Neurog1 can partially substitute for Atoh1 function in hair cell differentiation and maintenance during organ of Corti development

Neurosensory development requires the sequential, coordinated activation and cross-regulation of numerous transcription factors (TFs) to define precursors and initiate differentiation of the various cell types of the nervous and sensory system (Fritzsch et al., 2015; Imayoshi and Kageyama, 2014; Reiprich and Wegner, 2015). Molecularly dissecting these interactions requires model systems with limited cellular diversity and stereotyped cellular patterning. The organ of Corti (OC) is such a model system, with hair cells (HCs) and supporting cells (SCs) organized into the most stereotyped cell assembly of vertebrates (Slepecky, 1996), and is suited to detect minute aberrations (Jahan et al., 2013). The stereotyped cellular pattern may allow the molecular dissection of the intricate interaction of multiple basic helix-loop-helix (bHLH) proteins (Benito-Gonzalez and Doetzlhofer, 2014; Fritzsch et al., 2010b) to define HCs/SCs.

Targeted deletion studies in mice have demonstrated that three bHLH TFs (Neurog1, Neurod1, Atoh1) are essential to differentiate neurons and HCs of the inner ear (Bermingham et al., 1999; Fritzsch et al., 2006,, 2010b). These loss-of-function analyses also revealed cross-regulation among these bHLH TFs. For example, Neurog1 null mice showed loss of HCs (Ma et al., 2000; Matei et al., 2005), apparently through alteration of Atoh1 expression (Raft et al., 2007). Absence of Neurog1 reduced HCs, whereas absence of Neurod1 resulted in ectopic HCs in ganglia (Jahan et al., 2013). Changes in Atoh1 expression may be mediated through cross-regulation of Neurod1 downstream of Neurog1 (Jahan et al., 2010). Neurod1, in turn, may suppress Atoh1 in the ear (Jahan et al., 2013), comparable to its role in the cerebellum (Pan et al., 2009).

Other loss-of-function studies showed that Atoh1 drives HC differentiation (Bermingham et al., 1999; Fritzsch et al., 2005; Pan et al., 2011), and overexpression of Atoh1 transformed non-sensory cells into HCs (Kelly et al., 2012; Zheng and Gao, 2000). Atoh1 differentiates HCs in the ear, and the level and duration of Atoh1 expression regulate different types of HCs and their viability (Jahan et al., 2013; Sheykholeslami et al., 2013). Consistent with these data, self-terminating Atoh1 (Atoh1-Cre; Atoh1^f/f) mice initiated near normal HC differentiation but rapidly lost most HCs by 3 weeks (Pan et al., 2012a), a loss also reported in inducible Cre lines (Cai et al., 2013; Chonko et al., 2013). Replacing both alleles of Atoh1 with another bHLH TF, Neurog1 (Atoh1^{kiNeurog1/kiNeurog1}), resulted in very few immature HCs bearing microvilli with a central kinocilium (Jahan et al., 2012). This inability of Neurog1 to maintain and differentiate HCs beyond that achieved with even transient Atoh1 expression (Pan et al., 2012a) contrasts with work on retinal ganglion cells (RGCs), in which the Atoh1 paralog, Atoh7, was replaced by Neurod1 (Mao et al., 2008). Neurod1 can replace Atoh7, possibly because RGC precursors are pre-programmed to differentiate as RGCs independently of the type of bHLH TF (Mao et al., 2013). In molecular terms, the pairing of Atoh1/Neurog1 is as different as that of Atoh7/Neurod1 (52% versus 60% sequence similarity), suggesting that overall binding differences might not fully explain these very different results in eyes and ears. We reasoned that failure of functional replacement of Atoh1 by Neurog1 in the ear (Jahan et al., 2012), compared with Atoh7 by Neurod1 in the retina (Mao et al., 2008), or Atoh1 by fly atonal (Wang et al., 2002), could relate either to ‘self-regulation’ of Atoh1 via its enhancer (Helms et al., 2000) or to a unique functional requirement of Atoh1/Atonal protein to initiate HC differentiation.

To test these possibilities, we generated a new mouse model (Atoh1-Cre; Atoh1^f/kiNeurog1) with insertion of Neurog1 in one allele of Atoh1 (Jahan et al., 2012) and removal of the second floxed Atoh1 allele by self-termination with Atoh1-Cre (Pan et al., 2012a). This novel composite mouse mutant can differentiate most HCs and can rescue many HCs for up to 9 months. However, HCs stereocilia are variably disorganized, OC patterning is disrupted, and mice are deaf despite near normal numbers of rather well differentiated HCs.

