Transcriptional and epigenetic networks in haematological malignancy

Identification of transcription factors as prevalent targets affected by recurring chromosomal translocations has provided the first hint for the importance of transcriptional deregulation in haematological malignancies. However, the actual molecular functions of these leukaemia‐associated transcription factors on gene expression remained largely unknown until the recent discovery of their association with specific enzymatic activities that modify epigenetic codes (at DNA and/or histone levels) of downstream transcriptional targets. Intriguingly, while only just about half of acute leukaemia associates with recurring translocations, emerging evidence indicates that cryptic mutations identified in the “normal‐karyotype” leukaemia also frequently affect components of epigenetic machinery. We will review these recent findings and discuss their implications in understanding the biology of the disease and in development of effective cancer therapeutics.


Introduction
Recurrent chromosomal translocations that result in the formation of chimaeric fusion proteins are pivotal for the initiation and progression of human cancers, and in particular, in haematological malignancy [1][2][3]. In human leukaemia, many of the oncogenic fusion proteins encode transcription factors with aberrant transcription activities leading to dysregulation of their critical target genes and downstream transcriptional programmes that drive the oncogenesis. Targeting leukaemic fusion protein has led to few but significant therapeutic achievements in the past few decades for the treatment of human malignancy. In patients suffering from acute promyelocytic leukaemia (APL), differentiation therapy using all trans-retinoic acid (ATRA) and arsenic trioxide to induce terminal differentiation and apoptosis of leukaemic cells harbouring PML-RARa (promyelocytic leukaemia-retinoic acid receptor a) leukaemic fusion has transformed it from a fatal to a highly curable disease [4]. Similarly, the introduction of imatinib mesylate (Gleevec; STI571), a small molecule inhibitor for kinase activity of the BCR-ABL fusion in chronic myeloid leukaemia (CML), has converted a fatal cancer into a manageable chronic condition in patient [5]. However, the treatment for the majority of human leukaemias still relies heavily on the conventional chemotherapy with associated side effects and disease prognosis remain dismal for many subtypes of haematological malignancies such as MLL leukaemia. Hence, a better understanding of the underlying leukaemogenic mechanisms is critical for invention of rational therapeutic approach for the eradication and/or control of these highly malignant diseases. Emerging evidence indicates that an array of epigenetic enzymes plays important roles in the leukaemic transformation, and potentially represents a plethora of amendable targets for therapeutic intervention. In this review, we will examine and discuss the implications of some of the recent discoveries that important epigenetic machineries are hijacked or directly mutated in human leukaemia.

Histone code: the writers and the readers
The role of epigenetic modification, particularly DNA methylation, in the regulation of gene expression in normal and in cancer cells has been extensively studied in the past few decades. Recent advances in sequencing technology and the availability of highly specific antibodies have greatly facilitated progress in the study of the epigenetic landscape at both the nucleic acid and at histone levels [6][7][8]. Different covalent post-translational histone modifications of the histone octamers generates a myriad combination of histone marks, which are collectively termed the 'histone code' [9,10]. These modifications are made by a large array of histone ''writers'' with distinct enzymatic activities. Histones can be methylated, ubiquitinated, phosphorylated, proline-isomerized, ADPribosylated or modified by other means such as citrullination and by proteolytic cleavage [10][11][12][13]. These histone marks can be read and interpreted by other transcriptional regulators or epigenetic modifiers, which possess motifs that are capable of specifically interacting with the modified histones, e.g., bromodomain recognize acetylated lysine, plant homeodomain (PHD) finger and chromodomain bind methylated lysine [9,14,15]. These transcriptional complexes are usually comprised of multi-modular components that contain different copy numbers and combinations of histone recognition motifs to facilitate their interaction with histones [15].

