Animal body plans are the result of genomic mutations that augment specific developmental processes by modifying the spatial and temporal regulation of thousands of genes followed by selection of these modifications. Fossil and comparative anatomical data can raise significant questions about the progression of specific evolutionary changes but rarely provide enough material to generate a plausible theory. Most importantly, fossil data cannot provide answers to the underlying questions of genetic changes that drive differential development through altered cis-regulatory gene networks. It is only through comparative and functional analyses of developmental gene regulatory networks that one can directly discover genomic modifications leading to novel developmental variances and thus alteration of adult structure (Davidson and Erwin, 2006; Carroll, 2008). With a continual increase in the number of species whose genomes have been sequenced, comparative genomics allows us to examine gene multiplications, evolution of cis-regulatory elements, evolution of gene regulatory networks, conservation of existing genes, and their regulatory sequences among diverse species. It is these changes in the organization of the genomic code and thus gene regulation that causes alterations in the body plan on which natural selection can act (Davidson, 2006). In essence, the twenty-first century will reveal the molecular basis for innovations at the molecular, cellular, organ, and organismal level that will allow for the first time establishing the causality of gene changes with altered morphology. It is our view that regulating processes are evident and seemingly recapitulated throughout all levels of life. We, therefore, propose that the principle of duplication followed by segregation and diversification, well known at the level of both genes and population diversification, is also underlying the evolution of cells, tissues, and organs. This review attempts to provide a snapshot of our current limited understanding of the genetic, cellular, and organ changes in the ear toward the goal of establishing a molecular evolutionary theory of changes (Andreas, 2011) that produce the cell types and ear anatomy we see in extant vertebrate species today.
Here, we review the molecular basis of mechanosensory cell and mechanosensory organ development and evolution with an emphasis on the conservation of transcription factors and emerging data on conserved gene networks. The ear, the organ of vertebrates dedicated to the perception of sound and balance, perceives these stimuli with the use of mechanosensory cells. The developmental gene regulatory network used during mechanosensory cellular development has been conserved from ancient bilaterian cells, and modified for the extraction of specific mechanical stimuli resulting in phenotypic changes. In the vertebrate lineage, mechanosensory cells became specialized as gravistatic sensors after they became aggregated to form the ear. After this aggregation, growth, including duplication and segregation of existing neurosensory epithelia, gave rise to new epithelia and can be appreciated by comparing sensory epithelia from the inner ears of different vertebrates and their innervation by different neuronal populations. Novel directions of differentiation were apparently further expanded by incorporating unique molecular modules in newly developed sensory epithelia. For example, the saccule gave rise to the auditory epithelium and corresponding neuronal population of tetrapods, starting possibly in an aquatic environment. This novel sensory perception was followed by emergence of the central auditory nuclei and a selective cochlear nucleus projection. The data for this process is outlined and contrasted with other ideas dealing with a subset of the data. Anat Rec, 2012. © 2012 Wiley Periodicals, Inc.
Overview of Ideas Related to Ear and Lateral Line Evolution
Localized areas of epidermal proliferation and thickenings called placodes (von Kuppfer, 1895) develop along the head of most vertebrates (Baker and Bronner-Fraser, 2001; Northcutt, 2005) and to a limited extent may be present in chordates (Meulemans and Bronner-Fraser, 2007). Research into two of these placodes, the otic and lateral line, have historically been grouped together as both placodes have been claimed to possess a single developmental origin (Ayers, 1892; Wilson and Mattocks, 1897; Van Bergeijk, 1966, 1967) and their derivatives to terminate in the same primary hindbrain nuclei (Mayser, 1882; Herrick, 1897; Pearson, 1936; Larsell, 1967). Others (Wilson and Mattocks, 1897) speculated that both the lateral line and the labyrinth (ear) were derivatives of a single sensory placode originally dedicated to creating the lateral line and subsequently giving rise to the inner ear. Mayser (1882) examined the projection of the eighth cranial nerve in cyprinids and claimed that both inner ear and lateral line nerves project centrally to the “acoustic tubercle” of the medulla. Based upon the alleged convergence of the inner ear and lateral line nerves, Mayser suggested that the lateral line is an acoustic sensory organ renaming the “acoustic tubercle” the acousticolateral area. Following this idea, the inner ear's proposed origin from the lateral line system is known as the acousticolateralis hypothesis. It should be noted that at the turn of the century it was fashionable for scientists to regard the lateral line system as a starting point for all forms of sensation, and was hypothesized to give rise not only to the inner ear but also the senses of touch, smell, taste, and even whiskers, which bear a striking similarity to the lateral line in overall distribution. Likewise, vestibular sensation, as we now know, was not yet associated with the ear.
Taken together, the acousticolateralis hypothesis can be broken down into three key assumptions for the evolution of the ear from the lateral line (reviewed in Van Bergeijk, 1966): (1) Both the lateral line and ear develop from the same placode; (2) The lateral line and ear have identical sensory cell types; and (3) The lateral line and inner ear project to the same hindbrain nuclei.
The first assumption of the acousticolateralis hypothesis is that of ontological evidence, suggesting a common placode gave rise to both the lateral line and the inner ear. More recently, molecular evidence has been produced that shows this not to be the case (Schlosser, 2010). This adds to previous data that indicate a developmental time difference with the inner ear placode developing much earlier than the lateral line placode in amphibians (reviewed in: Fritzsch et al., 1998, 2005a). In addition, the lateral line has been lost in terrestrial vertebrates, indicating that these placodal regions can be independently modified. However, two ideas have been presented which deviate from this notion; one idea being that all placodes share a common ancestry and are phylogenetically derived from a single pan-placodal region that is later broken down into olfactory, lens, trigeminal, ear, and lateral line placode (Streit, 2001; Schlosser, 2010). The other idea being a common origin and physiological function for the lateral line, epibranchial, and otic placodes (Baker et al., 2008). Contrary to these ideas, placodes have also been thought to be mere aggregations of cells that are induced within a generalized field of ectoderm that is primed to produce placodes, but are not evolutionarily linked (Groves and Bronner-Fraser, 2000; Begbie and Graham, 2001). The latter view is supported by numerous placode-like appearances of organs such as taste buds, hairs, and feathers, simply indicating that forming placodal thickenings is the epidermis' way to generate a highly proliferative area that can be turned into different (neurosensory) organ systems depending on the specific overlap of molecular cues. Thus, how placodes enlarge or shrink, split or fuse during evolution is unclear. Be this as it may, only a heterochronic shift in relative development of ear and lateral line placodes could explain why the lateral line is typically developing last but should have evolved first.
