The deltaA gene of zebrafish mediates lateral inhibition of hair cells in the inner ear and is regulated by pax2.1

The inner ear consists of a complex series of chambers, each containing a specialized epithelium composed of sensory hair cells interspersed with support cells (Anniko, 1983; Lewis et al., 1985). Hair cells bear ciliary bundles that project into the lumen of the ear. Lateral displacement of the ciliary bundles due to sound vibrations or accelerational forces stimulates the hair cells to transmit neural signals associated with hearing or detection of motion and orientation. The role of support cells is less certain, but they appear to provide factors in trans required for normal hair cell function and survival (Riley and Grunwald, 1996; Riley et al., 1997; Haddon et al., 1998a), a role analogous to that of glial cells in the central nervous system. In addition, support cells can act as stem cells that divide asymmetrically to give rise to additional support cells and new hair cells (Corwin and Warchol, 1991; Presson et al., 1996). The latter function provides a mechanism for hair cell regeneration in adult tissues, but it remains to be established whether a similar process operates during embryonic development.

The first sensory epithelia to develop in the otic vesicle are the maculae, which are associated with dense crystalline otoliths that assist hair cells in sensing sound and linear acceleration (Haddon and Lewis, 1996). Subsequently, sensory epithelia called cristae develop in the semicircular canals, which primarily sense angular acceleration. Unlike maculae, cristae are not associated with otoliths.

Genetic studies in mouse and zebrafish have revealed a number of mutations affecting various stages of otic development, including induction of the otic placode, morphogenesis of the otic vesicle, and differentiation and function of sensory hair cells (Epstein et al., 1991; Lufkin et al., 1991; Chisaka et al, 1992; Mansour et al., 1993; Cordes and Barsh, 1994; Erkman et al., 1996; Torres et al., 1996; Xiang et al., 1997; Hadrys et al., 1998; Mendonsa and Riley, 1999; Moens et al., 1998; Self et al., 1998; Wang et al., 1998; Bermingham et al., 1999). Nevertheless, the mechanisms controlling the initial stages of hair cell and support cell differentiation remain largely unknown.

The alternating pattern of hair cells and support cells has led to the suggestion that their differentiation is coordinately regulated by interactions between Delta, a tethered ligand, and Notch, its receptor, in a process known as lateral inhibition or lateral specification (Campos-Ortega, 1994; Artavanis-Tsakonas et al., 1995; Lewis, 1996). This process forces neighboring cells belonging to the same equivalence group to adopt different cell fates. Typically, all cells in an equivalence group initially express low levels of both Delta and Notch, which later become mutually antagonistic. The balance in expression of these genes is disrupted when inductive signals or stochastic effects stimulate a subset of cells in the equivalence group to differentiate as the ‘default’ cell type and express high levels of Delta. Elevated Delta signaling from the default cell type increases activity of Notch receptors on neighboring cells, causing the latter to downregulate Delta expression and adopt an alternative cell fate. Thus, lateral inhibition provides a general mechanism of short-range signaling that produces ‘salt and pepper’ patterns of distinct cell types that differentiate in close proximity to one another. Mutations that disrupt the function of Delta or Notch result in overproduction of the default cell type. For example, in a wide range of animal species, disruption of Delta-Notch signaling causes ‘neurogenic’ phenotypes characterized by overproduction of early neural cell types at the expense of alternative cell types (Artavanis-Tsakonas et al., 1995; Chitnis et al., 1995; de la Pompa et al., 1997; Appel and Eisen, 1998; Haddon et al., 1998b). The hypothesis that hair cells and support cells are regulated by lateral inhibition is supported by recent findings in chick and zebrafish that differentiating hair cells, but not support cells, express high levels of multiple delta homologs (Adam et al., 1998; Haddon et al., 1998a). Furthermore, zebrafish embryos homozygous for the mind bomb (mib) mutation produce a large excess of hair cells but virtually no support cells. Although the mib gene has not been identified, the mutation is believed to impair Delta-Notch signaling because mib– mutants show a dramatic neurogenic phenotype that affects virtually the entire nervous system (Jiang et al, 1996; Schier et al., 1996).

