TLC assay

TLC is a classic method that separates different nucleotides or different modified forms of the same nucleotide based on their differential migration rates on TLC plates. Using the methylation-insensitive restriction enzyme MspI and a TLC assay, Tahiliani et al. (2009) estimated that the 5hmC level in mouse ES cells is ∼0.03% of total nucleotides. Upon leukemia inhibitory factor (LIF) withdrawal, 5hmC levels are reduced to ∼40% in differentiated mouse ES cells, consistent with a concomitant decrease in Tet1 and Tet2 mRNA levels (Koh et al. 2011). Using a two-dimensional (2D) TLC method and nearest-neighbor analysis, Kriaucionis and Heintz (2009) reported that 5hmC level is ∼0.6% and 0.2% of total nucleotides in Purkinje and granule neurons, respectively. However, under standard TLC buffer conditions, 5hmC and 5fC have almost identical migration patterns, and 5caC fails to migrate (Ito et al. 2011). To overcome this technical issue, we developed a modified 2D TLC assay in conjunction with the use of TaqI, a restriction enzyme that is insensitive to all cytosine modifications, allowing for clear separation of 5mC and its oxidation products (Fig. 2A; Ito et al. 2011).

Liquid chromatography (LC) and MS

To better quantify the global levels of 5mC oxidation derivatives, several groups have developed MS-based methods. Using stable isotope-labeled reference compounds and LC-MS, Gobisch et al. (2011) found that 5mC levels are relatively similar in all tested mouse tissues (∼4.3% of all cytosine or 43 × 10³ 5mC in every 10⁶ C). However, 5hmC levels varied significantly between different tissues, with the highest levels of 5hmC found in various brain tissues (3 × 10³ to 7 × 10³ 5hmC in every 10⁶ C); medium levels in kidney, nasal epithelia, bladder, heart, muscle, and lung (1.5 × 10³ to 1.7 × 10³ 5hmC in every 10⁶ C); and low levels in pituitary gland, liver, spleen, and testes (0.3 × 10³ to 0.6 × 10³ 5hmC in every 10⁶ C) (Munzel et al. 2010; Globisch et al. 2011). Using a similar approach, the same group recently discovered that 5fC is also present in mouse ES cells and estimated that 5%–10% of 5hmC in mouse ES cell DNA is further oxidized to 5fC (0.2 × 10³ 5fC in every 10⁶ C vs. 3.9 × 10³ 5hmC in every 10⁶ C) (Pfaffeneder et al. 2011). Upon differentiation by LIF withdrawal or in Dnmt3a/3b-deficient mouse ES cells, 5fC level is significantly reduced, reinforcing the notion that 5fC is generated from pre-existing 5mC/5hmC by Tet proteins. Using a quantitative LC-MS assay, our group recently characterized the enzymatic kinetics of the Tet-catalyzed iterative oxidation reaction and quantified the endogenous levels of 5hmC, 5fC, and 5caC (Fig. 2B; Table 1). This analysis indicated that the genomic content of these modified bases in mouse ES cells is ∼1.3 × 10³ 5hmC, 20 5fC, and three 5caC in every 10⁶ C, respectively (Ito et al. 2011). While 5hmC and 5fC are also present in various mouse tissues, 5caC can only be reliably detected in mouse ES cells (Table 1; Ito et al. 2011).

Modification-specific antibodies

DNA methylation patterns can be determined by using antibodies against 5mC or methyl-CpG-binding domains (MBDs). Similarly, antibodies specific for 5hmC, which are available from both commercial (e.g., Diagenode and Active Motif) and academic (Williams et al. 2011) sources, have been used for genome-wide mapping of 5hmC distribution by high-throughput sequencing or whole-genome tiling microarrays (Fig. 2C; Ficz et al. 2011; Jin et al. 2011; Stroud et al. 2011; Szulwach et al. 2011; Wu et al. 2011a; Y Xu et al. 2011). Antibodies specific for 5mC and 5hmC are also extensively used in immunostaining assays to examine cellular levels of these epigenetic marks in various biological systems, such as zygotes, developing embryos, and adult brains (Ficz et al. 2011; Iqbal et al. 2011; Ruzov et al. 2011; Wossidlo et al. 2011). In addition, Pastor et al. (2011) took advantage of the fact that bisulphite treatment converts 5hmC into cytosine 5-methylenesulphonate (CMS) and developed specific antisera against CMS. As specific antibodies against 5fC or 5caC become available, they can be used in a similar way to determine the cellular levels as well as the genome-wide location of these newly identified 5mC oxidation derivatives.

