Epigenetic Biomarkers and Their Therapeutic Applications in Colorectal Cancer

2.1 Histone modifications

Over the past decade, significant advances in our understanding of the CRC “epigenome” have revealed that most CRC cases harbor alterations in their histone modification states, particularly regarding aberrant histone methylation and acetylation [6, 15, 23]. Importantly, these abnormal histone marks are highly recurrent and have recently been used as biomarkers to predict the clinical outcome in CRC patients [2]. These include changes in the global patterns of specific histone modifications. For example, Tamagawa et al. showed that global changes in histone H3K4me2, H3K9ac, and H3K9me2 in metachronous liver metastasis correlated to overall survival of CRC patients [24]. Specifically, low H3K4me2 levels were shown to correlate with overall poor prognosis [24]. Likewise, other studies have identified reduced levels of H3K9me3 and H4K20me3 as diagnostic biomarkers for CRC in circulating nucleosomes which correlated with poor patient outcome [25]. Conversely, high H4K20me3 and H3K9me3, as well as low nuclear expression of H3K4me3, were associated with a better prognosis for early-stage CRC patients [26].

Furthermore, since reduction or enrichment of these marks frequently occurs at the promoters of key CRC-related oncogenes and tumor suppressors, this results in detrimental changes in gene expression that form the basis of tumorigenesis [15, 27]. For instance, H3K4me3, when found to be elevated in CRC primary tumors and cell lines, resulted in activated Wingless-type(WNT) signaling and target gene expression via interaction between SET domain-containing protein 1A(SETD1A) and β-catenin[28]. Meanwhile, another study revealed that low H3K4me1/2/3 levels were associated with hypoxia-induced silencing of MLH1in SW480 cells, which is a key event in the DNA mismatch defects linked to the development of sporadic CRC [29]. Yokoyama’s group also demonstrated a role for the well-recognized repressive mark H3K9me3, revealing that its increased levels in metastatic CRC patient-derived cells correlated with enhanced cell motility [30]. Interestingly, this coincided with repression of Ataxia-telangiectasia mutated(ATM) and p53-associated KZNF protein(APAK), leading to a defect in p53-dependent apoptosis [30]. Moreover, enrichment of another repressive mark, H3K27me3, was associated with poor CRC patient prognosis while elevated H3K79me2 was shown to enhance interleukin (IL)-22-induced stemness in CRC cells [31, 32]. Intriguingly, more recent studies have also shown that mutations in specific methylation sites could promote CRC development. For instance, the Shah and Lu groups identified histone 3 lysine 36-to-methionine (H3K36M) substitution mutations in CRC patient samples, which promoted more undifferentiated sarcomas in vivo[33, 34]. This suggests that H3K36 methylation potentially constitutes a major tumor suppressive epigenetic mark.

In addition to abnormal methylation, disruption of histone acetylation patterns also contributes to CRC pathogenesis, particularly relating to transcriptional inactivation of tumor suppressor genes and, sometimes, activation of oncogenes. For example, Richon et al. showed that hypoacetylation at the promoter of the tumor suppressor p21^WAF1 led to its repression, an effect that was reversed by inhibition of HDAC activity [35]. Conversely, mass spectrometry-based analyses used to quantify global alterations of histone modifications in CRC samples identified H3K27ac as a modification frequently upregulated in CRC [36]. In fact, one study highlighted the effects of aspirin in reducing the enrichment of H3K27ac in the promoters of inducible nitric oxide synthase(iNOS), tumor necrosis factor alpha(TNF-α), and IL-6[37]. This in turn corresponded to the dramatic reduction of the mRNA and protein levels of these genes, which suppressed inflammatory colitis symptoms and CRC tumor burden [37]. Taken together, these studies emphasize the differential abundance of key repressive and activating histone methylation and acetylation marks in CRC and suggests their role in regulating genes associated with CRC development and progression.

2.2 DNA methylation

DNA methylation constitutes the first recognized epigenetic alteration in CRC [38]. Usually, global DNA hypomethylation is frequently seen, which occurs gradually and early in the process of CRC carcinogenesis [38]. More precisely, global DNA hypomethylation mainly takes place on cytosine guanine (CpG) dinucleotides within pericentromeric regions. Initially, this hypomethylation in CRC was hypothesized to be associated mainly with widespread oncogene activation but has now been linked predominantly to increased genomic instability [3]. This increased accumulation of chromosome breakage and overall chromosomal instability contributes to a prevalent subtype of CRCs known as the CIN phenotype as we briefly described in Section 1 [39].

