Calsyntenin 1-mediated trafficking of axon guidance receptors regulates the switch in axonal responsiveness at a choice point

Attractive and repulsive guidance cues cooperate in the navigation of growing axons to their correct targets during development. On their way to the final target, axons may contact one or several intermediate targets. At each one of them, growth cones need to change their responsiveness in order to overcome the attraction derived from the intermediate target and to continue their journey.

The dI1 population of commissural neurons has been widely used to study the molecular mechanisms of axon guidance (Chédotal, 2011; Nawabi and Castellani, 2011). They extend their axons ventrally towards the floor plate, the structure that forms the midline of the neural tube. During this first stage of growth, axons are repelled by BMPs (Augsburger et al., 1999) and Draxin (Islam et al., 2009), chemorepellents derived from the roof plate. At the same time, axons are attracted by the floor plate-derived chemoattractants Netrin (Kennedy et al., 1994) and Shh (Charron et al., 2003). The interaction of axonin 1 (contactin 2) and NgCAM (L1CAM) is responsible for axonal fasciculation along the ventral pathway, whereas the interaction between axonal axonin 1 and floor plate NrCAM is responsible for floor plate entry (Stoeckli and Landmesser, 1995; Stoeckli et al., 1997). RabGDI (GDI1)-dependent insertion of Robo1 into the growth cone membrane triggers sensitivity to the midline-associated repellent Slit1 (Philipp et al., 2012). Upon floor plate exit, post-crossing commissural axons turn rostrally guided by two opposing morphogen gradients (Stoeckli, 2006). Wnt gradients in mouse (Lyuksyutova et al., 2003) and in chick (Domanitskaya et al., 2010; Avilés and Stoeckli, 2016) were shown to attract post-crossing axons. At the same time, axons were repelled by a rostral^low-caudal^high gradient of Shh (Bourikas et al., 2005). Shh shapes the Wnt gradient by inducing the expression of the endogenous Wnt antagonists Sfrp1 and Sfrp2 in a rostral^low-caudal^high gradient (Domanitskaya et al., 2010). Thus, Shh has multiple roles in commissural axon guidance: it attracts pre-crossing axons to the intermediate target in a Boc- and Smo-dependent manner (Charron et al., 2003; Okada et al., 2006). Then, it affects post-crossing axons both directly and indirectly by shaping the Wnt activity gradient. The direct repulsive effect on post-crossing axons is mediated by Hhip (Bourikas et al., 2005). Recently, we demonstrated that Shh itself regulates the expression of Hhip in a glypican 1-dependent manner (Wilson and Stoeckli, 2013).

The transcriptional regulation of Hhip by its own ligand, Shh, represents one mechanism explaining the switch in axonal responsiveness at the intermediate target. Furthermore, Shh was suggested to induce the responsiveness to semaphorin 3B, a floor plate-derived repellent (Parra and Zou, 2010). Semaphorin 3B binds to a receptor complex consisting of neuropilin 2 and plexin A1. On pre-crossing axons, plexin A1 is destabilized and therefore not expressed on the growth cone surface due to proteolysis by Calpain (Nawabi et al., 2010). Upon axonal arrival at the floor plate, Calpain is inhibited in an NrCAM- (Nawabi et al., 2010) and GFRα1/NCAM-dependent manner (Charoy et al., 2012). This, in turn, results in stabilization and surface expression of plexin A1 and thus responsiveness to the floor plate-derived repellent semaphorin 3B as well as Slit1 (Delloye-Bourgeois et al., 2015).

In the visual system, responsiveness to semaphorin 3A was shown to be regulated at the post-transcriptional level by miR124 (Baudet et al., 2012). Our observation of the regulation of Robo1 surface expression at the post-translational level by RabGDI suggested trafficking as yet another mechanism to switch axonal responsiveness at choice points (Philipp et al., 2012).

