Initiation of synapse formation by Wnt-induced MuSK endocytosis

Synapse formation requires the precise alignment of pre- and postsynaptic elements. In the case of neuromuscular synapses, this process involves proper navigation of motor axons from the spinal cord to their muscle targets as well as the spatial alignment of these motor axons with acetylcholine receptors (AChRs) on the muscle surface (reviewed in Wu et al., 2010). Neuromuscular synapse development can be subdivided into two functionally distinct phases: an early and a late phase (Flanagan-Steet et al., 2005; Panzer et al., 2006; Wu et al., 2010; Zhang et al., 2004). The late phase is characterized by the transformation of motor growth cones into presynaptic nerve terminals which release signals, including Agrin, that are essential for postsynaptic stabilization and maturation (Campanelli et al., 1991; Godfrey et al., 1984; Ruegg et al., 1992; Wu et al., 2010). By contrast, the early phase is independent of the nerve and nerve-derived factors (Lin et al., 2001; Yang et al., 2001; Yang et al., 2000). During the early phase, even before the arrival of motor axons, AChRs accumulate in the center of muscle cells, precisely where the first synapses will form in a process termed AChR pre-patterning (Lin et al., 2001; Yang et al., 2001). Importantly, ‘blocking’ AChR pre-patterning through temporal inactivation of the muscle-specific kinase (MuSK) does not eliminate the formation of functional, mature synapses (Jing et al., 2009). Instead of navigating the center of muscle cells, motor axons now invade lateral muscle cell territories where they form ectopic synapses, suggesting that events during the early phase of neuromuscular development determine where synapses will form (Jing et al., 2009).

The receptor tyrosine kinase MuSK is a key player in determining where along the muscle cell synapses will form (DeChiara et al., 1996; Glass et al., 1996). In both mice and zebrafish, MuSK is expressed in early postsynaptic muscle cells, and in embryos lacking MuSK, AChRs fail to cluster in the center of muscle cells, and motor axons invade lateral muscle territory they usually avoid (DeChiara et al., 1996; Lin et al., 2001; Yang et al., 2001; Zhang et al., 2004). Overexpression of MuSK is sufficient to establish muscle pre-patterning through ligand-independent autoactivation (Kim and Burden, 2008), and in zebrafish the secreted glycoprotein Wnt11r binds to the MuSK ectodomain, and functional knockdown of wnt11r leads to defects in AChR pre-patterning and motor axon guidance (Jing et al., 2009). Although these studies suggest that Wnt signals play a crucial role in vertebrate neuromuscular development, the functional requirement for Wnt has not been confirmed using genetic mutants, and the mechanism by which Wnt signals initiate synapse formation has not been established.

Here, we characterize null mutants for wnt11r and show that wnt11r and wnt4a are essential for triggering relocalization of the MuSK receptor from the cell membrane to recycling endosomes located in the center of early muscle cells beneath future synaptic sites. We provide compelling evidence that MuSK localization to recycling endosomes activates a signaling cascade best known for its role in mediating planar cell polarity (PCP). PCP components colocalize with the internalized MuSK receptor and inhibition of selective PCP components results in a reduction of AChR pre-patterning and axon guidance errors. We propose a model in which Rab11-mediated trafficking positions a signaling complex consisting of the MuSK receptor and PCP components to the center of muscle cells to initiate synapse formation.

All experiments were conducted according to an Animal Protocol fully approved by the University of Pennsylvania Institutional Animal Care and Use Committee (IACUC) on 2 February, 2011; protocol number 803446. Veterinary care is under the supervision of the University Laboratory Animal Resources (ULAR) of the University of Pennsylvania.

All embryos used in this study were raised at 28°C for the required amount of time. Wild-type fish used for experiments were TLF, and mutants used were musk (unp)^tbr307(Zhang and Granato, 2000; Zhang et al., 2004), wnt11r^fh224(G94*)(Banerjee et al., 2011) and Tg(mnx1:GFP)ml2 (Flanagan-Steet et al., 2005). Transgenic fish were used alone, in combination with each other or in combination with various mutant backgrounds.

Embryos were fixed and stained as described previously (Zeller et al., 2002) and labeling of AChRs was achieved via the method described by Jing et al. (Jing et al., 2009). The following antibodies and dilutions were used: znp-1 [1:200, Developmental Studies Hybridoma Bank (DSHB)], myc (9E10, 1:1000, Covance), F59 (1:20, DSHB). Embryos were imaged with LCS (Leica) and IX81 (Olympus) confocal microscopes.

