The calcium channel subunit α2δ-3 organizes synapses via a novel activity-dependent, autocrine BMP signaling pathway

Synapses are highly specialized for neurotransmitter signaling, yet activity-dependent growth factor release also plays critical roles at synapses. While efficient neurotransmitter signaling is known to rely on precise apposition of release sites and neurotransmitter receptors, molecular mechanisms enabling high-fidelity growth factor signaling within the synaptic microenvironment remain obscure. Here we show that the auxiliary calcium channel subunit α2δ-3 promotes the function of a novel activity-dependent autocrine BMP signaling pathway at the Drosophila NMJ. α2δ proteins have conserved synaptogenic activity, although how they execute this function has remained elusive. We find that α2δ-3 provides an extracellular scaffold for autocrine BMP signaling, suggesting a new mechanistic framework for understanding α2δ’s conserved role in synapse organization. We further establish a transcriptional requirement for activity-dependent, autocrine BMP signaling in determining synapse density, structure, and function. We propose that activity-dependent, autocrine signals provide neurons with continuous feedback on their activity state and are thus well poised to modulate synapse structure and function.


Introduction
Synapses are asymmetric cellular junctions underlying directional information transfer in the nervous system. Activity-dependent pathways are often invoked as regulators of synapse development, maintenance, plasticity, and elimination. While roles for activity-dependent cues at synapses are indisputable, the molecular links between activity and synapse differentiation remain poorly defined. To establish these connections, it is imperative to define when and where activity-dependent cues are released, and then how these extracellular signals impinge on synapse morphology and function.
Classical morphogens and neurotrophins have emerged as central regulators of synaptic development. A diverse collection of secreted cues including BMPs, Wnts, IGFs, FGFs, BDNF, Netrins, and Ephrins promote differentiation of the pre-or postsynaptic compartment (Cao et al., 2011;Glasgow et al., 2018;Johnson-Venkatesh and Umemori, 2010;Packard et al., 2002;Shen and Scheiffele, 2010;Terauchi et al., 2016). Such cues are often proposed to coordinate pre-and postsynaptic development by signaling across the synapse. Autocrine signals also serve important functions at the synapse by providing ongoing feedback to individual pre-or postsynaptic cells regarding their activity state (Herrmann and Broihier, 2018). As increasing numbers of synaptic signals with distinct functions and cellular origins are identified, the question arises of how neurons discriminate among these information channels. It is reasonable to hypothesize that proteins in the synaptic cleft serve crucial roles controlling the spatiotemporal availability of secreted proteins at the synapse to facilitate signal segregation.
α2δ proteins are well-positioned to impact the activity of synaptic signals. They are best studied as accessory calcium channel subunits and are composed of two disulfide-linked peptides: an entirely extracellular α2 peptide and a δ peptide predicted to be membraneassociated via a GPI link (Davies et al., 2010;Dolphin, 2018). They are important for voltagegated calcium (CaV) channel trafficking to synaptic terminals as well as for biophysical properties of CaV channels (Dolphin, 2018;Ferron et al., 2018;Kadurin et al., 2016).
Underscoring α2δ functions at mature synapses, α2δ proteins are cellular targets of Gabapentin (Field et al., 2006), a commonly prescribed treatment for epilepsy and neuropathic pain. All α2δ proteins contain a von Willebrand factor A (VWA) domain, a protein-protein interaction domain characteristic of ECM proteins such as matrilins and collagens (Whittaker and Hynes, 2002). Interestingly, α2δ proteins interact with synaptogenic proteins including Thrombospondin-1 (TSP-1) (Eroglu et al., 2009;Risher et al., 2018), suggesting functions beyond CaV channel localization and function. In line with this hypothesis, Drosophila α2δ-3, also called Straightjacket (Stj), is required for bouton morphogenesis at the embryonic neuromuscular junction (NMJ)-a function independent of the CaV channel Cacophany (Cac) . Functions for Drosophila α2δ-3 in subsequent aspects of NMJ differentiation are incompletely understood, but are reported to include synaptic assembly, synaptic function, and presynaptic homeostatic plasticity (Dickman et al., 2008;Ly et al., 2008;Wang et al., 2016).
In this study, we report a functional requirement for Drosophila α2δ-3 in activity-dependent autocrine BMP signaling. At the Drosophila NMJ, postsynaptic Gbb (Glass bottom boat), a BMP family member, is an early and permissive retrograde cue that initiates scaling growth of the NMJ (Berke et al., 2013;McCabe et al., 2003). We previously demonstrated that presynaptic Gbb is also trafficked to presynaptic terminals, where it is released in response to activity (James and Broihier, 2011;James et al., 2014). What is the function of this activity-dependent cue? Here we show that it is an autocrine signal required to maintain synapse structure and function. In the absence of presynaptic Gbb, active zone architecture, synaptic vesicle distribution, and baseline neurotransmitter release are all disrupted. Loss of the BMPresponsive transcription factor Mad results in similar phenotypes, arguing for a canonical BMP pathway. Embryonic bouton morphogenesis is also disrupted in BMP mutants, a phenotype identical to that seen in α2δ-3 mutants , prompting us to test for a link between BMP signaling and α2δ-3. Indeed, we find that structural and behavioral defects in α2δ-3 mutants are suppressed by activation of autocrine BMP signaling, suggesting that α2δ-3 mutant NMJs have reduced autocrine BMP signaling. Providing mechanistic insight into the function of α2δ-3 in autocrine BMP signaling, the extracellular α2 peptide of α2δ-3 prevents diffusion of Gbb following its activity-dependent release. We therefore propose that α2δ-3 is a key component of the microenvironment in the synaptic cleft that acts as a diffusion barrier to extracellular Gbb.

