Pharmacogenomics of cisplatin‐induced neurotoxicities: Hearing loss, tinnitus, and peripheral sensory neuropathy

Abstract Purpose Cisplatin is a critical component of first‐line chemotherapy for several cancers, but causes peripheral sensory neuropathy, hearing loss, and tinnitus. We aimed to identify comorbidities for cisplatin‐induced neurotoxicities among large numbers of similarly treated patients without the confounding effect of cranial radiotherapy. Methods Utilizing linear and logistic regression analyses on 1680 well‐characterized cisplatin‐treated testicular cancer survivors, we analyzed associations of hearing loss, tinnitus, and peripheral neuropathy with nongenetic comorbidities. Genome‐wide association studies and gene‐based analyses were performed on each phenotype. Results Hearing loss, tinnitus, and peripheral neuropathy, accounting for age and cisplatin dose, were interdependent. Survivors with these neurotoxicities experienced more hypertension and poorer self‐reported health. In addition, hearing loss was positively associated with BMIs at clinical evaluation and nonwork‐related noise exposure (>5 h/week). Tinnitus was positively associated with tobacco use, hypercholesterolemia, and noise exposure. We observed positive associations between peripheral neuropathy and persistent vertigo, tobacco use, and excess alcohol consumption. Hearing loss and TXNRD1, which plays a key role in redox regulation, showed borderline significance (p = 4.2 × 10−6) in gene‐based analysis. rs62283056 in WFS1 previously found to be significantly associated with hearing loss (n = 511), was marginally significant in an independent replication cohort (p = 0.06; n = 606). Gene‐based analyses identified significant associations between tinnitus and WNT8A (p = 2.5 × 10−6), encoding a signaling protein important in germ cell tumors. Conclusions Genetics variants in TXNRD1 and WNT8A are notable risk factors for hearing loss and tinnitus, respectively. Future studies should investigate these genes and if replicated, identify their potential impact on preventive strategies.


