GcJJDC

FTRO#

N

Bgi

Oo

Mrtqas(A

R5ah

IEm

nSAPmFiadafcth

1

enome-wide association study of electrocardiographiconduction measures in an isolated founder population: Kosrae. Gustav Smith, MD,�,‡,§ Jennifer K. Lowe, PhD,�,† Sirisha Kovvali, BA,� Julian B. Maller, BSc,�,‡,¶

acqueline Salit, MS,** Mark J. Daly, PhD,�,‡ Markus Stoffel, MD, PhD,�

avid M. Altshuler, MD, PhD,�,†,‡ Jeffrey M. Friedman, MD, PhD,** Jan L. Breslow, MD,#

hristopher Newton-Cheh, MD, MPH�,‡,§

rom the *Program in Medical and Population Genetics, Broad Institute of Harvard and Massachusetts Institute ofechnology, Cambridge, †Department of Molecular Biology, ‡Center for Human Genetic Research, and §Cardiovascularesearch Center, Massachusetts General Hospital, Boston, Massachusetts, ¶Department of Statistics, University ofxford, Oxford, United Kingdom, �Institute of Molecular Systems Biology, ETH Zurich, Zurich, Switzerland, and

Laboratory of Biochemical Genetics and Metabolism and **Laboratory of Molecular Genetics, Rockefeller University,

ew York, New York.

Gnpaswa1am

Csw

KG

(

ACKGROUND Cardiac conduction, as assessed by electrocardio-raphic PR interval and QRS duration, is an important electrophys-ological trait and a determinant of arrhythmia risk.

BJECTIVE We sought to identify common genetic determinantsf these measures.

ETHODS We examined 1604 individuals from the island of Kos-ae, Federated States of Micronesia, an isolated founder popula-ion. We adjusted for covariates and estimated the heritability ofuantitative electrocardiographic QRS duration and PR intervalnd, secondarily, its subcomponents, P-wave duration and PRegment. Finally, we performed a genome-wide association studyGWAS) in a subset of 1262 individuals genotyped using theffymetrix GeneChip Human Mapping 500K microarray.

ESULTS The heritability of PR interval was 34% (standard error [SE]%, P � 4 � 10�18); of PR segment, 31% (SE 6%, P � 3.2 � 10�13);nd of P-wave duration, 17% (SE 5%, P � 5.8 � 10�6), but the

rsvvdtrEaav

w0vvsarvard.edu. (Received January 12, 2009; accepted February 11, 2009.)

547-5271/$ -see front matter © 2009 Heart Rhythm Society. All rights reserved

WAS was performed only for the PR interval and its subcompo-ents. A total of 338,049 single nucleotide polymorphisms (SNPs)assed quality filters. For the PR interval, the most significantlyssociated SNPs were located in and downstream of the alpha-ubunit of the cardiac voltage-gated sodium channel gene SCN5A,ith a 4.8 ms (SE 1.0) or 0.23 standard deviation increase indjusted PR interval for each minor allele copy of rs7638909 (P �.6 � 10�6, minor allele frequency 0.40). These SNPs were alsossociated with P-wave duration (P � 1.5 � 10�4) and PR seg-ent (P � .01) but not with QRS duration (P �.22).

ONCLUSIONS The PR interval and its subcomponents showedubstantial heritability in a South Pacific islander population andere associated with common genetic variation in SCN5A.

EYWORDS Conduction; Electrocardiography; Electrophysiology;enetics; Ion channels

Heart Rhythm 2009;6:634–641) © 2009 Heart Rhythm Society.

eritablility of QRS duration was only 3% (SE 4%, P � .20). Hence, All rights reserved.

ntroductionlectrocardiographic PR interval and QRS duration areeasures of cardiac conduction time. The PR interval

Dr. Newton-Cheh was supported by National Institutes of Health granto. NIH K23-HL-080025, a Doris Duke Charitable Foundation Clinicalcientist Development Award, and a Burroughs Wellcome Fund Careerward for Medical Scientists. J.K. Lowe was supported by a Tostesonost-Doctoral Fellowship for Basic Research from the Massachusetts Bio-edical Research Corporation. Dr. Altshuler is a Burroughs Wellcomeund Clinical Scholar in Translational Research and a Distinguished Clin-

cal Scholar of the Doris Duke Charitable Foundation. The genome-widessociation scan on Kosrae was supported by grants from the Starr Foun-ation and Howard Hughes Medical Institute. Address reprint requestsnd correspondence: Christopher Newton-Cheh, M.D., M.P.H., Centeror Human Genetic Research, Cardiovascular Research Center, Massa-husetts General Hospital, 185 Cambridge Street, CPZN 5.242, Bos-on, Massachusetts 02114. E-mail address: cnewtoncheh@chgr.mgh.

eflects conduction of the electrical impulse from theinoatrial node through the conduction system to theentricular myocardium but primarily through the atrio-entricular (AV) node. It can be subdivided into P-waveuration and the PR segment, from the end of the P waveo the beginning of the QRS complex. The QRS durationeflects conduction through the ventricular myocardium.xperimental and clinical data suggest a role for globaltrial conduction slowing in the pathogenesis of atrialrrhythmias,1 and QRS prolongation is associated withentricular arrhythmias and sudden cardiac death.2