Chemical labeling of 5mC oxidation derivatives

While the antibody-based detection methods for 5mC oxidation derivatives are convenient to use and advantageous in immunostaining assays, it has been reported that the immunoprecipitation efficiency of 5hmC antibodies are partially dependent on the density of CpG sites (Pastor et al. 2011). Therefore, sparsely distributed 5hmC marks in genomic DNA may not be readily detected using the 5hmC antibody. To further improve the sensitivity of 5hmC detection methods, several groups developed selective chemical labeling approaches for 5hmC by exploiting the naturally occurring glucosylation process of 5hmC by β-glucosyltransferase (β-GT) in T-even bacteriophages (Fig. 2D; Szwagierczak et al. 2010; Pastor et al. 2011; Song et al. 2011). Such strategy allows for the addition of a glucose moiety to the hydroxyl group of 5hmC using β-GT to yield β-glucosyl-5-hmC. For instance, Song et al. (2011) used purified β-GT to transfer a custom-synthesized UDP-glucose analog (UDP-6-N₃-glucose) onto 5hmC. With an azide group present in the chemically modified glucose, a biotin group or other tags can be attached to 5hmC using click chemistry for various applications (Song et al. 2011). A similar approach, termed GLIB (glucosylation, periodate oxidation, and biotinylation), was developed by Rao and colleagues (Pastor et al. 2011). In vitro binding assays suggest that these chemical labeling methods may possess enhanced sensitivity for detecting 5hmC marks sparsely distributed in genomic DNA (Pastor et al. 2011). Importantly, chemical labeling strategy can also be used to add biotin tags to 5fC and 5caC (Ito et al. 2011; Pfaffeneder et al. 2011), providing a potential method to compare the relative abundances of different 5mC oxidation derivatives.

Single-molecule real-time (SMRT) sequencing

While modification-specific antibody or chemical labeling methods can detect the presence of 5mC oxidation derivatives within genomic DNA fragments (200- to 500-base-pair [bp] resolution), neither of the above approaches provides the modification status of cytosine at single-base-pair resolution. Thus, it is challenging to precisely map 5mC and its derivatives when these modified bases are within the same DNA fragments. A recent study has reported promising results on the development of SMRT sequencing, which uses the capacity of DNA polymerase to incorporate C, 5mC, and 5hmC with different kinetics (Flusberg et al. 2010). While the robustness of this technology as well as whether this technology can discriminate 5hmC from 5fC and 5caC remain to be seen, real-time sequencing technologies provide a potential tool to map the various cytosine derivatives at base-pair resolution without the need for bisulphite treatment.

