Dynamic instability of genomic methylation patterns in pluripotent stem cells
© Ooi et al; licensee BioMed Central Ltd. 2010
Received: 15 June 2010
Accepted: 24 September 2010
Published: 24 September 2010
Skip to main content
© Ooi et al; licensee BioMed Central Ltd. 2010
Received: 15 June 2010
Accepted: 24 September 2010
Published: 24 September 2010
Genomic methylation patterns are established during gametogenesis, and perpetuated in somatic cells by faithful maintenance methylation. There have been previous indications that genomic methylation patterns may be less stable in embryonic stem (ES) cells than in differentiated somatic cells, but it is not known whether different mechanisms of de novo and maintenance methylation operate in pluripotent stem cells compared with differentiating somatic cells.
In this paper, we show that ablation of the DNA methyltransferase regulator DNMT3L (DNA methyltransferase 3-like) in mouse ES cells renders them essentially incapable of de novo methylation of newly integrated retroviral DNA. We also show that ES cells lacking DNMT3L lose DNA methylation over time in culture, suggesting that DNA methylation in ES cells is the result of dynamic loss and gain of DNA methylation. We found that wild-type female ES cells lose DNA methylation at a much faster rate than do male ES cells; this defect could not be attributed to sex-specific differences in expression of DNMT3L or of any DNA methyltransferase. We also found that human ES and induced pluripotent stem cell lines showed marked but variable loss of methylation that could not be attributed to sex chromosome constitution or time in culture.
These data indicate that DNA methylation in pluripotent stem cells is much more dynamic and error-prone than is maintenance methylation in differentiated cells. DNA methylation requires DNMT3L in stem cells, but DNMT3L is not expressed in differentiating somatic cells. Error-prone maintenance methylation will introduce unpredictable phenotypic variation into clonal populations of pluripotent stem cells, and this variation is likely to be much more pronounced in cultured female cells. This epigenetic variability has obvious negative implications for the clinical applications of stem cells.
De novo DNA methylation occurs primarily in non-dividing germ cells in a sexually dimorphic manner . A key regulator of de novo methylation is the DNA methylation cofactor/adaptor DNMT3L (DNA methyltransferase 3-like). Genetic studies show that DNMT3L is required for the establishment of genomic imprints in growing oocytes  and for de novo methylation at retrotransposons in prospermatogonia . Although DNMT3L possesses the structural folds present in all catalytically active mammalian DNA methyltransferases , it lacks the functional domains required for catalytic activity, and is unable on its own to cause DNA methylation . Biochemical studies have demonstrated that DNMT3L can function as a regulator of the DNA methyltransferases DNMT3A and DNMT3B . DNMT3L is not expressed in differentiated somatic cells but is expressed in embryonic stem (ES) cells, which are known to be highly active in DNA methylation [7, 8]. We previously showed that DNMT3L forms a complex with DNMT3A2 and DNMT3B, and that this complex specifically binds to nucleosomes that are unmethylated at lysine 4 of histone H3 (H3K4) . Biochemical studies revealed that DNMT3L interacts via the N-terminal cysteine-rich region with the N terminal tail of histone H3, and that this interaction is abolished by di- or trimethylation of H3K4. This resulted in the postulation of the DNMT3L histone recognition hypothesis, which states that recognition of DNA methylation target sequences is dependent on the ability of DNMT3L to bind the histone H3 N-terminus and that regulation of H3K4 methylation plays a role in targeted de novo DNA methylation. It is interesting to note that genomewide analysis of DNA methylation and H3K4 methylation, particularly di- and tri-methylation, reveals a mutually exclusive distribution , supporting the notion that H3K4 methylation protects promoter regions from de novo methylation.
Maintenance methylation is very stable in differentiated/somatic cells, and DNA that is methylated in predetermined patterns maintains this methylation pattern for >80 cell divisions in transfected cells . This stability is a consequence of recognition of hemimethylated DNA after DNA replication by DNMT1 and the regulatory factor UHRF1 (ubiquitin-like, containing PHD and RING finger domains 1) . Both DNMT1 and UHRF1 bind to hemimethylated CpG dinucleotides, and deficiency in either factor results in genomewide demethylation and embryonic lethality [12–14]. Additional mechanisms are likely to be involved in the correct recruitment of both DNMT1 and UHRF1. The observation that UHRF1 is able to bind to histone H3 that is di- or trimethylated at lysine 9  implies the involvement of other chromatin factors.
