(576c) Beyond Structure Determination: Crystallisation As a Purification Unit Process for Proteins and Peptides

The purification of high molecular weight proteins and peptides, for their use in biopharmaceutical drug therapeutics, can still account for a significant proportion of the drug development cost, which averages at between $1-6 billion. Crystallisation of these materials has mainly been used only in the field of structural determination, using small scale methods such as hanging drop vapour diffusion. However, advances in the optimisation of crystallisation conditions has seen significant increase in the scalability of protein crystallisation, including the development of continuous crystallisation approaches. This has seen crystallisation emerge as a more economic and greener alternative to the existing chromatographic methods typically utilised in peptide and protein purification.

The Made Smarter review of 2017 identified that manufacturing industries should embrace transformative digital technologies to streamline their unit processes. Mechanistic process modelling, based on population balance equations to represent the underpinning physical phenomena, has proved a computationally efficient method of predicting optimal crystallisation conditions of the purification of small molecule active pharmaceutical ingredients. However, this has yet to be expanded into the crystallisation of biopharmaceutical proteins.

Here, we utilise a combined process modelling and experimental approach to scale up the crystallisation of lysozyme and insulin from the µl to the tens of ml scale. We successfully optimise conditions, such as stirring rate and operation, pH, concentration and templating additives to not only scale up, but also increase the achieved crystallisation yield.

Insulin, being more susceptible to denaturing, utilised shaking to induce mixing of the protein rich buffer and salt precipitant. Small scale hanging drop screening experiments found that the addition of amino acids could reproducibly encourage the crystallisation of insulin, offering the promise of control over the nucleation. Scale up experiments to 15ml allowed the investigation of the impact of shaker mixing from 50-150 rpm. The insulin case study also highlights how to circumvent common issues with protein crystallisation, such as working with extremely low solubility materials and identifying the nucleation and growth points during the crystallisation process.

The lysozyme protein is more resistant to denature than insulin, whereby it can be subjected to mechanical stirring. Here we show successful batch and continuous crystallisations, utilising both shaking and mechanical stirring, up to the 30ml scale. Induction times were measured and classical nucleation kinetics parameters were derived from this data, with the pre-exponential factor and the surface energies being consistent with previous studies. Further optimisation of stirring methods and initial protein concentration was investigated to optimise the size and shape of the produced lysozyme crystals, whereby their purity was compared to the as purchased lipolyzed powder.

Finally, process modelling techniques were applied to predict the crystallisation of lysozyme and insulin in the batch reactor conditions. Such models are well parameterised to describe the crystallisation of small molecule active pharmaceutical materials. However, the classical nucleation and growth equations, as well as the models to describe the liquid state, have been less commonly applied to biological materials. Kinetic parameters to describe the processes that were extracted from the literature were used as a starting point for the model, where careful optimisation of these parameters were used to identify problems and where discrepancies between modelled and experimental particle size distributions and concentration vs time data originated. The kinetic parameters that were optimised against reliable experimental data were confronted with further experimental data to examine the wider applicability of protein crystallisation process model.

It was found that the classical Mullin nucleation and power law growth kinetic equations could be successfully regressed against experimental concentration vs time data, even for long nucleation times (Figure 1).

Figure 1: Concentration vs time for a 30ml batch crystallisation of lysozyme, using a sodium acetate buffer and sodium chloride precipitant. Blue dots are experimental data with approximate 10% error from nanodrop UV vis measurements and the red line in the regressed fit using Mullin classical nucleation kinetics and power law growth kinetics

Figure 1 shows that a good fit of the process model to the experimental concentrations vs time data was achieved. It is interesting to observe that the classical two step growth kinetics failed to find a fit, where we speculate that the diffusion term associated with the large lysozyme molecule limits the growth of this material in the model. Further investigation of the exact parameters in the equations will be presented.

This study demonstrates a novel experimental and process modelling workflow for optimising the crystallisation of large biological molecules, paving a path for this to be a viable unit process in biopharmaceutical drug development. This study gives guidance and insights into the challenges of protein crystallisation and modelling of this process, along with what are the next steps to make both processes more robust for further scale-up.