Untangling the genomics of noise-induced hearing loss and tinnitus: Contributions of Mus musculus and Homo sapiens

Abstract

Acoustic trauma is a feature of the industrial age, in general, and mechanized warfare, in particular. Noise-induced hearing loss (NIHL) and tinnitus have been the number 1 and number 2 disabilities at U.S. Veterans hospitals since 2006. In a reversal of original protocols to identify candidate genes associated with monogenic deafness disorders, unbiased genome-wide association studies now direct animal experiments in order to explore genetic variants common in Homo sapiens. However, even these approaches must utilize animal studies for validation of function and understanding of mechanisms. Animal research currently focuses on genetic expression profiles since the majority of variants occur in non-coding regions, implying regulatory divergences. Moving forward, it will be important in both human and animal research to define the phenotypes of hearing loss and tinnitus, as well as exposure parameters, in order to extricate genes related to acoustic trauma versus those related to aging. It has become clear that common disorders like acoustic trauma are influenced by large numbers of genes, each with small effects, which cumulatively lead to susceptibility to a disorder. A polygenic risk score, which aggregates these small effect sizes of multiple genes, may offer a more accurate description of risk for NIHL and/or tinnitus.

I. INTRODUCTION

Chronic noise exposure is a side-effect of the industrial age and mechanized warfare, and noise-induced hearing loss (NIHL) is second only to age-related hearing impairment (ARHI) as a sensory injury. An estimated 22 × 10⁶ workers in the U.S. are exposed to noise greater than 85 dBA time-weighted average (TWA), and 18% of these have hearing loss, defined as pure-tone average threshold of 25 dB or more in either ear across 1, 2, 3, and 4 kHz (National Institute of Occupational Safety and Health Standards, 1998; Masterson et al., 2013). At least 20% of Europeans and 10% of Americans are noise-exposed for half of their workday (Konings et al., 2009a). Worldwide, 16%–24% of hearing loss is estimated to be noise-related (Feder et al., 2017). Hearing loss among U.S. teenagers, which can compromise educational achievement and communication skills (Arlinger, 2003; Gomaa et al., 2014; Shargorodsky, 2010), increased significantly from 3.5% to 5.3% between 1994 and 2006, possibly due to the effect of loud music through personal devices (Gregory et al., 1994; Shargorodsky, 2010). Besides impacting military operations, hearing loss is significantly associated with depression, isolation, reduced social activity, and cognitive decline (Dawes et al., 2015).

Currently, there is no prevention for ARHI; thus, it is important to distinguish between noise-induced trauma and age-related audiogram changes in order to ascertain the proportion of preventable damage and establish phenotype definitions that drive genomic studies. While age-related genome-wide association (GWAS) has routinely excluded those under 50 years old (Hoffmann et al., 2016; Vuckovic et al., 2018), studies indicate that pure-tone averages decline linearly beginning in a person's 30s and 40s (Dawes et al., 2014; Moore et al., 2014), making it difficult to distinguish NIHL from ARHI.

Unlike the inescapable ARHI, NIHL is avoidable with a combination of engineering solutions, administrative directives, i.e., limitation of time spent in noisy industries or turning down loud music, and personal protective devices such as earplugs and earmuffs (Yankaskas, 2013). A “noise notch” at 3, 4, and 6 kHz can differentiate the two, compared to a “down-sloping” audiogram with loss increasing at higher frequencies (Rabinowitz et al., 2006), and this notch needs to be quantified in order to assess it in any GWAS analysis. In order to identify acoustic-trauma-susceptible genes versus age-related changes, it will be necessary to compare GWAS results from a young exposed population, the military, or teens who listen to loud music, for example, to an older population relatively unexposed to noise.

While audiograms give some indication of distinction between ARHI and NIHL, tinnitus etiologies are distinguishable by history only due to the subjective nature of ringing in the ears. A separate auditory disorder with different sequelae but emanating from the same injury, tinnitus prevalence is estimated at 50 × 10⁶ cases in the U.S., with 16 × 10⁶ adults reporting ringing in their ears on a daily basis (Shargorodsky et al., 2010). Tinnitus and NIHL have been the Veteran Affair's (VA) number 1 and number 2 service-connected disabilities since 2006, now costing over two billion dollars per year in compensation, and in 2017 over 2.6 × 10⁶ veterans received disability payments for auditory damage (U.S. Department of Veterans Affairs, 2018). Traumatic brain injury (TBI), the signature injury in recent wars, more than doubles the risk of tinnitus (Yurgil et al., 2016).

Operationally, tinnitus impacts military mission completion in ways unconnected to hearing loss. Tinnitus leads to slower reaction times, poorer accuracy while dual tasking, anxiety, depression, and increased suicide risk (Asplund, 2003; Crönlein et al., 2016; Gomaa et al., 2014; Gopinath et al., 2010; Rossiter et al., 2006; Schecklmann et al., 2015; Tegg-Quinn et al., 2016; Trevis et al., 2016; Vogel et al., 2014). Even with normal hearing, tinnitus degrades cognition, dichotic listening, and speech-in-noise, an important factor during combat operations where the signal-to-noise ratio is diminished (Degeest et al., 2017; Jafari et al., 2012; Jain and Sahoo, 2014). It affects the lives of veterans both during and after separation from service. Sleep disturbance is the second most cited aeromedical factor in naval aviation mishaps and hazardous reports (HAZREPS), and 76% of tinnitus subjects complain of sleep dysfunction (Schecklmann et al., 2015). Sleep degradation is associated with slower reaction times and increased mistakes in recognition of targets as friend or foe (Smith et al., 2019).

