Neurog1, Neurod1, and Atoh1 are essential for spiral ganglia, cochlear nuclei, and cochlear hair cell development

We review the molecular basis of three related basic helix–loop–helix (bHLH) genes (Neurog1, Neurod1, and Atoh1) and upstream regulators Eya1/Six1, Sox2, Pax2, Gata3, Fgfr2b, Foxg1, and Lmx1a/b during the development of spiral ganglia, cochlear nuclei, and cochlear hair cells. Neuronal development requires early expression of Neurog1, followed by its downstream target Neurod1, which downregulates Atoh1 expression. In contrast, hair cells and cochlear nuclei critically depend on Atoh1 and require Neurod1 and Neurog1 expression for various aspects of development. Several experiments show a partial uncoupling of Atoh1/Neurod1 (spiral ganglia and cochlea) and Atoh1/Neurog1/Neurod1 (cochlear nuclei). In this review, we integrate the cellular and molecular mechanisms that regulate the development of auditory system and provide novel insights into the restoration of hearing loss, beyond the limited generation of lost sensory neurons and hair cells.


Introduction
Without a doubt, loss of hair cells, in combination with deprivation of sensory neurons and cochlear nuclei, results in severe aging-related hearing loss [1][2][3][4][5] . Various approaches to hearing restoration focus mostly on hair cell regeneration, often without a full appreciation of the apparent interaction of hair cells with sensory neurons and cochlear nuclei [6][7][8] . For instance, the loss of hair cells also reduces most, but not all, spiral ganglion neurons [9][10][11] . Furthermore, early loss of sensory neurons massively affects the cochlear nuclei 12 . Thus, the best way of approaching the development/regeneration of hair cells, sensory neurons, and cochlear nuclei neurons is to resolve their dependence on each other: how are the development of hair cells, sensory neurons, and cochlear nuclei related 13-18 ?
Three basic helix-loop-helix (bHLH) genes were shown to be crucial for hair cell, sensory neuron, and cochlear nucleus development: 1. Neurog1 plays a crucial role in sensory neuron development, affects hair cells 19,20 , and has a limited impact on cochlear nuclei 21 .
Sensory neurons exit the cell cycle from the base to the apex between embryonic day 10 (E10) and E12 in mice, followed by cochlear hair cells from the apex to base between E12 and E14 29 . In parallel, cochlear nuclei exit the cell cycle between E10 and E14 30 . Spiral ganglion neurons project to cochlear hair cells (from base to apex; E13-E16; Figure 1) and nearly simultaneously send central processes to cochlear nuclei (from base to apex; E12-E16) [31][32][33][34][35][36] . Neurons and hair cells have been suggested to have a clonal relationship because of similarities in bHLH gene expression. This relationship may play a role in neuronal pathfinding for at least the periphery 37 ; however, central targeting is less understood but may involve Neurod1 16 . Spiral ganglion neurons depend upon Neurog1 19 and Neurod1 22 . In contrast to Neurog1 null mice 19 , which showed a complete loss of neurons, Neurod1 null mice 23 showed residual spiral ganglion neurons extending centrally to smaller cochlear nuclei 16,22 . Unlike Neurog1, which is possibly transiently expressed in cochlear nuclei, Neurod1 was found massively expressed, overlapping with Atoh1 26 , Ptf1 38,39 , and Lmx1a/b 14,25 . Peripherally, it was established that cochlear hair cells critically depend on Atoh1 (Math1) 24 . Furthermore, the length of the cochlea depends on Neurog1 19 and Neurod1 22,23 . Neurog1 is upstream of Neurod1 20 , and both are upstream of Atoh1 28,40 . Neurog1 and Neurod1 truncate Atoh1 expression 19,27 . Similarly, in the cerebellum, Neurod1 negatively regulates Atoh1 41 , suggesting that these genes interact in many areas of neuronal development. Also, a loss or reduction of cochlear hair cells occurs following the absence of Gata3 42 , Pax2 43 , Eya1/Six1 44 , Foxg1 45,46 , and Lmx1a 47-49 , and many of these genes and others also affect the sensory neurons innervating them 31,42,43,[50][51][52][53] .
