The zebrafish fgf24 mutant identifies an additional level of Fgf signaling involved in vertebrate forelimb initiation

Vertebrate paired appendages, or limbs, develop from primordia called limb buds, that arise as localized outgrowths from the flank of the embryo. Two distinct populations of cells in the flank contribute to the limb bud. These are the mesenchymal cells of the lateral plate mesoderm (LPM), which will form the core of the bud, and the overlying ectodermal cells, which will form the epithelial jacket of the bud. Two important questions arising in the context of this development are: how is limb development initiated in these cells, and how is it coordinated spatially and temporally between the mesenchyme and the ectoderm (reviewed by Johnson and Tabin,1997; Martin,1998; Tickle and Munsterberg,2001; Capdevila and Izpisua Belmonte, 2001). Experimental removal of the intermediate mesoderm(IM), which lies between the somites and the LPM, leads to a reduction of limb structures in the chicken (Geduspan and Solursh, 1992), suggesting that the IM could be the source of a signal triggering limb development. The chicken FGF8 gene has been proposed to encode this signal, based on the observation that it is transiently expressed in the IM adjacent to the limbs, and that it, and similar Fgfs, can initiate the development of additional limbs from the chicken embryonic flank when ectopically expressed(Cohn et al., 1995; Ohuchi et al., 1995; Crossley et al., 1996; Vogel et al., 1996). Fgf8 is thought to exert its effect by activating the expression of WNT2B (at the forelimb level) and WNT8C (at the hindlimb level) in the IM of the chicken embryo (Kawakami et al.,2001). As fgf8 is expressed in the same tissue as its target genes, wnt2b and wnt8c, this is a local signaling event. Like Fgf8, Wnt2b and Wnt8c are capable of inducing ectopic limbs in the chicken, and are thus thought to mediate the limb inducing activity of Fgf8(Kawakami et al., 2001).

Another gene involved in limb induction is Fgf10. Fgf10 is expressed in the limb mesenchyme beginning at very early stages, and when ectopically expressed, can induce additional limbs in the chicken(Ohuchi et al., 1997). Furthermore, Fgf10 mutant mice fail to form limbs(Min et al., 1998; Sekine et al., 1999). These mutant embryos do not show activation of markers expressed in the limb bud ectoderm, suggesting that Fgf10 relays limb induction from the mesenchyme to the ectoderm. Fgf10 belongs in a subclass of the Fgf family with highest affinity for the Fgf receptor 2 isoform b, Fgfr2b(Ornitz et al., 1996), which is expressed in epithelial cells(Orr-Urteger et al., 1993)(reviewed by Xu et al., 1999). Fgfr2b mutant mice share many phenotypes with Fgf10 mutants(DeMoerlooze et al., 2000; Ohuchi et al., 2000), further supporting a model in which mesenchymally expressed Fgf10 activates Fgfr2b in the overlying ectoderm (reviewed by Xu et al., 1999). Fgf2, Fgf4 and Fgf8, however, have highest affinity for Fgfr2c (Ornitz et al.,1996), which is mesenchymally expressed(Orr-Urteger et al., 1993). This scenario suggests that the limb-inducing activity of Fgf8 and similar Fgfs is mediated by Fgf10, which relays the inductive event to the ectoderm. Consistent with this proposal, Fgf10 is able to induce ectodermal limb markers in the chicken flank even in the absence of mesenchyme, while Fgf2 and Fgf4 are not (Yonei-Tamura et al.,1999). The induction of Fgf4 and Fgf8 in the ectoderm by Fgf10 is not direct, and appears to be mediated in the chicken by WNT3A, which is activated in the ectoderm in response to Fgf10(Kengaku et al., 1998; Kawakami et al., 2001).

