(598d) Modeling of a Recombinant Protein Production System: Cell Culture Dynamics and Dynamic Metabolic Flux Analysis

Process development for new antibody products typically starts with selecting candidate cell lines that fit an existing platform process and then tailoring this process to improve antibody productivity, cell vitality and product quality with respect to critical quality attributes (CQAs). Process optimization to achieve the desired outputs relies on in house expertise and historical datasets, and usually consists of a trial-and-error approach. Besides requiring a large investment in time and resources, this approach can lead to only sub-optimal solutions. On the contrary, systems biotechnology can provide a more systematic and rational approach to optimization and aspire to predict how one system changes over time and under varying conditions.

In theory, first-principles models can be used for prediction. However, even for first-principles models the extent of prediction capability greatly depends on the information content and range of the experimental information the model is based on. This is especially true for unstructured models.

In this work, we aim to capture the physiological response of Chinese hamster ovary (CHO) cells to extracellular changes during antibody-producing fed-batch culture as well as to elucidate resource utilization under three different feeding strategies and during the different cell culture phases. To reach these goals we have developed a new model structure comprising two levels. At the extracellular level, cell culture dynamics (CCDyn) are described using a Monod-type kinetic model. Segregation among proliferating, resting and dead cell subpopulations is incorporated into this kinetic model, unlike the classic approach which is usually unsegregated. At the intracellular level, cell metabolism is described using a stoichiometric approach. This coupling is based on the assumption that intracellular reaction kinetics are much faster than the changes in the extracellular process (at least between two subsequent feeding times) so that pseudo-steady state can be assumed to apply for all intracellular reactions. The intracellular compartment is modeled through a fully determined dynamic metabolic flux analysis (dMFA) which describes, in a parameter-free manner, the most significant metabolite fluxes in both the cytoplasm and the mitochondrion. The flux model includes glycolysis, TCA cycle, amino acid metabolism, NS metabolism, as well as mAb glycosylation. One novelty compared to the existing models in CHO systems, is that the MFA includes the Aspartate-Malate shuttle, which directly links amino acid metabolism with TCA. The inputs to this model are the nutrients uptake and bio-products secretion rates calculated by the CCDyn model, while the outputs are the intracellular metabolite fluxes varying with respect to time as a function of changes in the extracellular environment.

Generally speaking, a stoichiometric modelling approach cannot be quantitatively predictive, even in the dynamic version we adopted. However, it is particularly useful to elucidate a metabolic network structure in the different cell culture phases and inform the structure of the CCDyn model. In particular, by including the Asp-Mal shuttle, the dMFA provides additional information on how amino acids (namely, asparagine and glutamate) are consumed towards energy metabolism. Moreover, our dMFA strategy directly links asparagine consumption with energy metabolism and ammonia secretion. Such information is particularly useful in the proposed approach because the CCDyn model is developed as dependent on the extracellular concentrations only. As a consequence, all the experimental data required to develop and validate the model come from the analysis of the supernatant only.

We will present our evaluation of the prediction capability of this new model structure against an independent dataset from fed-batch CHO cell cultures. We will show that the proposed CCDyn model can accurately predict key performance indicators without the need for parameter re-estimation for the new experimental conditions. We can conclude that the proposed approach, that is coupling a segregated Monod-kinetic model with a parameter-free dMFA, is able to give enough information at a reduced price in terms of computational and experimental effort. For this reason, this study represents a significant step towards predictive modeling of animal cell cultures.