Zebrafish foxi one modulates cellular responses to Fgf signaling required for the integrity of ear and jaw patterning

We are using the zebrafish to study the genetic basis of ear and jaw development. Development of the zebrafish ear begins as a thickening of the epidermal layer into a two-cell-thick placode at the nine-somite stage(Kimmel et al., 1995). The placode cavitates to form a lumen and hair cells begin to differentiate,projecting bristles into the liquid filled interior(Haddon and Lewis, 1996). Over the next few days, columns of tissue grow out from the walls of the otic vesicle, fusing in the center of the space to generate semicircular canals. Many genes have been shown either to be expressed in the developing zebrafish ear, or to be involved in the proper formation of the organ(Whitfield et al., 2002). pax2a and dlx3b are expressed in the developing placodes and roles in ear development for each of these genes have been demonstrated or inferred (Krauss et al., 1991; Lun and Brand, 1998; Solomon and Fritz, 2002). Fgf3 and Fgf8 have been implicated directly in the induction of the otic placode(Leger and Brand, 2002; Maroon et al., 2002; Phillips et al., 2001). Inhibition of either one of these signaling molecules causes marked reductions in the size of the placode, while inhibiting both causes an elimination of the ear primordia. pax8 is the earliest known molecular marker for cells that will become the otic placode, but its actual role in induction is not known (Pfeffer et al.,1998).

As in all vertebrates, the zebrafish ear serves the dual functions of linear acceleration detection and hearing(Bever and Fekete, 2002). Although functionally similar to the human ear, many structures are not present including the cochlea and middle ear. Remarkably, the middle ear structures first arose during the water-to-land transition and their origins are believed to be structures derived from the gills (spiracular pouch) and the jaw (hyomandibular bone) of fish(Webster et al., 1992).

The pharyngeal cartilages are derived from three streams of ventrolaterally migrating cranial neural crest (NC) cells. The pharyngeal skeleton shows both an anterior-to-posterior (AP) segmented polarity, with each segment deriving from an embryonic pharyngeal arch, as well as a dorsal-to-ventral (DV)polarity within individual AP segments(Trainor and Krumlauf, 2001). The first mandibular arch gives rise to a ventral Meckel's cartilage, which articulates with a dorsal palatoquadrate. The second hyoid arch is serially homologous to the first arch with a ventral ceratohyal that articulates via the interhyal with the hyosymplectic. The hyosymplectic is composed of a ventral symplectic rod region and a dorsal plate-like hyomandibular region. The arches and pouches form dynamically with the ectoderm forming an AP series of bilateral surface `in-pockets' complementary to bilateral endodermal out-pockets that together form the pharyngeal pouches separating adjacent arches along the AP axis. The migratory NC cells fill the spaces between the forming pouches adopting a cylindrical morphology encasing central cores of paraxial mesoderm (Kimmel et al.,2001).

Pharyngeal pouches likely signal to adjacent NC cells(Le Douarin and Ziller, 1993; Veitch et al., 1999) and essential roles for endothelin1 (edn1) in forming ventral cartilages has been shown in zebrafish and mice(Clouthier et al., 1998; Kurihara et al., 1994; Miller et al., 2000). Evidence for a dorsal cartilage-patterning signal is provided by the zebrafish fgf8 mutant, acerebellar (ace), which has a reduced hyomandibular region. However, early neural tube defects and absence of ace expression in the pouches makes interpretation difficult(Roehl and Nusslein-Volhard,2001). Studies on zebrafish casanova (cas)mutants and Fgf3 morpholino antisense oligonucleotide knockdowns suggest that loss of Fgf3 pouch expression results in early apoptotic elimination of posterior NC cells (Alexander et al.,1999; David et al.,2002). How an Fgf signal might function in DV patterning within an arch is still unclear.

In a large, insertional mutagenesis screen performed at MIT(Amsterdam et al., 1999; Golling et al., 2002), we isolated four insertional alleles of the zebrafish ortholog of the forkhead related transcription factor Foxi1 (FREAC6, FKHL10, HFH-3, Fkh10)(Avraham et al., 1995; Chen et al., 2002; Clevidence et al., 1993; Hulander et al., 1998; Larsson et al., 1995) that show specific defects in both ear and jaw development. The otic placodes are severely reduced in size, often split into two smaller placodes, and the semicircular canals fail to form properly, often resulting in a single large cavity instead of the normal three distinct chambers. The dorsal first arch derivative, the palatoquadrate, is mildly reduced in mutant animals, whereas the ventral Meckel's cartilage is indistinguishable from wild type. The dorsal second arch derivative, the hyomandibular region, is severely reduced in mutant animals while the ventral symplectic rod region and ceratohyal appear only mildly affected. Of the posterior arches, the third and fourth arches show substantial variable reductions while the fifth to seventh branchial arch cartilages appear relatively unaffected. Interestingly, targeted mutations of Foxi1 in mice show defects in ear development and homozygous mutant mice are born with both hearing and vestibular defects, as well as a lower survival rate after birth (Hulander et al.,1998), suggesting some conservation of the developmental pathway of the ears between zebrafish and mice.

