Beyond regulations at DNA levels: A review of epigenetic therapeutics targeting cancer stem cells

Abstract In the past few years, the paramount role of cancer stem cells (CSCs), in terms of cancer initiation, proliferation, metastasis, invasion and chemoresistance, has been revealed by accumulating studies. However, this level of cellular plasticity cannot be entirely explained by genetic mutations. Research on epigenetic modifications as a complementary explanation for the properties of CSCs has been increasing over the past several years. Notably, therapeutic strategies are currently being developed in an effort to reverse aberrant epigenetic alterations using specific chemical inhibitors. In this review, we summarize the current understanding of CSCs and their role in cancer progression, and provide an overview of epigenetic alterations seen in CSCs. Importantly, we focus on primary cancer therapies that target the epigenetic modification of CSCs by the use of specific chemical inhibitors, such as histone deacetylase (HDAC) inhibitors, DNA methyltransferase (DNMT) inhibitors and microRNA‐based (miRNA‐based) therapeutics.


| INTRODUC TI ON
Cancer is one of the leading fatal diseases that severely threaten human life. 1,2 Approximately 18 million people are diagnosed with cancer every year and 9.6 million will ultimately die of cancer. 3 However, traditional therapeutics are effective only for few malignant tumours 4 due to metastasis, recurrence, heterogeneity, resistance to chemotherapy and radiotherapy, and escape from immunological surveillance, 5 all of which might be explained by the properties of cancer stem cells (CSCs). 6 Initial studies indicating that cancer cells may have similar stem-like characteristics were conducted in teratomas, 7 which led to the first CSC hypothesis that tumours consist of malignant stem cells and their benign progeny, 8 and eventually the identification of a small population of leukaemia stem cells initiating leukaemia in mice. 9 CSCs, currently defined as initiating tumour cells, have been identified in various cancer types and are regarded as one of the most promising targets for cancer therapeutics because of their intrinsic potential to cause cancer initiation, relapse, metastasis, multidrug resistance and radiation resistance. 10 However, this level of cellular plasticity cannot be entirely explained by irreversible genetic alterations. Thus, the significance of reversible epigenetic modifications has gradually been discerned in terms of the activation of specific transcriptional networks underlying the diverse cellular states of CSCs. Epigenetic changes are covalent modifications to DNA or histones without altering the DNA sequence, including DNA methylation, histone modification (methylation, acetylation, phosphorylation), and non-coding RNA (ncRNA) expression. 11 The main types of epigenetic modifications that have been targeted by cancer treatment in recent years are DNA methylation, histone acetylation and ncRNA expression. Moreover, the gene expression patterns triggered by exact epigenetic modulations are unique to CSCs. Therefore, selective epigenetic tumour therapeutics based on a deeper understanding of epigenetic alterations will definitely benefit the development of novel cancer treatments.

| C SC s AND THEIR ROLE IN C AN CER INITIATI ON AND PROG RE SS I ON
Human cancer is a type of genetic disease that originates from a series of accumulating mutations or genomic alterations, some of which are only found in specific cancer types, such as c-KIT mutations in gastrointestinal stromal cancers, whereas mutations in TP53 occur in almost every type of cancer. These aberrant gene expressions eventually affect different pathways regulating cell signalling, cell growth, DNA repair and other cellular events leading to several changes in normal cells such as the acquisition of the ability to divide infinitely, aid angiogenesis, escape from growth-inhibitory signals, evade apoptosis, and promote invasion and metastasis. [12][13][14] The classical view of tumorigenesis argues that the majority of tumour cells are capable of proliferating extensively and initiating new cancer cells on their own. However, such points of view are unsatisfactory because they cannot explain the few colonies observed in vitro and the need for a large number of tumour cells to form new tumour cells in vivo. 15 Given these unexplained properties of cancer cells, studies are continuously being conducted and a large body of work has deepened our knowledge regarding tumorigenesis. A more comprehensive understanding of cancer was obtained after consensus of the fact that tumour cells are heterogeneous, suggesting that only a limited subset of cells have the potential to fuel cancer initiation and progression, which was first proven in acute myeloid leukaemia (AML). 9,16 Furthermore, subsequent research identified a small number of malignant stem cells with the ability to initiate solid tumours in mammary cancers. 17 These malignant cells are termed CSCs in accordance with their stemness or stem-like properties, including the capabilities of differentiation and self-renewal extensively shared with normal stem cells. In addition, the decisive difference between CSCs and normal stem cells is their potent tumour-initiation capacity, indicating the significance of eliminating all CSCs in order to achieve effective treatment. 17 Apart from the shared stem-like properties, another common characteristic is the similar signalling pathways collectively utilized by these two kinds of stem cells, which highlights the importance of specific signalling pathways in the course of cancer initiation and progression. 18,19

