Integration of Genetics into a Systems Model of Electrocardiographic Traits Using HumanCVD BeadChipClinical Perspective
Background—Electrocardiographic traits are important, substantially heritable determinants of risk of arrhythmias and sudden cardiac death.
Methods and Results—In this study, 3 population-based cohorts (n=10 526) genotyped with the Illumina HumanCVD Beadchip and 4 quantitative electrocardiographic traits (PR interval, QRS axis, QRS duration, and QTc interval) were evaluated for single-nucleotide polymorphism associations. Six gene regions contained single nucleotide polymorphisms associated with these traits at P<10−6, including SCN5A (PR interval and QRS duration), CAV1-CAV2 locus (PR interval), CDKN1A (QRS duration), NOS1AP, KCNH2, and KCNQ1 (QTc interval). Expression quantitative trait loci analyses of top associated single-nucleotide polymorphisms were undertaken in human heart and aortic tissues. NOS1AP, SCN5A, IGFBP3, CYP2C9, and CAV1 showed evidence of differential allelic expression. We modeled the effects of ion channel activity on electrocardiographic parameters, estimating the change in gene expression that would account for our observed associations, thus relating epidemiological observations and expression quantitative trait loci data to a systems model of the ECG.
Conclusions—These association results replicate and refine the mapping of previous genome-wide association study findings for electrocardiographic traits, while the expression analysis and modeling approaches offer supporting evidence for a functional role of some of these loci in cardiac excitation/conduction.
Standard measurement methods from the resting ECG trace define quantitative ECG measurements PR interval, QRS axis, QRS duration, and QTc interval (QT interval corrected for heart rate) that predict cardiovascular morbidity and mortality.1–3 These traits predict risk of arrhythmias and sudden death, particularly where there is prolongation of QTc or of QRS durations.4 PR interval, QRS axis, QRS duration, and QTc interval reflect a system of anatomically compartmented electrophysiological events and characteristics underpinned by voltage-gated ion channels, together acting in a cyclical fashion. Cardiac P wave represents atrial depolarization, QRS complex represents ventricular depolarization, and QTc interval represents ventricular repolarization. Some of these electrocardiographic traits have a significant genetic basis,5 with a wide diversity of rare mutations in more than 20 genes known to have monogenic effects on these traits and predispose to arrhythmia (online-only Data Supplement Table I).
Clinical Perspective on p 638
Less is understood of the contributions of common genetic variations that underpin population variation in these electrocardiographic traits. Recent genome-wide association studies (GWAS) have identified associated genomic regions, although the causal variants (or even genes) usually remain unknown from these analyses. The most widely replicated gene from these studies is NO synthase 1 (neuronal) adaptor protein (NOS1AP),6–10 but the identity and number of functional single-nucleotide polymorphisms (SNPs) in this gene remain unresolved. Following GWAS with SNP replication, it is proposed that gene-centric SNP panels that more densely capture the genetic architecture for specific loci, as well as representing important candidate genes, may help to refine association signals within candidate genes and also permit follow-up in a wider variety of cohorts and traits.11 For cardiovascular risk genes, the Illumina HumanCVD BeadChip was constructed for this purpose.11 In this study, we describe a meta-analysis of PR interval, QRS axis, QRS duration, and QTc interval using HumanCVD BeadChip data for 3 cohorts (n=10 526). For each of the main loci identified, we model the genotypic difference in gene expression that would be consistent with the phenotypic effects observed in our analysis. Ultimately, a systems model should be able to integrate genotypic, expression, conductance, electrophysiological, and clinical data.
Materials and Methods
Association analyses were carried out using 10 526 participants from 3 population-based cohorts: British Women’s Heart and Health Study (BWHHS), Genetic Regulation of Arterial Pressure of Humans in the Community (GRAPHIC), and the Whitehall II Study. Full details of these cohorts and collection of ECG data and blood samples for genetic analyses are presented in the online-only Data Supplement Materials and Methods.
