A computational approach to predict survival of cancer patients using aging-related genes in normal cells.

Abstract

As life expectancy around the world has steadily increased, the understanding of aging has been highlighted in recent decades. There have been few studies researching gene expression data from adjacent normal samples of cancer patients to explore aging, although cancer is one of the most threatening age-associated diseases. The present study aimed to identify tissue-specific gene expression changes in the normal tissue of cancer patients and to determine their effects on survival. We analyzed transcriptomic profiling data from normal tissue of 12 types of cancer patients in The Cancer Genome Atlas, including 587 samples of mRNA expression data and 575 samples of miRNA expression data. We catalogued genes and microRNAs where the expression level altered with chronological age in each tissue by using linear regression. We investigated the association between the expression of aging-related transcripts and survival status of cancer patients, showing a significant relationship in kidney renal cell carcinoma. In addition, we performed functional annotation analysis, yielding that aging-related genes were related to various biological processes, including the immune system, metabolic process, development process, and cell cycle. We constructed modules of aging-related genes via protein-protein interaction networks to reveal clusters which were made up of functionally associated genes, and investigated the relationship between survival. Furthermore, we identified 1,355 of differentially expressed genes in renal cancer, as well as finding links between aging and cancer. Among cancer-related pathways, several were found to be common with the aging-related pathways, giving a glimpse of how aging and cancer are associated. We built survival prediction models which employed the average expression of aging-related genes from normal tissue and/or differentially expressed genes in renal cancer from tumor tissue as input variables. The average expression of aging-related genes in normal tissue exhibited better prediction performance than differentially expressed genes in tumor tissue. In conclusion, our results suggested that aging-related genes from paired normal samples of cancer patients could be used as a prognostic indicator.