(653e) Development of a Computational Methodology for the Prediction of Mab Produced By CHO Cell Clones in Large Scale Bioreactors

Monoclonal antibody (mAb) treatments have been established as one of the most successful strategies to treat cancer, autoimmune diseases, and others over the past 20 years [1]. These are expensive therapeutic options, and there is a need to optimize mAb production to meet demand and reduce costs and cycle-times. Two-thirds of all approved mAbs in the market in 2017 were produced in CHO cells [2]. CHO cell-based production starts with Cell Line Development (CLD) to develop a CHO cell line that stably produces a large amount of mAb with the desired product quality profile, e.g. N-linked glycosylation [3]. The optimization of a process for Phase III and commercial production is time and resource intensive and is often done in high-throughput small, lab scale systems to limit costs [4]. However, scale up to larger systems results in longer blending times, thus higher spatial heterogeneities in important physiochemical components such as dissolved oxygen and pH, which are not observed in small scale experiments [5]. This in turn might impact the micro-environment that the cells experience within the bioreactor, and therefore the performance of the overall process. To optimize conditions for commercial production of mAbs, the spatial- and time-dependency of the environmental conditions must be considered.

In this study, to overcome these challenges a computational tool to predict mAb product titer and quality in large scale bioreactors with environmental heterogeneities is developed. The proposed modeling approach combines a metabolic kinetic model, built with small scale experimental data, with computational fluid dynamics (CFD) which predicts fluid flow and spatial heterogeneities in the large 2000L production bioreactor. The metabolic kinetic model is formulated to capture uptake/secretion rates of metabolites, mAbs and byproducts, such as ammonia and lactate, for multiple CHO cell lines producing the same molecule, media formulations, and process conditions based on industrial 15mL and 250mL scale experimental data. An N-linked glycosylation model uses the uptake and secretion rates from the metabolic model as inputs and predicts the glycan fractions produced for each CHO cell line with the various bioreactor conditions. Unique parameter sets for the same metabolic and glycosylation model structure are regressed for each cell line. Small scale one compartment scale-down simulator experiments supplement this data to capture the impact of oscillating physiochemical components expected at the large scale in dissolved oxygen and pH on cell metabolism, mAb production, and mAb quality on each cell line. CFD simulations are then performed with sources and sinks of these heterogeneous components to predict the formation of spatial gradients at the 2000L scale. Trajectories of tracer probes injected into the CFD simulations capture the time-dependence of oscillating conditions, mimicking the cellular experience in the 2000L bioreactor [6].

The ‘lifelines’ of predicted physiochemical exposures functionalize the model of CHO cell metabolism and glycosylation to predict the impact of heterogeneities expected in the large scale 2000L bioreactor on cell line specific metabolism, mAb production, and mAb quality. This computational approach could help identify cell lines that perform better at the large scale, reducing risks related to production and cost of large scale experiments.

References

1. Shepard et al. Clin. Med, 2017. DOI: 10.7861/clinmedicine.17-3-220
2. Dhara et al. BioDrugs, 2018. DOI: 10.1007/s40259-018-0319-9
3. Hong et al. Current Opinion Chem. Eng, 2018. DOI: 10.1016/j.coche.2018.08.002.
4. Li et al. Biotechnol. Prog, 2006. DOI: 10.1021/bp0504041
5. Lara et al., Mol. Biotechnol, 2006. DOI: 10.1385/MB:34:3:355
6. Haringa et al. Eng. in Life Sci, 2022. DOI: 10.1002/elsc.202100159