Simultaneous Assessment of Genetic and Epigenetic Contributions to the Plasma Lipid Levels with Respect to Cardiovascular Risk

Figures

• 

  Fig.1. Overview of the study design. An integrative genomic analysis involves GWAS array genotyping, target gene resequencing of five FH-related genes, and DNA methylation assays at selected CpG sites. The NCGM Biobank and BIO-CVD samples were reorganized into three subgroups and tested for DNA methylation (DNAm) changes. Individuals in the non-targeted populations (Supplementary Table 1) were used as reference genetic datasets to determine the Japanese Standard Polygenic Risk Score (PRS). Tohoku Medical Megabank (TMM) participants were used as in silico genetic datasets for the evaluation of rare variants of LDL-C levels. A general population subgroup (part of the KING study cohort) and a non-CAD subgroup were compared with the high LDL-C and CAD subgroups to develop a predictive model for dyslipidemia. In addition, some BIO-CVD subjects (n = 34) were subjected to the EWAS array assay to analyze changes in DNAm and the correlation between the DNAm sites due to aging and statin administration.

• 

  Supplementary Fig.1. Reduction rate in LDL-C according to medication. The treatment-induced reduction rate in LDL-C was calculated in Japanese by a retrospective analysis of data from 36 individuals with high LDL-C in whom plasma LDL-C measurements were available both before and within 18 months after initiating statins and ezetimibe. The arrows indicate the mean of the reduction rate in each treatment group. Thus, the average reduction rate of LDL-C achieved by statins in the analyzed Japanese sample was approximately 0.4, higher than the 0.3 previously reported in Western populations 7-9) .

• 

  Supplementary Fig.2. Estimating standardized impacts of a PRSLDL-C on plasma LDL-C level in Japanese. First, to define the standardized PRSLDL-C decile classes in Japanese (referred to as the ‘Japanese standard PRS for LDL-C’), we utilized PRSLDL-C data from three population-based cohorts (Amagasaki, Ehime, and Kita-Nagoya [KING]), whose areas are somewhat geographically distant from each other in Japan. Next, to convert the PRSLDL-C of high LDL-C and CAD subgroups into LDL-C predictions, we used the mean of the LDL-C measurements in each standardized PRSLDL-C decile class of the NCGM hospital cohort (i.e., an external reference group, N = 351) belonging to the same population as the above two subgroups recruited at NCGM. A: Histograms of PRSLDL-C in three cohorts. B: Box-and-whisker plots of LDL-C by PRSLDL-C decile in three cohorts. C: Line graphs of PRSLDL-C in Japanese at each decile point for the combined three-cohort data. Box-and-whisker plots of LDL-C measurements from the NCGM hospital cohort (D) and the high LDL-C and CAD subgroups (E) according to the Japanese standard PRS for LDL-C.

• 

  Supplementary Fig.3. Distribution of LDL-C and PRSLDL in two hospital-based cohorts at NCGM. Histograms of plasma LDL-C level (A) and PRSLDL-C (B) in the BIO-CVD (top panels) and NCGM hospital cohort (bottom panels) projects. BIO-CVD is a bioresource collection project focused on cardiovascular disease, including 306 CAD and 47 non-CAD, which were used for target gene resequencing (see Fig. 1). On the other hand, the NCGM hospital cohort (N = 351) was separately recruited without screening for cardiovascular disease. In the histograms of Fig. S3A, CAD (and non-CAD) subjects are colored in blue (and grey), and the arrows indicate the mean of LDL-C in individual project samples. Besides, non-CAD (N = 117) subjects from BIO-CVD were added to GWAS array genotyping (see Fig. 1).

• 

  Fig.2. Impacts of rare FH-related gene variants on LDL-C level and their prevalence. An LDL-C increase compared to the reference group (BIO-CVD subjects without notable rare variants, N = 305) was estimated for three categories, disruptive, damaging, and non-synonymous (see Methods about the classification), of five FH-related gene variants (A), based on LDL-C level according to rare variants that were identified by target gene resequencing (B). C: The percentage of ‘disruptive + damaging’ variant carriers is shown for populations with different morbidity statuses: high LDL-C (left) and CAD (middle) subgroups and a general population (right; TMM cohort) of Japanese descent. The ratio of carriers for disruptive and damaging variants was significantly (P＜0.05) higher among CAD subjects than in the general population (TMM).