The PR interval has evidence of a heritable componentith most previous heritability estimates in the range of.34–0.46, which could result from rare or common geneticariants or both.3–5 Rare coding6,7 and common noncoding8

ariants in the alpha-subunit of the cardiac voltage-gated

odium channel gene, SCN5A, have been reported to cause

. doi:10.1016/j.hrthm.2009.02.022

ddpedss

iaumcagtdmipcMmbp

goccdGKfls

MSIso12it(ltmTiPntAv

rinatoaiEpRtfA

ET1PddoYVPfMpl

GGHclScatS

PWp((iQdN

reaui

635Smith et al GWAS of Cardiac Conduction

elayed conduction and PR interval prolongation, but toate most of the PR interval heritability remains unex-lained. The PR interval has also been shown to be influ-nced by other factors including sex, age, heart rate, andromotropic drugs, such as those acting on the cardiacodium channel. QRS duration, has not shown evidence ofignificant heritability in most previous studies.5,9

A genome-wide association study (GWAS) can be used todentify genes (known or unknown) and variants in them thatre associated with cardiac conduction phenotypes. GWASses dense maps of common genetic single nucleotide poly-orphisms (SNPs) to examine whether genotype is signifi-

antly associated with differences in phenotype. The GWASpproach has only recently been made possible by advances inenotyping and genomic mapping and has successfully iden-ified genes involved in a number of common, multifactorialiseases including atrial fibrillation10 as well as genetic deter-inants of quantitative traits such as electrocardiographic QT

nterval.11 In aggregate, such studies have confirmed the hy-othesis that common genetic variants influence heritability ofommon diseases and observed variation in quantitative traits.any of the genes and pathways identified in this systematicanner, such as NOS1AP for QT interval,11 had not previously

een linked to the trait of interest and thus provide novelhysiological information.

We hypothesized that a systematic search for commonenetic variants with an influence on PR interval and, sec-ndarily, its subcomponents, as well as on QRS durationould identify novel molecular mechanisms involved inardiac conduction, which would then be potential can-idates to influence arrhythmogenesis. We performed aWAS in a community-based sample from the island ofosrae, Federated States of Micronesia, an isolated

ounder population. This population exhibits long-rangeinkage disequilibrium,12 potentially making associationcanning with high genomic coverage feasible.

ethodstudy samplenclusion of individuals and variables examined in the Ko-rae study has been described elsewhere.13 Briefly, residentsf Kosrae were examined at three different time points: in994, 2001, and 2003. We studied the sample examined in001, which included a total of 2170 individuals.14 Allndividuals were examined with recording of anthropometricraits, blood pressure, several blood biomarkers, medicationsthe island has a limited formulary), self-reported ethnicity, andifestyle habits. Self-reported familial relationships were inves-igated by a genetics counsellor who was able to construct aultigenerational pedigree that included most of the islanders.he relationships were later validated or revised based on

dentity-by-descent estimates from SNP data usingLINK.14,15 One individual was excluded for self-report ofon-Micronesian ancestry. For the three pairs of monozygoticwins identified, the average trait value was used in analyses.fter exclusions, there were 1604 individuals in the PR inter-

al heritability analysis and 1588 individuals in the QRS du- a

ation heritability analysis. PR interval GWAS was completedn the 1262 individuals free of exclusions with available ge-ome-wide genotypes passing quality control. In a secondarynalysis after our initial PR interval GWAS results were ob-ained, additional ECG measurements were made in the subsetf individuals free of exclusions with genotypes in the GWASnd included 1133 individuals with P-wave duration and 1131ndividuals with PR segment measures. Informed consent innglish or Kosraen was obtained from all subjects. The studyrotocol was approved by the institutional review boards ofockefeller University, Massachusetts General Hospital, and

he Massachusetts Institute of Technology. The authors hadull access to the data and take responsibility for their integrity.ll authors have read and agree to the manuscript as written.

CG trait measurementwelve-lead ECGs were recorded at a paper speed of 25 mm/s in756 individuals using a Universal ECG (QRS Diagnostic LLC,lymouth, MN). Deidentified paper ECGs were scanned intoigital images at high resolution. Quantitative measures of QRSuration, PR interval, P-wave duration, and PR segment werebtained using digital calipers in Rigel 1.7.3 (AMPS Llc., Nework) from two cardiac cycles of lead II and one cycle each from2 and V5. The PR interval was measured from the onset of thewave to the onset of the QRS complex, and the PR segment

rom the offset of the P wave to the onset of the QRS complex.easurements in all analyzed individuals were made at two time

oints with high agreement for mean PR interval (r � 0.90) butower for mean QRS duration (r � 0.60).

enotyping and quality controlenotyping was performed using the Affymetrix GeneChipuman Mapping 500K microarray at Affymetrix (San Fran-

isco). Genotypes were called using the Bayesian robustinear modeling using Mahalanobis distance algorithm.16

NPs were excluded for the following reasons: genotypeall rate below 95%, 12 or more Mendelian errors, minorllele frequency (MAF) below 0.01, or mapping to morehan one locus in the human genome. A total of 338,049NPs passed quality criteria on the 22 autosomes.

