Distinct roles for Fgf, Wnt and retinoic acid in posteriorizing the neural ectoderm

Studies of axis formation in Xenopus laevis have led to models in which two signaling events that are initiated by localized maternal determinants cause mesendoderm induction in the marginal zone, and organizer formation on its dorsal side at blastula-to-gastrula stages. Mesendoderm induction divides the embryo into three germ layers along the vegetal-to-animal axis, while signals from the organizer induce neural tissue at the dorsal side of the ectoderm. The neural ectoderm is patterned along its anteroposterior (AP) axis, coincident with, and subsequent to, its initial specification; this patterning is initiated by posteriorizing signals derived from prospective or definitive mesendodermal tissues (Nieuwkoop, 1950; Toivonen, 1968). Three candidate posteriorizing signals have been suggested: retinoic acid (RA) (Blumberg et al., 1997; Conlon, 1995; Durston et al., 1989; Sive et al., 1990); fibroblast growth factors (Fgfs) (Cox and Hemmati-Brivanlou, 1995; Kengaku and Okamoto, 1993; Kengaku and Okamoto, 1995; Koshida et al., 1998; Lamb and Harland, 1995) and Wnts (Fekany-Lee et al., 2000; Kazanskaya et al., 2000; Kelly et al., 1995; Kiecker and Niehrs, 2001; McGrew et al., 1995; Yamaguchi, 2001). Although it is likely that factors in all three families participate in this process, the precise role of each in the temporal and spatial aspects of neural patterning as well as the molecular consequences of their action have not been fully clarified.

In zebrafish, a region called the yolk syncytial layer (YSL), which is located beneath the blastoderm, may have a role in both mesendoderm and organizer induction (Mizuno et al., 1996). As a consequence of inductive signals that emanate from the YSL, the blastoderm margin forms the mesendoderm, one side of which develops into the organizer (which, in turn, induces neural specification within the dorsal ectoderm). As in Xenopus, there is evidence for signals emanating from the prospective mesendodermal layer that transform early neural ectoderm from an anterior to a posterior fate (Koshida et al., 1998).

One feature that adds complexity to the mechanisms of AP patterning is the repeated use of the same type of signal in different stages and regions of the embryo, with context-dependent consequences. Fgfs and Wnts are both expressed in undifferentiated mesendoderm from the blastula stage onwards, and in the area of the presumptive midbrain-hindbrain region, beginning at the late gastrula stage (Furthauer et al., 1997; Kelly et al., 1995; Phillips et al., 2001). Signals mediated by members of these two major classes of secreted factors are involved in early AP patterning in the neural ectoderm, as well as subsequent regional patterning processes within the developing brain (Houart et al., 2002; Kim et al., 2000; Reifers et al., 1998). To avoid having to consider a large range of these complexities, we have focused on the earliest manifestation of AP specification that is evident from the late blastula through gastrula stages.

We have searched for genes that may have an early role in axis formation as part of a random in situ screen for regionally expressed genes in zebrafish embryos (Kudoh et al., 2001). In this screen, we noted the anterior neural ectodermal expression of the gene cyp26/P450RAI, which encodes all trans retinoic acid 4-hydroxylase, an enzyme that degrades and inactivates RA. Zebrafish cyp26 has originally been cloned from regenerating fin tissue as an RA-responsive gene (White et al., 1996). We find that cyp26 is specifically expressed in the presumptive anterior neural ectoderm from a surprisingly early stage (from the late blastula onwards). At the early gastrula stage, the earliest known marker of posterior neural ectoderm, hoxb1b (Alexandre et al., 1996), is expressed in a complementary pattern to cyp26. We focused on the regulation of this earliest subdivision along the AP axis in the neural ectoderm, using cyp26 and hoxb1b as our primary tools of analysis. We show that posteriorization of the neural ectoderm has two separable steps: suppression of anterior gene expression and activation of posterior gene expression. Suppression of anterior gene expression in the posterior region is the first step of AP differentiation and is caused by Fgfs/Wnts in an RA-independent pathway. The activation of posterior gene expression is the second step, and this event is RA dependent. These two posteriorizing steps are linked through the regulation and function of the cyp26 gene.