Once the limb bud has formed, Fgf4 and Fgf8 are expressed in the apical ectodermal ridge (AER), a signaling center that directs outgrowth of the limb bud, and these Fgfs have been shown to mediate the activity of the AER in the chicken and the mouse(Laufer et al., 1994; Niswander et al., 1994; Sun et al., 2002). Fgf10 continues to be expressed in the mesenchymal cells underneath the AER, and forms a feedback loop of mutual dependence with the ectodermally expressed Fgfs (reviewed by Xu et al., 1999). Fgf10 thus also directs ectodermal expression of Fgf4 and Fgf8 during the outgrowth phase of the limb.

Tbx5 encodes a T-box transcription factor that is expressed in the forelimb mesenchyme at very early stages, and has been shown to participate in the specification of limb identity in the chicken(Rodriguez-Esteban et al.,1999; Takeuchi et al.,1999). Recent results have indicated that Tbx5 is also involved in limb bud initiation. Targeted knockdown of tbx5, or mutagenesis of the tbx5 locus, leads to zebrafish embryos that lack pectoral fin buds (Ahn et al.,2002; Garrity et al.,2002; Ng et al.,2002). These mutants fail to activate expression of ectodermal fin bud markers, which correlates with the absence of fgf10 in the mesenchyme. Furthermore, targeted knockdown of wnt2b causes loss of tbx5 expression in the zebrafish pectoral fin primordium, suggesting that Tbx5 acts downstream of Wnt2b to induce fgf10 during limb initiation (Ng et al.,2002).

The zebrafish has recently become established as a model system to study the development of vertebrate paired appendages, and a number of zebrafish mutants affecting the development of paired fins have been isolated in a large-scale genetic screen (van Eeden et al., 1996). We report the molecular and phenotypic analysis of one of these mutants, named ikarus (ika; znfn1a1— Zebrafish Information Network). We show that ika encodes fgf24, which is a new member of the Fgf8/17/18 subfamily of Fgf ligands, and has highest sequence similarity to Fgf18. In the absence of fgf24 activity, we observe activation of early mesenchymal fin bud markers, such as tbx5, but the absence of all genes expressed in the fin bud ectoderm. We show that fgf24 acts downstream of tbx5to activate fgf10 expression. These results identify an additional layer controlled by Fgf signaling in the genetic hierarchy initiating limb development.

The ika^tm127c allele has been previously described(van Eeden et al., 1996). We isolated a new ika allele, ika^hx118, in a screen for ENU-induced mutations, which has been described in Habeck et al.(Habeck et al., 2002). As ika^hx118 is the stronger of the two alleles, all experiments presented in this study have been performed with this allele.

We mapped the ika mutation to linkage group 14 by using a standard panel of SSLP markers (Knapik et al.,1998). For fine mapping single ika mutant embryos (938 in total) were tested against individual SSLP markers in the crucial interval. The cloning and physical mapping of fgf24 are described elsewhere(B.W.D., D. W. Stock and C. B. Kimmel, unpublished). For sequencing of the fgf24 gene, total RNA was extracted from wild-type or ikaembryos at 36 hpf. RT-PCR was performed using the Superscript kit (Invitrogen)with the following primers: fgf24up (5′-TCCGGGGTTTTGTTTGTGAG-3′)and fgf24down (5′-TCTTTTCGGTAGCCATTGTTTATT-3′). PCR products from four independent PCR reactions on two different RNA samples were sequenced on both strands and analyzed using the MacVector software.

Morpholinos were purchased from GeneTools LLC. The following morpholinos were used: anti-tbx5 oligonucleotide for the coding sequence, as described by Ahn et al. (Ahn et al.,2002). Anti-wnt2b oligonucleotide, as described by Ng et al. (Ng et al., 2002). For fgf24, we used an oligonucleotide targeted against the translation start site with the following sequence: 5′GACGGCAGAACAGACATCTTGGTCA-3′. As a control we used the standard control oligonucleotide available from GeneTools. All oligonucleotides were solubilized in 1×Danieau's solution and injected into one-cell stage zebrafish embryos at concentrations ranging from 5-10 ng/embryo.

Donor embryos were injected with 2.5% rhodamine-dextran, and cells transplanted into hosts at 30-70% epiboly. To target wild-type cells to the fin mesenchyme, transplantation was carried out as described previously(Ahn et al., 2002). To target the ectoderm, we transplanted cells to a region opposite the shield in early gastrula embryos.

