Mouse metastable epialleles are extremely rare

Abstract

Metastable epialleles (MEs) are genomic loci at which epigenetic marks are established stochastically during early embryonic development and maintained during subsequent differentiation and throughout life, leading to stable epigenetic and phenotypic variation among genetically identical individuals. Although MEs were first described in mice over 20 years ago, the extent of epigenetic metastability in the mouse genome remains unknown. We present the first unbiased genome-wide screen for MEs in mice. Using deep whole-genome bisulfite sequencing across tissues derived from the three embryonic germ layers in isogenic C57BL/6J mice, we identified only 29 MEs, precisely localizing them and documenting their rarity. Consistent with recent findings, we found no effects of maternal dietary methyl donor supplementation on ME methylation in the offspring, challenging previous assertions that MEs generally exhibit developmental plasticity. Most but not all MEs are associated with intracisternal A-particle (IAP) elements, tending to localize to the 5′ end of the IAP. Additionally, we discovered autosomal regions at which systemic interindividual variation in DNA methylation is associated with sex, providing insights into sex-associated epigenetic development that apparently precedes sexual differentiation. Our findings indicate that expression of transcription factors, including CCCTC-binding factor (CTCF) and specific KRAB zinc finger proteins during early embryonic development, plays a key role in orchestrating stochastic establishment and/or maintenance of DNA methylation at metastable transposable elements. Overall, these findings advance our understanding of the genomic determinants of epigenetic metastability and suggest that interindividual epigenetic variation at MEs is unlikely to be a major determinant of phenotypic variation among isogenic mice.

Graphical Abstract

Graphical Abstract.

Introduction

Over two decades ago, observations of epigenetic and corresponding phenotypic variation among isogenic mice led Whitelaw and colleagues to propose the term ‘metastable epiallele’ (ME) to describe alleles at which the epigenetic state is established probabilistically during development and, once established, mitotically inherited during subsequent differentiation [1]. The two most extensively characterized murine MEs are agouti viable yellow (A^vy) and axin fused (Axin^Fu). Epigenetic metastability at A^vy resulted from spontaneous retrotransposition of an intracisternal A particle (IAP) element ∼100kb upstream of the agouti promoter. Wide interindividual variation in A^vy DNA methylation and associated ectopic agouti expression leads to striking variation in coat color and hyperphagic obesity among isogenic A^vy/a mice [2–4]. (The non-agouti (a) allele produces no functional agouti protein, so the phenotype of each A^vy/a mouse is determined by the epigenetic state of the A^vy allele.) Likewise, at Axin^Fu, spontaneous insertion of an IAP element into an intron of Axin confers epigenetic metastability such that isogenic Axin^Fu heterozygous mice exhibit wide variation in Axin^Fu methylation associated with ectopic expression of a truncated Axin transcript, manifesting most obviously as a variably-expressive tail kink phenotype [5]. Similarly, at the Cabp^IAP ME, which was discovered by searching mouse cDNA databases for IAP long-terminal repeat (LTR) sequence [6], an IAP insertion induces a cryptic promoter that drives variable Cabp expression among isogenic C57BL/6J mice. It was shown that maternal dietary supplementation with methyl donors and cofactors before and during pregnancy permanently alters the coat color of A^vy offspring by increasing DNA methylation at the A^vylocus [3, 7], the first demonstration of developmental programming via an epigenetic mechanism. This finding was corroborated in Axin^Fu mice; the same dietary supplementation regimen increased Axin^Fumethylation in offspring, reducing the incidence and severity of tail kinking [8]. Maternal dietary supplementation with the naturally-occurring dietary phytoestrogen genistein before and during pregnancy, on the other hand, reduced Cabp^IAP ME methylation levels in the offspring [9].

These findings motivated interest in identifying MEs in the human genome. To advance this aim, researchers took advantage of the systemic (i.e. non-tissue-specific) nature of interindividual epigenetic variation at murine MEs. Initial studies identified systemic interindividual variation (SIV) in DNA methylation by genome-scale DNA methylation profiling in multiple tissues from multiple individuals. The first such screen for human SIV employed a reduced-representation approach to profile DNA methylation across two tissues (reflecting different embryonic germ layer lineages) in each of eight individuals [10], identifying ∼40 putative MEs. The authors used a natural experiment to test for effects of periconceptional nutrition at these loci, studying Gambian children conceived during different times of the year associated with divergent maternal nutritional status. At candidate human MEs, higher methylation was found in children conceived during the rainy season compared to those conceived in the dry season. Similar “two tissue” screens subsequently used the HM450 array [11] or whole-genome bisulfite sequencing (WGBS) [12] to identify additional candidate MEs, many corroborating the findings of the previous screen [10]. Silver et al. identified the gene for the imprinted non-coding RNA VTRNA2-1 as a human ME sensitive to periconceptional environment [12]. Subsequent studies [13, 14] confirmed a lack of genetic influence on epigenetic variation at VTRNA2-1 (also known as nc886), corroborating it as a human ME.

To date, the largest screen for human SIV involved deep WGBS in three tissues (representing all three embryonic germ layers) of each of 10 individuals, mapping out nearly 10 000 correlated regions of SIV (CoRSIVs) [15]. Regardless of potential genetic influences, the practical utility of SIV for population epigenetics motivates interest in this larger set of loci. Not surprisingly, studying methylation quantitative trait loci (mQTL) in a cohort of nearly 200 individuals demonstrated that most human CoRSIVs are influenced by genetic variation [16]. About 15% of human CoRSIVs, however, show minor mQTL effects, suggesting they may be MEs. Despite the genetic influences, establishment of DNA methylation at CoRSIVs is also generally sensitive to periconceptional environment, including maternal nutrition [15, 17], embryo culture during assisted reproduction [18] and environmental contaminants [19]. Together, these findings indicate that gene-environment interactions during critical periods of early human embryonic development determine the establishment of systemic and persistent DNA methylation states.