The coding region in cyp26 was amplified by PCR and subcloned into the pCS2+ expression vector. Cyp26-GFP fusion construct was made using PCR primers forward (F; ccggatcctcgaggtcgacccacgcg) and reverse (R; ccgaattcggtgtcagagcccaggatgg), which amplifies 82 bp of 5′ non-coding region with 375 bp of coding region. This PCR product was inserted to pCS2+ and subsequently EGFP cDNA was inserted into the 3′ end of cyp26 cDNA. mRNA was synthesized by mMESSAGE mMACHINE (Ambion). The following mRNAs were injected into zebrafish embryos: cyp26 (500pg), cyp26-EGFP (300pg), fgf3 (50pg) (Koshida et al., 2002), XFD (500pg) (Amaya et al., 1991) and dkk1 (50pg) (Hashimoto et al., 2000).

RA was stored at 10^–3 M in ethanol and diluted to 10^–6 M in fish water before use. Embryos were treated with RA from the 40% epiboly stage onward for 80 minutes, followed by thorough washing. For LiCl treatment, 50% epiboly stage embryos were exposed to 300 mM LiCl for about 8 minutes and washed immediately with fish water several times. These embryos were fixed at late gastrula stage and stained by in situ hybridization.

A sequence complementary to the region of cyp26 cDNA around the start codon was used to synthesize a morpholino antisense oligonucleotide, mCYP1, by Gene Tools (Philomath, USA). The sequence of mCYP1 is 5′-cgcaactgatcgccaaaacgaaaaa-3′. Five nanograms of morpholino were injected at the one- to two-cell stage into the yolk. For comparison, the same amount of standard control morpholino (Gene Tools) was injected.

Embryos were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) overnight at 4°C, manually dechorionated and stored in methanol up to several weeks. In situ hybridization was performed essentially as described previously (Kudoh and Dawid, 2001). Probes for the following genes were used: cyp26, iro1, meis3, hoxb1b, otx2 and ntl. These probes were obtained from our in situ-based screen (Kudoh et al., 2001). For injected and drug-treated embryos, 15 to 30 embryos were used in each experiment.

cyp26 is first expressed at the late blastula stage in two distinct regions, at the animal side of the dorsal ectoderm (Fig. 1A,A′, closed arrowhead) and at the blastoderm margin (Fig. 1A,A′, open arrowhead). These two domains continue to express cyp26 up to the bud stage at the end of gastrulation (Fig. 1A-D). At later gastrula and somitogenesis stages, ntl expression also marks the notochord (Fig. 1C,F). cyp26 expression in the anterior neural ectoderm decreases during gastrulation so that only weak expression in a region surrounding the neural plate is seen by the three-somite stage (Fig. 1D-F). By contrast, cyp26 expression in the blastoderm margin persists as this domain extends towards the posterior, contributing to the tail. This expression pattern suggests that cyp26 is important for establishing the anterior character of neural ectoderm during gastrula stages, as well as for the formation of the tail during later development.

The expression of cyp26 was compared with the posterior neural ectoderm expression of hoxb1b (originally named hoxa1) (Alexandre et al., 1996). At the late blastula stage, there is a small gap between the animal and marginal domain of cyp26 expression (Fig. 2A, asterisk); this gap is expanded subsequently through convergence-extension movements during gastrulation (Fig. 2B-E). At about the 60% epiboly stage, hoxb1b expression is initiated within this gap. When cyp26 and hoxb1b are doubly stained by single-color in situ hybridization, they show complementary expression patterns from the onset of hoxb1b expression to the end of gastrulation (Fig. 2K-M); a narrow gap persists between the cyp26 and hoxb1b domains, emphasizing the exclusiveness of these domains. In these double-labeled embryos, cells completely double negative for cyp26/hoxb1b are not observed at high magnification, suggesting that this gap is not a distinct domain but represents the border of the anterior and posterior domains where the expression of the two markers decreases. This complementarity suggests that initial AP patterning in the neural ectoderm is reflected by the anterior expression of cyp26 in the late blastula, while the expression of hoxb1b, the earliest posterior marker, constitutes a subsequent step in the evolving process.