Iterative 5mC oxidation followed by DNA glycosylase/BER

As discussed above, Tet proteins can generate 5hmC, 5fC, and 5caC through iterative oxidation of 5mC (He et al. 2011; Ito et al. 2011). These 5mC oxidation derivatives can potentially serve as substrates for mammalian DNA glycosylases or deaminases (see below). Indeed, recent studies have demonstrated that thymine DNA glycosylase (TDG) can efficiently excise 5fC or 5caC in the context of CpG sites (He et al. 2011; Maiti and Drohat 2011). Thus, subsequent repair of the resulting abasic site would regenerate unmethylated cytosine. Interestingly, TDG removes 5fC from the G:5fC substrate at a rate even faster (∼40%) than processing the G:T mismatch (Maiti and Drohat 2011), probably due to the weakened N-glycosidic bond between the nucleobase and the sugar (Bennett et al. 2006). In contrast, TDG has essentially no activity for 5hmC (Maiti and Drohat 2011). The robust excision activity of TDG toward 5fC and 5caC may be one explanation as to why 5fC and 5caC are present at very low levels in the genome compared with 5hmC (Ito et al. 2011; Pfaffeneder et al. 2011). In agreement with the proposal that TDG promotes active demethylation, deletion of TDG in mice leads to embryonic lethality and aberrantly elevated DNA methylation levels at a cohort of gene promoters (Cortazar et al. 2011; Cortellino et al. 2011). In contrast, other DNA glycosylases, such as single-strand-selective monofunctional U DNA glycosylase 1 (SMUG1) and MBD protein 4 (MBD4), have no significant activity for excision of 5fC and 5caC from DNA (He et al. 2011; Maiti and Drohat 2011), suggesting that TDG might be the only mammalian DNA glycosylase to have robust excision activity for 5fC and 5caC. Thus, these findings lead to a multistep active DNA demethylation model linking Tet1-mediated iterative oxidation of 5mC and the BER pathway initiated by the TDG DNA glycosylase (Fig. 3). However, it is important to note that although depletion of TDG in mouse ES cells resulted in an increase in the 5caC level (He et al. 2011), the amount of estimated 5caC under TDG-depleted conditions is still below the level of 5fC under normal conditions (He et al. 2011; Ito et al. 2011). Given the high efficiency of full-length Tet2 in converting 5mC to 5caC (He et al. 2011), it is anticipated that 5caC would accumulate at a much higher level under TDG depletion conditions if TDG is the only enzyme capable of processing 5caC. Therefore, it is possible that additional enzymes that possess 5caC processing activity exist in mammalian cells.

5hmC deamination followed by DNA glycosylase/BER

Another potential active DNA demethylation pathway that can be initiated by Tet proteins involves deamination of 5hmC followed by DNA repair. In this scenario, 5hmC may first be deaminated by the AID (activation-induced deaminase)/APOBEC (apolipoprotein B mRNA-editing enzyme complex) family of cytidine deaminases to produce 5-hydroxymethyluracil (5hmU), followed by 5hmU:G mismatch repair through the action of DNA glycosylases and the BER pathway (Fig. 3). This hypothesis is supported by recent studies demonstrating that DNA glycosylases such as TDG and SMUG1 exhibit robust excision activity against 5hmU:G in dsDNA (Cortellino et al. 2011), whereas they have no or very low 5hmC glycosylase activity in an in vitro biochemical assay (Cortellino et al. 2011; He et al. 2011). Furthermore, studies in human HEK293 cells demonstrated that TDG can directly interact with AID (Cortellino et al. 2011), suggesting that the deamination and DNA glycosylase activity may be coupled.

However, one open question that remains to be addressed is to what extent the AID/APOBEC family of cytidine deaminases is used to deaminate 5hmC to generate 5hmU in vitro and in vivo (Fig. 3). Whereas AID and APOBEC1 are shown to possess the capacity to deaminate 5mC to T in the context of ssDNA in vitro (Morgan et al. 2004), no direct biochemical evidence suggests that these deaminases exhibit robust activity for 5hmC. In addition, mass spectrometric analysis indicates that 5hmU, the predicted intermediate of this pathway, does not accumulate to a detectable level in mammalian cells (Globisch et al. 2011; Pfaffeneder et al. 2011). This indicates that either the deamination reaction does not occur extensively in vivo or 5hmU is extremely short-lived. In agreement with this, although AID may contribute to active PGC demethylation in vivo (Popp et al. 2010), it is probably responsible for only a small portion of this large-scale demethylation process, as substantial demethylation still occurs in the absence of AID. Nevertheless, 5hmC deamination followed by the BER pathway does seem to be a potential mechanism underlying loci-specific active demethylation. For example, a recent study has suggested that Tet1 and AID may cooperate in demethylating a methylated DNA duplex (Guo et al. 2011). In this study, cotransfection of plasmid DNA encoding various AID/APOBEC deaminases with a linearized 5mC- or 5hmC-containing reporter into HEK293 cells resulted in locus-specific demethylation of 5hmC- but not 5mC-containing dsDNA (Guo et al. 2011). Interestingly, Guo et al. (2011) also observed that 5hmC in a non-CpG context is more frequently demethylated than those in CpG dinucleotides (∼9% for 5hmCpHs vs. ∼2% for 5hmCpGs), and 5hmC in the nontranscribed strand seems to be preferentially demethylated.

