Abstract 7272: aPD1 response prediction in a human <i>ex vivo</i> tumor model: A case study in multiomic approaches for preclinical drug evaluation

Abstract Background: Anti-PD1 (aPD1) and other immunotherapies have demonstrated significant clinical impact in cancer treatment. However, the inability to accurately predict which patients will respond to specific therapies and to determine the correct indications or combinations for many immunotherapy drugs has led to clinical trial failures across many tumor indications. To address this, we developed a preclinical ex vivo patient derived-tumor fragment (PDTF) platform that leverages single-cell multiomics data and our expansive patient-derived single-cell database AMICATM to dissect drug mechanism of action (MOA), and identify baseline determinants of molecular ex vivo response, which can be used to predict therapeutic response in clinical samples. Using pembrolizumab as a use case, this approach identifies patient-level baseline predictive features of aPD1 therapeutic response and predicts the relative clinical efficacy of aPD1 across patient subgroups. Methods: PDTFs from a cohort of renal cell carcinoma (RCC) patients were treated with 10 ug/ml aPD1 or isotype control for 48 hr. Multiplex ELISA was performed on culture supernatants, while pre-culture and treated PDTFs were dissociated and sorted on live immune cells for CITE-seq analysis. The resulting multiomic dataset includes cell abundances, cell type specific RNAs and proteins, clonality of T cell subsets, and secreted proteins. These features were used to stratify patients into molecular responders and non-responders. The pre-culture samples from the same patients were then analyzed using the ImmunoDynamics EngineTM, a proprietary machine learning (ML) framework, to determine the baseline features of molecular response. To validate our approach, these baseline features were contextualized using samples from patients with known clinical response to PD1 blockade. Results: aPD1 elicits cytotoxicity and interferon pathway upregulation in PD1-expressing cell types. The MOA of aPD1 in our ex vivo system is consistent with several clinical datasets. Individual donors could be classified by molecular response to aPD1 across several multiomic features. Immune features correlating with response to aPD1 at baseline were identified and found to be predictable of molecular response with reasonable accuracy. Molecular response features were validated on baseline tumor samples from patients treated with PD1 inhibitors in the clinic and were found to be predictive of therapeutic prognosis. Conclusions: Our ex vivo model system and analytical framework uses several modalities to establish the MOA and clinical response features of an immunotherapy agent. Recapitulating the known biology and response of aPD1 builds conviction that extending this framework to other immunomodulatory drugs and their combinations could enhance patient selection and clinical trial design, ultimately improving clinical outcomes. Citation Format: Adam Yalin, Andras Sziraki, Devin Mediratta, Andrew Shuparski, Andy Tu, Mirko Andreoli, Patrick Petrossian, Konstantinos Vlachos, Tomas Zikmund, Dagmar Schierova, Alena Molnarova, Anna Dostalova, Eleni Mimitou, Adeeb Rahman, Lisa Freinkman, Florian Uhlitz, Evgeny Kiner. aPD1 response prediction in a human ex vivo tumor model: A case study in multiomic approaches for preclinical drug evaluation [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7272.