Independent genomewide screens identify the tumor suppressor VTRNA2-1 as a human epiallele responsive to periconceptional environment

Abstract

Background: Interindividual epigenetic variation that occurs systemically must be established prior to gastrulation in the very early embryo and, because it is systemic, can be assessed in easily biopsiable tissues. We employ two independent genome-wide approaches to search for such variants.

Results: First, we screen for metastable epialleles by performing genomewide bisulfite sequencing in peripheral blood lymphocyte (PBL) and hair follicle DNA from two Caucasian adults. Second, we conduct a genomewide screen for genomic regions at which PBL DNA methylation is affected by season of conception in rural Gambia. Remarkably, both approaches identify the genomically imprinted VTRNA2-1 as a top environmentally responsive epiallele. We demonstrate systemic and stochastic interindividual variation in DNA methylation at the VTRNA2-1 differentially methylated region in healthy Caucasian and Asian adults and show, in rural Gambians, that periconceptional environment affects offspring VTRNA2-1 epigenotype, which is stable over at least 10 years. This unbiased screen also identifies over 100 additional candidate metastable epialleles, and shows that these are associated with cis genomic features including transposable elements.

Conclusions: The non-coding VTRNA2-1 transcript (also called nc886) is a putative tumor suppressor and modulator of innate immunity. Thus, these data indicating environmentally induced loss of imprinting at VTRNA2-1 constitute a plausible causal pathway linking early embryonic environment, epigenetic alteration, and human disease. More broadly, the list of candidate metastable epialleles provides a resource for future studies of epigenetic variation and human disease.

Figures

Figure 1

Genomewide screen for human MEs. (a) DNA methylation in PBL is highly correlated across the two individuals included in the screen, C01 and C02. The density plot summarizes all 4.1 million 200 bp bins that were covered by sufficient read depth in both samples (R2 = 0.926). (b) Interindividual DNA methylation residuals (C01-C02) in HF versus those in PBL; 3.9 million 200 bp bins were informative in all four samples. The hyperbola delineates regions containing potential MEs. (c) Genomewide, most bins showed no evidence of genetic discordance between the two individuals. Regions of systemic interindividual variation (SIVI ≥20), however, were enriched for interindividual genetic variation. (d) HF versus PBL interindividual residual plot for the 4,852 filtered ME bins (SIVI ≥20, no genetic variation, no segmental duplication). The SIVI algorithm effectively targeted the regions indicated in panel (b). (e) Targeted analysis of Blueprint Epigenome data (DNA methylation in monocytes of six healthy individuals); ME bins with six or more CpG sites exhibit greatest interindividual variation. (f) Interindividual discordance of DNA methylation (C02 versus C01) of the 109 ME bins containing 6 or more CpG sites. (g) Manhattan plot of SIVI for all 200 bp bins with 6 or more CpG sites. Bins with SIVI ≥20 (candidate MEs) are crowned; gene-associated bins with SIVI ≥25 are labeled.

Figure 2

Distribution of CGIs and repetitive elements in ME versus non-ME genomic regions. In each pair of plots, 20 kb regions centered on ME bins (SIVI ≥ 20, n = 109, right) are compared with 20 kb regions centered on comparable non-ME bins genomewide (SIVI = -5 to 5, n = 298,979, left). For each 500 bp window, the normalized overlap score is the number of elements that overlap such windows, divided by the total number of bins. (a) ME regions are slightly depleted of CGIs (P = 2.5 × 10-6). (b) ME regions are depleted of SINE elements (P = 2.5 × 10-28). (c) ME regions are enriched for LINE elements (P = 7.0 × 10-8). (d) ME regions are enriched for ERVs (P = 3.5 × 10-15). All P-values based on chi-squared test.

Figure 3

Interindividual epigenetic variation at VTRNA2-1.(a) UCSC browser shot of the VTRNA2-1 region on chromosome 5. A cluster of five bins with high positive SIVI (top track) overlaps VTRNA2-1. Blueprint Epigenome DNA methylation data on monocytes from healthy individuals (orange) confirm interindividual variation in this same region. (b) Bisulfite pyrosequencing results for two individuals with discordant VTRNA2-1 methylation. T/C polymorphisms resulting from bisulfite conversion at three CpG sites are highlighted in gray. (c) Inter-tissue correlations of VTRNA2-1 methylation across kidney, liver, and brain of 17 Asian cadavers confirm systemic nature of interindividual variation. (d) Clonal bisulfite sequencing data on PBL DNA of two Gambian individuals (both A/A at SNP rs9327740) confirm pyrosequencing data and suggest interindividual variation in VTRNA2-1 methylation is not driven by local genetic variation. Columns and rows correspond to CpG sites and individual clones, respectively. Filled circles indicate methylation; gray circles indicate missing data.

Figure 4

Season of conception (SoC) and maternal periconceptional nutritional status predict methylation at VTRNA2-1. (a) Bisulfite pyrosequencing data on 215 Gambian children according to SoC. The rank plot (left) highlights the markedly different distribution according to SoC. The histogram (right) shows that individuals conceived in the dry season are under-represented for intermediate methylation expected at an imprinted locus (40 to 60%, highlighted) and over-represented for hypomethylation (P = 0.004). (b) In 80 Gambian infants with pyrosequencing data on both HF and PBL (left), VTRNA2-1 methylation in HF is highly correlated with that in PBL. Rank plot of average VTRNA2-1 methylation in HF of Gambian infants (right) shows that the SoC effect in HF is similar to that in PBL. (c) 450k array data on 120 Gambian children, according to SoC. Shown are 15 CpGs mapping to the VTRNA2-1 locus. The box highlights 10 CpGs corresponding to the imprinted DMR. The SoC effect on hypomethylation spans the entire imprinted DMR (P = 0.02, chi-squared test). (d) Rank plot of 450k array data at VTRNA2-1. Each box represents the methylation values across the 10 CpG sites spanning the imprinted DMR for one individual. (e) Seasonal variation in 13 methyl donor-related biomarkers and associated derivatives, back-extrapolated to time of conception and adjusted for gestation age (n = 164 pregnant mothers) [10]. Biomarkers are expressed as percentage of bi-season geometric mean. ANOVA P-values of seasonal differences: *<0.05; **<0.01, ***<0.001. (f) Maternal nutritional status biomarkers around the time of conception predict VTRNA2-1 hypomethylation (<40%) in her infant. Low maternal vitamin B2 or methionine (MET) status increases risk of VTRNA2-1 hypomethylation (P = 0.05 and P = 0.01, respectively). Low maternal dimethylglycine (DMG) is protective (P = 0.05). (g) Repeat measurements by bisulfite pyrosequencing in 55 Gambians indicate that VTRNA2-1 methylation in PBL is highly stable over a period of 10 years.