Whole-mount in situ hybridization was performed as previously described(Kishimoto et al., 1997),using the following probes: tbx5(Begemann and Ingham, 2000); msxc (Akimenko et al.,1995); dlx2 (Akimenko et al., 1994); pea3 and erm1(Roehl and Nusslein-Volhard,2001); fgf8 (Reifers et al., 1998); fgf10(Ng et al., 2002); shh (Krauss et al.,1993); bmp2(Kishimoto et al., 1997); and hand2 (Yelon et al.,2000). Alcian Blue staining was performed described previously(Grandel and Schulte-Merker,1998). Histological sections were obtained by staining cryosections with Methylene Blue (Humphrey and Pittman, 1974).

Of all the fin mutants from the Tuebingen I screen(van Eeden et al., 1996), ika shows the most severe reduction of pectoral fins. Larvae at 3 days post fertilization (dpf) have no detectable pectoral fins(Fig. 1A,B), and staining of the cartilage elements at this stage reveals that ika mutants lack all the skeletal elements of the pectoral fin, including the shoulder girdle(Fig. 1C,D). Examination of ika mutants at earlier stages reveals the absence of morphologically detectable pectoral fin buds at any time(Fig. 1E,F; data not shown). We also fail to detect any morphological signs of an AER in ika mutants(data not shown). Interestingly, even the strongest ika allele, ika^hx118 (which is likely to be a null mutant, see below),is homozygous viable, giving rise to adults that lack pectoral fins, but that have normal pelvic fins, and have no other obvious defects (data not shown). We used the ika^hx118 allele for all the experiments described in this study.

As the ika phenotype suggests it plays an important role during pectoral fin development, we were interested in identifying the molecular nature of ika. To this end, we mapped ika using SSLP markers(Knapik et al., 1998; Liao and Zon, 1999), and found it to be located on linkage group 14, between the markers z5435 and z4203(Fig. 2A,B). We found z5435 to be very tightly linked to the ika locus (1 recombinant out of 938 embryos tested) (Fig. 2A). Comparison of this interval of the genetic map with the radiation hybrid map(Geisler et al., 1999)revealed that a cDNA encoding fgf24 maps to a very similar location(Fig. 2B) (B.W.D., D. W. Stock and C. B. Kimmel, unpublished). Fgf24 is a new member of the Fgf8/17/18 subfamily of Fgf ligands and has highest sequence similarity to Fgf18 (B.W.D.,D. W. Stock and C. B. Kimmel, unpublished).

In order to determine whether ika encodes fgf24, we sequenced the fgf24 gene of the two ika alleles. In ika^hx118, we found the G at position 256 of the wild-type coding region changed to a T, causing the conversion of a glutamate codon to a stop codon at amino acid 86 (Fig. 2C and data not shown). In ika^tm127c, we found the G at position 271 of the wild-type coding region changed to a T, causing the conversion of a cysteine codon to a phenylalanine codon at position 124(Fig. 2C; data not shown).

The stop codon in ika^hx118 leaves a truncated protein lacking more than half of its C terminus(Fig. 2D), thus removing most of the core region, which has been shown to interact with the receptor in the case of other Fgf proteins (reviewed by Ornitz and Itoh, 2001). ika^hx118 is thus likely to be a null allele, which is supported by the full phenotypic penetrance observed with this allele. The cysteine converted in ika^tm127c to phenylalanine is one of six amino acids that are conserved in the core region of all Fgf proteins(reviewed by Ornitz and Itoh,2001), suggesting that it is crucial for normal function. Nevertheless, this allele appears to retain some activity, as it shows variable phenotypic penetrance and expressivity(van Eeden et al., 1996).

