Inner ear development: Building a spiral ganglion and an organ of Corti out of unspecified ectoderm

Abstract

The mammalian inner ear develops from a placodal thickening into a complex labyrinth of ducts with five sensory organs specialized to detect position and movement in space. In addition, the mammalian ear develops a spiraled cochlear duct containing the auditory organ, the organ of Corti (OC), specialized to translate sound into hearing. Developing the OC out of a uniform sheet of ectoderm requires an unparalleled precision in topological developmental engineering of four different general cell types, sensory neurons, hair cells, supporting cells, and general otic epithelium, into a mosaic of ten distinctly recognizable cell types in and around the OC, each with a unique distribution. In addition, the OC receives a unique innervation by ear-derived spiral ganglion afferents and brainstem-derived motor neurons as efferents, and requires neural crest-derived Schwann cells to form myelin and neural crest-derived cells to induce the stria vascularis. To achieve this transformation of a sheet of cells into a complicated interdigitating set of cells necessitates the orchestrated expression of multiple transcription factors that enable the cellular transformation from ectoderm into neurosensory cells forming the spiral ganglion neurons (SGN) while simultaneously transforming the flat epithelium into a tube, the cochlear duct housing the OC. In addition to the cellular and conformational changes to make the cochlear duct with the OC, additional changes in the surrounding periotic mesenchyme form passageways for sound to stimulate the OC. This article reviews molecular developmental data generated predominantly in mice. The available data are ordered into a plausible scenario that integrates the well described expression changes of transcription factors and their actions revealed in mouse mutants for formation of SGNs and OC in the right position and orientation with the right kind of innervation. Understanding the molecular basis of these developmental changes leading to the formation of the mammalian OC and highlighting the gaps in our knowledge may guide in vivo attempts to regenerate this most complicated cellular mosaic of the mammalian body to reconstitute hearing in a rapidly growing population of aging people suffering from hearing loss.

1. Introduction

Like all neurons of the brain and receptor cells of the eye, olfactory system and taste buds, the ear is a molecularly transformed embryonic ectoderm. This transformation in the ear is mediated by the expression of transcription factors that expand proliferation of the precursor population and transforms the cells into neurons or sensory cells (Fritzsch et al., 2006a, Gokoffski et al., 2011). In addition, complex changes of the single layer of cells into a set of tubes and recesses is needed to generate the inner ear labyrinth (Kopecky et al., 2012). The inner ear labyrinth of adult mammals consists of three semicircular canals, the utricle connecting all canals, and the saccular recess (Fig. 1). The base of each canal, the utricle, and the saccule house the five vestibular sensory organs that allow perception of position and movement in space (Lewis et al., 1985). In addition, only the mammalian ear has the coiled cochlear duct connected via the ductus reuniens with the saccule. The cochlear duct contains the organ of Corti (OC) that translates sound into hearing (Hudspeth, 2014) and is connected via the spiral ganglion neurons (SGNs) to the cochlear nuclei of the hindbrain (Nayagam et al., 2011).

Fig. 1.

Thin-sheet laser imaging microscopy (TSLIM) and 3D-reconstructions show a developing left ear, viewed from lateral (top row, anterior is to the left) and ventral (bottom row, anterior is to the left). Note the dramatic growth of the cochlea to become the largest duct in the mouse ear exceeding the semicircular canals in length and width. AC, anterior canal crista; C, cochlear duct; HC, horizontal (lateral) canal crista; PC, posterior canal crista; S, saccule; U, utricle. Bar indicates 100 μm. Compiled after (Kopecky et al., 2012).

To achieve a coordinated transformation of a flat embryonic epithelium into the cellular mosaic of the adult structure requires an orchestrated regulation of a multitude of transcription factors leading to the differential expression of about 17,000 genes that mark the adult inner and outer hair cells (Liu et al., 2014). We only now begin to understand the interactions of these factors mostly through analysis of expression patterns and phenotypes in mutant mice. The complexity of changes related to the transformation of a sheath of cells into the vestibular organs and cochlear duct as part of a 3-dimensional labyrinth of ducts and recesses (Chang et al., 2004, Fritzsch et al., 2013). The OC is arguably the most sophisticated cellular mosaic of the mammalian body with ten distinct types of cells (Slepecky, 1996, Zetes et al., 2012) uniquely distributed in and around the OC to ensure transformation of sound into electric signals (Hudspeth, 2014). In addition, the OC is innervated by a highly patterned set of afferents (Nayagam et al., 2011) and receives in addition a complicated innervation by efferents (Simmons, 2002) to modify apparently the sensitivity of the organ to loudness (Sienknecht et al., 2014).

The OC and the unique organization of its innervating sensory neurons into a distinct spiral ganglion (Mao et al., 2014, Nayagam et al., 2011, Sandell et al., 2014) as well as the organization of efferents into the olivo-cochlear system (Simmons et al., 2011) are all features found only in mammals. Interesting transitions of the basilar papilla of tetrapods into an OC exist in egg-laying mammals (Fritzsch et al., 2013). These changes may indicate that expansion of the cochlear duct subsequent to the loss or transformation of the lagena (Luo et al., 2011) was a crucial step to evolve the simpler checkerboard cellular mosaic of basilar papillae of other tetrapods into the complicated organization of the mammalian OC (Fritzsch et al., 2013). Interestingly, the OC of extant Monotremes (Platypus, Spiny Anteater) contains all the cell types of the Eutherian OC, but in a different arrangement suggesting that the transformation of the tetrapod basilar papilla into a coiled OC happened already in large part in mammalian ancestors. However, these cell types of the OC of Monotremes are within a short lagena duct that contains at its apex, like the lagena in reptiles and amphibians. Instead of one row of inner hair cells, two rows of pillar cells and three rows of Deiter's cells and outer hair cells throughout the OC, all these elements of the mammalian OC are mostly multiplied in Monotremes, except for the basal part that resembles closely to the Eutherian OC (Ladhams and Pickles, 1996). Beyond resolving the evolutionary origin of the basilar papilla as a separate sensory epithelium of the lagena recess of tetrapods (Fritzsch et al., 2013), understanding the molecular basis of this reorganization of afferents, hair cells and supporting cells to generate the uniquely mammalian OC out of the tetrapod basilar papilla with its innervating SGNs is the most important next step in understanding the evolution and development of the inner ear part of the mammalian auditory system. As all central auditory system development (Rubel and Fritzsch, 2002) presented in this volume hinges directly or indirectly on this evolutionary issue (Fritzsch et al., 2006c, Grothe et al., 2004), this chapter will provide the molecular development of OC, cochlear duct and SGN to set the stage for an understanding of the topics explored in the other chapters of this special issue.

