In all species, neurogenesis is confined to spatially restricted regions of the otic cup, partially overlapping with zones that produce the sensory cells but excluded from those fated for non-neurosensory tissues in the mature inner ear. In rodents and birds, the neurons of the inner ear arise in the anteroventral quadrant of the otic vesicle (Figs. 2.1 and 2.2). Although intuited from histological studies, which revealed an obvious region of delamination (Knowlton, 1967; Carney & Silver, 1983), in vitro fate mapping studies ultimately confirmed that this portion of the otocyst produces neurons when dissected and cultured in isolation (Li et al., 1978; Adam et al., 1998), but that more dorsal and more posterior regions do not (Li et al., 1978). Similarly, dye labeling of the chick otic cup (Hamburger Hamilton [HH] stage 12) showed that CVG neurons are produced from the anterior compartment (Abello et al., 2007). Moreover, cells in this region rarely mixed with cells in the neighboring “non-neuronal” compartment and respected gene expression boundaries, indicating that the region of neurogenesis is patterned at an early stage.

The neurogenic zone itself appears to be further patterned, as vestibular neurons are generated before auditory neurons and from spatially distinct populations, as evidenced by dye labeling experiments in chicks (Bell et al., 2008) and genetic tracing experiments in mice (Koundakjian et al., 2007). Indeed, auditory and vestibular neurons appear to delaminate from different regions of the otic vesicle, with vestibular neurons developing close to the vestibular sensory epithelia and most auditory neurons instead delaminating from the boundary between the cochlea and saccule and then from the middle and apical turns of the cochlea itself (reviewed in Yang et al., 2011). Such spatial segregation is particularly extreme in fish, in which there are two separate neurogenic zones, with the anterior region producing neurons that innervate the utricle and the posterior region producing neurons that innervate the saccule, with anterior neurogenesis occurring slightly earlier (Haddon & Lewis, 1996; Haddon et al., 1998; Sapede & Pujades, 2010).

Importantly, in each species, the neurogenic zone is closely associated with regions of the epithelium that produce sensory cells, namely the hair cells and supporting cells of the utricle, saccule, cristae, and cochlea. Hence, the neurogenic zone is actually contained within a larger “proneurosensory domain” (PNSD) that contains both neural and sensory progenitors, with the neurogenic region overlapping with the nascent sensory epithelia for the utricle and saccule (Cole et al., 2000; Raft et al., 2007). Other sensory areas, such as those for the cristae, can arise either from apparently non-neurogenic regions in the PNSD or outside of the PNSD altogether, as suggested by expression of sensory markers and fate mapping (reviewed in Fekete & Wu, 2002).

Although the otic vesicle is the primary source for inner ear neurons, there is a long history of studies considering the possibility of a contribution from the neural crest and/or neuroepithelium. In fact, Bartelmez argued strongly that auditory neurons derive from the neural tube based on his histological analysis of early human embryos (Bartelmez, 1922). In contrast, early embryological experiments using the larval salamander indicated that most if not all neurons derive from the otic placode, whereas the neural crest produces the Schwann cells that myelinate the inner ear neurons (Yntema, 1937). Similarly, when neural crest cells were transplanted from quails to chicks, many quail-derived glia populated the mature ganglia and eighth nerve, confirming that auditory Schwann cells share the same neural crest origin as those in the rest of the peripheral nervous system (D’Amico-Martel & Noden, 1983). More surprisingly, some quail-derived cells also appeared to develop as neurons, but these seemed more likely to be vestibular based on their location. However, it was not possible to rule out a contaminating non-neural crest cell population, nor were markers used to verify the neuronal identity of the quail-derived cells.