| INTRODUCTION
Cisplatin and other platinating agents represent the most widely used and successful class of cytotoxic drugs worldwide. More than 5.8 million patients (pediatric and adults) globally are diagnosed each year with cancers (e.g., testicular, ovarian, bladder, lung, head and neck, pancreas, breast, endometrium, esophagus, advanced cervical cancer, lymphomas, metastatic osteosarcoma, and others) for which first-line therapy can potentially include platinating agents. 1,2 Cisplatin is associated with over 95% 5-year survival rates in germ cell tumors, but can result in debilitating off target effects including ototoxicity, neurotoxicity, nephrotoxicity, and cardiometabolic abnormalities. 3,4 Unfortunately, there are no approved preventive measures and no FDA-approved drug therapies for these toxicities; however, patients would benefit from individualized risk assessments, allowing for detailed education and counseling, and the development of a personalized treatment and monitoring plan. There are attempts to identify individuals a priori who are more likely to develop these sequelae through studies of genetic risk factors and nongenetic comorbidities. 5,6 However, clinical implementation of assessing genetic biomarkers and counseling patients with nongenetic comorbidities with regard to potential toxicities has been limited.
To ascertain both genetic risk factors and modifiable comorbidities for cisplatin-induced toxicities, we constructed the Platinum Study, a well-phenotyped cohort of testicular cancer survivors (TCS) treated with homogenous cisplatin-based chemotherapy, 7,8 which consisted of primarily four cycles of EP (etoposide and cisplatin) or three cycles of BEP (bleomycin, etoposide, and cisplatin). 8 Testicular cancer is the most common malignancy among young men, predominantly of European descent. 9 Survivors can subsequently live upwards of 50 years following treatment, further accentuating the debilitating effects of iatrogenic neurotoxicities. Cisplatin-induced neurotoxicities are common, long term, irreversible adverse events in TCS, with 75%-80% of survivors developing hearing loss, 10,11 40% experiencing tinnitus, 12 and 56% reporting symptoms of peripheral sensory neuropathy. 13 Through an agnostic genome-wide association study (GWAS), a single nucleotide polymorphism (SNP) in WFS1 (rs62283056; p = 1.4 × 10 −8 ) was associated with increased susceptibility to cisplatin-induced hearing loss, 11 and replicated in an independent Canadian study of 229 TCS evaluating the same phenotype (p = 5.67 × 10 −3 ). 14 Although a traditional GWAS of cisplatin-induced tinnitus found no genome-wide significant signals, OTOS (rs7606353; p = 1.90 × 10 −6 ) was identified as marginally significant and functional studies in auditory cells indicated that knockdown of OTOS was associated with higher cellular sensitivity to cisplatin. 12 A GWAS of cisplatin-induced peripheral neuropathy identified no significant SNP associations. 13 Results: Hearing loss, tinnitus, and peripheral neuropathy, accounting for age and cisplatin dose, were interdependent. Survivors with these neurotoxicities experienced more hypertension and poorer self-reported health. In addition, hearing loss was positively associated with BMIs at clinical evaluation and nonwork-related noise exposure (>5 h/week). Tinnitus was positively associated with tobacco use, hypercholesterolemia, and noise exposure. We observed positive associations between peripheral neuropathy and persistent vertigo, tobacco use, and excess alcohol consumption. Hearing loss and TXNRD1, which plays a key role in redox regulation, showed borderline significance (p = 4.2 × 10 −6 ) in gene-based analysis. rs62283056 in WFS1 previously found to be significantly associated with hearing loss (n = 511), was marginally significant in an independent replication cohort (p = 0.06; n = 606). Gene-based analyses identified significant associations between tinnitus and WNT8A (p = 2.5 × 10 −6 ), encoding a signaling protein important in germ cell tumors. In the present study, we evaluated 1680 TCS from the Platinum Study to comprehensively investigate and quantify the extent to which nongenetic associations, including modifiable comorbidities (hypertension, hypercholesterolemia, tobacco use, excess alcohol consumption, and loud noise exposure) were related to all three neurotoxicities (hearing loss, tinnitus, and peripheral neuropathy). To identify novel genetic signals associated with cisplatininduced hearing loss, tinnitus, and peripheral neuropathy, we performed GWAS in over 1000 survivors. We also attempted to replicate previously identified SNP associations for each phenotype.

| Patients and study design
All 1680 patients were enrolled in the Platinum Study, which includes eight cancer centers in the United States, Canada, and United Kingdom. 7,8 Eligibility criteria are illustrated in Figure S1 and as previously described. [11][12][13] All survivors provided written consent for study participation, access to medical records, and genotyping. Study procedures were approved by each institution's Human Subject Review Board and conducted in accordance with the U.S. Common Rule. The overall study design and sample size for each analysis (GWAS and independent SNP analysis) are also shown in Figure S1.

