Actin assembly and non-muscle myosin activity drive dendrite retraction in an UNC-6/Netrin dependent self-avoidance response

Dendrite growth is constrained by a self-avoidance response that induces retraction but the downstream pathways that balance these opposing mechanisms are unknown. We have proposed that the diffusible cue UNC-6(Netrin) is captured by UNC-40(DCC) for a short-range interaction with UNC-5 to trigger self-avoidance in the C. elegans PVD neuron. Here we report that the actin-polymerizing proteins UNC-34(Ena/VASP), WSP-1(WASP), UNC-73(Trio), MIG-10(Lamellipodin) and the Arp2/3 complex effect dendrite retraction in the self-avoidance response mediated by UNC-6(Netrin). The paradoxical idea that actin polymerization results in shorter rather than longer dendrites is explained by our finding that NMY-1 (non-muscle myosin II) is necessary for retraction and could therefore mediate this effect in a contractile mechanism. Our results also show that dendrite length is determined by the antagonistic effects on the actin cytoskeleton of separate sets of effectors for retraction mediated by UNC-6(Netrin) versus outgrowth promoted by the DMA-1 receptor. Thus, our findings suggest that the dendrite length depends on an intrinsic mechanism that balances distinct modes of actin assembly for growth versus retraction.

Introduction Dendritic arbors are defined by the balanced effects of outgrowth which expands the structure versus retraction which constrains the size of the receptive field. Microtubules and filamentous actin (F-actin) are prominent dendritic components and have been implicated as key drivers of dendritic growth and maintenance by the finding that treatments that perturb cytoskeletal dynamics may also disrupt dendritic structure [1][2][3]. For many neurons, dendritic growth is highly exuberant with multiple tiers of branches projecting outward from the cell soma. Ultimately, growth may be terminated by external cues. For example, neurons with similar functions are typically limited to separate domains by tiling mechanisms in which mutual contact induces dendrite retraction. Similarly, the related phenomenon of self-avoidance is widely observed to prevent overlaps among sister dendrites arising from a single neuron [4]. Homotypic interactions between the membrane components Dscam and protocaderins can mediate the self-avoidance response [5][6][7]. Surprisingly, in some instances, soluble axon guidance cues and their canonical receptors are also required. For example, self-avoidance for the highlybranched Purkinje neuron depends on both protocadherins and also repulsive interactions between sister dendrites decorated with the Robo receptor and its diffusible ligand, Slit [8]. In another example, we have shown that UNC-6/Netrin and its cognate receptors, UNC-5 and UNC-40/DCC, mediate self-avoidance for PVD nociceptive neurons in C. elegans [9]. Although multiple cell-surface interactions are now known to trigger self-avoidance, downstream effectors that drive dendrite retraction in this mechanism are poorly understood.
The PVD neurons, one on each side of the body, build complex dendritic arbors through a series of successive 1˚, 2˚, 3˚and 4˚orthogonal branching events [10] [11][12][13]. Self-avoidance ensures that adjacent 3˚branches do not overgrow one another [9,11] (S1 Fig). Our previous results suggested a novel mechanism of self-avoidance in which UNC-40/DCC captures UNC-6/Netrin at the PVD cell surface and then triggers retraction by interacting with UNC-5 on the neighboring 3 o branch. UNC-40/DCC may also effect self-avoidance by acting in a separate pathway that does not involve UNC-6/Netrin [9]. Here we describe a cell biological model of dendrite retraction in the PVD self-avoidance response that depends on actin polymerization.
Our results show that UNC-34/Ena/VASP functions downstream of UNC-5 to mediate PVD self-avoidance. Genetic evidence detecting roles for the Arp2/3 complex and its upstream regulator, WASP/WSP-1, indicates that the formation of branched actin networks may contribute to dendrite retraction. A necessary role for actin polymerization is also indicated by the defective PVD self-avoidance response of mutants with disabled UNC-73/Trio or MIG-10/ lamellipodin. We show that PVD self-avoidance also requires NMY-1/non-muscle myosin II which we propose effects dendrite retraction in a contractile mechanism that drives the reorganization of the nascent actin cytoskeleton. The necessary role for myosin could explain how actin polymerization can result in shorter rather than longer dendrites in the self-avoidance response. Thus, we propose that UNC-6/Netrin triggers self-avoidance by simultaneously stimulating actin assembly and non-muscle myosin activity in 3˚dendrites.
Because dendrite growth depends on actin polymerization [24,25], our finding that the actin cytoskeleton is also necessary for dendrite self-avoidance points to different modes of actin polymerization for growth vs retraction. Recent work has shown that a multicomponent complex involving the PVD membrane proteins DMA-1 and HPO-30 drives dendritic growth by linking the PVD actin cytoskeleton to adhesive cues on the adjacent epidermis [26,27]. Dendrite elongation, in this case, depends on the WRC and TIAM-1, both of which are known to promote actin polymerization. In contrast, our work has revealed that effectors of actin polymerization that are required for 3˚branch self-avoidance (e.g., UNC-34/Ena/VASP, MIG-10/lamellipodin, UNC-73/Trio, Arp2/3 complex, WSP-1/WASP) are not necessary for 3˚dendrite outgrowth. Additionally, we show that UNC-6/Netrin signaling antagonizes the DMA-1-dependent mechanism of dendritic growth. These findings are significant because they suggest that overall dendrite length is defined by the relative strengths of opposing pathways that utilize separate sets of effectors to differentially regulate actin polymerization for either growth or retraction.

