INTRODUCTION

During vertebrate gastrulation, cell movements form three distinct germ layers and elaborate the body axes. In pregastrula stage teleost and amphibian embryos, precursors of the three germ layers are distributed at characteristic positions along the animal-vegetal axis. In zebrafish, for example, the cells near the blastoderm margin at the late blastula stage form mesoderm and endoderm (mesendoderm), and are the first to involute at the onset of gastrulation (Kimmel et al., 1990; Warga and Nüsslein-Volhard, 1999). Cells slightly farther from the margin involute at later stages and exclusively become mesodermal cell types. In more animal regions, cells adopt ectodermal fates and do not involute. Each germ layer is patterned along the dorsoventral axis, generating the diverse array of cell types characteristic of the vertebrate body plan. The initial animal-vegetal and dorsoventral asymmetries are established by maternal transcription factors, which regulate zygotic genes controlling cell fate specification and morphogenesis (Harland and Gerhart, 1997; Heasman, 1997; Moon and Kimelman, 1998; Schier and Talbot, 1998; De Robertis et al., 2000; Kimelman and Griffin, 2000; Schier, 2001). Although zygotic factors that can induce and pattern mesoderm have been identified, significant questions remain unanswered about how the mesoderm is patterned.

nodal-related genes encode zygotically acting TGFβ family proteins that are necessary for induction of mesoderm and endoderm (Conlon et al., 1994; Jones et al., 1995; Joseph and Melton, 1997; Schier and Shen, 2000; Takahashi et al., 2000; Schier and Talbot, 2001; Whitman, 2001). Mouse nodal mutant embryos lack a primitive streak and fail to form mesodermal derivatives (Conlon et al., 1994; Varlet et al., 1997). In zebrafish, there are two known nodal-related genes, squint (sqt; ndr1 — Zebrafish Information Network) and cyclops (cyc; ndr2 — Zebrafish Information Network) (Erter et al., 1998; Feldman et al., 1998; Rebagliati et al., 1998a; Rebagliati et al., 1998b; Sampath et al., 1998). Whereas defects in sqt and cyc single mutants are largely confined to dorsal axial structures (Hatta et al., 1991; Thisse et al., 1994; Heisenberg and Nüsslein-Volhard, 1997; Feldman et al., 1998; Warga and Nüsslein-Volhard, 1999), almost all mesendodermal derivatives are absent in sqt; cyc double mutants, including notochord, trunk somites, pronephros, heart, blood and gut (Feldman et al., 1998). Thus, these nodal-related genes have both overlapping and essential functions in mesendoderm development. Cell tracing experiments in sqt; cyc double mutants and in maternal-zygotic one-eyed pinhead (oep) (MZoep) mutants, which are completely unresponsive to Nodal signals (Gritsman et al., 1999), indicate that Nodal signals allocate marginal cells to mesendodermal fates (Feldman et al., 2000; Carmany-Rampey and Schier, 2001).

Although it is now firmly established that Nodal signals induce mesendoderm, a possible role for Nodal signals in specifying different mesendodermal fates along the dorsoventral axis is still controversial. In Xenopus, three classes of models have been proposed for the role of activin-like ligands, a group that includes Activin and Nodal signals, in mesoderm patterning (Harland and Gerhart, 1997; Heasman, 1997; McDowell and Gurdon, 1999; De Robertis et al., 2000; Kimelman and Griffin, 2000). One group of models proposes that a gradient of activin-like signals patterns the mesoderm along the dorsoventral axis. This view is supported by explant experiments in which low doses of activin-like signals induced ventrolateral mesodermal fates and high levels induced dorsal fates (Smith et al., 1988; Ruiz i Altaba and Melton, 1989; Green et al., 1992; Agius et al., 2000). Further support for graded action of nodal-related genes comes from analysis of the distribution of phosphorylated Smad2, a proposed transcriptional effector of Nodal signals that is initially elevated in the dorsal marginal region in Xenopus blastulae (Faure et al., 2000). In addition, Nodal inhibitors block mesoderm formation in different dorsoventral positions in a dosage-dependent manner when overexpressed in Xenopus embryos, consistent with asymmetric action of endogenous Nodal signals (Agius et al., 2000).

The second class of models proposes that activin-like signals act uniformly along the dorsoventral axis to induce mesoderm in the marginal region, while independent signals generate dorsoventral pattern (Christian et al., 1992; Kimelman et al., 1992; Clements et al., 1999). This view is supported by experiments with a synthetic activin/TGF-β responsive reporter, which found that the transcriptional output from these signals is uniform along the dorsoventral axis at the beginning of gastrulation (Watabe et al., 1995). Furthermore, promoter analysis indicates that input from both Wnt and activin-like signaling pathways is required for proper expression of dorsal-specific genes, such as goosecoid and siamois (Watabe et al., 1995; Crease et al., 1998).

Recent work (Lee et al., 2001) supports a third model emphasizing the dynamic action of activin-like signals. The spatiotemporal distribution of phosphorylated Smad2 suggests that these signals are active predominantly in dorsal regions in the late blastula, and that the activity shifts to ventral regions as gastrulation progresses. In addition, activin-like ligands acting in dorsal regions elicit responses at earlier stages than ventrally acting ligands (Lee et al., 2001; Schohl and Fagotto, 2002). These results suggest that dorsoventral patterning involves distinct temporal responses to activin-like signals in dorsal and ventral cells.

