Zebrafish fgf24 functions with fgf8 to promote posterior mesodermal development

In vertebrate embryos, the posterior body and tail develop in an anterior to posterior progression by the coordinated growth and morphogenesis of precursor cells located in the tail bud(Kanki and Ho, 1997; Davis and Kirschner, 2000). Studies in several organisms have established that the Fgf signaling pathway plays an essential role during the development of the posterior body, perhaps by maintaining a population of posterior precursors cells during embryogenesis. Inhibiting Fgf signaling in Xenopus or zebrafish embryos by overexpressing a dominant-negative Fgf receptor (dnFgfr) blocks the formation of posterior body structures, including all posterior mesoderm(Amaya et al., 1991; Amaya et al., 1993; Griffin et al., 1995). Similarly, mouse embryos mutant for the Fgf receptor 1 (Fgfr1), one of four known vertebrate Fgf receptors, produce limited amounts of posterior mesoderm (Yamaguchi et al.,1994; Deng et al.,1994). Fgfr1 is cell autonomously required for posterior mesodermal development, as Fgfr1 mutant cells transplanted into wild-type host embryos do not contribute to this tissue, and instead adopt neuronal fates (Ciruna et al.,1997; Ciruna and Rossant,2001). Thus, the Fgf signaling pathway appears to play a conserved role during development of posterior mesoderm in vertebrates.

To date, 23 Fgf ligands (Fgf1-23) have been described in tetrapods(reviewed by Ornitz and Itoh,2001) and several of these ligands are known to be expressed in early mesodermal progenitors in mice, including Fgf3(Wilkinson et al., 1988), Fgf4 (Niswander and Martin,1992; Drucker and Goldfarb,1993), Fgf5 (Haub and Goldfarb, 1991; Hébert et al., 1991) and Fgf8(Heikinheimo et al., 1994; Ohuchi et al., 1994; Crossley and Martin, 1995). Mutational analyses in mice, however, suggest that not all of these ligands are required for the development of mesoderm. For example, embryos mutant for Fgf3 and Fgf5 have only slight (Fgf3)(Mansour et al., 1993) or no(Fgf5) (Hébert et al.,1994) defects in posterior development. Conversely, Fgf8mutant embryos do not form posterior mesoderm, indicating that Fgf8activity can account for the majority of Fgf signaling required for posterior development in mice. A role for Fgf4 in posterior mesodermal development in mice has yet to be established, as Fgf4 mutants die prior to mesoderm formation (Feldman et al., 1995).

In addition to Fgf signaling, T-box genes, which function as transcriptional regulators, are also required for formation of the posterior body during vertebrate embryogenesis. The founding member of the T-box gene family, mouse T or Brachyury is expressed early in mesodermal precursors and then in the developing notochord(Herrmann et al., 1990). T is required for the development of these tissues, as Tmutant embryos fail to form a notochord and lack posterior body structures(reviewed by Smith, 1999; Papaioannou, 2001). The role of T in mesodermal development appears to be evolutionarily conserved in vertebrates, as T orthologs in several organisms have been shown to have similar expression patterns and functions. For example, Torthologs in Xenopus and zebrafish, called Xbra and no tail (ntl), respectively are expressed in mesodermal precursors and in the developing notochord (Smith, et al., 1991; Schulte-Merker et al., 1992), and are required(Halpern et al., 1993; Conlon et al., 1996) and sufficient (Cunliff and Smith, 1992; O'Reilly et al., 1995) for notochord and posterior mesodermal development.

The T-box gene VegT/spt has also been implicated in mesodermal specification in vertebrate embryos. VegT in Xenopus is expressed in mesodermal precursors and in developing posterior paraxial mesoderm, and is also expressed maternally(Horb and Thomsen, 1997; Lustig et al., 1996; Stennard et al., 1996; Zhang and King, 1996). Inhibition of maternal VegT function results in embryos that fail to form both mesoderm and endoderm, showing that VegT has an early role in germ layer formation (Zang et al., 1998). The function of zygotically expressed VegT has not been determined. In zebrafish, spadetail (spt; tbx16 - Zebrafish Information Network) is an ortholog of VegT and is similarly expressed in mesodermal precursors and in developing paraxial mesoderm. In contrast to VegT, however, spt is not expressed maternally(Griffin et al., 1998). spt mutant embryos lack paraxial mesoderm in the trunk, but not in the tail, and form a relatively normal notochord(Kimmel et al., 1989; Amacher et al., 2002). Thus, spt mutants have a phenotype that is nearly reciprocal to that of ntl mutants. Although both spt and ntl mutants form lateral and ventral mesodermal cell types, spt;ntl double mutant embryos fail to form all posterior mesoderm(Amacher et al., 2002). These results suggest that spt and ntl have distinct roles in promoting the development of specific mesodermal subtypes, as well as a presumed earlier, and redundant role in the specification of all posterior mesodermal precursors.

A link between Fgf signaling and T-box gene function in posterior mesodermal development was revealed when it was shown that T-box gene expression in mesodermal precursors is dependent on Fgf signaling. In Xenopus and zebrafish, expression of Xbra/ntl is inhibited when Fgf signaling is blocked(Amaya et al., 1991; Isaacs et al., 1994; Schulte-Merker and Smith,1995; Griffin et al.,1995) and ectopic activation of the Fgf signaling pathway leads to ectopic Xbra/ntl expression(Isaacs et al., 1994; Schulte-Merker and Smith,1995; Griffin et al.,1995). These and other results have led to the model that Fgf signaling and T-box genes form an auto-regulatory feedback loop during early mesodermal development, where the function of one component is necessary for the continued expression of the other. These interactions are thought to promote posterior development by maintaining and regulating the growth and morphogenesis of mesodermal precursors in the posterior region of the embryo(reviewed by Isaacs,1997).

In zebrafish, inhibiting Fgf signaling leads to a phenotype that is strikingly similar to that of spt;ntl double mutant embryos(Griffin et al., 1995; Amacher et al., 2002). Because expression of both spt and ntl in mesodermal precursors is Fgf dependent (Griffin et al.,1995; Griffin et al.,1998), it is possible to explain the mesodermal defects associated with blocking Fgf signaling as a loss of spt and ntlfunction. Although spt and ntl are key regulators of posterior development in zebrafish, little is known about which Fgf signaling components are required to maintain their expression.

The zebrafish fgf8 gene is expressed in mesodermal precursors and is therefore a candidate Fgf ligand for regulating posterior development. A mutation in fgf8 (or acerebellar)(Reifers et al., 1998), has been identified, but unlike embryos injected with a dnFgfr(Griffin et al., 1995), fgf8 mutants (Reifers et al.,1998), or embryos in which fgf8 function has been inhibited with morpholino oligonucleotides(Araki and Brand, 2001; Draper et al., 2001) have relatively mild defects in posterior development. A hypothesis that we explore here is that additional Fgf ligands function together with Fgf8 during development of the posterior body in zebrafish.