Despite this progress in humans, identification of murine MEs continues to be important. For example, although MEs were first described over two decades ago [1] we do not understand the DNA sequence features that confer epigenetic metastability. Unbiased identification of a broad panel of murine MEs could advance this goal. Identification of murine MEs could also help explain substantial phenotypic variation that has been observed among isogenic animals, across a broad range of anatomical, behavioral, and immunological outcomes [20], including interindividual variation in tissue-specific gene expression [21] and adiposity in response to an obesogenic diet [22] among male C57BL/6J mice. Our current understanding of SIV is that it results from establishment of DNA methylation states in the early embryo, prior to gastrulation. Broader identification of murine MEs would provide opportunities to test this directly by studying epigenetic ontogeny at these loci, as pioneered at the A^vy locus [23]. Discovery of additional murine MEs linked to clinically relevant phenotypes (such as the hyperphagic obesity of A^vy mice) could provide animal models for prevention and/or treatment of epigenetically based human disease. And, because murine MEs convey transgenerational epigenetic inheritance [24, 25], ME discovery in isogenic mice will facilitate the study of transgenerational epigenetic inheritance unconfounded by genetic inheritance.

Recent screens for murine MEs were biased towards IAPs, a subset of endogenous retroviruses (ERVs). Kazachenka et al. [26] analyzed WGBS data on B and T cells of C57BL6/J mice, identifying 104 “‘variably methylated IAPs” (VM-IAPs). However, they did not systematically assess the systemic nature of interindividual epigenetic variation at these loci. Elmer et al. refined the list to 51 “constitutive VM-IAPs” (cVM-IAPs), again without formal evaluation of SIV [27]. Bertozzi et al. [28] evaluated 11 cVM-IAPs for responsiveness to early environmental exposures, concluding they generally lack the developmental plasticity observed at A^vy and Axin^Fu [3, 7–9]. An expression microarray approach to screen for MEs [29] found no link between gene expression and DNA methylation. Oey et al. [30] used WGBS to identify 356 regions of interindividual variation in DNA methylation, many of which overlapped an ERV, but did not assess these for SIV. No unbiased genome-wide screens specifically designed to identify regions of SIV among isogenic mice have been reported.

To address this longstanding lacuna, here we performed the first screen for SIV in DNA methylation among isogenic mice. We performed deep WGBS to profile DNA methylation across 3 tissues representing all three embryonic germ layers in each of 10 C57BL/6J mice, identifying genomic regions of SIV. We conducted independent validation by quantifying DNA methylation in multiple tissues within a separate cohort of C57BL/6J mice. Our well-powered screen identified a small number of murine MEs and discovered several autosomal regions that exhibit substantial and systemic sex differences in DNA methylation. In addition to characterizing sequence features associated with epigenetic metastability, we assessed the impact of maternal dietary methyl donor supplementation on offspring methylation at a panel of the MEs we identified.

Materials and methods

Study population and sample collection

The ME screen utilized five male and five female C57BL/6J mice (Strain #000 664, The Jackson Laboratory); in addition to being the basis for the mouse reference genome [31], this strain was the focus of previous ME screens [6, 26, 30]. Mice were received at seven weeks of age and housed in our animal facility for one week prior to tissue collection. The screen was based on liver, kidney, and cerebral cortex of each mouse. These tissues were selected to represent the three embryonic germ layers—endoderm, mesoderm, and ectoderm, respectively. And, each is relatively homogeneous, simplifying standardization of tissue collection. Liver, kidney, and brain tissues were also used to document systemic interindividual epigenetic variation in our early studies of mouse MEs [3], our studies of human SIV [10, 12, 15], and in recent studies by other groups [26]. The mice were euthanized within a 1-h period in the morning, and tissues collected and stored until DNA isolation. Genomic DNA was isolated from flash-frozen tissue samples by mechanical disruption, enzymatic digestion, and organic extraction as previously reported [12, 32, 33]. This study was approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine, and animals were maintained in accordance with federal guidelines.

Dietary methyl donor supplementation study

Six-week-old male and female C57BL/6J mice were obtained from The Jackson Laboratory. They were maintained in our animal facility for two to three weeks prior to commencing the diet study. The diets used were an NIH-31 natural ingredient control diet (Envigo, TD.95262), and NIH-31 supplemented with folic acid, vitamin B12, choline chloride, and anhydrous betaine (Envigo, TD.220545); these were formulated precisely as in our previous studies [3, 8]. At age 8–9 weeks, females were randomly assigned to either the control or methyl supplemented diet. After 2 weeks on their respective diets, one male was placed in the cage with each female. As soon as each dam was visibly pregnant the male was removed from the cage and the cage was checked daily for a litter. When each litter was born (P0), the diet was switched to Picolab 5V5M natural ingredient mating/lactation diet. At weaning (P21) two male and two female pups from each litter, when available, were euthanized for tissue collection. Genomic DNA was isolated from liver, as described above, and quantitative bisulfite pyrosequencing was used to measure average % methylation at a panel of MEs. Because the supplementation was provided to the dams, the litter is the appropriate unit of analysis, and the individual pups within each litter are pseudo-replicates. Therefore, our statistical analysis comparing % methylation in supplemented vs. control offspring is based on litter averages.

Enrichment of transposable elements within and near SIV regions

Genomic enrichment of transposable elements at SIV loci was evaluated relative to matched control regions (see supplementary methods). Genome-wide data on repeat elements including their class, family and subfamily were downloaded from the UCSC genome browser RepeatMasker track for genome build GRCm38/mm10. TEs overlapping SIV or control regions, or located in their flanking regions 1 kb upstream and downstream, were enumerated using the bedtools software package [34]. The RepeatMasker annotation often breaks IAPs into multiple fragments which we counted individually, including instances where insertions were present. Paired t-tests were performed using the SciPy package [35] to assess the statistical significance of relative enrichment or depletion.

Regional enrichment of candidate cis-regulatory elements

The ENCODE candidate cis regulatory elements (cCREs) [36] track was downloaded from UCSC genome browser for genome build GRCm38/mm10. Within each cCRE category, we searched for associations between cCREs and SIVs. We tabulated directly overlapping cCREs on SIVs and matched control regions (see supplementary methods); heatmaps were used to illustrate the results. Proportions of MEs, mMEs, SASIVs and matched control regions with at least one regulatory element overlapping 10kb upstream and downstream flanking regions (by 1kb interval) were calculated and plotted using line graphs. Fisher's exact test (SciPy package [35]) was used to assess the statistical significance of enrichments or depletions.