Standard 12 lead ECGs were recorded on either Siemens 460 electrocardiographs (Camberley, Surrey, UK) or Burdick Eclipse (Manchester, UK) or Atria (Manchester, UK) models. Digital ECG data were transferred to the University of Glasgow ECG Core Lab at Glasgow Royal Infirmary and analyzed by the University of Glasgow ECG analysis program.12,13 This software meets all of the required specifications in terms of measurement accuracy and is used widely in various commercial products. All ECGs were reviewed manually, and technically unsatisfactory ECGs were excluded (eg, reversed limb lead connections, excessive artifact, etc). The measurements used in this study were: PR interval, QRS axis, QRS duration, and QTc interval (corrected for heart rate using the Hodges equation: QTc=QT+1.75×[RATE−60] ms).14 Outliers >3 SDs from the mean in each study were excluded from the analysis.
Genotyping and Quality Control
All 3 cohorts used the Illumina HumanCVD BeadArray (Illumina Inc, San Diego, CA), which comprises ≈50 000 SNPs in ≈2100 genes systematically selected as cardiovascular disease candidates by an international consortium.11 Genotypes were called using Illumina BeadStudio (v3) Genotyping Module (based on GenCall application, which incorporates the GenTrain clustering algorithm). Samples with genotype call rate <90% were removed from the analysis. SNPs were included in the association analysis if they satisfied the following criteria: SNP call rate ≥0.95, minor allele frequency >1%, and Hardy-Weinberg Equilibrium P>1×10−4. Based on HumanCVD principal components analysis, there is no evidence for significant population stratification in these 3 cohorts (this finding is consistent with self-reported ancestry).
SNP Association Analyses
Separate within-cohort linear regression analyses (n=3 cohorts) were performed for each trait using an additive genetic model relating the trait to genotype dosage (0–2 copies of the minor allele) for each SNP, adjusting for age, sex, body mass index, and systolic blood pressure (after correction for antihypertensive medication).15 Covariates were selected based on prior knowledge of physiological measurements affecting the ECG. In GRAPHIC, additional adjustment for age (using age2 because of the 2 generational structure of the cohort) and familial correlation (using Generalized Estimating Equations with an exchangeable correlation structure) was made to take into account the family structure. After verifying strand alignment, a meta-analysis of the results from the 3 studies was conducted using a fixed-effects model. A comparison with a random-effects model (DerSimonian-Laird)16 was also performed and showed similar results. Heterogeneity (measured using I2 17) was variable between SNPs. We adopted a P value threshold of 1×10−5 for the reporting of putative associations but only discuss those with a P value below 1×10−6 in detail. The 1×10−5 threshold represents a trade-off between maintaining a low false-positive rate and taking into account the higher prior odds of association that a selected, gene-centric approach offers over unbiased GWAS,11 as SNPs represented on the array were selected based on prior knowledge of cardiovascular disease loci. Based on a linkage disequilibrium (LD) cut-off of r2=0.2 in Whitehall II, there are ≈19 000 independent SNPs in the HumanCVD array. A Bonferroni correction based on 19 000 SNPs gives a P=0.05 equivalent of P=2.6×10−6. Association results between these 2 thresholds (1×10−6<P<1×10−5) should be interpreted with caution but are included for information. At each locus with >1 significantly associated SNP, an adjusted association analysis was performed to identify independent effects from the lead SNPs. For each locus identified, the lead SNP was added to the regression model as a covariate to identify additional SNPs that passed our significance threshold.
Expression Quantitative Trait Loci Analysis
The Advanced Study of Aortic Pathology (ASAP) consisted of patients undergoing aortic valve and aorta replacement surgery at the Karolinska University Hospital, Stockholm, Sweden.18 In this study, samples from heart (needle biopsy of left ventricle), aorta intima-media, and aorta adventitia were investigated. Altogether 399 samples from 215 different patients were included. RNA was extracted as described by Folkersen et al18 and hybridized onto Affymetrix GeneChip Human Exon 1.0 ST expression arrays to determine gene expression levels. Cel files were preprocessed in Affymetrix Power Tools (1.10.2) and normalized using Robust Multichip Average Normalization19 (with measurements log2 transformed as part of this process). The preprocessing and analysis are described in reference 18. DNA samples from ASAP patients were genotyped using Illumina Human 610 W-Quad Beadarrays and imputed using the MACH algorithm20 and data from the 1000 genomes project.
All SNPs associated with electrocardiographic traits at P<1×10−5 were investigated for expression quantitative trait loci (eQTL) effects with the closest gene and fold-change of gene expression per minor allele estimated using an additive linear model, that is, values >1 indicate increased expression with the minor allele of an SNP, and values <1 indicate decreased expression with the minor allele of an SNP. A false discovery rate of 5% for the 21 reported tests corresponds to an uncorrected P≤0.00532.