The second assumption on which the aousticolateralis hypothesis relies is that the sensory cells of the lateral line and inner ear are both hair cells with very similar overall morphology of the mechanotransduction machinery, the idea being that this highly sophisticated morphology could have evolved only once. However, based on this fact alone there is no compelling evidence to demonstrate that the ear arose from the lateral line. In fact, it is equally plausible that the lateral line arose from the inner ear, or that they share a common origin from an organ which no longer exists. Moreover, lateral line placodes give rise to the nonmechanosensory “hair cells” of the electroreceptive system (Northcutt et al., 1995) thus minimizing the significance of this insight. Indeed, the mechanosensory hair cell may evolutionarily predate the formation of placodes and we will further focus on this idea in molecular detail below.
As stated above, early studies claimed that, inner ear and lateral line nerves project to the same hindbrain nuclei and extensively overlap within this single nucleus (Mayser, 1882; Herrick, 1897; Pearson, 1936). However, it has been known for over 30 years that the nerves of these two systems project to distinct nuclei within the hindbrain (Maler et al., 1973, 1974; Northcutt, 1979, 1980; Fritzsch, 1981) and a novel sensory organ system unknown to the initial proponents of the acousticolateralis hypothesis was in part established based on the unique projection to the dorsal electroreceptive nucleus, the ampullary electroreceptors (Bullock and Heiligenberg, 1986). While these studies agree that the lateral line and inner ear fibers both enter the hindbrain alar plate they also show that this region is not composed of a single nucleus, but consists of a series of nuclei which vary in number dependent upon the species and the absence or presence of a lateral line system (with or without ampullary electroreceptive organs) or an auditory system (Fig. 1; Fritzsch et al., 2006b).
Thus, the embryological and connectional data do not support the acousticolateralis hypothesis. In addition, fossil evidence of the early vertebrate, Ostracoderm kiaeraspis, claimed presence of both inner ears and lateral line nerves (Stensio, 1927; Long, 1995; Janvier, 1996) and thus does not shed light on this issue. Of extant vertebrates, hagfish have underdeveloped lateral line organs without the typical mechanosensory stair case appearance (Braun and Northcutt, 1997) and lampreys have both ampullary organs and mechanosensory lateral line organs (Bullock and Heiligenberg, 1986), raising the possibility that we either look at an evolutionary progression of organ development (ear before lateral line before ampullary electroreceptors) or a regression series (hagfish have lost all ampullary electroreceptors and reduced the lateral line to almost undeveloped placodes). None of this data unequivocally supports the acousticolateralis hypothesis.
The second idea on ear evolution is the statocyst hypothesis put forth by Wever (1972). This idea is based upon an apparent morphological conservation of the vestibular portion of the ear across phyla and that these “statocysts” have similarities in function and predate evolution of the vertebrates. Wever suggested that the small statocysts containing ciliated sensory cells of ctenophores, coelenterates, echinoderms, and crustaceans are precursors to the vertebrate ear. This hypothesis suggests that “hair cells” and ears of vertebrates have a common origin and evolved together throughout time, an idea that was rejected by other contemporaries (Markl, 1974). The statocyst hypothesis has recently been further supported by the molecular comparison of tunicate atrial chambers, which contain mechanosensory receptors and supporting cells, with the vertebrate inner ear. These molecular comparison has focused on the Pax2/5/8 gene present in the atrial chamber and is the homolog of the mammalian Pax2 gene. Pax2 and Pax8 have been shown to be necessary for development of the zebrafish, chicken, and mammalian ears (Li et al., 2004; Bouchard et al., 2010; Padanad and Riley, 2011). This suggests that at a molecular level, ears and statocysts may be as much homologous as are various eyes in their dependency on Pax6 (Kozmik et al., 2003).
A relatively recent third idea has been put forth (Jorgensen, 1989; Fritzsch and Beisel, 2001, 2003; Caldwell and Eberl, 2002; Todi et al., 2005), which we name here the “mechanosensory cell first” hypothesis. This idea proposes, as in the acousticolateralis hypothesis that the mechanosensory receptor evolved first and evolved into the modern hair cell independent from the evolution of the ear. The difference is that this theory does not presume the prior evolution of lateral line organs, and in fact suggests that the hair cell and its precursor the ciliated mechanosensory cell were present well before the lateral line and inner ear as organ systems evolved. This hypothesis also does not presuppose that hair cells evolved within a statocyst organ, but rather as diffuse cells in the skin, which later became aggregated to evolve into the ear.
MECHANOSENSORY CELL FIRST HYPOTHESIS
The mechanosensory cell first hypothesis proposes that the mechanoreceptor and thus the molecular machinery necessary for the formation of the mechanoreceptor is the link between disparate organs not only of lateral line and inner ear but also of mechanosensory cells of invertebrates, such as those of the Johnston's organ of flies. In each lineage, this molecular machinery was co-opted to form a mechanosensory cell in very different organ types and adapted to specific mechanosensory stimuli resulting in receptor resting potential changes. We propose that an ancient molecular network to specify mechanosensory cells is being used in different environments and in the known mechanosensory organs of extant animals to generate mechanosensory receptors. This is analogous to a duplicated gene producing a protein in a different spatio-temporal context and accruing alterations over time toward a novel function. Thus while the patterning and morphological context of these mechanoreceptor cells vary, they all use an ancient molecular developmental regulatory system to generate a ciliated mechanosensory cell in an organ that may not be homologous.