Zebrafish were mutagenized, maintained and screened as previously described(Amsterdam et al., 1999). The retroviral insertion junction fragment cloning and animal genotyping procedures were as previously described(Golling et al., 2002; Sun and Hopkins, 2001). The primers used for genotyping carrier fish and embryos were:

• 1359A-link1 5′-CCGAGGACGCGCTTAAATTGGC-3′ and 1359A-link2 5′-GTCCACACACACACTGCATTTAAACATG-3′ in combination with nLTR3 5′-CTGTTCCATCTGTTCCTGAC-3′ for hi1321 and hi1359A;

• 1359A-link1 and 1321-0RT 5′-CTGATCCGTCCCGGTTGTATATGAGG-3′ in combination with RMN3′LTR 5′-GAGGAGACCCTCCCAAGGAACAG-3′ for hi3239;

• 1321-start 5′-CCATATGACCGTGTATTGCGACTCCAACTTC-3′ and 3239-gap2 5′-AGACTGGAAGTGTCCGCCAATTTAAGC-3′ in combination with RMN3′LTR for hi3747.

• Full-length cDNA for zebrafish foo was isolated by 5′-RACE and 3′-RACE as previously described(Golling et al., 2002). The primers used for 5′RACE amplification were 5′-GATGGTGCAGGTTCTGCTGGTACATGCTG-3′ followed by 5′-TCTGACCTGCTGAGTGTGCTCGTCTCTGAGG-3′. The primers used for 3′RACE amplification were 5′-GAGAAAAAGAAGAGCCGACGGAAAC-3′followed by 5′-GTCTGTGAAATCCGAGGACGCG-3′.

High-density filters to the CHORI-211 BAC library were hybridized with a radioactive probe made from the EcoRI-digested plasmid containing a partial foo cDNA. The positive BAC clones were identified by PCR amplifications and used for fingerprinting and automated sequencing.

For RT-PCR analysis of foo gene expression in wild-type and mutant animals, embryos were collected according to a previously described staging method (Kimmel et al., 1995)and RNA was prepared using Trizol reagent (ROCHE) with subsequent first-strand synthesis as previously described (Sun and Hopkins, 2001). The RT-PCR primers used for amplifying the sequence containing all foo proviral insertion sites were 1321-start and 1321-RT2 5′-CCTAAGATGACAAGAATGGGTGTTGG-3′. For RT-PCR amplification of the internal non-flanking sequence the primers used were 3239-gap1 5′-CGCTCAGTCAGATTTATCAGTACGTGG-3′ and 3239-gap2. The primers used for amplifying the b-actin control sequence were 5′-CATCAGCATGGCTTCTGCTCTGTATGG-3′ and 5′-GACTTGTCAGTGTACAGAGACACCCTG-3′.

The plasmid pCR2.1-foxi1 3′RACE was isolated by performing 3′RACE with the previously described primers. It was digested with NotI and antisense probe for in situ hybridization synthesized using T3 RNA polymerase as previously described(Sun and Hopkins, 2001). Previously described plasmids for probes were hand2(Angelo et al., 2000), dlx2 and dlx3b (Akimenko et al., 1994), gsc(Schulte-Merker et al., 1994), krox20 (egr2 – Zebrafish Information Network)(Oxtoby and Jowett, 1993), nkx2.3 (Lee et al.,1996) and pax2a(Krauss et al., 1991). Plasmid for making fgf3 probe was constructed by PCR amplification from 0-24 hpf cDNA using primers fgf3-f 5′-CTTGTTGTTACTGAGCTTCTTGGATCCGAG-3′and fgf3-r 5′-CCTCCAGATTTCAGTGTCAAACAATGCC-3′, followed by subcloning into pCRII-BLUNT to yield pCR-fgf3. To make antisense probe,pCR-fgf3 was digested with NotI and transcribed using SP6 RNA polymerase. Plasmid for making pax8 probe was constructed by PCR amplification from 0-24 hpf cDNA using primers pax8-f 5′-GCTTCCGGAGGTGATCCGGCAAAGG-3′ and pax8-r 5′-CTGGAGTTGGTGAATCTCCAGGCCTCG-3′ followed by subcloning into pCRII-BLUNT to yield pCR-pax8. To make antisense probe, pCR-pax8 was digested with BamHI and transcribed using T7 RNA polymerase.

Embryos were injected at the one- to four-cell stages using pulled glass needles and a picospritzer II (Parker Instrumentation) with the Fgf3-MO(Phillips et al., 2001). Calibration of injection needles indicated an average injection volume of 1 nl per 50 msecond pulse. Morpholino sequence was 5′-ACTCATGTTGACTACTCCTCCCACT-3′; four-base mismatch control was 5′-ACTCATCTTCACTACTGCTCCCAGT-3′.

Whole-mount in situ hybridization was performed as previously described(Thisse et al., 1994). Two-color reactions were performed as previously described(Hauptmann and Gerster, 1995; Sun and Hopkins, 2001).

