Differential Methylation of Xite and CTCF Sites in Tsix Mirrors the Pattern of X-Inactivation Choice in Mice

Abstract

During mammalian dosage compensation, one of two X-chromosomes in female cells is inactivated. The choice of which X is silenced can be imprinted or stochastic. Although genetic loci influencing the choice decision have been identified, the primary marks for imprinting and random selection remain undefined. Here, we examined the role of DNA methylation, a mechanism known to regulate imprinting in autosomal loci, and sought to determine whether differential methylation on the two Xs might predict their fates. To identify differentially methylated domains (DMDs) at the X-inactivation center, we used bisulfite sequencing and methylation-sensitive restriction enzyme analyses. We found DMDs in Tsix and Xite, two genes previously shown to influence choice. Interestingly, the DMDs in Tsix lie within CTCF binding sites. Allelic methylation differences occur in gametes and are erased in embryonic stem cells carrying two active Xs. Because the pattern of DNA methylation mirrors events of X-inactivation, we propose that differential methylation of DMDs in Tsix and Xite constitute a primary mark for epigenetic regulation. The discovery of DMDs in CTCF sites draws further parallels between X-inactivation and autosomal imprinting.

In mammals, one of two X-chromosomes is inactivated in the developing female embryo to ensure that XX and XY individuals have equal sex chromosome dosage (22). The questions of how a chromosome-counting mechanism differentiates the XX and the XY cell and how one X-chromosome is selected for silencing remain two key epigenetic problems in the X-chromosome inactivation (XCI) field (reviewed in reference 2). With regard to X-chromosome selection, silencing can take place randomly or in an imprinted fashion. In mice, imprinted XCI silences the paternal X (X^P) through preimplantation development and persists in the extraembryonic lineages (11, 30). In the epiblast lineage that gives rise to the embryo proper, the X^P is reactivated to enable random selection of one X-chromosome for silencing just after uterine implantation (24).

Three loci associated with noncoding RNA have been implicated in random X-chromosome choice. Xist, the gene that initiates silencing along the X (5), has been associated with skewed XCI ratios when specific mutations are incurred (25, 27, 33). Xist's antisense counterpart, Tsix, blocks Xist RNA accumulation (19) and is required for random XCI choice. Tsix deletions lead to biased silencing of the mutated X (19, 36). Required sequences lie within its CpG-rich 5′ terminus, a region that includes multiple binding sites for CTCF (6), a chromatin insulator known to regulate autosomal imprinting at the H19/Igf2 locus (40). XCI choice is also affected by Xite, a modulator of Tsix and a candidate locus for the Xce modifier (29). Xite deletions reduce Tsix expression in cis, thereby lowering the probability that the linked X will become the active X (Xa). Although Xist, Tsix, and Xite each play a part in random choice, the molecular underpinnings remain unknown.

Superficially, imprinted XCI appears different with regard to mechanism, since it leads to stereotyped inactivation of the X^P. One recent work has proposed that imprinted XCI consists of two phases: an Xist-independent phase that initiates silencing in the paternal germ line and is then propagated into the zygote, followed by a presumptively Xist-dependent phase that takes over the task of maintaining X^P silence in the early embryo (11, 12). While the Xist-independent phase is posited to find its basis in meiotic sex chromosome inactivation, the presumptively Xist-dependent mechanism is thought to require the more conventional mechanism of gametic imprinting, occurring specifically at the X-inactivation center (Xic). Such an Xic imprint would be acquired some time during gametogenesis. A priori, the gametic Xic mark could be of either parental origin. If maternal, the imprint would promote the activity of the X^M (maternal X); if paternal, the imprint would promote silencing of the X^P. Strong arguments have been made for a maternal imprint (8, 31). Nuclear transplantation experiments indicate that embryos carrying two X^M resist silencing despite having two Xs, while those carrying two X^P partially overcome imprinting to silence one X^P. This suggests that the X^M is rigidly imprinted, whereas the X^P is loosely imprinted. While the X^M imprint is acquired during the oocyte growth phase (38), the exact nature of the maternal and possible paternal imprints is uncertain. Recent evidence suggests an imprinting center at the 5′ end of Tsix, since its deletion on the X^M causes ectopic Xist expression in the placenta (15, 36).