Molecular Origin of Mechanosensory Cells Predates the Lateral Line and Ear
It has been proposed that cell types, which play a role in mechanosensation, have all evolved from molecular machinery present in a common flagellate ancestor (Nerrevang and Wingstrand, 1970; Jorgensen, 1989; Fritzsch et al., 2007). Mechanosensory cells can be found in both vertebrate and invertebrate species; however, the hair cell with its asymmetric apical specialization is unique to vertebrates (Fig. 2). The choanoflagellates have a single kinocilium surrounded by microvilli. In studied diploblasts, the kinocilium is surrounded by microvilli but is slightly polarized to one end. In vertebrates, the final arrangement is a polarized kinocilium at one end of the cell; however, it does not start here during development. Early in development, the kinocilium is centrally located and surrounded by microvilli; it is only over the course of development that it becomes polarized. Typically, invertebrates are also unique in that they possess primary mechanosensory cells, a sensory cell with a connecting axon (Fig. 3). Vertebrates have apparently duplicated the ancestral primary sensory cell and the two resulting cells have distinguished their functions with one (secondary sensory cell) taking on the role of ciliated mechanotransduction and the other the role of transmitting information centrally (Fig. 3; Fritzsch and Beisel, 2004). Invertebrate cephalopod sensory cells exist without axons indicating that primary and secondary mechanosensory cell types are not exclusive to each group (Budelmann, 1992). Sensory cells without axons and have been described in the ascidians (Burighel et al., 2008, 2011) and in amphioxus (Holland, 2005). In the past, the reconstruction of hair cell evolution has relied almost exclusively on comparison of adult and developmental features of extant invertebrates and chordates with vertebrates. Arguments have been made against a conserved molecular program for mechanoreceptor cells (Coffin et al., 2004), suggesting that vertebrate hair cells are unique and are similar to invertebrate mechanoreceptors only through convergent evolution. However, those divergent ideas have been surpassed by abundant molecular data to the contrary and are no longer tenable in light of the overwhelming molecular evidence, as will be discussed below.
We aim to establish a plausible scenario for the continuity of interacting genetic networks, as well as for their modification, to achieve the unique vertebrate outcome from an ancient bilaterian mechanosensory cell. To obtain this, we need to compare early development of hair cells in vertebrates with invertebrate sensory cell development and analyze whether similar developmental gene regulatory networks and homologous genes are used to specify the divergent sets of cells needed to form a ciliated mechanosensory cell in divergent phyla. In contrast to the master control gene hypothesis used by others to describe both the inner ear and eye evolution, we take the position that no single gene however necessary to a developmental process is sufficient to produce a state of differentiation, and will instead focus on how gene regulatory networks may be conserved, with variation, among bilateria.
Similar Genetic Networks Specify Invertebrate Mechanosensors and Vertebrate Hair Cells
When trying to specify cellular homology, at least a few homologous genes should be used as an indication that cells are homologous, but only when sub-circuits within the developmental gene regulatory networks are the same or closely similar should the cells be considered homologous (Davidson, 2006). We show below that the similarities in gene networks of mechanosensory cells between Drosophila and vertebrates are striking and beyond mere superficial and objectionable similarities.
Differentiation of neurons and sensory cells are heavily reliant on specific transcription factors, the proneural basic-helix-loop-helix (bHLH) genes (Bertrand et al., 2002; Kageyama et al., 2005). The atonal family of bHLH genes has evolved with multicellular organisms (Seipel et al., 2004). In vertebrates, Atoh1 has been shown to be necessary for hair cell formation in mammals and zebrafish (Bermingham et al., 1999; Millimaki et al., 2007). This line of data can lead to the assumption that Atoh1 is a master regulatory gene for hair cell development. It is currently unclear how Atoh1 is affecting hair cell development as only undifferentiated precursors form and subsequently degenerate in Atoh1 null mice (Bermingham et al., 1999; Chen et al., 2002; Fritzsch et al., 2005b; Pan et al., 2011). It is clear that Atoh1 is not only needed to initiate development of hair cells but sustained expression of an unknown level is apparently required to maintain hair cells (Pan et al., 2012). Arguing against a master control gene hypothesis some hair cell markers such as Myo7a are expressed in the absence of Atoh1 (Pan et al., 2011; Ahmed et al., 2012). Insect mechanosensory cells also require bHLH genes to form: atonal for scolopidial organs and hearing, and achaete and/or scute for bristles (Jarman and Ahmed, 1998; Caldwell and Eberl, 2002). The Atoh1 gene in mammals can be functionally replaced with Drosophila atonal (Wang et al., 2002). Likewise, Drosophila mechanosensory cells can be rescued in the absence of atonal by the expression of mammalian Atoh1 (Ben-Arie et al., 2000). Thus, the binding capacity and specificity of atonal homologous genes, and the cis-regulatory sequences for mechanotransduction span multiple bilaterian groups. It has been suggested that the atonal homolog lin-32 in C. elegans is not required for the development of mechanosensory cells, thus negating the role of atonal in establishing mechanosensory cells across phyla (Coffin et al., 2004). However, this has been shown to be not the case and lin-32 is expressed in, and necessary for, C. elegans mechanosensory cell formation (Mitani et al., 1993; Zhao and Emmons, 1995; Portman and Emmons, 2000; Du and Chalfie, 2001; Syntichaki and Tavernarakis, 2004).
In addition to atonal genes, Pou domain factors are essential and conserved transcription factors for cellular differentiation, including the hair cells of the ear (O'Brien and Degnan, 2002), which initially form but degenerate in Pou4f3 null mice (Xiang et al., 2003; Hertzano et al., 2004; Pauley et al., 2008). Downstream to Pou4f3 is another gene necessary for hair cell development, Gfi1 (Hertzano et al., 2004). The homolog of Gfi1 in insects, senseless, has been shown to be necessary for the differentiation of mechanosensory cells (reviewed in: Jafar-Nejad and Bellen, 2004).