Whole-mount immunohistochemistry was performed as previously described(Hanneman et al., 1988).

For TUNEL assays, fixed dehydrated embryos were rehydrated and acetone permeabilized for 10 minutes at –20°C, then rehydrated in PBST,washed in PBS + 0.25 mg/ml BSA and preincubated in TdT buffer for 2 hours at room temperature. TdT/fluorescein-dUTP reactions (Roche product 1-767-291)were incubated overnight at room temperature. Embryos were then washed twice for 45 minutes in PBST+1 mM EDTA at 65°C, 4 times for 45 minutes in PBST at room temperature, and processed with anti-fluorescein antibodies for alkaline phosphatase activity as previously described(Thisse et al., 1994).

Alcian Blue staining was performed as previously described(Golling et al., 2002).

We first identified foxi one (foo) as a mutant with a defect in ear development. The earliest morphologically detectable phenotype was a reduction in the otic placode size at around the 15-somite stage, with the placode appearing not only smaller, but split into two or more tiny placodes. These tiny placodes can later either fuse or remain split(Fig. 1B,D). When the placode was split into two smaller regions, an otolith typically formed in each otocyst (Fig. 1D).

Through established segregation analysis(Amsterdam et al., 1999), the mutant phenotype was linked to a single proviral integration for each of four independent isolates of foo (not shown). The genomic DNA adjacent to the proviral integration events was cloned, sequenced, and used to determine linkage of the insertions to the mutant phenotype(Fig. 1E) (see Golling et al., 2002) All alleles were crossed and failed to complement. Linkage data placed the mutation responsible for the foo/foo mutant phenotype within 0.9 cM of a proviral insertion. The sequences adjacent to all four proviruses show identity to a single zebrafish EST (Accession Number BI891904). We isolated several bacterial artificial chromosomes (BACs) and sequence from this genomic region showed significant homology at the DNA and protein level to the mouse forkhead related transcription factor Foxi1 (Fkh10) which has presumed orthologs in several species, including human (Fkhl10, FREAC6), rat (HFH-3)and frogs (Foxi1c). Among all species, including zebrafish, a genomic structure of two exons with a single intron (in the same location) is conserved (Fig. 1F). In one allele, hi3747, the mutagenic provirus lies in sequences encoding a poly-histidine stretch near the 5′ end of the gene and should result in a truncation before the conserved DNA-binding domain(Fig. 1F). We performed RT-PCR on mutant and wild-type embryos for each allele with two sets of primers: one set spanning the entire gene and located outside all four viral integration sites; and a second internal set that did not flank any of the viral integration sites (Fig. 1F). The longer amplicon was not successfully amplified from any of the alleles,although a significantly reduced amount of RNA was detected with the smaller amplicon in hi1321, hi3239 and hi3747. Almost no RNA was detected in the hi1359A allele. In situ hybridization of foo antisense RNA to mutant embryos was negative compared with wild-type siblings(Fig. 2D,E), suggesting a destabilized mRNA. The severity of the ear phenotype was variable within each mutant line, making comparisons of allele strength difficult. However, the allele with the most consistent phenotype was hi3747, the mutant with a proviral integration closest to the 5′ end of the gene and before the DNA binding domain. In hi3747, the otic vesicles were substantially reduced and often split (as in Fig. 1D).

Antisense oligonucleotides using morpholino chemistry have been shown to be effective, gene-specific, transcriptional inhibitors in zebrafish embryos(Ekker, 2000; Nasevicius and Ekker, 2000). As further confirmation that mutations in the foo gene are responsible for the observed phenotypes, we designed two morpholino antisense oligonucleotides to foo mRNA and injected them into embryos. In 50%of the embryos, the phenotype generated was indistinguishable from the mutant alleles (39/78) in ear or jaw defects, while the control morpholino showed no effect (data not shown).

By RT-PCR, foo transcripts were not detected in unfertilized oocytes but were readily detected by the sphere stage (data not shown). Location of foo expression by in situ hybridization was first detected at the dome stage (Fig. 2A,B). By the shield stage, expression was in the presumptive ventral ectoderm near the animal pole that is fate mapped to be placodal precursors (Fig. 2C)(Kozlowski et al., 1997; Woo et al., 1995). By 90%epiboly, foo expression was split into bilateral regions(Fig. 2D). As convergence continued, foo expression appeared to be restricted to a region destined to become the otic placode (Fig. 2F). Expression of foo precedes expression of pax8, making it the earliest known marker of otic placode formation.

