Retrotransposons as a major source of epigenetic variations in the mammalian genome

Abstract

Transcription of retrotransposons is usually repressed by DNA methylation, but a few elements, such as intracisternal A-particles (IAPs) associated with the Agouti and Axin-fused loci, partially escape this repression mechanism. The levels of this repression are also variable among individuals with an identical genome sequence, generating epigenetically different states of loci or ‘epialleles.’ In the current study, we tested the existence of additional retrotransposon-derived epialleles in the mouse genome. Using a series of bioinformatic approaches, 143 candidate epialleles were first identified from the mouse genome based on their promoter activity and association with active histone modification marks. Detailed analyses suggest that a subset of these elements showed variable levels of DNA methylation among the individual mice of an isogenic background, revealing their stochastic nature (metastability) of DNA methylation. The analyses also identified two opposite patterns of DNA methylation during development, progressive gaining vs. losing, confirming the dynamic nature of their DNA methylation patterns. qRT-PCR analyses demonstrated that the expression levels of these elements are indeed variable among the individual mice, suggesting functional consequences on their associated endogenous genes. Overall, these data confirm the presence of a number of new retrotransposon-derived epialleles with suggestions of the presence of more, and further identify retrotransposons as a major source of epigenetic variations in the mammalian genome.

Introduction

Epigenetic modification on the mammalian genomes is part of the normal process of development, and also plays a critical role in many biological processes, including transcriptional control of development and tissue-specific genes, X chromosomal inactivation, and genomic imprinting.¹ Epigenetic modification is usually mediated through specific enzymatic reactions, and thus influenced by many variables, such as substrates’ concentration and enzyme activity. As a result, even if the sequence of the human genome is identical as in the case of a monozygotic twin, these processes could easily derive several different versions of the human epigenome, providing a source for inter-individual variations.² According to recent epidemiological studies, the human population also displays very unusual inter-individual variations in terms of its susceptibility to several adult-onset diseases, including cancer, diabetes, obesity, and mental illness.^3-6 Yet, this phenotypic variability has often been observed even between human monozygotic twins and among other model organisms with isogenic backgrounds.^7,8 These observations suggest that epigenetic, but not genetic, variations are likely responsible for many of the phenotypic differences observed in humans and other model organisms.

It is currently unknown what genomic elements contribute most significantly to variations in the epigenome. However, according to the studies of two mutant mouse phenotypes, viable yellow agouti (A^vy) and axin-fused kinky (Axin^fused), retrotransposons of relatively recent origin tend to show inter-individual variations in their DNA methylation levels and also concurrent phenotypic variations.^4,6 Interestingly, both mutant phenotypes are caused by intracisternal A-particle (IAP) insertions and subsequent ectopic expression of the two endogenous genes, agouti and axin genes, driven by the IAP elements’ long-terminal repeat (LTR). In both loci, the IAP LTR partially escapes DNA methylation-mediated repression, and triggers the production of fusion transcripts of LTR and endogenous gene’s transcripts. These LTR-driven ectopic expressions subsequently cause disruptions of the cell type specific and time or stage specific transcription of the endogenous genes, resulting in coat color change and obesity for the agouti locus and tail kinkedness for the axin-fused kinky locus. The different levels of DNA methylation on the two LTRs also correlate well with the expression levels of fusion transcripts and associated phenotypes.

The majority of retrotransposons are usually repressed by DNA methylation, and the degrees of DNA methylation are complete and static in different somatic tissues. In contrast, the DNA methylation on the two LTRs of the agouti and axin loci is partial and metastable. These specific LTRs display a wide range of DNA methylation (30–90%) within the mice of an isogenic genetic background. The DNA methylation on these LTRs is also easily affected by nutritional supplements and other environmental agents particularly during early embryogenesis.^9-11 Such distinctive nature of these retrotransposons and their associated alleles earned them the term ‘metastable epialleles’ in order to distinguish them from traditional alleles.¹² This variable and metastable nature of DNA methylation in these retrotransposons further predicts that phenotypic variations in mice, humans and other mammals may be caused by epigenetic variations stemming from similar retrotransposons. To test this prediction, we performed a series of bioinformatic searches and DNA methylation analyses on a subset of potential epialleles in mouse. The results from the analyses of this data indicate a strong probability that mammalian genomes contain a large number of uncharacterized metastable epialleles that are derived from retrotransposons and, further, that these elements may be a major source of inter-individual variations in the mammalian epigenome.

Results

Identification of potential metastable epialleles using bioinformatic searches

A series of bioinformatic searches was performed to identify additional retrotransposons with similar epigenetic features as the two epialleles, the agouti and the axin-fused kinky loci, that have been mentioned earlier (Fig. 1 and Fig. S1). For our study, we excluded SINE elements due to the prediction that the majority of RNAP III-driven SINEs may not be a good source for promoter elements for RNAP II transcription, even though it is possible that some opportunistic SINEs could also function as promoters for RNAP II transcription.¹³ So here and onwards, we will be meaning only LINEs and LTRs when we mention retrotransposons. First, we looked for retrotransposon and mRNA pairs where the transcription start site (TSS) of the mRNA is located within 50 bp upstream and 250 bp downstream of the 5′-end of the retrotransposon by intersecting the entire retrotransposon set (1,767,082 elements containing LINEs and LTRs) with the transcriptome of the mouse (228,765 mRNAs). This search identified 5,679 retrotransposons that may initiate or influence the transcription of adjacent mRNAs. Second, this set of retrotransposons was further filtered using the following criteria. The size of a given retrotransposon needed to be equal to or greater than 50 bp in length and also the element needed to contain at least 3 CpG dinucleotides, which were necessary for DNA methylation analyses at the later stages. These criteria further trimmed down the initial set to a group of 1,982 retrotransposons (1,266 LTRs and 716 LINEs) with potential RNAP II promoter activity.

