Combinatorial Fgf and Bmp signalling patterns the gastrula ectoderm into prospective neural and epidermal domains

In vertebrates, the organiser and its axial derivatives have the ability to induce and pattern neural tissue when transplanted elsewhere within the embryo(Munoz-Sanjuan and Hemmati-Brivanlou,2002; Stern,2001). The identification of secreted inhibitors of Bmp proteins expressed in these organiser cells led to proposition of the `neural default model', which suggests that Bmp signals induce epidermal fate whereas Bmp antagonists induce neural fate(Munoz-Sanjuan and Hemmati-Brivanlou,2002). However, genetic analyses in zebrafish have revealed a more complex role for the Bmp pathway in early ectodermal patterning(Hammerschmidt and Mullins,2002). These and related studies in other species have shown that high levels of Bmp signalling promote epidermal identity, whereas neural crest, lateral neural and more medial neural identities are established at progressively lower levels of Bmp activity(Aybar and Mayor, 2002; Barth et al., 1999; Nguyen et al., 2000). In all vertebrate model organisms that have been examined, most neural tissue originates close to the organiser (on what is termed the dorsal side of the gastrula in fish and frogs), supporting the notion that neural induction is dependent upon organiser-derived signals including Bmp antagonists. However,spinal cord neural tissue of zebrafish embryos originates distant to the organiser (Kimmel et al.,1990) and in this study, we show that the initial steps in its specification appear to occur independent of organiser-derived signals.

Although the Bmp signalling pathway is likely to play a role in neural and epidermal development in all vertebrates, a variety of studies have challenged the idea that antagonism of Bmp activity is sufficient to induce neural tissue(Stern, 2001; Stern, 2002; Wilson and Edlund, 2001). Indeed, several other signals may have roles in neural induction(Bally-Cuif and Hammerschmidt,2003) and among these, Fgfs are perhaps the most extensively studied candidates (Akai and Storey,2003). Fgfs can induce neural (or prospective neural) identity both in stem cells (e.g. Ying et al.,2003) and when ectopically expressed in vivo (e.g. Bertrand et al., 2003; Furthauer et al., 1997; Kengaku and Okamoto, 1995; Koshida et al., 2002; Lamb and Harland, 1995; Sheng et al., 2003; Streit et al., 2000; Wilson et al., 2000); in addition, abrogation of Fgf activity can in some circumstances disrupt neural induction (e.g. Bertrand et al.,2003; Hongo et al.,1999; Streit et al.,2000; Wilson et al.,2000).

In this study, we attempt to resolve the relative contributions of organiser-derived Bmp antagonists and Fgf signals to the initial steps of neural induction in zebrafish embryos. We show that Fgf activity, rather than Bmp antagonism, initiates development of prospective vegetal neural tissue that contributes to trunk and tail CNS. In vegetal prospective neural tissue,Bmp activity does not antagonise the induction of prospective neural markers,rather it promotes the ability of cells to contribute to caudal neural ectoderm. Therefore, at gastrula stages, the role of Bmp activity in the animal and vegetal ectoderm is different. In animal ectoderm, high levels of Bmp activity push cells towards a non-neural rather than neural fate whereas in vegetal ectoderm, differential levels of Bmp activity influence the regional (rostral to caudal) character of the neural tissue.

RNA probes were synthesised, and in situ hybridisation procedures were performed as previously described (Itoh et al., 2002; Kudoh et al.,2002).

Zebrafish embryos were injected with caged fluorescein (Molecular Probes)at the one-cell stage. Photoactivation of fluorescent dye was performed as previously described (Rohr and Concha,2000) at around the 70-80% epiboly stage. Following photoactivation, DIC and fluorescent images were acquired using Openlab software (Improvision) with a cooled CCD camera (Hammamatsu) attached to an Axioplan microscope (Zeiss).

Capped RNA was synthesised by mMessage mMachine SP6 kit (Ambion) according to the manufacturer's instructions. RNA concentrations used for injection were: bmp2b (50 pg); fgf3 (50 pg); XFD (300 pg); chordin (50 pg); and eGFP (100 pg). mRNAs were injected into cells at one- to two-cell stage in all blastomeres while the morpholino for the chordin gene (500 pg) (Nasevicius and Ekker, 2000) was injected at the one- to two-cell stage in the yolk. More than 15 embryos were examined in each injection experiment.