Notably, hypomethylation typically occurs in concert with systematic and discrete DNA hypermethylation events at the promoters of genes involved in DNA repair, apoptosis, proliferation, angiogenesis, adhesion, and invasion [38, 40]. DNA hypermethylation is the most extensively characterized epigenetic alteration in CRC, occurring at CpG dinucleotide-dense regions, called CpG islands, which are present in about 60% of genes [6, 41]. Apart from CpG islands, DNA hypermethylation has also been extensively observed within the first exonic/intronic regions of some genes and generally results in transcriptional silencing [42]. Some of the most frequently hypermethylated genes in CRC include Adenomatous polyposis coli(APC), Cyclin-dependent kinase inhibitor 2A(p16^INK4a/CDKN2A), Tissue inhibitor of metalloproteinases 3(TIMP3), O-6-Methylguanine-DNA methyltransferase(MGMT), Secreted frizzled related protein 1(SFRP1), Transmembrane protein with epidermal growth factor(EGF) like and two follistatin like domains 2(TMEFF2), Heparan sulfate-glucosamine 3-sulfotransferase 2(HS3ST2/3OST2), Ras association domain family member 1(RASSF1A), and GATA binding protein 4(GATA4) [43].

Another subtype of CRCs with extensive patterns of promoter methylation, known as the CIMP phenotype as described in Section 1, is also characterized by aberrant DNA methylation at genes with roles in CRC initiation or progression [44]. For instance, using a qPCR-based technique, one group identified genes with the highest percentage of methylation in CRC patients including Runt related transcription factor 3(RUNX3), Protocadherin 10(PCDH10), Secreted frizzled related protein 5(SFRP5), Insulin-like growth factor 2(IGF2), and Hepatocyte nuclear factor 1 homeobox A(HNF1) homeobox B(Hnf1b) [45]. Moreover, these genes were observed to have the most promising biomarker potential because of the frequent gene repression patterns [45]. Other commonly hypermethylated genes, such as Sex-determining region Y(SRY)-related HMG-box(SOX17) and Apoptosis-associated speck-like protein containing a CARD(ASC)/target of methylation-induced silencing(TMS1), were differentially methylated based on the staging of the disease [46, 47, 48], whereas MLH1, p16, Death-associated protein kinase 1(DAP-kinase), Ras association domain family member 2(RASSF2A), and WNT inhibitory factor 1(Wif-1) were regarded as plasma or serum detection markers for CRC [49].

In summary, these data strongly support the promising utility of DNA methylation as a critical diagnostic marker for CRC. Unfortunately, this has not necessarily translated into their prognostic or predictive use in clinical practice [50]. This can be attributed to significant variability in sensitivity, specificity, and reproducibility between diverse patient cohorts and gene expression platforms, which ultimately impacts the prognostic value of many tests. Currently, two FDA-approved commercial tests, Epi proColon® and Cologuard, have been used for screening alterations in methylation of common genes, such as SEPT9, NDRG4, and BMP3, for early detection of CRC [51]. However, they also generally lack prognostic value and require improvements in terms of sensitivity and specificity. Several other methylation biomarker assays have also been suggested, but validation in independent and large population screening studies is still needed [52].

2.3 Noncoding RNAs

Another epigenetic regulatory mechanism frequently deregulated in CRC involves the role of noncoding RNA (ncRNAs). Specifically, aberrations of microRNAs (miRNAs) expression, a major class of ncRNAs, are often observed in CRC and are considered to play a major role in tumorigenesis and CRC progression [23, 53]. These observations are consistent with the fact that miRNAs tend to exert oncogenic or tumor-suppressive effects. For example, miRNAs, such as miR-141, miR-200c, miR-145, miR-373, miR-520c, miR-135a, and miR-135b, have all been shown to affect CRC by regulating epithelial differentiation, WNT signaling, and CRC cell migratory and invasive potential [54]. Other miRNAs implicated in CRC include miRNA-124a and miRNA-34b/34c, which were shown to regulate the cell cycle and TP53 pathway, respectively [55]. Several miRNAs are also associated with epithelial-to-mesenchymal transition (EMT) in CRC. miR-15/16, miR-140, and miR-200 family members were shown to be associated with suppression of EMT and tumor cell metastatic potential while miR-21 enhanced this process [56].

Other studies have highlighted that alterations in the expression pattern of miRNAs in CRC were considered diagnostic, prognostic, or chemosensitivity markers [57]. For instance, high levels of miR-320 and miR-498 were correlated with progression-free survival in stage II CRC [58], while miR-21 abundance was associated with poor patient response to 5-fluorouracil adjuvant chemotherapy [59]. On the other hand, induced suppression of miR-21 promoted the sensitization of CRC cells to chemotherapeutic regimens [60, 61]. Another study by Toiyama et al. demonstrated a correlation between elevated serum miR-200c levels and stage IV CRC compared to earlier stages. Furthermore, high serum miR-200c showed a significantly positive correlation with lymph node metastasis, distant metastasis, and prognosis [62]. A comprehensive look of miRNAs as CRC biomarkers is reviewed by several other sources [60].

There is also ample evidence of miRNAs being downregulated in CRC and thus playing tumor-suppressive roles [63]. Arndt et al. showed that reduced levels of miR-133a as well as enrichment of miR-224 were associated with CRC initiation [64]. Moreover, this study and others also revealed that CRC patients at the adenomatous and carcinoma stages consistently exhibited reduced steady-state levels of miR-143 and miR-145 [63, 64]. Another classic example includes the miR-34 family. Transfection of miR-34a into CRC cells led to induction of apoptosis and inhibition of cell proliferation in part by amplifying the p53-mediated apoptotic response [65, 66]. Intriguingly, p53 has been shown to regulate miR-34a, suggesting a positive feedback loop between the two in which miR-34a could partly mediate the tumor-suppressive roles of p53 [55].

