RNA-binding FMRP and Staufen sequentially regulate the Coracle scaffold to control synaptic glutamate receptor and bouton development

Fragile X syndrome (FXS) is a common heritable cause of intellectual and autism spectrum disorders (Crawford et al., 2001). FXS patients typically exhibit a fragile X mental retardation 1 (FMR1) 5′ untranslated region (UTR) CGG repeat expansion (typically ≥200), which causes epigenetic transcriptional silencing via FMR1 promoter hypermethylation (Garber et al., 2008; Hansen et al., 1992; Verkerk et al., 1991). The fragile X mental retardation protein (FMRP; FMR1) product is a very broadly expressed (e.g. neurons, muscles) mRNA-binding translation regulator (Drozd et al., 2018), which binds target transcripts via K homology (KH) domains and arginine-glycine rich (RGG) box (Blackwell and Ceman, 2011; Kenny and Ceman, 2016; Myrick et al., 2015; Ramos et al., 2003). FMRP regulates protein translation to modulate synaptic architecture (bouton/spine number) and glutamate receptor (GluR) levels (Comery et al., 1997; Connor et al., 2011). In the Drosophila FXS disease model, dfmr1 mutants likewise exhibit increased synaptic bouton formation and Glutamate receptor IIA (GluRIIA) levels at the neuromuscular junction (NMJ) model glutamatergic synapse (Pan and Broadie, 2007; Zhang et al., 2001). The molecular mechanism of FMRP-mediated synaptic regulation remains elusive; however, FMRP has been increasingly linked to other mRNA-binding proteins (Kenny et al., 2020; Price et al., 2006; Zhang et al., 2017). A key hypothesized partner is Staufen, a double-strand RNA-binding protein (dsRBP) repeatedly associated with FMRP function via both biochemical and genetic interaction studies (Barbee et al., 2006; Chu et al., 2019; Yu et al., 2012).

Staufen plays crucial roles in regulating mRNA localization, stability, translation and ribonucleoprotein (RNP) assembly (Dugré-Brisson et al., 2005; Micklem et al., 2000; Park and Maquat, 2013). In Drosophila, Staufen colocalizes with FMRP in neural RNP granules that mediate mRNA translational repression and mRNA decay, with genetic interaction regulating long-term memory consolidation (Barbee et al., 2006; Bolduc et al., 2008). Like FMRP, Staufen controls both synaptic bouton formation and GluRIIA levels at the Drosophila NMJ (Gardiol and St Johnston, 2014). In this mechanism, Staufen works by regulating local translation of the 4.1 ezrin-radixin-moesin (FERM) scaffold Coracle in the muscle postsynaptic domain (Gardiol and St Johnston, 2014). Consistently, mammalian Staufen also binds Coracle homolog 4.1 mRNA and is predicted to regulate its local translation (Furic et al., 2008). Coracle is suggested to link F-actin to GluRIIA C-termini to scaffold receptors within the postsynaptic membrane (Chen et al., 2005; McClatchey, 2012). Importantly, intercellular interaction between postsynaptic GluRIIA and the presynaptic bone morphogenic protein (BMP) receptor Wishful thinking (Wit) generates phosphorylated Mothers against decapentaplegic (pMad) retrograde trans-synaptic signaling to regulate presynaptic bouton formation (Chou et al., 2020; Sulkowski et al., 2014, 2016). Based on these studies, we hypothesized that FMRP works with Staufen to regulate postsynaptic Coracle scaffolding, which in turn acts to control postsynaptic GluRIIA accumulation and thereby GluRIIA-dependent presynaptic bouton development.

To interrogate this layered hypothesis, we first tested NMJ bouton number and GluRIIA levels in dfmr1 and staufen single mutants and RNA interference (RNAi) lines, to find that both FMRP and Staufen negatively regulate synaptic bouton formation and GluRIIA accumulation. We next made trans-heterozygous double mutants (dfmr1/+; staufen/+) to find that FMRP and Staufen operate in the same pathway to control synaptic development. Subsequently, we used RNA immunoprecipitation (RIP) to show that FMRP binds staufen mRNA to regulate transcript abundance in the postsynaptic muscle, and that Staufen in turn binds coracle mRNA. Consistently, Coracle expression in the NMJ postsynaptic domain was elevated in both dfmr1 and staufen mutants, as well as in trans-heterozygous double mutants. We found that postsynaptic Coracle overexpression (OE) and loss of function similarly increase bouton number and GluRIIA levels. Consistently, we employed NMJ electrophysiology recordings to show that dfmr1, staufen and coracle mutants all display increased synaptic strength. Moreover, postsynaptic knockdown of dfmr1, staufen and coracle all caused elevated presynaptic pMad levels, consistent with activation of GluRIIA–Wit retrograde trans-synaptic signaling to drive presynaptic bouton formation. Taken together, these findings suggest that FMRP and Staufen work sequentially to inhibit the Coracle scaffold controlling GluRIIA levels in postsynaptic domain, and that postsynaptic GluRIIA levels in turn signal presynaptic bouton development. This work provides insights into the molecular pathway by which FMRP regulates synapse formation, identifying potential new FXS treatment targets.

At the Drosophila NMJ, we have previously reported that viable dfmr1 nulls (dfmr1^50M) exhibit elevated synaptic bouton formation and GluRIIA levels (Pan and Broadie, 2007; Zhang et al., 2001). By contrast, staufen nulls are embryonic lethal owing to essential mRNA localization and translation roles (St Johnston et al., 1991), and a viable staufen mutant over a genomic deficiency [stau^HL/Df(2R)Pcl7B] reportedly develops fewer NMJ boutons and lower GluRIIA levels (Gardiol and St Johnston, 2014). The stau^HL mutant contains a T-A point mutation in dsRNA-binding domain 5 (Fig. S1A) that blocks local translation (Gardiol and St Johnston, 2014). As a first step, we re-tested dfmr1^50M and stau^HL mutants compared with matched genetic background controls (w¹¹¹⁸) for bouton number and GluRIIA level. We then tested trans-heterozygotes (dfmr1^50M/+; stau^HL/+) for a predicted interaction within the same pathway. We assayed wandering third-instar NMJs double labeled with anti-horseradish peroxidase (HRP) (Jan and Jan, 1982; Pan and Broadie, 2007), which recognizes neural presynaptic membrane, and anti-Discs large (DLG; Dlg1) (Kamimura et al., 2019; Menon et al., 2013), which recognizes muscle subsynaptic reticulum (SSR). Both total NMJ boutons and developing satellite boutons were counted in muscle 4 terminals in abdominal segment A3. The same genotypes were double labeled with anti-HRP and anti-GluRIIA (Pan and Broadie, 2007) at the same NMJ. GluRIIA labeling intensity was quantified at HRP-thresholded boutons.