Figure 1. Scheme of the bioinformatic approach for identification of potential epialleles from retrotransposons. The database of approximately 1.77 million retrotransposons and 0.2 million mRNAs (version mm9) were intersected by using custom scripts and applying a set of defined filters. The intersection identified 1,982 retrotransposon (1,266 LTRs and 716 LINEs) and mRNA pairs which were then further intersected with the combined histone 3 lysine 4 trimethylation (H3K4me3) ChIP-Seq peak data derived from four different tissues and cell lines: adult liver, ES cell, MEF cell and NP cell. This last intersection provided some 143 candidate metastable retrotransposons that have at least one mRNA originating in close proximity of its 5′-end and also have an overlapping H3K4me3 signal on it.

As an independent approach for finding retrotransposons with promoter activity, we performed a series of database searches using genome-wide ChIP-Seq data, which were designed to survey active (H3K4me3) regions in different types of cells, including ES (Embryonic Stem), NP (Neural Precursor), and MEF (Mouse Embryonic Fibroblast) cells,¹⁴ as well as adult liver.¹⁵ According to this survey, the majority of LINEs and LTRs (551,827) are not associated with an active histone mark (H3K4me3) but rather marked by repressive histone modifications (H3K9me3 or H3K20me3) and their genomic locations are mostly away from gene promoter regions (Fig. 2B). However, a small fraction of LINEs and LTRs (4,226) are modified with an active mark (H3K4me3), and a large fraction of this set of retrotransposons (32%) tends to be located in close proximity (within 10 Kb upstream) to the promoter regions of genes (Fig. 2A and B). This retrotransposon set with H3K4me3 was further intersected with the initially identified set of 1,982 retrotransposons with potential promoter activity (elements in proximity to mRNA TSS, where the TSS of the mRNA is within -50 to 250 bp of the 5′ end of the retrotransposon). This intersection identified a set of 143 retrotransposons that contained both potential promoter activity and an active histone mark (Fig. 1) and thereby can be considered excellent candidates for metastable epialleles (MEs). This set contains even greater fraction of retrotransposons (66%) located in close proximity (within 10 Kb upstream) to the promoter regions of genes (Fig. 2B). It is also interesting to note that many retrotransposons are associated with active histone marks mainly in two cell or tissue types, ES cells and adult liver, which might be indicative of the greater transcription rates in liver and ES cells, even though MEF and NP cells are not as fully differentiated as liver cells are (Fig. 2C).

Figure 2. Detailed bioinformatic analysis of the candidate and other relevant retrotransposons. (A) Verification of the presence of the H3K4me3 ChIP-Seq peak in the vicinity of a randomly chosen retrotransposon and its associated mRNA pair from the pool of 143 candidate metastable epialleles. The retro/mRNA pair associated with the gene Eps8R1 have been shown here. The figure shows a snapshot from the UCSC Genome Browser in the mm8 version of mouse genome where the ectopic transcript Eps8R1 was still annotated with the same name as the endogenous transcript Eps8l1. The current mm9 version of the UCSC Genome Browser, which has the proper annotation, does not have the “Broad H3 ChIPseq” track that was used to show the H3K4me3 ChIP-Seq peaks. The red arrow indicates the retrotransposon-driven ectopic transcript, the blue arrow indicates the endogenous transcript of Eps8R1 (annotated as Eps8l1), the orange arrow indicates the retrotransposon driving the ectopic transcript and the red box surrounds the confirmed H3K4me3 ChIP-Seq peak. This ChIP-Seq result also shows another peak around the endogenous promoter region marked by the blue box. (B) Percentage of retrotransposons in the three different pools based on their relative positions to annotated genes. The pools are: i) filtered set of retrotransposons that does not intersect with either mRNA or H3K4me3 ChIP-Seq peaks (551,827 LTRs and LINEs); ii) filtered set of retrotransposons that intersects with H3K4me3 ChIP-Seq peaks (4,226 LTRs and LINEs) and; iii) filtered set of retrotransposons that intersects with both mRNA and H3K4me3 ChIP-Seq peaks (143 LTRs and LINEs). (C) Four-way Venn diagrams showing the number of retrotransposons (LTRs and LINEs only) obtained for each type of cell lines or tissues during the separate intersection of H3K4me3 ChIP-Seq peak data with the retrotransposons having nearby mRNAs (143 elements) and with the retrotransposons that do not have nearby mRNAs (4,083 elements).

The identified set of 143 retrotransposons are comprised of 99 LTRs and 44 LINEs; ERVK and L1 subfamilies make up the major subfamilies for LTRs and LINEs, respectively, with 47 ERVKs and 24 L1s (Fig. S2). It is rather surprising, though, that this set of 143 retrotransposons do not contain any copy of the mouse ERV family, the Early Transposons (ETns), some of which have been demonstrated to show significantly variable DNA methylation.¹⁶ These retrotransposons are distributed among all the chromosomes, except the Y chromosome. The genomic features surrounding these 143 retrotransposons were further examined using the UCSC genome browser (Fig. 2A). According to this manual inspection, each of the 143 retrotransposons can be assigned exclusively to one of four groups based on their position relative to genes: a) 89 out of the 143 retrotransposons express transcripts from the promoter regions (within 10 Kb upstream) of endogenous genes. Thirty out of these 89 retrotransposons appear to function as completely alternative promoters for adjacent endogenous genes (Fig. 2A), ultimately generating fusion transcripts between the retrotransposons and endogenous genes, as seen in the viable yellow agouti locus. The rest 59 of the retrotransposons in this group are located very close to the endogenous promoters (within -1 to 0 Kb of the TSS or within the first exon); b) 21 out of the 143 retrotransposons also function as promoters for RNAP II transcription in more gene-poor non-genic regions producing transcripts, in most cases without any obvious open reading frames within their sequences (Fig. S3A). Thus, the majority of these transcripts are novel and likely non-coding RNA genes; c) 34 of the 143 retrotransposons express transcripts from the inside of other endogenous genes, as seen in the axin-fused kinky locus¹⁷ (Fig. S3B). Interestingly, 11 out of these 34 retrotransposons have their transcriptional direction opposite to that of the endogenous genes and, thus, the transcription of these elements might interfere with that of the associated endogenous genes; d) only 1 of the 143 retrotransposons expresses transcripts that originate from the downstream of an endogenous gene and is transcribed in the opposite direction of that gene. A list of the 143 retrotransposons (candidate MEs) along with their associated mRNAs and endogenous genes has been provided in Supplemental Data Set 1. Overall, the series of bioinformatic approaches described above were implemented to identify an initial set of 143 retrotransposons, which were investigated as potential candidates for metastable epialleles in the following sections.

