Genes establishing dorsoventral pattern formation in the zebrafish embryo: the ventral specifying genes

The pattern formation mechanisms establishing the body plan of vertebrates is best understood in Xenopus laevis. The dorsoventral axis of the frog is visible shortly after fertilization by rotation of the surface cortex and formation of the grey crescent (reviewed by Sive, 1993). By an unknown mechanism, this cortical rotation results in the local activation (or repression) of a maternal factor on the future dorsal side. This dorsally localized activity, the Nieuwkoop center, resides in the vegetal blastomeres and during the blastula stage induces overlying marginal blastomeres to form dorsal mesoderm, the future Spemann organizer. Similarly, ventral and lateral vegetal cells induce overlying marginal blastomeres to form ventral mesoderm. A further refinement of dorsoventral patterning occurs following the midblastula transition. The Spemann organizer (the induced dorsal mesoderm) produces a signal, which patterns adjacent mesoderm to form muscle and induces the neuroectoderm to form at anterior-dorsal positions (for reviews see Kessler and Melton, 1994; Sive, 1993; Kimelman et al., 1992). An additional ventral signal is implicated in inhibiting this dorsal signal and specifying ventral cell fates (reviewed in Harland, 1994).

Numerous molecules have been proposed to play roles in these signalling events. Members of the TGFβ and the FGF families are candidates for the signals derived from the vegetal pole (for review see Slack, 1994), while the products of genes such as noggin (Lamb et al., 1993; Smith et al., 1993), chordin (Hemmati-Brivanlou et al., 1994; Sasai et al., 1994, 1995), and follistatin (Kessler and Melton, 1995; Sasai et al., 1995), and those of the wnt gene family members (Cui et al., 1995; Ku and Melton, 1993; Smith and Harland, 1991) have been implicated in the dorsalization signalling and neural induction processes.

Another signalling molecule, bone morphogenetic protein-4 (BMP-4) of the TGFβ family, has been shown to be particularly important with respect to dorsoventral patterning of the mesoderm. Ectopic expression of BMP-4 in dorsal blastomeres results in a ventralization of the embryo: a reduction of dorsal tissues accompanied by an expansion of ventral cell types (Dale et al., 1992; Jones et al., 1992). Consistent with this finding, expression of a dominant negative form of a BMP receptor (Suzuki et al., 1994; Maeno et al., 1994; Graff et al., 1994; Schmidt et al., 1995a) or BMP-4 (Hawley et al., 1995), or injection of antisense BMP-4 RNA (Steinbeisser et al., 1995) into Xenopus embryos results in a dorsalization of ventral mesoderm. Additionally, BMP-4 has the ability to induce ventral mesoderm in animal cap assays (Köster et al., 1991; Dale et al., 1992; Jones et al., 1992; Hemmati-Brivanlou and Thomsen, 1995; Schmidt et al., 1995a). Mouse embryos mutant for the BMP receptor do not form any mesoderm (Mishina et al., 1995). Similarly, most BMP-4 mutant mice do not form mesoderm and arrest at the egg cylinder stage; however, some mutant embryos develop beyond this stage and exhibit a disor-ganization or truncation of posterior structures and a reduction in vental-lateral mesoderm (Winnier et al., 1995). These results implicate BMP-4 and the BMP receptor in mesoderm formation and ventral and posterior cell fate determination. As BMP-4 and the BMP receptor mRNAs are expressed both maternally and zygotically in Xenopus (Köster et al., 1991; Dale et al., 1992; Graff et al., 1994) their ventralizing action could be a part of the ventrovegetal center and/or a postblastula stage ventral activity that inhibits the dorsalizing signal of Spemann’s organizer (see Harland, 1994 for a review). While the maternal transcripts of both genes are not localized in the embryo, BMP-4 mRNA becomes restricted to ventral and lateral regions during gastrulation (Fainsod et al., 1994; Schmidt et al. 1995a; Hemmati-Brivanlou and Thomsen, 1995).

Considerably less is known about mesoderm induction and dorsoventral pattern formation in the zebrafish. An equivalent of the cortical rotation is not apparent in the fish zygote and the orientation of the dorsoventral axis does not become evident until shortly before gastrulation (Schmitz and Campos-Ortega, 1994). During gastrulation, involution and dorsal convergence lead to the formation of a bilayered embryo with a prominent thickening at its dorsal side, termed the embryonic shield (see Kimmel et al., 1995, for a staging series). It has been demon-strated that shield cells, when transplanted to an ectopic position, are able to induce a secondary axis (Ho, 1992). Thus, the shield appears to function as the Spemann organizer in the early zebrafish gastrula. The molecular events involved in establishing the shield are unclear, although a number of experiments suggest the yolk cell to be a source of inductive signals (Devillers, 1952; Kostomarova, 1969; Oppenheimer, 1936).

