DNA methylation markers detected in blood, stool, urine, and tissue in colorectal cancer: a systematic review of paired samples

Purpose Methylated cell-free DNA in liquid biopsies are promising non-invasive biomarkers for colorectal cancer (CRC). Optimal markers would have high sensitivity and specificity for early detection of CRC and could be detected in more than one type of material from the patient. We systematically reviewed the literature on DNA methylation markers of colorectal cancer, detected in more than one type of material, regarding their potential as contributors to a panel for screening and follow-up of CRC. Methods The databases MEDLINE, Web of Science, and Embase were systematically searched. Data extraction and review was performed by two authors independently. Agreement between methylation status in tissue and other materials (blood/stool/urine) was analyzed using the McNemar test and Cohen’s kappa. Results From the 51 included studies, we identified seven single markers with sensitivity ≥ 75% and specificity ≥ 90% for CRC. We also identified one promising plasma panel and two stool panels. The correspondence of methylation status was evaluated as very good for four markers, but only marginal for most of the other markers investigated (12 of 21). Conclusion The included studies reported only some of the variables and markers of interest and included few patients. Hence, a meta-analysis was not possible at this point. Larger, prospective studies must be designed to study the discordant detection of markers in tissue and liquid biopsies. When reporting their findings, such studies should use a standardized format. Electronic supplementary material The online version of this article (10.1007/s00384-020-03757-x) contains supplementary material, which is available to authorized users.


Introduction
Colorectal cancer (CRC) develops over years or decades through genetic and epigenetic alterations [1]. CRC is the second leading cause of cancer-related death in the Western world [2,3], and its treatment and follow-up have a massive impact on the quality of life [4,5]. Measures to increase CRC survival would be to diagnose the disease at an earlier stage, more reliably identify patients with residual disease after treatment, more accurately diagnose recurrence, and more closely monitor the effect of oncologic treatment. Sensitive and reliable biomarkers in blood, stool, or urine would be ideal for this purpose. Hence, much effort has been put into the identification of new and improved biomarkers for early detection and follow-up of CRC [6,7]. Cell-free DNA (cfDNA) methylation markers, especially in blood and stool, are considered promising biomarkers, with remarkably high sensitivity and specificity for CRC [8,9].
Methylation of cytosine to form 5-methylcytosine at CpG dinucleotides is a widespread and normal epigenetic modification of the DNA in humans. Increased CpG methylation in promoter regions of genes, especially at CpG-rich sequences termed CpG islands, is associated with transcription repression [10]. Increased methylation in classical tumor suppressor genes, genes regulating mitosis, and DNA repair is considered an early event in CRC tumorigenesis [1,11]. Many individual markers have been investigated to date [11][12][13][14][15][16]. However, it is considered that the most useful tool for detection and followup of CRC would be a panel of cell-free DNA markers with high sensitivity and specificity for early detection of the disease, with prognostic value, with the possibility to detect residual disease, and recurrence, as well as ability to change as a result of oncologic treatment [17]. Detection even at the adenoma level would be preferable [18]. Also, optimal markers would be possible to detect both in the blood ( b ), stool ( s ), urine ( u ), and tumor tissue ( t ) throughout the course of the disease [19].
Much of the research on specific methylation markers of CRC has focused on tumor-derived DNA. Lately, single markers and panels of markers have also been tested in tumor remote media such as blood, stool, and urine and have gained much attention as potential liquid biopsies (non-invasive cancer biomarkers) [20,21]. Somewhat challenging, detection of such markers in blood/urine/stool has not been concordant with detection in CRC tissue [22][23][24]. This inconsistency between liquid biopsies and tissue biopsies has been pointed out as one of the hurdles that need to be addressed before liquid biopsies can be taken into routine clinical use [25]. To overcome this issue, some studies have analyzed methylation markers in several different materials from the same individual. The present review focuses on analyses performed in more than one type of material from the same individual. The objective is to identify cell-free DNA methylation markers of colorectal cancer detected in more than one type of material from the same patient and systematically review their potential as contributors to a panel for screening and follow-up of CRC.

Methods
The search strategy, study inclusion and exclusion criteria, data extraction, study quality assessment, and data analysis were performed based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA), the variant for diagnostic test accuracy (PRISMA-DTA), and the quality assessment for diagnostic accuracy studies (QUADAS) [26][27][28][29].

Data extraction
Data were extracted in a pre-defined form including the type of specimen, sensitivity for CRC (n/N, %), sensitivity for adenoma (n/N, %), specificity (n/N, %), and method for methylation analysis. Data extraction and review was performed by two authors independently.