DNA methylation analyses on the identified retrotransposons

We analyzed the DNA methylation levels of 13 randomly chosen retrotransposons from the set of 143 retrotransposons or candidate MEs (Fig. 3 and Table 1). We also included two retrotransposons (associated the genes Rgs9 and Cog6) that have potential promoter activity but do not have any H3K4me3 mark (Table 1). Three additional loci were selected as controls for analysis: a house keeping gene (Gapdh), an imprinted gene (Peg3), and one IAP element (located within the 6th intron of Cdk5rap1), which was previously shown to be a metastable epiallele.¹⁸ The DNA methylation levels of these retrotransposons were analyzed using a set of mouse tail DNA derived from one litter of five 1-week-old pups with an isogenic background (C57BL/6J). DNA was first treated with the bisulfite conversion method¹⁹ and subsequently used for PCR amplification, for which one of the primers was designed from within the retrotransposon itself, while the other primer from the adjoining endogenous and non-repeat region. The amplified products were digested with their appropriate restriction enzymes, the digestion or non-digestion of which indicate the methylated or unmethylated status of the original DNAs²⁰ (COBRA, Combined Bisulfite Restriction Analysis). In order to test the completeness of digestion of the restriction enzymes, all the enzymes used in this experiment and throughout the study were made to digest specific PCR products (amplified from non-bisulfite treated genomic DNA) that had the restriction recognition site of the enzyme concerned (Fig. S4).

Figure 3. Combined Bisulfite Restriction Analysis (COBRA) of randomly selected candidate metastable retrotransposons in five 1-week-old littermates. Tail DNAs isolated from five 1-week-old littermates of the C57BL/6J strain were bisulfite-treated, PCR-amplified and digested with restriction enzymes. The relative methylation status of the CpG sites at the enzyme recognition sites for each PCR product were judged from the intensities or the relative proportions of the bands obtained after gel electrophoresis. The endogenous genes associated with the mRNA/retrotransposon pairs are indicated on the left and the restriction enzymes used are indicated on the right of the electrophoresis gel picture. The bands expected in case of methylated and unmethylated states of the PCR amplicon after the enzyme digestion are marked with M and U, respectively. The promoter region of Gapdh and the DMR of Peg3 were used as negative controls and the IAP-LTR element associated with Cdk5rap1 was used as a positive control. The mRNA/retrotransposon pairs associated with the endogenous genes Rgs9 and Cog6 are H3K4me3 mark-deficient.

Table 1. Candidate metastable retrotransposons tested in the COBRA of 1-week-old littermates.

Associated Endogenous UCSC Gene	Retrotransposon				Associated mRNA
	Coordinates	Name	Family	Class
Rgs9*	chr11: 109165513–109165759	IAPEY3_LTR	ERVK	LTR	AK166518
Cog6*	chr3: 52798218–52798593	IAPLTR2_Mm	ERVK	LTR	BC071236
Gimap9	chr6: 48625715–48626103	RLTR9F	ERVK	LTR	AK041336
AK145129	chr2: 118380501–118380977	IAPLTR2_Mm	ERVK	LTR	AK145129
AK011682	chr11: 113034330–113034803	IAPLTR2_Mm	ERVK	LTR	AK011682
Bcat2	chr7: 52826548–52826692	LTR65	ERV1	LTR	AK160159
Eps8R1	chr7: 4411952–4412304	IAPLTR1_Mm	ERVK	LTR	BC027043
Ogdh	chr11: 6192483–6192694	LTR31	ERV1	LTR	BC049104
Msl3l2	chr10: 55826590–55826754	RMER15	ERVL	LTR	BC037487
AK009289	chr10: 95143475–95143681	RLTR18B	ERVK	LTR	AK009289
Alkbh8	chr9: 3334707–3335338	Lx9	L1	LINE	BC050863
Gsta3	chr1: 21234684–21235118	RodERV21-int	ERV1	LTR	BC147272
Popdc2	chr16: 38376981–38377439	MTC	MaLR	LTR	AK171056
Repin1	chr6: 48548858–48549049	L3	CR1	LINE	AK085289
Spsb3	chr17: 25024291–25024393	L2	L2	LINE	AF403038

^*The retrotransposons associated with the genes Rgs9 and Cog6 are H3K4me3 mark-deficient ones and are not part of the candidate metastable retrotransposon list.

