Abstract

Candidate gene polymorphisms related to inflammation, thrombosis and lipid metabolism have been implicated in the development of ischemic stroke. Using DNA samples collected at baseline in a prospective cohort of 14 916 initially healthy American men, we genotyped 92 polymorphisms from 56 candidate genes among 319 individuals who subsequently developed ischemic stroke and among 2092 individuals who remained free of reported cardiovascular disease over a mean follow-up period of 13.2 years to prospectively determine whether candidate gene polymorphisms contribute to stroke risk. After adjustment for multiple comparisons and age, smoking, body mass index, hypertension, hyperlipidemia and diabetes, two related to inflammation [a val640leu polymorphism in the P-selectin gene (OR=1.63, 95% CI 1.22–2.17, P=0.001) and a C582T polymorphism in the interleukin-4 gene (OR=1.40, 95% CI 1.13–1.73, P=0.003)] were found to be independent predictors of thrombo-embolic stroke. In bootstrap replications, the inclusion of genetic information from these two polymorphisms improved prediction models for stroke based upon traditional risk factors alone (ROC 0.67 versus 0.64). Two polymorphisms related to thrombosis (an arg353gln polymorphism in the factor VII gene and a T11053G polymorphism in the plasminogen activator inhibitor type-1 gene) and one related to lipid metabolism [a C(-482)T polymorphism in the apolipoprotein CIII gene] achieved nominal significance, but were not found to be independent predictors after multiple comparison adjustment. Two inflammatory candidate gene polymorphisms were identified which were independently associated with incident stroke. These population-based data demonstrate the ability of prospective, epidemiological studies to test candidate gene associations for athero-thrombotic disease.

INTRODUCTION

Completion of the initial sequencing of the human genome project brings substantial promise to clinical medicine, including the potential to develop genetic panels for the assessment of disease risk (1). One approach to this issue is the evaluation of selected polymorphisms in genes, which are suspected of being associated with disease, either because they are known to encode for specific proteins related to the disease process or because they lie within chromosomal regions identified in linkage studies. Although retrospective case–control studies are frequently used to assess relationships between candidate markers and disease endpoints, a preferred setting for such analyses is large prospective population cohorts where an adequate number of individuals can be evaluated to ensure stable estimates of the background allele and genotype frequencies, and where the number of incident events is sufficient to provide both informative null data as well as narrow estimates for any observed positive effects. If the population under study derives from a homogeneous prospective cohort of initially healthy individuals such that case status is defined solely by the subsequent development of disease, inadvertent epidemiological bias arising from case selection criteria or from population stratification can be minimized (2,3).

With regard to occlusive cardiovascular events, candidate genes of interest include those associated with lipid metabolism, thrombosis and hemostasis, and inflammation. To begin an exploration of these issues, we undertook a large-scale evaluation of 92 candidate gene polymorphisms related to these processes within the context of a prospective cohort of initially healthy American men who were followed over a 13.2 year period for the occurrence of first ever cardiovascular events. In this report, we present our findings for the ischemic stroke component of this ongoing study.

RESULTS

Table 1 lists the 92 polymorphisms examined and the observed frequency of the less common allele at each site among the 2092 control subjects. Among the 71 polymorphisms with frequency >0.05, seven displayed Hardy–Weinberg disequilibrium with a P-value less than 0.05 (none were less than 0.01) before Bonferroni correction. After Bonferroni correction, all alleles tested (those with frequency >0.05) demonstrated Hardy–Weinberg equilibrium.

Baseline characteristics of participants who subsequently developed stroke (cases) and of those who remained free of cardiovascular disease during follow-up (controls) are shown in Table 2. As expected, participants who developed stroke had a higher prevalence of conventional athero-sclerotic risk factors at baseline.

A full listing of genotype frequencies for the case and control participants as well as calculated odds ratios associated with additive, dominant or recessive models of inheritance for each of the 92 individual polymorphisms evaluated is available from the authors in electronic format. For ease of presentation, Table 3 presents those polymorphisms which, in any mode of inheritance, were found to have a nominal univariable P-value less than 0.10 for association with incident thrombo-embolic stroke. As shown, 13 polymorphisms were found in this initial analysis.

Following our conservative a priori analysis plan, two marker associations remained promising after full adjustment for multiple comparisons and after adjustment for other risk factors, the dominant effect of the val640leu P-selectin polymorphism and the additive effect of the C582T interleukin-4 polymorphism. Specifically, in stepwise multivariable risk factor adjusted analyses, the val640leu polymorphism in the P-selectin gene (OR=1.63, 95% CI=1.22–2.17, P=0.001) and the C582T polymorphism in the interleukin-4 gene (OR=1.40, 95% CI=1.13–1.73, P=0.003) were found to be independent predictors of thrombo-embolic stroke. For these, the conservative Bonferroni P-values were 0.08 for both and the permutation-adjusted P-values were 0.09 for both, despite simultaneous adjustment for 92 comparisons. P-values for all other polymorphisms after the Bonferroni adjustment and the permutation test were greater than 0.75. Perhaps of greater relevance for a genetic analysis, the false discovery rate (FDR) for each was only 0.039. From a statistical perspective, this FDR implies that both are likely to represent true-positive findings (4,5).

Of the original 92 polymorphisms, only five (Table 4) were found to be independent genetic predictors of risk in the traditional age- and smoking-adjusted forward selection multivariable models with a P-value less than 0.05. These five included the two polymorphisms significant in the FDR analysis above as well as three others; an arg353gln polymorphism in the gene coding for coagulation factor VII (OR in a dominant model of 0.70, 95% CI=0.52–0.94, P=0.017); a C(-482)T polymorphism in the apolipoprotein CIII gene (OR in a dominant model of 1.31, 95% CI=1.03–1.66, P=0.029); and a T11053G polymorphism in the plasminogen activator inhibitor type-1 gene (OR in a recessive model of 0.71, 95% CI=0.51–0.98, P=0.040). However, in analyses using the BIC as the selection criterion, only the two polymorphisms significant in the FDR analysis remained in the model, providing further statistical justification for the importance of these two findings.

