Dynamin-2 regulates synaptic podosome maturation to facilitate neuromuscular junction development

Neuromuscular junctions (NMJs) govern rapid and efficient neuronal communication with muscle cells, which relies on the proper architecture of specialized postsynaptic compartments. However, the intrinsic mechanism in muscle cells contributing to elaborate NMJ development has been unclear. In this study, we reveal that the GTPase dynamin-2 (Dyn2), best-known for catalyzing synaptic vesicle endocytosis at the presynaptic membrane, is also involved in postsynaptic morphogenesis. We demonstrate that Dyn2 is enriched in the postsynaptic membrane of muscle cells and is involved in the maturation of neurotransmitter receptor clusters via its actin bundling ability. Dyn2 functions as a molecular girdle to regulate synaptic podosome turnover and promote morphogenesis of the postsynaptic apparatus. In Drosophila NMJs, Dyn2 is required to organize the postsynaptic actin cytoskeleton and to mediate its electrophysiological activities. Mechanistically, the actin binding, self-assembly, GTP hydrolysis ability, and Y597 phosphorylation of Dyn2 all regulate its actin bundling activity. Together, our study uncovers a role for Dyn2 in cytoskeleton remodeling and organization at the postsynaptic membrane of NMJs.


Introduction
Neuromuscular junctions (NMJs) control the final output of the nervous system to direct voluntary movements. NMJs are equipped with elaborate membranous and cytoskeletal structures developed from a stepwise process of morphogenesis, i.e., prepatterning of neurotransmitter receptors during embryogenesis and maturation of the pre-and post-synaptic apparatus postnatally (Sanes & Lichtman, 2001). Muscle prepatterning occurs from embryonic day (E) 12.5 to E14.5 of early mouse embryogenesis, which features acetylcholine receptor (AChR) clustering and subsequent motor neuron innervation (Lin, Burgess et al., 2001, Shi, Fu et al., 2012, Yang, Arber et al., 2001. After birth, these AChR clusters undergo a plague-topretzel morphological transition that requires extracellular matrix (ECM) signaling, motor neuron activity, and muscle-intrinsic machineries (Bernadzki, Rojek et al., 2014, Bezakova & Ruegg, 2003, Marques, Conchello et al., 2000, Shi et al., 2012. After maturation, NMJs are maintained to enable life-long motor performance with limited plasticity (Burden, Huijbers et al., 2018, Kummer, Misgeld et al., 2006, Shi et al., 2012. Among these processes, the intrinsic mechanism in muscle cells that contributes to NMJ maturation remains poorly understood. Several studies have identified the presence of a unique actin-based structure, the podosome, at the postsynaptic membrane of NMJs, which dictates the plague-to-pretzel morphological transition of AChR clusters in aneurally-cultured myotubes (Chan, Kwan et al., 2020, Proszynski, Gingras et al., 2009. Podosomes are protrusive actin structures equipped with ECM degradative abilities and they are also involved in proper tissue development, immune cell patrolling, and cancer cell metastasis (Linder, 2007, Linder & Aepfelbacher, 2003, Linder & Wiesner, 2016, Luxenburg, Geblinger et al., 2007. Unlike the podosomes in motile cells that facilitate cell motility, synaptic podosomes promote NMJ maturation by redistributing AChR and ECM components in cultured myotubes (Bernadzki et al., 2014, Kishi, Kummer et al., 2005, Proszynski & Sanes, 2013. Recently, activity of the podosome component MT1-MMP (a metalloproteinase) has been reported to be crucial for NMJ development in Xenopus and mouse (Chan et al., 2020). Therefore, although it has been well documented that synaptic podosomes play critical roles in NMJ development, their dynamics and architecture remain ill-defined.
Previously, we discovered that a GTPase, dynamin-2 (Dyn2), is enriched at podosomes in differentiated myoblasts where it stiffens the podosomes and promotes myoblast fusion.
Apart from its functions in catalyzing membrane fission, Dyn2 has also been found to localize and function as an actin remodeling protein at many actin-rich structures, including lamellipodia, dorsal membrane ruffles and podosomes (Ferguson & De Camilli, 2012, Menon & Schafer, 2013. Although it has been reported that Dyn2 can regulate actin rearrangement by directly binding actin filaments and interacting with actin polymerization regulators (Gu, Yaddanapudi et al., 2010, Mooren, Kotova et al., 2009, it is not clear what the molecular function of dynamin is at the synaptic podosomes of myotubes.
In this study, we reveal that Dyn2 forms a belt-like structure around the podosome core of myotubes through its actin bundling activity to promote the maturation and turnover of synaptic podosomes, thereby regulating the development and function of NMJs. The actin bundling activity of Dyn2 is derived from its actin-binding and self-assembly abilities, and is regulated by phosphorylation of residue Y597 and GTP hydrolysis. Together, our results reveal a pivotal role for Dyn2 in synaptic podosome maturation and turnover, providing important insights into its function at postsynaptic NMJs.

