Loss of cerebum function ventralizes the zebrafish embryo

In their experiments on the mechanisms of vertebrate gastrulation, Mangold and Spemann demonstrated the ability of cells from the amphibian dorsal blastopore lip to induce an ectopic embryonic axis upon transplantation to the ventral side of another embryo (Spemann and Mangold, 1924). The trans-planted dorsal tissue was termed the organizer because of its ability to induce surrounding ventral host cells to develop as dorsal mesoderm and neurectoderm. The response of the embryo to the organizer is graded, with the amount of dorsal mesoderm and neurectoderm induced in proportion to the amount of organizer tissue present (Stewart and Gerhart, 1990). The organizer appears to be a conserved feature of vertebrate development, as embryonic regions with similar inductive properties have been demonstrated for zebrafish, chicken and mouse embryos (Kintner and Dodd, 1991; Ho, 1992; Storey et al., 1992; Beddington, 1994; Shih and Fraser, 1996).

The organizer is thought to produce extracellular signaling molecules responsible for its patterning influence on adjacent mesoderm and ectoderm, and, in the absence of these signals, cells assume ventral fates (refer to Slack, 1994). Candidates for such substances include Chordin and Noggin, both of which are expressed by the Xenopus organizer and have the ability to dorsalize mesoderm and to induce neurectoderm in animal cap assays (Sasai et al., 1994; Smith and Harland, 1992).

More recently, the model of an actively dorsalizing organizer and a ventral default state was challenged by the discovery that several bone morphogenetic proteins (BMPs), members of the TGFβ superfamily of growth factors, act as ventralizing factors that can override the influence of the organizer. Overexpression of BMP-4 or BMP-2 promotes development of ventral mesoderm and epidermis at the expense of dorsal mesoderm and neurectoderm (Jones et al., 1992; Fainsod et al., 1994; Clement et al., 1995; Hemmati-Brivanlou and Thomsen, 1995; Schmidt et al., 1995; Wilson and Hemmati-Brivanlou, 1995). Conversely, the development of dorsal mesoderm and neural tissue is promoted by reducing the activity of BMP-4, either through injections of antisense RNA (Steinbeisser et al., 1995), ectopic expression of a truncated receptor (Graff et al., 1994; Schmidt, et al., 1995) or expression of a non-cleavable BMP-4 (Hawley et al., 1995).

A connection between BMP signaling and the organizerderived factors Noggin (Smith and Harland, 1992) and Chordin (Sasai et al., 1994) has come from the recent demonstration that both proteins directly bind BMP-4 and prevent it from binding its receptor (Piccolo et al., 1996; Zimmerman et al., 1996). Thus, it is likely that the dorsal patterning activity of the organizer in part arises from its ability to antagonize the activity of BMP-4 or other similar ventralizing factors. This interaction represents a highly conserved mechanism; bmp-4 is a vertebrate homologue of the Drosophila gene decapentaplegic (dpp), required for proper dorsal-ventral patterning, and dpp is antagonized by short gastrulation (sog), the Drosophila homologue of chordin. Although no Drosophila homologue has been identified for noggin, Xenopus noggin RNA is capable of antagonizing dpp in Drosophila embryos (Holley et al., 1996). Analysis of mouse mutants produced by targeted gene disruption of Bmp-4, in which formation of mesoderm is sub-stantially disrupted (Winnier et al., 1995), also provides genetic evidence for the importance of BMP-4 in early patterning of vertebrate embryos.

Studies of dorsal-ventral patterning in Xenopus have revealed a complex series of interactions, involving factors both maternally supplied and zygotically expressed, some having similar activities in misexpression assays (refer to Sive, 1993). Some genes have been identified that are thought to act upstream or downstream of BMP-4 signaling (Christian and Moon, 1993; Gawantka et al., 1995; Ladher et al., 1996; Schmidt et al., 1996). However, not all of the implicated genes have been easily placed in regulatory networks, nor have all of the genes involved been identified.