2. From placode to SGNs: molecular transformation of ectoderm to neurons that provide the innervation of the OC

The development of vertebrate neuroectoderm requires BMP4 downregulation combined with Fgf expression (Delaune et al., 2005, Fritzsch et al., 2006a, Streit et al., 2000). Consolidating the transformation of ectoderm into pro-neurosensory precursor producing neuroectoderm is overall driven by similar genes regulation, with Oct6/Pou3f1 being among the earliest transcription factors expressed, followed by Sox2 and other Sox genes to maintain proliferation in a committed population of neuronal precursors (Fritzsch et al., 2006a, Reiprich and Wegner, 2014). This continued proliferation of committed pro-neurosensory precursors is achieved in part by preventing premature signals from proneural basic helix-loop-helix (bHLH) transcription factors through the expression of Id's and Myc's (Zhu et al., 2014), reinforced by many signals that define the pro-neurosensory region of the otic placode/otocyst, such as Lmx1a (Nichols et al., 2008), Tbx1 (Raft et al., 2004) and various diffusible factors (Chen and Streit, 2013, Raft and Groves, 2014).

Unique to vertebrates is that all major sensory organs - eyes, ears, taste buds and olfactory system - develop from, or have contributions from placodes (Northcutt and Gans, 1983, O'Neill et al., 2012, Patthey et al., 2014). Placodes may have evolved through embryonic aggregation of ancestral single cell neurosensory precursors (Fritzsch and Straka, 2014), comparable to the apparent aggregation of all neurogenic potential of the ectoderm into a single continuous sheet of cells that form the central nervous system (Fritzsch and Glover, 2007, Pani et al., 2012). In mammals, the otic placode is initially part of a pan-placodal region that expresses transcription factors necessary to initiate the neurosensory transformation of embryonic ectoderm (Chen and Streit, 2013, Schlosser et al., 2014, Steventon et al., 2014). The size of the otic placode is determined mostly through Wnt signaling (Ohyama et al., 2007), and several other factors cooperate to invaginate the rapidly proliferating otic placode to form the otic cup and later the otocyst that is completely segregated from the ectoderm (Fritzsch et al., 1998, Romand and Varela-Nieto, 2014).

The future cochlear duct in mice begins its growth at the ventral tip of the elongated otocyst (Fig. 1) around embryonic day (E) 11.5 and finishes its elongation around birth (Kopecky et al., 2012). Similar growth of cochlear duct exists in humans, albeit at a different time table (Kopecky and Fritzsch, 2013). As the cochlear duct elongates, SGNs delaminate apparently from three principal areas of the cochlear duct (Fig. 2): the ductus reuniens that later separates the cochlear duct from the saccular recess, the middle turn and the apex (Yang et al., 2011). These delaminating neurons are characterized by the expression of neurotrophins (Durruthy-Durruthy et al., 2014, Farinas et al., 2001). The neurons continue to proliferate and eventually shut down neurotrophin expression after they become post-mitotic, and express neurotrophin receptors instead (Fritzsch et al., 2002). Neurotrophins released from the sensory epithelia and neurotrophin receptors expressed in developing SGNs are essential for neuronal survival (Fritzsch et al., 2004). This expression of neurotrophins in the prosensory region and in early delaminating neuroblasts suggests a possible lineage relationships of hair cells and sensory neurons (Farinas et al., 2001), now expanded to neuronal and hair cell differentiation transcription factors and their interactions (Fritzsch et al., 2010b, Ma et al., 2000, Pan et al., 2012b). The migrating SGNs have to digest their way through the basal lamina surrounding the developing cochlear duct possibly with enzymes similar to neural crest (Mao et al., 2014) and respond to stop signals provided by Schwann cells to reside in the Rosenthal's canal inside the developing ear (Yang et al., 2011). In the absence of Schwann cells, SGNs migrate in part outside the ear, like vestibular ganglion neurons (Mao et al., 2014). ‘Birthdating’ with 3H-Thymidine (Ruben, 1967) and BrdU (Matei et al., 2005) has established that SGN's become post-mitotic in a base to apex progression between E10.5 to E12.5. Several recent reviews have expertly summarized much of what is known about gene regulation of neuronal development of the ear (Appler and Goodrich, 2011, Coate and Kelley, 2013, Raft and Groves, 2014, Yang et al., 2011) and the reader is referred to these more detailed accounts. Here we concentrate only on the molecular regulation of SGN formation and development as is relevant for the viability and patterned projection of these neurons to connect precisely the OC with cochlear nuclei, the molecularly least understood aspect of SGN development.

Fig. 2.

The developing ear has neurosensory precursors that share expression of Bdnf (a-c), whether they develop into hair cells or neurons. In control animals, all sensory epithelia are positive for Bdnf (a), but only the canal cristae and the apex of the cochlea retain Bdnf expression in Atoh1 null mice (b), indicating that the fate of sensory cells is different in different epithelia. In contrast to Atoh1 null, Neurog1 null mice never develop ganglion neurons; show an enlargement of the utricle but a severe reduction of the saccule (c). While only canal cristae and apical precursors remain Bdnf positive in Atoh1 null mice (b), mice homozygotic for Atoh1^LacZ show profound expression in sensory precursors nearly equivalent to control mice at this early stage (d). Delaminating neurons from the ear are only positive for Bdnf (asterisks in a and b), but some neurons are also positive for eGFP expression driven by an Atoh1 enhancer element (Vgl in e) that also labels all vestibular hair cells (e). At later stages, this Atoh1-eGFP expression expands not only to all hair cells of OC but also in almost all inner pillar cells (IPC, f). Bar indicates 100 μm (a-e) and 25 μm (f). Compiled after (Fritzsch et al., 2005b, Matei et al., 2005).

Among genes that have been shown to be essential for the formation of SGNs and their patterned projection are mostly transcription factors but also several diffusible factors such as Shh (Bok et al., 2007), Wnts (Ohyama et al., 2007) and Fgfs (Wright and Mansour, 2003). Among the transcription factors are several more global, as well as locally expressed transcription factors that specify first the pre-placodal region and subsequently the otic placode and the prosensory region of the otocyst. One of the most important preplacodal transcription factors is Eya1/Six1, which is now known to regulate the expression of downstream pro-neurogenic factors such as Neurog1 (Ahmed et al., 2012). Several factors are selectively expressed in the otic placode such as Foxi factors that are essential for Fgf signaling (Edlund et al., 2014). Fgf's and their recetors are, in turn, essential for placode differentiation (Pirvola et al., 2000). Analogous to the role of Pax6 for eye development (Gehring, 2011, Lamb, 2013) are Pax2/8 genes for ear development (Bouchard et al., 2010). Mutational analysis has shown that loss of Pax2 results in loss of the cochlea (Burton et al., 2004) or its transformation into a featureless sack without SGNs (Bouchard et al., 2010). In contrast, the earlier expressed Pax8 has no direct effect on ear development, apparently due to compensation by the later expressed Pax2. Eliminating both Pax2 and Pax8 results in severe truncation of ear development at the otocyst stage without any formation of neurosensory cells, indicating a molecularly unclear connection of Pax2/8 genes to the formation of ear ganglion cells, possibly via interaction with Pou domain factors as in general neurogenesis (Zhu et al., 2014). Incidentally, Pax2 plays a similar role in retinal ganglion cell development (Cross et al., 2011, Torres et al., 1996) which require the bHLH transcription factor Atoh7, an ortholog of the hair cell differentiation transcription factor Atoh1, for normal differentiation (Mao et al., 2013, Prasov and Glaser, 2012). These similarities may relate to conservation of these molecular cascades in the sense of deep homology (Shubin et al., 2009) for eye and ear development and evolution (Fritzsch and Piatigorsky, 2005).