Dyn2 forms a belt-shaped structure around the actin core of podosomes
To explore the function of Dyn2 at synaptic podosomes, first we used immunostaining and z-sectioning confocal microscopy to determine the spatial distribution of endogenous Dyn2 in myotubes differentiated from C2C12 myoblasts on laminin-coated coverslips (Kummer , Misgeld et al., 2004, Proszynski et al., 2009. Podosomes are actin-based protrusive structures comprising an actin core, an actin-based cable network, and an integrin-based adhesive ring (Fig   1a). The actin core is enriched with a branched actin polymerization machinery, with the cable network emanating from the actin core to link with the adhesive ring composed of integrin and adaptor proteins (Linder & Aepfelbacher, 2003, Linder & Wiesner, 2016, Luxenburg et al., 2007. Similar to a previous finding (Proszynski et al., 2009), we observed that Dyn2 is enriched at the actin core by co-staining with phalloidin and the specific podosome scaffold protein, Tks5 (Seals, Azucena et al., 2005) (Fig. 1b). An orthogonal view of a podosome from a reconstructed z-stack image further showed that Dyn2 is specifically enriched at the edge of the actin core where it forms a belt-shaped structure (yz view in Fig. 1b). In contrast, the actin polymerization machinery, comprising Arp2/3 and cortactin, was evenly enriched throughout the actin core and it partially colocalized with the Dyn2-enriched belt. Both the actin cable network and adhesive ring, labeled by myosin IIA and vinculin respectively, were located outside of the Dyn2 belt ( Fig.   1c). Localization of the belt-shaped Dyn2 structure between the actin core and actin cable was further confirmed by stimulated emission depletion (STED) microscopy ( Supplementary Fig.   S1a, b). Therefore, despite direct interaction between Dyn2 and cortactin , their slightly different localizations in podosomes suggest that Dyn2 may play an additional role in those structures apart from promoting actin polymerization. Notably, we also observed belt-shaped Dyn2 structures around actin cores of podosome rosettes in cSrctransformed NIH3T3 fibroblasts ( Supplementary Fig. S1c).
Given that Dyn2 belts were not observed in all podosomes (arrows in Fig. 1b; Supplementary Fig. S1d), we sought to investigate the role of Dyn2 in podosomes. We quantified the sizes (height and width) of actin cores and categorized podosomes into three groups-those equipped with full, partial, or no Dyn2 belt-and found that the proportion of podosomes in each group was relatively similar (Fig. 1d). Statistical analysis showed that podosomes with a Dyn2 belt had taller and wider actin cores (2.83 ± 0.20 µm in height and 1.46 ± 0.25 µm in width) compared to those with partial Dyn2 belts (2.19 ± 0.22 µm in height and 1.02 ± 0.19 µm in width) or lacking a Dyn2 belt (1.65 ± 0.28 µm in height and 0.79 ± 0.17 µm in width) (Fig. 1e, f). These findings raise the possibility that Dyn2 may contribute to podosome growth.
Next, we performed live-cell imaging to explore the temporal distribution of Dyn2 together with other critical podosome components in myotubes (Fig. 1g, h). Consistent with the kinetics of podosome components in other cell types (Luxenburg, Winograd-Katz et al., 2012), we found that F-actin and cortactin accumulated synchronously during initiation of synaptic podosomes, which was followed by Tks5 recruitment (Fig. 1g; Supplementary Fig. S1e). This result supports the roles of cortactin in podosome initiation/formation and Tks5 in podosome maturation.
Defining podosome lifespan according to the period of F-actin appearance, we found that podosomes in myotubes exhibited heterogenous lifespans, with >90% of podosomes having lifetimes longer than 15 min, which differs from the short-lived (<15 min) podosomes in macrophages and osteoclasts (Destaing, Saltel et al., 2002, Guiet, Verollet et al., 2012 (Fig 1i, Supplementary video 1). Furthermore, more than 30% of the podosomes we assessed had a lifespan longer than 60 min. Dyn2 enrichment possesses ~48 % of the lifespan of podosomes ( Fig. 1j; Supplementary Fig. S1f), and transient Dyn2 appearance explained the occurrence of podosomes decorated with partial Dyn2 belts at a given time-point. We also noticed that, after a Dyn2 belt disappeared, the actin core (labeled by Lifeact-GFP) gradually dissembled and eventually vanished, suggesting a potential role for Dyn2 in podosome turnover.