RabGDI, which is encoded by a gene linked to human mental retardation (D'Adamo et al., 1998), is an essential component of the vesicle fusion machinery (Seabra et al., 2002; Pfeffer and Aivazian, 2004). RabGDI associates with Rab11-positive vesicles (Philipp et al., 2012). Likewise, an association of Rab11-positive vesicles with calsyntenin 1 was found in kinesin 1-dependent axonal transport (Konecna et al., 2006; Steuble et al., 2010). Calsyntenin 1 is a member of a family of three transmembrane proteins (Vogt et al., 2001; Hintsch et al., 2002) and acts as a linker between vesicular cargo and kinesin 1, the motor for anterograde axonal transport (Konecna et al., 2006). Two binding sites in the cytoplasmic domain of calsyntenin 1 were found to interact with the tetratricopeptides of kinesin light chain 1 (KLC1). Mutations in these domains of calsyntenin 1 result in reduced fast anterograde axonal transport (Konecna et al., 2006). In zebrafish, calsyntenin 1-mediated vesicle trafficking is required for branching of peripheral but not central axons of sensory neurons (Ponomareva et al., 2014).

We found calsyntenins to be expressed in a dynamic, partially overlapping pattern in the developing chicken spinal cord. In particular, we found calsyntenin 1 to be expressed in commissural neurons at the time when their axons reach the floor plate and cross the midline. Because calsyntenin 1 was found associated with Rab11-positive vesicles (Steuble et al., 2010, 2012) and because Robo1-positive vesicles requiring RabGDI for membrane insertion of their cargo were also Rab11 positive (Philipp et al., 2012), we analyzed these associations further and found a partial overlap of calsyntenin 1 with Robo1-, Rab11- and RabGDI-positive vesicles both in COS7 cells and in growth cones of commissural axons.

Downregulation of calsyntenin 1 resulted in similar axon guidance defects, as observed after silencing Robo1 or RabGDI. In contrast to Robo1 and RabGDI, loss of calsyntenin 1 had an additional effect on post-crossing commissural axons. Like Shh (Bourikas et al., 2005; Wilson and Stoeckli, 2013), Wnts (Domanitskaya et al., 2010; Avilés and Stoeckli, 2016), SynCAMs (Niederkofler et al., 2010) and semaphorin 6B (Andermatt et al., 2014), calsyntenin 1 was required for the rostral turning of post-crossing commissural axons. Our detailed in vivo analyses indicate a regulatory role of calsyntenin 1-mediated trafficking in midline crossing by controlling Robo1 expression and in longitudinal axon guidance by controlling frizzled 3 expression.

We first detected calsyntenin 1 mRNA in dI1 commissural neurons at HH21 (not shown), and more clearly at HH22, when the first axons have reached the ipsilateral floor plate border in the lumbar spinal cord (Fig. S1). We used in ovo RNAi to investigate a potential function of the calsyntenins in commissural axon guidance. To this end, we analyzed the trajectory of dI1 axons after downregulation of calsyntenin 1 in open-book preparations of spinal cords collected from embryos at HH25 (Fig. 1A). In untreated (Fig. 1B) and in control-injected (Fig. 1C) embryos dI1 axons had crossed the midline and turned rostrally along the contralateral floor plate border. After downregulation of calsyntenin 1 in dI1 neurons with a microRNA (miR)-based construct targeting calsyntenin 1 driven by the Math1 enhancer (miCst1, Fig. 1D) or by electroporation of double-stranded (ds) RNA (dsCst1, Fig. 1E), axon guidance was severely perturbed (Fig. 1F). Axons stalled in the floor plate and failed to turn into the longitudinal axis. On average, only 22.5±5.2% of the injection sites showed normal axonal navigation after injection and electroporation of miCst1. Similarly, after targeting calsyntenin 1 with dsCst1, only 27.7±2.8% of the injection sites showed normal axon guidance. No, or only minor, effects were observed when either calsyntenin 2 or 3 was silenced, as 72.0±4.4% (miCst2) or 56.2±9.3% (dsCst2) and 46.7±4.0% (miCst3) or 63.0±8.4% (dsCst3) of the injection sites were normal (not shown; Table 1). Calsyntenin 2, which is not expressed in dI1 neurons, served as a negative control, demonstrating the specificity of our approach. Although calsyntenin 3 is expressed in dI1 neurons, its downregulation did not affect commissural axon guidance (Table 1). The efficiency of downregulation was demonstrated by in situ hybridization (Fig. S1E,F, Table S1).

