Genetic polymorphisms for estimating risk of atrial fibrillation in the general population: a prospective study

Introduction

Atrial fibrillation (AF) is a common cardiac disease and major risk factor for stroke, heart failure and death. Tools for prediction of AF have been developed to identify individuals who might benefit from preventive therapies, incorporating conventional cardiovascular risk factors, and the effects of such risk factors have been evaluated across several cohorts.^1,2 Recently, a heritable component to AF has been reported and polymorphisms in three genetic regions have been reproducibly associated with AF: chromosome 4q25, located 150 kb from the closest gene - a transcription factor (PITX2) involved in cardiac development; chromosome 16q22, intronic to another transcription factor of unknown function, expressed in cardiac tissue (ZFHX3); and an amino acid-altering variant in KCNH2, one of the major cardiac voltage-gated potassium channels.^3–5 Rare genetic variants segregating with AF are typically exclusive to individual families and unlikely to contribute to AF prediction at the population level but genetic polymorphisms could provide important predictive information.

Methods

The single nucleotide polymorphism (SNP) with the strongest association at each of the three genetic regions reproducibly associated with AF in genome-wide^3,4 or candidate gene studies⁵ were genotyped in a large population-based cohort of middle-aged participants from southern Sweden (Malmö Diet and Cancer study, MDCS).

Data collection and clinical definitions have been described.⁶ Briefly, 30,447 randomly selected individuals (born 1923–1950) attended a baseline examination between 1991–1996 with sampling of peripheral venous blood, measurement of blood pressure and anthropometric measures, and filled out a questionnaire. Cardiac disease endpoints were ascertained from national registers (Swedish Cause of Death Register and Swedish Hospital Discharge Register).⁶ Follow-up for AF extended through January 1, 2009.

DNA extracted from peripheral blood cells was assigned to batches without regard to AF status or personal identity. Batches were genotyped with the same set of reagents using real-time polymerase chain reaction (PCR) with 2.5 ng DNA as PCR template for allelic discrimination on an ABI 7900HT (Life Technologies, Carlsbad, CA, USA). Genotype calls were obtained using SDS 2.3 software and fluorescence intensity plots curated manually.

Association of genotype with AF was studied using both cross-sectional and prospective study designs. In cross-sectional analyses, association of SNPs with AF diagnosed prior to baseline was examined using logistic regression analysis. In prospective analyses, association of SNPs with incident AF during follow-up was examined in individuals free of AF at baseline using Cox proportional hazards models with censoring at death, emigration or end of follow-up. Kaplan-Meier estimates of absolute AF risk per genotype were calculated. The proportionality of hazards assumption was confirmed using Schoenfeld’s global test.

Polymorphisms associated with AF were assessed for predictive discrimination using Harrell’s concordance (C) statistic, a generalization of the area under the receiver-operating characteristic (ROC) curve, with confidence interval estimates using a jackknife resampling method in the STATA package somersd. Model calibration was evaluated using the Gronnesby-Borgan test implemented in the STATA package stcoxgof. All analyses were performed using SAS 9.2 (SAS Institute, Cary, NC, USA) or STATA 11.1 (Statacorp, College Station, Texas).

Results

Baseline characteristics for the MDCS cohort have been published.⁶ Clinical data was available in 28,473 individuals of whom DNA was available in 26,946. The mean age was 58.1 years (SD 7.6) and a majority were women (60.6%). At baseline, 287 individuals had been diagnosed with AF (prevalence 1.0%). During up to 17.8 years follow-up (median 14.1, IQR 12.9–15.7) 2,050 individuals developed AF. The Kaplan-Meier estimate of cumulative AF incidence was 11.9% (95% CI=10.7–13.3).

The call rate was >95% for all three SNPs. Minor allele frequencies (MAF) were similar to previous studies and the European panel (CEU) of the HapMap project (4q25: T allele, MAF 10.1%; 16q22: A allele, MAF 17.4%; KCNH2: G allele, MAF 21.2%). The two SNPs from GWA studies were associated with both incident and prevalent AF, but the SNP in KCNH2 from candidate gene studies was not (Table). The Kaplan-Meier estimate of AF incidence was 27.7% (95% CI=11.6–57.4) for homozygotes of the risk allele T of rs2200733 and 11.4% (95% CI=9.9–13.1) for C allele homozygotes. For rs2106261, the Kaplan-Meier estimate of AF incidence was 18.4% (95% CI=12.8–26.0) for homozygotes of the risk allele T and 11.6% (95% CI=9.9–13.5) for C allele homozygotes. Few individuals were homozygous for risk alleles of both SNPs (n=11) but these individuals had a high prevalence (9%, n=1) and incidence (45%, n=5) of AF.

Table 1.

