Modelling the performance of membrane nanofiltration—application to an industrially relevant separation

Abstract

For traditional separation processes there are widely available process design methodologies for scale-up and optimisation. However, there is an increasing need for such a rational approach to membrane separation processes, identifying at an early stage operating limits and process options. Such predictive models will reduce development risk and time, thus promoting the wider use of membrane technology in process industries such as pharmaceutical manufacture. Design methods exist that have been verified experimentally at the laboratory scale for simple aqueous solutions. There is now a need for the application of the existing theoretical and experimental methods to separations of real industrial interest. In this paper we demonstrate this philosophy by describing the rationale for modelling the performance of membrane nanofiltration (NF) used in the isolation of N-acetyl- d-neuraminic acid (Neu5Ac) an industrially important example of an equilibrium-controlled biotransformation reaction. The chemoenzymatic synthesis of Neu5Ac is already achieved at large scale and its derivatives are important as precursors for a range of antiviral drugs such as the commercially available anti-influenza agent 4-guanidino-Neu5Ac-2-en (zanamivir, Relenza™). The separation involves the removal of pyruvate from the process stream, which is complicated by the fact that Neu5Ac and pyruvate have similar p K a values. A systematic study of the rejection of the components involved in the separation was carried out, firstly as individual components and then as mixtures. Excellent agreement between the calculated and experimental rejections of the solutes and ions was observed for batch nanofiltration. A membrane charge isotherm was developed from the experimental data for use in the theoretical process modelling. A comparison was made between a full model and a linearised model based upon the extended Nernst–Planck equation and the difference between the two was considered small. Subsequently, the linearised model was used for prediction of a diafiltration process to remove pyruvate and excellent agreement was achieved when allowing for the variation in effective membrane charge density as the pyruvate concentration in the solution changed as a function of time.