To directly assess the role of endogenous Delta proteins, we have analyzed a mutation affecting the deltaA (dlA) gene of zebrafish. The mutant allele, dlA^dx2, is weakly dominant and incompletely penetrant. In homozygous mutants, penetrance is more complete and the phenotype is much more severe. The mutation results from a mis-sense mutation in which a highly conserved cysteine residue in EGF-repeat 2 is replaced by tyrosine. The resulting mutant protein exhibits dominant-negative activity, causing a more severe neurogenic phenotype than a null mutation of dlA (Appel et al., 1998).

Here, we show that dlA^dx2/dx2 homozygotes develop with a large excess of hair cells and display a notable expansion in the number of cells expressing pax2.1 in the vicinity of the sensory epithelia. More detailed analysis confirmed that pax2.1 is a reliable marker of hair cell differentiation. Support cells, which do not express pax2.1, are grossly deficient in dlA^dx2/dx2mutant embryos. noi^tb21/tb21 mutants, which are disrupted for pax2.1 function (Brand et al., 1996; Pfeffer et al., 1998), also produce a weak neurogenic phenotype that appears to result from reduced expression of dlA and dlD. In contrast, noi^tb21 partially suppresses the dlA^dx2 phenotype by reducing expression of the dominant-negative dlA^dx2 protein. Together, these data indicate that dlA normally restricts the number of hair cells that differentiate in the inner ear, and pax2.1 is required for normal levels of dlA-mediated lateral inhibition.

The wild-type line was obtained by hybridizing the AB line (Eugene, OR) to a partially inbred line of genetically similar store-bought fish. The dlA^dx2 mutation was induced in the wild-type background with ethylnitrosourea (Appel et al., 1999). The T(msxB)^b220 mutation was induced in the AB line with gamma rays (Fritz et al., 1996). The noi^tb21 and mib^ta52b mutations were induced with ENU in the Tu wild-type strain (Brand et al., 1996; Jiang et al., 1996).

Embryos were developed in an incubator at 28.5°C. Developmental stages are expressed in terms of h (hours of development). dlA^dx2/dx2 homozygotes were identified at 24 h by several criteria: the floor plate of the neural tube is partially disrupted, the main body axis usually shows strong dorsal curvature, and the hindbrain and otic vesicles often show severe morphogenetic defects. The majority of dlA^dx2/+ heterozygotes show none of these defects at 24 h, although 1-5% show mild dorsal curvature. To identify dlA^dx2/+ heterozygotes, hair cells were visualized in live embryos at 22-24 h. Wild-type embryos invariably show two utricular hair cells during this period, whereas dlA^dx2/+ embryos usually show three to five utricular hair cells by 24 h. dlA^dx2/T(msxB)^b220trans-heterozygotes resemble dlA^dx2/dx2 homozygotes but are usually less severely affected. noi^tb21/tb21 homozygotes were identified at 24 h by disruption of the midbrain-hindbrain border (Brand et al, 1996). dlA^dx2/dx2; noi^tb21/tb21 double mutants resemble noi^tb21/tb21 single mutants, except that double mutants also show partial disruption of the floor plate. mib^ta52b/ta52b homozygotes resemble dlA^dx2/dx2 homozygotes, but mib^ta52b/ta52b mutants are usually more severely affected and show somite defects as well. mib^ta52b/ta52b; noi^tb21/tb21 double mutants resemble mib^ta52b/ta52b single mutants, but double mutants also show disruption of the midbrain-hindbrain border.

Embryos were fixed in MEMFA (0.1 M MOPS at pH 7.4, 2 mM EGTA, 1 mM MgSO₄, 3.7% formaldehyde). In situ hybridizations (Stachel et al., 1993) were performed at 67°C using probes for pax-2.1 (Krauss et al., 1991), dlA or dlD transcripts (Appel and Eisen, 1998; Haddon et al., 1998b).

