Modeling and Optimization of Reversed-Phase Chromatography : Effects of Modulators and Temperature

University dissertation from Department of Chemical Engineering, Lund University

Abstract: Many widespread diseases, such as diabetes, various types of cancer, and aggressive versions of influenza, are treated or prevented with biopharmaceuticals. Biopharmaceuticals are drugs that are based on proteins, peptides, antibodies, attenuated viruses (vaccines), and other biomolecules that are synthesized predominately in bacteria, yeast, and mammalian cells. The first biopharmaceutical was introduced to the market in the early 1980s, and in the past several years, approximately 10 new compounds have reached the market annually. If this trend continues, rapid development of production processes for these new biopharmaceuticals will be required.One important method by which biomolecules are purified is preparative chromatography. Although it is a well-established approach, the phenomena on which it is based are still incompletely understood. Knowledge about the effects of the process setup and operating conditions is crucial to design new chromatographic processes efficiently and streamline existing processes.In the work presented in this thesis, the influence of the adsorbent and process conditions on the chromatographic separation of three insulin variants was examined. Two adsorbents each for reversed-phase chromatography (RPC) and hydrophobic interaction chromatography (HIC) were tested, and the effects of temperature and the concentrations of the two modulators, KCl and ethanol, were examined.The retention of the insulin variants on the RPC adsorbents decreased as the temperature and concentrations of the modulators rose. On the HIC adsorbents, the retention declined with higher ethanol concentrations and increased with higher KCl concentrations. Consequently, KCl caused salting-in at the high ethanol concentrations that were required for elution from the RPC adsorbents and induced salting-out at the low ethanol concentrations that were needed to achieve retention on the HIC adsorbents. These data are consistent with predictions by other groups. Due to the severe self-association of insulin molecules in the HIC experiments, these two process setups were not examined further.In a comparison between the solubility data for insulin and its chromatographic retention, the influence of ethanol on the latter was significantly stronger and thus was attributed not only to its effect on the mobile phase — the most likely explanation is that ethanol molecules adsorbed onto the ligands and were displaced by adsorbing insulin molecules. A semi-empirical RPC model that was based on thermodynamic theories was derived from the adsorption equilibrium. This model assumed adsorption of ethanol and included the activity coefficients of all involved species.The effect of temperature on the equilibrium constant can be satisfactorily described by a linear variation of the change in Gibbs free energy on adsorption — i.e., assuming that the changes in enthalpy and entropy are temperature-independent. Because the estimated values of the enthalpy and entropy are negative, the adsorption must be enthalpy-driven. Apart from the effect of temperature on the equilibrium constant, the activity coefficients of the ethanol and insulin variants varied significantly with temperature. These effects should be separated if the temperature and modulator concentrations are varied and if several combinations of adsorbates, adsorbents, and modulators are compared.A satisfactory model fit was achieved for variations in the concentrations of KCl and ethanol with regard to calculation of the linear-range retention and the dynamic simulations at high load. The effect of changes in temperature is less well described, albeit sufficiently, by the model. Considering that the values of the model parameters that are related to the influence of the modulators were not adjusted to the data from the temperature study, the fit is impressive.The applicability of the final model was demonstrated in a model-based multi-objective optimization study. Pareto fronts, showing the optimal combinations of yield and productivity, were generated for both RPC adsorbents. Due to the higher selectivity between the insulin variants on the C18 versus C4 adsorbent, the former effected greater productivities at a higher yield. The effect of a constraint on the Pareto fronts, with regard to the solubility of the insulin variants, was examined by comparing Pareto fronts that were based on constrained versus unconstrained optimizations. The Pareto fronts diverged when the constraint became active, and the productivity was nearly constant, with decreasing yield for the constrained optimizations, whereas that for the unconstrained optimizations continued to rise steadily.Due to the halt in increased productivity, an alternative to performing constrained optimizations could be to select the operating point from an unconstrained optimization that lies just below the solubility limit, which yielded approximately the same result in this case study.