Transient localization of the Arp2/3 complex initiates neuronal dendrite branching in vivo

Neurons extend branched dendrites to establish appropriate connections within a circuit. Dendrite branching enhances the coverage of the receptive field of a neuron and modulates the initial processing of incoming stimuli. Complex dendrite patterns are achieved in vivo through a dynamic process of branch extension and retraction that initiates with the formation of a filopodium-like dendrite branchlet (Jontes et al., 2000; Kaethner and Stuermer, 1997; Sugimura et al., 2003; Wu et al., 1999). Branchlets are generally short processes of less than 30 µm in length, but their appearance and distinct dynamics vary among different types of neuron (Nagel et al., 2012). Only a subset of these dynamic branchlets is stabilized and can elongate into bona fide branches that remain for extended periods of time with the potential of branching further, while most disappear within a range of minutes to hours (Dailey and Smith, 1996; Heiman and Shaham, 2010; Niell et al., 2004). The initial step that allows the initiation of a dendrite branchlet is thus key to elaborating dendrite morphologies.

Actin is expected to play a central role in the formation of dynamic dendrite branchlets. Super-resolution microscopy recently revealed details of the organization of the actin cytoskeleton in dendrites, including periodical actin rings (D'Este et al., 2015). Additionally, patches of actin were observed in dendrites in electron microscopy preparations of primary hippocampal neuron cultures, as well as in pyramidal cells of the adult mouse cortex imaged in vivo with stimulated emission depletion (STED) nanoscopy (Korobova and Svitkina, 2010; Willig et al., 2014). These actin patches contain the actin nucleator complex Actin-Related Protein 2/3 (Arp2/3) and give rise to dendrite spines (Korobova and Svitkina, 2010; Saarikangas et al., 2015). However, the dynamic structure of actin supporting the process of bona fide dendrite branch formation has not been clarified. In Drosophila sensory class III dendritic arborisation (cIIIda) neurons, the formation of dendrite branchlets is predicted by the accumulation of a marker for dynamic F-actin: GMA (Andersen et al., 2005). GMA is a genetically encoded GFP-tagged actin-binding domain of moesin (Edwards et al., 1997) and its accumulation suggests a rapid remodelling or nucleation of actin at the site of branch formation. Additionally, GMA and LifeAct, an alternative F-actin marker, localize in dynamic puncta in Drosophila sensory class IV da (cIVda) neurons, at sites of new branch formation (Nithianandam and Chien, 2018). The Arp2/3 complex is a strong candidate for reshaping actin at the site of branch formation, as it promotes dendrite branching in cultured hippocampal neurons (Dharmalingam et al., 2009; Zhang et al., 2017) and is a major target of the small GTPase Rac1, a conserved key regulator of dendrite morphology (Govek et al., 2005).

Combining in vivo time-lapse analysis of fluorescently tagged actin regulators in da neurons, genetic analysis and electron tomography, we demonstrate that the Arp2/3 complex is the major actin nucleator involved in the initiation of dendrite branchlet formation, under the control of the suppressor of cAMP receptor/WASP family verprolin homology protein (SCAR/WAVE). Our data reveal a fundamental mechanism for dendrite branch formation via actin remodelling.

To address the specific regulatory events at the core of the process of initial dendrite branch formation, we concentrated on the Drosophila larval sensory da neurons (Gao et al., 1999; Jan and Jan, 2010; Singhania and Grueber, 2014). These fall into four morphologically and functionally distinct neuronal subclasses, which range from the simple cIda to the complex cIIIda and space-filling cIVda neurons (Grueber et al., 2002). We first investigated the dynamics of actin enrichment within differentiating cIVda neuron dendrites in live second instar larvae. Dynamic actin, highlighted by GMA (Edwards et al., 1997), was enriched within the distal dendrite branches, which exhibit a higher rate of de novo branchlet formation (Fig. 1A). It displayed a highly dynamic localization with GMA puncta, as well as longer stretches of GMA enrichment forming and disappearing within minutes (Fig. 1C,D; Movie 1). Although transient GMA enrichment did not per se predict a site of de novo branch formation, the appearance of a novel branchlet was preceded by a persistent punctum of GMA enrichment. Indeed, 90% of all analysed de novo branch formation events showed actin accumulation of more than 30% relative to the basal level (n=85) (Fig. 1B). These data indicate that polymerization of actin is highly dynamic during dendrite differentiation in cIVda neurons. Moreover, they suggest that actin polymerization is a required step but does not suffice for the formation of a novel branch.

