Anteroposterior patterning in the zebrafish, Danio rerio: an explant assay reveals inductive and suppressive cell interactions

During embryogenesis in the zebrafish, Danio rerio, the first morphological indicator of axial asymmetry appears at the beginning of gastrulation, with a thickening called the ‘embryonic shield’ on the dorsal side of the marginal zone. Fate mapping has shown that the embryonic shield contains future dorsal derivatives including mesodermal (presumptive notochord) and ectodermal (presumptive neural) lineages (Kimmel et al., 1990; Shih et al., 1995). As gastrulation progresses, cells move dorsally so that the entire dorsal side of the embryo, extending to the animal pole, becomes thicker than the ventral side (Kimmel and Warga, 1987; Kimmel et al., 1990). By the end of gastrulation, an obvious anteroposterior axis is present, with a posterior tailbud and anterior prechordal plate.

Transplantation assays indicate that germ layer restriction occurs during gastrulation (Ho, 1992b; Ho and Kimmel, 1993), although these assays have not been used to address the detailed timing of axial patterning. Molecular markers clearly show that dorsoventral asymmetry is established just after the mid-blastula transition, even before the shield appears. For example, expression of the homeobox gene goosecoid (gsc; Stachel et al., 1993; Schulte-Merker et al., 1994a; Thisse et al., 1994) and the forkhead gene axial (Strähle et al., 1993) is already restricted to the future dorsal side of the embryo shortly after the mid-blastula transition. Conversely, the homeodomain gene eve1 (Joly et al., 1993) is expressed exclusively in the ventrolateral marginal zone at early gastrula and the zinc finger gene gta3 (Neave et al., 1995) becomes ventrally restricted by mid-gastrula.

The early events in anteroposterior axis formation are less well illustrated by patterns of gene expression. As cell migration begins during gastrulation, cells expressing gsc move anteriorly, while cells expressing no tail (ntl) remain more posterior (Schulte-Merker et al., 1994a) defining an anteroposterior axis in the hypoblast (the deep cell layers) of the embryo. By mid-gastrula, otx1 and otx2 RNAs are restricted to anterior dorsal epiblast (superficial layers) and mesendoderm (hypoblast) (Li et al., 1994) confirming that an anteroposterior axis is present in both embryonic layers. By the end of gastrulation, many markers are activated in the epiblast at different anteroposterior positions. These include the paired domain gene pax6 (Krauss et al., 1991; Püschel et al., 1992), the homeodomain gene eng3 (Ekker et al., 1992), the zinc finger gene krox20 (Oxtoby et al., 1993) and notch (Bierkamp et al., 1993).

Despite these descriptions, the mechanisms that regulate zebrafish axis formation are not clear. It has been suggested that dorsoventral asymmetry is induced by signals from the yolk cell, with formation of the embryonic shield one result (Stachel et al., 1993). The shield is believed to be equivalent to the amphibian ‘organizer’ or amniote ‘node’, able to direct anteroposterior axis formation and induce neural tissue. This notion is supported by expression in the shield of genes also expressed in the Xenopus organizer and mouse or chicken node, such as goosecoid (for example: Blum et al., 1992; Cho et al., 1991; Izpisua-Belmonte et al., 1993; Schulte-Merker et al., 1994a; Stachel et al., 1993; Thisse et al., 1994) and lim1 (Taira et al., 1993; Barnes et al., 1994; Toyama et al., 1995). Where tested, genes affecting axial patterning in other vertebrates appear to carry out similar functions in fish (for example: Griffin et al., 1995; Halpern et al., 1993; Krauss et al., 1993; Schulte-Merker et al., 1994b; Toyama et al., 1995). A pivotal role for the shield in axial patterning is also supported by the finding that lithium treatment, which generates excess shield tissue, leads to overrepresentation of dorsal tissues, including neurectoderm (Stachel et al., 1993). Further, bifurcation of the shield in the janus mutant (Abdelilah et al., 1994) correlates with development of a twinned embryo. In other teleost fish, Fundulus, Salmo and Perca, shields transplanted into the ventral side of a host embryo led to formation of dorsal axial derivatives in the host, as would be expected from an organizing center (Luther, 1935; Oppenheimer, 1936, 1959). Despite this corroborative evidence, there has been little direct analysis of shield function in the zebrafish (see Ho, 1992a for a discussion of preliminary data) and it has been unclear precisely when and how this region may act during axial patterning.