For puncta and AChR counts, confocal images were projected into a single plane and converted to a 16-bit image using Metamorph (Molecular Devices). Puncta were counted using the ‘count nuclei’ function, with the following parameters for each condition: AChRs imaged at 20× magnification, AChRs imaged at 60× magnification and endocytosed puncta imaged at 60× magnification with minimum/maximum lengths of 5/30, 5/100 and 1/10, respectively, and minimum average intensities of 30, 50 and 100, respectively. The results were imported to Microsoft Excel and Graphpad Prism for statistical analysis and plotting. Single confocal slices of images of RNA-injected embryos were converted to 16-bit images using Metamorph, and the maximum pixel intensity at three membranes per image was recorded manually for both color channels.

Colorimetric in situ hybridization was performed as described by Schneider and Granato (Schneider and Granato, 2006) with the previously published wnt4a in situ probe ‘eu648’ (Thisse and Thisse, 2005) and the previously published full-length wnt11r probe (Jing et al., 2009).

Destination vectors containing either GFP- or myc-tagged constructs were microinjected into one-cell-stage embryos as previously described (Thermes et al., 2002). wnt11r, unplugged-GFP and unpluggedΔCRD-GFP mRNA were in vitro transcribed from NotI-linearized PCS2+ expression vectors using SP6 mMessage mMachine Kit (Ambion), diluted in 0.1 M KCl and microinjected into embryos at the one-cell stage. A previously published wnt4a translation-blocking morpholino (Matsui et al., 2005) targeting the following sequence was injected (7.2 ng) at the one-cell stage: CTCCGATGACATCTTTAGTGGAATC.

Rab11a-GFP and rab11(S25N) in PCS2+ (Clarke et al., 2006) were gifts from Clare Buckley (Kings College London, London, UK) and Gwyn Gould (University of Glasgow, Glasgow, UK). mDiversin in eYFP C1 was a gift from Tobias Schäfer (University Hospital Freiburg, Freiburg, Germany) (Haribaskar et al., 2009). Daam1-GFP in PCS2+ was a gift from Raymond Habas (Temple University, Philadelphia, PA, USA) (Liu et al., 2008). XDishevelled-GFP in PCS2+ was a gift from Peter Klein (University of Pennsylvania, PA, USA), dishevelled(DEP+) was cloned as previously described (Jing et al., 2009) and dishevelled(DIX) was subcloned from the full length construct using the following primers: AAAAGAATTCGCCACCATGGCGGAGACTAAAGTGATTTAC (forward) and AAAAAGAATTCTTAGAGCTCTCTGACCTCTGTAGCAGF (reverse). NDaam1-HA in pcDNA3 was a gift from Yasuyuki Kida (Tohoku University, Japan) (Kida et al., 2007). CamKII(K42M) and rhoA(N19) in pCS2+ were gifts from Juan Carlos Izpisua Belmonte (Salk Institute, CA, USA) (Matsui et al., 2005). Unplugged, unpluggedΔCRD (Jing et al., 2009) and wnt11r were amplified from genomic DNA and cloned into the PCS2+ vector. Once cloned into PCS2+ (or in the case of diversin-YFP and daam1b in eYFP C1 and pcDNA3 vectors, respectively), the constructs were tagged with either GFP or 5×Myc using standard cloning procedures. Fusion constructs were PCR amplified from PCS2+ and cloned into pENTR/D-TOPO (Invitrogen, USA) using the following primers: 5′-CACCCAAGCTACTTGTTCTTTTTGC-3′ (forward) and 5′-GGTACCGGGCCCAATGCATT-3′ (reverse). Constructs were then transferred to a homemade Gateway destination vector containing either the 3.8unp promoter (below) or the smyhc1 promoter using the Gateway LR Clonase II enzyme mix (Invitrogen, USA). The 3.8unp promoter, which includes the 5′UTR and first intron of musk, was amplified from Fosmid CH1073-179D18 (BACPAC Resources, Oakland, CA, USA) using primers 5′-TGGTCTTTCGTGGAGTTC-3′ (forward) and 5′-TGGTTGTGAGAAGTAAAGATCTG-3′ (reverse).

