The expression of a zebrafish gene homologous to Drosophila snail suggests a conserved function in invertebrate and vertebrate gastrulation

During gastrulation, the embryonic tissue of the blastula becomes subdivided into three distinct germ layers, the ectoderm, endoderm and mesoderm. Germ layer formation and cell specification during gastrulation is coupled to specific cell movements. In zebrafish, three morphogenetic movements can be distinguished: epiboly, involution and dorsal convergent extension (Warga and Kimmel, 1990). During epiboly, the blastoderm thins and spreads over the yolk towards the vegetal pole. During involution, cells migrate around the blastodermal margin, causing the blastoderm to fold under itself and to become bilayered. The outer layer, the epiblast, gives rise to ectodermal derivatives, the inner layer, the hypoblast, forms endodermal and mesodermal derivatives (Warga and Kimmel, 1990). Dorsally involuting cells give rise to anterior/dorsal structures, while laterally and ventrally involuting cells form posterior/ventral structures (Kimmel et al., 1990). Dorsal convergence brings the cells of the epiblast and the hypoblast to the dorsal side of the embryo, resulting in the formation and the elongation of the embryonic axis (Warga and Kimmel, 1990).

Two zebrafish mutants have been described that show defects in gastrulation and mesoderm formation. In spadetail (spt) mutant embryos, laterally involuting cells fail to converge to the dorsal side, causing a lack of trunk somites (Kimmel et al., 1989; Ho and Kane, 1990). no tail (ntl) mutant embryos fail to form sufficient mesoderm and lack the notochord and all posterior structures. Furthermore, they show defects in the muscle pioneer cells of the trunk somites, indicating that the differentiation of these cells depends on signalling properties of the differentiated notochord (Halpern et al., 1993). The phenotype of ntl is very similar to that of mouse embryos mutant in the brachyury (T) gene (Chesley, 1935; Yanagisawa et al., 1981; Herrmann et al., 1990). The zebrafish homologue of the T gene was cloned and shown to be identical with the ntl gene (Schulte-Merker et al., 1992, 1994b).

The control of gastrulation and mesoderm induction in vertebrates have been most extensively studied in Xenopus laevis. The initial mesoderm-inducing signals are probably maternally supplied and appear to come from vegetal cells which themselves do not contribute to mesodermal structures (Nieuwkoop, 1969). There seem to exist at least two signals, one inducing dorsal, the other inducing ventral mesoderm. Factors have been found that induce presumptive ectodermal cells to form mesoderm in the animal cap assay. Activin A, a transforming growth factor β (TGF β), preferentially induces anterior/dorsal mesoderm; the basic fibroblast growth factor (bFGF) induces posterior/ventral mesoderm (for reviews, see Smith, 1989). Studies with dominant-negative FGFand activin-receptors (Amaya et al., 1991; Hemmati-Brivanlou and Melton, 1992) provide evidence that the growth factors are involved in mesoderm induction in vivo.

In Drosophila, the control of gastrulation and mesoderm formation has been elucidated based on mutational analysis (for review see Anderson, 1987; St. Johnston and NüssleinVolhard, 1992). Zygotically expressed genes have been identified that control the invagination of the presumptive mesoderm in the ventral furrow of the embryo. Embryos mutant for the genes twist (Thisse et al., 1988) or snail (Boulay et al., 1987) fail to form a ventral furrow and completely lack mesoderm, suggesting that the genes regulate downstream target genes that are involved in the determination of mesodermal fates as well as in the morphogenetic movements of invagination (Leptin and Grunewald, 1990). After the finding that many developmental genes are conserved among invertebrates and vertebrates (see Discussion), snail homologues have been cloned and characterized from Xenopus (xsna, Sargent and Bennett, 1990), mouse (msna, Nieto et al., 1992; Smith et al., 1992) and zebrafish. Similar to snail in Drosophila, the vertebrate snail homologues are expressed in involuting cells, as they become mesodermal, suggesting a role in mesoderm formation and the morphogenetic movements of gastrulation similar to the function of snail in Drosophila.