Compared with the genetic background control (w¹¹¹⁸), dfmr1^50M mutants showed supernumerary synaptic bouton formation (Fig. 1A, top). The quantified total bouton number was significantly elevated (mean±s.e.m.: control 19.10±1.77, dfmr1 31.42±1.67; P<0.0001; Fig. 1C), with a parallel increase in satellite boutons (number/NMJ: control 0.86±0.27, dfmr1 2.65±0.47; P=0.003; Fig. 1D). Similarly, stau^HL mutants also developed consistently more NMJ boutons compared with w¹¹¹⁸ genetic controls (Fig. 1A, bottom). Quantification showed that the total NMJ bouton number was significantly increased in staufen mutants compared with controls (control 20.85±0.78, stau^HL 29.25±2.15; P=0.0003; Fig. 1C), with satellite boutons also elevated (control 0.89±0.21, stau^HL 4.64±0.72; P<0.0001; Fig. 1D). Assaying synaptic GluRIIA levels, dfmr1 mutants exhibited a clear increase throughout the NMJ terminal (Fig. 1B, top). GluRIIA fluorescence levels normalized to control were significantly higher in dfmr1 mutants (control 1.00±0.08, dfmr1^50M 1.616±0.11; P=0.0002; Fig. 1E). Likewise, GluRIIA levels were also increased in the staufen mutants compared with matched controls (Fig. 1B). Compared with levels in genetic controls, the normalized GluRIIA fluorescence levels in the staufen mutants were also significantly elevated (control 1.00±0.04, stau^HL 1.24±0.06; P=0.0014; Fig. 1B,E). These results indicate that FMRP and Staufen similarly regulate synaptic development.

To further test staufen phenotypes, we next used staufen RNAi as an independent knockdown method (Table S1, Fig. S1B-D). Studies with quantitative PCR (qPCR) showed ∼90% staufen mRNA loss with global UH1-Gal4 driving UAS-staufen RNAi [Vienna Drosophila Resource Center (VDRC) 106645; Fig. S1B]. Consistent with the above staufen mutants, staufen RNAi elevated both presynaptic bouton formation (Fig. S2A, top) and postsynaptic GluRIIA levels (Fig. S2A, bottom). Quantification of the knockdown showed that UH1>stau RNAi (VDRC 106645) increased all measurements, including total bouton number (UH1/+ 23.33±1.09, UH1>stau RNAi 28.87±1.30; P=0.004; Fig. S2B), satellite boutons (UH1/+ 1.33±0.43, UH1>stau RNAi 2.93±0.50; P=0.0273; Fig. S2C) and GluRIIA levels (UH1/+ 1.00±0.06, UH1>stau RNAi 1.28±0.07; P=0.0038; Fig. S2D). We repeated these analyses with an independent staufen RNAi line [Bloomington Drosophila Stock Center (BDSC) 31247]. Consistent with the above results, this second RNAi similarly caused a significant increase in synaptic bouton number (UH1/+ 24.33±0.79, UH1>stau RNAi 32.36±1.815, P=0.0004) and GluRIIA levels (UH1/+ 1.00±0.07, UH1>stau RNAi 1.66±0.16; P=0.0041). Thus, stau^HL and two independent staufen RNAi lines (VDRC 106645 and BDSC 31247) confirmed the same NMJ development phenotypes. We therefore conclude that Staufen loss increases synaptic bouton formation and GluRIIA levels, consistent with FMRP requirements.

To test the hypothesis that FMRP and Staufen co-regulate NMJ development in a common pathway, we next made dfmr1 and staufen double mutants. Homozygous double mutants were early larval lethal, but dfmr1^50M/+; stau^HL/+ trans-heterozygotes were viable and could be tested. Similar to dfmr1^50M and stau^HL single mutants, we found a clear elevation of total boutons in the trans-heterozygotes. Quantification showed bouton increases in dfmr1 (control 17.91±0.89, dfmr1/+ 25.88±1.81) and staufen (control 19.00±0.70, stau/+ 26.91±1.56) heterozygotes, and the trans-heterozygotes (control 17.56±1.17, dfmr1/+; stau/+ 28.70±1.55; P<0.0001; Fig. 2A, bottom). We next tested GluRIIA to find similar levels in dfmr1/+ and stau/+ heterozygotes compared with controls, but elevated levels in dfmr1/+; stau/+ trans-heterozygotes (Fig. 2A). Quantified GluRIIA levels were not changed in either dfmr1/+ or stau/+ single heterozygotes compared with control (Fig. 2A,B), but were significantly increased in trans-heterozygotes (normalized control 1.00±0.08, dfmr1/+; stau/+ 1.28±0.12; P=0.047; Fig. 2A,B). These findings showed that reducing FMRP and Staufen in parallel elevated synaptic GluRIIA levels, suggesting that the two RNA-binding proteins (RBPs) work in a common mechanism. Overall, we conclude that FMRP and Staufen negatively regulate synaptic development in the same direction, and to a similar degree, by functioning in the same pathway.

Cell-targeted RNAi studies have established that FMRP inhibits GluRIIA levels only postsynaptically (Pan and Broadie, 2007), but suppresses bouton development in both postsynaptic and presynaptic cells (Friedman et al., 2013; Zhang et al., 2001). Likewise, Staufen subcellularly localizes in the postsynaptic domain, to function postsynaptically in muscle, controlling mRNA localization and local translation (Gardiol and St Johnston, 2014). These previous studies, as well as the above non-complementation genetic interaction tests, suggest that FMRP interacts with Staufen in the postsynaptic compartment to regulate GluRIIA levels and presynaptic bouton formation. Previous antibody labeling shows Staufen in the postsynaptic muscle region immediately surrounding NMJ termini, with Staufen not detectable in presynaptic boutons (Gardiol and St Johnston, 2014). We therefore hypothesized Staufen that has a specific muscle postsynaptic function. To test this hypothesis, we used muscle-specific 24B-Gal4 (Kim et al., 2021) and neuron-specific elav-Gal4 (Kan et al., 2021) to drive UAS-staufen RNAi (VDRC 106645; Landskron et al., 2018). We also used postsynaptic 24B-Gal4 to drive wild-type UAS-staufen in both homozygous stau^HL and dfmr1^50M mutants. As above, synaptic bouton development was tested with presynaptic anti-HRP and postsynaptic anti-DLG double labeling, and GluRIIA levels with anti-HRP and anti-GluRIIA double labeling.