Basic principles of the histone code
Histone modification is a dynamic and reversible process regulated by various histone-modifying enzymes that can cooperate with or antagonise each other (Fig. 1). Of the complexity of covalent histone modifications, there are some basic principles regarding the functional consequences of certain histone marks [11,16,17]. Histone lysine acetylation mediated by histone acetyltransferases (HATs) is mostly associated with active transcription. The action of these enzymes can be reversed by histone deacetylases (HDACs) leading to histone hypoacetylation and is usually associated with transcriptional suppression [18,19]. Methylation of histone tails is much more complicated, as lysine residues can be mono-, di-or tri-methylated, whereas arginine residues can be mono-, di-, symmetrical or asymmetrically methylated [11,20]. These methylated lysine marks can be erased by the lysine specific demethylase 1 (LSD1) and Jumonji domain-containing family of histone lysine demethylases (KDM) [21][22][23] (Figs. 1 and 2). Some histone methylation marks are strongly associated with transcriptional activation, such as H3K4 trimethylation (H3K4me3) mediated by the MLL complex or H3K79 dimethylation (H3K79me2) mediated by the Dot1 complex [24,25]. The H3K27 trimethylation (H3K27me3) mark, mediated by Polycomb Repressor Complex 2 (PRC2), is critical for transcriptional repression of genes, in particular those essential for development and maintenance of cellular pluripotency [26]. H3K9 methylation is another histone modification usually associated with transcriptional repression and the formation of constitutive and stably repressed heterochromatic regions, such as the inactivated Xchromosome and in centromeric heterochromatin [27,28]. However, the demarcation of transcription status based on the histone modification is unclear in genes in the bivalent chromatin state where bivalent genes are usually associated with the presence of both active H3K4 trimethylation and the repressive H3K27 trimethylation marks in embryonic stem cells [26]. In this configuration, transcription activity is at a basal level, however, the bivalent gene is at a poised state whereby it can rapidly respond to developmental cues and other stimuli [26]. Although it is not clear if the same principle applies to somatic stem cells, studies using mixed populations of haematopoietic stem/progenitor cells show that many cell fate decision genes are held in a transcriptionally poised bivalent state until their commitment to differentiate [29].

