Integration of metabolome data with metabolic networks reveals reporter reactions

Tunahan Çakir^1,a, Kiran Raosaheb Patil^2,a, Zeynep Ilsen Önsan¹, Kutlu Özergin Ülgen¹, Betül Kirdar¹ & Jens Nielsen²

^aThese authors contributed equally to this work

Article highlights

Synopsis

Cellular metabolism, as reflected in the metabolite levels and fluxes, is an integrated result of mass balance constraints and regulation at several different levels. Consequently, analysis of cellular metabolite levels, generally referred to as metabolomics, is an important step in the post-genomic era towards understanding of the biological logic behind large-scale organization and operation of cellular metabolism. Although it is now possible to measure quantitatively many intracellular metabolites, interpreting such data is a difficult task owing to the high connectivity in the metabolic network and inherent interdependency between enzymatic regulation, metabolite levels and fluxes. Here, we present a hypothesis-driven algorithm (Figure 1) for the integration of metabolome data with topology of genome-scale metabolic models and thereby identify the reactions (reporter reactions) significantly responding to the environmental/genetic perturbations through changes in metabolite levels. The algorithm is analogous to the algorithm developed by us earlier for identification of reporter metabolites using transcriptome data (Patil and Nielsen, 2005). For demonstration of the algorithm, we use two recently collected metabolome data sets for the yeast Saccharomyces cerevisiae, corresponding to an environmental and a genetic perturbation (Villas-Bôas et al, 2005a; Devantier et al, 2005a), to illustrate the applicability of the algorithm.

Figure 1

Reporter reaction algorithm to identify differential reaction significance by integrating metabolome data with metabolic networks. Quantitative metabolome data obtained from perturbation experiments are interpreted in terms of significance of change, and mapped onto the stoichiometric network, which is represented as bi-partite undirected graph, to identify reporter reactions.

Full figure and legend (110K)Figures & Tables index

Analytical methods available to date for metabolome measurements cover only a small fraction of the metabolites present in genome-scale metabolic models. Consequently, lack of quantitative data for several metabolites presents a major hurdle in integration of metabolome data and network topology. We therefore apply a pathway analysis-based preprocessing of the genome-scale yeast model to derive a reduced model with increased fraction of measured metabolites. In this way, the yeast genome-scale model including three compartments (mitochondria, cytosol and extracellular space) with 844 metabolites and 1175 reactions was reduced to a two-compartment model (intracellular and extracellular space) with 178 metabolites participating in 139 reactions, which corresponded to more than 47% of the quantitative metabolome data used in this study (84 metabolites). The first data set (Villas-Bôas et al, 2005a) allowed the examination of the effect of a perturbation related to an altered redox metabolism resulting from a gene deletion and aerobic/anaerobic growth, whereas the second data set (Devantier et al, 2005a) was used for studying the effect of very high-gravity fermentation media on metabolic phenotype.

Significance of change for each of the measured metabolites was quantified as P-values calculated by using the U-test. To address the problems arising owing to unavailability of data for over 50% of the metabolites (after preprocessing), P-values estimated from uncharacterized peaks in GC–MS spectra were randomly assigned to the 94 metabolites that remained unmeasured in the reduced metabolic model. These P-values were then converted to Z-scores, which will be normally distributed for a random data set. Each reaction in the model was then scored by using the Z-scores of its neighboring metabolites: To account for the random assignment of scores to unmeasured metabolites, calculations were repeated 1000 times and the resultant scores were averaged. Thus, the final scores represent the significance of reactions partially independent of the true levels of the unquantified metabolites. Top scoring reactions are hereby termed reporter reactions.

As metabolite levels are governed by changes in fluxes and enzyme activities, reporter reactions indicate the significance of how those reactions respond to the perturbation under study. Reporter scores of reactions participating in selected pathway structures across the analyzed perturbations are consistent with the previously reported findings and/or the expectations based on the type of the perturbation. The here reported algorithm thus enabled identification of key reactions in the yeast metabolism affected by genetic and environmental perturbations. Reporter reaction analysis is an attempt to infer the differential reaction significance based on metabolite measurements, and hence provides a basis for understanding the underlying cellular processes responding to the perturbations.

We also show that our method, in combination with transcriptome data (Devantier et al, 2005b), may provide information on whether a given reaction is likely to be regulated at the metabolic level or at the hierarchical level (ter Kuile and Westerhoff, 2001). The Z-score of a reaction calculated by our approach can be treated as an indicator of metabolic regulation, whereas the degree of hierarchical regulation of reactions can be approximated by the Z-scores calculated based on the changes in gene expression levels. By comparing the Z-scores emerging from different omics approaches, in this case metabolomics and transcriptomics, the underlying reasoning for regulation of reactions included in our reduced model could be hypothesized. For the 121 reactions in the model having corresponding genes associated with them, the analysis allowed determination of the reactions with potential regulation at metabolic, hierarchical or at both levels. Our results indicate that although there are many metabolically regulated reactions in the network, regulation is predominantly hierarchical.

This study can be regarded as one of the first steps towards the integration of different types of omics data by using metabolic networks as a scaffold in order to understand the architecture of metabolic regulatory circuits. Furthermore, our model-driven analysis forms a platform for the integration of other types of omics data, such as proteomics, and hence allow genome-scale identification of regulation in the metabolism.

Figure 1. Reporter reaction algorithm to identify differential reaction significance by integrating metabolome data with metabolic networks. Quantitative metabolome data obtained from perturbation experiments is interpreted in terms of significance of change, and mapped onto the stoichiometric network, which is represented as bi-partite undirected graph, to identify reporter reactions.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.