For transplantation, donor embryos were injected with various RNA constructs and cells removed at blastula or early gastrula stages using a microelectrode connected to a Hamilton syringe as previously described(Houart et al., 2002). Donor cells were gently aspirated into various positions of similar stage host embryos. To detect donor cells expressing GFP, embryos were labelled by peroxidase conjugated anti-GFP antibody (×1000 dilution) (TP401, Torrey Pines Biolabs) and detected by DAB substrate reaction after the in situ hybridisation protocol. In cases where two populations of cells were co-transplanted, donors were distinguished by injection with either fluorescein dextran (green) or rhodamine dextran (red).

Donor embryos injected with constructs encoding both Fgf3 and the truncated Fgf receptor XFD, did not show an Fgf gain-of-function phenotype and instead showed a phenotype similar to embryos expressing XFD alone. This suggests that cells from these embryos produce, but are compromised in their ability to respond to, Fgf signals. In induction assays, it is likely that Fgf3 from these cells acts directly on host cells as opposed to acting indirectly by inducing the expression of other signals within the donor cells.

If the primary role of Bmp signalling is to antagonise neural development,then we reasoned that all neural tissue should form in regions of low Bmp activity. To determine if this is indeed the case, we used marker gene and fate mapping approaches to identify ectodermal cells destined to form CNS tissue in gastrula stage zebrafish embryos. Early markers of prospective neural tissue, such as sox3(Kudoh et al., 2001) (GenBank Accession Number, AY316135), sox31(Girard et al., 2001) and zic2.2 (Toyama et al.,2004), are not only expressed dorsally at gastrula stages, but also in a radial band of cells along the vegetal edge of the ectoderm,adjacent to the prospective mesoderm (Fig. 1A-F). Expression of the prospective epidermal markers, foxi1 (Solomon et al.,2003) and p63 is complementary to that of these prospective neural genes, leaving a band of ventral vegetal ectodermal cells free of transcripts (Fig. 1G,H)(Bakkers et al., 2002). On the ventral side of the gastrula, the expression of prospective neural genes overlaps with that of bmp2b and bmp4, as well as eve1, a putative target of Bmp signalling(Joly et al., 1993),suggesting that Bmp activity is compatible with prospective neural gene expression (Fig. 1I-K). Altogether, these results show that ventrovegetal ectoderm expresses prospective neural rather than epidermal markers, suggesting that some prospective neural tissue is specified in a region of potentially high Bmp activity.

To determine if ventral ectoderm does indeed form neural tissue, we performed cell fate mapping of ectoderm in midgastrula stage embryos. At 50%epiboly, cells at all DV positions of the ectoderm can give rise to neural tissue with at least some prospective dorsal spinal cord cells located ventrally (Kimmel et al.,1990). To extend these studies, we determined the fates of ectodermal cells at mid-gastrula (70-80% epiboly), a stage by which early prospective neural genes and epidermal genes are expressed in complimentary domains (Fig. 1). Cell-fate mapping was performed by uncaging fluorescein within vegetal ectodermal cells,or transplanting biotin-dextran-labelled cells into the vegetal ectoderm, and determining the fate of the labelled cells on the following day. These experiments showed that ventrally located vegetal ectoderm gives rise to caudal spinal cord, whereas progressively more dorsal vegetal ectoderm contributes to progressively more rostral spinal cord and hindbrain(Fig. 2A-F). Next, to determine if any dorsal cells give rise to tail neural tissue, we uncaged fluorescein within the entire dorsal vegetal ectoderm (and underlying mesendoderm)(Fig. 2G), In these cases(n=8), fluorescent cells were observed widely in the head and trunk,whereas in the posterior part of the tail, fluorescent cells were restricted to the floor plate, notochord and hypocord, and were absent from the rest of the neural tube (Fig. 2H). Therefore, the tail floor plate is exceptional in being derived from dorsal tissue (Amacher et al., 2002; Shih and Fraser, 1995; Shih and Fraser, 1996),whereas other regions of the tail neural tube are derived from ventral,vegetal ectoderm.

These experiments allow us to conclude that vegetal ectodermal cells on the ventral side of the gastrula, which are distant from the organiser and within a region of bmp expression, do contribute to the tail spinal cord.

