This article describes the rapid prediction of recovery process performance for a new recombinant enzyme product on the basis of a broad portfolio of computer models and highly targeted experimentation. A process model for the recombinant system was generated by linking unit operation models in an integrated fashion, with required parameter estimation and physical property determination accomplished using data from scale-down studies. This enabled the generic modeling framework established for processing of a natural enzyme from bakers' yeast to be applied. An experimental study of the same operations at the pilot scale showed that the process model gave a conservative prediction of recombinant enzyme recovery. The model successfully captured interactions leading to a low overall product yield and indicated the need for further study of precipitate breakage in the feed zone of a disc stack centrifuge in order to improve performance. The utility of scale-down units as an aid to fast model generation and the advantage of integrating computer modeling and scale-down studies to accelerate bioprocess development are highlighted.

Models for high pressure homogenisation developed using natural bakers' yeast were applied to a recombinant yeast strain. The models were found to be generic in nature and generally adequate for initial process design (30%). More accurate descriptions (10%) of the release of protein engineered enzyme product, the release of total soluble protein and the change in debris size distribution required a small amount of extra information which was readily acquired using a scale-down unit. Utilisation of such scale-down and modelling techniques will enable design calculations, unit integration and identi®-cation of optimum operating conditions to be carried out for the high pressure homogenisation of recombinant yeast early in process development with a high degree of con®dence.List of symbols b constant in Eq. (5) C u soluble protein assay result (mg/ml) C y yeast concentration on a wet weight basis (g/l) d particle diameter (lm) d 50 Boltzman average diameter (lm) d 50N0 whole cell Boltzman average diameter (lm) d à 50 dimensionless form of d 50 de®ned as d 50N0 À d 50 ad 50N0 À D fraction of cells disrupted À F aqueous fraction À f D S stress distribution imposed on cells by homogeniser À f S S cell strength distribution À f d NYP cumulative (undersize) volume PSD for a given size range at given P and N i number of data points À k constant in Eq. (7) K constant in Eq. (1) m constant in Eq. (5) n constant in Eq. (5) N number of passes À Q¯owrate (m/s 3 ) P pressure (barg) P t threshold pressure (barg) r regression coef®cient À R soluble protein release (mg protein/ g wet weight yeast) R max maximum soluble protein release (mg protein/g wet weight yeast) S effective strength À S m mean effective strength À Greek symbols a exponent in Eq. (1) r standard deviation À x Boltzman parameter in Eq. (6) x N0 whole cell Boltzman parameter in Eq. (6) x à dimensionless form of x de®ned as (x N0 À xx N0 À v chi statistic R centrifuge sigma factor (m 2 ) IntroductionThe use of the yeast Saccharomyces cerevisiae as a host organism for the production of recombinant proteins often results in an intracellular product [9]. Ef®cient cell disruption is therefore required to achieve product release and enable subsequent recovery. High pressure homogenisation is usually the preferred unit operation for achieving this disruption at a large scale [20]. In determining process feasibility and for subsequent engineering design it is essential that the homogenisation stage can be described by relevant mathematical models. With increasing pressure on biotechnology companies to bring bioproducts to market in the shortest possible time and at low cost, there is also a need to produce unit and process models early in the process development cycle from only small amounts of sample material and with a minimum of effort. This provides the incentive for scaledown research in which accurate mimics of key unit operations are developed which require only limited quantities of feed material but can generate useful process information [13].