Integrating pharmacogenomics into clinical trials of hearing disorders

PGx is the study of the role of genomic variation in drug response. This encompasses the effects of genetic variation on drug efficacy, metabolism, and toxicity (Ma and Lu, 2011). Ultimately, the goal is to identify the best candidate drug for a given individual and optimize pharmacotherapy with the minimum number of adverse events. Patients vary widely in their response to standard therapies, and this can lead to ineffective treatments or even life-threatening adverse events, which are a major barrier to successful pharmacotherapy (Schork, 2015). In recognition of the urgent need to optimize the current health care system through individualized therapy, the Obama administration launched the Precision Medicine Initiative in 2015. In contrast to a one-size-fits-all approach, precision medicine considers inter-individual genetic variation and response to a specific treatment, seeking to tailor the treatment for subgroups of patients (Collins and Varmus, 2015). Major technical advances over the past two decades have dramatically decreased the cost of sequencing and genotyping (Hodson, 2016), which has propelled the field of PGx (Wadelius and Alfirevic, 2011).

PGx has already had an impressive impact on health care (Roden and George, 2002). To date, 342 of 1800 U.S. food and drug administration (FDA)-approved drugs have PGx information that provides gene- or allele-specific recommendations for therapeutic indication and usage, adverse reactions, dosage, precautions, and/or contraindications on their labels, and of these, 230 medications have actionable PGx markers, for which genetic information should or could be used to guide the prescription of the relevant drug (Fig. 1) (Center for Drug Evaluation and Research, 2021; Kinch , 2014). Of note, these clinically actionable pharmacogenetic markers are highly prevalent in patients. The Vanderbilt PREDICT cohort with 10 000 patients found that 91% of all genotyped patients and 96% of African American patients carried at least one clinically actionable variant (Van Driest , 2014). Moreover, one or more clinically actionable variants were found in 96.19% of the 5000 subjects in the eMERGE‐PGx study (Bush , 2016), in 99% of the 1013 subjects in the RIGHT protocol (Ji , 2016), and in 99% of the veterans who use the Veterans Health Administration in the United States (Chanfreau-Coffinier , 2019). A few examples of drug-gene pairs with demonstrated clinical utility, such as trastuzumab and HER2, abacavir and HLA-B, and clopidogrel and CYP2C19, are summarized below as illustrative examples of current PGx labeling, which may be informative for future clinical studies in treating hearing disorders.

The first drug marketed with PGx data, trastuzumab, is a monoclonal antibody targeting the human epidermal growth factor receptor 2 (HER2), which is overexpressed in about 25% of breast cancers. In the activity-estimating trials, the efficacy of trastuzumab was minimal without stratification by HER2. However, significant clinical benefit has been shown when limited to the patients with HER2-positive breast cancer (Cobleigh , 1999; Slamon , 2001). With the enrichment strategy, it soon gained FDA approval for use in combination with a HER2 test for treating HER2-overexpressing breast cancer. To date, trastuzumab combined chemotherapy has gained favor in clinical practice for HER2-positive breast cancer patients (Perez , 2014).

Abacavir is a nucleoside reverse transcriptase inhibitor indicated for the treatment of human immunodeficiency virus (HIV) infection as part of a highly active antiretroviral therapy. Although highly efficacious, ∼5–8% of patients experience hypersensitivity reactions during the first 6 weeks of abacavir treatment (Martin , 2012). Multiple independent studies have demonstrated the association of the HLA-B allele HLA-B*57:01 and abacavir-induced hypersensitivity reactions in Caucasians (Hetherington , 2002; Mallal , 2002; Rauch , 2006) and African Americans (Saag , 2008). Moreover, the results from PREDICT-1 (Mallal , 2008), the first multi-center, prospective, randomized, double-blind trial of a genetic test to reduce hypersensitivity reactions, demonstrated that HLA-B*57:01 genotyping-guided treatment dramatically reduced abacavir-induced hypersensitivity reaction compared to standard-of-care. The convincing evidence has prompted the FDA to provide a black box warning that recommends genetic testing prior to abacavir use, which is an excellent example of PGx being integrated into routine medical practice to reduce adverse drug reactions (Martin , 2012).