Most models suggest that neural induction is dependent upon secreted molecules, notably Bmp antagonists, emanating from the dorsal organiser(Kodjabachian et al., 1999; Munoz-Sanjuan and Hemmati-Brivanlou,2002); but the presence of prospective neural tissue in the ventrovegetal region of the gastrula challenges this idea. We therefore performed several analyses to test if Bmp antagonists or other organiser-derived signals are necessary for the expression of markers of the vegetal, prospective neural ectodermal domain. Vegetal expression of the prospective neural markers sox3 and zic2.2 is unaffected following overexpression of bmp2b (n=55), in mutant embryos lacking the activity of the Bmp antagonist, Chordin (din)(n=9), and in ichabod^-/- embryos (n=20)that are defective in β-catenin signalling and entirely lack the organiser (Hammerschmidt and Mullins,2002; Kelly et al.,2000) (Fig. 3A-J). In all these situations, rostral neural tissue is reduced or absent, whereas expression of the epidermal marker p63 extends further dorsally consistent with previous observations (Fig. 3K-O) (Hammerschmidt and Mullins, 2002; Kelly et al.,2000). By contrast, irrespective of the severity of the ventralised phenotype, p63 expression does not expand vegetally,always retaining a gap between its expression domain and that of ntlin the nascent mesoderm (Fig. 3O). These results indicate that expression of markers of prospective vegetal neural tissue is not dependent upon the organiser. They also suggest that induction of these markers is either unaffected by high levels of Bmp signalling or that Bmp activity in vegetal regions of the gastrula can be antagonised by mechanisms independent of the organiser.

As vegetal ectodermal markers are expressed around the entire circumference of the gastrula, signals from the germ ring (the blastoderm margin) are likely to contribute to their induction. Fgf genes are good candidates for encoding these signals as several are expressed in the germ ring(Draper et al., 2003; Furthauer et al., 1997; Koshida et al., 2002) and previous studies have implicated this pathway in neural induction(Akai and Storey, 2003; Stern, 2001; Wilson and Edlund, 2001). We therefore assessed the requirement for Fgf signalling in the induction of vegetal sox3 expression.

Expression of vegetal ectodermal markers was examined in embryos that had been treated from the 1024-cell stage with SU5402, an inhibitor for Fgf receptor activity (Mohammadi et al.,1997) that should abrogate activity of zygotically encoded Fgf signals emanating from the germ ring. In embryos treated with 60 μM SU5402,vegetal sox3 expression is variably decreased, whereas the animal(anterior) domain of sox3 expression expands vegetally on the dorsal side of the embryo (Fig. 4A-D). With a higher dose of SU5402 (90 μM), more than 90% of embryos showed a severe phenotype equivalent to that shown in Fig. 4D (n>25). In these embryos, the vegetal ectodermal markers sox31 and zic2.2 behaved similarly to sox3, showing loss of ventrovegetal expression and retention of dorsal expression(Fig. 4E-H). Supporting the notion that the remaining prospective neural tissue on the dorsal side of these embryos has anterior character, expression of otx2, which is restricted to the prospective anterior neural tissue of gastrula stage wild-type embryos, is expanded vegetally on the dorsal side of the embryo(Fig. 4I,J) (see also Koshida et al., 1998; Kudoh et al., 2002). Complementing the changes in prospective neural genes, expression of the prospective epidermal marker, p63, was expanded ventrovegetally into the territory from which prospective neural marker gene expression is lost(Fig. 4K,L). Similar results(Fig. 5A-C), were obtained through overexpression of a truncated, dominant-negative Fgf receptor, XFD(Amaya et al., 1991; Griffin et al., 1995). These data support the idea that induction of vegetal prospective neural tissue is dependent on Fgf activity and that the residual neural tissue that remains after Fgf receptor blockade has anterior character.

Many previous studies have shown that enhanced Bmp activity during gastrulation can suppress development of rostral neural tissue(Fig. 3) (reviewed by Hammerschmidt and Mullins,2002), and the experiments described above show that suppression of zygotic Fgf activity can suppress caudal neural development. Consistent with these observations, embryos in which Fgf activity was suppressed, while Bmp activity was enhanced showed complete absence of sox3 expression(Fig. 5D).