We will provide a comprehensive review of the interplay of the three bHLH genes (Neurog1, Neurod1, and Atoh1) in the context of spiral ganglia, cochlear nuclei, and cochlear hair cells development. In addition, we will examine the role of other transcription factors (Eya1/Six1, Sox2, Pax2, Gata3, Foxg1, and Lmx1a/b) known to be involved in their development.

Spiral Ganglion Neurons
Crosstalk of Neurog1, Neurod1, and Atoh1 determines inner ear sensory neuron fate Both Neurog1 and Neurod1 play important roles in sensory neuron development and differentiation. All inner ear sensory neurons were lost in Neurog1 null mice 19 . Similarly, many sensory neurons were lost in Neurod1 null mice; however, not all neurons were lost 54 . More recent work in Neurod1 null mice showed that of those neurons that survived, there was an intermingled vestibular and auditory sensory neuron projection to cochlear hair cells 16,27 and showed a reduced and aberrant central projection to cochlear nuclei 10,16 .
What is unknown is whether there is a direct role of Atoh1 in sensory neuron development or whether it is indirect. Hair cells depend on neuronal innervation for long-term maintenance [55][56][57] . Similarly, neurons depend on hair cells and supporting cells for their maintenance 12 . Logically, one would assume that the absence of hair cells will eventually cause degeneration of many neurons because of a lack of neurotrophic support. Atoh1 null mouse embryos, which lack hair cells, showed reduced Bdnf-lacZ staining and reduced hair cell innervation in the basal turn of the cochlea ( Figure 2). The apex, which retained Bdnf-lacZ staining in undifferentiated cells of these mice, showed a denser spacing of spiral ganglion neurons, suggesting that Bdnf expression may not depend on Atoh1 in the apex 58 . Conditional deletion of Atoh1 resulted in residual innervation correlated to residual hair cell formation 11,27 , demonstrating that near-normal residual cochlear hair cells receive innervation from a surprisingly large number of neurons 27 .
null mice, which develop only immature hair cells and have limited expression of neurotrophins 59 , show little effect on innervation patterns beyond the lack of innervation to outer hair cells (OHCs) birth. The absence of inner hair cells (IHCs), through the loss of Atoh1 or in Bronx-waltzer mutants, results in spiral ganglion projections to OHCs and disorganized central projections 10,60,61 ( Figure 2). Interestingly, replacing an allele of Atoh1 with Neurog1 in Atoh1 kiNeurog1 mice showed a different pattern of spiral ganglia projections to reach out the organ of

The inactivation of both bHLH transcription factors in double
Atoh1/Neurod1 null mutants uncouples fiber growth and expansion of remaining neurons 27 that could be useful for hair cell restoration 3,5,65,66 . More recent data using Rosa CreER ; Rainbow mice showed clones of spiral ganglion neurons and hair cells in the organ of Corti, suggesting that they arose from a typical progenitor cell 67 . Initially, the meaning of the transient expression of apparently cochlear-derived neurons was unclear.
In contrast to the loss of spiral ganglion neurons in mice lacking Neurog1 19,28 , overexpression of Neurog1 in immortalized multipotent otic progenitors (a cellular system for spiral ganglion neuron differentiation) drives proliferation via increased Cdk2. It promotes neuronal differentiation through the expression of Neurod1 68 . These findings suggest that Neurog1 can promote proliferation or neuronal differentiation and possibly impact hair cells without affecting cochlear nuclei 68,69 . It appears that a set of data support the transformation of astrocytes into neurons in Neurod1 70 and Neurog2 71 . The induction of neuronal proliferation and otic progenitor cell transplantation is a potential strategy to replace lost spiral ganglion neurons.
Recent work on the characterization of neuronal and hair cell progenitors revealed insights into early gene expression during neuronal development 7,72 . Markers for spiral ganglion neurons, Isl1 73,74 and Gata3 9,75,76 , were detected in developing neurons, although Neurod1 was seen in only the youngest neurons 7 .