Because of the early expression of foo in the developing otic placode, we examined the expression in our mutant of other genes known to be expressed early in the otic placode. Mutant embryos showed reduced expression of dlx3b and pax2a that, in addition, was often split into several small patches of expression (Fig. 3D,F,H, arrows). Expression of dlx3b and pax2ain other areas of the developing embryo were unaffected(Fig. 3D,F,H, arrowheads). The strongest effect we detected was on the expression of pax8. At the one-somite stage, no pax8 expression could be detected in the presumptive otic placode of one quarter of the embryos(Fig. 3B). However, we saw normal expression of pax8 in the mesoderm of presumptive kidney(Fig. 3B, arrowhead). This suggests that foo is upstream of pax8 in the otic vesicle and is required for proper expression of pax8 in this region.

dlx2 is normally expressed in all three streams of migratory NC cells beginning around the two- to three-somite stage. NC cells migrating out of rhombomere 5 (r5) normally move ventrally in a posteriolateral arc around the otic placode (Fig. 4A,arrow). Because the foo/foo mutant embryos displayed a significant disruption in otic placode development, we examined markers of migratory neural crest cells to determine whether disrupting the otic placode might affect normal NC cell migration. In foo/foo embryos the intensity of dlx2 expression was normal but the third stream NC cells appeared slightly more anterior and no longer distinct from the second stream(Fig. 4B, arrowhead). krox20 is expressed only in third stream migratory NC cells exiting r5. In foo/foo embryos, these NC cells migrated ventrolaterally, but failed to make the posterior movements around the otic placode(Fig. 4, compare C with D,arrowheads). Many cells were observed migrating through the otic placode region. Notably, the anterior margin of the post-migrating third stream NC cells has only shifted minorly towards the anterior(Fig. 4E,F arrows). In addition, the migration of second stream NC cells was indistinguishable from wild type, as visualized by ephrin B2a (efnb2a) expression(data not shown). It is unclear whether these changes are the result of a reduced otic placode or loss of foo expression. Cell transplantations will be required to answer this question.

By 5dpf, foo/foo embryos display significant defects in the cartilaginous structures of the developing jaw(Fig. 5A,B), observable from both lateral and ventral views (Fig. 5C-F). Most notably in all hi3747 embryos, there is a severe reduction of the hyomandibular region (Fig. 5C,D, labeled hm). As it is often difficult to see this reduction,we flat-mounted the jaw cartilage to illustrate better the loss of the hyomandibular (Fig. 5G,H). We also saw variable reductions in the ceratohyals (ch) (strong example of reduction in Fig. 5F), and third and fourth gill arches (Fig. 5, numbers). Ceratohyal inversion is a common phenotype seen in back arch mutants. Therefore, inversion of the ceratohyals in foo/foo embryos probably results from the reduced third and fourth arches. There is a subtle, but consistent reduction in the overall measured size of the palatoquadrate (PQ)(Fig. 5I), although there is no apparent reduction in the more ventral Meckel's cartilage (M)(Fig. 5I). The embryos die after 6-7 days from unknown causes that are presumed to be related to the phenotypes either observed or yet to be determined. Injection of high concentration foo antisense morpholinos generated a morphant phenocopy of the jaw defects (data not shown).

To begin to investigate the roles foo plays during jaw development, we examined expression of foo at later developmental time points. By the 10 somites stage, expression in the otic placode has faded and a second region of expression more ventral and lateral is apparent(Fig. 6A,B). Double staining of 10-somite stage embryos with krox20 and foo shows that a stream of ventroanteriorly migrating foo-positive NC cells can be detected that are probably second stream associated NC cells. These foo-positive cells appear to initiate expression of foo as they approach their ventrolateral destinations, and suggest that foois expressed in at least a subset of the post-migratory second stream NC(Fig. 6E, arrows). Initiation of foo expression was likewise detected among post-migratory third stream NC cells. However, second arch associated NC cell expression of foo is significantly downregulated by the 15-somite stage and no longer detected by 24 hpf (Fig. 6B,D). By the 15-somite stage, foo expression is readily detected in the region of what will be endodermal precursor cells of the first and second pharyngeal pouches (pp1, pp2 in Fig. 6B,C). fooexpression remains strong in this region at least until 32 hpf, with downregulation proceeding in a medial-to-lateral and anterior-to-posterior fashion. Transverse sections through pp1 verified endodermal expression(Fig. 6F).

As the cartilages are derived from NC cells, we examined the expression patterns of early NC cell markers. dlx2 is expressed in both migrating NC cells and post-migratory arch-associated NC cells. At 28 hpf, we detected a substantial reduction of dlx2 expression in foo/foo embryos in the NC cells that populate the third and fourth arches, both of which are derived from third stream NC cells(Fig. 7A,B). Analysis of dlx3b expression, which is also expressed in post-migratory NC cells,revealed reduction of expression in the third and fourth arches in foo/foo embryos (data not shown). The endothelin 1 signaling pathway is essential for the proper formation of ventral cartilages and at least partly functions through the induction of the transcription factor hand2 in ventral NC cells (Miller et al., 2000). The ventral expression of hand2 appeared largely unaffected in the first and second arches consistent with the near normal formation of ventral cartilage elements in the anterior arches, but,mirroring the dlx2/3 defects, expression of hand2 was markedly reduced in the third and fourth arches of foo/foo embryos(Fig. 7C,D). Taken together,altered expression in foo/foo embryos of these three markers indicate a severe disruption in the proper patterning of the third and fourth arches.