Animal cap assays in Xenopus revealed that xsna and xbra, the Xenopus T-homologue, can be induced by activin A and bFGF, suggesting an in vivo regulation of the transcription factors by the peptide growth factors. xbra is induced directly as an immediate-early response (Smith et al., 1991), while it is not known whether the induction of xsna is mediated by another gene product (Sargent and Bennett, 1989). Ectopic expression of xbra leads to ectopic induction of xsna (Cunliffe and Smith, 1992), suggesting a regulatory cascade with activin inducing xbra (T), which in turn induces xsna. We show that sna-1 expression is induced by activin A in animal caps. However, this induction, as well as the early sna-1 expression in vivo, is independent of ntl. The early coexpression of zebrafish ntl and sna-1 in a simple pattern in the same population of cells suggests that these genes might instead respond to the same inducing signals in a parallel pathway.

A zebrafish gastrula cDNA library in Lambda Zap (Stratagene) was screened with a ³²P-labelled, randomly primed probe derived from a Xenopus xsna cDNA clone. Replica filter lifts were taken using positively charged nylon membranes (Pall Biodyne B). Hybridisation was performed overnight at 55°C in 5× SSPE, 1% SDS, 5× Denhardt’s solution, filters were washed at 58°C in 4× SSPE, 1% SDS. 13 phage clones were isolated and recombinant Bluescript SK plasmids were recovered by in vivo excision according to suppliers’ instructions (Stratagene). Eight different clones were identified by restriction analysis. About 300 nucleotides at both ends of the inserts were sequenced using vector specific primer. Clones with an open reading frame were completely sequenced by primer walking. Both strands of the 1.8 kb sna-1 cDNA of the plasmid pBS-Sn3 were sequenced. Sequencing was performed with an ALF sequencer (Pharmacia). Specific primers were synthesized with a Pharmacia Gene assembler.

All studies, unless noted otherwise, were performed with embryos from wild-type strains that have been kept and inbred several times in the Tübingen laboratory. The spadetail mutant line spt^b104 and the no tail mutant lines ntl^b160 and ntl^b195 were kindly provided by the laboratory of C. B. Kimmel (Eugene, Oregon) and were outcrossed several times to the Tübingen wild-type strains. spt mutant embryos and ntl mutant embryos of postgastrula stages used for in situ hybridisation were identified by their mutant phenotype, ntl mutant embryos of gastrula stages were diagnosed with anti-Ntl antibody.

Total RNA of 300 or more embryos and of adult males was prepared by the hot phenol method (Schulte-Merker et al., 1992). For northern analysis, 5-10 μg of total RNA were loaded on 1.4% agarose (BRL) formaldehyde gels and run under denaturing conditions. After high-salt transfer onto Hybond N membranes (Amersham), filters were baked at 80°C for 20 minutes, exposed to UV light for 1 minute and stained with methylene blue in order to visualize the rRNAs for loading control. Hybridisation was performed overnight at 68°C in 7% SDS, 0.3 M sodium phosphate buffer pH 7.0, 10 mM EDTA and filters were washed at 68°C in 1% SDS, 40 mM sodium phosphate buffer pH 7.0. ³²P-labelled probe was obtained by randomly primed DNA polymerisation using the BamHI-HindIII fragment of pBS-Sn3 (nucleotides 1-933) as template. Unincorporated nucleotides were removed by gel filtration (Sepharose CL 6B).

The His-tag vector system (Quiagen) was used to express the fulllength Sna-1 protein (pQE12-sn) and a fusion protein of dihydrofolate reductase and the N-terminal part of Sna-1 without the zinc finger domain (pQE16-snΔ114), respectively. Using the appropriate primers for oligonucleotide site-directed mutagenesis, the required restriction sites were introduced into the cDNA by polymerase chain reaction. Recombinant plasmids were transformed into E. coli M15 cells (Quiagen). Cells were harvested 5 hours after induction of protein expression. Recombinant Sna-1 protein was purified by affinity chromatography with Ni-NTA-Agarose under denaturing conditions according to suppliers’ instructions (Quiagen). For immunization of rabbits, Ni-purified protein was further purified by preparative SDS-PAGE.

