Zebrafish colourless encodes sox10 and specifies non-ectomesenchymal neural crest fates

The neural crest is a vertebrate tissue of developmental and medical importance. Developmentally, neural crest is intriguing because the cells are initially multipotent and subsequently form a great diversity of derivative cell types, including ectomesenchymal fates, such as craniofacial skeleton and fin mesenchyme, and non-ectomesenchymal fates, such as neurones, glia and pigment cells (Le Douarin, 1982; Smith et al., 1994). Medically, neural crest is important because some diseases, known as neurocristopathies and including diverse conditions such as albinism, neurofibromatosis and Hirschsprung’s disease (aganglionic megacolon), affect cell types derived from this tissue (Bolande, 1974).

Understanding the genetic and embryological basis of neurocristopathies has depended on animal models. Thus, models for Hirschsprung’s disease, in which individuals have few or no enteric ganglia in the colon, or the related Waardenburg-Shah syndrome, in which individuals combine Hirschsprung’s disease with pigmentary anomalies of the skin, hair and irises, have been described in several species, including mouse and zebrafish (Hosoda et al., 1994; Kelsh and Eisen, 2000). Analysis of such models in mice has identified three loci that are crucial for Waardenburg-Shah syndrome (Attié et al., 1995; Edery et al., 1996; Hofstra et al., 1996; Pingault et al., 1998; Puffenberger et al., 1994; Southard-Smith et al., 1999). Thus, mutations in loci encoding the G-protein-coupled transmembrane receptor protein endothelin receptor B (Ednrb) or its natural ligand endothelin 3 (Edn3) result in aganglionosis of terminal gut in homozygous mutants (Baynash et al., 1994; Hosoda et al., 1994), as do heterozygous mutations in the Sry-related transcription factor gene Sox10 (Herbarth et al., 1998; Southard-Smith et al., 1998). Homozygous Sox10 mutant animals show a more severe phenotype with aganglionosis of the whole gut (Herbarth et al., 1998; Southard-Smith et al., 1998). Additionally, mutations in all these genes affect body pigmentation (Lane and Liu, 1984; Mayer, 1965; Mayer and Maltby, 1964), but only Sox10 mutations result in widespread peripheral nervous system defects (Herbarth et al., 1998; Kapur, 1999; Southard-Smith et al., 1998).

The Sox gene family encodes a large family of transcription factors, with vertebrates likely to have more than 20 Sox genes each (Wegner, 1999). Their precise roles are not well understood, although many are presumed to function in cell fate specification (Pevny and Lovell-Badge, 1997). For example, the founding family member, Sry, is likely to be responsible for Sertoli cell specification, and thus male sex determination, in mammals. While Sox10 is clearly an important transcriptional regulator in neural crest cell (NCC) development, the cellular basis of the Sox10 mutant phenotype remains unclear. It has been suggested that peripheral nervous system and pigmentation defects result from loss of NCCs (Southard-Smith et al., 1998; Kapur 1999), although the developmental status of these cells at the time of loss is unknown. Furthermore, roles in defining regional identity in the cranial neural crest and in glial cell differentiation have also been proposed (Bondurand et al., 1998; Herbarth et al., 1998; Kuhlbrodt et al., 1998a; Pusch et al., 1998; Southard-Smith et al., 1998; Britsch et al., 2001).

Mutations at the colourless (cls) locus have been identified in zebrafish mutagenesis screens (Kelsh et al., 1996; Malicki et al., 1996). We have previously characterised the crest derivative defect displayed by cls mutants (Kelsh et al., 2000a; Kelsh and Eisen, 2000), noting extensive loss of pigment cells and enteric nervous system, together with large reductions in sensory and sympathetic neurones and putative satellite glia and Schwann cells. By contrast, we found little effect on ectomesenchymal derivatives, craniofacial skeleton and fin mesenchyme. Based on the severity and details of the phenotype, we proposed that cls functions in specification, proliferation or survival of a progenitor(s) for all non-ectomesenchymal crest derivatives.

The cls phenotype, and the cell-autonomy of cls gene action in pigment cell types (Kelsh and Eisen, 2000), suggested sox10 as a candidate gene. We provide an experimental test of this hypothesis. We report the mapping of the cls locus and cloning of a zebrafish sox10 homologue. We show linkage between cls and sox10, identify sox10 lesions in four mutant alleles and show rescue of the cls phenotype by sox10 expression. In addition, we describe iontophoretic labelling experiments to examine the precise cell-biological role of cls/sox10 gene function in neural crest development. We show in live embryos that NCC clones in cls and wild-type embryos differentiated into ectomesenchymal fates after migration to appropriate sites. Remaining NCC clones adopted non-ectomesenchymal fates in wild-type embryos. By contrast, in cls embryos differentiation to non-ectomesenchymal fates was rarely observed. Instead, most clones failed to migrate and underwent late cell death by an apoptotic mechanism. Finally, for the melanophore fate, we show disrupted expression in cls mutants of genes vital for melanophore specification and migration. Together, these data demonstrate a complex phenotype in cls embryos that can be explained by proposing that cls/sox10 has a primary role in specification of non-ectomesenchymal fates. Defects in cell migration, survival and differentiation are therefore likely to be secondary consequences of an inability of these cells to adopt specific fates.

