FGF signaling is required for β-catenin-mediated induction of the zebrafish organizer

FGF signaling is involved in patterning and tissue formation of early vertebrate embryos (reviewed by Sivak and Amaya, 2004; Böttcher and Niehrs, 2005). Expression of XFD, a dominant negative form of Xenopus FGF receptor 1 (FGFR1), in Xenopus and zebrafish embryos results in loss of trunk and tail tissues(Amaya et al., 1991; Amaya et al., 1993; Griffin et al., 1995). FGF signaling has also been shown to be required for neural induction(Streit et al., 2000; Wilson et al., 2000; Böttcher and Niehrs,2005), for organizer-independent induction of posterior neural tissue (Kudoh et al., 2002; Kudoh et al., 2004), anterior neurectoderm formation (Hongo et al.,1999) and posterior mesoderm development(Draper et al., 2003).

Perhaps due to the pleiotropic effects that result from inhibition of FGF signaling in early embryos, a role of FGF signaling in dorsal axis establishment - as distinct from mesoderm induction, posterior development and neural induction - has not been emphasized in recent reviews(De Robertis and Kuroda, 2004; Niehrs, 2004). Nevertheless,substantial evidence does indicate that FGF signaling is involved in establishment of the dorsal axis. In Xenopus, targeted injections of XFD RNA to blastomeres that give rise to the Spemann organizer but not to trunk mesoderm itself, result in trunk and tail deficiencies, but only slightly affect head development (Mitchell and Sheets, 2001). These results show that FGF signaling is required for trunk-inducing activities of the Spemann organizer. Ectopic expression of fgf8 in the zebrafish embryo can induce a partial secondary axis, but not head(Fürthauer et al., 1997),indicating that in this organism FGF signaling is sufficient for partial organizer function. Evidence for the importance of FGF signaling in axis formation also comes from studies on chordin (chd) and BMP expression. In both Xenopus(Mitchell and Sheets, 2001)and zebrafish (Londin et al.,2005), chd expression is dramatically reduced in embryos blocked in FGF signaling, and dorsalization of embryos by overexpression of fgf3 does not occur in embryos (dino mutants) lacking chd function (Koshida et al.,2002).

Also consistent with a role for FGFs in dorsal axis formation is the relationship between FGF gene expression and potential upstream signaling pathways. Organizer formation is fully dependent on β-catenin in amphibians, mice and zebrafish (Heasman et al., 1994; Wylie et al.,1996; Heasman,2000; Huelsken et al.,2000). In zebrafish embryos bred from female ichabodhomozygotes, a maternal effect mutation resulting in severe ventralization and loss of head and trunk, β-catenin fails to localize to dorsal YSL and blastomere nuclei (Kelly et al.,2000). These deficits are due to impairment of maternal expression of a second β-catenin gene, β-catenin-2, which maps in proximity to the ichabod mutation and is required for establishment of the early dorsal signaling center(Bellipanni et al., 2006). ichabod embryos injected with FGF8 RNA developed trunk and anterior structures such as forebrain and eyes, although a complete anterior brain and notochord did not form (Tsang et al.,2004). fgf3 and fgf8, as well as mkp3,a modulator and itself a target of FGF signaling, are first normally expressed in the dorsal margin of blastula-stage zebrafish embryos(Fürthauer et al., 1997; Tsang et al., 2004). ichabod embryos fail to express these genes in this early dorsal domain, suggesting that FGF ligands operate downstream of β-catenin in dorsal axis formation (Tsang et al.,2004). β-catenin is also required for expression of XFGF3 in both dorsal and non-dorsal regions of Xenopus blastulae, andβ-catenin overexpression causes stimulation of XFGF3 expression(Schohl and Fagotto, 2003). Activation of the MAPK pathway is the most common downstream response to FGF signaling (Böttcher and Niehrs,2005). MAP kinase phosphatase 3 (MKP3) has been shown to inhibit FGF/MAPK signaling activity by dephosphorylation of MAPK proteins (reviewed by Tsang et al., 2004) and expression of MKP3 in zebrafish embryos can eliminate FGF signaling and cause ventralization (Tsang et al.,2004). These experiments taken together indicate that FGF signaling operates downstream of β-catenin in dorsal axis establishment.

