The ImPrint Immune Signature to Identify Patients with High-Risk Early Breast Cancer Who May Benefit from PD1 Checkpoint Inhibition in I-SPY2.

514 Background: The remarkable increase of novel Immuno-Oncology drugs in many malignancies has led to the need for biomarkers to identify who would benefit. Various predictive biomarkers have been developed (PD-1/PD-L1 expression, mutations in mismatch repair genes and microsatellite instability, tumor mutational burden and immune infiltration), none have consistently predicted efficacy. The I-SPY2 consortium qualified several expression-based immune biology related signatures that predict response to PD1 checkpoint inhibition. Here we assessed whole transcriptome data of high-risk early-breast cancer (EBC) patients who received Pembrolizumab within the neoadjuvant biomarker-rich I-SPY2 trial (NCT01042379), aiming to migrate the I-SPY2 research findings to a robust clinical grade platform signature to predict sensitivity to PD1 checkpoint inhibition. Methods: Whole transcriptome microarray data were available from pre-treatment biopsies of 69 HER2- patients enrolled in the Pembrolizumab (4 cycles) arm of the I-SPY2 trial. All patients had a High-Risk 70-gene MammaPrint profile. Pathologic complete response (pCR) was defined as no residual invasive cancer in breast or nodes at the time of surgery. Of the 69 patients, 31 had a pCR (12 HR (hormonal receptor)+HER2-, 19 Triple Negative (TN)), while 38 (28 HR+HER2-, 10 TN) had residual disease (RD). To identify the most predictive genes associated with pCR, gene selection was performed comparing pCR and RD groups by iteratively splitting the dataset in training and test, balancing for HR status. Due to limited sample size, leave one out cross validation was used for performance assessment. Genes with effect size > 0.45 were considered significant. Results: A signature of 53 genes, named ImPrint, was identified with overall sensitivity and specificity > 90% and > 80% for predicting pCR to pembrolizumab in all patients. Sensitivity and specificity in TN were > 95% and ≥70%, and in HR+HER2- > 80% and > 85%, respectively. The Positive Predictive Value (PPV) is 77% for the HR+HER2- subgroup. Biological annotation of the 53 genes showed that over 90% of the genes have known immune system related functions, of which 63% were previously known to be involved in immune response (including genes coding PD-L1 and PD-1, as well as those identified in I-SPY2). Conclusions: In the signature development phase, ImPrint predicts pCR to Pembrolizumab in a set of 69 high risk EBC with high sensitivity and specificity. The signature features genes with immune-related functions known to be involved in immune response indicating that it might aid identifying patients with an immune-active phenotype. Importantly, ImPrint appears effective in identifying a subset of HR+HER2- patients who could benefit from immunotherapy. External validation in independent dataset(s) is ongoing and will be presented at the time of the meeting.