Fgf3 is required for dorsal patterning and morphogenesis of the inner ear epithelium

The inner ear is a morphologically complex sensory organ responsible for hearing and balance. In higher vertebrates, the sound-sensing cells are housed in the ventrally located cochlear duct and the head motion-sensing cells are housed within the dorsally located vestibular system. The vestibular system is further divided into the more dorsally located semicircular canal system, in which sensory patches devoted to angular motion detection (cristae) are found at the base (ampulla) of each of three non-sensory canal ducts, and sensory patches devoted to linear motion detection (maculae) are contained within the centrally located saccule and utricle. The inner ear also has a dorsally projecting non-sensory appendage, the endolymphatic duct and sac (EDS), which plays an important role in maintaining the unique fluid environment essential for normal sensory function of the inner ear. Dysfunction of the inner ear is among the most common congenital disorders, affecting at least 1 in 500 births(Smith et al., 2005; Morton and Nance, 2006) and up to 39% of sensorineural deafness is associated with inner ear malformations(Mafong et al., 2002; Wu et al., 2005).

In mice, inner ear formation initiates on embryonic day (E) 8.0 as a placodal thickening of head ectoderm adjacent to hindbrain rhombomeres (r)5-6. Subsequently, from E8.5-9.0, the placode invaginates, forming the otic cup, from which neuroblasts of the eighth ganglion (GVIII) delaminate anteroventrally and migrate medially. By E9.5, the cup closes and separates from the surface ectoderm, forming the spherical otic vesicle, or otocyst. During the next 6 days, this simple, fluid-filled epithelium acquires its mature and complex morphology and begins to differentiate its multitude of distinct sensory and non-sensory cell types. Morphogenesis of the vesicle initiates at E10.5 with a dorsomedially directed outgrowth of the EDS anlage and a ventrally directed outgrowth of the cochlear duct. At E11.5, the otic epithelium evaginates dorsolaterally to form the vertical canal plate, the precursor of the anterior and posterior semicircular canals, which are separated by the common crus, and shortly thereafter evaginates laterally to form the lateral canal plate, the precursor of the lateral semicircular canal. The canals are formed at E12.5 by fusion of cells within two central regions of the vertical plate and a single central region of the lateral canal plate,followed at E13.5 by resorption of the fused cells into the epithelium. During this period, there are additional evaginations of the central region to form the more dorsally situated utricle and the more ventrally situated saccule. By E15.5, the inner ear epithelium has acquired its mature morphology, but is still undergoing cellular differentiation within the six sensory patches(Kiernan et al., 2002; Barald and Kelley, 2004; Mansour and Schoenwolf, 2005; Fritzsch et al., 2006).

The initial induction and subsequent morphogenesis and differentiation of the otic epithelium involves signaling interactions within and between otic and non-otic tissues. In particular, fibroblast growth factor (FGF) signals from the endoderm induce a mesenchymal FGF that is required together with hindbrain FGFs for expression of otic placode genes and for otic placode induction and vesicle formation. Based on genetic data, these signals are likely to be provided in mice by FGF8, FGF10 and FGF3, expressed by the endoderm, mesenchyme and hindbrain, respectively(Alvarez et al., 2003; Wright and Mansour, 2003; Ladher et al., 2005; Zelarayan et al., 2007). WNT signals, presumably from the hindbrain, are required to limit the region of the ectoderm that forms the otic placode(Ohyama et al., 2006). In response to the inductive signals, the otic placode expresses several transcription factor genes uniformly, including Pax2, Dlx5 and Gbx2 (Wright and Mansour,2003; Ladher et al.,2005). By the otocyst stage, the expression domains of most otic markers become confined to restricted domains of the vesicle specified to give rise to different ear structures (Fekete and Wu, 2002). This division of the otocyst into gene expression domains required for normal morphogenesis also depends on signals from the hindbrain. Removal or rotation of the neural tube in the vicinity of the developing otic tissue or rotation of the vesicle disrupts otic vesicle molecular patterning and development, particularly the outgrowth of the EDS and cochlear duct (Hutson et al.,1999; Bok et al.,2005). Mutations in hindbrain-expressed genes, such as Mafb (Kreisler) and Hoxa1, which cause defects in the development of r5-6, also show aberrant patterning of the otic vesicle that presages inner ear malformations(Pasqualetti et al., 2001; Choo et al., 2006). WNT signaling from the hindbrain is sufficient to maintain the expression of some dorsal otic genes, and Wnt1 and Wnt3a, which are expressed in the dorsal hindbrain, are required redundantly for formation of the vestibular structures (Riccomagno et al.,2005).