Embryos were obtained through natural crosses and staged according to Kimmel et al. (Kimmel et al., 1995). We used 4 cls alleles (m618, t3, tw2 and tw11).

cls^tw11 heterozygous fish (Tübingen background), were crossed to wild-type strain WIK11 to produce a reference mapping cross. Heterozygous F₁ were incrossed and separate pools of F₂ homozygous cls fish and their wild-type siblings were used for simple sequence length polymorphism analysis (Knapik et al., 1996). Linkages from the pools were confirmed and refined by genotyping individual embryos, as described by Rauch et al. (Rauch et al., 1997).

RT-PCR was performed using total RNA of 19 hpf stage wild-type embryos using published conditions and degenerate primers (Yuan et al., 1995). Sequencing of resulting clones identified a sox10-like sequence. The sox10 clone was extended by RACE PCR using gene-specific primers (Clontech, SMART RACE kit) and sequenced on an ABI DNA sequencer. All primer sequences available on request. The full sox10 cDNA sequence is available in Genbank (Accession Number AF402677. The zebrafish sox10 homologue was mapped on the radiation hybrid panel LN54 (Hukriede et al., 1999) by PCR with primers 5′-ACCGTGACACACTCTACCAAGATGACC-3′ and 5′-CATGATAAAATTTGCACCCTGAAAAGG-3′, which generate a 931bp 3′ UTR fragment.

For phylogenetic analysis sequences were extracted from Genbank and coding sequences automatically extracted using Genetrans (within GCG9). These were translated and then aligned using ClustalX (Thompson et al., 1997). The nucleotide alignments were reconstructed from the protein alignments using MRTRANS (www.hgmp.mrc.ac.uk/Registered/Option/mrtrans.html). Tree-Puzzle (v 4.0.2) was used to construct an unrooted tree by maximum likelihood (Strimmer and von Haeseler, 1996). It automatically assigns estimations of support to each internal branch, figures for which are presented in the figure. To model the substitution process the Tamura and Nei (Tamura and Nei, 1993) model was employed and all sites were used. Gamma distributed variation in rates of evolution was permitted with eight variable sites and one invariable. Parameters were estimated using quartet sampling and the neighbour-joining tree.

Total RNA from 27 hpf homozygous embryos of mutant alleles cls^m618, cls^tw2 and cls^tw11 was prepared using TRI reagent (Sigma). First strand cDNA was generated using random hexamers and SuperscriptII RT (GibcoBRL). For each allele four overlapping RT-PCR fragments were sequenced to identify mutant lesions. For t3, genomic DNA was extracted from individual cls^t3 mutant embryos, a genomic fragment encoding the N-terminal of the Sox10 protein was amplified by PCR and sequenced commercially (Oswell, Southampton). The t3 insertion sequence is available in Genbank (Accession Number, AF404490).

The full coding region of cls^m618 was amplified by RT-PCR using primers ClaI-S21 (5′-CCATCGATACCTACCGAAGTCACCTGTGG-3′) and S27-XbaI (5′-GCTCTAGAGTTTGTGTCGATTGTGGTGC-3′). The 1615bp fragment was subcloned into the ClaI/XbaI site of the heatshock vector pCSHSP (Halloran et al., 2000) to generate hs>sox10(L142Q). The wild type sox10 construct, hs>sox10, was generated by site-directed mutagenesis of hs>sox10(L142Q) using QuikChange Site Directed Mutagenesis Kit (Stratagene). Sequencing confirmed the successful generation of both clones. DNA purified for injection using Microcon Filter Devices (Millipore) was diluted to a concentration of 25 ng/μl, with 0.1% Phenol Red. cls^m618 or their wild-type siblings were injected with 2 nl of either hs>sox10 or hs>sox10(L142Q) at the one- to two-cell stage and incubated at 28.5°C. As appropriate, embryos were heatshocked twice (at 10-13 hpf and 22-24 hpf), by incubation at 37°C for 1 hour. cls embryos were scored for rescue at 48 hpf using a MZ12 dissecting microscope (Leica). Rescue was defined as the presence of at least one melanophore of wild-type morphology; these are never seen in uninjected mutant embryos.

Iontophoretic labelling of individual neural crest cells was performed using essentially the method of Raible et al. (Raible et al., 1992), except that micropipette tips were filled with 3% lysinated rhodamine dextran and 3% biotinylated dextran (both 10×10³M_r; Molecular Probes) mixture dissolved in 0.2 M KCl and the needles backfilled with 0.2 M KCl. Embryos from heterozygotes for cls alleles t3, tw2 and tw11 were used; no phenotypic differences between these alleles were seen, so we do not distinguish them here. Individual NCCs were labelled by intracellular injection with dye. Premigratory cranial crest cells, between the posterior eye and the anterior boundary of somite 1, were labelled in 4-14 somite (11-16 hours post fertilisation, hpf) embryos, while premigratory trunk NCCs at the level of somite 7 were labelled in 16-22 somite (18-20 hpf) embryos. Premigratory trunk NCCs at the level of somite 14 were labelled in 22-25 somite (20-22 hpf) embryos. Embryos were recovered and raised for several hours at 28.5°C in embryo medium with 1% v/v penicillin/streptomycin solution (Gibco). Embryos were remounted and examined to identify those with only a single labelled NCC; only these embryos were analysed.

