Model based process design for bioprocess optimisation: case studies on precipitation with its applications in antibody purification

Developing a bioprocess model can not only reduce cost and time in process \ndevelopment, but now also assist the routine manufacturing and guarantee the \nquality of the final products through Quality by Design (QbD) and Process \nAnalytical Technology (PAT). However, these activities require a model based \nprocess design to efficiently direct, identify and execute optimal experiments for the \nbest bioprocess understanding and optimisation. Thus an integrated model based \nprocess design methodology is desirable to significantly accelerate bioprocess \ndevelopment. This will help meet current urgent clinical demands and also lower the \ncost and time required. This thesis examines the feasibility of a model based process \ndesign for bioprocess optimisation. A new process design approach has been \nproposed to achieve such optimal design solutions quickly, and provide an accurate \nprocess model to speed up process understanding. \nThe model based process design approach includes bioprocess modelling, model \nbased experimental design and high throughput microwell experimentation. The \nbioprocess design is based on experimental data and a computational framework \nwith optimisation algorithm. Innovative model based experimental design is a core \npart in this approach. Directed by the design objectives, the method uses D-optimal \ndesign to identify the most information rich experiments. It also employs Random \ndesign and Simplex to identify extra experiments to increase the accuracy, and will \niteratively improve the process design solutions. \nThe modelling and implementation method by high throughput experimentation was \nfirst achieved and applied to an antibody fragment (Fab’) precipitation case study. A \nnew precipitation model based on phase equilibrium has been developed using the \ndata from microwell experimentation, which was further validated by statistical tests \nto provide high confidence. The precipitation model based on good data accurately describes not only the Fab' solubility but also the solubility of impurities treated as a \npseudo-single protein, whilst changing two critical process conditions: salt \nconcentration and pH. The comparison study has shown the model was superior to \nother published models. The new precipitation model and the Fab' microwell data \nprovided the basis to test the efficiency and robustness of the algorithms in model \nbased process design approach. The optimal design solution with the maximum \nobjective value was found by only 5 iterations (24 designed experimental points). \nTwo parameterised models were obtained in the end of the optimisation, which gave \na quantitative understanding of the processes involved. The benefit of this approach \nwas well demonstrated by comparing it with the traditional design of experiments \n(DoE). \nThe whole model based process design methodology was then applied to the second \ncase study: a monoclonal antibody (mAb) precipitation process. The precipitation \nmodel was modified according to experimental results following modelling \nprocedures. The optimal precipitation conditions were successfully found through \nonly 4 iterations, which led to an alternative process design to protein A \nchromatography in the general mAb purification platform. The optimal precipitation \nconditions were then investigated at lab scale by incorporating a depth filtration \nprocess. The final precipitation based separation process achieved 93.6% (w/w) mAb \nyield and 98.2 % (w/w) purity, which was comparable to protein A chromatography. \nPolishing steps after precipitation were investigated in microwell chromatographic \nexperimentation to rapidly select the following chromatography steps and facilitate \nthe whole mAb purification process design. The data generated were also used to \nevaluate the process cost through process simulations. Both precipitation based and \nprotein A chromatography based processes were analysed by the process model in \nthe commercial software BioSolve under several relevant titre and scale assumptions. \nThe results showed the designed precipitation based processes was superior in terms \nof process time and cost when facing future process challenges.