Cross-regulation of Ngn1 and Math1 coordinates the production of neurons and sensory hair cells during inner ear development

During ear development, diverse components of a sensory pathway - neurons,mechanosensory hair cells, supporting cells and other non-sensory epithelial cells - derive from a simple ectodermal placode. These cell fate decisions are temporally and spatially regulated over a period when the placode, through growth and morphogenesis, transforms into an otocyst and, later, the inner ear labyrinth (see Fig. 1A)(Barald and Kelly, 2004; Kiernan et al., 2002). Early on, a subset of otic epithelial cells expresses neural fate markers(Raft et al., 2004; Cole et al., 2000; Adam et al., 1998), delaminates from the otocyst, and forms the VIIIth cranial ganglion rudiment with neural crest-derived glial precursors(D'Amico-Martel and Noden,1983). In mice, hair cell production begins about 3.5 days after the onset of neurogenic gene expression(Shailam et al., 1999). Hair cells and supporting cells remain epithelial and form six discrete sensory patches along the inner ear labyrinth(Sher, 1971). Five sensory epithelia mediate the sense of balance (the utricular macula, saccular macula,and three cristae) and one - the organ of Corti - detects sound.

Morphological and genetic homologies between the sensory organs of the vertebrate ear and insects raise the question of whether neurons and hair cells derive from a common progenitor cell type(Fekete and Wu, 2002; Adam et al., 1998). Direct evidence for this is limited to one recent study in chicken that found clonal relatives of VIIIth ganglion neurons within the utricular macula and adjacent non-sensory epithelium (Satoh and Fekete,2005). Existing evidence for such a relationship in mammals is circumstantial (Matei et al.,2005; Raft et al.,2004).

Otic neurogenesis and hair cell generation are each dependent on an Atonal-related bHLH transcription factor. neurogenin 1 (Ngn1; also known as Neurog1 - Mouse Genome Informatics) is necessary for the commitment of otocyst epithelial cells to a neural fate, as Ngn1^-/- embryos lack an VIIIth cranial ganglion and fail to express neural fate markers in the otocyst(Ma et al., 1998; Ma et al., 2000). By contrast, Math1 (also known as Atoh1 - Mouse Genome Informatics) is necessary and sufficient for hair cell generation(Bermingham et al., 1999; Zheng and Gao, 2000; Izumikawa et al., 2005). Thus,during ear development, Ngn1 and Math1 function as determination factors, but it is not known whether their expression is regulated in a coordinated manner. In other neural systems, bHLH genes cross-regulate to control the commitment of progenitor cells to alternative fates (Bertrand et al., 2002). Examples include the involvement of Mash/Math/Ngn1 genes in the sequential production of retinal neurons and glia(Inoue et al., 2002; Akagi et al., 2004), and in the simultaneous production of distinct neural subtypes in the forebrain and spinal cord (Fode et al.,2000; Gowan et al.,2001).

Here, we identify a progenitor cell field that produces both neurons and hair cells and describe genetic interactions mediating a neural-hair cell fate decision. We show that neural precursor and hair cell production overlap in the otic epithelium for several days and demonstrate by fate mapping that the Ngn1-expressing neurogenic region is transformed into the sensory maculae of the utricle and saccule. We propose that this transformation is governed by a mutual antagonism between Ngn1 and Math1. We also show that Ngn1 negatively regulates its own expression through Notch-mediated lateral inhibition, whereas Math1 positively regulates its own expression. Differential autoregulation of Ngn1 and Math1 provides an explanation for the progression toward sensory epithelial development over time. Ngn1, as well as generating neural precursors, functions via lateral inhibition to maintain an uncommitted progenitor cell population for sensory epithelial development; Math1,in turn, irreversibly commits these progenitors to a hair cell fate.