In order to study the timing of anteroposterior interactions during zebrafish development, we developed an explant assay for extended in vitro culture of tissue isolated from blastula and gastrula stage embryos. We present data showing that anteroposterior neural patterning can be induced in zebrafish and that this appears to be a multistep process. These data clarify normal zebrafish embryogenesis, and will be valuable in analysing mutant phenotypes resulting from genetic screens.

Adult zebrafish were maintained as described by Westerfield (1995). Embryos were generated by natural crosses. Staging was performed according to Kimmel et al. (1995).

Dissections and incubations were done at 29°C in 1× MBS+G (1× modified barth’s saline (Peng, 1991) with 50 μg/ml gentamicin). Animal caps consisted of about 50 cells from the animal pole of sphere to dome stage embryos (4-4.5 hpf). Individual explants were immediately aggregated into groups of 10 making them easier to handle. Shield explants encompassing about 60° of the circumference at the blastoderm margin and 30-40% of the distance from the margin to the animal pole were taken from shield stage embryos (6 hpf). A shield explant was about 5 times larger than a cap and could be cultured individually. Ventral marginal zone (VMZ) explants were of the same size as the shield explants, but were taken from the side of the embryo directly opposite the shield. Conjugates were made by combining animal caps with either shield or VMZ explants. First animal caps were dissected. Second, within 1 hour, shield or VMZ explants were cut. Third, an animal cap aggregate was cut in half and the newly exposed surface placed next to the internal face of a freshly cut shield or VMZ. Each resulting conjugate contained one shield or VMZ and five animal caps. This represented an approximately equal mass of each tissue.

Explants were dissociated by repeated pipetting in a small volume of PhoNaK buffer (Godsave et al., 1989). Cultured explants required the addition of collagenase (2 mg/ml). Cells were fixed in 4% formaldehyde (EM grade, EM Sciences) and, after at least 20 minutes at room temperature, nuclear stain (DAPI, 1 μg/ml final concentration) was added. Cells with intact nuclei were counted using a hemacytometer and a fluorescence microscope equipped with a UV light source.

The sequence of the primers used for each gene, the size of the amplification product (bp) and the source of the sequence information is given in Table 1 for each gene analyzed.

RNA was isolated according to the method of Chomczynski et al. (1987). One animal cap represented about 1/30th of an embryo. This ratio remained true during culture. First strand cDNA was synthesized with Superscript II Reverse Transcriptase (Gibco BRL, no. 18064-014) according to the manufacturers instructions. The linear range was determined for all primer pairs using cDNA made from about 2.5 explants or 0.075 embryos (representing an equal amount of template for each sample). Since all primer pairs were found to amplify linearly between 20 and 26 cycles, amplification was for 24 cycles of 1 minute at 94°C, 1 minute at 55°C and 1 minute 30 seconds at 72°C. The PCR reaction mixture contained 1× PCR buffer, 1 mM MgCl₂, 0.2 mM of each dNTP, 5 μCi [α-³²P]dCTP (800 Ci/mmol), 2.5 U Taq polymerase and 0.3 μM of each primer. Amplification samples were purified over Qiaquick spin columns (Qiagen, no. 28106) and eluted in 50 μl TE. A third of the sample was separated on a 5% acrylamide gel and analyzed autoradi-ographically. Since the same amount of template (see above) was added to all reactions, comparisons could be made between samples, giving information about the relative expression of a gene in different samples. However, since the efficiency of amplification may vary between primer pairs, it was not possible to compare levels of two different genes. Northern analysis with ntl, MyoD, wnt1, pax6 and cyt1 (not shown) as well as in situ hybridization with gsc, ntl, pax6, otx2, cyt1 and gta3 (Fig. 2C) was used to verify the RT-PCR quantitation.