Mounting evidence reveals that random choice and Xic imprinting may actually have a common mechanism. Most notably, Tsix and Xist mutations that affect random XCI also affect imprinted XCI (15, 19, 25, 36). It has been postulated that random XCI descended directly from imprinted XCI by passing the reigns of Xic control from parent to zygote (9, 16, 23). Genomic imprinting may have in fact originated on the X-chromosome to deal with dosage compensation (16). This idea suggests potential mechanistic parallels between XCI and genomic imprinting in general. At autosomally imprinted loci, it is known that DNA methyltransferases (14, 21) and differential methylation at CpG dinucleotides (reviewed in references 3 and 37) play a critical role in regulation. Furthermore, at the H19/Igf2 locus, differentially methylated domains (DMDs) occur within binding sites for CTCF (40), a chromatin insulator that also binds Tsix (6).

Given these striking coincidences, we now sought to determine whether DNA methylation differences also underlie Xic imprinting and choice. This idea makes very clear predictions. If true, there would be obvious methylation differences between sperm and oocyte. In anticipation of dynamic changes in XCI status during early zygotic development, these differences should remain through preimplantation development (X^P inactive), possibly persist in the placenta as well (X^P remains inactive), and then be erased in epiblast-derived cells (two active Xs) and zygotically reestablished in somatic lineages (one randomly chosen inactive X). To examine epiblast-derived cells, we use a mouse embryonic stem (ES) cell model that faithfully carries out random XCI during cell differentiation in culture.

Several CpG islands occur in the murine Xist-Tsix-Xite region. Analyses of the CpG cluster in the Xist promoter have produced variable results, with some studies demonstrating differential methylation (1, 28, 41) and others finding none (26; H. Sasaki, personal communication). Similarly, the DXPas34 cluster downstream of the Tsix start site was originally thought to display differential methylation (7) but was later refuted (34). Thus, the role of differential methylation in XCI has remained ambiguous. Here, we focus on Tsix and Xite, two loci that bias X-chromosome selection. We find strong DMDs whose patterns of methylation correlate with the ongoing events of XCI. Interestingly, the DMDs partially overlap with CTCF binding sites and further highlight close molecular parallels between XCI and autosomal imprinting.

MATERIALS AND METHODS

Collection and preparation of DNA from cells, embryos, trophoblast-derived stem cells (TS cells), and cell lines.

Sperm DNA was collected from C57BL/6J male mice (10). Oocytes were collected from 3- to 4-week-old superovulated B6SJLF1/J female mice, washed three times in Tris-EDTA (TE), and pooled in 20 μl of TE, and DNA was extracted (13). Fertilized eggs were recovered at 0.5 days postcoitus from the corresponding matings and cultured in vitro in KSOM (Specialty Media) to the appropriate stages. For allele-specific methylation analysis, embryos were derived from crosses between female Mus musculus castaneus (CAST/Ei) and male M. m. musculus (C57BL/6). A total of 25 to 30 morula-stage embryos were pooled for each methylation-sensitive restriction analysis (MSRA) experiment. Derivation and culture of TS cells were described previously (11). The ES lines 16.7, J1, male (CG7) and female (3F1) ΔCpG and ΔXite (ΔL-C2) have been described (18, 19, 29).

Southern blot analysis.

A total of 20 μg of genomic DNA was digested overnight with enzymes outside the region of interest and simultaneously digested with a methylation-sensitive enzyme in the region of interest. The DNA was then Southern blotted, and the relevant fragments detected by probes doubly labeled with [α-³²P]dCTP and [α-³²P]dATP.

MSRA.

