After the discovery of DNA and the double helix structure, classic genetics has long assumed that the sequences of DNA determine the phenotypes of cells. DNA is packaged as chromatin in cells, with nucleosomes being the fundamental repeating unit. Four core histones (H2A, H2B, H3, and H4) form an octamer and are then surrounded by a 147-base-pair (bp) segment of DNA. Nucleosomes are separated by 10–60 bp DNA. Researchers have gradually found organisms that share the same genetic information but have different phenotypes, such as somatic cells from the same individual that share a genome but function completely differently. The term epigenetics was first proposed and established in 1942 when Conrad Waddington tried to interpret the connection between genotype and phenotype.1 Later, Arthur Riggs and his group interpreted epigenetics as inherited differences in mitosis and meiosis, which could explain the changes in phenotypes. They were both trying to find the link between genotype and phenotype. Epigenetics is usually referred to as a genomic mechanism that reversibly influences gene expression without altering DNA sequences. Holliday assumed that epigenetics was also mitotically and/or meiotically heritable without DNA sequence change. Aberrant DNA methylation could be repaired via meiosis, but some patterns are still transmitted to offspring.2 This phenomenon covers a wide range of cellular activities, such as cell growth, differentiation, and disease development, and is heritable.3 Generally, epigenetic events involve DNA methylation, histone modification, the readout of these modifications, chromatin remodeling and the effects of noncoding RNA. The elements involved in different modification patterns can be divided into three roles, “writer,” “reader,” and “eraser”. The “writers” and “erasers” refer to enzymes that transfer or remove chemical groups to or from DNA or histones, respectively. “Readers” are proteins that can recognize the modified DNA or histones (Fig. 1). To coordinate multiple biological processes, the epigenome cooperates with other regulatory factors, such as transcription factors and noncoding RNAs, to regulate the expression or repression of the genome. Epigenetics can also be influenced by cellular signaling pathways and extracellular stimuli. These effects are temporary and yet long-standing. Given the importance of epigenetics in influencing cell functions, a better understanding of both normal and abnormal epigenetic processes can help to understand the development and potential treatment of different types of diseases, including cancer.

Fig. 1: Epigenetic regulation of DNA methylation, histone acetylation, and histone methylation.
figure 1

Gene silencing in mammalian cells is usually caused by methylation of DNA CpG islands together with hypoacetylated and hypermethylated histones. The “writers” (DNMTs, HATs, and HMTs) and “erasers” (DNA-demethylating enzymes, HDACs, and KDMs) are enzymes responsible for transferring or removing chemical groups to or from DNA or histones; MBDs and other binding proteins are “readers” that recognize methyl-CpGs and modified histones. DNMTs, DNA methyltransferases; MBDs, methyl-CpG binding domain proteins; HATs, histone acetylases; HDACs, histone deacetylases; HMTs, histone methyltransferases; KDMs, histone-demethylating enzymes.

The etiology of cancer is quite complicated and involves both environmental and hereditary influences. In cancer cells, the alteration of genomic information is usually detectable. Like genome instability and mutation, epigenome dysregulation is also pervasive in cancer (Fig. 2). Some of the alterations determine cell function and are involved in oncogenic transformation.4 However, by reversing these mutations by drugs or gene therapy, the phenotype of cancer can revert to normal. Holliday proposed a theory that epigenetic changes are responsible for tumorigenesis. The alteration of cellular methylation status by a specific methyltransferase might explain the differences in the probability of malignant transformation.5 In clinical settings, we noticed that although cancer patients share the same staging and grade, they present totally different outcomes. In tumor tissues, different tumor cells show various patterns of histone modification, genome-wide or in individual genes, indicating that epigenetic heterogeneity exists at a cellular level.6 Likewise, using molecular biomarkers is thought to be a potential method to divide patients into different groups. It is important to note that tumorigenesis is the consequence of the combined action of multiple epigenetic events. For example, the repression of tumor suppressor genes is usually caused by methylation of DNA CpG islands together with hypoacetylated and hypermethylated histones.7 During gene silencing, several hallmarks of epigenetic events have been identified, including histone H3 and H4 hypoacetylation, histone H3K9 methylation, and cytosine methylation.8,9

Fig. 2: Epigenetic regulations in cancer.
figure 2

Alterations in epigenetic modifications in cancer regulate various cellular responses, including cell proliferation, apoptosis, invasion, and senescence. Through DNA methylation, histone modification, chromatin remodeling, and noncoding RNA regulation, epigenetics play an important role in tumorigenesis. These main aspects of epigenetics present reversible effects on gene silencing and activation via epigenetic enzymes and related proteins. DNMTs, DNA methyltransferases; TETs, ten-eleven translocation enzymes; HATs, histone acetylases; HDACs, histone deacetylases; HMTs, histone methyltransferases; HDMs, histone-demethylating enzymes. MLL, biphenotypic (mixed lineage) leukemia.

Therefore, epigenetics enables us to investigate the potential mechanism underlying cancer phenotypes and provides potential therapy options. In this review, we focused and briefly expanded on three aspects of epigenetics in cancer: DNA methylation, histone acetylation and histone methylation. Finally, we summarized the current developments in epigenetic therapy for cancers.

DNA methylation

The DNA methylation pattern in mammals follows certain rules. Germ cells usually go through a stepwise demethylation to ensure global repression and suitable gene regulation during embryonic development. After implantation, almost all CpGs experience de novo methylation except for those that are protected.10 Normal dynamic changes in DNA methylation and demethylation based on altered expression of enzymes have been known to be associated with aging.11,12 However, inappropriate methylation of DNA can result in multiple diseases, including inflammatory diseases, precancerous lesions, and cancer.13,14,15 Of note, de novo methylation of DNA in cancer serves to prevent reactivation of repressed genes rather than inducing gene repression.16 Because researchers have found that over 90% of genes undergoing de novo methylation in cancer are already in a repressed status in normal cells.17 Nevertheless, aberrant DNA methylation is thought to serve as a hallmark in cancer development by inactivating gene transcription or repressing gene transcription and affecting chromatin stability.18

The precise mechanism by which DNA methylation affects chromatin structure unclear, but it is known that methyl-DNA is closely associated with a closed chromatin structure, which is relatively inactive.19 Hypermethylation of promoters and hypomethylation of global DNA are quite common in cancer. It is widely accepted that gene promoters, especially key tumor suppressor genes, are unmethylated in normal tissues and highly methylated in cancer tissues.20 P16, a tumor suppressor encoded by CDKN2A, has been found to gain de novo methylation in ~20% of different primary neoplasms.21 Mutations in important and well-studied tumor-suppressive genes, such as P53 and BRCA1, are frequently identified in multiple cancers.22,23,24 Studies have found that the level of methylation is positively associated with tumor size. In support of this, a whole-genome methylation array analysis in breast cancer patients found significantly increased CpG methylation in FES, P2RX7, HSD17B12, and GSTM2 coincident with increasing tumor stage and size.25 After analysis of long-range epigenetic silencing at chromosome 2q14.2, methylation of EN1 and SCTR, the first well-studied example of coordinated epigenetic modification, was significantly increased in colorectal and prostate cancers.26,27 EN1 methylation has also been observed to be elevated by up to 60% in human salivary gland adenoid cystic carcinoma.28 Of note, only ~1% of normal samples exhibited EN1 CpG island hypermethylation.26 Therefore, the significant difference between cancer cells and normal cells makes EN1 a potential cancer marker in diagnosis. In human pancreatic cancer, the APC gene, encoding a regulator of cell junctions, is hypermethylated by DNMT overexpression.29 During an analysis of colorectal disease methylation patterns, researchers found several genes that showed significant changes between precancerous diseases and cancers, including RUNX3, NEUROG1, CACNA1G, SFRP2, IGF2 DMR0, hMLH1, and CDKN2A.30 In the human colon cancer cell line HCT116, hMLH1 and CDKN2A always bear genetic mutation and hypermethylation of one allele, and this leads to inactivation of key tumor suppressors.31 It is known that p16, p15, and pax6 are usually aberrantly methylated in bladder cancer and show enhanced methylation in cell culture.32. Unlike gene promoter methylation, gene body methylation usually results in increased transcriptional activity.33 This process often occurs in CpG-poor areas and causes a base transition from C to T.34 The hypermethylation of specific CpG islands in cancer tissues is informative of mutations when the gene in normal tissues is unmethylated. One representative marker is glutathione S-transferase-π (GSTP1), which is still the most common alteration in human prostate cancer.35 Recently, DNA methylation in cancer has generally been associated with drug resistance and predicting response to treatment.36 For example, MGMT (O-6-methylguanine DNA methyltransferase) hypermethylation is still the best independent predictor of response to BCNU (carmustine) and temozolomide in gliomas because hypermethylation of MGMT makes tumor cells more sensitive to treatments and is associated with regression of tumor and prolonged overall survival.37,38 Similarly, MGMT is also a useful predictor of response to cyclophosphamide in diffuse large B-cell lymphoma39 (Table 1).

