DNA glycosylases: in DNA repair and beyond

Scanning the genome for damaged bases

Little is known about how DNA glycosylases find damaged bases in the genome. One idea, proposed on the basis of biochemical evidence and theoretical considerations, postulates the association of the glycosylase with undamaged DNA by non-specific interactions, facilitating sliding along the DNA duplex for a certain distance and scanning the sequence for irregular bases (Berg et al. 1981). Considering the structural and functional diversity of DNA glycosylases, however, there are likely to be different translocation mechanisms, variably involving features of tracking, diffusion, and hopping on the DNA (Blainey et al. 2009; Steinacher and Schär 2005; Friedman and Stivers 2010). Recently, an appealing DNA scanning concept was proposed for DNA glycosylases harboring an [4Fe4S] cluster. The underlying observation was that E. coli MutY and Nth change the oxidation state of their iron–sulfur cluster from [4Fe4S]²⁺ to [4Fe4S]³⁺ upon contact with DNA, which stabilizes the interaction. Thus, if such DNA glycosylases bind in the vicinity of each other, they might act as electron donors and acceptors for each other, making use of the DNA for charge transfer. This may facilitate the dissociation of one glycosylase upon binding of another by reduction of its [4Fe4S] cluster. If the electron transfer involved is perturbed by a base lesion between the two glycosylases, however, both will stay bound to the DNA, increasing the likelihood of damage detection (Boal et al. 2009). This way, [4Fe4S] clusters may support the search for base damage without a need of scanning the entire DNA sequence.

Detection of DNA base damage by DNA glycosylases ultimately requires a full examination of the chemical surface of single bases. To minimize the effort, DNA glycosylases employ strategies of damage pre-selection. OGG1 and UNG, for instance, were proposed to pre-select substrates by establishing superficial base contacts through conserved residues in close proximity to the mouth of their catalytic pocket. This allows potential substrates to be identified without fully inserting every base into the active site pocket (Fig. 2) (Banerjee et al. 2005; Parker et al. 2007). Consistently, NMR studies on human UNG showed the glycosylase to undergo a conformational change upon DNA binding, allowing for oscillation between an open form, loosely interacting with the DNA in an unspecific manner, and a closed form engaged in base examination without fully rotating the base out of the helix (Fig. 2) (Friedman et al. 2009). Base pre-scanning is likely to be facilitated by DNA breathing, considering that many base lesions affect base pairing dynamics to some extent and are therefore likely to enhance local DNA melting.

Open image in new window

Formation of a mature enzyme substrate complex

For final damage verification, the base needs to be flipped out of the DNA helix and accommodated in the active site cavity of the glycosylase. This increases the surface for molecular interactions, providing for a sensitive discrimination of even minor base alterations and, ultimately, catalysis of the base release. One of the best-studied glycosylases with regard to damage recognition is UNG. On its surface, UNG forms a positively charged groove that accommodates the minor groove of the DNA and contains the active site cavity. Upon encountering a uracil in DNA, UNG uses a hydrophobic side chain of a conserved leucine to push the uracil out of the base stack and insert it into the catalytic pocket (Fig. 3). The same residue stabilizes the double helix by occupying the vacated space (Mol et al. 1995; Savva et al. 1995). At the same time, UNG pinches and compresses the double-stranded DNA backbone, thereby inducing DNA bending by about 45° at the site of the damage (Fig. 1) (Parikh et al. 1998). Selectivity for uracil is mediated by several structural features: purine bases are sterically excluded from the active site by the overall narrow geometry of the binding pocket, while the entry of thymine and other 5-methylated pyrimidines is blocked by a side chain of a conserved tyrosine residue. Accordingly, UNG can be converted into a thymine-processing enzyme by an amino acid substitution that eliminates this sterical block (Kavli et al. 1996). Cytosine can enter the active site but, due to unfavorable hydrogen bonding at the bottom of the cavity, cannot be positioned correctly for catalysis (Kavli et al. 1996; Slupphaug et al. 1996).