henotype modelinge excluded from analyses of PR interval and its subcom-

onents individuals with Wolf-Parkinson-White patternn � 1), those using drugs that slow AV-nodal conductionn � 4), and those with atrial fibrillation (n � 1). Nondividuals examined for this study had a pacemaker. InRS analyses, we excluded individuals with mean QRSuration �120 ms or right bundle branch block (n � 36).ormality of traits was confirmed.We adjusted for covariates using multivariable linear

egression models. To create regression models, we firstxamined univariable regression models for relevant covari-tes. The following covariates were significant (P �.05) innivariable linear regression models: age, sex, body massndex, height, weight, waist, waist-to-hip ratio, body surface

rea, lean body mass, systolic blood pressure, diastolic

bhtt

im(ettsamswtitsS

HAaAd

GTslMastwmtnt(srisaio

e(tPiue

RPFThfacve

HT[foPo0stato

GAv3tif

T

ABS

D

L

H

T

TTSPQHPP

636 Heart Rhythm, Vol 6, No 5, May 2009

lood pressure, mean arterial blood pressure, pulse pressure,ypertension, heart rate, total cholesterol, low-density lipopro-ein, high-density lipoprotein, triglycerides, creatinine, diabe-es, fasting blood sugar, and HbA1c.

Covariates with P �.05 on univariable analysis werencluded in stepwise multiple regression models for eacheasure in cardiac cycle-specific and averaged trait values

from up to four measures). The variability explained byach model (as estimated by the coefficient of determina-ion, r2) was compared between the cycle-specific and meanrait models, under the assumption that more precise mea-ures would have a greater proportion of variation explain-ble by clinical covariates. For all traits, the r2 in regressionodels and the correlation of two independent sets of mea-

ures were greatest for the mean of all four measures, whichas therefore used for heritability estimation and associa-

ion. Residuals from fully adjusted models were standard-zed to mean 0, standard deviation (SD) 1 and served as therait for subsequent heritability and genotype-phenotype as-ociation analyses. Phenotype modeling was performed inPSS (ver. 11.5.0, SPSS Inc., Chicago).

eritabilitydditive heritability was estimated using variance components

nalysis as implemented in the Sequential Oligogenic Linkagenalysis Routines software (SOLAR v4.0.7, Southwest Foun-ation for Biomedical Research, San Antonio).

enome-wide association analysishe association strategy was designed to account for themall effective population size and the high degree of re-atedness that exists on Kosrae due to the founder effect.

ultiple strategies were examined and compared for type Ind type II error rates in simulation experiments.14 The finaltrategy tested for association within sibships drawn fromhe extended Kosrae pedigree and combines scores from bothithin- and between-sibship association testing as imple-ented in the QFAM-Total procedure in PLINK.15 An addi-

ive genetic model was tested in all analyses. We included aumber of singletons (“sibships” of one) with low relatednesso the rest of the sample as defined by identity-by-descentIBD) estimates �0.125 (equivalent to a first cousin relation-hip or less). We adjusted for test statistic inflation induced byelatedness between sibships by calculation of the genomicnflation factor and subsequent genomic control. Our analytictrategy thus attempts to maximize power by analyzing withinnd between components of association, while controlling thenflation in test statistics resulting from background relatednessn the island.

A quantile-quantile plot of the observed compared withxpected �log10(P-value) was generated in R, version 2.6.0http://www.r-project.org), and the �log10(P-value) acrosshe genome was plotted using Haploview version 4.0.17

lots of correlation patterns for selected loci were generatedn Haploview using pairwise r2 of SNPs from 133 individ-als drawn from the entire population who exhibit the low-

st pairwise relatedness (pairwise IBD estimates �0.08). s

esultshenotype modelingor clinical characteristics of the study sample, see Table 1.he final regression model for PR interval included age, sex,eart rate, weight, and fasting blood sugar, and the final modelor QRS duration included sex, heart rate, body surface area,nd pulse pressure. For P wave and PR segment, the sameovariates as for PR duration were used. The proportion ofariation in phenotype explained by covariates (r2) was mod-st: 17%, 6%, 5%, and 10%, respectively.

eritabilityhe estimated heritability of PR interval was 0.34 (standard error

SE] 0.05, P � 4 � 10�18), suggesting that additive geneticactors explain 34% of variability in PR interval. The heritabilityf P-wave duration was 0.17 (SE 0.05, P � 5.8 � 10�6), and ofR segment it was 0.31 (SE 0.06, P � 3.2 � 10�13), while thatf QRS duration was not significant (h2 � 0.03; SE 0.04, P �.20). The correlation to PR interval of its subcomponents wasubstantial (P wave r � 0.48, PR segment r � 0.86). Because ofhe significant heritability of PR interval and its subcomponentsnd the low heritability of QRS duration, genome-wide associa-ion testing was performed for PR interval and its subcomponentsnly.

enome-wide association studyGWAS of PR interval was performed using 1262 indi-

iduals from 168 sibships and 189 singletons and the38,049 SNPs that passed quality control. Mild inflation ofest statistics was observed (Figure 1), with a genomicnflation factor �GC of 1.16. After genomic control to corrector relatedness between sibships, there was a slight excess of

able 1 Sample characteristics

Female(n � 1049)