We applied an established model of electrical action potentials of human ventricular cells (see reference 21 for description and validation of this model) to simulate conduction of excitation wave across the transmural strand of human ventricle.22 The algorithm of Shaw and Rudy23 was used to simulate the effects of changes of ion channel expression or activity on electrocardiographic parameters. We used our data except for SCN10A, KCNE1, and KCNJ, which were not represented in HumanCVD BeadChip; for these we used the replication datasets as in references24 and 9. We modeled the effects of changes of ion channel expression on QRS duration and QTc interval but did not model PR interval as the model lacks consideration of atria. Eight distinct ion currents influencing the cardiac action potential25 and hence electrocardiographic traits have been considered. These currents are summarized in relation to the 4 phases of the cardiac action potential in Shaw and Rudy’s Figure 1.23
We assumed that most genotypic effects would be through an effect on expression level because there are no explanatory protein variants in most instances. Using the observed magnitude of genotypic effects to estimate likely size of molecular effect, we can read off from the model what the percentage change in current should be. Assuming a quantitative linear correspondence, this enables induction of the predicted magnitude of expression difference according to genotype.
Association analyses were performed on 10 526 participants; 3443 from the BWHHS, 2024 from GRAPHIC, and 5059 from the Whitehall II study. The cohort characteristics are presented in Table 1, and the correlations of phenotypes and covariates are presented in online-only Data Supplement Table II. The phenotypes were all approximately normally distributed (online-only Data Supplement Figure I). Quantile–quantile plots with Genomic Inflation Factor for each trait in each cohort are presented in online-only Data Supplement Figure II.
Association With Electrocardiographic Traits
In a fixed-effects meta-analysis across 4 phenotypes, 6 independent genomic regions were identified containing associated SNPs at a threshold of P<10−6 (top hits at each locus in Table 2, regional association plots in Figure 1, and full details of associated SNPs in online-only Data Supplement Table IV). In these data, the regression coefficient indicates change in phenotype (seconds) per minor allele of an SNP. SNPs in the gene for sodium channel, voltage-gated, type V, alpha subunit (SCN5A) were associated with both PR interval (top hit rs7372712, P=8.08×10−12) and QRS duration (top hit rs7374540, P=5.87×10−9). These 2 SNPs are essentially not in LD (r2=0.02 in HapMap Europeans). PR interval SNP rs7372712 is located in an LD block in the 5’ region of SCN5A.
SNP rs3807989 in the CAV1/CAV2 genomic region was also associated with PR interval (P=1.99×10−8), with all other significant associations in this region attributable to LD with rs3807989. In contrast with SNPs in SCN5A, SNPs in the CAV1/CAV2 locus did not exhibit any significant association with QRS duration. Conversely, in the CDKN1A gene region, SNP rs3176326 displayed evidence of association with QRS duration (P=1.41×10−7) but little evidence of association with PR interval (P=0.069).
For QTc interval, we observed 56 SNPs in NOS1AP associated at P<1×10−6, with adjusted analyses (see below) suggesting there may be >1 independent effect.
Nonsynonymous SNP rs1805123 (K899T) in KCNH2 was significantly associated with QTc interval (P=8.68×10−7, beta coefficient=−0.00174). However, the most significantly associated SNP in this region was rs3815459, an intronic SNP (P=9.44×10−9, beta coefficient=0.00219).
In KCNQ1, SNP rs12271931 was significantly associated with QTc interval (P=3.17×10−11, beta coefficient=−0.00292).
Analysis adjusted for the lead SNP effect at each locus (Figure 2) showed no strong evidence of additional independent SNP effects at the CDKN1A locus. For QRS duration, we observed independent signals for rs6797133 (P=1.42×10−7), rs7624535 (P=1.02×10−7), rs11710077 (P=6.75×10−4), and rs1805126 (P=7.59×10−4) adjusted for the top hit rs7374540 in SCN5A. For PR interval, we analyzed SNPs in SCN5A adjusted for the top hit rs7372712 and observed evidence of independence for rs7374540 (P=2.65×10−11), rs12053903 (P=1.52×10−10), and rs1805126 (P=2.00×10−7). For QTc interval, the strongest independent signal came from rs10918594 (P=4.00×10−8), with a number of other SNPs showing evidence of association in the analysis adjusted for rs12039600 in NOS1AP. We analyzed the SNPs in KCNH2 adjusted for our top hit rs3815459 and observed some evidence of independent effects from rs6947240 (P=3.83×10−4) and rs1805123 (P=6.74×10−4).
