Feed

Tumor cell heterogeneity is a fundamental driver of breast cancer aggressiveness, underlying recurrence, metastasis, and therapy resistance. Understanding the biological characteristics and functions of specific tumor cell clusters in the tumor microenvironment is crucial for advancing precision oncology.

We delineated breast cancer tumor cell heterogeneity by integrating single-cell transcriptomics, spatial transcriptomics, bulk transcriptomics, genomic and radiomic data. The oncogenic functions of the candidate gene TMSB10 were rigorously validated in vitro. To advance individualized patient management, we employed machine learning to develop a non-invasive MRI radiomic model for estimating tumor cluster abundance and a robust prognostic signature for risk stratification.

We discovered a poor-prognosis tumor cell cluster (C1 cluster). C1 cluster exhibited a late evolutionary state, metabolic reprogramming (OXPHOS/glycolysis), and active crosstalk with cancer-associated fibroblasts and endothelial cells. High abundance of C1 cluster was associated with poor survival, specific somatic mutations, and predicted superior response to immune checkpoint blockade, but not to chemo/radiotherapy. The radiomic model based on MRI images was exploratively established for estimating the abundance of C1, and the prognostic model based on C1-derived genetic features significantly stratified the survival risk of breast cancer in multiple cohorts. In vitro experiments confirmed that TMSB10, a C1 core gene, promotes proliferation, migration, invasion.

This study revealed C1 cluster as a key driver of breast cancer progression and its application for predicting immunotherapy response. Additionally, TMSB10 was identified as a functional effector of C1 cluster, providing a new and applicable clinical tool for non-invasive detection and prognostic stratification for breast cancer.