FGF signaling is also implicated in otic vesicle morphogenesis. FGF3 and FGF10 are expressed by prospective otic sensory tissues(Wilkinson et al., 1988; Mahmood et al., 1996; McKay et al., 1996; Pirvola et al., 2000; Pauley et al., 2003), and their major receptor, FGFR2b, is found in prospective non-sensory tissue(Pirvola et al., 2000). Global inhibition of FGF signaling by application of SU5402 to chick E2.5-3 otic vesicles just prior to semicircular canal pouch evagination causes dose-dependent inhibition of canal formation, suggesting that developing sensory structures control morphogenesis of non-sensory structures(Chang et al., 2004). Mouse mutants support this concept. Mice lacking FGFR2b form otic vesicles that subsequently develop dysmorphologies initiating at EDS outgrowth and include failure of semicircular canal formation(Pirvola et al., 2000). Most Fgf10 mutants completely lack semicircular canals(Pauley et al., 2003; Ohuchi et al., 2005). Fgf3-null mutants undergo normal otic vesicle formation, but then go on to develop highly variable and incompletely penetrant inner ear dysmorphologies that, like those of Fgfr2b mutants, appear to initiate at the EDS outgrowth stage. These mutants also show a reduction in the size of GVIII (Mansour et al.,1993).

To gain insight into the mechanism of Fgf3 function in otic morphogenesis, we determined the three-dimensional structure of Fgf3mutant ears and the changes in their molecular patterning. We show that homozygous mutants have variably penetrant and expressive inner ear morphologies that are characterized most typically by loss of the EDS and common crus, dilatation of the remaining epithelium and poor coiling of the cochlea, phenotypes that are very similar to those of Hoxa1, Mafb and Gbx2 mutants (Pasqualetti et al.,2001; Lin et al.,2005; Choo et al.,2006) as well as to some classes of hearing-impaired patients(Manfre et al., 1997; Kim et al., 2004; Wu et al., 2005). The initial molecular patterning of Fgf3 mutant otocysts was normal, but by E10-10.5 these ears lacked or had reduced domains of dorsal otic genes,suggesting a loss of the cells fated to form the EDS. Ventrally expressed genes important for cochlear development were not affected, but markers of the developing vestibular sensory domains, particularly those expressed in the posterior otic vesicle, were downregulated or absent and GVIII was reduced and displaced dorsomedially. Finally, hindbrain expression of Wnt3a was ventrally expanded in Fgf3 mutants. Taken together, our results show that Fgf3 is required for dorsal otic patterning and that its primary role is to sustain and focus the dorsal otic gene expression induced by WNT signals. We suggest that Fgf3 also serves to focus inductive WNT signals on the dorsal otic vesicle, highlighting a new example of cross-talk between the two signaling systems.

All mouse use complied with protocols approved by the University of Utah Institutional Animal Care and Use Committee. The targeted Fgf3^neo allele (Fgf3^tm1Mrc,MGI:1931059) was described previously and has an insertion of a promoter-containing Neo gene in the first protein-coding exon (1B)(Mansour et al., 1993). A new Fgf3 allele (Fgf3^Δ2),which has a deletion of exon 2, was generated by first targeting loxP sites on either side of exon 2 to generate a conditional allele and then deleting the intervening DNA using CRE recombinase (X.W. and S.L.M., unpublished). Fgf3^Δ2 transcripts contain exon 1 spliced directly to exon 3, which is in a different frame. Thus, they encode only the first 73 of 245 FGF3 amino acids, including only the first 28 of 137 amino acids of the FGF core homology domain.

Both Fgf3 alleles were studied on a mixed genetic background containing hybrid 129 and C57Bl/6 alleles. Genotypes of heterozygous intercross offspring were determined by PCR amplification of yolk sac or tail DNA using three-primer mixes that distinguish the mutant and wild-type alleles. The Fgf3^neo genotyping assay was performed as described previously (Wright and Mansour,2003). The Fgf3^Δ2genotyping mix, containing primers 455C(5′-CTGCCTATGTGCTATATCCATGG-3′), 456C(5′-GTAGATGACTGAGTGTGTAGG-3′) and 485B(5′-GGTTCCTCGATCAAACTCTGG-3′), was amplified in an MJR thermocycler for 35 cycles of 30 seconds at 94°C, 20 seconds at 60°C and 1 minute at 72°C and produced a wild-type band of 250 bp and a mutant band of 600 bp.

E15.5 paint-filled ears were prepared as described(Morsli et al., 1998). Otic and hindbrain marker gene expression was detected in somite-stage matched E9.5-10.5 (the day of vaginal plug detection was regarded as E0.5) specimens by whole-mount in situ hybridization with digoxigenin-labeled antisense cRNA probes as described (Henrique et al.,1995). Hybridization probes were generated from plasmids containing: Pax2 (Dressler et al., 1990), Dlx5(Depew et al., 1999), Gbx2 (Wassarman et al.,1997), Hmx3 (Nkx5.1)(Merlo et al., 2002), Lfng (Morsli et al.,1998), Fgf10 (Xu et al., 1998), Wnt1 and Wnt3a(Parr et al., 1993), Wnt2b (Grove et al.,1998), Bmp4 (Jones et al., 1991), Otx2(Simeone et al., 1993) Mafb (Cordes and Barsh,1994) and Gata3(George et al., 1994). Stained embryos were cryoprotected in sucrose and cryosectioned as described(Stark et al., 2000). Two to six embryos of each genotype were analyzed with each probe.

