HSPG synthesis by zebrafish Ext2 and Extl3 is required for Fgf10 signalling during limb development

The development of metazoan animals is directed by cell-cell signalling mediated by members of the Wnt, Hedgehog (Hh), Bmp and Fgf families of secreted signalling proteins. Recently, extensive biochemical and genetic studies have demonstrated that heparan sulphate proteoglycans (HSPGs) play a crucial role in regulating the extracellular distribution, movement and activity of these signalling factors (reviewed by Esko and Selleck, 2002; Nybakken and Perrimon, 2002; Lin, 2004). HSPGs are a diverse group of macromolecules associated with the cell-surface and extracellular matrix, and consist of a core protein to which heparan sulphate(HS) side chains are attached (reviewed by Esko and Selleck, 2002). HS chains are long unbranched polysaccharides consisting of repeating disaccharide units of uronic acid linked to glucosamine. HS chain synthesis occurs in the Golgi apparatus, and is initiated at the HS attachment sites of core proteins, followed by polymerisation and several modifications(Esko and Selleck, 2002; Nybakken and Perrimon, 2002; Lin, 2004).

The exostosin (ext) gene family has been shown to encode glycosyltransferases that synthesise the polymerisation of HS side chains of HSPGs (reviewed by Zak et al.,2002). Vertebrate exostosin genes include Ext1 and Ext2, as well as the exostosin-like genes Extl1, Extl2, and Extl3. In Drosophila, 3 exostosin genes have been identified: ext1, ext2 and extl3(Bornemann et al., 2004; Han et al., 2004; Takei et al., 2004). All three Drosophila exostosin genes participate in shaping the extracellular morphogen gradients of Hh, Dpp/Bmp and Wg/Wnt, as well as regulating their signalling activity (reviewed by Lin,2004). Interestingly, the effect of Exostosins on specific signalling proteins is context dependent. For example, although all three Drosophila exostosins are crucial for Dpp/Bmp signalling in the wing imaginal disc, they have no effect on Dpp/Bmp signalling during embryogenesis(The et al., 1999; Bornemann et al., 2004; Han et al., 2004; Takei et al., 2004). Likewise,neither Hh, nor Wg/Wnt signalling is defective in ext2 mutants during embryogenesis (The et al.,1999), indicating that the control of developmental signalling by HSPGs is both signal and context dependent.

Fibroblast growth factors (Fgfs) comprise a large family of signalling molecules involved in regulating many cellular responses during development,and often participate in reciprocal signalling across epithelial-mesenchymal boundaries (reviewed by Ornitz,2000; Itoh and Ornitz,2004). A well-studied process in which Fgfs mediate epithelial-mesenchymal interactions is during limb development (reviewed by Johnson and Tabin, 1997; Martin, 1998; Tickle and Munsterberg, 2001). Fgf10 is expressed in the limb bud mesenchyme at very early stages in mouse and chicken embryos, and is required for the activation of genes expressed in the overlying apical ectodermal ridge (AER), including Fgf4 and Fgf8(Ohuchi et al., 1997; Min et al., 1998; Sekine et al., 1999). Fgf10 binds with highest affinity to the Fgfr2b splice variant of Fgf receptor 2,which is expressed in epithelial cells, whereas Fgf4 and Fgf8 have highest affinity for mesenchymally expressed Fgfr2c(Orr-Urtreger et al., 1993; Ornitz et al., 1996). This scenario suggests a model in which Fgf10 signals to Fgfr2b in the overlying ectoderm, leading to activation of Fgf4 and Fgf8, which then signal back to the mesenchyme via Fgfr2c to maintain Fgf10 expression. Thus, a positive feedback loop is formed, based on mutual dependence (reviewed by Xu et al., 1999). As in mouse and chick, the zebrafish fgf10 gene is expressed in the mesenchyme of the pectoral fin buds, which are homologous to tetrapod limb buds(Ng et al., 2002). Following initiation of outgrowth, the limb field becomes patterned along three main axes (Johnson and Tabin, 1997; Martin, 1998; Capdevila and Izpisua Belmonte,2001). Signals from the AER control limb outgrowth along the proximodistal axis. The anteroposterior axis is patterned by Shh secreted from the zone of polarising activity (ZPA). Finally, the limb bud is patterned along its dorsoventral axis, leading to expression of Wnt7a and Eng1 in dorsal and ventral ectoderm, respectively.

