Genetic associations at 53 loci highlight cell types and biological pathways relevant for kidney function

Abstract

Reduced glomerular filtration rate defines chronic kidney disease and is associated with cardiovascular and all-cause mortality. We conducted a meta-analysis of genome-wide association studies for estimated glomerular filtration rate (eGFR), combining data across 133,413 individuals with replication in up to 42,166 individuals. We identify 24 new and confirm 29 previously identified loci. Of these 53 loci, 19 associate with eGFR among individuals with diabetes. Using bioinformatics, we show that identified genes at eGFR loci are enriched for expression in kidney tissues and in pathways relevant for kidney development and transmembrane transporter activity, kidney structure, and regulation of glucose metabolism. Chromatin state mapping and DNase I hypersensitivity analyses across adult tissues demonstrate preferential mapping of associated variants to regulatory regions in kidney but not extra-renal tissues. These findings suggest that genetic determinants of eGFR are mediated largely through direct effects within the kidney and highlight important cell types and biological pathways.

Figure 1. Discovery stage genome-wide association analysis.

Manhattan plots for eGFRcrea, CKD and eGFRcys. Previously reported loci are highlighted in light blue (grey labels). (a) Novel loci uncovered for eGFRcrea in the overall and in the non-diabetes groups are highlighted in blue and green, respectively. (b) Results from CKD analysis with highlighted known and novel loci for eGFRcrea. (c) Results from eGFRcys with highlighted known and novel loci for eGFRcrea and known eGFRcys loci.

Figure 2. Association eGFRcrea loci in subjects with and without diabetes.

Novel (a) and known (b) loci were considered. Displayed are effects and their 95% confidence intervals on ln(eGFRcrea). Results are sorted by increasing effects in the diabetes group. The majority of loci demonstrated similar effect sizes in the diabetes as compared with non-diabetes strata. SNP-specific information and detailed sample sizes are reported in Supplementary Table 5.

Figure 3. Bioinformatic analysis of eGFR-associated SNPs.

Connection of eGFR-associated SNPs to gene expression and variant function across a variety of tissues, pathways and regulatory marks was considered. (a) The DEPICT method shows that implicated eGFR-associated genes are highly expressed in particular tissues, including kidney and urinary tract. Shown are permutation test P values (see Methods). (b) Enrichment of eGFRcrea-associated SNPs in DHS according to discovery P value threshold. SNPs from the eGFR discovery genome-wide scan meeting a series of P value thresholds in the range 10⁻⁴–10⁻¹⁶ preferentially map to DHSs, when compared with a set of control SNPs, in 6 of 123 cell types. Represented are main effects odds ratios from a logistic mixed effect model. Cell types indicated with coloured lines had nominally significant enrichment (* indicate P values <0.05) at the P value <10⁻¹⁶ threshold and/or were derived from renal tissues (H7esDiffa2d: H7 embryonic stem cells, differentiated 2 days with BMP4, activin A and bFGF; Hae, amniotic epithelial cells; Hrce, renal cortical epithelial cells; Hre, renal epithelial cells; Hrgec, renal glomerular endothelial cells; Rptec, renal proximal tubule epithelial cells; Saec, small airway epithelial cells). (c) ENCODE/Chromatin ChIP-seq mapping: known and replicated novel eGFRcrea-associated SNPs and their perfect proxies were annotated based on genomic location using chromatin annotation maps from different tissues including adult kidney epithelial cells. P values from Fishers' exact tests for 2 × 2 tables are reported (significance level=5.6 × 10⁻³, see Methods). There is significant enrichment of variants mapping to enhancer regions specifically in kidney but not other non-renal tissues.

Figure 4. Gene set overlap analysis.

The 19 reconstituted gene sets with P value<10⁻⁵ were considered. Their overlap was estimated by computing the pairwise Pearson correlation coefficient ρ between each pair of gene sets followed by discretization into one of three bins: 0.3≤ρ<0.5, low overlap; 0.5≤ρ<0.7, medium overlap; ρ⩾0.7, high overlap. Overlap is shown by edges between gene set nodes and edges representing overlap corresponding to ρ<0.3 are not shown. The network was drawn with Cytoscape.