Epigenetic Alterations in Colorectal Cancer: Emerging Biomarkers

Abstract

Colorectal cancer (CRC) is a leading cause of cancer deaths worldwide. One of the fundamental processes driving the initiation and progression of CRC is the accumulation of a variety of genetic and epigenetic changes in colon epithelial cells. Over the past decade, major advances have been made in our understanding of cancer epigenetics, particularly regarding aberrant DNA methylation, microRNA (miRNA) and noncoding RNA deregulation, and alterations in histone modification states. Assessment of the colon cancer “epigenome” has revealed that virtually all CRCs have aberrantly methylated genes and altered miRNA expression. The average CRC methylome has hundreds to thousands of abnormally methylated genes and dozens of altered miRNAs. As with gene mutations in the cancer genome, a subset of these epigenetic alterations, called driver events, is presumed to have a functional role in CRC. In addition, the advances in our understanding of epigenetic alterations in CRC have led to these alterations being developed as clinical biomarkers for diagnostic, prognostic and therapeutic applications. Progress in this field suggests that these epigenetic alterations will be commonly used in the near future to direct the prevention and treatment of CRC.

INTRODUCTION

Colorectal cancer (CRC) is common worldwide and remains the second leading cause of cancer-related deaths in Western countries¹. Despite recent improvements in screening strategies and the development of more effective treatments for CRC, the prognosis of advanced CRC is still poor. Furthermore, the newest line of molecularly-targeted therapeutic agents, appear to only have activity in metastatic CRC and do not cure the patient, but they have exponentially increased treatment costs and the economic burden of CRC care. Therefore, robust diagnostic, prognostic and predictive biomarkers are clearly and urgently needed to detect advanced colon polyps and early stage CRC, which are most effectively treated with current therapies, and to identify the most effective treatments for specific CRC patients.

Gene mutations have long been known to be important in cancer formation. However, epigenetic alterations have only recently been recognized as significant contributors to cancer development. “Epigenetics” was first described by the developmental biologist Conrad H. Waddington, in 1942, as “the study of heritable changes in gene expression mediated by mechanisms other than alterations in primary nucleotide sequence of a gene” and is now considered as broadly referring to heritable alterations in gene expression that are not mediated by changes in the DNA sequence^{2, 3}. Epigenetic alterations frequently found in cancer include aberrant DNA methylation, abnormal histone modifications, and altered expression levels of various non-coding RNAs, including microRNAs (miRNAs). With regards to the role of epigenetic alterations in the normal-polyp-cancer sequence, as with gene mutations, it appears that a subset of the hundreds-thousands of alterations found in the typical cancer cell drives the initiation and progression of CRC formation through the sequential accumulation of genetic and epigenetic changes in key tumor-suppressor and oncogenes⁴.

The prevailing consensus suggests that epigenetic alterations in CRC occur early and manifest more frequently than genetic alterations. In addition, advances in genomic technologies have led to the identification of a variety of specific epigenetic alterations as potential clinical biomarkers for CRC patients. This review briefly outlines the fundamental basis of epigenetic alterations in cancer, and details the current state of the field regarding the promise and clinical usefulness of various epigenetic alterations as biomarkers for early detection, diagnosis, prognosis and management of CRC patients.

DNA METHYLATION

Overview of DNA Methylation in Cancer

The most widely studied epigenetic alteration in cancer is aberrant DNA methylation. Although DNA hypermethylation has received the most attention recently, DNA hypomethylation was the first reported DNA methylation abnormality in human cancer (1983)⁵. As illustrated in Figure-1, at this time global DNA hypomethylation was identified in both colorectal adenomas and CRC⁶. It was not until several decades later, that Baylin and colleagues identified site-specific hypermethylation- of the Calcitonin (CALCA) gene in lung cancer and lymphoma⁷. Later, RB (Retinoblastoma), a known tumor suppressor gene, was found to silenced by aberrant DNA methylation, providing support for a functional role for epigenetic alterations in oncogenesis^8–10. This landmark discovery was followed by the discovery of other hypermethylated tumor suppressor genes in cancer, including CDKN2A, MLH1, and CDH1^11–13.

Figure 1.

A historical perspective illustrating key milestones associated with the discovery of various epigenetic alterations in colorectal cancer from 1983 to the present. Individual epigenetic alterations are listed in color-coded boxes; aberrant DNA methylation (red), non-coding RNAs including microRNAs (green), and histone modifications (yellow).

DNA methylation is one of a host of epigenetic modifications that can regulate gene expression. In humans, DNA methylation occurs at cytosine residues that precede guanines, called CpG dinucleotides (C-phosphodiester-G bond)^{14, 15}. The majority of CpG dinucleotides in the human genome are methylated, however, there are CpG rich sequences called CpG islands, which are typically unmethylated in normal healthy cells. CpG islands are found in the promoter regions of ~40–60% of tumor suppressor genes, are generally 200–2000 bps long, have a CG content >50%, and are involved in the regulation of gene expression^14–16. Methylated CpG dinucleotides are usually found in gene bodies and large repetitive sequences (LINE-1, SINE/Alu sequences etc.) that are found in centromeres and other retrotransposon elements^17–19. In one sense, this normal methylation pattern is essentially reversed as cancer cells.

DNA methylation is mediated by DNA methyltransferases (DNMTs) that facilitate the catalytic addition of methyl groups to the 5^th position of the cytosine rings of CpG dinucleotides. DNA methyltransferases are classically considered to be either maintenance methyltransferases (e.g. DNMT1), which preserve existing methylation patterns during DNA replication, or de novo DNMTs (e.g. DNMT3A and DNMT3B), which preferentially catalyze the methylation of previously unmethylated CpGs³. Although the way in which methylation represses gene transcription is still under investigation, proposed mechanisms include alterations in chromatin complexes and the recruitment of methyl-CpG domain-binding proteins (MBD) around the CpG islands of the corresponding gene(s)^{20, 21}. MBD proteins at gene promoters impede access of the regulatory proteins required for active gene transcription. DNA methylation can directly inhibit cis-binding elements, including the transcriptional factors Activating Protein 2 (AP-2), CCAAT enhancer-binding protein C/EBF, cAMP response element-binding protein (CREB), E2 promoter binding factor (E2F), and nuclear factor kappa-light-chain-enhancer of B cells (NF-κB). Furthermore, emerging evidence indicates that extensive DNA methylation changes at CpG island “shores” ― regions with relatively lower CpG density and located within 2 kb of CpG islands, strongly correlates with loss of gene expression^22–24. CpG island shores are methylated in a tissue-specific manner, and contain 70% of the differentially methylated regions involved in cellular reprogramming. The mechanism behind differential methylation of CpG shores regulating gene expression remains poorly understood and is an area of active investigation.

Aberrant DNA Hypermethylation in Colorectal Cancer

Advances in our understanding of the molecular pathogenesis of CRC led to the initial observation that these neoplasms primarily arise through two major molecular pathways of genomic instability ― chromosomal instability (CIN) and microsatellite instability (MSI). However, more recently a third class of CRCs characterized by a high frequency of DNA hypermethylation has been identified. These cancers have been defined as having a “CpG island methylator phenotype (or CIMP)”, and their identification provided a significant advance in our understanding of the molecular mechanisms that orchestrate colorectal tumor formation.

The CIMP was first described in colorectal tumors in 1999 and was characterized as having an exceptionally high frequency of hypermethylated CpG dinucleotides²⁵. Weisenberger and colleagues later introduced the prevailing method used to identify CIMP in CRC, which is based on the methylation status of five genes, CACNA1G, IGF2, NEUROG1, RUNX3, and SOCS1²⁶. CIMP-positive tumors exhibit unique clinicopathological and molecular features, including a predilection for proximal location in the colon, female gender, poor and mucinous histology, and the presence of frequent KRAS and BRAF mutations^{25, 27–30}.

MSI generally results from inactivation of the DNA mismatch repair (MMR) system through hypermethylation (80% of MSI CRCs) or mutations in the genes MLH1, MSH2, MSH6, and PMS2 (20% of MSI CRCs). Inactivation of these genes results in the accumulation of DNA replication errors in repetitive microsatellite sequences, some of which are located in the exons of potential tumor suppressor genes. MSI CRCs represent ~12–15% of all tumors and >90% of familial Lynch syndrome CRCs, which account for 2–3% of all MSI CRC^{31, 32}. The remaining MSI CRCs (10–12%) are sporadic^{33, 34}. Eighty percent of sporadic CRCs with MSI harbor biallelic hypermethylated MLH1 alleles^{12, 26},^{12, 25}.

While there exists a significant overlap in the clinical features between sporadic MSI and Lynch syndrome cancers, patients with sporadic MSI have an older age of onset, and higher frequency of the BRAF V600E mutations and hypermethylated MLH1, while Lynch patients are generally younger and harbor rare BRAF mutations ^{35, 36}.

Aberrant DNA Methylation Occurs Early in “Traditional” and “Serrated” Polyp Pathways

The conventional model of CRC formation as initially proposed by Fearon and Vogelstein describes a stepwise normal-adenoma-cancer progression and considers adenomatous polyps as the principal pre-neoplastic lesions leading to CRC^{4, 37}. As described previously³⁸, the transition from normal mucosa to adenomatous polyp is marked by both genetic and epigenetic alterations, some of which deregulate central molecular pathways³⁹. These epigenetic alterations include hypermethylation of a variety of genes, such as SLC5A8, ITGA4, SFRP2, PTCH1, CDKN2A, HLTF, and MGMT, and some of these play a role in the initiation and progression of adenomas to CRC^{11, 40–45}. The identification of methylated genes in colon polyps has led to their use as biomarkers in early detection assays for CRC⁴⁶.

More recently recognized, the “serrated pathway” is another route for the formation CRC. It was originally described as “an alternative pathway” to the conventional adenoma-to-CRC pathway because of the unique morphological and histological characteristics of the sessile serrated polyps that give rise to CRCs via this route⁴⁷. Mutations in BRAF and KRAS, along with CIMP, are common features of polyps in the serrated pathway. In contrast to classic adenomatous polyps, sessile serrated polyps and traditional serrated adenomas do not typically have genetic alterations in APC or CTNNB1 (gene for β-catenin)^48–50 indicating that these two pathways are distinct and employ different molecular processes. In addition to BRAF/KRAS mutation-induced activation of the MAPK-ERK pathway⁵¹, serrated polyps evolve through methylation-mediated transcriptional inactivation of various genes belonging to the β-catenin/WNT pathway (SFRP family, CDX2, MCC)^52–54, p53 signaling pathway (IGFBP7)⁵⁵, cell cycle control proteins (CDKN2A)⁵⁶, and DNA Mismatch repair (MLH1)⁵⁵ family.

