Multivariate monitoring, modelling and control for stabilization of bioprocesses

Abstract: The obstacles to overcome low reproducibility and stability of bioprocesses are numerous. Underlying biochemical processes are inherently non-linear, complex and subject to shifting initial conditions. Problems with high variability are also associated to production strains and scale-up of a bioprocess to largescale bioreactors. Reliable on-line monitoring of key process variables is still a challenging task and hinders the closed-loop control of these key process variables. In this thesis, methods for stabilization of bioprocesses by means of multivariate on-line monitoring, modelling and control are studied.The foundation was laid with the development of integrated multivariate bioprocess monitoring, modelling and control within a real-time knowledge-based expert system. Thereby, a large number of signals from different advanced on-line analyzers ranging from mass spectroscopy via on-line HPLC to nearinfrared spectroscopy and electronic noses, could be used in combination with a variety of multivariate modelling and control tools for a flexible development of methods for stabilization of bioreactor processes. Subsequently, it could be shown how problems related to the initial conditions of a bioprocess can be solved by a multivariate assessment of the preculture quality. Furthermore, it was demonstrated how qualitative and quantitative key process variables can be made available and applied for process supervision; here, multivariate statistical process modelling and neural network sensor fusion from on-line monitoring of bioprocesses with advanced on-line analyzers were used. Finally, a closed-loop control method was presented, showing how feedback control of a multivariate key process variable trajectory can improve adherence to the specifications of the bioprocess. As model systems, aerobic fed-batch cultivations using recombinant Escherichia coli and anaerobic yoghurt batch fermentations have been used. The results provide general methods for multivariate stabilization of bioprocesses in precultivation steps, laboratory-scale and production-scale. They show that multivariate monitoring, modelling and control can provide a functional and versatile framework for reduced batch-tobatch variation and stabilization of bioprocesses with possible implications on product quality and process economics.