As in Xenopus, LiCl treatment of early cleavage stage zebrafish embryos causes a dorsalization of the embryo (Joly et al., 1993; Stachel et al., 1993), suggesting a common early pathway of dorsal induction in zebrafish and Xenopus. This notion is strengthened by the similarity of the fate maps of Xenopus (Dale and Slack, 1987) and zebrafish embryos (Kimmel et al., 1990): the notochord and head mesoderm, for example, map to the most dorsal marginal region of the blastoderm, somitic mesoderm encompasses a large portion of the lateral and ventral marginal zone, and pronephros and blood precursors are derived from ventrolateral and ventral positions, respectively. In both zebrafish and Xenopus, more posterior tissues are generally derived from more ventral regions in the blastoderm. For example, tail progenitor cells primarily arise from the ventral side of the blastoderm. Many vertebrate genes that are expressed at blastula or early gastrula stages have been cloned by homology in zebrafish and have similar expression patterns to their mouse or frog counterparts suggesting that determination processes are conserved between these vertebrates.

The functional assays used in amphibians have been extremely valuable for identifying genes with properties expected of mesoderm inducers or dorsalizing and ventralizing signals. However, they do not reveal whether the identified molecules function in this manner endogenously in the embryo. The most stringent test for a gene playing an essential role in a biological process is the analysis of mutants of this gene. This approach has been used successfully in mice to examine the function of a number of genes during early development, including nodal (Conlon et al., 1994), HNF-3β (Ang and Rossant, 1994; Weinstein et al., 1994), FGFR (Deng et al., 1994; Yamaguchi et al., 1994), Brachyury (Herrmann and Kispert, 1994; Beddington et al., 1992), lim1 (Shawlot and Behringer, 1995) and BMP-4 (Hogan et al., 1994; Winnier et al., 1995). Some of these genes are involved in formation of mesoderm and posterior pattern, but only BMP-4 has been reported to be involved in dorsoventral patterning of the early gastrula (Winnier et al., 1995).

Here we describe the identification of 22 mutants that show reductions in ventrally derived structures and expansions in dorsal tissues in the zebrafish embryo. These mutants define 6 genes, 3 of which contain multiple alleles. piggytail (pgy), and possibly somitabun (sbn), function maternally and zygotically in the establishment of the dorsoventral axis, while swirl (swr), snailhouse (snh), lost-a-fin (laf), and mini fin (mfn) appear to function only zygotically. Embryos mutant in all 6 genes show abnormal gastrulation movements. The products of these genes are good candidates for molecules contributing to a ventralizing factor(s) that inhibits the dorsalization signal emanating from the organizer.

Breeding and maintenance of fish and identification and isolation of mutants was done as described by Mullins et al. (1994) and Haffter et al. (1996). The following mutants were used for photography of live animals and staining, except where otherwise noted: laf^tm110b and pgy^dty40 as representative of laf and pgy, respectively, and swr^ta72 of swr, except for the embryos in Figs 3F and 4K which were swr^tc300 .

In situ hybridization and antibody staining were done as described by Schulte-Merker et al. (1992). The RNA probes that were used were myoD (Weinberg et al., 1996), gata1 (Detrich et al., 1995), eve1 (Joly et al., 1993), pax2 (Krauss et al., 1991b) and fkd3 (J. O., unpublished results). Antibodies that were used were an anti-Ntl rabbit polyclonal antiserum (Schulte-Merker et al., 1992). Embryos were photographed using a Zeiss Axiophot microscope. Pictures were scanned into a Macintosh computer using a Nikon slidescanner and processed using Adobe photoshop software.

Histological staining for blood was done as described by Iuchi and Yamamoto (1993).

In our large-scale screen for visible embryonic mutants we identified a total of 22 mutants, which display progressive reductions in the ventral tail fin, length of the tail and amount of blood, associated with progressive expansions of somitic mesoderm. Complementation tests have placed 18 of the mutants into 5 complementation groups, swirl (swr), snail-house (snh), piggytail (pgy), lost-a-fin (laf), and mini fin (mfn), while the analysis is not yet complete for 3 mutants and inconclusive for 1 mutant, which is tentatively called somitabun (sbn) (Table 1 and see below). A spectrum of phenotypes is observed among the mutants, with indistinguishable mutant phenotypes found for the alleles of more than one gene (Table 2). For the ease of describing the genes, we grouped the mutant phenotypes according to their strength into 5 classes, with 1 being the weakest and 5 the strongest. The strength of the phenotype is based on the degree to which the notochord, somites and tail are affected.

