Zebrafish Foxi1 provides a neuronal ground state during inner ear induction preceding the Dlx3b/4b-regulated sensory lineage

The vertebrate inner ear mediates hearing and balance through a complex arrangement of mechanosensory hair cells, nonsensory supporting cells and sensory neurons. All of these cell types derive from the otic placode, a transient ectodermal thickening adjacent to the developing hindbrain (Barald and Kelley, 2004; Fritzsch et al., 2006). Otic placode formation is a multistep process initiated by the establishment of the preplacodal region, which surrounds the anterior neural plate and contains precursors for all sensory placodes (Streit, 2007). Subsequently, the otic-epibranchial progenitor domain (OEPD), a common territory for otic and epibranchial precursors, is specified (Ladher et al., 2010; Chen and Streit, 2012). In zebrafish, the OEPD also appears to contain the progenitors of the anterior lateral line ganglion (McCarroll et al., 2012). Studies in various species have shown that OEPD formation is triggered by Fibroblast growth factor (Fgf) ligands secreted by the hindbrain and subjacent mesendoderm (Phillips et al., 2001; Léger and Brand, 2002; Maroon et al., 2002; Alvarez et al., 2003; Wright and Mansour, 2003; Ladher et al., 2005). In response to these signals, cells express Pax8 and Pax2, two members of the Pax2/5/8 transcription factor family, which are crucial regulators of OEPD formation and proper inner ear development (Brand et al., 1996; Pfeffer et al., 1998; Hans et al., 2004; Mackereth et al., 2005; Bouchard et al., 2010; Freter et al., 2012). The subsequent segregation of otic and epibranchial progenitors is mediated by Wnt signaling in a Pax2-dependent manner (Freter et al., 2008; McCarroll et al., 2012). With respect to otic fate, a two-phase model has been proposed to summarize genetic interactions during otic induction in zebrafish (Hans et al., 2004; Solomon et al., 2004). According to this model, the forkhead transcription factor Foxi1 enables expression of Pax8 during the early phase, and the homeodomain transcription factors Dlx3b and Dlx4b (Dlx3b/4b) provide competence to activate Pax2a during the second phase (Hans et al., 2004; Solomon et al., 2004). Previous work has shown that dlx3b, dlx4b and foxi1 are regulated initially independently in a BMP-dependent manner in the same region at late gastrula stages and whereas foxi1 is progressively restricted to the presumptive OEPD, dlx3b and dlx4b are maintained in a stripe corresponding to cells of the preplacodal region (Akimenko et al., 1994; Ellies et al., 1997; Nissen et al., 2003; Solomon et al., 2003; Solomon et al., 2004; Hans et al., 2007). Subsequently, downregulation of foxi1 in a Pax2a- and Pax8-dependent manner is required for proper otic development, whereas dlx3b and dlx4b are maintained in the cells of the future otic placode (Akimenko et al., 1994; Padanad and Riley, 2011). Loss of Foxi1 or Dlx3b/4b results in compromised otic induction and development of smaller otic vesicles (Solomon and Fritz, 2002; Liu et al., 2003; Nissen et al., 2003; Solomon et al., 2003), and combined loss of both factors eliminates all indications of otic specification (Hans et al., 2004; Solomon et al., 2004). However, the functional differences of Foxi1 and Dlx3b/4b during otic induction have not been addressed thus far and remain elusive.