Genes and genetic pathways that underlie susceptibility to tinnitus have not been identified, nor is there a cure or definitive treatment for either disorder, and we remain unable to predict the susceptibility of any individual to tinnitus or NIHL. While NIHL is often comorbid with tinnitus, the two disorders appear to have a separate pathophysiologic architecture. Injuries leading to both disorders appear to originate in the cochlea with primary neural degeneration in the high threshold cochlear-nerve fibers (Kujawa and Liberman, 2015). Hearing loss studies have concentrated on defects in the cochlea, while tinnitus is theorized to be associated with areas higher up in the brainstem and auditory cortex (Eggermont, 2015; Kaltenbach, 2011; Knipper et al., 2010; De Ridder et al., 2011; Ryan and Bauer, 2016). A study of one-half million Iraq and Afghanistan veterans with a history of TBI identifies 7.3% with a diagnosis of hearing loss, 6% with tinnitus, and another 5.6% with both (Swan et al., 2017). Thus, given the same exposure, some will be susceptible to tinnitus, others to hearing loss, some will sustain both, and some neither, indicating a genetic component to this environmentally induced injury.

This paper will review the literature addressing genetic susceptibility to tinnitus and NIHL in both human and animal studies. Its approach ranges from twin studies in humans to candidate and GWAS studies in humans and animals. The advantage of GWAS is that, unlike candidate studies, which select genes based on prior knowledge, they provide a survey of the genome in an unbiased manner.

Findings in mouse models have contributed predominantly to the genetics of hearing loss, while comparative studies on the genetics of tinnitus have proven less practical due to phenotyping limitations. Due to the types of mutations that may often underlie complex traits, the promise of mouse models for discovery and validation of human pro-NIHL genes has yet to be realized. We will outline possible strategies to identify the genetic architecture of tinnitus and NIHL, including markers—single nucleotide polymorphisms (SNPs), as well as an overall polygenic risk score (PRS) to ascertain susceptibility to acoustic trauma. We discuss strengths, pitfalls, and strategies to address these challenges.

II. PARALLEL PROGRESS IN HUMAN AND MOUSE ASSOCIATION STUDIES

Recently, correlation of findings in humans and animal models in the area of ARHI has led to identification of multiple novel variants. Based on summary data without stratification for noise exposure, authors compared United Kingdom Biobank (UKB) data to mice gene expression patterns and identified novel SNPs in genes related to Mendelian disorders, as well as other genes not previously reported (Kalra et al., 2019; Wells et al., 2019). They then utilized mouse models to relate these variants to expression in specific cells of the cochlea. Methods used in these papers may pave the way for future NIHL and noise-induced tinnitus investigations. In a reversal of previous protocols where laboratory findings are brought to human studies, human findings may increasingly guide the direction of analysis in animal labs.

A. Benefits of animal studies

Although human GWAS may direct animal experiments in new studies, it is worth revisiting why even novel approaches still must utilize animal studies. Mechanoreception is evolutionarily very old so that genetic similarities in the molecular machinery of hearing encompasses species as far-ranging as zebrafish and fruit flies (Senthilan et al., 2012; Whitfield, 2002). However, the far greater degree of similarity gained by working within mammals means that more mutations will impact overlapping aspects of anatomy and physiology down to the molecular level. In this regard, mice are unparalleled, representing the lowest phylogenetic model of mammalian hearing mechanisms. In addition, the economy and high degree of standardization of commercial mouse lines versus other species used for hearing research (e.g., guinea pigs, gerbils) are unmatched. Clearly, mapping of complex traits is difficult, regardless of species.

B. Mendelian versus complex traits

Original genetic investigations sought genes that underlie congenital hearing loss, which is typically inherited in a Mendelian manner whereby hearing loss is viewed as binary (present/absent), and the presence of particular mutations nearly guarantees hearing loss, most often inherited in a recessive fashion. Mapping of early deafness genes heavily leveraged anatomic, physiologic, and genetic similarities with respect to hearing between humans and laboratory mice (Bowl and Dawson, 2014; Ohlemiller et al., 2016; Steel, 2014). In addition to hearing changes, mice carrying putative deafness alleles can be used to localize the messenger ribonucleic acid (mRNA) or protein intracellularly, and characterize the resulting cochlear lesion. It helps that there exist over 400 inbred mouse strains, many of which carry naturally occurring disease-promoting mutations, plus large-scale efforts to generate and evaluate chemically induced mutations (El Hakam Kamareddin et al., 2015). Hearing loss extant in large repositories, such as the Jackson Laboratory, can be ascertained in organized screening programs. Using this mouse to human parallel approach, over 100 deafness genes have been identified, along with roughly an additional 100 loci.

Since the manifestations of tinnitus and NIHL depend on the interplay of cochlear injury and repair processes, many successful animal studies have focused on genes involved in protective or repair processes (Carlsson et al., 2005; Konings et al., 2009b; Sliwinska-Kowalska and Pawelczyk, 2013). If the elimination of such a gene in a knockout (KO) model produces a tinnitus- or NIHL-prone phenotype, that gene or an ortholog could be implicated in human association studies. Such studies were “candidate” since they tested specific suspects, and are the laboratory counterpart of clinical genetic screens such as Otoscope (Sloan-Heggen and Smith, 2016). The genes tested in such experiments or screens often included those highly expressed in the ear or genes that regulate the expression of those genes.

This approach netted candidates that pointed to important pathways, including oxidative stress and potassium ion channels and other tinnitus genes. However, few such genes have been replicated in human GWAS. It may be that although these variants and genes are related to acoustic injury, they are not common enough to be a problem in the general population.