RNA probes containing digoxigenin-11-UTP (Boehringer, no. 1209256) were synthesized from linearized plasmid DNA for otx2, pax6, eng3, krx20, cyt1, gta3, gsc and ntl as described (Harland, 1991). For double in situ hybridizations a fluorescein-labelled probe for ntl was prepared by substituting fluorescein-12-UTP (Boehringer, no. 1427857) for the digoxigenin-11-UTP. In situ hybridizations were performed as described by Oxtoby and Jowett (1993) except that most steps were carried out in 12-well dishes with Netwell inserts (Costar, no. 3481). Blocking and antibody incubations were done in 1× maleic acid buffer [1× MAB=100 mM maleic acid (Sigma M0375), 150 mM NaCl, pH 7.5] with 20% lambserum (heat inactivated at 60°C) and 2% BMB (Boehringer, no. 1096176). 10% polyvinyl alcohol (PVA; 98-99% hydrolyzed; Mw 31,000-50,000; Aldrich no. 36,313-8) was added during the developing step to accelerate the color reaction. For double probe in situ hybridization the two probes were included together in the hybridization step. First the digoxigenin-labelled probe was detected with anti-digoxigenin antibody (Boehringer, no. 1093274) and developed in the presence of BCIP (Sigma B8503), giving a light blue stain. Second, the fluoresceinated probe was detected with anti-fluorescein antibody (Boehringer, no. 1426338) and developed in the presence of NBT (Research Organics 0415N-2) and BCIP, giving a purple stain. For sectioning, samples were embedded in JB4 resin (Polysciences 00226). 5 μm sections were cut on a microtome using a dry glass knife and mounted in Crystal-mount (Biomeda M03).

5 nl lyseinated fluorescein-dextran (FLDX; 10,000 M_r; Molecular Probes D-1820) at 1% in 10 mM Tris, pH 7.6, 0.1 mM EDTA was injected into 1-8 cell stage embryos. Embryos were monitored under a fluorescence microscope prior to dissection to ensure uniform labelling. The lineage label was detected following in situ hybridization by anti-fluorescein antibody and developed in the presence of only BCIP, giving a light blue stain.

We initially set out to isolate a non-neural cell population that was a substrate for neural induction. Explants were made from the animal pole region, at late blastula (sphere stage, 4 hours postfertilization; hpf). ‘Animal cap’ explants consisted of only about 50 cells and were therefore cultured as groups of ten to increase the tissue mass (Fig. 1A and Materials and Methods). Caps survived in culture for at least 2 days and, consistent with the good health of the explants, cell division continued during the culture period with a 15-fold increase in total cell number after 20 hours of culture (Fig. 1B).

The specification state of animal caps was analysed using a quantitative reverse transcriptase-polymerase chain reaction (RT-PCR) assay for tissue-specific transcripts (Fig. 2A and Materials and Methods). Explants were harvested at the time of dissection (late blastula, 4 hpf), when control embryos initiated gastrulation (6 hpf, 2 hours in culture), when controls reached the end of gastrulation (10 hpf, 6 hours in culture) or when controls completed somitogenesis (24 hpf, 20 hours of culture). By the final timepoint control embryos have a readily apparent nervous system and express all the markers analyzed. At the time of isolation, animal caps contained very low levels (5% of the maximal expression in the embryo) of ntl RNA that is expressed throughout the marginal zone and later in the notochord (Schulte-Merker et al., 1992), or gsc RNA that is dorsally restricted (Stachel et al., 1993; Schulte-Merker et al., 1994a; Thisse et al., 1994) (Fig. 2B, lane 1). This may have represented maternal expression (Stachel et al., 1993) and was lost within 2 hours of culture (lane 2). Two other mesodermal markers, the ventrally restricted eve (Joly et al., 1993) and MyoD, which is muscle-specific (Weinberg et al., 1996), were not expressed in cultured caps (Fig. 2B).

