Cochlear hair cell regeneration after noise-induced hearing loss: does regeneration follow development?

Abstract

Noise-induced hearing loss (NIHL) affects a large number of military personnel and civilians. Regenerating inner-ear cochlear hair cells (HCs) is a promising strategy to restore hearing after NIHL. In this review, we first summarize recent transcriptome profile analysis of zebrafish lateral lines and chick utricles where spontaneous HC regeneration occurs after HC damage. We then discuss recent studies in other mammalian regenerative systems such as pancreas, heart and central nervous system. Both spontaneous and forced HC regeneration occurs in mammalian cochleae in vivo involving proliferation and direct lineage conversion. However, both processes are inefficient and incomplete, and decline with age. For direct lineage conversion in vivo in cochleae and in other systems, further improvement requires multiple factors, including transcription, epigenetic and trophic factors, with appropriate stoichiometry in appropriate architectural niche. Increasing evidence from other systems indicates that the molecular paths of direct lineage conversion may be different from those of normal developmental lineages. We therefore hypothesize that HC regeneration does not have to follow HC development and that epigenetic memory of supporting cells influences the HC regeneration, which may be a key to successful cochlear HC regeneration. Finally, we discuss recent efforts in viral gene therapy and drug discovery for HC regeneration. We hope that combination therapy targeting multiple factors and epigenetic signaling pathways will provide promising avenues for HC regeneration in humans with NIHL and other types of hearing loss.

1. Noise-induced hearing loss and its military relevance

Hearing impairments, including hearing loss and tinnitus, are the most common military service-connected disabilities. There are many sources of potentially damaging noise in military settings, including weapons systems, engines, and explosions. The prevalence of hearing problems in the US military operation in Afghanistan and Iraqi ranged from 7.3% to 26.6% among more than 90,000 veterans who received the Veterans Health Administration health care(Frayne et al., 2011). Hearing loss significantly affects performance, and is one of the top reasons why soldiers cannot be redeployed (Casali, 2010; Hawkins et al., 1980). Given the importance of auditory acuity in perceiving commands and sensing enemy activity, it is clear that even mild hearing loss increases the risk to military personnel. Service-related injuries that result in hearing loss have decreased with the advent of hearing protection, but many military personnel still develop NIHL by the end of their tour. Hearing loss that restricts or prevents continued military service can result in the loss of personnel, including some of the most effective and experienced officers and non-commissioned officers. Furthermore, NIHL is irreversible, resulting in a permanent disability. As a result, 4.1 million US veterans currently receive disability compensation and treatment for service-connected hearing disability at a total cost of more than $2 billion annually (US Department of Veterans Affairs, 2017, Page 53). Among those veterans who suffer from NIHL, a resulting breakdown in communication with caregivers and family members can impede their reintegration into society and exacerbate their depression and anxiety. In view of these consequences, the US Department of the Navy has recently declared “The Naval Global War on Noise” (Naval Safety Center, 2007).

One common but extreme cause of NIHL and tinnitus is explosions. The blast injuries occur frequently in situations where blast exposure cannot be predicted, where the intensity of the blast exceeds the protective capability of the available mechanical protective devices, or where protective devices are simply not available (Arnold et al., 2004; Arnold et al., 2003; Patterson et al., 1997). In wars or terrorist attacks, such blasts are particularly damaging to both military personnel and civilians. In addition to damaging the central nervous system (CNS), blast injuries most commonly cause disruption or damage to the auditory sensory system. The damage to the peripheral auditory sensory system due to the extreme physical force of a blast is often diverse and may include rupture of the tympanic membrane (TM, ear drum), fracture of the middle ear bones, dislocation of sensory hair cells from the basilar membrane, and loss of spiral ganglia that innervate hair cells (Hamernik et al., 1987; Patterson et al., 1997; Roberto et al., 1989). In studies of humans with blast injury, approximately 17% to 35% were afflicted with severe TM rupture (Garth, 1994; Gondusky et al., 2005; Xydakis et al., 2007), and 33% to 78% exhibited moderate to severe sensorineural hearing loss [hair cell (HC) and ganglion loss] as a result of exposure to terrorist or military explosions (Cave et al., 2007; Fausti et al., 2009; Gondusky et al., 2005; Hoffer et al., 2010; Nageris et al., 2008).

Regenerative therapies may provide a means of hearing restoration to allow individuals with NIHL (and other types of hearing loss) to continue in military service, thereby reducing the attrition of the armed forces. This is particularly important when blast injury or post-traumatic stress disorder is associated with the hearing loss.

In this review, we will first discuss the lessons from recent regeneration studies in non-mammalian vertebrates (zebrafish and chicken) and other mammalian regenerative tissues (pancreas, heart and CNS), and then focus on the recent progresses on cochlear HC regeneration in vivo, specifically highlighting the differences between SC-to-HC direct lineage conversion and HC development, and essential roles of epigenetics in HC regeneration. Finally, we will summarize recent development in gene therapy and drug delivery for regenerative purposes.

2. Regeneration in non-mammalian vertebrates and other mammalian tissues

NIHL is primarily caused by damage to mechanosensory HCs in the cochlea (Hamernik et al., 1989; Johnsson et al., 1976; Liberman et al., 1984; Wang et al., 2002). The cochlea is a coiled structure in the inner ear that contains the auditory epithelium known as the organ of Corti. The organ of Corti contains HCs, which detect incoming sound, and supporting cells (SCs) (Fig. 1). HCs are subdivided into inner hair cells (IHCs), which are responsible for transmitting detected sound waves to auditory regions of the brain, and outer hair cells (OHCs), which are responsible for amplifying the sound (Dallos, 2008; Dallos et al., 2008). Only mammals have OHCs, which are unique because they primarily receive efferent signaling from the brain (Dallos, 2008) and express the motor protein prestin, allowing them to be electromotile and resulting in their ability to modulate sound (Dallos, 2008). NIHL results in the preferential loss of OHCs; IHCs are lost only in severe cases of NIHL (Hamernik et al., 1989).

Fig. 1.

Diagram (cross section) of the mature cochlear organ of Corti, illustrating the major sensory and non-sensory cell types. IHC: inner hair cell; OHC: outer hair cell; IPhC: inner phalangeal cell; IBC: inner border cell; PC: pillar cell; DC: Deiters’ cell. Lgr5+ cells at neonatal age are highlighted with pink color.

Unfortunately, humans and other mature mammals cannot replace lost HCs in the organ of Corti, so the loss of HCs can lead to permanent deafness. However, birds, fish, and other non-mammals can recover lost hearing either by proliferating SCs and then converting them into functional HCs or by directly transdifferentiating SCs to HCs (Adler et al., 1996; Baird et al., 1996; Corwin et al., 1988; Jones et al., 1996; Ryals et al., 1988; Warchol et al., 1996). Interestingly, mature mammalian vestibular end organs, such as utricle, exhibit limited HC regeneration after damage, which occurs almost exclusively by transdifferentiation (Golub et al., 2012; Jung et al., 2013; Slowik et al., 2013). Notably, the regeneration of vestibular HCs declines with age (especially in the early developmental stage) and is not sufficient to compensate HC loss (Burns et al., 2012; Gopen et al., 2003; Kirkegaard et al., 2005; Rauch et al., 2001). Nevertheless, these natural processes provide a potential roadmap for mammalian cochlear HC regeneration.

2.1. HC regeneration in non-mammalian vertebrates

Fish, birds, and other nonmammalian vertebrates have the remarkable ability to regenerate cochlear HCs after damage. Zebrafish regenerate lost HCs almost exclusively from mitotic SCs (Ma et al., 2008); in birds, cochlear HC death induces adjacent SCs to proliferate and transdifferentiate into new HCs (Baird et al., 1996; Girod et al., 1989; Raphael, 1992). A detailed assessment of HC regeneration in zebrafish and chickens has been performed using cell- or organ-specific transcriptome profile analysis (Jiang et al., 2014; Steiner et al., 2014), with the aim of explaining why mammalian cochleae lack such regenerative ability. Activation of conserved signaling pathways involved in the spontaneous regeneration could be a key to reprogram the quiescent SCs in mature mammalian cochleae.