To further resolve and confirm expression domains, cyp26 was also analyzed in combination with iro1, krox20 and raldh2. iro1 is expressed as two pairs of bands at late gastrula stage; the anterior bands contain the presumptive midbrain-hindbrain boundary and possibly rhombomere 1 (Itoh et al., 2002). The anterior domain of iro1 and the cyp26 expression domain share a common posterior border of expression, suggesting that at this stage, the posterior end of cyp26 expression is around prospective rhombomere 1 (Fig. 2P). Double staining of cyp26 with krox20, which marks rhombomeres 3 and 5 at bud stage, revealed a gap between the posterior edge of cyp26 expression and rhombomere 3 (Fig. 2Q). This gap presumably represents the future rhombomere 2. raldh2, which encodes a major RA synthesis enzyme, is expressed in the posterior paraxial mesoderm in a pattern complementary to cyp26 along the AP axis, with a small but widening gap between the two expression domains (Fig. 2N,O). The complementarity of RA synthesis (raldh2) and degradation (cyp26) domains supports the idea that refinement of RA-positive and -negative domains of the ectoderm is achieved by the action of these opposing gene products (Chen et al., 2001; Swindell et al., 1999).

As cyp26 exhibits extremely early anterior neuroectodermal expression, we tested the regulation of its expression by Fgf, a candidate posteriorizing factor; the behavior of cyp26 was compared with another anterior neural gene, otx2, and to the posterior gene hoxb1b. To generate conditions that correspond to both gain and reduction of function of Fgf signaling, fgf3 and dominant-negative Fgf-receptor (XFD) (Amaya et al., 1991) mRNAs were injected into one- to two-cell stage embryos. fgf3 mRNA injection led to suppression of anterior expression of cyp26 and otx2 (Fig. 3B,J), whereas expression of the posterior gene hoxb1b was expanded anteriorly (Fig. 3F). XFD mRNA injected embryos showed the opposite phenotype in that cyp26 and otx2 domains were expanded posteriorly (Fig. 3D,L) (Koshida et al., 1998) and hoxb1b was suppressed (Fig. 3H arrowhead). These results indicate that Fgf signaling is necessary and sufficient for the suppression of the anterior genes cyp26 and otx2, as well as for the activation of the posterior gene hoxb1b during gastrulation.

RA has been shown to affect the specification of AP values in the developing nervous system of different vertebrates and is a candidate posteriorizing signal (Blumberg et al., 1997; Conlon, 1995; Durston et al., 1989; Niederreither et al., 1999; Sive et al., 1990). RA is also the substrate of the cyp26 gene product. To examine the role of RA in early AP pattern specification in the neural ectoderm, we examined the expression of the anterior genes cyp26 and otx2 and the posterior genes hoxb1b and meis3 in RA-treated embryos. Consistent with the designation of RA as a posteriorizing factor, the expression of hoxb1b and meis3 was expanded (Fig. 4D,F) and that of otx2 was suppressed (Fig. 4H). However, cyp26 expression was expanded in RA-treated embryos, in spite of the anterior character of its ectodermal expression domain (Fig. 4B). This outcome is most likely to be due to the fact that cyp26 is a direct target of RA and acts as an RA-responsive gene in different stages of embryogenesis such as fin formation (Loudig et al., 2000; White et al., 1996). As it encodes an RA-metabolizing enzyme, cyp26 may be part of a negative feedback loop that limits RA signaling in different biological contexts.

To analyze the function of cyp26 and its substrate, RA, in early AP patterning, cyp26 mRNA was injected into one blastomere of two-cell embryos, and the expression of AP marker genes was examined during gastrulation (Fig. 5). Consistent with the activity of its gene product as an RA-degrading enzyme, cyp26 injection showed the opposite effect to application of RA on the expression of the posterior genes hoxb1b and meis3, both of which were suppressed (Fig. 5B,D). We tested an additional gene, iro1, which has two expression domains, an anterior domain that overlaps with cyp26 and a posterior domain that overlaps with hoxb1b (Fig. 2P). Consistent with the results above, only the posterior domain of iro1 was suppressed by cyp26 (Fig. 5F, arrowhead).

