|Home | About | Journals | Submit | Contact Us | Français|
The QT interval, a measure of cardiac repolarization, predisposes to ventricular arrhythmias and sudden cardiac death (SCD) when prolonged or shortened. A common variant in NOS1AP is known to influence repolarization. We analyze genome-wide data from five population-based cohorts (ARIC, KORA, SardiNIA, GenNOVA and HNR) with a total of 15,842 individuals of European ancestry, to confirm the NOS1AP association and identify nine additional loci at P < 5 × 10−8. Four loci map near the monogenic long-QT syndrome genes KCNQ1, KCNH2, SCN5A and KCNJ2. Two other loci include ATP1B1 and PLN, genes with established electrophysiological function, whereas three map to RNF207, near LITAF and within NDRG4-GINS3-SETD6-CNOT1, respectively, all of which have not previously been implicated in cardiac electrophysiology. These results, together with an accompanying paper from the QTGEN consortium, identify new candidate genes for ventricular arrhythmias and SCD.
The lack of serologic biomarkers to predict ventricular tachycardia, ventricular fibrillation and sudden cardiac death has made genome-wide searches for common genetic variants influencing these traits extremely important1. Studies of the QT interval from the electrocardiogram (EKG), which captures major temporal and spatial aspects of the repolarization process, are particularly attractive as its prolongation or shortening reflects alterations of cardiac repolarization known to trigger ventricular tachycardia and ventricular fibrillation and predispose to SCD2. Importantly, QT interval prolongation has been associated with increased cardiovascular mortality in individuals with heart disease3, as well as in the general population4. Furthermore, mutations in genes associated with long- and short-QT syndromes (LQTS, SQTS) markedly increase the odds of SCD5.
Previous genetic analyses of QT interval have largely relied on family-based studies focused on rare mendelian QT syndromes. These have resulted in the identification of eleven genes in which mutations cause prolongation or shortening of the QT interval and SCD. Recently, in an initial genome-wide association study (GWAS) in the KORA community-based cohort, an association between common genetic variants in the NOS1AP (CAPON) gene and the QT interval was identified6. This association highlights the importance of the nitric oxide synthase pathway in myocardial function and has now been replicated in several studies7–9. Genetic variants at NOS1AP explain ~1% of the variance in QT interval, and common genetic variants in NOS1AP are also associated with SCD in populations of European ancestry10.
In this study, we carried out GWAS in five population-based cohorts from Europe and the United States (ARIC11, SardiNIA12, KORA13, GenNOVA14 and Heinz Nixdorf Recall (HNR)15) (Supplementary Table 1 online). In all the participating studies all individuals studied and all analyses on their samples were done according to the Helsinki declarations and were approved by the local medical ethics and institutional review committees. All participants gave signed informed consent. We used preexisting genome-wide SNP data of 15,842 individuals randomly ascertained within these cohorts to identify additional genes modulating the QT interval. Cohort descriptions are detailed in Supplementary Methods online. Samples had been genotyped using either Affymetrix Gene Chip Human Mapping Array Set 6.0 (ARIC, KORA S4) or 500K (KORA F3, SardiNIA) and Illumina HumHap300v2 (GenNOVA) or HumanHap 550v3 BeadChips (KORA S4, HNR): genotyping details and SNP quality control filters for each study are summarized in Supplementary Table 2 online. To facilitate comparison of results across studies, we imputed HapMap SNPs in all study participants using the HapMap CEU sample as a reference (HapMap release 21)16. We excluded individuals with atrial fibrillation, pacer and/or defibrillator implants, prolonged QRS intervals (>120 ms) indicating bundle branch blocks or other conduction disorders, as well as pregnant women. Heart rate (RR interval), age and sex were included as covariates and adjusted for in all analyses (Supplementary Table 3a online). The standard deviation of the adjusted QT interval meta-analyzed across all studies was ±17.8 ms.
