The HMG box transcription factors Sox1a and Sox1b specify a new class of glycinergic interneuron in the spinal cord of zebrafish embryos

The spinal cord of vertebrates consists of multiple classes of neurons and glial cells, which differentiate from distinct progenitor domains along the dorsoventral (DV) axis of the embryonic spinal cord (Dessaud et al., 2008). Initial differences along the DV axis are triggered by signals like sonic hedgehog from the notochord and floor plate forming a gradient in the ventral neural tube. Dorsally, other signals, like BMPs and Wnts, act to instruct neurogenesis. These external signals trigger the expression of combinations of transcription factors in the progenitor domains at distinct DV levels along the spinal cord. As a consequence, these levels produce different types of postmitotic neurons at different DV positions of the spinal cord (Goulding, 2009).

Further subdivisions into neuronal subtypes occur within the initially specified domains. V2 interneurons located in the ventral half of the spinal cord are derived from the p2 progenitor domain. In mouse, three different subtypes of V2 interneurons (termed V2a, V2b and V2c) are generated in this domain (Karunaratne et al., 2002; Li et al., 2005; Panayi et al., 2010; Smith et al., 2002; Zhou et al., 2000). Postmitotic V2a interneurons are characterized by expression of the homeobox transcription factor Vsx2 (Chx10) (Ericson et al., 1997; Kimura et al., 2006). V2b interneurons express the zinc-finger transcription factors Gata2 and Gata3, as well as the bHLH transcription factor Tal1 (Scl) (Karunaratne et al., 2002; Muroyama et al., 2005; Smith et al., 2002), and V2c interneurons express the HMG group transcription factor Sox1 (Panayi et al., 2010). V2a neurons are glutamatergic excitatory neurons, whereas V2b cells are GABAergic inhibitory neurons. No neurotransmitter type has been reported for V2c interneurons. Studies in mouse support a signalling cascade in which Foxn4 acts upstream of Gata2. Gata2 in turn activates Tal1 (also known as Scl), which finally induces Gata3, specifying V2b interneurons (Del Barrio et al., 2007; Muroyama et al., 2005; Nardelli et al., 1999). Sox1 is required for V2c development (Panayi et al., 2010).

In the spinal cord of zebrafish, the glutamatergic V2a and the GABAergic V2b interneurons are also present, suggesting a conserved structure of the vertebrate spinal cord with respect to these V2 interneurons (Batista et al., 2008). Because of their distinct axonal projections, V2a and V2b interneuron homologues were previously referred to as CiD and VeLD, respectively (Batista et al., 2008; Bernhardt et al., 1992). They express markers related to those of the mouse V2a and V2b cells (Batista et al., 2008; Kimura et al., 2008), and they differentiate at the p2 level by asymmetric divisions of a basally located V2a/b progenitor. The orientation of the division planes is stochastic with respect to the axes of the spinal cord (Kimura et al., 2008). The expression of vsx1 and vsx2 is detected in mature V2a interneurons, whereas mature V2b interneurons express gata2a (previously gata2), gata3, tal1 and tal2 (Batista et al., 2008; Kimura et al., 2006). Immediately after birth, however, the precursors can share gene expression patterns, e.g. vsx1 and gata2a are co-expressed in these precursors. Moreover, the persistence of GFP expressed from a vsx1:GFP transgene is an ideal tracer of cell lineage, as V2b cells also continue to harbour fluorescence, albeit at lower levels, for a significant time after birth (Kimura et al., 2008). The decision to develop into a V2a or V2b interneuron is affected by Notch signalling (Batista et al., 2008; Del Barrio et al., 2007; Kimura et al., 2008; Peng et al., 2007). Genetic inhibition of Notch signalling in homozygous mindbomb zebrafish mutants causes the V2a/b precursor cells to develop into V2a at the expense of V2b cells (Batista et al., 2008). By contrast, forced activation of Notch signalling shifts cell fate towards the V2b fate (Kimura et al., 2006). In the mouse, V2c neurons are derived from Gata3⁺ cells in the V2 domain (Panayi et al., 2010). It is not clear whether V2c interneurons exist in the zebrafish spinal cord, even though expression of sox1a and sox1b has been detected (Andrzejczuk et al., 2018; Armant et al., 2013; https://itgmv3.itg.kit.edu/ffdb/index.html). Interestingly, V2b regulatory genes are also expressed in more ventrally located, cerebrospinal fluid-contacting Kolmer-Agduhr interneurons in the zebrafish (Andrzejczuk et al., 2018; Yang et al., 2010). However, the jointly expressed transcription regulators appear to be connected into distinct regulatory networks (Yang et al., 2010).

Studies in mice and zebrafish indicate that V2 interneurons fine-tune the output of motor neurons (Ampatzis et al., 2014; Zhong et al., 2011). V2a neurons organized in different units activate distinct motor neuron pools during fast locomotion, such as gallop in the mouse or escape behaviour in zebrafish (El Manira, 2014; Song et al., 2018). V2b neurons in mouse are involved in limb-flexion movements (Britz et al., 2015). So far, nothing is known about the function of V2c interneurons.

Here, we have investigated the role of sox1a and sox1b in the specification of V2 interneurons in the zebrafish spinal cord. We provide evidence for the existence of a third V2 interneuron type in the zebrafish spinal cord, which is distinct from V2a and V2b. We refer to this neuronal cell type as V2s. This neuronal subtype initially and transiently shares marker expression with V2b cells. During the second day of development, however, V2s cells develop into glycinergic cells and present axonal projections distinct from V2a and V2b cells. Both sox1a and sox1b genes are required for specifying V2s interneurons. This dependence on sox1 gene activity is reminiscent of mouse V2c interneurons. Our data are consistent with a hierarchical model of two precursor types producing first a pair of cells giving rise to V2a cells and V2b,s precursors (gata2a⁺; gata3⁺; tal1⁺; tal2⁺; gad1b⁺; sox1a/b⁺) followed by a second fate decision giving rise to V2b and V2s cells. V2s cells require Notch signalling to differentiate. Unilateral ablation of V2s cells in zebrafish embryos causes a delay in startle movements upon touch, suggesting that V2s cells are required for fast escape responses.