After placode formation, otic tissue develops into the otic vesicle where sensory neurons and mechanosensory hair cells are born. The neuronal precursors delaminate as neuroblasts from an anterior-ventral position in the otic vesicle and give rise to the eighth cranial or statoacoustic ganglion, whereas hair cells are generated in the sensory epithelia of the otic vesicle (Haddon and Lewis, 1996; Rubel and Fritzsch, 2002). It is well established that neuroblast and hair cell formation require the activation of the proneural proteins Neurogenin 1 (Neurog1) and Atonal homolog 1 (Atoh1) (Ma et al., 1998; Bermingham et al., 1999; Andermann et al., 2002; Fritzsch et al., 2010). However, despite an understanding of proneural gene function and extensive research on otic placode induction, upstream events, including the underlying regulation of Atoh1 and Neurog1 expression in sensory and neuronal precursors, are just beginning to be understood. In zebrafish, it was shown that Dlx3b/4b is required for proper expression of atoh1a and atoh1b and subsequent formation of mechanosensory hair cells (Millimaki et al., 2007). However, clonal analyses have indicated only a limited relationship between neurons and hair cells with regional differences within the developing inner ear (Satoh and Fekete, 2005; Abello and Alsina, 2007; Sapède et al., 2012). Currently, it is unknown whether lineage commitment followed by proneural activation of Neurog1 and Atoh1 occurs in separate neuronal and sensory domains, or in a common neurogenic domain.

Here, we demonstrate that the competence to enter a neuronal or sensory fate is provided sequentially by Foxi1 and Dlx3b/4b very early during inner ear development. Initially, Foxi1, a known otic competence factor, induces the OEPD and provides competence to enter a neuronal fate. Consequently, inactivation of Foxi1 does not only result in compromised otic vesicle formation, but in a loss of all neuronal OEPD derivatives, including the statoacoustic, anterior lateral line and epibranchial ganglia, which has been reported previously (Lee et al., 2003), Subsequently, Dlx3b/4b, which is able to provide otic fate in the absence of Foxi1, restricts neuronal and promotes sensory fate.

All animal experiments were conducted according to the guidelines and under supervision of the Regierungspräsidium Dresden (permit AZ 24D-9168.11-1/2008-1 and AZ 24D-9168.11-1/2008-4). All efforts were made to minimize animal suffering and the number of animals used.

Zebrafish were raised and maintained as described previously (Brand et al., 2002). Zebrafish embryos were obtained by natural spawning of adult fish and staged according to hours post-fertilization (hpf) or standard criteria (Kimmel et al., 1995). The wild-type line used was AB. The conditional red-to-green reporter Tg(hsp70l:loxP-DsRed-loxP-EGFP), the DsRed gene trap into the pax8 locus in pax8^nia03Gt, the null allele foxi1^em1 and the deletion of dlx3b, dlx4b and sox9a in Df^b380 have been described previously (Fritz et al., 1996; Solomon et al., 2003; Ikenaga et al., 2011; Kroehne et al., 2011). For germ-line transformation, the pTol pax2a:CreER^T2 plasmid was injected into fertilized eggs, embryos were raised to adulthood and crossed to AB wild-type fish, and the resulting F1 embryos were screened by PCR as previously described (Hans et al., 2009).

Antibodies used were anti-GFP (Molecular Probes; 1:500) and peroxidase-conjugated goat anti-rabbit IgG (whole molecule) (Sigma: A0545; 1:500). cDNA probes that detect the following genes were used: neurog1 and neurod (Andermann et al., 2002); stm (Söllner et al., 2003); cdh6 and cdh10 (Liu et al., 2006); myo7aa (Ernest et al., 2000); atoh1b (Adolf et al., 2004); tbx1 (Piotrowski et al., 2003); pax2a (Krauss et al., 1991); phox2a and phox2bb (Guo et al., 1999); tlx3b (Langenau et al., 2002); and CreER^T2 (Hans et al., 2009). Antibody staining, probe synthesis and in situ hybridization were performed as previously described (Westerfield, 2000).

Morpholino oligomers (MOs) were obtained from Gene Tools. MOs for neurog1 and foxi1 were described previously (Andermann et al., 2002; Solomon et al., 2003). For tamoxifen (Sigma, T5648) treatment a 50 mM stock solution in DMSO was made and stored at -20°C. At early gastrulation stages (6 hpf), embryos, still in their chorions, were transferred into Petri dishes containing 0.5 μM tamoxifen solution and incubated in the dark until heat treatments were conducted at the indicated stages. For subsequent heat-shock treatment, embryos were transferred into fresh Petri dishes. After removal of excess embryo medium, 42°C embryo medium was added, incubated for 0.5 hours at 37°C, returned to 28.5°C and fixed at 24 hpf.