Labelled cells were monitored using Nomarski and fluorescence optics on a BX50WI microscope (Olympus), and documented on a Eclipse E800 microscope (Nikon) using low light level, video-enhanced fluorescence microscopy. Labelled cells were monitored twice daily for up to 3 days until progeny could be identified using published morphological criteria (Raible and Eisen, 1994; Raible et al., 1992; Schilling and Kimmel, 1994). Thus, melanophores contained melanin granules; xanthophores were visibly yellow and autofluoresced at a wavelength near that of fluorescein; iridophores contained iridescent granules; dorsal root ganglial neurones were identified by position ventrolateral to the neural tube combined with a visible neurite; sympathetic neurones were positioned ventral and lateral to the notochord and showed neurites; Schwann cells were elongated and positioned along axonal processes, e.g. trigeminal nerve; satellite glia were associated with the ganglionic sheath or appeared to be wrapped around neuronal stomata; craniofacial cartilage formed characteristic stacks of vacuolated cells in the jaw or gill arches; fin mesenchyme occupied a position within the dorsal fin fold and showed characteristic asymmetric organisation of projections (Smith et al., 1994). Cells which could not be assigned to any of the described groups, owing to their position in regions of limited optical resolution or lack of distinctive morphologies, were classified as ‘unidentified’. In cranial regions, these cells undoubtedly included cells of connective tissue fates (Schilling and Kimmel, 1994).

TUNEL (terminal deoxynucleotidyl transferase (TdT)-mediated deoxyuridinetriphosphate (dUTP) nick end-labelling) of double-strand DNA fragmentation was used to confirm apoptosis in cells with apoptotic morphology. cls^t3 and their wild-type siblings were fixed overnight at 4°C in 4% paraformaldehyde and TUNEL performed using fluorescein dUTP and developed using 4-Nitroblue tetrazolium chloride and 5-Bromo-4-chloro-3-indolyl-phosphate (Boehringer Mannheim; Reyes, 1999).

AB wild-type embryos at 25, 30 and 35 hpf were injected with either a high (16.5 ng) or low (9 ng) dose of a morpholino designed to knock-down sox10, as described previously (Dutton et al., 2001). Effects on nac/mitf expression in melanoblasts at 25 hpf were evaluated by counting nac/mitf-expressing cells in one half of the trunk in each of 20 embryos at each dose.

RNA in situ hybridisation was performed as described by Kelsh and Eisen (Kelsh and Eisen, 2000), on cls^m618, cls^tw11 and cls^t3, their wild-type siblings and morphants. Probes for the following genes were used: nac/mitf (Lister et al., 1999), spa/kit (Parichy et al., 1999); dopachrome tautomerase (dct) (Kelsh et al., 2000b); dlx2 (Akimenko et al., 1994); forkhead 6 (fkd6) (Odenthal and Nuesslein-Volhard, 1998).

Antibody staining with anti-Hu, mAb 16A11 (Marusich et al., 1994), was performed using peroxidase-antiperoxidase (VECTASTAIN® Elite ABC kit) and DAB substrate.

The strong phenotypic similarity between sox10^D°^m mice and cls mutants suggested a zebrafish sox10 homologue as a candidate gene for cls. We used RT-PCR to clone a partial sox10-like HMG box and RACE RT-PCR to clone 5′ and 3′ regions. Sequencing these clones revealed an open reading frame encoding a sox10 homologue, which we refer to as sox10 (Fig. 1A,B).

Genetic mapping of 274 meioses placed cls on LG 3 within a 3.9 cM interval between markers z872 and z13387 (Fig. 1C). Two oligonucleotide primers amplified a 931 bp fragment from zebrafish, but not from control mouse genomic DNA. Amplification from the DNAs in the LN54 radiation hybrid mapping panel (Hukriede et al., 1999) with these primers mapped sox10 to LG 3, 0cR from the marker z8492 (LOD score=17.6; Fig. 1C). The striking linkage of cls and sox10, together with in situ hybridisation experiments showing sox10 expression in neural crest (see below), strongly supported our hypothesis that cls might encode sox10.

We used RT-PCR to amplify the sox10-coding region from 27 hpf homozygous mutants of 3 cls alleles. Sequencing these PCR products identified sequence differences from wild-type consistent with them causing the mutant phenotype (Fig. 1D,E). Two alleles show an A→T transversion, resulting in a premature Stop codon; the third is a non-conservative substitution (L142Q) of a fully-conserved residue in the HMG domain. A fourth allele, t3, showed highly reduced RNA expression using a 3′ probe for whole-mount RNA in situ hybridisation. PCR from genomic DNA identified a 1.5 kb insertion in all t3 mutants that was not present in wild types. Sequencing genomic DNA from t3 homozygotes identified a 1397 bp insertion with sequence homology to a transposon first identified in a mutant no tail allele (data not shown) (Schulte-Merker et al., 1994). The insertion interrupts the sox10-coding sequence upstream of the HMG domain and adds eight novel amino acids before prematurely truncating the protein (Fig. 1D).

