Dscam2 suppresses synaptic strength through a PI3K-dependent endosomal pathway

Cell recognition molecules are key players in the development of the nervous system. Some regulate connections between neurons by promoting axon guidance, target selection, and boundary formation, whereas others critically regulate synaptic physiology, growth, and maintenance (Shapiro et al., 2007). The number of cell recognition molecules encoded by the human genome is small (Almén et al., 2009) compared with the trillions of connections in the brain. Several different mechanisms could account for these numerical differences including combinatorial interactions (Thu et al., 2014), alternative splicing (Nilsen and Graveley, 2010), and the ability of these proteins to perform multiple functions (Neuhaus-Follini and Bashaw, 2015).

Dscam2 has been primarily studied in the Drosophila visual system, where it functions as a tiling receptor and is required for both postsynaptic specificity and dendritic targeting of lamina neurons (Millard et al., 2007; Millard et al., 2010; Tadros et al., 2016). Alternative splicing of Dscam2 produces two distinct extracellular protein isoforms, Dscam2A and Dscam2B. These proteins differ at a single immunoglobulin domain, and this domain confers biochemical specificity for homophilic recognition (Millard et al., 2007). Alternative splicing of Dscam2 is cell type specific; most neurons analyzed thus far express either Dscam2A or Dscam2B, but not both (Lah et al., 2014; Tadros et al., 2016). This cell type specificity is driven in part by the splicing factor Muscleblind, which represses the selection of Dscam2 exon 10A and is necessary for the selection of exon 10B (Li and Millard, 2019). Because homophilic binding does not occur between different isoforms (Millard et al., 2010; Özkan et al., 2013), only neurons that express the same isoform can induce signaling downstream of homophilic interactions. Thus, cell-specific expression of Dscam2 isoforms could be analogous to having two different genes that perform similar functions in different cells. One prediction of this model is that expression of a single isoform in all Dscam2-positive cells should generate the same gain-of-function phenotypes with either isoform. Indeed, this is what was previously observed in visual system neurons (Lah et al., 2014; Kerwin et al., 2018; Tadros et al., 2016).

In addition to regulating connectivity between neurons, cell recognition molecules can also modulate synaptic processes (Thalhammer and Cingolani, 2014). For example, FasII was originally discovered as an axon guidance molecule required for fasciculation (Grenningloh et al., 1991; Lin et al., 1994) but was later shown to play a role in synaptic plasticity (Stewart et al., 1996; Schuster et al., 1996). The ability of neurons to modulate neurotransmitter release within a dynamic range is critical for processes that require synaptic potentiation and depression (Yang and Calakos, 2013), such as learning (Lynch, 2004). If a synapse were to function at maximal capacity, an increase in synaptic strength (potentiation) would not be possible. Thus, intracellular signaling pathways that inhibit neurotransmitter release function as “brakes” that permit further synaptic potentiation (Cooper and Bear, 2012), and cell surface receptors that regulate these pathways form a line of communication between neurons and their extracellular environment. Several studies have demonstrated that endosomes play a central role in this process by integrating intracellular signals that modulate neurotransmitter release (Hauswirth et al., 2018). Rab5, for example, is required to maintain early endosome identity and morphology but has been found to also act as a rate-limiting molecule for presynaptic neurotransmitter release (Wucherpfennig et al., 2003). High levels of Rab5 at synapses are associated with increased release, and impairment of Rab5 causes a decrease in neurotransmitter output (Wucherpfennig et al., 2003). Further, loss of the Rab35 GTPase-activating protein, Skywalker, results in increased neurotransmitter release and aberrant endosomal morphology (Uytterhoeven et al., 2011). Although we are beginning to understand the downstream intracellular processes that enact changes to synaptic strength in order to maintain neuronal plasticity (Delvendahl and Müller, 2019; Frank, 2014), much less is known about the upstream cell recognition molecules that regulate their activity.

Previous studies indicate that Dscam2 predominantly functions to regulate neuronal connectivity during development. Whether it can influence the physiology of synapses after they form is currently not known. In this study, we discovered that Dscam2 is expressed in motor neurons of Drosophila larvae and exploited this well-characterized system to ask whether Dscam2 influences various aspects of synaptic physiology. When we stimulated motor neurons and recorded postsynaptic muscle responses, we found that Dscam2 mutant synapses were stronger than controls. Interestingly, this effect of suppressing synaptic strength was also observed when motor neurons expressed the incorrect isoform of Dscam2 (Dscam2A). We identified the PI3K enhancer, Centaurin Gamma 1A (CenG1A), as a component of the Dscam2-mediated synaptic pathway and found changes in specific phosphoinositides and endosomal compartments in the Dscam2 mutants that were previously implicated in regulating synaptic strength (Uytterhoeven et al., 2011; Fernandes et al., 2014; Wucherpfennig et al., 2003). To understand how these changes in endosomal compartments led to increased neurotransmitter release, we used electron microscopy to analyze synaptic vesicle distribution at active zones and found an increase in the Dscam2 mutants. Our results indicate that Dscam2 regulates a phosphatidylinositol-3 kinase (PI3K)-dependent endosomal pathway that suppresses the deposition of synaptic vesicles at active zones, providing a compelling example of how an alternative isoform of a developmental molecule can be temporally reemployed for unique purposes in the nervous system.

Using previously described isoform-specific transcriptional reporter lines (Lah et al., 2014), we first characterized Dscam2 isoform expression within the larval ventral nerve cord (VNC) and peripheral nervous system (PNS). Dscam2A expression was restricted to the VNC, whereas Dscam2B was expressed in the VNC and in neurons that project through peripheral nerves (Fig. 1, A and B′). This was confirmed by co-labeling neuromuscular junctions (NMJs) with the neuronal membrane marker anti-HRP and the Dscam2A reporter (Fig. 1, C–C′′). By contrast, Dscam2B was expressed in the VNC (Fig. 1 B′), motor neurons (Fig. S1 A), and peripheral body-wall sensory neurons (Fig. S1 A′). Dscam2B was detectable in ∼50% of motor neurons (Table S1), including the type Ib motor neuron that contacts muscles 6 and 7, known as MN6/7-Ib (Fig. 1, D and D′′). Interestingly, Dscam2B was not detectable in the type Is motor neuron that also connects to these muscles, known as MNSNb/d-Is (Fig. 1, D–D′′).

