Maps of open chromatin guide the functional follow-up of genome-wide association signals: application to hematological traits

Abstract

Turning genetic discoveries identified in genome-wide association (GWA) studies into biological mechanisms is an important challenge in human genetics. Many GWA signals map outside exons, suggesting that the associated variants may lie within regulatory regions. We applied the formaldehyde-assisted isolation of regulatory elements (FAIRE) method in a megakaryocytic and an erythroblastoid cell line to map active regulatory elements at known loci associated with hematological quantitative traits, coronary artery disease, and myocardial infarction. We showed that the two cell types exhibit distinct patterns of open chromatin and that cell-specific open chromatin can guide the finding of functional variants. We identified an open chromatin region at chromosome 7q22.3 in megakaryocytes but not erythroblasts, which harbors the common non-coding sequence variant rs342293 known to be associated with platelet volume and function. Resequencing of this open chromatin region in 643 individuals provided strong evidence that rs342293 is the only putative causative variant in this region. We demonstrated that the C- and G-alleles differentially bind the transcription factor EVI1 affecting PIK3CG gene expression in platelets and macrophages. A protein-protein interaction network including up- and down-regulated genes in Pik3cg knockout mice indicated that PIK3CG is associated with gene pathways with an established role in platelet membrane biogenesis and thrombus formation. Thus, rs342293 is the functional common variant at this locus; to the best of our knowledge this is the first such variant to be elucidated among the known platelet quantitative trait loci (QTLs). Our data suggested a molecular mechanism by which a non-coding GWA index SNP modulates platelet phenotype.

Figure 1. Functional follow-up of the 7q22.3 locus associated with platelet volume and function.

(A) Regional plot of the 7q22.3 locus showing data from the discovery GWA meta-analysis as reported by Soranzo et al. The SNP rs342293 is indicated in purple (P=6.75×10⁻¹³).Values for r² are based on the 1000 Genomes Project (Pilot 1, CEU). The data was plotted with LocusZoom v1.1 (http://csg.sph.umich.edu/locuszoom/). (B) Sites of open chromatin marked by FAIRE across the locus in the CHRF-288-11 (MK) and K562 (EB) cell lines (data uploaded as UCSC Genome Browser custom tracks). (C) Site of open chromatin found in MK but not EB cells harboring the common variants rs342293 and rs342294. (D) In silico annotation of transcription factor binding sites at the open chromatin region described in (C).The predicted binding events are shown as transcription factor matrices (MatInspector v8.01). Note that some predicted binding sites are overlapping and not visible.

Figure 2. The effect of rs342293C>G on transcription factor binding.

(A) Nucleotide sequence of the synthetic probe harboring rs342293 and in silico prediction of transcription factor binding sites. The major (C-) allele of rs342293 is shown in bold. The minor (G-) allele of rs342293 is predicted to disrupt the binding motifs of the transcription factors GATA1 (− strand) and EVI1 (−). RUNX1 (+) and HHEX (+) transcription factor binding sites are predicted to be located 9 bp and 1 bp apart from rs342293, respectively. (B) Electrophoretic mobility shift assays (EMSA) in nuclear extracts from the megakaryocytic cell line CHRF-288-11 showed differential binding of the two alleles of rs342293. Competition reactions were performed with a 200-fold molar excess of unlabeled probes. Only the probe harboring the reference allele was able to shift the protein-DNA complex when incubating with EVI1 antibodies. Supershift experiments using GATA1 and RUNX1 antibodies are shown in Figure S6.

Figure 3. Association of rs342293 genotypes with PIK3CG transcript levels in platelets and macrophages.

In platelets (A) and macrophages (B), we observed a genotypic effect on PIK3CG transcript levels (P=0.0542 and P=0.0018, respectively). The associations are presented as box-and-whisker plots. In macrophages, the proxy-SNP rs342275A>G (r²=0.94, Phase II HapMap, CEU) was used for the eQTL analysis.

Figure 4. Protein–protein interaction network centered on PIK3CG.

A protein-protein interaction network was constructed centering on the human orthologous proteins encoded by the 220 differentially expressed transcripts between Pik3cg knockout and wild type mice with a fold-change of at least ±1.5. The color of these ‘core’ proteins represents the fold-change of gene transcript levels between Pik3cg^−/− and wild type mice on a continuous scale from over- (red) to underexpression (blue) of transcripts in knockout mice. Core proteins, which do not exhibit first-order interactors in the Reactome database and are disconnected from the largest connected component, are not shown. First-order interactors of PIK3CG that form network nodes and edges are shown in grey and were obtained from Reactome. Interactors that are not expressed in MK cells based on the HaemAtlas data are omitted. The resulting network consists of 45 core proteins with 642 nodes and 1067 edges.

Figure 5. Model of the mechanism by which rs342293 may affect platelet volume.

Based on our data, we propose that rs342293C>G affects EVI1 transcription factor binding at the megakaryocyte-specific site of open chromatin at chromosome 7q22.3. The underlying DNA sequence around rs342293 (indicated in red) is shown and transcription factor binding motifs for RUNX1 (green) and EVI1 (purple) are highlighted. The transcriptional complex of EVI1 is a well-described repressor of gene transcription. In platelets, individuals with homozygous rs342293-C have lower PIK3CG gene expression levels compared to subjects with homozygous rs342293-G. In Pik3cg knockout mouse models, transcripts that encode key proteins for platelet membrane biogenesis were found to be downregulated, which may ultimately affect platelet phenotype.