The strongest class of defects, class 5, is displayed in homozygous swr and sbn embryos. At 70% epiboly a thinning of the wild-type gastrula occurs in ventral anterior regions as a result of epiboly and dorsal convergence (80% is shown in Fig. 1A, Warga and Kimmel, 1990). Class 5 mutant embryos are iden-tifiable at this stage by a thickened ventral side (Fig. 1E), probably due to reduced dorsal convergence. At the bud stage, the embryo is ovoid rather than round and the notochord appears broadened (Fig. 1B,F). The phenotype becomes very striking at somite stages as the somites extend from the notochord around the circumference of the embryo fusing on the ventral side (Fig. 1C,G). The tailbud does not proceed around the yolk toward the ventral side, as in wild-type embryos, but instead extends off the yolk (Fig. 1C,G). This could reflect an attempt of the vegetal cells around the circumference of the embryo to extend around the yolk, which would normally be exhibited only by the dorsally lying cells (Kimmel et al., 1995). A narrowing of the circumference of the embryo occurs in posterior regions, probably caused by the somites circling around the embryo. As a consequence, the yolk becomes extruded anteriorly. This posterior constriction of the embryo and anterior expulsion of the yolk becomes more severe as development proceeds. The increased amount of yolk now present in the anterior part of the embryo appears to cause too great a hydrostatic pressure, so that by the 13 somite stage the yolk spills out of the embryo and the embryo lyses (Fig. 1H).

The second strongest mutant phenotype, class 4, is found in homozygous snh and dominant sbn mutant embryos (Table 2). As with class 5 mutants, the shape of the embryo becomes ovoid just prior to the bud stage, the notochord is broader (Fig. 1I), and the tailbud extends off the yolk prematurely (Fig. 1J). The first 2 somites are expanded laterally, but do not extend to ventral positions (as with class 5 embryos), whereas the posterior somites do (Fig. 1J). The posterior constriction in class 4 mutants does not lead to a bursting of the embryo, probably because the anterior somites are not as strongly expanded as in class 5 mutants. In later segmentation stages, the trunk axis twists around itself, like a snail shell (compare Fig 1D,K). Two tissues of ventral origin in the blastoderm, the tail and blood are not apparent, (Fig. 2A,B), although small amounts of blood can be detected in some mutant embryos in a histological stain (data not shown). The anteriormost region of the embryo is comparatively normal; the eyes are developed and the midbrain-hindbrain boundary of the brain is visible (Fig. 2B). However, other structures of the head are disorganized and the otic placodes are found fragmented at various positions on the yolk sac (data not shown). Additionally, the pectoral fin buds are improperly located on the yolk sac. The heart tubes are not joined at the midline and two beating heart chambers, often consisting of only a few beating cells, are seen in more lateral positions (data not shown). The embryos survive for 2 to 3 days and then necrose and die. A dorsalization of the embryo is strongly suggested by the phenotype of class 4 mutant embryos: a reduction in cell types of ventral origin in the blastoderm (blood and tail) and an expansion of dorsally derived tissues (anterior somites and notochord).

A phenotype of intermediate strength, class 3, is represented by homozygous mutant embryos of the strongest pgy allele, pgy^dty40. The mutant phenotype is visible at the bud stage as a slight protrusion of the tailbud (Fig. 1L). In early somite stages the phenotype is apparent as an enlarged tailbud that remains in a more vegetal position rather than extending around the yolk (Fig. 1M,N), similar to class 4 and 5 mutants. The anterior somites in class 3 mutant embryos are expanded laterally (Fig. 1M), but only beginning with somite 11 do the somites circle around the embryo (data not shown). In class 5 and 4 embryos this circling begins with the first and third somite, respectively. In contrast to class 4 mutants, the tail of class 3 mutants is present, but shortened and twisted (Fig. 2C) and the ventral tail vein and fin are missing (Fig. 2E,F). Blood is reduced in amount (data not shown). Blood circulation does not occur in 1-day old embryos as in wild type, probably due to the absence of the ventral tail vein. Class 3 mutant embryos necrose and die by day 3. Although a less strong phenotype than class 4, the class 3 phenotype is also suggestive of a dorsalization of the embryo: a reduction in blood and tail length and an expansion of somitic mesoderm.

A weak phenotype, restricted to the tail region of the embryo, is displayed by embryos carrying the recessive alleles of laf and mfn, weak alleles of pgy and the dominant alleles of pgy, swr, and sbn (see Table 2). We distinguish two weak phe-notypic classes. The class 2 phenotype is developed in mutants homozygous for laf and some pgy alleles (Table 1). The phenotype is visible by the 3 somite stage as a slight enlarge-ment of the tailbud, which becomes more pronounced in later somite stages (Fig. 2I-K). As in class 3 mutants, the tailbud does not extend around the yolk as far as in the wild type, but lies in a more vegetal position and becomes enlarged (Fig. 2I,K). The tail is of about normal length, while the ventral tail vein and fin are missing (Fig. 2D). The embryos do not develop blood circulation, which likely results in failure of brain ventricle formation seen in many embryos (Fig. 2D). The mutant embryos necrose and die by day 3.