Calsyntenin 1 had a direct effect on axonal navigation, as its downregulation did not affect spinal cord patterning. The expression of cell type-specific marker proteins did not differ between experimental and untreated control embryos (not shown). Furthermore, the phenotype was not due to a delay in axon outgrowth or a slower growth rate, as axons reached the floor plate at the appropriate stage (Fig. S1G,H) and axons in embryos sacrificed 1 day later still failed to turn into the longitudinal axis (Fig. S1I-L). Similarly, neurite length did not differ between control-treated sensory neurons and neurons lacking calsyntenins grown on laminin (Fig. 1G-K).

Taken together, our in vitro and in vivo experiments demonstrate a role for calsyntenin 1 in commissural axon guidance during floor plate crossing and in the subsequent turning into the longitudinal axis.

The aberrant axon guidance phenotype observed in the absence of calsyntenin 1 strongly resembled the phenotype observed in the absence of RabGDI (Philipp et al., 2012). Because calsyntenin 1 was shown in biochemical studies to be associated with Rab11-positive vesicles (Steuble et al., 2010, 2012) and because our previous in vivo studies indicated that Robo1-containing vesicles regulated by RabGDI were also Rab11 positive (Philipp et al., 2012), we tested for a role of calsyntenin 1 in the regulation of Robo1 surface expression.

We verified partial colocalizations of Robo1, Rab11, RabGDI and calsyntenin 1 in commissural neurons (Fig. 2). Because antibodies for the detection of endogenous proteins are not available, we expressed tagged proteins in commissural neurons in vivo at HH17. After 2 days, we sacrificed the embryos and dissected commissural neuron explants to visualize individual growth cones. We did not include Robo2 in these analyses because the lack of Robo2 was shown to result primarily in ipsilateral turns and, thus, a phenotype different from that obtained after downregulation of calsyntenin 1, Robo1 or RabGDI (Philipp et al., 2012). We also excluded calsyntenin 2 and 3, as their downregulation did not cause any significant changes in commissural axon guidance (Table 1).

We found partial overlap between calsyntenin 1 and Robo1 (Fig. 2A), RabGDI (Fig. 2D) and Rab11 (Fig. 2F). We confirmed our previous observations of a colocalization of Robo1 and Rab11 (Fig. 2B) and RabGDI and Robo1 (Fig. 2C), as well as RabGDI and Rab11 (Fig. 2E) (Philipp et al., 2012). The Pearson's colocalization coefficients of the double-stained axons are given in Fig. 2H. The values from both positive and negative controls (Fig. 2G,H) were significantly different from all other conditions (ANOVA, P<0.05). As a positive control, we used an HA-Robo1-myc construct that was stained for both tags (HA in green, myc in red). As a negative control, we stained growth cones for tubulin (green) and neurofilament (red). We also used triple staining to confirm partial colocalization of calsyntenin 1, Robo1, RabGDI and Rab11 (Fig. S2).

Taken together, our colocalization studies suggest the existence of calsyntenin 1-positive vesicles containing Robo1 as cargo. In agreement with previous studies, Robo1-containing vesicles are associated with RabGDI and Rab11, suggesting that at least a subpopulation is also positive for calsyntenin 1.

To obtain functional evidence for a potential cooperation between calsyntenin 1 and RabGDI in the regulation of Robo1 surface expression, we electroporated miR-based constructs at subthreshold or ‘hypomorphic’ doses. If calsyntenin 1 and RabGDI act in the same pathway, then the use of low doses should reproduce the phenotypes observed after efficient downregulation of either RabGDI or Robo1 alone. For a direct comparison, we repeated the silencing of RabGDI and Robo1 obtained previously with dsRNA (Philipp et al., 2012) but using miR constructs (Fig. S3).

As expected, downregulation of Robo1 (Fig. S3A) or RabGDI (Fig. S3B) with 700 ng/µl Math1-driven miR constructs resulted in axonal stalling in the floor plate, as seen previously when we electroporated dsRNA derived from Robo1 or RabGDI (Table 1) (Philipp et al., 2012). When we used only low doses of miRs (300 or 350 ng/µl), downregulation of Robo1 (Fig. S3C), RabGDI (Fig. S3D) or calsyntenin 1 (Fig. S3E) was less effective and did not result in axonal pathfinding errors (Table 1, Fig. S3F).