Male(n � 702)

Total(n � 1751)

ge, years 37.3 (15.3) 37.9 (17.1) 37.6 (16.1)ody mass index 31.1 (6.5) 28.9 (5.7) 30.2 (6.3)ystolic bloodpressure, mmHg

110.1 (18.2) 115.3 (16.8) 112.1 (17.9)

iastolic bloodpressure, mmHg

73.4 (11.5) 76.3 (11.1) 74.6 (11.4)

ow-densitylipoprotein, mmol/L

2.5 (0.7) 2.6 (0.8) 2.5 (0.8)

igh-densitylipoprotein, mmol/L

1.1 (0.3) 0.9 (0.2) 1.0 (0.3)

otal cholesterol,mmol/L

3.9 (0.9) 3.9 (1.0) 3.9 (0.9)

riglycerides, mmol/L 0.9 (0.4) 1.1 (0.6) 1.0 (0.5)ype 2 diabetes, % 16.6 16.0 16.3moking, % 0.8 28.5 11.9R interval, ms 164.4 (21.5) 170.1 (24.8) 166.7 (23.0)RS interval, ms 97.4 (9.3) 101.0 (9.2) 98.8 (9.4)eart rate, bpm 72.1 (9.7) 66.8 (10.1) 69.9 (10.2)-wave duration, ms 104.2 (11.7) 107.0 (12.0) 105.3 (11.9)R segment, ms 65.8 (18.7) 67.9 (20.4) 66.6 (19.4)

Mean and standard deviation are reported for each sex and the total

ample. Percentage of cases is reported for dichotomous traits.

llstat(maeraiSlrSb0aoelaTra2

fp

ldr(

SQSwtsaa.21e

DMWdEsveppteScam

HTsftAiSbhpeoPdhceQtre

SRc

F3t(oc

637Smith et al GWAS of Cardiac Conduction

ow P-values (high �log[P-value]). The 10 results with theowest P-values were all for noncoding SNPs at chromo-ome 3p22.2 (Figure 2) at the 3= end of the alpha-subunit ofhe cardiac sodium channel gene, SCN5A (MIM 600163)nd downstream (Figure 3, Supplementary Table 1). Theop SNP result was for rs7638909 in intron 27 of SCN5AMAF 0.40), which was associated with an effect size of 4.8s (SE 1.0) or a 0.23-SD increase in PR duration for each

dditional minor allele copy (P � 1.6 � 10�6; Table 2),xplaining 2.0% of variation in PR interval. The second bestesult was for SNP rs2070488 (MAF 0.33), which wasssociated with a 5.0-ms (SE 1.1) or 0.24-SD decrease in PRnterval per minor allele (P � 3.8 � 10�6). The majority of theNPs with the lowest P-values in the region are highly corre-

ated and fall in one large block of strong linkage disequilib-ium (Figure 4A). We tested for association of the top twoNPs in a multivariable linear regression model and foundoth to be significant (rs7638909, P � 0.001; rs2070488, P �.04), suggesting possible independent effects. However, ex-mination of the haplotypes across the locus suggests that thepposite direction of effect for the top two SNPs may bexplained by the fact that the minor allele of rs7638909 isocated on a complementary haplotype containing the majorllele of rs2070488 (Figure 4B, 4C). As can be seen inable 2, only one other SNP than the 10 SNPs at SCN5A

eached a P-value below 10�5, rs2461751 (4.5 ms/minorllele, SE 1.0, MAF 0.44, P � 8.0 � 10�6) on chromosomeand more than 200 kb from any known gene.GWAS for P-wave duration and PR segment was per-

ormed in 1133 and 1131 individuals, and genomic control waserformed for inflation factors of 1.34 and 1.27, respectively.

For PR segment, the most significant SNP, rs2008242, isocated near MSX1 (P � 3 � 10�6), and for P-waveuration we note that a SNP, rs10925451, in the cardiacyanodine receptor (RYR2) ranks among the top 10 SNPs

igure 1 Quantile-quantile plot of PR interval association results for38,049 autosomal SNPs under an additive genetic model. Plotted arehe expected versus observed �log10(P-value) before (black) and afterred) genomic control from the PR interval GWAS. The measure ofverdispersion of the test statistics, �GC, was 1.16 before genomicontrol.

