Fragmentomic Liquid Biopsy Enables Non-invasive Detection, Molecular Subtyping and Lymph Node Assessment in Early Breast Cancer

Breast cancer (BC) remains a leading global health concern in women. While mammography is the standard screening tool, its efficacy is limited by the high breast density and accessibility challenges in China. Here, we conducted a multicenter case-control study (NCT06016790) involving 503 patients with malignant BC and 289 benign controls across seven hospitals to develop TuFEst, a machine learning model utilizing multi-feature cell-free DNA (cfDNA) fragmentomics. TuFEst showed excellent early detection performance (95.0% sensitivity and 78.3% specificity), maintaining 96.2% accuracy in an imaging-pathological inconsistency cohort (n=26). To broaden its clinical application, we extended TuFEst to molecular subtyping (TuFEst-MS) and lymph node metastasis prediction (TuFEst-LN). TuFEst-MS yielded AUCs of 0.906 (ER ⁺ /PR ⁺ HER2 ⁻ ), 0.925 (HER2 ⁺ ), and 0.891 (triple-negative) with 85.7% accuracy in the oligometastatic validation cohort (n=21). TuFEst-LN achieved a negative predictive value (NPV) of 95.2%, which improved to 97.6% in an independent cohort (n=124) with discordant axillary imaging pathology. RNA-seq of paired bulk tumor samples (n=79) demonstrated that elevated TuFEst-derived cancer scores were associated with aggressive tumor characteristics, particularly enriched immune responses and epithelial-mesenchymal transition (EMT) signatures, emphasizing the clinical importance of early detection. Our study established cfDNA fragmentomics as an integrated liquid biopsy solution for BC management, enabling concurrent detection, molecular subtyping, and lymph node evaluation with transformative clinical potential.