Targeted disruptions of Math1(Ben-Arie et al., 1997), Ngn1 (Ma et al.,1998), or Pofut1 (Shi and Stanley, 2003) loci were maintained on a CD1 background. Ngn1-GFP and Ngn1-CreER BAC transgenic mice were constructed with RPCI-23-457E22(http://bacpac.CHORI.org/),which contains a genomic insert of 184 kb with 113 kb 5′ and 71 kb 3′ of the Ngn1 coding sequence. The Math1-GFP BAC transgenic was constructed with RPCI-23318G16, which contains a genomic insert of 181 kb with 75 kb 5′ and 104 kb 3′ of the Math1 coding sequence. Homologous recombination in bacteria (Yang et al., 1997) was used to replace endogenous coding sequences with coding sequence for either nuclear-localized EGFP (Clontech) or CreER^T2 (Feil et al.,1997). Details of Ngn1CreER BAC transgene construction (E.J.K. and L.V.G., unpublished) are available on request. Genotyping was accomplished by PCR using the following oligos: Ngn1-GFP,5′-CGAAGGCTACGTCCAGGAGCGCAC-3′ and 5′-GCACGGGGCCGTCGCCGATGGGGGTGT-3′; Ngn1-CreER,5′-AGCCCATTCACTCCCTGAG-3′ and 5′-ATCAACGTTTTCTTTTCGGA-3′; Math1-GFP,5′-CTGACCCTGAAGTTCATCTGCACC-3′ and 5′-TGGCTGTTGTAGTTGTACTCCAGC-3′; Math1 mutants,5′-GAACCCAAAGACCTTTTGCAC-3′ and 5′-CACGAGACTAGTGAGACGTG-3′; Ngn1 mutants,5′-AAGAGTGCCATGCCCGAAGG-3′ and 5′-AAGGCCGACCTCCAAACCTC-3′; Pofut1 mutants,5′-GGGTCACCTTCATGTACAAGTGAGTG-3′ and 5′-ACCCACAGGCTGTGCAGTCTTTG-3′. The Math1-GFP transgenic line carrying a 1.4 kb Math1 enhancer fragment was genotyped as previously described (Chen et al., 2002; Lumpkin et al., 2003). The Ngn1-CreER transgene was maintained in males on a Z/EG reporter strain(Novak et al., 2000);double-transgenic embryos were obtained by breeding these to Z/EG reporter females. Tamoxifen (Sigma T5648) in corn oil was administered by gavage twice daily to alert, pregnant females (1.0 mg per 40 g of body weight per administration).

Embryos were immersion-fixed (4% paraformaldehyde), cryoprotected in 30%sucrose in PBS, embedded in OCT compound (Tissue-Tek) and cryosectioned. Immunohistochemistry was performed by a standard protocol of blocking (5%donkey serum, 0.1% Triton X-100 in PBS), incubations and washings. Antibodies included: anti-myosin VIIa rabbit polyclonal (obtained from Tama Hasson, UCSD,CA; 1:400), anti-GFP chicken polyclonal (Chemicon, 1:100), anti-GFP rabbit polyclonal (Molecular Probes, 1:1000), anti-islet1/2 mouse monoclonal (mAb4D5,Developmental Studies Hybridoma Bank, straight supernatant) and anti-human/mouse active caspase 3 (R&D Systems, 1:500). Secondary antibodies (Jackson ImmunoResearch) were used at 1:200. In situ hybridization was performed by standard methods (Stern,1998). For combined in situ hybridization and GFP immunohistochemistry (with anti-GFP rabbit polyclonal, Molecular Probes),proteinase K digestion (5 μg/ml in PBS) was reduced to 3 minutes. Digoxygenin-labeled RNA probes for NeuroD (Lee et al., 1994), Math1 (Helms and Johnson,1998), Ngn1 (Ma et al., 1998) and Dll1(Beckers et al., 1999) were prepared by standard methods. TUNEL reactions were performed according to manufacturer's instructions (Roche).

Serial sections were photographed and imported into Photoshop CS2 (Adobe). Occurrence data were obtained by digitally marking all positive cells identified in TIFF files of complete, consecutive 12 μm serial sections for a given structure and automated tallying (Image Processing Toolkit, Reindeer Graphics). Data were tested for significance with multiple t-tests(two-tailed; unequal variance). Domain areas and relative cell positions were quantified with NIHimage/ImageJ. Space filling models were generated by aligning serial sections with Autoaligner (Bitplane AG), tracing the regions of interest, and importing stacks of tracings into Imaris (Bitplane AG).

We performed RNA in situ hybridization for NeuroD and Math1 to characterize any temporal and spatial overlap between neurogenesis and hair cell production (Fig. 1A) (Fritzsch et al.,2002). NeuroD (also known as Neurod1 - Mouse Genome Informatics), a neural differentiation gene(Bertrand et al., 2002), is expressed under the control of Ngn1 in the otic epithelium(Ma et al., 1998) and marks cells that are committed to the neural lineage. We also hybridized a NeuroD RNA probe to tissue from a Math1-GFP transgenic reporter mouse(Lumpkin et al., 2003). Math1 is the earliest known specific marker of hair cells(Bermingham et al., 1999).

The number of NeuroD⁺ cells within the epithelium increases between E9 and E10.5 (Raft et al., 2004) (Fig. 1A). Between E11.5 and E12.5, NeuroD expression split into two distinct regions of neurogenesis that would ultimately lie within the developing utricle and saccule (Fig. 1B,C, cyan). Neurogenesis declined from E11.5 onward(Fig. 1D,E), but a few delaminating NeuroD⁺ cells were still present in the utricle as late as E17.5 (5±4 cells, n=3 ears; data not shown). A comparison of NeuroD and Ngn1 expression in the otic epithelium revealed no differences in their patterning (see Fig. S1A,D,E,H in the supplementary material).