A number of biochemical studies have implicated HS in the regulation of Fgf signalling (reviewed in Ornitz,2000). Structural studies have led to the proposal that either one HS chain forms a ternary complex with one Fgf and one Fgfr molecule(Schlessinger et al., 2000),or that one HS chain binds to two Fgf molecules to form a dimer bridging two receptors (Pellegrini et al.,2000). Interestingly, as distinct Fgfs differ in the amino acid sequence of their heparin-binding sites, they may have distinct requirements for HS to exert their biological activities(Bellosta et al., 2001). However, in contrast to the wealth of biochemical data, there is a scarcity of in vivo studies addressing the role of HS in Fgf signalling. The first genetic evidence for a role of HSPGs in Fgf signalling came from the observation that Drosophila mutants in UDP-glucose dehydrogenase, an enzyme which catalyses the formation of an essential building block of HS polysaccharides,are defective for Fgf signalling (Lin et al., 1999). Recently, a mouse mutant disrupting the same gene was shown to cause gastrulation defects owing to abrogation of Fgf8 signalling(Garcia-Garcia and Anderson,2003).

Exostosins have so far not been directly linked to Fgf signalling in vivo. Ext1 knockout mice die early, as a result of defective gastrulation(Lin et al., 2000). In these mutants, Indian Hedgehog (Ihh) protein fails to associate with the surface of target cells, suggesting a role for Ext1 in Hh signalling. This proposal is further supported by the observation that Ext1 hypomorphic mutations result in an increase in the range of Ihh signalling during bone development(Koziel et al., 2004). Conditional inactivation of Ext1 in the mouse brain leads to a number of defects, some of which may be caused by a reduction of Fgf8 signalling(Inatani et al., 2003). The zebrafish mutants dackel and boxer disrupt ext2 and extl3 respectively, and these genes are required for axon sorting in the optic tract (Lee et al.,2004). Ext2 and Extl3 are broadly and uniformly expressed during embryogenesis, and disruption of both genes causes a global reduction in HS levels (Lee et al., 2004). Both dackel and boxer were originally isolated on the basis of their defective pectoral fin development(van Eeden et al., 1996), and dackel is known to be required for AER maintenance in the pectoral fin bud (Grandel et al.,2000).

In this study, we investigate how ext2 and extl3 mutants affect Fgf signalling during limb development. We show that like ext2,extl3 is required for AER maintenance, although its phenotype is weaker and more variable than that of ext2. Both ext2 and extl3 show a very similar phenotype to daedalus, a novel zebrafish pectoral fin mutant. We find that daedalus disrupts fgf10, thus suggesting that Fgf10 signalling is affected by Ext2 and Extl3. Consistent with this hypothesis, we find that a partial reduction of fgf10 levels leads to a strong enhancement of the extl3 limb phenotype. Furthermore, application of Fgf10 protein rescues target gene expression in fgf10 mutants, but not in ext2 or extl3 mutants, suggesting that activity of these genes is necessary for Fgf10 signalling. Interestingly, application of Fgf4 protein can activate target genes in both ext2 and extl3 mutants, thus revealing an unexpected specificity for HSPGs in regulating signalling by distinct vertebrate Fgf ligands.

The following alleles were used: daedalus(dae^tbvbo; dae^t24030); boxer(box^tw24) (van Eeden et al., 1996); and dackel (dak^tw25e)(van Eeden et al., 1996). dae^–/–;box^–/– double mutants were generated by crossing together dae^tbvbo and box^tw24. The dae^tbvbo and dae^t24030 alleles have very similar phenotypes, but as dae^tbvbo encodes a null allele, we used these mutants for phenotypic characterisation. In some cases, dae^tbvboembryos were genotyped using a SNP. Genomic DNA was amplified with the PCR primers GCTCTTCCCAGTTTTCCGAGCTCCAGGACAATGTGCAAATCG (forward) and TCCGTTCTTATCGATCCTGAG (reverse), followed by digestion with Taq1. Wild-type embryos generated a band of 260 bp. dae^tbvbomutant embryos were not digested by Taq1 and produced a band of 300 bp; heterozygous embryos were identified as having two bands (300 bp and 260 bp) following electrophoresis on a 2% agarose gel. The residual 40 bp band formed by digestion of wild-type or heterozygous embryos was run off the gel and is not shown in Fig. 7. Embryos were cultured in E3 medium, with or without the addition of 0.003%1-Phenyl-2-thiourea (PTU, Sigma) to inhibit pigmentation. Embryos were staged according to hours post fertilisation (hpf)(Westerfield, 1995).

