Process models for CHO bioprocess optimization and scale-up

Abstract

Bioprocess development for biologics primarily relies on data-driven tools and models. While these methods enable a good description of early phase development data, they do not necessarily result in enhanced understanding of the developed process. This also holds true for later development steps, when the process needs to be transferred from laboratory to production scale. Although there are engineering rules for scale-up, process performance rarely matches between small and large-scale. Pinpointing the source for variations is often unfruitful, since little is known about how the large-scale environment impacts the cells. This thesis aims to address these challenges throughout process development. For the early stages of process development, a mechanistic, rather than a data-driven approach was explored for process optimization. Furthermore, a scale-down simulator was developed to simulate the large-scale environment of a production reactor to increase understanding of how the cells are affected by this environment. Design of Experiment (DoE) studies are the golden standard to generate data-driven models during the early stages of process development. Drawbacks of this approach include limited extrapolation ability, challenges in evaluating dynamic processes, as well as the necessity of a high number of experiments. To avoid these disadvantages, mechanistic modeling is explored as an alternative to the DoE approach. The goal of the model is to determine improved time points for pH and temperature shift to increase culture longevity as well as the volumetric productivity. Based on the model predictions, an increased final product concentration of 14% was achieved in comparison to an already established industrial fedbatch process with the same cell line. By applying the mechanistic model, the number of required experiments was reduced in comparison to a generic DoE approach and extended process understanding was generated by predicting time windows for similar process performance. Therefore, the mechanistic model is a good alternative to the DoE approach for process optimization.Upon successful optimization of the process at lab-scale, it needs to be transferred to production scale. The volume of a production scale bioreactor can however be up to 10.000times bigger than that of a lab-scale bioreactor. This results in higher mixing times for the large vessels and can lead to the formation of gradients, which are not present in the lab scale bioreactor. Due to these differences between the scales, process performance can be negatively affected at production scale. Since investigations at large-scale are impractical and expensive, a scale-down simulator was developed, which replicates the large-scale bioreactor by allowing the exposure of only a fraction of the cells to gradients. Particularly pH gradients have the potential to negatively impact process performance, since mammalian cells are sensitive to pH changes. Therefore, cells were exposed to reoccurring pH amplitudes up to a magnitude of 0.4 pH units (pH 7.3). This resulted in a decreased viable cell count, as well as the absence of the lactate metabolic shift, which has been associated with improved process performance. Product quality was also influenced by the pH excursions. Particularly less galactosylated glycoforms were observed, when cells were exposed to the amplitudes. Even the introduction of smaller pH amplitudes, with a magnitude of only 0.1 pH units (pH 7.00), had an impact on process performance, showing the sensitivity of the cells to pH excursions. These results show the necessity to better understand the response of mammalian cells to the large-scale bioreactor environment to ensure a successful scale-up process.