The class 1 phenotype is the weakest and is displayed in mutants homozygous for mfn, those with weak alleles of pgy, and dominant alleles of pgy, swr, and sbn (see Table 2). In these mutant embryos the ventral tail vein is present, while the ventral tail fin is entirely or partially absent (Fig. 2G,H). In other respects, the class 1 phenotype is very similar to that of class 2 embryos (Fig. 2L). Many of the embryos with weak phenotypes survive to adulthood and frequently have a short tail and no tail fin (Fig. 2M,N).

We investigated the phenotypes in more detail using a variety of molecular markers. In the first series of experiments, genes expressed in the axial and paraxial mesoderm were examined. The expression of the goosecoid gene, which marks the first involuting cells of the gastrula and the most dorsal-anterior mesoderm (Stachel et al., 1993; Schulte-Merker et al., 1994), appears normal at shield stage in class 3, 4 and 5 (pgy, snh, and swr) mutant embryos (Fig. 4J-M and data not shown). Thus the most anterior axial mesoderm appears normal.

The more posterior axial mesoderm of the notochord is expanded laterally; however, the primary cause of this expansion appears to be a reduced amount of dorsal convergence. In strong and intermediate phenotypes (classes 3, 4 and 5), the enlargement of the notochord, which is visible in live embryos, was confirmed by analyzing the expression pattern of Ntl (Schulte-Merker et al., 1992) in mutant embryos at the bud stage (Fig. 3A-D). By this stage in wild type, the notochord precursors have formed 2 layers of cells as a result of dorsal convergence. In class 5 mutant embryos (swr) at this stage, only a single layer of cells is observed for most of the notochord anlage and the anlage is shortened in its anterior-posterior length. Indeed counting the number of Ntl-positive nuclei in wild type and swr mutant embryos reveals that the majority of the broadened midline in swr is due to a reduced amount of dorsal convergence and extension and not to an increased number of notochord precursors.

In the mutant embryos, a mediolateral expansion of adaxial cells is revealed by myoD expression (Weinberg et al., 1996). In wild type at bud stage, myoD is expressed in 2 stripes of paraxial mesoderm (Fig. 3E,′, Weinberg et al., 1996), which correspond to the adaxial cells (Felsenfeld et al., 1991; Thisse et al., 1993). The adaxial cells form 4 rows of cells, one on top of the other, which lie directly adjacent to the notochord (M. Westerfield and S. Devoto, personal communication). This 1-2 cell wide region in wild-type embryos (Weinberg et al., 1996; Fig. 3E,′) is broadened to 4-5 cells in width in class 5 (swr), 3-4 cells in class 4 (snh), and 2-3 cells in class 3 (pgy) embryos (Fig. 3F,′-H′). In addition, in mutant embryos myoD is strongly expressed posteriorly in an adjoining half-ring in the tailbud, which is not seen in the wild type (Fig 3E-H). The anterior-posterior length of the adaxial expression is shorter in class 3, 4 and 5 embryos than in wild-type embryos (Fig. 3E-H), in parallel to the shortened notochord region. The mediolateral expansion of myoD in the mutants may be due to a reduced amount of dorsal convergence, rather than an enlarged population of adaxial cells. In class 2 (laf) mutant embryos the adaxial myoD expression looks normal at the bud stage (data not shown).

At the 5-6 somite stage, myoD begins to be expressed more extensively in the somitic mesoderm (Weinberg et al., 1996) and is dramatically expanded in mutant embryos, consistent with the morphology seen in live embryos. In class 5 mutants (swr and sbn) all somites extend around the circumference of the embryo (Fig. 3M,N,P, data not shown). The first 2 somites in class 4 mutant embryos (snh) extend more laterally than in wild type, but do not reach the ventral side. However, somites posterior to somite 2, do extend around the circumference (Fig. 3Q,R). In class 3 embryos (pgy) the first 5 somites extend about twice the length laterally as the equivalent wild-type somites (Fig. 3S,T), while in the case of class 2 mutant embryos (laf), myoD expression is broadened only in the tail (Fig. 3I,J). Since the anterior somitic mesoderm is derived from laterally positioned cells in the blastoderm (Kimmel et al., 1990; Shih and Fraser, 1995), a reduced amount of dorsal convergence cannot be the cause of the ventrally located somites in class 5 and 4 (swr, sbn and snh) mutant embryos. The expanded somitic myoD expression in these mutants is likely due to an enlarged population of somitic mesodermal cells. For class 2 and 3 mutants (laf and pgy), determination of the number of somitic mesodermal cells is necessary to assess whether this popula-tion of cells is enlarged.