To test further whether cls encodes sox10, we attempted to rescue the cls phenotype with ectopic sox10 expression under heat shock control. We took advantage of the consistent absence of large, stellate melanophores in every cls embryo. Ectopic expression of wild-type sox10 rescued 1-40 melanophores to a wild-type morphology in 48% of cls^m618 mutants, while ectopic mutant sox10(L142Q) failed to do so (Table 1; Fig. 2). Furthermore, while cls melanophores always remain dorsal to the neural tube, rescued melanophores frequently migrated, even to very distal positions (Fig. 2).

We used in situ hybridisation to examine sox10 expression in wild-type and cls embryos. In wild types, expression was first detected at the one-somite stage in cells in the lateral neural plate (data not shown). Throughout somitogenesis stages, strong sox10 expression was seen in premigratory NCCs and extended progressively more caudally in older stages (Fig. 3A-C,F). Double RNA in situ hybridisation studies show that there is extensive, but incomplete, overlap of sox10 and fkd6, a marker expressed widely in premigratory NCCs (Fig. 3O,P) (Odenthal and Nuesslein-Volhard, 1998). sox10 expression was maintained in some migrating NCCs on the medial migration pathway (Fig. 3C,F). By 30-40 hpf sox10 expression on the medial pathway is organised in segmentally arranged clusters adjacent to the notochord, presumably developing Schwann cells associated with the segmental nerves (Fig. 3Q); sox10 expression is lost from this site by 48 hpf. Although NCCs are found extensively on the lateral pathway from 24 hpf (Raible et al., 1992), counts of sox10-expressing cells show that expression was essentially absent from cells on this pathway (Fig. 3M,N). This demonstrates the rapid downregulation of sox10 from pigment cell precursors, as NCCs on this pathway form only pigment cells (Raible and Eisen, 1994). Consistent with this, xanthophores and most pigmented melanophores show no detectable sox10 expression (Fig. 3L), although some weakly expressing melanophores were noted at earlier stages (Fig. 3K). Expression was not seen in fin mesenchyme nor in dlx2-expressing craniofacial cartilage precursors (Fig. 3D,E), although differentiating jaw cartilage shows weak expression by 60 hpf (data not shown). Cells expressing sox10 accumulated in clusters corresponding to the forming cranial ganglia and extending along the posterior lateral line nerve by 24 hpf (Fig. 3F,H), in a pattern reminiscent of fkd6 expression (Kelsh et al., 2000a), suggesting that sox10 is expressed in satellite glial and Schwann cells. Consistent with this interpretation, double labelling with anti-Hu antibody confirmed the non-overlapping expression of sox10 and this neuronal marker (Fig. 3J). sox10 is maintained in developing Schwann cells on the posterior lateral line nerve up to 60 hpf (data not shown). sox10 expression was prominent in enteric nervous system precursors at 60 hpf (Fig. 3S).

At early stages, premigratory crest showed equivalent patterns of sox10-positive cells in wild-type and cls^tw11 and cls^m618 mutant embryos. Counts at 24, 30 and 35 hpf revealed that although in wild type sox10-expressing cells were rapidly lost from the premigratory position, they remained here in cls^m618 mutants; by contrast, sox10 expression in migrating trunk NCC was broadly comparable between wild type and cls^m618 mutants (Fig. 3M,N). Additionally, cls mutants were distinguishable from 24 hpf by the clustered, not scattered, distribution of sox10-positive cells in the head (Fig. 3F,G) and a variable reduction in the number of sox10-positive cells in cranial ganglia and on cranial nerves (Fig. 3H,I). By 35 hpf, cls mutants were readily distinguished from wild types by their reduced number of sox10-positive cells, concentrated in a premigratory position dorsal to the neural tube or clustered near the posterior lateral line ganglia; putative Schwann cells were missing from the posterior lateral line nerve. cls mutants showed reduced expression in the putative Schwann cells found as segmental clusters of sox10-positive cells (Fig. 3Q,R). Enteric nervous system precursors were absent at 60 hpf (Fig. 3T) in cls mutants.

In contrast to the other mutant alleles examined, cls^t3 mutants consistently show highly reduced sox10 transcripts when examined using the 3′ sox10 probe at all stages. When examined using a probe lying 5′ to the insertion site, sox10 expression levels were comparable with those in other mutant alleles, suggesting that insertion of the transposon does not disrupt transcription of 5′ sequences (data not shown).

Outside neural crest, sox10 expression was particularly strong in otic placode and otic vesicle, and from 11-somite stage onwards (Fig. 3C,F,H), was detected in pectoral fin and in some spinal cord cells from 36 hpf. Expression in the ear was significantly weaker in cls mutants by 40 hpf (data not shown).

