 or 3 of all different experimental sets were taken using a Zeiss LSM 780 confocal microscope with a Plan-Apochromat 63x/1.4 Oil DIC M27 objective under a 1 mm interval setup and equal laser power and laser exposure time. For data quantification, consecutive Z-plane images spanning whole NMJ were projected under maximal fluorescence intensity. All type Ib boutons in individual Z-projection image were outlined according to PLC d1 -PH-EGFP-stained regions. PLA or antibody immunostaining signal intensities within the boutons and background staining signals in surrounding muscles were counted using ImageJ and averaged. One individual data-point was obtained by muscular background signal subtraction. Image processing was achieved using LSM Zen.
Fluorescence signal was detected using a Microplate Reader M1000 pro (Tecan) with two different emission spectrum filters, that is 500-540 nm for Nluc and 550-630 nm for BODIPY-TMR Phosphatidylinositol 4,5-bisphosphate. For competition assay, 30 ml of the reaction solution was included with 1 mM brain phosphatidylinositol 4,5-bisphosphate (Avanti). The BRET signal was calculated according to the following formula:

Live imaging
For PLC d1 -PH-EGFP imaging, third-instar larvae were dissected in a zero-calcium HL-3 solution at room temperature. For groups stimulated with electric pulses, larval fillets were bathed in a solution of 2 mM Ca 2+ (70 mM NaCl, 5 mM KCl, 10 mM MgCl 2 , 10 mM NaHCO 3 , 5 mM trehalose, 5 mM HEPES (pH 7.4), 115 mM sucrose, 2 mM CaCl 2 ). High concentrations of glutamate were used to desensitize glutamate receptors, thereby reducing muscle contraction when stimulated. A cut axonal bundle was sucked into the tip of a glass capillary electrode and then stimulated at 20 or 40 Hz for 3 min. Stimulus strength was set at 5 V and 0.5 ms duration by means of pClamp 10.6 software (Axon Instruments Inc). Ten images were taken from larval fillets at rest. After stimulation for 2 min, muscle contraction significantly decelerated. Thus, we captured 60 consecutive snapshot images every second from the third minute. Muscles 6 and 7 of abdominal segment three were imaged. Under the condition of high K + stimulation, larval fillets were bathed in a solution of 90 mM K + /2 mM Ca 2+ /7 mM glutamate (25 mM NaCl, 90 mM KCl, 10 mM MgCl 2 , 10 mM NaHCO 3 , 5 mM trehalose, 5 mM HEPES (pH 7.4), 30 mM sucrose, 2 mM CaCl 2 , 7 mM monosodium glutamate) for 5 min. Sixty consecutive snapshot images were captured every second from the fifth minute of stimulation. Images were taken using a long working distance water immersion objective (XLUMPLFLN20XW, Olympus) and EMCCD camera (iXon, Andor) mounted on a SliceScope Pro 6000 (Scientifica) microscope and employing MetaFluor software (Molecular Devices). For GCaMP6f imaging, third instar larvae were dissected in a zero-calcium HL-3 solution at room temperature. Ten images were taken from larval fillets at rest. Subsequently, larval fillets were stimulated with a solution of 90 mM K + /2 mM Ca 2+ /7 mM glutamate for 5 min. Sixty consecutive snapshot images were captured every second from the fifth minute of stimulation. The lexA/lexAop2 binary system was used to stably express a comparable level of GCaMP6f in the presynaptic compartment of NMJ boutons for all tested genotypes, allowing us to compare Ca 2+ imaging results when the resting Ca 2+ levels were potentially affected by differences in genetic background. We stained for the GCaMP6f protein using a-GFP antibody and confirmed comparable GCaMP6f levels among the different genotypes tested in each dataset. Evoked Ca 2+ levels were calculated by subtracting the resting GCaMP6f fluorescence from the GCaMP6f fluorescence induced by high K + stimulation. The NMJs were derived from muscles 6 and 7 of abdominal segment 2/3. Images were taken using a water immersion objective (W Plan-Apochromat 40x/1.0 DIC M27, Zeiss). For each imaging experiment, at least three focused images for the same boutons under resting, stimulation, or post-stimulation conditions were used for data quantification. Fluorescence intensities of PLC d1 -PH-EGFP or GCaMP6f within the boutons were quantified using ImageJ and averaged to serve as one individual data-point. Image processing was achieved using LSM Zen.

