Pax8 and Pax2a function synergistically in otic specification, downstream of the Foxi1 and Dlx3b transcription factors

The vertebrate inner ear is the sensory organ that provides the auditory and vestibular functions responsible for hearing and balance. It develops from a transient embryonic structure, the otic placode, a thickening of head ectoderm adjacent to the developing hindbrain. Through interactions with adjacent tissues, the otic placode develops into the otic vesicle that subsequently forms epithelial and neuronal cells of the inner ear(Noden and van de Water, 1986; Couly et al., 1993; Fritzsch et al., 1997; Barald and Kelley, 2004).

Studies in various species suggest that signals from the underlying mesoderm and adjacent hindbrain induce ectodermal cells to form the otic placode (reviewed by Fritzsch et al.,1997; Torres and Giráldez, 1998; Baker and Bronner-Fraser, 2001; Whitfield et al., 2002). In zebrafish, Fgf3 and Fgf8 appear to have overlapping functions in otic placode induction. The genes are expressed in the future hindbrain by late gastrula stages, and fgf3 is also expressed at this stage in the underlying mesendoderm. Loss of either fgf3 or fgf8 leads to a reduction in ear size and loss of both fgf3 and fgf8together results in near or total ablation of otic tissue(Phillips et al., 2001; Maroon et al., 2002; Leger and Brand, 2002). Furthermore, Fgf signaling is sufficient and necessary for otic induction,indicating a direct role for Fgf3 and Fgf8(Phillips et al., 2004). In the mouse, Fgf3 and Fgf10 act as redundant signals during otic induction(Wright and Mansour, 2003; Alvarez et al., 2003). We have previously shown that Fgf signals are required for the preotic expression of some, but not all, of the transcription factors involved in otic induction(Liu et al., 2003). For example, the four transcription factors Dlx3b(Ekker et al., 1992), Dlx4b(Stock et al., 1996; Ellies et al., 1997), Sox9a(Chiang et al., 2001; Yan et al., 2002) and Sox9b(Chiang et al., 2001; Li et al., 2002) are all required for otic placode specification. sox9a and to some extent sox9b require Fgf signaling for their proper expression in the preotic region, whereas dlx3b and dlx4b do not. When Fgf3 and Fgf8 functions are removed, dlx3b and dlx4b gene expression is unaffected during induction and early patterning stages. Later expression of dlx3b and dlx4b in the otic anlagen is reduced when Fgf signals are blocked, but this is probably an indirect effect caused by the loss of sox9a function(Liu et al., 2003).

Fate-mapping experiments at mid-gastrula stages indicate that precursors of cranial placodes are arranged in an anteroposterior order at the lateral border of the prospective anterior neural plate(Kozlowski et al., 1997). We have previously shown that dlx3b and dlx4b are both expressed in late gastrula stage embryos in this same region, a stripe corresponding to cells of the future neural plate border; expression of both genes becomes restricted to cells of the future olfactory and otic placodes by the beginning of somitogenesis (Akimenko et al., 1994; Ekker et al.,1992; Ellies et al.,1997). We also showed by fate mapping that a subset of cells in the anterior part of the dlx3b stripe later contribute to the olfactory placodes (Whitlock and Westerfield, 2000). Knockdown of dlx3b and dlx4bcauses a severe loss of otic tissue even in the presence of functional Fgf signaling (Solomon and Fritz,2002; Liu et al.,2003), indicating that these genes are required to specify the competence of cells to form the ear. Expression of the forkhead class winged helix transcription factor, foxi1, is progressively restricted at late gastrula stages to bilateral domains, including the presumptive otic placode, and, subsequently, foxi1 expression is downregulated prior to placode formation. Disruption of foxi1 leads to severe defects in otic placode formation and highly variable ear phenotypes(Solomon et al., 2003; Nissen et al., 2003),suggesting that foxi1 also influences otic competence.

In addition to these genes implicated in otic development, some Pax genes that encode paired box transcription factors are expressed at the right time and place to be involved in otic specification. The Pax gene family is subdivided into four distinct classes based on sequence similarities(Noll, 1993; Mansouri et al., 1996). The Pax2, Pax5 and Pax8 genes constitute one such class and encode highly related transcription factors with similar biochemical activities (Pfeffer et al.,1998). Pax2-Pax5-Pax8 genes have important roles in embryonic development and organogenesis of the eye, ear, kidney and thyroid(Dressler et al., 1990; Nornes et al., 1990; Plachov et al., 1990). The best-studied Pax2-Pax5-Pax8 gene function is in development of the midbrain-hindbrain boundary (isthmus). In the mouse, Pax2 or Pax2 and Pax5, depending on the genetic strain, tops a hierarchy of genes that function together to form the isthmus. Pax2 and Pax5 act at multiple stages in this process and are also required for maintenance of Pax2 (Urbanek et al.,1994; Torres et al.,1996; Mansouri et al.,1998). Implantation of Fgf8-soaked beads into chick embryos showed further that Fgf8 acts in this positive feedback loop maintaining Pax2 expression in the isthmus(Martinez et al., 1999). In zebrafish, pax2a has been shown to function in this process; loss of pax2a leads to failed formation of the isthmus(Brand et al., 1996; Lun and Brand, 1998). Similar to amniote embryos, maintenance but not induction of zebrafish pax2adepends on both pax2a and fgf8(Lun and Brand, 1998; Reifers et al., 1998). Gene replacement in the mouse has shown that Pax5 can functionally substitute for Pax2, indicating that the Pax2-Pax5-Pax8 proteins are interchangeable (Bouchard et al.,2000). Combined, redundant gene function has also been shown for Pax2 and Pax8 during development of the mouse urogenital system (Bouchard et al.,2002).