During the decline in neurogenesis, Math1 mRNA expression begins and is maintained in all hair cells through to at least E17.5(Shailam et al., 1999). At E11.5, two Math1⁺ stripes appeared within the NeuroD domain along its lateral and medial borders(Fig. 1B,F,G). Between E11.5 and E12.5, Math1-GFP⁺ cells in these stripes increased in number by 8- to 10-fold and formed the nascent maculae of the utricle and saccule. Initially, Math1-GFP⁺ and NeuroD⁺ cells intermingled (Fig. 1H,I), but later lay on either side of a border that delineates the macula and its adjacent neurogenic domain (Fig. 1J). Math1 expression associated with the cristae first appeared as separate foci outside the NeuroD⁺ domain at around E12 (data not shown). We found no temporal overlap of Math1and NeuroD expression in the cochlea (data not shown), suggesting that neurogenesis and hair cell generation do not coincide in the auditory end-organ. Taken together, our data reveal that neurogenesis is maintained through stages of hair cell production in the utricle and saccule, but declines sharply as Math1 expression and hair cell production increase.

Our expression analysis raised the possibility that sensory maculae of the utricle and saccule, but not the cristae or organ of Corti, derive from the neurogenic region of the otocyst. To test this, we permanently labeled cells of the neurogenic region using a BAC transgenic mouse line (Ngn1-CreER)(E.J.K. and L.V.G., unpublished) that expresses a tamoxifen-inducible form of Cre recombinase (CreER) under the control of Ngn1 regulatory elements(see Fig. S1B,F in the supplementary material). This allowed us to identify cells transiently expressing Ngn1 and their progeny after sensory epithelia have formed. As expected, when the Ngn-CreER mouse was crossed with the Z/EG reporter line (Novak et al.,2000) and tamoxifen administered, roughly 50% of VIIIth cranial ganglion neurons were permanently labeled in double-transgenic embryos(Fig. 2A; see Fig. S1I in the supplementary material). Importantly, we also found Ngn1 derivatives to be present in sensory and non-sensory inner ear epithelia of embryos that had been sacrificed after neurogenesis was largely complete(Fig. 2B-G).

To follow the fate of Ngn1-expressing cells in the ear, we administered tamoxifen twice daily from E8.5 until E13.5 to pregnant females of Ngn1-CreER × Z/EG matings. We analyzed 20 double-transgenic right ears from seven litters ranging in age from E13.5-16.5 and identified over 5000 labeled epithelial cells. Sensory epithelia were identified by the presence of myosin VIIa protein, and the resulting distribution pattern of epithelial Ngn1 derivatives is summarized in Fig. 2G. Supporting the hypothesis that maculae are the only sensory epithelia to derive from neurogenic epithelium, Ngn1 derivatives were present in the utricular and saccular maculae of all specimens analyzed(Fig. 2B,C). For embryos sacrificed at E14.5, tamoxifen administration from E8.5-13.5 yielded an average of 157±25 Ngn1 derivatives per utricular macula (n=6 ears from two litters). We found a lower occurrence of such cells in the saccular macula (82±24; n=6 ears from two litters). By contrast, only one ear out of the 20 analyzed had Ngn1 derivatives in the lateral crista (eight labeled supporting cells), and no such cells were detected in the other cristae in our cohort of specimens. No Ngn1 derivatives were detected in the organ of Corti. We were able to classify macular Ngn1 derivatives as differentiated myosin VIIa⁺ hair cells(Fig. 2H), undifferentiated myosin VIIa⁺ epithelial cells migrating within the apical-basal plane of the epithelium (Fig. 2I), or as myosin VIIa^- pseudostratified epithelial cells (Fig. 2J). By E16.5, many of these myosin VIIa^- Ngn1 derivatives exhibited morphological features of supporting cells (Fig. 2K, arrowheads).

Regions of non-sensory epithelium flanking the maculae of E13.5-16.5 ears also contain labeled cells (Fig. 2D,E), but we found no Ngn1 derivatives in the semicircular canals. In the auditory portion of the ear, Ngn1 derivatives were detected in 82% (14/17) of the cochleae analyzed (E13-16.5), but showed extreme variability in their numbers (93±146 cells, n=8 cochleae at E14.5). These cells commonly occupied the greater epithelial ridge (GER), a non-sensory region of the cochlea that is adjacent to the organ of Corti(Fig. 2F). However, as described above, no Ngn1 derivatives were detected in the organ of Corti itself.

