Wnt/PCP proteins regulate stereotyped axon branch extension in Drosophila Free

Branching morphology is crucial for nerve function. Acting as conduits, branched axons send nerve signals to multiple targets that are sometimes separated by large distances and are functionally distinct within the nervous system or periphery. Conversely, dendrite branches collate disparate nerve inputs and their integrative function is key to signal processing. Branched structures along a nerve segment also have physiological properties that propagate, disperse and spatially represent nerve signals (Baccus et al., 2000; Huguenard, 2000; Bollmann and Engert, 2009; Foust et al., 2010). How branched structures form is unclear. Given their diverse shapes, the mechanisms might be quite heterogeneous. Some branched forms are likely to be regulated by intrinsic factors. Others might become more or less elaborate in response to extracellular growth factors, neural activity or injury. Axon branch dynamics are also intimately linked to synaptogenesis (Alsina et al., 2001; Javaherian and Cline, 2005; Waites et al., 2005; Meyer and Smith, 2006; Stettler et al., 2006). Both imaging and anatomical studies indicate that branching parameters, such as number, length and branch localization points, are also influenced by neuronal type, global branch distribution and connection properties (Stepanyants et al., 2004; Portera-Cailliau et al., 2005; Snider et al., 2010). Recent computational approaches also suggest that Ramon y Cajal’s principle of the economic use of wiring space, cytoplasmic material and conduction times are key trade-offs in branch patterning (Budd et al., 2010; Cuntz et al., 2010). Inevitably, all these parameters rely on the very diverse context of function(s) in individual neurons.

Given the extensive branching morphologies seen in vivo, existing studies do not give a full understanding on this subject. In many examples of studies in which branching phenotypes have been reported, the defects represent a range of morphological criteria, affecting the number, length, order, stereotypy, formation or removal of branches. For example, brain-derived neurotrophic factor (BDNF) acts together with activity-dependent processes to regulate the terminal branch number and length of Xenopus retinal axons (Cohen-Cory and Fraser, 1995; Cohen-Cory, 1999). The n-methyl-d-aspartic acid (NMDA) receptor controls activity-dependent processes that determine terminal branch patterning (Ruthazer et al., 2003). Other factors, such as Slits and Ephrins, also regulate axon branching. In zebrafish, once retinal axons have reached the tectal target, Slit1a inhibits terminal arborization to reduce excessive branching (Campbell et al., 2007). Slit can also positively increase branch numbers of dorsal root ganglion axons in mammals (Wang et al., 1999). Slit and the Slit receptor Roundabout mutant mice exhibit multiple axon branching phenotypes, in which branch number, length, order and stereotyped bifurcation are affected in different neuronal types (Ma and Tessier-Lavigne, 2007). Ephrins play different roles in axon branching. Depending on the positional origin of retinal ganglion cells, their axons branch in different regions of the target zone, by a process termed interstitial branching. Ephrin signals are responsible for branch suppression at inappropriate sites along the axonal shaft and for the subsequent removal of branches distal to the terminal zone (Yates et al., 2001; McLaughlin and O’Leary, 2005). In addition, the ephrinA5 ligand has branch-promoting activity through interactions with the TrkB receptor (Marler et al., 2008). Based on geometry, axon and dendrite branches can display stereotyped and non-stereotyped patterns. Recent studies show the small, natriuretic peptide CNP regulates the stereotyped bifurcation of sensory axons in the spinal cord (Schmidt et al., 2009; Zhao and Ma, 2009). A different example of stereotyped branch patterning was recently found in Netrin pathway mutants in Caenorhabditis elegans, in which ventral branching is lost in some neurons (Hao et al., 2010).

As branch formation might involve cell-polarizing mechanisms, I decided to study components of the Wnt/Planar Cell Polarity (PCP) pathway. The Wnt/PCP pathway represents a major class of regulators that control planar cell polarity (also termed tissue polarity or planar polarity). This has been extensively studied in Drosophila patterning of the adult cuticle in the wing, abdomen and notum, in the Drosophila eye and in sensory hair cells of the inner ear in mice (Klein and Mlodzik, 2005; Jones and Chen, 2007; Wang and Nathans, 2007; Strutt, 2009; Vladar et al., 2009). The Wnt/PCP pathway acts to orient cells along an axis so that all cells exhibit a grouped polarity. These genes can act both cell-autonomously (intracellularly) and non-autonomously (intercellularly between neighboring cells).