Six neural markers; zash1a, zash1b (Allende and Weinberg, 1994), pax6 (Krauss et al., 1991; Püschel et al., 1992), wnt1 (Molven et al., 1991), eng3 (Ekker et al., 1992) and krx20 (Oxtoby and Jowett, 1993), representative of different positions along the anteroposterior axis, were also not expressed in cultured cap explants (Fig. 2B). The anterior marker otx2 (Li et al., 1994), expressed both in epiblast (ectoderm) and hypoblast (mesendoderm) during gastrulation, and later in the anterior nervous system, was expressed transiently and at low levels (about 10% of the whole embryo level after 6 hours in culture (lane 3), with no expression by 20 hours (lane 4) of culture.

In contrast, we observed high levels of two non-neural markers, cyt1 (Sagerström and Sive, unpublished data) and gta3 (Neave et al., 1995) (Fig. 2B). cyt1 is a novel type I cytokeratin gene that is first expressed at early gastrula stages. Its expression increases at least until the end of somitogenesis (24 hpf) and is localized primarily to the enveloping layer (Sagerström and Sive, unpublished data). In explants, cyt1 expression proceeded with an onset and timecourse similar to that in the whole embryo (compare lanes 1-4 with lanes 5-8). gta3 is expressed on the entire ventral side of the embryo from early gastrula, with later expression in pronephros and some neurons (Neave et al., 1995). gta3 was expressed at high levels in explants both at the time of dissection and during culture, similar to the whole embryos (Fig. 2B). Since mesodermal- and neural-specific genes were generally not expressed in cultured caps, we conclude that gta3 expression in caps solely represented the ventral component of whole embryo expression.

In agreement with the RT-PCR results, whole-mount in situ hybridization on cultured caps verified the absence of gsc, ntl and pax6 RNAs during culture (Fig. 2C, a, b and c respectively). otx2 was detected at low levels (d), whereas cyt1 and gta3 were highly expressed (e and f respectively). Sections through stained explants showed expression of cyt1 primarily at the surface (g) and gta3 internally (h), analogous to the expression pattern of these genes in the embryo, suggesting that the explants achieved the correct internal and external organization.

These analyses showed that zebrafish animal caps did not contain neural or mesodermal precursors, but that they could activate non-neural marker gene expression with the same time course as the whole embryo. We concluded that blastula stage animal caps required further signals for neurogenesis and axial patterning.

The embryonic shield may be an axial organizing region in zebrafish (see Introduction), and was therefore the initial tissue we tested for the ability to induce neural gene expression in animal caps. In order to better define the shield itself, we analysed its fate after isolation from shield stage embryos (6 hpf) and culture (see Fig. 3A). After 5 hours of culture (until control embryos reached the end of gastrulation), 100% of individually cultured shields elongated and developed a broad region at one end, that expressed otx2 (Fig. 3B, a) and gsc (b). At the other end, a smooth protruding ‘knob’ formed that expressed ntl (c). Expression of gsc and otx2 and ntl at opposite ends of the explant was confirmed by double whole-mount in situ hybridization (d). Since in control embryos, gsc RNA is anteriorly restricted and ntl RNA is posteriorly restricted, we assign the broad end of the explant as anterior and the knob end as posterior.