A total of 500 ng of DNA (or DNA from the average of 200 to 500 oocytes) was doubly digested overnight with a methylation-insensitive enzyme that cuts outside of region to be amplified (PvuII or EcoRI) and with a methylation-sensitive enzyme of interest. DNA was amplified during 35 cycles of amplification at 95°C for 15 s, 56°C for 45 s, and 72°C for 60 s and analyzed on agarose gels. Real-time PCR was performed on a Bio-Rad iCycler machine in triplicate in SYBR green buffer (Stratagene) or with a TaqMan sequence-specific probe. The template was amplified for 50 cycles of 95°C for 15 s, 56°C for 45 s, and 67°C for 60 s. The fluorescence data were collected during the extension phase. The fluorescence threshold for determination of Cτ is the average standard deviation of the background × 10.00. The cycle number at the threshold is the “Cτ.” Samples digested with methylation-sensitive enzymes will have Cτ values which correspond to the undigested samples if methylated and Cτ values that diverge from undigested samples if unmethylated. Application of the formula, ΔCτ = undigested Cτ − digested Cτ, normalizes the Cτ value of each sample to that of the undigested control (the cycle number at the threshold is the Cτ.) In all MSRA assays, we interpreted differences of less than one cycle as essentially equal amplification (because this is within the standard error of the assay), suggesting that the template is methylated. Differences of between one and two cycles are interpreted as partially methylated or equivalent to 25 to 50% of template being methylated. Differences of greater than two cycles are unmethylated, indicating that <25% of template are methylated. Site A sequence was amplified with primer pair CC3-3B/CC3-3C; the real-time PCR analysis was done by using either the TaqMan probe 3B3C or SYBR green buffer. Site C sequence was amplified with the primer pair CC3-1C/CC3-1B; the real-time PCR analysis was done by using either the TaqMan probe 1B1C or SYBR green buffer. DHS6 was amplified with the primer pair NG-1/NG-2, DHS1+2 was amplified with the primer pairs Apa-5/Apa-5.1 and Apa-6/Apa-6.1, DHS3 was amplified with the primer pairs Apa-3/3.1 and Apa-4/4.1, and DHS4 was amplified with the primer pairs Apa-1/Apa-1.1 and Apa-2/Apa-2.1. Primers sequences are indicated in the Table 1.

TABLE 1.

Sequences of primers and TaqMan probes used in MSRA and bisulfite analysis

Primer or probe	Sequence (5′-3′)
MRSA primers
CC3-1A	GTACCTGCAAGCGCTACACA
CC3-1B	CGGGAACGTGGCATGTATGT
CC3-1C	AATGCCTGCGTAGTCCCGAA
CC3-3A	GGCGGGCAAGGCAAAACAAT
CC3-3B	GGAGCCTAAACCTGTCTGTC
CC3-3C	GCTACCTGTGTCTCTGTATC
CC3-3R	CGAATTTGACTGCATATGTAC
CC3-3S (DMR1-for)	CCTCAAACTCGCAAAGCTCTG
DMR1-rev	GTCTGCCTACTAACACAGGT
CC3-3T	GAGAATTCATTTCCCTGCCCT
NG-1	CGCACGCACATATCTGTACGCA
NG-2	GCGTTTTACCTTTTTAGGACAGAG
Apa-1	GCATTCCATGAGTAAATCCCAC
Apa-1.1	AGCCCACATCCCTGATGATC
Apa-2	GAGACAACTGGAAATGGGA
Apa-2.1	GGTGTAGCTTAGTTTCATCAGCTTC
Apa-3	CTCCTCAGCCTTCCAGATTT
Apa-3.1	CAATCCTTGAAGCCAGTGAAC
Apa-4	AACGGTAGACTGGTTTTGACC
Apa-4.1	AGTTGCCATGGGGAGAGGGGA
Apa-5	CGCTGGATATAGCCTGGATTA
Apa-5.1	CACCACACCAGGCCAAAGCC
Apa-6	GTGCTCAGTTCTAAATGGAGC
Apa-6.1	TATCAGAGGGCCACACCATTAAC
TaqMan probes
1B1C	(6FAM)-TGGCTCCAGGACCCAGCAGAC
3B3C	(VIC)-TAGCTCAAGAGGCTAGACCTCTC-(MGB)
3S3T	(VIC)-CTAGTAATTGCCTTCCTCTATGTCT-(TAMRA)
Allele-specific PCR probes
KHp4	TAACCACCTGTAAGGGACAG
KHp388	GTCTTCTGATTGTAACATGCTG
Bisulfite sequencing primers
bCC3-1F	TATATAAACTATCTTTCCAAATCCCTAAC
bCC3-1G	GTAGAAAGATTTTTAAAATGTTTGTTAGT
bCC3-3F	AAATAAAACCTAAACCTATCTATCTCTT
bCC3-3G	TTTATTAAGTTTTTATTATTTGGGATAAG
bCC3-3H	TAACTACATATATACATAAAACCTTTAAATAA
bCC3-3I	AGTTTTGGAAGTTGAGATTTAAAATTTTGGTA
bCC3-3L	AGAAAAGGAATTATGAGATAGGTTAAGGTA
bCC3-3N	CTTCTATAAATCTAAAAAATTCATTTCCCT
bCC3-3O	GTTTAGAAAGGAGGGAAATTAATAATTAGT
bCC3-3U	GTTTTATTTTTGAGATAAGGTTTTAGTAAATAAT
bCC3-3V	TAAAAAATCAATATTTATCTACCTACTAACACAAA
bNG-3	TATATTTTAGATTAGATTGGATTTAAATTT
bNG-3.1	CTATACTCAAACTAAAAACTTACAAACTAA
bNG-4	ATATATAAACATAAATAATAAAATCAAATA
bNG-4.1	TTAGTTTGTAAGTTTTTAGTTTGAGTATAGAT
bApa-1	ATTTTTAGAAGAGTAGTTGGTGTTTTTT
bApa-1.1	TTAAAAATTTATAACCCACATCCCTAATAATC
bApa-2	CAAATAATACCTAAAACCTAATAAAATA
bApa-2.1	TTTTGGTATTATGAATTGAGTTAGATAGT
bApa-3	TAAATTTTTATGAATGTTTATTTTTTAATGT
bApa-4	CATCTAATTATAATTAACATAATAAAAATA
bApa-4.1	GTTTATTGGTTTTAAGGATTGAGGAGTTGT
bApa-5	TTATTAAATTTATTTAAATAAATTTTTTAAG
bApa-6	ATATCACACAATTACATACAAACCTCAAAA
bApa-6.1	GATTTTGTAATTTGGGGAATGTTATTTTTTTG