Despite extensive studies, how they function is unclear and several distinct models are proposed, based principally on Drosophila results (Klein and Mlodzik, 2005; Lawrence et al., 2007; Strutt and Strutt, 2009; Vladar et al., 2009). One key finding is that a seven-pass transmembrane protein Frizzled (Fz) is essential (Gubb and Garcia-Bellido, 1982; Vinson and Adler, 1987; Vinson et al., 1989). How Fz regulates tissue polarity is much debated. One model suggests ‘core’ components of the Wnt/PCP pathway act to localize Fz asymmetrically within cells (Strutt, 2001; Strutt, 2003; Klein and Mlodzik, 2005; Strutt and Strutt, 2009; Vladar et al., 2009). Acting as an anti-Fz factor, the four-pass transmembrane protein Strabismus (Stbm, also known as Van Gogh or Vang) is thought to act intracellularly to restrict Fz localization to the pole opposite of Stbm (Jenny et al., 2003; Amonlirdviman et al., 2005; Klein and Mlodzik, 2005; Vladar et al., 2009). Interestingly, in gain- or loss-of-function and clonal analysis experiments, Drosophila fz and stbm mutants often show opposite misoriented tissue polarity phenotypes (Strutt, 2003; Lawrence et al., 2004; Klein and Mlodzik, 2005). Via intercellular signaling, further models suggest that Stbm also acts downstream to amplify Fz protein asymmetry (Jenny et al., 2003; Amonlirdviman et al., 2005; Wu and Mlodzik, 2008). Alternative models suggest that Fz and Stbm effects are inter-dependent, based on intercellular bi-directional signaling that also promotes asymmetric localization across neighboring cells (Lawrence et al., 2004; Strutt and Strutt, 2007; Chen et al., 2008). Therefore, Fz and Stbm have antagonistic as well as cooperative roles in tissue polarity.

Other core components, such as the cytoplasmic factors Prickle (Pk) and Dishevelled (Dsh), are also involved and similarly asymmetrically localized. Pk is linked functionally to Stbm and acts intracellularly to inhibit Fz-Dsh signals (Tree et al., 2002; Jenny et al., 2003; Jenny et al., 2005). Pk is also required downstream to amplify Fz dependent signals via intercellular signaling (Tree et al., 2002; Amonlirdviman et al., 2005). However, this is debated because, in contrast to fz and stbm, pk effects are largely cell-autonomous (Gubb and Garcia-Bellido, 1982; Gubb et al., 1999; Adler et al., 2000) and others find it is not required for fz and stbm non-autonomous functions (Lawrence et al., 2004; Strutt and Strutt, 2007). Previous studies suggest Dsh is a Fz transducer that is antagonized by Stbm and Pk (Krasnow et al., 1995; Boutros et al., 1998; Tree et al., 2002; Jenny et al., 2003). However, this is also debated as some reports show that Dsh is not required for Fz intercellular signaling (Strutt and Strutt, 2002; Strutt and Strutt, 2007). In addition, Dsh also interacts physically with Stbm and is purported to act downstream of Stbm (Bastock et al., 2003; Torban et al., 2004; Jenny et al., 2005).

Asymmetrically organized Fz is thought to be essential because this helps explain the polarized nature of Fz signaling, which is an essential feature in fly wing tissue polarity (Adler et al., 1997; Adler, 2002). However, the role of PCP protein asymmetry is debated as some Drosophila studies show that cells can still exhibit tissue polarity even without core components such as Dsh, Pk or Diego (Strutt and Strutt, 2002; Lawrence et al., 2004; Strutt and Strutt, 2007). In addition, timing studies suggest that some PCP-dependent events occur early, prior to obvious PCP protein asymmetry (Strutt and Strutt, 2002; Classen et al., 2005; Strutt and Strutt, 2007; Doyle et al., 2008). Nonetheless, whether as a basis of the cause or the consequence of tissue polarization, vertebrate PCP proteins also show subcellular asymmetry in polarized tissues (Wang et al., 2005; Montcouquiol et al., 2006; Wang et al., 2006; Deans et al., 2007; Jones and Chen, 2007; Qian et al., 2007; Wang and Nathans, 2007; Devenport and Fuchs, 2008; Sugiyama et al., 2010), suggesting that PCP protein asymmetry is a conserved feature in tissue polarity.

The Wnt/PCP pathway also regulates axon development and vertebrate homologs of the Drosophila proteins, such as Flamingo, Frizzled, Dishevelled and Stbm, have different roles in axon guidance (Wang et al., 2002; Lyuksyutova et al., 2003; Tissir et al., 2005; Zhou et al., 2008; Jing et al., 2009; Fenstermaker et al., 2010; Shafer et al., 2011). Interestingly, Drosophila studies show that Wnt/PCP factors (such as Flamingo/Starry night) can function independently of other ‘core’ components to regulate axon and dendrite patterning (Gao et al., 2000; Senti et al., 2003; Steinel and Whitington, 2009). In addition, Frizzled and Dishevelled can signal through distinct Wnt components (Frizzled 2 and Wnt4) to control axon guidance in the fly visual system (Sato et al., 2006).