Open image in new window

While UNG is highly specialized to the excision of DNA uracil, TDG is a good example of a broad spectrum UDG showing mismatch dependency. Unlike UNG, the bacterial and human TDG orthologs have rather spacious pyrimidine-binding pockets, accommodating a large variety of base derivatives (Barrett et al. 1998, 1999; Baba et al. 2005; Maiti et al. 2008), although the preferred substrates are G-mismatched deaminated pyrimidines (Waters and Swann 1998; Hardeland et al. 2003). Because TDG acts on thymine, it needs to be able to discriminate between a regular thymine in DNA (A•T) and one that resulted from deamination of 5-meC (G•T). Substrate selection thus has to take into account not only the base structure itself but also the opposite base. Structural and biochemical studies of E. coli Mug and human TDG (Barrett et al. 1998; Maiti et al. 2008, 2009) have shed light on how this might be achieved. In contact with DNA, TDG undergoes a conformational change in its N-terminal domain, forming a clamp-like structure that permits TDG to track along the DNA (Fig. 2) (Steinacher and Schär 2005). Upon encountering a G-mismatched substrate (G•T), the substrate base is pushed out of the DNA helix by an insertion loop wedging into the DNA helix. This same wedge stabilizes the base stack and forms specific hydrogen bonds with the widowed guanine, mimicking Watson–Crick base pairing (Barrett et al. 1998, 1999; Maiti et al. 2008). These interactions then cooperate with the non-specific DNA binding activity of the N terminus to stabilize the TDG–substrate complex for efficient base excision (Hardeland et al. 2000; Steinacher and Schär 2005).

Another DNA glycosylase well studied with regard to damage recognition is OGG1. Following successful pre-selection of a potential 8-oxoG substrate (Fig. 2) (Banerjee et al. 2005), the oxidized base is flipped into the active site cavity of OGG1, inducing DNA bending of about ~70° due to a tyrosine residue that wedges between the opposing C and its 5′ neighbor. Unstacking of the widowed C facilitates enzyme–DNA contacts that maintain opposite base selectivity. The void generated by 8-oxoG rotation is occupied by a conserved asparagine residue, which contributes to hydrogen bonding with the opposing cytosine. While A, C, and T are excluded from the active site pocket by several strategically positioned amino-acids, a conserved glycine is the only residue discriminating between G and 8-oxoG (Bruner et al. 2000).

Monofunctional DNA glycosylases

To illustrate the monofunctional mode of action, we will focus on UNG and TDG as two well-studied UDGs showing contrasting catalytic features. Upon recognition, uracil is tightly fitted into the active site pocket of UNG (Fig. 3). Interactions between the uracil and amino-acid residues at the bottom of the pocket position the N-glycosidic bond for hydrolysis (Mol et al. 1995; Savva et al. 1995; Slupphaug et al. 1996). Structure and mutational analyses have established a catalytic mechanism involving the polarization of the N-glycosidic bond by a conserved histidine to make it susceptible for nucleophilic attack, and the positioning and deprotonation of a water molecule by a conserved catalytic aspartate, which then attacks the C1 of the deoxyribose (Fig. 2) (Mol et al. 1995; Slupphaug et al. 1996). N-glycosidic bond cleavage is completed by addition of the water nucleophile to uracil, resulting in a free base and an AP-site.

