Abstract A033: Identifying logic-gated combinatorial antigens from single-cell transcriptomics data of patient tumors for effective and safe CAR therapy design

Abstract Introduction: Chimeric antigen receptor (CAR) T-cell therapy has transformed the treatment of hematological malignancies. However, its application to solid tumors remains limited by antigen heterogeneity and on-target, off-tumor toxicities. One critical emerging approach to combat this challenge is logic-gated multi-antigen targeting, for both enhanced safety and efficacy. Methods: To address these challenges, we developed a novel genetic algorithm, termed LogiCAR, and applied it to analyze more than a dozen different breast cancer (BC) patient cohorts of single-cell (scRNA-seq) and single-nucleus (snRNA-seq) transcriptomics data from patient tumors, and the Tabula Sapiens healthy reference tissue atlas. This approach identifies logical combinations of surface antigens that are selectively expressed in cancer cells vs. noncancerous cells by using the logical operators "AND," "OR," and "NOT." The efficacy of a given logic-gated antigen combination is defined as the proportion of tumor cells it targets, while its safety is defined as the proportion of non-tumor cells estimated to be spared by that combination in the tumor microenvironment (TME) and twenty-four Tabula Sapiens atlas healthy tissues. Results: As a benchmark, we first evaluated the single-cell-derived safety and efficacy scores of single-antigen clinical CAR targets. Overall, single-antigen clinical targets have strong safety profiles when evaluated on the Tabula Sapiens cohort. However, when tested on our sc/snRNA-seq BC cohorts, they have low efficacy scores. As benchmarks against which to compare our newly identified combination LogiCARs, we identified EPCAM as the overall most efficacious target (mean patient efficacy = 0.43), and EGFR as the lowest overall safety target (mean patient safety = 0.76). When tested on more than a dozen breast cancer cohorts, including a new multi-ethnicity snRNA-seq cohort of 82 breast cancer patients that we have generated at the NCI, the LogiCAR algorithm identified numerous triplet antigen combinations that outperform the clinically approved/tested single-antigen targets in efficacy measures, while maintaining high safety, displaying the potential to overcome both intra- and intertumor heterogeneity. The top LogiCAR combinations identified across all patients exhibited strong performance across cohorts, and strikingly, across all breast cancer subtypes (HR+, HER2+, and TNBC). Conclusion: LogiCAR successfully identifies highly efficacious and safe combinations in breast cancer, promising to facilitate the clinical development of logic-gated CAR circuits. Citation Format: Sanna Madan, Tiangen Chang, Alexandra Harris, Huaitian Liu, Saugato Rahman Dhruba, Sanju Sinha, Sheila Rajagopal, Aravind Srinivasan, Stefan Ambs, Chi-Ping Day, Alejandro A. Schäffer, Eytan Ruppin. Identifying logic-gated combinatorial antigens from single-cell transcriptomics data of patient tumors for effective and safe CAR therapy design [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Functional and Genomic Precision Medicine in Cancer: Different Perspectives, Common Goals; 2025 Mar 11-13; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(5 Suppl):Abstract nr A033.