(528c) Hybrid Kinetic-Stoichiometric Model of CHO Cell Fed-Batch Process

Mammalian cells are used to produce half of the commercially available therapeutic proteins, with Chinese Hamster Ovary (CHO) cells representing the major fraction of the used cells. CHO cell systems enable the application of certain post-translational modifications to obtain specific structures of protein products that are compatible with humans. Production processes involving CHO cell lines typically operate in fed-batch mode, as it allows cells to be fed throughout the cycle, thus preventing the culture from nutrient depletion before production goals are met.

Process models of fed-batch CHO cell cultures offer little predictive power due to having been fitted for a particular set of culture and process conditions. Whenever process or culture conditions change, the models must be updated with new experimental data, and new kinetic parameters must be estimated. This works aims to reduce this burden by developing a hybrid kinetic-stoichiometric model, based on a state-of-the-art genome-scale metabolic model.

A single CHO cell contains thousands of metabolites, genes and reactions. Thus, including a full description of the metabolism of a CHO cell would make the process model computationally intractable. As a result, this work also investigates the trade-off between the granularity, the computational burden and the predictive accuracy of metabolic network models.

The proposed hybrid model combines a fed-batch CHO cell culture model with a CHO cell metabolic model. The model of the fed-batch process is given by the total biomass and key metabolites balance, written as a system of ordinary differential equations. The metabolic model takes the form of a genome-scale model, which is a constraint-based model, that relies on a system of linear equations to describe the mass balance of each metabolite. Given the high degrees of freedom, the system of equations is solved as an optimization problem, where the metabolic fluxes are constrained by experimental data.

The proposed framework establishes the link between the metabolic and process models via the fluxes of secreted metabolites. The dynamic formulation of the metabolic model provides the fluxes of secreted metabolites, biomass and target protein produced as a function of time. By having a set of shared variables, we are able to connect both the fed-batch and metabolic models.

The proposed framework is capable of simulating a fed-batch CHO cell culture process, with an embedded dynamic genome-scale model. Given a starting state of process conditions, the model is also able to predict the uptake and secretion rates of key metabolites, as well as the biomass and protein produced.

In this work, we propose a hybrid modelling framework that overcomes the parameter estimation burden of fitting several kinetic parameters, by taking advantage of state-of-the-art genome-scale models. This represents a step forward for accurately modelling fed-batch CHO cell cultures.