Systematic identification of functional plant modules through the integration of complementary data sources

Abstract

A major challenge is to unravel how genes interact and are regulated to exert specific biological functions. The integration of genome-wide functional genomics data, followed by the construction of gene networks, provides a powerful approach to identify functional gene modules. Large-scale expression data, functional gene annotations, experimental protein-protein interactions, and transcription factor-target interactions were integrated to delineate modules in Arabidopsis (Arabidopsis thaliana). The different experimental input data sets showed little overlap, demonstrating the advantage of combining multiple data types to study gene function and regulation. In the set of 1,563 modules covering 13,142 genes, most modules displayed strong coexpression, but functional and cis-regulatory coherence was less prevalent. Highly connected hub genes showed a significant enrichment toward embryo lethality and evidence for cross talk between different biological processes. Comparative analysis revealed that 58% of the modules showed conserved coexpression across multiple plants. Using module-based functional predictions, 5,562 genes were annotated, and an evaluation experiment disclosed that, based on 197 recently experimentally characterized genes, 38.1% of these functions could be inferred through the module context. Examples of confirmed genes of unknown function related to cell wall biogenesis, xylem and phloem pattern formation, cell cycle, hormone stimulus, and circadian rhythm highlight the potential to identify new gene functions. The module-based predictions offer new biological hypotheses for functionally unknown genes in Arabidopsis (1,701 genes) and six other plant species (43,621 genes). Furthermore, the inferred modules provide new insights into the conservation of coexpression and coregulation as well as a starting point for comparative functional annotation.

Figure 1.

Delineation of functional gene modules. A, Four different primary data sets were processed to extract functional gene modules, resulting in 1,563 nonredundant modules. Data types are in roman font, and methods are in italic font. B, Biological properties (functional coherence, EC, and cis-regulatory coherence) of gene modules were characterized. Dotted lines indicate gene-GO associations and nonsignificant PCCs for the functional coherence and the EC panels, respectively. In the cis-regulatory coherence panel, the blue triangle represents an enriched motif.

Figure 2.

Basic properties of the derived functional gene modules. A, Number of different module types per gene. B, Number of different input data types per module edge. C, Overlap between the different types of modules. D, Gene size distribution for the set of 1,563 nonredundant gene modules.

Figure 3.

Functional, expression, and cis-regulatory coherence. A, Comparison of EC scores between the modules from different input data types. The EC scores are shown for both the general compendium (dotted lines) and the compendium showing the maximum EC (solid lines). The vertical dotted line indicates the threshold for significant EC. B, GO-BP and motif enrichment statistics for the modules delineated using the different input data types.

Figure 4.

Overview of GO-BP slim biological processes in which modules were predicted to be involved. Modules with multiple GO-BP annotations can be present in different GO slim categories.

Figure 5.

Regulatory complexity of genes in modules. A, Number of modules in which a gene is present. Asterisks denote values higher than zero. B, Number of motifs per module. C, Number of motifs per gene promoter. D, Regulatory complexity, defined as a combination of the number of modules in which a gene is present and the number of motifs in its promoter. All 13,142 genes are included, and the number of genes at each coordinate is given as a colored size scale. The gray circle indicates the average regulatory complexity for all 13,142 genes. The dotted line is the function f(x) = x.

Figure 6.

Example of a delineated module with true-positive genes. A, Cell wall biogenesis. B, Xylem and phloem pattern formation. C, DNA endoreduplication. D, Cell cycle regulation. Edges with EC conservation in less than three species are hidden. E, Response to GA. Modules can be explored in detail using the additional data Web site.