Dyn2 is required for the growth of podosomes
To investigate the function of Dyn2 in podosomes, we downregulated Dyn2 in C2C12derived myotubes by means of lentiviral shRNAs ( Fig. 2a; Supplementary Fig. S2a) and then examined the morphology of podosomes by z-stack confocal microscopy. Dyn2 knockdown significantly reduced both the mean height and width of podosome cores from 1.81 m and 1.52 m to 1.03 m and 0.86 m, respectively, without affecting podosome density (Fig. 2b, c).
Consistent with the defect in podosome growth, podosome lifespan was also significantly diminished in Dyn2-depleted myotubes (Fig. 2d, e). These results demonstrate an essential role for Dyn2 in podosome growth, but not podosome initiation.
In contrast, acute treatment with the dynamin GTPase inhibitor dynasore induced an accumulation of Dyn2 at podosomes, which resulted in time-dependent actin core elongation (Supplementary Fig. S2b,c), suggesting that the GTPase activity of Dyn2 is critical for podosome turnover. However, we noted that prolonged dynasore incubation (12 h) resulted in a reduction of podosome numbers in myotubes, likely due to indirect effects or a pleiotropic cellular response caused by blockage of dynamin-mediated endocytic pathways ( Supplementary   Fig. S2d).

Dysregulated Dyn2 activity affects podosome growth and turnover
To understand which biochemical activity of Dyn2 is responsible for the growth of podosomes, we utilized adenovirus to transiently overexpress different dominant-negative mutants of Dyn2 in wild-type or Dyn2-depleted myotubes. These mutants included a hyperassembling CNM-linked mutant (A618T), a membrane fission-defective CMT-linked mutant (G537C), a mutant with lower actin binding ability (K/E) or with higher actin binding activity (E/K), as well as a GTPase-defective mutant (K44A) (Chin et al., 2015, Gu et al., 2010, Kenniston & Lemmon, 2010 (Fig. 3a, b). Further detailed information and references for these mutants are provided in Fig. 3b and Supplementary Table 1. Exogenously-expressed wild type Dyn2 (Dyn2 WT ), Dyn2 G537C and Dyn2 E/K were enriched in podosomes and rescued the size of podosome actin cores in Dyn2-depleted myotubes ( Fig. 3c-f).
In contrast, neither Dyn2 K/E nor Dyn2 K44A were enriched in podosomes, nor could they rescue the size of actin cores. Intriguingly, Dyn2 A618T did localize to podosomes and restored actin core height but not width. Similar effects of these Dyn2 mutants on synaptic podosomes were also observed in wild type myotubes, with Dyn2 A618T , Dyn2 G537C and Dyn2 E/K localizing to podosomes, but Dyn2 K/E and Dyn2 K44A not doing so ( Fig. 3g; Supplementary Fig. S3a-d).
Importantly, the Dyn2 A618T mutant protein (which exhibits hyper-self-assembly) led to a significant increase in podosome height, whereas the membrane fission-defective Dyn2 G537C mutant protein did not affect any apparent morphological feature of podosomes. These results suggest that the actin binding activity of Dyn2 is critical for its enrichment at podosomes, and hyper-self-assembly of Dyn2 can result in abnormal podosome morphology.
To further examine the effect of Dyn2 on podosome turnover, we co-transfected LifeAct-GFP and mCherry-tagged Dyn2 mutants into myoblasts and induced their differentiation into myotubes to measure the lifespan of podosomes. Statistical analysis revealed that whereas expression of Dyn2 G537C -mCherry, Dyn2 K/E -mCherry, and Dyn2 E/K -mCherry did not significantly alter podosome lifespan, Dyn2 A618T -mCherry expression prolonged podosome lifetimes, manifesting as a dramatic increase in long-lived podosomes (> 2 hr in Fig. 3h).
Moreover, unlike the temporal emergence of Dyn2 WT in podosomes (Fig. 1h), Dyn2 A618T belts were very stable throughout the lifetime of podosomes (Supplementary video 2). In Supplementary Fig. S3e, we present an example of podosomes decorated with Dyn2 A618T belts for 14.5 hr with an overall lifetime of ~15 hr.
Regrettably, we were unable to analyze the effect of Dyn2 K44A -mCherry on podosome lifetime in myotubes due to its strong dominant-negative effect on endocytosis and myoblast differentiation (Chuang, Lin et al., 2019), which prevented us from generating myotubes that expressed Dyn2 K44A -mCherry. Instead, we analyzed the effect of Dyn2 mutants on podosome rosettes in cSrc-transformed NIH3T3 fibroblasts and observed an increase in podosome height and lifespan in Dyn2 A618T -expressing cells, whereas there was a notable decrease in podosome size and lifespan in Dyn2 K/E -and Dyn2 K44A -expressing cells ( Supplementary Fig. S3f-i).
Together, these results show that the actin binding, self-assembly, and GTP hydrolysis activities of Dyn2, but not its membrane fission ability, are involved in regulating podosome growth and turnover.

