Biomass performance : monitoring and control in bio-pharmaceutical production

The primary concern in the pharmaceutical industry is not the optimisation of product yield or the reduction of manufacturing cost, but the production of a product of consistently high quality. This has resulted in 'process monitoring' becoming an integral part of process operation. In this thesis process monitoring is one of the central themes, from monitoring the environment of the micro-organisms to monitoring the micro-organisms themselves. The latter is called monitoring biomass performance . Subsequently, in this thesis, monitoring is applied to predict the future trajectory of the cultivation and thus the time to harvest. Furthermore, it is applied to indicate process consistency and thus to give assurance about product quality. Finally, monitoring is applied to control the specific growth rate in order to improve process consistency and thus to improve assurance of product quality.One of the key metabolic indicators suitable for monitoring the performance of cell cultures is the respiration quotient. Usually gas analysis is used to monitor this variable. However, in bicarbonate buffered media the carbon dioxide balance is affected by accumulation and hence the respiration quotient can not directly be calculated from gas measurements. A Kalman filter is introduced that estimates the carbon dioxide evolution rate and copes with these buffering capacities. The model used by the Kalman filter lumps all carbonate in the liquid to one term in order to eliminate the role of a priori knowledge of cell and medium kinetics without affecting the performance. A reference experiment verified the performance of the software sensor and subsequent experiments with insect cells showed the progress of the respiration quotient during cultivation.To take advantage of the newest techniques in monitoring and control, more knowledge of the micro-organism involved is necessary. The specific growth rate is an important indicator of biomass performance . An off-line algorithm is developed that is capable of determining the specific growth rate from off-line biomass samples. The technique is based on combining subsequent backward and forward extended Kalman filtering to give a smoothed and optimal estimation. It can be used as a powerful tool for various organisms since again no a priori knowledge of the micro-organism is needed. This estimator gained improved knowledge of the growth characteristics of Bordetella pertussis and Neisseria meningitidis . Both organisms seemed to be simultaneously limited by two substrates. However, for the case of B. pertussis , neither interactive nor non-interactive modelling seemed appropriate and a model that combines essential and enhanced kinetics was developed based on experimental observation. Instead of fitting all model parameters at once, a step-wise experimentation procedure was used and the accuracy of the dual-substrate model was shown by two cultivations.The next step in monitoring the cultivation step in pharmaceutical production is monitoring the biomass concentration and its growth rate on-line. Due to the lack of reliable and cheap sensors, they cannot often be measured directly or estimated from related variables, such as the concentrations of substrates or products. A sequential observer is introduced to estimate the specific growth rate and the biomass concentration for processes where the measurement of oxygen uptake rate is available on-line. The applicability of the algorithm is proven by the good agreement between the sequentially estimated values and the measured values for the cultivation process of B. pertussis .One of the applications of this monitoring system is the combination of the current estimated state of the cultivation with the dual-substrate model of B. pertussis in order to reconstruct on-line, the total current state of the cultivation (e.g. biomass and substrate concentrations). With the estimated current state, the future trajectory of the cultivation is computed by forward simulation. This prediction algorithm successfully aids the operator in determining and scheduling the right moment of harvest on-line. Thus, this prediction algorithm gains the advantage of letting the moment of harvest being dictated by the state of the organisms instead of by a predetermined time-schedule, which ultimately leads to improved process consistency.Another application of the monitoring system is the use of the available on-line estimates to track whether the process remains within specifications. The sequential estimator gains more information than regular process monitoring. Now biomass performance is available on-line, and can thus be used to indicate consistency at a metabolic level. For the case of B. pertussis , the final product test is statistically hardly capable of indicating batch-to-batch variation. Monitoring on-line process parameters provides a better assurance of consistency and quality, thus giving the opportunity for batch release and regulatory approval without the use of the final animal tests.To further improve process consistency during cultivation, the metabolic state of the micro-organisms in the bioreactor has to be kept constant. By controlling the specific growth rate of biomass such metabolic stability is guaranteed. This thesis demonstrates the combined use of the sequential estimator, the dual-substrate model, and a feed-forward-feedback-controller to keep the specific growth rate constant for B. pertussis . Successful tests were performed in real experiments where the specific growth rate was controlled at reduced specific growth rate over a period of more than 15 hours, quadrupling the total amount of biomass. In order to apply this controller to other processes it is necessary to conduct a relatively small number of experiments providing enough data to determine and tune the parameters.All techniques in this thesis are proposed in a modular approach where each tool or module is implemented in a separate software routine, such that they can be tuned, tested, improved and validated separately. Connecting the modules results in one of the mentioned combined applications.