Assessment of overlaps with KRAB zinc finger binding protein (KZFP) motifs

To ascertain binding motifs directly overlapping SIV regions, we used the top binding motifs for 45 murine KZFPs (as MEME output files) identified in a previous study by Wolf et al. [37]. We applied “Find Individual Motif Occurrences” (FIMO) [38] to search the DNA sequences of SIV and 87 matched control regions (three matched controls for each region, see supplementary methods) for the most statistically significant motif of each of the 45 KZFPs. Resulting significant hits (q ≤ 0.05) for SIVs and matched control regions were counted and plotted in heatmaps. Analogous heatmaps were generated using a set of random controls matched only for chromosome and lengths of the regions (i.e. not matched for CpG density). Subsequently, we set out to investigate whether there are KZFP binding motifs within the IAP elements that directly overlap SIV and matched control regions. We downloaded the UCSC genome browser RepeatMasker track and extracted IAPs that directly overlap SIVs and matched control regions using bedtools [34]. After applying FIMO, significant KZFP binding motif hits (q ≤ 0.05) on specific TE subfamilies (IAPEz-int and IAPLTR1) directly overlapping SIV and matched control regions were counted and plotted in heatmaps. Statistical significance of enrichments was calculated using Fisher's exact test (SciPy package [35]).

Results

Identification and validation of SIV regions in isogenic mice

To identify genomic regions of SIV among isogenic individuals, we performed WGBS on liver, kidney, and cerebral cortex from each of five male and five female C57BL/6J mice (Fig. 1A). Generating deep sequencing data on all 30 libraries, with an average of ∼1.35B uniquely mapped 150-bp paired-end reads per sample after QC filtering (Supplementary Fig. S1A), yielded an average genome-wide read depth of 40X. Analysis of read depth on the Y chromosome (Supplementary Fig. S1B) confirmed correct sex for all 30 libraries, indicating reliable sample handling. The coverage files obtained from Bismark methylation extraction are at single CpG resolution [39]. We analyzed CpG methylation at 100-bp resolution, focusing on the ∼12M 100-bp autosomal bins (GRCm38/mm10) containing at least one CpG site and yielding adequate read depth (see supplementary methods). Bin level methylation was obtained by averaging methylation across all CpG sites within each bin. As expected, given the cell-type specificity of DNA methylation, for a random subset of 10 000 such informative bins, bin-level average DNA methylation in the 30 libraries clustered by tissue (Fig. 1B), underscoring the reliability of the WGBS data. Our analysis followed the same two-step computational approach we recently developed to identify human CoRSIVs [15]. In the first step, all reads from the three tissues of each mouse were pooled, maximizing read depth for the calculation of individual-level average methylation for each autosomal 100 bp bin. Evaluating these across the 10 mice, regions of correlated methylation (bin-bin R ≥ 0.71 (R² ≥ 0.5)) were built. One such region within an intron of Snd1 is depicted in Fig. 1C. Sex chromosomes were omitted from the analysis due to the limited sample size. In the second step of the screen, for each region of correlated methylation, inter-tissue correlation (ITC) of average methylation was assessed across all three tissue-type pairs (i.e. brain versus liver, kidney versus liver, and kidney versus brain). Regions yielding a minimum ITC ≥ 0.71 were identified as correlated regions of SIV in DNA methylation. This approach identified 114 loci as potential SIV regions (Supplementary Table S1), including the Snd1 intronic region (Fig. 1D).

Figure 1.

Screen for SIV identifies a small number of MEs, but also sex-associated SIV regions (SASIVs). (A) In each of 10 mice, genomic DNA from liver, kidney, and cerebral cortex was profiled by deep WGBS, yielding 30 methylomes. These tissues represent the three embryonic germ layer lineages. (B) Unsupervised clustering of bin-level average methylation (random sample of 10 000 100-bp bins) groups the 30 samples by tissue. (C) Step 1 of the SIV screen identified blocks of correlated methylation. One such block at the Snd1 gene body is shown; the blue triangle indicates correlated methylation across 12 100-bp bins. The color of each small square indicates correlation of methylation across adjacent 100 bp bins. (D) Step 2 of the screen assessed inter-tissue correlations (ITCs) of block-level average methylation. Each point represents one mouse. Here, all three ITCs at the Snd1 gene body are ≥0.71 (R² ≥ 0.5) identifying it as an SIV region. (E) Interindividual methylation range vs. number of CpGs per block for all genomic blocks satisfying the SIV criteria. We focused on those including at least 5 CpGs and yielding an interindividual range ≥ 20% (dashed lines). (F) Consistent with their systemic nature, unsupervised clustering of average methylation at these 36 SIV regions groups the 30 libraries by individual. Also note separation by sex. The colors indicating individual mice are consistent across panels D and F.

Previous studies attempting to identify mouse MEs used various approaches to provide evidence of SIV [26, 27, 30]. Accurate clustering of tissue-level methylation data by individual is an unbiased indicator of SIV. We therefore utilized both our current mouse WGBS data and WGBS data on three tissues from each of 10 humans [15] to determine the ITC cutoff required to document SIV (see methods). Remarkably, a Rand index of 1.0 (indicating perfect clustering) [40], was achieved at nearly the same ITC cut-off, 0.71 and 0.70, in mice and humans respectively (Supplementary Fig. S2). We therefore propose that measuring DNA methylation in three tissues derived from different embryonic germ layers and applying a minimum ITC cutoff of 0.71 (as we have in several previous studies) [10, 12, 15] is a reliable standard to substantiate SIV. Most regions passing the ITC criterion showed only modest interindividual variation, and many included only a small number of CpG sites (Fig. 1E). To focus on the most biologically robust variants, and consistent with our screen for human CoRSIVs [15], regions that each include at least five CpGs and exhibit an inter-individual range ≥20% were considered in subsequent analyses. ITC scatter plots for the 36 potential SIV regions are provided in Supplementary File 1. In contrast to genome-wide bins (Fig. 1B), across these 36 potential SIV regions unsupervised clustering of average methylation grouped the 30 samples perfectly by individual (Fig. 1F, Supplementary Table S3), consistent with their systemic nature. Interestingly, they also clustered by sex. Using liver, kidney, and cerebral cortex genomic DNA from an independent cohort of 24 C57BL/6J mice, SIV was assessed by quantitative bisulfite pyrosequencing in eight regions which were selected to be representative of the full set of 36, with inter-individual methylation ranges from 22% to 67%. Seven of these (∼87%) were validated in bisulfite pyrosequencing (i.e. at least one tissue pair showed R² ≥ 0.5) (Supplementary Table S4). One potential SIV region (6_1184) at Vwde was not validated, leaving 35 for downstream analysis. Pyrosequencing assay details, including data on their quantitative validation, can be found in Supplementary Table S5.

