Nonmuscle myosin II regulates presynaptic actin and neuronal mechanobiology in Drosophila

Neuromuscular junctions (NMJs) and axons must weather continuous mechanical strain from muscle contractions, while maintaining contact with significantly more rigid muscle tissue (Tyler, 2012). The NMJ offers an excellent tissue mechanics model to study evolutionarily conserved mechanisms by which “soft” neurons sustain their shape, function, and plasticity while facing substantial mechanical challenges. NMJs are susceptible to aging and neurodegenerative disorders, and synaptic terminal retraction precedes axon degeneration and neuronal cell body loss in many diseases (Iyer et al., 2021). Therefore, understanding how mechanical forces are generated, sensed, and transmitted within neurons and between neurons and muscle is an important and understudied topic.

The actin cytoskeleton underlies multiple conserved mechanisms by which cells generate and respond to cellular forces (Michelot and Drubin, 2011). For example, endocytosis is supported by submicron-sized (∼200 nm), short-lived (<30 s) force-generating actin patches, containing Arp2/3-nucleated branched filamentous actin (F-actin) (Akamatsu et al., 2020; Del Signore et al., 2021). On the other hand, cell- and tissue-scale forces are often mediated by the actin-based motor nonmuscle myosin II (NMII) (Murrell et al., 2015). NMII is a hexamer composed of two heavy chains (NMII^HC), two essential light chains (NMII^ELC), and two regulatory light chains (NMII^LC), and exists in folded or extended conformations (Sellers and Heissler, 2019). The transition from a folded, autoinhibited state to an extended state is controlled via NMII^LC phosphorylation, which leads to self-assembly of contraction-competent NMII into bipolar filaments that bind antiparallel actin filaments (Sellers, 1991). In this way, NMII induces filament sliding, generating contractile forces for various cellular needs (Vicente-Manzanares et al., 2009). For example, ventral stress fibers are micron-sized, long-lived assemblies of linear F-actin that recruit NMII and play key roles in cellular mechanics and force sensing in cultured cells (Livne and Geiger, 2016). Thus, actin assemblies with variable sizes, lifetime, and molecular composition support different cellular processes.

Numerous actin and myosin structures in axons and presynaptic terminals could play important roles in the mechanical properties of neurons. The axonal submembranous cytoskeleton (SMC) is composed of periodic repeats of spectrin, actin rings, and NMII (Berger et al., 2018; Costa et al., 2020; Xu et al., 2013). This structure is conserved in mammalian and invertebrate neurons (He et al., 2016; Qu et al., 2017), and supports periodic organization of many molecules important for neurotransmission, cell adhesion, and molecular transport (Zhou et al., 2022). The SMC includes radial actomyosin, which regulates axonal diameter and conduction (Costa et al., 2020), and the passage of large cargos (Costa et al., 2020; Wang et al., 2020). Axons also feature a longitudinal NMII structure (Vassilopoulos et al., 2019), which may bridge neighboring actin rings, associate with other axonal linear actin assemblies such as trails (Ganguly et al., 2015) or F-actin bundles (Gallo, 2006; Micinski and Hotulainen, 2024), or have other functions. SMC periodicity diminishes at synaptic boutons, where actin forms mesh-like assemblies at active zones, rails between active zones and reserve pools, corrals around presynaptic compartments (Bingham et al., 2023), and transient puncta with features of endocytic sites (Del Signore et al., 2021). Whether these assemblies recruit NMII or contribute to the mechanical properties of presynaptic terminals remains unknown.

Functional studies implicate actomyosin structures in NMJ physiology. At the Drosophila larval NMJ, neuronal NMII depletion disrupts synaptic vesicle organization (Seabrooke et al., 2010), suggesting this motor associates with presynaptic actin structures. Both presynaptic NMII overexpression and knockdown decrease spontaneous vesicle release, suggesting that balanced NMII levels are necessary for presynaptic synaptic transmission (Seabrooke and Stewart, 2011). At Drosophila embryonic NMJs, changing mechanical tension via axotomy modulates vesicle organization and synaptic plasticity (Siechen et al., 2009). Moreover, neuronal NMII depletion at the larval NMJ leads to nascent bouton formation by blebbing (Fernandes et al., 2023). These nascent boutons depend on neuronal activity and muscle contraction and can mature into functional boutons (Ataman et al., 2008; Fernandes et al., 2023; Vasin et al., 2019), representing a form of activity-dependent structural plasticity. Thus, actomyosin contractility maintains bouton integrity during muscle contraction and regulates growth of new synaptic arbors. Taken together, these findings highlight the ability of the NMJ to respond to mechanical forces. However, the subcellular organization and functions of presynaptic actomyosin assemblies underlying these events remain largely unknown.

Actomyosin assemblies coordinate with other multi-protein machineries to control cellular mechanosensing and mechanotransduction. Ventral stress fibers are physically coupled to focal adhesion complexes (Livne and Geiger, 2016) along ventral membranes of adherent cells. Central to focal adhesion–mediated force-tuning are the metazoan integrin receptors, which bridge cytoplasmic cytoskeletal components to extracellular matrix (ECM) molecules and counter-receptors on other cells (Chastney et al., 2021). Via biochemical and mechanical activation, integrins enable cells to sense the mechanical features of their surroundings and transduce forces into biological signals by inducing focal adhesion assembly and maturation. Integrins play crucial roles in synaptic biology in vertebrates (Park and Goda, 2016) and at Drosophila NMJs, where their disruption (Beumer et al., 1999; Orr et al., 2022) or targeting their interactors (components of the ECM [Tsai et al., 2012; Wang et al., 2018], enzymes involved in ECM glycosylation [Dani et al., 2014], an integrin activator [Lee et al., 2017], or the cytoskeletal linker talin [Orr et al., 2022]), induces morphological defects and affects activity-dependent synaptic plasticity. These studies highlight the importance of integrins in neuronal function and position these molecules as candidates to control transsynaptic neuron-to-muscle communication and NMJ mechanobiology.

In this study, we identify a previously unrecognized linear actomyosin core at Drosophila larval NMJ presynaptic terminals that responds to changes in NMII levels and activity, as well as mechanical cues. This structure likely maintains mechanical continuity between neurons and muscle through integrin-based adhesions, providing a potential mechanism for transsynaptic mechanotransduction.

At the Drosophila larval NMJ, presynaptic terminals innervate mechanically active postsynaptic muscles, making this synapse an ideal model to study biological solutions to mechanical challenges at the neuron–muscle interface (Fig. 1 A). Previously, we demonstrated that a subpopulation of patch-like, submicron-sized, presynaptic actin is amenable to live confocal imaging and quantitative analysis (Del Signore et al., 2021). To improve visualization of these small and transient actin structures, we redesigned a broadly used, genetically encoded F-actin marker GMA (GFP-moesin actin-binding domain) (Bloor and Kiehart, 2001; Del Signore et al., 2021), by tagging it with the bright mNeonGreen fluorescent protein (mNGMA) (Shaner et al., 2013). To assess presynaptic actin structure and function, we generated QF-driven, QUAS-mNGMA (QmNGMA) transgenes. This allowed genetic combination of two binary expression systems (UAS/Gal4 and QUAS/QF) (Brand and Perrimon, 1993; Potter et al., 2010) in a single fly, to visualize presynaptic actin while independently manipulating actin-associated proteins via UAS-RNAi. Finally, to improve the resolution of the presynaptic actin structures of interest, we used Airyscan microscopy.