Expression of multiple neural markers was also activated in 100% of cultured shields with a timing similar to that seen in the whole embryo. pax6 and krx20 RNAs were detected after 9 hours of culture (when control embryos reached the 12 somite stage; Fig. 3B, f and g) and 17 hours in culture (when sibling embryos had reached the end of somitogenesis; i and k), at which time eng3 was also strongly expressed (j). otx2 expression was maintained at 9 and 17 hours of culture (e and h) and is likely to reflect neural expression, as seen in embryos of equivalent age (Li et al., 1994). Neural genes were expressed in a detailed anteroposterior pattern similar to that seen in the whole embryo (Fig. 3C, a-c, and summarized in d). otx2 RNA was present at the anterior tip of the explant and extended about one third of the length, presumably reflecting the forebrain expression of this gene. pax6 showed one broad stripe of expression in explants, overlapping the otx2 staining, presumably reflecting forebrain expression of the whole embryo, as well as a fainter stripe more posteriorly, reflecting the normally weaker hindbrain expression. Two stripes of krx20 RNA were located just anterior to the posterior band of pax6 staining, reflecting expression in rhombomeres 3 and 5 of the hindbrain. After 17 hours in culture, eng3 was expressed in a single band as it is at the midbrain/hindbrain border of the embryo. However, additional morphological changes in the explants made it difficult to distinguish anterior and posterior ends.

These data showed firstly, that the early gastrula shield contained specified mesodermal and neural precursors. Secondly, despite its small size, the shield was always able to organize a detailed anteroposterior axis along which genes were expressed in similar relative positions to those in the whole embryo. These findings also suggested that the shield may contain the signals required to induce and pattern neural tissue in animal caps.

We tested whether the embryonic shield could induce animal caps to a neural fate in a conjugation assay. A group of 5 fused blastula animal caps was conjugated with a freshly cut early gastrula shield as shown in Fig. 4A, and cultured before being assayed for specific RNAs. Shield was distinguished from animal cap by labelling the shield with fluoresceinated dextran (FLDX; see Materials and Methods), visualized as light blue in the conjugates (Fig. 4B). Lineage labelling was required since the shield contains intermingled mesodermal and neural precursors (Shih and Fraser, 1995), making it impossible to separate the two populations by dissection. Consistent with this there is no detectable morphological boundary between the epiblast and hypoblast at this stage.

While the shield portion of the conjugate elongated during culture, similarly to shields cultured alone, the animal cap portion did not elongate. ntl and gsc were induced in the cap portion of only about 10-15% of the conjugates (Table 2; arrowheads in Fig. 4B, a and b) after 5 hours of culture (when control embryos reached the end of gastrulation). Induced gsc was detected as a small patch in the cap, adjacent to gsc RNA in the shield while induced ntl was seen as a patch separate from ntl RNA in the shield. In the embryo, gsc expression has a transient neural component at late gastrula (Thisse et al., 1994), and it was not clear whether the small amount of induced gsc represented mesodermal or neural expression.

In contrast, extensive and high level expression of otx2 was induced in about 60% of the conjugates after culture for 5 hours (Table 2) or 9 hours (when controls reached the 12 somite stage; Fig. 4B, arrowheads in d; Table 2). otx2 induced after 5 hours may reflect both hypoblast (mesendodermal) and epiblast (neural) expression, while otx2 induced after 9 hours probably represents solely neural expression as in embryos at the equivalent stage (Li et al., 1994). pax6 was also induced in the cap portion in a high percentage of conjugates after culture for 9 hours (approximately 80% of conjugates; Table 2; Fig. 4B, arrowheads in c) or 17 hours (60% of the conjugates; when sibling controls reached the end of somitogenesis; Table 2). Induced pax6 and otx2 expression was generally aligned with expression of the same marker in the shield (Fig. 4B, c and d). When the conjugates were oriented as described for the shield explants (Fig. 3C) it became clear that the anterior stripe of pax6 RNA was contiguous with that induced in the animal cap (data not shown).

More posterior markers were induced less frequently in the cap portion of conjugates. After 17 hours of culture krx20 was induced in 5% and eng3 in 10% of the conjugates (Table 2; Fig. 4B, e and f), while the posterior stripe of pax6 was not induced (data not shown). Induced krx20 and eng3 expression was also less extensive than that of the more anterior markers. These data showed that the embryonic shield could efficiently induce anterior neural markers in blastula stage animal caps. However, the shield did not induce posterior neural markers efficiently, showing that anterior and posterior neural induction are separable events.