Embryos were fixed in MEMFA and washed for 1-2 hours at room temperature with PBT.LS (0.8% sodium chloride, 0.02% potassium chloride, 0.02 M sodium phosphate, pH 7.3, 2 mg/ml bovine serum albumin, 10% lamb serum and 0.1% Triton X-100). Embryos older than 24 h were initially washed with PBT.LS containing higher concentrations of Triton X-100: 30 h embryos were washed with 0.5% Triton X-100, and 48 or 60 h embryos with 2.5% Triton X-100. In all subsequent washes and incubations, PBT.LS with 0.1% Triton X-100 was used. Embryos were incubated with a 1:100 dilution of primary antibody directed against mouse Pax2 (Berkeley Antibody Company) or acetylated tubulin (Sigma T-6793). Embryos were then washed and incubated with one or more of the following secondary antibodies: Cy3-conjugated sheep anti-mouse IgG (Sigma C-2181, diluted 1:50), Cy3-conjugated sheep anti-rabbit IgG (Sigma C-2306, diluted 1:50), Alexa 488-conjugated goat anti-mouse IgG (Molecular Probes # A-11001, diluted 1:50), or HRP-conjugated goat anti-rabbit IgG (Sigma A-0545, diluted 1:200). For sectioning, embryos were embedded in Immunobed resin (Polysciences No. 17324) and cut into 4 μm sections.

mRNA was extracted from groups of 40 embryos (24 h) or 30 embryos (30 h) dissolved in 800 μl Tri Reagent (Sigma no. T9424) according to manufacturer specifications. Reverse transcription was performed in 20 μl reactions using Superscript II (Gibco-BRL) with 2 μg total RNA and 2 pmole of the downstream dlA-specific primer (see below). 1 μl (5%) of the RT-PCR reaction product was then PCR amplified with Amplitaq (Perkin-Elmer) and 2 pmole each of dlA-specific primers using the following program: 105 seconds at 95°C; 21 cycles of 94°C for 45 seconds, 55°C for 55 seconds, 72°C for 45 seconds; and a final extension for 6 minutes at 72°C. Sequences for dlA-specific primers are: Upstream primer, TCAGAGTCAAGGTATTCCG. Downstream primer, TCAGTA-CAGAGAACCAGCTC.

To analyze inner ear defects in dlA^dx2 mutant embryos, we examined expression of pax2.1, an early marker of otic placode and vesicle development. At 24 h, pax2.1 transcripts are localized to the medial portion of the otic vesicle in both wild-type and dlA^dx2/dx2 mutant embryos, with increased expression levels seen in clusters of cells at the anterior and posterior limits of this expression domain (Fig. 1A-C). In dlA^dx2/dx2 embryos, the number of cells with elevated pax2.1 expression is variably but consistently increased. These cells lie within the regions of the otic vesicle where sensory epithelia first develop, suggesting that the extra pax2.1-expressing cells in dlA^dx2/dx2 mutants might reflect overproduction of hair cells. Use of a polyclonal antibody directed against mouse Pax2 produces a pattern identical to the pax2.1 expression pattern and reveals increased numbers of Pax2-expressing cells in the anterior and posterior ends of the otic vesicle in dlA^dx2/dx2 mutants (Fig. 1E,F). Co-staining with a second antibody directed against acetylated tubulin confirms that virtually all of the cells with elevated Pax2 staining are indeed hair cells.

Because two homologs of pax2 have been identified in the zebrafish, pax2.1 and pax2.2 (Pfeffer et al., 1998), we wished to determine whether the mouse antibody recognizes both zebrafish proteins. Therefore, we examined the pattern of Pax2 antibody staining in no isthmus (noi) mutants, which are disrupted in pax2.1 function. The allele studied here, noi^tb21, produces a truncated protein lacking a significant portion of the carboxy terminus to which the antibody normally binds (Lun and Brand, 1998). Nevertheless, noi^tb2/tb211 mutants cannot be distinguished from wild-type embryos by anti-Pax2 staining at 14 h (not shown). However, noi^tb2/tb211 mutants show diminished staining in all pax2.1-expressing tissues by 18 h (Fig. 1F-I, and data not shown). Residual staining at 18 h does not simply reflect cross-reactivity to pax2.2 since staining is similarly reduced in the pronephros, which does not express pax2.2. Moreover, Pax2 staining is undetectable in noi^tb2/tb211 mutants by 24 h (Fig. 1J and data not shown), even though expression of pax2.2 is not diminished in noi– mutants (Pfeffer et al., 1998). These data suggest that the mouse Pax2 antibody primarily recognizes zebrafish pax2.1 and that the reduced staining observed in noi^tb2/tb211 embryos results from failure to maintain normal levels of pax2.1 expression (Brand et al., 1996).

