Clinical and Genome-wide Analysis of Cisplatin-induced Tinnitus Implicates Novel Ototoxic Mechanisms

Approximately 9.6% of the U.S. population experiences tinnitus, the phantom perception of sound characterized by persistent ringing, buzzing, beeping, or hissing in the ears (1). Several studies have linked tinnitus to increased anxiety, depression, insomnia, and reduced productivity and quality of life (QoL; refs. 2, 3). Nevertheless, the pathogenic mechanisms of this disorder remain poorly understood (4), and no pharmacologic agents are approved to treat tinnitus (5); treatment is typically limited to cognitive behavioral therapy and management of associated anxiety (5, 6). Despite its etiologic ambiguity, several risk factors have been shown to contribute to the development of tinnitus, including age, hearing loss, and the administration of ototoxic drugs (7, 8).

Cisplatin is one of the most commonly used chemotherapeutics, and is known for its potent ototoxic effects. It can lead to irreversible bilateral hearing loss in up to 80% of patients, with many experiencing permanent tinnitus (9–11). Although few studies have rigorously assessed tinnitus following cisplatin-based chemotherapy, approximately 40% of patients appear to experience at least some degree of tinnitus (6). This is substantially higher than the rate observed in the general U.S. population, and in patients with testicular cancer who did not receive cisplatin-based chemotherapy (12%; ref. 9). Severe cases of tinnitus are also markedly increased in cisplatin-treated patients when compared with the general population (22% vs. 1%–2%; refs. 4, 9). One study of 1,409 testicular cancer survivors (TCS) noted that 19% of TCS treated with 100–400 mg/m² cisplatin exhibited symptoms of moderate/severe tinnitus, with that prevalence reaching 42% in the dose-intensive cisplatin chemotherapy group (12), suggesting a dose-dependent effect. It has been previously demonstrated that cancer survivors with neuropathy, hearing loss, and tinnitus were more likely to report poorer quality of life (QoL) than those with neuropathy alone (13). Another investigation confirmed worse perceived stress among cancer survivors with tinnitus (14).

Despite this documented effect on QoL, little is known about the risk factors and mechanisms of cisplatin-induced tinnitus (CisIT). The subjective nature of the disorder and lack of model systems hamper the elucidation of its pathophysiology. While genotype–phenotype associations promise to identify underlying molecular alterations, to our knowledge, no genome-wide association study (GWAS) of CisIT has yet been conducted. We previously characterized ototoxicity in our North American cohort of TCS receiving homogeneous cisplatin-based treatment. Among 488 TCS, 40% experienced some degree of tinnitus and 15.8% reported moderate/severe tinnitus. Furthermore, tinnitus was significantly correlated with reduced hearing at each frequency examined (P < 0.001; ref. 9). In this study, we comprehensively evaluate the role of clinical characteristics and modifiable risk factors on the development of tinnitus after cisplatin-based chemotherapy in a cohort of 762 TCS, and identify genetic determinants of CisIT through GWAS. In addition, we conduct pathway analyses and in vitro experiments to validate and contextualize our results.

All patients were enrolled in the Platinum Study, which includes eight cancer centers in the United States and Canada. Eligibility criteria included: men diagnosed with histologically or serologically confirmed germ cell tumor (GCT), age <55 years at diagnosis and >18 years at enrollment, treatment with first-line cisplatin-based chemotherapy regimen for either initial GCT or recurrence after active surveillance, with no antecedent chemotherapy for another primary cancer, no salvage chemotherapy, and routine follow-up at the participating site. All patients were disease-free at the time of clinical evaluation and provided written consent for study participation, access to medical records, and genotyping. Study procedures were approved by the Human Subjects Review Board at each institution. The studies were conducted in accordance with the U.S. Common Rule.

Patient data were either ascertained at clinical follow-up or abstracted from medical records according to a standardized protocol (Supplementary Fig. S1). Data abstracted from medical charts included: age at GCT diagnosis, GCT tumor characteristics, treatment regimen with cumulative dose and number of cycles, and BMI at the initiation of treatment. Data collected at clinical evaluation included age and BMI, audiometric studies, self-reported questionnaires, and genotyping. Pure-tone air conduction thresholds were measured bilaterally on a subset of 602 survivors at the frequency range 0.25–12 kHz to quantify hearing. The geometric mean of hearing thresholds across cisplatin-affected frequencies (4–12 kHz) was used to quantify cisplatin-induced hearing loss as described previously (9). Self-reported ototoxicity and neurotoxicity were assessed with two validated questionnaires: the EORTC-Chemotherapy Induced Peripheral Neuropathy 20 item questionnaire (CIPN20) and the Scale for Chemotherapy-Induced Neurotoxicity (SCIN; ref. 15).