The zebrafish provides the means for exploring dorsal-ventral specification and its importance in vertebrate gastrulation using a genetic approach. Mutagenesis screens for early embryonic phenotypes (Kimmel, 1989; Mullins et al., 1994; Solnica-Krezel et al., 1994) have already identified genes proven to be important in the development of other vertebrates (Halpern et al., 1993; Schulte-Merker et al., 1994b; Talbot et al., 1995), demonstrating that genetic pathways are likely to be conserved. Furthermore, characterization of zebrafish mutations affecting dorsal-ventral patterning will not only facilitate studies on the interactions among known genes, but will reveal previously unidentified components of signaling pathways.

During a mutagenesis screen for early embryonic phenotypes, we identified cerebum (crm) mutants that exhibit defects in the morphogenesis and differentiation of mesodermal and ectodermal derivatives. As initial steps toward molecular identification of crm, we have mapped 5 gamma (γ)-ray-induced deficiencies to zebrafish linkage group 15 (LG 15) and we have ordered DNA markers in the crm chromosomal region. Analysis of the early phenotype of crm mutants suggests that the primary defect lies in dorsal-ventral specification at gastrulation, when ventral mesoderm and ectoderm are expanded at the expense of lateral mesoderm and neurectoderm. The crm loss-of-function phenotype reveals the existence of ventralizing signals in the zebrafish gastrula and provides genetic evidence for antagonists of those signals.

Techniques for the care and breeding of zebrafish were followed that have been previously described in detail (refer to Westerfield, 1993). The cerebum deficiency alleles (b305, b386, b409, c4, c47) are maintained as intercross lines or as outcrosses to wild-type (WT) fish of the Oregon AB background (provided by C. Walker and C. B. Kimmel). Each generation, heterozygous adults were identified by mating. In later experiments, carriers of crm deficiencies b386 and b409 were crossed to homozygotes for the dominant mutation jaguar (jag), which produces distinct body pigment patterns in heterozygotes and homozygotes (T. Mason and S. L. Johnson, personal communication) and, as described below, maps to the same chromosome arm as crm. Doubly heterozygous progeny (crm⁺; jag⁻/crm⁻; jag⁺) were maintained as interbreeding stocks. In these stocks, the jag⁻/jag⁺ het-erozygotes, identified by their disrupted body pigment stripes and lack of tail pigment stripes, were always found to be crm⁻/crm⁺ and were readily identified from their sibling jag⁻; crm⁺ homozygotes.

Embryo culture and staging

Embryos were collected from natural matings and maintained in embryo medium (15 mM NaCl, 0.5 mM KCl, 1 mM CaCl₂, 1 mM MgSO₄, 0.15 mM KH₂PO₄, 0.05 mM Na₂HPO₄, 0.7 mM NaHCO₃) at 28.5°C. Embryos were staged according to standard morphological criteria (Kimmel et al., 1995).

γ-mutagenized stocks were generated essentially as described (Chakrabarti et al., 1983; Westerfield, 1993). Sperm was collected from WT (AB) males and pooled in ice-cold Hank’s solution. A dose of 250-300 rads was delivered from a ¹³⁷Cs source and the sperm used to fertilize WT eggs expelled by squeezing of WT females.

To map the crm locus, we followed techniques developed by Postlethwait et al. (1994) that were used to produce a linkage map for the zebrafish based on random amplified polymorphic DNA (RAPD) markers in two partially inbred strains, AB and Darjeeling (DAR). A map cross panel of DNA was obtained by mating a b305/+ carrier with the AB background to a DAR fish. Haploid embryos were derived from the resultant F1 AB/DAR crm⁻/crm⁺ females by fertilizing their eggs in vitro with UV-irradiated sperm (Streisinger et al., 1981). Genomic DNA was prepared from individual mutant and WT haploid embryos as described (Postlethwait et al., 1994). Typically, 0.2% of a sample was used in an individual polymerase chain reaction (PCR). Since the original mutagenesis was performed on AB fish, closely linked AB-specific RAPD markers only appeared in crm mutants, while closely linked DAR-specific markers were only present in WT haploids.