Cell transplants were performed as described previously (Zeller and Granato, 1999) with the modification that embryos were fixed at 20 somites, permeabilized for 5 minutes with 0.1% collagenase, stained for 1 hour with fluorescently conjugated bungarotoxin and mounted in Vectashield for confocal imaging.

We reported previously that Wnt11r binds the MuSK receptor and that morpholino-mediated knockdown of wnt11r recapitulates the pre- and postsynaptic defects observed in musk (unp)-null mutants (Jing et al., 2009). These results suggest that Wnt11r is a ligand for MuSK, yet the incomplete phenotypic penetrance observed in wnt11r morphants left open the possibility of a partial morpholino knockdown, maternal contribution and/or additional ligands. To distinguish between these possibilities, we characterized AChR pre-patterning and axon guidance in embryos carrying an early stop codon in wnt11r, likely to represent a wnt11r-null allele (G94^*) (Banerjee et al., 2011). Prior to the arrival of motor axons (20-somite stage), AChR clusters accumulate in a very narrow area in the center of individual muscle fibers (Fig. 1A). By contrast, in wnt11r mutants AChR clusters occupied a much broader area (∼61% of hemisegments; Fig. 1E, quantified in 1L) or completely failed to accumulate AChRs to the muscle center (∼39% of hemisegments; Fig. 1F). This failure to accumulate AChR clusters was observed in 100% of musk (unp) mutant hemisegments (Fig. 1C). Importantly, overall muscle morphology in wnt11r-null mutants was indistinguishable from wild-type siblings (supplementary material Fig. S1A,B).

Analysis of motor axons at 26 hours post fertilization (hpf) revealed that in wnt11r mutants, motor axons branched and invaded the lateral muscle territories from which they are excluded in wild-type embryos (Fig. 1G; n=338, ∼26% of hemisegments), identical to the phenotype observed in musk (unp) mutants (Fig. 1D; n=148, ∼85% of hemisegments). To test whether maternally provided wnt11r function might account for the quantitatively weaker pre- and postsynaptic phenotypes [compared with musk (unp) mutants], we examined mutant embryos derived from homozygous mutant wnt11r parents. These embryos displayed AChR and axonal phenotypes at frequencies very similar to those derived from wnt11r heterozygote parents (23% guidance defects in mutants derived from a heterozygote cross compared with 26% in mutants from a homozygote mutant cross; n=100 and n=338, respectively), suggesting that maternal wnt11r contributions are dispensable for synapse initiation. Thus, wnt11r-null mutants recapitulate the pre- and postsynaptic defects observed in musk (unp) null mutants (Fig. 1C,D), albeit with lower penetrance, suggesting the existence of additional MuSK ligands.

Given the great extent of redundancy among vertebrate Wnts, we hypothesized that other Wnts might be compensating in the absence of wnt11r. Based on its mRNA expression pattern in the vicinity of muscle cells that first form the AChR pre-pattern, we focused on wnt4a (supplementary material Fig. S1E,F). Using a previously characterized translation-blocking morpholino (Matsui et al., 2005), we found that knockdown of wnt4a by itself did not produce detectable perturbations of AChR pre-patterning or axon guidance (Fig. 1H,I). By contrast, injection of the same dose of wnt4a morpholino into wnt11r mutants exacerbated the partial wnt11r-null phenotype. In all wnt11r/wnt4a-deficient somitic segments, AChR receptors failed to accumulate in the center and instead formed small aggregates randomly along muscle fibers (Fig. 1J). These aggregates resemble AChR ‘hot spots’ known to form spontaneously along aneural muscle cells (Fischbach and Cohen, 1973; Sytkowski et al., 1973). Concomitantly, in 70% of wnt11r/wnt4a-deficient somitic segments, motor axons displayed musk (unp)-like branching defects (Fig. 1K), again without affecting overall muscle cell morphology (supplementary material Fig. S1C,D). Thus, loss of wnt11r and wnt4a mimics the AChR pre-pattern and axonal defects observed in musk (unp)-null mutants (Zhang and Granato, 2000), consistent with the idea that, in vivo, AChR pre-pattern requires the MuSK receptor and multiple Wnt ligands.