In an expression screen of transcription regulators in the 24 h post fertilization (hpf) zebrafish embryo (Armant et al., 2013), we detected expression of two closely related HMG box transcription factor genes, sox1a and sox1b, in the brain and in neurons of the spinal cord (https://itgmv3.itg.kit.edu/ffdb/index.html). sox1a and sox1b mRNAs are expressed in the ventral telencephalon, ventral diencephalon, the lens, the spinal cord and the lateral line (Fig. 1A,B, Fig. S1A-C′, data not shown). In transverse sections through the spinal cord of 24 hpf embryos, the mRNA of sox1a and sox1b was detectable in single cells in and next to the lateral floor plate, and in slightly more dorsal cells (Fig. S1C,C′). The ventral cells are in contact with the ventricle and therefore represent Kolmer-Agduhr KA′ (Fig. S1C,C′) and KA″ (Fig. S1C,C″) interneurons. A third type of sox1a- and sox1b-expressing cell was located more dorsally, close to the pial surface (Fig. S1C,C′). We mapped the dorsal sox1a- and sox1b-expressing cells relative to the V1/V0 and the motor neuron domain by staining Tg(dbx1b:eGFP) and Tg(olig2:eGFP) transgenic animals using sox1a and sox1b antisense mRNA (Fig. S1D-E″). The location of sox1a- and sox1b-expressing cells relative to olig2 and dbx1b markers suggests that the dorsal sox1a⁺ and sox1b⁺ cells belong to the V2 class of interneuron. We next assessed whether sox1a and sox1b are co-expressed in KA′, KA″ and V2 domain cells by staining embryos with sox1a and sox1b antisense mRNA. Although differences in onset of expression in the spinal cord were noted during earlier development (Fig. S1B,B′), sox1a and sox1b have similar patterns of expression in the spinal cords of 24 hpf embryos (Fig. 1C-E). Among all sox1a- and sox1b-expressing cells in the V2 domain, around 80% co-express sox1a and sox1b mRNA. Similar percentages of co-expression were noted for KA′ and KA″ cells (Fig. 1D).

We next investigated whether sox1a⁺ and sox1b⁺ cells express markers of the known V2 neuronal subtypes, V2a and V2b, in the zebrafish spinal cord (Table S1). The sox1a⁺ and sox1b⁺ cells only rarely express the mRNA of the V2a interneuron marker vsx2 (Fig. S2A-A″,F-F″) (Batista et al., 2008; Ericson et al., 1997). At 24 hpf, sox1a⁺ and sox1b⁺ cells partially co-express the transcription factors gata2a, gata3 and tal2, which are known to be expressed in V2b cells (Batista et al., 2008; Yang et al., 2010) (Table S1; Fig. S2B-D″,G-I″). Glutamic acid decarboxylase (gad1b), a marker for GABAergic neurons such as V2b interneurons, was expressed in 90% of sox1a⁺ and 80% of sox1b⁺ cells in the V2 domain (Fig. S2E-E″,J-J″). Thus, sox1a⁺ and sox1b⁺ cells do not express markers of V2a cells but share the expression of markers characteristic for V2b cells at 24 hpf (Table S1).

In a functional screen of conserved regulatory sequences in genes expressed in the central nervous system (Naville et al., 2015; data not shown), we noted that an integrated Tg(dmrt3a-gata2a:eGFP) chimeric reporter expressed eGFP in a pattern very similar (Fig. S3A) to that seen for the sox1a and sox1b genes (Fig. 1A,B). We verified this notion by staining transgenic embryos with sox1a and sox1b mRNAs. The overall pattern of expression matches that of sox1a and sox1b (Fig. S3C-F). This pattern, however, does not match that of the dmrt3a gene (https://itgmv3.itg.kit.edu/ffdb/index.html) (Satou et al., 2013). Sequencing of the genome of the transgenic embryos revealed an insertion of the transgene into the exon of the sox1a gene (Fig. S3B,B′). The reporter therefore appears to be expressed under the control of the endogenous sox1a gene.

In time-lapse studies of sox1a:eGFP embryos, eGFP expression is initiated in single cells in the V2 domain that did not divide anymore prior to extension of axons (Movie 1). We next assessed the morphology of transgene expressing cells in the V2 domain at 60 hpf sox1a:eGFP embryos. The eGFP-expressing cells in the ventral spinal cord are too densely packed to assess the projections of individual neurons (Fig. 2A). We thus reduced the number of eGFP-expressing cells by knocking out eGFP with a guide RNA/Cas9 approach. The eGFP⁺ cells in the V2 domain have an oval cell nucleus located at the pial surface and extend an axon ventrally (Fig. 2B,D, 100%, n=50 cells). This axon bifurcates in all examined cases with a short branch ending at the ventral region of the spinal cord (Fig. 2B′-D′, yellow arrows). The other branch descends ipsilaterally for five or six somites rising to an intermediate dorsoventral level (Fig. 2D, arrowheads). This pattern of axon branching and extension is different from the patterns of the V2a (CiD) and V2b (VeLD) interneurons (Batista et al., 2008; Kuwada et al., 1990).