Interestingly, miRNA deregulation can also induce aberrant activity of many of the components of the epigenetic machinery [67]. Take DNMT3Afor example, which has been identified as a miR-143 target and is associated with CRC via downregulation of miR-143 and subsequent increase in DNMT3Aexpression levels [68]. Other examples include miR-140 and miR-449, which have been shown to target and downregulate HDAC1 and 4, respectively, thus exerting their tumor-suppressive effects [69, 70]. Taken together, these findings underscore the importance of miRNAs in exerting both oncogenic and antitumor roles in CRC, which may serve as the basis for the development of novel prognostic and therapeutic markers.

3.1 Microsatellite instability phenotype (MSI)

Approximately 15–20% of all CRC cases harbor MSI [72]. Microsatellites are defined as repetitive one to six base pair DNA sequences distributed along coding and noncoding regions of the genome. Importantly, the repetitive nature of these regions makes them particularly susceptible to mismatch errors [72]. Tumors with the MSI phenotype are therefore driven by the inactivation of MMR genes, which are involved in repairing DNA recombination and replication errors as well cellular responses to DNA damage [12]. The net effect of defective MMR machinery is accumulation of single base-pair mismatches, which results in a hypermutable cellular state [72].

MSI tumors are typically classified as MSI-high (MSI-H) or MSI-low (MSI-L). In MSI-H CRCs, usually two or more of the five microsatellite markers show instability, whereas in MSI-L tumors, only one of the five markers shows instability [12]. If none of the markers show instability, however, these are classified as microsatellite stable (MSS) CRC tumors, which account for approximately 80–85% of CRC patients [12]. Although majority of MSI-H tumors sporadically arise, a few are also linked to a familial hereditary syndrome known as Lynch syndrome or hereditary non-polyposis CRC (HNPCC) which account for about 3–5% of all CRC cases [12]. Sporadic MSI tumors are generally affected by epigenetic inactivation at the MLH1promoter via CpG island hypermethylation, whereas Lynch syndrome is caused by germline mutations in MLH1, PMS2, MSH6, or MSH2[12].

Importantly, unlike MSS, MSI tumors are poorly differentiated and are more often located in the proximal colon. MSI tumors also harbor a mucinous or signet ring type histology and increased number of tumor-infiltrating lymphocytes [72]. Additionally, when MSI was first identified in CRC patients, it was shown that MSI-H patients tended to have a better patient prognosis compared with MSS tumors and had an overall lower tumor stage at diagnosis [12]. Moreover, randomized phase III clinical trials along with several prospective studies have shown that the MMR status of these patients is also predictive of response to adjuvant 5-FU-based chemotherapy [12, 73]. The consensus was that patients with MSI-H tumors did not benefit from adjuvant 5-FU therapy compared to their non-MSI-H counterparts [73]. Furthermore, these data are consistent with other studies that revealed that human CRC cell lines with MSI-H phenotypes displayed resistance to DNA damaging agents, such as 5-FU, which could be overcome by the restoration of normal MMR function [74, 75].

3.2 Chromosomal instability (CIN) phenotype

The acquisition of genomic or chromosomal instability is a key feature in CRC development [39]. In fact, CIN has been found in approximately 85% of CRC cases and is characterized by increased chromosomal losses and gains as well as increased loss of heterozygosity [13, 14]. Although the exact mechanisms underlying CIN remain incompletely understood, it has been attributed to defects in genes related to the DNA damage response, telomere stability, and chromosomal segregation [39]. Unfortunately, standardizing the precise quantitative criteria that define a “CIN-positive” tumor has been challenging due to difficulties in the detection approaches of chromosomal instability [39]. The approaches currently in use include cytometry, karyotyping, and loss of heterozygosity analyses [76].

Along with the typical chromosomal abnormalities, accumulation of a characteristic set of mutations in specific tumor suppressor genes and oncogenes is also a prominent feature observed in CIN tumors [39]. These include mutations in APC, p53, Cyclooxygenase-2 (COX-2), and KRASas well as 18q alterations [39]. Interestingly, many studies have sought to determine the prognostic value of KRAS, TP53, or 18q alterations. So far, evidence of increased risk of relapse or mortality in CRC patients with KRASmutations has been presented, but other studies have failed to confirm this correlation [77]. Consequently, some of these putative individual prognostic markers are still undergoing rigorous study. However, several compelling studies indicate that the overall CIN phenotype is associated with a less favorable outcome in patients than those with the MSI phenotype, and unlike MSI tumors, it is not significantly influenced by adjuvant therapy in patients with stage II–III CRC [78].