As shown in Figure 3, the promoter region of a housekeeping gene (Gapdh) was not digested by TaqI, indicating the unmethylated status of this gene. In contrast, the Differentially Methylated Region (DMR) of one imprinted gene (Peg3) showed about 50% digestion with HphI and thus 50% methylation, consistent with the fact that only one allele, the maternal allele, is methylated.^21,22 As expected, the known metastable IAP (Cdk5rap1) showed a partial methylation pattern with inter-individual variation: Mouse #1 and 5 had much lower methylation levels than their littermates, as evident from the intensities of the bands expected for methylated and unmethylated products after digestion. Among the 13 candidate MEs (10 LTRs and 3 LINEs) tested, which do not include the H3K4me3 mark-deficient Rgs9 and Cog6, only one locus was completely unmethylated (Gimap9), whereas the other 12 loci showed partial methylation patterns, mostly unmethylated, with potential inter-individual variations based on the detection of different levels of digestion with the same enzyme for a specific locus in some of the cases (Eps8R1, AK145129, Msl3l2, Gsta3 and Repin1). We also performed a second set of COBRA for some of the 13 candidate MEs and also Rgs9 depending upon the availability of a different enzyme site within the retrotransposons (Fig. S5). As observed in the first set of COBRA, most of the loci were unmethylated while the Repin1 locus showed mostly methylation. Since COBRA, however, is limited to the methylation status of a single CpG site, we further followed up this possibility through cloning and sequencing the PCR products of five of the loci (Bcat2, Eps8R1, Ogdh, AK011682, and AK145129) for more extensive and accurate DNA methylation analysis (Fig. 4). The results confirmed variable levels of DNA methylation in the retrotransposons among the individual mice: 0 to 44% for Bcat2, 26 to 72% for Eps8R1, 0.3 to 53% for AK011682, and 19 to 46% for AK145129. Even though the retrotransposon associated with Ogdh shows no significant variable methylation by itself (0 to 6%), the retrotransposon, along with the flanking regions, does show some variation in DNA methylation: 15 to 26%. Analyses of the DNA methylation of the flanking regions of the retrotransposons show some distinct patterns for at least three of the loci. The flanking region upstream of the retrotransposon in the Bcat2 locus shows a methylation pattern independent of that of the retrotransposon itself, while the methylation pattern of the downstream flanking region mimics more or less that of the retrotransposon. The downstream flanking region (the only sequenced flanking region) of the retrotransposon in the Eps8R1 locus shows methylation pattern that is similar to that of the retrotransposon itself. The methylation of the flanking regions of the retrotransposon in the Ogdh loci show some interesting patterns: the upstream flanking region has similar methylation pattern as the retrotransposon itself while the downstream flanking region shows substantial and variable methylation with an abrupt boundary at the border of the retrotransposon. In order to evaluate the statistical significance of the methylation variations of the CpG sites between individual mice for each of the five loci, we performed the Mann-Whitney U-test (also known as the Wilcoxon rank-sum test) on the bisulfite sequences of two individuals at a time in all combinations using the free online tool QUMA.²³ The p values obtained from the tests showed significant to moderate methylation variations among the individuals for the Bcat2, Eps8R1, and AK011682 loci, while the Ogdh and AK145129 loci did not show any statistically significant variations (Supplemental Data Set 1). In contrast to these five retrotransposons with variable DNA methylation patterns, the two retrotransposons without any H3K4me3 mark (associated with the genes Rgs9 or Cog6) showed almost complete methylation with very minute intensities for unmethylated fragments in COBRA (both sets in case of Rgs9) and this was confirmed by cloning and sequencing the PCR product of one of those loci, Cog6, which showed almost complete methylation for all five littermates (Fig. 3; Figs. S5 and S6). Overall, the preliminary results confirmed that 12 out of the 13 chosen candidate MEs are partially methylated and are either equally or more unmethylated. The results, particularly the bisulfite sequencing, also demonstrated that some of these retrotransposons indeed display inter-individual variations in their DNA methylation levels, further supporting the feasibility of our approach.

Figure 4. Bisulfite sequencing of five highly potential candidate metastable retrotransposons in the tail of 1-week-old littermates. The PCR products of the bisulfite-treated tail DNA of the five 1-week-old littermates were cloned and sequenced to check the methylation status of the CpG sites of the potential metastable retrotransposons associated with the genes Bcat2, Eps8R1, Ogdh, AK011682 and AK145129. The bisulfite sequencing result has been shown using a bubble chart. Each row represents a different clone and each column represents a different CpG site. Filled and open circles indicate methylated and unmethylated cytosines, respectively. The CpG sites are arranged in the 5′ to 3′ direction of the sequenced region while going from left to right in each clone. The numbers on the far left of the figure represent the mouse number corresponding to the COBRA (Fig. 3). The red boxes encompass the CpG sites that are within the retrotransposon. In case of AK011682, all the CpG sites shown are within the retrotransposon. The percentage (without the parentheses) on the right of each bubble chart indicates the percentage methylation of the CpG sites within the retrotransposon itself and the percentage within the parentheses indicates the percentage methylation of the CpG sites within the retrotransposon and adjoining regions in that specific sequenced DNA sample. The CpG sites analyzed in the first set of restriction enzyme digestion are marked by brown inverted triangles, while those analyzed in the second set are marked by blue inverted triangles.

We also analyzed the DNA methylation levels of the five candidate MEs that were sequenced in more detail along with the two H3K4me3 mark-deficient retrotransposons using the DNA derived from different tissues of two adult mice (2- and 5-mo-old, Fig. 5). The two H3K4me3 mark-deficient LTRs (Rgs9 and Cog6) had almost complete methylation patterns in all the tissues tested for both ages, which is similar to the patterns seen in 1-week-old mice (Fig. 3). Two candidate MEs (AK011682 and AK145129) had similar levels of partial methylation patterns throughout all different tissues at both ages, which is also similar to the patterns seen in the neonatal set. However, the DNA methylation levels of the three remaining candidate MEs (Bcat2, Eps8R1, and Ogdh) showed quite different patterns of DNA methylation among the tissues analyzed. The candidate ME associated with the Bcat2 gene had almost no methylation in lung and spleen at both ages in addition to the tail at 5-mo-old stage, whereas partial methylation was observed in the other tissues. The candidate ME associated with the Ogdh gene had almost no methylation in liver (particularly the 5-mo-old liver), lung, and kidney at both ages, whereas it contained partial methylation in brain, tail, and spleen. However, the candidate ME associated with the Eps8R1 gene did not appear unmethylated like the other two loci but was significantly unmethylated in lung at both ages as well as in tail, kidney, and spleen at 5-mo-old stage, whereas partial methylation was observed in the other tissues. Overall, the four retrotransposons analyzed (Rgs9, Cog6, AK011682, and AK145129) seem to maintain similar patterns of DNA methylation in the adult tissues, as was previously observed in the viable yellow agouti locus, but, interestingly, three of the retrotransposons (Bcat2, Eps8R1, and Ogdh) displayed tissue-specific, fluctuating DNA methylation patterns, which might be reminiscent of tissue dependent methylation profiles observed in the axin-fused kinky locus.¹¹ The possible reason for the hypomethylation at some of the 5-mo-old stage tissues could be a combination of two phenomena: a progressive loss of methylation during the aging process and the presence of inter-individual variations among the 2-mo- and 5-mo-old mice.

Figure 5. COBRA of the identified metastable retrotransposons in different tissues from 2- and 5-mo-old mice. DNA methylation analysis of the identified metastable retrotransposons during different growth stages was done by COBRA on DNA obtained from six different tissues of two individual mice at two different ages (2 and 5 mo). The tissues were chosen from three germ layers: ectoderm (brain and tail), endoderm (liver and lung), and mesoderm (kidney and spleen). As in the previous COBRA, Gapdh and Peg3 were used as negative controls whereas the IAP-LTR associated with Cdk5rap1 was used as a positive control. The names of the tissues are indicated at the top of each column. The associated genes, the restriction enzymes used and the expected bands in case of methylation and unmethylation are mentioned as in Figure 3.