Previous characterisation of cls embryos catalogued a strong reduction in non-ectomesenchymal neural crest derivatives (Kelsh and Eisen, 2000). To analyse the cell biological basis for loss of these neural crest derivatives, we used iontophoretic labelling of single NCCs. We labelled premigratory NCCs in cls mutants and their wild-type siblings in two regions that generate ectomesenchymal fates, and scored them for the fate(s) adopted by their progeny (Table 2). In wild-type embryos, almost all (148/160; 93%) labelled cells survived throughout the experiment, and all major derivatives, both ectomesenchymal and non-ectomesenchymal, were identified among the clones. Consistent with our previous analyses, labelled cells generated a similar proportion of ectomesenchymal fates in cls mutants (4/58; 7%) and wild-type siblings (11/160; 7%). Furthermore, even in cls mutants, cells in these clones migrated and differentiated normally. In wild-type embryos, most cells where fates were identifiable (111/122; 91%) differentiated into recognisable non-ectomesenchymal derivatives. By contrast, most identifiable clones in cls embryos (45/49; 92%) died late on the second day (∼35-45 hpf). Most showed no sign of morphological differentiation or pigmentation (34/49; 69%). The only exception, clones that developed some xanthophore pigmentation (11/49; 22%), always had abnormal morphology, being rather rounded and blebbed, and died soon after differentiation (Fig. 4N). A series of single cell injections in NCCs at the axial level of somite 7 gave similar results, with 92% of cls cells dying by around 48 hpf, in contrast to only 2% in wild-type siblings (data not shown); note that at this axial position no ectomesenchymal fates have been reported (Raible and Eisen, 1994). Our results are consistent with the severe reduction in differentiated non-ectomesenchymal fates in cls embryos and suggest that extensive NCC death is associated with the cls phenotype.

The morphological appearance of dying cells in our clonal analyses suggested an apoptotic mechanism of death. We combined TUNEL with biotin-detection of the labelled clone in each of five cls embryos and showed that the dying cells had fragmented DNA (Fig. 4A-C). We conclude that many non-ectomesenchymal clones in cls embryos die by an apoptotic mechanism.

Analysis of cell survival in our cranial NCC data set shows that death of labelled NCCs occurs within a relatively discrete time-window (Fig. 4L). In both cls and wild-type embryos, around 10% of labelled clones had died by the morning after injection. In wild-type embryos, the number of surviving clones was essentially unchanged at later times. However, in cls mutants, the number of surviving clones dropped precipitously to only 40% within the period of around 5-10 hours between the two observational time-points on the second day after labelling (equivalent to ∼40 hpf), but then remained constant. We interpret the early (overnight after injection) loss of wild-type and cls clones as representing death due to damage during labelling. The subsequent late loss of clones in cls embryos corresponds to the time of appearance of apoptotic cells and we attribute this to the cls phenotype.

To examine the extent to which apoptosis contributes to the cls phenotype, we used whole-mount TUNEL to examine apoptosis of NCCs (Fig. 4D-K). We saw a notable concentration of apoptotic cells dorsal or dorsolateral to the neural tube in cls embryos, but not in wild-type control embryos. A timecourse of TUNEL between 20 and 60 hpf indicated that cell death in NCCs becomes apparent by 35 hpf and continues to approximately 45 hpf.

Our in vivo clonal studies revealed that, while wild-type NCCs migrated extensively on both medial and lateral pathways, most cls NCCs failed to leave the premigratory crest area. Thus, excluding ectomesenchymal derivatives (which always migrated normally), only 2/31 (6%) of cranial NCCs appeared to migrate away from their initial positions. At the level of somite 14, all wild-type labelled cells left the premigratory area, and more than 75% migrated at least 50 μm from their original position. Only fin mesenchyme clones migrated normally in cls mutants. Of the remaining clones, only 4/23 (17%) migrated away from the premigratory crest; another cell extended towards the horizontal myoseptum, but maintained contact with the premigratory area. Of these five cells, two migrated on the lateral pathway, but underwent apoptosis after moving about half way to the horizontal myoseptum. The other three cells migrated on the medial path. In one case, the cell did not divide, migrated to a position appropriate for a dorsal root ganglion, then died before overt differentiation. The other two cells divided once; in each clone, one sister cell remained dorsal to the neural tube, and may have been an extramedullary cell (Kelsh and Eisen, 2000), while the other migrated to a dorsal root ganglial position, but failed to undergo axonogenesis, consistent with the impaired touch response of 5 dpf larvae. We obtained similar results for labelled NCCs from the region of somite 7 (data not shown). This direct observation of failed NCC migration in cls embryos is consistent with our observation of dct-expressing melanoblasts and sox10-expressing NCCs concentrated dorsal to the neural tube in mutant embryos (Kelsh et al., 2000b) (this work).

cls mutants combine an extensive failure of NCC migration with late apoptotic death of non-ectomesenchymal derivatives. Three recent studies suggested a possible molecular genetic mechanism for these aspects of the phenotype. sparse (spa), a kit homologue, is crucial for survival and migration of melanophores and melanoblasts (Parichy et al., 1999; Kelsh et al., 2000b) and nacre (nac), a microphthalmia transcription factor (mitf) homologue, is crucial for melanophore specification and required for spa expression (Lister et al., 1999). We used in situ hybridisation at 20-35 hpf to ask whether spa or nac expression was disrupted in cls embryos (Fig. 5). We found that both spa and nac expression were absent from NCCs, even at a time when many dct-positive cells are present, suggesting that the cls melanophore phenotype might result from loss of nac expression. By contrast, spa expression in intermediate cell mass and nac expression in the pigmented retinal epithelium are unaffected in cls mutants.