Both Drosophila and mammalian mechanosensors rely on zinc finger proteins of the GATA binding variety. In Drosophila, the Gata zinc finger pannier has been shown to activate proneural bHLH genes necessary for mechanosensory development. Similarly, a Gata zinc finger protein in mammals, Gata3, has been shown to be necessary for all hair cell development in the inner ear except for those of the saccule (Duncan et al., 2011; Haugas et al., 2012). It remains to be seen if there is an equivalent zinc finger protein in the saccule necessary for hair cell development. The effect Gata3 has on bHLH gene upregulation or function has not been worked out in mouse as pannier has in Drosophila. If it does have an effect on bHLH genes, it would further bolster the connection between Gata3 and pannier as homologs and their role as a crucial node in mechanosensory cell development.
Two other zinc finger transcription factors necessary for mechanosensory cell development in Drosophila are spalt and spalt-related. These genes are coexpressed and function in parallel to atonal and may be required for its upregulation. The vertebrate homologs of these genes the Sall genes are expressed in the developing ear, and while mutations in Sall1 have been associated with inner ear defects, their specific roles are unknown (Kohlhase et al., 1998). Likewise, the homeobox genes of the Iroquois complex are necessary for insect bHLH gene regulation (Garcia-Bellido and de Celis, 2009; Fritzsch et al., 2011) and have been described in the developing ear of Xenopus (Alarcon et al., 2008), but their function has not yet been evaluated due to the complex interaction and redundancy of the six members in vertebrates (Tena et al., 2011).
In addition to transcription factors, many downstream genes are conserved between invertebrates and vertebrates. The Drosophila crinkled gene encodes a homolog of the vertebrate Myosin VIIa gene; an unconventional myosin expressed primarily in sensory hair cells (Hasson et al., 1997). In humans, defects in Myosin VIIa underlie Usher syndrome type 1B, characterized partially by sensorineural deafness. Mutations in Myosin VIIa are also the cause of two additional forms of nonsyndromic deafness, DFNB2 and DFNA11 (Libby and Steel, 2000). Crinkled is expressed in the Drosophila chordotonal organs in addition to Johnston's organ. Mutations within crinkled lead to deafness and ataxia, which is the result of malformed mechanosensory organs (Todi et al., 2004, 2005, 2008). Prestin a specific outer hair cell gene is at least partially responsible for the active mechanical amplification that occurs in the mammalian cochlea. In Drosohpila, an ortholog of this gene is known to be expressed within the mechanosensory cell of the Johnston's organ, but its physiologic role has not been clarified (Weber et al., 2003) and is unlikely to be similar to vertebrates (Okoruwa et al., 2008).
It appears that mechanosensory cell development is governed by a conserved expression of transcription factors that cooperate to ensure complete differentiation and maintenance of such cells. Neither the individual functions nor interactions of transcription factors required to achieve the desired outcome are completely understood (Fritzsch et al., 2006b; Kelley, 2006). The “mechanosensory cell first” hypothesis could be further strengthened by detailed examination of the regulatory relationships of subnetwork genetic interactions occurring in both Drosophila and vertebrate mechanosensory cells, and would be further reinforced by molecular analysis in other bilaterians.
The high similarity within mechanosensory cells is not so obvious for organ development. In mammals, Sox2 regulates, directly or indirectly, expression of Atoh1 as no Atoh1 expression has been reported in extreme Sox2 hypomorphs (Kiernan et al., 2005). How Sox2 expression is regulated, however, remains unclear despite a tremendous understanding of the Sox2 promoter region (Uchikawa et al., 2003). Moreover, in insects there is no evidence for the expression of sox genes in neurosensory precursors (McKimmie et al., 2005), suggesting that Sox2 in the PNS is a vertebrate acquisition that may relate to the clonal expansion of neurosensory precursors required to form large patches of sensory epithelia such as found in the ear (Fritzsch et al., 2007). In contrast, flies seem to use EGFr for a somewhat similar clonal expansion (Caldwell and Eberl, 2002; Eberl and Boekhoff-Falk, 2007). Knowing the specific regulators of Sox2 in vertebrates and EGFr in Drosophila, and their downstream genes, will help understanding if these genes fit within a conserved regulatory network or are novelties in their respective mechanosensory development. Most importantly, the relationship of Eya1 and Pax2/5/8, both early markers of the otic placode (Bouchard et al., 2010), to Sox2 and EGFr expression need to be established for flies and mice to understand similarities and differences.
Coevolution of Hair Cells and Sensory Neurons
As stated earlier the major difference between mechanosensory cells of invertebrates and vertebrates is the splitting of the primary sensory cell into two distinct cell types within the vertebrate lineage. Due to the widespread presence of mechanosensory cells with their own axon among triploblastic and diploblastic organisms, it has been suggested that a single mechanosensory cell with an axon is the ancestral feature (Fig. 3), and the presence of secondary sensory cells with a separate neuron to conduct information likely occurred independently several times (Budelmann, 1992; Mackie and Singla, 2003). While invertebrate primary mechanosensory cells rely on atonal (as discussed above; Fig. 3), vertebrates hair cells rely on Atoh1 and the neuronal cell, which transmits information from the hair cell to the brain relies on Neurogenin1 (Neurog1) a bHLH gene that is very closely related to atonal\Atoh1 (Fritzsch et al., 2010). As previously detailed in (Fritzsch and Beisel, 2004), the transformation of pro-neural clusters, which give rise to a single mechanosensory cell, into clusters that give rise to two separate cell types including sensory neurons, could have been accomplished by an additional round of cell division creating the two distinct cell types and specific bHLH genes to govern the development of each cell type. The most parsimonious explanation is the ancestral mechanosensory cell expressed the ancestral forms of atonal and neurogenin with Drosophila secondarily losing the necessity of neurogenin in mechanosensory cells. However, the Drosophila genome did not lose a functional neurogenin (Tap), but the cis-regulatory elements controlling the expression of ancestral neurogenin have been changed within the invertebrate lineage in order for Tap to be expressed within other sensory systems and the CNS but not mechanosensory organs (Fig. 3). An alteration of downstream cis-regulatory sequences of neurogenin in invertebrates necessary for axonal development would have been altered to bind atonal. In this scenario, cis-regulatory elements downstream of atonal and neurogenin would have remained constant within the vertebrate lineage. In the vertebrate lineage, the cis-regulatory elements controlling expression of Atoh1 and Neurog1 would have been altered to limit their expression to hair cells and sensory neurons, respectively. This scenario is further supported by the fact that Atoh1 is inhibited in inner ear neurons by the transcription factor Neurod1 as recently demonstrated (Jahan et al., 2010b). In the absence of Neurod1, some neurons turn into hair cells indicating the necessity of shutting down Atoh1 expression. In this scenario hair cells and sensory neurons are derived from ciliated primary mechanosensory cells, and suggest a clonal relationship between vertebrate hair cells and sensory neurons (Fritzsch et al., 2000; Hassan and Bellen, 2000; Fritzsch and Beisel, 2004; Raft et al., 2007).