Mice were anesthetized using 0.02 ml/g avertin. Auditory brainstem response(ABR) thresholds for click stimuli (47 μsecond duration, 29.3/second)presented to each ear individually, were determined using high-frequency transducers controlled and analyzed by SmartEP software (Intelligent Hearing Systems) according to Zheng et al. (Zheng et al., 1999).

We showed previously that Fgf3^neo/neo mice have defects of inner ear morphogenesis that are incompletely penetrant and variably expressive. The earliest morphologic defect was a failure of EDS outgrowth between E9.5 and E10.5 (Mansour et al.,1993). Here we present a more detailed analysis of the inner ear dysmorphology in Fgf3 mutants and an analysis of otic marker genes using the previously described Fgf3^neo allele and a newly targeted Fgf3^Δ2 allele(Fig. 1A).

To determine whether the new Fgf3^Δ2 allele was similar to the Fgf3^neo allele, we first compared the phenotypes of adult homozygotes. Offspring of Fgf3^+/neo and Fgf3^+/Δ2 intercrosses were observed and genotyped at weaning. In both cases, although all three genotypes were recovered, the genotypic distribution deviated significantly from the normal mendelian expectation (Fig. 1B). Only 10% of Fgf3^+/neo and 9% of Fgf3^+/Δ2 intercross offspring were homozygous mutant, showing that there is significant and similar lethality associated with both alleles. In addition, all homozygous mutants exhibited the characteristic short, curly tails described previously(Fig. 1B)(Mansour et al., 1993). Thus,the two Fgf3 alleles behave similarly at the overt phenotypic level.

Some of the previously described Fgf3^neo/neo animals had an abnormal Preyer's reflex and exhibited circling behavior, which suggested auditory and vestibular dysfunction(Mansour et al., 1993). Therefore, we measured auditory brainstem response (ABR) thresholds in each ear in 6-week-old animals of all three Fgf3^Δ2 genotypes. No significant differences in auditory thresholds were observed between wild-type(n=7) and heterozygous (n=11) animals and all of these controls had normal behavior. By contrast, only one Fgf3^Δ2/Δ2animal had normal thresholds in both ears and eight of nine Fgf3^Δ2/Δ2animals had ABR thresholds that were significantly elevated (>2 standard deviations above same-side controls) in one (n=3) or both(n=5) ears (Fig. 1C). All five bilaterally affected homozygous mutants showed circling behavior,whereas the three unilaterally affected animals had only a slight head tilt. Histological sections of normal-hearing heterozygous inner ears showed typical gross morphology and normal cochlear structure(Fig. 1D,D′). By contrast, inner ear sections of a deaf homozygote showed only rudimentary partitioning into distended cochlear and vestibular chambers. Nevertheless,the otic epithelium could be identified and showed evidence of sensory differentiation, albeit abnormal (Fig. 1E,E′). Thus, as was concluded using morphologic studies of Fgf3^neo homozygotes(Mansour et al., 1993), Fgf3^Δ2 homozygotes have auditory and vestibular phenotypes that are incompletely penetrant and variably expressive.

To visualize more clearly the consequences of loss of Fgf3 to inner ear morphogenesis, we filled E15.5 Fgf3 control(Fig. 2A,B) and mutant(Fig. 2C-I) inner ear epithelia with latex paint. Both Fgf3 alleles were studied, but no obvious differences between them were apparent, so the results for a total of 120 mutant ears were combined. At the gross anatomical level, 50/120 (42%) of mutant ears had a normal morphology, consistent with incomplete penetrance of the adult Fgf3 auditory and vestibular phenotypes(Fig. 2C). The remaining 70(58%) showed graded morphologic defects that were assigned to groups based on the classification of Gbx2-null ears(Lin et al., 2005).

