Coordinate control of terminal dendrite patterning and dynamics by the membrane protein Raw

The directional flow of information in the nervous system relies on the compartmentalization of neurons. On a basic level, a neuron has two compartments apart from the soma: dendrites, which receive inputs, and axons, which transmit signals. To meet their respective functions, axons and dendrites have distinct morphological and molecular properties. For example, the microtubule cytoskeleton is organized differently in axons and dendrites, allowing distinct modes of trafficking (Baas et al., 1988; Horton and Ehlers, 2003). Beyond this basic level of compartmentalization, neurons display extreme diversity in axon and dendrite morphology, and both axons and dendrites contain structurally and functionally distinct subdomains (Katsuki et al., 2011; Masland, 2004).

Several lines of evidence support the existence of compartments within dendrites. First, some neurons have morphologically distinct dendrites. For example, mammalian olfactory bulb mitral cells extend a tufted primary dendrite radially and structurally and functionally distinct lateral dendrites horizontally (Imamura and Greer, 2009). Likewise, apical/basal dendrites of hippocampal neurons and ipsilateral/contralateral dendrites of motoneurons are morphologically and biophysically distinct. Second, specialized structures are asymmetrically distributed in many dendrites. Notable among these is the dendritic spine, an isolated compartment that is electrically and biochemically distinct from the rest of the dendrite arbor, and dendritic spines likewise have distinctive microdomains (Chen and Sabatini, 2012; Yuste, 2013). Many organelles are selectively deployed in dendrites, including a satellite secretory pathway containing endoplasmic reticulum and Golgi outposts; the number and location of these organelles locally influences dendrite growth and dynamics (Aridor et al., 2004; Gardiol et al., 1999; Horton and Ehlers, 2003; Ye et al., 2007). In highly branched dendrite arbors, for example cerebellar Purkinje neurons and insect sensory neurons, major dendrites and terminal dendrites have distinct cytoskeletal compositions and growth properties (Fujishima et al., 2012; Jinushi-Nakao et al., 2007). Finally, dendrite arbors often contain functionally distinct domains as well. For example, the proximal-distal compartmentalization of chloride co-transporters underlies directional selectivity in starburst amacrine cells (Gavrikov et al., 2006). Whereas axon/dendrite compartmentalization can be attributed to neuronal polarization, the developmental origin of dendrite subcompartments is less well understood.

Drosophila peripheral nervous system (PNS) class IV dendrite arborization (C4da) neurons have highly branched dendrite arbors, consisting of major dendrites emanating radially from the soma and terminal dendrites that fill in the space in the receptive field (Grueber et al., 2002). Main branches and terminal arbors have distinct growth properties in these neurons, with terminal dendrites exhibiting dynamic growth and containing a cytoskeleton largely devoid of microtubules (Grueber et al., 2002; Jinushi-Nakao et al., 2007), suggesting that different cellular programs pattern main dendrites and terminal arbors. To identify the developmental bases for compartment-specific patterning in these dendrites, we used a genetic screen to identify mutations that affect distinct dendritic compartments. From this screen, we identified mutants that selectively affected terminal dendrites, including their placement along the proximal-distal axis and their patterning, suggesting that distance from the soma and branch type (major or terminal dendrite) are two key pieces of positional information in the patterning of dendrite arbors. Mutations in raw were unique in that they simultaneously affected multiple aspects of terminal dendrite patterning, suggesting that raw coordinately controls multiple aspects of terminal dendrite growth. Indeed, we found that Raw regulates terminal dendrite adhesion and elongation via distinct pathways, the former involving the Trc kinase and the latter involving cytoskeletal remodeling and the RNA-binding protein AGO1. Thus, Raw appears to be a crucial component of a spatially localized program controlling terminal dendrite patterning.

To identify the developmental bases for compartment-specific patterning in dendrites, we used mosaic analysis with a repressible cell marker (MARCM) to screen for mutations that differentially affected different regions of C4da dendrite arbors (Lee and Luo, 1999). From this screen, we identified two phenotypic groups that define different levels of organization of these dendrite arbors. First, we identified mutations that differentially affected terminal dendrite growth/distribution along a proximal-distal axis relative to the cell body (supplementary material Fig. S1). Most commonly, as in jj329 mutant neurons, mutants in this group caused exuberant terminal dendrite branching in proximity to the soma and reduced terminal branching distally (supplementary material Fig. S1). Notably, this phenotype is similar to dendrite phenotypes caused by mutations in Dlic (Satoh et al., 2008; Zheng et al., 2008), and thus we hypothesize that organization along the proximal-distal axis involves spatial information conferred by microtubule-based transport. Consistent with this hypothesis, jj329 is an allele of CG12042, which encodes Dynactin subunit 4, and several other mutants in this group impinge on microtubule-based processes. Second, we identified a large group of mutants that preferentially affect terminal dendrite patterning without affecting major dendrites. Within this group, we identified mutants that affected the number, length and/or orientation of terminal dendrites. For example, jj472 had simplified terminal arbors; jj835 mutants were devoid of higher order branches; and jj599 mutants affected terminal dendrite self-avoidance (supplementary material Fig. S1). The identification of mutants that show deficits in the formation of terminal dendrite arbors (jj835) and distinct aspects of terminal patterning (length, jj472; avoidance, jj599) suggests that multiple pathways function at different levels of control to pattern terminal dendrites.