Aberrant DNA Hypomethylation in Colorectal Cancer

While DNA hypermethylation can silence tumor-suppressor genes, global DNA hypomethylation is believed to influence CRC development by inducing chromosomal instability and global loss of imprinting (Figure-2A)^57–60. Genome-wide hypomethylation generally occurs within repetitive transposable DNA elements such as the LINE-1 or short interspersed nucleotide elements (SINE, or Alu) sequences in many cancers, including CRC^61–65. LINE-1 hypomethylation inversely associates with MSI and/or CIMP^{65, 66}. Furthermore, a number of studies have demonstrated that a high degree of LINE-1 hypomethylation correlates with worse patient survival^67–70. One hypothesis is that hypomethylation of LINE/SINE sequences may induce inadvertent activation of potential proto-oncogenes⁷¹, which implies that LINE-1 hypomethylation has a functional role in CRC formation⁷².

Figure 2.

An illustration of various epigenetic alterations in colorectal cancer. A) This figure illustrates the concept of aberrant DNA hypermethylation in the context of a “cancer cell”. Double helix DNA represents a tumor suppressor gene, with CpG islands and CpG shores in its promoter region. Hypermethylation of CpG sites (shown as black lollipops) leading to gene silencing and closed chromatin in the cancer cells is shown. In contrast, CpG dinucleotides within introns as well as in intergenic regions are frequently hypomethylated, which may lead to the increased expression of oncogenes and oncomiRNAs, and resulting open chromatin conformation. B) This figure illustrates histone modifications in a cancer cell. The left panel depicts heterochromatin, which is a closed chromatin conformation that is often associated with DNA methylation and inactive gene transcription. In contrast, the euchromatin state is in an open conformation and associates with active gene transcription, presumably secondary to increased transcription factor binding. C) A schematic demonstrating miRNA biogenesis in cancer cells and how miRNAs inhibit and/or cause degradation of their mRNA targets. D) This figure illustrates various activities of long-noncoding RNAs (lncRNAs) in cancer cells, including i) their ability to regulate chromatin conformation; ii) induce transcription by binding to appropriate transcription factors; iii) function as decoy and inhibit gene transcription; iv) act as miRNA sponges; v) cause mRNA decay and vi) induce mRNA translation.

With regards to the mechanism responsible for cancer associated DNA hypomethylation, the prevailing belief is that it results from a passive process secondary to inadequate maintenance of methylation during DNA replication in most cancers. Recent studies have demonstrated that DNA hypomethylation in cancer can be secondary to an active process in some tumors. The ten-eleven translocation enzymes (TET) can catalyze the formation of 5-hydroxymethylcytosine (5-hmC) from 5-methylcytosine, which is recognized by the Base Excision Repair (BER) proteins for removal and replacement with unmethylated cytosine. The ten-eleven translocation (TET) family members TET1 and TET2 are mutated in leukemia and a biochemically associated enzyme, isocitrate dehydrogenase (IDH1) is mutated in some gliomas. It has been proposed that mutant TET1 and TET2 and mutant IDH1 are the mechanisms for DNA hypermethylation in leukemias and gliomas, respectively.^73–75. Of interest, global loss of 5-hmC is reported in several types of solid tumors, including CRC⁷⁶. Recently, downregulation of TET1 expression was found in early stages of colon cancers, and loss of its expression was shown to inhibit WNT signaling pathway and suppression of tumor proliferation^{77, 78}. However, the biological significance of 5-hmC and the TET proteins with regards to LINE-1 hypomethylation in CRC still remains unclear, and requires further investigation. IDH1 and TET1 or TET2 mutations do not appear to be a common cause of aberrant DNA methylation in CRC⁷⁹.

Methylated DNA as a Biomarker for Colorectal Cancer

The term “biomarker” or “biological marker”⁸⁰ was defined in 1998 by The National Institutes of Health Biomarkers Definition Working Group as “a characteristic that is objectively measured and evaluated as an indicator of normal biological and pathological processes, or pharmacologic responses to a therapeutic intervention”⁸¹. However, the definition of biomarker remains dynamic, and is constantly under revision in response to our evolving understanding of cancer. Based on current clinical criteria, CRC biomarkers can be described as substances: a) that are measured easily and inexpensively to identify a patient’s cancer; b) that identify a patient’s prognosis independent of conventional classifications (i.e. TNM classification and tumor markers) to improve their outcome; and c) that predict a patient’s response to a specific treatment, thereby improving their prognosis and quality of life.

Methylated DNA as a Diagnostic Biomarker

Despite recent advances in treatment options, early detection and removal of precancerous lesions (advanced colorectal adenomas, A-CRA) remains the most effective strategy for reducing mortality associated with CRC. However, current screening modalities are limited by invasiveness, expense and poor patient compliance, leading to the late detection of cancers and subsequently unfavorable prognoses. The most widely used fecal screening tests, FOBT and FIT, have suboptimal diagnostic accuracy, highlighting the need for more robust and reliable non-invasive biomarker assays for the detection of early CRC.

During early stages of colorectal carcinogenesis, epigenetic alterations appear to exceed the frequency of genetic mutations, suggesting their greater potential for the next generation of diagnostic biomarkers for the detection of colonic polyps and cancers.

Stool-based biomarkers

Specific biofluids, such as blood (plasma or serum) and feces are the most common analytes used in CRC screening tests. Since the initial discovery by Sidransky and colleagues of mutant KRAS in fecal specimens from patients with CRC⁸², numerous studies have supported using fecal DNA for potential screening assays for the early detection of CRC. Studies of methylated SFRP2, SFRP5, PGR, CALCA, and IGFBP2, in fecal DNA from two independent cohorts identified methylated SFRP2 as a diagnostic biomarker for CRC detection with high sensitivity (77–90%) and specificity (77%)⁸³. A later study using 111 fecal specimens, including 21 colorectal adenomas, demonstrated that methylated SFRP2 can also identify patients with precancerous colonic polyps⁸⁴.

Another well-studied fecal DNA biomarker for CRC is methylated VIM, the gene for Vimentin. Methylated VIM specifically occurs in CRC tissues and is detected in fecal DNA with reasonably high sensitivity (46%) and specificity (90%)⁸⁵. A number of studies have demonstrated the potential for using methylated VIM as a biomarker for the early detection of CRC^{86, 87}, and these studies led to the development of an assay that detects methylated VIM as one of the first commercial fecal-DNA screening test for CRC (ColoSure™, Lab Corp, Burlington, NC). To date, a large number of hypermethylated genes including APC, ATM, BMP3, CDKN2A, SFRP2, GATA4, GSTP1, HLTF, MLH1, MGMT, NDRG4, RASSF2A, SFRP2, TFPI2, VIM, and WIF1 have been analyzed in fecal DNA for the early detection of CRC^{85, 88–95}.

As illustrated in Table 1, a large number of stool-based methylation biomarkers, for use in CRC early detection assays have been identified in Phase I and II biomarker studies⁹⁶. Perhaps of most clinical importance, this line of investigation has culminated in the development of an FDA approved, clinically available stool-based CRC screening test – Cologuard^® (Exact Sciences Corporation). This stool DNA based assay, which detects methylated BMP3, methylated NDRG4, and mutant KRAS, was recently compared to FIT and to colonoscopy in nearly 10,000 patients enrolled at 90 sites⁹⁷. This fecal DNA based assay showed an overall sensitivity of 92% (95% CI, 83–97.5%) for CRC and 93% (95% CI 83.8–98.2%) for stage I-III CRC, compared to sensitivity of FIT at 74% (95% CI, 61.5–84%) and 73% (95% CI, 60.3–83.9%), respectively (p=0.002). For advanced adenomas and sessile serrated polyps, the sensitivity of the test increased proportionately with lesion size and grade. The molecular assay was significantly more sensitive than FIT for advanced adenomas: 42% (95% CI, 38.9–46%) vs. 24% (95% CI, 20.8–27%), respectively, for those ≥1 cm and 66% vs 43% for those ≥2 cm (p< 0.001). Sessile serrated polyps ≥1 cm were detected at a rate of 42% for the molecular assay compared 5% for FIT (p< 0.001); In this study the test specificity was based on the primary and secondary study endpoints, specifically the detection of CRC and advanced pre-cancers. For these endpoints combined, the specificity was 87%⁹⁷.

Table 1.

List of aberrantly methylated genes/loci as potential non-invasive markers for diagnosing colorectal cancer