Allele-specific PCR analysis.

Allele-specific PCR was done by digesting the MSRA conventional PCR amplified products with specified restriction enzymes that are sensitive to single nucleotide polymorphisms to distinguish between the parental alleles followed by Southern blot analysis. M. m. castaneus (CAST/Ei) DNA was used as the control maternal sample and M. m. musculus (C57BL/6) DNA was the control paternal sample for allele specific polymorphic digestion to compare with assayed DNAs. DHS6 was amplified with the primer pair NG-1/NG-2 for allele-specific PCR and probed with KHp388. Site C was amplified with the primer pair CC3-1B/CC3-1C and probed with KHp4. All primer and probe sequences are shown in Table 1.

Bisulfite sequencing.

A total of 500 ng of genomic DNA, or DNA from an average of 500 oocytes, was treated with the CpGenome DNA modification kit (Intergen) as directed by the supplier, with the modification of additional denaturation steps at 95°C during the bisulfite incubation. PCR amplification was done on 2 μl of treated sample DNA using primers devoid of CpG dinucleotides which would only amplify converted sequence, as is standard practice. Site A was amplified by two rounds of PCR using either primer pair bCC3-3F/bCC3-3L or primer pair bCC3-3H/bCC3-3L (with similar results). Site C was amplified by two PCR rounds with bCC3-1F/bCC3-1G. DHS6 was amplified by using bNG-3/bNG-3.1 and bNG-4/bNG-4.1, DHS4 was amplified by using bApa-1/bApa-1.1 and bApa-2/bApa-2.1, and DHS3 was amplified by using bApa-4/4.1. Primer sequences are shown in the Table 1. Amplified PCR products were cloned into the pGEM-T vector (Promega Corp.), sequenced by using an Applied Biosystems Taq DyeDeoxy Terminator cycle sequencing kit, and resolved on an ABI 3700 Prism automated sequencer.

RESULTS

Gamete-specific methylation differences in the CTCF binding sites of Tsix.

We first focused on the 5′ end of Tsix wherein lies a CpG island implicated in imprinting and random choice. This CpG island includes the Tsix promoter and CTCF binding sites in and around DXPas34 (34) (Fig. 1A). To search for DMDs, we performed MSRA, whereby the methylation status is assessed by whether genomic DNA is digestible (unmethylated) or not (methylated) with methylation sensitive enzymes. As a preliminary screen, we performed Southern analysis by digestion of gametic, ES cell, and somatic genomic DNAs with up to four different methylation-sensitive enzymes. Oocyte DNA is limiting so female gametes were not included in this initial analysis. DNA was digested with enzymes whose sites fall within the region of interest and with a flanking methylation-insensitive enzyme, PvuII (to reduce fragment sizes for electrophoresis).

FIG. 1.

