Recent Advances on DNA Base Flipping: A General Mechanism for Writing, Reading, and Erasing DNA Modifications

Abstract

The modification of DNA bases is a classic hallmark of epigenetics. Four forms of modified cytosine—5-methylcytosine, 5-hydroxymethylcytosine, 5-formylcytosine, and 5-carboxylcytosine—have been discovered in eukaryotic DNA. In addition to cytosine carbon-5 modifications, cytosine and adenine methylated in the exocyclic amine—N4-methylcytosine and N6-methyladenine—are other modified DNA bases discovered even earlier. Each modified base can be considered a distinct epigenetic signal with broader biological implications beyond simple chemical changes. Since 1994, several crystal structures of proteins and enzymes involved in writing, reading, and erasing modified bases have become available. Here, we present a structural synopsis of writers, readers, and erasers of the modified bases from prokaryotes and eukaryotes. Despite significant differences in structures and functions, they are remarkably similar regarding their engagement in flipping a target base/nucleotide within DNA for specific recognitions and/or reactions. We thus highlight base flipping as a common structural framework broadly applied by distinct classes of proteins and enzymes across phyla for epigenetic regulations of DNA.

12.1. Introduction

Chemical modifications of DNA bases have fundamental biological roles in virtually every living organism. In both prokaryotes and many eukaryotes, cytosine can be methylated at carbon-5 (C5) position by cytosine C5 methyltransferases (MTases) to generate 5-methylcytosine (5mC) (Kumar et al. 1994; Bestor 2000) (Fig. 12.1a). In higher eukaryotes, ten-eleven translocation (TET) 5mC dioxygenase enzymes utilize α-ketoglutarate (αKG) and Fe (II) to oxidize the methyl group of 5mC to generate 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC), and 5-carboxylcytosine (5caC) via discrete reactions (Kriaucionis and Heintz 2009; Tahiliani et al. 2009; Ito et al. 2011; He et al. 2011). In prokaryotes, 5mC and 5hmC can be introduced de novo into the genome during phage invasions, as both modified bases can be synthesized prior to incorporation into the phage genome during DNA synthesis (Warren 1980). After DNA synthesis, phage glucosyltransferases can modify 5hmC within the genome to generate glucosylated 5hmC (5ghmC) (Lehman and Pratt 1960; Kornberg et al. 1961; Lunt et al. 1964). Beyond cytosine C5 modifications, exocyclic amine groups of adenine and cytosine can be methylated in prokaryotes to generate N6-methyladenine (N6mA) and N4-methylcytosine (N4mC), respectively (Malone et al. 1995) (Fig. 12.1b, c). Crystal structures of DNA modification enzymes to date have consistently shown that the target nucleotide is flipped out of the double helix for reactions in a process called base flipping.

Fig. 12.1.

Chemical modifications of nucleic acids. (a) DNA cytosine C5 modifications: enzymes and proteins involved in writing, reading, and erasing the modifications via base-flipping mechanisms. (b) Adenine N6 methylation in DNA or RNA: enzymes and proteins involved in writing, reading, and erasing adenine N6 methylation. (c) Cytosine N4 methylation

In addition to the modification writers, modified base readers have also been shown to flip the target base for recognition. Mammalian SET- and RING-associated (SRA) domains recognize 5mC within genome by base flipping (Arita et al. 2008; Avvakumov et al. 2008; Hashimoto et al. 2008), and have been characterized as non-enzymatic base flippers. Since the first discovery in eukaryotes, SRA has been rediscovered in prokaryotes, recognizing 5mC, 5hmC, and/or 5ghmC to coordinate restriction activity in a modification-dependent manner (Horton et al. 2014a–c). In addition to SRA, the bacterial modified cytosine restriction B enzyme (McrB) also flips 5mC for recognition but is structurally distinct from other known base flippers (Sukackaite et al. 2012). Structural homologs of McrB across different phyla may recognize modified bases in a similar way.

A brief survey of DNA base modifications in both prokaryotes and eukaryotes reveals that two major families of enzymes, methyltransferases, and dioxygenases are involved in writing DNA modifications in the four forms of modified cytosine: 5mC, 5hmC, 5fC, and 5caC. In plants, 5mC DNA glycosylase repressor of silencing 1 (ROS1) can excise 5mC and 5hmC (in vitro) (Gong et al. 2002; Jang et al. 2014; Hong et al. 2014), and mammalian thymine DNA glycosylase (TDG) can excise 5fC and 5caC (He et al. 2011; Maiti and Drohat 2011; Zhang et al. 2012; Hashimoto et al. 2012a). These discoveries effectively link the base excision repair pathway, including AlkB homologs (see below), to DNA demethylation/demodification, by which epigenetic signals encoded in the modified cytosines can be reversed. DNA glycosylases represent the most structurally diverse family of enzymes that are involved in base flipping (also known as nucleotide flipping) (Brooks et al. 2013). Thus, base flipping is not restricted to writers and readers, but has been adopted by DNA glycosylases for erasing DNA modifications as well. Together, structural characterizations of writers, readers, and erasers of DNA base modifications in prokaryotes and eukaryotes effectively showcase base flipping as a general mechanism for regulating and translating fundamental epigenetic signals as well as for maintaining genome integrity (i.e., DNA damage repair).

12.2. Base Flipping for Methylation of DNA Bases

12.2.1. Bacterial DNMTs (HhaI, TaqI, Dam, CcrM, and CamA)

Biological methylation is widely engaged in various regulations, and it uses S-adenosyl-l-methionine (AdoMet or SAM) as a primary methyl donor. The methyl group of AdoMet is bound to a positively charged sulfur atom predisposed to a nucleophilic attack. During the methylation reaction, AdoMet loses the methyl group and becomes S-adenosyl-l-homocysteine (AdoHcy or SAH). A number of different families of MTases use AdoMet as cofactor, targeting diverse substrates ranging from small molecules to large macromolecules such as DNA, RNA, proteins, lipid, and polysaccharides. The atoms subjected to methylation also vary, including carbon (C), nitrogen (N), oxygen (O), sulfur (S), and several metals. AdoMet-dependent DNA MTases were first discovered in bacterial restriction-modification systems (Roberts et al. 2015). The known structures of AdoMet-dependent DNA MTases share a common “MTase fold” characterized by mixed seven-stranded β sheets (6↓ 7↑ 5↓ 4↓ 1↓ 2↓ 3↓) in which strand 7 is inserted between strands 5 and 6 antiparallel to the others (Cheng 1995; Schubert et al. 2003).