Complementing the Fgf loss-of-function studies, increasing Fgf3 activity leads to induction of ubiquitous ectodermal expression of sox3, sox31and zic2.2 (Fig. 5E,H,K) (Koshida et al.,2002; Furthauer et al.,1997) and this induction is unaffected by overexpression of bmp2b (Fig. 5F,I,L)(>90%, n>15). This suggests either that Fgf can induce expression of prospective neural genes in the presence of high levels of Bmp activity or that Fgf activity can block Bmp signalling.

Previous studies have shown that one mechanism by which Fgf signalling can indirectly induce neural tissue is by antagonising Bmp activity(Furthauer et al., 1997; Koshida et al., 2002; Pera et al., 2003; Wilson et al., 2000). We consequently asked if abrogation of Bmp signalling by Fgf activity is necessary for Fgf activity to induce sox3 expression. Overexpression of fgf3 leads to widespread induction of chd expression,which contributes to suppression of bmp4 expression [and presumably lowered Bmp activity (Koshida et al.,2002); Fig. 5N,Q;>90%; n>30]. In this situation, the expanded expression of prospective neural markers (Fig. 5E,H,K) could, therefore, be due to suppression of Bmp activity. However, co-expression of bmp2b in embryos overexpressing fgf3 suppresses chd expression and expands bmp4expression (Fig. 5O,R)(>80%, n>30), and yet despite this, the ectoderm still ubiquitously expresses prospective neural markers(Fig. 5F,I,L). Therefore, the Bmp pathway can be activated in the presence of Fgf signalling without affecting induction of prospective neural gene expression. These data indicate that although Fgf activity can suppress Bmp signalling by promoting expression of Bmp antagonists [and by other means (e.g. Pera et al., 2003)], this is not necessary for Fgf signalling to be able to induce markers of prospective neural identity.

To further explore the epistatic relationships between the Fgf and Bmp pathways, we examined the consequences of locally activating or suppressing Fgf signalling. fgf3-expressing ectodermal cells transplanted into animal pole regions of host embryos induce sox3 non-autonomously in surrounding host cells (Fig. 6A; >80%; n>20). This induction still occurs if the donor cells are from embryos co-expressing a truncated Fgf receptor(Fig. 6B), suggesting that the Fgf signal from the donor cells acts directly on the host (100%; n=19). When both donor and host cells are overexpressing bmp2b,fgf3-expressing cells still induce sox3 and suppress expression of the epidermal marker gene, foxi1(Fig. 6C,D; 100%; n=22), again suggesting that exogenous Bmp activity does not block induction of prospective neural marker genes by Fgf.

Next, we locally suppressed Fgf signalling by transplanting XFD-expressing donor cells to various positions in the prospective neural ectoderm of host embryos. sox3 expression was suppressed in XFD-expressing cells transplanted to dorsal, lateral or ventral vegetal ectoderm(Fig. 6E-J) (>80%n=30), and foxi1 was ectopically induced in transplants targeted to ventral-vegetal ectoderm (Fig. 6K,L) (>70%; n=27). These results suggest that ventral vegetal ectoderm needs to receive Fgf to express sox3, otherwise it expresses the prospective epidermal marker, foxi1.

To directly assess if Fgf signals are essential for vegetal ectoderm to form neural tissue, we traced the eventual fate of XFD-expressing donor cells transplanted to wild-type hosts. In these experiments, labelled wild-type cells were co-transplanted with XFD-expressing cells to the same locations in the vegetal ectoderm of unlabelled host embryos at the end of blastula stage. When the donor cells were transplanted to dorsal side, wild-type cells primarily contributed to the hindbrain, whereas the XFD-expressing cells localised more anteriorly, mainly in the midbrain (n=14). However,when transplanted to ventral vegetal ectoderm, wild-type donor cells contributed to spinal cord and muscle (see also Kimmel et al., 1990) whereas XFD-expressing cells were excluded from the CNS and found in tissues such as the epidermis and fin (n=8). These results suggest that Fgf signalling is required for vegetal ectoderm to contribute to caudal neural tissue. They also suggest that the consequences of suppression of Fgf signalling in cells in dorsal and ventral domains of the vegetal ectoderm are different: dorsally, cells with compromised Fgf signalling frequently move into anterior neural tissue; ventrally, cells move into the prospective epidermis and are excluded from neural tissue. These results are consistent with analyses of embryos in which XFD is expressed ubiquitously (e.g. Griffin et al., 1995)(Fig. 3) and which show loss of posterior neural structures and anteriorisation of remaining CNS tissue on the dorsal side of the embryo.