The initial assembly of an actin filament requires actin nucleators that stabilize the first seed of G-actin molecules (Firat-Karalar and Welch, 2011; Rottner et al., 2017). Therefore, to reveal the molecular machinery underlying branch formation, we performed a late cell-autonomous knockdown in cIda, cIIIda and cIVda neurons of actin nucleators, including the formins Capu, dDAAM, Dia, Fhos, FMNL and Formin3, the WH2 domain protein Spire, and components of the Arp2/3 complex (Dietzl et al., 2007). This approach allowed bypassing the initial phase of neuronal polarity establishment and primary neurite elongation to observe specific effects during terminal dendritic branching. The knock down of subunits of the Arp2/3 complex, including Arp2, Arp3 or Arpc1, strongly reduced the number of dendrite branches in cIVda (Fig. S1A,D) and cIIIda neurons (Fig. S1B,E). Although the primary branches (Fig. S1D-F) were not modified, other branch orders were reduced in number in cIVda and cIIIda neurons (Fig. S1D,E).

We thus addressed the in vivo role of Arp2/3 in dendrite branchlet formation. Using mosaic analysis with a repressible cell marker (MARCM) (Lee and Luo, 1999), we generated single cIVda neuron clones carrying null mutations of the Arp2/3 subunit Arpc1 (Arpc1^Q25sd or Arpc1^R337st) (Hudson and Cooley, 2002; Zallen et al., 2002). These two alleles led to a strong reduction in overall branch density in cIVda neurons, validating the RNAi phenotype (Fig. 2A-C,E). In addition, in Arpc1^Q25sd mutant clones, high order branch number was reduced and average branch length increased (Fig. 2E). The cIVda neuron phenotype was rescued by expressing Arpc1-GFP in the mutant neurons (Hudson and Cooley, 2002), demonstrating that the dendrite branching phenotype derives from the loss of Arpc1 function and that Arpc1 is required cell-autonomously (Fig. 2D,E).

To distinguish whether the reduction in the number of branches observed upon loss of Arp2/3 function is due to a defect in branch formation or to an increased rate of branch retraction, we studied dendrite branching dynamics using time-lapse imaging of class IV ddaC neurons of second instar larvae expressing control or Arp2 RNAi constructs. ddaC neurons in which Arp2 was knocked down had a much lower rate of de novo dendrite branch formation (0.28±0.26 in comparison with 2.37±0.70 newly formed branchlets per 100 μm, n=5), whereas the rates of extension and partial or complete retraction were not affected (Fig. S2A,B,D; Movie 2). Hence, Arp2/3 complex promotes branching in cIVda neurons primarily by initiating de novo formation of dendrite branches.

To investigate the localization of the Arp2/3 complex during dendrite differentiation, we expressed GFP-tagged Arp3 (Arp3-GFP) in cIVda neurons of second instar larvae, under the control of Gal4^ppk, a condition that does not alter the morphology of cIVda neurons (data not shown). Arp3-GFP was enriched at discrete puncta (Fig. 2F; Movie 3). Newly forming branches were almost invariably marked by the presence of Arp3-GFP at the branching point (Fig. 2I, black line). Furthermore, Arp3-GFP localization marked the site of subsequent branch formation 60 s in advance (Fig. 2G-I). Thus, Arp3 localization strongly correlated with the formation of a new branchlet and predicted the site of branchlet formation. The localization was transient and the signal decreased rapidly again after the initiation of banchlet elongation (Fig. 2I). Consistently, we observed a decrease of Arp3-GFP fluorescence when branchlets disappeared (Fig. 2I). Taken together, the phenotypic characterization of Arp2/3 complex loss of function, together with the Arp3-GFP localization observed in live-cell imaging, suggest that the local recruitment of the Arp2/3 complex is an early step towards dendrite branch formation.