The relative roles of Pax2 and Pax8 in otic specification are still somewhat unclear. Pax8 is one of the earliest known markers of otic cells in vertebrates, showing onset of otic expression before Pax2(Pfeffer et al., 1998; Heller and Brandli, 1999; Hutson et al., 1999; Groves and Bronner-Frasier,2000). In mouse, loss of Pax8 does not prevent expression of Pax2 or proper inner ear formation(Mansouri et al., 1998) and loss of Pax2 has no effect on otic induction but variably affects formation of the cochlea (Torres et al.,1996). Zebrafish have two Pax2 genes, pax2a and pax2b, both expressed in the preotic region with pax2aexpressed at higher levels several hours earlier than pax2b(Pfeffer et al., 1998). Loss of pax2a, pax2b or both alters hair cell development but does not hinder otic placode induction (Riley et al., 1999; Whitfield et al.,2002). Cells of the zebrafish otic placode also express pax8 (Pfeffer et al.,1998); functional studies have not previously been described.

We show that Pax8-depleted zebrafish embryos, like pax2a^– mutants, have only mild ear defects. By contrast, Pax8 depleted pax2a^– mutants fail to form a differentiated otic vesicle or inner ear, although a few residual cells express genes characteristic of otic fate. These data demonstrate that Pax2a and Pax8 have similar functions required for the correct expression of sox9a, sox9b and pax2a, and to a much lesser extent, dlx3b. However, pax8 and pax2a are regulated by two independent factors, Foxi1 and Dlx3b, respectively; removal of both factors is required to block the establishment of otic fate. Our results integrate pax2a and pax8 gene functions into the previously known genetic pathways that regulate otic placode induction; Foxi1 and Pax8 mediate early Fgf dependent otic specification, whereas Dlx3b and Pax2a mediate later Fgf signaling required for maintained development.

Embryos were obtained from the University of Oregon zebrafish facility,produced using standard procedures(Westerfield, 2000) and staged according to standard criteria (Kimmel et al., 1995) or by hours post fertilization at 28°C (h). The wild-type line used was AB. The lines, no isthmus^tu29a, a null allele of pax2a, and acerebellar^ti282a, a strong hypomorphic allele of fgf8, have been described previously(Brand et al., 1996) and we refer to the homozygous mutants as pax2a^– and fgf8^–, respectively. The mutation in foxi1was described by Solomon et al. (Solomon et al., 2003), and we refer to the homozygous mutants as foxi1^–. Homozygous pax2a^–mutants were either scored by the loss of the midbrain-hindbrain boundary or by the use of Pax2 antibody (α-Pax2; Covance); homozygous foxi1^– mutants were scored by absence of pax8 expression or by PCR(Solomon et al., 2003). Homozygous pax2a^–;foxi1^–double mutants were scored for both phenotypes.

Approved gene and protein names that follow the zebrafish nomenclature conventions(http://zfin.org/zf_info/nomen.html)are used.

Antibody staining was carried out as described previously(Westerfield, 2000) with some modifications. Primary antibodies were used in the following concentrations:α-Pax2 (Covance), 1:100; α-Dlx3b(Liu et al., 2003), 1:50;rabbit polyclonal IgG α-Myc (Santa Cruz Biotechnology), 1:500. The following secondary antibodies were used: goat α-mouse Alexa Fluor 488(Molecular Probes), goat α-mouse Alexa Fluor 568 (Molecular Probes),1:100; goat α-rabbit Alexa Fluor 488 (Molecular Probes), 1:100. Embryos were analyzed using a Zeiss Axiophot 2 microscope.

cDNA probes that detect the following genes were used: dlx3b(Ekker et al., 1992); sox9a (Chiang et al.,2001); sox9b (Chiang et al., 2001; Li et al.,2002); egr2b (previously krox20)(Oxtoby and Jowett, 1993); cldna (Kollmar et al.,2001); and pax2a(Krauss et al., 1991). For detection of pax2a in dnpax2a-myc-injected embryos, the 5′ and 3′ UTRs of pax2a were amplified by PCR, subcloned into pBluescript, linearized with NotI and EcoRI,respectively, and transcribed with T7 RNA polymerase. Probe synthesis and single or double-color in situ hybridization was performed essentially as previously described (Thisse et al.,1993; Jowett and Yan,1996; Whitlock and Westerfield, 2000). We purified the in vitro synthesized mRNA and probes using an RNeasy mini column (Qiagen GmbH). In vitro mRNA synthesis was performed using an SP6 RNA synthesis kit (Ambion). The construct encoding dnPax2a-myc was generated by PCR amplification of the pax2a gene coding for the first 295 amino acids, which were fused in-frame with six Myc-epitopes and cloned into the CS2+ vector(Turner and Weintraub, 1994). For RNA injections, 1-3 nl of a 300 ng/μl solution was delivered into the cytoplasm of one cell at the two-cell stage.