Leukaemic fusions as aberrant transcriptional and epigenetic regulators
Cytogenetic analyses and positional cloning have significantly contributed to the identification of numerous chromosomal translocations in human leukaemias and other cancers, which has now been further facilitated by the technological advances in whole genome sequencing [30][31][32]. Oncogenic chimaeric transcription factors resulting from reciprocal translocations account for about half of acute leukaemias [33]. Abnormal activity of the resulting leukaemic chimaeric fusion protein can lead to the deregulation of the target genes of their normal counterparts by altering the chromatin status via aberrant recruitment of epigenetic regulators. We can broadly classify chimaeric transcription factors into activating and repressive leukaemic transcription factors based on their predominant transcriptional activities on their direct downstream target genes, which lead to distinct gene expression profiles [34,35] (Fig. 3).
Repressive leukaemic transcription factors transform haematopoietic cells through the recruitment of the co-repressor complexes that result in the gain of repressive histone marks accompanied by the loss of active marks. PML-RARa fusion is the best-characterised repressive leukaemic fusion protein. It transforms haematopoietic cells mostly though repression of its target genes resulting in an accumulation of abnormal promyelocytes [4]. Its aberrant transcription repressive function is mediated by: (1) the acquisition of the ability to homo-oligomerise [36,37]; (2) the recruitment of DNA binding cofactors such as RXRa [38,39]; and more importantly, (3) its specific association with co-repressor complexes such as NCoR/HDAC, DNA methyltransferase DNMT3A and PRC2 [40][41][42]. The RARa moiety of this oncogenic fusion protein is essential for its interaction with co-repressor complexes and is a focal point for drug targeting. Although the key responsible epigenetic enzymes remain unknown, the binding of therapeutic drug ATRA to RARa releases the co-repressor complex and induces terminal differentiation of the leukaemic cells by de-repression of target genes [40,41]. In contrast, activating leukaemic transcription factors, generally are involved in the aberrant recruitment of epigenetic enzymes that deposit active histone marks and prevent the accumulation of repressive marks. Mechanisms for the recruitment of co-repressor complexes by repressive transcription factors have been reviewed in the past, thus the current article will mainly focus on leukaemic fusion proteins, which predominantly activate transcription [4,41,43]. HOX, which have an essential role on the orchestration of the embryonic development, are one of the most frequently dysregulated targets in human leukaemias and other malignancies [44,45]. HOX genes such as HOXA9, HOXA10, HOXA11, HOXC13, HOXD3 or HOXD13 have been found in translocations involving NUP98 to generate chimaeric fusions in leukaemia patients [46]. Over-expression of HOX genes is frequently found in leukaemia patients, particularly in those chromosomal translocations, which involve MLL and MOZ [34,47] (discussed in the next section). HOXA9  Activating leukaemic transcription factors activate their critical target genes by recruiting histone-modifying enzymes that leave activation histone marks and prevent the accumulation of repressive marks. Histone acetylation is mediated by histone acetyltransferases (HATs) such as CBP, p300 and MOZ. Increased active lysine methylation marks can be induced by lysine methyltransferases (KMTs), e.g. DOT1L, NSD1 and MLL. Repressive lysine methylation marks may be removed by lysine demethylases (KDMs) such as JMJD3/UTX (H3K27) and JMJD2 (H3K9) family members. Arginine methylation (H4R3) mediated by PRMT1 can also facilitate transcriptional activation. Activation complexes may also recruit TET family proteins to remove the repressive promoter CpG methylation. b. Repressive leukaemic transcription factors recruit co-repressor complexes and other histonemodifying enzymes to leave repressive histone marks and impede the deposition of active marks on their target genes. Histone deacetylases (HDACs) erase the active acetylation marks thereby lead to promoter hypoacetylation and transcriptional repression. The JARID1 family and LSD1 family KDMs both contribute to the removal of the active H3K4 methylation marks. H3K9 KMT (G9a, SETDB1) and PRC2 cause increased repressive H3K9 and H3K27 methylation marks, respectively. Gene silencing can be stabilised by monoubiquitination of H2A by the PRC1 complex. Repressor complexes can also recruit DNA methyltransferases (DNMTs) to repress target gene expression by inducing promoter CpG methylation. is one of the single most significant poor prognostic markers for acute myeloid leukaemia (AML) [48]. Knockdown of HOXA9 in human MLL leukaemia cells suppresses leukaemic cell survival in vitro and reduces tumour burden in mouse models [49]. Furthermore, over-expression of Hoxa9, together with its interaction partner Meis1, is sufficient to immortalise haematopoietic progenitor cells and induces acute myeloid leukaemia in vivo [50]. Together these studies strongly support a pivotal role of HOX genes in acute leukaemogenesis.
In embryonic stem cells (ESC), Hox genes are regulated and maintained in the bivalent chromatin state by two antagonistic transcription complexes, namely the activating Trithorax (e.g., MLL family protein) complexes and the silencing PRC1/2 [26,51,52]. MLL protein, which regulates Hox gene expression via its H3K4 methyltransferase activity, is also required for normal embryonic development and haematopoiesis [53,54]. The methylated H3K4 residue can be removed by the KDM1/LSD1 and KDM5/ JARID1 demethylase family [21][22][23] (Figs. 1 and 2). PRC2 comprises of several core components that mediate gene silencing by trimethylation of H3K27 [55,56] (Fig. 1). The SET domain-containing proteins, EZH1 and EZH2, are the catalytic subunits for H3K27 histone methylation whereas SUZ12 and EED are essential for the activation of EZH enzymatic activity [55][56][57]. PRC1, which comprises the E3 ubiquitin ligase RING1/RING1B, BMI1 and other components, are essential for mono-ubiquitination of H2A and maintenance of the repressive state [55,58]. PRCs are recruited to their target genes through binding to specific DNA sequence elements known as the Polycomb Response Element [55,59]. Recently, Jarid2, the founding member of Jumonji domain-containing protein, was found to be a component of the PRC2 and is involved in fine-tuning the activity of the complex [60][61][62][63][64]. The repressive H3K27me3 mark on HOX genes can be specifically erased by the histone lysine demethylases, JMJD3 and UTX, and provide an additional regulatory pathway to antagonise PRC-mediated gene silencing [65][66][67] (Figs. 1 and 2). H3K27 demethylation can be followed by CBP/p300-mediated histone acetylation to further reinforce the transcription activation status [68]. In leukaemia, this dynamic epigenetic regulation of HOX genes is altered by a number of fusion proteins such as MLL, MOZ and NUP98 leukaemia fusions.