| Assessments
Patient data were determined at clinical follow-up or collected from medical records following a standardized protocol as previously described. 8 Data collected from medical records included: treatment regimen with cumulative dose and the number of cycles of each chemotherapeutic, and age at diagnosis of TC. Data collected during a physical examination included: age, weight, and height. At this clinical evaluation, the patient also completed a validated self-reported questionnaire, and blood was collected for genotyping. In addition, puretone air conduction thresholds were measured bilaterally at the frequency range 0.25-12 kHz to quantify hearing.
Information regarding hypertension, hypercholesterolemia, persistent dizziness/vertigo, overall health conditions, noise exposure, alcohol consumption, and tobacco use were also obtained from these questionnaires as previously described (Methods S1). [11][12][13] 2.3 | Establishment of the cisplatin-induced hearing loss, tinnitus and peripheral sensory neuropathy phenotypes Cisplatin-induced hearing loss was modeled as a quantitative phenotype using the geometric mean of bilateral average air conduction thresholds measured at frequencies between 4 and12 kHz as described previously (n = 1258; previous study analyzed 488 subjects of this cohort 11 ).
Cisplatin-induced tinnitus was defined as previously described (n = 1217, previous study analyzed 762 subjects of this cohort 12 ), based on response to the question, "Have you had in the last 4 weeks: ringing or buzzing in the ears?" from the validated SCIN questionnaire. 15 Answers included: not at all, a little, quite a bit, very much. Survivors were dichotomized to tinnitus case/control groups. Controls were survivors who responded "not at all." Cases were only survivors who responded "quite a bit" or "very much" with the exclusion of those who answered "a little" to establish a more rigorous phenotype. In a separate question, survivors were asked, "Do you have ringing or buzzing in the ears?" For consistency, tinnitus cases responding "no" to this question, but "quite a bit" or "very much" to the SCIN question were excluded from the analysis.
For cisplatin-induced peripheral neuropathy, the frequency of sensory neuropathy was evaluated using eight items in the validated EORTC-CIPN20 (Table S1). 16 We converted sample responses to a 0-3 numeric scale: 0 for "none", 1 for "a little", 2 for "quite a bit", 3 for "very much." We created four categories to represent the severity of peripheral neuropathy for 1653 TCS using a summary statistic mathematically equivalent to the standard scoring algorithm 17 and combined groups 2 and 3 due to small sample sizes as described in our previous study that analyzed 680 subjects of this cohort. 13

| Analysis of phenotypes with patient characteristics
Associations between survivors' characteristics and cisplatin-induced hearing loss, tinnitus, and peripheral neuropathy were evaluated using linear, logistic, and proportional odds ordinal logistic regression, respectively. Models were adjusted for age at clinical examination and cumulative cisplatin dose, which were used as continuous variables.
To investigate cumulative cisplatin dose threshold, regression analyses were also conducted adjusting for age at clinical examination for each phenotype with cumulative cisplatin dose as a categorical variable (<300, 300, 400, >400 mg/m 2 ). Survivors receiving >300 and < 400 mg/ m 2 cumulative cisplatin dose were excluded due to small sample size (n = 69). In addition, the majority of TCSs receive typical homogeneous treatments with cumulative cisplatin dose of 300 or 400 mg/m 2 . We further performed a two-proportion z-test for each phenotype, comparing the proportion of survivors with the phenotype. For cisplatininduced hearing loss, the two-proportion z-test was applied on patient proportions with hearing threshold >20 and ≤20 dB; for tinnitus, the population proportions of the case and control were compared; and the proportion of patients with severe cisplatin-induced peripheral neuropathy was compared to the proportion of the remaining cohort.
All tests were two-sided at a significance threshold of p <0.05. All statistical analyses were performed in R 3.6.1 and plotted with ggplot2 unless otherwise specified.

| Genotyping and imputation
At the time of clinical evaluation, DNA was extracted from peripheral blood of survivors. Genotyping was performed on the Infinium Global Screening Array-24 chip (GSA-24v1-0_A1; Illumina) at Regeneron Pharmaceuticals. Sample-level and SNP-level quality control criteria were consistent with our previous studies [11][12][13] and illustrated in Figures S2-S4.The minor allele frequency (MAF) threshold was set to 0.01 for tinnitus and 0.05 for hearing loss and peripheral neuropathy. Imputation was done on the University of Michigan Imputation Server. SNPs and samples passing QC criteria comprised the input set for imputation with EAGLE phasing using the Haplotype Reference Consortium. [18][19][20] SNPs with imputation R 2 < 0.8, MAF < MAF threshold, HWE p < 1 × 10 6 , and INFO scores > 1.05 or <0.6 were excluded ( Figures S2-S4).