We previously demonstrated that Neurog1 replacement in both alleles of Atoh1 resulted in expression of Neurog1 in HC precursors that differentiate multiple microvilli (Jahan et al., 2012). Like Atoh1 null mutant mice (Bermingham et al., 1999), homozygous Neurog1 knock-in mice die after birth, precluding postnatal analysis. Atoh1 autoregulates its own expression by activating an Atoh1-specific enhancer (Helms et al., 2000). We speculated that replacement of Atoh1 by Neurog1 in both alleles might have failed to activate the enhancer for adequate Neurog1 expression under Atoh1 promoter control (Jahan et al., 2012). To overcome this problem, we generated a novel mouse model that combined Neurog1 knock-in in one Atoh1 allele with a floxed second Atoh1 allele, the latter to be excised by Atoh1-Cre (Atoh1-Cre; Atoh1^f/kiNeurog1; Fig. 1A). This newly developed Atoh1-Cre; Atoh1^f/kiNeurog1 model resulted in viable mice, overcoming the neonatal lethality of the homozygous Neurog1 knock-in (Jahan et al., 2012). Atoh1^f/+ or Atoh1^f/f mice without Atoh1-Cre were used as a control in this study except for the RT-qPCR analysis, where Atoh1^+/kiNeurog1 mice were used as control for comparison of Neurog1 expression with Atoh1-Cre; Atoh1^f/kiNeurog1 mice.

In situ hybridization (ISH) showed that Neurog1 knock-in into the Atoh1 locus (Atoh1-Cre; Atoh1^f/kiNeurog1) resulted in expression of Neurog1 in OC HCs not detected in control littermates (Fig. 1B,B′). At P0, Atoh1-Cre; Atoh1^f/kiNeurog1 mice showed very little Atoh1 expression in the apex (Fig. 1C,C′). By contrast, a downstream target of Atoh1, Pou4f3, was strongly expressed in the HCs of P0 Atoh1-Cre; Atoh1^f/kiNeurog1 OC (Fig. 1D,D′), whereas expression of Barhl1 was delayed compared with the vestibular epithelia at P0 (Fig. 1E,E′). At P7, qPCR showed that relative mRNA expression for Neurog1, Pou4f3 and Barhl1 in Atoh1-Cre; Atoh1^f/kiNeurog1 cochlea was reduced compared with Atoh1^+/kiNeurog1 cochlea, but the reduction was not statistically significant (Atoh1^+/kiNeurog1 expression was normalized to 1, Fig. 1F). In self-terminating Atoh1-Cre; Atoh1^f/f mice, ISH of both Pou4f3 and Barhl1 showed patchy loss at P0 (Pan et al., 2012a). By contrast, Neurog1-misexpressing mice revealed expression of both Pou4f3 and Barhl1 until P7 (Fig. 1F). Partial expression of Pou4f3 and Barhl1 might improve HC viability, as HCs are rapidly lost in either Pou4f3 (Hertzano et al., 2004; Xiang et al., 2003) or Barhl1 (Li et al., 2002) null mutants.

Immunohistochemistry of Myo7a (an HC-specific marker) at P7 revealed the formation of an almost continuous row of inner hair cells (IHCs) and two to three rows of outer hair cells (OHCs) in the Neurog1-misexpressing mice (Fig. 2). There were occasionally two rows of IHCs. Extra rows of IHCs coincide with absence of inner pillar (IP) cells, as shown by immunohistochemistry of acetylated tubulin (which is abundant in SCs), next to the extra row of IHCs (Fig. 2B-C‴). There was no indication that the length of the cochlea was changed (supplementary material Table S1). The third row of OHCs disappeared toward the base of the cochlea, including the formation of some gaps in OHCs (Fig. 2C-D‴). Tubulin immunohistochemistry labeled differentiated IP cells, outer pillar (OP) cells and many Deiters' cells at P7 (Fig. 2). Overall, Neurog1 misexpression resulted in a considerable increase in HC and SC formation as compared with Atoh1-Cre; Atoh1^f/f mice, which mostly had one row of OHCs and only a few IHCs at P7 (Fig. 3).

We conclude that Neurog1 misexpression combined with transient Atoh1 expression can partially substitute for Atoh1 to differentiate and maintain HCs, as compared with the massive loss of nearly all HCs in self-terminating mice (Pan et al., 2012a). Diminished long-term viability of HCs correlates with reduced expression of TFs known to be essential for HC viability (Pou4f3, Barhl1).

We next quantified the Myo7a-positive HCs, identifying IHCs as being medial and OHCs lateral to the tubulin-immunopositive IP/OP cells. We counted all cells in a 300 µm stretch near the apex (10%), near the middle (50%) or near the base of the cochlea (90%) in comparable segments from control, Atoh1-Cre; Atoh1^f/kiNeurog1 and Atoh1-Cre; Atoh1^f/f littermates (Fig. 3D,E; supplementary material Table S2; n=6). The numbers of IHCs were significantly higher in Atoh1-Cre; Atoh1^f/kiNeurog1 than in Atoh1-Cre; Atoh1^f/f mice (Fig. 3D; P<0.01). IHC counts between Atoh1-Cre; Atoh1^f/kiNeurog1 mice and control littermates showed no significant difference (Fig. 3D). In summary, replacement of one allele of Atoh1 by Neurog1 combined with a self-terminating second Atoh1 allele rescued most IHCs as compared with the massive loss of IHCs in the Atoh1-Cre; Atoh1^f/f mouse.