MLL gene rearrangement and HOX deregulation
The wild-type MLL protein, which has a C-terminal SET domain that methylates H3K4 to activate transcription of target genes such as HOX genes, forms a macromolecular nuclear transcription complex with other core components (ASH2, WDR5, RBBP5, LEDGF, MENIN, etc.) [69][70][71][72]. MLL located at chromosomal 11q23 is one of the most frequently mutated genes in human leukaemia. MLL can be de-regulated in four different ways in leukaemia, namely by gene amplification, partial tandem duplication (PTD), internal deletion or through fusion to a partner gene (Fig. 4). MLL-PTD mutations are observed in approximately 8% of AML patients with normal cytogenetics and in 25-54% of AML patients where trisomy 11 is the only identified abnormality [73,74]. MLL-PTD involves an in-frame partial duplication of the exons encoding the DNA binding domain (encompassing the AT hook and CXXC domain) of MLL. In the MLL-PTD knockin mouse model, MLL-PTD cells upregulate Hoxa9 expression with a concomitant increase in H3K4 methylation and H3 acetylation at the locus, suggesting a dominant gain-of-function effect on transcription activation [75]. HOXA9 overexpression can also be detected in patients with the 11q23/MLL gene amplification, which accounts for approximately 1% of AML and some myelodysplastic syndromes (MDS) patients [76,77]. In contrast to MLL internal deletion, which can only be found in rare cases of T-cell acute lymphoid leukaemia (T-ALL), MLL fusion proteins, which account for 5-10% of adult acute leukaemia, 25% of therapy related leukaemia, and around 80% of infant leukaemia, are one of the best studied leukaemias with a remarkable number of publications detailing their biochemical mechanisms, mouse models and clinical data [78][79][80][81][82]. In MLL leukaemia patients, the translocation usually involves only a single copy of MLL and the wild-type MLL allele is presumably functional in the leukaemic cells. This remaining wild-type MLL is required for the H3K4 trimethylation of Hoxa9. MLL leukaemic fusion is unable to immortalise haematopoietic cells where the MLL gene has been down-modulated by knockout or shRNA approaches [83].
MLL is a promiscuous gene that can fuse with over 50 different partner genes of both nuclear and cytoplasmic proteins [80,[84][85][86]. This results in the generation of chimaeric MLL proteins with gain of transactivation functions provided by the fusion partners. The translocation breakpoints of MLL usually cluster between the exons encoding the CXXC-RD2 motif and the PHD fingers (Fig. 4). This leads to the retention of some essential motifs of MLL N-terminus, such as the DNA-binding AT-hook motif, and the deletion of MLL C-terminus which contains the PHD domain and SET domain, which specifically methylates H3K4 [87]. The position of the breakpoint cluster region of MLL is critical for the oncogenic activity of MLL fusion proteins. This was first illustrated by the synthetic inclusion of the third PHD finger motif (PHD3), which is invariably absent from the MLL fusion proteins, and results in compromised transformation [88,89]. Structure-function studies of the PHD3 with its adjacent bromodomain, revealed a critical functional recruitment of cyclophilin 33 (CyP33) for the induction of a specific conformational change in MLL by catalysing the cis-trans isomerisation of proline in the linker between PHD and bromodomain of MLL which is required for co-repressor complex recruitment [90]. Interestingly, the internal deletion of the MLL PHD finger reported in T-ALL suggests that this loss-of-function mutation could be leukaemogenic [91]. Recruitment of epigenetic regulators mediated by the MLL N-terminus has been recently reported by Muntean et al., whereby they identified the recruitment of the polymerase associated factor complex (PAFc) by the CXXC and RD2 motif of MLL. PAFc plays a critical role in mediating H2B ubiquitination, which has been shown to promote H3K4 and H3K79 methylation for transcriptional activation [92,93]. Recruitment of the PAFc histone modifying complex is essential for HOX gene activation and leukaemic transformation [92].

Deregulation of the epigenetic writers
Leukaemic transcription factors acquire novel functions provided by the fusion partners. For many leukaemic transcription factors, this gain of function is attributed to the acquisition of additional epigenetic modifiers or writers that modify histone marks on their target genes.

Non-SET domain lysine methyltransferase-DOT1L
Among the bewildering number of MLL fusion partners, recent studies have highlighted some unified themes regarding their leukaemogenic mechanism (Fig. 5). The first histone modifying enzyme recruited by MLL fusion partner was reported by Okada et al. [94]. They demonstrated that DOT1L, a histone methyltrasferase that specifically methylates lysine 79 of histone H3, binds to AF10, a fusion partner of both MLL and CALM [94,95]. Loss-of-function of DOT1L in cells expressing these leukaemic fusions compromised their oncogenic transformation ability [94][95][96]. The involvement of DOT1L in MLL fusion-mediated transformation was further underscored by the identification of its potential association with multiple MLL fusion partners including ENL, AF4/AFF1, LAF4, AF5q31/AFF4 as well as P-TEFb components, which phosphorylates the CTD domain of RNA polymerase II [97,98]. Methylated H3K79 is associated with transcriptional activation and is highly enriched in transcribed regions [6,25]. An increased H3K79 methylation mark was consistently observed in many critical MLL target genes in both leukaemic cell lines and patient samples carrying MLL fusions [99][100][101]. However, co-purification experiments using AF4/AFF1 isolated the AEP complex (AF4/ENL/ P-TEFb) and AF5q31/AFF4 but not DOT1L [102]. Interestingly, another MLL fusion partner, ELL1, recruited elongation complexes containing AF4/AFF1, AF5q31/AFF4, AF9, ENL and P-TEFb, but not DOT1L [103]. The discrepancy in the presence or absence of DOT1L in different MLL fusion complexes implies that there are at least two distinct types of fusion complexes involving these elongation factors, i.e., DOT1L or non-DOT1L containing complexes (Fig. 5). It remains doubtful whether the recruitment of DOT1L is sufficient for MLL fusion-mediated transformation as the transforming activity of MLL-DOT1L fusion is disputed [94,102]. It is possible that the recruitment of DOT1L is a direct effect of the MLL fusion protein or it is just coincidental to the initiation of transcription elongation. Even though DOT1L and its H3K79 dimethylation marks are present at most of MLL fusion target gene loci, its functional role in the MLL fusion-mediated leukaemogenesis may require further investigation [99,100].