We next investigated how wnt11r and wnt4a initiate synapse formation. We had shown previously that ubiquitous expression of MuSK-Myc along the entire length of muscle cells did not alter the centrally localized AChR pre-pattern (Jing et al., 2010), suggesting that additional factors or processes might determine where along the muscle cell MuSK signals. In the early zebrafish embryo, Wnts recruit the Frizzled receptor and the obligate intracellular signaling component Dishevelled to the plasma membrane, thereby generating local cell contacts important for gastrulation movements (Witzel et al., 2006). To test whether Wnts might alter the cellular distribution of MuSK, one-cell-stage embryos were injected with mRNAs encoding GFP-tagged MuSK and the membrane marker mCherry-CAAX, in the absence or presence of wnt11r mRNA. Four and a half hours later, at dome stage, injected embryos were analyzed for the cellular localization of MuSK-GFP protein (Fig. 2). Importantly, these early embryos do not express wnt11r endogenously, providing a ‘Wnt-naive’ environment (Nojima et al., 2010).

In the absence of wnt11r mRNA, MuSK-GFP is highly enriched at the cell membrane, as expected for a receptor tyrosine kinase (Fig. 2A-A″). By contrast, in the presence of wnt11r mRNA, levels of MuSK-GFP at the cell membrane were dramatically reduced (Fig. 2B-B″, quantified in 2E). Moreover, deletion of the cysteine-rich domain (CRD), which is essential for Wnt11r binding to the MuSK-GFP ectodomain (Jing et al., 2010), did not affect steady-state membrane levels of MuSK-GFP in the absence of wnt11r (Fig. 1C-C″), but reduced wnt11r responsiveness (Fig. 1D-D″, quantified in 1F). These data demonstrate that exposure to wnt11r alters membrane localization of MuSK, suggesting that Wnt11r affects MuSK protein stability or dynamics, possibly by modulating localization at the plasma membrane.

To explore the relationship between Wnt exposure and MuSK localization in the relevant cellular context, we examined MuSK localization in early muscle cells, the adaxial muscle cells, which, in the zebrafish embryo, form the first AChR pre-pattern. musk mRNA is detectable prior to and during the time period of AChR pre-patterning in adaxial muscle cells and becomes undetectable soon afterwards (Zhang et al., 2004). To express musk in adaxial muscle cells, we cloned 3.8 kb of genomic DNA directly upstream of the musk translational start site (3.8unp) and used this to drive the expression of GFP or musk-GFP.

Injection of smyhc1:GFP, a bona fide promoter for adaxial muscle (Elworthy et al., 2008), or the 3.8musk promoter driving GFP, resulted in GFP expression throughout the cytoplasm of stochastically labeled wild-type adaxial muscle cells (Fig. 3A,B). Stochastic labeling results in individual cells that retain and express the injected DNA construct, surrounded by cells which do not retain the DNA and hence do not express the construct (Downes et al., 2002). Injection of 3.8musk driving musk-GFP (3.8musk:MuSK-GFP) into musk (unp)-null mutant embryos restored AChR pre-patterning, demonstrating that this construct is functional (supplementary material Fig. S2B,B′). Injection of 3.8musk:MuSK-GFP into wild-type embryos resulted in stochastic expression of the fusion protein in adaxial muscle cells, but unlike expression of only GFP under the same promoter, MuSK-GFP localized to puncta in the center of adaxial muscle cells before and during the time period of AChR pre-patterning, but not post-AChR pre-patterning, when these cells undergo a medial-to-lateral migration (Flanagan-Steet et al., 2005) (Fig. 3C-E). Live imaging revealed that these MuSK-GFP-positive puncta were highly dynamic and that they appeared to be actively maintained in the muscle center, as puncta that strayed from the center rapidly moved back to the central position (supplementary material Movie 1).

To determine whether these puncta consist of cytoplasmic protein aggregates or whether they represent a regulated pool of endocytosed protein, we co-injected into wild-type embryos 3.8musk:MuSK-RFP with rab7-GFP, a late endosome marker, and rab11-GFP, a marker for exocytic/recycling endosomes (reviewed by Hutagalung and Novick, 2011). Rab7-positive endosomes were present throughout the entire length of individual adaxial muscle cells, and only a very small fraction of Rab7-positive endosomes colocalized with MuSK (supplementary material Fig. S2D-D″). Conversely, Rab11-positive vesicles were restricted to the central portion along the anterior-posterior axis of both wild-type and musk (unp) mutant muscle cells, and the majority colocalized with MuSK (Fig. 3F-F″′, quantified in supplementary material Fig. S2G). Thus, MuSK expressed under its endogenous promoter localizes just before and during the time of AChR pre-patterning to Rab11-positive vesicles at the center of adaxial muscle, precisely where AChRs accumulate.

