Abstract Background: Germline variants in microRNAs (mirSNPs) have been found to shape tumor biology and to predict both outcome and response to therapy. Here, we tested the hypothesis that mirSNPs predict high versus low lymph node burden in breast cancer patients. Methods: DNA was isolated from microdissected tumor specimens from 73 breast cancer patients with invasive ductal carcinoma (IDC), who underwent axillary lymph node dissection. The mean age at diagnosis was 59.1 years (SD 12.9), and the cohort was racially diverse: 43.8% White, 23.3% Asian, 12.3% Black, 12.3% Hispanic, and 8.2% Middle Eastern. The average BMI was 27.3 (SD 7.3). The majority of patients had HR+/HER2- breast cancer (72.6%), with fewer HER2+ (15.1%) or triple-negative (12.3%) disease. Tumors were primarily grade 2 (41.1%) or grade 3 (46.6%), with a mean Ki67 proliferation index of 29.1% (SD 20.2). T2 was the most common clinical T stage (57.5%), and the mean tumor size at diagnosis was 28.3 mm (SD 19.7). Most patients were clinically node (cN) positive (91.8%), and 57.5% had four or more positive lymph nodes. Low nodal burden was defined as N1 (3 axillary lymph nodes) and high nodal burden N2/N3 (4 axillary lymph nodes). The germline mirSNP dataset was filtered prior to modeling, with 22 SNPs demonstrating an association with nodal burden at a Fisher exact p-value cutoff of 0.2. mirSNPs were further reduced by iteratively removing the bottom five predictors by AUC impact and re-fitting models, resulting in a final set of 17 mirSNPs. A set of clinical variables was selected by iterative boosted tree modeling, with features retained if they had non-zero median relative importance across five leave-one-out cross-validation (LOOCV) rounds. This resulted in retention of age at diagnosis, tumor size, BMI, race, Ki67, clinical T stage, cancer subtype (TNBC, HER2+, HR+/HER2-), tumor grade, IDC with lobular features, and use of adjuvant therapies (endocrine, chemotherapy, radiotherapy). These selected clinical features were then used alone and in conjunction with mirSNP genetic data in subsequent predictive modeling. Results: Elastic net regression using the 17 mirSNPs yielded strong predictive accuracy for high nodal burden, with AUCs ranging from 0.76 to 0.79 and specificity values between 0.81 and 0.87. Importantly, the predictive accuracy did not correlate with the number of nodes sampled, and misclassifications by the model were predominantly concentrated around the clinical threshold of four positive nodes- rather than being biased by the depth of nodal sampling. When mirSNPs were combined with clinical variables in the predictive models, the resulting performance was lower than for genetics alone, with AUCs ranging from 0.67 to 0.71. Models using only clinical variables performed the worst, achieving an AUC of 0.55. The most significant mirSNP in the model was in P2RX7, a gene whose expression has been previously associated with the risk of bone metastases in breast cancer, and the second most significant mirSNP in the model was in CSMD1, a gene that is linked to the spread of breast cancer to lymph nodes. Conclusions: Our findings support the potential of mirSNPs as an additional tool to help identify patients at increased risk of high nodal burden in breast cancer. Our mirSNP model outperformed clinical models in this data set. Studies investigating the benefit of incorporating mirSNPs with other biomarkers of lymph node burden, the association of this signature with outcome, as well as further validation of these findings are ongoing. Citation Format: J. Weidhaas, K. McGreevy, J. Le, M. Alcaraz, E. Rietdorf, M. Ensenyat-Mendez, J. Baker, M. Dinome, D. Marzese, D. Telesca. Germline microRNA-based variants and lymph node burden in breast cancer abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS4-05-13.
Paik et al. (Tue,) studied this question.