Southern blot analysis of the Tsix region reveals potential DMDs. (A) Map of the Tsix and Xite regions. Oval region, CTCF motifs. DHS, DNase I-hypersensitive sites. Inset: H, HpaII; A, AciI. (B) Genomic Southern blot analysis of the 5′ end of Tsix using the combination of enzymes indicated. Probe, 1.1-kb AgeI-HincII fragment at the 5′ end of Tsix; Sp, sperm; M, male liver; F, female liver; filled circles, methylated bands; open circles, unmethylated bands. (C) Genomic Southern analysis of undifferentiated (day 0) ES cells and differentiated (day 12) EB cells using the 1.1-kb AgeI-HincII probe. Filled circles, methylated bands; open circles, unmethylated bands. Asterisks, doublet lower bands associated with the female ES and EB DNAs resulted from DXpas34 repeat length polymorphisms between the 129-derived and Mus castaneus-derived Xs in 16.7 ES cells. (D) Genomic Southern analysis of DXPas34 using the same probe. ES cells are female. The restriction sites (HpaII and AciI) are those indicated in the panel A inset. Asterisks indicate sites in ES cells that have become hypermethylated and have shifted upward and lost intensity or disappeared in EB cells.

The results revealed a variegated pattern of DNA methylation across the region (Fig. 1B). No tissue-specific differences were observed at NarI and AgeI: NarI appeared uniformly methylated, and AgeI was undermethylated in the three tissues examined. In contrast, HgaI and SalI revealed potential differences: at HgaI, the sperm and male liver were hypermethylated, whereas female liver appeared to contain methylated and unmethylated epialleles. At SalI, sperm DNA was unmethylated, while the male and female liver DNA looked similar in being partially methylated. HgaI and SalI also exhibited differences in an ES model (Fig. 1C), with undifferentiated male and female ES cells being undermethylated and differentiated male and female EBs being hypermethylated. (Note that the doublet lower bands associated with the female ES and EB DNAs resulted from DXpas34 repeat length polymorphisms between the 129-derived and M. m. castaneus-derived Xs in 16.7 ES cells.) Intriguingly, SalI and HgaI coincide with elements previously shown to bind CTCF in vitro in a methylation-sensitive manner (sites A and B) (6). Furthermore, Southern analysis of DXPas34 suggested that AciI sites might be differentially methylated, since sites present in ES cells carrying two Xa become hypermethylated upon cell differentiation and initiation of XCI (asterisks, Fig. 1D; bands present in ES cells shifted upward and lost intensity or disappeared in EB). Because several CTCF motifs in Tsix have AciI sites and HgaI and SalI are within CTCF sites as well, these results led us to ask whether CTCF sites might constitute regulatory DMDs.

To test this possibility, we compared the methylation status of CTCF sites of oocytes to that of sperm (Fig. 2). Because oocyte DNA was limiting, we modified the MSRA to include a PCR step after digestion with a methylation-sensitive enzyme. If the site is methylated, the DNA resists digestion and amplifies. If unmethylated, digestion would render the DNA unamplifiable. We designed primers to flank site A and another CTCF site previously designated site C (6) (Fig. 1A). Other sites, such as site B, could not be tested because they lie within the DXPas34 repeat, precluding design of locus-specific primers.

FIG. 2.

CTCF sites A and C of Tsix are differentially methylated in gametes. (A) MSRA at sites A and C using conventional PCR. (B) Representative MSRA of at site A using quantitative real-time PCR analysis. Each curve represents the amplification of a single sample. Horizontal orange lines, fluorescence threshold. HgaI-digested (red arrows) and undigested (black arrows) samples were run in triplicates. (C) Bisulfite analysis of site A. Filled circle, methylated; open circle, unmethylated. n, number of strands exhibiting each methylation pattern.

At both sites A and C, PCR analysis showed that oocytes amplified poorly compared to the undigested control, whereas somatic cells and sperm amplified robustly (Fig. 2A). The results of conventional PCR were confirmed by quantitative real-time PCR analysis (Fig. 2B). Sperm DNA remained resistant to cutting at both sites. These findings demonstrated a gamete-specific methylation difference in two assayable CTCF sites, with the maternal allele being undermethylated and the paternal allele being hypermethylated.

To determine how many Cs are differentially methylated, we carried out bisulfite sequencing analysis, whereby genomic DNA is treated with bisulfite which converts C's to U's (read as T's) when the C is unmethylated. After PCR and sequencing with locus-specific primers, the methylation pattern on the original DNA strand can be inferred. Bisulfite analysis of site A revealed that the single CpG dinucleotide coinciding with the HgaI site is differentially methylated (Fig. 2C). Whereas sperm was consistently methylated, oocytes were undermethylated. Interestingly, several asymmetric C's within site A showed differential methylation (data not shown). Although CTCF binding can be modulated by non-CpG methylation in vitro (6), the significance of non-CpG methylation remains unclear, since it was predominantly observed in oocytes and not in other cell types (data not shown).