Results for the consequent genome-wide analysis are summarized in Figures 1 and and2.2. The genomic control factor (λ) for this analysis was 1.016, indicating that unmodeled relatedness and population structure had no appreciable impact on our results17. The quantile-quantile plot in Figure 1 (inset) shows a clear excess of extreme P values, indicating the presence of true associations. After having identified ten main association signals across the genome (P <5 × 10−8) we carried out a second round of genome-wide analysis after adjusting for the main signals. This round led to the identification of independent secondary signals in two of the ten loci (P < 5 × 10−8).
Consistent with a previous GWAS study6, the strongest main association signal maps to the NOS1AP locus. There, the most significant association was at SNP rs12143842 (P = 1.62 × 10−35, Figs. 1 and and2b,2b, Table 1, and Supplementary Table 4 online). In the NOS1AP locus we identified an independent secondary signal at rs4657178 which was in low linkage disequilibrium (LD) to the main signal (r2 = 0.001 to rs12143842 in HapMap CEU, P = 1.02 × 10−22 before and 9.98 × 10−13 after adjustment for main association signals, Table 2) supporting the notion of at least two common QT-modifying variants at that locus. The SNP rs10494366, indicating the most significant association signal in the previous report6, was in moderate LD with the main signal, and to a lesser degree also with the secondary signal (Table 3). After adjustment for the ten main association signals a significant reduction in the association with this SNP occurs, indicating that it does not represent an independent tertiary signal but rather a marker for the main signal at rs12143842.
After excluding SNPs in a 1-Mb region surrounding NOS1AP, we still observed a clear excess of small P values distributed across several genetic loci (Fig. 1, inset), indicating the presence of additional genomic loci associated with the QT interval. Overall, nine additional loci show association at P < 5 × 10−8, corresponding to genome-wide significance at 5% after adjustment for ~1 million independent tests, the estimated multiple testing burden of HapMap SNPs in samples of European ancestry16.
Notably, four of these associations map in or near genes known to harbor both LQTS- and SQTS-causing mutations: KCNQ1 (LQT1, SQT2, 11p15.5, rs12296050, P = 8.52 × 10−9), KCNH2 (LQT2, SQT1, 7q36.1, rs2968863, P = 3.79 × 10−9), KCNJ2 (LQT7, SQT3, 17q24.3, rs17779747, P = 3.36 × 10−8) and SCN5A (LQT3, Brugada syndrome, 3p22.2, rs11129795 with P = 3.67 × 10−8). SCN5A encodes a sodium channel, and, in previous studies, the nonsynonymous variant S1103Y with allele frequency 8% only in individuals of African ancestry has been associated with increased risk of arrhythmias18. The three other genes encode potassium channel α-subunits in the cardiomyocyte plasma membrane.
Previous candidate gene analyses provided evidence of association between common variants and QT interval for three of these loci: KCNQ1, KCNH2 and SCN5A. We investigated the previously identified variants with respect to our data on the basis of their surrounding LD patterns and effect sizes before and after adjustment for the main signals (Table 3 and Supplementary Fig. 1 online). The nonsynonymous coding KCNH2 variant K897T (rs1805123)19–21, the variant in intron 1 of KCNQ1 (rs757092)20 and the noncoding SCN5A variant D1819D (rs1805126)22 all seem to be in LD with the main association signals we detected. After adjustment for the main signals, with which they were in moderate to high LD, these SNPs were no longer significantly associated. In contrast, the signal at the SCN5A variant H558R (rs1805124)22 did not reach genome-wide significance in our data (P = 4.06 × 10−3) but remained almost unchanged after adjustment, suggesting a small but independent association signal. Additionally, the previously published secondary association signals in KCNH2 show independent but slightly less than genome-wide significant signals in our data: rs3815459 (ref. 20) (P = 5.07 × 10−8) and rs3807375 (ref. 21) (P = 4.68 × 10−7) (Table 3). Both seem to be in LD with the strongest secondary signal we identified in KCNH2 at rs3778873 bordering genome-wide significance (P = 5.07 × 10−8 before and P = 7.90 × 10−5 after adjustment for main signals) (Supplementary Table 5 online).
These results emphasize that in many loci we are likely to identify genes with an allelic series comprising both common variants influencing QT interval with modest effects (~3–6 ms per locus in this study) in healthy volunteers as well as rare variants with a more marked effect (>100 ms) in individuals with a genetic syndrome23.