Diagnostic markers
Markers	Source	Study Design	Sample size		Target: Adenoma or CRC	Sensitivity (%)	Specificity (%)	Reference
			Cases	Control
Blood-based biomarker
ALX4	Serum	R	30	52	CRC	83.3	70	Ebert et al. 109
ALK4	Plasma	R V	28/5 (CRC), 45/49 (polyp)	55/22	CRC, AD	40 (CRC)/45(polyp)	82	Tänzer et al. 295
HLTF	Serum	R	49	41	CRC	32.7	92.7	Leung et al. 110
HLTF	Serum	R	38	20	CRC	55	100	Wallner et al. 111
hMLH1	Serum	R	19	-	CRC	33	100	Grady et al. 296
hMLH1	Serum	R	49	41	CRC	42.9	97.6	Leung et al. 110
hMLH1	Serum	R	38	20	CRC	18.4	100	Wallner et al. 111
HPP1	Serum	R	38	20	CRC	21.1	100	Wallner et al. 111
NEUROG1	Serum	R V	252 (15/95/45/97)	93 (32/-/16/45)	CRC	61	91	Herbst et al. 112
SEPT9	Plasma	R	133	179	CRC	69	86	Lofton-Day et al. 106
SEPT9	Plasma	R V	252/126 (CRC),168 (polyp)	102/183	CRC, AD	55.5(CRC), 14.9(AD)	91	Grützmann et al. 297
SEPT9	Plasma	R V	97/90	172/155	CRC	72/68	93/89	deVos et al. 298
SEPT9	Plasma	R V	28/5 (CRC), 45/49 (polyp)	55/22	CRC, AD	73(CRC), 29 (AD)	91	Tänzer et al. 295
SEPT9	Plasma	PT V	50 (CRC) / 104 (AD)	94/53	CRC/AD	90(CRC)/12 (AD)	88/97	Warren et al. 299
SEPT9	Plasma	R	30 (CRC), 22 (AD)	49	CRC, AD	60(CRC), 14 (AD)	73	Ahlquist et al. 300
SEPT9	Plasma	PT	53 (CRC), 315 (AD)	938	CRC, AD	48.2(CRC), 11.2(AD)	91.5	Church et al. 107
SEPT9	Plasma	R	34 (CRC), 26 (AD)	24	CRC, AD	88.2(CRC), 30.8(AD)	91.7	Tóth et al. 301
SEPT9	Plasma	PT	1544		CRC	68	80	Potter et al. 302
SEPT9	Plasma	PT V	102/2(CRC), 26(AD)	-/94	CRC	73.3	81.5	Johnson et al. 303
Vimentin	Plasma	R	81	110	CRC	59	93	Li et al. 113
APC, MGMT, RASSF2A, WIF1 (panel)	Plasma	R	243 (CRC), 64 (AD)	276	CRC, AD	86.5(CRC), 74.6(AD)	92.1(CRC), 91.3(AD)	Lee et al. 95
TMEFF2	Plasma	R	133	179	CRC	69	65	Lofton-Day et al. 106
NGFR	Plasma	R	133	179	CRC	51	84	Lofton-Day et al. 106
Stool-based biomarker
ALX4	Stool	R	90 (CRC+AD)	157	CRN (CRC+AD)	10.6	98.7	Amiot et al. 304
CDKN2A	Stool	R	29	25	AD	31	84	Petko et al. 93
CDKN2A	Stool	R	25	20	CRC	20	100	Abbaszadegan et al. 305
CDKN2A	Stool	-	30(CRC), 25(AD)	31	CRC, AD	40 (CRC), 24 (AD)	96.8 (panel)	Chang et al. 306
GATA4 (Train set/Valid set)	Stool	PR V	28/47	45/30	CRC	71/ 51	84/93	Hellebrekers et al. 91
ITGA4	Stool	R	13	28	AD	69	79	Ausch et al. 307
ITGA4	Stool	-	30(CRC), 25(AD)	31	CRC, AD	36.7(CRC), 16 (AD)	96.8 (panel)	Chang et al. 306
MGMT	Stool	R	29	25	AD	48	73	Petko et al. 93
MGMT	Stool	R	52(CRC), 21(AD)	24	CRC, AD	48.1 (CRC), 28.6 (AD)	100	Huang et al. 308
MGMT	Stool	R	60(CRC) 52(AD)	37	CRC, AD	51.7 (CRC), 36.5 (AD)	-	Baek et al. 309
NDRG4 (Train set/ Valid set)	Stool	R V	28/46	45/30	CRC	61/53	93/100	Melotte et al. 92
SFRP2 (Train set/ Valid set)	Stool	R	10/13	13/13	CRC	90/77	77/77	Muller et al. 83
SFRP2	Stool	R	52(CRC), 21(AD)	24	CRC, AD	94.2 (CRC), 52.4 (AD)	93	Huang et al. 308
SFRP2	Stool	R	69(CRC), 34(AD)	30	CRC, AD	87.0 (CRC), 61.8 (AD)	93.3	Wang et al. 89
SFRP2	Stool	R	29 (AD)	6	AD	46	100	Oberwalder et al. 310
SFRP2	Stool	R	84(CRC), 56(AD)	113	CRC, AD	63.1 (CRC), 32.1 (AD)	92	Nagasaka et al. 311
SFRP2	Stool	-	30(CRC), 25(AD)	31	CRC, AD	60 (CRC), 44 (AD)	96.8 (panel)	Chang et al. 306
SFRP2	Stool	R	48 (CRC), 35 (AD)	30	CRC, AD	56.3(CRC), 51.4(AD)	100	Zhang et al. 312
TFPI2 (Pilot/Training/Validation)	Stool	R V	11/26/47(CRC), 7/19(AD)	12/45/30	CRC, AD	73/89/76(CRC), 43/- /21(AD)	100/79/93(CRC), 100/-/93(AD)	Glockner et al. 90
TFPI2	Stool	R	60(CRC), 20(AD)	30	CRN (CRC+AD)	68.3	100	Zhang et al. 313
RASSF2	Stool	R	84(CRC), 56(AD)	113	CRC, AD	45.3(CRC), 12.6(AD)	94.7	Nagasaka et al. 311
Vimentin	Stool	R	94	198	CRC	46	90	Chen et al. 85
Vimentin	Stool	R	40	122	CRC	72.5	86.9	Itzkowitz et al. 86
Vimentin (Train set/Valid set)	Stool	PR V	40/42	122/241	CRC	73/81	87/82	Itzkowitz et al. 87
Vimentin	Stool	R	22(CRC), 20(AD)	38	CRC, AD	41 (CRC), 45(AD)	95	Li et al. 113
Vimentin	Stool	R	60(CRC), 52(AD)	37	CRC, AD	38.3 (CRC), 15.4 (AD)	-	Baek et al. 309
Vimentin	Stool	R	9 (UC-CRC)10 (UC-dysplasia)	35 (25, UC; 10 ,CD)	UC-CRC/dysplasia	89	94	Kisiel et al. 314
Vimentin	Stool	R	90 (CRC+AD)	157	CRN (CRC+AD)	32.6	100	Amiot et al. 304
WIF1	Stool	R	90 (CRC+AD)	157	CRN (CRC+AD)	19.3	99.4	Amiot et al. 314
WIF1	Stool	R	48(CRC), 35(AD)	30	CRC, AD	60.4(CRC), 45.7(AD)	96.7	Zhang et al. 312
BMP3 NDRG4 TFPI2 Vimentin KRAS(mutation) (panel)	Stool	R V	170/82 (CRC), 89/44 (AD)	197/96	CRC, AD	85(CRC), 56(AD)/78(CRC), 48(AD)	90/85	Ahlquist et al. 315
BMP3 NDRG2 KRAS(mutation) (panel)	Stool	PT	65(CRC), 757(precancerous)	6274	CRC, AD	92.3(CRC), 42.4(AD)	86.6	Imperiale et al. 97

R, Retrospective study; PR, Retrospective analysis of prospectively collected samples in clinical trial; PT, Prospective trial; CRC, Colorectal cancer; CRN, colorectal neoplasia; AD, adenoma

Assays in bold have been approved by regulatory agencies in U.S., E.U. or China for clinical use.

Blood-based biomarkers

Due to accessibility and high patient acceptance, blood is invariably felt to be the most ideal analyte for cancer biomarkers. Table 1 summarizes aberrantly methylated genes discovered in the plasma or serum of CRC patients, which are candidate biomarkers consequently.

“Circulating DNA biomarkers” were first described in the 1970s when abnormally high concentrations of DNA were discovered in the sera of patients with various cancers⁹⁸. Initial studies focused on discovering somatic mutations^{99, 100}; however, since somatic mutations are relatively rare compared to DNA methylation alterations in the early stages of CRC tumorigenesis, a great deal of effort has gone into the development of blood-based diagnostic assays based on circulating methylated DNA.

Following the initial reports of methylated CDKN2A in circulating DNA in a variety of human cancers in 1999^101–103, a growing number of studies have examined the potential of methylated genes to be a blood-based biomarkers for CRC patients. Currently, the most established methylated DNA blood biomarker is methylated Septin 9 (SEPT9), which belongs to the gene family that encodes a group of GTP-binding¹⁰⁴ and filament-forming proteins¹⁰⁵ involved in cytoskeletal formation. Lofton-Day and colleagues first identified methylated SEPT9 as a non-invasive diagnostic biomarker for CRC¹⁰⁶. These authors reported that methylated SEPT9 had 69% sensitivity and 86% specificity for discriminating CRC patients from healthy individuals¹⁰⁶. Subsequent studies validated the clinical significance of methylated SEPT9 as a potential biomarker for CRC screening, which is now commercially-offered as a blood-based screening test in various assays including EpiproColon^® 1.0 (Epigenomics, Seattle, WA), ColoVantage^® (Quest Diagnostics, Madison, NJ) and RealTime mS9 (Abbott Laboratories, Des Plaines, IL). Furthermore, a recent prospective clinical trial (PRESEPT) demonstrated a comparable sensitivity and specificity of this assay for CRC vis-à-vis the conventional fecal occult blood test (FOBT), confirming its potential usefulness as a blood-based biomarker for CRC¹⁰⁷. This test was also recently approved the Chinese FDA for use as a CRC screening assay. However, methylated SEPT9 has a limited sensitivity for the detection of advanced adenomas (11%), underscoring the need for further improvement of this test for implementation for population-based screening of colorectal neoplasia. A recent study demonstrated that the methylated SEPT9 assay was superior to fecal immunochemical (FIT) at detecting CRC neoplasms, but both approaches were suboptimal for diagnosing patients with advanced adenomas¹⁰⁸. To date, several other blood-based diagnostic methylation biomarkers have been identified for CRC detection, including ALX4¹⁰⁹, APC⁹⁵, CDKN2A⁹³, HLTF¹¹⁰, HPP1¹¹¹, hMLH1¹¹⁰, MGMT⁹⁵, NEUROG1¹¹², NGFR, RASSF2A⁹⁵, SFRP2, VIM¹¹³, and WIF1⁹⁵. It is conceivable that a robust biomarker panel of methylated genes will be developed into a clinically viable CRC screening method in the near future.

Methylated DNA as a Prognostic Biomarker

Currently, the most accurate means for assessing CRC patient prognosis requires pathological staging of the tumor and the assessment of specific histological features of the tumor. However, the heterogeneity of survival times in patients with the same stage of CRC is well known and highlights the need for a more accurate system for determining CRC patient prognosis. As shown in Table 2, multiple, large and sufficiently powered clinical studies with independent external validation cohorts have demonstrated the feasibility of using specific methylated DNA signatures for developing prognostic biomarkers in CRC.

Table 2.

List of aberrantly methylated genes/loci as potential prognostic and predictive markers in colorectal cancer