We investigated whether the expansion of the dorsal-lateral mesoderm is accompanied by a reciprocal reduction of cell types normally derived from ventral regions. We examined the expression of the gata1 gene (Detrich et al., 1995), normally expressed in the blood precursors, mesodermal cells known to arise from a ventral region in the zebrafish blastoderm fate map (Kimmel et al., 1990; Lee et al., 1994; Stainier et al., 1993). At the 8 somite stage, gata1 is expressed in 2 lateral stripes that fuse in a ventral position, just posterior to the tailbud (Fig. 4A,B). In mutant embryos gata1 expression is absent or reduced. In all class 5 (swr) and most class 4 (sbn and snh) mutant embryos examined, no gata1 expressing cells were detectable (Fig 4C-F, data not shown). However, a small domain of expression is sometimes observed in class 4 mutant embryos (Fig 4G, data not shown). In class 3 (pgy) mutants, the gata1 expression domain is shortened along its anterior-posterior length (Fig. 4H), while the width of the expression domain is normal, indicating a reduced number of cells expressing gata1 in class 3 embryos. The gata1 expression results are consistent with the absence or reduction of blood found in these mutant embryos in a histological stain (data not shown). In class 2 (laf) mutant embryos, gata1 expression looks normal at the 8 somite stage (Fig. 4I).

The expression of eve1, a marker of ventral and lateral regions at early gastrula stages (Joly et al., 1993), is reduced in class 5 and 4 mutant embryos (swr and snh). At shield stage eve1 expression is strongly reduced in class 5 mutant embryos (Fig. 4J,K). In class 4 mutant embryos a moderate reduction in eve1 is apparent at 70% epiboly (Fig. 4L,M). The extent of the expression domain appears normal in both class 4 and 5 embryos, but the amount of expression is decreased. It appears that the level expressed in dorsolateral regions is extended to the ventral midline. This is consistent with the expansion of anterior somites, which are derivatives of dorsolateral cells of the blastoderm, to ventral positions. A reduction in eve1 expression is also seen in class 3 (pgy) and class 2 (laf) mutant embryos at the 5 somite stage (data not shown).

In Xenopus, signals involved in the dorsalization of the mesoderm also play a role in neural induction, while signals ventralizing the mesoderm repress neural induction processes (Hawley et al., 1995; Sasai et al., 1995; Wilson and Hemmati-Brivanlou, 1995). In order to investigate whether the expansion of dorsal fates in the mesoderm is linked to an enlargement of the presumptive neural region, we examined the expression of fkd3 and pax2.

fkd3 is expressed dorsally in the presumptive neuroectoderm at early gastrulation stages (J.O., unpublished) (Fig. 4N). In class 5 (swr) mutant embryos fkd3 expression is no longer limited to dorsal regions at 70% epiboly, but instead encircles the circumference of the embryo (Fig. 4O). In class 4 (snh) embryos the domain is expanded to ventrolateral regions (Fig. 4P). These results clearly show an expansion of neuroectoderm in class 4 and 5 mutant embryos.

In wild-type embryos at the 10 somite stage, pax2 (formerly paxb) is expressed in a variety of cell types, which include the optic stalk, parts of the optic vesicle, the otic placode, the posterior part of the midbrain, isolated neuronal precursors of the hindbrain and spinal cord, and the ventrolateral mesodermal progenitors of the pronephros (Fig. 4Q; Krauss et al., 1991a). Strikingly, in 10 somite stage class 5 mutant embryos (swr) the midbrain expression domain of pax2 extends around the entire circumference of the embryo (Fig. 4R). The optic vesicle expression is present in the middle of the ring of pax2 staining; it does not lie superficially as in wild type, but deep in a mass of tissue. This is consistent with morphological observations at this stage where the optic placodes come to lie in a ball of tissue at the animal pole (data not shown). The otic vesicle, pronephric, and neuronal precursor expression are not visible in class 5 mutant embryos (Fig. 4R). However, small amounts of sporadic pax2 expression are seen in posterior ventral and ventrolateral regions, which could correspond to any one of these cell types (data not shown).

In class 4 mutant embryos (snh) the optic vesicle expression of pax2 appears normal, however, all the posterior expression domains are altered (Fig. 4S). The midbrain expression is broadened to about 3 times its width in wild type. The presumptive otic vesicle expression of pax2 lying posterior to the midbrain is not visible. Patches of stained cells appear to be spread out laterally over the yolk consistent with the morphology in 2-day old embryos where the otic vesicles are found fragmented on the yolk (data not shown). The two stripes of neuronal staining in the spinal cord are present but lie in more lateral positions than in wild type, probably due to an enlarged neural plate. In addition, more neuronal precursor cells are observed. The pronephric expression of pax2 is reduced to a small patch on the yolk lying just posterior to the twisted axis.

In class 3 (pgy) mutant embryos the anterior pax2 expression domains of the optic vesicle, midbrain, otic placode and neuronal precursors appear normal (Fig. 4T). The pronephric expression domain, however, is altered in two ways. First it is shortened in its anterior-posterior length, but in addition the stripes are broadened mediolaterally in class 3 mutant embryos. It is difficult to interpret the meaning of this result. In class 2 (laf) mutant embryos, pax2 expression appears normal (Fig. 4U).

