Muscle cofilin alters neuromuscular junction postsynaptic development to strengthen functional neurotransmission Free

Skeletal muscle is essential for daily functions, including respiration, posture control, and coordinated movement. Muscle contraction and relaxation require many levels of organization: each cell is made up of myofibrils, which are concatenations of the contractile sarcomeres. The sarcomere is composed of thin filaments anchored at Z-discs and thick filaments containing myosin motors that enable filament sliding. Thin filaments comprise filamentous actin (F-actin) and actin-binding proteins that anchor, maintain length, and regulate myosin binding.

The actin depolymerizing factor (ADF)/cofilin family maintains filament length by inducing a unique conformational change at ADP-actin filaments to promote actin severing and prevent G-actin nucleotide exchange (Galkin et al., 2011; Hayden et al., 1993). The vertebrate family includes ADF/destrin, cofilin 1 and cofilin 2, with cofilin 2 predominant in postnatal and mature skeletal muscle (Mohri et al., 2000; Ono et al., 1994; Vartiainen et al., 2002). Cofilin 2 is important for skeletal muscle development, maintenance and regeneration, and for determining thin filament length (Agrawal et al., 2012; Gurniak et al., 2014; Kremneva et al., 2014; Ono et al., 1994; Thirion et al., 2001). Variants of 12 actin and sarcomere-related genes, including CFL2, have been implicated in nemaline myopathy (NM), a skeletal muscle disease that presents with progressive weakness, particularly during the perinatal period in its most severe forms (Christophers et al., 2022; Sewry et al., 2019). NM diagnosis is based on myopathic clinical disease hallmarks, including hypotonia and proximal muscle weakness, as well as the presence of actin-rich accumulations (known as nemaline rods or bodies) and myofibrillar disruption on pathology. Individuals with CFL2 NM have generalized muscle weakness; delayed motor skills; stiff spine; kyphoscoliosis; and joint contractures (Agrawal et al., 2007; Fattori et al., 2018; Ockeloen et al., 2012; Ong et al., 2014). CFL2 variants have been described as recessive and hypomorphs and lead to reduced levels of cofilin activity in muscle.

Cofilin is conserved across species, and cofilin-deficient disease models have been developed in mice, Caenorhabditis elegans and Drosophila melanogaster. Several strategies have been used to study cofilin 2 in mouse models, including constitutive knockout, muscle-specific excision, chimeras that have a combination of knockout and wild-type cells, and knock-in of an NM mutation (Agrawal et al., 2012; Gurniak et al., 2014; Mohri et al., 2019; Morton et al., 2015; Rosen et al., 2020). Mice from all these models are small and have myofibrillar organization defects. Most animals die in the early postnatal period, which indicates that cofilin 2 is not essential for initial myofibrillogenesis, but is important for postnatal muscle maintenance. In C. elegans, the cofilin isoform UNC-60B is required for incorporating actin into developing myofibrils, and mutants develop actin aggregates in body wall muscle (Ono et al., 1999, 2003). In Drosophila, we previously characterized the effect of muscle-specific knockdown (KD) of the cofilin homolog twinstar, the only cofilin homolog in Drosophila (Balakrishnan et al., 2020), hereafter referred to as DmCFL. DmCFL KD leads to progressive muscle deterioration over the larval developmental stages. Actin accumulation occurs as a result of improper sarcomere addition at the growing muscle ends, and, eventually, the accumulations form throughout the cell, reminiscent of the defects seen in humans with NM. Introducing wild-type human CFL1 or CFL2 rescued this deterioration phenotype in DmCFL KD, but that of a human CFL2 NM variant did not, supporting the use of this system to understand human NM disease.

To investigate further the DmCFL KD model, we conducted RNA sequencing, which identified changes in genes related to the neuromuscular junction (NMJ), the specialized synapse where the presynaptic motor neuron and postsynaptic muscle communicate. Early case reports examining NM muscle biopsies provide evidence of NMJ disruption, describing abnormal motor end plates and synaptic clefts (Fukuhara et al., 1978; Heffernan et al., 1968). Moreover, electromyography shows a myopathic pattern in individuals with NM, although it is unclear whether there is an additional superimposed NMJ transmission defect. Cytoplasmic actin and other actin-binding proteins are present at the vertebrate NMJ (Berthier and Blaineau, 1997; Bloch and Hall, 1983; Hall et al., 1981). Additionally, in vertebrates, actin podosomes are crucial for shaping the complex organization of postsynaptic proteins (reviewed by Bernadzki et al., 2014). Thus, it is possible that disruption of actin and actin-binding proteins would affect the NMJ in humans with NM. Recently, a case report of two individuals with NM with an alpha-actin variant describes abnormal postsynaptic muscle membrane at the NMJ, but it is unknown whether this finding is generalizable to all people with NM (Labasse et al., 2022). At the widely studied Drosophila larval NMJ, proper actin organization at the muscle postsynapse is important for synaptic development, neurotransmitter receptor localization, and neurotransmission (Chen et al., 2005; Pielage et al., 2006; Wang et al., 2010). Based on the results of our RNA-sequencing experiment and these previous Drosophila studies, we sought to improve our understanding of the NMJ in the DmCFL KD model and its contribution to disease development.

Here, we report that DmCFL is present near the muscle postsynaptic membrane, where it is important for proper actin organization in this region. We find that, in DmCFL KD, although the motor neuron properly innervates the muscle, postsynaptic structural organization is disrupted prior to and in parallel with muscle structure deterioration. Although transcriptomic analysis suggested changes in many synaptic components, we did not see defects in the presence of active zones or total glutamate receptors. However, DmCFL is important for the localization of glutamate receptor subunit A (GluRIIA) at the postsynapse, and DmCFL KD results in decreased NMJ neurotransmission. Together, these findings show that cofilin is important for NMJ postsynaptic structure, actin organization, GluRIIA localization, and synaptic signaling.

We have previously developed a DmCFL muscle KD model that has disrupted sarcomere structure, protein aggregates, and reduced muscle function, which are features of NM pathology (Balakrishnan et al., 2020). We used the GAL4/UAS system with a muscle-specific Myosin heavy chain promoter (Mhc-Gal4) driving an RNA interference (RNAi) construct targeting mCherry (control, UAS-mCherry RNAi) or twinstar (UAS-tsr RNAi), the only DmCFL gene. MhcGal4>UAS-tsr RNAi (DmCFL KD) larvae have decreased DmCFL RNA and DmCFL protein in larval body wall muscles (Fig. S1A-C). DmCFL levels within the muscle are reduced to the same extent in all DmCFL KD muscle classes relative to control (Fig. S1D).

