Active DNA demethylation—The epigenetic gatekeeper of development, immunity, and cancer

Abstract DNA methylation is a critical process in the regulation of gene expression with dramatic effects in development and continually expanding roles in oncogenesis. 5‐Methylcytosine was once considered to be an inherited and stably repressive epigenetic mark, which can be only removed by passive dilution during multiple rounds of DNA replication. However, in the past two decades, physiologically controlled DNA demethylation and deamination processes have been identified, thereby revealing the function of cytosine methylation as a highly regulated and complex state—not simply a static, inherited signature or binary on‐off switch. Alongside these fundamental discoveries, clinical studies over the past decade have revealed the dramatic consequences of aberrant DNA demethylation. In this review we discuss DNA demethylation and deamination in the context of 5‐methylcytosine as critical processes for physiological and physiopathological transitions within three states—development, immune maturation, and oncogenic transformation; and we describe the expanding role of DNA demethylating drugs as therapeutic agents in cancer.

(DNMT1), but also capable of de novo placement of methylation sites (DNMT3A and DNMT3B). The significant medical result of uncovering the process of maintenance and de novo methylation was the discovery of two cytidine analogs, 5-azacytidine (azacitidine), and its deoxy derivative 5-aza-2 0 -deoxycytidine (decitabine). These two analogs were originally developed as anticancer antimetabolites, and can irreversibly inhibit DNMTs through covalent binding, ultimately leading to hypomethylation.
While the inhibition of DNMT enzymes induces a gradual, accumulated reduction in methylation, a physiological system of removing cytosine methylation has long been suspected, but remained controversial until approximately 10 years ago. The identification of dioxygenases of the ten-eleven translocation (TET) family and of the DNA repair enzyme thymine DNA glycosylase (TDG) as the main enzymes responsible for active DNA demethylation in mammals confirmed this suspicion. Over the past decade, much progress has been made in delineating the mechanism of DNA demethylation in both biochemical and clinical contexts. A comprehensive view of the current data suggests that DNA demethylation plays a critical role in physiological and physiopathological transitions, with expanding potential as a therapeutic target.

| MECHANISMS OF DNA DEMETHYLATION
So far three mechanisms for demethylation have been identifiedactive demethylation (replication independent), passive demethylation mediated by TET (replication dependent), and 5mC deamination ( Figure 1). Although far less established, and probably less prevalent F I G U R E 1 Pathways of DNA demethylation through oxidation or deamination. Bases are signified by colored shapes (circles = cytosine variants, square = thymine/hydroxymethyluracil), the stem signifying the deoxyribose moiety. A, Replication-independent DNA demethylation: teneleven translocation (TET) enzymes oxidize 5mC (black) to 5hmC (orange) and then sequentially further oxidize to 5fC (red) and 5caC (burgundy). All products are stable, and the latter two are substrates for thymine DNA glycosylase (TDG). Excision by TDG results in an abasic site that is processed in a replication-independent manner by AP endonuclease and downstream base excision repair activities to produce an unmethylated cytosine (white). B, Replication-dependent DNA demethylation: 5hmC is the most prevalent product of TET enzyme activity and is largely found at CpG sites. DNA containing fully methylated CpGs is converted to strands of hemi-methylated DNA during replication (right side of panel). The DNMT1/UHRF complex maintains DNA methylation during replication by targeting hemi-methylated DNA and methylating the newly incorporated, unmethylated cytosine (white) in the complementary strand to 5mC (black). 5hmC (orange), however, is not/poorly recognized by the DNMT1/UHRF complex, as indicated (left side of panel). The "hemi-hydroxymethylated" DNA is either further oxidized by TET enzymes during the remaining cell cycle or "diluted" by subsequent rounds of replication. Thus, in the absence of TET enzyme activity, methylated DNA remains undiluted by replication. C, DNA demethylation by deamination: When methylated cytosine is deaminated (as opposed to oxidized in A and B), thymine (blue) is produced. The presence of thymine opposite guanine, the natural base partner of cytosine, creates a DNA mismatch. TET enzymes are also capable of oxidizing thymine, resulting in 5hmU (fuchsia). Both thymine and 5hmU, when mismatched with G, are specific substrates for TDG; mismatched thymine and 5hmU are also specific substrates of MBD4 and SMUG1, respectively than the two TET-mediated pathways, evidence continues to grow suggesting the presence and importance of deamination in regulation of demethylation.