Table 1 Key regulatory factors of DNA methylation in cancer.

DNA methyltransferases (DNMTs)

DNA methylation is a covalent modification of DNA and is one of the best-studied epigenetic markers. It plays an important role in normal cell physiology in a programmed manner. The best-known type of DNA methylation is methylation of cytosine (C) at the 5th position of its carbon ring (5-mC), especially at a C followed by a guanine (G), so-called CpG sites. Non-CpG methylation, such as methylation of CpA (adenine) and CpT (thymine), is not common and usually has restricted expression in mammals.40 CpG islands traverse ~60% of human promoters, and methylation at these sites results in obvious transcriptional regression.41 Meanwhile, among the ~28 million CpGs in the human genome in somatic cells, 60–80% are methylated in a symmetric manner and are frequently found in promoter regions.42,43 The process of DNA methylation is regulated by the DNA methyltransferase (DNMT) family via the transfer of a methyl group from S-adenosyl-L-methionine (SAM) to cytosine.44 There are five members of the DNMT family: DNMT1, DNMT2, DNMT3a, DNMT3b, and DNMT3L. DNMT1 is responsible for the maintenance of methyl-DNA, recognizes hemimethylated DNA strands and regenerates the fully methylated DNA state of DNA during cell division.45 In a recent study, DNMT1 with Stella, a factor essential for female fertility, was responsible for the establishment of the oocyte methylome during early embryo development.46 DNMT3a and DNMT3b are regarded as de novo methylation enzymes that target unmethylated CpG dinucleotides and establish new DNA methylation patterns, but they have nonoverlapping functions during different developmental stages.47,48 DNMT2 and DNMT3L are not regarded as catalytically active DNA methyltransferases. DNMT2 functions as an RNA methyltransferase, while DNMT3L contains a truncated inactive catalytic domain and acts as an accessory partner to stimulate the de novo methylation activity of DNMT3A. The DNA methyltransferase-like protein DNMT3L can modulate DNMT3a activity as a stimulatory factor.49

During aberrant DNA methylation, DNMTs play an important role. Compared with DNMT1 and DNMT3a, DNMT3b was significantly overexpressed in tumor tissues.50 Overexpression of DNMT1, DNMT3a, and DNMT3b has been observed in multiple cancers, including AML, CML, glioma, and breast, gastric, colorectal, hepatocellular, pancreatic, prostate, and lung cancers. In cervical cancer patients, DNMT1 was expressed in more than 70% of cancer cells, whereas only 16% of normal cells expressed DNMT1. The higher level of DNMT1 expression was also associated with worse prognosis.51 The expression of DNMT1, DNMT3a, and DNMT3b has been observed to be elevated in acute myeloid leukemia (AML) and various solid cancers. These three methyltransferases do not show significant changes in the chronic phase of chronic myeloid leukemia (CML), but they are significantly increased during progression to the acute phase in CML.52,53 Notably, downregulation of DNMTs can also lead to tumorigenesis (Table 1).

Methyl-CpG recognition proteins

How DNA methylation leads to gene repression has been considered in many studies. Several hypotheses have been proposed. Three methyl-CpG binding domain protein (MeCP) families can read the established methylated DNA sequences and in turn recruit histone deacetylases, a group of enzymes responsible for repressive epigenetic modifications, to inhibit gene expression and maintain genome integrity.10,54 The first group is methyl-CpG binding domain (MBD) proteins, including MeCP2, MBD1, MBD2, and MBD4. MeCP1 is a complex containing MBD2, the histone deacetylase (HDAC) proteins HDAC1 and HDAC2, and the RbAp46 and RbAp48 proteins (also known as RBBP7 and RBBP4).55 MBD3 is unlike the other four family members and is not capable of binding to methylated DNA but instead binds to hydroxymethylated DNA.56 The zinc-finger and BTB domain-containing protein family is the second group and comprises three structurally different proteins, KAISO (ZBTB33), ZBTB4 and ZBTB38, which bind to methylated DNA via zinc-finger motifs. The third family includes two ubiquitin-like proteins with PHD and RING finger domains, UHRF1 and UHRF2, which recognize 5-mC via RING finger-associated (SRA) domains. On the other hand, methylation of DNA can also be a barrier for certain transcription factors to bind to promoter sites such as AP-2, c-Myc, CREB/ATF, E2F, and NF-kB.13

As for methyl-group binding proteins, many studies have investigated their roles in various cancers, but the mechanism underlying these alterations remains unclear. MBD proteins cooperate with other proteins to regulate gene transcription.57,58 However, the role of MBD1 and MBD2 has not been identified in human lung or colon cancer, with only limited mutations being detected.59 Furthermore, loss of MBD1 did not show any carcinogenic effect in MBD−/− mice.60 Compared with MBD1, MBD2 shows more effect on tumorigenesis. Deficiency of MBD2 strongly suppresses intestinal tumorigenesis in APCMin-background mice.61 A possible reason is that many important signaling pathways are downregulated in colorectal cancer, and loss of MBD2 leads to reexpression of these genes.62 Meanwhile, inhibition of MBD2 shows promising effects on suppression of the tumorigenesis of human lung cancer and colon cancer.63 Although MBD3 does not directly bind to methylated DNA, it regulates the methylation process via interactions with other proteins, such as MBD2 and HDAC. For example, application of an HDAC inhibitor in lung cancer cells upregulated p21 (also known as CDKN1A) and downregulated ErbB2, leading to inhibition of cancer cell growth. Silencing of MBD3 blocked the effects of an HDAC inhibitor.64 MBD3 and MBD2 form a complex, nucleosome remodeling and deacetylase (NuRD), which interacts with histone-demethylating enzymes to regulate gene expression in cancer.65 Mutation of MBD4 has been found in colorectal cancer, endometrial carcinoma and pancreatic cancer.66 Furthermore, this mutation unexpectedly affects the stability of the whole genome, not only CpG sites.67 Knockout of MBD4 indeed increased tumorigenesis in APCMin-background mice, which makes MBD4 a tumor suppressor.68 MBD4 is important in DNA damage repair, given the interaction between MBD4 and MMR.69 In contrast, the expression of MeCP2 and the UHRF family tends to promote tumor growth.70,71,72,73,74 In the KAISO family, KAISO directly binds to p120ctn, a protein with an alternative location in some cancer cells, and they together regulate cell adhesion and motility.75,76 However, deficiency of ZBTB4 contributes to tumorigenesis77 (Table 1).

DNA-demethylating enzymes

DNA methylation is a stable and highly conserved epigenetic modification of DNA in many organisms.78 However, loss of 5-mC and DNA demethylation have been identified in different biologic processes. For example, DNA demethylation is important for primordial germ cells (PGCs) to gain pluripotent ability.79,80 DNA demethylation is actively regulated by the TET protein family (ten-eleven translocation enzymes, TET1-3) via the removal of a methyl group from 5-mC. These three proteins differ from each other in terms of expression depending on the developmental stage and cell type.18 TETs oxidize 5-mC in an iterative manner and catalyze the conversion of 5-mC to 5-hydroxymethylcytosine (5-hmC), which is a key intermediate in the demethylation process.81 5-hmC, as a relatively stable intermediate substrate, is less prone to further oxidation by TET proteins than 5-mC.82 However, overexpression of only TET1 and TET2 can cause a global decrease of 5-mC.18 Stepwise oxidation of 5-hmC by TET proteins can yield two products: 5-formylcytosine (5-fC) and 5-carboxylcytosine (5-caC).83 These two molecules can be excised by thymine-DNA glycosylase (TDG) and eventually be repaired to unmodified C.84 DNA demethylation or restoration of the unmodified cytosine can also occur passively through replication-dependent dilution of 5-mC.

Disruption of normal DNA demethylation is thought to be associated with oncogenesis. TET proteins were initially associated with leukemia. Researchers have found that in a small number of AML patients, TET1 is fused to MLL via the chromosome translocation t(10;11)(q22;q23).85 Further studies found that TET2 was more widely expressed in different tissues than TET1 and TET3. Analyses revealed that mutation or deficiency of TET2 occurred in ~15% of patients with myeloid cancers, including myelodysplastic syndrome (MDS), myeloproliferative disorders, and AML.86 In patients with CML, mutation of TET2 has been detected in ~50% of patients.87 Although TET2 mutations have been found in several myeloid malignancies, their prognostic effect remains controversial. Based on the phenomenon that mutation of TET2 was elevated in patients whose disease transformed from chronic myeloid malignancy to AML, researchers considered that TET2 loss was important for cells to regain the ability to self-renew.88 The role of TET proteins has also been investigated in several solid tumors. Compared with surrounding normal tissues, 5-hmC is significantly reduced in human breast, liver, lung, pancreatic, and prostate cancers with reduced expression of TET family proteins.89 Deficiency of TET1 in prostate and breast cancer is associated with tumor cell invasion and breast xenograft tumor formation via the inhibition of the methylation of metalloproteinase (TIMP) family proteins 2 and 3.90 Loss of 5-hmC is an epigenetic hallmark of melanoma, and thus, introducing TET2 into melanoma cells results in suppression of tumor growth and increased survival in an animal model91 (Table 1).