Compared with transgenic controls (24B-Gal4/+), muscle-targeted staufen RNAi (24B>stau RNAi) resulted in more synaptic boutons (Fig. 3A, top). With quantification, total boutons were significantly increased in 24B>stau RNAi (24B/+ 21.94±1.23, RNAi 25.43±0.86; P=0.02; Fig. 3C), with more developing satellite boutons (24B/+ 2.09±0.31, RNAi 3.46±0.34; P=0.0047; Fig. 3D). By contrast, neural staufen knockdown (elav>stau RNAi) had no effect, with no change in bouton number (Fig. 3A, bottom). Quantification showed no significant difference in synaptic bouton formation between transgenic control (elav-Gal4/+) and neural staufen RNAi (P=0.91; Fig. 3C), indicating that presynaptic Staufen has no detectable role. Assaying GluRIIA levels, muscle-targeted knockdown (24B>stau RNAi) resulted in clearly higher fluorescence than in the transgenic controls (24B-Gal4/+), with elevated GluRIIA levels at synaptic boutons (Fig. 3B, top). Quantified GluRIIA measurements showed that muscle-targeted staufen RNAi strongly upregulated GluIIA levels compared with normalized controls (24B/+ 1.00±0.04, RNAi 1.33±0.05; P<0.0001; Fig. 3E). By contrast, neural staufen knockdown (elav>stau RNAi) resulted in no detectable change in GluRIIA synaptic fluorescence (Fig. 3B, bottom), with quantified results showing no role in determining GluRIIA levels (P=0.46; Fig. 3E). These findings suggest that Staufen acts in the postsynaptic muscle to regulate NMJ development.

To further test this conclusion, we next performed complementary staufen rescue experiments in the postsynaptic muscle. Compared with transgenic control (24B-Gal4/+), muscle UAS-staufen expression in staufen^HL (stau^HL) homozygous mutant (stau^HL; 24B>stau) showed strongly rescued synaptic development (Fig. S3). With targeted postsynaptic UAS-staufen, the presynaptic bouton number was restored to the control level (24B/+ 23.80±0.86, stau^HL; 24B>stau 26.89±1.24), with no significant difference remaining (P=0.113). Assaying GluRIIA levels revealed an even stronger effect, with muscle-targeted rescue (stau^HL; 24B>stau) resulting in clearly reduced fluorescence compared with that of transgenic controls (24B-Gal4/+), showing lower GluRIIA levels at synaptic boutons (Fig. S3A). Quantification revealed a >40% reduction in GluRIIA receptors normalized to transgenic controls (24B/+ 1.00±0.067, stau^HL; 24B>stau 0.57±0.04), which is a significant decrease (P<0.0001; Fig. S3B). Consistent with this postsynaptic requirement, the same muscle staufen OE in the null dfmr1 homozygous mutant (dfmr1; 24B>stau) suppressed GluRIIA expression to levels comparable with those of the transgenic control (normalized 24B/+ 1.00±0.07, dfmr1; 24B>stau 1.21±0.19; P=0.21). Taken together, these findings suggest that FMRP interacts with Staufen in the muscle postsynaptic domain to regulate GluRIIA levels.

FMRP and Staufen both bind mRNA directly to regulate local protein translation (Bonnet-Magnaval, 2016; Laver et al., 2013; Liu et al., 2018; Tsang et al., 2019), and therefore could operate either in parallel or sequentially in protein-mRNA interactions limiting synaptic development. Importantly, FMRP has been predicted to bind staufen mRNA (D'Annessa et al., 2019), and we therefore hypothesized that FMRP regulates Staufen translation in a sequential mechanism. To test the predicted FMRP and staufen mRNA interaction, we used UH1-Gal4 to express UAS-dfmr1::YFP (Cziko et al., 2009), and then pulled down FMRP-RNA complexes from larval lysates using anti-YFP beads (Nagai et al., 2002; Rana et al., 2018). In parallel, non-tagged w¹¹¹⁸ third-instar lysates served as the immunoprecipitation (IP) negative control. In both cases, α-Tubulin (αTub85E; FMRP does not bind) was the negative control and Futsch (a known FMRP target) was the positive control (Zhang et al., 2001). Immunoprecipitated mRNAs were reverse transcribed using random hexamers, followed by specific primer PCR amplification to produce ∼200 bp PCR fragments (Table S2). Downstream of hypothesized mRNA binding, we also assayed staufen mRNA levels with qPCR measurements in dfmr1 null mutants and with muscle-targeted dfmr1 RNAi (Flockhart et al., 2006) to test the postsynaptic interaction in isolated muscle analyses.

The FMRP IP pulled down staufen mRNA from larval lysates (Fig. 4A, IP, top). Consistently, the futsch mRNA positive control was precipitated in parallel, with no detectable binding to the α-tubulin mRNA negative control (Fig. 4A). As expected, the genetic negative control w¹¹¹⁸ (no YFP) showed no immunoprecipitated bands (Fig. 4A, IP, bottom). These results indicated that FMRP binds to staufen mRNA from the wandering third instar, with the controls confirming the binding interaction specificity. RNA binding protects transcripts from degradation by increasing RNA stability, so we hypothesized that FMRP binding should increase staufen mRNA levels. To test this idea, we performed qPCR to measure staufen mRNA levels in genetic background controls (w¹¹¹⁸) compared with dfmr1 nulls (Fig. 4B). Quantification showed that staufen mRNA levels were significantly reduced in dfmr1 mutants normalized to controls (control 1.00±0.05, dfmr1 0.68±0.08; P=0.002; Fig. 4B). This finding suggested that FMRP stabilizes staufen mRNA through protein-RNA binding. To test postsynaptic roles, we used muscle 24B-Gal4 to drive dfmr1 RNAi, and then isolated body muscles for mRNA extraction (Fig. 4C). Quantified qPCR results showed that staufen mRNA levels were strongly downregulated in 24B>dfmr1 RNAi muscles normalized to the transgenic control (24B/+ 1.00±0.09, RNAi 0.61±0.10; P=0.009; Fig. 4C). These findings suggest that FMRP binding stabilizes staufen mRNA in the postsynaptic muscle.

At the NMJ, postsynaptic Staufen is required for the localization and translation of coracle mRNA, which encodes a mammalian 4.1 ortholog functioning as a GluRIIA anchoring scaffold (Gardiol and St Johnston, 2014). To test Staufen and coracle mRNA interaction, we used UH1-Gal4 to drive UAS-staufen::GFP (Barbee et al., 2006; Laver et al., 2013) and pulled down RNA complexes from larval lysates using anti-GFP beads (Fig. 4D). As above, w¹¹¹⁸ (no GFP) was the IP negative control. As a positive control, Staufen binds α-tubulin mRNA (Laver et al., 2013), whereas peptidyl-α-hydroxyglycine-α-amidating lyase 1 (Pal1) mRNA reportedly is not bound by Staufen (Laver et al., 2013) and was therefore selected as a negative control. Staufen IP pulled down coracle mRNA, but also pulled down α-tubulin and Pal1 mRNA (Fig. 4D). We also tested GAPDH (Gapdh2), RP49 (RpL32) and Gal4 mRNAs, and all of these were also immunoprecipitated. Repeated trials with increasing transfer RNA (tRNA) concentrations (300 μg, 600 μg, 900 μg, 1 mg) or even highly elevated tRNA (10 mg, 20 mg) all showed continued mRNA pulldown. Thus, Staufen binds coracle mRNA, but lacks binding specificity. Staufen is also predicted to bind dfmr1 mRNA (Laver et al., 2013), but FMRP levels did not change in staufen mutant muscle (control 1.00±0.04, stau^HL 0.91±0.07; P=0.30; Fig. S4A,B). We therefore suggest that there is a directional pathway of FMRP binding staufen mRNA to control postsynaptic muscle levels, with Staufen in turn binding coracle mRNA.

