Generally speaking, in vitro studies are the first step in the discovery of new epigenetic biomarkers. Researchers often compare profiles of CRC cell lines with normal colorectal cells and then compile a list of candidate biomarkers for further study. Similarly, identifying valuable prediction biomarkers in CRC patients often begins with preclinical studies using xenograft tumors, which allow one to observe tumor growth and how it responds to different therapies. Nevertheless, a significant shortcoming of this approach is that the tumor and vasculature are of mouse origin rather than human. Studying tumors in a different growth environment makes it difficult to explain the results accurately and translate them into clinical application. However, in vitro and preclinical studies are still the foundations on which most clinical studies are built.

In the human genome, DNA methylation typically occurs on the cytosine of the sequence 5’-CpG-3’, which is found in promoter regions of approximately 70% of genes[6]. In this biochemical process, a methyl group (-CH₃) is added to cytosine nucleotides by a DNA methyltransferase (DNMT). A large body of evidence has demonstrated that promoter hypermethylation is associated with gene silencing, while hypomethylation results in gene-product upregulation. In this section, in vitro studies will be discussed first, followed by clinical studies that utilize blood or stool to identify DNA methylation in CRC patients.

Khamas et al[7] conducted a genome-wide screen of 15 CRC cell lines and 23 paired tumor and normal samples from CRC patients to identify a set of methylation-silenced genes in CRC. Gene expression studies were then used to confirm whether the methylated genes were really regulated by their methylation status. The results of this study revealed that 139 genes showed greater than 1.5-fold up-regulation in at least one 5-aza-2’-deoxycytidine-treated cell line and no less than a 1.2-fold change in other treated CRC cell lines. Among them, eight genes, DCAF4L1, DDX43, ICAM1, MSX1, PGF, PTPRO, ZFP42 and the cancer-germline antigen families, had previously been reported to be up-regulated by demethylation in CRC and were thus excluded from the analysis. Twenty genes with poor annotation, 20 genes located on the X chromosome, 16 genes with duplicated probes, two genes with no CpG islands, 8 genes with unknown function, 23 without a relevant function in tumorigenesis and 22 genes with potential oncogenic activity were also excluded, leaving 20 candidates (CAMK2B, CHAC1, THSD1, CSTA, COL1A1, GADD45B, DMRTB1, COL6A1, GAS5, GPRC5A, GPSM1, KLHL35, LTBP2, NAA11, RBP4, SEMA7A, SYCP3, TBRG1, TNFSF9 and TXNIP) that had not been previously reported to be affected by epigenetic mechanisms in CRC. Therefore, from the 54613 genes analyzed, a much smaller set of genes was isolated as potential biomarkers for CRC.

In this study, two genes, THSD1 and GADD45B, were selected for further analysis. THSD1 methylation appeared to have the potential for diagnostic, prognostic or therapeutic use. Thrombospondin type-1 domain-containing protein 1 (THSD1) is located in a region that is strongly associated with the progression of colorectal adenoma to carcinoma and encodes a transmembrane molecule containing a thrombospondin type 1 repeat that might be involved in cell adhesion and angiogenesis. High THSD1 expression positively correlated with better distant metastasis survival in breast cancer. Therefore, its loss may be associated with metastatic tumor spread. Additionally, as one of the consensus radiation-response genes in primary human fibroblasts, THSD1 may play a role in radiation response in cancer stem cell. Moreover, a recent study has shown that THSD1 was expressed in CRC classified as D in Duke’s classification scheme for CRC and thus may be relevant to tumor progression[8]. GADD45B functions as a tumor suppressor in many cancers, can inhibit cell proliferation at different stages and induce cell apoptosis, but its function in CRC is unknown.

In addition, Schuebel et al[9] described another genome-wide, expression array-based approach for the identification of genes silenced by promoter hypermethylation in human CRC, and approximately 500 hypermethylated genes were identified. They analyzed the top-tier hypermethylome of each cell line (HCT116 and SW480) and then made a comparison of hypermethylation frequencies in cell lines, normal human tissues and human tumor samples. They found that BOLL, DKK3, CABYR, EFEMP1, GNB4, GSTM3, FOXL2, HOXD1, JPH3, NEF3, NEURL, PPP1R14A, RAB32, TLR2, SALL4, TP53AP1 and ZFP42 were hypermethylated and underexpressed in both CRC cell lines and in colon cancers, but not in normal tissues. These genes possess great promise as useful biomarkers for molecular diagnostics, early detection and CRC therapy. Recently, Yi et al[10] also reported that hypermethylation of promoter DNA in the FBN2 and TCERG1L genes might provide excellent biomarkers for early detection of CRC. Both genes showed a high frequency of methylation in colon cancer cell lines, adenomas and carcinomas.