Several animal studies have utilized KO mice to study the genetic architecture of tinnitus, both in the cochlea and in higher centers. The availability of mouse KO models for the majority of genes, either as universal or conditional KOs from consortia, such as the International Mouse Phenotyping Consortium (IMPC; (Koscielny et al., 2014), shortens the path from hypothesis to experimental testing. As an example, KCNQ2/3 channel KO mice allowed identification of therapeutic targets in cerebral cortical pyramidal neurons and hippocampal neurons for prevention of tinnitus (Kalappa et al., 2015). In another study, increased spontaneous firing of fusiform cells was identified as necessary for the induction of tinnitus. Mice lacking ZnT3, a vesicular zinc transporter in the dorsal cochlear nucleus (DCN) of the auditory brainstem, were not inhibited from zinc-inhibited spontaneous firing of fusiform cells, indicating that non-vesicular zinc inhibited the spontaneous activity of the DCN (Perez-Rosello et al., 2015). Last, nAChR KO mice fail to show suppression of cochlear responses, and overexpressing mice demonstrated reduced acoustic injury from noise exposures. nAChR constitutes the nicotinic receptor channel for acetylcholine, the primary efferent neurotransmitter at the inner and outer hair cells (Lustig, 2006), which transmits through the olivocochlear pathways to inhibit and modulate the afferent response. This olivocochlear pathway is strongly implicated in the generation of tinnitus (Riga et al., 2017).

In the search for “tinnitus genes,” rare monogenic disorders have been noted to be associated with complaints of ringing in the ears, as seen in the Online Mendelian Inheritance in Man (OMIM¹), but no linkage studies have been performed and, again, the findings have not been replicated in the larger environmentally exposed population. While the separate etiologies of noise-induced, age-related, and TBI-associated tinnitus may have distinct genetic architecture, extant studies have been too small to examine individual etiologies separately (Clifford and Rogers, 2017; Sand et al., 2007). In these candidate studies, originally, exonic SNPs were examined until it became evident that introns, promoters, enhancers, and other genetic regulatory elements were at least as important in the genomics of acoustic trauma. In fact, since there is such a large environmental component in these disorders, it becomes critical to examine the elements that react and interact with acoustic damage.

C. Translation of lab results to humans—Expression quantitative trait loci in animals (eQTLs)

Recent animal studies have focused on transcription factors that drive quantitative genetic expression within eQTL studies, which means mapping the expression profiles of hearing-related genes (Breschi et al., 2017; Lavinsky et al., 2016; Lavinsky et al., 2018). Genetic regions where these functions co-map may implicate a transcription factor or mutations in regulatory sequences adjacent to a conventional gene. Only about 1% of cataloged quantitative trait loci (QTLs) have been identified from human or animal studies (Ermann and Glimcher, 2012; Flint and Eskin, 2012). New and emerging mouse resources (see the following) may be expected to increase this success rate, yet, there is no anticipated substitute for the requirement to phenotype hundreds, even thousands, of mice to reduce the often large number of candidate genes within identified mapping intervals. This is not a mouse problem, but a result of the sheer complexity of the genome, only a small fraction of which performs according to our early understanding. Of the rest, about half is transcribed, much of it into multiple types of poorly understood non-coding regulatory ribonucleic acids (nc RNAs; Breschi et al., 2017).

Despite the success of applying mice to the genetics of Mendelian hearing loss, it will not necessarily follow that they are equally useful for mapping complex traits. Studies of gene expression indicate that mouse orthologs of hearing-related genes may be expressed in different cochlear cell types in mice and non-human primates (Hosoya et al., 2016a; Hosoya et al., 2016b). More broadly, comparisons of specific transcription factor binding indicate that no more than 20% of transcription factor binding sites are conserved (Breschi et al., 2017). Notably, the changes that are found often allow for the binding of additional transcription factors, suggesting evolutionary pressures to refine gene expression, perhaps more than gene function.

Since the genetics of complex traits seem likely to reflect gene regulatory networks that can differ hugely by species, it is reasonable to ask how likely it is that the same loci will account for a given complex trait across species (Gasch et al., 2016). Attempts to address this experimentally should incorporate as much mouse genetic diversity as possible. Humans differ genetically at an average 4–5 × 10⁶ SNPs, or about 0.5% of the genome (Breschi et al., 2017). All inbred mouse strains, long preferred for hearing studies, are bred to be as genetically identical as possible, and genetic variation within strain is nearly negligible. Genetic differences across classic inbred strains, the appropriate metric for comparison with individual humans, are somewhat greater than for humans, although this estimate pales when compared to “mousedom” as a whole. Mice may hold advantages over humans in terms of their sheer numbers and rapid generational turnover.

The key to mapping any complex trait with GWAS is the phenotype's statistical association with a particular SNP or other marker. Different populations carry different markers, as a result of geographic isolation and the steady pace of new mutations. Every meiosis offers an opportunity to create new and unique marker combinations, and the shorter the chromosomal distance over which this occurs, the greater the potential mapping resolution. Thus, the size and dynamism of mouse populations, compared to humans, hold the potential for more sensitive detection of loci associated with complex traits. This is most true for studies that incorporate wild-derived inbred strains. The more heavily utilized classic strains possess less diversity, and more importantly, stronger selection for particular traits. This may inadvertently select for particular combinations of alleles in linkage disequilibrium (LD) so that recombination is diminished and mapping resolution suffers. The bottom line is that sufficiently large mouse populations, chosen to maximize genetic diversity and minimize LD, should yield more genetic associations at finer resolution than most human GWAS studies. What cannot be known at the outset is whether the entire mouse genome actually contains orthologous SNPs for the same complex traits, such as tinnitus or NIHL, under study in humans (Clifford et al., 2016).