| Genome-wide analyses
All GWAS were conducted using methodology similar to previous studies, [11][12][13] with significantly more samples and use of a different Illumina genotyping array. GWAS for hearing loss was done with linear regression for 1071 survivors with 5,385,324 SNPs using cumulative cisplatin dose, age at clinical examination, and the first 10 genetic principal components as covariates. 11 GWAS for tinnitus was performed with logistic regression on 1037 TCS and 7,657,611 SNPs using cumulative cisplatin dose, age at diagnosis, the first five genetic principal components as covariates as described previously. 12 We also adjusted for overall noise exposure in the tinnitus GWAS, because noise is a known risk factor for tinnitus 21 and because regression analyses from the current study demonstrated strong association between tinnitus and noise exposure (noise exposure at work: OR = 1.9, 95% CI = 1.4-2.6, p < 0.0001; outside of work: OR = 2.2, 95% CI = 1.6-3.0, p < 0.0001; both at work and outside of work: OR = 1.5, 95% CI = 1.2-1.8, p < 0.0001; Figure 1). GWAS for tinnitus and hearing loss was done in PLINK v1.9. Using age at diagnosis and the first 10 European genetic principal components as covariates, GWAS for peripheral neuropathy was performed for 1397 survivors with 4,875,644 SNPs by ordinal logistic regression in R 3.6.1 with the MASS package. 22 All GWAS assumed additive effects and had a genome-wide significance threshold set to p < 5 × 10 −8 .
For the gene-based association analysis, the aggregated effect of all SNPs within a gene was analyzed simultaneously in the functional mapping and annotation of GWAS (FUMA) platform using the multi-marker analysis of genomic annotation (MAGMA) method that is based on a multiple regression model that efficiently incorporates linkage disequilibrium between SNPs. 23,24 Summary statistics were then uploaded to FUMA for gene-based association analysis and for region-based plotting. SNPs were mapped to 18,544, 18,819, 18,106 protein coding genes for hearing loss, tinnitus, and peripheral neuropathy respectively, producing a significance threshold of 2.7 × 10 −6 . After collecting the candidate genes that were genome-wide significant or nearly significant, we extracted SNPs with a GWAS p < 1 × 10 −5 and within 25 kb upstream or downstream of these genes. We then searched these SNPs in GTEx portal for splicing quantitative trait loci (QTL) or expression QTLs without specifying tissue types, but associated with the candidate genes.

| Replication of candidate SNPs
Using independent cohorts, we evaluated two SNPs previously associated with cisplatin-induced neurotoxicities: rs62283056 in WFS1 for hearing loss 11 and rs7606353 in OTOS for tinnitus. 12 We performed independent SNP association tests adjusting with covariates using linear regression and assuming linear additive SNP effect. Excluding participants in the previous studies, linear regression, and logistic regression were performed on completely independent replication cohorts for hearing loss (n = 606) and tinnitus (n = 325), respectively. The covariates were consistent with the original GWAS analysis, and the significance threshold is 0.05.

| Evaluation of cisplatin sensitivity based on gene expression in silico
Gene expression data for two genes (TXNRD1, WNT8) from the FUMA gene-based analysis in central nervous F I G U R E 1 Associations between survivor characteristics and cisplatin-induced neurotoxicities. Forest plots of regression coefficients and 95% confidence interval (95% CI) for: (A) cisplatin-induced hearing loss, (B) cisplatin-induced tinnitus, (C) cisplatin-induced peripheral sensory neuropathy. All models were adjusted for age at clinical evaluation and cumulative cisplatin dosage. Bolded regression coefficients (95% CI) are significantly associated at α = 0.05. *p ≤ 0.05; **p ≤ 0.001 system (CNS) was acquired from the Cancer Cell Line Encyclopedia. 25 Cisplatin sensitivity, which was measured as the area under the dose-response curve, was obtained from the Genomics of Drug Sensitivity in Cancer Project. 26 We then performed Spearman correlation and linear regression to analyze the associations between gene expression and drug sensitivity of cancer cell lines. Genes with missing expression data were excluded. There were only four non-missing gene expression values in CNS tumor cell lines from the Cancer Cell Line Encyclopedia 25 available for WNT8, preventing an analysis of gene expression and cell sensitivity to cisplatin. These analyses were done in R 3.6.1.