2.1.1 The zebrafish lateral line

Two recent studies using genome-wide transcriptome analysis of presumptive HC progenitor cells in the zebrafish lateral line have demonstrated that multiple signaling pathways are necessary to activate the regenerative responses after HC injury in this species (Jiang et al., 2014; Steiner et al., 2014). Here, the zebrafish lateral line makes the high temporal resolution analysis possible because HCs can be killed within 30 min by copper sulfate or neomycin and regeneration occurs rapidly within 24 h. Analysis at multiple time points after HC injury indicated that the correct order of multiple signaling pathways is crucial. For example, Notch signaling and FGF signaling were downregulated immediately after HC loss (within 1 h), whereas cell cycle and Jak1/Stat3 pathways were activated at the same time. In contrast, Wnt/β-catenin was activated much later (at 12 h) (Jiang et al., 2014). It suggests that in the early stage of spontaneous regeneration in zebrafish, the immediate inhibition of Notch signaling downregulates Sox2 and activates Atoh1 to enable SCs for further proliferation and/or differentiation. However, because Notch needs to be reinstated later to specify the fate of SCs and control the organ size, it is reasonable that Notch target genes (i.e., sox2, her4, and notch3) were later upregulated at 3 to 5 h, consistent with the findings of several previous studies at later time points in zebrafish, guinea pigs, and chickens (Hori et al., 2007; Ma et al., 2008; Stone et al., 1999). At the same time, FGF and Jak1/Stat3 pathways may contribute to the early proliferation status as Wnt/β-catenin pathway-regulated proliferation likely occurs in the later stage. In a following study, the same group clarified spatially restricted Notch and Wnt pathways in the zebrafish lateral line -- in distinct tissue compartments, Notch signaling inhibits proliferation through Wnt-dependent and -independent pathways and suppresses differentiation via other, unidentified mechanisms (Romero-Carvajal et al., 2015). According to these results, the authors speculated that the lack of immediate and transient Notch downregulation is the key difference between mammalian cochleae and zebrafish lateral lines with regard to HC regeneration and that the temporal inhibition of Notch would be fruitful for mammalian HC regeneration (Jiang et al., 2014). Indeed, genetic (e.g., by knocking out one core component of Notch signaling pathway, RBP-J) or pharmaceutical (e.g., by using γ-secretase inhibitors) suppressing Notch signal in normal or damaged cochleae induced transdifferentiation from SCs (excluding pillar cells) to HCs (Doetzlhofer et al., 2009; Korrapati et al., 2013; Mizutari et al., 2013; Yamamoto et al., 2006). However, these studies only agree on the effect of Notch in neonatal mice, while HC conversion in adult cochleae still needs more evidence. In contrast to what Mizutari et al (Mizutari et al., 2013) reported (a γ-secretase inhibitor at extremely high doses produced new hair cells and partially restored hearing), Maass et al (Maass et al., 2015) thoroughly investigated the mRNA level of Notch ligands, receptors, and downstream effectors, and found dramatic downregulation of Notch signaling after postnatal day 6. The contradictory findings may result from non-specific effect of the drug used by Mizutari et al. Alternatively, the mouse has only one allele of endogenous Sox2 gene and the haploinsufficiency of Sox2 may genetically interact with Notch signaling (Li et al., 2012; Walters et al., 2015b). Moreover, the ablation of Notch signal is deleterious to differentiating Deiters’ cells (Campbell et al., 2016), and this toxic effect must be considered before any clinical application.

Besides these well-characterized pathways, several less characterized signaling pathways have been implicated in mantle-cell regenerative responses; for example, atg2bl, a putative mediator of autophagy, is upregulated (Steiner et al., 2014). However, further investigation is required to validate their roles in HC regeneration.

2.1.2. The chick utricle

Avian sensory epithelia provide enriched information sources for HC regeneration. The chick utricle culture is believed to resemble the utricle in vivo as regards to its regenerative responses and thus provide important insights into signaling pathways during HC regeneration. After an initial report in which transcriptome in chick utricle cultures were analyzed during the initial 48 h after laser or neomycin damage (Hawkins et al., 2007), Lovett and colleagues further investigated the gene expression profiles using RNA-seq in chick utricle cultures up to 168 h after a one-day aminoglycoside antibiotic treatment (Ku et al., 2014). They uncovered three sequential but overlapping phases of HC regeneration: DNA replication/cell cycle control; conversion; and regenerative proliferation. As expected, the Notch and FGF signaling pathways play critical but complex roles in these time windows. Specifically, Notch signaling was elevated after the damage possibly as a consequence of upregulation of Atoh1, followed by a transient inhibition at 48–72 hour time window when DNA replication signal and the SC-to-HC conversion are at the peak. Meanwhile, the FGF signaling is more complicated – different FGF family members showed variable expression patterns during the regeneration process. With additional experiments (adding exogenous FGF20 or a FGFR inhibitor), the authors concluded that the FGF signaling (especially FGF20) was a negative mediator for regenerative proliferation. Indeed, FGF signaling function depends on cellular context – although mostly demonstrated for their roles in promoting proliferation (e.g. (De Moerlooze et al., 2000)), the FGFR3 signal drives differentiation while suppresses proliferation in bone development (Colvin et al., 1996). Notably, in early developmental stage (i.e. E10.5 – E12.5) of the murine cochlea, FGF20 is required for sensory progenitor cell proliferation (Huh et al., 2015).

Interestingly, some novel regeneration regulator candidates were identified by examining the expression pattern of transcription factor (TF) genes. In the total of 212 differentially expressed TF genes, 8 TF genes including MAMLD1 and RBPJL (which are activated by the Notch signal and increase Hes genes while repress Atoh1 ) were upregulated during the 48- to 72-h window in which phenotypic conversion dominates, while 18 TF genes including BTG1 (a gene that functions as a coactivator of cell differentiation and a suppressor of proliferation) were downregulated at the same time window. Similar to the transcriptome studies in zebrafish, a number of unexpected pathways were also identified, including cytokines, interleukins, and interleukin receptors (Ku et al., 2014). Given the recent findings on the importance of fractalkine signaling and macrophages in the mammalian inner ear (Kaur et al., 2015) and the discovery of complement components associated with innate immune responses after noise trauma (Patel et al., 2013), resident immune cells within the mammalian inner ear may play roles in regenerative responses.

As mentioned earlier, many pathways identified in the zebrafish lateral line and the chick utricle have been validated in mammalian cochleae. For example, the inhibition of Notch signaling initiated HC regeneration in mammalian cochleae (Korrapati et al., 2013; Mizutari et al., 2013; Tona et al., 2014), while enforced Wnt/β-catenin signaling increased the proliferative responses (Chai et al., 2012; Kuo et al., 2015; Shi et al., 2013). However, signaling pathways identified by these transcriptome studies may be limited by the usage of a mixed cell population. The two studies in zebrafish showed varied results due to their using different sources for the purified SCs, i.e., mantle cells and inner SCs (Jiang et al., 2014) versus mantle cells alone (Steiner et al., 2014). These two cell types play different roles during lateral line regeneration (Romero-Carvajal et al., 2015). Similarly, the analysis using the entire chicken utricle culture also diluted the HC conversion signals in a small population of cells with the dominant proliferation signals (Ku et al., 2014). Moreover, it remains to be determined whether mantle cells and inner SCs in zebrafish and SCs in avian utricles are indeed the equivalent of SCs or progenitors in the mammalian inner ear. The transcriptome studies on zebrafish inner ear and chick basilar papilla, as well as single-cell RNA-seq could be necessary to address these questions. In addition, forcedly regenerated HCs in mammalian cochleae are far away from native HCs or spontaneously regenerated HCs in fish and chicken in term of morphology and function. These limitations may result from the loss of stemness for mature SCs or the epigenetic barriers. Moreover, studies using different genetic models raised controversy (e.g. different effects of the Notch inhibitor, proliferation vs. differentiation induced by the Wnt signal). We will discuss these in sections 3 and 4.