D. Mouse mapping panels for GWAS

To maximize the potential of mouse GWAS studies, new “panels” of strains have been developed that draw in as much genetic variation as possible (Flint and Eskin, 2012; Su et al., 2010). The success of using these depends more on the number of strains involved, and less so on the number of mice tested from each strain, particularly for robust data such as auditory brain response (ABR) thresholds. Some panels, e.g., the Collaborative Cross (Churchill et al., 2004), involve the generation of new recombinant inbred (RI) strains formed from selected classic inbred strains. Other panels, such as the Hybrid Mouse Diversity Panel (Lusis et al., 2016), combine many genetically divergent inbred strains with selected RI strains. The beauty of these panels is that all of the mice of any particular strain are genetically identical. Therefore, all the experimenter need do is to phenotype a few mice from, preferably, at least 30–40 strains, and then compare the distribution of phenotypes with online databases of strain-specific markers. There is no need to genotype every animal for markers, potentially saving much time and money. Other panels leverage crosses of non-inbred stocks (e.g., Diversity Outbreds; Churchill et al., 2012), but each resulting animal is unique and must be genotyped for markers. While inbred mice are preferred for most basic hearing studies, many outbred mouse strains such as CD1, ICR, NMRI, and SW exist, and have found uses in mapping due to their genetic diversity (Yalcin and Flint, 2012). These mice also must be individually genotyped. These mapping resources somewhat obviate the selection pressure that has shaped virtually all commercial inbred mouse strains (even wild-derived), increased LD, and reduced the resolution of trait mapping.

Each mouse QTL mapping endeavor involves a tradeoff of sensitivity versus inclusiveness. Consider the case of two inbred strains that differ significantly in ABR threshold and are to be used to map the genes involved using an intercross. If their phenotypic differences reflect one or a few loci, the alleles carried by the two strains will be present in a ∼50:50 mix, and the intercross is likely to reveal those loci. However, the more closely related the two strains, the more markers they will share, reducing potential mapping resolution and the odds of isolating the causal gene. Also, only the allele(s) that accounts for the poor hearing of one strain can be revealed in such a study. The study cannot cast a broad net for the many mutations carried by all mice that may affect ABR thresholds.

The newly developed mapping panels in mice are designed to detect many loci using extant variation across mice as a species, yet, at high resolution. The hope is that GWAS involving a few hundred mice can complement far costlier human GWAS involving tens of thousands of people. While studies of complex traits have often failed to replicate from mouse to humans or account for total estimated heritability of various pathologies, age-related hearing loss studies that began with human GWAS and, subsequently, replicated in mice have been more successful in identifying variants and genes relevant to humans (Buchner and Nadeau, 2015; Fransen et al., 2015).

III. DEFINING THE NOISE-INDUCED TINNITUS PHENOTYPE

A. Human studies of tinnitus

The definition of the phenotype of tinnitus in any particular genetic study drives gene identification in human analysis since GWAS is essentially a statistical regression of the phenotype on individual SNPs. In the absence of an objective measure, currently, tinnitus is defined by self-report. H2 refers to broad-sense heritability and reflects genetic variance, epistatic, maternal, and paternal effects, as opposed to SNP heritability or h2. H2 is most often studied in twins, where monozygotic and dizygotic pairs can be compared to ascertain heritability and environmental components. Imaging studies of tinnitus subjects show enhanced brain connections between the auditory cortex and arousal systems, frontal lobe, amygdala, pain centers, and others, indicating connecting gene pathways and isoforms more indicative of emotional and cognitive issues (Minen et al., 2014). Thus, it is vital to separate the perception of ringing in the ears from its psychological context in order to glean genetic information specific to tinnitus. The importance of the definition of the phenotype became evident when a large-scale study of tinnitus identified 0.11 “H2” after posing a question about “bothersome tinnitus” (Kvestad et al., 2010). Several years later, other population studies noted up to 0.43–68 H2 for unilateral and bilateral tinnitus when another study posed a differently worded question about the perception itself, i.e., “Do you have buzzing in the ears?” (Bogo et al., 2017; Maas et al., 2017).

Noise-induced tinnitus is modeled as a latent unobserved variable, defined as a vulnerability which then manifests itself as perceived tinnitus given environmental exposure. Whether the subject sustains “enough” acoustic trauma to lead to the disorder being reported is dependent upon some measure of acoustic trauma sustained by the auditory system. Concomitantly, h2 or narrow sense heritability consists of only an additive genetic portion, referring to hard-wired deoxyribose nucleic acid (DNA) that affects a trait. Narrow-sense heritability is measured via linkage disequilibrium score regression (LDSC) either on summary or individual SNP data (Zheng et al., 2016). The UKB, a cohort of 500 000 individuals aged 40–69 years old, is one of the first researcher-available large datasets (see Fig. 1). The UKB in association with Neale labs at Massachusetts General Hospital and the Broad Institute has performed a basic GWAS test on separate answers to the UKB's Touchscreen Questionnaire² and summary data.³ Briefly, half a million participants in England were asked questions about their lifestyles, including questions on hearing and ringing in their ears. The phenotype of tinnitus was treated as a categorical variable, and the “now most or all of the time” answer was calculated to explain 0.137 of the total variance at p = 1.34e-07, still short of the H2 heritability of 0.43 identified in large population studies.

FIG. 1.

(Color online) Question on tinnitus from the UKB TouchScreen. In answer to the question regarding noises in the ear, participants answered according to seven categories shown. Figures in parentheses are numbers of participants who answered within that category.

On the other hand, subjects in the Million Veteran Program (MVP), a cohort of over 700 000 U.S. veterans, were asked a more specific binomial question—whether they had been “diagnosed with tinnitus or ringing in the ears” (yes/no; Gaziano et al., 2016). Whether this difference in phenotype definition will lead to identification of different significant SNPs and genes is yet to be determined.

B. Animal models of tinnitus

In contrast to humans, where a diagnosis of tinnitus can be self-reported, reliable induction and detection of tinnitus in an animal model is exceptionally challenging. Even more complex are the changes in the perception of tinnitus depending on background noise, stress, fatigue, and other factors more difficult to measure in the animal.