To study axon branching, I analyzed mushroom body (MB) neurons in the Drosophila brain. MB axons of the αβ lineage follow a simple pattern, extending a single branch dorsally and another medially (Fig. 1A-D) (Lee et al., 1999). Previous results show that inactivation of Rac GTPases results in MB axon branching defects with the loss of either the dorsal or the medial branch, with the other branch remaining properly targeted (Ng et al., 2002). Searching for additional genes, I found that the Wnt/PCP pathway also controls axon branching. Some Wnt/PCP components showed a strong preference for dorsal or medial branch phenotypes. Although Wnt/PCP components interact functionally, here I suggest that their actions are separable. These proteins are present in axons but are not asymmetrically localized to distinct branches. Interestingly, I found that Prickle branch localization is dependent on Frizzled and Strabismus.

The following fly strains were used: dsh¹ (Axelrod et al., 1998; Boutros et al., 1998); fz^K21 (fz³⁰) (Adler et al., 1990); fz²¹, fz^R52 (fz²³), fz^H51 (fz¹⁵) (Jones et al., 1996); stbm^A3 (Vang^A3) (Taylor et al., 1998); stbm⁶ (Wolff and Rubin, 1998); pk-sple¹³, UAS-pk^sple (Gubb et al., 1999); UAS-stbm (lines 6.2 and 10.2 constitute 2× expression) (Bastock et al., 2003); UAS-GFPstbm (line 5) (Bellaiche et al., 2004); UAS-FzGFP (line 29) (Strutt, 2001); UAS-dshGFP (lines 27.1a and 27.2a, both gave similar results) (Axelrod et al., 1998); dshGFP (two lines located on the second and third chromosome) (Axelrod, 2001). All PCP loss-of-function alleles are null with respect to PCP phenotypes. fz²¹, stbm⁶ and pk-sple¹³ are protein-null. Additional allele descriptions are in Flybase (www.flybase.org). To analyze dsh homozygous mutants, dsh^1/Y males and dsh^1/3 females were used. dsh³-null animals are non-viable, and were therefore not analyzed. RNAi lines were obtained from the Vienna Drosophila RNAi Center [VDRC (v)] or the Bloomington Drosophila Stock Center [BDSC (bl)], or from David Strutt, Sheffield University, UK (Bastock and Strutt, 2007). Under standard conditions, flies were raised at 25°C. For RNAi analysis, animals were raised at 29°C and with co-expression of Dcr-2 for a higher RNAi efficiency (Dietzl et al., 2007).

MARCM clones were performed as previously described (Lee et al., 1999). The main GAL4 driver used was OK107, which labels all MB neurons in addition to small populations of other neurons in the antennal lobe, optic lobe and pars intercerebralis (Connolly et al., 1996; Ng and Luo, 2004; Aso et al., 2009). Additional MB GAL4 lines were described previously (Akalal et al., 2006; Tanaka et al., 2008; Aso et al., 2009).

Antibodies used were anti-Fas2 (1:20, 1D4 from G. Tear, King’s College London, UK), anti-GFP (1:200, Molecular Probes, A11122), anti-mouse CD8 (1:100; eBioscience, 53-6.7), anti-Pk [1:250, from J. Axelrod, Stanford University, USA (Tree et al., 2002)] and anti-Dsh [1:1000, from T. Uemura, Kyoto University, Japan (Shimada et al., 2001)] and staining was performed as described previously (Wu and Luo, 2006). Fz and Stbm antibodies were also tested. Staining patterns in the adult brain were either diffuse with indistinguishable MB [using two independently generated Fz antisera, 1:50-250, from D. Strutt and R. Nusse (Stanford University, USA), and anti-Stbm, 1:200-500, from T. Wolff, Washington University, St Louis, USA (Rawls and Wolff, 2003)], or showed MB signals that were also detected in protein-null mutants [anti-Stbm, 1:100, from D. Strutt (Bastock et al., 2003)]. These antisera were not included in this study. A recent, independent study showed Fz and Stbm expression in MB axons (Shimizu et al., 2011). The nature of these discrepancies is currently unclear but might reflect the differences in antibody batches, preparative methods used and/or the stages analyzed.

To estimate the bias in branching phenotypes, a preference index (PI) was calculated by dividing the greater percentage value by the lower value (D^–M⁺/D⁺M^– or D⁺M^–/D^–M⁺, where D is the dorsal branch and M is the medial branch). Mutants showing a preference for the dorsal branch defect were given a positive value, whereas a preference for the medial branch defect is indicated with a negative value. Genotypes that showed no preference were designated a value of 1.0. When no comparative numbers were available, the index was left blank even though a bias might exist. A two-tailed binomial test with P=0.5 was used to infer the probability of a bias in dorsal or medial branching phenotypes. As indicated by the asterisks, P<0.001 (^***), P<0.01 (^**) and P<0.05 (^*) indicate threshold P-values for a preference for dorsal or medial branch defect, the lower values represent a stronger likelihood for a bias. Non-significant (NS) values (P>0.05) indicate no significant bias for either phenotype. Despite having the same gene lesion, Wnt/PCP stocks from different sources showed some variability in the extent of branching defects. To minimize the non-specific genetic backgrounds, mutants were analyzed either as transheterozygotes or as trans-combinations using stocks obtained from different laboratories or out-crossed strains that I generated. Differences in genetic background could account for the low penetrance in pk-sple¹³ branching reported in the Shimizu et al. study (Shimizu et al., 2011).