The nucleating activity of the Arp2/3 complex is intrinsically low and requires activation via interaction with the VCA (verprolin-homology, central, acidic) domain of Wilskott-Aldrich syndrome protein (WASP) family of proteins (Campellone and Welch, 2010). Using RNAi we knocked down the WASP family proteins characterized in fly in da neurons and found that only a reduction of the SCAR/ WAVE function resulted in a similar phenotype to the one observed upon loss of Arp2/3 function (Fig. S3). Cell type-specific knockdown of WAVE resulted in a reduction of branch number and density in cIIIda and cIVda neurons, and a reduction of branches in cIda (Fig. S3A-F). To validate the knockdown data, we generated single wave^Δ37 mutant cell clones by MARCM (Zallen et al., 2002). The densities of secondary, tertiary and higher-order branches in wave^Δ37 mutant class IV ddaC neurons were clearly reduced (Fig. 3B,C). Interestingly, a membrane-tagged WAVE construct lacking the VCA domain for Arp2/3 activation and acting as a dominant-negative (WAVE^ΔVCAmyr) (Stephan et al., 2011) reproduced the wave mutant phenotype that displayed a reduction in the number and density of secondary, tertiary and higher-order branches (Fig. S4C,D). RNAi-mediated wave knockdown in ddaC neurons produced a strong reduction of de novo branch formation. The remaining dendritic branches appeared to be more stable and less dynamic than the control (Fig. S2C,D; Movie 2). Thus, WAVE, like Arp2/3, is primarily required for de novo formation of dendrite branchlets. The activation and membrane recruitment of WAVE is in turn under tight regulation by the conserved multiprotein WAVE regulatory complex (WRC) (Chen et al., 2010). Within the WRC, the Sra-1 and Nap-1 dimer is essential for both Rac1-mediated regulation of the WRC and its recruitment to the membrane (Chen et al., 2010). Sra-1 knockdown by RNAi yielded a similar effect to the loss of Arp2/3 or loss of WAVE, namely a clear reduction in the density of secondary and high-order branches of cIVda neurons (Fig. S4F,G). Together, these data suggest that WAVE is a major activator of Arp2/3 during dendrite branch formation in cIVda neurons.

The small GTPase Rac1 is a central determinant of cytoskeleton organization through the regulation of multiple proteins, including the WRC (Steffen et al., 2004; Tahirovic et al., 2010). Furthermore, da neurons overexpressing Rac1 display an increased number of terminal dendrite branchlets (Andersen et al., 2005; Emoto et al., 2004; Lee et al., 2003; Nagel et al., 2012). We therefore asked whether Rac1 drives the activation of WAVE and Arp2/3, and promotes dendrite branchlet formation. Single cIVda neuron MARCM clones mutant for Rac1 (Rac1^J10) (Hakeda-Suzuki et al., 2002) displayed a reduction in the number of tertiary branches (Fig. 3D-F), indicating that appropriate branching of cIVda neurons requires Rac1 (Lee et al., 2003; Nagel et al., 2012; Soba et al., 2015). Next, we used a photoactivatable form of a constitutively active mutated Rac1 protein (PA-Rac1CA) (Wang et al., 2010; Wu et al., 2009). When applying localized photoactivation in a restricted region of interest (ROI) of cIVda neurons, this construct was de-suppressed and produced strong and rapid formation of single branchlets and of groups of branchlets (Fig. 3G,I; Movie 4). Such an effect was not elicited in unilluminated ROIs upon overexpression of a dominant-negative PA-Rac1DN or of a light-insensitive LI PA-Rac1CA (Fig. S4H-J; Movie 5). A large fraction of the branchlets formed after illumination appeared along the length of existing dendrites, rather than at dendrite tips, and remained stable (Fig. 3I). Importantly, the dendrite overbranching locally induced by PA-Rac1CA was largely suppressed in the background of cell-autonomous wave RNAi (Fig. 3H,I; Movie 6). Thus, Rac1-induced branchlet formation requires WAVE in da neurons.

The Arp 2/3 complex binds to already existing actin filaments and is the only actin nucleator generating characteristic branched F-actin networks (Mullins et al., 1998). We thus attempted to locate the activity of the Arp2/3 complex in a structural context and carried out electron microscopy of dendrite branching points in primary neuronal cultures (Small et al., 2008). We obtained striking images of branching points of dendrites of larval brain neuron primary cultures (Fig. 4). In dendrite branches, we observed a few microtubules and long actin filaments extending parallel to the cortical membrane (Fig. 4B), similar to what has been reported for cultured rat hippocampal neurons (Xu et al., 2013). In contrast, the branchlets were filled by a tight bundle of actin filaments (Fig. 4F). However, we did not observe regions of branched actin filaments within the dendrite branchlets, as reported for the filopodia that give rise to dendrite spines (Korobova and Svitkina, 2010). At branching points, actin filaments appeared more dispersed and less bundled than along the main dendrites (Fig. 4B). Splayed microtubule filaments were present at branching points (Fig. 4B). In all analysed tomograms of branching points in dendrites (n=5 fully traced tomograms out of 12 sets), we observed several patches of short highly branched actin filaments (Fig. 4C-E,G). To follow the actin filaments in 3D and to unambiguously trace actin branching points, we performed EM tomography of dendrite branching points and reconstructed actin and microtubules (Fig. 4H; Movie 7). In the tilted images we could define unequivocally sites of actin branching (Movie 7), a clear indication of Arp2/3 activity. Single filaments derived from a patch of branched actin converged into actin bundles forming the core of a branchlet (Fig. 4F). Thus, patches of branched F-actin were specifically present at the base of dendritic branching points, supporting a role for Arp2/3 in dendrite branching.