Drawing parallels from the work on activin-like signals in Xenopus, analysis of Nodal pathway mutants has suggested at least two possible models of Nodal function in zebrafish mesendoderm patterning. The first model proposes graded action of Nodal signals along the dorsoventral axis, with high levels inducing dorsal mesendoderm and low levels inducing ventrolateral mesendoderm (Fig. 1A). In support of this view, dorsal axial structures are reduced in mutants with impaired Nodal signaling, such as sqt, cyc and schmalspur (sur) single mutants (Hatta et al., 1991; Heisenberg and Nüsslein-Volhard, 1997; Feldman et al., 1998; Pogoda et al., 2000; Sirotkin et al., 2000a). The expression patterns of sqt and cyc are consistent with an elevated dorsal requirement for nodal-related gene function. Soon after the onset of zygotic transcription, sqt is expressed specifically in the dorsal marginal region, where dorsal mesendoderm originates, and cyc is strongly expressed in the axial mesendoderm during gastrulation (Erter et al., 1998; Feldman et al., 1998; Rebagliati et al., 1998a). Moreover, dorsal mesoderm is most sensitive to overexpression of lefty1 (also known as activin), an inhibitor of activin-like signals (Bisgrove et al., 1999; Thisse and Thisse, 1999; Meno et al., 1999; Thisse et al., 2000). Finally, overexpression of high levels of sqt or cyc induces presumptive ectodermal cells to initiate dorsal mesendodermal gene expression, while lower levels induce pan-mesodermal but not dorsal mesodermal markers (Erter et al., 1998; Sampath et al., 1998; Chen and Schier, 2001). Models in which Nodal activity gradients pattern the dorsoventral axis predict a transformation of cell fates along the dorsoventral axis as Nodal dose is lowered (one scenario is depicted in Fig. 1A, right), but this prediction has not yet been tested by fate-mapping experiments.

The second model proposes that Nodal signals act uniformly along the dorsoventral axis to induce mesendoderm, while independent signals instruct fates along the dorsoventral axis (Fig. 1B). In support of this possibility, dorsoventral asymmetry is established even in the absence of Nodal signals. For example, the dorsal genes chordin and bozozok (dharma — Zebrafish Information Network) are expressed in sqt; cyc double mutants and in MZoep mutants (Gritsman et al., 1999; Shimizu et al., 2000; Sirotkin et al., 2000b). The sqt and cyc genes are expressed uniformly along the dorsoventral margin at the late blastula stage (Erter et al., 1998; Feldman et al., 1998; Rebagliati et al., 1998a; Sampath et al., 1998), consistent with an equivalent requirement in dorsal and ventrolateral regions. Additional evidence derives from the analysis of zygotic oep (Zoep) mutants (Schier et al., 1997), in which the response to Nodal signals is impaired but not eliminated (Gritsman et al., 1999). Zoep mutants lack endodermal derivatives from both the dorsal and ventral margin. Fate mapping of the dorsal marginal region in Zoep mutants indicates that the marginal-most cells (prechordal plate precursors) are transformed to a slightly more animal fate (notochord precursors), suggesting that different levels of Nodal signaling distinguish these cell fates (Gritsman et al., 2000). Similarly, low levels of exogenous antivin block endoderm but not mesoderm formation (Thisse and Thisse, 1999; Thisse et al., 2000). If Nodal signals are required uniformly around the margin, reductions in Nodal levels should shift all marginal derivatives toward more animal fates (Fig. 1B, right panels) — a prediction that has not been tested by fate mapping cells in ventrolateral regions.

To test these models and to explore possible stage-specific requirements for nodal-related genes, we have assessed mesendodermal patterning and fate specification in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- embryos, in which the action of endogenous Nodal signals is reduced but not eliminated. Our results indicate that Nodal signals act in the marginal region to pattern the animal-vegetal axis and that ventrolateral mesendodermal fates can be induced by a lower level of nodal-related gene function than dorsal mesendoderm. Furthermore, we find that differential regulation of the cyclops gene in dorsal and ventrolateral cells contributes to the different requirements for nodal-related gene function in these cells. In addition, our analysis shows that dorsal mesendodermal precursors are competent to respond to Nodal signals over a surprisingly long period, ranging from late blastula through at least early gastrula stages.

MATERIALS AND METHODS

Mutant alleles and photography

sqt^cz35 and cyc^m294are presumed null alleles that have been described previously (Feldman et al., 1998; Sampath et al., 1998). Embryos were collected from natural crosses and staged as described (Kimmel et al., 1995). Embryos shown in Fig. 4D and Fig. 2A-O were derived from crosses of sqt^cz35/+; cyc^m294/+ × sqt^cz35/+; cyc^m294/+ parents. After photography, DNA for genotype analysis was extracted from living embryos as described by Gates et al. (Gates et al., 1999), or from fixed and stained embryos as described by Sirotkin et al. (Sirotkin et al., 2000b). For analysis of gsc, embryos with defective gsc expression at each stage were counted and a representative sample was photographed; after photography, DNA was extracted for genotyping as previously described. All other mutant embryos, except the sqt mutants shown in Fig. 9G-J and Fig. 3E,J were derived from repeated crosses of a small group of sqt^cz35/+; cyc^m294/+ × sqt^cz35/+ parents, in which the majority of sqt^cz35/sqt^cz35; cyc^m294/+ progeny consistently had a truncated notochord and fused somites, as shown in Fig. 2P-U. For quantitation of foxa2/axial expression at 8 hours post-fertilization (h), stained embryos were photographed as wholemounts in canada balsam: methyl salicylate (11:1), rehydrated in PBT (1× PBS and 0.1% Tween-20), and equilibrated in 80% glycerol. The embryos were then dissected along the ventral midline, photographed under low magnification, and DNA was extracted to determine the genotype by PCR.