In order to test whether the ability to induce neural gene expression was unique to the dorsal shield, we conjugated animal caps to explants from the ventral marginal zone (VMZ) of shield stage embryos (6 hpf) as shown in Fig. 5A. Unlike their dorsal counterparts, VMZ explants did not elongate, did not express any neural markers, nor did they induce pax6 expression in animal caps (data not shown). Instead we found that the VMZ suppressed the low level of otx2 expression seen in animal caps (Fig. 5B, a and b). 90% of animal cap explants cultured alone for 9 hours expressed a diffuse patch of otx2, however, when the caps were conjugated to VMZ shortly after dissection, and assayed after culture for 9 hours, only about 10% of the conjugates expressed detectable otx2 (Table 3). These data showed that the ventral marginal zone actively inhibited the expression of otx2, that at later stages is dorsoanterior-specific.

Using explant assays, we have directly demonstrated neural induction and analysed anteroposterior patterning in zebrafish. Our data indicates that this process is likely to involve several steps, where anterior neural induction is separable from more posterior induction, and where ventrally derived signals may counteract dorsally derived inducers. These results clearly indicate that the embryonic shield can induce patterned gene expression, and confirm that this region is an active organizing center, equivalent to the amphibian organizer and amniote node. Explant culture assays have been used very little for analysing development in zebrafish (Schulte-Merker et al., 1992) and other teleost fish (Bozhkova et al., 1994; Wittbrodt et al., 1994).

The extensive array of zebrafish embryonic mutants defined (Mullins et al., 1993; Driever et al., 1994) will be extremely useful for analysing axial patterning. However, a description of normal development is essential to fully understand any mutant phenotype (Keller, 1991; Kimmel et al., 1995; Shih and Fraser, 1995). Zebrafish gastrulation occurs differently to that of other vertebrates. For example, unlike the amphibian organizer, mesodermal and neural precursors are intermingled in the zebrafish shield (Shih and Fraser, 1995). Further, unlike amniotes, zebrafish have no primitive streak, although cell ingression apparently occurs that is similar to amniotes and dissimilar to amphibians (Slack, 1991; Kimmel et al., 1995). These differences made it necessary to address experimentally whether developmental strategies in zebrafish were similar to those of other vertebrates.

Our animal cap explants were made at late blastula, earlier than the first fate maps (Kimmel et al., 1990; Woo and Fraser, 1995) raising the question of what lineage(s) was specified in the caps. A type I cytokeratin gene, cyt1, was activated in the outer enveloping layer of the caps during culture, as it is in the normal embryo, confirming that the outer cells were committed to their normal fate by late blastula (Kimmel et al., 1990). The pattern of cyt1 expression is similar to epidermal cytokeratin gene expression in the outer (epithelial) layer of Xenopus animal caps and intact embryos (Jamrich et al., 1987). Since, in the animal pole region, the Xenopus epithelial layer is considered to be part of the ectoderm (Keller, 1975), by this criterion, the outer layer of the zebrafish caps is specified ectoderm at late blastula.

The inner animal cap cells expressed high levels of gta3 indicating that they were specified as ventral tissue. However, caps also expressed low levels of otx2, that at later stages is a dorsoanterior marker (Li et al., 1994). These data indicate that the caps have both strong ventral and weak dorsoanterior character. These characteristics also appear in Xenopus animal caps, which express both the ventral marker Bone Morphogenetic Protein-4 (BMP-4; Fainsod et al., 1994) and low levels of otx2 (Pannese et al., 1995; J. Kuo and H. Sive, unpublished data). otx2 is also expressed in prestreak epiblast from mice (Simeone et al., 1993; Ang et al., 1994), suggesting that the commitment state of these tissues is similar in different organisms.

