Using prediction models to identify miRNA-based markers of low dose rate chronic stress

Abstract Robust biomarkers of exposure to chronic low dose stressors such as ionizing radiation, particularly following chronic low doses and dose-rates, are urgently needed. MicroRNAs (miRNA) have emerged as promising markers of exposure to high dose and dose-rate. Here, we evaluated the feasibility of classifying γ-radiation exposure at different dose rates based on miRNA expression levels. Our objective was to identify miRNA-signatures discriminating between exposure to γ-radiation or not, including exposure to chronic low dose rates. We exposed male CBA/CaOlaHsd and C57BL/6NHsd wild-type mice to 0, 2.5, 10 and 100 mGy/h γ-irradiation (3 Gy total-dose). From an initial screening of 576 miRNAs, a set of 21 signature-miRNAs was identified based on differential expression (>± 2-fold or p  To identify miRNAs that distinguished exposed mice from controls, nine prediction methods; i.e., six generalized regression, random-forest, boosted-tree and nearest-shrunken-centroid (PAM) were used. Considering the small dataset used, the generalized regression methods seem to outperform the other prediction methods for classification of irradiated and control samples. Using the 21-miRNA panel in the prediction models, we identified sets of candidate miRNA-markers that predict exposure to γ-radiation. Among the top10 miRNA predictors, contributing most in each of the three γ-irradiated groups, three miRNA predictors (miR-140-3p, miR-133a-5p and miR-145a-5p) were common. Three miRNAs, miR-188-3p/26a-5p/26b-5p, were specific for lower dose-rate γ-radiation. Similarly, exposure to the high dose-rates was also correctly predicted, including mice exposed to X-rays. Our approach identifying miRNA-based signature panels may be extended to classify exposure to environmental, nutritional and life-style-related stressors, including chronic low-stress scenarios.