• 

  Fig.3. Development of a conversion formula from PRSLDL-C to LDL-C value for PRSLDL-C-based risk prediction. First, to define the standardized PRSLDL-C decile classes, data for three population-based cohorts (Kita-Nagoya, Amagasaki, and Ehime) were combined: distribution of PRSLDL-C (A) and actual LDL-C for each PRSLDL-C decile class, thus standardized (B). Then, data for another hospital-based study cohort (NCGM hospital cohort) were used to convert PRSLDL-C to LDL-C values for the individuals newly recruited at NCGM for rare variant search in addition to PRSLDL-C and MRSLDL-C calculations (see also Supplementary Fig.2). C: PRSLDL-C distribution is stratified based on the standardized PRSLDL-C decile classes for the high LDL-C (top) and CAD (bottom) subgroups at the NCGM. Correlations are shown between actual LDL-C levels and those predicted by the standardized PRSLDL-C decile class alone (D), composite genetic risk (E), and composite genetic risk + MRSLDL-C (F) in newly recruited individuals at the NCGM (BIO-CVD + NCGM Biobank) with a complete dataset.

• 

  Supplementary Fig.4. DNA methylation changes in lipid-associated CpGs with aging and statin administration. A: Characteristics of 34 BIO-CVD subjects used for an EWAS array assay to examine temporal changes in DNA methylation of lipid-associated CpGs. B: Blood lipid changes in each subgroup. C: Temporal DNA methylation changes with statin treatment at SC4MOL cg05119988, previously reported to be significantly associated with statin use in EWAS 31). Of 93 selected lipid-associated CpGs, temporal DNA methylation changes are nominally significant (P＜0.05) in 38% (35 of 93 CpGs) with aging (D) and 11% (10 of 93 CpGs) with statin treatment (E). Known CpG sites with age-related changes in methylation rates (e.g., SLC1A5 cg22304262) are included in the former comparative analysis (D).

• 

  Fig.4. CpG-trait association and prediction of lipid traits by risk scores. A: The effect sizes for CpG-trait association are shown for triglycerides (left; at 11 CpGs), HDL-C (middle; at 7 CpGs), and LDL-C (right; at 6 CpGs), where comparisons were made between this study and previous multi-ethnic EWAS 10) (top panels) and between high LDL-C and CAD subgroups (bottom panels). B: The impacts of two risk scores, PRS and MRS, on lipid traits are estimated by R2 in the high LDL-C (left) and CAD (middle) subgroups and the general population (right; KING study cohort) of Japanese descent. ＊P＜0.05, ＊＊P＜0.001, ＊＊＊P＜10-4.

• 

  Supplementary Fig.5. Pairwise correlation of lipid-associated CpGs with respect to DNA methylation levels. A: The schematic diagrams indicate that CpG pairwise correlations for DNA methylation (DNAm) levels measured at a single point on each CpG (left) and the difference in DNA methylation levels between two points (ΔDNAm) (right) are examined for each lipid trait. B-E: The strength of the correlation is shown as the coefficient of determination (R²) on the lower left of the diagonal and the P-value in linear regression on the upper right of the diagonal. The types of lipid trait-associated CpGs and DNAm examined in each panel are as follows: B: HDL-C-associated CpGs and DNAm measured at a single point; C: HDL-C-associated CpGs and ∆DNAm; D: triglycerides-associated CpGs and ΔDNAm; E: LDL-C-associated CpGs and ΔDNAm.

• 

  Supplementary Fig.6. Distribution of metabolic traits in three subgroups newly recruited. Histograms of lipid traits and body mass index (BMI) are shown for three subgroups—high LDL-C (≥ 160 mg/dl without CAD), CAD (regardless of LDL-C level), and non-CAD individuals—newly recruited via two bioresource projects (BIO-CVD and NCGM Biobank). LDL-C was imputed for subjects taking lipid-lowering medications (statins or ezetimibe), and the imputed values (second column from left) were utilized in multivariate regression analyses for each subgroup (Supplementary Fig. 9 and Supplementary Table 5C). The arrows indicate the mean of each metabolic trait in individual study groups.

• 

  Supplementary Fig.7. Correlation of DNA methylation measurement between ddPCR and EPIC array. Scatter plots show a fair correlation between two analytical methods—ddPCR (in the x-axis) and EPIC array (in the y-axis)—at 17 CpGs examined in this study.

• 

  Supplementary Fig.8. Distribution of MRSs for lipid traits calculated with 17 CpGs in high LDL-C, CAD, and general subject groups. Histograms of MRSs for lipid traits are shown for two subgroups—high LDL-C (≥ 160 mg/dl) and CAD individuals—at NCGM and a general subject group (part of the KING study cohort). The arrows indicate the mean of lipid MRSs in individual study groups. We made an inter-group comparison of MRSs between the high LDL-C and CAD subgroups.