P � 2.9 � 10�5) (Supplementary Table 2). Q

CN5A in P-wave duration, PR segment, andRS durationince the heritability of QRS duration was nonsignificant,e did not perform a genome-wide association scan for this

rait. However, we did examine whether the SNP with thetrongest association with PR interval, rs7638909, was alsossociated with QRS duration. We did not find a significantssociation between SCN5A SNPs and QRS duration (P �20, Table 2). We did find this SNP to be associated with a.2-ms (0.19 SD) increase in P-wave duration (P � 1.5 �0�4) and a 2.6-ms (0.14 SD) increase in PR segment forach minor allele copy (P � .01).

iscussionain findingse have examined the heritability and common genetic

eterminants of cardiac conduction time as measured byCG intervals in an isolated founder population. We ob-erved heritability estimates for PR interval similar to pre-ious studies in other populations and significant heritabilitystimates for P-wave duration and PR segment, the subcom-onents of PR interval, but not for QRS duration. Weerformed a GWAS of PR interval and found the SNPs withhe lowest P-values to be located in the 3= end of the genencoding the alpha-subunit of the cardiac sodium channelCN5A, a strong candidate gene to modulate this ECGorrelate of cardiac conduction. The most strongly associ-ted SNP increases the PR interval by 4.8 ms/copy of theinor allele and explains 2.0% of PR interval variability.

eritabilityhe PR interval heritability estimate on Kosrae (34%) isimilar to that previously reported in other populations. Aamily-based study of 1951 individuals of European ances-ry from the United States estimated the heritability at 34%.3

family-based study in 820 individuals from the Tokelauslands in Polynesia estimated heritability to be 46%.18

tudies in twins have also shown significant heritabilityut of varying degree: a study in 355 twin pairs estimatederitability to 34%,5 while a study in 20 Norwegian twinairs estimated heritability to be 78%.4 To our knowl-dge, no previous studies have examined the heritabilityf the PR interval subcomponents P-wave duration andR segment. Studies examining the heritability of QRSuration are scarce but have not demonstrated significanteritability.5,9 The lack of heritability in these studiesould be attributable to limited power to detect suchffects due to modest sample sizes or poor precision ofRS measurements. Alternatively, additive genetic fac-

ors as assessed by heritability might not play much of aole in determination of QRS duration compared withnvironmental factors.

CN5A in cardiac conductionare loss-of-function variants in SCN5A have been found toause delayed conduction with prolonged PR interval and

RS duration (MIM 600163.0016),6 atrial fibrillation,19

s6chrrmiams

kpm

ScBtrtB

Fo hromo

Fkbi

638 Heart Rhythm, Vol 6, No 5, May 2009

udden cardiac death,20 sick sinus syndrome (MIM08567),21 Brugada syndrome (MIM 601144),22 dilatedardiomyopathy (MIM 601154),23 and progressive familialeart block (MIM 113900).7 These variants are likely toesult in a reduced action potential upstroke velocity, whichesults in slowed impulse propagation.24 Mouse knockoutodels of SCN5A exhibit 50% reduced sodium conductance

n patch clamp experiments and impaired atrioventricularnd intramyocardial conduction,25,26 whereas transgenicice with constitutive overexpression of SCN5A exhibit

horter P-wave duration and PR interval.27 Heterozygous

igure 2 PR interval association results for 338,049 SNPs across 22 aun the y-axis is �log10(P-value), and on the x-axis, physical position by c

igure 3 Regional association plot for PR interval at a locus on chromosob downstream. Positions are from NCBI build 35 and recombination rates aslue diamond is the SNP with the lowest P-value. Diamond color represents

ndicates strong correlation. Recombination rate is plotted in the background, and kn

nockout mice were found to resemble the human disease ofrogressive familial heart block with a progressive impair-ent of atrial and ventricular conduction.28

Bezzina et al8 described a common haplotype in theCN5A promoter (immediately 5= of the gene) in a sampleonsisting of 102 healthy Japanese subjects and 71 Japaneserugada syndrome patients without coding SCN5A muta-

ions which was associated with PR interval and QRS du-ation in both groups. We find it unlikely that our associa-ion signal is the same as the promoter signal reported byezzina et al8 since the SNPs in our study are all located in

s under an additive genetic model. Each dot represents one SNP. Plottedsome.

The plot covers the genomic region from 150 kb upstream of SCN5A to 300ed from HapMap phase II. SNPs are represented by diamonds, and the largeion with the top SNP: faint red indicates weak correlation, and brighter red

tosome

me 3p.estimatcorrelat

own genes in the region are shown in the bottom of the plot.

tSw.hsLpfgrdafHpSvH5p

boditwdtPdsas

OFrm

tAmf

nThmictmmr

SGierlcaHtttdrv

Gdposci

T

R

11

omo and M

639Smith et al GWAS of Cardiac Conduction

he 3= end of the SCN5A gene, a gene that spans 101 kb. NoNP in the 5= end of SCN5A reaches P �.33, and the SNPith the lowest P-value within 50 kb 5= of the gene has P �

05 (Supplementary Table 1). Additionally, a recombinationotspot separates our association signal from the SNPs withtronger association in the 5= end of the gene (Figure 3).astly, none of the SNPs within 50 kb of the SCN5Aromoter in HapMap Chinese from Beijing or Japaneserom Tokyo samples or in the Kosraen sample has an r2

reater than 0.05 to the top SNP identified in the currenteport (rs7638909). To our knowledge no other studies haveescribed an association between common SCN5A variantsnd cardiac conduction. However, previous studies haveound an association between a common missense variant,558R, in SCN5A and atrial fibrillation in an Asian sam-le29 and a variant associated with ventricular arrhythmia,1102Y, in an African American sample.30 The H558Rariant has low correlation with the top SNP in our report inapMap (r2 � 0.03) and is not included on the Affymetrix00K GeneChip. The S1102Y variant is very rare in sam-les of Asian or European ancestry.

