While intratumor heterogeneity (ITH) is recognized as a fundamental driver of treatment resistance and disease progression in triple-negative breast cancer (TNBC), existing prognostic models often fail to quantify ITH as a biomarker robustly. This study obtained a training set from the METABRIC, while validation sets were sourced from the GEO database. ITH scores were calculated using the DEPTH2 algorithm, a computational method to infer heterogeneity from bulk transcriptomic data. Patients were stratified into distinct high-ITH and low-ITH groups based on these scores. Ten machine learning algorithms were integrated to generate 101 modeling approaches in the training set. The StepCoxbackward + GBM model, incorporating seven ITH-related signature (IRS) genes (IRF8, MAGEA4, IL18R1, IL7R, TBC1D10C, TEX261, and CD6), was selected as the optimal prognostic model. Then, we validated the cellular distribution and intercellular communication networks of core genes by single-cell sequencing data. Finally, quantitative real-time PCR (qRT-PCR) was performed on matched TNBC tumor and adjacent normal tissues to experimentally validate the differential expression of the IRS genes. The IRS model demonstrated robust prediction of patient overall survival (OS) and disease-free survival (DFS) in three independent cohorts. Patients in the high IRS score group showed worse prognosis, elevated Tumor Immune Dysfunction and Exclusion (TIDE) scores, lower Microsatellite Instability (MSI) scores, reduced expression of immune checkpoints, lower immune cell infiltration. And high IRS score was associated with poorer immunotherapy response in the GSE194040, GSE173839, GSE124821, and IMvigor210. The high and low-IRS score groups also exhibited differences in related functions and pathways. The high-risk group was significantly enriched in metabolic pathways, particularly N-Glycan biosynthesis and O-Glycan biosynthesis. Conversely, the low-risk group was predominantly associated with robust immune-related pathways, including adaptive immune response and immune response-regulating cell surface receptor signaling pathways. The IRS core genes were specifically distributed in CD8 + T cells and macrophages, and the interaction networks between CD8 + T cells and macrophages in the GALECTIN and CD45 signaling pathways were elucidated. Drug sensitivity analysis showed that the high-risk group was more sensitive to chemotherapeutic agents, especially to poly ADP-ribose polymerase (PARP)inhibitors. qRT-PCR validation confirmed the differential expression of seven critical genes between TNBC and adjacent normal tissues. Our study established an IRS model integrating machine learning and single-cell RNA sequencing. This model improves the accuracy of prognosis prediction for TNBC patients and effectively distinguishes risk subgroups with different immune microenvironment characteristics and treatment responses, providing preliminary data for precision medicine for TNBC.
Li et al. (Sat,) studied this question.