Comprehensive preclinical evaluation of CAR-T cell therapeutics in organoid-fibroblast co-cultures

Introduction: Preclinical assessment of CAR-T cell therapies requires models that recapitulate the complex tumor microenvironment (TME), but conventional 2D cultures fail to model stromal barriers. Patient-derived (xenograft) organoids (PDOs/PDXOs) offer superior clinical relevance, but analyzing 3D immune-tumor-stromal interactions is challenging with bulk assays like LDH release. High-Content Imaging (HCI) overcomes these limitations, enabling high-throughput, multi-parametric quantification of immune infiltration and cytotoxicity. We utilized an HCI-based organoid co-culture platform to evaluate CAR-T therapies in both standard and complex, fibroblast-containing TME models. Methods: Organoids were co-cultured with PBMCs or engineered CAR-T cells in a 384-well format suitable for automation. HCI analysis allowed quantification of different cell populations and several key readouts, including organoid volume (via actin and DAPI staining), fibroblast network branching (via Cy5 staining) at 96h, and T-cell infiltration at 48h. For a head-to-head lead selection study, seven distinct Claudin18.2-targeting CAR-T candidates were screened against three pancreatic organoid models to comparatively assess their cytotoxic potential. To model a complex TME, ovarian model OV9522B was co-cultured with Normal Human Lung Fibroblasts (NHLF). Triple co-cultures were treated with Mesothelin (MSLN)-targeting CAR-T cells at various Effector-to-Target (E:T) ratios (20:1, 10:1, 5:1). Results: Baseline Graft-versus-Tumor (GvT) effects were established by co-culturing PBMCs from four healthy donors with four organoid models (LU9906B, CR5082B, CR20155B, PA20077B) revealing variable donor- and model-dependent cytotoxicity (5%-20%). The platform enabled comparison and ranking of seven Claudin18.2-targeting CAR-T candidates against three models. This screen revealed a wide range of killing potency: the most potent CAR-T induced >80% volume reduction in the most sensitive pancreatic model, while the weakest CAR-T showed ~10% reduction in a poor responsive model. Finally, in the triple co-culture model, MSLN-targeting CAR-T cells induced potent, dose-dependent killing of OV9522B organoids (high MSLN expression). Despite a stromal network, a significant reduction (>90%) in organoid volume and visualized CAR-T engagement were observed, demonstrating fibroblasts did not affect cytotoxic activity in this case. Conclusion: Integration of HCI with organoid co-cultures provides a comprehensive, scalable, and clinically relevant platform for evaluating cell therapies. Visualizing and quantifying key parameters like targeted killing in presence of stromal components is a critical advantage and enables confident ranking of therapeutic candidates and provides mechanistic insights, de-risking and accelerating clinical development for next-generation cell therapies.