| Cohort characteristics
Cohort characteristics stratified by each cisplatininduced phenotype, hearing loss, tinnitus, and peripheral neuropathy are provided in Table 1 and Table S2.
c Following conversion of the Likert scale: "none, a little, quite a bit, very much" to a 0-3 numeric scale, we created four categories to represent the severity of peripheral neuropathy using a summary statistic and combined groups 2 and 3. 13 Phenotype excludes 27 participants for whom the variables were not stated. T A B L E 1 (Continued) but no association was found between peripheral neuropathy and hypercholesterolemia (Figure 1).

| Effect of cumulative cisplatin dose on cisplatin-induced hearing loss, tinnitus, and peripheral neuropathy
Risk of more severe hearing loss for survivors who received 400 mg/m 2 was greater compared with those receiving 300 mg/m 2 (β = 0.1, 95% CI = 0.01-0.2, p = 0.02); a stronger increase of risk was found between 400 mg/m 2treated-TCS and >400 mg/m 2 -treated-TCS (β = 0.3, 95% CI = 0.06-0.5, p = 0.01; Figure 2A). Survivors who received >400 mg/m 2 were 3.1-fold more likely to have tinnitus (95% CI: 1.8-5.3-fold, p < 0.0001) compared with those receiving 400 mg/m 2 ( Figure 2B) with no significant differences in tinnitus risk for survivors treated with <300 mg/m 2 versus 300 mg/m 2 or for 300 mg/m 2 versus 400 mg/m 2 cisplatin. Performing the same analysis for peripheral neuropathy, survivors who were treated with 400 mg/m 2 were at 1.4-fold increased risk compared with those receiving 300 mg/m 2 (95% CI: 1.2-1.7-fold, p = 0.003) with no significant differences for <300 mg/m 2 versus 300 mg/m 2 and 400 mg/m 2 versus >400 mg/m 2 ( Figure 2C). To further confirm the dose threshold of cisplatininduced neurotoxicities, two-proportion z-tests were performed to compare patient proportions in dose groups. We only compared dose groups that demonstrated significant difference in neurotoxicity risks according to the regression analyses described above. The proportion of survivors with hearing loss and tinnitus was significantly greater following treatment with >400 mg/m 2 compared to 400 mg/m 2 cumulative cisplatin dose (hearing loss: 65% vs. 46%, p = 0.01; tinnitus: 43% vs. 19%, p < 0.0001). Although proportions of survivors with hearing loss treated with 400 mg/m 2 and 300 mg/m 2 were not significantly different, significant proportional differences were observed comparing survivors with peripheral neuropathy (16% vs. 8%, p < 0.0001).

| GWAS of cisplatin-induced hearing loss
In a GWAS of hearing loss for 1071 TCS, no SNP met genome-wide significance (Table S3); however, TXNRD1 was nearly genome-wide significant for the gene-based analysis (p = 4.2 × 10 −6 ; Figure 3A Table S4. Further, a SNP on chromosome 12, rs4406890, located in the intronic region of TXNRD1 was borderline significant Cisplatin sensitivity, measured as the area under the cisplatin dose-response curve, for all CNS tumor cell lines extracted from CancerRX. Gene expression data were downloaded from the Cancer Cell Line Encyclopedia. Expression data were rank normalized to fit a normal distribution prior to analysis. Correlation was assessed nonparametrically using the Spearman rank method, as well as by linear regression with a p-value of 5.7 × 10 −6 . This intronic variant is a sQTL that regulates alternative splicing of TXNRD1 in skeletal muscle as reported by the GTEx Portal.