Administration of tamoxifen only at placode/otocyst stages (E8.5 and E9.5)resulted in the same spatial distribution of Ngn1 derivatives as described above. Together, our results indicate that the utricular and saccular maculae,as well as some non-sensory epithelium flanking these structures, derive from the neurogenic region of the otocyst, and that Ngn1-expressing otocyst cells or their descendants can differentiate as hair cells, supporting cells, or as structural epithelial cells.

To understand how the sensory maculae and their surrounding tissue arise from neurogenic tissue, we analyzed changes in Ngn1 expression over time using two different transgenic lines. A series of Ngn1-CreER × Z/EG litters received initial tamoxifen exposures at progressively later developmental time points from E8.5 onwards; once begun, tamoxifen administration was continued twice per day until E13.5 and all litters were sacrificed on E14.5. By quantifying GFP⁺ cells in the utricle of these embryos, we confirmed that starting tamoxifen administration at progressively later times leads to diminishing numbers of labeled Ngn1 derivatives (Fig. 3C). From these experiments, we mapped the distributions of Ngn1 derivatives in the utricular macula and its flanking non-sensory tissue(Fig. 3E-E‴). The distribution of cells actively expressing Ngn1 at E14.5 was obtained from a different BAC transgenic line (Ngn1-GFP) that reports directly on Ngn1 promoter activity (see Fig. S1C,G,J in the supplementary material).

Our results for the utricle are consistent with a stereotyped reduction in the area of Ngn1 expression over time. At E14.5, the active neurogenic domain of the utricle, defined by expression of the Ngn1-GFP reporter, lay medial to and centered along the anteroposterior axis of the macula (Fig. 3A,B). When tamoxifen was administered to Ngn1-CreER;Z/EG litters from E8.5-13.5(Fig. 3E), Ngn1 derivatives were present throughout the neurogenic domain (white region), the macular sensory epithelium (gray region, defined by myosin VIIa expression), lateral non-sensory tissue (between the utriclar macula and lateral crista, yellow region), and posterior non-sensory tissue (between the utricular and saccular maculae, yellow region, bottom). When tamoxifen was administered from E10.5 or E11.5 to E13.5, Ngn1 derivatives were present at all these sites, except for posterior non-sensory tissue (between the maculae, Fig. 3E′). Finally, when tamoxifen was administered from E12.5-13.5, Ngn1 derivatives were present only within the neurogenic domain and a portion of the macular epithelium nearest its border with the neurogenic domain (Fig. 3E″). Contraction of the Ngn1 expression domain is therefore directional, occurring largely from lateral to medial in the utricle. We observed similar Ngn1 expression dynamics in the saccule(see Fig. S2 in the supplementary material), although contraction occurred largely along the anteroposterior axis of this structure, rather than along the medial-lateral axis as in the utricle. Importantly, all tamoxifen start times tested (ranging from E8.5-12.5) resulted in a cohort of macular Ngn1 derivatives comprising both hair cells and pseudostratified epithelial cells. These data, together with results described in the previous section, indicate that macular sensory cells can derive from cells that express Ngn1 at any time between E9 and E14.

Our results indicate that the domain of Ngn1⁺ precursor cells is gradually transformed from a purely neurogenic region into sensory epithelia of the utricular and saccular maculae. Since functional antagonism between related bHLH transcription factors has been described in other systems(Fode et al., 2000; Gowan et al., 2001; Akagi et al., 2004), we tested whether Math1 (which is required for sensory epithelial differentiation)(Bermingham et al., 1999)suppresses neurogenesis by inhibiting Ngn1 transcription in otic epithelial cells. We found that Math1 function is required for the normal contraction of epithelial Ngn1 expression characterized in the previous section (Fig. 4A,B;data not shown). We quantified expression of the Ngn1-GFP reporter and NeuroD in Math1-null homozygote embryos and observed large(>6×) increases over wild type in the numbers of neural precursors within the developing utricle and saccule; a less severe form of this phenotype was found in Math1 heterozygotes(Table 1). Excess neural precursors were seen to delaminate and migrate away from the mutant epithelium to form an VIIIth cranial ganglion that was larger than wild type (see Fig. S3B asterisk, E-H, in the supplementary material). Ectopic neurogenesis in Math1^-/- epithelia localized specifically to parts of the utricle and saccule that normally differentiate as sensory maculae, and was not detected in the cristae, cochlea, or any non-sensory epithelia(Fig. 4D,F; see Fig. S3A-D in the supplementary material; data not shown); in Math1 heterozygotes,it occurred only at the interface of neurogenic and sensory regions(Fig. 4E, bracket). In Math1^-/- epithelia, marked excess neurogenesis at E14.5 and E15.5 followed a partial decline in neurogenic activity through E13.5(Fig. 4N). This initial declining trend in neurogenic activity in mutants, which is similar to that of wild type, suggests that factors in addition to Math1 contribute to early neurogenic suppression.

