Background

Cardiovascular disease (CVD) is a leading cause of morbidity and mortality in the United States [1]. It is increasingly recognized that CVD is a life-course disease, with overt events being antedated by subclinical cardiovascular target organ damage [2, 3]. Current research indicates a fundamental role of left ventricular (LV) chamber size, wall thickness (LV remodeling) and mass (LVM) in the pathogenesis of high blood pressure [4, 5], and clinical CVD [6, 7], including stroke [8, 9] and heart failure [10–12]. On a parallel note, exercise treadmill stress testing (ETT) is used routinely to evaluate patients with chest pain suggestive of ischemic etiology and for identifying individuals at intermediate pre-test probability of CVD who are more likely to develop clinical events [13]. Likewise, endothelial dysfunction, as assessed via brachial artery (BA) flow-mediated dilation (FMD), has emerged as a fundamental component of atherosclerosis and a precursor of overt CVD [14–16]. Thus, traits obtained via echocardiography (Echo), testing for BA endothelial function and ETT can serve as intermediate phenotypes in the pathway from standard risk factor to overt CVD. Such intermediate phenotypes have been studied extensively to characterize their clinical and genetic correlates, have been reported to be heritable traits [14, 17–28], and have been linked to select genetic loci in several reports [29–31].

More recently, several investigators have proposed genome-wide association studies (GWAS) as a strategy to map causal genes with modest influences on traits associated with complex diseases such as CVD [32, 33]. The availability of 100K genotype data on a subset of related Framingham Heart Study participants [34] provides a unique opportunity to conduct both genome-wide association and linkage analyses to explore the genetic underpinnings of LV remodeling, endothelial function and exercise performance in a community-based sample.

Methods

The design and selection criteria of the Original Framingham Study [35] and the Offspring Study [36] have been described elsewhere. As detailed in the Overview [37], 1345 participants (1087 Offspring and 258 Original Cohort) underwent genotyping using the Affymetrix GeneChip Human Mapping 100K single nucleotide polymorphism (SNP) set [34]. Participants were eligible for the present investigation if they had available genotypes and the echocardiographic, vascular and ETT traits of interest (as defined below). The Institutional Review Board at Boston University Medical Center approved the study and all participants gave written informed consent (including for genetic research).

Measurement of phenotypes

Echocardiography

All attendees underwent routine transthoracic two-dimensionally-guided M-mode echocardiography at the second (1979–1982), fourth (1987–1990), fifth (1991–1995) and sixth (1996–1998) Offspring cohort examinations. Echocardiographic equipment for image acquisition varied across these examinations: at examination cycle 2, a Hoffrel 201 ultrasound receiver (and Aerotech transducer) was used; at examinations 4 and 5, a Hewlett Packard (model 77020AC) ultrasound machine was used; at examination 6 images were acquired using a Sonos 1000 Hewlett-Packard machine. At all four examinations, however, measurements of LV internal dimension in diastole (LVDD) and systole (LVDS), the thicknesses of the posterior wall (PW) and interventricular septum (IVS), and the diameters of the aortic root (all measured at end-diastole) and left atrium (LA) at end-systole were obtained by using a 'leading edge' technique [38], averaging measurements in 3 cardiac cycles according to the American Society of Echocardiography guidelines. We calculated LV wall thickness (LVWT) as the sum of PW and IVS measurements. LV mass was calculated by using the formula 0.8[1.04(LVDD+IVS+PW)3 - (LVDD)3] + 0.6 [39]. The reproducibility of Echo measurements was systematically assessed at the sixth examination [40].

ETT measures

At the second Offspring examination (1978–1981), all attendees underwent submaximal exercise test according to the standard Bruce protocol for up to five incremental 3-minute stages. The test was terminated (without a cool down period) when participants reached their target heart rate (85% age-predicted peak heart rate). Blood pressure measurements and electrocardiograms were recorded during exercise at the midpoint of each 3-minute exercise stage, and for each minute for up to 4 minutes into the recovery period.

