Pedigrees
Replicates 4, 10, and 21 were used in all analysis. Families with more than 20 individuals were excluded from the data set to reduce computational time. Two hundred and seventy seven pedigrees were analyzed with a mean of 11 members (minimum 7 and maximum 20 members). Of the 3155 subjects in the pedigrees, 33% had genotype data available.
Systolic blood pressure phenotypes
Six phenotype models describing SBP were defined; hypertension, a cross-sectional measure of SBP, and four models derived from longitudinal data, Models 1–4. Absence or presence of hypertension was used as a qualitative phenotype; individuals were defined as affected if they had a diagnosis of hypertension at any examination. Eighty-five families, containing enough affected members with genotype data, contributed to linkage analysis in each of the three replicates.
Taking the SBP value at the first examination for each subject generated a cross-sectional measure of systolic SBP. No one in replicates 4, 10, or 21 was recorded as receiving antihypertensive treatment at the first time interval; therefore, no adjustment for treatment was necessary.
Longitudinal SBP data were analyzed with a subject-specific approach (e.g., ref [2]). All systolic BP measures taken over time were included in the analysis. However, some subjects on some occasions were receiving antihypertensive medications. Since their recorded systolic BP was lower than it would have been if they had not been on treatment, these observations were treated as censored. Therefore, to account for the presence of both repeated measurements and right censoring, a mixed probit-normal or Tobit model [3] with a subject-specific random intercept was estimated using the program GLLAMM (generalized linear latent and mixed models) in STATA 7 [4].
For an untreated SBP measurement y
it
made on individual i at time t the model was specified as y
it
= βX
it
+ u
i
+ e
it
, where X
it
is a vector of covariates (including a constant) that may vary over time, β is a vector of regression coefficients, u
i
is a N(0, σu2) subject-specific random effect, eit are N(0, σe2) disturbance terms with corr(u,e.) = 0 for all t and corr(e.s,e.t) = 0 for s ≠ t. For censored observations Pr[Y >y
it
] = Φ ((βX
it
+ u
i
- y
ij
) / σ
e
,1), where Φ(.,1) is the standard normal cumulative density function.
The empirical Bayes' estimates of the individual random effects (the subject level residuals {
}) were extracted for use as adjusted longterm SBP phenotypes for input to SOLAR [5]. Fitted to males and females separately, four different adjustments for covariates were considered: Model 1: age, age squared and body mass index; Model 2: covariates in Model 1 plus cohort; Model 3: covariates in Model 2 plus smoking and alcohol consumption; Model 4: covariates in Model 3 plus cholesterol level and fasting glucose. Covariates were selected for inclusion in a step-wise manner, starting with subject-specific factors known to strongly affect SBP, then an allowance for any possible cohort effect, followed by further environmental covariates. Finally, we included possible intermediate phenotypes that contribute to the variation in SBP, including such covariates may reduce the power to detect linkage for genes contributing to these intermediate phenotypes.
The models for repeated censored data yielded consistent covariate effects across replicates for cigarettes per day and the absence of effect for alcohol, moderate consistency for cholesterol, and weak consistency for glucose.
Heritability and linkage analysis
Heritability estimates were obtained using variance components analysis as implemented in the SOLAR package [5]. Multipoint quantitative linkage analyses were conducted on the cross-sectional SBP phenotype and the standardized residuals for the longitudinal systolic SBP values.
GENEHUNTER version 2 [6] was used to carry out a nonparametric linkage (NPL) analysis of the whole genome, treating hypertension as a qualitative trait, as described above.
LOD scores > 1 were considered as nominal evidence of linkage, LOD scores > 2.2 as suggestive evidence, LOD scores > 3.6 as genome-wide evidence and LOD scores > 5.4 as confirmed linkage.
Pedigreed disequilibrium test (PDT) analysis
Microsatellite markers under areas of increased allele sharing (p < 0.1) were tested for evidence of association with the hypertension phenotype. The PDT method [7] was used, as this is a test for linkage and association in general pedigrees and can currently only be applied to qualitative data. In total 99 markers were analyzed.