DNA demethylation pathways: Additional players and regulators

Genetic data implicate AID in demethylation, but the substrate is not clear

Activation induced deaminase (AID) deaminates cytosine to uracil and to a lesser extent 5mC to thymine, by simple hydrolysis. AID has been named for its role in class switch recombination, hypermutation, and gene conversion in activated B-cells 30, 31, mediated by its deaminase activity against cytosines in DNA 32. AID is mainly expressed in lymphoid cells 33, and only scarcely and transiently in other tissues 34, except in the context of malignancy 35-37. Consistent with the expression pattern of Aid in germinal centers, lack of Aid abrogates class switch recombination and somatic hypermutation in the mouse 30. In humans, AID deficiency causes the autosomal recessive form of hyper-IgM syndrome (HIGM2) 38. Both the class switch recombination and hypermutation defects can be explained without reference to DNA demethylation, but additional experiments show that Aid deficiency may nevertheless impair DNA demethylation.

Expression of Aid in embryonic stem (ES) cells 39 and the role of Aid in demethylation during reprogramming experiments involving ES or iPSC cells are controversial. In heterokaryon experiments (involving the fusion of ES cells and differentiated cells), two independent groups found that Aid knockdown reduced the efficiency of reprogramming, and impaired demethylation of promoters of pluripotency associated genes 39, 40, whereas a third group reported no expression of Aid in ES cells and concluded that Aid was not required for reprogramming 41. Reports about the involvement of AID in the generation of induced pluripotent stem cells (iPSCs) are equally contradictory. The original report concluded that AID was required early in reprogramming for the demethylation of selected promoters of pluripotency genes 42. A second study corroborated AID involvement, but found that the enzyme regulated the induction of pluripotency first negatively and then positively 43. Two further studies concluded that Aid did not play any role in reprogramming 44, 45 (Fig. 1).

Studies addressing the role of deamination in DNA demethylation in animals have also not yet led to consensus conclusions. A genetic study of methylation dynamics of germinal center B-cells from mice convincingly supports a role of the enzyme in DNA demethylation 46. However, the same authors found no effect of Aid on DNA methylation levels when using cells in culture, in agreement with the results of at least two other groups 47, 48. A pathway for DNA demethylation in zebrafish embryos involving Gadd45a, Aid/Apobec, and Mbd4 (to excise G:T mispairs) was proposed 49 based on a correlation between Aid/Apobec levels and DNA demethylation activity. However, the experimental findings have since been challenged 50. Slightly higher levels of DNA methylation were found in primordial germ cells of Aid null mice compared to controls, but it was not demonstrated that the effect was direct 51, and results may be influenced by different genetic backgrounds of Aid deficient and control mice 47. Demethylation in the mouse brain by linked deamination and BER has been suggested based on the detection of a complex of Tdg, Gadd45a, and Aid, but Aid activity on 5mC containing DNA could not be directly demonstrated 23 (Fig. 1).

Many reports on a role of AID in demethylation are hard to reconcile with prevailing views about AID expression patterns. Moreover, they conflict with biochemical results, particularly with the tacit assumption that AID acts directly on the 5mC base. Several reports agree that AID deaminates C much more efficiently than 5mC 52-54. Zebrafish Aid is a better 5mC deaminase than other orthologs, but still prefers a C over a 5mC 55. Moreover, AID works only on single stranded DNA (after treatment with RNase to clear RNA-DNA hybrids), but not on double-stranded DNA or RNA-DNA heteroduplexes 56.

Several investigators have sought to resolve the difficulty of poor AID activity toward 5mC by postulating the involvement of alternative deaminases. At least one human APOBEC3 paralog (3A, but not 3G) indeed deaminates 5mC in DNA faster than AID 57, but the single murine APOBEC3 also discriminates against 5mC 52. Some prokaryotic methyltransferases can catalyze deamination of Cs to uracils 58 and of 5mCs to thymines 59 in the absence of S-adenosylmethionine (SAM), but mammalian DNA methyltransferases are not known to catalyze this reaction, either in vitro or in vivo. Moreover, the implication of alternative deaminases does not explain the genetic data that implicate specifically Aid and not deaminases in general in DNA demethylation.

The conflicting genetic and biochemical data suggest that AID acts indirectly. AID may deaminate C, not 5mC, and “regional” repair may then lead to the replacement of methylated cytidines by unmethylated ones in the vicinity 28, 60, 61 (Fig. 2). The model is appealing, because it is consistent with both genetic and biochemical data. Moreover, deamination of C rather than 5mC avoids the formation of a legitimate, but miscoding (and therefore highly mutagenic) DNA base during DNA demethylation. For more detail, and additional arguments for and against a deamination-based DNA demethylation pathway, the reader is directed to other reviews in the literature 62-64.