Among other genes essential for SGN development are Gata3 (Karis et al., 2001), Neurog1 (Ma et al., 2000), Neurod1 (Kim et al., 2001), Tbx1 (Raft et al., 2004) and Pou4f1 (Huang et al., 2001). Gata3 is uniquely expressed in the ear and mutations truncate the cochlear duct and result, if eliminated early enough, in complete absence of all neurosensory development (Duncan and Fritzsch, 2013). In contrast, later conditional deletion of Gata3 in the differentiated SGNs causes severe disruption of their projection (Appler et al., 2013, Duncan and Fritzsch, 2013). Another gene uniquely expressed in the ear is Foxg1 and its loss truncates severely the formation of both the sensory epithelia and the sensory neurons, consistent with its alleged function in enhancing proliferation of precursors (Pauley et al., 2006). Other early expressed genes of the ganglion neurons are known, but their function has not been analyzed in detail and they may mostly function to modify the proneural signal such as Delta/Notch and the downstream bHLH transcription factors (Raft and Groves, 2014, Yang et al., 2011).

While it is clear that Neurog1 is an essential gene to induce neuronal development in the ear, alternative interpretations for the origin of neurons from the ear and neural crest have been proposed recently (Freyer et al., 2011). To put this recent claim about dual origin of the inner ear neurons into perspective, one need to first realize that neurons have been traced to delaminate from the ear using neurotrophins (Fig. 2). Moreover, inner ear neurons require factors for their development or express genes that are only present in the developing ear, not in the adjacent neural crest [Pax2, Pax8, Gata3, Foxg1, Bdnf (Bouchard et al., 2010, Pauley et al., 2006, Rubel and Fritzsch, 2002)]. However, some genes, such as Neurog1, are common to neural crest derived and ear derived neurons (Ma et al., 1998). Consistent with the ubiquitous expression of some markers of ear-derived neurons, enhancers isolated from transcription factors expressed in neural crest have been claimed to show that a variable number of neurons and even hair cells derive from neural crest (Freyer et al., 2011). Unfortunately, alternative interpretations for these results, such as altered expression of the artificial construct, have not been ruled out. More recent analysis of neural crest contributions to the neurosensory cells of the developing ear using more specific techniques could not verify such contributions (Sandell et al., 2014). Most recently, genetic ablation of all neural crest failed to show noticeable deficits in neurosensory development of the ear (Mao et al., 2014) as predicted based on the claims of substantial contributions of neural crest to neurosensory ear development (Freyer et al., 2011). At the moment it appears that the numerous studies that directly show the delamination of neurons from the ear using ear specific markers (Bdnf, Ntf3; Fig. 2) or show loss of neurons using ear specific transcription factors (Gata3) in combination with ear specific Cre lines such as Foxg1 (Duncan and Fritzsch, 2013), neither of which is expressed in the neural crest (Karis et al., 2001, Pauley et al., 2006) are not compatible with a neural crest derived contribution to the neurosensory cells of the ear. In addition, the neural crest-derived hypothesis cannot explain why nearly all neurons of the vestibular system depend on TrkB (Fritzsch et al., 1995), much like other placode derived neurons (Fritzsch et al., 1997b), but unlike neural crest derived neurons (von Bartheld and Fritzsch, 2006). In fact, all inner ear sensory neurons are uniquely dependent on two neurotrophin receptors (TrkB and TrkC) and their two ligands (Bdnf and Ntf3) (Fritzsch et al., 2006b, Rubel and Fritzsch, 2002) and this unique dependency is not shared with any neural crest-derived sensory neurons (von Bartheld and Fritzsch, 2006).

Transformation of ectoderm into sensory neurons depends on the expression of a number of genes, but some fotm an interesting cascade of gene expression. Sox2 is needed for Neurog1 expression (Puligilla et al., 2010) that, in turn, regulates the expression of Neurod1, a gene necessary for the differentiation of most, but not all sensory neurons (Kim et al., 2001, Krüger et al., 2006) as initially claimed (Liu et al., 2000). Among the genes regulated by Neurod1 to ensure proper differentiation are Pou4f1 (Brn3a), a gene needed for proper fiber growth (Huang et al., 2001), and the neurotrophin receptors TrkB/TrkC, which are needed for survival of sensory neurons (Farinas et al., 2001). Many other genes expressed in differentiating neurons have been identified but their role in the differentiation and development of targeted projections of SGNs has not yet been fully clarified through genetic analysis in mutant mice (Appler and Goodrich, 2011, Yang et al., 2011).

One feature unique to SGNs is their specific topographically precise order of projection to the cochlear nuclei and the orderly projection to the two types of hair cells of the OC to form the neuroanatomical basis of the tonotopic map of the auditory system (Nayagam et al., 2011). The vast majority of SGNs differentiates as Type I SGNs that terminates exclusively at the single row of inner hair cells in an overlapping fashion of 10-30 fibers converging on a single hair cell. A minority of 5-8% of SGNs develop into Type II SGNs that project seemingly exclusively to the three rows of outer hair cells, each fiber innervating many outer hair cells over a distance of several hundred micrometers. Despite this diffuse peripheral projection (Fritzsch, 2003), the central projection in the cochlear nuclei appears to be in parallel and as well organized for Type II as that for Type I SGNs (Nayagam et al., 2011). Numerous ideas of how this peripheral segregation of Type I fibers to inner and Type II fibers to outer hair cells comes about have been proposed (Bulankina and Moser, 2012, Echteler et al., 2005). However, detailed analysis of this process revealed that the segregation of Type I and Type II neurons may begin much earlier (Bruce et al., 1997, Koundakjian et al., 2007) than previously suggested. Indeed, certain mutations indicate deviation from normal development of Type II afferents prior to them reaching the outer hair cells (Fritzsch et al., 2010a), indicating that Type II SGNs may be genetically determined to project differently rather than being randomly sorted upon reaching the OC with their processes as some previous ideas implied.