For fine mapping of dae, SSLPs were generated by using a zebrafish SSR search website(http://danio.mgh.harvard.edu/markers/ssr.html)in combination with the Sanger genome database. The closest SSLP marker to the dae mutation uses the primer pair TCGTCTGTCAGCTCAACCCTA (forward) and GGTACTAAGTGAAGCACTCTTACTCT (reverse), at a distance of 0.286 cM (2/698 meioses) upstream of the mutation. The PCR primers CAGACACGATCACTACGGACGCTTTAC(forward) and AGCTTGACTAAATTCGGATGGTAGGAT (reverse) were designed to amplify a 1061 bp fragment of DNA that included the entire fgf10 open reading frame. rtPCR was performed using cDNA from both sibling and mutant embryos,followed by cloning of the fragment in the TOPO TA vector (Invitrogen) for sequence analysis.

Fgf10 splice morpholino oligonucleotide (MO) was purchased from GeneTools. The MO, designed to target the exon2-intron2 splice junction, has the sequence GAAAATGATGCTCACCGCCCCGTAG (e2i2 MO). A MO stock solution was formed by dilution in water and was stored at –20°C prior to use. Embryos were injected at the single cell stage with 0.125 mM MO, allowed to develop for 3 days and were then scored for pectoral fin phenotype. Embryos were snap frozen in liquid nitrogen and RNA was subsequently extracted. To confirm splicing defects following MO injection, rtPCR was carried out using the Superscriptase II kit (Invitrogen) and the Fgf10F and Fgf10R primers described above.

In situ hybridisation was performed as previously described(Macdonald et al., 1994). The following mRNA in situ probes were used: bmp2b(Martinez-Barbera et al.,1997), dlx2a (Akimenko et al., 1994), dusp6(Kawakami et al., 2003), eng1a (Ekker et al.,1992), erm1 (Roehl and Nusslein-Volhard, 2001), fgf4(Grandel et al., 2000), fgf8 (Reifers et al.,1998), fgf10 (Ng et al., 2002), fgf24(Fischer et al., 2003), pea3 (Roehl and Nusslein-Volhard,2001), shh (Krauss et al., 1993), sp8(Kawakami et al., 2004), sp9 (Kawakami et al.,2004), wnt3l (Krauss et al., 1992) and wnt7a (see below). Alcian Blue staining of cartilage was performed as described previously(Grandel and Schulte-Merker,1998). Histological sections were obtained by staining cryosections with Methylene Blue (Humphrey and Pittman, 1974).

A novel gene encoding a zebrafish wnt7a orthologue was identified as lying between 21837919 and 21842385 bp on chromosome 11 using the zebrafish genome server(http://www.ensembl.org/danio_rerio). rtPCR was performed using the Superscriptase II kit (Invitrogen) and the following primers: GCCGCTGGATTTTTCACAT (wnt7aF); TGTGTACACTTCTGTCCGTTCACT(wnt7aR). The amplified fragment was cloned in the TOPO TA vector (Invitrogen)and was sequenced and analysed to confirm its identity (see Fig. S1 in the supplementary material).

Bead implantation was carried out as described previously(Grandel et al., 2000). Recombinant human Fgf4 and Fgf10 protein (R&D systems) was dissolved at a concentration of 1 μg/μl in phosphate-buffered saline with 0.1% bovine serum albumin. All batches of beads loaded with Fgf10 were tested by implantation into dae embryos, and assayed for gene rescue, before using the same batch to implant beads into dak or box mutant embryos.