When analyzing loss-of-function mutants of the Wnt/PCP pathway, dishevelled (dsh), prickle (pk, also known as prickle-spiny legs or pk-sple), frizzled (fz) and strabismus (stbm, also known as Van Gogh or Vang) mutants all showed αβ axon branching defects (Fig. 1E-J). These defects included loss of the dorsal (D^–M⁺) or the medial (D⁺M^–) branches. Quantitative analyses showed that both phenotypes were not equally affected (Table 1). In fz mutants, the bias was significantly towards dorsal defects. For stbm mutants, medial branch defects predominated. In dsh mutants, preference for dorsal defects was detected with a strong allelic combination (dsh^1/3). No bias was detected in dsh¹ animals. No bias was detected in pk-null mutants either. MARCM analysis at single-cell resolution showed that only single branches were fully extended in fz and stbm mutant axons (Fig. 1I-J; n=8-10 clones each), often with a shorter branch (sometimes multiple branches) close to the wild-type branching point. This suggests that these defects reflect failure of branch extension but not of initiation of branch formation.

One way to explain the biases in fz and stbm branching defects is to invoke the PCP models above. Therefore, if dorsal branch projections rely on Fz and medial branch projections on Stbm, Fz-Stbm cross-regulatory events might control branch formation. This was investigated in two ways. First, fz, stbm double homozygous mutants were generated. From these (and any further) genetic interactions, I measured whether there was any change in the extent or bias in dorsal or medial branch defects. Rather than resulting in a default fz or stbm phenotype, as would be expected if either gene was dominant, any preference for dorsal or medial branch defect was lost in fz, stbm mutants (Table 2). Compared with single mutants, the extent of branching errors was also higher in fz, stbm as wild-type (D⁺M⁺) branching was reduced. So, despite the bias in branch contributions in each mutant, both Fz and Stbm independently control dorsal and medial branch projections in a partially redundant manner. Furthermore, Stbm is likely to act in parallel, being neither upstream nor downstream of Fz (Fig. 5B).

This was explored further using gain-of-function experiments (Fig. 2). When FzGFP (Strutt, 2001) was overexpressed in MB neurons, branching defects were observed with the loss of dorsal or medial lobes (Fig. 2B,C). However, dorsal defects were detected more often. This result is surprising as the predicted fz gain-of-function should be opposite to its loss-of-function phenotype, as reported in Drosophila wing (Adler et al., 1997; Lawrence et al., 2004), and is expected act like the loss of stbm (Taylor et al., 1998), i.e. to give predominantly medial branch defects. This was also observed using a different UAS-Fz transgene (Adler et al., 1997) (not shown). Strong Stbm overexpression (Bastock et al., 2003) also resulted in a branching phenotype, which was opposite to its loss-of-function phenotype, in which only dorsal lobes were disrupted (Fig. 2D,E).

Next, I tested whether these phenotypes would change under different PCP settings (Fig. 2E). Overexpression of FzGFP in stbm-null neurons resulted in a preference for dorsal branch defects. This suggests that the fz gain-of-function is dominant over the stbm loss-of-function phenotype and that stbm perturbation has little effect. This is in contrast to studies that suggest fz gain-of-function signals require stbm (Lawrence et al., 2004; Strutt and Strutt, 2007). I also performed the converse experiment of expressing GFP-tagged Stbm (GFPstbm) (Bellaiche et al., 2004), using a transgenic line that does not significantly perturb branching in a wild-type background. In fz-null animals, this enhanced dorsal branching defects, phenocopying the strong Stbm overexpression phenotype. This suggests that endogenous Fz might counteract Stbm gain-of-function signals. This result is also unexpected given that previous studies show that stbm gain-of-function signals also require fz (Lawrence et al., 2004; Strutt and Strutt, 2007).

When GFPstbm was co-expressed with FzGFP, there was an increase in medial branch defects and the bias for dorsal defects was lost. Any bias in branching defects was also eliminated when strong Stbm lines were co-expressed with FzGFP, which together showed a strong enhancement of branching phenotypes. Thus, high Fz levels together with high Stbm expression can result in the same branching phenotypes as was observed in the fz, stbm double null mutants. Together these results suggest that Fz and Stbm not only have independent redundant functions, they also show mutually cooperative behaviors (Fig. 5B; supplementary material Fig. S3). Stbm does not behave like an anti-Fz factor. However, under Stbm gain-of-function conditions, endogenous Fz has anti-Stbm activities.