An interesting set of data seemed to provide a solution for this problem by indicating that neurotrophins have differential effects on Type II versus Type I fibers (Ernfors et al., 1995). However, more detailed investigations concluded that this differential effect is only a quantitative correlation and does not fit to the distribution pattern of either neurotrophins or their receptors (Farinas et al., 2001, Yang et al., 2011). Moreover, genetic manipulations that lead to the almost complete ablation of inner hair cells indicates that the majority of spiral ganglion afferents have the capacity to grow to outer hair cells, if forced to (Pan et al., 2012a), but receive normally some stop signal around inner hair cells that is possibly related to the unique co-expression of Ntf3 and Bdnf in inner hair cells that develops shortly before birth (Farinas et al., 2001) or to expression of EphA4 (Defourny et al., 2013). Markers for Type II fibers such as Peripherin are able to distinguish Type II fibers only after they have projected differentially to outer hair cells and thus cannot be causally linked to the segregation process (Nayagam et al., 2011). Despite plausible ideas on this subject, the molecular basis of the segregation of these types of afferents remains still unclear. Indeed, more recent data point out that a combination of early neuronally-expressed genes such as Prox1 (Fritzsch et al., 2010a), as well as interactions of growing afferents with neural crest derived Schwann cells (Mao et al., 2014), ultimately regulate the growth of the two types of spiral ganglion fibers to the OC and within the OC (Fig. 3). Given the role of Schwann cells in guiding afferents to the OC, it seems reasonable to assume that afferent interactions with the Schwann cell-like supporting cells will play a major role in the afferent fiber segregation as is so obvious in mouse mutants with supporting cell development defects (Puligilla et al., 2007). Until those molecules have been identified beyond the single candidate gene (Defourny et al., 2013), more data on effects of upstream regulators of such genes will be found to affect proper afferent growth such as Gata3, Pax2, Foxg1 and others.

Fig. 3.

The normal pattern of OC innervation around birth is shown (a, d) as well as aberrations induced with three mutations (b,c,e). The normal pattern is a very regular set of tunnel crossing fibers (TC) that project to three bundles forming between adjacent Deiter's cells (a,d). Note that fibers as they grow to outer hair cells always turn toward the base (d) and that the majority of radial fibers (RF) end at inner hair cells (IHC). Targeted deletion of Prox1 in SGNs using Nestin-Cre results in aggregation of all tunnel crossing fibers (TC) between the outer pillar cells (OP) and the third row of Deiter's cells (D3). Note that most inner pillar cells (IPC), outer pillar cells (OP) and Deiter's cells remain positive for Prox1, indicating a selective effect of Prox1 expression in SGNs to drive the fiber sorting. Altering the properties of supporting cells in Fgfr3 null mice results in disorganization of afferents to hair cells (c). Most disruptive for the pattern of innervation is the lack of Schwann cells (e). Radial fibers extend from the more centrally migrated SGNs to either bypass entirely the hair cells of the OC to end up at the lateral wall (LW) or to form a disorganized innervation of the OC. Bar indicates 20 μm (a-c) and 100 μm (d,e). Data are from (Mao et al., 2014, Yang et al., 2011).

SGNs carry restricted local hair cell activity to the cochlear nuclei to translate the frequency of sound dispersed along the OC into a tonotopic map used by the central auditory system to derive cues relevant for directional hearing and other aspects of sound processing. Numerous ideas about how such a map can develop based on local electric signals have been proposed (Bulankina and Moser, 2012, Tritsch et al., 2007). However, the central projections of SGNs to the cochlear nuclei develop as early as E12.5 in mice [Fig. 4; (Fritzsch, 2003, Karis et al., 2001)] and have already basal and apical segregation at embryonic day 14.5 (Jahan et al., 2010a). Based on these data (Fig. 4), a topological representation of the OC onto the cochlear nuclei develops in early embryos many days prior to the onset of such electrical phenomena (Fritzsch, 2003, Tritsch et al., 2007). Even disruption of hair cell differentiation in certain mutants has little effect on the overall initial topology of the peripheral target onto the cochlear nuclei (Xiang et al., 2003), suggesting that developmental mechanisms that do not require electrical activity might generate at least a coarse tonotopic map that may be refined by activity later in life (Leake et al., 2002). Mutations in transcription factors such as Neurod1 (Jahan et al., 2010a) or Gata3 (Appler et al., 2013, Duncan and Fritzsch, 2013) cause disruption of central and peripheral projections prior to the onset of electric activity and with or without defecting the OC. It should be noted that Type II afferents have a very different peripheral distribution extending to many outer hair cells instead of a single inner hair cell (Fritzsch, 2003). Despite this different peripheral organization, the central projection seems to reflect a similar topological restriction that is parallel to, and nearly identical to, the adjacent Type I afferent fibers (Nayagam et al., 2011), indicating that the central projections are sorted independently from peripheral fiber distribution. Such topological segregation of afferents might reflect the birth dates of SGNs, as has been suggested for afferent segregations in other developing systems (Fritzsch et al., 2005a). Another factor seems to be the competition for connections that can be relaxed in neurotrophin mutants and result in an expansion of central projections in areas devoid of afferents from other areas of the OC (Fritzsch et al., 1997a). Coordinated analysis of afferent birth dates with development of central and peripheral connections are needed to substantiate these suggestions.

Fig. 4.

Spiral ganglion axons reach the cochlear nuclei around E12.5 (a) and are from the initial projection clearly separated from vestibular axons of the posterior vertical canal crista (PC). Vestibular afferents extend early past the boundaries of the cochlear nuclei into the brainstem and the cerebellum whereas cochlear afferents in the anteroventral (AVCN) and dorsal cochlear nuclei (DCN). A segregated projection of base and apex is already established in E14.5 embryos (b) but appears to become refined in early neonates (c). Abbreviations: AVCN, anteroventral cochlear nucleus; Cne, cochlear nerve; Co, cochlear application; DCN, dorsal cochlear nucleus; eff, olivocochlear efferent bundle; PC, posterior canal crista application. Bar indicates 100 μm. Modified after (Fritzsch et al., 2005a, Jahan et al., 2010a, Maklad and Fritzsch, 2003).

In summary, SGNs are specified in early ear development as a subset of cells of the antero-ventral neurosensory part of the developing ear. Proliferating precursors delaminate mostly from three areas of the growing cochlear duct to assemble in the spiral ganglion that parallels and partially drives the spiraling of the cochlear duct. SGNs develop in embryos a topological projection to the cochlear nuclei (Fig. 4) that predates formation of, and is independent of, hair cell differentiation using unclear molecular cues. SGNs interact with Schwann cells to stop their migration and develop ordered projections to the OC. Within the OC, afferents apparently interact with supporting cells to sort out the Type II projection to outer hair cells (Fig. 3). The detailed projections both to the cochlear nuclei (Fig. 4) and to the OC (Fig. 3) require a partially understood set of factors for normal development but neither the full complement of factors needed, nor the detailed interactions to generate the tonotopic projection to the cochlear nucleus and the sorting of Type I versus Type II afferents are clear at the moment.