In summary, the known deletion of spiral ganglion neurons in Neurod1 and Neurog1 null mice 27,28 suggests these as potential genes for the induction of new neurons with or without inducing hair cells 7,68 and is consistent with predictions of various cell types that require independent inducers 9,10 . Understanding how the expansion of neuronal projections in the absence of hair cells could be helpful to restore lost innervation 3,5,72,77,78 , in particular, understanding how to reinnervate the flat epithelia after long-term hearing loss, will be beneficial 79 .
Deletion of Sox2 and other genes affect spiral ganglion neuron development Initially, deletion of Sox2 was thought to eliminate all sensory neurons 80,81 ; however, a transient development of vestibular neurons was recently shown 31 . A delayed loss of Sox2 in Isl1-cre; Sox2 f/f mice showed a transient development of spiral ganglion neurons with abnormal innervation to disorganized hair cells in the base but no hair cells or sensory neurons in the apex 73 . That the later-forming neurons in the apex never developed suggests that Sox2 is essential for late neuronal development. Any similarities between different Sox2 deletions (Lcc, Ysb, Isl1-cre; Foxg1-cre) remain to be investigated. Eya1/ Six1 induces Sox2 expression to promote proneurosensorylineage specification. Ablation of the ATPase-subunit Brg1 or both Eya1/Six1 results in loss of Sox2 expression and lack of neurosensory identity, leading to abnormal apoptosis within the otic ectoderm. Brg1 binds to two of three distal 3′ Sox2 enhancers occupied by Six1, and Brg1 binding to these regions depends on Eya1/Six1 activity 82 . Recent work provides insight into SOX2 and NEUROD1 protein expression dynamics during neuronal differentiation. Quantification of the fluorescence intensity of nuclear proteins in immortalized multipotent otic progenitors showed expression dynamics of SOX2 and NEUROD1 from a progenitor into differentiated neurons. During neuronal differentiation, SOX2 levels decreased while NEUROD1 levels increased 69 . Evaluation of Neurog1 was excluded because of its dual roles in both proliferation and neuronal differentiation 68 . The increase of Neurod1 expression is in line with what is known for Neurod1 in collaboration with Sox2 10,31 . Understanding the expression dynamics of crucial transcription factors helps design replacement strategies for lost sensory neurons 69 .
The deletion of Pax2 resulted in a near absence of spiral ganglion neurons 43 , comparable to the significant loss of spiral ganglion neurons in Isl1-cre; Sox2 f/f mice 73 . Many additional genes derail the development of the inner ear and its innervation 9,83-86 . For example, disorganized projections to the cochlea are shown with Sox10 deletion in Schwann cells 87 . In addition, partial loss of hair cells reorganizes the remaining afferents and efferents 75,88,89 . These data provide a baseline of various deficits that require further examination, including the disorganized innervation in conditional deletions of Gata3 9,32,90 . Other genes, such as those involved in Wnt signaling, affect afferent innervation to OHCs 85 , but more work is needed to fine-tune the different effects. Finally, Lmx1a loss results in a delayed upregulation of Atoh1 combined with a transformation of basal turn hair cells into a mix of cochlear and vestibular hair cells 10,13 . In summary, Sox2 is essential for sensory neuron development 31 in combination with other downstream neuronal inducers (Neurog1 and Neurod1) known to interact with Atoh1 16,27 .

Cochlear Nuclei
Neurod1 and Atoh1 are expressed in the cochlear nuclei Beyond a transient and limited expression of Neurog1 expression in vestibular nuclei 21,91,92 , the other bHLH genes, Atoh1 and Neurod1, are expressed in cochlear nuclei 18,93,94 . Atoh1 is expressed in developing cochlear nuclei, and the dorsal cochlear nucleus specifically requires Neurod1 22,23 . Atoh1 is expressed dorsally in the central nervous system and its deletion disrupted spinal cord, brainstem, and cerebellum development 95,96 . Rhombomere-specific deletion of Atoh1 demonstrates that the cochlear nucleus forms from cells in rhombomeres 3-5 17,97 . Atoh1 expression is negatively regulated by Neurod1 in the cerebellum 41,98 , the cochlear hair cells and neurons 10 , and the intestine 99 but has not yet been shown for the cochlear nucleus. An additional bHLH gene, bHLHb5 97 , is also necessary to properly form the dorsal cochlear nucleus. Both bHLHb5 and another gene, Ptf1a, are strongly expressed in the dorsal cochlear nucleus 39,100 ; however, details on central projections for losing either of those two genes have not yet been provided 94,101 . Loss of Atoh1 or Ptf1a resulted in a loss of excitatory or inhibitory cochlear nuclei neurons, respectively, suggesting that both genes are important for regulating cell fate determination 38,39 . Recent molecular work on Atoh1 and Ptf1a lineage contributions to cochlear nuclei development show conserved and divergent origins across species 15,102 .