Primers and conditions for genotyping sqt^cz35 and cyc^m294 have been described (Feldman et al., 1998; Sampath et al., 1998). In crosses with no cyc homozygotes, we used a primer pair that amplified a fragment specific to the cyc^m294 allele to distinguish cyc^+/+ and cyc^m294/+ embryos: forward, 5′-GGTGGACATGCATGTGGATTT-3′ and reverse, 5′-TCGGGCAGGCCCCCTCCCG-3′.

mRNA synthesis and embryo injection

The zβ-catenin expression vector has been described previously (Kelly et al., 1995). Transcripts for injection were synthesized using the Message Machine kit (Ambion). Wild-type embryos were injected with 100 pg β-catenin mRNA, while embryos from an intercross of sqt^cz35/+ parents were injected with 500 pg β-catenin mRNA. In this experiment (Fig. 9G-J), mutant genotypes were inferred from phenotypes rather than assessed by PCR. In the control injection 24% (12/49) of the embryos had reduced gsc expression typical of sqt mutants (Fig. 9J), whereas the remaining 76% (37/49) displayed normal gsc (Fig. 9H). β-catenin overexpression induced ectopic or expanded gsc (Fig. 9G) in over half of the wild-type embryos (46/77). β-catenin did not induce high levels of gsc expression in sqt mutants, because 26% (16/62) of injected embryos displayed reduced gsc typical of uninjected sqt mutants (Fig. 9I).

Immunocytochemistry and in situ hybridization

For analysis of the distribution of β-catenin protein, wild-type embryos were fixed at 15 minute intervals during the first 4 hours of embryogenesis, processed for immunocytochemistry with β-catenin antibodies and sectioned. Polyclonal anti-β-catenin antibodies (Schneider et al., 1996) were used at a 1:1000 dilution. Immunostaining was carried out as described (Schier et al., 1997). For sectioning, embryos were embedded in Eponate-12 resin and sectioned at 3 μ m. Twenty embryos were analyzed at each time-point, and two or three of these were sectioned to confirm the presence or absence of nuclear β -catenin. Although the majority of embryos after the 1000-cell stage exhibited β-catenin protein in dorsal nuclei (in both the YSL and blastomeres), only three embryos at the 128-cell stage, three embryos at the 256-cell stage and three embryos at the 512 stage exhibited β-catenin protein in dorsal nuclei. At all stages, staining of membrane localized β -catenin protein served as a positive control for the antibody reaction.

Synthesis of probes and in situ hybridization were conducted as described (Sirotkin et al., 2000b). After in situ hybridization, the number of mutant embryos from each cross was counted and representative examples of each phenotype were photographed and genotyped by PCR, except when the genotype could be inferred from morphology as in the analysis of gsc at 10 h in sqt^-/-; cyc^+/- and sqt^-/-; cyc^-/- embryos and sox17 expression in sqt^-/-; cyc^+/- mutants. The number of mutants was usually close to the expected values, and the same mutant phenotypes were observed in repeated crosses of the same parents.

Lineage-tracing and fete map analysis

Embryos of sqt^cz35/+; cyc^m294/+ × sqt^cz35/+ parents were injected (Kimmel et al., 1990) with the lineage tracer dye tetramethylrhodamine-isocyanate dextran (Molecular Probes, Eugene, OR; 10×10³ M_r, diluted to 5% (wt/vol) in 0.2 M KCl) in single blastomeres of the surface enveloping layer (EVL) at the 1000-cell stage. Injected embryos were mounted in 0.125% agarose in embryo medium and oriented so that the fluorescent cells faced toward the microscope objective; the position of the clone along the animal-vegetal axis was then determined. The clonal position relative to the dorsoventral axis was determined at 6 h, when the dorsal side is first morphologically apparent in wild-type embryos, or at 8 h, when the dorsal side is morphologically apparent in sqt^-/-; cyc^+/- embryos. Some embryos were filmed during gastrulation with time-lapse photography as previously described (Warga and Kimmel, 1990). After filming, embryos were removed from agarose and the fates of the resulting clones were determined at 24 h and 48 h, after which all embryos were genotyped by PCR.

RESULTS

Genetic interaction between nodal-related genes reveals dosage-sensitive function of cyclops

To understand the role of Nodal signaling in patterning the mesendoderm, we analyzed the phenotypes of embryos with reduced nodal gene function, including sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants. At 6 h, sqt^-/-; cyc^+/+ single mutants lack a morphologically visible embryonic shield, the site of presumptive dorsal mesendoderm (Table 1) (Feldman et al., 1998). Kupffer's vesicle, a derivative of the dorsal forerunner cells (Cooper and D'Amico, 1996; Melby et al., 1996), was reduced or absent at 12 h (Fig. 2T). Forerunner cells were also reduced or absent in sqt^-/-; cyc^+/+ mutants (see Fig. 5J-L, arrows), as revealed by expression of sox17 (Alexander and Stainier, 1999). Thus, most or all sqt^-/-; cyc^+/+ single mutants have defects in dorsal marginal structures at the onset of gastrulation. At 24 h, a variable fraction of sqt mutants display mild cyclopia and reduction of ventral diencephalon (Table 1; Fig. 2I) and some sqt^-/-; cyc^+/+ single mutants are morphologically indistinguishable from wild type (Table 1; data not shown). The amelioration of the sqt mutant phenotype at later stages depends on the function of cyc, because sqt; cyc double mutants lack all mesendodermal derivatives in the head and trunk at 24 h (Fig. 2A-C) (Feldman et al., 1998).