M.HhaI was the first DNA MTase to be structurally characterized (Cheng et al. 1993) (Fig. 12.2a). It contains an N-terminal MTase domain and a C-terminal target recognition domain (TRD). M.HhaI is a cytosine C5 MTase that methylates the first cytosine within 5′-GCGC-3′ recognition sequences and prevents R.HhaI restriction activity at this site (Roberts et al. 1976; Horton et al. 2020). Before the structure was available, the proposed mechanism predicted that catalytic Cys81 would make a nucleophilic attack on C6 of cytosine to form a covalent complex, followed by transferring the methyl group from AdoMet to cytosine C5 and releasing the covalent intermediate (Wu and Santi 1985, 1987). In 1994, the crystal structure of M. HhaI-DNA with AdoMet was solved as a trapped covalent enzyme-DNA intermediate using 5-fluorocytosine and directly supported the proposed mechanism, presenting the catalytic cysteine covalently linked to C6 and showing methylated C5 adjacent to AdoHcy (Klimasauskas et al. 1994). Yet, the most striking aspect of the structure was that both the MTase and the TRD of the enzyme work simultaneously to bind DNA and flip the target base into the active-site pocket. The mechanism of DNA base access by base flipping has since been described as the framework for other DNA MTases (Cheng and Roberts 2001).

Fig. 12.2.

Examples of bacterial DNA MTases. (a) M. HhaI, a 5mC MTase, (b) M.TaqI, a class γ MTase, (c) EcoDam, a class α MTase, and (d) CcrM, a class β MTase. The MTase domain (green) binds cofactor (in ball model), and the target recognition domain (TRD) is colored in dark blue. The DNA recognition strand containing the flipped target base is in magenta, and the complementary strand in yellow

After the first structure of M.HhaI-DNA was solved, many crystal structures of DNA MTase-DNA complexes have been solved. Besides cytosine C5 methylation, adenine exocyclic N6 methylation is also a critical modification in prokaryotic DNA (Fig. 12.2b–d) and in eukaryotic RNA (Anton and Roberts 2021; Wei and He 2021). The structure of the adenine N6 MTase M. TaqI in complex with DNA and a non-reactive AdoMet analog was solved in 2001 (Goedecke et al. 2001) (Fig. 12.2b). The enzyme methylates adenine within 5′-TCGA-3′ sequence and harbors a similar two-domain organization as M.HhaI, with the conserved N-terminal MTase domain, but a quite distinct C-terminal TRD. The ternary structure is remarkably reminiscent of M.HhaI, involving a flipped adenine in the active site, where the methyl group from the AdoMet analog is positioned near N6 of the flipped adenine. Instead of the catalytic cysteine residue as in M. HhaI, the asparagine 105 side chain and the following proline backbone oxygen make hydrogen bonds with the adenine N6 amine group, potentially modulating the direct transfer of the methyl group from AdoMet. A similar mode of interaction is also seen in the active site of the T4 phage DNA adenine MTase (T4 Dam) that flips adenine in 5′-GATC-3′ sequence, and an aspartate residue (Asp171) contacts adenine N6 (Horton et al. 2005).

Dam is an orphan MTase (Fig. 12.2c), a type of MTase that acts alone without associated cognate restriction endonuclease as part of R-M system (Roberts et al. 2015). Besides Dam, cell cycle-regulated DNA MTase (CcrM) in Caulobacter crescentus (Zweiger et al. 1994) (Fig. 12.2d) and newly discovered Clostridioides difficile adenine MTase A (CamA) (Oliveira et al. 2020) are also orphan MTases, although they belong to different subgroups depending on the sequential order of conserved motifs (Malone et al. 1995; Woodcock et al. 2020a). The kinetics and structural studies revealed these orphan MTases shared similarities while functioning quite differently. Both Dam (α-group) and CcrM (β-group) are responsible for the post-replication maintenance of daughter strand adenine methylation of symmetrical GATC sequence or near-symmetric GAnTC (n=any base) (Messer and Noyer-Weidner 1988; Stephens et al. 1996). CamA (γ-group), predominantly presented in all C. difficile genomes, methylates an asymmetric 6-bp sequence: CAAAAA (underlining indicates the target A) (Oliveira et al. 2020; Zhou et al. 2021). Although the target adenine residue is flipped out during the catalysis for all three enzymes, each individual orphan MTase has its own distinct mechanism. Dam-bound DNA conformation has intact intrahelical paired bases (Horton et al. 2005), CcrM pulls the two DNA strands apart and creates a bubble comprising four unpaid bases for enzyme recognition (Horton et al. 2019), and CamA squeezes out the target adenine by base pair rearrangement (Zhou et al. 2021). In addition, Dam and CamA make base-specific contacts to both DNA strands, whereas CcrM only contacts the bases in the target strand. These unique features allow CcrM to methylate not only double-stranded but also single-stranded DNA (Reich et al. 2018; Konttinen et al. 2020), while both Dam and CamA are inactive on single-stranded DNA.