Several ongoing phase I and II clinical trials are underway to therapeutically target pathways that directly or indirectly initiate and perpetuate CIN. Some of these include small-molecule inhibitors of COX-2, Polo-Like Kinases(Plks), Eg5, and Centromere protein E(CENP-E) [39]. Swanton et al. also showed that CIN-positive tumors are intrinsically resistant to taxanes due to the similarity between pathways that regulate the chromosomal segregation and those implicated in the taxane response [79]. These and other studies have collectively prompted the Chromosomal Instability and Anti-Tubulin Response Assessment (CINATRA) trial to assess whether patients with MSI-positive solid tumors derive benefit from EPO906 (new microtubule stabilizer) compared to patients with CIN-positive cancers [80]. Overall, phase I trials showed encouraging tumor control and response rates in patients with metastatic CRC (mCRC), although the trial was prematurely closed due to toxicity issues [80]. In summary, these data support the role of the CIN pathway in guiding patient stratification and the clinical management of CRC. However, more studies to better define the mechanisms underlying CIN and determine how CIN influences progression will be critical to advance our understanding of the most common form of genetic instability in CRC. Moreover, the feasibility of standardizing detection of CIN-positive tumors and thus be able to target chromosomally unstable cells, will be critical.

3.3 CpG island methylator phenotype (CIMP)

The term “CpG island methylator phenotype,” or CIMP, was coined in 1999 by Toyota and Issa to denote the CRC tumor subtypes characterized by widespread promoter DNA hypermethylation at certain tumor suppressor genes [81]. More than 50% of genes have promoters found within CpG islands [44]. Hence, the frequency of CIMP CRCs depends on which promoters are examined for methylation, with some promoters being more beneficial than others for identifying CIMP. Several studies have revealed that this methylation is common at the promoters of a diverse spectrum of genes, including Phosphatase and tensin homolog(PTEN), RUNX3, and Unc-5 netrin receptor C(UNC5C), making these key genes part of the expression signature profile in the evolution of CIMP CRCs [44]. Moreover, based on a panel of CIMP-specific markers coupled with the B-Raf proto-oncogene(BRAF) mutational status, CIMP tumors may be further classified according to the fraction of promoters that exceed a certain threshold of DNA methylation as being CIMP high, low, or negative [44]. Although CIMP-high and CIMP-low CRCs are significantly associated with biological sex as well as BRAF and KRAS mutational status, these classifications need additional refinement [44]. Nonetheless, it is reminiscent of the classification of CRCs based on degrees of MSI [82].

Notably, several of the clinicopathological characteristics of CIMP-high tumors have also been correlated to MSI tumors [82, 83]. For example, like MSI, CIMP tumors also represent a clinically distinct group characterized by epigenetic instability, distinct histological and pathological features, and discrete precursor lesions [84]. Pathologically, CIMP tumors also originate similarly to MSI tumors in the proximal colon, with a mucinous and poorly differentiated histological type and are frequently diagnosed in elderly and female patients [84, 85]. However, the determination of which specific methylated loci should be used to define CIMP remains a major challenge in the evaluation of CIMP tumors. Currently, several studies have proposed the classic panel containing the genes hMLH1and p16as well as Munc-18-interacting (MINT) proteins MINT1, MINT2, and MINT31. This panel has been further developed to contain the genes Calcium voltage-gated channel subunit alpha1 G(CACNA1G), Cellular retinoic acid binding protein 1(CRABP1), IGF2, Neurogenin 1(NEUROG1), RUNX3, Suppressor of cytokine signaling 1(SOCS1), Hypermethylated in cancer 1(HIC1), IGF-binding protein 3(IGFBP3), and Werner syndrome ATP-dependent helicase(WRN) [86].

In recent years, the use of DNA hypermethylation of specific genes to predict CRC patient outcome and therapeutic approach has received much attention. Although further validation is warranted, many studies have found a correlation between MSS CIMP+ CRC patients and poor prognosis [87]. Furthermore, the correlation between CIMP status and response to chemotherapy has been investigated. CIMP status predicts poor survival in metastatic MSS CRC patients treated with chemotherapeutic agent 5-FU [88]. Overall, patients who did not receive chemotherapy treatment but had tumors identified as MSS and CIMP had a worse survival outcome [89]. In contrast, two separate studies reported better outcomes for patients with CIMP tumors, a conflict that could be attributed to differences in the criteria used across the studies to define CIMP status. Additionally, this also suggests that the heterogeneous nature of CIMP tumors may warrant further classification [44].

Intriguingly, many studies have also found an association between CIMP status and other important epidemiological factors [90]. For instance, reports of an association between cigarette smoking, obesity, and CIMP showed that the number of cigarettes smoked as well as body mass index (BMI) had a significant relationship to CIMP tumor development [90, 91]. Furthermore, associations of CIMP status with smoking and obesity were evident only for females [90]. Taken together, the above evidences support a critical role of the CIMP pathway in the pathogenesis of CRC, which has also become a significant part of the current management of CRC. In the future, it will also be essential to have a consensus on a standardized panel of loci to define CIMP, similar to that utilized to identify MSI CRCs.