To detect RAPD markers by PCR, 25 μl reactions were set up in 96-well plates. The final reaction mix contained 10 mM Tris-HCl, pH 8.3, 1.5 mM MgCl₂, 50 mM KCl, 0.01% gelatin, 1 mg/ml BSA, 1 mM each dNTP, and 25 ng of the 10-base primer. Primers were purchased from Operon Technologies (San Diego, CA), as part of their standard 10-mer kits, and their primer designations are used in naming mapped RAPD markers. Reactions were taken through 37 cycles of 5 seconds at 94°C, 55 seconds at 92°C, 60 seconds at 36°C, 105 seconds at 72°C, followed by 7 minutes at 72°C. Pools of crm mutant and WT genomic DNA were screened by PCR for the approximately 400 previously mapped RAPD markers (Postlethwait et al., 1994), and comparisons were made of mutant and WT amplification products upon gel electrophoresis. Segregation of markers with the crm mutant phenotype was confirmed by retesting of individual haploids.

Recently, another mapping panel of zebrafish DNA was constructed based on simple sequence length polymorphism (SSLP) markers (Knapik et al., 1996, and E. W. Knapik, personal communication), and is being correlated with the original RAPD-based linkage map (J. Postlethwait and S. Horne, personal communication). SSLP markers were detected by PCR of individual haploid or diploid DNA samples, with 100 nM each of forward and reverse primers, purchased from Research Genetics (Huntsville, AL). Amplifications were performed as described (Knapik et al., 1996), with the omission of [γ³²P]ATP. Products were resolved by 5% Metaphor (FMC Bioproducts, Rockland, ME) agarose gel electrophoresis.

To amplify the 3′-UTR of lim-1, 25 μl reactions were set up as above with 500 ng of forward and reverse primers (provided by R. Toyama and I. Dawid). Reactions were taken through 30 cycles of 1 minute at 94°C, 1 minute at 50°C and 2 minutes at 72°C.

RAPD primers typically amplify several bands, each of which is potentially an independently segregating marker. To make PCRprimers specific for the LG15 15N.875 band, a sample of genomic SDNA was amplified with the N15 RAPD primer and electrophoresed in low melting point agarose (FMC Bioproducts, Rockland, ME). The 875 bp band was excised and ligated in the pCR™II vector, using the TA Cloning Kit (Invitrogen). After DNA sequencing, primers were made to the ends of the fragment and used in amplifications with an annealing temperature of 55°C. The primer sequences are GTT-TAACCCTTTAACAGG (forward) and TCAGTCATGCAGAATGGA (reverse), and the converted RAPD marker is referred to as 15N.875^*.

Digoxigenin-labeled antisense RNA probes were synthesized from linearized DNA templates using T3 [krox-20 (Oxtoby and Jowett, 1993), α-collagen II (Yan et al., 1995)] or T7 [lim5 (Toyama et al., 1995a), lim1 (Toyama et al., 1995b) otx1 and otx2 (Li et al., 1994) islet1 (Korzh et al., 1993), goosecoid (Thisse et al., 1994), no tail (Schulte-Merker et al., 1992), eve1 (Joly et al., 1993)] RNA poly-merase (Boehringer Mannheim). The gata2 RNA probe (Detrich et al., 1995) was provided by Len Zon and Steve Pratt. Whole-mount in situ hybridizations were performed essentially as described (Thisse et al., 1993), using homozygous crm^b305 or crm^c4 mutant embryos. For in situ hybridizations on embryos prior to tail bud stage, before mutants can be sorted by morphology, embryos from intercrosses of crm^c4/crm⁺ carriers were used. Abnormal patterns of expression were present in the expected 25% of embryos. In some cases, following in situ hybridization and probe visualization, embryos were genotyped by DNA extraction and PCR amplifcation of the 15N.875* marker, which is absent in crm^c4 homozygotes (data not shown).

To detect cell death in mutant embryos and WT siblings, solutions and terminal transferase provided in the Apoptag Kit-Peroxidase (Oncor, Inc.) were used. Embryos were fixed and dehydrated as for in situ hybridization. After rehydration, they were incubated 1 hour at room temperature in equilibration buffer, then in working strength terminal transferase at 37°C for 2 hours. After several washes in stop/wash buffer, embryos were incubated with anti-digoxigenin antibody conjugated to alkaline phospatase (Boehringer Mannheim) and visualized as for RNA in situ hybridization.