We have described the dlx3b-MO, fgf3-MOs and fgf8-MOs previously (Liu et al.,2003; Maves et al.,2002). Splice-blocking pax8-MOs were: E2/I2,5′-GTGTGTGTTCACCTGCCCAGGATCT; E3/I3, 5′-GTGTGTACCGGTTGATGGAGCTGAC;E4/I4, 5′-CACAGCACTTACTCAGTGTGTGTCC; E5/I5,5′-TTTCTGCACTCACTGTCATCGTGTC; and E9/I9,5′-ACCGGCGGCAGCTCACCTGATACCA. About 1-3 nl of MO-solution was injected into the cytoplasm of one-cell stage embryos. The concentration of the pax8 splice-blocking MOs was 1 μg/μl each for pax8-E5/I5 and pax8-E9/I9.

In zebrafish, the otic anlagen express pax2a, pax2b and pax8 prior to formation of the placode. Expression of pax8is initiated at 85-90% epiboly and upregulated at bud stage [for a detailed description, see Phillips et al. (Phillips et al., 2001)]. Strong expression persists until the 9- to 10-somite stage (Fig. 1A,B),when the otic placode becomes morphologically visible. Subsequently, pax8 is rapidly downregulated(Fig. 1C) and we are no longer able to detect pax8 transcripts in the otic placode(Fig. 1D), in contrast to previous reports describing a low level expression of pax8 until formation of the otic vesicle at 18-somite stage(Pfeffer et al., 1998; Phillips et al., 2001). pax2a expression begins in presumptive otic cells at the three-somite stage. High levels of pax2a can be detected as the placode develops and expression is maintained throughout the placode(Fig. 1A-D). After formation of the otic vesicle, pax2a transcripts become localized to the ventromedial region of the otic vesicle and are eventually retained only in sensory hair cells (Riley et al.,1999). To determine whether the same population of preotic cells expresses pax2a and pax8, we double labeled for both pax2a and pax8 from preplacodal to placodal stages. The preotic pax8 expression domain overlaps with pax2a from the onset of pax2a expression until pax8 is downregulated. The two domains share the same medial border that abuts the hindbrain, but the pax8 domain extends farther lateral than does the pax2adomain (Fig. 1E-H). pax2b is also expressed in the preotic region overlapping with pax2a (not shown). However, pax2b is expressed later, just prior to formation of the otic placode, and at low levels. It was therefore not included in this report.

Previous studies have suggested that Fgf3 and Fgf8 play overlapping roles in otic induction and that both pax2a and pax8 require Fgf3 and Fgf8 signaling for their proper expression(Phillips et al., 2001; Maroon et al., 2002; Leger and Brand, 2002). However, we have shown that even in the absence of Fgf3 and Fgf8 signaling,some residual cells express pax2a(Liu et al., 2003). We, thus,examined whether this is also true for pax8.

We find that knockdown of Fgf3 and Fgf8 in wild-type embryos or knockdown of Fgf3 in fgf8^– mutants significantly reduces pax8 expression in the preotic region, although weak residual expression can still be detected (Fig. 1I,J). This finding is similar to the observations of Maroon et al. (Maroon et al., 2002), but contrasts with the results of Phillips et al.(Phillips et al., 2001) and Leger and Brand (Leger and Brand,2002) who reported complete loss of pax8 when Fgf3 and Fgf8 are knocked-down by antisense morpholino oligonucleotide (MO) injection. Although the discrepancies between these studies might be explained by incomplete effectiveness of the MOs, the Fgf receptor blocking drug SU5402 also led to differing results: Leger and Brand(Leger and Brand, 2002)reported a complete loss of pax8 expression whereas Maroon et al.(Maroon et al., 2002) found that pax8 was unaffected. Thus, taken together, these results could indicate either that pax8 is highly sensitive to Fgf signaling and only complete loss of the Fgf signal leads to loss of pax8 expression or that pax8 expression is regulated partly by some factor in addition to Fgf. foxi1^– mutants fail to initiate pax8 expression (Solomon et al.,2003; Nissen et al.,2003), suggesting that this forkhead-related transcription factor may be the other regulator of pax8 expression. To test this interpretation, we examined whether foxi1 acts independently of Fgf signaling. We found that knockdown of Fgf3 and Fgf8 in wild-type embryos or knockdown of Fgf3 in fgf8^– mutants has no significant effect on foxi1 expression(Fig.1K,L), supporting the hypothesis that foxi1 expression is independent of Fgf signaling, but required for cells to respond to Fgf signaling(Nissen et al., 2003).