We anticipated that we would identify mutations that selectively affect primary dendrite growth; however, we have not identified any such mutations, possibly indicating that primary dendrite growth is a prerequisite for terminal dendrite branching in these neurons (Parrish et al., 2006). Nevertheless, results from our screen suggest that dendrite arbor development in C4da neurons is compartmentalized in at least two ways: growth of terminal dendrites along a proximal-distal axis emanating from the cell body, and in patterning of main and terminal dendrites.

Among the mutants that preferentially affected terminal dendrites, jj102 was of particular interest because it affected multiple aspects of terminal dendrite patterning, suggesting that it might coordinately regulate multiple signaling pathways that influence terminal dendrites (Fig. 1). jj102 mutant C4da neurons exhibited four characteristic terminal dendrite defects. First, major dendrites were decorated with short, mostly unbranched terminal dendrites, whereas major dendrites in control neurons had extensively branched terminal arbors. Thus, although the total branch number and distribution along the proximal-distal axis were comparable in jj102 mutants and wild-type controls (Fig. 1E, left; supplementary material Fig. S2A), jj102 mutant neurons exhibited reduced complexity in distal regions of the arbor, which normally contain highly branched dendrites (Fig. 1E, middle and right). Consequently, the average path length from soma to dendrite tip was significantly shorter in jj102 mutants (Fig. 1F). Second, the average terminal dendrite length was significantly reduced in jj102 mutants (Fig. 1G); consequently, total dendrite length was also reduced (supplementary material Fig. S2B). Third, terminal dendrite orientation was altered: both the local branch angle and three-dimensional (3D) placement of terminal dendrites were defective in jj102 mutants. In wild-type neurons, branch angles of terminal dendrites approached 90°, contributing to the radial arrangement of dendrite arbors, but terminal dendrites of jj102 mutants branched at more acute angles (Fig. 1H). As a result, terminal dendrites often grew in proximity to primary dendrites, rather than growing away from them. Fourth, dendrite-dendrite crossing events were common in jj102 mutant neurons; whereas dendrites of wild-type neurons are confined to a two-dimensional (2D) area and rarely cross over one another (Fig. 1A′) (Emoto et al., 2004; Han et al., 2012; Kim et al., 2012), terminal dendrites in raw mutant neurons occupy larger 3D areas and frequently cross over one another (Fig. 1B′,I). Thus, jj102 defines a potential point of convergence for the developmental control of multiple aspects of terminal dendrite patterning.

Homozygous jj102 mutants arrest as embryos with dorsal closure defects, and jj102 mapped to an interval on chromosome 2 containing one known regulator of dorsal closure, raw (Nusslein-Volhard et al., 1984). raw encodes a novel protein that modulates cell-cell signaling in dorsal closure (Bates et al., 2008; Bauer Huang et al., 2007; Byars et al., 1999) and Cadherin-based interactions between somatic gonadal precursor cells and germ cells (Jemc et al., 2012). Consistent with jj102 affecting raw function, chromosome deficiencies that span raw and raw reduction-of-function alleles (raw¹, raw^k01021) failed to complement jj102 lethality. Further, an independently derived raw allele (raw^k01021) phenocopied the jj102 dendrite defects (Fig. 1C,F-I). Finally, jj102 carries a nonsense mutation in raw (CAA→TAA, Q777Stop in Raw-short), which should lead to a premature stop prior to a predicted transmembrane domain (Fig. 1J) (Jones et al., 1994).

