Journal home
Advance online publication
Current issue
Press releases
Free Association (blog)
Guide to authors
Online submissionOnline submission
For referees
Free online issue
Contact the journal
Reprints and permissions
About this site
For librarians
NPG Resources
Nature Biotechnology
Nature Cell Biology
Nature Medicine
Nature Methods
Nature Reviews Cancer
Nature Reviews Genetics
Nature Reviews Molecular Cell Biology
Nature Conferences
RNAi Gateway
NPG Subject areas
Clinical Medicine
Drug Discovery
Earth Sciences
Evolution & Ecology
Materials Science
Medical Research
Molecular Cell Biology
Browse all publications
Nature Genetics 38, 644 - 651 (2006)
Published online: 30 April 2006; | doi:10.1038/ng1790

A common genetic variant in the NOS1 regulator NOS1AP modulates cardiac repolarization

Dan E Arking1, 2, 16, Arne Pfeufer3, 4, 16, Wendy Post2, 5, W H Linda Kao5, Christopher Newton-Cheh6, 7, 8, Morna Ikeda1, Kristen West1, Carl Kashuk1, Mahmut Akyol3, 4, Siegfried Perz9, Shapour Jalilzadeh3, 4, Thomas Illig10, Christian Gieger10, Chao-Yu Guo6, 11, Martin G Larson6, 11, H Erich Wichmann10, 12, Eduardo Marbán2, Christopher J O'Donnell6, 7, 8, Joel N Hirschhorn7, 13, 14, Stefan Kääb15, Peter M Spooner2, Thomas Meitinger3, 4 & Aravinda Chakravarti1

1 McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA.

2 Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA.

3 Institute of Human Genetics, Technical University Munich, D-81675 Munich, Germany.

4 Institute of Human Genetics, GSF National Research Center of Environment and Health, D-85764 Neuherberg, Germany.

5 Department of Epidemiology, Johns Hopkins University, Bloomberg School of Public Health, Baltimore, Maryland 21205, USA.

6 National Heart, Lung, and Blood Institute's Framingham Heart Study, Framingham, Massachusetts 01702, USA.

7 Broad Institute of Harvard and MIT, Cambridge, Massachusetts 02139, USA.

8 Cardiology Division, Massachusetts General Hospital, Boston, Massachusetts 02114, USA.

9 Institute of Medical Informatics, GSF National Research Center of Environment and Health, D-85764 Neuherberg, Germany.

10 Institute of Epidemiology, GSF National Research Center of Environment and Health, D-85764 Neuherberg, Germany.

11 Department of Mathematics and Statistics, Boston University, Boston 02215, Massachusetts, USA.

12 Institute of Information Management, Biometry and Epidemiology, Ludwig-Maximilians-Universität, D-81377 Munich, Germany.

13 Divisions of Genetics and Endocrinology and Program in Genomics, Children's Hospital, Boston 02115, Massachusetts, USA.

14 Department of Genetics, Harvard Medical School, Boston, Massachusetts 02115, USA.

15 Department of Medicine I, Ludwig-Maximilians-Universität, D-81377 Munich, Germany.

16 These authors contributed equally to this work.

Correspondence should be addressed to Aravinda Chakravarti

Extremes of the electrocardiographic QT interval, a measure of cardiac repolarization, are associated with increased cardiovascular mortality. We identified a common genetic variant influencing this quantitative trait through a genome-wide association study on 200 subjects at the extremes of a population-based QT interval distribution of 3,966 subjects from the KORA cohort in Germany, with follow-up screening of selected markers in the remainder of the cohort. We validated statistically significant findings in two independent samples of 2,646 subjects from Germany and 1,805 subjects from the US Framingham Heart Study. This genome-wide study identified NOS1AP (CAPON), a regulator of neuronal nitric oxide synthase, as a new target that modulates cardiac repolarization. Approximately 60% of subjects of European ancestry carry at least one minor allele of the NOS1AP genetic variant, which explains up to 1.5% of QT interval variation.
The electrocardiographic (ECG) QT interval, a measure of cardiac repolarization, is a genetically influenced quantitative trait with approx30% heritability1, 2, 3. The QT interval has considerable medical relevance, as both high and low values are associated with increased risk of cardiovascular morbidity and mortality4, 5, 6, 7, 8, 9, 10. Moreover, extremely long or short QT intervals occur in a heterogeneous collection of mendelian disorders (long-QT syndrome (LQTS) and short-QT syndrome (SQTS)) and are usually due to rare, highly penetrant mutations in ion channel genes and are associated with increased risk of sudden cardiac death (SCD)11.

Despite major advances in understanding the etiology of cardiovascular disease and concomitant reduction in cardiovascular disease–related mortality, the incidence of SCD has remained largely unchanged. Familial clustering of SCD has been observed, but the vast majority of subjects who are at risk for SCD do not have mutations in the known genes for LQTS or SQTS. To date, common variants in these genes have been associated with disease in only a few studies in selected populations12, 13. Thus, the major genetic mechanisms by which an altered QT interval and other arrhythmogenic conditions may contribute to SCD risk remain incompletely characterized. To identify previously unknown genetic pathways that may modulate SCD risk, we examined the QT interval directly, as opposed to the SCD phenotype. Several reasons prompted this choice: (i) the QT interval is a genetically modulated intermediate trait for SCD with modest heritability; thus, genes affecting the QT interval have been implicated in the pathogenesis of SCD; (ii) the QT interval is a quantitative trait that can be accurately and reliably measured in large samples from standard ECG recordings14; (iii) quantitative rather than qualitative traits are more powerful for genetic analysis and (iv) large populations of uniformly ascertained individuals with SCD are presently unavailable. Notably, the QT interval has been examined in large numbers of healthy volunteers so that findings can be replicated in population-based settings.