Dyn2 is required for synaptic podosome function
The function of synaptic podosomes is to promote NMJ development through their ECM degradative ability, a critical feature of mature podosomes (Chan et al., 2020, Linder, 2007, Proszynski et al., 2009. To assess if proper activity of Dyn2 is required for the functionality of synaptic podosomes, we performed an ECM degradation assay on Dyn2-depleted myotubes reexpressing either wild-type or different mutant proteins. Similar to their effects on podosome morphology, Dyn2 WT , Dyn2 A618T , Dyn2 G537C and Dyn2 E/K could restore the ECM degradative ability of Dyn2 knockdown myotubes, but Dyn2 K/E or Dyn2 K44A mutants could not (Fig. 4a, b).
Furthermore, expression of Dyn2 A618T in wild type myotubes resulted in a significant increase in the area of degradation (Fig. 4c, Supplementary Fig. S4a). Given that podosome density in Dyn2 A618T -expressing myotubes remained unchanged ( Supplementary Fig. S3d), this increase in ECM degradation indicates that those podosomes enclosed by Dyn2 A618T not only have taller actin cores and longer lifetimes, but also exhibit a better ECM degradative ability. Thus, our results demonstrate that Dyn2 promotes podosome maturation and turnover, and that the CNMassociated Dyn2 A618T mutant protein has a dominant effect on the stability of synaptic podosomes.
To further explore the effect of Dyn2 on synaptic podosome-mediated NMJ development, we assessed AChR distribution by staining with Alexa488-conjugated bungarotoxin (BTX). The AChR clusters in Dyn2 WT and Dyn2 G537C myotubes were perforated and had similar areas relative to control, whereas the AChR clusters in Dyn2 A618T -expressing myotubes were perforated but with less cluster area (Fig. 4d, e). Decreased AChR area in Dyn2 A618T -expressing myotubes might be due to higher ECM remodeling ability of Dyn2 A618T (Fig. 4c). This finding suggests that Dyn2 plays a critical role in synaptic podosome and AChR cluster maturation, and that the CNM-associated Dyn2 mutant variant may perturb their development.

Dyn2 regulates actin-based postsynaptic cytoskeletal organization and postsynaptic development
To establish if Dyn2 A618T could affect the development and electrophysiological function of NMJs, we utilized Drosophila as it is a commonly used system for studying synaptic development and function. We generated UAS transgenes of HA-tagged wild-type and mutant human Dyn2, and expressed each of them predominantly in muscles using the MHC-GAL4 driver (Schuster, Davis et al., 1996). We observed that Dyn2 WT was localized in postsynaptic NMJs and colocalized with α-spectrin, an actin-binding protein associated with the postsynaptic plasma membrane and that is involved in establishing the postsynaptic subsynaptic reticulum (SSR) ( Supplementary Fig. S5a) (Pielage, Fetter et al., 2006). The Dyn2 A618T and Dyn2 G537C mutant variants also presented similar enrichment at postsynaptic NMJs, but the pattern and intensity of -spectrin signal was significantly reduced upon expression of Dyn2 A618T ( Fig. 5a- with RNAi knockdown of αor -spectrin altering SSR membrane organization and synaptic transmission (Pielage et al., 2006). To examine the functional consequences of altered spectrin organization, we recorded the membrane potential of postsynaptic NMJs from Dyn2 A618Texpressing Drosophila larvae and observed increased amplitudes of miniature excitatory junction potential (mEJP) and evoked EJP ( Fig. 5c-f), reminiscent of the effect of spectrin depletion in Drosophila (Pielage et al., 2006). However, mEJP frequency and Quantal content were not altered in these mutants ( Supplementary Fig. S5e, f).
Greater mEJP amplitude could result from an increased level or activity of postsynaptic glutamate receptors, or both. However, we noted that neither the expression level nor the cluster size of GluRIIA, the major subunit of glutamate receptors (Marrus, Portman et al., 2004), was altered by expression of Dyn2 A618T (Supplementary Fig. S5g-j). Given that expression of Dyn2 A618T alters postsynaptic organization and electrophysiological activities, we examined SSR ultrastructure using transmission electron microscopy (TEM) and observed that Dyn2 WT or Dyn2 A618T did not cause any apparent change in the area or thickness of SSR ( Fig. 5f-h).
However, expression of Dyn2 A618T , but not Dyn2 WT , reduced SSR density (SSR layers/SSR thickness), indicative of a looser membrane structure (Fig. 5f, i). Thus, our ultrastructural and electrophysiological data indicate that the hyper self-assembly activity of the Dyn2 A618T mutant protein alters actin-dependent cytoskeletal organization, thereby disturbing the electrophysiological function of the postsynaptic compartment of Drosophila NMJs. It is worth noting that, similar to Drosophila NMJs, endogenous Dyn2 is also enriched at mouse NMJs ( Supplementary Fig. S5k, l).

