The dual developmental origin of spinal cerebrospinal fluid-contacting neurons gives rise to distinct functional subtypes

Chemical and mechanical cues from the cerebrospinal fluid (CSF) can affect the development and function of the central nervous system (CNS). How such cues are detected and relayed to the CNS remains elusive. Cerebrospinal fluid-contacting neurons (CSF-cNs) situated at the interface between the CSF and the CNS are ideally located to convey such information to local networks. In the spinal cord, these GABAergic neurons expressing the PKD2L1 channel extend an apical extension into the CSF and an ascending axon in the spinal cord. In zebrafish and mouse spinal CSF-cNs originate from two distinct progenitor domains characterized by distinct cascades of transcription factors. Here we ask whether these neurons with different developmental origins differentiate into cells types with different functional properties. We show in zebrafish larva that the expression of specific markers, the morphology of the apical extension and axonal projections, as well as the neuronal targets contacted by CSF-cN axons, distinguish the two CSF-cN subtypes. Altogether our study demonstrates that the developmental origins of spinal CSF-cNs give rise to two distinct functional populations of sensory neurons. This work opens novel avenues to understand how these subtypes may carry distinct functions related to development of the spinal cord, locomotion and posture.

There is evidence that spinal CSF-cNs do not constitute a homogeneous population of neurons. In particular, these cells originate from distinct progenitor domains and are specified differentially by several cascades of transcription factors 26,[29][30][31][32] . In zebrafish, CSF-cNs are subdivided into the ventral population, referred to as KA", originating from the progenitor domain p3 and a dorsal population, referred to as KA' , originating from pMN 29,30 . In mouse, spinal CSF-cNs were recently shown to originate from the p3 and p2 progenitor domains 26 . In addition, some secreted compounds have been previously reported in a restricted number of spinal CSF-cNs, such as the somatostatin 33,34 or serotonin 35,36 .
The zebrafish has emerged as an ideal model organism to study the development 13,29,30 , the morphology and physiological roles of CSF-cNs in vivo due to its transparency at early stages 12,15,37 . Yet, few functional markers of CSF-cNs related to their sensory or secretory functions have been identified in this species. Here, we investigate whether the two types of spinal CSF-cNs defined by distinct developmental origins can be discriminated by morphology, projection on neuronal targets and expression of secreted compounds. Using a quantitative measure of cell shape, we show that ventral and dorsal CSF-cNs have differently shaped apical extensions as well as different axonal projections and consequently project onto distinct neuronal targets within the spinal cord. By electron microscopy (EM), we found that both populations of CSF-cNs exhibit large granular vesicles in accordance with secretory properties. We demonstrate that these two cell types express distinct modulators and peptides: while ventral CSF-cNs transiently express serotonin, dorsal CSF-cNs express the sst1.1 somatostatin paralog. We observed that the Pkd2l1 channel is not required for the differentiation of CSF-cN axonal projections, for the expression of serotonin or somatostatin. Altogether, our results show that spinal CSF-cNs constitute two distinct functional cell types that differ in apical and axonal morphology, neuronal targets within the spinal cord as well as in the transient expression of secreted compounds.

CSF-cNs are neurosecretory cells with an apical extension into the central canal.
We characterized CSF-cN ultrastructure in transverse sections of the 2.5 days post fertilization (dpf) zebrafish spinal cord by transmission electron microscopy (TEM). Without specific markers, we observed cells around the central canal that bore an apical extension with microvilli (Supplemental Fig. 1). To demonstrate these cells were CSF-cNs, we transiently expressed the engineered genetically-encoded peroxidase APEX2 under the pkd2l1 promoter (see Materials and Methods, Supplemental Fig. 2). In APEX2-TagRFP + larvae, we performed diaminobenzidine (DAB) staining (see Material and Methods, and Supplemental Fig. 3). DAB was detected only within CSF-cNs in the spinal cord, making the cells appear darker (Supplemental Fig. 3b). The specific DAB staining (Figs 1 and 2) revealed that CSF-cNs were either round (Figs 1a and 2a-b) or pear-shaped (Fig. 1b). Ventral and dorsal CSF-cNs bear an apical extension with several microvilli directed toward the lumen of the central canal (Figs 1c and  2c). Among actin-based microvilli, we observed, as previously described in one cell 37 , a single cilium directed toward the CSF (Fig. 1d, n = 12 ventral CSF-cNs, Fig. 2d, n = 3 dorsal CSF-cNs). This cilium bears two central microtubule singlets along the axoneme (Figs 1d and 2d-e), often seen in a kinocilium 38,39 . In APEX2 + CSF-cNs, we observed large granular vesicles (Figs 1e and 2f), suggesting that these cells either release or uptake peptides or neuromodulators into or from the CSF. Interestingly, we observed one axo-somatic symmetric synaptic contact at the basal pole of a CSF-cN (Fig. 1f), suggesting that CSF-cNs receive inhibitory synaptic inputs. Indeed, asymmetric synapses usually contain glutamate, are non-GABA immunoreactive and are therefore considered excitatory, while symmetric synapses contain GABA and are considered inhibitory [40][41][42] . Altogether, our TEM data show that both populations of CSF-cNs exhibit properties of sensory and secretory cells.
