Integrative genomic analyses in adipocytes implicate DNA methylation in human obesity and diabetes

DNA methylation variations are prevalent in human obesity but evidence of a causative role in disease pathogenesis is limited. Here, we combine epigenome-wide association and integrative genomics to investigate the impact of adipocyte DNA methylation variations in human obesity. We discover extensive DNA methylation changes that are robustly associated with obesity (N = 190 samples, 691 loci in subcutaneous and 173 loci in visceral adipocytes, P < 1 × 10-7). We connect obesity-associated methylation variations to transcriptomic changes at >500 target genes, and identify putative methylation-transcription factor interactions. Through Mendelian Randomisation, we infer causal effects of methylation on obesity and obesity-induced metabolic disturbances at 59 independent loci. Targeted methylation sequencing, CRISPR-activation and gene silencing in adipocytes, further identifies regional methylation variations, underlying regulatory elements and novel cellular metabolic effects. Our results indicate DNA methylation is an important determinant of human obesity and its metabolic complications, and reveal mechanisms through which altered methylation may impact adipocyte functions.


Supplementary Figure 2B: Effect of genetic factors on methylation-phenotype associations in visceral adipocytes.
Comparisons of associations between visceral adipocyte DNA methylation and extreme obesity at each sentinel site using linear regression analysis: i. the base model (X-axis: methylation difference model); and ii. the base adjusted for the effects of genetic variants (Y-axis: methylation difference model + SNPs, combined discovery and replication cohorts). As in subcutaneous adipocytes, genetic factors did not systematically affect DNA methylation-obesity associations, indicating underlying environmental origins. Top panels: adjusted for the first 5 principal components from PCA analysis of >8M SNPs carried out in all study participants. Middle panels: adjusted for the genotype dose of the top cis-SNP (±500-kb) associated with change in methylation at each sentinel site (Top mQTL, additive model). Bottom panels: adjusted for the genotype dose of all cis-SNPs (±500-kb) associated with change in sentinel methylation at FDR<0.01 (additive model). Left panels: association effect sizes (betas). Right panels: -log10 P values for association. Blues lines: Null of no systematic difference. Dark red lines: Significance threshold (P<0.05, Bonferroni adjusted for the number of sentinels).

Supplementary Figure 3A: Evaluation of potential adipocyte impurity by contaminating gene expression in subcutaneous adipocytes.
Gene expression results from RNA sequencing were used to evaluate the effects of potential contaminating cells (impurity) on DNA methylation-obesity associations by comparing base models without and with adjustment for gene expression levels (subcutaneous, replication cohort). Effects of 14 individual genes were evaluated (12 potential contaminating and 2 control genes), as were principal components (PCs) from PCA analyses of all 12 potential contaminating cell genes. Analysed by linear regression. No systematic effects on DNA methylation-obesity associations at the 691 subcutaneous sentinel sites were observed. Red lines: Null of no systematic difference. Identifiers: CD14 (monocyte/macrophage); ITGAM (CD11b, broad immune cell); CD68 (monocyte/macrophage); PTPRC (CD45, broad immune cell); CD19 (B lymphocyte); IL2RA (T lymphocyte); TNFRSF8 (CD39, broad lymphocyte); CD4 (T lymphocyte subtype); CD8A (T lymphocyte subtype); CD2 (T lymphocyte and NK cell); NCAM1 (NK cell); CD34 (endothelial and precursor cell); PPARG (adipocyte); FABP4 (adipocyte); PC1-PC5 (principal components from PCA analysis of the 12 immune and stromovascular genes). Analyses were carried out with variance stabilizing gene expression counts. Figure 3B: Evaluation of potential adipocyte impurity by contaminating gene expression in visceral adipocytes.

Supplementary
Effect of contaminating cell gene expression results from RNA sequencing (impurity) on DNA methylation-obesity associations (visceral replication cohort, 12 potential contaminating and 2 control genes, and principal components (PCs) from PCA analyses of all 12 potential contaminating cell genes). Analysed by linear regression. No systematic effects were observed at the 173 visceral sentinel sites. Red lines: Null of no systematic difference. Identifiers: CD14 (monocyte/macrophage); ITGAM (CD11b, broad immune cell); CD68 (monocyte/macrophage); PTPRC (CD45, broad immune cell); CD19 (B lymphocyte); IL2RA (T lymphocyte); TNFRSF8 (CD39, broad lymphocyte); CD4 (T lymphocyte subtype); CD8A (T lymphocyte subtype); CD2 (T lymphocyte and NK cell); NCAM1 (NK cell); CD34 (endothelial and precursor cells); PPARG (adipocyte); FABP4 (adipocyte); PC1-PC5 (principal components from PCA analysis of the 12 immune and stromovascular genes). Analyses carried out with variance stabilizing gene expression counts. Surrogate variable analysis was used to evaluate the effects of unknown confounders, in particular potential contaminating cell impurity, on associations between DNA methylation and extreme obesity in A. subcutaneous and B. visceral adipocytes (combined discovery and replication cohorts). Inclusion of SV1 and SV2 in the base linear regression models did not systematically alter methylation-phenotype effect sizes, and the majority of sentinels remained significantly associated with obesity after adjustment for SV1 and SV2 in the base model. Left panels: association betas. Right panels: -log10 P values. Analysed by linear regression. Blues lines: Null of no systematic difference. Dark red lines: Significance threshold (P<0.05, Bonferroni adjusted for the number of sentinels).