Dynamic methylation changes during cell differentiation.

We next examined methylation in undifferentiated female ES cells, which would have reactivated the X^P and therefore would carry two active Xs, and differentiated EB derivatives, which randomly reinactivate one of two Xs. Bisulfite sequencing showed that site A was undermethylated in ES cells (Fig. 3A). The minority of undifferentiated clones showing methylation may be due to partial differentiation, as ES cells are prone to differentiate in culture. A similar result was observed at site C (Fig. 3B). MSRA PCR confirmed that ES cells were undermethylated at sites A and C (Fig. 3C and D).

FIG. 3.

Developmentally regulated methylation profiles at CTCF sites A and C. (A) Bisulfite analysis at site A. The methylation state of the single CpG is shown. (B) Bisulfite analysis at site C. CpG residues examined are numbered across; the AciI recognition site is noted. Filled circle, methylated; open circle, unmethylated; n, number of strands exhibiting each methylation pattern. (C) MSRA using real-time PCR of various cell lines and tissues. Filled circle, methylated; half-filled circle, partially methylated; open circle, unmethylated (see the text). ES cells are undifferentiated day 0 (d0) cultures; EB cells are differentiated cultures at days 7 to 12, a time when a vast majority of cells have undergone XCI (19). (D) Representative real-time PCR runs of the MSRA performed in triplicates. Black arrows, undigested samples; red arrows, digested samples. (E) Allele-specific MSRA at site C in ES and EB cells. Samples were digested (+) or not digested (−) with AciI before PCR amplification with primers CC3-1B and CC3-1C and probing with end-labeled primer, KHp4.

Differentiation of ES cells for 7 to 12 days into EB cells resulted in de novo DNA methylation at sites A and C. Bisulfite analysis showed various degrees of new methylation at seven CpGs at site C in males and females (Fig. 3A and B). These results were supported by MSRA (Fig. 3C and D). To determine which allele was methylated, we carried out quantitative allele-specific PCR after digestion with methylation-sensitive restriction enzymes. The allele-specific PCR was based on a single-nucleotide polymorphism that creates an AluI restriction site on the castaneus chromosome (Fig. 3E). In the 16.7 female ES line, the 129 X is maternally derived and the castaneus X is paternally derived. The analysis showed that female cells were undermethylated on both maternal and paternal alleles before the onset of XCI (undifferentiated ES cells). Cell differentiation and the onset of XCI then resulted in hypermethylation of either or both alleles. Thus, in epiblast-derived ES cells, methylation is lost from the previously methylated paternal allele (spermatic methylation), and this timing of demethylation correlates with the developmental stage when the XCI imprint is known to be erased. In males, the single X in undifferentiated cells was undermethylated, a finding consistent with its previously unmethylated state in the oocyte (Fig. 3E). We conclude that DNA methylation is therefore a reasonable candidate for the gametic XCI imprint.

We then tested specific embryonic and somatic tissues. Liver DNA exhibited hypermethylation by MSRA (Fig. 3C) and full methylation of all strands tested by bisulfite sequencing (data not shown). Furthermore, MSRA and bisulfite analysis of female extraembryonic yolk sac endoderm at 7.5 days postcoitus showed that site A was hypermethylated and site C was partially methylated (Fig. 3C) and that, in female trophoblast-derived stem cell line (i.e., TS cells), both parental alleles were also methylated (data not shown). In morulae, both alleles were slightly methylated (data not shown). These findings demonstrated that, in both tissues subject to imprinted XCI and those subject to random XCI, both Tsix alleles are at least partially methylated in differentiated cells. However, in extraembryonic tissues, DNA methylation does not correlate completely with the XCI status, a finding consistent with studies showing that epigenetic regulation in extra-embryonic tissues may depend less on DNA methylation and rely more on other chromatin modifications (20, 35, 39).

Taken together, these data showed that DNA methylation at Tsix is differentially established in cells that carry the primary imprints for choice (oocyte, sperm) and lost in cells that have erased the imprint (ES cells). Remethylation occurs upon cell differentiation and becomes biallelic in somatic cells and also in TS cells. The latter result counters the prediction that the Xi would be uniquely methylated. However, it is known that Tsix expression is eventually turned off in cells that have established XCI. Therefore, it is possible that biallelic methylation in differentiated and TS cells is the result of two sequential events: a primary event on the future Xi and a secondary event on the established Xa after the loss of Tsix expression.