Table Prediction of AF with genetic polymorphisms and conventional risk factors

A. Risk factor	Cross-sectional		Prospective
	OR (95 % CI)	p-value	HR (95 % CI)	p-value
Age	2.12 (1.75–2.57)	<0.001	2.77 (2.57–2.98)	<0.001
Male sex	1.94 (1.48–2.54)	<0.001	1.79 (1.63–1.97)	<0.001
Body mass index	1.21 (1.03–1.42)	0.07	1.29 (1.22–1.37)	<0.001
Hypertension	2.91 (1.89–4.49)	<0.001	1.46 (1.29–1.65)	<0.001
History of Diabetes	1.80 (1.13–2.87)	0.02	1.25 (1.01–1.54)	0.04
History of MI	1.59 (0.95–2.67)	0.04	1.63 (1.31–2.02)	<0.001
History of HF	18.55 (9.86–34.91)	<0.001	3.22 (1.88–5.53)	<0.001
4q25 (rs2200733)	2.15 (1.69–2.74)	<0.001	1.47 (1.33–1.62)	<0.001
16q22 (rs2106261)	1.28 (1.02–1.61)	0.03	1.13 (1.04–1.12)	0.003
KCNH2 (rs1805123)	0.86 (0.68–1.09)	0.22	1.08 (1.00–1.17)	0.06
B. Model	C-statistic	Calibration	C-statistic	Calibration
Basic model
Age, sex	0.737 (0.711–0.763)	14.5 (p=0.07)	0.738 (0.728–0.748)	15.2 (p=0.08)
Genetic polymorphisms
Age, sex, rs22700733	0.751 (0.724–0.779)	11.7 (p=0.17)	0.742 (0.732–0.753)	15.9 (p=0.07)
Age, sex, rs2106261	0.740 (0.713–0.767)	9.5 (p=0.30)	0.739 (0.729–0.750)	18.8 (p=0.03)
Age, sex, rs22700733, rs2106261	0.751 (0.724–0.779)	5.8 (p=0.67)	0.743 (0.733–0.754)	14.4 (p=0.11)
Conventional risk factors
Age, sex, hypertension	0.755 (0.730–0.779)	4.1 (p=0.85)	0.743 (0.733–0.753)	19.9 (p=0.02)
Age, sex, BMI	0.745 (0.719–0.771)	6.0 (p=0.65)	0.747 (0.737–0.756)	8.4 (p=0.49)
Age, sex, diabetes	0.738 (0.711–0.765)	13.0 (p=0.11)	0.738 (0.727–0.748)	17.4 (p=0.04)
Age, sex, history of MI	0.743 (0.717–0.769)	16.6 (p=0.03)	0.740 (0.730–0.750)	22.7 (p=0.007)
Age, sex, history of HF	0.750 (0.724–0.777)	14.5 (p=0.07)	0.739 (0.730–0.749)	17.2 (p=0.05)
Age, sex, all conventional risk factors	0.776 (0.750–0.802)	2.1 (p=0.98)	0.750 (0.741–0.762)	10.5 (p=0.31)
Conventional risk factors and genetic polymorphisms
Age, sex, conventional risk factors, rs22700733	0.784 (0.757–0.812)	3.2 (p=0.92)	0.754 (0.743–0.765)	10.4 (p=0.32)
Age, sex, conventional risk factors, rs2106261	0.776 (0.749–0.804)	8.1 (p=0.42)	0.751 (0.741–0.762)	4.0 (p=0.91)
Age, sex, conventional risk factors, rs22700733, rs2106261	0.785 (0.757–0.813)	5.2 (0.73)	0.755 (0.744–0.766)	9.6 (p=0.39)

The upper panel (A) of the table presents effect estimates (OR, odds ratio; HR, hazard ratio) with 95% confidence intervals per risk factor from multivariable models including conventional risk factors and genetic polymorphisms. Cross-sectional results refer to logistic regression models of prevalent cases at baseline and prospective results refer to Cox proportional hazards models of incident cases during follow-up. Effect estimates for genetic polymorphisms are shown per risk allele, for age per 10 years and for body mass index per 5 units. P-values refer to Wald χ² tests. The lower panel (B) presents C-statistics with 95% confidence intervals and calibration statistics with p-value for each model. Calibration refers to Hosmer-Lemeshow tests for cross-sectional analyses and Groennesby-Borgan Likelihood-ratio tests for prospective analyses.

Discrimination of AF with genotypes and conventional risk factors is shown in the Table. Age and sex at baseline showed high discrimination for prevalent (C-statistic 0.737) and incident AF (0.738). Small but non-significant improvements in discrimination were observed with addition of single conventional risk factors or genetic polymorphisms. The addition of all conventional risk factors to age and sex improved C-statistics modestly for prevalent (0.776) and incident AF (0.750). Addition of genetic polymorphisms further improved C-statistic for prevalent (0.785) and incident AF (0.755), although this improvement was not significant.

Comment

In this large, prospective study, two genetic polymorphisms with high prevalence in the population predicted AF independently of and with similar risk magnitude to single clinical risk factors. However, genetic polymorphisms did not significantly improve predictive accuracy when added to clinical risk factors. Findings do not support utility of clinical genotyping for AF risk prediction with these SNPs, currently marketed by commercial companies for direct-to-consumer genetic testing with provision of absolute genotypic risk estimates.

The association with the K897T missense variant in KCNH2 was not replicated, and also recently failed to replicate in a large case-control sample.⁷ These results do not support the large effect described in the initial report, but cannot rule out a small effect.

Additional, independent SNPs on 4q25 have been associated with AF and a polymorphism on chromosome 1q21 was recently associated with lone AF. Although these polymorphisms with smaller effects are unlikely to improve predictive accuracy, future studies will be needed to evaluate the predictive information content of genome-wide SNP data. Furthermore, recent studies have highlighted that asymptomatic episodes of AF may not be uncommon and confer increased stroke risk. Characterization of populations for such episodes might reveal genotypic risks based on clinical AF to be underestimates.