To further test the possibility that ika encodes fgf24,we designed an antisense morpholino oligonucleotide to block translation of fgf24 (Nasevicius and Ekker,2000). Injection of morpholinos targeted against fgf24generated larvae that specifically lack pectoral fins (n=72), but otherwise appear normal, thus phenocopying the ika mutation(Fig. 2D) (B.W.D., D. W. Stock and C. B. Kimmel, unpublished). We injected morpholinos within the range of 5-10 ng per embryo, and within this range, all injected embryos displayed complete absence of pectoral fins. The injection of a control morpholino had no effect (data not shown), and yielded individuals identical to the wild-type larva shown in Fig. 1A.

From these data, we conclude that ika encodes fgf24 and hereafter refer to the mutant by its molecular name. The GenBank Accession number for fgf24 is AY204859.

To better define the role played by fgf24 during pectoral fin development, we localized the fgf24 transcript using in situ hybridization. We first detect fgf24 in the region of the pectoral fin primordia at 18 hpf (Fig. 3A). Apart from tbx5, which is first detected in this region at 17 hpf (Begemann and Ingham,2000; Ruvinsky et al.,2000), fgf24 is the earliest marker for the pectoral fin primordia. At this stage, the transcript is detectable only in the mesenchyme(Fig. 3A). Expression is still present in the fin bud mesenchyme at 24 hpf(Fig. 3B), but becomes downregulated in these cells between 28 and 30 hpf(Fig. 3C). At the same time, fgf24 is activated in the overlying AER(Fig. 3C).

As both the fgf24^-/- phenotype and the fgf24expression pattern suggest that it plays a role during early stages of pectoral fin development, we examined the expression of several genes expressed at these stages in fgf24 mutants.

tbx5 is the earliest markers for pectoral fin development, and is expressed in the mesenchymal compartment of the fin buds(Fig. 4A)(Begemann and Ingham, 2000; Ruvinsky et al., 2000). We find that it is also expressed in these cells at 24 hpf in fgf24mutants, although the expression appears weaker than in wild-type embryos(Fig. 4B).

msxc is expressed dynamically during pectoral fin development. At 24 hpf, it is expressed throughout the mesenchyme of the pectoral fin bud primordia (Fig. 4C and data not shown). The expression of msxc in fgf24 mutants is indistinguishable from that in wild-type embryos at this stage(Fig. 4D). Between 28 and 32 hours, expression becomes weaker and then disappears completely in fgf24 mutants (data not shown).

dlx2 is an early marker for the AER in zebrafish. We find that dlx2 is activated in the ectoderm of wild-type pectoral fin buds already at 20-22 hpf (Fig. 4E;data not shown), which precedes formation of the AER(Grandel and Schulte-Merker,1998). At early stages, dlx2 appears to be expressed throughout the entire fin bud ectoderm, and later becomes restricted to the AER (data not shown). We find that dlx2 expression is not detectable in the pectoral fin buds of fgf24 mutants at any time(Fig. 4F; data not shown).

shh is first expressed in pectoral fins at 24 hpf. We detect no shh expression in the pectoral fins of fgf24 mutants at 24 hpf (Fig. 4G,H), or at later stages (see Fig. 8A,B).

These data suggest that at least some mesenchymally expressed genes are activated in fgf24 mutant pectoral fin buds, but not ectodermally expressed genes. We further addressed this possibility by examining marker gene expression at 30 hpf, when the mesenchymal and ectodermal compartments of the pectoral fin buds are easier to distinguish.

erm1 and pea3 encode transcription factors that have been shown to be dynamically expressed in many tissues during zebrafish development(Roehl and Nuesslein-Volhard,2001). We find that both genes are expressed throughout the pectoral fin bud mesenchyme, and that pea3 is additionally expressed in the ectoderm (Fig. 5A,C). Both erm1 and pea3 are expressed in the mesenchyme of fgf24 mutants, but pea3 fails to be activated in the ectoderm (Fig. 5B,D). Like msxc, erm1 and pea3 expression starts to fade around 30 hpf in fgf24 mutants, and is no longer detectable at later stages (data not shown).