Several of the genes produce both recessive and dominant phe-notypes, as well as variability in the mutant phenotype.

swr^ta72 shows a dominant zygotic phenotype

swr^ta72 displays dominant class 1 and 2 zygotic phenotypes at varying frequencies (Table 2, Fig. 2J). It is likely that the penetrance of the dominant phenotype is dependent on the genetic background, as in our recent generation of fish that had undergone several outcrosses, almost no dominance is observed. In reciprocal crosses, a maternal influence on the frequency or strength of the dominant phenotype is not obvious. All embryos displaying a dominant phenotype that survived and were fertile, were heterozygous for the mutation (18/18 fish).

A dominant phenotype is also observed in pgy. In this case, the dominant embryonic phenotype is displayed only if both the maternal and zygotic genotypes are heterozygous for the mutation. About 50% of the embryos of crosses between heterozygous females with the strongest pgy allele, pgy^dty40, and wild-type males display class 1 and 2 phenotypes. No dominant phenotype is observed in the progeny of pgy^dty40/+ males crossed to wild-type females, while 50% of the embryos show the dominant phenotype in intercrosses between two heterozygous parents (Table 2). Many of the embryos with weak defects survive to adulthood and of the 36 adults tested, all were pgy^dty40 heterozygotes. A similar genetic behavior of zygotic-maternal dependence is observed in the epiboly mutants described by Kane et al. (1996).

The dominant ventral tail fin phenotypes observed in swr and pgy mutant embryos are likely due to partial loss of function of these genes, as opposed to a gain-of-function defect. This is clearest with pgy for which 6 alleles were found, 3 of which are completely recessive and also show the weakest homozygous phenotypes resulting from the 6 alleles. As the strength of the recessive phenotype increases for the 3 strongest alleles, so does the strength and penetrance of the dominant zygotic-maternal phenotype (Table 2), which appears identical to the recessive phenotypes produced by other pgy alleles. These data along with the gradation in homozygous phenotypes displayed by mutants with the 6 pgy alleles strongly suggest that the dominant phenotypes are a result of a partial loss of function of the pgy gene product. Thus proper development of the embryo is very sensitve to the dosage of the pgy gene product.

A mutation with unique properties, sbn, produces a completely penetrant dominant maternal-effect phenotype, where all progeny of heterozygous females crossed to wild-type males show a class 4 phenotype (Fig. 1I,J,K, Table 2). The progeny of a cross between a heterozygous male and wild-type female also display a weak dominant zygotic class 1 or 2 phenotype of varying penetrance (Table 2). Homozygous sbn embryos develop the strongest phenotype observed in our series (class 5). Due to both the maternal and zygotic dominance of sbn, complementation tests have not been conclusive. Trans-heterozygous swr/sbn embryos show phenotypes stronger than the dominant zygotic, but weaker than the recessive phenotype of swr (Table 2). Interestingly, a cross between a pgy^tc227a/+ female and sbn/+ male shows a moderate class 3 phenotype, stronger than the recessive pgy^tc227a and dominant zygotic sbn phenotype (Table 2). sbn could be a very strong allele of pgy or an unusual allele of swr. Alternatively, sbn may be a gain-or loss-of-function mutation in an independent gene, which interacts with both swr and pgy. Due to the ambiguity in the complementation test results, we tentatively regard sbn as a distinct gene. Placing these mutations on the genetic map (Postlethwait et al., 1994) will clarify whether they are mutations in the same or different genes.

The only snh allele results in a strong recessive phenotype of little variance between embryos within the same egg lay. We observe, however, a variation in the phenotypic strength among individual fish (Table 2, note the class 3 phenotype in one snailhouse cross and class 4 phenotype in two crosses). This is probably due to differences in genetic background.

In our large-scale screen for visible embryonic mutants, we identified a total of 22 mutants displaying expansions of tissues normally derived from dorsolateral cells of the blastoderm at the expense of ventrally derived cell types. These mutants all have similar phenotypes, although the extent to which each particular mutant is affected differs. The phenotypic spectrum observed ranges from a very weak defect restricted to the posterior of the embryo to a dramatic expansion of dorsolater-ally derived structures around the entire circumference of the embryo. The genetic parameters of these mutants indicate that both maternal and zygotic contributions are required for establishing proper dorsoventral pattern formation in the zebrafish embryo and that they are required in a dosage sensitive process.

A partial dorsalization of the embryo is reflected in the reduction of expression of ventrally expressed genes accompanied by the ventrolateral expansion of normally dorsally restricted gene expression. In class 4 and 5 mutant embryos, eve1 expression is decreased and the pax2 pronephric and gata1 blood precursor expression is absent or reduced, demonstrating a reduction of cell types derived from ventral regions of the blastoderm. The paraxial mesoderm is expanded, with the effect on the adaxial cells being less pronounced than on the somitic mesoderm, which extends to the ventral side. Only a slight enlargement of the posterior axial mesoderm is observed, while the region expressing goosecoid, the anterior axial mesoderm, does not appear enlarged. The dorsally derived neuroectoderm is expanded to ventrolateral regions. Although this has to be confirmed with more molecular markers, it appears that there is a progression in the strength of the defects from dorsal to ventral positions, with the most dorsal positions the least affected. The altered expression patterns at very early gastrulation stages indicate that the primary defect in these mutants is in pattern formation, as opposed to morphogenesis. The data together demonstrate a dorsalization phenotype in swr, snh, and sbn mutant embryos (classes 5 and 4) where pattern formation is affected along most of the dorsoventral axis. In pgy, laf and mfn mutant embryos (classes 3, 2 and 1) the dorsalization is less dramatic and restricted to more posterior regions. The phenotypic series of the dorsalized mutants suggests the involvement of a gradient determining position along the dorsoventral axis.