Developmental profiling of DNA methylation patterns of selected retrotransposons

The DNA methylation levels of the same set of retrotransposons were further analyzed using sperm, oocyte, and blastocyst-stage embryos derived from the C57BL/6J mice (Fig. 6). For this series of COBRA, we isolated sperm from 4 male littermates (individually), a pool of 60 mature oocytes from three female littermates and, finally, a pool of 10 blastocyst-stage embryos after timed mating. The DNA purified from these samples was subsequently analyzed in a similar manner as described above. To monitor the purity of germ cells, we first analyzed the DMRs of a control set, the two imprinted genes H19 (maternally expressed) and Peg3 (paternally expressed). As shown in Figure 6, the DMRs of H19 and Peg3 were completely unmethylated in the oocyte and sperm, respectively, confirming the purity of the germ cells. Thus, we analyzed the DNA methylation levels of the 7 retrotransposons using the DNA purified from germ cells and early-stage embryos.

Figure 6. COBRA of the identified metastable retrotransposons in germ cells and blastocyst. Analysis of DNA methylation status of the newly identified metastable retrotransposons at different developmental stages. COBRA was performed on DNA obtained from sperm of four male littermates, combined oocyte from three female littermates and blastocyst-stage embryos from mating pairs of the same litter. Gapdh, Peg3 (in sperm), and H19 (in oocyte) were used as negative controls whereas the IAP-LTR associated with Cdk5rap1 was used as positive control. The associated genes, the restriction enzymes used and the expected bands in case of methylation and unmethylation are mentioned as in Figure 3.

We summarized the results from this series of analyses (Fig. 6) along with the results from the neonates and the tissues of adult stages (Figs. 3 and 5), which subsequently provide a developmental overview of DNA methylation patterns for the tested retrotransposons (Fig. 7). According to the three result sets, the retrotransposons display four developmental patterns of DNA methylation. First, the two H3K4me3 mark-deficient LTRs (Rgs9 and Cog6) display complete methylation throughout development, consistent with the fact that these two LTRs are not associated with H3K4me3 mark in any cell types examined. Second, one candidate ME (AK011682) appears almost completely unmethylated throughout development, including germ cells and blastocysts (Fig. 6). Third, three candidate MEs (Bcat2, Ogdh and AK145129) show a pattern of progressive gaining of DNA methylation, although the candidate MEs associated with AK145129 appear to be partially methylated in the sperm. The endogenous genes associated with this group (Bcat2 and Ogdh) are expressed ubiquitously and, thus, the unmethylated status of the associated candidate MEs in germ cells and blastocysts somewhat agrees with the ubiquitous expression pattern of the endogenous genes. Fourth, the two candidate MEs (Eps8R1 and Cdk5rap1) start with complete methylation in the germ cells and/or blastocysts, but progressively lose their DNA methylation levels in the tissues of neonatal and adult stages. The endogenous genes associated with this group are tissue- and stage-specific genes and, thus, the complete methylation of the associated candidate MEs in early stages also reflects the promoter activity of the endogenous genes. Overall, it is prudent to note that the DNA methylation patterns of the latter two groups of retrotransposons are somewhat consistent with the expression patterns of the associated endogenous genes. Furthermore, according to the results from detailed sequencing (Fig. 4), the retrotransposons with relatively greater levels of inter-individual variations belong to these two groups, the groups with progressive gaining and losing of methylation. This observation suggests that the fluctuating DNA methylation patterns embedded in the associated endogenous genes might be a contributing factor to the inter-individual variations observed in the associated retrotransposons.

Figure 7. Model of progression of DNA methylation status of retrotransposons through different developmental and growth stages. Four different patterns of methylation change throughout the developmental and growth stages have been proposed for retrotransposons after detailed DNA methylation analysis for the candidate metastable retrotransposons (those associated with genes or mRNAs: Bcat2, Eps8R1, Ogdh, AK011682 and AK145129), the established metastable retrotransposon (Cdk5rap1), and for the retrotransposons that do not intersect with any H3K4me3 data (those associated with Rgs9 and Cog6). The methylation state for each representative colored pie chart has been described in the figure.

Ectopic and variable expression of endogenous genes by the associated retrotransoposons

The three candidate MEs (Bcat2, Eps8R1 and Ogdh) were further analyzed to test whether they trigger ectopic expression of their associated endogenous genes and whether that ectopic expression is variable at the same time. For this series of analyses, three tissues (brain, kidney and lung) from three 5-mo-old littermates were used for isolating total RNA and subsequent qRT-PCR analyses (Fig. 8). All of the three candidate MEs indeed trigger transcription based on the detection of the PCR products that are derived from fusion transcripts of the LTRs and the associated endogenous genes (Fig. 8A). LTR-driven ectopic expressions were detected in all three tissues examined in at least two out of the three animals. The relative expression levels of the ectopic vs. endogenous transcripts were further measured for Bcat2 and Ogdh since the endogenous expression of these two genes was detected in all three tissues. The expression levels of the ectopic expression were very minimal, with ranges of 0.18 - 7.0% (Bcat2) and 0.61 - 4.3% (Ogdh) of the respective endogenous genes, based on their relative C_t (threshold cycle) values. Given the very low expression levels, the LTR-driven ectopic expression is thought to have minimal impact on the function of the Bcat2 and Ogdh loci. In the case of Eps8R1, out of the three mice tested, the expression of the endogenous transcript was detected only in the kidney of Mouse #2, as well as in all three tissues of Mouse #3; yet, the LTR-driven expression was readily detectable in all three tissues from all three mice (Fig. 8A). Thus, the LTR-driven ectopic expression might have a significant impact on the function of the Eps8R1 gene.