Hypertension, hypercholesterolemia, use of psychotropic medication (antidepressants, anxiolytics, and antipsychotics), overall health, tobacco use, alcohol consumption, and noise exposure were also ascertained at follow-up as patient-reported outcomes. Hypertensive subjects indicated receiving a diagnosis of high blood pressure and taking prescription medication for it at evaluation. Hypercholesterolemia subjects answered “yes, current” to the question: “Have you ever taken prescription medication for high cholesterol?” The use of psychotropic medication was assessed as in Fung and colleagues (16), and was based on whether a patient had been taking psychotropic medication for at least one month at the time of clinical evaluation. Self-reported health was rated on a poor–excellent scale in response to the question “How would you rate your health?” Values were numerically converted for association testing (poor/fair = 1, good = 2, very good = 3, excellent = 4). Exposure to noise was assessed using two questions: “Have you ever had a job where you were exposed to loud noise for 5 or more hours a week? (loud noise means noise so loud that you had to speak in a raised voice to be heard)” and “Outside of a job, have you ever been exposed to steady loud noise or music for 5 or more hours a week? (examples are noise from power tools, lawn mowers, farm machinery, cars, trucks, motorcycles, or loud music)”. After numeric conversion (Yes = 1, No = 0), overall noise exposure was computed as the sum of noise exposure in and outside of work. Alcohol consumption was assessed as the self-reported response to the question “During the past year, how many drinks of alcoholic beverage have you consumed on average? [1 drink = 12 oz. beer (1 can or bottle), 4 oz. glass of wine, 1 mixed drink or shot of liquor]” with the following options: rarely/never (0), 1–3/month (1), 1/week (2), 2–4/week (3), 5–6/week (4), 1/day (5), 2–3/day (6), 4–5/day (7), and 6+/day (8). Tobacco use was assessed as the response to the following two questions: “Have you EVER smoked cigarettes?” with “Yes” (1) and “No” (0) options and “Do you CURRENTLY smoke cigarettes?”

Genotyping was done as described previously (17, 18). DNA was extracted from peripheral blood mononuclear cells. SNPs were genotyped in two phases at the RIKEN Center for integrative Medical Science (Yokohama, Japan). Set 1 was genotyped on the HumanOmniExpressExome-8v1-2_A chip (Illumina) and set 2 was genotyped on the InfiniumOmniExpressExome-8v1-3_A chip (Illumina). Samples were plated with inter- and intra-plate duplicates at random. Standard quality control (QC) was implemented in PLINK, and is outlined in Supplementary Fig. S2. Sample-level QC criteria included: sample call rate > 0.99, pairwise identity by descent < 0.125, coefficient of inbreeding F < 6 SDs from the mean, and genetically European as determined by principal component analysis (performed using SMARTPCA). SNP-level QC included: call rate > 0.99, and Hardy–Weinberg equilibrium (χ²P > 1 × 10⁻⁶). QC was first performed in each set independently and then the two sets were merged followed by additional QC. QC for batch effects was performed using a batch pseudo-GWAS where variants associated with batch pseudo-phenotype (10 genotypic PC-adjusted P < 10⁻⁴) were excluded. SNPs and samples passing QC criteria comprised the input set for imputation with EAGLE phasing, performed on the Michigan Imputation Server using the Haplotype Reference Consortium (18–20). SNPs with imputation R² < 0.8, MAF < 0.01, and INFO scores > 1.05 or < 0.3 were excluded. Only subjects who passed QC were included in this study (Supplementary Fig. S2).

Using SCIN (15), TCS were dichotomized to tinnitus cases/controls based on responses to the question: “Have you had in the last 4 weeks: Ringing or buzzing in the ears?” Cases responded “quite a bit/very much”, and controls responded “not at all”. Those answering “a little” were not included. We labeled “not at all”, “a little”, and “quite a bit/very much” as none, mild, and moderate/severe tinnitus. TCS were also asked: “Do you have: ringing and buzzing in the ears?” Tinnitus cases responding “no” to this question were excluded from analysis.

GWAS was done in PLINK v1.9 (22, 23) with logistic regression assuming additive effects, with cumulative cisplatin dose, age at diagnosis, self-reported noise exposure, and the first 5 genetic principal components (SMARTPCA (23)) as covariates. We performed imputation of missing covariates to avoid sample exclusion. Eight individuals with missing age at diagnosis were assigned an imputed age calculated as age at clinical follow-up minus median follow-up duration (5.2 years). Twelve individuals were missing values for cumulative cisplatin dose received, but each of these patients had information on the cycles of cisplatin received. Because one cycle of cisplatin is equivalent to 100 mg/m², these patients were assigned an imputed cumulative dose by multiplying the number of cycles they received by 100 (i.e., 3 cycles × 100 = 300 mg/m²). No subjects were missing data on noise exposure. Genome-wide significance was set to P < 5 × 10⁻⁸. A list of expression quantitative trait loci (eQTL) was extracted from GTEx v7 using all 48 tissues with eQTL analysis (25). Conditional analysis was performed by adjusting for the lead single nucleotide polymorphism (SNP) as a covariate. SNP family enrichment was assessed using the empirical Brown combined P value (26). Narrow-sense heritability was estimated with a genetic relationship matrix variance component model as implemented by GCTA with the same covariates as GWAS (27). Pathway over-representation analysis was performed using Gene Ontology (28, 29) pathways with the g:Profiler package (30) at a significance threshold of FDR q < 0.05 after attributing genes the P value of the most significantly associated SNP in or within 25 kilobases of gene start/end sites (GENCODE GRCh37.p12).