However, co-injection of low concentrations of all these miRs together did result in the expected phenotype, i.e. axons stalled in the floor plate and those that did reach the contralateral floor plate border failed to turn rostrally along the longitudinal axis (Fig. 3A,E, Table 1). Similarly, reducing both calsyntenin 1 and RabGDI together by electroporation of low doses was sufficient to induce floor plate stalling (Fig. 3B,E). These results indicate that calsyntenin 1 and RabGDI cooperate in commissural axon guidance.

The combination of low doses of calsyntenin 1 and Robo1 was less efficient in inducing floor plate stalling (at 30.7±7.0% of the injection sites per embryo) but still interfered markedly with correct axon guidance at the floor plate exit site in comparison with control-treated embryos (Fig. 3C,E). At HH18, Robo1 mRNA is already detected in dI1 neurons (Philipp et al., 2012). Thus, we reasoned that the weaker effect on axonal midline crossing might be due to the presence of Robo1 protein in vesicles already at the time of electroporation at HH18. Thus, we repeated the electroporation of miCst1 together with miRobo1 at HH15. In agreement with our hypothesis, we now found floor plate stalling at 55.6±3.3% of the injection sites (Table 1).

In summary, the in ovo perturbation experiments together with our previous findings on the role of RabGDI in Robo1 trafficking (Philipp et al., 2012) suggest a cooperation of calsyntenin 1 and RabGDI in the regulation of Robo1 trafficking to the growth cone surface.

To demonstrate that loss of RabGDI and calsyntenin 1 function indeed affected axon guidance by preventing Robo1 surface expression we assessed the surface levels of double-tagged Robo1 under different conditions (Fig. 4). After downregulation of both calsyntenin 1 and RabGDI, we found significantly less Robo1 on the surface of dI1 growth cones (Fig. 4A,E). As shown previously (Philipp et al., 2012), downregulation of RabGDI alone also interfered with Robo1 surface expression (Fig. 4B,E). The same was true for downregulation of calsyntenin 1 (Fig. 4C,E).

Further evidence for the specificity and the underlying mechanism of Robo1 trafficking was provided by rescue experiments (Fig. 5). The effect on midline crossing induced by a miR targeting chicken RabGDI was rescued by co-electroporation of a plasmid encoding human RABGDI that was not targeted by miRabGDI (Fig. 5A). The effect of loss of calsyntenin 1, induced by dsRNA derived from the 3′ UTR, was reversed by co-electroporation of a plasmid encoding EGFP-tagged calsyntenin 1 that was not targeted (Fig. 5B). However, the effect of calsyntenin 1 silencing could not be rescued by a mutated version of calsyntenin 1 that lacked the binding sites to kinesin 1 (Konecna et al., 2006) and, therefore, was unable to mediate linkage of the vesicle to the transport system (Fig. 5C). Taken together, these results indicate that Robo1 delivery and surface expression on commissural growth cones depend on both RabGDI and calsyntenin 1. Furthermore, the function of calsyntenin 1 depends on its role in vesicle trafficking.

Comparison of the phenotypes seen after silencing calsyntenin 1 or RabGDI revealed qualitative differences. In the absence of RabGDI most axons failed to reach the contralateral floor plate border. Therefore, the behavior of post-crossing axons was difficult to assess. Silencing calsyntenin 1 also affected midline crossing, but more axons reached the contralateral floor plate border. However, most post-crossing axons failed to turn into the longitudinal axis. Thus, we analyzed the behavior of post-crossing axons in more detail.