For affinity purification of the antisera, 10 mg of Ni-purified recombinant full-length Sna-1 protein were coupled to CNBrSepharose (Sigma) in the presence of 100 mM NaHCO₃, 500 mM NaCl, 6 M guanidine hydrochloride pH 8.3. Antisera were loaded onto the affinity column in the presence of 0.5 M NaCl, 0.1% Tween 20, and the antibodies were eluted with 0.1 M glycine pH 2.5. Collected fractions were neutralized with Tris-HCl pH 8.0, and BSA, glycerol and azide were added.

12.5% acrylamide gels were run at constant current (30 mA) using the buffer system described by Laemmli (1970). Blotting was performed for 45 minutes at constant current (400 mA) using immobilon-P membranes (Millipore) essentially as described by Towbin et al. (1979). After transfer, membranes were blocked at least 2 hours in blocking buffer (BB: 100 mM Tris-HCl pH 9.2, 300 mM NaCl, 0.1% Tween 20) + 10% milk powder, followed by incubation with 10³-fold diluted antibody in BB + 5% milk powder. After four washes in BB + 5% milk powder, signal detection was carried out with peroxidase-coupled goat anti-rabbit antibodies and the ECL system (Amersham) according to suppliers’ instructions.

Whole-mount in situ hybridisation with digoxigenin-labeled RNA probes and whole-mount antibody stainings were performed as previously described (Schulte-Merker et al., 1992). For sna-1/Ntl double stainings, embryos first underwent complete in situ hybridisation, followed by complete antibody staining.

To show that sna-1 expression starts at the dorsal side, domestage embryos were divided into three fractions. One fraction was hybridised with a sna-1 probe, the second with a gsc probe (Schulte-Merker et al., 1994a) and the third with both the sna-1 and the gsc probe. The fact that, in all three fractions, the in situ signal came up at one side of the embryos was taken as indication for an expression of sna-1 and gsc at the same side of the embryo. Embryos were orientated with the help of bridged coverslips or spread after manual removal of the yolk. Sections of embryos were taken after whole-mount staining. In situ hybridized embryos first were incubated for 2 hours at room temperature in 4% paraformaldehyde/PBS in order to fix the stain of the alkaline phosphatase colour reaction. Embryos were embedded in Technovit 7100 (Kulzer, FRG) and 5–8 mm sections were cut. Sections were mounted in DPX mountant (Fluka). Photographs were taken at an Axiophot photomicroscope (Zeiss).

Plasmid pBSCT-ZFc1 (Schulte-Merker et al., 1992) was linearized with BamHI, pBS-Sn3 with KpnI. Following the protocol of Driever et al. (1990), capped sense RNAs were obtained using T3 RNA polymerase. The transcripts thus obtained could be translated in vitro. Embryos were injected as described by E. S. Weinberg in Westerfield (1993). Stability of the injected RNA was controlled on northern blots and its in vivo translation was tested on western blots and by whole-mount antibody stainings.

The animal-most quarter of 3.5-hour-old embryos was cut off using an eyelash and transferred to agarose-coated Petri dishes containing incubation medium with or without 50 u/ml human activin A. The incubation medium used was LDF supplemented with 1% FCS (Gibco), 0.4% trout serum and 25 mg/ml trout embryonic extract (Collodi et al., 1992). After incubation for 4-5 hours, animal caps were subjected to immunostaining as described for whole embryos (Schulte-Merker et al., 1992).

The xsna cDNA was used to screen a zebrafish gastrula cDNA library at low stringency. Of 3×10⁵ phage clones screened, two code for a snail homologue, designated sna-1. The longer 1.8 kb clone (pBS-Sn3, Fig. 1) contains 115 nucleotides of 5′ untranslated region, a 3′ poly(A)-tail and an open reading frame encoding a putative protein of 260 amino acids with a calculated relative molecular mass of 28.6×10³ and four zinc fingers in its C-terminal part. There is an in frame stop codon 7 triplets upstream of the first Met codon, indicating that this ATG triplet is the initiation codon for translation of zebrafish sna-1. The Nterminal part of Sna-1 protein contains small regions shared with mouse Msna and Xenopus Xsna (Fig. 2), but bears no significant homology to the Drosophila Snail and Escargot proteins (Whiteley et al., 1992). However, all proteins show significant homology in the zinc finger region, although the first zinc finger is missing in the mouse and the zebrafish gene (Fig. 2). Amino acid sequence comparison of zinc fingers 2-5 reveals 88% identity to xsna, 83% to msna, 71% identity to Drosophila snail and 82% to Drosophila escargot.