Previous fate-mapping experiments using Kaede and caged fluorescein-dextran have shown that the Pax2-positive OEPD contributes to the otic vesicle, the epibranchial and presumably the anterior lateral line ganglia (McCarroll et al., 2012). Similarly, genetic lineage tracing in mice using the Cre/loxP recombination system revealed that OEPD Pax2-expressing cells give rise to the otic placode, as well as to epidermis (Ohyama and Groves, 2004b). Recently, we showed that Cre/loxP is also highly efficient in zebrafish (Hans et al., 2009; Hans et al., 2011; Kroehne et al., 2011). In common genetic lineage-tracing analyses, reporter expression is usually driven by a constitutive promoter and directly linked to Cre-mediated recombination. Consequently, all Cre-positive cells activate the reporter and indicate the cell fates of the entire Cre domain at later stages (Fig. 1A,B). By contrast, we devised a novel Cre-mediated lineage-tracing method called pioneer tracking (PioTrack) that allows fate mapping specifically of the first Cre-expressing cells of a nascent Cre domain, if the domain subsequently expands by de novo Cre expression in neighboring cells. PioTrack employs a conditional promoter, which can disconnect Cre-mediated recombination from reporter activation. We used the zebrafish temperature-inducible hsp70l promoter, which is expressed strongly and ubiquitously only during heat treatment (Halloran et al., 2000). Application of heat during the early stages of a nascent Cre domain results in reporter expression in only those cells that have undergone recombination, and persistence of the reporter protein reveals the fates of these cells at later stages (Fig. 1C). Application of heat at later stages, after Cre is expressed throughout the entire domain, activates reporter expression and indicates cell fates of the entire Cre domain in a manner similar to common genetic lineage tracing (Fig. 1D). To use PioTrack in early otic development, we generated the transgenic line Tg(pax2a:CreER^T2)^#31 driving CreER^T2 in the OEPD, using a pax2a promoter fragment that faithfully recapitulates OEPD expression and has been used in previous studies (Picker et al., 2002; Hans et al., 2009; McCarroll et al., 2012). To confirm the utility of Tg(pax2a:CreER^T2)^#31, we performed in situ hybridizations with pax2a and CreER^T2 probes (supplementary material Fig. S1). Compared with pax2a, CreER^T2 is absent in the midbrain-hindbrain boundary and ectopic expression is found in rhombomeres 3 and 5 during early segmentation stages as previously reported (Picker et al., 2002; Hans et al., 2009). By contrast, CreER^T2 expression in the OEPD closely recapitulates the endogenous pax2a expression in a temporal and spatial manner (supplementary material Fig. S1A-D,F-I) as confirmed by double color in situ hybridization (Fig. 1E-G′). Only after otic placode formation does pax2a and CreER^T2 expression diverge when CreER^T2 is not downregulated in Tg(pax2a:CreER^T2)^#31 anteriorly to the otic placode (supplementary material Fig. S1E,J). We crossed Tg(pax2a:CreER^T2)^#31 with the conditional red-to-green reporter Tg(hsp70l:loxP-DsRed-loxP-EGFP) (Kroehne et al., 2011), applied tamoxifen to elicit immediate Cre-mediated recombination as soon as CreER^T2 is expressed, divided the embryos into different groups and provided heat treatments at various developmental stages (Fig. 1H). Using this strategy, we find that enhanced green fluorescent protein (EGFP)-labeled cells end up as delaminated neuroblasts and in an anterior-ventral position within the otic vesicle after heat treatments at early OEPD stages, including the 3- and the 6-somite stages (Fig. 1I,J). Heat treatments at late OEPD stages, including the 9-somite stage, enlarge the domain, but no labeled cells populate the posterior-dorsal region (Fig. 1K). By contrast, EGFP-labeled cells are present throughout the otic vesicle after heat treatment at placodal or vesicle stages (Fig. 1L; data not shown). Taken together, our results show that early OEPD pax2a-positive cells give rise predominantly to the anterior-ventral part of the otic vesicle, including the neurogenic region, but do not contribute randomly to the entire otic vesicle.