We verified these data by a different way of cell labelling to exclude adverse effects of the transgene integration or the CRISPR/Cas9 approach to reduce the number of eGFP-expressing cells. We labelled a sox1a-encoding BAC clone by recombineering with a GFP reporter cassette [TgBAC(sox1a:eGFP)]. After microinjection of TgBAC(sox1a:eGFP) into fertilized eggs, transgene expression was obtained in single V2 domain cells in the spinal cord of the G0 generation at 48 hpf (Fig. 2C). The eGFP-expressing V2 cell showed an oval-shaped soma and a ventrally extending axon with a short branch stretched towards the ventral midline of the spinal cord and a long ipsilaterally descending axon branch (Fig. 2C,C′, n=20 cells). These data fully support the observations obtained with the stably integrated reporter in the sox1a:eGFP transgenic line. The distinct patterns of axonal projections suggest that the sox1a- and sox1b-expressing neurons in the V2 domain are different from V2a and V2b interneurons described previously (Bernhardt et al., 1992; Kimura et al., 2006) at this location of the zebrafish spinal cord. We will refer to these cells as V2s cells for their distinguishing expression of sox1a and sox1b.

We next assessed whether the V2s cells can be distinguished from V2a and V2b cells by their neurotransmitter production. V2a and V2b interneurons are glutamatergic and GABAergic, respectively (Batista et al., 2008; Kimura et al., 2008). sox1a- and sox1b-expressing cells do not significantly co-express markers of glutamatergic V2a cells (Fig. S2A-A″,F-F″). About 78% of sox1a:eGFP⁺ cells co-expressed gad1b, a marker for GABAergic cells, at 24 hpf (Fig. S4A-A″). Comparable results (92%) were scored in double-labelling experiments using sox1a antisense RNA instead of the eGFP fluorescence as a readout of sox1a⁺ cells in the V2 domain (Fig. S2E-E″). We also noted, however, that about 64% of sox1a⁺ cells co-expressed solute carrier family 6 (neurotransmitter transporter), member 5 (slc6a5, previously glyt2), a marker for glycinergic inhibitory neurons (Fig. 3A-A‴, Fig. S4B-B″). Thus, at 24 hpf a large number of sox1a-expressing cells are both GABAergic and glycinergic. This ratio changed over the next few hours of development (Fig. 3B, Fig. S4). At 30 hpf, about 45% of the sox1a⁺ cells expressed gad1b and about 70% expressed slc6a5 (Fig. S4C-D″). However, by 48 hpf, the majority of sox1a⁺ cells have stopped to express the GABAergic marker gad1b (Fig. S4E-E″) and have turned on expression of the glycinergic marker slc6a5 (Fig. 3B, Fig. S4F-F″). This suggests that mature V2s cells are inhibitory neurons employing the neurotransmitter glycine. Altogether, these results bring further support for the notion that V2s cells are different from the previously described V2a and V2b interneurons (Bernhardt et al., 1992; Kimura et al., 2006).

We next wondered whether such glycinergic V2 interneurons could be found in the spinal cord of mouse embryos. We performed immunodetection of Sox1 and in situ hybridization with Gad1 and Slc6a5 probes on adjacent transverse sections of embryonic day E11.5 and E12.5 mouse embryos. As shown previously (Adams et al., 1995; Katarova et al., 2000), Slc6a5 and Gad1 mRNAs were not detected in the ventral part of the spinal cord at E11.5, and hence in Sox1⁺ cells (data not shown). Conversely, at E12.5, Gad1 was detected in clusters of ventral cells that match those where Sox1⁺ interneurons (INs) were located (Fig. S4i-ii′,iv-v′). Slc6a5 mRNA was also detected in these clusters but only at more developed brachial levels (Fig. S4iii,iii′,vi,vi′). Taken together, these data support the idea that the mouse spinal cord contains glycinergic Sox1⁺ V2 cells.

Precursor cells at 24 hpf appear to co-express to a large degree markers of V2b cell fates (Andrzejczuk et al., 2018; this study). Our data suggest, however, that mature V2s cells differentiate from these co-expressing progenitors. To address this issue further, we double stained 24 and 30 hpf embryos with gata2a and sox1a probes, and counted sox1a⁺;gata2a⁺ double-positive cells in a five-somite-long segment of the V2 domain of the spinal cord over the yolk extension. At 24 hpf, 65% of sox1a⁺ cells (n=71) co-express gata2a (Fig. 4A, Fig. S5A-A″). In contrast, only 32% cells (n=34) are still positive for both markers at 30 hpf (Fig. 4A, Fig. S5B-B″). Over the same time course, the total number of cells expressing sox1a or gata2a are increased by 92% (n=111) and 62% (n=244), respectively. Thus, between 24 and 30 hpf, sox1a⁺ cells stop co-expressing the V2b marker gata2a, consistent with the appearance of glycinergic sox1a⁺ cells that no longer express the neurotransmitter marker gad1b (Fig. 3B). In contrast to these V2b markers, the gene NK1 transcription factor related 2-like,b (nkx1.2lb, also known as sax2; Bae et al., 2004) tightly follows sox1a expression. In 80% of cases, sox1a-expressing cells are positive for nkx1.2lb mRNA at both 24 hpf and 30 hpf (Fig. 4A, Fig. S5C-D″). Thus, nkx1.2lb is tightly linked to sox1a expression and V2s differentiation.

As V2s cells express initially V2b markers, V2s cells appear to derive from a common V2b,s precursor pool. We took advantage of the persistence of vsx1:GFP expression, which marks the V2a/V2b progenitors (Kimura et al., 2008) and their derivatives for some time, to assess the expression of sox1a and gata2a at 22, 24 and 26 hpf (Fig. 4B-E‴, Fig. S6). We detected four different combinations of marker gene expression (Fig. 4B-E‴, Fig. S6). All four classes of cells expressed vsx1:GFP, although at varying levels (Fig. 4B-E‴, Fig. S6), which presumably reflects the common origin but also the difference between cells actively expressing the transgene at this stage and cells in which transcription of the transgene has been stopped (Fig. 4B-E‴). Cells expressing vsx1:GFP only represent most likely terminally committed V2a interneurons (Fig. 4B-E‴). Except for a slight increase in vsx1:GFP⁺; sox1a⁺ cells (Fig. 4D-D‴, Fig. S6) the overall proportion of the four types of cells does not differ strongly across the short period of observation. The group of increasing vsx1:GFP⁺; sox1a⁺ cells are most likely committed V2s cells (Fig. 4D-D‴). In addition, we detected vsx1:GFP⁺; gata2a⁺ cells (Fig. 4C-C″) and cells that co-expressed vsx1:GFP, gata2a and sox1a (Fig. 4E-E‴). These patterns represent committed V2b and uncommitted V2b,s precursors, respectively. Thus, V2s cells derive from the V2 domain. These data are consistent with the hypothesis that the V2 domain harbours two precursor types: a V2a/V2b and a V2b,s precursor.