| Replication of candidate SNPs
Performing individual SNP association analysis on completely independent cohorts by excluding survivors in previous studies, 11,12 the association between hearing loss and rs62283056 in WFS1 was borderline significant (p = 0.06). The overall hearing threshold distributions did not differ between discovery and replication cohorts (Table S2), however when analyzing-specific genotypes, we observed lower median hearing threshold for survivors carrying the risk allele in the replication set (Median: 15.4 dB; Range: 3.1-73.7 dB) compared to the discovery set (Median: 36.0 dB; Range: 6.8-86.9 dB; Figure S5). 11 This observation is also a potential explanation of the difference in association effect direction (β replication = − 0.34, β discovery = 0.11). No association (p = 0.7) was observed between tinnitus and rs7606353 in OTOS (Table S7). 12

| DISCUSSION
Based on the largest study to date of cisplatin-induced neurotoxicities, we demonstrate that a substantial proportion of TCS treated with cisplatin-based chemotherapy experience hearing loss, tinnitus, and peripheral neuropathy, and those that do experience neurotoxicities were more likely to have hypertension and describe their health as poor. Hearing loss, but not tinnitus or periperhal neuropathy, was associated with survivors with greater BMI. Conversely, tinnitus and peripheral neuropathy, but not hearing loss, was associated with persistent vertigo or dizziness. Similar results were observed in evaluating modifiable risk factors demonstrating survivors who are either chronic or current smokers are more likely to experience tinnitus or peripheral neuropathy, but not hearing loss. We identified a marked increase in risk for peripheral neuropathy (1.3-fold) and significantly more hearing loss when comparing 300 mg/ m 2 against 400 mg/m 2 cisplatin dose. In contrast, there is no significant difference for the risk of tinnitus, yet there is a significant risk (3.1-fold) above 400 mg/m 2 compared to 400 mg/m 2 . In addition, gene-based association analysis identified WNT8A significantly associated with tinnitus.

| Genetic findings for cisplatininduced tinnitus and hearing loss
For the first time, we identified WNT8A to be genomewide significant for tinnitus and TXNRD1 as having borderline genome-wide significance for hearing loss following cisplatin treatment through a gene-based association analysis. Human WNT8A mRNA is expressed in NT2 cells with neuronal differentiation potential 27 and was reported to be involved in the development of early embryos as well as germ cell tumors through activation of the WNT βcatenin-TCF pathway. 28 Mattes et al. 29 have illustrated that WNT8A plays a key role in the Wnt/β-catenin signaling pathway, inducing cell proliferation in a variety of cancer types, including colorectal cancer, pancreatic cancer, and gastric cancer. Wnt8a (mouse homolog) is expressed in the hindbrain and is involved in early inner ear development in mice. 30 Using the gEAR database, Wnt8a and Txnrd1 are expressed in mouse cochlea. 31 To inhibit the Wnt signaling pathway, several antineoplastic therapies have been developed, and many agents are currently in early phase oncology clinical trials. 32 Nevertheless, the role WNT8A in relation to tinnitus and/or de novo tinnitus is not yet established. TXNRD1, associated with cisplatin-induced hearing loss, is critical for redox regulation, antioxidant defense, and synthesis of deoxyribonucleotides. 33 Both increasing age and more severe hearing loss in a mouse model are associated with downregulation of the Txnrd1 gene in the auditory portion of the inner ear and cochlea. 34

| Cisplatin-induced tinnitus
The association between tinnitus and persistent dizziness/vertigo in our cisplatin-treated TCS has also been shown for de novo tinnitus, as inner ear problems are strongly implicated in balance disorders, including vertigo. 35 Both hypertension 36,37 and hypercholesterolemia 38,39 have also been demonstrated to be risk factors for de novo tinnitus. Przewoźny et al., 40 suggested that hypertension causes damage to the stria vascularis by reducing cochlear oxygen partial pressure and disrupting the recycling of potassium ions in the cochlea. The relationship with hypercholesterolemia may be related to evidence that cholesterol levels can impair cochlear microcirculation 41 and the function of cochlear outer hair cells. 42 Accordingly, these findings suggest that lowering abnormally high serum lipid levels and adequate control of hypertension may decrease the likelihood or severity of tinnitus.

