Background

Cardiovascular disease (CVD) and its most common manifestations, coronary heart disease (CHD), stroke, heart failure (HF), and atrial fibrillation (AF) are major causes of morbidity and mortality. In many industrialized countries CVD claims more lives each year than any other disease. In the United States, for example, heart disease and stroke are the first and third leading causes of death [1]. At age 40 the lifetime risk of developing CHD is one in two for men and one in three for women [2], the lifetime risk for stroke is one in six for men and one in five for women [3], the lifetime risk for HF is one in five in men and women [4] and the lifetime risk for AF is one in four in both sexes [5].

Prior Framingham Heart Study research points to strong familial patterns of CVD, HF, and AF [6–8] and such evidence is consistent with a genetic effect. Several single gene variants associated with CHD and atherosclerotic CVD have been reported [9–13]. A substantial body of research has also identified a number of genetic variants associated with HF and AF [14, 15].

We report results of a genome-wide association study of four CVD outcomes in community-based Framingham Heart Study participants who were enrolled without regard to disease status. Analysis for each specific outcome was restricted to those free of the condition at baseline. We also provide association results for previously reported candidate genes and candidate regions for these CVD outcomes.

Methods

Study sample

In 1948, 5209 men and women from Framingham, Massachusetts, who were between 28 and 62 years of age, were recruited to participate in the Framingham Heart Study [16]. Periodic clinic visits, performed every two years, included a medical history, physical examination focusing on the cardiovascular system, laboratory tests, and electrocardiogram. The offspring cohort of the Framingham Heart Study began in 1971, with the enrollment of 5124 offspring and spouses of offspring of original participants [17]. Repeated examinations of the offspring cohort occurred approximately every 4 years, except for an 8 year interval between their initial and second visit. At each clinic visit, participants gave written informed consent. The consent documents and the examination content were approved by the Institutional Review Board at Boston University Medical Center (Boston, Massachusetts).

Phenotype definition & methods

All participants in both cohorts who were free of a specific condition at enrollment were analyzed for onset of that endpoint during follow up through the end of 2004. All suspected CVD events were reviewed and adjudicated by a panel of three Framingham physician investigators after review of all available Framingham Heart Study examination records, hospitalization records, and physician notes, using previously published criteria [18].

For these analyses, we considered four groups of events: major CHD events included recognized myocardial infarction, coronary insufficiency, and death due to CHD; major atherosclerotic CVD events included major CHD plus atherothrombotic stroke; the remaining groups were HF and AF. Myocardial infarction was diagnosed by the presence of 2 out of 3 clinical criteria: new diagnostic Q-waves on ECG, prolonged ischemic chest discomfort, and elevation of serum biomarkers of myocardial necrosis. CHD death was established upon review of all available records, if the cause of death was probably CHD and no other cause could be ascribed.

Atherothrombotic brain infarction was defined as a non-embolic acute-onset focal neurological deficit of vascular etiology that persisted for more than 24 hours or an ischemic infarct was documented at autopsy.

History of interim hospitalizations and symptoms of HF were obtained at each clinic examination; outside medical records were evaluated for participants who did not attend an examination. Three physicians reviewed all suspected interim events using Framingham Heart Study clinic notes, external physician reports and hospitalization records. HF was diagnosed when at least two major criteria were present, or one major and two minor criteria. Major criteria were paroxysmal nocturnal dyspnea, pulmonary rales, distended jugular veins, enlarging heart size on chest radiography, acute pulmonary edema, hepatojugular reflux, third heart sound, jugular venous pressure of 16 cm or greater, weight loss of 4.5 kg or greater in response to diuresis, pulmonary edema, visceral congestion, or cardiomegaly on autopsy. Minor criteria counted only if not attributed to another disease. Minor criteria were bilateral ankle edema, nocturnal cough, shortness of breath on ordinary exertion, hepatomegaly, pleural effusion, vital capacity decreased by one third from previous maximum, and heart rate ≥120 beats/min.

AF was diagnosed when, upon review by a study cardiologist, AF or atrial flutter was present on an ECG obtained from a routine Framingham clinic examination or from a hospital or physician record. HF was defined on the basis of review of medical records and the finding of concurrent presence of two major or one major plus two minor criteria [19].

Genotyping methods

The accompanying Overview [20] provides details of the genotyping methods used in this investigation. The Affymetrix 100K chip with 112,990 autosomal SNPs was used to genotype individual participant DNA on the Framingham Heart Study family plate set. SNPs were excluded for minor allele frequency < 0.1 (n = 38062); call rate < 0.8 (n = 2346); Hardy Weinberg equilibrium p value < 0.001 (n = 1595). After these exclusions, 70,987 SNPs were available for analysis.