None of the ectodermal markers we tested are activated in the pectoral fin buds of fgf24 mutants. These include pea3, dlx2, fgf4, and fgf8 (Fig. 5A,B,E-J), msxc and bmp2 (Fig. 7C-F), and fgf24 (data not shown).

Taken together, these results suggest that activation of genes expressed in the ectoderm of the pectoral fin buds fails to occur in the absence of fgf24 activity. The expression of some genes expressed in the fin bud mesenchyme, however, appears to be independent of fgf24 activity.

The fgf24^-/- pectoral fin phenotype bears strong resemblance to that of zebrafish embryos lacking wnt2b or tbx5 activity (Ahn et al.,2002; Garrity et al.,2002; Ng et al.,2002). To investigate the relationship of fgf24 with these two genes, we examined the expression of fgf24 in the absence of wnt2b or tbx5 activity, by using a morpholino knockdown approach (Nasevicius and Ekker, 2002). We fail to detect fgf24expression at 20 hpf or 24 hpf in the pectoral fin bud primordia in the absence of either wnt2b or tbx5 activity(Fig. 6A-C; data not shown). Injection of control morpholinos yielded embryos showing wild-type fgf24 expression identical to that shown in Fig. 6A. Together with the observation that tbx5 expression is initiated in fgf24mutants (Fig. 4B), these results indicate that fgf24 acts downstream of tbx5 in the genetic cascade initiating pectoral fin bud development. As tbx5fails to be activated in the absence of wnt2b activity(Ng et al., 2002), and as fgf24 activation depends on tbx5(Fig. 6B), it is not surprising that fgf24 also depends on wnt2b activity.

Like fgf24, the fgf10 gene depends on tbx5 for its activation. To clarify the regulatory relationship between fgf24and fgf10, we examined the expression of fgf10 in fgf24 mutants. We fail to detect pectoral fin expression of fgf10 at all stages in the absence of fgf24 activity(Fig. 6D,E; data not shown).

Taken together, these results suggest that fgf24 acts downstream of tbx5 to activate fgf10 expression during limb bud initiation (Fig. 6F).

As fgf24 is expressed both in the mesenchyme and in the AER, we wished to determine in which cells fgf24 activity is required for fgf10 activation. We therefore transplanted wild-type cells into fgf24 mutant embryos. We observed rescue of fin bud outgrowth when wild-type cells were located in the fin bud mesenchyme(Table 1, n=3). These fin buds also showed an AER at 36 hpf, and fgf10 expression. In one case, we did not observe rescue when wild-type cells were located in the lateral plate mesoderm at the level of the pectoral fin bud (data not shown). This may be due to the fact that not enough wild-type cells were present in this case. We never observed rescue of the fgf24 phenotype when wild-type cells were located in the ectoderm at the pectoral fin level(Table 1, n=6).

These results suggest that fgf24 activity is required in the mesenchyme, and not the ectoderm, for fin bud initiation and the activation of fgf10.

As tbx5 is required for the movement of mesenchymal cells in the lateral plate mesoderm to the pectoral limb bud(Ahn et al., 2002), and as fgf24 is activated downstream of tbx5, we asked whether fgf24 might play a role in mediating this effect of tbx5. To this end, we compared the distribution of tbx5-expressing cells in the lateral plate mesoderm in wild-type and fgf24 mutant embryos between 18 hpf and 32 hpf. At 18 hpf, tbx5 expression is indistinguishable in wild-type and fgf24 mutants(Fig. 7A,B). However, at 27 hpf, the tbx5-expressing cells have congregated towards the pectoral fin bud in wild-type embryos, but remain dispersed in fgf24 mutants(Fig. 7C,D). This phenotype is even more striking at 32 hpf (Fig. 7E,F).

This observation indicates that fgf24 activity is required for the correct movement of tbx5-expressing cells in the lateral plate mesoderm to the pectoral fin bud.