We calculated c-statistics as a method to compare the clinical efficacy of stroke risk prediction models based upon genetic data from the P-selectin and interleukin-4 polymorphisms and traditional cardiovascular risk factors to risk-prediction models based upon traditional cardiovascular risk factors alone. In this analysis, the c-statistic for the final risk factor adjusted model for stroke which included these two polymorphisms as compared to those found in models not using genetic data were 0.68 and 0.66, respectively, a marginal gain in discriminatory accuracy. In bootstrap replications of the final stroke model, these c-statistics became 0.67 and 0.64, respectively; a negligible correction was thus observed after the bootstrap replications. In addition, using internally cross-validated estimates of the log likelihoods, the contribution of the two gene polymorphisms found to be associated with stroke remained highly significant overall (P=0.0004).

DISCUSSION

In this initial report from a prospective, population-based genetic–epidemiological evaluation of initially healthy American men, we found two inflammatory polymorphisms to be strongly associated with incident stroke, after adjustment for multiple comparisons.

From a research perspective, we believe these data provide several insights, which if corroborated in other cohorts, could provide direction in the search for new targets for therapeutic intervention. In particular, the observation in these data of statistically significant effects on stroke in association with polymorphisms in the P-selectin and interleukin-4 genes is of considerable patho-physiologic interest given prior data from this and other cohorts implicating plasma levels of several cell adhesion molecules, cytokines, and other inflammatory mediators as key determinants of athero-thrombotic risk. The fact that the contribution of these two gene polymorphisms remained independent and highly significant upon cross-validation (P<0.001) suggests that the products of these genes—both of which appear involved in several endothelial processes—may represent important targets in the prevention of thrombo-embolic stroke.

For interleukin-4, the encoded protein is a pleiotropic cytokine produced by activated T cells, is a ligand for the interleukin-4 receptor, and also binds to interleukin 13. Serum levels of interleukin-4 have been found to be significantly elevated in patients with cerebral infarction (6,7). In addition, prior work regarding the interleukin-4 C582T polymorphism has been shown to play a role in various inflammatory and immune disorders (813).

With regard to P-selectin, this platelet membrane protein—also called GMP-140 or CD62—mediates the interaction of platelets and leukocytes with activated endothelial cells. Genetic variants in P-selectin have been suggested as risk factor for myocardial infarction (1416), suggesting a similar functional role in cerebro-vascular disorders.

However, another possible explanation is that the observed association of the two significant polymorphisms found with incident stroke may result from linkage disequilibrium with a yet-to-be-identified nearby susceptibility locus(i) or gene(s). This would require linkage disequilibrium analysis of the gene considered for association.

In addition to these two polymorphisms associated with inflammation, two polymorphisms associated with hemostasis and thrombosis were also found to be associated with incident stroke at nominal significance levels. One of these, an arg353gln polymorphism in the factor VII gene, has previously been reported to influence plasma levels of expressed factor VII activity (17,18). The current data thus provide independent cohort support for prior results regarding the potential role of this polymorphism as a risk factor for athero-thrombosis (19,20), a finding which has been controversial in the published literature (18,21). We also found that a T11053G polymorphism in the plasminogen activator inhibitor (PAI-1) gene is associated with incident stroke, with the common T allele more frequently observed in cases than controls; an intriguing finding as PAI-1 excess has long been associated with athero-thrombotic risk (22). While these latter results are suggestive, more data are needed to provide statistical support for an association.

From clinical and epidemiological perspectives, our evaluations of model fit and c-statistic data provide preliminary evidence that genetic markers may in the future prove to have clinical utility. However, clinical application of this concept will require several additional steps. Most importantly, confirmation of these data in other cohorts and in different populations is required since, despite the low false discovery rates observed and the high level of significance found in our internal cross-validation analyses, these data must be viewed as hypothesis generating. Our results should not be generalized to women or to other groups with different absolute levels of stroke risk, such as stroke at young age.

Despite the intriguing nature of these findings, we in general believe appropriate clinical caution should be used when interpreting positive results from any association study. Epidemiological limitations, which have the potential to lead to false-positive findings, include inadequate sample size, failure to ensure that affected and unaffected subjects derive from the same source population, over-reliance on post hoc subgroup analyses, and selective presentation of results without consideration of the adverse effects which can arise due to multiple comparisons. With regard to these concerns, strengths of our study design include its adequate sample size and the fact that we used a closed prospective cohort in which the determination of case status was based solely on the subsequent development of disease rather than on any arbitrary selection criteria designed by the investigators. Further, our genomic control analysis revealed no evidence of population stratification in these data. We also chose, on an a priori basis, to adjust for multiple comparisons using several different techniques, and to present all our data simultaneously rather than focusing on any one specific finding.

It is important to remember that many of our null findings were considered on an a priori basis to have a reasonable likelihood of success. For example, although we found some evidence of association between a C(-482)T polymorphism in the apolipoprotein CIII gene and incident stroke, we found little evidence for other lipid-based polymorphisms. Power to detect an effect, particularly after multiple comparison adjustment, requires consideration. However, based on our sample size, at a nominal alpha level of 0.05 as in the stepwise models, we have 80% power to detect an odds ratio of 1.47 with an allele frequency of 0.10 and an odds ratio of 1.31 with an allele frequency of 0.30. The probability of inclusion in Table 3, with a nominal significance level of 0.10, is even higher. Thus, polymorphisms that are false negatives after our conservative multiple comparison adjustment are likely to be among those in Table 3, which are worthy of investigation in other data.

In summary, in this prospective, population-based study, two candidate gene polymorphisms were identified which were independently associated with incident stroke. It is particularly intriguing that these two polymorphisms with the lowest false-discovery rates relate to interleukin-4 and P-selectin, key inflammatory mediators known to play central roles in vascular occlusion. These data and the statistical approaches applied to them thus demonstrate the ability of genetic-epidemiological studies to assist in the evaluation of candidate gene polymorphisms for athero-thrombotic disease.

MATERIALS AND METHODS

Participants

We studied prospectively collected DNA samples from a cohort of apparently healthy middle-aged men participating in the Physicians' Health Study (23). In brief, 22 071 predominantly Caucasian (>92%) US male physicians 40–84 years old who were free of prior myocardial infarction, stroke, transient ischemic attack and cancer were enrolled, of whom 14 916 (68%) provided baseline blood samples which were used for genetic analysis. Over an average period of 13.2 years, follow-up has been complete for all deaths and for 99.4% of all reported morbid events.