Most methods for detection of tinnitus in animal models rely on Pavlovian operant conditioning (OC) or changes in acoustic behaviors. With OC methods, animals are trained to refrain from or initiate a behavior when a broad or narrowband noise is stopped for a variable period of time (von der Behrens, 2014; Jones and May, 2017). The type of induced tinnitus noise is thought to depend on the type of noise exposure, i.e., narrowband noise induces tonal tinnitus (Kiefer et al., 2015). After OC and tinnitus induction, the assumption is that animals with tinnitus will not perceive or detect silent periods in experimental sound stimuli. Because the method needs to distinguish hearing loss (the animal never heard the sound) from tinnitus (an internal sound that interferes with gap detection), these studies require (a) intact hearing in the frequency of the tinnitus and (b) induction of the tinnitus (most often using pharmacologic intervention or noise exposure) only after the OC training is complete. The resulting accuracy varies based on the training efficiency and assumes no extinction of the behavior. OC approaches are not practical for screening of mouse models with genetic mutations that may lead to tinnitus, due to the difficulty of operant training in mice. If applied to studies aiming to identify genetic propensity to develop tinnitus after a noise exposure, their low throughput presents a significant impediment.

An alternative to the OC approaches is the gap prepulse inhibition of the acoustic startle (GPIAS) reflex (Turner et al., 2006; Turner et al., 2012). This approach applies background noise with a brief silent gap just prior to the presentation of a loud “startle” sound, and does not require training. Instead, the output metric is the amplitude of the startle reflex, as measured by the force it generates. The strength of this reflex is reduced by sudden changes in the acoustic environment prior to the startle sound. These changes may include either increments or decrements (e.g., silent periods) in the background stimulus. According to the theory, animals without tinnitus will detect the silent gap and be less startled by the loud sound, whereas animals with tinnitus will not detect the gap and are hence expected to exhibit a greater startle amplitude. This is a “within subject” method as the animals are exposed to the loud sound with and without a prior silent gap. The GPIAS ratio (NoGap/Gap) is used to infer the presence of tinnitus. This method is relatively high throughput; however, some controversy exists regarding its validity. The notion that tinnitus fills a perceptual silent gap was not supported in human experiments using the same method (Fournier and Hébert, 2013). Also, the method requires normal hearing so that any deafening protocol used to induce tinnitus must be applied unilaterally and is sometimes performed under anesthesia, thus, introducing a number of potential confounds. A more recently introduced alternative, sound-based avoidance detection (SBAD) requires training, resulting in similar drawbacks to the OC techniques (Zuo et al., 2017).

Numerous studies in mice and other animals have revealed processes that may play a role in the development of tinnitus. These range from structural, immunohistochemical, and electrophysiological studies in the central and peripheral auditory system in models of acquired tinnitus (Basta et al., 2005; Hickox and Liberman, 2014; Hwang et al., 2011a; Hwang et al., 2011b; Llano et al., 2012; Lowe and Walton, 2015; Ma et al., 2006; Middleton et al., 2011; Nansel et al., 1992; Zhang et al., 2014) but also include gene expression studies (Im et al., 2007). The latter could be leveraged to pinpoint candidate genes for validation in human GWAS. In addition, for many of these mouse models, auditory phenotypes are available, whereas for others the auditory phenotyping is in progress (Bowl et al., 2017). The multidimensionality of tinnitus—including variation in laterality, loudness, apparent source location, sound quality, proximate cause, and constancy—has made it extremely difficult to classify individual human cases, let alone in animals, wherein such qualities cannot be determined with any certainty. Tinnitus is difficult to characterize, and the propensity to develop tinnitus will be a thorny complex trait to map. Given that genetic studies are only as good as the phenotyping of subjects, attempts to map “pro-tinnitus” loci are likely to be hobbled by the metrics used, or the way “cases” are combined.

IV. DEFINING THE NIHL PHENOTYPE

A. Humans

Considerations for a genetic phenotype of NIHL for their effect on genetic studies include the definition of hearing loss and the characterization of acoustic exposure. These stem from evidence that NIHL has both genetic and environmental components with an H2 genetic contribution estimated at 0.67–0.70 (Wingfield et al., 2007). Complicating this analysis is that heritability appears to be frequency specific both in animals and humans (Lavinsky et al., 2016; Wingfield et al., 2007). Suitably low and high frequencies must be applied with weighting toward those frequencies that may be noise vulnerable (Rabinowitz et al., 2006). Frequencies beyond 8 kHz are not currently standard, but should be, given the heightened vulnerability of the cochlear base.

Although the biological substrate of NIHL is well studied in humans and animals, the definition of hearing loss in the literature has not been standardized. Studies that utilize “binaural pure-tone average for the frequencies of 1000, 2000, 3000 and 4000 Hz of greater than 25 dB” (Concha-Barrientos et al., 2004) in an occupational setting may not account for genes that affect high frequencies (Wingfield et al., 2007). A normal hearing definition of “no hearing threshold greater than 25 dB HL at any hearing level frequency” (HL; Grondin et al., 2015) does not take noise notches into account (McBride and Williams, 2001; Rabinowitz et al., 2006). Recently, principal components analysis (PCA; Anwar and Oakes, 2012) has been utilized to encompass the shape and other derived parameters of the audiogram. In the large cohorts currently available, the UKB administered a digits-in-noise test. How well this test will correlate with the standard audiogram in other large studies is unknown (Moore et al., 2014). The MVP has audiology data available for analysis, and the definition of hearing loss will be critical, as these different definitions of NIHL covariates may elicit different findings on GWAS. Speech detection thresholds, speech reception thresholds (Hoffmann et al., 2016), ICD-9 codes, and self-reported hearing difficulty (Moore et al., 2014) are also available for use in GWAS studies.