Using these tools, we imaged presynaptic actin in live dissected larvae with intact brains and axons, at rest. Interestingly, along with numerous spot-like F-actin assemblies, our improved signal to noise and resolution revealed a previously unrecognized linear, cable-like actin structure traversing the NMJ (Fig. 1 B and Video 1). To enable unbiased analysis, we built a WEKA machine learning–based classifier (Arganda-Carreras et al., 2017) in the image analysis software FIJI (Schindelin et al., 2012) that specifically segments round from linear F-actin structures at the NMJ (Fig. 1 B and Fig. S1). Next, we verified the existence of the linear actin core with the F-actin marker, Lifeact (Riedl et al., 2008), tagged with Halo self-labeling protein (UAS-Lifeact::Halo). Neuronally expressed Lifeact::Halo (C155-Gal4>Lifeact::Halo) labeled the linear actin core and the previously observed spot-like structures at the NMJ (Fig. 1 C). Thus, we confirmed that the newly identified linear actin assemblies can be visualized with at least two independent actin markers in live preparations. We used Lifeact and GMA interchangeably in our experiments, as both markers label comparable F-actin structures at the NMJ.

During live imaging, we noticed that the presynaptic actin core is stable (in space and time) over minutes (Fig. S2 A), unlike transient, spot-like structures, which have lifetimes in the tens of seconds and likely represent endocytic events (Del Signore et al., 2021) (Fig. S2 B). Fluorescence recovery after photobleaching (FRAP) of QmNGMA demonstrated immediate relabeling of the same, practically unchanged presynaptic actin core structure, supporting the idea that these are long-lived assemblies on which the F-actin marker can readily bind and unbind (Fig. S2, C and D). To explore actin dynamics directly, we generated a QUAS transgene carrying the Actin5C Drosophila isoform, tagged with the bright red fluorescent protein mScarlet-I (QAct5C::mScarI) (Bindels et al., 2017). Live imaging of QAct5C::mScarI showed that tagged actin integrated into spot-like and linear F-actin structures along the core, albeit less continuously than QmNGMA (Fig. 1 D and Fig. S2 E). FRAP analysis of QAct5C::mScarI demonstrated that F-actin assemblies constituting the presynaptic actin core have slow but detectable recovery (Fig. 1, E and F), suggesting that they do not turn over quickly. To independently test dynamics, we evaluated linear actin core sensitivity to the actin-sequestering drug latrunculin A (LatA). Incubating dissected larvae for 30 min in 1 µM LatA led to disappearance of distinct actin structures at the NMJ, including the core, while DMSO solvent alone did not cause actin structure loss (Fig. S2, F and G). These results indicate that the core undergoes dynamic turnover. In sum, we identified different presynaptic actin assemblies that vary in size and lifetime and are probably involved in distinct functions at synaptic terminals.

Next, we asked whether linear F-actin structures at the NMJ resemble previously described axonal linear F-actin structures (D’Este et al., 2015; Gallo, 2006; Ganguly et al., 2015; Micinski and Hotulainen, 2024). We assessed QmNGMA-labeled actin in axonal bundles proximal to cell bodies in the ventral ganglion, and at distal axons exiting the nerve bundle to connect to NMJs (Fig. S3, A and B). In proximal axon bundles, QmNGMA and free GFP were indistinguishable both visually and by analysis of WEKA-segmented structures, possibly due to narrow axonal diameter, precluding any conclusion about actin structures in this region (Fig. S3 C). In contrast, in distal axons, we observed stretches of linear F-actin structures aligned with the axonal shaft, while neuronally expressed GFP filled the entire shaft and NMJ without labeling specific structures (Fig. S3 B). Thus, linear F-actin structures can be found in axons, but it remains inconclusive whether they exist throughout axons or represent a continuum from cell bodies to the actin core at NMJs.

Actin-associated proteins define the assembly, shape, and stability of actin structures (Pollard, 2016). For example, actin nucleators such as the Arp2/3 complex and formins facilitate the generation of branched and linear actin filaments, respectively. Previously, we demonstrated that most endocytic actin patches at the NMJ colocalize with the Arp2/3 complex (Del Signore et al., 2021). On the other hand, pharmacological formin inhibition disrupted presynaptic protein composition, vesicle cycling, and endocytosis in induced synapses and hippocampal neurons (Bingham et al., 2023; Wen et al., 2016). These studies suggest that distinct actin subpopulations might be generated via different F-actin nucleators. We assessed the presence of nucleation machinery along the presynaptic actin core in larvae co-expressing Lifeact::Halo (to label F-actin) and GFP-tagged Arp3 or the formin Dia. A fraction of both Arp3::GFP (Fig. S4 A and Video 2) and Dia::eGFP (Fig. S4 B and Video 3) puncta were visible along the linear actin core, unlike diffuse control cytosolic GFP distribution (Fig. S4 C and Video 4). We did not detect distinct enrichment of Arp3 or Dia on the presynaptic F-actin structures, suggesting that both nucleation mechanisms could contribute to their assembly.

To further evaluate the molecular constituents of linear presynaptic actin structures, we investigated nonmuscle tropomyosin localization. Tropomyosins are important F-actin stabilizers that copolymerize with actin in linear structures such as yeast actin cables (Alioto et al., 2016) and stress fibers in cultured mammalian cells (Tojkander et al., 2011). Tropomyosins are components of the axonal SMC and determinants of neuronal morphology and function (Abouelezz et al., 2020; Brettle et al., 2016). To ask whether Drosophila nonmuscle tropomyosin 1 (Tm1) associates with the linear presynaptic actin core, we generated transgenic flies containing QUAS-driven, mNeonGreen-tagged versions of two of the 18 predicted Drosophila Tm1 isoforms (QmN-Tm1-A and QmN-Tm1-L). When expressed in neurons (with nSyb-QF) and imaged live, both QTm1 isoforms labeled linear structures traversing the NMJ (Fig. S5, A and B). Compared with control cytosolic GFP, Q-Tm1-A colocalized significantly more strongly with the actin marker Lifeact::Halo along the actin core (Fig. S5, C and D). Q-Tm1-L signal was sensitive to the same LatA treatment that depolymerized Q-mNGMA–labeled actin structures (Fig. S5, E and F), consistent with Tm1 association with the F-actin core. We performed immunostaining with an antibody recognizing Tm1-A and Tm1-L (Cho et al., 2016) and observed presynaptic actin core-like distribution in control animals (Fig. S5 G). This Tm1 signal was strongly decreased at NMJs of larvae expressing RNAi targeting all Tm1 isoforms (panTm1⁴³⁵⁴²) or an independent line targeting Tm1-A (Tm1⁵⁶⁸⁶⁹) (Fig. S5, G and H), confirming specificity. Importantly, this indicates that the core is an endogenous structure independent of F-actin marker expression. In sum, the linear presynaptic actin core is decorated by the F-actin stabilizer tropomyosin, giving this structure a molecular signature resembling stress fibers (Tojkander et al., 2011).