UNC-6/Netrin mediates PVD sister dendrite self-avoidance
To visualize PVD morphology, we utilized a previously characterized GFP marker driven by a PVD-specific promoter (PVD::GFP) (S1A and S1B Fig) [11]. Each PVD neuron adopts a stereotypical morphology characterized by a series of orthogonal junctions between adjacent branches: 1˚dendrites extend laterally from the PVD cell soma; each 2˚dendrite projects along the dorsal-ventral axis to generate a T-shaped junction comprised of two 3˚dendrites each with either an anterior or posterior trajectory; terminal 4˚dendrites occupy interstitial locations between the epidermis and body muscles. The net result of this branching pattern is a series of tree-like structures or menorahs distributed along the 1˚dendrite. In contrast to the highly branched dendritic architecture, a single axon projects from the PVD cell soma to join the ventral nerve cord [10,11,26]. Despite the complexity of this network, dendrites rarely overlap (< 5%) in the mature PVD neuron (S1A Fig) [9]. Self-avoidance depends on an active process in which sister PVD dendrites retract upon physical contact with one another (S1C Fig). In the wild type, 3˚branches arising from adjacent menorahs initially grow outward toward one another but then retract after mutual contact. The net result is a characteristic gap between the tips of 3˚dendrites in neighboring menorahs [11]. We have previously shown that 3˚dendrite self-avoidance depends in part on the diffusible cue UNC-6/Netrin and its canonical receptors, UNC-40/DCC and UNC-5 [9]. Mutations that disable unc-6, for example, result in a significant increase in the fraction of 3˚dendrites that fail to retract and remain in contact (S1B Fig, Fig 3D). Similar self-avoidance defects were observed for mutants of either unc-40 or unc-5. Our results are consistent with a model in which UNC-40 captures UNC-6 for contact with UNC-5 at the tips of adjacent 3˚dendrites (S1D Fig) [9]. Here we address the question of how the activated UNC-5 receptor then modifies the actin cytoskeleton to drive dendrite retraction.

Constitutively active UNC-5 delays dendrite outgrowth
We have proposed that UNC-6 initiates dendrite retraction by activating UNC-5 [9]. This model predicts that a constitutively active form of UNC-5 should retard net outgrowth of 3d endrites by triggering an opposing retraction mechanism. To test this idea, we attached a myristoylation sequence to the cytoplasmic domain of UNC-5 [27]. The resultant myristyolated UNC-5 intracellular domain (MYR::UNC-5) is effectively tethered to the cytoplasmic membrane which is sufficient to activate UNC-5-dependent downstream signaling [27]. Expression of MYR::UNC-5 in PVD results in substantially shorter 3˚dendrites in comparison to wild type at the L3 stage (Fig 1A-1C). This phenotype suggests that constitutive activation of UNC-5 impairs normal outgrowth of 3˚dendrites. We tested this idea by scoring the dynamic status of 3˚dendrites. To quantify movement, we measured the distance from a fiduciary point to the tip of a 3˚dendrite at intervals over time (delta t = 240 sec) (Fig 1E) (S1 and S2 Movies). Examples of these measurements are shown in Fig 1E. We observed that wild-type 3˚dendrites typically display saltatory movement with periodic bouts of extension and retraction that result in net elongation. In contrast, 3˚dendrites of PVD neurons expressing MYR:: UNC-5 grow more slowly and rarely show periods of active extension and retraction (Fig 1E  and 1F). The higher overall motility of 3˚dendrites in the wild type vs PVD::MYR::UNC-5 is evident from a comparison of the standard deviation of the net movement of 3˚dendrites in each strain (Fig 1F). Despite slower outgrowth, 3˚dendrites expressing MYR::UNC-5 eventually produce a normal-appearing PVD architecture by the late L4 stage (Fig 2A). MYR::UNC-5 lacks the extracellular UNC-6-binding domain ( Fig 1D) and thus should function independently of UNC-6. We confirmed this prediction by demonstrating that MYR::UNC-5 prevents the self-avoidance defect observed in an unc-6 mutant in L4 stage animals (Fig 2B-2D). We attribute this effect to the constitutive activation of dendrite retraction by MYR::UNC-5 that effectively prevents overgrowth of adjacent sister 3˚dendrites in an unc-6/Netrin mutant. Together, these results are consistent with the hypothesis that UNC-5 functions downstream of UNC-6/Netrin to trigger dendrite retraction in the self-avoidance response. Given the wellestablished role of UNC-5 as an UNC-6/Netrin receptor [28,29], we favor this model over the alternative hypothesis that UNC-5 functions in parallel to UNC-6/Netrin which is also consistent with these epistasis results.