OTOS expression data in different cancer cell lines was obtained from the Cancer Cell Line Encyclopedia (CCLE; ref. 31). Drug sensitivity to cisplatin and other antineoplastic agents, measured as the area under the dose–response curve, was obtained from the Genomics of Drug Sensitivity in Cancer Project (32). Spearman correlation and linear regression was performed between expression and sensitivity of cancer cell lines with nonmissing OTOS expression in R 3.3.2.

Data are presented with medians (ranges). Age was modeled continuously and in quartiles (ordinal). Cisplatin dose was treated continuously and in dose categories: <300, 300, 400, and >400 mg/m². Prevalence ratios (PR) were calculated to compare tinnitus prevalence by treatment or allele group, with normal approximation for 95% confidence intervals (CI) and χ²P values reported. Logistic models were constructed to assess associations (OR) with tinnitus in univariable and multivariable contexts. All tests were two-sided at a significance threshold of P < 0.05. Results are presented as OR/PR (95% CI). All statistical analyses were performed in R 3.3.2 and plotted with ggplot2 (33) unless otherwise specified.

HEI-OC1, a mouse auditory cell line derived from the inner ear organ of Corti, was cultured in complete media of DMEM (high glucose, l-glutamine and pyruvate) with 10% FBS (Corning Inc.). L929, a mouse fibroblast cell line, was cultured in complete media of EMEM with 10% FBS and 1% penicillin/streptomycin. Cells were passaged three times/week, plated at 1 × 10⁶ cells/T25 culture flask, and grown in a humidified chamber. HEI-OC1 was grown at 33°C with 10% CO₂ and L929 at 37°C with 5% CO_2. The authenticity of the cell lines was confirmed by IDEXX BioResearch by measurement for interspecies contamination or misidentification, Case # 26495-2018 (HEI-OC1) and #32923-2018 (L929). Cell lines were also assessed for Mycoplasma contamination using a MycoAlert Mycoplasma Detection Kit (Lonza) at the time of acquisition and prior to performing in vitro experiments.

OTOS protein vector, pOTOS (pPM-N-D-C-His; catalog no. PV401573) and blank control (pControl; PV001) vector were purchased from Applied Biological Materials, Inc., with the blank control vector sequence being verified by the manufacturer. The plasmids were amplified into competent DH5α bacteria (Life Technologies Corporation), expanded using 50 μg/mL kanamycin selection, and isolated using PureYield Plasmid Midiprep (Promega Corporation). Confirmation of OTOS cDNA sequence was performed at the University of Chicago DNA Sequencing Core.

After 24 hours of plating each cell line at 31,250 cells/cm² into 96-well plates, the cells were transfected in Opti-MEM (Life Technologies) with the addition of a complexed DNA mixture consisting of 1 part Lipofectamine 3000 (Thermo Fisher Scientific) with 3 μg pOTOS or pControl, as per manufacturer's instructions. Transfection continued for 24 hours. Cisplatin (Sigma-Aldrich) was dissolved in a darkened hood using 0.9% saline. The solution was vortexed for 1 minute, sonicated for 10 minutes with no heat, and then filtered through a 0.2-μm syringe filter to obtain a 10 mmol/L stock solution. Transfection media was exchanged for 5, 7.5, 10, 25, or 50 μmol/L cisplatin in complete media or vehicle control. At 48 or 72 hours postdrug treatment, 100 μL CellTiter-Glo (Promega Corporation) was added to each well, as per manufacturer's instructions. A total of 135 μL of each well was transferred to a white Costar assay plate (Thermo Fisher Scientific) and luminescence was measured on a Synergy-HT plate reader (BioTek Corporation). Four independent experiments were performed and within an experiment, each concentration was evaluated in triplicate. IC₅₀ values were calculated through nonlinear regression of the log cisplatin concentration versus normalized response curve using Prism6 software.