Forced activation of Notch signalling abolished the V2a fate (Kimura et al., 2008), whereas loss of Notch signalling in the mindbomb mutant shifted cell fate towards V2a cells at the expense of V2b cells (Batista et al., 2008). Thus, we asked whether Notch signalling is also involved in specifying V2s cells. We reasoned that this second window of Notch requirement is delayed with respect to the first one. We therefore chose a conditional approach by exposing sox1a:eGFP transgenic embryos to the γ-secretase inhibitor LY411575, which blocks Notch processing after ligand binding (Fauq et al., 2007). We determined the number of sox1a:eGFP/gad1b or sox1a:eGFP/slc6a5 co-expressing cells at 30 hpf in embryos exposed to LY411575 from 16 to 24 hpf (Fig. 4F). Blocking Notch signalling in this time window increased the number of GABAergic sox1a:eGFP⁺;gad1b⁺ cells by 2.3-fold (Fig. 4F, Fig. S5E-F″). At the same time, it reduced the number of glycinergic sox1a:eGFP⁺;slc6a5⁺ cells by 43% (Fig. 4F, Fig. S5G-H″). Thus, specification of V2s cells depends on Notch signalling.

We next tested when precursors of V2s cells become post-mitotic by monitoring incorporation of the nucleotide analogue EdU. We exposed embryos to EdU either between 16 and 24 hpf or between 24 and 30 hpf (Fig. 4G). Sox1a:eGFP transgenic embryos were labelled for incorporated EdU and cells were counted in a five-somite-spanning segment of the V2 domain in the spinal cord over the yolk extension (Fig. 4H-I″,J). More than 70% of sox1a:eGFP⁺ cells were positive for EdU incorporation in embryos exposed to the nucleotide analogue from 16 to 24 hpf (Fig. 4H-H″,J). In contrast, when embryos were exposed from 24 to 30 hpf, only 1% of the sox1a:eGFP⁺ cells had incorporated EdU (Fig. 4I-I″,J). Thus, the sox1a:eGFP⁺ cells that became gradually negative for V2b markers between 24 and 30 hpf were born before 24 hpf.

Disrupting either gata2a or gata3 translation alone produced at best a marginal effect on V2b marker expression (Yang et al., 2010). However, double knockdown of gata2a and gata3 reduced tal2-expressing V2b interneurons (Fig. 5A-A″) and expression of gad1b (Fig. 5B-B″) was also decreased in the V2 domain. Thus, gata2a and gata3 are required for differentiation of V2b cells. Reduced gad1b expression was recently also reported in gata2a;gata3 double mutants (Andrzejczuk et al., 2018). These genetic data underscore the specificity of our gata2a and gata3 morpholino knockdown approaches employed in this and a previous report (Yang et al., 2010). The expression of sox1a and sox1b in the V2 region was significantly reduced in the absence of Gata2a and Gata3 (Fig. 5C-D″). Thus, gata2a and gata3 are also required for sox1a and sox1b expression.

As V2b and V2s cells are derived from preceding fate decisions of common V2a/b,s progenitors that also transiently express gata2a (Batista et al., 2008; Kimura et al., 2008), we asked whether the V2a cells were affected by loss of Gata factors. Knockdown of gata2a and gata3 did not affect expression of vsx2, a marker of V2a cells (Fig. 5E-E″). Thus, Gata2a and Gata3 appear to be essential for controlling the V2b and V2s fates without influencing V2a interneuron differentiation. The distinct and specific effects on different neurons (Fig. 5A″-E″) underscore the specificity of the gata2a and gata3 morpholinos. This is further supported by the analysis of gata2a and gata3 mutants (Andrzejczuk et al., 2018).

We next assessed the function of sox1a and sox1b by CRISPR/Cas9-directed gene knockout (Fig. S7A,B). Homozygous single mutants of both Sox1 genes were viable and produced offspring with apparent normal spinal cord patterning. We therefore bred double-homozygous mutant embryos. Double mutants failed to inflate their swim bladders (Fig. S7C,D) and died during early larval stages. sox1a^−/−;sox1b^−/− mutant embryos showed differences in expression patterns of V2b-specific genes (Fig. 6A-D″). Counts of gata2a⁺ cells in whole spinal cords showed an increase of 22% in sox1a^−/−;sox1b^−/− mutants (Fig. 6A-A″). The numbers of cells expressing tal1, tal2 and gata3 in the V2 domain at 24 hpf increased by 83%, 75% and 35%, respectively (Fig. 6B-D″). One interpretation of this increase in V2b cells is that V2s cells fail to form, thereby increasing the number of V2b,s precursors differentiating into tal1⁺, tal2⁺ and gata3⁺ V2b cells. In agreement, the sox1a^−/−;sox1b^−/− mutants show a 2.3-fold increase in the number of gata3-expressing cells (Fig. 6F-F″) and a decrease of nkx1.2lb⁺, slc6a5⁺ co-expressing cells by 22% (Fig. 6G-I) in the V2 domain at 30 hpf. The expression of the V2a marker vsx2 was unchanged in the mutants (Fig. 6E-E‴), suggesting that sox1a and sox1b only affect the fate decision between V2b and V2s.

