Study design and population
The original (FHS) and offspring (FHSO) cohorts were launched in 1948 and 1970, respectively (Splansky et al., 2007; Govindaraju et al., 2008; Cupples et al., 2009). Briefly, the FHS includes N = 5209 respondents aged 28–62 years at baseline who have been biannually followed during about 60 years comprising 28 examinations. The FHSO respondents (N = 5124) aged 5–70 years at baseline were biological descendants (N = 3514), their spouses (N = 1576), and adopted offspring (N = 34) of the FHS participants who have been examined at seven visits. The FHS/FHSO participants have been followed for the onset of CVD through regular examinations at the FHS clinic, surveillance of hospital admissions, and death registries (Splansky et al., 2007; Govindaraju et al., 2008), currently through 2008. Biospecimens were mostly collected in the late 1980s and through 1990s from surviving participants (Myers et al., 1996; Lahoz et al., 2001; Cupples et al., 2009). The procedure used for the APOE genotyping is described in Lahoz et al. (2001). The data include information on the APOE e2/3/4 polymorphism for 1258 FHS and 3924 FHSO participants. Genome-wide data are available for 1529 FHS and 3750 FHSO participants.
We use data on longitudinally followed FHS and FHSO participants to comprehensively characterize the role of lipid-related genetic variants discovered in candidate gene (the APOE e2/3/4 polymorphism) and genome-wide (Teslovich et al., 2010) studies, aging-related processes, human generations as a proxy for environmental changes, and TC as a proxy for lipids in the onset of CVD. First, we characterized the role of genetic markers in the onset of CVD. Then, we elucidated the role of lipids in genetic effects on CVD. Next, we addressed direct associations of the selected genetic markers with TC. These analyses were conducted using data from different FHS/FHSO examinations to ensure that the detected associations are: (i) not the results of a specific stochastic realization; (ii) robust to longitudinal attrition of the samples at risk of CVD; and (iii) not the result of disproportional (i.e., genotype-specific) survival selection of robust individuals until biospecimen collection. Finally, to better characterize the role of the aging-related processes in associations of the selected genetic markers with TC, these associations were evaluated in the same longitudinally followed individuals at different chronological ages.
Associations of genetic markers with ages at onset of CVD in genotyped subjects were characterized by Kaplan–Meier estimator and the Cox proportional hazard regression model at different FHS/FHSO examinations. The time variable in the analyses was ‘age at onset of CVD (cases) or age at censoring (at death or the end of follow-up in 2008)'. Individuals who developed CVD prior to an examination in question were excluded from the respective analyses. Both the Kaplan–Meier estimator and the Cox regression model evaluate the probability of remaining free of CVD by each given age for each individual who was free of CVD at the start of the follow-up period. We used the robust sandwich estimator of variances in the Cox model to account for potential clustering (e.g., familial).
To address the role of lipid metabolism in genetic associations with CVD, we used two models. We evaluated individual and additive effects of TC and genetic markers on CVD in the same sample, i.e., with individuals missing TC measurements excluded, using the Cox regression model. Direct associations of the selected genetic markers with TC were evaluated using linear mixed effects regression models with unstructured covariance matrices parameterized in terms of variances and correlations to account for clustering. We used TC (log-base-10 transformed to adjust for deviations from normality) in the mixed effects model as a dependent variable.
The Cox and mixed effects models were adjusted for current age and sex, when applicable. These analyses were conducted separately for different examinations in the FHS/FHSO, using the data for all genotyped participants at each selected examination.
To characterize the role of the aging-related processes in genetic effects on TC, we focused on individuals who participated in all selected examinations, i.e., individuals who missed at least one examination were excluded from these analyses. We verified that inclusion of individuals who missed few examinations did not make qualitative difference. These analyses evaluated the associations in the same sample of individuals as they truly aged. We also evaluated the associations at different examinations. An advantage of the examination-specific estimates compared with modeling longitudinal changes is that the former approach is free of constraints on parameterization of longitudinal changes in TC.
Twenty-five SNPs discovered in the genome-wide association study by Teslovich et al. (2010) as associates of blood lipids were directly genotyped in the FHS. Of these, one single nucleotide polymorphism (SNP) was potentially important for our analyses, rs1042034 (chromosome two; nonsynonymous coding variant in the APOB gene) because it showed marginal significance with coronary artery disease and was associated with lipids (triglycerides and high-density lipoprotein cholesterol) in Teslovich et al. (2010), and had proportional hazards over age domain for different genotypes in the FHS cohort (examined in this study using the Kaplan–Meier empirical curves for the probability of staying free of CVD). This SNP was retained for the current analyses.
Although the phenotype-limited access data available for this study include information on TC and other lipids, the most comprehensive longitudinal measurements are available for TC only. Given our focus on the longitudinally followed population, only the TC has been retained for the current analyses, as a proxy for other lipids.
Blood lipids in early examinations of the FHS were measured without controlling for fast. In some FHS examinations, information on the fasting status of a limited number of individuals is available. We verified that fasting status makes no qualitative difference in our analyses [see Supplementary Information (SI), Table S1]. Accordingly, the entire genotyped samples, regardless of fasting status, were used in the analyses.
Basic characteristics of the genotyped study participants are given in Table S2 for each genotype of the APOE gene and rs1042034. Because the FHS cohort was followed for about 60 years, Table S2 indicates substantial differences in the proportions of diseased participants of the FHS and FHSO. The analyses were focused on carriers of the e4 (risk; e2/4, e3/4, and e4/4) and non-e4 (e2/2, e2/3, and e3/3) alleles (Kulminski et al., 2011). Empirical screening suggested the major allele-dominant model for the rs1042034. Because both genetic variants were selected based on prior evidence, corrections for multiple comparison were not required. Statistical analyses were conducted using sas (release 9.3, Cary, NC, USA).