The finding that early pax2a-positive OEPD cells are spatially restricted within the developing inner ear suggests that neuronal fate might already be specified within the OEPD very early on. In zebrafish, the transcription factors Foxi1 and Dlx3b/4b play pivotal roles during otic induction, and Dlx3b/4b is essential for sensory lineage development (Nissen et al., 2003; Solomon et al., 2003; Millimaki et al., 2007). To test whether Dlx3b/4b is also required for the neuronal lineage during otic development, we examined neurogenic differentiation (neurod) and cadherin 6 (cdh6) expression (Andermann et al., 2002; Liu et al., 2006). In control embryos at 24 hpf, neurod is expressed in the anterior-ventral region of the otic vesicle and in delaminated neuroblasts, whereas cdh6 is expressed in delaminated neuroblasts only (Fig. 2A,E). At this stage, both neurod and cdh6 also label progenitors of the anterior lateral line ganglia (gALL) that are closely associated with the delaminated neuroblasts of the otic vesicle (Andermann et al., 2002; Liu et al., 2006). In the absence of Dlx3b/4b, neurod and cdh6 expression are slightly increased or unaffected (Fig. 2D,H; Table 1), rather than being reduced, despite smaller otic vesicles as indicated by starmaker (stm) expression, a marker of the entire epithelium of the otic vesicle (Söllner et al., 2003). By contrast, removal of Foxi1 activity (Fig. 2C,G), which also results in smaller otic vesicles, severely reduces or even abolishes expression of both neurod and cdh6 in a manner similar to the removal of Neurog1 (Fig. 2B,F), which is required for the development of all zebrafish cranial ganglia, including the statoacoustic and anterior lateral line ganglia (Andermann et al., 2002). At vesicle stages, homeo box (H6 family) 3 (hmx3) is also expressed in delaminated neuroblasts and in an anterior-ventral position of the otic vesicle; however, in contrast to neurod, hmx3 is already initiated in the anterior portion of the developing otic placode by the 12-somite stage (15 hpf), prior to expression of Neurog1 (Adamska et al., 2000). Compared with control embryos, loss of Neurog1 or Dlx3b/4b activity does not affect placodal hmx3 expression, whereas loss of Foxi1 completely abolishes hmx3 at placodal stages (Fig. 2I-L). We also readdressed the functions of Foxi1 and Dlx3b/4b in sensory lineage development using myosin VIIAa (myo7aa), a marker of sensory hair cells (Ernest et al., 2000), and atonal homolog 1b (atoh1b), a proneural gene required for hair cell formation (Adolf et al., 2004; Millimaki et al., 2007). In 24 hpf control embryos, myo7aa is expressed in discrete anterior and posterior domains of the otic vesicle corresponding to the prospective utricular and saccular maculae, which are indicated at the 12-somite stage by atoh1b expression (Fig. 2M,Q). Both myo7aa and atoh1b expression are present in the absence of Neurog1 or Foxi1, which, however, frequently results in the formation of only one sensory patch (Fig. 2N,O,R,S). By contrast, and as previously reported (Millimaki et al., 2007), removal of Dlx3b/4b activity results in the complete loss of myo7aa and atoh1b (Fig. 2P,T). We also examined T-box 1 (tbx1), which is expressed in the non-neurogenic otic territory during placodal and vesicle stages (Piotrowski et al., 2003; Radosevic et al., 2011). We found that Dlx3b/4b-depleted embryos at the 12-somite stage also show a complete absence of tbx1 compared with control embryos or embryos lacking Neurog1 or Foxi1 activity (Fig. 2U-X).