Ngn1 loss-of-function causes a failure of neural precursor generation in the otic epithelium and absence of the VIIIth cranial ganglion(Ma et al., 1998; Ma et al., 2000). If antagonism between Math1 and Ngn1 during ear development is reciprocal, then Ngn1 loss-of-function should result in ectopic hair cells. Our analysis of sensory marker expression in the Ngn1^-/- utricle confirmed this prediction. As indicated by both Math1 and myosin VIIa expression, the Ngn1^-/- utricular macula at E13.5 and E14.5 was expanded medially into the region that is normally neurogenic(Fig. 4C,F,H,K,M). At E13.5,the total area of Math1 expression in the Ngn1^-/-utricle exceeded that of wild type by 39% (P<0.002, n=3 epithelia per genotype), and there were more Math1⁺ cells in the Ngn1^-/- utricle than in wild type at E13.5 and E14.5 (Table 1). Growth abnormalities of the Ngn1^-/- mutant ear(Matei et al., 2005; Ma et al., 2000) confounded our analysis of other structures. The saccule was profoundly hypoplastic (see Fig. S4A in the supplementary material), and we correlated this to a period of intense, region-specific apoptosis in the otic epithelium between E11.5 and E12.5 (see Fig. S4B,C in the supplementary material; data not shown). Ngn1^-/- cristae were also smaller than their wild-type counterparts, which was evident from our quantification of Math1⁺ cells in the Ngn1-null mutant lateral crista (Table 1).

The inner ear of Ngn1 heterozygotes was not grossly dysmorphic and therefore provided a context for analyzing cell fate changes in response to reduced Ngn1 levels. We quantified Math1⁺ cells in the utricle, saccule and lateral crista of Ngn1 heterozygote and wild-type littermates. In the Ngn1^+/- utricle and saccule,numbers of Math1⁺ cells were increased by 16-29% compared with wild type (Table 1). This was due to an expansion in the area of Math1⁺ macular domains in heterozygotes (Fig. 4K,L) and an increased density of Math1⁺ cells within heterozygote maculae compared with wild type(Fig. 4O). Unlike the maculae,the Ngn1^+/- lateral crista - a structure that does not express Ngn1 during normal development - showed no increase in Math1⁺ cells compared with wild type(Table 1). These results suggest that the numbers of Math1⁺ cells are increased in response to Ngn1 hemizygosity specifically at sites where the two genes are co-expressed (utricle and saccule).

Although, as described above, myosin VIIa expression domains are expanded in Ngn1 mutants (Fig. 4F-H), we found fewer myosin VIIa⁺ cells in the Ngn1^-/- utricle and Ngn1^+/- utricle and saccule than in wild type (Table 1). This was owing to a lower density of myosin VIIa⁺cells in the mutants as compared with wild type and, in addition, myosin VIIa⁺ hair cell size and shape were abnormal in the mutants (see Fig. S4D-F in the supplementary material). Neither TUNEL, nor anti-caspase 3 immunohistochemistry, at E14.5 provided evidence that these abnormalities were due to apoptosis (data not shown). Since the onset of myosin VIIa expression normally follows Math1 expression, and all of the mutant sites under consideration have more Math1⁺ cells than wild type, a considerable number of Math1⁺ macular cells in Ngn1 mutants (greater than 35%) must fail to express myosin VIIa⁺. Taken together, our results indicate that at sites of Ngn1 and Math1 co-expression (utricle and saccule), reduced Ngn1 gene dose causes excess numbers of Math1⁺cells to form, but many of these cells do not properly differentiate as hair cells.

Ngn1 is thought to function analogously to the Drosophilaproneural genes (Ma et al.,1996; Ma et al.,1998; Cornell and Eisen,2002). Consistent with a proneural function for Ngn1, we found that Ngn1-GFP signal varies in intensity between neighboring cells, and is strongest in delaminating cells (Fig. 5A). We therefore tested whether Ngn1 negatively regulates its own transcription in the otic epithelium by comparing expression patterns of the Ngn1-GFP BAC transgene in Ngn1-null homozygote and wild-type littermates prior to the appearance of ectopic Math1expression. At E12.5, Ngn1-GFP is normally expressed in a speckled pattern by a subset of cells in the neurogenic region(Fig. 5B). By contrast, in Ngn1^-/- embryos, the GFP signal was present in all cells of the region, with little variability in signal strength between cells(Fig. 5D). Loss of the speckled expression pattern for the Ngn1-GFP transgene did not occur on a Math1^-/- background(Fig. 5C), indicating that this phenotype is specific to the loss of Ngn1. We also found excess Ngn1-GFP⁺ and NeuroD⁺ epithelial cells in the Ngn1^+/- utricle and saccule compared with wild type(Table 1). In Ngn1^-/- ears, expression of the Ngn1-GFP transgene was completely abolished by E14.5, presumably owing to inhibition by ectopic Math1 expression (Fig. 4H,M).