The quantification of OHCs demonstrated that significantly more OHCs form in Atoh1-Cre; Atoh1^f/kiNeurog1 compared with Atoh1-Cre; Atoh1^f/f mice (Fig. 3E; in the apex, P<0.05; in the middle and base of the cochlea, P<0.01). In contrast to IHCs, OHCs in Atoh1-Cre; Atoh1^f/kiNeurog1 mice were significantly reduced compared with control littermates (Fig. 3E; in the apex, P<0.05; in the middle and base of the cochlea, P<0.01). Misexpression of Neurog1 increased survival preferentially of IHCs, with the greatest reduction in the third row of OHCs (Fig. 2C-D‴, Fig. 3B,E). Previous work on self-terminating Atoh1 conditional null mice (Pan et al., 2012a) and Atoh1 hypomorphs (Sheykholeslami et al., 2013) showed a differential loss of OHCs, with the third row being most susceptible, whereas tamoxifen-induced Atoh1-Cre wiped out all HCs rapidly (Cai et al., 2013; Chonko et al., 2013).

We next investigated the longevity of these HCs using Myo7a and tubulin immunohistochemistry (Fig. 4). At 9 months, many Myo7a-positive HCs remained in the Atoh1-Cre; Atoh1^f/kiNeurog1 mice. In particular, many IHCs (medial to the tubulin-positive IP cells) survived, but OHCs showed patchy loss (Fig. 4B-B″). This contrasted with severe HC loss in Atoh1-Cre; Atoh1^f/f mice, where only a few HCs survived up to 9 months (Fig. 4C-C″), essentially being largely lost by 3 weeks (Pan et al., 2012a). In addition to HCs, SCs also persisted longer in Atoh1-Cre; Atoh1^f/kiNeurog1 mice and many more tubulin-positive pillar cells and Deiters’ cells were observed compared with Atoh1-Cre; Atoh1^f/f mice (Fig. 4B′,B″,C′,C″).

The proper arrangement of cells is essential for the function of the OC to enable sound perception (Cai et al., 2003; Jacobo and Hudspeth, 2014). To assess hearing function in Neurog1-misexpressing mice, we measured the auditory brainstem response (ABR) to click stimuli at P30 (Fig. 5). Atoh1-Cre; Atoh1^f/kiNeurog1 mice showed no ABR, in contrast to control littermates (Fig. 5). We concluded that the expression of Neurog1 facilitates the development and maintenance of many HCs in a non-functional OC. We next investigated possible reasons for this functional defect, focusing first on the stereocilia of HCs.

HC function depends on the normal development of the stereocilia bundles that mediate mechanotransduction (Hudspeth, 2014; Jacobo and Hudspeth, 2014; Sienknecht et al., 2014). Loss, alteration or reduction of Atoh1 results in abnormal stereocilia bundle formation (Chonko et al., 2013; Pan et al., 2012a), if any bundle forms at all (Pan et al., 2011). Atoh1 dosage and timing of expression are important for proper stereocilia development and maturation (Jahan et al., 2013). Several mutations that interfere with stereocilia bundle organization and homeostasis are known (Hertzano et al., 2008; Kitajiri et al., 2010; Mogensen et al., 2007; Sekerková et al., 2011) and many of these result not only in aberrant bundle morphology but also in the death of HCs (Kersigo and Fritzsch, 2015; Self et al., 1998; Ueyama et al., 2014). Consistent with a role of Atoh1 in stereocilia differentiation, forced expression of Atoh1 can restore hair bundles (Yang et al., 2012).

To extend our findings beyond the obvious increase in HC survival (Figs 2–4) and to better understand the apparent deafness (Fig. 5), we next investigated how replacement of Atoh1 with Neurog1 influences stereocilia formation and thus the function of HCs. We examined at least three cochlea of the Atoh1-Cre; Atoh1^f/kiNeurog1 mutant and littermate controls, each at three different stages (P1, P7 and P22), by scanning electron microscopy (SEM) (Figs 6 and 7). We found that the onset of stereocilia bundle differentiation in Neurog1-misexpressing mice was delayed (Fig. 6A-B″). The bundles were immature, with the formation of microvilli that were all similar in length surrounding a central kinocilium at P1 (Fig. 6B-B″). Control littermates already displayed the staircase pattern of progressively increasing length of stereocilia modiolar to an acentric kinocilium (Fig. 6A). This stunted stereocilia growth was more obvious in OHCs than IHCs, especially in the second and third rows of OHCs of Atoh1-Cre; Atoh1^f/kiNeurog1 mice. In Atoh1-Cre; Atoh1^f/kiNeurog1 mice at P7, the bundles formed a nearly normal staircase pattern in the apex of the cochlea, whereas the basal HCs displayed aberrations such as fusion or loss of stereocilia, predominantly in the third row of OHCs (Fig. 6C-D″).