Rab11 is a member of the large family of small GTPases, and regulates both the exocytic biosynthetic and the recycling pathway (Chen et al., 1998; Ren et al., 1998; Satoh et al., 2005; Sonnichsen et al., 2000; Ullrich et al., 1996). To determine whether MuSK localization to exocytic/endocytic vesicles depends on rab11 function, we expressed a dominant-negative, GDP-restricted rab11, rab11(S25N) (Ren et al., 1998), stochastically in adaxial muscle cells. To ensure high levels of expression that are essential for maximal efficacy of dominant-negative proteins, we used the ‘stronger’ smyhc1 adaxial promoter rather than the weak 3.8unp promoter. As shown in Fig. 4, disruption of rab11 function resulted in uniform MuSK-RFP distribution along the entire length of adaxial muscle cells, suggesting that MuSK protein localization depends on rab11-mediated trafficking (Fig. 4A,A′, quantified in 4E). Importantly, Rab11 localization is unaffected in musk (unp) mutant adaxial cells (supplementary material Fig. S2G). Similarly, ‘global’ overexpression of wild-type rab11-GFP via mRNA injection or high level expression of wild-type rab11-GFP in individual muscle cells via the smyhc1 promoter did not perturb AChR pre-patterning or musk-mKate localization (supplementary material Fig. S2E-F′; data not shown).

If the localization of MuSK to Rab11-positive vesicles is functionally important for wnt11r- and wnt4a-induced synapse initiation, then several predictions can be made. First, the disruption of rab11-mediated processes in adaxial muscle cells should affect AChR pre-patterning and motor axon guidance. Indeed, when we expressed rab11(S25N)-myc in individual adaxial muscle cells, we observed a decrease in AChR clustering in these cells but not in the surrounding wild-type cells (38%, n=13 cells in five embryos; Fig. 4B,B′). Similarly, expression of rab11(S25N)-myc in individual muscle cells resulted in musk (unp)-like axonal defects in 32% of hemisegments (n=50).

A second prediction is that in the absence of Wnt signals, the subcellular localization of MuSK to recycling endosomes would be affected. Consistent with our findings that AChR pre-patterning requires both wnt11r and wnt4a function (Fig. 1), analysis of wnt11r mutant/wnt4a morphant embryos revealed a nearly complete shift in the subcellular localization of MuSK-GFP from Rab11-positive vesicles to the cell membrane (Fig. 4C-C″,D-D″, quantified in 4F). These results demonstrate that Wnt signaling is required to traffic MuSK-GFP from the cell membrane into Rab11-positive vesicles. Moreover, the accumulation of MuSK-GFP at the cell membrane of wnt11r mutant/wnt4a morphant embryos strongly argues that MuSK-GFP and Rab11 double-positive vesicles are not exocytic vesicles, but do indeed represent recycling endosomes. Combined, our data show that wnt11r- and wnt4a-dependent trafficking of MuSK from the cell membrane to Rab11-positive endosomes is essential for AChR pre-patterning.

Given the striking subcellular localization of MuSK-GFP in response to Wnt signals, we examined the localization of several components of the noncanonical Wnt signaling pathway. We previously used a dominant-negative version of dishevelled, dishevelled(DEP+), that specifically blocks noncanonical Wnt signaling (Axelrod et al., 1998; Heisenberg et al., 2000; Jing et al., 2009; Wallingford et al., 2000), to demonstrate that dishevelled-dependent noncanonical signaling in adaxial muscle is crucial for restricting AChR clusters and motor axons to the central portion of adaxial muscle cells. Although it is known that Dishevelled can bind the juxtamembrane/kinase domain of the MuSK receptor (Luo et al., 2002), the subcellular localization of Dishevelled and other PCP core components during vertebrate neuromuscular synapse formation has not been established. To examine the subcellular localization of PCP core components, we expressed Daam1-GFP, Ankrd6-YFP and Dishevelled-GFP in individual adaxial muscle cells and examined their localization just prior to and during the period of AChR pre-patterning. To avoid nonspecific protein aggregates, we used the 3.8musk promoter, which drives protein expression at much lower levels than the ‘stronger’ smyhc1 promoter. Indeed, expression of most PCP core proteins under the control of the smyhc1 promoter resulted in nonspecific protein aggregates (data not shown).

