(270a) Systematic Development of Predictive Mathematical Models for Animal Cell Cultures

Unstructured models of animal cell cultures have been proposed as an appropriate basis for applying control, optimization and process development techniques to the production process of proteins (Dowd et al., 1999). Their main advantages include that they involve extracellular culture variables that are typically monitored during a culture, as well as that their simulation and subsequent in silico applications are computationally tractable. One of the major problems associated with unstructured models though is their limited range of applicability, since they are usually validated with experimental data from batch cultures. However, fed-batch cultures, which are known to increase levels of protein production and are preferred industrially, exhibit different behaviours, particularly in terms of cellular metabolism. It is therefore desirable to develop a systematic framework that creates predictive, unstructured or hybrid, tractable models of these processes.

In this study, a hybrid model of antibody-producing animal cell cultures is developed based on the relevant literature. This involves expressions for cell growth and death, metabolism of glucose and glutamine with the resulting generation of lactate and ammonia, and intracellular synthesis and production of monoclonal antibodies (MAbs). Batch cultures are performed using the CRL-1606 hybridoma cell line, which produces IgG1 MAbs against fibronectin from human plasma. The data generated is used to validate the model structure and provide initial estimates for model parameters. Those parameters to which the model output (MAb concentration) is most sensitive are identified through the Sobol method of global sensitivity analysis (Kontoravdi et al., 2005).

The accurate estimation of these parameters is then targeted through D-optimal model-based experimental design. The later dictates the appropriate feeding strategy for a fed-batch experiment so that the information content of the collected data is maximized. The experiment is then performed and model parameters are estimated, resulting in good agreement between model simulation results and the experimental profiles of cell, nutrient, metabolite and antibody concentrations. The predictive capability of the model is then checked against data from an independent set of fed-batch experiments. Results show that the proposed model correctly predicts the process variables monitored during the fed-batch experiments: viable and dead cell concentrations, glucose, glutamine, lactate and ammonia concentrations, and MAb concentration in the extracellular medium. These findings support the adequacy of simple models for process systems applications, as their extended applicability to fed-batch cultures is demonstrated (starvation conditions are not examined in this work).

References · Dowd J.E., I. Weber, B. Rodriguez, J.M. Piret and K.E. Kwok (1999). Predictive control of hollow-fiber bioreactors for the production of monoclonal antibodies. Biotechnology and Bioengineering 63(4): 484-492. · Kontoravdi C., S.P. Asprey, E.N. Pistikopoulos and A. Mantalaris (2005). Application of global sensitivity analysis to determine goals for design of experiments: an example study on antibody-producing cell cultures. Biotechnology Progress 21:1128-1135.