According to the overall study's nested case–control design, each participant who provided an adequate sample of whole blood at baseline and had a confirmed stroke, myocardial infarction, or venous thrombo-embolism during follow-up was matched wherever possible to two or three controls. The controls were study participants who had also provided a baseline blood sample and who remained free of any reported cardiovascular disease at the time the index event occurred in the case patient. Controls were selected at random from among those who met the matching criteria of age (±2 years), smoking habits (former, current or never) and time since study entry. As described elsewhere (24), for all reported incident vascular events occurring after study enrollment, hospital records, death certificates and autopsy reports were requested and reviewed by an end-points committee using standardized diagnostic criteria. With specific regard to stroke, the diagnosis was confirmed if the patient had a new, abrupt focal neurologic deficit and if the symptoms and signs persisted for more than 24 h. Computed tomography or magnetic resonance scans were available in all confirmed cases and were used, along with the clinical history, to exclude from analysis primary hemorrhagic stroke events. To increase power for statistical comparison, all controls were pooled in a common reference group. Employing these criteria, a total of 319 cases of incident ischemic stroke and 2092 controls were available for analysis.

For each case and control, whole blood collected and frozen at baseline was thawed and underwent DNA extraction (25,26). Genotyping was performed using multiplex PCR and linear immobilized probe array assays for candidate markers of cardiovascular disease, immune response, and inflammation (Roche Molecular Systems, Alameda, CA, USA), essentially as described previously (27,28). In brief, the candidate genes examined were selected from biochemical pathways that have been implicated in the development and progression of cardiovascular disease. In addition to the biological relevance of the selected candidate genes, the polymorphisms were further selected based on prior evidence of potential functionality, validated allele frequency and heterozygosity, and sequence-proven allelic variation. The selected genetic polymorphisms focused broadly on the athero-sclerotic pathway, including genes involved in lipid metabolism, inflammation, cell adhesion, thrombosis and hemostasis, and platelet function.

Extracted DNA samples from cases and controls were analyzed together in blinded fashion with the sequence varied at random within groups to avoid systematic bias. To evaluate accuracy of genotyping, we compared results for specific polymorphisms which had previously been evaluated in this cohort (25,26,29,30) with those obtained in the current analysis. Reproducibility was in excess of 99% with most instances of error being due either to transcription or data entry. Where discrepancy was found, samples underwent repeat genotyping. The study was approved by the Brigham and Women's Hospital's Institutional Review Board for Human Subjects Research.

Statistical analysis

We sought evidence of association between each of the 92 polymorphisms evaluated and incident thrombo-embolic stroke in a multi-stage process.

First, for descriptive purposes, we calculated crude allele and genotype frequencies for each individual polymorphism and evaluated Hardy–Weinberg equilibrium using a one-degree of freedom goodness-of-fit test among controls (31). We then compared the genotype frequencies between case and control participants using logistic regression analyses to compute relative risks and 95% confidence intervals. To address the use of a pooled control group, all analyses were unmatched and adjusted for the matching factors of age and smoking status. To avoid assumptions regarding modes of inheritance, all analyses were performed using additive (homozygote-common versus heterozygote versus homozygote-rare), dominant (homozygote-common versus heterozygote and homozygote-rare) or recessive (homozygote-common and heterozygote versus homozygote-rare) modes for each polymorphism. Acknowledging the possibility of falsely excluding a potential true positive effect, we present results for those polymorphisms with a nominal P-value of 0.10 or less in the initial results.

Second, we performed forward-stepwise multivariable logistic regression analyses using a nominal P-value cut-point of 0.05 to evaluate evidence that specific polymorphisms were independently associated with thrombo-embolic stroke, after adjusting for age and smoking; entry to the model was a nominal P-value of 0.10, with a final cut-point of 0.05 to remain. Stepwise-selected polymorphisms were then entered into a risk factor-adjusted analysis which further controlled for baseline body mass index, hypertension, hyperlipidemia, diabetes, and randomized treatment assignment to aspirin or beta-carotene. In addition, we used the Bayes Information Criterion (BIC) to select variables for inclusion after adjustment for the number of terms selected (32). We recognized a priori that the large number of polymorphism analyzed could lead to a high number of false positive findings. There are several statistical methods which address this issue including Bonferroni adjustment which is highly conservative and low in power. Permutation testing is somewhat less conservative than the Bonferroni approach in that it allows and adjusts for correlations among variables considered. Both of these procedures control the experiment-wise error rate (33). An alternative approach based on controlling the FDR (4) has been advocated for analyses of genetic markers (5,34). This procedure, computed using PROC MULTTEST of SAS, does not control the experiment-wise error rate, but estimates the proportion of errors among the rejected hypotheses. For the purposes of epidemiological comparison, we chose on an a priori basis to calculate all three of these measures separately. Because the mode of transmission is uncertain for most of the alleles considered, we performed these tests on the additive effects, thus adjusting for 92 independent comparisons.

Finally, we examined and adjusted measures of fit for the multivariable models. C-statistics, representing the area under the receiver operator characteristic (ROC) curve, were used to estimate discrimination, the probability that a given final model correctly assigned high-risk (35). Because participants were age- and smoking-matched, the c-statistics do not reflect the overall fit of a predictive model including these terms. A comparison of c-statistics, however, can indicate the genetic contribution to model fit. In order to cross-validate the fit of these models, we also computed estimates of the c-statistics and likelihood ratios based on 100 bootstrap samples (36,37). The bootstrap procedure selects random samples of size n with replacement from the original data. Repeating the sampling procedure a large number of times provides information on the variability and validity of the parameter estimate.

On a post hoc basis, we used genomic control to examine the potential impact of population stratification in these results (38). In these data, the median chi-square value for additive effects of all 92 polymorphisms was 0.38, indicating that no correction was necessary (39).

ACKNOWLEDGEMENTS

This work was supported by grants from the National Heart Lung and Blood Institute (HL-58755, HL-63293), the Doris Duke Charitable Foundation and the American Heart Association. We would like to thank the participants, staff and investigators of the Physicians' Health Study for their long-term commitment to this project. We also wish to thank Michael Grow and Arkadiy Silbergleit for their work in developing and optimizing the RMS CVD panel, and Karen Walker and Gabriele Zangenberg for their work in developing and optimizing the RMS Infla panel used for this study.

Table 1.