Neurod1 deletion is shown to affect the central targeting of inner ear neurons massively. Not only are auditory neuron projections aberrant, but there is also an overlap of cochlear and vestibular projections 16 . Furthermore, the central projections are disorganized to the inferior colliculi 16 , expanding previous work on defects generated with Hoxb2 mutants 103 . In contrast, Atoh1 null mutants, which lack cochlear nuclei, show nearnormal central projections 104 , suggesting that neither Atoh1 nor the cochlear nuclei themselves have a notable role in afferent pathfinding centrally. The conditional deletion of Atoh1 in the ear, but retaining Atoh1 expression in cochlear nuclei, shows near-normal segregation of central projections 27 , expanding the critical independence of Atoh1 in neuronal pathfinding. Not surprisingly, then, Atoh1/Neurod1 double null mice had little additional disorganized projection of cochlear afferents beyond that of Neurod1 alone 27 (Figure 3). Atoh1/Neurod1 forms a complex interaction in the cerebellum 41,98,105 , which is useful for Neurod1 to convert astrocytes and Schwann cells into neurons 70, 106,107 . Details are needed to determine whether deviations of central projections (Figure 1) would occur in older stages after cochlear nuclei are formed 30 and dependence of cochlear nuclei on neuronal input declines 12 . Recent data suggest plastic reinnervation of cochlear nuclei 108 , but it remains unclear whether this plasticity is permanent.
These data implicate several different bHLH genes (Atoh1, Neurod1, Ptf1a, and bHLHb5) in cochlear nuclei development. The interactions of these genes in cochlear nuclei development and innervation remain to be fully characterized.

Sox2 and Lmx1a/b are expressed in cochlear nuclei
Sox2 is essential for proneuronal regulation throughout the entire brain 109,110 and is broadly expressed in cochlear nuclei, but its role has not been detailed by selective Sox2 deletion in cochlear nuclei. Lmx1a/b double null mutants lack cochlear nuclei and choroid plexus and have a hindbrain reminiscent of a spinal cord 13 . In these mice, central projections of spiral ganglion neurons are lost, and vestibular fibers project bilaterally to the dorsal hindbrain and interdigitate with contralateral vestibular fibers 13 . The presence of these bilateral projections correlated with the expression of other genes, such as Wnt3a and In summary, the expression of Lmx1a/b for the proper formation of the hindbrain is essential and the deletion of Lmx1a/b causes aberrational projections. In contrast to the detailed description of Lmx1a/b loss, there is limited information on the role of Sox2 and other genes (Npr2, Prickle1, Fzd3, and Wnt3a) on central projections.