To determine further how PCP genes control axon branching, genetic mosaic experiments were performed. In each hemisphere, there are ∼2000 MB neurons (Kenyon cells) of which ∼1000 are branched αβ neurons (Akalal et al., 2006; Aso et al., 2009). MB neurons are arranged in concentric layers whereby newborn neurons emerge from the central core region and, by displacement, earlier-born neurons are located in the peripheral regions, which is further divided into sub-regions based on birth order and projection patterns (Kurusu et al., 2002; Tanaka et al., 2008). Each neuron is derived from one of four dividing neuroblasts (Nb) in each hemisphere. By MARCM labeling, an adult Nb clone contains ∼400 neurons of which ∼200 are αβ neurons (Lee et al., 1999). When such large Nb clones of PCP mutant genotypes were generated (corresponding to about a quarter of MB neurons per hemisphere) in animals of heterozygous background, branching phenotypes were either absent or severely reduced (Fig. 3A-D; Table 3). Although this suggests that Wnt/PCP branching functions are non-MB-autonomous, when the corresponding RNAi transgenes were expressed in all MB neurons, branching defects were observed (Fig. 3E-H; Table 3). Therefore, Wnt/PCP functions are likely to be MB-autonomous but their loss in a subpopulation does not result in branching defects.

This non-autonomous effect was tested further using stbm rescue experiments (Fig. 3I-L; Table 3). GFPstbm expression in all stbm^–/– MB neurons rescued the branching defects, consistent with an MB-autonomous requirement. Using MARCM, when GFPstbm expression was restricted to single large Nb clones, this only partially suppressed the branching phenotype. A partial rescue effect was also observed when GFPstbm expression was restricted to Nb clones of only αβ neurons. Note that whenever branching phenotypes were observed, mutant and non-mutant axons always exhibited the same phenotype (Fig. 3J,K,L compared with 3J′,K′,L′, respectively; n=20, 15, 10 for large Nb, small αβ and single cell clones, respectively). Together, this suggests that PCP proteins regulate axon branching in a collective manner.

To determine further Wnt/PCP cell-autonomous functions in axon branching, I used the GFPstbm rescue paradigm to test its requirement in different MB subpopulations (supplementary material Table 1). Different GAL4 drivers label MB neuron subpopulations according to lineage, developmental stage and cell number (Akalal et al., 2006; Tanaka et al., 2008; Aso et al., 2009). The GAL4 driver OK107, labeling all MB neurons, provided an almost full restoration of wild-type, branched projections when GFPstbm was expressed in stbm⁶ animals. With c739-GAL4, GFPstbm expression in most αβ neurons similarly provided a robust rescue of wild-type projections. By contrast, its expression in γ (201Y and H24) and α′β′ (c305a) lineages failed to significantly alter the branching errors. Using 17D-GAL4, which labels 40% of αβ neurons (Akalal et al., 2006), this provided a partial rescue. This is consistent with the genetic mosaic results that show expression of GFPstbm in single neuroblast clones only partially rescues the branching defect (Table 3). The αβ lobe is developmentally subdivided between the core region, where growing axons form, and the surface and posterior region, where mature axons are located (Tanaka et al., 2008; Aso et al., 2009). GFPstbm expression in the core region alone (using NP7175) was sufficient for a robust rescue. By contrast, expression in the mature region (NP5286) only gave a partial rescue. GFPstbm expression in all mature MB neurons (MB247) also failed to rescue the stbm defects. These results suggest that PCP-dependent axon branching occurs early in development in the core region of αβ lobes. For NP7175-GAL4, only males were analyzed owing to stronger GAL4 expression (Aso et al., 2009).

Studies from Drosophila eye and wing development show that PCP proteins are partitioned to discrete subcellular locations, at points along the axis where tissue polarity develops (Usui et al., 1999; Axelrod, 2001; Shimada et al., 2001; Strutt, 2001; Strutt et al., 2002; Tree et al., 2002; Yang et al., 2002; Das et al., 2002; Bastock et al., 2003; Jenny et al., 2003; Rawls and Wolff, 2003; Das et al., 2004). To determine whether PCP axonal functions involve asymmetric localization to specific branches, I analyzed their adult MB localizations. UAS-GAL4 methods were used to analyze Fz and Stbm, as the antibodies tested could not clearly visualize native proteins in the brain (see Materials and methods). Dsh localization was analyzed using a Dsh-GFP fusion protein expressed under a dsh promoter (Axelrod, 2001). dshGFP fully rescued dsh¹ axon branch defects and its expression closely matched the endogenous Dsh brain expression pattern (Fig. 4A; 100% wild type, n=56 hemispheres; not shown). Pk localization was analyzed using Pk antibodies. At the gross level, all proteins were detected in dorsal and medial branches under wild-type conditions (Fig. 4), so it is unlikely that they need to be restricted to particular branches to execute branching functions. Also, it is unlikely that protein asymmetry was adopted at early stages as similar patterns were observed in pupal brains (supplementary material Fig. S1).