If Bmp signalling can still occur in vegetal ectoderm without suppressing sox3 expression, then this raises the issue of what Bmp activity is doing in this domain. Within animal pole ectoderm, Bmp activity promotes expression of markers of ventral, non-neural, ectodermal fates (such as p63 and foxi1) at the expense of expression of dorsal neural markers [such as otx2(Hammerschmidt and Mullins,2002); Fig. 7B,parts i,iv; C, parts i,iv]. In vegetal ectoderm, DV organisation does not correspond to neural versus non-neural fates, but rather is predictive of the anterior-to-posterior (AP) position that neural cells will occupy within the caudal CNS (Fig. 2). This raises the possibility that in vegetal ectoderm, Bmp activity promotes regional specification within the prospective neural ectoderm rather than neural/non-neural fate specification. Supporting a global role for Bmp signalling in promoting ventral fates in the gastrula, irrespective of their neural or non-neural destiny, expression of the ventral vegetal marker eve1 is induced by overactivation of Bmp signalling, whereas the dorsal vegetal marker hoxb1b is suppressed[Fig. 7B, parts ii,iii; C, part ii,iii; similar results have been observed in mutants or embryos with other modulations affecting Bmp signalling (e.g. Bakkers et al., 2002; Hammerschmidt et al., 1996; Mullins et al., 1996)]. As cells within the domain of eve1 expression of wild-type embryos give rise to more posterior neural fates than cells within the ectodermal domain of hoxb1b expression, then this suggests that graded Bmp activity in vegetal ectoderm contributes to the allocation of fates along the AP axis of the neural tube rather than to a neural versus non-neural fate choice.

These results suggest that the consequences of graded Bmp activity are different in vegetal gastrula stage ectoderm, which is exposed to Fgf signals from the germ ring than in animal ectoderm (which we predict is unaffected by these Fgf signals). To test this hypothesis further, we manipulated the levels of Bmp activity in embryos in which Fgf signalling was either ubiquitously activated or was suppressed. When Fgf activity is suppressed, the entire ectoderm expresses either the prospective epidermal marker p63 or the prospective forebrain marker otx2, depending on whether Bmp activity is enhanced or suppressed (Fig. 7D,G). These data are consistent with the role of endogenous Bmp signalling in animal pole ectoderm of wild-type embryos. Conversely,ubiquitous activation of Fgf3 signalling leads to either ubiquitous eve1 or hoxb1b expression, again depending on the levels of Bmp activity [Fig. 7E,F;induction of hoxb1b by fgf3 is also reported elsewhere(Koshida et al., 2002)]. These data are consistent with the role we predict for endogenous Bmp signalling within vegetal ectoderm of wild-type embryos.

To explore the possibility that in vegetal regions of the ectoderm, Bmp signalling promotes the adoption of caudal neural fate, we examined marker gene expression and fates of vegetal ectoderm in din^-/-mutants/morphants in which Bmp activity is enhanced. In din^-/- embryos, the caudal marker, eve1 is expanded to lateral vegetal ectoderm(Hammerschmidt et al., 1996)(Fig. 8B), whereas vegetal sox3 expression is not significantly altered(Fig. 3B, Fig. 8D). Furthermore, while exogenous Fgf3 induces sox3 throughout the ectoderm both in din^-/- and in wild-type embryos(Fig. 8G,H), eve1 is suppressed in wild-type embryos (n=37), whereas expression expands throughout most of the sox3-expressing ectoderm in din^-/- mutants (n=13)(Fig. 8E,F) with the exception of the dorsal most ectoderm which retains hoxb1b expression in this condition (Koshida et al.,2002). These data suggest that a key role for Chordin in the vegetal ectoderm is to suppress eve1 and tail formation, and raise the possibility that cell fate in the lateral vegetal ectoderm may be changed from trunk neural to tail neural fate in din^-/- mutants. Indeed, the size of trunk may be smaller and tail bud larger in din^-/- embryos compared with wild-type embryos(Hammerschmidt et al., 1996; Myers et al., 2002).