In a mutagenesis screen for early embryonic phenotypes, cerebum mutants (crmb305) were identified due to their dramatically reduced brains and eyes, and abnormal accumulation of cells caudal to the anus. During subsequent mutagenesis screens, three more γ-ray-induced mutations, b386, b409 and c47, were isolated by the remarkably similar phenotypes they produced. These were later confirmed to be allelic by failure to complement crm^b305. An additional allele, c4, was discovered by screening for noncomplementation of the crm mutant phenotype in matings of b305/+ heterozygotes to progeny of γ-irradiated sperm.

With the exception of b409 (see Fig. 3C,F), all the crm alleles produced indistinguishable phenotypes as haploids (b305, b386, and c47), as homozygous diploids (b305, b386, and c4; Fig. 1B), or as transheterozygotes. The mutant phenotype was morphologically apparent during segmentation stages, when the tailbud was noticeably enlarged and the anterior neural keel shallower than in WT siblings (Fig. 1C,D). The observed frequency of crm mutant progeny from intercrosses of heterozygous fish, or in haploids produced from heterozygous females, indicated that the c4 allele segregates with simple Mendelian ratios (Table 1). In the case of crm^b305 and crm^b386, we observed a lower frequency of crm mutants in haploid clutches and in intercross progeny of heterozygous carriers (Table 1). In addition, there was a second distinct cosegregating mutant phenotype present in both the haploid and diploid siblings of b305 and b386 mutants. These data suggest that each of these alleles results from a balanced reciprocal translocation. Additional evidence for this has come from genetic mapping studies.

The crm locus was placed on the zebrafish genetic map using PCR techniques that allow the identification of polymorphic DNA markers that segregate with a given mutant phenotype (Postlethwait et al., 1994). From the examination of 400 pre-viously mapped random amplified polymorphic DNA (RAPD) markers, the linkage group (LG) 15 marker 9AB.930 was found to segregate with WT haploid siblings of crm^b305 haploid mutants (data not shown). Subsequently the more distal marker 15N.875^* was discovered to be deleted in genomic DNA of crm^b305 haploids (Fig. 2A). A size polymorphism in 15N.875^*, present in some WT fish (AB), demonstrated that haploid embryos displaying the distinct, early embryonic lethal phenotype (ghost) that accompanies crm^b305 have the expected duplication of the LG 15 region that is deleted in crm mutants (Fig. 2A). Thus, mapping analysis supports the segregation data (Table 1), suggesting that crm^b305 is maintained as a balanced reciprocal translocation.

DNA isolated from diploid or haploid mutant embryos was tested for the presence of additional LG 15 RAPD and simple sequence length polymorphism (SSLP) markers, and the data were used to generate a comparative map of the crm alleles (Fig. 2C). Mapping results indicated that all crm alleles are deficiencies, deleting variable extents of LG 15. This analysis has also allowed us to order SSLP markers. The absence of Z24 and the presence of Z5393 in c4 indicate that Z5393 is distal to Z24 on the LG 15 map. Finally, the presence of markers Z732 and Z1195 in b409 has allowed us to narrow the LG 15 region responsible for the crm phenotype to approximately 17 cM.

Mapping of crm deficiencies has revealed useful information about neighboring loci. The dominant viable pigmentation mutation jaguar (jag), which produces distinct heterozygous and homozygous phenotypes, had been previously mapped to LG 15 proximal to crm (Johnson et al., 1996). From evidence that jag maps near or distal to the breakpoints in b409 and b386 (see Fig. 1D), we predicted that recombination would be sup-pressed at the jag locus in carriers of these deficiencies. This was confirmed and has made it possible to maintain inter breeding stocks with the b386 and b409 alleles, without mating fish to identify crm/⁺ carriers, as described in Materials and Methods.

The homeodomain gene lim-1 was also recently mapped to LG 15 (J. Postlethwait, personal communication), proximal to jag. We find that the lim1 gene is disrupted in crm^b409 (Fig. 2B), potentially causing the more severe brain phenotype of b409 homozygous mutants (see below).

The crm deficiencies lead to a decreased development of the anterior central nervous system (CNS), obvious by morphology (Fig. 3A,B). An enhancement of this phenotype was observed in homozygous mutants bearing the crm^b409 allele, which completely fail to develop recognizable forebrain, midbrain or eyes (Fig. 3C). In confirmation of the morphological defects, krox-20 hindbrain expression is present (data not shown), but the diencephalic expression of lim5 is absent in b409 homozygotes (Fig. 3F). This is in contrast to b305 homozygotes, in which the anterior domain of lim5 expression is largely intact despite the decreased size of the brain (Fig. 3E).

