Abstract Genomic mutations progressively arise in our tissues, accumulating throughout life. The frequency and location of these mutations are influenced by lifestyle and genetic factors. When several mutations arise in cancer-driving genes, progressive phases of carcinogenesis may begin. Accurate cancer risk assessment in clinically normal appearing tissues (CNATs) allows time for preventative intervention to take place, lowering the overall treatment burden on patients. Therefore, it is essential for us to understand the correlation between mutation burden and cancer risk in CNATs. Ultradeep sequencing is required to study mutations in CNATs. However, ultradeep sequencing is currently only feasible in targeted genomic areas. It is well-known that mutation hotspots exist in a variety of cancer types. Our study aimed to assess whether mutation hotspots found in cancer are suitable for assessing cancer risk in CNAT. In our current work, we identified a high-quality mutation dataset from cancer and CNAT of the skin, bladder, and lung. The number of samples per CNAT dataset was 123, 555, and 515 in skin, bladder, and lung, respectively. We used our previously presented “hotSPOT” computational tool to identify the 10% of exome regions containing the most mutations in each cancer dataset. Each region found by our tool was labeled as a unique cancer mutation hotspot (CMH). We split the CNAT datasets into high and low-risk subsets based on history of sun exposure (skin), and history of cancer (bladder, lung). The distribution of mutations in CMH was compared in high and low-risk CNATs of each organ site. Finally, we tested the ability of mutations in CNATs located within CMHs to classify unlabeled samples based on risk. Each dataset was split 80%/20% into training and test datasets. The training datasets were fit to a neural network and the test datasets were used to measure the overall prediction accuracy. We identified 8 CMHs with significantly more mutations in high-risk skin compared to low-risk skin including CMHs in GRM3, SALL1, and TP53 (p < 0.05 - p < 0.01). In bladder, 29 CMHs captured significantly more mutations (p < 0.01 - p < 0.001) and 38 CMHs captured significantly less mutations (p < 0.01 - p < 0.001) in high-risk samples compared to low-risk. In our lung dataset, a single CMH in ZNF479 was found to capture significantly (p < 0.01) more mutations and 3 CMHs were found capture less mutations in high-risk samples compared to low-risk including areas of TP53 (p < 0.001, p < 0.05), and CST8 (p < 0.01). The neural network risk prediction models yielded an accuracy of 66.67%, 92.04%, and 66.99% in skin, bladder, and lung, respectively. Variability in mutation distribution of CNATs within CMHs and prediction model accuracy may be due to differences in dataset size and the number of relevant CMHs in each organ site. We found prominent differences in the mutation distribution of high and low-risk CNATs within CMHs of multiple organ sites. Our findings indicate that genomic tools could be developed to predict cancer risk in CNATs. Citation Format: Sydney R. Grant, Megan E. Fitzgerald, Spencer R. Rosario, Prashant K. Singh, Barbara A. Foster, Wendy J. Huss, Lei Wei, Gyorgy Paragh. Mutation hotspots in skin, bladder, and lung cancer as targets of carcinogenic risk assessment in clinically normal-appearing tissue. [abstract]. In: Proceedings of the AACR Special Conference: Precision Prevention, Early Detection, and Interception of Cancer; 2022 Nov 17-19; Austin, TX. Philadelphia (PA): AACR; Can Prev Res 2023;16(1 Suppl): Abstract nr P012.
Read full abstract