Histone modification

Histone modification can occur to the flexible tails as well as the core domain of histones, including those sites that are buried by DNA. In particular, the flexible histone tails are enriched with basic Lys/Arg and hydroxyl group-containing Ser/Thr/Tyr residues, thereby being hotspots for hallmark histone modifications. The tails extend from the surface of the nucleosome and are readily modulated by covalent posttranslational modification (PTM). PTMs modify histones by adding or removing chemical groups and regulate many biological processes via the activation or inactivation of genes. These processes mainly include acetylation and methylation of lysines (K) and arginines (R), phosphorylation of serines (S) and threonines (T), ubiquitylation, and sumoylation of lysines. In addition to those mentioned and discussed above, histone modifications also include citrullination, ADP-ribosylation, deamination, formylation, O-GlcNAcylation, propionylation, butyrylation, crotonylation, and proline isomerization at over 60 amino acid residues.157,158 In addition to conventional PTMs, novel PTM sites are also found outside of the N-terminal tails.

Histone modifications at certain sites, such as promoters and enhancers, are thought to be largely invariant, whereas a small number of these sites remain dynamic. H3K4me1 and H3K27ac, two dynamic modifications, were identified to activate enhancers and regulate gene expression.159 H3K9ac and H3K9me3 are two common modifications at promoters.160,161 Appropriate histone modifications are important in gene expression and human biology; otherwise, alterations in PTMs may be associated with tumorigenesis. Analysis of cancer cells reveals that they exhibit aberrant histone modifications at individual genes or globally at the single-nuclei level.6,162 Understanding histone modification patterns in cancer cells can help us to predict and treat cancers. Thus far, most studies have focused on aberrant modifications within an individual site, such as H4K20me3 or H4K16ac, rather than enzymatic activity-associated abnormalities. Generally, alterations in histone modifications occur at an early stage and accumulate during tumorigenesis.163

Histone acetylation (lysine)

Histone acetylation occurs at multiple lysine residues at the N-terminus via the catalysis of histone acetyltransferases (HATs), also named lysine acetyltransferases (KATs). Histone acetylation regulates the compaction state of chromatin via multiple mechanisms, such as neutralizing the basic charge at unmodified lysine residues, and is associated with active transcription, especially at gene promoters and enhancers and the gene body; it also facilitates the recruitment of coregulators and RNA polymerase complexes to the locus.157,164 To date, HATs and histone deacetylases (HDACs) are the two of the best characterized groups of enzymes involved in histone PTMs. HATs transfer the acetyl groups from acetyl-CoA cofactors to lysine residues at histones, whereas the role of HDACs is the opposite, which makes histone acetylation a highly reversible process.

Histone acetyltransferases

HATs are predominantly located in the nucleus, but multiple lines of evidence have shown lysine acetylation in the cytoplasm, and their acetylation is associated with key cellular events.165 In addition, lysine acetylation found outside histones reminds us of the role of HATs in nonhistone PTMs.166 The first HAT was identified in yeast, and was named HAT1,167 and was then isolated from tetrahymena as HAT A by Allis and coworkers.168 In humans, HATs can be roughly divided into three groups: general control nondepressible 5 (GCN5)-related N-acetyl transferase (GNAT) (based on the protein Gcn5 found in yeast; including GCN5 and PCAF), MYST (based on the protein MOZ; including MOZ, MOF, TIP60, and HBO1), and p300/cAMP-responsive element-binding protein (CBP).169 Other HATs, including nuclear receptors and transcription factors, such as SRC1, MGEA5, ATF-2, and CLOCK, also harbor the ability to acetylate histones. Notably, a number of acetyltransferases also perform protein acetylation outside histones, such as TFIIB, MCM3AP, ESCO, and ARD1.170 Knockout of CBP/p300 is lethal for early embryonic mouse models.171,172 The acetyl group transfer strategies for each HAT subfamily are different. For the GCN5 and PCAF family, the protein crystal structure shows a conserved glutamate in the active site. Blockade of this amino leads to a significantly decreased acetylation function.173,174 Similarly, there is also a conserved glutamate plus a cysteine residue located at active sites of MYST family proteins.175 Unlike the other two families, the p300/CBP HAT subfamily has two other potential conserved residues, a tyrosine and a tryptophan.176 Generally, their catalytic mechanisms of acetyl group transfer can be divided into two groups. The GNAT family depends on a sequential ordered mechanism, whereas the members of the MYST family use a so-called ping-pong (i.e., double displacement) catalytic mechanism, which means that the acetyl groups are first transferred to a cysteine residue and then transferred to a lysine residue.177 In addition to differences in the acetyl transfer mechanism, HAT subfamilies, even different proteins in the same family, also have remarkable diversity in targeting sites.

Appropriate acetylation within cells is important since upregulation or downregulation of HATs is associated with tumorigenesis or poor prognosis.162,178 Compared with solid tumors, the association between histone modifications and cancer has been widely investigated in hematological malignancies. Germline mutation of CBP results in Rubinstein-Taybi syndrome along with an increased predisposition to childhood malignancies. Meanwhile, loss of another family member, p300, has also been associated with hematological malignancies.179,180 Therefore, both CBP and p300 seem to function as tumor suppressors. During cancer development, the expression of HAT genes can be disrupted by chromosomal translocations, although these are rare events. Generation of the fused protein CBP-MOZ is the result of the t(8,16)(p11,p13) translocation in AML.181 Translocation of t(10;16)(q22;p13) leads to the CBP-MORF chimera.182 Similarly, p300-MOZ, MLL-CBP, and MLL-p300 (MLL, mixed lineage leukemia) have also been identified in hematological malignancies.183,184,185 Generally, chromosomal rearrangements involving CBP are more common than those involving p300. Researchers have also investigated solid tumors, which are less mutated. The expression of translocated P300 in laryngeal squamous cell carcinoma (LSCC) tissue is much higher than that in adjacent normal tissue and is associated with advanced stage and poor prognosis.178 Missense point mutations in p300 are found in colorectal adenocarcinoma, gastric adenocarcinoma and breast cancer with quite low incidences.186,187 Rare inactivating mutations in CBP and PCAF have only been identified in cancer cell lines but not primary tumors.188 Based on these findings, we hypothesize that the differences between cell lines and primary tumors cannot be ignored. Amplified in breast cancer 1 (AIB1), also frequently called NCOA3 (nuclear receptor coactivator 3) or SRC3 (steroid receptor coactivator 3), is overexpressed in ~60% of human breast cancers, and increased levels of AIB1 are associated with tamoxifen resistance and decreased overall survival.189 Steroid receptor coactivator 1 (SRC1) is also associated with the chromosomal translocation t(2;2)(q35;p23), which results in PAX3–NCOA1 gene fusion in rhabdomyosarcoma without a consistent genetic abnormality during embryonic development190 (Table 2).

Table 2 Important enzymes or proteins that regulate histone acetylation in cancer.

Acetyl-lysine recognition proteins

The bromodomain (BRD) motif is an ~110-amino-acid conserved protein module and is regarded as the first and sole histone-binding module that contains a hydrophobic pocket to identify acetyl-lysine.191 The specificity of different BRDs depends on the sequences within the loops that form the hydrophobic pocket. Therefore, each BRD has a preference for different histones.192,193 In addition to their recognition of acetyl-lysine, BRDs are also capable of interacting with other chromatin molecules, such as plant homeodomain (PHD) finger motifs or another BRD. To date, 42 proteins containing bromodomains and 61 unique bromodomains have been discovered.194,195 Based on the sequence length and sequence identity of BRDs, the human BRD family can be divided into nine groups and one additional set of outliers, which has been well illustrated in published papers.169,194 Different BRD-containing proteins contain one to six BRDs. Intriguing, the most notable and well-studied bromodomain proteins are also HATs, such as PCAF, GCN5, and p300/CBP. Yaf9, ENL, AF9, Taf14, Sas5 (YEATS), and double PHD finger (DPF) have also been discovered to be acyl-lysine reader domains.191,196 Human MOZ and DPF2 are two proteins containing the DPF domain. Mutations in the YEATS and DPF domains are associated with cancer. For example, mutation of AF9 has been found in hematological malignancies, and ENL dysregulation leads to kidney cancer.197,198

Another important family is the BRD and extraterminal domain (BET) protein family, including BRD2, BRD3, BRD4, and BRDt, and this family shares two conserved N-terminal bromodomains and a more divergent C-terminal recruitment domain.199,200 These bromodomain proteins are critical as mediators of gene transcriptional activity.201 Of note, bromodomains have also been found in some histone lysine methyltransferases, such as ASH1L and MLL. BRDs are promiscuous domains and have been discussed in other well-constructed papers.169,194 In this review, we focus on the role of BRDs in tumorigenesis.

