Abstract PS1-01-05: Pneumonitis but not other immune related adverse events (irAEs) associate with pre-treatment immune signatures in early-stage breast cancer patients receiving neoadjuvant immunotherapy (IO): results from 5 IO arms in I-SPY2

Abstract Background Neoadjuvant immunotherapy has become standard of care for triple negative (TN) breast cancer; and I-SPY2 and other trials have shown efficacy of IO in HR+HER2-, especially in immune+ subsets. However, IO agents carry risk of immune-related toxicities, some long-term. Previously we developed Imprint, an IDE-enabled immune classifier predicting response to IO now being used in I-SPY2.2 as part of the Response Predictive Subtypes. Here we pool irAE data from 5 IO arms in I-SPY2, where both TN and HR+HER2- patients enrolled, to assess whether irAE development may be predicted by gene expression signatures in pre-treatment tumor biopsies. Long-term, the goal is to develop clinical decision support tools to help balance likelihood of response to IO with likelihood of irAE development. Methods 356 patients (206 HR+HER2- and 150 TN) across 5 IO arms (69 pembrolizumab (Pembro), 76 Pembro/SD101, 73 durvalumab/olaparib, 62 cemiplimab (Cemi), and 78 Cemi/LAG3) with irAE and pCR data were analyzed. irAEs considered included adrenal insufficiency (10%; n=36), colitis (2.3%; n=8), diabetes mellitus (0.3%; n=1), hyperthyroidism (4.2%; n=15), hypothyroidism (15%; n=53), hypophysitis (1% ; n=4), pneumotitis (5.3%; n=19), and thyroiditis (1.4%; n=5) (unresolved grade 1 or grades 2/3). 32 mostly-immune (7 immune checkpoints, 14 immune cells, 3 TcClassII, 4 chemokine/cytokine, 1 interferon, 1 TGFB, 1 ESR1/PGR, and 1 proliferation) genes/signatures were evaluated from pre-treatment mRNA from Agilent 44K arrays for 344 patients. Associations between irAEs (overall/any and individual) and signatures were assessed using logistic regression, overall and within HR+HER2-, TN, and ImPrint+/- subsets. P-values were adjusted for multiple hypothesis testing using the Benjamini-Hochberg method (significance BH p