To maximize our chance of identifying QT interval genetic factors, we focused on discovering alleles with a large genetic effect or with a genetic effect that could be amplified by study design. Some alleles of large effect have been identified for categorical traits15, 16, 17, 18. For a quantitative trait such as the QT interval, we examined the extremes of the distribution, as these can be expected to be enriched for the alternate alleles at many quantitative trait loci (QTLs), thereby creating a large detectable effect by design and maximizing the statistical power for detecting such alleles19, 20. These models assume that the alleles underlying QTLs are biallelic, common, additive and of small effect but cannot exclude the existence of numerous, rare variants of large effect at the extremes as well. Currently, the impact of rare variants can be comprehensively tested only by DNA sequencing of known candidate genes, as recently shown for low high-density lipoprotein (HDL) cholesterol21. However, common variants can now be efficiently tested by a genome-wide association study, with the added advantage that new pathways can be uncovered.

A multistage design for genome-wide association
Our genetic experiment had three key features. First, we studied subjects from a population-based survey of volunteers aged 25–75 years from Germany (n = 3,966 from the KORA S4 survey22). Second, we performed the genome-wide analysis only on a subset of the 2,001 women from this survey. This strategy was designed to avoid the heterogeneity due to sex in the QT interval22, 23 and because women have a lower prevalence of cardiovascular disease, which could confound QT interval measurement. Third, using a three-stage study design, we attempted to minimize false positive findings yet maximize power and efficiency by examining samples with phenotypic means of decreasing deviations from the population average but with increasing sample size, for SNPs significantly associated in the previous stage (Fig. 1). Multiple-stage designs have been shown to be both powerful and cost-efficient in such settings24. Indeed, simulation studies show that in our specific design, the statistical power to identify a variant that explains 5% of QT interval variation exceeds 40% and 87% for alleles with frequency of 20% and 40%, respectively; for a variant that explains 2% of QT interval variation, the corresponding power is 7% and 22%, respectively, assuming that we are typing either the functional variant or a marker in complete linkage disequilibrium (LD). The power estimates are for each QTL that might exist, and, thus, under the assumption of multiple contributing variants, the power to identify any QTLs of such magnitude is roughly proportional to the number of QTLs. Fourth, to reduce false-positive reporting, we performed replication studies in two population-based samples of European ancestry.

Figure 1. Genome-wide association study of the QT interval.
Figure 1 thumbnail

In stage I, genome-wide genotyping was performed on 100 females from each extreme of QTc_RAS. Two analysis approaches were taken: (i) a genome-wide scan (above the arrows) and (ii) a candidate gene scan (below the arrows), with significance criteria indicated for the following stage. In stage II, an additional 200 females from each extreme were genotyped, and the combined stage I and II samples were analyzed. In stage III, all samples not typed in stages I and II were genotyped, including both males and females, and both combined and stratified analyses were performed.

Full FigureFull Figure and legend (31K)
In stage I, we selected 100 women from each extreme of the QT interval distribution in the KORA S4 cohort, corrected for the covariates known to influence QT interval: heart rate, age and sex (termed 'QTc_RAS'). This selection corresponds to QT intervals below the 7.5 percentile (385.7 plusminus 7.7 ms) or above the 92.5 percentile (444.8 plusminus 3.6 ms). These samples were genotyped using Affymetrix Centurion arrays containing probes for approx115,000 SNPs, with an average heterozygosity of 0.30 and with average spacing of 23.6 kb.

Each of the 88,500 SNPs that passed quality criteria and was polymorphic (see Methods) was tested individually for association with QTc_RAS using a truncated measure test25, 26, under the recessive, dominant and additive genetic models, with the most significant value retained. No single SNP in stage I reached genome-wide significance at the alpha = 5.6 times 10-7 level, based on permutation testing. Nevertheless, for follow-up in stage II, we selected the best SNP from each of the ten most significant loci, all of which showed nominal P values <10-4 (Fig. 2; Supplementary Table 1 online). Approximately nine false positives were expected by chance, so further follow-up was critical. Statistical tests of population stratification confirmed that the observed allele frequency differences between the samples at the extremes were not from inherent population substructure (FST = 0.0009 plusminus 0.0759 across the genome).

Figure 2. Genome-wide significance of QTc_RAS.
Figure 2 thumbnail

The analysis compared 100 females from each extreme of QTc_RAS (stage I). The x-axis is genomic position, and the y-axis is the negative base-10 logarithm of the P value. None of the SNPs showed genome-wide significance, but the top ten positives had P < 10-4.

Full FigureFull Figure and legend (61K)
Based on our current understanding of the biology of cardiac repolarization, we also selected a priori 45 candidate genes that have been implicated in SQTS or LQTS or cardiac cellular electrophysiology or that are homologous to the selected genes (Supplementary Table 2 online) and that each had at least one SNP represented on the array within 10 kb of its 5' or 3' UTR. We used a less stringent significance threshold to choose candidate gene SNPs for follow-up in stage II (P < 0.01), as their prior probability of involvement was higher than that for anonymous markers.

Each SNP selected for follow-up was supplemented with a partially correlated 5' and 3' flanking SNP (see Methods). As we were unlikely to have identified the causal SNPs in stage I, the addition of flanking SNPs in the neighborhood of a positive signal could identify a SNP more strongly associated with QTc_RAS and thus more highly correlated with a causal variant.