To search for possible dosage-sensitive functions for squint and cyclops, we compared the phenotypes of sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- embryos, which differ by a single copy of cyc. Like sqt^-/-; cyc^+/+ single mutants, sqt^-/-; cyc^+/- embryos did not form an embryonic shield or Kupffer's vesicle (Table 1; Fig. 2U). At later stages, however, sqt^-/-; cyc^+/- embryos had a significantly stronger phenotype than sqt^-/-; cyc^+/+ embryos (Fig. 2, compare D-F to G-I). For example, the notochord was severely reduced or truncated in most sqt^-/-; cyc^+/- embryos (Fig. 2E), although it formed normally in the majority of sqt^-/-; cyc^+/+ single mutants (Fig. 2H). In sqt^-/-; cyc^+/- embryos with truncated notochords, the somites were often fused across the midline (Fig. 2R), as were adaxial and paraxial domains of myod expression (Weinberg et al., 1996) (Fig. 6A-F). In addition, sqt^-/-; cyc^+/- embryos have more pronounced cyclopia than sqt^-/-; cyc^+/+ embryos (Fig. 2F,I), suggesting a greater reduction in prechordal plate mesoderm. The severity of these defects varies with the genetic background, but within a group of siblings the typical sqt^-/-; cyc^+/- embryo has a stronger phenotype than the typical sqt^-/-; cyc^+/+ sibling. These results indicate that the formation of dorsal axial structures in sqt mutants is strongly dependent on cyc gene dosage, such that the phenotype of sqt^-/-; cyc^+/+ embryos is compounded by the loss of a single copy of cyc (i.e. in sqt^-/-; cyc^+/- embryos). Conversely, we did not note any differences between the phenotypes of sqt^+/+; cyc^-/- and sqt^+/-; cyc^-/- embryos (Fig. 2J-L; Table 1; data not shown).

Dorsal expression of cyc is activated by sqt and by autoregulation

Maintenance of sqt and cyc expression depends on the ability of cells to respond to Nodal signaling (Meno et al., 1999; Pogoda et al., 2000; Sirotkin et al., 2000a). Therefore we investigated the role of crossregulation in the dosage-sensitive interaction between sqt and cyc by examining cyc expression in sqt mutants and sqt expression in cyc mutants. Expression of sqt appeared normal in a time course of cyc mutant embryos (compare Fig. 3F-I with 3A-D), but was reduced in sqt mutants (Fig. 3E,J). cyc expression was altered in sqt^-/-; cyc^+/+ mutants (compare Fig. 3L-O with 3P-S). Although cyc expression initiated in sqt mutants (Fig. 3P), it was reduced or absent at the dorsal margin at 5 h (Fig. 3Q) and was evident at the dorsal midline at reduced levels during gastrulation (Fig. 3R,S). Expression of cyc was reduced significantly in sqt^-/-; cyc^+/- mutants (Fig. 3T-V) in comparison with sqt^-/-; cyc^+/+ embryos (Fig. 3Q-S). At the onset of gastrulation in sqt^-/-; cyc^+/- mutants, cyc transcripts were present in only two weak dorsolateral patches (Fig. 3U); the dorsal gap between these patches closed as gastrulation progressed, such that a narrow strip of cyc expression was visible at the dorsal midline at 8 h (Fig. 3V). These results show that dorsal cyc expression is activated by sqt at 5 h, the first stage where a reduction of cyc expression is evident in sqt mutants. The reduction of cyc expression in sqt^-/-; cyc^+/- embryos when compared with sqt^-/-; cyc^+/+ mutants suggests that cyc expression also activates its own transcription. It is not likely that a reduction of mutant mRNA stability is the cause of reduced cyc expression in sqt^-/-; cyc^+/- mutants, because the level and pattern of cyc mRNA expression was not altered in sqt^+/+; cyc^+/- embryos when compared with wild type (data not shown).

Mesendoderm induction requires different levels of nodal-related gene function in dorsal and ventrolateral regions

To examine the effect of reduced Nodal signaling on dorsoventral patterning of the mesendoderm, we analyzed the time course of expression of an array of marker genes in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants. For these experiments, we analyzed clutches of embryos in which most sqt^-/-; cyc^+/- mutants had strong cyclopia, truncated notochords, and medially fused somites. Embryos from appropriate crosses were harvested at blastula and gastrula stages, analyzed by whole-mount in situ hybridization, assigned to phenotypic classes based on expression patterns, and then genotyped by PCR assays to determine which genotypes constituted each phenotypic class. As expected from morphological analysis, markers for dorsal mesendoderm such as gsc and axial/foxa2 were reduced more severely in sqt^-/-; cyc^+/- embryos than in sqt^-/-; cyc^+/+ single mutants. Expression of gsc, which marks the prechordal plate during gastrulation (Stachel et al., 1993), initiated in the dorsal marginal region in sqt^-/-; cyc^+/+, sqt^-/-; cyc^+/- and sqt^-/-; cyc^-/- embryos (Fig. 4B1,C1,D1), but it rapidly faded in embryos of all three mutant genotypes (Fig. 4B2,C2,D2). By 5 h (40% epiboly), gsc transcripts were prominent at the dorsal margin of wild-type embryos (Fig. 4A2), but were detectable in only a few cells in sqt^-/-; cyc^+/+ mutants (Fig. 4B2). We were unable to detect gsc expression in sqt^-/-; cyc^+/- and sqt^-/-; cyc^-/- embryos at this stage (Fig. 4C2,D2). gsc expression recovered during gastrulation in sqt^-/-; cyc^+/+ mutants (Fig. 4B3-B6), but not in sqt^-/-; cyc^+/- and sqt^-/-; cyc^-/- mutant siblings (Fig. 4C3-C5,D3-D5). After gastrulation, weak ectodermal gsc expression initiated in both sqt^-/-; cyc^+/- and sqt^-/-; cyc^-/- mutants (Fig. 4C6,D6). Like gsc, expression of axial/foxa2 in axial mesoderm (Strähle et al., 1993) is reduced to a much greater extent in sqt^-/-; cyc^+/- embryos than sqt^-/-; cyc^+/+ siblings (Fig. 5A-F). And like gsc, axial/foxa2 expression at the midline later recovers in sqt^-/-; cyc^+/+ embryos (data not shown). Thus, early dorsal mesendodermal gene expression is reduced in sqt^-/-; cyc^+/+ single mutants, and the further loss of a single copy of the cyclops gene compounds this defect. The recovery of dorsal mesoderm gene expression in some sqt mutants suggests that dorsal marginal cells can adopt dorsal mesendoderm fates in response to cyc signaling during gastrulation, more than 2 hours after dorsal mesendoderm induction occurs in wild type.