| 5-Methylcytosine demethylation-replication independent
In the "replication-independent" active demethylation, TET dioxygenases sequentially oxidize 5mC to 5-hydroxymethylcytosine (5hmC), and further to 5-formylcytosine (5fC), and finally to 5-carboxylcytosine (5caC). [8][9][10] 5fC and 5caC are then removed by base excision repair (BER), namely by DNA glycosylases (Figure 2). As a result, the loss of methylated cytosine is not due to the lack of binding of the DNMT1-UHRF in a replication-coupled process (see below), but direct excision through BER. The main DNA glycosylase capable of excising the TET-generated 5fC and 5caC is TDG [9][10][11] ; in addition, Nei-like 1 DNA glycosylase (NEIL1) can also remove 5caC, but not 5fC. 12,13 Of note, the partial redundancy of TDG and NEIL1 appears to be mechanistically relevant, as NEIL1 directly interacts and enhances TDG activity on 5caC. 13 Upon removal of 5fC and 5caC, the resulting apurinic/apyrimidinic/abasic (AP site) is cleaved by AP endonuclease (APE); the one residue gap is filled by DNA polymerase β and finally the nick is sealed by DNA ligase. 14 This pathway is very significant in development, likely due to the need for rapid programming of cellular state changes for pathway commitment, for example, myeloid vs lymphoid, neural vs neural crest.
In another word, aggressive demethylation is needed since cells are rapidly proliferating but also malleable-a critical point). The clear evidence of the developmental significance of this pathway is the embryonic lethality phenotype of TDG null mice. 15,16 2.2 | 5-Methylcytosine demethylation-replication dependent Characterization of TET enzyme function and recruitment has shown that the most prevalent process of demethylation is linked with replication and mediated by gene expression and reprogramming via transcriptional states. [17][18][19] The model is as follows: (a) frequent transcription of genes results in the recruitment of TET enzymes by transcription factors; (b) TET enzymes induce the oxidation of 5mC to 5hmC; (c) 5hmC bases are poorly recognized by the DNMT1-UHRF complex, which targets and fully methylates the transiently hemimethylated strands during replication; (d) after oxidation, the strand becomes hemi-methylated following one round of replication, and then fully demethylated after two rounds of replication. Thus, this mechanism of demethylation is likely a reflection of transcriptional activity, rather than induction of transcription per se.