The activation of shh expression in the posterior mesenchyme of chick limb buds depends on Fgfs secreted from the AER(Laufer et al., 1994; Niswander et al., 1994; Sun et al., 2002). Because fgf24 mutants do not show any sign of AER formation, we examined whether shh is activated in the absence of fgf24 activity. We find that shh expression is not detectable in fgf24mutants at any stage, consistent with the absence of AER in these mutants(Fig. 4G,H; Fig. 8A,B). msxc is initially expressed throughout the mesenchyme of the pectoral fin buds, but expression becomes restricted to the anterior by 28-30 hpf in wild-type embryos (Fig. 4A; Fig. 8C). In fgf24mutant pectoral fin buds, this repression in the posterior mesenchyme fails to occur (Fig. 8D). This is probably due to the absence of Shh, because the same phenotype is observed in shh mutant pectoral fin buds(Neumann et al., 1999). hoxd13 and hoxa13 also fail to be activated in fgf24 mutant pectoral fin buds (data not shown). Again, this phenotype is similar to what is observed in the absence of shhactivity (Neumann et al.,1999).

bmp2 is expressed both in the AER and in the posterior mesenchyme(Fig. 8E). We find that bmp2 is activated in the posterior mesenchyme in fgf24mutants, but not in the ectoderm (Fig. 8F). This correlates with the observation that activation of bmp2 in the posterior mesenchyme is independent of shhactivity (Neumann et al.,1999). The shh-independent anterior/posterior (AP)polarity of pectoral fin buds has been shown to be directed by the hand2 gene, which encodes a bHLH transcription factor(Yelon et al., 2000). Consistent with this observation, we find normal posterior activation of hand2 in fgf24 mutants(Fig. 8G,H), although hand2 expression gradually fades after 30 hpf (data not shown).

These data suggest that the early aspects of AP polarity are established in the fgf24 mutant pectoral fin primordium. Those aspects of AP polarity that are lost in fgf24 mutants are ones which have been shown to depend on shh activity(Neumann et al., 1999), which correlates well with the failure to activate shh expression in fgf24 mutant fin buds, while the shh-independent AP polarity is unaffected by the loss of fgf24.

The initiation of vertebrate limb development involves the transfer of positional information from the intermediate mesoderm (IM) to the lateral plate mesoderm (LPM), and from there to the ectoderm. This process is based on a cascade of inductive events that depends on secreted signaling proteins sequentially activated in all three of these tissues. Fgf8 is thought to initiate this cascade at the level of the chicken forelimb by activating Wnt2b in the IM, which then signals to the LPM to activate Tbx5 expression. Tbx5 activates Fgf10 in the LPM,which in turn signals to the overlying ectoderm to activate Wnt3a. Wnt3a completes the cascade by inducing Fgf4 and Fgf8 in the chicken AER.

In this study, we have shown that the zebrafish fgf24 gene, which is disrupted by the ikarus mutation, acts downstream of tbx5to activate fgf10 in the LPM, thus identifying another Fgf acting early in the cascade of limb induction. Consistent with this model, fgf24 is expressed very early during pectoral fin development, and is activated in the LPM at a similar stage as tbx5 (this study)(Begemann and Ingham, 2000; Ruvinsky et al., 2000). Together with the observation that Tbx5 is activated within 1 hour of implanting an Fgf-soaked bead into the flank of a chick embryo(Isaac et al., 2000), these results suggest that Tbx5 and Fgf24 act early in the limb-inducing cascade. Fgf10, however, is not activated until 17 hours after Fgf application(Ohuchi et al., 1997),consistent with the proposal that its induction requires an additional signaling event. A more direct effect of mouse TBX5 on Fgf10transcription has been recently proposed, based on the finding that there is at least one potential TBX5-binding site upstream of the mouse Fgf10promoter, and that the Fgf10 promoter can be upregulated by co-expressing Tbx5 in cultured cells(Agarwal et al., 2003). However, this direct model does not explain why the activation of Fgf10 by Fgf bead application takes so much longer than the activation of Tbx5, and it remains to be seen what the role of the TBX5-binding sites are in vivo. One possibility is that the Fgf10promoter integrates several different signals, and that direct binding by TBX5 is necessary, but not sufficient for activation in vivo. Our data indicate that one of the additional requirements for Fgf10 activation in the zebrafish is the exposure of these cells to Fgf24.