The mismatch-specific uracil glycosylases, e.g., Mug and TDG, interact less specifically with the substrate base within the catalytic pocket and use a less potent mechanism of catalysis, as first revealed in the crystal structure of E. coli Mug (Barrett et al. 1998, 1999). In place of the catalytic aspartate in UNG, Mug/TDG enzymes have an asparagine. This asparagine positions a water molecule but, unlike the aspartate in UNG, is unable to protonate it for an efficient nucleophilic attack on the N-glycosidic bond (Fig. 2). Also, an appropriate residue for polarization of the N-glycosidic bond appears to be missing in Mug/TDG, altogether explaining the comparably low catalytic efficiency of the MUG enzymes (Hardeland et al. 2000; Maiti et al. 2009). These differences illustrate that the mode of catalysis can vary considerably even within one DNA glycosylase superfamily, most likely reflecting the requirement to fine tune substrate spectrum and catalytic efficiency in the evolution of subfamilies with distinct biological functions.

Bifunctional DNA glycosylases

DNA glycosylases that use an amino group of a lysine side chain as a nucleophile for base cleavage form a covalent Schiff-base intermediate with the substrate. The resolution of this reaction intermediate incises the DNA 3′ to the product AP-site, generating a strand break with 3′ phosphate and 5′ OH ends. Thus, these enzymes couple base excision with an AP-lyase step, as best illustrated by the bifunctional mechanism proposed for OGG1 (Sun et al. 1995; Nash et al. 1997). Once an 8-oxoG is stably fitted in the active site cavity of OGG1, the side chain amino group of a suitably positioned catalytic lysine is activated as a nucleophile to attack the C1 of the deoxyribose. The resulting rearrangement to a covalent Schiff-base intermediate releases the damaged base from the DNA, which is then held in position to further participate in the catalysis of the β-lyase reaction, cleaving the DNA strand at the 3′ phosphate. Notably, all steps of the OGG1 catalyzed base release involve the transfer of protons and this is promoted by the excised base itself (Bruner et al. 2000; Fromme et al. 2003). An interesting variation on the bifunctional mode of action is displayed by the mammalian NEIL proteins and their E. coli counterpart Nei. These enzymes couple base excision to beta and delta elimination, incising the DNA strand both 3′ and 5′ to the AP-site (Takao et al. 2002). Moreover, unlike OGG1, E. coli Nei ejects the excised base from the catalytic site immediately, i.e., is capable to perform the AP-lyase reaction without contribution of the damaged base (Zharkov et al. 2002).

DNA glycosylases providing innate immunity

Proteins of the human APOBEC3 subfamily were shown to inhibit replication of a variety of retroviruses, including human immunodeficiency viruses (HIV) (reviewed in Holmes et al. 2007). APOBEC3G was discovered by its ability to restrict replication of a mutant HIV-1 lacking the viral infectivity factor (Vif) (Sheehy et al. 2003). Cells infected with such a virus package APOBEC3G into HIV-1 virions. When these infect new cells, APOBEC3G will deaminate multiple cytosines in the viral cDNA during reverse transcription, which inactivates the provirus (reviewed in Holmes et al. 2007). While APOBEC3G appears to intervene with the viral life cycle at several steps, the antiviral activity mediated by uracilation of the viral genome is coupled to the action of UNG2 and APE1. UNG2 is thought to introduce AP-sites into the deaminated viral cDNA, triggering the cleavage and thus degradation by APE1 (Harris et al. 2003; Yang et al. 2007). Consistently, inhibition and/or downregulation of either UNG2 or APE1 was shown to decrease the antiviral potency of APOBEC3G (Yang et al. 2007). Given that HIV-1 evolved Vpr, a small protein specifically targeting UNG2 and SMUG1 for degradation by the ubiquitin–proteasome system, supports a critical role of uracil base excision in antiviral defense (Schrofelbauer et al. 2005, 2007).

DNA glycosylases providing adaptive immunity

The cooperation of cytidine deaminase and UDG activities is also central to the genetic transactions associated with antibody diversification in the adaptive immune system, i.e., somatic hypermutation (SHM) and class switch recombination (CSR) (Fig. 4) (Muramatsu et al. 2000; Rada et al. 2002).

