Identification of mutation hotspots to measure carcinogenic risk in Barrett's Esophagus using seq.hotSPOT and compSPOT
Year: 2023
Type: other
Abstract: In 2022, an estimated 20,000 new cases of esophageal adenocarcinoma (ESAD) were reported, with little decline over the last few decades, highlighting the need for cancer prevention. Barrett’s Esophagus (BE) is a precursor lesion to ESAD, with a reported risk of progression from 0.2-13.4% per patient per year. There is a lack of tools to accurately differentiate BE likely to progress to advanced disease. ESAD and BE have shown clonal mutations in known cancer-causing genes, including TP53, CDKN2A, ARID1A, PIK3CA, and SMAD4. We have developed a computational tool, seq.hotSPOT, to identify highly mutated genomic regions in various tissue types. Using whole-exome sequencing data from 6 cBioPortal datasets of ESAD and esophageal squamous cell carcinoma (ESCC) samples, we identified the top 10% most mutated genomic regions in each cancer subtype with seq.hotSPOT. Our tool compSPOT was then used to identify significantly mutated genomic hotspots in ESAD and ESCC, showing 5 hotspots in TP53 for ESCC and 9 hotspots within TP53 and CDKN2A in ESAD. Highly mutated genomic regions showed a significant correlation (p < 0.00001) between samples of clinically normal esophageal epithelium and both ESAD and ESCC. The correlation of mutated regions between BE found to progress to ESAD and BE with no observed disease progression were compared to ESAD samples. A significant correlation (p < 0.00001) was found in both cases, however, a higher correlation was observed in progressive BE compared to non-progressive (R = 0.836, R = 0.4924, respectively). The number of mutations present in BE samples found within the 9 mutation hotspots in ESAD was measured for progressive and non-progressive BE, revealing significantly more mutations within the progressive BE samples (p < 0.01). The ESAD mutation hotspots were then used as features in a machine learning model to classify BE samples as either progressive or non-progressive. The BE dataset was split into training and test data, following training of a neural network model, the performance of the model was measured based on the test dataset yielding an AUC of 0.7857. The relative variable importance was calculated for each hotspot, and the most influential hotspot in this classification model was located on TP53 chr17:7577022-7577121. seq.hotSPOT and compSPOT provide an efficient pipeline for identifying genomic areas of high mutation sensitivity, and future studies with larger cohort sizes may identify additional significant hotspot markers for the prediction of disease progression. The addition of personal risk factors of Esophageal Adenocarcinoma as features in machine learning model may increase the accuracy of a classification algorithm.
Authors: Sydney R. Grant, Spencer R. Rosario, Ella M. Sampson, Barbara A. Foster, Lei Wei, Wendy J. Huss, Gyorgy Paragh
Institution: Roswell Park Comprehensive Cancer Center
Language: English
Subfield: Surgery
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W4416559819