Metabolic pathways variability and sequence/networks comparisons
© Tun et al; licensee BioMed Central Ltd. 2006
Received: 07 July 2005
Accepted: 18 January 2006
Published: 18 January 2006
In this work a simple method for the computation of relative similarities between homologous metabolic network modules is presented. The method is similar to classical sequence alignment and allows for the generation of phenotypic trees amenable to be compared with correspondent sequence based trees. The procedure can be applied to both single metabolic modules and whole metabolic network data without the need of any specific assumption.
We demonstrate both the ability of the proposed method to build reliable biological classification of a set of microrganisms and the strong correlation between the metabolic network wiringand involved enzymes sequence space.
The method represents a valuable tool for the investigation of genotype/phenotype correlationsallowing for a direct comparison of different species as for their metabolic machinery. In addition the detection of enzymes whose sequence space is maximally correlated with the metabolicnetwork space gives an indication of the most crucial (on an evolutionary viewpoint) steps of the metabolic process.
The concept of network in the sense of a set of mutually interacting elements whose collectivebehaviour gives rise to emergent properties is a very fruitful metaphor . Thus we can readabout gene networks , protein interaction networks , metabolic networks , ecological networks  or even protein folding networks .
A lot of both theoretical and experimental works [21, 4, 24, 14], were devoted to network analysis in trying to identify network invariants important for predicting collective behaviour. We concentrate on the between organisms variability in metabolic networks wiring patterns: we developed a simple method to estimate the similarities among different organisms metabolic networks in a way formally identical to biopolymer sequence comparisons. The topic is not new, many groups developed quantitative methods to compare the topological similarities of metabolic networks [6, 10, 25], the originality of our proposal relies on the simplicity and amplitude of application range of the method together with the definition of a general strategyfor comparing the network and sequence metrics. We will show how the relation between network and sequence similarity spaces is a potentially fruitful new avenue for evolution studies, providing a system level phenotypic character (network wiring) amenable to quantitative phylogenetic analysis and easily comparable to phylogenetic structures generated by molecular level approaches like biopolymer sequences. The correlation between network and sequence spaces can give useful hints about the relative importance of the involved enzymes for the entire pathway functioning.
Results and discussion
Component loadings profile relative to the metabolism dissimilarity matrix. Bolded values refer to elements significantly loaded on the relative component. The component loadings correspond to the correlation coefficients between original variables and the extracted components. The loadings corresponding to the variables most important for the component interpretation are bolded.
% of explained variance
The minor components account for relatively secondary features of the dissimilarity structure of the data set, with the second component mainly linked to the Streptococcus species specificity and the third component describing the Bacillus genus singular behaviour.
By superimposing the different metabolic networks, we observed that the main difference between the networks posited at the two opposite first component ends, is the presence (absence) of the lateral branching of the Glycolisis/Gluconeogenesis module deputed to the utilization of Arbutin and Salicin as substrates for Glycolisis (ellipse of Fig. 2). This difference in the use of substrates for energy production is the main physiological determinant of the metabolic network wiring differences in our data set. The 25 analysed organisms share 11 common enzymes of the Glycolisis/Gluconeogenesis pathway (bolded in Fig. 2) thus, each of these enzymes allows for a specific sequence space to be constructed and compared with the metabolic network space. It is well known that the correlation of two protein distance spaces is a marker of some form of interaction between the two molecules , and this feature is routinely adopted for inferring protein-protein interactions . The need to support a viable interaction between two proteins imposes a mutual constraint (resulting into a covariation) to the random mutation drift of the two systems. This covariation results into correlated dissimilarity matrices (trees) relative to the two proteins as for a suitable set of organisms .
Component loadings profile relative to the sequence space correlation of the 11 common enzymes of the Glycolisis/Gluconeogenesis pathway. Bolded values refer to variables (Major principal component for each single sequence space) significantly loaded on the relative component. The position of the enzymes in the module can be checked in Fig.2.