We have utilized the ichabod mutant embryo system to work out the pathway relationships between these factors, the Nodal protein gene squint (sqt), and the homeodomain protein gene bozozok/dharma/nieuwkoid (boz). sqt(Feldman et al., 1998; Erter et al., 1998; Rebagliati et al., 1998) and boz (Yamanaka et al.,1998; Koos and Ho,1998; Fekany et al.,1999) act in parallel downstream of β-catenin(Shimizu et al., 2000; Kelly et al., 2000) to establish the organizer (Sirotkin et al.,2000; Shimizu et al.,2000). As embryos bred from homozygous ichabod mothers fail to form a Nieuwkoop center equivalent region(Kelly et al., 2000), they can be used to determine which factors and signaling pathway components are downstream of β-catenin, and which aspects of embryonic patterning are dependent on this pathway (Kelly et al.,2000; Tsang et al.,2004; Kudoh et al.,2004; Gore et al.,2005). An advantage of this system is that individual downstream components of the pathway can be expressed in the absence of expression of other β-catenin-dependent factors. Using this approach, we show that FGF signaling is absolutely required for β-catenin activation of the organizer and dorsal axis, that at least one mode of dorsal axis formation proceeds through the pathway β-catenin→Squint→FGF signaling→Chordin, and that FGF signaling is required for dorsal axis formation at additional steps downstream of β-catenin, including the accumulation of boz transcript and boz induction of chd.

To obtain ichabod^p1 heterozygous and homozygous fish,we crossed ichabod homozygous males with ichabodheterozygous females. Embryos derived from crosses of ichabodhomozygous females and brass or AB strain males are essentially all ventralized and are referred to as `ichabod embryos' in this study. Only those matings that yielded predominantly the most severely ventralized phenotypes (Class 1; see Fig. 1B) (Kelly et al.,2000) were used in this work. Embryos derived from crosses of ichabod heterozygous females and brass or AB strain have wild-type phenotype and served as wild-type controls.

Antisense RNA probes were synthesized and in-situ hybridization procedures were performed as previously described(Hashimoto et al., 2004), with BM purple alkaline phosphatase substrate (Roche) used instead of NBT/BCIP as chromogen in some experiments. Capped mRNAs were synthesized using the appropriate mMessage mMachine Kit (Ambion), following the manufacturer's protocol. Approximately 1 nl of RNA solution was injected through the intact chorion into the yolk at the base of the blastomeres of 4-8-cell stage embryos(except for the case of Fig. 8S). The amounts of mRNA that were injected were as follows: fgf8 (0.1 pg), fgf3 (1 pg), β-catenin-1 (50 pg), bozozok (50 pg), squint (10 pg), XFD (200 pg), d50,(200 pg), and mkp3 (200 pg). The following morpholino antisense oligonucleotides (MOs) were obtained from Gene Tools (Philomath, OR):MOsqt (5′-ATGTCAAATCAAGGTAATAATCCAC-3′), MOcyc(5′-GCGACTCCGAGCGTGTGCATGATG-3′), MOchd(5′-ATCCACAGCAGCCCCTCCATCATCC-3′). The pNA-grip antisense oligonucleotide, pNAboz (N-TTCAAGTGTAGGGGTGCC-C), was used to inhibit bozozok expression. These specific reagents have all previously been shown to specifically phenocopy the respective mutant phenotypes(Feldman and Stemple, 2001; Karlen and Rebagliati, 2001; Nasevicius and Ekker, 2000; Urtishak et al., 2003), and we confirmed that they produced fully penetrant phenotypes in wild-type embryos in our experiments and also yielded these phenotypes when co-injected withβ-catenin RNA into ichabod embryos (data not shown). The amounts of these antisense reagents that were injected were: MOsqt (8 ng),MOcyc (8 ng), MOchd (1 ng) and pNAboz (2.2 ng).