Figure 8. Verification and quantification of the ectopic transcripts with accompanying DNA methylation analysis. (A) Quantitative RT-PCRs were performed for both the retrotransposon-driven ectopic transcripts and the endogenous transcripts for the genes Bcat2, Eps8R1, and Ogdh in brain, kidney, and lung of three 5-mo-old littermates. The bands shown are the qRT-PCR products from the brain of the three mice (#1, 2, and 3) with the internal control β-actin at the top followed by the ectopic and the endogenous transcripts of the three loci as mentioned on the left. The exon structure of each gene is shown at the top of the gel bands with the location of the primers used for amplifying the ectopic and the endogenous transcripts marked by the red and green arrows, respectively. The exon 1 of the ectopic and the endogenous transcripts are indicated by the orange and blue boxes, respectively. (B) The fold differences of the ectopic transcripts with respect to the control β-actin are shown for the gene Eps8R1 in the three different tissues and three different animals (Ec #1, Ec #2, and Ec #3) as described before. (C) The COBRA data (using the restriction enzyme HpyCH4IV) of the three different tissues and the bisulfite sequencing data of the brain Eps8R1 ectopic transcript in the three different animals. In the electrophoresis gel picture, the bands expected after digestion in case of methylated and unmethylated CpGs are indicated by M and U, respectively. The bubble charts depict the sequencing results and the percentages indicate the percentage methylation in the same way as described in Figure 4. The red boxes encompasses the CpG sites that are within the retrotransposon while the numbers (#1, 2, and 3) refers to corresponding individual animals in both the pictures.

Given the observations described above, we further tested the presence of potential inter-individual variations associated with the Eps8R1 locus. For each tissue, the expression level of the LTR-driven transcript was first normalized with that of β-actin, and later compared between the normalized levels of the three individual mice. As shown in Figure 8B, the expression levels of the LTR-driven transcript were relatively higher in Mouse #1 and 3 than in Mouse #2 for all three tissues examined. We also analyzed the DNA methylation levels of the LTR using the DNA isolated from the three adult mice (Fig. 8C). As depicted by the COBRA of all three tissues and the bisulfite sequencing of the brain DNA from all three animals, the DNA methylation levels of the LTR were indeed much lower in Mouse #1 and 3 than Mouse #2 in all three tissues, which is consistent with the higher expression levels observed in Mouse #1 and 3. Such an observation might indicate that methylation might have been set prior to the divergence of the three germ layers from which the three tissues were derived from. Any relation of the sex of the mouse with the methylation status can also be ruled out because Mouse #1 and 2 are males and Mouse #3 is female. Overall, the results described above clearly demonstrate that the identified retrotransposons have inter-individual variations in terms of their DNA methylation levels and, further, that some of these variations in DNA methylation levels have concurrent outcomes on the transcription levels of the nearby endogenous genes.

Discussion

In the current study, we have identified a set of potential metastable epialleles (MEs) that are derived from retrotransposons using a series of bioinformatic approaches. We also analyzed a subset of these potential metastable epialleles with DNA methylation and expression analyses. According to the results, some of the analyzed candidates are most likely epialleles with functional consequences on the nearby genes. This further implicates that the mammalian genomes may contain a large number of uncharacterized metastable epialleles. Also, given the sheer number of retrotransposons in the mammalian genome, these elements may be a major source of inter-individual epigenetic variations resulting in more subtle phenotypic or physiological variations.

The bioinformatic approach used in this study has been successful for finding many epialleles. Two criteria are thought to be instrumental for this success. First, we sought to identify retrotransposons (LINEs and LTRs only) that are located very close to the 5′-end of mRNAs. This approach revealed that about 2.5% of the entire transcriptome (5,679 out of 228,765) is likely initiated or influenced by these retrotransposons (Fig. 1). At the beginning, this estimate was somewhat surprising that retrotransposons could affect such a large number of the transcriptome, but this fact has been independently confirmed by other studies.²⁴ Thus, mammalian retrotransposons may have reshaped not only the landscape of genomes but also the functions of individual genes.^25-28 Despite this surprising estimate, however, our follow-up DNA methylation analyses suggest that many in this group of retrotransposons might be completely methylated in normal tissues without any inter-individual variations (Rgs9 and Cog6 in Fig. 3). This might reflect the fact that this group of retrotransposons likely triggers RNAP II transcription only in very unusual cell populations, such as cancer cells. This observation is consistent with the fact that a large fraction of the transcriptome is indeed derived from these types of cells. Second, as a measure to correct this bias in the transcriptome, we decided to utilize active histone modification (H3K4me3) profiles derived from ES and other cells, which likely have a significant impact on the DNA methylation levels in normal somatic tissues. A large number of retrotransposons (4,226) are indeed found to be associated with active histone marks. Furthermore, a minor fraction (143) of this group was already proven to function as promoters for RNAP II transcription based on their fused exon structures between the 5′-side exons derived from retrotransposons and the 3′-side exons from endogenous genes. Interestingly, some of the retrotransposons that have been tested in this small set of 143 turns out to be epialleles with variable levels of DNA methylation among littermates of an inbred mouse (Figs. 3 and 4). Surprisingly, though, it was shown in a recent study that for the viable yellow agouti (A^vy) mice, groups of yellow and pseudoagouti mice did not exhibit any variations of the H3K4me3 mark in the LTR of the A^vy metastable epiallele in spite of showing variable DNA methylation.²⁹ It is also important to note that some of the retrotransposons that have been proven to function as promoters for RNAP II transcription might not have associated active histone marks. Nonetheless, combining the two criteria, filtering retrotransposons based on their close proximity to the 5′-ends of mRNA and their association with active histone marks, turns out to be a good method to predict epialleles with a high rate of success.