Mitotic inheritance of genomic methylation patterns has been reported to be less faithful in ES cells than in differentiated somatic cells. A study of imprinted loci by Dean et al.  and Humphreys et al.  reported that methylation imprints are gained and lost at high rates in clonal populations of ES cells, although the mechanism of this was not apparent. Zvetkova et al.  reported spontaneous loss of methylation at imprinted and repeat sequences specifically in female ES cells; this was attributed to lower levels of DNMT3A/DNMT3B in XX cells.
We report here that mouse ES cells that lack DNMT3L lose methylation during culture, unlike non-stem cells, which maintain methylation patterns in the absence of DNMT3L. Loss of DNA methylation is much more rapid in female than in male mutant ES cells, even though levels of DNA methyltransferases and DNMT3L are the same in male and female ES cells. We also found that human ES and induced pluripotent stem (iPS) cells tend to lose DNA methylation spontaneously in a process that is independent of sex and passage number. Whereas maintenance methylation in non-stem cells is mediated by the faithful copying of methylation patterns at S phase, stem cell-specific maintenance of genomic methylation patterns involves dynamic demethylation and de novo methylation, which leads to heterogeneous methylation within clonal cell populations. This instability has the potential to cause dysregulation of imprinted genes and other gene expression abnormalities. Epigenetic instability is likely to introduce unpredictable phenotypic variation into clonal populations of ES and iPS cells, and the effect will be more severe when the cells are female.
It was surprising to find that female (XX) ES cells were much less proficient at provirus methylation than were male (XY) ES cells (Figure 1d), both in the presence and absence of DNMT3L. Also surprising was the DNMT3L-independent de novo methylation of the Oct4 promoter (Figure 1e), which normally occurs when ES cells are induced to differentiate . However, there are both DNMT3L-dependent and -independent de novo methylation events in germ cells [2, 3].
ES and embryonic carcinoma cells are known for their ability to potently restrict retroviral expression [7, 8], which involves two phases. The initial phase, which immediately follows retroviral integration, depends on the interaction of retroviral DNA sequences with host restriction factors, which include TRIM28  and ZFP809 . Maintenance of this repression is subsequently thought to rely on epigenetic mechanisms, primarily DNA methylation. Clonal studies in ES cells using murine stem cell virus (MSCV) transduction followed by knockdown of DNMT3A and/or DNMT3B showed that maintenance of DNA methylation is important for stable proviral silencing , and 5-azacytidine-induced demethylation of previously methylated and silent MSCV provirus resulted in their reactivation . Both these observations suggest that DNA methylation is necessary to enforce provirus silencing. By starting with a population of ES cells in which integrated provirus has escaped the initial silencing system by virtue of replacement of the primer binding site to prevent binding of ZFP809, we investigated the role of DNMT3L-mediated DNA methylation acquisition in the gradual silencing of active retrovirus. We found that the ability to silence over time is dependent on the ability to acquire methylation at proviral LTRs.
Whereas DNMT3L was found to be necessary for de novo methylation of newly integrated proviral DNA, it was dispensable for de novo methylation at a promoter after induction of differentiation. In vitro differentiation of ES cells is known to coincide with de novo methylation at over 300 CpG-poor regions that are in proximity to gene promoters . This is evidence of DNMT3L-independent de novo methylation, which had been previously reported [2, 3]. It is not clear whether the low density of DNA methylation actually represses transcription or whether the de novo methylation of the CpG-poor Oct4 promoter is actually involved in Oct4 regulation.
