A frequency-based gene selection method to identify robust biomarkers for radiation dose prediction

Purpose : A fast, radiation-specific and highly accurate prediction of the radiation dose of accidentally exposed individuals is essential for medical decision-making. The aim of the present study is to identify small gene signatures allowing the discrimination between low and medium dose exposure of low linear energy transfer (LET)-radiation. Material and methods : We developed a framework for dose prediction using a frequency-based gene selection approach, based on a p-value and fold-change criterion applied to microarray expression data. A repeated cross-validated classification guarantees unbiased performance results. Human blood from six healthy donors was irradiated ex vivo with 0.5, 1, 2 and 4 Gy (Cs-137 gamma-rays). Expression levels of isolated blood lymphocytes were measured at 6, 24 and 48 h after irradiation. Results : We identified radiation-responsive genes, most of them functionally linked to apoptosis, DNA-damage or cell-cycle regulation. We extracted small subsets of genes, with which 95.7% of all samples can be correctly predicted, regardless of the time post irradiation. Seven of these genes were used for validation by Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR). Conclusion : The genes identified are potential robust biomarkers, which are particularly suitable for dose level discrimination at a window of time that would be appropriate for life-saving medical triage.