3. From placode to OC: placing and differentiating the OC in the right position in the cochlear duct to translate sound into hearing

Like the SGNs, the OC appears to derive entirely from the cochlear duct and thus exemplifies an in situ transformation of pro-neurosensory epithelial cells into the most complex of all cellular mosaics in the mammalian body. To understand the necessary steps needed to achieve this transformation and to appreciate the outcome of genetic manipulations in various mouse mutants, it is paramount to fully understand the cellular organization of the OC (Slepecky, 1996). In contrast to some theoretical attempts to explain the cellular mosaic of the OC as a simple checkerboard mosaic of alternating hair cells and supporting cells (Sprinzak et al., 2011), a pattern that applies to vestibular organs or the basilar papilla of vertebrates, the OC is only partially organized like a checkerboard (Fig. 5). More to the point, the OC is only in the outer compartment organized in a nearly checkerboard fashion with outer pillar cells, Deiter's cells of row 1 and 2 and outer hair cells forming a checkerboard except for the most extreme aspects of the base and the apex of the OC (Fig. 5). In contrast, the inner compartment -consisting of inner pillar cells, inner hair cells, inner phalangeal cells and border cells - shows a very different cellular distribution pattern. In the human cochlea, a continuous row of ~6000 inner pillar cells abuts an equally continuous row of ~4000 outer pillar cells (Fig. 5). Obviously, simple lateral inhibition via the Delta/Notch system can drive the checkerboard mosaic of the outer compartment (Sprinzak et al., 2011). However, the continuity of as a single row of adjacent, broadly contacting inner pillar cells cells (Zetes et al., 2012) that has no numerical match to any other cell types of the OC is not easily explained by simple lateral inhibition models. Likewise, adult inner hair cells form a single row of broadly contacting cells (Slepecky, 1996), again difficult to reconcile with the lateral inhibition model. Obviously, these broad cellular contacts of inner hair cells and inner pillar cells must develop despite the Delta/Notch activity that should counteract formation of adjacent and broadly contacting. Clearly, Delta/Notch based models requires significant modifications (Moody, 2007) to serve as an explanation for the boundary between the inner and outer compartments and continuous approximation of inner hair cells and inner pillar cells.

Fig. 5.

The organ of Corti (OC) is often depicted by most theoretical papers as a simple checkerboard of alternating hair cells and supporting cells (a), in which each hair cell is surrounded by four supporting cells and each supporting cell is surrounded by four hair cells. This checkerboard pattern exists, however, only in the ‘outer compartment’ of the OC with alternating rows of outer hair cells (OHC) and Deiter's cells/outer pillar cells (D1-3; OPC). This outer compartment is separated by a single row of adjacent inner pillar cells (IPC) from the inner compartment, consisting of two alternating cell types, the inner hair cell (IHC) and the inner phalangeal cells (IPhC). Note that inner phalangeal cells seem to be organized in pairs flanking either medially or laterally the IHCs according to more recent TEM data. Note also the numerical match of outer compartment elements in humans whereas neither IHC nor IPC fit to the numbers of the outer compartments (b). In particular the continuity of IPCs in a single row and the fact that two supporting cells (IPC, OPC) are adjacent to and touching each other is difficult to reconcile with simple interpretations of the Delta/Notch interaction of lateral inhibition. Modified after (Slepecky, 1996, Spoendlin and Schrott, 1988, Zetes et al., 2012).

Comparative data show that the basilar papilla of tetrapods, which is the precursor of the mammalian OC, has the checkerboard organization of vertebrate vestibular organs and both can regenerate hair cells upon loss (Golub et al., 2012). How the exclusive organization of the single row of broadly contacting inner hair cells and equally broadly contacting inner pillar cells, uniquely characterized by p75 (Ngfr) expression (von Bartheld et al., 1991), evolved and how this altered organization of the OC might relate to its inability to regenerate is unclear. While acknowledging that this major aspect of OC cellular development and evolution remains unresolved at the moment, we will summarize here our molecular understanding beyond recent reviews summarizing certain aspects covered here only in part (Fritzsch et al., 2011, Groves et al., 2013, Raft and Groves, 2014). Delta/Notch signaling (Imayoshi and Kageyama, 2014, Zine et al., 2014) is undoubtedly an important player for certain aspects of the patterning of the OC. However, we will concentrate on the proneural genes, which are ultimately responsible for much of the cellular neurosensory development (Imayoshi and Kageyama, 2014). As will be apparent in this overview, proneural genes are also responsible for the continued patterning of the OC and the proper differentiation of hair cells. Indeed, the initial upregulation of an important gene for hair cell differentiation, Atoh1, can happen in the complete absence of the canonical Delta/Notch signaling, but requires lateral inhibition for sustained expression (Basch et al., 2011). Unfortunately, it remains unclear what defines the position of the OC as well as the distribution of the inner and outer hair cells. In contrast to the loss of function effects, ubiquitous expression of Atoh1 leads to nearly ubiquitous differentiation of hair cells that nevertheless form some semblance of an OC in the right position (Kelly et al., 2012).

As outlined in multiple reviews, proper positioning of the OC and its cellular elements is crucial for the function of the OC. To emphasize this point about the importance of cell type and positioning, even slight changes in numbers and distribution of outer pillar cells through manipulation of Fgfr3 signaling (Puligilla et al., 2007), or reduction or loss of the inner hair cells in Srrm4 mutants (Nakano et al., 2012) or loss of the outer most row of outer hair cells in Fgf20 mutants (Huh et al., 2012) can render the OC dysfunctional. Attempts to regenerate the OC require an understanding of the signals that position the OC properly and differentiate the right hair cell and supporting cell type in the right position in the correct numbers and correct proximity, aspects of OC development we just begin to understand (Jahan et al., 2013, Kopecky et al., 2013). Given the unique distribution and number of inner pillar cells, it is fair to say that understanding OC development requires minimally a causal explanation on how the position and distribution of the single row of inner pillar cells and inner hair cells (Fig. 5) is achieved, in addition to the checkerboard patterning of the outer hair cell compartment that appears to fit so well to the lateral inhibition model (Sprinzak et al., 2011).

Beyond the position of cells and their distribution pattern in a radial organization is the systematic longitudinal variation of cells within the OC (Slepecky, 1996). While these variations are functionally meaningful as they may help tuning the OC to different frequencies (Slepecky, 1996), there is only limited understanding how this comes about molecularly, which is mostly correlative with causality still remaining to be explored. One such correlation is cell cycle exit and onset of differentiation (Jahan et al., 2013).