Clopidogrel is widely prescribed for acute coronary syndromes for patients undergoing percutaneous coronary intervention. However, a considerable number of patients do not respond to clopidogrel effectively (Angiolillo , 2007; Gurbel , 2005). It is formulated as a prodrug that requires a two-step activation to form its active metabolite, which then exerts an antiplatelet effect by inhibiting the adenosine diphosphate P2Y12 receptor. The activation is catalyzed by cytochrome P450 2C19 (CYP2C19) and several other CYP enzymes. The association between clopidogrel response and CYP2C19 genotypes has been extensively reported (Brandt , 2007; Harmsze , 2010; Mega , 2010; Shuldiner , 2009; Sibbing , 2010; Zabalza , 2012) and further validated through a recent large prospective randomized trial (Claassens , 2019). These studies demonstrated that a PGx-guided clopidogrel therapy is superior to reduce bleeding risk relative to conventional therapy with ticagrelor and prasugrel. Given the solid evidence indicating the benefit of PGx-guided clopidogrel therapy, a number of institutions have implemented this approach in clinical practice (Shahin and Johnson, 2013).

There are currently no FDA-approved drugs on the market for hearing disorders such as tinnitus, age-related hearing impairment (ARHI), or noise-induced hearing loss (NIHL), which collectively affect approximately 1 in every 5 individuals globally (McCormack , 2016). The lack of drugs is not for lack of effort, as numerous clinical studies have been conducted over the past several decades. Rather, the large number of disappointing outcomes from these clinical trials likely reflects the challenges of drug development to treat each symptom of hearing disorders, which may consist of different pathological conditions. For example, tinnitus is a symptom with several different underlying pathologies, reflecting the heterogeneity of the condition (Tunkel , 2014). Thus, it is not unexpected that drug therapies were not successful in large clinical trials without subtyping (Asnis, 2020; Piccirillo , 2020). In fact, in most clinical trials for tinnitus, patient populations have not been stratified based on tinnitus pathology. Perhaps one of the best examples can be illustrated by the use of carbamazepine for the treatment of “typewriter tinnitus” (Levine, 2006; Mardini, 1987). In typewriter tinnitus, patients often refer to their tinnitus as a staccato clicking much like a typewriter. In a recent review of case studies, Sunwoo (2017) revealed that 22 of 22 patients suffering from typewriter tinnitus experienced immediate relief from short-term treatments of carbamazepine (CBZ). The authors further conclude that responsiveness to low-dose and short-term CBZ treatment could be a primary diagnostic for neurovascular compression of the cochlear nerve (NVC-C) (Sunwoo , 2017). Interestingly, a different HLA-B allelic variant from the one discussed above for abacavir sensitivity, HLA-B*15:02, has been associated with CBZ-sensitivity (Fang , 2019). Thus, pre-screening patients for HLA-B*15:02 prior to assessing CBZ responsiveness to the drug could reduce adverse events in the use of CBZ to treat typewriter tinnitus. In addition, PGx data may serve as an important bridge in identifying patient populations most likely to benefit from a specific drug treatment. For example, when the efficacy of CBZ was tested in a clinical study as a therapy for a broader class of tinnitus sufferers, those experiencing chronic non-pulsate tinnitus, the effects were not significantly different from a placebo control (Gerami , 2012). Thus, despite a 100% success rate for typewriter tinnitus, the drug fails as a general therapy for tinnitus (Gerami , 2012; Hulshof and Vermeij, 1985). Thus, by first stratifying a patient population using either PGx or pathological markers, it may be possible to use specific subtyping as an inclusion or exclusion criterion in the design of the clinical study for hearing disorders. For instance, patients with biomarkers for microvascular compression (Lassaletta , 2019) could be prescreened as an inclusion criterion for a study of the CBZ-mediated treatment of tinnitus.