The Staufen dsRNA-binding domain 5 is specifically required for Coracle local translation (Gardiol and St Johnston, 2014). Disruption of this domain in stau^HL over the genomic deficiency [stau^HL/Df(2R)Pcl7B] reportedly impairs postsynaptic accumulation of Coracle protein via loss of local translation, without affecting coracle mRNA localization (Gardiol and St Johnston, 2014), suggesting that Staufen regulates local Coracle translation specifically within the NMJ postsynaptic domain. Coracle binds the GluRIIA C-terminus to scaffold receptors in the postsynaptic membrane, with tight stoichiometry between Coracle and GluRIIA levels within muscle (Chen et al., 2005). As FMRP and Staufen both repress GluRIIA accumulation, we hypothesized that both proteins should inhibit postsynaptic Coracle expression. To test how FMRP and Staufen might regulate Coracle, alone and in combination within the FMRP-Staufen pathway, we used an anti-Coracle antibody (Gomez et al., 2012) to measure levels in anti-HRP-labeled NMJs in wandering third instars. Coracle levels were measured in genetic background controls (w¹¹¹⁸), dfmr1 and staufen homozygous mutants, dfmr1/+ and staufen/+ heterozygotes, and dfmr1/+; stau/+ trans-heterozygous double mutants. The expression was quantified postsynaptically surrounding anti-HRP thresholded synaptic boutons, within a dilated 1 μm region of interest to capture the postsynaptic SSR domain.

Coracle encircled NMJ boutons, with a more intense ring in dfmr1 mutants than in controls (Fig. 5A). Quantification showed that normalized Coracle levels were significantly elevated in dfmr1 mutants (control 1.00±0.18, dfmr1 1.68±0.13; P=0.0081; Fig. 5B). To test possible feedback regulation, we assayed anti-FMRP in coracle mutant muscle, but found no detectable change (control 1.00±0.10, cora 1.14±0.12; P=0.36; Fig. S4C,D), showing that FMRP acts upstream of Coracle. Compared with genetic controls (w¹¹¹⁸), stau mutants also had more intense Coracle rings around boutons (Fig. 5C). Quantification showed that staufen mutants also exhibited significantly more postsynaptic Coracle expression (control 1.00±0.08, stau 1.32±0.05; P=0.0018; Fig. 5D). These findings indicate that FMRP and Staufen similarly limit Coracle in the NMJ postsynaptic domain. To test FMRP and Staufen action in the same pathway, we assayed dfmr1/+; stau/+ trans-heterozygotes. Neither dfmr1/+ nor stau/+ single heterozygotes showed any detectable difference in Coracle levels compared with controls (Fig. S5A-D). By contrast, the dfmr1/+; stau/+ trans-heterozygotes had clearly enhanced postsynaptic Coracle rings around NMJ boutons (Fig. 5E). Quantification showed that the double mutant had a significant 50% increase in normalized Coracle levels compared with the control (control 1.00±0.14, dfmr1/+; stau/+ 1.53±0.20; P=0.034; Fig. 5F). We suggest that FMRP and Staufen act sequentially to inhibit Coracle GluRIIA-scaffold enrichment in the postsynaptic domain.

Null coracle mutants are embryonic lethal (Lamb et al., 1998), and total Coracle loss impairs GluRIIA accumulation at the embryonic NMJ (Chen et al., 2005). As a GluRIIA-binding scaffold, Coracle anchors receptors to underlying F-actin cytoskeleton in the postsynaptic domain (Chen et al., 2005). Consistently, our above results predicted that Coracle OE should be causally associated with an increase in postsynaptic GluRIIA levels. However, many scaffolds like Coracle show similar phenotypes with loss and OE (McCarthy, 2010), including scaffolds at intercellular junctions (Tokuda et al., 2014) and specifically at neuronal synapses (Fulterer et al., 2018). We therefore hypothesized that disrupting Coracle levels in either direction could generate elevated GluRIIA levels and, secondarily, supernumerary bouton formation. To test this hypothesis, we assayed in parallel a larval viable coracle hypomorphic mutant (cora¹⁴; Khadilkar et al., 2017; Lamb et al., 1998) and muscle-targeted 24B-Gal4 coracle RNAi (Jiang et al., 2019), as well as coracle OE (Ward et al., 1998). For all three conditions and matched controls, we used double labeling with presynaptic anti-HRP and postsynaptic anti-DLG to assay NMJ architecture and quantify synaptic bouton number. We also double labeled with anti-HRP and anti-GluRIIA to assay synaptic GluRIIA expression and quantify receptor level based on fluorescence intensity.

Both coracle mutants and muscle-targeted coracle RNAi produced enlarged NMJs with more synaptic boutons (Fig. 6A). Quantification showed that bouton numbers increase in coracle mutants (control 21.00±1.10, cora¹⁴ 32.86±1.79; P<0.0001) and with muscle coracle RNAi (24B/+ control 19.20±1.06, RNAi 26.21±1.69; P=0.0014; Fig. 6B). NMJ bouton formation was also elevated by muscle-specific coracle OE (Fig. 6A, bottom). Quantification showed that bouton number was significantly elevated by 24B-Gal4-targeted coracle OE (24B/+ 21.33±0.64, cora OE 30.00±1.12; P<0.0001; Fig. 6B). Thus, both Coracle loss and gain in the postsynaptic muscle similarly restricts presynaptic development. Similarly, coracle mutants, muscle-targeted coracle RNAi and OE all had more postsynaptic GluRIIA than controls (Fig. 6C). Quantification showed that normalized GluRIIA levels were significantly higher in coracle mutants (control 1.00±0.05, cora¹⁴ 1.32±0.10; P=0.0085) and muscle-specific 24B-Gal4>cora RNAi (24B/+ 1.00±0.06, RNAi 1.30±0.13; P=0.03; Fig. 6D) than in controls. Supporting our hypothesis, quantification likewise showed that normalized GluRIIA levels were highly elevated by coracle OE in muscle (24B/+ 1.00±0.06, cora OE 1.35±0.096; P=0.004; Fig. 6D). Taken together, these results suggest that postsynaptic FMRP restricts Staufen to restrict Coracle to restrict GluRIIA levels and thus presynaptic bouton formation, with loss and gain of Coracle phenocopying each other within this GluRIIA regulative mechanism.

