Abstract 1714: A multifactorial model of T-cell expansion and durable clinical benefit in response to a PD-L1 inhibitor

There is an urgent need to improve prediction of patient response to checkpoint inhibitor immunotherapy. Checkpoint inhibitors have achieved unprecedented successes in treating late-stage cancers, but response rates vary across patients, and known biomarkers for response such as high mutation load are not predictive for every patient. Predicting response to checkpoint inhibitors is a critical priority for avoiding adverse responses, identifying additional patients who may benefit, and accelerating the development of new treatments. A key challenge for predicting response is modeling features of the immune system and cancer simultaneously. Recently, clinicians have begun to collect a wealth of molecular tumor and immune system data before and during immunotherapy, but researchers have yet to model how molecular and clinical features interact to affect response. To address this challenge, we developed multifactorial models for response to checkpoint inhibitors. Our approach is based off of the elastic net–a machine learning method for regression that automatically selects informative features from the data–and models clinical, tumor, and immune system features simultaneously. We applied our model to data from Snyder et al. (PLoS Medicine, 2017), who measured mutations and gene expression in the tumor and T-cell receptor (TCR) sequences in the tumor and blood in 29 urothelial cancers treated with anti-PD-L1. Rather than model the clinical response of each patient directly, we modeled the response of each patient9s immune system and used the predicted immune responses to stratify patients based on expected clinical benefit. By modeling the immune response, we have the advantage of predicting fine-grained, molecular measurements that are predictive of clinical response. Our models predict immune response with high accuracy, and that predicted immune response is associated with durable clinical benefit (DCB). In held-out patients, our model explains 79% of the variance of the log number of T-cell clones in the tumor that expand in the blood post-therapy. The magnitude of the predicted expansion is associated with DCB, as 100% of held-out patients with DCB have predicted scores above the median score of held-out patients without DCB. Notably, mutation load alone did not demonstrate significant association with DCB. We also evaluated the importance of the tumor, immune system, and clinical features to our model. Our model requires all three feature classes to make accurate predictions and can explain at most 23% of the held-out patient variance when at least one of the classes is removed. Taken together, these results show that models for immune response may be useful for predicting clinical response to immunotherapy and that noninvasive TCR sequencing in the blood may be an effective way to monitor patient response. Citation Format: Mark D. Leiserson, Vasilis Syrgkanis, Amy I. Gilson, Miroslav Dudik, Samuel A. Funt, Alexandra Snyder, Lester Mackey. A multifactorial model of T-cell expansion and durable clinical benefit in response to a PD-L1 inhibitor [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1714.