CSF-cNs display different shapes of apical extensions and axonal projections. As a quantitative analysis of CSF-cN morphology is difficult using serial EM, we turned to fluorescence for a quantitative comparison of the apical extension and the axonal projections of dorsal and ventral CSF-cNs. First, we tested whether these cells differ in the morphology of their apical extension. We investigated the shape of the CSF-cN apical extension by labeling the membrane of CSF-cNs in Tg(pkd2l1:Gal4;UAS:TagRFPCAAX;cmlc2:eGFP)icm22 larvae or F-actin itself in Tg(pkd2l1:Gal4;UAS:LifeAct-GFP;cryaa:V)icm28 larvae or larvae transiently expressing (pkd2l1:Gal4) and (UAS:LifeAct-TagRFP;cryaa:C). Using these different approaches, we found that the shape of the apical extension differs between ventral and dorsal CSF-cNs at 3 and 6 dpf (Fig. 3a,b and Material and Methods). The apical extension of dorsal CSF-cNs (Fig. 3a,b) was more extended along the border of the central canal than for ventral cells (Fig. 3a,b). The apical extension in ventral CSF-cNs was overall more compact with a small proportion displaying a more extended apical extension similar to dorsal CSF-cNs (13.3%, n = 6 out of 45 ventral CSF-cNs, Fig. 3a, empty arrow). Next, to assess whether ventral and dorsal CSF-cNs differ in their axonal projections, we injected the (pkd2l1-TagRFP) DNA construct to label single cells (Supplemental Fig. 4a) and reconstructed their axon (Fig. 3c, Supplemental Fig. 4b,c). All CSF-cN axonal projections were ventral, ipsilateral and ascending, yet they were heterogeneous along the rostrocaudal axis of the spinal cord (n = 54, Fig. 3c). We measured significant differences between the axonal projections of the two populations of CSF-cNs in terms of the area of the axonal arborization, the length of the axon, the dorso-ventral coverage within the spinal cord and the number of axonal branches (Fig. 3d). The larger axonal length, arborization and number of branches in ventral CSF-cNs compared to dorsal cells suggest these ventral cells possibly differentiate before dorsal ones. Nonetheless, the fact that the dorsal and ventral populations of CSF-cNs project on different domains along the dorso-ventral axis of the spinal cord suggests that these two populations may target different neuronal types within the ventral spinal cord.