In stage II, we included an additional 400 females having QTc_RAS below the 15th (n = 200) or above the 85th percentile (n = 200). All 600 women who had their QTc_RAS means approx2 s.d. apart and a mean trait difference of 45.5 ms were genotyped for SNPs that passed stage I criteria and flanking SNPs. The second stage should lead to fewer false positives, because under the null hypothesis of no association, adding samples will decrease significance and should lead to greater power, as we used a sample size three times larger than the initial screening set while still maintaining a sample enriched with subjects from either tail of the phenotypic distribution. In stage III, we genotyped anonymous SNPs significant at P < 0.005 and candidate gene SNPs with P < 0.01 in stage II in the remaining 3,366 subjects of both genders. Importantly, we performed significance tests separately on the men and women specific to stage III (that is, excluding the 600 females analyzed in stages I and II), allowing stage III to serve as a validation study for stages I and II. This replication approach has comparable power to the joint analysis proposed by others27, given the smaller sample size of the first two stages compared with the third stage and the small proportion of markers chosen for follow-up in the third stage.

Genome-wide analysis identifies NOS1AP
Twelve of 57 SNPs (rs2282428 was in common among the anonymous and candidate gene SNPs) were significant in stage II, representing seven SNPs from four anonymous loci studied at P < 0.005 (Supplementary Table 3 online) and five SNPs from four candidate gene loci at P < 0.01 (Supplementary Table 4 online). Two SNPs, rs945713 (NOS1AP) and rs7341478 (CACNA2D1), showed increased significance in the stage II analysis compared with stage I. Notably, rs10494366, the 5' flanking SNP of rs945713, achieved genome-wide significance (P < 2.57 times 10-8) with a P value three orders of magnitude lower than that for rs945713, the sentinel SNP on the Centurion array, supporting our rationale for incorporating flanking SNPs in stage II (Supplementary Table 3). From each of the eight loci (four anonymous and four candidate genes), we selected the most significant SNP from stage II for validation in stage III (Supplementary Tables 3 and 4); seven of these SNPs, all in Hardy-Weinberg equilibrium, were successfully genotyped in the remainder of the KORA S4 sample.

We identified three loci based on sex-pooled analyses of stage III samples with nominal significance (P < 0.05; Table 1): these loci correspond to rs10494366 at NOS1AP (P < 10-7), rs1559578 at QTc_5.3 (P < 0.004) and rs7341478 at CACNA2D1 (P < 0.024). NOS1AP (CAPON) is the C-terminal PDZ domain ligand to neuronal nitric oxide synthase (nNOS, encoded by the NOS1 gene)28 and affects NMDA receptor–gated calcium influx. It has not been previously suspected to have a role in cardiac repolarization. However, using RT-PCR, we found NOS1AP expression in human left ventricular heart tissue (data not shown). CACNA2D1 encodes an L-type voltage-dependent calcium channel regulatory subunit expressed in the heart29, 30. The third locus, QTc_5.3, does not correspond to a known gene, but rather to a GeneScan31 prediction, so its potential biological relationship to the QT interval is unknown. None of the four remaining loci in stage III was significant in the overall sample, but KCNK1 (P approx 0.005) showed significant effects in females only. KCNK1 (TWIK1) is the weakly inward-rectifying potassium channel subfamily K member 1 and may be involved in the control of background potassium membrane conductance. The gene is transcribed in many tissues but is particularly highly expressed in the brain and heart32. However, given the number of hypotheses tested (seven SNPs, sex-pooled and sex-specific), only the NOS1AP SNP achieves genome-wide significance after correcting for multiple testing. This finding was not dependent on the genetic model, with a similar result obtained from model-free analysis (P < 10-5).

Table 1. Genetic effects of QT interval–associated polymorphisms in 3,966 individuals from the KORA S4 sample
Table 1 thumbnail

Full TableFull Table
Validation in the KORA F3 cohort
We genotyped a separate sample of 2,646 subjects from the KORA F3 for the seven SNPs from stage III (Table 2). Power estimation by permutation testing revealed that for the non–sex specific effects, we had >95% power to replicate a true finding from stage III and >85% power to replicate the sex-specific effect seen in KCNK1. rs10494366 in NOS1AP was highly significant in the replication cohort even after adjusting for the number of SNPs tested (P < 10-10), but no SNP from the other six loci was significant. The nominal significance observed for rs7341478 in CACNA2D1 was actually for an effect in the direction opposite from that observed in the S4 samples, suggesting a false positive association. Nevertheless, these results clearly demonstrate that the multistage genome-wide approach was able to unequivocally identify at least one common variant of a gene previously unrecognized as being associated with cardiac repolarization.

Table 2. Genetic effects of stage III SNPs with QT interval in 2,646 individuals from the KORA F3 replication sample
Table 2 thumbnail

Full TableFull Table
The average genetic effect (delta), measured as the difference in means of QTc_RAS between the two homozygotes, is 4.9 ms for NOS1AP in the total S4 sample and 7.9 ms in the F3 sample, with NOS1AP accounting for 1.2% and 1.9% of the variance, respectively. The minor allele frequency was the same in both populations (36%), and genetic effects were observed in both males and females.

Validation in the Framingham Heart Study
We genotyped SNPs from the seven loci of stage III in 1,805 participants from the Framingham Heart Study (FHS), a population-based sample of predominantly European ancestry (Table 3). We confirmed the association of rs10494366 in NOS1AP with QT interval in the sex-pooled analysis (the prespecified primary test of replication as supported by the S4 and F3 results) with P = 0.004. The average genetic effect for QT interval adjusted for heart rate (RR interval), age and sex was 4.0 ms. In secondary analyses, the effect of rs10494366 on QT interval in FHS was stronger in women (delta = 6.5 ms), with a more modest effect in the same direction in men (delta = 1.1 ms). However, because findings in the KORA S4 and F3 samples did not show a clear difference by sex, the apparent differences in effect between men and women observed in the FHS sample may reflect chance statistical fluctuations, although a small influence of gender in the effect of the NOS1AP SNP cannot be excluded.

