Abstract 4150: Enriching for adenosine antagonist patient responses through deep learning

Abstract

Abstract Background: Adenosine signaling is a key metabolic pathway regulating tumor immunity. The conversion of inflammatory extracellular ATP into immunosuppressive adenosine invokes signaling of the A2a receptor (A2AR) in the tumor microenvironment. This dampens immune responses and creates a pro-tumor niche. A number of novel IO drugs targeting the adenosine pathway through inhibition of the ectonucleotidases CD39 and CD73 or the A2AR/BR are in clinical trials, including our non-CNS penetrant A2AR-selective antagonist, EXS21546 (NCT04727138). While early clinical results from other adenosine receptor-inhibitors (A2ARi) have shown modest monotherapy activity in nonspecific patient populations, we believe greater success can be achieved by leveraging methods that enable the evaluation of single-cell effects in patient samples preclinically. Using our deep learning driven image analysis platform, we define an adenosine-induced, tumor protective immunosuppression biomarker to augment A2AR antagonist responder identification. Here we describe efforts to transcriptionally and functionally map the adenosine suppressed immune potential and activation by inhibition of A2AR with EXS21546 in primary material. The goal is to reveal gene signatures indicative of adenosine immunosuppression deployable in clinical studies, increasing the likelihood of trial success by identifying patients that have the highest efficiency potential for A2AR targeted therapy. Methods: Leveraging patient material as disease relevant model systems, and collecting baseline and treatment condition transcriptomics, we model the patient specific anti-cancer immune repertoire and validate patient selection methods functionally with a translatable high content imaging platform amenable to primary human material supported by end-to-end deep learning driven image analysis. Results: Combining single cell transcriptomic and functional response data, we demonstrate preclinical mechanistic studies of A2AR antagonism on infiltrating immune cells, with the ultimate aim of discovering predictive algorithms to enrich patients more likely respond to adenosine pathway inhibitors. Patient selection gene signatures are functionally validated using a high content imaging platform with proven translational capabilities (Kornauth et al, Cancer Disc., 2021), demonstrating the association of anti-cancer immune activity with inhibition of adenosine signaling by EXS21546. Signatures and patient selection algorithms are cross-validated with publicly available data. Discussion: Gathering multiple layers of data from primary tumor tissues, we reveal and map the association of immune response potential to A2AR inhibition in cancer. Patient stratification gene signatures identified have the aim to be implemented in future studies of our candidate A2ARi EXS21546, to deliver the right drug at the right time in the right patients. Citation Format: Isabella Alt, Robert Shelke, Anna Lobley, Claudia Baumgaertler, Maja Stulic, Pierre Fons, Mark Whittaker, Klaus Hackner, Lucia Dzurillova, Edga Petru, Laudia Hadjari, Judith Lafleur, Josef Singer, Nikolaus Krall, Lukas Hefler, Thorsten Füreder, Christina Taubert, Christophe Boudesco, Andrew Payne, Gregory Ian Vladimer. Enriching for adenosine antagonist patient responses through deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 4150.