daedalus (dae) is a novel zebrafish mutant isolated during a recent large scale genetic screen(Habeck et al., 2002) on the basis of its pectoral fin morphology. At 5 days post fertilisation, mutant larvae have severely truncated pectoral fins(Fig. 1B). They show no other apparent defects, but fail to inflate their swim bladder and most mutants die at 2 weeks of age. In order to examine the fin endoskeleton in daemutants, we performed Alcian Blue staining(Fig. 1C,D)(Grandel and Schulte-Merker,1998). Wild-type larvae have a pectoral fin skeleton that consists, in proximal-to-distal sequence, of a scapulocoracoid, postcoracoid process and an endoskeletal disk (Fig. 1C). In contrast to this, dae mutant embryos lack the entire endoskeletal disk and have a dysmorphic scapulocoracoid with most of the postcoracoid process missing (Fig. 1D). To further characterise the pectoral fin defect in dae, we examined transverse sections of 40-hour mutant and sibling embryos stained with Methylene Blue (Fig. 1E,F). Wild-type fin buds have a morphologically distinct apical ridge at this stage, which corresponds to the AER in tetrapods(Fig. 1E)(Grandel and Schulte-Merker,1998). In contrast to this, mutant fin buds appear smaller and undifferentiated, and do not have an apical ridge, although there is a slight apical thickening (Fig. 1F). These observations indicate that dae disrupts a gene required for both the development of distal structures and the integrity of the apical ridge of the zebrafish pectoral fin bud.

To define the molecular function of dae, we identified the gene disrupted by the dae mutation(Fig. 2). Initial bulk segregant analysis of pools of 48 sibling and 48 mutant embryos placed the dae locus on linkage group 21(Fig. 2A). Fine mapping using both previously available and novel simple sequence length polymorphisms(SSLPs), placed dae within an interval that corresponds to 0.97 cM. Of five genes located between the two closest SSLP markers, fgf10 was the best candidate for the mutated gene(Fig. 2A). We cloned and sequenced the entire fgf10 open reading frame of both daealleles. dae^tbvbo was found to have a lysine (aaa) to stop(taa) change at amino acid position 5, thus generating a protein null allele. dae^t24030 encodes an amino acid substitution of methionine(atg) to valine (gtg) at position 170 (Fig. 2B). To further confirm that disruption of fgf10 causes the dae phenotype, we designed a morpholino (MO) to target the exon2-intron2 splice junction of fgf10 (e2i2 MO; Fig. 2B) and injected this into wild-type embryos at the single cell stage. Injected embryos showed a striking dae phenocopy (52% injected embryos; Fig. 2D). We analysed splicing defects following MO injection by extracting RNA from morphant embryos and performing rtPCR. As expected, the morphant rtPCR reaction predominantly generated a smaller band than the full-length transcript found in uninjected siblings (Fig. 2E), confirming that aberrant splicing had occurred (Draper et al., 2001). Taken together, these data indicate that dae disrupts the zebrafish fgf10 gene.

As mouse Fgf10 is required for the establishment of the AER(Min et al., 1998; Sekine et al., 1999), we analysed the expression of AER markers by in situ hybridisation in fgf10/dae mutant and sibling embryos. fgf24 acts upstream of fgf10 during fin development(Fischer et al., 2003). We therefore examined the early expression of fgf24 in dae. At 28 hpf, expression of fgf24 in the fin bud is similar in wild-type and dae (Fig. 3A). However, by 38 hpf, expression of fgf24 is absent in dae(Fig. 3B,C), suggesting that fgf10 activity is necessary to maintain fgf24 expression during development. At early stages of fin development (28 hpf), expression of the ectodermal markers bmp2b (Fig. 3D) and sp8 (Fig. 3H) is present at reduced levels in fgf10/dae. However, sp9, dlx2a and wnt3l expression(Krauss et al., 1992; Akimenko et al., 1994; Kawakami et al., 2003)(Fig. 3K,L,O,P,S,T) is strongly reduced even at early stages, indicating that zebrafish Fgf10 already contributes to signalling from the mesenchyme to the ectoderm at this stage. By 38 hpf of development, expression of all AER markers analysed, including bmp2b (Fig. 3E,F), sp8 (Fig. 3I,J), dlx2a (Fig. 3Q,R), fgf8 (Fig. 3W,X), sp9 (Fig. 3M,N), wnt3l (Fig. 3U,V) and fgf4 (data not shown) is absent from the fin ectoderm in dae. Taken together, these results indicate that zebrafish fgf10/dae is crucial for AER induction and maintenance. At early stages, there is a low level of sp8 and bmp2b expression in the AER of fgf10/dae mutants, while all AER markers are completely lost in fgf10/dae mutants at later stages.

