Association of Gene Variants With Incident Myocardial Infarction in the Cardiovascular Health Study
Objective— We asked whether single nucleotide polymorphisms (SNPs) that had been nominally associated with cardiovascular disease in antecedent studies were also associated with cardiovascular disease in a population-based prospective study of 4522 individuals aged 65 or older.
Methods and Results— Based on antecedent studies, we prespecified a risk allele and an inheritance model for each of 74 SNPs. We then tested the association of these SNPs with myocardial infarction (MI) in the Cardiovascular Health Study (CHS). The prespecified risk alleles of 8 SNPs were nominally associated (1-sided P<0.05) with increased risk of MI in White CHS participants. The false discovery rate for these 8 was 0.43, suggesting that about 4 of these 8 are likely to be true positives. The 4 of these 8 SNPs that had the strongest evidence for association with cardiovascular disease before testing in CHS (association in 3 antecedent studies) were in KIF6 (CHS HR=1.29; 90%CI 1.1 to 1.52), VAMP8 (HR=1.2; 90%CI 1.02 to 1.41), TAS2R50 (HR=1.13; 90%CI 1 to 1.27), and LPA (HR=1.62; 90%CI 1.09 to 2.42).
Conclusions— Although most of the SNPs investigated were not associated with MI in CHS, evidence from this investigation combined with previous studies suggests that 4 of these SNPs are likely associated with MI.
Cardiovascular disease is a complex disease with a genetic component,1 and many genetic polymorphisms have been reported to be associated with cardiovascular disease.2 However, to confirm these associations, they should be examined in other populations, ideally in population-based prospective studies that have sufficient power to detect the hypothesized associations. One such population-based prospective study is the Cardiovascular Health Study (CHS), a study of American men and women 65 years and older sponsored by the National Heart, Lung, and Blood Institute.3,4 CHS offers several strengths, including a large population-based cohort, collection of baseline data for traditional risk factors, long follow-up, and central adjudication of cardiovascular events.
We have been investigating the association between cardiovascular disease and single nucleotide polymorphisms (SNPs) using a panel of ≈12 000 mostly nonsynonymous SNPs.5–7 The discovery studies for these investigations were conducted in case–control studies that included patients enrolled by investigators at the Cleveland Clinic Foundation (CCF) and the University of California, San Francisco (UCSF)5–7; the association between 9 of these SNPs and cardiovascular disease in multiple discovery studies has been previously described.5–9 We have used these 9 SNPs to build multiplex assays that are suitable for genotyping thousands of samples even when only a limited quantity of DNA is available for each sample. These multiplex assays also contain assays for 65 additional SNPs that were found to be associated with cardiovascular disease in one or more of the discovery studies (for these 65 SNPs, the results of the antecedent discovery studies are presented in the online supplement of this article, available online at http://atvb. ahajournals.org). We investigated whether the risk allele that was identified for each of these 74 SNPs in the antecedent studies would be associated with increased risk of MI in CHS.
Cardiovascular Health Study
CHS is a prospective observational study of risk factors for cardiovascular disease in older adults. Men and women aged 65 years and older were recruited from random samples of Medicare eligibility lists in 4 US communities (Sacramento County, Calif; Washington County, Md; Forsyth County, NC; and Pittsburgh, Pa) and from age-eligible participants in the same household. Potential participants were excluded if they were institutionalized, not ambulatory at home, under hospice care, receiving radiation or chemotherapy for cancer, not expected to remain in the area for at least 3 years, or unable to be interviewed. CHS enrolled 5201 participants in 1989 to 1990; an additional 687 African American participants entered the cohort in 1992 to 1993. The combined cohort of 5888 was 57.6% female and 15.7% African American. The mean age at enrollment was 72.8 years (standard deviation 5.6). Participants who did not donate DNA or who did not consent to the use of their DNA for studies by private companies (n=514) were excluded from the present study. Participants for whom DNA samples were inadequate (n=130) were also excluded. The institutional review board at each site approved the study methods, and all participants gave written informed consent. Details of CHS recruitment3 and design4 have been reported.