Prognostic markers
Marker	Study Design	Source	Sample size	Stage	Status associated with poor prognosis	Assay	Survival Analysis	Reference
Tissue-based biomarker
CDKN2A	PR	Tissues	86	Duke’s A-C	Hypermethylation	MSP	Univariate	Esteller et al. 316
CDKN2A	PR	Tissues	902	I-IV	Hypermethylation	Methylight	Univariate	Shima et al. 317
CDKN2A	R	Tissues	90	II-IV	Hypermethylation	Methylight	Univariate	Maeda et al. 318
CDKN2A	R	Tissues	84	T3N0M0 (Dukes B2)	Hypermethylation	MSP	Univariate	Liang et al. 319
CDKN2A	-	Tissues	555	I-IV	-	MSP	Univariate	Ward et al. 320
CDKN2A	R	Tissues	582	I-IV	Hypermethylation	MSP	Univariate	Barault et al. 321
CDKN2A	PR	Tissues	188	IV	Hypermethylation	MSP	Univariate	Shen et al. 114
CDKN2A	R	Tissues	422	I-IV	Hypermethylation	Pyroseq and MSP	Multivariate	Bihl et al. 322
CDKN2A	R	Tissues	381 (Rectal cancer)	I-IV	Hypermethylation	Methylight	Multivariate	Kohonen-Corish et al. 323
CDKN2A	R	Tissues	285 (Rectal cancer:131)	I-IV	Hypermethylation	Pyrosequencing	Multivariate	Kim et al. 115
CDKN2A	R	Tissues	151	I-IV	Hypermethylation	qMSP	Multivariate	Mitomi et al. 324
CHFR	PR, V	Tissues	Trainig:173 (stageII:66); Validation:569 (Stage II:136)	I-IV	Hypermethylation	qMSP	Multivariate	Cleven et al. 325
EVL	R	Tissues	219 (Training 147; Validation:72)	I-IV	Hypermethylation	qMSP	Multivariate	Yi et al. 326
IGFBP3	R	Tissues	219 (Training 147; Validation:72)	I-IV	Hypermethylation	qMSP	Multivariate	Yi et al. 326
KiSS1	R	Tissues	342 (Training:62;Testing:100;Validation:190)	I-IV	Hypermethylation	MSP	-	Moya et al. 327
LINE-1	PR	Tissues	643	I-IV	Hypomethylation	Pyrosequencing	Multivariate	Ogino et al. 67
LINE-1	R	Tissues	343	I-IV	Hypomethylation	Pyrosequencing	Univariate	Antelo et al. 68
LINE-1	R	Tissues	207(MSI)	I-IV	Hypomethylation	Pyrosequencing	Multivariate	Rhee et al. 70
LINE-1	R	Tissues	161	III	Hypomethylation	Pyrosequencing	Multivariate	Ahn et al. 69
MGMT	R	Tissues	111	I-IV	Hypomethylation	Pyrosequencing	Univariate	Nilsson et al. 125
MGMT	R	Tissues	90	I-IV	Hypomethylation	MSP	Multivariate	Nagasaka et al. 126
IGF2	PR	Tissues	1033	I-IV	Hypomethylation	Pyrosequencing	Multivariate	Baba et al. 124
RET	PR, V	Tissues	Series 1:716, Series 2:393, Series3:294	I-IV/II+III/II	Hypermethylation	qMSP and pyroseq	Multivariate	Draht et al. 127
TFAP2E	R	Tissue	311	I-IV	Hypomethylation	MS-HRM	Multivariate	Zhang et al. 128
TFAP2E	R	Tissue	193	I-III	Hypomethylation	MS-HRM	Multivariate	Park et al. 129

CIMP marker	Study design	Specimens	Sample size	Stage	Marker genes	CIMP category	Significance	Survival analysis	Reference
CIMP	R + two arm study	Tissues	206	III	CDKN2A, MINT2, MDR1	CIMP −/+	CIMP(+) correlated with poor prognosis in surgery alone group	Univariate	van Rijnsoever et al.118
CIMP	-	Tissues	605	I-IV	MINT1,MINT2,MINT12, MINT31,CDKN2A	CIMP −/+	CIMP(+) correlated with poor prognosis in MSS CRC	-	Ward et al. 320
CIMP	PR	Tissues	816	I-IV	MINT 1, MINT 2, MINT 31, CDKN2A, and hMLH1	CIMP low/high	CIMP(+) correlated with poor prognosis in MSS CRC	Univariate	Samowitz et al. 328
CIMP	PR	Tissues	31	IV	CACNA1G, CDKN2A, CRABP1, IGF2,hMLH1, NEUROG1, RUNX3, SOCS1, MINT1, MINT31, IGFBP3, MGMT, WRN.	CIMP 0/low/high	CIMP High correlated with poor prognosis in metastatic CRC with MSS	Univariate	Ogino et al. 116
CIMP	PR	Tissues	188	IV	MINT1, MINT2, MINT31, hMLH1, p14, CDKN2A	CIMP −/+	CIMP(+) correlated with poor prognosis in CRC	Multivariate	Shen et al. 114
CIMP	R	Tissues	134	I-IV	hMLH1, CDKN2A, MINT1, MINT2, and MINT31	CIMP −/+	CIMP-High correlated with poor survival (OS) in MSS CRC	Univariate	Lee et al. 119
CIMP	R	Tissues	582	I-IV	hMLH1, CDKN2A, MINT1, MINT2, and MINT31	CIMP 0/low/high	CIMP-High correlated with poor survival (OS) in MSS CRC	Multivariate	Barault et al. 321
CIMP	PR	Tissues	649	I-IV	CACNA1G, CDKN2A (p16), CRABP1, IGF2, hMLH1, NEUROG1, RUNX3, SOCS1	CIMP 0/low/high	CIMP-High is independent factor of low colon-cancer mortality	Multivariate	Ogino et al. 36
CIMP	R	Tissues	320	I-IV	CACNA1G, CDKN2A (p16), CRABP1, IGF2, hMLH1, NEUROG1, RUNX3, and SOCS1.	CIMP −/+	CIMP(+) correlated with poor prognosis in CRC	Univariate	Kim et al. 329
CIMP	PR	Tissues	604	I-IV	hMLH1, CDKN2A, CACNA1G, NEUROG1, RUNX3, SOCS1, IGF2, CRRABP1	CIMP 0/low/high	CIMP-High correlated with poor survival (CSS) in MSS CRC	Multivariate	Dahlin et al. 330
CIMP	R	Tissues	161	III	MINT1,MINT2,MINT31,hMLH1,CDKN2A,p14,SFRP1,SFRP2,WNT5A	CIMP −/+	CIMP1 correlated with poor DFS in proximal CRC	Multivariate	Ahn et al. 69
CIMP	R	Tissues	337	-	CRABP1, hMLH1, CDKN2A, CACNA1G, NEUROG1	CIMP 0/low/high	CIMP-High correlated with poor prognosis in CRC	Univariate	Zlobec et al. 117
CIMP	PR	Tissues	150 (Rectal)	II/III	CACNA1G, IGF2, NEUROG1,RUNX3, SOCS1	CIMP −/+	CIMP(+) correlated with poor DFS in CRC	Univariate	Jo et al. 331
CIMP	R	Tissues	207 (MSI)	I-IV	CACNA1G, CDKN2A, CRABP1, IGF2,hMLH1, NEUROG1, RUNX3, and SOCS1	CIMP 0/low/high	CIMP-High correlated with poor prognosis in MSI CRC	Multivariate	Rhee et al. 70
CIMP	PR	Tissues	509	I-IV	CACNA1G, IGF2, NEUROG1, RUNX3, SOCS1	CIMP −/+	CIMP(+) correlated with poor prognosis in early follow up timinig	Multivariate	Simons et al. 332
CIMP	R	Tissues	734	I-IV	CACNA1G, CDKN2A, CRABP1, IGF2, hMLH1, NEUROG1, RUNX3 and SOCS1	CIMP 0/low/high	CIMP-High correlated with poor prognosis in rectal cancer	Multivariate	Bae et al. 333
CIMP	R	Tissues	50	II/III	p14, hMLH1, CDKN2A, MGMT, MINT1	CIMP −/+	CIMP(+) correlated with poor prognosis (DFS) in CRC	Univariate	Wang et al. 334
CIMP	PR+ two arm study	Tissues	615	III	CACNA1G, IGF2, NEUROG1, RUNX3, SOCS1	CIMP −/+	CIMP(+) correlated with poor prognosis in CRC	Univariate	Shiovitz et al. 132

Blood-based biomarker
CDKN2A	R	Serum	99	I-IV	Hypermethylation	qMSP	Univariate	Nakayama et al. 335
HTLF	R	Serum	77	I-IV	Hypermethylation	qMSP	Multivariate	Wallner et al. 111
HTLF	R	Serum	106	I-III	Hypermethylation	-	Multivariate	Herbst et al. 336
HTLF	R	Serum	311	I-IV	Hypermethylation	qMSP	Multivariate	Philipp et al. 337
HPP1	R V	Serum	142 (38,104)	I-IV	Hypermethylation	qMSP	Multivariate	Wallner et al. 111
HPP1	R	Serum	311	I-IV	Hypermethylation	qMSP	Multivariate	Philipp et al. 337

Predictive markers for response to treatment
Marker	Study design	Specimen	Sample size	Purpose	Treatment	Significance	Endpoint	Reference
Tissue-based biomarker
TFAP2E	PR V	Tissues	220 (CohortI:74, CohortII:36,CohortIII:42,CohortIV:68)	Adjuvant/Palliative	5-FU based chemotherapy	Resistance to treatment with TFAP2E hypermethylation	Response	Ebert et al. 134
LINE-1	R	Tissues	131 (MSS/CIMP(-))	Adjuvant		No benefit of 5-FU chemotherapy in patients with High LINE1 methylation in CRC	Survival benefit	Kawakami et al. 338
EGFR	R	Tissues	52 (EGFR-positive, K-RAS wild-type stage IV)	Palliative	EGFR	No benefit if hypermethylation of EGFR	Survival benefit	Scartozzi et al. 339
MGMT	PT	Tissues	65	Palliative	Dacarbazine	Hypermethylation correlates with good response	Response	Amatu et al. 340
MGMT	R	Tissues	90 (I-IV)	Adjuvant/Palliative	oral fluoropyrimidines	Unmethylated MGMT correlates with poor survival	Survival benefit	Nagasaka et al. 126
miR-148a	PR	Tissues	273(II-IV)	Adjuvant/Palliative	5-FU based chemotherapy	Hypermethylation indicates poor response and prognosis	Response and Survival benefit	Takahashi et al. 273
SRBC	R V	Tissues	189 (IV; Discovery:131, Validation:58)	Palliative	Oxali-based chemotherapy	Resistance to treatment with SRBC hypermethylation	Survival benefit	Mountiho et al. 135
PCDH10,SPARC,UCHL1 (panel)	PR + two arm study	Tissues	143 (stage II)	Adjuvant	5FU/LV or Surveillance	Hypomethylation of panel could identify the stage II patients with benefit of treatment	Survival benefit	Heitzer et al. 341

CIMP marker						CIMP Marker	Significance
CIMP	PR + two arm study	Tissues	206 (Stage III)	Adjuvant	5-FU/LV	CDKN2A, MINT2, MDR1	Treatment benefit in CIMP positive	Survival benefit	Van Rijnsoever et al. 118
CIMP	R	Tissues	245 (Stage II/III: 196)	Adjuvant	5-FU or Capecitabie	CACNA1G, IGF2, NEUROG1,RUNX3, SOCS1	Treatment benefit in CIMP positive	Survival benefit	Min et al. 130
CIMP	PR + two arm study	Tissues	302 (Stage II/III: 196)	Adjuvant	5-FU/LV or Surveillance	CACNA1G, hMLH1, NEUROG1, RUNX3, SOCS1	No benefit in CIMP positive	Survival benefit	Jover et al. 131
CIMP	PR + two arm study	Tissues	615 (Stage III)	Adjuvant	5FU/LV or IFL	CACNA1G, IGF2, NEUROG1, RUNX3, SOCS1	IrinotecanTreatment benefit in CIMP positive MSS CRC	Survival benefit	Shiovitz et al. 132

Among all biomarker candidates, CIMP status has undoubtedly been the most promising indicator for prognosticating CRC patients. CIMP-positive cancers correlate with an overall unfavorable prognosis^{70, 114–117}. Rijnsoever and colleagues showed in a cohort of 206 stage III CRC patients that CIMP-positive status associated with poor survival¹¹⁸. Another independent study analyzed more than 600 CRC patients and also found that CIMP associated with poor prognosis in MSS CRC patients¹¹⁹. Some studies suggested that poor prognosis in CIMP-positive CRCs is from coexisting V600E BRAF mutations^{36, 119–121}, however, in addition to CIMP, MSI status remains an important confounding factor that likely underlies the difference in prognosis of CIMP-positive MSS vs. MSI cancers^{122, 123}. These data highlight that the prognosis of patients with CIMP CRCs also depends on the MSI status of the tumor.