More recently, an argument for a neural crest contribution was made based on genetic fate mapping studies in mice (Freyer et al., 2011). In these experiments, neural crest cells were permanently marked by Cre-mediated recombination of fluorescent reporters. In mice that produce Cre under the control of the Pax3 promoter, which is active throughout the dorsal neural tube, fluorescent cells could be seen moving from the neural tube into the otic vesicle, eventually contributing to the ganglion, maculae, and cochlea. However, these types of experiments can be difficult to interpret, as even undetectably low and likely physiologically irrelevant levels of Cre protein might be sufficient to induce recombination. In addition, Pax3 expression is not restricted to the neural crest, so many of these cells may be occasional neuroepithelial precursors that accidentally found themselves in the otic cup, which is pressed up against the hindbrain, but were nonetheless able to proliferate and differentiate within this new environment. In fact, dye labeling of cells in the early embryonic chick hindbrain revealed a similar contribution to various tissues of the inner ear, including the CVG (Ali et al., 2003). Hemond & Morest, (1991b) further noted a possible contribution from migratory cells formed at the boundary of the otic cup in chicks. These so-called “otic crest” cells seemed to stream toward multiple ganglia, but a specific contribution to the CVG was not defined.

In fact, many experiments argue that any non-otic contribution to the neurons of the inner ear is minimal or even nonexistent. For instance, complementary fate mapping studies using multiple independent Cre lines with expression in the otic vesicle, namely Foxg1–Cre (Hebert & McConnell, 2000), Pax2–Cre (Ohyama & Groves, 2004), and Pax8–Cre (Bouchard et al. 2004), suggest that the vast majority if not all inner ear neurons do in fact derive from the otic epithelium. In addition, fate mapping with Wnt1–Cre as well as a more restricted Pax3–Cre driver did not reveal any substantial contribution to the inner ear besides Schwann cells (Sandell et al., 2014). The conflicting results for Wnt1–Cre could reflect variation in the level of Cre activity in combination with different reporters and on different genetic backgrounds. Because the currently available “otic” Cre lines also mediate recombination in the neuroepithelium, albeit within highly restricted regions (Hebert & McConnell, 2000; Bouchard et al., 2004; Ohyama & Groves, 2004), generation of a truly otic-specific Cre line may be necessary for final resolution of this issue. Nonetheless, most studies indicate that the otic vesicle does indeed serve as the primary source for inner ear neurons.

Initial efforts to understand how the PNSD is patterned showed that the anterior–posterior axis is fixed remarkably early. For instance, anterior pieces of the chick otic epithelium are able to generate neurons in vitro, even in the absence of any surrounding tissues (Adam et al., 1998). Similarly, when otic cups were transplanted with a reversed orientation in HH stage 16 chick embryos, the PNSD remained in its original position (Wu et al., 1998). However, a similar manipulation between HH stages 10–12 caused the L–fng domain to form in what was originally the posterior half of the otic cup (Bok et al., 2005). Thus, signals from the surrounding tissue initially influence where the PNSD will form, but the axis is fixed just after the 16 somite stage (HH stage 12). In contrast, the dorsal–ventral axis remains sensitive to changes in orientation for longer, with important cues provided by the notochord, floorplate, and dorsal neural tube (Wu et al., 1998; Bok et al., 2005). Changes to the dorsal–ventral axis can also affect the position of the neurogenic domain (Bok et al., 2005; Riccomagno et al., 2005), highlighting the complexity of the interactions that ultimately shape the PNSD. However, the timing of events suggests that signals along the anterior–posterior axis provide the primary cues for PNSD formation.

Embryological experiments have helped narrow down the possible source of the cues that pattern the otic cup. The hindbrain does not appear to provide essential anterior-posterior information, as rotations prior to the 16 somite stage did not affect the location of the PNSD (Bok et al., 2005). Instead, signals seem to come from the nearby periotic ectoderm and somatic mesoderm (Bok et al., 2011). A major component of this signal is retinoic acid (RA), which prevents neurogenesis when ectopically expressed in chicks or mice. Conversely, blockade of RA signaling causes expanded neurogenesis. Further, RA and other components of the pathway are present and actively maintained in tissues surrounding the early otic cup in chicks. A similar role for RA has been described in zebrafish (Radosevic et al., 2011), suggesting this is an evolutionarily conserved mechanism, though differences among species likely exist in terms of the source of RA, how the gradient is established, and how the signal is interpreted. Indeed, the complex expression patterns and wide range of teratogenic effects across species indicate that RA can affect multiple aspects of inner ear development (reviewed in Romand et al., 2006).

