Background and Aims: Colorectal cancer (CRC) is an aggressive disease with late diagnosis and poor prognosis. There is growing evidence suggesting a prominent correlation between immune signature and CRC. This study was aimed to establish an immune-related gene pairs (IRGP) signature for predicting the outcomes of CRC patients. 
Methods: The gene expression profiles and clinical information of CRC patients were extracted from TCGA and GEO databases, and were then divided into a training dataset (TCGA-COAD) and a validation dataset (GSE39582). Immune-related genes (IRGs) were downloaded from the ImmPort database for screening of the IRGs in the training dataset and the validation dataset. Paired comparison of the expression values of IRGs in each sample was performed, and the final immune-related gene pairs (IRGP) signature [immune-related gene pair index (IRGPI)] was constructed by Lasso-Cox proportional hazard model with an iteration number of 1 000. Then, the ROC curve of the IRGP signature was applied to split CRC patients into high and low-risk groups, followed by analysis of the survival states of the two groups of patients using Kaplan-Meier curves and Log-rank test. Simultaneously, the predictive ability of the signature was evaluated using univariate and multivariate Cox proportional hazards regression models. Subsequently, the infiltration conditions of immune cells in CRC patients were identified on the tumor samples using CIBERSORT through deconvolution algorithms. Finally, functional annotation and analyses of the model were performed to further understand its biological functions.
Results: Twenty IRGPI containing 28 IRGs were successfully constructed (CXCL14|BST2, RBP1|STC2, RBP7|PTGS2, RBP7|ARG2, RBP7|IL7, APOD|IL17RB, GNAI1|GRP, CCL4|INHBB, CCL28|INHBB, ABCC4|GRP, ARG2|GRP, CCR7|INHBB, CD86|IL7, OLR1|IL7, C5AR1|NR3C2, INHBB|PDGFC, STC2|HNF4G, IL10RA|TNFRSF11A, RORC|PRKCQ, TNFRSF11A|LCK), which were significantly associated with the prognosis of CRC patients. In the training dataset, the overall survival of CRC patients in high-risk group was shorter than that of CRC patients in the low-risk group (P=1.295×10–11, HR=6.51, 95% CI=3.79–11.21), and the similar result was obtained in the validation dataset (P=0.000 1, HR=1.82, 95% CI=1.36–2.44). Univariate and multivariate Cox analysis verified IRGPI as an independent prognostic factor for CRC (training dataset: HR=3.270, 95% CI=2.555–4.186, P<0.001 and HR=3.008, 95% CI=2.295–3.941, P<0.001; validation dataset: HR=1.278, 95% CI=1.10–1.474, P<0.001 and HR=1.189, 95% CI=1.024–1.380, P=0.023). According to the analysis of tumor-infiltrating immune cells, the regulatory T cells (P=0.007) and macrophages (P=0.024) in high-risk group were significantly higher than those in low-risk group, while the resting dendritic cells (P=0.006), resting memory CD4+T cells (P=0.006), and resting macrophages (P=1.784×10e–05) were significantly increased in low-risk group. The results of functional annotation indicated that IRGs correlate with certain biological processes, which included the leukocyte migration, cell chemotaxis, cytokine-cytokine-receptor interaction, etc. Further, the related pathways with significant differences between high and low-risk groups included the keratinocyte differentiation, keratinization, epidermal cell differentiation, and skin development, etc.
Conclusion: A IRGP signature for evaluating the prognosis of CRC patients is successfully constructed, which may provide novel insights into the diagnosis and treatment of CRC.