Inference of transcriptional networks in Arabidopsis through conserved noncoding sequence analysis

Abstract

Transcriptional regulation plays an important role in establishing gene expression profiles during development or in response to (a)biotic stimuli. Transcription factor binding sites (TFBSs) are the functional elements that determine transcriptional activity, and the identification of individual TFBS in genome sequences is a major goal to inferring regulatory networks. We have developed a phylogenetic footprinting approach for the identification of conserved noncoding sequences (CNSs) across 12 dicot plants. Whereas both alignment and non-alignment-based techniques were applied to identify functional motifs in a multispecies context, our method accounts for incomplete motif conservation as well as high sequence divergence between related species. We identified 69,361 footprints associated with 17,895 genes. Through the integration of known TFBS obtained from the literature and experimental studies, we used the CNSs to compile a gene regulatory network in Arabidopsis thaliana containing 40,758 interactions, of which two-thirds act through binding events located in DNase I hypersensitive sites. This network shows significant enrichment toward in vivo targets of known regulators, and its overall quality was confirmed using five different biological validation metrics. Finally, through the integration of detailed expression and function information, we demonstrate how static CNSs can be converted into condition-dependent regulatory networks, offering opportunities for regulatory gene annotation.

Figure 1.

Overview of CNS Properties. (A) Length distribution of significantly conserved footprints. All footprints are grouped in bins of size 10 bp. (B) Overview of significantly conserved footprints in relation to the number of species in which the footprint was conserved. For all conservation scores, the relative percentage of significant footprints is shown (gray boxes) as well as a cumulative distribution (black line). (C) Breakdown of CNS over different genomic regions. [See online article for color version of this figure.]

Figure 2.

Recovery of AtProbe Elements and Comparison of CNSs from Different Phylogenetic Footprinting Studies. (A) Overview of the recovery of experimental AtProbe elements in four different CNS studies. Black boxes show the percentage of recovered elements, and white boxes shows the percentage of uniquely recovered elements. Diamonds depict fold enrichments, which are defined as the ratio of the observed overlap over the expected overlap by chance. (B) Genome-wide coverage of CNSs. Black boxes show the total number of nucleotides assigned to CNSs per study, while white boxes show the number of nucleotides in CNSs that are unique to a single study.

Figure 3.

Recovery of in Vivo Functional Targets Using CNS Information. White and black boxes show fold enrichments for CNSs and naïve motif mapping, respectively. White and black diamonds show the fraction of recovered elements for CNSs and a simple motif mapping approach, respectively.

Figure 4.

Evaluation of the Biological Relevance of the Predicted Network Using Different Biological Metrics Assessing Functional and Expression Coherence. GO annotations, MapMan annotations, and functional modules together with a stress and developmental expression compendium were used to evaluate the biological relevance of the predicted GRN. A comparison of fold enrichment is depicted between the predicted network (black bars) and the experimental network (white bars). All reported fold enrichments are significant (P value < 0.05). Numbers in parentheses report the number of regulatory interactions in the two networks and the number of genes having functional or expression information, respectively.

Figure 5.

A Condition-Specific Secondary Cell Wall Gene Regulatory Network. Nodes and edges depict genes and regulatory interactions, while condition-specific seed, flower, and hormone coexpression edges are shown using orange, green, and blue lines, respectively. Experimentally confirmed interactions are shown using an arrow line. Red diamonds are the source TFs, gray diamonds are target genes that are TFs, and rounded rectangles are other target genes. Target genes with a gray border are known to be involved in secondary cell wall biosynthesis based on GO.