Evaluation of stromal cell infiltration in the tumor microenvironment enable prediction of treatment sensitivity and prognosis in colon cancer

Abstract

Current clinical factors for screening candidates that might benefit from adjuvant chemotherapy in colon cancer are inadequate. Tumor microenvironment, especially the stromal components, has the potential to determine treatment response. However, clinical translation of the tumor-associated stromal characterization into a practical biomarker for helping treatment decision has not been established. Using machine learning, we established a novel 31-gene signature, called stromal cell infiltration intensity score (SIIS), to distinguish patients characterized by the enrichment of abundant stromal cells in five colon cancer datasets from GEO (N = 990). Patients with high-SIIS were at higher risk for recurrence and mortality, and could not benefit from adjuvant chemotherapy due to their intrinsic drug resistance; however, the opposite was reported for patients with low-SIIS. The role of SIIS in detection of patients with high stromal cell infiltration and reduced drug efficiency was consistently validated in the TCGA-COAD cohort (N = 382), Sun Yat-sen University Cancer Center cohort (N = 30), and could also be observed in TCGA pan-cancer settings (N = 4898) and four independent immunotherapy cohorts (N = 467). Based on multi-omics data analysis and the CRISPR library screen, we reported that lack of gene mutation, hypomethylation in ADCY4 promoter region, activation of WNT-PCP pathway and SIAH2-GPX3 axis were potential mechanisms responsible for the chemoresistance of patients within high-SIIS group. Our findings demonstrated that SIIS provide an important reference for those making treatment decisions for such special patients.

Keywords

• Stromal cell infiltration
• Machine learnin
• Gene signature
• Treatment sensitivity
• CRISPR library screen
• Colon cancer