In the 17q24.3 region the strongest association signal was missed in previous candidate gene studies probably because the associated signal at SNP rs17779747 maps ~300 kb away from the KCNJ2 gene. Although the LD block it resides on extends toward the gene, SNP rs17779747 is not in high LD with any common variant within the KCNJ2 coding sequence, the strongest being to the synonymous coding rs173135 (L382L, r2 = 0.014). KCNJ2 remains the best prior candidate in the region as mutations in it are known to cause Andersen syndrome (MIM170390), a condition characterized by periodic paralysis, dysmorphic features and cardiac ventricular arrhythmias triggered by repolarization disturbances. Alternatively, other genes from the region, including the nearby paralog KCNJ16, may have a role24.
We identified two new loci with genes encoding proteins with well-established myocardial electrophysiological functions: in chromosome 1q24.2 the strongest signal was within ATP1B1 (Na+/K+ ATPase beta subunit 1, rs10919071, P = 2.18 × 10−12, Fig. 2c). ATP1B1 encodes a transmembrane protein that has a crucial role in the maintenance of Na+ and K+ gradients across membranes, thus regulating electrical excitability of muscles, and may also be involved in the regulation of blood pressure25. These data make ATP1B1 a strong functional candidate, although NME7, BLZF1, C1orf114 and SLC19A2 cannot be excluded without functional validation.
In chromosome 6q22.31 a broad association signal covers PLN (phospholamban), SLC35F1 and C6orf204, with the strongest signal at the intergenic SNP rs11970286 (P = 1.96 × 10−16, Fig. 2e). PLN is the strongest regional candidate, as it is a regulator of the sarcoplasmic reticulum Ca2+ ATPase (encoded by ATP2A2, also known as SERCA2) responsible for diastolic lowering of the cytoplasmic Ca2+ concentration. Mutations in this gene have previously been associated with inherited cardiomyopathies and congestive heart failure26. Neither SLC35F1 nor C6orf204 has previously been shown to have a functional role in the myocardium. In the 6q22.31 region we identified an independent second variant (rs12210810, intergenic, P = 3.24 × 10−13 before and P = 4.50 × 10−8 after adjustment for main signals). The low LD to the main signal (r2 = 0.067) and the fact that the rare allele at this second signal was associated with shorter QT intervals strengthen the assumption of two independent causal variants in or near PLN. At each of these two loci, there were no additional compelling biological candidates that we could identify.
At the remaining three loci discovered, there was no obvious biological candidate. One of these loci in chromosome 1p36.31 overlaps LITAF (rs8049607, P = 2.90 × 10−8, Fig. 2h), which encodes a DNA-binding protein thought to have a role in the regulation of TNFA expression. Mutations in this gene have previously been implicated in Charcot-Marie-Tooth type 1C neuropathy and are sometimes associated with reduced nerve conduction velocity27. Another candidate gene in the region is TXNDC11, which encodes thioredoxin domain-containing protein 11 and harbors a nonsynonymous SNP in weak LD with the leading variant (V756L, rs3190321, r2 = 0.022), as well as SNN, encoding stannin (rs8191288, r2 = 0.022). A second newly identified locus was in chromosome 1p36.31, with a nonsynonymous coding SNP in RNF207 as the main signal (G603A, rs846111, P = 3.56 × 10−9, Fig. 2a). Its best proxy, rs709209, is another nonsynonymous SNP in the same gene (N573S, r2 = 0.673). RNF207 encodes a RING-type zinc-finger protein of unknown function. Other regional candidate genes are GPR153 (rs4908542, intronic, r2 = 0.445), CHD5 (rs12754299, r2 = 0.021), ICMT (rs846108, r2 = 0.020), HES2 (rs932402, r2 = 0.018) and the gene for the shaker-related potassium channel KCNAB2 (rs2294934, r2 = 0.028). None of the SNPs in the neighboring genes were coding and none of the genes have been previously implicated in myocardial pathology.