Virtually all affected Fgf3 mutant ears had a distended membranous labyrinth (n=65/70). In most cases, the swollen cochlear duct showed incomplete partition and poor coiling. All of the expected structures could be identified in type Ia ears, but the endolymphatic duct appeared truncated dorsally, lacking the endolymphatic sac(Fig. 2D, n=8/70). Type Ib ears were similar to type Ia, but entirely lacked the EDS appendage(Fig. 2E, n=12/70). Type II was the most common phenotype seen in affected Fgf3 mutant ears and included absence of the EDS as well as of the common crus, such that the anterior and posterior canals appeared as a single continuous canal(Fig. 2F, n=32/70). In addition, the utricle and saccule were fused with the enlarged cochlear duct. Smaller numbers of mutant ears were even more severely affected. Type III ears had the EDS and common crus aplasia typical of type II ears, together with absence or severe truncation of the anterior and posterior semicircular canals(Fig. 2G, n=5/70). Most of the associated ampullae could be identified, but one ear lacked the posterior ampulla and another lacked the anterior ampulla. These ears all retained the lateral canal and ampulla. The type III cochlear duct was severely malformed and shortened, with less than one coil. The type IV phenotype was the most severe, with defects in all inner ear structures. The only discernible formation was a cochlea-like structure(Fig. 2H, n=7/70). Taken together, these data show that the abnormal morphologies of most Fgf3 mutant ears included hypoplasia or aplasia of the EDS, swelling of the remaining labyrinth and, in many cases, aplasia of the common crus. These are similar to the phenotypes seen in Hoxa1, Mafb and Gbx2 mutant ears (Pasqualetti et al., 2001; Lin et al.,2005; Choo et al.,2006).

A few affected Fgf3 mutant ears retained a normal endolymphatic appendage and were not distended. These were grouped into type O(n=5/70). Three type O ears lacked the posterior canal and ampulla(Fig. 2I). One type O ear lacked the lateral canal and one had an abnormally thickened and posteriorly positioned common crus (data not shown). Thus, type O ears have dysmorphologies that are also seen in Fgf10- and Fgfr2b-null mutants (Pirvola et al., 2000; Pauley et al., 2003; Ohuchi et al., 2005) and are similar to those of Fgf10 hypomorphs (X.W., T.J.W. and S.L.M.,unpublished).

To correlate Fgf3 expression with major steps in early otic morphogenesis, we detected Fgf3 transcripts by in situ hybridization of E8.5-10.5 embryos. Consistent with results described previously, at 3-4 somites Fgf3 was expressed weakly in a dorsoventrally oriented stripe within the presumptive otic placode and strongly in the adjacent neurectoderm(data not shown) (but see McKay et al.,1996; Wright and Mansour,2003). Ectodermal expression of Fgf3 disappeared as the placode was induced and formed a cup (5-12 somites, data not shown), whereas neurectodermal expression remained strong throughout the otic cup stage and was restricted to r5-6 (Fig. 3A-D) (Mahmood et al.,1996; McKay et al.,1996; Wright and Mansour,2003). As the cup closed to form the otic vesicle, hindbrain Fgf3 became restricted to r6, expression was induced in an anteroventrolateral region of the vesicle and in delaminating neuroblasts of GVIII (Fig. 3E-H). EDS outgrowth was first evident morphologically at 27 somites, just as r6 expression of Fgf3 diminished(Fig. 3I-L). At E10.5, Fgf3 expression was maintained in anteroventrolateral otic cells from which both neuroblasts and sensory cells are derived(Fig. 3J,K)(Li et al., 1978; Carney and Silver, 1983). The neuroblasts of GVIII, however, no longer expressed Fgf3. Thus, Fgf3 shows a dynamic pattern of expression in the hindbrain, GVIII and prospective sensory domains during the transition from the otic cup stage to the initiation of EDS outgrowth.

The disruption of EDS development in the majority of abnormal Fgf3mutant ears suggested that molecular patterning of the dorsal otocyst could be perturbed. Thus, we analyzed expression of several dorsal otic markers at E9.5-10.5, the stage when otic vesicle morphogenesis is initiated by EDS outgrowth and when Fgf3 expression is found in the hindbrain and initiates in the prosensory domain (Fig. 3). Again, both Fgf3 mutant alleles were examined, but no allele-specific differences were apparent, so the results were combined(Fig. 4).

First, we analyzed genes whose mutant phenotypes included an EDS abnormality (Gbx2 and Dlx5)(Depew et al., 1999; Lin et al., 2005) and/or were expressed in the developing EDS (Wnt2b). Gbx2 expression,which marks a broad dorsomedial region of the vesicle, including the prospective EDS, as well as more anterior and posterior dorsal otic tissue,was generally not affected in E9.5 Fgf3 mutants(Fig. 4A-B′). Only one such Fgf3 mutant showed a slight reduction in the extent of the medial Gbx2 domain (2/8 mutant ears, data not shown). The dorsomedial Gbx2 expression domain persisted in E10.5 control inner ears(Fig. 4C,C′). By contrast, otic Gbx2 expression was strongly affected by E10.5 in the Fgf3 mutant otocysts, which were noticeably smaller than control otocysts (Fig. 4D,D′). In eight of 12 Fgf3 mutant ears, no Gbx2 staining was found in the otic epithelium (Fig. 4D,D′, right ear), and in the remaining four ears, weak staining was detected in the dorsomedial wall of the otocyst; however, the expression domain was severely reduced(Fig. 4D′, left ear). At E10.5, Gbx2 was also expressed in the mid-hindbrain region(Lin et al., 2005). This Gbx2 domain was not affected in any Fgf3 mutant(Fig. 4C-D′). At E9.5, Dlx5 was found in the dorsomedial otocyst of both controls and Fgf3 mutants (Fig. 4E-F′). At E10.5, control otocysts showed initiation of EDS outgrowth and expression of Dlx5 in the dorsal half of the otocyst(Fig. 4G,G′). However, in Fgf3 mutants there was a slight reduction in the ventral boundary of the Dlx5 domain, both medially and laterally, possibly owing to physical loss of the prospective EDS domain (n=6/6 ears; Fig. 4H,H′). Wnt2b was first detected in E10.5 control otocysts in a dorsomedial domain, marking the outgrowing EDS (Fig. 4I,I′). By contrast, Wnt2b was entirely absent from the dorsal otocyst in four of ten Fgf3 mutant ears, and was severely reduced, showing only weak staining on the medial otocyst wall, in six of ten mutant ears, suggesting a failure of EDS specification(Fig. 4J,J′).