We next analysed the expression of several genes known to be targets of Fgf signalling during limb development. shh, which is expressed in the zone of polarising activity (ZPA), depends on Fgf4 and Fgf8 signalling from the AER (Sun et al., 2002). At 28 hpf, we detected normal expression of shh in fgf10/daefin buds (Fig. 4A,B), but we observed a strong reduction of shh expression by 38 hpf(Fig. 4C,D). Similarly, the expression of direct Fgf target genes such as pea3, erm1(Roehl and Nusslein-Volhard,2001) and dusp6 (formerly mkp3)(Kawakami et al., 2003) was present at 28 hpf, but absent by 38 hpf of development(Fig. 4E-O), although expression of pea3 and dusp6 is already weakly reduced in fgf10/dae at 28 hpf (Fig. 4E,F,L,M). We then analysed markers of the dorsoventral (DV) axis expressed in the ectoderm, which might depend on Fgf10 signalling from the underlying mesenchyme. The expression of eng1a in the ventral ectoderm is weakly reduced at 28 hpf (Fig. 5A,B), but virtually absent by 38 hpf(Fig. 5C,D). We also examined the expression of zebrafish wnt7a, which is expressed in the dorsal ectoderm, as observed in other vertebrate species(Capdevila and Izpisua Belmonte,2001). We find that, similar to eng1a, wnt7a expression is present in fgf10/dae mutants at 28 hpf(Fig. 5E,F), but is absent at 38 hpf (Fig. 5G,H). Together,these results indicate that Fgf-dependent marker gene expression is initially established in zebrafish fgf10 mutants, but is lost by around 36 hpf of development. Similarly, expression of eng1a in the ventral ectoderm, and wnt7a in the dorsal ectoderm is initiated normally in the absence of Fgf10, but is subsequently downregulated.

The fgf10/dae pectoral fin phenotype described here is similar to that of the zebrafish dackel (dak) mutant(Grandel et al., 2000), which has recently been shown to disrupt the ext2 gene(Lee et al., 2004). The zebrafish boxer (box) mutant disrupts extl3,another exostosin family member and shares several phenotypes with ext2/dak (van Eeden et al.,1996; Lee et al.,2004). Therefore, we compared the pectoral fin phenotype of extl3/box to that of ext2/dak and dae in more detail. At early developmental stages (28 hpf), ext2/dak mutants show only very weak effects on fin development, and strongly resemble wild-type siblings (Grandel et al.,2000). We find that extl3/box mutants show even weaker phenotypes at this stage and are virtually indistinguishable from wild-type siblings (data not shown). At 48 hpf, we find that expression of the AER markers dlx2a and fgf8 is strongly reduced in extl3/box mutants (Fig. 6A,B,D,E). Similarly, expression of the Fgf-dependent marker shh, and the expression of eng1a is reduced in extl3/box mutants at this stage(Fig. 6G,H,J,K). The reduction of these markers is not as severe as in fgf10/dae or in ext2/dak mutants (Fig. 6C,F,I,L), but the phenotype of extl3/box mutants is more variable than that of the other two mutants. In a few strongly affected extl3/box mutants, marker gene reduction is as severe as in fgf10/dae or in ext2/dak mutants (data not shown). These results indicate that extl3/box has a similar, although on average weaker, phenotype to fgf10/dae and ext2/dak, thus raising the possibility that these three genes function in the same pathway.

To further explore the possibility that Fgf10 and Extl3 function in the same genetic pathway, we crossed the fgf10/dae mutation into the extl3/box mutant background. As the extl3/box phenotype is weaker than that of the fgf10/dae and ext2/dak mutants, we reasoned that a low level of Fgf10 signalling is retained in extl3/box mutants. If so, further reduction of Fgf10 signalling, by reducing the level of Fgf10 protein through genetic removal of one copy of the fgf10/dae gene, should cause an enhancement of the extl3/boxphenotype (Fig. 7). Indeed, we observe that box^–/–;dae^+/–larvae have a much stronger pectoral fin reduction than box mutants alone, resembling fgf10/dae or ext2/dak mutants in severity(Fig. 7E). We identified double mutants both on the basis of their much severer phenotype (73 out of 150, or 49% of total mutants scored, corresponding to the expected Mendelian ratio)and by SNP genotyping (Fig. 7F). Following PCR amplification and digestion with Taq1,extl3/box larvae with a strong reduction of fin tissue were confirmed to be dae^+/– by the presence of two bands visible on an agarose gel. This strong genetic interaction between extl3 and fgf10 during limb development further indicates that these genes act in the same pathway.