Distinct neuronal targets for ventral and dorsal CSF-cNs within the spinal cord. Differences in the axonal projection of ventral and dorsal CSF-cNs suggest these cells may project onto distinct targets within the spinal cord. Our lab recently demonstrated that CSF-cNs project onto caudal primary (CaP) motor   . The apical extension of dorsal CSF-cNs extends more than for ventral CSF-cNs (two-sample t-tests, p < 5 · 10 −7 ) and this difference persists at later stages (6 dpf, p < 0.002). (c) The reconstruction from dorsal (light shade) and ventral (dark shade) CSF-cNs from different segments (Seg) illustrates the diversity of axonal morphologies CSF-cNs between the two types along the spinal cord (n = 11 for each type). Vertical black bars represent the dorso-ventral limits of the spinal cord. Cells are positioned according to their dorso-ventral (D-V) position with dorsal edge set to 1 and ventral to 0. (d) Comparison of ventral and dorsal CSF-cNs for axonal arborization area, total axon length, number of branches and axonal arborization dorso-ventral range (n = 39 and 15 cells respectively). Ventral CSF-cNs have a wider axonal arborization (p < 0.003), a longer axon (p = 0.0014), reach more ventral domains of the spinal cord (p < 9 · 10 −4 ), and cover a larger dorso-ventral (D-V) range (p < 2 · 10 −4 ) with more axonal branches (p < 0.02). Two-sample t-tests were performed to compare the two populations. Scale bar = 10 μm (a) and 20 µm (c). neurons and commissural primary ascending (CoPA) sensory interneurons within the escape circuit 43 as well as onto V0-v interneurons within the slow swimming circuit 15 . We investigated which CSF-cN type projects onto these targets using mosaic labeling in transgenic lines labeling CaP (Tg(parg mnet2 -GFP) 43,44 ), V0-v (Tg(vglut2a:lox:DsRed:lox:GFP) 15,45 ) or CoPA (Tg(tbx16-GFP) 43,46 ). We observed that only ventral CSF-cNs form the stereotypical basket structure around the soma of CaP motor neurons (Fig. 4a, n = 7 out of 8 ventral CSF-cNs found projecting onto CaP and 0 out of 5 dorsal CSF-cNs). In contrast, we found only dorsal CSF-cNs contacting ventrolateral glutamatergic cells, putatively V0-v based on their location and our previous findings 15, [47][48][49] (Fig. 4b, n = 3 out of 3 dorsal CSF-cNs and 0 out of 5 ventral CSF-cNs). Interestingly, we observed that both ventral and dorsal CSF-cNs project onto CoPA sensory interneurons (Fig. 4c,d, n = 4 dorsal and 5 ventral CSF-cNs). Altogether, our observations demonstrate a complex connectivity pattern from CSF-cNs onto their neuronal targets in the spinal cord. While some targets receive projections from both CSF-cN types (such as CoPA sensory interneurons), others seem to only receive inputs from either ventral (CaP motor neurons) or dorsal (ventrolateral glutamatergic neurons, most likely V0-v interneurons) CSF-cNs.
Investigation of secreted compounds expressed in dorsal and ventral CSF-cNs. We tested whether ventral and dorsal CSF-cNs differentially express secreted compounds previously reported in a restricted number of CSF-cNs, namely somatostatin 14,21,33,34 and serotonin 35,36 . By taking advantage of pkd2l1 expression in CSF-cNs 13 , we used the Tg(pkd2l1:GCaMP5G)icm07 transgenic line to selectively target CSF-cNs 37 . We demonstrated the specificity of this line using fluorescent in situ hybridization (FISH) for pkd2l1 mRNA coupled to GFP immunohistochemistry (IHC) to amplify the endogenous GCaMP5G signal from 24 hours post fertilization (hpf) to 5 dpf (Supplemental Fig. 5).
In zebrafish, somatostatin immunoreactivity has previously been reported 12 , but 6 different paralogs exist (sst1.1, sst1.2, sst2, sst3, sst5 and sst6) 50,51 . Testing the expression of all of them, we only detected expression of sst1.1 in CSF-cNs. Expression of sst1.1 was observed only in dorsal CSF-cNs ( Fig. 5a,b, arrows) in the rostral part of the spinal cord, from segment 1 to 13 at 24, 48, 72 hpf and in the adult spinal cord (data not shown). Since sst1.1 had been reported as expressed transiently in motor neurons from 19 hpf until 55 hpf 52 , we showed using FISH for sst1.1 with GFP IHC in 24 hpf Tg(mnx1:GFP) embryos 53 that sst1.1 expression was excluded from motor neurons (Fig. 5c). Altogether, our results show that sst1.1 is expressed in dorsal CSF-cNs and absent in ventral ones at early stages and restricted as well to a subpopulation of CSF-cNs in the adult. Next, we tested whether zebrafish CSF-cNs were serotoninergic by performing double IHC for 5-HT and GFP on Tg(pkd2l1:GCaMP5G)icm07 embryos and larvae from 24 hpf to 72 hpf. At 24 hpf, no 5-HT immunostaining was detected in the zebrafish spinal cord (data not shown). At 48 hpf, we detected 5-HT in ventral CSF-cNs (Fig. 5d, arrowheads) in the rostral part of the spinal cord (from segment 1 to 24). This expression was restricted to a subset of ventral CSF-cNs (Fig. 5d). The proportion of 5-HT + CSF-cNs at 48 hpf decreased along the rostrocaudal axis of the spinal cord (segments 3 to 6: 88.6 ± 16.0%, 96 cells; segments 10 to 13: 73.2 ± 27.3%, 100 cells; segments 23 to 26: 31.2 ± 33.3%, 94 cells). At 3 dpf, although 5-HT was still expressed along the entire spinal cord in other cells, the staining was absent in CSF-cNs in the rostral spinal cord (Fig. 5e). In conclusion, dorsal CSF-cNs express sst1.1 while a subpopulation of ventral CSF-cNs population is transiently serotoninergic. Hence, these two populations originating from different progenitor pools differentiate into distinct cell types expressing specific secreted compounds. pkd2l1 null mutation does not impact on the differentiation of spinal CSF-cNs. We previously showed that the PKD2L1 channel is expressed in spinal CSF-cNs across multiple vertebrate species 13 and that the channel appears necessary to mediate detection of spinal bending by CSF-cNs in the larva 37 . In pkd2l1 mutants, we observed defects in locomotion that we interpreted as being caused by the lack of sensory responses in CSF-cNs 37 .