Open image in new window

SHM introduces point mutations in the light chain variable (V) region of immunoglobulin loci. Mutagenesis is initiated by AID, which is targeted to these loci to deaminate multiple cytosines in single-stranded DNA occurring during transcription (reviewed in Pavri and Nussenzweig 2011). Mutations can arise from these U•G mismatches in several ways (Fig. 4): (1) in the absence of uracil excision, replication across the U•G mismatch would generate a C → T mutation in one of the daughter strands; (2) uracil excision by a UDG and subsequent replication across the resulting AP-site can potentially give rise to any type of base substitution at the site of deamination, although replicative DNA polymerases preferentially insert A opposite an AP-site; (3) long-patch BER, initiated by a UDG, or MMR coupled to error-prone DNA synthesis would cause mutations in proximity to the deaminated cytosine, allowing for mutations to occur also at A•T base pairs.

An involvement of UNG2 in antibody diversification was first implicated by a general perturbance of both SHM and CSR in an UNG2 inhibited chicken B-cell line as well as in UNG-deficient mice (Di Noia and Neuberger 2002; Rada et al. 2002). While the loss of UNG mainly affected SHM at G•C base pairs, inactivation of the mismatch repair system (MMR) was found to diminish hypermutation at A•T pairs (Rada et al. 1998; Wiesendanger et al. 2000). This suggested that error prone MMR contributes to mutagenesis at sites away from the deaminated cytosine (Wilson et al. 2005). Later work then implicated that MMR can indeed operate at an AID induced G•U mismatch if assisted by UNG2, providing a nick at a nearby G•U for initiation of strand excision (Frieder et al. 2009; Schanz et al. 2009).

AID, UNG2, and APE1 were also shown to play a crucial role in the initiation of CSR, a specialized recombination process switching the Ig isotype of an antibody without affecting its antigen specificity. CSR rearranges the antibody constant region of heavy chain genes (C_H) by strand breakage in and joining of two selected switch (S) regions located upstream of every C_H-coding segment. The DNA strand breaks required for initiation of CSR were shown to arise through cytosine deamination, subsequent uracil excision, and AP-site cleavage by AID, UNG2, and an AP-endonuclease, respectively (Fig. 4) (Imai et al. 2003). Consistently, murine B-cells lacking UNG2 show a severe reduction of CSR and in humans, recessive mutations in the UNG gene have been associated with the hyper-IgM syndrome caused by a deficiency in CSR (Rada et al. 2002; Imai et al. 2003). It is also reported, however, that the catalytic activity of UNG2 is dispensable for efficient CSR, while an N-terminal sequence motif appears to be important, suggesting that UNG2 is not responsible for the DNA cleavage step of CSR (Begum et al. 2007, 2009). It is possible that UNG2 simply marks AID-induced G•U mismatches for further processing by other factors, whereby the N terminus is required to mediate specific protein–protein interactions.

Residual CSR and SHM in UNG2-deficient mice hint at a possible contribution of other UDGs. Indeed, overexpression of SMUG1 in MSH2 UNG2 double-deficient cells could partially restore SHM and CSR. A biological role of SMUG1 in antibody diversification, however, is questionable as it is downregulated upon B-cell activation (Di Noia et al. 2006). Likewise, the U•G mismatch-directed glycosylase MBD4 is unlikely to play a major role in either SHM or CSR as a knockout of the gene in mouse showed no effect on either processes (Bardwell et al. 2003). The situation is less clear for TDG. While the impact of a TDG-deficiency on antibody diversification remains to be investigated, its upregulation in activated mouse B-cells hints at a specific function in B-cell maturation (Christophe Kunz and Primo Schär, unpublished data). In this context, TDG might be simply required to prevent mutations at non-Ig genes arising from mistargeted AID. It might, however, also directly contribute to SHM and CSR. Considering its tight interaction with AP-sites, TDG would be optimally suited to delay processing of these repair intermediates, thereby favoring error-prone translesion synthesis and recombination repair.