The DmCFL KD model shows a progressive muscle deterioration phenotype that we separate into three classes. Class 1 muscles retain sarcomeric actin organization in the muscle cell, class 2 muscles develop accumulations of actin at the cell poles with retained sarcomeric organization, and class 3 muscles lose overall muscle and sarcomeric integrity, in addition to having actin accumulations throughout the cell (Balakrishnan et al., 2020). Our data also showed a deterioration progression from class 1 into class 2 muscles, which, ultimately, give rise to class 3 muscles. At the final wandering third instar, all three muscle classes are present in an individual larva, with classes 2 and 3 predominating. In the present study, we analyzed ventral longitudinal muscles VL3 and VL4 (also known as muscles 6 and 7, respectively). These paired muscles are well-characterized, easily accessible, and share the same motor neuron innervation. We first performed a series of imaging and transcriptomic experiments to gain a better understanding of the DmCFL KD phenotype.

First, to characterize the frequencies of the different muscle class combinations, we analyzed the F-actin sarcomeric structure at the wandering third-instar stage based on phalloidin labeling. Despite sharing the same orientation, body position, and motor neuron, we found that these two muscles can be different classes, and thus, are not dependent on each other for their structural integrity (Fig. 1A, Fig. S1E). We considered the combination of VL3/4 classes as groups. Group 1 pairs had only class 1 muscles, group 2 pairs had at least one class 2 muscle, and group 3 pairs had at least one class 3 muscle. Most muscle pairs in wandering third-instar larvae consisted of two class 3 muscles (62%), yet even at this late stage some muscle pairs consisted of two class 1 muscles (16%; Fig. 1B). We used confocal microscopy to image live larvae expressing a DmCFL::GFP protein, which showed that DmCFL localized to both the muscle sarcomere and myonuclei (Fig. 1C-E). Using immunofluorescence and structured illumination microscopy (SIM), we confirmed that DmCFL is present at the sarcomere Z-line and H-zone (Fig. 1F). Importantly, DmCFL was reduced at the sarcomere in DmCFL KD (Fig. 1F; Balakrishnan et al., 2020).

To determine the transcriptomic profile of DmCFL KD muscles, we dissected third-instar larvae to produce muscle-enriched preparations for bulk RNA sequencing (Fig. 2; Fig. S2). Differential expression analysis using the DESeq2 package revealed a total of 1417 differentially expressed genes with at least a twofold change in DmCFL KD compared with control: 558 genes with decreased and 850 genes with increased expression (Fig. 2A; Fig. S2C,D). Reduced DmCFL expression was further confirmed in DmCFL KD muscle (Fig. S2B). RNA sequencing identified changes in genes related to actin binding or regulation. Over-representation analysis using the Gene Ontology (GO) Biological Processes gene sets showed that protein degradation processes were significantly reduced (Fig. 2B, Table S1; Ashburner et al., 2000; Gene Ontology Consortium et al., 2023). This finding was consistent with our previous work in which increasing proteasomal activity in DmCFL KD larvae improved the muscle deterioration phenotype (Balakrishnan et al., 2020). Prominently, upregulated pathways were related to synaptic signaling (Fig. 2C, Table S1), as a result of increased expression of several genes acting at the NMJ (Fig. 2D). These gene products have been previously reported to be functional on the presynaptic (i.e. motor neuron) and the postsynaptic (i.e. muscle) sides of the NMJ (Banerjee et al., 2017; Bergquist et al., 2010; Broadie et al., 1995; Chen et al., 2010; Daniels et al., 2004; Featherstone et al., 2000; Ganesan et al., 2011; Ganetzky, 1984; Ganetzky and Wu, 1983; Hewes et al., 1998; Humphreys et al., 1996; Kambadur et al., 1998; Kaplan and Trout, 1969; Liebl et al., 2008; Li et al., 2007; Megighian et al., 2010; Nakayama et al., 2014; Romero-Pozuelo et al., 2007; Roos et al., 2000; Simon et al., 2009; Suzuki, 2006; Torroja et al., 1999; Wagh et al., 2006).

Given these identified transcriptomic changes related to motor neuron–muscle communication at the NMJ, we next examined DmCFL and actin cytoskeleton in the postsynaptic domain.

The canonical structures investigated here are shown in a schematic of the Drosophila larval NMJ (Fig. 3A). The specialized postsynaptic muscle membrane, known as the subsynaptic reticulum (SSR), can be visualized using an antibody against the Discs-large (Dlg; Dlg1) scaffold protein (Lahey et al., 1994). The motor neuron presynaptic membrane can be visualized using anti-horseradish peroxidase (HRP; Jan and Jan, 1982). The SSR apposes the motor axonal endings (boutons), with distinctive Dlg rings around HRP signal (Fig. 3B). To test the presence of DmCFL at the postsynapse, both live and fixed imaging approaches were used. For the following experiments, we focused our analyses on muscle pairs that had at least one class 1 or 2 muscle, because the postsynapse could not be reliably quantified in muscle pairs with both class 3 muscles owing to their severe muscle deterioration.

Live imaging in DmCFL::GFP larvae showed that the protein is enriched at the muscle SSR in a pattern consistent with the ring patterns of the postsynaptic density (Fig. 3C). In fixed samples, immunofluorescence and confocal microscopy showed that DmCFL is present at the postsynapse (Fig. 3D). DmCFL activity is tightly regulated in cells, where it is inactivated by phosphorylation (Niwa et al., 2002). Inactive phosphorylated cofilin (p-DmCFL) was similarly present at the postsynaptic domain (Fig. 3E). SIM imaging confirmed that DmCFL is found at the postsynapse in opposition to the boutons and that this local postsynaptic DmCFL is reduced in DmCFL KD muscles (Fig. 3F).

Given that the antibody against DmCFL also recognized DmCFL in the motor neuron, we sought to restrict our analysis to the postsynaptic domain while also minimizing signal from the underlying sarcomeres (approach described in detail in Fig. S3). We found that postsynaptic total DmCFL and p-DmCFL were both reduced in KD muscles: 69.2% and 48.4% of control levels, respectively (Fig. 3G,H).

Together, these data suggest that DmCFL and inactive p-DmCFL are enriched at the postsynaptic domain, and both are reduced in this region in the DmCFL KD model.