Table 3. Genetic effects of stage III SNPs with QT interval in 1,805 individuals from the FHS replication sample
Table 3 thumbnail

Full TableFull Table
Fine mapping of the common variant at NOS1AP
To identify the underlying functional variant(s), we undertook fine-scale association mapping of the NOS1AP locus by genotyping 13 SNPs in the region in the 600 stage II samples; the P values for each SNP by genomic position are plotted in Figure 3. We found the strongest association in the 5' upstream region of the NOS1AP gene, at rs4657139, although the entire region between the SNPs rs10494366 and rs2880058 (approx120 kb) shows strong LD and, consequently, strong association with QT interval. Similar results were observed for a subset of SNPs typed in the entire KORA S4 cohort, including the sentinel SNP from the Affymetrix array (Supplementary Table 5 online). The rapid drop in association between QT interval and SNPs further upstream (toward OLFML2B) strongly suggests that a functional variant in the NOS1AP gene is mediating the observed variation in QT interval. This observation is supported by reference genotyping in the HapMap CEU sample, which demonstrates strong correlation of SNPs in the NOS1AP gene with a significant drop in association 3' of OLFML2B. To identify the mutational site, we sequenced all exons and, in addition, 13 noncoding conserved sequences presumed to have regulatory function (see Methods) localized to this 120-kb region, in ten subjects of each rs4657139 homozygous genotype. We did not find any missense mutations in the exons, but we identified three SNPs associated with QT interval (rs12096347, rs4656349, rs11579080) in the noncoding conserved sequences (Fig. 3), and functional analysis of these SNPs is a high priority. Thus, there is a high likelihood that the NOS1AP functional allele is in noncoding DNA. This observation is consistent with the increasing evidence that regulatory SNPs have a significant role in complex inheritance15.

Figure 3. Fine mapping of the NOS1AP gene.
Figure 3 thumbnail

The lower panel shows pairwise LD between SNPs at NOS1AP. The value within each diamond represents the pairwise correlation between SNPs (measured as D') defined by the top left and the top right sides of the diamond. Diamonds without a number correspond to D' = 1. Shading represents the magnitude and significance of pairwise LD, with a red-to-white gradient reflecting higher to lower LD values. NOS1AP exons 1 and 2 are shown in orange. The upper panel shows significance for each SNP, with genomic position on the x-axis and the negative base-10 logarithm of the P value on the y-axis, indicating that the most likely location of the underlying functional variant is in the 5' region of NOS1AP. SNPs detected by sequencing conserved regions in individuals from the QT interval extremes are shown in red.

Full FigureFull Figure and legend (81K)
Previous attempts to dissect genetic contributions to the QT interval have focused on monogenic LQTS disease genes22 or on family-based linkage studies of the quantitative trait1, 3. The latter approach is well suited to uncovering rare alleles with large to moderate effects. Instead, we have used a genome-wide association study to uncover common polymorphisms of small effect that are capable of explaining a greater degree of the population QT interval variation.

We have identified NOS1AP as a gene that is significantly associated with QT interval variation in a general population of approx4,000 German adults and have replicated this finding in a second sample of approx2,700 adults from the same population and in a third sample of approx1,800 American adults of European ancestry. Differences in effect size and statistical significance across the three populations could reflect methodological differences in measurement (see Methods) or statistical fluctuation around a common effect. SNPs at the other six loci were not confirmed in the replication samples.

These results emphasize that genome-wide association studies, which are not limited by our current understanding of cardiac repolarization biology, can be used to identify common variants that show previously unanticipated genetic associations. Indeed, the involvement of the NOS1AP gene in QT interval variation was unsuspected, yet it explains 1.5% of the variance in our combined sample of approx6,600 German adults and 0.6% in the FHS sample. Although the variant identified explains a small fraction of the total variation in QT duration, the identification of a gene not previously known to be implicated in myocardial repolarization opens up a completely new area for QT interval biology and brings to attention drug targets that could be of benefit to SCD patients. NOS1AP is a regulator of neuronal nitric oxide synthase effected by forming a ternary complex with PSD95 (membrane-associated guanylate kinase28) and Dexras1 (member of the Ras family of small monomeric G proteins33). Notably, NOS1 has been shown recently to have a role in cardiac contractility34. Consequently, nitric oxide signaling may be an important effector of cardiac repolarization and seems to have a role in balancing nitric oxide and superoxide production35. In addition, several cardiac and neuronal ion channel genes contain PDZ domains capable of binding NOS1AP36, 37, 38. Through such binding, direct modulation of channel activity may occur, or NOS1AP may displace other PDZ-binding regulators of channel expression or function.

Keeping in mind that the ultimate goal is a comprehensive genome-wide scan, it is important to ask how much of the genome our study has missed. The Phase I HapMap data suggests that an efficient set of 100,000 SNPs with minor allele frequency >5% would cover approx70% of the genome for subjects of Northern European origin with an r2 0.8 (ref. 39). The marker set we used has a smaller number of SNPs, and they were not chosen with particular regard to their linkage disequilibrium patterns to neighboring SNPs; thus, we believe that we have covered closer to 50% of the genome. Our scan also did not interrogate a number of gene loci known to be involved in regulating cardiac electrogenesis: for example, the depolarizing sodium channel SCN5A (Supplementary Table 2). Nevertheless, a few conclusions are warranted: (i) we did not find a major QTL (> 5% explained variance); (ii) the NOS1AP SNP explains a larger percentage of the variance than previous findings for the QT interval based on a candidate gene approach22 and (iii) given that the heritability of QT interval is approx30% (refs. 1, 2, 3), this would suggest that many more genes with small effects are likely to be involved. If this effect size holds true for other complex phenotypes, it would suggest that the vast majority of studies are substantially underpowered, and sample sizes will need to be much larger than those currently studied.