To further characterize the CNS defects associated with the crm mutant phenotype, we examined regionally localized patterns of gene expression by whole-mount in situ hybridization. An early marker of hindbrain rhombomeres 3 and 5, krox-20, is first expressed at the end of gastrulation in a broad band of cells constituting presumptive rhombomere 3 and, later, is also expressed in rhombomere 5 (Oxtoby and Jowett, 1993). The rhombomere-specific domains of krox-20 expression were present in the crm mutant hindbrain, although they were less intense and more variable in width than in WT (Fig. 3G,H). The homeodomain gene, lim5, a marker for the diencephalon (Toyama et al., 1995a), had a normal anterior limit of expression in crm mutants (Fig. 3D,E), as did the general forebrain marker otx1 (Li et al., 1994) (Fig. 3I,J); however, both genes were expressed in reduced domains. Thus, anterior-posterior patterning of the crm⁻ brain is retained, although there is a pronounced reduction in cell number.

The islet 1 (isl1) gene is a marker for subsets of identified neurons (Thor et al., 1991) and an indicator of dorsal-ventral polarity in the spinal cord. In zebrafish embryos, isl1 is normally expressed in dorsal Rohon-Beard sensory neurons and in a subset of ventral primary motor neurons (Korzh et al., 1993; Inoue et al., 1994; Appel et al., 1995). In the crm mutant spinal cord, isl1-expressing cells corresponding to the dorsal Rohon-Beard cells and to the ventral motor neurons were present (Fig. 3K,L); however, the number of motor neurons per spinal cord segment was more variable compared to WT, with fewer isl1-expressing cells overall. The ventral midline cells of the spinal cord, the floor plate, were also present in crm mutants (see Fig. 4D). Differentiation of floor plate, motor neurons and dorsal sensory neurons suggests that dorsal-ventral patterning is largely preserved in the crm mutant spinal cord.

In summary, the CNS defects of crm⁻ embryos can be best characterized as a general decrease in cell number, with anterior structures most severely affected; however, anterior-posterior and dorsal-ventral patterning of the nervous system is largely intact.

To learn how early the CNS phenotype was apparent, we examined expression of neurectodermal markers at gastrulation. For example, otx2, a marker for the presumptive forebrain and eyes (Li et al., 1994) showed a decreased expression domain in crm mutants at gastrulation [80% epiboly] (Fig. 3M,N). In WT gastrulae, the goosecoid (gsc) gene is first expressed in the involuting prechordal mesoderm and, later, an additional domain of expression arises in the overlying neurectoderm (Thisse et al., 1994). Mesodermal gsc expression was intact in crm mutants but the neurectodermal component was greatly reduced (Fig. 3O-R). Expression of these two early markers of neurectoderm suggests that the overall reduction in size of the crm⁻ brain stems from a reduced number of neurectodermal cells at gastrulation.

One possible explanation is that the decrease in neurectoderm observed in crm⁻ gastrulae results from increased cell death. Although morphologically we could not observe an overabundance of dying cells, we used a more sensitive whole-mount terminal transferase assay to directly visualize apoptotic cells in crm^c4 mutants. In progeny from crm^c4/+ intercrosses, there was no noticeable increase in cell death during gastrulation or during early segmentation stages when crm^c4 mutants could be readily distinguished from their WT siblings (data not shown). In contrast to the mutant CNS, by midsegmentation there was a pronounced increase in apoptotic cells in the crm⁻ ventral tail (Fig. 3S,T). Thus, an increase in cell death in the neurectoderm is not the underlying cause of the reduced crm⁻ nervous system. Instead, mis-specification during gastrulation could account for the reduction in neurectoderm.