| Other findings
We observed greater risk of both tinnitus and peripheral neuropathy for chronic smokers and current smokers. Tobacco use is known to be associated with poorer response to cancer treatments, increased cancer mortality, and risk of second primary cancer. 43 Continued smoking in patients with cancer is associated with significant incremental costs for further cancer treatment when first-line therapy fails. 44 In addition, previous investigations also demonstrated significant increases in the neurotoxic effects of chemotherapy in smokers, supporting the findings of the current study. 12,45 The 2020 Surgeon General's Report 46 indicates that smoking cessation improves patients' quality of life and adds up to 10 years to their lifespan. In addition, with clear evidence that smoking cessation is beneficial both before and after cancer treatment for many cancer types, this report 46  Cisplatin-based therapy consists primarily of either three cycles of BEP or four cycles of EP resulting in a cumulative dose of 300 or 400 mg/m 2 cisplatin, respectively although occasional patients may receive doses of cisplatin greater than 400 mg/m 2 . Our study evaluated the risk of toxicity at these cumulative doses to provide physicians valuable information to make an informed decision since both BEPX3 and EPX4 are considered curative for testicular cancer. 47 Our observation showed disproportionally increased risks of hearing loss (β = 0.4) or tinnitus (3.1-fold) following >400 mg/ m 2 cumulative cisplatin versus 400 mg/m 2 . A previous study showed that patients treated with >400 mg/m 2 cumulative cisplatin dose were 2.6-fold more susceptible to tinnitus than those treated with ≤400 mg/m 2 12 . In addition, Brydøy et al. found that compared with survivors receiving one to four cycles of cisplatin-based chemotherapy, TCS given five or more cycles had a 2.2fold greater incidence of severe hearing impairment and tinnitus. 48 We identified a marked increase in risk for peripheral neuropathy when comparing 300 mg/m 2 cumulative cisplatin dose with 400 mg/m 2 (1.3-fold). Consistent with our results, previous studies suggest that significant peripheral neuropathy occurs when cumulative cisplatin dose exceeds 300-400 mg/m 2 . 49 Krarup-Hansen et al. 50 conducted clinical and electrophysiological experiments in 16 TCS treated with BEP, to study the primary site of damage of cisplatin-induced peripheral sensory neuropathy. The authors reported that the amplitudes of sensory nerve action potentials (SNAP) were reduced 50-60% from the feet and fingers for TCS who received >300 mg/ m 2 cisplatin; and the conduction velocities of SNAP were reduced 10%-15% for survivors treated with cumulative doses of 400-700 mg/m 2 cisplatin compared to patients treated with <300 mg/m 2 . Among 80 women with recurrent ovarian cancer, Van den Bent and colleagues 51 reported an increased incidence of peripheral neuropathy development with >420 mg/m 2 cumulative cisplatin dose compared to 300-420 mg/m 2 . After a cumulative cisplatin F I G U R E 5 Genome-wide single nucleotide polymorphism (SNP) and gene-based association studies of cisplatin-induced peripheral sensory neuropathy (A) Manhattan plot of genome-wide association study (GWAS) results for cisplatin-induced peripheral sensory neuropathy. (B) Quantile-Quantile plot of GWAS results for cisplatin-induced peripheral sensory neuropathy. Covariates in both GWAS include age at diagnosis and 10 European genetic principal components accounting for population substructure. (C) Manhattan plot of the gene-based association analysis identifies no genome-wide significant genes. Summary statistics for SNP-based GWAS were uploaded to functional mapping and annotation to run a gene-based association analysis based on a multiple linear principal components regression to determine the aggregated effect of all SNPs within a gene. Inputted SNPs were mapped to 18,106 protein coding genes, producing a significance threshold of p = 0.05/18,106 (2.8 × 10 −6 ). (D) Quantile-Quantile plot of results from the gene-based association analysis dose of 600 mg/m 2 , all patients (n = 80) developed some degree of peripheral neuropathy with 30%-40% of patients having moderate sensory neuropathy, and 10% having severe and disabling neuropathy. Taken together, these prior reports [49][50][51] suggest that a dose threshold exists for disproportionally increased risk of peripheral neuropathy after cisplatin-based chemotherapy.
In comparison with our previous study on 511 survivors where rs62283056 in WFS1 was significantly (p = 1.4 × 10 −8 ) associated with hearing loss, 11 in this study of 606 survivors, the association was of borderline significance. A potential explanation of the lack of statistical significance is that there were more survivors carrying the risk allele that experienced little or no hearing loss in the replication cohort compared to the discovery cohort. 11 We did not successfully replicate rs7606353 in OTOS likely due to small sample size of the replication cohort (n = 325). In addition, the MAF (0.032) was lower than that of 0.04 in the previous study (n = 762 12 ), which constrained the statistical power of the replication analysis. We evaluated the genome-wide significance of rs7606353 among the entire cohort of 1037 survivors and found the p-value to be 1.5 × 10 −4 .

