Unraveling transcriptional control in Arabidopsis using cis-regulatory elements and coexpression networks

Abstract

Analysis of gene expression data generated by high-throughput microarray transcript profiling experiments has demonstrated that genes with an overall similar expression pattern are often enriched for similar functions. This guilt-by-association principle can be applied to define modular gene programs, identify cis-regulatory elements, or predict gene functions for unknown genes based on their coexpression neighborhood. We evaluated the potential to use Gene Ontology (GO) enrichment of a gene's coexpression neighborhood as a tool to predict its function but found overall low sensitivity scores (13%-34%). This indicates that for many functional categories, coexpression alone performs poorly to infer known biological gene functions. However, integration of cis-regulatory elements shows that 46% of the gene coexpression neighborhoods are enriched for one or more motifs, providing a valuable complementary source to functionally annotate genes. Through the integration of coexpression data, GO annotations, and a set of known cis-regulatory elements combined with a novel set of evolutionarily conserved plant motifs, we could link many genes and motifs to specific biological functions. Application of our coexpression framework extended with cis-regulatory element analysis on transcriptome data from the cell cycle-related transcription factor OBP1 yielded several coexpressed modules associated with specific cis-regulatory elements. Moreover, our analysis strongly suggests a feed-forward regulatory interaction between OBP1 and the E2F pathway. The ATCOECIS resource (http://bioinformatics.psb.ugent.be/ATCOECIS/) makes it possible to query coexpression data and GO and cis-regulatory element annotations and to submit user-defined gene sets for motif analysis, providing an access point to unravel the regulatory code underlying transcriptional control in Arabidopsis (Arabidopsis thaliana).

Figure 1.

EC scores for genes functionally annotated using GO (A) and AraCyc (B). BP, MF, and CC refer to Biological Process, Molecular Function, and Cellular Component categories, and the number of categories is indicated in parentheses. [See online article for color version of this figure.]

Figure 2.

Functional enrichment of GO and cis-regulatory element annotation for guide gene cluster AT5G59220. Lines indicate coexpression relationships, and colored circles show the functional annotation for the individual genes. Enrichment analysis is performed using the hypergeometric distribution and Bonferroni correction for multiple hypotheses testing. Results are shown for the ATH95 benchmark coexpression network. [See online article for color version of this figure.]

Figure 3.

Functional predictive power for three benchmark coexpression networks built using different expression similarity thresholds (ATH90, ATH95, and ATH99). A to C show cumulative SN scores for a subset of GO categories, and D shows overall cumulative SN scores.

Figure 4.

Examples of cis-regulatory motifs showing significant associations with one or more GO categories. GO motif networks reveal for different GO categories the fraction of genes having the motif in their promoter (P value < 0.05 using the hypergeometric distribution). The line thickness reflects the motif coverage per GO category and varies from 6% to 70%. Known motifs from AGRIS or PLACE are indicated in italics.