To establish that jj102 dendrite defects were caused by loss of raw function, we used MARCM to resupply functional raw to jj102 mutant neurons. To this end, we generated a transgene encoding a GFP-Raw fusion protein (UAS-GFP-raw, Fig. 1J) using the short isoform of raw, which has been implicated in PNS development (Prokopenko et al., 2000). UAS-GFP-raw fully rescued the jj102 dendrite defects (Fig. 1D,F-I). By contrast, overexpressing UAS-GFP-raw in wild-type C4da neurons had no effect on terminal branch patterning (data not shown), and thus raw is necessary for terminal dendrite patterning but not sufficient to promote exuberant growth. Altogether, these results demonstrate that raw functions cell-autonomously to regulate terminal dendrite orientation and elongation, and that Raw-short can support dendrite development; we have not investigated the function of other isoforms in the PNS.

Homozygous raw mutants arrest before PNS dendrite outgrowth begins and raw influences the development of several embryonic tissues (Byars et al., 1999; Jack and Myette, 1997; Jemc et al., 2012; Kania et al., 1995). Therefore, to ascertain the developmental origin of the raw mutant dendrite defects, we monitored dendrite phenotypes of raw mutant MARCM clones during early larval development. At each stage that we observed (clones become visible in late first instar larvae), raw mutant neurons exhibited characteristic terminal branching defects while major dendrites grew comparably to controls (supplementary material Fig. S2). Thus, raw is required for dendrite development from early stages of larval development and appears to act primarily on terminal dendrites.

The raw mutant dendrite branching defects could reflect a failure in terminal branch elongation, increased branch retraction, or some combination of both. To distinguish between these possibilities, we monitored terminal dendrite dynamics in wild-type or raw mutant C4da MARCM clones using time-lapse microscopy. Initially, we chose a 4-h window beginning at 96 h after egg laying (AEL), when larval growth is nearly complete and C4da dendrite arbors are no longer expanding (Parrish et al., 2009). During this time-lapse most terminal dendrites exhibit dynamics with equivalent rates of branch extension and retraction (Fig. 2A,C,D). By contrast, a significantly larger proportion of terminals exhibited dynamics in raw mutant neurons, with dynamic branches retracting more frequently than they grew (Fig. 2B-D). Additionally, the extent of dynamics was significantly increased in raw mutants (3.8 µm on average, compared with 1.8 µm for control neurons) (Fig. 2E). Thus, raw appears to regulate terminal dendrite stability.

Our developmental analysis of raw mutant neurons suggested that it is required for the stabilization of terminal branches throughout development. We therefore monitored dendrite dynamics over an earlier time-lapse when arbors are growing rapidly and adding new terminal dendrites. Between 72 and 76 h AEL, control neurons exhibited an increased frequency and a different distribution of dynamics: terminal growth occurred almost twice as frequently as retraction during the early interval, whereas growth and retraction occurred at equivalent rates during the late interval (Fig. 2C,D). As in the late time-lapse, raw mutants exhibited significantly more dynamic behavior than wild-type controls between 72 and 76 h AEL (Fig. 2C). Notably, whereas the dynamic behavior of control neurons changed during development, the rate of dynamics and the frequency of growth/retraction in raw mutant clones remained similar throughout both time-lapses, suggesting that terminal dendrite growth is constantly offset by retraction in raw mutants (Fig. 2D). Thus, raw affects terminal branch stability; at each time point examined, terminal dendrites of raw mutants were more dynamic than those of controls, with growth matched by retraction. Hence, terminal dendrites of raw mutants rarely branch or elongate, preventing the elaboration of terminal arbors.

To gain insight into Raw control of terminal dendrite dynamics, we monitored the intracellular distribution of GFP-Raw in raw mutant C4da neuron clones. GFP-Raw rescues the raw^jj102 dendrite defects, and thus we reasoned that GFP-Raw localization in these clones is likely to reflect the site of action of Raw. In dendrites, GFP-Raw is concentrated in puncta, most of which localize to branch-points along major dendrites (Fig. 3A, green arrowheads), consistent with Raw locally regulating terminal dendrite patterning. However, GFP-Raw was rarely detected in terminal dendrites, suggesting that Raw might function at branch-points to stabilize nascent dendrites or recruit other factors that function in the terminals.

Terminal dendrites rapidly turn over in raw mutants; therefore, we monitored whether GFP-Raw localization correlated with terminal branch stabilization. Using time-lapse microscopy, we monitored dendrite dynamics during an 18-h interval in GFP-Raw-expressing raw^jj102 mutant C4da MARCM clones and examined whether the presence of GFP-Raw puncta at branch-points was associated with branch stabilization (Fig. 3B). We sampled over 500 dynamic terminal dendrites from six neurons and found that the growth behavior of terminal dendrites originating in branch-points containing GFP-Raw puncta [χ²=23.80, degrees of freedom (d.f.)=2, P<0.01] or lacking GFP-Raw puncta (χ²=14.90, d.f.=2, P<0.01) was significantly different from that of the population of terminal dendrites as a whole. Specifically, branches associated with GFP-Raw puncta were more likely to grow or remain the same length and less likely to retract, whereas branches emanating from branch-points lacking GFP-Raw were more likely to retract, consistent with Raw promoting terminal dendrite stabilization/elongation or marking stabilized terminals (Fig. 3C).