A subset of mouse SIV regions associates with sex

At classic murine MEs A^vy and Axin^Fu, the distribution of DNA methylation shows both wide (nearly full-range) interindividual variation and multiple modes (i.e. distinct clusters of individuals with similar methylation levels) (Supplementary Fig. S3A & B); we therefore asked if any of the 35 SIV regions likewise show multi-modal distributions (Supplementary File 2). We identified 12 such regions (Supplementary Fig. S3C & D); at six of these, methylation was associated with sex (Supplementary Fig. S3C, Supplementary Fig. S4, Supplementary Table S6). We call these regions of sex-associated systemic interindividual variation (SASIVs); our WGBS data on three SASIVs are shown in Fig. 2A–C. The most likely cause of this variation is genetic differences in the sex chromosome complement between males and females. SASIVs, therefore, are not MEs. The six SIV regions with multimodal DNA methylation distributions not associated with sex, on the other hand, were classified as multimodal MEs (mMEs) (Supplementary Fig. S3D). These most resemble the classic murine MEs A^vy and Axin^Fu loci (Supplementary Fig. S3A & B). Indeed, the interindividual range of mMEs is greater than that of the remaining MEs (two-sided t-test, P = 2.9 × 10⁻⁷). To summarize, the repertoire of SIV regions we discovered includes 23 MEs, six mMEs and six SASIVs (Supplementary Fig. S5, Supplementary Table S2). At all six SASIVs, average methylation is higher in females than in males (Supplementary Fig. S6A–F). This agrees with previous reports of sex differences in autosomal methylation [41–43]. SASIVs display consistent sex differences in methylation across all three tissues (brain, kidney, and liver) derived from different embryonic germ layers (Supplementary Fig. S6A–F). To determine if these might result from reads originating from sex chromosomes aberrantly mapping to autosomal regions, we used BLAT [44] to test for significant homology to pseudoautosomal regions on either the X or Y chromosome, finding none.

Figure 2.

Some multimodal SIV regions reflect sex differences in DNA methylation. Each row shows WGBS data on one genomic region. (A  –C) CpG-level methylation in liver samples at (A) sex-associated SIV region (SASIV) 5_10 111_Gm22011, (B) SASIV 6_17 515_C1s2 and (C) SASIV 6_13 992_Gm44169. Each color represents a mouse (females—dotted lines; males solid lines). (D  –F) At SASIVs, most heterogeneity of methylation occurs among DNA molecules. Tanghulu plots portray read-level methylation patterns (in kidney) for the segments of SASIVs showing sex differences in methylation; data are shown for one male and one female mouse; each row represents a WGBS sequencing read, and each column a CpG site. Filled circles depict methylated CpGs. (G–I) Box plots representing methylation haplotype load (MHL) of brain, kidney, and liver at SASIVs. MHL was calculated over the same genomic coordinates as shown in D–F. MHL is higher in females than males in all three tissues (P= 3.2 × 10⁻¹⁷, 1.5 × 10⁻⁶, and 1.0 × 10⁻⁷ in panels G, H and I, respectively).

To better characterize the molecular nature of sex differences at SASIVs, we utilized our WGBS data to analyze their read-level methylation patterns. Examples of tanghulu plots are shown in Fig. 2D–F, and corresponding data on methylation haplotype load (MHL), a measure of the degree of co-methylation on single DNA molecules [45], are shown in Fig. 2G–I. These show that sex differences in methylation at SASIVs largely reflect variation among sequence reads rather than heterogeneity within the reads. Because each WGBS read originates from a single DNA molecule, SASIVs likely reflect genomic regions exhibiting cellular mosaicism in methylation, at which a larger proportion of cells are highly methylated in females. This characteristic was found to be consistent across all tissue types (Supplementary Fig. S7A–F). Interestingly, this variation among sequence reads appears to be a characteristic of multimodal SIV regions in general. Whereas mMEs show a preponderance of either unmethylated or fully methylated reads (Supplementary Fig. S8A & B), those mapped to unimodal MEs exhibit a broader range of intermediate methylation values (Supplementary Fig. S8C & D). This distinction is also apparent in the context of MHL [45] (Supplementary Fig. S8). These data suggest that different mechanisms may mediate epigenetic metastability at mMEs and unimodal MEs.

