Analysis of DNA Methylation in Bowel Lavage Fluid for Detection of Colorectal Cancer

Colorectal cancer is one of the most commonly occurring malignancies worldwide, and early detection is essential for its successful treatment. Large population studies have shown that the fecal occult blood test (FOBT) is a highly cost-effective screening method that reduces colorectal cancer–related mortality (1). Moreover, the performance of the immunochemical FOBT (iFOBT or Fecal Immunochemical Test, FIT) has been improved (2, 3) and is now widely used for colorectal cancer screening in Japan and Europe. However, the FOBT continues to have limitations, especially for detection of early-stage colorectal cancers. Several other methods, including colonoscopy and barium enema, have been available for years, but none of these methods has been established as a gold standard for colorectal cancer screening.

Fecal DNA tests are a noninvasive and potentially effective means of screening for both early colorectal lesions and advanced colorectal cancers (4, 5). As such, the feasibility of detecting genetic mutation of oncogenes or tumor-suppressor genes, such as APC, KRAS, TP53, and BAT-26, has been extensively tested, but the diagnostic performance of these assays remains unsatisfactory (6, 7). Epigenetic alterations are also commonly observed in colorectal cancers. Because of its high frequency and the wide variety of affected genes, aberrant DNA methylation has emerged as a new biomarker for stool-based colorectal cancer screening. For instance, SFRP2 methylation occurs in approximately 90% of primary colorectal cancers (8), and was one of the first epigenetic markers reported in fecal DNA (9). More recently, a variety of other genes have been identified as potential biomarkers for stool-based methylation testing, including VIM, GATA4, TFPI2, PHACTR3, AGTR1, WNT2, and miR-34b/c (10–15).

In an earlier study, we demonstrated that DNA methylation is detectable in the mucosal wash fluid from colorectal tumors, which can be collected during colonoscopy (16). Importantly, wash fluid from invasive cancers exhibited significantly higher levels of methylation of tumor-related genes than noninvasive tumors. This prompted us to postulate that wash fluid from invasive tumors contained greater numbers of exfoliated tumor cells, and that the methylation was a potential biomarker predictive of tumor invasiveness. Our results also suggested that a DNA methylation test might complement the diagnostic performance of colonoscopy and that intestinal wash fluid could be a useful source for analysis of tumor-derived DNA methylation. We therefore hypothesized that oral bowel lavage fluid (BLF) might contain tumor-derived DNA, and thus molecular alteration in BLF specimens could be a useful biomarker for colorectal cancer screening. To test that idea, in this study, we analyzed DNA methylation of tumor-related genes in BLF specimens from patients with colorectal tumors and healthy individuals, and examined its clinical utility for cancer detection.

All samples were collected from Japanese patients who underwent colonoscopy at Akita Red Cross Hospital (Akita, Japan) because of abdominal symptoms or a positive FOBT. Informed consent was obtained from all patients before collection of the specimens. Approval for this study was obtained from the Institutional Review Board of Akita Red Cross Hospital and Sapporo Medical University (Sapporo, Japan). Before colonoscopy, patients were pretreated with 2 L of polyethylene glycol lavage solution and 10 mL of BLF specimens were collected from the rectum at the beginning of the colonoscopy (Fig. 1A). BLF samples were initially classified into four groups according to the Boston bowel preparation scale (BBPS; Fig. 1B; ref. 17). Then, on the basis of colonoscopic and histologic findings, the BLF samples were divided into four groups: patients with colorectal cancer, patients with advanced adenoma, patients with minor polyp, and individuals without colorectal lesions. Advanced adenomas were defined as being 1 cm or more in diameter, and/or with villous components, and/or with high-grade dysplasia. Minor polyps were defined as being adenomas that did not satisfy the above criteria. A total of 508 BLF samples from 56 patients with colorectal cancer, 53 patients with advanced adenoma, 209 patients with minor polyp, and 190 individuals with a normal colon were collected. In addition, biopsy specimens were collected from 44 of the 56 patients with colorectal cancer. BLF and tissue specimens were suspended in ThinPrep PreservCyt solution (Hologic) and stored at 4°C until DNA extraction. Genomic DNA was extracted using the standard phenol–chloroform procedure. FIT was performed in 349 individuals, including 17 patients with colorectal cancer. Samples were randomly sorted into two groups (training set and test set) for validation analysis (Table 1).