SNPs with strong evidence of association with electrocardiographic traits in our meta-analysis were investigated for their impact on expression of genes in cardiac tissue (Table 3). The most associated SNPs for each trait (rs12039600 in NOS1AP for QTc interval; rs1934968 in CYP2C9, and rs2132570 in IGFBP3 for QRS axis; rs7372712 in SCN5A and rs3807989 in CAV1 for PR interval) show evidence of association with expression levels (Table 3 and online-only Data Supplement Figure III).
Comparison with Mendelian Conduction Disorders
We compared our results with those reported for single-gene arrhythmia disorders and GWAS results, and we present a review of rare and common variation influencing electrocardiographic traits in online-only Data Supplement Table I. There is no evidence of a role of common variation at the population level for a number of genes in which rare mutations have reported pathology (within the limitations of HumanCVD coverage). We confirm previously reported findings from a number of GWAS for both Mendelian disorder genes (eg, SCN5A) and genes for which no Mendelian disorder is known (eg, NOS1AP).
We modeled the effects of changes of ion channel expression or activity on electrocardiographic parameters for comparison with the observed genotypic effects on these traits. Figure 3 illustrates the channels and genes considered, with magnitude of phenotypic effect shown in milliseconds/allele. In Table 4, we show estimates of percentage changes in expression, which would be necessary to achieve the percentage changes in each specific current consistent with the observed phenotypic effects of SNPs in the relevant current-influencing gene. This table is based on an explicit assumption that links changes in expression level to current magnitude (note that for NOS1AP other channels may also be influenced by NO).26 Online-only Data Supplement Figure IV provides illustrative data on the predicted effect of INa change on QRS duration. The genotypic effects we observe on QRS duration (SCN5A and SCN10A) are consistent with an 11% to 12% predicted difference in expression per allele, whereas for QTc interval the genotypic effects are consistent with anywhere from a 1% per allele (KCNE1) to a 10% per allele (NOS1AP) difference in expression.
Using the HumanCVD BeadChip we identified 6 regions containing SNPs with evidence of association with electrocardiographic traits in cohorts representative of the general population, broadly consistent with previous reports.9,10,27 The high density of SNPs in HumanCVD BeadChip for some of these regions offers additional insight into possible causal sites or haploblocks around these genes.
SNPs in the gene for the voltage-gated, sodium channel type V, alpha subunit (SCN5A) gave highly significant signals for both PR interval and QRS duration. The most significant SNPs, which were not in LD (r2=0.02 in HapMap Europeans), differed for each trait, respectively, rs7372712 and rs7374540 (online-only Data Supplement Figure V). SNP rs7372712, located in an LD block in the 5’ region of SCN5A, appears to mark a novel independent effect poorly represented in earlier chip-based analyses. Additionally, rs7372712 is essentially in linkage equilibrium with the coding SNPs in SCN5A claimed to influence electrocardiographic traits, specifically rs1805126 (D1819) and rs1805124 (H558R) in Europeans (HapMap r2=0.003 and r2=0.011, respectively). In a separate gene-focused study, Newton-Cheh et al10 reported SCN5A rs11720524, an intron 1 SNP, as the highest ranking marker for association of SCN5A genotype with sudden cardiac death. There is weak LD (r2=0.112) between rs7372712 and rs11720524, and these 2 SNPs colocalize at the 5’ end of SCN5A (intron 1, 10.8kb apart). This finding indicates that variation in the 5’ region of SCN5A may exert considerable influence on both electrocardiographic traits, being our leading SNP for PR duration, and clinical outcomes.