High level expression of pax2.1 in differentiating hair cells has not been previously reported, so we examined this correlation in more detail at various stages of inner ear development. pax2.1 is expressed uniformly in cells throughout the medial half of the nascent otic vesicle at 18.5 h (Fig. 1G) but dramatically upregulates in hair cells by 24 h (Fig. 1D). At 30 h, the number of cells in the developing maculae that express high levels of pax2.1 increases roughly two-fold from that seen at 24 h (Fig. 2A), which agrees with the observed increase in the number of hair cells (Haddon and Lewis, 1996; Riley et al., 1997). In contrast, pax2.1 expression begins to downregulate in cells surrounding the presumptive hair cells. Sectioning of embryos stained with anti-Pax2 and anti-acetylated tubulin confirmed that all hair cells express high levels of pax2.1 whereas support cells express little or none (Fig. 2B). By 48 h, additional cells expressing high levels of pax2.1 accumulate in maculae, reflecting comparable changes in the number and distribution of hair cells. Maculae are also surrounded by rings of cells expressing pax2.1 at lower levels (Fig. 2C). This lower expression probably marks nascent hair cells at the margins of the maculae, as it presages the future dimensions of macular growth. Sections of 60 h embryos confirm that all macular hair cells continue to express high levels of pax2.1 (Fig. 2D). Although support cells do not express detectable levels of pax2.1, the margins of maculae often contain isolated cells that fully span the epithelium and also express low levels of pax2.1. The identity of these cells is uncertain but they probably correspond to nascent hair cells, which develop from cells that superficially resemble support cells (Haddon et al., 1998a). Cristae, which are the sensory patches associated with the semicircular canals, begin to develop at 60 h, and individual cells within the cristae also express pax2.1 (Fig. 2E). Analysis of specimens double stained for pax2.1 and acetylated tubulin confirms that these, too, are hair cells, and that support cells in the developing cristae do not express pax2.1 (Fig. 2F). Thus, pax2.1 appears to be induced during early stages of hair cell differentiation and is later maintained at high levels in mature hair cells. In contrast, support cell do not express detectable levels of pax2.1.

In dlA^dx2/dx2 mutants at 30 h, presumptive hair cells expressing high levels of pax2.1 become so numerous that the anterior and posterior maculae often become contiguous (Fig. 3A). Sections of dlA^dx2/dx2 embryos reveal that these expanded maculae develop with a gross excess of hair cells, all of which express high levels of pax2.1 (Fig. 3C). Support cells, on the contrary, are almost totally missing. The imbalance in the ratio of hair cells to support cells severely distorts the shape of the maculae and alters the morphology of the otic vesicle (Fig. 3C,D). Observation of live dlA^dx2/dx2 embryos reveals that hair cells in the anterior and posterior maculae often spread together to form a contiguous lawn, in agreement with the pattern of pax2.1 expression.

Otolith formation is usually delayed by several hours in dlA^dx2/dx2 embryos (not shown), probably reflecting a severe reduction in support cell function(s) that are normally required for deposition of otolith precursor materials (Riley and Grunwald, 1996; Riley et al., 1997). By 30 h, otoliths are usually present but are variable in number and morphology (Fig. 3D).

In severely affected dlA^dx2/dx2 embryos (Fig. 3A,C,D), the semicircular canals and cristae do not develop properly. However, the dlA^dx2 mutation is variable in its effects and permits some homozygotes to develop more normally. Analysis of moderately affected dlA^dx2/dx2 embryos revealed that, as with the maculae, cristae also develop with an excess of hair cells (Fig. 3E). Thus, the dlA^dx2 phenotype strongly supports the hypothesis that lateral inhibition mediated by dlA normally establishes the interspersed pattern of hair cells and support cells.

Hair cell differentiation in no isthmus mutants

Expression of pax2.1 in hair cells suggested that pax2.1 might play an essential role in hair cell differentiation. To address this issue, we examined hair cell differentiation in noi^tb21/tb21 mutant embryos by staining with antibodies to detect both pax2.1 and acetylated tubulin. As seen in both whole-mounts and sections of stained specimens (Fig. 4A,B), numerous hair cells are produced in noi^tb21/tb21 embryos by 30 h, despite the absence of detectable pax2 protein. Indeed, noi^tb21/tb21 embryos display a weak neurogenic phenotype in the ear, typically producing nearly twice the normal number of hair cells by 30 h (Table 1). By comparison, dlA^dx2/dx2 embryos typically produce about a five-fold excess of hair cells by 30 h.