I tested whether these proteins undergo changes in branch locations in PCP mutant backgrounds. From Drosophila and mammalian studies, disrupting one PCP component will lead to the mislocalization of other PCP proteins (Strutt, 2003; Jones and Chen, 2007; Wang and Nathans, 2007). Dsh, Fz and Stbm branch localizations were unchanged when different PCP components were removed (Fig. 4; n≥10 for each genotype). Under wild-type conditions, Pk immunostaining revealed two levels of expression in MB axons: a higher level in the central core region and a weaker level throughout the αβ lobes (Fig. 4M,M′). In fz mutants, Pk levels were reduced in both dorsal and medial lobes, particularly in the core region (Fig. 4N). In stbm mutants, although Pk was consistently observed in the dorsal lobes, it was often absent from medial lobes (Fig. 4O; in a blind study, 13 out of 33 lobes showed abnormal medial staining in stbm⁶ animals whereas five out of five lobes showed wild-type staining in wild-type controls). Pk immunolabeling was detected in both branches in dsh¹ mutants (Fig. 4P). Therefore, with the exception of Pk, this suggests that the branch-localized PCP proteins do not require other PCP factors to act as targeting complexes.

As shown above, pk mutants do not exhibit any preference for dorsal or ventral branch defects. This is surprising when compared with stbm and fz, which show clear stereotyped biases in branching defects. To understand how pk branching functions relate to other PCP genes, genetic interactions were performed. As shown in Tables 1 and 2, compared with single mutants, fz, stbm double mutants show more frequent branching errors, together with a loss of bias in dorsal or medial branching defects. In pk, fz double homozygous mutants, whereas pk strongly synergized with fz with ∼40-50% increase in branching errors, the fz dorsal bias remained. This suggests that pk acts additively to fz, as it affects only the extent but not the directional bias in branch defects. One surprise was dsh, pk double mutants, which showed a strong ‘switch’ towards medial branch defects, compared with the single mutants that had no significant bias. This suggests that, rather than acting as an anti-Dsh factor (Tree et al., 2002; Jenny et al., 2003; Jenny et al., 2005), Dsh and Pk have a convergent preference in regulating the medial branch that is masked in the single mutants. Despite this cooperative behavior, compared with single mutants, the extent of branching errors was not significantly altered in dsh, pk mutants. In pk, stbm double mutants, there was a small (∼10%) increase in branch error. Nonetheless, a bias towards medial branch defects remained (Table 2). This suggests that pk also functions additively, though not identically, to stbm.

pk overexpression also resulted in branching defects (supplementary material Fig. S2). Under different PCP conditions, further examples of cooperative and antagonistic behavior were also observed. For example, Pk overexpression led to the preferential loss of the medial branches, but when coupled with Dsh overexpression, the extent of branch errors and the bias were reduced. When Pk was overexpressed in Fz loss-of-function or overexpression backgrounds, the bias was lost, but there were more severe axonal phenotypes that were associated with growth and guidance defects. When tested under stbm settings, Pk co-expression with GFPStbm also resulted in the loss of bias, together with axon growth and guidance defects. In stbm-null animals, Pk overexpression did not result in strong alterations to axonal phenotypes, although more growth phenotypes were observed and the medial bias was reduced, when compared with either condition alone.

Together with the loss-of-function paradigms, these results suggest that Pk effects are separable from Stbm and Fz. Pk has complex, context-dependent interactions with other PCP components, is able to enhance and suppress the effects of dorsal and medial branching, and has additional roles in axon growth and guidance (Fig. 5; supplementary material Fig. S3).

The following results show that Dsh axonal functions are likely to be complex and unlikely to act in a single Fz or Stbm pathway. When dsh, fz double mutants were analyzed, the extent of branch errors were slightly increased and biased towards dorsal defects (Table 2). Although not the focus of this report, additional axonal phenotypes characteristic of growth and guidance defects were observed. This suggests that Dsh and Fz have separable but convergent axon branching, growth and guidance functions. I found that dsh, stbm double homozygous mutants also showed an increase in branching errors. However, there was a significant preference for medial defects (Table 2). These results suggest that dsh acts additively to both fz and stbm. The observation that dsh, stbm double mutants maintain a bias in branching phenotypes suggests that dsh does not behave identically to fz, as would be expected if Dsh acts simply as a Fz transducer.

This was confirmed further with genetic interactions using gain-of-function experiments (supplementary material Fig. S2). DshGFP overexpression resulted in a preference for dorsal branch defects, similar to Fz and Stbm overexpression (Fig. 2). When co-expressed with FzGFP, strong growth and guidance defects were observed, similar to observations in dsh, fz double mutant animals (Table 2). GFPstbm co-expression did not alter the DshGFP gain-of-function phenotype. In stbm-null animals, DshGFP overexpression showed strong axon growth and guidance phenotypes that were not observed in stbm-null or DshGFP overexpressing animals. As a bias for dorsal branch defects remained, this suggests that the Dsh gain-of-function signal is dominant over the stbm-null phenotype. In the reverse situation, GFPstbm overexpression did not significantly alter the dsh¹ phenotype. This is in contrast to GFPstbm expression in fz²¹ mutants (Fig. 2), again suggesting that Dsh signals are distinct from Fz.