As histone acetylation “readers”, bromodomain proteins play important roles in tumorigenesis. BRD4 recruits the positive transcription elongation factor complex (P-TEFb), a validated target in chronic lymphocytic leukemia associated with c-Myc activity.202,203,204 Chromosomal translocation of BRD4, via the t(15;19) translocation, results in the generation of the fusion protein BRD4-NUT (nuclear protein in testis), which is found in NUT midline carcinoma (NMC). Importantly, inhibition of BRD4-NUT induces differentiation of NMC cells.205 Moreover, BRD4 is required for the maintenance of AML with sustained expression of Myc206 (Table 2).

Histone deacetylases

Histone deacetylases (HDACs) have recently attracted increasing attention. In humans, the genome encodes 18 HDACs. In contrast to the function of HATs, HDACs usually act as gene silencing mediators and repress transcription. Similarly, HDACs are expressed not only in the nucleus but also in the cytoplasm, and their substrates are also not limited to histones. Based on sequence similarity, HDACs can be divided into four classes: class I HDACs, yeast Rpd3-like proteins, are transcriptional corepressors and have a single deacetylase domain at the N-terminus and diversified C-terminal regions (HDAC1, HDAC2, HDAC3, and HDAC8); class II HDACs, yeast Hda1-like proteins, have a deacetylase domain at a C-terminal position (HDAC4, HDAC5, HDAC6, HDAC7, HDAC9, and HDAC10); class III HDACs are yeast silent information regulator 2 (Sir2)-like proteins (SIRT1, SIRT2, SIRT3, SIRT4, SIRT5, SIRT6, and SIRT7); and class IV involves one protein (HDAC11). The class IV protein shares sequence similarity with both class I and class II proteins.207,208 Classes I, II, and IV are included in the histone deacetylase family, whereas class III HDACs belong to the Sir2 regulator family.209 The catalytic mechanisms for these two families are different; classes I, II, and IV are Zn2+-dependent HDACs, whereas sir2-like proteins (sirtuins) are nicotinamide adenine dinucleotide (NAD+)-dependent HDACs and are also capable of mono-ADP-ribosyltransferase activity, another pattern of histone modification.210 Intriguingly, SIRT4 is thought to have more mono-ADP-ribosyltransferase activity than HDAC activity. SIR2 and SIRT6 seem to have equal levels of both mono-ADP-ribosyltransferase and HDAC activities.211,212 Moreover, after revealing the crystal structure of SIRT5, researchers found that SIRT5 is also a lysine desuccinylase and demalonylase.213 Therefore, the diversity of the sirtuin family makes them a group of multifunctional enzymes.

So far, the major known recognition sites of each HDAC are different, and these largely remain to be uncovered. For example, HDAC3 is thought to deacetylate H4K8 and H4K12,214 but in an HDAC3-knockout HeLa cell line, the acetylation levels of H4K8 and H4K12, even the overall acetylation levels of H3 and H4, were comparable with those in wild-type cells.215 Nevertheless, HDAC1 or HDAC3 siRNA can indeed increase the acetylation levels of H3K9 and H3K18.215 Therefore, partially because of the functional complementation and diversity within HDAC families, especially in class I, II, and IV, it is difficult to identify the specific substrates of certain HDACs. However, the substrates of the sirtuin family are quite clear. It is notable that because SIRT4 and SIRT5 are only located in mitochondria, they have no effect on histones. However, nonhistone lysine acetylation is also prevalent, since more than 3600 acetylation sites on 1750 proteins have been identified.166 The tumor suppressor p53 and the cytoskeletal protein α-tubulin are two representative substrates of HDACs.216,217,218 Notably, HDACs are also capable of regulating gene transcription by deacetylating other proteins that are responsible for epigenetic events, such as DNMTs, HATs, and HDACs.166,219 Another phenomenon is that some HDACs have to form a complex along with other components to function as transcriptional corepressors, which provides ideas and methods to design novel HDAC inhibitors. The Sin3, NuRD, and CoREST complexes are three complexes containing HDAC1 and HDAC2. Studies have found that purified HDAC1 or HDAC2 without associated components shows fairly weak deacetylation activity in vitro.220 HDAC3 interacts with the corepressors SMRT/NCoR to form the functional complexes, which significantly increases HDAC3 activity. NCoR also interacts with HDAC1, HDAC2 and the class II deacetylases HDAC4, HDAC5, and HDAC7, but usually not in the form of a complex.221,222 Deleted in breast cancer 1 (DBC1) and active regulator of SIRT1 (AROS) are two proteins that are able to bind to SIRT1, whereas their interactions present opposite functions. The DBC1/SIRT1 complex inhibits the deacetylation activity of SIRT1, whereas the combination of AROS and SIRT1 stimulates the activity of SIRT1.223,224

HDACs not only are able to deacetylate histones and nonhistone proteins but also interact with other epigenetic-associated enzymes, which gives them a vital role in tumorigenesis.162,178 Alterations in HDACs in cancers usually result in aberrant deacetylation and inactivation of tumor suppressor genes. For example, hypoacetylation of the promoter of p21, a tumor suppressor encoded by CDKN1A, can be reversed by HDAC inhibitors, resulting in an antitumor effect.225 A screen of the mutations in several HATs and HDACs, such as CBP, PCAF, HDAC1, HDAC2, HDAC5, HDAC7, and SIRT1, in more than 180 cancer samples including primary tumors and cancer cells indicated that the expression profiles of HDAC1, HDAC5, HDAC7, and SIRT1 are distinctive for colorectal cancers and normal colorectal mucosa, and the expression profiles of HDAC4 and CBP are capable of distinguishing breast cancer tissue from normal tissues226 (Table 2).

Histone methylation (lysine and arginine)

Similar to the process of histone acetylation, histone methylation also consists of three important components: “writers”, histone methyltransferases (HMTs), “readers”, histone methylation-recognizing proteins, and “erasers”, histone demethylases (HDMs). Methylation of histones occurs at arginine and lysine residues. Arginine and lysine both can be monomethylated or dimethylated, whereas lysine is also capable of being trimethylated. Histone methylation can either promote or inhibit gene expression, which depends on the specific situation. For example, lysine methylation at H3K9, H3K27, and H4K20 is generally associated with suppression of gene expression, whereas methylation of H3K4, H3K36, and H3K79 induces gene expression.360 Mutation of H3K27M (lysine 27 to methionine) and H3K36M are two important oncogenic events, and H3K27M and H3K36M serve as drivers of pediatric gliomas and sarcomas. H3K27M has been identified in more than 70% of diffuse intrinsic pontine gliomas (DIPGs) and 20% of pediatric glioblastomas, which results in a global reduction in the trimethylation of H3K27 (H3K27me3).361,362,363 However, the H3K36M mutation impairs the differentiation of mesenchymal progenitor cells and generates undifferentiated sarcoma, leading to increased levels of H3K27me3 and global loss of H3K36 (me2 and me).364,365 Meanwhile, depletion of H3K36 methyltransferases results in similar phenotypes to those seen with H3K36M mutation.364 To date, KMTs (lysine methyltransferases) have been better studied than arginine methyltransferases (PRMTs) due to their sequence of discovery, different prevalence and impact. Their targets are not limited to only histones, they also modify other key proteins, such as the tumor suppressor p53, TAF10, and Piwi proteins.366,367,368