In addition, we studied Hmx3 (Nkx5.1), a marker that broadly identifies the vestibular domain and is required redundantly with Hmx2 (Nkx5.2) for vestibular development(Wang et al., 1998; Wang et al., 2001; Wang et al., 2004). Hmx3 was expressed dorsolaterally in E10.5 control otocysts(Fig. 4K,K′), marking the prospective EDS and lateral canal plate regions. In Fgf3 mutants,however, Hmx3 expression shifted medially (n=8/8 ears; Fig. 4L,L′). Finally, Gata3, which is expressed in both dorsolateral and ventromedial otocyst domains (Lawoko-Kerali et al.,2002; Lillevali et al.,2004) and is required for both dorsal and ventral otic development(Karis et al., 2001; Lillevali et al., 2006), was unaffected in E10 Fgf3 mutants (data not shown), suggesting that its role in otic morphogenesis is independent of Fgf3.

As the cochlear ducts in affected E15.5 Fgf3 mutant inner ears were greatly enlarged and poorly coiled, we analyzed the expression patterns of ventral otic markers. Pax2 marks the ventromedial region of the otocyst and is important for the development of the cochlear duct(Torres et al., 1996; Burton et al., 2004). Pax2 expression was similar in control and Fgf3 mutant embryos at E9.5 (Fig. 5A-B′) and E10.5 (Fig. 5C-D′). At E10.5, Otx2 is normally expressed in a posterior ventrolateral patch (Morsli et al., 1998) (Fig. 5E,E′). In most Fgf3 mutant ears, Otx2expression was unchanged (n=4/6 ears; Fig. 5F,F′). However, in one specimen, Otx2 was reduced in one ear and absent in the other(data not shown). Finally, ventromedial Gata3 expression was unaffected in Fgf3 mutants (data not shown). Thus, it is likely that Fgf3 mutant cochlear phenotypes are largely secondary to disturbances of dorsal patterning and morphogenesis.

Expression of Fgf3 in the anteroventrolateral otocyst domain that gives rise to sensory cells and GVIII neuroblasts, together with the reduced size of GVIII in E11.5 Fgf3^neo mutants, suggested roles for Fgf3 in sensory organ and ganglion development(Mansour et al., 1993). Therefore, we assessed the expression of sensory markers, Lfng and Bmp4, and of a sensory and ganglion marker, Fgf10, in Fgf3 mutant otocysts.

Lfng is normally expressed in the anteroventral otocyst - in the precursors of the cochlear, saccular and utricular sensory organs(Morsli et al., 1998). No significant differences in the distribution of Lfng-expressing cells were detected between control (Fig. 6A-A″) and mutant (Fig. 6B-B″) specimens at E10.5. Bmp4 marks the presumptive cristae (Morsli et al.,1998). At E10.5, Bmp4 expression was detected in two patches: one anterodorsolateral (Fig. 6C,C′) and the other posteroventral(Fig. 6C,C″). The anterior Bmp4 patch contains precursors of the anterior and lateral cristae, whereas the posterior Bmp4 domain(Fig. 6C, arrowhead) houses posterior crista precursors (Morsli et al., 1998). The anterior Bmp4 domain was present in all E10.5 Fgf3 mutant embryos (Fig. 6D,D′); however, there was a reduction in the extent of the staining and a ventral shift of that domain. The posterior Bmp4domain was not detected in four of six mutant inner ears(Fig. 6D,D″). Thus, loss of Fgf3 can affect posterior crista development.