Northern blot analysis of RNA (Fig. 3) reveals a single sna-1 transcript of about 1.9 kb in length. Sna-1 message can be detected in cleavage and early blastula stages before zygotic transcription starts (Kane and Kimmel, 1993), suggesting that the transcripts are supplied maternally. In the sphere stage, slightly after midblastula transition, the abundance of sna-1 mRNA is significantly higher, suggesting that zygotic sna-1 transcription starts in blastula stages.

The abundance of sna-1 mRNA continuously increases until the onset of gastrulation, is maximal in stages of midgastrula and drops continuously during later gastrula, neurula and somite stages. sna-1 mRNA can still be detected in 1-, 2and 3-day-old zebrafish larvae, but cannot be found in older embryos and adult males.

Affinity-purified polyclonal antibodies raised against fulllength Sna-1 protein expressed in E. coli specifically detect recombinant Sna-1 protein on western blots (Fig. 4A) and specifically immunoprecipitate Sna-1 protein expressed in reticulocyte lysate (not shown). Western blot analysis with zebrafish embryonic extracts (Fig. 4B) shows that the temporal distribution of the Sna-1 protein reflects that of the mRNA.

Further analysis of temporal and spatial sna-1 expression was performed by whole-mount in situ hybridisation and whole-mount antibody stainings.

In cleavage and blastula stages, before and after the onset of zygotic sna-1 transcription, the sna-1 mRNA is ubiquitously distributed in the embryo (Fig. 5A). In apparently all cells of the blastula embryos, Sna-1 protein is localized in the nucleus. Differences in the nuclear Sna-1 staining among different cells of later blastula stages are due to the asynchrony of the cell cycle: in noncleaving cells the nuclear DAPI fluorescence is quenched by the Sna-1 antibody staining, while cells lacking Sna-1-positive nuclei show unquenched DAPI-positive mitotic figures (Fig. 5B). Shortly before gastrulation (dome stage), the ubiquitous expression of sna-1 mRNA decreases, until it is completely gone by the onset of gastrulation.

Patterned sna-1 expression begins in the dorsal marginal zone, before rapidly spreading circumferentially along the whole margin of the embryo. In the dome stage, sna-1 mRNA is restricted to the dorsal margin of the embryos (Fig. 5C). The dorsal side was identified by simultaneous staining for goosecoid (gsc) mRNA, which is restricted to the dorsal side (Schulte-Merker et al., 1994a) (see Material and methods, not shown). By 30% epiboly, sna-1 is expressed as a ring around the whole marginal zone of the embryo with highest expression at the dorsal side. There, marginal cells express sna-1 prior to their involution (Fig. 5D). After onset of involution at 50% epiboly, sna-1 is uniformly expressed around the whole ring (Fig. 5E) in the cells of the hypoblast, during and after their involution (Fig. 5F).

By 60% epiboly (Fig. 5G), sna-1 mRNA can no longer be detected in involuting cells at the dorsal midline. However, sna-1 continues to be expressed in involuting cells in more lateral and ventral positions. Among the cells that expressed sna-1 during their involution, sna-1 expression is maintained in cells forming the paraxial trunk mesoderm, while sna-1 mRNA can no longer be detected in all other cells shortly after their involution (Fig. 5H). In somitogenesis stages, one row of cells within the paraxial trunk mesoderm adjacent to each side of the presumptive notochord becomes more prominently positive for sna-1 (Fig. 6A,C,E,F). From this cell population, the muscle pioneer cells are drawn (Felsenfeld et al., 1991). Additionally, sna-1 is highly expressed in the tailbud, which develops after closure of the germ ring and where the tail mesoderm is formed (Fig. 6A,E,F,J).

