Commissural neurons transgress the CNS/PNS boundary in absence of ventricular zone-derived netrin 1

Netrin 1 is a secreted protein that controls cell-cell interactions in many organs and species, during development and in pathological conditions (Mehlen et al., 2011). In the central nervous system (CNS), netrin 1 promotes axon outgrowth to the midline, axon attachment to their targets and neuronal migration (Akin and Zipursky, 2016; Serafini et al., 1994, 1996; Yee et al., 1999). Netrin 1 is secreted, but acts locally by promoting cell adhesion and haptotaxis (Akin and Zipursky, 2016; Li et al., 2004; Moore et al., 2009). In the mouse hindbrain and spinal cord, netrin 1 is not only produced by the floor plate, but is also released at the pial surface by neural precursors of the ventricular zone (Dominici et al., 2017; Kennedy et al., 1994; Varadarajan et al., 2017). This suggests that netrin 1 accumulation in the basal lamina provides a permissive substrate for axon extension. In the spinal cord, netrin 1 and its receptor deleted in colorectal cancer (Dcc) influence the confinement of commissural axons to the CNS (Laumonnerie et al., 2014). In the spinal cord, two repulsive guidance cues, netrin 5 and Sema6A, also act as gate keepers at the CNS/PNS border (Bron et al., 2007; Garrett et al., 2016; Mauti et al., 2007). Both cues are expressed by so-called boundary cap (BC) cells, which constrain motor neurons and oligodendrocyte soma to the spinal cord and prevent them from migrating along motor nerves into the PNS (Kucenas et al., 2009; Vermeren et al., 2003). Whether such mechanisms are at play at the level of the hindbrain is unknown.

Interestingly, several classes of hindbrain neurons preferentially migrate tangentially under the pial surface in a netrin 1-rich domain (Stanco and Anton, 2013). This occurs in the case of the pontine nucleus, one of the four hindbrain precerebellar nuclei that contain neurons projecting to the cerebellum. Pontine neurons are born in the rhombic lip, a dorsal neuroepithelium that lines the fourth ventricle (Wullimann, 2011). Pontine neurons form a compact and superficial migratory stream that first progresses anteriorly before turning ventrally towards the floor plate (Geisen et al., 2008; Kratochwil et al., 2017; Zelina et al., 2014). Pontine neurons fail to migrate ventrally in Ntn1 hypomorphic mutants and their number is reduced (Yee et al., 1999; Zelina et al., 2014). Here, we show that netrin 1, acting at least in part through the Dcc receptor, prevents pontine neurons and other classes of hindbrain commissural neurons from exiting the CNS through sensory nerve roots.

A few commissural axons project outside the spinal cord in netrin 1 hypomorph embryos (Ntn1^ßgeo/ßgeo) (Laumonnerie et al., 2014). We first confirmed this observation using immunostaining for Robo3, a marker of commissural neurons (Friocourt and Chédotal, 2017; Sabatier et al., 2004) on E11 embryos (Fig. 1A-L). To facilitate the analysis, whole-mount immunostaining was performed followed by 3DISCO clearing and three-dimensional (3D) imaging with light-sheet fluorescence microscopy (LSFM) (Belle et al., 2014). In Ntn1^ßgeo/+ embryos, Robo3⁺ axons were restricted to the spinal cord, extending dorso-ventrally and crossing the midline (Fig. 1A,D, Fig. S1A,B, Movie 1), whereas in Ntn1^ßgeo/ßgeo embryos, Robo3⁺ axons were also seen outside the spinal cord, within dorsal root ganglia (DRG) labeled with anti-islet 1 (Fig. 1B,E and Fig. S1C,D). A few Robo3⁺ axons were also seen in the ventral roots (data not shown). Next, we studied a null allele of Ntn1 (see Materials and Methods), Ntn1^−/−. Robo3⁺ axons also left the spinal cord in Ntn1^−/− embryos (Fig. 1C,F, Fig. S1E,F and Movie 2). Importantly, unlike in control embryos (Fig. 1G,H), Robo3⁺ axons were detected outside the CNS in the hindbrain, both in Ntn1^ßgeo/ßgeo embryos (Fig. 1I,J) and Ntn1^−/− embryos (Fig. 1K,L). As in the spinal cord, commissural axons escaped the CNS via sensory roots and this was particularly striking at the trigeminal and vestibular nerve roots (Fig. 1K,L and Fig. S1G-J). The amount of Robo3⁺ axons invading the PNS at the hindbrain level was significantly lower in Ntn1^ßgeo/ßgeo compared with the Ntn1^−/− embryos (Fig. S1S and Table S1).