The mechanistic basis by which a subset of retrotransposons escapes DNA methylation is still poorly understood and, furthermore, the reasons why the levels of this DNA methylation are variable among different individuals are also unknown. However, given the large number of potential epialleles identified in this study, we surmise that these unusual epigenetic features may be closely associated with the three factors described below. First, the location of a given retrotransposon appears to be important based on the fact that a large number of epialleles are located closely to the promoter regions of endogenous genes. An opportunistic insertion into a promoter region likely protects the inserted retrotransposon from DNA methylation. Second, this opportunistic position may also be accountable for the variable levels of DNA methylation observed from epialleles. The vast majority of retrotransposons are destined to be repressed by DNA methylation in the mammalian genomes.^25-28 This general repressive trend might be in direct conflict with the unusual but opportunistic position of promoter proximity, resulting in an arms race between DNA methylation vs. protection from methylation. This opposing pressure might be a reason for the variable levels of DNA methylation that has been observed in some of the retrotransposon-derived epialleles. Third, the variable levels of DNA methylation observed in retrotransposons may also be influenced by the transcriptional programs embedded within their associated endogenous genes. Among several candidate MEs (Figs. 3 and 4), the retrotransposons that are associated with tissue- and stage-specific genes (Eps8R1 and Cdk5rap1) tend to show greater degrees of variations in DNA methylation levels among individual mice. Dynamic transcriptional programs in these tissue- and stage-specific genes during development might provide the associated retrotransposons more opportunities to accumulate DNA methylation variations than the other type of genes with static transcriptional programs. Overall, the position of retrotransposons and the transcriptional patterns of the associated genes are thought to be major contributing factors to the metastable nature of DNA methylation associated with epialleles.

It is currently unknown to what extent epigenetic variations contribute to phenotypic variations that are often observed among human populations. Many recent studies, however, suggest that a large fraction of human genetic disorders, in particular disorders with low penetrance, might be accountable for by the epigenetic variations.^3-6 This is particularly likely since traditional genetic studies have focused mainly on genetic variations but not on epigenetic variations. In a similar line, it is equally important to note that the candidate and the confirmed epialleles identified in the current study are derived from retrotransposons, which tend to be ignored by traditional genetic studies as junk DNA with no functional relevance to human health. Nevertheless, the results from the current study highlight the possibility that retrotransposons may have a much greater impact than previously thought on the epigenetic setting of mammalian genomes. In this regard, it is prudent to note that about 4,000 retrotransposons have been identified from the mouse genome, based on their unusual association with active histone marks (Fig. 2). A good number of retrotransposons in this group might be epigenetically unstable and likely contribute to epigenetic variations in the mouse genome. We also predict that a similar number of retrotransposons may exist as epialleles in the human genome since all the mammalian genomes have similar genomic makeup in terms of repeat and gene content. It will be very interesting to test whether the majority of these elements are indeed epigenetically metastable. If that is the case, this group of retrotransposons should provide a large set of epigenetically useful DNA elements, some of which might already explain potential phenotypic variations based on their chromosomal linkage to endogenous genes with disease connections. At the same time, this group of retrotransposons might also serve as epigenetic biomarkers for the diagnosis and/or prognosis of adult onset diseases, such as diabetes and cancers, since the epigenetic status of these DNA elements is prone to be affected by environmental exposure during development.³⁰

Materials and Methods

Data set construction

Mouse (mm9) mRNA data set (track: ‘Mouse mRNAs’; table: ‘all_mrna’) and repeat element data set (track: ‘RepeatMasker’; table: ‘rmsk’) were downloaded from Table Browser of the UCSC Genome Bioinformatics website (http://genome.ucsc.edu/cgi-bin/hgTables?org=Mouse&db=mm9). The mRNA data set contains coordinates of the pre-spliced mRNA only. Genome-wide histone lysine 4 trimethylation (H3K4me3) ChIP-Seq data set of adult mouse liver was downloaded from the website of Canada's Michael Smith Genome Sciences Centre (http://www.bcgsc.ca/data/histone-modification/histone-modification-data).¹⁵ The data set was already available as ChIP-Seq peaks which were generated by the tool FindPeaks 2.0.³¹ The H3K4me3 ChIP-Seq data sets for the V6.5 embryonic stem (ES) cell, embryonic day 13.5 mouse embryonic fibroblast (MEF) cell and ES-derived neural precursor cell (NP) were downloaded from the GEO data repository (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE12241).¹⁴ Since the authors for these ChIP-Seq data sets supplied the peaks that were called using a different tool than the one used for the liver ChIP-Seq data set, we regenerated the peaks for these cell lines using the same Findpeaks tool but using a newer version (FindPeaks 4.0) that is available now. All the peak data sets were subsequently converted to the mm9 version from their original mm8 version using the Lift-Over tool of the Galaxy genome analysis website (http://main.g2.bx.psu.edu/). Stretches of 100 bp were then added to both ends of each peak coordinates of all the ChIP-Seq data sets.

Data set intersections and computational analyses

In the very first step of this bioinformatic approach, the mRNA and the repeat element data sets were uploaded to a MySQL database in order to run a series of coordinate intersections and filtering steps to identify the LTRs and LINEs that intersects with the transcription start sites (TSS) of mRNAs. The identified intersection result must also fit into the following criteria: 1) the TSS of the mRNA is within 250 bp downstream and 50 bp upstream of the 5′-end of the retrotransposon; 2) the retrotransposon must be at least 50 bp in length and; 3) the retrotransposon must have at least 3 CpG sites in its sequence. In the cases where multiple mRNAs transcribing in the same orientation could have been mapped to a particular retrotransposon, only the mRNA that had its TSS closest to the retrotransposon 5′-end and also gave the longest transcription product was kept. For the retrotransposons that had two mRNA transcribing in opposite orientation, only one of the mRNAs was retained after a manual inspection of the surrounding genomic features. A full schematic representation for the preparation of this retro/mRNA pair data set has been shown in Figure S1. For the intersection with histone modification data, the retro/mRNA pair data set and all the H3K4me3 ChIP-Seq peak data sets were uploaded to the Galaxy genome analysis website (http://main.g2.bx.psu.edu/) and intersected using the ‘Intersect’ tool of the ‘Operate on Genomic Intervals’ section of the website. For the analyses of the pool of retrotransposons in which the retrotransposons did not intersect with the mRNA TSS and/or histone modification data sets, the LTR and LINE pool was created prior to those intersections from our constructed MySQL repeat element database using the following criteria: i) the retrotransposon must be at least 50 bp in length and ii) the retrotransposon must have at least 3 CpG sites in its sequence. These retrotransposon data sets were also intersected with the histone modification data sets using Galaxy in the same way mentioned earlier. In the analyses of the positions of the retrotransposons relative to the UCSC Known Gene data set, the retrotransposons were considered to be in the promoter regions if their 5′-ends were located within 10,000 bp upstream of the TSS of the gene and the 3′-end of the Exon 1 of the gene.