Study population.
The KORA S3 and S4 surveys are representative samples from the general population living in or near Augsburg, Germany and were conducted between 1994 and 2004. Consequent to informed consent, each of the surveys sampled subjects from ten strata according to sex (equal ratio) and age (range 25–75 years) with a minimum stratum size of >400 subjects. KORA procedures and samples have been previously described extensively22, 40. Briefly, for sample S4, 4,261 probands were studied between 1999 and 2001, and for sample S3, 4,856 subjects were studied between 1994 and 1995. In 2003 and 2004, 2,974 participants from S3 returned for follow-up (KORA F3). For this analysis, we excluded subjects with atrial fibrillation, pacemaker implant and/or pregnancy. We included 3,966 subjects from S4 for the three-stage genome-wide study, and 2,646 subjects from F3 were used for an independent replication. All studies involving humans were performed according to the declarations of Helsinki and Somerset West and were approved by the local medical ethics committees in Germany and the US.

The FHS, the second replication sample, is a prospective epidemiologic study established in 1948 to evaluate potential risk factors for coronary heart disease. In 1971, 5,124 subjects were entered into the Framingham Offspring Study, including children or spouses of the children of the original cohort. We measured the QT interval in electrocardiograms (ECG) from offspring participants of the Framingham Heart Study examined between 1971 and 1975 who were free of atrial fibrillation or QT-influencing medication and survived to provide DNA collected during 1995–1998. This community-based cohort is predominantly of European ancestry (mean age 36.9 years; 51.4% women).

QT interval measurement.
QT interval in all KORA samples was measured in ms from 10-s, 12-lead digitally recorded resting ECGs (S4: Bioset 9000, Hörmann Medizinelektronik; F3: Mortara Portrait, Mortara) as previously described22. In S4, QT intervals were determined by computerized analysis of an averaged cycle computed from all recorded cycles after exclusion of ectopic beats using the Hannover ECG analysis software (v 3.22-12). The QT interval determined by this algorithm represents the earliest start of depolarization until the latest deflection of repolarization between any two leads. QT measurements over short- and long-term time intervals have been investigated and shown to be highly reproducible14. In F3, QT intervals were determined by the proprietary algorithm implemented in the ECG system41. Absolute measurement values for QT are known to depend strongly on the individual algorithm used, which explains the significant differences in mean QT interval measurements between KORA S4 and F3. In contrast, the relative differences in QT interval between subjects (the measurement relevant to QTL studies) have been shown to be well preserved across ECG measurement platforms42. The raw QT interval measured in the ECG has several significant covariates that need to be normalized to perform genetic studies; we used a multivariate linear regression model including heart rate (RR interval), sex and age. Correction factors were determined separately for each gender, as dictated by our sampling strategy, and the resulting QT interval, corrected for heart rate (R), age (A) and sex (S) was termed QTc_RAS. The correction formulas were as follows:

S4 males:

S4 females:

F3 males:

F3 females:

where RR denotes RR interval in ms.

QT intervals in the FHS sample were measured using digital calipers in leads II, V2 and V5 from digitized electrocardiograms, as previously published3. A single cycle from each lead was regressed for each sex separately on age and RR interval in linear models. QT residuals were standardized to mean 0 and s.d. 1 and averaged across the three leads. These average residuals represent the age, sex and RR-adjusted QT phenotype studied and have a demonstrated heritability of 35% in the FHS sample3. Although minor differences between the QT trait definition exist, adjustment for age, sex and RR interval were comparable, and the association of a variant with NOS1AP using either QT trait definition attests to the robustness of the finding and the applicability of the results to QT intervals measured using multiple methods.

Genome-wide assays and SNP genotyping.
Stage I genome-wide analysis was performed using Affymetrix oligonucleotide arrays containing 115,571 SNPs, which were hybridized with genomic DNA as described43. Genotypes were determined using the software tool GDAS3.0, with a setting of 0.05 for both homozygous and heterozygous genotype calls. Fourteen (7%) arrays with <85% overall genotyping call rates (across all SNPs) as well as 9,616 (8.3%) SNPs with overall genotyping call rates of <85% (across samples) were removed from the data set because their accuracy was 99.5%, as determined by extensive internal validation of repeat samples. We also removed 17,367 SNPs with minor allele frequency (MAF) <2.5%, as they would have no power under any study design. These procedures left us with 186/200 subjects (93.0%) and 88,548/115,571 SNPs (76.6%) for analysis. Additional genotyping was performed in S4, F3 and FHS using either TaqMan Assays on Demand or Assays by Design (Applied Biosystems) or primer extension MALDI-TOF genotyping technology (Autoflex HT, Sequenom), according to the manufacturers' protocols. Sequenom primer sequences used are available in Supplementary Table 6 online.

Testing for population stratification.
All SNPs were tested for Hardy-Weinberg equilibrium using methods previously described44. FST for each SNP, with no missing data in the stage I analysis (11,431 SNPs), was calculated as the complement of the observed to expected heterozygosity.