A presynaptic, autocrine BMP pathway at the Drosophila NMJ
Seminal studies demonstrate that BMP signaling orchestrates NMJ development and physiology. Constitutive loss of BMP signaling causes a reduction in overall NMJ size as judged by bouton number-as well as ultrastructural defects, reduced evoked glutamate release, and impaired homeostatic plasticity (Goold and Davis, 2007;Marques et al., 2002;McCabe et al., 2003;. These widespread defects raise the question of whether the phenotypes have a common root or, if instead, they reflect separable cell-specific roles for BMP signaling. Early work suggested at least partially separable pre-and postsynaptic BMP pathways; while expression of Gbb in the postsynaptic muscle rescues bouton number in gbb constitutive nulls, it does not rescue evoked neurotransmitter release. Expression of Gbb in the presynaptic neuron is required to restore proper glutamate release to gbb nulls (Goold and Davis, 2007;James et al., 2014;McCabe et al., 2003). These findings suggest that Gbb is released by the presynaptic motor neuron. Lending support to this idea, we found that Gbb is trafficked to presynaptic terminals, where it is subject to activity-dependent release (James et al., 2014).
We hypothesized that this presynaptic pool of Gbb regulates synapse formation or maintenance. To explore this idea, we first established the effect of constitutive loss of Gbb. As expected, overall NMJ size, as assessed by bouton number, is significantly decreased in gbb nulls ( Figure S1 A-B, H). Each bouton contains many synapses, or individual presynaptic glutamate release sites precisely aligned to postsynaptic glutamate receptor clusters. We utilized the ELKS-related protein Bruchpilot (Brp) as a presynaptic marker, GluRIII as a postsynaptic marker-and defined a synapse as a pair of Brp/GluRIII puncta (Fouquet et al., 2009;Kittel et al., 2006;Marrus et al., 2004;Wagh et al., 2006). We scored Brp-positive synapse density (synapse number/µm 2 ) to exclude differences in synapse number arising as a secondary consequence of altered overall NMJ size. We find that loss of Gbb results in a 30% decrease in Brp-positive synapse density (Figure 1 A-B, H), indicating that Gbb regulates synapse development.
We defined the relevant cellular source of Gbb by selectively removing either pre-or postsynaptic ligand. To test necessity of the postsynaptic pool we used muscle-specific gbb RNAi (gbb 1 /+; BG57>gbb RNAi). In both this RNAi experiment and those described below, knock-down was performed in gbb heterozygotes. gbb heterozygosity does not impact synapse density in an otherwise wild-type background (Figure 1 H). To confirm this RNAi-based approach and to test sufficiency of the presynaptic pool, we overexpressed Gbb in motor neurons in gbb nulls (gbb 1/2 ; D42>Gbb). As expected, muscle-derived ligand is required for NMJ expansion, since bouton number is significantly reduced in both backgrounds (Marques et al., 2002;McCabe et al., 2003) (Figure S1 A, C, H). In contrast, we do not detect a change in Brp/GluRIII density (Figure 1 A, C, H), indicating that muscle-derived Gbb does not otherwise regulate synapse development. We gained genetic access to motor neuron-derived Gbb using analogous strategies. We utilized motor neuron-specific RNAi (gbb 1 /+; D42>gbb RNAi) to probe a requirement for presynaptic ligand, while to test sufficiency of the postsynaptic ligand, we expressed Gbb in muscle in a gbb null (gbb 1/2 ; BG57>Gbb). As expected, bouton number is unchanged in these backgrounds (Figure S1 A, D, H), indicating that motor neuron-derived ligand does not regulate NMJ growth. However, the density of Brp/GluRIII puncta in both backgrounds is decreased comparably to gbb nulls (Figure 1 A, D, H). Thus, the synapse density phenotype observed in gbb nulls is attributable to neuron-derived ligand.
Wishful thinking (Wit) is the neuronal Type II BMP receptor mediating Gbb function in NMJ growth (McCabe et al., 2004;Marques et al., 2003). We were curious if Wit is also required for Gbb's synapse density function. As expected, loss of Wit results in fewer boutons ( Figure S1 A, E, H). In addition, we observe a 40% decrease in Brp-positive synapse density (Figure 1 A,

E, H), indicating that Wit regulates synapse development. To determine whether presynaptic
Wit carries out this function, we assayed whether Wit expression in motor neurons in wit nulls (wit A12/B11 ; D42>Wit) rescues the decrease in synapse density. Indeed, synapse density is fully rescued by motor neuron expression of Wit (Figure 1 A, F, H), indicating a presynaptic requirement. The genetic requirement of ligand and receptor in the presynaptic neuron, and not in the muscle, argues for an autocrine BMP signaling loop. Lastly, we were interested in whether this autocrine pathway requires Mad, the canonical transcription factor in the BMP pathway. Loss of Mad results in a 45% reduction in synapse density, supporting a canonical BMP pathway (Figure 1 A, G-H). Together, these findings uncover a canonical, presynaptic and autocrine BMP pathway that promotes Brp-positive synapse density at the NMJ.

Presynaptic, autocrine BMP signaling regulates active zone architecture and composition
In the preceding analysis, we had the impression that loss of autocrine BMP signaling results in larger and irregular Brp puncta (Figure 1). When labeled with a C-terminal Brp antibody (nc82) and imaged in planar orientation at higher resolution, Brp puncta resolve into roughly 200 nm diameter rings (Figure 2 A) (Fouquet et al., 2009). Loss of key active zone cytomatrix components alter Brp ring geometry (Bruckner et al., 2017;Liu et al., 2011;Müller et al., 2012;Owald et al., 2010; making it a sensitive probe for assembly of the highly ordered cytomatrix. Thus, we analyzed Brp ring size and distribution utilizing extended resolution microscopy to gain insight into gbb mutant phenotypes. In controls, we find that the average diameter of isolated Brp rings is roughly 200 nm, consistent with published reports (Figure 2 A, H, I, J) (Bruckner et al., 2017;Kittel et al., 2006;Müller et al., 2012;Owald et al., 2010;. In contrast, mean Brp ring diameter increases by 23% in gbb nulls (Figure 2 B, J). We investigated the relevant cellular source of Gbb using celltype specific knockdown and rescue genotypes described in detail above. We find that loss of neuronal Gbb results in increased Brp ring diameter analogous to that seen in gbb nulls, while loss of muscle-derived Gbb has no effect (Figure 2 A-D, J). We next analyzed whether Wit and Mad also regulate cytomatrix architecture and find that loss of either one results in increased Brp ring diameter (Figure 2 A, E, G, J). The wit mutant phenotype is rescued by neuronal expression of Wit, again pointing to an autocrine signaling pathway (Figure 2 A, F, J). The increased Brp ring diameter in autocrine BMP mutants indicates that the pathway ensures precise construction and/or maintenance of the active zone cytomatrix.
Our analysis also revealed a clear defect in the distribution of Brp rings in autocrine BMP pathway mutants. In controls, the majority of Brp rings are separated from each other and observed as isolated rings, while in autocrine BMP mutants, many Brp rings are present in interconnected, disorganized clusters. We quantified this phenotype and found that roughly 31% of Brp rings are interconnected in controls, relative to 66% in gbb nulls (Figure 2 A-

