While genomic assays such as Oncotype DX (ODX) effectively guide adjuvant chemotherapy in ER+/HER2- breast cancer (BC), they inherently overlook prognostic drivers within the tumour microenvironment (TME), leaving a subset of clinically 'low-risk' patients vulnerable to recurrence. While the vascular niche is a critical regulator of tumour progression, the prognostic significance of perivascular mural cell signalling in early-stage BC remains undefined.

In a discovery cohort of 302 ER+/HER2- BC patients profiled by ODX, mural β3-integrin protein abundance was quantified using Cell Dive multiplex imaging and HALO AI-driven analysis. Spatial transcriptomics (NanoString GeoMx) in ODX low-risk tumours was employed to derive a stromal gene signature defined by differential expression between mural β3-integrin ‘high’ and ‘low’ patients. The prognostic utility of this signature was subsequently validated in the METABRIC cohort, restricted to the clinically relevant ER+/HER2- population, utilising in silico ODX scores derived via the genefu package.

Perivascular mural β3-integrin protein expression signposts a stromal gene signature that sub-stratifies patients considered clinically ‘low-risk’. In the discovery cohort, high mural β3-integrin protein expression independently predicted relapse within the ODX low-risk population (HR 9.28, P = 0.005). Spatial transcriptomic profiling of this population identified an analogous stromal 6-gene signature that achieved complete risk stratification, with no relapse events observed in the ‘low’ signature group. Validation in the METABRIC cohort, filtered for ODX low-risk patients (n = 274), confirmed the signature’s robustness, with ‘high’ signature expression predicting worse relapse-free survival (HR 2.38, P = 0.015), independent of standard clinicopathological features.

These findings highlight a distinct, stroma-driven risk profile that operates independently of standard epithelial assays. Consequently, this signature provides a clinically viable classifier to refine ODX stratification, guiding therapeutic escalation for BC patients currently considered at low risk of recurrence.

For sample collection, permission has been obtained from the relevant regulatory authority and properly informed consent given where appropriate. Alex Jordan is supported by a Cancer Research UK programme grant (CRUK DRCNPG- May21/100004). We gratefully acknowledge the invaluable support of Luke Gammon and Ryan Wallis (Phenotypic Screening Facility, Queen Mary University of London) for their technical assistance.