During somite formation, the paraxial sna-1 expression in the hypoblast is organized into a segmented pattern (Fig. 6E). In an intermediate stage, no sna-1 expression can be detected in the loosely organized cells inside the forming somites, while the epithelial cells at the anterior and the posterior boundary of the somites and the muscle pioneer precursors express it at low level (Fig. 6E, small photo). Shortly after formation of the somite, the expression in the muscle pioneer precursors and in the cells of the anterior somite epithelium decreases, whereas the sna-1 expression in the posterior region of the somites becomes significantly stronger (Fig. 6F).

In later forming somites from the 7th somite onwards, the complete rearrangement of the sna-1 expression pattern is accomplished 40 minutes after somite formation, the time required to form two new somites. In contrast, the rearrangement of the first 5–6 somites arrests in the intermediate stage described above (Fig. 6E, small photo). These somites continue the rearrangement synchronously at the 7to 8somite stage (Fig. 6E). This is in line with the synchronous onset of the expression of an engrailed-like gene in the muscle pioneer cells of the first 7-8 somites in 10to 11somite-stage embryos (Hatta et al., 1991; Ekker et al., 1992).

Somitic sna-1 mRNA disappears in a rostrocaudal wave beginning at 18 hours after fertilization. In 1-day-old embryos, sna-1 is expressed in ventral lateral positions of the head region except the region of the trigeminal ganglion and the otic placode (Fig. 7A). From lineage analysis studies, it is known that the cells of this region derive from the neural crest and the paraxial head mesoderm (Schilling, 1993). In hatched larvae (2 days after egg deposition), sna-1 mRNA is found in the lateral and medial parts of the mandibular and the hyoid (1. and 2. pharyngeal arch), while the central regions of the arches are devoid of sna-1 mRNA (Fig. 7B). sna-1 mRNA is also detected in medial regions of the gill arches (Fig. 7C). The sna-1 expression reflects the course of cartilage differentiation, which starts in central regions of the arches and proceeds laterally and medially (Schilling, 1993), suggesting that sna-1 is expressed in presumptive cartilage cells of the arches and turned off as the cells differentiate. Furthermore, sna-1 is expressed in two parallel bands of muscle precursor cells in the pectoral fin buds of larvae 2.5 days after egg deposition (Fig. 7D). The myoblasts of the fin buds are supposed to derive from invading myotomal cells as in chick (Solursh et al., 1987).

As shown above, cells of the paraxial trunk mesoderm express sna-1 from the time that they involute in lateral positions. In spt mutant embryos, laterally involuting cells fail to converge during gastrulation. They finally end up in an enlarged tailbud so that somites are missing in the trunk, but are present in the tail (Kimmel et al., 1989; Ho and Kane, 1990). In spt mutant embryos of the 4-somite stage, the sna-1 staining in paraxial positions and the sna-1-positive lines adjacent to the notochord are absent (Fig. 6H). Only in the posterior-most region of the hypoblast can two faint lateral sna-1-positive stripes be detected. These might correspond to paraxial cells that have reached their position after closure of the germ ring independent of convergent movements and which give rise to the tail somites. The tailbud of mutant embryos is much more strongly stained than in wild-type siblings (Fig. 6H,M versus E,J). Thus, the alteration of the sna-1 pattern in gastrulating spt mutant embryos exactly reflects the described defects of the mutants in the morphogenetic movements of gastrulation.

A number of genes have been identified that are expressed in the presumptive mesodermal region during early gastrulation. ntl, the zebrafish homologue of the mouse T (brachyury) gene, is expressed in the germ ring in the same pattern as sna-1. While descendants of lateral regions of the germ ring abolish the ntl expression after their involution, its expression is maintained in the presumptive notochordal cells at the dorsal midline (Schulte-Merker et al., 1992). At that time, this region is devoid of sna-1 expression. We analyzed the relationship of ntl and sna-1 expression more thoroughly by staining for sna-1 mRNA and Ntl protein in the same embryos.

Ntl protein can be detected in marginal cells on the dorsal side of late blastulas before the onset of the margin-specific sna-1 transcription. In dome-stage embryos, when the first sna-1 transcripts can be detected in dorsal marginal cells, the ntl expression has spread around the whole margin of the embryo. In germ ring embryos (Fig. 8D), sna-1 and ntl are expressed in all involuting cells of the germ ring. Cross sections through the germ ring (Fig. 8C) reveal that cells in the marginal epiblast before their involution express ntl, but not yet sna-1, the expression of which is restricted to the cells in the hypoblast. This indicates that also in these later involuting cells sna-1 expression is preceded by expression of ntl.