BA endothelial function

As described previously [14], BA flow-mediated dilation (FMD; percent change in diameter from baseline; i.e. 100 * [hyperemic diameter at 1 minute - baseline diameter]/baseline diameter) and mean hyperemic flow velocity (cm/sec) were determined during the seventh clinical examination cycle (1998–2001). A Toshiba SSH-140A ultrasound system with a 7.5 MHz linear array transducer and commercially available software (Brachial Analyzer version 3.2.3, Medical Imaging Applications) were used. Investigators, blinded to participant clinical and genetic data, determined brachial artery diameter at baseline and 1 minute after reactive hyperemia induced by 5-minute forearm cuff occlusion. The coefficient of variation for baseline and hyperemic diameters were 0.5% and 0.7%, respectively.

Doppler flow was assessed at baseline and during reactive hyperemia using a 3.75 MHz carrier frequency and with correction for the insonation angle [41]. Mean baseline and hyperemic flow velocities were analyzed from digitized audio data using semiautomated signal averaging (Cardiovascular Engineering, Waltham, MA). Baseline and deflation flow measurements were reproducible on repeated analysis of 30 subjects with correlations of >0.98.

Genotyping methods

The Overview [37] details the genotyping performed with the Affymetrix 100K SNP GeneChip http://gmed.bu.edu/about/genotyping.html[34] and with the Marshfield STR marker set at the Mammalian Genotyping Service http://research.marshfieldclinic.org/genetics.

Statistical methods

We generated normalized sex-specific residuals adjusting for the following covariates: for the echocardiographic phenotypes, age, sex, height, weight, smoking, systolic and diastolic blood pressure, hypertension treatment; for ETT measures, age, sex, body mass index, baseline heart rate, diabetes, smoking, ratio of total to high-density lipoprotein cholesterol, and treatment for hypertension (additional adjustments for select variables is detailed in Table 1); for BA function, a set of 15 covariates previously reported [14] to be associated with endothelial function in our sample (see Table 1). Covariates were from the same exam as the phenotype measures. Next, we used residuals for the phenotypes of interest to test for potential association with 100K SNPs using additive family-based association tests (FBAT) and linear regression models with general estimating equations (GEE; additive genetic models) to account for correlation among related individuals from nuclear families, as detailed in the Overview [37]. We chose 70,987 SNPS for association analysis that met the following criteria: autosomal SNPs with genotypic call rate ≥80%, minor allele frequency ≥10%, Hardy-Weinberg equilibrium test p ≥0.001, and ≥10 informative families for FBAT. The choice of an 80% genotyping call rate threshold may appear unusually liberal. We chose this threshold to be more inclusive in terms of associations reported. Also, the algorithm for the genotype calls was the Dynamic Modeling algorithm, which is less precise than other algorithms that have been introduced more recently. The association analyses were complemented by linkage analyses that used variance components methods and a subset of 100K markers and Marshfield STRs; the selection of markers and methods for calculating identity-by-descent are also described in the Overview [37].

Table 1 Echocardiographic, exercise testing and brachial artery function traits analyzed in participants with 100K genotype data
Full size table

We used an unfiltered approach and report the top 25 SNPs associated with echocardiographic traits, ETT measures and BA phenotypes (15, 5 and 5 SNPs, respectively; relative proportions chosen empirically because of the larger number of echocardiographic traits analyzed) according to their degree of statistical significance (lowest p values) in GEE and FBAT models separately. For echocardiographic phenotypes, we analyzed the mean of values for traits averaged across the 4 examinations, as well as traits at individual examinations separately. In order to evaluate potential pleiotropic effects, we examined SNP associations across related sets of traits and listed the top 25 SNPs with the lowest geometric mean of p values for all echocardiographic traits (averaged across the four examinations), for all ETT measures and for all BA traits (15, 5 and 5 SNPs, respectively, for the 3 groups). Because we analyzed individual echocardiographic traits at each of the four examinations, we also listed the top SNP associations based on the geometric mean of p values for these individual echocardiographic traits across the four examinations. Thus, we use the term 'pleiotropic effects' to refer to whether there were SNPs that were associated with multiple traits within the 3 subgroups. Additionally, we examined associations of SNPs in or within 200 Kb of the start or terminus of six selected genes (ACE, AGT, AGTR1, ADRB1, VEGF, NOS3) that have been previously reported to be associated with Echo, ETT and BA function phenotypes [30, 42–51]. We view these analyses as exploratory because the coverage of the Affymetrix 100K GeneChip for these genes was quite limited.

