henhaobio
Sign in

medRxivGenetic & Genomic MedicinePreprint

Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants.

Krishna G. AragamiD, Tao JiangiD, Anuj GoeliD, Stavroula KanoniiD, Brooke N. WolfordiD, Elle M. WeeksiD, Minxian WangiD, George HindyiD, Wei ZhouiD, Christopher GraceiD, Carolina RoselliiD, Nicholas MarstoniD, Frederick KamanuiD, Ida L. SurakkaiD, Loreto Muñoz VenegasiD, Paul SherlikeriD, Satoshi KoyamaiD, Kazuyoshi IshigakiiD, Bjørn Olav ÅsvoldiD, Michael R Brown, Ben Michael BrumptoniD, Paul S. de VriesiD, Olga GiannakopoulouiD, Panagiota GiardoglouiD, Daniel Fannar GudbjartssoniD, , Adam S. ButterworthiD

PDFDOIData

In-depth summary

Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (2 Jun 2021) · CC BY 4.0

A coronary artery disease GWAS of 181,522 cases among 1,165,690 participants yields 241 associations (54 new); an eight-predictor consensus nominates 185 candidate causal genes, 94 with ≥3 predictors.

Discovery of coronary artery disease (CAD) loci has moved faster than work on how they act, and in silico tools for assigning causal genes can give spurious or conflicting answers. This work pairs a larger GWAS with chromatin-informed fine-mapping, the Polygenic Priority Score (PoPS, which uses genome-wide signal that earlier similarity-based methods left unused) and an eight-predictor consensus, producing ranked candidate variants and genes plus a risk score that improves on one from a 2015 GWAS.

Key results

  1. ShownThe meta-analysis found 241 conditionally independent genome-wide significant CAD associations at 198 loci. Of these, 54 are new, and 30 lie in regions not previously linked to CAD. Adding Biobank Japan gave 38 further new loci.

    Genome-wide threshold p≤5.0×10^-8; new means r^2<0.2 with reported variants. 46 of 54 new sentinels are common (minor allele frequency >0.05), odds ratios 1.03-1.07. In UK Biobank phenome-wide scans, 130 of 241 (54%) had no association with lipids, blood pressure, type 2 diabetes or adiposity. Effects correlated r=0.59 with Biobank Japan over 199 shared sentinels. · Discovery of known and novel CAD loci; Allelic architecture; Trans-ethnic comparison and meta-analysis

  2. ShownA genome-wide polygenic risk score (PRS) built with LDpred from these statistics (the 2021 PRS) predicted incident and recurrent coronary events better than the best score from a 2015 GWAS of about 60,000 cases.

    Malmo Diet and Cancer held-out subset (n=5,685; 815 incident cases): age/sex-adjusted hazard ratio per SD 1.61 (95% CI 1.50-1.72) vs 1.49 (1.39-1.59); 2021 PRS stays at 1.54 after risk-factor adjustment; top vs bottom decile 5.7-fold vs 3.8-fold. FOURIER placebo arm (n=7,135; 673 events): 1.20 (1.11-1.29) vs 1.13 (1.04-1.22); deciles 1.7-fold vs 1.4-fold. · Association of polygenic risk scores with…; Derivation and training of polygenic risk scores; Primary event prediction analyses in the Malmo…; Secondary event prediction analyses in the…

  3. ShownFGWAS fine-mapping reweights association statistics by how enriched Roadmap chromatin states are in 10 CAD-relevant tissues. It narrowed many signals and sometimes favoured a variant other than the sentinel.

    116 of 197 regions enriched in at least one tissue (most often adipose, 28; liver, 24; aorta, 21). 95% credible sets held 1,456 variants; 42 regions had fewer than five, 49 had a variant with posterior probability of association (PPA) ≥0.5, and 13 resolved to one variant. At 10 loci the top variant was not the sentinel, e.g. MAFB rs1883711 PPA 0.77 vs 0.13 for rs2207132. · Functionally-informed fine-mapping; Functional GWAS analysis

  4. ShownAn unweighted consensus of eight locus- and similarity-based predictors recovered 28 of 30 established CAD genes and nominated 185 candidate causal genes. For 25 of them, the nominated gene is not the one nearest the sentinel.

    Positive-control accuracy: nearest gene 87%, PoPS 80%, expression QTL (eQTL) 79%, mouse knock-out 95%. 206 of 241 associations (85.5%) had at least two concordant predictors; 94 genes had three or more. Non-nearest picks include APOC3, PLTP and LOX. Pairwise concordance ranged from 30/41 (73%) for nearest gene with missense variant to 5/17 (29%) for monogenic genes with eQTL. · Systematic prioritization of putative causal…; Systematic integration of gene prioritization…

  5. SuggestedThe PoPS features behind gene prioritization pointed to lipid homeostasis, vascular cell proliferation, extracellular matrix, early development and cell-cycle signalling. PoPS also named MFGE8 (not the nearest gene) and candidates at new loci (ITGA1, LIPC, MYO9B).