Mouse MEs overlap with those identified by other groups

It has been observed that there is “little to no overlap” among mouse genomic regions identified in previous ME screens [26]. To compare our list of MEs with those identified by other groups, we excluded SASIVs and considered only the set of 29 MEs. Of these, 13 overlap constitutive VM-IAPs (cVM-IAPs) identified by Elmer et al. [27] (Fig. 3A, Supplementary Table S7), many more than expected by chance (Fisher's exact test, P= 1.52 × 10⁻⁴⁷). Fig. 3B-D illustrates three such overlaps, at Eps8l1, Rnf157, and Rab6b respectively. Notably, whereas Elmer et al. annotated cVM-IAPs as extending across entire IAP elements in most cases, our results reveal that SIV most often occurs only at one end of each IAP. For example, at Eps8l1 (Fig. 3B), only the 5′ end of the IAP LTR, which overlaps the Eps8l1 transcription start site, shows interindividual variation. At Rnf157, on the other hand, although both ends of the IAP show interindividual variation, the variation is systemic (i.e. high ITC) only at its 5′ end (Fig. 3C). One exception is at the ME ∼3kb upstream of Rab6b (Fig. 3D), which encompasses the entire cVM-IAP identified by Elmer et al., and additional flanking regions. Of the 13 overlaps of our MEs and the cVM-IAPs identified by Elmer et al., the ME overlaps only the 5′ end of the cVM-IAP element in seven. Of the remaining six, four overlap the center region of a cVM-IAPs, but neither their 5′ nor 3′ LTRs. The remaining two (including the one upstream of Rab6b) encompass the complete cVM-IAP including both the 5′ and 3′ LTRs. Hence, none of the MEs overlaps only the 3′ end of a cVM-IAP. We next evaluated the overlap of our MEs with ERV-associated inter-individual differentially methylation regions (ERViiDMRs) reported by Oey et al. [30]. Of the 29 MEs we identified, 11 overlap with ERViiDMRs (Fig. 3A, Supplementary Table S8), again many more than expected by chance (Fisher's exact test, P= 9.72 × 10⁻⁴⁵). For eight of these, the ME fully encompasses the ERViiDMR. Of the remaining three MEs, two overlap only the 5′ end of the ERViiDMR, and one overlaps only the 3′ end of the ERViiDMR. Of all the MEs that directly overlap IAP elements, eight overlap only the 5′ end, and none overlap only the 3′ end (two-sided t-test, P = 0.03). Using our WGBS data, we evaluated interindividual range and ITCs at all cVM-IAPs and ERViiDMRs, both those that do and do not overlap with our MEs (Supplementary Fig. S9). The cVM-IAPs and ERViiDMRs that do not overlap our MEs generally exhibit substantial interindividual variation but do not meet our ITC criterion, meaning the interindividual variation is not systemic. Overall, the extensive overlap of MEs with cVM-IAPs and ERViiDMRs supports the reliability of our screen.

Figure 3.

Several MEs we identified overlap previously discovered cVM-IAPs. (A) Venn diagram representing genomic coordinate overlap between all 29 mouse MEs, cVM-IAPs, and ERViiDMRs. The legend to the right is for panels B-D, which portray MEs identified in or near (B) Eps8l1, (C) Rab6b, and (D) Rnf157, respectively. Above each, the ITC heat map indicates the minimum block-level inter-tissue correlation (of brain versus liver, kidney versus liver, and kidney versus brain). Purple and gray bars indicate the locations of MEs identified by our screen and cVM-IAPs described by Elmer et al. [27], and the yellow and brown segments indicate the assay coordinates used for their validation, respectively. Dot plots show bin-level % methylation data in each region (from WGBS) for each of the 10 mice in our screen. The RepeatMasker track shows the ERV LTR subclasses in the vicinity of each ME.

No effect of maternal dietary methyl-donor supplementation on ME methylation

We conducted a methyl-donor supplementation experiment, using the same diets and approaches as previously reported with A^vy and Axin^Fu mice [3, 8]. Starting two weeks prior to mating and continuing throughout pregnancy, female mice were provided either a control diet (13 litters) or the control diet supplemented with folic acid, vitamin B12, betaine, and choline (12 litters) [3, 8, 46] (See methods and Supplementary Table S9 for details). DNA methylation was measured in hepatic DNA of the offspring, using quantitative bisulfite pyrosequencing. The findings (Fig. 4) showed no significant effect of maternal supplementation on methylation at the MEs evaluated. Given the discrepancy between these results and those observed at A^vy and Axin^Fu [3, 8], we explored the hypothesis that effects of maternal supplementation on establishment of ME methylation may depend upon gene expression during pre-implantation development. Using a publicly available dataset (Database of Transcriptome in Mouse Early Embryos) [47], we analyzed gene expression profiles for ME-associated genes—Ahi1, Eps8l1, Rab6b, Prss16, Rims2, and Rnf157—alongside Agouti (Asp) and Axin1 (Supplementary Fig. S10). Agouti and Axin1 showed low and high expression, respectively, in the cleavage-stage embryo, whereas the newly identified MEs showed variable intermediate expression. Hence, expression patterns in the early embryo do not appear to correlate with the supplementation effect.

Figure 4.

No significant effects of maternal dietary methyl donor supplementation on offspring ME methylation were observed. Box plots summarize methylation data for each of 13 control litters (48 mice total, yellow) and 12 supplemented litters (48 mice total, teal). A–F show box plots of % methylation in liver within ME regions overlapping or near Snd1, Rims2, Rab6b, Ahi1, Eps8l1, and Rnf157, respectively.

Mouse MEs associate with LTR transposable elements

Classical mouse MEs such as A^vy and Axin^Fu were induced by retrotransposition of IAPs (a class of murine LTR retrotransposon) [4, 5]. We therefore asked whether the MEs we identified are associated with transposable elements. To evaluate this, we generated a set of control regions (Supplementary Table S10) each matched to one of the 29 MEs based on chromosome, size, and number of CpGs. Given the small number of MEs, to improve the precision of our analysis we generated three matched control regions for each; enrichments were adjusted for the larger number of control regions. Compared to the control regions, the 29 mouse MEs (including mMEs) were significantly enriched for direct overlaps with LTR retrotransposons (paired t-test, P = 0.0005) (Fig. 5A & B). This finding is consistent with observations at A^vy and Axin^Fu. Although there were a few overlaps with TEs from other repeat classes (Fig. 5A & B), these were not significantly enriched. Unlike MEs, SASIVs were enriched for direct overlaps with simple repeats (paired t-test, P = 0.01) (Fig. 5A & B, Supplementary Fig. S11A & B). The association of MEs with LTRs reflected enrichment at two IAP subfamilies: IAPEz-int (paired t-test, P = 0.02) and IAPLTR1 (paired t-test, P = 0.003) (Fig. 5C & D). Genomic coordinates for MEs and control regions that directly overlap either IAPEz-int or IAPLTR1 elements are listed in Supplementary Table S11. A summary of these, in comparison with the VM-IAPs identified by Elmer et al. [27], is shown in Fig. 5E; most the MEs and mMEs identified by our unbiased screen are associated with the same IAP subfamilies as are VM-IAPs. In particular, the strong overlap of mMEs with IAPEz-int and IAPLTR1 elements implicates these IAP subfamilies in epigenetic metastability. SASIVs, on the other hand, do not overlap these elements, further distinguishing them from MEs.

Figure 5.