Because different noise exposures appear to impart different types of injury (e.g., metabolic versus mechanical trauma), it may be important to combine subjects within and across studies by the type of noise exposure (Gaussian versus kurtotic; Grondin et al., 2015; Konings et al., 2009c). Typically there will be no way to identify the “cause” in terms of cellular substrate. That is, while one individual's 20 dB hearing loss may reflect outer hair cell loss, another's could conceivably replicate as noise-related strial injury. These would likely have different underlying genetics. Clinical audiograms have the advantage of being universal and fairly standardized but are not diagnostic for affected cell type. However, as the most global output metric, audiograms are more likely to detect hearing problems than, say, distortion product otoacoustic emissions (DPOAEs), which are less standardized and will not detect inner hair cell or neural pathology.

In regard to the nature of specific noise exposures, cohorts useful for GWAS studies often consist of workers or military personnel who shared the same environment for known periods, and whose HLs can be reasonably assumed to reflect occupational noise exposure. Thus, there must be baseline threshold data plus later assessments at fixed times. Cumulative noise exposure in humans is difficult to quantify, and here a surrogate measure is required. In the UKB cohort, occupation may serve as an effective surrogate for noise exposure. Participants answered questions regarding exposure in industrial settings as well as exposure to loud music. In the U.S., noise exposure associated with many occupations is still poorly defined (Dobie, 2008). The Occupational Safety and Health Administration (OSHA) regulations put in place in the 1980s pose both a blessing and a hindrance: Exposures in some of the largest occupational groups have been capped at levels that pose minimal risk. This is good news for workers but inconvenient for researchers, who may be tempted to conduct exposure studies in countries with less effective protections. In the U.S. MVP, where military occupations are invariably noisy, the military occupational specialty (MOS) can be used as a surrogate for noise exposure. Some data-mining of exposure in the electronic health record will be necessary to separate the noise-induced component of hearing loss from any age-related component (Gaziano et al., 2016).

B. Animals

In animals, both the subjects and exposures can be rigorously controlled, especially in genetically defined populations such as inbred rodent models. From such experiments, it has become clear that both the extent and form of cochlear injury caused by noise can vary with genetics (Ohlemiller and Gagnon, 2007). Recent research has emphasized progressive injury to the inner hair cell-afferent synapse, or “cochlear synaptopathy,” which may follow even mild noise exposure (Kujawa and Liberman, 2015). Such injury does not cause threshold shifts, but may impair supra-threshold capabilities, such as speech perception in noise. This phenomenon fits well with the single major complaint of the aged, namely a decline in sound quality (Frisina, 2009), and may represent the substrate of Schuknecht's “neural presbycusis.” However, whether moderate noise exposure can produce this type of injury is controversial, as its severity varies widely with animal model, and the results of human studies are mixed (Guest et al., 2016). With the possible exception of non-human primates, animals appear more susceptible to noise than are humans (Guest et al., 2018). Generally, the smaller the mammal, the more noise vulnerable. This does not discount the value of animals for characterizing noise injury and NIHL risk, but animal studies cannot be used to set specific exposure guidelines.

Across species, there appears broad consensus that a major driver of noise-related permanent threshold shifts is outer hair cell injury and loss. Other types of injury, such as injury to the cochlear lateral wall, endocochlear potential reduction, and neuronal loss appear more variable and clearly genetically modulated (Ohlemiller and Gagnon, 2007). Thus, grouping human subjects by the extent of NIHL may erroneously lump together different types of injuries and susceptibilities.

Other variables have the potential to complicate—or contaminate—genetic studies of noise susceptibility. Animal studies indicate that age and sex can impact noise vulnerability (Milon et al., 2018; Saunders and Chen, 1985). “Adolescent” animals appear more susceptible, although this aspect of noise vulnerability in animals oddly lacks a human research counterpart. Both the form and extent of cochlear damage depend on age in animals, strongly suggesting that NIHL at different ages represents distinctly different processes, and very likely has different genetic foundations (Ohlemiller et al., 2018). Human and animal studies that do not control for this risk will obtain misleading results. The manner of this measurement of hearing loss, moreover, may depend upon the hearing metric one applies, such as DPOAE levels versus hearing thresholds. Whether the approach is whole-genome sequencing, linkage studies, or GWAS, proper classification of ‘controls’, cases, and the form and extent of NIHL will hugely impact results.

Clearly, consensus is required for the guidance of research into pharmaceutical and gene therapy. The mouse has been the primary model organism used for modeling genetic human diseases and in particular disorders of the peripheral auditory system (Bowl and Brown, 2018). Genetic studies, however, depend on exquisitely accurate phenotyping, particularly when dealing with phenotypes that may represent complex traits. Accurate phenotyping is the sine qua non of any genetic study.

V. MAPPING SUSCEPTIBILITY GENES

A. Human studies

1. Noise-induced tinnitus

To our knowledge, the largest GWAS on tinnitus published until now included fewer than 200 subjects, identified no significant genes of p-value < 5.0e-08, was too small to separate etiologies, i.e., noise-induced versus age-related, and explained only 3.2% of the SNP variance (h2; Gilles et al., 2017), leaving unanswered the question of where the remainder of the 40% general heritability is, as noted in twin studies (Bogo et al., 2017). As an answer to this conundrum, it has become increasingly clear that common disorders, including cardiovascular disease, hypertension, breast and prostate cancer, and type 2 diabetes, involve a large number of genes, each with a small effect in the range of less than a 1.5 odds ratio (OR; Fransen et al., 2015; Timpson et al., 2017; Torkamani et al., 2018). A PRS attempts to quantify these small effect sizes and includes a wider number of genes, and consists of an individual's summation of their genotype weighted by effects sizes on the trait of interest. In order to attempt to identify as much as possible of heritability variation, typically, PRS is calculated on a large independent data set, and then used as a predictor in a target population in regression analysis for verification.