In the hindbrain, commissural neurons are produced at least until E16 (Pierce, 1966; Zelina et al., 2014). Therefore, we next studied Ntn1 mutant embryos at E13 and E16. In control embryos, Robo3 axons were only found in the CNS (Fig. 1M,N and Movie 3), whereas in Ntn1^ßgeo/ßgeo embryos and Ntn1^−/− embryos, Robo3⁺ axons massively invaded trigeminal and vestibular nerves and ganglia (Fig. 1O-R and Movie 4). This defect was more pronounced in Ntn1^−/− than in Ntn1^ßgeo/ßgeo mutants (Fig. S1T and Table S1). At E16, only pontine neurons still express Robo3 in the hindbrain of control embryos (Fig. 2A-C and Movie 5) (Marillat et al., 2004; Zelina et al., 2014), suggesting that some of the Robo3⁺ axons leaving the brain in Ntn1^ßgeo/ßgeo (Fig. 2D-F) and Ntn1^−/− (Fig. 2G-I) E16 embryos could belong to pontine neurons. This hypothesis was tested using immunostaining for Barhl1 and Pax6, two markers of migrating pontine neurons (Benzing et al., 2011; Zelina et al., 2014). In controls, the auditory and trigeminal nerves and ganglia did not contain Barhl1⁺/Robo3⁺ or Pax6⁺ neurons (Fig. 2A-C,J-N and Fig. S1K,L,O,P). By contrast, streams of Barhl1⁺/Robo3⁺ or Pax6⁺/Robo3⁺ cells were seen inside these nerves in Ntn1^ßgeo/ßgeo (Fig. 2O-S and Fig. S1M,Q) and Ntn1^−/− embryos (Fig. 2T-X, Fig. S1N,R and Movie 6), that could be traced back to the pontine migratory stream in the hindbrain (Fig. 2D-I). In wild-type embryos, Barhl1⁺/Robo3⁺ neurons are absent from trigeminal and vestibular ganglia, which contains Sox10⁺ sensory neurons (Fig. 2J-N). By contrast, there was a significant colonization of trigeminal and vestibular nerves and ganglia by streams and clusters of Barhl1⁺ neurons in Ntn1^ßgeo/ßgeo embryos and Ntn1^−/− embryos (Fig. 2O-X and Fig. S1U and Table S2). Ectopic Barhl1⁺ neurons were not immunoreactive for Sox10 (Fig. 2P,Q,U,V) but expressed Robo3 (Fig. 2R,S,W,X), supporting their pontine neuron identity.

To confirm that these neurons originated from the CNS, we electroporated a plasmid encoding green fluorescent protein (GFP) into the rhombic lip of Ntn1^−/− E13.5 embryos (n=4) and collected them at E16. This selectively drives GFP expression in migrating pontine neurons (Kawauchi, 2006; Zelina et al., 2014). In all controls (n=10), GFP⁺ neurons were restricted to the hindbrain (Fig. 2Y), whereas in all Ntn1^−/− embryos (Fig. 2Z,Z′ and Movie 7), many GFP⁺/Robo3⁺ processes and cell bodies were found within the auditory and trigeminal nerves. Together, these data show that neurons maintaining a pontine identity transgress the PNS/CNS boundary in absence of netrin 1 and migrate along nerve roots. Their long-term fate could not be assessed as both types of Ntn1 mutants die at birth.

Netrin 1 has either growth-promoting or growth-inhibiting activity and could act as a repulsive barrier at sensory nerve roots. However, Ntn1 mRNA was absent from DRG and hindbrain sensory ganglia (Fig. S2A,C) and was only found in the floor plate, ventricular zone progenitors and cochlea, as previously described (Abraira et al., 2008; Dominici et al., 2017; Laumonnerie et al., 2014; Serafini et al., 1994). These results were confirmed by monitoring β-galactosidase and netrin 1 protein expression in Ntn1^ßgeo/+ E13 embryos (Fig. S2B,D). β-Gal was present in floor plate and ventricular zone precursors, and netrin 1 accumulated at the pial surface, as recently described (Dominici et al., 2017; Varadarajan et al., 2017). It was also detected in the inner ear (Nishitani et al., 2017) and in some mesenchymal cells but not in sensory ganglia. Netrin 1 levels were high in nestin⁺ radial glia endfeet but stopped dorsally at the level of the vestibular and trigeminal nerve roots (Fig. S2E,F). Netrin 1 staining was abrogated in Ntn1^−/− embryo but nestin⁺ glial endfeet were organized normally as previously described (Fig. S2G,H) (Dominici et al., 2017). Importantly, netrin 1 and β-gal were absent from commissural neurons, including migrating pontine neurons. The absence of netrin 1 in sensory ganglia and nerves suggests that it is unlikely to act as a repulsive barrier.