Murine MEs are associated with specific classes of LTR transposable elements. (A) Heatmap summarizing counts of different classes of transposable elements directly overlapping SIV regions. Each column represents one of the 35 SIV regions; dotted lines separate MEs, multimodal MEs (mMEs), and sex-associated SIV regions (SASIVs). (B) Analogous heatmap for matched control regions (three control regions for each SIV region), in the same order as their matching SIV regions in panel A. Compared to control regions, MEs are enriched for long terminal repeats (LTRs) (paired t-test, P = 0.0005). (C) Heatmap summarizing counts of different LTR subfamilies overlapping SIV regions. (D) Analogous heatmap for matched control regions, in the same order as in panel C. Compared to control regions, MEs are enriched for IAPEz-int and IAPLTR1 elements (P_IAPEz-int= 0.02, P_IAPLTR1= 0.003). (E) Bar plot representing proportions of autosomal cVM-IAPs, MEs, multimodal MEs (mMEs), SASIVs and control regions that directly overlap various ERV subfamilies. Only subfamilies with a proportion ≥ 0.1 in at least one category are shown. Bars sum to > 1.0 because LTR overlaps are not mutually exclusive. Pie charts (inset) show the percentages of ERV overlaps for each category.

Mouse MEs show significant associations with distal enhancers

To further probe genomic determinants of epigenetic metastability, we investigated associations between ENCODE cCREs and MEs. Counting individual cCREs that directly overlap MEs or their matched controls (Supplementary Fig. S12A & B) revealed an enrichment of distal enhancers within MEs (Fisher’s exact test, P = 0.02). Extending this analysis to the 10kb flanking regions of MEs, mMEs, SASIVs, and their respective matched controls (Supplementary Fig. S12C & D) showed that, at mMEs, the enrichment of distal enhancers extends throughout the 10kb upstream flanking region (Fisher’s exact test, P = 0.0003) (Supplementary Fig. S12C). An apparent enrichment of proximal enhancers within 2kb upstream of mMEs (Supplementary Fig. S12D) was not statistically significant. We additionally asked whether specific sequence motifs are enriched in the vicinity of MEs, using “Multiple Expectation maximization for Motif Elicitation” (MEME) [48] but detected only associations with motifs contained within IAP consensus sequences. Compared to matched control regions, we found higher percentages of ME and mME loci containing cCREs overlapping TEs (Supplementary Fig. S13A). Among these, and consistent with previous findings by Costello et al. [49], distal enhancers and proximal enhancers tended to overlap with LTRs and SINEs, respectively (Supplementary Fig. S13B).

IAPEz-int elements in MEs may be targeted by specific KZFPs

Krüppel-associated box (KRAB) zinc finger proteins (KZFPs) comprise a large family of transcription factors known to target transposable elements to orchestrate their silencing during early embryonic development [37, 50–52]. To gain greater insight into the molecular mediators of epigenetic metastability, we investigated whether SIV regions contain known KZFP binding sites. We obtained MEME [48] output files containing the top binding motifs for numerous murine KZFPs from Wolf et al. [37] and utilized “Find Individual Motif Occurrences” (FIMO) [38] to screen SIV and matched control regions for KZFP binding motifs. Of the 45 KZFP binding motifs with data, 33 had at least one motif directly overlapping MEs (Supplementary Fig. S14A). While similar patterns of binding motifs corresponding to multiple KZFPs were detected in matched control regions (Supplementary Fig. S14B), binding motifs of KZFPs Gm13152, Zfp661, and Zfp759 were significantly enriched on MEs compared to matched controls (Fisher’s exact test, P= 0.02, P = 4.6 × 10⁻⁵, and P = 1.9 × 10⁻⁶, respectively).

Considering the enrichment of IAPEz-int and IAPLTR1 elements directly overlapping MEs (Fig. 5C & D), we next set out to determine whether these harbor KZFP binding motifs. We used FIMO [38] to search for KZFP binding motifs within the sequences of IAPEz-int and IAPLTR1 elements directly overlapping MEs. Binding motifs of three KZFPs (Gm13152, Gm13157 and Gm13251) were significantly enriched on IAPEz-int elements directly overlapping MEs (paired t-test, P= 0.01, P = 0.003, and P = 0.002, respectively) (Supplementary Fig. S15A & B). Within IAPLTR1 elements directly overlapping MEs, on the other hand, we found eight distinct KZFP binding motifs. None of the matched controls contained an IAPLTR1 element, yielding significant enrichments for seven out of eight KZFP binding motifs (paired t-test, P_Gm13157 = 0.005, P_Gm13251 = 0.04, P_Gm4631 = 0.03, P_Zfp599 = 0.003, P_Zfp661 = 0.003, P_Zfp738 = 0.02, P_Zfp759 = 0.01). Considering the concordance across IAPEz-int and IAPLTR1 elements, particularly Gm13157 and Gm13251, these results suggest that specific KZFPs play a role in regulating early embryonic maintenance or establishment of DNA methylation at murine MEs.

Mouse MEs are enriched for CTCF binding motifs

CTCF is an important developmental transcription factor whose binding to consensus sites is regulated by DNA methylation [53]. Enrichment of CTCF binding sites upstream of VM-IAPs was previously observed [27]. Using the same CTCF ChIP-seq dataset on liver of eight adult individuals [27], we evaluated enrichment of CTCF binding sites within the 1kb vicinity of MEs. Relative to matched control regions, we detected a statistically significant enrichment (paired t-test, P= 0.007) for CTCF binding motifs directly overlapping ME and mME loci (Supplementary Fig. S16A & B). SASIVs showed no enrichment, again distinguishing them from MEs. MEs with directly overlapping IAPs were particularly enriched for CTCF binding (Fisher’s exact test, P= 0.001, Supplementary Table S12).

Eps8l1 gene expression is associated with methylation at its ME

For gene-associated MEs at Eps8l1, Rab6b, Ahi1, Snd1, and Prss16, we investigated whether interindividual variation in methylation influences gene expression. In liver, kidney, and brain tissues from our validation cohort of 24 C57BL/6J mice, methylation was measured by quantitative bisulfite pyrosequencing and gene expression was measured by qPCR. Only Eps8l1 showed a significant correlation between methylation and expression, in both brain and liver (Supplementary Fig. S17). The negative correlation is consistent with the location of the ME at the Eps8l1 promoter (Fig. 3B). These findings indicate that, at least at some of the newly identified MEs, stochastic interindividual variation in DNA methylation influences gene expression.