Ventral stress fibers are actomyosin structures that sense and coordinate the adhesion of cultured cells to their substrates (Livne and Geiger, 2016). We asked whether the linear presynaptic actin core similarly recruits NMII. Drosophila NMII includes a NMII^HC encoded by Zipper/Zip, and a NMII^LC encoded by Spaghetti squash/Sqh. We used a well-established Sqh::GFP transgene, expressed from its own promoter (Royou et al., 2004) to evaluate colocalization with neuronally driven Lifeact::Halo. In live larvae, we observed Sqh::GFP as punctate submicron-sized structures localized partially along F-actin and dispersed throughout the bouton (in contrast to a free GFP control, which was uniformly distributed), suggesting that Sqh::GFP associates with discrete NMJ domains (Fig. S4 D and Video 5).

Next, we examined the distribution of endogenous NMII^HC/Zip via immunostaining of control NMJs with a previously described antibody (Sokac and Wieschaus, 2008) (Fig. 2 A). We validated antibody specificity by confirming decreased Zip signal in neurons and muscles upon expression of Zip RNAi (Zip⁶⁵⁹⁴⁷) in these tissues (Fig. 2, B and C; and Fig. S6, E–G). We detected Zip as puncta of various sizes enriched in and around HRP-labeled neuronal terminals, some of which aligned along the NMJ core similar to F-actin linear structures (Fig. 2 A). We also identified a subpopulation of larger aggregate-like Zip structures with smooth edges, localized predominantly in muscles (Fig. 2 A). Next, we examined whether Zip associates with presynaptic F-actin assemblies by neuronally co-expressing Zip::GFP and Lifeact::Halo (C155-Gal4>Lifeact::Halo>Zip::GFP^OE) (Fig. 2, D and E; and Video 6). Similar to endogenous NMII^HC/Zip staining, Zip::GFP labeled puncta of different sizes throughout boutons, and largely colocalized with Lifeact::Halo in both round and linear F-actin assemblies (Fig. 2, D–F). Interestingly, some of the Zip::GFP puncta were organized as doublets, resembling NMII bipolar filament organization (Fig. 2 E). These Zip doublets were spaced with an average distance of ∼260 nm in maximum intensity projection images (Fig. 2, E and G), slightly lower than the bipolar filament range of 280–400 nm in stress fibers (Svitkina et al., 1989). This tighter spacing may reflect different bipolar filament orientations along the z axis, or a contracted state (Russell et al., 2011). Some Zip::GFP doublets were detectable along linear F-actin structures, suggesting that they could form actomyosin assemblies, while others were not obviously correlated with F-actin and might represent constitutive NMII minifilament assembly (Vasquez et al., 2014). These results suggest that the linear actin structure and NMII together might form a presynaptic actomyosin core with contractile properties.

Ventral stress fiber contractility and structural integrity depend on NMII recruitment and mechanical tension (Livne and Geiger, 2016). To test how NMII regulates the presynaptic actin cytoskeleton, we imaged QmNGMA-labeled F-actin in control- and Zip RNAi/Zip⁶⁵⁹⁴⁷-expressing larvae, and observed reorganization of both round and linear presynaptic actin assemblies (Fig. 3, A and B; and Videos 7 and 8). Using our WEKA classifier, we segmented and analyzed spot-like and linear assemblies as separate categories (see Fig. S1, A–C for details). Although overall F-actin levels were comparable in both genotypes (Fig. 3 C), we found fewer actin spots (Fig. 3 D) and reduced linear F-actin coverage of the NMJ in Zip⁶⁵⁹⁴⁷ mutants (Fig. 3 E). The remaining round F-actin structures in Zip⁶⁵⁹⁴⁷ mutants were comparable in size and brightness to controls (Fig. 3 D), but linear actin assemblies were smaller and dimmer (Fig. 3 E; and Videos 7 and 8). Fixed Zip⁶⁵⁹⁴⁷ larvae showed similar presynaptic actin phenotypes (Fig. S7, A and B). The neuronal overexpression of Zip::GFP similarly reduced the area and brightness of both round and linear presynaptic Lifeact::Halo F-actin structures, compared with GFP alone (Fig. S7, C–E), suggesting that presynaptic actin organization depends on regulated NMII^HC/Zip levels.

To test how NMII activation state affects presynaptic actin, we manipulated NMII^LC/Sqh phosphorylation at conserved residues Thr20/Ser21, which control the transition from folded/inactive to unfolded/active NMII (Jordan and Karess, 1997; Sellers, 1991). Using Lifeact::Halo to visualize F-actin, we expressed either nonphosphorylatable (Sqh^A20A21/Sqh^DN) or phosphomimetic (Sqh^E20E21/Sqh^CA) Sqh mutants to alter the balance between inactive and active NMII (Fig. 3, F–H). Total F-actin levels at Sqh phosphomutant NMJs were comparable to GFP-expressing controls (Fig. 3 I), and the size, intensity, and number of round F-actin structures were not significantly changed (Fig. 3 J). However, both mutants significantly reduced the area covered by linear F-actin, with Sqh^DN causing more severe reductions in size and intensity (Fig. 3 K). Together with our Zip knockdown results, these findings show that NMII levels and activation state regulate presynaptic actin organization, particularly the linear actin core.

While exploring presynaptic actomyosin structures, we detected dynamic Sqh::GFP puncta at NMJs and in the postsynaptic muscle near boutons (Fig. S4 D and Video 5). This area contains the subsynaptic reticulum, a postsynaptic specialization enriched with ion channels, cell adhesion molecules, and receptors that facilitate neuron-to-muscle contact (Ataman et al., 2006). Similar postsynaptic Sqh was also present in fixed controls (Fig. 4 A). Neuron-specific Zip knockdown with two independent RNAi lines (Zip⁶⁵⁹⁴⁷ and Zip³⁶⁷²⁷) reduced Sqh at NMJ terminals (Fig. 4, A and B), and in axons (Fig. S6, A and B). Surprisingly, postsynaptic Sqh::GFP signal also decreased upon neuronal Zip knockdown, both near boutons and throughout muscles (Fig. 4, A and B; and Fig. S6 C). We ruled out leaky RNAi expression in the muscle by confirming that Sqh did not decrease in the absence of the neuronal driver C155-Gal4 (Fig. S6 D). Next, using WEKA-based analysis of Sqh::GFP particles in the α-HRP–delineated presynaptic and postsynaptic NMJ areas (Fig. 4 C), we confirmed a significant decrease in the number, size, and brightness of the NMII^LC/Sqh structures upon presynaptic Zip knockdown (Fig. 4, D and E). These results indicate that neuronal NMII^HC/Zip depletion results in transsynaptic decrease of the regulatory light chain Sqh in the muscle.

