Molecular Subtypes and Risk Prediction Model Based on Malignant Cell Differentiation Trajectories in Breast Cancer.

Abstract

Breast cancer (BRCA) is characterised by complex cellular heterogeneity and differentiation hierarchies, which play a crucial role in bone metastasis and therapeutic resistance. However, existing classification systems remain inadequate in capturing these complexities, limiting their effectiveness in guiding treatment strategies. To address this gap, we integrated single-cell RNA-seq profiles, spatial transcriptomes, along with 1097 bulk RNA-seq profiles of TCGA-BRCA cohort to dissect the molecular landscape of BRCA. By performing UMAP analysis, we identified nine tumour clusters and three spatially distinct spot types (immune, stromal and malignant spots) and further delineated 11 differentiation states from 2493 malignant spots. Through clustering, monocle 2 pseudo-time and prognostic analyses, we identified the prognostic BRCA cell fate-related genes, then constructed a novel BRCA stratification system (four subtypes) with differential prognosis, biological plausibility and clinical significance. Also, least absolute shrinkage and selection operator (LASSO) regression analysis was performed for the BRCA cell fate-related genes in constructing a prognostic model. The model has modest accuracy and accordance (AUC = 0.708), which could distinguish BRCA patients into high-risk or low groups. With correlation analysis, regulation networks were constructed for different subtypes based on the key cell fate-related genes, transcription factors, metastasis-related pathways, immune components and so on, to investigate the regulatory relationships between primary BRCA and BRCA bone metastasis. Afterwards, we identified the most significant inhibitors (puromycin, MS-275, megestrol, aesculetin) for bone metastatic BRCA, which might have potential translational significance. In all, we developed a novel molecular stratification system for BRCA based on the cell fate-related markers of malignant cells, which offered strong translational potential for diagnosis, prognosis and personalised therapeutic interventions.