12.2.2. Mammalian DNMTs (DNMT1, DNMT3A/3L)

Structural features of classic prokaryotic cytosine C5 MTases are extensively shared by mammalian DNA MTases: DNMT1, DNMT3A, and DNMT3B. They are all cytosine C5 MTases containing the MTase domain with the catalytic cysteine and the TRD. DNMT1 is primarily implicated in methylation of the daughter strand during DNA replication to maintain the methylation pattern encoded in the mother strand by preferentially recognizing hemi-methylated DNA in CpG dinucleotide context (Li et al. 1992). On the other hand, DNMT3A and DNMT3B are considered de novo MTases that can methylate CpG sites as well as non-CpG sites (Okano et al. 1999; Ramsahoye et al. 2000; Gowher and Jeltsch 2001). Such differences in substrate specificities are partly due to the involvement of other domains outside the catalytic fragment. For example, the CXXC and BAH1 domains within DNMT1 hinder methylation of unmethylated CpG sites (Song et al. 2011), whereas DNMT3A and DNMT3B do not contain such domains and can readily methylate them.

Moreover, it appears there exist strong influences of flanking sequences on the CpG and non-CpG methylation activity of mammalian DNA MTases and resulting patterns of methylation. Deep enzymology experiments, which utilize substrates with partly randomized sequences and then analyze the methylation levels by bisulfite conversion coupled with deep next-generation sequencing (NGS) readout, along with structural information, delineate an intricate interplay between flanking sequence, the enzyme-mediated base flipping, and the dynamic landscape of DNA methylation (Jeltsch et al. 2021). Flanking sequences of CpG directly influence catalysis by affecting base-flipping mechanisms which are accompanied by conformational changes in the MTase.

For DNMT1 (Fig. 12.3a), there are strong effects of the ± 2 flanking sequences of a CpG on its activity, with an about 100-fold difference in methylation rates of N₋₂N₋₁CpGN₊₁N₊₂ sites with the best and worst flanking sequences (Adam et al. 2020). Three crystal structures augment this study as a DNMT1 complex with DNA with a highly favorable flanking C₋₁ only showed flipping of the target base, while another complex with a less favorable A₋₁ showed the target base flipping together with rotation of the orphaned G. Finally, a structure with the least favorable G₋₁ displayed a rotation of the target base and invasion of the orphaned G into the −1 flank base pair followed by the formation of a GG non-canonical base pair and flipping of the C normally in the base pair with the G₋₁ residue (Song et al. 2011; Adam et al. 2020). Differences in positioning of the helix that follows the catalytic loop of DNMT1 could be observed with the different nucleotides at the N₋₁ position as well: with A and C at that position, the helix of DNMT1 predominantly adopted a kinked conformation, and with G, the helix mainly adopted a straight conformation moving the active-site loop away from the DNA. Differences of residues at the plus-side flank of the CpG appear to affect the minor groove width which may also influence the equilibrium between the two alternative conformations. Structural transition between these two states of active-site loop-helix is required for DNMT1 activity, suggesting that targeting this transition with small compounds could reduce DNA methylation in cancer cells (Ye et al. 2018). Excitingly, this has been shown to be the case with a new class of reversible DNMT1-selective inhibitors containing a dicyanopyridine moiety (Pappalardi et al. 2021) (Fig. 12.3c).

Fig. 12.3.

Human DNMT1. (a) In the absence of DNA. (b) In the presence of DNA. (c) In the presence of DNA- and DNMT1-specicifc inhibitor

The flanking sequence preferences of DNMT3A and DNMT3B have also been investigated using the deep enzymology approach (Gao et al. 2020). Methylation levels were averaged for 4096 CpG sites with randomized ± 3 flanks, i.e., N₋₃N₋₂N₋₁CpGN₊₁N₊₂N₊₃ sites. The highly methylated sites of DNMT3A showed a preference for C at the +1 site, while DNMT3B preferred G/A at the +1 site and a G at the +2 site DNMT3B while both enzymes prefer T at the −2 site. In addition, the ratio of flanking sequence preferences of DNMT3A and DNMT3B for sites extended a more than 100-fold range, illustrating the noticeable divergence in flanking sequence preferences between the two enzymes.

12.2.3. Implications of DNA Methyltransferase Oligomers (DNMT3A/3L, DNMT3A/3B3, EcoP15I, CcrM, and MettL3–14)

The genomic targeting of DNMT3A and DNMT3B is further regulated by additional factors, including their N-terminal domains and DNMT3L. Besides being a catalytic domain, the MTase domain can participate in protein–protein interactions as exemplified by the DNMT3A MTase domain interacting with a naturally inactive MTase-like domain of DNMT3L, a scaffold protein that binds histone tail H3 to guide DNMT3A activities by forming a tetramer of 3L-3A-3A-3L (Jia et al. 2007; Ooi et al. 2007) (Fig. 12.4a). Moreover, DNMT3L can also form a similar linear assembly with DNMT3B (Lin et al. 2020) (Fig. 12.4b), and 3B-3B and 3B-3L interfaces share some conserved residues with those of the DNMT3A-3L complex.

Fig. 12.4.

Human DNMT3 family. (a) DNMT3A-3L tetramer in complex with DNA. (b) DNMT3B-3L tetramer in complex with DNA. (c) DNMT3A-3B3 tetramer in complex with nucleosome

As a key accessory protein of de novo DNA methylation, Dnmt3L predominantly exists in early embryos and embryonic stem cells and is silenced upon differentiation. Dnmt3b3, a catalytically inactive Dnmt3b isoform, which is ubiquitously expressed in differentiated cells and contains the exact 63-residue deletion corresponding to that of Dnmt3L, carries out the regulatory role during late embryonic development and in somatic cells. Dnmt3b3 can positively regulate the catalytic activities of both Dnmt3a2 and Dnmt3b2 either in vitro or in vivo (Duymich et al. 2016; Zeng et al. 2020). The stimulatory effect of Dnmt3b3 is highly dependent on the direct interaction with active form of Dnmt3a or Dnmt3b proteins, and the optimal stimulation is reached at equal molar stoichiometry. The newly reported cryo-electron microscopy structure defined the architecture of a ternary complex of Dnmt3a2-Dnmt3b3 heterotetramer with a nucleosome core particle flanked by linker DNA (Xu et al. 2020) (Fig. 12.4c). This complex contains two monomers each of Dnmt3a2 and Dnmt3b3, forming a tetramer with 3b3-3a2-3a2-3b3, similar to the arrangement of 3L-3a-3a-3L complex. The 3b3-3a2-3a2-3b3 tetramer interacts asymmetrically with the nucleosome with one of the Dnmt3b3 molecules bound to the acidic patch of the nucleosome. The contact point orients both Dnmt3a2 catalytic domains and the second Dnmt3b3 to follow the path of the linker DNA and constrain the arrangement of Dnmt3a target recognition domain with linker DNA. This model suggested a crucial role of Dnmt3b3 in targeting the nucleosome core and driving the de novo methylation on a genome-wide scale.