We next asked whether neuronal NMII depletion affects endogenous NMII^HC/Zip levels and distribution in muscle (Fig. 5 A), using our validated Zip antibody, Zip RNAi lines (Fig. 2, B and C; and Fig. S6, E–G), and Sqh RNAi lines (Fig. S6, A and B). Due to the density of Zip structures near the α-HRP–marked NMJ, we could not definitively isolate presynaptic Zip at our imaging resolution. We therefore focused our analysis on Zip organization deeper in muscles. WEKA-based image analysis of muscle Zip immunofluorescence distinguished two Zip populations differing in size by an order of magnitude: small Zip spots and large Zip aggregates. We tested transsynaptic effects of neuronal NMII depletion on these structures. Upon the C155-Gal4 expression of NMII^HC/Zip⁶⁵⁹⁴⁷ and NMII^LC/Sqh³²⁴³⁹ (Fig. 5 A), we observed increased abundance, size, and brightness of Zip spots near boutons (Fig. 5 B), and more Zip aggregates compared with controls (Fig. 5 C). Similar Zip aggregates were previously reported in Sqh null and phosphomimetic mutants and likely reflect NMII minifilament assembly due to altered Sqh/Zip stoichiometry, rather than protein misfolding (Jordan and Karess, 1997; Royou et al., 2004; Vasquez et al., 2014). These results demonstrate that neuronal NMII depletion triggers transsynaptic changes in both light and heavy chain organization, reducing Sqh::GFP puncta while promoting Zip aggregate formation in muscles.

The presynaptic actin core shares key features with mechanosensitive structures like stress fibers, including molecular composition, dependence on NMII, and ability to propagate cytoskeletal changes between cells. Thus, we hypothesized that presynaptic actomyosin drives the transsynaptic effects on NMII levels and organization. Since stress fibers transmit mechanical signals through integrin-based adhesions, we investigated whether the actomyosin core similarly engages transmembrane integrin receptors by examining the distribution of the Drosophila integrin-β PS subunit myospheroid (Mys). We visualized integrin-β using a ubiquitously expressed transgene (ubiMys::YFP) in live, dissected larvae (Yuan et al., 2010) (Fig. 6 A), and with an established antibody in fixed larvae (Dani et al., 2014; Wang et al., 2018) (Fig. 6 B). A subset of integrin-β structures aligned along the Lifeact::Halo-labeled linear actin core (Fig. 6 A, Fig. S8 A, and Video 9), and similar linear integrin-β formations were detectable in fixed larvae (Fig. 6 B), consistent with a prior report (Wang et al., 2018). We also observed diverse spot-like integrin-β entities both pre- and postsynaptically (Fig. 6 B), possibly reflecting different maturation stages (Kechagia et al., 2019; Livne and Geiger, 2016; Orr et al., 2022) and suggesting fine-tuning of cellular forces at the NMJ.

To test whether NMII regulates integrin-β organization, we knocked down NMII^HC/Zip⁶⁵⁹⁴⁷ in neurons and found significantly reduced integrin-β levels, as well as a decrease in the abundance, size, and intensity of linear and spot-like integrin-β structures in the α-HRP–delineated NMJ (Fig. 6, C–F). NMII^LC/Sqh³²⁴³⁹ expression similarly perturbed NMJ integrin-β entities (Fig. 6, C–F). Both Zip⁶⁵⁹⁴⁷ and Sqh³²⁴³⁹ significantly decreased postsynaptic integrin-β levels near boutons and reduced the size and intensity of postsynaptic spots (Fig. 6, D and G). These effects were reproduced with an independent Zip³⁶⁷²⁷ RNAi line (Fig. S8, B and C). Two controls confirmed the specificity of our finding: Zip RNAi without the Gal4 driver had no effect on integrin-β levels (Fig. S8, D–F), and integrin-β levels at myotendinous junctions (in the muscle but far from the NMJ) were unchanged in C155-Gal4>Zip⁶⁵⁹⁴⁷ larvae (Fig. S8, G–H). These results suggest that neuronal NMII depletion triggers local integrin-β reorganization near the NMJ.

To assess whether integrin-β also acts transsynaptically, we expressed Mys³³⁶⁴² RNAi in neurons with C155-Gal4. This effectively reduced neuronal integrin-β in axons (Fig. S9, A and B) but not at the NMJ, where presynaptic integrin-β is likely obscured by closely apposed postsynaptic integrin-β (Fig. S9, C and D). Overall postsynaptic and muscle integrin-β levels remained unchanged (Fig. S9 D). Notably, this rules out ectopic C155-Gal4 expression in the muscle as a trivial cause for NMII depletion in our previous experiment. Strikingly, postsynaptic integrin-β structures were smaller and dimmer with neuronally driven Mys³³⁶⁴² (Fig. S9, E and F), indicating transsynaptic effects on postsynaptic integrin-β organization. Together with our NMII knockdown results, these data suggest that presynaptic actomyosin is mechanically coupled to integrin receptors, propagating force-dependent signals across the NMJ.

Axotomy is widely used in NMJ studies as it prevents excessive muscle contraction in larval fillets, while the preparation maintains muscle potential and capacity for evoked neurotransmitter release (Feng et al., 2004). To complement our chronic NMII depletion studies, we examined how acute disruption of neuronal tension affects the presynaptic actin core by performing axotomy near the ventral ganglion (Fig. 7 A), similar to studies of mechanical tension effects on synaptic vesicle distribution in embryos (Siechen et al., 2009). We analyzed presynaptic actin assemblies labeled with neuronal QmNGMA, in larvae with intact versus cut axons; larvae were imaged live within ∼20 min after axotomy (Fig. 7 B; and Videos 10 and 11). While total F-actin levels and structure numbers remained unchanged after axotomy (Fig. 7 C), WEKA-based analysis revealed that severed axons had larger and brighter round actin structures (Fig. 7 D) but smaller and dimmer linear F-actin assemblies (Fig. 7 E). This reorganization resembles NMII depletion effects. To test whether NMII mediates this acute response, we performed axotomy in Zip⁶⁵⁹⁴⁷ larvae (Fig. S8 A). Despite reduced overall F-actin levels (Fig. S8 B), axotomy in NMII-depleted neurons increased round F-actin structures but did not alter linear F-actin (Fig. S8, C and D), suggesting NMII is required for tension-dependent remodeling of the presynaptic actin core.

We examined whether acute loss of axonal tension affects integrin-β organization by analyzing NMJs 15 min after axotomy (Fig. 7 F). While total integrin-β levels at the α-HRP–delineated and postsynaptic area at the NMJ remained unchanged (Fig. 7 G), we observed fragmentation of the linear integrin-β structures (Fig. 7 H), similar to neuronal NMII depletion. Strikingly, overall muscle integrin-β levels decreased within this 15-min window (Fig. 7 G), demonstrating that muscles rapidly sense and respond to changes in neuronal tension. Since these acute effects parallel our observations from chronic genetic manipulations, they suggest that synaptic integrin reorganization reflects a physiological response to neuronal actomyosin changes rather than an artifact of prolonged genetic manipulation.

To test the effects of increased mechanical tension along axons, we subjected larval preparations to a static stretching by gently pulling the brain for 15 min (Fig. 8 A), First, we examined whether stretching affected presynaptic actin organization using larvae expressing neuronal QmNGMA. Samples were fixed using cytoskeleton-preserving buffer (Jimenez et al., 2020) immediately after stretching. Compared with unstretched controls, stretched larvae showed no significant alterations in presynaptic F-actin organization (Fig. 8, B–E), indicating that at the resolution of our assay, the actin core remains structurally stable following increased tension. We next examined whether increased tension affected integrin-based adhesions using control larvae (not expressing fluorescent markers). Under identical stretching conditions, these larvae maintained comparable integrin-β levels across all analyzed compartments (Fig. 8, F–H). Interestingly, we observed a significant decrease in the size and fluorescence intensity of linear integrin-β structures. These findings demonstrate that mechanical manipulations along axons are rapidly sensed at the terminals, within 15 min of application. Axotomy (decreased tension) leads to fragmentation of actin structures and integrin adhesions, while static stretching (increased tension) similarly reduces integrin-based adhesion at the neuron–muscle interface. These results further confirm the capacity of this system to dynamically respond to changes in mechanical tension.