To strengthen the conclusions of this previous study, we sought to precisely assess whether CSF-cNs develop properly in the pkd2l1 mutants. We found that in mutants, CSF-cNs were still GABAergic, ventral ones were still serotoninergic, that the number of CSF-cNs was not impacted, and that the morphology of their axonal projections was not different from WT (Supplemental Fig. 6). We also wanted to determine whether the functional connection between ventral CSF-cNs and CaPs 43, 54 was impacted and found that light-mediated activation of mutants CSF-cNs was still able to induce monosynaptic inhibitory postsynaptic currents in CaPs (Supplemental Fig. 7). Altogether, our results show that pkd2l1 is not required for CSF-cN differentiation.

Discussion
Spinal CSF-cNs were previously shown to originate from distinct progenitor domains characterized by distinct pools of transcription factors in the embryo 26,29,30 . Here we demonstrate that these two domains give rise to two cell types of CSF-cNs with probably distinct functional properties at the larval stages (Supplemental Fig. 8). First, the morphology of these two cell types differs both at the apical extension level as well as at the axonal level. Furthermore, we show that these differences in axonal projections lead to differences in neuronal targets reached in the spinal cord. Second, our ultrastructure data indicate that both ventral and dorsal CSF-cNs bear dense granules under the apical extension, suggesting possible roles in release or uptake of compounds from the CSF. We show that ventral and dorsal populations express distinct secreted factors, in particular 5-HT and sst1.1, during development. Finally we show that Pkd2l1 is not necessary for the differentiation of CSF-cNs. Altogether, our results demonstrate that the two cell types of CSF-cNs previously identified with different developmental origins segregate into distinct populations bearing specific functional properties.
Due to their peculiar apical extension contacting the CSF, it had been hypothesized that CSF-cNs may release compounds into the CSF 7, 55 . The presence of large granular vesicles (LGV) within CSF-cNs in our EM data, combined with expression of sst1.1 and 5-HT for dorsal and ventral CSF-cNs respectively, reinforces this hypothesis.
Previous works have shown that CSF-cNs are chemosensory 11, 24, 56, 57 as well as mechanosensory cells 37,43,58 . In zebrafish larvae, we previously showed that dorsal CSF-cNs respond to lateral bending of the spinal cord 37 while ventral CSF-cNs respond to longitudinal contractions 43 . The difference in shape of the apical extension between dorsal and ventral CSF-cNs may reflect a structural difference relevant for either their mechanosensory functions (direction of maximum sensitivity to bending of the tissue or CSF flow) or their secretory functions (with possibly larger surface area for dorsal CSF-cNs).  There are indications that spinal CSF-cNs harbor different morphologies in other species as well. In rat, spinal GABAergic CSF-cNs were first classified into three subtypes according to the shape of their soma 59 . Four morphological types of CSF-cNs were then described based on the shape of their soma, their axonal projection, and on the expression of the peptide Met-Enk-Arg-Gly-Leu 60 . These indications together with our data reveal a high level of heterogeneity among spinal CSF-cNs. Future studies will investigate the role of CSF-cN structural differences in sensory and secretory functions.