We previously reported (Tsang et al.,2004) that ichabod embryos do not express fgf3and fgf8 in the prospective organizer region and that they can be partially rescued by FGF3 or FGF8 RNA injection to form eyes and/or some anterior neurectoderm, but not notochord of fully normal anterior brain(Tsang et al., 2004). As targeted injection of FGF RNA into one cell at the 4-8-cell stage can elicit secondary axis formation (Fürthauer et al., 1997), and rescue of ichabod is more efficient with β-catenin RNA if injected in this fashion, we repeated our experiments with injections into one cell of ichabod embryos at this stage. By contrast with our earlier experiments, in which at most only 7% of the injected embryos formed normal appearing nervous systems, we obtained a high percentage of embryos with no neural defects (50% for FGF8 and 21% for FGF3), although notochords still did not form(Fig. 1A). The higher degree of rescue could have resulted from a higher local concentration of ligand. Injecting a combination of the two FGFs did not increase the percentage of embryos with completely normal anterior neurectoderm and did not induce notochord. These experiments show that FGF signaling was sufficient to induce normal-appearing neural tissue of both anterior and posterior identity, a novel finding. Based on these results, all mRNA and antisense oligonucleotide injections reported in the following experiments (with the exception of the RNA injection in Fig. 8S) were carried out by injection into one cell at the 4-8-cell stage.

To better characterize the phenotype of FGF mRNA-injected ichabodembryos, we examined the expression of genes involved in dorsoventral (DV) and anteroposterior (AP) patterning (Fig. 1G-Z). FGF3 and FGF8 could induce otx2(Fig. 1H-J) as well as hoxb1b (Fig. 1M-O)during gastrulation in ichabod embryos, indicating that FGF signaling is sufficient for correctly patterned expression of these two markers in the absence of dorsal β-catenin. Both FGFs could induce the formation of at least a partial organizer in ichabod embryos, as shown by the expression of goosecoid (gsc)(Fig. 1R-T) and floating head (flh) (data not shown) in the embryonic shield region at 50% epiboly. These embryos went on to express gsc in the prechordal plate at tailbud stage (Fig. 1W-Y). FGF8 is more effective than FGF3 in inducing all of these markers, in agreement with the greater degree of rescue of ichabodembryos by FGF8 RNA (Fig. 1A). FGF expression did not result in no tail (ntl) or flh expression in a recognizable notochord in injected ichabod embryos, although weak expression of flh was observed in the midline mesoderm at tailbud stage (data not shown). FGFs appear to be sufficient to induce notochord precursor cells, but not for their maintenance. By contrast, FGF expression was sufficient to induce and maintain posterior and anterior neurectoderm, organizer and prechordal plate in embryos deficient in dorsal β-catenin-mediated signaling.

To examine whether FGF signaling is required for β-catenin activation of organizer formation, we tested the effect of FGF signaling pathway inhibitors on β-catenin RNA rescue of ichabod embryos. To block FGF signaling, we injected RNA for the dominant negative version of FGFR-1,XFD (Amaya et al., 1991), along with β-catenin RNA into a single blastomere at the 4-8-cell stage. The ability of β-catenin to dorsalize and rescue ichabod embryos was completely inhibited by XFD, as shown by morphological examination of 24 hours post fertilization (hpf) injected embryos(Fig. 2A,E,F). To confirm the specificity of the XFD effect, we also injected embryos with RNA encoding the non-functional FGF receptor, d50 (Amaya et al., 1991), which did not inhibit β-catenin dorsalization and rescue (Fig. 2A,E,G). We also examined gsc expression at shield stage in embryos injected with these RNAs. β-catenin can induce gsc in ichabod embryos(Kelly et al., 2000)(Fig. 2J), but XFD completely blocked this induction, whereas d50 had no effect(Fig. 2K,L). This result suggests that FGF signaling is required for an early step in formation of the organizer in response to β-catenin. To confirm that FGF signaling is essential for β-catenin dorsalization and rescue, we repeated the experiment of Fig. 2A,injecting RNA for MKP3, a protein effective in eliminating FGF signaling in zebrafish embryos (Tsang et al.,2004). In agreement with the results of XFD expression, MKP3 strikingly inhibited β-catenin rescue of ichabod(Fig. 2B). We also utilized a third method of inhibiting FGF signaling, treatment with SU5402, an inhibitor of FGF receptor activity (Mohammadi et al., 1997). In ichabod embryos co-injected withβ-catenin RNA and SU5402, induction of gsc was severely inhibited (Fig. 2J,M). On the basis of three independent methods of inhibiting FGF signaling, we conclude that such signaling is required for β-catenin to induce organizer genes and rescue organizer-deficient embryos.