Protein arginine methyltransferase (PRMT)
In addition to lysine methylation, recent studies also indicate that MLL fusions regulate its target gene expression though histone arginine methylation. Arginine residues are methylated by protein arginine methyltransferases (PRMTs), where type I and type II PRMTs catalyse asymmetric and symmetric dimethylation, respectively [104,105]. PRMT-mediated histone arginine methylation is essential for transcriptional regulation [106]. PRMT1, the major type I arginine methyltransferase involved in various biochemical processes, asymmetrically dimethylates its substrates, which harbour glycine-arginine rich motifs and arginine 3 of histone 4 (H4R3) and is associated with transcriptional activation [107,108]. CARM1/PRMT4 is recruited by nuclear receptors to activate their target genes through histone arginine methylation and/ or by methylation of the splicing complexes [109,110]. Arginine methylation has been shown to cooperate with other epigenetic modifications such as lysine acetylation involved in the transcriptional regulation. Recently, the identification of the mutually exclusive nature of symmetrical dimethylation of H3R2 meditated by PRMT6, and MLL-mediated H3K4 methylation further highlights the complexity of histone crosstalk involving arginine methylation [111][112][113].
The critical role of PRMT function in leukaemogenesis was first reported in the study of the MLL-EEN fusion which transforms haematopoietic progenitor cells and induces leukaemia in vivo [114]. Structure-function analyses demonstrated that SH3 domain of EEN was the minimal transforming domain, which mediated the interaction with Prmt1 and three major arginine methylated RNA binding proteins i.e., SAM68, NONO and PSF [115][116][117] (Fig. 5). SAM68, which is also a PRMT1 substrate, belongs to the KH-domain containing RNA-binding protein family implicated in RNA splicing regulation and signal transduction [118]. PSF and NONO heterodimerise and interact with non-coding RNA (ncRNA) NEAT1 to form the subnuclear ribonucleoprotein body known as paraspeckle [119]. SAM68, which serves as an adaptor to link PRMT1 and CBP for H4R3 arginine methylation and histone acetylation, respectively, was essential for MLL-EEN-mediated transformation and target gene expression [114]. Importantly, the direct fusion of PRMT1, but not the catalytic mutant, to MLL could transform haematopoietic progenitor cells and therefore strongly supported the role of PRMT1 in leukaemogenesis [114]. Similar to lysine methylation, arginine methylation is an reversible process in which methylated arginine can be antagonised by a deimination reaction mediated by PADI4, or erased by arginine demethylase JMJD6 [120,121]. However, their physiological roles in epigenetic regulation and leukaemogenesis remain poorly understood.
The functional significance of the recruitment of PSF and NONO in MLL-EEN-mediated leukaemogenesis is still unknown. Interestingly, both PSF and NONO have been identified in oncogenic translocations in renal cell carcinoma whereby their N-terminal RNA recognition motifs (RRM) were fused with the DNA-binding domain of transcription factor, TFE3 [122]. The gain of RNA-binding activity by these transcription factors may facilitate oncogenesis by enhancing the coupling of transcription and splicing, or alternatively, enable their interaction with other RNA species for transcriptional or epigenetic regulation. The functional importance of ncRNA in regulating Hox expression and metastasis is highlighted by a recent discovery of the long intervening ncRNA HOTAIR originating from the HOXC locus [123][124][125]. HOTAIR forms a ribonucleoprotein complex with the PRC2 and the LSD1/CoREST/REST repressor complexes, and represses HOXD gene expression through epigenetic silencing. However, the role of ncRNA in leukaemogenesis has not been explored and it is unclear whether the novel RNA-binding activity acquired by oncogenic fusion proteins can facilitate transcriptional activation via RNA-mediated epigenetic regulator recruitment.