Genetic and biochemical data implicate TETs in DNA oxidative demethylation

Ten-eleven-translocation (TET) proteins are Fe(II)-dependent dioxygenases that oxidize DNA using molecular oxygen and α-ketoglutarate as co-substrates, generating oxidized DNA, succinate, and carbon dioxide (CO₂) as the co-products. TETs act on modified cytosines. They convert 5mC to 5hmC 65, 66, 5hmC to 5-formylcytosine (5fC), and 5fC to 5-carboxylcytosine (5caC) 67. Simplistically speaking, TET enzymes split molecular oxygen into oxygen atoms (with two unpaired electrons and a carbene-like tendency to insert into other bonds). One oxygen atom inserts into succinic semi-aldehyde, the formal product of α-ketoglutarate decarboxylation, to form succinate, the other into the 5-substituent of the C-base, converting 5mC to 5hmC, 5hmC to 5fC (as a result of the equilibrium between gem-diol and aldehyde by dehydration/hydration), and 5fC to 5caC (Fig. 3A). More realistic models of the reaction mechanism of α-ketoglutarate-dependent dioxygenases in general 68 and of TETs in particular 69 agree that oxygen is split while bound to the iron in the TET active site. The iron resting state is Fe(II). Reaction with oxygen leads to a Fe(IV) = O species involved in oxidizing the DNA base (Fig. 3B).

TETs were originally identified in the context of hematologic malignancies. The first paralog, TET1, is named for the karyotype aberration t(10;11)(q22;q23), that creates a TET1-MLL fusion protein 70. The same study also showed that there are three TET paralogs in humans, as was later found to be true also for mouse and zebrafish. Xenopus has only Tet2 and Tet3, presumably due to lineage specific loss of Tet1. Functional Tets have also been found in invertebrates 71, and even in the protozoan Naegleria gruberi 72, 73, but not in plants 74. Tets are large multidomain proteins (typically in excess of 1,000 amino acids). Tet1 consists of a CXXC domain, a Cys-rich domain (distinct from the CXXC domain, which is also cysteine rich), and a double stranded β-helix (DSBH) domain, which is interrupted by a spacer of unknown function. The Cys-rich and DSBH domains form the catalytic domain harboring the dioxygenase activity. The CXXC domain is involved in Tet targeting. In the case of Tet2, the CXXC domain is provided on a separate polypeptide chain termed IDAX, due to an ancestral chromosomal gene inversion 75. Tet3 occurs in several isoforms with and without CXXC domain 76.

The link between TET activity and demethylation is supported by both genetic and biochemical data. As TETs oxidize 5mC, they obviously reduce the level of 5mC, and conversely, a loss of TETs causes hypermethylation 77. This does not necessarily imply that all oxidized nucleobases are ultimately converted back to the unmodified ones. In the early embryo, 5hmC has been shown to persist through the early divisions 11. Particularly in the brain, 5hmC levels are relatively high 78, and a plethora of 5hmC reader proteins, with no obvious connection to DNA repair, suggest a role for 5hmC as an epigenomic mark in its own right 79. However, there is solid evidence for the entire pathway from 5mC ultimately back to C 22, 80. Recently, the entire pathway has been reconstituted in vitro from defined components 81. TET knockout phenotypes are informative and consistent with a demethylation defect.

Triple Tet knockout ES cells and murine embryonic fibroblasts (MEFs) are viable, but have differentiation and de-differentation defects 77, 82. Triple Tet KO ES cells have problems with teratoma formation, contribute poorly to chimeric embryos, and cannot support embryonic development 77. MEFs have problems with de-differentiation under the influence of defined reprogramming factors (related to the original Yamanaka factors 83). The defect was traced to an impaired mesenchymal-to-epithelial transition (MET), and could be rescued by overexpression of miRNAs that were suppressed by the TET deficiency 82. Conversely, TET overexpression facilitates reprogramming. Tet1 can replace Oct4 in the standard cocktail for iPSC induction 84, and together with Oct4 suffices for iPSC generation in the absence of other exogenous reprogramming factors 85. The positive effects of Tets on reprogramming and differentiation are likely mediated by the control of the methylation state of regulatory loci, particularly enhancers that need to be activated to bring about cell-type changes 14, 86, 87.