B, K).
This phenotype is apparent when presynaptic, but not postsynaptic, ligand is lost (Figure 2 A, C-D, K), and likewise requires Mad and neuronal Wit (Figure 2 A, E-G, K). These analyses suggest that autocrine BMP signaling regulates Brp allocation among active zones. Together, these light-level analyses suggest that the observed 30% decrease in synapse density in autocrine BMP mutants may be explained by altered Brp distribution among active zones.
In other words, some active zones may have excess Brp while others have none. Brp is an essential component of electron-dense T-bars visible at the EM level (Fouquet et al., 2009).
Thus, if autocrine BMP signaling regulates Brp allocation to individual active zones, mutants might exhibit altered T-bar distribution. To test this hypothesis, we analyzed presynaptic ultrastructure in gbb mutants. We first asked if Gbb regulates the number of synapses, defined as electron-dense tightly apposed pre-and postsynaptic membranes. We scored active zone number per µm bouton perimeter in control, gbb null, as well as in animals with cell-type specific loss of Gbb and do not detect any alterations (Figure 2 T). Thus, Gbb is not required to establish active zones. However, presynaptic Gbb promotes adhesion of pre-and postsynaptic synaptic membranes, since membrane ruffling increases in mutant backgrounds (from <5% in controls to roughly 20% in presynaptic Gbb mutants), consistent with membrane ruffling reported in constitutive BMP mutants (Aberle et al., 2003;McCabe et al., 2003).
In contrast, we find a marked reduction in the percentage of active zones containing T-bars. In but have been reported with constitutive loss of the BMP pathway (Aberle et al., 2002;McCabe et al., 2003). We propose that these detached aggregates of T-bar material underlie the unopposed Brp puncta observed at the light level (arrows in Figure 1 B, D, E, G). Taken together, these analyses indicate that autocrine BMP signaling regulates the architecture and composition of the presynaptic compartment.

Presynaptic, autocrine BMP signaling regulates synaptic vesicle distribution
The preceding analyses demonstrate that autocrine BMP signaling regulates organization of the active zone cytomatrix. Might this pathway regulate additional features of the presynaptic compartment? We turned to the question of small synaptic vesicle (SSV) localization since tight regulation of vesicle number and distribution underpins presynaptic function. We analyzed SSV distribution using the vesicular glutamate transporter (VGLUT) as an SSV marker (Daniels et al., 2004). SSVs cluster at active zones for activity-dependent release and are enriched at presynaptic membranes, leaving bouton interiors relatively less populated with SSVs (Figure 3 A). In contrast, in gbb mutants, we observed that SSVs are more uniformly distributed (Figure 3 B). To quantify this phenotype, we assessed relative SSV distribution by drawing a line through the bouton center and analyzing the fluorescence intensity profile along the line. We defined proper SSV distribution as corresponding to a 50% decrease in VGLUT fluorescence intensity in the bouton interior (Figure 3 H, I). We analyzed only boutons of at least 2 µM diameter to avoid small boutons lacking well-defined SSV distributions. We also calculated integrated VGLUT intensity as an estimate of overall vesicle number.
Indicating that BMP signaling does not regulate overall SSV number at the NMJ, integrated VGLUT fluorescence intensity is not altered in any BMP pathway backgrounds ( We find that loss of Mad results in a more intermediate phenotype, with 48% of boutons displaying normal SSV distribution (Figure 3 A, G, J). Together, these findings argue that presynaptic, autocrine Gbb signaling regulates SSV distribution. Interestingly, LOF mutations in the conserved synaptic adhesion molecule Neurexin-1 display the same phenotype (Rui et al., 2017), suggesting that large-scale spatial organization of SSVs within boutons could be a common function of synapse organizing pathways.

Presynaptic Gbb is necessary for baseline synaptic function
Genetic rescue experiments establish that neuronal Gbb is sufficient to restore normal baseline glutamate release to gbb nulls (Goold and Davis, 2007;James et al., 2014;McCabe et al., 2003).
While these studies demonstrate that neuronal Gbb can support normal transmission, they do not demonstrate that neuronal Gbb normally does serve this function. The abnormalities in active zone structure and vesicle distribution that we observe with loss of neuronal Gbb are consistent with a role for this ligand pool in transmission.
To address this question directly, we measured spontaneous and evoked synaptic potentials via intracellular recordings in gbb RNAi backgrounds. We find that mutants with Gbb removed selectively from motor neurons exhibit a 53% reduction in evoked EJP amplitude, while mutants with Gbb removed from the muscle have a 26% reduction (Figure 4 A-E). The 26% reduction in EJP amplitude with muscle-specific Gbb knockdown was expected as these NMJs display a 22% reduction in overall NMJ size without increased synapse density (Figures 1 and S1). In contrast, the 53% reduction in EJP amplitude with loss of neuron-specific Gbb knockdown cannot be explained by a change in gross NMJ morphology since bouton number is unchanged in this background ( Figure (Goold and Davis, 2007;James et al., 2014;McCabe et al., 2003) to establish a functional requirement for presynaptic Gbb in evoked neurotransmitter release.