The actin bundling activity of Dyn2 is regulated by GTP hydrolysis
To directly investigate how Dyn2 remodels actin cytoskeleton, we utilized in vitro reconstitution to explore the aforementioned biochemical activities of Dyn2 and its effect on actin bundle formation. Similar to previous reports (Chuang et al., 2019, Gu et al., 2010, we observed prominent F-actin bundling activity for Dyn2, but much less bundled actin for Dyn1 To better visualize how addition of GTP affects Dyn2-mediated actin bundling, we utilized negative-stain TEM to image the Dyn2-actin bundles with or without nucleotides. We found that unlike the aligned and bundled actin filaments surrounded by ordered Dyn2 rings in the absence of GTP, the Dyn2 rings were disassembled and the actin bundles became dispersed upon addition of GTP, but not GMPPCP (Fig. 6f). We also analyzed the effect of other GTP analogs, including GDP and GDP:AlF4‾ (which mimics the transition state of GTP hydrolysis), and found that actin bundles remained stable, but Dyn2-actin assemblages were less ordered in the presence of GDP relative to GMPPCP ( Supplementary Fig. S6c).
The GTP hydrolysis-induced Dyn2 disassociation from actin is reminiscent of Dyn2 disassembly from the membrane upon GTP hydrolysis (Bashkirov, Akimov et al., 2008, Chin et al., 2015, Pucadyil & Schmid, 2008. Furthermore, we occasionally observed a partially packed Dyn2 displays prominent bundling activity on branched actin, which is also regulated by GTP hydrolysis (Fig. 6g-k). The morphological differences of Dyn2-bundled linear and branched actin were better visualized under confocal microscopy ( Fig. 6l). Together, these results show that Dyn2 bundles both types of actin filaments that are present in podosomes, and that assembly of Dyn2 oligomers on actin filaments is regulated by GTP hydrolysis.

CNM-associated Dyn2 mutants are insensitive to GTP hydrolysis
Given the differential effect of Dyn2 A618T and Dyn2 G537C on cytoskeleton remodeling in vivo (Fig. 5a), we examined if these two mutants exert different impacts on actin bundles. Using an actin sedimentation assay, we found that both Dyn2 A618T and Dyn2 G537C present comparable actin bundling activities (Fig. 7a, b). However, whereas actin filaments bundled by Dyn2 WT and Dyn2 G537C dissociated upon GTP hydrolysis, those bundled by Dyn2 A618T were resistant to GTP addition and remained bundled (Fig. 7a, b). Importantly, both linear and branched actin bundled by other CNM-associated Dyn2 mutant proteins, Dyn2 R465W and Dyn2 S619L , were also insensitive to GTP (Fig. 7c-g, Supplementary Fig. S7). These findings indicate that Dyn2 can bundle both unbranched and branched actin, which is terminated by GTP hydrolysis. Thus, Dyn2 assembly and sensitivity to GTP hydrolysis play decisive roles in actin organization, and these biochemical features are critical for podosome turnover as well as NMJ morphology.