We next tested whether injection of morpholinos directed against sox1a and sox1b would result in the same effects. The two single exon genes are so similar that it is impossible to achieve gene-specific effects. Knockdown of sox1a/b translation resulted in a significant increase of tal2-, gad1b-, gata2a- and gata3-expressing cells in the V2 domain at 24 hpf (Fig. S8A-D″). Thus, overall the morpholinos reproduced the effect seen in the double mutants. The only exception was a significant increase of gad1b expression in the morphants at 24 hpf (Fig. S8B-B″), which we could not measure in the double mutants at the similar stage (data not shown). In addition, the number of vsx2-expressing V2a cells was unaffected (Fig. S8E-E″). This agrees with the conclusion that sox1a/b is involved in the V2b/V2s fate decision. This dependence on sox1 activity is reminiscent of V2c cells of the mouse spinal cord.

To verify that lack of sox1a and sox1b leads to loss of V2s cells, we knocked down Sox1a and Sox1b protein translation by injecting the morpholinos, and asked whether the numbers of sox1a or sox1b mRNA-expressing cells were reduced. A sixfold reduction in the number of sox1a⁺;gad1b⁻ cells was noted (P<0.006; n>8 animals for control and knockdown). At the same time, a twofold increase in the number of sox1a⁺;gad1b⁺ cells was counted (P<0.005, n=9 animals for control and knockdown). A 3.2-fold increase was noted for sox1a⁺;gata3⁺ cells (P=0.001, n>8 animals for control and knockdown) and a 2.4-fold increase in the number of sox1b⁺;gad1b⁺ double-labelled cells (P=0.004, n>10 animals for control and knockdown) in the sox1a and sox1b morphants. Thus, precursors develop into V2b cells (gata3⁺, gad1b⁺) in the absence of sox1a and sox1b function. This is consistent with the notion that sox1a and sox1b specifies V2s cell fate from a common V2b,s precursor that follows the V2b fate as default in the absence of sox1a and sox1b activity.

Given the predominant expression of sox1a and sox1b in the central nervous system, we assessed whether sox1a and sox1b double mutants show aberrant movement behaviour. We recorded spontaneous movement every 2 h between 96 and 108 hpf. By using an automated imaging system, minimum disturbance of embryos during the period of recording was assured. Double mutants moved less frequently than wild-type or heterozygous siblings (Fig. 7A). They tend to lie at the bottom of the Petri dish for longer periods than their wild-type or heterozygous siblings.

sox1a and sox1b genes are not only expressed in multiple neuronal subtypes in the spinal cord but also in the brain (Fig. 1, Fig. S1). The functional relevance of V2s neurons may therefore not be detectable in a phenotypic analysis of double mutants. To address more directly the role of V2s neurons, we ablated the V2s neurons on one side of the spinal cord of wild-type embryos. We used the sox1a:eGFP line to specifically identify the V2s neurons at 2 days of development. All sox1a:eGFP-expressing V2s neurons on one side of the spinal cord were ablated using a two-photon laser. These treated embryos, as well as mock-treated embryos, were subjected to analysis of swimming movement upon touch stimulation using a high-speed camera. The overall swimming movements were indistinguishable between ablated, mock-ablated and untreated embryos (data not shown). However, ablated embryos showed a significant delay in response to touch relative to mock-treated embryos (Fig. 7B). This suggests that V2s neurons are required for a fast escape reaction to touch-evoked stimuli.

We report here the identification of a novel type of V2 interneuron in the spinal cord of the zebrafish embryo. This V2s interneuron depends on sox1a and sox1b, has a unique pattern of axonal projections and is glycinergic. It differentiates from a common pool of V2 progenitors during the second day of development. Moreover, ablation of V2s interneurons delays the touch-provoked escape reaction, suggesting that the neurons are required for a fast motor response.

The two Sox1 genes are expressed in overlapping patterns in the spinal cord, including neurons in the V2 domain. At around 24 hpf, precursors share expression of sox1a, sox1b and V2b-specific markers. During the second day of development, however, cells acquire the glycinergic V2s fate characterized by expression of slc6a5 and loss of the GABAergic neuron marker gad1b. At the same time, the number of V2b neurons that express only gad1b and other V2b markers, such as tal1 and gata3, but not sox1a/b increases. Hence, the V2s and the V2b neurons start to differentiate into their terminal neuronal subtypes during the second day. Clearly at 24 hpf, a large proportion of gata2a⁺ and gad1b⁺ cells express sox1a and slc6a5, the markers of V2s cells. Knockdown of gata2a and gata3 abolished differentiation of both V2b and V2s interneuronal types. These observations are in agreement with the recent analysis of gata2a/gata3 double mutants (Andrzejczuk et al., 2018). Thus, a common gata2a/gata3-dependent precursor referred to as V2b,s expresses a chimeric gene programme that is representative of both V2 subtypes. Lack of sox1a and sox1b function leads to an increase in the number of V2b cells and a loss of V2s cells. This suggests that sox1a and sox1b are required to shift a proportion of cells from the initial chimeric precursor V2b,s state (gad67⁺; gata3⁺; gata2a⁺; sox1a/b⁺; slc6a5⁺) to a V2s cell state (gad67⁻; gata3⁻; gata2a⁻; sox1a/b⁺; slc6a5⁺). In the absence of sox1a and sox1b, more precursors develop into V2b cells (gad67⁺; gata3⁺; gata2a⁺; sox1a/b⁻; slc6a5⁻). Markers for V2a neurons (vsx1⁺ and vsx2⁺) were unaffected in sox1 mutants/morphants, showing that sox1a and sox1b are not required for differentiation of V2a interneurons.