Because pax8^– mutants are not available, we used MOs to knock down gene function in a gene-specific manner(Nasevicius and Ekker, 2000). In addition to their ability to block the translation of mRNAs in the cytoplasm, MOs can inhibit pre-mRNA splicing(Draper et al., 2001), thus interfering with the transport of transcript from the nucleus to the cytoplasm(Yan et al., 2002). This inappropriate retention of transcripts in the nucleus can be used as an assay for MO efficacy in the absence of an antibody to test for the production of a translated product (Yan et al.,2002). We used this splice-blocking strategy for pax8because neither a Pax8 antibody nor the sequence of the 5′ terminus of the pax8 mRNA (Pfeffer et al.,1998) was available. We determined the sequence of several introns, designed specific splice-blocking MOs(Fig. 2A), and injected the MOs alone and in various combinations. As a control for the efficacy of the MOs,we visualized the nuclear localization of pax8 messenger by in situ hybridization; the most efficacious combination of MOs was used in this study(Fig. 2A-C).

Injection of pax8-MOs into wild-type embryos has only a subtle effect: the morphological development of the otic placode is delayed(Fig. 2F,J) compared with wild-type embryos injected with control MOs or to un-injected wild-type embryos (Fig. 2D,H). Consistent with their eventual otic fates, cells in the embryos injected with pax8-MOs express otic markers such as Dlx3b, claudin a(cldna) and fibronectin 1 (fn1)(Fig. 2N,R; not shown), again showing only a slight developmental delay compared with control embryos(Fig. 2L,P). The lack of a stronger phenotype could be due to compensation for pax8 by pax2a. pax2a^– mutants exhibit only a weak neurogenic phenotype in the ear that probably results from reduced Delta signaling(Riley et al., 1999) but are otherwise relatively unaffected (Fig. 2E,I,M,Q). Injection of pax8-MOs into pax2a^– mutants produces a different and highly penetrant phenotype, including severe disruption of otic development. Most of the injected pax2a^– embryos (79%, 44/56) do not form an otic vesicle (Fig. 2G,K),although a few (12/56) form an extremely small vesicle that lacks otoliths(not shown). Nevertheless, analysis of the otic markers, Dlx3b, cldnaand fn1 reveals the presence of residual otic cells, even in embryos with no visible otic vesicles (Fig. 2O,S; not shown). Thus, pax2a and pax8 have overlapping and, apparently, synergistic functions required for formation of the ear, although removal of both gene functions is insufficient to eliminate all otic cells.

To study the placement of pax2a and pax8 in the genetic pathway regulating otic development, we examined the expression of several otic markers at preplacodal, placodal and vesicle stages. We concentrated on the transcription factor genes, sox9a, sox9b and dlx3b(Yan et al., 2002; Chiang et al., 2001; Akimenko et al., 1994), that are essential for formation of the ear(Liu et al., 2003). We also examined pax2a expression that is reduced in pax2a^– mutants at later stages(Brand et al., 1996).

Loss of Pax2a, together with Pax8 knockdown, affects sox9a and sox9b expression. sox9a is broadly expressed in the preotic region at the three-somite stage (Fig. 3A) and expression is maintained in the placode(Fig. 3C) and in the vesicle(not shown). In pax2a^– mutants injected with pax8-MOs, the expression of sox9a is reduced in extent and level (Fig. 3B). At the 12-somite stage, when the placode is morphologically visible in wild-type embryos, we detect no sox9a expression in pax2a^– mutants injected with pax8-MOs(Fig. 3D). The sox9aduplicate, sox9b, is also expressed from preplacodal to vesicle stages, although it is initiated later in development(Fig. 3E,G). Like sox9a,sox9b expression is compromised at preplacodal stages in pax2a^– mutants depleted of pax8(Fig. 3E) and absent at the 12-somite stage (Fig. 3H).

dlx3b expression is less affected by loss of Pax2a and Pax8 knockdown. By the end of gastrulation, a band of cells expresses dlx3b surrounding the neural plate, particularly in the region that corresponds to the future otic placode(Fig. 3I). dlx3bexpression persists throughout the placode until formation of the otic vesicle(Fig. 3K). In pax2a^– mutants injected with pax8-MOs, the dlx3b expression domain is smaller, but the concentration of cells in the region where the placode would normally form can still be recognized(Fig. 3J). In contrast to sox9a and sox9b, dlx3b expression can still be detected in some cells in pax2a^– mutants depleted of pax8 at the 12-somite stage (Fig. 3L) and even at the stage when the vesicle would form in normal embryos (Fig. 2O). Previously,we have shown that loss of Sox9a leads to strong reduction of Dlx3b in the preotic domain (Liu et al.,2003) and because the phenotype of these embryos is essentially the same as pax2a^– mutants depleted of Pax8, we conclude that reduction of dlx3b gene expression in the preotic domain in the absence of Pax2a and Pax8 is presumably due to concomitant reduction of Sox9a.