In addition to hypermethylation of various genes/loci, growing evidence suggests that DNA hypomethylation status associates with the prognosis of CRC patients. Ogino and colleagues have reported a correlation between LINE-1 hypomethylation and poor survival in prospective cohort studies of CRC patients⁶⁷. Subsequent studies not only validated this association for LINE-1 hypomethylation and CRC prognosis^68–70, but also identified other potential genes that correlate with adverse outcomes^124–129. In aggregate, these studies provide evidence that aberrantly methylated DNA has potential for use as prognostic biomarkers for CRC; however, further investigation is required to develop clinically robust, “locked-down” assays based upon standardized and reproducible prognostic biomarkers in order for these assays to be used in clinical care.

Methylated DNA as a Predictive Biomarker for Response to Treatment

Despite recent advances in the development of cancer therapeutics, the current generation of chemotherapeutic drugs are suboptimal because of modest efficacy and intrinsic or acquired resistance. This limitation highlights the need for predictive biomarkers that can be used to identify patients with low or high likelihoods of a response to specific treatments in order to maximize the overall treatment success with these chemotherapeutic drugs.

Over the last decade, a number of aberrantly methylated genes have been proposed to serve as predictive biomarkers for CRC patients undergoing various chemotherapeutic regimens (Table 2). The majority of these studies have not progressed beyond Phase I/II discovery phase and thus will not be discussed further in this article. CIMP as a predictive marker has been intensively studied for more than a decade. In 2003, van Rijnsoever and colleagues first reported that CIMP-positive CRC patients benefited from 5-flurouracil (FU) based adjuvant chemotherapy and that this association with survival was independent of MSI or TP53 mutation status¹¹⁸. These findings were validated in a subsequent study when stage II and III CRC patients with CIMP-positive tumors were also shown to derive survival benefit following 5-FU treatment¹³⁰. However, these results were challenged in a later study that examined a large patient cohort and reported that those patients with CIMP-positive tumors that did not receive 5-FU treatment survived longer when compared to patients with CIMP-negative CRCs¹³¹. Another study showed CIMP-positive CRC patients had a shorter DFS after 5-FU treatment compared with patients with CIMP-negative tumors¹³¹. Recently, in another prospective study, the addition of irinotecan to adjuvant 5FU/LV in stage III, MSS, and CIMP-positive colon cancers improved overall survival¹³². Although the differences between these studies may reflect differences in CIMP criteria or inherent differences between patient cohorts; these studies suggest promise for the use of CIMP as a predictive marker and also highlight the need for additional studies of the interaction between CIMP status and therapeutic response to various treatments.

Although still in the early phase of development, some promising pharmaco-epigenetic biomarkers, such methylated MGMT for temozolamide treatment in gliomas, have been identified in various cancers¹³³. A recent study showed the feasibility of using hypermethylated Transcription Factor AP-2 Epsilon (TFAP2E) as a predictive biomarker for response to 5-FU based chemotherapy in CRC patients¹³⁴. Furthermore, DNA methylation microarray profiling of oxaliplatin sensitive vs. resistant CRC cell lines revealed that oxaliplatin-resistant cells exhibited hypermethylation of the BRCA1 interactor SRBC gene; which was subsequently shown to associate with poor progression free survival (PFS) in CRC cohorts treated with oxaliplatin¹³⁵. The ability to develop reliable assays for methylated genes and the results of early phase studies using methylated genes as predictive markers will continue to drive the investigation of methylated genes as response predictors for CRC therapy.

Methylated DNA as a Colorectal Cancer Risk Biomarker: “Field Cancerization” and ‘Epigenetic Drift’

The concept of “field cancerization” (or field effect) was first proposed in 1953 by Slaughter et al¹³⁶. Field cancerization is characterized by the occurrence of genetic and epigenetic alterations in histologically normal-appearing tissues, and is believed to lead to an increased risk for synchronous or metachronous primary tumors. While genetic alterations are common in CRC cells, these are believed to be rare in normal cells. In contrast, some studies suggest somatic epigenetic dysregulation occurs not only in cancer tissues, but also in non-cancerous and pre-neoplastic tissues. Considering that epigenetic alterations could contribute to the early events predisposing to malignant transformation, these studies suggest epigenetic events are potentially more promising somatic CRC risk markers than are gene mutations. Methylation changes in tumor-suppressor genes and the estrogen receptor (ER) occurs in an age- and region-specific manner in normal colonic mucosa. The proposed age-related epigenetic alterations in tumor-suppressor genes may be one of the earliest events that predispose normal mucosa to tumorigenic transformation in CRC¹³⁷. Loss of the insulin-like growth factor-II (IGF2) gene imprinting occurs at a higher frequency in the normal mucosa adjacent to cancer tissue, compared to normal mucosa in patients without CRC¹³⁸, emphasizing the potential of IGF2 imprinting as a molecular signature to identify patients at greater risk for CRC development. In line with this concept, other studies have revealed that both hypermethylation of tumor-suppressive genes such as SFRP, ESR1, MYOD, and MGMT, as well as LINE-1 hypomethylation in normal colonic mucosa correlates with an increased risk of CRC, in contrast to patients without these traits^139–142.

Accumulating evidence supports that the landscape of DNA methylation can be modified as a “function of age”. DNA methylation has been proposed to result from a gradual stochastic age-dependent dysregulation caused by a combination of external environmental factors and internal spontaneous random errors in the maintenance of methylation. This process of age-dependent alterations in methylation is defined as “epigenetic drift”¹⁴³. Interestingly, such age-associated DNA methylation often targets the promoters of tumor-suppressive genes^{144, 145}. In monozygotic twins, epigenetic divergence with age suggests the underlying epigenetic drift may in part help explain the disease discordance^146–148. Since this new concept closely relates to the “field effect”, identification of biomarkers that overlap both the “epigenetic drift” and “field effect” in colorectal mucosa may allow development of next-generation biomarkers for determining risk for CRC development in the future.

HISTONE MODIFICATIONS: A POTENTIAL CLASS OF CRC BIOMARKERS

DNA in eukaryotic cells is found in chromatin, which is a complex of macromolecules consisting of DNA, RNA and protein. The primary functions of chromatin are to facilitate DNA compaction, to reinforce the DNA macromolecule during mitosis, to protect against DNA damage, and to control gene expression and DNA replication. The primary protein components of chromatin are histones, which regulate DNA compaction and gene expression. Histones are protein octamers that consist of two copies of four core proteins - H2A, H2B, H3, and H4. Each histone octamer contains approximately 147 bp of DNA to form the nucleosome¹⁴⁹. As illustrated in Figure-2B, each core histone protein has specific histone tails, which are subject to modifications including acetylation, methylation, ubiquitination, phosphorylation, and sumoylation¹⁵⁰. Various modifications alter the three-dimensional structure of the nucleosome and affect the transcriptional control of associated genes by creating an “inactive”, compacted, heterochromatin state, or an “active”, open chromatin, euchromatin conformation. For example, the active transcriptional state is recognized by di- and tri-methylation of histone H3 lysine 4 (H3K4me2, and H3K4me3), histone H3 lysine 36 (H3K36me2, and H3K36me3), and acetylation of H3 and H4, while trimethylation of H3 lysine 9 and 27 (H3K9me3 and H3K27me3) are enriched in heterochromatin and associate with inactive transcription¹⁵¹. After the initial discovery of the dysregulation of histone modifications in CRC¹⁵², studies have revealed that this dysregulation likely alters gene expression patterns in CRC. However, due to the technical limitations of assays that assess the post-translational histone modification state, it has been difficult to determine the histone modification state in primary cancer tissues. Therefore, efforts to determine if histone modifications can be used as disease biomarkers have been limited (Supplementary Table 1). Global alterations of specific-histones in primary tissues have been the focus for biomarker development in CRC. Studies of H3K4me2, H3K9ac, and H3K9me2 alterations detected by immunohistochemical staining in liver metastases suggest that high H3K4me2 expression levels inversely correlate with poor prognosis¹⁵³. Additionally, other studies in CRC suggest that histone modifications, such as acetylation of H3 lysine 56 and di- or tri-methylation of H3 lysine 9 and 27, have potential to be prognostic markers in CRC.^151–156. Similarly, studies of histone modifications in circulating nucleosomes have identified reduced levels of H3K9me3- and H4K20me3 as potential diagnostic biomarkers for CRC^{157, 158}. However, thus far, these studies are all Phase I biomarker studies and should be considered “proof of principle”. Further research is needed to determine whether any of these modifications will be clinically useful diagnostic or prognostic biomarkers in CRC.

NONCODING RNAS

Noncoding RNA overview

The central dogma of molecular biology, which describes the sequential transfer of genetic information and the concept that “DNA makes RNA and RNA makes protein”, was developed in 1956 and provided a fundamental framework for modern molecular biology until recently¹⁵⁹. Advances in our understanding of the regulation of gene expression has been provided by the Encyclopedia of DNA Elements Consortium (ENCODE) transcriptome project, which recently revealed that protein-coding genes represent less than 2% of total genome and approximately 80% of the genome is actively transcribed into non-coding RNA (ncRNA)¹⁶⁰. Although ncRNAs were previously believed to be “background transcriptional noise” related to “junk DNA”, mounting evidence indicates that ncRNAs play a significant role in many biological processes, including the regulation of oncogenes and tumor suppressor genes in cancer¹⁶¹. NcRNAs are broadly categorized into two groups based upon their size: small ncRNAs, that are shorter than <200 nucleotides, comprising of miRNAs (miRNAs), piwi-interacting RNAs (piRNAs), and small nucleolar RNAs (snoRNAs); and long ncRNAs (lncRNA) that are longer than 200 nucleotides¹⁶². Recent studies have elucidated the functional role of ncRNAs, particularly miRNAs, in human cancers and have not only revolutionized our understanding of their biological contribution to cancer pathogenesis, but have also provided important insights into the feasibility of their use as clinically relevant biomarkers in cancer.

