EXTH-66. Logic-Guided Antigen Pairing and TEV-Responsive CAR Designs for Precision Immunotherapy in Glioblastoma

Résumé fourni par la source

Abstract Chimeric antigen receptor (CAR) T cell therapies for glioblastoma (GBM) have shown limited clinical efficacy, hindered by extensive intratumoral heterogeneity, dynamic antigen expression, and the immunosuppressive landscape of the central nervous system. Current CAR designs often rely on individual scFvs developed against historically available targets, rather than arising from a rational analysis of tumor-specific antigen landscapes. As a result, many clinical constructs fail to address incomplete tumor coverage or pose risks of off-tumor toxicity in sensitive tissues such as the brain. To overcome these limitations, we established a data-driven framework for antigen selection and synthetic receptor design that integrates single-cell RNA sequencing (scRNA-seq), machine learning, and programmable CAR logic architectures. We constructed a high-resolution atlas incorporating malignant GBM, normal adult tissue (Tabula Sapiens), fetal brain tissue, and brain-resident immune and stromal populations to profile surface antigen expression across relevant compartments. Using Boolean logic gates (AND, OR, NOT), we computationally modeled antigen pairings that maximize tumor cell recognition while minimizing overlap with normal tissue expression. We applied machine learning to score over 200 candidate surface proteins based on expression intensity, co-expression frequency, and tissue specificity, enabling prioritization of optimal antigen combinations. While commonly used GBM targets such as EGFR and IL13RA2 failed to achieve broad tumor coverage or sufficient exclusivity, alternative combinations featuring underutilized candidates like PTPRZ1, SLC1A3, NRCAM, CD44, and LIFR exhibited improved tumor selectivity. Notably, NOT-gated designs consistently outperformed AND and OR architectures in minimizing predicted off-tumor activation. To translate these insights into functional circuits, we engineered TEV protease-responsive NOT CARs, in which detection of a “safe” antigen induces proteolytic inactivation of the CAR’s signaling domain at various response thresholds. Collectively, this work presents a systematic approach to intelligent CAR design in GBM, guided by high-dimensional data and refined by logic-gated synthetic biology. Our platform enables precision targeting through antigen pair selection and tunable receptor control, offering a pathway to safer and more effective T cell therapies for GBM.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.