Differential metabolic and coexpression networks of plant metabolismby Nooshin Omranian, Sabrina Kleessen, Takayuki Tohge, Sebastian Klie, Georg Basler, Bernd Mueller-Roeber, Alisdair R. Fernie, Zoran Nikoloski

Trends in Plant Science


Plant Science


LXVIII. The Abbot and Convent of Woburn to the King

Thabbot and convent of Woburn

Regulatory approaches to the control of environmental mutagens and carcinogens

Members and Consultant of Committee

Modelling Metabolic Networks—The Theories of Metabolism

Jean-Pierre Mazat, Bertrand Beauvoit

Systems Analysis of Plant Functional, Transcriptional, Physical Interaction, and Metabolic Networks

G. W. Bassel, A. Gaudinier, S. M. Brady, L. Hennig, S. Y. Rhee, I. De Smet

Systematic Comparison of C3 and C4 Plants Based on Metabolic Network Analysis

Chuanli Wang, Longyun Guo, Yixue Li, Zhuo Wang


Differential metabolic and co

T li nb


Techniques & Applications

TRPLSC-1260; No. of Pages 3be physically clustered in Arabidopsis thaliana, soybean (Glycine max), and sorghum (Sorghum bicolor) and observed that specialized and primary metabolism genes of

Arabidopsis differ in their coexpression patterns. Using state-of-the-art approaches with the comprehensive resources provided in Chae et al., here we aim to demonstrate that strong caution in interpreting biological findings from network-based comparative analyses is needed, since the results may largely depend on the network representations and the computational approaches employed. also affect the path-based network properties (e.g., betweenness centrality [9]).

However, the stoichiometric matrix accurately captures the weighted directed hypergraph structure of metabolic networks [10], as illustrated in Figures 1A,B, without removal of any currency metabolites. Direct comparison between the 16 plant metabolic networks can be readily conducted by employing singular value decomposition of the corresponding stoichiometric matrices without removal of the key currency metabolites [11]. To this end, we relied on the distribution of scaled singular values determined from the singular value decomposition. We used the Kolmogorov–Smirnov test statistic on the resulting distributions of scaled singular values to obtain a distance matrix and to test the differences between the distributions from every pair of plant metabolic networks (see the supplementary material online). The distance matrix indicates that there are 1360-1385/  2015 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/ j.tplants.2015.02.002

Corresponding author: Nikoloski, Z. (nikoloski@mpimp-golm.mpg.de).

Keywords: plant specialized metabolism; metabolic networks; gene coexpression; differential network analysis.plant metabolism

Nooshin Omranian1,2, Sabrina Kleessen3,

Georg Basler4, Bernd Mueller-Roeber1,2, A 1Max Planck Institute of Molecular Plant Physiology, Am Mu¨hle 2 Institute of Biochemistry and Biology, University of Potsdam, 3Targenomix GmbH, Am Mu¨hlenberg 11, Potsdam, Germany 4Estacio´ n Experimental del Zaidı´n CSIC, 18008 Granada, Spain

Recent analyses have demonstrated that plant metabolic networks do not differ in their structural properties and that genes involved in basic metabolic processes show smaller coexpression than genes involved in specialized metabolism. By contrast, our analysis reveals differences in the structure of plant metabolic networks and patterns of coexpression for genes in (non)specialized metabolism. Here we caution that conclusions concerning the organization of plant metabolism based on network-driven analyses strongly depend on the computational approaches used.

Plant metabolic network structure and gene coexpression

In a recent computational study, Chae et al. [1] collected experimentally validated and predicted enzymatic functions from 16 plants and reconstructed reaction-centric networks. The considered enzymes included those that participate in pathways for the synthesis of building blocks necessary for survival, termed primary metabolism, as well as those that are involved in the production of specific metabolites facilitating various other tasks, termed secondary metabolism. Although these networks differ in the included reactions, many of the salient network properties were found to be similar across the investigated species.

Chae et al. also determined that genes involved in some of the specialized metabolism pathways were more likely toexpression networks of akayuki Tohge1, Sebastian Klie3, sdair R. Fernie1, and Zoran Nikoloski1 erg 1, Potsdam, Germany sdam, Germany

Accurate metabolic representations reveal structural differences due to specialized metabolism

Metabolism encompasses the entirety of largely enzymecatalyzed reactions transforming the set of nutrients into molecules that support various functions. Metabolic reactions do not operate in isolation and comprise functional, connected networks capable of bearing flux. Accurate metabolic modeling relies on mass/charge-balanced biochemical reactions and considers their thermodynamic properties [2,3]. In plants, metabolic reactions are coordinated across organelles, with some duplication of metabolic processes or reactions (e.g., from glycolysis and photorespiration [4–6]). The considered reactions usually differ between cell types and environments [4,7]. The reconstructions of genome-scale metabolic networks presented by Chae et al. do not consider these important aspects of large-scale metabolic modeling.

The functionality of a metabolic network is partially ensured by the connectedness of the reactions. However, in the reaction-centric network reconstructions of Chae et al. (Figure 1C), after the removal of arbitrarily chosen, socalled ‘currency’ metabolites, 11.4–17.5% of the reactions are not in the largest component [8], suggesting that they may operate orthogonally to the rest of the network. This number decreases to 0.68–1.53% if the metabolite–reaction network representation is used (Figure 1D). Therefore, removal of the currency metabolites has a dramatic effect on the number and size of connected components and canTrends in Plant Science xx (2015) 1–3 1 3ree i (dire 3b) re

TRPLSC-1260; No. of Pages 3statistically significant structural differences between the considered networks (Figure 2A). Interestingly, the resulting matrix is similar to that obtained from the

Jaccard distance of the reaction node sets (Figure 2B), supported by the RV coefficient of 0.96 (P = 3.456e 5, permutation test; see the supplementary material online). In contrast to Chae et al., our findings were determined without the need for subjective grouping of reactions and, more importantly, without the removal of any metabolites. Since 730 reactions are shared across the 16 species, the differences are due to the speciesspecific reactions predominantly participating in specialized metabolism and involving functionally redundant, multifunctional, and functionally diverse enzymatic gene families (e.g., phenylpropanoid metabolism [12]).