Iterative 5mC oxidation followed by decarboxylation

Both mechanisms described above involve glycosylases and the DNA repair pathway. Such mechanisms are unlikely to be used in large-scale DNA demethylation processes observed in PGCs and zygotes, as genome-wide repair of mismatches in such a limited time window will put tremendous pressure on genome stability. Therefore, we previously proposed a simple mechanism involving Tet-mediated iterative oxidation followed by decarboxylation (Wu and Zhang 2010). Given that only two enzymes are required to convert 5mC to C in this hypothetical mechanism (Fig. 3), it obviates the need for a DNA strand break that is required for DNA repair-based mechanisms, thereby relieving the pressure of genome instability associated with BER. Despite the simplicity of this proposed demethylation pathway and the precedent of a decarboxylase in thymidine salvage reactions, the existence of the putative decarboxylase capable of removing carboxyl group from 5caC to regenerate unmodified cytosine remains to be revealed.

Role of Tet3-mediated 5mC oxidation in zygotic DNA demethylation

While DNA methylation patterns are relatively stable in somatic cells, rapid erasure of global DNA methylation patterns has been observed in developing PGCs as well as zygotes (Sasaki and Matsui 2008; S Feng et al. 2010; Wu and Zhang 2010). In one-cell zygotes, the 5mC mark in the paternal pronucleus quickly disappears before the first cell division (depicted in orange in the top panel of Fig. 5; Mayer et al. 2000; Oswald et al. 2000), whereas the maternal genome is proposed to be protected from zygotic demethylation by DNA-binding proteins, such as Stella/Dppa3 (Nakamura et al. 2007). Using specific antibodies against 5hmC or 5mC, recent studies have shown that the paternal and maternal genomes of mouse zygotes at late pronuclear (PN4–5) stages are predominantly marked by 5hmC and 5mC, respectively, and that appearance of 5hmC (around PN3 stage) coincides with the loss of 5mC marks in the paternal genome (summarized in the bottom panel of Fig. 5; Iqbal et al. 2011; Wossidlo et al. 2011). The asymmetric marking of 5hmC in the paternal pronucleus is also observed in bovine and rabbit zygotes (Wossidlo et al. 2011), suggesting a conserved role for oxidation of 5mC in the erasure of paternal 5mC immediately after fertilization. Gene expression analysis indicated that Tet3, but not Tet1 and Tet2, is expressed at high levels in oocytes and zygotes, but rapidly decreases at two-cell and later stages (Iqbal et al. 2011; Wossidlo et al. 2011). In agreement with the notion that Tet3 is responsible for 5mC oxidation in the paternal pronucleus, either conditional deletion of Tet3 from the female germ cells or siRNA-mediated down-regulation of zygotic Tet3 impedes the conversion of 5mC to 5hmC in the paternal genome (Gu et al. 2011; Wossidlo et al. 2011). Furthermore, Tet3 protein appears to be specifically enriched in the paternal pronucleus at the zygotic stage (Gu et al. 2011). Subsequent bisulphite sequencing analysis suggests that zygotic Tet3 may be required for active DNA demethylation at specific loci in the paternal genome, including Line1 transposons and gene regulatory regions of pluripotency factors (Oct4 and Nanog), which are hypermethylated in sperm. However, loss of Tet3 does not affect the methylation status of several imprinting loci (Gu et al. 2011). Interestingly, Tet3 is dispensable for germ cell maturation, fertilization, and preimplantation development, but deletion of Tet3 from either PGCs (TNAP-Cre) or growing oocytes (Zp3-Cre) resulted in increased frequency of developmental failure of embryos, possibly due to delayed/attenuated activation of the paternal allele of genes important for embryogenesis (Gu et al. 2011). In addition, somatic cell nuclear transfer (SCNT) using Tet3-deficient oocytes, but not control oocytes, resulted in failure in 5mC oxidation of somatic genomes, impaired demethylation at the somatic Oct4 promoter, and activation of an Oct4-EGFP transgene (Gu et al. 2011). Together, these results suggest that Tet3 acts as a critical factor for zygotic epigenetic reprogramming by initiating the global conversion of 5mC to 5hmC and possibly other oxidation forms.