By contrast, expression of the anterior gene otx2 was not affected by cyp26 (Fig. 5H). To confirm the differential effect of cyp26 on anterior and posterior neural markers, control and injected embryos were double stained with otx2 and hoxb1b, showing that hoxb1b expression was decreased, whereas that of otx2 was neither suppressed nor expanded (Fig. 5J). However, injection of cyp26 could at least partially restore otx2 expression in the presence of RA (Fig. 5L).

The effects of cyp26 on gene expression illustrated in Fig. 5 for mid-to-late gastrula stages were also seen at early gastrulation (data not shown). These results suggest that RA is necessary for the induction of posterior gene expression, but not for the suppression of anterior gene expression. This lack of an effect of cyp26 in promoting anterior gene expression in the posterior domain represents a difference between RA and Fgf in their action as posteriorizing agents (compare Fig. 3L with Fig. 5H,J).

To investigate the relationship of the posteriorizing signals delivered by Fgf and RA, we applied agonists and antagonists of these two signals to the same embryos. When fgf3 and cyp26 mRNAs were co-injected, fgf3-mediated activation of hoxb1b expression was strongly suppressed (Fig. 6B,C). This result suggests that Fgf3-dependent induction of hoxb1b is mediated by RA. By contrast, Fgf3-mediated suppression of otx2 expression was not affected by co-injection of cyp26 (Fig. 6E,F). Thus, the effect of Fgf in suppressing otx2 does not require RA, and suppression of the RA signal is not sufficient for the induction of otx2.

In complementary experiments, embryos injected with XFD mRNA were treated with RA. XFD blocks Fgf signaling, and consequently inhibits hoxb1b expression (Fig. 6H). Consistent with the results above, this XFD-induced suppression of hoxb1b was overcome by the addition of RA (Fig. 6I). Furthermore, XFD led to an expansion of otx2 towards the posterior (Fig. 6K), and both the expanded and normal expression of otx2 were suppressed in the presence of RA (Fig. 6L). Thus, RA can repress otx2 expression in either the anterior or posterior compartment, but reduction of RA alone is not sufficient to induce ectopic otx2 expression (Fig. 5H, Fig. 6F). These results support the view that RA acts downstream of Fgf in inducing posterior genes in the neural ectoderm.

Next, we examined the involvement of Wnt signaling in the regulation of expression of these early marker genes. LiCl inhibits the kinase activity of GSK3 and consequently activates β-catenin, a downstream effector of the canonical Wnt pathway (Klein and Melton, 1996). Activation of the Wnt pathway in the early gastrula by treatment with LiCl at the 50% epiboly stage led to suppression of the anterior markers, cyp26 and otx2(Fig. 7A,B,F,G). However, the posterior marker hoxb1b was not induced in the anterior-most area of the embryo, although some anterior expansion was observed (Fig. 7K,L). To reduce canonical Wnt signaling, dkk1 mRNA was injected. Dkk1 slightly expanded otx2 and cyp26 expression, and reduced that of hoxb1b, but there was always residual posterior character near the margin (Fig. 7C,M).

In order to elucidate the relationship between Wnt and Fgf signals in regulating AP polarity, we concomitantly altered activity of both pathways. XFD injected embryos were treated with LiCl to suppress Fgf activity, while enhancing Wnt signaling. In these embryos, cyp26 and otx2 as well as hoxb1b were suppressed (Fig. 7D,I,N). This suggests that activation of Wnt signaling can suppress anterior markers in the absence of Fgf signaling but that activation of posterior markers requires Fgf.

As a complementary experiment, fgf3 and dkk1 were co-expressed to enhance Fgf while suppressing Wnt signaling. In these embryos, cyp26 and otx2 were suppressed, indicating that Dkk1 is not sufficient to promote anterior markers in the presence of exogenous Fgf signaling (Fig. 7E,J). Dkk1 did, however, inhibit the ability of Fgf to expand hoxb1b expression into anterior regions (Fig. 7O).