Results

Table 1 lists the phenotypes analyzed from the three groups (Echo, ETT and BA endothelial function), the number of traits evaluated within each group, and the covariates included in regression models to create residuals. We observed moderate to high heritability of most of the traits evaluated (Table 1; estimates are multivariable-adjusted, for covariates noted above under methods, and listed in table footnote). Heritability estimates were 52% for aortic root dimension, 36–40% for LV mass, internal dimensions and LVWT, and 25% for LA size. Estimates for ETT measures varied from 41% for post-exercise recovery heart rate, 28% for exercise systolic blood pressure, and 16–25% for other phenotypes. For BA function, baseline flow velocity and vessel diameter were most heritable (32 and 25%, respectively) and hyperemic flow the least (6%), with intermediate values for FMD (19%).

Results of all association analyses and detailed linkage results are web-posted at http://ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?id=phs000007, the database of genotype and phenotype public repository (dbGaP) at the National Center for Biotechnology Information. Overall, nine SNPs yielded a p value <10-5 in either GEE or FBAT models for Echo traits. Eleven SNPs were associated with ETT traits with a p value <10-5, and one SNP yielded a p value below this threshold for BA function traits. A conservative Bonferroni correction for the number of statistical tests (0.05/1,000,000) yields an approximate threshold of genome-wide significance of 5*10-8.

Table 2A displays the 25 most significantly associated SNPs in GEE analyses (15 for echo phenotypes; 5 each for ETT and BA phenotypes; additive genetic models) sorted by p values along with the corresponding p value using FBAT. The top SNPs associated with averaged Echo traits included: rs1379659 and rs666088 (both in SLIT2) with LV diastolic dimension, rs1935881 (FAM5C) and rs10493389 (PDE4B) with LA diameter, and rs10504543 (KCNB2) with LV systolic dimension. The top SNPs associated with ETT traits included: rs6847149 (NOLA1) and rs2056387 (RYR2) with stage 2 exercise heart rate, and rs2553268 (WRN) with stage 2 exercise systolic blood pressure. The top 5 SNPs associated with BA traits included rs3814219 (OBFC1) and rs10515508 (NRG2) with BA baseline flow, and rs4148686 (CFTR) with FMD. Only one of these SNPs (SLIT2) had a p value < 10-3 in FBAT.

Table 2 Top associations for Echo, ETT and BA function traits based on lowest p value for GEE test (2a), FBAT (2b), and Linkage (2c)*
Full size table

Table 2B lists the top 25 SNPs associated with phenotypes in FBAT along with corresponding p values in GEE models. Only one of these SNPs (rs1959290 associated with LV systolic dimension) had a p value < 10-3 in GEE models. The top 5 SNPs associated with ETT traits included rs1029947 and rs1029946 (both in PRKAG2) with heart rate at 3 minutes of post-exercise recovery. The top 5 SNPs associated with BA traits included rs2389866 (PDE5A).

Table 2C lists the magnitude and the location of loci with LOD scores that exceeded 2.0. The peak LOD scores were: for Echo traits, 4.38 (chromosome 5) for LV mass and 4.23 (chromosome 3) for aortic root size; for ETT traits, 2.93 (chromosome 5), 2.46 (chromosome 1), and 2.43 (chromosome 22) for heart rate during stage 2 of exercise; for BA traits, 2.14 (chromosome 8) for baseline flow velocity.

Table 3 evaluates the potential pleiotropic effect of SNPs by evaluating the geometric mean of p values for associations across averaged echo, and single-exam ETT and BA traits, and by relating them to the individual Echo traits across 4 examinations. SNPs associated with 4 genes (SLIT2, WDR72, UBE2L3 and KCNB2) were related to individual echo traits across examinations, as well as associated with the averaged Echo traits. Likewise, RYR2 was associated with a low geometric p value when related to all ETT traits across the examination.