In addition, methylation of the hMLH1, p16INK4A, APC, MGMT, sFRP1, GATA-5, sFRP4, sFRP5, GATA-4, B4GALT1 TFPI2, SOX17 and TMEM25 genes has been described in several studies[11-17]. These genes are hypermethylated and downregulated in CRC and thus may serve as excellent candidate biomarkers. In addition, insulin-like growth factor-binding protein 3 and Enah/Vasp-like have been validated as prognostic biomarkers for CRC and found to be useful in stratifying high-risk CRC patients who would benefit from adjuvant chemotherapy[18]. PPP2R2B was also found to be hypermethylated in CRC and was connected to therapeutic resistance[19]. These genes could serve as candidate biomarkers for prognosis. However, clinical studies are required to confirm these results, and it remains to be seen if these alterations can be detected in blood or stool.

Biomarkers detected in patient blood samples would provide the most practical screening tool for CRC because of the ease with which these samples can be acquired. It has been well documented that genetic material can shed from tumor cells, and aberrant DNA methylation can be specifically quantified in blood despite the large amounts of normal DNA in circulation. Bisulfite treatment and methylation specific polymerase chain reaction (PCR) are the two most commonly used techniques. A blood biomarker with a high sensitivity and specificity for CRC can not only be used to segregate high-risk patients for further clinical tests but also be an excellent tool for monitoring CRC recurrence in patients who have undergone tumor resection (Table 1)[20-30].

The SEPT9 gene, encoding a guanosine triphosphate enzyme involved in cytokinesis and cell cycle control, has been reported to be associated with several cancers. The v2 region of the Septin 9 (SEPT9) promoter has been shown to be methylated in CRC tissue compared with normal colonic mucosa. Using highly sensitive real-time PCR assays, methylated SEPT9 was first detected in the plasma of CRC patients with an overall sensitivity of 72% and a specificity of 90%[24]. Significant validation has been performed for this methylation biomarker, and Warren et al[27] have confirmed a sensitivity of up to 90% and a specificity of up to 88% for SEPT9. Based on these results, SEPT9 methylation appears to have the highest probability of correctly distinguishing between the blood of cancerous and non-cancerous persons for CRC detection. Currently, two CRC detection kits using plasma SEPT9 methylation analysis are marketed for clinical application. Combining SEPT9 with other methylation biomarkers would improve the detection rate[31]. Further studies are needed to compare these panels and kits and discover their advantages and limitations. Ultimately, the most effective ones should be chosen for clinical use.

Other genes, such as APC, hMLH1, ALX4, TMEFF2, NGFR, NEUROG1, SFRP2, CDKN2A/P16, TPEF/HPP1 and RUNX3, have also emerged as serum methylation markers for CRC, with sensitivities ranging from 6% to 83% and specificities ranging from 69% to 100% (Table 1). Among them, ALX4, TMEFF2 and NEUROG1 showed better performance relative to the others, and the use of these markers in combination can improve detection accuracy[25,31].

In addition to the successful identification of DNA methylation-based blood biomarkers, it is important to find genes that have prognostic value in the blood of patients with CRC. Methylation of helicase-like transcription factor (HLTF) has shown a strong correlation with tumor size, metastatic disease and tumor stage and is also associated with an increased risk of disease recurrence in CRC patients. Therefore, the methylation of this gene can serve as an independent biomarker for the identification of CRC with an increased risk of death. These results indicate that detection of HLTF methylation in the blood of CRC patients has the potential as a pretherapeutic predictor of patient outcome[32]. Deafness, autosomal dominant 5 (DFNA5) is another candidate biomarker for the noninvasive screening and monitoring of CRC. DFNA5 methylation has been observed in DNA from the peripheral blood (PB) of CRC patients at a high frequency (48% or 12/25) relative to healthy controls (only 12% or 3/25). Moreover, the methylation of DFNA5 in PB samples from CRC patients was significantly correlated with lymph node metastasis and distant metastasis (P = 0.027)[5], which suggests that DFNA5 could potentially be an independent prognostic serum biomarker for CRC patients. It is clear, however, that further validation in large-scale prospective trials is necessary before these biomarkers are ready for use in the clinic[26].

As a more attractive alternative to tissue sampling, biomarkers from feces could be of great clinical value because sampling is noninvasive and has much higher specificity. These properties offer a distinct advantage over endoscope- and FOBT-based screening strategies for the detection of both CRC and critical precursor lesions. Over the past decade, numerous studies have engaged in the development of methylation-based detection assays for stool biomarkers of CRC (Table 2)[33-66], though the fecal biomarker detection can only be performed in only less than 50% of patients due to very limited compliance. The best-studied and top-performing methylation biomarkers are secreted frizzled-related protein 2 (SFRP2) and vimentin.

