Accelerating process development through analysis of cell metabolism

Abstract

Background Biopharma process development accumulates growing collections of physicochemical product information and fermentation data, partly in response to initiatives like Process Analytical Technology (PAT) and Quality by Design (QbD) backed by regulatory authorities. The key goal of these initiatives is to ensure robust processes and consistent product quality based on a scientific and mechanistic understanding of the product compound itself and of the manufacturing process. Biopharma companies gather much high-quality data during fermentations including online measurements and omics data such as transcript and metabolite measurements. Typically, these data will be stored for documentation purposes only whereas data evaluation and interpretation lags behind and often is sporadic at best. Cellular network models can tap this underused resource for predicting fermentation outcomes and for analyzing why certain fermentations failed or succeeded based on a mechanistic representation of cell physiology. In particular, network models of cell metabolism upgrade metabolomics data by enabling predictions of cell behavior from concentration time series of extracellular and – if available – intracellular metabolites. Model simulations can be used for rapid hypothesis testing, e.g. to evaluate the impact of changes in feeding on intracellular metabolism, growth, or product formation. Identifying suitable metabolic target genes for cell line engineering represents another application area of such models. Here, we illustrate this approach using the prediction of optimal media compositions for a Chinese hamster ovary (CHO) cell line employing a genomebased CHO network model as example. Methods The CHO stoichiometric metabolic network was reconstructed using information from public databases as well as from primary literature and accounts for the specific amino acid composition and glycoform structure of the product molecule. In a first step, we applied the network model to a comprehensive metabolic characterization of the existing fermentation process. Rates of cellular nutrient uptake, growth, and product formation in physiologically distinct process phases were determined from concentration time series of extracellular metabolites during a fermentation run. These cell-specific rates served to compute intracellular flux distributions using the CHO network model. Comparing flux distributions for different process phases provided insight as to when and where in intracellular metabolism significant changes occur during the fermentation. This is often not obvious from inspection of concentration time series alone. Especially for fed-batch processes, multiple feed streams and volume changes due to pH control and sampling impede interpretation of raw data. If desired, further information about the usage of alternative intracellular pathways and in vivo reaction reversibilities can be obtained from labeling experiments combined with transient C-Metabolic Flux Analysis [1,2], which is applicable to industrial fed-batch settings. Intracellular flux distributions also provide an ideal starting point for process optimization. Distinct optimal media compositions were computed for different fermentation phases based on the observed nutrient demand of the clone inferred from flux distributions. The chosen optimization approach combines stationary and dynamic model simulations on high-performance computing clusters. For dynamic simulations, the stoichiometric CHO network representation was transformed into a kinetic model. Model parameters were determined using evolutionary strategies and cluster * Correspondence: dirk.mueller@insilico-biotechnology.com Insilico Biotechnology AG, D-70563 Stuttgart, Germany Muller et al. BMC Proceedings 2011, 5(Suppl 8):P81 http://www.biomedcentral.com/1753-6561/5/S8/P81