The observation that absence of Foxi1 frequently results in smaller otic placodes or vesicles containing only one sensory patch suggested that Foxi1 might regulate hair cell formation, either directly or indirectly, and we thus analyzed sensory lineage development in foxi1 mutants in more detail. In control siblings, a broad and robust atoh1b expression can be detected during early OEPD stages, and over time this single domain resolves into two smaller patches (Millimaki et al., 2007), which give rise to the prospective utricular and saccular maculae (supplementary material Fig. S2A,D,G). By contrast, all foxi1 mutants show reduced atoh1b expression at OEPD stages although the degree of reduction is highly variable (supplementary material Fig. S2B,C). Subsequently, a strong correlation of otic placode or vesicle size with sensory patch number can be observed with small or medium-sized placodes or vesicles displaying one or two sensory patches, respectively (supplementary material Fig. S2E,F,H,I). Thus, the neuronal lineage of the inner ear is regulated by Foxi1. By contrast, the sensory lineage is regulated by Dlx3b/4b and only indirectly by Foxi1; frequently, loss of Foxi1 reduces otic induction resulting in smaller atoh1b domains that do not separate into two domains during subsequent otic development.

To corroborate our results, we analyzed a chromosome deficiency called b380, which lacks the genomic loci of dlx3b, dlx4b and sox9a physically and the OEPD expression of pax2a genetically (Fritz et al., 1996; Liu et al., 2003). Homozygous b380 mutants completely lack otic placodes and fail to undergo otic vesicle morphogenesis, although a few residual cells, constituting an epithelial ball, express genes characteristic of the developing inner ear (Fritz et al., 1996; Liu et al., 2003). In contrast to Dlx3b/4b, Sox9a and Pax2a, which are all relevant to the second phase of the proposed otic induction model, expression of Foxi1 and Pax8, crucial for the first phase, are unaffected in b380 mutants (Solomon and Fritz, 2002; Hans et al., 2004). At the onset of OEPD induction, pax8 is strongly expressed but rapidly downregulated just prior to otic placode formation (Hans et al., 2004). However, persistence of DsRed protein in a gene trap in which the coding sequence of DsRed is inserted into the pax8 locus (Ikenaga et al., 2011) allows us to track Pax8:DsRed-positive cells to later stages. To confirm the utility of this gene trap, we performed in situ hybridization with probes for DsRed and pax8. Compared with pax8, pax8:DsRed expression is elevated and more stable owing to the SV40 polyA signal contained in the gene trap vector, but the temporal and spatial patterns closely recapitulate endogenous pax8 expression during otic development (supplementary material Fig. S3A-H). Live imaging of Pax8:DsRed-positive cells in wild-type embryos shows that OEPD Pax8-expressing cells give rise to the otic placode, which is morphologically visible at the 12-somite stage (Fig. 3A,E). Subsequently, Pax8:DsRed-positive cells label the entire otic vesicle and its associated neuroblasts, which have already delaminated by 24 hpf (Fig. 3C,G). By sharp contrast, Pax8:DsRed-positive cells are present but only loosely aggregated at the 12-somite or 24 hpf stages in b380 mutants that do not show any morphological signs of otic placode or vesicle formation (Fig. 3B,D,F,H). Thus, OEPD Pax8-expressing cells are maintained after otic induction in b380 mutants, and to examine their fates, we analyzed the expression of neurod, cdh6, hmx3, myo7aa, atoh1b and tbx1. As previously reported (Millimaki et al., 2007), b380 mutants fail to form hair cells, demonstrated by the lack of myo7aa due to the loss of atoh1b, and they also lack otic tbx1 expression in comparison with control siblings at 24 hpf or the 12-somite stage (Fig. 4M-R). By contrast, although no otic vesicle is formed, a large number of cells express neurod and cdh6 in b380 mutants (Fig. 4A,B,E,F). Furthermore, neurod expression within the otic vesicle, which is confined to an anterior-ventral position in control siblings, can be detected throughout the remaining epithelial ball in b380 mutants (Fig. 4A,B). At the 12-somite stage, expression of hmx3 can also be detected, although not in a discrete domain as in control siblings, but rather in a dispersed band of cells in b380 mutants (Fig. 4I,J). To demonstrate that the underlying pathways leading to the expression of neurod, cdh6 and hmx3 are equivalent in control siblings and b380 mutant embryos, we compromised the function of Neurog1 or Foxi1. We find that, similar to control embryos, loss of Neurog1 activity in b380 mutant embryos completely abolishes neurod and cdh6 but not hmx3 expression at 24 hpf or 12-somite stages, respectively (Fig. 4C,G,K). Consistently, depletion of Foxi1 in b380 mutant embryos results in the absence of neurod, cdh6 and hmx3 in the same manner as depletion of Foxi1 in control embryos at 24 hpf or the 12-somite stage (Fig. 4D,H,L). These results show that, in the absence of Dlx3b/4b and Sox9a, OEPD-dependent neurogenesis occurs and that the genetic pathways underlying OEPD-dependent neuronal development in b380 mutants are identical to those in control embryos.