SFRP2 was the first reported DNA methylation marker in stool, has shown a sensitivity of 77%-90% and specificity of 77%[66] and has since been studied extensively. SFRP2 methylation has been shown to be the most sensitive biomarker for CRC, with detection rates ranging from 77% to 94% (Table 2). When SFRP2 methylation was used in a multigene, fecal methylation panel, detection of CRC and a small number of advanced adenomas reached a sensitivity and specificity of 96%[58]. A follow-up study found that SFRP2 methylation was detectable in the stool of almost half of all patients with hyperplastic polyps or colorectal adenomas[53], further supporting its use in the detection of premalignant lesions. Fecal SFRP2 methylation also drops dramatically after surgery [postoperative: 8.7% (6/69) vs preoperative: 87% (60/69)][52], suggesting its possible utility as a biomarker for recurrence.

The vimentin gene, which encodes an intermediate filament protein involved in cell attachment, migration, and signaling, was identified in the stool of 83% of CRC patients with a specificity of 90%[62]. Since then, many studies have been devoted to vimentin methylation. Follow-up studies have obtained similar results and thus have reinforced the utility of vimentin as a standalone biomarker[47,54,55,59,61]. This has led to the commercialization of a single-gene stool kit for CRC detection based on vimentin methylation. More recently, vimentin methylation has been used in combination with other methylation markers to further increase detection rates, and vimentin has also been found in urine, suggesting an alternative method of detection[34,48,67]. Vimentin has a low detection rate in serum, however, and is thus most likely not suitable for use as a serum biomarker for CRC. Recently, Ahlquist et al[26] reported that a panel of methylation markers from stool that includes vimentin has shown a significantly higher sensitivity for CRC, primarily because of higher detection rates in stage I-III CRC (91% vs 50%).

In addition to SFRP2 and vimentin, several other methylation biomarkers have been identified; these include GATA4, HIC1, ITG4, NDRG4, OSMR, TFPI2, ESR1, SLIT2, PHACTR3, SPG20, 3OST2 and MGMT. These genes have sensitivities for CRC ranging from 38% to 89% and specificities ranging from 79% to 100%. The combination of different methylation biomarkers (combinations of 2 to 7 genes including APC, ATM, CDKN2A, GSTP1, HLTF, hMLH1, HPP1, MGMT, RASSF1, SFRP2, MAL, P16, or vimentin) increased sensitivity from 55% to 100% and increased specificity from 87% to 100% (Table 2). However, more clinical studies are required to confirm these results.

In recent years, miRNA has been a relatively new but rapidly expanding field, as is evidenced by the increasing number of assays in development. miRNAs are small non-coding RNA molecules that function in transcriptional and post-transcriptional regulation of gene expression and control various cellular functions. Currently, more than 1000 miRNAs have been discovered in the human genome, and their activities and regulatory mechanisms are being intensively investigated. miRNAs typically function via base pairing with complementary sequences in mRNA molecules, resulting in gene silencing via translational repression or target degradation. It has been well documented that many miRNAs are regulated by the methylation of their promoter region, and some miRNAs target epigenetic activity. For example, miR-29b has been reported to induce DNA hypomethylation and the re-expression of tumor suppressor genes in acute myeloid leukemia by targeting DNMT[68]. These results suggest that there is a strong relationship between miRNA expression and epigenetic mechanisms. Notably, many miRNAs have been found in CRC, and researchers have quantified specific miRNAs for the purpose of CRC diagnosis and prognosis in patient blood, stool and tissue samples. In vitro studies have also been conducted to identify any correlation between epigenetic aberrations and therapy response.

Currently, 54 miRNAs have been identified that are regulated either up or down in CRC cells relative to non-tumor cells (Table 3)[69,70]. Of these, miR-17, miR-20, miR-21, miR-31, miR-92a, miR-93, miR-183 and miR-203 were upregulated in CRC cells, while miR-30a, miR-30c, miR-133a, miR-143, miR-145 were downregulated. These observations have been validated in subsequent studies. The upregulated miRNAs were associated with chromosomal regions that are often amplified in CRC, and the downregulated miRNAs often associated with chromosomal regions that were typically deleted. These changes may be closely related to genetic alterations as well as epigenetic modification.