To further characterize netrin 1 function at the CNS/PNS boundary, we next performed selective genetic ablation of netrin 1 from various cellular sources using specific Cre-recombinase driver lines and a Ntn1 conditional allele (Ntn1^fl/fl) (Dominici et al., 2017). Cre expression was confirmed using a Rosa^tdTomato reporter line (Fig. S2I-L). As in Ntn1^fl/+ (Fig. 3A-C), no Robo3⁺ axons were detected in the PNS of Shh:Cre;Ntn1^fl/fl E13 embryos (Fig. 3D-F), which completely lack netrin 1 at the floor plate (Dominici et al., 2017) (Fig. S2I). Next, we studied Nes:Cre;Ntn1^fl/fl E13 embryos in which netrin 1 is deleted from neural cells in the CNS and PNS but maintained in floor plate and inner ear (Fig. S2K) (Dominici et al., 2017). Interestingly, a massive invasion of peripheral nerve roots by commissural axons was seen in Nes:Cre;Ntn1^fl/fl E13 embryos, with Robo3⁺ axons extending far in the trigeminal nerve branches and in the vestibular nerve (Fig. 3G-J). The phenotype was as severe as in Ntn1^−/− and Ntn1^ßgeo/ßgeo mutants (Fig. S1T and Table S1). At E16, streams of Barhl1⁺/Robo3⁺ and Pax6⁺ pontine neurons were also detected within the trigeminal and auditory nerves (Fig. 3K-N, Fig. S1U and Table S2). Interestingly, a subset of Foxp2⁺ cells, most likely corresponding to inferior olivary neurons (Dominici et al., 2017), also escaped the CNS in Nes:Cre;Ntn1^fl/fl and Ntn1^−/− E13 embryos to enter the vagus nerve (Fig. 3O-R). Together, these data show that a massive exit of hindbrain commissural neurons from the CNS is caused by the absence of netrin 1 from CNS cells.

Boundary cap (BC) cells might prevent commissural neuron from escaping the CNS as they do for motor neurons. To visualize BC cells in control and Ntn1 mutant embryos, we performed in situ hybridization for Prss56, which encodes a potentially secreted trypsin-like serine protease highly expressed by BC cells (Coulpier et al., 2009). Prss56⁺ BC cells were found at the level of all nerve roots in the spinal cord and hindbrain both in wild-type E11 and E13 embryos (Fig. S2M,N,Q,R) but also in Ntn1^−/− mutant embryos (Fig. S2O,P,S,T). This shows that the invasion of the PNS by commissural neurons is not due to a lack of BC cells.

Netrin 1 has several receptors, including Dcc and Unc5s (Unc5a-d) (Ackerman et al., 1997; Kolodziej et al., 1996). It has previously been shown that commissural axons also enter the DRGs in Dcc knockouts but not in Unc5a/Unc5c knockouts (Laumonnerie et al., 2014). We could confirm this result using whole-mount immunolabeling for Robo3 on E11 Dcc^−/− embryos. As in Ntn1 mutants, Robo3⁺ axons were detected within the trigeminal and vestibular nerves at E11 (Fig. 4A, Table S3), E13 (Fig. 4B-D, Table S3) and E16 (Fig. 4E). Some were pontine neurons, as shown with Pax6/Barhl1 immunolabeling (Fig. 4E-F). A similar defect was seen in a second Dcc null allele, Dcc^Δ/Δ, resulting from the intercross of Dcc^fl/fl mice to a line expressing cre in the germline (see Materials and Methods; Fig. 4G). To determine whether Dcc acts in pontine neurons to constrain their migration to the CNS, we next crossed the Dcc^fl/fl mice to Wnt1:Cre line (Danielian et al., 1998; Zelina et al., 2014). As in Dcc^−/− embryos, pontine neurons were unable to migrate towards the floor plate in Dcc^Δ/Δ and Wnt1:Cre;Dcc^fl/fl E16 embryos (Zelina et al., 2014 and data not shown). In addition, Robo3⁺ axons and Barhl1⁺/Pax6⁺ neurons were found in the trigeminal and vestibular ganglia of Wnt1:Cre;Dcc^fl/fl E16 embryos (Fig. 4H-K,L), indicating that a the lack of Dcc in pontine neurons induces some of them to exit the CNS (Fig. 4M).