To determine whether Dscam2 functions in the development of the VNC and peripheral neurons, we analyzed Dscam2 isoform expression throughout embryonic development. Most motor neurons pass major developmental milestones between 12 and 16 h after egg laying (AEL), including responsivity to the major excitatory neurotransmitter acetylcholine and maturation of currents that underlie action potential firing (Broadie and Bate, 1993; Baines and Bate, 1998). Dscam2B was detectable in two small groups of segmentally repeating peripheral and central neurons beginning at 12–16 h AEL (Fig. S1, G–J′), but both isoforms were much more broadly expressed beginning at larval hatching 8 h later (Fig. S1, B–F′). This was interesting, given that by this stage all motor neurons have undergone dendritogenesis and have formed functional synaptic connections with their target muscles (Landgraf et al., 1997). Thus, detectable expression of Dscam2 does not largely coincide with any major developmental milestone in motor neurons other than hatching. This makes it unlikely that Dscam2 plays a critical role in the development of motor neuron synapses.

Dscam2 regulates specificity at photoreceptor synapses in collaboration with Dscam1 (Millard et al., 2010), and cell-specific regulation of its extracellular isoforms is necessary for attaining appropriate axon terminal size in lamina neurons (Lah et al., 2014). Whether Dscam2 is directly required for synaptic maintenance or function, however, has not been investigated. We therefore performed intracellular electrophysiological recordings at the NMJ (Fig. 1 E), where both synaptic morphology and physiology have been well characterized (Menon et al., 2013). To investigate a role for Dscam2 in synaptic physiology and whether the two different isoforms function equally in this process, we assessed three different Dscam2 genotypes: null animals (Dscam2^null) and two different transgenic lines of flies that express a single isoform from the endogenous Dscam2 locus (Dscam2A and Dscam2B; a full description of single isoform knock-in lines appears in Lah et al. (2014). This allowed us to determine whether any changes to neuronal physiology were identical in the two single isoform lines, as would be expected if the isoforms function similarly in different cells, and to distinguish between loss- and gain-of-function effects by comparing them with Dscam2^null animals.

To record spontaneous (miniature) excitatory junctional potentials (mEJPs), we used sharp electrodes to impale muscle 6 (Fig. 1 E). Control and Dscam2^null animals exhibited similar mEJP frequencies, whereas the mEJP frequency was dramatically increased in Dscam2A and Dscam2B (Fig. 2, A and C). We also observed an increase in the mEJP amplitude in both Dscam2A and Dscam2B single isoform animals, but not Dscam2^null, relative to controls (Fig. 2, B and C). These observations suggest that Dscam2 is not required to regulate spontaneous neurotransmitter release per se. Instead, removing cell-specific Dscam2 isoform expression causes a gain-of-function phenotype that promotes spontaneous neurotransmitter release. This effect is likely caused by inappropriate Dscam2 interactions between neurites within the VNC, such as motor neuron dendrites and interneuron axons, which normally express different isoforms.

To test whether Dscam2 plays a role in regulating evoked neurotransmission, we stimulated motor neurons while recording postsynaptic excitatory junctional potentials (EJPs) in muscle. To investigate different properties of neurotransmission, we tested a range of extracellular calcium concentrations ([Ca²⁺]_o; 0.5, 0.75, 1.5, and 3 mM). If Dscam2 were to influence action potential propagation or impair exocytic machinery, then changes in the amplitude of evoked potentials would be expected at higher, saturating levels of [Ca²⁺]_o, whereas if it were to directly affect neurotransmitter release, effects would be most apparent at lower ranges of [Ca²⁺]_o (del Castillo and Katz, 1954a; Jan and Jan, 1976). We did not observe any significant effects at high [Ca²⁺]_o (1.5 and 3 mM) for the Dscam2^null and single isoform lines, suggesting that Dscam2 does not regulate action potential propagation or exocytic machinery in motor neurons (Fig. S2 A). Conversely, strong effects were observed at low [Ca²⁺]_o (0.5 mM), suggesting that Dscam2 influences calcium-dependent aspects of neurotransmitter release. At this concentration of [Ca²⁺]_o, used in all following experiments, unless otherwise stated, the EJPs, corrected for nonlinear summation (EJP′) in Dscam2^null animals, were increased by 50% compared with controls (Fig. 2, D and E), demonstrating that Dscam2 normally weakens the evoked synaptic response. Further, Dscam2A animals displayed a 74% increase in EJP′ amplitude, whereas Dscam2B animals had responses that were not significantly different from controls (Fig. 2, D and E). We found that muscle resting membrane potential, EJP width at half maximum, and rise time were not different between any of the genotypes (Fig. S2, D, E, and F, respectively), highlighting the specificity of Dscam2 for neurotransmitter release. Furthermore, since Dscam2A single isoform animals exhibited a loss-of-function phenotype and motor neurons normally express Dscam2B, this result suggested that Dscam2B is required to suppress synaptic strength.

We next calculated the number of synaptic vesicles released in response to a single presynaptic depolarization (quantal content) by dividing the mean amplitude of the EJPs by the mean amplitude of the mEJPs for each genotype. For the Dscam2 single isoform mutants, we used the median of mEJP amplitudes because they displayed skewed amplitude frequency distributions (Fig. S2 C). Quantal content was higher in both Dscam2^null and Dscam2A animals relative to control and was unchanged in Dscam2B animals (Fig. 2 F). Variance-amplitude plots indicated that this change was predominantly driven by an increase in the probability of neurotransmitter release (Fig. S2 B). These data demonstrate that Dscam2 normally acts to restrict synaptic strength at the NMJ by suppressing the number of vesicles released in response to an action potential and that Dscam2B, the isoform normally expressed in motor neurons, but not Dscam2A is able to carry out this function.

To validate the synaptic effects observed in Dscam2 mutant animals, we used a paired pulse assay, which measures the extent to which evoked release is influenced by changes in release probability and presynaptic calcium-dependent priming (del Castillo and Katz, 1954b; Dudel and Kuffler, 1961). We found that across all measured interstimulus time intervals (20, 50, and 200 ms) the paired-pulse ratio (PPR) was not significantly different between control and Dscam2B animals; however, at 20 ms and 50 ms, it was significantly reduced in both Dscam2^null and Dscam2A animals (Fig. 2 G). Previous studies have shown that weak synapses tend to facilitate when exposed to paired stimuli, whereas strong synapses do not (Kurdyak et al., 1994). Thus, the PPR results suggest that loss of Dscam2 results in stronger synapses and provide further evidence that Dscam2B can perform a function in motor neurons that Dscam2A cannot.