    21,407 of 57,543 features passed selection, forming 3,149 clusters; the top cluster was blood-lipid homeostasis. At intergenic rs1807214, PoPS ranked MFGE8 (108kb upstream) first of ten genes, above nearest gene ABHD2, matching CRISPR deletion data the authors cite. At ITGA1, intronic rs4074793 had PPA 0.95 and led a STARNET liver cis-eQTL (p=1.8×10^-73). · Polygenic prioritization of candidate causal…; Candidate loci with converging variant- and…

Methods, data and reproducibility

Inverse-variance meta-analysis (METAL) of ten new imputed array-genotyped studies with CARDIoGRAMplusC4D summary statistics, then GCTA joint analysis, UK Biobank phenome-wide and rare-variant tests, sex-differentiated and Biobank Japan meta-analyses, PRS training and validation, FGWAS fine-mapping, and gene prioritization from eight predictors including PoPS, eQTLs, ClinVar and mouse knock-outs.

  • Ten de novo case-control studies (imputed genotyping arrays) meta-analysed with CARDIoGRAMplusC4D summary statistics; over 95% European ancestry · human · n = 181,522 CAD cases among 1,165,690 participants
  • UK Biobank: phenome-wide association scan, gene-based missense/loss-of-function tests, LD reference · human · n = PheWAS 393,461; gene tests 33,941 cases / 438,394 controls
  • Biobank Japan CAD GWAS summary statistics (East Asian ancestry) for comparison and trans-ethnic meta-analysis · human · n = 29,319 cases, 183,134 controls
  • Malmo Diet and Cancer Study, used for PRS training and held-out validation of incident CAD · human · n = training 22,872 (3,307 incident cases); validation 5,685 (815 cases)
  • FOURIER trial genetic substudy, placebo arm, recurrent major coronary events · human · n = 7,135 participants; 673 events
  • Sex-stratified GWAS summary statistics for sex-differentiated meta-analysis · human · n = 77,080 CAD cases
  • Functional annotation: NIH Roadmap Epigenomics ChromHMM states (10 cell/tissue types), STARNET cis-eQTLs (7 tissues) and GTEx v7 cis-eQTLs (8 tissues) · human · n = 600 CAD patients (STARNET)
  • Single-gene knock-out phenotypes from the International Mouse Phenotyping Consortium (release 10.1) and Mouse Genome Informatics (July 2019) · mouse
Methods & tools
METAL, GCTA-COJO, SAIGE / SAIGE-GENE, GWAMA, LDpred, PLINK2, FGWAS, PoPS, MAGMA, Ensembl Variant Effect Predictor (LOFTEE), qvalue (R package), ClinVar
Reagents & environment
LDpred v1.0; SAIGE-GENE v0.35.8.5; Variant Effect Predictor v96.0 with LOFTEE; PLINK v2.0; R 3.6.1; GENCODE v19; GTEx v7; Haplotype Reference Consortium v1.1 imputation panel. No container or requirements file stated.
Reproducibility gaps
Code is available only on reasonable request to the corresponding authors. GWAS summary statistics (CARDIoGRAMplusC4D, GWAS Catalog) and PRS weights (PGS Catalog) are promised upon publication, not at posting.
Limitations (6)
  • Participants were over 95% European ancestry; the authors say it remains to be studied how far adding non-European participants would improve PRS portability across ancestry groups. · stated by the authors
  • The prioritized variants and genes are candidates; the authors state that experimental evidence is required to confirm causal mechanisms and suggest adding more cell types (e.g. vascular smooth muscle cells) and CRISPR screens. · stated by the authors
  • The authors expect larger European-ancestry GWAS to find few additional large-effect common variants and to improve PRS prediction only modestly. · stated by the authors
  • Rare-variant gene tests were run only in UK Biobank and were not extended to other studies because only PCSK9 gave a strong signal. · stated by the authors
  • The 30 positive-control genes were chosen by the authors. Established CAD genes probably carry drug, Mendelian randomization, monogenic or mouse evidence that also serve as predictors, so the 28 of 30 recovery may overstate accuracy on less-studied loci. · noted in this summary
  • The incident-CAD PRS was trained and validated within a single cohort (Malmo Diet and Cancer). The FOURIER recurrent-event count is 673 in the Results but 1,074 in the Figure 2 legend. · noted in this summary

Read this if you work on CAD or cardiometabolic genetics and want prioritized variants and genes to test, a stronger CAD PRS, or a worked example of combining locus- and similarity-based gene predictors.

Citations
67

Abstract

Rapid progress of the discovery of genetic loci associated with common, complex diseases has outpaced the elucidation of mechanisms pertinent to disease pathogenesis. To address relevant barriers for coronary artery disease (CAD), we combined genetic discovery analyses with downstream characterization of likely causal variants, genes, and biological pathways. Specifically, we conducted a genome-wide association study (GWAS) comprising 181,522 cases of CAD among 1,165,690 participants. We detected 241 associations, including 54 associations and 30 loci not previously linked to CAD. Next, we prioritized likely causal variants using functionally-informed fine-mapping, yielding 42 associations with fewer than five variants in the 95% credible set. Combining eight complementary predictors, we prioritized 185 candidate causal genes, including 94 genes supported by three or more predictors. Similarity-based clustering underscored a role for early developmental processes, cell cycle signaling, and vascular proliferation in the pathogenesis of CAD. Our analysis identifies and systematically characterizes risk loci for CAD to inform experimental interrogation of putative causal mechanisms for CAD.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISJacobsen '26Schnitzler '22Barbera '24Matsunaga '20Ramste '25Hartmann '22
83% semantic matchGenetic & Genomic Medicine
Jacob Thoft Jacobsen, Peter Loof Møller, Palle Duun Rohde·medRxiv·Apr 2, 2026