Pharmacogenomic studies of aminoglycoside and cisplatin-induced hearing loss have also shown promise in recent years (Lee , 2016; McDermott , 2021; McDermott , 2022; Mukherjea and Rybak, 2011; Tserga , 2019). For instance, at least eight single nucleotide polymorphisms (SNPs) from five genes in repeated studies have shown significant associations with cisplatin ototoxicity. These genes are associated with anti-oxidant regulation, neurotransmission, or auditory function and can serve as a foundation for developing diagnostics tests for susceptibility to ototoxicity (Tserga , 2019). Studies of aminoglycoside toxicity have identified a single mitochondrial variant (m.1555A>G) present in the gene encoding the 12s rRNA subunit. This variant likely acts by predisposing the human mitochondrial 12s rRNA to aminoglycoside toxicity in much the same way the drug targets the bacterial 16s rRNA subunit, providing mechanistic insight into the mode of action of the drug. Importantly, a rapid diagnostic test has also been developed enabling a tailored antibiotic regime to those carrying the sensitive allele (allele frequency of approximately 1:500), thus, avoiding aminoglycoside hearing loss in a significant portion of the population (McDermott , 2021; McDermott , 2022).

NIHL is the most common form of work-related sensorineural hearing loss (Conway , 1993). It is estimated that close to 900 million adults will suffer from disabling hearing loss by 2050 and over 1 billion children are at risk for permanent hearing loss due to unsafe listening practices (Olusanya , 2019). NIHL is one of the leading causes of hearing impairment globally and is a complex trait determined by both genetic and environmental variables (Le Prell and Bao, 2012; Sliwinska-Kowalska and Pawelczyk, 2013). The etiology of NIHL reflects this complexity with multiple anatomical, physiological, and cellular components contributing to the pathology (Guthrie and Bhatt, 2021; Kobel , 2017; Kurabi , 2017).

One of the greatest challenges in developing robust therapies to treat or prevent NIHL is that noise exposure may elicit a range of cellular and molecular damage based on its intensity and duration. This was elegantly demonstrated in a comprehensive study of NIHL in mice revealing a clear distinction between cochlear pathologies based on noise intensity that ranged from a selective damage of only hair cell bundles to a loss of both outer hair cells (OHCs) and inner hair cells (IHCs) due to the rupture of the Organ of Corti (Wang , 2002). Additional animal studies clearly demonstrated the potential for cochlear synaptic loss without damage to other cochlear structures when the sound level is not extremely loud [for a recent review, see Kujawa and Liberman (2019)]. Since the rodent cochlear structure and many molecular signaling pathways are similar to those in humans, we have proposed at least four unique NIHL subtypes, from hair bundle damage to loss of IHCs, that underlie NIHL (Fig. 2). It would be difficult to determine whether similar NIHL subtypes are present in humans, although identification of biomarkers such as prestin in the blood may provide an avenue for testing this hypothesis (Emre , 2022; Iliadou , 2021; Parker , 2022). Human studies of possible interactions between NIHL and aging did suggest at least the presence of certain subtypes, although a mixed loss of cochlear synapse and OHCs is common in the aging population (Wu , 2020a; Wu , 2021). For example, cochlear synaptic loss is present at all cochlear frequencies during aging and was exacerbated by NIHL. OHC loss is also greatly exacerbated at high frequencies by NIHL (Wu , 2019; Wu , 2021). Certain age-related hearing loss (ARHL) pathologies may be different from NIHL. For example, IHC loss during aging is mainly observed at high frequencies and unaffected by noise at either low or high frequencies (Wu , 2021). Thus, to determine whether these four subtypes of NIHL are present in humans, it would be useful to develop non-invasive functional assays to identify these different NIHL subtypes so that subtyping of NIHL can be used in the design of future clinical studies. Indeed, studies have shown that loss of hair cells can be independent of loss of spiral ganglion neurons or loss of efferent synapses during aging (Han , 2006; Jin , 2011), which strongly suggests the need to develop treatments targeting specific subtypes of cellular damage.