Interestingly, a multi-subunit prokaryotic DNA N6mA methyltransferase, EcoP15I, contains a DNA MTase dimer in which one monomer is involved in target base flipping and the other in the recognition of DNA base context (Gupta et al. 2015) (Fig. 12.5a). The “division of labor” might be a conserved feature among class-beta MTases, including M.EcoGII (Murray et al. 2018) and M.EcoP15I (Gupta et al. 2015) from Escherichia coli, Caulobacter crescentus cell cycle-regulated DNA methyltransferase (CcrM) (Horton et al. 2019) (Fig. 12.5b), the MTA1-MTA9 complex from the ciliate Oxytricha (Beh et al. 2019) (Fig. 12.5c), and the mammalian MettL3-MettL14 complex (Fig. 12.5d) (reviewed in Woodcock et al. 2020a). These MTases all generate N6-methyladenine in DNA, with some members having activity on single-stranded DNA as well as RNA. The beta class of MTases has a unique multimeric feature, forming either homo- or hetero-dimers, allowing the enzyme to use division of labor between two subunits in terms of substrate recognition and methylation. Thus, dimerization of two structurally comparable proteins for divergent functionalities may be a mechanism for fine-controlling DNA/RNA modifications.

Fig. 12.5.

Examples of dimeric class β MTases. (a) M.EcoP15I-DNA complex. (b) CcrM-DNA complex. (c) A model of Oxytricha MTA1-MTA9. (d) Human MettL3-MettL14 in the absence of substrate RNA or DNA

12.2.4. Plant DNMTs

Plant DNA MTases show similar functionalities to their mammalian counterparts. Met1 is homologous to mammalian DNMT1 and is responsible for the maintenance CpG methylation, whereas domains rearranged methyltransferase 2 (DRM2) is involved in de novo DNA methylation (Law and Jacobsen 2010). DRM2 contains a rearranged MTase domain, such that its N-terminal half is equivalent to the C-terminal half of the conventional MTase fold and vice versa. A structural study of DRM2 family MTase domain has revealed that the rearranged domain still forms a classic MTase structure and functions as a homodimer (Zhong et al. 2014) analogous to the DNMT3A-3L heterodimer. In addition to Met1 and DRM2, plants also have plant-specific DNA MTases, such as CMT2 and CMT3 that are specifically involved in CNG methylation (Stroud et al. 2014; Lindroth et al. 2001; Zemach et al. 2013). The higher diversity of the MTase family within plants compared to the mammalian family suggests that DNA methylation may be more dynamically regulated in plants than in mammals.

12.3. Base Flipping in Oxidative Modifications of Methylated Bases

12.3.1. Eukaryotic TET Enzymes

The 5mC is by far the most widely studied modified base. Yet, if 5mC has been considered “the fifth” base of the genetic alphabet, 5hmC is increasingly being labeled as “the sixth” base and has garnered much attention. The existence of 5hmC in bacteriophage, modified from 2′-deoxycytidine before integration into the viral genome (Warren 1980), was first reported in the early 1950s (Wyatt and Cohen 1952, 1953). In 1993, a novel J base (β-d-glucosyl hydroxymethyluracil) was discovered in trypanosome, in which J-binding proteins (JBP1 and JBP2) are involved in oxidizing the C5 methyl group of thymine during J base synthesis by using αKG and Fe(II) as cofactors to generate 5-hydroxymethyluracil (Gommers-Ampt et al. 1993; Borst and Sabatini 2008). In 2009, mammalian JBP homolog TET enzymes were discovered to oxidize the methyl group of 5mC to generate 5hmC (Tahiliani et al. 2009). Further analysis revealed that TET enzymes could further oxidize 5hmC to 5fC and then to 5caC (Ito et al. 2011; He et al. 2011). Also, TET enzymes have been shown to convert thymine (5-methyluracil) to 5-hydroxymethyluracil by oxidizing the C5 methyl group of thymine (Pfaffeneder et al. 2011; Pais et al. 2015).

Eukaryotic JBP/TET homologs are present across many eukaryotic organisms including amoeboflagellate Naegleria gruberi (Iyer et al. 2013; Hashimoto et al. 2014a, 2015a). Crystal structures of Naegleria gruberi TET-like (NgTET) (Fig. 12.6a) and human TET2 (hTET2) (Fig. 12.6b) in complex with 5mC-, 5hmC, and 5fC-containing DNA have been characterized (Hashimoto et al. 2014a, 2015a; Hu et al. 2013, 2015). All TET structures show a flipped base positioned in the active-site pocket close to N-oxalylglycine (NOG)—an inactive αKG analog—and a divalent metal such as Fe(II) or Mn(II) for stalling the enzyme in the pre-reaction state. Some of the features of the flipped base recognition observed in DNMT-DNA complex structures (Cheng and Roberts 2001) can also be seen in the structures of TET-DNA complexes. The flipped base in the active site of a TET enzyme in complex with DNA is stabilized by π stacking interactions involving an aromatic residue such as Phe295 in NgTET (Hashimoto et al. 2014a) and Tyr1902 in hTET2 (Hu et al. 2013). Also, polar residues such as Asn147, His297, and Asp234 in NgTET contact O2, N3, and N4, respectively, to guide substrate specificities (Hashimoto et al. 2014a), and the methyl or the hydroxymethyl group is oriented toward NOG and Fe(II)/Mn(II) (Hashimoto et al. 2015a; Hu et al. 2015). Often, active-site pockets for flipped bases not only contain residues for base recognition, but also specifically orient the base for distinct reactions depending on the type of enzymes. Base flipping is therefore a common mechanism applied by different classes of enzymes, such as AdoMet-dependent methyltransferases and αKG- and Fe(II)-dependent dioxygenases to recognize and stabilize the target base for specific reactions.