Additional critical patterning information appears to be provided by FGFs. In chicks, FGF8 is expressed close to the otic territory prior to the onset of neurogenesis and can promote neurogenesis when expressed ectopically (Abello et al., 2010). Conversely, broad pharmacological inhibition of FGF signaling caused a loss of PNSD markers, accompanied by an expansion of the nonsensory marker Lmx1b. Similar alterations in BMP signaling had no effect on neurogenesis, but did influence Lmx1b expression, confirming a role in anterior–posterior patterning. The expansion of a nonsensory marker in the absence of an effect on neurogenesis indicates that FGF and BMPs may act independently to pattern the axis, with additional signals such as RA influencing the final outcome. For instance, as in chicks, an FGF ligand is required for anterior-posterior patterning in fish (Hammond & Whitfield, 2011), but some of these effects may be due to changes in expression of genes required for RA metabolism (Radosevic et al., 2011).

Whether FGFs regulate PNSD formation in mice remains unclear, due in part to differences in how and when neuronal development has been assessed. The PNSD appears to be present in FGF3 mutants, though the CVG is noticeably smaller once it forms (Hatch et al., 2007; Vazquez-Echeverria et al., 2008). However, loss of both FGF3 and FGF10 can cause expanded neurogenesis (Vazquez-Echeverria et al., 2008). Similarly, Neurod1-positive cells form in ectopic locations in Kreisler mutant mice, which suffer from hindbrain patterning defects including reduced expression of FGF3 and FGF10, and this phenotype can be partially rescued by restoration of either FGF. On the other hand, loss of the FGF2R (IIIb) receptor, which binds both FGF3 and FGF10, does not impair early CVG development (Pirvola et al., 2000; Pauley et al., 2003). However, because early PNSD development was not assessed, it is possible that a similar phenotype was missed in these animals (Pirvola et al., 2000). Altogether, the exact contribution of FGF3/FGF10 signaling is difficult to pinpoint due to the variability of reported double mutant phenotypes (Wright & Mansour, 2003b; Vazquez-Echeverria et al., 2008) and the fact that the phenotypes do not recapitulate what is seen in FGF2R mutants (Pirvola et al., 2000; Pauley et al., 2003). Why FGFs may antagonize neurogenesis in some contexts yet induce neurogenesis in others, as well as the specific contributions of various family members, will be important to work out in the future.

Although anterior–posterior signaling pathways function first, the final shape and size of the PNSD is also influenced by cues that pattern the dorsal–ventral axis. Shh from the notochord and floor plate provides a potent ventralizing signal that is necessary for normal expression of sensory and nonsensory markers (Brown & Epstein, 2011). However, loss of the PNSD from Shh mutant mice does not appear to be a direct effect on otic vesicle patterning, as otic-specific ablation of the Shh receptor Smoothened did not reproduce this effect. Instead, the change in PNSD gene expression appears to be secondary to expanded Wnt signaling from the neural tube, which is dorsalized in Shh mutants (Chiang et al., 1996; Riccomagno et al., 2005). In support of this idea, activation of Wnt signaling also inhibits neurogenesis (Ohyama et al., 2006; Brown & Epstein, 2011). However, importantly, Wnt signaling is not necessary for neurogenesis per se (Ohyama et al., 2006). Hedgehog proteins may play a more direct role during anterior–posterior patterning of the zebrafish otic vesicle, where they appear to work in parallel with—yet independently of—FGFs (Hammond et al., 2003, 2010; Hammond & Whitfield, 2011). The basis of these apparent differences among species is unclear but could reflect unique features of different types of vertebrate ears.

Although signals extrinsic to the ear likely initiate the events that place the PNSD in the anterior otic cup, transcriptional networks and local cell–cell interactions play an important role in reinforcing these decisions and ensuring the neurosensory competence of cells within this domain. The transcriptional networks serve two complementary functions: to ensure expression of genes necessary for that cell’s needs at that point in time and to control the activity of the network itself. The Cell–cell interactions coordinate these intrinsic events with the surrounding tissue.