Cells were plated at 31,250 cells/cm² into 12-well plates for qPCR collections. After 24, 48, 72, and 96 hours of transfection (corresponding to 0, 24, 48, and 72 hours after cisplatin treatment), cells were pelleted following a brief trypsinization and stored at −80°C. RNA was isolated using the RNeasy Plus Kit (Qiagen) followed by reverse transcription of 200 ng RNA into cDNA with the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). Mouse Otos (Mm01292235_g1) expression was tested in HEI-OC1 and L929 cells, and was found to be minimally expressed. Expression of human OTOS was determined with qPCR using TaqMan primer OTOS (Hs00964785_m1) and compared with mouse Actβ (Mn00607939_s1), the endogenous control. Comparative ΔΔC_t method was used to determine fold change of human OTOS at each timepoint. Four independent experiments were performed and each sample was run in triplicate on a Life Technologies Viia7 PCR machine.

Of 1,029 TCS who passed GWAS QC criteria, 608 (59.1%), 265 (25.8%), and 154 (15.0%) reported none, mild, and moderate/severe tinnitus, respectively. Subjects were dichotomized to cases (moderate/severe) and controls (none). Two cases were excluded for inconsistent responses. Cohort characteristics are provided in Table 1. Median age at diagnosis and evaluation was 31 (15–54) years and 39 (18–75) years, respectively. Median follow-up time was 5.2 (1–37) years. A total of 91.6% were treated with either BEP (n = 444, 58.3%) or EP (n = 254, 33.3%). The remaining 64 (8.4%) received vinblastine (n = 6), carboplatin (n = 5), or ifosfamide (n = 45) in combination with cisplatin, or had missing cisplatin dose data (n = 12).

Cumulative cisplatin dose was significantly associated with tinnitus [OR per 100 mg/m² = 1.38; 95% confidence interval (CI): 1.1–1.7; P = 0.007]. Tinnitus risk for 400 mg/m²-treated TCS (n = 358) was not significantly greater than those receiving 300 mg/m² (n = 300; OR = 1.14; 95% CI: 0.8–1.6; P = 0.41). However, survivors who received >400 mg/m² (n = 30, median dose = 580 mg/m², range = 402–828 mg/m²) were at 2.61(95% CI: 1.8–3.9)-fold increased risk compared with those receiving ≤400 mg/m² (P < 0.0001, Fig. 1A). Tinnitus was not associated with cumulative doses of bleomycin (OR = 1.01, P = 0.13), etoposide (OR = 1.02, P = 0.33), or ifosfamide (OR = 1.06, P = 0.87) when cumulative cisplatin dose was included as a covariate. Three of five TCS receiving both carboplatin (median dose = 1,800 mg/m², range = 5–3,646 mg/m²) and cisplatin (median dose = 200 mg/m², range = 200–527 mg/m²) had tinnitus (OR = 7.37; 95% CI: 1.2–58.5, P = 0.034). Tinnitus was associated with increasing age at diagnosis as a continuous variable (OR by decade = 1.31; 95% CI: 1.1–1.6, P = 0.006) and by quartile (OR = 1.22, P = 0.013; Fig. 1B). Age at evaluation was also associated with tinnitus (OR = 1.30; 95% CI: 1.1–1.5; P = 0.002). No association with follow-up duration was observed (age-adjusted OR = 0.99; P = 0.57).

Self-reported overall health was significantly lower in cases than in controls (OR = 0.54; 95% CI: 0.4–0.7; P < 0.0001, Fig. 2A). Tinnitus cases also reported using psychotropic medications more frequently than controls (OR = 2.40; 95% CI: 1.3–4.4; P = 0.003, Fig. 2B). Audiometry was performed on 602 (78.4%) subjects. TCS with tinnitus had significantly worse hearing at every frequency examined (0.25 kHz–12 kHz, P < 0.0001; Fig. 3A). Tinnitus was also associated with self-reported hearing loss (OR = 6.38; 95% CI: 4.9–8.5; P < 0.0001, Fig. 3B), problems hearing in a crowd (OR = 8.28; 95% CI: 5.5–12.5; P < 0.0001, Fig. 3C), and persistent dizziness or vertigo (OR = 6.40; 95% CI: 3.2–12.9, P < 0.0001; Fig. 3D), as well as symptoms of sensory neuropathy (OR range 1.66–2.72; P <0.0001, Fig. 3E).

Tinnitus was significantly associated with hypertension (age-adjusted OR = 2.07; 95% CI: 1.3–3.4; P = 0.003), and hypercholesterolemia (age-adjusted OR = 1.96; 95% CI: 1.2–3.2; P = 0.009). In an age at clinical examination-adjusted model with both variables, tinnitus was significantly associated with hypertension (OR = 1.77; 95% CI: 1.02–3.0; P = 0.039), but not hypercholesterolemia (OR = 1.6, 95% CI: 0.9–2.8, P = 0.09). Tinnitus was associated with noise exposure at work (age-adjusted OR = 1.93; 95% CI: 1.3–2.8; P = 0.0005), and outside of work (age-adjusted OR = 2.28; 95% CI: 1.6–3.3; P < 0.0001). Risk was also associated with chronic (>15 years) smoking (age-adjusted OR = 2.07; 95% CI: 1.2–3.8; P = 0.005), but not increased among ever smokers (OR = 1.26; 95% CI: 0.9–1.7; P = 0.20) or with alcohol consumption (OR = 0.92; 95% CI: 0.5–1.5; P = 0.12). Association with chronic smoking remained significant after adjusting for age, hypertension, and noise exposure (OR = 1.79; 95% CI: 1.01–1.04; P = 0.04).