Dorsal cells that involute in later stages of gastrulation and lack sna-1 mRNA maintain ntl expression after their involution (Fig. 8E,F). However, in paraxial cells, which coexpressed ntl and sna-1 during involution, only sna-1 expression can be detected later on. This leads to a clear separation of ntl and sna-1 expression in the developing mesoderm. Ntl is expressed in the axial mesoderm and sna-1 in the paraxial trunk mesoderm (Fig. 8F,G,H).

Ntl mutant embryos lack the tail and a differentiated notochord. Both isolated ntl alleles, ntl^b160 and ntl^b195, have ntl mRNA, but lack functional Ntl protein (Schulte-Merker et al., 1992, 1994b). The observation that the onset of sna-1 transcription in involuting cells is slightly preceded by the onset of ntl expression (Fig. 8C), as well as the induction of ectopic xsna transcription after ectopic xbra expression in Xenopus (Cunliffe and Smith, 1992), suggested that ntl might activate the sna-1 transcription. However, in ntl mutant gastrulas, the expression pattern of sna-1 is very similar to that in wild-type siblings, although the sna-1 signals appear to be fainter (Fig. 8A,B versus D,E). The weaker expression of sna-1 in ntl mutant embryos is significant: in gastrulating embryos of the shield stage, when mutant and wild-type embryos cannot yet be distinguished by morphological criteria, we could enrich for ntl mutants by sorting embryos with weaker sna-1 staining after in situ hybridisation (60% ntl in the sorted fraction versus 27% total, Table 1). As an unambiguous identification, however, was not possible, the influence of ntl on the level of sna-1 expression is low. From late stages of gastrulation onwards, ntl and wild-type embryos are morphologically distinguishable. In these stages, the sna-1 expression in ntl embryos is significantly lower than in wild-type siblings (Fig. 6B versus A). The expression pattern, however, is relatively normal: as in the wild type, two rows of cells adjacent to the dorsal midline predominantly express sna-1 (Fig. 6D,G versus C,E), followed by sna-1 expression in the posterior part of differentiated somites (Fig. 6L). Striking differences in the sna-1 expression pattern are visible in older embryos: while the tailbud is less prominently stained in ntl mutant embryos after closure of the germ ring, no sna-1 staining at all is left in the tip of the ‘tail’ of 22h ntl embryos (Fig. 6L versus J). However, this does not necessarily mean that in these late stages ntl is required for sna-1 transcription. More likely, the absence of sna-1 mRNA is due to a cessation of mesoderm formation in the tail tip of ntl mutant embryos, which also is indicated by the absence of ntl mRNA as a marker for newly forming mesoderm (Schulte-Merker et al., 1994b; not shown). Thus, although the level of sna-1 expression is progressively reduced in ntl mutant embryos, the sna-1 expression pattern itself appears to be virtually independent of ntl.

In order to test whether the lower degree of sna-1 expression in ntl mutant embryos might be due to a maintenance or enhancement of sna-1 transcription by Ntl protein, Ntl protein was ectopically expressed by the injection of ntl sense RNA into one cell of 1to 4-cell-stage wild-type embryos. Injected embryos were fixed at different stages and double stained for Ntl protein and sna-1 mRNA. In embryos of the germ ring stage, the intensity of the Ntl antibody staining was similar in nuclei ectopically expressing Ntl and in marginal nuclei with endogenous Ntl expression, suggesting that ectopic Ntl was expressed at physiological levels. Ntl and sna-1 are coexpressed in the germ ring as in uninjected embryos. In more animal regions, ectopic Ntl expression does not lead to the production of sna-1 mRNA (Fig. 9A). However, an effect of ntl overexpression on sna-1 mRNA levels can be demonstrated in embryos of pregastrula stages, when the marginal expression of ntl and sna-1 is still low: in the injected region, the sna-1 staining is significantly stronger than in the uninjected region. The stronger expression is restricted to the margin, the expression domain of endogenous sna-1 (Fig. 9B). Thus, at least in early stages of gastrulation, ntl is not sufficient to initiate, but appears to enhance the sna-1 transcription.