Ongoing autocrine BMP signaling organizes the presynaptic compartment
Gbb and Mad are required exclusively during the first instar larval stage for NMJ growth (Berke et al., 2013), arguing for an early critical period for the retrograde pro-growth cue. To define the temporal requirement of presynaptic Gbb in organizing the presynaptic compartment, we utilized the GeneSwitch Gal4 system (GS-Gal4) for inducible expression of Gbb in all neurons via elav-GS-Gal4 (Osterwalder et al., 2001). Our aim was to define the critical period of autocrine Gbb signaling in active zone organization by inducing Gbb expression in neurons in gbb nulls at defined stages of larval development. Thus, we first tested if proper synapse density is rescued by inducing Gbb in neurons during the second instar larval stage (L2) and find that it is restored to wild-type levels ( Figure  Together these findings argue that early retrograde Gbb signaling initiates scaling growth of the NMJ while continuous autocrine Gbb signaling organizes presynaptic terminals. The separable temporal requirements raise the possibility that the autocrine pathway is the predominant BMP pathway active at the L3 NMJ. In this event, autocrine signaling might more efficiently drive BMP signal transduction at this stage. We tested this hypothesis by comparing the effectiveness of pre-and postsynaptic ligand to drive nuclear pMad localization. We find that pMad localization in motor neuron nuclei is strongly reduced in gbb nulls ( Figure S3 A-B, E). And in accordance with our model, while expression of Gbb in the muscle weakly rescues pMad localization in gbb nulls, neuronal Gbb provides significantly greater rescue ( Figure S3 A-E). These results argue that autocrine Gbb signaling is the primary BMP pathway at the L3 stage.

Loss of Gbb phenocopies loss of α2δ-3 at the embryonic NMJ
Our findings argue that Gbb regulates synapse maintenance; however, they do not address whether it regulates initial synapse formation. At 21 hours after egg laying (AEL), embryonic NMJs display hallmarks of mature NMJs such as rounded boutons and localized presynaptic components, including Brp (Rheuben et al., 1999;Yoshihara et al., 1997). Thus, to address whether Gbb is required for embryonic synapse formation, we assessed Brp localization in gbb mutants. In wild type, we find approximately 40 Brp-positive puncta at NMJ 6/7 at 21 h AEL (Figure 6 A, E). We do not find a reduction in the number of Brp-positive puncta in either gbb or wit null embryos (Figure 6 A-C, E), suggesting that Gbb signaling is not required for initial synapse formation. While these NMJs contain an appropriate number of Brp puncta, their gross morphology is markedly abnormal. Whereas wild-type NMJs at this stage are characterized by clusters of boutons, NMJs in gbb and wit mutants are elongated with fewer, smaller boutons. Indeed, we find a roughly 50% reduction in bouton number in gbb and wit mutants relative to controls (Figure 6 F). The motor neuron-specific and muscle-specific drivers in our larval analysis did not enable us to map the cellular requirement of Gbb in this context; however, these findings demonstrate an overall function for BMP signaling in embryonic bouton formation.
This embryonic phenotype caught our attention because it mirrors the phenotype displayed by strong α2δ-3 alleles (Figure 6 D-F) ). α2δ-3 is an auxiliary Ca 2+ channel subunit that traffics and localizes Cacophany (Cac), the Drosophila α1 subunit of mammalian N-and P/Q-type voltage-gated calcium channels, to active zones (Dickman et al., 2008). α2δ-3 proteins contain two disulfide-linked peptides, α2 and δ. While the C-terminal δ peptide is membrane-associated and necessary for α1 subunit association, the large, heavily glycosylated N-terminal α2 domain is exclusively extracellular (Dolphin, 2018). Intriguingly, α2δ proteins are also conserved regulators of synapse formation and function (Dolphin, 2018;Eroglu et al., 2009;Hoppa et al., 2012;Pirone et al., 2014;Tong et al., 2017;Wang et al., 2016;). Yet how α2δ proteins carry out their distinct functions in Ca 2+ channel localization versus synapse formation is not well understood.

Gbb and α2δ-3 have related functions in presynaptic organization
The phenotypic similarity between BMP mutants and α2δ-3 mutants at embryonic NMJs raised the possibility of an underlying functional interaction. Hence, we tested if loss of α2δ-3 results in presynaptic deficits, making use of well-characterized α2δ-3 alleles. α2δ-3 DD106 is a null allele; α2δ-3 DD196 is a hypomorphic allele harboring a stop codon following the α2 peptide; and α2δ-3 k10814 (here called α2δ-3 k ) is a hypomorphic P element allele with reduced expression of wildtype protein (Dickman et al., 2008;Kurshan et al., 2009) (Figure 7 E). Importantly, while bouton formation is blocked in α2δ-3 DD106 homozygotes, α2δ-3 DD196 mutants have normal embryonic NMJs, suggesting a specific role for the extracellular α2 peptide in bouton morphogenesis ). α2δ-3 DD106 homozygotes are late-stage embryonic lethal, so to characterize presynaptic organization at the L3 stage, we analyzed this allele in trans to the weak α2δ-3 k allele, as well as α2δ-3 DD196 in trans to α2δ-3 k . Consistent with α2δ-3's role in Ca 2+ channel localization, both allelic combinations (α2δ-3 DD106/k and α2δ-3 DD196/k ) display strong reductions of Cac at active zones ( Figure S4 A-D).
We next assessed presynaptic organization in α2δ-3 DD106/k and α2δ-3 DD196/k animals. We find that The observed decrease in synapse density in α2δ-3 DD106/k mutants is similar to that reported for another strong α2δ-3 allelic combination (Dickman et al., 2008). α2δ-3 DD106/k mutants also display altered SSV distribution akin to presynaptic Gbb mutants (Figure 7 K-O). In contrast, presynaptic organization appears normal in α2δ-3 DD196/k mutants (Figure 7 A-O). These findings argue that α2δ-3 regulates active zone cytomatrix architecture and SSV distribution and further suggest a selective requirement for the extracellular α2 domain.
Given the phenotypic similarities, we assayed for dominant genetic interactions between BMP pathway mutants and α2δ-3 mutants. In line with a common pathway, α2δ-3 DD106 , gbb 1 double heterozygotes and α2δ-3 DD106 , wit A12 double heterozygotes display decreased Brp-positive synapse density similar to individual single mutants (Figure 7 P). Notably, the strength of the interaction between α2δ-3 DD106 and gbb or wit is as strong as the genetic interaction between gbb and wit (Figure 7 P). Moreover, α2δ-3 DD106 , wit A12 double heterozygotes have deficits in SSV distribution akin to that of the single mutants (Figure 7 Q). Surprisingly, α2δ-3 DD106 , gbb 1 double heterozygotes have normal SSV distribution (Figure 7 Q), hinting at genetic redundancy.
Together, these genetic interactions suggest that α2δ-3 functions with the autocrine BMP signaling pathway.