We have identified and characterized a second Fgf ligand-encoding gene in zebrafish that is expressed in mesodermal precursors. This ligand is a new,but distinct, member of the fgf8/17/18 subclass of Fgf ligands, for which there is no ortholog among the 23 known Fgfs in tetrapods. We therefore designate this gene fgf24. We show that fgf24 is expressed in a domain that overlaps extensively with that of fgf8, ntl and spt in mesodermal precursors during gastrulation, and that fgf8 and fgf24 are together required for the formation of most posterior mesoderm. Furthermore, we present both gene expression and genetic data showing that interactions between the Fgf signaling pathway and the ntl and spt T-box genes are essential for posterior mesoderm development in zebrafish. Last, we show that fgf24 is also required for initiation of the pectoral fin bud, a role that appears similar to that of Fgf10 in mice (reviewed by Martin, 1998).

Degenerate primers for RT-PCR of fgf8-related genes(5′-GCCGGGATCCACNAGYGGNAARCAYGTNCA-3′ and 5′-GCCGGAATTCGGNARNCKYTTCATRAARTG-3′, where the underlined sequences represent restriction sites added for cloning) were designed from an alignment of tetrapod Fgf8 sequences. PCR was carried out on cDNA produced from mRNA isolated from 5-day-old larvae. PCR products were cloned into pBluescript II SK⁺ (Stratagene, La Jolla,CA) and 34 independent clones were sequenced. Of these, seven were identified as fgf8, two as fgf17, seventeen as fgf18 and eight as fgf24, based on phylogenetic analyses of the 105-106 encoded amino acids. An fgf24 cDNA was isolated by using one of the cloned RT-PCR fragments as a probe to screen a gastrula-stage cDNA library (a gift from T. Lepage and D. Kimelman). Additional fgf18 cDNA sequence was isolated by 3′ and 5′ RACE using the First Choice RLM-RACE kit (Ambion,Austin, TX) following the manufacturer's instructions. The fgf24 gene structure was determined by partially sequencing a PAC clone containing the fgf24 gene. This clone was identified by screening a PAC library(Amemiya and Zon, 1999) by PCR with the gene specific primers 5′-CAGGAGTGCGTCTTCGTGGAG-3′ and 5′-GTGCCCTTCGTGTCCTTTTCG-3′ (231 bp fragment). Temporal expression profiles were determined by RT-PCR, as previously described(Draper et al., 2001) using the following primers (5′ primer/3′ primer): fgf8,CACATTTGGGAGTCGAGTTCG/GTGCTCTGCGATTTGGTGTCC (288 bp fragment); fgf24,GCAAGAAGATTAACGCCAATGG/TTTAGGTCGACCCTTTCG (272 bp fragment); fgf18,GACGACGGAGATAAATATGCC/CGTACCATCCTGTGTAGCGC (221 bp fragment); and odc, ACACTATGACGGCTTGCACCG/CCCACTGACTGCACGATCTGG (309 bp fragment). GenBank Accession Numbers for the cDNA sequences are: fgf18,AY243514; fgf24, AY204859.

Phylogenetic relatedness of fgf24 and fgf18 were determined by aligning sequences with the ClustalX program, and constructing trees from the alignments using the neighbor-joining method. Prior to alignment, we used the Signal IP program(Nielsen et al., 1997) to identify the most probable cleavage site of the signal peptide that comprises the N-terminal 25-30 amino acids of the proteins. The tree was then constructed using an alignment that contained only those sequences that were predicted to be present in the mature Fgf proteins. As an outgroup, the distantly related zebrafish Fgf10 protein sequence was used.

The positions of fgf18, fgf24 and the EST fi43f07 (GenBank Accession Number AW174476; M. Clark and S. Johnson, WUZGR; http://zfish.wustl.edu)in the zebrafish genome were determined by mapping on the Goodfellow T51 radiation hybrid (RH) panel (Kwok et al.,1998) (Research Genomics) using the following primers pairs(5′ primer/3′ primer): fgf18,CCGGGACTCAAACCAGCGACC/GTCCTGCTGGTTGGGAAGCG (411 bp fragment); fgf24,same primers as those used for PAC isolation (see above); fi43f07,GTTCACCGACGGGTTTCCATTTTCA/TCCTGCATCTTTAGCCCGCGTTTAC (199 bp fragment). Following PCR, fragments were separated by electrophoresis and scored as described by Geisler et al. (Geisler et al., 1999). The RH data was converted to a map position using the Instant Mapping program(http://134.174.23.167/zonrhmapper/instantMapping.htm).

Adult zebrafish stocks and embryos were maintained at 28.5°C as described previously (Westerfield,1995). Embryos were produced by natural matings of the appropriate adult fish. Embryos were collected and sorted at early cleavage stages and maintained in embryo medium (Westerfield et al., 1995) at 28.5°C until the desired developmental stages according to Kimmel et al.(Kimmel et al., 1995). The following alleles were used for this study: fgf8/acerebellar^ti282, spt^b104 and ntl^b195. The spt^b104(Griffin et al., 1998) and ntl^b195(Schulte-Merker et al., 1994)alleles are null, while the fgf8^ti282 allele is probably a hypomorph (Draper et al.,2001). spt maps to LG8 (S. L. Amacher, unpublished), Fgf8 maps to LG13 (Woods et al.,2000) and ntl maps to LG19(Postlethwait et al.,1998).

Fish doubly heterozygous for spt^b104 and fgf8^ti282, or for fgf8^ti282 and ntl^b195 were generated by crossing single heterozygotes of the appropriate genotype to produce F₁ offspring. The genotypes of adult F₁ fish were scored by crossing to tester fish of known genotype. Homozygous double mutant embryos were then produced by crossing doubly heterozygous fish. Embryos from such crosses were sorted into phenotypic classes based on morphology at 24 hpf using a dissecting microscope. For two unlinked mutations segregating in a Mendelian fashion,four phenotypic classes should be obtained in a ratio of 9:3:3:1 (wild type:mut1^-/-:mut2^-/-:mut1^-/-:mut2^-/-). In crosses between spt^+/-;fgf8^+/-fish, the following phenotypic classes were found in the ratios of 9.07:3.08:2.91:0.94 (wild type:spt^-:fgf8^-:spt^-;fgf8^-, n=1,076, x²=0.612, P>0.80). In crosses between fgf8^+/-;ntl^+/-fish, the following phenotypic classes were obtained in the ratios of 8.83:3.13:2.94:1.10 (wild type:fgf8^-:ntl^-:fgf8^-;ntl^-, n=609, x²=0.76, P>0.80). In crosses between fgf8^+/-;ntl^+/- double heterozygotes and fgf8^+/- single heterozygous fish, the following phenotypic classes were obtained in the ratios of 6.03:0.94:1.03(wild type:fgf8^-:fgf8^-short tail, n=382, x²=0.196, P>0.90).