Clustered postsynaptic GluRIIA channels mediate excitatory ion influx during neurotransmission (Han et al., 2015; Müller and Davis, 2012). GluRIIA levels are thus positively correlated with enhanced excitatory synaptic strength (Petzoldt et al., 2014). We therefore hypothesized that impairment of the FMRP–Staufen–Coracle pathway should elevate function. To test this hypothesis, two-electrode voltage-clamp (TEVC) recordings in dfmr1, staufen and coracle mutants were compared with those in genetic background controls (w¹¹¹⁸). Evoked excitatory junction current (EJC) amplitude provides a measure of overall NMJ neurotransmission efficacy dependent on postsynaptic GluRs precisely juxtaposed to the presynaptic active zone glutamate release sites (Clarke et al., 2012; Hong et al., 2020; Marrus, 2004). To make EJC recordings, suprathreshold stimuli (0.5 ms) were applied with a motor nerve suction electrode at 0.2 Hz (Kopke et al., 2020). Ten sequential evoked traces were recorded in 1 mM [Ca²⁺] and averaged to generate each data point (Kopke et al., 2020). Miniature EJC (mEJC) recordings assay spontaneous synaptic vesicle fusion events, with frequency indicating presynaptic fusion probability and amplitude correlated with postsynaptic GluR function (Harris and Littleton, 2015). In these recordings, mEJCs were analyzed at 10 kHz in continuous gap-free configuration (Kopke et al., 2020).

Compared with the EJC amplitude of the genetic control (w¹¹¹⁸), dfmr1, staufen and coracle mutants all showed consistently elevated EJC amplitudes (Fig. 7A). Relative to the control EJC amplitude (126.9±11.75 nA), NMJ synaptic strength was significantly greater in dfmr1 (190.0±19.19; P=0.0069), staufen (185.0±11.24; P=0.0091) and coracle (206.7±12.59; P=0.0002) mutants (Fig. 7B), consistent with their GluRIIA accumulation. Spontaneous mEJC events failed to reveal any obvious changes in these mutants (Fig. 7C). The mEJC frequency was not detectably altered, and quantification showed no change in mutants (staufen: P=0.92; coracle: P=0.67; Fig. 7D), indicating that synapse number and presynaptic vesicle fusion probability were unaltered. The mEJC amplitude was also unchanged in mutants (Fig. 7C), and quantification showed no change in mutants (staufen: P=0.41; coracle: P=0.93; Fig. 7E), indicating that GluR conductance was unaffected. Similar EJC-specific phenotypes have been repeatedly reported at the Drosophila NMJ (e.g. Wang et al., 2017), which might reflect postulated differences between evoked and spontaneous fusion mechanisms (Horvath et al., 2020; Kavalali, 2015) or compensatory interactions between GluRIIA number and conductance (Petzoldt et al., 2014; Renden and Broadie, 2003). We conclude that FMRP, Staufen and Coracle all repress postsynaptic GluRIIA accumulation and functional neurotransmission strength, but we still need a mechanistic link to the increase in presynaptic bouton formation via this pathway.

Functional GluRIIA accumulation in the dfmr1 mutant induces presynaptic bouton development via a non-canonical BMP trans-synaptic signaling pathway (Sulkowski et al., 2016; Kamimura et al., 2019). This may not involve a BMP ligand, but rather a direct interaction between postsynaptic GluRIIA and presynaptic BMP receptor (Sulkowski et al., 2016). Intercellular signaling triggers synaptic phosphorylation of pMad to induce bouton formation (Kamimura et al., 2019; Sulkowski et al., 2016; Upadhyay et al., 2017). As dfmr1, staufen and coracle mutants all showed increased GluRIIA levels and bouton formation, we hypothesized that all mutants activate GluRIIA-dependent signaling of presynaptic pMad. To test this idea, we triple labeled all three mutants with anti-HRP (to mark neuronal presynaptic membrane), anti-Bruchpilot (Brp; to mark presynaptic active zones) and anti-pMad (Kamimura et al., 2019). The Brp-positive active zones and pMad levels were assayed within anti-HRP thresholded boutons using laser-scanning confocal microscopy (LSM). However, this approach has restricted X-Y resolution to visualize the small, closely spaced active zones (∼500-600 nm diameter; Guggenheim et al., 2016; Pielage et al., 2006; Wegel et al., 2016). Therefore, to better resolve pMad around Brp puncta, we also used higher resolution structured illumination microscopy (SIM) (Guggenheim et al., 2016).

In LSM imaging, Brp-positive active zone numbers in dfmr1, staufen and coracle mutant NMJs were all comparable to those in control NMJs, whereas the surrounding pMad labeling was consistently elevated in all three mutants (Fig. 8A). Note that Brp did not colocalize with pMad, indicating that pMad surrounds the presynaptic active zones but is not present within each synapse. Compared with genetic control (w¹¹¹⁸) normalized pMad levels (intensity: 1.00±0.061), fluorescence quantification showed that presynaptic pMad was significantly elevated in dfmr1 (1.52±0.14; P=0.005), staufen (1.64±0.14; P=0.001) and coracle (1.78±0.14; P<0.0001) mutants (Fig. 8C). At higher resolution, SIM imaging clearly revealed elevated pMad levels surrounding presynaptic active zones in all dfmr1, staufen and coracle mutants compared with controls (Fig. 8B). Note in single NMJ boutons that Brp and pMad labeling was non-overlapping, but adjacent. In all three of the mutants, presynaptic pMad aberrantly accumulated around Brp-positive active zones. Importantly, quantification of the active zones compared with matched control (Brp puncta density/µm²: 1.44±0.07) showed no significant change in the coracle (1.54±0.05; P=0.54), staufen (1.46±0.02; P=0.99) or dfmr1 (1.63±0.11; P=0.11) mutants (Fig. 8D). These findings suggest that elevated pMad levels in all three mutants correlate with increased GluRIIA-dependent retrograde trans-synaptic signaling from the postsynaptic domain, rather than presynaptic active zone density.