Isolation and purification of sperm, oocyte and blastocyst-stage embryos

Mature sperms were isolated from the epididymis of adult C57BL/6J mice by following the ‘swim up’ method.³² Briefly, freshly dissected and chopped epididymis was incubated in sterile PBS at 37°C for 30 min to allow the sperms to swim up to the top of the container. Following the incubation, PBS from the upper part of the container was transferred to a new container and centrifuged at 600 rpm for 5 min. The sperms were again allowed to swim up by having an incubation period at 37°C for 30 min. Following this round of incubation, the PBS in the upper part of the container was extracted and a small volume of it was observed under the microscope to check for purity (less somatic cells). Successive rounds of centrifugation, incubation and transfer were done until sperms of high purity (devoid of almost any somatic cells) were obtained. Finally the sperms were concentrated by centrifugation and resuspended in PBS for DNA methylation analyses.

For the isolation of mature oocytes in the MII phase, we followed the widely used superovulation protocol.^33,34 In brief, female C57BL/6J mice were first injected subcutaneously with 5 IU of the Pregnant Mare’s Serum (PMS) (Cat. G4877, Sigma-Aldrich), and the same mice were injected again 48 h after the initial injection with 5 IU of the human Chorionic Gonadotropin (hCG) hormone (Cat. C1063, Sigma-Aldrich). These mice were sacrificed 12 h after the second injection, and mature oocytes were isolated from the swollen ampullae of the oviducts. The isolated eggs were incubated in hyaluronidase solution (Cat. H3506, Sigma-Aldrich) for several minutes to separate them from the cumulus cells, and subsequently washed three additional times to remove potential somatic tissue contamination. Finally, the oocytes were used to isolate DNA.

The blastocyst-stage embryos were also isolated using the same ovulation protocol as above. After the second injection, however, the female mice were paired with male littermates (both were of C57BL/6J strain). The female mice were then sacrificed after 3 d following the breeding setup, and the blastocysts were isolated from their uterus by flushing with PBS.³⁴ The isolated blastocysts were further examined under the microscope to estimate their developmental stages and purity, and subsequently washed three additional times to remove potential somatic tissue contamination. The blastocysts were then used for DNA isolation.

COBRA (COmbined Bisulfite Restriction Analysis) and bisulfite sequencing

The methylation levels of the CpG sites of a specific locus were analyzed using two widely used DNA methylation analyses, COBRA (COmbined Bisulfite Restriction Analysis) and bisulfite sequencing analysis. For the purpose of isolation and purification of genomic DNA from the six different adult tissues (brain, tail, liver, lung, kidney, and spleen), sperm, oocyte, and blastocyst-stage embryos homozygous C57BL/6J mice obtained from the Jackson Laboratory (http://www.jax.org/) were used. All experiments were performed in accordance with the National Institutes of Health guidelines for care and use of animals. For each type of DNA sample, 2 µg of the genomic DNA was modified using the bisulfite conversion reaction according to the manufacturer’s protocol (EZ DNA methylation^TM kit, Zymo Research). The converted DNA (1 μl) was used as template for PCR reactions (Maxime PCR Premix Kit, Intron Biotech) using specific primer sets that lacked any CpG dinucleotides in the original pre-converted sequence and were designed to specifically amplify the bisulfite-converted (C-to-T converted) DNAs. Information regarding the primer sequences and detailed PCR conditions for each tested loci has been provided in Supplemental Data Set 2. The amplified DNAs were further used for the COBRA and bisulfite sequencing analysis.

For the COBRA, the amplified DNAs were digested using appropriate restriction enzymes (New England Biolabs) according to the manufacturer’s protocol in order to measure the methylation levels of particular CpG sites of the specific loci. All the genomic coordinates and restriction enzymes for the regions tested are available in Supplemental Data Set 2.

For the bisulfite sequencing analysis, the amplified DNAs were separated by agarose gel electrophoresis and then purified by MEGA-spin^TM Agarose Gel Extraction Kit (Intron Biotech). After cloning the purified DNAs into pGEM-T Easy Vector Systems (Promega), the cloned DNAs were isolated by using the DNA-spin^TM Plasmid DNA Purification Kit (Intron Biotech) and sequenced with the ABI 3130XL DNA Sequencer (AME Bioscience). The sequences were analyzed and aligned using the software BiQ Analyzer v2.00 which also ensured that 1) clones with identical methylation patterns were in fact different clones by comparing the positions of the unconverted C’s in the clones; 2) each bisulfite sequence has a sequence identity of at least 80% with the genomic sequence and; 3) the bisulfite conversion rate of each sequence exceeds a minimum of 90%.³⁵

RT-PCR and quantitative real time PCR

Total RNA was isolated from the brain, kidney and lung of three 5-mo-old littermates (Mouse #1 and 2: male; Mouse #3: female) using the Trizol (Invitrogen) reagent as per the manufacturer’s protocol. The isolated RNA was reverse-transcribed using the M-MLV Reverse Transcriptase (Invitrogen) and other compatible reagents from the manufacturer. PCR amplifications of the cDNA were performed with specific gene primer sets (Supplemental Data Set 2) using the Maxime PCR premix kit (Intron Biotech). Quantitative real-time PCR was also performed with the iQ SYBR Green Supermix (Bio-Rad) using the iCycler iQTM multicolor real-time detection system (Bio-Rad). All qRT-PCR reactions were performed for 40 cycles under the standard PCR conditions. β-actin was used as the internal control. The results were analyzed based on the threshold (C_t) value. Each reaction was checked for the presence of the desired PCR product by gel electrophoresis and any product for which the C_t value exceeded the value of 31 was discarded. The experiments were performed in triplicate (for each transcript) and for each replicate a ΔC_t value was first calculated by subtracting the C_t value of that given replicate from the average C_t value of the internal control (β-actin). Next, the fold difference was determined for each replicate by raising 2 to the ΔC_t powers and then the average of those 3-fold differences were used for that particular transcript. The standard deviations were calculated from the fold differences of the replicates.