raw encodes a protein with a putative C-terminal transmembrane domain (Jones et al., 1994), and thus raw might encode a membrane protein that coordinately regulates multiple aspects of terminal dendrite development in response to extracellular cues. We therefore investigated whether Raw is membrane associated. We fractionated Drosophila S2 cells expressing GFP-Raw and probed the cellular fractions with antibodies to GFP, α-Tubulin (cytoplasmic marker) and Robo (membrane marker). Indeed, GFP-Raw was present in the cytoplasmic and membrane fractions, demonstrating that some of the GFP-Raw was membrane associated (Fig. 4A). Similarly, the C. elegans Raw ortholog (OLRN-1) is membrane associated, but OLRN-1 has multiple putative transmembrane domains that are not present in Raw (Bauer Huang et al., 2007).

Our membrane topology prediction suggested that the N-terminal ∼700 amino acids of Raw constitute an extracellular domain (ECD), whereas the short C-terminus is located intracellularly (Jones et al., 1994). We therefore examined Raw distribution and topology using surface staining of ‘rescued’ raw mutant MARCM clones expressing GFP-Raw, reasoning that GFP should be surface exposed if Raw is targeted to the plasma membrane. As a negative control, we first assayed for surface-exposed GFP in MARCM clones expressing EB1-GFP, a microtubule-associated protein that should not be surface exposed. Although EB1-GFP was dispersed throughout the dendrite arbor, no GFP surface staining was evident, demonstrating that the plasma membrane was not permeabilized by our staining protocol (Fig. 4B). By contrast, C4da neurons expressing GFP-Raw exhibited punctate GFP surface staining (Fig. 4C), and these puncta frequently occurred at dendrite branch-points (green arrowheads). Similarly, GFP was surface exposed in S2 cells expressing GFP-Raw (data not shown). We conclude that Raw can associate with membranes and that the N-terminal domain can be exposed to the extracellular environment, with the potential to locally influence terminal dendrite development in response to extracellular cues.

We next set out to characterize downstream pathways by which Raw regulates terminal dendrite patterning. In dorsal closure and gonadal ensheathment, raw negatively regulates Jnk signaling to modulate cell-cell interactions (Byars et al., 1999; Jemc et al., 2012). We hypothesized that raw similarly regulates Jnk signaling in C4da neurons to pattern terminal dendrites. We therefore monitored effects of raw on phospho-Jnk accumulation, on expression of the Jnk pathway target puckered, and on AP-1 reporter expression in C4da neurons (Chatterjee and Bohmann, 2012), and in each case we observed no effect (supplementary material Fig. S3). Likewise, modulating Jnk activity had no apparent effect on C4da dendrite patterning. We therefore conclude that raw is likely to signal through distinct pathways in the epidermis and C4da neurons, with raw function in dendrite patterning being largely independent of Jnk signaling.

To identify genes that function with raw to regulate terminal dendrite patterning, we assayed for genetic interactions between raw and known regulators of terminal dendrite development, including genes in the Trc signaling pathway, that regulate dendrite-dendrite repulsion and terminal dendrite adhesion: turtle (tutl), which regulates dendrite-dendrite repulsion; Dscam, which regulates dendrite self-avoidance; and myospheroid (mys), which is required for dendrite-extracellular matrix (ECM) interactions (Emoto et al., 2004; Han et al., 2012; Kim et al., 2012; Koike-Kumagai et al., 2009; Long et al., 2009; Matthews et al., 2007; Soba et al., 2007). On its own, heterozygosity for mutation in raw or any of the other genes had no significant effect on the number of terminal dendrite crossing events, but larvae doubly heterozygous for mutations in raw and Trc signaling pathway genes exhibited significant increases in dendrite-dendrite crossing (Fig. 5), suggesting that Raw functions together with the Trc pathway to regulate terminal dendrite patterning. Similar to raw, trc regulates multiple aspects of terminal dendrite patterning, namely terminal dendrite number and adhesion, and these activities are genetically separable (Emoto et al., 2004). However, we observed significant changes in dendrite-dendrite crossing but not dendrite number in raw trc double heterozygotes, suggesting that the different functions of raw in terminal branching are likewise genetically separable, with terminal dendrite adhesion involving trc.

