Digitalization platform and supervisory control of an integrated continuous bioprocess

Full title: 

Digitalization platform and supervisory control of an integrated continuous bioprocess

Abstract:

Integrated continuous manufacturing is entering into the biopharmaceutical industry. The main drivers range from improved economics, manufacturing flexibility, and more consistent product quality. However, studies on fully integrated production platforms have been limited due to the large complexity of such processes, limited process information, disturbances, and drift sensitivity, as well as difficulties in digital process integration so far. This work presents a digital platform of an end-to-end integrated process consisting of a perfusion bioreactor, continuous capture (CaptureSMB), virus inactivation, and two polishing steps, producing an aggregation sensitive antibody with an instable mammalian cell line. The platform enables the digital integration of the unit operations and analyzers, collects and centrally stores all process data, embeds advanced data analysis and modeling algorithms, as well as allows process-wide monitoring and control. The model-based supervisory control holistically controls the process and adjusts the downstream cascade in response to changes in the bioreactor. Despite significant process disturbances and drifts, a constant process performance and consistent product quality can be guaranteed, which is fundamental for continuous manufacturing. 

Furthermore, the application of Raman spectroscopy in upstream processing (USP) is elaborated and a complete predictive model approach is developed, including several pretreatment and preprocessing techniques, a unique calibration, and external testing strategy as well as several advanced machine learning based regression techniques. This is essential to build accurate and reliable predictive models for glucose, lactate, titer, and viable cell density, which reveals the immense potential of the Raman technology for online monitoring in USP. Motivated by the results in USP, a flow cell is developed, which meets the requirements (e.g., measurement in flow, reduce dead volume, signal intensification) to implement the Raman technology in downstream processing (DSP). Despite clear advantages compared to other spectroscopic techniques (e.g., MIR or NIR), the application of Raman spectroscopy has never been reported for DSP of biologics. The flow cell in combination with a unique modeling approach is applied to on-line detect the mAb concentration in the breakthrough of a protein A capture step. Due to the high impurity content within the sample, the measured spectra of harvest show no distinct peak profiles, and, therefore the multivariate model calibration is indispensable to analyze the titer. The estimation performance shows very promising results for a medium harvest titer (mAb < 2.82 mg/mL). Hence, the concept is repeated for a very low titer harvest (mAb < 0.42 mg/mL), where the Raman technology can indeed follow the shape of the breakthrough curves, but with a considerable noise. Therefore, a smart sensor approach is developed, using a chromatographic process model as backbone, while the Raman-based real-time information updates the state estimates. The approach is superior in terms of reducing noise, diminishing offsets, and increasing the robustness against model input perturbations. This allows the monitoring in even low titer harvests, which are common in perfusion cultivations. Hence, it is concluded that the Raman technology is also very promising for DSP and more applications should be investigated. Implemented into an integrated continuous bioprocess, Raman spectroscopy could deliver a vast amount of on-line information of which the process-wide control could greatly benefit.

Finally, two model-based strategies are developed to investigate the resin aging of protein A, which is the commonly used, though most expensive resin, applied in capturing. The strategies include a statistical and a hybrid model approach. The former strategy allows the qualitative monitoring of the column state and predictions of column performance indicators (e.g. yield, dynamic binding capacity and impurities) by using chromatographic on-line data. The latter provides important insights onto the prevailing aging mechanism in dependence of different cleaning procedures (CP). Furthermore, it can be applied for model-based CP optimization and yield forecasting with the capability of state estimation corrections based on on-line information. Both approaches show promising potential to extend the resin lifetime and greatly support in terms of predictive maintenance once embedded into a digital platform. This work proofs that a digital platform, allowing a process-wide monitoring and control structure, equipped with advanced data analysis and modeling based techniques, supplied with on-line process information from modern process analyzers, is crucial to enable and leverage the advantages of integrated continuous manufacturing. Moreover, the concept is highly supported by regulatory authorities and will greatly accelerate the digital transformation within the biopharmaceutical industry.