Genomic DNA (1 μg) was modified with sodium bisulfite using an EpiTect Bisulfite Kit (Qiagen), after which methylation analysis was carried out as described previously (18). PCR for MethyLight assays was run in a 20-μL volume containing 50 ng of bisulfite-treated DNA, 625 nmol/L each primer, 250 nmol/L TaqMan-MGB probe, and 1× TaqMan Fast Universal PCR Master Mix (Applied Biosystems). Fast real-time PCR was done using a 7500 Fast Real-Time PCR System according to the manufacturer's instructions (Applied Biosystems). The PCR protocol entailed 20 seconds at 95°C followed by 40 cycles of 3 seconds at 95°C and 30 seconds at 60°C. The Alu repetitive element was used as an endogenous control, and the percentage of methylated reference (PMR) was calculated as described previously (19, 20). Sequence information for the primers and probes used for miR-1-1, miR-9-1, miR-9-3, miR-34b/c, miR-124-1, miR-124-2, miR-124-3, miR-137, SFRP1, SFRP2, DKK2, WIF1, LOC386758, and ZNF582 is listed in Supplementary Table S1; those used for APC is described elsewhere (20).

Quantitative variables were analyzed using the Student t test. The Fisher exact test and the χ² test were used for analysis of categorical data. The Pearson correlation coefficient was used to evaluate correlations between continuous data. Receiver–operating characteristic (ROC) curves for the diagnosis of colorectal cancer were constructed on the basis of the methylation levels, followed by calculation of the area under the curve (AUC). The best cutoff PMR value for each gene was defined as the point on the ROC curve closest to the upper left corner. A diagnostic scoring system using a panel of selected marker genes was constructed by analyzing a training set using the following three-step algorithm: (i) methylation status of marker genes in BLF was assessed; (ii) the number of methylated genes was determined, which we termed the methylation score (M-score); and (iii) the samples were classified into four groups based on the M-score. Values of P < 0.05 (two-sided) were regarded as significant. All statistical analyses were performed using the SPSS Statistics 18 (IBM Corporation) and GraphPad Prism ver. 5.0.2 (GraphPad Software).

After collecting 10-mL BLF specimens from the rectums of the study participants at the beginning of their colonoscopy, we successfully extracted sufficient amounts of genomic DNA to perform a methylation analysis (Fig. 1A). To determine the best time to obtain the BLF specimens, we scored the BLF samples using the BBPS (Fig. 1B; ref. 17). Among the 268 BLF samples initially collected, 58 were scored as 3, 154 were scored as 2, 46 were scored as 1, and 10 were scored as 0. BLF samples without residual stool (BBPS scores 2 and 3) contained significantly smaller amounts of genomic DNA than those with residual stool (BBPS scores 0 and 1; Fig. 1C). However, MethyLight assays revealed that the endogenous control Alu element was detected at lower threshold cycle (C_t) values in BLF specimens with high BBPS scores than in those with residual stool (Fig. 1C). This suggests that the relative fraction of human genomic DNA is larger in higher BBPS score BLF, most likely because of the smaller amount of contaminating bacteria-derived DNA. We then analyzed BLF specimens from selected patients with colorectal cancer, comparing the detectability of DNA methylation between specimens with lower and higher BBPS scores. As shown in Fig. 1D, methylation of representative genes was readily detectable in BLF specimens with a higher BBPS score, whereas it was undetectable in specimens with a lower score (Fig. 1D). For these reasons, we collected BLF specimens after sufficient bowel preparation for the next analysis.

Our training set consisted of 355 BLF specimens obtained from patients with colorectal cancer (n = 39), advanced adenomas (n = 31), or minor polyps (n = 135), as well as individuals with no colorectal lesions (n = 150; Table 1). Using these specimens, we performed quantitative MethyLight assays to assess the methylation status of 15 genes known to be frequent targets of aberrant CpG island methylation in colorectal cancer (miR-1-1, miR-9-1, miR-9-3, miR-34b/c, miR-124-1, miR-124-2, miR-124-3, miR-137, SFRP1, SFRP2, APC, DKK2, WIF1, LOC386758, and ZNF582; refs. 8, 21). The methylation levels of the respective genes were calculated as PMR values, and we generated ROC curves to assess their clinical utility for detection of colorectal cancer (Supplementary Table S2 and Supplementary Fig. S1). Among the candidate marker genes analyzed, we found that miR-124-3, LOC386758, and SFRP1 were highly discriminative between patients with colorectal cancer and those without colorectal cancer (Supplementary Table S2). The most discriminating PMR cutoffs for miR-124-3, LOC386758, and SFRP1 were 11.1, 0.0003, and 1.1, while the most sensitive setting (PMR > 0) also achieved high sensitivity and specificity (Supplementary Table S2).