Table 3 Top associations ordered by geometric mean of GEE p-values across traits (3 groups) and across examinations (echo traits)
Full size table

Table 4A–C displays results of association of Echo, ETT and BA function traits with SNPs in proximity to 6 genes (within 200 Kb of the start or terminus) chosen from the published literature; as noted earlier, the Affymetrix 100K GeneChip does not cover SNPs within these genes adequately. We observed weak associations of several SNPs in proximity to the genes of interest and Echo, ETT and BA function traits, none of which had a p value < 10-3). It is noteworthy, though, that several SNPs in proximity to ADRB1, AGT, and AGTR1 were associated weakly (p value of 0.05 to 10-3) with several Echo, ETT and BA function phenotypes.

Table 4 Associations of traits with SNPs in or near (up to 200 kb away) 6 well-replicated genes in the published literature with a p-value < 0.05 in either FBAT or GEE.
Full size table

Discussion

Principal findings

We report results of GWAS of Echo, ETT and BA function traits in a moderate-size community-based sample using several complementary analytical approaches. Our principal findings are five-fold. First, we observed modest to strong evidence of heritability for several Echo, ETT and BA function traits, underscoring the contribution of additive genetic effects to interindividual variation in these traits. Our heritability findings confirm prior reports for some of the traits [18, 20, 22, 23, 27, 28, 52], including from our group [14, 24]. Second, notwithstanding the modest-to-high heritability, none of the SNP-trait associations we observed achieved genome-wide significance (conservative Bonferroni correction p of 5*10-8). Therefore, any associations presented should be viewed as hypothesis-generating, with need for replication in additional samples. Third, our investigation highlights some of the challenges inherent in the interpretation of GWAS results. We did not observe any overlap between the top SNPs noted in GEE-based versus FBAT-based analyses, in part due to the inherent differences in the two analytical methods (see Overview for details [37]). Fourth, notwithstanding the lack of genome-wide statistical significance, our data do suggest several interesting biological candidates among the SNPs most strongly associated with different traits in the various analytical approaches (see discussion below). Fifth, we were quite limited in our ability to replicate findings for genetic variants previously associated with the traits that we investigated because specific coverage of such genetic variation in these candidates was limited in the Affymetrix 100K GeneChip. Therefore, the lack of replication of SNPs in proximity to 6 genes previously reported to be associated with Echo, ETT and BA traits should be interpreted with great caution. It is interesting that several weak associations (p between 0.05 and 10-3) were observed between traits in the three groups and SNPs in proximity to selected candidate genes evaluated (ADRB1, AGT and AGTR1).

Potential biological candidates among observed associations

In our GWAS of Echo traits, a SNP in SLIT2 was associated with Echo LV diastolic dimension in several analyses. SLIT2 is an evolutionarily highly conserved gene that encodes a putative secreted protein, which contains conserved protein-protein interaction domains including leucine-rich repeats and epidermal growth factor-like motifs [53]. The gene has multiple effects but has been recently identified to have a novel role in vascular function by contributing to migratory mechanisms in vascular smooth muscle cells [54]. Likewise, the associations of LV mass with HSPA8, and of LA size with PDE4B are consistent with the key role of heat shock protein expression [55] and T-cell mediated immune responses [56], respectively, in myocardial hypertrophic responses to insults or hemodynamic overload.

Analyses of ETT traits provided some interesting results. The association of a SNP in RYR2 with exercise heart rate responses in multiple analyses is quite consistent with the fundamental role of the ryanodine receptor on the sarcoplasmic reticulum in calcium trafficking during cardiac muscle excitation-contraction coupling [57]. Furthermore, RYR2 has been implicated in exercise-induced polymorphic ventricular tachyarrhythmias [58]. Using FBAT, SNPs in PRKAG2 were associated with heart rate during the recovery period post-exercise. Mutations in PRKAG2, an enzyme that modulates glucose uptake and glycolysis [59], are associated with glycogen-filled vacuoles in cardiomyocytes. The phenotypic manifestations include cardiac hypertrophy, ventricular pre-excitation and conduction system disturbances, encompassed together in the Wolff-Parkinson-White syndrome [60].

Genetic linkage analyses of ETT traits identified peaks on chromosomes 5 and 22 for exercise heart rate. The 1.5 LOD support intervals for these peaks included MEF2C and MAPK1, respectively. MEF2C is a critical regulator of cardiac morphogenesis [61]. Additionally, overexpression of MEF2C in experimental studies is associated with disturbances in extracellular matrix remodeling, ion handling, and metabolism of cardiomyocytes [62]. The peak on MAPK1 is of interest because a recent investigations highlighted the role of MAPK signaling in mediating the responses of skeletal muscles to exercise training [63].