In previous studies, we characterized the role of morphogen signaling in post-crossing commissural axon guidance. Both Shh (Bourikas et al., 2005; Wilson and Stoeckli, 2013) and Wnt signaling (Domanitskaya et al., 2010; Avilés and Stoeckli, 2016; see also Lyuksyutova et al., 2003) were shown to be required for the rostral turn of post-crossing commissural axons along the longitudinal axis of the spinal cord. Because pre-crossing axons are not sensitive to the Shh and Wnt gradients, the surface expression of guidance receptors needs to be temporally regulated in a precisely controlled manner. The expression of Hhip, the receptor mediating the repulsive response to Shh, was shown to be regulated at the transcriptional level (Bourikas et al., 2005; Wilson and Stoeckli, 2013). A colocalization of Hhip and calsyntenin 1 was therefore not expected. Indeed, that is what we found, when Hhip and calsyntenin 1 were expressed in COS7 cells (Fig. S4).

By contrast, immunoreactivities of calsyntenin 1 and frizzled 3 (Fzd3), the receptor for Wnts, strongly overlapped in COS7 cells (Fig. 6A) and in commissural axons and growth cones (Fig. 6B). Fzd3 mRNA was found in dI1 commissural neurons already at HH18 (Fig. 6C). Therefore, regulation of Fzd3 at a post-transcriptional level was required to explain why post-crossing, but not pre-crossing, axons were responsive to Wnt. In agreement with the temporal expression pattern, downregulation of Fzd3 with dsRNA at HH18/19 did not effectively prevent the rostral turn of post-crossing commissural axons, as normal axon pathfinding was seen at 88.7% of all injection sites. By contrast, after silencing Fzd3 at HH14/15, we found robust interference with the rostral turning of post-crossing commissural axons. Normal pathfinding was only seen at 21.8% of the injection sites. These values were similar in embryos electroporated with a miR targeting Fzd3 (Fig. 6E,H). We found axons that failed to turn rostrally along the contralateral floor plate border at 69.3±6.2% (miFzd3) of all injection sites. These results are in agreement with previous studies in the mouse (Lyuksyutova et al., 2003).

To assess whether calsyntenin 1 is required for the regulation of Fzd3 expression, we again used in ovo RNAi with hypomorphic concentrations of miRs (Fig. 6F,H). We decreased the amount of miR until the injection of either the miR targeting Fzd3 (67.0±8.5% of the injection sites with normal axon guidance) or that targeting calsyntenin 1 alone (64.9±5.2% of the injection sites with normal axon guidance) did not significantly interfere with post-crossing commissural axon guidance. However, when we co-injected the low levels of miRs against calsyntenin 1 and Fzd3 most axons reached the contralateral border of the floor plate but failed to turn rostrally at 70.6±5.5% of the injection sites (Fig. 6G,H, Table 1).

Taken together, these results indicate a cooperation of calsyntenin 1 and Fzd3 in post-crossing commissural axon guidance.

Based on the temporal expression of Fzd3 and the role of calsyntenin 1 in vesicle trafficking, we expected calsyntenin 1 to regulate Fzd3 surface expression. Using a similar approach as that detailed above, we analyzed Fzd3 trafficking to the growth cone surface (Fig. 7). Downregulation of calsyntenin 1 together with RabGDI did reduce Fzd3 on the surface but only to the same extent that was achieved with loss of calsyntenin 1 alone. Silencing RabGDI alone did not change the surface expression of Fzd3 compared with the control (compare Fig. 7B with D, quantified in E).

The absence of an effect of RabGDI on Fzd3 surface expression was corroborated by functional data. When we electroporated low amounts of miR constructs targeting Fzd3 and RabGDI, we did not observe any increase in axon guidance errors compared with control embryos (Fig. 7F,H; 65.9±1.5%). Thus, we concluded that RabGDI is not required for Fzd3 trafficking to the growth cone surface.

Our findings were further supported by the detailed analysis of Robo1 and Fzd3 expression. Although both Robo1-positive and Fzd3-positive vesicles were mostly immunopositive for calsyntenin 1, Robo1 and Fzd3, immunoreactivity barely overlapped (Fig. 8). These findings support our in vivo data implicating calsyntenin 1 in both Robo1 and Fzd3 trafficking. However, in contrast to Robo1, Fzd3 trafficking is independent of RabGDI. Most likely, calsyntenin 1 is also required for the trafficking of additional vesicles containing as yet unknown cargoes, as many vesicles that were calsyntenin 1-positive were not associated with either Fzd3 or Robo1.