3.4 Consensus molecular subtypes

The three molecular pathways described so far also fall within several consensus classification systems for CRC. These systems vary in terms of the number of proposed subtypes, which can range from three to six depending on the combinations of genetic, epigenetic, clinical, and histopathological parameters used as well as the extent of the overlap between the three molecular pathways. For instance, the Consensus Molecular Subtypes (CMS) consortium has been suggested as one of the most robust classification systems and describes four groups (CMS1–4) based on expression profiling data from multiple studies [92]. While the majority of MSI-H CRCs fall into the CMS1 category, CMS2–4 display higher CIN. However, CMS3 samples have a distinctive profile compared with other CIN tumors. They tend to have lower CIN, higher prevalence of CIMP and close to 30% of the tumors are hypermutated which confers significant overlap with MSI status tumors [93]. Additionally, the Cancer Genome Atlas study also demonstrated that CIMP overlaps with the MSI pathway because of the fact that sporadic MSI-H CRCs usually harbor CIMP-high clinicopathological features [94]. Meanwhile, CMS4 are defined as CIN-heterogeneous tumors with mesenchymal characteristics that occur in later disease stages [93].

It is also noteworthy that many CRC tumors demonstrate mixed characteristics compatible with two or more of these subtypes, which may represent a transition phenotype or intratumoral heterogeneity, while others cannot be precisely classified into any of these pre-defined subgroups [95]. Furthermore, these classifications often lack incorporation of the molecular markers used for traditional TNM staging of CRC [96]. Taken together, these challenges as well as the existing incongruity between the various systems illustrate the need to further refine these consensus classifications by developing more progressive and integrated approaches.

4.1 Histone methyltransferases and demethylases

As discussed so far, aberrant changes in epigenetic modifications can significantly contribute to CRC progression. It is therefore unsurprising that many of the epigenetic enzymes mediating these modifications are themselves deregulated during the initiation and progression of CRC. Here, we describe the significance of changes in the expression levels of two such families of enzymes that oppose each other in terms of function, namely histone methyltransferases (HMTs) and demethylases (HDMs). Although changes in the expression or activity levels of several methylation-related enzymes have been linked to CRC, in most cases only a limited knowledge regarding the molecular mechanisms by which these enzymes contribute to disease development exists [15]. We summarize current knowledge regarding some of the preclinical validated implications of these enzymes as proof of principle for the employment of epigenetic agents in CRC. We also briefly discuss potential mechanisms of action of these enzymes as well as the advantages of targeting them using combinatorial over monotherapy approaches.

Histone lysine methyltransferases (HKMTs) have been widely studied across multiple solid tumor types including CRC [97]. For instance, studies in a preclinical model of CRC found that increased expression and activity of SET and MYND domain containing 3(SMYD3), a well-known HKMT, was strongly correlated with tumorigenesis. Moreover, RNAi-mediated depletion of SMYD3significantly impaired CRC cell proliferation, indicating a crucial role of SMYD3in maintaining CRC malignancy [98]. More recent studies suggest a putative mechanism by which this overexpression might occur by demonstrating that hypomethylation of the SMYD3promoter was observed in CRC tumor tissues compared to adjacent normal tissues. Further subgroup clinicopathological analyses showed that this hypomethylation was observed with stage III and IV tumors as defined by moderate to well-differentiated histology and positive lymph node metastasis [99].

Another well-studied HKMT, enhancer of zeste 2(EZH2), is also frequently deregulated in CRC. Both mRNA and protein levels of EZH2were found to be significantly increased in CRC tissues compared to non-cancerous counterparts [16]. Additionally, increased EZH2expression was directly correlated with tumor size, metastases, and overall worse disease-free survival of CRC patients [100]. He et al. also showed that siRNA-mediated depletion of EZH2inhibited the proliferation and migration of SW620 CRC cells, while inducing apoptosis and G0/G1 cell cycle arrest [101]. Another mechanistic study also revealed that knockdown of EZH2significantly reduced CRC cell invasion and matrix metalloproteinases 2/9(MMP2/9) secretion in vitrowhile promoting increased overall survival and decreased lung metastasis in vivo[102]. Furthermore, this EZH2-induced CRC cell invasion was mediated by direct binding of Signal transducer and activator of transcription 3(STAT3) to the EZH2promoter, resulting in downregulation of the vitamin D receptor (VDR) [102]. Interestingly, an association between a missense variant in EZH2and risk of CRC was discovered by the Li group. They identified that the presence of the rs2302427 variant showed a significant association with increased CRC susceptibility [103]. Recent studies point to other mechanistic roles of HKMTs in CRC. For example, depletion of SETD1A, a member of the trithorax (TrxG) family of HMTs, inhibited CRC cell growth and colony formation in part by decreasing expression of approximately 50% of Wnt/β-catenintarget genes [28]. Finally, in a mouse model, IL-22-mediated activation of disruptor of telomeric silencing 1-like(DOT1L) promoted CRC stemness and tumorigenic potential and was considered a predictor of poor survival outcome in CRC patients [32].