Statistical methods

Proportional-hazards models were used to analyze time to each endpoint, stratified by cohort, using covariate values obtained at enrollment. Models were adjusted for (i) sex and age, or (ii) sex, age and multiple covariates. For CVD and CHD, covariates included smoking, diabetes, systolic BP, anti-hypertensive treatment and total cholesterol; for HF, covariates were smoking, diabetes, systolic BP, anti-hypertensive treatment and body mass index; for AF, covariates were diabetes, systolic BP, anti-hypertensive therapy and valve disease. Deviance residuals estimated from each model were standardized (mean 0, variance 1) to form the phenotypes analyzed with genetic models. For genotype-phenotype association analyses, we assumed an additive-allele model of inheritance and we conducted association tests using regression models with generalized estimating equations (GEE), as well as family-based association testing using FBAT. Due to relatively small numbers of outcome events and non-normality of the deviance residuals, we decided a priori not to perform linkage analysis on outcomes residuals. The distribution of observed p values for the four CVD outcomes was compared to that which would be expected under the null hypothesis of no genetic associations with outcomes.

Candidate gene analyses

GEE and FBAT additive genetic effect models also were run for SNPs in or near candidate genes for each of the CVD outcomes. Candidate genes were selected after separate literature searches for each outcome. All SNPs across the interval extending from 200 Kb proximal to the start to 200 kb beyond the end of each gene were eligible if the minor allele frequency was ≥0.1, the genotype call rate was ≥0.8, and the Hardy-Weinberg equilibrium p value was ≥0.001.

Results

Four primary phenotypes were analyzed: major atherosclerotic CVD (n = 142), major CHD (n = 118), HF (n = 73), and AF (n = 151). Covariates for each outcome are listed in Table 1. In this sample, deviance residuals from multivariable models generally had low heritability: HF, 0.023 (SE = 0.054); Major CVD, 0.036 (SE = 0.058), Major CHD, 0.085 (SE = 0.061); and AF, 0.135 (SE = 0.058).

Table 1 Phenotype definitions
Full size table

GEE additive genetic models yielded six associations with p values < 10-5 and another 31 with p values < 10-4 (see Table 2a for best 25). The lowest p-values for each CVD phenotype were as follows: major CVD, rs499818, p = 6.6 × 10-6; major CHD, rs2549513, p = 9.7 × 10-6; AF, rs958546, p = 4.8 × 10-6; HF: rs740363, p = 8.8 × 10-6. Of note, rs10501920 in CNTN5 was associated with AF (p = 9.4 × 10-6) and HF (p = 1.2 × 10-4). Three SNPs near PHACTR1 were associated with major CVD (rs499818, rs1512411, rs507369; lowest p = 6.6 × 10-6) and one of these was associated with major CHD (rs1512411; p = 6.3 × 10-5). Among GEE results for HF was rs939698 (p = 3.6 × 10-4) in RYR2, which has been implicated in arrhythmogenic right ventricular dysplasia/cardiomyopathy [21], a rare familiar cardiomyopathy.

Table 2 Additive Genetic Model – ordered by GEE (2a) and FBAT (2b) p-value Results
Full size table

Results of FBAT are provided in Table 2b. The lowest p values for each phenotype were: major CVD, rs39312 in WNT2, p = 4.4 × 10-5; major CHD, rs10505879, p = 3.1 × 10-5; AF, rs10511311 in CD200, p = 4.5 × 10-5; and HF, rs10515869, 4.72 × 10-5.

The distribution of observed GEE p values is presented in Table 3. Note that the ratio of observed to expected numbers is inflated only at very low p values.

Table 3 Distribution of Observed and Expected P Values from GEE models
Full size table

Association results for 408 SNPs in 46 candidate genes (Table 4) revealed suggestive evidence for major CHD events for ALOX5AP (23 SNPs, 7 with p < 0.05 by GEE or FBAT), GJA4 (14 SNPs, 6 with p < 0.05), MEF2A (5 SNPs, 2 with p < 0.05), and PCSK9 (11 SNPs, 3 with p < 0.05). For HF, 4 SNPs in PLN and 2 each in ADRB2 and TPM1 had p values < 0.05. There was little evidence of association of AF with SNPs in specified candidate genes. Overall, 538 candidate-SNP association tests were carried out because there were 130 SNPs common to both major CHD and major CVD. Results with GEE p < 0.05 were obtained for 28 tests (5.2%) and p < 0.01 for 5 tests (0.9%), similar to the overall distribution in Table 3. Lack of consistency between GEE and FBAT results may be due to lower power of FBAT compared with GEE tests.