The data presented here delineate a model of the mechanism underlying dendritic branching. This involves the initial activation of Rac1, leading to recruitment and local activation of the WRC. The subsequent transient recruitment of the Arp2/3 complex promotes the nucleation of branched actin filaments close to the plasma membrane that can thus be pushed forward to form an initial process.

One major trait of neuronal morphology is the density of dendrite branches (Stuart et al., 2007), which is likely produced by a combination of frequency of de novo branch formation and incidence of stabilization of the produced branchlets (Ziv and Smith, 1996). Our data indicate that the distribution of sites of Arp2/3 complex activation plays a major role in the control of branch formation frequency. Rac1 activity promotes actin recruitment at sites of dendrite spine formation and of dendrite filopodium initiation (Cheadle and Biederer, 2012; Galic et al., 2014; Korobova and Svitkina, 2010). Furthermore, Rac1 is responsible for the localization of WAVE at the growth cone plasma membrane of mouse cerebellar granule neurons (Tahirovic et al., 2010). We thus hypothesize that activated Rac1 might recruit WRC at various locations along the length of a dendrite, but once a crucial concentration of active WAVE is reached, Arp2/3 can locally remodel the actin cytoskeleton to promote the formation of an actin patch. This cascade closely resembles the initiation of axonal branch formation (Armijo-Weingart and Gallo, 2017). In addition, the localization of WRC can be supported by a large family of molecules containing a short WRC interacting receptor sequence (WIRS) (Chen et al., 2014). In fact, molecular complexes including WIRS-containing proteins were recently shown to contribute to the localization of axon or dendrite branching sites in C. elegans HSN and PVD neurons, respectively, in part by binding to the WRC (Chia et al., 2014; Zou et al., 2018). The transmembrane receptor Robo, which promotes branching in cIVda neurons, contains a WIRS domain and might represent a potential candidate for the localization of WRC in these neurons to support branch formation (Chen et al., 2014; Dimitrova et al., 2008). How the localization of WIRS-containing activators might in turn become spatially restricted to initiate branch formation at discrete locations along the dendrite is not clear.

Taken together, we have revealed the details of the initial step into the formation of a dendrite branchlet in vivo with high temporal and spatial resolution. We suggest that, given the potential pathways of activation of WRC, in different neurons a concentration of active Arp2/3 that is sufficient for initiating branchlet formation might be achieved by stochastic enrichment of the activators or by the tethered localization of activator complexes at specific sites.

All flies were kept at 25°C unless otherwise noted. The fly strains used are listed in the supplementary Materials and Methods. MARCM was performed as described by Nagel et al. (2012) (see supplementary Materials and Methods). For all RNAi experiments, animals carried a da neuron class-specific driver (Gal4^ppk, Gal4^IG11 or Gal4^C161), UAS-mCD8 GFP, UAS-Dcr2, and a target or control RNAi. Embryos were allowed to develop at 25°C until examined as third instar larvae.

Live late third instar larvae expressing UAS-mCD8GFP in subsets of da neurons were mounted in 70% glycerol and immobilized with a cover slip for visualization with a Leica TCS SP2 confocal microscope. For time-lapse imaging, second instar live larvae expressing UAS-mCD8GFP in subsets of da neurons were immobilized in an imaging chamber (Dimitrova et al., 2008) filled with halocarbon oil to allow oxygen exchange, and confocal z-stacks (six images spaced by 1 µm) were obtained using a Carl Zeiss LSM7 confocal microscope every 5 min over a period of 30 min. Two-colour imaging of Gal4^109(2)80/ +; UAS-mCD8-Cherry/UAS-GMA or Gal4^109(2)80/ +; UAS-mCD8-Cherry/UAS-Arp3-GFP larvae was performed as previously described (Nagel et al., 2012) with several modifications. Briefly, for GMA, imaging stacks spaced by 1 µm were obtained every 30 s over a period of 10 min with a dual camera system; for Arp3 GFP, images were obtained every min over a period of 10-15 min with a single camera. In both cases, a Yokogawa Spinning-Disc on a Nikon stand (Andor) with two back-illuminated EM-CCD cameras (Andor iXON DU-897) was used. Two colour images of dual camera recordings were aligned automatically using Huygens essentials (Scientific Volume Imaging). Abdominal segment A6 was imaged for GMA and segments A1-2 for Arp3 GFP. Larvae expressing Arp3 GFP were kept at 29°C until imaging. To determine a point of new branch formation, newly forming branches were observed during the time-lapse images and their first appearance was set as the time-point 0 s. Accumulation of GMA was measured at the time point −30 s as the maximum intensity of a round region of interest (ROI) of 7 pixels diameter. A threshold for background GMA signal was set by averaging 20 randomly set ROIs along dendrite branches for each video (n=11). The spatiotemporal analysis of the Arp3-GFP signal along individual branches is described in detail in the supplementary Materials and Methods.