One promising avenue to subtype NIHL pathologies is to develop novel non-invasive functional assays from standard otological diagnostic tests. Pure-tone audiometry is still the standard clinical tool for determining hearing loss (Rabinowitz, 2000) and is subject to large error estimates associated with the need for behavioral responses from the patient or other variables in the procedure (e.g., ±2–5 dB shifts). Furthermore, standard audiometric measurements often do not extend to high frequencies where ARHL and most NIHL hearing loss are first detected. Distortion product otoacoustic emissions (DPOAEs) are now often used in clinical trials to monitor the OHC function. Non-invasive measurements of hearing, including auditory brainstem response (ABR) and electrocochleography (ECochG), can be used to detect cochlear synaptopathy in the absence of OHC damage in animal models, although outcomes of human studies are not yet conclusive (Harris and Bao, 2022). ECochG features are determined primarily by the peripheral auditory system, presenting two prominent waves: the summating potential (SP), which is a direct current generated by receptor potentials of IHCs, and the compound action potential, which is generated by action potentials at the distal auditory nerve (equivalent to wave I of the ABR) (Gibson, 2017). ABR and ECochG measures are similar between humans and rodent species with slight discrepancies in waveform morphology and latencies (Akil , 2016). Clinically, ABR/ECochG evaluations currently involve simplified analyses of response latencies and peak amplitude or ratios between different wave peaks (i.e., latency difference between wave I and wave V). However, high variability of peak amplitudes dramatically reduces the sensitivity of these diagnostic measurements. One solution is to improve ABR/ECochG data collection and analysis in animal models with identified NIHL pathological types and then apply them for human diagnosis. Interestingly, machine learning algorithms have been deployed as an effective statistical technique to identify multiple features associated with complex phenomena, including those in auditory research (e.g., Bramhall , 2018; McKearney and MacKinnon, 2019). As discussed in this issue (Harris and Bao, 2022), artificial intelligence-based algorithms could be trained to identify novel features from standard audiological assessments, such as ABR/ECochG waveforms associated with specific cochlear pathology, such as cochlear synaptopathy.

Current limitations of non-invasive human diagnostic tests to define underlying noise-induced cochlear damage limit efforts to subtype NIHL based on genome-wide association studies (GWASs). A lack of sensitivity in hearing assessment methodologies also reduces our chances at drawing associations between genotypic variation and patient response to drug candidates. Furthermore, a lack of standardization in hearing assessments limits our ability to cross-reference and validate PGx outcomes. For example, in one of the earliest comparisons of NIHL–gene associations, 23 loci were associated with NIHL in one of two studies of Swedish (Carlsson , 2004; Carlsson , 2005) and Polish (Śliwińska-Kowalska , 2006) workers exposed to high noise work environments. However, SNPs in only two of these genes were found to be significantly associated with NIHL in both populations (Konings , 2009). Although there are many potential confounding variables in cross-study comparisons, the use of different methodologies to compare hearing thresholds certainly is a major contributor to this variation. In the Swedish study, populations were stratified based on hearing thresholds at 3 kHz, whereas stratification in the Polish study was based on hearing assessments at 4 and 6 kHz, to better reflect the demographics of each of the respective groups (Konings , 2009). Thus, hearing assessments were not directly comparable in the two studies, and more importantly, the underlying pathologies associated with long-term hearing loss at 3 kHz may be fundamentally different from the cellular mechanisms underlying hearing loss at 4 or 6 kHz (Taylor , 1965).