To assess GVIII development together with the prospective otic sensory domains, we examined Fgf10 expression. At E9.5, Fgf10 was expressed throughout both control and Fgf3 mutant otocysts (data not shown). At E10.5, Fgf10 was detected in the anterior pole of control vesicles, which includes the prospective sensory domain, and in delaminating neuroblasts of GVIII. Fgf10 was also expressed in a second, smaller posterior patch, which gives rise to the posterior crista(Fig. 6E-E″)(Pauley et al., 2003). The posterior Fgf10 domain appeared absent in eight of 12 Fgf3mutant ears (Fig. 6F,arrowhead), but examination of sections showed that it was shifted anteriorly and medially (Fig. 6F″). There was also a dorsomedial shift of the Fgf10-expressing anterior domain together with a corresponding positional shift of GVIII in six of 12 mutant ears (Fig. 6F′).

WNT signaling from the dorsal neural tube is required for dorsal otocyst patterning, with Wnt1 and Wnt3a identified as the specific ligands (Riccomagno et al.,2005). We therefore examined Wnt1 and Wnt3aexpression in Fgf3 mutants. As expected at E9.5, both Wnt1and Wnt3a expression marked the dorsal-most region of the developing neurectoderm in control embryos (Fig. 7A,A′,C,C′) (Parr et al., 1993). Wnt1 expression was unchanged in all three Fgf3 mutants examined (Fig. 7B,B′). Similarly, Mafb expression in r5-6 was unaffected in Fgf3 mutants (data not shown). By contrast, all six Fgf3 mutants examined showed a significant ventral expansion of Wnt3a expression in r5-6 (Fig. 7D,D″), but not in hindbrain regions anterior(Fig. 7D′) or posterior(Fig. 7D‴) to the otic vesicle. These results show that Fgf3 negatively regulates Wnt3a, but not other genes required for otic development, in the dorsal hindbrain.

This study provides new morphologic data on aberrant inner ear morphogenesis in Fgf3 mutants. These data, together with an analysis of changes in molecular patterning of Fgf3 mutant otocysts and a comparison of similar studies of other mutants that affect inner ear morphogenesis, allow us to propose a placement for Fgf3 in the genetic cascade leading to EDS formation and draw parallels with human inner ear dysmorphogenesis and deafness.

Although homozygotes for another Fgf3-null allele(Fgf3^tm1Sng, MGI:3027990) from which all coding sequences were removed, failed to show any inner ear abnormalities(Alvarez et al., 2003), we found that both Fgf3 alleles examined here gave similar otic phenotypes. The phenotypic differences between our alleles and those of Alvarez and colleagues could arise either from differences in genetic background, or from the particular nature of the mutations, the contributions of which will require additional experiments.

We found that otic epithelia from two Fgf3 mutant alleles exhibited a graded series of morphologies ranging from normal to complete loss of all vestibular structures. Common to all affected Fgf3 mutant ears, except for those few members of the type O group, was hypoplasia or aplasia of the EDS, global swelling of the entire epithelium and poor cochlear coiling. Since the expression of ventral otocyst genes required for cochlear development was largely unperturbed, it is likely that the gross distortion of cochlear development was secondary to the effects on EDS and dorsal development. The association between abnormal EDS development and otic epithelial swelling is characteristic of both syndromic and non-syndromic human hearing loss (Wu et al.,2005) and is common to mouse mutations that disrupt the hindbrain signaling required for EDS induction (e.g. Mafb, Hoxa1, Wnt1/Wnt3adouble mutants) (Pasqualetti et al.,2001; Riccomagno et al.,2005; Choo et al.,2006) or disrupt transcription factors that are expressed early in the dorsal otic vesicle and are required for EDS formation (e.g. Gbx2,Hmx2, Hmx3, Dlx5, Dlx5/6) (Wang et al., 2001; Merlo et al.,2002; Lin et al.,2005). Furthermore, mutations in Foxi1, which encodes an EDS-expressed transcription factor, or in its genetic target, Slc26a4(encoding the anion transporter, pendrin), cause abnormal EDS function,swelling of the entire epithelial space, loss of endocochlear potential and deafness (Everett et al.,2001; Hulander et al.,2003). Indeed, mutations in SLC26A4 are a common cause of syndromic deafness (Pendred syndrome) and have also been found in non-syndromic deafness (DFNB4) (Morton and Nance, 2006). These patients show an enlarged vestibular aqueduct(the structure that houses the EDS) and may show the Mondini defect(dilatation and incomplete partitioning of the cochlear duct), similar to the Fgf3 type Ia ears described here(Phelps et al., 1998; Fitoz et al., 2007). Thus, it appears that any disruption of EDS development, whether leading to EDS hypoplasia or aplasia, as when hindbrain-expressed or early otic-expressed genes are disrupted, or to overall dilatation of a labyrinth that retains an EDS, as in the case of later-acting EDS-expressed genes, can cause deafness. The EDS is thought to participate in regulation of endolymph homeostasis,which is crucial for hearing function in the mature ear, but the phenotypes exhibited by Fgf3 and the other mutants with similar dysmorphologies suggest that EDS function might be required from the earliest stages of otic morphogenesis. It is interesting to speculate that Meniere's disease, an adult condition characterized by both auditory and vestibular symptoms and associated with enlarged endolymphatic spaces, could be caused either by mutations in EDS-expressed genes that function only after completion of morphogenesis, or by mutations in `developmental' genes that, like Fgf3, have a wide spectrum of penetrance and expressivity.