The third identified locus maps to chromosome 16q21, the strongest signal being an intronic SNP within the CNOT1 gene (rs7188697, P = 1.25 × 10−12, Fig. 2i), encoding a subunit of the CCR4-NOT transcription complex. This locus is highly conserved and syntenic in many vertebrates and contains other potentially causal genes such as SETD6, encoding SET domain–containing protein 6 (rs37036, r2 = 0.920), NDRG4, encoding vascular smooth muscle cell–associated protein 8 (SMAP-8) (rs40186, r2 = 0.468), and GINS3, encoding GINS complex subunit 3 (rs8054945, r2 = 0.041). A recent genetic screen in zebrafish mutants of resistance and sensitization to dofetilide, a drug causing atrioventricular blocks, uncovered a GINS3 mutation that modifies cardiac repolarization (personal communication, D. Milan, Massachusetts General Hospital and Harvard Medical School). NDRG4, a gene known to be expressed in the heart during zebrafish development that regulates proliferation and growth of cardiomyocytes28, is also a plausible candidate. Human NDRG4 has three alternative promoters, all of them containing consensus binding sequences for the Tbx5 transcription factor29, which is known to operate in conjunction with Nkx2.5 in human heart development30. Additionally, a combined effect of several genes may exist given the high degree of conserved synteny in that locus. For all the loci we identified, but particularly for these last three, more experimental evidence will be required to identify likely functional mechanisms. This could include the sequencing of genes in individuals with unexplained long-QT syndrome and functional screens in model organisms.
The nine newly identified main association signals identified here increase the proportion of explained variance in heart rate-, sex- and age-adjusted QT interval from 1.0% by the NOS1AP SNP rs12143842 signal alone to 3.3% in a meta-analysis across all studies; a more comprehensive analysis of all variation at these loci, including secondary signals, is likely to increase this quantity further. In addition, there are likely rarer variants at each of these genes that modulate QT intervals to a greater extent but have been missed in our screen for common variants, as has been observed for high-density-lipoprotein cholesterol levels31.
Variation in demographic factors such as age and sex also contributes to the QT interval phenotype32. We carried out association analysis for males and females separately, and for younger and older subgroups of individuals (stratified by age ≤50 y). As shown in Table 4, the main signal at NOS1AP shows a stronger effect in females, supporting a recently published study that demonstrated a sex-specific effect for SNP rs10494366 (refs. 9,33). In addition, we observed data supporting the possibility of age-specific effects at RNF207 and PLN. In RNF207 the QT-prolonging effect of the rare C allele of rs846111 is 1.277 ms larger in young individuals (P = 0.014), whereas in PLN the QT-shortening effect of the rare C allele of rs12210810 is 1.809 ms larger in older individuals (P = 0.036). We replicated the previously published sex-specific effect at NOS1AP; no other age- or sex-specific effects remained significant after adjustment for the number of performed tests.
Finally, our analyses should be compared with the results of an independent yet similar GWAS by the QTGEN consortium34 that adds further confidence in these results. All the loci that reach P <5 × 10−8 in our screen also show evidence for association in their study of 13,685 individuals (each with P < 1 × 10−4). Furthermore, our data provides evidence for association with the QT interval at chromosome 17q12 near the genes LIG3 and RFFL (rs2074518, P = 3.03 × 10−6), a locus that exceeded the genome-wide significance threshold in the QTGEN consortium. Owing to reduced genotyping coverage of the KCEN1 locus as compared to QTGEN, our data do not support the association with the nonsynonymous D85N variant in KCNE1 (refs. 20,22,34–36).
To assess the cumulative effect of the ten main QT-associated SNPs, we summed up the number of QT interval–prolonging alleles for each participant in the studies of unrelated individuals (N = 10,563) (Supplementary Methods). Overall, we observed a 1.53 ± 0.08 ms (P = 1.79 × 10−88) increase in the mean QT interval for each additional QT-prolonging allele, with a difference of 18.1 ms between individuals with a QT score of ≤6 or ≥16 (Fig. 3). The 58% of individuals with a score of ≥11 had an OR of 1.49 for prolonged QT using clinical thresholds (males ≥440 ms, females ≥450 ms) when compared with the 42% of individuals with a score of ≤10 (95% CI = 1.27–1.76, P = 1.54 × 10−6). Further at the extremes, the ~8% of individuals with a score of ≥14 had an OR of 2.52 for prolonged QT when compared to the ~10% individuals with a score ≤8 (95% CI = 1.74–3.66, P = 4.83 × 10−7).