In the next set of experiments, we examined the relationship between the RA and Wnt signaling pathways. RA activity was suppressed by overexpression of cyp26, and subsequently LiCl immersion at 50% epiboly stage was used to activate Wnt signaling. In these embryos both otx2 and hoxb1b were suppressed, indicating that Wnt signals inhibit anterior development in the absence of RA (Fig. 8B,E). As a complementary experiment, dkk1 mRNA-injected embryos were treated with RA. In these embryos, otx2 was largely suppressed and hoxb1b was ectopically expressed up to the animal pole (Fig. 8C,F). Although there is otx2 expression remaining in the dorsal-most area (Fig. 8C, arrowhead), because these cells are hypoblast they are possibly the expanded axial mesoderm. These results suggest that Wnts and RA both can suppress otx2 in the anterior neural ectoderm but that activation of hoxb1b depends on RA.

One possibility is that Wnts promote hoxb1b expression through regulating the production of RA. To assess if this might be the case, expression of the gene encoding an RA synthesis enzyme, raldh2, was examined in conditions in which Wnt signaling was increased or decreased. At the 50% epiboly stage, raldh2 is expressed in nascent paraxial mesoderm at the blastoderm margin. In embryos in which the Wnt antagonist dkk1 was injected, raldh2 expression was not suppressed although the dorsal raldh2-negative domain was slightly expanded, most probably because Dkk1 activity is known to enhance axial mesodermal fates (Hashimoto et al., 2000). This suggests that Wnt signaling is not necessary for the initiation of raldh2 expression. However, by late gastrula stage, raldh2 expression remained restricted to the blastoderm margin in dkk1-injected embryos, whereas in control embryos, raldh2 was highly expressed in the involuted mesoderm as it migrated anteriorly (Fig. 8I,J). This limitation of raldh2 expression could account for the similar restriction of hoxb1b expression to the marginal zone of dkk1-injected embryos. raldh2 expression was also examined in LiCl-treated embryos and shown to retain normal expression (data not shown). Together, these results suggest that Wnt signals indirectly influence hoxb1b expression through the modulation of RA activity.

To examine the physiological role of cyp26 in early AP patterning, morpholino antisense oligonucleotide (mCYP1) was injected into zebrafish embryos. In mCYP1-injected embryos, otx2 expression was decreased (Fig. 9B) and hoxb1b expression was expanded anteriorly (Fig. 9D). Complementing this, the expression domains of meis3 and iro3 were both shifted in an anterior direction (Fig. 9F,H). These results suggest that the activity of endogenous Cyp26 is required to restrict the expression of posterior genes at their anterior border and to protect anterior genes from repression by RA. Cyp26 is likely to carry out these functions by degrading RA molecules that encroach into the anterior ectodermal compartment, thereby limiting their presence to the posterior compartment.

The efficacy of mCYP1 was confirmed in experiments using a cyp26-GFP fusion construct. GFP fluorescence was strongly suppressed by the co-injection of mCYP1 but not by the control morpholino (Fig. 9I,J); thus mCYP1 at this concentration efficiently suppressed the translation of Cyp26 fusion protein. In addition to the changes in expression of AP marker genes, morpholino injected embryos were more ovoid in shape by late gastrulation (Fig. 9B,D,F,H). A similar shape change was observed in RA-treated embryos at the end of gastrulation (not shown) supporting the notion that endogenous RA activity is increased by the mCyp1.

