A signature predictive of early versus late recurrence after radiation (RT) for breast cancer that may inform the biology of early, aggressive recurrences.

112 Background: Unmet clinical needs in breast cancer (BC) management include the identification of patients at high risk to fail locally despite standard local therapy and an understanding of the biology of these recurrences. We previously reported a radiation response signature and here extend those studies to identify a signature predictive of timing of recurrence after RT. Methods: 2 independent patient cohorts were used for training (119 pts) and validation (112 pts). All patients received RT after BCS and systemic therapy as appropriate. Spearman’s rank correlation to correlate gene expression to recurrence time was used for feature selection. Significant genes were used to train a linear model which was locked before validation. Cox regression was used for both UVA and MVA. Results: Spearman’s correlation identified 485 genes whose expression was significantly associated with recurrence time (+/-3 yrs). Feature reduction refined the list to 41 genes retained within the signature. In training, the correlation of score to recurrence time was 0.85, p-value < 1.3x10-31; AUC of 0.91. External validation in an independent BC validation set accurately identified patients with early vs. late recurrences (correlation= 0.75, p-value = 0.001, AUC = 0.92, sens.=0.75, spec.= 1.0, PPV = 1.0, NPV = 0.8). Unique associations of breast cancer intrinsic subtype to timing of local recurrence were found. In UVA and MVA the signature remained the most significant factor associated with recurrence. GSEA analysis of the 41 genes retained within the signature identified proliferation and EGFR concepts associated with early recurrences and luminal and ER-signaling pathways associated with late recurrences. Knockdown of genes associated with the early and late recurrences demonstrated novel effects on proliferation and clonogenic survival, respectively. Conclusions: We report a BC gene expression signatures that may be useful in identifying patients unlikely to respond to adjuvant RT and may be used to predict timing of recurrences, with implications for potential treatment intensification and duration of follow-up for women with breast cancer treated with RT.