SNP selection for stage II and III genotyping.
SNPs representing the ten most significant loci from the genome-wide screen and P < 0.01 from the candidate genes were selected for follow-up in stage II. For loci with multiple SNPs in high linkage disequilibrium (LD) (r2 > 0.4), only the most significant SNP was selected. Flanking SNPs were chosen from the International HapMap project39. LD was measured (r2) for each HapMap SNP (genotyped in a sample of 60 independent subjects of Northern European origin) within 500 kb of the target SNP, and one was chosen on each side of the target SNP with an r2 value between 0.4 and 0.8 and MAF 0.2. In those cases where there was no flanking SNP within the r2 limits, the closest SNP with MAF 0.2 was chosen. The SNPs showing P < 0.005 and P < 0.01 from the stage II genome-wide and candidate gene analyses, respectively, were selected for stage III genotyping.

Statistical analyses of genetic effects.
Stage I analysis (200 women), including 100 from each extreme of QTc_RAS, was performed using a truncated measure analysis of variance (ANOVA)25, 26 under additive, dominant and recessive genetic models, with best P values retained. Stage II analysis (600 women) added the next 200 women from each extreme of QTc_RAS and was performed as above. Stage III samples (3,366 subjects comprising both men and women) and the replication sample F3 (2,646 subjects comprising both men and women) were analyzed using standard ANOVA under additive, dominant and recessive genetic models. P values for the best fit model are reported; however, P values are adjusted for the testing of multiple genetic models by permutation tests. Model-free analyses were performed using ANOVA with 2 degrees of freedom. Analyses were performed separately for males and females, as well as for the combined sample. To estimate the variance explained, we calculated R2 using linear regression. Regression analyses were performed using SPSS ver13.0.

The FHS analyses of SNPs at the seven loci identified in stage II involved testing the association of genotypes at these loci with adjusted QT interval duration in 1,805 unrelated FHS participants, using linear regression testing dominant, additive and recessive models (SAS v 8.1). Nominal P values for the best genetic model are reported for the replication samples.

Identification and sequencing of conserved noncoding regions.
Conserved noncoding regions were identified using the phastCons45 track from the University of California, Santa Cruz genome browser with a threshold of lod 25. Automated dideoxy sequencing was performed on an ABI3100 with the BigDye Terminator Sequencing Kit according to the manufacturer's protocol (Applied Biosystems). Primer sequences are available in Supplementary Table 7 online.

Power calculations.
Power was estimated by Monte Carlo simulation. Phenotypes (P) for 2,000 subjects were randomly generated from a normal distribution. To simulate a genetic effect (delta) under an additive model, genotypes were generated under Hardy-Weinberg equilibrium, and phenotypes were simulated with the following means: AA = P + (delta / 2), Aa = P; aa = P - (delta / 2). These subjects were ranked by phenotype and ANOVA was performed on the top and bottom 100 subjects (stage I analysis). If P < 10-4 was obtained, the next 200 ranked subjects at the top and bottom were added and ANOVA performed (stage II). If P < 0.005 was obtained, an additional 2,000 subjects were simulated as above, and ANOVA was performed on the 3,400 subjects not tested in stages I and II (stage III). A positive result was assigned if P < 0.005, which was empirically derived and corresponds to a genome-wide type I error of 0.05. Power to replicate findings in the F3 population (nominal P < 0.05) was determined by simulating 2,700 subjects as described above, using the genetic effect observed in the S4 population.

The International HapMap project39:; University of California, Santa Cruz genome browser: See for further details on mapping in Figure 3.

Note: Supplementary information is available on the Nature Genetics website.


The KORA GWA study was designed by D.E.A., A.P., W.P., W.H.L.K., H.E.W., E.M., S.K., P.M.S., T.M. and A.C. Phenotype assessment was performed by A.P., W.P., S.P., C.G. and S.K. T.I. and H.E.W. were responsible for the management of KORA data and biological samples. Affymetrix genotyping was conducted by K.W., M.I. and D.E.A. Sequenom genotyping and analysis was performed by M.A., S.J. and A.P.; T.M. and A.C. supervised the resequencing and all marker typing. Statistical analyses were performed by D.E.A., A.P., W.H.L.K. and C.K. under the supervision of A.C. The Framingham replication study was designed and carried out by C.N.-C. under the supervision of J.N.H. and C.J.O.; the statistical analyses were conducted by C.-Y.G. under the supervision of M.G.L.

Received 27 February 2006; Accepted 29 March 2006; Published online: 30 April 2006.