Riboprobes for in situ hybridization were synthesized using the MaxiScript kit according to the manufacturer's instructions (Ambion, Austin, TX). With the exception of fgf24 (this paper), the probes used have been described previously as follows: pax2.1(Krauss et al., 1991), krx20 (egr2 - Zebrafish Information Network)(Oxtoby and Jowett, 1993), myod (Weinberg et al.,1996), ntl(Schulte-Merker et al., 1992), spt (Griffin et al.,1998), fgf8(Fürthauer et al., 1997; Reifers et al., 1998) and shh (Krauss et al.,1993). The fgf24 in situ probe was transcribed from the full length cDNA. Immunohistochemical staining with anti-Ntl(Schulte-Merker et al., 1992)and anti-Spt (Amacher et al.,2002) were preformed as detailed by Amacher et al.(Amacher et al., 2002). For in situ hybridization experiments using embryos older than 24 hpf, melanogenesis was inhibited by raising embryos in embryo medium containing 0.003% PTU(1-phenyl 2-thiourea) (Westerfield,1995). For sectioning, embryos were embedded in epon and 7.5 μm sections were cut.

The splice-site targeted morpholino oligonucleotide (MO) fgf24-E3I3 was obtained from GeneTools (Corvalis, OR) and has the following sequence: 5′-AGGAGACTCCCGTACCGTACTTGCC-3′. MO injections were performed as previously described(Draper et al., 2001). RT-PCR analysis shown in Fig. 3 was performed essentially as described above, but using the fgf24specific PCR primer pair (5′ primer/3′ primer):CGGCAAACGCTGGAAACAGG/GTCTCTGTCTCCACCACAAGC (wild-type fragment 300 bp). RNase protection assays were performed using the RPA III kit (Ambion, Austin, TX) as previously described (Draper et al.,2001). A template for making antisense fgf24 probe was generated by amplifying a fragment of the fgf24 cDNA using the primer pair ATGTCTGTTCTGCCGTCAAGG/GTCTCTGTCTCCACCACAAGC, and cloning into the pCRIITOPO TA vector (Invitrogen, Carlsbad, CA).

One-month-old fish were cleared and stained for bone (with Alizarin Red)and cartilage (with Alcian Green) as described by Grandel and Schulte-Merker(Grandel and Schulte-Merker,1998).

In zebrafish, the mesodermal and endodermal germ layers form from precursor cells located at the margin of the early gastrula embryo(Kimmel et al., 1990). During gastrulation, these cells involute at the margin to form the hypoblast layer under the overlying epiblast layer, which contains ectodermal precursors(Warga and Kimmel, 1990; Warga and Nusslein-Volhard,1999). fgf8 has previously been shown to be expressed in mesendodermal precursor cells during gastrulation in zebrafish(Fürthauer et al., 1997; Reifers et al., 1998). We sought to identify additional Fgf ligands that are expressed in mesendodermal precursors during gastrulation and focused on identifying fgf8-related genes. Using degenerate oligonucleotide primers that were designed to amplify genes closely related to fgf8, we isolated four distinct cDNA fragments. In addition to fragments corresponding to fgf8 (Fürthauer et al.,1997; Reifers et al.,1998) and fgf17(Reifers et al., 2000), we identified fragments from two genes whose sequence appeared most closely related to tetrapod Fgf18(Ohbayashi et al., 1998; Hu et al., 1999).

To further characterize these two genes, we identified and sequenced cDNAs,and used their conceptually translated protein sequences to construct a phylogenetic tree (Fig. 1A,B). Sequence comparison to the 22 known human FGF ligands confirmed that these two genes are members of the FGF8/17/18 subfamily (henceforth referred to as 'Fgf8 subfamily') (reviewed by Ornitz and Itoh, 2001) and are most closely related to FGF18(not shown). Of the two zebrafish proteins, one shared 73% amino acid identity with human FGF18, while the other shared only 65% identity(Fig. 1A,B). For simplicity, we shall refer to these genes as fgf18 and fgf24 respectively. To better determine the relationship of fgf18 and fgf24 to known genes, we mapped them using the T51 radiation hybrid panel(Kwok et al., 1998) and found that they both localized to LG14, ∼18 cM apart(Fig. 1C). Human FGF18has been mapped to chromosome 5q34(Whitmore et al., 2000), and previous studies have found that LG14 contains regions with conserved synteny to human 5q31-5q35 (Woods et al.,2000). Using these data, together with the map positions of additional zebrafish orthologs of human genes known to map in the 5q31-5q35 interval, we determined that there was significant conserved synteny between regions containing zebrafish fgf18 and human FGF18. For example, the human gene encoding the F-box WD40 protein FBXWB1, and its zebrafish ortholog, referred to by its GenBank Accession Number AW174476,are closely linked to fgf18 (Fig. 1C). By contrast, we found no significant syntenic conservation between the map location of fgf24 and any region of the human genome. Because Fgf24 protein sequence is as distantly related to Fgf18 orthologs as Fgf8 orthologs are to Fgf17 (Fig. 1B), we propose that Fgf24 defines a new clade in the Fgf8/17/18 subfamily, and for which a tetrapod ortholog has not been described.

We used RT-PCR to compare the temporal expression profiles of fgf18 and fgf24 with that of fgf8. Similar to fgf8, we found that fgf18 and fgf24 transcripts could be detected in one-cell stage embryos(Fig. 1D), indicating that these genes are maternally expressed. By contrast, fgf24, but not fgf18, is also expressed throughout gastrulation (6-10 hpf; Fig. 1D) during the period of mesoderm specification and involution. For the remainder of this study, we focus only on characterizing the expression and function of fgf24. The expression and function of fgf18 will be reported elsewhere(B.W.D. and D.W.S., unpublished).

Analysis of the Fgf24 protein sequence using the SignaIP program(Nielsen et al., 1997)indicates that the C-terminal 30 amino acids encode a probable signal sequence, arguing that fgf24 is a secreted protein. We determined the intron/exon boundaries of the fgf24 gene by partial sequencing of the genomic locus and found that they are in positions that are conserved within the Fgf8 subfamily (Xu et al., 1999) (Fig. 1A).

We determined the expression pattern of fgf24 transcripts in gastrula-stage embryos by whole-mount in situ hybridization. We first detected localized fgf24 transcripts at the beginning of epiboly (6 hpf) in the dorsalmost cells of the blastula margin (not shown) and, soon after,expression extends completely around the margin with no obvious dorsoventral bias (Fig. 2A,B). fgf24 expression continues in marginal cells throughout gastrulation(Fig. 2C-F) and by the end of gastrulation is localized to the tail bud(Fig. 2G). Thus, fgf24has a similar expression pattern to that of fgf8 in early embryos(Fürthauer et al., 1997; Reifers et al., 1998).