% of explained variance
Metabolic networks wiring can be considered as a 'complex' phenotype, crucially related to both organism evolution history and ecological niche. The strong correlation of protein sequence and wiring topology based phylogenetic trees points to the possibility to investigate the pleiotropic effects of mutations on a quantitative evolutionary bases.
It is worth noting that when we tried and correlate the metabolic dissimilarity matrices with the different number of genes codifying, in each species, for the different involved enzymes, we did not find any significant correlation. This is in line with the finding of Papp and colleagues  indicating multiplicity of isozymes to be essentially linked more to flux regulation than to basic wiring topology. Due to absence of reliable kinetic data, we adopted a purely topological view of the metabolic network, nevertheless the method could in principle be applied to kinetic data by substituting the Hamming metrics (see Methods) based on the absence/presence of a given reaction with the Euclidean metrics based on the values of a kinetic costant (or any other convenient weight coefficient attached to the different arcs).
The attaining of a strong species specificity (much higher than the one attained by nucleic acid comparisons) and the stability of classifications, indicates the pure topological wiring allows for a meaningful picture of the studied organisms.
The striking similarity between phosphopyruvate hydratase sequence space and the general metabolic network space is consistent with the recently discovered relevance of the so called non-hub-connectors  as well as with our results in yeast pointing to a preferred periphery location of essential enzymes .
There are in literature some other network comparisons methods: one is PathBlast that has been used on protein interaction network with success . However, PathBlast does not deal with network divergence in sufficient detail and is not explicitly based on metabolic networks. Moreover it is definitively more complex in terms of computation. The seminal work on network quantitative comparison is, to our knowledge, the Dandekar and colleagues one . The method proposed by the authors is based on the decomposition of the analysed pathway into its elementary modes. This is not a pre-requisite of our method that, on the contrary can be immediately applied on the adjacency matrix without further complications. This allows for the analysis of the whole set of metabolic reactions without being limited to specific modules. The Zhu and Qin  approach is not based on pathway alignment: the authors derive some general numerical descriptors of the network wiring topology (number of edges, clustering coefficient and so forth) allowing to associate to each network a numerical vector having as components the above mentioned descriptors. The between networks comparisons are thus based on the comparisons of the relative profiles in terms of the descriptors: this is particularly convenient for mechanistic studies but does not allow for an accurate phylogenetic analysis. The Hong et al.  approach is the most similar to ours among all the currently adopted methods of pathway alignment: it allows for the comparison of very large networks, but again it is based on the segmentation of a pathway into the constituent sub-pathways, while this intermediate step is not required by our methodology.
The most important feature of our method is the possibility to put on quantitative basis, thanks to the filtering action of principal component analysis, the relation between the shape of metabolic networks and the involved enzymes sequence space, thus finding the 'crucial' elements of the network. A last methodological remark is linked to the robustness of the attained classifications when in presence of 'noise' due to errors in wiring pattern: we demonstrated (data not shown) that the classification tree relative to the entire metabolome remains stable until 5% of 'errors' in the form of elimination/addition of arcs to the original network.
A lot of applications can be imagined for comparative network studies, going from the correlation between metabolic network shapes and the pattern of sensitivity to specific antibiotics to large scale environmental studies of ecological communities so to correlate metabolic similarities to food-web structures.
A metabolic network can be represented as a binary square adjacency matrix having as rows and columns the intervening metabolites and taking 1/0 values depending on the presence/absence of an arc between the corresponding elements. In the case of metabolic networks, an arc corresponds to the presence of one (or more) enzyme catalysing a chemical reaction transforming a metabolite into another. The irreversibility of some reactions make the adjacency matrix not symmetrical, Figure 1 reports the adopted formalization stressing the relation holding between the di-graph and matrix notations.
The adjacency matrix formalization was very useful in describing a lot of network structures, particularly rich is the literature adopting this notation for organic molecules, in this case an arc represents a chemical bond between two atoms and the matrix is isomorphic to the structural formula of the molecule [15, 2].