F I G U R E 2 Biochemical details of pathways of DNA demethylation. Schematic of DNA demethylation pathways. 5caC, 5-carboxylcytosine; 5fC, 5-formylcytosine; 5hmC, 5-hydroxymethylcytosine; 5hmU, 5-hydroxymethyluracil; 5mC, 5-methylcytosine; AID, activation-induced deaminase; AP site, apurinic/apyrimidinic site; APE, AP endonuclease; APOBEC3A, apolipoprotein B RNA-editing catalytic component 3A; C, cytosine; DNMTs, DNA methyltransferases; MBD4, methyl-binding domain 4; NEIL1, Nei-like 1 glycosylase; SMUG1, single-strand selective monofunctional uracil DNA glycosylase 1; T, thymine; TDG, thymine DNA glycosylase; TETs, ten-eleven translocation dioxygenases; U, uracil; UDG, uracil DNA glycosylase This model is consistent with the fact that "CpG islands"-large intragenic CpG-rich regions that act as cis-elements at gene promoters-are largely unmethylated, that is, methylation is inversely proportional to gene expression. Accordingly, the role of TET enzymes in mediating passive demethylation is very significant during differentiation, when a lot of transcriptional activities occur. This is likely due to the need to maintain lineage and pathway commitment once established (ie, once the ball is rolling, keep the pace since the cells are proliferating quickly), as is the case for differentiation of immune progenitors to mature immune cells. 20 One potential problem with this model is that DNMT1 activity is affected by the oxidation state of the hemi-methylated strand-its activity is only about a third on the Cs opposite to 5hmCs compared to those Cs opposite to 5mC. 21

| Deamination of cytosine and 5-methylcytosine
Deamination of cytosine is one of the most characterized processes of cytosine modification, but the least understood in terms of pathway involvement. While pyrimidines are known to be prone to spontaneous deamination, 22 The authors convincingly argue that both processes contribute equally as the rate of A/T mutagenesis is only significantly decreased in the Pms2-Ung double-knockout mice, compared to respective single knockout. 24 This "error-prone" system is utilized in a controlled fashion to diversify the antibody affinity during immunoglobulin switch and somatic hypermutation (SHM), and the nontrivial role of DNA glycosylase repair may allude to the involvement of DNA methylation. 24 Deamination has an inherently important role in the context of DNA methylation. In fact, deamination of 5mC produces thymine, generating a G:T mismatch in the CpG context. Only two enzymes are known to excise thymine in this unique context, TDG and MBD4.
While TDG is known for its causing significant embryonic lethality phenotype when mutated, and its defined role in active demethylation, 15,16 Mbd4 mutant mice develop normally and do not show a dramatic increase in cancer susceptibility [25][26][27] ; however, an increased number of C to T transition mutations at CpG sites are noted, that is, CpG to CpA or CpG to TpG, as expected due to unrepaired G:T mismatches. 25,26 In addition, when Mbd4 mutant mice were crossed into mice mutant for the tumor suppressor gene Apc or the MMR gene Mlh1, an increased tumor burden and rate of tumor progression is noted. 25,26,28 This suggests that MBD4 may serve as a possible link between deamination and demethylation.

| LINKS BETWEEN DEMETHYLATION AND DEAMINATION
Recently, a demethylation pathway involving TET1, AID, and MBD4 was uncovered in mouse embryonic cells; the loss of function of any of these genes resulted in hypermethylation of the transfected CpG island reporter, but not endogenous CpG islands. 29 The model suggested by the authors is that in contrast to maintenance of the overall bimodal methylation pattern, a specific pathway can be activated for "resetting" methylation patterns during somatic cell reprogramming. It is interesting that this putative pathway might be at the intersection between the TET-mediated oxidation and AID/APOBECmediated deamination of methylated CpGs.
Indeed, previous studies have correlated AID with DNA demethylation. For example, our group has shown in vivo the direct interactions between AID and TDG, 16 but has not functionally tested the involvement of AID or other deaminases in demethylation. MBD4 was shown to be involved in active DNA demethylation in zebrafish, acting in concert with AID and the adaptor protein GADD45, 30 but some of these data is considered controversial. 31 However, since many of these studies were conducted prior to the discovery of TET enzyme function and their establishment as prominent mediators of active demethylation, the functional significance of these interactions needs to be reexamined.
Most work investigating AID/APOBEC enzymes in the context of CpG methylation has focused on characterizing the substrate specificities for 5mC and TET-oxidized derivatives. The results of in vitro biochemical studies demonstrated that most AID/APOBEC family enzymes have physiologically insignificant affinities for 5mC, largely dismissing the concept of demethylation through deamination. 32 However, studies over the past 5 years reintroduced the concept of deamination as affecting cytosine methylation, through either a direct or indirect biochemical process. APOBEC3A was recently shown as being capable of significantly deaminating 5mC, providing a direct molecular mechanism of generating T:G mismatches from 5mC. 33 Sabag et al, as mentioned previously, showed that the maintenance of the demethylated state in CpG islands in embryonic stem cells is disrupted in TET1 knockout cells, as expected, as well as in AID, MBD4, and GADD45a null lines. 29 As further studies showed the interactions of AID and GADD45a with TDG, 16 a link between demethylation and deamination has been reinforced.