To develop a more efficient diagnostic system for detection of colorectal cancer, we constructed a scoring system based on the methylation of miR-124-3, LOC386758, and SFRP1. Using the number of methylated genes (PMR > 0), we classified the samples into four groups based on their M-score (Fig. 2A). A ROC curve was then constructed to evaluate the ability of the scoring system to distinguish samples obtained from patients with colorectal cancer by plotting the sensitivity over 1 − specificity at each point (Fig. 2B). We then validated the diagnostic system by analyzing an independent test set (Table 1 and Fig. 2A and B). AUCs in the training and test sets were 0.834 and 0.808, respectively, confirming the accuracy of our system for detecting colorectal cancer.

The association between the clinical characteristics and M-scores is summarized in Table 2 and Supplementary Table S3. Higher M-scores were significantly associated with colorectal cancer, but their association with advanced adenomas or minor polyps was limited (Fig. 2C and Supplementary Table S3). M-scores were not significantly associated with tumor location, size or stage (Table 2). We also did not find a correlation between M-scores and age in patients with colorectal cancer, though we observed a tendency for higher M-scores to be associated with older age in non–colorectal cancer individuals, perhaps due to age-dependent methylation (Supplementary Fig. S2). These results suggest that the M-score system is able to reveal the presence of colorectal cancers, irrespective of the tumor's location, size, or clinical stage, but greater age may increase the false-positive rate.

Although the results summarized above demonstrate the clinical utility of BLF methylation for colorectal cancer screening, the system failed to detect 5 of the 56 patients with colorectal cancer (Supplementary Table S3). We therefore tested whether the apparent absence of methylation in those 5 BLF specimens actually reflects the unmethylated status of the genes in tumor tissues. For this purpose, we analyzed biopsy specimens from 41 patients with colorectal cancer with different M-scores (score 3, n = 20; score 2, n = 10; score 1, n = 8; and score 0, n = 3), and found that the majority of the tumors exhibited methylation of all three genes (miR-124-3, LOC386758, and SFRP1), irrespective of the M-score (Supplementary Fig. S3). MethyLight assays revealed that the C_t values for the endogenous Alu tended to be higher in BLF specimens with low M-scores, indicating that the apparent absence of BLF methylation may be result of too little tumor-derived DNA in the sample.

We next assessed whether BLF methylation could be used to detect upper gastrointestinal tract cancers. Among the individuals enrolled in this study were 294 who underwent upper gastrointestinal endoscopy; of those, 21 were found to have a gastric cancer. BLF methylation was detected in 12 of the 21 patients with gastric cancer, and a majority of the positive cases showed only a minimal number of methylated markers (M-score 0, n = 9; score 1, n = 8; score 2, n = 1; and score 3, n = 3). Five of the 8 patients with gastric cancer with minimal methylation (M-score, 1) also had minor colorectal polyps, which could also have been the source of the methylated DNA. Interestingly, 2 of the 4 patients with gastric cancer with high M-scores (>2) were also found to have colorectal cancers, while the remaining 2 patients showed no colorectal lesions. These results suggest it would be difficult to use BLF methylation as a biomarker for upper gastrointestinal cancers.

FIT was performed in 349 of the study participants, including 17 patients with colorectal cancer (Table 3). Most of the patients with colorectal cancer were positive on the FIT (14 of 17), while a significant number of colorectal cancer–free individuals also showed positive results (142 of 332). For that reason, we next tested whether the diagnostic performance of FIT could be improved by combining it with the BLF methylation test. In the FIT-negative group (n = 193), which included 3 patients with colorectal cancer, all the patients with colorectal cancer were detected using the M-score system (Table 3). In the FIT-positive group, most of the patients with colorectal cancer (12 of 14) exhibited BLF methylation (M-score ≥ 1), while a majority of the BLF methylation-negative subjects were colorectal cancer–free (80 of 82). Thus, the combination of FIT and the BLF methylation test significantly improved the positive predictive value (PPV) in both the FIT-negative and FIT-positive groups.