To directly test the postsynaptic to presynaptic signaling mechanism, all three genes were knocked down with muscle-targeted RNAi (Fig. S6). Compared with the driver control (24B-Gal4/+), postsynaptic dfmr1 RNAi caused a clear and consistent increase in presynaptic pMad levels (Fig. S6A). Quantification normalized to the control showed a significant elevation in the knockdown (control 1.00±0.086, dfmr1 RNAi 1.40±0.12; P=0.0098), with no change in BRP-marked active zone density (P=0.85; Fig. S6B). Similarly, targeted knockdown of postsynaptic staufen resulted in an obvious elevation in pMad levels (Fig. S6C). Quantification again indicated a significant increase in pMad fluorescence (control 1.00±0.085, stau RNAi 1.44±0.13; P=0.0072), with no change in synapse number (P=0.78; Fig. S6D). Finally, muscle-targeted coracle RNAi also drove heightened presynaptic pMad levels (Fig. S6E). Quantification likewise showed that there is a significant increase in pMad in the presynaptic boutons (control 1.00±0.084, cora RNAi 1.28±0.10, P=0.049), with no change in BRP-marked active zone density (Fig. S6F). These findings indicate that targeted loss of all three genes in the postsynaptic muscle causes trans-synaptic elevation of pMad surrounding neuronal presynaptic active zones. Taken together, these results suggest that accumulated postsynaptic GluRIIA in dfmr1, staufen and coracle mutants activates presynaptic pMad signaling, which in turn induces new bouton formation during NMJ synaptic development.

This study reveals the mechanism of the established FMRP negative regulation of postsynaptic GluRIIA receptors and presynaptic bouton formation in the Drosophila FXS disease model (Pan and Broadie, 2007; Zhang et al., 2001). Specifically, the mRNA-binding FMRP-positive translational regulator binds to staufen mRNA as predicted (D'Annessa et al., 2019), within the postsynaptic cell. Consequently, both dfmr1 and staufen mutants share the elevated GluRIIA level and bouton number phenotypes based on a common postsynaptic pathway function, and genetically interact as trans-heterozygotes to reproduce these phenotypes. Staufen acts as a dsRBP (Banerjee and Barraud, 2014) to bind coracle mRNA as predicted (Laver et al., 2013); both dfmr1 and staufen mutants exhibit elevated postsynaptic Coracle levels, and genetically interact as trans-heterozygotes to reproduce this phenotype. Coracle acts as a GluRIIA-binding anchoring scaffold within the postsynaptic domain to regulate local receptor accumulation (Chen et al., 2005). Consequently, dfmr1, staufen and coracle mutants all increase NMJ synaptic functional differentiation to elevate neurotransmission strength. Finally, the elevated postsynaptic GluRIIA levels mediate retrograde BMP receptor trans-synaptic signaling that induces pMad to drive new presynaptic bouton development (Kamimura et al., 2019; Sulkowski et al., 2016). dfmr1, staufen and coracle mutants all exhibit elevated presynaptic pMad levels, thereby linking the postsynaptic GluRIIA accumulation and presynaptic supernumerary bouton formation defects shared by all of these mutants.

The staufen mutant increased synaptic Coracle levels, GluRIIA levels and bouton number are all internally consistent. In a previous study (Gardiol and St Johnston, 2014), opposite phenotypes were measured in staufen^HL/Df(2R)Pcl7B, which reduces another 14 genes in heterozygous deficiency, including loci involved in neuronal development (e.g. grh, nopo; Almeida and Bray, 2005; Bakshi et al., 2020; Ketosugbo et al., 2017; Merkle et al., 2009). Importantly, we similarly found reduced synaptic protein levels and bouton number in staufen^HL/Df(2R)Pcl7B, suggesting that heterozygosity of one or more of the neighboring genes impairs synaptic development (Mutsuddi et al., 2004; Tsou et al., 2015; Yilmazer et al., 2016). However, we showed that a staufen RNAi that reduces transcript levels by ∼90% replicates the staufen mutant NMJ phenotypes of increased GluRIIA levels and synaptic bouton numbers. We also replicated this with a second, independent staufen RNAi line. Moreover, we showed that the effect is entirely restricted to postsynaptic muscle RNAi, with no effect from presynaptic neuron RNAi, consistent with restricted postsynaptic Staufen function (Gardiol and St Johnston, 2014). In addition, postsynaptic staufen rescue of the staufen mutant restored normal synaptic bouton formation, with OE reducing GluRIIA levels in staufen mutants and rescuing GluRIIA levels in dfmr1 mutants. Both staufen mutants and postsynaptic staufen RNAi also share the arrested supernumerary satellite bouton development characterizing dfmr1 null mutants (Friedman et al., 2013). These many independent lines of evidence confirm our results, and are consistent with the known parallel FMRP role in restricting GluRIIA levels and synaptic bouton formation (Pan and Broadie, 2007; Zhang et al., 2001).

To regulate Staufen, FMRP binds staufen mRNA and protects targeted staufen transcripts from degradation. FMRP contains at least three distinct RNA-binding domains (RBDs) (Kenny and Ceman, 2016), and Staufen has five RBDs (Laver et al., 2013). Staufen reportedly binds a specific RNA hairpin structure formed by long 3′ UTRs (Gardiol and St Johnston, 2014; Laver et al., 2013), but our RIP shows that Staufen also binds mRNAs that are not predicted to generate this secondary structure (Ramos et al., 2000). Although the decreased staufen mRNA levels in both dfmr1 mutants and muscle-targeted dfmr1 RNAi are predicted to be due to the lack of FMRP binding, it is also possible that other unregulated interactors cause the downregulated staufen mRNA expression (Shah et al., 2020). Localized labeling with an anti-Staufen antibody has been reported in the postsynaptic NMJ (Gardiol and St Johnston, 2014), which we can confirm, but we could not reduce labeling in staufen hypomorphic mutants. We therefore have not shown Staufen labeling in the current study. Moreover, western blots have been reported with the same anti-Staufen antibody (St Johnston et al., 1991); however, our attempts were unsuccessful. We therefore used qPCR to measure staufen mRNA levels. Staufen binds to coracle mRNA, but does so in a non-selective manner. This result is consistent with Staufen acting as a very broad spectrum dsRBP (Heraud-Farlow and Kiebler, 2014; Laver et al., 2013), and suggests that Staufen likely acts with a translational regulator partner to generate specificity. FMRP is very well established to partner with other RBPs to mediate the translational regulation of its target transcripts (Bardoni et al., 1999, 2003; Didiot et al., 2009; Kenny et al., 2014).