UNC-34/Ena/VASP mediates self-avoidance in actin-containing PVD dendrites
To explore potential modifications of the PVD cytoskeleton as effectors of the UNC-6/Netrindependent self-avoidance response, we used live cell markers to visualize actin (PVD::ACT-1:: GFP) and microtubules (pdes-2::TBA-1::mCherry) in PVD [30]. We observed a robust ACT-1::GFP signal throughout the PVD neuron including the axon and all dendrites ( Fig 3A). In contrast, the TBA-1::mCherry signal was strongest in the PVD axon and 1˚dendrite. This result confirms an earlier report suggesting that dendritic microtubules are most abundant in the 1˚branch (Fig 3A) [30]. We thus considered the possibility that the self-avoidance response could depend on the regulation of actin dynamics. To test this idea, we examined the PVD self-avoidance phenotype for a mutation that disables the actin binding protein, Ena/ VASP. We selected Ena/VASP for this test because it has been shown to function as a cytoplasmic effector of actin polymerization downstream of unc-6/Netrin signaling [12,31,32]. In addition, the C. elegans Ena/VASP homolog, UNC-34 [33], is enriched in a gene expression profile of PVD neurons [11]. A strong loss-of-function allele of unc-34/Ena/VASP [34] alters overall PVD morphology (e.g., truncated or misplaced 1˚dendrite, fewer 2˚branches) ( [24,25]. PVD neurons in unc-34 mutant animals show the opposite effect of adjacent 3˚dendrites that fail to self-avoid and overgrow one another (Fig 3B and 3C). PVD-specific expression of mCherrytagged UNC-34 protein rescued this Unc-34 mutant phenotype suggesting that unc-34 function is cell-autonomous for the self-avoidance response ( Fig 3C). unc-34 mutant 1˚and 2b ranch defects were not rescued by PVD expression of UNC-34, however, which could be indicative of an additional UNC-34 function in nearby cell types that influence PVD morphogenesis ( Fig 3B, S2 Fig) [35,36]. Thus, our results suggest that UNC-34/Ena/VASP is selectively required in PVD neurons for dendrite retraction in the self-avoidance response but not for dendrite outgrowth. The previously established role of UNC-34/Ena/VASP as a downstream effector of UNC-5 [20] predicts that unc-34 should be epistatic to the MYR::UNC-5 gain-offunction phenotype. We confirmed this prediction by showing that the delayed outgrowth PVD dendrites in the PVD::MYR::UNC-5 strain is fully suppressed in unc-34 mutant backgrounds which display the Unc-34 self-avoidance defect ( Fig 3C). This result stands in striking contrast to our finding above that PVD::MYR::UNC-5 is epistatic to the Unc-6 self-avoidance phenotype (Fig 2B-2D). Thus, our genetic results are consistent with a model in which UNC-6/Netrin acts through UNC-5 to activate UNC-34/Ena/VASP for dendrite retraction in the self-avoidance response.

Mutations that disable actin-polymerizing proteins disrupt dendrite selfavoidance
Genetic tests of additional regulators of actin polymerization also revealed PVD self-avoidance phenotypes. Lamellipodin (Lpd) interacts with ENA/VASP at the tips of filopodia and lamellopodia to extend actin filaments [12,19]. Mutants of mig-10/Lpd display robust PVD self-avoidance defects (Fig 3B and 3E). Similarly, self-avoidance is disrupted by a mutation that eliminates the Rac GEF activity of UNC-73/Trio, a potent regulator of actin dynamics [15]. The predicted downstream function of UNC-73/Trio in the UNC-5 signaling pathway [27] is confirmed by our finding that the self-avoidance defects of unc-73/Trio, like unc-34/Ena/VASP mutants, are fully epistatic to the delayed outgrowth phenotype of constitutively active PVD:: MYR::UNC-5 ( Fig 3C). We note that both mig-10/Lpd and unc-73/Trio mutants also show other defects in PVD morphology (e.g., displaced 1˚dendrite, fewer 2˚branches) (S2A-S2C To test for potential roles for branched actin polymerization, we examined a null allele of wsp-1/WASP (Wiskott-Aldrich Syndrome protein) and observed defective PVD self-avoidance. If wsp-1/WASP functions in a common pathway with unc-6, then the self-avoidance defect of the double mutant, wsp-1; unc-6, should be no more severe than that of either wsp-1 or unc-6 single mutants. We confirmed this prediction by scoring PVD self-avoidance in wsp-1; unc-6 mutant animals ( Fig 3D). An independent study confirmed the role of wsp-1 in PVD self-avoidance. This work also reported, however, that wsp-1 does not enhance the self-avoidance defect of a parallel-acting pathway involving MIG-14/Wntless and thus concluded that wsp-1 functions downstream of MIG-14/Wntless [39]. Additional experiments are needed to resolve the question of whether WSP-1/WASP functions exclusively in the UNC-6/Netrin vs MIG-14/Wntless self-avoidance pathways.
Because WASP is known to activate the Arp2/3 complex [14,40], we also performed RNAi knock down of the conserved Arp2/3 component ARX-5/p21 and detected significant PVD self-avoidance defects ( Fig 3E). The canonical roles of WASP and the Arp2/3 complex suggest that self-avoidance depends on the formation of branched F-actin networks. Recent studies have shown that actin and actin regulators like TIAM-1/GEF and components of the Wave Regulator Complex (WRC) are required for PVD dendrite outgrowth [26,27] Here our genetic results point to the paradoxical idea that elongation and branching of F-actin filaments are also necessary for dendrite retraction and that this mechanism is activated by UNC-6/Netrin in the self-avoidance response.