SET domain containing lysine methyltransferase-NSD family
The nuclear-receptor-binding SET-domain-containing (NSD) family of proteins (NSD1, NSD2 and NSD3) are histone lysine methyltransferases that harbour the SET domain, PHD and PWWP histone recognition motifs [126,127]. Aberration of NSD1 has been reported in both AML and in Sotos syndrome, which has predisposition to cancer [128]. NSD1 specifically trimethylates lysine 36 of histone H3 and is generally correlated with actively transcribed regions [6]. NSD1 is directly fused with the nucleoporin NUP98 gene in AML [129]. Retroviral transduction of NUP98-NSD1 transforms mouse haematopoietic progenitor cells and induces AML in vivo [126]. The immortalisation activity of this fusion is dependent on the H3K36 HMT activity of the NSD1 moiety and may mediate the transcriptional activation of hoxa genes [126]. However, the status and pathological role of Hox genes in the associated human disease has not been investigated. In multiple myeloma (MM), recurrent chromosomal translocations result in the placement of the multiple myeloma-associated SET (MMSET/WHSC1/NSD2) gene under the regulation of the IgH enhancer element leading to its transcriptional dysregulation [130]. Aberrant expression of NSD2 is critical for the oncogenicity of MM cells as its suppression by gene knockdown and knockout inhibited tumour cell growth in vitro and reduced tumour formation of MM xenografts [131]. NSD3 is also a fusion partner to NUP98 in AML and is amplified in breast cancer cells. Together, these findings suggest that the gain of function associated with the NSD SET domain may represent a common oncogenic mechanism [132,133].

Lysine acetyltransferase and its crosstalk with PRMTs
Several members of the HAT family are involved in the chromosomal translocations in leukaemia patients (Fig. 6). Both CBP and p300 reciprocally translocate to MLL, which results in an in-frame fusion containing a large proportion of the CBP/p300 protein [134][135][136]. Structure-function analyses using the MLL-CBP fusion identified both the bromodomain and HAT domain of CBP are required for transformation of haematopoietic cells [137]. Conditional MLL-CBP knockin mice develop myeloproliferative disorders and leukaemia upon c-irradiation or chemical mutagen treatment [138]. In AML, both CBP and p300 have been shown to be involved in fusions with MOZ, a member of MYST family HAT protein required for maintenance of haematopoietic stem cell and hox gene regulation [139][140][141][142][143]. The chimaeric MOZ fusion protein consists of the HAT domain of MOZ and the majority of the CBP/p300 protein. Indirect recruitment of CBP was also reported to be associated with the MOZ-TIF2 leukaemic fusion whereby MOZ is fused with the transactivation domains of TIF2 (Transcription intermediary factor 2) [144,145]. MOZ-TIF2 transformed haematopoietic progenitor cells in vitro and was able to induce AML in vivo. The leukaemogenic activity of the fusion is dependent on both the HAT activity of MOZ and the recruitment of CBP by the AD1 transactivation domain of TIF2 [145]. These leukaemic fusion proteins induce histone hyperacetylation and result in increased Hox gene expression [145,146] (Fig. 6). Crosstalk of CBP mediated histone acetyla-tion and arginine methylation has been reported in nuclear receptor-mediated transcription activation, whereby CARM1 and PRMT1 cooperate with CBP in the activation of p53 expression [147]. Interestingly, this mechanism has seemingly been hijacked by the MLL-EEN leukaemic fusion, which recruits both CBP and PRMT1 to activate Hox gene expression. Recruitment of HAT is not only limited to leukaemic fusions. Wild-type MLL protein also recruits MOF via its C-terminal domain to mediate H4K16 acetylation [148]. Therefore, targeting the activity of different HATs recruited by leukemic fusions or aberrant MLL protein, such as MLL-PTD, may provide another opportunity for therapeutic intervention.