Cochlear Hair Cells
Neurog1, Neurod1, and Atoh1 interaction in developing hair cells Without a doubt, the development of all hair cells depends upon Atoh1 expression 24 . Atoh1 expression initiates in the cochlea at the upper-middle turn around E13.5 and progresses bilaterally toward the base and apex. Atoh1 expression shows a delayed upregulation in the apex compared with the base 24,58 , combined with very late apical hair cell differentiation at E18.5 112,113 . Interestingly, inner pillar cells were positive for Atoh1, suggesting that Atoh1 expression does not always result in a hair cell fate 28,114 . In contrast to differentiation of hair cells starting near the base and progressing toward the apex, hair cells exit the cell cycle first in the apex, at E12.5, and progress toward the base 28,29,115 . Furthermore, cell exit progresses radially from IHCs to OHCs 10,116,117 , as was shown initially using green fluorescent protein (GFP) labeling 118 . Loss of Neurog1 results in hair cells exiting the cell cycle two days earlier than controls 28 . Furthermore, there is a premature Atoh1 upregulation in an atypical apex-to-base progression in hair cells following Neurog1 loss 19,28 . Likewise, in Neurod1 null mice, early upregulation of Atoh1 from apex to base resulted in the formation of IHC-like cells in the region of OHCs, suggesting a transformation of OHCs into IHCs because of increased Atoh1 expression 16,23 . The cellular processes driving remodeling of the prosensory domain during cochlear development indicate that combinations of cellular growth contribute to base-to-apex cochlear extension, allowing different interpretations of OHC progression 10,88,116,117,119,120 . Despite its prominent role in hair cell differentiation, Atoh1 (Figure 4) does not seem to have a role in cochlear length determination 27 . In contrast, Neurog1 deletion resulted in a 50% reduction in cochlear length, a reduction in the size of vestibular epithelia 28 , and ectopic hair cells in the utricle 9,121 . Likewise, loss of Neurod1 (Figure 4) shortened the cochlea by about 50% 16,23 . Atoh1/Neurod1 double knockout added minimally to the cochlear length reduction in Neurod1 loss alone 27 . Although this suggests a possible interaction of bHLH genes, the reduction in length may be influenced simply by the loss of Shh normally generated by spiral ganglion neurons 122 , which would be absent or reduced in number in Neurog1 or Neurod1 null mice. The reduction of the organ of Corti is affected by several deletions of Shh 123 , Gata3 75 , Foxg1 45,124 , and Lmx1a 47,49 in addition to Neurog1 and Neurod1.
Conditional deletion of Atoh1 using Pax2-cre showed that most hair cells were lost during late embryonic development; however, some undifferentiated cells express Myo7a in postnatal stages and are targeted by neurons. A "self-terminating" system (Atoh1-cre; Atoh1 f/f ), in which a transient expression of Atoh1 results in some initial hair cell development, demonstrated progressive loss of IHCs and most OHCs shortly after birth 11 . However, some Myo7a-positive OHCs remained in adults in these mice. This suggests that most hair cells depend upon continued Atoh1 expression for at least some time. Various other conditional deletions of Atoh1 established that continued Atoh1 expression is essential for hair cell survival and maturation 100,125 . Interestingly, generating a transgenic mouse in which Neurog1 replaces Atoh1 (Atoh1 kiNeurog1/kiNeurog1 ) showed that, although Neurog1 cannot fully rescue the Atoh1 null hair cell loss phenotype, it does form additional patches of undifferentiated "hair cells" rather than a flat epithelium 63 . In addition, heterozygote mice expressing one copy of each gene  (Atoh1 kiNeurog1/+ ) showed some disorganization of hair cell distribution ( Figure 2 and Figure 4) not observed in Atoh1 heterozygotes, suggesting cross-interaction between Atoh1 and Neurog1. Using an ingenious system to overexpress Atoh1, in which the Atoh1 coding sequence is under the control of a tetracycline response element (TRE), generated viable ectopic "hair cells" in early postnatal mice 126 in line with an upper induction of proliferation 127 .
Loss of Neurod1 resulted in the formation of Atoh1-positive "hair cell"-like cells within intraganglionic vesicles (Figure 4) in the vestibular ganglion 54 , suggesting a potential conversion of vestibular sensory neurons into hair cells. The ectopic hair cells are forming in addition to the saccule and utricle and are positive for several genes-such as Atoh1, Fgf8, and Nhlh1that generally are expressed outside the hair cells (Figure 4). This finding indicates the normal suppression of Atoh1 by Neu-rod1 in these neurons and implies that Neurod1 might suppress hair cell fate in sensory neurons 16 . Similar Neurod1-Atoh1 interactions were reported in the cerebellum 41,98 and the intestine 99 and were used to transform astrocytes to neurons 106,107 . In the absence of both Atoh1 and Neurod1 in double null mutants, these "ectopic hair cells" are not formed 27 , suggesting that Neurod1 and Atoh1 interact upregulate neurons into ectopic hair cells after the loss of Neurod1.