In animals, Tet deletion has species-dependent effects. In the mouse, deletion of Tet3 causes neonatal lethality. Germline-specific ablation of Tet3 abolishes the appearance of oxidized 5mC derivatives in the paternal pronucleus 10. However, the consequences of Tet3 ablation from the oocyte are mild. Tet3 ablation reduces fertility, but surprisingly embryos from mating of Tet3 null mothers and wild-type fathers show no overt defects in pre-implantation development and are viable, suggesting that demethylation of the paternal pronucleus is not essential 88. In the mouse, loss of either Tet1 or Tet2 is compatible with embryonic and postnatal development 89, 90. Double knockout of Tet1 and Tet2 leads to partly penetrant mid-gestation abnormalities. Nevertheless, some overtly normal mice are born, albeit with reduced fertility of females 91. Triple Tet KO mice exhibit defects already at the gastrulation stage. The phenotype is akin to the Nodal gain of function phenotype and probably due to hypermethylation of the promoters and consequent decreased expression of the Lefty genes that antagonize Nodal signaling 92. In Xenopus, Tet3 deficiency causes defects in eye and neural development 93. In zebrafish, Tet3 deficiency alone is inconsequential, but combined loss of Tet2 and Tet3 genes, or of all three Tet genes, is incompatible with survival beyond the larval period 94.

Impaired TET activity alone does not seem to cause malignancies, but it clearly contributes, particularly in the hematopoietic system. TETs were named for a translocation affecting the TET1 gene, but the case for the involvement of TETs in malignant transformation is now clearest for a genetic loss of TET2 95-97. Interestingly, impaired TET activity can play a role in malignancies also in the absence of genetic alterations of TET genes. Perturbations of the tricarboxylic acid cycle (TCA) lead to accumulation of genuine or aberrant TCA metabolites, which inhibit TETs, but also other α-ketoglutarate-dependent dioxygenases 98, 99 (Fig. 4 and Box 1).

AID and TETs may cooperate, but are unlikely to act on the same base

The clear evidence for the involvement of both AID and TETs in DNA demethylation has led to suggestions that they may act sequentially on the same DNA base 23. In principle, AID could act upstream or downstream of TETs. Models that AID acts upstream require the formation of the legitimate DNA base thymine as a demethylation intermediate and imply a pathway that should be highly mutagenic. Moreover, they have to be reconciled with the weak activity of AID on 5mC (compared to C) 52. Models that AID acts downstream require the action of AID on 5hmC, 5fC, or 5caC, and conflict directly with biochemical data on AID 52. Moreover, they are hard to reconcile with the observation that most 5-hydroxymethyluracil (5hmU) in the genome appears to be formed by oxidation of T, not deamination of 5hmC (Fig. 5 and Box 2).

Direct reversion of modified DNA bases to C does not seem to occur in vivo

The products of C or 5mC deamination by AID can only be resolved by repair (because deamination decreases similarity to C and may affect base close to, but different from 5mC). In contrast, at least some of the products 5mC oxidation could in principle revert back to C. The search for enzymes that could catalyze such reversion reactions has paradoxically led to the DNA methyltransferases. In vitro, thiol reagents and the de novo mammalian DNA methyltransferases DNMT3a and DNMT3b can convert 5hmC (with loss of formaldehyde) and 5caC (with loss of CO₂) to C in the absence of the methyl donor S-adenosylmethionine 15-17, 100. De novo DNA methyltransferases are present both in the germline and in the oocyte (judging from defects in germline development in their absence 101-104), but a role of these enzymes in demethylation would require the improbable absence of SAM. 5caC decarboxylation activity has been reported for stem cell nuclear extracts 105. Orotidine 5-phosphate decarboxylase (ODC) decarboxylates orotidine 5-phosphate (6-carboxyuridine 5-phosphate) very efficiently 106, but lacks activity on 5caC containing DNA. The stem cell decarboxylase activity therefore remains unconfirmed and its source unknown. Thus, DNA repair appears to be the only in vivo pathway to resolve DNA demethylation intermediates.

Various BER glycosylases are involved in the nucleotide replacement step

Uracil, generated from C by deamination and mispaired with guanine, can be excised by UNG2, SMUG1, TDG, and MBD4 107. Thymine, generated by deamination of 5mC and mispaired with guanine, is excised by TDG and MBD4. The reactions are well documented in the context of somatic hypermutation. In the case of Ung2 61 and Tdg 28, there is also evidence for the involvement of these reactions in the removal of nearby 5mC bases, by either long patch base excision or non-canonical mismatch repair.