2.2. Regeneration in other mammalian systems

Cardiac, pancreatic, and CNS regeneration have been well studied (Fig. 2A). Here, we will outline similarities among these regenerative systems with the aim of deriving principles to guide our efforts in the field of cochlear HC regeneration.

Fig. 2.

Examples of direct lineage conversion in other mammalian regenerative systems (A) and in the cochlea (B). The original differentiated cells (at left) were converted to another differentiated cell type (at right). The lineage-specific transcription factors are listed for each conversion.

2.2.1. The heart

In spite of different origins from the cochlea (endoderm vs. ectoderm), the mammalian heart presents several similarities during regeneration, including inefficient cell reprograming, age-dependence, immature/nonfunctional cells, etc. After injury, the neonatal mouse cardiomyocytes have already exited the cell cycle normally and exhibit little proliferation, but they start to proliferate after injury (Porrello et al., 2011). In the adult heart, only very limited turnover of cardiomyocytes (1% per year) can be detected (Bergmann et al., 2009). A massive loss of cardiomyocytes due to an infarction cannot be spontaneously repaired; instead, a scar is formed to prevent further damage (Sutton et al., 2000).

Because of the limited regenerative capacity of the heart, several approaches were used to regenerate cardiomyocytes, including transplantation of cells differentiated from induced pluripotent stem cells (iPSCs) or adult stem cell therapy, and direct conversion of other cell types in the heart. The latter strategy may avoid complications of cell transplantation including failure to engraft appropriately, short survival period, and cardiac arrhythmias (Lambers et al., 2016). In fact, several groups have succeeded in converting cardiac fibroblasts to functional cardiomyocytes with an efficiency of 7% to 20% by using a combination of multiple cardiac-lineage TFs (Gata4, Hand2, Mef2c, and Tbx5 or GHMT) (Qian et al., 2012; Song et al., 2012)(Fig. 2A). Recently, progress has been made toward increasing the conversion efficiency three to five folds (Wang et al., 2015) by using polycistronic vectors, indicating the stoichiometry of the reprogramming TFs is essential for the conversion efficiency. Epigenetic factors also play important roles in cardiac regeneration. BAF60c, a chromatin remodeler, when combined with TFs (Gata4 and Tbx5), transdifferentiates mouse mesoderm cells into cardiac myocytes (Takeuchi et al., 2009); microRNAs (miRNA-1, -133, -208, and -499) convert fibroblasts to cardiomyocytes, and the conversion efficiency has been increased to 10-fold by a JAK inhibitor (Jayawardena et al., 2012).

Notably, although iPSC differentiation and direct conversion are promising, the resulting cardiomyocytes, either in vitro or in vivo, still possess immature features of embryonic cardiomyocytes, namely higher proliferation rates, smaller and more rounded morphology lacking t-tubules and binucleation, and abnormal potassium currents that are responsible for arrhythmias (Lambers et al., 2016).

2.2.2. The pancreas

Pancreatic β-cells secrete insulin and play a central role in controlling blood glucose levels and metabolism. Direct lineage conversion from non-β-cells in situ has produced results of considerable interest to researchers investigating cochlear HC regeneration: the regeneration efficiency can be determined by reprograming resources, combined TFs, age, and injury-induced signals. However, full functional restoration of a mechanosensory cell, whose structure is highly-ordered, may be different from that of an insulin secreting cell.

The efficiency of conversion in vitro and in vivo depends on the starting cell types and the TFs used. One TF, Pdx1, could convert 0.3% to 6% of transduced hepatocytes (which start with the same endodermal lineage) to insulin secreting cells (Ferber et al., 2000), but adding other TFs (NeuroD, Ngn3/betacellulin, Mafa, and Pax4) substantially enhanced the conversion efficiency (Berneman-Zeitouni et al., 2014; Yang et al., 2013). In contrast, pancreatic non-β-cells, like acinar cells, are closer to the β-cell lineage and more advantageous to be converted. A set of three TFs (Ngn3, Pdx1, and Mafa) has been effective for converting non-β-cells to β-cells and improving insulin secretion in the adult pancreas, a process that occurs rapidly within the first 10 days with an efficiency of 10% to 20% (Zhou et al., 2008) (Fig. 2A). Using a polycistronic vector to control the stoichiometry, the conversion efficiency has been increased to 50% and conversion occurs rapidly within the first 10 days and gradually progresses for more than 7 months at the global transcriptome and DNA methylation levels so that the converted cells begin to resemble endogenous β-cells (Li et al., 2014). Interestingly, even at 7 months, converted β-cell–like cells, though fully functional in normalizing glucose levels in diabetic mice, remain hybrid-like cells that are part-way between acinar and β-cells, expressing only 70% of β-cell–specific genes and still retaining a significant number (> 1,209) of acinar cell–specific genes. Another interesting feature of this in vivo conversion is that islet-like structures surrounding newly converted β-cells play a key role in their continued maturation; the ability to form such islet-like structures is lost when native β-cells are dissociated, thereby losing their function, and can be facilitated by inter-islet signaling pathways (i.e., Eph/ephrin). Additionally, treating adult mice with EGF and CNTF led to the appearance of β-cell–like cells in diabetic mice, suggesting that these factors could play a role in reprogramming and/or maturation during β-cell conversion, at least from acinar cells (Baeyens et al., 2014).

Another aspect of β-cell conversion from pancreatic alpha cells is the dependency of the process on age and the amount of damage incurred. A severe loss of β-cells could stimulate the conversion of alpha cells to β-cells -- the near complete loss of β-cells caused by genetic expression of diphtheria toxin fragment A (DTA) triggered proliferation and co-expression of glucagon and insulin in alpha cells. The mass and the hybrid feature of these islet cells maintained for more than 10 months (Thorel et al., 2010). However, β-cell loss caused by streptozocin expression invoked no such conversion (Cavelti-Weder et al., 2013), suggesting that the mode of β-cell death could determine whether the conversion of alpha cells to β-cells is triggered. Interestingly, DTA-induced extreme β-cell loss in juveniles before puberty invoked a different type of conversion: from delta cells to β-cells (Chera et al., 2014). It would be interesting to identify the signals that trigger the cell conversion in these different β-cell loss scenarios.

2.2.3. The CNS

The CNS is one of the most difficult systems to regenerate because the vast majority of cells in the adult brain are postmitotic and the cells are extremely diverse, with only a small number of each cell type being present. These two features are shared by the cochlea. Moreover, the neurons and the hair cells are both derived from ectoderm, so the regeneration may also share similar molecular mechanisms.

As in the pancreas and heart, iPSCs, embryonic stem cells (ESCs), and adult stem cells (i.e., mesenchymal stem cells) can differentiate into neurons in vitro; however, the drawbacks of stem cell–based therapy include its potential tumorigenicity, the difficulty of generating diverse neuronal cell types, and the lack of full functionality in the induced neurons (Huang et al., 2015; Petersen et al., 2016). The direct conversion of glia/astrocytes to neuron-like cells in situ offers an alternative and promising strategy. This strategy is largely based on the successful conversion of fibroblasts into neuron-like cells in vitro: the most prominent set of TFs for converting fibroblasts into neurons is designated BAM (Brn2, Ascl1, and Myt1l), and the conversion efficiency in vitro is approximately 20% from mouse cells (Vierbuchen et al., 2010). When applied to human cells, although BAM efficiently promoted ESC differentiation, they only convert very small population of human postnatal fibroblasts (HPFs) to immature neurons. Addition of another TF, NeuroD, however, doubled the conversion efficiency (to 2–4%) to neurons (Pang et al., 2011). Interestingly, although the conversion with different factors from different original cells has various efficiencies, additional factors, especially epigenetic factors or small molecules targeting certain pathways, in general achieve higher efficiency. For example, miR-9/9* and miR-124 suppress the BAF53a subunit of the SWI/SNF-like BAF chromatin remodeling complex and control multiple genes regulating neuronal differentiation (Yoo et al., 2011). Thus, addition of miR-9/9* and miR-124 to TF Ascl1, Myt1l and NeuroD1 increased the conversion efficiency from human fibroblasts to neurons to 80% (Yoo et al., 2011); while applying a GSK-3β inhibitor and a SMAD pathway inhibitor together with TFs Ascl1 and Ngn2 successfully converted about 85% HPFs to neurons (Ladewig et al., 2012). On the other hand, direct conversion of residual glia/astrocytes into neurons may be advantageous due to the native physiologic environment. In fact, several such conversions have been achieved experimentally: from astrocytes to striatum neurons by BAM expression (Torper et al., 2013), from neonatal callosal projection neurons to corticofugal projection neurons by Fezf2 expression (Rouaux et al., 2013), and from astrocytes to striatum neuroblasts by Sox2 expression and subsequently to mature neurons by BDNF and Noggin expression (Niu et al., 2013). The last example is particularly important, as it suggests that a single-lineage TF is insufficient for the full maturation of converted neuron-like cells and that multiple steps involving trophic factors are required.