As the cardiac sodium channel is expressed widely inoth ventricles and atria, we also examined the associationf SCN5A variants with the ventricular component of car-iac conduction, QRS duration, and did not find any signif-cant association. This might be explained by differences inhe properties of atrial and ventricular sodium channels,hich have been described elsewhere,31 lack of power toetect modest genetic effects, or a limited genetic contribu-ion to variation in QRS duration. We found the strongestR interval SNP to be strongly associated with P-waveuration, with a weaker association with PR segment, con-istent with a general action of the variant identified in bothtrial conduction tissue and the remainder of the conductionystem.

ther GWAS resultsor PR interval, the only SNP outside the SCN5A locus toeach a P-value below 10�5 is located on chromosome 2,

able 2 Association results from PR interval GWAS with P �10

ank SNP Position Alleles MAF

1 rs7638909 Chr3:38569977 T/G 0.402 rs2070488 Chr3:38417494 G/A 0.333 rs2284820 Chr3:38537583 T/C 0.334 rs1065800 Chr3:38541484 A/G 0.335 rs2268757 Chr3:38480857 C/T 0.336 rs1058945 Chr3:38507515 G/A 0.337 rs1002676 Chr3:38419936 G/T 0.338 rs2051215 Chr3:38535349 G/A 0.339 rs7373828 Chr3:38499315 A/G 0.330 rs2067082 Chr3:38413186 G/C 0.341 rs2461751 Chr2:176114826 A/G 0.44

Additive genetic models were tested for all 338,049 autosomal SNPs paf the reference sequence in NCBI build 35. SNP identification numbers reinor allele with the SE shown in parentheses. The correlation for each SNf the 133 most distantly related Kosraeans genotyped (see the Materials

ore than 200 kb from any known gene. However, one of d

he two closest genes is activating transcription factor 2,TF-2, which was previously implicated in experimentalodels of cardiac hypertrophy.32 This finding could merit

urther examination if replicated in other studies.For PR segment, the most significant SNP was located

ear MSX1. MSX1 was recently shown to interact with-box factors in regulating connexin expression in theeart,33 which contributes to electrical coupling of cardio-yocytes and has been found to have a role in embryolog-

cal development of the AV myocardium, making it a strongandidate for follow-up.34 For P-wave duration, we notedhat a SNP in RYR2, the calcium channel that couplesyocardial excitation to contraction and is responsible foronogenic cardiomyopathy and ventricular tachycardia,

anks among SNPs with the strongest association.

trengths and limitationsenetic screens in isolated populations can be valuable tools in

dentifying Mendelian disease genes due to the low genetic andnvironmental heterogeneity. We have previously shown thateduced genetic heterogeneity and long-range linkage disequi-ibrium on Kosrae also makes GWASs with high coverage ofommon variation feasible, with as much as 80% coverage ofll common variation at r2 �80% using only 110,000 SNPs.12

owever, our study is underpowered to fully take advantage ofhe greater coverage provided by over 300,000 SNPs in iden-ifying common genetic variants of the low effect sizes thatypify many common variants that alter the risk of humanisease or physiology. The effect size seen in SCN5A SNPs iselatively large for a common variant, explaining 2.0% of traitariability.

The significance thresholds that should be used inWASs to control the false-positive rate are a matter ofebate. The Wellcome Trust Case Control Consortium em-loyed a threshold of 5 � 10�7.35 We have recently rec-mmended a threshold of 5 � 10�8 in European populationamples, accounting for roughly 1,000,000 independentommon variant tests in the human genome.36 On Kosrae, its likely that the increased correlation among SNPs (linkage

ct PR P-value QRS effect QRS P-value r2

1.0) 1.7 � 10�6 0.42 (0.34) .22 —1.1) 3.7 � 10�6 �0.35 (0.36) .33 0.291.1) 4.4 � 10�6 �0.37 (0.37) .33 0.321.1) 4.6 � 10�6 �0.31 (0.36) .38 0.321.1) 4.6 � 10�6 �0.34 (0.37) .36 0.301.1) 4.7 � 10�6 �0.32 (0.36) .36 0.321.1) 5.1 � 10�6 �0.35 (0.38) .35 0.291.1) 5.5 � 10�6 �0.33 (0.36) .36 0.321.1) 6.7 � 10�6 �0.32 (0.35) .35 0.331.1) 6.7 � 10�6 �0.22 (0.38) .56 0.311.0) 8.0 � 10�6 — — —

ality control. All SNPs are noncoding. Positions are on the forward strandbSNP. Alleles are shown as major/minor. Effects are milliseconds/copy ofe chr3 locus with the top SNP as measured by r2 is based on examinationethods section).

�5

PR effe

4.8 (�5.0 (�4.9 (�4.9 (�5.0 (�4.9 (�4.9 (�4.9 (�4.9 (�4.8 (4.54 (

ssing qufer to dP in th

isequilibrium), as was previously shown12 and exemplified

hpsttrbgbta

sllrmfic

dKtli

CWtuicist

AT

FdcecctH

640 Heart Rhythm, Vol 6, No 5, May 2009

ere by the reduced number of Kosraen haplotypes com-rised of SNPs in SCN5A as compared with Asian HapMapamples (Figure 4), makes this threshold overly conserva-ive because of the more limited number of independentests across the genome in Kosraens. Lastly, the existence ofare SCN5A mutations in Mendelian forms of conductionlock and a common variant effect in one Asian study givesreater confidence that the current findings are unlikely toe spurious. Ultimately, proof of true association will needo come from replication in additional samples of variousncestral backgrounds.