To examine how the endoderm is affected in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants embryos, we analyzed expression of axial/foxa2 and sox17 (Fig. 5A-L) (Schier et al., 1997; Alexander and Stainier, 1999). In order to quantify endodermal progenitors, we made fillets of mid-gastrula stage wild-type and mutant embryos stained for axial/foxa2 expression and counted the total number of labeled endodermal cells (Fig. 5M). At mid-gastrulation, wild-type embryos (n=4) had an average of 270 (±54) axial/foxa2-expressing endodermal cells, sqt^-/-; cyc^+/+ embryos (n=5) had an average of 164 (±35), and sqt^-/-; cyc^+/- embryos (n=5) had only 65 (±12). A similar distribution of endodermal cells was observed in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants at 10 h and 12 h, as detected by sox17 expression (data not shown). The reduction in endodermal precursors was not uniform along the dorsoventral axis (Fig. 5N). In wild-type embryos, previous work showed that endoderm is asymmetrically distributed at mid-gastrulation (Warga and Nüsslein-Volhard, 1999), and we found an average of 13 rows of axial/foxa2-expressing endodermal cells dorsally, compared with only four rows ventrally (Fig. 5N). This asymmetry was less pronounced in sqt^-/-; cyc^+/+ embryos, which had an average of only seven tiers of axial-expressing cells dorsally and 2.5 tiers ventrally. Dorsal endoderm was completely eliminated in sqt^-/-; cyc^+/- embryos, although one to three tiers of axial/foxa2-expressing cells remained ventrally and laterally. Thus, like mesoderm, endoderm in dorsal locations in sqt mutants is more sensitive to reductions in cyc gene dosage than endoderm in lateral and ventral positions.

We next investigated whether the loss of dorsal mesendodermal markers is accompanied by a corresponding alteration in ventrolateral gene expression (Fig. 6). The ventrolateral mesoderm markers spadetial (spt; tbx16 — Zebrafish Information Network), tbx6 and vox (Hug et al., 1997; Griffin et al., 1998; Kawahara et al., 2000; Melby et al., 2000) are each expressed around the margin but are largely excluded from dorsal marginal regions (Fig. 6G,J,M,P). Ventrolateral expression of these genes was not significantly altered in sqt^-/-; cyc^+/+ or sqt^-/-; cyc^+/- mutant embryos (Fig. 6H,K,N,Q), except that each of these genes was excluded from a larger dorsal sector in sqt^-/-; cyc^+/- embryos (Fig. 6I,L,O,R). The different effects on dorsal and ventrolateral mesendodermal gene expression in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants suggests that the formation of mesoderm and endoderm requires higher levels of cyc function at the dorsal margin than at the ventrolateral margin.

Dorsal marginal cells adopt neural fates in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants

As dorsal marginal cells fail to express markers for both dorsal and ventrolateral mesendoderm in sqt^-/-; cyc^+/- mutants, we wished to determine the fates of dorsal marginal cells in these embryos. Expression of the pan-mesodermal marker ntl/brachyury (Schulte-Merker et al., 1992), was reduced on the dorsal side of sqt^-/-; cyc^+/+ embryos (Fig. 6T) and absent from a dorsal sector in sqt^-/-; cyc^+/- embryos (Fig. 6U), raising the possibility that dorsal marginal cells adopt a neurectodermal fate in sqt^-/-; cyc^+/- mutants. Accordingly, the neural plate marker cyp26 (White et al., 1996; Thisse and Thisse, 1999; Kudoh et al., 2001) was positioned closer to the margin in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants than in wild type (Fig. 6W,X, arrowheads).

The gene expression studies prompted us to determine directly the fates of marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants. We labeled single blastomeres at the 1000-cell stage by injection of a lineage tracer dye and ascertained the fates of their progeny between 1 and 3 d, when most cells have differentiated. Dorsal cells closest to the margin in wild-type embryos exclusively form mesodermal and endodermal derivatives, including hatching gland, foregut and notochord (Kimmel et al., 1990; Melby et al., 1996; Warga and Nüsslein-Volhard, 1999). By contrast, some dorsal marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants adopt neural fates, including spinal cord, hindbrain and midbrain (Fig. 7B,C upper panels), which derive from more animal regions in wild-type embryos (Fig. 7A). In two cases, mutant dorsal marginal clones exclusively generated neural fates (sqt^-/-; cyc^+/+ n=1; sqt^-/-; cyc^+/- n=1), which is never observed in wild type. In addition, dorsal marginal cells formed foregut endoderm and hatching gland rarely if at all in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants, although these cells can form notochord. As previously reported (Feldman et al., 2000; Carmany-Rampey and Schier, 2001), all dorsal marginal cells adopt neural fates in the absence of Nodal signaling (Fig. 7D).