Histone methyltransferases

All KMTs contain a 130-amino-acid conserved domain, the SET (suppressor of variegation, enhancer of Zeste, trithorax) domain, except for DOT1L. The SET domain is responsible for the enzymatic activity of SET-containing KMTs. Instead of methylating lysine residues in histone tails, DOT1L methylates lysine in the globular core of the histone, and its catalytic domain is more similar to that of PRMTs.369,370 The enzymatic activity of KMTs results in the transfer of a methyl group from S-adenosylmethionine (SAM) to a the ε-amino group of a lysine residue. The first identified KMT was SUV39H1, which targets H3K9.371 Sequentially, more than 50 SET-containing proteins have been identified with proven or predicted lysine methylation potential. Of note, KMTs are highly specific enzymes, meaning that they are highly selective for lysine residues they can methylate and the specific methylation degree they can achieve. For example, SUV39H1 and SUV39H2 specifically methylate histone 3 at lysine 9 (H3K9), and DOT1L only methylates H3K79.371 Based on their structure and sequence around the SET domain, generally, KMTs can be divided into six groups, SUV39, SET1, SET2, EZH, SMYD, and RIZ (PRDM) (reviewed by Volkel and Angrand372). The Pre-SET domain of the SUV39 family contains nine conserved cysteines that coordinate with three zinc ions to function. The SET1 family members share a similar Post-SET motif that contains three conserved cysteine residues. The SET2 family possesses an AWS motif that contains 7–9 cysteines. Their SET domain is located between the AWS motif and a Post-SET motif. The members of the enhancer of zeste homolog (EZH) family are the catalytic components of polycomb repressive complexes (PRCs), which are responsible for gene silencing. EZH proteins have no Post-SET motif but have 15 cysteines in front of the SET domain and show no methylated activity as isolated proteins.373 PRC2 shows lysine methylation activity through its catalytic components, EZH2 or its homolog EZH1.374 EZH2 can methylate not only histone H3 but also histone H1 at lysine 26.375 The SMYD family members, which are SET and MYND domain-containing proteins, possesses a MYND (myeloid-nervy-DEAF1) domain, a zinc-finger motif responsible for protein–protein interaction.376 The RIZ (PRDM) family is a large family containing a homolog of the SET domain, the PR domain. The PR and SET domains share 20–30% sequence identity and are both capable of inducing histone H3 methylation.377 However, most members of the RIZ family responsible for histone methylation are still unknown. So far, two of them have been proven to induce the methylation of histones: PRDM2 (RIZ1) is associated with H3K9 methylation; and Meisetz, the mouse homolog of PRDM9, trimethylates H3K4.378 Meanwhile, PRDM1 has been identified to interact with EHMT2, a member of the SUV39 family. PRDM6 acts as a transcription suppressor by interacting with class I HDACs and EHMT2 to induce cell proliferation and inhibit cell differentiation.379 Meanwhile, the recruitment of EHMT2 is based on the formation of a complex with PRDM1.380 Due to the lack of a characteristic sequence or structure flanking the SET domain, other SET-containing KMTs, such as SET7/9, SET8, SUV4-20H1, and SUV4-20H2, cannot be classified into these families. Notably, some KMTs contain more than one domain, which allows them to interact with other proteins, especially other epigenetic modifying proteins. SUV39H1 possesses a chromodomain that directly binds to nucleic acids and forms heterochromatin.381 MLL1 recognizes unmethylated DNA through its CpG-interacting CXXC domain. SETDB1 contains an MBD that interacts with methylated DNA.382 The Tudor domain in SETDB1 may potentially recognize the methylation of lysine residues.383 ASH1 is able to interact with CBP, a HAT, via a bromodomain within ADH1.384

Protein arginine methyltransferases (PRMTs) can be divided into two groups. Among the nine PRMTs, only PRMT5, PRMT7, and PRMT9 are type II PRMTs, and the other five PRMTs, except for PRMT2, are type I PRMTs. PRMT2 was identified by sequence homology385 but has not shown any catalytic activity during investigations, although PRMT2 acts as a strong coactivator for androgen receptor (AR), which is thought to be associated with arginine methylation.386 Both types of PRMTs first catalyze the formation of monomethylarginine as an intermediate. However, sequentially, type I PRMTs can form asymmetric dimethylarginine (ADMA, Rme2a), but type II PRMTs form symmetric dimethylarginine (SDMA, Rme2s). Rme2a means two methyl groups on one ω-amino group, whereas an Rme2s has one methyl group on each ω-amino group. PRMT1-PRMT8 were investigated by Herrmann and Fackelmayer,387 and FBXO11 was identified as PRMT9, which symmetrically dimethylates arginine residues.388

Most enzymes for histone methylation are substrate-specific proteins; therefore, alterations in the aberrant expression of enzymes are usually associated with specific histone residue mutations. One of the best-known examples of alterations in tumorigenesis is H3K4me3, which is associated with biphenotypic (mixed lineage) leukemia (MLL). The location of the MLL gene is where chromosomal translocations in AML and ALL usually occur.389 When the MLL gene is translocated, the catalytic SET domain is lost, which results in MLL translocation-generated fusion proteins, which recruit DOT1L.390 Maintenance of MLL-associated ALL depends on the methylation of H3K79 catalyzed by DOT1L.391 Therefore, DOT1L is usually associated with hematological malignancies rather than solid tumors. Alteration of the EZH2-induced methylation of H3K27 has been observed in multiple cancers, including various solid tumors (prostate, breast, kidney, bladder, and lung cancers) and hematological malignancies.392 Meanwhile, overexpression of EZH2 has been found in multiple cancers and is associated with poor prognosis.393 Different mechanisms have been proposed to describe the role of EZH2 in tumorigenesis (Table 3).

Table 3 Important enzymes or proteins that regulate histone methylation in cancer.

Methyl-histone recognition proteins

“Readers” of histone methylation contain several specific domains recognizing lysine or arginine methylation, such as a chromodomain,394 the WD40 repeat, the MBT (malignant brain tumor) domain, the Tudor domain395 and the PHD (plant homeodomain) finger motif.396 Representative chromodomain-containing proteins in humans are HP1 and Chd1, which can recognize H3K9me and H3K27me, respectively.394,397 WDR5 is a protein containing WD40 repeats. In addition to H3K4me, WDR5 prefers to bind to H3K4me2 via a histone-methylating complex and is required for maintaining H3K4me3.395 Later, WDR5 was shown to directly read H3R2, a “WIN” motif of MLL1, as well as symmetrical H3R2 dimethylation through the WD40 domain.398 L3MBTLs are a group of proteins containing three MBT repeat domains. L3MBTL1 represses gene expression via monomethylation or dimethylation of H4K20 or H1BK26.399 BPTF, RAG2, PYGO, and the tumor suppressor ING2 are representative proteins containing PHD finger motifs. They are all able to recognize and bind to H3K4me3.400 Intriguingly, DNMT3L and BHC80 also possess a PHD finger motif, but they selectively bind to unmethylated H3K4.401,402 There are a number of proteins containing Tudor domains, with a representative protein being JMJD2A. JMJD2A is a histone demethylase that equally binds to H3K4me3 and H4K20me3403 (Table 3).

Histone demethylases

The identification of histone demethylases (HDMs or KDMs) has lagged behind that of HMTs. Thus far, KDMs can be classified into two groups. The amine-oxidase type lysine-specific demethylases (LSDs) and the highly conserved JumonjiC (JMJC) domain-containing histone demethylases. LSD1 and LSD2, also known as KDM1A and B, are flavin adenine dinucleotide (FAD)-dependent amine oxidases that can only demethylate monomethylated and dimethylated lysine residues. LSD1 has been identified to specifically activate androgen receptor (AR) target genes along with AR by demethylating H3K9.404 The human genome codes more than 30 JMJC-containing KDMs that are able to remove methyl groups from all three methyl-lysine states. JHDM1A was the first characterized JMJC domain-containing HDM and specifically demethylates H3K36me2 and H3K36me1.405 Not all JMJC domain-containing proteins are able to demethylate histone proteins, such as HIF1AN and the transmembrane phosphatidylserine receptor PTDSR. JMJC-containing HDMs can be divided into six families:360 the JHDM1, JHDM2 (JMJD1), JHMD3 (JMJD2), JARID, PHF, and UT families. Notably, not all of these families possess the ability of histone demethylation. However, some JMJC-containing proteins, including those that are not included in these six families, contain one or more methylated-histone-binding domains. Their potential to demethylate methyl-lysine or methyl-arginine must be investigated. In addition to demethylases for lysine residues, JMJD6 is the first described arginine demethylase and lysine hydroxylase. It can remove methyl groups from H3R2 and H4R3.406 Another kind of protein is peptidylarginine deiminases (PADs or PADIs) or protein-arginine deiminases, which are able to convert arginine and monomethylated arginine to citrulline.407

LSD1 (KDM1A) is one of the best-studied KDMs and has been found to be increased in multiple cancers. Inhibition of LSD1 leads to global H3K4 methylation and promotes differentiation of neuroblastoma cells.408 Unlike KDM1A, KDM1B is mostly involved in growing oocytes with a restricted expression pattern.409 Similar to the dual roles of LSD1, members of the KDM2 family can either promote tumor formation or inhibit tumorigenesis.410 Through dimethylating H3K36 in DUSP3 (dual specific phosphatase 3), KDM2A activates ERK1/2 expression in lung cancer cells.411 Knockout of KDM2B in breast cancer downregulates the tumor stem cell markers ALDH and CD44 via the repression of polycomb complexes. KDM2B is also overexpressed in pancreatic ductal adenocarcinoma (PDAC) and cooperates with KrasG12D to promote PDAC formation in mouse models.412 The LSD1 and KDM2 family possesses context-dependent tumor-promoting and -inhibiting functions, which might depend on the different features of various cancers and the specific substrates of the enzymes. Therefore, further studies should take the dual roles of these enzymes into consideration. KDM3A, induced by hypoxia and nutrient starvation within the tumor microenvironment, shows carcinogenic effects via the promotion of tumor cell migration and invasion. Inhibition of KDM3A downregulates tumor-associated angiogenesis and macrophage infiltration.413,414 KDM3C is required for MLL-AF9 leukemia maintenance and is mutated in patients with intracranial germline tumors.415,416 KDM4A, KDM4B, and KDM4C have shown increased expression in prostate cancer with decreased levels of H3K9me2/3 and increased levels of H3K9me1.417 H3K9me3 is thought to be a hallmark of heterochromatic areas of the genome. In addition, KDM4 family members were the first identified demethylases targeting trimethylated lysines. Aberrant expression of KDM4 family members might lead to instability of the genome and become involved in tumorigenesis.410 Members of the KDM6 family usually act as tumor suppressors and are thought to cause cell growth arrest.418 For example, the tumor suppressor proteins p16INK4A and p14ARF, encoded by the INK4A-ARF locus, are repressed by H3K27me3. When stimulated by oncogenic factors, KDM6B is recruited to the INK4A-ARF locus and activates the transcription of these two tumor suppressors.419 In colorectal cancer, KDM7C is required for the efficacy of oxaliplatin and doxorubicin and for the activation of p53420 (Table 3).