The OC shows a unique pattern of cell cycle exit followed by a different pattern of gene expression that regulates the differentiation process. Whereas the proliferation of hair cells progresses from the apex toward the base (Ruben, 1967), the expression of Atoh1 and differentiation of hair cells starts near the base and progresses bi-directionally towards the base and the apex (Chen et al., 2002, Matei et al., 2005). Importantly, changes in expression of Atoh1 result invariably in disorganization of hair cells and hair cell types in areas with such alterations of expression. For example, in Lmx1a mutants there is a delayed expression of Atoh1 in the basal turn and the entire basal turn of the cochlea develops into a hybrid set of disorganized vestibular and cochlear like hair cells that do not show the patterning of the OC into recognizable inner and outer hair cells (Nichols et al., 2008). In contrast, the apex shows more normal Atoh1 expression timing and develops into a near normal OC. Similar but less profound disorganization of the OC occurs when Atoh1 is prematurely expressed in mutants null for Neurog1 (Matei et al., 2005) or Neurod1 (Jahan et al., 2010b). Manipulation of the cell cycle by mutating the bHLH transcription factor N-Myc results in premature cell exit but delayed expression of Atoh1 and disorganization of apical hair cells so that no OC organization is apparent (Kopecky et al., 2013). Combined, these data point out that timing of cell cycle exit relative to expression of genes that affect timing and level of expression of Atoh1 is crucial for normal OC development (Jahan et al., 2013). This basic idea has recently been used to explain the limited changes in the OC observed after combined Hey1 and Hey 2 knockout (Benito-Gonzalez and Doetzlhofer, 2014)

If this aspect of regulated and differential delay between hair cell cycle exit and expression of Atoh1 to differentiate hair cells is so critical for the normal development of the OC, one would expect that several transcription factors need to cooperate to stabilize this pattern against all changes to prevent misguided differentiation leading to a non-functional OC. Despite significant efforts to detect the regulatory network for cell cycle exit we have only factors in our hand that maintain the non-proliferative state of hair cells once they exited the cell cycle (Chen et al., 2002), factors that are needed for overall OC development such as Sox2 (Kiernan et al., 2005), various Delta/Notch ligands (Doetzlhofer et al., 2009, Kiernan et al., 2006), Pax2 (Bouchard et al., 2010), micro RNA (Kersigo et al., 2011, Soukup et al., 2009), and Gata3 (Duncan and Fritzsch, 2013, Karis et al., 2001). Levels of Gata3 and Sox2 define hair cell differentiation in interaction with Atoh1 (Dabdoub et al., 2008, Duncan and Fritzsch, 2013) and micro RNA is crucial for normal hair cell differentiation (Groves et al., 2013, Kersigo et al., 2011). In contrast to this data, there is no data on the molecular definition of cell cycle exit beyond effects of other bHLH genes such as Neurog1 (Matei et al., 2005) and N-Myc (Kopecky et al., 2013). Clearly, neither gene defines hair cells but plays a yet to be specified role at some point in the precursors of hair cells. In mutants for either gene, ectopic hair cells form in the greater epithelial ridge of the middle turn and the ductus reuniens, both areas with extensive delamination of SGNs (Fig. 2).

The hypothesis of neurosensory evolution and development of the vertebrate hair cells and neurons as a split of an ancestral neurosensory cell through gene duplication and diversification (Fritzsch et al., 2006a, Fritzsch and Straka, 2014, Pan et al., 2012b) has been confirmed for the vestibular part of the ear through lineage tracing but thus far, not yet for the cochlea (Raft and Groves, 2014, Raft et al., 2007). However, more clear-cut data are needed for the more derived development of the OC despite obvious indirect evidence to its validity (Matei et al., 2005). This inability to verify the clonal relationship of SGNs and OC cells adds to the overall unusual development of the OC that might tie into its unusual pattern of cells and its inability to regenerate whereas mammalian vestibular organs have some capacity to regenerate (Golub et al., 2012). In this context it needs to be stressed that simple manipulations such as removal of Neurod1 can change surviving sensory neurons into hair cells, indicating lasting plasticity of neurons to differentiate into hair cells, a potential that is normally suppressed by Neurod1 (Jahan et al., 2010b). These data support the notion that cellular decision making is not a simple process of expressing a single gene that drives differentiation of a specific cell type but rather is specified by an intricate cascade of interacting transcription factors with a coordinated change in expression over time to guide development of different cell types (Fritzsch et al., 2006a, Imayoshi and Kageyama, 2014, Reiprich and Wegner, 2014).

The simple fact that expression of the bHLH gene Atoh1 can induce differentiation in almost any cell of the developing ear (Kelly et al., 2012), including developing neurons, if not suppressed by other factors (Jahan et al., 2010b), has led to the simplified perspective that proper expression of Atoh1 is all that is needed to generate and, by logical extension, to regenerate hair cells of the OC (Woods et al., 2004). While this certainly applies for sensory epithelia with a simple checkerboard mosaic of hair cells and supporting cells, such simple mosaic is in the OC limited to the outer compartment (Fig. 5). Consistent with the unusual organization of the inner compartment is an unusual distribution of several factors that play a role in neurosensory development: Bmp4 and various Fgfs. While Bmp4 and Fgf10 are, for example, overlapping in canal cristae (Chang et al., 2004, Pauley et al., 2003), they are differentially distributed in the OC (Fig. 6). Mutations of Bmp4 (Ohyama et al., 2010) and Fgf20 (Huh et al., 2012) result in differential loss of hair cells and supporting cells, indicating that the patterned expression of these two diffusible factor classes has different positional information for the various cell types of the OC, presumably in interaction with local cell-cell interactions (Doetzlhofer et al., 2009). Consistent with this idea is the finding that alteration in the patterning of the OC always results in a changed expression pattern of Fgf8 (Jahan et al., 2010b, Nichols et al., 2008), which in turn regulates differentiation of supporting cells into a different phenotype (Jahan et al., 2013). It appears to be essential for the normal patterning of the OC to have the positional distribution of these diffusible factors to the medial and lateral edge of the OC (Groves and Fekete, 2012). Intriguingly, the expression of some Fgfs is in turn maintained by the normally differentiating OC (Pan et al., 2012a, Pan et al., 2011). It appears possible that the developing OC is part of a feedback loop whereby diffusible factors outside the OC define in their interactive gradients (Srinivasan et al., 2014) both position and type of cells. Apparently, the differentiating OC releases a yet-to-be identified signal to maintain such expressions for continued reinforcement of differentiation, fine-tuned by the proper cellular interactions via the Delta/Notch lateral inhibition system.

Fig. 6.

The distribution of OC cells and the origin and hypothetical gradients of several diffusible factors (Fgfs, BMP4) are shown in control (a,b) and Neurod1 null ears (c,d). Note that the conversion of outer hair cells into Fgf8 positive hair cells in Neurod1 null mice results in transformation of surrounding Deiter's cells into pillar cell like cells (white arrows in d), indicating that a second center of Fgf8 diffusion can disrupt the cellular mosaic of the outer compartment of the OC (c,d). In essence, diffusible factors cooperate with lateral inhibition and selected expression of other genes to coordinate the normal development of the OC. Misexpression of only one diffusible factor (Fgf8 in some OHCs) can override all other interactions leading to a disrupted development of the OC (Jahan et al., 2010b). Bar in b and d indicates 100 μm. a,c,d, is modified after (Groves and Fekete, 2012, Jahan et al., 2013), b is unpublished data.