  1. Busjahn, A. et al. QT interval is linked to 2 long-QT syndrome loci in normal subjects. Circulation 99, 3161–3164 (1999). | PubMed | ISI | ChemPort |
  2. Carter, N. et al. QT interval in twins. J. Hum. Hypertens. 14, 389–390 (2000). | Article | PubMed | ISI | ChemPort |
  3. Newton-Cheh, C. et al. QT interval is a heritable quantitative trait with evidence of linkage to chromosome 3 in a genome-wide linkage analysis: The Framingham Heart Study. Heart Rhythm 2, 277–284 (2005). | PubMed | ISI |
  4. Schouten, E.G. et al. QT interval prolongation predicts cardiovascular mortality in an apparently healthy population. Circulation 84, 1516–1523 (1991). | PubMed | ISI | ChemPort |
  5. Dekker, J.M. , Schouten, E.G. , Klootwijk, P. , Pool, J. & Kromhout, D. Association between QT interval and coronary heart disease in middle-aged and elderly men. The Zutphen Study. Circulation 90, 779–785 (1994). | PubMed | ISI | ChemPort |
  6. Elming, H. et al. The prognostic value of the QT interval and QT interval dispersion in all-cause and cardiac mortality and morbidity in a population of Danish citizens. Eur. Heart J. 19, 1391–1400 (1998). | Article | PubMed | ISI | ChemPort |
  7. Sharp, D.S. , Masaki, K. , Burchfiel, C.M. , Yano, K. & Schatz, I.J. Prolonged QTc interval, impaired pulmonary function, and a very lean body mass jointly predict all-cause mortality in elderly men. Ann. Epidemiol. 8, 99–106 (1998). | PubMed | ISI | ChemPort |
  8. de Bruyne, M.C. et al. Prolonged QT interval predicts cardiac and all-cause mortality in the elderly. The Rotterdam Study. Eur. Heart J. 20, 278–284 (1999). | Article | PubMed | ISI | ChemPort |
  9. Okin, P.M. et al. Assessment of QT interval and QT dispersion for prediction of all-cause and cardiovascular mortality in American Indians: the Strong heart study. Circulation 101, 61–66 (2000). | PubMed | ISI | ChemPort |
  10. Dekker, J.M. , Crow, R.S. , Hannan, P.J. , Schouten, E.G. & Folsom, A.R. Heart rate-corrected QT interval prolongation predicts risk of coronary heart disease in black and white middle-aged men and women: the ARIC study. J. Am. Coll. Cardiol. 43, 565–571 (2004). | Article | PubMed | ISI |
  11. Priori, S.G. & Napolitano, C. Genetics of cardiac arrhythmias and sudden cardiac death. Ann. NY Acad. Sci. 1015, 96–110 (2004). | Article | PubMed | ChemPort |
  12. Yang, P. et al. Allelic variants in long-QT disease genes in patients with drug-associated torsades de pointes. Circulation 105, 1943–1948 (2002). | Article | PubMed | ISI | ChemPort |
  13. Splawski, I. et al. Variant of SCN5A sodium channel implicated in risk of cardiac arrhythmia. Science 297, 1333–1336 (2002). | Article | PubMed | ISI | ChemPort |
  14. Perz, S. et al. Does computerized ECG analysis provide sufficiently consistent QT interval estimates for genetic research? in Analysis of Biomedical Signals and Images Vol. 17 (eds. Jan, J., Kozumplik, J. and Provaznik, I.) 47–49 (VUTIUM Press, Brno, Czech Republic, 2004).
  15. Emison, E.S. et al. A common sex-dependent mutation in a RET enhancer underlies Hirschsprung disease risk. Nature 434, 857–863 (2005). | Article | PubMed | ISI | ChemPort |
  16. Klein, R.J. et al. Complement factor H polymorphism in age-related macular degeneration. Science 308, 385–389 (2005). | Article | PubMed | ISI | ChemPort |
  17. Haines, J.L. et al. Complement factor H variant increases the risk of age-related macular degeneration. Science 308, 419–421 (2005). | Article | PubMed | ISI | ChemPort |
  18. Edwards, A.O. et al. Complement factor H polymorphism and age-related macular degeneration. Science 308, 421–424 (2005). | Article | PubMed | ISI | ChemPort |
  19. Robertson, A. The nature of quantitative genetic variation. in Heritage from Mendel (ed. Brink, R.A.) 265–280 (Univ. of Wisconsin Press, Madison, Wisconsin, 1967).
  20. Risch, N.J. & Zhang, H. Mapping quantitative trait loci with extreme discordant sib pairs: sampling considerations. Am. J. Hum. Genet. 58, 836–843 (1996). | PubMed | ISI | ChemPort |
  21. Cohen, J.C. et al. Multiple rare alleles contribute to low plasma levels of HDL cholesterol. Science 305, 869–872 (2004). | Article | PubMed | ISI | ChemPort |
  22. Pfeufer, A. et al. Common variants in myocardial ion channel genes modify the QT interval in the general population: results from the KORA study. Circ. Res. 96, 693–701 (2005). | Article | PubMed | ISI | ChemPort |
  23. Bezzina, C.R. et al. A common polymorphism in KCNH2 (HERG) hastens cardiac repolarization. Cardiovasc. Res. 59, 27–36 (2003). | Article | PubMed | ISI | ChemPort |
  24. Satagopan, J.M. , Venkatraman, E.S. & Begg, C.B. Two-stage designs for gene-disease association studies with sample size constraints. Biometrics 60, 589–597 (2004). | Article | PubMed | ISI |
  25. Boerwinkle, E. , Chakraborty, R. & Sing, C.F. The use of measured genotype information in the analysis of quantitative phenotypes in man. I. Models and analytical methods. Ann. Hum. Genet. 50, 181–194 (1986). | PubMed | ISI |
  26. Page, G.P. & Amos, C.I. Comparison of linkage-disequilibrium methods for localization of genes influencing quantitative traits in humans. Am. J. Hum. Genet. 64, 1194–1205 (1999). | Article | PubMed | ISI | ChemPort |
  27. Skol, A.D. , Scott, L.J. , Abecasis, G.R. & Boehnke, M. Joint analysis is more efficient than replication-based analysis for two-stage genome-wide association studies. Nat. Genet. 38, 209–213 (2006). | Article | PubMed | ISI | ChemPort |
  28. Jaffrey, S.R. , Snowman, A.M. , Eliasson, M.J. , Cohen, N.A. & Snyder, S.H. CAPON: a protein associated with neuronal nitric oxide synthase that regulates its interactions with PSD95. Neuron 20, 115–124 (1998). | Article | PubMed | ISI | ChemPort |