Protein arginine methyltransferases (PRMTs), although studied to a lesser extent, have also been shown to play critical roles in CRC malignancy via activation of Wnt/β-cateninand NF-ĸB signaling [104]. CARM1, for example, is an important positive modulator of Wnt/β-catenintranscription and was found to promote survival and anchorage-independent growth of CRC cells with aberrantly activated Wnt/β-cateninsignaling [105]. Meanwhile, our lab and others have shown that protein arginine methyltransferase 5(PRMT5) was overexpressed in CRC cells and patient-derived primary tumors, which correlated with increased cell growth, migration, invasion, and NF-ĸB activation as well decreased overall patient survival [106, 107, 108, 109]. The enzymes catalyzing removal of methylation marks, HDMs, are perhaps the least studied among the enzymes mentioned thus far and only a few have been implicated as playing tumor suppressive or oncogenic roles in CRC. LSD1, KDM4B, KDM4C, and KDM5Bhave all been shown to play pro-tumorigenic roles by promoting CRC cell growth and metastasis, whereas HDMs, such as JMJD3and JMJD1B, have been implicated as tumor suppressors [15]. Taken together, these data provide strong support for the continued development of selective and potent small-molecule inhibitors against these methylation-modifying enzymes as promising therapeutic agents for CRC.

4.2 Targeting HMTs and HDMs in CRC

Disruption of epigenetic regulation in CRC mediated by deregulated HMTs, DNMTs, and HDMs has garnered increasing interest in recent years. In this section, we aim to review the current status on the development of therapeutic strategies to modulate histone methylation for CRC treatment. The current therapeutics including pre-clinical and clinical agents that target epigenetic enzymes in CRC are listed inTable 1. Thus far, more than 20 histone-methylation enzymes have been found to be clinically relevant to CRC, including 17 oncoproteins and 8 tumor suppressors, although their exact mechanisms of action are not fully understood [15]. Furthermore, more than 20 small-molecule inhibitors targeting HMTs, DMNTs, and HDMs have been employed for preclinical or clinical studies. For example, treatment of DLD1 colon cell line and primary CRC cells with a potent HKMT inhibitor EPZ004777 (anti-DOT1L) resulted in significant reduction in sphere formation in vitro, thus inhibiting cell growth [32]. Other HKMT inhibitors, such as BCI-121 and Chaetocin, have significantly suppressed CRC cell growth and migration by inhibiting SMYD3and SUV39H1, respectively [98]. Notably, inhibitors against the HKMT EZH2have yielded some of the most promising results for treating CRC. DZNep, an indirect EZH2inhibitor, induced apoptosis in CRC cell lines and stem cells, while GSK346 impaired the migratory potential of CRC cells and reduced H3K27me3 levels in Colo205 and HT-29 cells (Table 1) [110].

Table 1.

Overview of pre-clinical and clinical drugs that target epigenetic enzymes in CRC.

Unlike HKMTs, the development of inhibitors against PRMTs has only recently gained prominence in the cancer field, and only a couple of these have made it to the clinical trial phase thus far. AMI-1, which inhibits PRMT1and PRMT5, demonstrated antiproliferative activity in CRC cells and xenograft mouse models [106]. However, further in vivovalidation studies are needed, and it has not entered clinical trial yet. Another promising PRMT5inhibitor that recently made it to Phase I clinical trials is GSK3326595, which potently inhibited tumor growth in vitroand in vivo[111]. Trials with GSK3326595 are currently being conducted in adult subjects with relapsed and/or refractory solid tumors (NCT02783300). Additionally, inhibitors targeting HDMs are even fewer in number and have shown limited efficacy in suppressing CRC cell growth. For example, KDM4A/Cinhibitors were ineffective in blocking HCT116 CRC cell growth when used in isolation [112]. However, they exhibited potent antiproliferative effects in combination with another HDM inhibitor, NCL-2, which targets LSD1[113]. These data suggest a potential for synergy between the two classes of HDM inhibitors.

Finally, the use of DNMT inhibitors for CRC treatment has also shown some exciting promise. In studies using CRC cell lines, suppression of DNMT1and DNMT3Bresulted in significant reduction in methylation, which correlated with the re-expression of tumor suppressor genes. This also resulted in induction of apoptosis as well as reduced cell proliferation and stemness [114]. Notably, studies with the DNMT1inhibitor, 5-aza-2′-deoxycytidine (decitabine), exhibited its ability to re-sensitize colorectal tumors to both irinotecan and 5-FU, thus becoming a major component of the treatment regimen for CRC in the clinic [19]. Another recent preclinical study showed that combination of the anti-EGFRinhibitor, gefitinib and decitabine showed highly synergistic inhibition of CRC cell proliferation and migration [115]. Additional combination regimens are outlined in Table 1.

4.3 Acetyltransferases and deacetylases

Acetylation of histones by acetyltransferases (HATs) and removal of these acetyl marks by HDACs are essential events for the maintenance of normal chromatin organization and function [116]. However, as is often the case in cancer, these enzymes are dysregulated, leading to increased chromosomal instability and aberrant gene expression changes [117]. To date, only a handful of HATs have been reported as contributing to the pathogenesis of CRC. Here, we describe the role of a few of these HATs namely p300/CREB-binding protein(p300/CBP), GCN5, N-Acetyltransferase 10(Nat10), and Human males absent on the first(hMOF). Assessment of 262 CRC samples from patients receiving 5-FU treatment demonstrated that low expression of p300/CBPin CRC tissue was closely associated with poor clinical response to 5-FU based-chemotherapy [118]. Furthermore, low p300/CBPexpression also correlated with poor disease-free survival and increased early disease progression in the same patients [118]. Mechanistic studies also uncovered that 5-FU induced degradation of p300/CBPwhich was dependent on chaperone-mediated autophagy involving heat-shock cognate protein 70 kDa(hsc70) and lysosomal-associated membrane protein 2A(LAMP2A). In short, degradation of p300/CBPwas found to be relevant to chemoresistance to 5-FU, since blocking this degradation also enhanced 5-FU’s cytotoxicity in CRC cells [118].