Genetic polymorphisms evaluated in the study

Gene Symbol Polymorphism Chromosome Frequency-controls (less common allele) 
Alpha-adducin ADD1 gly460trp 4p16.3 0.187 
Angiotensin II receptor type 1 AGTR1 A1166C 3q21–25 0.297 
Angiotensinogen AGT met235thr 1q42–43 0.438 
Angiotensin-converting enzyme ACE del/ins 17q23 0.446 
Apolipoprotein(a) LPA C93T 6q26–27 0.145 
  G121A  0.176 
Apolipoprotein AIV APOA4 thr347ser 11q23-ter 0.194 
  gln360his  0.087 
Apolipoprotein B APOB thr71ile 2p24 0.284 
  arg3500gln  0.00024 
Apolipoprotein CIII APOC3 C(−641)A 11q23.2–23.3 0.386 
  C(−482)T  0.266 
  T(−455)C  0.378 
  C1100T  0.262 
  C3175G  0.104 
  T3206G  0.391 
Apolipoprotein E APOE cys112arg 19q13.2 0.139 
  arg158cys  0.083 
Atrial natriuretic factor NPPA G664A 1p36 0.049 
  T2238C  0.140 
Beta-2 adrenergic receptor ADRB2 gly16arg 5q31–32 0.378 
  gln27glu  0.410 
  thr164ile  0.013 
Beta-3 adrenergic receptor ADRB3 trp64arg 8p12–11.1 0.070 
Beta-3 guanine nucleotide-binding protein GNB3 C825T 12p13 0.317 
Chemokine receptor 2 CCR2 val62ile 3p21 0.096 
Chemokine receptor 3 CCR3 pro39leu 3p21.3 0.006 
Chemokine receptor 5 CCR5 ins(32bp)del 3p21 0.109 
  G59029A  0.454 
Cholesteryl ester transfer protein CETP C(−630)A 16q13 0.069 
  C(−628)A  0.495 
  ile405val  0.325 
  asp442gly  0.00024 
  G(+1)A−(+3)ins T  — 
Eotaxin SCYA11 ala23thr 7q21.1–21.2 0.178 
Factor II F2 G20210A 11p11–q12 0.019 
Factor V F5 arg506gln 1q23 0.025 
Factor VII F7 del/ins 13q34 0.146 
  arg353glu  0.138 
Fc fragment of IgE, High affinity I, receptor for; beta polypeptide MS4A1 glu237gly 11q12–13 0.034 
Fibrinogen beta FGB G(−455)A 4q28 0.207 
Hepatic lipase LIPC C(−480)T 15q21–22 0.230 
Intercellular adhesion molecular 1 ICAM1 lys56met 19p13.2 0.006 
  gly241arg  0.097 
Interferon gamma receptor 1 IFNGR1 val13met 6q23–24 0.00048 
Interleukin 1 alpha IL1A C(−1203)T 2q13 0.298 
  C(−889)T  0.309 
Interleukin 1 beta IL1B C4336T 2q13–21 0.237 
Interleukin 4 IL4 C582T 5q23–31 0.173 
Interleukin 4 receptor IL4R ile50val 16p12.1 0.449 
  gln576arg  0.204 
Interleukin 5 receptor alpha IL5RA G(−80)A 3p26–24 0.259 
Interleukin 6 IL6 G(−174)C 7p21–15 0.374 
Interleukin 9 IL9 A553C 5q22–32 — 
  thr113met  0.126 
Interleukin 10 IL10 C(−571)A 1q31–32 0.238 
Interleukin 13 IL13 gln130arg 5q31 0.203 
Lipoprotein lipase LPL T(−93)G 8p22 0.021 
  asp9asn  0.012 
  asn291ser  0.014 
  ser447term  0.112 
Low density lipoprotein receptor LDLR +NcoI 19p13.2 0.301 
5,10-methylenetetrahydrofolate reductase MTHFR C677T 1p36.3 0.353 
Monocyte differentiation antigen CD14 CD14 C(−260)T 5q22–32 0.484 
Nitric oxide synthase 3 NOS3 A(−922)G 7q36 0.369 
  C(−690)T  0.081 
  glu298asp  0.310 
Paraoxonase 1 PON1 leu55met 7q21–22 0.368 
  gln192arg  0.304 
Paraoxonase 2 PON2 ser311cys 7q21.3 0.218 
Peroxisome proliferator activated-receptor gamma PPARG pro12ala 3p25 0.123 
Plasminogen activator inhibitor type 1 PAI1 4G/5G 7q22.1–22.3 0.468 
  T11053G  0.443 
Platelet glycoprotein Ia ITGA2 G873A 5q23–31 0.394 
Platelet glycoprotein IIIa ITGB3 leu33pro 17q21.32 0.147 
Selectin E SELE ser128arg 1q23–25 0.109 
  leu554phe  0.046 
Selectin P SELP val640leu 1q23–25 0.110 
  thr756pro  0.101 
  ser330asn  0.204 
Sodium channel, epithelial, alpha subunit SCNN1A trp493arg 12p13 0.019 
  ala663thr  0.332 
Stromelysin 1 MMP3 6A/5A 11q22.2–22.3 0.493 
Transforming growth factor, beta 1 TGFB1 C(−509)T 19q13.2 0.345 
Tumor necrosis factor alpha TNF G(−376)A 6p21.3 0.019 
  G(−308)A  0.154 
  G(−244)A  0.002 
  G(−238)A  0.060 
Tumor necrosis factor beta LTA thr26asn 6p21.3 0.310 
Tumor necrosis factor receptor 1 TNFRSF1A pro(A)12pro(G) 12p13 0.421 
Uteroglobin UGB G(+38)A 11q12.3–13.1 0.337 
Vascular cell adhesion molecule 1 VCAM1 T(-1594)C 1p32–31 0.183 
Gene Symbol Polymorphism Chromosome Frequency-controls (less common allele) 
Alpha-adducin ADD1 gly460trp 4p16.3 0.187 
Angiotensin II receptor type 1 AGTR1 A1166C 3q21–25 0.297 
Angiotensinogen AGT met235thr 1q42–43 0.438 
Angiotensin-converting enzyme ACE del/ins 17q23 0.446 
Apolipoprotein(a) LPA C93T 6q26–27 0.145 
  G121A  0.176 
Apolipoprotein AIV APOA4 thr347ser 11q23-ter 0.194 
  gln360his  0.087 
Apolipoprotein B APOB thr71ile 2p24 0.284 
  arg3500gln  0.00024 
Apolipoprotein CIII APOC3 C(−641)A 11q23.2–23.3 0.386 
  C(−482)T  0.266 
  T(−455)C  0.378 
  C1100T  0.262 
  C3175G  0.104 
  T3206G  0.391 
Apolipoprotein E APOE cys112arg 19q13.2 0.139 
  arg158cys  0.083 
Atrial natriuretic factor NPPA G664A 1p36 0.049 
  T2238C  0.140 
Beta-2 adrenergic receptor ADRB2 gly16arg 5q31–32 0.378 
  gln27glu  0.410 
  thr164ile  0.013 
Beta-3 adrenergic receptor ADRB3 trp64arg 8p12–11.1 0.070 
Beta-3 guanine nucleotide-binding protein GNB3 C825T 12p13 0.317 
Chemokine receptor 2 CCR2 val62ile 3p21 0.096 
Chemokine receptor 3 CCR3 pro39leu 3p21.3 0.006 
Chemokine receptor 5 CCR5 ins(32bp)del 3p21 0.109 
  G59029A  0.454 
Cholesteryl ester transfer protein CETP C(−630)A 16q13 0.069 
  C(−628)A  0.495 
  ile405val  0.325 
  asp442gly  0.00024 
  G(+1)A−(+3)ins T  — 
Eotaxin SCYA11 ala23thr 7q21.1–21.2 0.178 
Factor II F2 G20210A 11p11–q12 0.019 
Factor V F5 arg506gln 1q23 0.025 
Factor VII F7 del/ins 13q34 0.146 
  arg353glu  0.138 
Fc fragment of IgE, High affinity I, receptor for; beta polypeptide MS4A1 glu237gly 11q12–13 0.034 
Fibrinogen beta FGB G(−455)A 4q28 0.207 
Hepatic lipase LIPC C(−480)T 15q21–22 0.230 
Intercellular adhesion molecular 1 ICAM1 lys56met 19p13.2 0.006 
  gly241arg  0.097 
Interferon gamma receptor 1 IFNGR1 val13met 6q23–24 0.00048 
Interleukin 1 alpha IL1A C(−1203)T 2q13 0.298 
  C(−889)T  0.309 
Interleukin 1 beta IL1B C4336T 2q13–21 0.237 
Interleukin 4 IL4 C582T 5q23–31 0.173 
Interleukin 4 receptor IL4R ile50val 16p12.1 0.449 
  gln576arg  0.204 
Interleukin 5 receptor alpha IL5RA G(−80)A 3p26–24 0.259 
Interleukin 6 IL6 G(−174)C 7p21–15 0.374 
Interleukin 9 IL9 A553C 5q22–32 — 
  thr113met  0.126 
Interleukin 10 IL10 C(−571)A 1q31–32 0.238 
Interleukin 13 IL13 gln130arg 5q31 0.203 
Lipoprotein lipase LPL T(−93)G 8p22 0.021 
  asp9asn  0.012 
  asn291ser  0.014 
  ser447term  0.112 
Low density lipoprotein receptor LDLR +NcoI 19p13.2 0.301 
5,10-methylenetetrahydrofolate reductase MTHFR C677T 1p36.3 0.353 
Monocyte differentiation antigen CD14 CD14 C(−260)T 5q22–32 0.484 
Nitric oxide synthase 3 NOS3 A(−922)G 7q36 0.369 
  C(−690)T  0.081 
  glu298asp  0.310 
Paraoxonase 1 PON1 leu55met 7q21–22 0.368 
  gln192arg  0.304 
Paraoxonase 2 PON2 ser311cys 7q21.3 0.218 
Peroxisome proliferator activated-receptor gamma PPARG pro12ala 3p25 0.123 
Plasminogen activator inhibitor type 1 PAI1 4G/5G 7q22.1–22.3 0.468 
  T11053G  0.443 
Platelet glycoprotein Ia ITGA2 G873A 5q23–31 0.394 
Platelet glycoprotein IIIa ITGB3 leu33pro 17q21.32 0.147 
Selectin E SELE ser128arg 1q23–25 0.109 
  leu554phe  0.046 
Selectin P SELP val640leu 1q23–25 0.110 
  thr756pro  0.101 
  ser330asn  0.204 
Sodium channel, epithelial, alpha subunit SCNN1A trp493arg 12p13 0.019 
  ala663thr  0.332 
Stromelysin 1 MMP3 6A/5A 11q22.2–22.3 0.493 
Transforming growth factor, beta 1 TGFB1 C(−509)T 19q13.2 0.345 
Tumor necrosis factor alpha TNF G(−376)A 6p21.3 0.019 
  G(−308)A  0.154 
  G(−244)A  0.002 
  G(−238)A  0.060 
Tumor necrosis factor beta LTA thr26asn 6p21.3 0.310 
Tumor necrosis factor receptor 1 TNFRSF1A pro(A)12pro(G) 12p13 0.421 
Uteroglobin UGB G(+38)A 11q12.3–13.1 0.337 
Vascular cell adhesion molecule 1 VCAM1 T(-1594)C 1p32–31 0.183 
Table 2.