In late gastrula embryos, gsc expression in the prechordal plate is separated from ntl expression in the notochord (Schulte-Merker et al., 1994a). Expression of gsc and ntl also separated in explanted shields, with otx2 and gsc expressed on a broad anterior region likely equivalent to the prechordal plate, and ntl expressed on a posterior knob reflecting notochord and, possibly, tailbud development. Since morphogenesis in explants and whole embryos occurred at about the same time (when embryos completed gastrulation) normal gastrulation movements may take place in culture.

The shield contains neural precursors (Kimmel et al., 1990; Shih and Fraser, 1995), however, much presumptive neural tissue lies outside the shield region, with anterior (forebrain) regions towards the animal pole and posterior regions towards the blastoderm margin (Kimmel et al., 1990; Woo and Fraser, 1995). Although forebrain apparently does not map to the shield piece that we explanted (Woo and Fraser, 1995), we consistently saw high levels of forebrain markers in cultured shields. This striking difference between the specification state and the fate map of the shield may be because, in the embryo, prechordal plate mesoderm migrates away from the shield during gastrulation and induces anterior neural lineages near the animal pole. In explants, prechordal plate cells cannot migrate out of the shield, and instead induce the anterior fates they would have in the intact embryo, but in a different ecto-dermal population. Hindbrain precursors map lateral to the shield region we cultured (Woo and Fraser, 1995), suggesting that expression of krx20 in shield explants may also have arisen by regulative induction of cells different from those that would form hindbrain in the embryo.

In conjugates, pax6 and otx2 were induced in the part of the animal cap directly adjacent to the region expressing each of these genes in the shield. Often the precise boundaries of expression in the shield and the overlying cap were coincident, implying a direct transfer of positional information between shield and cap. The simplest interpretation of these data is that during normal zebrafish development a vertical signal can be transmitted between the hypoblast, as it migrates towards the animal pole, and uncommitted cells in the epiblast of the animal hemisphere (Li et al., 1994; Thisse et al., 1994), however our data does not address the role played by inductive signals trans-mitted in the plane of the explants (reviewed by Doniach, 1993). Neural markers in the caps could have been induced firstly, directly by mesoderm in the shield, or secondly, homeogenetically, by neurectoderm in the shield. Since the intermixing of mesodermal and neural precursors in the shield at the time of dissection (Shih and Fraser, 1995) made it impossible to separate the two germlayers, we could not distinguish between these possibilities. However, it was clear that anterior neural induction did not require induction of mesoderm in the cap since very few explants expressed either gsc or ntl.

The shield induced anterior neural markers in animal caps much more readily than it activated posterior neural genes. Consistently, in conjugates, caps did not elongate or form the ‘knob’ that was an indicator of posterior morphogenesis. Since the shield itself went on to express posterior neural genes, blastula animal caps were probably not competent to respond to posterior inducing signals from the shield. This implies that ectoderm may require a preparatory signal to become responsive to posterior inducers.

A similar situation may hold in other vertebrates. In Xenopus, retinoids and fibroblast growth factor can induce posterior neural genes but generally only when applied to previously induced ectoderm (for example Kengaku and Okamoto, 1995; Lamb and Harland, 1995; Sharpe, 1991; Sive et al., 1990). Classical embryological data has suggested that early during amphibian gastrulation, anterior type inductions occur in the posterior ectoderm, which is subsequently reprogrammed to a more posterior fate (for a review see Slack and Tannahill, 1992). These events can be reproduced in conjugates of both Xenopus (Sive et al., 1989) and mice (Ang et al., 1994), however, the data do not address whether anterior induction is required for more posterior induction, or whether more general preparatory inductions are sufficient.