In most cases, it remains to be demonstrated whether direct lineage conversion to neurons in vivo can be completed or whether the converted neurons are identical to endogenous neurons. Recent genome-wide transcriptome analyses of isolated converted neurons have demonstrated that, as in pancreatic β-cell conversion, converted neuron-like cells are not identical to endogenous neurons (Li et al., 2014; Ye et al., 2015). It is likely that the surrounding microenvironment/niche plays a key role in the maturation of converted cells.

3. Recent progress in HC regeneration in mammalian cochleae

Because of the lack of HC regeneration in mature mammalian cochleae, it is difficult to directly compare their morphologic changes and transcriptome profiles with those for HC generation in zebrafish or chickens. It has, therefore, been difficult to pinpoint the “missing” keys for mammalian cochlear HC regeneration. However, from studies in zebrafish, chicken, mammalian heart, pancreas, and CNS, we have learned 1) that most regenerated cells retain the certain transcriptomic and epigenetic profiles of the cell sources, while they can still be functional; 2) that resident progenitor or stem cells in situ are the likely sources of HC regeneration; and 3) that multiple factors are required for efficient conversion and maturation of converted cells, including TFs, epigenetic factors, trophic factors, the stoichiometry of the factors, and the architectural integrity of the organ of Corti.

In the last decades, significant progress has been made towards HC regeneration in mammalian cochleae. Here, we focus on recent developments (mainly within the past 2–3 years) in the field of HC regeneration in mammalian cochleae in vivo, with some mention of ex vivo or in vitro studies that support in vivo conclusions. It becomes notable that results of ex vivo and in vitro studies using cochlear explants, stem cells, or other cell lines are sometimes inconsistent with those of in vivo experiments. One line of striking evidence in this regard is that when Wnt/β-catenin is overexpressed in neonatal mouse cochlear SCs, both proliferation and conversion occur ex vivo, whereas only proliferation occurs in vivo using the same genetic model (Chai et al., 2012). Comparably, in CNS studies, ex vivo studies of cerebellar granule cell migration concluded that the weaver mutation was non–cell autonomous in nature, whereas it was subsequently proven in vivo to be a true cell–autonomous effect (Goldowitz et al., 1995).

3.1. Spontaneous regeneration in mammalian cochleae

In contrast to what had been commonly believed, neonatal cochleae exhibited spontaneous HC regeneration in vivo after being damaged (Cox et al., 2014b). In two independent mouse models, cochlear HCs were ablated by ectopic expression of a toxin at birth (Postnatal day 0 or P0); surprisingly, a wave of newly regenerated HCs formed between P3–P6. However, most regenerated HCs did not survive and a 2^nd wave of death occurred after P7 (Cox et al., 2014b). This transient wave of HC regeneration was further demonstrated to be derived from SCs within the cochlear sensory epithelium by genetic fate mapping. Moreover, both transdifferentiation and mitotic regeneration were observed for the spontaneous HC regeneration, demonstrating the mechanistic conservation of HC regeneration in mammalian cochleae and utricles, zebrafish, and chickens. However, such spontaneous regenerative capacity declines after P7 in vivo. It is important to note that the results from ex vivo studies are consistent with those obtained in vivo (Bramhall et al., 2014). Many more questions remain, however: Why do newly regenerated HCs die eventually? Why do cochleae older than P7 lose their spontaneous regenerative capacity? What mechanisms govern such regeneration? Molecular analysis of this spontaneous regeneration is, therefore, warranted. Interestingly, deletion of Rb induces proliferation of neonatal HCs, while the proliferating cells undergo apoptosis rapidly (Sage et al., 2005; Weber et al., 2008). Thus, the Rb pathway may be involved in the self-regeneration and apoptosis process.

Neonatal mouse cochlear SCs can proliferate with appropriate genetic modifications. Specifically, SCs can proliferate when cell cycle inhibitors (i.e. p27^kip1 and Rb) are removed (Liu et al., 2012b; Oesterle et al., 2011; Yu et al., 2010) or when canonical Wnt signaling is increased by stabilizing β-catenin (Chai et al., 2012; Shi et al., 2013). It should also be emphasized that mature cochleae exhibit no proliferative effects when these same genes are manipulated in a similar manner (Cai et al., 2014; Cai et al., 2015; Kelly et al., 2012; Liu et al., 2012a; Walters et al., 2013).

Apart from the SC proliferation for HC regeneration, some neonatal SCs can themselves be regenerated after damage (Mellado Lagarde et al., 2014). Inner phalangeal cells (IPhCs) and inner border cells (IBCs) (Fig. 1) can be regenerated, after toxin-induced autonomous cell loss, via the direct conversion of neighboring SCs in the GER regions medial to IPhCs/IBCs (Mellado Lagarde et al., 2014), whereas the loss of Deiters’ cells (DCs) or pillar cells (PCs) at similar neonatal ages cannot (Mellado Lagarde et al., 2013). Not surprisingly, such regenerative capacity is lost again after P7 in vivo. That such a remarkable regenerative capacity is exhibited only in neonatal mouse cochleae suggests the immature and stem cell–like nature of these SCs, a feature that is also supported by single-cell RNA-seq profile analysis (Burns et al., 2015; Waldhaus et al., 2015).

Notably, in neonates, certain cell populations of the organ of Corti are multipotent; they can proliferate and generate otic spheres, as well as differentiate and express HC markers, after being mechanically isolated and grown in culture in vitro. Stem cell markers, including p27^Kip1, Sox2, Lgr5, and Axin2, are expressed consistently in the developing neonatal cochlea. By using genetically modified mouse models carrying fluorescent protein labeling, cells expressing p27^Kip1, Sox2, Lgr5, and Axin2 were purified from neonatal mouse cochleae, which displayed significant potency for self-renewal (Chai et al., 2012; Jan et al., 2013; Kiernan et al., 2005; Shi et al., 2012; Shi et al., 2013; White et al., 2006). Moreover, the decrease in the level of these markers during aging (for example, Lgr5 is broadly expressed in GER cells, IPhs, IPCs, and the third DCs after birth but becomes restricted to the third DCs in adulthood; see Fig. 1) correlates with the reduced regeneration ability in SCs. However, more evidence is needed to prove cells expressing these genes are indeed cochlear progenitor cells. For example, only about 30% of Lgr5-positive cells purified from neonatal murine cochleae expressed HC markers under the differentiation condition, suggesting the heterogeneity of these cells (Chai et al., 2012). Recently, the application of single-cell PCR and RNA-seq has allowed the high-resolution analysis of different cochlear cell populations. For example, Waldhaus et al. compared apical inner pillar cells (IPCs) and outer pillar cells (OPCs) (Fig. 1) and revealed that decreased Notch signaling and increased Wnt signaling may account for the different potencies of IPCs and OPCs for regeneration (Waldhaus et al., 2015). However, the epigenetic signature for these potential progenitor cells is still lacking, simply because of the technical barrier posed by the small number of cells. ESCs or progenitor cells have quite different chromatin properties and DNA methylation status compared with differentiated cells (see Fig. 3 for the epigenetic landscape) (Layman et al., 2014). These epigenetic features can be keys to the lineage differentiation and regeneration in mature cochleae.

Fig. 3.