Our data indicate that fgf10/dae activity is necessary for maintenance of eng1a expression in the ventral ectoderm. However, as Fgf10 signalling also activates fgf4 and fgf8 expression in the AER, it is presently not clear if eng1a depends directly on Fgf10 from the underlying mesenchyme, or if it instead depends on Fgf4/8 signalling from the AER. In order to distinguish between these possibilities, we performed gain-of-function experiments by applying either Fgf10 or Fgf4 protein to fgf10/dae mutant limb buds. We find that implantation of Fgf10-soaked beads into fgf10/dae mutants results in the rescue of fgf8 expression in the AER (4 out of 7 dae; Fig. 8A). This treatment also rescues shh expression in the posterior mesenchyme (5 out of 5 dae; Fig. 8B), eng1a expression in the ventral ectoderm (4 of 7 dae: Fig. 8C) and wnt7a in dorsal ectoderm (5 out of 5 dae: Fig. 8D) of fgf10/daemutant limb buds. Implantation of control beads soaked in PBS has no effect on mutant limb buds (Fig. 8E). Interestingly, implantation of Fgf4-soaked beads also leads to rescue of shh (6 out of 9 dae; Fig. 9B), eng1a (8 out of 11 dae; Fig. 9C) and wnt7a (3 out of 3 dae; data not shown) expression in fgf10/daemutants, but is unable to rescue fgf8 expression (0 out of 6 dae; Fig. 9A). These results indicate that Fgf4 is able to activate eng1a in the absence of fgf10 activity, suggesting that the effect of fgf10 on eng1a expression is mediated by the activation of fgf4expression in the AER.

Since ext2/dak, extl3/box and fgf10/dae all show a similar disruption of the AER during pectoral fin development, this raises the possibility that HSPG synthesis by Ext2 and Extl3 is necessary for Fgf10 signalling. To test this hypothesis directly, we implanted Fgf10-soaked beads into both ext2/dak and extl3/box mutant limb buds. We compared the effect of this treatment to that of Fgf4 soaked beads, which have previously been shown to activate several target genes in ext2/dakmutants (Grandel et al.,2000). We find that implantation of Fgf10 soaked beads into either ext2/dak or extl3/box mutant limb buds fails to rescue expression of fgf8 (0 out of 5 dak; Fig. 10A), shh (0 out of 9 dak; Fig. 10B; 0 out of 3 box; Fig. 10E), eng1a (0 out of 4 dak; Fig. 10C) or wnt7a (0 out of 3 dak; Fig. 10D), whereas implantation of the same batch of beads into fgf10/dae mutant embryos does lead to activation of these markers(Fig. 8; data not shown). By contrast, implantation of Fgf4-soaked beads is able to rescue expression of all genes analysed: eng1a (3 out of 4 dak; Fig. 11A), wnt7a (4 out of 4 dak; Fig. 11B) and shh (5 out of 5 box; Fig. 11C) in ext2/dakor extl3/box mutant limb buds. Collectively, these results indicate that HSPG synthesis by Ext2 and Extl3 is required for Fgf10, but not for Fgf4,signalling.

In this study, we have shown that the daedalus mutation disrupts zebrafish fgf10 and that, in the absence of fgf10 activity,development of the pectoral fin bud is severely compromised. Our analysis suggests that some early steps in fin development appear to be intact in fgf10/dae mutants. Three observations support this conclusion: first,a pectoral fin bud with a small apical thickening is present in fgf10/dae mutants, which gives rise to proximal endoskeletal elements(Fig. 1). This is in contrast to the fgf24 mutant, in which the bud fails to grow out at all, and no endoskeletal elements are formed. Second, several early AER markers,including bmp2b and sp8(Fig. 3), are initially activated in fgf10/dae mutants, albeit at reduced levels, and are subsequently lost. Third, the expression of both ZPA markers and Fgf target genes, such as shh, pea3, erm1 and dusp6(Fig. 4), is initiated in the absence of Fgf10, and then lost later on indicating that initiation of Fgf signalling occurs normally in the absence of fgf10 activity.

As AER expression of sp8 and bmp2b is already reduced in fgf10 mutants at early stages, our results suggest that fgf10 contributes to initial AER induction, and is then uniquely required for AER maintenance. In agreement with this, the majority of AER expressed genes we examined (including sp9, dlx2a and wnt3l)were strongly reduced by 28 hpf. The early expression of shh and Fgf target genes in the zebrafish may thus be independent of the AER, and might instead be directed by Fgf24 in the mesenchyme. This would be in contrast to the situation in tetrapods, where shh activation depends on the AER from the very beginning.