Because computed tomographic colonography (CTC) has emerged in recent years as a noninvasive screening method for colorectal cancer (22), we examined the feasibility of using BLF methylation testing to complement the diagnostic performance of CTC. Among the subjects enrolled in this study, 9, including 5 patients with colorectal cancer, were examined using CTC (Table 4). CTC detected four colorectal cancers, while all 5 patients with colorectal cancer were positive for BLF methylation (M-score, > 2). Notably, 1 patient (case 5) developed a laterally spreading tumor (LST) that consisted of a histologically benign polypoid component and a flat adenocarcinoma component. CTC detected only the polypoid component, so the lesion was diagnosed as a minor polyp (Table 4 and Supplementary Fig. S4). Our results suggest that combining assessment of BLF methylation with CTC may improve diagnostic performance, though further study of a larger population will be necessary to confirm the clinical utility of this combination.

Numerous studies have shown that aberrant methylation of DNA in the stool is a promising biomarker suitable for noninvasive colorectal cancer screening. For instance, VIM, SFRP2, and TFPI2 are reported to be useful single-gene markers for a fecal DNA methylation test (9, 10, 12). In addition, other groups have shown that combinations of multiple markers improve the diagnostic efficacy of stool DNA methylation (14). In this study, we demonstrated that aberrant DNA methylation is detectable in the wash fluid of oral bowel lavage collected from the rectum of patients with colorectal cancer. Earlier studies have shown that methylation of DNA in body fluids, including pancreatic juice (23), saliva (24), and gastric juice (25), has the potential to serve as a biomarker for cancer detection and risk assessment, yet there have been no studies assessing the feasibility of using BLF for molecular screening for colorectal cancer. Importantly, we found that the utility of BLF depends on successful bowel preparation, and that residual stool may interfere with sensitive detection of tumor-derived DNA methylation. Although the total amount of extracted DNA is small, BLF specimens with sufficient bowel preparation appear to contain a greater proportion of tumor-derived DNA than those with insufficient treatment.

In this study, we tested a set of genes known to be frequently methylated in colorectal cancer, and selected the three genes with the highest sensitivities for detection of colorectal cancer (miR-124-3, SFRP1, and LOC386758). The miR-124 family consists of three members (miR-124-1, miR-124-2, and miR-124-3), all of which are reportedly methylated in multiple types of human malignancy, including colorectal cancer and gastric cancer (26, 27). SFRP1 encodes secreted frizzed-related protein 1, a negative regulator of Wnt signaling, and the promoter CpG island of SFRP1 is frequently methylated in various cancers, including colorectal cancer, gastric cancer, and esophageal cancer (28–30). LOC386758 is a frequent target of aberrant methylation newly identified in our recent epigenome analysis in colorectal cancer, though its function remains unknown (manuscript in preparation). Although BLF methylation of each of these genes could be used to detect colorectal cancer with relatively high sensitivity and specificity, we found that combining them improved diagnostic accuracy. Importantly, BLF methylation was not affected by tumor size, location, or stage, suggesting that it could potentially serve as a biomarker for both proximal and distal colon cancers.

However, the BLF methylation system failed to detect a small number of colorectal cancers as well as more than half of the precancerous lesions (minor polyps and advanced adenomas). We confirmed that the negative result was not due to the absence of methylation in the tumor tissues. In addition, we and others have previously shown that many of the 15 genes analyzed in this study are frequently methylated in colorectal premalignant lesions (31, 32). Although the true reason for the false-negative finding remains uncertain, we suspect that the presence of a too small number of exfoliated cells in the BLF is the cause. We have previously shown that DNA methylation in colonoscopically obtained mucosal wash fluids could be a predictive biomarker of tumor invasiveness (16). By performing quantitative bisulfite-pyrosequencing, we detected elevated levels of DNA methylation of tumor-related genes (miR-34b/c, SFRP1, SFRP2, and DKK2) in the mucosa of invasive tumors, though these genes were equally methylated in noninvasive and invasive tumors. Early during this study, we found that we were unable to detect BLF methylation using bisulfite-pyrosequencing, so we switched to the more sensitive MethyLight assay. We therefore suggest that the numbers of exfoliated cells and the amount of cell-free DNA in BLF specimens are far smaller than in the colonoscopically obtained mucosal wash fluid. Moreover, BLF specimens with high M-scores tended to show lower C_t values for Alu elements with MethyLight, which is indicative of the relative abundance of human genomic DNA (Fig. 1C). These results suggest that successful detection of BLF methylation is highly dependent on the amount of tumor-derived DNA in the BLF specimens. In addition, we speculate that differences in the sample collection steps and the methods used for methylation analysis could be major reasons why the best marker genes differed between our early studies and the present one (16).

