A rapid patient-derived organoids platform guides the evaluation of ERBB-targeting bispecific T-cell engagers

Background: The ERBB receptor family (EGFR/HER1, HER2, HER3, HER4) are well-validated oncogenic drivers. The clinical translation of ERBB-targeted bispecific antibodies (e.g., EGFRxCD3, HER2xCD3,HER2 degrading BsAbs) is bottlenecked by the lack of models that can rapidly and concurrently predict patient-specific efficacy and on-target, off-tumor toxicity. Optimizing these drugs including fine-tuning parameters like affinity, valency, and epitope to maximize the therapeutic window, which requires a predictive platform that provides integrated feedback on both anti-tumor activity and on-target toxicity in a clinically relevant timeframe. Methods: We established a comprehensive biobank of patient-derived organoids (PDOs) from breast, gastric, and non-small cell lung cancers, alongside matched normal organoids. We developed a high-throughput, standardized co-culture assay with PBMCs or T cells to evaluate a panel of ERBB-targeting bispecific antibodies with varying molecular formats. The entire workflow was designed for speed, generating parallel data on tumor organoid killing and normal organoid toxicity within days. Results: Our platform delivered robust, quantitative data on the therapeutic index for a series of BsAb candidates within a 3-week window. We demonstrated that affinity-tuning towards tumor-associated antigen levels could preferentially spare normal organoids with low ERBB expression while maintaining potent tumor killing. Furthermore, comparing different BsAbs revealed that certain molecular formats or epitope choices were associated with reduced cytokine release and less severe toxicity in normal organoids, without compromising efficacy in target-positive tumors. This provides a direct, rapid strategy for lead candidate selection and optimization. Conclusion: We have developed a rapid, reproducible PDO-based platform that not only predicts the therapeutic window of ERBB-targeting BsAbs but also provides critical insights for their molecular optimization. This "fast-feedback" system can significantly de-risk and accelerate the translation of BsAbs by enabling data-driven decisions on affinity, format, and epitope selection early in the drug development process. The ability to rapidly profile multiple candidates against a backdrop of human tumor and normal tissues makes this an invaluable tool for designing safer and more effective bispecific antibodies. This study presents a transformative preclinical optimization tool. By generating predictive safety and efficacy data in weeks, our platform moves beyond mere prediction to active guidance, empowering the rational design of next-generation BsAbs with an inherently improved therapeutic profile for clinical trials.