We characterized in more detail the expression of fgf24 in gastrula-stage embryos by analyzing parasagittal sections of in situ hybridized mid-gastrula stage embryos (8 hpf). We compared the localization of fgf24 transcripts (Fig. 2H,L) in the paraxial level of the germ ring to that of fgf8 (Fig. 2I,M), and the mesodermally expressed T-box genes ntl(Fig. 2K,O) and spt(Fig. 2J,N). Although we found the expression domains of fgf24 and fgf8 have significant overlap, they are not identical. Specifically, cells expressing the highest levels of fgf24 localize to the hypoblast layer of the germ ring(Fig. 2L), and in this regard,expression of fgf24 is most similar to the expression of spt(compare Fig. 2L with 2N)(Griffin et al., 1998). By contrast, cells expressing the highest levels of either fgf8 or the T-box gene ntl localize to the epiblast layer of the germ ring(compare Fig. 2M with 2O)(Fürthauer et al., 1997; Schulte-Merker et al.,1992).

In addition to the germ ring, fgf8 is also expressed in the presumptive brain beginning at 8 hpf in a domain that spans from the future midhindbrain junction (MHB) posteriorly to rhombomere 4(Fig. 2I,M)(Reifers et al., 1998; Maves et al., 2002). At this stage of development, fgf24 is not expressed in the presumptive brain, though weak staining can be seen in dispersed cells within the presumptive spinal cord (Fig. 2E). Thus, the co-expression of fgf8 and fgf24in the germ ring, but not in the presumptive brain, supports the hypothesis that fgf24 functions with fgf8 during posterior mesoderm production and can readily explain why fgf8 mutant embryos have only mild defects in the development of posterior mesoderm, yet have significant defects in the development of the MHB(Reifers et al., 1998).

We directly tested the hypothesis that fgf24 and fgf8function redundantly during the development of posterior mesoderm by knocking down fgf24 gene function with antisense morpholino oligonucleotides(MOs) (Nasevicicius and Ekker,2000) targeted to a splice junction site in the fgf24pre-mRNA. Splice site-targeted MOs have been shown to alter pre-mRNA splicing when injected into zebrafish embryos, and have the advantage that their efficacy can be quantified by ribonuclease protection(Draper et al., 2001). We obtained a MO targeted to the splice donor site located at the junction of exon 3 and intron 3 (henceforth referred to as fgf24-E3I3; Fig. 3A). We first asked if fgf24-E3I3 could alter splicing of fgf24 pre-mRNA using RTPCR. We injected 5 ng of fgf24-E3I3 into one- to four-cell stage embryos and harvested RNA at 24 hpf. Using primers that span exon 3(Fig. 3A), we found that injection of fgf24-E3I3 results in two aberrant splice forms, one of which causes an ∼100 bp deletion in the fgf24 cDNA when compared with cDNA amplified from control embryos(Fig. 3B). We sequenced this RT-PCR product and found that the deletion results from the aberrant use of a cryptic splice donor site located 98 bp upstream of the correct exon 3 splice donor (Fig. 3C). Splicing at this cryptic splice donor shifts the reading frame of fgf24 mRNA such that only 19 of the 178 amino acids that are predicted to form the secreted Fgf24 protein are encoded. This severely truncated form of Fgf24 is predicted to be non-functional (Fig. 3C).

We next quantified the ability of fgf24-E3I3 to reduce the amount of correctly spliced fgf24 mRNA by ribonuclease protection. We injected fgf24-E3I3 into one- to four-cell stage embryos at doses ranging from 1.3-5.0 ng MO/embryo and harvested RNA at 24 hpf. Using a riboprobe that detects correctly spliced message(Fig. 3D) we found that injection of fgf24-E3I3 reduced the amount of wild-type fgf24 mRNA in a dose-dependent manner(Fig. 3E,F). Because injection of 5 ng fgf24-E3I3 per embryo reduced the amount of wild-type message to levels that were undetectable in our assay, we chose this amount for all subsequent experiments involving MO-induced knockdown of fgf24.

We asked what effect reducing fgf24 gene function had on the development of posterior mesoderm by injecting fgf24-E3I3 into one-to four-cell stage wild-type embryos to generate fgf24^MO embryos. We compared the amount of posterior mesoderm produced by injected and control siblings at the 12-somite stage by staining fixed embryos for marker genes that are expressed in restricted domains of posterior mesoderm. We assayed the production of axial mesoderm using anti-Ntl antibodies, which reveal cells in the notochord and tail bud (Schulte-Merker et al.,1992), and paraxial and intermediate mesoderm by in situ hybridization using probes specific for myod(Weinberg et al., 1996) and pax2.1 (Krause et al., 1991), respectively(Fig. 4A). We found that we could not detect differences in marker gene expression when comparing fgf24^MO embryos with wild-type control embryos(compare Fig. 4A with 4B). In addition, we compared the morphology of live fgf24^MO embryos and wild-type embryos at 24 hpf and again could not detect any significant differences (compare Fig. 4E with 4F). Thus,reducing the level of fgf24 mRNA to undetectable levels in early zebrafish embryos appears to have no detectable effect on the development of posterior mesoderm under our assay conditions.

To test the possibility that lack of fgf24 function in fgf24^MO embryos is compensated for by fgf8 function, we injected fgf24-E3I3 into fgf8mutant embryos. We will refer to fgf8^- embryos that have been injected with fgf24-E3I3 MO as fgf8^-;fgf24^MO embryos. At the 12 somite stage, fgf8^- embryos can be identified by their reduced expression of pax2.1 in the MHB(Fig. 4C)(Reifers et al., 1998). In addition to the MHB defects, fgf8 single mutants produce less somitic mesoderm than wild-type embryos (Fig. 4C) (Reifers et al.,1998). In contrast to the effect observed after injection of fgf24-E3I3 into wild-type embryos, injection into fgf8^- embryos resulted in severe defects in posterior mesoderm development. At the 12-somite stage (13 hpf) fgf8^-;fgf24^MO embryos had a significantly truncated notochord, and produced very few myod- and pax2.1-expressing cells (Fig. 4D) when compared with control wild-type embryos(Fig. 4A). When live embryos were examined at 24 hpf, fgf8^-;fgf24^MOembryos lacked the MHB (Fig. 4H), a phenotype that is identical to fgf8 single mutants(Fig. 4G), and additionally had severely truncated tails relative to wild-type embryos, fgf8^- or fgf24^MO embryos(Fig. 4F,G). In this respect, fgf8^-;fgf24^MO embryos more closely resembled embryos in which Fgf signaling had been inhibited by expression of a dnFgfr (Griffin et al., 1995)or spt;ntl double mutant embryos(Amacher et al., 2002).

