Network-based analysis of genome wide association data provides novel candidate genes for lipid and lipoprotein traits

Abstract

Genome wide association studies (GWAS) identify susceptibility loci for complex traits, but do not identify particular genes of interest. Integration of functional and network information may help in overcoming this limitation and identifying new susceptibility loci. Using GWAS and comorbidity data, we present a network-based approach to predict candidate genes for lipid and lipoprotein traits. We apply a prediction pipeline incorporating interactome, co-expression, and comorbidity data to Global Lipids Genetics Consortium (GLGC) GWAS for four traits of interest, identifying phenotypically coherent modules. These modules provide insights regarding gene involvement in complex phenotypes with multiple susceptibility alleles and low effect sizes. To experimentally test our predictions, we selected four candidate genes and genotyped representative SNPs in the Malmö Diet and Cancer Cardiovascular Cohort. We found significant associations with LDL-C and total-cholesterol levels for a synonymous SNP (rs234706) in the cystathionine beta-synthase (CBS) gene (p = 1 × 10(-5) and adjusted-p = 0.013, respectively). Further, liver samples taken from 206 patients revealed that patients with the minor allele of rs234706 had significant dysregulation of CBS (p = 0.04). Despite the known biological role of CBS in lipid metabolism, SNPs within the locus have not yet been identified in GWAS of lipoprotein traits. Thus, the GWAS-based Comorbidity Module (GCM) approach identifies candidate genes missed by GWAS studies, serving as a broadly applicable tool for the investigation of other complex disease phenotypes.

Fig. 1.

Schematic representation of GCM approach. (1): Mapping of GLGC GWAS meta-analysis SNPs to genes. (2): Construction of a human interactome by pooling protein interaction data from different sources and mapping seed genes within the network. (3): Identification of candidate genes associated with lipid/lipoprotein traits using molecular triangulation (MT). (4): Identification of seed and neighbouring gene modules using the jActiveModule (jAM) method, pruning of MT candidate gene sets. (5): Selection of phenotypically coherent (GCM) modules of seed and candidate genes using comorbidity analyses. (6): Validation of MT, jAM and GCM gene set outputs and comparison to CANDID and MetaRanker methods. (7): Selection of SNPs, representing GCM candidate genes, for genotyping in the MDC-CC. GCM genes were prioritized based on their co-expression with seed genes and hierarchical criteria including the genomic locations of SNPs, if the SNPs were synonymous variations, if the SNPs were in conserved regions of the genome, and GLGC GWAS-meta-analysis p-values (p < 0.05).

Fig. 2.

Performance of CANDID, MetaRanker, MT, jActiveModule and GCM with respect to benchmarking dataset. The histograms show an increase in specificity, precision and accuracy with each of the steps.

Fig. 3.

GO term enrichments for gene sets. A, Mean numbers of genes, anywhere in the genome, associated with GO terms for which significant enrichments were found. B, Mean numbers of genes, within the sets of genes being tested, found to be associated with GO terms for which the gene sets were enriched. C, Counts of GO terms for which sets of candidate genes were enriched. D, Median odds ratios of GO term enrichment as a measure of enrichment effect size.

Fig. 4.

Percentage of overlapping candidate genes between CANDID or MetaRanker and each of the GCM steps.

Fig. 5.

GCM module with CBS gene and the associated diseases. Combination schema including protein-protein interactions (purple), metabolic interactions (red), and transcriptional interactions (yellow), gene-disease associations (dashed black), and relative risk associations between diseases greater with magnitude greater than 1 (black line). Seed genes (red ovals), CBS GCM genes (dark blue ovals) and diseases (gray) are linked within a highly interconnected module that includes Homocystinuria, venous embolism and thrombosis diseases associated with CBS gene in OMIM.