Overview of miRNAs

Over the last decade, in the family of ncRNAs, the role of miRNAs has been best established in the context of carcinogenesis. MiRNAs are endogenous single-stranded small RNAs that are 18–25 nucleotides in length that were first discovered in 1993 as negative post-transcriptional regulators in Caenorhabditis elegans^{163, 164}. During their biogenesis, premature-miRNAs are exported from the nucleus to the cytoplasm. Subsequent processing of the pre-miRNA generates mature-miRNA, which binds to 3’UTR “seed sequence” of target mRNAs, a process that is catalyzed by the RNA-induced silencing complex (RISC). The binding of miRNA to the target mRNA can result in degradation of the target mRNA or inhibition of its translation into protein, with the degree of sequence complementarity between the miRNA and mRNA determining which mechanism is employed^165–167 (Figure 2C). The role of miRNAs in cancer was first shown in 2002 by Croce and colleagues, who demonstrated decreased expression of miR-15 and miR-16 in patients with chronic lymphocytic leukemia¹⁶⁸. Since then, hundreds of miRNAs have been shown to be deregulated in other human malignancies, some of these deregulated miRNAs appear to have a functional role in tumorigenesis by regulating the expression of important oncogenes and tumor-suppressor genes¹⁶⁹.

Dysregulation of miRNA Expression in CRC: Role in “Traditional” and “Serrated” Pathways

Commonly deregulated miRNAs have been identified that appear to participate during each step of the “traditional” (normal-adenomatous polyp-cancer) and “serrated” (normal-serrated polyp-cancer) pathways of CRC development. For instance, the miR-17-92a cluster, miR-135b, miR-143, and miR-145 regulate WNT/β-Catenin signaling pathway, which is involved in colorectal neoplasia initiation ^170–173. Genes associated with the RAS-MAPK and PI3K/AKT cascades, that appear to drive the transition from early to advanced adenoma, are found to be regulated by specific miRNAs (e.g. RAS-MAPK by miR-143, let-7, miR-21, and miR-31; and PI3K/AKT by miR-1, miR-21 and miR-143) ^174–180. Likewise, p53, a tumor suppressor protein frequently inactivated/silenced during the evolution from advanced adenoma to adenocarcinoma, is controlled by miR-34a/b/c, miR-133a, miR-143, and miR-145^{174, 181–183}. In contrast, genes such as LIN28 drive CRC progression through inhibition of let-7 miRNA biogenesis, highlighting the complexity of miRNA-gene interaction in cancer^{184, 185}. In addition, miR-21, miR-155 and miR-200 family members regulate the TGF-ß pathway ^186–188.

miRNAs also appear to play a role in the serrated pathway of carcinogenesis. The identification of increased expression of miR-21 and miR-181 in hyperplastic polyps (HP) and sessile serrated adenomas (SSAs), yielded the first clues for a potential role of miRNAs in the serrated pathway¹⁸⁹. A recent, comprehensive miRNA expression profiling study of CRCs with or without BRAF mutations identified elevated miR-31-5p expression in BRAF mutant cancers¹⁹⁰. A significant correlation between miR-31 overexpression and specific types of serrated polyps, sessile serrated adenomas (SSAs) vs. traditional serrated adenomas (TSAs) ¹⁹¹ highlights the possible functional role for this miRNA in the serrated pathway¹⁹⁰. These findings were corroborated in a follow-up study that analyzed 381 serrated and 222 non-serrated adenomas and identified a CIMP-independent association between miR-31 overexpression and BRAF mutations¹⁹². Interestingly, expression of miR-31 was progressively increased in the histologically more advanced lesions arising from SSAs but not from TSAs. Furthermore, miR-31 expression, BRAF mutation, CIMP-positive status, and MLH1 methylation showed a gradual increase from rectum to cecum in SSA lesions ^{193, 194}.

MiRNAs as Clinically Useful Biomarkers for Colorectal Cancer

Over the last five years, miRNA biomarker research in human cancer has increased exponentially. The underlying reasons for this burgeoning interest are based upon some of their unique characteristics. First, miRNAs are remarkably stable under a variety of experimental and laboratory conditions. Second, due to their small size and the hairpin-loop structure, miRNAs are protected from RNase-mediated degradation¹⁹⁵, and thus are easily extractable from a wide variety of clinical specimens, including formalin-fixed paraffin embedded (FFPE) tissues, and a variety of body fluids including blood, saliva, urine, feces etc. Third, cell-free miRNAs are often protected from degradation because of being in high density lipoprotein particles, apoptotic bodies, microvesicles, and exosomes, and through their binding to argonaute-2 (Ago-2), which results in increased stability^196–199. Fourth, in addition to the stability in various human specimens, miRNAs are actively secreted by cancer cells into the circulatory system and digestive tract^{200, 201}. Taken together, the stability of miRNAs coupled with their presence in a variety of compartments in the body (blood, feces, cancer cells, cells near the cancer, etc.), has led to an intense line of research on the use of miRNAs as cancer biomarkers.

MiRNAs as Diagnostic Biomarkers in Colorectal Cancer

Detection of pre-malignant polyps and early-stage neoplasms currently is one of the major goals of CRC screening strategies. The identification and treatment of polyps and cancers in their earliest stages leads to the most successful outcomes. In the following sections and Table 3, we summarize some of the key findings with regards to circulating cell free-miRNAs and fecal-miRNAs and their potential as diagnostic biomarkers in CRC.

Table 3.

List of miRNAs as potential noninvasive diagnostic markers in colorectal cancer

miRNA	Source	Study Design	Sample size		Endpoint: Adenoma or CRC	Sensitivity (%)	Specificity (%)	AUC (95% CI)	Normalizer	Reference
			Cases	Control
Blood based biomarker
miR-17-3p	plasma	R V	5/25/90	5/20/50	CRC	64	70	0.72	RNU6B	Ng et al. 203
miR-18	plasma	R	78	86	CRC	73.1	79.1	0.8	RNU6B	Zhang et al. 211
miR-21	plasma	R V	30/20	30/20	CRC	90	90	0.82/0.91	U6	Kanaan et al. 207
miR-21	plasma	R	29	29	CRC	-	-	0.65	miR-16	Zanutto et al. 342
miR-21	plasma	R	49	49	CRC	76.2	93.2	0.88	cel-miR-39	Du et al. 343
miR-21	Serum	R V	12/186(CRC), 43(AD)	12/53	CRC, AD	91.9(CRC) 81.1(AD)	81.1(CRC) 76.7(AD)	0.92(CRC) 0.81(AD)	cel-miR-39	Toiyama et al. 200
miR-21	Serum	R	200(CRC), 50(AD)	80	CRC, AD	-	-	0.8(CRC) 0.71(AD)	miR-16	Liu et al. 208
miR-21	Serum (Exosome)	R	88	11	CRC			0.8	miR-451	Ogata-Kawata et al. 201
miR-21	Serum	R	40	40	CRC	77	78	0.87	RNU6B	Basati et al. 344
miR-29a	plasma	R V	20/100(CRC),37(AD)	20/59	CRC, AD	69(CRC) 62.2(AD)	89.1(CRC) 84.7(AD)	0.84(CRC) 0.77(AD)	miR-16	Huang et al. 204
miR-92a	plasma	R V	5/25/90	5/20/50	CRC	89	70	0.89	RNU6B	Ng et al. 203
miR-92a	plasma	R V	20/100(CRC),37(AD)	20/59	CRC, AD	84(CRC) 64.9(AD)	71.2(CRC) 81.4(AD)	0.84(CRC) 0.75(AD)	miR-16	Huang et al. 204
miR-92a	Serum	R	200(CRC), 50(AD)	80	CRC, AD	-	-	0.77(CRC) 0.7(AD)	miR-16	Liu et al. 208
miR-155	Serum	-	146	60	CRC	-	-	0.78	-	Lv et al. 209
miR-200c	plasma	R	78	86	CRC	64.1	73.3	0.75	RNU6B	Zhang et al. 211
miR-221	plasma	R	103	37	CRC	86	41	0.61	-	Pu et al. 210
miR-18a,-20a,-21,-29a,-92a,- 106b, -133a,-143,-145	plasma	R V	50/80(CRC), 50(AD)	50/194	CRC	-	-	0.75	miR-16	Luo et al. 212
miR-15b,-17,-142-3p, -195, - 331, -532,-532-3p, -652	plasma	R V	20/45(CRC), 9/16(AD)	12/26	AD	88	64	0.87(AD)	U6	Kanaan et al. 207
miR-21, 31, 92a, 181b, 203, let-7g (panel)	Serum	R V	30/83	30/59	CRC	96.4	88.1	0.92	miR-16	Wang et al. 214
miR-7, -93, -409-3p (panel)	plasma	R V	47/55/22	33/57/27	CRC	91/82	88/89	0.87/0.9	miR-1228	Wang et al. 215
Fecal-based biomarker
miR-18	Stool	R	198	198	CRC	61	69	0.67	-	Yau et al. 345
miR-21	Stool	R	19	10	CRN (CRC+AD)	-	-	-	miR-16/ miR-26b	Link et al. 216
miR-21	Stool	R	88	101	CRC	55.7	73.3	0.64	-	Wu et al. 217
miR-17-92 cluster	Fecal colonocyte	R	197	119	CRC	69.5	81.5	-	U6	Koga et al. 219
miR-92a	Stool	R	88	101	CRC	71.6	73.3	0.78	-	Wu et al. 217
miR-106a	Stool	R	19	10	CRN (CRC+AD)	-	-	-	miR-16/ miR-26b	Link et al. 216
miR-106a	Stool	R	107	117	CRC	34.2	97.2	-	miR-24	Koga et al. 218
miR-135b	Fecal colonocyte	R	197	119	CRC	45.7	95	-	U6	Koga et al. 219
miR-135b	Stool	R	104(CRC), 169(AD)	109	CRC	78(CRC) 65(AD)	68	0.79(CRC) 0.71(AD)	-	Wu et al. 346
miR221	Stool	R	198	198	CRC	62	74	0.73	-	Yau et al. 345

R, Retrospective study; PR, Retrospective analysis of prospectively collected samples in clinical trial; PT, Prospective trial; CRC, Colorectal cancer; CRN, colorectal neoplasia; AD, adenoma

Blood-based biomarkers

The discovery of miRNAs in extracellular body fluids²⁰² triggered a growing number of studies that evaluated dysregulated expression of circulating miRNAs in various cancers. The first systematic and comprehensive miRNA expression profiling study was conducted by Ng and colleagues²⁰³, who evaluated miRNA expression alterations in tissue and plasma samples from CRC patients and healthy subjects. This study revealed that high expression of two miRNAs, miR-92a and miR-17-3p, could discriminate patients with CRC from healthy study subjects (sensitivity: 64%, 89%, specificity: 70%, 70%, AUC: 0.72, 0.89, for each miRNA respectively). This landmark study further reported that the plasma levels of both miRNAs decreased significantly following surgical resection of the primary tumors, and that plasma miR-92a levels were also elevated in patients with gastric cancer and inflammatory bowel disease. Their findings have been replicated in other studies of CRC patients²⁰⁴.