Our results illustrate the power of GWAS to identify common variants both at loci previously unsuspected of involvement in cardiovascular function as well as at loci with a documented role in the regulation of QT interval. Although the statistical evidence of these associations is compelling mostly owing to the size of current GWAS, we are aware of limitations of this approach. As population stratification may confound our association findings, we have calculated a genome-wide Fst statistic between populations37. This indicates an overall Fst statistic of 0.004 between all cohorts, comparatively minor relative to the value of 0.12 obtained by the International HapMap Project when comparing Europeans, Africans and Asians (Supplementary Table 6 online). We report cohort-specific results in Supplementary Table 7 online.
Also we did not account for some known covariates of QT interval, as these either did not contribute significantly to the model fit (Supplementary Table 3b) or were not uniformly available across studies and the increase of explained variance of QT interval was small at the population level (such as for example, for underlying cardiac pathologies, serum potassium levels and medication use) (Supplementary Table 3c). We may potentially have missed some association signals especially in the presence of interaction. As most complex genetic associations identified today perform well in log linear models38 and as the accompanying study34 did not identify any loci that our study would dismiss, we do not consider this limitation a major one. A post hoc analysis of the ten main association signals identified in a model fitting additional covariates indicated no major improvement in model fit (Supplementary Table 8 online): six out of the ten loci slightly gained in significance level attained, whereas four loci decreased. Other limitations include the overestimation of effect sizes in initial discoveries due to the ‘winners curse’ phenomenon and the inability of the association approach to identify underlying genes or mechanisms in the regions of association signals.
The identification of causal genes and mechanisms at each of the loci remains a major task. This may be performed by functional experiments, but genetic studies are integral to scrutinize loci for causal mutations and establish pathways that regulate myocardial function. Promising genetic methods include SNP fine mapping at each locus at a resolution higher than that of the HapMap, and resequencing the target locus in individuals from the extremes of the QT interval distribution and in individuals diagnosed with long-QT syndrome in order to identify additional common as well as rare variants that may reveal the causal genes. The approach of establishing an allelic series at a given locus including both common and rare variants seems particularly promising. Four of ten loci we identified by association do overlap with genes known to harbor mutations in LQTS and SQTS. Loci showing allelic series have also been identified in genome-wide association studies of lipid levels39, height40 and uric acid41. In the future, it may be worthwhile to include all the loci identified here in sequencing efforts in cases of long- and short-QT syndromes and in subjects where these syndromes are induced after exposure to specific medications. More immediately, our results point to a specific set of loci that are associated with QT interval and provide further targets for molecular studies of susceptibility to QT-triggered ventricular arrhythmias, sudden cardiac death and cardiovascular function in general.
For a description of methods used, see Supplementary Methods.
We gratefully acknowledge all participants in the community-based studies of ARIC, KORA, SardiNIA, GenNOVA and Heinz Nixdorf Recall Study, and all the members of our laboratories for helpful discussion of this study. We thank G. Fischer for help with genotype imputation and data management for the KORA samples, Y. Li for her help on statistical analysis, K. Tarasov for in silico promotor analysis, A. Cao for his valuable advice and support in the SardiNIA project and all ARIC, KORA, SardiNIA, GenNOVA and Heinz Nixdorf Recall Study investigators for study design and continued operation.
We also thank C. Egger and Y. D’Elia for the valuable support in data management and data administration; S. Melville and M. Facheris for the important work of drug classification in the GenNOVA project; and the primary care practitioners R. Stocker, S. Waldner, T. Pizzecco, J. Plangger, U. Marcadent and the personnel of the Hospital of Silandro (Department of Laboratory Medicine) for their participation and collaboration in the GenNOVA project. ARIC is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, N01-HC-55022, R01HL087641, R01HL59367 and R01HL086694; National Human Genome Research Institute contract U01HG004402; and National Institutes of Health contract HHSN268200625226C. Infrastructure was partly supported by Grant Number UL1RR025005, a component of the National Institutes of Health and NIH Roadmap for Medical Research. In addition, we acknowledge support from NHLBI grants HL86694 and HL054512, and the Donald W. Reynolds Cardiovascular Clinical Research Center at Johns Hopkins University for genotyping and data analysis relevant to this study. A.K. is supported by a German Research Foundation Fellowship.