SEM revealed progressive deformations of the stereocilia bundles in Atoh1-Cre; Atoh1^f/kiNeurog1 mice (Fig. 7). Some HCs in the Neurog1-misexpressing mice had stereocilia bundles that lacked the staircase organization, and some stereocilia were fused with each other or showed stunted growth (cyan arrows in Fig. 7). SEM also revealed that the thickness of the stereocilia of some IHCs in Atoh1-Cre; Atoh1^f/kiNeurog1 mice was altered to resemble OHCs, indicating partial cell fate switching of some IHCs to OHCs (red arrows in Fig. 7). Beyond the fusion and ectopic stereocilia formation, SEM in the Neurog1-misexpressing mice also revealed disruption of the boundaries between adjacent HCs (Fig. 7). Some HCs touched each other without intervening SCs in Atoh1-Cre; Atoh1^f/kiNeurog1 mice (green arrows in Fig. 7). At P7 and P22, we found ectopic stereocilia bundles protruding from the apical surfaces of the IP cells (yellow arrows, Fig. 7). This altered pattern of HCs and SCs indicated disruptions of cell-cell interactions. We therefore further analyzed the molecular basis of OC pattern formation (Fritzsch et al., 2014b; Groves and Fekete, 2012), focusing on genes known to affect OC development.

We previously demonstrated conversion of OHCs into IHCs in Neurod1 conditional deletion mice (Jahan et al., 2010). Absence of Neurod1 resulted in premature and altered Atoh1 and Fgf8 expression and transformation of thin OHC stereocilia into thick IHC stereocilia (Jahan et al., 2010, 2013). Essentially, Atoh1 and Fgf8 expression failed to be suppressed by Neurod1 in Neurod1 conditional deletion mice (Jahan et al., 2010). We investigated the expression of Neurod1 and Fgf8 in the Atoh1-Cre; Atoh1^f/kiNeurog1 mice, as they showed the opposite effect, i.e. conversion of stereocilia of IHCs to OHCs. Misexpression of Neurog1 resulted in stronger and earlier expression of Neurod1 in E16.5 Atoh1-Cre; Atoh1^f/kiNeurog1 cochlea as compared with control littermates (Fig. 8A-B″). Fgf8 expression was patchy and much reduced in some IHCs of the E16.5 Atoh1-Cre; Atoh1^f/kiNeurog1 mice (Fig. 8D′; supplementary material Fig. S1B′). The alteration of HC stereocilia diameter in Atoh1-Cre; Atoh1^f/kiNeurog1 mice might relate to the locally altered level of Atoh1/Neurog1 signaling that distorted the specificity of HC types. How alteration of Atoh1 signal level mechanistically regulates the thickness of the stereocilia bundles, possibly through modulation of whirlin (Mogensen et al., 2007), requires further work.

Among microRNAs (miRNAs), miR-96 was specifically reported as being required for the proper maturation and organization of the stereocilia bundle (Kuhn et al., 2011; Lewis et al., 2009). Single base mutation in the MIR96 gene results in non-syndromic, progressive hearing loss in humans (Kuhn et al., 2011; Lewis et al., 2009). miR-96 is expressed in the sensory epithelia of the mouse cochlea (Weston et al., 2011). We previously reported very limited miR-96 expression only in the apex of Atoh1^{kiNeurog1/kiNeurog1} cochlea (Jahan et al., 2012) or its near absence in conditional Atoh1 null mice (Pan et al., 2011). Both mutants showed no HC differentiation beyond precursors cells. In the Atoh1-Cre; Atoh1^f/kiNeurog1 mice, miR-96 expression was almost normal, except for the basal hook region where it showed some gaps (Fig. 8E-F′). How possible alterations in some miRNAs tie into stereocilia development through the repression of relevant genes remains speculative.

Given the formation of stereocilia bundles on IP cells (Fig. 7), we next investigated the distribution of an IP cell-specific marker, p75 (Ngfr) (von Bartheld et al., 1991). ISH of p75 confirmed its selective expression in IP cells in P0 control mice (Fig. 9A). In P0 Atoh1-Cre; Atoh1^f/kiNeurog1 mice, p75 expression was patchy near the base (Fig. 9B′,B″) but continuous in the apical half (Fig. 9B) of the cochlea. Consistent with stereocilia bundles on IP cells (Fig. 7A-C, E), immunohistochemistry for the HC-specific Myo7a subsequent to p75 ISH showed that the gaps in the p75-positive IP cells were filled with Myo7a-positive HCs (Fig. 9B″). We also performed dual immunohistochemistry using p75 and Myo7a antibodies with different fluorophore labeling of the secondary antibodies. This also showed that the gaps in p75-immunopositive IP cells were filled with Myo7a-positive putative HCs (Fig. 9C-D″). Combined with stereocilia formation in the apical surfaces of the IP cells of Atoh1-Cre; Atoh1^f/kiNeurog1 mice, our data suggested that expression of Neurog1 transformed some IP cells into a hybrid of IP and HC.

To further assess this possibility, we quantified the tubulin-immunopositive IP cells in P7 Atoh1-Cre; Atoh1^f/kiNeurog1 mice (Fig. 9E). There was a reduction of tubulin-positive IP cells in the Neurog1-misexpressing mice relative to control littermates in the apex (P<0.05) and middle (P<0.01) turn of the cochlea. Despite this presumed transformation, overall IP cell formation was increased in Atoh1-Cre; Atoh1^f/kiNeurog1 relative to Atoh1-Cre; Atoh1^f/f mice (P<0.01). Although p75 is an excellent marker for IP cells, differentiation of IP cells remains normal in p75 null mice and its function in IP cells remains obscure (Tan et al., 2010).

