(233e) Improving the Batch Cycle in Bioreactors: From Media Makeup to Cleaning

On a bioprocess plant where batch fermentation is the means of production, improving overall productivity requires consideration of all aspects of the batch operation: from charging media, through optimizing reaction, to completing reactor clean out. Achieving consistency in operation and demonstrating strict adherence to product quality specifications has always been of paramount importance to the pharmaceutical industry. There have been significant changes to the manner in which this achieved, with pressures now to move from a situation where the same recipe is followed to one where deeper process understanding informs the actions taken to achieve consistency. This new philosophy is termed Quality by Design (QbD) where fundamental knowledge on process behavior guides design decisions and operational policy. In adopting the QbD philosophy it is necessary to consider all the phases of batch operation and for each ensure that there is sufficient information available to operate in a responsive, robust and reliable manner. The increasing acceptance of on-line process analytical technologies (PAT) has significantly enhanced the information content available contributing to fundamental knowledge and industrial moves towards the QbD philosophy has ensured that new technology is accepted.

This paper considers the phases of batch operation and how process measurement, control and optimization strategies may be exploited for business benefit. The foundation for improvement is the utilization of the broad range of process measurements to provide the fundamental insight into process condition that is critical to the QbD concept. Typically maximizing information content involves process data fusion from spectroscopic type instruments and combing possible multiple signals from such devices with conventional process data. The paper presents strategies by which this may be achieved and demonstrates the additionality gained compared to using single data sources. Several approaches are contrasted for data compression and fusion adopting either linear or non-linear algorithms to extract pertinent information. How they cope with typical practical characteristics such as varying sample frequencies, batch length variations and process noise is presented. The importance of appropriate data pre-processing is demonstrated prior to the application of compression and fusion algorithms in order to make the analysis more tractable.

During operation, awareness of process state is essential information for the operators. One approach is to translate complex spectral signatures into concentrations measures using calibration modeling. An alternative is to use the available process data as a ?signature' of condition and apply data compression methods to indicate condition in an interpretable manner for the operators. Both strategies have their place as calibration models provide useful information to control systems to respond to process disturbance or demand changes whereas the fingerprinting approach is useful for proof of consistency in a comprehensive multivariate sense taking into account the numerous concentration changes that could potentially impact on behavior. This paper contrasts different strategies for calibration modeling and fingerprinting with emphasis on ease of use and interpretability. Again, linear and non-linear variants of the algorithms are considered.

Addressing the phases that make up the batch cycle involves consideration of the make-up of media and inoculation of a seed, the monitoring and control of batch progression and selection harvest time and finally the clean-out of the process plant prior to starting the next cycle.

In batch operation, achieving consistent initial conditions as far as possible involves attention to media preparation and seed inoculum transfer. PAT type measurements have found application in raw material acceptance testing but for fermentations making use of natural media components a degree of variability inevitably arises. The application of PAT to determine seed transfer time to inoculate the production vessel offers distinct advantages. Here moving from time based transfer where seed condition varies to a transfer based on seed condition where inoculum age varies offers greater consistency in production performance. PAT measurement provides an indication of the physiological state which is a more pertinent indicator of performance on transfer to the production vessel. Indeed, previous studies have indicated that consistency in seed is crucial to production vessel behavior as sub-standard seeds cannot be compensated for by subsequent operational changes. The use of PAT devices and NIR in particular for seed transfer and seed quality performance impact will be demonstrated in the presentation.

Following inoculation, the availability of new measurement devices now provides greater insight into state profiles giving vital information on accumulation of media components or where levels are falling to limiting concentrations. The improvement of all phases of batch operation involves ensuring reaction proceeds in robust and consistent manner and disturbances, whether from media effects or process deviations, are adequately compensated for. Whereas previously off-line analysis was the primary source of such data, on-line spectroscopic measurements mean greatly enhanced frequency and significantly reduced measurement delay and consequently they facilitate a move from open-loop or limited closed loop through operator intervention to one where closed loop control of broth compositions is feasible. Again, examples of the use of spectroscopic devices for state profile measurement and control will be provided. Success depends upon the availability of accurate calibration models and the methods of data fusion described previously are shown to be beneficial in obtaining more precise measurements.

Following reaction, vessel emptying and clean out is undertaken. Ensuring this aspect of the cycle is optimized is also crucial to overall plant performance. In some instances, the clean out phase is significantly longer than the reaction phase hence focus on improvement by minimizing cleaning times can have a significant impact on overall plant yield. In cleaning, tracking of process state involves monitoring the approach to cleanliness and determining when the plant achieves this condition. The traditional strategy has been to adopt an open loop approach and clean the process in the same manner from batch to batch. The consequence of using this strategy is that while the process plant is clean, it is likely that excessive time and energy have been used. The QbD concept applied to the cleaning operation would necessitate a move from open loop to closed loop operation where cleaning progress is monitored and cleaning terminated when cleanliness is achieved. The risk based aspect of the QbD philosophy requires that cleaning termination time determination involves specification of the risk of failure. The benefits come from moving from an open-loop strategy where risk is minimized but not quantified or awareness of risk may be limited to the closed loop strategy where risk is understood and balanced against other operational considerations such as time taken and cost. Strategies for monitoring the progression of cleaning and forecasting the end-point are presented.

The paper draws on results from a number of industrial case studies both in pilot plant and production facilities to show how, with appropriate measurements and methodologies to interpret the data, operational benefits may be gained. Emphasis is placed on the practicality of the instrumentation and the information it provides, addressing in particular the different challenges faces during process development and process manufacture.