Using GCTA's linear mixed model approach (27), we found that additive SNP effects explained a large fraction of CisIT variance (h² = 0.81 ± 0.42; P = 0.006). In GWAS, no SNP met genome-wide significance (Supplementary Table S1; Supplementary Fig. S3). The first independent signal identified (rs7532231, OR = 0.45, P = 1.1 × 10⁻⁶) is 116 kilobases upstream of the lncRNA RP11–364B6.1. The second (rs6671895, OR = 4.32, P = 1.2 × 10⁻⁶) is intronic to the lncRNA RP5–884C9.2. The third (rs7606353, P = 1.9 × 10⁻⁶, Fig. 4A) is a SNP 14 kilobases downstream of OTOS (otospiralin), a gene implicated in auditory function and survival of cochlear fibrocytes (34, 35). This variant is in near perfect linkage disequilibrium with rs74002277 in Europeans (CEU R² = 0.96), a SNP in the 3′-UTR region of OTOS and which alters several transcription binding motifs (35). The minor allele (G, 4% frequency) increased CisIT risk (OR = 4.20; 95% CI: 2.3–7.5). Because there has been prior evidence that OTOS protects cochlear fibrocytes against cisplatin cytotoxicity (37), we hypothesized that variants associated with higher OTOS expression lowered CisIT risk. In GTEx, OTOS was significantly expressed only in pituitary and thyroid tissues (Supplementary Fig. S4). We found that OTOS eQTLs were significantly enriched in GWAS independently from rs7606353 (P = 0.018). One haplotype block containing six thyroid/pituitary eQTLs was driving this enrichment (Supplementary Table S2; Supplementary Fig. S4). The minor allele (A) of the most significant variant within this block (rs10190781, 1.5% frequency) was associated with higher CisIT risk (OR = 3.42; 95% CI: 1.5–9.5; P = 0.007; Fig. 4B) and lower OTOS expression (Fig. 4C), suggesting that higher OTOS expression is protective. Tinnitus risk increased with the number of risk alleles in either locus (OR = 3.77; 95% CI: 2.3–6.2; P = 9.5 × 10⁻⁸; Fig. 4D).

We tested the association between OTOS expression and cellular sensitivity to cisplatin (as measured by area under the dose–response curve) in central nervous system (CNS) tumor lines (30, 31) and found a significant positive correlation (Spearman ρ = 0.46, P = 0.03; R² = 0.25, P = 0.01; Fig. 4E), further supporting a protective OTOS function against cisplatin-induced damage. To examine the specificity of this correlation, we compared the sensitivity of CNS tumor cell lines to eight antineoplastic agents [5-fluorouracil (5-FU), bleomycin, bortezomib, cisplatin, cytarabine, docetaxel, etoposide, and vinblastine] based on OTOS expression levels (Supplementary Table S3). We also examined the relationship between cisplatin sensitivity and OTOS expression in seven cancer cell line types (aerodigestive, breast, CNS, digestive system, lung, skin, and urogenital system; Supplementary Table S4). Of the cytotoxic drugs tested, only cisplatin sensitivity demonstrated a statistically significant correlation with OTOS expression in CNS lines. Furthermore, cisplatin sensitivity was associated with OTOS expression only in CNS tumor lines and not cell lines of other tissue origins. These data suggest a CNS tumor–specific association between OTOS expression and cisplatin sensitivity.

To functionally validate the role of OTOS gene expression in cellular sensitivity to cisplatin, we overexpressed OTOS in two cell lines, HEI-OC1 (a mouse auditory cell line) and L929 (a mouse fibroblast cell line), and compared gene expression levels and cellular proliferation following cisplatin treatment in OTOS-transfected cells and plasmid-transfected control cells (Fig. 5A). Overexpression of human OTOS in HEI-OC1 resulted in increased cell viability following cisplatin treatment at 5, 7.5, 10, and 25 μmol/L concentrations for 48 hours (P = 0.004; Fig. 5B) and 72 hours (P = 0.008; Fig. 5C). Supplementary Figure S5 illustrates the effect of cisplatin on HEI-OC1 cells transfected with human OTOS and plasmid control. Furthermore, the cisplatin IC₅₀ value of HEI-OC1 cells transfected with human OTOS (18.68 μmol/L, 95% CI: 13.95–25.01 μmol/L) was 1.8-fold higher than cells transfected with the control plasmid (10.66 μmol/L, 95% CI: 6.04–18.82 μmol/L) at 48 hours. Overexpression of OTOS was confirmed via qPCR, with human OTOS expression increasing 69-fold at the time of drug treatment, 24 hours following pOTOS transfection (Fig. 5C, inset). Similarly, overexpression of OTOS in L929 fibroblasts also markedly reduced cellular sensitivity to cisplatin following treatment for 48 hours (P = 0.006; Fig. 5D) and 72 hours (P = 0.03; Fig. 5E) concomitant with increased expression of OTOS (Fig. 5E, inset). The cisplatin IC₅₀ value in OTOS-transfected cells (16.30 μmol/L, 95% CI: 13.25–20.04 μmol/L) was 1.5-fold higher than cells transfected with the control plasmid (10.96 μmol/L, 95% CI: 8.82–13.61 μmol/L) at 48 hours.