Participants completed a baseline clinic examination4 that included a medical history interview, physical examination, and blood draw.10 Baseline self-reports of MI or stroke were confirmed by information from the clinic examination or by review of medical records or physician questionnaires.11 Genotypes of the CHS participants were determined using a multiplex method that combines polymerase chain reaction (PCR), allele-specific oligonucleotide ligation assays, and hybridization to oligonucleotides coupled to Luminex 100TM xMAP microspheres (Luminex, Austin, TX). See online supplemental materials for details.
Diabetes mellitus was defined by fasting serum glucose of at least 126 mg/dL or the use of insulin or oral hypoglycemic medications.12 Impaired fasting glucose was defined as a fasting glucose of 110 to 125 mg/dL. Hypertension was defined by systolic blood pressure of at least 140 mm Hg, diastolic blood pressure of at least 90 mm Hg, or a physician’s diagnosis of hypertension plus the use of antihypertensive medications. Body mass index (BMI) was defined as body weight in kilograms divided by the square of height in meters.
Cardiovascular events during follow-up were identified at semiannual contacts, which alternated between clinic visits and telephone calls. Suspected events were adjudicated according to standard criteria by a physician review panel using information from medical records and, in some cases, interviews with the physician, participant, or a proxy informant.13 Medicare utilization files were searched to ascertain events that may have been missed. In this analysis, MI was defined as definite or probable nonfatal MI or definite fatal MI.
Prespecification of Risk Alleles and Inheritance Models for SNPs Investigated in CHS
For each of the 74 SNPs that were genotyped in CHS we prespecified a risk allele and an inheritance model based on antecedent data (supplemental Table I). For 9 of the 74 SNPs, genotypic association results have been previously published.5–9 The remaining 65 SNPs were associated with MI in at least 1 of 2 case–control studies described in the online supplementary text (supplemental text and supplemental Table II). An inheritance model for each SNP was prespecified using the following 3 rules: (1) for SNPs that had been previously reported to be associated with cardiovascular disease the inheritance model was based on the published data; (2) for SNPs that were nominally associated (P<0.1) in the 2 antecedent MI case–control studies reported in the online supplement (supplemental Table I) and had the same inheritance model in both studies, we used that model; (3) for all other SNPs we used an additive inheritance model. For example, if a SNP had the same risk allele in both studies and was nominally associated with MI (P<0.1) using an additive model in one study and using a recessive model in the second study, we used an additive model. We also used an additive model for SNPs that were associated with MI in only one of the two antecedent studies of MI.
Analyses excluded participants with a baseline history of MI (n=517 of the 5244 participants with genotype data) or stroke (n=222). Participants who were neither White nor African American were also excluded (n=30). Participant characteristics at baseline were described by counts and percents or means and standard deviations (Table 1).
Hardy–Weinberg equilibrium (HWE) tests were run for each SNP using the “genhw” procedure14 in Stata15 with corresponding Pearson chi-square tests; if either homozygote count was 5 or less, an exact test was used.
Because genetic risk factors can have different magnitude in Whites and in African Americans, we investigated the association of SNPs with incident MI in CHS in each race separately.
We conducted analyses of time to incident MI. Follow-up began at CHS enrollment and ended on the date of incident MI, death, loss to follow-up, or June 30, 2003, whichever occurred first. The median time at risk was 11.3 years for incident MI (12.7 years for the 1989 to 90 cohort and 10.1 years for the African American cohort).