The hyomandibular region of the hyosymplectic is reduced in foo/foo embryos. This element is derived from dorsal second arch NC cells. Because the dlx2 and dlx3b data indicate a reduction in dorsal second arch NC cell expression, the dorsally and ventrally second arch restricted transcription factor gsc was examined for potential dorsal-specific defects (Schulte-Merker et al., 1994). gsc expression in wild-type embryos shows distinct dorsal NC cell expression in the first and second arches flanking either side of pp1 as well as a distinct second arch ventral expression domain(Miller et al., 2000). Outside the arches, gsc is also expressed within the developing otic vesicle. Strikingly, the dorsal first and second arch gsc expression domains were not detected in foo/foo embryos(Fig. 7E,F, arrows). Expression in the otic vesicles was likewise not detected in foo/foo embryos. Conversely, the olfactory and ventral second arch gsc expression domains appeared unaffected in mutants. Consistently, the homeobox transcription factor hoxa2, which is uniformly expressed in the second arch NC cells, also showed a dorsally restricted second arch NC cell defect (data not shown). Thus, defects in markers for the dorsal, but not ventral, NC cells of the first and second arches precedes the hyomandibular defect.

To determine whether the loss of expression of NC cell markers and absence of corresponding cartilages was the result of NC cell apoptosis, we performed terminal deoxynucleotidyl transferase fluorescein-dUTP nick end labeling(TUNEL) assays to visualize apoptotic cells in developing embryos. When compared with wild-type embryos, foo/foo embryos display a substantial, transient increase in apoptosis in the pharyngeal arch region(Fig. 8A-G). The transient wave of cell death peaks around 26 hpf at ∼20-fold over the intensity observed in wild-type embryos, and is then reduced by 28 hpf(Fig. 8H). Notably, no increased apoptosis was observed at 22 hpf or at earlier time points during otic development (Fig. 8H; data not shown). Based on their sub-ectodermal location in transverse sections, at least some of these TUNEL-positive cells appear to be post-migratory NC cells(Fig. 8E-G, sections located as approximately indicated in 8D). However, we cannot rule out that some of these TUNEL-positive cells might be endodermal pouch cells.

Fibroblast growth factors (Fgfs) are known to have mitogenic activity and fgf3 is known to be expressed in the posterior half of pharyngeal pouches (David et al., 2002). Therefore, we examined the dynamic expression of fgf3 in the pharyngeal pouches. By 15 somites to at least 34 hpf of development, fgf3 was expressed in the endoderm two pouches at a time with downregulation in the most anterior pouch coincident with up regulation in the next most posterior pouch on a roughly four hour cycle(Fig. 8I,K,M,O, arrowheads;data not shown). At 16- and 20-somite stages, expression of fgf3 was not significantly different between wild-type and mutant animals indicating that initiation of fgf3 expression was not compromised, and that fgf3-positive cells are at least initially present in mutant animals(Fig. 8I,J). However, by the 24-somite stage, foo/foo embryos showed a premature loss of fgf3 expression from pp1, the anterior-most pharyngeal pouch(Fig. 8L, red arrowhead). Examination of later developmental time points also shows premature loss of fgf3 expression from the anterior-most pouch(Fig. 8N,P, blue then first black arrowhead). Thus, although initiation of fgf3 pouch expression is always observed, the maintenance of fgf3 pouch expression appears compromised. Expression of fgf3 in other regions of the embryo, such as the isthmic region, was unaffected, which indicates the specificity of this defect.

As morpholinos that inhibit fgf3 are known to cause posterior pharyngeal arch defects similar to those observed in foo/foo embryos,we analyzed fgf3 morpholino injected animals for hyomandibular defects (David et al., 2002). Consistent with personal communications (L. Maves and C. B. Kimmel,unpublished) we observed not only the posterior gill arch defects but also a reduction of the hyosymplectic (Fig. 8Q,R, arrows). Out of 39 fgf3 morpholino-injected animals selected at 30 hpf for a reduced otic vesicle phenotype and raised to 4 days of age, 10 showed all cartilages severely reduced (not shown), 16 displayed the strong phenotype (Fig. 8R)and 13 showed the mild phenotype (Fig. 8Q). We also analyzed fgf3 morpholino injected animals for apoptosis by TUNEL assay and observed substantially increased levels of apoptosis in morphant animals compared with controls. However, extensive cell death in fgf3 morpholino injected animals made conclusive interpretations of the TUNEL assays regarding specific NC cell apoptosis effects difficult (data not shown).

To determine the extent of the pharyngeal pouch patterning defect, we examined the staining pattern of zn-8, a monoclonal antibody that recognizes the pouch ectoderm, and we examined the expression pattern of the transcription factor nkx2.3, a later marker for posterior pouch endoderm. At 34 hpf, the zn-8 antibody effectively labeled pp1-pp4 in the wild-type animal but foo/foo embryos displayed significant reductions in pp2-pp4 (Fig. 9A,B). Although slightly reduced, pp1 was labeled by zn-8 in foo/foo embryos indicating that complete patterning of pp1 may depend upon additional factors,reflecting the higher complexity of the cartilage structures developing adjacent to this pouch. Consistent with this interpretation, ephrin B2a expression in pp1 was unaffected in foo/foo embryos (data not shown). The disrupted expression of nkx2.3 from pp2-pp5 provides further indication that posterior pouch patterning requires foo(Fig. 9C,D).