Interestingly, asymmetric marking of 5hmC in sperm-derived chromosomes persists into two-cell stage embryos (Iqbal et al. 2011), suggesting that 5hmC is not rapidly removed on a genome-wide scale. Thus, it is important to determine the fate of paternal 5hmC during preimplantation development. As discussed above (Fig. 3), 5hmC/5fC/5caC can be actively processed through enzymatic reactions or be passively diluted through DNA replication. In a recent study, careful analysis of high-resolution mitotic chromosome spreads that are positive for 5hmC staining at the one-, two-, four-, and eight-cell stages indicates that 5hmC marks on the paternal chromosomes are gradually lost from the one- to eight-cell stage, and this process appears to be mediated by replication-dependent dilution, as, on average, half of the chromosomes retain the 5hmC mark after each round of cell division (Inoue and Zhang 2011). The above results support the model in which global erasure of paternal 5mC is first initiated by Tet3-mediated conversion of 5mC to 5hmC in the male pronucleus, followed by replication-dependent passive loss of 5hmC during preimplantation development. Similarly, a replication-dependent passive demethylation mechanism has been proposed to be responsible for global loss of 5mC marks on the maternal chromosomes (Rougier et al. 1998). While these findings cannot exclude the possibility that 5mC oxidation derivatives may also serve as intermediates for replication-independent active DNA demethylation pathways (Fig. 3), whether and how specific loci in paternal and maternal genomes undergo actual DNA demethylation (conversion of 5mC/5hmC to C, not just to 5fC or 5caC) in zygotes and preimplantation embryos awaits further investigation, as bisulphite sequencing analysis cannot discriminate 5caC from unmethylated cytosine (He et al. 2011). Of note, a recent study identified the elongator complex as a potential regulator of zygotic paternal genome demethylation (Okada et al. 2010; Wu and Zhang 2010). It will be of interest to determine whether elongator and Tet3 proteins cooperate to mediate the erasure of paternal 5mC. Given that developing PGCs also undergo global erasure of 5mC (Hajkova et al. 2002; Sasaki and Matsui 2008), it will be interesting to determine whether Tet-mediated 5mC oxidation is also involved in PGC demethylation.

Role of Tet1 in regulating pluripotency and lineage differentiation of mouse ES cells

Concomitant with the rapid reduction of Tet3 expression at the two-cell stage, the expression of Tet1 is rapidly up-regulated at later preimplantation stages (Iqbal et al. 2011; Wossidlo et al. 2011). As the zygote develops into the blastocyst, Tet1 and Tet2 are expressed at high levels in the inner cell mass (ICM). Consistently, Tet1 and Tet2, but not Tet3, are expressed in mouse ES cells. Upon in vitro differentiation of mouse ES cells, Tet1 and Tet2 expression is rapidly decreased, whereas Tet3 expression is up-regulated (Tahiliani et al. 2009; Ito et al. 2010; Koh et al. 2011). Given that both Tet1 and Tet2 are direct downstream targets of a cohort of pluripotency factors (Koh et al. 2011) and that Tet1 binds to gene promoters of some key ES cell transcription factors (Wu and Zhang 2011), Tet proteins are likely to be part of the pluripotency regulatory circuit.