The consequences of altering Fgf, RA and Wnt signals upon early AP patterning of the ectoderm are summarized as a cartoon in Fig. 10. In general, all anterior and all posterior markers behaved similarly in each set of experiments, with the exception of the cyp26 gene, which was induced by RA even though it is normally expressed in the anterior region of the neural ectoderm (see Discussion). In the cartoon, we summarize the results obtained from analysis of the expression of other markers, primarily otx2 and hoxb1b. From this, it is evident that anterior gene and posterior gene expression domains never overlap. However, although in some cases, the entire embryo acquired anterior or posterior identity (i.e. fgf3 or XFD injection), in other cases, anterior and posterior genes retained complementarity of expression but with a shifted boundary (i.e. dkk1 injection). Finally, in some situations we observed that anterior or posterior character could be specifically lost in one region (i.e. after cyp26 injection or LiCl treatment) or throughout the entire embryo (i.e. after fgf3+cyp26 injection or cyp26 injection plus LiCl treatment). The latter case is particularly instructive: excessive signaling by either Fgf or Wnt eliminates anterior character, but establishing posterior character depends on RA. Thus, the combination of ectopic Fgf or Wnt signals and suppression of RA leaves the presumptive neuroectoderm without any AP identity.

Our study has revealed that AP regionalization of the prospective neural ectoderm occurs very early and that the cyp26 gene appears to be involved in this early patterning. We have analyzed the expression pattern of zebrafish cyp26 and found that its dorsoanterior expression begins around 30-40% epiboly in the late blastula. This observation is of interest as it suggests that AP patterning of the neural ectoderm is initiated well before the onset of gastrulation. Expression in the anterior neural domain persists through gastrulation but disappears at early somitogenesis stages, suggesting that cyp26 affects AP patterning mainly from the blastula through the gastrula stages. cyp26 has been identified in Xenopus and in the mouse, and is also expressed in the anterior neural ectoderm in these species (de Roos et al., 1999; Fujii et al., 1997; Hollemann et al., 1998).

As Cyp26 degrades RA it is expected to limit the range of RA-mediated posteriorization in the embryo. Our Cyp26 overexpression and loss-of-function studies support this idea. RA is known to be involved in the expression of posterior genes including Hox genes (Kolm et al., 1997; Kolm and Sive, 1995; Niederreither et al., 1999; Simeone et al., 1990). In our experiments, RA induced posterior genes such as hoxb1b and meis3, while cyp26 injection suppressed expression of these genes, presumably by reducing the endogenous RA concentration. However, the expression domain of the anterior gene otx2 was not expanded by cyp26 mRNA injection, suggesting that RA-independent pathways suppress otx2 in the posterior neural ectoderm. We believe that limiting the range of RA action is the in vivo function of Cyp26 as morpholino antisense oligonucleotide injection caused anterior expansion of posterior genes such as hoxb1b, and a shrinking of the expression domain of the anterior gene, otx2. Cyp26 thus has an important role in defining the border between anterior and posterior neural ectoderm.

cyp26 is also expressed at the blastoderm margin with a gap between its two domains of expression. The blastoderm margin expression persists into the developing tail region during somitogenesis, suggesting that suppression of RA signals is also required in tail development. Consistent with this view, RA treatment in Xenopus and zebrafish causes truncation of tail as well as head structures (Durston et al., 1989) (data not shown). Recently, mice carrying targeted mutations in cyp26 were generated and these animals showed loss of tail and adjacent posterior structures, stressing the importance of RA signal suppression by cyp26 for tail development (Abu-Abed et al., 2001; Sakai et al., 2001).

Overexpression experiments using cyp26 in Xenopus showed otx2 expression was slightly expanded in a posterior direction (Hollemann et al., 1998). This is somewhat different from our observation that otx2 was not expanded by cyp26 overexpression. The differences may be due to the fact that phenotypes in frog were examined at later stages, when many other CNS patterning events are likely to have occurred and modified early effects of exogenous Cyp26 activity. Our observation that RA caused the expansion of hoxb1b expression in anterior neural ectoderm differs from results reported previously by Alexandre et al., who described more subtle phenotypes following RA treatment (Alexandre et al., 1996). This difference can be explained by the different concentration of RA in the two studies. When we applied 10^–7 M RA to zebrafish embryos, hoxb1b was induced only in axial mesoderm, as reported by Alexandre et al. (Alexandre et al., 1996). However we used 10^–6 M RA because this or higher doses generate anterior truncations in Xenopus (Durston et al., 1989; Sive et al., 1990). Furthermore, the internal concentration of RA under these conditions is not known. Recently, the zebrafish mutants neckless and nofin, which have a decrease in hindbrain size, have been shown to carry mutations in the RA synthesis enzyme, Raldh2 (Begemann et al., 2001; Grandel et al., 2002). Begemann et al. have shown that the mutant phenotype in the hindbrain could be rescued by 5×10^–7 M or 10^–6 M RA, but not by lower concentrations. This suggests that when applied externally, 10^–6 M RA may mimic physiological conditions that are achieved by normal RA synthesis in the wild-type embryo (Begemann et al., 2001).