A recent study further advanced the role of GADD45a in demethylation, and possibly deamination, by demonstrating a mechanistic link with TET1. The prevailing paradigm for TET enzyme targeting for demethylation is through its recruitment by transcription factors to transcriptionally active regions. However, details beyond proteinprotein interactions have been largely unknown. Recent work by Niehrs and collegues makes a significant progress in identifying a process of TET1 recruitment to CpG islands in promoters through DNA-RNA hybrids, or R-loops. 34 Through antisense transcription of the region, a long noncoding RNA (lncRNA) is produced resulting in Rloop formation that is then bound by GADD45a, which subsequently recruits TET1 and induces demethylation. Whether this process is complementary or parallel to the role of transcription factor-induced TET recruitment is an important question that may point to a role for deamination.
Arguably the most significant limitation in the initially proposed deamination-demethylation pathways was the assumption that oxidized uracil compounds, 5-hydroxymethyluracil (5hmU), 5-formyluracil, and 5-carboxymethyluracil, which are considered to be putative substrates for TDG, MBD4 and SMUG, were derived from deamination of the TET-oxidized 5mC derivatives. However, a functional study investigating 5hmU showed that the very TET enzymes could oxidize thymine to produce 5hmU physiologically. 35 In vivo studies support the presence of a deamination-mediated demethylation pathway, possibly intersecting with TET enzyme function.
Guo et al demonstrated that overexpression of TET1 in HEK293 cells increased 5hmC, as expected. Intriguingly, the increase in 5hmC was reduced when the deaminase AID was co-overexpressed with TET1. 36 To investigate whether deaminases were involved in 5hmC demethylation, the authors transfected HEK293 cells with PCR products containing 5hmC. Using bisulfite-sequencing, a genomic DNA methylation analysis method, they found that, when co-transfected with AID or APOBEC family expression constructs, the PCR products showed increased rates of demethylation. The sequence context of 5hmC demethylation showed a 2.4-fold higher rate of demethylated residues in the WRC vs SYC context (where W is A or T, R is A or G, S is G or C and Y is C or T-IUPAC nucleotide codes), reflective of the AID sequence selectivity for WRC motifs. Finally, overall distribution of the 5hmC demethylation deviated significantly from the expected Poisson distribution, suggestive of a processive (ie, by scanning), and not distributive (ie, by random targeting), mechanism of demethylation. 36 The concept of deamination and processive demethylation has since been investigated, but remains significantly understudied overall. Franchini et al used a DNA targeting domain (GAL4) fused to AID and a Xenopus extract system to show that deamination in the vicinity of 5mC residues induced demethylation. 37 The group then applied the same targeting approach in vivo, breeding transgenic GAL4-AID mice to transgenic mice bearing the GAL4 binding sites (UAS) into the first of four paternal (methylated) H19 differentially methylated regions (DMRs). The results showed demethylation throughout the region, irrespective of GAL4 binding motif context, that is, not just of the first DMR, making AID induced deamination of each CpG very unlikely and instead suggesting involvement of a processive DNA repair pathway (ie, long patch BER or MMR). 37 While the functional relations between deamination and demethylation are intriguing, how the respective reactions might co-exist is an open question. Why do enzymes such as SMUG1 and TDG act on 5hmU and related residues, if it is unlikely that AID/APOBEC deaminases generate them in the first place? If TDG, an evolutionarily conserved enzyme with an embryonic lethal knock-out phenotype, has a primary role in demethylation, then why does it have overlapping roles in excising both 5hmC oxidation products (5fC, 5caC) and potential deamination products (T, 5hmU)? One simple yet uninvestigated possibility is that the oxidation reactions by TET enzymes may exacerbate the effects of aberrant deamination. If aberrant deamination of 5mC occurs (ie, by APOBEC3A, thereby producing thymine) in regions of high TET activity, then 5hmU becomes an additional source of DNA damage. In short, AID/APOBEC enzymes may increase the need for a "guardian" of TET (over-)activity; with this perspective, TDG might have a well-designed role in maintaining the genomic code ( Figure 3).