We identified 5 clearly independent genes, swr, snh, pgy, laf and mfn, which are required for specification of ventral regions in the zebrafish embryo. Due to the dominance of sbn and its strong genetic interaction with both pgy and swr, it is difficult to assess its allelism and so we tentatively consider it a sixth gene. Multiple alleles were isolated for swr, pgy, and mfn genes, while only a single allele was found for snh, sbn, and laf. Complementation testing is incomplete for 3 mutations (Table 1). Homozygous mutant swr and sbn embryos show the strongest phenotype (class 5) followed by those of snh (class 4) and pgy (class 3). The weakest defects are seen in laf (class 2) and mfn (class 1) mutant embryos. For none of these genes do we know if a null mutation exists, although it appears that all of the mutations are loss-of-function alleles, with the possible exception of sbn. Three of the genes, pgy, sbn and swr show dominant zygotic pheno-types. These dominant defects, which consist of reductions in ventral tail tissues, are semi-lethal and consequently result in a selective disadvantage in the propagation of the mutation. Thus, null or strong loss-of-function mutations of pgy, sbn and swr may have been selected against in our inbreeding scheme (Mullins et al., 1994) and therefore may be underrepresented in our mutant collection.

A remarkable feature of pgy, and perhaps also sbn, is that it is required both maternally as well as zygotically for dorsoventral pattern formation. Due to the zygotic recessive lethality of the mutations, we were not able to raise homozygous mutant pgy females to look at the phenotype of the more complete loss-of-function of pgy (or sbn). Hence, the mutant phenotypes described are only partial loss-of-function phenotypes and the complete loss-of-function (which can be made by producing chimeric females containing homozygous mutant germline cells (Lin et al., 1992)) would show a more severe phenotype.

LiCl treatment of cleavage stage Xenopus or zebrafish embryos causes a dorsalization of the embryo. One class of LiCl-treated embryos in zebrafish, termed ‘bustled’ (Stachel et al., 1993), displays defects resembling the class 4 phenotype. Similar to these mutants the ‘bustled’ embryos have relatively normal dorsoanterior structures, an apparently enlarged notochord, and a twisted dorsoposterior axis lying above the yolk. During gastrulation some LiCl-treated embryos also resemble the dorsal-ized mutants in their ovoid shape at the bud stage, a posterior reduction in the circumference of the embryo in somite stages, and extension of the tailbud prematurely off the yolk. The defects are not identical, however, as the LiCl-treated embryos, for example, show an expanded domain of goosecoid expression.

The most dorsalized class of LiCl-treated zebrafish embryos, in which multiple dorsal axes are formed (‘radialized’ embryos), was not seen in our mutant collection. It is likely that this is in part due to the residual maternal (and possibly also zygotic) component of pgy or another maternally functioning gene(s) that contributes to dorsoventral patterning. Alternatively, LiCl may disrupt the function of multiple pathways involved in pattern formation of the early embryo, which could not be mimicked by the loss of a single gene product.

The similar and overlapping phenotypes found for mutations in these 6 genes suggest that they function in the same pathway. swr and sbn function early in the embryo and phenotypic effects are visible as early as the shield stage. snh and pgy function by late gastrulation stages, while laf and mfn may be acting only late in the dorsoventral patterning of the tail, along with the other genes. All these genes are candidates for a ventralizing molecule that opposes the dorsalizing signal emanating from the Spemann organizer. The phenotypes of these mutants are those expected from a mutation in a component of a ventralization pathway. In the dorsalized mutant embryos, the activity of the dorsal signal would extend to more ventral regions due to a loss of its repression by a ven-tralizing factor. Possibly, a graded activity emanating from the Spemann organizer could be repressed by a nongraded or graded ventralizing molecule. The products of the 6 genes identified here could be the ventralization factor itself or could be involved in generating or executing its function.

