The study protocol was approved by the Johns Hopkins Institutional Review Board, and all subjects gave written informed consent. Subjects were identified from European American and African American families with premature coronary artery disease (CAD) through a proband with documented CAD prior to 60 years of age. Apparently healthy siblings of the probands, the adult offspring of both the probands and their siblings, and the coparents of the offspring were recruited for the Genetic Study of Aspirin Responsiveness (GeneSTAR). GeneSTAR was designed to examine genetic and environmental determinants of platelet function in response to low-dose aspirin therapy. Eligible subjects were ≥ 21 years of age and had no history of any coronary heart disease, thrombotic event, peripheral vascular disease, stroke, transient ischemic attacks, known derangement in hematologic profiles (aplastic anemia, Sickle Cell Disease, von Willebrand's Disease, Factor V Leiden), renal or hepatic failure, autoimmune diseases, glucocorticosteroid use, hemorrhagic event, measured blood pressure ≥ 180/105 mmHg or current pregnancy. Subjects were also excluded if they had an allergy or intolerance to ASA, baseline platelet count ≤ 100,000 or ≥ 500,000 cells/μL, hematocrit ≤ 30%, or white blood cell count ≥ 20,000 cells/μL. Anticoagulants, ASA, nonsteroidal anti-inflammatory agents, illicit drugs, were proscribed for 14 days before the baseline visit and during the study interval. Consumption of tea and coffee and flavenol-rich foods (egg, wine, grape juice, chocolate) and other dietary items known to affect platelet function (fish rich in omega-3 fatty acids and garlic) were prohibited for 24 hours and smoking for 12 hours prior to assessments. Subjects with zero aggregation of platelets to arachidonic acid in both whole blood and platelet-rich plasma (PRP) at the baseline visit and questionable adherence to the list of proscribed substances, were omitted from the study. At the end of the first screening visit, eligible subjects were given a supply of 81-mg ASA tablets and instructed to take 1 pill each day for 14 days. Adherence to ASA use during the study was assessed at the post-ASA visit using pill counts and an adapted standardized medication adherence questionnaire .
Blood for the measurement of platelet function in whole blood (WB) and platelet rich plasma (PRP) was obtained by venipuncture at the same time of day at baseline and following 14 days of ASA therapy. Baseline and post-ASA measures included aggregation, slope, and lag time in response to selected doses of 4 known platelet agonists, and urinary excretion of the thromboxane B2 metabolite 11- dehydrothromboxane B2 (Tx-M). Optical aggregation was measured in PRP in a PAP-4 aggregometer (Horsham, PA) after samples were stimulated with arachidonic acid (1.6 mmol/L, collagen (1, 2, 5 μg/mL), ADP (2, 10 mmol/L), and epinephrine (2, 10 μmol/L). Whole blood impedance aggregometry was measured in a Chrono-Log dual-channel lumiaggregometer (Havertown, PA) after samples were stimulated with arachidonic acid (0.5 mmol/L), collagen (1, 5 μg/mL), and ADP (10 μmol/L). Platelet function under shear stress was determined by the platelet function analyzer (PFA) test (PFA-100, Dade- Behring, Newark, DE). Whole blood was loaded into standard proprietary cartridges (Dade-Behring) containing collagen and epinephrine, and aperture closure time was recorded in seconds (maximum of 300 seconds).
Plasma fibrinogen was assayed in the Johns Hopkins Clinical Coagulation Laboratory and von Willebrand factor (vWF) was measured using enzyme-linked immunosorbent assays (DiaPharma, West Chester, OH).
Urine was collected at the same time of day for pre- and post-ASA measurements of Tx-M using an enzyme-linked immunosorbent assay (Cayman Chemical Co., Ann Arbor, MI) and values were normalized to urinary creatinine.
Cardiac Risk Factor Evaluation
Blood pressure was measured according to methods of the American Heart Association and hypertension was defined as the average of 4 resting blood pressures ≥ 140/90 mm Hg and/or the taking of antihypertensive medications. Current smoking was defined as any reported cigarette smoking within the past 30 days and nonsmoking status was validated by exhaled CO levels < 8 ppm on the average of 2 measurements. Height and weight were measured and body mass index was calculated as the weight (kg)/height squared (m2). Plasma glucose, total cholesterol, triglyceride, and HDL-cholesterol levels were measured after patients had fasted for 12 hours overnight. LDL cholesterol was estimated using the Friedewald equation. Diabetes was defined as self-reported diabetes with the use of diabetes medications, or measured glucose levels ≥ 126 mg/dl.