The actin cytoskeleton is highly dynamic in growing and retracting PVD dendrites
To test the idea that the actin cytoskeleton is differentially modulated for both outgrowth and retraction we expressed LifeAct::GFP in PVD neurons to monitor actin dynamics during dendrite morphogenesis. Although LifeAct::GFP binds both filamentous actin (F-actin) and monomeric G-actin with high affinity, bright LifeAct::GFP-labeled foci typically correspond to F-actin-containing structures [41]. We observed that the LifeAct::GFP signal was consistently brightest nearest the tips of growing dendrites in comparison to internal dendritic regions ( Fig  5B and 5C) as also reported in other studies [26,27,40]. Time-lapse imaging of PVD::LifeAct:: GFP by spinning disk confocal microscopy during the late L3/early L4 stage detected striking, dynamic fluctuations in the LifeAct::GFP signal (S3 Movie). We quantified this effect by comparing time-dependent changes in LifeAct::GFP fluorescence versus that of a PVD::mCherry cytosolic marker. The relatively constant intensity of PVD::mCherry signal in these experiments favors the idea that the observed changes in LifeAct::GFP fluorescence could be due to actin dynamics in growing dendrites (Fig 5A, 5D and 5E).
Because genetic knockdown of actin polymerizing proteins (UNC-34/Ena/VASP, MIG-10/ Lpd, UNC-73/Trio, WSP-1/WASP, ARX-5/P21) (Fig 3) disrupts self-avoidance and thus suggests that F-actin is required, we monitored the LifeACT::GFP signal at the tips of 3˚dendrites during the self-avoidance response (S4 Fig) (S3 Movie). The cytoplasmic mCherry marker was used to detect contact events as instances in which the measured gap between left and right 3d endrites approached zero. Although time-lapse imaging detected dynamic fluctuations in LifeAct::GFP fluorescence both before and after contact, we did not observe a quantitative increase in the LifeACT::GFP signal with retraction (S4 and S5 Figs). Notably, for these experiments, we used the marker strains (i.e., PVD::LifeAct::GFP and PVD::mcherry), imaging strategy (i.e., 40 X objective, 35 s sampling interval) and method of quantitative analysis described in an earlier publication that reported increased LifeAct::GFP fluorescence after contact (N = 14, p = 0.54, two-way ANOVA with Bonferroni correction [39] (S4A-S4C Fig). In an additional approach, we used a higher resolution objective (100X) and longer sampling interval (1-1.3 min) and also did not detect elevated LifeAct::GFP fluorescence during retraction (N = 8, p = 0.74, two-way ANOVA with Bonferroni correction) (S4D Fig and S5 Fig). The simplest explanation of our results is that actin polymerization drives both outgrowth and retraction which are tightly coupled over time at dendrite tips and thus difficult to resolve as separate events.

NMY-1/Non-muscle myosin II drives dendrite retraction in the selfavoidance mechanism
Our imaging results indicate that F-actin is the dominant cytoskeletal structure in growing PVD dendrites. In considering a mechanism to account for dendrite shortening in the selfavoidance response, we hypothesized that a newly synthesized actin cytoskeleton could mediate dendrite shortening by providing a substrate for non-muscle myosin II to drive contraction. To test this idea, we examined loss-of-function alleles that disable the C. elegans nonmuscle myosins, NMY-1 and NMY-2. This experiment determined that mutations in nmy-1 but not nmy-2 [42] show PVD self-avoidance defects (Fig 6A-6D). This effect is cell autonomous for nmy-1 because PVD-specific RNAi of nmy-1 also results in overlapping 3˚dendrites and PVD-specific expression of GFP-tagged NMY-1 is sufficient to rescue the nmy-1 selfavoidance defect (Fig 6B-6D). We conclude that a specific non-muscle myosin, NMY-1, promotes dendrite retraction in the self-avoidance mechanism and could act by driving the translocation of F-actin bundles at the tips of 3˚dendrites.

Phosphomimetic activation of MLC-4 shortens PVD 3˚dendrites
Non-muscle myosin II motor proteins are composed of myosin heavy chain (MHC) and light chains, the essential light chain (ELC) and regulatory light chain (RLC) [43,44]. Myosin motor activity is triggered by RLC phosphorylation that induces a conformational change to allow assembly of the myosin complex into active bipolar filaments (Fig 7A) [45]. Having shown that NMY-1/non-muscle myosin is required for self-avoidance, we next performed an additional experiment to ask if constitutive non-muscle myosin activity is sufficient to induce dendrite retraction. For this test, we expressed a phosphomimetic mutant of MLC-4, the C. elegans RLC homolog. MLC-4/myosin regulatory light chain contains serine/S and threonine/T phosphorylation sites in a highly-conserved domain. Both S and T residues were mutated to Aspartate/D and the resultant phosphomimetic construct MLC-4DD was fused to GFP for transgenic expression in PVD (Fig 7B and 7C) [46]. PVD neurons that expressed MLC-4DD showed significantly shorter 3˚dendrites at the L3 stage in comparison to wild type as well as a concurrent increase in the width of gaps between 3˚dendrites from adjacent menorahs ( Fig  7D-7G). These phenotypic traits are consistent with the idea that normal dendrite outgrowth is slowed by constitutive activation of non-muscle myosin by MLC-4ADD. By the L4 stage, PVD architecture in the MLC-4DD strain resembles that of the wild type with normal appearing menorah morphology and spacing (Fig 8A and 8B). Similar results were observed for PVD neurons that express constitutively active MYR::UNC-5 which we have proposed functions downstream of UNC-6/Netrin to trigger the self-avoidance response (Fig 2D). Our finding that MLC-4DD rescues the self-avoidance defect of unc-6 mutants suggests that NMY-1/nonmuscle myosin II also functions downstream of unc-6 for dendrite retraction (Fig 8D). Alternatively, rescue of the Unc-6 phenotype by MLC-4DD could be indicative of NMY-1/nonmuscle myosin II acting in parallel to unc-6.

