Method for Determining Radioresistance of Cancer Cell Lines Based on Cluster Analysis of Clonogenic Cell Survival Data

Abstract

Background: The outcome of radiation therapy, the duration and quality of life of cancer patients significantly depend on the radiosensitivity (RS) of a cancerous tumor, and the duration of the patient’s relapse-free period is largely determined by the degree of its radioresistance (RR). Today the results of molecular mechanism investigation of cancer radioresistance and the classification of cancer cells according to their radiophenotypes mostly contribute to improving prognosis methods of treatment outcomes and increasing effectiveness of radiation therapy. In this work, we developed a classification method of cancer cells according to their radiosensitivity using machine learning based on the data analysis of clonogenic cell survival under ionizing radiation. Material and methods: The method consists of clustering parameters of experimental dose-effect relationships, which were approximated using the equation of a linear-quadratic (LQ) model, which is used to evaluate RS of cancer cells in radiobiology. The training of the statistical model included published experimental dataset of 96 cancer cell lines, for which parameters a, b and their ratio a/b of the LQ model were determined. Classification of cancer cells according to their radiosensitivity was carried out based on principal component analysis (PCA) in the parameter space (a, a/b), k-means clustering and hierarchical clustering methods. Results: Application of the developed statistical model to a large dataset of cancer cells made it possible to reliably separate radiosensitive and radioresistant (RR) cells into two clusters according to the parameters a and a/b. Application of the model to cancer cells with acquired RR, in which RS was suppressed as a result of exposure to irradiation or hypoxia, allowed tracing the shift of parent cells’ parameters from the RS cluster to the RR cell cluster. To study the genetic mechanisms of radiosensitivity, we performed bioinformatic analysis of the mutation distribution in genes encoding proteins in the cellular signalling pathways of cancer cells, i.e. proliferation, apoptosis, repair of damaged DNA molecules and antioxidant defence cellular system. Conclusion: The developed statistical model of radiophenotypic classification of cancer cells based on their radiosensitivity can be used in the development of radiation therapy treatment plans taking into account radiosensitivity of patient’s tumour. The model may be also helpful in a joint analysis of the phenotypic and genotypic characteristics of cancer cells, aiming at the elucidation of the molecular and genetic mechanisms of radiosensitivity and development of biomarkers of radioresistance.