In addition to their marked CNS defects, crm⁻ embryos develop smaller somites and an accumulation of cells in the tail. To determine the extent to which mesodermal derivatives are affected, we examined expression patterns of mesodermal genes. Hatching gland-1 (hgg-1) is an early marker for the hatching gland, a derivative of the prechordal plate (Thisse et al., 1994). The intensity and number of hgg-1-expressing cells appeared normal in crm mutants, although labeled cells were distributed more closely to the midline than in WT (Fig. 4A,B). Consistent with the early expression of hgg-1, the hatching gland appears morphologically normal at later stages (unpublished data). In the WT trunk, α-collagen II (col2a1) is expressed in floor plate, notochord, and hypochord, a row of axial cells lying just beneath the notochord (Yan et al., 1995). Expression of col2aI in these midline structures was essentially normal in crm⁻ embryos (Fig. 4C,D).

In contrast to axial mesodermal derivatives, derivatives of paraxial and intermediate mesoderm were reduced in crm mutants. Somites are much smaller although they form at the same rate as in WT siblings (unpublished data). Accordingly, myoD, a marker for the myotomal component of somites (Weinberg et al., 1996), is expressed to a much lesser extent during segmentation stages (Fig. 4E,F). After gastrulation, the homeodomain gene lim1 (Toyama et al., 1995b) is expressed in notochord and in the developing pronephric ducts derived from intermediate mesoderm (R. Toyama, personal communication). In crm mutants, the axial domain of lim1 expression was largely normal, while the pronephric duct expression was dramatically reduced (Fig. 4G,H).

Blood precursor cells arise from the ventral mesoderm and eventually migrate to the site of the blood islands in the tail (Detrich et al., 1995). The transcription factor gata2 is expressed in the blood precursors both prior to and after their migration to the blood islands (Detrich et al., 1995). In crm⁻ embryos, an expanded pool of gata2-expressing cells was found in the tail (Fig. 4I,J), enabling some of the accumulated cells caudal to the anus to be identified as blood precursors. Other extra cells are probably derivatives of ventral epiblast, as the expression domain of the homeobox gene eve1, a marker of ventral epiblast and the developing tailbud (Joly et al., 1993), was also greatly increased in crm mutants (Fig. 4K,L). In summary, gene expression analyses have revealed a range of mesodermal defects in crm mutants, with dorsal axial meso-dermal derivatives least affected, paraxial and intermediate mesoderm reduced, and ventral mesoderm increased.

crm⁻embryos are ventralized at gastrulation

The overlapping crm deficiencies produced an invariant embryonic phenotype, differentially affecting derivatives of both mesoderm and ectoderm. Decreased cell numbers in some regions of the mutant embryo and a corresponding increase in other regions, led us to examine cell distribution at gastrulation. The zebrafish Brachyury homologue no tail (ntl) is expressed in mesodermal precursor cells around the margin prior to involution, after which its expression is turned off in all but axial mesoderm and the tail bud (Schulte-Merker et al., 1992). In crm mutants, marginal expression of ntl was broadened ventrally and narrowed dorsally adjacent to the developing notochord (Fig. 5A,B), suggesting that fewer mesodermal cells had converged toward the dorsal side of the embryo. The basis for this difference may be a shift in mesodermal identities. For example, early gata2 expression in WT embryos is confined to a crescent of marginal cells at the ventral side of the gastrula (Detrich et al., 1995), thus defining ventral mesoderm. However, in crm mutants, the gata2 expression domain was dramatically increased around the margin, expanded to include all but a small region encompassing the dorsal embryonic shield (Fig. 5C,D).

At the beginning of gastrulation, eve1 is expressed in the epiblast over the ventral margin (Joly et al., 1993). In crm mutants, eve1 was expressed in a greater circumference around the germ ring at early shield stage (Fig. 5E,F). For comparison, a double in situ hybridization was performed for eve1 and for gsc, an early marker of the dorsal embryonic shield (Stachel et al., 1993; Schulte-Merker et al., 1994a) (Fig. 5G,H). By the late shield stage, eve1 expression in the mutants had expanded to the edge of the gsc expression domain, while the extent of gsc expression was the same in mutant and WT (brackets). Thus, at the onset of gastrulation in crm⁻ embryos, cells in lateral regions of the embryo have inappropriately adopted properties of more ventral cells, while dorsal axial cells appear unaffected.