Fig. 12.6.

Examples of Fe(II)- and αKG-dependent dioxygenases. (a) Naegleria gruberi Tet in complex with DNA. (b) Human Tet2 in complex with DNA. (c) E. coli AlkB in complex with DNA. The metal ion is shown as an orange ball in the active site, αKG is colored in blue, and the flipped nucleotide is in green

Like the mammalian DNMTs, flanking sequence preferences of TET1 and TET2 have been analyzed by a deep enzymology approach (Adam et al. 2022) and revealed that TET enzymes show up to 70-fold differences in oxidation rates of either 5mC or 5hmC in CpG target sites embedded in different N₋₂N₋₁CpGN₊₁N₊₂ flanking contexts. For TET1 and TET2 and both substrates, an A is strongly preferred at the −1 site and G is strongly disfavored. At the +1 site, C is generally disfavored, and at the +2 site, TET2 prefers T. TET1 prefers a TA dinucleotide at the −2/−1 site, particularly with 5mC substrates. Yet, a TG at the same place is disfavored especially by TET2. Moreover, a TT dinucleotide is preferred at the +1/+2 site, more so by TET2 and with 5hmC substrates. It was also found that sites with a high genomic 5hmC/5mC ratio are preferentially observed within an A₋₁ and T₊₁ context, while sites with a low genomic 5hmC/5mC ratio are associated with G₋₁ and C₊₁ flanks.

Inspection of structures of TET can help justify these findings. For instance, while only a water-mediated contact is formed with a TA base pair at the +1 site in a TET2 structure (Hu et al. 2015), in another complex with a CG base pair at the +1 site (Hu et al. 2013), a conserved arginine residue makes a direct hydrogen bond to G(+1′) in the minor groove. In turn, the CG(+1) base pair is shifted by 1.5 Å in the direction of the helix axis which may increase the stacking of the target C base in the helix, making base flipping of the target residue more difficult, and explaining the disfavor for C at the +1 flank position. Interestingly, CACGTG appears to be the best sequence which also is the canonical E-box motif, a well-known recognition site for many helix-loop-helix (bHLH) and basic zipper leucine domain (bZIP) transcription factors (Ravichandran et al. 2021). In the case of MAX, a binding partner of the oncogenic transcription factor MYC, MAX exhibits the greatest affinity for a 5caCpG containing E-box, and much reduced affinities for the corresponding 5mC, 5hmC, or 5fC forms (Wang et al. 2017). In the case of TCF4, which binds the E-box element in the context of CG₀-CA₁-CG₂-TG₃ (where the numbers indicate successive dinucleotides), modification of the central CG₂ has very little effect on TCF4 binding, the CA₁ modification has a negative influence on binding, while modification of the flanking CG₀, particularly carboxylation, has a strong positive impact on TCF4 binding to DNA (Yang et al. 2019).

12.3.2. AlkB and Homologs

Similar to TET enzymes, eukaryotic homologs of E. coli AlkB (Fig. 12.6c) such as FTO and ALKBH5 are also αKG- and Fe(II)-dependent dioxygenases that can oxidize the methyl group of N6mA within mRNA to yield demethylated adenine (Jia et al. 2011; Zheng et al. 2013; Zhu and Yi 2014). Another ALKBH family member, ALKBH1, appears to be a DNA N6mA demethylase in mammals with preference of N6mA nucleotide within a bubbled or bulged DNA with flanking duplex stems (Zhang et al. 2020). Interestingly, the human METTL3–METTL14 heterocomplex N6mA MTase (Liu et al. 2014a) has been shown by in vitro methylation assays to methylate ssDNA and unpaired DNA regions with flanking duplex DNA, as well as having activity on dsDNA containing cyclopyrimidine dimers, which are the major UV radiation-induced photoproducts (Woodcock et al. 2019; Yu et al. 2021; Qi et al. 2022).

TET-DNA complex structures are remarkably comparable to that of the AlkB-DNA complex, and both TET and AlkB enzymes are Fe(II)- and aKG-dependent dioxygenases using base flipping for reactions (Hashimoto et al. 2014a; Hu et al. 2013; McDonough et al. 2010). Common structural folds include two twisted β-sheets in the core where the active site is formed (Fig. 12.6). However, the two enzymes differ in an important mechanistic aspect. TET enzymes oxidize CH₃ attached to an inert carbon atom (cytosine or thymine C5). The resulting product (5hmC or 5hmU) is very stable and can undergo further oxidations in subsequent rounds of reactions to generate further oxidized products. On the other hand, FTO and ALKBH1/5 likely generate N6-hydroxymethyladenine intermediate in which the oxidized carbon is attached to a reactive nitrogen atom (adenine N6). This intermediate spontaneously releases the hydroxymethyl group as formaldehyde and decomposes to adenine—the final “demethylated” product (Hashimoto et al. 2015b) (Fig. 12.1b). Therefore, AlkB and its homologs are demethylases, while TET enzymes should not be characterized as demethylases, but would rather be appropriately understood as “writers” that generate additional modifications on 5mC within genomes to alter epigenetic signals.

