Integration of omic networks in a developmental atlas of maize

Abstract

Coexpression networks and gene regulatory networks (GRNs) are emerging as important tools for predicting functional roles of individual genes at a system-wide scale. To enable network reconstructions, we built a large-scale gene expression atlas composed of 62,547 messenger RNAs (mRNAs), 17,862 nonmodified proteins, and 6227 phosphoproteins harboring 31,595 phosphorylation sites quantified across maize development. Networks in which nodes are genes connected on the basis of highly correlated expression patterns of mRNAs were very different from networks that were based on coexpression of proteins. Roughly 85% of highly interconnected hubs were not conserved in expression between RNA and protein networks. However, networks from either data type were enriched in similar ontological categories and were effective in predicting known regulatory relationships. Integration of mRNA, protein, and phosphoprotein data sets greatly improved the predictive power of GRNs.

Fig. 1. Comparison of transcriptome and proteome data sets

(A) FPKM distribution of mRNA abundance (red). FPKM values of transcripts corresponding to quantified proteins (blue), phosphopeptides (green), syntenic genes conserved between maize and sorghum (gray), and nonsyntenic genes (black) are shown. Data are the average expression from the 23 tissues profiled. (B) Percentage of quantified mRNA and proteins in the annotated filtered (high-confidence gene models) and working (all gene models) gene sets. (C) Breakdown of detected mRNA and proteins, based on annotations. (D) Percentages of all annotated genes that are transcribed and percentages of all transcribed genes that are translated, for both the syntenic and nonsyntenic gene sets.

Fig. 2. Coexpression network analyses

(A) Hypothetical undirected coexpression subnetwork showing conserved (solid lines) and nonconserved (dotted lines) coexpression edges between mRNA and protein networks. (B) Venn diagram depicting edge conservation (solid lines in Fig. 2A) between the two coexpression networks. (C) Number of edges a given gene (node) has in the protein (x axis) and mRNA (y axis) coexpression networks. Nodes above the 90th percentile for the number of edges are considered hubs and are colored according to whether they are hubs in the protein (blue) or mRNA (red) network or both (green). Black dots represent non-hub nodes.

Fig. 3. Categorical enrichment analysis of coexpression modules

Coexpression modules were determined by WGCNA and functionally annotated using MapMan categories. Categories enriched (Benjamini-Hochberg adjusted P value ≤ 0.05) in one or more modules are represented by vertical bars and labeled with the bin number and name. For each category, the genes accounting for the enrichment were extracted separately from mRNA and protein modules. Only functional categories with at least 20 genes are shown. Colored bars represent the proportion of genes in each enriched category that are specific to one network (mRNA, red; protein, blue) or shared between the networks (green).

Fig. 4. Unsupervised GRN analyses

(A) Hypothetical GRN subnetwork depicting a TF regulator (square) and potential target genes (circle) quantified as mRNA (red) or protein (blue). GRN-specific and -conserved predictions are depicted by dotted and solid lines, respectively. (B) Overlap of the true-positive predictions from the top 500 true GRN predictions for KN1 quantified as mRNA, protein, or phosphopeptide. True KN1 targets were identified by Bolduc et al. (24). (C) Overlap of the top 1 million TF target predictions between the GRNs reconstructed using TF abundance quantified at the mRNA, protein, or phosphopeptide level. (D) ROC curves and (E) precision-recall curves generated using known Kn1 and O2 target genes for a mRNA-only GRN (red) and a fully integrated GRN built by combining mRNA, protein, and phospho-protein data into a single GRN (blue).