The cerebum chromosomal region is an efficient target for γ-ray mutagenesis. Genetic mapping analyses demonstrate that the five γ-ray-induced crm alleles result from partial deletions of the distal arm of LG 15. The c4 allele segregates as a simple deficiency; however, several lines of evidence indicate that crm^b305 and crm^b386 are maintained as balanced reciprocal translocations. For each of these alleles, a separate mutant phenotype is found among sibling embryos from intercrosses or from haploids produced from crm/+ carriers, and the observed frequency of mutant classes is consistent with trans-mission of balanced translocations. Additionally, in the case of crm^b305, we have shown that haploids with the cosegregating mutant phenotype ghost have an extra copy of a DNA marker that is deleted in crm⁻ haploids. The lesion in crm^b409 is more complex; the deletion of sequences in two unlinked regions, with intervening sequences intact, is suggestive of an inversion with an accompanying loss of some chromosomal material. We hypothesize that the more severe brain phenotype in crm^b409 is due to the added deletion of lim1, as the absence of forebrain and midbrain in b409 homozygotes is similar to the phenotype described for Lim1 homozygous mutants produced by targeted mutagenesis in the mouse (Shawlot and Behringer, 1995).

With the exception of b409, the overlapping crm deficiencies exhibit a single, recognizable early embryonic phenotype, despite the different amounts of genetic material deleted. This suggests that crm is the earliest zygotically acting gene within the overlapping deleted intervals. In support of this, ENU-induced alleles of dino have been recently recovered, which produce an early embryonic phenotype very similar to crm (Hammerschmidt et al., 1996a). In crosses to crm^c4, dino failed to complement in 41/146, or 28% of the progeny (M. Hammerschmidt and S. Fisher, unpublished data), suggesting that dino and crm are allelic, and sup-porting the hypothesis that the crm⁻ gastrulation phenotype is due to the loss of a single gene. Additional genetic analysis and the mapping of the dino alleles will confirm whether the same gene is responsible for the ventralized mutant phenotype. If this proves to be the case, the cerebum locus will be renamed dino.

Comparative analysis of the crm deficiencies has allowed us to order markers, and has limited the interval in which crm resides to approximately 17 cM, equivalent to approximately 9 megabases (Johnson et al., 1996). From the analysis of other vertebrate genomes (Miklos and Rubin, 1996), we estimate that several hundred genes would map to a chromosomal region of this size. However, far fewer would result in embryonic lethality or encode transcripts expressed at the onset of gastrulation, as we expect of crm.

Ectodermal and mesodermal derivatives are both affected in crm⁻embryos

Deficiencies of crm are pleiotropic, with multiple derivatives of ectoderm and mesoderm affected. The size of the mutant CNS is decreased; however, genes in the brain and spinal cord are expressed in appropriate anterior-posterior and dorsal-ventral patterns. Derivatives of paraxial and intermediate mesoderm are also decreased but, despite their reduced size, somites develop at the same rate as in WT, indicating that some aspects of mesodermal patterning are intact in crm⁻ embryos.

In contrast to the reduction of cells in other regions of the crm⁻ embryo, the tail is greatly enlarged. Aspects of the crm⁻ phenotype are reminiscent of zebrafish spadetail (spt) mutants, in which trunk somites fail to develop and cells accumulate abnormally in the tail (Kimmel et al., 1989). Interestingly, cells that accumulate in the tail of spt mutants express eve1 (Joly et al., 1993), as was observed for some of the excess cells in the crm mutant tailbud, and both mutations result in cell death in caudal regions where cells inappropriately accumulate. Mutations in spt are thought to affect cell fate and subsequent movements at gastrulation, leading to the abnormal morphology of older embryos (Kimmel et al., 1989; Ho and Kane, 1990). We expect that altered cell fate and similar, early morphogenetic defects lead to the pleiotropic phenotype of crm mutants.

Our analysis of crm mutants has revealed defects in dorsal-ventral patterning at gastrulation that could account for the later phenotypic severity and range of affected tissues. The reduced size of the CNS likely stems from a decrease in presumptive neurectoderm at gastrulation, since increased cell death is not observed in crm mutants until mid-segmentation, and then primarily in the tail. While we have not directly demonstrated a change in ecto-dermal cell fates accompanying the reduced neurectoderm, crm⁻ gastrulae clearly exhibit a shift in the identity of mesodermal cells. By 50% epiboly, cells in lateral regions of the crm⁻ gastrula margin inappropriately assume properties of more ventral cells. Later, a corresponding increase in ventral derivatives, such as blood, is observed. In contrast, axial midline structures derived from the dorsal embryonic shield, the zebrafish equivalent of the Xenopus organizer (Ho, 1992; Shih and Fraser, 1996) are largely unaffected in crm mutants.

