Telomere Length Dynamics and Chromosomal Instability for Predicting Individual Radiosensitivity and Risk via Machine Learning

Abstract

The ability to predict a cancer patient’s response to radiotherapy and risk of developing adverse late health effects would greatly improve personalized treatment regimens and individual outcomes. Telomeres represent a compelling biomarker of individual radiosensitivity and risk, as exposure can result in dysfunctional telomere pathologies that coincidentally overlap with many radiation-induced late effects, ranging from degenerative conditions like fibrosis and cardiovascular disease to proliferative pathologies like cancer. Here, telomere length was longitudinally assessed in a cohort of fifteen prostate cancer patients undergoing Intensity Modulated Radiation Therapy (IMRT) utilizing Telomere Fluorescence in situ Hybridization (Telo-FISH). To evaluate genome instability and enhance predictions for individual patient risk of secondary malignancy, chromosome aberrations were assessed utilizing directional Genomic Hybridization (dGH) for high-resolution inversion detection. We present the first implementation of individual telomere length data in a machine learning model, XGBoost, trained on pre-radiotherapy (baseline) and in vitro exposed (4 Gy γ-rays) telomere length measurements, to predict post radiotherapy telomeric outcomes, which together with chromosomal instability provide insight into individual radiosensitivity and risk for radiation-induced late effects.

Telomeres and aging: on and off the planet!

“…We and others have shown that longitudinal analyses of telomere length within individuals is more informative than cross-sectional comparison at single time points (across individuals); an illustrative example involved predicting potential risk of degenerative effects following radiation therapy in prostate cancer patients undergoing IMRT (Luxton et al 2021 ), which was consistent with data that telomere length changes most rapidly in proliferative cell populations (e.g., blood). Furthermore, a large Mendelian randomization collaboration (Telomeres Mendelian Randomization et al 2017 ) and recent quantitative estimates suggest that both short and long telomeres are associated with approximately equal degrees of increased disease risk (Protsenko et al 2020 ; Stone et al 2016 ).…”

Telomere fragility in radiology workers occupationally exposed to low doses of ionising radiation

“…To the best of our knowledge, this is the first study of telomere fragility in persons occupationally exposed to ionising radiation. In view of the fact that dysfunctional telomeres coincide with many late effects or radiation ( 44 , 45 ), we believe our findings provide one more argument to use telomere fragility as a biomarker of health risks associated with occupational exposure to ionising radiation.…”

A correlation graph attention network for classifying chromosomal instabilities from histopathology whole-slide images

“…Significant progress has been made in the application of machine learning in various scenarios, including the prediction of CIN status in tumor patients. 7 , 8 , 9 , 10 Most of these studies are based on gene expression data; the correlation between CIN status and gene expression is used to explore cancer treatment options and drug development. Singh et al.…”

“…In addition, Luxton et al. 11 used the XG boost model to analyze the relationship between CIN and genomic instability, thereby enabling risk assessment of the response of cancer patients to radiotherapy and adverse late health effects. Although machine learning-based methods have obtained relatively successful results for the association between the CIN and tumors, these studies were based on dozens of samples.…”