| Strengths and weaknesses
Major strengths of our investigation include the detailed phenotypic data collected in a large cohort of survivors treated with fairly uniform cisplatin-based chemotherapy. As a result, we were able to better quantify the associations between cisplatin-induced neurotoxicities and various comorbidities, and also analyze new contributors. Intrinsic limitations of any cross-sectional study, that also applies here, is the inability to infer causation between comorbidities and phenotypes (cisplatin-induced neurotoxicities). The definition of loud noise exposure is another limitation, as the participants self-reported their exposure, and details with regard to noise duration, loudness, or frequency (or timing in relationship to chemotherapy) was not queried. In addition, teasing out whether hearing loss is due to cisplatin, noise exposure or the combination is a challenge. Furthermore, when analyzing the effect of cumulative cisplatin dose on toxicity, the sample size for <300 and >400 mg/m 2 treatment groups were significantly smaller compared to the other groups. Lastly, this study focused on a European-only cohort and the results may not be generalized in other population.

| Clinical implications
Our findings emphasize the importance of clinical characteristics associated with cisplatin-induced neurotoxicities.
Clinicians should ensure that patients are aware of these iatrogenic toxicities and their associated comorbidities, including hypertension and hypercholesterolemia. Associated clinical conditions and behavioral characteristics should be addressed before receiving cisplatin-based chemotherapy, such as recommending close monitoring of blood pressure and cholesterol level, and encourage smoking cessation.
In addition, clinicians should be aware that a cumulative cisplatin dose of >400 mg/m 2 may result in disproportionally more survivors with ototoxicity than groups receiving lower doses. As noted previously, follow-up hearing assessment guidelines exist for children given cisplatin-based chemotherapy; however, no similar recommendations have been developed for adult-onset cancer survivors. 10 In Frisina et al., 10 the authors concluded that healthcare providers should at a minimum annually query patients about hearing status, consulting with audiologists as indicated. The increase in risk for peripheral neuropathy (1.3-fold) when comparing 300 mg/m 2 against 400 mg/m 2 cumulative cisplatin dose suggests for some patients (i.e., diabetic, pianist), a physician might consider the BEPX3 regimen over the EPX4 regimen. Additional research efforts are needed to replicate and validate the genetic risk factors, and their potential for new drug development to either prevent or treat these long-lasting neurotoxicities. Further implementation of individualized risk assessments to identify patients a priori who are more susceptible to developing neurotoxicities, thus allowing for personalized education, counseling, treatment, and monitoring plans, are needed.

ETHICAL APPROVAL STATEMENT
Study procedures were approved by each institution's Human Subject Review Board and conducted in accordance with the U.S. Common Rule. All participants provided written consent for study participation, access to medical records, and genotyping.

DATA AVAILABILITY STATEMENT
In addition to GWAS data in Supplemental Table 3, 5 and 6, full summary statistics for GWAS of cisplatin-induced hearing loss, GWAS of cisplatin-induced tinnitus and GWAS of cisplatin-induced peripheral neuropathy are available (https:// apps.cancer.iu.edu/plati num/publi shed-resea rch.php).