The most striking example of the relationship between (de-)methylation and deamination is in their effects on genomic CpG mutagenesis. While the concept of demethylation driven solely by deamination has clearly not been established, the likely intersection between TET and AID/APOBEC enzyme function can clearly be seen in their mutagenic patterns (Table 1). Further studies will reveal whether TETmediated oxidation and AID/APOBEC-mediated deamination function together as a synchronized pathway or independently to induce controlled CpG mutagenesis.
F I G U R E 3 Blurred boundaries between physiological DNA modifications and pathological DNA damage: the underlying central role of thymine DNA glycosylase (TDG) and base excision repair in active demethylation. The production of stable 5hmC oxidation derivatives (5fC and 5caC) by ten-eleven translocation dioxygenase (TET) enzymes has been well established in vivo but their function remains unresolved. Conversely, 5hmU has long thought to be a form of oxidative DNA damage, yet the regulation of deaminating enzymes is understudied, despite evidence of significant effects on the global DNA methylation state. Enzymatic DNA oxidation and deamination are conceptually linked through the production of substrates specifically excised by TDG. 5hmU, 5-hydroxymethyluracil; 5mC, 5-methylcytosine; 5hmC,5-hydroxymethylcytosine; 5fC, 5-formylcytosine; 5caC, 5-carboxylcytosine; T, thymine. DNA bases are color coded as in Figure 1 A significant element of complexity in uncovering the overlap between methylation and deamination in the genomic context is the reliance on the bisulfite sequencing method, currently the gold standard for identifying methylated cytosines. 50,51 Bisulfite sequence analysis identifies methylated cytosines by deaminating all unmethylated but not methylated cytosines. However, as a result of this in vitro deamination, any cell-produced uracil is indistinguishable from bisulfite-generated uracil. One approach to address this problem is to compare untreated to bisulfite-treated sequences with singlebase resolution in "normal" vs AID/APOBEC contexts, allowing a complete assessment of the cytosine modification state.

| THE SIGNIFICANCE OF 5mC OXIDATION DERIVATIVES
The biochemical pathway of sequential oxidation of 5mC clearly points to TDG as the terminal enzyme, due to its specific excision of 5fC and 5caC. [9][10][11] The presence of this system is reflected in vivooverexpression of TET enzymes increases 5fC and 5caC, while loss of TET activity leads to the reduction of 5hmCas well as 5fC and 5caC. [8][9][10]52 Accordingly, loss of TDG leads to an increase in 5fC and 5caC, in support of its function as the primary excision enzyme of these substrates.
Shortly after being identified as the terminal enzyme in 5mC oxidation, the possibility of TDG affecting DNA methylation was investigated by us as well as by Schär's group. Using knockout strategies, both groups identified TDG as being essential for embryonic development in mice, and showed that DNA methylation increases at developmentally relevant promoter and enhancer sites. 15,16 In addition, our group also described the demethylation defect of homozygous Tdg knock-in mice bearing a point mutation (Tdg N151A ) that inactivated its glycosylase activity, thus providing compelling genetic evidence on the existence of an active, enzymatically driven DNA demethylation process. 16,53 Furthermore, our recent work shows evidence of increased de novo methylation following TDG knockdown in melanoma cell lines 54 and mouse adenomas bearing the Tdg N151A knock-in mutation, 55 respectively. In summary, the studies using three genetic systems across two mammalian species strongly support the role of TDG in preventing aberrant DNA methylation in the form of elevated 5mC.