We mapped SNPs to genes by proximity and performed pathway enrichment with Gene Ontology terms. We identified 134 coding genes with P < 5 × 10⁻⁴ (Supplementary Table S5). Ten pathways were significantly over-represented (Supplementary Table S6), including nervous system process (q = 0.005), potassium ion transmembrane transport (q = 0.007), and sensory perception of sound (q = 0.005). Potassium transport most strongly implicated KCNQ1 (rs2283200, P = 8.9 × 10⁻⁶), which maintains high potassium concentrations in the endolymph of the inner ear (38, 39).

To our knowledge, this is the first investigation to consider the effects of clinical, genetic, and lifestyle factors on CisIT risk. Here 25.8% and 15.2% of cisplatin-treated TCS reported mild and moderate/severe tinnitus at a median of 5 years after treatment. Survivors treated with 100–400 mg/m² cisplatin appeared to have similar CisIT risk, but those given >400 mg/m² were at 2.6-fold increased risk, suggesting a nonlinear threshold–dose response. Follow-up time was not associated with CisIT, suggesting irreversibility, but longitudinal data are needed to support this observation. Significant associations with tinnitus were observed for hearing loss, vertigo, sensory neuropathy, poorer overall health, and use of psychotropic medications, as well as for hypertension, noise exposure, and chronic smoking. GWAS revealed independent signals implicating OTOS that had relatively large effect sizes (ORs of 3.5–4.5) and were relatively rare (minor allele frequencies 1%–5%). We verified protective effects of OTOS through eQTL analysis and functional validation in two cell lines, demonstrating that decreased OTOS expression increases susceptibility to CisIT. Whether thyroid/pituitary eQTLs coregulate expression in cochlear fibrocytes is unknown, and human cochlear transcriptomic databases are necessary for more precise interpretations. However, we provide further evidence for OTOS's protective functions in silico using central nervous system cell line data, and in vitro by overexpressing human OTOS in a mouse auditory cell line routinely used for in vitro drug ototoxicity screening (40) and a mouse fibroblast cell line. Pathway analysis implicated potassium transport genes, including the cochlear potassium transporter KCNQ1 (38, 39).

Despite adequate statistical power, significant differences in tinnitus risk between TCS given 3 versus 4 cycles of cisplatin-based chemotherapy were not evident. Brydøy and colleagues previously showed that TCS treated with >4 cycles of cisplatin-based chemotherapy had a 2.21-fold increased risk of tinnitus compared with those given 1–4 cycles (12), consistent with our findings. Bokemeyer and colleagues showed a dose-dependent increased risk for ototoxicity after cisplatin-based chemotherapy, but did not specifically assess tinnitus or compare by dosage group (41).

An approximate 2-fold association between hypertension and CisIT was apparent among our TCS. The relationship between hypertension and tinnitus in the general population is well-established, with a recent meta-analysis estimating a pooled OR of 1.37 (P < 0.05; ref. 42). It has been hypothesized that hypertension causes damage to either the cochlear microcirculation and/or the stria vascularis, which supplies blood to the organ of Corti (43). This could explain the observed association.

Multiple studies have identified a relationship between tobacco use and tinnitus in the general population (44–46) and we found a significantly increased risk of tinnitus in survivors with >15 years of tobacco use. It is possible that long-term tobacco use may reduce peripheral blood flow, including to the inner ear microcirculation. The role of tobacco use in CisIT was examined by Brydøy and colleagues, but significant associations were not apparent for either current or former smokers compared with never smokers (40), similar to our results.

Noise exposure is a known risk factor for tinnitus (47), but its role in the development of the hearing disorder in other cisplatin-treated cohorts has not been evaluated to our knowledge. Although Bokemeyer and colleagues reported an association between a history of noise exposure and cisplatin-induced ototoxicity, this term included both tinnitus and hearing loss (41).