We have recently shown that injection of sox10 morpholino oligonucleotides phenocopies the cls mutant phenotype (Dutton et al., 2001). The morphant phenotype varies depending upon the amount of morpholino injected, with high doses phenocopying all aspects of the phenotype of the strong cls alleles. Injection of lower doses results in a phenocopy of the weak cls allele phenotype, with embryos showing some melanophores, concentrated in dorsal positions (Dutton et al., 2001; Malicki et al., 1996). The number of nac/mitf-expressing cells in embryos injected with a low dose of sox10 morpholino (median number=78; n=20) is significantly higher than in embryos injected with a high dose (median number=20; n=20; Mann-Whitney U-test, P<0.05; Fig. 5C,D), consistent with the differences in melanophore phenotype.

Studies in mice have led to identification of Sox10 as a key gene in human Waardenburg-Shah syndrome. We have shown that the zebrafish cls locus is a sox10 homologue predicted to encode a protein with all the major domains identified in mammals (Fig. 1). We have identified the molecular lesion likely to cause the mutant phenotype in 4 cls alleles. In two cases, tw2 and tw11, both originating from the same mutagenised male (Haffter et al., 1996; Kelsh et al., 1996), the lesions are identical and are presumably independent isolations of the same allele. The zebrafish Lys376Stop lesions resemble the human 059 allele (delGA) in lacking the transactivation domain, although they lack the C-terminal extension present in the human allele. In transient transfection assays human 059 mutant protein has been shown to have no transcriptional activation (Bondurand et al., 2000; Kuhlbrodt et al., 1998b; Liu et al., 1999). The m618 allele substitutes Leu142 in the second alpha helix of the HMG domain and there is no similar mammalian mutation. The t3 allele results in a severely truncated protein that lacks both the DNA-binding and transcriptional activation domains. It is thus reminiscent of the human Y83X allele which has been proposed to be a functional null (Kuhlbrodt et al., 1998b; Potterf et al., 2000). All four mutant alleles described here show similar, strong phenotypes (Kelsh et al., 1996; Malicki et al., 1996) (data not shown). These considerations, combined with the similarity of these phenotypes with the maximal morphant phenocopy generated with a sox10 morpholino, lead us to suggest that these alleles are all likely null alleles. Analysis of the activities of these mutant proteins will be interesting to test this proposal.

Zebrafish sox10 expression is consistent with cell types affected in cls mutants and strongly reminiscent of that in mammals. We have taken advantage of the higher resolution of such studies in zebrafish, to define more precisely the extent of sox10 expression throughout the neural crest at different stages. In wild type, expression is extensive in premigratory neural crest, but comparison with fkd6 expression shows that a minority of premigratory NCCs lack sox10 expression. Expression persists in some migrating cells on the medial migration pathway, but is rapidly downregulated in differentiating pigment cells. Glia of the developing peripheral nervous system show strong expression, as has also been described in mice (Kuhlbrodt et al., 1998a; Britsch et al., 2001). Crest cells in forming branchial arches and fin mesenchyme do not show expression, in agreement with mouse, but not human, expression studies (Bondurand et al., 1998; Kuhlbrodt et al., 1998a; Southard-Smith et al., 1998). Late expression in cranial cartilages is intriguing, but will require further study to evaluate its role. In cls mutants, sox10-expression in premigratory NCCs is initially unaffected but, unlike in wild-type siblings, sox10-expressing cells soon accumulate in this position. In the trunk sox10-expressing cells migrating on the medial pathway are seen and presumably contribute to the neurones and glia of the dorsal root ganglia, which are detectable in mutants (Kelsh and Eisen, 2000) (this study). NCCs tend to become clustered in premigratory positions in cls mutants, consistent with the observed defect in NCC migration revealed by our single-cell labelling studies.

Outside the neural crest, sox10 expression is also largely conserved. Thus, expression in the developing inner ear epithelium is seen in zebrafish as well as in mammals (Bondurand et al., 1998; Southard-Smith et al., 1998). In zebrafish, this expression is remarkably strong and persistent in wild type, but is downregulated in cls mutants by 40 hpf. As pronounced otic defects are morphologically detectable in cls mutants by 48 hpf (R. N. K., unpublished), we suggest that sox10 is crucial for development of otic epithelium. Limited expression in embryonic central nervous system is seen in zebrafish and mice (Kuhlbrodt et al., 1998a), although detailed studies will be required to establish the cell types involved in each case.