Baseline characteristics of study participants who subsequently developed ischemic stroke (cases), and those who remained free of vascular disease during follow-up (controls)

 Controls (n=2092) Thrombo-embolic stroke (n=319) P 
Age (years) 58.4±0.2 60.7±0.5 0.01 
Smoking status (%)   0.15 
 Never 43.5 39.3  
 Past 41.3 43.4  
 Current 15.3 17.3  
Body mass index (kg/m225.0±0.1 25.6±0.2 0.002 
Blood pressure (mmHg)    
 Systolic 127.4±0.3 133.6±0.8 <0.001 
 Diastolic 79.0±0.2 82.1±0.4 <0.001 
Hyperlipidemia (%) 8.3 12.7 0.014 
Diabetes (%) 2.7 10.7 <0.001 
Family history of premature CAD (%) 11.3 11.7 0.82 
Aspirin use (%) 49.5 50.5 0.75 
Beta-carotene use (%) 52.7 50.2 0.39 
 Controls (n=2092) Thrombo-embolic stroke (n=319) P 
Age (years) 58.4±0.2 60.7±0.5 0.01 
Smoking status (%)   0.15 
 Never 43.5 39.3  
 Past 41.3 43.4  
 Current 15.3 17.3  
Body mass index (kg/m225.0±0.1 25.6±0.2 0.002 
Blood pressure (mmHg)    
 Systolic 127.4±0.3 133.6±0.8 <0.001 
 Diastolic 79.0±0.2 82.1±0.4 <0.001 
Hyperlipidemia (%) 8.3 12.7 0.014 
Diabetes (%) 2.7 10.7 <0.001 
Family history of premature CAD (%) 11.3 11.7 0.82 
Aspirin use (%) 49.5 50.5 0.75 
Beta-carotene use (%) 52.7 50.2 0.39 

Mean±SE.