Pillar cell formation is abnormal in Fgf8 null mice (Jacques et al., 2007; Mueller et al., 2002). As previously reported (Jacques et al., 2007; Pirvola et al., 2000), Fgf8 is expressed in all IHCs. In the Atoh1-Cre; Atoh1^f/kiNeurog1 mice, downregulation of Fgf8 expression might affect the strength of diffusible signal from IHCs to nearby SCs (Groves and Fekete, 2012). Given the defects in IP cells, we next performed double ISH for Fgf8 and p75 and revealed that the reduction of Fgf8 expression correlated with that of p75 in Atoh1-Cre; Atoh1^f/kiNeurog1 mice (supplementary material Fig. S1A-C′). Previous work also demonstrated that Atoh1 is transiently expressed in IP cells (Driver et al., 2013; Matei et al., 2005). Signals that prevent IP cells from differentiating as HCs might be disrupted in the Atoh1-Cre; Atoh1^f/kiNeurog1 mice, resulting in partial transdifferentiation of some IP cells into HCs (Fig. 7).

We also investigated the Fgf10 and Bmp4 expression that flanks the medial and lateral borders of the OC, respectively (supplementary material Fig. S1D,D′). In P0 Atoh1-Cre; Atoh1^f/kiNeurog1 mice, Fgf10 was drastically reduced and Bmp4 showed expanded expression toward OC, with gaps where OC cells were replaced by simple epithelium (supplementary material Fig. S1D,D′).

To obtain further molecular insight into SC differentiation, we studied Prox1 expression by ISH in the Neurog1-misexpressing mice (Fritzsch et al., 2010a). Prox1 was expressed in almost all SCs, except for some reduction in the IP cells in P0 Atoh1-Cre; Atoh1^f/kiNeurog1 mice (Fig. 10A-B″). These data suggested that several reliable markers of SC differentiation were altered, lost or even replaced by HC markers, indicating that expression of Neurog1 in HCs affected other cells of the OC.

Atoh1 expression in HCs regulates the expression of genes in the Delta/Notch signaling pathway that are necessary for the development of the surrounding SCs and maintains the proper patterning of the OC (Kobayashi and Kageyama, 2014; Sprinzak et al., 2011; Yamamoto et al., 2014). Lack of the Notch ligand Jag1 results in extra rows of IHCs and the loss of OHCs (Brooker et al., 2006; Kiernan et al., 2006). The Hes/Hey factors, which are downstream target genes of Notch, play important roles in the proper specification of SCs by negatively regulating pro-neuronal bHLH TFs (Doetzlhofer et al., 2009; Zine and de Ribaupierre, 2002). Given the lack of proper patterning of HCs/SCs in the OC, we investigated the expression of Jag1 and Hes5 by ISH in Atoh1-Cre; Atoh1^f/kiNeurog1 mice (Fig. 10). In the P0 control mice, Jag1 was widely expressed in the greater epithelial ridge (GER) and in SCs, particularly in the IP cells (Fig. 10C,C′). In Atoh1-Cre; Atoh1^f/kiNeurog1 mice, the expression of Jag1 was slightly reduced in both the GER and SCs, with some patchy loss toward the cochlear base (Fig. 10D,D′).

At E14.5, Hes5 was found in the mid-base of control cochlea (supplementary material Fig. S2A). E14.5 Atoh1-Cre; Atoh1^f/kiNeurog1 mice had delayed Hes5 expression (supplementary material Fig. S2D); Hes5 was upregulated at later stages (supplementary material Fig. S2E,F) and showed patchy expression in the GER and in SCs (Fig. 10E-F″; supplementary material Fig. S2E,F). We combined Hes5 ISH with Myo7a immunohistochemistry to detect whether the downregulation of Hes5 was associated with HC differentiation in the Neurog1-misexpressing mice (supplementary material Fig. S2G-H″), as previously reported after Hes5 loss (Zine and de Ribaupierre, 2002). Myo7a-positive HCs showed no correlation with the patchy loss of Hes5 expression in this mutant (supplementary material Fig. S2G-H″).

Alteration of Jag1 and Hes5 expression in Atoh1-Cre; Atoh1^f/kiNeurog1 mice indicated the potential effects of Neurog1 on the expression of these genes, and thus on Delta/Notch signaling. This altered HC-SC communication might disrupt OC patterning (Fig. 7), but did not result in an overproduction of HCs (supplementary material Fig. S2H-H″), unlike in Hes/Hey mutants (Benito-Gonzalez and Doetzlhofer, 2014; Zine et al., 2001). How changes in intracellular HC gene expression through replacement of Atoh1 by Neurog1 alter intercellular signaling and disrupt the mosaic of the OC remains unclear.