Concepts of DNA glycosylase-mediated demethylation in vertebrates

First evidence for an involvement of DNA glycosylases in active demethylation in vertebrates came with the discovery of a 5-meC DNA glycosylase activity in extracts of chicken embryos (Jost et al. 1995). The responsible enzyme later revealed itself as a homolog of the human TDG (Zhu et al. 2000b). 5-meC DNA glycosylase activity was then also reported for the human TDG and MBD4 proteins by the same group (Zhu et al. 2000a, 2001). Yet, as several attempts to reproduce these finding were unsuccessful, the idea of TDG and MBD4 acting as demethylating glycosylases did not achieve broad acceptance. Recent observations, however, indicate that the difficulty in reproduction might lie in the necessity of unknown co-factors that either boost the catalytic efficiency of these glycosylases or convert 5-meC into a more favorable substrate.

MBD4 was recently reported to control CpG methylation in the context of parathyroid (PTH) hormone-induced gene activation. This was shown for the CYP27B1 promoter, which undergoes active demethylation upon hormone stimulation (Kim et al. 2009). Both promoter activation and cytosine demethylation coincided with and depended on the physical association of MBD4 and downstream BER factors, but not of the functionally related TDG. Remarkably, this study further showed that phosphorylation of MBD4 by PKC may potentiate its activity to process 5-meC, suggesting that posttranslational modification might be required to unleash a potentially harmful but dormant 5-meC glycosylase activity in certain DNA glycosylases for targeted demethylation under specific conditions (Fig. 5a). This might apply as well to TDG, given its propensity to posttranslational modification by SUMOylation (Hardeland et al. 2002), ubiquitylation (Hardeland et al. 2007), phosphorylation (Um et al. 1998), and acetylation (Tini et al. 2002). It will thus be necessary to revisit TDG's activity as a 5-meC glycosylase under conditions that support the formation of these posttranslational modifications.

Open image in new window

Other lines of investigation support demethylation scenarios that involve the conversion of 5-meC to more favorable substrates for DNA glycosylases. Deamination of 5-meC by a cytidine deaminase is one possibility. This would generate a G•T mismatch that can be acted on by MBD4 or TDG (Fig. 5b). Along these lines, it was shown that the estrogen responsive pS2/TFF1 promoter undergoes cyclic DNA methylation and demethylation during transcriptional activation, involving 5-meC deamination by the methyltransferase Dnmt3a/b itself and subsequent TDG-dependent BER (Metivier et al. 2008; Kangaspeska et al. 2008). The demethylation complex also includes the RNA helicase p68, implicating an RNA component in either targeting and/or stabilization of the complex, consistent with previously reported observations on the chicken 5-methylcytosine DNA glycosylase (Jost et al. 1997; Schwarz et al. 2000). Additional support for a deamination-coupled demethylation pathway came from a study in zebrafish embryos, implicating AID in the deamination of 5-meC and MBD4 in the excision of the resulting G•T mismatch, as well as Gadd45 as an auxiliary factor (Rai et al. 2008). The same concept was adopted more recently to explain TDG-mediated 5-meC demethylation as it may occur during somatic differentiation of the developing mouse embryo (Cortellino et al. 2011). This, however, was mainly inferred from the co-immunoprecipitation of overexpressed AID and Gadd45 with TDG and therefore needs to be corroborated by more direct functional evidence. Finally, an AID-dependent and, thus, deamination-mediated mechanism has been associated with global DNA demethylation occurring in primordial germ cells as well as during the reprogramming of somatic cell nuclei towards pluripotency (Bhutani et al. 2010; Popp et al. 2010). Obviously, genome-wide 5-meC deamination would generate massive amounts of G•T mismatches, and all of these would have to be repaired by the G•T-directed DNA glycosylases MBD4 or TDG. A strong prediction of a deamination-based demethylation model is therefore that a failure of G•T repair would result in genome-wide C → T mutagenesis. This has not been observed so far in MBD4 and/or TDG-deficient cells (Cortazar et al. 2011) and will have to be tested more carefully.