Because previous studies have shown that the OEPD contributes to the statoacoustic, epibranchial and presumably anterior lateral line ganglia (McCarroll et al., 2012), we analyzed the identity of the neuronal progenitors in b380 mutants. At 24 hpf, the proneural gene neurogenin1 (neurog1) is expressed in a distinct pattern delineating the neurogenic placodes with strong expression in the progenitors of the anterior lateral line ganglion and in a subset of cells within the otic vesicle that will delaminate to form the statoacoustic ganglion (Andermann et al., 2002). In comparison with control or Dlx3b/4b-depleted embryos, neurog1 is expressed in the progenitors of the anterior lateral line ganglion in b380 mutants and a second strong expression domain is present despite the absence of the otic vesicle (Fig. 5A-C). To distinguish anterior lateral line and statoacoustic ganglion progenitors, we examined T-cell leukemia, homeobox 3b (tlx3b) and cadherin10 (cdh10) expression (Langenau et al., 2002; Liu et al., 2006). Whereas tlx3b is exclusively expressed in trigeminal and anterior lateral line ganglion progenitors, as well as in the developing hindbrain (Langenau et al., 2002), cdh10 is initially expressed in anterior lateral line ganglion progenitors, but is restricted to statoacoustic ganglion progenitors at 24 hpf (Liu et al., 2006). In b380 mutants, expression of cdh10 is reduced in comparison with control or Dlx3b/4b-depleted embryos at 24 hpf, whereas tlx3b appears increased in the absence of Dlx3b/4b as well as in b380 mutants compared with control embryos at 24 hpf (Fig. 5D-I). To identify epibranchial ganglion progenitors, we examined the expression of paired-like homeobox 2a (phox2a) and paired-like homeobox 2bb (phox2bb) (Guo et al., 1999). At 24 hpf, expression of phox2a is initiated in the progenitors of the geniculate ganglion (gVII) in Dlx3b/4b-depleted embryos, b380 mutants and control siblings (Fig. 5J-L). In comparison with control siblings at 48 hpf, phox2bb expression revealed that the petrosal (gIX) and nodose ganglion (gX) progenitors are also generated in a similar manner in the absence of Dlx3b/4b activity or in b380 mutants (Fig. 5M-O). Taken together, our results show that all neuronal derivatives of the OEPD, the statoacoustic, anterior lateral line and epibranchial ganglia progenitors, are generated in the absence of Dlx3b/4b or in Dlx3b/4b- and Sox9a-deficient b380 mutants.