To investigate potential interactions of Ngn1 with Notch signaling, we assayed expression of the Notch ligand Dll1 in the Ngn1-null homozygote. We found reduced Dll1 expression in the primitive utricle and saccule of mutants as compared with wild type(Fig. 5E,F) (see also Ma et al., 1998). To test whether Notch activity suppresses Ngn1 transcription within the otic epithelium, we analyzed Ngn1 expression in embryos lacking Pofut1. This gene encodes the protein O-fucosyltransferase 1, which glycosylates epidermal growth factor-like repeats within the extracellular domain of the Notch receptor (Lei et al.,2003; Okajima et al.,2003; Okajima et al.,2005). Pofut1 loss-of-function abolishes ligand-induced Notch signaling and causes phenotypes similar to those of embryos lacking downstream effectors of all Notch receptors, including mid-embryonic lethality(Shi and Stanley, 2003). We therefore assayed Ngn1 expression in early Pofut1 embryos(E9-9.5), when the otic placode invaginates to first form an otocyst. At these stages, Ngn1 mRNA signals were increased in Pofut1^-/- embryos as compared with wild-type littermates in the otic epithelium, midbrain, trigeminal placode, epibranchial placodes and spinal cord (Fig. 5G,H),indicating that Ngn1 transcription in all these embryonic regions is negatively regulated by canonical Notch signaling. Thus, the negative autoregulation of Ngn1 in the otic epithelium might be controlled by Notch-mediated lateral inhibition.

Math1 positively autoregulates its transcription at particular sites in the embryo (Helms et al.,2000). To determine whether Math1 is subject to positive autoregulation during ear development, transgenic reporters of Math1promoter activity were compared across wild-type and Math1^-/- backgrounds. A BAC that expresses GFP under the control of Math1 regulatory elements mimics patterns of Math1 mRNA expression in the developing ear(Fig. 6A), hindbrain region(Fig. 6A, inset), spinal cord,and other sites in the embryo (J.E.J., unpublished). By contrast, we were unable to detect GFP signal in the ears of Math1^-/-::Math1-GFP BAC embryos at stages E13.5 through E15.5 (Fig. 6B; data not shown), although GFP expression was clearly present at other sites of expression, such as the hindbrain (Fig. 6B, inset). A second transgenic line, which carries a 1.4 kb Math1 enhancer with an E-box site that is essential for Math1 binding and autoregulation in other tissues (Helms et al., 2000), also mimics Math1 expression in the developing wild-type ear (Chen et al.,2002; Lumpkin et al.,2003). As with the Math1-GFP BAC, we found no GFP reporter signal in the sensory epithelia of these Math1^-/-::Math1-GFP embryos at stages E13.5 through E15.5 (Fig. 6C,D). The complete lack of reporter expression in the cristae of both Math1^-/-::Math1-GFP reporter lines indicates that the phenotype is not due solely to inhibition by ectopic expression of Ngn1.

It has been unclear how neurogenic tissue of the otocyst relates to sensory epithelia of the inner ear. Gene expression studies in both chicken and mouse suggest that neurogenesis occupies only part of a larger sensory-competent domain of the otocyst (Morsli et al.,1998; Adam et al.,1998; Cole et al.,2000; Fekete and Wu,2002), and a recent lineage study in the chicken shows that vestibular and spiral (auditory) neurons of the VIIIth ganglion can be clonally related to utricular epithelial cells (both macular and non-sensory)(Satoh and Fekete, 2005). By fate mapping in mouse, we show that Ngn1⁺ otic epithelial cells can differentiate as vestibular or spiral ganglion neurons, hair or supporting cells of the utricular and saccular maculae, or non-sensory epithelial cells surrounding the maculae. Although the question remains open as to whether, in mouse, these cell types descend clonally from a common progenitor, our work unambiguously traces the origin of specific sensory(macular) and non-sensory cells to the neurogenic domain of the otocyst.