The Xite locus contains three DMDs.

We next examined the methylation status of Xite. Xite contains two clusters of CpG-rich sites, coinciding with two intergenic transcription start sites and associated DNase I-hypersensitive sites, DHS1 to DHS6 (Fig. 4A). For MSRA, we designed primer pairs to span the DHS and compared the PCR efficiency of cut versus uncut (control) samples. Several DMDs were uncovered (Fig. 4B). Between DHS1 and DHS2, analysis of AgeI, HpaII, and AciI showed that oocytes were undermethylated, correlating with a single Xa in oocytes. In contrast, sperm was hypermethylated, a finding consistent with a paternal X marked as Xi. The AvaI site at DHS4 and the HpaII, AciI, and HgaI sites at DHS6 all demonstrated a similar pattern of differential methylation. Bisulfite sequencing analysis confirmed that the oocyte DNA is considerably undermethylated relative to sperm DNA (Fig. 4C).

FIG. 4.

Differential methylation at the Xite locus correlates with events of XCI. (A) Map of Xite. Methylation-sensitive restriction sites are indicated above the axis; amplicons are indicated below the axis. Black rectangles, novel DMDs. (B) MSRA using real-time PCR at Xite. The deduced methylation status is shown; open circle, unmethylated; filled circles, methylated; half-filled circles, partial methylation. ES cells are day 0 (d0) cultures, EB cells are from days 7 to 12. ND, no data. (C) Bisulfite analysis at DHS6. CpG residues examined are numbered across; the enzyme recognition sites are noted. Filled circle, methylated; open circle, unmethylated; n, number of strands exhibiting each methylation pattern. (D) Allele-specific MSRA of ES and EB cells at DHS6. Enzyme indicates samples that were digested (+) with either HpaII, AciI, or HgaI or undigested (−) before PCR amplification with primers NG-1 and NG-2 and then probing with end-labeled primer KHp388. (E) Allele-specific MSRA of morula and female TS cells at DHS6 as described in panel D. For the TS cells, the maternal allele is of 129 origin, the paternal allele of castaneus origin. For morulae, the maternal allele is of castaneus origin, and the paternal allele is of 129 origin.

We then examined Xite methylation in ES and EB cells. Undifferentiated pre-XCI ES cells were relatively undermethylated, a finding consistent with there being one Xa in male ES cells and two Xa in female ES cells (Fig. 4B). Differentiated post-XCI EB cells became considerably hypermethylated, with methylation being most pronounced and biallelic in somatic cells (liver) (Fig. 4B and data not shown). Thus, like at the Tsix locus, a secondary methylation event of Xite may occur on both alleles following Tsix repression after XCI (29). To determine which ES and EB alleles were methylated, we carried out quantitative allele-specific PCR after digestion with various methylation-sensitive restriction enzymes (HpaII, AciI, and HgaI; Fig. 4D). This allele-specific PCR was based on a single-nucleotide polymorphism which creates a BstUI restriction site on the 129 X-chromosome. We found that female ES cells were vastly undermethylated on both the maternal and the paternal alleles before the onset of XCI. Cell differentiation and the onset of XCI then resulted in hypermethylation of either or both alleles. Thus, as was the case for Tsix, DNA methylation is lost from the previously methylated paternal Xite allele, and this demethylation occurred around the developmental stage when the XCI imprint is known to be erased. DNA methylation at Xite is therefore also a good candidate for the gametic XCI imprint.

We also examined allelic methylation patterns in TS cells and morulae (Fig. 4E). In TS cells, the HpaII and AciI sites was biallelically methylated, whereas the HgaI site was undermethylated on both alleles. These results showed that DNA methylation does not correlate completely with the XCI status in this extraembryonic tissue, a finding consistent with studies showing that epigenetic regulation in extraembryonic tissues may depend less on DNA methylation and rely more on histone modifications (20, 35, 39). In pooled morulae, while HpaII and AciI appeared to be biallelically methylated, the HgaI site showed residual paternal-specific methylation, indicating that the methylation pattern in sperm is partially retained in the preimplantation embryo.

The DMDs are not affected by mutations in Tsix and Xite.