shRNA-mediated Tet1 knockdown in mouse ES cells suggests that Tet1 deficiency leads to a decrease in total 5hmC levels (∼50%) and increased DNA methylation at the Nanog-proximal promoter, which is accompanied by a decrease in Nanog expression and impaired proliferation of ES cells (Ito et al. 2010). Subsequent genome-wide location analyses in mouse ES cells have confirmed that Tet1 and 5hmC are enriched at promoter regions of several pluripotency factors, including Nanog, Tcl1, and Esrrb (Wu and Zhang 2011). RNA-seq analysis of Tet1 and Tet2 double knockdown mouse ES cells also showed that several genes related to pluripotency are down-regulated in the absence of Tet proteins (Ficz et al. 2011). However, other studies using different sets of shRNAs to down-regulate Tet1 and Tet2 suggest that Tet1/2 deficiency does not affect the expression of pluripotency factors and mouse ES cell proliferation (Koh et al. 2011; Williams et al. 2011). These discrepancies between in vitro experiments are possibly due to differences in mouse ES cell background, culturing conditions, and/or off-target effects of shRNAs (Williams et al. 2011; Wu and Zhang 2011). Further analysis of lineage-specific gene expression and teratomas indicate that Tet1 deficiency leads to increased spontaneous differentiation toward trophoectoderm and mesoendoderm lineages (Ito et al. 2010; Ficz et al. 2011; Koh et al. 2011). The above results, together with recent findings on the dual functions of Tet1 in both Polycomb repression of developmental genes and transcriptional activation of pluripotency genes, suggest that Tet1 is potentially required for orchestrating the balance between pluripotency maintenance and lineage commitment.

To further study the function of Tet1 in ES cell maintenance and in vivo development, Jaenisch and colleagues (Dawlaty et al. 2011) recently generated Tet1-null mouse ES cells and mice. In Tet1-null mouse ES cells, a decrease in 5hmC levels (∼35%) is observed and 221 genes are found to be significantly dysregulated. However, Tet1-null ES cells maintain a normal morphology, express proper levels of key pluripotency factors, and support embryonic development in a tetraploid complementation assay (Dawlaty et al. 2011). While Jaenisch and colleagues (Dawlaty et al. 2011) also observed altered expression of lineage-specific marks (Brachyury and Pax6) in differentiating mutant ES cells and skewed differentiation toward trophectoderm in Tet1-null teratomas, Tet1 mutant ES cells fail to exhibit similar phenotypic defects in vivo. These results suggest that in vitro differentiation defects exhibited in mutant ES cells and teratoma assays are less pronounced in the context of embryonic development in vivo. Further analysis of Tet1-null mice indicate that mutant mice are viable, but 75% of the homozygous mutant pups display a smaller body size at birth, suggesting loss of Tet1 leads to a developmental delay. Moreover, both male and female germ cells are generated in homozygous mutant embryos, but the number of progenies derived from the intercross of Tet1-null males and females are significantly reduced, indicating that either a subset of Tet1-null mice are embryonic lethal or germ cell development is impaired in homozygous Tet1-null mice (Dawlaty et al. 2011). Given that Tet1 is highly expressed in developing PGCs (Hajkova et al. 2010), further investigations are required to elucidate the role of Tet1 in epigenetic reprogramming of PGCs and gametogenesis.

Role of Tet proteins in brain development and activity-dependent DNA demethylation

Brain-specific deletion of DNA methyltransferases results in premature death, neural developmental defects, and impaired neuronal functions (Fan et al. 2001, 2005; Nguyen et al. 2007; J Feng et al. 2010; LaPlant et al. 2010; Wu et al. 2010). Loss-of-function mutations of the X-chromosome-linked gene encoding MECP2 causes Rett syndrome, an autism spectrum disorder predominantly found in girls (Amir et al. 1999). These findings suggest that precise control of DNA methylation patterns is required for proper brain development. Among various mouse cell types and tissues, neurons in the CNS appear to contain the highest levels of 5hmC, almost 10 times higher than that of mouse ES cells, suggesting that Tet proteins may play a critical role in brain development and maturation.