Deregulation of the epigenetic readers
An alternative mechanism contributing to leukaemogenesis is the misreading of the histone code by leukaemic fusions due to the aberrant acquisition of a histone reader [149,150]. This is best illustrated by the leukaemic fusion generated between NUP98 and H3K4 lysine demethylase JARID1A [151]. The resultant fusion encodes the FG repeats of NUP98 and the PHD finger 3 of JARID1A. The resultant NUP98-JARID1A lacks the Jmjc enzymatic domain but shows specific binding to H3K4me2/me3. The specific H3K4 binding property of the PHD finger is required for haematopoietic transformation mediated by NUP98-JARID1A [150]. PHD of the fusion protein may protect those critical target genes from H3K4 demethylation and gene silencing, which is consistent with the detection of increased H3K4 methylation and expression of hoxa9 and Meis1 [150]. The transformation ability of NUP98-JARID1A may also be dependent on CBP and other novel epigenetic mechanisms recruited through the FG repeats of NUP98 to reinforce tran-scription activation as recent studies show chromatin binding of NUP98 in actively transcribed regions [152,153].
The importance of histone recognition motifs in leukaemogenesis is further highlighted by the identification of the essential role of the PWWP motif of Lens-Epithelial Derived Growth Factor (LEDGF) in MLL-mediated leukaemogenesis [72]. Yokoyama et al. show that MENIN interacts with the N-terminus of MLL and serves as an adaptor linking MLL to LEDGF [71,72]. This MENIN-LEDGF complex is essential for MLL fusion-mediated transformation, as the loss of either function compromises transformation [71,72]. Detailed mapping of LEDGF has shown that the methyl-lysine binding PWWP domain is necessary and sufficient to transform haematopoietic cells when fused directly to a non-transforming MLL-AF9 deletion mutant that lacks the ability to recruit the ME-NIN-LEDGF complex [72]. The PWWP domain is a methyl-lysine recognition motif that is also present in other histone modifying enzymes and epigenetic regulators [154,155]. Biochemical analyses of the PWWP domain of BRPF1 (Bromodomain and PHD finger-containing) protein demonstrated specific binding to the trimethylated H3K36 mark, which is known to coordinate events associated with the elongation phase of transcription and is also emerging as an important epigenetic regulator of pre-mRNA processing [154,156]. Interestingly, BRPF1, which is required for Hox regulation during development, was also identified to complex with MOZ [157][158][159]. In addition to the methyl-lysine binding motif, the bromodomain is also necessary for MLL-CBP mediated leukaemogenesis [137]. Thus, understanding how different leukaemic transcription factors read the histone marks on target genes with their specific histone recognition motifs may create new opportunities for therapeutic targeting. The potential cross-talk between MLL and MOZ complexes is also illustrated.

Mutation of epigenetic regulators in haematological diseases
The intricate mechanisms utilised by the different leukaemic fusions in the activation of their target genes underscores the importance of epigenetic regulation during carcinogenesis. Perturbation of the epigenetic network by the acquisition of somatic mutations in some critical histone modifying enzymes can result in the global gene deregulation of multiple signalling pathways that may facilitate clonal evolution of malignant cells. In this section, the mutations affecting the physiological function of the epigenetic regulators will be discussed.

Epigenetic regulators of H3K27 methylation: EZH2 and UTX
Enzymes regulating H3K27 methylation are frequently mutated in haematological malignancies. Somatic mutation of the histone methyltransferase, EZH2, was first detected in follicular and diffuse large B-cell lymphomas of germinal-centre origin [160]. This mutation involves a specific replacement of tyrosine 641 by histidine at the catalytic SET domain that results in the reduced H3K27 methyltransferase activity in vitro [160]. Similarly, EZH2 inactivation mutations were also independently identified in MDS, a heterogeneous group of acquired haematological disorders characterised by dysplasia of the myeloid, erythroid and megakaryocytic lineages [161][162][163]. While the pathological significance and underlying transformation mechanism of EZH2 mutation is still largely unknown at this stage, the loss or reduced PRC2 activity due to EZH2 mutation could potentially disturb the dynamics of epigenetic regulation, in particular, for those engaged in the bivalent chromatin state. Gene expression profiling of leukaemic cells and other human malignancies reveals the similarity of their transcription profiles to embryonic stem cells [164,165]. The majority of bivalent genes lose the active H3K4me3 mark and get silenced by the PRC2-mediated H3K27 methylation after lineage commitment and haematopoietic differentiation [29]. Therefore, defective EZH2 silencing may facilitate the oncogenic transformation of cells by inadvertently allowing the aberrant transcriptional activation of genes involved in self-renewal and blocking haematopoietic differentiation. Alternatively, it is also possible that loss of an antagonistic epigenetic repressor may reduce the epigenetic threshold and facilitate the activation of some critical target genes by oncogenic stimuli, thus making cells harbouring those epigenetic defects more susceptible to transformation.
The genetic data linking EZH2 mutation to haematological malignancy is in stark contrast to those reported in solid tumours [166]. EZH2 is overexpressed in breast and prostate cancer and its overexpression is associated with metastatic progression in prostate cancer [167,168]. Overexpression of EZH2 may augment PRC2-mediated gene silencing and cause obligate transcriptional repression. Obligate gene silencing could also be contributed by the inactivation mutation of an antagonistic H3K27 demethylase, such as UTX, whereby a loss-of-function mutation has been identified in some human myeloma cell lines [169]. Reduced or loss of H3K27 demethylase activity or hyperactive PRC2 activity may lead to the obligate silencing of tumour surveillance genes such as tumour suppressor genes. Deregulation of tumour suppressor genes at the INK4A-ARF loci is highly plausible with studies showing that the H3K27 methylation status of INK4A-ARF loci are regulated by the interplay of PRC2 complex and H3K27 demethylase JMJD3 [170][171][172].