The data presented here and elsewhere suggest that zebrafish Wnt2b, Tbx5,Fgf24 and Fgf10 act sequentially in a linear pathway in which Wnt2b induces Tbx5 expression, which then induces Fgf24, which in turn induces Fgf10(Fig. 6F). In addition to its role in this linear cascade, zebrafish Tbx5 has been shown to be required for the correct migration of lateral plate mesenchymal cells to the pectoral fin primordium (Ahn et al., 2002). Here we have shown that Fgf24 is also required for this process, as tbx5-expressing cells fail to congregate to the pectoral fin primordium in the absence of fgf24 activity. The dispersal of tbx5-expressing cells is not as dramatic in the absence of fgf24 activity as it is in the absence of tbx5 activity,suggesting that Fgf24 does not mediate all effects of Tbx5 on this migratory event. This is consistent with the observation that tbx5 activity is cell-autonomously required for the correct movement of cells to the fin bud(Ahn et al., 2002).

The pectoral fin phenotype of fgf24 mutants is similar to that of zebrafish raldh2 mutants (Begemann et al., 2001; Grandel et al.,2002). As raldh2 has been shown to act upstream of tbx5 activation in the fin bud(Begemann et al., 2001), it is also likely to act upstream of fgf24 activation.

The Fgf24 protein is a new member of the Fgf8/17/18 subfamily of Fgf ligands (B.W.D., D. W. Stock and C. B. Kimmel, unpublished). Furthermore,Fgf24 also shows functional overlap with Fgf8, as both genes are expressed at early stages in the posterior mesoderm of the embryo, and fgf8/fgf24 double mutants display developmental defects in this region that are not observed in either fgf8 or fgf24single mutants, thus indicating that these two Fgfs have very similar activities (B.W.D., D. W. Stock and C. B. Kimmel, unpublished). Hence, the analysis of the fgf24 mutant phenotype provides the first loss-of-function data demonstrating a role for an Fgf8-like gene in limb initiation, and complements the gain-of-function experiments which show that ectopic application of Fgf2, Fgf4 and Fgf8 can trigger the development of additional limbs (Cohn et al.,1995; Ohuchi et al.,1995; Crossley et al.,1996; Vogel et al.,1996; Yonei-Tamura et al.,1999).

Members of the Fgf8 subclass of Fgfs appear to be expressed in all three tissues involved in limb initiation, and seem to have important functions at three distinct steps in early limb development. First, Fgf8, which is expressed in the chicken IM, initiates WNT2B expression in the IM. Secondly, Fgf24, which is expressed in the zebrafish LPM, activates fgf10 expression in the LPM, and thirdly, Fgf4 and Fgf8, which are expressed in the chicken and mouse AER, direct outgrowth of the limb bud and maintain Fgf10 expression in the mesenchyme. Interestingly, our data suggest that Fgf24 signals in an autocrine manner to activate fgf10in the LPM, as both genes appear to be co-expressed in the same region,although the activation of fgf24 precedes that of fgf10. However, on the basis of these data, we cannot distinguish whether Fgf24 produced by an individual cell signals to the same cell, or whether there is signaling between cells in the same tissue. Our data also suggest that mesenchymally expressed Fgf24 activates mesenchymal shh expression at 24 hpf. This situation is clearly different from the chick and the mouse,where Fgf8 secreted from the AER has been shown to activate shh in the mesenchyme (Laufer et al., 1994; Nieswander et al., 1994; Sun et al., 2002). Consistent with this observation, the zebrafish fgf8 gene is activated in the AER at a much later stage than in tetrapods (12 hours after the activation of shh in the mesenchyme), and zebrafish fgf8 mutants have no effect on fin development (Reifers et al.,1998), although this may be a hypomorphic mutation.