The NMJ represents an impressive metazoan solution to mechanical challenges posed by interactions between structurally and mechanically distinct cell types. Despite its importance, the molecular mechanisms that support NMJ mechanical integrity remain poorly understood. Neuronal actomyosin structures likely contribute to force sensing and adaptation, but have been difficult to study since their submicron size necessitates super-resolution imaging (Berger et al., 2018; Vassilopoulos et al., 2019), and the actin cytoskeleton is highly sensitive to fixation (Pereira et al., 2019). Moreover, resolving cytoskeletal elements between closely apposed neurons and muscles poses an additional challenge. We addressed these challenges using the Drosophila larval NMJ, by combining tissue-specific genetic tools, live imaging–compatible actin markers, high-resolution Airyscan microscopy, and machine learning–based image analysis to reveal distinct presynaptic actin structures.

We identified a linear actin core traversing the NMJ center that represents a candidate mechanosensitive structure. Live imaging and FRAP experiments demonstrated long-lived, low-turnover actin sensitive to LatA depolymerization. The core’s properties—linear organization, lifetime, turnover dynamics, tropomyosin content, and lack of nucleator enrichment—closely resemble ventral stress fibers (Livne and Geiger, 2016; Tojkander et al., 2015; Tojkander et al., 2011). Earlier studies noted structural similarities between axonal F-actin and stress fibers but dismissed functional parallels due to the apparent lack of focal adhesions (Gallo, 2006). Our findings validate this comparison, as the presynaptic actomyosin core couples to muscles through integrin-based adhesions.

The core’s dependence on NMII further supports its similarity to contractile structures. NMII^HC/Zip overlaps with the core in puncta and larger assemblies that resemble bipolar filaments spaced ∼260 nm apart, similar to stress fibers (Svitkina et al., 1989). NMII depletion or altered phosphorylation disrupted core organization, suggesting antiparallel actin filament organization required for contractility. However, we cannot rule out that some filaments may have uniform polarity like yeast actin cables (Moseley and Goode, 2006). The core’s NMII-dependent response to axotomy indicates mechanosensitivity to tension loss. In contrast, this structure did not exhibit notable reorganization at the endpoint of our experiment involving mechanical stretching to elevate neuronal tension. Whether the core undergoes rapid contractile responses followed by steady-state stabilization, as reported for stress fibers (Ni et al., 2023), requires future investigation.

We found that neuronal NMII downregulation caused striking transsynaptic changes in muscle NMII levels and organization. The reduction in NMII^LC/Sqh::GFP could result from mechanically responsive transcriptional pathways (Dupont and Wickström, 2022) or altered proteostasis following tension changes (Höhfeld et al., 2021). The Zip aggregations we observed might represent a physiological mechanism for NMII^HC storage when its activation is disrupted (Jordan and Karess, 1997; Royou et al., 2004). Interestingly, similar accumulations occur in myosin storage myopathies (Tajsharghi and Oldfors, 2013), suggesting sustained mechanical stress could trigger pathological myosin aggregation (Höhfeld et al., 2021).

We also detected reduced levels and altered organization of integrins at NMJs and muscles following neuronal NMII downregulation or axotomy. Since both manipulations disrupted the presynaptic actin core, this suggests the core mediates transsynaptic effects on muscle mechanoproteins. Like stress fibers, the presynaptic actomyosin core likely couples to integrin-based adhesions through mechanosensitive linkers like talin and vinculin (Goult et al., 2018). This inside-out activation of integrins likely promotes both cytoplasmic effector binding that drives adhesion maturation (Schiller and Fässler, 2013) and extracellular conformational changes that facilitate interactions with ligands in the ECM (Shattil et al., 2010) or on the neighboring muscle. Without proper presynaptic actomyosin function, integrin activation appears disrupted, reducing foci size and number both pre- and postsynaptically while sparing distant myotendinous junctions. Muscle contractions may reciprocally provide mechanical feedback through outside-in integrin mechanosensing (Chen et al., 2012), creating a feedback loop contributing to activity-dependent synaptic plasticity (Dani et al., 2014; Lee et al., 2017; Tsai et al., 2012). Conversely, axonal stretching showed no detectable changes in presynaptic actin organization but fragmented linear integrin-β structures, suggesting additional molecular players in mechanocoupling. Optimizing the stretching protocol to deliver and measure quantifiable changes in tension will provide further insights into the mechanisms of mechanotransduction at the neuron–muscle interface.

Interestingly, we observed both linear and round integrin-β assemblies at the NMJ, likely reflecting distinct organizational states and ligand interactions. This complexity aligns with the diversity of integrin heterodimers and ECM interactions in Drosophila (Humphries et al., 2006; Moreno-Layseca et al., 2019). The varying sizes of integrin foci even in wild-type conditions may reflect different maturation states (Kechagia et al., 2019), with recent work linking integrin/talin foci expansion to homeostatic synaptic plasticity at the Drosophila NMJ (Orr et al., 2022). We found that the linear integrin-β populations partially align with the presynaptic actin core, consistent with actin-dependent integrin organization (Swaminathan et al., 2017). Further, both linear and round integrin assemblies are disrupted by neuronal NMII depletion, suggesting the actomyosin core regulates their organization. Together, these findings suggest that synapses employ multiple modes of integrin-based adhesion to maintain mechanical continuity between neurons and muscle. An interesting question that arises from our study is how changes in muscle integrin levels arise. The relative amount of active integrins is thought to be controlled via two major mechanisms, both affected by changes in cell membrane tension. One controls the activation of the proteins already at the cell surface, while the other mechanism couples acute mechanical stress to integrin recycling and activation (Lolo et al., 2022). Thus, both mechanisms might be at play in muscles experiencing chronic neuronal actomyosin depletion, or acute changes in tension via axotomy.

The axon and presynaptic terminal form a mechanical continuum that must withstand both contractile muscle forces and axonal tension. Prior studies have identified distinct actomyosin organization in axons that may relate to our findings at the NMJ. The submembranous cytoskeleton (SMC), composed of periodic actin rings and spectrin, organizes molecules essential for neurotransmission, adhesion, and transport (Zhou et al., 2022), and serves as a mechanical scaffold and tension buffer system for strained axons (Dubey et al., 2020). NMII modulates axonal diameter and conduction (Costa et al., 2020; Zhou et al., 2022), and bulky cargo transport (Wang et al., 2020) without disrupting the SMC’s periodicity, indicating a role in radial contractility. In contrast, our NMJ data show NMII-dependent rearrangement of linear, longitudinal presynaptic actin, suggesting a distinct function from that in the SMC.

