Factors underlying variable DNA methylation in a human community cohort

Abstract

Epigenetics is emerging as an attractive mechanism to explain the persistent genomic embedding of early-life experiences. Tightly linked to chromatin, which packages DNA into chromosomes, epigenetic marks primarily serve to regulate the activity of genes. DNA methylation is the most accessible and characterized component of the many chromatin marks that constitute the epigenome, making it an ideal target for epigenetic studies in human populations. Here, using peripheral blood mononuclear cells collected from a community-based cohort stratified for early-life socioeconomic status, we measured DNA methylation in the promoter regions of more than 14,000 human genes. Using this approach, we broadly assessed and characterized epigenetic variation, identified some of the factors that sculpt the epigenome, and determined its functional relation to gene expression. We found that the leukocyte composition of peripheral blood covaried with patterns of DNA methylation at many sites, as did demographic factors, such as sex, age, and ethnicity. Furthermore, psychosocial factors, such as perceived stress, and cortisol output were associated with DNA methylation, as was early-life socioeconomic status. Interestingly, we determined that DNA methylation was strongly correlated to the ex vivo inflammatory response of peripheral blood mononuclear cells to stimulation with microbial products that engage Toll-like receptors. In contrast, our work found limited effects of DNA methylation marks on the expression of associated genes across individuals, suggesting a more complex relationship than anticipated.

Fig. 1.

Variable DNA methylation in PBMCs derived from a human cohort. (A) Distinct distribution of mean DNA methylation levels dependent on the context of CpG site. All CpG sites were classified into LC regions, IC regions, and HC regions. CpG sites were further divided based on mean DNA methylation levels into hypomethylated (less than 20% β-values, dark gray), heterogeneously methylated (β-value levels between 20% and 80%, gray), and hypermethylated (more than 80% β-values, light gray). (B) Mean DNA methylation levels are represented by M-values and divided according to CpG density categories. M-values are log transformations of methylated intensity over unmethylated intensity and resulted in a much more uniform variability. A M-value of 0 is equivalent to a β-value of 0.5, with negative M-values indicating less than and positive M-values indicating more than a β-value of 0.5. (C) Population distribution of the 99 CpG loci with a mean SD > 1, with CpG density category indicated by colors as in B. Each column depicts DNA methylation levels at one specific CpG site, with every individual in our study graphically represented by one dot. The most variable CpG site is shown in the leftmost column, with CpG sites arranged in columns of decreasing variability going from left to right.

Fig. 2.

Demographic and psychosocial factors were associated with DNA methylation. Graphical representations of P-value distributions. In each case, the dashed line represents the uniform distribution that was expected by chance. The skewed distributions with an enrichment of CpG sites having small P values suggested that sex (A), ethnicity (B), and early-life SES (C) were correlated with DNA methylation. (D) This contrasted with the lack of correlation suggested by the uniform P-value distribution of current SES. Furthermore, cortisol output (E) and perceived stress (F) were both correlated with DNA methylation. Testing for correlations was done using either Wilcoxon tests (A–D) or Spearman ρ statistics (E and F).

Fig. 3.

DNA methylation was predictive of PBMC ex vivo response. As judged by the skewed P-value distributions, IL-6 production in PBMCs was associated with DNA methylation on ex vivo stimulation for 6 h by ODN (A) and LPS (B). Dashed lines represent the uniform distribution expected by chance. Testing for correlations was done using Spearman ρ statistics.

Fig. 4.

PCA revealed covariant DNA methylation patterns in a population. (A) Percentage of variation in DNA methylation accounted for by the top 20 eigen-probes. (B–D) DNA methylation signatures for the top 3 eigen-probes over the population of individuals. Using a methylation variation (M-value) cutoff of ±0.1 allowed us to create groups of correlated individuals for each eigen-probe that we could then test for enrichment for particular traits (main text).

Fig. 5.

DNA methylation and gene expression were not tightly linked across individuals. (A) Histogram of correlations between DNA methylation and expression of the associated genes within each individual across all 16,419 CpG sites demonstrated the canonical negative correlation between DNA methylation and expression. Average correlation is shown in red. (B) Q-Q plot shows the association of DNA methylation and mRNA expression of associated genes across individuals. Although the majority of significant correlations were negative, a substantial fraction was unexpectedly positive. (C) Representative example of two CpG loci in the promoter of the DDX43 gene that had a strong negative correlation with expression across individuals. These sites were also differentially methylated between males and females. Lines of least square and Spearman correlation between DNA methylation at each site and mRNA expression of DDX43 gene are shown on the graph. (D) Only a minority of variable CpG sites had a significant correlation with mRNA expression across individuals. The extent of variation in DNA methylation is shown as SD on the y axis, whereas the correlation between DNA methylation and expression of the associated gene is shown on the x axis. Each circle indicates one CpG site. Testing for correlations was done using Spearman ρ statistics, and the red circles in B and D indicate CpG sites that survive FDR correction at a q value <5%, whereas the black circles indicate CpG sites with q values between 5% and 25%.