Differentiation of V2s from the precursor pools requires Notch signalling. Our inhibition experiments suggest a differential temporal requirement of Notch signalling for specification of the different V2 domain interneurons. Conditional inhibition of Notch signalling between 16 and 24 hpf reduced the number of V2s cells, while V2b-expressing cells were increased. Inhibition of Notch signalling before 16 hpf resulted in a shift towards V2a differentiation with loss of V2b cell fate (Batista et al., 2008), whereas forced activation of Notch signalling shifted cells towards a V2b fate (Kimura et al., 2008). V2s cells are predominantly born between 16 and 24 hpf. However, the differentiation from the chimeric V2b,s precursor state into the distinct V2b and V2s subtypes appears to require most of the second day of development. Our data are consistent with the notion that Notch-dependent cell fate specifications are required at least twice consecutively in the V2 domain: first, in the decision of the V2a/V2b precursor cell (Batista et al., 2008; Kimura et al., 2008) and then in that of the V2b,s precursor (Fig. 8). At 22 to 26 hpf, many V2b,s precursors express traces of vsx1:GFP, suggesting that these precursors are derived from the vsx1:GFP-expressing V2a/b progenitors (Kimura et al., 2008). At least some of the primary progenitors may therefore be referred to as V2a/b,s progenitors. Symmetric cell fate decisions were also noted for some of the primary V2a progenitors (Kimura et al., 2008). A key question is: are V2b and V2s cells generated by division of the V2b,s cell? Our time lapse analysis of sox1a:eGFP-expressing V2 cells suggests that cells marked by the transgene did not divide. We therefore favour a model where V2s and V2b cells arise post-mitotically from the V2b,s precursor without cell division (Fig. 8). However, eGFP expression slightly lags behind endogenous sox1a mRNA expression. Thus, we cannot totally exclude the possibility that V2b/V2s differentiation is initiated by division of the progenitor. To address these issues, sophisticated cell-tracing experiments will be required.

The zebrafish spinal cord shares many similarities with the organization of the mouse spinal cord, including aspects of the underlying mechanisms of cell specification (Jessell, 2000; Lewis, 2006). The developmental path leading to differentiation of V2s interneurons has remarkable resemblance to the pathway outlined for the development of V2c interneurons in the mouse (Panayi et al., 2010). Like murine V2c neurons (Panayi et al., 2010), zebrafish V2s interneurons require sox1 activity for differentiation. V2b and V2c interneurons are derived from common precursors in the mouse. Moreover, the V2c progenitors express transiently gata3 in both organisms. In contrast to mouse (Panayi et al., 2010), however, both gata3 and gata2a are required for expression of the V2s marker sox1a in the zebrafish spinal cord. V2s cells are glycinergic. This trait is likely to be also conserved as some Sox1⁺ cells expressed glycinergic markers at brachial levels in the E12.5 mouse spinal cord. V2s cells could thus be homologues of V2c cells of the murine spinal cord. However, functions and axonal projections of murine V2c cells have not been characterized. Although the expression domains of sox1a and sox1b largely overlap, there is a proportion of cells in the V2 domain that expresses only sox1a or sox1b. In addition, in the mouse spinal cord not all Sox1⁺ cells are glycinergic (V.R., unpublished). The possibility cannot be excluded that more V2 interneuron subtypes exist in the spinal cord of fish and mice (e.g. murine V2d; Francius et al., 2016). It is currently also not possible to unequivocally answer the question of whether V2s cells are the zebrafish homologues of mouse V2c interneurons, even though some similarities of the developmental path leading to differentiation of V2c and V2s cells in the two organisms may suggest so. Clearly, sox1 is co-expressed with gad1 and slc6a5 in a large proportion immediately after birth of sox1⁺ cells in the zebrafish, whereas post-mitotic Sox1⁺ cells in the murine E11.5 spinal cord do not seem to express glycinergic markers until E12.5 in anterior brachial levels. Thus, the mechanism of differentiation is either different between zebrafish and mice or we are looking at non-homologous V2 IN cell types.

Concomitant with the development of a V2s-specific gene expression programme by loss of V2b-specific gene expressions, axons grow out from V2s interneurons during the second day of development, forming stereotypic patterns of projections. A ventrally growing axon splits into a short side branch before growing ipsilaterally and caudally over the next five to six somites. This long descending axon gradually rises from a ventral to a mid-spinal cord level. This pattern of axonal projection resembles that of V2a (CiD) interneurons, which sometimes also develop a short ventral branch (Kimura et al., 2006). However, in contrast to V2a cells, which develop an ascending axon branch (Kimura et al., 2006), we never observed ascending branches in V2s cells. Clearly, the V2s axonal projections are also different from those of V2b (VeLD) neurons (Batista et al., 2008; Bernhardt et al., 1992). V2s neurons are glycinergic and therefore inhibitory, like the GABAergic V2b neuron, unlike the glutamatergic V2a interneurons. The axonal connections of V2s cells remain to be determined in detail.

sox1a and sox1b double mutants are significantly less active in unprovoked swimming tests than wild-type embryos. Double mutants fail to inflate the swim bladder, which may be an indirect consequence of the reduced motility, as immotile mutants fail to gulp air from the surface to fill their swim bladder (Nicolson et al., 1998; Schoppik et al., 2017). sox1a and sox1b genes are expressed in several different structures (telencephalon, cerebellum, spinal cord, lateral line, etc.) in the nervous system. It is therefore likely that the reduced motility is not only due to impaired activity of the spinal cord but also to defects in higher order brain structures. This notion is supported by our ablation experiments, where we ablated the V2s neurons unilaterally in the spinal cord. These embryos show normal unprovoked swimming behaviour. However, upon touch, the V2s ablated larvae show significantly longer delays of escape movements in comparison with mock-ablated controls. Thus, V2s neurons are required in a circuit for efficient touch-evoked escape.