Because the phenotype of fgf8^-;fgf24^MO embryos is similar to that of spt;ntl double mutants, we asked if the defects in posterior mesoderm development were associated with defects in the expression of ntl and spt. We first compared the expression patterns of ntl transcripts and Spt protein in eight-somite stage fgf8^-;fgf24^MO embryos with those of wild-type, fgf8^- and fgf24^MOembryos. We found that in comparison with wild-type embryos(Fig. 5A), fgf8^- (Fig. 5B), but not fgf24^MO embryos (data not shown), had reduced numbers of Spt protein-expressing cells in the presomitic mesoderm, and nearly 1/3 of the embryos had gaps in the axial mesodermal expression domain of ntl(Fig. 5B). Thus, loss of fgf8 function alone, but not fgf24, is sufficient to cause reduced levels of spt and ntl expression in developing posterior mesoderm, an observation that could explain why fgf8 single mutants have defects in somitogenesis (Fig. 4C) (Reifers et al.,1998). In contrast to single mutant embryos, we found that all fgf8^-;fgf24^MO embryos had severe defects in spt and ntl expression in posterior mesoderm. Although all of fgf8^-;fgf24^MO embryos had expression of ntl in anterior notochord cells(Fig. 4D, Fig. 5C), we could not detect expression of either ntl or Spt in more posterior regions(Fig. 5C) (see also supplemental Fig. S1). These data together suggest that fgf8^-;fgf24^MO embryos at the 10-somite stage do not contain mesodermal precursors in the tail bud.

We next asked at what stage expression of ntl and sptbecome dependent on the function of fgf8 and fgf24. We analyzed the expression of ntl and spt at the beginning of gastrulation, and then again in midgastrula stage (8 hpf) embryos. At the beginning of gastrulation, we could not distinguish differences in the expression of the T-box genes in fgf8^-;fgf24^MO embryos relative to wild-type embryos (data not shown). In mid-gastrula stage embryos, however, we found that fgf8^-;fgf24^MO embryos had markedly reduced expression of ntl relative to wild-type embryos,with the ventral germ ring having the most dramatic reduction (compare Fig. 5D with 5E). Similarly, we found that fgf8^-;fgf24^MO embryos had markedly reduced expression of spt in both the germ ring and presomitic mesoderm (compare Fig. 5F with 5G). Reducing the activity of fgf8 or fgf24alone did not result in significant decreases in ntl or sptexpression at the embryonic stages analyzed here (data not shown). Thus,cooperative function of fgf8 and fgf24 in the germ ring is required for continued high level expression of ntl and sptin mesodermal precursors, but they are not required for the initial expression of the T-box genes at early gastrula stages.

It had been proposed that Fgf signaling and T-box genes form an auto-regulatory feedback loop, where the expression of one maintains the expression of the other (reviewed by Smith, 1999). We therefore asked what effect loss of spt and ntl function had on the expression of fgf8 and fgf24 during gastrulation. We found that mid-gastrula-stage (8 hpf) ntl mutants had reduced expression of both fgf8 (Fig. 5I)and fgf24 (Fig. 5M) in axial, but not ventral mesoderm compared with wild-type embryos(Fig. 5H and 5L, respectively). Surprisingly, spt mutant embryos also failed to express fgf8in axial mesoderm, but had apparently normal expression of fgf8 in non-axial domains (Fig. 5J). By contrast, expression of fgf24 in spt mutant embryos was reduced ventrally, but not dorsally (Fig. 5N). Finally, we examined the expression of fgf8 and fgf24 in spt;ntl double mutant embryos, and found that expression levels of fgf8 were further reduced in the dorsal and lateral but not the ventral, germ ring(Fig. 5K). By contrast, we found that expression of fgf24 was reduced both dorsally and ventrally, but not laterally in spt;ntl double mutants (Fig. 50). These data show that wild-type function of spt and ntl are required for some, but not all, fgf8 and fgf24 expression in mesodermal precursors.

We have so far provided only indirect evidence based on phenotypic analysis and gene expression that interactions between the Fgf ligands Fgf8 and Fgf24,and the T-box genes spt and ntl are required for posterior mesoderm development in zebrafish. We tested this hypothesis more directly by asking if we could detect genetic interactions between the Fgf ligands and the T-box genes. We therefore constructed and analyzed fgf8;ntl and spt;fgf8 double mutants and used fgf24-E3I3 MO to create fgf24^MO;ntl and spt;fgf24^MOmutant embryos. In comparison with wild-type embryos(Fig. 6A), embryos single mutant for either fgf8 (Fig. 6B) or ntl (Fig. 6D) produce significant amounts of paraxial mesoderm, as revealed by the expression of myod at the 12-somite stage(Reifers et al., 1998; Halpern et al., 1993). By contrast, we found that fgf8;ntl double mutants produced significantly less paraxial mesoderm than would have been expected from simple addition of their single mutant phenotypes(Fig. 6E). Similarly, in comparison with wild-type embryos (Fig. 6K), embryos mutant for either fgf8(Fig. 6L) or spt(Fig. 6M) produce significant amounts of axial mesoderm, as revealed by the expression of Ntl protein in the nuclei of notochord cells (Fig. 6I-K). By contrast, we found that spt;fgf8 double mutants produce significantly less axial mesoderm than would have been expected from simple addition of their single mutant phenotypes(Fig. 6N). Fig. 6F-J,O-R give representative examples of live embryos at 24 hpf for each genotypic class(see Materials and methods for segregation frequencies). We did not observe significant differences in the amount of mesoderm produced by either fgf24^MO;ntl or spt;fgf24^MOembryos when compared with ntl or spt single mutants,respectively (see supplemental Fig. S2). These data provide direct evidence that fgf8 genetically interacts with ntl and spt during the development of posterior mesoderm.