Our results provide important insights into the acquisition of neuronal competence during inner ear development. In zebrafish, fate mapping of early OEPD cells using live imaging of a fluorescent reporter is challenging because endogenous pax2a expression is highly dynamic. The new conditional Cre recombination procedure we used here, PioTrack, importantly allows labeling of the first Cre-expressing cells of a nascent Cre domain owing to a conditional reporter that disconnects Cre-mediated recombination and reporter activation. The zebrafish temperature-inducible hsp70l promoter is inactive at normal temperatures and is only strongly and ubiquitously expressed during heat treatment (Halloran et al., 2000). Consequently, cells that have undergone recombination show reporter expression after heat treatment, and subsequent reporter persistence reveals the fate of these cells at later stages. By contrast, neighboring cells that initiate Cre expression after heat treatment undergo Cre-mediated recombination, but do not show reporter activity unless a second heat treatment is administered (Fig. 1). Furthermore, Cre-mediated recombination is a stochastic event, meaning that recombination does not happen in all Cre-expressing cells at the same time, and the probability of a recombination event increases as the concentration of Cre protein builds up over time (Nagy, 2000). Hence, our observation that EGFP-labeled cells are present throughout the otic vesicle only after heat treatment at placodal or later stages is due to delayed recombination and does not indicate that dorsal and posterior otic cells are added after otic placode formation. Using Kaede and caged fluorescein-dextran to label small groups of cells of the pax2a-positive OEPD showed that cells contributing to the otic vesicle are distributed throughout the pax2a-positive OEPD with some bias of anterolateral cells labeling the statoacoustic/anterior lateral line ganglia (McCarroll et al., 2012), which partly conflicts with our data. However, in our analysis, onset of pax2a expression regulated by Pax8 (Hans et al., 2004; Mackereth et al., 2005) correlates with the spatial distribution of labeled cells within the otic vesicle. Given that heterogeneous levels of Pax2a and Pax8 are found in the OEPD and high Pax2a and Pax8 levels correlate with otic fate (McCarroll et al., 2012), we assume that, in the Kaede and caged fluorescein-dextran approach, neighboring cells with differential Pax2a and Pax8 levels have been labeled that subsequently contribute to the entire otic vesicle. In this respect, our results are also consistent with the previous finding that high Pax2a and Pax8 levels correlate with otic fate (McCarroll et al., 2012), because early OEPD pax2a-positive cells always contribute to the anterior lateral line ganglia progenitors or otic lineage but never enter an epibranchial fate.

Previous studies in chick have shown that neural specification takes place only in the anterior region of the placode (Adam et al., 1998; Abelló et al., 2010), and our data indicate that the crucial determinative event occurs earlier, concurrent with OEPD induction. Previously the Foxi1-Pax8 pathway has been suggested to act as an early ‘jumpstart’ mediating the initial Fgf-dependent otic induction that occurs over a much shorter time period in zebrafish than in mice (Hans et al., 2004). We now show that in addition to otic induction, Foxi1 also provides competence to embark on a neuronal fate. Loss of Foxi1 results in a loss of pax8 and in a patchy and variable onset of pax2a expression in the OEPD, which might explain why foxi1 mutants develop small otic vesicles with no or just one otolith, whereas others have two small lumina each containing a single otolith (Nissen et al., 2003; Solomon et al., 2003). In contrast to this variable phenotype, compromised neurogenesis was robustly observed in all foxi1 mutants examined (Table 1). Size reduction of the otic vesicle cannot alone explain this finding, because neurogenesis is not impaired in the absence of Dlx3b/4b activity, which causes a similar otic vesicle size reduction. Furthermore, other OEPD-derived neuronal derivatives, i.e. the progenitors of the anterior lateral line ganglion, are also absent in foxi1 mutants. Taken together with previous findings showing that Foxi1 is required to specify OEPD-derived epibranchial progenitors (Lee et al., 2003), we propose that Foxi1 converts naive cells of the preplacodal region into the OEPD and provides neuronal competence for all OEPD-derived neuronal progenitors (Fig. 6). Expression analysis showed that at the end of gastrulation foxi1 and dlx3b, a marker of the preplacodal region, share the same medial border, but foxi1 extends further laterally (Solomon et al., 2003), supporting results that cells outside of the preplacodal region also contribute to epibranchial ganglia (Padanad and Riley, 2011; McCarroll et al., 2012). Subsequently, Dlx3b/4b, which is able to initiate otic induction in the absence of Foxi1, represses neuronal and imposes sensory fate by the activation of atoh1b. Consistently, it has been shown that neuronal differentiation is repressed after overexpression of Dlx3 (Pieper et al., 2012). However, although sensory hair cell formation is completely abolished and neuronal fate is increased in Dlx3b/4b-depleted embryos, nonsensory epithelial cells are still present and give rise to a size-reduced otic vesicle. This indicates that additional factors are required to repress Foxi1-mediated neuronal competence, and Sox9 is a likely candidate. Zebrafish sox9a and sox9b genes are duplicate orthologs of the human SOX9 gene and both genes are expressed in the otic placode (Chiang et al., 2001). In b380 mutants, sox9a is deleted and additional removal of Sox9b activity completely blocks all signs of otic specification and formation of the residual epithelial ball (Liu et al., 2003). It is unknown which OEPD-derived neuronal fate is still present in Sox9b-depleted b380 mutants. We now show that all neuronal progenitors of the OEPD, the statoacoustic, anterior lateral line and epibranchial ganglia, are established in b380 mutants. However, the number of statoacoustic ganglion progenitors is reduced whereas anterior lateral line progenitors are increased, indicating that in the absence of properly formed otic placode/vesicle, OEPD-derived cells embark on an anterior lateral line progenitor fate. Nevertheless, neurod expression within the otic vesicle, which is confined to an anterior-ventral position in wild-type embryos, is present throughout the remaining epithelial ball in b380 mutants, showing that all otic cells adopt a neuronal fate in this genotype. Cell death as well as proliferation might also play an important role for proper neuronal progenitor formation, because Pax2 has been shown to play an important role in proliferation in the OEPD (Freter et al., 2012).