Table 3.

Estimated effects for polymorphisms selected in univariable, forward-stepwise multivariable, and in risk factor-adjusted analyses of ischemic stroke

Polymorphism Mode Crudea Stepwise multivariableb Stepwise multivariable, risk-factor adjustedc FDRd 
  OR 95% CI P OR 95% CI P OR 95% CI P  
Apolipoprotein(a) C93T Recessive 0.35 0.11–1.14 0.081        
Apolipoprotein A4 gln360his Additive 0.75 0.54–1.04 0.082        
 Dominant 0.73 0.51–1.04 0.079        
Apolipoprotein C3 C(−482)T Additive 1.23 1.03–1.47 0.026        
 Dominant 1.32 1.04–1.67 0.024 1.31 1.03–1.66 0.029 1.28 0.99–1.65 0.061 0.59 
Atrial natriuretic factor G664A Additive 1.42 1.00–2.00 0.047        
 Dominant 1.39 0.97–1.99 0.075        
Chemokine receptor 5 G59029A Recessive 1.28 0.97–1.70 0.084        
Cholesteryl ester transfer protein C(−628)A Additive 0.85 0.72–1.01 0.062        
 Recessive 0.77 0.58–1.02 0.073        
Factor 7 del/ins Additive 0.78 0.60–1.01 0.059        
 Dominant 0.75 0.57–1.00 0.050        
arg353gln Additive 0.73 0.56–0.96 0.022        
 Dominant 0.71 0.53–0.95 0.020 0.70 0.52–0.94 0.017 0.65 0.47–0.89 0.008 0.59 
Interleukin 1 alpha C(−1203)T Additive 0.83 0.69–1.01 0.058        
 Dominant 0.79 0.62–1.00 0.051        
C(−889)T Additive 0.84 0.70–1.01 0.068        
 Dominant 0.81 0.64–1.02 0.076        
Interleukin 4 C582T Additive 1.41 1.15–1.73 0.001 1.42 1.16–1.74 0.001 1.40 1.13–1.73 0.003 0.039 
 Dominant 1.48 1.16–1.89 0.002        
 Recessive 1.73 1.00–2.99 0.050        
Plasminogen activator inhibitor type 1 T11053G Additive 0.84 0.71–1.00 0.051        
 Recessive 0.69 0.50–0.97 0.030 0.71 0.51–0.98 0.040 0.72 0.51–1.02 0.064 0.61 
Selectin P val640leu Additive 1.48 1.18–1.87 0.001        
 Dominant 1.59 1.22–2.08 0.001 1.58 1.21–2.06 0.001 1.63 1.22–2.17 0.001 0.039 
Polymorphism Mode Crudea Stepwise multivariableb Stepwise multivariable, risk-factor adjustedc FDRd 
  OR 95% CI P OR 95% CI P OR 95% CI P  
Apolipoprotein(a) C93T Recessive 0.35 0.11–1.14 0.081        
Apolipoprotein A4 gln360his Additive 0.75 0.54–1.04 0.082        
 Dominant 0.73 0.51–1.04 0.079        
Apolipoprotein C3 C(−482)T Additive 1.23 1.03–1.47 0.026        
 Dominant 1.32 1.04–1.67 0.024 1.31 1.03–1.66 0.029 1.28 0.99–1.65 0.061 0.59 
Atrial natriuretic factor G664A Additive 1.42 1.00–2.00 0.047        
 Dominant 1.39 0.97–1.99 0.075        
Chemokine receptor 5 G59029A Recessive 1.28 0.97–1.70 0.084        
Cholesteryl ester transfer protein C(−628)A Additive 0.85 0.72–1.01 0.062        
 Recessive 0.77 0.58–1.02 0.073        
Factor 7 del/ins Additive 0.78 0.60–1.01 0.059        
 Dominant 0.75 0.57–1.00 0.050        
arg353gln Additive 0.73 0.56–0.96 0.022        
 Dominant 0.71 0.53–0.95 0.020 0.70 0.52–0.94 0.017 0.65 0.47–0.89 0.008 0.59 
Interleukin 1 alpha C(−1203)T Additive 0.83 0.69–1.01 0.058        
 Dominant 0.79 0.62–1.00 0.051        
C(−889)T Additive 0.84 0.70–1.01 0.068        
 Dominant 0.81 0.64–1.02 0.076        
Interleukin 4 C582T Additive 1.41 1.15–1.73 0.001 1.42 1.16–1.74 0.001 1.40 1.13–1.73 0.003 0.039 
 Dominant 1.48 1.16–1.89 0.002        
 Recessive 1.73 1.00–2.99 0.050        
Plasminogen activator inhibitor type 1 T11053G Additive 0.84 0.71–1.00 0.051        
 Recessive 0.69 0.50–0.97 0.030 0.71 0.51–0.98 0.040 0.72 0.51–1.02 0.064 0.61 
Selectin P val640leu Additive 1.48 1.18–1.87 0.001        
 Dominant 1.59 1.22–2.08 0.001 1.58 1.21–2.06 0.001 1.63 1.22–2.17 0.001 0.039 

aCrude=univariable logistic regression analyses adjusting for age and smoking.

bMultivariable=forward step-wise multivariable logistic regression analyses adjusted for age and smoking.

cMultivariable-risk factor adjusted=further adjusted for BMI, history of hypertension and hyperlipidemia, presence or absence of diabetes, and randomized treatment assignment to aspirin or beta-carotene.

FDRd=false discovery rate adjusting for multiple comparisons in univariable tests of additive effects among 92 polymorphisms.

Table 4.