The recent discovery of DNA dioxygenases acting on 5-meC in DNA introduced another conceptual framework for active demethylation. The principle of oxidative demethylation of DNA bases was first described for the bacterial DNA repair protein AlkB. AlkB belongs to a large superfamily of Fe(II)/2-oxoglutarate (2-OG)-dependent hydroxylases and catalyzes the hydroxylation of N-methylated bases like 1-methyladenine (1-meA) and 3-methylcytosine (3-meC), ultimately resulting in demethylation upon release of the hydroxymethyl moiety as formaldehyde (Falnes et al. 2002; Trewick et al. 2002). The superfamily of Fe(II)/2-OG hydroxylases also contains the kinetoplastid base J binding proteins (JBP). Base J stands for β-d-glucosyl(hydroxymethyl)uracil, an abundant base in the genome of kinetoplastida, synthesized through a 5-hmU intermediate generated by enzymatic hydroxylation of thymine by JBP1/2 (reviewed in Borst and Sabatini 2008). Computational analyses identified the mammalian oncogenic TET proteins as close relatives of the JBPs (Iyer et al. 2009). The ultimate finding that TETs comprise a catalytic domain capable of catalyzing the oxidation of 5-meC to 5-hmC uncovered a functional link between these proteins and DNA methylation, possibly demethylation (Tahiliani et al. 2009; Ito et al. 2010). Indeed, 5-hmC was suggested to trigger passive demethylation through inhibition of the maintenance methyltransferase DNMT1 (Valinluck and Sowers 2007). Besides that, 5-hmC might represent an intermediate of active DNA demethylation through stepwise oxidation of 5-meC coupled to either excision repair or decarboxylation (Fig. 5d).

Direct excision of 5-hmC by a DNA glycosylase would seem a plausible scenario (Fig. 5c). A 5-hmC DNA glycosylase was reported to be active in calf thymus extracts (Cannon et al. 1988). The responsible protein, however, has never been purified, nor has 5-hmC glycosylases activity been associated with any known mammalian DNA glycosylase. Given their activity on 5-hmU, TDG and SMUG1 would seem good candidates for 5-hmC processing, but recent evidence shows that at least TDG fails to do so at an appreciable rate (He et al. 2011; Maiti and Drohat 2011). Thus, while direct 5-hmC excision by DNA glycosylases may occur and contribute to DNA demethylation, the underlying enzymatic pathway remains to be clarified.

Another possible route of demethylation would be the further conversion of 5-hmC to an intermediate for DNA glycosylase mediated excision. The deamination of 5-hmC by a specific deaminase (e.g., AID), for example, would give rise to 5-hmU mismatched with G (Rusmintratip and Sowers 2000), which is a substrate for SMUG1 and TDG (Fig. 5e) (Boorstein et al. 2001; Hardeland et al. 2003; Cortellino et al. 2011). As discussed above for 5-meC deamination, however, the concept of demethylation through a pro-mutagenic 5-hmU intermediate is debatable as this would require an immensely efficient and accurate coupling of the deamination and repair processes, particularly in densely methylated DNA sequences, if genome integrity is to be maintained.

Less problematic in this regard seems a more recently implicated mode of TET-mediated demethylation. The key discovery was that TET proteins do not only generate 5-hmC but can oxidize this intermediate further to 5-formylcytosine (5-fC), and 5-carboxylcytosine (5-caC) (Ito et al. 2011), which are good substrates for excision by TDG (He et al. 2011; Maiti and Drohat 2011) (Fig. 5d). Given this, a DNA glycosylase-mediated conversion of 5-meC to C may thus occur without a need to deaminate, i.e., to generate a mutagenic intermediate. However, such a pathway would still trigger massive DNA incision activity in the context of genome-wide active demethylation, which seems a genetically risky and energetically wasteful way to erase an epigenetic mark. A more straightforward way to eliminate 5-caC under these conditions would be to couple the 5-meC oxidation cascade with a decarboxylation step to generate C and CO₂ as final products. A 5-caC decarboxylase is thus an activity to watch out for.