In the case of NIHL and tinnitus, ideally, a PRS would target an individual's susceptibility to noise prior to exposure, so that this person could be counseled and afforded extra hearing protection and/or the opportunity to change professions to an environment less prone to this environmental exposure. As an example in other medical fields, PRS has been applied successfully to predict pre-morbid disposition to Alzheimer's disease (Logue et al., 2018) and serves to identify susceptibility as a latent variable prior to exposure. In addition, PRS has been successfully combined with clinical risk factors, including elevated cholesterol, to advise earlier treatment with statins, and has been shown to lower risk of first heart attack (Timpson et al., 2017). PRS helps to answer questions about “missing heritability,” that is, the difference between H2 and h2 heritability. For instance, H2 analysis identified 0.43 tinnitus heritability in a large population, while GWAS analysis in the UKB cohort identified only 0.137 h2. PRS thus provides an estimate of individualized vulnerability to a trait, and PRS for common disorders appears to be equivalent to the risk of a monogenic disorder (Khera et al., 2018).

2. NIHL

To our knowledge, the only human GWAS study in the literature to specifically examine NIHL identified a SNP in the gene nucleolin (Grondin et al., 2015). This small study awaits replication. Before the advent of large population studies, cochlear injury, as evidenced in both ARHI and NIHL in mice, were used to search for relevant genes and pathways. Because of significant lab findings, approaches to finding pro-NIHL genes in humans initially targeted genes involved in protection, repair, inflammation, redox chemistry, or ion homeostasis.

Replicated findings in candidate gene studies have included SNP variants of genes encoding antioxidants catalase (CAT), superoxide dismutase type 2 (SOD2), paraoxonase 2 (PON2), and glutathione transferases GSTM1 and GSTT1 (Abreu-Silva et al., 2011; Carlsson et al., 2005; Fortunato et al., 2004; Yang et al., 2015). A candidate gene polymerase chain reaction (PCR) study found a significant SNP in the inflammatory gene interleukin-6 (IL-6) related to age and noise-exposure (p = 0.004; Doi et al., 2015). Pawelczyk et al. (2009) found KCNE1 to be associated with tinnitus independent of hearing loss, although this study has not been replicated. Sand et al. (2007) reported that GDNF and BDNF jointly predicted tinnitus severity in women only (p = 0.04). Also studied in candidate genes included stress-related protein chaperones HSP70–1 and HSP70–2, and K⁺ channel genes KCNQ1/KCNE1, KCNQ4, and KCNJ10 (Clifford et al., 2016). The latter serve to gate K⁺ flux into scala media (KCNQ1/KCNE1 complex), K⁺ efflux from outer hair cells into scala tympani (KCNQ4), and K⁺ efflux from strial intermediate cells (KCNJ10). Further implicated were gap junctional proteins within the organ of Corti and lateral wall GJB1, GJB2, and GJB4, which may help prevent K⁺ from accumulating within the organ of Corti during periods of intense sound. While some types of injury-promoting genes might be anticipated, disease-causing genes often follow no convenient line of reasoning. Examples in the case of NIHL include protocadherin 15 (PCDH15, DFNB23), which mediates calcium-dependent cell-cell adhesion, and MYH14 (DFNA4), which is a myosin isoform (Konings et al., 2009c). Some studies have supported nonlinear (epistatic) interactions among multiple variants (Clifford et al., 2016). Such interactions are difficult to detect, but are likely to help explain how many loci that may contribute only slightly to NIHL risk by themselves appear more significant in some individuals. Not considered further here are genes and alleles that may promote medical conditions that may in turn promote NIHL, such as hypertension and hypercholesterolemia (Starck et al., 1999). These studies all included less than 1000 subjects, relied on already-studied mouse models, and to date have not been replicated in unbiased GWAS studies nor in other ancestries.

An example of the Manhattan plot associated with the hearing loss question administered is shown in Fig. 2, and demonstrates a large number of significant hits. While the UKB has published summary data and identified genes associated with hearing loss, whether these are specific for age-related loss or will overlap with those that promote NIHL is currently unknown and awaits analysis combined with noise-exposure data.

FIG. 2.

(Color online) Manhattan plot of question on hearing loss from the United Kingdom cohort publicly available data. Participants answered “yes” or “no” to the question, “Do you have any difficulty with your hearing?.” Significant SNP hits are seen as a vertical line that extends above the accepted significance P-value of 5E-08. Significant genes are labeled, while those lines that are not labeled may indicate hits that fall outside of known genes.

B. Animal studies—Tinnitus and hearing loss

To date, only one study in mice has correlated a specific genetic mutation with tinnitus (Yu et al., 2016). Using a GPIAS to detect tinnitus, this study indicated that mice lacking the gene encoding the glutamate aspartate transporter (GLAST) had an increased propensity to develop tinnitus following salicylate exposure. Whether salicylate produces a pathophysiologic disorder consistent with noise-induced tinnitus is unclear, and the temporary bilateral hearing loss caused by salicylate is likely to pose confounds as well.

Alleles that act in a dominant manner are more prone to cause delayed, progressive hearing loss that may causally overlap with ARHI (Ohlemiller and Frisina, 2008; Ohlemiller, 2006). In the case of acoustic injury, the loss is acquired in adulthood in a probabilistic manner, so that statistical associations are sought using different analytical methods, and require much larger samples.