Supplementary Figure 5: Genomic annotation of subcutaneous and visceral sentinels.
Localisation of DNA methylation sentinels in subcutaneous and visceral adipocytes to functional/active genomic regions. Subcutaneous sentinels (N=691) were enriched in enhancers from various datasets and in CpG sparse genomic regions (open sea) but were underrepresented in promoter CGIs. Visceral sentinels (N=173) showed generally similar trends but did not reach statistical significance in any feature. CGI: CpG island track from UCSC. Presented as -log10 enrichment or depletion Q value, and number of observed counts, for each feature (Fisher's exact test, two-sided). Roadmap adipocytes and WAT: human adipocyte (E025) and adipose tissue nuclei chromatin states from the Roadmap epigenomics consortium (E063). Reg2map adipocytes and WAT: human adipocyte and adipose tissue regulatory features called using Roadmap and Encode epigenomes (-log10 P value ≥10). Fantom 5: human enhancer tracks called from Fantom 5 CAGE enhancer-promoter co-expression. GeneHancer: multifaceted human enhancer-target gene inference database.  A. Subcutaneous and visceral adipocyte sentinel-target gene associations at FDR<0.01 grouped by annotation method and presented as proportion of associations in each Roadmap chromatin state. Genic: sentinel in a promoter, 5/3'UTR or exon. Functional: intronic/intergenic sentinel sharing a functional interaction with a distal target gene. TAD: intronic/intergenic sentinels and distal target genes sharing a topologically associated domain in human adipocytes. Null: methylation-target gene associations at FDR>0.01. E025: Human adipocytes. E063: human adipose tissue nuclei. All sentinel DNA methylation-target gene expression analyses were carried out in combined subcutaneous and visceral adipocyte samples.

B.
Visceral adipocyte sentinel-target gene associations at FDR<0.01 (analysed by fixed-effects linear regression), grouped by annotation method, presented as distance between the sentinel and its target gene TSS, and coloured by roadmap chromatin state (E025 adipocytes). Scatter plots of distance and -log 10 pvalue for association split according to direction of effect (methylation-expression). Density plots of each sentinel-target gene association relative to: i. distance (top); and ii. -log10 pvalue (right border). Fold change: expression fold change per unit change in methylation. C. Association between expression of 6 TFs predicted to bind to Motif 2, and expression of the predicted target genes of each sentinel methylation site corresponding to Motif 2 (combined subcutaneous and visceral adipocyte samples). X-axis: association betas without adjustment for sentinel methylation level. Y-axis: association betas with adjustment for sentinel methylation level. With regression line (blue) and 95% confidence intervals (dark grey). Adjustment for sentinel methylation level systematically impacted TF-target gene association relationships involving SNAI2, TCF3, SNAI1 and SMAD3, but not TCF12 and ZEB2. Studies of lipid accumulation and expression of key genes involved in adipocyte metabolism at day 6 of differentiation, in 3T3-L1 adipocytes transfected with siRNA against Prrc2a, Limd2 or non-silencing (NS) control at day 2 after initiation of differentiation.

Supplementary Figure 11: Adipocyte functional studies off-target evaluation.
Studies of lipid accumulation, gene expression and cell viability in 3T3-L1 adipocytes transfected with two distinct siRNA against Prrc2a (78, 79), Limd2 (72, 74) or non-silencing (NS) control. This approach was used to confirm the absence of siRNA mediated off-target effects, as the distinct siRNA were designed to share no sequence, and thus target different regions of the Prrc2a and Limd2 mRNA.

C.
Effects of Prrc2a and Limd2 siRNA silencing using distinct siRNA on cell viability measured using the RealTime-Glo™ luminescence assay (Relative Light Units, RLU) to quantify metabolic activity in live/healthy cells at 4-hr, 24-hr, 48-hr and 72-hr after siRNA treatment. N=6 per condition in undifferentiated 3T3-L1 adipocytes. All values are presented as mean ± SEM (Two-Way repeated measures ANOVA test, Dunnett's test multiple comparisons); source data are provided in the Source Data file.

Supplementary Figure 12: Targeted methylation sequencing sensitivity analyses.
Precision and concordance analyses of 8 samples sequenced in replicate. A. Mean and standard deviation of methylation measurements (%) at paired sites (N=~13K) grouped by depth of coverage. Targeted methylation sequencing even at >150x coverage lacked the precision of array-based and whole-methylome sequencing (when compared to our previous work 120 ). B. Concordance of methylation measurements (%) at paired sites. Rep1 and Rep2: Replicate 1 (pilot) and Replicate 2 (main). C. Differences in methylation measurements (%) between replicates relative to mean coverage at paired sites.
Comparisons of methylation measurements using Array (Illumina HumanMethylation450 and EPIC BeadChips combined) and targeted methylation sequencing (TMS) platforms at 839 sites present in both datasets. D-E. Methylation measurements (%) on different platforms correlated strongly but showed platform specific biases, similar to those observed in whole-methylome sequencing results 119 .
Associations with obesity in 89 subcutaneous and visceral adipocyte samples evaluated by targeted methylation sequencing (TMS). F. Effect sizes at sentinel and non-sentinel sites present in targeted sequencing and array datasets (combined discovery and replication cohorts, %-methylation difference in obesity) were highly concordant (Pearson correlation coefficient (R), two-sided). G. Sentinel methylation sites were enriched for association with obesity in targeted methylation sequencing data compared to: i. non-sentinel sites present on the array; and ii. non-sentinel sites not present on the array. Observed: observed association P values in TMS results (linear regression). Expected: expected association P values based on the null hypothesis of no associations.