Noncoding RNA

Epigenetic related noncoding RNAs (ncRNAs) include microRNAs (miRNAs), small interfering RNA (siRNAs), Piwi-interacting RNA (piRNAs), and long noncoding RNAs (lncRNAs). MiRNAs, one of the most studied ncRNAs, are small RNAs between 19 and 22 nucleotides in length that play important roles in the regulation of gene expression by controlling mRNA translation. Intriguingly, the regions that miRNAs usually target are frequently associated with carcinogenesis.567 Generally, they can be divided into tumor-promoting and tumor-suppressing miRNAs. During tumorigenesis, oncogenic miRNAs such as miR-155, miR-21 and miR-17-92 are usually overexpressed, and tumor-suppressive miRNAs such as miR-15-16 are downregulated.568 There is another type of miRNA, cellular context-dependent miRNAs, functioning in tumorigenesis. For example, miR-146 has been shown to be overexpressed in multiple cancers, whereas a recent study has proven that miR-146 can reduce the expression of BRCA1.568,569 Meanwhile, the expression of proteins and enzymes is also regulated by certain miRNAs. MiR-101 directly represses EZH2, and abnormal downregulation of miR-101 has been observed in cancers.570,571 The expression of the miR-29 family is inversely correlated with that of DNMT3A and -3B in lung cancer tissues. Forced expression of miR-29 inhibits tumorigenesis by inducing reexpression of methylation-silenced tumor suppressor genes.572 LncRNAs are another large group of noncoding RNAs that play a vital role in tumorigenesis. Some lncRNAs are cancer type-specific, such as PCGEM1 in prostate cancer and HEIH in hepatocellular carcinoma.573,574 Many aberrant lncRNAs have been discovered in various cancers. Dysregulation of HOTAIR has been found in lung, pancreatic, and colorectal cancer.575,576,577

Therefore, ncRNAs can either be directly involved in tumorigenesis or indirectly affect tumor development by participating in other epigenetic events.

Inhibitors and clinical trials

Unlike genetic mutations, epigenetic alterations are reversible. Given the importance of epigenetic marks in tumorigenesis, the availability of corresponding inhibitors has attracted extensive attention. Meanwhile, epigenetic regulation of a gene usually requires more than one epigenetic event. Currently, there are six epigenetic drugs approved for clinical use by the FDA (Table 4).

Table 4 Epigenetic drugs approved by the FDA.

Targeting DNA methylation

Blockade of DNMTs is the most effective way to prevent aberrant DNA hypermethylation. However, until now, targeting of the methyltransferase enzymes still lacks specificity and even causes hypomethylation of the global genome.578 Complete deletion of DNMT1 in mice results in embryonic lethality.579 Knockout of DNMT1 in fibroblast cells causes aberrant expression of 10% of genes and p53-dependent death.580 Administration of DNA methylation inhibitors results in tumorigenesis in male Fischer rats.581 Regulation of DNA methylation is vital in cell survival and function, and in addition to the specificity needed and the side effect associated, it is hard to identify proper drugs.

DNA methylation inhibitors can be divided into two groups: nucleoside analogs and nonnucleoside analogs. Nucleoside analogs have a modified cytosine ring and can be turned into nucleotides and incorporated into newly synthesized DNA or RNA. DNA methyltransferases are bound by covalent complexes with the analogs, which inhibits DNA methylation. 5-Azacitidine (5-Aza-CR) and 5-aza-2′-deoxycytidine (5-Aza-CdR) are currently the two most studied and promising demethylation agents.582 5-Aza-CR and zebularine are ribonucleoside analogs that can be phosphorylated to be able to incorporate into RNA. However, they can also be incorporated into DNA via the ribonucleotide reductase pathway. 5-Azacitidine, an analog of cytidine, is an injectable suspension for the treatment of myelodysplastic syndromes (MDSs). It promotes cell differentiation, demethylation, and reexpression of inactivated genes.583 The 5-azacitidine side effects include fetal abnormalities584 and decreased male fertility, especially at high doses, but its analog, 6-azacytidine, does not show such effects.585 Notably, after treating the noninvasive breast cancer cell lines MCF-7 and ZR-75-1 with azacytidine, the cells gained invasive abilities due to the hypomethylation of several prometastasis genes.586 Decitabine (5-Aza-CdR) and 5-fluoro-2′-deoxycytidine (5-F-CdR) are deoxyribonucleoside analogs that are capable of incorporating into DNA following phosphorylation. Decitabine (5-aza-2′-deoxycytidine) inhibits DNA methylation in a dosage-dependent manner. It can reactivate silenced genes at low doses but gains cytotoxicity at high doses, while myelosuppression is the major side effect at all doses.587. Dihydro-5-azacytidine (DHAC) is a biologically active and chemically stable analog of 5-azacitidine with decreased toxicity.588,589 Because of its hydrolytic stability, it may be administrated via prolonged i.v. infusion, potentially eliminating the acute toxicities caused by administration of 5-azacytidine.590 Zebularine is a potential oral DNA-demethylating drug with stability in acidic environments and in aqueous solutions.591 However, the near millimolar dose requirements and the limited bioavailability in rodents (<7%) and primates (<1%) leave zebularine far from clinical translation.592

Among the drugs discussed, 5-Aza-CR593 and 5-Aza-CdR594 have already been approved by the US Food and Drug Administration (FDA) for the treatment of certain subtypes of MDS and chronic myelomonocytic leukemia. Because of their intrinsic preference for newly synthetic DNA, they tend to affect dividing cells, i.e., cancer cells.595 Ongoing preclinical experiments and clinical trials are exploring their efficacy in solid tumors. The common side effects of these nucleoside-like analogs are mutagenic risk and genomic instability. Nonnucleoside analogs are capable of avoiding these side effects.

Currently, many nonnucleoside analogs have been developed to prevent DNA from aberrant hypermethylation. These drugs are usually small molecular inhibitors and directly target catalytic sites rather than incorporating into DNA. Based on a three-dimensional model of DNMT1, RG108 was designed to block the activity of this enzyme and cause demethylation.596 Psammaplin is a group of natural extracts from the sponge Pseudoceratina purpurea and is capable of inhibiting both DNA methyltransferases and histone deacetylases with mild cytotoxicity.597 Similarly, EGCG ((-)-epigallocatechin-3-gallate) is the major polyphenol from green tea and reversibly demethylates methyl-DNA, resulting in the reactivation of multiple key genes, including hMLH1, P16, and RA, in colon, esophageal, and prostate cancer cell lines.598 Both hydralazine and procainamide, two drugs associated with lupus-like autoimmune diseases, can inhibit DNA methylation and induce self-reactivity in cloned T-cell lines.599 They have promising tumor suppressor-reactivating and antitumor actions in breast cancer.600,601 Another strategy is developing antisense oligonucleotides to inhibit DNMT transcription. MG98 is a second-generation phosphorothioate antisense oligodeoxynucleotide that prevents DNMT1 mRNA translation effects but has no obvious antitumor effect.602 It has been under investigation in preclinical experiments and phase I/II clinical trials, especially in solid tumors.603,604 Of note, in a systemic analysis comparing nonnucleoside inhibitors with 5-Aza-CdR, the latter showed better efficacy in DNA demethylation inhibition.605

To date, hundreds of clinical trials have investigated the effects of anti-DNA methylation therapy for various cancers (Table 5).

Table 5 Important ongoing clinical trials with DNA methylation-targeted therapies.

Inhibitors of histone modifications

Compared with DNA methylation, histone modifications have been investigated in broader areas of diseases, including solid tumors, hematological malignancies, and even many inflammatory diseases (such as viral infection, diabetes and inflammatory lung diseases). During the process of gene silencing, lysine deacetylation and demethylation of H3K4 rather than demethylation of H3K9 or cytosine methylation might be the primary causative event.606 Therefore, histone modification plays an essential role in the regulation of gene expression, which also makes it a promising target for disease treatment. Clinical trials targeting histone acetylation and histone methylation are listed in Table 6 and Table 7, respectively.