To directly assess if the expansion of eve1 in lateral vegetal ectoderm of embryos lacking Chordin activity correlates with a switch from prospective trunk to prospective tail identity, the fate of lateral vegetal ectoderm was assessed in din morphants. As expected, lateral vegetal ectodermal cells contributed primarily to the trunk spinal cord in wild-type embryos (Fig. 2D,E,F; Fig. 8I,J,K). By contrast, the majority of equivalently positioned vegetal ectoderm cells in dinmorphants contribute to the tail, including the most caudal spinal cord, a region that does not receive contribution from lateral ectodermal cells in wild-type embryos (n=11) (Fig. 8L-N). Therefore in the absence of Chordin function, equivalently positioned ectodermal cells tend to contribute to more caudal regions of the CNS than in wild type, supporting the idea that Bmp activity affects the allocation of regional AP fates in the vegetal ectoderm.

Our study has revealed that combinatorial activity of the Bmp and Fgf signalling pathways during blastula and gastrula stages contributes to the subdivision of the embryonic ectoderm into prospective neural and epidermal domains. We show that in vegetal ectoderm, initiation of neural fate specification requires Fgf activity and can occur in the presence of Bmp signalling activity (Fig. 9). Indeed, the regulation of prospective neural/non-neural marker gene expression in vegetal ectoderm appears to be independent of all dorsal organiser activity, as it still occurs in the most severely ventralised ichabodmutant embryos in which β-catenin-dependent development of dorsal structures fails to occur (Kelly et al.,2000). This study shows that both Fgfs and Bmp antagonists can promote neural development and may help reconcile the differing mechanisms proposed to regulate early neural development and patterning in amniotes and anamniotes.

We find that Bmp activity promotes fates of cells positioned on the ventral side of the gastrula ectoderm irrespective of whether or not these cells are destined to contribute to neural or non-neural structures. In vegetal ectoderm, Bmp activity promotes tail neural fate at the expense of trunk neural cell fates whereas towards the animal pole, it promotes epidermal at the expense of neural fates (Fig. 9). Other studies have shown that Bmp activity promotes caudal and ventral mesodermal fates (Munoz-Sanjuan and Hemmati-Brivanlou, 2001; Myers et al., 2002), and so Bmp signalling may have graded activity throughout all germ layers, promoting all fates that lie on the ventral side of the gastrula. Our data helps to reconcile how high levels of Bmp activity during gastrulation can promote development of both epidermis and tail neural tissue by showing that these alternative consequences are dependent upon the level of Fgf activity. Given that the consequences of activation of the Fgf and Bmp pathways differ in different regions of the gastrula, it will be important to resolve how these two signalling pathways influence cell fate.

Given that Bmp activity may not directly specify cell identity (as it promotes several distinct fates for ventrally positioned cells), it is intriguing to consider how this pathway can promote the adoption of caudal neural fate. An attractive possibility is that it does so by regulating the movements of gastrulating cells. Solnica-Krezel and her colleagues(Gonzalez et al., 2000; Myers et al., 2002) have analysed movements of gastrulating cells in embryos with different levels of Bmp activity. They have shown that laterally positioned cells in the wild-type gastrula move dorsally, whereas the ventral-most cells fail to converge dorsally and consequently end up in the tail bud area. In din^-/- embryos, the domain of non-convergence is expanded laterally with the likely consequence that more cells end up in the tail bud. Therefore, if Bmp-dependent regulation of dorsal convergence also occurs within the ectoderm, then this model provides an attractive explanation of why laterally positioned prospective neural cells have a higher likelihood of ending up in the tail spinal cord in din^-/- embryos.

If the role of the Bmp pathway in caudal development is primarily to regulate cell movements, then other pathways may cooperate with Bmps to impose caudal cell fates. Several candidates are well documented(Munoz-Sanjuan and Hemmati-Brivanlou,2001), and in fish, Agathon and colleagues(Agathon et al., 2003) have recently shown that entire tail structures could be induced by simultaneous activation of Bmp, Nodal and Wnt pathways. Although a role for Wnt signalling in posterior development is supported by many other studies (e.g. Erter et al., 2001; Hashimoto et al., 2000; Marlow et al., 2004), a requirement for Nodal signalling to form tail tissue is less clear given that fish embryos lacking Nodal activity do still form tails that contain neural tissue (Feldman et al., 2000). Nevertheless, it is likely that cooperation between several signalling pathways (e.g. Haremaki et al.,2003) is required for ventrally positioned cells to move and adopt appropriate fates within the tail.