Our findings also suggest that BLF methylation could be used to complement current colorectal cancer screening methods. FIT is one of the most commonly used and cost-effective screening tests, but its low PPV may lead to a low compliance rate among FIT-positive individuals receiving medical advice to go for secondary screening. When combined with FIT, a BLF methylation test could significantly improve PPV and more effectively select individuals who should be strongly encouraged to undergo total colonoscopy. Moreover, our data demonstrated that BLF methylation of multiple genes could be an indicator of colorectal cancer, even among FIT-negative individuals.

As compared with stool DNA tests, the biggest disadvantage of the BLF methylation test is that it requires bowel preparation. Therefore, combination with endoscopies is another feasible clinical application of BLF methylation. For instance, when combined with sigmoidoscopy, a BLF methylation test may complement the diagnostic performance for detection of proximal colon cancers. Similarly, BLF methylation could provide supportive information for patients with unsuccessful total colonoscopy. In addition, we propose that BLF methylation may improve the diagnostic performance of CTC. Although the sensitivity of CTC for detection of some colorectal cancers is equivalent to colonoscopy, its ability to detect small or flat lesions is more limited (33–35). Moreover, it is sometimes difficult to distinguish between early-stage cancers and benign adenomas using CTC. The fact that BLF methylation is highly specific for malignant tumors indicates that it could increase the ability to detect colorectal cancers using CTC. In this study, we compared BLF methylation with CTC findings in 9 individuals, including 5 patients with colorectal cancer. Using CTC, four of the colorectal cancers were successfully detected, but a flat type cancer was diagnosed as a minor polyp. In contrast, BLF methylation (M-score, > 2) was detected in all 5 patients with colorectal cancer. These results suggest that combining assessment of BLF methylation with CTC may improve the diagnostic performance, but further prospective study of a larger number of patients will be necessary to evaluate the diagnostic performance of this combination.

In sum, our results demonstrate the feasibility of using aberrant DNA methylation in BLF specimens for noninvasive colorectal cancer screening. We also found that using a panel of several marker genes further improved the sensitivity and specificity of this diagnostic system. It is noteworthy that DNA methylation was readily detectable in BLF specimens with no purification or capture of human genomic DNA. Thus, combination with other colorectal cancer detection methods that require bowel preparation, including sigmoidoscopy or CTC, would be a suitable application of the BLF methylation test. Further technical refinements, including easier bowel preparation, single-molecule detection of methylated DNA, and identification of better marker sets, would also enhance the practicality of this test.

No potential conflicts of interest were disclosed.

Conception and design: T. Harada, E. Yamamoto, T. Niinuma, T. Sugai, K. Imai

Development of methodology: H.-O. Yamano, T. Sugai

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Harada, E. Yamamoto, H.-O. Yamano, K. Kumegawa, K. Yoshikawa, T. Kimura, E. Harada, R. Takagi, Y. Tanaka, H. Aoki, M. Nakaoka, K. Shimoda

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Harada, E. Yamamoto, M. Nojima, R. Maruyama, A. Tuyada, M. Kai, T. Sugai

Writing, review, and/or revision of the manuscript: T. Harada, E. Yamamoto, A. Tuyada, H. Suzuki

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.-O. Yamano, M. Ashida, T. Kimura, M. Nishizono

Study supervision: E. Yamamoto, Y. Shinomura, T. Sugai, K. Imai, H. Suzuki

The authors thank Tomo Hatahira for technical assistance and Dr. William F. Goldman for editing the article.

This study was supported, in part, by Grants-in-Aid for Scientific Research (i) from the Japan Society for Promotion of Science (to H. Suzuki, K. Imai), Grants-in-Aid for Young Researchers (ii) from the Japan Society for Promotion of Science (to T. Harada, E. Yamamoto), and the Project for Developing Innovative Research on Cancer Therapeutics (P-DIRECT; to H. Suzuki).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.