When a pair of fish heterozygous for both ntl and fgf8(i.e. ntl^+/-;fgf8^+/-), are mated, four phenotypic classes are expected [wild type(Fig. 6F), fgf8 single mutant (Fig. 6G), ntlsingle mutant (Fig. 6I) and ntl;fgf8 double mutant (Fig. 6J)] that segregate in the ratio of 9:3:3:1, respectively. In this cross, however, we observed that the fgf8 single mutant class, which was distinguished by lacking the MHB but producing somites and a notochord,could be further sorted into two phenotypic subclasses based on tail length at 24 hpf, or by the amount of notochord produced when assayed for marker gene expression at the 12-somite stage; 1/3 of these embryos were indistinguishable from fgf8 single mutants (Fig. 6B,G) while 2/3 produced only anterior notochord and had tails that were intermediate in length between fgf8 single mutants and fgf8;ntl double mutants (Fig. 6C,H). Based on these segregation frequencies and the fact that both phenotypic classes produced notochord, we reasoned that the fgf8homozygotes that had short tails and reduced notochord development were heterozygous for the ntl mutation (i.e. fgf8^-/-;ntl^+/-), whereas those with long tails were ntl homozygous wild type (i.e. fgf8^-/-;ntl^+/+). We tested this hypothesis by crossing fgf8^+/-;ntl^+/-double heterozygous animals to fgf8^+/- single heterozygotes. In this cross, 1/2 of the fgf8^-/- embryos will also be genotypically ntl^+/-. Again, we found that we could divide the fgf8 homozygotes into two phenotypic classes: 1/2 of the fgf8 mutants segregating in this cross were indistinguishable from fgf8 single mutants, while the other 1/2 had short tails and patchy notochord, identical to the animals in Fig. 6C,H (see Materials and methods for segregation frequencies). These data show that a loss-of-function ntl allele can dominantly enhance the phenotype of fgf8mutant embryos, providing further support that ntl and fgf8interact genetically.

After the completion of gastrulation, fgf24 expression can be detected in a variety of tissues during somitogenesis and larval development. Expression of fgf24 in the tail bud mesenchyme can be detected in 12-18 hpf embryos (Fig. 7A,D,F), but it is no longer expressed in this domain at 24 hpf(Fig. 7H). fgf24 is expressed in the otic placode beginning around the twosomite stage (10.5 hpf,not shown) and is clearly visible at 12 hpf(Fig. 7A) as bilateral patches adjacent to rhombomere 5 (Fig. 7B). Co-labeling with the otic placode marker pax2.1(Krauss et al., 1991)indicates that fgf24 is uniformly expressed in the otic placode at 12 hpf (Fig. 7C). Expression of fgf24 in the developing ear is dynamic and by 24 hpf is localized to a discrete domain in the posterior otic epithelium(Fig. 7I). In addition to the otic placode, fgf24 is expressed in anterior neuroectoderm at 12 hpf,in a location that has been fate mapped to form the olfactory placode(Fig. 7A)(Whitlock and Westerfield,2000) and in 18 hpf embryos, expression can be seen in the forming nasal organs (Fig. 7E). Expression of fgf24 persists in the nasal organ through 52 hpf (the latest time point analyzed), at which point expression can also be detected in the olfactory bubs (Fig. 7M,O). Last, at 12 hpf, fgf24 expression is detected in bilateral stripes of cells that appear to be located in lateral mesoderm(Fig. 7A). We correlated this expression domain with the expression of pax2.1, which at this stage also labels the forming pronephric ducts, an intermediate mesodermal derivative (Krauss et al.,1991), and found that cells expressing fgf24 lay medial to, and do not appear to overlap with, those expressing pax2.1(Fig. 7D). This expression domain may identify precursors of regions of the gut because, at later stages, fgf24 is also expressed in the developing gut(Fig. 7F,H,J).

Beginning at the 16-somite stage (18 hpf), fgf24 expression is detected in a restricted domain in the medial retina(Fig. 7E) and in the caudal fin ectoderm (Fig. 7F). Expression in the caudal fin persists through 52 hpf(Fig. 7H and data not shown). Additionally, fgf24 is detected in bilateral domains of trunk mesoderm beginning at 18 hpf (not shown) which by 20 hpf(Fig. 7G) appear to mark the mesenchyme that will contribute to the developing pectoral fin bud. Thin transverse sections of 24 hpf larva confirm that this expression domain is restricted to the mesenchyme, and not the overlying surface ectoderm(Fig. 7I,J). By 52 hpf, when the developing pectoral fin is clearly visible, fgf24 expression is no longer detected in the fin mesenchyme, but is instead restricted to the apical ectodermal ridge (Fig. 7K), similar to the expression of fgf8(Reifers et al., 1998). At 24 hpf, fgf24 is also expressed in the pharyngeal pouches(Fig. 7I). In 52 hpf embryos, fgf24 expression continues in the pharyngeal arches, though the expression domains in the pouches become more restricted to their lateral tips(Fig. 7L-N). Additional pharyngeal arch expression domains are seen as well. For example, in the first arch, fgf24 is expressed in bilateral patches adjacent to the mouth opening (Fig. 7N), similar to what has been reported for fgf8(Reifers et al., 1998). In the second arch, fgf24 is expressed in the posterior ectodermal margin(Fig. 7M), similar to fgf8 in chicks (Wall and Hogen, 1998). Last, fgf24 is expressed in two posterior pharyngeal arch domains that appear to be tooth germs (Fig. 7M). fgf4,the mammalian ortholog of which marks a subset of the dental epithelium, is expressed in similar domains at this stage (B.W.D. and D.W.S., unpublished). In zebrafish, as in all cypriniforms, teeth form only on the most posterior(seventh) pharyngeal arch (Kimmel et al.,1995).

We used the fgf24-E3I3 MO to address the function of fgf24 in later development. As shown previously, fgf24^MO embryos at 24 hpf are morphologically indistinguishable from their control siblings(Fig. 3E,F). We therefore allowed fgf24^MO embryos to develop to various stages past 24 hpf, and assayed for morphological phenotypes. We found that at 33 hpf, fgf24^MO embryos were indistinguishable form their control sibling embryos, with the exception that they did not have visible pectoral fin buds, which are easily scored at this stage of development as discrete epidermal bumps on the dorsal yolk (not shown)(Kimmel et al., 1995; Grandel and Schulte-Merker,1998), or by their expression of shh(Fig. 8A,B)(Krauss et al., 1993). Surprisingly, we found that injected embryos could survive to adulthood, but they never develop pectoral fins.

We investigated the pectoral fin phenotype in more detail by analyzing skeletal preparations of 1-month-old wild-type and fgf24^MO fish stained with Alizarin Red and Alcian Green to visualize bone and cartilage, respectively. The skeleton of the paired pectoral fins in zebrafish consist of fin rays or lepidotrichia, that support the visible part of the fin, and a pectoral girdle located internally that provides support for the fin rays as well as articulation with the skull(Grandel and Schulte-Merker,1998). We analyzed pectoral skeletal morphology in wild-type(Fig. 8C,D) and fgf24^MO fish(Fig. 8E,F) and found that elements of the pectoral girdle, the cleithrum and postcleithrum, could be found in both wild-type and fgf24^MO fish. By contrast, elements derived from the fin bud (i.e. scapula, radials and lepidotrichia) could only be identified in wild-type(Fig. 8C), but not in fgf24^MO fish(Fig. 8E). Thus, loss of fgf24 function appears to affect a very early stage of pectoral fin development. A more detailed analysis of the role of fgf24 in pectoral fin development is presented elsewhere(Fischer et al., 2003).

