Integrative profiling of multi-modal plasma cfRNA signatures enables detection and prognostic risk stratification in breast cancer.

Abstract

Breast cancer requires non-invasive biomarkers for accurate detection and risk stratification. We comprehensively profiled plasma cell-free RNA (cfRNA) from 41 patients with malignant and 42 with benign breast lesions using SLiPiR-seq. Multiple cfRNA subtypes displayed distinct expression patterns, and machine-learning models were developed with repeated stratified four-fold cross-validation. The integrated cfRNA model achieved a mean AUC of 0.795, while the cf-miRNA model performed best (AUC: 0.814) and was externally validated in an independent cohort (AUC: 0.867). A three-gene tissue expression signature comprising DLST, DOCK4, and EGFL7 further stratified patients by overall survival in TCGA-BRCA. High-risk tumors showed increased PIK3CA mutations, PI3K-AKT pathway activation, and altered immune features. Single-cell analysis revealed distinct localization of the three genes across epithelial cells, tumor-associated macrophages, cancer-associated fibroblasts, and T-cell populations. Collectively, multi-modal plasma cfRNA profiling may enable non-invasive breast cancer detection and provide candidate markers for prognostic risk stratification and future precision oncology applications.