Maintenance of pax2a expression depends strongly on Pax2a or Pax8. In pax2a^– mutants, pax2a transcription initiates normally and we cannot distinguish between wild-type, pax2a^– mutants and pax2a^–mutants injected with pax8-MOs at preplacodal stages (not shown). However, there is severe downregulation of pax2a in pax2a^– mutants injected with pax8-MOs by the 12-somite stage (Fig. 3N)and pax2a is completely lost by vesicle stages(Fig. 3P). Labeling for fgf3 or fgf8 expression and egr2b or mafb(previously valentino) expression, both downstream targets of Fgf signaling (Maves et al.,2002), reveals that expression of fgf3 or fgf8and patterning of the hindbrain occurs normally in pax2a^– mutants depleted of Pax8(Fig. 3N, and not shown). These results indicate that Pax2a and Pax8 act synergistically downstream of Fgf3 and Fgf8; when Pax2a and Pax8 functions are both compromised, otic induction is weaker and otic fate is not maintained, even in the presence of normal Fgf signaling.

Because we see some residual specification of otic cells in Pax2a and Pax8 knockdown embryos, we were concerned that some Pax protein function remained. Structure-function analyses have shown that proteins of the Pax2-Pax5-Pax8 family have overlapping biochemical activities; their DNA-binding specificities are highly similar and they can substitute for each other(Bouchard et al., 2000). Thus,to block all Pax2-Pax5-Pax8 function, we generated a dominant-negative form of Pax2a (dnpax2a-myc) by replacing the C-terminal transactivation-inhibitory domain with six Myc epitope tags(Fig. 4A). We injected mRNA from this pax2a variant into wild-type embryos and analyzed subsequent otic development. To assess the effectiveness of dnpax2a-myc, we examined eng3 expression in injected embryos. Expression of eng3, a downstream target of Pax2a in the midbrain-hindbrain boundary region, is initiated at the one-somite stage in wild-type embryos (Fig. 4B) but is never activated in strong pax2a^– mutants(Lun and Brand, 1998). In dnpax2a-myc mRNA-injected embryos, we identify regions expressing the variant protein by the presence of the Myc-epitopes. In these regions, eng3 transcription is severely reduced or completely absent(Fig. 4C). The nuclear localization of the Myc-epitopes and the severe downregulation of eng3 show that this Pax2a variant enters the nucleus and competes with the endogenous Pax2-Pax5-Pax8 proteins for binding sites. We expect that,owing to its abundance, this construct is able to out compete the endogenous proteins and act in a dominant-negative fashion.

Injection of dnpax2a-myc strongly affects initial otic development. Expression of both sox9a and sox9b is severely reduced or absent in the presence of the Pax2a variant, interfering with expression not only in the preotic domain but also in the neural crest and neural tube (Fig. 4D′,D″,E′,E″). Because pax8 is the only member of the Pax2-Pax5-Pax8 family known to be expressed in the preotic region at this early developmental stage, our result suggests that the dominant-negative Pax2a construct interferes with Pax8 function and that Pax8 is required for correct early otic expression of sox9a and sox9b. In addition, pax2a transcription is affected in the preotic domain of these injected embryos but not in the isthmus(Fig. 4F′,F″)indicating that otic pax2a, but not isthmic pax2a,expression depends on Pax2-Pax5-Pax8 proteins at this stage. By contrast, dlx3b expression is less affected in the presence of the dominant-negative form of Pax2a; the preotic dlx3b domain is slightly reduced, presumably owing to loss of sox9a, but can still be easily recognized (Fig. 4G′,G″). Thus, injection of the dominant-negative form of Pax2a affects expression of sox9a, sox9b and dlx3b,similar to loss of pax2a together with Pax8 knockdown. By contrast, pax8-MO injection into pax2a^– mutants has only a mild effect on initiation of pax2a expression, whereas injection of dnpax2a-myc blocks pax2a induction. This discrepancy could indicate that the pax8-MOs are only partially effective in blocking Pax8 function, or alternatively that a Pax protein other than Pax8 is required for initiation of pax2a expression. Analysis of the transcript structure of pax8 (B. Riley, personal communication)shows that pax8 is subject to alternative splicing, producing some transcripts that are not targeted by these pax8-MOs; thus, incomplete depletion of Pax8 is likely. Nevertheless, these results together demonstrate that Pax8 function is required for correct otic expression of sox9a,sox9b, and pax2a and, to a much lesser extent, dlx3b.

Recent studies implicate combined functions of Fgf3 and Fgf8 in otic pax2a induction (Phillips et al.,2001; Maroon et al.,2002; Leger and Brand,2002). In addition, Pax2a and Fgf8 are both required to maintain pax2a expression in the isthmus(Lun and Brand, 1998; Reifers et al., 1998). To test whether Pax8 also acts together with Fgf signals to promote pax2aexpression, we injected pax8-MOs into fgf8^–mutants. pax2a expression in fgf8^– mutants alone shows normal timing, but the size of the otic expression domain is significantly reduced (Fig. 5A,B) (Phillips et al.,2001), and the otic vesicle is subsequently smaller(Fig. 5D,E)(Phillips et al., 2001). fgf8^– mutants depleted of pax8 show an even stronger reduction of pax2a expression in the preotic region(Fig. 5C) and form an even smaller ear (Fig. 5F). This observation suggests that Fgf8 and probably Fgf3 act through and in parallel with Pax8 to promote proper pax2a expression.