Several biochemical observations suggest that modified cytosines beyond 5mC may form distinct epigenetic signals. Many 5mCpG readers such as methyl-CpG binding domain (MBD) proteins have shown significantly reduced binding affinity toward 5hmC when compared to 5mC within CpG context (Hashimoto et al. 2012b, 2015b), whereas some proteins may preferentially bind 5hmC (Zhou et al. 2014). DNMT1 has a significantly reduced activity toward hemihydroxymethylated DNA substrate compared to hemi-methylated DNA (Hashimoto et al. 2012b), suggesting that methylation marks altered by TET enzymes can be lost in subsequent DNA replications. In addition, the RNA polymerase II transcription rate can be specifically reduced by 5fC and 5caC (Kellinger et al. 2012; Wang et al. 2015). These findings strongly point to the possibility that modifications beyond 5mC are distinct signals, and future work is needed to elucidate how the modified bases are differently implicated in larger biological contexts. Several 5caC reader proteins have been structurally characterized in the context of specific sequences (Wang et al. 2017; Yang et al. 2019; Hashimoto et al. 2014b).

12.4. Base Flipping in the Recognition of Modified Bases

12.4.1. Eukaryotic SRA Domains

The function of 5mC and N6mA in prokaryotes was classically understood in the context of restriction-modification systems, in which methylated bacterial DNA is protected from restriction digestion (Wilson and Murray 1991). Effects of DNA methylation are fundamentally determined by the way the methyl groups alter various protein–DNA interactions. In eukaryotes, genomic 5mC bases are considered widely involved in various regulatory processes to control gene expression, chromatin states, and genomic stability that are highly relevant in the human disease context (Robertson 2005). Such penetrating biological implications can be partly attributed to the large number of protein–DNA interactions that are potentially affected by DNA methylation in a direct manner. Evidence shows that several transcription factors are prevented from DNA binding when their binding site is methylated (Tate and Bird 1993), whereas several MBD family proteins are specific 5mCpG readers, as previously mentioned (Klose and Bird 2006). Furthermore, a few 5caC readers have been recently characterized (Yang et al. 2020). The interface between modified DNA and its biological effects can be further complicated by the involvement of histone modifications which DNA methylation is profoundly associated with (Cedar and Bergman 2009; Hashimoto et al. 2010).

The initial discovery of 5mC-binding proteins has raised the possibility of other readers involved in modified base recognitions. In 2007, another family of 5mC readers was discovered in plants and was termed SET- and RING-associated (SRA) domain as a part of VIM1 (Woo et al. 2007). A mammalian homolog to VIM1 is UHRF1, which can associate with DNMT1 during post-replicative maintenance of DNA methylation (Bostick et al. 2007; Sharif et al. 2007). In the following year, three crystal structures of the mammalian UHRF1 SRA domain in complex with 5mC-containing DNA were reported (Arita et al. 2008; Avvakumov et al. 2008; Hashimoto et al. 2008). The structures have revealed that SRA recognizes 5mC by base flipping, although it is not a DNA-modifying enzyme such as MTases or dioxygenases. SRA is also structurally distinct from other base flippers and is characterized by a twisted β-sheet fold resembling a half-moon shape (Fig. 12.7a). Remarkably, the 5mC-binding pocket of SRA features familiar modes of base recognitions exemplified by π stacking interactions, recognitions of the N3 and N4 by Asp474 side chain, and a van der Waals’ contact of the C5-methyl group of flipped 5mC by Ser486 Cβ.

Fig. 12.7.

Examples of reader domain proteins using base flipping. (a) Mouse UHRF1 SRA with 5mC, (b) MspJI SRA with 5mC, (c) E. coli McrB-N with N4mC, (d) Thermococcus McrB-N, and (e) human YTHDC1 YTH with N6mA. The conserved strands are in green and one conserved helix (colored in red) is behind the arch. The other helices are in gray. DNA strands are in ribbon with the flipped base in stick presentation. The modified bases are bound in a cage formed by 2–3 aromatic residues

Interestingly, the SRA of UHRF2 binds 5hmC with a slightly higher preference compared to 5mC, and the crystal structure of UHRF2 SRA in complex with 5hmC-containing DNA is available (Zhou et al. 2014). In the structure, 5hmC is flipped and stabilized, and the OH moiety of the hydroxymethyl group is contacted by the backbone carbonyl groups of Thr508 and Gly509 in the active-site pocket which is slightly larger in size compared to that of UHRF1 SRA. Therefore, eukaryotic SRA has been characterized as a base-flipping domain that recognizes both 5mC and 5hmC.

In addition, SRA domains have been rediscovered in prokaryotes in families of modification-dependent restriction enzymes that recognize modified bases and introduce a double-stranded break in some distances away. MspJI was among the first such enzymes to be reported, which recognizes hemi-modified 5mC or 5hmC by the N-terminal SRA-like domain and restricts the DNA by the C-terminal endonuclease domain (Cohen-Karni et al. 2011). The crystal structure of MspJI has been solved with substrate DNA, revealing an SRA-like structure in the N-terminal modification recognition domain that flips the target 5mC (Fig. 12.7b) (Horton et al. 2014b). Despite the lack of amino acid sequence conservation between eukaryotic UHRF1/2 SRA and MspJI SRA, all SRA domains feature a twisted β-sheet fold with a half-moon shape.

As more modification-dependent restriction enzymes have been identified, some of them are found with different specificities toward 5mC, 5hmC, and 5ghmC. AbaSI, unlike MspJI, has an N-terminal Vsr-like endonuclease domain and a C-terminal SRA-like domain (Horton et al. 2014a; Borgaro and Zhu 2013). Its SRA domain seems to preferentially recognize 5ghmC and 5hmC compared to 5mC, as the relative rate of cleavage of DNA containing the corresponding modification is 5ghmC:5hmC:5mC=8000:500:1 (Wang et al. 2011). Structural features within SRA domains that fine-tune such specificities await future characterizations.

12.4.2. EcMcrB-N Homologs as 5mC and N6mA Readers

Modification-dependent restriction enzymes also utilize yet another 5mC recognition domain. The N-terminus of Escherichia coli McrB (EcMcrB-N) recognizes 5mC next to adenine within 5′-ACCGGT-3′ sequences, and McrC associates with McrB to provide endonuclease activity (Stewart et al. 2000). The crystal structure of EcMcrB-N in complex with 5mC-containing DNA shows a flipped 5mC in the active site (Fig. 12.7c), revealing a novel fold distinct from all other known base flippers (Sukackaite et al. 2012). The active site displays familiar π stacking of the flipped 5mC via aromatic residues and van der Waals’ contact of the C5-methyl group via the side chain of Leu68.