To our knowledge, a source of tail neural tissue distant from the dorsal organiser has not been shown for other model species and it will be of interest to determine if there are similar roles for Bmp activity in promoting caudal neural fates in these species.

We show that Fgf signalling promotes expression of prospective neural markers and is required for ectodermal cells to contribute to caudal CNS structures. This is consistent with data in other species that suggest Fgfs are important regulators of neural induction(Bertrand et al., 2003; Kengaku and Okamoto, 1995; Lamb and Harland, 1995; Streit et al., 2000; Wilson et al., 2000; Ying et al., 2003). However,there is a larger body of literature that indicates Fgf signalling promotes caudalisation of neural tissue (Bally-Cuif and Hammerschmidt, 2003; Cox and Hemmati-Brivanlou, 1995; Koshida et al., 1998; Kudoh et al., 2002) and it will be important to determine if these two activities are separable.

In this study, we show that abrogation of Fgf activity leads to loss of prospective neural tissue on the ventral side of the embryo (supporting a role for Fgf signalling in an early step of neural induction) and also anteriorisation of the residual neural tissue on the dorsal side of the embryo(supporting a role for Fgf in posteriorisation) (see also Koshida et al., 1998; Kudoh et al., 2002). Some of our data and other reports suggest that Fgf-dependent induction of prospective neural fate and posteriorisation of induced neural tissue are separable events. For example, induction of expression of the posterior neural gene, hoxb1b by Fgf3 depends on Chordin function(Koshida et al., 2002) and is probably mediated indirectly by retinoic acid(Kudoh et al., 2002). By contrast, induction of the prospective neural markers sox3, sox31 and zic2 is independent of Chordin activity (this study). Furthermore,Koshida et al. (Koshida et al.,2002) have shown that XFD-expressing cells transplanted to wild-type hosts are still able to express hoxb1b, supporting an indirect role for Fgf on hoxb1b expression. However, our results suggest that ventral vegetal ectodermal cells may directly need to receive Fgf signalling to specify neural fate (to induce sox3, to suppress foxi1 and to contribute to caudal CNS). Therefore, at least in part,the roles of Fgf signalling in neural induction and posteriorisation by Fgf seem to be separable.

Although the regulation of early neural/non-neural marker gene expression in vegetal ectoderm is independent of Bmp antagonism, this does not discount the possibility that Bmp antagonists may be contributing to the maintenance and/or further differentiation of neural plate. Although there has not been any detailed analysis of mature neural tissue in ichabod mutant or other severely ventralised embryos, it appears that relatively few neurones are present in these fish (Gonzalez et al., 2000) (T.K., unpublished), despite the early expression of vegetal prospective neural markers. This suggests that exposing ectodermal cells to Fgf signalling alone at blastula and gastrula stage is not sufficient to maintain and promote further differentiation of the neural tube. Bmp antagonists such as Chordin and Noggin 1 may therefore contribute to later steps in the induction, maturation and/or maintenance of neural tissue. Indeed, the idea that Chordin is not necessary for the initial phase of neural induction but rather functions at a later step of neural development has been suggested from neural induction assays in chick(Stern, 2002; Streit et al., 1998).

In our studies, we noticed that although exogenous Fgf induced early prospective neural marker genes it actually appeared to inhibit the expression of markers of mature neuronal identity such as huC (data not shown). This is of interest in light of recent observations that, in chick, levels of Fgf activity are believed to be high in immature neural tissue and low in mature neural tissue (Diez del Corral et al., 2003). Indeed Diez del Corral et al. propose that inhibition of Fgf activity is a necessary step in the maturation of neural tissue that allows neuronal differentiation to occur. Thus, although exogenous Fgf activity may promote early steps in neural development (induction of expression of prospective neural markers and suppression of prospective epidermal markers), it may inhibit this prospective neural tissue from progressing to a fully differentiated state. Furthermore, it has recently been shown that Sox protein activity in undifferentiated neural tissue inhibits neurogenesis, and that this activity needs to be suppressed through the activity of proneural bHLH proteins(Bylund et al., 2003). Therefore as expression of Sox genes is promoted by Fgf signalling (this study) (Streit et al., 2000),this may provide an explanation of how Fgf can promote early neural fate yet inhibit later neuronal differentiation.