Expression of pax8 has been reported in the hyoid arch in mice,but has not been reported in the zebrafish hyoid arch(Pfeffer et al., 1998). To determine whether the foo-pax8 pathway is present in the pharyngeal pouches we examined pax8 expression in foo/foo embryos. Although we were unable to detect expression of pax8 in pp1 or pp2 at the 15-somite stage (data not shown), there was clear expression of pax8 in the posterior part of pp2 at the 21-somite stage and that expression domain was absent in foo/foo embryos(Fig. 9E,F, blue arrowhead). Likewise, at 30 hpf, pax8 expression was observed in more posterior pouches of normal sibling embryos but absent in foo/foo embryos (data not shown). This apparent AP progression of pax8 pouch expression was reminiscent of both the foo and fgf3 expression profiles. Interestingly, at 34 hpf, pax8 expression in the thyroid precursor cells was detected in foo/foo embryos indicating that not all pharyngeal endoderm expression of pax8 is dependent on fooactivity (data not shown).

We have demonstrated that embryos mutated in foo show a severe defect in the formation of the otic placode. foo gene expression begins early with mRNA detectable by the sphere stage in a restricted ventral region near the animal pole. By fate mapping, this region corresponds to presumptive placodal tissue (Kozlowski et al., 1997; Woo et al.,1995). By the tail bud stage, foo expression was localized in two distinct bilateral regions that mark the future otic placodes. Prior to this study, pax8 was the earliest known marker of the otic placodes. Remarkably, not only does foo expression in presumptive placodal tissue precede pax8 expression, but foo/foo embryos also show a complete loss of pax8 expression in the otic placode as well as reduction and disorganization of later otic placode markers. This places pax8 downstream of foo, which makes it the earliest known marker of otic placode development and gives foo a demonstrated role in the induction of a specific otic vesicle marker.

At tail bud stage, dlx3b is normally expressed in all placodal tissue forming a discrete lateral band of expression that encircles the anterior neural plate with higher expression over the future otic placode(Akimenko et al., 1994). By the 10-somite stage, dlx3b is normally strongly expressed in the otic placode, although its expression is no longer detected laterally between the otic placode and the anteriormost domain. Although 10 somites stage foo/foo embryos show the reduced otic placode expression of dlx3b, the mutant embryos apparently also retain from the tail bud stage some of the lateral band of dlx3b expression surrounding the anterior neural plate (see Fig. 3), suggesting that foo may function in restricting or maintaining the coherence of the expression domains of placodal markers. This failure to tightly restrict the expression domains of otic placode markers such as dlx3b and pax2a might explain why foo/fooembryos often display split and/or duplicated otic vesicles at pharyngula and later stages. Thus, this Foxi1-dependent pathway has an essential function for the developing otic placode but not a significant role in initial induction.

Three recent papers show that induction of the otic placode requires both fgf3 and fgf8 (Leger and Brand, 2002; Maroon et al.,2002; Phillips et al.,2001). The combined data from the three papers suggest that pax8 expression is induced by fgf3 signaling, but the amount of signal required to induce pax8 is much lower than other otic placode markers.

Removing fgf3/8 signals completely eliminates otic induction, but foo/foo embryos still show an induction of several markers. Although other models are possible, the simplest model places foo activity between fgf3/8 and pax8 in the induction. Because of the observed phenotype, we propose that fgf3/8 signaling to the otic placode induces parallel pathways. The Foxi1-independent pathway begins very early and would be responsible for the initial induction and expansion of the otic placode. As foo/foo embryos completely lacked pax8expression in the otic placode, yet were still able to induce placodal structures that ultimately formed an ear, our data suggests that in zebrafish, pax8 and foo function are not required for otic placode induction. There is early pax8 expression in mouse otic placodes(Pfeffer et al., 1998), and a mouse line with a targeted deletion of pax8 has been made, but there no data have yet been reported regarding ear defects in these mice(Mansouri et al., 1998).

The second, Foxi1-dependent pathway has at least the one downstream effect:that of inducing pax8 expression. Based on data obtained by treating embryos with the Fgf receptor inhibitor SU5402(Leger and Brand, 2002),initiation of this pax8 induction would take place sometime after 70%epiboly, as SU5402 added at 70% epiboly can block pax8 expression. The role of the earlier initiating Fgf signal would be to provide an activating signal for foo-expressing cells to induce pax8and perhaps other unidentified downstream genes. It is also important to note that foo expression is likely to be independent of fgf3/8signaling even though pax8 expression is not.