Phosphorylation of Dyn2 residue Y597 is important for its podosome targeting and actin bundling activity
Dyn2 is recruited to plasma membrane clathrin-coated pits by interacting with several SH3 domain-containing proteins (Meinecke, Boucrot et al., 2013). However, despite there being a direct interaction between Dyn2 and the podosome components cortactin and Tks5 , Oikawa, Itoh et al., 2008, we have already shown that the spatiotemporal distribution of Dyn2 is somewhat distinct from those core components (Fig. 1), raising the possibility that targeting of Dyn2 to podosomes may be regulated by a mechanism other than protein-protein interactions. Src tyrosine kinase is known to be a critical podosome-initiating enzyme (Gimona, Buccione et al., 2008, Murphy & Courtneidge, 2011. Dyn2 has been reported to be a substrate of Src kinase and is phosphorylated at residue Y597, both in vivo and in vitro (Ahn, Kim et al., 2002). Interestingly, Src-mediated phosphorylation was shown to induce Dyn1 self-assembly in vitro, as well as promote the function of Dyn2 on Golgi apparatus (Ahn et al., 2002, Weller, Capitani et al., 2010. To test if Y597 phosphorylation of Dyn2 is responsible for its targeting to podosomes, we expressed a phospho-deficient mutant fusion construct, Dyn2 Y597F -mCherry, in myotubes and observed a significant reduction in podosome association for this mutant (Fig. 8a, b). We also noted that expression of a phospho-mimetic mutant, Dyn2 Y597E , interfered with myoblast differentiation, thus preventing us from examining its effect on synaptic podosomes. Instead, we used cSrc-transformed NIH3T3 fibroblasts to examine the effect of mutant Dyn2 Y597 . Similar to its effect in myotubes, Dyn2 Y597F also presented reduced enrichment at podosomes and induced smaller podosome rosettes, whereas Dyn2 Y597E clearly targeted to podosome rosettes ( Supplementary Fig. S8a, b). Moreover, in the actin sedimentation assay, more of the bundled actin assembled by phospho-mimetic Dyn2 Y597E remained in the presence of GTP, indicating a slower dissociation rate of this mutant variant (Fig. 8c, d). Consistent with that outcome, much Dyn2 Y597E -bundled actin could still be observed under TEM upon GTP treatment (Fig. 8e).
Similar to Src-phosphorylated Dyn1 (Ahn et al., 2002), Dyn2 Y597E also displayed enhanced selfassembly in low salt buffer ( Supplementary Fig. S8c), recapitulating the hyper self-assembly of the Dyn2 A618T mutant. Accordingly, our results demonstrate that Y597 phosphorylation is the molecular trigger for the actin bundling activity and podosome targeting of Dyn2.