Confocal analysis revealed that Daam1-GFP, Ankrd6-YFP and Dishevelled-GFP localize to vesicle-like puncta that are strongly enriched in the central region of adaxial muscle cells, identical to the localization of MuSK-GFP (Fig. 5A-C, quantified in supplementary material Fig. S3A). Moreover, stochastic expression of a dominant-negative version of daam1 (NDaam1) (Kida et al., 2007) and rhoA [rhoA(N19)] (Matsui et al., 2005) in adaxial muscle cells reduced AChR pre-patterning and caused musk (unp)-like axonal guidance defects, confirming that PCP pathway components play a crucial role in this process (supplementary material Fig. S3E,F,I-J′). Importantly, blocking the canonical Wnt pathway or the Wnt/Ca²⁺ pathway did not affect AChR clustering or axon guidance (supplementary material Fig. S3C,D,G-H′). Finally, co-expression of dishevelled-GFP and musk-mKate revealed a high degree of colocalization between both proteins, strongly suggesting that in adaxial muscle cells, PCP core components such as Dishevelled associate with recycling endosomes (Fig. 5E-E″).

In Drosophila, PCP core components are dependent on one another for their correct subcellular localization. For example, loss of one PCP core component, such as the frizzled receptor, disrupts localization of other core proteins (reviewed in Wu and Mlodzik, 2009). The analogous prediction for this system would be that blocking the MuSK receptor should interfere with Dishevelled-GFP localization, and vice versa. Remarkably, the subcellular localization of Dishevelled-GFP to the center of adaxial muscle cells is completely abolished in musk (unp) mutants, and instead Dishevelled-GFP is diffusely distributed throughout the cytoplasm (Fig. 5C,D, quantified in 5H). Conversely, blocking dishevelled-dependent noncanonical signaling in adaxial muscle cells leads to a significant redistribution of MuSK-GFP from the endosomal compartment in the muscle center to the cell membrane (Fig. 5F-G″), identical to what we observed when blocking wnt11r and wnt4a function (Fig. 4D-D″). Importantly, localization of Rab11-GFP is independent of dishevelled function, suggesting that rab11 probably functions ‘upstream’ of dishevelled, and that the dependence of MuSK localization on dishevelled is highly specific (supplementary material Fig. S2G). Thus, several core PCP proteins colocalize with MuSK to presumptive endosomal organelles in the center of adaxial muscle cells, precisely where the AChR pre-pattern forms. Moreover, MuSK and Dishevelled are dependent on one another for proper localization, consistent with the idea that both participate in a PCP-like signaling cascade to initiate synapse development.

Genetic mutants and knockouts have demonstrated that Wnts play a crucial role in the assembly and plasticity of central synapses (Budnik and Salinas, 2011; Ciani and Salinas, 2005; Hall et al., 2000; Lucas and Salinas, 1997; Wu et al., 2010; Yang et al., 2010). Wnts are also involved at the invertebrate neuromuscular synapse (reviewed by Davis and Ghosh, 2007; Klassen and Shen, 2007). However, the situation at the vertebrate neuromuscular junction (NMJ) is less well defined. Mice mutant for the frizzled co-receptor lrp4 lack all AChR pre-patterning and mature NMJ synapses (Kim, N. et al., 2008; Weatherbee et al., 2006; Zhang et al., 2008), and knockdown of zebrafish wnt11r significantly reduces AChR pre-patterning (Jing et al., 2009). Although these reports strongly suggest that Wnts are crucial for vertebrate neuromuscular synapse development, Wnt knockouts or genetic mutants with defects in neuromuscular synapse development have not yet been reported.