To explore the extent to which the Dscam2 loss-of-function synaptic effect was autonomous to motor neurons, we knocked down Dscam2 using upstream activating sequence (UAS) RNAi driven by different GAL4s and measured mEJP, EJP′, and quantal content. We found that both pan-neuronal and motor neuron–specific knockdown of Dscam2 phenocopied the null phenotype, whereas knockdown in muscle (BG487-GAL4) had no effect (Fig. 2, H–J). These data highlight the specificity for the requirement of Dscam2 in regulating synaptic strength and suggest that this requirement is autonomous to motor neurons.

NMJ structure is a strong determinant of synaptic strength. One possible explanation for the neurotransmission defects observed in Dscam2 mutants is that Dscam2 might regulate growth of the NMJ, which in turn produces downstream changes in synaptic strength. If this is the case, the increased EJP′ and quantal content effects in Dscam2^null larvae may be a result of larger NMJs. However, anti-HRP revealed a small but significant reduction in the total number of boutons per NMJ in Dscam2^null, Dscam2A, and Dscam2B animals relative to control (Fig. 3, A–E). Muscle size, measured using phalloidin, was also modestly reduced in the single isoform lines, but not in Dscam2^null larvae (Fig. 3 E′). Given that there were fewer boutons in Dscam2^null and Dscam2A animals, these data argue against Dscam2 regulating synaptic strength via NMJ morphology.

We next assessed a panel of synaptic markers in the Dscam2 mutant lines, including the scaffolding protein brp, which marks active zones (Fig. 3, K–N), the synaptic vesicle protein Synaptotagmin 1 (Syt1; Fig. S3, F–I), and postsynaptic proteins discs-large (dlg; Fig. S3, A–D) and glutamate receptor subunit IIA (GluRIIA; Fig. 3, P–S). None of these markers were different in Dscam2^null larvae compared with controls, failing to provide any structural or molecular correlates for the increased synaptic strength in these animals (Fig. 3, J, O, and T; and Fig. S3, E and J). Conversely, effects were observed in the single isoform lines, commonly at a similar magnitude for both, including increased bouton size (Fig. 3 J), increased number of brp-positive puncta per bouton (Fig. 3 O), and increased diameter of dlg immunoreactivity surrounding boutons (Fig. S3 E). An increase in the number of active zones per bouton in Dscam2A and Dscam2B is a significant synaptic change given that it increases the number of sites at which neurotransmitter release can occur. This potentially explains why we detected an increase in mEJP frequency in the single isoform lines. These effects were not observed in the type Is motor neurons that also contact muscles 6/7, but do not express Dscam2 (Fig. 3 J′, O′, T′; and Fig. S3, E′ and J′), suggesting that the gain-of-function effects are autonomous to Dscam2-expressing neurons.

To gain insight into the signaling pathways through which Dscam2 suppresses synaptic strength, we searched the literature for mutations that result in altered evoked neurotransmitter release without affecting spontaneous release or NMJ morphology. The resultant list was short, given how few genes have been found to suppress synaptic strength in Drosophila. We then performed a reverse genetic screen by crossing these mutations into a Dscam2 mutant background and assessing spontaneous and evoked release at the NMJ. Of the eight genes we tested, we identified Centaurin Gamma 1A (CenG1A), a PI3K enhancer with a GTPase-activating domain (Homma et al., 2014), as a candidate (Fig. S3 K). CenG1A has homology to vertebrate PIKEs, which play numerous roles in regulating cell signaling by enhancing PI3Ks, binding to membrane receptors, and suppressing apoptotic pathways (Chan and Ye, 2012). A loss-of-function allele of Drosophila CenG1A (Gross et al., 2015) was previously shown to increase EJP amplitude and quantal content at the larval NMJ (Homma et al., 2014). We confirmed these results and found that the magnitude of the EJP′ increase was very similar between CenG1A and Dscam2 single mutants (Fig. 4, A–C).

We next tested heterozygotes (Dscam2^null/⁺ and CenG1A^EY01217/⁺) and surprisingly found that EJP' amplitude was consistently elevated in each heterozygote mutant alone. This made it difficult to determine whether the genes were in the same pathway using double-heterozygote combinations because the difference between hetero- and homozygotes was small; therefore, additive and synergistic effects would be expected to be the same. Indeed, double-heterozygotes exhibited an increase in EJP′ amplitude and quantal content that was similar to that of either mutant alone and significantly different from that of controls (Fig. 4, A–C). Consistent with a genetic interaction between these two genes, however, double mutants for Dscam2 and CenG1A were not different from either single homozygous mutant alone (Fig. 4, A–C). This argued against two parallel pathways that independently regulate synaptic strength, as additive effects would have resulted in an EJP′ increase that was 2× higher than the single mutants alone. Thus, our genetic results suggest that Dscam2 and CenG1A may act within the same pathway to suppress synaptic strength.

One possible model for the interaction between Dscam2 and CenG1A is that CenG1A expression, localization, or activity is regulated by Dscam2. Given that CenG1A is a PI3K enhancer, loss of this gene is expected to affect the levels of phosphoinositides associated with PI3K activity (Gross et al., 2015). In particular, PI(4,5)P₂ (PIP₂) is a substrate for PI3K, so a decrease in PI3K activity should result in its accumulation. We therefore investigated PIP₂ levels at control and Dscam2^null NMJs by driving expression of a well-characterized PIP₂ reporter (UAS-PLCδPH-mCherry) in motor neurons (Verstreken et al., 2009). Quantification of PLCδPH-mCherry immunofluorescence intensity revealed a dramatic enrichment (threefold increase in MN6/7-Ib terminals) of PIP₂ at Dscam2^null boutons relative to controls (Fig. 4, D–F), providing evidence that Dscam2 promotes PI3K activity, likely through CenG1A, at motor neuron axon terminals. Indeed, PIP₂ levels were also increased at CenG1A mutant terminals (Fig. 4, G–I′). Interestingly, the increase in PIP₂ in Dscam2^null animals was observed to a lesser extent in Is boutons (Fig. 4 I′), suggesting that this effect is nonautonomous to Dscam2-expressing cells.