Our fate mapping suggests that the other functional class of vestibular sensory epithelia, the cristae, and their associated semicircular canals do not derive from the Ngn1⁺ domain of the otocyst. This confirms previous gene expression studies tracing the origin of cristae to a Bmp4⁺ region outside of, and adjacent to, the neurogenic domain (Morsli et al., 1998; Raft et al., 2004)(Fig. 7A′). The very rare occurrence (a few cells in one of 20 ears) of Ngn1 derivatives in the lateral crista suggests that mixing of cells between the neurogenic and Bmp4domains occurs infrequently. Whether this lack of mixing is due to differential affinity between the two regions, or whether neurogenesis is actively suppressed within the Bmp4 domain, is not clear. In support of the latter hypothesis, Tbx1, a T-box gene that inhibits Ngn1 and maintains Bmp4 expression in the otocyst epithelium, is expressed continuously and from very early stages in the presumptive and definitive cristae (Arnold et al., 2006; Raft et al.,2004; Vitelli et al.,2003).

Our mapping of Ngn1-GFP and NeuroD expression domains revealed no evidence of active neurogenesis in the definitive cochlea. However, we did find Ngn1 derivatives in a non-sensory region of the cochlea (the GER) in the majority of ears analyzed. Based on its location in the ear, the GER might derive from the most posteroventral-medial edge of the otocyst neurogenic region. Interestingly, the GER lies immediately adjacent to the organ of Corti, within which we found no Ngn1 derivatives. This result, the common occurrence of Ngn1 derivatives in non-sensory tissue between the utricula macula and the anterior/lateral cristae (but not in the cristae)(Fig. 2E,G), and the initiation of macular Math1 expression as stripes just within opposite borders of the neurogenic domain, support the hypothesis that sensory epithelia are induced at or near compartment boundaries in the otocyst(Fekete, 1996; Brigande et al., 2000).

We show that neurogenesis and hair cell production, long considered strictly sequential, actually overlap in the developing utricle and saccule for several days of gestation. During this period, neural precursors and nascent hair cells initially intermingle and later sort out across well-defined borders. Functionally, we show that Math1 and Ngn1 mutants have complementary inner ear phenotypes, supporting the hypothesis that mutual antagonism between these genes coordinates neurogenesis and hair cell production (Matei et al.,2005). Loss of Math1, which is normally expressed in all sensory regions of the ear, leads to excess and ectopic neurogenesis only in sensory regions with a history of Ngn1 expression (utricular and saccular maculae). This effect is gene dose-sensitive, as Math1heterozygotes exhibit a neurogenic phenotype intermediate to those of the Math1-null homozygote and wild type. Conversely, Ngn1hemizygosity causes excess and ectopic Math1 expression specifically in the utricle and saccule, and although Ngn1-null homozygosity causes growth abnormalities of the ear, the Ngn1^-/-utricle still shows a phenotype of excess and ectopic Math1expression. These effects are seen only at sites in the developing ear where Ngn1 and Math1 are co-expressed, and we propose that they result from a disruption of close-range cross-inhibition. Cross-inhibition might influence multiple steps in the process, whereby an Ngn1⁺ progenitor field of dual competence (neural and sensory epithelial) is gradually restricted to producing only sensory epithelial cells. These include: (1) Math1 domain establishment within opposite borders of the Ngn1⁺ region(Fig. 7A′); (2) Math1 domain expansion and decline in Ngn1 expression(Fig. 7A′,A″); and(3) compartmentalization of the region into a pair of adjacent Math1(sensory) and Ngn1 (neurogenic) domains(Fig. 7A″). The potential basis for the competitive advantage of Math1 over Ngn1 in this system is discussed below.

Criteria for proneural function include early, broad expression of transcript in all cells of a germinal epithelium and subsequent refinement of transcription to a subset of cells by lateral inhibition(Jan and Jan, 1993; Lewis, 1996). We find no evidence of these features in our studies of Math1 expression in the vestibular system of the mouse. Of the two genes relevant to this study, it is Ngn1 and not Math1 that initially marks the prospective maculae and exhibits the variegated expression among neighboring cells that is characteristic of proneural genes (Fig. 5A,B). Furthermore, using two different transgenic reporter lines,we show that Math1 is required for detectable levels of Math1 reporter expression in the otic epithelium, suggesting that Math1 promoter activity is amplified and maintained by positive autoregulation. One possible consequence of this is a rapid and irreversible commitment of progenitors to the hair-cell fate once Math1transcription surpasses a threshold for positive autoregulation. Our results thus support the view that Math1 functions as a hair-cell commitment factor rather than a proneural (or `prosensory') gene(Chen et al., 2002) (for a review, see Kelley, 2006). Interestingly, in zebrafish, which has two atoh1 genes, differences in the timing and autoregulation of Math1/atoh1 genes from that described here lead to the opposite conclusion(Millimaki et al., 2007). For example, zebrafish atoh1a and 1b are required for hair cell generation, but their expression precedes that of ngn1(Andermann et al., 2002) and marks the prospective maculae from very early stages. Gene duplication and evolutionary pressure on the regulatory genome might therefore dictate the precise functions of Math1/atoh1 during ear development in different species.