To better characterize the extent of hair cell hyperplasia, we determined the time course of hair cell production during the early stages of otic vesicle development in +/+, dlA^dx2/dx2 and noi^tb21/tb21 embryos. The first hair cells to differentiate in the otic vesicle are termed ‘tether cells’, which constitute a precocious cell type analogous to primary neurons. Although tether cells initially appear morphologically immature and fully span the epithelium, they nevertheless possess functional kinocilia that serve to localize otolith accretion over the developing maculae (Riley et al., 1997). In wild-type embryos, we invariably observe two tether cells in both anterior and posterior maculae at the onset of otic vesicle formation at 18.5 h (Riley et al., 1997). Tether cells eventually acquire a more typical hair cell morphology that is indistinguishable from that of later forming hair cells, which begin to accumulate after 24 h. By 30 h, the anterior macula normally possesses around five hair cells, which includes the two mature tether cells. In noi^tb21/tb21 embryos, the number of tether cells produced during early stages of otic vesicle development is normal (Fig. 4C) whereas later forming hair cells are produced in excess, yielding about twice the normal number of hair cells by 30 h (Figs 4A, 5; Table 1). In dlA^dx2/dx2 embryos, both tether cells (Fig. 4D) and later forming hair cells are produced in excess, with about a five-fold excess of hair cells becoming evident by 30 h (Fig. 5; Table 1). Thus, although they are not as severely affected as dlA^dx2/dx2 embryos, noi^tb21/tb21 mutants produce a significant excess of hair cells. These data suggest that pax2.1 is not essential for hair cell differentiation but that it may play a role in limiting the number of cells that differentiate as hair cells.

To examine this further, we generated dlA^dx2/dx2; noi^tb21/tb21 double mutants to determine whether the two mutations interact genetically. To our surprise, inner ear development in dlA^dx2/dx2; noi^tb21/tb21 double homozygotes is nearly identical to that of noi^tb21/tb21 homozygotes: double mutants initially produce a normal number of tether cells but produce just over twice the normal number of hair cells by 30 h (Figs 4E, 5; Table 1). Thus, noi^tb21 strongly suppresses the dlA^dx2 phenotype, suggesting that pax2.1 and dlA participate in the same developmental pathway.

To better understand the effects of dlA^dx2 and noi^tb21 on hair cell development and lateral inhibition, we analyzed expression of dlA and dlD in the inner ear of wild-type and mutant embryos. dlA and dlD are initially expressed in hair cell precursors at the anterior and posterior ends of the otic placode by 14 h (Haddon et al., 1998a). Expression in tether cells is evident in the newly formed otic vesicle at 19 h in wild-type embryos and, as expected, this population of cells is overproduced in dx2 mutants (Fig. 6A,C,D,F). By 30 h of development, dlA and dlD are downregulated to very low or undetectable levels in mature hair cells. However, at the margin of each macula in wild-type embryos, we typically observe one or two cells that express these genes at relatively high levels (Fig. 7A,E). These cells, which fully span the epithelium, appear to be nascent hair cells undergoing early stages of differentiation (Haddon et al., 1998a). In dlA^dx2/dx2 embryos, nascent hair cells with high levels of dlA and dlD expression are produced in excess (Fig. 7C,G). These data provide further evidence that dlA^dx2 disrupts lateral inhibition, thereby increasing the number of cells differentiating as hair cells.