The history of the discovery of base flippers suggests a strong possibility of its structural homologs present in a wide spectrum of phyla. EcMcrB-N is poorly conserved among the wide array of McrBC homologs and other domains exist in other homologs for binding modified bases. Indeed, T. gammatolerans N-McrB is structurally distinct from the EcMcrB DNA-binding domain, adopting a YTH domain fold commonly found in eukaryotic proteins (Wu et al. 2017). It binds and flips methylated base out of the DNA (Hosford et al. 2020) (Fig. 12.7d). Similarly, it has been observed that, in addition to N6mA in RNA, the N6mA reader domain of human YTHDC1 binds N6mA in either a single-stranded DNA or a single-base gap between two canonical DNA helices (Yu et al. 2021; Woodcock et al. 2020b) (Fig. 12.7e).

12.4.3. 5mC and N6mA Readers Use Non-Base-Flipping Recognition

While base flipping seems to be a major mechanism by which a modified DNA base can be recognized, it should be noted that many DNA-binding proteins recognize modified bases in a sequence-dependent manner without involving base flipping. Along with previously mentioned MBD family proteins that recognize 5mC within the simple dinucleotide CpG sequence, mammalian DNA-binding proteins such as Kaiso (Buck-Koehntop et al. 2012), Zfp57 (Liu et al. 2012), Klf4 (Liu et al. 2014b), Egr1 (Hashimoto et al. 2014b; Zandarashvili et al. 2015), and AP-1 (Hong et al. 2017) bind 5mC within specific sequences via a common structural motif (Liu et al. 2013). In addition, transcription factors WT1 (Hashimoto et al. 2014b), MAX (Wang et al. 2017), Tcf3-Ascl1 heterodimer (Golla et al. 2014), and TCF4 (Yang et al. 2019) can specifically bind 5caC within their consensus sequences. In prokaryotes, DpnI harbors a C-terminal WH domain that recognizes the methyl group of N6mA within 5′-GATC-3′ sequence via Trp138 involving van der Waals’ interactions (Mierzejewska et al. 2014). Therefore, DNA modifications may regulate transcription-binding sites in much more dynamic and selective manners than previously understood.

12.5. Base Flipping in Removing Modified and Unmodified Bases

12.5.1. Mammalian Thymine DNA Glycosylase (TDG)

The discovery of TET-mediated modified cytosine bases has provided a fresh insight into a long-sought-after pathway of 5mC demethylation/demodification within mammalian genomes (Zhu 2009). In the base excision repair pathway, DNA glycosylases cleave the glycosidic bond between the ribose and the target base and represent the most structurally diverse family of base-flipping enzymes (Brooks et al. 2013). Initially, it was hypothesized that 5mC (and 5hmC) is removed by mammalian 5mC/5hmC DNA glycosylase activities (Vairapandi and Duker 1993; Cannon et al. 1988; Vairapandi et al. 2000; Vairapandi 2004). However, the glycosylase(s) involved was never identified. After the discovery of TET enzymes, mammalian TDG that generally removes uracil or thymine mismatched to guanine was surprisingly revealed to excise 5fC and 5caC to establish genome-wide DNA demethylation (He et al. 2011; Maiti and Drohat 2011; Hashimoto et al. 2012a). The crystal structure of the human TDG catalytic domain in complex with 5caC-containing DNA was solved (Fig. 12.8a), presenting the flipped base in the active site with an Asn140Ala mutation where the C5-carboxyl moiety of 5caC is specifically recognized by the side chain of Asn157 and the Tyr152 amide backbone (Zhang et al. 2012). In another crystal structure of TDG bound to DNA with a non-cleavable (2′-fluoroarabino) analog of 5-formyldeoxycytidine flipped into its active site (Pidugu et al. 2019), TDG provides a hydrogen bond (2.8 Å) from the Tyr152 backbone N−H to the 5fC formyl oxygen. A nucleophilic water molecule is bound by Asn140 and the backbone oxygen of Thr197, supporting an essential catalytic role for Asn140. The 1.6 Å high-resolution structures of TDG and its N140A mutant bound to DNA with 5caC flipped into the active site suggest that acid-catalyzed 5caC excision is facilitated by two water molecules and contact with Asn191, resulting in a protonated form of 5caC that would be ineffective for C, 5mC, or 5hmC (Pidugu et al. 2019). The discovery of TDG excising 5fC and 5caC has effectively linked the base excision repair pathway to DNA demethylation in mammalian systems.

Fig. 12.8.

Examples of DNA glycosylases. (a) Human TDG flipping 5caC opposite guanine. (b) Geobacillus stearothermophilus Endonuclease III in complex with DNA. Iron-sulfur cluster is colored in orange and yellow. (c) Pyrococcus abyssi PabI uses double base flipping. (d) Bacillus cereus AlkD uses no-base flipping

12.5.2. Plant ROS1

In plants, paradoxically, bone fide 5mC DNA glycosylases were clearly demonstrated and identified in 2002 (Gong et al. 2002), approximately a decade before TET and TDG were implicated in DNA demethylation. In Arabidopsis, four closely related 5mC DNA glycosylases exist: ROS1, DME, DML2, and DML3 (Gehring et al. 2006; Morales-Ruiz et al. 2006; Ortega-Galisteo et al. 2008). They have a catalytic glycosylase domain homologous to E. coli endonuclease III (Fig. 12.8b), a helix-hairpin-helix (HhH) fold DNA glycosylase that harbors an iron-sulfur cluster-binding site (Ponferrada-Marin et al. 2009, 2010, 2011, 2012). Thus, plant ROS1 and mammalian TDG have mutually exclusive substrate specificities for 5mC, 5hmC, 5fC, and 5caC: the first two specific for ROS1 and the latter two for TDG (Hong et al. 2014; Hashimoto et al. 2012a) (as shown in Fig. 12.1a). One of the most surprising aspects of plant 5mC DNA glycosylases is that they excise the target base only when both the catalytic glycosylase domain and the C-terminal domain are present (Hong et al. 2014; Mok et al. 2010). The C-terminal domain of ROS1 is conserved only among plant 5mC DNA glycosylases and has been shown to strongly associate with the catalytic domain, suggesting that domain–domain interactions are important for target base recognition and excision (Hong et al. 2014).