Here, we demonstrate that wnt11r-null mutants display profound pre- and postsynaptic defects during the early phase of synapse development, without obvious defects in muscle development. Interestingly, we find that only the combined inactivation of wnt11r and wnt4a results in a complete loss of AChR pre-patterning, almost identical to that observed in musk (unp) mutants (Fig. 1C,J). Wnt11r mRNA is expressed in the lateral mesoderm (Groves et al., 2005; Matsui et al., 2005) (supplementary material Fig. S1G,H), consistent with a paracrine mode of action, whereas wnt4a is expressed in adaxial muscle, suggesting an autocrine mode of operation (supplementary material Fig. S1E,F). At the Drosophila NMJ, Wg and Wnt5 function in a well-documented autocrine manner (Liebl et al., 2008; Packard et al., 2002) and future experiments are required to determine whether, in zebrafish, muscle-derived wnt4a activates the muscle-expressed MuSK receptor directly. Combined, our results uncover the first genetic requirement for Wnts during vertebrate NMJ synapse development and demonstrate that multiple Wnt signals initiate synapse formation.

Once thought to terminate transmembrane receptor signaling, the endocytic pathway is becoming increasingly appreciated for its ability to activate signaling of surface receptors such as the Notch, EGF and Eph receptors (reviewed by Gagliardi et al., 2008; Pitulescu and Adams, 2010), as well as chemokine receptors such as Cxcr2. For example, it has been shown that in response to ligand stimulation, the Cxcr2 receptor is internalized into Rab11-positive recycling endosomes, and that reducing Cxcr2 recycling diminishes Cxcr2-mediated chemotaxis (Fan et al., 2004). Similarly, Wnt-mediated activation of Frizzled3 on axonal growth cones triggers Frizzled3 endocytosis and thereby initiates PCP signaling within growth cones (Shafer et al., 2011). This and other experiments have contributed to the current view that upon ligand binding, receptor endocytosis can lead to degradation or to signaling, depending on the endocytic compartment to which the receptor is sorted (reviewed in Gagliardi et al., 2008). In the case of receptor tyrosine kinases, there is accumulating evidence that receptor endocytosis is required for signal transduction (Jing et al., 1992; Vieira et al., 1996; Yang et al., 2005).

Previous studies have shown that in response to ligand stimulation, MuSK undergoes rapid endocytosis in cultured myoblasts (Zhu et al., 2008). We find that in vivo and during the time of synapse initiation, MuSK is internalized from the cell membrane into Rab11-positive endosomes, and that this process is dependent on wnt11r and wnt4a. Moreover, rab11 function is important for proper MuSK protein localization and AChR pre-patterning, demonstrating the importance of endosome recycling to initiate MuSK-dependent processes (Fig. 4). Although we cannot exclude the possibility that rab11 exerts its function directly on AChR trafficking, recycling of AChR receptors has been observed only at mature NMJs and only in response to muscle stimulation (Bruneau and Akaaboune, 2006; Martinez-Pena y Valenzuela et al., 2010), whereas at the stage of AChR pre-patterning when we examined rab11-dependent trafficking, NMJs have not yet formed. Combined with the observations that MuSK is crucial for localization of Dishevelled to presumptive endosomes, and that Dishevelled function in muscle cells is crucial for AChR pre-patterning and axon guidance (see below) (Jing et al., 2009), we favor the idea that rab11 regulates primarily MuSK localization and function.

Finally, we showed previously that forcing MuSK expression throughout the entire surface of muscle cells did not alter synapse initiation or position, suggesting the existence of additional regulators (Jing et al., 2009). Our finding that MuSK internalization to endosomes spatially restricts MuSK function and, hence, positions AChR clusters to the muscle center explains our previous results and strongly suggests that MuSK internalization to the spatially restricted Rab11-positive endosomal compartment determines where along the muscle cell future synapses will form.

The localization of MuSK to Rab11-positive endosomes and the requirement of rab11-mediated trafficking for AChR pre-pattern initiation suggests that the signaling proteins downstream of MuSK might localize to and be active at Rab11-positive endosomes. In fact, we find that several core components of the PCP pathway, including Dishevelled, localize to the center of muscle cells, precisely where AChR clusters accumulate, and we find that Dishevelled localization is MuSK-dependent (Fig. 5C,D). Conversely, blocking dishevelled function affects membrane-bound MuSK, preventing it from internalizing to the endosomal compartment, which ultimately results in defects in AChR pre-patterning and motor axon guidance (Fig. 5F-G″) (Jing et al., 2009). This reciprocal functional dependency is intriguingly reminiscent of how the PCP pathway patterns Drosophila wing disc cells, although the precise mechanism initiating PCP signaling there is still unresolved (reviewed by Strutt and Strutt, 2009). Consistent with our results that intracellular trafficking plays a vital role in establishing cellular polarity are recent reports on the role of sec24b, a component of the COPII complex that is essential for intracellular endoplasmic reticulum (ER)-to-Golgi protein transport (Sato and Nakano, 2007). Genetic and biochemical analyses show that during mouse embryogenesis, sec24b plays a crucial role in establishing planar cell polarity through selective trafficking of Vangl2 (Merte et al., 2010; Wansleeben et al., 2010). Our results provide compelling evidence for a model in which Wnt ligands initiate a MuSK-dependent signaling cascade in Rab11-positive recycling endosomes, where core PCP components modulate the cytoskeleton to determine the precise location of future synapses along the anterior-posterior axis of muscle cells (Fig. 6).