Gbb overexpression in motor neurons suppresses α2δ-3 mutant phenotypes
α2δ proteins are large, glycosylated proteins residing in the synaptic cleft (Dolphin, 2018), raising the possibility that α2δ-3 influences BMP signaling at the extracellular level. Moreover, our phenotypic analysis of α2δ-3 alleles supports a specific requirement for the extracellular α2 peptide in autocrine BMP signaling. We considered two hypotheses for how α2δ-3 and BMP signaling might interact. First, α2δ-3 might be required for autocrine BMP signaling, perhaps by serving as an obligate co-receptor. Alternatively, α2δ-3 might act in the extracellular space to potentiate autocrine BMP signaling by regulating Gbb levels or diffusion.
To distinguish between these hypotheses, we tested if over-activation of autocrine BMP signaling is sufficient to suppress α2δ-3 phenotypes. We reasoned that if α2δ-3 facilitates autocrine BMP pathway activity by promoting its extracellular availability, then elevated Gbb release might suppress α2δ-3 mutant phenotypes. In contrast, if α2δ-3 is an essential component of the BMP pathway, then elevated Gbb release is not predicted to ameliorate α2δ-3 mutant phenotypes. We find that the decrease in Brp-positive synapse density observed in α2δ-3 DD106/k animals is suppressed by overexpression of Gbb in motor neurons (Figure 8 A-D). Similarly, proper SSV enrichment around the bouton perimeter is restored to α2δ-3 mutants by Gbb overexpression in motor neurons (Figure 8 E-H), suggesting that this phenotype is caused by reduced autocrine Gbb signaling. Together, these findings are consistent with the hypothesis that α2δ-3 normally facilitates autocrine BMP signaling.
Can motor neuron overexpression of Gbb also improve motor function of α2δ-3 mutant animals? To address this question, we compared larval locomotion of α2δ-3 mutants and α2δ-3 mutants with neuronal Gbb overexpression. We find that α2δ−3 DD106/k mutants exhibit dramatically reduced larval locomotion relative to wild-type larvae, moving only 18% of the distance as wild-type larvae over a three-minute interval (Figure 8 I- Together, these data support the hypothesis that α2δ-3 facilitates BMP signaling.

α2δ-3 limits diffusion of Gbb following its activity-dependent release
Extracellular proteins serve diverse functions in BMP pathways. They can antagonize BMP signaling by sequestering BMP ligands from their receptors, or alternatively, promote BMP signaling by serving as scaffolds to facilitate formation of ligand-receptor complexes (Sedlmeier and Sleeman, 2017). We hypothesized that the large glycosylated α2 peptide is wellpositioned to modulate the extracellular distribution of Gbb following its activity-dependent release. In this case, we might observe a shift in extracellular Gbb in the presence and absence of α2δ-3.
To test this prediction, we utilized an assay of activity-dependent Gbb release (James et al., 2014). Presynaptic Gbb release is triggered by high K + stimulation and quantified by comparing Gbb-HA intensity within the presynaptic membrane in unstimulated and stimulated preparations. Following stimulation, we find a roughly 50% decrease in mean Gbb-HA intensity within the presynaptic membrane, indicative of Gbb release (Figure 9 A-B, G; James et al., 2014). We next investigated if loss of α2δ-3 interferes with presynaptic release of Gbb in α2δ-3 mutant backgrounds. We do not find alterations in presynaptic Gbb-HA levels in the absence of stimulation in either α2δ-3 mutant background. Moreover, stimulation leads to comparable reductions of intracellular Gbb-HA in both backgrounds (Figure 9 A-G). The finding that presynaptic Gbb levels and activity-dependent release are independent of α2δ-3 is consistent with the model that α2δ-3 is an extracellular modulator of autocrine BMP signaling.
We then assessed if α2δ-3 acts subsequently to shape the distribution of Gbb following its activity-dependent release. At control NMJs, released Gbb-HA is visible as a cloud surrounding the presynaptic membrane following stimulation (Figure 9 A-B''). We quantified extracellular Gbb by defining a 400-nm margin from the intracellular face of the presynaptic membrane as labeled by HRP into the extracellular space (between dashed lines in A'-D'). At control NMJs, we find a two-fold increase in mean Gbb-HA intensity in this membraneproximal domain following stimulation (Figure 9 H) (James et al., 2014). In contrast, we do not detect an increase in membrane-proximal Gbb following stimulation at α2δ-3 DD106/k mutant NMJs (Figure 9 C-D'', H). To address a specific function for the extracellular α2 peptide, we analyzed the distribution of Gbb-HA following stimulation in α2δ-3 DD196/k mutants since these mutants are predicted to retain the α2 peptide. Unlike α2δ-3 DD106/k mutants, the extracellular distribution of Gbb-HA following stimulation is normal at α2δ-3 DD196/k mutant NMJs (Figure 9 E-F'', H). These findings argue that α2δ-3 serves as a physical barrier to Gbb diffusion to promote activity-dependent autocrine BMP signaling and further support a specific function for the extracellular α2 peptide in the pathway.