Regulation of F-actin by actin-binding proteins is important at the SSR membrane for proper cytoskeletal organization (Blunk et al., 2014; Pielage et al., 2006; Wang et al., 2010). In the DmCFL KD model, we previously described aberrant F-actin accumulation at the fiber poles in class 2 and actin aggregates in class 3 muscles (Balakrishnan et al., 2020). Similarly, we now show that F-actin accumulates in the postsynaptic region in class 2 and 3 DmCFL KD muscles (Fig. 4A). In three-dimensions, the actin cytoskeleton formed a halo that enveloped the HRP-positive synaptic bouton, even if there were intact sarcomeres present deeper in a class 2 muscle (Movie 1). These data indicate that the F-actin network at the postsynaptic domain was affected prior to the complete muscle sarcomere deterioration seen in class 3 muscles. Postsynaptic F-actin accumulation occurred in both class 2 and 3 muscles independently of the class of its adjacent muscle in the VL3/4 muscle pair, suggesting that the effect is autonomous to muscles that have transitioned to a progressively deteriorated state. For example, in a pair in which VL3 is class 1 and VL4 is class 3, only the class 3 muscle had aberrant postsynaptic F-actin accumulation. Quantitatively, the level of postsynaptic actin was increased in all DmCFL KD muscle classes, but especially in class 2 and 3 muscles (Fig. 4B). Therefore, the F-actin cytoskeleton at the muscle postsynapse is initially impacted and progressively deteriorates, similar to the deterioration seen in the sarcomeres in DmCFL KD muscles.

Transmission electron microscopy (TEM) shows areas of cytoplasmic F-actin disorganization in DmCFL KD muscle (Balakrishnan et al., 2020). In this study, we used TEM to examine the ultrastructure of the postsynaptic domain. Samples were dissected and sectioned longitudinally to reveal an orthogonal view of the neuronal boutons and underlying muscle, including the postsynaptic actin network and sarcomeres (sagittal sectioning approach shown in Fig. S4). In control muscles, we observed actin filaments below the bouton organized parallel to each other (Fig. 4C). The postsynaptic actin organization below the bouton was lost in DmCFL KD. In addition, we found that there is a deformation of the synaptic cleft, the space between the bouton and the SSR membrane, in the DmCFL KD muscle. Importantly, the actin-capping protein tropomodulin (Tmod) was clearly present in the control, but it was greatly reduced in all DmCFL KD classes and became disorganized in classes 2 and 3 (Fig. 5A). α-Spectrin is also present at the postsynapse and plays a role in the membrane's integrity (Pielage et al., 2006). Similar to Tmod, levels of α-Spectrin were visibly reduced surrounding the boutons in DmCFL KD muscle compared with control (Fig. 5B). Thus, the actin-binding proteins Tmod and α-Spectrin did not accumulate at the postsynapse in DmCFL KD muscles, despite the observed actin accumulation in the postsynaptic domain.

As visualized by confocal microscopy and TEM, postsynaptic actin accumulates and is disorganized in DmCFL KD deteriorated muscles, surrounding the boutons in class 2 and 3 but not in class 1 muscles.

Genetic manipulations of other muscle actin-binding and regulatory proteins affect NMJ development leading to morphology abnormalities, including reduced bouton number and Dlg disorganization (Lee and Schwarz, 2016; Pielage et al., 2006; Xing et al., 2018). Thus, we sought to characterize NMJ morphology in DmCFL KD muscles. The motor neuron properly innervated all DmCFL KD groups by the wandering third-instar stage (Fig. 6A). The NMJ span along the muscle cell was unchanged (Fig. 6B). We did not observe any changes in the number of large type Ib boutons in DmCFL KD muscles (Fig. 6C). These boutons are prominent at the VL3/4 NMJ and provide the neurotransmission for acute muscle contraction (Atwood et al., 1993). Similarly, there was no difference in the presence of branches of small type Is bouton pattern, which also synapse onto VL3/4 muscles and are stained by anti-HRP.

To evaluate the postsynaptic SSR, we measured SSR area by taking the ratio of Dlg-positive area to the combined cell area of the muscle pair. Area quantification was carried out for group 1 or 2 VL3/4 muscle pairs, as cell area is difficult to quantify reliably when muscle cell integrity is severely compromised as in group 3. We did not find a significant difference in SSR coverage when comparing the DmCFL KD with control (Fig. 6D). Morphologically, however, we detected differences in the postsynaptic organization: despite Dlg being properly organized around the boutons in class 1 KD muscles, it became increasingly disorganized in class 2 and 3 muscles, showing a more diffuse pattern (Fig. 6E). The disorganization in class 2 muscles was not enough to alter the mean SSR coverage compared with control. Quantification of Dlg and HRP intensity levels across the diameter of single boutons in the control revealed a pattern whereby the most intense Dlg signal peaks were found on either side of the peak HRP bouton signal. In DmCFL KD, class 1 muscles showed a similar pattern to control, whereas class 3 muscles displayed a plateau in Dlg intensity across the diameter irrespective of the HRP peak (Fig. 6F). The range of disruption in Dlg organization over the three classes suggests that postsynaptic structural integrity deteriorates progressively, similar to defects seen in postsynaptic F-actin and the sarcomeres in the DmCFL KD model.

Together, these data indicate normal presynaptic innervation when DmCFL is reduced in the muscle, whereas the postsynaptic architecture became progressively disorganized as the muscle deteriorates over time in the DmCFL KD model.

Excitation-contraction coupling is the process by which NMJ synaptic signaling from the motor neuron bouton is communicated to the muscle to drive contraction. In Drosophila, this neurotransmission relies on glutamatergic signaling, whereas in vertebrates excitation occurs via acetylcholine. At the presynaptic active zone, a consolidation of proteins, including the core active zone scaffold Bruchpilot (Brp), orchestrate synaptic vesicle docking and the fusion release of glutamate (Wagh et al., 2006). In the postsynaptic SSR membrane, the excitatory neurotransmitter binds to glutamate receptors (GluRs). These receptors are heterotetramers, with essential subunits GluRIIC, D and E, and an optional fourth subunit of either GluRIIA or GluRIIB (DiAntonio, 2006). To test whether changes in NMJ signaling machinery occur in DmCFL KD muscles, we first analyzed Brp and the essential GluRIIC subunit.

Although our RNA sequencing results indicated an over threefold increase in brp gene expression (Fig. 2D), we found that Brp protein in NMJ presynaptic boutons was not changed in DmCFL KD compared with control (Fig. S5A,B). Likewise, postsynaptic GluRIIC protein levels were not changed in DmCFL KD muscles, suggesting that there was no change in the total amount of glutamate receptors in the postsynaptic membrane (Fig. S5C). No differences were found in Brp-to-GluR apposition in DmCFL KD (Fig. S5A).