Using a chemical labeling strategy to determine the total amount of 5hmC in the mouse cerebellum at different stages of development, Song et al. (2011) report a gradual increase from postnatal day 7 (P7; 0.1% of all nucleotides) to adult stage (0.4%). They also determined the genome-wide distribution of 5hmC in P7 and adult cerebellum (Song et al. 2011). This analysis identified 5425 genes acquiring 5hmC during aging. Among them, genes related to age-dependent neurodegenerative disorders, angiogenesis, and hypoxia response are enriched. Previous studies show that active DNA demethylation at specific gene promoters (e.g., Bdnf and Fgf1) may be required for activity-dependent adult neurogenesis (Ma et al. 2009; Wu and Sun 2009). A more recent study indicates that Tet1 may contribute to this process by initiating 5mC hydroxylation followed by AID/APOBEC-mediated deamination and BER (Guo et al. 2011). These results raise the possibility that 5hmC may play a critical role in postnatal neural development, age-related neurodegeneration, and transcriptional regulation of activity-dependent neural plasticity.

Role of Tet2 in hematopoiesis and leukemia

Aberrant DNA methylation pattern is one of the hallmarks of cancer cells (Esteller 2008; Gal-Yam et al. 2008). In general, cancer cells display global hypomethylation and promoter hypermethylation of tumor suppressor genes. Consistently, accumulating evidence suggests that somatic mutations in DNA methyltransferases and 5mC-modifying enzymes, such as TET proteins, are associated with oncogenic transformation. For instance, mutations in the de novo DNA methyltransferase DNMT3A were recently found in a significant fraction of patients with AML (Shah and Licht 2011; Yan et al. 2011).

The first evidence implicating TET proteins in tumorigenesis was the identification of TET1 as a rare fusion partner of MLL in patients with AML (Ono et al. 2002; Lorsbach et al. 2003). The chromosome 4q24 harbors the human TET2 gene and displays recurrent microdeletions and copy-neutral loss of heterozygosity in patients with myeloid malignancies (Viguie et al. 2005). In 2009, two studies identified somatic TET2 mutations in patients with myeloproliferative neoplasms (MPNs) and myelodysplastic syndromes (MDSs) (Delhommeau et al. 2009; Langemeijer et al. 2009). Subsequent investigations of larger cohorts of leukemia patients suggest that TET2 mutations are frequently observed in a diverse spectrum of myeloid malignancies, including MDS (19%–26%), MPN (12%–37%), chronic myelomonocytic leukemia (CMML; 20%–50%), de novo AML (7%–23%), and secondary AML (sAML) (Abdel-Wahab et al. 2009; Delhommeau et al. 2009; Langemeijer et al. 2009; Tefferi et al. 2009a,b; Abdel-Wahab 2011; Pronier et al. 2011). Most recently, recurrent TET2 mutations were also identified in B- and T-cell lymphoma (Quivoron et al. 2011). For detailed discussion of clinical aspects of TET2 mutations in hematopoietic malignancies, please refer to a recent review by Aifantis and colleagues (Cimmino et al. 2011).

To better understand the mechanism by which somatic TET2 mutations contribute to leukemia, several studies examined the effect of disease-associated mutations on TET2 catalytic activity and global epigenetic profiles. Ko et al. (2010) demonstrated that TET2 mutations found in patients with myeloid malignancies impair its enzymatic activity. Consistently, genomic DNA derived from patient bone marrows with TET2 mutations contains significantly lower levels of 5hmC compared with that from healthy controls (Ko et al. 2010). Interestingly, samples from patients with low 5hmC levels also display a hypomethylated state relative to controls at the majority of differentially methylated CpG sites (Ko et al. 2010). However, a recent integrative genetic and methylation analysis of a large, homogeneous cohort of AML patients showed that TET2 mutations are predominantly associated with a DNA hypermethylation state (Figueroa et al. 2010). The discrepancies between these two studies are probably due to differences in disease subtypes (diverse myeloid malignancies vs. de novo AML only) and methylation profiling platforms (Illumina Infinium methylation assay vs. HpaII tiny fragment enrichment by ligation-mediated PCR [HELP] assay). Given that both methylation assays used in previous studies only assess a subset of CpG sites at gene promoter regions, unbiased genome-wide mapping of 5mC and 5hmC distribution is required to better understand the extent of alterations in DNA methylation associated with TET2 mutations.