The molecular mechanisms by which Fgfs and Bmp antagonists interact to regulate prospective neural/non-neural gene expression are undoubtedly complex. In our study, we have shown that Fgf3 can induce sox3 in ectodermal cells without suppressing the ability of the Bmp pathway to induce bmp4 and suppress chordin expression. This indicates that Fgf activity does not lead to a comprehensive block of Bmp signalling. However, data from other studies have shown intracellular crosstalk between Fgf and Bmp pathways by which Fgf pathway activation can antagonise Bmp pathway activity. For example, in frogs, Map kinase, a downstream effector of Fgf signalling, phosphorylates a linker domain in Smad1 thereby suppressing Bmp signalling (Pera et al.,2003). Assuming similar Fgf-dependent regulation of Smad1 activity occurs in fish, then either this is insufficient to abrogate Bmp signalling,or there may be several independently regulated intracellular responses downstream of Bmp receptor activation. In addition to intracellular crosstalk between these two pathways, Fgf signalling can profoundly affect Bmp signalling activity through the regulation of expression of secreted Bmp antagonists such as Chordin (Koshida et al., 2002; Furthauer et al.,1997) (this study).

In our study and in related experiments (e.g. Griffin et al., 1995; Koshida et al., 1998), XFD injection and SU5402 treatment both suppressed posterior neural induction,whereas anterior neural fate was retained in both situations. This is consistent with some experiments in other species showing a requirement for Fgf activity only in posterior neural development(Munoz-Sanjuan and Hemmati-Brivanlou,2001). By contrast, in chick and ascidian (and from some experiments in frogs), it has been proposed that Fgf activity is crucial for induction of neural tissue, including anterior domains(Bertrand et al., 2003; Hongo et al., 1999; Hudson and Lemaire, 2001; Streit et al., 2000; Wilson et al., 2000). One possibility is that in fish, Fgf activity may be required in prospective anterior neural tissue at lower levels or at earlier stages than in more caudal neural tissue. If so, then sufficient early Fgf activity might still be present in manipulated embryos in our experiments to allow development of rostral neural tissue. Indeed, very recent studies have shown that early activation of Fgf signalling plays an important role in the regulation of the Bmp signalling pathway in prospective anterior neural tissue(Furthauer et al., 2004) and preliminary data suggest that abrogation of Fgf activity at stages earlier than reported here leads to more severe depletion of anterior neural tissue(T.K., unpublished).

In the animal (prospective anterior) gastrula stage ectoderm, Bmp antagonists emanating from the organiser, its derivatives and other tissues are crucial for neural development, whereas in prospective caudal regions,prospective neural fate appears to be specified by Fgfs emanating radially from the germ ring (Fig. 9). Therefore, we suggest that the organiser and the germ ring constitute two distinct and at least partially independent sources of signals that promote neural development. We predict that the combined activity of germ ring and organiser signals establishes prospective neural tissue from the dorsal animal ectoderm through to the ventral vegetal ectoderm. One consequence of this view of early neural induction is that the neural to non-neural ectodermal fate choice should not only be considered as occurring between dorsal and ventral ectoderm but also, on the ventral side of the gastrula, between animal and vegetal ectoderm.

Two very recent papers examine early activity and function of the Fgf pathway in zebrafish (Furthauer et al.,2004; Tsang et al.,2004). We briefly cite the Furthauer paper in the Discussion, but it contains additional data pertinant to our study.

We thank Eric Weinberg for providing ichabod mutant embryos;colleagues in the scientific community for providing reagents; Claudio Stern,Les Dale, Masa Tada, Anukampa Barth and members of our laboratories for discussions of this work; and the fish facility staff for care of the fish. This work was supported by grants from the Wellcome Trust and BBSRC to S.W. S.W. is a Wellcome Trust Senior Research Fellow. T.K. was initially funded by HFSP and JSPS fellowships. M.C. receives funding from FONDECYT (1020902) and is a Wellcome Trust IRDA Fellow.