Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Sep 12;114(37):E7841-E7850.
doi: 10.1073/pnas.1707375114. Epub 2017 Aug 29.

Exploring Regulation in Tissues With eQTL Networks

Affiliations
Free PMC article

Exploring Regulation in Tissues With eQTL Networks

Maud Fagny et al. Proc Natl Acad Sci U S A. .
Free PMC article

Abstract

Characterizing the collective regulatory impact of genetic variants on complex phenotypes is a major challenge in developing a genotype to phenotype map. Using expression quantitative trait locus (eQTL) analyses, we constructed bipartite networks in which edges represent significant associations between genetic variants and gene expression levels and found that the network structure informs regulatory function. We show, in 13 tissues, that these eQTL networks are organized into dense, highly modular communities grouping genes often involved in coherent biological processes. We find communities representing shared processes across tissues, as well as communities associated with tissue-specific processes that coalesce around variants in tissue-specific active chromatin regions. Node centrality is also highly informative, with the global and community hubs differing in regulatory potential and likelihood of being disease associated.

Keywords: GTEx; GWAS; bipartite networks; eQTL; expression quantitative trait locus.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Fig. 1.
Fig. 1.
Structure of eQTL networks. (A) The eQTL network from heart left ventricle. Each circle represents a community. The nodes, both SNPs and genes, are located around each circle. Gray lines represent network edges (significant cis- and trans-eQTL associations). (B) Modularity of the eQTL network from each of the 13 tissues. Modularity assesses the strength of division of the network in communities and corresponds to the fraction of edges observed within each community minus the expected fraction if edges were randomly distributed. B, Right shows the number of communities (Com) in each network. (C) Structure of communities within the eQTL heart left ventricle network. The heart left ventricle network is represented as a matrix with SNPs in columns and genes in rows. Each network edge is represented by a point. Intracommunity edges are plotted in blue and intercommunity edges in black. Community structure for the other 12 networks is presented in SI Appendix, Fig. S4.
Fig. 2.
Fig. 2.
Some network communities are enriched for biological functions shared across tissues. A complete list of the significantly overrepresented biological processes in each community and each network can be found in Dataset S3. (A) Heatmap clustering the similarity of GO biological processes in communities from all tissues. Only GO terms that were significant in at least 12 tissues are included. (B) Sankey diagram linking clusters from the heatmap to the tissues that contain at least one community enriched for genes involved in the clustered functions. A ribbon’s thickness is proportional to the number of communities enriched for each cluster of GO terms in each TS network.
Fig. 3.
Fig. 3.
Close-up of a community enriched for TS GO biological processes: the heart ventricle community 86. (A) Circos diagram for heart left ventricle eQTL in community 86. From the outside to the inside: chromosomes (in black), genes from community 86 (in blue, dark blue bars represent genes associated with cellular metabolism), SNPs (in green, with dark green bars indicating an association with metabolic traits), and SNP–gene assocations (in gray: all associations, in red: eQTL linked to genes involved in cellular respiration or SNPs involved in metabolism). Details about GWAS annotation of the SNPs and genes are given in Dataset S4. (B) Heart left ventricle community 86 is enriched for genes involved in cellular respiration.
Fig. 4.
Fig. 4.
Characteristics of TS communities. (A) Enrichment in TS genes, SNPs, and edges among communities with genes involved in TS GO biological processes compared with communities with genes involved in shared pathways. Barplots represent the ratio of mean proportion of unique elements in TS vs. shared communities. A, Right shows P values obtained using the Mann–Whitney U test and correcting for LD. (B) TS SNPs are more likely to be located in TS activated chromatin regions among TS communities compared with shared communities. P values were obtained using the Fisher test and correcting for LD. (A and B) ADS, adipose subcutaneous; ATA, aorta; ATT, artery tibial; EMC, esophagus mucosa; EMS, esophagus muscularis; FIB, fibroblast; HRV, heart left ventricle; LNG, lung; NS, nonsignificant; SKN, skin; SMU, skeletal muscle; TNV, tibial nerve; THY, thyroid; WBL, whole blood (WBL); *P<0.05, **P<0.01, ***P<0.001.
Fig. 5.
Fig. 5.
Network hub SNPs are more likely to lie in active chromatin regions than nonhub SNPs. (A and B) Symbols shows the odds ratios across all eight tissues with matching chromatin states from the Roadmap Epigenomics Project. Odds ratios are combined across tissues, using conditional logistic regression. Bars represent the 95% confidence interval. Each odds ratio measures the enrichment of central SNPs in a particular functional category, corrected for number of genes within 1 Mb of the SNP. (A) Enrichment in each chromatin state for SNPs with a core score in the top 25%. (B) Enrichment in each chromatin state for SNPs with network degree greater than 10. Odds ratios and P values for each TS network are listed in Dataset S5. Enrichment in each chromatin state for all trans-eQTL are presented in Dataset S1. (C) Example of a core SNP located in an active TSS region: rs4072037, associated with esophageal and gastric cancer. (D) Example of a high-degree SNP located in an active enhancer: rs2546765. The chromatin-state tracks are based on results from the Roadmap Epigenomics Project core 15-states model for esophagus mucosa (E079) for C and heart left ventricle (E095) for D.
Fig. 6.
Fig. 6.
Network properties of GWAS SNPs associated with autoimmune diseases in whole blood. (A) Ratio of observed vs. expected number of SNPs associated with autoimmune diseases by GWASs depending on their network degree. P values were obtained using 1,000 resamplings, taking into account gene density around each SNP. Ratios including all GWAS traits and diseases in each TS network are in SI Appendix, Fig. S9. (B) Distribution of core scores for SNPs associated (in blue) or not (in gray) to autoimmune diseases by GWAS. P values were obtained using a LRT and pruning for SNPs in LD. Distributions for all TS networks including all GWAS traits and diseases are in SI Appendix, Fig. S10.

Similar articles

See all similar articles

Cited by 28 articles

See all "Cited by" articles

Publication types

MeSH terms

LinkOut - more resources

Feedback