It is also noteworthy in this context that fgf24 is activated in the AER after limb bud initiation has taken place, and at the same time is downregulated in the mesenchyme (Fig. 3C). This observation raises the possibility that Fgf24, together with the other Fgfs expressed in the AER, continues to direct fgf10expression in the mesenchyme during limb outgrowth, this time by a paracrine mechanism. This scenario is not supported by our transplantation results,however, because the fgf24 mutant phenotype is rescued only by wild-type cells located in the fin mesenchyme, but not in the ectoderm, and rescue leads to restoration of fgf10 expression even at 36 hpf, when fgf24 is expressed in the AER, and not the mesenchyme. The possibility remains that at this stage, Fgf24 functions redundantly with other Fgfs expressed in the AER, such as Fgf4 and Fgf8.

It is interesting to note that some genes are specifically required for the development of either forelimbs or hindlimbs, while other genes function in both types of limbs. Fgf24, for example, is only required for the development of pectoral fins, but not pelvic fins, and shares this characteristic with chicken Tbx5 and Wnt2b (Rodriguez-Esteban et al., 1999; Takeuchi et al.,1999; Kawakami et al.,2001). Fgf10, however, appears to play an equivalent role in both fore- and hindlimbs in mice and chicken(Min et al., 1998; Sekine et al., 1999). It is likely that chicken Tbx4 and Wnt8c, or related genes, play a role in hindlimb development that is similar to the role of Tbx5 and Wnt2b, respectively, in forelimb development (Rodriguez-Esteban et al., 1999; Takeuchi et al.,1999; Kawakami et al.,2001). Hence, our data suggest that a zebrafish Fgf closely related to Fgf24 fulfills a similar function during pelvic fin development.

There appears to be strong evolutionary conservation of the developmental mechanism of limb bud initiation. For example, inactivation of the Tbx5 gene leads to similar reductions of the forelimbs in chicken,zebrafish, human and mouse (Newbury-Ecob et al., 1996; Ahn et al.,2002; Garrity et al.,2002; Ng et al.,2002; Agarwal et al.,2003). In addition, the role of Wnt2b in limb initiation appears highly conserved in chicken and zebrafish(Kawakami et al., 2001; Ng et al., 2002). By contrast,however, there appears to be no role for Wnt2b in initiating mouse limb development (Ng et al.,2002), nor is there a role for Wnt3a in inducing the mouse AER (Barrow et al.,2003). Nevertheless, the phenotype of the Lef1/Tcf1 double mutant mouse is consistent with WNT signaling being necessary for AER formation in the mouse(Galceran et al., 1999). These results suggest that different Wnt proteins, comparable in activity to the chicken Wnt2b and Wnt3a proteins, fulfill their respective roles in the mouse. Consistent with this proposal, it has recently been shown that the mouse Wnt3 gene fulfills the function of chicken WNT3A during limb development (Barrow et al.,2003).

This idea is similar to the one that different genes of similar activity fulfill the same function during fore- and hindlimb development, and suggests that the existence of gene families with similar activities has allowed the functional replacement of specific genes during evolution. By the same reasoning, our data suggest that an Fgf comparable in activity with Fgf24 occupies a similar position in the genetic cascade that initiates limb development in other vertebrate species. However, as both the mouse and human genomes have been sequenced, it is clear that they contain no ortholog of the zebrafish fgf24 gene. Although it is possible that zebrafish Fgf24 fulfills the role attributed to tetrapod Fgf8 in limb initiation,this is unlikely, because Fgf8 is never expressed in limb mesenchyme in any vertebrate examined to date. An alternative possibility is that another tetrapod Fgf family member with similar activity to Fgf8 is transiently expressed in the limb bud mesenchyme, and that this expression has so far gone undetected. Future experiments will hopefully resolve this issue.

We thank Pascal Haffter and Robert Geisler for allowing us to use their mapping facility, Christiane Nüsslein-Volhard for her support, Marcus Dekens for sharing expertise, Gillian Brunt for technical assistance, Henry Roehl and Deborah Yelon for sharing cDNAs, and Sandrine Pizette and Stephen Cohen for critical comments on the manuscript