Our single cell labelling studies at three rostrocaudal positions make clear that defects in cls mutants are not limited to one axial position, but instead affect non-ectomesenchymal rather than ectomesenchymal crest derivatives. Most NCCs in cls mutants show restricted migration and poor or no differentiation before dying by an apoptotic mechanism within a discrete time-window. We interpret the dying cells as being those that would in wild-type siblings form non-ectomesenchymal fates. Thus, in cls embryos, NCCs that would normally yield these missing neural crest derivatives are present in premigratory neural crest in normal numbers, but then die before differentiating, usually without migrating. This confirms and extends reports that putative NCCs apoptose during migration in mouse sox10 mutants (Kapur, 1999; Southard-Smith et al., 1998).

The sox10 expression pattern in cls mutants may appear contradictory to our single cell label results, as the expression studies show normal numbers of sox10-positive NCCs migrating on the medial pathway, while the single cell labelling studies show a strong migration defect in the neural crest. However, our data suggest that sox10 is expressed in almost all premigratory cells, but is rapidly downregulated in ectomesenchymal and pigment cell precursors as they start to migrate. Thus, sox10 expression in migrating cells reflects just the neural precursors. Our studies show that these cells are much less defective in migration in cls mutants. Consistent with this, the labelled cells in cls mutants that did migrate normally all took the medial migration pathway and, if they survived, adopted a position consistent with a dorsal root ganglial fate. By contrast, most labelled NCC clones generated pigment cells, consistent with former studies in zebrafish. These pigment cells are severely defective in migration, as shown by molecular markers for xanthophore (gdh) (Parichy et al., 2000) and melanophore (dct; Fig. 5) cell fates (M. Hawkins and R. N. K., unpublished) (Kelsh and Eisen, 2000). Further work will be required to investigate the specification status of the migrating neural precursors in wild-type and mutant embryos.

We have demonstrated the complexity of NCC defects in cls embryos, with survival, migration and differentiation of non-ectomesenchymal cell types all affected. Further, a small decrease in clone size suggests that proliferation may also be somewhat affected (K. A. D. and R. N. K., unpublished). This spectrum of defects is not readily consistent with a primary function for wild-type cls/sox10 in NCC survival, proliferation or differentiation. Although a dying cell might show abnormal migration and proliferation, the timing of appearance of defects in cls mutants conflicts with cell survival being the primary defect. We can first distinguish cls mutants at 20 hpf (by lack of nac/mitf expression), but do not see NCC apoptosis by TUNEL until around 35 hpf; previous work in zebrafish has suggested that the delay between induction of apoptosis and detectable morphological changes and DNA fragmentation is at most 3-4 hours (Ikegami et al., 1999). Likewise the cls mutant phenotype cannot be explained as primarily a failure of migration, with subsequent failure of exposure to required trophic factors, because for three subsets of affected cells (three pigment cell types in the stripe dorsal to the neural tube), the premigratory position is also a final location and hence the necessary trophic factors must be available. Despite this, in cls mutants all pigment cells, including those of the dorsal stripe, fail to develop properly. Instead, we propose that the primary role of cls gene function is in specification of non-ectomesenchymal cell fates; all other defects in cls mutants would then be secondary effects of a failure to become properly specified. Thus, in cls mutants NCCs are unable to adopt non-ectomesenchymal fates and so fail to differentiate. This differentiation process would include expression of growth and trophic factor receptors, and hence secondary defects might include impaired proliferation and later apoptosis. Our identification of the cls gene as a sox10 transcription factor gene is consistent with this interpretation.

Indeed, for the specific case of melanophore fate, data presented here and elsewhere permit us to propose a molecular model consistent with this interpretation (Fig. 6). In both zebrafish and mammals, Nac/Mitf homologues have been shown to be important transcription factors for specifying melanophore fate (Lister et al., 1999; Opdecamp et al., 1997; Tachibana et al., 1996). We show here that cls mutants lack expression of nac (indeed it is the earliest defect we have identified). Thus, sox10 is required for nac expression and thus for melanophore lineage specification. In addition, in nac mutants, nac-positive cells show a failure to migrate highly reminiscent of the cls melanoblast phenotype (Lister et al., 1999). Likewise, in both cls and nac mutants, spa/kit expression is lost in melanoblasts. spa function is necessary for both melanoblast migration and melanoblast and melanophore survival (Kelsh et al., 2000b; Parichy et al., 1999). Furthermore, the timing of loss of melanoblasts in spa mutants, beginning between 30 and 36 hpf, is consistent with the timing of decrease in melanoblast numbers and with the timing of NCC apoptosis in cls mutants (Kelsh et al., 2000b) (this study). Hence, a simple model can largely explain the cls melanophore phenotype. cls/sox10 has a key role, by direct or indirect activation of nac/mitf, in melanophore specification and consequently cls mutants lack expression of genes activated by nac, including spa/kit, mediating survival and migration, and melanogenic enzymes, mediating pigmentation. In support of this model, injection of sox10 mRNA into early embryos results in ectopic nac/mitf expression and expression of nac in cls embryos is sufficient to rescue melanophores (S. E. and R. N. K., unpublished). It will be of interest to expand this model to include other targets of sox10 and nac, as well as to extend this model to other non-ectomesenchymal neural crest derivatives affected in cls mutants. Our model predicts that the immediate targets of sox10 will be either master regulators of neural crest fate (e.g. nac) or a multitude of diverse gene products required within specific crest derivatives for the cellular functions of trophic support, migration and differentiation. Thus, it will be important to identify the direct targets of Sox10 transcriptional regulation.