Discussion
In this study we uncovered a novel functional interaction between α2δ-3 and an activitydependent, autocrine BMP signaling pathway at the Drosophila NMJ (Figure 10). We find that α2δ-3 mutants display synaptic defects analogous to autocrine BMP mutants, and that α2δ-3 mutant phenotypes are suppressed by over-activating autocrine BMP signaling. We provide evidence that the α2 extracellular domain of α2δ-3 acts as a physical barrier to corral Gbb at the presynaptic membrane upon its activity-dependent release. Here we examine the relationship among α2δ proteins, autocrine BMP signaling, and the synaptic cleft microenvironment in maintaining synapse structure and function.  (Sulkowski et al., 2016). While accumulation of synaptic pMad depends on Wit, it is independent of Gbb, indicating that it is separable from the Gbbdependent autocrine pathway described in this study.

Multiple BMP signaling pathways at the Drosophila NMJ
How are the Mad-dependent pathways distinguished? In other words, how does retrograde BMP signaling regulate Mad-dependent gene transcription to promote scaling growth while autocrine BMP signaling regulates Mad-dependent gene transcription to maintain synapse structure and function? The distinct temporal requirement for the two pathways likely plays a key role to enable signal segregation. While the L1 stage is the critical period for the retrograde pro-growth signal, (Berke et al., 2013), we show that ongoing autocrine BMP signaling maintains synaptic structure. Thus, the signals are largely temporally separable. However, sustained muscle overexpression of Gbb during larval development fails to rescue baseline glutamate release indicating that temporal differences alone do not explain the distinct functions of the two pathways (Goold and Davis, 2007;McCabe et al., 2003). Moreover, we found that at L3, muscle-specific Gbb overexpression is less efficient at driving nuclear pMad localization than motor neuron Gbb overexpression ( Figure S3).
We propose that the temporal dynamics of activity-dependent release of BMP from presynaptic terminals promote robust induction of the downstream BMP cascade. This hypothesis is consistent with work demonstrating that pulsed TGF-β signaling dramatically increases the Smad transcriptional response (Sorre et al., 2014;Warmflash et al., 2012). This model is also supported by our work on the neuronal Gbb-binding protein Crimpy (Cmpy), which traffics Gbb to dense core vesicles (DCVs) for activity-dependent release (James et al., 2014). In the absence of Cmpy-mediated Gbb release, the ability of presynaptic Gbb to promote baseline neurotransmission is impaired. As expected, cmpy LOF mutants share some of the phenotypes described here including aberrant SSV distribution and disrupted embryonic bouton formation (Hoover and Broihier, unpublished). However, cmpy mutant NMJs are dramatically overgrown (James and Broihier, 2011), complicating analyses of synapse structure and function. Interestingly, Cmpy contains an IGF-binding domain and is poised to modulate the function of additional growth factors. Thus, here we focused on Gbb. Together, our findings provide a foundation for delineating how neurons discriminate among diverse secreted cues in their environment.

BMPs, α2δ proteins, and the synaptic cleft microenvironment
The signaling range of BMP ligands in the extracellular space is tightly controlled by a large number of interacting proteins, including ECM components (Sedlmeier and Sleeman, 2017;Umulis et al., 2009;Wharton and Serpe, 2013). Not surprisingly, the ECM serves contextdependent functions in BMP signaling. For example, fibrillins sequester BMPs in the ECM to inhibit signaling (Sengle et al., 2008;, while collagen type IV interacts with BMPs to augment local BMP signaling (Sawala et al., 2015;Wang et al., 2008). Specifically, collagen type IV immobilizes the BMP ligand Dpp to enable short-range signaling in Drosophila. More broadly, the ECM is proposed to generate a structural barrier through which BMP diffusion is limited (Sedlmeier and Sleeman, 2017). Such a barrier function could reasonably play a critical activating role in autocrine BMP pathways, such as the one described here.
Our findings argue that α2δ-3 likely serves as a diffusion barrier to Gbb following its activitydependent release from neurons. α2δ proteins are present in the cleft and contain VWA domains, protein-protein interaction domains commonly found in extracellular scaffolds. The α2 subunit is glycosylated at multiple residues (Dolphin, 2012;, raising the possibility that it could bind Gbb directly, since glycosylated proteins are well-established interactors of BMP ligands in the extracellular space that regulate their range and activity (Hayashi et al., 2009;Negreiros et al., 2018). Regardless of whether α2δ-3 and Gbb interact directly, or via an intermediary, we propose that α2δ-3 is a component of an extracellular scaffold that corrals BMP near the presynaptic membrane to facilitate the ligand-receptor interaction.
In mammals, α2δ-1 promotes excitatory synaptogenesis by acting as a postsynaptic receptor for glial thrombospondin (Risher et al., 2018;Eroglu et al., 2009). At the Drosophila NMJ, we found that overexpression of neuronal Gbb suppresses α2δ-3 mutant phenotypes (Figure 8), indicating that α2δ-3 is not an obligate BMP receptor. Instead, we argue that α2δ-3 facilitates local BMP signaling. Is α2δ-3's only function in the cleft local modulation of autocrine BMP signaling? Mammalian α2δ-1 binds a large number of synaptic proteins in addition to thrombospodin, including integrin subunits, neurexin, neuroligin, and cadherins (Brockhaus et al., 2018), suggesting that α2δ proteins interact widely with synaptic signaling or adhesion complexes. While we found clear phenotypic similarities between autocrine BMP mutants and α2δ-3 mutants, we could not analyze strong allelic combinations of α2δ-3 in late larval development as its function in Cac localization renders strong alleles late embryonic lethal. To determine if α2δ-3 modulates other extracellular signals at the Drosophila NMJ, it would be informative to generate α2δ-3 alleles that specifically delete the α2 domain to distinguish α2δ-3's function in Cac localization from its Cac-independent roles in modulating extracellular signaling or adhesion.