Next, we assayed for physical association between Raw and Trc. We co-expressed GFP-Raw and Trc-mCherry-HA in S2 cells and examined whether the epitope-tagged proteins co-immunoprecipitated. Indeed, Raw co-immunoprecipitated with Trc and vice versa (Fig. 6A), and a truncated Raw protein containing the transmembrane domain and intracellular domain (TM-ICD) interacted with Trc, whereas the Raw ECD did not (Fig. 6B). Trc activation is potentiated by membrane association (Hergovich et al., 2005; Koike-Kumagai et al., 2009); thus, our findings that Raw is a membrane-associated protein, that Raw physically associates with Trc, and that the Raw TM-ICD mediates this interaction, suggested that Raw might play some role in Trc phosphorylation/activation.

To assess the functional relevance of the Raw-Trc interaction, we developed antibodies that allowed us to monitor Trc phosphorylation on threonine 449 (T449), the residue associated with maximal kinase activation (Fig. 6C; supplementary material Fig. S4) (Tamaskovic et al., 2003). First, we examined the relationship between Raw and Trc activation in S2 cells. In control or GFP-raw-transfected S2 cells, Trc P-T449 was present at low levels (Fig. 6D; for overexposed blot see supplementary material Fig. S4), suggesting that Raw is not sufficient to promote Trc phosphorylation. To examine whether Raw could facilitate Trc phosphorylation, we treated S2 cells with okadaic acid (OA) and monitored the effect of Raw on Trc P-T449 accumulation. As expected, OA treatment induced Trc phosphorylation (Koike-Kumagai et al., 2009; Millward et al., 1999), and Raw overexpression led to a 62% increase in OA-induced Trc phosphorylation (Fig. 6D). Thus, Raw can potentiate Trc activation in S2 cells. Further, the Raw TM-ICD was sufficient to potentiate Trc phosphorylation whereas the ECD exhibited no activity in this assay (Fig. 6E), suggesting that Raw enhances Trc phosphorylation by promoting Trc membrane association/proximity.

To examine whether Raw influences Trc activity in vivo, we assayed effects of raw mutation on Trc phosphorylation in C4da neurons. In the larval PNS, Trc P-T449 immunoreactivity was present at high levels in C4da dendrites, where it appeared to accumulate in puncta (Fig. 6F, raw^jj102/+ heterozygote), consistent with prior reports that Trc is required for dendritic tiling in these neurons (Emoto et al., 2004). Trc P-T449 was also detectable in Class III da neurons, albeit at much lower levels, and trc is required for dendrite morphogenesis in these neurons as well (Fig. 6F, double arrowheads). Finally, Trc P-T449 was present at high levels in the axons of da neurons (Fig. 6F, bracket). By contrast, in raw mutant C4da MARCM clones, Trc P-T449 levels were substantially reduced to levels comparable to those in Class III da neurons (Fig. 6G). Thus, raw appears to potentiate Trc phosphorylation in C4da neurons as well as in S2 cells.

If raw functions through Trc to regulate dendrite-dendrite crossing, we reasoned that ectopic Trc activation in raw mutant neurons should mitigate the raw mutant dendrite crossing phenotype. This is indeed what we found. Although overexpressing wild-type Trc in raw mutant neurons had a dominant-negative effect (Fig. 6H,J), overexpression of myristoylated Trc, which is membrane targeted and hence constitutively active (Koike-Kumagai et al., 2009), significantly reduced dendrite-dendrite crossing in raw mutant neurons (Fig. 6I,J). Likewise, overexpression of a Trc phosphomimetic (UAS-trc T449E) partially suppressed the raw mutant dendrite-dendrite crossing defect (Fig. 6J). Thus, the raw mutant dendrite-dendrite crossing defect is likely to be caused, in part, by deficits in Trc activation, which Raw may potentiate by promoting Trc membrane association.

Trc can modulate dendrite adhesion to the ECM (Han et al., 2012); therefore, we investigated whether Raw can likewise modulate adhesion. We compared the adhesion of control or GFP-raw-expressing S2 cells to different ECM components (Fig. 7A). GFP-raw expression significantly enhanced S2 cell adhesion to collagen and, to a lesser degree, fibronectin (Fig. 7B). Since Raw can modulate Trc activity, and one output of Trc is cell adhesion (Han et al., 2012), we tested whether Raw promotes adhesion in a Trc-dependent manner. Indeed, a dominant-negative version of Trc (Trc K112A; Emoto et al., 2004) abrogated the ability of Raw to enhance adhesion to collagen (Fig. 7C).