One important player in the PNSD is the transcription factor Tbx1, whose expression is restricted to the posterolateral (i.e. nonsensory) region of the otocyst as early as the 10 somite stage in mice (Raft et al., 2004). Neurog1 and Neurod1, which mark neural progenitors in the PNSD, are expressed in a complementary pattern to Tbx1 in the otocyst (Fig. 2.2). In mice with excess Tbx1, neurogenesis is reduced, particularly in the posterior of the PNSD (Raft et al., 2004; Freyer et al., 2013). Conversely, in Tbx1 mutant mice, Neurog1 and Neurod1 expression is increased (Raft et al., 2004; Xu et al., 2007), likely due to an expansion of the PNSD (Raft et al., 2004). Moreover, in Tbx1 mutants, cells from the Tbx1 domain populate the ganglion, something that never occurs in wild-type animals (Xu et al., 2007). Thus, Tbx1 actively represses the PNSD fate in nonsensory cells, thereby confining neurogenesis to a restricted domain. Extra Neurog1– and Neurod1-positive cells also develop in Tbx1 mutant fish (Radosevic et al., 2011), and spiral ganglion defects have been reported in human patients with velocardiofacial/DiGeorge syndrome (Schmidt, 1985), which is linked to TBX1 mutations (Yagi et al., 2003; Zweier et al., 2007). Hence, Tbx1 appears to play a basic, evolutionarily conserved role in determining where neurogenesis will occur.

If Tbx1 acts to prevent the neurosensory fate within the nonsensory region of the otocyst, then what factors promote this fate in the PNSD? The positively acting pathways are more complex, involving multiple transcriptional networks that interact with each other to drive production not only of neurons, but also of hair cells and supporting cells that will form the organ of Corti. One factor that appears to participate in many of these fate decisions is the SoxB1 family member Sox2. In mice, Sox2 is expressed in the ventral rim of the otic cup by E8.5 (Wood & Episkopou, 1999; Zou et al., 2008) and then in the ventrolateral otocyst at E9.5 (Mak et al., 2009). Subsequently, Sox2 can be detected both in delaminating neuroblasts and in the developing sensory epithelia (Kiernan et al., 2005; Mak et al., 2009). Sox genes show a similar expression in the PNSD in chicks, with expression complementing Tbx1 and Lmx1b by the 10 somite stage (Abello et al., 2007, 2010).

Just as Tbx1 promotes the nonsensory fate, so does Sox2 drive cells down the neurosensory path. In mice harboring mutations in Sox2, neurosensory development is severely disrupted, with an early loss of both the prosensory domain (Kiernan et al., 2005) and of neurons (Puligilla et al., 2010). The loss of both cell types suggests that Sox2 may be required for the initial specification of precursors in the PNSD. Moreover, in chicks, ectopic expression of Sox3 inhibits Lmx1b, hinting that mutually antagonistic interactions may solidify the nonsensory versus PNSD fate decision (Abello et al., 2010): wherever Tbx1 is on, Sox2 will be off and vice versa. Interestingly, Tbx1 expression was not affected. Similarly, Lmx1a but not Tbx1 shows reduced expression in Kreisler mutant mice, indicating there may be multiple parallel pathways that influence the nonsensory fate (Vazquez-Echeverria et al., 2008). Indeed, in Lmx1a mutant mice, Sox2 expression expands, followed by the formation of fused sensory epithelia and a larger CVG (Nichols et al., 2008; Koo et al., 2009).

As early acting factors such as Sox2 direct cells toward the neurosensory fate, additional transcriptional networks cooperate to determine whether individual progenitors will produce neurons, sensory cells, or both (Fig. 2.3). Unraveling the logic of this progressive fate restriction has been complicated by the fact that there seem to be multiple types of neurosensory progenitors within the PNSD. In favor of this idea, fate mapping in zebrafish has revealed three different types of precursors: neurosensory, neural, and sensory (Sapede et al., 2012). Similarly, in chicks, there is solid evidence that a common neurosensory progenitor produces both hair cells and neurons in the utricular maculae, but not in the other sensory organs, consistent with the presence of a heterogeneous precursor population (Satoh & Fekete, 2005).