Although an increase in PIP₂ can be interpreted as evidence for reduced PI3K activity, it might conversely be the result of increased activity of proteins involved in PIP₂ synthesis, such as PI4P5K (Ishihara et al., 1996). To more directly test the involvement of PI3K, we recorded EJP′ amplitude in control and Dscam2^null animals in the presence of two different PI3K inhibitors, Wortmannin and LY294002. The effects of PI3K suppression on neurotransmitter release have been well characterized and demonstrate different outcomes depending on whether PI3K is acutely or chronically inhibited. For example, acute suppression of PI3K using drugs typically results in reduced neurotransmitter release (Cousin et al., 2003; Thyagarajan et al., 2014), whereas chronic suppression using transgenic approaches has the opposite effect, resulting in increased neurotransmitter release (Howlett et al., 2008). Our results with Dscam2 are consistent with a chronic inhibition of PI3K, resulting in an increase in synaptic release. We reasoned that chronic inhibition of PI3K in Dscam2^null mutants would make these animals less sensitive to PI3K inhibitors. We found that in controls increasing concentrations of both Wortmannin and LY294002 caused a reduction in EJP′ amplitude, consistent with previous studies (Cousin et al., 2003; Thyagarajan et al., 2014). In Dscam2^null mutants, however, this decrease was not observed, even at concentrations of LY294002 that produced an ∼60% drop in EJP′ amplitude in controls (Fig. 4, J and K). Thus, loss of Dscam2 results in a robust synaptic resistance to PI3K inhibitors, strongly implicating PI3K in the Dscam2 pathway for regulation of synaptic strength.

Endosomal dynamics have been intimately linked to phosphoinositide levels (Jones and Clague, 1995; Tan et al., 2015) and synaptic strength (Uytterhoeven et al., 2011; Fernandes et al., 2014). We therefore next explored whether the increase in synaptic strength in the Dscam2 mutants could be linked to changes in the synaptic endosome system. Early endosomes contain phosphatidylinositol (PI) phosphorylated at position 3, PI(3)P, on their membranes, and this lipid plays a critical role in recruiting components of the endosome fusion machinery (Wucherpfennig et al., 2003). PI(3)P is produced when PI3K phosphorylates PI on endosome membranes (Gillooly et al., 2003). To visualize PI(3)P in Dscam2^null animals, we expressed a well-characterized marker consisting of tandem FYVE domains fused to GFP (2xFYVE-GFP) that bind specifically to PI(3)P (Wucherpfennig et al., 2003). Quantification of 2xFYVE-GFP immunofluorescence intensity in Dscam2^null boutons revealed a significant decrease compared to control (Fig. 5, A–E′). The penetrance of this phenotype was variable, with some boutons exhibiting undetectable labeling whereas others were labeled quite well, and the average decrease was ∼30% (Fig. 5, E and E′). Together, these experiments demonstrated that loss of Dscam2 causes depletion of PI(3)P at motor neuron axon terminals.

After exocytosis, one pathway for membrane recovery is through an endocytic process that produces preendosomal intermediates, which then fuse with endosomes. Synaptic vesicle proteins are sorted at these stations, and new vesicles form and are added to the releasable pool in the nerve terminal (Wucherpfennig et al., 2003; Watanabe and Boucrot, 2017). Early endosome accumulation is therefore dependent on (1) the amount of traffic from the early endosome to synaptic vesicle pool, (2) endocytic processes that generate preendosomal intermediates, and (3) activity of protein complexes that promote fusion between these intermediates and the endosome (Gorvel et al., 1991; Wucherpfennig et al., 2003; Helenius et al., 1983). We reasoned that if the decrease in 2xFYVE-GFP immunofluorescence intensity in Dscam2^null mutants was due to increased traffic from endosomes to the synaptic vesicle pool, then stimulating exocytosis should result in a further decrease due to increased demand for generation of synaptic vesicles. To test this, we measured 2xFYVE-GFP immunofluorescence intensity following a KCl-mediated stimulation paradigm to promote robust exocytosis. Interestingly, following stimulation with KCl, 2xFYVE-GFP immunofluorescence intensity was no longer different between control and Dscam2^null animals (Fig. 5, E and E′). This suggested that the decrease must be explained by defects in either endocytosis or in fusion of preendosomal intermediates (Wucherpfennig et al., 2003; Watanabe et al., 2013a, 2013b). To distinguish between these two options, we performed a KCl-mediated FM4-64 styryl dye uptake assay, which is a reliable method for investigating gross defects in endocytosis (Verstreken et al., 2008). Following 1 min of stimulation and dye uptake, we found that the average immunofluorescence intensity of FM4-64 in axon terminals was not different between Dscam2^null and control animals (Fig. 5, F–H′). This strongly suggested that the decreased 2xFYVE-GFP signal could not be explained by defective endocytosis, leaving defective fusion between preendosomal intermediates and maturation to PI(3)P-positive endosomes as the most likely explanation for the decreased 2xFYVE-GFP in Dscam2^null mutants.

To determine if there were ultrastructural correlates to the endosomal effects that we had observed at the level of light microscopy, we turned to electron microscopy (Fig. 6, A and B). We analyzed synaptic vesicles, dense-core vesicles (DCVs; Fig. 6 C), and several structures attributed to endosomal trafficking including irregularly shaped/tubular endosomes (Fig. 6 D), multivesicular bodies (MVBs; Fig. 6 E), and large (70–80-nm diameter) vesicles (Fig. 6 F). These large vesicles have been shown by others to represent preendosomal intermediates (Wucherpfennig et al., 2003; Watanabe et al., 2013a, 2013b). The density of MVBs was unaffected (Fig. 6 G), suggesting late-stage endosomal maturation is not under the control of Dscam2. Conversely, Dscam2^null boutons displayed a significant increase in the density of both large vesicles and irregularly shaped/tubular endosomes compared with control boutons (Fig. 6, A, B, H, and I). The near-twofold increase in large vesicle density in Dscam2^null mutants is consistent with a model wherein Dscam2 promotes fusion of preendosomal intermediates by controlling the level of PI(3)P on their membranes. PI(3)P is necessary for recruiting critical components of the fusion machinery, such as Rab5 (Gillooly et al., 2003). Interestingly, we also found an approximately threefold increase in the density of DCVs in Dscam2^null boutons compared with controls (Fig. 6 J). Given that exocytosis of DCVs is dependent on PI(3)P (Meunier et al., 2005), this finding is consistent with Dscam2 and CenG1A promoting the activity of PI3Ks at synapses. Interestingly, we also found increased density of DCVs in Dscam2A and not in Dscam2B single isoform larvae (Fig. S4 F), providing additional evidence for this model.