Antagonistic pathways regulate dendrite outgrowth vs self-avoidance
Our time lapse imaging studies detected dynamic actin polymerization in growing PVD dendrites (S3 Movie). This observation and the recent finding that PVD dendrites are truncated and misdirected by mutations that disable either a the actin structural gene, act-4, or specific components (WRC and TIAM/GEF) that promote actin polymerization, suggest that dendritic outgrowth depends on actin assembly [26,27] (S2D Fig). Actin assembly is also likely required for dendrite retraction given results showing that genetic knockdown of a separate set of actinbinding proteins (i.e., UNC-34/Ena/VASP, MIG-10/Lpd, UNC-73/Trio, WSP-1/WASP, ARX-5/p21) (Fig 3), blocks PVD self-avoidance but not dendrite outgrowth [39]. A shared role for actin polymerization in both outgrowth and retraction argues that the net length of PVD dendrites must depend on an intrinsic mechanism that regulates actin assembly to balance these competing effects.
Our results suggest that UNC-6/Netrin signaling promotes dendrite retraction in a mechanism involving F-actin assembly. PVD dendritic outgrowth is driven by a separate pathway in which DMA-1, a membrane protein expressed in PVD, interacts with a multicomponent receptor anchored in the adjacent epidermis [13,[45][46][47][48][49]. Dendritic growth depends on binding of DMA-1 and the claudin-like protein, HPO-30, to specific regulators of actin assembly (i.e., WAVE regulatory complex, TIAM-1/GEF) in the PVD cytoplasm [26,27,50]. These interactions may be modulated by the Furin KPC-1 which antagonizes DMA-1 surface expression in 3˚dendrites show mCherry::UNC-34 puncta, ��� p<0.001, Fischer's exact test, N = 33 dendrite tips in 9 animals. Both iSIM and Airyscan images (S3 Fig, S1 Methods) were used in this analysis. https://doi.org/10.1371/journal.pgen.1008228.g004 Netrin Signaling regulates the actin cytoskeleton for dendrite self-avoidance  [47][48][49]. kpc-1 activity is proposed to temporally weaken adhesion with the epidermis to facilitate changes in trajectory at specific stages in dendritic outgrowth [49,50]. For example, the tips of 3˚branches typically execute a right-angle turn after the self-avoidance response to produce a 4˚dendrite [11]. This morphological transition largely fails in kpc-1 mutants in which 3˚dendrites elongate abnormally and show severe self-avoidance defects due to over-   [49,50]. Thus, we considered the idea that self-avoidance is achieved by the combined effects of kpc-1-dependent downregulation of DMA-1 and a separate mechanism that antagonizes outgrowth and requires UNC-6/Netrin. We performed a series of genetic experiments to test this model.
The hypomorphic loss-of-function allele, kpc-1(xr58), results in over-expression of DMA-1 and a consequent, robust self-avoidance defect with~42% of sister 3˚branches overlapping one another (Fig 9A and 9D) [48,49]. This defect is not observed, however, in kpc-1/+ heterozygotes (Fig 9D). Similarly, self-avoidance in unc-6/+ heterozygous animals is indistinguishable from wild type (Fig 9D). Thus, a 2-fold reduction in gene dosage for either kpc-1 or unc-6 does not perturb self-avoidance. We reasoned, however, that if expression of UNC-6/Netrin and KPC-1 were simultaneously reduced by half, then self-avoidance should be at least partially disabled since both unc-6 and kpc-1 normally function to antagonize dendrite outgrowth. To test this prediction, we constructed double heterozygous mutants of kpc-1 and unc-6 (e.g., kpc-1/+; unc-6/+) and quantified the self-avoidance phenotype. This experiment revealed that 30% of adjacent sister 3˚dendrites fail to self-avoid in kpc-1/+; unc-6/+ animals which is significantly greater than that of either kpc-1/+ or unc-6/+ single mutants (Fig 9B and 9D). This genetic "enhancer" effect is consistent with the idea that both KPC-1 and UNC-6 function to limit growth of 3˚dendrites. If this effect is due to elevated levels of DMA-1 arising from reduced KPC-1 activity in the kpc-1/+; unc-6/+ background, then a concomitant diminution of DMA-1 activity should restore normal self-avoidance. We tested this idea by constructing the triple heterozygote, kpc-1 +/+ dma-1; unc-6/+, and detected robust suppression of the selfavoidance defect observed in kpc-1/+; unc-6/+ animals (Fig 9C and 9D). Thus, these results suggest that DMA-1-mediated adhesion to the epidermis antagonizes UNC-6/Netrin-dependent dendrite retraction and that the relative strengths of these opposing pathways must be finely tuned to achieve the self-avoidance effect (Fig 9E). The key role of actin polymerization in both UNC-6/Netrin-mediated retraction and DMA-1-dependent outgrowth argues that this balancing mechanism likely regulates both modes of actin assembly.

Discussion
Dendrites arising from the same neuron typically do not overlap. This phenomenon of dendrite self-avoidance is universally observed and derives from an active mechanism in which contact between sister dendrites triggers mutual retraction [4,48]. As might be predicted for an event that depends on physical proximity, cell-surface associated proteins have been shown to mediate dendrite self-avoidance [39,51]. These include Dscam and protocadherins which are expressed in multiple alternative forms to distinguish self from non-self among contacting dendrites [5][6][7]. Self-avoidance for certain neurons can also depend on the interaction of diffusible cues with their cognate receptors [8,51]. For example, we have shown that the axon guidance signal, UNC-6/Netrin, and its receptors, UNC-5 and UNC-40/DCC, function together to mediate self-avoidance for PVD sensory neurons in C. elegans. We have proposed that UNC-40/DCC captures UNC-6/Netrin on the membrane surface to facilitate repulsion through physical interaction with UNC-5 on PVD dendrites [9] (S1D Fig). Little is known, and human myosin RLCs. (C) Phosphomimetic construct, PVD::GFP::MLC-4-DD, fused to GFP with a linker peptide for expression with PVD promoter. (D-E) Representative images (top) and schematics (bottom) of PVD morphology anterior to cell body (asterisk) denoting 3˚dendrites for (D) wild type (red) and (E) MLC-4DD (blue) (L3 stage larvae). Wild-type image also used in Fig 2A. Scale bars are 10 μm. 3˚dendrite length (F) is shorter and gaps between 3˚dendrites (G) is wider for MLC-4DD versus wild type at the L3 stage. Gaussian fit was plotted with bin center values (μm) on the X-axis and the number of values that fall under each bin center on the Y axis, ��� p<0.001, Kolmogorov-Smirnov test for cumulative distributions of PVD 3˚dendrite lengths and length of gaps in MLC-4DD (n = 141, 6 animals) versus wild type (n = 156, 8 animals). Scale bar is 10 μm. https://doi.org/10.1371/journal.pgen.1008228.g007 Netrin Signaling regulates the actin cytoskeleton for dendrite self-avoidance Mean and SEM are shown. n � 9 animals per genotype. Quantified with PVD::mCherry. Scale bar is 10 μm. wild-type and unc-6(ev400) results from Fig 2D. The image for unc-6(ev400) is also shown in Fig 2C. https://doi.org/10.1371/journal.pgen.1008228.g008 Netrin Signaling regulates the actin cytoskeleton for dendrite self-avoidance Here we report new results that are consistent with a model in which UNC-6/Netrin promotes actin-polymerization to propel dendrite retraction. Our finding that non-muscle myosin is also involved resolves the paradoxical idea that UNC-6/Netrin-dependent actin polymerization shortens rather than lengthens PVD 3˚dendrites. We propose that non-muscle myosin engages the nascent F-actin cytoskeleton at the tips of PVD dendrites to drive retraction.