Table 6 Important ongoing clinical trials with histone acetylation-targeted therapies.
Table 7 Important ongoing clinical trials with histone methylation-targeted therapies.

Inhibitors for HATs and BETs

Generally, there are two strategies for preventing aberrant histone acetylation, including altering interactions within the active sites within HATs or using mimetic products of enzymatic substrates. To date, many inhibitors targeting BRD proteins have been investigated in clinical trials, whereas there are no clinical trials investigating inhibitors for HATs.

Bisubstrate inhibitors are selective inhibitors for PCAF, p300, and TIP60. They mimic two substrates of HATs: the cofactor acetyl coenzyme A (Ac-CoA) and a peptide resembling the lysine substrate.607,608 However, due to their peptidic nature and size, they are not membrane-permeable and require the assistance of a delivery system. Based on inhibitory strategies for HATs, nonpeptide small molecular inhibitors have been developing as potential therapeutic agents. Several small molecule inhibitors are natural products, including garcinol, curcumin, and anacardic acid.609,610,611 These natural HAT inhibitors lack selectivity between HATs and often have other targets. Therefore, structurally modified and synthetic compounds have been reported. Α-Methylene-g-butyrolactones are small molecular inhibitors of HATs with selectivity for either GCN5L2 or PCAF.612 Isothiazolone is another HAT inhibitor targeting p300 and PCAF.613 However, high reactivity towards thiolates limits the application of HAT inhibitors in biological systems. Other inhibitors of HATs, such as thiazide sulfonamide and C646, have been gradually identified and show promising effects in multiple cancers. Another strategy to inhibit HAT activity is to target protein–protein interactions between HATs and their interaction partners. This method is dependent on the function of the interactions rather than the acetylation activity of HATs. ICG-001 and PRI-724 are representatives of this kind of inhibitor. Appropriately applying HAT agonists is also important to correct aberrant acetylation during diseases. CTPB is derived from anacardic acid and selectively activates p300, resulting in gene transcription.609 TTK21 and SPV106 are two other agonists based on anacardic acid.

Binding to BRDs and blocking acetylated lysine recognition is another mechanism that inhibits acetylation. JQ1 and I-BET762 are two representative inhibitors of the BET family. JQ1 is a cell-permeable small molecule and can competitively bind to BRD4 fusion oncoproteins, such as BRD4-NUT, resulting in cancer cell differentiation and apoptosis.614 Similarly, I-BET762 is also a synthetic mimic of and competes with BRD4.615 Other compounds, such as MS417, OTX-015, RVX-208, OXFBD, I-BET151, PFI-1, MS436, and XD14, are also BET inhibitors and have been well illustrated in other published papers.616 We will focus on the associations between these compounds and cancers. However, a number of non-BET proteins containing BRDs have attracted considerable attention. Many non-BET bromodomain inhibitors are based on a structure called the “WPF shelf” and a “gatekeeper” residue located at the start of the C helix.617 Several HATs have a BRD, such as Gcn5, PCFA, p300, and CBP. Inhibitors for CBP include MS2126, MS7972, ischemin, SGC-CBP30 and I-CBP112; optimized 1-(1H-indol-1-yl) ethanone derivatives have also shown promising results in inhibiting CBP and p300.618 BAZ2A/B bromodomain inhibitors include BAZ2-ICR and GSK2801. The quinolone-fused lactam LP99 was the first synthetic selective inhibitor for BRD7/9. I-BRD9 was identified by GlaxoSmithKline (GSK) and is a selective inhibitor of BRD9, which has more than 200-fold selectivity for BRD9 over BRD7 and 700-fold selectivity for BRD9 over BET family members.619 PFI-3 is a potential inhibitor of SMARCA4 and PB1 with a stronger affinity for the bromodomain of SMARCA4. However, Vangamudi et al. identified that the ATPase domain within SMARC4 bypassed the anticancer effects related to the bromodomain since PFI-3 did not inhibit cell proliferation.620 The BRPF1 (bromodomain and PHD finger-containing 1) protein is part of the BRPF family, which is a component of MYST family complexes. The inhibitors of BRPF1 include PFI-4, OF-1, and NI-57. 1,3-Dimethyl benzimidazolones were the first selective inhibitors of BRPF1. PFI-4 and OF-1 are two close analogs of 1,3-dimethyl benzimidazolone that have been identified by the Structural Genomics Consortium (SGC). Another BRPF1 inhibitor, NI-57, was discovered by the SGC based on a new quinolinone scaffold. Both NI-57 and OF-1 are thought to interact with BRPF1-3 as pan-BRPF bromodomain inhibitors. Based on the bromodomain contained within both TRIM24 (tripartite motif containing protein 24) and BRPF1, a dual inhibitor, IACS-9571, has been identified.621 Bromosporine is a panbromodomain inhibitor with good cellular activity, whereas in a recent study, researchers noticed that bromodomain inhibitors only targeted the BET family rather than other BRDs.622

Inhibition of HDACs

Given that multiple methods can regulate HDAC activity, the designation of HDAC inhibitors has its own advantages. In the 1970s, butyrate was found to induce the accumulation of acetylated histones in cancer cells, which is thought to be associated with the inhibition of deacetylation.623 Later, a natural extract, trichostatin A (TSA), was identified to inhibit the activity of partially purified HDACs and induce cancer cell differentiation and apoptosis.624 Gradually, more natural and synthetic compounds have been identified to inhibit histone deacetylation. A study reported that administration of HDAC inhibitors only regulates a small number of genes (1–2%) but induces an obvious and rapid decrease in c-Myc gene expression, which indicated that a restricted set of cellular genes was uniquely sensitive to regulation of histone acetylation.625 The combination of two HDAC inhibitors, SAHA and TSA, induced melanoma cell growth arrest by upregulating p21, p27 and NF-κB, and MG132 can enhance the effect of TSA.626 The inhibition of HDACs has been investigated in various cancers, with promising antitumor effects.627,628 Based on the characteristics of their chemical structures, HDAC inhibitors can be divided into five groups: short-chain fatty acids, hydroxamic acids, benzamides, cyclic peptides, and hybrid molecules. In addition to those included in the five groups, some new synthetic compounds also act as inhibitors of HDACs.

The short-chain fatty acid group contains sodium butyrate, valproic acid (VPA), sodium phenylbutyrate, and AN-9 (pivaloyloxymethyl butyrate). The effective concentration of butyrate is usually at the micromolar level. The group of hydroxamic acids includes more than ten members and is the best-studied class. Structural analyses of TSA and suberoylanilide hydroxamic acid (SAHA) show that they are noncompetitive inhibitors of HDACs since they share significant homology with class I and class II HDACs, which makes them mimics of the lysine substrates.629 In addition, they chelate the active zinc ion in a bidentate manner, which is crucial for enzymatic activity.624 Hexamethylene bisacetamide (HMBA) is a representative of the hybrid polar compounds (HPCs), whereas second-generation HPCs, such as oxamflatin, SAHA, suberic bishydroxamic acid (SBHA), and m-carboxycinnamic acid bishydroxamide (CBHA), have shown better inhibition of HDACs and anticancer effects than first-generation agents.630 Oxamflatin, scriptaid, and amide are analogs of TSA and show anticancer effects.631,632,633 Benzamide inhibitors (MS-275, MGCD0103, and CI-994) are well-studied and show promising effects in the treatment of diseases, especially cancers. They inhibit histone deacetylation via binding to catalytic zinc ions within HDACs through carbonyl and amino groups. Inhibition of HDACs by benzamide inhibitors is thought to be reversible, but the bond may become tight and pseudoirreversible in a time-dependent manner.634,635 However, benzamide inhibitors have less activity than members of the hydroxamate or cyclic peptide families, with an effective concentration around the micromolar range.636 Cyclic peptides can be further divided into two groups: cyclic tetrapeptide containing a 2-amino-8-oxo-9, 10-epoxy-decanoyl (AOE) moiety (HC-toxin, trapoxin) and cyclic peptides without the AOE moiety (apicidin and romidepsin). The epoxyketone group is essential for the inhibitors to bind to active zinc ions, but the epoxyketone-based bond is irreversible. Trapoxin is a fungal cyclic peptide and can irreversibly inhibit the activity of HDACs.637 Romidepsin, also known as FK228, most likely relies on one of the thiol groups to coordinate to the active site zinc ion.638 Garlic-associated derivatives, such as diallylsulfide and allylmercaptan, are capable of generating a thiol group that makes them potential inhibitors of HDACs.639 K-trap, an analog of trapoxin, and other derivatives, including 9-acyloxyapicidins and 9-hydroxy, have been under investigation. Depudecin is a natural epoxide derivative isolated from the fungus Alternaria brassicicola. Psammaplins is isolated from a marine sponge Pseudoceratina purpurea. These two natural extracts can inhibit the activity of HDACs.