How these known and unknown factors combine to regulate the spatiotemporal expression progression of Atoh1, which in turn seems to define major aspects of OC development, is not yet established in its detailed causality. Indeed, some aspects of Atoh1 expression defy simple explanations of OC patterning: the expression of Atoh1 in a large subset of inner pillar cells, a unique single row of adjacent supporting cells (Fig. 5). The initial observation of this expression of Atoh1 in these cells (Matei et al., 2005) could have been a problem of the incomplete regulation of the enhancer fragment used in the reporter line despite the fact that it requires prior expression of Atoh1 protein to drive this enhancer (Chen et al., 2002). However, subsequent data using different molecular strategies have meanwhile confirmed the original finding (Driver et al., 2013, Yang et al., 2010), raising the intriguing possibility that Atoh1 expression, if counteracted by numerous co-expressed supporting cell differentiating bHLH genes (Benito-Gonzalez and Doetzlhofer, 2014, Doetzlhofer et al., 2009), is mostly compatible with inner pillar cell differentiation. In other words, even in the ear Atoh1 expression is only driving hair cell differentiation if the cellular context of gene expression is right. This, in turn, stresses the intricate interaction of levels of various bHLH gene expressions as the driver for differential hair cell type development of the OC (Jahan et al., 2013). These data also stress the uniqueness of inner pillar cells, already outlined at the introduction to this section. It appears possible that inner pillar cells are the single row of Atoh1-lacZ positive cells remaining in Atoh1 null mice (Fritzsch et al., 2005b) as suggested also in conditional deletions of Atoh1 prior to hair cell differentiation (Pan et al., 2011). In fact, inner pillar cells may play a unique role in the extension of the OC (Yamamoto et al., 2009).

In summary, the mammalian OC is molecularly highly derived vestibular end organ in which the ubiquitous Delta/Notch signaling has been modified through the action of unknown factors, including re-patterning of diffusible factors such as Fgfs and Bmps. This re-patterning generates continuous rows of adjacent hair cells and supporting cells with broad contact to each other in the inner compartment and a near ideal checkerboard pattern of alternating outer hair cells and Deiter's cells in the outer compartment (Fig. 5). Available evidence suggests that the overall patterning of the OC is tied into the unconventional cell cycle exit and Atoh1 upregulation mode of the OC relative to spiral ganglion cells (Matei et al., 2005), but details of causality are unclear. A notable cell, unique to the OC, is the inner pillar cell, which also has unique molecular signatures such as p75 and Atoh1 expression that are not shared with other supporting cells. Moreover, the continuous contact with surrounding supporting cells (basal and apical inner pillar cells in the longitudinal extent, inner phalangeal cells and outer pillar cells in the radial plane) makes this cell highly unusual. It is therefore noteworthy that the inner pillar cells seems to be the only cell remaining for a long term in otherwise completely degenerated OC of Pou4f3 (Brn3c) null mutant mice (Pauley et al., 2008, Xiang et al., 2003) or mice with conditional deletion of Atoh1 (Pan et al., 2012a, Pan et al., 2011). Given this highly derived development of the inner compartment of the OC, it seems necessary to understand how this becomes organized to have the possibility of regenerating a functional OC for which the position of the inner pillar cell over the bony lip of Rosenthal's canal with the single inner hair cell being immediately medial to it appear to be essential for its function. In essence, we need to be able to explain causally how a cell that expresses Atoh1 (Fig. 2) can be flanked by other supporting cells and does not differentiate as a hair cell but remains the inner pillar cell. Solving that problem requires generation of mutant mice in which the fate of the inner pillar cells is altered to understand the molecular conditions that normally specify this crucial cell for OC function and development, possibly including the extension of the OC of eutherian mammals.

4. From OC to a ‘placode-like’ epithelium and back: neurosensory loss in transcription factor mutants, the aging population and attempts to regenerate an OC

Hair cells are progressively lost with age, typically starting from the base and progressing toward the apex. Various chemicals as well as loud sound can accelerate this process and several models using chemicals and sound to ablate hair cells have been designed to study what happens in the OC upon sudden or progressive loss of hair cells. We concentrate here on hair cell loss following mutation in genes that are demonstrably necessary for continued viability of hair cells. Chief among those genes are four transcription factors, Atoh1, Pou4f3, Gfi1 and Barhl1. Mice with mutations of these genes lose all hair cells but in a different pattern of temporal and spatial progression. Similarly to these genes and their potential regulation, directly or indirectly by Atoh1 (Ikeda et al., 2014), manipulations of Atoh1 with different approaches lead to somewhat different results. Loss of Atoh1 through replacement with a LacZ reporter or conditional deletion prior to differentiation leads to absence of any hair cell differentiation but the retention of many LacZ positive cells (Bermingham et al., 1999) along the entire length of the OC (Fritzsch et al., 2005b) and not to the complete loss of all cells, as previously reported (Chen et al., 2002). While many cells die via apoptosis In Atoh1 null mutants, the overall cochlear extension, attributed by some to convergent extension and intercalation of OC cells (Kelly and Chen, 2009) is normal. Likewise, the expression of factors expressed in and adjacent to the OC such as neurotrophins (Fritzsch et al., 2005b), Fgf10 and Bmp4 (Pan et al., 2011) is initially normal in Atoh1 null mutants. However, loss of hair cells causes loss of Fgf10 expression in the GER as well as absence of Fgf8 expression and expansion of Bmp4 positive Claudius cells to replace the OC by a flat epithelium composed of cells that normally never differentiate as part of the OC (Pan et al., 2012a). Most intriguing is the persistence of a single row of cells along the entire length of the OC that could possibly present the inner pillar cells (Fritzsch et al., 2005b, Pan et al., 2011), now well characterized as being mostly Atoh1 positive (Driver et al., 2013, Matei et al., 2005) and important for extension of the OC (Yamamoto et al., 2009). Interestingly, in Foxg1 null mutant, the OC is truncated and composed of multiple rows of hair cells whereas the pillar cells are barely recognizable (Pauley et al., 2006). It is possible that the lack of pillar cell differentiation to drive extension of the OC, and not the inability of the multiple rows of HCs to conduct convergent extension movements, is the reason for this short and wide OC.