These results suggest that the DmCFL KD affects neither protein levels nor organization of presynaptic active zone Brp or the postsynaptic GluRs at the NMJ, despite highly increased RNA expression of NMJ genes being identified by transcriptomic analysis. Given that there is diversity in GluR subunit composition that impacts GluR channel properties, we pursued further NMJ analyses in DmCFL KD muscle.

A previous screen examining lethal mutants qualitatively identified various cytoskeleton-related genes that specifically affect GluRIIA subunit levels detected by immunofluorescence (Liebl and Featherstone, 2005). GluRIIA-containing receptors in the postsynaptic domain are selectively regulated by Coracle via the cortical actin cytoskeleton (Chen et al., 2005; Song et al., 2022). Moreover, actin monomer-binding protein twinfilin mutants have decreased abundance of GluRIIA-containing receptors at the NMJ (Wang et al., 2010). Given that DmCFL is an actin-binding and severing protein, we hypothesized that muscle DmCFL KD would selectively affect GluRIIA receptor levels.

To test this hypothesis, we quantified GluRIIA using immunofluorescence and found there was less GluRIIA in DmCFL KD muscles compared with control (66%; Fig. 7A). Stratifying the data by muscle pair class combination showed that GluRIIA is only decreased in group 2 or 3 muscle (Fig. 7B). SIM imaging revealed that glutamate receptors containing GluRIIA were properly located at the NMJ postsynaptic domain in class 1 muscles, but the field of these receptors was expanded and mislocalized at some postsynaptic densities in class 2 and class 3 muscles (Fig. 7C). GluRIIA, unlike other GluR subunits, is anchored to the actin cytoskeleton by Coracle (Chen et al., 2005). Consistent with this Coracle regulatory mechanism, we found that DmCFL KD muscles have reduced Coracle in the postsynaptic domain, suggesting a possible mechanism for altered GluRIIA localization at the NMJs of DmCFL KD muscles (Fig. 7D).

GluRIIA-containing receptors mediate the strength of the muscle postsynaptic response, with reduced GluRIIA selectively impairing NMJ neurotransmission strength (DiAntonio et al., 1999; Petersen et al., 1997). To examine neurotransmission in DmCFL KD muscle, we employed two-electrode voltage-clamp (TEVC) recordings to measure glutamate release and GluR activation (Leahy et al., 2023). The motor neuron was stimulated at suprathreshold voltages with a glass suction electrode at 0.2 Hz, with ten consecutive evoked excitatory junction current (EJC) traces recorded (Fig. 7E) and then averaged to calculate the mean neurotransmission strength in the DmCFL KD compared with control (Fig. 7F). To get a varied sample of the electrophysiological profile of DmCFL KD muscles, a mix of the different muscle class combinations were tested by TEVC in a recording configuration in which the operator was unaware of the muscle classes being tested. Compared with the control, the average overall EJC amplitude of DmCFL KD muscles was significantly reduced (∼60% reduced; Fig. 7E,F). Interestingly, all DmCFL KD muscles had a decreased EJC amplitude and, therefore, similarly reduced NMJ neurotransmission strength. This indicates that NMJ function was impaired when muscle DmCFL levels are reduced.

These findings indicate that DmCFL KD muscles have decreased levels of GluRIIA-containing receptors in the NMJ postsynaptic domain, with mislocalization of the receptors into extra-synaptic muscle regions and a concomitant reduction in NMJ synaptic transmission amplitude shown by direct voltage-clamp electrophysiology recording.

Cofilin 2, an actin-severing protein, has been linked to NM. Although NM is often characterized as a disease of the sarcomere, other actin-dependent structures and functions in the muscle have not been investigated. Consistent with its known role in human NM, we have shown that DmCFL plays a crucial role in sarcomerogenesis and muscle maintenance (Balakrishnan et al., 2020). Here, we investigated an additional role of CFL in regulating actin at the NMJ. We show that DmCFL is located at the postsynaptic domain and that loss of DmCFL in muscle results in buildup of postsynaptic F-actin, disorganization of SSR membrane proteins, and mislocalization of the key GluRIIA glutamate receptor subunit. Consistent with these findings, we see reduced NMJ neurotransmission by electrophysiology. Interestingly, these defects occur prior to the onset of advanced sarcomeric decline. We propose that DmCFL is required for proper postsynaptic actin dynamics, and, in its absence, the SSR membrane and its components become progressively disorganized.

As a first step to study the role of DmCFL at the Drosophila NMJ and to determine its postsynaptic localization and reduction in DmCFL KD muscles, we visualized DmCFL in living and fixed tissues using different approaches. DmCFL is found at the postsynapse, which can only be appreciated by examining thin sections and suggests that its localization is fairly restricted in space. We used an innovative three-dimensional approach to quantify proteins in the postsynaptic domain to distinguish from the high levels of DmCFL in the underlying sarcomere. This approach could be applied to other studies of the postsynapse. Postsynaptic quantification, in addition to western blotting of overall muscle levels, revealed that DmCFL is reduced in DmCFL KD. Similarly, phosphorylated cofilin 2 is absent in the muscles of an individual with NM harboring a CFL2 variant (Agrawal et al., 2007).

In humans, mice and Drosophila with muscle cofilin defects, muscle structure progressively deteriorates and function declines. Progressive muscle weakness is reported in CFL2 NM cases with some individuals showing a deterioration after early childhood and others having severe symptoms present at birth (Agrawal et al., 2007; Fattori et al., 2018; Ockeloen et al., 2012; Ong et al., 2014). Cfl2 mouse models also show a progressive reduction in muscle function and an increase in nemaline rods seen on histology (Agrawal et al., 2012; Gurniak et al., 2014; Mohri et al., 2019; Rosen et al., 2020). Even in chimeric mice in which only some muscle fibers harbor Cfl2 mutations, the mutant fibers deteriorate, indicating that the health of surrounding cells does not impact muscle disease progression (Mohri et al., 2019). In Drosophila, DmCFL is knocked down in all larval muscles, yet the progression of the deterioration phenotype does not occur in a coordinated fashion across all muscles. We show that a single pair of ventral longitudinal muscles (VL3 and 4), which are innervated by the same branch of intersegmental nerve B, do not necessarily exhibit the same extent of muscle deterioration by the end of larval development, as evidenced by the existence of various groups of phenotypic classes. This finding mirrors the progressive deterioration seen in the chimeric Cfl2 mice, as cofilin 2 is only affected in select muscles rather than in all tissues like in other Cfl2 mouse models.