Cox regression was used to estimate hazard ratios associated with each SNP in each race. Multivariate analyses were adjusted for baseline age and sex. Additional analyses were also adjusted for traditional risk factors: BMI, current smoking, diabetes or impaired fasting glucose, hypertension, LDL cholesterol, and HDL cholesterol. Because the expected direction of the effect (risk allele) was prespecified, we used a 1-sided probability value to test the significance of the coefficient associated with each SNP. Correspondingly, we estimated 90% confidence intervals for the hazard ratios (for hazard ratios greater than one, there is 95% confidence that a true risk estimate is greater than the lower bound of a 90% confidence interval). Data were analyzed using Stata statistical software.15
The expected influence of multiple testing was evaluated using 2 approaches. First, we used false discovery rates (FDR) as described by Benjamini and Hochberg16 to estimate the expected fraction of false-positives in a group of SNPs with probability values below a given threshold. For the 8 pairs of SNPs that are located in the same gene (rs529038 and rs619203 in ROS1; rs11016076 and rs10082504 in MKI67; rs3129196 and rs3130210 in LOC651870; rs7439293 and rs12510359 in PALLD; rs3813135 and rs892145 in PGLYRP2; rs428785 and rs402007 in ADAMTS1; rs2296436 and rs1804689 in HPS1; rs3749817 and rs13183672 in FSTL4), we included only the SNP with the higher (less significant) probability value in FDR calculations, which were performed with R statistical software.17 Second, false-positive report probabilities were calculated as described by Wacholder et al.18 Because assigning a prior probability is subjective, we used a range of prior probabilities to calculate a range of false-positive report probabilities for each SNP. The assumptions we used to determine the range of prior probabilities are described in the online supplement. The prior probability is directly proportional to the assumptions: alternative false-positive report probability estimates can be calculated by choosing different prior probability assumptions.
During 13 years of follow-up, 539 (12%) of the 4522 CHS participants in this analysis had an incident MI. We tested 74 SNPs separately in Whites and in African Americans for deviation from the genotype distribution expected under HWE and we found that 8 SNPs (5 in Whites and 3 in African Americans) deviated from HWE expectations (P<0.05, supplemental Table III). Had we adjusted the HWE test for multiple testing using a Bonferroni correction, none of the SNPs in African Americans, and only 3 of the SNPs in Whites would have deviated from HWE expectations. In Whites the 5 SNPs that nominally deviated from HWE expectations were rs3027309 in ALOX12B, rs11538264 in BAT2, rs11758242 in LY6G5B, rs402007 in ADAMTS1, and rs35690712 in SLC39A7. In African Americans the 3 SNPs were rs220479 in ITGAE, rs1804689 in HPS1, and rs3813135 in PGLYRP2. Because none of these SNPs deviated from HWE expectations in both Whites and African Americans, this deviation is unlikely to be attributable to genotyping error. Therefore we included all 74 SNPs in the analysis. Table 2⇓ lists all 74 SNPs and the genes in which they are located.
In Whites, 8 SNPs in 7 genes were nominally associated (P<0.05) with incident MI after adjustment for age and sex (Table 3). The associations between all 74 SNPs and MI in Whites are available in the supplemental Table IV. The 8 nominally associated SNPs were in KIF6, PGLYRP2 (2 SNPs), LPA, MCM10, VAMP8, DCC, and TAS2R50. We estimated the FDR for these 8 SNPs to be 0.43, indicating that about 4 of these SNPs are expected to be false-positives. When we considered the evidence for association with cardiovascular disease before testing in CHS, 4 of these 8 SNPs were among those with the strongest prior evidence (association in 3 studies after adjustment for multiple testing; supplemental Table I). The false-positive report probabilities for these 4 SNPs were all ≤0.01 (KIF6 [0.0005; range 0.0005 to 0.08], VAMP8 [0.005; range 0.002 to 0.31], TAS2R50 [0.005; range 0.003 to 0.33], and LPA [0.01; range 0.01 to 0.66]), suggesting that they are unlikely to be false-positives. In contrast, 2 of the SNPs (in MCM10 and DCC) had high false-positive report probability (> 0.9) indicating that they are likely to be false-positives, and the remaining 2 SNPs (both in PGLYRP2) had false-positive report probabilities that were intermediate (0.3). Adjustment for traditional risk factors did not appreciably change the risk estimates for these 8 SNPs although the association of the SNP in LPA was no longer nominally significant (P=0.069). Because we had previously observed that this LPA SNP was associated with plasma Lp(a) levels,7 we investigated the association of the LPA SNP with Lp(a) and found that carriers of the risk allele had a higher median level of Lp(a) (63 mg/dL) than non carriers (42 mg/dL, P<0.00005). However, for the MI end point, adjustment of the risk estimate of the LPA SNP to account for Lp(a) levels did not appreciably change the hazard ratio (HR=1.64, 90%CI; 1.10 to 2.45).
