Comparing Condition Specific Genome-Scale Metabolic Networks Obtained from Multi-Omics Data: Exploration of the Metabolic Shifts Occurring during the Differentiation of a Human Hepatic Cell Line

In this study, we combined multi-omics data and in silico methods to better characterize the metabolic capacities of HepaRG, a human hepatoma cell line increasingly used in toxicity studies. This cell line has the particularity to differentiate from bipotent progenitor cells to fully differentiated hepatocyte-like cells. We exploited transcriptomic and metabolomic data in the context of the global human genome-scale metabolic network Recon2, to compute relevant sub-networks, more specifically representing the functional metabolic network of the HepaRG cells at each developmental stage, in order to explore the metabolic shifts occurring during differentiation.

Gene expression and metabolomic (¹H NMR) data were obtained on the HepaRG cell line in both developmental stages. In a first step, we applied the iMAT algorithm developed by Shlomi et al. to identify, based on gene expression data, the sub-networks of reactions specifically active in HepaRG cells at each stage. The goal was to find a coherent model that maximizes the number of reactions whose activities match the expression state of the associated genes. Metabolomic data were then used to further select the most adequate sub-networks, by selecting those for which a higher number of observed metabolites were predicted to be present

Using this pipeline, we were able, for each differentiation stage, to identify several sub-networks of active reactions, having an equivalent adequacy to experimental data. We applied classification analysis methods (cluster analysis and principal component analysis) to explore the intra-stage and between-stages variability in the computed sub-networks of active reactions. We showed that, within a particular developmental stage, the heterogeneity between computed sub-networks was mainly caused by the occurrence of several alternative reactions in a same pathway (e.g., keratin sulfate degradation pathway) or by the scarce number and/or contradictory aspect of transcriptomic data in a pathway (e.g., heme synthesis). Classification and pathway enrichment analyses performed on all sub-networks from both stages enabled to identify several sets of reactions that were differently active in these two stages. Interestingly, about 70% of these reactions could not have been directly inferred from transcriptomic data and were « newly » inferred from the computational model. Comparison of reactions predicted to be specifically active in differentiated cells compared to progenitor cells provides key information about the metabolic changes occurring during the transition between these two stages. These reactions mostly belong to pathways specific to the development of the hepatic activity (bile acid synthesis, amino acid metabolism…) and anabolic/catabolic processes including detoxification functions (cytochrome metabolism). Such context-specific functional reconstruction is a first step in the determination of the metabolic capacities of key cell lines used in toxicology.