The mechanism of how TDG affects DNA methylation physiologically is an important but unanswered question. The simplest explanation is that the accumulation of 5fC and 5caC results in negative feedback producing 5mC. In this scenario, an increase in global 5fC and 5caC would precede a global increase in 5mC. While the first part of this possibility has been consistently demonstrated, the kinetics of 5mC accumulation has not, and thus warrants more in-depth investigation. Additional possibilities are more complex and imply a direct effect of 5fC and 5caC. One study demonstrated that 5caC accumulation in TDG null mouse embryo fibroblasts induces ectopic CTCF binding sites across the genome. 56 CTCF is a multifunctional transcription factor that binds to DNA in a sequence specific manner influenced by CpG methylation, and has insulator activity, blocking enhancer-promoter communications.
The best characterized example of its function is in imprinting regulation of IGF-2 expression: when the CTCF binding site is demethylated, CTCF binds and induces a methylation change downstream, thereby repressing IGF-2 expression. Thus, the presence of ectopic CTCF sites can be predicted as having dramatic changes in methylation patterns that need not be reflected as a global increase in 5mC. Perhaps the strongest evidence for 5fC and 5caC oxidation derivatives functioning as markers of active chromatin is in the context of p300, an essential histone acetyltransferase. Song et al used two methods of 5fC genome-wide profiling to show the enrichment at enhancers as well as a correlation with acquired sites containing p300. 66 As the authors note, this may reflect active 5mC oxidation to facilitate p300 binding. In line with this concept, studies have shown TDG to enhance CBP/p300 mediated transcription through allosteric interactions. [67][68][69] Similarly, absence of TDG in vivo results in a reduction of histone acetylation, with in vitro studies confirming TDG as a co-activator of p300 activity. 70 However, the possibility that 5fC and 5caC are inevitable oxidation byproducts of TET enzyme activity cannot be ruled out. Evidence for this lies in characteristics of these bases resembling oxidative damage. 71 and are mutually exclusive with TET2 mutations in AML. 133 When analyzing the cumulative incidences (Table 2), few notable observations can be made. The incidence of mutations is by far most significant in low-grade glioma, consistent with the brain having the highest enrichment of 5mC oxidation derivatives. TET2 mutations are more frequent in "mature" type hematopoietic cancers, consistent with its defined function in terminal differentiation/maturation within both myeloid and lymphoid lineages. 132 The incidence patterns of specific cancers in patients carrying germline mutations appear to resemble the somatic mutation patterns, likely in further support of respective developmental functions. One notable exception is the high incidence of  Table 2 are conserved and bolded. Associated cancers from Table 2 are in red. Solid blue arrows represent transitions reported to involve changes in 5fC/5caC levels. [82][83][84]131 Solid purple arrows represent transitions reported to involve ten-eleven translocation dioxygenase (TET)/thymine DNA glycosylase (TDG) enzymes, discussed in text. Dashed blue arrows represent transitions predicted to involve changes in 5fC/5caC levels T A B L E 2 Pooled prevalence of DNMT3a, TET2, and IDH1/2 mutations in specific cancers interacts with MMR to facilitate CSR. 153,154 Notably, a less publicized but physiologically critical role for AID and deamination is in the suppression of autoreactivity. As AID was originally identified as the driver of antibody diversification through the previously mentioned processes (ie, SHM, CSR), expression and function of AID was thought to be limited to activated and mature B

cells. Interestingly, low level of AID expression in immature B-cells in
