Stemness Refines the Classification of Colorectal Cancer With Stratified Prognosis, Multi-Omics Landscape, Potential Mechanisms, and Treatment Options

Abstract

Background: Stemness refers to the capacities of self-renewal and repopulation, which contributes to the progression, relapse, and drug resistance of colorectal cancer (CRC). Mounting evidence has established the links between cancer stemness and intratumoral heterogeneity across cancer. Currently, the intertumoral heterogeneity of cancer stemness remains elusive in CRC.

Methods: This study enrolled four CRC datasets, two immunotherapy datasets, and a clinical in-house cohort. Non-negative matrix factorization (NMF) was performed to decipher the heterogeneity of cancer stemness. Multiple machine learning algorithms were applied to develop a nine-gene stemness cluster predictor. The clinical outcomes, multi-omics landscape, potential mechanisms, and immune features of the stemness clusters were further explored.

Results: Based on 26 published stemness signatures derived by alternative approaches, we decipher two heterogeneous clusters, low stemness cluster 1 (C1) and high stemness cluster 2 (C2). C2 possessed a higher proportion of advanced tumors and displayed worse overall survival and relapse-free survival compared with C1. The MSI-H and CMS1 tumors tended to enrich in C1, and the mesenchymal subtype CMS4 was the prevalent subtype of C2. Subsequently, we developed a nine-gene stemness cluster predictor, which robustly validated and reproduced our stemness clusters in three independent datasets and an in-house cohort. C1 also displayed a generally superior mutational burden, and C2 possessed a higher burden of copy number deletion. Further investigations suggested that C1 enriched numerous proliferation-related biological processes and abundant immune infiltration, while C2 was significantly associated with mesenchyme development and differentiation. Given results derived from three algorithms and two immunotherapeutic cohorts, we observed C1 could benefit more from immunotherapy. For patients with C2, we constructed a ridge regression model and further identified nine latent therapeutic agents, which might improve their clinical outcomes.

Conclusions: This study proposed two stemness clusters with stratified prognosis, multi-omics landscape, potential mechanisms, and treatment options. Current work not only provided new insights into the heterogeneity of cancer stemness, but also shed light on optimizing decision-making in immunotherapy and chemotherapy.

Keywords: colorectal cancer; immunity; machine learning; multi-omics; stemness.