As β-catenin can induce dorsal expression of sqt and boz in the zebrafish embryo and expression of either sqt or boz can fully rescue ichabod(Kelly et al., 2000; Shimizu et al., 2000), we next examined the relationship between FGF signaling and expression of these two genes during β-catenin-induced organizer formation. Dependence of FGF expression on Nodal signaling has previously been shown for mesodermal precursors in the zebrafish embryo(Mathieu et al., 2004). Overexpression of the Nodal proteins Sqt or Cyclops (Cyc) is sufficient to induce fgf3 and fgf8, and early expression of these FGF genes in presumptive mesoderm does not occur in MZoep embryos lacking Nodal signaling (Mathieu et al.,2004).

We first examined whether FGF3 or FGF8 could induce sqt and boz in ichabod embryos and found that neither gene was induced by either of the FGFs (data not shown). By contrast, Sqt and Boz both could induce fgf3 and fgf8 in ichabod embryos(Fig. 3A-E,G-K). As we had previously shown that Boz can induce Sqt in both wild-type and ichabod embryos (Kelly et al.,2000), we tested whether Boz might function through Sqt to induce fgf3 and fgf8 expression. Co-injection of boz RNA and an MO targeting sqt (MOsqt) into ichabodembryos completely blocked the induction of fgf3 and fgf8obtained by injection of boz RNA alone(Fig. 3E,F,K,L). Thus, asβ-catenin can induce both sqt and boz, and bozcan induce sqt under these conditions, β-catenin induction of FGFs proceeds through induction of Sqt and subsequent Nodal signaling.

To test this further, we examined the effects of inhibiting sqt or boz on β-catenin induction of fgf3 and fgf8 in ichabod embryos (Fig. 3M-T). In this case, MOs for both sqt and cycwere injected along with β-catenin RNA, in case the two Nodal proteins could redundantly transduce the β-catenin activation effect. Treatment with the two Nodal MOs completely eliminated β-catenin-dependent FGF gene expression (Fig. 3O,S), whereas co-injection of boz antisense PNA (pNAboz) (at a concentration that phenocopied the boz mutation when injected into ichabod embryos along with β-catenin) had only a slight effect(Fig. 3P,T). These results show that Nodal proteins are required for β-catenin induction of FGFs and that boz is not essential for this part of the pathway.

The experiments described thus far suggest that β-catenin rescue of ichabod embryos proceeds at least in part via the linear pathway,β-catenin→Nodals→FGFs→organizer function. As Sqt, but not Boz, can directly induce FGF genes, we compared the ability of Sqt and Boz to rescue ichabod under conditions of inhibition of FGF signaling. When XFD RNA was co-injected with sqt RNA into ichabod embryos,the ability of Sqt to rescue and dorsalize was completely inhibited(Fig. 4A-C,I), whereas d50 expression had no effect (Fig. 4D,I), consistent with the linear pathway mentioned above. By contrast, Boz could restore a significant degree of dorsalization as well as anterior neural tube formation (class 3^* and 4 embryos; Fig. 4J), but not notochord. Many of the rescued embryos, however, had abnormal trunk and tail morphologies, even when they developed two eyes (e.g. Fig. 4G). Thus, FGF signaling is absolutely necessary for the induction of dorsal and anterior tissues by Sqt in ichabod embryos, but Boz induction of both dorsal and anterior structures - although not notochord - is at least partially independent of FGF signaling.

We next investigated events downstream of FGF signaling. As several studies have shown that FGFs can induce the BMP antagonist gene, chd(Mitchell and Sheets, 2001; Koshida et al., 2002; Londin et al., 2005), we tested whether FGFs could stimulate the expression of chd in ichabod embryos. Injection of 1 pg of FGF3 or 0.1 pg of FGF8 RNA,sufficient for rescue of the complete AP pattern of neurectoderm in ichabod embryos (Fig. 1), induced expression of chd(Fig. 5D-F,H-L) and repressed bmp2b (Fig. 5P-R,T-X)and bmp4 (data not shown) in these embryos. Although we found that FGF8 and FGF3 both induced chd and repressed bmp2b during gastrulation, only FGF8 had this effect at 30% epiboly(Fig. 5D,G,J). Expression of FGF3 after injection of 1 pg of RNA neither induced chd nor repressed bmp2b at this stage (Fig. 5P,S,V). Thus, under conditions of axis formation, we were unable to dissociate chd induction from bmp2b repression. We did confirm work of others (Fürthauer et al., 2004; Londin et al.,2005) that FGF3 could repress bmp2b at 30% epiboly if 50 pg of RNA was injected (Fig. 5D′), and we also found that this repression was independent of chd expression by co-injecting the RNA and a morpholino oligonucleotide against chd (MOchd)(Fig. 5F′). However, the physiological significance of effects using this high concentration of RNA is not clear, as such embryos are hyperdorsalized and do not form a recognizable axis.

