Human non-CG methylation

Abstract

Non-CG methylation is well characterized in plants, where it appears to play a role in gene silencing and genomic imprinting. Although strong evidence for the presence of non-CG methylation in animals has been available for some time, both its origin and function remain elusive. In this review we discuss available evidence on non-CG methylation in animals in light of evidence suggesting that the human stem cell methylome contains significant levels of methylation outside the CG site.

In plant cells non-CG sites are methylated de novo by Chromomethylase 3, DRM1 and DRM2. Chromomethylase 3, along with DRM1 and DRM2 combine in the maintenance of methylation at symmetric CpHpG as well as asymmetric DNA sites where they appear to prevent reactivation of transposons.1 DRM1 and DRM2 modify DNA de novo primarily at asymmetric CpH and CpHpH sequences targeted by siRNA.2

Much less information is available on non-CG methylation in mammals. In fact, studies on mammalian non-CG methylation form a tiny fraction of those on CG methylation, even though data for cytosine methylation in other dinucleotides, CA, CT and CC, have been available since the late 1980s.3 Strong evidence for non-CG methylation was found by examining either exogenous DNA sequences, such as plasmid and viral integrants in mouse and human cell lines,4,5 or transposons and repetitive sequences such as the human L1 retrotransposon6 in a human embryonic fibroblast cell line. In the latter study, non-CG methylation observed in L1 was found to be consistent with the capacity of Dnmt1 to methylate slippage intermediates de novo.6

Non-CG methylation has also been reported at origins of replication7,8 and a region of the human myogenic gene Myf3.9 The Myf3 gene is silenced in non-muscle cell lines but it is not methylated at CGs. Instead, it carries several methylated cytosines within the sequence CCTGG. Gene-specific non-CG methylation was also reported in a study of lymphoma and myeloma cell lines not expressing many B lineage-specific genes.10 The study focused on one specific gene, B29 and found heavy CG promoter methylation of that gene in most cell lines not expressing it. However, in two other cell lines where the gene was silenced, cytosine methylation was found almost exclusively at CCWGG sites. The authors provided evidence suggesting that CCWGG methylation was sufficient for silencing the B29 promoter and that methylated probes based on B29 sequences had unique gel shift patterns compared to non-methylated but otherwise identical sequences.10 The latter finding suggests that the presence of the non-CG methylation causes changes in the proteins able to bind the promoter, which could be mechanistically related to the silencing seen with this alternate methylation.

Non-CG methylation is rarely seen in DNA isolated from cancer patients. However, the p16 promoter region was reported to contain both CG and non-CG methylation in breast tumor specimens but lacked methylation at these sites in normal breast tissue obtained at mammoplasty.11 Moreover, CWG methylation at the CCWGG sites in the calcitonin gene is not found in normal or leukemic lymphocyte DNA obtained from patients.12 Further, in DNA obtained from breast cancer patients, MspI sites that are refractory to digestion by MspI and thus candidates for CHG methylation were found to carry CpG methylation.13 Their resistance to MspI restriction was found to be caused by an unusual secondary structure in the DNA spanning the MspI site that prevents restriction.13 This latter observation suggests caution in interpreting EcoRII/BstNI or EcoRII/BstOI restriction differences as due to CWG methylation, since in contrast to the 37°C incubation temperature required for full EcoRII activity, BstNI and BstOI require incubation at 60°C for full activity where many secondary structures are unstable.