Using a single cell labelling approach, we show that ventral CSF-cNs have on average a longer and broader axonal arborization and cover a higher dorso-ventral spinal cord range with more axonal branches than dorsal CSF-cNs. These differences in axonal projections suggest the two CSF-cN populations might have distinct targets within the spinal cord. We had shown that CSF-cNs form active GABAergic synapses onto glutamatergic descending V0-v interneurons 15 as well as CaP primary motor neurons and CoPA sensory interneurons 43 . Here we show that only ventral CSF-cNs form the characteristic basket-like contact onto CaP soma while we only found dorsal CSF-cNs contacting ventrolateral glutamatergic neurons, most likely V0-v interneurons [47][48][49] previously shown to receive inputs from CSF-cNs 15 . Interestingly CoPA sensory interneurons received innervation from both dorsal and ventral CSF-cNs. Our results suggest that targets of the slow swimming circuits receive innervation from dorsal CSF-cNs while targets of the fast swimming circuit receive innervation from ventral CSF-cNs, and sensory interneurons involved in processing feedback receive innervation from both ventral and dorsal CSF-sNs. Further investigations will be necessary to establish whether other targets of the slow and fast swimming circuits follow the same dichotomy, receiving inputs from dorsal and ventral CSF-cNs respectively.
Multiple markers have been found in subsets of CSF-cNs across vertebrate species: somatostatin 12, 14, 21, 33, 34 , dopamine 61-64 and serotonin 35,36,65 . The expression of most markers was reported in a restricted fraction of CSF-cNs. Here we tested whether some of these markers showed restricted expression patterns among CSF-cNs. We found that ventral CSF-cNs were at 48 hpf transiently serotoninergic as previously described in other species 35,36,65 in accordance with findings from Montgomery et al. 66 , in zebrafish. In addition, we demonstrated that, even at 48 hpf, not all ventral CSF-cNs expressed 5-HT. This suggests that there may be different subtypes of ventral CSF-cNs based on 5-HT expression. Moreover, Montgomery et al. 66 , also showed that the rate-limiting enzyme involved in 5-HT synthesis tryptophan hydroxylase 1 a (tph1a) 67,68 is expressed in the ventral spinal cord at 24 and 48 hpf suggesting that ventral CSF-cNs could constitute a transient source of serotonin at early stages. One possible explanation for the disappearance of 5-HT in ventral CSF-cNs at later stages could be the emergence of descending fibers from 5-HT + neurons originating from the brainstem raphe nuclei from 48 hpf onwards 67 . 5-HT from descending fibers has been previously shown in mammals to suppress the monoaminergic expression of CSF-cNs [69][70][71] . The physiological relevance of the transient 5-HT expression among this ventral population of CSF-cNs remains to be elucidated.
Regarding the origin of somatostatin in zebrafish CSF-cNs, we found that among the six SST paralogs, sst1.1 was the only one expressed. We demonstrated that the expression of this gene is specific and restricted to the dorsal population of CSF-cNs, identifying for the first time a specific marker of dorsal CSF-cNs. We also demonstrated that this peptide is excluded from motor neurons contrary to what previous studies suggested 52 . The physiological relevance of sst1.1 expression by dorsal CSF-cNs is not known. In mammals and lamprey, it has been shown that somatostatin can reduce locomotor frequency 58,72,73 . The release of SST by dorsal CSF-cNs could impact the frequency of locomotor events along with the release of neuropeptides of the UII family, urp1 and urp2 by ventral CSF-cNs only 74 as previously suggested by the presence in coho salmon of a UII-like immunoreactive ventral population of CSF-cNs and a distinct somatostatinergic one 75 . Taken together, our results confirm that ventral and dorsal CSF-cNs express distinct peptides and neuromodulators. The transient nature of the expression of 5-HT suggest that it could have developmental roles to be further investigated.

Material and Methods
Animal care. Zebrafish (Danio rerio) adults and larvae were maintained and raised on a 14/10 hour light cycle. Fish lines used in this study are referenced in Table 1. Water was regulated at 28.5 °C, conductivity at 500 μS and pH at 7.4. All embryos and larvae under 6 dpf were anesthetized in 0.02% tricaine methane sulfonate (MS 222) (Sandoz, Levallois-Perret, France) and euthanized in 0.2% MS 222 prior to fixation. Experimental and animal protocols were approved by the Institut du Cerveau et de la Moelle Épinière in agreement with the French National Ethics Committee (Comité National de Réflexion Éthique sur l'Éxpérimentation Animale; Ce5/2011/056) and European regulations.Genotype protocol for the pkd2l1 icm02 mutants was described in ref. 37 (Böhm et al. 2016).