The daedalus phenotype appears to be weaker than that of mouse Fgf10 mutants, in which the AER is never established, and ZPA marker genes such as shh fail to be activated(Min et al., 1998; Sekine et al., 1999). These observations suggest that the early and late roles played by Fgf10 in the mouse, i.e. initial AER induction, followed by subsequent AER maintenance, are regulated differently in the zebrafish: initial AER induction is directed by fgf10 plus a second gene, whereas AER maintenance depends entirely on fgf10. In this respect, fgf10 acts as a classical apical ectodermal maintenance factor (AEMF)(Zwilling et al., 1959; Ohuchi et al., 1997).

If fgf10 is not uniquely required for AER induction, what is the nature of the additional AER-inducing signal in zebrafish? An in silico search of the zebrafish genome did not reveal a second fgf10 orthologue(W.H.J.N., unpublished), suggesting that redundancy following duplication cannot explain this phenotype. Previous results indicate that fgf24may provide this function (Fischer et al.,2003). Zebrafish fgf24 mutants show no morphological sign of pectoral fin bud formation, have no apical thickening and fail to activate expression of any AER markers. The expression of fgf24 is present in the lateral plate mesoderm prior to that of fgf10, and fgf10expression is dependent on fgf24 activity(Fischer et al., 2003). Furthermore, activation of fgf24 expression is not dependent on fgf10 activity (Fig. 3). The complete absence of pectoral fin buds in fgf24mutants raises the issue of whether fin bud development is initiated at all in the absence of fgf24. The normal initial activation of tbx5and msxc expression in the lateral plate mesoderm fgf24mutants, however, indicates that these cells have been correctly specified as early limb bud cells, but that subsequent relay of the limb inducing signal to the ectoderm fails in the absence of fgf24(Fischer et al., 2003).

Intriguingly, no fgf24 orthologue is present in tetrapod genomes,although it can be found in sharks (Draper et al., 2003). As sharks diverged from the ancestors of teleosts before tetrapods, this suggests that fgf24 was initially present, but then lost during the evolution of land vertebrates. To allow for this loss,other Fgf genes must have taken over the function of fgf24 in tetrapods. These Fgf genes include fgf8 during posterior mesoderm development (Draper et al.,2003), and fgf10 during early limb development(Fischer et al., 2003) (this study).

Our data demonstrate that expression of eng1a and wnt7a,respective markers of the ventral and dorsal limb ectoderm, depend on Fgf10 signalling as they are lost in fgf10/dae mutants. However, as AER signalling is also abrogated in these mutants, it is not clear how direct this effect is. Our gain-of-function experiments show that Fgf4, which is expressed in the AER, can rescue both eng1a and wnt7a expression in the absence of fgf10/dae activity. Although we cannot exclude the possibility that Fgf10 also activates these genes directly, our results show that failure of Fgf4 signalling from the AER is sufficient to explain why eng1a and wnt7a are lost in fgf10 mutants. This implies that the effect of Fgf10 signalling on eng1a and wnt7a expression is mediated by activation of fgf4expression in the AER, which in turn signals to dorsal and ventral ectoderm. Fgf4 protein is unable to rescue fgf8 expression in the AER of fgf10/dae mutants, consistent with the proposal that mesenchymal Fgf10 signals to the overlying AER through Fgfr2b, whereas Fgf4 signals through Fgfr2c. Thus, our results confirm that Fgf4 is unable to replace Fgf10 signalling to the AER, and indicate that this response can only be triggered via Fgf10. It remains to be determined through which receptor Fgf4 activates wnt7a and eng1a expression in the ectoderm.

We have shown here that the failure of AER maintenance in fgf10/dae mutants is very similar to that previously described in ext2/dak mutants (Grandel et al.,2000). In addition, we have shown that extl3/box mutants have similar defects in AER maintenance, although their phenotype is weaker than that of the other two mutants. These results suggest that these three genes act in the same pathway required for AER maintenance. This proposal is further supported by the observation that genetic removal of one copy of fgf10/dae dramatically enhances the severity of the extl3/box phenotype. The weaker phenotype of extl3/boxmutants correlates well with the observation that extl3/box mutants have a weaker reduction of HS levels than ext2/dak mutants(Lee et al., 2004). As Ext2 and Extl3 are required for the polymerisation of HS side chains on HSPGs,these results suggest that HSPG synthesised by Ext2 and Extl3 is required for Fgf10 function. Our gain-of-function data provide direct evidence for this possibility, as application of Fgf10 protein is able to rescue target gene expression in fgf10/dae mutants, but not in ext2/dak or extl3/box mutants. Furthermore, transplantation of wild-type cells into the epidermis of ext2/dak fin buds has been shown to enable a local rescue of AER development (Grandel et al., 2000). Taken together, these results suggest that HSPG synthesis by Ext2 and Extl3 in the fin bud ectoderm is required for ectodermal cells to respond to Fgf10 protein secreted by the underlying mesenchyme.