A SNP in NRG2 was associated with BA flow velocity at rest, and also was in proximity to the top LOD peak for LV mass, raising the possibility of pleiotropic effects of this gene on ventricular and vascular remodeling and function. NRG2, which encodes neuregulin-2, is a member of the epidermal growth factor (EGF) family and binds to ErbB receptors. ErbB signaling has been implicated in angiogenesis and endothelial cell proliferation [64]. Of interest, a SNP in CFTR was associated with FMD. It is noteworthy that CFTR is expressed in vascular smooth muscle cells and activation of CFTR chloride channels regulates contraction and relaxation of smooth muscle cells; disruption of the CFTR gene prevents cAMP-dependent vasorelaxation in experimental studies [65]. CFTR is also expressed in endothelial cells where it functions as a cyclic nucleotide-regulated chloride channel [66]. Of interest, a SNP in PDE5A was associated with BA hyperemic flow velocity in FBAT analyses. Phosphodiesterase 5 (PDE5) hydrolyzes cyclic guanosine monophosphate (cGMP) and cyclic adenosine monophosphate (cAMP), is widely expressed in the vasculature, and is best known as the target of sildenafil, a drug used to treat erectile dysfunction [67]. PDE5 degrades cGMP in smooth muscle cells so as to maintain the contracted state of blood vessels [67]. PDE5A may also play a critical role in the growth promoting effects of Angiotensin II on vascular smooth muscle cells [68].

Strengths and limitations

The moderate-sized community-based sample, availability of longitudinal Echo measurements, routine ascertainment of standardized and reproducibly-measured traits, and evaluation of multiple complementary analytical methods including assessment of potential pleiotropic genetic effects strengthen our investigation. By web-posting unfiltered aggregate data we provide a resource for the scientific community to conduct in silico replication. Nonetheless, several limitations must be emphasized. As noted previously, the lack of genome-wide significance for any association observed given the extent of multiple statistical testing does not exclude a potential role of genetic influences on the traits studied. We had limited statistical power to detect modest genetic effects, given the sample size and the extent of multiple testing. As detailed in the Overview paper [37], for a conservative alpha level such as 10-8, we have more than 90% power to detect an association with a SNP explaining 4% or more total phenotypic variation when 80% or more individuals are phenotyped. We also had limited ability to replicate previously reported findings, in view of the partial coverage of genetic variation in select candidates with the Affymetrix 100K gene chip. Additionally, genetic variants may influence phenotypes in a context-specific manner [69], being modulated by environmental influences. For instance, the associations of ACE and AGTR2 with LV mass were reported to vary according to dietary salt intake in one investigation [48]. We did not undertake an investigation of gene-environmental interactions in the present study. Likewise, some of the moderately strong associations may represent false-positive results, notwithstanding the evidence suggesting that some of the associated SNPs may be reasonable biological candidates. We averaged echocardiographic traits across multiple examinations, with a view to characterizing the phenotype better over a period of time using several observations. Such a strategy could limit regression dilution bias, if the examinations are repeated over a short period of time. In our study, however, these examinations spanned a time period of twenty years, and the examinations used different echocardiographic equipment that may introduce misclassification. Further, such averaging assumes that similar sets of genes and environmental factors influence traits over a wide age range. Such an assumption may not be true, i.e., age-dependent gene effects may be masked by averaging of observations across ages in participants. Lastly, our sample was white and of European descent. The generalizability of our findings to other ethnicities is unknown.

Conclusion

In hypothesis-generating GWAS of Echo, ETT response and BA vascular function in a moderate size community-based sample, we identified several SNPs that are potential candidates for replication. Overall, our investigation provides a scientific framework for analyzing and interpreting GWAS of phenotypes fundamental to our understanding of cardiac and vascular remodeling and hemodynamic responses to exercise testing. We expect the Framingham 100K SNP data to serve as a valuable scientific resource by virtue of the web-posting of unfiltered aggregate data