The recent report by Lister et al.14 confirmed a much earlier report by Ramsahoye et al.15 suggesting that non-CG methylation is prevalent in mammalian stem cell lines. Nearest neighbor analysis was used to detect non-CG methylation in the earlier study on the mouse embryonic stem (ES) cell line,15 thus global methylation patterning was assessed. Lister et al.14 extend these findings to human stem cell lines at single-base resolution with whole-genome bisulfite sequencing. They report14 that the methylome of the human H1 stem cell line and the methylome of the induced pluripotent IMR90 (iPS) cell line are stippled with non-CG methylation while that of the human IMR90 fetal fibroblast cell line is not. While the results of the two studies are complementary, the human methylome study addresses locus specific non-CG methylation. Based on that data,14 one must conclude that non-CG methylation is not carefully maintained at a given site in the human H1 cell line. The average non-CG site is picked up as methylated in about 25% of the reads whereas the average CG methylation site is picked up in 92% of the reads. Moreover, non-CG methylation is not generally present on both strands and is concentrated in the body of actively transcribed genes.14

Even so, the consistent finding that non-CG methylation appears to be confined to stem cell lines,14,15 raises the possibility that cancer stem cells16 carry non-CG methylation while their nonstem progeny in the tumor carry only CG methylation. Given the expected paucity of cancer stem cells in a tumor cell population, it is unlikely that bisulfite sequencing would detect non-CG methylation in DNA isolated from tumor cells since the stem cell population is expected to be only a very minor component of tumor DNA. Published sequences obtained by bisulfite sequencing generally report only CG methylation, and to the best of our knowledge bisulfite sequenced tumor DNA specimens have not reported non-CG methylation. On the other hand, when sequences from cell lines have been reported, bisulfite-mediated genomic sequencing8 or ligation mediated PCR17 methylcytosine signals outside the CG site have been observed. In a more recent study plasmid DNAs carrying the Bcl2-major breakpoint cluster18 or human breast cancer DNA13 treated with bisulfite under non-denaturing conditions, cytosines outside the CG side were only partially converted on only one strand18 or at a symmetrical CWG site.13 In the breast cancer DNA study the apparent CWG methylation was not detected when the DNA was fully denatured before bisulfite treatment.13

In both stem cell studies, non-CG methylation was attributed to the Dnmt3a,14,15 a DNA methyltransferase with similarities to the plant DRM methyltransferase family19 and having the capacity to methylate non-CG sites when expressed in Drosophila melanogaster.15 DRM proteins however, possess a unique permuted domain structure found exclusively in plants19 and the associated RNA-directed non-CG DNA methylation has not been reproducibly observed in mammals despite considerable published20–23 and unpublished efforts in that area. Moreover, reports where methylation was studied often infer methylation changes from 5AzaC reactivation studies24 or find that CG methylation seen in plants but not non-CG methylation is detected.21,22,25,26 In this regard, it is of interest that the level of non-CG methylation reported in stem cells corresponds to background non-CG methylation observed in vitro with human DNA methyltransferase I,27 and is consistent with the recent report that cultured stem cells are epigenetically unstable.28

The function of non-CG methylation remains elusive. A role in gene expression has not been ruled out, as the studies above on Myf3 and B29 suggest.9,10 However, transgene expression of the bacterial methyltransferase M.EcoRII in a human cell line (HK293), did not affect the CG methylation state at the APC and SerpinB5 genes29 even though the promoters were symmetrically de novo methylated at ^mCWGs within each CCWGG sequence in each promoter. This demonstrated that CG and non-CG methylation are not mutually exclusive as had been suggested by earlier reports.9,10 That observation is now extended to the human stem cell line methylome where CG and non-CG methylation co-exist.14 Gene expression at the APC locus was likewise unaffected by transgene expression of M.EcoRII. In those experiments genome wide methylation of the CCWGG site was detected by restriction analysis and bisulfite sequencing,29 however stem cell characteristics were not studied.