The basis of the observed precursor heterogeneity remains unclear. One possibility is that hair cells and neurons derive from a similar early neurosensory progenitor that gradually shifts its potential over time, with the latest remnants of this population showing an extended ability to produce both hair cells and neurons in the maculae. Such a progenitor would be difficult to detect using standard fate mapping techniques. Alternatively, there may be a spatial segregation within the PNSD, with common progenitors limited to one subregion. Indeed, across species, proven bipotent progenitors exist in regions where the sensory domain, presaged, for example, by expression of BMP4, is found within the PNSD itself (Cole et al., 2000; Raft et al., 2007; Sapede et al., 2012).

Within the otocyst, local cell–cell interactions reinforce the intrinsic pathways that ultimately determine which cells will contribute to the neuronal lineage, influencing both the initial formation of the PNSD and the subsequent segregation of neuronal and sensory precursors. Both of these binary fate decisions appear to be under the control of the Notch pathway. As initially established in Drosophila, the Notch receptor interacts with a transmembrane ligand called Delta to mediate lateral inhibition and thereby promote the acquisition of distinct cell fates within a field of multipotent progenitors (reviewed in Schwanbeck et al., 2011). A second related ligand, Jagged, plays a similar role. In mammals, the basic pathway is conserved, with four Notch receptors, three Delta-like ligands, and two Jagged ligands. Ligand binding induces cleavage of the intracellular domain (ICD) of the Notch receptor protein in the neighboring cell. The Notch-ICD then enters the nucleus to directly regulate expression of target genes. Among the target genes are additional transcription factors that feedback to increase expression of Notch itself while simultaneously decreasing Delta production in that cell. The overall consequence is that neighboring cells ultimately express either Delta or Notch and therefore adopt one of two possible fates, for instance, whether to become a neuron. Because of the lateral inhibition mechanism, early uniform expression of the ligand and receptor is often followed by a more salt-and-pepper–like appearance within that field of cells, reflecting the gradual emergence of two distinct cell fates. Thus, activation of the Notch receptor by a Delta-family ligand offers a direct way to convert an extrinsic signal into an intrinsic change in gene expression.

Understanding the precise effects of Notch signaling in otic neurogenesis has been challenging because of the presence of multiple ligands and receptors, as well as differences in the effects of these molecules throughout development (reviewed in Kiernan, 2013). However, several lines of evidence suggest that Notch signaling promotes neurogenesis during inner ear development. For instance, like Lfng (Morsli et al., 1998), which is a known regulator of the Notch pathway, the Delta homolog Delta–like 1 (Dll1) is expressed in the PNSD at early stages (Abello et al., 2007; Daudet et al., 2007). In addition, pharmacological inhibition of Notch signaling in chicks can increase the number of Dll1+ cells (Daudet et al., 2007). Conversely, when Dll1 is absent in mice, too many neurons develop, consistent with a loss of lateral inhibition between Dll1+ neurons and the surrounding cells (Brooker et al., 2006). Interestingly, the production of extra neurons in Dll1 mutant mice apparently comes at the expense of the saccular and utricular maculae, whereas there is an increase in hair cell number in the cochlea, providing further evidence that there is a defined common neurosensory progenitor for only a subset of hair cells and neurons. Jagged ligands, on the other hand, may not be involved in inner ear neurogenesis (Zhang et al., 2000; Brooker et al., 2006; Neves et al., 2011).