Longitudinal actin structures like “actin trails”—dynamic, bidirectional filaments terminating at boutons—have been observed in multiple neuron types (D’Este et al., 2015; Ganguly et al., 2015; Sood et al., 2018; Unsain et al., 2018), and may serve as actin delivery vehicles and reservoirs. If generated via mechanisms similar to ventral stress fibers, through reorganization of existing filaments rather than de novo assembly, the presynaptic core could serve a dual role as both a mechanosensitive structure and an actin reservoir. Interestingly, axonal actin trails were sensitive to moderate concentrations of SMIFH2, which is a formin inhibitor with off-target effects on NMII activity (Ganguly et al., 2015), raising the possibility that trails are NMII-dependent like the NMJ actin core. Earlier work also identified NMII-dependent longitudinal actin bundles involved in growth cone collapse and axon retraction (Gallo, 2006), while similar structures were recently associated with activity-dependent plasticity in the axon initial segment (Micinski and Hotulainen, 2024). Likewise, the Drosophila NMJ actin core requires NMII activity and reorganizes following axotomy. Notably, NMII activity is also essential for axon retraction and F-actin rearrangements after severing (Gallo, 2004; Phillips et al., 2019). Our work, along with these previous findings, suggests that neurons employ specialized actomyosin arrays in both axons and presynaptic terminals to maintain mechanical integrity.

We show that axotomy in wild-type neurons mimics the actin fragmentation observed in NMII-depleted NMJs and that further fragmentation does not occur upon NMII depletion. We propose that actomyosin-based elements, including the presynaptic actin core, contribute to a spring-like tension sensor spanning from soma to terminal. While this core may be part of a broader mechanical continuum, local cytoskeletal remodeling might also buffer strain (Mutalik et al., 2018). For instance, axonal actomyosin protects against mild mechanical stress through reversible beading and diameter changes, a capacity lost upon actomyosin inactivation, leading to calcium dysregulation and degeneration (Pan et al., 2024). We propose that the linear actin structures we describe might serve a similar protective function, helping to balance tension along neurons and at the neuron–muscle interface while providing mechanical flexibility to the neuromuscular system.

This presynaptic actin structure is a promising model for studying actin-linked neuromuscular disorders in Drosophila (Ermanoska et al., 2023). Our data further suggest that myopathy can contribute to neuromuscular disorders traditionally classified as peripheral neuropathies. For example, patients with MYH14 mutations, affecting the nonmuscle myosin NMIIC isoform, exhibit combined peripheral/cranial neuropathy and myopathy (Choi et al., 2011). Interestingly, NMII-C forms sarcomere-like structures on the periphery of epithelial cells that are precisely aligned with similar structures in the neighbor cell, forming a transcellular contractile network (Ebrahim et al., 2013). Thus, actomyosin rearrangements and mechanotransduction at the interface between different cells, including neuron and muscles, offer potentially valuable therapeutic possibilities.

We employed multiple complementary approaches and controls to address the technical challenges inherent to studying actomyosin dynamics at intact synapses in an animal. Visualizing presynaptic F-actin at the NMJ is particularly difficult, as phalloidin staining is incompatible with live imaging and, more importantly, prominent muscle F-actin obscures smaller presynaptic assemblies in our in vivo system. We relied on genetically encoded F-actin markers that we previously validated for this purpose (Del Signore et al., 2021), though we cannot fully exclude potential effects on actin dynamics (Kumari et al., 2020; Montes-Rodriguez and Kost, 2017; Spracklen et al., 2014). Our confidence in these tools is supported by prior findings showing that three distinct markers produced identical F-actin dynamics (Del Signore et al., 2021). Additionally, endogenous and tagged tropomyosin traversed the NMJ in patterns consistent with the actin core, providing marker-independent validation of this structure.

Studying protein localization and mechanical forces posed further challenges. Immunostaining, limited to fixed samples, can be confounded by dense postsynaptic signals, which we addressed using WEKA-based image analysis. While GFP-tagged NMII subunits are established tools, their overexpression can disrupt NMII biology (Heissler and Sellers, 2015); indeed, Zip::GFP altered presynaptic actin organization. We therefore complemented these experiments with endogenous Zip staining. Finally, since our genetic manipulations affect neurons throughout development, some phenotypes may reflect systemic rather than direct effects. While we addressed this through acute manipulation via axotomy or axonal stretching, we still lack tools to directly measure or quantitatively manipulate tension within neurons or muscles. Further, while our stretching experiments did not reveal obvious structural changes in the presynaptic actin core, this does not preclude more subtle molecular rearrangements or transient responses that might occur on faster timescales than we measured (Ni et al., 2023). Development of tension biosensors and methods for combining precise mechanical manipulation with high-resolution imaging will be crucial for understanding how mechanical forces shape synaptic organization and function.

Drosophila melanogaster stocks were raised on a standard cornmeal medium, or on molasses formulation for experiments in Fig. 2, Fig. 3, F–K, Fig. 5, Fig. 6, C–G, Fig. 7, F–H, Fig. 8, Fig. S3, Fig. S6, E–G, Fig. S7, C–E, and Fig. S10. Crosses were maintained at 25°C, on 12-h light–dark cycles. Detailed genotype information (including sex) of larvae is listed in Tables S1 and S2.

The following clones were generated by nucleotide synthesis at VectorBuilder, Inc.: mNeonGreen::GMA (GMA sequence described in Bloor and Kiehart [2001])), mNeonGreen::Tm1 isoform A (NCBI Accession: NP_524360.2), mNeonGreen::Tm1 isoform L (Accession: NP_996216.1), and mScarlet-I::Act5C (NCBI Accession: NP_001284915.1). These constructs were then subcloned into pQUAST-attB (DGRC Stock 1438; https://dgrc.bio.indiana.edu//stock/1438; RRID:DGRC_1438). All plasmids were sequence-verified (IDG and Plasmidsaurus, Inc.). Transgenic flies carrying these constructs were generated at BestGene Inc., using the PhiC31 system at attP40 or attP2 landing sites (see Table S2).

All larvae dissected are third instar wandering larvae, coming from at least three different crosses both for control and for mutant conditions. The crosses are set with 4–6 virgin female flies to avoid larval overcrowding. Third instar larvae with desired genotypes were held by two pins (anterior and posterior), dissected in Ca²⁺-free HL3.1 (pH 7.2) (Feng et al., 2004) in a Sylgard polymer–covered plastic petri dish. Internal organs were carefully removed to preserve muscles, brain, and axons from damage. After dissection, one or two larvae were immediately transferred to a microscopic coverslip (ventral side down) in a drop of ∼50 μl HL3.1. The larvae were carefully sandwiched, using permanent double-sided tape spacers (Scotch 3M ID CBGNHW011141) and a No. 1.5 glass coverslip, and imaged immediately. When two different conditions were tested (e.g., LatA treatment and DMSO, or intact brain and axotomy), we mounted pairs of larvae, and alternated the order in which the two conditions were imaged between pairs. For the LatA treatment, 1 mM InSolution Latrunculin A (428026; Sigma-Aldrich) in DMSO was brought to a concentration of 1 µM with HL3.1 dissection medium. As a vehicle control, we used 100% DMSO, similarly dissolved 1000× in HL3.1. Two larvae were dissected as described above, with one undergoing a 30-min incubation in LatA while the other was incubated in DMSO (in separate Sylgard-coated small petri dishes). After this pre-incubation, the larvae were immediately mounted in a drop of LatA or DMSO containing HL3.1 on a microscopic slide for imaging, as described above. To visualize the Lifeact::Halo actin marker, dissected larvae were incubated in HL3.1 with 2 µM Janelia Fluor HaloTag Ligand-549 for 5 min, and directly mounted in ∼50 μl HL3.1 without additional washing. For the axotomy experiment, two larvae were dissected in parallel, and then just before mounting, we cut the axons (close to the ventral ganglion) from one larva. Larvae were imaged within ∼20 min of axotomy.