The isolated population study design also introducesome potential challenges. The long-range linkage disequi-ibrium makes fine-mapping and resequencing of identifiedoci more difficult. In addition, as rare alleles may haveisen to high frequencies in founder populations, findingsay not be generalizable to other populations. Hence, ournding of SCN5A as the top SNP in the PR interval GWAS

igure 4 Panel A shows a linkage disequilibrium (LD) plot including alownstream. Values in the boxes are pairwise SNP correlations (r2); boxorrelation (r2 � 100%), and a clear box indicates low correlation. The LDt al as implemented in Haploview. The strongest and second strongest assoalculated in the 133 most distantly related Kosraens (see the Materials aonstructed in Haploview in which each line represents one haplotype folloo the number for each SNP as given in the LD plot. Panel C shows haploaplotypes with frequencies below 0.02 were excluded.

ould be the result of a generally rare variant such as those a

escribed elsewhere6 that has risen to higher frequency onosrae as recently shown for phytosterolemia.37 However,

he high MAF of the variants reported here make this lessikely. Further work, including fine-mapping and resequenc-ng, is required to exclude this possibility.

onclusione have identified a common variant at the SCN5A locus

hat influences PR interval in a South Pacific founder pop-lation. Replication in other cohorts and extension to clin-cal samples is needed to define the ultimate clinical impli-ations of these findings. Our results add to those of QTnterval duration demonstrating that systematic geneticcans of ECG traits may identify genes and variants impor-ant in electrophysiological function.

cknowledgmentshe authors thank Itsik Pe’er and Benjamin Neale for valu-

with P �10�5 and intervening SNPs in the genomic region of SCN5A andeflects the degree of correlation. An empty, dark box indicates completecluding most of the associated SNPs was defined by the criteria of GabrielNPs are indicated with red and blue boxes, respectively. Correlations werehods section). Panel B shows haplotypes with all 19 SNPs in the regionits frequency on Kosrae and the numbers above each column correspond

f the same SNPs from the Asian populations CHB/JPT in HapMap rel21.

l SNPscolor r

block inciated Snd Metwed bytypes o

ble assistance in developing the genome-wide association

ap

ASSt

R

1

1

1

1

1

1

1

1

1

1

2

2

2

2

2

2

2

2

2

2

3

3

3

3

3

3

3

3

641Smith et al GWAS of Cardiac Conduction

nalyses, the Kosrae Department of Health Services, and theeople of Kosrae, who made this study possible.

ppendixupplementary dataupplementary data associated with this article can be found, in

he online version, at doi:10.1016/j.hrthm.2009.02.022.

eferences1. Platonov PG. Interatrial conduction in the mechanisms of atrial fibrillation: from

anatomy to cardiac signals and new treatment modalities. Europace 2007;9(Suppl 6):vi10–vi16.

2. Kashani A, Barold SS. Significance of QRS complex duration in patients withheart failure. J Am Coll Cardiol 2005;46:2183–2192.

3. Newton-Cheh C, Guo CY, Wang TJ, et al. Genome-wide association study ofelectrocardiographic and heart rate variability traits: the Framingham HeartStudy. BMC Med Genet 2007;8(Suppl 1):S7.

4. Moller P, Heiberg A, Berg K. The atrioventricular conduction time—a heritabletrait? III. Twin studies. Clin Genet 1982;21:181–183.

5. Havlik RJ, Garrison RJ, Fabsitz R, et al. Variability of heart rate, P-R, QRS andQ-T durations in twins. J Electrocardiol 1980;13:45–48.

6. Tan HL, Bink-Boelkens MT, Bezzina CR, et al. A sodium-channel mutationcauses isolated cardiac conduction disease. Nature 2001;409:1043–1047.

7. Schott JJ, Alshinawi C, Kyndt F, et al. Cardiac conduction defects associate withmutations in SCN5A. Net Genet 1999;23:20–21.

8. Bezzina CR, Shimizu W, Yang P, et al. Common sodium channel promoterhaplotype in Asian subjects underlies variability in cardiac conduction. Circu-lation 2006;113:338–344.

9. Russell MW, Law I, Sholinsky P, et al. Heritability of ECG measurements inadult male twins. J Electrocardiol 1998;30:S64–S68.

0. Gudbjartsson DF, Arnar DO, Helgadottir A, et al. Variants conferring risk ofatrial fibrillation on chromosome 4q25. Nature 2007:448:353–357.

1. Arking DE, Pfeufer A, Post W, et al. A common genetic variant in the NOS1regulator NOS1AP modulates cardiac repolarization. Nat Genet 2006;38:644–651.

2. Bonnen PE, Pe’er I, Plenge RM, et al. Evaluating potential for whole-genome studies inKosrae, an isolated population in Micronesia. Nat Genet 2006;38:214–217.