The KORA study was funded by grants by the German Federal Ministry of Education and Research (BMBF) in the context of the German National Genome Research Network (NGFN), the German National Competence network on atrial fibrillation (AFNET) and the Bioinformatics for the Functional Analysis of Mammalian Genomes program (BFAM) by grants to S.K. (NGFN 01GS0499, 01GS0838 and AF-Net 01GI0204/N), A.P. (NGFN 01GR0803, 01EZ0874), H.–E.W. (NGFN 01GI0204) and to T.M. (NGFN 01GR0103). S.K. is also supported by a grant from the Fondation Leducq. The KORA platform is funded by the BMBF and by the State of Bavaria.
The SardiNIA team was supported by Contract NO1-AG-1-2109 from the National Institute on Aging and in part by the Intramural Research Program of the US National Institute on Aging, NIH. The efforts of G.R.A. were supported in part by contract 263-MA-410953 from the National Institute on Aging to the University of Michigan and by research grants from the National Human Genome Research Institute and the National Heart, Lung, and Blood Institute (to G.R.A.). The GenNOVA study was supported by the Ministry of Health of the Autonomous Province of Bolzano and the South Tyrolean Sparkasse Foundation. The Heinz Nixdorf Recall Study was funded by a grant of the Heinz Nixdorf Foundation (Chairman: G. Schmidt).
COMPETING INTERESTS STATEMENT
The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturegenetics/.
Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/
Note: Supplementary information is available on the Nature Genetics website.
AUTHOR CONTRIBUTIONSDesigned the scientific rationale of the work: A.P., S.S., D.E.A., R.J.P., S.S.N., A.A.H., D.S., T.M., M.U., J.C., S.K., G.R.A., A.C. Obtained funding: A.P., R.E., K.-H.J., S.N., G.S., A.A.H., E.L., H.-E.W., D.S., E.B., T.M., M.U., J.C., S.K., G.R.A., A.C. Analyzed EKG readings: A.P., S.S., D.E.A., C.F., M.O., S.P., M.F.S., B.M.-M., C.H., S.S.N., S.M., F.M., E.L. Performed genotyping: A.P., S.S., D.E.A., T.W.M., A.A.H., P.P.P., T.M., A.C. Performed genotype imputation: A.P., S.S., D.E.A., M.M., C.P., B.P., T.W.M., B.M.-M., A.A.H. Performed statistical analysis: A.P., S.S., D.E.A., M.M., C.F., C.P., G.B.E., C.G., F.M. Performed metaanalysis: A.P., S.S., D.E.A., V.G., M.M., C.P. Drafted manuscript: A.P., S.S., D.E.A., G.R.A., A.C. Critically revised manuscript: M.M., M.O., A.K., G.U., M.B., M.L., A.S., M.F.S., C.H., T.W.M., M.D., S.M., L.C., R.E., K.-H.J., S.N., G.S., F.M., A.A.H., B.M.-M., P.P.P., H.-E.W., D.S., E.B., T.M., M.U., J.C.
List of investigators by cohort:
ARIC (Atherosclerosis Risk in Communities Study): D.E.A., G.B.E., A.K., M.B., M.L., R.J.P., W.H.L.K., E.B., J.C., A.C.
GenNOVA: F.M., C.P., C.F., A.A.H., P.P.P.
KORA (Kooperative Gesundheitsforschung in der Region Ausgsburg): A.P., M.M., S.P., M.F.S., C.H., C.G., B.P., B.M.-M., G.S., S.K., H.-E.W., T.M.
HNR (Heinz Nixdorf Recall Study): T.W.M., S.M., R.E., K.-H.J.
SardiNIA: S.S., V.G., M.O., G.U., A.S., S.N., M.D., L.C., S.N., E.L., D.S., M.U., G.R.A.