Feed optimization of a cell culture fed-batch process using a mechanistic model

Abstract

Computational models are a powerful tool with numerous applications, also in pharmaceutical industry. In general, modelling biopharmaceutical processes is more challenging than traditional processes due to their high complexity. However, because of the multitude of applications models are becoming increasingly popular in bioprocess engineering, too.The aim of this study is to develop a mechanistic model for a fed-batch culture of CHO cells that produces a therapeutic monoclonal antibody. Therefore, a target-oriented macroscopic mechanistic modelling approach was chosen. It is generic and can also be used for other cell cultures independent of cell line, product and scale. Further benefits of this approach are that no model libraries are necessary and that it requires comparatively little prior process knowledge. As target variable the viable cell count was defined.The model follows two purposes in this study. On the one hand the amount of the daily added feed is optimized by means of the model in order to maximize the integral of viable cells and, thus, product titre.On the other hand it is used to create further process understanding by identifying both limiting substrates and inhibitors. Since interpretable mechanistic links are applied, critical levels of these substances can be determined directly from model parameters.The final model is based on aspartic acid and alanine acting as limiting substrate and inhibitor, respectively. As well, the total cell count is considered. Within the normal operating range, the final model estimates the viable cell count throughout the whole process duration of 14 days with an error less than 10 %. This criterion is also fulfilled in a variation outside the normal operating range. Other modelled states show an acceptable quality.The application of the model is limited to the investigated process and mentioned limitation and inhibition.Model-based feed optimization does not lead to an improved result showing that feeding already seems to be close to its optimum regarding given constraints.The model is intended to be used to design model-based experiments for process characterization.Robustness due to applied quality checks make the adapted model also suitablefor online applications such as process monitoring or process control. Specific adaptation of model parameters on a particular run moreover improves accuracy of model predictions.