Study Population. ELITE was a single-center, randomized, double-blind, placebo-controlled clinical trial (ClinicalTrials.gov number NCT00114517) testing effects of HT on progression of subclinical atherosclerosis as a function of time-since-menopause. Participants were healthy postmenopausal women without diabetes and clinical evidence of CVD who had no regular menses for at least 6 months or who had surgically induced menopause, as well as a serum estradiol level lower than 25pg/mL (92pmol/L). Women in whom time since menopause could not be determined, or who had fasting plasma TG levels > 500 mg/dL, diabetes mellitus or fasting serum glucose levels > 140 mg/dL, serum creatinine level > 2.0 mg/dL, uncontrolled hypertension, untreated thyroid disease, life-threatening disease with prognosis < 5 years, a history of deep vein thrombosis, pulmonary embolism, breast cancer, or current use of postmenopausal HT within 1 month of screening were excluded. A total of 643 women were stratified according to time-since-menopause (< 6 years [early] or ≥ 10 years [late]) and randomized to receive either HT or placebo using a 1:1 ratio of stratified blocked randomization, resulting in four treatment groups: early/placebo, early/HT, late/placebo, and late/HT. HT consisted of oral micronized 17b-estradiol 1mg/day with 4% vaginal micronized progesterone gel 45mg/day for 10 days each month (among women with intact uterus). ELITE demonstrated that, compared with placebo, HT reduced CIMT progression in women who were within six years of menopause but not women who were 10 or more years from menopause (12). Additional details on the design, methods, and results of the trial have been described previously (12, 31). ELITE was approved by the University of Southern California institutional review board and all participants provided written informed consent.
Whole-genome DNA Methylation Profiling. To maximize the likelihood of identifying associations between CpG sites and HT as a function of time-since-menopause, we selected subset of 48 women from the early/HT with the lowest 36-month rate of CIMT progression and an equivalent number of women from each of the early/placebo, late/placebo, and late/HT groups with the highest CIMT progression for methylation profiling. Genomic DNA was extracted from buffy coats of this subset of 192 ELITE participants obtained at baseline and 36 months following randomization using DNeasy kits (Qiagen, Valencia, CA) and bisulfite treated with the Zymo EZ DNA Methylation Kit (Zymo Research, Orange, CA). Quantitative levels of DNA methylation were obtained for > 850,000 CpG sites using the Infinium Human Methylation EPIC BeadChip (Illumina, San Diego, CA) according to the manufacturer’s protocols.
Methylation Data Processing and Normalization: Prior to analysis, the meffil package in R (32) was used to carry out several quality control (QC) steps, including filtering of samples and CpG sites, identifying batch effects, and normalizing sample quantiles. Meffil is a comprehensive and integrated toolkit that utilizes multiple previously developed R packages for methylation analysis, such as minfi (33), illuminaio (34), and noob (35). Background and dye-bias correction was first performed using raw probe signal intensities as the input. The noob background normalization method (35) was used to account for technical variation in background fluorescence signal, which capitalizes on a new use for the Infinium I design bead types to measure nonspecific fluorescence in the color channel opposite of their design (Cy3/Cy5). Poor quality CpGs were removed using the illuminaio R package (34) and the ChAMP R package (36–38) was used to identify and exclude SNP-related CpG probes based on previously reported annotations (39). Low quality samples were removed if they were outliers for methylated/unmethylated levels or control probe means, had too many undetected probes or low bead number probes (32). Functional normalization (FN) as implemented with the minfi R package (33) was used to minimize technical variation based on control probes present in the EPIC BeadChip that do not exhibit biological variation and whose only source of variation is due to technical artifacts. FN was also used to identify the number of principal components (PCs) of methylation matrix to include in the normalization that minimizes the residual variance unexplained by the given number of PCs, and to remove technical artifacts by normalizing sample quantiles using additional fixed and random effects (32). Quantile normalization was performed using meffil where slide, plate, and array were treated as random effects, and first 10 PCs were included as fixed effects. The Houseman algorithm (40) as implemented in meffil was used to estimate fractions of six different white blood cell populations (B cells, CD4 T cells, CD8 T cells, granulocytes, monocytes, and natural killer cells) using GSE35069 as the cell type reference (41). Leukocyte fraction estimates were subsequently used as covariates in the EWAS analyses. Methylation levels (β values) at each CpG site were determined by calculating the ratio of fluorescence intensities between methylated (signal A) and un-methylated (signal B) sites using the formula β = Max(M,0)/[Max(M,0) + Max(U,0) + 100]. Thus, β values range from 0 (completely un-methylated) to 1 (completely methylated). Prior to final analysis, β values were transformed to M-values (log2 ratio of methylated vs. unmethylated probe) using ‘beta2m’ function in the lumi package in R (42). The final dataset included 186 ELITE participants in whom methylation data at 748,567 CpG sites were available at both visits (total of 372 methylation profiles at baseline and 36 months after trial randomization).
Differential Methylation Analysis.
EWAS analyses with smoking were carried out at the baseline and 36-month visits out using linear regression models that were fitted using limma (43) as implemented in meffil (32). Participants were categorized as never, former, and current smokers (coded as 0, 1, 2) and methylation M-values were compared across categories using an analysis pipeline in meffil with adjustment for age, ethnicity, and estimated blood cell fractions. Longitudinal EWAS analyses with epigenetic data at both the baseline and 36-month visits were used to investigate the effect of HT (treated vs. placebo) and time-since-menopause (< 6 years [early] or ≥ 10 years [late]) on methylation levels. Changes in methylation at CpG sites were tested using lmrse package in R (44) designed to fit linear models with cluster robust standard errors across high-dimensional data to evaluate methylation trajectories. Participants for repeated measures analysis were categorized into four groups based on early or late post-menopause and randomized to HT or placebo. P-values for methylation changes at each CpG site were obtained from tests of interaction between these latter four categories and a time variable (baseline vs. 36 months) with adjustment for age, ethnicity, and estimated blood cell fractions.
Measurement of Subclinical Atherosclerosis. Rate of change in far wall intima–media thickness of the right distal common carotid artery was assessed by computer image processing of B-mode ultrasonograms. At baseline, two examinations were conducted (averaged to obtain baseline CIMT values) and every 6 months during trial follow-up (12). High-resolution B-mode ultrasonographic imaging and CIMT measurements were performed with use of standardized procedures and in-house technology that was specifically developed for longitudinal measurements of changes in atherosclerosis (31). Coefficient of variation for baseline CIMT measurements was 0.69% (12). The relationship between changes in CIMT progression and methylation levels (M-values) at CpGs identified through interactions with HT and time-since-menopause was assessed using partial Spearman’s correlation with adjustment for age and estimated blood cell fractions at baseline and at 36 months after treatment.