The forkhead domain transcription factor, Foxi1, has been implicated as a regulator of pax2a and pax8(Solomon et al., 2003). foxi1^– mutants display highly variable otic morphologies, initiation of otic pax8 fails and otic pax2aexpression is severely delayed. Early dlx3b expression is unaffected in foxi1^– mutants, but concentration of dlx3b-expressing cells in the preotic region is delayed and patchy(Solomon et al., 2003; Nissen et al., 2003). Previously, we have shown that Dlx3b is required for the correct temporal onset of pax2a expression in the preotic region(Liu et al., 2003). Thus,reduced expression of Dlx3b could explain the delayed and patchy expression of pax2a in foxi1^– mutants. To determine whether the residual patchy Pax2a is co-expressed with Dlx3b, we used double fluorescent antibody labeling (Fig. 6I-J′). In wild-type embryos at the 12-somite stage, all cells of the otic placode express Pax2 and Dlx3b(Fig. 6I-I″). In foxi1^– mutants, all otic cells that express Dlx3b are also Pax2 positive and when Dlx3b expression appears in two or more smaller patches, Pax2 expression is affected in an identical manner(Fig. 6J-J′).

To place Foxi1 in the genetic pathway regulating otic specification, we examined whether foxi1 is required for correct expression of the two Sox9 genes. In foxi1^– mutants, both the size of the expression domain and the level of expression of sox9a in the preotic region are significantly reduced at the three-somite stage(Fig. 6B) in comparison with wild-type embryos (Fig. 6A). These reductions are even more significant than in pax2a^– mutants injected with pax8-MOs(compare with Fig. 3B). In foxi1^– mutants at the 12-somite stage, sox9a is variably expressed in the otic region; tightly aggregated patches of cells express high levels of sox9a(Fig. 6D), but the total number of sox9a-expressing cells is reduced in comparison with wild-type embryos (Fig. 6C). The sox9a duplicate, sox9b, has a similar expression pattern in foxi1^– mutants. At the five-somite stage, we detect very low levels of sox9b expression in the preotic region(Fig. 6F), even lower than in pax2a^– mutants depleted of Pax8 (compare with Fig. 3F). By the 12-somite stage, sox9b expression somewhat recovers in foxi1^– mutants(Fig. 6H), although not to wild-type levels (Fig. 6G).

Together, these results show that in the absence of Foxi1 function, and hence also in the absence of pax8 expression, only weak induction of otic sox9a and sox9b occurs, and otic pax2aexpression is restricted to Dlx3b-positive cells.

Our analysis of foxi1^– mutants, dominant-negative Pax2a and pax2a^– mutants depleted of Pax8 suggest that Pax2a may provide a Foxi1-independent pathway for otic specification. To test this hypothesis, we generated foxi1^–;pax2a^– double mutants and analyzed them for otic specification. The ears of pax2a^– single mutants are virtually indistinguishable from wild-type embryos by morphology(Fig. 2D,E,H,I) or by gene expression (Fig. 2L,M,P,Q). foxi1^– single mutants are highly variable; some mutants develop a small lumen with only one or no otolith, whereas others have small split lumens, each with a single otolith(Solomon et al., 2003; Nissen et al., 2003). The foxi1 mutant allele we used typically forms a small lumen with one otolith (Fig. 7B,G) and consistently retains some expression of the otic markers Dlx3b, cldnaand fn1 (Fig. 7L,O;not shown).

foxi1^–;pax2a^– double mutants never show any morphological sign of otic specification(Fig. 7C,H) and the otic markers Dlx3b and fn1 are lost(Fig. 7M; not shown). Labeling for cldna expression, however, reveals the presence of some residual`otic' cells that can be distinguished from cldna-expressing cells of the anterior and posterior lateral line placodes by expression levels and position (Fig. 7P). At the 12-somite stage, foxi1<sup>–</sup>;pax2a<sup>–</sup> double mutants are similar to pax2a<sup>–</sup> mutants injected with pax8-MOs: they lack sox9a and sox9b expression and have reduced dlx3b expression (not shown). pax2a expression,however, is more severely reduced or completely lost in the foxi1<sup>–</sup>;pax2a<sup>–</sup> embryos compared with pax2a<sup>–</sup> mutants injected with pax8-MOs (compare Fig. 7U with Fig. 3N). Thus, embryos with compromised Foxi1 and Pax2a show an even stronger loss of otic specification than pax2a<sup>–</sup> mutants depleted of Pax8, although a few residual cells still assume an `otic' fate. These results further support our interpretation that Pax2a acts synergistically with Pax8 in otic specification and demonstrate that the majority of otic cells specified in the absence of Foxi1 are Pax2a dependent.