Previous studies in mouse ES and iPS cells have reported that the presence of two X chromosomes causes genomewide hypomethylation [18, 26]. Our quantitative studies examined the expression levels of DNMT1, DNMT3A and DNMT3B in XX and XY ES cells, and revealed that the increased rate of loss of DNA methylation in XX versus XY ES cells cannot be attributed to reduced amounts of DNA methyltransferase proteins. It is instead consistent with some inherent difference between XX and XY ES cells, which affects DNMT recruitment and general regulation of DNA methylation. XX cells may have a higher rate of loss of methylation or a lower rate of remethylation, or both. The observation that DNMT3L deficiency results in hypomethylation at retrotransposons and minor satellite sequences is also in contrast to previously published results, which claimed that DNMT3L was dispensable for their methylation . Given that we observed a passage-dependent effect on the ability to maintain methylation in the absence of DNMT3L (Figure 2a, b), we propose that this discrepancy might be attributable to lower passage numbers of the DNMT3L-deficient ES cells used in the earlier analysis.
Two previous studies reported that the active methyltransferases DNMT3A and DNMT3B were required for methylation content to be maintained at normal levels [28, 29]. Our study is the first to demonstrate that the catalytically inactive adaptor DNMT3L is required for normal DNA methylation in pluripotent stem cells. DNMT3L is not expressed in differentiated somatic cells, yet unlike Dnmt3L-deficient ES cells, they are able to maintain genomic methylation patterns with high fidelity. These findings indicate that methylation patterns in ES cells are the product of the dynamic gain and loss of DNA methylation, rather than passive clonal inheritance as occurs in differentiated cells. This places a higher load on non-maintenance methylation-based mechanisms, which involve DNMT3 family members. We speculate that there are at least two possible explanations why maintenance methylation-based mechanisms (that is, those involving DNMT1 and UHRF1) are less effective in ES cells. First, ES cells contain combinations of histone modifications not observed in differentiated somatic cells, which could adversely affect recruitment of DNMTs and other factors involved in maintenance methylation. Among these is bivalent methylation of H3K4 and H3K27, two methylation markers that are usually mutually exclusive . Although it has been shown that UHRF1 binds to di- and trimethylated H3K9 , the consequences of H3K27 methylation and of other ES cell-specific patterns of chromatin modifications are unknown but are likely to be responsible for some of the epigenetic instability that occurs in pluripotent stem cells. Second, 5-hydroxymethyl cytosine (hm5C) is present in DNA of ES cells,  and structural models indicate that UHRF1 cannot bind to hm5C . If hm5C occurs within CpG dinucleotides, this could lead to inefficient maintenance methylation. However, the sequence contexts in which hm5C occurs in vivo are not known, and the role of this modified base in maintenance methylation is unclear.
The most significant feature of unstable genomic methylation patterns in pluripotent stem cells may be the introduction of stochastic phenotypic variation into clonal cell populations, particularly with regard to genome destabilization, selection of cells that have increased expression of genes that stimulate cell growth, and the unpredictable gain and loss of imprinted gene expression. It should be noted that cultured ES cells are derived from cell types that exist only transiently in vivo. Selective pressures for high genetic  or epigenetic stability are therefore low in vivo. The forced ex vivo propagation of ES cells for a far greater number of cell divisions than are undergone by their in vivo counterparts renders cultured stem cells - both ES and iPS cells -vulnerable to increased genetic and epigenetic instability.
DNMT3L-deficient ES cells were derived from crosses between Dnmt3L +/- animals  using a previously described protocol . Dnmt1-/- and Dnmt3a -/- ;Dnmt3b -/- ES cells (generously provided by E. Li, Novartis, MA, USA) have also been described previously . Mouse ES cells were cultured on gelatinized tissue culture plates in ES cell media (Dulbecco modified Eagle medium (DMEM) supplemented with 15% fetal bovine serum, 100 IU/mL penicillin, 100mg/mL streptomycin, 2 mmol/L L-glutamine, MEM non-essential amino acids, 0.12 mmol/L β-mercaptoethanol, and leukemia inhibitory factor (LIF) from the conditioned medium of LIF-secreting cells. Human ES and iPS cells were cultured on γ-irradiated (CF1-derived) mouse embryonic fibroblasts (GlobalStem Inc., Rockville, MD, USA) in standard human ES media (DMEM/F-12; Stem Cell Technologies Inc., Vancouver, BC, Canada) supplemented with 20% knockout serum, 1 mM L-glutamine and 100 μM MEM-nonessential amino acids (Invitrogen Corp., Carlsbad, CA, USA), 100 μM 2-mercaptoethanol (Sigma Chemical Co., St Louis, MO, USA) and 10ng/ml recombinant human bFGF (Invitrogen). Both mouse and human ES and iPS cell DNA was purified using a commercial kit (DNeasy® Blood and Tissue Kit; Qiagen Inc., Valencia, CA, USA).
