Development of a microRNA-based prognostic model for accurate prediction of distant metastasis in breast cancer patients

Abstract

Background: The attempt to exploit molecular subtyping for risk stratification in breast cancer patients has been only partially successful with a limited application in the clinical practice. In the BREMIR study, we aimed to identify a panel of miRNAs as prognostic biomarkers for breast cancer. We first confirmed the association of previously linked miRNAs with critical clinical parameters, then adopted a discovery-driven approach to identify novel biomarkers. Methods: miRNA expression was analyzed using the Affymetrix Gene Chip 4.0 array in a discovery cohort of 34 patients (3 with synchronous metastases, 14 who developed metastases after 10 years, and 17 who remained metastasis-free) and 6 controls. RT-qPCR validated selected miRNAs in an extended cohort (n = 223) with a median follow up of 6.6 years. A stepwise logistic regression model incorporating miRNA levels and clinicopathological features was developed to predict metastasis risk. Additionally, miRNA expression was assessed in plasma extracellular vesicles (EVs) using digital PCR in an independent cohort (n = 39). In silico enrichment analyses explored the functional role of relevant miRNAs in metastasis development. Results: Eight differentially expressed miRNAs were identified in the discovery cohort. In the extended cohort, miR-3916 and miR-3613-5p were the most effective in distinguishing patients who developed metastases. Higher miR-3916 expression was associated with reduced metastasis risk (OR = 0.42, 95%CI 0.23-0.70, p = 0.002), while higher miR-3613-5p expression was linked to increased risk (OR = 2.06, 95%CI 1.27-3.50, p = 0.005). Adding these miRNAs to a model with clinicopathological features improved discrimination (AUC = 0.85 vs. AUC = 0.76, p = 0.001). The model was effective across all breast cancer subtypes. In extracellular vesicles, miR-3613-5p was more abundant in tumors than benign lesions (p = 0.039), while miR-3916 was lower in metastatic samples than in non-metastatic tumors (p = 0.020). In-silico pathway enrichment analyses indicates their involvement in critical steps of the metastatic process including EMT plasticity, DNA damage response and metastatic niche formation. Conclusions: This is the first study integrating miRNA expression with clinicopathological features in a logistic model for breast cancer prognosis. While further validation is needed, our model shows promise as a prognostic tool across all breast cancer subtypes. In silico pathway enrichment analysis highlights miR-3613-5p and miR-3916 as critical regulators of metastasis development, underscoring the need for further investigation.