The pectoral fin phenotype of ext2/dak and extl3/boxmutants strongly resembles that of fgf10/dae mutants, but not that of fgf24 mutants. Therefore, HSPGs appear to be differentially required for Fgf10 signalling during limb development. In direct support of this proposal, we find that both ext2/dak and extl3/box mutants are able to respond to application of Fgf4 protein, but are unable to respond to Fgf10 protein. This agrees well with the observation that distinct Fgfs differ in the amino acid composition of their heparin-binding residues(Bellosta et al., 2001). Taken together, these results indicate that distinct Fgfs have different requirements for HS in vivo.

Interestingly, both Fgf4 and Fgf24 belong to the subgroup of Fgf ligands with preference for Fgfr2c, whereas Fgf10 belongs to the subgroup with preference for Fgfr2b (Orr-Urtreger et al., 1993; Ornitz et al.,1996; Fischer et al.,2003), raising the possibility that these classes of Fgfs may have differential requirements for HS in vivo. Alternatively, the different receptor subtypes might determine the role played by HS during receptor binding and activation. Arguing against this hypothesis is the observation that conditional removal of Ext1 activity from the mouse CNS results in several phenotypes that may be caused by abrogated Fgf8 signalling(Inatani et al., 2003), as Fgf8 also signals preferentially through Fgfr2c. Another possibility could be that Fgf signalling to the mesenchyme might be much less HSPG dependent than signalling to the ectoderm. This is unlikely, however, given that Fgf4 can activate ectodermal eng1a and wnt7a expression in ext2/dak mutants.

The Drosophila Exostosin genes have been shown to be crucial for Hh distribution and signalling during imaginal disc development (reviewed by Nybakken and Perrimon, 2002; Lin, 2004). Similarly, mouse mutations in Ext1 affect Ihh distribution and signalling(Lin et al., 2000; Koziel et al., 2004). It is therefore surprising that none of the phenotypes of ext2/dak and extl3/box can be linked to Hh signalling. During limb development,signalling by Shh is clearly not affected in these mutants, as rescue of Shh expression in ext2/dak mutant fin buds, by application of Fgf4 beads,leads to normal activation of the Hh dependent target genes hoxd11and hoxd13 (Grandel et al.,2000).

Although it is possible that some signalling factors do not require the presence of any HSPGs for their function, at least in some cellular contexts,an alternative possibility is that different factors require different levels of HSPGs in different contexts. As overall HS levels are strongly reduced, but not absent, in ext2/dak and extl3/box mutants(Lee et al., 2004), this might be an indication that signalling by Fgf10 requires much higher levels of HS than other signalling events during limb development.

Taken together, these results indicate that the effect of HSPGs on cell-cell signalling is both signal and context dependent. This provides an explanation of why the phenotypes of ext2/dak and extl3/boxmutants are so discrete and specific, even though both genes are broadly expressed (Lee et al., 2004),and their disruption causes a global reduction of HS levels. Because of this specificity in the control of developmental signalling by HSPGs, it will be critical to identify the exact signalling factors modulated by HSPGs in different organs and cell types in vivo, in order to better understand the biological relevance of this mechanism of signal regulation.

We thank Katherine Brown, Gillian Brunt, Filippo Del Bene, David Hipfner,Nadia Mercader, Maria Sakkou and Valia Stamataki for critically reading an earlier version of the manuscript, and members of the Neumann and Wittbrodt groups for stimulating discussions. The Wnt sequence alignments were assembled with help from Aidan Budd. wnt3l was a generous gift from Dr Arne Lekven. We thank Sabine Fischer and Gillian Brunt for excellent technical assistance and fish room care. J.L. is funded by the Swedish SSF through the EMBL Alumni Association.