Abstract 3413: Transfer learning identifies common cellular determinants of immune checkpoint inhibitor response between preclinical tumor models and patients

While immune checkpoint inhibitors (ICIs) have revolutionized cancer therapy, only a small percentage of patients demonstrate therapeutic response. Understanding the cellular and molecular mechanism of action of ICIs could help identify novel biomarkers of ICI response and therefore select patients who are more likely to obtain a clinically durable response. To identify the unknown molecular interactions between cancer and immune cells that shape therapeutic response, we leveraged single-cell RNA-sequencing (scRNA-seq) of CD45+ intratumoral cells in a preclinical sarcoma model. Our unsupervised learning method CoGAPS detected gene signatures associated with distinct cellular responses to anti-PD-1 and anti-CTLA-4 treatment, either monotherapy or in combination. Notably, this analysis identified unanticipated synergy between PD1 and CTLA4 blockade in combination, including upregulation of antigen presentation and downregulation of regulatory T cell pathways. A subset of dual treated immune cells also revealed a signature consisting of autoimmune and steroidogenic genes, which we hypothesize may be involved in immune related adverse effects. Annotating the cellular landscape of these molecular patterns identified novel interactions between NK cells and CTLA4 blockade. Importantly, NK cells in the monotherapy treated group showed upregulation of granzymes and perforin, suggesting enhanced activation and degranulation as a mechanism of ICI response. To examine the relevance of this signature in humans, we first assessed the correlation between CTLA4 expression and CIBERSORT cell type classification in cancers FDA approved for ICI therapy from The Cancer Genome Atlas. This revealed a positive correlation between CTLA4 and NK cell activation. RT-qPCR and western blot analysis in human NK cell lines and donor cells confirmed CTLA4 expression. We are currently expanding these results to assess the functional role of CTLA4 in the anti-tumor response of human NK cells during ICI treatment. Additionally, we relate learned therapeutic response signatures to patients through our transfer learning method, projectR. We queried the occurrence of our preclinical signatures in scRNA-seq of human CD45+ intratumoral cells collected from 48 ICI treated melanoma patients. The NK signature was highest in patients responsive to CTLA4 blockade, either single or dual ICI. We also identified T cell specific transcriptional signatures associated with upregulation of T cell trafficking and proliferation pathways enriched in patients who were responsive to ICI. Notably, these signatures were present prior to treatment, indicating their potential to predict clinical responders to ICI. Collectively, our data identify clinically relevant cellular perturbations in ICI responders and contribute to our mechanistic understanding of ICI response. Citation Format: Emily F. Davis-Marcisak, Allison A. Fitzgerald, Neeha Zaidi, Elizabeth M. Jaffee, Louis M. Weiner, Elana J. Fertig. Transfer learning identifies common cellular determinants of immune checkpoint inhibitor response between preclinical tumor models and patients [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3413.