Generation of transgenic lines.
In order to generate the Tg(UAS:TagRFP-CAAX;cm lc2:eGFP)icm22 line, the TagRFP-CAAX sequence was amplified using a TagRFP-for ward  (5′-CCCGGGATCCACATGGTGTCTAAGGGCGAAG-3′) and reverse primer ( 5′-GA TC GC GG CC GC TC  AG GA GA GC AC AC AC T T GC AG CT CA TG CA GC C G GG GC CA CT CT CA TC AG GAGGGT TCA  GCTTATTAAGTTTGTGCCC-3′) and inserted into the pME-MCS vector 76 via BamHI/NotI restriction digestion. The resulting pME-TagRFP-CAAX vector was recombined via a Gateway reaction (MultiSite Gateway Three-Fragment Vector Construction Kit) with p5E-4nrUAS 77 , p3E-pA and pDest-Tol2; cmlc2:eGFP 76 resulting in (p4nrUAS:TagRFP-CAAX-pA-Tol2;cmcl2:eGFP). Using a similar approach, the pME_LifeAct-GFP plasmid was generated from the plasmid pCS2_LifeAct-GFP (kind gift from Nicolas David) using the SP6 promoter sequencing primer (5′-ATTTAGGTGACACTATAG-3′) and inserted into the pME_MCS vector via a BamHI/NotI restriction digestion. The final three-way gateway reaction used the pME_LifeAct-GFP, pDest_cryaa:V and p5′E_polyA plasmids to generate the (UAS:LifeAct-GFP;cryaa:V) construct. Microinjection of these plasmids was performed with Tol2 mRNA (25 ng/µl) following standard protocols. Transgenic founder fish Tg(UAS:TagRFP-CAAX;cmlc2 :eGFP)icm22 where screened based on GFP expression in the heart. These two lines were also screened based on transactivation of their respective transgene when crossed with various Gal4 lines.

Electron microscopy.
To label CSF-cNs for electron microscopy, we followed the procedure described in refs 37, 78 and Supplemental Figs 2 and 3. 2.5 dpf larvae selected for TagRFP expression in CSF-cNs were primarily anaesthetized in 0.02% tricaine methane sulfonate (MS 222) then euthanized in 0.2% tricaine prior to fixation in 2% glutaraldehyde in 100 mM sodium cacodylate buffer pH 7.4 to which 2 mM of CaCl 2 was added for 45 min. Whole embryos were then rinsed in the same buffer 5 × 2 min each. Functional aldehyde excess was blocked by treating embryos for 5 min with 20 mM glycine in the sodium cacodylate buffer. Other rinses in buffer (5 × 2 min) were then carried out. The APEX2 peroxidase was further revealed in diaminobenzidine (DAB, 0.5 mg/ml) to which 10 mM hydrogen peroxide (30%) was added (see Lam et al.), in 100 mM cacodylate buffer. After 5 min, the reaction was stopped by rinsing embryos twice in cold buffer before post fixation in 2% osmium tetroxide (OsO 4 ) in the same buffer for 45 min. Embryos were then rinsed in cold distilled water, stained in 2% uranyl acetate in distilled water overnight in a cold chamber. Animals were returned to room temperature, rinsed in distilled water, then dehydrated in a graded series of ethanol, cleared in acetone, and embedded in epoxy resin (EMBed812, Electron Microscopy Science, France). Embryos were oriented coronally in resin molds cured at 60 °C for 48 hr. Whole embryos were imaged using a Leica DMRB microscope to assess the presence of DAB positive CSF-cNs and determine their location within the animals. Embryos were first cut in 1 µm semi-thin sections with an ultramicrotome (Ultracut E, Leica). Sections were picked up every µm, stained with toluidine blue. Semi-thin sectioning was performed until the first level of DAB + CSF-cN cell appears. Then, serial ultra-thin sections (~70 nm thick) were collected onto copper grids (about 8 sections per grid). 12 ventral CSF-cNs, and 3 dorsal CSF-cNs were analyzed. They were contrasted in urany-less solution (Delta Microscopies, France) for 1-2 min, rinsed in distilled water, dried for at least 1 hr. Observations were made with an Hitachi HT 7700 electron microscope operating at 70 kV. Electron micrographs from DAB + CSF-cNs were taken at low (×6200), medium (×22,000) and high (×53,000) magnifications, using an integrated AMT XR41-B camera (2048 × 2048 pixels). In all the images displayed, dorsal is up and rostral is left.