Many alternative functions can be envisioned for non-CG methylation, but the existing data now constrains them to functions that involve low levels of methylation that are primarily asymmetric. Moreover, inheritance of such methylation patterns requires low fidelity methylation. If methylation were maintained with high fidelity at particular CHG sites one would expect that the spontaneous deamination of 5-methylcytosine would diminish the number of such sites, so as to confine the remaining sites to those positions performing an essential function, as is seen in CG methylation.30–33 However, depletion of CWG sites is not observed in the human genome.34 Since CWG sites account for only about 50% of the non-CG methylation observed in the stem cell methylome14 where methylated non-CG sites carry only about 25% methylation, the probability of deamination would be about 13% of that for CWG sites that are subject to maintenance methylation in the germ line. Since mutational depletion of methylated cytosines has to have its primary effect on the germ line, if the maintenance of non-CG methylation were more accurate and more widespread, one would have had to argue that stem cells in the human germ lines lack CWG methylation. As it is the data suggests that whatever function non-CG methylation may have in stem cells, it does not involve accurate somatic inheritance in the germ line.

The extensive detail on non-CG methylation in the H1 methylome14 raises interesting questions about the nature of this form of methylation in human cell lines. A key finding in this report is the contrast between the presence of non-CG methylation in the H1 stem cell line and its absence in the IMR90 human fetal lung fibroblast cell line.14 This suggests that it may have a role in the origin and maintenance of the pluripotent lineage.14

By analogy with the well known methylated DNA binding proteins specific for CG methylation,35 methylated DNA binding proteins that selectively bind sites of non-CG methylation are expected to exist in stem cells. Currently the only protein reported to have this binding specificity is human Dnmt1.36–38 While Dnmt1 has been proposed to function stoichiometrically39 and could serve a non-CG binding role in stem cells, this possibility and the possibility that other stem-cell specific non-CG binding proteins might exist remain to be been explored.

Finally, the nature of the non-CG methylation patterns in human stem cell lines present potentially difficult technical problems in methylation analysis. First, based on the data in the H1 stem cell methylome,40 a standard MS-qPCR for non-CG methylation would be impractical because non-CG sites are infrequent, rarely clustered and are generally characterized by partial asymmetric methylation. This means that a PCR primer that senses the 3 adjacent methylation sites usually recommended for MS-qPCR primer design41,42 cannot be reliably found. For example in the region near Oct4 (Chr6:31,246,431), a potential MS-qPCR site exists with a suboptimal set of two adjacent CHG sites both methylated on the + strand at Chr6:31,252,225 and 31,252,237.14,40 However these sites were methylated only in 13/45 and 30/52 reads. Thus the probability that they would both be methylated on the same strand is about 17%. Moreover, reverse primer locations containing non-CG methylation sites are generally too far away for practical bisulfite mediated PCR. Considering the losses associated with bisulfite mediated PCR43 the likelihood that such an MS-qPCR system would detect non-CG methylation in the H1 cell line or stem cells present in a cancer stem cell niche44,45 is very low.

The second difficulty is that methods based on the specificity of MeCP2 and similar methylated DNA binding proteins for enriching methylated DNA (e.g., MIRA,46 COMPARE-MS47) will discard sequences containing non-CG methylation since they require cooperative binding afforded by runs of adjacent methylated CG sites for DNA capture. This latter property of the methylated cytosine capture techniques makes it also unlikely that methods based on 5-methylcytosine antibodies (e.g., meDIP48) will capture non-CG methylation patterns accurately since the stem cell methylome shows that adjacent methylated non-CG sites are rare in comparison to methylated CG sites.14

In summary, whether or not mammalian stem cells in general or human stem cells in particular possess functional plant-like methylation patterns is likely to continue to be an interesting and challenging question. At this point we can conclude that the non-CG patterns reported in human cells appear to differ significantly from the non-CG patterns seen in plants, suggesting that they do not have a common origin or function.

Acknowledgements

This work was supported in part by grants from the Congressionally Directed Medical Research Program of the US ARMY (W81XWH-08-1-0517) and the US National Cancer Institute of the National Institutes of Health (CA102521 and CA 136055) to S.S.S. and by grant 09-08-00687 to O.D. and 10-04-00037 to Y.B. from the Russian Foundation of Fundamental Research.