In none of these experiments did we find evidence that disruption of nodal-related genes shifted the fates of dorsal marginal cells along the dorsoventral axis. There was instead clear evidence that marginal cells adopt fates characteristic of more animal positions. This animal-vegetal fate shift was most apparent at the dorsal margin, where cells adopt neural fates in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants (Fig. 7B,C, upper panels). There is also evidence to suggest an animal-vegetal fate shift at the ventrolateral margin, where muscle precursors are located closer to the margin than in wild type (Fig. 7B,C, middle panels).

Abnormal morphogenesis of dorsal marginal cells in sqt^-/-; cyc^+/- mutants

In addition to changes in cell fate, we observed that dorsal marginal cells undergo an aberrant morphogenetic behavior in sqt^-/-; cyc^+/- embryos, consistent with their altered cell fates (Fig. 7 and Fig. 8B). At the onset of gastrulation, dorsal marginal cells enter the mesendodermal germ layer and rapidly move toward the animal pole, forming the prechordal plate (n=16 clones, Fig. 8A) (Warga and Nüsslein-Volhard, 1999). Instead of undergoing normal involution movements and moving toward the animal pole, dorsal marginal cells in sqt^-/-; cyc^+/- embryos dispersed along the margin. Many dorsal cells in mutants rolled inwards, temporarily altered their course and then continued migrating toward the vegetal pole (Fig. 8B n=3 clones in sqt^-/-; cyc^+/- embryos). Despite their anomalous behavior, some of these dorsal cells were able to contribute to trunk notochord, much as they did in wild type (Fig. 7, Fig. 8A″,B″, arrows). In contrast to dorsal marginal cells, the morphogenetic program of cells on the ventrolateral margin appeared normal in sqt^-/-; cyc^+/- embryos (Fig. 8D, n=33 wild-type and 3 sqt^-/-; cyc^+/- embryos).

β-catenin activates dorsal expression of sqt and cyc

The pronounced reduction of dorsal but not ventrolateral mesendoderm in sqt^-/-; cyc^+/- mutants correlates with the elevated expression of both sqt and cyc at the dorsal margin (Erter et al., 1998; Feldman et al., 1998; Rebagliati et al., 1998a; Sampath et al., 1998). To understand how the asymmetry of sqt and cyc expression is generated, we asked if cyc is controlled by the maternal dorsal determinant β-catenin, which has been implicated as a regulator of sqt and Xenopus xnr-1 (Hyde and Old, 2000; Shimizu et al., 2000). β-catenin mRNA overexpression induced expanded or ectopic cyc expression at 6 h in about half (32/60) of injected wild-type embryos (Fig. 9A), similar to the fraction of injected embryos (12/18) with ectopic or expanded gsc expression at 5 h (Fig. 9G). Although β-catenin did not affect cyc expression prior to 6 h (Fig. 9C-F), it induced ectopic or expanded sqt expression at 3.3 h (79/112) (Fig. 9K), as previously reported (Shimizu et al., 2000). β-catenin had no effect on sqt expression at 5 h (Fig. 9M). Thus, activation of sqt and cyc by β-catenin follow distinct time courses, suggesting that sqt may be a direct target, while cyc could be an indirect target of β-catenin, mediated by genes such as sqt or boz that are induced at earlier stages (Fekany et al., 1999).

Because squint expression is rapidly upregulated by β-catenin, we asked if it is required for β-catenin function. We injected β-catenin mRNA into embryos from sqt^+/- parents and analyzed gsc expression at 5 h (40% epiboly). At this stage, gsc is reduced to only a few cells in control injected and uninjected sqt mutants (Fig. 9J). Although β-catenin mRNA can induce mutant secondary axes in the absence of sqt, as indicated by extra patches of cells expressing low levels of gsc (Fig. 9I, arrow), it cannot induce the high levels of gsc typical in wild-type embryos (Fig. 9G) (Kelly et al., 1995; Pelegri and Maischein, 1998). Thus, β-catenin requires squint function to induce high levels of dorsal mesodermal gene expression. Nevertheless, β-catenin, acting directly or through other factors, is sufficient to induce low levels of ectopic gsc expression in sqt mutants.

Consistent with the possibility that β-catenin activates early sqt expression, we found that β-catenin protein is localized to the nuclei of presumed dorsal blastomeres prior to 3 h (Fig. 10A-C), when sqt transcripts are first detected (Erter et al., 1998; Feldman et al., 1998). These results extend a prior analysis of β-catenin distribution, which reported that β-catenin accumulates in dorsal nuclei at 3.3 h (Schneider et al., 1996). Moreover, we detected nuclear β-catenin in embryos as early as the 128-cell stage (2.25 h) (Fig. 10A), prior to formation of the yolk syncytial layer (YSL), an extra-embryonic structure proposed to have a role in establishing dorsoventral asymmetry in the overlying blastoderm (reviewed by Schier and Talbot, 1998). Nuclear β-catenin was detected in only a small fraction (15%, n=3/20, at the 128-cell stage) of the embryos prior to the 1000-cell stage, perhaps because of intermittent accumulation of nuclear β-catenin during rapid cleavage divisions. Although there are no other markers of dorsoventral polarity in zebrafish embryos younger than 3 h, the dorsal accumulation of nuclear β-catenin after 3 h has been confirmed by colocalization with dorsal specific genes (Koos and Ho, 1998). Therefore, we presume that β-catenin accumulates in dorsal nuclei prior to 3 h, as it does after 3 h. Our results show that dorsoventral asymmetry is established in the blastoderm before the YSL forms.