Although we were able to link loss of Dscam2 to changes in endosomes at the NMJ, we had yet to understand how this relationship affected neurotransmission. To investigate this aspect at the level of synaptic ultrastructure, we analyzed the density and size of synaptic vesicles in boutons and the number of synaptic vesicles clustered around active zones, which are marked by electron-dense T-bar structures in flies (Meinertzhagen and O’Neil, 1991). We found that synaptic vesicle density and size was not different between control and Dscam2^null (Fig. 6, K and N), consistent with the lack of mEJP frequency and amplitude phenotypes in mutant animals. Interestingly, however, we observed an ∼40% increase in the number of synaptic vesicles within 200 nm of active zones in Dscam2^null animals relative to control (Fig. 6, L, M, and O). An increase in the number of vesicles around release sites has been shown to correlate with increased quantal content at Drosophila synapses (Bruckner et al., 2012). To confirm this finding in a different Dscam2 background, we also analyzed the single isoform lines. Given that Dscam2A, but not Dscam2B, larvae exhibited an increase in EJP′, we expected that the former would have more synaptic vesicles at active zones whereas the latter would be similar to control. This is indeed what we observed (Fig. S4, O, P, and R). We conclude that Dscam2 suppresses synaptic strength through a PI3K-dependent endosomal pathway that reduces the number of synaptic vesicles at active zones.

Cell recognition molecules play a crucial role in wiring the brain, but a major challenge is to understand how they promote and maintain specific connections between different neurons. One way to generate diversity in cell recognition molecules is through alternative splicing, which can produce distinct proteins from the same gene. These alternative protein isoforms can then provide distinct functions in different cell types (Aoto et al., 2013; Lah et al., 2014; Fuccillo et al., 2015). Here, we report that Drosophila Dscam2, a cell recognition molecule classically associated with axonal tiling and self-recognition during development, functions in regulating intimate facets of synaptic function. Only one of its two extracellular isoforms can carry out this function. Our results reveal that Dscam2B suppresses synaptic strength through a PI3K-dependent endosomal pathway that inhibits the number of synaptic vesicles at active zones.

Because the two isoforms of Dscam2 differ at a single immunoglobulin domain that confers homophilic binding specificity, we originally hypothesized that these proteins would have similar functions in different cells. If this is the case, then both of the single isoform lines, which express one isoform of Dscam2 in all Dscam2-positive cells, should generate similar phenotypes. This is what we observed for mEJP effects and NMJ morphology, consistent with our previous analysis in the visual system (Lah et al., 2014). For example, lamina neurons L1 and L2 express distinct Dscam2 isoforms, and their axon terminals contact each other. When these cells express the same isoform of Dscam2, axon terminal size decreases significantly, presumably due to inappropriate homophilic repulsion (Lah et al., 2014). Analogous relationships between neurons in the VNC likely exist. For example, sensory neurons that express Dscam2B could provide input into interneurons expressing Dscam2A, which in turn connect with motor neurons that express Dscam2B. If pre- and postsynaptic cells repelled each other, this simple circuit would be severely disrupted in the single isoform lines. However, other mechanisms are also possible, and demonstrating that inappropriate repulsion drives these gain-of-function phenotypes will require high-resolution circuit analysis within the VNC.

What then, is the role of Dscam2 in the neurons of these simple circuits? It likely functions as a repulsive cue to maintain boundaries between neighboring neurons expressing the same Dscam2 isoform as the larvae grow and may also act as a self-avoidance cue to help dendritic and axonal arbors maximize their coverage as they increase in size. Adhesive roles for the two isoforms are also possible, and each of these functions has been previously observed in the visual system (Millard et al., 2007; Millard et al., 2010; Lah et al., 2014; Tadros et al., 2016).

In contrast to the gain-of-function effects described above, the loss-of-function synaptic effects in Dscam2^null and Dscam2A animals were unexpected and delineate a functional dichotomy between the two isoforms. How Dscam2B, but not Dscam2A, can suppress synaptic strength in motor neurons is currently a mystery. One possibility is that alternative Dscam2 cytoplasmic domains, with unique signaling properties, are required for regulating synaptic strength. Seven different cytoplasmic variants of Dscam2, produced through a combination of alternative splicing and stop codon suppression (Jungreis et al., 2011), have been identified at the mRNA level using techniques such as RNA-sequencing (Graveley et al., 2011), but whether they are expressed at the protein level is not known. We deem it unlikely that a subset of these variants is specifically expressed in motor neurons to regulate synaptic strength, as their splicing would need to be intimately coordinated with that of isoform B (presumably through the splicing factor Muscleblind). We previously analyzed this using semi-quantitative RT-PCR and found no evidence for changes in expression of the Dscam2 cytoplasmic isoforms in muscleblind mutant animals (Li and Millard, 2019). In addition, most of the flanking sequences that likely regulate cell-specific splicing of exons 10A and 10B were removed in the single isoform lines (Lah et al., 2014), which would perturb coordination between Muscleblind and the selection of the specific cytoplasmic exons. Thus, although a subset of these cytoplasmic tails may transmit the signal that regulates synaptic strength, it is unlikely that a specific variant is linked to Dscam2B. A second, more likely possibility is that the functional differences between isoforms are dependent on the variable immunoglobulin domain. Since Dscam2 is not expressed in muscle, Dscam2 homophilic interactions between the motor neuron terminal and muscle cannot regulate synaptic strength. However, an alternative ligand that binds specifically to Dscam2B could either induce or suppress Dscam2 signaling within motor neurons. Secreted or membrane-bound cues may interact specifically with Dscam2B, although no alternative Dscam2 ligands were found in an in vitro screen with other cell recognition molecules (Özkan et al., 2013). A coreceptor that binds to Dscam2B, but not Dscam2A, in cis is another realistic possibility. In this scenario, the ligand for the coreceptor could be regulating Dscam2 signaling. Future studies to identify Dscam2-interacting proteins are required to resolve this issue. Finally, although our phenotypes strongly suggest that Dscam2 is functioning in the axon terminal, we have been unable to detect it in this compartment using antibodies. It is therefore possible that Dscam2 suppresses synaptic strength from the dendritic compartment of the motor neuron, although we deem this unlikely.