In conclusion, near normal numbers of HCs can be generated by combining transient Atoh1 expression with Neurog1 misexpression, but these HCs have variably defective stereocilia bundles and the OC is disorganized. Despite long-term viable HCs, the abnormal OC organization causes deafness. OC disorganization may be a consequence of changes in Fgf8 expression, which might lead to the formation of secondary signaling centers comparable to the midbrain hindbrain boundary (Fritzsch et al., 2014a; Lee et al., 1997), or may also be due to altered cellular interactions (Groves and Fekete, 2012).

A fundamental aim in the study of neurosensory development, and in the context of regeneration in particular, is understanding the molecular basis for the generation of topologically distinct cell fates from uniform progenitor populations (Fritzsch et al., 2014a; Imayoshi and Kageyama, 2014; Reiprich and Wegner, 2015). Precise spatial and temporal control of gene expression by different combinations of TFs establish the molecular code that determines cell fate (Guillemot, 2007). Molecular dissection of this complexity requires models of limited cellular diversity in a stereotyped arrangement to evaluate minute deviations from normal; for example, in the ommatidia of flies (Johnston and Desplan, 2014). The OC of the mammalian inner ear is another excellent model organ with stereotyped cellular patterning that allows the exploration of molecularly induced deviations of developmental processes mediated by intracellular and extracellular patterning processes (Fritzsch et al., 2014b; Groves and Fekete, 2012). Our data suggest a profound effect of intracellular signals via cell-cell interactions and alterations in diffusible factors on the cellular and organ patterning process.

One approach to probing the signal specificity of a given TF in developing gene regulation networks is to replace them by closely related TFs (Guillemot, 2007). Replacing Atoh7 with Neurod1 leads to normal differentiation of RGCs, whereas replacing Neurod1 by Atoh7 alters the cell fate of amacrine and photoreceptor cells into RGCs (Mao et al., 2013,, 2008), indicating context dependency of gene actions. We knocked Neurog1 into the Atoh1 locus to test whether a bHLH TF that is exclusively associated with proliferative precursors in the ear and brain (Imayoshi and Kageyama, 2014; Ma et al., 2000) can function in differentiation to maintain or alter HC precursor differentiation. We previously showed that Atoh1^{kiNeurog1/kiNeurog1} can effectively drive Neurod1 in HC precursors (Jahan et al., 2012) but can neither initiate normal HC development nor maintain the viability of HC precursors. We also showed that self-terminating Atoh1 results in very limited viability of HCs, with incomplete stereocilia differentiation (Pan et al., 2012a). Our quantitative assessment of HC formation in Atoh1-Cre; Atoh1^f/kiNeurog1 mice in this study indicates that Neurog1 misexpression partially rescues HCs and maintains HCs for a longer period, as compared with self-terminating Atoh1 conditional null mice (Pan et al., 2012a). Neurog1 cannot maintain the basal third row of OHCs that depends on Fgf20 released from SCs (Huh et al., 2012), a factor that might be affected in our mice due to alteration in SC development.

Our data suggest that HC precursors behave like RGCs (Mao et al., 2008) and show limited flexibility to respond to the distantly related bHLH TF Neurog1 beyond enhanced HC differentiation relative to that of transient Atoh1 expression. This differs from spiral ganglion neurons, which differentiate readily as HCs if suppression of Atoh1 is removed by eliminating Neurod1 (Jahan et al., 2010). Atoh1 is required for an uncharacterized initial step of HC differentiation that neither Neurog1 nor Neurod1 can replace, and thus no HCs differentiate in homozygotic Atoh1^{kiNeurog1/kiNeurog1} mice (Jahan et al., 2012). The fly atonal gene can replace Atoh1 (Wang et al., 2002) and HCs can partially differentiate with the reduced (Sheykholeslami et al., 2013) or transient (Pan et al., 2012a) expression of Atoh1, or even in the absence of Atoh1 in chimeric mice (Du et al., 2007). Elucidating the molecular basis of this critical step of downstream gene activation (Cai et al., 2015; Pan et al., 2012b), in which Atoh1 expression cannot be substituted by Neurog1, could help to transform stem cells more effectively into HCs (Ronaghi et al., 2014; Zine et al., 2014).

Atoh1 cooperates with unknown factors to fully express Pou4f3 (Ahmed et al., 2012), which, in turn, cooperates with Atoh1 to maintain HCs (Chen et al., 2015; Hertzano et al., 2004; Masuda et al., 2012; Xiang et al., 2003). In contrast to Pou4f3, Barhl1 expression depends exclusively on Atoh1 (Chellappa et al., 2008). Barhl1 is required for HC survival even in the presence of Atoh1 and Pou4f3 (Li et al., 2002). Atoh1-Cre; Atoh1^f/kiNeurog1 mice have near normal expression of Pou4f3, but show a delay and progressive reduction of Barhl1 (Fig. 1). We suggest that Neurog1 might activate a second pathway for near normal Pou4f3 expression (Ahmed et al., 2012; Masuda et al., 2012), but fails to fully activate Barhl1 needed for HC maintenance (Fig. 11). Our mouse model will prove useful in testing the ability of putative regulators of downstream genes to enhance HC viability through the regulation of Pou4f3 and Barhl1 expression.