SNP rs12053903 in SCN5A was the second most significant SNP for PR interval and corresponds with previously reported SNP effects.27,28 These SNPs in the 3’ region of SCN5A are in 1 LD block and may represent the same functional mechanism or (unknown) causal SNP. Our data suggest independent effects from variants in the functionally related SCN10A.27,29 SCN5A SNPs are also reportedly associated with QTc interval; the same SNP and 3’ SCN5A block that we also observed to affect PR duration has been associated with QTc interval.10
We replicated a PR interval specific association of rs380798927 located in the genomic region of CAV1 and CAV2 (P=1.99×10−8). All other significant SNP associations in this region were explicable by their linkage disequilibrium with this SNP. Caveolins act as scaffolds in caveolae, cell membrane pits found in almost all myocardial cell types. Receptors, ion channels, and endothelial nitric oxide synthase localize at caveolae, thus influencing signaling and excitation–contraction coupling. Altered caveolin activity alters NO modulation of membrane excitability.30 This genetic effect, as for NOS1AP on QTc interval (see below), may therefore involve NO, but equally plausibly could involve ryanodine receptors and calcium or sodium channels.31 A systems view of the NO system is presented in the online-only Data Supplement.
SNPs in CDKN1A were uniquely associated with QRS duration consistent with previous reports.24,32 It is likely that our lead SNP marks the same effect as previously reported. CDKN1A encodes cyclin-dependent kinase inhibitor 1A, which is involved in cell cycle regulation.33 The mechanism by which a cell cycle gene could differentially influence QRS duration relative to other electrocardiographic traits remains obscure but may reflect an influence on ventricular development or remodeling, given that the QRS complex represents ventricular depolarization. Alternatively, it could influence ventricular depolarization through an effect on peripheral circulatory resistance.
Numerous studies have reported association of common variation in NOS1AP with QTc interval, and our results are in broad accord with data that suggest 3 independent effects.10 Other SNPs in our data and reported in the literature do not appear to be independent of these. NOS1AP encodes carboxyl-terminal PDZ ligand of neuronal nitric oxide synthase protein, a protein that binds neuronal NO synthase (NOS1) via a C-terminal binding domain and other proteins such as Dexras1 and synapsins via an N-terminal phosphotyrosine-binding domain. In the heart and ventricular myocytes, carboxyl-terminal PDZ ligand of neuronal nitric oxide synthase protein interacts with NOS1 to accelerate cardiac repolarization by inhibition of L-type calcium channels, which generate a slow depolarizing current.26 NOS1AP association plots for all 4 traits are presented in online-only Data Supplement Figure VI.
At the KCNH2 locus, we replicated findings for a nonsynonymous SNP, rs1805123 (K899T) widely reported for association with QTc and with electrophysiological effects in vitro.34–38 We also replicated findings for SNPs emergent from published GWAS,9,10,24,29 where our best proxy was rs6972137 (r2=0.083 with rs1805123 in HapMap Europeans). Our lead SNP effect, rs3815459 (intronic), is inconsistent with rs1805123 being the only functional site in KCNH2. Our data support reports of independent effects in KCNH2.10 We note that 1 overarching LD block spans both KCNH2 and NOS3 (online-only Data Supplement Figure VI), separated by ≈10 kb, and speculate that intronic or intergenic SNPs ascribed to KCNH2 could alternatively mark effects of NOS3 (also a plausible candidate for effects on QTc, particularly considering the recent discovery of unexpected NOS1AP association with QTc).10
Our results also broadly replicate earlier observations for KCNQ1.9,24 Numerous reports have implicated KCNQ1 variation in type 2 diabetes mellitus risk39–42 and also in association with adrenaline, adenosine diphosphate,43 and alpha2-macroglobulin levels.44 The diabetes mellitus SNP is not represented in our data; however rs179429, reportedly43 associated with adrenaline levels, is present but showed no association with QTc. It seems unlikely that KCNQ1 effects on QTc could be explained by indirect effects through adrenaline level rather than by direct effects on cardiac outward potassium current.