In African Americans, 3 SNPs were nominally associated with incident MI after adjustment for age, sex, and traditional risk factors (P<0.05, Table 4). The association between all 74 SNPs and MI in African Americans are available in supplemental Table V. One of these 3 SNPs (rs2213948) is located in an intergenic region; the other 2 SNPs are located in AQP10 and FCAR. This risk allele of the SNP in FCAR had been previously reported to be associated with increased risk of cardiovascular disease in the placebo arms of CARE and WOSCOPS.8 The estimated FDR for this set of 3 SNPs was 0.67. For the SNPs in VAMP8 and KIF6, which had the lowest false-positive report probabilities in White participants of CHS, the risk estimates in African Americans were high (1.71 [CI 0.92 to 3.19] for VAMP8 and 4.14 [CI 0.79 to 21.77] for KIF6) but did not reach statistical significance (P=0.08 for both).
We investigated the association between MI and 74 SNPs in CHS and found that 8 SNPs were nominally associated (P<0.05) with MI among White participants of CHS. The false discovery rate for these 8 SNPs was 0.43, suggesting that about 4 of these 8 are truly associated with MI.
Of these 8 SNPs, 4 had strong evidence for association with cardiovascular disease prior to testing in CHS. These 4 SNPs are located in KIF6, TAS2R50, VAMP8, and LPA. The strongest prior evidence for association with cardiovascular disease was for the SNPs in KIF6 and VAMP8. The SNP in KIF6 had been previously found to be associated with cardiovascular disease in the placebo arms of 2 statin trials, and the association remained significant after a Bonferroni correction for multiple testing.9 The SNP in VAMP8 had been found to be associated with MI in 3 case–control studies, with an FDR <0.1 in the third study.6 The risk alleles of these 2 SNPs have also been found to be associated with increased risk of coronary heart disease in the Atherosclerosis Risk in Communities (ARIC) study,19,20 a large population based prospective study of middle-aged Americans. Thus the SNPs in KIF6 and VAMP8 had been found to be consistently associated with cardiovascular disease prior to testing in CHS, and the associations found in CHS further strengthen the evidence for these associations. Furthermore, in African American participants of CHS, the risk estimates for the SNPs in VAMP8 and KIF6 were high (1.71 for VAMP8 and 4.14 for KIF6), although they did not reach statistical significance (P=0.08 for both). However, there was a smaller number of African American participants in this study (673 African Americans compared with 3849 Whites), and consequently, the power to detect association was lower among African Americans than among Whites, which could partially account for the lack of statistical significance of these risk estimates.
The SNPs in TAS2R50 and LPA were not associated with MI among African American participants of CHS. However, there are considerable differences in the LD structure of the LPA and TAS2R50 regions between Yoruba in Ibadan and CEPH (Utah residents with ancestry from Northern and Western Europe) populations.21 Thus, different SNPs in these 2 genes should be explored in African American populations to test whether other variants of these genes are associated with MI in this population.
For the SNP in LPA, the prior evidence was association with coronary stenosis in 3 case–control studies, association that remained significant after a Bonferroni correction for multiple testing in the third study.7 For the SNP in TAS2R50, the prior evidence was association with MI in 3 case–control studies, with a false discovery rate of <0.1 in the third study.5 However, these 2 SNPs (in LPA and TAS2R50) were not associated with coronary heart disease in ARIC.19
Because we tested 74 SNPs in CHS, we were concerned that multiple testing may have resulted in false-positive associations. To reduce the extent of multiple testing, we prespecified the risk allele and inheritance model for each SNP based on antecedent studies. Thus we tested a single hypothesis for each SNP. We used 2 different approaches to evaluate the extent to which multiple testing resulted in false-positives. The first method, FDR, is a frequentist approach that estimates the expected fraction of false-positives in a group of SNPs with probability values below a certain threshold.16 The FDR is computed from the nominal probability values and the number of independent tests. The group of 8 SNPs that were nominally associated (P<0.05) with MI in White participants of CHS had an FDR of 0.43, suggesting that about 4 of these SNPs are expected to be false-positives. However, none of the SNPs we tested in the white participants of CHS had an FDR lower than 0.1.