Our observation that a few residual cells express otic markers even in foxi1^–;pax2a^– double mutants suggests the possibility that an additional factor participates in otic specification; Dlx3b is a likely candidate. To test this possibility, we compromised Dlx3b function using morpholino injection. Reduction of Dlx3b in wild-type embryos impairs complete maturation of the otic vesicle and most embryos form a smaller vesicle with only one otolith(Fig. 7D,I)(Solomon and Fritz, 2002; Liu et al., 2003) and reduced cldna expression (Fig. 7Q). By contrast, foxi1^– mutants injected with dlx3b-MO show no morphological signs of otic specification (Fig. 7E,J) and no otic expression of cldna (Fig. 7R). In wild-type embryos depleted of Dlx3b, otic specification is delayed as indicated by delayed onset and reduced expression of pax2a(Fig. 7V). However, in foxi1^– mutants injected with dlx3b-MO, otic pax2a expression is undetectable(Fig. 7W). Taken together,these results show that removal of the two factors, Foxi1 and Dlx3b, leads to a complete absence of otic specification.

Previous studies have suggested that Pax2, Pax5 and Pax8 have overlapping functions (Urbanek et al.,1994; Torres et al.,1996; Mansouri et al.,1998; Bouchard et al.,2002). To define the roles of pax2a and pax8 in otic induction, we analyzed the knockdown of pax8 by morpholino injection into wild-type embryos and into pax2a^–mutants. Our results show that removal of both Pax2a and Pax8 together prevents the formation of the otic vesicle and leads to a substantial loss of otic tissue (Fig. 2), whereas neither the single null mutation of pax2a nor the knockdown of pax8 alone is sufficient to block otic placode induction. This result demonstrates that Pax2a and Pax8 have overlapping functions in otic development.

Otic vesicle formation in pax2a^– mutants is virtually the same as in wild-type embryos; pax2a^–mutants show a weak neurogenic phenotype, probably owing to reduced Delta signaling (Riley et al.,1999). This result indicates that in the absence of Pax2a, Pax8 is sufficient for most aspects of otic development, even though there are differences in the expression patterns of these two Pax genes. Specifically, pax8 is expressed prior to pax2a, and in contrast to pax8, pax2a transcription continues after the otic placode becomes morphologically visible (Fig. 1) until it is subsequently restricted to sensory hair cells(Riley et al., 1999). The presence of the pax2a duplicate, pax2b, cannot account for the absence of a more dramatic phenotype because depletion of Pax2a and Pax2b together leads to no loss of otic structures(Whitfield et al., 2002). These observations suggest either that Pax8 protein is stable and can provide sufficient function at later stages after transcription has ended or that Pax2-Pax5-Pax8 function is not required after placode formation. We are unable to distinguish between these two interpretations because no antibody against Pax8 is currently available.

The otic phenotype of pax2a^– mutants depleted of Pax8 is similar to embryos depleted of both Fgf3 and Fgf8. Previous studies have shown that knockdown of Fgf3 and Fgf8 in wild-type embryos or knockdown of Fgf3 in fgf8^– mutants causes a synergistic loss of otic tissue, indicating that fgf3 and fgf8 encode overlapping functions required for otic specification(Phillips et al., 2001; Maroon et al., 2002; Leger and Brand, 2002; Liu et al., 2003). The failure of otic tissue formation in the absence of Fgf function is preceded by a strong reduction of pax2a and pax8 expression(Phillips et al., 2001; Leger and Brand, 2002). Furthermore, compromising Fgf signals leads to an absence of sox9aexpression and significant reduction of sox9b expression in the preotic region (Liu et al.,2003), similar to the Fgf dependence of Sox9 in Xenopus(Saint-Germain, 2004). By contrast, early dlx3b(Leger and Brand, 2002; Liu et al., 2003) and foxi1 (Fig. 1)expression is less affected by loss of Fgf signaling and effects on dlx3b expression are caused at least in part by reduced levels of sox9a at early stages and by reduced levels of sox9a and sox9b at later stages (Liu et al., 2003). Together, these observations lead to the conclusion that pax2a and pax8 act downstream of Fgf3 and Fgf8, but upstream of sox9a and sox9b, and that Pax2a and Pax8 are mediators of Fgf signals during otic placode induction(Fig. 8). Recent experiments in Xenopus have led to the suggestion that Sox9 may act upstream of Pax8, although the results reported do not rule out the possibility that Sox9 and Pax8 interact to maintain each other's expression(Saint-Germain et al.,2004).