It is notable that common variants are observed affecting 5 of the 8 or so ionic currents (Figure 3). We observe a strong correspondence between the leading genotypic effects on PR, QRS, and QTc intervals and the major genes encoding products influencing those traits (eg, INa: SCN5A on PR and QRS duration, IKr: KCNH2 on QTc, and IKs: KCNQ1 on QTc). CACNA1C was represented by 193 SNPs in the HumanCVD array, but none showed any evidence of association (all P>0.05). However, NOS1AP, which influences ICa(L), showed strong evidence of association with QTc interval. From a modeling perspective (online-only Data Supplement Figure IV), percentage changes in QRS duration are an order of magnitude greater per percent change in INa (eg, through possible differential level of expression according to genotype). In Table 4 it is evident that remarkably small (down to 1%) effects can be detected for QTc interval compared with PR and QRS duration, possibly reflecting the relatively smaller SD of QTc. There is no evidence of effect on QTc for SCN5A SNPs with prominent effects on QRS and PR duration. From the predicted QRS duration based on percentage change in INa, the observed ≈1 ms/allele effect of SCN5A rs7374540 on QRS duration would correspond with 12% change in INa, possibly represented as a 12% difference of SCN5A allelic expression were INa to directly correlate with SCN5A expression. Such a difference would be predicted to exert a small effect on QTc, creating a 0.3% difference per allele. Mean QTc was ≈0.42 seconds, which predicts a beta of ≈0.001; however, we did not observe this finding in our association results.
Generally, the various potassium and calcium channel currents have little impact on QRS duration, thus the genotypes influencing QTc are unlikely to show detectable effects on QRS duration. We also cannot readily integrate CAV1/2 or CDKN1A to this model except to note CAV1/2 interaction with channels, receptors, and NO signaling. Finally, our assumption of a quantitative linear correspondence between expression and phenotypic effect is a potential limitation of the approach.
We tested the top associated SNPs for each gene (Table 2) for differential expression by genotype. rs12039600 in NOS1AP, rs7372712 in SCN5A, rs2132570 in IGFBP3, rs1934968 in CYP2C9, and rs3807989 in CAV1 all showed evidence of eQTL effects (Table 3), with per-allele effects of 4%, 5%, 14%, 5%, and 9%, respectively. Because the SNP associations with electrocardiographic traits do not appear to reflect coding variants, the ECG effects may operate through expression levels. Our findings support this hypothesis for these SNPs. rs12039600 is in intron 1 of NOS1AP, and rs7372712 is in the 5’ region of SCN5A. These may therefore mark a haplotypic effect on allelic expression in the promoter regions of these genes. In our eQTL analyses for NOS1AP rs12039600, a 4% difference of allelic expression was observed. The G allele leads to lower expression, which would imply less inhibition of the depolarizing L-type calcium channel, leading to a longer QTc interval. This finding is consistent with the observed G allele association with QTc. The IGFBP3, CYP2C9, and CAV1 eQTL effects were greater than might be anticipated from their regression coefficients (eg, 9% effect per allele on expression vs. 1% per allele for rs3807989 on PR interval); however, their mechanisms of effect are complex and cannot readily be modeled. It should be noted that we did not find complete consistency of allelic imbalance by SNP across different vascular tissues. Also, eQTL analysis for the ideal tissue (eg, ventricular myocardium for QRS) was not available.
We have replicated across the range of electrocardiographic traits a number of previously reported SNPs. Our data point to a previously unrecognized allele and region effect in SCN5A on PR and QRS durations and confirm multiple allelic effects of NOS1AP on QTc. We also identify relevant eQTL effects in NOS1AP, SCN5A, CYP2C9, IGFBP3, and CAV1. We consider the genetic findings in the light of both biological knowledge and a general systems model of electrocardiographic traits, representing a first step toward an integrative genetic model of the ECG as a complex trait. The impact of specific allelic differences in this model may be of value in predicting the effects of drugs targeting specific gene products influencing electrocardiographic traits.
We acknowledge the work of Shahid Latif, Louise Inglis, Kathryn McLaren, and Jean Watts at the University of Glasgow ECG Core Lab for the processing and analysis of ECGs.
