Transcriptional coordination of the metabolic network in Arabidopsis

Abstract

Patterns of coexpression can reveal networks of functionally related genes and provide deeper understanding of processes requiring multiple gene products. We performed an analysis of coexpression networks for 1,330 genes from the AraCyc database of metabolic pathways in Arabidopsis (Arabidopsis thaliana). We found that genes associated with the same metabolic pathway are, on average, more highly coexpressed than genes from different pathways. Positively coexpressed genes within the same pathway tend to cluster close together in the pathway structure, while negatively correlated genes typically occupy more distant positions. The distribution of coexpression links per gene is highly skewed, with a small but significant number of genes having numerous coexpression partners but most having fewer than 10. Genes with multiple connections (hubs) tend to be single-copy genes, while genes with multiple paralogs are coexpressed with fewer genes, on average, than single-copy genes, suggesting that the network expands through gene duplication, followed by weakening of coexpression links involving duplicate nodes. Using a network-analysis algorithm based on coexpression with multiple pathway members (pathway-level coexpression), we identified and prioritized novel candidate pathway members, regulators, and cross pathway transcriptional control points for over 140 metabolic pathways. To facilitate exploration and analysis of the results, we provide a Web site (http://www.transvar.org/at_coexpress/analysis/web) listing analyzed pathways with links to regression and pathway-level coexpression results. These methods and results will aid in the prioritization of candidates for genetic analysis of metabolism in plants and contribute to the improvement of functional annotation of the Arabidopsis genome.

Figure 1.

Coexpression p and r² values for genes from the AraCyc database of metabolic pathways. A, Logarithm (base 10) of regression p values plotted against corresponding r² values obtained from regressing expression values for 1,330 AraCyc genes against all genes on the ATH1 expression microarray. B to D, Example plots showing positive (D) and negative (B) linear relationships between normalized expression values (log base 2) for weakly (C) and strongly (B and D) coexpressed genes.

Figure 2.

Relative proportions of high-confidence coexpression links. Each column reports the relative numbers of coexpression pairs with p values indicated on the x axis, where the paired genes are both in the same AraCyc pathway (I), both are in different AraCyc pathways (N), or only one is in an AraCyc pathway (O).

Figure 3.

Coexpression network analysis. A, Schematic showing PLC analysis for identifying functionally relevant candidate genes outside a functional grouping (e.g. a metabolic pathway) using their connections to genes within the group. Bait genes that are members of the same functional group (rectangular area) are connected via coexpression relationships (dotted lines) to genes not currently annotated as belonging to the group. Some of the genes outside the group are connected to more than one group member and are selected as candidates for further analysis. B, Numbers of coexpressed genes selected under different coexpression p value cutoffs (ranging from 1E-40 to 1E-120) or requiring increasingly large numbers of coexpression partners within a pathway (2–15). No gene pair is counted more than once per column.

Figure 4.

Transcriptional organization of metabolic pathway genes. A, Connectivity in the glycolysis pathway based on coexpression p value threshold 10⁻⁸⁰. Lines connect groups of coexpressed genes. Only genes with nonpromiscuous probe sets are shown. B, Distribution of connections for pathways with six or more pathway steps and at least two coexpressed pathway genes. The x axis shows distance in pathway steps. Genes catalyzing the same step are counted as zero pathway steps, genes catalyzing adjacent steps are counted as one pathway step, and so on. The y axis gives the number of coexpression links for positive (dark) and negative (lighter) coexpression relationships. Pathways included in the analysis are listed in Supplemental Table S2.

Figure 5.

Coexpression links per gene. The distributions for positive (A) and negative (B) coexpression links (p < 10⁻⁸⁰) per AraCyc pathway genes are shown. C, Link frequency distribution for positive (+) and negative (o) coexpression networks, p value < 10⁻⁸⁰. The y axis indicates the number of genes that are coexpressed with the number of genes shown on the x axis.

Figure 6.

Within- and across-pathway coexpression patterns. Coexpression relationships are viewed as a grayscale heatmap between a subset of pathways listed in Table I. Cells are shaded according to the negative logarithm (base 10) of coexpression p values between genes from corresponding rows and columns. Coexpression p values less than or equal to 10⁻¹⁰⁰ are shown using the darkest shade. When rows and columns represent the same gene but intersect off the diagonal (due to shared genes across pathways), the corresponding cells are colored using the lightest shade. A version labeled with gene identifiers and pathway names is available as Supplemental Figure S1. Pathway designations and annotations are from AraCyc 2.1.

Figure 7.

Patterns of coexpression with photosynthesis pathway genes. Genes that are coexpressed (p value < 1E-60) with each gene in the photosynthesis light reaction pathway are shown.

Figure 8.

Metabolic network connectivity. A, Network of positively coexpressed metabolic pathways using a p value cutoff of 1E-200. B, Network of negatively coexpressed metabolic pathways using a cutoff of 1E-80. Connected pathways share at least seven pairs of coexpressed genes.