| Cancer initiation
CSCs regularly serve as a small population of primary cells that fuel the initiation of diverse types of solid cancers. 20 Taking the initiation of head and neck squamous cell carcinoma (SCC) as an example, this course can only be triggered by gene mutations, such as those that cause the overexpression of Kras and p53 or affect the Tgfb and Pten signalling pathways, in undifferentiated stem-like cells of the epithelium. 21,22 Likewise, the aberrant expression of oncogenes, such as Sox2 and Stat3, in undifferentiated basal cells leads to oesophageal SCC initiation; however, this does not affect differentiated cells. 23 In colon cancer, the downregulation of APC associated with Wnt signalling and subsequent activation of Ras and phosphoinositide 3-kinase signalling results in cancer development. The low rate of these mutations and the time needed for the process of cancer development are almost certainly due to the hallmarks of CSCs. 24,25 Although all of these studies collectively suggest that mutations in CSCs may lead to cancer initiation, it is noteworthy that differences in the cells that were originally mutated are likely to have a paramount impact on cancer type. For example, in breast tumour models established in mice, loss of p53 along with BRCA1 in basal stem cells resulted in the development of malignant adenomyoepitheliomas, which is a tumour type rarely seen in human breast cancer patients, whereas the same mutations in luminal progenitors led to adenocarcinomas. 26 In conclusion, both mutated genes and the original cells are decisive for tumour type.

| Epithelial to mesenchymal transition, cancer metastasis and chemotherapy drug resistance
CSC and stem cell signals play important roles in cancer metastasis, 27 according to which CSCs undergo the process of epithelial to mesenchymal transition (EMT) and obtain the capability to transfer from the primary site to distal tissues or organs. In general, EMT is a continuous process that reduces adhesion between cells at first and then decreases cell polarity and enhances cell motility, and finally provides CSCs with invasive mesenchymal properties. 28 The EMT process of CSCs is generally associated with intrinsic epigenetic changes in these cancer-initiating cells. For example, the chromatin of genes fuelling EMT is more reachable and active in SCC stem cells derived from hair follicle stem cells in comparison with the less metastatic, more differentiated cell populations of SCC arising from epidermal stem cells. 29 Moreover, studies have revealed that CSCs with EMT features are more resistant to chemotherapy drugs than other cancer cells. 30 Traditional chemotherapy drugs such as cisplatin, gemcitabine and 5-fluorouracil are less effective in pancreatic tumour cell lines with an EMT-like phenotype. 31 The increased chemotherapeutic drug resistance is mainly mediated by the overexpression of drug efflux transporters such as multidrug resistance protein 1 (MDR1), multidrug resistance-associated protein 1 (MRP1), and ATP-binding cassette sub-family G member 2 (ABCG2), whose function is to expel drugs from cells using ATP against concentration gradients. 32,33 The overexpression of these transporters is likely caused by various pathways and mechanisms encompassing epigenetic changes, which are attracting increasing attention. For instance, the downregulation of histone deacetylase 1 (HDAC1) or increased H3S10 phosphorylation, H3K4 tri-methylation and histone H3 acetylation can all lead to the activation of ABCG2 transcription, finally resulting in enhanced drug efflux capability, 34 suggesting that epigenetic alterations may be potential targets for cancer treatments.

