Background: Cisplatin is a major chemotherapeutic agent, which is highly effective in the treatment of many different cancers. One of the most common adverse drug reactions associated with this chemotherapeutic is ototoxicity (hearing loss). Various genome-wide association studies of cisplatin-induced ototoxicity (CIO) have shown that 38-47% of the variability has been attributed to genetics and that susceptibility can be traced to multiple loci. It is therefore important to use genomics data to develop predictive models and guide the development of preventative treatments.

Methods: Given that the genetic architecture of hearing loss and CIO overlap, we used large-scale hearing loss data from the UK Biobank to identify genetic factors contributing to diverse hearing traits. We used SBayesR to develop a polygenic score to predict hearing loss and S-PrediXcan to perform a transcriptome-wide association study to identify heritable gene expression profiles associated with hearing loss. We subsequently refined these results using publicly available data to increase the relevance of these findings to i) hearing - based on murine inner ear single-cell RNA sequencing data and ii) CIO - based on evidence of a drug-gene relationship, indicated by a correlation (P<0.05) between cisplatin cytotoxicity and gene expression. The relevance of the resulting polygenic scores to a pediatric CIO cohort was examined using ReACt. Additionally, through drug repurposing analyses, CIO-associated gene expression profiles were compared to the gene expression profiles induced by various perturbagens housed in CLUE.

Results: We found that a heritable hearing loss polygenic score was significantly associated with CIO (P= 3.85x10-3). While the novel filtering strategy did not improve the predictive capabilities of the score, this polygenic score was shown to outperform a previously developed polygenic score. A transcriptome-wide association study provided 178 credible genes that were associated with hearing loss that were used as input into drug repurposing analyses. When testing filtered gene sets, drug repurposing analyses revealed that the most promising otoprotectants (i.e., those showing the greatest dissimilarity in gene expression to input) were identified for the gene expression profile that had undergone the most stringent filtering for CIO. These analyses also uncovered perturbagens which induced highly similar gene expression profiles. These potential ototoxic targets, ADSS, GATAD2A and ZNF451, had been previously established in the scientific literature for involvement in hearing loss.

Conclusion: To the best of our knowledge, this is the first polygenic score developed to predict the risk of CIO. In addition, our novel filtering strategy showed an increase in confidence for drug repurposing results when applied to these data. Future studies should investigate this polygenic score in an independent cohort of cisplatin-treated patients. Further, the potential otoprotectants identified through this study should be tested in in vivo mouse models.