Closer investigation of the cochlea of Neurog1-misexpressing mice revealed various irregularities in stereocilia bundles and their distribution in the OC (Figs 6 and 7): the irregular length of individual stereocilia within the bundles; the uncoupling of stereocilia diameter from HC type; the appearance of ectopic stereocilia bundles on IP cells; and stereocilia bundles being abnormally fused with each other.

Stereocilia are necessary for hearing (Müller and Barr-Gillespie, 2015). Despite the rescue of overall HC formation, Atoh1-Cre; Atoh1^f/kiNeurog1 mice show no ABR and are deaf (Fig. 5). Whether this is due to some defect in the HCs or to the disorganization of the OC requires future single-cell recordings on isolated HCs. We presume that the irregularities in stereocilia bundles, which are essential for mechanoelectric transduction (Hudspeth, 2014), preclude the normal function of many HCs. Loss of Pou4f3, but not of Barhl1, causes stereocilia bundle aberrations (Chellappa et al., 2008; Hertzano et al., 2004). A reduced level of Pou4f3 transcripts at or after P7 might contribute to the bundle aberration that develops mostly in late postnatal stages in Neurog1-misexpressing mice (Fig. 7).

Actin is a major protein component of stereocilia (Müller and Barr-Gillespie, 2015) and stereocilia homeostasis is essential for HC function and viability (Kersigo and Fritzsch, 2015; Self et al., 1999; Ueyama et al., 2014). Dysregulation of actin bundling has been associated with HC dysfunction and loss (Mogensen et al., 2007; Perrin et al., 2010; Rzadzinska et al., 2009; Taylor et al., 2015). Various myosins, actins and a rich variety of actin-bundling proteins are regulated downstream of Atoh1 to transform microvilli into stereocilia during development (Kitajiri et al., 2010; Schwander et al., 2010). Loss of miRNAs is also associated with derailed stereocilia development (Weston et al., 2011). Neurog1 misexpression results in near normal miR-96 expression, but other miRNAs could be altered that also affect stereocilia formation. Once causality between Atoh1 and downstream signals associated with various bundle aberrations is clearer, our mouse model could help to eliminate spurious relationships as it has partially uncoupled stereocilia differentiation from HC topology.

The OC is highly stereotyped in organization, with two distinctly patterned compartments separated by a single row of adjacent IP cells, which express Atoh1 (Driver et al., 2013; Fritzsch et al., 2014b; Matei et al., 2005) without differentiating into HCs. Expression of multiple Hes/Hey factors (Benito-Gonzalez and Doetzlhofer, 2014; Doetzlhofer et al., 2009; Petrovic et al., 2015) may block these cells from HC development. Notch inhibition (and thus reduced expression of Hes/Hey) may lead to HC differentiation of IP cells (Mizutari et al., 2013). Neurog1 misexpression affects IP cells in multiple ways, including the formation of HC-like stereocilia bundles and replacement by the HC marker Myo7a in p75-positive IP cells. These instabilities in IP cells might relate to the alteration in Notch signaling, as observed (Figs 9 and 10).

In addition, the disruption of OC patterning is in part due to alterations in the Delta/Notch signaling pathway. For example, expression of Jag1 and of the downstream target gene Hes5 are altered. Previous work on Jag1 loss showed a reduction in the total number of HCs (Kiernan et al., 2006). Neurog1 misexpression results in discontinuity of Jag1 and delayed upregulation and differential downregulation of Hes5 in the cochlea. These expression changes vary radially and longitudinally, providing additional modulations of the variable signals of diffusible factors such as Fgf8, Fgf10 and Bmp4. Combined with altered SC response properties through changes in p75 and Prox1, these local changes might relate to the random OC patterning defects.

In summary, we demonstrate that misexpression of Neurog1 provides partial functional replacement of Atoh1 in the developing HC and improves HC maintenance. Neurog1 misexpression changes the signaling pattern of diffusible factors originating from HCs, as well as cell-cell interactions via Delta/Notch; both alterations contribute to the disorganization of the OC. Previous reports on the deletion of different combinations of Hes1, Hes5 or Hey2 show changes in sensory patterning in support of the lateral inhibition model, but mutants mostly maintain the HC and SC mosaic formation (Doetzlhofer et al., 2009; Zine et al., 2001). Changes of proneural bHLH genes in HCs alters the overall patterning of the OC, presumably by altering intercellular interactions via diffusible factors (Fgf8) and cell-cell interactions.

Understanding the causalities of these alterations and translating such understanding into regeneration (Zine et al., 2014) could help to restore hearing in deaf patients, with deafness now constituting the fastest growing ailment of the elderly worldwide (Kersigo and Fritzsch, 2015; Müller and Barr-Gillespie, 2015). Beyond the addition or deletion of entire rows of HCs obtained with previous mutations (Doetzlhofer et al., 2009; Zine et al., 2001), we provide here a new model of a dysfunctional OC that can help in OC restoration endeavors. Converting, through additional manipulation, the dysfunctional OC of our new mouse model into a functional OC that is able to restore hearing could provide proof of principle for fledgling attempts that aim to transform a partially defunct elderly OC into a fully functional OC.