Autocrine signals can provide neurons with continuous feedback on their activity state
In this work, we demonstrate that an activity-dependent autocrine BMP pathway maintains synapse structure and function at the Drosophila NMJ. This pathway is spatially and temporally separable from the long-studied retrograde BMP pathway that initiates scaling growth (Berke et al., 2013;James et al., 2014;McCabe et al., 2003). Target-derived retrograde cues serve critical, conserved functions regulating neuronal survival, connectivity, and synaptogenesis (Zweifel et al., 2005). However, the broad expression patterns and pleiotropic phenotypes of many conserved signaling proteins has obscured a detailed view of their synaptic functions. The increasing ease of generating genetically mosaic animals, in which the protein-of-interest is deleted in a cell-type specific manner, has elucidated the cellular requirements of secreted synaptic signals, and revealed a number of autocrine pathways (reviewed in Herrmann and Broihier, 2018).
The autocrine BMP pathway we describe here is predicted to be transcriptional since it requires Mad. Thus, it does not appear to be a local synapse-specific signal, but rather a transcriptional cascade that regulates expression of genes involved in maintaining synapse structure and function. We propose that this activity-dependent autocrine cue provides ongoing feedback to motor neurons permitting them to maintain their activity state. It will therefore be essential to identify the cohort of Mad transcriptional targets induced by this activity-dependent cascade. Interestingly, BMP4 was recently found to be a presynaptic, autocrine cue in the mammalian hippocampus that signals through a canonical Smad pathway (Higashi et al., 2018). And similar to presynaptic Gbb in Drosophila, BMP4 is trafficked to DCVs and released in response to activity. However, in contrast to the synaptic maintenance function we describe here, BMP4 signaling destabilizes synapses for synaptic elimination.
Thus, activity-dependent autocrine BMP signaling plays a conserved role in regulating synapse density during development, though whether the pathway has a stabilizing or destabilizing effect on synapses is context-dependent. It will be essential to determine the molecular mechanisms through which activity-dependent BMP signal transduction exerts its broad range of effects on synapse structure and function.

Immunohistochemistry and immunofluorescence
Wandering third-instar (L3) larvae were dissected and processed for immunofluorescence as previously described (James and Broihier, 2011). Briefly, larvae were dissected in PBS and For embryonic experiments, late-stage embryos were transferred using a small paintbrush to piece of double-sided tape. The embryos were rolled out of their chorion and placed on a plate in a small drop of PBS. Late-stage embryos have trachea fully formed and were removed from their vitelline membrane with forceps. Embryos were pinned at the head and tail using sharpened pins. Tungsten wire was used to pierce the body wall, and scissors were used to make a lateral cut down the stretched embryo. PBS was replaced with Bouin's fixative for 5 min. The embryo was then rinsed with PBS and placed into a mesh basket surrounded by PBS in a small Petri dish. The embryos in the basket were washed three times for 15 min in PBT while slowly shaking. Embryos were blocked with PBT with 10% NGS (PBTN) for 10-30 min and incubated with primary antibody diluted in PBTN overnight at 4°C. Samples were then washed in PBT for 45 min while shaking, blocked in PBTN for 10-30 min, and incubated with secondary antibody diluted in PBTN for 2 hr at RT. Embryos were washed in PBT three times for 15 min each and PBS twice for 5 min each prior to being mounted in 60% glycerol.
For larval CNS experiments, wandering third-instar larvae were dissected and processed for immunofluorescence as previously described (Politano et al., 2019). Briefly, larvae were dissected in Schneider's media (Sigma) supplemented with 10% FBS (Sigma) and 10U/mL Penicillin-Streptomycin (Thermo Fisher Scientific), then fixed in 4% PFA for 20 min. They were transferred to borosilicate tubes and washed in PBS and blocked in PTN (PBS, 0.1% Triton X-100, and 1% NGS) at RT on a rocker for at least 1 hr prior to overnight incubation in primary antibodies diluted in PTN at 4°C. Secondary antibody incubation in PTN was done for 3 hr at RT after primary antibody was washed out with PBS. Samples were mounted in ProLong Gold Antifade with DAPI (Thermo Fisher Scientific) after washing in PBS.

RU486-GeneSwitch
Induction of Gal4-dependent transcription in GeneSwitch-Gal4 lines is induced by feeding animals with RU-486 since Gal4 is rendered steroid-inducible via fusion with the human progesterone receptor. Per Osterwalder et al., 2001, adults and their larval offspring were raised on non-RU486 food. L2 or early L3 larvae were transferred to molasses plates containing 25 mg/mL RU486 (mifepristone; Sigma Aldrich) and topped with a yeast paste made from 1 g dried yeast and 2 mL 50 mg/mL RU486 in water. RU486 was prepared as a stock solution at 10 mg/mL in ethanol. Control larvae were fed food containing equivalent concentrations of ethanol as vehicle at the same developmental stage. We ensured that in the absence of RU486, gbb -/-; Elav-GS>Gbb animals display active zone phenotypes similar to that observed in gbb nulls. Indeed, without drug, we saw identical structural phenotypes as observed in gbb nulls, confirming RU486-dependence of transgene expression.

Gbb-HA release following neuronal stimulation
Each larva was pinned at the head and tail while in cold hemolymph-like HL3 saline. An incision was made laterally, and one pin was moved closer to the other, so the body wall can loosen. The HL3 was replaced with either cold 2 mM K + or 90 mM K + buffer for 5 min. The larva was next washed with cold PBS and quickly dissected.