One of the key consequences of Notch signaling is to induce expression of potent basic helix loop helix (bHLH) transcription factors, which act both to regulate Notch–Delta production and to induce cell-type–specific programs of gene expression. Similarly, whereas many genes are uniformly expressed in the PNSD, Dll1 shows a more irregular pattern, apparently reflecting upregulation in early neuronal precursors (Adam et al., 1998; Abello et al., 2007; Daudet et al., 2007) which express the proneural bHLH transcription factors Neurog1 and Neurod1. In Neurog1 mutant mice, no inner ear neurons form and Dll1 expression is lost, consistent with the classic model of lateral inhibition, with Neurog1 acting both to enhance production of Dll1 and promote the neuronal fate (Ma et al., 1998; Raft et al., 2007). The Dll1- cells, on the other hand, likely adopt a prosensory fate, which is promoted by a different bHLH transcription factor, Atoh1 (formerly Math1), which is required for hair cell development in mice (Bermingham et al., 1999). When signaling downstream of Notch is prevented, Neurog1 expression increases, as would be expected if Dll1 is no longer able to inhibit expression of Neurog1 in neighboring cells (Raft et al., 2007). In addition, pharmacological inhibition of Notch in chicks enhances the local upregulation of Dll1, indicating that lateral inhibition normally segregates Notch+ and Dll1+ populations within the PNSD (Daudet et al., 2007). Indeed, ectopic expression of the Notch-ICD is sufficient to create ectopic sensory regions and can divert neuroblasts to the hair cell fate in vivo (Daudet & Lewis, 2005; Hartman et al. 2010; Pan et al., 2010; Liu et al., 2012).

Because Neurog1 and Atoh1 are essential for neuronal and sensory development respectively, one attractive idea is that these two transcription factors participate in mutually antagonistic interactions within PNSD precursors that ultimately produce dedicated neuronal (i.e., Neurog1+) or sensory (i.e., Atoh1+) progenitors. If this model is true, then Neurog1 and Atoh1 must be coexpressed, at least at low levels, in any common neurosensory progenitor. This clearly appears to be the case in the utricle and saccule, where stripes of Atoh1 expression appear within the neurogenic domain (Raft et al., 2007). Moreover, descendants of Neurog1+ cells, as marked by genetic fate mapping, do indeed populate the maculae, exactly as predicted both by gene expression studies and viral fate mapping (Morsli et al., 1998; Cole et al., 2000; Satoh & Fekete, 2005). Additional evidence has come from analysis of Neurog1 and Atoh1 mutant mice: loss of Neurog1 is accompanied by an increased number of Atoh1-positive hair cell precursors in the developing utricular macula, whereas there are more neural precursors in the maculae of Atoh1 mutant mice (Raft et al., 2007). Ultimately, however, fewer hair cells develop in Neurog1 mutants, particularly in the saccule, indicating that the extra Atoh1-positive progenitors may not differentiate properly (Ma et al., 2000; Raft et al., 2007).

Together, these findings provide strong support for a common neurosensory progenitor in the maculae. Notably, the three major drivers of neuronal and hair cell fates in vertebrates—Neurog1, Neurod1, and Atoh1—are all bHLH factors that are closely related to each other and to bHLH factors with similar functions in sensory development in invertebrates. Indeed, it has been proposed that transient coexpression and cross-regulation of Neurog1, Neurod1, and Atoh1 may reflect an evolutionarily ancient series of interactions, with an expansion within the family eventually leading to a situation where individual bHLH factors segregate to neural versus sensory precursors (Pan et al., 2012).

The situation in the cochlea is less clear. First, Atoh1 expression does not appear to overlap with the neurogenic region of the cochlea (Raft et al., 2007). Accordingly, genetic fate mapping revealed no contribution of Atoh1-positive progenitors to the CVG (Yang et al., 2010). Conversely, the Neurog1 population does not produce any cochlear hair cells (Raft et al., 2007), despite massive labeling of the spiral ganglion neurons (Koundakjian et al., 2007; Raft et al., 2007). Nevertheless, there are some tantalizing phenotypes in the cochlea that suggest that Atoh1–Neurog1 interactions may occur, though perhaps only transiently. During normal development, neurons exit the cell cycle before hair cells (Matei et al., 2005). In contrast, in Neurog1 mutants, hair cell precursors exit the cell cycle prematurely (Matei et al., 2005), and extra rows of hair cells develop in some regions (Ma et al., 2000). This phenotype can be explained in part by the absence of SGNs, which normally produce a Shh cue that prevents hair cells from exiting the cell cycle (Bok et al., 2013). However, intrinsic effects might also contribute: when Neurog1 is lost, a progenitor that would normally become a neuron may instead select the alternate fate and differentiate as a hair cell (Matei et al., 2005).