All flies expressing photoactivatable constructs were raised under standard conditions but in full darkness. They carried: ppk-Gal4 UAS-mCherry and UAS-PA Rac1 CA, UAS-PA DN Rac1, UAS-LI Rac1 or UAS-PA Rac1 CA and UAS Wave RNAi. Living late third instar larvae were prepared for live confocal imaging under a dimmed red light. Photoactivation of PA-Rac1 and live-cell imaging were performed using a Carl Zeiss LSM5 confocal microscope. To photoactivate the Rac1 constructs (CA, LI or DN), the indicated region of interest (ROI) of the dendritic tree of a ddaC neuron in living L3 larvae was repeatedly illuminated using a 5 mW 445 nm laser at 0.2% power for 20 s in a photobleaching mode and the membrane marker mCherry was imaged in the red channel every minute (Kato et al., 2014; Wu et al., 2009).

Quantification of a dendritic tree complexity was performed by manual tracing of dendritic branches from maximum projections of original image stacks in ImageJ using the NeuronJ plug-in and FIJI. Dendrites were separated in specific clusters (Figs S1, S3) according to their morphology and distance from the cell soma. Density was measured as number of branches of a given order divided by total length of lower order branches. Bar graphs and tables express the data of dendritic length and number as mean±s.e.m. Unless otherwise noted, data were analysed for statistical significance using two-tailed unpaired Student t-test for comparing two groups or one-way ANOVA for more than two groups at P<0.05. If normal distribution of the dataset was confirmed using the Shapiro-Wilk normality test, the ANOVA Bonferroni's post test was used for multiple comparison, otherwise the Krustal-Wallis test with a Dunn's post test was used. Asterisks in graphs and tables represent the significance of P-values comparing indicated groups with control (*P<0.05; **P<0.01; ***P<0.001; ****P<0.0001). For time-lapse analysis, every time point was analysed separately. At least five animals per genotype were used for quantification.

Electron microscopy experiments were obtained with primary neuronal culture of larval mushroom body neurons. Dissociated neuronal cultures from Drosophila brain tissue were prepared as described previously (Kraft et al., 2006) and cultured on formvar-coated gold finder EM grids for 48 h at 25°C. The cells were examined and preselected on an inverted Zeiss Observer epifluorescence microscope and extracted at room temperature with cytoskeleton buffer [10 mM MES buffer, 150 mM NaCl, 5 mM EGTA, 5 mM glucose, 5 mM MgCl₂ (pH 6.8)] containing 0.25% glutaraldehyde and 0.5% Triton for 40 s. For electron tomography, neurons were fixed with 2% glutaraldehyde in the same buffer for 15 min and grids were stained with 6-8% sodium silicotungstate (SST), including 1 mg/ml phalloidin. Electron tomography was obtained using an FEI Tecnai F30 Helium (Polara) microscope, and actin filaments were manually tracked using IMOD, as previously described (Mueller et al., 2014; Vinzenz et al., 2012).

We dedicate this manuscript to the family of our dear colleague Anastasia Tatarnikova, whom we prematurely lost during the course of this work. We are grateful to Dr Fikret-Gürkan Agircan for his great experimental support during revision. We thank J. Zallen for sharing fly lines and the Bloomington Stock Center for providing reagents. We are grateful to Rüdiger Klein (MPI of Neurobiology) for hosting A.T., to Michael Sixt (IST Austria) for his support, and to Frank Bradke and Peter Soba for helpful discussions and comments on the manuscript. Members of the Tavosanis lab provided many helpful comments.