We found that postsynaptic actin and SSR membrane organization deteriorate progressively in DmCFL KD muscles. In vertebrates, actin isoforms have been defined based on their expression: skeletal muscle (α_sk and α_ca), smooth muscle (α_sm and γ_sm) and cytoplasmic (β_cyto and γ_cyto) (reviewed by Perrin and Ervasti, 2010; Lubit and Schwartz, 1980). In Drosophila, the different isoforms are expressed during particular developmental stages and in specific tissues (Fyrberg et al., 1981, 1983; Röper et al., 2005; Wagner et al., 2002). Although these actins are typically categorized as cytoplasmic or muscle actins, there is evidence that all of the isoforms are present at the sarcomere of larval muscles (Röper et al., 2005). Röper and colleagues speculate that the formation and localization of particular actin structures within the cell is driven by actin-binding proteins with different affinities (Röper et al., 2005). Despite the vertebrate sarcomeric α-actin isoforms being predominant in vertebrate skeletal muscle, the non-sarcomeric γ-actin networks also play myriad roles in the muscle cell. This isoform is needed for anchoring myofibrils to the muscle membrane, localizing organelles such as mitochondria and sarcoplasmic reticulum, and forming the sarcomere Z-discs (Craig and Pardo, 1983; Gokhin and Fowler, 2011; Nakata et al., 2001; Papponen et al., 2009; Pardo et al., 1983; Rybakova et al., 2000). Additionally, γ_cyto is needed for proper maintenance of the cytoskeleton in muscle but not its development; nevertheless, its absence does lead to a phenotype similar to centronuclear myopathy (Belyantseva et al., 2009; Sonnemann et al., 2006). Given that there are different pools of actin in the muscle cell, cofilin may play a role at the sarcomere engaging with α-actin and at other locales, such as the NMJ, by regulating γ_cyto actin. The fact that actin is required in the muscle for more than sarcomere structure invites further inquiry into how effects on non-sarcomeric actin contributes to NM and defects at the postsynapse.

Importantly, there is a greater increase in F-actin than G-actin seen in the muscles of a Cfl2 knockout mouse model (Agrawal et al., 2012). The postsynaptic actin accumulations in DmCFL KD muscle appear as filamentous swirls surrounding the motor neuron boutons, similar to the structures reported in the act^E84K mutant, in which the actin isoform Act57B is affected, and in the twf¹¹⁰ mutant, in which the actin-binding protein Twinfilin is mutated (Blunk et al., 2014; Wang et al., 2010). We found defects in F-actin at the NMJ of class 2 muscles occurred before the sarcomeric deterioration deeper in the muscle, indicating that it is unlikely that sarcomeric actin is the only source of actin accumulating at the NMJ. F-actin accumulations surround only some boutons in class 2 muscles, whereas all boutons are surrounded by actin in class 3 muscles, confirming that the deterioration is progressive. It is possible that the postsynaptic deterioration may also be progressive in humans with NM, and that accumulations may not be as obvious as the rods seen emanating from the sarcomere because they do not include Z-disc proteins often examined in samples from individuals with NM. Even using TEM, we found disorganized actin filaments in the postsynaptic region of DmCFL KD muscle, which is reminiscent of the cytoplasmic actin disorganization shown in the biopsy from an individual with CFL2 NM (Fattori et al., 2018). We find that actin-binding proteins Tmod and α-Spectrin are reduced at the DmCFL KD postsynaptic domain. This reduction may be from altered structural organization owing to actin accumulation or altered actin dynamics, making access for these actin-binding proteins more difficult. The reduction of Tmod may be due to sequestering of Tmod at the cell poles, as we have previously reported (Balakrishnan et al., 2020).

Several Drosophila models have shown that the actin cytoskeleton is important for proper SSR formation (Pielage et al., 2006; Wang et al., 2011). Although the postsynaptic membrane forms properly initially, the expansion of Dlg at the postsynapse in DmCFL KD class 3 muscles suggests that actin dynamics, regulated by DmCFL severing, are important in maintaining the SSR. We did not see an increase in postsynaptic SSR area in group 1 and group 2 pairs likely because only some boutons in class 2 muscles have affected Dlg organization. Dlg is thought to not interact directly with actin, instead being part of a complex with other actin-binding proteins, including adducin/Hts and Spectrin (Wang et al., 2014). Thus, Dlg disorganization at the DmCFL KD postsynapse in the most deteriorated muscles is likely due to an effect on the overall complex, rather than the actin accumulation affecting Dlg directly. Studies of Cfl2 knockout and mutant mice report no obvious NMJ defects; it is possible that a subtle change on the postsynaptic membrane would not be appreciated when only visualizing the motor neuron and acetylcholine receptors (Agrawal et al., 2012; Gurniak et al., 2014). Nevertheless, some studies of biopsies of humans with NM imaged by TEM report alterations at the postsynapse, including collapsed or dilated primary and secondary NMJ synaptic clefts (Fukuhara et al., 1978; Heffernan et al., 1968; Karpati et al., 1971). We see a collapse of the synaptic cleft by TEM in the DmCFL KD model. Together, these data suggest that maintaining a certain amount and/or dynamic state of postsynaptic actin is important for retaining postsynaptic SSR structure, and, thus, the increase in actin due to cofilin reduction in the Drosophila model leads to progressive structural deterioration.

Cofilin and the actin cytoskeleton are implicated in the trafficking of ionotropic AMPA receptors to dendritic membranes in vertebrates. In fact, it has been suggested that ADF/cofilin activity is needed both for clearing actin that may be in the way of AMPA receptor insertion and for regulating the new microfilaments that guide the receptors to the membrane (Gu et al., 2010; Zhou et al., 2001). Experimentally, increasing ADF/cofilin in hippocampal neuron culture led to increased addition of AMPA receptors to dendritic spines, whereas a decrease resulted in their removal (Gu et al., 2010). Work in a Xenopus culture model found that actin is more dynamic near membrane acetylcholine receptors compared with the actin within the myofibrils, and that a balance of active and inactive cofilin near the membrane is required for their proper addition (Lee et al., 2009). Pharmacological stabilization of actin at the Drosophila NMJ reduces GluRIIA cluster size (Chen et al., 2005). From the current studies in mouse, it is unclear whether there is a reduction in acetylcholine receptor levels in Cfl2 knockout mice. Our DmCFL KD model indicates no reduction in the levels of total glutamate receptors as indicated by the GluRIIC subunit present in all receptors, but selectively reduced GluRIIA. Using SIM, we show that, in class 2 and 3 muscles, GluRIIA-containing receptors are being pulled away from or are not properly delivered to the postsynaptic density directly opposing the motor neuron boutons. Coracle links the GluRIIA subunit to the actin cytoskeleton, and its loss reduces levels of GluRIIA (Chen et al., 2005). We found there are reduced Coracle levels at the DmCFL KD postsynapse, which is consistent with the reduction in GluRIIA. A similar decrease in GluRIIA and Coracle levels is seen in twf mutants (Wang et al., 2010). Live-imaging experiments have revealed that glutamate receptors enter the postsynaptic density from pools around the plasma membrane; it is possible that without a dynamic actin cytoskeleton, GluRIIA receptors are not transported efficiently to the postsynapse (Rasse et al., 2005).