MiR-21 is another well-characterized oncogenic miRNA, which is considered as one of the promising non-invasive biomarkers for the early detection of CRC owing to the following attributes: a) dysregulation of miR-21 occurs frequently in early stages of the adenoma-carcinoma sequence²⁰⁵ b) miR-21 is one of the most highly expressed miRNAs in CRC²⁰⁶, and c) miR-21 is highly secreted by cancer cells and can be measured in exosomes or as free miRNAs in plasma or serum^{200, 201}. One of the first studies utilizing miRNA expression profiling on primary CRC tissues and the adjacent normal mucosa identified miR-21 as differentially expressed in CRC (n=30)²⁰⁷. Validation studies in an independent set of plasma samples (“test” set; 20 CRCs and 20 healthy controls) demonstrated that plasma miR-21 could be used to discriminate patients with CRC from normal control patients with high sensitivity (90%) and specificity (90%). Furthermore, the diagnostic and prognostic potential of serum miR-21 in CRC patients was recently addressed in another study²⁰⁰, which revealed that tumor-derived circulating miR-21 expression not only accurately discriminated patients with CRC from healthy subjects, but allowed identification of patients with advanced adenomas, which are “bonafide target lesions” for an ideal CRC screening test (sensitivity: 91.1%, 81.1%, specificity: 81.1%, 76.7%, AUC: 0.92, 0.81, for cancer and adenoma detection, respectively). This study also confirmed the significant association between lower miR-21 expression in serum and CRC tissues following curative resection of the primary tumor. Consequently, several studies confirmed the potential of miR-21 for use as a single miRNA biomarker for the early detection of CRC^{201, 208}.

Although increasing number of miRNAs have been identified as potential biomarkers for the early diagnosis of CRC^{204, 209–211}, it seems unrealistic that a single miRNA will adequately capture the underlying disease heterogeneity in colorectal polyps and cancers. Accordingly, several studies have proposed combining miRNAs into a biomarker panel to improve the detection accuracy of colorectal neoplasms^212–215.

Although miRNAs appear to be promising CRC biomarkers, there are several challenges that must be borne in mind while considering their potential as diagnostic CRC biomarkers. The lack of consistency between biomarker panels in independent studies highlights a major obstacle for the development of robust miRNA biomarkers. This variability in miRNA biomarkers/biomarker panels among studies may in part be due to lack of standardized sample handling and processing steps, use of inconsistent normalization approaches as well as to the differences in ethnic and racial makeup of patient populations in the various studies. Once these issues are resolved, validation studies utilizing standardized assays in large population-based cross-sectional and prospective cohort studies are needed to identify optimal miRNAs and marker panels that can be used in clinical care.

Stool-based biomarkers

Similar to blood-based biomarkers, the use of stool-based miRNA biomarker assays for the early detection of colorectal neoplasia has been assessed recently. In 2010, one of the first studies of stool-based miRNAs demonstrated the feasibility of a one-step miRNA extraction and amplification method defined as “direct miRNA analysis” (DMA), and showed dysregulated expression of miR-21 and miR-106a as potential candidate biomarkers for CRC screening²¹⁶. A subsequent study that assessed fecal miR-21 and miR-92a levels found that fecal miR-92a expression could differentiate patients with CRC or adenoma from those with lower-risk polyps or healthy subjects²¹⁷. A more recent study evaluated fecal miRNA expression in residual material collected from FOBT kits and determined that miR-106a expression enhanced the sensitivity of FOBT in identifying patients with CRC²¹⁸.

Another potential stool-based miRNA biomarker assay employed the extraction of RNA from fecal colonocytes using immuno-magnetic beads conjugated with the epithelial cell adhesion molecule (EpCAM) monoclonal antibody. This assay was used to study a cohort of 197 CRC patients and 119 healthy volunteers²¹⁹ and the expression of the miR-17-92a cluster and miR-135b in feces was found to discriminate patients with CRC from healthy subjects with a high sensitivity (69.5%, 45.7%) and specificity (81.5%, 95%), further highlighting the potential use of stool as a source for miRNA biomarkers in CRC screening tests.

MiRNAs as Prognostic Biomarkers in Colorectal Cancer

Recognition of the probable functional role miRNAs play in human cancers and their remarkable stability in a variety of clinical specimens has made them attractive candidates as prognostic biomarkers in CRC²²⁰. The first study in this regard was conducted by Schetter and colleagues in 2008²⁰⁵, who used a microarray-based approach to evaluate the expression levels of 389 miRNAs in 84 CRC and matched normal colonic tissues. This study identified and validated 37 differentially expressed miRNAs, including miR-20a, miR-21, miR-106a, miR-181b and miR-203. A seminal finding of this study was that high expression of miR-21 in CRC patients associated with poor survival, which has since been independently confirmed in several other reports^{200, 221–225}. As shown in Table 4, although several other overexpressed (miR-10b, miR-17–92a cluster, miR-29a, miR-31, and miR-182)^{190, 226–236} and under-expressed miRNAs (miR-143 and miR-124)^237–239 have been proposed to be prognostic biomarkers, currently miR-21 has the best potential to be a clinically useful miRNA-based prognostic biomarker in CRC.

Table 4.

List of miRNAs as potential prognostic and predictive markers in colorectal cancer

Prognostic markers
miRNA		Source	Dysregulation in CRC	Notable confirmed target genes*
Tissue-based b iomarker
Mature miRNA ID	Previous miRNA ID
miR-10b-5p	miR-10b	Tissue	Up regulated	TIP30, SDC1, NF1, KLF4, HOXD10, PPARA, CDKN2A, CDKN1A, TFAP2C, BCL2L11
miR-17-5p	miR-17	Tissue	Up regulated	PTEN, DLC1, ZBP1, p130, CDKN1A, PTPRO, PKD2, BCL2L11, E2F1, MAPK9, TGFBR2, VIM, CCND1
miR-21-5p	miR-21	Tissue	Up regulated	PTEN, PDCD, RECK, JAG1, BCL2, TIMP3 DKK2, SOX5, MTAP, TGFBR2, E2F1, TPM1, APAF1
miR-29a-3p	miR-29a	Tissue	Up regulated	KLF4, DNMT3a, DNMT3b, CDK6, BACE1, MCL1, BCL2, PPM1D, LPL1
miR-31-5p	miR-31	Tissue	Up regulated	FIH1, RhoBTB1, RASA1, FOXP3, ITGA5, MMP16, RDX, FZD3, DKK1, CREG1, RASA1
miR-92a -3p	miR-92a	Tissue	Up regulated	PTEN, P63, RECK, DUSP10, ITGA5, BMPR2, KAT2B
miR-155-5p	miR-155	Tissue	Up regulated	MLH1, MSH2, MSH6, APC, TP53INP1, TAB2, MEIS1, MECP2, SOCS1, INPP5D, SMAD5, HIVEP2, BACH1
miR-181a-5p	miR-181a	Tissue	Up regulated	PTEN, E2F5, SMAD7, ATM, BCL2, PROX1, KAT2B, CDKN1B, RNF2, RALA, KLF6
miR-182-5p	miR-182	Tissue	Up regulated	NDRG1, RECK, MTSS1, SMAD4, CDNK1A, FOXO3, FOXO1, MITF, CYLD, BCL2, CCND2
miR-191-5p	miR-191	Tissue	Up regulated	TIMP3, BASP1, CDK6, SATB1, NDST1
miR-221-3p	miR-221	Tissue	Up regulated	SUN2, RECK, CDKN1B, BMF, FOXO3, KIT, CDKN1C, TMED7, DDIT4, BNIP3L, TBK1
miR-224-5p	miR-224	Tissue	Up regulated	SMAD4, CXR4, PHLPP1, PHLPP2, KLK10, CDC42, AP15, EYA4, EDNRA, DIO1
let-7i-5p	let-7i	Tissue	Down regulated	TLR4, SOCS1, BMP4, IL13
miR-16 -5p	miR-16	Tissue	Down regulated	BMI1, WIP1, CDK6, HMGA1, ACVR2A
miR-34a-5p	miR-34a	Tissue	Down regulated	CD44, MYCN, PD-L1, MET, JAG1, MYC, MET, CDK4, CCNE2, MYCN, CCND1, E2F3, SIRT1, BCL2, NOTCH1
miR-124-3p	miR-124	Tissue	Down regulated	SLUG, STAT3, EZH2, IL6R, EFNB1, CEBPA
miR-133a/b	miR-133a/b	Tissue	Down regulated	LASP1, RFFL, FSCN1, CXCR4, MET, KCNQ1, PNP, KCNH2,
miR-137	miR-137	Tissue	Down regulated	LSD1, FMNL2, CDC42, CDK6, KDM1A, NCOA2, CTBP1, MITF
miR-143-3p	miR-143	Tissue	Down regulated	KRAS, DNMT3A, TLR2, MACC1, MYO6, MAPK7, FNDC3B, FSCN1, SERPINE1, JAG1, AKT1
miR-148a-3p	miR-148a	Tissue	Down regulated	BCL2, MMP7, ROCK1, DNMT1, TG1F2, DNMT3, RPS6KA5
miR-149-5p	miR-149	Tissue	Down regulated	FOXM1, ZBTB2
miR-200a-3p	miR-200a	Tissue	Down regulated	CTNNB1, ZEB2, SIRT1, ZEB1, ZFPM2, GDAP1, MAPK14, KEAP1
miR-340-5p	miR-340	Tissue	Down regulated	MET, PLAT
miR-378a-3p	miR-378	Tissue	Down regulated	IGF1R, MYC, VEGFA, NPNT, GALNT7
Blood-based biomarker
miR-21-5p	miR-21	Serum	Up regulated	PTEN, PDCD, RECK, JAG1, BCL2, TIMP3 DKK2, SOX5, MTAP, TGFBR2, E2F1, TPM1, APAF1
miR-29c-3p	miR-29c	Serum	Up regulated	KLF4, DNMT3a, DNMT3b, CDK6, BACE1, MCL1, BCL2, PPM1D, LPL1
miR-92a-3p	miR-92a	Serum	Up regulated	PTEN, P63, RECK, DUSP10, ITGA5, BMPR2, KAT2B
miR-141-3p	miR-141	Plasma	Up regulated	ZEB2, ZEB1, SFPQ, CLOCK BRD3, UBAP1, PTEN, ZFPM2, E1F4E
miR-155-5p	miR-155	Serum	Up regulated	MLH1, MSH2, MSH6, APC, TP53INP1, TAB2, MEIS1, MECP2, SOCS1, INPP5D, SMAD5, HIVEP2, BACH1
miR-200c-3p	miR-200c	Serum	Up regulated	TUBB2, BMI1, ZEB2, ZEB1, FN1, ZFPM2, PTPN13, RNF2, BRD7
miR-221-3p	miR-221	Plasma	Up regulated	SUN2. RECK, CDKN1B, BMF, FOXO3, KIT, CDKN1C, TMED7, DDIT4, BNIP3L, TBK1
miR-345-5p	miR-345	Blood	Up regulated	CDKN1A, ABCC1
miR-885-5p	-	Serum	Up regulated	CDK2, MCM5