To test to what extent chd was required for FGF3 and FGF8 partial rescue of ichabod embryos, we co-injected MOchd along with 1 pg of FGF3 RNA or 0.1 pg of FGF8 RNA into these embryos. Inhibition of chd expression caused a complete block in FGF-dependent dorsalization(Fig. 6A,B,E,F,H,I). However,co-injection of MOchd and β-catenin RNA into ichabodembryos resulted in the much less severe dino mutant phenotype(Fig. 6C,N), indicatingβ-catenin is able to induce trunk, anterior neurectoderm and dorsal mesoderm by chd-independent mechanisms.

In the section above, we demonstrated that FGFs can induce chd and that chd is required for FGF partial rescue of ichabodembryos. As it has been shown that either sqt or boz gene function is sufficient for some degree of chd expression(Sirotkin et al., 2000; Shimizu et al., 2000), we tested whether β-catenin induction of chd in ichabodembryos is similarly dependent on Sqt and Boz. ichabod embryos were co-injected with β-catenin RNA and either MOs against sqt and cyc, pNAboz, or a combination of the three antisense inhibitors. As expected, chd expression, assayed at 70% epiboly, was decreased, but not abolished, in embryos blocked in Sqt and Cyc function and also in embryos blocked in Boz function, and the combination of inhibition of Nodals and Boz resulted in complete inhibition of β-catenin-induction of chd, demonstrating an absolute dependence on these factors (data not shown). We also checked to what extent Boz and Sqt could each independently induce chd in ichabod embryos. Co-injection of bozRNA and sqt and cyc MOs, or co-injection of sqt RNA with pNAboz resulted in an intermediate level of induction of chd (data not shown). These experiments verify that the previously characterized roles of Boz and Sqt in chd expression(Sirotkin et al., 2000; Shimizu et al., 2000) are operative in ichabod embryos.

We next tested to what extent FGF signaling is required for induction of chd and repression of bmp4 by β-catenin, Sqt or Boz(Fig. 7). β-catenin(compare Fig. 7A,B with 7C,F),Sqt (compare Fig. 7A,B with 7I,L), and Boz (compare Fig. 7A,B with 7O,R) can induce chd and repress bmp4in ichabod embryos at 70% epiboly. The induction of chd and repression of bmp4 by both β-catenin(Fig. 7C-H) and Sqt(Fig. 7I-N) was blocked by expression of XFD, but not d50 control. Boz induction of chd was also blocked by XFD and not d50 (Fig. 7O-Q), although XFD failed to inhibit the repressive effect of Boz on bmp4 expression (Fig. 7R-T), consistent with a direct inhibitory effect of Boz on bmp transcription (shown for bmp2b)(Leung et al., 2003a), and a result compatible with the roles of Boz as an antagonist of vox, ventand ved (Melby et al.,2000; Imai et al.,2001; Shimizu et al.,2002), which are most probably independent of FGF signaling and chd expression. The lack of chd expression even when bmp4 is inhibited by Boz expression, is the one instance of dissociation of chd induction and BMP gene repression revealed in our studies, indicating that inhibition of BMP expression is not always sufficient for chd induction. In this case, FGF signaling appears to be required for Boz induction of chd in addition to any mechanism that results in BMP gene repression. The ability of Boz to repress bmp4 expression in the absence of FGF signaling and chd expression is consistent with the rescue of anterior and dorsal tissues in ichabod embryos under these conditions (Fig. 4E-H,J).

The effects of XFD on chd induction and bmp4 repression described above can only partially explain why β-catenin and Sqt cannot dorsalize or rescue ichabod under conditions of inhibition of FGF signaling. If these were the only effects of XFD, β-catenin and Sqt rescue of ichabod embryos should yield the comparatively mild dino phenotype, rather than being fully blocked by the inhibitor(e.g. Fig. 2A,F, Fig. 4C,I). In the section below, we present an additional effect of inhibition of FGF signaling that may explain the complete block to β-catenin or Sqt rescue of ichabodby XFD.