For sorting experiments, green fluorescent protein-positive ES cells were purified on a cell sorter (FACSAria Cell Sorter; BD Biosciences, San Jose, CA, USA), and analyses performed on an automated cell analyzer (FACSCalibur Cell Analyzer; BD Biosciencs).
Retroviral preparation and transduction was performed as described previously . For analysis of Mo-MLV40bp/GFP, bisulfite conversion using the method described by Hajkova and colleagues  was used. For Oct4 promoter analysis, DNA was converted using a commercial kit (EZ DNA Methylation Gold™Kit; Zymo Research Corp., Orange, CA, USA). Analysis and statistical comparison of bisulfite data was performed using QUMA software http://quma.cdb.riken.jp/. For methylation-sensitive Southern blots, DNA was subjected to 2 rounds of digestion with either methylation-sensitive HpaII or the methylation-insensitive isoschizomer MspI (New England Biolabs Inc., Ipswich, MA, USA) to ensure complete digestion. Briefly, DNA was purified from cell pellets (fresh or frozen) using a commercial kit (DNeasy Kit; Qiagen) according to the manufacturer's protocol and quantified, before digesting using a 10-fold unit excess of enzyme. After digestion, DNA was precipitated with ethanol and digested a second time. Digestions were performed for between 4 and 6 hours. Digested DNA was resolved in 1% agarose gels, before being transferred onto a nylon membrane. After ultraviolet-induced cross-linking, membranes were incubated at 65°C with prehybridization solution (6 × saline sodium citrate buffer, 10 × Denhardt solution, 1% sodium dodecyl sulfate, 10% dextran sulfate). LINE1 and Intracisternal A Particle (IAP) probes were used as described previously . The minor satellite probe used has also been described previously . Probes were incubated with membranes overnight, before washing and exposure to phosphor screens (Phosphorimager; Molecular Dynamics, Sunnyvale, CA, USA).
All statistical comparisons were carried out using the non-parametric two-tailed Mann-Whitney test.
For western blotting, the antibodies used were: anti-DNMT1 rabbit polyclonal (pATH52)  1: 800; anti-DNMT3A (SC-20703; Santa Cruz Biotechnologies, Santa Cruz, CA, USA) 1:100; anti-DNMT3B (SC-52922; Santa Cruz Biotechnologies) 1: 200; anti-tubulin mouse monoclonal (T6199; Sigma Chemical Co. St. Louis, MO, USA,); anti-UPF1 Rent1 (H300) (SC-48802; Santa Cruz Biotechnologies), anti-FLAG M2, mouse monoclonal (F3165; Sigma Chemical Co.) 1: 400. Horse radish-conjugated secondary antibodies were obtained from Sigma Chemical Co. IR-800 antibodies used for the Li-Cor detection system were obtained from Rockland Immunochemicals (Gilbertsville, PA, USA).
Primer sets used in the experiments
Nested primers for bisulfite analysis of Mo-MLVGFP long terminal repeat
Bisulfite analysis of Oct4 promoter
This article is dedicated to the memory of our friend and colleague Dan Wolf (July 15, 1977-September 24, 2009). We thank E Li for the gift of 7aabb mutant ES cells, CS Lin and Z Wu for XX and XY ES cells, SP Lin for derivation of Dnmt3L -/- ES cells and Z. A. Hilbert for assistance. Supported by grants from the NIH (to S G, G Q D and T H B). S. P G was supported by a grant from the State of New York Dept. of Health NYSTEM program. D W was an associate and S P G and G Q D are Investigators of the Howard Hughes Medical Institute.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.