Construction of the immune landscape of durable response to checkpoint blockade therapy by integrating publicly available datasets

Abstract

Background Immune checkpoint blockade (ICB) has revolutionized cancer treatment. However, long-term benefits are only achieved in a small fraction of patients. Understanding the mechanisms underlying ICB activity is key to improving the efficacy of immunotherapy. A major limitation to uncovering these mechanisms is the limited number of responders within each ICB trial. Integrating data from multiple studies of ICB would help overcome this issue and more reliably define the immune landscape of durable responses. Towards this goal, we formed the TimIOs consortium, comprising researchers from the Society for Immunotherapy of Cancer Sparkathon TimIOs Initiative, the Parker Institute of Cancer Immunotherapy, the University of North Carolina-Chapel Hill, and the Institute for Systems Biology. Together, we aim to improve the understanding of the molecular mechanisms associated with defined outcomes to ICB, by building on our joint and multifaceted expertise in the field of immuno-oncology. To determine the feasibility and relevance of our approach, we have assembled a compendium of publicly available gene expression datasets from clinical trials of ICB. We plan to analyze this data using a previously reported pipeline that successfully determined main cancer immune-subtypes associated with survival across multiple cancer types in TCGA.1 Methods RNA sequencing data from 1092 patients were uniformly reprocessed harmonized, and annotated with predefined clinical parameters. We defined a comprehensive set of immunogenomics features, including immune gene expression signatures associated with treatment outcome,1,2 estimates of immune cell proportions, metabolic profiles, and T and B cell receptor repertoire, and scored all compendium samples for these features. Elastic net regression models with parameter optimization done via Monte Carlo cross-validation and leave-one-out cross-validation were used to analyze the capacity of an integrated immunogenomics model to predict durable clinical benefit following ICB treatment. Results Our preliminary analyses confirmed an association between the expression of an IFN-gamma signature in tumor (1) and better outcomes of ICB, highlighting the feasibility of our approach.

Conclusions In line with analysis of pan-cancer TCGA datasets using this strategy (1), we expect to identify analogous immune subtypes characterizing baseline tumors from patients responding to ICB. Furthermore, we expect to find that these immune subtypes will have different importance in the model predicting response and survival. Results of this study will be incorporated into the Cancer Research Institute iAtlas Portal, to facilitate interactive exploration and hypothesis testing.