Genotype frequencies of the five stepwise-multivariable regression selected polymorphisms

 Controls Cases P 
Apolipoprotein C3 C(−482)T   0.026 
 CC 0.545 0.477  
 CT 0.380 0.433  
 TT 0.076 0.091  
Factor 7 arg353gln   0.022 
 Arg–Arg 0.743 0.803  
 Arg–Gln 0.239 0.185  
 Gln–Gln 0.018 0.013  
Interleukin 4 C582T   0.001 
 CC 0.688 0.602  
 CT 0.277 0.345  
 TT 0.034 0.053  
Plasminogen activator inhibitor type 1 11053G   0.051 
 TT 0.310 0.342  
 TG 0.494 0.511  
 GG 0.196 0.147  
Selectin P val640leu   0.001 
 Val–Val 0.795 0.712  
 Val–Leu 0.189 0.263  
 Leu–Leu 0.016 0.025  
 Controls Cases P 
Apolipoprotein C3 C(−482)T   0.026 
 CC 0.545 0.477  
 CT 0.380 0.433  
 TT 0.076 0.091  
Factor 7 arg353gln   0.022 
 Arg–Arg 0.743 0.803  
 Arg–Gln 0.239 0.185  
 Gln–Gln 0.018 0.013  
Interleukin 4 C582T   0.001 
 CC 0.688 0.602  
 CT 0.277 0.345  
 TT 0.034 0.053  
Plasminogen activator inhibitor type 1 11053G   0.051 
 TT 0.310 0.342  
 TG 0.494 0.511  
 GG 0.196 0.147  
Selectin P val640leu   0.001 
 Val–Val 0.795 0.712  
 Val–Leu 0.189 0.263  
 Leu–Leu 0.016 0.025  