Modified “Waddington landscape” for cochlear HC development and regeneration. The original (shaded) cell at the top of the “mountain” represents a zygotic cell. Below, along the “valley,” is an ES/iPS cell representing a pluripotent embryonic stem cell or induced pluripotent stem cell that can give rise to an otic progenitor cell (PC), which in turn gives rise to either a supporting cell (SC) or a hair cell (HC) by following the normal developmental paths (blue arrows). A differentiated fibroblast (Fb) can be reversely converted to iPS by the four Yamanaka factors, following an “uphill” path (green arrows) as opposed to the normal “downhill” developmental path. Importantly, the direct conversion of an SC to an HC takes a path (red arrows) that differs from that of indirect conversion in order to avoid the PC state; there is, therefore, a unique, hybrid cell type (red cell) that is different from a PC or ES/iPS.

3.2. Multiple factors for HC regeneration

Embryonic, neonatal, and juvenile SCs can be converted to HCs by the forced expression of Atoh1 in vivo (Atkinson et al., 2014; Gubbels et al., 2008; Izumikawa et al., 2008; Izumikawa et al., 2005; Kawamoto et al., 2003; Kelly et al., 2012; Kraft et al., 2013; Kuo et al., 2015; Liu et al., 2014b; Liu et al., 2012a; Pan et al., 2013; Wu et al., 2013). Such conversion was first described in a pioneering report in which ectopic expression of Atoh1 successfully converted SCs in the GER region to HCs in rat cochlear explant culture (Fig. 2B) (Zheng et al., 2000). Together with a study in which fibroblasts were converted to skeletal muscle cells by MyoD in vitro (Davis et al., 1987), this study was one of the first on direct lineage conversion by lineage specific TFs. However, the effect of Atoh1 in HC regeneration has been controversial. In several reports, Atoh1 overexpression is capable to convert SCs to functional HCs in the cochlea of mice and guinea pigs by in utero gene transfer or viral delivery to adult inner ears (Gubbels et al., 2008; Izumikawa et al., 2005; Kawamoto et al., 2003; Kraft et al., 2013). In contrast, we and others showed that ectopic expression of Atoh1 in SCs in neonates and juveniles elicited only partial conversion, leading to immature HCs in vivo (Kelly et al., 2012; Liu et al., 2014b; Liu et al., 2012a). The immature features of these HCs included the lack of terminal differentiation markers (i.e., prestin, oncomodulin, or vGlut3), abnormal morphology (i.e., a larger cell body size), immature electrophysiologic responses, and immature hair-bundle morphology (Liu et al., 2014b; Liu et al., 2012a). Moreover, the efficiency of Atoh1-mediated conversion was low (6%–20%), and the mature cochlear SCs did not respond to ectopic expression of Atoh1 (Kelly et al., 2012; Liu et al., 2014b; Liu et al., 2012a). Recently, Atkinson et al. showed that transduction with adenoviruses containing Atoh1, even when combined with neurotrophin-3 (Nt3) or BDNF, resulted in an increase in immature HCs but no increase in synaptic ribbons after aminoglycoside treatment in adult guinea pigs (Atkinson et al., 2014). This contradiction may be partially attributed to the varied responsiveness to Atoh1 in SC subtypes.

A clinical trial of this Atoh1 gene therapy was approved by the FDA for HC regeneration (CGF166 by Novartis and GenVec). It was paused for further review of the trial’s Data Safety Monitoring Board (GenVec Press Release Jan. 20, 2016) and continued afterwards (GenVec Press Release May. 2, 2016). Referring to the regeneration in other systems, we believe additional factors are required to facilitate more efficient and fully functional HC conversion in vivo. One example is that co-activation of canonical Wnt signaling with Atoh1 overexpression increased the number of converted HCs in vivo by nearly 10 folds (Kuo et al., 2015). Because proliferation and conversion are both involved in zebrafish and chicken HC regeneration, it is reasonable to combine Atoh1 and β-catenin overexpression in SCs in vivo. Indeed, both proliferation and conversion of SCs to HCs were detected in neonates, and not surprisingly, new HCs were found in the apical cochlear turns in mature animals—at least 10-fold more than were observed with Atoh1 or β-catenin ectopic expression alone (Kuo et al., 2015). Interestingly, the authors only detected β-catenin-regulated proliferation but not differentiation, whereas a previous report demonstrated that high levels of β-catenin increased cell proliferation and low levels facilitated HC differentiation (Shi et al., 2013). These controversies may result from the different models used by the two groups. Shi et al. used the Sox2-CreER line to turn on stabilized β-catenin, while Kuo et al. used Lgr5-CreER and PLP-CreER lines. One possibility is that β-catenin activates negative feedback signals (e.g. Notch signaling) and suppresses HC conversion, while this suppression does not exist in the Sox2-CreER line due to Sox2 haplodeficiency (Kuo et al., 2015; Shi et al., 2013). Notably, these synergistic effects of Atoh1 and β-catenin did not improve HC maturation and hearing in vivo. Whether Wnt signaling is active in normal or injured mature cochleae remains debatable, but active studies in this area may lead to increased regeneration of HCs. Given that Notch inhibition occurs early in zebrafish HC regeneration (Jiang et al., 2014), Atoh1 is a target for both Notch and β-catenin (Lanford et al., 2000; Shi et al., 2010), and upregulation of Wnt signaling activates Notch pathway in cochleae (Kuo et al., 2015), it is reasonable to further investigate the roles of Notch, Wnt, and Atoh1 in mature HC regeneration, especially their synergistic effect.

What could be the underlying mechanism for that Atoh1 plus additional factors induce direct lineage conversion to functional HCs? According to the literature, most cell fate conversion requires more than one factor (Fig. 2) (One exception is a single TF, MyoD, induced conversion from fibroblasts to myoblasts (Davis et al., 1987)). Genome-wide analysis of TF occupation showed that TFs physically interact with only a low percentage of their consensus binding sites, suggesting that the chromatin structure is a stronger effector than the DNA sequence (Carr et al., 1999; Iyer et al., 2001; Kaplan et al., 2011; Yang et al., 2006). One or a few TFs (namely pioneer factors) usually bind the genomic DNA first, modify the chromatin structure, and facilitate the interaction of other factors(Davis et al., 1987; Liang et al., 2008; Soufi et al., 2012; Vierbuchen et al., 2010). As a result, these factors together generate the lineage-specific transcriptome. For example, FoxA, a TF, is critical for hepatocyte differentiation and regeneration. FoxA resembles the linker histone structure, and the binding of FoxA increases the accessibility of the DNA to the other TFs (Sekiya et al., 2009; Zaret et al., 2011). In HC development and regeneration, Atoh1 displayed some pioneer factor features like spontaneous binding (Abdolazimi et al., 2016). However, it is not clear whether the binding of Atoh1 to DNA facilitates the DNA-protein interaction of other TFs. Interestingly, recent studies identified GATA3 and TFE2 as co-factors of Atoh1 to promote HC conversion from GER cells in explant (Masuda et al., 2012), and earlier studies showed GATA3 and TFE2 to be involved in increasing DNAase I hypersensitivity and mediating nucleosome positioning (i.e. features of pioneer factors) (Eeckhoute et al., 2007; Ishii et al., 2004). It would be interesting to clarify whether GATA3 and TFE2 are necessary for Atoh1 to bind more target sites or Atoh1 forms a pioneer-factor complex with GATA3 and/or TFE2 and, consequently, initiates HC regeneration.

Could other factors contribute to cell proliferation and conversion to HCs in vivo? One interesting TF is Pou4f3, which has been recognized as a survival promoter rather than a differentiation factor (Erkman et al., 1996; Xiang et al., 1997). However, overexpression of Eya1 and Six1 can convert SCs to HC-like cells in embryonic cochlear explants (Ahmed et al., 2012), in which a significant portion (> 50%) of newly converted Myo7a-expressing HCs did not express Atoh1 but did express Pou4f3. This observation suggests that Pou4f3 may play an equally important role as Atoh1 in HC regeneration. In support of this hypothesis, the overexpression of a combination of three TFs (Atoh1, Pou4f3, and Gfi1) can convert mouse embryoid body (mEB) cells to HC-like cells in vitro (Costa et al., 2015) (Fig. 2B). In addition to Pou4f3, Lin28b is an Atoh1 target that, when overexpressed, can itself induce HC formation in explant culture (Golden et al., 2015). The overexpression of Sox4/11 can similarly convert SCs to HC-like cells in utricle explants, whereas their activity in the cochlea is unknown (Gnedeva et al., 2015). The lack of ephrin-B2 signaling led to transdifferentiation from SCs to HCs in vivo, which was possibly due to SC translocation to HC layer and conversion to HC fate (Defourny et al., 2015).