The second method we used to account for multiple testing was a Bayesian approach—false-positive report probability—that takes into consideration not only the observed probability value but also the power of the study to detect association and the prior probability of the SNP being associated with disease.18 We found that the false-positive report probabilities for the SNPs in KIF6, VAMP8, LPA, and TAS2R50 were all ≤0.01, suggesting that these 4 SNPs are unlikely to be false-positives. For the SNP in KIF6, even the high-end of the false-positive report probability range (0.08) suggests a low probability of being a false-positive. However, the high-end of the false-positive report probability range of the SNPs in VAMP8 (0.23), TAS2R50 (0.31), and LPA (0.66) indicated an intermediate probability of being false-positives when the more conservative end of the prior-probability range was used to estimate the false-positive report probability.
We have previously discussed the potential role LPA, VAMP8, TAS2R50, and KIF6 in cardiovascular disease,5–7,9 however the mechanisms by which the variants of these genes influence the pathophysiology of disease is unknown. Briefly, the SNP in LPA encodes the apolipoprotein(a) protein portion of the Lp(a) particle, a known risk factor for cardiovascular disease.22,23 We had previously reported that this SNP in LPA was associated with increased plasma levels of Lp(a).7 We have now confirmed this finding in CHS Whites. We also found that in CHS, the risk associated with this LPA SNP remains unchanged after adjustment for Lp(a) levels. The protein encoded by the VAMP8 gene plays a role in platelet degranulation.24 The TAS2R50 gene is a bitter taste receptor, and thus might be involved in food preference and diet.25 KIF6 encodes a member of the kinesin superfamily that plays a role in microtubule-mediated intracellular transport; however, its potential role in cardiovascular disease is unknown.
This study has several limitations. The antecedent studies that provided the prior evidence for the 74 SNPs were case–control studies, which might have resulted in selection and survival bias. Furthermore, because DNA limitations required the use of multiplexed assays for genotyping the CHS subjects, not all SNPs that were associated with disease in the antecedent studies were tested in CHS and some of the SNP included in the multiplexed assays had only been associated with cardiovascular disease in a single antecedent study. In this genetic study of CHS, we have not formally tested for population stratification, which could confound genetic association studies. However, because none of the 4 SNPs that are most likely to be true positives deviated from HWE expectations, these associations are unlikely to be confounded by population stratification. Additionally, participants in CHS were older than 65 at baseline (median 72 years); therefore, because cardiovascular disease heritability decreases with age,1 it may be more difficult to identify genetic associations in this population. Furthermore, in this older population gene variants might be associated with MI because they affect disease pathways that are particularly important in older individuals.
In summary, we found that 4 gene variants that have strong prior evidence for their association with cardiovascular disease were also associated with incident MI in CHS. This study suggests that even in older adults, genetic variation may affect cardiovascular risk.
The authors thank the CHS participants. A full list of participating CHS investigators and institutions can be found at http://www.chs-nhlbi.org.
Sources of Funding
The research reported in this article was supported by contracts N01-HC-15103, N01-HC-35129, N01-HC-45133, N01-HC-55222, N01-HC-75150, N01-HC-85079 through N01-HC-85086, and U01 HL080295 from the National Heart, Lung, and Blood Institute, with additional contribution from the National Institute of Neurological Disorders and Stroke. R.P.T. was supported by NIH RO1 HL077499, and J.P.K. and M.J.M. were supported by the Leducq Foundation.
D.S., L.A.B., C.M.R., J.Z.L., A.R.A., O.I., M.M.L., B.A.Y., and J.J.D. have employment and ownership interests. E.S.O., M.J.M., J.P.K., and R.P.T. and received research grants, and S.G.E. has consulting fees.
Original received August 16, 2007; final version accepted October 22, 2007.
Lusis AJ, Fogelman AM, Fonarow GC. Genetic basis of atherosclerosis: part I: new genes and pathways. Circulation. 2004; 110: 1868–1873.