Regulators of actin polymerization are required for dendrite self-avoidance
The proposed role for actin polymerization in the self-avoidance response derives from our finding that multiple regulators of F-actin assembly are necessary for efficient dendrite retraction. For example, a mutation that disables UNC-34, the C. elegans homolog of Ena/VASP, disrupts PVD self-avoidance. Ena/VASP promotes actin polymerization by preventing capping proteins from blocking the addition of actin monomers to the plus-end of elongating actin filaments. In axonal growth cones, Ena/VASP localizes to the tips of actin bundles in growing filopodia [12,38]. We observed punctate Ena/VASP localization at the tips of PVD dendrites (Fig  4 and S3 Fig). Our results showing that an unc-34 mutant suppresses the retarded dendrite outgrowth phenotype arising from the constitutively active MYR::UNC-5 ( Fig 3C) argues that unc-34 functions downstream to mediate UNC-6/Netrin-dependent self-avoidance. Additional genetic evidence indicates that WSP-1/WASP also acts in a common pathway with UNC-6/Netrin (Fig 3D). The established role for WASP of activating the Arp2/3 complex [52] is consistent with our finding that RNAi knockdown of arx-5/p21, a conserved Arp2/3 component, partially disables the self-avoidance mechanism (Fig 3D). Dual roles for UNC-34 and WSP-1 in actin polymerization, as suggested here, are also consistent with previous work indicating that Ena/VASP can activate Arp2/3 through direct interaction with WASP proteins [33,[53][54][55]. Thus, taken together, our results suggest that Ena/VASP and WSP-1 coordinate the creation of branched actin networks at the tips of retracting PVD dendrites in the UNC-6/ Netrin dependent self-avoidance response (Fig 9E).
A role for Ena/VASP in dendrite retraction, as proposed here, parallels earlier findings of a necessary function for Ena/VASP in growth cone repulsion. For example, axon guidance defects arising from ectopic expression of UNC-5 in C. elegans touch neurons requires UNC-34/Ena/VASP [56], a finding consistent with our observation that the dendritic shortening phenotype induced by MYR::UNC-5 is prevented by a loss-of-function mutation in unc-34. Ena/VASP is also required for axon repulsion mediated by Slit and its receptor Robo [20,57,58]. Although Ena/VASP function can be readily integrated into models of axonal attraction, the well-established role of Ena/VASP in F-actin assembly has been more difficult to rationalize in mechanisms of axonal repulsion which are presumed to involve actin depolymerization [12,59]. A partial explanation for this paradox is provided by a recent study showing that Slit, acting through Robo, induces transient filopodial outgrowth in migrating axonal (arrowheads) are noted. (D) Quantification of self-avoidance defects for each genotype. Note that all mutant alleles are recessive but that kpc-1/+; unc-6/+ double heterozygotes show a strong self-avoidance defect that is suppressed in kpc-1 +/+ dma-1; unc-6/+ triple heterozygotes. ��� p<0.0001, one way ANOVA with Tukey's posthoc test. Mean and SEM are indicated. ��� p <0.0001, N �7 animals per genotype. PVD is labeled with ser2prom3::GFP for all images. Scale bar is 10 μm. Mutant alleles were: kpc-1(xr58), unc-6 (ev400) and dma-1(wy686). (E) Model of opposing pathways that promote actin polymerization to define dendrite length. KPC-1 antagonizes surface expression of DMA-1 that interacts with an epidermal receptor composed of SAX-7/CAM, MNR-1 and LECT-2 and functions with HPO-30 to activate TIAM-1/GEF and the Wave Regulatory Complex (WRC) for actin polymerization and dendrite outgrowth. UNC-6/Netrin is captured by UNC-40/DCC for interaction with UNC-5 to activate UNC-34/Ena/VASP which functions with WSP-1/WASP and the Arp2/3 complex to drive F-actin assembly for NMY-1/myosin-dependent dendrite retraction in the self-avoidance response. (UNC-73/Trio and MIG-10/Lamellipodin are not shown to simplify the diagram, see Results). Homotypic interaction of MIG-14/Wntless in contacting 3 O dendrites also drives retraction possibly through WSP-1.
https://doi.org/10.1371/journal.pgen.1008228.g009 growth cones and that this effect requires Ena/VASP [58]. The functional necessity for filopodial growth in this mechanism is unclear, but the central role of Ena/VASP-induced actin assembly in the overall repulsive response to Slit mirrors our results suggesting that dendrite retraction in PVD neurons requires UNC-34/Ena/VASP and F-actin assembly.