Specifically, we propose a model (Fig. 6) in which secreted Wnt ligands, such as wnt11r, trigger translocation of the MuSK receptor from the cell surface into Rab11-positive recycling endosomes. Importantly, these endosomes are enriched and actively maintained in the center of the muscle fiber. Once endocytosed, MuSK associates with Dishevelled (Luo et al., 2002), which is known to stimulate signaling following Wnt-induced receptor endocytosis (Chen et al., 2003; Gao and Chen, 2009). For example, upon wnt11-mediated endocytosis of the Frizzled receptor, Dishevelled triggers noncanonical signals crucial for Xenopus gastrulation (Kim, G. H. et al., 2008).

Downstream of dishevelled, the noncanonical Wnt pathway bifurcates into several branches, including into a daam1-dependent branch that mediates cytoskeletal reorganization (reviewed in Montcouquiol et al., 2006). We find that in addition to binding MuSK, Dishevelled also colocalizes with and binds to Ankrd6 (previously known as Diversin) and Daam1, two proteins with well-characterized functions in PCP signaling (Habas et al., 2001; Itoh et al., 2009; Kida et al., 2007; Moeller et al., 2006). Daam1 is a member of the Formin family of proteins and is a central player in cytoskeletal reorganization (Alberts, 2002; Wallar and Alberts, 2003), in part through RhoA (Habas et al., 2001), a small GTPase known for its ability to modify the actin cytoskeleton (Maekawa et al., 1999; Matsui et al., 1996). Consistent with this view, we find that inhibiting RhoA function in adaxial muscle cells impairs AChR clustering (supplementary material Fig. S3), although it remains unclear how modifications of the actin cytoskeleton affect AChR clustering during synapse initiation in vivo. In cultured myotubes, actin dynamics have been shown to regulate AChR trafficking and accumulation (Dai et al., 2000; Lee et al., 2009; Pato et al., 2008). Regardless of the precise mechanism, our results demonstrate that core members of the PCP pathway known to regulate cytoskeleton dynamics colocalize with MuSK and play in vivo roles during synapse initiation.

One outstanding question is the mechanism by which Rab11- and MuSK-positive endosomes localize to the center of muscle cells. Rab11 retains its subcellular localization even in the absence of musk (unp) and dishevelled (supplementary material Fig. S2G), which suggests that muscle cells have an inherent, MuSK-independent ‘polarity’. Moreover, Rab7-positive late endosomes are distributed throughout muscle cells, excluding a ‘general’ endosome localization mechanism. The restricted localization of Rab11 is reminiscent of the situation in Drosophila sensory organ precursor (SOP) cells, where Rab11 is localized to only one of two daughter cells following cell division, independent of the Par complex (Emery et al., 2005). Instead, Rab11 localization requires Nuclear fallout (also known as Arfophilin, Eferin or Rab11-FIP3), which has been proposed to provide a link between rab11 and the dynein light intermediate chain (Horgan et al., 2010). Alternatively, the position of the nucleus might determine Rab11 localization, and thereby synapse position, although altering muscle nuclei position through Syne-1 knockdown does not interfere with neuromuscular synapse maturation (Grady et al., 2005). If rab11 effector proteins and/or additional factors play a role in the localized accumulation of AChRs and ultimately synapse initiation remains an open question. Nonetheless, our results suggest a basic mechanism by which Wnt signals shape synapse localization through localized receptor endocytosis.

Thanks to C. Buckley, G. Gould, T. Schafer, R. Habas, Y. Kida, J. C. Izpisua Belmonte and P. Klein for providing constructs.