|Zusammenfassung||The rate of nonalcoholic fatty liver disease (NAFLD) such as steatosis and nonalcoholic steatohepatitis (NASH) in populations is continuing to grow vigorously and became a worldwide public health issue. To understand the liver disease progression one needs to investigate complex interactions occurring within biological systems. Systems biology tries to understand the interactions within biological systems by means of mathematical models. Exploiting this approach I want to describe interactions of genes, proteins and metabolites that are involved in nonalcoholic fatty liver disease. Molecular data from liver tissue samples of three mouse strains (A/J, C57Bl6 and PWD) under two different conditions: 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC)-treated and untreated (control) were analyzed. Each of these mouse strains shows different degrees of the disease under DDC treatment displaying high, low, and no steatohepatitis-like phenotypes for A/J, C57Bl6 and PWD, respectively.
In this work I performed pathway analysis using gene expression data of mouse liver samples and identified metabolism of histidine, beta-alanine, purine along with glycolysis and gluconeogenesis pathways as top hit candidates that may be involved in liver dysfunction. Furthermore, gene expression and metabolite data of the arachidonic acid metabolism were found to be deregulated and this pathway was used for kinetic modeling. Genes and metabolites of S-adenosylmethionine (SAMe) metabolism were found to be perturbed under DDC-treatment. In addition, I have developed a novel enrichment analysis approach that may be used for identification of the most relevant fully coupled modules in a disease context. The approach includes three steps: 1) obtain fully coupled reactions which represent a module, 2) use gene expression data of a disease context to obtained marked correlated modules and 3) select modules in which at least one gene is differentially expressed between normal and disease conditions. Aforementioned steps are used to identify liver disease specific modules such as modules of pentose phosphate pathway and hepatic SAMe metabolism which are linked to oxidative stress. Furthermore, I also identified a module of cholesterol metabolism which is linked to apoptosis along with a module of pyrimidine catabolism, for which experimentally measured genes and metabolites were also found to be deregulated. The goal was to identify modules for which genes and metabolites are perturbed under DDC-treatment. The identified modules may be involved in liver disease and they can be used to build kinetic models for better understanding of the liver disease progression. Thus, in addition to enrichment analysis of fully coupled modules, I developed an approach which is based on elementary flux modes (EFMs). In this approach initially differentially regulated metabolites due to DDC-treatment were identified. Then reactions which can produce differentially regulated metabolites were used as a target set. For each reaction in the target set, 50 EFMs that contain the reaction were identified. After that, gene expression data of mouse liver samples were used to select important EFMs that may be involved in the liver disease progression. I identified two EFMs: one EFM comprises differentially regulated metabolites L-arginine, ornithine and putrescine and another EFM comprises differentially regulated metabolites D-glucose, L-glutamine and L-asparagine.
I introduced a mGX-FBA method which is a modified version of the previously published GX-FBA method. Differences between metabolic flux levels among mouse strains may provide a better understanding of the reasons behind NAFLD. To address this, I performed an in silico flux-based analysis using E-Flux and the modified mGX-FBA method by incorporating gene expression data of the mouse model. Furthermore, during the course of my thesis I compared the results of both methods, E-Flux and mGX-FBA, and validated the results with experimental data. The change of flux through metabolic pathways may change metabolic concentrations. Due to the absence of experimental flux data for mouse liver samples it is difficult to assess in silico predicted flux regulation. However, in silico flux regulation may give a hint about the regulation of metabolic concentrations. Hence, to observe the flux regulation I colored metabolic maps with the deviation of fluxes between DDC-treated versus control. Different degrees of flux regulation was identified through cholesterol biosynthesis among all three mouse strains. The concentration of desmosterol that is a downstream metabolite of cholesterol biosynthesis was found to be regulated at different degrees. The regulation of desmosterol concentration is inline with the flux regulation of cholesterol biosynthesis among all three strains. In addition, to understand whether one can speculate about the prediction of flux regulation in metabolic pathways based on gene regulation I used bile acid synthesis and cholesterol biosynthesis pathways. For these pathways, I observed that based on gene expression data it is difficult to estimate flux regulation, but the integration of gene expression data using E-Flux improves the prediction of flux regulation.
I introduced novel objective functions for measuring the readout of steatosis and steatohepatitis. To construct an objective function for steatosis I used metabolites which are involved in the formation of lipid droplets (LDs), while for steatohepatitis an objective function integrating both metabolites that are involved in LDs formation and metabolites involved in oxidative stress was used. Gene expression data of liver samples of all three mouse strains were incorporated to a mouse metabolic model and obtained objective values were validated with strain's phenotypic data. I also constructed an objective function with metabolites which may be involved in cell proliferation. Cell proliferation is used as a readout of hepatocellular carcinoma (HCC). Applying these functions I performed an in silico drug target analysis in which potential drug candidates for steatosis, steatohepatitis and HCC were identified. Cholesterol metabolism and triacylglycerol synthesis were found as top hits that contain the largest number of potential drug target candidates. Out of 78 identified potential drug candidates, 7 were found to be approved by the Food and Drug Administration (FDA) as anticancer drugs.
Metabolite concentrations can be viewed as end points of perturbations occurring at the gene level, so that changes of gene expression might explain changes in metabolite concentrations. I proposed a novel hypothesis to predict changes in metabolite concentrations between two conditions based on gene expression data. To address this, I have developed a Petri net-based method (MPN) and used it to simulate the arachidonic acid model. As an alternative to MPN, Monte Carlo ODE-based simulation was used, but both methods cannot predict the metabolic concentrations in the real range of experimental data. To overcome this, I have developed a fitted detailed kinetic model of the arachidonic acid metabolism that comprises metabolites that are markedly deregulated due to DDC-treatment in all three mouse strains.