Files
Abstract
As cancer treatment has advanced, there is an increasing number of cancer survivors. Cancer survivors face unique health challenges due to the long-term effects of aggressive treatments. Cisplatin, a commonly used chemotherapy agent worldwide, can cause peripheral sensory neuropathy and ototoxicities such as tinnitus, hearing loss, and vertigo. Long-term consequences of these neurotoxicities include cognitive decline, decreased quality of life, and increased mortality, with no FDA-approved treatments or preventive therapies for adult-onset cancer survivors. Therefore, it is critically important to understand the risk factors for these toxicities.
Although single nucleotide polymorphisms (SNPs) have been found to be associated with cisplatin-induced neurotoxicities through genome-wide association studies (GWAS), there are major gaps in knowledge that must be addressed before findings are clinically applicable. Individual SNPs, even those significantly associated with toxicities, are not sufficient for predicting risk. Thus, we trained and tested machine learning prediction models for cisplatin-induced peripheral sensory neuropathy, tinnitus, and hearing loss, utilizing a variety of clinical and genetic inputs as predictors. Models for predicting severe cisplatin-induced peripheral sensory neuropathy were particularly effective, with strong sensitivity and specificity. Models with high sensitivity were also trained for predicting overall cisplatin-induced peripheral sensory neuropathy. Models predicting cisplatin-induced hearing loss showed a significant correlation between hearing thresholds predicted by the model and actual hearing thresholds. Analyzing the input features used to train models highlighted the importance of dyslipidemia and altered lipid measurements in the risk for cisplatin-induced peripheral sensory neuropathy.
The second area of interest was to determine whether there were population disparities in cisplatin-induced neurotoxicities. We found that survivors of African ancestry were significantly more likely to have peripheral sensory neuropathy and vertigo than those of European ancestry, with no significant differences in clinical and lifestyle factors (smoking, drinking, and medication use for blood pressure, cholesterol, and diabetes) between these populations. Filtering SNPs to find independent, functional ancestry-informative markers and assessing the association between these variants and toxicities identified expression quantitative trait loci for RNF24 in nerve tissue that were among the SNPs with the strongest p-values of association with neuropathy. SNPs impacting MFSD4B expression in nerve tissue and REV3L expression in brain tissue and SNPs impacting DDX25 expression in nerve tissue were among those with the strongest p-values of association with vertigo. Based on in silico validation, MFSD4B and REV3L expression were significantly correlated with cisplatin sensitivity in cancer cell lines.
This project provides important insights into the clinical and genetic risk factors for cisplatin-induced neurotoxicities. In the future, our work will help inform clinical risk prediction of cisplatin-induced neurotoxicities prior to treatment. It can also contribute to more effective dosing and monitoring in adult-onset cancer patients treated with cisplatin as well as the development of targeted interventions.