The postsynaptic Coracle scaffold acts in a GluRIIA local anchoring mechanism, presumably to link the receptors to the underlying actin cytoskeleton (Chen et al., 2005). The jointly elevated Coracle and GluRIIA levels in both dfmr1 and staufen mutants are consistent with this scaffold function. Because the dfmr1/+; staufen/+ trans-heterozygotes share this correlated Coracle and GluRIIA upregulation in the postsynaptic domain, a single common signaling pathway is indicated. Coracle also restricts terminal branching development in peripheral sensory neurons (Jiang et al., 2019; Tenenbaum et al., 2017). Both coracle mutants and sensory neuron-targeted coracle RNAi also display increased dendritic branch and termini numbers. These phenotypes are similar to the expanded NMJ terminals and increased synaptic bouton development reported here. Importantly, both coracle loss of function (mutants and muscle-targeted RNAi) and gain of function (muscle-targeted OE) increase postsynaptic GluRIIA levels and generate supernumerary boutons. Likewise, the knockdown and OE of many other similar scaffolds are known to cause phenocopying defects (McCarthy, 2010). Some examples include the muscle chaperone UNC-45 (Landsverk et al., 2007), the tight junction scaffold zonula occludens-1 (ZO-1) (Tokuda et al., 2014) and synaptic UNC-13 (Fulterer et al., 2018). Indeed, both coracle loss and OE similarly cause increased dendritic crossing in Drosophila sensory neurons (Tenenbaum et al., 2017), similar to the phenocopy of developmental defects reported here. Combining the roles of postsynaptic FMRP–Staufen–Coracle in GluRIIA clustering, we reasoned that this pathway must be a regulatory determinant of synaptic functional development.

Removing FMRP, Staufen and Coracle strongly enhances functional synaptic differentiation and NMJ neurotransmission strength. This is consistent with expectations from the postsynaptic GluRIIA accumulation in all of these mutants (Harris and Littleton, 2015). Elevated GluRIIA levels are well known to be associated with increased evoked functional responses and prolonged channel open times (DiAntonio et al., 1999; Schmid et al., 2008). A GluRIIA pore sequence (MQQ) critically required for the Drosophila channel Ca²⁺ permeability is conserved in mammalian receptors (Petersen et al., 1997). This selectivity allows Ca²⁺-dependent participation in spontaneous (mEJC) and evoked (EJC) neurotransmission (Han et al., 2015). Although enhanced evoked EJC amplitudes are typically accompanied by mEJC alterations (Karunanithi et al., 2020; Sandstrom, 2011; Tsurudome et al., 2010), we find that mEJC amplitude and frequency are unchanged in both the staufen and coracle mutants, and show only minimal changes in the dfmr1 mutants (Zhang et al., 2001). Classically, both evoked and spontaneous neurotransmission were thought to be mediated by the same vesicles (del Castillo and Katz, 1954; Groemer and Klingauf, 2007); however, more recent evidence has indicated that spontaneous and evoked neurotransmission have distinct machinery and vesicle pools (Groffen et al., 2010; Horvath et al., 2020; Ramirez et al., 2012; Sara et al., 2005). Postsynaptic receptors can be segregated into different compartments that are activated by either spontaneous or evoked release (Atasoy et al., 2008). Our work supports this growing body of evidence for differential regulation. Importantly, GluRIIA has unique functions, modulating both presynaptic glutamate release and presynaptic bouton development (Bogdanik et al., 2004; Kamimura et al., 2019).

The dfmr1, staufen and coracle mutants all showed upregulated presynaptic pMad correlated with postsynaptic activated GluRIIA accumulation. GluRIIA activation triggers presynaptic pMad signaling via BMP receptors surrounding active zones, which, in turn, stabilizes GluRIIA receptors in the postsynaptic domains (Sulkowski et al., 2016). This trans-synaptic signaling mechanism induces new presynaptic bouton development. The targeted postsynaptic RNAi for all three genes confirms this intercellular link. Synaptic BMP signaling involves both the type I serine/threonine kinase receptors and the type II receptor Wit (Upadhyay et al., 2017). Although BMP ligand Glass bottom boat (Gbb) signaling via Wit presynaptic receptors is well established at the NMJ to modulate synaptogenesis (Ellis et al., 2010; Kim et al., 2019; McCabe et al., 2003), the mechanism of presynaptic bouton formation induced by activated GluRIIA signaling does not involve canonical BMP signaling via Gbb (Friedman et al., 2013; Kamimura et al., 2019). In the dfmr1 mutants, we suggest that postsynaptic GluRIIA accumulation induces presynaptic bouton development via non-canonical GluRIIA-Wit trans-synaptic retrograde signaling (Sulkowski et al., 2016). Similarly, the muscle postsynaptic glypican Dally-like protein (Dlp) (Kamimura and Maeda, 2021) negatively regulates NMJ synaptic development by inhibiting this same non-canonical BMP pathway through decreased activated GluRIIA expression (Kamimura et al., 2019). Postsynaptic GluRIIA clustering can thus trigger presynaptic bouton formation, although supernumerary boutons do not always induce reciprocal GluRIIA changes (Sulkowski et al., 2016). We conclude that an FMRP–Staufen–Coracle–GluRIIA–pMad pathway regulates intertwined structural and functional glutamatergic synapse development.

All stocks were reared at 25°C on standard cornmeal/agar/molasses food. The genetic background control was w¹¹¹⁸. The viable dfmr1 mutant was w; dfmr1^50M, in which the dfmr1 locus has been completely removed via a P-element imprecise excision deletion (Zhang et al., 2001). The larval viable staufen mutant was w; stau^HL, which has a point mutation in the dsRNA-binding domain 5 specifically required for local mRNA translation (Fig. S1A; St Johnston et al., 1991). The viable coracle mutant was w; cora¹⁴, which has a nonsense mutation (Arg¹⁶⁰⁷) reducing function (Lamb et al., 1998; Ward et al., 2001). Transgenic drivers were ubiquitous daughterless UH1-Gal4 (Rohrbough et al., 2004), neuronal elav-Gal4 (Kan et al., 2021) and muscle-specific 24B-Gal4 (Kim et al., 2021). All genetic crosses and recombinations to make double mutant lines were done using standard approaches. Transgenic UAS lines used in this study are listed in Table S1.

Staged wandering third instars were dissected in phosphate-buffered saline (PBS). For FMRP, DLG, Brp and pMad labeling, tissues were fixed in 4% paraformaldehyde (PFA) diluted in PBS for 10 min at room temperature (RT). Coracle labeling was performed with 20 min fixation at RT. To label GluRIIA, larvae were fixed in 100% Bouin's fixative (Karr et al., 2009) for 5 min at RT. Fixed preparations were blocked for 1 h at RT in PBS with 0.2% Triton X-100 (PBST) plus 1% bovine serum albumin (BSA). Primary antibody incubation was done overnight at 4°C and secondary antibody incubation was done for 2.5 h at RT. Primary antibodies used were as follows: mouse anti-DLG [Developmental Studies Hybridoma Bank (DSHB), 4F3, 1:50], mouse anti-GluRIIA (DSHB, 8B4D2, 1:50), mouse anti-Coracle (DSHB, C566.9, 1:50), mouse anti-FMRP (Abcam, 10299, 1:250), mouse anti-Brp (DSHB, NC82, 1:100), rabbit anti-Smad3 (phospho S423+S425, Abcam 52903, 1:500), Cy5-conjuagted goat anti-HRP (Jackson ImmunoResearch, 147967, 1:250) and Cy3-conjugated goat anti-HRP (Jackson ImmunoResearch, 137589, 1:250). Secondary antibodies used were as follows: goat 488 anti-mouse (Invitrogen, A11001, 1:250), goat 488 anti-rabbit (Invitrogen, A11008, 1:250), donkey 555 anti-mouse (Invitrogen, A31570, 1:250) and donkey 488 anti-mouse (Invitrogen, A21202, 1:250).