In summary, our in vivo studies demonstrate a crucial role for calsyntenin 1 in the specific trafficking of guidance receptors to the growth cone surface at a choice point. The storage of guidance receptors in Rab11-positive vesicles in growth cones allows for their rapid insertion into the membrane. In turn, the surface expression of novel guidance receptors changes the responsiveness of the growth cone to the intermediate target and, thus, forces the axon to continue with its navigation toward the next intermediate or its final target. Furthermore, the rapid insertion of guidance receptors at a choice point, in this case the floor plate, rather than their accumulation on the surface over time explains why axons acquire responsiveness to the guidance cues for the longitudinal axis only after midline crossing.

Midline crossing by dI1 commissural axons represents an easily accessible model with which to study the molecular mechanisms underlying the required switch in axonal responsiveness at a choice point. Pre-crossing axons enter the floor plate due to the predominance of positive signals derived from the interaction of the growth cone with guidance cues expressed by floor plate cells. However, in order to leave the floor plate and move on towards the final target, axons need to overcome this attraction. Therefore, growth cones need to change their surface receptors to recognize previously undetectable repulsive cues associated with the intermediate target. At the floor plate, these negative cues have been identified as Slits (Brose et al., 1999; Yuan et al., 1999) and class 3 semaphorins (Nawabi et al., 2010; Zou et al., 2000). Furthermore, axons need to ignore the guidance cues directing them along the longitudinal axis on the ipsilateral floor plate border but readily detect them upon floor plate exit. Both floor plate crossing and turning into the longitudinal axis therefore depend on the precisely regulated expression of guidance receptors on the growth cone surface.

Responsiveness to semaphorin 3B and Slit was shown to depend on the stabilization of the surface receptor component plexin A1 (Nawabi et al., 2010; Charoy et al., 2012; Delloye-Bourgeois et al., 2015) and to involve Shh-dependent sensitization (Parra and Zou, 2010). The sensitivity to Slit1 was triggered by the RabGDI-dependent trafficking of Robo1 to the growth cone surface (Philipp et al., 2012). In mouse, a switch in Robo3 isoform expression was suggested to contribute to the regulation of Robo1-mediated sensitivity to Slit of pre-crossing versus post-crossing commissural axons (Chen et al., 2008). However, these findings have been questioned recently (Zelina et al., 2014).

Similarly, the regulation of axonal responsiveness to guidance cues for the longitudinal axis depends on the precisely controlled surface expression of guidance receptors. An example of regulation at the transcriptional level is provided by our findings that Shh induces the expression of its own receptor Hhip for the guidance of post-crossing axons in a glypican 1-dependent manner (Wilson and Stoeckli, 2013; Bourikas et al., 2005). By contrast, responsiveness of post-crossing axons to Wnts is regulated by trafficking of Fzd3 to the growth cone surface (this study). This trafficking must not be confused with the endocytosis of Fzd3 shown to be part of Wnt signaling in axon guidance (Onishi et al., 2013; Shafer et al., 2011). In response to Wnt binding, Fzd3 was shown to be internalized in an Arf6-dependent manner for signal transduction. In our study, we describe the molecular basis of trafficking for the initial delivery of Fzd3 to the growth cone surface, explaining why growth cones do not respond to the Wnt gradient along the ipsilateral floor plate border.

Both our previous in vivo studies characterizing the role of RabGDI in Robo1 regulation (Philipp et al., 2012) and the studies characterizing the role of calsyntenin 1 in Robo1 and Fzd3 regulation reported here, indicate that RabGDI and calsyntenin 1 are not required for axonal growth but specifically for axon guidance.

Loss-of-function and rescue experiments indicate that the role of calsyntenin 1 in commissural axon guidance depends on its binding to kinesin, the motor for anterograde axonal transport of vesicles. An effect of calsyntenin 1 on the selective transport of vesicle subpopulations containing guidance receptors as cargo is in agreement with our findings that calsyntenin 1-positive puncta colocalized with puncta positive for either Robo1 or Fzd3, both axon guidance receptors whose surface expression needs to be precisely regulated at the floor plate to prevent premature reactivity to the respective ligands Slits and Wnts. Furthermore, the expression of neither Robo1 nor Fzd3 is regulated at the transcriptional level.