We have described the identification and function of zebrafish fgf24, a new member of the fibroblast growth factor (Fgf) 8/17/18 subfamily of signaling molecules. Our results show that fgf24 is expressed in posterior mesodermal precursors during gastrulation where it functions cooperatively with fgf8 to promote mesodermal development,in part by maintaining the expression of the mesodermal T-box genes ntl and spt. We have presented double mutant analyses that reveal genetic interactions between the T-box genes and Fgf signaling. These results provide compelling evidence that these genes function in a genetic pathway that promotes posterior mesodermal development in zebrafish. Last, we have shown that fgf24 is expressed in a wide variety of tissues after gastrulation, including the early fin bud mesenchyme, and is required for an early stage of pectoral fin bud development.

With the addition of fgf24, at least 22 distinct Fgf-encoding genes have been identified in vertebrates (human FGF19 and mouse Fgf15 may be orthologous genes). Based on sequence relatedness, the Fgf superfamily can be subdivided into seven subfamilies of more closely related genes (reviewed by Ornitz and Itoh, 2001). The genes encoding the ligands Fgf8, Fgf17,Fgf18, and Fgf24 define one such subfamily and mouse members of this subfamily have been shown to have very similar Fgf receptor specificity profiles (Xu et al., 2000). It is therefore likely that in zebrafish Fgf8 and Fgf24 have similar activities.

Because Fgf24 so far appears to be unique to zebrafish, it is necessary to consider its origin. There is increasing evidence that a whole-genome duplication event occurred in the rayfinned fish lineage after it diverged form the terrestrial vertebrate lineage(Amores et al., 1998; Postlethwait et al., 1998; Force et al., 1999; Postlethwait et al., 2000). It is therefore possible that fgf18 and fgf24 are paralogs that arose by duplication of an ancestral fgf18 during this proposed event. However we have shown that fgf24 and fgf18 are closely linked on LG14, whereas paralogs that resulted from a genome duplication event are expected to be unlinked (see Woods et al., 2000). In addition, we found compelling evidence of conserved gene synteny around the zebrafish and human Fgf18 loci. By contrast, the fgf24 locus does not appear to be in a region with conserved synteny with any region of the human genome. Finally, the grouping of zebrafish Fgf24 with zebrafish Fgf18 is contradicted by a node with 97% bootstrap support in our phylogenetic analysis of Fgf protein sequences. Thus, our data argues that fgf24 and fgf18 are not paralogous genes that resulted from the ray-finned fish-specific genome duplication.

We instead favor the model that fgf24 and fgf18 are paralogs that resulted from a gene duplication event that predates the divergence of ray-finned fish and terrestrial vertebrate lineages. Based on protein sequence relatedness, Fgf24 is as similar to Fgf18 orthologs,as Fgf17 orthologs are to Fgf8. It is therefore possible that a single ancestral gene, following two sequential duplication events,gave rise to the four members of the Fgf8 subfamily. A similar hypothesis has been proposed for the origin of the four tetapod Hox clusters(discussed by Furlong and Holland,2001). In support of this model, a probable fgf24ortholog has been identified in a shark (D.W.S., unpublished), arguing that fgf24 arose early in gnathostome (jawed vertebrate) evolution. It is therefore likely that an fgf24 ortholog was lost at some point in the terrestrial vertebrate lineage after its divergence from ray-finned fishes. Similar examples of lineage-specific gene loss have already been described,including the loss of functional copies of the hox paralogs hoxb10, hoxc1 and hoxc3 in the mammalian lineage, but not in zebrafish (Amores et al., 1998; Prince et al., 1998; Postlethwait et al.,1998).

Our results show that fgf8 and fgf24 are components of the Fgf signaling pathway that regulates posterior mesoderm development in zebrafish. We found that fgf8 and fgf24 are expressed in mesodermal precursors and that fgf8^-;fgf24^MO embryos produce very little posterior mesoderm. Although the function of fgf8 and fgf24 can account for much of the Fgf signaling activity that is known to be required for posterior mesoderm development in zebrafish, we observed that fgf8^-;fgf24^MO embryos produce significantly more mesoderm than do embryos overexpressing the dnFgfr(Griffin et al., 1995). Because the dnFgfr is likely to block all Fgf signaling in early embryos(Ueno et al., 1992), ligands in addition to Fgf8 and Fgf24 are likely to contribute to early mesoderm formation in zebrafish. In addition to fgf8 and fgf24, fgf3is the only other Fgf gene in zebrafish that is known to be expressed in mesodermal precursors during gastrulation(Fürthauer et al., 2001). Although fgf3 may account for some of the Fgf activity present in fgf8^-;fgf24^MO embryos, it is not likely to account for all; injection of fgf3 MO(Maves et al., 2002) into fgf8^-;fgf24^MO embryos does not appear to decrease the amount of posterior mesoderm produced relative to fgf8^-;fgf24^MO embryos alone (L. Maves and B.W.D., unpublished).

We cannot rule out the possibility that the mesoderm produced by fgf8^-;fgf24^MO embryos is due to residual activity of fgf8 and/or fgf24 in these embryos. The single fgf8 allele that has been isolated, fgf8^ti282(Reifers et al., 1998), is likely to be a hypomorph (Draper et el.,2001). However, using fgf8 MOs, which reduce the expression of functional fgf8 below the level produced by the fgf8 mutation (Draper et al.,2001), in combination with the fgf24 MO, does not appear to increase the severity of the phenotype relative to the fgf8^-;fgf24^MO embryos (B.W.D.,unpublished). Similarly, it is possible that our fgf24 MO does not completely eliminate fgf24 function, although our RNase protection results argue against this. Last, fgf8(Reifers et al., 1998; Draper et al., 2001), fgf24 and fgf18 (this study) are expressed maternally and these maternal mRNAs persist for several hours after fertilization. It is therefore possible that sufficient amounts of Fgf protein are produced from wild-type maternal transcripts to allow partial mesoderm development in the absence of zygotic fgf8 and fgf24 function. As only a few orthologs of the known vertebrate Fgf ligands have been identified in zebrafish, it remains to be seen how many other ligands participate in posterior mesodermal development.

Current models for how Fgfs and T-box genes interact during mesodermal development have proposed that they form an auto-regulatory feedback loop,where the function of one component maintains the expression of the other(reviewed by Isaacs, 1997). Although it is not yet clear how Fgf signaling regulates T-box gene expression, there is evidence in Xenopus that Xbra, the ortholog of ntl, can directly regulate the expression of embryonic(e)Fgf, an Fgf4 ortholog(Casey et al., 1998). This model predicts that wild-type expression patterns of fgf8 and fgf24 should require ntl and spt function, and indeed we found this to be true. However, ntl and spt can not be the only regulators of fgf8 and fgf24 expression during early mesoderm formation, as expression of fgf8 and fgf24 persist in the germ ring of early spt;ntl double mutant embryos. In addition to spt and ntl, the spt-related gene tbx6 is also expressed in mesodermal precursors during gastrulation (Hug et al., 1997). tbx6 is unlikely to contribute to Fgf regulation in the absence of spt and ntl function, however,because it is not expressed in spt;ntl double mutants(Griffin et al., 1998).