Somewhat paradoxically, although FGF signaling is essential forβ-catenin or Sqt to rescue ichabod embryos, and Chd is essential for FGFs to partially rescue these embryos, β-catenin is able to restore most of the AP and DV pattern in ichabod embryos that are blocked in chd expression. These findings predict that FGF signaling is required for dorsal axis formation at steps downstream of β-catenin in addition to activation of chd. As expression of β-catenin can induce boz in ichabod embryos(Kelly et al., 2000) and Boz expression can restore a significant degree of dorsalization (as well as anterior neural tube formation) in the absence of FGF signaling(Fig. 4F,G.J), it was unclear why inhibition of FGF signaling would result in a complete block in the ability of β-catenin or Sqt to induce an axis in ichabodembryos. The simplest explanation would be that FGF signaling is required forβ-catenin-dependent boz expression. To test this hypothesis, we examined boz expression at sphere stage in β-catenin RNA-injected ichabod embryos in the presence and absence of XFD(Fig. 8A-D) and found that boz transcripts were not detected when XFD was expressed, whereas d50 had no inhibitory effect. In contrast to boz expression, we found no effect of XFD on sqt expression, indicating that FGF signaling was not required for all β-catenin-activated gene expression. In further support for a role of FGF signaling in β-catenin-dependent bozexpression, we co-injected MKP3 and β-catenin RNAs into ichabodembryos (Fig. 8K,N,Q), or injected MKP3 RNA alone into wild-type embryos(Fig. 8S). In injected ichabod embryos, MKP3 completely abolished boz expression at 3.5 and 4 hpf sphere stage (compare Fig. 8M with 8N, and 8P with 8Q), and in the wild-type embryos, MKP3 greatly reduced the level of boz transcripts at this stage (compare Fig. 8R,S). However, at 3.0 hpf, the earliest stage we could detect boz expression inβ-catenin RNA-injected ichabod embryos, we found no inhibition of boz expression by MKP3 (compare Fig. 8J,K). Taken together, the results with both XFD and MKP3 indicate that FGF signaling is essential for accumulation of boz transcript, but is not required for initial transcription of this gene. The complete block of β-catenin and Sqt rescue of ichabod embryos when FGF signaling is inhibited can thus be explained by the finding that FGF signaling is essential for both bozand chd expression.

We have shown, using expression of the FGF signaling inhibitors XFD, MKP3 and SU5402 that FGF signaling is required for β-catenin induction of a dorsal axis in ichabod embryos devoid of a Nieuwkoop center equivalent. β-catenin induction of both gsc and chd and repression of bmp4 were completely inhibited in ichabodembryos blocked in FGF signaling by XFD(Fig. 2I-K, Fig. 7C-H). Moreover,dorsalization and dorsal axis formation in otherwise highly ventralized ichabod embryos were completely blocked by both XFD and MKP3(Fig. 2A-F). XFD or MKP3 expression also prevented β-catenin-dependent accumulation of boz transcript (Fig. 8A-D,L-Q). In addition, treatment with SU5402 inhibited theβ-catenin induction of gsc in these embryos(Fig. 2O). These results demonstrate that FGF signaling is necessary for β-catenin induction of the zebrafish organizer and in formation of both anterior and posterior neurectodermal tissues.

In agreement with these results, β-catenin can induce early dorsal expression of fgf3 and fgf8 in ichabod embryos,which lack expression of these genes in the early prospective dorsal domain(Tsang et al., 2004). Also indicating a role for FGF signaling, we found that injection of FGF3 or FGF8 RNA into ichabod embryos resulted in early expression of gsc(Fig. 1R-T) and flh,properly patterned expression of otx2 and hoxb1b during gastrulation (Fig. 1H-J,M-O),and subsequent formation of a dorsal axis, with many embryos developing a fully AP patterned neural tube (Fig. 1A). These results are consistent with the ability of ectopic FGF8 to induce a secondary axis in zebrafish embryos(Fürthauer et al.,1997).

By contrast to the effects of inhibition of FGF signaling by interfering with receptor or downstream transduction function, ventralization has not been reported on loss of function of fgf3 and fgf8 in the zebrafish (Phillips et al.,2001; Maroon et al.,2002; Léger and Brand,2002). This failure might be due to redundancy with fgf17b and fgf24, which are also expressed in the embryonic shield (Cao et al., 2004; Draper et al., 2003). However,our attempts to co-inject MOs specific for any three or all four of these transcripts always resulted in non-specific early developmental arrest.