In noi^tb21/tb21 embryos, the overall pattern of dlA and dlD expression is similar to that seen in wild-type embryos (Figs 6B,E, 7B,F). However, the level of expression in nascent hair cells in noi^tb21 embryos is variably reduced compared to the wild type. Moreover, the number of nascent hair cells with detectable expression of dlA and dlD is not elevated in noi^tb21/tb21 embryos, even though they produce twice the normal number of hair cells. Similarly, dlA^dx2/dx2; noi^tb21/tb21 double mutants express dlA and dlD at lower levels, and in fewer cells, than do dlA^dx2/dx2 mutants (Fig. 7D,H). Together, these observations suggest that the noi^tb21 mutation reduces the magnitude and/or duration of maximal dlA and dlD expression in nascent hair cells. Quantitative PCR analysis of whole embryo RNA confirms that noi^tb21/tb21 mutants show a 25% decrease in overall dlA mRNA levels compared to wild-type embryos (Fig. 8). Moreover, results of in situ hybridization experiments indicate that noi^tb21 preferentially reduces dlA expression levels in tissues that also express pax2.1, especially the otic vesicle and lateral hindbrain (Fig. 7 and data not shown). Hence, the level of dlA expression in hair cells is likely to be reduced by considerably more than 25% in noi^tb21/tb21 mutants. Because the efficiency of Delta-Notch signaling is highly sensitive to changes in Delta protein levels, it is likely that the reduced expression of dlA and dlD observed in noi^tb21/tb21 embryos is sufficient to weaken signals required for lateral inhibition, thus providing an explanation for why these mutants produce supernumerary hair cells. Furthermore, reduced levels of transcription from the mutant dlA^dx2 locus could also explain the ability of noi^tb21 to suppress the dlA^dx2phenotype. Since dlA^dx2 encodes a dominant negative protein, reducing the level of the mutant protein would be expected to ameliorate its effects.

If noi^tb21 affects lateral inhibition in hair cells by reducing expression of dlA and dlD, then T(msxB)^b220, a deletion that removes the dlA locus (Fritz et al., 1996), should enhance the neurogenic effects of noi^tb21 by further reducing dlA levels. Analysis of T(msxB)^b220/b220 homozygotes is confounded by the phenotypic severity of the large T(msxB)^b220 deletion. T(msxB)^b220/+ heterozygotes, on the contrary, appear normal and produce a normal number of hair cells (Table 1). This indicates that there is sufficient redundancy in delta gene function to compensate for the loss of a single copy of dlA. Nevertheless, populations of embryos with equal numbers of noi^tb21/tb21; T(msxB)^b220/+ double mutants and noi^tb21/tb21 single mutants show a 20% increase in mean hair cell production compared to noi^tb21/tb21 alone (Table 1). This increase is highly reproducible and statistically significant (P<0.025). Moreover, it probably underestimates the effect of T(msxB)^b220 since non-T(msxB)^b220 carriers cannot be identified and excluded from such populations. Thus, halving the level of dlA alone is not sufficient to disrupt lateral inhibition, but is sufficient to enhance the noi^tb21 phenotype. This suggests that once Delta-Notch signaling drops below a critical threshold, quantitative changes in the efficiency of lateral inhibition are more easily observed such that haploinsufficiency of dlA enhances the neurogenic effects of noi^tb21.

Additional evidence that dosage of the wild-type and mutant alleles of dlA affects lateral inhibition of hair cells comes from analysis of interactions between dlA^dx2 and T(msxB)^b220. Unlike T(msxB)^b220/+ heterozygotes, dlA^dx2/+ heterozygotes show a weak neurogenic phenotype, producing an average of 70% more hair cells than normal (Table 1). dlA^dx2/T(msxB)^b220trans-heterozygotes produce nearly three-fold more hair cells than normal (Fig. 3B; Table 1), a value that lies between those seen in dlA^dx2/dx2 homozygotes and dlA^dx2/+ heterozygotes. Thus, expressing two copies of the dlA^dx2 allele causes a more complete disruption of lateral inhibition than does a single copy, and expressing one copy of dlA^dx2 alone is more severe than co-expressing dlA^dx2 along with the wild-type allele of dlA. These data are consistent with notion that dlA^dx2 encodes an antimorphic protein that interferes with the function of wild-type dlA, as well as other delta homologs, and that lateral inhibition is a sensitive function of total levels of Delta-Notch signaling.