Overall, the HC regeneration research has made significant accomplishments in identifying new factors and improving regeneration efficiency in vivo. Especially, the genome-wide and single-cell resolution studies provide invaluable information to investigate the cochlea development and regeneration (Burns et al., 2015; Jiang et al., 2014; Ku et al., 2014; Liu et al., 2014a; Scheffer et al., 2015; Steiner et al., 2014; Waldhaus et al., 2015). It is the time now to revisit several conceptually important questions related HC regeneration in more detailed.

4. Differences between HC regeneration and HC development

It has long been debated whether HC regeneration needs to follow a molecular path similar to that in normal HC development. Significant effort has been made in studying HC developmental programs in the hope of providing guidelines for cochlear HC regeneration, but the difference between regeneration and development is becoming more acceptable (Atkinson et al., 2015; Srivastava et al., 2016). In the best example, mouse ESCs or otic progenitor cells were redirected, in multiple steps, toward the HC fate in vitro by mimicking the natural HC developmental environment in vivo (Koehler et al., 2013; Oshima et al., 2010). Although a recent study showed organoid HCs achieved from a 3D stem cell culture system recapitulated electrophysiological properties of non-mammalian or mammalian vestibular hair cells (Liu et al., 2016), most of the resulting HC-like cells exhibited diverse electrophysiologic and morphologic properties of HCs, and never phenocopied cochlear HCs (Koehler et al., 2013; Liu et al., 2016; Oshima et al., 2010). However, a combination of four TFs (Oct3/4, Sox2, Klf4, and c-Myc, also known as Yamanaka factors) successfully reversed the development and generated ESC-like cells from terminally differentiated fibroblasts (Takahashi et al., 2006). This breakthrough discovery makes us to reconsider the regeneration strategy compared with normal development.

4.1. The Waddington landscape model for development and regeneration

Historically, the cell fate lineage specification between different cell states has been modeled as the “Waddington landscape” (Waddington, 1957), where “peaks” with higher energy represent more pluripotent cell states, whereas “valleys” represent paths of normal developmental cell differentiation (Fig. 3). The Yamanaka factors converted cells (i.e. fibroblasts) at their low-valley points to more progenitor-like, high-peak states (iPS in Fig. 3) through a reverse path in the mountainous developmental landscape (Takahashi et al., 2015). Similarly, cochlear SCs are terminally differentiated cells at low-valley points, whereas otic progenitor cells (PC in Fig. 3) are supposed to be downstream from, but closer to, ESCs (ES in Fig. 3) at higher peaks. Therefore, SCs in the low-energy starting state may not have to dedifferentiate to the higher-energy state of progenitor/stem cells and thus save energy by taking different paths to reach the common final destination of mature, differentiated HCs (Takahashi et al., 2015) (Fig. 3). Thus, a new hybrid cell state between the SC and HC states may exist in the direct conversion path (indicated by the red circle and path in Fig. 3). These two paths (direct conversion and normal development) may or may not converge before reaching the final destination of mature HCs.

In fact, there are opposing views on whether conversion bypasses progenitor cell state. During the retina regeneration in zebrafish, Müller glia cells first dedifferentiate to a progenitor stage before these injury-induced progenitor cells proliferate and differentiate to rod receptors following the developmental path (Fausett et al., 2006; Powell et al., 2013); however, this assertion is only based on expression of several progenitor cell markers in converting Müller cells. In contrast, the mature glia cells can be induced to neurons without dedifferentiating to progenitor cells (Vierbuchen et al., 2010) and direct conversion occurs without returning to progenitor cell states in pancreas, heart and other regenerative systems as described above.

In the auditory system, a recent study in the zebrafish lateral line showed that fish with a retroviral integration-disrupted gene, Pho, exhibited normal HC development but defective HC regeneration (Behra et al., 2009). Specifically, the retroviral insertion into the first exon of Pho results in a null mutation zebrafish line. The fish developed normal lateral line, but after damage induced by copper or neomycin, the pho mutant showed significantly lower proliferation rate of support cells and severely impaired HC regeneration. The distinct roles of Pho in HC development and regeneration are one of the first lines of genetic evidence to highlight different molecular pathways between HC development and HC regeneration. The same investigators recently screened more than 200 gene-targeted mutant zebrafish lines and discovered nine that also exhibited specific defects in HC regeneration but showed no obvious developmental defects and morphological defects (Pei and Burgess, unpublished; (Burgess et al., 2016)). In addition, in zebrafish, retinoic acid pathway regulates FGF activity during HC development but they have no crosstalk in regeneration (Rubbini et al., 2015). These results strongly favor different paths of HC regeneration and HC development.

4.2. Breaking the epigenetic barriers

At molecular levels, terminally differentiated cells such as SCs have different genome-wide transcriptome and methylation profiles from those of progenitor/stem cells and differentiated final cell state such as mature HCs. Theoretically, either directly converting SCs to HCs or dedifferentiating SCs to progenitors is energetically unfavorable. One of energy barriers is epigenetic that separates the lineage fates, including histone modification (methylation, acetylation, and phosphorylation), DNA methylation and hydroxymethylation, and chromatin remodeling (Fig. 4).

Fig. 4.

Possible epigenetic regulatory mechanisms during HC regeneration. The compact chromatin structure becomes accessible to transcriptional mechanisms by histone modification, including methylation (Me) and acetylation (Ac), and chromatin remodeler binding. Pioneer TFs are able to interact with the condensed DNA and facilitate the targeting of other TFs. DNA methylation also controls the initiation of transcription.

For example, histone acetylation has been broadly demonstrated in cell conversion and regeneration. Histone deacetylase (HDAC) inhibition dramatically facilitated iPSC induction by Yamanaka factors (Huangfu et al., 2008). In regenerative liver, an altered pattern of histone acetylation was found decades ago (Pogo et al., 1968). By using pharmacologic and genetic methods, investigators determined that zinc-dependent HDACs are required for liver regeneration (Huang et al., 2013) and that overexpressing SIRT1 (a Class III HDAC) also suppressed hepatocyte proliferation by impairing farnesoid X receptor activity (Garcia-Rodriguez et al., 2014). Similarly, it has been thoroughly reviewed that epigenetic factors regulate the development and regeneration of pancreas (Avrahami et al., 2012; Migliorini et al., 2014), heart (Quaife-Ryan et al., 2016), and CNS (Arlotta et al., 2014; Petersen et al., 2016).

Our increasing knowledge of epigenetic regulation provides new insights into HC differentiation and regeneration. Class I histone deacetylases are expressed in otic vesicles in mouse embryos and in the organ of Corti in neonatal animals (Layman et al., 2013; Murko et al., 2010). Interestingly, histone methyltransferase EZH2, histone demethylase KDM1A, and NuRD complex core CHD4 all show a similar developmental expression pattern in the murine organ of Corti. The coordinated changes in epigenetic regulators and critical genes (including Atoh1 and mTRET) during HC development indicate that HC differentiation and maturation may be under epigenetic control (Layman et al., 2013). Indeed, HDAC1 knockdown disturbed inner ear development in zebrafish, and the pharmacologic inhibition of LSD1 resulted in fewer sensory HCs in zebrafish neuromasts during lateral line development (He et al., 2016; He et al., 2013). In mammals, a recent study on the epigenetic status of the Atoh1 locus revealed that histone modifications at that locus contribute to the changes in the expression level of Atoh1 mRNA during development (Stojanova et al., 2015). During the progression from E14.5 otic precursors to E17.5 HCs, the bivalent chromatin structure markers H3K4me3 and H3K27me3 were downregulated at the Atoh1 enhancer region, indicating a poised status for Atoh1 expression. During the same period, the acetylation of the active enhancer marker H3K27 increased and elevated Atoh1 expression, which correlated with HC differentiation. Afterwards, H3K9me3 suppressed Atoh1 expression until the HCs were mature at P6. Although in vivo evidence is still lacking as to whether histone modifications are sufficient and necessary for Atoh1-mediated HC development, this report is the first to show that the epigenetic regulation of a specific gene could play a role in HC differentiation.

