A Novel Computational Framework for Predicting the Survival of Cancer Patients With PD-1/PD-L1 Checkpoint Blockade Therapy

Abstract

BackgroundImmune checkpoint inhibitors (ICIs) induce durable responses, but only a minority of patients achieve clinical benefits. The development of gene expression profiling of tumor transcriptomes has enabled identifying prognostic gene expression signatures and patient selection with targeted therapies.MethodsImmune exclusion score (IES) was built by elastic net-penalized Cox proportional hazards (PHs) model in the discovery cohort and validated via four independent cohorts. The survival differences between the two groups were compared using Kaplan-Meier analysis. Both GO and KEGG analyses were performed for functional annotation. CIBERSORTx was also performed to estimate the relative proportion of immune-cell types.ResultsA fifteen-genes immune exclusion score (IES) was developed in the discovery cohort of 65 patients treated with anti-PD-(L)1 therapy. The ROC efficiencies of 1- and 3- year prognosis were 0.842 and 0.82, respectively. Patients with low IES showed a longer PFS (p=0.003) and better response rate (ORR: 43.8% vs 18.2%, p=0.03). We found that patients with low IES enriched with high expression of immune eliminated cell genes, such as CD8+ T cells, CD4+ T cells, NK cells and B cells. IES was positively correlated with other immune exclusion signatures. Furthermore, IES was successfully validated in four independent cohorts (Riaz’s SKCM, Liu’s SKCM, Nathanson’s SKCM and Braun’s ccRCC, n = 367). IES was also negatively correlated with T cell–inflamed signature and independent of TMB.ConclusionsThis novel IES model encompassing immune-related biomarkers might serve as a promising tool for the prognostic prediction of immunotherapy.