All confocal imaging was performed on a Zeiss LSM 510 META laser-scanning confocal microscope and projected in Zen software (Kopke et al., 2020). The NMJ areas and fluorescent intensities were analyzed via blinded z-stack sum projection in FIJI software (Guillen et al., 2020). GluRIIA levels were quantified in HRP-labeled NMJ areas with eliminated muscle intensity background, while Coracle levels were quantified in dilated 1 μm rings surrounding individual NMJ boutons. For SIM, samples were imaged on a Nikon N-SIM microscope in 3D SIM mode (Kopke et al., 2020). Fluorophores were activated by 647 nm, 561 nm and 488 nm diode lasers. With a SR Apo TIRF 100× oil objective (1.49 NA WD 0.12) and an Andor iXon Ultra DU-897 EMCCD monochrome camera, samples were reconstructed through NIS-Elements (Nikon) with 0.12 μm step-size stacks. Stack reconstructions of the blinded raw data were done before image rendering and measurement analyses.

The total RNA from wandering third instars was isolated according to the instructions in the RNeasy Plus Mini Kit (Qiagen 74134). RNA (2 µg) measured by a Nanodrop 2000C was then reverse transcribed into complementary DNA (cDNA) with random hexamers using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4368814). Resulting single-strand cDNA was then subjected to real-time PCR employing the SYBR Green Master Mix (Applied Biosystems, A25742) and using the CFX384 Touch Real-Time PCR system (Bio-Rad). Targeted transcripts were normalized to reference gene cDNA (GAPDH2). The cDNA primers used in this study are listed in Table S2.

Fifty wandering third instars of each genotype were homogenized in 500 µl RNase-free lysis buffer [20 mM HEPES, 100 mM NaCl, 2.5 mM EDTA, 0.05% (v/v) Triton X-100, 5% (v/v) glycerol] with 1% β-mercaptoethanol, 1× protease inhibitor cocktail (cOmplete mini EDTA-free Tablets, Sigma-Aldrich), and 400 U RNase inhibitor (Applied Biosystems, N8080119). To preclear the supernatant, the centrifuged lysates were incubated with 20 µl Protein G Dynabeads for 1 h at 4°C. In parallel, 200 µl Protein G Dynabeads were incubated in blocking buffer (1×PBS, 0.2% TWEEN 20, 0.1% tRNA and 5% BSA) for 1 h at 4°C, followed by coating with 10 µg of the primary antibody. Next, the precleared supernatant was incubated with antibody-bead conjugates overnight at 4°C. To reduce non-specific RNA binding in larval lysates, tRNA (Sigma-Aldrich, 10109541001) was added in lysis buffer as specified (e.g. 1 mg tRNA per 500 μl IP reaction). To purify bound RNAs, washed bead-protein-RNA complexes were incubated with a 500 μl TRIzole and chloroform mixture (Ambion, 15596026) for 10 min. Subsequently, 3 µl glycogen was applied to carry RNAs for the precipitation by mixing with 100 µl 2-propenol. The precipitated RNA was then reverse transcribed into single-strand cDNA and subjected to primer-specific PCR (Table S2). We used 2% agarose gels to analyze the PCR products. Primary antibodies used were mouse anti-Venus YFP (Sigma-Aldrich, MABE1906) and mouse anti-GFP (Sigma-Aldrich, G6539).

Wandering third instars were dissected along the dorsal midline, internal organs were removed and body walls were glued down (Vetbond, 3 M). Next, all peripheral motor nerves were cut at the base of the ventral nerve cord. TEVC recordings were performed at 18°C in physiological saline (128 mM NaCl, 2 mM KCl, 4 mM MgCl₂, 1 mM CaCl₂, 70 mM sucrose and 5 mM HEPES, pH 7.2). Synaptic currents were recorded from ventral longitudinal muscle 6 of abdominal segments 3/4 under a Zeiss Axioskop microscope using a 40× water-immersion objective (Kopke et al., 2020). Muscles were impaled with two microelectrodes (1 mm outer diameter borosilicate capabilities; World Precision Instruments) of ∼15 MΩ resistance filled with 3 m KCl, and then clamped at a command voltage of −60 mV employing an Axoclamp-2B amplifier (Axon Instruments; Kopke et al., 2020). To make the EJC recordings, the motor nerve was stimulated using a fire-polished glass suction electrode with 0.5 ms suprathreshold voltage stimuli at 0.2 Hz (Grass S88 stimulator). Data were filtered at 2 kHz. To quantify EJC amplitude, ten consecutive traces were recorded and averaged, with the average peak amplitude being reported. Spontaneous mEJC events were recorded in continuous 2 min sessions at 10 kHz, and analyzed using a 200 Hz low-pass filter (Kopke et al., 2020). Clampex 9.0 was used for data acquisition, and Clampfit 9 was used for data analysis (Axon Instruments).

All statistics were performed using GraphPad Prism software (v8.0). All data sets were subject to ROUT outlier tests with Q set to 1%. All paired data comparisons (i.e. bouton numbers, GluRIIA levels, FMRP levels, Coracle levels and qPCR measurements) were assayed using unpaired two-tailed Student's t-tests for two-way comparison with 95% confidence. All data sets of more than two comparisons (i.e. electrophysiology results, pMad levels and active zone numbers) were analyzed using one-way ANOVA. Dunnett's multiple comparison tests were performed to compare the mean of each experimental data set with the control mean. All figures show mean±s.e.m., with P≤0.05 considered significant.

We are grateful to the BDSC (Indiana University, Bloomington, IN, USA) and the VDRC (Vienna, Austria) for essential genetic lines, and the DSHB (University of Iowa, Iowa City, IA, USA) for essential antibodies. We thank Daniel St Johnston (The Gurdon Institute, University of Cambridge, Cambridge, UK) and Richard Fehon (University of Chicago, Chicago, IL, USA) for staufen and coracle stocks, respectively. We thank Julian Hillyer, He Huang and Jim Patton (Vanderbilt University, Nashville, TN, USA) for qPCR equipment use, qPCR advice and RIP advice, respectively. We thank the Broadie Laboratory for extensive input.