DshGFP overexpression in pk mutants resulted in growth and guidance defects. However, the branching errors in this genotype showed a significant ‘switch’ towards medial defects, similar to the dsh, pk double homozygous mutant result (Table 2). Surprisingly, Dsh overexpression in fz-null animals resulted in strong enhancement of dorsal branch defects. Also, dsh¹ mutants can enhance FzGFP overexpression phenotypes, resulting in strong guidance and growth defects.

These results suggest that a balance of Dsh and Fz signals are important in axonogenesis and that there might be cross-inhibitory signals between them. These observations are in contrast to linear models of Fz-Dsh signaling in Drosophila wing PCP (Krasnow et al., 1995; Tree et al., 2002). They suggest that Dsh axonal functions are likely to be context-dependent on other PCP protein functions, i.e. able to promote or suppress dorsal and medial branch projections, as well as influence axon growth and guidance phenotypes (Fig. 5; supplementary material Fig. S3).

Branching morphogenesis is a key feature in axons and dendrites. This study shows that Wnt/PCP proteins, known for their roles in tissue polarity, are involved in two aspects of axon branch formation: stereotyped extension and collective decision-making. I found that some PCP mutants have a preference for dorsal or medial branching phenotypes. As the genetic results suggest that Fz and Stbm functions are separable, this is distinct from some tissue polarity models that have been presented for Drosophila patterning. The results suggest that PCP proteins have complex cross-regulatory effects that include the spatial regulation of Pk. A balance of their functions is key to stereotyped branching and to axon growth and guidance.

An independent study recently reported similar results for Wnt/PCP components in MB axon branching, guidance and growth (Shimizu et al., 2011). When comparing the results with those described here, most of the data are consistent in both studies. Slight differences, such as the immunohistochemistry results and the role of Prickle, might be a result of differences in preparative methods, stages at which animals were analyzed and the genetic background of the animals used (see Materials and methods).

Branched axons and dendrites tend to follow a bifurcation motif (Acebes and Ferrus, 2000). In MB axons, this corresponds to a single dorsal and a single medial projection. Here, I show that PCP proteins control the extension of axon branches, rather than the initial branch formation. The working model is that growing axons can initially reach the axon peduncle and bifurcate independently of the Wnt/PCP pathway (Fig. 5A). Once the nascent branches have formed, Fz and Stbm direct the stereotyped extension of branches. Fz is the major determinant of dorsal projections (Fig. 5B). By contrast, Stbm predominantly controls medial projections. My results suggest that these are partly redundant systems; dorsal and medial projections also require minor contributions from Stbm and Fz, respectively (Fig. 5B). In the absence of both, branching errors are increased and randomized. In some cases, both extensions are lost (D^–M^–, also categorized as growth). Axon branching relies mainly on mutual cooperation between Fz and Stbm. In some cases, antagonism between Stbm and Fz exists, although there is no evidence of Stbm inhibiting Fz signals (supplementary material Fig. S3). These results are reminiscent of observations in Drosophila showing that Fz and Stbm can exhibit redundant and cooperative PCP functions in the wing as well as in sensory organ precursor cell fate specification (Strutt and Strutt, 2007; Chen et al., 2008; Gomes et al., 2009).

MB axons follow a branching mode that reflects collective behavior. The initial results suggest that PCP mutant phenotypes are recessive. Next to wild-type axons, PCP mutant axons tend to adopt a wild-type branched pattern. However, with the GFPstbm rescue paradigm, the results together suggest that it is the collective mix of mutant and non-mutant neurons that determines whether an axon branches or not. How is this regulated? One possibility is that as MB axons are fasciculated, cell-cell contacts can regulate collective branching behavior. Although axons are long and thin, it is unlikely that 1000 branched αβ neurons in each hemisphere are in contact with each other. It is more likely that, as in other Drosophila tissues, it is the behavior of cells that are widely interlinked together that is controlled by PCP factors (Classen et al., 2005; Zallen, 2007). The non-autonomous results presented here differ significantly from previous PCP studies. Previous results in Drosophila wing and eye show that PCP mutations result in both cell-autonomous and non-autonomous tissue polarity defects (Adler, 2002; Strutt, 2003; Lawrence et al., 2007; Vladar et al., 2009). Here, PCP branch functions are not fully cell-autonomous. Furthermore, if domineering non-autonomy exists, it does not strongly affect wild-type axon branching.