Because mib is the only other neurogenic mutation yet described in zebrafish, we wished to compare the effects of mib^ta52b to those of dlA^dx2 and noi^tb21. The gene affected by the mib mutation has not yet been identified, but results of several studies strongly suggest that mib disrupts Delta-Notch signaling and lateral inhibition (Jiang et al., 1996; Schier et al., 1996; Haddon et al., 1998a). The mib^ta52b phenotype is similar to that of dlA^dx2, although mib^ta52b is more severe and affects a wider range of tissues. In the developing inner ear, mib^ta52b/ta52b mutants produce a 10-fold excess of pax2.1-expressing hair cells by 30 h (Table 1). mib^ta52b/ta52b; noi^tb21/tb21 double mutants produced the same number of hair cells as mib^ta52b.ta52b single mutants (Fig. 4F; Table 1). Thus, mib^ta52b is epistatic to noi^tb21, suggesting that mib^ta52b is unaffected by changing levels of delta gene expression. Furthermore, mib^ta52b also appears to be epistatic to dlA^dx2: Intercrosses between dlA^dx2/+; mib^ta52b/+ double heterozygotes yield the expected ratio of mib^ta52b/ta52b progeny but too few dlA^dx2/dx2 progeny. For example, among progeny pooled from three such intercrosses, 112/411 (27%) appeared as mib^ta52b/ta52b homozyotes, but only 70/411 (17%) showed the dlA^dx2/dx2 phenotype. We infer that 1/4 of mib^ta52b/ta52b progeny are also homozygous for dlA^dx2, but that the more severe mib^ta52b/ta52b phenotype masks the presence of dlA^dx2. Accordingly, we observed no difference in hair cell production in populations containing presumptive dlA^dx2/dx2; mib^ta52b/ta52b double mutants compared to mib^ta52b/ta52b single mutants (Table 1). Together, these data could indicate that the mib^ta52b mutation blocks reception of Delta signals.

In this study, we have shown that dlA and pax2.1 interact to regulate the initial differentiation and diversification of hair cells and support cells within the otic sensory epithelium. pax2.1 is initially expressed in preotic cells 2-3 hours prior to formation of the otic placode, but its expression does not become regionally restricted within this domain until much later when the otic vesicle forms (Krauss et al., 1991). In contrast, dlA, dlB and dlD are expressed in the nascent otic placode in regions that later give rise to the sensory epithelia, constituting one of the earliest events yet reported for initial differentiation of the sensory epithelium (Haddon et al., 1998a). The developmental signals required for localized induction of the sensory epithelia have not been identified, but several studies suggest that signals from cephalic mesendoderm are involved. In zebrafish mutants lacking cephalic mesendoderm, induction of the otic placode is delayed, as revealed by belated expression of pax2.1, and the sensory epithelia are poorly formed (Mendonsa and Riley, 1999). In chick embryos, explanting otic placodes without surrounding periotic mesenchyme results in formation of vesicles with deranged sensory epithelia (Noden and Van de Water, 1986; Swanson et al., 1990). However, in all cases reported, local arrangements of support cells and hair cells are relatively normal. These data suggest that signals from surrounding mesendoderm are required for normal positioning of the sensory epithelia within the placode and vesicle, but that endogenous cues generate the interspersed arrangement of hair cells and support cells. Delta-Notch signaling appears to be the principle mechanism for controlling the latter function (Adam et al., 1998; Haddon et al., 1998a; this report).

In other species, induction of neural potential and subsequent delta gene expression require prior expression of proneural genes, including homologs of atonal or members of the achete-scute complex in Drosophila (Campos-Ortega, 1994; Artavanis-Tsakonas et al., 1995; Lewis, 1996; Bermingham, 1999). The proneural genes encode bHLH transcription factors that stimulate transcription of delta genes. Presumably, one or more proneural genes also regulate delta gene expression in the zebrafish inner ear. Although such proneural genes have not yet been identified, it is likely that they are expressed quite early in otic development, possibly in response to the same signals that induce expression of pax2.1 in the preotic placode.

Although pax2.1 expression precedes delta gene expression in developing hair cells, and disruption of pax2.1 perturbs delta gene expression (Fig. 7), the function of pax2.1 is not strictly analogous to that of a proneural gene. Expression of proneural genes is typically extinguished in neural cells as they begin to differentiate (Campos-Ortega, 1994; Lewis, 1996), whereas pax2.1 is retained in hair cells following differentiation. Furthermore, the noi mutant phenotype indicates that pax2.1 is not required for induction of the otic placode or the sensory epithelia, nor is it required for induction of delta gene expression. Instead, pax2.1 is only required for optimal expression of delta genes in nascent hair cells. Expression levels of dlA and dlD are often reduced in nascent hair cells in noi^tb21/tb21 mutants, and the number of delta-expressing cells is lower than expected based on the number of mature hair cells that eventually form. These data suggest that, in noi^tb21/tb21mutants, maximal delta expression in differentiating hair cells is either delayed or prematurely attenuated. This misregulation is probably sufficient to weaken signals required for lateral inhibition, providing an explanation for why noi^tb21/tb21 mutants produce supernumerary hair cells. Reducing expression from the dlA locus also explains the ability of noi^tb21 to suppress the phenotype of dlA^dx2, which encodes a dominant-negative protein. Indeed, halving the dose of dlA^dx2 by generating trans-heterozygotes carrying a single copy each of dlA^dx2 and T(msxB)^b220 (a deletion that removes the dlA locus) also partially suppresses dlA^dx2. In contrast, halving the dose of wild-type dlA by generating T(msxB)^b220/+ heterozygotes enhances the neurogenic effects of noi^tb21.

