As one of the most recent members of the omics family, large-scale quantitative metabolomics data are currently complementing our systems biology data pool and offer the chance to integrate the metabolite level into the functional analysis of cellular networks. Network-embedded thermodynamic analysis (NET analysis) is presented as a framework for mechanistic and model-based analysis of these data. By coupling the data to an operating metabolic network via the second law of thermodynamics and the metabolites' Gibbs energies of formation, NET analysis allows inferring functional principles from quantitative metabolite data; for example it identifies reactions that are subject to active allosteric or genetic regulation as exemplified with quantitative metabolite data from Escherichia coli and Saccharomyces cerevisiae. Moreover, the optimization framework of NET analysis was demonstrated to be a valuable tool to systematically investigate data sets for consistency, for the extension of sub-omic metabolome data sets and for resolving intracompartmental concentrations from cell-averaged metabolome data. Without requiring any kind of kinetic modeling, NET analysis represents a perfectly scalable and unbiased approach to uncover insights from quantitative metabolome data.
Systems biology strives to gain a quantitative genome-scale understanding of the complex and highly interrelated cellular processes and phenomena. Such in-depth understanding will ultimately be achieved by a tight interplay between the two prominent pillars of systems biology: mathematical models and omics data. In the context of the latter, owing to the recent development of affordable and powerful mass spectrometers, large-scale sets of quantitative metabolome data are currently complementing our data pool (Goodacre et al, 2004; Nielsen and Oliver, 2005).
In order to fully exploit the wealth of information contained in large-scale data sets and to convert data into a body of knowledge, integration into mathematical models is required. For quantitative metabolome data, kinetic models describing enzyme reaction rates would represent the natural way for computational analysis. However, because of the lack of comprehensive knowledge about in vivo reaction mechanisms and parameters, and the still existing challenges on the measurement side as well as on the computational analysis side, it is very unlikely that large-scale kinetic models will become available in the near future. Until today, large-scale sets of quantitative metabolome data cannot be assimilated into mathematical models (Nielsen and Oliver, 2005) and thus, insight, for instance into underlying regulatory mechanisms, can hardly be inferred.
In this work, we present a computational thermodynamics-based framework for the analysis of quantitative metabolome data, whereby the mapping onto a stoichiometric reaction network and a coupling to fluxome data allow for extraction of novel insight from the data without requiring any kind of kinetic modeling. More specifically, in the developed network-embedded thermodynamic analysis (NET analysis), experimentally determined intracellular fluxes and metabolite concentrations are coupled to each other via the second law of thermodynamics and the metabolites’ Gibbs energies of formation, whereas an optimization algorithm is employed to resolve network-constrained, feasible ranges of Gibbs energies of reaction along with feasible ranges of unmeasured concentrations (Figure 1).
We first examined a small set of measured metabolite concentrations obtained from an Escherichia coli chemostat culture to illustrate the concept and the application of NET analysis and to demonstrate its ability to extract insight from even limited metabolite data. First, we showed that NET analysis could serve as a tool to check thermodynamic consistency of a data set. Thermodynamic consistency was approved for the analyzed data set, although several other published data sets were found to be infeasible, which emphasizes the need for quality analyses of metabolome data before they enter databases or are used in modeling efforts.
In a next step, we investigated whether NET analysis could also be used for prediction of unmeasured metabolite concentrations. In the analyzed data set, besides a few measured metabolites, the measurement provided only pooled concentrations for several isobaric molecules. With NET analysis, it was possible to resolve narrow concentration ranges for the individual metabolites. Moreover, concentration ranges were also predicted for some unmeasured metabolites. It can be envisioned that this predictive capability of NET analysis will support the development of more efficient analytical methods, as computable concentrations do not need to be determined experimentally.
Measured metabolite concentrations hardly provide any insights into the organization of metabolism, that is, the regulatory structure responsible for routing of matter via the different metabolic pathways, the result of which is a certain intracellular flux distribution. A flux distribution is established by the fact that in comparison with the neighboring reactions, the rates of some reactions, are limited by the available catalytic activity, so that at branch points, mass flux is accordingly distributed into the possible pathways. A limited catalytic activity of a reaction manifests itself in a large Gibbs energy of reaction. Reactions operating far from equilibrium are more likely to impose flux control (Wang et al, 2004), and it is assumed that such reactions are more likely to be regulated by the cell (Crabtree et al, 1997).
With NET analysis, reactions under putative active genetic or allosteric regulation can be identified from (even incomplete) metabolome data. For the data considered here, the respective results are provided in Figure 4. In perfect agreement with earlier findings, the pyruvate kinase and phosphofructokinase reactions were identified as regulatory sites in glycolysis, and also most other findings comply with our current knowledge, indicating that NET analysis of metabolome data is indeed able to provide correct regulatory insight.
Another finding, exemplifying the method's ability to uncover unknown functional relationships in the metabolic network, is related to E. coli's cytoplasmic transhydrogenase (udh): in glucose-limited continuous cultivations, compared to the biosynthetic demands, an excess of NADPH is produced (Nanchen et al, 2006). In order to eliminate excess NADPH, the udh-transhydrogenase reaction converts NADPH into NADH. In the considered experiment, NET analysis revealed that this reaction operates far from equilibrium, indicating that it is subject to active regulation. Furthermore, we found that a regulatory control of the udh-transhydrogenase is indeed required for physiological reasons: a further equilibration of the reactants, corresponding to a shift of the NAD(H) pool to the reduced state, would render the normal operation of catabolic (i.e. NAD-dependent) dehydrogenases infeasible.
Finally, we tested whether our approach is also applicable to more complex systems such as organisms with subcellular structure. For this, we extended the method and analyzed the largest available quantitative data set from Saccharomyces cerevisiae. Also here, after having approved thermodynamic consistency of the employed data set, we were able to identify active regulatory sites, consistent with our current knowledge. Furthermore, it was found that NET analysis can also be used to infer compartmental differences from cell-averaged metabolome data: operational differences between reactions occurring in various compartments could be uncovered and even intracompartmental concentrations for a series of metabolites could be resolved.
In summary, to gain insight from quantitative metabolome data, a coupling to mechanistic models is required. As an integrative analysis with kinetic models will remain a major challenge at least in the near future, we presented a new methodology (NET analysis) for the model-based analysis of quantitative metabolite data. NET analysis is easy to apply, perfectly scalable to the systems level and it only relies on indisputable thermodynamic facts. Thus, besides being an instrument for quality assessment, the NET analysis’ capability to identify putative regulatory sites, to unravel interrelations between different parts of metabolisms and to resolve metabolic functionalities based on compartmentalization underpins its power for uncovering system properties from quantitative metabolome data. We envision that NET analysis will significantly assist systems biology research and also will support more applied fields such as metabolic engineering.