human bone marrow is essential for suppressing autoantibody development through induction of apoptosis. 155,156 The consequence of increased autoreactive B-cell clones following AID knockdown may very well allude to a phenotype-modifying role in lymphoma; indeed, autoimmune paraneoplastic syndromes are a common yet highly variable occurrence in lymphomas. 157 Although the mechanisms of demethylation through deamination remain unresolved, recent studies continue to demonstrate their functional connections. AID expression has recently been shown to be directly regulated by the TET enzymes. 158  In the context of MDS, TET2 mutations do not appear to carry a significant prognostic value. 163 Two independent meta-analysis by Guo et al and Lin et al showed an incidence of 18.34% and 19.19%, respectively. In both studies, no significant difference in overall survival of the pooled cohorts was observed when comparing the patients with TET2 mutations to those without them. 94,164 The role of TET2 as a tumor suppressor is not limited to myeloid derived malignancies. TET2 mutations are present in an estimated 56% of patients with angioimmunoblastic T-cell lymphoma and 46% of peripheral T-cell lymphomas, suggesting that the TET2 loss has a more dramatic malignant phenotype in the lymphoid derivatives compared to the myeloid line. [101][102][103][104][105]165,166 Notably, TET2 mutations are far less prevalent in diffuse B-cell lymphomas at 7% of reported cases. 109,167 The clonal expansion of leukemic-driver mutations in clinically healthy patients has been defined as "clonal hematopoiesis of indeterminate potential" (CHIP), which was first proposed by Steensma et al. 168 In line with their developmental and oncogenic significance, TET2 and DNMT3a mutations represent 93% of driver mutations in clonal hematopoiesis. 169 Underscoring the systemic influence of the hematopoietic system, the significance of CHIP has been demonstrated not as an oncogenic precursor, but instead as an accelerator of atherosclerosis and chronic heart failure. [170][171][172] The mechanism by which the TET2 loss-of-function mutations While TET2 appears to have a clear role as a tumor suppressor, the function of TET1 in oncogenesis is equivocal as a tumor suppressor vs a driving oncogene. As mentioned above, the significance of TET1 was first noted as a fusion partner with the MLL gene in AML with the t (10, 11) translocation. 160,161 A subsequent study showed that TET1 was essential to the oncogenesis resulting from this fusion, primarily through regulation of the TET1-MLL fusion protein with its significant target genes. 175 The upregulation of TET1 coinciding with a mechanistic role is strongly supported by the global increase in 5hmC levels within these cells. 176 In the case of solid tumors, a recent bioinformatic study reported elevated expression of TET1 in a subset of triple negative breast cancer cases in the Cancer Genome Atlas (TCGA; in comparison to normal breast tissue and hormone-dependent tumors), which coincided with reduced methylation levels; in addition, TNBC cases with elevated TET1 expression levels and reduced methylation had worse prognosis. 177 As mentioned previously, TET1 has also been shown to be an oncogenic driver in IDHwild-type cholangiocarcinoma. 135 In lymphoid lineage malignancy, however, TET1 appears to play a tumor suppressor role comparable to TET2. TET1 deletion in mouse models, both in isolation and in combination with TET2 deletion, produces B-cell lymphomas. 40,178 Notably, TET1 as a tumor suppressor of B-cell malignancy has also been supported by evidence of reduced TET1 expression in human non-Hodgkin's lymphomas. 40

| AID and APOBEC3A
As discussed earlier, the mechanistic role of AID/APOBEC family enzymes in demethylation is still unclear. However, growing evidence continues to show how these deaminating enzymes play a direct role in immune development and oncogenesis. As with the DNMT and TET enzymes, this effect appears to be in both myeloid and lymphoid lineage cancers.