DISCUSSION

The role of Nodal signals in mesendodermal patterning

We have examined the role of the zebrafish nodal-related genes squint and cyclops in mesendoderm induction and patterning. Using marker gene analysis and cell tracing, we characterized the fates of dorsal and ventrolateral mesendodermal progenitors in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants, in which endogenous Nodal signals are reduced but not eliminated. Our data support a role for nodal-related genes in patterning marginal blastomeres along the animal-vegetal axis. At the dorsal margin, Nodal signals act to prevent dorsal marginal cells from adopting neurectodermal fates. In wild-type embryos, dorsal marginal cells exclusively form mesendodermal derivatives such as hatching gland, notochord and foregut. By contrast, dorsal marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants often gave rise to neural fates, including spinal cord, hindbrain and midbrain (Fig. 7B,C), and neural marker gene expression was shifted toward the margin (Fig. 6V-X). At the ventrolateral margin, expression of endodermal markers was reduced in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants (Fig. 5), and muscle precursors were located closer to the margin than in wild-type embryos (Fig. 7). In addition, endoderm was more severely disrupted in sqt^-/-; cyc^+/- than in sqt^-/-; cyc^+/+ mutants, supporting the proposal that different levels of Nodal signaling establish the mesoderm and endoderm at different positions along the animal-vegetal axis (Gritsman et al., 2000; Thisse et al., 2000; Chen and Schier, 2001). The analysis of sqt^-/-; cyc^+/+ or sqt^-/-; cyc^+/- mutants extends previous fate mapping studies of Nodal pathway mutants (Feldman et al., 2000; Gritsman et al., 2000; Carmany-Rampey and Schier, 2001) by demonstrating a dosage-sensitive role for Nodal genes in patterning the animal-vegetal axis in dorsal and ventrolateral marginal regions.

Our genetic analyses indicate that mesendodermal progenitors at different positions along the dorsoventral axis have distinct requirements for nodal-related genes. In sqt^-/-; cyc^+/- mutants, dorsal axial structures and dorsal expression of early mesendodermal genes such as gsc, axial, sox17, ntl and cyc, are strongly reduced or eliminated, whereas ventrolateral fates are relatively mildly affected (Figs 2, 5 and 6). This shows that a nodal gene dosage sufficient to induce mesendoderm at the ventrolateral margin is insufficient to induce dorsal mesendoderm. In addition, we noted that dorsal, but not ventrolateral, mesendoderm development is quite sensitive to nodal-related gene dosage. Dorsal mesendoderm is reduced in sqt^-/-; cyc^+/+ mutants, reduced more strongly in sqt^-/-; cyc^+/- mutants, and, as shown in previous studies (Feldman et al., 1998; Gritsman et al., 1999), completely lacking in the absence of Nodal signaling. These differences along the dorsoventral axis are not anticipated by simple models in which Nodal signals act uniformly to induce dorsal and ventrolateral mesendoderm (Fig. 1B).

Differential regulation of sqt and cyc in dorsal and ventrolateral mesendoderm

Dorsal sqt expression is activated by β-catenin (Fig. 9) (Shimizu et al., 2000; Kelly et al., 2000). Previous work (Schneider et al., 1996) has shown that β-catenin localized to dorsal nuclei in the YSL at the 2000-cell stage (3.3 h). We found that β-catenin was detectable in the nuclei of dorsal blastomeres as early as the 128-cell stage (2.25 h) (Fig. 10), prior to formation of the YSL. These results show that dorsoventral asymmetry is established in the blastoderm before the YSL forms. In addition, RNase treatment experiments indicate that RNAs in the YSL are not essential for specification of dorsal identity in overlying blastomeres (Chen and Kimelman, 2000). Taken together, these results establish that β -catenin in dorsal blastomeres specifies dorsal expression of target genes including sqt and boz (Shimizu et al., 2000; Kelly et al., 2000) and show that the YSL is not required for the initial dorsal identity within the blastoderm. By contrast, the YSL is required for expression of sqt and cyc in ventrolateral marginal blastomeres (Chen and Kimelman, 2000).

We found that differential regulation of sqt and cyc expression in dorsal and ventrolateral marginal cells of the late blastula accounts, at least in part, for the differential requirement of nodal-related genes along the dorsoventral axis. In the late blastula, cyc is expressed in a marginal ring that includes cells at all positions along the dorsoventral axis, but our results show that different mechanisms control cyc expression in dorsal and ventrolateral cells. Several observations indicate that sqt activates dorsal cyc expression in response to the maternal dorsal determinant β-catenin. Overexpression of β-catenin induces sqt expression soon after the midblastula transition (Fig. 9K) (Shimizu et al., 2000), and nuclear β-catenin protein is present at the correct time and place to directly activate sqt expression (Fig. 10). In addition, β -catenin requires sqt function for high-level activation of dorsal gene expression (Fig. 9G-J). Dorsal expression of cyc in the late blastula is reduced in sqt mutants (Fig. 3Q), and overexpression of β-catenin does not activate cyc expression until the early gastrula stage (Fig. 9A,C,E). This suggests that β-catenin directly activates sqt, which in turn induces dorsal cyc expression in the late blastula. At later stages, dorsal cyc expression is dependent on an autoregulatory loop, as evidenced by the strong reduction of dorsal cyc mRNA in sqt^-/-; cyc^+/- when compared with sqt^-/-; cyc^+/+ mutants (Fig. 3) and by previous studies of cyc expression in Nodal pathway mutants (Meno et al., 1999; Pogoda et al., 2000; Sirotkin et al., 2000a).

By contrast, ventrolateral expression of cyc is induced independently of Nodal signals but requires Nodal activity to achieve normal levels. cyc expression is induced normally in ventrolateral marginal cells in sqt^-/-; cyc^+/+ embryos, indicating that this expression does not depend on sqt function (Fig. 3). In embryos lacking all nodal gene function, cyc expression is initiated in ventrolateral marginal cells, but does not reach normal levels, whereas dorsal cyc expression is not detectable (Meno et al., 1999). These observations indicate that ventrolateral cyc expression is induced by as yet unknown factors and maintained at normal levels by an autoregulatory loop. Thus cyc expression is controlled differently in dorsal and ventrolateral marginal cells, with dorsal cyc expression entirely dependent on sqt and cyc gene function.