Since Raw promotes S2 cell adhesion to collagen together with Trc, and Trc promotes Integrin-based attachment of C4da dendrites to a collagen-rich ECM (Han et al., 2012), we examined whether defects in Integrin-based adhesion contribute to the raw mutant dendrite phenotype. If raw modulates dendrite adhesion, we reasoned that increasing dendrite-ECM attachment should suppress the raw mutant dendrite-dendrite crossing defects, as Integrin overexpression suppresses dendrite-dendrite crossing defects of Trc pathway mutants (Han et al., 2012). To test this prediction, we assayed the effects of neuronal overexpression of Integrins (UAS-mys+UAS-mew) on terminal dendrite patterning in raw mutant C4da MARCM clones. Integrin overexpression suppressed dendrite-dendrite crossings in raw mutant neurons (Fig. 7F), suggesting that dendrite-ECM attachment is compromised in raw mutants, and significantly increased terminal branch number (Fig. 7G), which is likely to be the result of stabilizing the exuberant, short-lived branches found in raw mutant C4da neurons. Indeed, Integrin overexpression altered raw mutant terminal dendrite dynamics, increasing the fraction of stable terminals (Fig. 7H). However, Integrin overexpression did not rescue the raw mutant terminal dendrite elongation defect (Fig. 7I), suggesting that raw regulates branch stabilization and elongation via distinct mechanisms, with the former involving Trc.

Raw modulates F-actin assembly in cuticular hairs (Blake et al., 1999), and we noted that raw cell-autonomously regulated the formation of Actin-rich protrusions in oenocytes (supplementary material Fig. S5), suggesting that changes in Actin assembly might contribute to the dendrite extension defects of raw mutants. To investigate this possibility, we monitored F-actin distribution using GMA-GFP in wild-type and raw mutant C4da MARCM clones (supplementary material Fig. S6) (Bloor and Kiehart, 2001). In wild-type C4da dendrites, GMA-GFP was evenly distributed throughout the arbor with occasional concentrations at branch-points and in terminal dendrites. By contrast, terminal dendrites in raw mutant C4da neurons exhibited substantially higher levels of GMA-GFP than major dendrites. Thus, raw mutant terminal dendrites appear to be enriched in F-actin, which is likely to contribute to the dynamic behavior of these terminals. We examined whether Raw physically interacts with Actin or Tubulin, but we were unable to detect either Actin or Tubulin in Raw immunoprecipitates (data not shown). We also found that raw had little effect on the dendritic microtubule cytoskeleton visualized with Tubulin-GFP (supplementary material Fig. S6). Thus, we conclude that Raw indirectly regulates cytoskeletal composition to influence terminal dendrite elongation. Consistent with Raw regulating branch elongation independently of the trc/fry pathway, we did not observe GMA-GFP accumulation in terminal dendrites of fry mutants (supplementary material Fig. S7).

If indeed raw regulates terminal dendrite adhesion and elongation via distinct pathways, we reasoned that mutations in the signaling pathway(s) involved in branch elongation would selectively enhance the dendrite elongation phenotype of raw mutant neurons. To test this hypothesis, we screened our collection of terminal branching mutants for genetic interactions with raw. We identified one mutant (jj472) that genetically interacts with raw to affect terminal dendrite length and higher order branching without affecting the 3D orientation of terminal dendrites (Fig. 8A-C), further demonstrating that raw functions in terminal dendrite adhesion and elongation are genetically separable. Similar to raw mutant neurons, raw jj472 double heterozygotes had shorter terminal dendrites and less complex terminal arbors; thus, the mean path length from the soma to branch endings was reduced in raw jj472 double heterozygotes, as in raw mutant neurons (data not shown). On its own, homozygosity for jj472 caused a significant decrease in terminal branch number and length, but only a modest increase in dendrite-dendrite crossing (Fig. 8D,E). jj472 is loss-of-function allele of AGO1, which encodes an RNA-binding protein involved in miRNA-mediated translational repression (Förstemann et al., 2007; Kataoka et al., 2001), as jj472 fails to complement AGO1^k00208 and AGO1^k08121. jj472 carries a premature stop codon (Q319Stop) that truncates AGO1 before the catalytic domain (Hock and Meister, 2008), and AGO1 RNAi and AGO1^k08121 phenocopy the dendrite defects of jj472 (Fig. 8F,G; data not shown).