G•T-directed DNA glycosylases and epigenetic stability

Active demethylation of 5-meC in mammalian cells occurs under different circumstances for different purposes, such as genome-wide epigenetic reprogramming, activation of tissue-specific genes during embryogenesis, and the maintenance of active and bivalent chromatin states during and after cell-lineage commitment. These circumstances generate specific demands for a demethylation mechanism (e.g., targeting, catalysis, processivity, efficiency) which are most likely addressed by distinct pathways.

Any of the above considered routes of DNA repair-mediated active demethylation of 5-meC requires an enzyme capable of recognizing and excising a cytosine derivative (5-meC, 5-hmC, 5-caC, T, 5-hmU) in a base (mis)pairing configuration with guanine. On the basis of their substrate spectra, the two DNA glycosylases MBD4 and TDG appear to be most suitable for this purpose, and both have been implicated in one way or another in DNA demethylation as discussed. The biological functions associated with these activities, however, still need to be clarified. Considering, for instance, the reported involvement of MBD4 in hormone-induced promoter demethylation, and the requirement of such demethylation for derepression of select genes during embryogenesis, it is surprising that Mbd4 knockout mice develop normally and display no apparent epigenetic abnormality (Millar et al. 2002; Wong et al. 2002). So, either promoter demethylation during embryogenesis is not developmentally important, or MBD4 is not or only redundantly involved in these processes. The phenotype of Tdg knockout mice, however, has provided more direct evidence for an epigenetic function.

Disruption of TDG in mouse causes embryonic lethality (Cortazar et al. 2011; Cortellino et al. 2011), most likely because TDG-deficient cells fail to properly establish and/or maintain cell-type-specific gene expression programs during cell lineage commitment (Cortazar et al. 2011). This phenotype coincides with the occurrence of aberrant chromatin modifications at promoters of misregulated genes: a loss of active histone marks (H3K4me²), a gain of repressive histone marks (H3K9me³, H3K27me³) and, nota bene, a gradual accumulation of CpG methylation (Cortazar et al. 2011). Together with evidence for a TDG-dependent engagement of BER at affected gene promoters (Cortazar et al. 2011; Cortellino et al. 2011), this suggests that TDG protects CpG-rich promoters from aberrant hypermethylation by active demethylation of erroneously methylated cytosines.

TDG may also contribute to active demethylation of originally methylated sequences. Tissues of TDG deficient embryos showed hypermethylation at the Alb1 enhancer and the Tat glucocorticoid-responsive unit, both undergoing demethylation in the process of tissue specific gene activation (Cortellino et al. 2011). These results can, however, be interpreted in two ways; while the hypermethylation measured in the absence of TDG can indeed be explained by inefficient active CpG demethylation during tissue differentiation, it can equally well be accounted for by a lack of TDG-dependent maintenance of the unmethylated state following successful active demethylation through a different pathway. Hence, while the concept of TDG-mediated active demethylation in the maintenance of hypomethylated states at CpG-rich gene promoters is well supported, its potential contribution to demethylation of methylated sequences requires further validation.

The functions of MBD4 and TDG in mediating DNA demethylation are clearly distinct but there might be a partial overlap; TDG might compensate for the loss of MBD4 in knockout mice but obviously not vice versa. A plausible scenario would be that MBD4 is primarily involved in the demethylation of methylated sequences in the context of developmental gene activation, whereas TDG’s function is to protect unmethylated promoter sequences across the genome from aberrant de novo methylation. This would be supported by the ability of MBD4 to bind to methylated CpGs, and the preferential association of TDG with unmethylated gene promoters.