Novel Data-Driven Mechanistic Modeling of Untargeted Metabolome Data Reveals Feed Component Effects in CHO Cell Bioprocess Using Column Generation-Based EFMs

Abstract

This study presents a novel approach for applying mechanistic metabolic modeling to untargeted metabolomics data. The approach was applied to the production process of a difficult-to-express enzyme by CHO cells, to identify key feed medium component candidates responsible for improved productivity through feed modification. The exploitation of untargeted metabolomics implies no prior decision of the metabolites or pathways and thus allows screening of metabolic phenomena and bringing an objective perspective. However, such exploitation is challenging due to the high-dimensionality, complexity, relative quantitative information, and high analysis cost of the data, leading to data scarcity. A combination of untargeted metabolomics data exploration and mechanistic modeling was developed to leverage metabolomics data. The study analyzed LC/MS/MS metabolomics data (563 cellular and 386 supernatant metabolites) to determine the key metabolites involved in the productivity increase associated with a feeding modification. The metabolome data was utilized to expand the original stoichiometric reaction network of 127 reactions to 370 reactions. Mechanistic modeling using elementary flux modes-based column generation identified and simulated the underlying metabolic pathways. Twenty-one key metabolites significant for productivity improvement were revealed. This included several unexpected metabolites, such as citraconate and 5-aminovaleric acid, in addition to well-known components, as well as their underlying metabolic pathways. This study offers a novel approach for investigating nutrient supplementation in terms of metabolic fluxes and process performance, paving the way for rational process optimization supported by mechanistic understanding.