Despite the uncertainty about its role, asymmetric organization of PCP proteins is a key feature in tissue polarity (Klein and Mlodzik, 2005; Jones and Chen, 2007; Wang and Nathans, 2007; Strutt and Strutt, 2009; Vladar et al., 2009). However, there are variations to this theme. For example, although Fz and Stbm are often on opposite poles, in mouse eye lens fiber cells they appear in the same tissue pole (Sugiyama et al., 2010). In zebrafish and ascidian models in which PCP signals control the convergence and extension of cells in the developing notocord, PCP protein localizations are variable (Jiang et al., 2005; Ciruna et al., 2006). These and other observations have led to the suggestion that in motile cells, PCP proteins might be dynamically regulated (Zallen, 2007). In short, despite PCP mutants having different preferences in dorsal or medial branch defects, no obvious PCP protein asymmetry can be found between axon branches. One possibility is that these proteins are regulated dynamically, therefore such differences are not readily observable in fixed tissue. Another possibility is that distinct branching fates might arise from differences in PCP effector functions between branches.

The observation that Pk axonal branch levels are reduced in fz and stbm mutants suggests that Pk regulation forms part of PCP branch functions. This is likely to be post-translational and axonal-specific as Pk localization in MB dendrites is unaffected by fz or stbm (J.N., unpublished). This is similar to Drosophila wing patterning for which Fz loss also results in a cell-autonomous reduction of Pk (Tree et al., 2002). However, in the wing, Dsh, Fz and Stbm also regulate the intercellular accumulation of Pk at the cell cortex, along the pole vertices (Tree et al., 2002; Strutt, 2003). Also, in contrast to some wing models (Tree et al., 2002; Amonlirdviman et al., 2005), this study shows that Fz and Stbm branch functions can still occur independently of Pk.

Several results suggest that PCP proteins function early during axonal development. First, Pk expression is higher in the αβ core region, where newborn developing MB axons are located (Kurusu et al., 2002; Tanaka et al., 2008). Further evidence from the stbm rescue studies shows that GFPstbm expression in the αβ core alone is sufficient to rescue axon branching (supplementary material Table S1). Also, branching defects were apparent in PCP mutants at early stages of development (supplementary material Fig. S1). Lastly, Pk and Dsh αβ localization can also be detected at these stages (supplementary material Fig. S1).

How Wnt/PCP proteins interact has been studied extensively in Drosophila. Here, I show that the MB axon phenotypes reveal some novel features of PCP signaling interactions. As already mentioned, Fz and Stbm play partly redundant, parallel roles. Although Pk controls axon branching and is regulated by Fz and Stbm, it does not account for all Fz and Stbm functions. This is also evident from other PCP studies (Lawrence et al., 2004; Strutt and Strutt, 2007). However, Pk does act additively to Fz and Stbm and has convergent as well as antagonistic functions with Dsh. Also, I show that Dsh does not simply transduce Fz or Stbm signals. This was also reported in some studies (Strutt and Strutt, 2002; Lawrence et al., 2004; Strutt and Strutt, 2007). The working model is that both Fz and Stbm are likely to function in part through Dsh and Pk (Fig. 5B). Based on further experiments, their effects are likely to be complex and context-dependent on the activity levels of other PCP proteins (Fig. 5B; supplementary material Fig. S3).

Previous reports suggest that these factors interact physically with each other. There is evidence that the extracellular regions of Stbm and Fz can physically bind to each other (Wu and Mlodzik, 2008), although this is debated (Chen et al., 2008; Lawrence et al., 2008; Strutt and Strutt, 2008). Physical associations between Stbm-Dsh and Fz-Dsh have been documented in vertebrate and fly proteins (Park and Moon, 2002; Bastock et al., 2003; Wong et al., 2003; Torban et al., 2004; Jenny et al., 2005). Apart from Stbm, Pk also interacts physically with Dsh (Tree et al., 2002; Jenny et al., 2005). Recent reports suggest that mouse Stbm (Vangl2) and Dsh (Dvl1) interact antagonistically to regulate Frizzled3 (Fzd3) phosphorylation (Shafer et al., 2011). Whether these interactions affect axon branching is currently unknown. Nonetheless, novel Wnt/PCP interactions are involved in MB axon patterning.

In conclusion, this study demonstrates that Wnt/PCP proteins are key regulators in the stereotyped growth of axon branches. Given that branched growth is intrinsic to neuronal morphogenesis, Wnt/PCP components and their reported interactions might also be involved synapse and dendrite formation. Such highly cross-regulated programs are probably essential for the intricate patterning of the nervous system.

J. Axelrod, R. Nusse, M. Povelones, A. Penton, D. Strutt, F. Schweisguth, H. Tanimoto, T. Uemura, T. Wolff, the TRiP project at Harvard Medical School, the Vienna Drosophila RNAi (VDRC) and the Bloomington Drosophila Stock Center made critical reagents available for this study. I thank G. Jefferis for support and statistical analysis, together with J. Clarke, L. Luo and D. Strutt, who gave feedback on the manuscript. R. Gonzalez-Quevedo helped with the phenotypic analysis.