Early HDAC inhibitors were nonselective because of the high homology of the structure and catalytic mechanism of HDACs within each group. The first selective HDAC inhibitor was tubacin, which targets HDAC6 with increased tubulin acetylation but not histone acetylation.640 PCI-34051, a specific inhibitor of HDAC8, can induce caspase-dependent apoptosis in T-cell lymphoma but does not increase histone acetylation.641 Another benzamide inhibitor, SHI-1:2, shows HDAC1/HDAC2-specific inhibitory activity that is >100-fold more selective than that of other HDACs.642 New synthetic chemicals, such as SK7041 and splitomicin, selectively target class I HDACs and sir2-like family members, respectively. The same efforts have been made to develop inhibitors for sirtuins, the class III HDACs. Nicotinamide, a byproduct of the sirtuin enzyme reaction, is a widely used inhibitor of all sirtuins. Other compounds, such as cambinol, salermide, tenovin, EX-527, suramin, and AGK2, have also been reported as sirtuin inhibitors. Sirtuin inhibitors (such as nicotinamide) function via interactions with the NAD+ within the active site of sirtuins or through binding to acetyl-lysine.

Of note, second-generation HDACs, including hydroxamic acids (vorinostat (SAHA), belinostat (PXD101), LAQ824, and panobinostat (LBH589)) and benzamides (entinostat (MS-275), tacedinaline (CI-994), and mocetinostat (MGCD0103)), are currently in clinical trials, and some of them have already been approved for disease treatment. The success of romidepsin in phase I clinical trials in cutaneous and peripheral T-cell lymphoma accelerated the development of HDAC inhibitors as anticancer drugs. In 2006, SAHA (vorinostat) was first approved by the US Food and Drug Administration (FDA) for the treatment of cancer, restricted to patients with cutaneous T-cell lymphoma (CTCL), as an HDAC inhibitor.643 Romidepsin (Istodax) was the second approved HDAC inhibitor, which was approved in 2009. Three members of the benzamide family have also shown clinical significance in anticancer drug development. Belinostat (Beleodaq, previously known as PXD101) was approved in 2014 by the US FDA and European Medicines Agency to treat peripheral T-cell lymphoma. Another HDAC inhibitor, panobinostat, is a nonselective HDAC (pan-HDAC). It has shown promising effects in anticancer treatments; therefore, the FDA accelerated its approval for the treatment of patients with multiple myeloma. Intriguing, as we mentioned before, truncating mutations in HDAC2 have been found in sporadic carcinomas and colorectal cancer and result in resistance to traditional HDAC inhibitors.644 Mutations in other HDACs also exist; therefore, screening of these mutations in cancer can improve the efficacy of HDAC inhibitors.

Inhibitor of HMTs and HDMTs

EPZ004777 was the first identified selective inhibitor of DOT1L and selectively kills MLL-translocated cells over those without MLL translocation.645 However, due to its poor pharmacokinetic properties, a second generation of EPZ004777, EPZ-5767, was developed with a cyclobutyl ring replacing the ribose moiety.646 EPZ-5767 also shows synergistic effects with cytarabine, daunorubicin, and the DNMT inhibitor azacitidine in treatments for ALL with MLL translocation. EPZ-5767, though still showing low oral bioavailability, has been investigated in clinical trials for the treatment of leukemia with MLL rearrangement.647 There are several inhibitors of EZH2. 3-Deazaneplanocin A (DZNep), a derivative of the antibiotic neplanocin-A, is one of the most studied compounds. In fact, DZNep is a SAH-hydrolase inhibitor and decreases EZH2 expression via upregulation of SAH, which leads to degradation of PRC2 in a feedback inhibition mechanism.648,649 Another kind of inhibitor is SAM competitive inhibitors. SAM is responsible for the methyl moiety of KMTs. EI1, a small molecular inhibitor of EZH2, inhibits EZH2 activity by directly binding to EZH2 and competing with SAM.650 GSK343 and GSK126 are two other SAM competitive inhibitors that have been investigated in clinical trials.651,652 EPZ005687, a potent inhibitor of EZH2, significantly reduces H3K27 methylation in lymphoma cells with point mutations at the Tyr641 and Ala677 residues of EZH2 without obvious effects on the proliferation of wild-type cells.653 EPZ-6438, which shows similar effects and superior oral bioavailability, was developed next.654 CPI-1205 is a novel inhibitor of EZH2 that belongs to the pyridone family.

Tranylcypromine (TCP) is an approved drug for depression due to its ability to inhibit monoamine oxidase (MAO) activity. The structures of LSD enzymes and MAOs share many similarities. Therefore, the side effects of TCP as an HDMT inhibitor, including orthostatic hypotension, dizziness, and drowsiness,655 are mostly caused by targeting of MAO. Administration of TCP in MLL-AF9 leukemia promotes tumor cell differentiation and apoptosis.656 TCP is also capable of resensitizing non-acute promyelocytic leukemia (APL) AML cells to all-trans retinoic acid (ATRA) treatment via increasing H3K4me2 and the expression of myeloid-differentiation-associated genes.657 Several derivatives of TCP have been developed to achieve better bioavailability and selectivity, including OG-002, RN-1, SP2509, and GSK690.658,659,660 Another LSD1 selective inhibitor, ORY-1001, can also promote the differentiation of leukemia cell lines, especially cells with translocations in MLL, and has good oral bioavailability.661 To date, three LSD1 inhibitors, including TCP, ORY-1001, and GSK2879552, have been under investigation in clinical trials for the treatment of cancer patients. Daminozide (N-(dimethylamino) succinamic acid, 160 Da), a plant growth regulator, selectively inhibits KDM2/7 by chelating the active site metal.662 Daminozide and siRNA can similarly downregulate KDM7 expression and then regulate tumor-repopulating cells via demethylation of H3K9.663 GSK-J1 was the first identified KDM6 inhibitor with restricted cellular permeability, which resulted from its highly polar structure. Its ethyl ester, GSK-J4, possesses an improved ability to enter cells.664 However, scientists have found that GSK-J1 shows compatible selectivity for the KDM6 and KDM5 families and that GSK-J4 is also a potential inhibitor for KDM5B and KDM4C.665 EPT-103182, a selective inhibitor of KDM5B, has shown promising results in terms of antiproliferative effects in hematological and solid cancer cells. KDM8 and JMJD6 share homology and can be inhibited by a broad spectrum inhibitor, NOG.661

Specific inhibitors usually have similar selectivity to closely related homologs within a group, and even across different groups, which needs to be taken into consideration when using compounds that are not highly selective.

Combined therapy

Epigenetic regulation during tumorigenesis is complicated and involves multiple steps. Therefore, the combination of two or more therapies targeting various epigenetic events seems helpful. This combination synergistically inhibits the expression of tumor-growth-promoting genes and promotes the reexpression of tumor suppressor genes. 4SC-202 is a small molecular drug with dual effects that can inhibit HDAC1/2/3 and LSD1 with similar low micromolar potency. This drug is under clinical investigation. Other studies have administered two or more kinds of epigenetic drugs for anticancer therapy. Relevant clinical trials are listed in Table 8.

Table 8 Important ongoing clinical trials with combination therapies including DNA methylation and histone modification.


Although more specific mechanisms need to be investigated, it is well accepted that epigenetic events are important in normal biological processes as well as in tumorigenesis and that the epigenetic status is usually widely altered during cancer initiation. This makes epigenome-targeted therapy a promising strategy for the treatment of cancer. Based on the complexity of cancer, epigenetic alterations have influenced multiple aspects in cancer, such as the expression of oncogenes and tumor suppressor genes and signal transduction, resulting in enhanced cancer growth, invasion and metastasis. Although epigenetic therapy has a rational and profound basis in theory, some problems remain to be discussed and solved. The first and most important is the problem of selectivity. Epigenetic events are ubiquitously distributed across normal and cancer cells. In fact, some cancers depend on certain epigenetic alterations and can be sensitive to this regulation, whereas under usual regulation, normal cells have the ability to compensate for these epigenetic changes. Therefore, the priority is to determine the most important epigenetic alterations for different cancers. The second problem extends from the first problem. Thus far, epigenetic therapy has obtained impressive results in hematological malignancies but not in solid tumors. The properties of hematological malignant cells and solid tumor cells are different. However, researchers have still investigated the appropriate strategies for solid tumors. Since epigenetic alterations have effects on the sensitivity of small molecule targeted therapy and chemotherapy or radiotherapy, epigenetic-targeted therapy seems to be an important adjunctive therapy. The combination of epigenetic therapy and immunotherapy has also been investigated in preclinical and clinical trials.

Based on the achievements obtained, epigenetic-targeted therapy is a promising strategy for anticancer treatment. Epigenomes in cancer are related to many aspects during cancer initiation. A better understanding of the specific mechanisms underlying those alterations in different cancers is necessary. Meanwhile, optimized treatment options, including a variety of combinations, still remain to be discovered.