The expression patterns we observed for fgf8 and fgf24 in wild-type embryos and in embryos mutant for either ntl or spt suggest that their expression in the germ ring is not regulated by an identical genetic network. First, the expression patterns of fgf8 and fgf24 in wild-type embryos, while overlapping, are not identical. We found that cells expressing the highest levels of fgf8 localize to the epiblast layer (similar to ntl),whereas those expressing the highest levels of fgf24 localize to the hypoblast layer (similar to spt). As might be expected from these expression patterns, expression of fgf8 and fgf24 also have non-identical requirements for ntl and spt function. However, we did not observe a simple one-to-one correlation between an Fgf expression domain and a T-box gene. Instead, we found that the expression of fgf8 in dorsal mesodermal precursors requires both ntl and spt function, while neither gene was required for fgf8expression in ventral precursors. By contrast, expression of fgf24 in dorsal mesodermal precursors requires ntl, but not spt,whereas ventral expression requires spt, but not ntl. Although it is not possible at present to derive an accurate pathway that explains the regulatory relationships that exist between these Fgfs and T-box genes, our data are consistent with the proposed feedback loop because we have found that reduction of Fgf signaling leads to a reduction of T-box gene expression and vice versa.

Data supporting the model that posterior development is promoted by a regulatory network between Fgfs and T-box genes has come largely from analyzing gene expression defects in single mutant embryos (e.g. Yamaguchi et al., 1994; Deng et al., 1994; Sun et al., 1999) or in embryos overexpressing single network components (e.g. Isaacs et al., 1994; Schulte-Merker and Smith,1995). We have provided genetic evidence that directly links Fgf signaling and T-box gene function in a genetic pathway that promotes posterior development. We have shown that fgf8;ntl and spt;fgf8 double mutants had phenotypes that were more severe than would be expected from the simple addition of either single mutant phenotype. For example, neither fgf8 nor ntl has severe defects in trunk somite formation,as assayed by myod expression, whereas fgf8;ntl double mutants produce few myod-positive cells. Because trunk somite formation is known to require spt function cell-autonomously(Ho and Kane, 1990), we propose that the muscle phenotype observed in fgf8;ntl embryos results from attenuated spt function. Similarly, we found that spt;fgf8 double mutant embryos appear to have attenuated ntlfunction as notochord development was reduced in double mutant embryos, but not in spt or fgf8 single mutant embryos. These results indicate that fgf8 cooperates with ntl to maintain spt function, and similarly with spt to maintain ntl function.

It is interesting that the expression of pax2.1, which marks the developing pronephric tubules (Krauss et al., 1991), is largely unaffected in either fgf8;ntl or spt;fgf8 mutants. Pronephric tubules develop from intermediate mesoderm and spt and ntl are redundantly required for their formation (Amacher et al.,2002). It is possible that pronephric development requires lower levels of T-box gene activity relative to that required for the development of the notochord and somites. Alternatively, Fgf8 signaling may promote the expression of dorsal-specific factors that function in combination with ntl and spt to promote the development of dorsal mesodermal derivatives, such as notochord and somites, but not the development of more intermediate derivatives, such as pronephros. In support of this, fgf8 is expressed at higher levels in dorsal mesoderm than ventral mesoderm and fgf8 overexpression can strongly dorsalize early zebrafish embryos (Fürthauer et al.,1997).

In addition to the interactions described above, we found that ntlmutations dominantly enhance the phenotype of fgf8 homozygotes: fgf8^-/-;ntl^+/- embryos produced less posterior mesoderm than fgf8^-/-;ntl^+/+embryos. Because ntl heterozygotes alone are phenotypically wild type, this result suggests that ntl function is attenuated in fgf8 single mutants, and consistent with this, we found that a third of the fgf8 single mutants have reduced ntl expression in axial mesoderm. These results imply that in the absence of fgf8function, fgf24 function alone is not sufficient to maintain wild-type levels of ntl activity. Interestingly, T null mutations in mice, but not in zebrafish, are semi-dominant as T^+/- heterozygotes have shorter tails than do wild-type mice (Dobrovolskaïa-Zavadskaïa,1927). Because wild-type expression of T in mouse mesodermal precursors is known to be dependent on Fgf8 function(Sun et al., 1999), it is possible that the apparent differences between the phenotypes of T/ntl heterozygotes in mice and zebrafish are due simply to differences in the quantitative levels of Fgf signaling in posterior tissue between these two organisms. In contrast to dominant interactions between ntl and fgf8, we could not find evidence that reduction of spt function could dominantly enhance the phenotype of fgf8mutant embryos, suggesting that the interactions between fgf8 and ntl are stronger than those between fgf8 and spt. Genetic interactions between the Fgf signaling pathway and T-box transcription factors is becoming a common theme in vertebrate development, as similar interactions have been proposed to play key roles in development of limbs(Ng et al, 2002) the cardiovascular system (Vitelli et al.,2002) and lungs (Cebra-Thomas et al., 2003).

Last, we have shown that fgf24 expression is not restricted to developing posterior mesoderm, but is also expressed in a wide variety of tissues during larval growth. However, the only defect we could identify in fgf24^MO embryos was in the development of pectoral fins. We found that fgf24^MO fish never produced a morphological fin bud, nor did they produce any skeletal elements of the external pectoral fin. Thus, the phenotype of fgf24 appears similar to that of mice mutant for either fgf10(Min et al., 1998; Sekine et al., 1998) or its likely receptor, Fgfr2(Xu et al., 1998).

We thank Thierry Lepage and David Kimelman for generously sharing their cDNA library; Karen Larison and Jewel Parker for expert sectioning assistance;Brian Summers for excellent technical assistance; Angel Amores for help with PAC library screening; Craig Miller for assistance identifying expression domains in larva and for many thoughtful discussions; members of the Kimmel laboratory for thoughtful discussions; The University of Oregon Zebrafish Facility for excellent fish care; David Kimelman, Cecilia Moens, Craig Miller and Lisa Maves for excellent comments on the manuscript; and Kenneth Weiss for research support of D.W.S. This work was supported by the NIH (5 PO1 HD22486 to C.B.K.), the Damon Runyon Cancer Research Foundation (B.W.D.) and the NSF(SBR 9408402 to K. M. Weiss).