Rapid and Efficient Antibody-Drug Conjugate Design Using Mechanistic Bottom-Up Modeling from In Vitro to Human.

Antibody-drug conjugates (ADCs) are complex molecules, and many fail clinically despite promising preclinical data. Here, a modular, bottom-up modeling strategy employing mechanistic PK-PD models was developed to translate ADC efficacy and toxicity from bench to bedside and to guide ADC design. The models handle various antigens, payloads, and ADCs, validated using six ADCs (Enhertu, Kadcyla, Trodelvy, RC48, RN927C, and Datroway). Using determined cellular distribution of payloads/ADCs, payload killing parameters, and systemic parameters (assay volume, cell doubling time), ADC in vitro potency was predicted with 94.64% accuracy within a 2-fold range. Incorporating ADC and payload pharmacokinetic parameters enables prediction of in vivo efficacy comparable to cell line- and patient-derived xenograft data from the literature. Scaling parameters from mouse to human (PK, tumor volume, and tumor doubling time) led to the reproduction of clinical efficacy trends. Beyond efficacy, the approach predicts key hematologic toxicities such as neutropenia and thrombocytopenia, demonstrated for Kadcyla and Enhertu. Application to the clinically failed RN927C demonstrated how our modeling approach could have flagged issues and enabled suggestions for design modifications to widen the therapeutic window and prevent the clinical failure. In conclusion, our presented modeling strategy delivers accurate efficacy and toxicity translation from in vitro to humans utilizing easily accessible parameters as the foundation and deepens understanding of ADCs and their individual components, thereby supporting ADC design and candidate and patient selection and accelerating ADC development.