With the aim of investigating the nature of the signals that initiate differential expression of cyp26 and hoxb1b, we examined the role of Fgfs in this process, as Fgfs have been implicated in posteriorization of the neural ectoderm (Griffin et al., 1995; Koshida et al., 1998). Experiments that involved injection of mRNAs for Fgf3 and the truncated receptor XFD into zebrafish embryos showed that Fgf signaling is both necessary and sufficient, directly or indirectly, for the suppression of the anterior genes cyp26 and otx2 and the induction of the posterior gene hoxb1b.

Like Fgf signaling, Wnt activity is known to affect the early specification of AP polarity in the neural ectoderm (Fekany-Lee et al., 2000; Kelly et al., 1995). We found that activation of Wnt signaling suppressed otx2 and cyp26, while the expression of hoxb1 was partly expanded by Wnt activation and largely suppressed by inhibition of Wnt signaling.

Interactions between the posteriorizing pathways were studied by combined activation of one signal and inhibition of another, as summarized in Fig. 10. These experiments indicate that either an Fgf or a Wnt signal can suppress anterior genes such as cyp26 and otx2 even when the other pathway is inhibited. Therefore, Wnts and Fgfs can act independently of each other in the suppression of anterior genes; this suppression was also independent of the activity of RA, but RA can suppress anterior genes even in the presence of Fgf or Wnt pathway inhibitors. By contrast, activation of posterior gene expression could only be initiated by Fgf or Wnt signals in the presence of an intact RA signaling pathway. It is likely that RA acts directly on the hoxb1b promoter, as RA-responsive elements are present in the mouse Hoxb1 gene and have been shown to have a crucial role in regulating its expression (Marshall et al., 1994; Ogura et al., 1996a; Ogura et al., 1996b). The expansion of hoxb1b expression after ectopic Wnt activation may in part be explained by the suppression of cyp26, which results in expansion of the range of RA activity in an anterior direction. This interpretation is in agreement with the similar expansion of the hoxb1b domain after injection of a cyp26 morpholino antisense oligonucleotide. By contrast, the more extensive expansion of hoxb1b expression by Fgf overactivation is unlikely to simply be due to abrogation of Cyp26 activity. Instead we suggest that other factors must contribute to the ectopic production of an RA signal in anterior regions of Fgf injected embryos.

A model for the mechanism of early AP patterning, based on the observations described in this paper, is presented in Fig. 11. A key feature of this model is that promotion of posterior fates and suppression of anterior fates are treated as separable events.

Fgfs and Wnts can suppress anterior genes in an RA-independent pathway. The set of genes sensitive to this suppression includes cyp26, which encodes an enzyme that degrades RA. While suppression of RA signaling is not necessary/sufficient for activation of anterior genes, we suggest that it is needed to prevent activation of posterior gene expression within the prospective anterior neural plate. The factors important for activation of anterior gene expression remain unknown, but many studies have suggested that ‘anterior’ is a default fate for induced neural tissue (Nieuwkoop, 1950; Toivonen, 1968).

Within prospective posterior neural tissue, RA is necessary and sufficient for the activation of at least some posterior genes. Furthermore, the ability of Fgfs and Wnts to promote expression of these posterior genes depends upon RA. In part, this is likely to be due to the ability of Fgfs and Wnts to promote RA activity through suppression of cyp26 expression, but is also likely to be due to additional regulatory events, such as promotion of raldh2 expression.

