Model-based prediction of monoclonal antibody retention in ion-exchange chromatography

• The retention factors of mAbs in ion-exchange chromatography were predicted.
• A colloidal model accounting for pH and salt dependence was applied.
• Structural information about protein and stationary phase was used in the model.
• Retention is calculated with the protein net charge using the amino acid sequence.
• The predictions are useful to reduce the experimental screening effort.

In order to support a model-based process design in ion-exchange chromatography, an adsorption equilibrium model was adapted to predict the protein retention behavior from the amino acid sequence and from structural information on the resin. It is based on the computation of protein–resin interactions with a colloidal model and accounts for the contribution of each ionizable amino acid to the protein charge. As a verification of the protein charge model, the experimental titration curve of a monoclonal antibody was compared to its predicted net charge. Using this protein charge model in the computation of the protein–resin interactions, it is possible to predict the adsorption equilibrium constant (i.e. retention factor or Henry constant) with an explicit pH and salt dependence. The application of the model-based predictions for an in silico screening of the protein retention on various stationary phases or, alternatively, for the comparison of various monoclonal antibodies on a given cation-exchanger was demonstrated. Furthermore, considering the structural differences between charge variants of a monoclonal antibody, it was possible to predict their individual retention times. The selectivity between the side variants and the main isoform of the monoclonal antibody were computed. The comparison with the experimental data showed that the model was reliable with respect to the identification of the operating conditions maximizing the selectivity, i.e. the most promising conditions for a monoclonal antibody variant separation. Such predictions can be useful in reducing the experimental effort to identify the parameter space.