We suggest that foo/foo mutant embryos demonstrate an uncoupling of otic placode induction from its subsequent morphogenesis(Fig. 10). This organizational or `integrity-maintenance' activity would hold the placode together with the predicted effect of this model being similar to what we actually see in the foo mutants, that is, smaller dissociated placodal precursors. Because the regions of pax2a and dlx3b appear to be somewhat reduced, we also propose that foo-expressing cells generate positive feedback to the Fgf signals (Fig. 10). This is an idea we will return to more strongly in the discussion of the role of foo in the developing jaw.

We have demonstrated that foo/foo embryos display severe cartilage defects and shown how analysis of mutants provides mechanistic insight into how foo activity can differentially affect dorsal and ventral cartilage shapes through affecting the maintenance of fgf3expression. By the 10-somite stage, foo/foo embryos displayed a disorganized third stream of NC cells that invaded the otic placode region. As the otic placode is thought to serve as an inhibitory guidance cue(Sechrist et al., 1993; Smith et al., 1997), the observed third stream migration defect is perhaps secondary to the previously described otic placode defects. Without the maintained integrity of tight cell-cell interactions, the otic placode may no longer be capable of deflecting the ventral movement of third stream NC cells. The extent of the third stream migration defect appeared limited to invasion of otic placode territory because no krox20-expressing third stream cells were detected in post-migratory second stream territory(Fig. 4D, dorsal view, the area lateral to r2-r4). It is important to emphasize that a significant fraction of the third stream NC cells do not undergo apoptosis because, although reduced,posterior cartilages are readily detected in foo/foo embryos. Alternatively, these migration defects could be intrinsic to NC cells lacking Foo activity, a question that will require cell transplantations and cell autonomy studies for a clear answer.

Because first and second stream NC cell migration appeared normal in foo/foo embryos, the hyomandibular defects probably arose from defects at later stages of jaw development. By the 15-somite stage, foo expression was readily observed in the endoderm of the pharyngeal pouches. At around 20 somites stage, foo/foo embryos showed a loss of pax8 expression from the pouches followed shortly at the 24 somites stage by a premature loss of fgf3 expression. The combined action of strong Fgf signaling and foo activity would prefigure the pouch pax8 expression domains, as fgf3 is expressed in only the posterior half of pp2 as is pax8. Consistent with this observation, pax8 expression in the otic placode is strongly dependent on both Fgf signaling and foo activity. Thus, in both the ear and the jaw, we propose that foo expression is responsible for the modification of downstream cellular responses to Fgf signaling leading to the induction of pax8 expression.

Interestingly, loss of pax8 expression in foo/foo embryos precedes the loss of fgf3 maintenance in pp2. It is possible that foo and pax8 function to maintain pouch endoderm integrity of markers such as nkx2.3 in a manner analogous to otic placode integrity (Fig. 10). A future direction will be to determine whether feedback maintenance of Fgf signaling depends on pax8 function. fgf3 is expressed in the pharyngeal pouches in a roughly AP progressive wave. From 15 somites until around 34 hpf, fgf3 expression is maintained in two pouches at a time with attenuation of the anterior-most pouch coinciding with the up regulation of the next posterior-most pouch. In foo/foo embryos, initial induction of fgf3 expression in any particular pharyngeal pouch appeared to be only modestly affected. However, the onset of signal attenuation appeared prematurely, which is consistent with a role for foo in the maintenance of fgf3 pouch expression. Thus, foo expression is not required for initial induction of fgf3in pouches but instead is part of a transient positive feedback mechanism that maintains fgf3 signaling. It is possible that this maintenance is either at the transcriptional level causing cells to continue to express fgf3,or at the survival level, providing positive feedback signals that prevent fgf3 cells from apoptosis. Further experiments will be required to determine which of these is true.

At 26 hpf, a 20-fold increase in NC cell apoptosis was observed in the arches. This transient wave of cell death is probably a direct result of the lost fgf3 maintenance because both cas mutants, which lack all endoderm (including fgf3 pouch expression), and SU5402-treated embryos, in which Fgf receptor signaling is blocked, display extensive NC cell apoptosis. The apoptosis in foo mutant embryos is not as severe as these other cases because some residual fgf3 signaling is still present. Consistently, fgf3-MO animals appear to display jaw defects that are very similar to and slightly more severe than those observed in foo/foo embryos (Figs 5, 8). Therefore, the failure to maintain fgf3 expression in foo/foo embryos is what causes the widespread but transient NC cell apoptosis, causing reductions in the dorsal structures of both the mandibular and hyoid arches, and variable reductions in the more posterior branchial arches.

The apoptotic death of NC cells probably explains the later absence of the dorsal first and second arch gsc expression domains as well as the loss of posterior arch defects in dlx2, dlx3b and hand2expression. Notably, dlx2 expression is likewise lost in posterior arches of cas/cas and fgf3-MO animals(David et al., 2002). The unperturbed patterning of the ventral NC cells is indicated by the normal expression of dlx2, dlx3b, hand2, gsc and hoxa2 gene expression in ventral NC cells as well as the equivalence in size of Meckel's cartilage between foo/foo and wild-type embryos. Interestingly,ventral NC cells of both the mandibular and hyoid arches do not require an fgf3-mediated signal for survival and/or proliferation, yet are absent in SU5402-treated embryos, implicating an essential role for a different as yet unidentified Fgf ligand. Similarly, it is worth noting that in spite of the fgf3 maintenance defects, which extend into the more posterior pouches such as pp3 to pp5, that the most posterior fifth to seventh arches appear unaffected in foo/foo embryos. A possible explanation for the normal patterning of these posterior arches is an undetermined Fgf signal that can compensate for the loss of fgf3. Consistent with this explanation, fgf3-MO animals often show normal cartilages from the fifth to seventh arches that are missing in SU5402-treated embryos(David et al., 2002).