In Fig. 11B, a spatial and temporal model underlying these AP patterning events is outlined. At the late blastula stage, genes in the Fgf and Wnt families are expressed in the prospective mesoderm at the blastoderm margin; fgf3, fgf8 and wnt8 are known to be expressed in this pattern (Furthauer et al., 1997; Kelly et al., 1995; Koshida et al., 2002; Phillips et al., 2001). We suggest that an early role for these margin-derived signals is to suppress, directly or indirectly, the expression of cyp26 and possibly other anterior genes in dorsal ectoderm adjacent to the margin, thereby initiating the specification of this region as presumptive posterior neural ectoderm at the late blastula stage (Fig. 11B, part i). The initial suppression of these genes is likely to be achieved through a planar signal because, at the 30-40% epiboly stage when localized cyp26 expression is first seen, the mesendodermal layer has not yet involuted below the ectodermal layer. This view is consistent with the report that posterior neural specification in dorsal ectoderm is observable at the shield stage (50-55% epiboly) but not in the blastula at the 30% epiboly stage (Grinblat et al., 1998).

Our model further proposes that the widening Cyp26-free area allows the accumulation of RA and the consequent induction of RA-dependent posterior genes, including hoxb1b, at early-to-mid gastrula stages (Fig. 11B, part ii). Subsequently, ongoing convergence-extension movements move lateral posterior ectoderm cells in a dorsal direction, causing further anteroposterior expansion of the Cyp26-negative, RA-positive area, thereby maintaining and enhancing the expression of hoxb1b and other posterior genes. Cyp26 maintains its expression anteriorly, thereby defining the rostral limit of expression of the posterior genes (Fig. 11Biii).

The RA-synthesizing enzyme, Raldh2, is likely to be involved in these early patterning events. In several species, including zebrafish, Raldh2 is expressed in the posterior mesoderm (Begemann et al., 2001; Berggren et al., 1999; Chen et al., 2001; Grandel et al., 2002; Niederreither et al., 1997; Swindell et al., 1999). Raldh2 mutations in mouse and zebrafish reduce posterior neural ectoderm with concomitant downregulation of Hox genes (Begemann et al., 2001; Grandel et al., 2002; Niederreither et al., 1999). In a complementary manner, increased Raldh2 levels posteriorize the nervous system of Xenopus embryos (Chen et al., 2001). Therefore, two mechanisms might regulate RA accumulation, one mediated by the synthetic enzyme Raldh2, which augments RA in the posterior neural ectoderm, the other mediated by Cyp26, which degrades RA in the anterior neural ectoderm (Chen et al., 2001; Swindell et al., 1999). Mutations in the Cyp26 gene in mouse do indeed lead to a moderate anterior expansion of the Hoxa1 expression domain (Abu-Abed et al., 2001; Sakai et al., 2001), a result similar to that obtained by cyp26 morpholino injection in fish. As Raldh2 is expressed only in the posterior region, RA may not significantly accumulate in the anterior-most region, even in the absence of Cyp26, possibly explaining the relatively mild phenotypes after abrogation of Cyp26 activity in mouse and zebrafish. Therefore, although cyp26 is widely expressed in the anterior neural ectoderm, its major role may be the definition of the boundary in the presumptive hindbrain beyond which expression of posterior genes such as hoxb1b and meis3 does not expand.

In summary, we show in the present paper that AP patterning is initiated in the presumptive neural ectoderm in the late blastula at 30-40% epiboly stage. We can distinguish two posteriorization steps in this process. The earliest step involves the Wnt/Fgf-dependent, RA-independent suppression of anterior genes, including cyp26 in the presumptive posterior domain. The next step involves the activation of genes such as hoxb1b and meis3 in the posterior domain; this step is mediated by RA signaling. The antagonism between cyp26 activity and RA signaling links the initial and the subsequent steps of AP patterning, thereby contributing to the establishment of the earliest known border between anterior and posterior neural ectoderm.

We thank Elizabeth Laver and the UCL Fish Facility personnel for assistance with zebrafish husbandry; H. Takeda for expression constructs; and A. Chitnis, M. Itoh and M. Tsang for comments on the manuscript. T. K. was supported by JSPS and Fogarty fellowships and is currently supported by Wellcome Trust funding. S. W. is a Wellcome Trust Senior Research Fellow.