In addition to improving diagnostic hearing assessments to subtype NIHL based on the underlying pathology, we must also develop new biomarkers for clinical studies. Very few clinical trials for hearing disorders have incorporated biomarkers as part of the study design. However, based on animal studies (Bao , 2013), Bao and colleagues identified several promising drug repurposing candidates to prevent and treat NIHL. A phase II clinical study is now underway to monitor the effectiveness of zonisamide (ZNS) in NIHL prevention and treatment (see NCT047685569). As a component of this program, patient DNA and RNA samples will be surveyed to identify novel genetic associations and miRNA biomarkers, respectively, to monitor drug efficacy and characterize the cellular damage associated with surgical drilling. These data should be able to better inform both inclusion and exclusion criteria for future clinical studies as well as inform drug labeling should an association between drug efficacy and patient genotype be revealed. ZNS, a voltage-dependent sodium and T-type calcium channel blocker (Biton, 2007; Kito , 1996), is primarily metabolized through the action of liver-localized cytochrome P450 family member CYP3A4 (Leppik, 2004; Martínez-Ávila , 2018; Nakasa , 1996). It is also rapidly absorbed after oral administration with a bioavailability of near 100% after single dose administration with a peak concentration at 3.2 h and a long half-life of 52 h (Kochak , 1998). Significant inter-subject variability in C_max suggests that genotypic variation likely affects both absorption and elimination rates. Thus, drug efficacy may be determined in part by genetic variation at loci encoding cytochrome P450 enzymes (Nakasa , 1996; Okada , 2008). Additional genetic variation may include loci necessary for drug absorption, distribution, metabolism, and excretion (ADME) (Innocenti , 2011; Schadt , 2008; Schröder , 2013). Although these data will likely not inform labeling requirements due to the limited power of the sample size, they can be extremely helpful in defining subsequent clinical investigations for this drug or other candidates if gene or miRNA associations are drawn. It may also be possible to identify patient populations most likely to benefit from drug therapies through associations of genetic variation linked to NIHL.

Characterizations of diverse collections of mice have revealed a number of genes that may underlie genetic predisposition to hearing loss after short-term exposures to loud [e.g., >100 dB sound pressure level (SPL)] noise (Crow , 2015; Lavinsky , 2016; Myint , 2016). These studies have identified a number of potential candidate genes that may underlie differential sensitivities associated with noise exposure in human populations. Human twin studies have also revealed a genetic component to noise susceptibility (Heinonen-Guzejev , 2005). However, given the potential for temporary threshold shifts to lead to hidden hearing loss (Kujawa and Liberman, 2009), intentional short-term exposures of high noise treatments in human populations is deemed unethical (Bramhall , 2021; Ryan , 2016), and thus, similarly controlled GWAS experiments as conducted in mice cannot be performed in human populations. To circumvent these direct interventions, several groups have conducted genome-wide associations to identify loci that may predispose individuals working in high noise environments to hearing loss (e.g., Jiang , 2021; Konings , 2009; Niu , 2021; Śliwińska-Kowalska , 2006; Sliwinska-Kowalska and Pawelczyk, 2013). A recent catalog of genes identified through these association studies and others has recently been assembled (Mao and Chen, 2021). As highlighted in Table I, many independent populations have now been studied, and several genes appear to mediate sensitivity across ethnic groups. However, as noted previously, single gene association studies are inherently problematic to extrapolate to broader conclusions. This point is highlighted by a recent meta-analysis of dozens of studies that failed to detect a significant association between loss-of-function alleles of GSTM1 and GSTT1 and increased risk of sensorineural hearing loss (Zong , 2019). In addition, the vast majority of these studies were conducted on largely male populations and, thus, do not capture variation in women that may contribute to enhanced or reduced risk of NIHL [e.g., see Kim (2010)]. Below, we have focused on a few case studies to highlight some innovative approaches to mine genome sequence data for associations with NIHL.

In a recent study of Chinese shipbuilders exposed to loud noise environments, Jiang (2021) used machine learning to stratify workers into low and high risk groups based on a number of demographic and potential risk factors (age, sex, cumulative noise exposure, smoking and drinking status). They then identified the outliers in each group as potentially more susceptible or resistant to NIHL. For instance, an individual classified as low risk, but with severe hearing loss, would be considered an extreme phenotype who is susceptible to NIHL. Through this extreme phenotyping, a subset of susceptible and resistant individuals were identified, and their DNA was subject to whole exome sequencing. These data revealed significant associations of alleles of two genes, CDH23 and WHRN, that were independently validated in a larger cohort. Thus, through extreme phenotyping, this study highlights how studies conducted with relatively small population sizes can provide new insight into the pathology of NIHL and can potentially identify allelic variants that will serve as inclusion or exclusion criteria in clinical trials.