Interestingly, recent mutational analyses of AML patients have shown that TET2 and the isocitrate dehydrogenase gene IDH1/2 mutations are mutually exclusive (Figueroa et al. 2010). IDH1/2 are NADP-dependent enzymes that catalyze the conversion of isocitrate to 2OG. Recurrent somatic mutations of IDH1/2 are frequently found in glioma and AML patients (Parsons et al. 2008; Mardis et al. 2009; Yan et al. 2009; Gross et al. 2010; Marcucci et al. 2010; Ward et al. 2010). Subsequent studies showed that mutant IDH enzymes predominantly acquire a neomorphic function to produce 2-HG, an oncometabolite that impairs catalytic activity of many Fe(II)- and 2OG-dependent dioxygenases, including the JmjC family of histone demethylases (Klose et al. 2006) and the Tet family of DNA hydroxylases. 2-HG inhibits the enzymatic activity of these enzymes by competing with their cosubstrate, 2OG (Dang et al. 2009; Gross et al. 2010; Ward et al. 2010; W Xu et al. 2011). These studies suggest that IDH1/2 mutations may alter DNA methylation patterns by inhibiting TET2 enzymatic activity. Indeed, methylation profiling of samples from AML patients with IDH1/2 mutations has demonstrated that patient samples exhibited elevated DNA methylation at differentially methylated regions when compared with control samples (Figueroa et al. 2010). Collectively, these results suggest that loss-of-function TET2 mutations and neomorphic IDH1/2 mutations in hematopoietic malignancies may share a common disease mechanism and pathoetiology that lead to alteration in 5hmC and 5mC patterns (Fig. 6). Thus, therapies modulating catalytic activity of TET proteins might be beneficial for the treatment of patients with TET2 mutations.

Frequent mutations of TET2 in leukemia patients suggest that TET2 may function as a physiological regulator of hematopoiesis. To elucidate the functions of TET2 in hematopoietic differentiation and homeostasis, several groups have used various genetic approaches to investigate the effect of Tet2 deficiency on normal hematopoiesis. The approaches include shRNA-based Tet2 knockdown and targeted deletion of Tet2 in mice (Figueroa et al. 2010; Ko et al. 2010, 2011; Li et al. 2011; Moran-Crusio et al. 2011; Quivoron et al. 2011). Despite different phenotypic results between reports using in vitro knockdown of Tet2 (Figueroa et al. 2010; Ko et al. 2010; Moran-Crusio et al. 2011), all Tet2-null mouse models generated highly similar phenotypes. Although all three Tet family proteins are expressed in hematopoietic systems, deletion of Tet2 alone is sufficient to cause significant loss of 5hmC in genomic DNA. Importantly, Tet2 deletion led to gradual enlargement of the stem/progenitor cell pool in a cell-autonomous manner, as shown by increased expression of stem cell marker genes (e.g., c-Kit) as well as markedly elevated self-renewal capacity. Intriguingly, all Tet2-null mouse models progressively develop myeloid neoplasms, despite minor differences in disease onset time and progression kinetics. Consistent with the high frequency of TET2 mutations in human CMML, Tet2-null mouse models exhibited predominantly CMML-like disease. Furthermore, functional analysis showed that Tet2 haploinsufficiency is sufficient to confer increased self-renewal to stem/progenitor cells and promote myeloproliferation in vivo (Moran-Crusio et al. 2011; Quivoron et al. 2011), suggesting clinically relevant gene dosage effects. Thus, Tet2 plays a critical role in regulating normal hematopoietic differentiation and homeostasis, and deletion of Tet2 is sufficient to initiate myeloid transformation.