Chance features of commercial inbred mouse strains—and good educated guesses—led to the discovery of the first genes that promote NIHL. First among these was the Ahl allele of cadherin 23 (Cdh23^753G>A; Johnson et al., 1997; Johnson et al., 2000). This gene was the first identified QTL (Ahl) shown to impact both ARHL and NIHL, and ignited the “mouse revolution” in hearing. Cdh23 encodes a portion of the stereociliary tip link, which is subject to mechanical and other stresses that accumulate with age. It accounts for at least part of the hearing phenotype in many inbred mouse strains (Zheng et al., 1999), including C57BL/6, which is the most commonly used strain for much of biomedical research. The destabilizing process by which the Ahl allele operates gave rise to a host of studies in KO models for other genes predicted to exert similar destabilizing effects. Accordingly, positive results were found in KO models for antioxidant, regulatory, and repair-related genes (Ohlemiller, 2006). These studies laid the foundation for much human work, and yielded support for some of the genes listed in Sec. V. A genetic mapping study seeking to explain the apparent noise resistance of 129S6/EvTac and MOLF/EiJ inbred mice using intercrosses and chromosomal substitution (Street et al., 2014) yielded ten significant QTLs for noise resistance (nr1–nr10) but no confirmed genes. In addition, an intercross of four inbred strains (UM-HET4, featuring crosses of MOLF/Ei, C3H/HeJ, FVB/NJ, and 129/SvImJ) yielded significant QTLs on three chromosomes, but none of these has yet been identified (Schacht et al., 2012). Mouse GWAS have taken advantage of newly developed panels, such as the Hybrid Mouse Diversity Panel, in mapping of both hearing thresholds and NIHL. Most of these have addressed ARHL and are not described here (Friedman et al., 2009). The few NIHL extant GWAS have applied a fixed noise exposure to a few mice from many HDMP strains, and then compared strain distribution patterns in ABR threshold with dense marker sets from the same strains. Presently, a single gene nicotinamide adenine dinucleotide phosphate (NADPH oxidase 3, Nox3) has been identified and confirmed, at least for mice, using a conditional KO model (Lavinsky et al., 2015). In keeping with a theme in aging and noise injury of stress and redox genes, Nox3 impacts inflammation and the generation of reactive oxygen species.

Any phenotype can be probed by GWAS, and the intense interest surrounding cochlear synaptopathy has given rise to at least one study of noise-induced synaptopathy in mice (Lavinsky et al., 2018). Again using the HDMP, and ABR wave 1 magnitude as a readout for synaptopathy, Friedman and colleagues identified Cd44 and Slc1a2 as loci potentially influencing cochlear synaptopathy (Lavinsky et al., 2018). The former is expressed in pillar cells, while the latter may help remove glutamate from the inner hair cell afferent synapse. Gene identification was leveraged by eQTL data that placed transcription factor binding within the QTL region. The study requires replication and further testing of these genes and other candidates. One possible limitation is that wave 1 amplitude was measured at a fixed sound level, instead of a fixed sensation level. Threshold shifts can confound this type of analysis.

VI. CONCLUSIONS

Consistent phenotype definitions of both tinnitus and hearing loss will be important in comparing future large-scale studies of susceptibility to acoustic trauma. Because there is no recognized objective testing currently, tinnitus is constrained to self-report, and methods for detection and characterization in animals are controversial. Other phenotypic challenges include identification of a definition of hearing loss with the largest possible heritability measure, as well as the lack of exposure data constant across large cohorts.

Previously, candidate studies in humans of mutations that promote the complex traits of tinnitus and NIHL predominantly addressed exonic SNPs. Most candidate genes were identified in targeted studies based on expected roles for homeostatic, ion regulatory, antioxidant, and anti-inflammatory processes in the stressed cochlea as identified in mice. While candidate genes could be tested readily in mice using congenic lines, chromosomal substitution strains, and transgenic and KO lines, these studies in general did not lead to genes that were significantly related to NIHL or tinnitus in humans. In contrast, human GWAS in age-related hearing loss has identified predominantly intronic SNPs and awaits application to populations exposed to noise. Success for genetic information in noise-induced acoustic trauma in GWAS is limited by the need for large cohorts with consistent exposure histories that are also well controlled for variables such as age, sex, and duration of exposure.

Human and animal studies indicate that results will often contain multiple loci, each of which may promote only a small amount of disease risk or heritability. In fact, recent studies reveal that most common disorders, instead of being caused by monogenic anomalies, are composed of a large number of genes, leading to an accumulation of small effect sizes. A PRS might better characterize an individual's susceptibility to acoustic trauma, and this PRS has been used successfully to aid in the application of personalized medicine. The polygenicity of a disorder presents a quandary for therapies aimed at reducing risk, and the most effective therapeutic targets may be risk genes involved in epistatic interactions that amplify the risk associated with other loci.

GWAS for NIHL in mice have been performed based on the assumption that many orthologous genes and regulatory sequences promote the same pathology in humans and mice. However, regulatory sequences are not universally conserved between mice and men, and correlations must be analyzed cautiously. Thus, because most extant inbred strains reflect strong selection pressures, LD and mapping resolution may be more limiting for mouse GWAS than for human GWAS. Mouse eQTL studies have focused on the possibility that pro-risk loci are expected to include more regulatory sequences than conventional protein-coding genes, and human GWAS has born this out by identifying predominantly intronic and intragenic areas of significance.

Because of the large regulatory sequence differences between rodents and mankind, in the unique case of genes leading to common disorders such as noise-induced tinnitus and hearing loss, a reversal of protocols may be required. Studies are most likely to be relevant to humans if the gene(s) of interest can be shown to be expressed in the same cell types.

Prior to this time, animal research drove human research. However, with the current level of expertise in genomics, in order to identify relevant genes, human GWAS may first provide an unbiased survey of significant SNPs with their effect sizes. Identification of significant human genes may then direct animal studies aimed at clarification and molecular analysis.