In our cohort, tinnitus was positively associated with symptoms of sensory neuropathy. This could suggest that inherent susceptibility to cisplatin-induced neurophysiological damage may partially underlie various neuro-otological symptoms, possibly involving pharmacokinetic pathways, intrinsic cisplatin sensitivity, reduced regenerative capacity, or altered neural plasticity. Another possible explanation is that of patient perception. Self-reported tinnitus, hearing loss, and neuropathy in cancer survivors have been linked to the Impact of Events Scale-Revised (IES-R) hyperarousal scores (14), suggesting that differences in perception and appraisal could partially explain our findings. One study showed that audiometric thresholds correlated with tinnitus awareness time and loudness, but that tinnitus-induced annoyance was associated with proneness to exhibit anxious responses (48, 49). These investigations highlight how both cochlear and central nervous mechanisms contribute to tinnitus.

While single-variant analysis did not identify significant genetic signals, we found that cumulatively, additive SNP effects explain a large portion of variance in CisIT, further supporting the hypothesis that CisIT is a polygenic trait with complex genetic underpinnings (4). Several mechanisms of tinnitus have been proposed, but no consensus on its etiology has been reached. One model postulates that disruptions to the endocochlear potential result in inhibitory–excitatory imbalances affecting spontaneous cochlear activity and lead to tinnitus (50). Inner-ear fibrocytes supply potassium ions to strial cells which pump them into the endolymph to maintain the endocochlear potential, the “cochlear battery” driving auditory transduction (51). The link between endocochlear potential and spontaneous auditory nerve activity has been established (52). Our GWAS points to the potential importance of this hearing mechanism in CisIT.

OTOS mRNA and protein expression localize to spiral limbus and spiral ligament fibrocytes (34), and overexpression protects primary spiral ligament fibrocytes against cisplatin cytotoxicity (37). In support of the protective effect of OTOS against cisplatin sensitivity, our in silico analysis indicated that cisplatin sensitivity is positively associated with increased OTOS expression in CNS tumor cell lines, a rough approximation of cell types that would be affected by cisplatin-induced neurotoxicity. While CNS tumor cell lines are not an ideal proxy for assessing toxicity to inner ear cells, we further validated the protective effects of OTOS by demonstrating that its overexpression in vitro reduces cisplatin sensitivity in two cell lines (HEI-OC1 and L929), with HEI-OC1 more accurately reflecting inner ear cells that are adversely affected by cisplatin.

In support of our findings, two prior studies have elucidated the importance of OTOS expression in mammalian hearing. Inducible knockdown of OTOS in guinea pigs caused severe degeneration of the organ of Corti and deafness, along with damage to fibrocytes (34). Interestingly, although Otos^−/− mice had damaged fibrocytes, they did not exhibit histologic abnormalities of sensory hair cells, and experienced only moderate deafness (35). Delprat and colleagues attributed this difference to long-term adaptation to congenital Otos absence (34). Therefore, it is plausible that inducible knockdown, modeling key aspects of direct auditory insults such as cisplatin administration, will acutely alter the ionic composition of the endolymph, disrupting hair cell metabolism and survival.

Breglio and colleagues recently showed a reduction of endocochlear potential following cisplatin administration in mice, and preferential accumulation of cisplatin in the stria vascularis (53). We linked potassium transport to CisIT in our pathway analysis, which most strongly implicated KCNQ1, an important cochlear ion channel. This voltage-gated channel localizes to marginal cells of the stria vascularis where it maintains high K+ concentrations in the endolymph (38). Mutations in KCNQ1 result in deafness and long QT syndrome (54).

Taken together, these findings point to the function of supporting cells in the cochlear lateral wall in cisplatin ototoxicity and hair cell survival. Impaired maintenance of the endolymph causes hair cell death and permanent hearing loss (55), which may be partially mediated by a reduced endocochlear potential. However, prior research has largely focused directly on cisplatin's cytotoxic mechanisms in hair cells (56). Our new findings, together with those in the literature, suggest that novel strategies aimed at protecting supporting cells in the cochlear lateral wall, perhaps by stimulating OTOS expression, may prove beneficial. Very little is known about the protein product of the gene, its functions, and its interactions. Better understanding of the functional protein and its cellular localization and interactions is warranted. A potential approach would be to use a gene expression–based high-throughput screen to identify small molecules or siRNA capable of markedly increasing OTOS expression without perturbing the viability of auditory cells, such as HEI-OC1. The identification of compounds or siRNAs resulting in increased OTOS expression could be evaluated for their effect on protecting cells from cisplatin damage. Importantly, the utility of this approach depends on the modulation not impacting the antineoplastic activity of cisplatin.

It remains unclear whether the effects of OTOS on tinnitus occur directly or are mediated by effects on hearing loss. However, if OTOS protects against tinnitus indirectly by preventing hearing loss, it remains a viable candidate for cisplatin otoprotection. It is also unclear whether the actions of OTOS are uniquely pertinent to cisplatin or generalizable to other ototoxic insults like noise, although evidence of direct protection against cisplatin cytotoxicity is presented here and in the literature (37). Future studies should elucidate the cellular functions and interactions of otospiralin to assess its viability as a target for cisplatin otoprotection.