Image acquisition and quantification
Larval NMJs were imaged on a Zeiss LSM 800 confocal microscope with 100x 1.4 NA and 40x 1.3 NA oil-immersion objectives. Complete z-stacks were acquired with optimized settings to ensure that oversaturation did not occur. Confocal settings were held constant across all of the genotypes being tested in an experiment. After imaging samples for Brp rings, Zeiss ZEN deconvolution software was used on images. Maximum z-stack projections were created by ImageJ (National Institutes of Health). No modifications to any images were made prior to quantification. Larval NMJ4 boutons in segments A2-A4 were manually counted using an Axioplan 2 equipped with a Colibri 2 illumination system using the 100x 1.4 NA oil-immersion objective.
Synapses, defined as the tight apposition between Brp and GluRIII puncta, were manually scored at NMJ4 terminal boutons, and synapse density was calculated by the number of Brp+ synapses per bouton area in square microns. For all Brp + synapse density quantifications, n is the number of NMJ4 boutons scored.
The diameters of isolated planar-oriented Brp rings were measured from confocal images of Brp-stained terminal boutons at NMJ4 that had been deconvolved using Zeiss ZEN software.
A line was manually drawn through each Brp ring in ImageJ, and the Brp mean pixel intensity was plotted across the line. The distance from intensity maximum to intensity maximum was determined in the plot window, and this calculation was determined to be the Brp ring diameter (Owald et al., 2012). For all Brp ring analyses, n is the number of NMJ4 boutons.
DVGLUT localization was assessed by manually drawing a line through each bouton in ImageJ and DVGLUT mean pixel intensity was plotted across the line. If there were two peaks in the expression plot, and the trough of the plot was less than half of the smallest peak, the bouton was marked as having properly localized DVGLUT. We performed this analysis only on boutons larger than 2 microns in diameter, as smaller boutons were often completely filled with synaptic vesicles. The percent of boutons exhibiting properly localized synaptic vesicles compared to the total number of boutons analyzed at single NMJ4 was calculated. For all DVGLUT localization analyses, n is the number of NMJs scored. Concurrently, we measured the integrated pixel intensity over the same drawn line as an estimate of the number of synaptic vesicles. For all integrated DVGLUT analyses, n is the number of NMJ4 boutons larger than 2 microns in diameter.
To determine pMad intensity within motor neuron nuclei, VNCs were imaged at the same confocal settings and equal thickness maximum projections of the dorsal portion of the VNCs were made for each. Regions of interest were determined using Eve immunoreactivity. For each Eve+ motor neuron nucleus within a single VNC, the pMad and Eve pixel intensities were each measured. After subtracting background immunofluorescence from each channel, the pMad intensity was divided by the Eve intensity, providing a ratio of pMad to Eve intensity for each Eve+ motor neuron in a single VNC. Ratios were then averaged for each genotype. For the pMad analysis, n is the number of motor neuron cells scored.
To quantify intracellular Gbb-HA, mean pixel intensity was determined inside the inner neuronal membrane for samples incubated in either the 2 mM or 90 mM K + buffer.
Background HA fluorescence was averaged and subtracted from each projection. The neuronal membrane as defined by HRP staining is roughly 200 nm itself. To calculate Gbb-HA proximal to the neuronal membrane, Gbb-HA mean pixel intensity was measured between the intracellular face of the presynaptic membrane as labeled by HRP and a distance of 200 nm away from the extracellular face of the neuronal membrane, a distance of 400 nm in total.
Again, background HA fluorescence was averaged and subtracted from each projection.

Electron microscopy and analysis
Third-instar larvae were dissected and fixed for 2 hr in 2% glutaraldehyde and 2.5% paraformaldehyde. Following post-fixation in 1% osmium tetroxide, the specimen was dehydrated through a graded ethanol series and embedded in Eponate 12. The block was sectioned with a Leica EM UC6 ultramicrotome and stained with uranyl acetate and lead citrate. Images were collected on a FEI Tecnai G2 Spirit BioTWIN transmission electron microscope at the Cleveland Clinic. Electron micrographs of NMJs were taken from NMJ 6/7 and segments A2-A3 in at least two larvae of each genotype. Active zones were identified as linear electron densities found between pre-and postsynaptic membranes. High-magnification images (>125,000) were used to determine the presence of membrane ruffling and/or a T-bar, which was defined as an electron-dense rod surrounded by vesicles and localized to the presynaptic membrane.
For each cell, mean miniature excitatory junctional potential (mEJPs) amplitudes were calculated using Mini Analysis (Synaptosoft) from all consecutive events during a one-minute recording in the absence of stimulation. EJPs were stimulated at a frequency of 0.2 Hz using an isolated pulse stimulator 2100 (A-M Systems), with intensity adjusted to consistently elicit compound responses from both type Ib and Is motor neurons. At least 25 consecutive EJPs were recorded for each cell and analyzed in pClamp to obtain mean amplitude. Quantal content was calculated for each recording using the ratio of mean EJP amplitude to mean mEJP amplitude. Recordings were limited to cells with a resting membrane potential between -50 mV and -85 mV and muscle input resistance greater than 5 MΩ.

Locomotion assay
Prior to imaging, 150 mm Petri dishes were prepared, each with a thin layer of 1% agarose and black fountain pen ink (Pelikan) as a dye. L3 larvae were grown at 25°C and 4-7 larvae were then transferred to the middle of a Petri dish with a few drops of dH2O to keep the larvae moist. Larvae were imaged for 3 min. Movies were converted to 0.1 Hz (0.1 frames per second) and then larval crawling distance was measured every 10 seconds using the ImageJ plugin Manual Tracking. For the locomotion analysis, n is the total number of larvae scored per genotype in over four experimental trials.

Statistical analyses
All statistical analyses were performed using Prism 6 (GraphPad Software). Each graph presents the data as mean + SEM with the n value at the bottom of each column. All pairwise comparisons were made using a Mann-Whitney test. For determining significance between groups of three or more genotypes, a one-way ANOVA followed by Dunnett's post hoc test was applied on normally distributed data. A nonparametric Kruskal-Wallis test followed by a Dunn's multiple comparison test was performed on datasets that did not have a normal distribution. Analyses were conducted with the relevant Gal4 lines or heterozygotes as controls. In all figures, p-values are as follows: *p<0.05, **p<0.01, and ***p<0.001. No significant difference is denoted as n.s.