Table 4 Association Results for Pre-Specified Candidate Genes
Full size table

Additionally, we examined all association results for major CHD and major CVD in the region of chromosome 9 that was recently reported to be associated with MI and CHD [22, 23], We found that 7 SNPs in a 76 Kb region had p < 10-5 for one or both outcomes.

Discussion

Cardiovascular disease is the leading cause of death in industrialized countries and will soon be the leading cause of death in the developing world [24]. Genome-wide association studies provide an opportunity to extend our understanding of CVD pathogenesis and improve public health. The identification of novel genes and pathways that play a causal role in CVD is an essential objective for the development of new therapies for the prevention and treatment of CVD. Finding genetic associations with CVD risk that are robust across multiple studies will aid in the personalization of medicine by identifying high risk individuals who can be targeted for early and aggressive preventive care.

We provide results of genome-wide association for 4 CVD outcomes of great public health impact: major CVD, major CHD, AF, and HF. No associations attained genome-wide significance [4.4 × 10-8 = 0.05/(70,987 SNPs × 4 major traits × 2 adjustment levels × 2 association models)] in our analyses using GEE or FBAT additive genetic models. With dramatic declines in the cost of high throughput genotyping, selective genotyping of SNPs with suggestive evidence of association can be considered. Two-stage approaches – genome-wide association followed by selective genotyping – have been adopted as a practical and efficient strategy for pursuing initial genome-wide results [25, 26].

Results of GEE and FBAT associations pointed to few candidate genes of obvious interest for any CVD outcomes. One intriguing result was the association of RYR2 (rs939698, p = 3.6 × 10-4) with HF. The ryanodine receptor has been implicated in arrhythmogenic right ventricular dysplasia/cardiomyopathy [21, 27], a rare familial cardiomyopathy.

The lowest p values we identified may be purely by chance. The number of events (maximum of 142 for major CVD) was small to detect association, but would be sufficient to detect a SNP with high minor allele frequency in linkage disequilibrium with a causal variant that contributed high risk. This was the case for a genome-wide association study of age-related macular degeneration – only 96 cases and 50 controls were sufficient to identify genome-wide association with complement factor H [28]. Sometimes multiple SNPs in the same chromosomal region had low GEE p values for a trait; for example, Table 2a has SNP clusters on chromosomes 6, 9, 11, 13, 15 and 17. Linkage disequilibrium exists for those clustered SNPs (typically, pair-wise r2 above 0.80) and it is uncertain whether the concordant results represent statistically correlated chance findings or indicate regions of heightened interest.

Candidate gene results for the 4 CVD outcomes provided suggestive confirmation of prior associations reported for ALOX5AP (23 SNPs, 7 with p < 0.05 by GEE or FBAT), GJA4 (14 SNPs, 6 with p < 0.05), MEF2A (5 SNPs, 2 with p < 0.05), and PCSK9 (11 SNPs, 3 with p < 0.05) in relation to CHD risk. In contrast, candidate gene results for HF and AF provided little evidence of replication of previously reported associations. Null results of these associations may be due in part to poor coverage of the candidates by the SNPs on the 100K chip and the modest number of events available for analysis. Our results can be compared with other genome-wide associations of similar phenotypes. We observed strong association of major CVD with 3 SNPs in the region of chromosome 9 that was recently reported to be associated with MI and CHD in multiple samples [22, 23]. This provides convincing evidence that, despite modest numbers of events, we were able to identify true associations.

This investigation has several limitations. This study used CVD cases that were identified through careful surveillance of a community-based sample with multigenerational participation. Recruitment of original and offspring cohort participants began long before DNA collection, which occurred in recent years. Thus, most CVD cases were prevalent at the time of DNA collection. For CVD outcomes (such as these) with substantial mortality risk, a survival bias may have been introduced by this study design; individuals with early CVD events had to survive and attend a later clinic examination at which DNA was collected. Another limitation is the modest number of events included in analyses, in particular for HF, where only 73 events were available for analysis. For continuous traits, we had 78% power to detect a SNP with QTL heritability of 1% at significance level 10-3, and at significance level 10-6 we had 84% power for QTL heritability 2% [20]. In the setting of a limited number of outcome events, those are large effect sizes. The negative results of candidate gene analyses may underestimate associations for genes that are incompletely covered by the SNPs used in this investigation. Lastly, a large proportion of the results are likely to be due to chance. Replication studies are needed to determine which, if any, of the results we report are indicative of true associations of causal variants with disease outcomes.

These association results for major CVD outcomes extend experience with genome-wide association studies. Replication studies are needed and will be used to guide future genotyping and resequencing efforts. Finding genetic variants associated with CVD may facilitate the identification of high risk patients and aid in identifying targeted future approaches to prevention and treatment of CVD.