Wild-type zebrafish were obtained from the ABO line, an inbred line initially derived from an intercross between the AB and OX lines (European Zebrafish Resource Centre). Wild-type and transgenic zebrafish [Tg(olig2:eGFP)^VU12 (Shin et al., 2003), Tg(dbx1b:eGFP) and sox1a^{Tg(dmrt3a-gata2a:eGFP)ka705} (this study)] were maintained on a 14 h/10 h light-dark cycle at 28.5°C in a recirculation system (Schwarz) and fed commercial food and in-house hatched brine shrimp as described previously (Westerfield, 2000). Embryos were cultured in embryo medium and staged according to Kimmel et al. (1995).

sox1a^ka701 and sox1b^ka702 mutant alleles were generated using CRISPR/Cas9 genome editing (Jao et al., 2013) by co-injection of sox1a guide RNA (target sequence GGGGCAAACGGGTCCAAATT) and sox1b guide RNA (target sequence GGAGTGGAAACTCATGTCCG) with Cas9 mRNA into one-cell stage embryos. A G1 founder with a 7 bp deletion in the sox1a gene 72 bp downstream of the ATG and a 2 bp indel in the sox1b gene 233 bp downstream of the ATG was identified by fin clipping (Fig. S7A,B) and sequencing of the subcloned PCR product amplified from the target region of the guide RNA. This results in frameshifts upstream the HMG and the SOX domain (interaction site for partner proteins), suggesting that these alleles are likely to cause a complete loss of function (Fig. S7A,B). The G1 founder with both mutations in sox1a and sox1b was outcrossed with wild-type fish and genotyped (Etard et al., 2017). G2 fish carrying sox1a^ka701 and sox1b^ka702 alleles were incrossed and sox1a^ka701/ka701;sox1b^ka702/ka702 double mutants were identified by genotyping.

DNA was isolated from anesthetized adults via fin biopsy or from fixed embryos via dissection of the head. Fin biopsy of adults was performed as previously described (Etard et al., 2017) and genomic DNA was extracted using the HotSHOT method using 100 µl of 50 mM NaOH and 10 µl Tris-HCl (pH 7.5). Heads of fixed embryos were dissected in PTW (PBS, 0.1% Tween 20) with pins, and embryo trunks were stored in PTW at 4°C for later analysis. The sox1a^ka701 and sox1b^ka702 alleles were identified by restriction enzyme digestion, as both mutations disrupt endogenous restriction enzyme sites. Genomic DNA extraction from dissected heads was performed as described previously (Juárez-Morales et al., 2016) or by directly inserting parts of dissected heads into the PCR reaction mix. The genomic region encompassing the mutation site was PCR amplified using the following conditions: initial denaturation step at 94°C for 7 min; 35 cycles of 94°C for 30 s, 57°C for 30 s and 72°C for 30 s; and a final elongation step at 72°C for 2 min. For sox1a, a 349 bp amplicon encompassing the mutation site was generated using the following primers: forward, TTCCACACTTCATCGGAGCT; reverse, TGCTGAGTGGAAGGTGATGT. This amplicon was digested with ApoI-HF to yield fragments of 191 bp and 158 bp (wild-type allele) or 342 bp (mutant allele). For sox1b, a 346 bp amplicon encompassing the mutation site was generated using the following primers: forward, AGAGACCCATGAACGCCTTT; reverse, GGCCAGAGGTTAGAGAGTCC. This amplicon was digested with NlaIII to yield 168 bp, 111 bp and 54 bp fragments (wild-type allele) or 168 bp and 166 bp fragments (mutant allele).

Whole-mount in situ hybridization was performed as described previously (Yang et al., 2010). Antisense RNA probes for tal2 (Pinheiro et al., 2004), gata2a (Detrich et al., 1995), gad1b (previously known as gad67, a mix of GAD67a and GAD67b probes), slc6a5 (previously known as glyt2, a mix of glyt2a and glyt2b probes) (Higashijima et al., 2004), gata3 (Yang et al., 2010), vsx2 (Passini et al., 1997), and sox1a and sox1b (Armant et al., 2013) (Table S2) were labelled with digoxigenin (Roche) or DNP-11-UTP (PerkinElmer).

In double fluorescent in situ hybridizations, the RNA probes were detected using anti-digoxigenin-POD (Roche, 1:250) and anti-DNP-HRP (PerkinElmer, 1:200). The DNP-probe was stained first with either TSA Plus Cyanine 5 Tyramide Reagent (PerkinElmer) or Alexa Fluor 488 Tyramide Reagent (Invitrogen) as a substrate following the manufacturers’ instructions. For the second staining, the embryos were incubated in 1% hydrogen peroxide for 1 h to quench horseradish peroxidase followed by washes in 1× PBS, 1% Tween-20. The second staining (Cyanine 3) was developed following the manufacturer's instructions for the TSA Plus Cyanine 3 (Cy3) kit (PerkinElmer).

Chicken anti-GFP (1:1000, Aves Labs, GFP-1020) or rabbit anti-GAD65+GAD67 (anti-Gad1b/Gad2, 1:500, Abcam, ab11070) antibodies were added to embryos with blocking buffer (PerkinElmer blocking buffer of the TSA Plus Cy3 kit) overnight at 4°C followed by incubation with goat anti-chicken or anti-rabbit antibody conjugated with Alexa Fluor 488 or Alexa Fluor 647 (1:1000, Invitrogen). The yolk sac of stained embryos was removed manually, and the embryos were embedded laterally in Aqua-Poly/Mount (Polysciences) followed by analysis with a TCS SP5 confocal microscope (Leica).

For sectioning, stained embryos were embedded in 3% agarose and 25 µm sections were cut using a vibratome (Leica). Sections and whole embryos (without yolk sac) were mounted in 100% glycerol and imaged under a compound microscope (DM5000B, Leica) with DIC optics. The intensity, contrast and exposure time of the pictures were adjusted using Photoshop and ImageJ software.

Counts of gad1b-, gata2a-, gata3-, sox1a-, sox1b-, tal2- and vsx2-positive cells in whole-mount embryos were derived from the entire trunk or from the first somite (V2 interneuron domain) to the tail on both sides of the spinal cord (Fig. 5, Fig. S8). Counts of gata2a-positive cells in the sox1a^−/−; sox1b^−/− mutant and wild type were derived from the fourth somite to the tail on both sides of the spinal cord. Counts of tal1-, tal2-, gata3- and vsx2-positive cells in the nkx1.2lb; scl6a5 double FISH experiment analysing the sox1a^−/−; sox1b^−/− mutant and wild type were derived from both sides of the spinal cord above the yolk extension over a five-somite distance (Fig. 6).