Shiffman D, Ellis SG, Rowland CM, Malloy MJ, Luke MM, Iakoubova OA, Pullinger CR, Cassano J, Aouizerat BE, Fenwick RG, Reitz RE, Catanese JJ, Leong DU, Zellner C, Sninsky JJ, Topol EJ, Devlin JJ, Kane JP. Identification of four gene variants associated with myocardial infarction. Am J Hum Genet. 2005; 77: 596–605.
Shiffman D, Rowland CM, Louie JZ, Luke MM, Bare LA, Bolonick JI, Young BA, Catanese JJ, Stiggins CF, Pullinger CR, Topol EJ, Malloy MJ, Kane JP, Ellis SG, Devlin JJ. Gene variants of VAMP8 and HNRPUL1 are associated with early-onset myocardial infarction. Arterioscler Thromb Vasc Biol. 2006; 26: 1613–1618.
Luke MM, Kane JP, Liu DM, Rowland CM, Shiffman D, Cassano J, Catanese JJ, Pullinger CR, Leong DU, Arellano AR, Tong CH, Movsesyan I, Naya-Vigne J, Noordhof C, Feric NT, Malloy MJ, Topol EJ, Koschinsky ML, Devlin JJ, Stephen G Ellis SG. A Polymorphism in the protease-like domain of apolipoprotein(a) is associated with severe coronary artery disease. Arterioscler Thromb Vasc Biol. In press.
Iakoubova OA, Tong CH, Chokkalingam AP, Rowland CM, Kirchgessner TG, Louie JZ, Ploughman LM, Sabatine MS, Campos H, Catanese JJ, Leong DU, Young BA, Lew D, Tsuchihashi Z, Luke MM, Packard CJ, Zerba KE, Shaw PM, Shepherd J, Devlin JJ, Sacks FM. Asp92Asn polymorphism in the myeloid IgA Fc receptor is associated with myocardial infarction in two disparate populations: CARE and WOSCOPS. Arterioscler Thromb Vasc Biol. 2006; 26: 2763–2768.
Iakoubova OA, Tong CH, Rowland CM, Kirchgessner TG, Young BA, Arellano AR, Shiffman D, Sabatine MS, Campos H, Packard CJ, Pfeffer MA, White TJ, Braunwald E, Shepherd J, Devlin JJ, Sacks FM, MD. Association of the Trp719Arg polymorphism in kinesin-like protein 6 with myocardial infarction and coronary heart disease in two prospective studies: the Cholesterol and Recurrent Events Study (CARE) and the West of Scotland Coronary Prevention Study (WOSCOPS). J Amer Coll Cardiol. 2007; 27: 2030–2036.
Cushman M, Cornell ES, Howard PR, Bovill EG, Tracy RP. Laboratory methods and quality control in the Cardiovascular Health Study. Clin Chem. 1995; 41: 264–270.
American Diabetes Association. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care. 1997; 20: 1183–1197.
Cleves MA. Hardy-Weinberg Equilibrium Tests and Allele Frequency Estimation. STATA Technical Bulletin. No. 48, March 1999, 34–37. URL http://www.biostat-resources.com/stata/.
StataCorp. Stata Statistical Software: Release 9. College Station, TX: StataCorp LP, 2005.
Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B. 1995; 57: 289–300.
R Development Core Team. R: A language and environment for statistical computing, reference index version 2.3.0. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www. R-project.org, 2005.
Wacholder S, Chanock S, Garcia-Closas M, El Ghormli L, Rothman N. Assessing the probability that a positive report is false: an approach for molecular epidemiology studies. J Natl Cancer Inst. 2004; 96: 434–442.
Morrison AC, Bare LA, Chambless LE, Ellis SG, Malloy MJ, Kane JP, Pankow JS, Devlin JJ, Willerson JT, Boerwinkle E. Coronary heart disease risk prediction using a genetic risk score: the Atherosclerosis Risk in Communities (ARIC) Study. Am J Epidemiol. 2007; 166: 28–35.
Berg K. Lp(a) lipoprotein: an overview. Chem Phys Lipids. 1994; 67–68:9–16.
Polgar J, Chung SH, Reed GL. Vesicle-associated membrane protein 3 (VAMP-3) and VAMP-8 are present in human platelets and are required for granule secretion. Blood. 2002; 100: 1081–1083.