Strengths of our study include the collection of detailed treatment data, homogeneous cisplatin-based chemotherapy, long-term follow-up, quantitative hearing evaluations, the use of validated instruments to assess neurotoxic symptoms, and the novel implementation of genome-wide scans to nominate cellular mechanisms and targets. Our results point to cochlear mechanisms that have not yet been widely explored in cisplatin ototoxicity, and may provide novel opportunities for targeted otoprotection.

An intrinsic limitation of all cross-sectional studies is the inability to make causal inferences. Furthermore, questionnaires with more detailed information on the experience of tinnitus, including baseline measurements pre-chemotherapy, would provide more conclusive evidence with regard to associations noted here. However, given the young median age of patients at treatment, it is unlikely that tinnitus was pre-existing. It is important to note that the results of our translational research could potentially impact millions of patients worldwide annually diagnosed with cisplatin-treated cancers who are at risk for CisIT. Our findings may also inform tinnitus research in nonplatinum–treated patients. In view of our results, health care providers can improve management of cancer survivors experiencing CisIT by recommending monitoring of blood pressure, encouragement of smoking cessation, and avoidance of additional ototoxic insults such as noise exposure.

B. Mapes is an employee of and has ownership interests (including patents) at Tempus Inc. D.R. Feldman reports receiving commercial research support from Novartis, Seattle Genetics and Decibel, and receives royalties from review chapter of UpToDate. R.J. Hamilton is a consultant/advisory board member for Bayer, Janssen, Amgen, Astellas, and Abbvie, and reports receiving commercial research support from Janssen and Bayer. C. Kollmannsberger is a consultant/advisory board member for Pfizer, Astellas, Bristol-Myers Squibb, EMD Serono, Ipsen, Eisai, Janssen, Merck and Roche. E.R. Gamazon is a consultant/advisory board member for City of Hope/Beckman Research Institute. L.H. Einhorn has ownership interests (including patents) at Amgen and Biogenidec, and is a consultant/advisory board member for Celgene Advisory Board. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O.E. Charif, B. Mapes, H.E. Wheeler, R.D. Frisina, D.J. Vaughn, C. Kollmannsberger, M. Kubo, N.J. Cox, L.B. Travis, M.E. Dolan

Development of methodology: O.E. Charif, B. Mapes, H.E. Wheeler, C. Wing, R.D. Frisina, E.R. Gamazon, M.E. Dolan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.R. Trendowski, C. Wing, D.R. Feldman, R.J. Hamilton, D.J. Vaughn, C. Fung, C. Kollmannsberger, T. Mushiroda, M. Kubo, R. Huddart, L.B. Travis, M.E. Dolan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O.E. Charif, B. Mapes, M.R. Trendowski, H.E. Wheeler, C. Wing, P.C. Dinh Jr., R.D. Frisina, D.R. Feldman, D.J. Vaughn, C. Kollmannsberger, E.R. Gamazon, P. Monahan, S.D. Fossa, L.H. Einhorn, L.B. Travis, M.E. Dolan

Writing, review, and/or revision of the manuscript: O.E. Charif, B. Mapes, M.R. Trendowski, H.E. Wheeler, P.C. Dinh Jr., R.D. Frisina, D.R. Feldman, R.J. Hamilton, D.J. Vaughn, C. Fung, C. Kollmannsberger, M. Kubo, E.R. Gamazon, N.J. Cox, R. Huddart, S. Ardeshir-Rouhani-Fard, P. Monahan, S.D. Fossa, L.H. Einhorn, L.B. Travis, M.E. Dolan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Mapes, P.C. Dinh Jr., D.R. Feldman, R.J. Hamilton, M. Kubo, L.B. Travis

Study supervision: M. Kubo, S.D. Fossa, L.B. Travis, M.E. Dolan

The HEI-OC1 cell line was a generous gift from Dr. Federico Kalinec, House Ear Institute, Los Angeles, CA. The L929 cell line was a generous gift from Dr. Wendy Stock at the University of Chicago. This work was supported by the NIH Genetic Susceptibility and Biomarkers of Platinum-related Toxicities grant (R01 CA157823), and Pharmacogenomics Research Network (PGRN)-RIKEN Global Alliance which is supported by the RIKEN Center for Integrative Medical Science and the NIH Pharmacogenomics Research Network (GM115370). The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by the NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. The data used for the analyses described in this manuscript were obtained from the GTEx Portal on 04/28/18. R.J. Hamilton acknowledges support of The Josh Carrick Charity and the Institute of Cancer Research and Royal Marsden NIHR Biomedical Research Centre.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.