However, there are some discrepancies between studies. For example, miR-34a, miR-191 and miR-378 were reported to be upregulated in some studies[71-73] and yet were down regulated in others[74-76]. This may have been caused by heterogeneity between the different studies with regards to tumor stage, tumor location, genetic background and technical issues. We believe that the accumulation of further studies will allow us to determine which miRNAs will be the most effective biomarkers and also better understand their role in colorectal cancer.

It is widely believed that miRNAs can shed from tumor cells via exosomes and survive in a stable form in the circulation. Many studies have been performed to quantify miRNAs in the blood for use as a biomarker (Table 4)[77-88]. miR-92a, located on chromosome 13q13, is a member of the miR-17-92 gene cluster. This cluster promotes cell proliferation, suppresses apoptosis, induces angiogenesis and accelerates tumor progression. miR-92a was first identified by Ng et al[84] as a potential noninvasive biomarker for CRC detection with a sensitivity of 89% and specificity of 70%. miR-17-3p, another member of the miR-17-92 gene cluster, was also evaluated in this study as a detection biomarker. This miRNA produced a sensitivity of 64% and a specificity of 70%.

To follow this study, Huang et al[82] performed a receiver-operating characteristic (ROC) analysis on 120 CRC patients, 37 patients with advanced adenomas and 59 healthy controls. In this analysis, the researchers found that they could not only discriminate CRC from controls (miR-29a yielded an area under the curve (AUC) of 0.844, and miR-92a yielded an AUC of 0.838), but also discriminate advanced adenomas from controls (the AUC was 0.769 for miR-29a and 0.749 for miR-92a). Furthermore, combined ROC analyses using these two miRNAs revealed an increased AUC with an 83.0% sensitivity and 84.7% specificity in discriminating CRC, and an AUC demonstrating 73.0% sensitivity and 79.7% specificity in discriminating advanced adenomas. These results suggested that plasma miR-29a and miR-92a have potential as novel noninvasive biomarkers for CRC detection and that a combination of different miRNAs may provide a higher sensitivity and specificity than a single miRNA.

More recently, miR-21, miR-601, miR-760 and miR-221 from plasma were also reported to be potential CRC biomarkers. In these studies, miR-221 and miR-21 were up-regulated in the plasma of CRC patients compared to healthy controls[79,83], while miR-601 and miR-760 were down-regulated[78]. Moreover, a study conducted in two independent CRC cohorts suggested that high levels of plasma miR-141 could predict poor survival, and thus miR-141 may serve as an independent prognostic factor for advanced CRC patients[81].

Stool-based miRNA detection has been widely studied as a noninvasive screening method for CRC (Table 4). Koga et al[87] conducted an miRNA expression analysis of exfoliated colonocytes isolated from the feces of 197 CRC patients and 119 healthy controls. They analyzed the miRNA expression of the miR-17-92 cluster (including miR-17, miR-18a, miR-19a, miR-19b, miR-20a and miR-92a), miR-21, and miR-135 by quantitative real-time PCR and found that expression of the miR-17-92 cluster and miR-135 was much higher in CRC patients than in healthy controls (P < 0.0001). miR-21, on the other hand, could not discriminate between the two groups. The miR-17-92 cluster detected distal tumors better than proximal tumors, as the sensitivity of miRNA expression for these tumors was 81.5% and 52.9%, respectively.

In another study, Wu et al[85] evaluated the feasibility of miR-21 and miR-92a detection in stool samples from 88 patients with CRC, 57 patients with colorectal polyps and 101 healthy controls. These results showed that patients with CRC had significantly higher levels of miR-21 (P < 0.01) and miR-92a (P < 0.0001) in their stool compared with normal controls. miR-92a levels provided a higher sensitivity for distal rather than proximal CRC (P < 0.05). In addition, stool miR-21 and miR-92a levels decreased significantly (P < 0.01) after surgical resection of tumor, which suggests that miR-92a and miR-21 from stool samples could serve as screening biomarkers for colorectal cancer.

In addition, miR-144* and miR-106a were found to be significantly overexpressed in adenomas and in the stool of CRC patients compared with healthy individuals[86,89]. These studies have confirmed that miRNAs from stool samples require validation as diagnostic biomarkers for CRC.

miRNAs have been closely linked to colorectal cancer development. They can serve as screening and diagnosis markers for CRC and also as potential prognostic and predictive markers. As a rough outline for the reader, we provide here a table to display the relationship between currently identified miRNAs and screening, diagnosis, prognosis and treatment in colorectal cancer (Table 5). As research continues, more miRNAs correlated with CRC will be discovered, and the mechanism of miRNA regulation will be deciphered. Therefore, it is highly likely that more effective miRNA biomarkers for CRC patients will be found in the future.