Third instar larvae were dissected in HL3.1 (pH 7.2) in a Sylgard-coated dissection dish, fixed for 10 min in 4% PFA in HL3.1, washed three times for 15 min in 1xPBS, and permeabilized with 0.1 % PBX (1xPBS with Triton X-100) for 3 × 5 min. Incubation with the primary antibody was performed either for 2 h at room temperature, or overnight at 4°C. After a wash (3 × 5 min in 0.1% PBX), the samples were incubated with secondary antibody for 1 h, followed by a final wash in 0.1% PBX. Larvae were mounted in ProLong Diamond Antifade Mountant (Invitrogen) and cured at room temperature for 48–72 h before imaging. HRP-Red Rhodamine was used at 1:250. The GFP signal of Sqh::GFP was enhanced by staining with FluoTag-X4 anti-GFP (1:250; NanoTag Biotechnologies). Mouse monoclonal anti-integrin beta PS (CF.6G11, 1:50; DSHB) primary, and Alexa Fluor 488-AffiniPure Goat Anti-Mouse IgG (H+L) (Jackson ImmunoResearch) or CF488A Donkey Anti-Rabbit IgG (H+L), Highly Cross-Adsorbed Antibody secondary antibodies were used to detect integrin at the NMJ. Rabbit anti-Tm1 (D. Montell, UCSB, Santa Barbara, CA, USA) 1:500, precleared by incubation with dissected, fixed, and permeabilized w¹¹¹⁸ larvae) and Alexa Fluor 488 AffiniPure Goat Anti-Rabbit IgG (H+L) (Jackson ImmunoResearch) secondary antibodies were used to detect Tm1-A/L at the NMJ. Rabbit anti-Zip (A.M. Sokac, UIUC, Urbana, IL, USA) 1:1,000, precleared by spinning at 150, 000 g for 1.5 h) and CF488A Donkey Anti-Rabbit IgG (H+L), Highly Cross-Adsorbed secondary antibodies (Biotium, Inc.) were used to detect NMII^HC/Zip. Rhodamine Red-X AffiniPure Goat Anti-Horseradish Peroxidase (Jackson ImmunoResearch) was used to detect the neuronal membrane of NMJs and axons.

Third instar larvae were loosely pinned and dissected in HL3.1 solution, leaving the brain intact. Two larvae per condition (stretched and unstretched) were processed in the same Sylgard-coated dish. At the end of dissection, brains were gently anteriorly displaced by 400–500 µm using a dissection pin to induce mild axonal stretching. Axons, approximated as elastic cables, distribute tension along their length, with possible nonuniformities arising at anchor points such as the CNS, branch sites, and NMJs. Local geometry shapes axonal paths, with posterior branches possibly experiencing less axial and more curvature-induced stress. Tension should be highest near the CNS and taper distally. NMJs on dorsolateral internal muscle 4 in segments A3 and A4, innervated by the intersegmental nerve, were analyzed due to their intermediate position and sensitivity to stretch. Both stretched and nonstretched preparations were incubated for 15 min, followed by fixation. Larvae expressing the F-actin marker QmNGMA were fixed in 4% PFA prepared in cytoskeleton-preserving buffer (Jimenez et al., 2020), containing 80 mM PIPES, 5 mM EGTA, 2 mM MgCl2, pH 6.8, for 5 min. Larvae were then washed in PBS, mounted on coverslips with mounting medium, and imaged the same day. Larvae expressing the CTRL³⁵⁷⁸⁵ RNAi line neuronally were fixed and stained for integrin-β as described above in the immunohistochemistry procedures.

Imaging was performed on an inverted Zeiss LSM 880 or LSM 900 microscope with an Airyscan 2 detector, using a 63X oil immersion objective (NA 1.4). Raw images were subjected to 3D Airyscan processing with default settings in Zen Black software. Z-stacks of NMJs in fixed larvae were captured at a voxel size (0.044 × 0.044 × 0.185) µm. Z-stacks of NMJs in live larvae were captured with a pixel size 0.044 × 0.044 µm, and Z-step varying from 250 to 500 µm. For the FRAP experiment, we used NMJs with clearly visible actin core, photobleached them at 100% laser power, and followed the recovery of the signal for 90 s with 1-Hz sampling frequency.

All image analyses were performed in FIJI (Schindelin et al., 2012). Unless otherwise stated, we used maximum intensity projections for all image analyses.

For measuring the fluorescence intensity per µm² of the signals of interest in the presynaptic compartment of fixed samples, we segmented and masked NMJs using the HRP signal. The HRP signal was also used to clean parts of images containing axons and debris when analyzing the presynaptic compartment of the NMJ. Before analyzing images with Sqh::GFP and endogenous Zip, we cropped out sections HRP-labeled NMJ in proximity to glia, trachea, or axons (all compartments where Sqh::GFP under its own promoter or Zip is enriched in a distinct pattern from the spot-like signal at the NMJ). This way we prevented measuring of irrelevant nonneuronal or nonmuscle Sqh::GFP and Zip signal. To obtain a mask, the following steps were run on maximum intensity projections: Gaussian blur with sigma = 2 > Auto Threshold method—Huang > Convert to Mask > Erode.

When measuring fluorescence per µm² of dim signals present in both the pre- and postsynaptic compartments of the NMJ (like Sqh::GFP), the background fluorescence per µm² (measured in areas where no signal enrichment was observed) was subtracted from the neuronal fluorescence per µm².

For the remaining image analyses, NMJs were masked using the fluorescently tagged actin or actin marker, or Tm1 (using the HRP masking workflow). To specifically analyze objects in the 1-µm ring surrounding the boutons, we used the “enlarge” function to extend the neuronal mask by 1 µm, and used the XOR function in FIJI to intersect that specific region.

In the data presented in Fig. S2, F and G (LatA treatment of larvae expressing the actin marker QmNGMA in neurons), we generated a WEKA classifier (Arganda-Carreras et al., 2017) that masks the NMJ core, and determined changes in F-actin fluorescence in this area of interest.

We built a two-class WEKA classifier by manually annotating round and linear F-actin (Fig. S1, A–C). The classifier was applied to the first frame of the captured movies, and separate probability maps of round and linear F-actin assemblies were obtained as 32-bit images. In the next step of the image analysis pipeline, we used batch processing to perform particle analysis in Fiji. Round and linear F-actin particles were masked and analyzed separately.