It has been previously shown that ntl is induced by activin in animal caps (Schulte-Merker et al., 1992). After our finding that sna-1 activation is independent of Ntl protein in vivo, we studied whether sna-1 and ntl might be induced in a parallel pathway responding to the same signals. Animal caps incubated in activin A-containing medium express Ntland Sna-1 protein (Fig. 9C), while animal caps incubated in medium lacking the activin do not (Fig. 9D). Expression of Sna-1 protein is also induced in animal caps derived from ntl mutant embryos, which were identified by their failure to express Ntl protein upon treatment with activin (not shown).

We have described a zebrafish gene, sna-1, that encodes a zinc finger protein with striking similarity to the zinc finger domains of Xenopus xsna, mouse msna and the Drosophila genes snail and escargot. The zebrafish and mouse proteins lack a counterpart to the first zinc finger of xsna, snail and escargot. While to date only one snail-related gene was described in mouse and Xenopus, a second zebrafish snailrelated gene has been isolated in addition to the gene described here (B. and C. Thisse and J. Postlethwait, personal communication), suggesting that, as in Drosophila, sna-1 might be a member of a family of snail-related genes. However, the conservation of some regions of the msna, xsna and sna-1 genes outside of the zinc finger domain might support the notion that sna-1 is the homologue of the Xenopus and the mouse gene. Despite the higher structural similarity of sna-1, xsna and msna to Drosophila escargot, we assume that they are functionally more closely related to Drosophila snail, because all of them are expressed in the mesoderm, whereas escargot shows a complex and highly dynamic expression pattern primarily in ectodermal derivatives (Whiteley et al., 1992).

sna-1 expression in early zebrafish gastrulas and snail expression in early gastrulating Drosophila embryos both are restricted to involuting/invaginating cells, which give rise to mesodermal and endodermal derivatives. In gastrula stages of zebrafish embryos, all presumptive endodermal and mesodermal cells express sna-1 during their involution, except dorsally involuting cells of stages from 60% epiboly onwards, which give rise to the notochord (Kimmel et al., 1990). Similarly, in gastrulating Drosophila embryos, snail is expressed in the invaginating presumptive mesodermal cells of the ventral furrow and in cells of the anterior midgut invagination, which give rise to anterior endodermal derivatives. Thus, the snail/sna-1 expression in cells with similar fates in zebrafish and Drosophila suggests that, despite the different temporal and spatial organisation of gastrulation, the genes might have a conserved function in mesoderm and endoderm formation and in the principle morphogenetic movements of gastrulation.

In the recent years, many other developmental genes have been found to be conserved between Drosophila and other organisms. Most strikingly, the system of the homeotic genes, which specifies positional identities along the anteroposterior axis in Drosophila embryos, is conserved in vertebrates, where it apparently serves a similar function. Furthermore, other homeobox genes like caudal/Cdx and even-skipped/Evx as well as members of the MyoD family, the achaete scute/MASH family and the forkhead/HNF family are found both in Drosophila and vertebrates and show some similarities in the expression patterns with respect to cell types or positions within the embryo of both species (reviewed in McGinnis and Krumlauf, 1992).

However, other structurally conserved genes like the Drosophila segmentation genes Krüppel, engrailed, hairy and wingless and their vertebrate counterparts have unrelated expression patterns in Drosophila and vertebrates, making a functional conservation of the genes unlikely. The zebrafish twist-related gene twi, like twist homologues in Xenopus (Hopwood et al., 1989) and mouse (Wolf et al., 1991), is expressed in the presumptive notochordal cells in the axial hypoblast after the cells have involuted (R. Riggleman and M. H., unpublished results). This suggests that twi in zebrafish is involved in the regional specification of parts of the mesoderm but, in contrast to twist in Drosophila, not in the initial mesoderm formation.