In summary, using progenitor cells for spiral ganglia and hair cell replacement seems to be a possible way forward for hearing restoration 7,68 , in addition to various other approaches 6,8,77,128 . Unfortunately, generation of new hair cells in later stages beyond the earliest stages has not yet been achieved 127 . Understanding how to generate new hair cells at later stages is needed for older animals and humans with aging-related hearing loss 1,2 . Fully understanding the various mutations and putting them into the context of different cell fates require identifying certain steps necessary to initiate specific distributions of sensory hair cells 10,113,129,130 . What remains is understanding the various interactions of Neurog1, Neurod1, and Atoh1 for the complete formation of all hair cells.
Sox2 interacts with other genes for hair cell expression Sox2 is also essential for hair cell formation 52 , likely through the activation of Atoh1 expression 109,110,131 . Interestingly, two independent approaches using delayed deletion of Sox2 53,73,131 showed different results. In one, a delayed loss of Sox2 using Sox2-cre-ER demonstrated effects in the apex only 131 . In the other study, conditional deletion of Sox2 using Islet1-cre resulted in the loss of hair cells in the apex and a delayed loss in the base, showing unusual basal turn hair cells/supporting cells and inner pillar cells 73 , suggesting a role for the timing of Sox2 expression. As expected, the timing of Sox2 expression was later demonstrated to be essential for sensory development 81,132 . Furthermore, a complete deletion of Sox2 in the ear using Foxg1-cre showed the overall cochlear reduction and no hair cell development 31 . These combined studies provide an essential role of Sox2, although the interaction of Sox2 with Atoh1 is not fully understood 6,8,68,76,77,88,117 .
Other genes are also crucial for inner ear and hair cell development. For example, Eya1/Six1 is essential for early ear development and is needed to form the cochlea 44,50,53 and induces Sox2 expression, as described earlier 82 . Another gene, Pax2, is necessary for organ-of-Corti formation 43 and cooperates with Sox2 to activate Atoh1 expression 51 . Conditional deletion of Gata3 using Pax2-cre showed deletion of many hair cells and a complete loss of all hair cells with an earlier deletion of Gata3 using Foxg1-cre 42,75 . In these latter mice, levels of Atoh1 expression were significantly reduced, and genes downstream of Atoh1 were not detected following this early deletion of Gata3. Mice mutant for another gene, Lmx1a, showed a delayed expression of Atoh1 followed by transforming some organ-of-Corti hair cells into differentiated vestibular hair cells 2,13,47,133 . Foxg1 null mice show a reduced cochlear length and a disorganized apex of multiple rows of hair cells with disoriented polarities 45,46,124,134 . A somewhat similar phenotype is reported for n-Myc null mutants accompanied with apical cell fate changes 46,57,[135][136][137] .
The partial deletion of some, but not other, hair cells is an exciting perspective that needs to be explored. Inactivation of Fgfr1 in the inner ear by Foxg1-Cre-mediated deletion leads to an 85% reduction in the number of auditory hair cells 138 . Likewise, Sox2 omission shows a partial loss of hair cells in the Yellow submarine (Ysb) mutation 52 . Using Pax2-cre to conditionally delete Dicer 89 resulted in incomplete hair cell loss compared with the total hair cell loss with Foxg1-cre conditional deletion, comparable to the equivalent conditional deletions of Gata3 75,139 . Finally, Bronx-waltzer mice, which are mutant for the gene Srrm4 (Figure 4), lose IHCs and vestibular hair cells but retain OHCs 60,61 . OHCs, meanwhile, express Srrm3 independent of the Srrm4 gene downstream of REST 61 .
These data show that cochlear hair cells are affected by single gene deletions and complex interactions of several genes, including compound analysis of partial deletions 10 , primarily unexplored in detail 7,72 . While Atoh1 alone is the dominant gene 24 , interactions with other genes need to be worked out 44,77,78 .

Summary and conclusion
Inner ear sensory neurons, cochlear nuclei, and cochlear hair cells all require bHLH genes for their proper development. Atoh1 is essential for cochlear hair cell and cochlear nuclei development. Neurog1 and Neurod1 are vital for sensory neuron development and differentiation. All three genes play crucial roles in a feedback network to regulate specific cell fate appropriately and in coordination with other genes. Some of these additional genes interact with the bHLH genes in these contexts, such as Lmx1a/b, requiring more detailed investigation.