Are epigenetic factors also involved in HC regeneration? A series of studies by He et al. suggest that LSD1 and HDACs are required for HC regeneration in zebrafish. Small-molecule inhibitors of LSD1 or HDAC reduced HC regeneration by suppressing cell proliferation (He et al., 2014; He et al., 2016). However, because SCs in adult mammalian cochleae lose their proliferation ability, HDAC and LSD1 may then play different roles than during development. Although HDAC inhibitors facilitate the conversion of MEFs to iPSCs, the application of HDAC inhibitors consistently, and surprisingly, had no effect on Atoh1-mediated conversion in adult mouse cochleae in vivo (Layman et al., 2015). Thus, although HDAC inhibitors are yet to be tested in younger mouse cochleae, it is possible that less dynamic epigenetic regulatory mechanisms (e.g., DNA methylation) dominantly regulate lineage-specific genes for HC fate.

Moreover, post-transcriptional regulation, especially microRNA-induced gene silence, has been implicated in ear development and HC regeneration. MicroRNAs are broadly expressed in mouse cochleae (Weston et al., 2006), and conditional knockout mice for dicer (which processes pre-miRNA to functional miRNA) showed impaired HC development and premature cell death (Friedman et al., 2009). Further studies identified the specific miRNAs and their downstream targets. In these studies, the miRNA 183 family, miRNA 124, miRNA 210, and let7 were shown to play essential roles in HC formation during inner ear development (Golden et al., 2015; Huyghe et al., 2015; Li et al., 2010; Weston et al., 2011). The miRNA 183 family also facilitates Atoh1-promoted HC regeneration in vitro (Ebeid et al., 2016).

Notably, although the majority of TF-induced conversion cannot totally erase the epigenetic memory (Hiler et al., 2015; Polo et al., 2010), a recent study indicates TF-induced conversion may be capable to modify the epigenetic landscape. The investigators profiled the DNA methylation of converted pancreatic β-cells from acinar cells in vivo over a period of up 12 months (Li et al., 2014). The results clearly demonstrated that the different DNA methylation regions compared to those of endogenous β-cells rapidly decreased within the first 10 days. Indeed, besides the direct manipulation on the global epigenetic regulators (e.g. HDACs), pioneer factors, as reviewed above, can modify chromosome structure and help overcome the epigenetic barriers. Identification of proper pioneer factors could be essential to improve the conversion for HC regeneration.

5. Clinical applications

With much progress having been made toward understanding the molecular genetics of HC regeneration in various species, the next question is how to use this knowledge to benefit people who have lost their cochlear HCs and are already hearing impaired, such as those with NIHL. To address this important translational question, we will review the current status of therapeutics for inner ear diseases.

5.1. Viral delivery of factors for HC regeneration

Many recent studies have successfully used viral vectors to deliver genes in vivo in order to correct genetic defects within the cochlea, thus making viral gene therapy a promising avenue for HC regeneration.

One of the first such studies used AAV1 to deliver vGlut3 (a gene necessary for IHC synaptic ribbon structure and function) into the cochlea of vGlut3-knockout deaf mice, thereby successfully restoring hearing and reversing the morphologic changes in the afferent nerve innervation to the IHCs (Akil et al., 2012). The AAV1-vGlut3 expression is specific to neonatal IHCs in the knockout mice, despite the broad viral uptake within the cochlea, and the hearing tests used included ABR, DPOAE, and startle behavioral response tests. Similarly, Chien et al. used AAV8-whirlin (a gene involved in hair-bundle structure and function) to infect neonatal whirlin-deficient deaf mice, resulting in stereocillia development in infected HCs and the enhanced survival of IHCs in vivo; however, the hearing of infected whirlin-deficient mice was not improved (Chien et al., 2016). In another study using AAV2/1-MsrB3, Kim et al. successfully infected embryonic otocysts of MsrB3-deficient mice at E12.5 and thereby restored hearing and hair-bundle morphology in adults (Kim et al., 2016). Similar approaches have been successful in vivo for Gjb2 (Iizuka et al., 2015; Yu et al., 2014), Cx30 (Gjbx) (Miwa et al., 2013), and (Askew et al., 2015).

Given such successes having been achieved with gene therapy in animal models, one would expect viral delivery of HC regeneration genes into the organ of Corti in vivo to have a bright future. However, despite the robust effect reported in the original report (Izumikawa et al., 2005), researchers (even including the same group) either failed to convert SCs in mature cochleae or only observed immature and nonfunctional HCs after Atoh1 overexpression using either viral infection or transgenic lines (Izumikawa et al., 2008; Kawamoto et al., 2003; Kelly et al., 2012; Kraft et al., 2013; Liu et al., 2012a; Yang et al., 2012a). Since Atoh1 is able to repair or regenerate the damaged or lost stereocilia of the HC, the improved-hearing and observed “regenerated” HCs may came from this repair effect (Yang et al., 2012b). Another explanation is that Atoh1 has different impact on different types of SCs. For example, adult IPh and IB cells are more responsive to Atoh1 induction than mature DCs and PCs (Walters et al, unpublished). Another complication is about the role of Atoh1 in HC survival: Atoh1 is required for differentiated HC survival (Cai et al., 2013; Pan et al., 2012), but continuously high level of Atoh1 in converted HCs may result in cell death (Kelly et al., 2012; Liu et al., 2012a).

Clearly, many questions remain regarding effective and safe gene therapy for HC regeneration. For example, how effective and specific are various viruses, including adenovirus, lentivirus, and many subtypes of AAVs, in infecting the mature organ of Corti? Many studies showed inconsistent transduced cell groups and varied transduction efficiency (Akil et al., 2012; Duan et al., 2010; Han et al., 1999; Kho et al., 2000; Kilpatrick et al., 2011; Konishi et al., 2008; Liu et al., 2005; Luebke et al., 2001; Mondain et al., 1998; Stone et al., 2005), resulting from different ex vivo or in vivo conditions, different animal models, different expressing vectors, different viral diffusion rates, etc. Moreover, will the virus affect other cells and cause permanent genetic changes? Although AAV has minimal toxicity and rarely causes inflammation, its vector has smaller gene packaging capacities. Furthermore, since AAVs do not pass the round window membrane (RWM), a surgery is required to instill the virus into the cochlea, increasing the risk of hearing loss associated with surgical trauma and inflammation. The gelatin sponge or RWM digestion may be a potential good solution for this (Jero et al., 2001; Wang et al., 2012). Nevertheless, viral gene therapy remains promising as one of the main methods for HC regeneration. Recent advances in CASPR/Cas9 and NgAgo offer alternative promising avenues to correct genetic deafness mutations (Gao et al., 2016; Zuris et al., 2015). Notably, as an alternative delivery method, researchers are developing non-viral vector transfer approaches. Recent studies use nanoparticles to deliver plasmids into the cochlea and greatly improve the efficiency compared to the old lipid-based method (Zhang et al., 2011; Zuris et al., 2015).

5.2. Drug discovery for HC regeneration

Recently, the pharmacologic conversion of fibroblasts into neural stem cells has become feasible (Zhang et al., 2016). A cocktail of nine drugs targeting several known signaling pathways has been demonstrated to induce the conversion in vitro and in vivo, thus giving a reason to expect that a cocktail of different drugs could help regenerate HCs in vitro and in vivo.