It is not yet clear how pax2.1 affects delta gene expression. pax2.1 could augment the actions of other transcription factors that are required for expression of delta genes. Alternatively, pax2.1 could inhibit transcription of Notch-activated genes in nascent hair cells, making them insensitive to the effects of Delta signals from other cells. Because Notch activity typically inhibits delta expression in the receiving cell (Campos-Ortega, 1994; Artavanis-Tsakonas et al., 1995; Lewis, 1996), pax2.1-mediated inhibition of N signal transduction would tend to intensify and prolong delta gene expression in nascent hair cells. Thus, pax2.1 could help to induce or amplify initial biases in Delta-Notch signaling associated with lateral inhibition. Finally, it is possible that pax2.1 plays a more general role that only indirectly affects Delta-Notch signaling. Although pax2.1 is not required for hair cell differentiation per se, it may regulate the sequence or duration of early events in the differentiation process. In this case, disruption of pax2.1 could prematurely activate a relatively advanced phase of hair cell differentiation during which delta genes are normally downregulated. Analysis of delta promoter sequences and identification of additional markers of hair cell and support cell differentiation will help to resolve these issues.

It is possible that a relationship between pax function and Delta-Notch signaling has been conserved between invertebrates and vertebrates. A pax2 homolog was recently identified in Drosophila, and two previously described mutations, sparkling and shaven, were found to disrupt distinct promoter elements that normally control regional expression of pax2 (Fu and Noll, 1997; Fu et al., 1998). sparkling mutants develop with a rough eye phenotype resulting from defects in formation of cone and pigments cells. shaven mutants develop with a deficiency of sensory bristles due to a defect in the formation of shaft cells. Similar phenotypes are caused by mutations that alter Delta-Notch signaling, and genetic studies confirm that pax2 and Delta-Notch cooperate to regulate differentiation of shaft cells in Drosophila (Kavaler et al., 1999). In C. elegans, eg-l38 encodes a pax2/5/8 homolog and mutants display defects in development of ventral uterine cells (Chamberlin et al., 1997). Disruption of lin-12, a notch homolog, perturbs development of the same cell population (Newman et al., 1995), suggesting that, here too, pax function might cooperate with lateral inhibition during normal development.

It is also possible that Pax2 plays a similar role in mouse. Mouse Pax2 is expressed in the developing inner ear and mice homozygous for a Pax2 null mutation fail to produce a functional cochlea (Torres et al., 1996). Neurogenic defects were not reported in Pax2 mutant mice, but such defects may have been relatively subtle, as we have observed in noi^tb21/tb21mutants.

In contrast to its effects on dlA^dx2, noi^tb21 does not suppress the neurogenic effects of mib^ta52b. The gene affected by the mib^ta52b mutation has not yet been identified, but it seems likely that it affects some aspect of notch function or downstream signal transduction. The mib^ta52b phenotype is more severe and penetrant than that of dlA^dx2, as would be expected if mib^ta52b were to block reception of all Delta signals. This would also explain why mib^ta52b is epistatic to both dlA^dx2 and noi^tb21 (Table 1). Identification of the mib gene will greatly aid in our understanding of Delta-Notch signaling pathways in zebrafish, and genetic screens for second site modifiers of dlA^dx2 and mib^ta52b could identify additional elements in the signaling pathway.

This work was supported by grants from The Texas Advanced Research Program (010366-080FI), March of Dimes (#1-FY98-0126), and the National Institutes of Health (#1 R29 DC03405-01A1). We thank Alex Schier and Michael Brand for providing noi^tb21. We thank Bruce Appel, Julian Lewis and Catherine Haddon for providing mib^ta52b and clones of dlA and dlD. We also thank Bruce Appel for thoughtful discussions of the work.