Discussion

When Rakyan et. al. [1] introduced the term ME over 20 years ago, they proposed that the mouse genome might harbor a large number of such regions, offering a potential explanation for unexplained phenotypic variation within inbred populations of mice. To address this definitively, we designed a well-powered and unbiased screen, obtaining information on 95% of the ∼ 11.5M 100 bp bins containing CpG sites in the C57BL/6J genome. Our multi-tissue screen directly assessed SIV throughout the mouse genome and was based on the proven approach used to identify nearly 10 000 human CoRSIVs [15]. Nonetheless, we identified only 29 murine MEs. That 18 of these overlap previously identified cVM-IAPs [27] or ERViiDMRs [30] supports both the reliability of our screen and the conclusion that mouse MEs are extremely rare. Our study additionally discovered a small number of SASIVs.

We and others have shown that maternal dietary methyl donor supplementation increases DNA methylation at the A^vy and Axin^Fu MEs in the offspring [3, 8, 54–56] providing evidence for an epigenetic basis for developmental programming and suggesting that developmental establishment of DNA methylation at MEs is generally sensitive to maternal methyl donor nutrition. At the MEs discovered here, however, despite using the same diets and approaches as in the previous studies [3, 8], we detected no such effect. This result is consistent with recent findings of Bertozzi et al. [28] who, in addition to maternal dietary methyl donor supplementation, tested cVM-IAPs for sensitivity to other early environmental influences including maternal diets that were obesogenic (high fat) or supplemented with the endocrine disrupter bisphenol A. The lack of sensitivity to early environmental effects at mouse MEs contrasts with widespread developmental plasticity documented at human CoRSIVs. This is ironic, as the goal of identifying MEs in humans [10, 12, 17] led to the identification of human CoRSIVs [15]. Across our human studies, establishment of DNA methylation at both candidate MEs [10, 12, 17] and CoRSIVs [15] was found to be particularly sensitive to maternal periconceptional nutrition. Subsequently, studies examining a wide range of prenatal environmental exposures including maternal gestational diabetes [57], maternal famine exposure [58–60], maternal obesity [61, 62], maternal age [13, 14, 63], embryo culture during assisted reproduction [18] and maternal environmental contaminant exposure [64–66] document that at human CoRSIVs, establishment of DNA methylation in the early embryo is sensitive to its environment. Moreover, a recent study in dairy cattle [67] found that establishment of DNA methylation at CoRSIVs is likewise sensitive to embryo culture during assisted reproduction. Hence, whilst our current findings do not support our earlier hypothesis [68] that the stochastic nature of methylation establishment at murine MEs confers developmental plasticity, CoRSIVs in outbred mammals, at which DNA methylation is often associated with cis genetic variation [15], are generally sensitive to early embryonic environment.

We anticipated that identifying a representative cross-section of murine MEs would help elucidate the sequence features that confer epigenetic metastability. Across the many sequence features analyzed, including TEs, cCREs, DNA sequence motifs, and transcription factor binding sites, we found only limited associations with epigenetic metastability. For example, we found that murine MEs are associated with distal enhancers, both directly overlapping and in the several kb upstream (Supplementary Fig. S12). The strongest association we detected, linking MEs to IAP elements, is consistent with the classic mouse MEs A^vy and Axin^Fu [4, 5]. And yet, not all murine MEs are within or near IAPs (Fig. 5). Some are associated with other subfamilies of LTRs or even other classes of TEs. According to the “escapee model” [69], reprogramming of DNA methylation in the sperm and egg genomes during early zygotic development does not completely erase all DNA methylation marks. In mice, many of these regions that escape developmental reprogramming overlap IAP elements [69]. Notably, the A^vy locus does not appear to be among them [23]. Moreover, other regulatory elements and aspects of chromatin structure may confer resistance to reprogramming [70]. Hence, as demonstrated at A^vy [71] and Axin^Fu [25], the murine MEs we have identified may be substrates for transgenerational epigenetic inheritance. More broadly, the association of MEs with IAPs and the latter’s propensity for escaping epigenetic reprogramming call for a re-evaluation of the conceptual framework of epigenetic metastability. Whilst epigenetic metastability is generally thought to reflect stochastic establishment of DNA methylation states in the early embryo [1], it could instead reflect stochastic interindividual differences in the transgenerational maintenance of methylation inherited from the sperm and/or egg genome.

The enrichments for binding motifs of KZFPs Gm13152, Gm13157 and Gm13251 on IAPEz-int elements at ME loci (Supplementary Fig. S15) suggests that these might actively target IAPEz-int elements. IAPLTR1 elements at ME loci are also enriched for binding motifs of KZFPs Gm13157 and Gm13251. Interestingly, all these KZFP genes fall within a previously identified KZFP cluster located on chromosome 4 [37, 50]. However, in a ChIP-seq screen performed by Wolf et al. [37] assessing binding sites of KZFPs in mouse embryonic carcinoma cells, KZFPs Gm13152, Gm13157 and Gm13251 did not exhibit affinity for IAPEz-int elements. Rather, their data [37] show that IAPEz-int elements are targeted by seven (of 19) KZFP genes in a cluster on chromosome 2. Nevertheless, studies from Ferguson-Smith and colleagues using Chr4-cluster knockout mice concluded that the Chr4 KZFP cluster is vital for the establishment of interindividual variation in methylation at multiple VM-IAPs [50]. Our finding that three Chr4-cluster KZFP binding sites are enriched at ME loci is consistent with this hypothesis. Further investigation is needed to determine whether expression of specific KZFPs plays a role in determining interindividual epigenetic variation at MEs. Our results documenting enrichment of CTCF binding sites at MEs also agree with the previous work from the Ferguson-Smith group [26, 27] and work related to human SIV [72]. Together, our findings are consistent with earlier reports [27, 50] in suggesting that early embryonic expression of CTCF, specific KZFPs, and potentially other transcription factors are crucial for mediating the stochasticity of early embryonic establishment and/or maintenance of DNA methylation at TEs, leading to epigenetic metastability. The small number of murine MEs we identified, however, limited our ability to identify cis sequence features that are both necessary and sufficient to confer epigenetic metastability.