MOLECULAR BASIS FOR THE EVOLUTION OF PLACODES TO FORM ORGANS
If the mechanosensory cell first hypothesis is indeed true, the logical extension is that placodes are not a necessary precursor for mechanosensory (hair) cell formation. The question thus becomes why are hair cells restricted to placodal lineages in vertebrates? An insight may be obtained by reviewing the evolution of the central nervous system (CNS). It may be that the precursor to the CNS was a diffuse basiepithelial nerve net, which evolved to be concentrated in a specific region (Holland, 2003). The similarities between neural plate and placode formation are striking in terms of both molecular and morphologic criteria. This is nowhere more prevalent than in the otic placode in which entire reviews have been dedicated to the subject (Mansour and Schoenwolf, 2005) and thus will only briefly be dealt with here. The embryonic precursor to the CNS, the neural plate, and the embryonic precursor to the inner ear, the otic placode, are a localized thickening of epithelium induced by diffusible morphogens and characterized by increased proliferation. During development, both tissue types undergo folding and pinching off from the overlying ectoderm to become situated under the epithelium. It has been previously proposed that hair cells were most likely single diffusely distributed mechanosensors within the ectoderm of a vertebrate precursor much like single or groups of cells found in the lancelet (Fritzsch et al., 2006a). Thus, the evolution of placodes must be interpreted as an adaptation to ensure development of mechanosensory cells within distinct localities, thus forming aggregated epithelia and consequently distinct organs, paralleling similar processes in the formation of a CNS out of a diffuse nervous system.
In organ evolution, the presence of a particular transcription factor is most likely to be first used for upregulation of genes necessary for differentiation, and only secondarily acquires a function necessary for morphogenesis. An example is Pax6 in eye development and morphogenesis. It is most likely that the original role of Pax6 was as an activator of differentiation genes in visual spots, and because it was expressed at the visual spot location, it was eventually incorporated, independently, in the different programs for eye morphogenesis (Davidson, 2006). This was possible not because the PAX6 protein evolved, but the cis-regulatory sequences in genes necessary for morphogenesis evolved to be inclined to PAX6 binding, and thus upregulated in the location of PAX6 expression. This logic can be expanded to the formation of placodes and the presence of mechanosensory cells within placode-derived organs.
Evidence for this process can be seen in the ear with transcription factors being necessary for both histogenesis of neurosensory cell types and morphogenesis of the ear. The mollusk Pax2/5/8 gene is associated with the mechanosensory containing statocyst (O'Brien and Degnan, 2002), and the inner ear of mammals Pax2 and Pax8 are needed for both hair cell development and morphogenesis. Specifically, the cochlea in the absence of Pax2 abnormally forms as an enlarged sac with no sensory cells and very unusual neurons (Bouchard et al., 2010). Gata3 is another example of a gene that may have been originally necessary for mechanosensory specification and differentiation (as described above) and secondarily coopted for inner ear morphogenesis. Gata3 is necessary in the mouse for full elongation of the cochlea, and formation of vestibular canals (Duncan et al., 2011).
ADDITION OF EAR NEUROSENSORY EPITHELIA IS A RESULT OF DUPLICATION AND DIVERSIFICATION OF EXISTING NEUROSENSORY PATCHES
Although there is much controversy regarding the initial formation of the inner ear, the idea that new neurosensory epithelia form by changes in already present neurosensory epithelia has been suggested for more than a century (Ayers, 1890; Wever, 1972; Duncan and Fritzsch, 2012) and currently has much support. After coopting the developmental gene regulatory network to form hair cells for morphogenesis of the ear there was most likely a primitive ear consisting of a single sensory macula, probably a gravistatic sensor as found in most free swimming animals (Markl, 1974). From this simple ear arose all of the complex three-dimensional labyrinths seen in extant vertebrates today. Looking at extant vertebrates shows a vast array of different numbers of sensory epithelia culminating with the gymnophiona inner ear containing nine sensory epithelia: three semicircular canal crista, utricle, saccule, lagena, basilar papilla, neglected papilla, and amphibian papilla; which all most likely evolved out of a single common macula and two canal cristae in a single torus as seen in the hagfish (Fig. 4). The hagfish ear being the most simple of the vertebrates is most likely closest to the ancestral ear (Fritzsch and Beisel, 2004) but could also be secondarily derived and simplified (see: Northcutt, 1981 for a differing point of view). Each additional sensory patch is formed from the duplication of an existing patch of sensory epithelia and this new epithelia was coopted for a new beneficial use (Fig. 4). This idea implies that during the evolution of the vertebrates there were most likely transient nonfunctional or redundant sensory patches that arose and were subsequently lost as much as gene duplications can lead to pseudogenes through rapid accumulation of mutations. It is only after segregation that sensory epithelia can develop unique molecular properties that allow a novel function. An example of this is the various acellular coverings that facilitate detection of specific stimuli: cupulae (canal crista), otoconia (gravistatic organs), and tectorial membranes (auditory organs).
There are several examples of where new sensory epithelia may have originated. Indeed, the waxing and waning of the neglected papilla presents an excellent example. Analysis of amphibian species that possess an amphibian papilla related to the novel function of sound detection indicate that it is ontogenetically derived from the neglected papilla (Fritzsch and Wake, 1988a).