Transmission electron microscopy
Third instar larval fillets were prepared in zero-calcium HL-3 solution at room temperature. For the resting conditions, the fillets were bathed in zero-calcium HL-3 solution at room temperature for another 10 min before fixation. For the high K + stimulation conditions, fillets were bathed in a solution of 90 mM K + and 2 mM Ca 2+ for 10 min. The stimulation was terminated by washing three times with zero-calcium HL-3 solution, followed by fixation. For recovery conditions, following high K + stimulation, fillets were bathed in zero-calcium HL-3 solution at room temperature for 10 or 20 min before fixation. Larval fillets were fixed for 12 hr at 4˚C in 4% paraformaldehyde/1% glutaraldehyde/ 0.1 M cacodylic acid (pH 7.2), rinsed with 0.1 M cacodylic acid (pH 7.2), and postfixed with 1% OsO 4 and 0.1 M cacodylic acid at room temperature for 3 hr. These samples were then subjected to a series of dehydration steps using 30-100% ethanol. After 100% ethanol dehydration, the samples were sequentially incubated with propylene, a mixture of propylene and resin, and pure resin. Finally, the samples were embedded in 100% resin. TEM images were captured using Tecnai G2 Spirit TWIN (FEI Company) and a Gatan CCD Camera (794.10.BP2MultiScanTM). NMJ boutons were captured at high magnifications. For each condition, NMJ bouton images were taken from at least five different NMJs of each third-instar larvae, and three to five larvae were used. Quantifications were performed using ImageJ. For diaminobenzidine (DAB) polymerization, third instar larvae were dissected at room temperature in zero-calcium HL-3 medium, followed by a 10 min incubation in 5 mM K + /0 Ca 2+ mM solution or a 10 min stimulation of 90 mM K + /2 mM Ca 2+ . Next, the samples were subjected to 30 min fixation in ice-cold 4% paraformaldehyde/1% glutaraldehyde/0.1 M cacodylic acid (pH 7.2). Subsequently, the samples were transferred to Eppendorf tubes for 15 min incubation with a solution of 0.5 mg/ml DAB solution, followed by incubation with a solution of 0.5 mg/ ml DAB and 0.006% H 2 O 2 for 15 min at room temperature. This latter step was repeated once to ensure DAB polymerization. Samples were washed three times with 1xPBS buffer for 10 min and then fixed with a solution of 4% paraformaldehyde/1% glutaraldehyde/0.1 M cacodylic acid (pH 7.2) for 12 hr at 4˚C, followed by fixation with a solution of 1% OsO 4 /0.1 M cacodylic acid at room temperature for 3 hr. Then, standard dehydration, embedding, and imaging were performed. For data quantifications of DAB intensities, the display color of TEM images was reverted to grayscale using ImageJ. Average DAB staining intensity on each individual bulk endosome was quantified. Then, the average DAB staining intensity on 50-100 surrounding SVs from the same bouton image was used to assess the relative level of HA-Fwe-APEX2 on bulk endosomes vs SVs.

Statistics
All data analyses were conducted using GraphPad Prism 8.0, unless stated otherwise. Paired and multiple datasets were compared by Student t-test or one-way ANOVA with Tukey's post hoc test, respectively. Individual data values are biological replicates. Samples were randomized during preparation, imaging, and data processing to minimize bias.

Data availability
All data generated or analysed during this study are included in the manuscript and supporting files. Source data files have been provided. Continued on next page