The mechanism through which Dscam2 regulates synaptic strength is remarkably specific and requires one of its two extracellular isoforms. Removal of Dscam2 results in increased evoked, but not spontaneous, neurotransmitter release. Similar effects have been observed following modulation of endosomal regulators. For example, Rab5 levels are rate-limiting for synaptic strength, and loss of the GTPase-activating protein Skywalker (Sky) increases synaptic strength at the Drosophila larval NMJ (Wucherpfennig et al., 2003; Uytterhoeven et al., 2011). Both Rab5 and sky mutants have overlapping phenotypes with Dscam2 mutants. For example, large (70–80 nm) vesicles accumulate in all three of these mutants. However, in sky null NMJs, this is only observed in nerve terminals that have been stimulated (Uytterhoeven et al., 2011). A change in synaptic strength is another common theme. Both sky and Dscam2 mutants exhibit an increase in evoked, but not spontaneous, release. In motor neurons expressing a dominant negative Rab5 (Rab5S43N), spontaneous release is also normal and there is a decrease in evoked release. In contrast, overexpression of Rab5 leads to increased evoked release (Wucherpfennig et al., 2003), demonstrating that it is rate-limiting and likely a major target of regulation. Given that Rab5 binds to PI(3)P (Gillooly et al., 2003), a product of PI3K, Dscam2 and CenG1A have the ability to modulate Rab5 on membranes. Similarities in SV pools are also observed in these mutants. We observed more synaptic vesicles at active zones in the Dscam2 mutants, and there is an increase in the releasable pool of synaptic vesicles in sky mutants (Uytterhoeven et al., 2011). This was not measured directly for Rab5 overexpression, but docked vesicles and spontaneous release were unchanged in these animals (Wucherpfennig et al., 2003), leading the authors to postulate increased pools of releasable vesicles at these NMJs.

The model of how sky mutants increase synaptic strength is by increasing traffic through the endosomal system, which results in younger synaptic proteins and consequently more fusogenic vesicles (Fernandes et al., 2014). Our data demonstrate that regulation of PI3K plays a critical role in this process, which is consistent with the dependence of endosomal sorting on specific phosphoinositides (Jähne et al., 2015). Dscam2 is necessary for promoting PI3K activity, and it carries out this function through CenG1A. Consistent with this, Dscam2 mutants and CenG1A mutants exhibit increases in PIP₂, a major substrate of PI3K. The accumulation of large vesicles in the Dscam2 mutants combined with the decrease in PI(3)P in NMJs at rest suggest that fusion between endosome intermediates and the early endosome is impaired. How this leads to increased traffic through the endosomal pathway is unclear, as we would expect it to have the opposite effect. One possibility is that an impairment in endosomal fusion reduces the number of SV proteins traveling through the endosome compartment, and the neuron compensates by increasing the number of de novo synthesized SV proteins that traffic to the terminal. These young synaptic vesicle proteins then become incorporated into new vesicles and trafficked to the releasable pool.

Another difference between Dscam2 and sky mutants (Fernandes et al., 2014; Uytterhoeven et al., 2011) is that the Dscam2 phenotypes occur in unstimulated neurons, where demands on synaptic vesicle recycling are less than in KCl- or high frequency–stimulated preparations. It is possible that Dscam2 regulates the endosomal pathway in a PI3K-dependent manner under moderate levels of activity. Following large increases in synaptic vesicle release, bulk endocytic pathways contribute significantly to membrane recovery and collaborate with endosome recycling pathways (Cousin, 2009). Under these conditions, Dscam2 would either be inhibited or irrelevant due to the activation of alternative mechanisms that rapidly produce synaptic vesicles. This could explain the recovery of PI(3)P labeling in activated Dscam2 mutant terminals (Fig. 5 E).

In conclusion, we have demonstrated that Dscam2 regulates neurophysiology through two distinct mechanisms. The first comprises regulated isoform expression in different cell types that presumably involves the spatial organization of neurites through isoform-specific homophilic interactions, as has been described previously (Millard et al., 2007; Lah et al., 2014; Tadros et al., 2016). The second mechanism, which decreases synaptic strength in motor neurons, is isoform specific. Dscam2B decreases neurotransmitter release in motor neuron terminals by inhibiting the number of synaptic vesicles that get trafficked to active zones through a PI3K-dependent endosomal pathway (Fig. 6, P and Q). Our study demonstrates how cell-specific expression of protein isoforms can provide unique functions to different cells and thus explains how an incredibly complex organ like the brain can operate effectively using a small toolkit of genes.

Flies were reared at room temperature (22–25°C) and humidity on standard cornmeal medium. The following lines were acquired from Bloomington Drosophila Stock Centre: CenG1A^EY01217 (15502), Lat⁶ (5571), Leo^12BL (9572), Psn¹⁴³ (8297), Df(2L)BSC6 (6338), Df(2L)BSC105 (8671), Df(2L)BSC694 (26546), Df(2L)Exel7080 (7853), Df(3L)BSC669 (26521), Df(3L)BSC671 (26523), Df(3L)Exel6092 (7571), elav-GAL4 (8765), nSyb-GAL4, OK6-GAL4 (64199), OK371-GAL4 (26160), UAS-Dscam2-RNAi (51839), UAS-mCD8::GFP (5137), UAS-PLCδPH-mCherry (51658), and UAS-2xFYVE-myc-GFP (42712). The following fly lines were generated in the Millard laboratory and have been previously described: Dscam2^null-1 and Dscam2^null-3 (Millard et al., 2007), Dscam2A-GAL4, and Dscam2B-GAL4 (Lah et al., 2014). All animals selected for experiments were on a w¹¹¹⁸ background, and larvae of either sex were selected only if in wandering third-instar stage. Experiments used larvae and embryos raised at 25°C unless otherwise specified. Embryos of either sex were staged to within a 2-h window.

Dissections of third-instar larvae were performed in hemolymph-like 3 (HL3) saline (Stewart et al., 1994) following the magnetic body-wall muscle procedure described in Ramachandran and Budnik (2010). In brief, an incision was made along the dorsal length of third-instar larvae, and the body wall was pinned open to expose internal organs and musculature. Organs were carefully removed, taking care not to damage the underlying muscles, and if required, peripheral nerves were cut to remove the brain. Bath [Ca²⁺] was dependent on the type of assay performed.