Our electrophysiological profile of DmCFL KD muscles is consistent with decreased levels of GluRIIA-containing receptors at the postsynapse. DmCFL KD muscles exhibit reduced neurotransmission strength based on TEVC recordings of evoked postsynaptic currents following motor nerve stimulation. Consistent with this, electromyogram (EMG) studies of individuals with NM often reveal a myopathic pattern with low amplitudes. Moreover, one longitudinal study of 13 individuals with NM found that after age 9 EMG studies of distal muscles began to show neuropathic changes in addition to myopathic changes (Wallgren-Pettersson et al., 1989). This suggests that as the disease progresses, there are degenerative changes of the motor units with some individuals reported to lack motor units; it is also possible that reinnervation by adjacent motor units occurs to result in higher amplitude potentials. One case report speculated that there was a disruption in functional innervation that could contribute to the formation of rods in the extrafusal muscle fibers (Karpati et al., 1971). Of note, we did not discover any defect in motor axon pathfinding to VL3/4 muscle targets or any change on the presynaptic active zones. However, the upregulation in presynaptic genes identified by RNA sequencing may suggest that an attempt by the motor neuron to compensate for impaired NMJ synaptic function could occur during late-stage progression in the DmCFL muscle KD model. We are unable to investigate further disease model progression because the third-instar larvae begin the process of pupation commencing with the wandering stage.

Although our study focuses on a cofilin NM model, case reports suggest that NMJ defects may be found in NM more broadly. Future studies should examine NM biopsy and electrophysiological results with more consideration for the specific gene affected, which was not possible in previous decades. The DmCFL KD model should encourage further study into the role of the NMJ in NM disease progression and potential therapeutics (Fisher et al., 2022). There are reports from some individuals with NM that use of acetylcholinesterase inhibitors can lead to clinical improvement (Natera-de Benito et al., 2016). One possibility for the improvement is that increased presence of acetylcholine in the synaptic cleft increases the probability that the neurotransmitter will bind receptors along the simplified postsynaptic membrane, thereby increasing NMJ synaptic transmission. New acetylcholinesterase inhibitors, such as C-547, that are more specific for the NMJ region show promise as they would limit off-target side effects of treatment (Petrov et al., 2018). Further study is needed into which pharmacological treatments may work for NM. The fact that the Drosophila NMJ is a glutamatergic synapse poses a challenge in testing potential treatments that impact acetylcholine; however, Drosophila models allow a simple system in which to interrogate further the basic underpinnings of disease mechanism. Our findings suggest that NMJ defects occur when muscle cofilin is reduced, which brings forth the alternative that druggable targets of actin-binding proteins could be screened in the Drosophila disease model. In addition, we observed NMJ structural deterioration in the most affected muscles, and thus EMG in tandem with nerve conduction studies and biopsies from older individuals with CFL2 NM would be informative about disease progression and health of the NMJ.

In conclusion, this work reveals defects at the NMJ, specifically in the postsynaptic domain, as being a part of the deterioration that results from dysregulated muscle cofilin levels. We propose that altered non-sarcomeric actin dynamics may affect other muscle functions that then contribute to the progression of muscle deterioration seen in NM, opening new avenues for understanding and ameliorating the condition.

All stocks and crosses were raised on standard cornmeal at 25°C on a 12-h light/12 h dark light cycle under humidity control. All experiments were carried out in wandering L3 larvae of both sexes reared at 25°C. DmCFL was knocked down specifically in muscle, as in our previous study (Balakrishnan et al., 2020), by leveraging the Gal4-UAS system (Brand and Perrimon, 1993). The muscle-specific driver Mhc-Gal4 (Bloomington Drosophila Stock Center, #67044) was used to drive UAS-mCherry RNAi (for control; Bloomington Drosophila Stock Center, #35785) or UAS-tsr TRiP RNAi (for DmCFL knockdown; Bloomington Drosophila Stock Center, #65055) generated by the Transgenic RNAi Project (TRiP) (Perkins et al., 2015). Live-imaging experiments were performed with larvae expressing tsr::GFP (ZCL2393; Kyoto Stock Center DGRC, #110875, referred to in the text as DmCFL::GFP) generated by the FlyTrap: GFP Protein Trap Database (Morin et al., 2001).

RNA sequencing of control and DmCFL KD larvae was performed as described previously (Zapater I Morales et al., 2023). Eight to ten late third-instar larvae of each genotype were dissected in triplicate.

Read counts data were assessed and plotted with the integrated Differential Expression and Pathway analysis (iDEP) web application (versions 0.96 and 1.1) (Ge et al., 2018). DESeq2 analysis was performed with the following parameters: false discovery rate of 0.05 and minimum twofold change (Love et al., 2014). Over-representation analysis was performed using the Gene Set Enrichment Analysis (GSEA) method to identify the top 20 enriched pathways defined by GO Biological Processes gene sets when considering the top 2000 genes at a false discovery rate of 0.05.