It has also been suggested that addition of a third semicircular crista in jawed vertebrates decreased the time and processing necessary to detect horizontal angular acceleration, which could be extracted through sophisticated computation from the two crista in agnathans. This third semicircular crista is also associated with a new canal (Fig. 4). In the absence of Otx1, this canal never forms (Morsli et al., 1998). As in hagfish, lampreys have only two canal cristae. Like Otx1 null mice, lampreys lack Otx expression in the ear, suggesting that the entire horizontal canal depends on Otx1 both during development and in evolution. Close examination revealed some displaced patches of hair cells remain in the Otx1 null ear leading to the thought that this remaining hair cell patch is reminiscent of the lamprey dorsal papilla. If this were true, it would indicate a two-step evolution of the horizontal canal system. First, a multiplication of hair cells and a segregation of the dorsal sensory patch would occur. This would have been followed by the evolution of a distinct horizontal canal, for which Otx plays a role. The finding that another gene, Foxg1, is necessary for the normal development of the horizontal canal cristae (Pauley et al., 2006) is in line with this suggestion. This implies that recruitment of two forebrain genes (Otx1 and Foxg1) to the ear provided the molecular driving force to generate a new canal and a new sensory epithelium associated with this new canal. Obviously, other genes must exist that regulate the expression of Otx1 and Foxg1 in the ear and that interact or are regulated by these transcription factors to drive neurosensory development and canal morphogenesis.
Similar developmental segregations of various sensory epithelia have been noticed in amphibians and certain bony fish, and resulted in the proposition that splitting of sensory epithelia is a general mechanism to form new endorgans within the ear that are originally functionally uncommitted and allow for specification in the course of evolution. It has been demonstrated that fiber connections are the easiest way to establish homology across sensory epithelia, even if they are highly modified such as the innervation of the amphibian papilla in modern frogs (Fritzsch, 1992; Fritzsch et al., 2002). In fact, development of the lateral line system shows how such new sensory epithelia patches can form and more recent data suggest some of the molecular mechanisms in this process (Chitnis et al., 2011).
Molecular Basis for the Evolution of Hearing
The segregation of the cochlea of mammals or basilar papilla of other sarcopterygian vertebrates from vestibular organs is a great example of inner ear sensory epithelia addition. It is likely that the basilar papilla\organ of Corti, evolved through embryonic transformation of parts of the saccule as the saccular recess expanded to form an additional recess (Fig. 4). This has been shown developmentally by markers indicating the progressive segregation of the mouse cochlear sensory epithelium from the saccule during development, Lfng and Bdnf (Cole et al., 2000). After the initial evolution of the lagena and its segregation from the saccule, the two epithelia continued to be closely associated with each other, sharing in several species the same otoconial mass (Fritzsch and Wake, 1988b). It is only in those sarcopterygian species where the lagena is situated at the end of its own recess (and therefore further away from the saccule) that a basilar papilla appears (Fig. 4; Fritzsch, 1987). It has been suggested that the prosensory saccular/lagenar anlage extended into a novel recess as the lagenar sensory epithelium (Duncan and Fritzsch, 2012; Fritzsch et al., 2011). This increase of prosensory area combined with formation of a new recess enabled the development of a new endorgan, which subsequently became segregated from both the saccular and lagenar sensory epithelia. It is unclear how the other properties needed for the extraction of sound could have evolved including association with the perilymphatic space that conduct the sound in the vestibular and tympanic scalae from the oval to the round window (Shute and Bellairs, 1953).
After the organ of Corti\basilar papilla became segregated the evolutionary transformation of auditory hair cells from a vestibular to an auditory receptor required changes in molecules to govern the emergence of novel properties. Some of the genes necessary for the segregation of saccular from cochlear epithelia in the mouse are Lmx1a, Wnt5a, and Mycn. In the absence of each of these genes, there is a fusion of the sensory epithelia such that both saccule and cochlear hair cells seem to lie in one continuous recess (Fig. 4). With the deletion of the LIM-homeodomain transcription factor Lmx1a, the mouse inner ear sensory epithelia are conjoined and not separated by non-sensory tissue. Specifically the utricle, saccule, and cochlea are fused into a single patch of hair cells (Nichols et al., 2008). This patch does have molecularly distinct regions as the area, which comprises the organ of Corti and utricular regions, are Gata3 positive with clear delineations from the saccule. However, morphologically the boundaries between regions are blurred. The basal portion of the organ of Corti displays a vestibular like organization with respect to hair cells and supporting cells while remaining normally innervated by the spiral ganglion. The apical region of the organ of Corti displays only minor disruption in the normal organization of cochlear hair cells and supporting cells. Indicating that after segregation many other molecular changes are necessary for the new epithelia to become unique and acquire a novel function. Wnt5a null mice also show a fusion of the utricle and saccule. There is an alteration of the number of rows of outer hair cells, but inner hair cells and outer hair cells are distinguishable from each other (Qian et al., 2007).
The basilar papilla, if present in sarcopterygians, is within a recess shared with the lagenar epithelium located at the ventral end of the basilar epithelium. In monotremes the basilar papilla is elongated and has a slight curve; the lagena continues to be located just beyond the proximal tip of the basilar papilla (Fig. 4). It is only within therians that the cochlear duct is substantially longer and coils. Correlated to this cochlear duct elongation is the loss of the lagenar epithelium. It is possible that the cochlear epithelium expanded into the lagenar epithelium during its elongation, and there was a fusion of the two epithelia. Consistent with this idea Mycn null mutants have a peculiar cochlear epithelium. At the apex of this epithelium is a ball of hair cells, which create circularized sensory epithelia (Kopecky et al., 2011). This would be reasonable given that the elongation of the basilar papilla into the cochlear duct requires additional rounds of cell division to create the added cells. With Mycn being a positive controller of cell cycle, it is possible with fewer rounds of cell division the expansion of cochlear precursors into the lagena does not occur leading to a respecified unique epithelium.