Sources of Funding
The British Women’s Heart and Health Study is supported by funding from the British Heart Foundation (BHF) and the Department of Health Policy Research Programme (England), with HumanCVD genotyping funded by the BHF [PG/07/131/24254]. Recruitment and genotyping of the Genetic Regulation of Arterial Pressure of Humans in the Community (GRAPHIC) Study was funded by the BHF. The GRAPHIC Study is part of the research portfolio supported by the Leicester National Institute for Health and Research Biomedical Research Unit in Cardiovascular Disease. The Whitehall II study has been supported by grants from the Medical Research Council (MRC); BHF; Health and Safety Executive; Department of Health; National Heart Lung and Blood Institute [grant number NHLBI: HL36310] and National Institute on Aging (AG13196), US, National Institutes of Health; Agency for Health Care Policy Research [grant number HS06516]; and the John D and Catherine T MacArthur Foundation Research Networks on Successful Midlife Development and Socio-economic Status and Health. The Advanced Study of Aortic Pathology study was supported by the Swedish Research Council [grant number 12660]; the Swedish Heart-Lung Foundation [grant number 20090541]; the European Commission [FAD, Health F2, 2008 grant number 200647]; and a donation by Fredrik Lundberg. N. Samani and S. Humphries hold Chairs funded by the BHF. Dr Hingorani, S. Humphries, Dr Kivimaki, and Dr Talmud were supported by the BHF [PG07/133/24260, BHFPG08/008] and Dr Whittaker by the BHF [PG07/133/24260]. I. Adeniran was funded by a BHF studentship to Drs Zhang and Casas [FS/08/021]. M. Tobin has been supported by MRC fellowships [G0501942, G0902313]. Dr Gaunt, Dr Lawlor, and I. Day work in a centre funded by the UK MRC [G0600705] and University of Bristol. The views in the article are those of the authors and not necessarily any funding body. All data collection, analyses, and interpretation of results were done independently of any funding body.
Dr Whittaker is employed by GlaxoSmithKline, owns GlaxoSmithKline shares, and holds a UK MRC project grant (G0801414). Dr Kivimaki is principal investigator of National Heart Lung and Blood Institute grant R01HL036310. S. Humphries has 5 grants from the British Heart Foundation and European Union Seventh Framework Programme, has received payment for speaking at the Genzyme Meeting (Amsterdam, November 2011), and is a consultant for StoreGene. The other authors report no conflicts.
The online-only Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGENETICS.112.962852/-/DC1.
↵* These authors contributed equally to this work.
- Received February 8, 2012.
- Accepted October 22, 2012.
- © 2012 American Heart Association, Inc.
- Kestenbaum B,
- Rudser KD,
- Shlipak MG,
- Fried LF,
- Newman AB,
- Katz R,
- et al
- Aarnoudse AJ,
- Newton-Cheh C,
- de Bakker PI,
- Straus SM,
- Kors JA,
- Hofman A,
- et al
- Macfarlane PW,
- Devine B,
- Clark E
- Hodges M,
- Salerno D,
- Erlien D
- Higgins JP,
- Thompson SG,
- Deeks JJ,
- Altman DG
- Folkersen L,
- van’t Hooft F,
- Chernogubova E,
- Agardh HE,
- Hansson GK,
- Hedin U,
- et al
- Bolstad BM,
- Irizarry RA,
- Astrand M,
- Speed TP
- ten Tusscher KH,
- Noble D,
- Noble PJ,
- Panfilov AV
- Shaw RM,
- Rudy Y
- Chang KC,
- Barth AS,
- Sasano T,
- Kizana E,
- Kashiwakura Y,
- Zhang Y,
- et al
- Feron O,
- Balligand JL
- Bezzina CR,
- Verkerk AO,
- Busjahn A,
- Jeron A,
- Erdmann J,
- Koopmann TT,
- et al
- Pfeufer A,
- Jalilzadeh S,
- Perz S,
- Mueller JC,
- Hinterseer M,
- Illig T,
- et al
- Newton-Cheh C,
- Guo CY,
- Larson MG,
- Musone SL,
- Surti A,
- Camargo AL,
- et al
- Takeuchi F,
- Serizawa M,
- Yamamoto K,
- Fujisawa T,
- Nakashima E,
- Ohnaka K,
- et al
Electrocardiographic traits are substantially heritable determinants of risk of arrhythmias and sudden cardiac death. We describe a genetic association analysis of PR interval, QRS axis, QRS duration, and QTc interval in 3 population cohorts. Genotyping was performed using the HumanCVD high-density array, and electrocardiographic traits were derived from digital electrocardiographic data using the Glasgow ECG analysis program. Our analyses confirmed a number of previously reported genetic associations and identified a novel independent genetic locus near the voltage-gated sodium channel gene SCN5A. Genetic associations were followed up by analysis of the effects of these loci on gene expression in heart tissue and by modeling the expected effects on electrocardiographic parameters using an established model of electrical action potentials of human ventricular cells. These analyses offered additional insight into the molecular mechanisms underlying variability in electrocardiographic traits. The findings contribute toward the development of an integrated description of the molecular system and common human variation underpinning electrocardiographic pathophysiology.