Asxl1
Mutations of Additional Sex Combs-Like 1 (ASXL1), a gene related to the polycomb family, were reported in MDS, AML and CML [173][174][175][176]. ASXL1, a mammalian homolog of Drosophila additional sex combs, is involved in embryonic development and is required for both the activation and repression of Hox genes [177].
Asxl1 mutant mice mis-express Hox genes resulting in defective haematopoiesis with variable defect in the lymphoid and myeloid compartments [178]. Interestingly, Asxl1 knockout mice did not develop haematological malignancy, suggesting that ASXL1 mutation may not be sufficient for leukaemia induction [178]. However it is also noted that mutation in Asxl1 knockout mice is slightly different from the human ASXL1 mutation, which clusters predominately within exon 12 and results in the expression of a truncated protein lacking only the C-terminal PHD finger with a possible dominant negative effect on the wild-type or other related proteins [173,174,178]. Recently, a polycomb repressive complex, PR-DUB, was identified to contain ASXL1 and the histone deubiquitinase BAP1, which removes monoubiquitin from H2A, but not from H2B [179]. Therefore, it is also tempting to speculate that the mutant ASXL1 may affect the activity, formation or targeting of PR-DUB complexes leading to the perturbation of H2A ubiquitination dynamics and the deregulation of critical PRC target genes such as Hox.

TET family
The TET family proteins include three members, TET1, TET2 and TET3, which share a conserved cysteine-rich region and the dioxygenase motif involved in Fe(II) and a-ketoglutarate binding. Recent studies have shown that TET1 enzymatically converts 5-methylcytosine (5mC) to 5-hydroxylmethylcytosine (5-hmC) [180]. Loss of Tet1 function resulted in increased promoter DNA methylation, suppression of Nanog expression and induced ESC differentiation [181]. TET1 is a translocation partner of MLL in AML [182,183], whereas TET2 is mutated in patients with MDS and other myeloid malignancies [184][185][186][187][188][189][190][191]. TET2 loss-of-function mutations in myeloid malignant cells resulted in uniformly low levels of 5hmC and, surprisingly, hypomethylation instead of hypermethylation of genomic DNA and the majority of the differentially methylated CpG sites compared to bone marrow samples from healthy controls [192]. Thus while it is plausible that deregulation of TET proteins may result in perturbation of global DNA methylation leading to transcriptional deregulation of critical target genes for leukaemogenesis, the actual functional consequence of TET mutation and 5hmC in gene expression is still unclear.

Jak2
Gain-of-function mutations of Janus kinase 2 (JAK2) leads to constitutive tyrosine kinase activity and is frequently detected in myeloid malignancies [193]. JAK2 phosphorylates and activates the transcription factor, signal transducer and activator of transcription 5 (STAT5), a major substrate for JAK2, to mediate its biological response [193]. The role of JAK2 in epigenetic regulation was elusive until Dawson et al. showed that nuclear JAK2 phosphorylates tyrosine 41 of histone H3 that resulted in the release of the transcription repressor, heterochromatin protein 1 a (HP1a), from chromatin. The exclusion of HP1a results in the transcription of genes normally repressed by HP1a, such as LMO2 [194]. JAK2 has also been reported to collaborate with JMJD2C to activate gene expression critical for lymphoma survival [195]. Given the prevalence of kinase deregulation in human cancer, it will be interesting to investigate the potential nuclear function of other kinases and their roles in the epigenetic regulation of transcription.

Prospective
Although our knowledge on the genetics aspect of leukaemogenesis has been improved significantly in the last few decades, the epigenetic mechanisms utilised by various leukaemic fusion proteins are yet to be explored. In addition to the epigenetic writers and epigenetic readers described in some detail above, the landmark discovery of histone demethylases (epigenetic erasers) and their roles in the epigenetic regulation will provide opportunities for further scientific pursuit. Identification of specific mutation of epigenetic regulators and the altered epigenomes in leukaemias pave the way for extended screening and identification of putative therapeutic targets with the use of the synthetic lethality approach [196][197][198]. In the near future, a better understanding of the gain or loss-of-function of those oncogenic transcription factors and epigenetic modifying enzymes essential for pre-leukemic or leukemic stem cells will provide important insights into the identification of amenable targets for therapeutic intervention [199].