The ablation of otx2 expression from the entire animal cap by the ventral marginal zone indicated either that an inhibitory signal diffused over the diameter of the cap, about 50 μm, or activated an intercellular signalling cascade. How does this reflect what takes place in the whole embryo? As gastrulation proceeds, the ventral marginal zone involutes to approach anterodorsal regions (Westerfield, 1995). We postulate that in zebrafish, ventrally derived signals may counteract dorsally derived inducers, define expression boundaries for dorsoanterior genes such as otx2, and so help position the front of the embryo. The data presented here is the first demonstration of this phenomenon in a zebrafish explant assay. Indeed, little direct data showing that ventral tissues isolated from normal embryos can alter the fate of dorsal regions has been obtained in any vertebrate (for exceptions see: Sive and Bradley, 1996; Zaraisky et al., 1992; Zhang and Jacobson, 1993).

An active role for ventral regions in axial patterning is supported by assays for specific gene function in fish and frog embryos. In Xenopus, wnt8 is expressed in ventrolateral mesendoderm from early gastrula (Christian et al., 1993), while in zebrafish a similar gene is also not expressed at the dorsal midline during gastrulation (Kelly et al., 1995). When expressed after the mid-blastula transition, wnt8 is a potent inhibitor of dorsal differentiation (Christian and Moon, 1993; Kelly et al., 1995). BMP-4 plays an important role in Xenopus ventral patterning (for examples, see Dale et al., 1992; Fainsod et al., 1994; Graff et al., 1994; Maeno et al., 1994; Schmidt et al., 1995), however it is not yet clear whether BMPs act similarly in zebrafish.

We have formulated a speculative multistep model for neural induction and anteroposterior patterning in the zebrafish gastrula (Fig. 6). We suggest that the first step takes place in the shield early during gastrulation, when a mixture of presumptive anterior and posterior mesodermal cells begin inducing neural precursors. Neural precursors may be intermingled amongst mesendodermal cells at the time of induction and subsequently sort into the epiblast, or they may segregate into the epiblast before induction (Shih and Fraser, 1995). Any positional identity in these early induced precursors is unclear (this work). Concurrently, or as a second step, deep cells in the shield (that are gsc positive) migrate towards the animal pole (Stachel et al., 1993; Schulte-Merker et al., 1994a; Thisse et al., 1994) and induce an anterior neural fate in the overlying epiblast (Li et al., 1994; Thisse et al., 1994). Thirdly, more posterior neural lineages are induced in cells derived from the original shield (this work) and in cells that converge dorsally from lateral regions (Woo and Fraser, 1995). Posterior inducers may require previously induced cells as a substrate (this work; see Discussion above). Fourthly, inhibitory signals originating from ventral regions may counteract some of the dorsally derived inductive signals to help position the front of the embryo (This work; Kelly et al., 1995).

The unique aspects of zebrafish development (Kimmel et al., 1995) suggested that zebrafish anteroposterior patterning may also use unusual strategies. However, our data suggest that many of the developmental mechanisms used by amphibians and mice are shared by zebrafish. The model we propose will be refined by further exploration of cell interactions involved in anteroposterior patterning, by isolating genes that are early markers of this process, and in conjuction with embryonic mutants that have defects in neural induction and patterning (Kimmel, 1989; Rossant et al., 1992; Mullins and NüssleinVolhard, 1993; Driever et al., 1994).

We are indebted to many colleagues for giving us probes prior to publication: Eric Weinberg, Nigel Holder, Bob Riggleman and David Grunwald, Greg Conway and Walter Gilbert. We thank Nancy Hopkins for fish, for encouragement and for reading the manuscript, Frank Solomon and Wolfgang Driever for their useful comments on the manuscript and the Sive lab members working on Xenopus for providing both scepticism and helpful criticism. We thank Vladimir Apekin for expert care of zebrafish. This work was supported by the Searle Scholars Program/ Chicago Community Trust in an award to H. L. S., and we gratefully acknowledge this contribution. C. G. S. was supported by a Sokol Post-doctoral Award and is currently an American Cancer Society Senior Post-doctoral Fellow. Y. G. is supported by a postdoctoral fellowship from the NIH.