Discussion
In this study, we reveal a mechanochemical role for Dyn2 in synaptic podosome maturation and turnover, which facilitates the development of NMJs. We provide evidence of Dyn2 enrichment and function at the NMJs of Drosophila and mouse, as well as its role in postsynaptic membrane development during synaptogenesis. Upon phosphorylation of residue Y597, Dyn2 directly targets to and bundles actin filaments to form belt-shaped structures around podosomal actin cores, which promotes podosome maturation, whereas GTP hydrolysis-induced Dyn2 disassembly triggers podosome turnover (Fig. 8f). Thus, Dyn2 activity is involved in the development and electrophysiological activity of NMJs through its regulation of synaptic podosome maturation and turnover (Fig. 8g).
An essential role for Dyn2 in actin organization was demonstrated a decade ago (Bruzzaniti, Neff et al., 2005, Destaing, Ferguson et al., 2013, Ochoa, Slepnev et al., 2000. It was generally assumed that Dyn2 functions to promote actin polymerization through its abilities to remove capping proteins and direct interactions with cortactin, Nck, and profilin (Gu et al., 2010, Mooren et al., 2009, Schafer, 2004. Here, we discovered a structural role for Dyn2 in regulating stress during protrusion. Based on our data, we hypothesize that Dyn2 binds and assembles around the actin cores to stiffen them, facilitates their assembly into the columnar architecture and, finally, triggers podosome turnover via GTP hydrolysis-induced disassembly (Fig. 8f) (Chuang et al., 2019). Therefore, we speculate that Dyn2 functions as a molecular girdle to maintain the structure of actin cores when podosomes encounter physical stress.
In contrast to our understanding of podosome formation, relatively little is known about how podosome turnover is regulated. To date, only myosin II, Supervillin and fascin have been reported to regulate podosome turnover (Bhuwania, Cornfine et al., 2012, Van Audenhove, Debeuf et al., 2015, van den Dries et al., 2013. Myosin II and Supervillin enable podosome turnover by increasing actomyosin contractility, whereas fascin facilitates podosome disassembly by inhibiting Arp2/3-mediated actin branching. Here, we have shown that GTP hydrolysisinduced Dyn2 disassembly is also involved in podosome turnover. These observations demonstrate that both formation and turnover of podosomes are tightly regulated in cells and that Dyn2 plays roles in both processes. Further studies are needed to better understand the multiple regulatory mechanisms controlling podosome turnover.
Although the function of Dyn2 at the presynaptic membrane has been well studied (Chung, Barylko et al., 2010, Hayashi, Raimondi et al., 2008, Newton, Kirchhausen et al., 2006, the role of Dyn2 at the postsynaptic membrane is largely unknown. Consistent with the function of synaptic podosomes in NMJ maturation (Proszynski & Sanes, 2013), we have shown here that postsynaptic expression of Dyn2 mutants with a defect in podosome turnover results in abnormal AChR cluster morphology in cultured myotubes and a disorganized pattern of spectrin distribution in Drosophila NMJs. Defects of postsynaptic spectrin and actin have been reported to disrupt SSR integrity, active zone spacing, glutamate receptor clustering, and electrophysiological activity (Blunk, Akbergenova et al., 2014, Pielage et al., 2006, Proszynski et al., 2009. Given that postsynaptic expression of Dyn2 A618T impaired the organization of postsynaptic spectrin, we reason that Dyn2 A618T expression causes a milder defect in Drosophila NMJs than spectrin knockdown.
CNM-associated Dyn2 mutations are hypermorphic alleles resulting from loss of autoinhibitory regulation, which promotes self-assembly (Faelber, Gao et al., 2013, Hohendahl, Roux et al., 2016. Hyper-assembled Dyn2 is hyperactive, resistant to GTP hydrolysis-induced disassembly, and presents enhanced membrane fission activity (Chin et al., 2015, Kenniston & Lemmon, 2010. Similar to observations of hyper-assembly of CNM-associated Dyn2 mutant protein on membrane, we show that Dyn2 A618T has stronger actin bundling ability, less sensitivity to GTP, and also extended podosome lifespan. Importantly, the phospho-mimetic Dyn2 Y597E mutant had a similar effect to Dyn2 A618T , demonstrating that self-assembly is a critical switch of Dyn2 function in cells. The structure of dynamin oligomers around membrane templates has been beautifully resolved; it binds to membrane via the PH domain and self-assembles into helixes through its stalk region (Antonny et al., 2016, Kong, Sochacki et al., 2018. Interestingly, our negative-stain TEM images revealed similar ring-like Dyn2 oligomers surrounding actin filaments. Akin to Dyn2 spirals on lipid templates, these Dyn2 oligomers on actin filaments are also responsive to GTP hydrolysis, which induces Dyn2 disassembly. Combined, these findings demonstrate that Dyn2 is a unique actin-binding protein that aligns and packs actin filaments together by forming ring-like oligomers around them. Recently, it was reported that the Dyn2 homolog in Drosophila, shibire, is also a multifilament actin-bundling protein, with shibire rings being located within the actin bundles (Zhang, Lee et al., 2020). Given that dynamin-actin bundling assays were conducted in non-physiological salt concentrations of 50 mM and 75 mM KCl in their and our experiments, respectively, we are conscious of the artificial conditions we both used. Therefore, parallel in vivo experiments, together with careful interpretation of in vitro reconstitution results, are critical to study the interplay between dynamin and cytoskeleton proteins (Shpetner & Vallee, 1989). Since our results demonstrated recruitment of Dyn2 to synaptic podosomes after formation of actin cores, we speculate that Dyn2 binds to actin filaments of podosome cores and assembles around them.
In summary, Dyn2 is an evolutionarily conserved actin bundler that regulates synaptic podosome maturation and turnover, where it participates in the development of NMJs. Many questions remain unanswered regarding exactly how Dyn2 binds and assembles on actin filaments, whether Dyn2 assembly at podosomes is regulated by other interacting protein, and if Dyn2 contributes to postsynaptic maturation in synapses other than NMJs. Our study highlights a distinct regulatory molecule involved in podosome turnover and the necessity for further studies on podosome kinetics and mechanical properties.
The laminin solution was aspirated completely before plating cells. Cells were seeded at 80% confluency one day before differentiation. After differentiation, cells cultured on laminin-coated coverslips or Permanox slides were ready for immunostaining, and cells cultured on laminincoated glass-bottom dishes were ready for time-lapse imaging.

Transfection, lentiviral and adenoviral infection
For transfection, cells at 70% confluency were transfected with target DNA using

Matrix degradation assay
To make FITC-gelatin-coated coverslips, acid-washed coverslips were first coated with 0.01% poly-D-lysine (#P7280, Sigma-Aldrich) for 1 h at room temperature and washed three times with PBS. We added 0.5% glutaraldehyde (#G5882, Sigma-Aldrich) to coverslips on ice for 15 min and washed with cold PBS. Pre-warmed coating solution composed of 0.1 mg/ml FITC-gelatin (#G-13187, Invitrogen) and 10 g/ml laminin (#23017-015, Invitrogen) were added to the coverslips, which were then placed in the dark for 10 min at room temperature.
After washing with PBS, 5 mg/ml NaBH4 (#452882, Sigma-Aldrich) was added for 15 min to inactivate residual glutaraldehyde. The FITC-gelatin-coated coverslips were washed with PBS and stored in 70% ethanol.
Day 3-differentiated C2C12 myotubes were plated on FITC-gelatin-coated coverslips with or without adenovirus induction. After 16 h, the cells were fixed, stained for F-actin and HA-Dyn2, and imaged by confocal microscopy. The thresholding feature in Metamorph software (Molecular Devices) was used to analyze matrix degradation areas and cell-containing areas.