Non-muscle myosin II mediates dendrite retraction in the self-avoidance response
Our genetic analysis detected a cell-autonomous role for NMY-1/non-muscle myosin II in dendrite self-avoidance (Fig 6A-6D). Because our results revealed F-actin at the tips of PVD dendrites (Fig 5), we suggest that NMY-1 mediates retraction by interacting with F-actin to drive a contractile mechanism. For example, NMY-1 could accelerate retrograde flow, a nonmuscle myosin-dependent effect that also drives filopodial retraction in growth cones and at the leading edge of migrating cells [60,61]. Alternatively, NMY-1 could shorten PVD dendrites by inducing the reorganization of the nascent actin cytoskeleton into a more compact structure [62]. Phosphorylation of myosin regulatory light chain (RLC) activates non-muscle myosin [45] and we showed that MLC-4DD, a phosphomimetic, and, thus, constitutively active form of the C. elegans homolog of RLC, results in delayed outgrowth of PVD 3˚dendrites ( Fig  7A-7G). The retarded dendrite outgrowth resulting from chronic activation of MLC-4 suggests a model in which NMY-1 function is triggered by contact between sister dendrites and subsequent UNC-6/Netrin signaling. A downstream function for non-muscle myosin II in the UNC-6/Netrin pathway is consistent with our finding that the Unc-6 self-avoidance defect is rescued by MLC-4DD (Fig 8D). Parallel roles for these components in axon guidance are suggested by recent results showing that non-muscle myosin II is required for Slit and Netrindependent midline avoidance in the developing vertebrate spinal cord [63]. Similarly, nonmuscle myosin II mediates Semaphorin-induced axonal repulsion. The mechanism of Semaphorin action in this case also involves the assembly of actin bundles in the axon shaft presumably to facilitate non-muscle myosin-driven retraction [64][65][66]. Thus, our findings suggest that key drivers of axonal repulsion including nascent actin assembly and non-muscle myosin may also mediate dendrite retraction in the self-avoidance response.
RLC phosphorylation activates non-muscle myosin II by releasing myosin monomers for assembly into bipolar structures or myosin "stacks" that interact with actin filaments to induce translocation [45,67]. Although our results are consistent with a necessary role for non-muscle myosin in dendrite retraction, the tips of PVD 3˚dendrites are likely too narrow (~50 nm) [10] to accommodate myosin stacks (~300 nm) [67,68]. We suggest that the actin-branching function of Arp2/3, which we have shown mediates PVD dendrite retraction, could induce transient expansion of the dendritic tip to allow assembly of myosin stacks. In this model, actin-polymerization involving Arp2/3 could effectively limit ectopic NMY-1 activation in the dendritic shaft by restricting the assembly of myosin stacks to the tips of contacting sister dendrites. Alternatively, recent evidence indicates that myosin can also induce contraction in its monomeric or unpolymerized form [69] which should have ready access to the dendrite tip. This idea is consistent with our observation that GFP-tagged NMY-1 is active (e.g., rescues the self-avoidance defect of nmy-1) and did not display the punctate appearance characteristic of myosin stacks (Fenix et al., 2016) but is diffusely localized throughout the PVD cytoplasm ( Fig  6B).

Parallel-acting pathways regulate dendrite self-avoidance
Although self-avoidance is perturbed in UNC-6/Netrin pathway mutants, a significant fraction (~70%) (Fig 3D) of PVD 3˚dendrites show apparently normal self-avoidance behavior [9]. One explanation for this observation is provided by the recent discovery of an independent pathway, mediated by MIG/14/Wntless, that is necessary for self-avoidance in an additional fraction of PVD 3˚dendrites [39]. Parallel acting pathways also mediate self-avoidance in mammalian Purkinje neurons where slit-robo signaling acts in concert with protocadherins (Gibson et al., 2014;Lefebvre et al., 2012). The likelihood of additional effectors of self-avoidance is also suggested by the incompletely penetrant effects of Dscam mutants in Drosophila sensory neuron self-avoidance (Matthews and Grueber, 2011). Thus, our results in C. elegans mirror findings in other organisms which together suggest that multiple mechanisms have evolved to insure dendrite self-avoidance.

The actin cytoskeleton is highly dynamic in growing and retracting PVD dendrites
We used time-lapse imaging with LifeAct::GFP, a live-cell marker for actin [41], to monitor actin PVD dendrites. The LifeAct::GFP signal is typically brighter near the distal ends of growing branches in comparison to interstitial regions (Fig 5B and 5C) and is highly dynamic (S3 Movie). The distal localization and rapidly fluctuating LifeAct::GFP signal in PVD dendrites resembles recently reported "actin blobs" that actively migrate in highly branched Drosophila sensory neurons where they appear to presage points of branch initiation [70]. A strong Life-Act::GFP signal at the tips of emerging PVD dendrites is also consistent with genetic evidence that actin polymerization is required for dendrite outgrowth [24,25]. Although our genetic results (Fig 3) suggest a model of dendrite retraction in which nascent actin polymerization is necessary for the self-avoidance response, our measurements did not detect a quantitative increase in LifeAct::GFP in retracting vs growing 3˚dendrites as previously reported by others [39]. This result suggests that the overall level of actin polymerization is high during both outgrowth and retraction but may be incorporated into distinct structures, regulated by distinct groups of effectors (see below) that our imaging methods do not resolve.