Conversely, another HAT GCN5has been implicated in promoting CRC cell growth via its upregulation rather than downregulation. One study found that GCN5overexpression in human colon adenocarcinoma tissues was attributed to the activities of the transcription factors, c-Mycand E2F transcription factor 1(E2F1) [119]. Depletion of c-Mycinhibited CRC cell proliferation mainly by downregulating GCN5 transcription, an effect that was rescued by ectopic expression of GCN5. However, ectopic overexpression of E2F1had the opposite effect by suppressing GCN5levels, thus inducing cell death. Furthermore, inhibition of GCN5with CPTH2, a HAT inhibitor, also suppressed CRC cell growth, revealing an avenue of great therapeutic potential [119]. Other HATs implicated in CRC include Nat10and hMOF, which were downregulated in CRC tissues. Particularly, recent studies showed that Nat10downregulation and subcellular redistribution were associated with increased cellular motility and invasion in CRC cells [120]. Meanwhile, low expression of hMOFcorrelated with clinicopathological features of CRC such as lymph node metastasis and advanced tumor stage [121].

In CRC, HDACs are also frequently overexpressed and represent another attractive class of targets for anticancer therapy. HDAC1–3and HDAC5–8have emerged as some of the most relevant deacetylases in CRC. Although all are highly overexpressed in CRC, only few studies have explored the relevance of this overexpression to disease [23]. For example, knockdown of HDAC1, 2, and 3reduced the growth of several CRC cells by largely unknown mechanisms [122]. Interestingly however, a mechanistic link between HDAC2expression and sensitivity of CRC cells to other anticancer agents was recently established. Alzoubi et al. demonstrated that depletion of HDAC2specifically enhanced the combined anti-tumor effect of the pan-HDAC inhibitor suberoylanilide hydroxamic acid (SAHA) and the DNA-damaging agents, 5-FU and oxaliplatin, in SW480 and HT29 cells. On the other hand, overexpression of HDAC2conferred resistance to these agents, which were independent of the p53 mutational status [123]. In summary, these findings strongly suggest that HATs and HDACs are critical biomarkers for CRC and influence the sensitivity of CRC cells to certain therapeutics as evidenced by their frequent combination with other anticancer agents (Table 1).

4.4 Targeting HATs and HDACs in CRC

Like methylation, several studies have demonstrated that inhibitors targeting HATs and HDACs also induce epigenetic alterations that modulate the expression of genes or pathways critical for CRC treatment. One study showed that direct inhibition of p300/KAT3Bhistone acetyltransferase, a coactivator of β-Cateninwith rimonabant, induced downregulation of Wnt/β-catenintarget genes in HCT116 cells [124]. Furthermore, rimonabant also significantly reduced tumor growth in HCT116 xenografts [124]. The general HAT inhibitors such as garcinol and anacardic acid have also been shown to inhibit CRC tumor growth in mice as well as sensitize cancer cells to irradiation [125].

Compared to HATs, a far greater number of studies have been dedicated to investigating the efficacy of HDAC inhibitors at both the preclinical and clinical level. Overall, the use of HDAC inhibitors in preclinical solid tumor models has shown some early promise albeit their progress to the clinic has been hindered by serious limitations including ineffective concentrations and cardiac toxicity [126]. In CRC specifically, these inhibitors are mainly being administered as combination therapy with conventional chemotherapy or other agents [127]. In pre-clinical models for instance, treatment of irinotecan-resistant CRC cells with HDAC inhibitors conferred sensitization of these cells to irinotecan, whereas HDAC inhibitor CG2 showed an additive effect when used with irinotecan, 5-FU, or oxaliplatin in HCT116 xenografts [128]. Meanwhile, a small molecular inhibitor of HDAC6, ACY-1215, was able to enhance the anticancer activity of oxaliplatin by promoting apoptosis and blocking cell proliferation in CRC cells and xenograft models [129].