In contrast to deamination products, oxidized methylcytosine derivatives are not mispaired, at least when 5fC and 5caC are present in their dominant tautomeric forms 108, 109. However 5caC:G pairs may resemble T:G mispairs when 5caC when 5caC adopts a minor and still debated 109, 110 alternative tautomer 111. 5fC and 5caC nucleotides also resemble BER substrates in many other ways. As formyl and carboxyl groups are electron-withdrawing, both 5fC and 5caC nucleotides have weakened glycosidic bonds 110, 112 like many other oxidatively and otherwise damaged DNA nucleotides 113. Base pairing is weaker for 5caC:G than for C:G 112, presumably leading to a higher rate of spontaneous flipping, as for many damaged DNA bases 114. Weakened glycosidic bonds and increased flipping rates suggest that 5fC and 5caC may be excised by BER glycosylases, and indeed, this is the case (Fig. 6).

Thymine DNA glycosylase (TDG), a monofunctional DNA glycosylase, has been named for its activity against T bases arising from 5mC deamination mispaired to G. TDG was also the first enzyme found to be able to excise 5fC and 5caC 20, 80. Favorable interactions between TDG and the flipped 5caC base 115, and faster excision of 5fC than T from pairs with G 20 have prompted speculation that the “main” role of TDG may be in DNA demethylation rather than in deamination repair. Tdg is barely expressed in oocytes or zygotes (judging from the RNA levels), and not required for demethylation of the paternal pronucleus 116. It is essential for MET 82, which plays a role in somitogenesis and organogenesis (e.g. nephrogenesis, cardiogenesis, or hepatogenesis). The timing of MET fits nicely with the appearance of a Tdg deficiency phenotype around embryonic day 11.5 in the mouse 23, 117, but causality remains to be established.

Nei-like 1 (NEIL1) and Nei-like 2 (NEIL2), both bifunctional DNA glycosylases, excise oxidatively damaged pyrimidines (5-hydroxycytosine, 5-hydroxyuracil, thymine glycol) and purines chemically degraded to bases structurally resembling damaged pyrimidines (hydantoins, formamidopyrimidines) 118. In several screens, NEIL1 or NEIL2 contributed to reactivation of epigenetically silenced reporter plasmids, although to a lesser extent than TDG 22, 119. Leonhardt and co-workers concluded that NEIL1 and NEIL2 could act redundantly with TDG 22. Niehrs and co-workers dispute glycosylase activity on 5fC and 5caC containing DNA, and suggest instead that TDG recruits NEIL1 or NEIL2 (instead of APEX) to excise the (deoxy)ribonucleosides 119. A physiological demethylation process requiring either NEIL1 or NEIL2 remains to be discovered. Neil1 knockout (and heterozygotic) mice develop metabolic syndrome, a combination of severe obesity, dyslipidemia, fatty liver disease, and a tendency for hyperinsulinemia 120, but none of the phenotypes has been linked to a demethylation defect. Neil2 knockdown leads to a neural crest phenotype in Xenopus 119 not seen in Neil2 deficient mice 121.

NEIL3 (unlike NEIL1 and NEIL2) is a monofunctional glycosylase lacking β, δ-lyase activity. Reports are either inconclusive 21 or conflict on whether 22 or not 119 NEIL3 can excise 5caC from DNA. Neil3 is highly expressed in the mouse oocyte, the unfertilized ovum, and the zygote, but then the expression falls strikingly after the zygote stage 122, suggesting a possible involvement in zygotic DNA demethylation (together with UNG2, see below). Neil3 is also expressed in the developing mouse brain, particularly in regions where neurogenesis takes place 122. Neil3 deficient mice are viable and fertile, but exhibit a loss of neural progenitors 123, pointing to a role of the enzyme in rapidly dividing cells, rather than in active DNA demethylation.