The regulation of vesicle trafficking and thus the specificity of cargo delivery to the growth cone surface in a temporally precisely regulated manner represents a potent molecular mechanism underlying the switch in axonal responsiveness at an intermediate target. Our results indicate that the regulatory mechanisms for the trafficking of Robo1 and Fzd3 differ. Robo1 surface expression depends on both RabGDI and calsyntenin 1 (Fig. 4) (Philipp et al., 2012). By contrast, RabGDI was not required for the regulation of Fzd3 surface expression on post-crossing versus pre-crossing axons (Fig. 7).

Thus, the most parsimonious explanation of these results is the regulation of trafficking of vesicles with specific guidance receptors as cargo. One type of vesicle contains Robo1 receptors. This type of vesicle requires both RabGDI and calsyntenin 1 for fusion with the plasma membrane and insertion of Robo1 into the growth cone membrane. Another type of vesicle that contains Fzd3 as cargo is independent of RabGDI but requires calsyntenin 1 for trafficking to the growth cone membrane. These results are in perfect agreement with a recent study in zebrafish, where loss of calsyntenin 1 was found to affect branching of peripheral but not central sensory axons (Ponomareva et al., 2014). These findings were linked to the observation that calsyntenin 1 was associated with a specific subpopulation of vesicles. The same concept – subpopulations of vesicles with specific cargo – is also found in our study demonstrating a regulatory role of vesicular trafficking of guidance receptors to the growth cone surface to orchestrate the switch in axonal behavior at intermediate targets or choice points.

Fertilized chicken eggs obtained from a local hatchery were windowed after incubation at 38.5°C for 2 or 3 days. At the appropriate Hamburger and Hamilton (HH) stage of development (Hamburger and Hamilton, 1951), we removed the extra-embryonic membranes to inject plasmids or dsRNA into the central canal of the spinal cord (Wilson and Stoeckli, 2012). All experiments were in accordance with the regulations of the Cantonal Veterinary Office Zurich. We used miR-based constructs encoding GFP as a transfection marker followed by the miR sequence at 250 ng/µl for β-actin-driven constructs or 700 ng/µl for Math1-driven constructs (Wilson and Stoeckli, 2011) (for miR constructs see Table S2). Alternatively, we used a combination of dsRNA (300 ng/μl) and a plasmid encoding GFP (25 ng/μl) in PBS with 0.02% Fast Green (AppliChem). For electroporation, we used the same settings as described previously (Pekarik et al., 2003). Embryos were sacrificed at the desired HH stage and their spinal cords were analyzed as open-book preparations as described (Perrin and Stoeckli, 2000). Injections and electroporations of dsRNA or miR were carried out at HH13-15 (E2) or at HH17/18 (E3). Analyses of commissural axon pathfinding were performed at HH25 (E5), except for the experiment reported in Fig. S1, where the analysis was carried out 1 day later (E6/HH28).

The polymerase II-driven miR constructs were cloned as described previously (Wilson and Stoeckli, 2011). We inserted the miR hairpin-loop structure using NheI and MluI (both NEB) to create either β-actin-driven or Math1-driven plasmids (Table S2). We used bp 4283-4995 of the 3′ UTR of chicken calsyntenin 1 (NM_001197050.1) to produce dsRNA (dsCst1) according to our published protocol (Bourikas et al., 2005). For dsCst2 we used bp 498-1436 of chicken calsyntenin 2 (XM_422633.4) and for dsCst3 bp 2175-3169 of chicken calsyntenin 3 (XM_416520.4).

To mimic double-heterozygous approaches, which are often used in genetic analyses, we lowered the injected amount of each construct to levels that did not effectively interfere with axon guidance. The co-injection of two miR constructs targeting different genes was expected to interfere with axonal navigation if the two target genes work in the same pathway. The feasibility of this approach has been demonstrated previously in our analyses of the role of Shh in post-crossing commissural axon guidance (Wilson and Stoeckli, 2013). To determine the hypomorphic amounts of the injected miR constructs, we typically used half of the concentration (350 ng/µl) that was used for effective gene silencing. An exception was Math1-miRobo1, where we had to lower the concentration to 300 ng/μl.