Antagonistic pathways regulate dendrite outgrowth vs retraction
Dendritic architecture is defined by the combined effects of outgrowth which expands the arbor vs retraction which limits the size of the receptive field. This interaction of positive and negative effects is readily observed in the development of 3˚PVD dendrites. Initially, 3˚dendrites emanating from adjacent menorahs grow out along the body axis until contacting one another and then retract to avoid overlap [9,11]. Outgrowth in this case is promoted by a multicomponent receptor-ligand complex that mediates adhesive interactions with the adjacent epidermis [71][72][73][74] (Fig 9E). When this pathway is dysregulated by over-expression of the DMA-1 receptor in PVD neurons, for example, 3˚dendrites continue to adhere to the epidermis and overgrow one another [49,75]. 3˚branch self-avoidance also fails in mutants that disable UNC-6/Netrin signaling which we have shown promotes dendrite retraction [9]. Thus, these results suggest that the net length and placement of each 3˚branch depends on the balanced effects of locally acting signals that either extend (e.g., DMA-1) or shorten (e.g., UNC-6/ Netrin) the dendrite. We confirmed this idea in genetic experiments that detected strong dose-sensitive interactions between these opposing pathways. For example, a 50% reduction of UNC-6/Netrin expression in a heterozygous unc-6/+ mutant does not perturb self-avoidance. Similarly, a genetic mutant (kpc-1/+) that partially elevates DMA-1 also shows normal 3b ranch outgrowth. The combination of both mutations in the double heterozygote, unc-6/+; kpc-1/+, however, produces a strong self-avoidance defect (Fig 9A-9D). Our genetic results identified multiple regulators of actin polymerization (UNC-34/Ena/VASP, WSP-1/WASP, ARX-5/p21, MIG-10/Lamelipodin, UNC-73/Trio) that are required for self-avoidance (Fig 3). We thus propose that UNC-6/Netrin promotes actin polymerization to effect dendrite retraction and that non-muscle myosin II interacts with a nascent actin cytoskeleton to translocate each 3˚dendrite away from its neighbor. F-actin is also abundant and highly dynamic in growing PVD dendrites. Notably, a distinct set of F-actin promoting factors (WRC, TIAM/GEF) interacts with the DMA-1 receptor complex to drive PVD dendritic growth (Fig 9E) [26,27]. Additional genetic evidence suggests that the WRC could also function in self-avoidance [39]. Thus, the challenge for future studies is to elucidate the cell biological machinery through which opposing pathways manipulate the actin cytoskeleton to effect either dendrite growth or retraction.

Strains and genetics
All the strains used were maintained at 20˚C and cultured as previously described (Brenner, 1974). We used the N2 Bristol strain as wild-type.
Confocal microscopy and image analysis. Animals were immobilized with 15 mM levamisole/ 0.05% tricaine and mounted on 2% agarose pads in M9 buffer as previously described [11]. Z stack images were collected using a Nikon TiE inverted fluorescent microscope equipped with an A1R point scanning confocal and a Plan Apo 40X, 1.3NA oil immersion objective or on a Leica TCS SP5 confocal microscope with a 40X or 63X objective. Z stacks were collected at either 0.5 or 1μm depth from the position of ventral nerve cord to the top of PVD cell body. Maximum projection images were collated with either LAS-AF (Leica) or with NIS elements software (Nikon). Dendrite length was measured using the line tool in Fiji/image J. Co-localization images were assembled using image J. Z stack images that were collected using the LAS-AF or NIS elements software was imported into ImageJ/Fiji for further analysis.
iSIM super-resolution imaging and analysis. Animals were immobilized with 0.4 mM levamisole and mounted on 10% agarose pads in M9 buffer. Z stack images were captured using a Hamamatsu sCMOS camera on a iSIM microscope (VisiTech International) equipped with a 100X, 1.45 NA objective. Images were deconvolved and processed using Metamorph and Fiji/ImageJ. Spinning disk confocal microscopy. Animals were immobilized with 0.4mM levamisole and mounted on 10% agarose pads in M9 buffer. Volumes over time were captured using an automated TiE inverted fluorescence microscope platform with an X1 spinning disk head (Yokogawa) and DU-897 EM-CCD (Andor Technology), piezo-electric Z stage (Mad City Labs) and Apo TIRF 100x 1.49 NA oil immersion objective (Nikon Instruments, Inc.). Analysis of fluorescence intensity was accomplished using NIS-Elements and Fiji/Image J software.
Quantification of fluorescence intensity in growing vs. non-growing dendrites. L2-L3 larvae were immobilized with 0.4mM levamisole and mounted on 10% agarose pads in M9 buffer for imaging with the spinning disk confocal microscope. Z-stacks of PVD dendrites were imaged at either 1 min or 1.30 minute intervals with the 100X 1.49 NA oil immersion objective. PVD dendrites (2˚, 3˚and 4˚) undergoing outgrowth with a succession of rapid extensions and retractions during the length of the time lapse videos were analyzed. NIS elements was used for measurements of LifeAct::GFP from a 1μm ROI at the tips of growing dendrites. In the same videos, fluorescence intensity was measured for an identical 1μm ROI of PVD branches (1˚, 2˚and 3˚) that were not undergoing outgrowth. Fluorescence intensity measurements for the outgrowth and no-outgrowth datasets were normalized against the maximum overall LifeAct::GFP value (Fig 5).
Statistics. Student's t test was used for comparisons between 2 groups and ANOVA for comparisons between three or more groups with either Tukey posthoc or Bonferroni correction for multiple comparisons. Kolmogorov-Smirnov test was performed for comparisons between frequency distributions of 3˚dendrite length and gaps between 3˚dendrites (Figs 1C, 8F and 8G).