| EPI G ENE TI C REG UL ATI ON OF C SC S
There is no doubt that DNA encodes all the biological information essential to living creatures, whose mutations may lead to alterations in cellular differentiation and improper development. DNA inside the nucleus, packaged into chromatin, forms a compact nucleoprotein structure in which the nucleosome is the smallest functional unit, composed of 147 base pairs of DNA wrapped around a core of eight histone proteins. 35 This octamer consists of two copies each of the H2A, H2B, H3 and H4 proteins whose amino-terminal lysine-rich tails protrude out of the nucleosome and potentially and post-translation. 37 Here, we discuss normal epigenetic regulation in terms of three interrelated processes: chromatin modification mainly referring to histone acetylation, DNA methylation and regulation of ncRNA expression. F I G U R E 1 (A) Nucleosomes encompass eight histone proteins including two each of H2A, H2B, H3 and H4. The lysine residues in the aminoterminal tails of histones protruding from the octamer can either be acetylated by HATs or be deacetylated by HDACs. (B) DNA can also be epigenetically modified by DNMT-based methylation. This process is mediated by several DNMTs such as DNMT1, DNMT3A and DNMT3B through catalysing a methyl group to CpG dinucleotides. K: lysine residues; AC: acetyl group; Me: methyl group

| Histone acetylation
Playing an active role in the regulation of cellular processes such as cell differentiation, proliferation, angiogenesis and apoptosis, acetylation is the most common modification among these epigenetic changes. As a consequence, aberrant acetylation is believed to be relevant to various cellular events in cancer pathologies; notably, the global hypoacetylation of H4 is one of the most common hallmarks of cancer. 38 The level of histone acetylation is mainly regulated by HDACs and histone acetyltransferases (HATs). HATs can acetylate lysine residues in histone tails, while HDACs can remove an acetyl group from the ε-amino groups of lysine residues ( Figure 1A).  48 whereas DNMT3A and DNMT3B are prone to de novo DNA methylation by catalysing the methylation of unmethylated DNA, with the assistance of the catalytically inactive DNMT3 L. 49,50 It is noteworthy that hypermethylated CpG islands are observed in or near promoter regions, whereas gene bodies become hypomethylated in tumours with abnormal methylation. 51 Various types of mutations in DNMTs contribute to divergent routes of aberrant DNA methylation. Taking initiating mutations as an example, more than 20% of samples derived from patients suffering from AML had mutations in DNMT3A. In addition, more than half of the mutations occurred at amino acid R882, 52 which were later confirmed to be dominant-negative, leading to decreased catalytic activity of DNMT3A and focal hypomethylation, whereas wild-type DNMT3A R882 showed a hypermethylation pattern in AML DNA. 53,54 DNMT3A with initiating mutations is capable of creating an ancestral or founder preleukaemic clone, establishing an environment suitable for additional mutations forming malignant clones. 55,56 Then, subclones of overt leukaemia arise, providing that oncogenes have undergone further mutation. [57][58][59] In addition, ancestral clones are consistently present both when patients are in remission or suffering from relapse. 60

| CircRNAs
CircRNAs are a single-stranded closed circular RNAs that lack 5'-3' ends and poly (A) tails. 73 In recent years, there have been an increasing number of reports that circRNAs might be related to the pathogenesis of silicosis, 74 diabetes, 75 osteoarthritis, 76 Alzheimer's disease, 77 cardiovascular diseases, 78  The miRNA biogenesis begins with their transcription by RNA polymerase II which produces primary miRNA (pri-miRNA) as an end product. Then a type III RNase Drosha along with its cofactor protein DGCR8 binds to the pre-miRNA to generate precursor miRNA (pre-miRNA) by mediating enzymatic cleavages. And the pre-miRNA is exported to the cytoplasm via the exportin 5-RNA•GTP complex. Next, the RNase III Dicer binds to the pre-miRNA to cut the terminal loop which generates miRNA duplex. In the next step, the RNA-induced silencing complex (RISC) is incorporated by the miRNA duplex mediated by the AGO family proteins. Depending on whether the mature miRNA is partially or perfectly complementary to the target mRNA, this leads to inhibited translation or degradation, respectively also reverse DNA methylation by recruiting TET1 DNA demethylase to the FLI1 promoter in order to induce DNA demethylation. 97 On the other hand, some circRNAs have been found to regulate methyltransferase EZH2 by interacting with miRNA and subsequently regulating histone methylation indirectly. 98 For instance, hsa_circ_0020123 and circBCRC4 can promote the activity of EZH2 by sponging miRNA-144 99 and miRNA-101, 100 respectively.