Although previous studies in which FGF signaling was inhibited by XFD expression throughout whole zebrafish and Xenopus wild-type embryos did not eliminate formation of anterior tissues(Amaya et al., 1991; Amaya et al., 1993; Griffin et al., 1995; Kroll and Amaya et al., 1996),there is substantial evidence that FGF signaling has a crucial role in DV patterning in these embryos (Mitchell and Sheets, 2001; Fürthauer et al., 1997; Fürthauer et al., 2004; Koshida et al.,2002; Londin et al.,2005). Here we show, in addition, that FGF signaling is required for β-catenin-mediated dorsal axis formation, a finding analogous to the demonstration that FGF signaling is required for Xenopus trunk organizer function (Mitchell and Sheets,2001), but we also show that inhibition of FGF signaling in the zebrafish embryo prevents β-catenin induction of anterior tissue.

We describe above a requirement for FGF signaling in at least three distinct steps in the network of response to β-catenin that leads to formation of the dorsal axis (see Fig. 9): (1) chd induction and BMP gene repression by Nodal proteins; (2) chd induction (but not BMP repression) by Boz; and (3)maintenance of β-catenin-induced boz transcript.

First, FGF signaling has an essential role in the pathway,β-catenin→Sqt→FGFs→Chd. We found that β-catenin can induce zygotic expression of sqt(Kelly et al., 2000), that Sqt induces fgf3 and fgf8(Fig. 3D,J) and that Boz induction of fgf3 and fgf8(Fig. 3E,K) requires sqt (Fig. 3F,L).β-catenin induction of fgf3 and fgf8 is eliminated by sqtMO plus cycMO but not by pNAboz(Fig. 3N-P,R-T). We conclude that that Nodal gene expression is necessary and sufficient for early dorsal fgf3 and fgf8 expression. We found that β-catenin, Sqt,FGF3 and FGF8 can each induce chd and repress bmp4 in ichabod embryos (Fig. 5; Fig. 7A-C,F,I,L), that XFD inhibits chd induction and bmp4 repression by β-catenin and Sqt(Fig. 7C-N), and that partial rescue of ichabod by FGF3 or FGF8 is dependent on chdexpression (Fig. 6). As FGF3 or FGF8 is capable of inducing a dorsal axis with full AP pattern in ichabod embryos (Fig. 1A), we conclude that FGF signaling is necessary and sufficient for chd expression under conditions in which it is sufficient to induce a partial dorsal axis. Sqt has previously been shown to induce chd in zebrafish mesodermal precursors(Mathieu et al., 2004), and FGFs have been shown to induce chd in early embryos(Koshida et al., 2002; Londin et al., 2005). Dorsal axis formation is thus dependent on theβ-catenin→Sqt→FGFs→Chd pathway, with each component sufficient for expression of the next downstream factor.

Second, FGF signaling is required for Boz-stimulated chdexpression. It was previously shown that chd expression is dependent on both Sqt and Boz in zebrafish embryos, and that each of these factors is sufficient to obtain some chd expression(Sirotkin et al., 2000; Shimizu et al., 2000). Based on the ability of XFD to completely eliminate chd expression in Sqt or Boz RNA-injected ichabod embryos, we conclude that FGF signaling is required for Sqt and Boz stimulation of chd(Fig. 6I-K,O-Q).