The adjacency matrix allows for a straightforward metrics to compare different metabolic networks to be developed. The considered networks are relative to the same pathway (or module, like glycolisis, purine metabolism, aminoacid biosynthesis...) in different organisms thus giving rise to a specific adjacency matrix for each organism. All these matrices have the same set of rows and columns correspondent to the maximal coverage of the whole set of intervening metabolites (it is sufficient a given metabolite is present in a single network to be included). The distance between each pair of networks will be simply set to the Hamming distance between the two networks, i.e. to the number of discrepancies (1 vs. 0 or 0 vs. 1) scored in the corresponding elements of the two networks. The Hamming distance corresponds to the classical Nei distance proposed for evolutionary studies .
In order to make the metrics independent of the number of analysed variables (metabolites) we divide the sum of the discrepancies by the total number of variables (maximal attainable distance) and multiply the ratio by 100. Thus we obtain a 'percentage of dissimilarity' ranging from 0 (complete equivalence of the two networks) to 100.
This operation, when applied to a set of n different networks will end into a symmetric nXn dissimilarity matrix conveying all the information linked to the pairwise similarities between the correspondent organisms in terms of the metabolic module analysed.
Being the dissimilarity matrix fully quantitative, it can be analysed by means of the whole range of multidimensional statistical techniques (multidimensional scaling, principal component analysis...) as well as to be the basis for the construction of similarity trees.
In this work we analysed different sets of microrganisms as for different metabolic networks, moreover we computed the correlation between distance matrices based on metabolic network superposition with the corresponding distance matrices relative to the aminoacid sequences of some enzymes acting in the network. The Jukes-Cantor score on the maximum likelihood estimate of the number of substitutions between two sequences was adopted for protein data. All the metabolic networks were downloaded from KEGG database  using Cellware platform . A total of 25 organisms were randomly selected from five prokaryotic genera based on the availability of metabolic data (Figure 3a). The total data space consisted of 31,945 metabolites (average 1277.8/organism, range 350–1943), and 1901 pathways (average 76.04/organism, range 32–111).
We wish to thank to anonymous reviewers for driving our attention to the existing literature on network comparisons and to the other possible extensions of our method.
- Benigni R, Giuliani A: Quantitative modeling and biology: The multivariate approach. Am Journ Physiol 1994, 266: R1697-R1704.Google Scholar
- Benigni R, Passerini L, Pino A, Giuliani A: The Information Content of the Eigenvalues from Modified Adjacency Matrices: Large Scale and Small Scale Correlations. Quant Struct -Act Relat 1999, 15: 449–455.Google Scholar
- Bork P, Jensen LJ, Von Mering C, Ramani AK, Lee I, Marcotte EM: Protein interaction networks from yeast to human. Curr Opin Struct Biol 2004, 14: 292–299. 10.1016/j.sbi.2004.05.003View ArticlePubMedGoogle Scholar
- Christensen B, Nielsen J: Metabolic network analysis. A powerful tool in metabolic engineering. Adv Biochem Eng Biotechnol 2000, 66: 209–231.PubMedGoogle Scholar
- Dhar P, Meng TC, Somani S, Ye L, Sairam A, Chitre M, Hao Z, Sakharkar K: Cellware, a multi-algorithmic software for computational systems biology. Bioinformatics 2004, 20: 1319–1321. 10.1093/bioinformatics/bth067View ArticlePubMedGoogle Scholar