Microsatellite genotyping was performed at deCODE Genetics, Inc. with the standard deCODE 550 STR marker set (average spacing = 8 cM). There were 574 successfully released STR markers with an overall duplicate error rate (3 CEPH sample duplicates on each of 18 plates) of 0.1% and an overall Mendelian error rate of 2% on 99.7% of samples. The STR genotyping data were checked for Mendelian inconsistencies using PEDCHECK , and all persons with inconsistencies were removed prior to analysis. The most likely relationship between pairs of relatives was inferred using RELCHECK , and these were used to verify self-reported relationships. SIBPAIR (v 0.99.9) was used to calculate allele frequencies where the contribution of each pedigree is weighted by the number of founders it contains.
SNP genotyping was performed at deCODE Genetics, Inc. using the Human 1Mv1_C array from Illumina, Inc. and 1,044,977 markers were released with an average call rate per sample of 99.65% and an overall missing data rate of 0.35%. Using 25 duplicate pairs of CEPH samples, the reproducibility rate was >99.95% for all duplicate pairs. Analyses at deCODE Genetics revealed Mendelian errors (> 5%) in 14 samples, which were eliminated from further analysis. Finally, 9 samples with gender discrepancies and an additional 3,427,500 inconsistency calls over all SNPs were eliminated from the final data prior to analyses. While all markers were analyzed, SNPs were flagged for closer examination where minor allele frequency was low (2%) and/or deviation from Hardy Weinberg Equilibrium was severe (p-value < 10-6).
Given the large number of potentially different biologically-related platelet function cascades examined [10
], principal components analysis (PCA) was used to define a set of independent factors that explained a large proportion of the phenotypic co-variance. PCA was run separately for European Americans and African Americans within each of three major groups of outcome phenotypes:
baseline, representing native platelet function
post-aspirin, representing platelet function measures after 2 weeks of daily aspirin, and
post-aspirin platelet function adjusted for pre-aspirin platelet function, representing the change attributable to aspirin, or aspirin responsiveness
Prior to PCA, all platelet variables were first adjusted for age and sex, levels of LDL cholesterol, fibrinogen, and body mass index, and for the presence of diabetes, hypertension, and current smoking using linear regression models. For PFA test only, von Willebrand's factor was included in the adjustment. Following adjustment for covariates, there were 37 pre-ASA platelet function variables, 32 post-ASA variables, and 27 post-adjusted for pre-ASA variables within each race with distributions that were adequate to enable calculation of accurate Z-scores. All Z-scored variables were then used for PCA in PROC Factor in SAS version 9.1 implementing orthogonal varimax rotations. Eight components with Eigenvalues > 1 were identified for the pre-ASA variables, post ASA and post-adjusted for pre-ASA variables within each race. The components were labeled by their primary platelet phenotypic variables, defined as loadings > 0.4. These final PCA components were used in the tests for linkage and additional association analyses described below.
Linkage analysis was performed with the Haseman-Elston regression approach [15
] in SAGE (v 5.1.0) for each principal component and each STR for each ethnic group separately. In these analyses the traditional Haseman-Elston analysis for a quantitative trait (i.e. the regression of the squared traits difference on the IBD estimate) was performed on full sib relationships under a multipoint approach. This approach has been shown to be adequately robust and powered even for sample sizes with less than 300 sib-pairs [16
]. We used p-values defined by Lander and Kruglyak [17
] to represent specific LOD score thresholds:
suggestive linkage with p value = 0.00074 and corresponding LOD = 2.2
significant linkage with p-value = 0.00002 and corresponding LOD = 3.6; and
highly significant linkage with p- value = 0.0000003 and corresponding LOD = 5.4
Associations between SNPs on the GWAS panel and factors that had evidence for linkage based on the analyses described above were tested using linear mixed effects models (LME). In the formulation of the LME model, SNP genotypes were included as fixed effects setting the genotypes to be additive in effect, and family identification number was included in the random effects (essentially treating the correlations between all pairs of individuals in the family as equal). We tested whether the additive effects of each SNP was different from zero. The LME in SAS (v. 9.1.3 for Linux OS) was applied with PROC MIXED using the option for EMPIRICAL variance.
In an attempt to offset the diminished power to detect association that arises as a result of multiple testing issues inherent in genomic searches and to determine the best set of loci to pursue in further follow-up studies (i.e. to control false positive association signals) in the absence of external replication data, we used the False Discovery Rate (FDR) approach proposed by Roeder et al . This new weighted FDR methodology involves weighting the association test p-values on the basis of prior data derived from linkage. A combined map was obtained; interpolating the STRs with the SNP map by assigning a physical location (in Mb) to the mid-point of the STR. Using p-values from the linkage scan at these assigned physical locations, continuous linkage traces were derived from the standard normal cumulative distribution and used to weight the association p-values prior to the calculation of the FDR threshold. In this analysis we used Storey's FDR approach  which can be more powerful than that proposed by Benjamini and Hochberg  under the same error rate (here, alpha = 0.05).