4.5 Benefits and pitfalls of epigenetic enzyme inhibitors

Despite their potential, a large gap still remains between the biological activity of epigenetic enzyme inhibitors in preclinical studies and their potential clinical utility. For example, the development of HAT inhibitors poses several challenges [130]. Because of their function in complexes consisting of many proteins which play multiple roles in HAT target specificity, this significantly limits inhibitor use in vivo[130]. Additionally, many undesired effects such as reactivity or lack of selectivity between different HAT subtypes are often associated with HAT inhibitors [130]. Hence, studies geared towards the development of more potent and selective inhibitors by carefully studying the catalytic mechanism and enzyme kinetics of various HATs are needed. As far as HDAC inhibitors are concerned, they have shown preferential efficacy against hematological malignancies, and therefore, drugs such as vorinostat (SAHA) and romidepsin (FK228) have achieved FDA approval for the treatment of cutaneous T cell lymphoma [131, 132]. Unfortunately, the evidence regarding HDAC inhibitors efficacy for solid tumors has not been as convincing and encouraging although they are well tolerated at low but not high doses. Currently, adverse side effects and inadequate clinical efficacy are the major limitations to their use, and more efforts are underway to generate specific HDAC inhibitors for solid tumors such as CRC [133]. Nonetheless, a few early phase clinical trials using vorinostat in combination with other chemotherapeutic agents have shown some early promise for mCRC patients. These include combinatorial regimens of vorinostat with 5-FU, leucovorin, and oxaliplatin (FOLFOX) as well as randomized phase II trial studies investigating the efficacy of vorinostat and hydroxychloroquine or regorafenib in refractory mCRC patients [134, 135]. Other regimens are outlined in Table 1. Similarly, while DNMT inhibitors have also met with some degree of success for treating blood cancers such as myelodysplastic syndrome (e.g., decitabine and 5-azacitidine), the major drawbacks of these compounds in solid tumors are harsh side effects and transient demethylation, which revert after drug removal [136, 137]. Interestingly, however, some studies have suggested that this transient demethylation that occurs with DNMT inhibition (e.g., 5-azadeoxycytidine) potentially creates a therapeutic window that can be leveraged for epigenetic reprogramming and/or combinatorial therapies with cytotoxic drugs [138].

Other general limitations regarding the use of epigenetic therapy in solid tumors deal with the unfavorable pharmacokinetic properties of these drugs, including instability, toxicity, and short half-life [137]. Some of these invariably contribute in some way to common toxicities associated with HDAC and DNMT inhibitors in CRC including thrombocytopenia, neutropenia, diarrhea, nausea, vomiting, and fatigue [139]. Furthermore, maintaining therapeutically relevant levels of the drugs necessary for clinical benefit is particularly difficult, and as of yet, no FDA-approved epigenetic treatments exist for CRC despite promising preclinical studies. This signifies the overall marginal clinically compelling responses to these agents in CRC patients. To overcome some of these limitations, newer formulations have been made to render these inhibitors more bioavailable, stable, and ultimately usable at lower doses with less toxicity and greater therapeutic efficacy. Examples of these include the oral HDAC inhibitor, entinostat, used in in vitroand in vivomodels of CRC and an orally active formulation of 5-azacitidine, cc-486 [140].

4.6 Emerging immunomodulatory and epigenetic combinatorial therapies

Compared to MSS tumors, there are an exponentially higher number of mutations acquired in MSI-H CRCs. Interestingly, these mutations have the potential to elevate the production of neo-antigens [141]. The result is increased tumor immunogenicity, which is further complemented by the fact that these tumors also harbor a high number of tumor-infiltrating lymphocytes. Within this context, CRC patients with MSI-H represent a subgroup more likely to benefit from immune checkpoint inhibitors compared to those with MSS tumors. Immune checkpoint inhibitors have shown unprecedented benefit across multiple tumor types. These agents specifically target the proteins programmed death ligand-1(e.g., durvalumab) and programmed death-1(e.g., nivolumab, pembrolizumab) and are administered as monotherapies or in combination with other anticancer agents. At the present time, several ongoing early and late-phase II and III clinical trials investigating the efficacy of immune checkpoint inhibitors in MSI-H and MSS CRC patients are being extensively explored including pembrolizumab (Keytruda), which recently obtained FDA approval (e.g., NCT01876511 and NCT02060188) [142].

Moreover, the possibility of combining epigenetic therapy and immunotherapy has also been recently explored, and several ongoing clinical trials in CRC investigating the combination strategies of HDACi and DNMTi with checkpoint inhibitors have been undertaken. Specifically, these epigenetic therapies have been shown to augment the effect of checkpoint inhibitors and are currently in early and late phase clinical trials [143]. However, since MSS subtypes represent the larger fraction of CRC cases, the marginal activity displayed by drugs such as pembrolizumab for treating MSS CRCs has been less than encouraging [144]. Hence, overcoming the clinical ineffectiveness of this class of drugs for this subtype remains an important need. Intriguingly, however, recent studies showed that treatment with 5-azacitidine and entinostat in CRC cell lines conferred a shift towards a CIMP+ signature, which would predictively convert them into a more immunogenic state [145]. This increased sensitivity to immunotherapy has prompted a clinical trial evaluating this strategy, with the combination regimen of romidepsin (HDAC inhibitor) and cc-486 with pembrolizumab in MSS-CRC patients (NCT02512172) [145]. Finally, romidepsin was also found to potentiate 5-FU cytotoxicity in HCT-116, HT29, and SW48 cells by inducing apoptosis and cell cycle arrest [146]. Interestingly, MHC class II gene expression was also induced with this combination, once again supporting the possible cooperation of epigenetic therapy with immunomodulatory agents [146]. In summary, the above evidences support a cooperative role between epigenetic and immune therapies, although further efforts to optimize the epigenetic control of immune-related gene expression will be necessary to successfully translate these notions to the clinic.