At early developmental stages, neural crest cells migrate out of the neural tube to colonize the embryo to form most of the PNS and a variety of tissue and organs. However, after neural crest cell migration is completed, the PNS and the CNS segregate and newly born CNS cells remain confined to the CNS. Sensory axons from the PNS can still enter the CNS but at specific locations (such as the dorsal root entry zone in the spinal cord). In the CNS, motor axons will cross the CNS/PNS boundary to project to their target muscles, but boundary cap cells prevent motor neurons from entering the nerves. In vitro evidence also suggest that meninges might also control the CNS/PNS border (Suter et al., 2017). However, the cellular and molecular mechanisms that shape the CNS/PNS interface are not well characterized. Here we show that netrin 1, which is secreted at the CNS basal lamina by neural precursor endfeet, prevents various populations of hindbrain commissural neurons, in particular pontine neurons, from migrating into the PNS through nerve roots. In the spinal cord and in the hindbrain, a subset of commissural axons is misguided from early developmental stages and project into nerve roots. Therefore, it is possible that these first escapers lead the way for later-born commissural neurons in particular precerebellar neurons that migrate close to the pial surface and in the vicinity of trigeminal and auditory nerve roots. We propose that commissural neurons do not actively avoid nerve roots in a repulsive manner, but that they preferentially extend on netrin 1, which appears largely absent from the nerve roots. Without netrin 1, the growth of commissural axons and the migration of pontine neurons is more erratic and randomized, and they can invade the nerve roots. Although motor neurons express Dcc, like commissural axons, their axons can extend into the PNS. However, it has previously been shown that Dcc is inactive in motor neurons as it is cleaved by presenilin, and that they are unresponsive to netrin 1 within the spinal cord (Bai et al., 2011).

Interestingly, during normal development pontine neurons exhibit features of so-called collective migration, previously described for neural crest cells and lateral line neurons, among others (Friedl and Mayor, 2017). Pontine neurons migrate along each other in a compact stream from the rhombic lip to the floor plate. We show that, in the absence of netrin 1, pontine neuron cohesion appears affected and some escape from the main stream to invade the nerve roots. This suggests that netrin 1 might control collective cell migration in this system. Our results suggest that Dcc mediates netrin 1 function at the CNS/PNS boundary. However, the milder pontine neuron emigration defects observed in Dcc KO compared with Ntn1 KO indicate that another receptor, such as the Dcc paralogue neogenin (Keino-Masu et al., 1996) could also contribute. Unc5 receptors are unlikely to be involved, as pontine neurons remain in the CNS in Unc5b and Unc5c knockouts (Di Meglio et al., 2013; Kim and Ackerman, 2011). Together, our results reveal a novel molecular mechanism controling the establishment of the CNS/PNS boundary (Fig. 4M).

Ntn1^ßgeo (Serafini et al., 1996) and Dcc (Fazeli et al., 1997) knockout lines have been previously described and genotyped by PCR. The Ntn1 conditional knockout and the Ntn1-null allele were generated as described elsewhere (Dominici et al., 2017). To generate a null allele of Dcc, Dcc^fl/fl mice (Krimpenfort et al., 2012) were crossed to Krox20:Cre mice, which express Cre recombinase in both male and female germlines after sexual maturity (Voiculescu et al., 2000).

To ablate netrin 1 and Dcc expression from their different sources, we used different Cre lines: for the floor plate cells we used the Shh:Cre line (Harfe et al., 2004) (Jackson Laboratories); for the ventricular zone precursors we used the Nestin:Cre line (Tronche et al., 1999); and, finally, for the rhombic lip derivatives (pontine neurons), we used the Wnt1:Cre line (Danielian et al., 1998). The Ai9 Rosa^tdTomato reporter line (Rosa^Tom; Jackson Laboratories) was used to analyze the Cre expression driven by the different lines. All mice were kept in C57BL/6 background and the day of females vaginal plug was counted as embryonic day 0.5 (E0.5). Mice were anesthetized with ketamine (100 mg/ml) and xylazine (10 mg/ml). All animal procedures were carried out in accordance to institutional guidelines and approved by the UPMC University ethic committee (Charles Darwin). Embryos of either sex were used.

Antisense RNA probes were labeled with digoxigenin-11-d-UTP (Roche Diagnostics) as described elsewhere (Marillat et al., 2004), by in vitro transcription of cDNA encoding mouse Ntn1 (Serafini et al., 1996), mouse Ntn1 exon 3 (Dominici et al., 2017) and mouse Prss56 (Coulpier et al., 2009).

