Modeling of hydrophobic interaction chromatography for the separation of antibody-drug conjugates and its application towards quality by design

Antibody-drug conjugates (ADCs) are hybrid molecules based on monoclonal antibodies (mAbs) with covalently attached cytotoxic small-molecule drugs. Due to their potential for targeted cancer therapy, they form part of the diversifying pipeline of various biopharmaceutical companies, in addition to currently seven commercial ADCs. With other new modalities, ADCs contribute to the increasing complexity of biopharmaceutical development in times of growing costs and competition. Another challenge is the implementation of quality by design (QbD), which receives a lot of attention. In order to answer these challenges, mechanistic models are gaining interest as tools for enhanced process understanding and efficient process development. The drug-to-antibody ratio (DAR) is a critical quality attribute (CQA) of ADCs. After the conjugation reaction, the DAR can still be adjusted by including a hydrophobic interaction chromatography (HIC) step. In this work, we developed a mechanistic model for the preparative separation of cysteine-engineered mAbs with different degrees of conjugation with a non-toxic surrogate drug. The model was successfully validated for varying load compositions with linear and optimized step gradient runs, applying conditions differing from the calibration runs. In two in silico studies, we then present scenarios for how the model can be applied profitably to ensure a more robust achievement of the target DAR and for the efficient characterization of the design space. For this, we also used the model in a linkage study with a kinetic reaction model developed by us previously. The combination of the two models effectively widens system boundaries over two adjacent process steps. We believe this work has great potential to help advance the incorporation of digital tools based on mechanistic models in ADC process development by illustrating their capabilities for efficient process development and increased robustness. Mechanistic models can support the implementation of QbD and eventually might be the basis for digital process twins able to represent multiple unit operations.