Very intriguingly, a few pillar cells surrounded by Myo7a positive cells differentiate even in the complete absence of Atoh1 in areas that show in earlier stages persistent expression of Sox2 (Pan et al., 2011). These data show that Atoh1 is a critical differentiation factor but some differentiation toward an OC can occur transiently in its absence. This point is particularly obvious in mutants in which Atoh1 is replaced by Neurog1, a related bHLH factor expressed in neurosensory precursors but not in differentiated hair cells (Jahan et al., 2012). These mice show a pattern of Neurog1 expression consistent with the pattern of LacZ markers replacing Atoh1 (Fritzsch et al., 2005b) with a major difference: there is a patchy development of cells that show a large tuft of microvilli consistent with undifferentiated hair cells generated in micro RNA deficient mice (Soukup et al., 2009). Moreover, loss of Neurog1 expressing cells is patchy instead of showing a continuous row as in Atoh1 null mice (Fritzsch et al., 2005b), indicating that there is some residual signaling of Neurog1 to partially differentiate cells of an OC through lateral inhibition via the Delta/Notch signaling pathway (Jahan et al., 2012).

This partial differentiation of an OC is even more obvious in mice in which there is delayed deletion of Atoh1 through a Cre driven by its own enhancer (Matei et al., 2005), generating a ‘self-terminating’ Atoh1 expression (Pan et al., 2012a). Interestingly enough, in these mice there is a near complete loss of inner hair cells and the third row of outer hair cells, suggesting the differential need of levels of Atoh1 expression for inner hair cells which in turn may induce proper differentiation of surrounding inner phalangeal cells to release Fgf20 needed for development of the third row of outer hair cells (Huh et al., 2012). While all these data on conditional and full deletion mutants of Atoh1 are inherently consistent, some more recent data do not fit well with these earlier data. Conditional deletion of Atoh1 using inducible Cre results in rapid and complete loss of all hair cells, both already partially differentiated ones and undifferentiated ones (Cai et al., 2013, Chonko et al., 2013), suggesting a critical phase of continued dependency on Atoh1 once differentiation has been initiated. It appears that Atoh1-positive differentiating hair cells are necessary for the continuous differentiation of supporting cells, which rapidly die in these mice. Why this induced Cre also affects the apical hair cells that have not yet expressed significant amounts of Atoh1 (Chen et al., 2002, Matei et al., 2005) remains unclear. It is possible that the induced expression of Cre can have adverse effects on undifferentiated hair cells as previously reported for neurons (Forni et al., 2006) and additional work is needed to rule that out.

In summary, these data suggest a difference in overall response of the OC development when Atoh1 is completely absent in mice null for Atoh1 (Bermingham et al., 1999, Fritzsch et al., 2005b) or with deletion of Atoh1 prior to its expression (Pan et al., 2011) as compared to delayed loss of Atoh1 using different non-inducible (Pan et al., 2012a) and inducible Cre lines (Cai et al., 2013, Chonko et al., 2013). Much of the interpretation presented here on various Atoh1 mutants is in line with data collected in mice with a delayed deletion of genes necessary for hair cell maintenance, which we discuss next.

Pou4f3 (aka Brn3c, Brn3-1) is a Pou domain transcription factor that is essential for hair cell maintenance and, indirectly, for sensory neuron maintenance (Xiang et al., 2003). Without this transcription factor, hair cells form and start to differentiate, but eventually all degenerate in late embryos and early neonates (Hertzano et al., 2004, Xiang et al., 2003). Despite the complete and rapid loss of all hair cells, which results in a nearly completely flat epithelium, some pillar cells, possibly inner pillar cells judging from the distance to the habenula perforata, survive for as much as six months. Most interesting is the expression change of Atoh1 in these mutants. First, Atoh1 stays expressed in what may be undifferentiated hair cells for several weeks (Pauley et al., 2008). In addition, cells surrounding the Tubulin positive pillar cells express Atoh1-LacZ even in six months old animals. What remains unclear is if the loss of hair cells results in altered expression of Atoh1-LacZ in these cells, as suggested for tamoxifen induced conditional Atoh1 deletion (Cai et al., 2013) or whether some hair cells survive as undifferentiated cells in the absence of Pou4f3. Pou4f3 is regulated by multiple transcription factors, including Atoh1 (Ikeda et al., 2014). It appears that the ability of Pou4f3 to induce hair cell differentiation is directly proportional to the binding of multiple transcription factors to various enhancer elements (Ikeda et al., 2014), providing a glimpse how altered expression levels of Atoh1 (Jahan et al., 2013) might affect differentiation of hair cells. Clearly, expression of Atoh1 without Pou4f3 cannot initiate late differentiation of hair cells in six month old animals, raising issues concerning the use of Atoh1 expression to differentiate hair cells in humans carrying mutant Pou4f3 genes.

Overall similar effects of a delayed loss of hair cells have been reported in mice null for the zinc finger transcription factor Gfi1 (Hertzano et al., 2004, Wallis et al., 2003). Like with the loss of Pou4f3, which seems to regulate expression of Gfi1, there is a progressive base to apex loss of hair cells in late embryos and neonates. In contrast to all these factors, the homeodomain-containing transcription factor Barhl1 shows a different progression both in time and place. Hair cell degeneration starts after onset of hearing and progresses from the apex toward the base in Barhl1 mutant mice (Li et al., 2002). Moreover, like Atoh1 (Jahan et al., 2012, Pan et al., 2012a), Barhl1 seems to positively regulate its own expression (Chellappa et al., 2008).

These data indicate that hair cell loss can be caused by several factors, progressing either from base to apex or vice versa. More detailed analysis of how the loss of hair cells affects overall retention of supporting cells is needed to understand better the clinical significance of loss of hair cells in human patients with mutations in these factors. However, it is remarkable that simple delayed loss of Atoh1 or loss of several other transcription factors can cause rapid demise of hair cells leading to the nearly flat epithelium, also observed in many human cases of long term hair cell loss. Equally remarkable is that in most of these mice with genetically induced hair cell loss there is substantial and long lasting innervation of parts of the OC (Fritzsch et al., 2005b, Pan et al., 2011, Xiang et al., 2003), which resembles more with human hearing loss conditions compared to hair cell loss induced by other means (Alam et al., 2007).

Why do these data matter for the current attempts to bring hearing back to the deaf that have lost their hair cells? There are many ways in which limited expression or hypomorphic function of these and several other genes could be combined in a single person to lead to progressive hair cell loss that will be different in spatio-temporal progression. In part such loss of hair cells will depend on the unique genetic predisposition as recently demonstrated (Ishimura et al., 2014), but will also in part depend on the unique combined exposure to loud sound and ototoxic chemicals throughout one's life. We are not yet in a position to go beyond personalized medicine and move toward an individualized age-related hearing loss assessment by combining both genetic predisposition and life accumulations of insults into a predictability matrix to estimate overall risk. Such information is needed to induce the right therapy for a given person at a given age. Depending on the residual presence of hair cells or remaining supporting cells, including in particular inner pillar cells, different strategies could be designed to restore iteratively the lost hair cells and reconstitute a semblance of an OC that is functionally meaningful (Zine et al., 2014). Attempts to understand the molecular composition of the different types of hair cells through hair cell specific gene expression analysis(Liu et al., 2014) will help to understand how the different hair cell types develop (Jahan et al., 2013).