In diffuse large B-cell lymphoma (DLBCL), a recent study has shown that expression of the activation-induced cytidine deaminase (AICDA) gene that encodes AID is a driver of global methylation heterogeneity and hypomethylation, the features that are correlated with inferior clinical outcome. 179 Another study investigated the methylation levels of the promoter region of the Fanconi anemia complementation group A (FANCA) gene, which is highly expressed in DLBCL. By correlating AID with recruitment of TET2, subsequent demethylation of the promoter, and increased FANCA gene expression, the complex but significant synergy between AID and TET2 is highlighted, mirroring the similar process discussed earlier that is seen in HSC. 159,180 Coupled with the findings of Lio et al showing TET enzymes as directly modulating the expression levels of AID through the "AICDA superenhancer," the framework for an inherent feedback loop appears in place. 158 As the prognostic and therapeutic value of tumor mutational burden becomes much clearer in both solid and liquid tumor types, the mutagenic role of the deaminase class enzymes beyond AID remains an important question. 181 Fraumeni syndrome. Referred to as the "Brazilian variant," the pR337H mutation was discovered as a founder mutation in Brazil and has been associated with the increased incidence of adrenocortical carcinomas (ACC) in children. 191 Interestingly, increased TDG expression was observed in the ACC patients with the presence of the germline pR337H mutation, further suggesting a possible endocrine/ hormonal relationship between TDG and cancer. 192 We have recently identified TDG as an anticancer target in melanoma. TDG knockdown induced cell cycle arrest and killing of melanoma cell lines 54 ; cells that escaped these two fates accumulated >4n DNA content and underwent senescence. 54 Concomitantly, gene set enrichment analysis of TDG knockdown melanoma cells revealed the enrichment of E2F target genes, which are known to be involved in cell cycle progression and DNA replication and repair. 54 Unexpectedly, upregulation of transcripts was also found, encoding immune regulators, secreted cytokines and proteases belonging to the inflammatory and immune response. 54 This suggests that the normal function of TDG suppresses inflammatory responses, presumably through modulation of DNA methylation levels.

| MBD4
Similar to TET1, MBD4 was discovered to have a foundational role in linking cytosine methylation with oncogenesis. Early clinical reports showed a strong correlation between MMR deficiency and secondary MBD4 mutations in Lynch syndrome spectrum cancers (CRC, endometrial, and pancreatic cancer). 28,[193][194][195][196] Subsequent research highlighted the epigenetic significance of MBD4 as protecting against 5mC transition mutations to thymine, specifically through the binding of MBD4 to 5mC deamination products as well as to MLH1 protein directly. 25,[197][198][199][200]  Several reports have alluded to the potential for hypomethylating agents in combination therapy beyond de novo AML. In one report of therapy-related AML in patients with advanced solid tumors, the complete remission rate of the combination of venetoclax and decitabine was comparable to the rates in de novo AML. While therapy-related AML is uncommon, it is associated with aggressive cytogenetic features, and is frequently disqualified from clinical trials. 214 Hypomethylating agents have intriguing and theoretical potential for combination therapy. Hypomethylation has an established effect of endogenous retroviral activation and expression of other antitumor genes, with an overall immune response that has the potential to convert immunotherapy resistant AML to sensitive one. 215 A nonrandomized open label study investigating azacitadine combined with nivolumab in relapsed/refractory AML reported improved remission rates (22%), compared to the reported rates of 10% to 16% for hypomethylating agents used as single-agent therapy. 216 Currently, there are five ongoing clinical trials investigating combination immunotherapy with hypomethylating agents for AML. For more detail regarding immunotherapy for AML, with and without hypomethylating agent combination, readers are referred to recent reviews. 215,217 Combination hypomethylating agents and immunotherapy outside of AML is also being investigated, at both the in vitro and the clinical level.
Patient-derived xenograft studies on microsatellite stable colorectal cancer, which is less sensitive to immunotherapy than the microsatellite instability counterpart, showed an improved response to the programmed death-1 (PD-1) blockade due to the increased expression of antigen presentation-related genes. 218 In a very recent clinical report, combining decitabine with camrelizumab, an anti-PD-1 antibody, demonstrated significantly higher complete remission rates in relapsed/refractory classical Hodgkin lymphoma compared to camrelizumab alone. 219

| CONCLUSIONS
As summarized in this review, multiple epigenetic mechanisms modulate cytosine within the genome to produce significant effects on development, immune function, and oncogenesis. As studies continue to demonstrate the effects of cytosine methylation on each of these processes, the opportunities to apply hypomethylating therapies increase. Similarly, as the relationships between immune maturation and cancer are strengthened, the potential for aberrant deamination to play a significant role has begun to emerge. By combining the perspectives of physiological C/5mC modification and the processes of deamination and demethylation, the potential for novel therapies as well as the optimization of therapies that are currently available will likely be more effectively realized.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.