Commonly used screening systems include cell lines, stem cells, primary cells, explants, and zebrafish -- each has its advantages and disadvantages. Cell lines (e.g., HeLa and HEI-OC1) are easy to culture and expand, but they are far away from real cochlear cells. Screening for p27^kip1 inhibitors, for example, was undertaken via two different approaches. The first approach was performed at the transcriptional level by directly screening for p27^kip1 transcriptional inhibitors in Hela cells (Walters et al., 2014). By using a p27^kip1 promoter–driven luciferase reporter assay, Walters et al. screened a library of 4,350 bioactive compounds, including 835 FDA-approved compounds, and discovered four inhibitors (Alsterpaullone, 2-Cyanoethyl or A2CE, Cerivastatin, Resveratrol, and Triacetylresveratrol) that can inhibit p27^kip1 transcription in various cell lines and cochlear explants. One compound, A2CE, was shown to inhibit the p27^kip1 upstream regulator Sirt2. It remains to be determined whether these compounds function in vivo when delivered to animal models. The second approach was to screen p27^kip1 protein inhibitors (Iconaru et al., 2015). Traditionally, this approach had been considered impossible because p27 normally behaves as an intrinsically disordered protein (IDP) that takes various forms under different conditions. Iconaru et al. utilized a novel method (a NMR-based fragment screening) to identify fragment binding hits (which are much smaller than small-molecule compounds but bind more easily to disordered proteins with lower affinity). They discovered two groups of fragments from which they hope to develop better compounds with higher affinity for p27 protein in cells.

Although the primary otic progenitor is a good model to study HC differentiation and regeneration, it has a limited number of cells and passages, and is not ideal for large-scale screening (Oshima et al., 2010). Thus, methods have recently been developed to overcome this. Conditionally reprogrammed otic stem cells (CR-OSCs), “pseudo-immortalized” cochlear progenitor cells, have been developed using the “conditional reprogramming” technique (Walters et al., 2015c). These cells can bypass the senescence that is inherent to cochlear progenitor cells without genetic alteration, and more than 15 million such cells can be generated from a single cochlea. The CR-OSCs can be further differentiated in vitro with feeder cells, and they upregulate both early and terminal differentiation genes associated with HCs, including prestin. CR-OSCs also respond to known HC cues, including the upregulation of HC genes in response to Atoh1 overexpression and the upregulation of prestin expression after thyroid hormone administration. A hair cell line capable of regulated expression of HC genes can easily be recreated in any laboratory from any mouse strain of interest. These CR-OSCs, which are now available upon request (licensed and distributed via Applied Biological Materials, Inc.), make it feasible to test candidate compounds or genes for HC regeneration ability. It is also possible that future refinements of CR-OSCs will allow large-scale drug screenings for HC regeneration.

Similarly, immortalized multipotent otic progenitor (iMOP) cells, a fate-restricted cell type, were created by the transient expression of c-Myc in Sox2-expressing otic progenitor cells from mouse embryonic cochleae (Kwan et al., 2015). Subsequently, the endogenous c-Myc was activated, and the existing Sox2-dependent transcripts were amplified to promote self-renewal and even retain the ability of the cells to differentiate into functional HCs and neurons. This progenitor-derived immortalized cell line are also suitable to screen and test drugs and factors for HC regeneration.

Inducing the expression of Atoh1, Pou4f3, and Gfi1 together in mouse embryoid body cells generated cells with certain HC markers, and some of the induced cells exhibit stereociliary bundles (Costa et al., 2015). This system also provides a potential platform for regeneration drug screening, but further studies are needed to investigate the molecular landscape of HC induction from stem cells by TFs, which may be quite different from that of HC conversion from SCs as reviewed above.

By using zebrafish lateral line models, Namdaran et al. identified compounds that modulate HC regeneration (Namdaran et al., 2012). From a total of 1,680 compounds screened, two enhancers and six inhibitors have been identified for HC regeneration in this model. It remains to be determined whether these promising compounds also work as expected in mammalian cochleae, considering fundamental differences between zebrafish and mammals.

To validate the effect of drugs from high-throughput screening or genetic studies of p27^Kip1, Atoh1, and Notch in HC regeneration, several groups have characterized small-molecule compounds (inhibitors of p27 or Notch and activators of Atoh1 or Wnt) using mammalian deaf models including mice and guinea pigs. However, it is notable that different methods that induce hearing loss also have different impacts on SCs. For example, diphtheria toxin (DT) induces HC death (through the artificially expressed DT receptors in HCs) and triggers robust Wnt signal, which facilitates spontaneous HC regeneration in the neonatal cochlea, while neomycin treatment (which is toxic to many cells in the cochlea) also leads to HC ablation, but only turns on weak Wnt signal, which fails to regenerate HCs (Hu et al., 2016). Moreover, although noise, aminoglycoside, and cisplatin trigger similar apoptotic pathways in HCs (Cheng et al., 2003; Devarajan et al., 2002; Nicotera et al., 2003; Wang et al., 2004), it remains largely unknown whether the SCs endure similar signals upon damage and whether this affects their regeneration ability.

The best-known in vivo study found that using the Notch inhibitor LY411575 (a γ-secretase inhibitor at a dose of 4 mM) in noise-damaged mouse cochleae resulted in the activation of Atoh1 and the induction of new HC formation, leading to an 8-dB improvement in hearing (Mizutari et al., 2013). However, as we discussed previously, this effect is still under debate. Using the Notch inhibitor DAPT (an analogue of LY411575) or a blocking antibody to the Notch1 receptor, Maass et al only detected HC regeneration in neonatal mice but not in adult animals. Considering the slightly different drugs used in the two studies, as well as the other distinct experimental conditions (i.e. the Sox2-CreER mouse vs. the wild type mouse), additional Notch inhibitors must be tested by lineage tracing of SC conversion to HCs followed by noise damage in vivo (Tona et al., 2014).

Since Notch inhibition results in different outcomes (e.g. regeneration or no effect, purely differentiation or proliferation) (Daudet et al., 2009; Maass et al., 2015; Mizutari et al., 2013; Wu et al., 2016). Efforts have also been made to screen for activators of Atoh1 (Edge et al, 2012, US Patent #: US8188131B2) that could complement Atoh1 viral gene therapy and consolidate Notch inhibitor effects because activation of Atoh1 is considered as the potential mechanism of Notch inhibition (Doetzlhofer et al., 2009).

It is now conceivable that the development of appropriate drugs will enable effective HC regeneration in vivo. However, many hurdles remain. For example, how can these reagents be delivered effectively to the mature organ of Corti in vivo? Drugs are normally difficult to pass through the blood-labyrinth barrier (BLB). In general, many drug candidates are poorly characterized in terms of their ability to cross the BLB and their pharmacokinetics and pharmacodynamics (PK/PD) in the inner ear fluids. For example, doxycycline cannot effectively penetrate into the cochlea of adults (Cox et al., 2014a; Walters et al., 2015a), despite several reports to the contrary (Kelly et al., 2012; Tarang et al., 2015). Moreover, Dox is not easily cleared by the inner ear fluids after it does get in (Cox et al., 2014a; Walters et al., 2015a). The most effective method for drug delivery is local delivery (by intra- or trans-tympanic or intracochlear injection), which is the most commonly used and effective method of drug delivery in patients with middle or inner ear diseases (El Kechai et al., 2015). Such a method resembles the intravitreal injections into the eyes that are used to treat certain ophthalmic diseases. For effective drug delivery to stimulate HC regeneration in vivo, it is highly desirable that these obstacles be further studied and overcome.

Highlights.

• Spontaneous and forced hair cell regeneration occurs in mammalian cochleae in vivo;

• Multiple factors can enhance hair cell regeneration;

• Paths of regeneration may be different from those of development;

• Epigenetic memory is a key to hair cell regeneration;

• Drug discovery holds promise in clinical therapy for hearing loss.

Abbreviations

NIHL

noise induced hearing loss

HC

hair cell

SC

supporting cell

OHC

outer hair cell

IHC

inner hair cell

CNS

central nervous system

TM

tympanic membrane

RNA-seq

RNA sequencing

TF

transcription factor

iPSC

induced pluripotent stem cell

ESC

embryonic stem cell

HPF

human postnatal fibroblast

mEB

mouse embryoid body