Third, FGF signaling is required for β-catenin dependent accumulation of boz transcript (Fig. 8A-D,L-Q). As β-catenin can induce boz in ichabod embryos, and Boz expression can dorsalize as well as induce anterior neural tube formation in the absence of FGF signaling(Fig. 4G,J), it was unclear why inhibition of FGF signaling would completely block β-catenin rescue of ichabod embryos. One possibility, that FGF signaling was required forβ-catenin-stimulation of boz, was indeed shown to be the case,as XFD and MKP3 blocked the accumulation of β-catenin-induced boz transcript in early ichabod embryos(Fig. 8A-D,L-Q) and MKP3 was effective in inhibiting boz expression in wild-type embryos(Fig. 8R,S). As boz is thought to be a direct target of β-catenin-activated transcription, based on the presence of Tcf/Lef consensus binding sites in the bozpromoter and analysis of expression of boz promoter fusion constructs(Ryu et al., 2001; Leung et al., 2003b), a role for FGFs in boz expression was unexpected. Indeed, we did not detect an effect of inhibition of FGF signaling on the initial expression of boz just before high stage (Fig. 8J,K). FGF signaling thus appears to be required for stability and maintenance of boz transcript in the early embryo but not for initial transcriptional activation. The profound combination of loss of chdand boz function shown here is very similar to that observed in boz;chd double mutants, which lack both head and trunk(Gonzalez et al., 2000). These embryos have more ventralized phenotypes than boz;sqt, boz;cyc, or even boz;sqt;cyc embryos(Sirotkin et al., 2000; Shimizu et al., 2000), and are almost as severely affected as ichabod Class 1 mutants(Kelly et al., 2000).

Chd is essential for FGF partial rescue of ichabod embryos, but not for β-catenin-dependent dorsal axis formation. Loss of function of chd alone (i.e. dino mutants) results in a moderately ventralized phenotype (Hammerschmidt et al., 1996; Fisher et al.,1997; Schulte-Merker et al.,1997) and β-catenin rescue of ichabod in the presence of chdMO produces the same phenotype(Fig. 6C,J). The severity of phenotype of loss of chd function is undoubtedly mitigated by the expression of other secreted inhibitors of BMP and Wnt ligands (reviewed by De Robertis and Kuroda, 2004; Schier and Talbot, 2005), all of which are expressed in the organizer region. As chdMO can completely prevent FGF rescue in ichabod embryos, FGFs are probably not sufficient for expression of BMP antagonists other than Chd.

Fig. 9 summarizes theβ-catenin inducible, FGF signaling-dependent pathways that operate in the zebrafish embryo to induce chd and repress BMP function. As discussed above, the pathway has two parallel branches, with one being directly dependent on induction of sqt and FGF genes, the other on boz. Two versions of these pathways are illustrated in Fig. 9, one in which FGFs induced by Squint induce chd as a consequence of inhibition of BMP signaling (Fig. 9A), the other in which chd is induced more directly by FGFs(Fig. 9B); our data are consistent with either of these models.

A potential unifying mechanism that would explain the repressive effects of FGF signaling on BMP repression and consequent induction of chd is provided by observations that FGF signaling promotes MAPK phosphorylation of the linker region of Smad1, counteracting the effects of BMP receptor-mediated C-terminal Smad1 phosphorylation (Kretzschmar et al., 1997; Pera et al., 2003; Kuroda et al., 2005). As BMP gene expression is dependent on a positive regulatory loop(Kishimoto et al., 1997; Nguyen et al., 1998; Schmid et al., 2000), and upregulation of chd has been found to be a consequence of low BMP levels, even in the absence of functional Chd protein(Schulte-Merker et al., 1997; Yabe et al., 2003), the FGF-mediated chd induction we observed may be dependent on interference with BMP signaling by MAPK phosphorylation of Smad1. Further experiments, however, are required to rule out a more direct effect of FGF signaling on chd expression in the sqt branch of the pathway(as in Fig. 9B). FGF-dependent BMP repression in the absence of chd expression in the early zebrafish embryo has been reported(Fürthauer et al., 2004; Londin et al., 2005), but these experiments were carried out using high levels of injected FGF RNA(25-50 pg), which we found to produce disorganized, albeit dorsalized, embryos that did not form a recognizable axis.

Most of our data are consistent with this model, as chd induction was always accompanied by loss of BMP transcripts(Fig. 5D-X). However, the finding that chd induction was not observed even though bmp4expression was repressed when embryos were injected with boz RNA under conditions of inhibition of FGF signaling(Fig. 6O-T) suggests that Boz induction of chd cannot solely be mediated by FGF pathway repression of BMP gene expression. In Fig. 9A,B, Boz is thus shown to have both FGF-dependent chd-inducing and FGF-independent BMP-repressing effects.

We thank Joshua Bradner, Jiangyan He, Richard Gill, Jr and Carol Hu for fish care and general laboratory assistance; Tetsuru Kudoh for reagents; and Michael Tsang, Mary Mullins and Daniel Kessler for helpful suggestions. This work was supported by NIH grant HD39272.