Stage 16 embryos were manually dechorionated and devitellinated by gently rolling on doubled-sided tape. They were then transferred to a drop of HL3 on a glass slide and filleted using sharp glass capillaries created using a model P-97 puller (Sutter Instruments) by taking advantage of the adhesion between stage 16 embryo cuticle and glass. Embryos were fixed, stained, and imaged on the same glass slide.

Dissected third-instar larval fillets were fixed in PBS containing 4% PFA (wt/vol) for 20 min for all antibody combinations used with the exception of GluRIIA, which required fixation in 100% methanol for 5 min. Following fixation, immunohistochemistry was performed as previously described (Brent et al., 2009). Fixed filleted larvae were first washed in PBS, then incubated in blocking buffer for 1 h at room temperature. Blocking buffer consisted of PBS containing 5% goat serum and 0.5% Triton-X. Following this, the filleted larvae were incubated in blocking buffer containing primary antibodies overnight at 4°C. Antibody dilutions were used as follows: Cy3-conjugated anti-HRP (1:500; 123–165-021; Jackson ImmunoResearch Laboratories); mouse anti-brp (1:100, nc82; Developmental Studies Hybridoma Bank [DSHB]); mouse anti-dlg (1:500, 4F3; DSHB), mouse anti-GFP (1:500, 4745–1051; Bio-Rad), mouse anti-Synapsin (anti-Syn; 1:100, 3C11; DSHB), mouse anti-GluRIIA (1:100, MH2B; DSHB), mouse anti-Syt1 (1:100, 3H2 2D7; DSHB), and Alexa Fluor 568 Phalloidin (1:500, A12381; Thermo Fisher Scientific). After incubating with primary antibodies overnight, filleted larvae were washed in PBS and then incubated in blocking buffer containing secondary antibodies overnight at 4°C. Secondary antibodies used included Alexa Fluor 488–conjugated goat anti-mouse IgG (1:800, A-11001; Thermo Fisher Scientific) and Alexa Fluor 647–conjugated goat anti-mouse IgG (1:800, A28181; Thermo Fisher Scientific). Antibody incubations were performed overnight at 4°C. Lastly, after incubating with secondary antibodies, filleted larvae were washed in PBS and mounted onto glass slides in glycerol.

For FM4-64 experiments, methods according to Verstreken et al. (2008) were closely followed. In brief, larvae were dissected in Ca²⁺-free HL3 and then incubated with HL3 containing 90 mM KCl and FM4-64FX (4 µM, F34653; Thermo Fisher Scientific) for 1 min. Following this, larvae were washed rigorously by alternating positive and negative pressure with fresh Ca²⁺-free HL3 in bath over 5 min and at least three solution changes. Preparations were then fixed, mounted on glass slides in glycerol, and imaged within 24 h.

Microscopy was performed at the School of Biomedical Sciences Imaging Facility and Queensland Brain Institute Advanced Microscopy Facility. Imaging of larval preparations was predominantly performed on an Olympus FV1000 upright scanning confocal microscope with NA 1.35 60× or NA 1.4 100× oil immersion objectives via photomultiplier tubes at room temperature. Fluoview software (Olympus America Inc.) was used to acquire images taken on the FV1000 system. Imaging was also performed on an Axio Observer Z1 (Zeiss) equipped with a CSU-W1 spinning-disk head (Yokogawa Corporation of America), ORCA-Flash4.0 v2 sCMOS camera (Hamamatsu Photonics), and NA 1.4 100× oil immersion objective. Huygens software (Scientific Volume Imaging) was used to acquire images taken on the spinning-disk system. Optical sectioning and laser settings were kept constant across all image data acquisition sessions for like experiments and depended on the type of experiment being performed. Images were taken from NMJs formed upon muscles 6 and 7 by MN6/7-Ib and MNSNb/d-Is in abdominal segments 3 and 4 (A3 and A4). No more than two NMJs were imaged per larva.

Analysis of all images was done using ImageJ (National Institutes of Health). To measure the number of boutons per NMJ, anti-Syn immunoreactivity was used to label individual boutons, which were manually counted. To measure the maximal cross-sectional area of muscles, a perimeter was traced around the perpendicular muscle cross section from maximal z-stack projections of phalloidin. Cross-sectional muscle area was then calculated from the traced perimeters. To measure bouton and subsynaptic reticulum (SSR) diameters, a line was drawn across the longest axis perpendicular to the main branch of HRP and anti-dlg immunoreactivity from maximal z-stack projections. All boutons on at least one Ib or Is branch per NMJ were measured and averaged to generate representative values. To measure GluRIIA fluorescence intensity, HRP immunostaining was used to generate “synaptic” regions of interest (ROIs) from which raw anti-GluRIIA pixel intensity was measured. Three random regions of similar cross-sectional area that did not intersect with synaptic ROIs were then determined as “nonsynaptic” ROIs and were averaged to determine the background fluorescence. The synaptic GluRIIA raw pixel values were then divided by nonsynaptic pixel values to generate normalized GluRIIA fluorescence values. These results were confirmed by manually measuring the size and average pixel values of at least 20 individual GluRIIA puncta per NMJ for every animal. A similar approach was taken for measuring relative Syt1 immunofluorescence intensity with the exception of having to define and measure puncta, as Syt1 immunoreactivity filled the entire bouton.

Measurement of PLCδPH-mCherry and 2xFYVE-GFP average pixel intensity per bouton was performed using ROIs that were made by drawing perimeters around HRP-positive boutons. For these genotypes, immunostaining was performed to amplify signals of the tags. The same methodology was used to quantify FM4-64FX immunofluorescence intensity, but ROIs were generated based on the FM4-64FX signal directly, instead of from HRP.

Dissected larvae were fixed in 4% PFA and 2.5% glutaraldehyde solution overnight. After a PBS rinse, the samples were stained with 1% osmium tetroxide and 2% uranyl acetate using a Pelco Biowave. They were then dehydrated in ethanol series (50%–100%) followed by infiltration with increasing concentrations of epon resin (25%–100%), in which they were left immersed for 24 h at room temperature. Samples were subsequently processed in the resin with the Biowave high vacuum function before being embedded in fresh resin and polymerized in a 60°C oven for 48 h. Formvar-coated, one-slot grids were used to collect thin sections (50 nm) obtained via a Leica Ultracut UC6 Ultramicrotome by taking transverse sections of muscles at intervals of 2.5 µm over 35–150 µm. Sections were then post-stained using Reynold’s lead citrate solution and 4% uranyl acetate. Boutons were photographed with a JEOL 1011 electron microscope at 80 kV. ImageJ was used to quantify substructures from electron micrographs in a blinded manner.