- Dandekar T, Schuster S, Huynen M, Bork P: Pathway alignment: application to the comparative analysis of glycolytic enzymes. Biochem J 1999, 343: 115–124. 10.1042/0264-6021:3430115PubMed CentralView ArticlePubMedGoogle Scholar
- Fiehn O, Weckwerth W: Deciphering metabolic networks. Eur J Biochem 2003, 270: 579–588. 10.1046/j.1432-1033.2003.03427.xView ArticlePubMedGoogle Scholar
- Goh CS, Cohen FE: Co-evolutionary analysis reveals insights into protein-protein interactions. J Mol Biol 2002, 324: 177–192. 10.1016/S0022-2836(02)01038-0View ArticlePubMedGoogle Scholar
- Guimera R, Nunes Amaral LA: Functional cartography of complex metabolic networks. Nature 2005, 433: 895–900. 10.1038/nature03288PubMed CentralView ArticlePubMedGoogle Scholar
- Hong SH, Kim TY, Lee SY: Phylogenetic analysis based on genome-scale metabolic pathway reaction content. Appl Microbiol Biotechnol 2004, 65: 203–210. 10.1007/s00253-004-1641-3PubMedGoogle Scholar
- Kanehisa M, Goto S: KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Research 2000, 28: 27–30. 10.1093/nar/28.1.27PubMed CentralView ArticlePubMedGoogle Scholar
- Lässig M, Bastolla U, Manrubia SC, Valleriani A: Shape of Ecological Networks. Phys Rev Lett 2001, 86: 4418–4421. 10.1103/PhysRevLett.86.4418View ArticlePubMedGoogle Scholar
- Legrain P, Wojcik J, Gauthier J-M: Protein – protein interaction maps: a lead towards cellular functions. Trends Genet 2001, 17: 346–352. 10.1016/S0168-9525(01)02323-XView ArticlePubMedGoogle Scholar
- Light S, Kraulis P: Network analysis of metabolic enzyme evolution in Escherichia coli. BMC Bioinformatics 2004, 5: 15–20. 10.1186/1471-2105-5-15PubMed CentralView ArticlePubMedGoogle Scholar
- Lukovits I: A compact form of the adjacency matrix. J Chem Inf Comput Sci 2000, 40: 1147–1150. 10.1021/ci990152hView ArticlePubMedGoogle Scholar
- Nei M: Molecular Evolutionary Genetics. New York: Columbia University press; 1987.Google Scholar
- Oltvai ZN, Barabasi AL: Systems biology. Life's complexity pyramid. Science 2002, 298: 763–764. 10.1126/science.1078563View ArticlePubMedGoogle Scholar
- Palumbo MC, Colosimo A, Giuliani A, Farina L: Functional essentiality from topology features in metabolic networks: A case study in yeast. FEBS Letters 2005, 579: 4642–4646. 10.1016/j.febslet.2005.07.033View ArticlePubMedGoogle Scholar
- Papp B, Pal C, Hurst LD: Metabolic Network analysis of the causes and evolution of enzyme dispensability in yeast. Nature 2004, 429: 661–664. 10.1038/nature02636View ArticlePubMedGoogle Scholar
- Rao F, Caflisch A: The protein folding network. J Mol Biol 2004, 342: 299–306. 10.1016/j.jmb.2004.06.063View ArticlePubMedGoogle Scholar
- Savinell JM, Palsson BO: Optimal selection of metabolic fluxes for in vivo measurement. I. Development of mathematical methods. J Theor Biol 1992, 155: 201–214.View ArticlePubMedGoogle Scholar
- Sharan R, Suthram S, Kelley RM, Kuhn T, McCuine S, Uetz P, Sittler T, Karp RM, Ideker T: Conserved patterns of protein interaction in multiple species. Proc Natl Acad Sci USA 2005, 102: 1974–1979. 10.1073/pnas.0409522102PubMed CentralView ArticlePubMedGoogle Scholar
- Shen-Orr SS, Milo R, Mangan S, Alon U: Network motifs in the transcriptional regulation network of Escherichia coli. Nature Genet 2002, 31: 64–68. 10.1038/ng881View ArticlePubMedGoogle Scholar
- Wittmann C, Heinzle E: Genealogy profiling through strain improvement by using metabolic network analysis: metabolic flux genealogy of several generations of lysine-producing corynebacteria. Appl Environ Microbiol 2002, 68: 5843–5859. 10.1128/AEM.68.12.5843-5859.2002PubMed CentralView ArticlePubMedGoogle Scholar
- Zhu DX, Qin Z: Structural comparison of metabolic networks in selected single cell organisms. BMC Bioinformatics 2005, 6: 8. 10.1186/1471-2105-6-8PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.