The 32-bit probability map image of linear F-actin after applying the WEKA classifier gave good coverage of predominantly linear structures (Fig. S1 B). Next, the 32-bit image was transformed to a 16-bit image, which was further used to generate a binary mask. The probability map to binary image transition included more F-actin structures than the linear population. By optimizing the subsequent steps, we designed a pipeline where the binary mask of linear F-actin was further despeckled and skeletonized. Further particle analysis of the skeletonized mask was performed with a size cutoff = 0.02 µm²–infinity and a circularity cutoff = 0.00-0.20, ensuring that the majority of analyzed particles represented linear F-actin structures. The 32-bit probability map image of round F-actin after applying the WEKA classifier similarly gave good coverage of spot-like structures (Fig. S1 C), while the transition to binary mask did not introduce many linear actin structures. To better segment individual round assemblies, the binary mask underwent despeckling and erosion. All round F-actin particles larger than 0.0018 µm² were included in the analysis. In general, our image analysis pipeline is powerful in providing unbiased and high-throughput analysis of specific, individual F-actin assemblies at the synapse. However, we think that it underestimates the size and brightness of these structures as a result of our efforts to segment and analyze them as individual, largely nonoverlapping entities.

We built single-class WEKA classifiers that were trained by manually annotating the Sqh or integrin structures. The classifiers were applied to the maximum intensity projections of the Sqh or integrin channels, accordingly. The obtained masked Sqh::GFP particles were despeckled and eroded before finally analyzed with no size or circularity filters. To distinguish between linear and round integrin entities, the following cutoffs were applied at the particle analysis step: for spot-like integrin, a size cutoff of 0.0018 µm² (1 pixel²) and a circularity cutoff of 0.51-1; for linear integrin analysis, the segmented entities were analyzed as particles with a size higher than 0.0018 µm² and a circularity between 0 and 0.5.

Three-class WEKA classifier (“genuine” spots, aggregates, and background Zip contained in residual axons or trachea) was manually trained to recognize these structures. The classifier was applied to the maximum intensity projections of the Zip channel. The obtained masked Zip particles were analyzed with the following cutoffs: genuine Zip spots with a size higher than 0.0018 µm² (1 pixel²) and Zip aggregates with a size higher than 0.01 µm². Zip doublets were manually distinguished as paired spot-like structures with each having comparable fluorescence intensity, and thus considered to likely represent formations of bipolar NMII filaments. We then measured the distance between the maxima of the two spots.

Colocalization analysis was performed in Fiji, using a customized macro that segments the whole NMJ as a region of interest (ROI), and then applies the Coloc 2 plugin to individual slices of the two-channel z-stacks. Segmentation of the ROI was based on the Lifeact::Halo signal that fills the whole NMJ. The Coloc 2 analysis calculates numerous parameters, including Pearson’s R value, which we used in our study to score for colocalization of two fluorophores.

We used GraphPad Prism v9 for statistical analyses and generation of graphs. Data were not tested for normality, and nonparametric tests were used for analyses. Mann–Whitney tests were used when analyzing two conditions, and one-way ANOVA with the Kruskal–Wallis or Šídák’s tests with multiple comparisons was used to determine the statistical significance of the data from more than two different conditions. To determine the effect size, we measured Hedges’ g value, which we also added to graphs. We calculated Hedges’ g using the following formula: $g=M1−M2/SDpooled$⁠, where M₁ – M₂ is the difference in means, while SD*_pooled is the pooled and weighted standard deviation. We used bar graphs or box-and-whisker graphs to represent our data. In the bar graphs, the error bars represent the standard error of the mean. In the box-and-whisker graphs, the whiskers are drawn down to the 10^th percentile and up to the 90^th percentile. Data points below or above the whiskers are shown as individual points. The line passing through the box represents the median. We used a logarithmic scale along the y axis to display the wide range of values for the area and fluorescence of the different actin assemblies, Sqh and integrin particles in a compact fashion. The FRAP curves in Fig. 1 F and Fig. S2 D were constructed after measuring the fluorescence intensity per µm² within the bleached region, as well as in a non-photobleached region to account for the background photobleaching during imaging. The measured values were normalized to the maximum fluorescence intensity value and fitted into a double-exponential curve in GraphPad.

This article contains supporting information (10 supplemental figures and two tables). Fig. S1 shows image analysis pipeline using a two-class WEKA classifier to segment round and linear F-actin assemblies (associated with Fig. 1). Fig. S2 shows time-lapse insets of neuronally expressed actin markers, FRAP analyses of the markers, and analyses of changes in fluorescence and distribution of the QmNGMA actin marker upon LatA treatment (associated with Fig. 1). Fig. S3 shows live images and assessment of linear F-actin in axons of Drosophila fillet preparations (associated with Fig. 1). Fig. S4 shows live images of actin-associated proteins at the presynaptic actin core (associated with Fig. 1). Fig. S5 contains images of the F-actin–stabilizing protein tropomyosin with linear F-actin structures in the presynaptic core in live and fixed preparations, in controls and upon neuronal expression of Tm1 RNAi lines (associated with Fig. 1). Fig. S6 contains images and analyses supporting the validation of UAS-RNAi lines targeting NMII^HC/Zip and NMII^LC/Sqh in neurons and muscles (associated with Fig. 2). Fig. S7 shows presynaptic actin assemblies upon downregulation and overexpression of NMII^HC/Zip in neurons (associated with Fig. 3). Fig. S8 shows live images demonstrating that NMII^HC/Zip facilitates changes of presynaptic F-actin upon mechanical severing of larval axons (associated with Fig. 7). Fig. S9 contains analyses and controls supporting that neuronal depletion of nonmuscle myosin II rearranges integrin receptors at the NMJ (associated with Fig. 6). Fig. S10 shows the organization of integrin receptors at the NMJ upon integrin-β neuronal downregulation (associated with Fig. 6). Table S1 describes genotypes, sample size, and statistics by dataset. Table S2 describes Drosophila strains, antibodies, and other reagents used in this study.

All data are available upon request.

We thank the Drosophila Genomics Resource Center (NIH Grant 2P40OD010949); Bloomington Drosophila Stock Center (Indiana University, Bloomington, IN, NIH 595 P40OD018537); Guy Tanentzapf (UBC) for PUbi-mys.YFP; Denise Montell (UCSB) for the Tm1 antibody; Adam Martin (MIT) for UAS-Zip::GFP flies; Anna Marie Sokac (UIUC) for Zip antibody; Bob Asselbergh and Simona Manzella (VIB/CMN/UAntwerp) for help with Zeiss LSM 900 microscopy; and the Brandeis University Light Microscopy Core Facility, RRID:SCR_025892. B. Ermanoska is grateful for immense support from I.J. and K.J.

B. Ermanoska is currently supported by Marie Sklodowska Curie Postdoctoral fellowship 101107344; J. Baets is supported by the Research Fund—Flanders (FWO) via a Senior Clinical Researcher mandate 1805021N. This work was supported by kBOF FFB240038 to B. Ermanoska; Research Fund—Flanders (FWO) Research Grant (G071723N) and EU Horizon 2020 program/Solve-RD under grant agreement 779257 to J. Baets; and NINDS grant R01 NS116375 to A.A. Rodal. J. Baets is a member of the European Reference Network for Rare Neuromuscular Diseases (ERN EURO-NMD). J. Baets and B. Ermanoska are members of the µNEURO Research Centre of Excellence of the University of Antwerp.

Author contributions: B. Ermanoska: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, software, supervision, validation, visualization, and writing—original draft, review, and editing. J. Baets: project administration, resources, and supervision. A.A. Rodal: conceptualization, data curation, funding acquisition, methodology, project administration, supervision, and writing—review and editing.