(313h) Scale Up of Membrane Affinity Chromatography Processes

Membrane chromatography represents one of the emerging technologies for downstream processing in the biotechnology industry. This process is currently used in polishing steps for antibody manufacturing, while its application is still uncommon for the capturing step. To evaluate and to promote its application in large scale processes, it is crucial to use a reliable simulation tool able to describe the performance of membrane adsorbers at all scales in a predictive way.

In this work, the physical model for the description of protein purification with affinity membrane chromatography has been used to predict the performance of scaled up systems and compared with the simplified lumped model, in which axial dispersion is not accounted for explicitly but is rather lumped into the binding kinetic term. Two commonly used binding kinetics have been implemented in the models, namely the Langmuir and the bi-Langmuir equations.

It is known that by considering experimental data at bench scale only, both lumped model and physical model can describe equally well the observed behaviors. However the hypothesis to lumped all transport and kinetic processes into a single term leads to the use of “binding kinetic” parameters which are strongly flow rate dependent, in contrast with the usual physical properties of such terms. That throws motivated doubts to the validity of the lumped model for scale up purposes, which are further supported by the fact that the two models have different dimensionless equations.

In the present work we inspect the different scale-up predictions obtained by using the physical and lumped model, whose parameters have been assessed for the proper description of bench scale experimental data of h-IgG purification using B14-TRZ-Epoxy2 affinity membranes at different concentrations and different flow rates. The comparison of the lumped model with a validated physical model represents a valuable tool to assess the suitability of the lumped approximation. Although the two models describe equally well the experimental concentration profiles measured in a bench scale module, their up-scaled simulations at the size of commercially available adsorbers reveal a large discrepancy even at the early stages of breakthrough. The lumped model always overestimates the breakthrough point with respect to the physical model and, at 10% breakthrough, the difference in dynamic binding capacity evaluated by the two models presented is remarkably in the order of 23-28%, definitely impressive for the high added value of the antibody to be produced.

The physical model is based only on the actual physical phenomena present in the chromatographic module, the dispersion coefficient is measured separately and independently from the chromatographic cycle, and the conclusion is that it is the appropriate model for scale-up purposes.

Acknowledgements

This work was financially supported by MIUR, Italian Ministry of Education, University and Research (PRIN 2008) and by the University of Bologna, Italy.