In late blastula stages, ntl (Schulte-Merker et al., 1992) and sna-1 are coexpressed in marginal cells, suggesting that they respond to the same inducing signals. Both ntl and sna-1 are induced by activin A in the animal cap assay (SchulteMerker et al., 1992 and this paper). We have shown that the induction of sna-1 expression in animal cap cells by activin A is independent of ntl. Furthermore, ntl is not necessary to activate sna-1 transcription in vivo, as in ntl mutant embryos the pattern of sna-1 expression is normal except for a reduction in the level of expression. Additionally, our studies of ectopic ntl expression, in contrast to the results of Cunliffe and Smith (1992), indicate that in gastrulating embryos ntl is not sufficient to activate sna-1 transcription. However, in line with the reduction of sna-1 expression in ntl mutant gastrulating embryos, ntl does appear to enhance sna-1 expression. The discrepancy between our results and those of Cunliffe and Smith could be due to experimental differences: they determined the sna-1 expression in animal caps corresponding to stage 12 embryos (midto late gastrula), whereas we used intact embryos of early gastrula stages. Thus, the data might suggest that the nature of ntl control on sna-1 expression changes during the course of gastrulation and that ntl becomes sufficient to activate sna-1 transcription in later gastrula stages. However, from our results in ntl mutant embryos, we at least know that also in these late gastrula stages ntl is not necessary for sna-1 transcription.

In late gastrula and somite stages, sna-1 is expressed in the presomitic mesoderm with prominent expression in the muscle pioneer precursor (MPP) cells (Felsenfeld et al., 1991). The MPP cells stop sna-1 expression shortly after somite formation and shortly before their differentiation during which they change their size, shape and relative position. At the same time, sna-1 becomes more strongly expressed in the posterior part of the newly formed somites. A rostrocaudal polarity of the expression pattern was described for a few other genes in chicken somites. These genes are probably involved in guiding innervation and neural crest migration (e.g. Layer et al., 1988; Bronner-Fraser et al., 1992). However, in contrast to chick (Rickmann et al., 1985; Keynes and Stern, 1984), these processes in zebrafish occur in a nonpolar fashion in the middle of the somites (Myers et al., 1986; Raible et al., 1992). So far, it is not known whether the polar distribution of sna-1 is necessary for and how it might contribute to normal somite development.

In ntl mutant embryos of late gastrula and somite stages, the two sna-1-positive lines of muscle pioneer precursor cells are present, although they are not as well organized as in wild-type siblings. This suggests that the determination of muscle pioneer precursors, at least at the level of sna-1 expression, does not depend on ntl and a differentiated notochord. It might either be totally independent of the notochord or be induced by notochord precursor cells, which can be detected in ntl mutant embryos and which are supposed to induce other tissues like the floor plate independent of Ntl protein (Halpern et al., 1993). However, the differentiation of the precursors to muscle pioneers does depend on ntl or a differentiated notochord (Halpern et al., 1993).

Xsna in Xenopus and msna in mouse in addition to their mesodermal expression are expressed in neural crest cells before and during their migration (Nieto et al., 1992; Smith et al., 1992; Essex et al., 1993). In contrast to xsna and msna, the earliest neural crest-specific expression of zebrafish sna-1 can be detected in cranial non-neurogenic cells only after they have migrated to ventrolateral positions. Thus, zebrafish sna-1 may provide only part of the function of Xenopus xsna and mouse msna.

We thank Drs K. Helde and R. Riggleman for the zebrafish gastrula cDNA library, Dr R. Riggleman for the twi cDNA, Dr N. Hopwood for the Xenopus xsna cDNA, Dr C. J. Kimmel and colleagues for providing the mutant lines ntl^b160, ntl^b195 and spt^b104, Dr S. Schulte-Merker for the anti-Ntl antibody and technical advice, Dr J. C. Smith for human activin A, Dr M. Furutani-Seiki for providing animal cap incubation medium, Drs B. and C. Thisse and Dr J. Postlethwait for communicating unpublished results, Dr T. F. Schilling for his help in the interpretation of the sna-1 expression pattern in the head region and Dr M. Brand, Dr R. Kelsh, Dr M. Leptin and R. Warga for their comments on earlier versions of this manuscript. M. H. is a fellow of the Boehringer Ingelheim Fonds.

The sequence shown in Fig. 1 has been submitted to the EMBL database and has the accession number X 74790.

We would like to apologize for using the incorrect abbreviation of the gene name in this paper. According to the zebrafish naming conventions it should be sna1 instead of sna-1.