A reduction in a single factor in swr and snh could account for both the expansion of dorsal mesoderm and neuroectoderm observed in these mutant embryos. In Xenopus, the ectopic expression of BMP-4 leads to the ventralization of dorsal mesoderm and to the repression of neural differentiation and the induction of epidermal cell fates (Dale et al., 1992; Jones et al., 1992; Schmidt et al., 1995a; Sasai et al., 1995; Wilson and Hemmati-Brivanlou, 1995), while a dominant negative form of BMP-4 has the opposite effect (Hawley et al., 1995). In swr mutant embyros the neural plate extends around the circumference of the embryo and in snh it is expanded laterally. The inductive and repressive properties of BMP-4 in Xenopus would predict that the expansion in neuroectoderm in swr and snh occurs at the expense of epidermis. This has not yet been tested. Several cloned genes are candidates for the genes described here. wnt8, which has ventralizing effects in Xenopus embryos when expressed after the midblastula stage (Christian and Moon, 1993), and eve1, are expressed zygotically in ventral and lateral regions (Christian and Moon, 1993; Joly et al., 1993) and are candidates for swr, snh, laf, and mfn. In Xenopus, BMP-4, the BMP receptor (Graff et al., 1994), and gsk-3 (He et al., 1995; Pierce and Kimelman, 1995) and wnt8 in zebrafish (Kelly et al., 1995) are expressed maternally and zygotically making them candidates for sbn and pgy and possibly also swr, snh, laf, and mfn. The maternal components of pgy and sbn make them candidates for the blastula stage vegetal signal that induces ventral mesoderm (the ventral counterpart of the Nieuwkoop center). BMP-4 and gsk-3 have ventral mesoderm-inducing abilities (He et al., 1995; Pierce and Kimelman, 1995) and thus are candidates for this signal and for pgy and sbn. Isolation of the zebrafish homologues of these genes, followed by a linkage analysis between the genes and the mutations will identify if any of these cloned genes correspond to the genes described here. This work is currently in progress.

Two genes we identified in our large-scale screen, dino and mercedes, result in ventralized phenotypes reciprocal to those found in dorsalized mutant embryos (Hammerschmidt et al., 1996). dino mutant embryos have a reduced head and trunk, an increased number of blood cells, an enlarged tail, and multiple ventral tail fins. These defects are in striking contrast to the dorsalized phenotypes described here, suggesting that the 2 groups of genes function antagonistically. Double mutant combinations between the ventralized and dorsalized mutants will reveal the relationships between these 2 groups of genes in dorsoventral pattern formation.

In Drosophila at least 7 genes function zygotically in setting up the dorsoventral axis. Two of these genes, short gastrulation (sog) and decapentaplegic (dpp), function antagonistically to each other in establishing pattern in the dorsal region of the Drosophila embryo (Ferguson and Anderson, 1992; Wharton et al., 1993; François et al., 1994). Related genes to sog (François et al., 1994) and dpp (Padgett et al., 1987) in Xenopus, chordin (Sasai et al., 1994; François and Bier, 1995) and BMP-4, respectively, function in an opposite manner to their Drosophila counterparts: chordin is a dorsalizing factor (Sasai et al., 1994), while BMP-4 is ventralizing. dpp has been shown to function in a similar manner to BMP-4 in promoting ventral cell fates in ectopic expression experiments in Xenopus (Holley et al., 1995). Likewise, human BMP-4 can substitute for dpp in Drosophila (Padgett et al., 1993). These and other studies have provided molecular evidence for an old hypoth-esis that the dorsoventral axis is inverted between vertebrates and arthropods (Nübler-Jung and Arendt, 1994; Holley et al., 1995; Schmidt et al., 1995b; Hogan, 1995; De Robertis and Sasai, 1996).

In Drosophila dpp acts as a morphogen in establishing the positional information in the dorsal half of the embryo. Highest activities determine the most dorsal cell fates, intermediate levels dorsolateral cell types, and the lowest activities are necessary for lateral cell fate specification (Ferguson and Anderson, 1992; Wharton et al., 1993). dpp function is very sensitive to its gene dosage. Null alleles of dpp show dominant zygotic lethality, producing embryos with a weakly ventralized phenotype (Irish and Gelbart, 1987). The parallels between the loss-of-function defects of dpp in Drosophila and those of the dorsalized mutants in zebrafish are worth noting. Both display a phenotypic series, show dominant defects and are extremely sensitive to gene dosage. It is tempting to speculate that BMP-4 functions in a similar manner in vertebrates to dpp in Drosophila and that the genes we identified here are functional components of a pathway where BMP-4 acts as a morphogen to establish ventral positional information in the zebrafish embryo.

We would like to thank: Eric Weinberg, and David Ransom and Leonard Zon for providing the myoD and gata1 cDNA clone, respectively, prior to publication; David Ransom for pointing out the histological blood staining method; J. S. Joly for the eve1 cDNA clone; Mary Ellen Lane, Siegfried Roth, Stefan Schulte-Merker, Herbert Steinbeiser for helpful comments on the manuscript; and very special thanks to Karin Finger-Miller for technical support, as well as to Hans-Martin Maischein, Amanda Wilson, Jörg Zeller, and Cosima Fabian. This work was supported by an NIH postdoctoral fellowship to M. C. M.