If DNA methylation were the primary mark for the imprinting and choice decision, it must act genetically upstream of Xite and Tsix. To test this possibility, we examined whether mutations in Tsix and Xite expression affected DNA methylation patterns by MSRA. When the Tsix CpG island, including the major antisense promoter, is deleted in ES cells (Tsix^ΔCpG; Fig. 5A), the pattern of DNA methylation at the Xite locus did not deviate significantly from the wild type (Fig. 5C). Conversely, when a 12-kb region of Xite including both intergenic promoters was deleted in ES cells (Xite^ΔL), Tsix methylation patterns were not altered (Fig. 5B). Bisulfite sequencing analysis also revealed no obvious methylation differences in either mutant cell line (data not shown). Thus, these results are consistent with the model that differential methylation acts genetically upstream of Tsix and Xite expression.

FIG. 5.

Mutant analyses suggest that DNA methylation works genetically upstream to Tsix and Xite. (A) The Tsix^ΔCpG and Xite^ΔL positions are shown. Proposed DMDs are indicated as filled black rectangles. (B) Effects of Xite^ΔL on CTCF sites A and C methylation at Tsix. MSRA analysis using real-time PCR of wild-type and mutant female cells lines at sites A and C. (C) Effects of Tsix^ΔCpG on Xite DMDs. MSRA analysis using real-time PCR of wild-type and mutant cell lines at the Xite DMDs. The deduced methylation status is shown. Female mutants are heterozygous. Open circles, unmethylated; filled circles, methylated.

DISCUSSION

In this study, we have tested whether DNA methylation could serve as a primary mark for Xic imprinting and choice. The results of MRSA and bisulfite analysis revealed DMDs within both Tsix and Xite. Our observations are consistent with the hypothesis in several ways. First, changes in DNA methylation patterns tend to precede XCI and correlate with ongoing events: the marks are differentially imprinted in oocyte and sperm, and the marks are lost in epiblast-derived ES cells, where it is known that the gametic imprint is erased in full. De novo methylation occurs in differentiated cells, in day 7.5 yolk sac endoderm, and in TS cells, with methylation becoming biallelic in these tissues. Biallelic methylation is consistent with biallelic repression of Xite and Tsix in differentiated cells. Second, the DNA methylation marks are not affected by mutations in Tsix and Xite expression. Current understanding of XCI places Xite genetically upstream of Tsix (29), which in turn lies upstream of Xist (19). Therefore, if DNA methylation lay further upstream than all three loci, mutations in the downstream genes would not be expected to affect the DMDs.

Finally, the DMDs are located within two genes—Xite and Tsix—known to be involved in the imprinting and/or choice decisions (15, 19, 29, 36). Whether Xite plays a role in imprinting XCI has not yet been examined. For random XCI, genetic analysis in ES cells suggests that Xite increases the likelihood of the Xa state by prolonging expression of the linked Tsix allele. In turn, Tsix designates the future Xa by blocking Xist RNA accumulation in cis. The observation that the Tsix DMD coincides with CTCF binding sites raises interesting parallels to autosomal imprinting. At H19/Igf2, the differential sensitivity of CTCF to DNA methylation sets up the pattern of parent-specific expression (40). At the Xic, CTCF has also been proposed as a candidate trans-acting factor for regulating allelic choice (6). By analogy to regulation at H19/Igf2, CTCF binding to Tsix could act as either a transcriptional activator for Tsix or a chromatin insulator that blocks Xist from upstream enhancers. In the case of transcriptional activation, DMD methylation could preclude binding of the CTCF activator to Tsix. If CTCF sets up a chromatin insulator, DMD methylation could nullify the insulator and allow upstream enhancers to engage Xist. Thus, gamete-specific methylation patterns in the CTCF sites of Tsix could provide the means for control of imprinted XCI. In the epiblast, demethylation could enable a zygotic mechanism to reset allelic choice in a stochastic fashion. Interestingly, these domains in Xite and Tsix have recently also been implicated in X-chromosome counting (17), so the potential role of DNA methylation in this aspect of XCI would be of future interest.

The XCI phenotype seen in ES cells and mice carrying mutations in Dnmt1, the maintenance methyltransferase, is consistent with a role for DNA methylation in random XCI. Male Dnmt1^−/− ES cells and mice show inappropriate expression of Xist and ectopic Xi formation in some cells (4, 32). In extraembryonic cells, Dnmt1 mutations do not cause a major disruption to imprinted XCI (35). Therefore, if DNA methylation plays a primary role in these cells, other Dnmt genes may be responsible. Ultimately, necessary confirmation of DNA methylation as the primary mark will come with specific mutagenesis of the DMDs in Tsix and Xite.