Another informative case study involves surveys of catalase (CAT) polymorphisms in Chinese factory workers exposed to loud noise environments. In a comprehensive study of Swedish workers, no associations of CAT with NIHL were detected (Carlsson , 2005). However, when more markers were developed to enable haplotype screening across the entire CAT locus, associations were detected with NIHL that were dependent on noise exposure level (Konings , 2007; Yang , 2015). Importantly, the alleles that displayed associations in the Swedish and Polish populations were present at very low frequencies in the Chinese Han population study (Yang , 2015), requiring the development of new marker sets. The use of whole exome sequencing or low pass genomic sequencing to survey population variation could enable a much more detailed assessment of gene–NIHL associations and provide a more comprehensive survey of ethnic variation.

Biomarkers to monitor cellular damage within the cochlea will provide valuable tools for associating specific cellular damage with NIHL. One promising class of biomarkers are miRNAs that regulate the transcriptional stability or translation of mRNA and play essential roles in development, differentiation, and diseases of the inner ear (Friedman , 2009; Geng , 2018; Hu , 2018; Ushakov , 2013) and have been implicated in the regulation of oxidative stress associated with NIHL (Forouzanfar and Asgharzade, 2020; Miguel , 2018; Prasad and Bondy, 2017). This includes circulating blood miRNAs that can be more readily accessible as biomarkers for clinical studies (Ding , 2016; Ha , 2020; Lee , 2018; Li , 2017; Miguel , 2018). In a recent study of idiopathic sudden sensorineural hearing loss (ISSNHL), Ha (2020) compared peripheral blood samples from 22 patients and controls to identify miRNAs associated with this sudden onset hearing loss. They monitored the expression of six miRNAs (miR-23a, miR-34a, miR-15a, miR-18b, miR-143, and miR-210) previously associated with ISSNHL (Li , 2017) and two (miR-205 and miR-183) from a mouse study of ototoxicity (Lee , 2018) using quantitative PCR. Two of the miRNAs examined were expressed at significantly higher levels in the ISSNHL patients relative to healthy controls (miR-183 and miR-210), and three were expressed at levels significantly lower than healthy controls (miR-18b, miR23a, and miR143). Of these potential markers, four of them (miR-183, miR-210, miR-15a, and miR-18b) were correlated with both the occurrence and treatment of ISSNHL and thus could serve as markers for potentially predicting the severity of ISSNHL and prognosis for treatment (Ha , 2020). Further confirmation, however, will be necessary to determine how robust these markers are across a broader demographic and whether they would be valuable as markers for other forms of hearing loss.

The field of PGx is rapidly advancing in part due to the proliferation of inexpensive genome sequencing enabling scientists to draw associations between genotypic variation and drug response. However, a number of significant challenges remain for the development of drugs to prevent or treat hearing disorders. This includes the lack of universal standards for the diagnosis of hearing disorders and a range of methodologies that are currently used to assess hearing thresholds and the auditory signaling pathway. New technologies promise to not only increase the sensitivity of detection but also to reduce subjectivity and operator error, enabling a more precise quantification of hearing ability. This is particularly important in studies of hearing loss that have to date been largely restricted to retrospective studies of individuals subjected to noise environments over a number of years or multi-year prospective studies where multiple confounding environmental signals often limit the statistical analysis of the study. Assessments that can detect the first molecular signatures of hearing loss (e.g., high frequency hearing loss) through both biomarkers and improved methods of detection should facilitate genotype by phenotype associations. Importantly, the incorporation of PGx in early phases of clinical studies has the potential to better inform downstream studies and potentially identify subpopulations that selectively benefit from the candidate drug.

T.P.B. and X.W. contributed equally to this work. The work presented here was supported by the National Institute on Deafness and Other Communication Disorders of the National Institutes of Health under Grant Nos. DC017406-01 and DC018759 (J.B.) and also by the U.S. Department of the Army Grant No. W81XWH19C0054 (J.B.). The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health and the U.S. Department of the Army. We thank Dr. John Hawks and an anonymous reviewer for helpful criticisms and comments on the manuscript. T.P.B. and J.B. are employed by Gateway Biotechnology Inc. for this project. X.W. declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.