Methodologies for model calibration to assist the design of a preparative ion-exchange step for antibody purification

Abstract

This work proposes methodologies using a model-based approach to gain knowledge on and assist the development of an ion-exchange step in a protein purification process; the separation of IgG from a mixture containing IgG, insulin and transferrin. This approach is suitable for capture and intermediate steps in a process. Both methods involve four consecutive steps. Firstly, the retention of the different protein components is determined giving a retention map of the system. From this the optimal pH and buffer can be determined. Secondly, additional salt gradient experiments are performed at the selected pH. Thirdly, experimental breakthrough curves have to be generated for the protein if the adsorption capacity of the medium for each component is not known. Fourthly, a validation experiment is performed. In method 1, where the capacity for the medium is assumed to be known, the protein adsorption is described by Langmuir kinetics with a mobile phase modulator (MPM). In this description salt is considered to be inert. In method 2 the adsorption behavior is described by steric mass action (SMA), where the salt component competes with the proteins for the available binding sites. Both methods use a dispersion model to describe transport in the mobile phase in the column. The methods are able to predict the separation and loading behavior of the three components. The methods can, with reasonable accuracy, predict the breakthrough of transferrin in a mixture of insulin, IgG and transferrin. Method 1 requires fewer experiments and predicts the mean volume of breakthrough for the loading step in the validation experiment more accurately than method 2. On the other hand, method 2 has a better accuracy to predict the position of 10% breakthrough and the shape of the breakthrough curve. The methods suggested in this work are shown to be efficient in process development. Some additional experiments have to be performed to obtain the unknown parameters in the models. However, the predictability that is achieved results in less experimental work in the process design as a whole.