The marked dependence of dorsal, but not ventrolateral, cyc expression on Nodal signaling results in a non-uniform reduction of Nodal signals in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants. Some dorsal marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants adopt neural fates because they are exposed to little or no Nodal signals before gastrulation. By contrast, ventrolateral marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants are exposed to sufficient levels of cyc at the late blastula stage to induce ventrolateral mesendoderm. Thus, the role of autoregulation in the reinforcement of dorsal cyc expression can explain the severe reduction of dorsal but not ventrolateral fates in sqt^-/-; cyc^+/- mutants.

Experiments in zebrafish, frogs and mice indicate that autoregulation is a conserved feature of the transcriptional control of nodal-related genes (this work) (Meno et al., 1999; Hyde and Old, 2000; Osada et al., 2000; Norris et al., 2002). Our results extend this observation by demonstrating that, in zebrafish, cells differ in their sensitivities to the autoregulatory feedback loop depending on their position. cyc expression was completely eliminated from dorsal marginal cells in late blastula stage sqt^-/-; cyc^+/+ embryos, but was present at reduced levels in ventrolateral marginal cells (Fig. 3P-S). This aspect of the autoregulatory control of nodal-related gene expression may also be conserved in other systems, introducing a point of caution in interpreting experiments in which Nodal signaling activity is reduced by genetic methods or by overexpression of Nodal antagonists. It is possible that some cell types are preferentially affected in these experiments.

Dorsoventral patterning of mesendoderm is independent of Nodal signals

Our cell-tracing and gene expression experiments are inconsistent with models proposing that the graded action of Nodal signals patterns the dorsoventral axis (Fig. 1A). Such models predict that dorsal marginal clones should contain fates normally arising at ventrolateral positions, such as muscle, when Nodal signaling is reduced. Rather than muscle or other ventrolateral derivatives, dorsal marginal cells in sqt^-/-; cyc^+/+ and sqt^-/-; cyc^+/- mutants often adopt neural fates (Fig. 7B,C). In addition, ventrolateral gene expression was not expanded into dorsal regions (Fig. 6), as would be expected if dorsal marginal cells adopted ventrolateral fates in these mutants. Thus, despite the elevated dorsal requirement for sqt and cyc, we found no evidence supporting a role for nodal-related genes instructing cell fates along the dorsoventral axis. Instead it seems that factors including Chordin, Boz and BMP signals, which are expressed in the absence of Nodal signaling (Gritsman et al., 1999), are able to pattern the dorsoventral axis without an additional instructive role for Nodal signals.

Overlapping and unique functions of squint and cyclops

Despite the absence of the embryonic shield and dorsal mesendodermal gene expression in all sqt^-/- embryos at early stages, many sqt mutants have dorsal mesodermal derivatives such as notochord at 24 h, and some survive to adulthood (Table 1). The basis of the variability in the sqt mutant phenotype at late stages is not known. Nevertheless, the recovery of dorsal mesendoderm in some sqt mutants depends on cyc function, because all sqt; cyc double mutants lack head and trunk mesendodermal derivatives (Fig. 3) (Feldman et al., 1998). Thus, cyclops activity can compensate for the loss of squint function, either directly or perhaps by activating a parallel pathway that can overcome the loss of sqt. This genetic evidence for overlapping functions of the sqt and cyc genes contrasts with their distinct expression profiles and different activities in misexpression assays (Rebagliati et al., 1998a; Erter et al., 1988).

In misexpression experiments, Cyc acts only over short distances to induce mesodermal gene expression, whereas Sqt acts as a morphogen, directly specifying patterned gene expression over long distances (Chen and Schier, 2001). Thus, the formation of dorsal mesendoderm in sqt mutants demonstrates that Cyc activity can support normal patterning of the late gastrula and viability in the absence of the Sqt morphogen. This suggests that both Sqt and Cyc may act over long distances in their endogenous context, or that the long-range action of a Nodal morphogen is not always essential for viability.

Timing of developmental response to Nodal signals

In light of the different expression profiles of sqt and cyc, it is interesting that Cyc activity can compensate for loss of sqt function. sqt expression initiates at the midblastula stage in dorsal marginal cells, while cyc transcripts first appear in the late blastula in dorsal and ventral marginal cells. In wild-type embryos, cyc expression accumulates to high levels in involuting axial mesoderm, but this expression is reduced and delayed in sqt mutants (Fig. 3). Expression of dorsal mesendodermal markers closely follows cyc expression in sqt^-/-; cyc^+/+ embryos (Fig. 4). For example, gsc expression is nearly absent in the late blastula, when cyc mRNA is greatly reduced in the dorsal marginal region of these embryos. gsc transcripts begin to accumulate at the onset of gastrulation, when cyc expression begins to increase at the dorsal midline. Thus, late exposure of dorsal marginal cells to Nodal signals is sufficient for them to adopt dorsal mesendodermal fates. Conversely, dorsal mesendoderm in sqt^+/+; cyc^-/- embryos is induced entirely by sqt function and sqt expression in these mutants is mostly gone by the beginning of gastrulation (Fig. 3). Thus, exposure of dorsal marginal cells to sqt function before gastrulation, or cyc function after gastrulation has initiated, is sufficient to specify dorsal mesendoderm. This suggests that the timing of the response to Nodal signaling is not a crucial factor in the specification of dorsal mesendoderm in zebrafish.