The pattern of the organ of Corti was recently summarized by Jahan et al. (2015).

All animal procedures were carried out according to the recommendations and guidelines of the University of Iowa Institutional Animal Care and Use Committee (IACUC) under approved protocol ACURF #1309175.

Construction of the Neurog1 knock-in plasmid and generation of the Atoh1^kiNeurog1 mouse model were described previously (Jahan et al., 2012). To generate the Atoh1-Cre; Atoh1^f/kiNeurog1 line, we bred heterozygous Neurog1 knock-in mice (Atoh1^+/kiNeurog1) (Jahan et al., 2012) with mice carrying the Atoh1-Cre transgene and one Atoh1 floxed allele (Atoh1-Cre; Atoh1^f/+) as described previously (Pan et al., 2012a).

Mice were genotyped using tail DNA for standard PCR amplification as described previously (Jahan et al., 2012; Pan et al., 2012a). For further details, see the supplementary Materials and Methods.

ISH was performed as previously described (Jahan et al., 2012) using RNA probes labeled with digoxigenin. A detailed description is provided in the supplementary Materials and Methods.

For quantitative reverse transcription PCR (RT-qPCR), we used cochlea from P7 Atoh1-Cre; Atoh1^f/kiNeurog1 mice and their heterozygous Atoh1^+/kiNeurog1 littermates as control. After sedation with 2,2,2 tribromoethanol, the mice were hemisected and the cochleae were dissected out within 2-3 min in RNase-free conditions and stored in RNAlater (Ambion) at −80°C. Total RNA extraction was performed using the Direct-zol RNA Mini-Prep Kit (Zymo Research) and RNA concentration and 260/280 ratio were obtained with a Nanodrop spectrophotometer. On-column DNase I treatment was performed according to the Direct-zol Kit protocol. 1 µg total RNA was reverse transcribed and cDNA was synthesized with Anchored-oligo (dT)₁₈ primer using the Transcriptor First Strand cDNA Synthesis Kit (Roche). For qPCR, the primers and probes were designed using the Roche Universal ProbeLibrary Assay Design Center and primers were obtained from Integrated DNA Technologies. Primer sequences and priming conditions are listed in supplementary material Table S3. qPCR was performed in a 96-well plate using Roche LightCycler 480 Probes Master Mix and a Roche Light Cycler 480 real-time PCR machine. For all target genes, qPCR was performed for at least three to four biological replicates and three technical replicates including no-template controls for each sample following MIQE guidelines (Bustin et al., 2009). qPCR data were analyzed in Microsoft Excel and ΔΔC_T was calculated to determine the relative expression in Atoh1-Cre; Atoh1^f/kiNeurog1 cochlea compared with control normalized to Actb reference transcript. Statistical analysis was performed with Student's t-test in GraphPad Prism 6 software.

Immunohistochemistry was performed as described previously (Jahan et al., 2012). IHCs, OHCs and IP cells were counted in the P7 control, Atoh1-Cre; Atoh1^f/kiNeurog1 and Atoh1-Cre; Atoh1^f/f mice in 300 µm stretches in comparable regions in the apex, middle and base of the cochlea after performing immunohistochemistry for Myo7a and tubulin. Quantification was performed in six independent cochleae from each genotype. For further details, see the supplementary Materials and Methods. The data obtained from the quantification of IHCs, OHCs and IP cells were analyzed using Student's t-test. P<0.05 was considered statistically significant.

Following sedation, ABR recording was performed in 1-month-old control and Atoh1-Cre; Atoh1^f/kiNeurog1 littermate mice. A loudspeaker was placed 10 cm from the pinna of the test ear and computer-generated clicks were given in an open field environment in a soundproof chamber. Click responses were averaged and recorded signals were bandpass filtered (300 Hz-5 kHz) with a 60 Hz notch filter. The sound level was decreased in 10 dB steps from a 96 dB sound pressure level until there was no noticeable response. For further details, see the supplementary Materials and Methods.

SEM was performed as previously described (Jahan et al., 2012). A detailed description is provided in the supplementary Materials and Methods.

We thank Dr Fernando Giráldez for constructive and valuable comments on the manuscript; Qiufu Ma (Neurog1), Huda Y. Zoghbi (Atoh1), Mengqing Xiang (Barhl1 and Pou4f3), Jacqueline E. Lee (Neurod1), Ulla Pirvola (Fgf8), Thomas Gridley (Jag1), Brigid L. M. Hogan (Fgf10), Doris K. Wu (Bmp4), Andy Groves (Hes5) and Guillermo Oliver (Prox1) for providing plasmids for in situ hybridization; the Central Microscopy Research Facility for carrying out SEM; the Roy J. Carver Center for Imaging for use of the Leica TCS SP5 confocal microscope; the Roy J. Carver Center for Genomics for use of the Roche 480 LightCycler; and the Iowa Center for Molecular Auditory Neuroscience (P30; ICMAN) for the use of the ABR facility at the University of Iowa.