Five to ten wandering late third-instar larvae were dissected in HL3.1 buffer as has been previously described (Brent et al., 2009) to produce muscle-enriched preparations, which were then lysed in larval lysis buffer (50 mM HEPES, pH 7.5, 150 mM NaCl, 0.5% NP40, 0.1% SDS) supplemented with cOmplete mini protease inhibitor cocktail (Roche, 11836153001) and PhosSTOP phosphatase inhibitor cocktail (Roche, 4906837001). Ten micrograms from control and DmCFL KD lysates were run on a 12.5% polyacrylamide gel, then transferred to a nitrocellulose membrane (Thermo Fisher Scientific, PI88018). Blocking was carried out in 5% milk or 5% bovine serum albumin (BSA) in TBST (Tris-buffered saline+0.1% Tween 20) for 1 h at room temperature. Primary antibody staining was performed in Stamina Antibody Dilution Buffer (Kindle Biosciences, R2004) overnight at room temperature using rabbit anti-Twinstar (DmCFL; gift from Tadashi Uemura, Kyoto University, Kyoto, Japan; 1:1000); rabbit anti-phospho-Twinstar (p-DmCFL; gift from Tadashi Uemura; 1:1000); and mouse anti-β-Actin (Cell Signaling Technology, 4987; 1:1000). Secondary antibody incubation was performed in 5% milk for 1 h at room temperature using anti-rabbit-HRP (Jackson ImmunoResearch, 711-035-152; 1:5000) or anti-mouse-HRP (Jackson ImmunoResearch, 715-035-151; 1:5000). The blot was imaged using a KwikQuant Imager (Kindle Biosciences, D1001) using 1-Shot Digital-ECL (Kindle Biosciences, R1003) and intensities were quantified using the Fiji Gel Analyzer tool (NIH). Protein expression was normalized to β-Actin loading control within each sample and repeated in triplicate.

Total RNA was extracted from ten late third-instar wandering larvae muscle-enriched preparations (dissected as described in for western blotting) using TRIzol reagent (Thermo Fisher Scientific, 15596026) and was subsequently cleaned using the TURBO DNA-free Kit (Ambion, AM1907). Reverse transcription was carried out to synthesize cDNA using the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen, 18080-051) kit. PCR reactions were run on a CFX Opus 96 Real-Time PCR System using SYBR Select Master Mix for CFX (Applied Biosystems, 4472937) in biological and technical triplicate. Primers used for tsr analyses: forward GCTCTCAAGAAGTCGCTCGT; reverse GCAATGCACAGTGCTCGTAC. The ΔΔCt method (Livak and Schmittgen, 2001) was used to calculate fold changes with RpL32 as the normalization control. Reported values represent the log₂ fold change of the gene in DmCFL KD compared with control samples.

Third-instar larvae at the wandering stage were dissected as previously described (Azevedo et al., 2016; Brent et al., 2009) to expose the body wall muscles. Fixation was carried out using 4% paraformaldehyde in HL3.1 buffer for 15 min at room temperature for all antibodies except GluRIIA, for which Bouin's fixative was used for 5 min at room temperature. Samples were blocked in BSA-PBT (PBS supplemented with 0.1% BSA and 0.3% Triton X-100) for 30 min at room temperature, then incubated with primary antibody overnight at 4°C and subsequently washed in BSA-PBT. Samples were then incubated with Alexa Fluor-conjugated secondary antibodies, phalloidin and goat Alexa-647 conjugated HRP (Jackson ImmunoResearch, 123-605-021) at a concentration of 1:400. Alexa Fluor 555 conjugated secondary antibodies were used for all intensity quantifications. Final washes were carried out in PBT prior to mounting in Prolong Gold (Invitrogen, P36930). All slides were cured for at least 24 h at room temperature prior to imaging.

Primary antibodies used in this study include rabbit anti-Tsr (DmCFL; 1:500, gift from Tadashi Uemura; Niwa et al., 2002) against the C-terminal peptide CREAVEEKLRATDRQ; rabbit anti-p-Cofilin (1:500; gift from Tadashi Uemura) against an N-terminal peptide [acetyl-A(pS) GVTVSDC]; mouse anti-Discs large [1:100; Developmental Studies Hybridoma Bank (DSHB), 4F3]; mouse anti-Bruchpilot (1:250; DSHB, nc82); rabbit anti-GluRIIC (1:1000; gift from Aaron DiAntonio, Washington University, St. Louis, MO, USA); mouse anti-GluRIIA (1:100; DSHB, 8B4D2, MH2B); rat anti-Tmod (1:200; gift from Velia Fowler, University of Delaware, Newark, DE, USA); and guinea pig anti-Coracle (1:1500; gift from Richard Fehon, University of Chicago, Chicago, IL, USA).

All samples for comparison were imaged with the same settings between genotypes. Pairs of ventral longitudinal muscles 3 and 4 (muscles 6 and 7) from abdominal hemisegments 2-4 were imaged for all experiments. For DmCFL KD, muscle pairs were only imaged if one muscle was class 1 or 2. z-stack images were acquired using an upright Leica Stellaris 5 laser-scanning confocal microscope with dry HC PL Apo 20×/0.75 CS2, oil HC PL Apo 63×/1.40 CS2, or oil HC PL Apo 100×/1.40 CS2 objectives and HyD S detector (Leica Microsystems). Images were acquired sequentially by stack scanning bidirectionally at 400 Hz, with a pixel size of 283.95 nm×283.95 nm and an area size of 2048×1024 pixels in Leica LASX software and saved as LIF files. For all images, the pinhole size was 92.53 µm, calculated at 1 A.U. for 561 nm emission. Images were acquired with step sizes of 1 μm (for 20×) or 0.5 μm (for 63× or 100×). For all images for postsynaptic intensity, z-stacks were generated using the HRP channel with the start before and the end 1 μm below the last appreciable HRP signal. Fiji (NIH) was used to create sum slices or maximum intensity z-projections.

Using Fiji, a standard region of interest was drawn in a region of the muscle cytoplasm. The first slice was defined as the slice in the z-stack in which >50% of the frame contained F-actin-positive signal. Fluorescence intensity of the DmCFL channel within the region of interest was recorded over ten frames starting with the first slice; the intensity measurements for each slice were summed for each sample.

Live samples were imaged using a Leica Stellaris 5 laser-scanning confocal microscope with a water HC FLUOTAR L VISIR 25×/0.95 objective and HyD S detector (Leica Microsystems). Larvae were dissected and pinned to expose ventral muscles and imaged live at ambient temperature while the larvae were maintained in ice-cold HL3.1 buffer. Images were acquired sequentially by stack scanning bidirectionally at 400 Hz, with a pixel size of 283.95 nm×283.95 nm and an area size of 2048×1024 pixels in Leica LASX software and saved as LIF files. For all images, the pinhole size was 92.53 µm, calculated at 1 A.U. for 561 nm emission. z-stacks were taken using a 1 μm step size and sum slices z-projections were created using Fiji.

All samples were imaged with the same settings on a Zeiss Elyra 7 with Lattice SIM² confocal microscope with a Plan-Apochromat 63×/1.4 Oil DIC M27 objective. Images were acquired sequentially by stack with a pixel size of 0.06 µm×0.06 µm and an area size of 80.1 µm by 80.1 µm in ZEN Black software and saved as CZI files. Images were taken with 13 phases and reconstructed using ZEN Black software at the ‘strong’ sharpness setting (Zeiss). The z-stacks were acquired at a step size of 0.329 μm and sum slices z-projections were made using Fiji.