The complexity of the pharyngeal endoderm patterning pathways is further highlighted by the dual observations of lost pp2 pax8 expression around the 20 somites stage (Fig. 9E,F) and retained thyroid follicular precursor cell pax8expression in 34 hpf foo/foo embryos (data not shown). As both tissues are pharyngeal endoderm derived, alternate pathways must exist for inducing pax8 expression not only in tissues such as the eyes, MHB and pronephric ducts but also in more medial regions of the pharyngeal endoderm. It is possible or likely that foo expression in the pharyngeal pouches has other important functions that will require further experimentation to establish.

There are several noteworthy implications for the observed failure to maintain fgf3 expression on possible mechanisms of establishing DV or AP polarity within an arch. First, it suggests that the relative size of a particular cartilage element can be directly influenced by the duration of a particular Fgf signal; the shorter the Fgf pulse, the fewer cells survive/proliferate and the smaller the structure. It is therefore possible that either a different Fgf signal might mediate survival of a different population of cartilage precursors or that a short but intense pulse of an Fgf signal could facilitate survival near the point source but with little effect at a distance. Second, it suggests that regional functionality of Fgf signaling could direct cartilage shape. The patterned induction of a variety of Fgf antagonist molecules, such as sef or sprouty family members, could direct cartilage shape by eliminating specific Fgf signals from specific NC cells. In support of this, sprouty4 has been shown to be expressed in the pouches (Furthauer et al., 2001). The combined action of multiple patterned Fgf agonists and multiple antagonists could therefore generate fairly complex DV, AP and ML matrixes of survival, proliferation and death cues. Furthermore, we suggest small changes in this complex interplay of signals could be one of the mechanisms for the evolutionary adaptation of jaw structures.

Our results show that in zebrafish, foo is a very early regulator of both ear and jaw development and is upstream of pax8 induction in both structures. pax2, pax5 and pax8 are believed to have arisen from a single ancestral pax2/5/8 gene(Pfeffer et al., 1998). Two recent publications have identified a single pax2/5/8 gene in ascidians and amphioxus that is expressed in the region of the primordial pharynx, a region analogous to the pharyngeal arches of vertebrates(Kozmik et al., 1999; Wada et al., 1998). Given the observed commonality of gene expression between the otic vesicle and the arches, we argue for a synthesis between two interpretations of the pax2/5/8 expression data in these primitive organisms(Kozmik et al., 1999; Wada et al., 1998). The ascidian cupulae could be considered gill slits that developed a placode-like function that differentiates into the ciliated primary sensory cells involved in feeding. The functional and spatial separation of these sensory regions away from the underlying arches as chordates evolved might explain why certain molecular markers that characterized the original gill slit ectoderm are maintained in vertebrate placode-derived structures. Because at least one vertebrate, the zebrafish, co-expresses foo and pax8 in the otic placode ectoderm and also in the pouch endoderm, one might imagine that the primordial gill slit would express foo and pax8 in both ectoderm and endoderm layers. The ability to perform gene knockdown studies in ascidians will allow for direct testing of conservation of gene regulatory networks (Heasman, 2002; Satou et al., 2001).

Even if the jaw and ear do not have a common origin, sharing gene regulatory networks helps explain how, across the water-to-land transition,the hyomandibular might transform into the stapes(Webster et al., 1992). If the hyoid NC cells lost purpose in the jaw but were still responsive to many of the same signaling molecules, such as Fgfs, then it is possible that the presumptive hyoid NC cells began responding to signals relating to the otic placode instead of the forming jaw. Without the similarities in gene expression, the mechanism for this jaw to middle ear NC cell shift would have to have happened in a more random and uncontrolled way. To address this issue directly is challenging; however, a simple, testable prediction of this jaw-ear crossover theory is that Foxi1^–/– mice should display middle ear bone defects, perhaps specific to the stapes.

We thank Kirsten Sadler for critical reading of the manuscript; J. G. Crump, L. Maves and C. B. Kimmel for helpful discussions; B. Riley for fgf3-MO; and S. Farrington and the technical staff of the Hopkins laboratory for support. We also thank the Developmental Studies Hybridoma Bank(maintained by the Department of Biological Sciences, The University of Iowa)for the zn-8 antibody, and A. M. Caron of the MIT CCR Core Histology Facility for technical assistance on sections. R.M.N. was supported by a post-doctoral fellowship from the NIH. This work was also supported by grants from Amgen and NIH, and the intramural program of the NHGRI.