We also conclude that Foxi1 has no direct role in sensory hair cell formation, although absence of Foxi1 frequently results in smaller otic vesicles containing only one sensory patch. foxi1 mutants display a highly variable otic phenotype that is foreshadowed by delayed and reduced expression of pax2a during OEPD induction (Nissen et al., 2003; Solomon et al., 2003). In wild-type embryos, sensory hair cell formation is initiated at OEPD stages highlighted by the expression of atoh1b, a proneural gene required for proper hair cell formation (Adolf et al., 2004; Millimaki et al., 2007). atoh1b expression is initiated prior to pax2a in a single prosensory domain that subsequently activates Delta-Notch feedback to split the domain into separate utricular and saccular primordia (Millimaki et al., 2007). In foxi1 mutants, however, the initial single prosensory domain is frequently severely reduced in size and fails to subdivide into two separate domains.

It will be interesting to learn whether there is a similar developmental pattern in mammals. Mouse Foxi1 is expressed and required at a much later stage of inner ear development (Hulander et al., 2003), but expression of two other mouse Foxi family members, Foxi2 and Foxi3, coincide spatiotemporally with the period of otic induction. In particular, Foxi3 is expressed in a broad ectodermal region before and during otic induction and is subsequently downregulated in the developing placode similar to Foxi1 downregulation in zebrafish (Ohyama and Groves, 2004a). Thus, analysis of Foxi3 mutant mice will be necessary to learn whether this developmental pattern extends to mammals. Furthermore, how neuronal competence downstream of Foxi1 is executed remains elusive. The SoxB1 transcription factor Sox3 is expressed early in the OEPD region of chick and zebrafish and regulates the formation of the sensory-neural domain of the otic vesicle (Sun et al., 2007; Dee et al., 2008; Abelló et al., 2010). Furthermore, in zebrafish, Foxi1, but not Dlx3b/4b, is required for proper sox3 expression, which is also unchanged in b380 mutant embryos (Sun et al., 2007). Although a recent study reports no perceptible deficiency in neuronal precursor formation in the absence of Sox3 (Padanad and Riley, 2011), the finding that SoxB1 proteins control neural differentiation by directly regulating Neurog1 expression makes Sox3 a promising candidate (Okuda et al., 2010).

We thank Drs Q. Liu and M. Westerfield for sharing reagents; M. Fischer and J. Michling for excellent zebrafish care; the members of the Brand laboratory for discussions; and S. Linn and Dr M. Westerfield for comments on the manuscript.