Our experiments reveal a profile of Ngn1 autoregulation very different from that of Math1. We demonstrate that Ngn1 is required to limit its own transcription within the otic epithelium. We also extend a previous observation that proper otic expression of the Notch ligand Dll1 is dependent on Ngn1(Ma et al., 1998) and show a pattern of increased Ngn1 expression in the early otic epithelium of Pofut1^-/- embryos, which are deficient in canonical Notch signaling (Shi and Stanley,2003). These results and the expression of Ngn1 in neural and sensory progenitors from very early stages fulfill several criteria for proneural function. Likely consequences of the proneural activity of Ngn1 are control over the pace of neurogenesis during otocyst stages and preservation of an uncommitted progenitor cell population for sensory development. These functions are consistent with recently reported effects of conditional Dll1 loss-of-function in the developing ear, which include an enlarged ganglion rudiment and specific hypoplasia of the utricle and saccule (Brooker et al.,2006), and with blockade of Notch signaling in the chicken, which causes excess neurogenesis at the expense of sensory epithelial precursors(Daudet et al., 2007).

Given our evidence for a mutual antagonism between Ngn1 and Math1, how does Math1 exert the stronger inhibitory activity so that sensory epithelia replace an active neurogenic region? Our model states that Ngn1 promotes a neural fate cell-autonomously and keeps its own expression low or off in neighboring cells through Notch-mediated lateral inhibition (Fig. 7B,C). Cells expressing high levels of Ngn1 delaminate from the epithelium as neural precursors. Cells remaining within the neurogenic epithelium constitute a dynamic mix of committed neural precursors and uncommitted progenitors. The latter group may adopt a neural fate in subsequent rounds of delamination or may remain uncommitted for several days, after which they adopt hair or supporting cell fates in response to Math1 induction within the region (Fig. 7B′,B″,C′). This is supported by our fate mapping results, as Ngn1 derivatives can have any of these identities. Once Math1 transcription exceeds a particular threshold, positive autoregulation irreversibly commits progenitors to a hair cell fate, and committed hair cells may then induce the supporting cell phenotype through intercellular signaling (Woods et al.,2004). Since strongly Ngn1⁺ cells continuously delaminate from the epithelium, Math1-expressing cells are left to interact with epithelial progenitors expressing lower levels of Ngn1. These features might bias the mutual antagonism between Ngn1 and Math1, thereby promoting sensory epithelial differentiation at the expense of continued neurogenesis.

We have implicated cross-regulation between bHLH genes and differential autoregulation as mechanisms for converting a neurogenic epithelium into specialized mechanosensory receptors. A novel aspect of this work - and one that is potentially relevant to other systems - is the dynamic nature of the patterning processes described. We show that progressive regionalization of bHLH genes through cross-inhibition can result in a sequential and overlapping production of distinct cell types, and that differential autoregulation might provide the driving force for such a transition.

Many questions remain unanswered. For example, does cross-inhibition between Ngn1 and Math1 occur within a single cell, through intercellular signaling, or by a combination of these two mechanisms?Cell-autonomous cross-inhibition might convert a weakly Ngn1⁺ cell directly into a Math1⁺nascent hair cell. Alternatively, if the antagonism occurs through intercellular signaling, Ngn1⁺ cells might pass through a`sensory-restricted progenitor' state before committing to the hair-cell fate(Fig. 7C′). We find the latter alternative attractive given that embryonic maculae contain many Ngn1 derivatives with a pseudostratified epithelial (non-hair-cell) phenotype. Molecular and cellular mechanisms underlying the apparent compartmentalization of sensory and neurogenic regions also warrant scrutiny, as there is abundant evidence that Notch-mediated intercellular signaling occurs at nascent boundaries during development (Irvine,1999). In summary, our results form a basis for understanding how progenitors are allocated to various cell fates during inner ear development.

We thank Huda Zoghbi and Pamela Stanley for providing Math1 and Pofut1 mutant mice; Jackie Lee, Qiufu Ma and Johannes Becker for in situ probes; and Juan Llamas, Welly Makmura and Sheri Juntilla for excellent colony maintenance. This work was supported by grants from the Mathers Charitable Foundation, the Alfred P. Sloan Foundation and the Medical Foundation (L.V.G.), and by NIH grants F32 DC007247 (S.R.), F31 DC007775(E.J.K.), HD037932 and NS048887 (J.E.J.) and DC006185 (A.K.G. and N.S.).