Altogether, our results support a model in which raw regulates terminal dendrite adhesion/stability and elongation by distinct pathways (Fig. 8J). Knockdown of one of these pathways causes terminal dendrite elongation defects (AGO1-RNAi, Fig. 8G) or dendrite-dendrite crossing defects (trc-RNAi, Fig. 8H), whereas knockdown of both pathways has an additive effect, resulting in dendrite defects similar to those of raw mutants (Fig. 8I). We have not yet characterized the raw-AGO1 interaction in detail, as AGO1 overexpression causes severe patterning defects and occasional neuron death, and we have been unable to generate double-mutant MARCM clones, so their respective functions in dendrite elongation remain to be determined. One possibility is that Raw interacts with AGO1 to influence the local translation of key effectors of terminal dendrite growth, as miRNAs are known to regulate local translation in dendrites in some neurons (Vo et al., 2010) and several factors with roles in local translation affect terminal dendrite growth in C4da neurons (Olesnicky et al., 2014).

Although the concept of positional information was first applied to embryonic development (Wolpert, 1969), intracellular positional information governs morphogenesis of individual cells as well. For example, positioning the nucleus at the cell center and growth zones at the cell periphery depends on positional information from the microtubule cytoskeleton in Schizosaccharomyces pombe (Bähler and Pringle, 1998; Castagnetti et al., 2007; Hagan and Yanagida, 1997). Several lines of evidence support the existence of distinct subcompartments in axons and dendrites, but the forms of intracellular positional information and the coordinate systems that guide the development of these subcompartments have not been extensively characterized. Results from our screen and other studies suggest that at least two types of positional information govern C4da dendrite patterning. First, terminal branch distribution along the proximal-distal axis depends on microtubule-based processes; perturbing microtubule-based transport leads to a distal-proximal shift in the distribution of terminal dendrites in C4da arbors (Satoh et al., 2008; Zheng et al., 2008). Interestingly, modulating the activity of the F-actin nucleator Spire also affects terminal dendrite positioning along the proximal-distal axis (Ferreira et al., 2014), suggesting that multiple pathways contribute to the fidelity of branch placement. Second, terminal dendrites rely on dedicated programs that may act locally to regulate terminal dendrite patterning. Our observation that different pathways regulate different aspects of terminal dendrite development suggests that multiple signaling systems exist for the local control of dendrite growth.

We identified raw as a key regulator of terminal dendrite patterning. raw encodes a membrane protein that accumulates at branch-points and coordinately regulates terminal dendrite adhesion/stability via a pathway that involves Trc and terminal dendrite elongation via a pathway that is likely to involve cytoskeletal remodeling and AGO1. Raw therefore provides a potential point of integration for external signals that regulate these downstream growth programs. These pathways could be responsive to the same signal – for example, Raw association with an extracellular ligand or a co-receptor – or could be spatially/sequentially segregated. Identification of additional raw-interacting genes should help clarify the architecture of these signaling pathways.

Raw regulates cell-cell signaling (Bates et al., 2008; Bauer Huang et al., 2007; Byars et al., 1999; Jemc et al., 2012), and in gonad morphogenesis Raw modulates Cadherin-based interactions between somatic gonadal precursor cells and germ cells, in part by localizing Armadillo to the cell surface (Jemc et al., 2012). Likewise, our data support a role for Raw in promoting Trc activation by localizing Trc to the plasma membrane. Thus, one plausible model for Raw function in dendrite development is that it interacts with an extracellular signal, which might be a component of the ECM or a cell surface protein on epithelial cells, and signals together with a co-receptor to stimulate downstream pathways for adhesion and cytoskeletal remodeling. Several analogous signaling systems involving interactions with the epidermis that influence terminal dendrite or sensory axon patterning have been described (Chiang et al., 2011; Dong et al., 2013; Han et al., 2012; Kim et al., 2012; Salzberg et al., 2013), but how many of these signaling systems are at work in a given neuron, and how Raw interfaces with other signaling pathways, remain to be determined.

Although Raw has no obvious vertebrate counterpart, stretches of the ECD bear similarity to mucins and leucine-rich repeat proteins, one of which might serve an analogous function. Moreover, components of both downstream signaling pathways that we identified are conserved in vertebrates and play known roles in dendrite patterning, including roles in the local control of dendrite growth: the Trc orthologs NDR1/2 (STK38/STK38L) regulate aspects of dendrite branch and spine morphogenesis (Ultanir et al., 2012), and Argonaute proteins mediate miRNA-mediated control of dendrite patterning, in part through local effects on translation (Vo et al., 2010). Additionally, dendrites contain structures related to P-granules, and Argonaute proteins may influence local translation in P-granules as well (Cougot et al., 2008). Thus, versions of the Raw-regulated signaling pathways might control terminal dendrite patterning in vertebrates.

A list of alleles used in this study and details of the genetic screen are provided in supplementary material methods and Table S1.