Postsynaptic intensity measurements were carried out using Imaris 10.0 (Bitplane). A three-dimensional surface for the HRP source channel was generated for each z-stack image with a surface detail grain level of 0.18 μm, smoothing enabled, and auto-thresholding (Fig. S3A,B). Small surfaces that were not part of the NMJ were removed. For Brp quantification, the sum of the sum intensity of the Brp channel was normalized to the total HRP volume and compared between genotypes. For DmCFL, p-DmCFL, GluRIIC and GluRIIA intensity measurements, an expanded volume was used. A mask was created from the HRP surface using the ‘Distance Transform’ setting. A second surface was made using the distance transform mask, with a surface detail level of 0.18 and a manual threshold of 0-0.5 to limit the surface to a shell extending from the edge of the HRP surface to 0.5 μm away (Fig. S3C-F). The sum of the sum intensity within this expanded surface was normalized to the expanded volume. For the postsynaptic actin intensity measurements, a manual region of interest was outlined corresponding to the VL3 or VL4 half of the NMJ. The HRP volume was created only within this region of interest and then all subsequent analysis steps were performed in a consistent manner to the other postsynaptic intensity measurements.

The Dlg channel was used to quantify NMJ morphology in Fiji. The number of boutons in a z-stack image was manually counted using the ‘Cell Counter’ plugin. NMJ span was measured by drawing a line parallel to the length of the NMJ and normalizing to the length of muscle VL3 (6) or VL4 (7) (whichever was longer). Cell length was determined by drawing a polygon around the phalloidin-positive shape of each muscle cell and then using the ‘length’ measurement. SSR area was defined as Dlg-positive area normalized to total summed cell areas of muscles 6 and 7. Dlg-positive area was measured by creating a binary mask of the Dlg channel using the Yen thresholding method and recording the area. Cell area was determined by drawing a polygon around the phalloidin-positive shape of each muscle cell and then using the ‘length’ measurement. Dlg/HRP intensity levels surrounding individual boutons were measured by: (1) drawing a line along the diameter of a bouton; (2) recording both the Dlg and HRP intensities along the length of the line; (3) normalizing length values to the total length; and (4) normalizing the Dlg and HRP intensities to the average intensities of that marker for that bouton.

Wandering L3 larvae were dissected as described above and fixed overnight in 2.5% glutaraldehyde at 4°C. Larval filets were trimmed to remove the head and tail and to retain the ventral body wall muscles on both sides. Samples were washed three times in 0.1 M sodium cacodylate buffer and then post-fixed in 1% osmium tetroxide for 1 h at room temperature on a rotator. Next, samples were washed three times in 0.1 M sodium cacodylate buffer for 10 min each wash at room temperature. A dehydration series in ethanol was conducted for 10 min at room temperature in each of the following: 30, 50, 70, 85, 95, 100 and 100% ethanol. Infiltration was carried out in four steps: (1) incubation in a solution of 1:1 acetonitrile and 100% ethanol for 10 min at room temperature; (2) infiltration in only acetonitrile for 10 min at room temperature; (3) incubation in a 1:1 mixture of acetonitrile and Embed-812 resin for 30 min at room temperature; and (4) incubation in only Embed-812 resin overnight. In a semi-hardened resin block, larvae were oriented so that the longitudinal edge was along the cutting face of the block. The block was polymerized for 48 h at 60°C. Thick sagittal sections of 5-10 μm were taken and stained with Toluidine Blue (0.1% of 1% solution in sodium borate; Electron Microscopy Sciences) until muscles that were oriented in a longitudinal direction were identified (Fig. S4). A Leica Ultracut UCT ultramicrotome with diamond knife was used to take 70 nm ultrathin sections, which were then collected on 3 mm diameter mesh copper grids. Images of individual boutons and the underlying muscle were taken on a JEOL JEM-1400 transmission electron microscope at 100 kV.

TEVC recordings were made on dissected wandering third instars as previously reported (Leahy et al., 2023). Briefly, all recordings were carried out at 18°C in physiological saline (in mM): 128 NaCl, 2 KCl, 4 MgCl₂, 1.0 CaCl₂, 70 sucrose and 5 HEPES (pH 7.2). Longitudinally dissected larvae had internal organs removed and peripheral motor nerves cut at the ventral nerve cord base. The body walls were glued down (Vetbond, 3M). The preparation was imaged with a 40× water-immersion objective on a Zeiss Axioskop microscope. Ventral longitudinal muscle 6 in abdominal segments 3 or 4 was impaled with two intracellular electrodes (1 mm outer diameter borosilicate capillaries; World Precision Instruments, 1B100F-4) of ∼15 MΩ resistance (3 M KCl). The muscle was voltage clamped at −60 mV using an Axoclamp-2B amplifier (Axon Instruments), and the motor nerve stimulated with a fire-polished glass suction electrode using 0.5 ms suprathreshold voltage stimuli at 0.2 Hz from a Grass S88 stimulator. Nerve stimulation-evoked EJC recordings were filtered at 2 kHz. To quantify EJC amplitude, ten consecutive traces were averaged, and the average peak value recorded. Clampex 9.0 was used for all data acquisition, and Clampfit 10.7 was used for all data analyses (Axon Instruments).

Pairwise comparisons between groups were performed using a two-tailed Student's t-test with an alpha of 0.05 using R statistical software (R Core Team, 2021). Plots were generated using the ggplot2 R package (Wickham, 2016). Figures show the mean±s.d. in addition to sample size.

We thank the Baylies lab members, M. Balakrishnan, M. Lopez, P. Agrawal, A. Beggs and V. Gupta for helpful discussions, and T. Uemura, A. DiAntonio, V. Fowler, and R. Fehon for providing antibodies. We also thank the Weill Cornell Medicine Microscopy Core Facility for their assistance in the electron microscopy experiment and the Bioinformatics Cores at MSKCC for assistance in the RNA-sequencing experiment. We are grateful to the Bloomington Drosophila Stock Center and the Kyoto Drosophila Stock Center for genetic lines and the Developmental Studies Hybridoma Bank for antibodies. The National Institutes of Health provides funding for Weill Cornell–Sloan Kettering (T32HD060600) and the Weill Cornell–Rockefeller–Sloan Kettering Tri-Institutional MD-PhD Program (T32GM007739) and the Memorial Sloan Kettering Cancer Center receives funding from the National Cancer Institute (P30 CA 008748).