This specific model is likely to be applicable to mammalian neural crest too. Thus, Sox10 directly regulates Mitf expression in mouse (Bondurand et al., 2000; Lee et al., 2000; Potterf et al., 2000; Verastegui et al., 2000). Although Dct-positive melanoblasts are not detectable in Sox10Dom mice, some evidence of impaired or delayed migration of some NCCs has been presented (Southard-Smith et al., 1998). Similarly, Kit function is critically involved in promoting melanoblast migration and survival, is rapidly lost from melanoblasts in Mitf mutants and is known to be a direct target of Mitf, at least in cultured mast cells (Opdecamp et al., 1997; Wehrle-Haller and Weston, 1995; Wehrle-Haller and Weston, 1997). In Sox10 mutant mice, Kit expression in melanoblasts may be absent, although the presence of unaffected cells in the skin, suggested to be mast cells, means that this result needs to be confirmed (Bondurand et al., 2000).

sox10 alone cannot be sufficient to specify each of the non-ectomesenchymal fates. We believe that sox10, while crucial for each of them, is not sufficient; instead, we propose that sox10 is one of a combination of genes that together specify non-ectomesenchymal fates from neural crest. This model helps explain why cells that express molecular markers for affected cell types can be identified in cls mutants (Kelsh and Eisen, 2000). For example, although melanophores barely express melanin, xanthophores show much pigment in cls mutants. If a combination of transcription factors together specify a cell type, each factor might regulate only partially overlapping subsets of the characteristics of individual fates. In the absence of sox10 gene function, those aspects of melanophore and xanthophore fate requiring sox10 function would be affected (e.g. synthesis of melanin and trophic machinery, respectively), while others would remain at least partly unaffected (e.g. dct expression and pteridine pigment synthesis, respectively). A directly analogous mechanism has been proposed for autonomic neurone specification from the neural crest, with different genes regulating pan-neuronal and subtype specific cell properties (Anderson et al., 1997). Factors that interact with Sox10 to regulate target genes have been described, including the paired box transcription factor Pax3 and the POU-domain protein Tst-1/Oct6/SCIP (Bondurand et al., 2000; Kuhlbrodt et al., 1998a). However, it remains to be seen whether these co-factors have any role in distinguishing non-ectomesenchymal fates. Identification of combinations of transcription factors required to specify each fate will be a promising line of future research.

Our specification model predicts that sox10 function is required only in NCCs fated to non-ectomesenchymal derivatives. fkd6 is expressed very broadly in premigratory NCCs (Odenthal and Nuesslein-Volhard, 1998) and shows extensive, but incomplete, overlap with sox10. Further studies will be required to test whether sox10-; fkd6+ NCCs represent ectomesenchymal precursors. However, our results do show that, consistent with our model, expression is, at least, rapidly downregulated in cells that adopt ectomesenchymal fates. In mammals, Sox10 has sometimes been assumed to be a generic NCC marker (Pattyn et al., 1999), but our results show that, in fish, overlap with other NC markers is incomplete, even at premigratory stages.

In many respects, zebrafish neural crest development is typical of neural crest development in all vertebrates (Raible et al., 1992). In contrast to mouse and human Sox10 mutations, none of the fish sox10 mutations show dominant effects on melanophore number or pattern (R. N. K., unpublished). In mice, expressivity of the sox10 haploinsufficiency phenotype is dependent on the genetic background (Lane and Liu, 1984). Presumably, in the AB background in zebrafish heterozygotes have sufficient Sox10 for normal melanophore development. Nevertheless, the conserved expression pattern and homozygous phenotypes in the two species suggest that the detailed cell-biological basis for the cls phenotype proposed here illuminates a possible mechanism behind Waardenburg-Shah syndrome and Hirschsprung’s disease, in which reduced SOX10 function may reduce the number of melanoblasts and enteric precursors specified.

We thank J. James for fish husbandry, L. D. Hurst for generation of Fig. 1B, D. Raible for pCSHSP, and C. Miller, D. Parichy and D. Raible for cDNA clones. We also gratefully acknowledge J. Eisen, D. Raible, J. M. W. Slack, R. Adams, W. Bennett, A. Ward and J. R. Dutton for helpful criticism and discussion of the manuscript. R. G. thanks S. Rudolph-Geiger for technical assistance. This work was supported by a Praxis XXI PhD Studentship (S. S. L.), a University of Bath studentship and ORS award (T. J. C.), a Royal Society Research Grant, Wellcome Trust Project Grant 050602/Z, a MRC Co-operative Group Component Grant G9810985, and MRC Co-operative Group Grant G9721903. R. G. and P. H. were supported by the German Human Genome Project (DHGP Grant 01 KW 9919).