We note a striking similarity between the phenotype of crm− gastrulae and experimentally ventralized Xenopus embryos, in which a range of defects can be induced with subsets of meso-dermal and ectodermal derivatives affected in concert (reviewed in Gerhart et al., 1991). A molecular basis for ventralization of the Xenopus embryo is suggested by the overexpression of BMP-4 or related signals, which can override organizer-derived dorsalizing signals (Jones et al., 1992; Fainsod et al., 1994; Clement et al., 1995; Hemmati-Brivanlou and Thomsen, 1995; Schmidt et al., 1995; Wilson and Hemmati-Brivanlou, 1995).

Evidence for the existence of ventralizing signals in the zebrafish gastrula has come from interactions between co-cultured regions in an explant system, similar to the animal cap assays widely used with Xenopus embryos (Sagerström et al., 1996). Loss of function of crm also reveals the existence of ventralizing signals in the zebrafish embryo, and further suggests that crm⁺ may normally antagonize those signals. Because of the role that the organizer plays in dorsalizing the Xenopus embryo, and because the equivalent region appears to be spared in crm⁻ embryos, we propose that the crm locus encodes a dorsalizing factor produced by the embryonic shield, which acts to influence the fate of surrounding mesoderm and ectoderm.

Candidates for crm include the zebrafish homologues of genes encoding the dorsally expressed factors Noggin and Chordin. In Xenopus embryos, both proteins can induce dorsal mesoderm and neurectoderm and presumably act through their ability to bind BMPs with high affinity and prevent them from activating their receptors (Piccolo et al., 1996; Zimmerman et al., 1996). Furthermore, ectopic expression of a dominant negative BMP-2/4 receptor can rescue the dino mutant phenotype (Hammerschmidt et al., 1996b), implicating the wild-type function of this locus in regulating BMP signaling.

Using primers specific to a zebrafish noggin gene (Fritz et al., 1996), we have amplified the expected PCR product from DNA of the crm deficiency mutants, ruling out this noggin homologue as a candidate for crm (S. Fisher, unpublished data). We have isolated a partial clone of a zebrafish gene with significant homology to Xenopus chordin and demonstrated that it is absent from DNA of c4 and b305 homozygotes, making it a strong candidate for the crm gene (V. Miller and S. Fisher, unpublished data). Molecular characterization of additional mutant alleles will determine whether crm encodes a zebrafish chordin homologue.

Lack of crm⁺ activity leads to an alteration in specification of the gastrula, such that an expansion of ventral epiblast and mesoderm is accompanied by a decrease in dorsolateral mesoderm and neurectoderm. Analysis of the crm loss-of-function phenotype has provided genetic evidence confirming the existence of ventralizing signals in the zebrafish gastrula, and has suggested a role for crm⁺ as an antagonist of those signals, further demonstrating the similarity of mechanisms that underlie patterning of fish and frog embryos.

The authors gratefully acknowledge the help and hospitality of John Postlethwait and members of his laboratory in the initial experiments to map crm (supported by NIH grants RR/RG10715 and 1PO1HD22486). We thank Charline Walker and Chuck Kimmel, our Oregon colleagues in initial gamma mutagenesis screens (supported by NIH grant 1PO1HD22486); and Steve Johnson for providing map cross DNA, unpublished information and valuable advice on genetics. We especially thank Matthias Hammerschmidt, Reiko Toyama and Igor Dawid, Andreas Fritz, John Postlethwait and Wolfgang Driever, Eliza Shah, Mark Fishman, Howard Jacobs and Ela Knapik for sharing information prior to publication. We thank Chen-Ming Fan, Andy Fire, Amy Rubinstein, Valarie Miller and Steve Johnson for critical readings of the manuscript, and Don Brown for his helpful input. This work was supported by an NIH Clinical Investigator Development Award (NS01850-02) to S. F., and an American Cancer Society postdoctoral fellowship (PF-4087) to S. L. A.; M. E. H. is a Pew Scholar in Biomedical Research.