Open-book preparations were imaged using an Olympus BX61 spinning disc microscope. A person blind to the experimental condition scored 10-15 open-book preparations with 11±3 injection sites per treatment group. Only injection sites with GFP expression were included. We distinguished between three phenotypes: normal floor plate crossing and turning rostrally along the contralateral floor plate border; floor plate stalling; and no axonal turns at the contralateral floor plate border. Because it is impossible to count individual axons, we scored an injection site as showing floor plate stalling only when at least 50% of the DiI-labeled axons failed to reach the contralateral floor plate border. Similarly, for the ‘no turning’ phenotype at least 50% of the axons at the floor plate exit site had to fail to turn rostrally. Obviously, these two phenotypes are not completely independent of each other, as stalling of all axons in the floor plate would prevent the analysis of their turning behavior at the floor plate exit site. Therefore, we only compared the percentages of injection sites per embryo exhibiting normal axon guidance in our quantitative analyses. Statistical analysis of the data was performed with SPSS (IBM). Normal distribution of the values was verified with the Shapiro-Wilk test (P≥0.01). P-values were calculated with one-way ANOVA and Tukey's post-hoc tests.

Explants of commissural neurons were dissected from untreated, control-treated or experimental embryos at HH25. Neurons were plated in 8-well Lab-Tek dishes coated with polylysine (20 µg/ml) and laminin (10 µg/ml). Dorsal root ganglion neurons at a density of 10,000 cells/cm² and commissural neurons were cultured for 24-36 h as described previously (Stoeckli et al., 1996, 1997).

To determine surface expression levels, cells were stained with primary antibody against the N-terminal tag before fixation and permeabilization. In all other experiments, cells were fixed and then permeabilized for immunohistochemistry using goat anti-Flag (Abcam), rabbit anti-HA (Rockland) and mouse anti-myc (9E10; Developmental Studies Hybridoma Bank), visualized by goat anti-mouse Cy3 (Jackson ImmunoResearch), donkey anti-rabbit Cy3 (Jackson ImmunoResearch), goat anti-mouse Alexa 488 (Molecular Probes), donkey anti-goat Alexa 488 (Invitrogen) or donkey anti-rabbit Alexa 488 (Jackson ImmunoResearch). For triple staining we also used goat anti-mouse Alexa 350 (Molecular Probes) and goat anti-rabbit Alexa 350 (Invitrogen). Details of antibodies are provided in Table S4. Immunolabeled cells were imaged with a Leica SP2 confocal microscope using a 63× oil-immersion objective (NA=1.4) and an Olympus BX61 spinning disc microscope with a 60× oil-immersion objective (NA=1.42). Image stacks (embracing the Nyquist criterion) were processed using Huygens 3D deconvolution and analysis software. At least 20 cells per condition were included in the quantitative analysis. Neurite length measurements and image analyses were performed with Imaris (Bitplane). Values were analyzed for normal distribution with the Shapiro-Wilk test and P-values were calculated with one-way ANOVA and Tamhane's post-hoc tests.

COS7 cells grown in DMEM (Invitrogen) supplemented with 2% fetal calf serum (FCS) were regularly passaged to avoid confluence. For transfection, cells were incubated in 150 μl DMEM with 10% FCS and 50 μl transfection solution containing 250-400 ng DNA (for single transfection) and 1.25% Lipofectamine (Invitrogen) in Opti-MEM (Gibco). For a list of transfected plasmids see Table S3.

In situ hybridization was performed as previously described (Mauti et al., 2006). Spinal cord patterning was assessed as previously described (Avilés and Stoeckli, 2016). Commissural axons were stained with a rabbit anti-axonin 1 antibody (Table S4).

We thank Dr Peter Sonderegger (Department of Biochemistry, University of Zurich) for the calsyntenin plasmids; Dr Yimin Zou (Division of Biological Sciences, UC San Diego) for the Fzd3 plasmid; Urs Ziegler (Center for Microscopy and Image Analysis of the University of Zurich) for help with image analysis; Dr Beat Kunz and Tiziana Flego for excellent technical assistance; and members of the E.T.S. lab for helpful comments and discussions.