| THER APIE S TARG E TING EPI G ENE TI C MOD IFIC ATI ON S OF C SC S
Given the significant role of epigenetic regulation, it is not surprising that HDAC and DNMT, which play a pivotal role in epigenetic erasers, and writer enzymes have attracted increasing attention and are continuously being studied in the search for cancer therapy. [101][102][103][104] In addition, the effects of ncRNAs in cancer are also receiving much attention. In this respect, it is natural to think about how we can interfere with the course of events mentioned above, using the corresponding chemical inhibitors.
In order to eradicate all cancer cells, targeting only the primary cancer cells is inadequate. Therefore, it is crucial to target a small population of CSCs. As mentioned above, epigenetic regulation mechanisms are indispensable in the progression of tumour cells and with a better understanding of the epigenome, which provides potential targets for the application of novel therapeutics against different tumour types, it has become increasingly important for us to target these using a variety of specific drugs and inhibitors to improve tumour therapy. For instance, both romidepsin and vorinostat have been proven to provide efficacy and a lasting response in patients with cutaneous T-cell lymphoma in Phase 2 multi-centre trials; however, few of the desired goals were achieved when these two drugs were utilized as single-agent drugs during the treatment of several solid tumours in clinical trials, [119][120][121][122][123][124][125][126][127][128][129][130] suggesting that haematological malignancies are more sensitive to HDAC inhibitors, and the combination of HDAC inhibitors and other anticancer drugs and/or radiotherapy may show promise in other cancer treatments. 44 Importantly, this kind of anticancer drug is associated with several adverse effects, not only those wide-ranging, easily controlled adverse effects, but also serious and life-threatening effects such as various cardiac effects, diarrhoea, and myelosuppression. Furthermore, extra caution is needed when such epigenetic modifiers are applied in children whose epigenetic profiles may be associated with their development. Apart from the very modest effect on solid tumours and serious adverse effects, another challenge in the development of novel HDAC inhibitors is that many patients develop resistance to HDAC inhibitors, which is also regularly observed for other types of anticancer drugs.

| Targeting HDAC and DNMT
Recent research shows that HDAC3 is effective for the con-

| DNMT inhibitors
As mentioned above, DNMTs are considered promising targets for the epigenetic treatment of tumours; therefore, it is not surprising that DNMT inhibitors have aroused substantial attention in recent years with respect to the regulation of aberrant DNA methylation patterns. Generally, DNMT inhibitors function by inhibiting DNA methylation in order to decrease the level of promoter hypermethylation and enable abnormally silenced tumour suppressor genes such as P15 or CDKN2B, P16 or CDKN2A, MLH1, and RB to reexpress. 136 There are two types of DNMT inhibitors: nucleoside DNMT inhibitors and non-nucleoside DNMT inhibitors. 137 The nucleoside analogues work by incorporating into the DNA and trapping DNMTs to DNA covalently, whereas the non-nucleoside analogues are capable of targeting the catalytic region of DNMTs to affect their activity. 137  which was administered at a recommended dose of 75 mg/m 2 , administered over a prolonged period of 7 days in a 4-week cycle, was first proven to be better than best supportive care (BSC) in MDS patients. 138 According to the data from clinical trials using azacytidine either in combinatorial therapies or as a single agent for a 15-year period, azacytidine demonstrated less toxicity, but similar or even better overall survival, compared with current AML treatments. 139 Thus, azacytidine is recommended for AML treatment, especially for elderly patients who cannot bear intensive chemotherapy regimens.

| Nucleoside DNMT inhibitors
Decitabine showed a prolonged median time to progression (TTP) to AML or death, while the overall survival was similar to BSC, and higher cytotoxicity was observed in clinical trials using decitabine either in combinatorial therapies or as a single agent for a 17-year period. 139 That being said, the development and approval of these two nucleoside DNMT analogues occurred well before the complexity of methylation patterns had been deciphered. 140,141 This type of inhibitor is prone to result in a genome-wide decline of DNA methylation levels and eventually induces the reactivation of genes randomly, including those with potentially deleterious effects. Moreover, nucleoside DNMT inhibitors lack single-agent efficacy in the treatment of solid tumours, probably because of hypoxia 142 and drug infiltration in solid tumours, and are cytotoxic to normal cycling cells. Concerns regarding the specificity and toxicity of nucleoside DNMT inhibitors relate to their intrinsic mechanisms, which hinder their clinical development. 143 There are currently only two other nucleoside DNMT inhibitors undergoing assessments in clinical trials at present: SGI-110 in Phase III and 5-F-CdR in Phase I. 144,145 An oligonucleotide antisense inhibitor of DNMT1, called MG98, is also in a Phase I study.
Given the shortage of these agents, nucleoside DNMT inhibitors are usually used at low doses to reprogramme and sensitize tumour cells to diverse radiotherapy, chemotherapy and immunotherapy regimens in clinical trials, some of which show good prospects. 112