EDS aplasia and general membranous swelling together with common crus aplasia was the most prevalent phenotypic complex (type II) observed in affected Fgf3 mutant inner ears and has also been noted in some Hoxb1, Mafb and Gbx2 mutants(Pasqualetti et al., 2001; Lin et al., 2005; Choo et al., 2006). Such a congenital phenotype has also been reported in patients(Manfre et al., 1997; Kim et al., 2004), but its cause is unknown. If EDS aplasia and consequent membranous swelling is explained by failure of FGF3-mediated signaling from the hindbrain to the dorsal otocyst, how might common crus aplasia arise? Since the common crus is formed when two distinct central regions of the vertical canal plate fuse and then resorb to form distinct anterior and posterior canals, one possibility is that Fgf3, expressed from the developing cristae, normally functions relatively directly to limit the fusion boundaries or the process of cellular resorption. Alternatively, it could be that the abnormal swelling of the epithelium consequent to failed or incomplete induction of the EDS is itself the cause of excessive cellular fusion and/or resorption in the vertical canal plate. Given the diversity of mutant genes that cause common crus aplasia, and the fact that most of these do not appear to be expressed in the otic epithelium, the latter possibility seems more likely. Conditional ablation of Fgf3 in the otic epithelium versus hindbrain (work in progress) could help to address this issue.

To understand the effects of the Fgf3 mutations at the molecular level, we surveyed expression of regionally restricted otocyst genes in mutant embryos at a stage just prior to the first morphologic sign of aberrant vesicle development - namely, at EDS induction. Dorsally expressed genes were affected (absent or shifted), whereas ventrally expressed marker genes were not strongly affected, suggesting that Fgf3 is a component of the dorsal determination pathway and EDS specification(Fig. 8). Just as the dysmorphologies of Fgf3 mutant ears were similar to those of Mafb, Hoxa1 and Gbx2 mutants, so too were the trends for most of the molecular changes. Since Mafb and Hoxa1 are required for Fgf3 expression in the hindbrain(Carpenter et al., 1993; Frohman et al., 1993; McKay et al., 1996; Pasqualetti et al., 2001), and both Mafb and Fgf3 are required for normal dorsal otic expression of Gbx2 (Choo et al.,2006) (this study), this places Fgf3 between Mafb and Gbx2 in the EDS induction pathway. However, because Gbx2 expression in Fgf3 mutants was only strongly affected at E10.5, and not at E9.5, when Gbx2 is affected in Wnt1/Wnt3a double mutants (which have no dorsal development at all)(Riccomagno et al., 2005), we suggest that WNTs provide the initiating signal for dorsal patterning of the otic vesicle, and that FGF3 directly or indirectly sustains or reinforces expression of dorsal otic genes. This might explain, at least in part, the reduced penetrance and severity of Fgf3 otic morphologic and molecular phenotypes relative to those of Gbx2.

Expression of other dorsal otic genes downstream of Gbx2 was also perturbed in Fgf3 mutants. Wnt2b, the gene that most specifically marks the developing EDS, but the otic function of which is unknown, is absent from Gbx2 mutant ears(Lin et al., 2005) and was strongly reduced or absent from Fgf3 mutant ears, consistent with the EDS defects in both mutants. Dlx5, which normally marks the entire dorsal half of the otocyst and is required for both EDS and semicircular canal development (Merlo et al.,2002), was also affected in both mutants, but in slightly different ways. In Gbx2 mutants, Dlx5 is lost from the entire dorsomedial domain (Lin et al.,2005), whereas in Fgf3 mutants, only the ventral-most region of both the dorsomedial and dorsolateral Dlx5 domains were absent. This suggests that Dlx5 might not be a direct target of FGF3 signaling, but is downstream of Gbx2 and in this instance is revealing the loss of the dorsomedial EDS domain and the consequent dorsal shifts of its most ventromedial and ventrolateral domains. Finally, Hmx3 is expressed in the dorsolateral otocyst and subsequently in the semicircular canals and is required for canal, but not for EDS, development(Wang et al., 2004). Hmx3 expression is unchanged in Gbx2 mutants(Lin et al., 2005) and slightly downregulated in Wnt1/Wnt3a double mutants(Riccomagno et al., 2005). By contrast, in Fgf3 mutants there is a medial expansion of Hmx3 expression, suggesting that Fgf3 might normally function directly to restrict expansion of Hmx3 into the medial otocyst or might function indirectly by restricting the Wnt3a domain dorsally.

