Improving the prediction for the response to radiotherapy of clinical tumor samples by using combinatorial model of MicroRNA expression

Purpose: Radiation therapy (RT) is one of the main treatments for cancer. The response to radiotherapy varies widely between individuals and some patients have poor response to RT treatment due to tumor radioresistance. Stratifying patients according to molecular signatures of individual tumor characteristics can improve clinical treatment. In here, we aimed to use clinical and genomic databases to develop miRNA signatures that can predict response to radiotherapy in various cancer types. Methods: We analyzed the miRNAs profiles using tumor samples treated with RT across eight types of human cancers from TCGA database. These samples were divided into response group (S, n = 224) and progressive disease group (R, n = 134) based on RT response of tumors. To enhance the discrimination for S and R samples, the predictive models based on binary logistic regression were developed to identify the best combinations of multiple miRNAs. Results: The miRNAs differentially expressed between the groups S and R in each caner type were identified. Total 47 miRNAs were identified in eight cancer types (p values < 0.05, t-test), including several miRNAs previously reported to be associated with radiotherapy sensitivity. Functional enrichment analysis revealed that epithelial-to-mesenchymal transition (EMT), stem cell, NF-kappa B signal, immune response, cell death, cell cycle, and DNA damage response and DNA damage repair processes were significantly enriched. The cancer-type-specific miRNA signatures were identified, which consist of 2-13 of miRNAs in each caner type. Receiver operating characteristic (ROC) analyses showed that the most of individual miRNAs were effective in distinguishing responsive and non-responsive patients (the area under the curve (AUC) ranging from 0.606 to 0.889). The patient stratification was further improved by applying the combinatorial model of miRNA expression (AUC ranging from 0.711 to 0.992). Also, five miRNAs that were significantly associated with overall survival were identified as prognostic miRNAs. Conclusion: These mRNA signatures could be used as potential biomarkers selecting patients who will benefit from radiotherapy. Our study identified a series of miRNA that were differentially expressed between RT good responders and poor responders, providing useful clues for further functional assays to demonstrate a possible regulatory role in radioresistance.