Uracil DNA glycosylases occur in mitochondrial (UNG1) and nuclear (UNG2) isoforms that are generated by alternative splicing. The enzymes excise uracils arising from cytosine deamination. Ung2 was recently identified in an unbiased screen for glycosylases involved in TET-dependent gene reactivation. In cultured cells, Ung2, but not an inactive Ung2 variant, prevented the accumulation of 5caC in genomic DNA resulting from Tet2 (catalytic domain) overexpression 21. The finding was unexpected, because in vitro UNG2 had been reported to lack activity against 5caC containing substrates 80. Whether weak UNG2 activity in vitro against 5-carboxyuracil (5caU) containing DNA 21 explains the cell culture data is not clear, because catalyzed conversion of 5caC to 5caU has not been demonstrated yet. As Ung transcript levels are high in the zygote and early embryo, a role of the enzyme in DNA demethylation at this stage was tested. Ung deficiency in the zygote impaired demethylation at some loci (Nanog and Line1 elements, some maternally hypermethylated regions), but did not perturb global levels of oxidized 5mC bases 21. As Ung2 has also been reported to cooperate with Aid in active demethylation in the zygote 61, it is not clear whether loss of Ung impairs oxidation or deamination based demethylation. Ung deficient mice have elevated levels of uracil in DNA, but develop normally into adulthood with no overt phenotype 124.

It is generally assumed that standard steps of BER 125 operate downstream of glycosylase action in DNA demethylation. However, little attention has been paid to whether a single or more nucleotides are replaced. For BER downstream of AID, only long patch repair is productive for DNA demethylation, if indeed AID deaminates Cs and not 5mCs. For BER downstream of TETs, short and long patch repair could be productive for demethylation. Short patch repair with Polβ alone has been reconstituted 81, but whether this is the dominant pathway downstream of TETs in vivo is not clear.

Other damage repair factors and pathways may contribute

In addition to the BER pathway, other DNA damage repair pathways may be involved in active DNA demethylation.

Gadd45a (growth arrest and DNA damage inducible) participates in various pathways that affect genome integrity 126, 127. The gene was also identified in a screen for factors involved in DNA demethylation, and its overexpression was reported to cause activation of methylation silenced reporter constructs and global DNA demethylation 128. Moreover, based on experiments in zebrafish, it was suggested that Gadd45 acted in deamination based DNA demethylation 49. Both studies have been challenged 50, 129, but the involvement of Gadd45 proteins in active DNA demethylation has been confirmed also in several other systems 13, 130. It is not entirely clear which DNA demethylation pathways are stimulated by Gadd45 proteins. Initial data pointed to an involvement of Gadd45 in AID based DNA demethylation 49. More recently, Gadd45a was shown enhance oxidation based DNA demethylation 128. Several DNA repair pathways may be coopted by Gadd45 proteins. Gadd45a interacts directly with the BER glycosylase TDG and enhances oxidation based demethylation. The protein also binds the NER 3′-endonuclease XPG and cooperates with it in demethylation 27.

Nucleotide excision repair (NER) repairs primarily bulkier lesions than BER, or lesions that interfere with transcription (TC-NER). At first sight, the formyl and carboxyl groups of 5fC and 5caC appear too small to trigger NER. However, 5fC causes DNA under-winding 108, like many DNA lesions that are repaired by the nucleotide excision repair pathway 131. 5fC and 5caC in the template strand interfere with transcription 132, 133, suggesting that they may initiate transcription coupled nucleotide excision repair (TC-NER) 134 (Fig. 6). Indeed, the NER 3-endonuclease XPG was shown to be required for demethylation of a reporter in Xenopus oocytes. Moreover, the NER factors XPA, XPG and XPF were shown to be required for demethylation of an rRNA promoter, and it was additionally shown that the catalytic activity of XPG (and not just the XPG protein itself) was required for demethylation 130.

Non-canonical mismatch repair (ncMMR) has also been considered as a possible pathway for the repair step of DNA demethylation. The process is normally geared towards the removal of mismatches starting from a nick elsewhere in the DNA, and leads to replacement of nucleotide patches from the nick or DNA end to ∼150 nucleotides past the mismatch site 135. In a recent study, it was demonstrated that uracils in DNA could trigger the replacement of 5mC nucleotides in DNA in a nick-dependent manner 28. Although the process competes with BER for U:G mismatches, it is appears to be enhanced by BER, because UDG or TDG create the nicks at U sites that are required to prime new DNA synthesis in ncMMR 28. Due to the role of U bases in the process, ncMMR would appear to be involved in AID dependent demethylation. On the other hand, a report that MutSα preferentially binds DNA with 5caC:G pairs over unmodified DNA suggests possible involvement of ncMMR in oxidation based DNA demethylation 111, which has not been corroborated in a functional assay 28. At present, the involvement of ncMMR in DNA demethylation has not yet been demonstrated in physiological circumstances, and it remains unclear how the genome integrity hazards associated with the combined action of AID and ncMMR could be avoided (Box 3).