Fixation was performed by embryo immersion in 4% paraformaldehyde in 0.12 M phosphate buffer (pH 7.4) (PFA) overnight at 4°C. Samples were cryoprotected in a solution of 10% sucrose, for E11 and E13 embryos, and 30% sucrose for E16 embryos, in 0.12 M phosphate buffer (pH 7.2), frozen in isopentane at −50°C. Immunohistochemistry was performed on cryostat sections (20 µm) after blocking in 0.2% gelatin in PBS containing 0.25% Triton-X100 (Sigma). Sections were then incubated overnight at room temperature with the following primary antibodies: goat anti-human Robo3 (1:250, R&D Systems, AF3076), goat anti-Dcc (1:500, Santa Cruz, sc-6535), rat anti-mouse netrin 1 (1:500, R&D Systems, MAB1109), mouse anti-Nestin-Alexa488 (1:1000, Abcam, ab197495), rabbit anti-β-gal (1:500, Cappel, 55976), rabbit anti-Dsred (1:500, Clontech, 632496), rabbit anti-Pax6 (1:500, Millipore, AB2237), rabbit anti-Barhl1 (1:500, Sigma, HPA004809), rabbit anti-mouse Islet1 (1:500, Abcam, ab20670), anti-Sox10 (1:500, Santa Cruz, sc-17342), goat anti-FoxP2 (1:500, Santa Cruz, sc-21069) and rabbit anti-FoxP2 (1:500, Abcam, ab16046). Corresponding secondary antibodies directly conjugated to fluorophores (Cy-5, Cy-3, Alexa-Fluor 647 from Jackson ImmunoResearch, or from Invitrogen) were incubated during 2 h. For netrin 1 immunostaining, an antibody retrieval treatment was performed as described previously (Dominici et al., 2017). Sections were counterstained with Hoechst (1:1000, Sigma). Slides were scanned with a Nanozoomer (Hamamatsu) and laser scanning confocal microscope (FV1000, Olympus). Brightness and contrast were adjusted using Adobe Photoshop.

Whole-mount immunostaining and 3DISCO clearing procedures have been previously described (Belle et al., 2014, 2017). 3D imaging was performed with a light-sheet fluorescence microscope (Ultramicroscope I, LaVision BioTec) using Inspector Pro software (LaVision BioTec). Images and 3D volume were generated using Imaris ×64 software (Bitplane).

In utero electroporation of PN neurons was performed as described previously (Zelina et al., 2014), with some modifications. Endotoxin-free plasmid DNA of pCX-EGFP (1 μg/μl) (provided by Dr M. Okabe, Osaka University, Japan) alone was diluted in PBS containing 0.01% Fast Green. Diluted DNA (1 μl) was injected with a glass micropipette into the fourth ventricle of E13.5 mouse embryo. Five electric pulses (45 V, 50 ms, 950 ms interval between pulses) were applied with CUY21EDIT or NEPA21 electroporators (NepaGene) using 5 mm diameter electrodes (CUY650-5, Nepagene). Electroporated embryos were partially dissected at E16 followed by whole-mount labeling using goat anti-human Robo3 (1:250, R&D Systems AF3076) and chicken anti-GFP (1:1000, Abcam, ab13970).

Two different individuals, blinded to the experimental conditions, performed Robo3⁺ axon volume and Barhl1⁺ cell quantifications. There was no randomization in the groups and any statistical method was used to predetermine sample sizes. Graphical representations show mean values±s.d. Statistical significance was measured using one-sided unpaired tests for non-parametric tendencies (Kruskal–Wallis and Mann–Whitney). For Robo3⁺ volume quantifications, a background subtraction was performed followed by 3D volumetric analysis, using Imaris ×64 software, to determine the total volume of Robo3⁺ fibers in the trigeminal nerve. The number of Barhl1⁺ cells in the auditory and trigeminal nerve roots was quantified within a rectangular area (340×380 µm) in five different sections. Two sections were taken at the auditory nerve root and three others at the trigeminal nerve root. Control embryos were from the same litters than the mutants. For both types of quantifications, at least four embryos of each genotype were quantified, from at least two different litters. In both quantifications, we considered differences to be significant when P<0.05 (see all statistical values in Tables S1-S4). All statistical analyses of the mean and variance were performed with Prism7 (GraphPad Software).

We thank Dr Piotr Topilko for providing the Prss56 cDNA, Dr Marc Tessier-Lavigne for providing the Ntn1βgeo and Dcc knockouts, and Dr Anton Berns for providing the Dcc conditional knockout line.