We quantified several parameters including bouton size; SSR size; and number of mitochondria, dense core vesicles, MVBs, synaptic vesicles, large endosomal vesicles, tubular/irregular structures, and multimembranous structures. Mitochondria were defined by the presence of cristae and double outer membranes. DCVs were defined by the presence of circular-shaped electron dense material enclosed by a single membrane. MVBs were defined by the presence of multiple (more than one) clear membrane-bound objects within a single clear membrane-bound object. Multimembranous structures were defined by multiple (more than one) lamellar-appearing membranes enclosing a clear core.

To define and measure the size of synaptic vesicles, large endosomal vesicles, and tubular/irregular endosomes in micrographs, we drew perimeters around all clear, membrane-bound structures within boutons. Of these structures, synaptic vesicles were defined as having circularity values >0.75, aspect ratios <2, and diameters <70 nm. Large vesicles were defined as having circularity values >0.75, aspect ratios <2, and diameters >70 nm. Tubular/irregular structures were defined as structures with circularity values <0.75 and aspect ratios >2. Multi-membranous structures were identified by the presence of at least one membrane-bound structure within another membrane-bound structure.

Larvae were dissected in HL3 containing variable [Ca²⁺] depending on the type of assay being performed. All peripheral nerve bundles were severed, and the central nervous system was removed. Muscle 6 in A3 or A4 was chosen for impalement using thick-walled, high-resistance (80–100 MΩ), sharp borosilicate electrodes filled with a 2:1 ratio of 3 M KCH₃COO and 3 M KCl. Impaled muscles were left unstimulated until the resting membrane potential reached stability, as defined by fluctuations of <10% within 5 min. Floating ends of the severed nerve bundles were recruited using wide-tipped (10–12 µm) thin-walled borosilicate electrodes containing HL3 (Kurdyak et al., 1994). Synaptic responses were elicited by stimulating recruited nerves with 0.3-ms-long square wave pulses at a frequency of 0.5–1 Hz using a SD9 square pulse stimulator (Grass Instruments). Spontaneous mEJPs were recorded for no less than 3 min, and EJPs were recorded for 1–2 min. No more than two muscles were recorded per larva. All intracellular responses were digitized using the LabChart 7.0 software (AD Instruments) at a sampling rate of 40 kHz and were amplified using an Axoclamp 2B amplifier (Molecular Devices), direct current (DC) high-pass bessel, and 3-kHz low-pass bessel. All EJPs and mEJPs were measured manually in LabChart7.

EJP amplitudes were corrected for nonlinear summation as described by Martin (1976) and elaborated upon by McLachlan and Martin (1981). Quantal content was calculated by dividing average EJP amplitude corrected for nonlinear summation (EJP′) by average mEJP amplitude for control and Dscam2^null recordings. For Dscam2 single isoform recordings, average EJP′ amplitude was instead divided by median mEJP amplitude since most individual frequency-amplitude histograms revealed skewed distributions. Amplitude-variation plots were generated by fitting standard quadratic equations to plotted values of average, SEM, and n for EJP′ amplitude and trial-to-trial variation of EJP′ per [Ca²⁺]_o explored.

Wortmannin (19545–26-7; Sigma-Aldrich) and LY294002 (S1105; Selleck Chemicals) stocks were dissolved in DMSO and maintained at −20°C. For experiments, drugs in DMSO were dissolved into HL3 solution up to 24 h before use and stored at 4°C. Larvae were dissected in drug-free HL3 and then incubated with Wortmannin (1 h) or LY294002 (30 min) before recordings. Total DMSO in HL3 did not exceed 0.001% (vol/vol).

GraphPad Prism 7 was used to perform all analyses and generate graphs. The D’Agostino–Pearson omnibus test for normality was used to test whether or not data assumed a Gaussian distribution unless the number of values collected was too small, in which case the Shapiro–Wilk test for normality was used instead. To determine statistical significance between groups that all assumed Gaussian distributions, parametric tests were employed (unpaired Student’s t test, one-way ANOVA with Tukey’s multiple comparisons, and two-way ANOVA with Sidak’s multiple comparisons). To determine statistical significance between groups in which at least one population did not assume a Gaussian distribution, nonparametric tests were employed instead (Mann–Whitney rank-sum test, Kruskal–Wallis test with Dunn’s multiple comparison, and two-way ANOVA with Sidak’s multiple comparisons). Significance was determined at P < 0.05. All error bars represent average ± SEM. Numbers (n) indicated in graphs or figure legends represent number of NMJs analyzed. No more than two NMJs were taken per larva.

Fig. S1 shows expression of Dscam2B in a larval abdominal hemi-segment. It also shows expression of Dscam2A and Dscam2 throughout embryonic development. Fig. S2 shows extended electrophysiological analyses of those presented in Fig. 2. Fig. S3 shows analysis of dlg and Syt1 fluorescence intensity and the results from a genetic interaction screen against Dscam2^null heterozygotes using EJC amplitude as a readout. Fig. S4 shows analysis of electron micrographs taken from Dscam2 single A and B isoform larvae in extension of the results shown in Fig. 6. Table S1 shows known properties of all larval abdominal segment motor neurons including expression of Dscam2B.

We thank Yong Qi Lin and Shanker Karunanithi for their technical assistance with neuromuscular synaptic recordings; Erica Lovas, Kathryn Green, and Matthias Floetenmeyer for providing electron microscopy technical assistance; and Frederic Meunier (Universisty of Queensland, Brisbane, Australia) for generously providing the FM styryl dyes.

G. Lorenzo Odierna was supported by an Australian Postgraduate Award, and project funding was provided by a National Health and Medical Research Council grant to S. Sean Millard (APP1105792).

Author contributions: G. Lorenzo Odierna performed all experiments and analyzed all data with the exception of electron microscopy data collection by L.E. Harris and CenG1A PLCδPH-mCherry data collection and analysis by S.K. Kerwin. G. Lorenzo Odierna and S. Sean Millard wrote the manuscript, P.G. Noakes provided reagents and expertise for the EM experiments. G.J. Shin provided critical feedback on the manuscript, S. Sean Millard secured funding and provided reagents, and P.G. Noakes, N.A. Lavidis, and S. Sean Millard provided expertise.

The authors declare no competing financial interests.