Mouse embryo fixation, embedding, cryo-sectioning and immunolabelling protocols have been previously described (Briscoe et al., 2000). For further details, see the supplementary Materials and Methods

Injections were performed using a gas-driven microinjector (Tritech Research) as described previously (Muller et al., 1999). The antisense morpholino directed against sox1a and sox1b (5′-CCGTTTCCATCATCATGCTATACAT-3′) was designed by GeneTools. Morpholinos containing five mismatches (mismatches indicated by small letters) were used as controls: 5′-CCcTTTgCATgATgATGgTATACAT-3′. The efficiency of morpholinos was examined by knockdown of GFP expressed from a reporter construct (data not shown). Other morpholinos used in this study have been described previously (Yang et al., 2010). The morpholinos were resuspended in water with 0.1% Phenol Red to 0.25 or 0.5 mM and were injected into the yolk of one- to two-cell stage embryos. For single gene knockdown experiments, morpholinos were injected at a concentration of 0.5 mM. In double knockdown experiments, individual morpholinos were used at a concentration of 0.5 mM and 0.25 mM, respectively. For controls, the same concentrations of the respective five mismatch control morpholinos were injected.

The zebrafish bacterial artificial chromosome (BAC) DNA for dbx1b (CH211-7E10) and for sox1a (CH211-207-L5) was purchased from Children's Hospital Oakland Research Institute. The BAC modification strategy for transgenesis was adapted from Shin et al. (2003) and Lam et al. (2009). In the first homologous recombination, the first exon of dbx1b and sox1a was replaced with eGFP. In the second homologous recombination, an ampicillin resistance gene flanked by Tol2 sites was introduced into the modified BAC, as previously described (Suster et al., 2011). The modification of BAC DNA was confirmed by PCR, and bacteria colonies positive for eGFP and Tol2 arms were selected for culture and further BAC DNA extraction using a Nucleobond AX-100 maxi-prep kit (Macherey-Nagel). The modified BACs were injected into one-cell stage embryos at 10 ng/µl. Embryos expressing GFP at 24 h post fertilization (hpf) were raised to maturity. The stable transgenic lines (F0) were identified by out-crossing injected fish with wild-type fish. Identified F0 were out-crossed with wild-type fish to obtain stable progenies that express the transgene in the F1 generation. We obtained a stable line for Tg(dbx1b:eGFP). TgBAC(sox1a:eGFP) was used only in transient expression experiments.

sox1a^{Tg(dmrt3a-gata2a:EGFP)ka705} (sox1a:eGFP for short) is a gene trap allele in which a Tol2-based Gateway destination enhancer test vector (Navratilova et al., 2009) was inserted fortuitously into the 5′ untranslated region of the sox1a gene (Fig. S3B). The integration was verified by whole-genome sequencing with an Illumina Hiseq1500 at a coverage of 7.1× (see supplementary Materials and Methods).

sox1a:eGFP transgenic embryos were injected at the one-cell stage with a guide RNA (final concentration: ∼100 ng/µl) against eGFP (Auer et al., 2014) and Cas9 protein (Invitrogen, final concentration: ∼300 ng/µl), resulting in embryos with mosaic eGFP expression that were raised under normal conditions until imaging.

LY411575 (Fauq et al., 2007) was reconstituted with dimethyl sulfoxide (DMSO) to a stock concentration of 25 mM. Embryos raised in embryo medium were dechorionated manually and placed in agarose-coated Petri dishes. The embryos were incubated in embryo medium containing 10 µM LY411575, 0.04% DMSO or 0.04% DMSO alone from 16 to 24 hpf, then washed with embryo medium containing 0.04% DMSO and raised until 30 hpf.

For long-term in vivo time lapse imaging (more than 16 h), sox1a:eGFP embryos were embedded as previously described (Middel et al., 2016) in LMP agarose (0.5%) with agarose removed from the tail (for elongation) in a 6 cm Petri dish and covered with 10 ml 1× E3 medium (60× stock: 5 mM NaCl, 0.17 mM KCl, 10 mM HEPES, 0.33 mM MgSO₄7H₂O and 0.33 mM CaCl₂6H₂O) containing 0.02% MESAB. Multiple positions of the lateral spinal cords were imaged under a Leica TCS SP5 upright microscope (HCX APO L 63×/1.2 W U-V-I objectives) using the bright field and 488 nm channel. Z-stacks covering 40-90 μm were acquired every 0.8 μm at each position (two positions per spinal cord). Each position was imaged every 10 min for 16 h in 512×512 pixel format with a depth of 8-bit by bidirectional resonant scanning at 8 kHz. Images with an extended depth of field were generated in Fiji (Schindelin et al., 2012) through maximum projections for the temporal analysis of the fluorescence.

For ablation of V2s neurons, sox1a:eGFP embryos were incubated in 0.003% phenylthiourea (PTU) containing E3 medium from 24 hpf onwards and were laterally embedded at 2 dpf in LMP agarose (0.5%) as previously described (Middel et al., 2016). Using a confocal microscope (TCS SP2, Leica), single V2s neurons on one side of the spinal cord were damaged at 32× zoom using a Ti:Sa laser (MaiTai, Spectra Physics) set to 864 nm (gain 65%, offset 26%) using the region-of-interest (ROI) set to 0.25-2 µm. Control and V2s-ablated embryos were removed from the agarose and incubated in E3 medium for recovery. At 4 dpf, single embryos were transferred to 6 cm Petri dishes and a touch evoked escape response was stimulated with a pin and recorded with a high-speed camera.

We thank N. Borel, her team and M. Rastegar for technical help; and T. Dickmeis for critical comments on the manuscript. We deeply thank C. Wyart for antibodies.