Functional constraints resulted in divergence between auditory and vestibular hair cells and they are morphologically different. One of the pathways that have been suggested in separation of these two cell types is Wnt signaling, which has been shown to be necessary for the development of the vestibular hair cell phenotype. When constitutively active β-catenin (part of the Wnt pathway) was expressed in the chicken inner ear, hair cells, and supporting cells in the auditory epithelia developed characteristics consistent with a vestibular phenotype (Stevens et al., 2003). It may be the case that the evolutionarily young auditory epithelium of tetrapods down regulated the Wnt pathway and in doing so shed some of its vestibular molecular burden, allowing for a new phenotype.
While all hair cells, including both vestibular and cochlear, require the bHLH transcription factor Atoh1 very few distinguishing markers are known for individual hair cell types. Fgf8 is known to be necessary for mouse and chick otic induction (Ladher et al., 2005). During later development, the inner hair cell row of cochlear hair cells is known to express Fgf8. Its deletion or overexpression does not influence hair cell number but rather other cell types of the organ of Corti, the inner and outer pillar cells (Jacques et al., 2007). Nevertheless, Fgf8 is a good marker of inner hair cells, but also of a subset of hair cells in gravistatic organs. One FGF receptor Fgfr1 has been shown to have specific effects on the auditory epithelia of mice (Pirvola et al., 2002). With the full null and conditional deletion with the ear, specific Foxg1cre there is a loss of hair cells specifically within the organ of Corti while the vestibular system remained normal. However, misexpression of Fgf8 is compatible with later cell phenotype change (Jahan et al., 2010a). In addition, specific to the cochlea seems to be morphogenetic gradients necessary for the correct positioning and development of the sensory epithelia within the cochlear duct. As previously suggested (Morsli et al., 1998; Pauley et al., 2003), counter gradients of Fgfs and BMPs seem to provide this information (Groves and Fekete, 2012). BMP signaling defines the non-neuronal boundary of the sensory epithelia, and upregulates genes necessary for outer sulcus fate. Moderate amounts of BMP signaling can increase hair cell numbers and suggests that sensory progenitors require an intermediate amount of BMP signaling to form (Ohyama et al., 2010). Fgfs are thought to mark the neuronal boundary of the cochlear sensory epithelia. Fgf10 is expressed asymmetrically on the neuronal side of the cochlea; however, its elimination alone is not enough to elicit effects (Pauley et al., 2003). This may be due to redundant signaling by other fgfs in the cochlea, possibly Fgf20 (Huh et al., 2012). How these counter gradients are initiated, how far their molecules diffuse, and how they are interpreted is currently unknown. Further research into these diffusible gradients will need to be done to understand fully the patterning of the cochlea.
Wnt signaling may also be involved in the transformation of an otoconia-based structure covering the hair cells to a tectorial membrane. Many proteins necessary for the formation and integrity of both types of matrices are shared between the two structures. However, minute alterations, perhaps mediated by the absence of Wnt induced differentiation of the sensory epithelium, could be involved in turning the otoconia covering into a tectorial membrane; otoconia being the major necessary component for the saccule to receive transduction by force of gravity, or other linear acceleration, and allowing a switch from gravitational input to the reception of auditory stimuli. Subsequent changes adapted the cochlea to a high frequency range characteristic of mammals (Manley, 2010).
Auditory Neurons are Evolutionary Derived from Vestibular Neurons
Just as auditory sensory epithelium is ontogenetically derived from the saccular epithelia so too are the auditory neurons derived from vestibular neurons. To become distinct, the newly derived auditory neurons must reach auditory rather than vestibular epithelia at the periphery, and project to auditory rather than vestibular nuclei in the central nervous system. Both auditory and vestibular neurons show dependence on both Neurog1 and Neurod1, and the absence of either result in loss of all or most vestibular and auditory afferent neurons (Ma et al., 2000; Kim et al., 2001; Jahan et al., 2010a). However, there is a preferential loss of auditory neurons compared to vestibular neurons in the Neurod1 null mouse. One gene shown to be preferentially expressed in the auditory neurons is Gata3 (Karis et al., 2001; Lu et al., 2011). Gata3 has been shown to modulate the expression of Neurod1 and thus may be linked to the preferential reliance of auditory neurons on this transcription factor (Duncan et al., 2011). Gata3 null mice lack all auditory neurons, but retain some vestibular neurons. One of the two fly orthologs for Gata3, grain, has also been shown to be necessary for specification and pathfinding properties for subsets of neurons (Garces and Thor, 2006). Because of the lack of auditory neurons in the mouse, conditional mutants for Gata3 in spiral ganglia are needed to assess its later possible role in spiral neuron pathfinding and identity.
Spiral and vestibular ganglion neurons may originate from distinct areas of the otocyst (Fritzsch, 2003; Yang et al., 2011), and this development from exclusive areas may make it easier to bestow unique identities to the vestibular and spiral ganglion neuron populations. It is also possible that neurons easily find their target because they project back to where they delaminated from an idea based upon the developmental and ontological relationship between hair cells and sensory neurons described above. A developmental lineage relationship has been established for the chicken, but it remains unclear if this has a direct effect on neuronal path finding (Satoh and Fekete, 2005; Bell et al., 2008). However, in the absence of hair cells there seems to be an initial outgrowth of fibers to the prosensory area signifying that hair cells are not necessary for initial projection of fibers (Pan et al., 2011). Clearly, more work is needed to establish the degree to which sensory neurons derive from sensory epithelia they are projecting to and require the sensory epithelia to guide initial outgrowth.