Predictive markers for response to treatment
miRNA		Specimen	Purpose of therapy	Endpoint	Treatment
Tissue-based biomarker
Mature miRNA ID	Previous miRNA ID
miR-21-5p	miR-21	Tissue	NeoAdjuvant/Adjuvant/Palliative	Prognosis	5FU-based chemotherapy
miR-148a-3p	miR-148a	Tissue	Adjuvant/Palliative	Prognosis	5FU/5FU+oxaliplatin
miR-150-5p	miR-150	Tissue	Adjuvant	Prognosis	5FU with/without LV/levamisole/CDDP
-	miR-200 family	Tissue	Adjuvant	Prognosis	Fluoropyrimidines
-	miR-215	Tissue	Adjuvant	Prognosis	5FU-based chemotherapy
let-7c, g	let-7c, g	Tissue	Palliative	Response, Prognosis	S-1 with or without CDDP, anti-EGFR
miR-31-3p	miR-31*	Tissue	Palliative	Prognosis	anti-EGFR
miR-99a-5p,-3p	miR-99a, -99a*	Tissue	Palliative	Prognosis	5FU-based chemotherapy, anti-EGFR
miR-107	miR-107	Tissue	Palliative	Prognosis	5FU-based chemotherapy
miR-125b-5p	miR-125b	Tissue	Palliative	Prognosis	anti-EGFR
miR-126-3p	miR-126	Tissue	Palliative	Response, Prognosis	XELOX, bevacizumab (anti-VEGF-A)
miR-181a,b	miR-181a,b	Tissue	Palliative	Response, Prognosis	S-1 with or without CDDP, anti-EGFR
miR-625-3p	miR-625*	Tissue	Palliative	Response, Prognosis	XELOX/FOLFOX
Blood-based predictive biomarker
miR-27b, -106a, -130b, -148, -326, -484		Plasma	Adjuvant	Response/Prognosis	5FU+Oxaliplatin
miR-19a-3p	miR-19a	Serum	Palliative	Response	FOLFOX

^*miRTarBase Ver. 4.5 (http://mirtarbase.mbc.nctu.edu.tw) was used to list confirmed miRNA target genes

Based upon current guidelines, the majority of stage II CRC patients are treated surgically without adjuvant chemotherapy. However, a significant proportion (approximately 15%) of these patients experience tumor recurrence and death due to disease progression^{240, 241}, highlighting the need for biomarkers that can identify high-risk stage II CRC patients who could benefit from adjuvant chemotherapy. Schepeler and colleagues found that miR-320 and miR-498 miRNA- distinguish high-risk from low-risk stage II patients²⁴² and correlate with recurrence free survival. Similarly, miR-21 expression appears to identify high-risk population in stage II CRC patients, which has been confirmed in several independent investigations^{222, 243–245}.

In addition to individual markers, a miRNA panel has been developed to identify stage II CRC patients with a high risk of recurrence. The expression of 1849 miRNA probes in 40 paired stage II colon tumors and adjacent normal mucosa tissues was examined²²⁵. A miRNA-based classifier comprising of miR-20a-5p, miR-21-5p, miR-103a-3p, miR-106a-5p, miR-143-5p and miR-215 was developed, which discriminated high risk of stage II CRC patients in a “testing cohort” (138 patients) and an independent “validation” cohort of patients (460 patients)²²⁵.

Although several miRNAs have thus far been identified as potential prognostic biomarkers in CRC, this field is still rapidly evolving. In addition to high-risk stage II patients, identification of low-risk stage III CRC patients is another potential use of miRNA risk stratification/prognostic biomarkers..

Predictive Biomarkers for Response to Treatment in Colorectal Cancer

The treatment options for advanced CRC patients have improved considerably over the last decade as a result of the development of novel targeted therapies²⁴⁶. 5-FU based regimens remain the mainstay for adjuvant therapy, while advanced metastatic disease is often treated with newer antibody-based drugs targeting the vascular endothelial growth factor (VEGFA) and epidermal growth factor receptor (EGFR)^247–249. In spite of therapeutic advances, the prognosis of patients with unresectable CRC still remains poor, with the median overall survival of only 18–21 months²⁵⁰. The development of biomarkers that can accurately predict a patient’s response to a specific chemotherapeutic regimen prior to initiation of the chemotherapy is of clear value and an area of active investigation.

Currently, the majority of the results related to the role of miRNAs in drug resistance are based on in vitro studies and remain to be assessed in clinical sample sets (Table 4). MiRNAs shown to mediate 5-fluorouracil resistance include miR-10b²²⁶, miR-19b²⁵¹, miR-20a²⁵², miR-21^253–255, miR-23a²⁵⁶, miR-31²⁵⁷, miR-34^{258, 259}, miR-129²⁶⁰, miR-140²⁶¹, miR-145²⁶², miR-192/-215²⁶³, miR-200 family²⁶⁴, and miR-497²⁶⁵. MiRNAs have also been identified that mediate irinotecan-resistance (miR-21²⁵⁴ and miR-451²⁶⁶) and oxaliplatin resistance, (miR-20a²⁵², miR-21²⁵⁴, miR-133a¹⁸³, miR-143²⁶⁷, miR-153²⁶⁸, miR-203²⁶⁹ and miR-1915²⁷⁰). However, as noted above, most of these findings are primarily based upon in vitro studies, and clinical data supporting these miRNA biomarkers as drug resistance markers are limited or nonexistent. Let-7g and miR-181b expression in primary tissues has been shown to associate with S-1-based response to chemotherapy²⁷¹, while overexpression of miR-21 in CRC tissues has been associated with a poor response to 5-FU-based chemotherapy^{205, 243, 272}. Furthermore, both expression and methylation levels of miR-148a are linked with lack of response to 5-FU and oxaliplatin chemotherapies in advanced CRC patients²⁷³. A recent comprehensive array-based interrogation of tumor tissues collected from KRAS wild-type metastatic CRC (mCRC) patients treated with anti-EGFR therapy identified miR-31-3p as a negative predictor of progression free survival (PFS)²⁷⁴.

Importantly, the studies of predictive miRNA biomarkers have been done using retrospective study designs and archived tissue samples. This is a limitation in our understanding of miRNA biomarkers as these studies have not accounted for intratumoral heterogeneity in the primary tissues and carry the biases inherent in retrospective study designs. Nonetheless, based on the in-vitro studies and on their presence in blood, circulating miRNA biomarkers in serum or plasma have potential to be used in blood based assays for predicting or monitoring response to chemotherapy. Although in its infancy, the development of promising miRNA biomarkers for predicting response to chemotherapeutic treatment in CRC patients could usher in a new era in the era of personalized management of cancer patients.

Long Non-coding RNA Biomarkers in Colorectal Cancer

Recent improvements in our understanding of the role of ncRNAs in carcinogenesis have led to their assessment as CRC biomarkers. Long non-coding RNAs (lncRNAs) are the second most commonly studied ncRNAs, following miRNAs. In 2012, the ENCODE Project Consortium identified 9640 lncRNA loci in the human genome and this number continues to grow^{160, 275}. LncRNAs are classified generally into five broad categories based on their orientation and genomic location with relation to protein-coding genes: sense, antisense, bidirectional, intergenic, and intronic^275–277. Although the biological role of most lncRNAs still is unknown, a growing body of literature suggests that they have a wide variety of roles in controlling the expression of genes and miRNAs in cancer. These studies have shown that lncRNAs are involved in a variety of regulatory activities including chromatin remodeling, transcriptional activation, decoy (transcriptional repressor), and RNA degradation. They can also act as miR sponges and affect translational efficacy (Figure 2D)²⁷⁸.

In addition, dysregulation of lncRNA expression occurs in a tissue-specific and organ-specific manner^279–281, and appear to contribute to CRC tumorigenesis²⁸², as summarized in Supplementary Table 2. In 2001, lncRNA dysregulation was identified in CRC in the setting of loss of imprinting of the long QT intronic transcript 1 (LIT1/KCNQ1OT1)²⁸³. Emerging evidence indicates that the aberrant expression of lncRNAs may have a functional role in CRC pathogenesis which would have obvious clinical implications ²⁸⁴. For instance, the oncogenic lncRNA, HOX Antisense Intergenic RNA (HOTAIR) is a 2158 bp gene residing in the mammalian HOXC locus on chromosome 12q13.13. HOTAIR binds PRC2 and LSD1 complexes in the 5’ and 3’ regions, and represses transcription of the HOXD cluster by acting as a scaffold for histone²⁸². Koga and colleagues were the first to demonstrate that high expression of HOTAIR significantly correlated with distant metastasis and poor prognosis in CRC patients²⁸⁵, a finding later validated by another group²⁸⁶. Another lncRNA, Metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) was first identified as a predictive biomarker of metastasis in non-small cell lung cancer patients²⁸⁷, and subsequently was found to be prognostic in CRC patients²⁸⁸. LncRNAs were recently detected in extracellular vesicles and were successfully measured in the serum or plasma^289–294, indicating their potential to serve as minimally-invasive biomarkers in CRC.

CONCLUSION AND PERSPECTIVES

Advances in our understanding of the natural history of CRC and the epigenetics of colon polyps and CRC has led to the development of epigenetic biomarker assays for CRC diagnosis, prognosis, and prediction of treatment response (Figure 3). The last two decades of research have demonstrated the potential of aberrant DNA methylation and alterations in noncoding RNAs to be used as biomarkers for colon polyps and CRC. Continued investigation of these promising class of biomarkers promises to lead to an high performance assays that can be used to prevent and manage patients with CRC.

Figure 3.

A schematic view of bench-to-bedside aspects of colorectal cancer epigenetics. This figure illustrates how a normal colonic epithelium undergoes a series of genetic and epigenetic alterations and transitions into an adenomatous polyp (via the “traditional pathway), or a serrated polyp (via the “serrated pathway’). Thereafter, these polyps acquire additional epigenetic alterations and genetic alterations and develop into primary CRCs. This normal-polyp-cancer multi-step cascade is governed by the acquisition of gene mutations and epigenetic alterations, including aberrant DNA methylation and dysregulated expression of several miRNAs. The molecular alterations during each step of colorectal cancer development can be measured in tissues (tissue-based biomarkers), as well as non-invasively in serum/plasma (blood-based biomarkers) and stool (stool-based biomarkers) and thus have potential to be diagnostic, prognostic and predictive biomarkers for colorectal cancer.