  29. De Jongh, K.S. , Warner, C. & Catterall, W.A. Subunits of purified calcium channels. Alpha 2 and delta are encoded by the same gene. J. Biol. Chem. 265, 14738–14741 (1990). | PubMed | ChemPort |
  30. Klugbauer, N. , Marais, E. & Hofmann, F. Calcium channel alpha2delta subunits: differential expression, function, and drug binding. J. Bioenerg. Biomembr. 35, 639–647 (2003). | Article | PubMed | ISI | ChemPort |
  31. Burge, C. & Karlin, S. Prediction of complete gene structures in human genomic DNA. J. Mol. Biol. 268, 78–94 (1997). | Article | PubMed | ISI | ChemPort |
  32. Lesage, F. et al. TWIK-1, a ubiquitous human weakly inward rectifying K+ channel with a novel structure. EMBO J. 15, 1004–1011 (1996). | PubMed | ISI | ChemPort |
  33. Fang, M. et al. Dexras1: a G protein specifically coupled to neuronal nitric oxide synthase via CAPON. Neuron 28, 183–193 (2000). | Article | PubMed | ISI | ChemPort |
  34. Massion, P.B. , Pelat, M. , Belge, C. & Balligand, J.L. Regulation of the mammalian heart function by nitric oxide. Comp. Biochem. Physiol. A Mol. Integr. Physiol. 142, 144–150 (2005). | PubMed | ISI |
  35. Khan, S.A. et al. Neuronal nitric oxide synthase negatively regulates xanthine oxidoreductase inhibition of cardiac excitation-contraction coupling. Proc. Natl. Acad. Sci. USA 101, 15944–15948 (2004). | Article | PubMed | ChemPort |
  36. Murata, M. et al. SAP97 interacts with Kv1.5 in heterologous expression systems. Am. J. Physiol. Heart Circ. Physiol. 281, H2575–H2584 (2001). | PubMed | ISI | ChemPort |
  37. Leonoudakis, D. , Mailliard, W. , Wingerd, K. , Clegg, D. & Vandenberg, C. Inward rectifier potassium channel Kir2.2 is associated with synapse-associated protein SAP97. J. Cell Sci. 114, 987–998 (2001). | PubMed | ISI | ChemPort |
  38. Kim, E. & Sheng, M. Differential K+ channel clustering activity of PSD-95 and SAP97, two related membrane-associated putative guanylate kinases. Neuropharmacology 35, 993–1000 (1996). | Article | PubMed | ISI | ChemPort |
  39. The International HapMap Consortium. A haplotype map of the human genome. Nature 437, 1299–1320 (2005). | Article |
  40. Wichmann, H.E. , Gieger, C. & Illig, T. KORA-gen–resource for population genetics, controls and a broad spectrum of disease phenotypes. Gesundheitswesen 67 (Suppl.), S26–S30 (2005).
  41. Mortara, D.W. Source consistency filtering. Application to resting ECGs. J. Electrocardiol. 25(Suppl.), 200–206 (1992). | Article |
  42. Bailey, J.J. et al. Recommendations for standardization and specifications in automated electrocardiography: bandwidth and digital signal processing. A report for health professionals by an ad hoc writing group of the Committee on Electrocardiography and Cardiac Electrophysiology of the Council on Clinical Cardiology, American Heart Association. Circulation 81, 730–739 (1990). | PubMed | ISI | ChemPort |
  43. Matsuzaki, H. et al. Genotyping over 100,000 SNPs on a pair of oligonucleotide arrays. Nat. Methods 1, 109–111 (2004). | Article | PubMed | ISI | ChemPort |
  44. Weir, B. Genetic Data Analysis II (Sinauer Associates, Cumberland, Massachusetts, 1996).
  45. Siepel, A. et al. Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res. 15, 1034–1050 (2005). | Article | PubMed | ISI | ChemPort |
The authors wish to thank G. Tomaselli, J. Nathans, S. Lin and D.J. Cutler for numerous helpful discussions and D. Levy, E. Benjamin and R. D'Agostino at FHS for contributions to the electrocardiographic QT measurement study. This work was supported in part by the D.W. Reynolds Clinical Cardiovascular Research Center, Johns Hopkins University, the US National Institutes of Health, and the German Federal Ministry of Education and Research (BMBF) both in the context of the program Bioinformatics for the Functional Analysis of Mammalian Genomes (BFAM) and the German National Genome Research Network (NGFN). The authors want to thank H. Löwel, C. Meisinger, R. Holle and J. John from the KORA Study Group. The FHS replication study is a contribution from the Framingham Heart Study of the National Heart, Lung, and Blood Institute (NHLBI) of the National Institutes of Health and Boston University School of Medicine, supported by NHLBI's Framingham Heart Study (Contract No. N01-HC-25195) and the Cardiogenomics Program for Genomic Applications (5U01HL066582), a GlaxoSmithKline Competitive Grants Award Program for Young Investigators (CNC) and NIH (K23HL080025, CNC). Some of the electrocardiographic measurements were supported by an unrestricted grant from Pfizer, Inc.

Competing interests statement:  The authors declare competing financial interests.

Previous | Next
Table of contents
Download PDFDownload PDF
Send to a friendSend to a friend
rights and permissionsRights and permissions
CrossRef lists 252 articles citing this articleCrossRef lists 252 articles citing this article
Save this linkSave this link
More articles like this
Figures & Tables
Competing financial interests
Supplementary info
Export citation
Export references



Search buyers guide:

Nature Genetics
ISSN: 1061-4036
EISSN: 1546-1718
Journal home | Advance online publication | Current issue | Archive | Press releases | Supplements | Focuses | For authors | Online submission | Permissions | For referees | Free online issue | About the journal | Contact the journal | Subscribe | Advertising | work@npg | naturereprints | About this site | For librarians
Nature Publishing Group, publisher of Nature, and other science journals and reference works©2006 Nature Publishing Group | Privacy policy