Despite abnormal cochlear development in Fgf3 mutants, expression of Pax2 and Otx2, ventral otic genes downstream of SHH signaling and necessary for cochlear development(Morsli et al., 1999; Riccomagno et al., 2002; Burton et al., 2004), were largely unaffected, except for one case of loss of Otx2 expression. This was somewhat unexpected, given that both Mafb and Gbx2mutants show a medial expansion of otic Otx2 expression(Lin et al., 2005; Choo et al., 2006). However,since Fgf3 appears to be required to maintain rather than to initiate Gbx2 expression, it is possible that the early phase of Gbx2expression is sufficient to establish the Otx2 domain and that FGF3,on its own, is not a major player in cochlear development. The possibility for redundant roles with Fgf10, such as occurs during otic induction,remains and will be tested using conditional mutants.

The Bmp4 and Fgf10 posterior expression domains marking the developing posterior cristae were absent or shifted dorsomedially,respectively, in Fgf3 mutants. Although the vast majority of Fgf3 mutant ears had a phenotype most consistent with a loss of FGF3 signaling resulting in a failure to properly induce the EDS, a small number of affected mutants apparently underwent normal EDS induction, but nevertheless developed with abnormal or absent posterior semicircular canals (type O). This phenotype is consistent with previous suggestions that FGF signals from the developing sensory patches induce non-sensory (canal) development(Chang et al., 2004) and further suggests that Fgf3 might participate in posterior sensory domain specification. However, addressing this issue directly and assessing the relative contributions of Fgf3 and Fgf10 to posterior sensory and non-sensory development will require analysis of double conditional mutants in which the hindbrain requirement for Fgf3 in EDS induction is bypassed.

As in the original study of Fgf3^neo mutants, GVIII, as marked in this study by Fgf10 expression, was notably smaller in affected mutant ears. We also consistently observed a dorsal shift of the ganglion in affected ears, as if either the neurogenic domain itself or the expression domain for an otocyst-derived attractant for delaminating GVIII neuroblasts had shifted. It is interesting that a similar dislocation of GVIII and the Fgf3 expression domain was noted in studies of Mafbmutants (McKay et al.,1996).

Our results show that Fgf3 negatively regulates Wnt3a in the region of r5-6, preventing a ventral expansion of Wnt3atranscripts. This suggests an important role for Fgf3 not only in reinforcing the inductive effect of WNT signals on the dorsal otocyst, but also in regulating localization of WNT signals to the dorsal-most region of the neural tube. Since WNT proteins are unlikely to diffuse over long distances, this localization may serve to limit the effects of WNTs to the dorsal otic vesicle.

FGF/WNT cross-talk is emerging as an important mechanism regulating various biological processes (Dailey et al.,2005), including brain, tooth and kidney development(Moon et al., 1997). In some cases, the parallel activation of FGF and WNT pathways causes developmental changes different from the individual effects of each factor, as exemplified by FGF/WNT interactions to specify neural and epidermal fate in the chick epiblast (Wilson et al.,2001). Evidence to support the role for FGFs in modulating WNT signaling comes from mouse genetic experiments showing that cross-regulation of FGF and WNT signaling is fundamental to normal skull development, when FGFs inhibit WNTs and subsequently osteoblast differentiation(Dailey et al., 2005). There are additional roles for cross-talk between the WNT and FGF pathways in otic development. During chick otic induction, mesodermal FGF19 induces neurectodermal expression of Wnt8c(Ladher et al., 2000). Both types of signals also participate in otic placode induction(Ladher et al., 2000; Wright and Mansour, 2003; Ladher et al., 2005), with WNT signals needed for stabilizing the otic placode cell state by enhancing and sensitizing the response of ectoderm to inductive FGF signaling(Ohyama et al., 2006). Given that WNT and FGF signaling pathway components continue to be expressed in the inner ear epithelia during the later stages of otic morphogenesis and sensory organ patterning, and that roles for each pathway are beginning to be defined(Pirvola et al., 2002; Pirvola et al., 2004; Wang et al., 2006; Hayashi et al., 2007), it will not be surprising if further intersections between these pathways are uncovered in the ear.

We thank Doris Wu for paint-fill instruction. cDNA clones newly obtained for this study were generously shared by Drs Eva Bober [Hmx3(Nkx5.1)], Doris Wu (Lfng), Andy McMahon (Wnt1),Jeffery Barrow (Wnt3a), Elisabeth Grove (Wnt2b), Mario Capecchi (Bmp4), Laure Bally-Cuif (Otx2) and Doug Engel(Gata3). Fred Grimmer provided helpful discussions of human inner ear dysmorphogenesis, Lisa Urness provided the Fgf3-hybridized embryos and Diana Lim created the model (Fig. 8). This manuscript was improved by suggestions from Lisa Urness and Gary Schoenwolf. This work was supported by grants from the NIDCD(R01DC05608 and R01DC04185) and by NIH Genetics Training Grant T32GM07464.