References

1
Collins, F.S. (
1999
) Shattuck Lecture—medical and societal consequences of the human genome project.
New Engl. J. Med.
 ,
341
,
28
–37.
2
Ridker, P.M. and Stampfer, M.J. (
1999
) Assessment of genetic markers for coronary thrombosis: promise and precaution.
Lancet
 ,
353
,
687
–688.
3
Bogardus, S.T., Concato, J. and Feinstein, A. (
1999
) Clinical epidemiological quality in molecular genetic research. The need for methodological standards.
JAMA
 ,
281
,
1919
–1926.
4
Benjamini, Y. and Hochberg, Y. (
1995
) Controlling the false discovery rate: a practical and powerful approach to multiple testing.
J. R. Stat. Soc. B
 ,
57
,
289
–300.
5
Weller, J.I., Song, J.Z., Heyen, D.W., Lewin, H.A. and Ron, M. (
1998
) A new approach to the problem of multiple comparisons in the genetic dissection of complex traits.
Genetics
 ,
150
,
1699
–1700.
6
Elkarim, R.A., Mustafa, M., Kivisakk, P., Link, H. and Bakhiet, M. (
1998
) Cytokine autoantibodies in multiple sclerosis, aseptic meningitis and stroke.
Eur. J. Clin. Invest.
 ,
28
,
295
–299.
7
Kim, H.M., Shin, H.Y., Jeong, H.J., An, H.J., Kim, N.S., Chae, H.J., Kim, H.R., Song, H.J., Kim, K.Y., Baek, S.H., Cho, K.H., Moon, B.S. and Lee, Y.M. (
2000
) Reduced IL-2 but elevated IL-4, IL-6, and IgE serum levels in patients with cerebral infarction during the acute stage.
J. Mol. Neurosci.
 ,
14
,
191
–196.
8
Nanavaty, U., Goldstein, A.D. and Levine, S.J. (
2001
) Polymorphisms in candidate asthma genes.
Am. J. Med. Sci.
 ,
321
,
11
–16.
9
Maksymowych, W.P., Suarez-Almazor, M.E., Buenviaje, H., Cooper, B.L., Degeus, C., Thompson, M. and Russell, A.S. (
2002
) HLA and cytokine gene polymorphisms in relation to occurrence of palindromic rheumatism and its progression to rheumatoid arthritis.
J. Rheumatol.
 ,
29
,
2319
–2326.
10
Nakayama, E.E., Meyer, L., Iwamoto, A., Persoz, A., Nagai, Y., Rouzioux, C., Delfraissy, J.F., Debre, P., McIlroy, D., Theodorou, I. and Shioda, T.; SEROCO Study Group (
2002
) Protective effect of interleukin-4 −589T polymorphism on human immunodeficiency virus type 1 disease progression: relationship with virus load.
J. Infect. Dis.
 ,
185
,
1183
–1186.
11
Kwa, D., van Rij, R.P., Boeser-Nunnink, B., Vingerhoed, J. and Schuitemaker, H. (
2003
) Association between an interleukin-4 promoter polymorphism and the acquisition of CXCR4 using HIV-1 variants.
AIDS
 ,
17
,
981
–985.
12
Bijlsma, F.J., van Kuik, J., van Hoffen, E., de Jonge, N., Tilanus, M.G., Gmelig-Meyling, F.H. and de Weger, R.A. (
2002
) Acute cardiac transplant rejection is associated with low frequencies of interleukin-4 producing helper T-lymphocytes rather than with interleukin-4 promoter or splice variants.
Hum. Immunol.
 ,
63
,
317
–323.
13
Bijlsma, F.J., vanKuik, J., Tilanus, M.G., de Jonge, N., Rozemuller, E.H., van den Tweel, J.G., Gmelig-Meyling, F.H. and de Weger, R.A. (
2002
) Donor interleukin-4 promoter gene polymorphism influences allograft rejection after heart transplantation.
J. Heart Lung Transplant.
 ,
21
,
340
–346.
14
Bray, P.F. (
2000
) Platelet glycoprotein polymorphisms as risk factors for thrombosis.
Curr. Opin. Hematol.
 ,
7
,
284
–289.
15
Tregouet, D.A., Barbaux, S., Escolano, S., Tahri, N., Golmard, J.L., Tiret, L. and Cambien, F. (
2002
) Specific haplotypes of the P-selectin gene are associated with myocardial infarction.
Hum. Mol. Genet.
 ,
11
,
2015
–2023.
16
Herrmann, S.M., Ricard, S., Nicaud, V., Mallet, C., Evans, A., Ruidavets, J.B., Arveiler, D., Luc, G. and Cambien, F. (
1998
) The P-selectin gene is highly polymorphic: reduced frequency of the Pro715 allele carriers in patients with myocardial infarction.
Hum. Mol. Genet.
 ,
7
,
1277
–1284.
17
Bernardi, F., Marchetti, G., Pinotti, M., Arcieri, P., Baroncini, C., Papacchini, M., Zepponi, E., Ursicino, N., Chiarotti, F. and Mariani, G. (
1996
) Factor VII gene polymorphisms contribute about one third of the factor VII level variation in plasma.
Arterioscler. Thromb. Vasc. Biol.
 ,
16
,
72
–76.
18
Lane, A., Green, F., Scarabin, P.Y., Nicaud, V., Bara, L., Humphries, S., Evans, A., Luc, G., Cambou, J.P., Arveiler, D. and Cambien, F. (
1996
) Factor VII arg/gln353 polymorphism determines factor VII coagulant activity in patients with myocardial infarction (MI) and control subjects in Belfast and in France but is not a strong indicator of MI risk in the ECTIM study.
Atherosclerosis
 ,
119
,
119
–127.
19
Iacoviello, L., Di Casstelnuova, A., de Knijff, P., D'Orazio, A., Amore, C., Arboretti, R., Kluft, C. and Benedetta Donati, M. (
1998
) Polymorphisms in the coagulation factor VII gene and the risk of myocardial infarction.
New Engl. J. Med.
 ,
338
,
79
–85.
20
Girelli, D., Russo, C., Ferraresi, P., Olivieri, O., Pinotti, M., Friso, S., Manzato, F., Mazzucco, A., Bernardi, F. and Corrocher, R. (
2000
) Polymorphisms in the factor VII gene and the risk of myocardial infarction in patients with coronary artery disease.
New Engl. J. Med.
 ,
343
,
774
–780.
21
Doggen, C.J.M., Cats, V.M., Bertina, R.M., Reitsma, P.H., Vandenbroucke, J.P. and Rosendaal, F.R. (
1998
) A genetic propensity to high factor VII is not associated with the risk of myocardial infarction in men.
Thromb. Haemost.
 ,
80
,
281
–285.
22
Hamsten, A., Wiman, B., Defaire, U. and Blomback, M. (
1985
) Increased plasma level of a rapid inhibitor of tissue plasminogen activator in young survivors of myocardial infarction.
New Engl. J. Med.
 ,
313
,
1557
–1563.
24
Steering Committee of the Physicians' Health Study Research Group (
1989
) Final report of the aspirin component of the ongoing Physicians' Health Study.
New Engl. J. Med.
 ,
321
,
129
–135.
25
Ridker, P.M., Hennekens, C.H., Lindpaintner, K., Stampfer, M.J., Eisenberg, P.R. and Miletich, J.P. (
1995
) Mutation in the gene coding for coagulation factor V and the risk of myocardial infarction, stroke, and venous thrombosis in apparently healthy men.
New. Engl. J. Med.
 ,
332
,
912
–917.
26
Lindpaintner, K., Pfeffer, M.A., Kreutz, R., Stampfer, M.J., Grodstein, F., LaMotte, F., Buring, J. and Hennekens, C.H. (
1993
) A prospective evaluation of an angiotensin-converting-enzyme gene polymorphism and the risk of ischemic heart disease.
New Engl. J. Med.
 ,
332
,
706
–711.
27
Cheng, S., Pallaud, C., Grow, M., Scharf, S.J., Erlich, H.A., Klitz, W., Pullinger, C.R., Malloy, M.J., Kane, J.P., Siest, G. and Visvikis, S. (
1998
) A multilocus genotyping assay for cardiovascular disease.
Clin. Chem. Lab. Med.
 ,
36
,
561
–566.
28
Cheng, S., Grow, M.A., Pallaud, C., Klitz, W., Erlich, H.A., Visvikis, S., Chen, J.J., Pullinger, C.R., Malloy, M.J., Siest, G. and Kane, J.P. (
1999
) A multilocus genotyping assay for candidate markers of cardiovascular disease risk.
Genome Res.
 ,
9
,
936
–949.
29
Ridker, P.M., Hennekens, C.H. and Miletich, J.P. (
1999
) G20210A mutation in the prothrombin gene and the risk of myocardial infarction, stroke, and venous thrombosis in a large cohort of US men.
Circulation
 ,
99
,
999
–1004.
30
Zee, R.Y.L., Lindpaintner, K., Struk, B., Hennekens, C.H. and Ridker, P.M. (
2001
) A prospective evaluation of the CD14 C(-260)T gene polymorphism and the risk of myocardial infarction.
Atherosclerosis
 ,
154
,
699
–702.
31
Weir, B.S. (
1996
)
Genetic Data Analysis II: Methods for discrete population genetic data
 . Sinauer Associates, Sunderland, MA.
32
Schwartz, G. (
1978
) Estimating the dimension of a model.
Ann. Stat.
 ,
6
,
461
–464.
33
Westfall, P.H. and Young, S.S. (
1989
) P value adjustments for multiple tests in multivariable binomial models.
J. Am. Stat. Assoc.
 ,
84
,
780
–786.
34
Mosig, M.O., Lipkin, E., Khutoreskaya, G., Tchourzyna, E., Soller, M. and Friedman, A. (
2001
) A whole genome scan for quantitative trait loci affecting milk protein percentage in Israeli-Holstein cattle, by means of selective DNA pooling in a daughter design, using an adjusted false discovery rate criterion.
Genetics
 ,
157
,
1683
–1698.
35
Hanley, J.A. and McNeil, B.J. (
1982
) The meaning and use of the area under a receiver operating characteristic (ROC) curve.
Radiology
 ,
143
,
29
–36.
36
Efron, B. and Tibshirani, R. (
1993
)
An Introduction of the Bootstrap
 . Monographs on Statistics and Applied Probability. Chapman and Hall, New York.
37
Steyerberg, E.W., Harrell, F.E., Borsboom, G.J.M., Eijkemans, M.J.C., Vergouwe, Y. and Habbena, J.D.F. (
2000
) Internal validation of predictive models: efficiency of some procedures for logistic regression analysis.
J. Clin. Epidemiol.
 ,
54
,
774
–781.
38
Devlin, B. and Roeder, K. (
1999
) Genomic control for association studies.
Biometrics
 ,
55
,
997
–1004.
39
Bacanu, S.A., Devlin, B. and Roeder, K. (
2000
) The power of genomic control.
Am. J. Hum. Genet.
 ,
66
,
1933
–1944.