An immune infiltration signature to predict the overall survival of patients with colon cancer

Immune infiltration of tumors has been increasingly accepted as a prognostic factor in colon cancer. Here, we aim to develop a novel immune signature, based on estimated immune landscape from tumor transcriptomes, to predict the overall survival of patients with colon cancer. The compositions of 22 immune cell subtypes from three microarray datasets were characterized with the CIBERSORT deconvolution algorithm. A prognostic immunoscore (PIS) model for overall survival prediction was established by using least absolute shrinkage and selection operator (LASSO) penalized regression analysis. A total of 17 immune cell markers were screened out in the LASSO model and were then aggregated to generate the PIS. In the training cohort (n = 490), patients with high PIS exhibited a remarkably poorer overall survival than those with low PIS. Similar results were obtained in patients with different TNM stages and in patients receiving adjunctive chemotherapy or not. Multivariate Cox regression indicated that the PIS was an independent predictor for overall survival in colon cancer (hazard ratio: 2.734, 95% confidence interval: 2.052-3.643, p < .001). The prognostic capability of PIS was also confirmed in the testing cohort (n = 245) and the entire cohort (n = 735). As for biological implications, the PIS was significantly associated with some immune checkpoints, inflammatory factors, epithelial-mesenchymal transformation regulators, and many known signaling pathways in cancer. The results of our study provide a novel and promising immune signature for overall survival prediction of patients with colon cancer.