Imaging was performed as described (Jiang et al., 2014). For time-lapse analysis, larvae were imaged at the indicated time, recovered to yeasted agar plates with vented lids, aged at 25°C, and imaged again.

Larval fillets were dissected and processed as described (Grueber et al., 2002) and stained with HRP conjugated with Cy5 (1:250; 123-175-021, Jackson ImmunoResearch), anti-mCD8 (1:100; MCD0800, Life Technologies), anti-GFP (1:500; A11122, Life Technologies), anti-β-gal (1:500; A11132, Life Technologies), anti-Trc-P-T449 (1:500; this study), and secondary antibodies from Jackson ImmunoResearch (112-225-003, 111-295-144; 1:250).

Antibodies from rabbits immunized with a KLH-conjugated phosphorylated peptide (CKDWVFINY-pT-YKRFE) were affinity purified with bead-conjugated phospho-peptide. Non-phospho-specific antibodies were removed by absorption against a non-phosphorylated version of the antigen (Yenzym). In Trc phosphorylation assays, total Trc levels were assessed with anti-Trc antibodies [1:1000; courtesy of K. Emoto; Koike-Kumagai et al. (2009)] and protein input was assessed with anti-Actin (1:5000; ab8224, Abcam) or anti-Histone H3 (1:2000; H0164, Sigma) antibodies.

Formaldehyde (4%) fixed fillets were washed/stained without detergent to prevent permeabilization. Rhodamine-conjugated secondary antibodies (1:250; 123-175-021; Jackson Immunoresearch) were used for surface staining to ensure that surface-exposed GFP could be differentiated from total GFP.

S2 cells were grown as described (Rogers and Rogers, 2008), transfected using Effectene (Qiagen), and treated with 100 nM okadaic acid (Sigma) for 30 min before harvesting, where indicated.

Two days post-transfection, cells were lysed in NP-40 buffer. One milligram of extract was incubated with primary antibodies to GFP (2 μg; A11120, Life Technologies) or HA (2 μg; 11867431001, Roche) for 2 h, followed by Protein-G Agarose (Roche) for 1 h. Beads were washed five times in lysis buffer and bound proteins were analyzed by SDS-PAGE and western blotting with antibodies to GFP (1:500; A11122, Life Technologies) or HA (1:2000; 11867431001, Roche).

Transfected S2 cells were plated for 1 h in a 6-well cell culture plate coated with collagen I, fibronectin, or laminin (Life Technologies). Cells were washed twice with PBS and counted in four fields of view, then washed three time with PBS and counted as before, from which we calculated an adhesion index as the ratio of GFP-positive cells before/after washing.

Transfected S2 cells were lysed in fractionation buffer [250 mM sucrose, 20 mM HEPES pH 7.4, 10 mM KCl, 1.5 mM MgCl₂, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, Complete protease inhibitor (Life Technologies)] with Dounce homogenization. Lysates were fractionated by centrifugation: 720 g for 5 minutes (nuclear), 10,000 g 5 for minutes (mitochondria) and 100,000 g for 1 h for membrane (pellet) and cytosolic (soluble) fractions. Fractions were analyzed by SDS-PAGE and western blotting with antibodies against GFP (1:500; A11122, Life Technologies), Robo and Tubulin (13C9 at 1:200, 12G10 at 1:2000, respectively; both from Developmental Studies Hybridoma Bank).

Raw-short was PCR amplified from cDNA clone GH23250 (Drosophila Genomics Resource Center, Bloomington, IN, USA) and cloned into pUAST (Drosophila Genomics Resource Center) with N-terminal EGFP derived from pEGFP-N1 (Clontech). Transgenics services were provided by BestGene.

Dendrite length, crossing and dynamics were analyzed as described previously (Jiang et al., 2014); for details, see supplementary material methods.

Differences between group means were analyzed by ANOVA with a post-hoc Dunnett's test; pairwise comparisons of group means were performed with Student's t-test. Binomial tests were used to evaluate whether terminal dynamics differed between control and raw mutant neurons (Fig. 2D). Chi-squared tests were used to evaluate whether the presence/absence of GFP-Raw at branch-points influenced terminal dynamics (Fig. 3C).

We thank the Bloomington Stock Center, Giovanni Bosco, Simon Collier, Kazuo Emoto, Eric Lai, Anthea Letsou, Peter Soba and Tadashi Uemura for fly stocks; Kazuo Emoto and the Developmental Studies Hybridoma Bank for antibodies; the Drosophila Genomics Resource Center for cDNA clones; Ashley Lau and Hannah Lampert for screening and stereology assistance; David Parichy, Michael Kim and Peter Soba for critical reading of the manuscript.