While TDG and ROS1 have been clearly implicated in DNA demethylation pathways, jury is still out on the possibility of other pathways to DNA demethylation. In addition to the previously mentioned mammalian 5mC DNA glycosylase activities, 5hmC DNA glycosylase activity was observed in a calf thymus extract (Cannon et al. 1988). A proteomic study has revealed that several mammalian DNA glycosylases such as NTH1, OGG1, NEIL1, and NEIL2 bind 5mC- and 5hmC-containing DNA in a modification-specific manner (Spruijt et al. 2013), though they by themselves do not have the glycosylase activity against 5mC or 5hmC (Hong et al. 2014).

The 5mC DNA glycosylase activity by ROS1 is interesting from a standpoint of historical characterization of DNA glycosylases as DNA damage repair enzymes. In a given genome, there can be many types of damaged bases, and their diversity is on par with many classes of DNA glycosylases that are structurally distinct (Brooks et al. 2013). On the other hand, 5mC in plants is not considered a damaged base and exists in substantial amounts in the Arabidopsis genome (Zhang et al. 2006). Thus, ROS1 must be regulated and specifically targeted to a certain genomic location to initiate DNA demethylation (Zheng et al. 2008; Qian et al. 2012). In addition to 5mC, ROS1 is comparably active on thymine mismatched to guanine and on some damaged pyrimidines, suggesting that ROS1 can be involved in both DNA demethylation and DNA damage repair (Ponferrada-Marin et al. 2009, 2010). Such dual functionality can be applied to TDG, which not only excises thymine or uracil mismatched to G during the process of DNA mismatch repair, but also excises 5fC and 5caC base-paired with guanine for DNA demodification in mammals.

12.5.3. Archaeon PabI Activity as Adenine DNA Glycosylase

Interestingly, the archaeal Pyrococcus abyssi PabI enzyme was initially thought to be a restriction endonuclease but has been re-characterized as a sequence-specific adenine DNA glycosylase (Miyazono et al. 2014, 2020). PabI is comparable to MutY family mismatch repair DNA glycosylases that excise target adenine mismatched to 8-oxoguanine (Fromme et al. 2004). However, PabI is remarkably distinct from MutY, because PabI excises adenine correctly base-paired to thymine in a targeted manner (Fig. 12.8c). It is therefore possible that DNA glycosylases have adapted to function in more processes than DNA damage repair by removing benign bases for various biological regulations.

12.6. Conclusions

First observed in 1994 in the crystal structure of M.HhaI with DNA, base flipping is now understood as a common mode of protein–DNA/RNA interactions adopted by structurally and functionally distinct classes of proteins across various phyla. Base flipping is the only known mechanism for establishing DNA modifications in a targeted manner via DNA MTases and TET dioxygenases. What used to be considered a eukaryote-specific base-flipping SRA 5mC reader has later been shown to be a widely prevalent domain in prokaryotic systems for recognizing several modified bases including 5mC, 5hmC, and 5ghmC. In addition to SRA, more structurally diverse classes of modified base readers have been discovered in prokaryotes, such as the base-flipping McrB-N 5mC reader and the N6mA-recognizing WH domain of DpnI (using non-base-flipping mechanism). Also, DNA glycosylases are base flippers primarily characterized as DNA repair enzymes, though not all DNA glycosylases flip a base/nucleotide for base excision, as presented in the example of bacterial AlkD (Mullins et al. 2015) (Fig. 12.8d). Today, DNA demodification is considered a bone fide output of the base excision repair pathway through DNA glycosylases, such as mammalian TDG and plant ROS1 whose mechanism of action again involves base flipping. In an era in which DNA modifications are considered critical and increasingly complex epigenetic signals, this simple but elegant structural mechanism for protein–DNA interaction is preserved as a truly ubiquitous framework.

Abbreviations

5caC

5-carboxylcytosine

5fC

5-formylcytosine

5ghmC

Glucosylated 5-hydroxymethylcytosine

5hmC

5-hydroxymethylcytosine

5mC

5-methylcytosine

αKG

α-ketoglutarate

AdoHcy

S-adenosyl-l-homocysteine (SAH)

AdoMet

S-adenosyl-l-methionine (SAM)

AlkB

E. coli Alkylated DNA repair protein AlkB

ALKBH5

Alkylated DNA repair protein AlkB homolog 5 in human

CMT2

Chromomethylase 2 (plant specific)

CMT3

Chromomethylase 3 (plant specific)

DME

Demeter (plant)

DML3

Demeter-like protein 3 (plant)

DNMT1

Mammalian DNA methyltransferase 1

DNMT3A

Mammalian DNA methyltransferase 3A

DNMT3L

Mammalian DNA methyltransferase 3-like

DRM2

Domain rearranged methyltransferase 2 (plant)

FTO

Fat mass and obesity-associated protein

HhH

Helix-hairpin-helix

JBP

J-binding protein

MBD

Methyl-CpG binding domain

McrB

Modified cytosine restriction B

Met1

DNA methyltransferase 1 (plant)

MTase

Methyltransferase

N4mC

N4-methylcytosine

N6mA

N6-methyladenine

NOG

N-oxalylglycine

ROS1

Repressor of silencing 1 (plant specific)

SRA

SET- and RING-associated

TDG

Thymine DNA glycosylase

TET

Ten-eleven translocation

TRD

Target recognition domain

Uhrf1

Ubiquitin-like-containing PHD and RING finger domains protein 1

WH

Winged helix