Whereas our screen was motivated by knowledge of classical MEs like A^vy and Axin^Fu [3, 8], most of the MEs we identified (and candidate MEs previously identified [27, 30]) do not demonstrate the broad and multimodal interindividual variation exemplified by those loci. Six of the loci we identified, however, which we termed mMEs, do exhibit broad and multimodal interindividual variation in % methylation. Due to the high density of repetitive elements in the vicinity of several of these, we were able to design quantitative pyrosequencing assays for only two (11_25 132_Rnf157, 15_4744_Rims2); both of these validated. These two were also assessed in our methyl donor supplementation experiment, but did not show any effect of maternal supplementation on offspring methylation. Despite this negative result, the patterns of inter-read heterogeneity we observed at mMEs (Supplementary Fig. S8) are reminiscent of those found at the A^vy locus in multiple early developmental stages [23]. Because co-suppression of retrotransposons through methylation, which occurs prior to gastrulation, is a stochastic event specific to each retrotransposon copy, Whitelaw and Martin [73] proposed that TE-induced interindividual variation in methylation will lead to epigenetic mosaicism, consistent with our findings. Indeed, all six mMEs directly overlap an LTR retrotransposon (Fig. 5A), and five of these belong to the IAPEz-int or IAPLTR1 sub-families (Fig. 5C). Future studies will be required to better understand the apparently distinct processes of early embryonic establishment of DNA methylation at mMEs and unimodal MEs.

Mouse autosomal regions that display sexually dimorphic methylation were reported several years ago [42, 43], but these were almost entirely tissue specific. A more recent study [43] used WGBS to identify Y chromosome-dependent differentially methylated regions (yDMRs). Similar to SASIVs, at some of these yDMRs sex-associated methylation differences were consistent across different tissues representing different embryonic germ layer lineages; none, however, overlapped with SASIVs. In humans, Grant et al. [74] used the Illumina EPIC array to investigate autosomal sex differences in DNA methylation in whole blood, identifying 396 sex-associated differentially methylated CpG sites. Across all of these studies [41–43, 74], females generally show higher methylation, consistent with our findings at SASIVs. Sex DMRs [42] were found to be enriched for ERVK repetitive elements, and nearly all of 19 yDMRs [43] overlap either a LINE1 or ERVK TE. SASIVs, on the other hand, are associated with simple repeats (Supplementary Fig. S11). SIV is thought to result when methylation states are established prior to gastrulation and maintained during subsequent differentiation [8, 10] suggesting that, at SASIVs, sex differences in methylation are established prior to sexual differentiation, which occurs in mid-gestation. Although most SASIVs are intergenic, one is located in the promoter region of C1s2. C1s, the human ortholog of C1s2, is associated with two autoimmune diseases, systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA) [75]. Interestingly, SLE is nine times more common [76] and RA is 2–3 times more common in females than males [77]. Future studies will be required to determine if sex-associated methylation differences at C1s2 are conserved across humans and mice and implicated in the sex biases for these diseases.

Although we identified a small number of MEs, some gene-associated MEs could contribute to interindividual phenotypic variation. For example, Eps8l1 (Fig. 3B), also known as DRC3, is linked to male infertility in both humans and mice. As the Eps8l1 ME overlaps the promoter, and its methylation is inversely associated with expression (Supplementary Fig. S17), male mice with high methylation at the Eps8l1 ME may exhibit fertility defects. Another ME is located in the promoter region of Ahi1. In humans, mutations in Ahi1 are known to cause Joubert syndrome-related disorders, a neuromuscular constellation involving impaired cilia function and ataxia [78]. Mouse models show that knockout of Ahi1 impairs motor and muscle development [79], so interindividual variation in Ahi1 methylation could affect motor development and physical activity in mice. We identified three ME regions within the gene body of Rnf157 (one is illustrated in Fig. 3C). Rnf157 is strongly expressed in the brain, and in vitro studies indicate it regulates important aspects of neuronal development [80], suggesting that epigenetic metastability at Rnf157 may be associated with neurological outcomes. We identified a 200bp ME overlapping the mouse Ephx2 gene on chromosome 14, which encodes a soluble epoxide hydrolase. Human Epx2 variants have been associated with cholesterol metabolism and atherosclerosis [81] and cardiac disorders, male infertility, and kidney function [82]. Whereas interindividual variation in DNA methylation occurs systemically at MEs, effects on expression may be tissue- and/or developmental stage-specific. All the aforementioned ME-associated genes are widely expressed, so characterizing effects on expression and phenotypic variation will require studying a wide range of tissues and developmental stages. Future studies should determine if interindividual differences in DNA methylation at these loci among isogenic mice associate with gene expression and phenotypic outcomes.

In conclusion, this study marks a major advance in the understanding of epigenetic metastability in mice. As murine MEs have been known for over 50 years [2, 83], an unbiased assessment of epigenetic metastability in the mouse was long overdue. Our screen provided precise coordinates of highly reliable mouse MEs and documents that they are extremely rare in C57BL/6J mice. Consistent with the recent findings of Bertozzi et al. [28], we found no evidence for effects of maternal methyl donor supplementation at the MEs we identified. However, rather than contradicting classic observations of dietary effects at A^vy and Axin^Fu, our interpretation is that developmental plasticity may be limited to a subset of murine MEs. It is likely that at some of the MEs we identified, interindividual epigenetic variation will have phenotypic consequences. Given their rarity, however, MEs are not likely to explain the many examples of phenotypic variation that have been observed among isogenic mice [20–22]. Future studies may expand upon our screen to assess other inbred mammalian species (such as rats and guinea pigs) and additional strains of mice to gain further insights into the sequence features that drive epigenetic metastability and the phenotypic consequences of such variation.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

Funding for this project was provided by NIH/NIDDK (1R01DK125562) and the USDA/ARS (CRIS 3092-5-001-059) to R.A.W. J.N.W. was supported by NIH R35-GM122550. Y.Z. was supported by NIH R01-DK136619, R01-DK136532 and USDA/ARS (CRIS 58-3092-5-008). Funding to pay the Open Access publication charges for this article was provided by 1R01DK125562.

Data availability

The raw WGBS data from the 30 methylomes have been deposited in GEO (PRJNA1070014) https://www.ncbi.nlm.nih.gov/bioproject/1070014.