Analysis of the apical dendritic extension of CSF-cNs.
To image the apical dendritic extension at a high resolution, we used different labeling strategies relying on transverse sections of the spinal cord. At 3 dpf, we used the stable Tg(pkd2l1:Gal4;UAS:TagRFP-CAAX;cmlc2:eGFP) transgenic line with membrane TagRFP and the transient expression of (pkd2l1:Gal4) 15 and (UAS:LifeAct-TagRFP;cryaa:C) DNA constructs injected into Tg(pkd2l1:GCaMP5)icm07 eggs at the one cell stage to label F-actin with LifeAct 79 . At 6 dpf, we took advantage of the stable Tg(pkd2l1:Gal4;UAS:LifeAct-GFP;cryaa:V)icm28 line. 3 and 6 dpf larvae were screened for expression in CSF-cNs and fixed using 4% paraformaldehyde (PFA) during 4 hours at 4 °C and immunostained as described below. We sliced 50 µm-thick transverse sections on larvae mounted in 3% low melting point agarose using a vibratome (HM 650 V Microtome, Thermo Scientific). For each cell, Z stacks (step size 0.25 µm) were acquired to image the entire apical extension and maximum projection was performed in Fiji 80 . The shape of the apical extension (S) was calculated as the ratio of the width (W, basal extension along the central canal) over the height (H, vertical extension within the central canal). W and H were extracted by drawing a polygon around the apical extension using the polygon tool in Fiji and estimated by the parallel and perpendicular axis, respectively, of the best fitting ellipse (Fig. 4b).
Single cell labeling. To assess whether pkd2l1 mutation led to a disruption of CSF-cN axonal refinement, we injected at 25 ng/μl the (pkd2l1-TagRFP) construct generated with a three-fragment Gateway recombineering reaction into single-cell stage embryos from pkd2l1 icm02/+ incrosses (see ref. 15). 3 dpf larvae selected for single CSF-cN expression were fixed with 4% PFA for 3 hours and immunostained following standard procedures. Genotyping of the larvae was performed after immunostaining. To assess the connectivity of ventral and dorsal CSF-cNs, the same procedure was followed when the (pkd2l1-TagRFP) construct has been injected in the Tg(par g mnet2 -GFP) 44 and Tg(tbx16-GFP) 46 lines and the (pkd2l1:Gal4) with (UAS:synaptophysin-GFP) 81 constructs in the Tg(vglut2a.lox-DsRed-lox-GFP)line 45 . Some of these animals have been imaged live.
Analysis of the axonal arborization of isolated CSF-cNs. To label and trace individual CSF-cNs, we followed the same procedure as described in ref. 15, in Supplemental Information and in Supplemental Fig. 4.
Fluorescent in situ (FISH). The pkd2l1 ISH probe was generated as previously described 13   Imaging. Images were acquired using an Olympus FV1000 confocal microscope equipped with a 20 and 40x water and 60X oil immersion objectives using the 405, 473 and 543 nm laser lines or using an upright microscope (Examiner Z1, Zeiss) equipped with a spinning disk head (CSU-X1, Yokogawa) and a modular laser light source (LasterStack, 3i Intelligent Imaging Innovations). To determine the overlap of GFP in the Tg(pkd2l1:GCaMP5G)icm07 transgenic embryos and larvae with pkd2l1, sst1.1 FISH or GABA and 5-HT IHC, fish were mounted laterally in 1, 5% agarose covered of Vectashield Mounting Medium (Vectorlabs, CA, USA). To analyze apical extensions, slices were transferred into Vectashield mounting medium as well (Vectorlab, CA, USA).
Electrophysiology and optogenetic stimulation. Procedures for optogenetic stimulation of CSF-cNs and electrophysiological recordings of CaP motor neurons are described in Supplemental Information and Supplemental Fig. 7.

Statistics. We used Student's t-tests for the morphological comparison of ventral versus dorsal CSF-cNs in
WT and pkd2l1 icm02/icm02 mutants and two-way ANOVAs to test the interaction between the genotypes and the cells investigated. The level of significance was p < 0.05 for all datasets. p values are represented as the following: (*)p < 0.05; (***)p < 0.001; (****)p < 0.0001.