| Non-nucleoside DNMT inhibitors
Owing to the boundedness of nucleoside DNMT analogues, accumulating research is concentrating on the design and development of non-nucleoside DNMT analogues. There are five main sources for obtaining novel non-nucleoside DNMT inhibitors, as follows: (1) repurposing traditional drugs such as procaine, 146  and its derivatives. [165][166][167] As for non-nucleoside DNMT inhibitors other than hydralazine and EGCG, the others are still in the preclinical stages. 144,168 Although the selectivity and efficacy of novel non-nucleoside DNMT analogues seem to have improved slightly according to their inhibitory activities (IC 50 /EC 50 ), the main weakness in the majority of these agents is the poor DNA demethylation capability in cells and/or relatively low bioactivity at micromolar levels in comparison with the nucleoside DNMT analogues. 153

| Targeting ncRNAs
Enormous challenges exist for targeting ncRNAs to develop new drugs for treatment. On the one hand, though the functions and regulatory roles that lncRNAs and circRNAs play in cancer cells make them potential targets for tumour therapeutics, they are still far from being recommended for diagnosis and treatment. As for lncR-NAs, there are no feasible or experimental therapies that directly target lncRNAs as yet due to their low expression and the lack of effective tools for adaptation to their particular features. 175,176 With respect to circRNAs, precise mechanisms related to cancer initiation and progression other than miRNA sponges remain to be elucidated, and more controlled and large-scale clinical studies are needed before cancer-specific circRNAs can be applied in clinical practice. That being said, on the other hand, tremendous progress has been made in terms of therapeutics targeting miRNAs, especially in the treatment of various solid tumour types using nanoparticle-conjugated miRNA mimetics. 177 In the following section, we focus on miRNA- Furthermore, in order to increase the amount of antitumour drug uptake by the target site in cancer treatment, the polysaccharide hyaluronic acid can be conjugated to the CD44 marker, which is upregulated in CSCs. [202][203][204]

| CON CLUS ION
Despite that the early attempts of cancer therapeutics to target CSCs were disappointing, we have learned much about CSCs and have begun to translate this understanding into the clinic. In this well-defined clinical context, epigenetic therapies also provide evidence that this strategy against CSCs can be promising and effective. In addition, many potential druggable epigenetic targets remain to be revealed. However, epigenetic modifications are likely to be much more complex than our initial imaginations might suggest. For example, we must consider the existing histone milieu because the cross-talk between histone modifications has an impact on biological response and protein recruitment when using HDAC inhibitors.
In addition, several enzymes thought to function epigenetically may act through non-epigenetic mechanisms, such as the methylation of cytosolic substrates.
Given this, a clear understanding of the hurdles that inhibit CSC-targeting epigenetic therapeutics will contribute to the development of better cancer treatments, and these are (1) the properties of many CSCs are not well identified in some cancer types 205 ; (2) the isolated CSCs used in most studies cannot simulate the complex biological microenvironment 206,207 ; and (3) whether CSCs should be activated or dormant in cancer therapy, which currently remains unclear. 208 Therefore, the future of CSC-targeting epigenetic therapeutics requires further exploration and substantial effort.

CO N FLI C T O F I NTE R E S T
No potential conflicts of interest are disclosed.

AUTH O R S ' CO NTR I B UTI O N S
WBY, YJG and SHZ contributed to conception and design. SHZ contributed to manuscript writing and figures making. YJG revised the figures. CJL, WBY and LJL critically viewed, edited and approved the manuscript. All authors read and approved the final manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analysed in this study.