The persistence and probably initial induction of pax2a expression depend upon Fgf signaling and Pax function. Normal activation of pax2a requires Pax8 in cooperation with Fgf8(Fig. 5) and, because pax8 expression is not maintained after the placode forms(Fig. 1), long-term maintenance of pax2a expression is probably supported by Pax2a activity. This interpretation is consistent with studies in mouse, chick and zebrafish that indicate a positive feedback loop in the isthmic region that requires Pax2 and Fgf8 activity for maintenance of Pax2 expression(Urbanek et al., 1994; Torres et al., 1996; Mansouri et al., 1998; Martinez et al., 1999; Brand et al., 1996; Lun and Brand, 1998; Reifers et al., 1998). The presence of a high-affinity Pax2-Pax5-Pax8-binding site conserved among human,mouse and fugu in the upstream promoter sequence of the Pax2gene is consistent with the auto-regulatory function of Pax2(Pfeffer et al., 2002). This element is also conserved in the zebrafish pax2a promoter (data not shown). Despite the importance of Pax protein for pax2a expression,induction may occur in the absence of Pax activity. In foxi1^– mutants(Solomon et al., 2003; Nissen et al., 2003)(Fig. 7), where pax8expression is undetectable in the preotic region, pax2a expression nevertheless appears, although delayed and variable. These observations may suggest that when Pax8 is reduced, longer exposure to Fgf is required to induce pax2a expression. Consistent with this interpretation, we find that embryos injected with the dominant-negative Pax2a construct completely lack otic pax2a expression (Fig. 4), presumably because Pax function is more effectively blocked.

Our results are consistent with those of Nissen et al.(Nissen et al., 2003) and Solomon et al. (Solomon et al.,2004) and suggest that Foxi1 is required for the initial Fgf-dependent induction of pax8. In foxi1^–mutants, early expression of sox9a and sox9b is also severely affected (Fig. 6),similar to the effects of Pax8 depletion in pax2a^–mutants. However, unlike embryos lacking both Pax2a and Pax8, foxi1^– mutants later recover sox9a and sox9b expression (Fig. 6). This recovery is due to Pax2a(Fig. 7); once expression begins, Pax2a protein maintains its own expression and activates downstream sox9 target genes. Thus, it is likely that variability in the onset of pax2a expression, in the absence of Foxi1 and, hence, Pax8(Fig. 8), produces the highly variable phenotype of foxi1^– mutants. Supporting this interpretation, foxi1^–;pax2a^– double mutants exhibit consistent, more severe reduction of otic tissue(Fig. 7).

Our data indicate that Pax2a and Pax8 participate in the same otic developmental pathway: Foxi1 and Pax8 mediate the initial Fgf dependent induction that includes initiation of Dlx3b-dependent pax2aexpression. Pax2a subsequently maintains its own expression. This model contrasts somewhat from previous suggestions(Riley and Phillips, 2003)primarily based on studies in mouse where loss of Foxi1(Hulander et al., 1998; Hulander et al., 2003) or Pax8(Mansouri et al., 1998) does not prevent otic Pax2 expression or early patterning and morphogenesis of the otic vesicle. This apparent discrepancy in Foxi1 function between zebrafish and mouse may be due to temporal differences in development. Otic induction in response to Fgf signals occurs over a much longer time period in mice than in zebrafish, which provides more time for cells in mammalian embryos to respond to Fgf signals, even in the absence of Pax8. Analysis of Pax2;Pax8 double mutant mice will be necessary to test this interpretation definitively.

Our results also provide further insight into Fgf-dependent and-independent processes and the mechanisms underlying competence in otic development. Previously, we have demonstrated that loss of either Fgf3 and Fgf8 or loss of Dlx3b, Dlx4b and Sox9a results in nearly complete loss of otic tissue, although a few residual cells express otic markers including pax2a, fn1 and cldna(Liu et al., 2003). Loss of both Fgf signals, and all three of these transcription factors completely blocks all indications of otic induction, suggesting that Fgf-dependent and Fgf-independent processes of otic induction act synergistically. We propose that induction of otic fate by Fgf signals takes place only when cells are competent to respond, and that this competence is provided by Foxi1 and Dlx3b(Fig. 8). A direct role for Foxi1 and Dlx3b in competence needs to be demonstrated, for example by ectopic expression and transplantation experiments. Foxi1 and Dlx3b function by regulating pax8 and pax2a expression, respectively, in an Fgf-dependent fashion. In Dlx3b-deficient embryos, expression of pax8is indistinguishable from that in wild-type embryos, presumably owing to normal Foxi1 and Fgf signaling. However, otic pax2a expression is initiated only very late and weakly(Solomon and Fritz, 2002; Liu et al., 2003). By contrast, otic pax8 expression fails and pax2a expression is present although delayed in foxi1^– mutants(Solomon et al., 2003; Nissen et al., 2003). Inhibition of both factors, Foxi1 and Dlx3b, completely blocks otic specification even in the presence of functional Fgf signaling(Fig. 7). By activating Pax8,Foxi1 thus provides competence to otic precursor cells to respond to early Fgf signaling; Dlx3b and Pax2a subsequently maintain this competence(Fig. 8).

We thank Sandra Brown for technical assistance; Andreas Fritz and Bruce B. Riley for sharing unpublished results; Andreas Fritz and Robert M. Nissen for materials; and Katharine E. Lewis, Kate F. Barald and Tanya T. Whitfield for critical reading of the manuscript. This work was supported by NIH DC04186 and HD22486. S.H. is a recipient of a Feodor Lynen fellowship of the Alexander von Humboldt foundation. D.L. was supported by a postdoctoral fellowship of the Canadian Institutes of Health Research. Dedicated to José A. Campos-Ortega, who died on 8 May 2004.