Differential neurotrophin expression could explain the specific projection of inner ear neurons. All vestibular and cochlear sensory neurons express both Ntrk2 and Ntrk3 (previously TrkB and TrkC); however, each cell type seems to be uniquely reliant on a particular receptor/ligand combination; vestibular innervation is more dependent on Bdnf and its receptor Ntrk2, and cochlear innervation is more dependent on Ntf3 and its receptor Ntrk3. In contrast, differential innervation of a given epithelium is not dependent on receptor or ligand specificity but rather on spatiotemporal expression of a specific neurotrophin resulting in differential elimination of basal or apical sensory neurons or canal cristae innervation (Farinas et al., 2001; Tessarollo et al., 2004). The high level of Ntf3 dependency of spiral ganglion neurons may be an additional reinforcement to project selectively to cochlear nuclei as indicated by recent data suggesting dependency of spiral ganglion neurons on cochlear nucleus derived neurotrophins (Maricich et al., 2009).
Centrally the auditory and vestibular fibers must also segregate from each other and project distinctly into the hindbrain, reaching their own information processing areas. Vestibular fibers can be rerouted to the cochlea (Tessarollo et al., 2004) and cochlear and vestibular fibers may be incompletely segregated both peripherally and centrally in certain mutants (Jahan et al., 2010a; Kopecky et al., 2011), indicating an incomplete segregation of selection of targets in these mutants. In particular, Neurod1 may play a role in the central projection of the auditory neurons. In its absence, the cochlear projection into the hindbrain becomes disrupted and fibers do not project properly to their targets but rather show an overlapping vestibular/cochlear nucleus projection (Jahan et al., 2010a).
ADDITION OF HINDBRAIN NUCLEI
Some have speculated a transformation occurred of the lateral line nuclei into auditory nuclei with the introduction of terrestrial hearing (Larsell, 1967). To look more closely at this we must shed light on the two systems, which make up the lateral line: the ampullary electroreceptive and neuromast mechanosensitive lateral line systems. Although these two systems are distinct in their cell types and function, they have been shown to originate from the same embryonic placode (Northcutt et al., 1995). During development, many species of amphibians will lose either or both of these systems (Fig. 1). The electroreceptive neuromasts are thought to have arisen twice during evolution (Bullock and Heiligenberg, 1986), thus its presence or absence is malleable. Only anurans and other vertebrates, which never develop electroreceptive ampullary organs, form a distinct auditory nucleus (Fig. 4; Will and Fritzsch, 1988; Fritzsch, 1999). Thus the simultaneous presence of both the lateral line mechanosensory nucleus and inner ear auditory nucleus invalidates the theory that with the loss of the lateral line and input into the mechanosensory lateral line nucleus the newly developed auditory fibers innervated this recently vacated area. Because of the coincidental loss of electroreception and the gain of an auditory nucleus, there are two possible scenarios, which are not mutually exclusive. In one scenario, some neurons, which previously contributed to the electroreceptive nucleus, were respecified and contribute instead to the auditory nucleus (Fritzsch, 1992, 1999). Another possibility is that other cell sources because of increased proliferation during development gave rise to the auditory nucleus, possibly those of the vestibular nuclei, which underwent functional respecification (Fritzsch, 1992, 1999; Grothe et al., 2004).
To assess the origin of the auditory nucleus data must be gleaned from comparative developmental biology. Reports are now becoming evident on the location of cochlear nucleus precursor populations in the mouse (Farago et al., 2006). This data has shown that the cochlear nucleus develops from multiple populations that are both genetically and physically different from one another. While this data has made major inroads into mammalian cochlear nucleus development and provides the tools to better characterize its developmental gene regulatory network, comparable molecular data from other vertebrates is lacking. Morphological data from chicken indicate that auditory precursor cells come from slightly different rhombomeric regions (Marin and Puelles, 1995; Cambronero and Puelles, 2000; Cramer et al., 2000). However, it remains to be seen if these differences are only superficial as found with other chick-mouse rhombomeric comparisons, or if there are true molecular differences. Only when the molecular data from several vertebrates is obtained, will we begin to have a clear picture on the ontogenetic origins of the auditory nucleus, solving relationships of the dorsolateral nucleus (DLN) of amphibians with the auditory nuclei of amniotes.
It is unclear in either scenario what directs the three organs of the combined inner ear and lateral line to project in a segregated, nonoverlapping fashion all within the alar plate of the hindbrain, and specifically what directs the auditory projection to end specifically on the new or respecified cells of the auditory nucleus and not the vestibular nucleus or mechanosensory lateral line nucleus. Evidence for genetic and biochemical specificity are abundant in the visual system and spinal cord. Particularly interesting are the sensory axons projecting to distinct types of motor neurons within the spinal cord. When the motor neuron identity is lost or altered sensory axons project to their normal dorsoventral tiers, indicating that these sensory cells are projecting to distinct areas instead of being allured by specific cell types (Sürmeli et al., 2011). However, when Atoh1 is deleted from the cochlear nucleus, spiral ganglion fibers show abnormal projections within the nucleus (Maricich et al., 2009). In addition, (Maricich et al., 2009) showed that some spiral ganglion fibers are dependent upon cochlear nucleus neurons for viability, indicating that within a particular nucleus chemical cues are necessary for guidance and viability.
We have critically evaluated numerous hypothesis proposed to explain the evolution of mechanosensory cells, the inner ear, and the auditory system. The strongest molecular evidence in line with emerging general principles of evolution such as gene duplication followed by segregation and diversification indicates that all mechanosensory cells derived from a single ancestor and coalesced into the placode developmental precursor of the ear through mechanisms comparable to those aggregating the diffuse neuronal network into the central nervous system. Once formed, the ear expanded through increase of sensory epithelia followed by splitting and functional reassignment to generate the sets of angular acceleration, linear/gravistatic acceleration and auditory sensation specific sensory epithelia through added changes of hair cells and supporting structures such as scalae and tectorial membrane/otoconia/cupula. Accompanying segregation and functional diversification of sensory epithelia is a functional segregation of sensory neurons and diversification and functional segregation in brain stem nuclei. Initial steps to characterize the molecular alterations in hair cell, sensory epithelium, sensory neuron, and sensory nuclei diversification are presented, but more data is needed to consolidate the presented evolutionary scenario into a systematic progression at all levels.
The authors wish to thank Karen Elliott for feedback on a pervious draft of the manuscript and Benjamin Kopecky for providing a copy of his data.