Tumor microenvironment subtypes and immune-related signatures for the prognosis of breast cancer

Abstract To better understand the heterogeneity of tumor microenvironment (TME) and establish a prognostic model for breast cancer in clinical practice, the leukocyte infiltrations of 22 cell types of interest from 2620 breast cancer patients were quantitatively estimated using deconvolution algorithms, and three TME subtypes with distinct molecular and clinical features were identified by unsupervised clustering approach. Then, we carried out systematic analyses to illustrate the contributing mechanisms for differential phenotypes, which suggested that the divergences were distinguished by cell cycle dysfunction, variation of cytotoxic T lymphocytes activity. Next, through dimensionally reduction and selection based on random-forest analysis, least absolute shrinkage and selection operator (LASSO) analysis, and uni- and multivariate COX regression analysis, a total of 15 significant genes were proposed to construct the prognostic immune-related score (pIRS) system and, in combinations with clinicopathological characteristics, a predictive model was ultimately built with well performance for survival of breast cancer patients. Comparative analyses demonstrated that proactivity of CD8 T lymphocytes and hyper-angiogenesis could be attributed to distinct prognostic outcomes. In conclusion, we retrieved three TME phenotypes and the curated prognostic model based on pIRS system for breast cancer. This model is justified for validation and optimized in the coming future.