Fly stocks and genetics
Fly stocks and GAL4 lines were obtained from the Bloomington Drosophila Stock Center and maintained on normal food medium. The parental strain ZH-51D was used to generate transgenic flies by injecting the pUAST-based constructs into Drosophila embryos, thereby integrating them into the attP (second chromosome) landing site.
They were subsequently fixed in 1% OsO4/0.1 M cacodylic acid solution at room temperature for 3 h. The samples were subjected to a series of dehydration steps, i.e., from 30% to 100% ethanol.
After the 100% ethanol dehydration step, the samples were incubated in propylene, a mixture of propylene and resin, and then in pure resin. Lastly, they were embedded in 100% resin. The images of type Ib boutons were captured using a Tecnai G2 Spirit TWIN system (FEI Company) and a Gatan CCD Camera (794.10.BP2 MultiScan) at ≥4,400× magnification. We identified type Ib boutons by the multiple layers of subsynaptic reticulum, and the size and layers of SSR of type Ib boutons were measured using Image J (NIH) accordingly to previous reports (Budnik, Koh et al., 1996, Lee & Schwarz, 2016. In brief, SSR thickness was measured in ImageJ as follows: (1) the center of mass was determined by drawing a region of interest around the periphery of a bouton; (2) four lines were drawn 90 degrees apart from one another emanating from the center of mass; and (3) SSR thickness was then determined based on the average length between SSR edge and bouton edge along each line. SSR layers were determined by the plot profile feature in ImageJ across the segment of SSR on each line.

Electrophysiology
Evoked excitatory junctional potential (EJP) was recorded as previously described (Peng, Lin et al., 2019). Briefly, third instar larvae were dissected in calcium-free HL3 buffer at room temperature, followed by incubation in 0.5 mM Ca 2+ HL3 solution for 5-10 min prior to recording. The mean resistance value for the recording electrode was ~40 MΩ when 3 M KCl solution was used as the electrode solution. All records were obtained from muscle 6 in the A3 hemisegment. Resting membrane potentials of muscles were held at less than −60 mV. EJPs were amplified using an Axoclamp 900A amplifier (Axon Instruments) under bridge mode and filtered at 10 kHz. EJPs were analyzed using pClamp 10.6 software (Axon Instruments).
Averaged EJP amplitude was calculated from the amplitudes of 80 EJPs in one consecutive recording. Miniature EJP recordings were performed in HL3 solution containing 0.5 mM Ca 2+ and 5 μM tetradotoxin (TTX) and also analyzed using pClamp 10.6 software.

Image analysis
Immunostained images were analyzed in Metamorph (Molecular Devices) and ZEN (Carl Zeiss) software. Cell area and Drosophila NMJ area were selected manually and the area and signal intensity were measured in Metamorph. Matrix degradation area as well as GluRIIA and AChR cluster area were selected using threshold features in Metamorph and quantified automatically using the same software. Podosome diameter and height were analyzed manually in xz orthogonal view in ZEN.

F-actin bundle sedimentation assay
Dynamin proteins were expressed in Sf9 cells transiently transfected with various constructs and purified as previously described, then snap-frozen in buffer containing 20 mM HEPES, 150 mM KCl, 1 mM EGTA, 1 mM DTT and 10% glycerol (Liu, Neumann et al., 2011). The actin bundling assay was performed as described previously ( Proteins were visualized by Coomassie blue staining, and band intensities were quantified using ImageJ.

Transmission EM
To visualize actin bundles, 5 M filamentous actin was incubated with or without 1 M Dyn2 at room temperature for 30 min. The mixture was then diluted 2-fold and adsorbed onto carbon-coated grids and stained with 2% uranyl acetate. Images were collected using a Hitachi H-7650 electron microscope at 75 kV and a nominal magnification of 120,000. To image Dyn2actin bundles upon addition of GTP, the actin mixture was placed on parafilm before GTP addition. GTP was added to the actin mixture with a final concentration of 1 mM. The carboncoated grids were placed on top of the mixture one min before being subjected to 2 % uranyl acetate staining. Negative-stained samples of actin bundles in the presence of other nuceotides were prepared and captured by TEM as described above.

F-actin
---+ + + S P S P S P S P S P S P + + + + + + e muscle D y n 2 W T D y n 2 Y 5 9 7 F D y n 2 Y 5 9 7 E D y n 2 W T D y n 2 Y 5 9 7 F