3. Shmulewitz D, Auerbach SB, Lehner T, et al. Epidemiology and factor analysisof obesity, type II diabetes, hypertension, and dyslipidemia (syndrome X) on theIsland of Kosrae, Federated States of Micronesia. Hum Hered 2001;51:8–19.

4. Lowe JK, Maller JB, Pe’er I, et al. Genome-wide association studies in anisolated founder population from the Pacific Island of Kosrae. PLoS Gen2009;5:e1000365.

5. Purcell S, Neale B, Todd-Brown K, et al. PLINK: a tool set for whole-genomeassociation and population-based linkage analyses. Am J Hum Genet 2007;81:559–575.

6. Rabbee N, Speed TP. A genotype calling algorithm for affymetrix SNP arrays.Bioinformatics 2006;22:7–12.

7. Barrett JC, Fry B, Maller J, et al. Haploview: analysis and visualization of LDand haplotype maps. Bioinformatics 2005;21:263–265.

8. Griggs LH, Chapman CJ, McHaffie DJ. Inheritance of atrioventricular conduc-tion time in Tokelau Islanders. Clin Genet 1986;29:56–61.

9. Darbar D, Kannankeril PJ, Donahue BS, et al. Cardiac sodium channel (SCN5A)variants associated with atrial fibrillation. Circulation 2008;117:1927–1935.

0. Albert CM, Nam EG, Rimm EB, et al. Cardiac sodium channel gene variants andsudden cardiac death in women. Circulation 2008;117:16–23.

1. Benson DW, Wang DW, Dyment M, et al. Congenital sick sinus syndromecaused by recessive mutations in the cardiac sodium channel gene (SCN5A).J Clin Invest 2003;112:1019–1028.

2. Chen Q, Kirsch GE, Zhang D, et al. Genetic basis and molecular mechanism foridiopathic ventricular fibrillation. Nature 1998;392:293–296.

3. Olson TM, Michels VV, Ballew JD, et al. Sodium channel mutations andsusceptibility to heart failure and atrial fibrillation. JAMA 2005;293:447–454.

4. Smits JP, Koopmann TT, Wilders R, et al. A mutation in the human cardiacsodium channel (E161K) contributes to sick sinus syndrome, conduction diseaseand Brugada syndrome in two families. J Mol Cell Cardiol 2005;38:969–981.

5. van Veen TA, Stein M, Royer A, et al. Impaired impulse propagation in Scn5a-knockout mice: combined contribution of excitability, connexin expression, andtissue architecture in relation to aging. Circulation 2005;112:1927–1935.

6. Papadatos GA, Wallerstein PM, Head CE, et al. Slowed conduction and ven-tricular tachycardia after targeted disruption of the cardiac sodium channel geneScn5a. Proc Nat Acad Sci U S A 2002;99:6210–6215.

7. Zhang T, Yong SL, Tian XL, et al. Cardiac-specific overexpression of SCN5Agene leads to shorter P wave duration and PR interval in transgenic mice.Biochem Biophys Res Commun 2007;355:444–450.

8. Royer A, van Veen TA, Le BS, et al. Mouse model of SCN5A-linked hereditaryLenegre’s disease: age-related conduction slowing and myocardial fibrosis.Circulation 2005;111:1738–1746.

9. Chen LY, Ballew JD, Herron KJ, et al. A common polymorphism in SCN5A isassociated with lone atrial fibrillation. Clin Pharmacol Ther 2007;81:35–41.

0. Splawski I, Timothy KW, Tateyama M, et al. Variant of SCN5A sodium channelimplicated in risk of cardiac arrhythmia. Science 2002;297:1333–1336.

1. Li GR, Lau CP, Shrier A. Heterogeneity of sodium current in atrial vs epicardialventricular myocytes of adult guinea pig hearts. J Mol Cell Cardiol 2002;34:1185–1194.

2. Lim JY, Park SJ, Hwang HY, et al. TGF-beta1 induces cardiac hypertrophicresponses via PKC-dependent ATF-2 activation. J Mol Cell Cardiol 2005;39:627–636.

3. Boogerd KJ, Wong LY, Christoffels VM, et al. Msx1 and Msx2 are functionalinteracting partners of T-box factors in the regulation of Connexin43. Cardio-vasc Res 2008;78:485–493.

4. Chen YH, Ishii M, Sucov HM, et al. Msx1 and Msx2 are required for endothe-lial-mesenchymal transformation of the atrioventricular cushions and patterningof the atrioventricular myocardium. BMC Dev Biol 2008;8:75.

5. Genome-wide association study of 14,000 cases of seven common diseases and3,000 shared controls. Nature 2007;447:661–678.

6. Pe’er I, Yelensky R, Altshuler D, et al. Estimation of the multiple testing burdenfor genomewide association studies of nearly all common variants. Genet Epi-demiol 2008;32:381–385.

7. Sehayek E, Yu HJ, von Bergmann K, et al. Phytosterolemia on the island of

Kosrae: founder effect for a novel ABCG8 mutation results in high carrier rateand increased plasma plant sterol levels. J Lipid Res 2004;45:1608–1613.