Statistical analysis
Agreement between methylation status in tissue and other materials (blood/stool/urine) was analyzed using the McNemar test and Cohen's kappa. If the p value of the McNemar test was insignificant (p ≥ 0.05) and the p value of the Kappa statistic was significant (p < 0.05), the methylation status was considered as being in agreement. The criteria for the strength of agreement were as follows: K < 0.2 poor, K 0.21-0.40 fair, K 0.41-0.60 moderate, K 0.61-0.80 good, K 0.81-1.00 very good [30]. All statistics were performed using SPSS 25 for windows (SPSS Inc., Chicago, IL, USA)

Results
The systematic literature search identified 728 publications of potential relevance and 85 publications were added by hand search. The abstracts of these 813 publications were reviewed and 381 publications were excluded. Among the remaining, 283 publications considered methylation only in tissue, 51 publications considered methylation only in the blood, 25 publications considered methylation only in stool, and three publications considered methylation only in the bowel lavage ( Fig. 1). No publications considered urine only. Methylation in more than one type of material was analyzed in 70 publications, among which only 51 publications analyzed methylation in more than one type of material from the same individual.
The 51 included studies demonstrate that some of the wellknown cell-free DNA methylation markers of colorectal cancer can be detected in more than one type of material from the same patient. The extracted data are presented in full as supplementary material (Supplementary Table 1 and 2). Single markers with sensitivity for CRC of 75% or above in tissue and at least one other material were BMP3 b , EFHD1 b , ITGA4 s , NDRG4 s , OSMR b PPP1R3C b , SEPT9 b , SFRP1 b , SFRP2 s , SPG20 b,s , TFPI2 s , and VIM u ( Table 1). Among these, ITGA4, SEPT9, SFRP1, SFRP2, SPG20, TFPI2, and VIM had a specificity of 90% or above ( Table 1).
When investigating the case-by-case relationship for single methylation markers analyzed in more than one type of material from the same patient, we found a very good agreement between tissue and other materials for four markers (CDH4 in the blood, ERCC1 in the blood, p16 INK4a in the blood, and SPG20 in stool) ( Table 3). Most of the markers (12 of 21) showed marginal reproducibility (k < 0.4) between methylation status in tissue and other materials (Table 3).

Discussion
The present systematic review identified 51 publications analyzing cell-free DNA methylation markers in more than one type of material from the same individual. The markers analyzed in these studies are well known from previous studies in tissue, blood, and stool [11][12][13][14][15][16]. We will here discuss the potential role of each of the identified markers, as contributors to a panel for screening and follow-up of CRC (i.e., sensitivity and specificity for early detection, prognostic value, detection of residual disease and recurrence, ability to reflect ongoing oncologic treatment, and detectability in more than one type of material, i.e., blood (b), stool (s), urine (u), and tumor tissue (t)).
Bone morphogenetic protein 3 (BMP3) is a member of the transforming growth factor-beta (TGFβ) superfamily of cytokines, binding to cell-surface receptors, activating a cascade of cell signaling, ultimately regulating the transcription of SMAD4 target genes to achieve growth suppression [60]. Downregulation of the BMP3 tumor suppressor gene is an MethylaƟon only in blood n=51 MethylaƟon only in stool n=25 MethylaƟon only in bowel lavage n=3 MethylaƟon only in urine n=0 MethylaƟon only in Ɵssue n=283 Material from only one paƟent n=51 Material from >1 paƟents n=19 Fig. 1 Flowchart for selection of publications early and frequent event in colorectal cancer [60]. In the present systematic review, methylation of BMP3 was identified in 75% of plasma samples and 82% of tissue samples from patients with colorectal cancer [31]. However, methylation of BMP3 was also detected in 11 of the 37 plasma samples from healthy controls referred to colonoscopy with no evidence of CRC or other cancer [31]. Tissue from healthy controls was not analyzed. BMP3 has also been analyzed in other studies where the sensitivity and specificity for CRC in tissue were 57% and 93%, respectively [61], where the sensitivity and specificity for CRC in plasma varied from 29 to 40% and 89 to 94%, respectively [62,63], and where methylation in stool DNA varied from 40 to 100% in CRC and 33 to 70% in advanced adenoma samples [64,65]. BMP3 is one of two methylation markers (the other is NDRG4) included in the commercially available fecal Cologuard® test, approved by the FDA in 2014 as a colon cancer screening test [46,66]. EFHD1 (EF-hand domain family member D1) encodes a calcium-binding protein involved in mitosis, synaptic transmission, and cytoskeletal rearrangement [67]. In the present systematic review, methylation of EFHD1 had 79% sensitivity and 22% specificity for CRC in plasma [32]. When combined with the analysis of PPP1R3C methylation, sensitivity was 53% and specificity reached 96% [32]. No patients with ns, not specified; MSP, methylation-specific PCR; qMSP, quantitative methylation-specific PCR; nMSP, nested methylation-specific PCR; pSEQ, pyrosequencing adenoma were included in this study. We have not found other studies on EFHD1 and its value as a potential biomarker for CRC is still unclear.
The ITGA4 gene encodes a membrane protein (integrin alpha 4) and is considered a risk marker for inflammationassociated colon cancer [68]. One study identified in the present systematic review found this marker to have 80% sensitivity and 100% specificity for CRC in stool, and methylated ITGA4 was found in 89% of CRC tissue samples and 88% of adenoma tissue samples [33]. In a stool only study, the sensitivity for adenoma was 29% and the specificity was 69% [69]. Another stool only study found 70% sensitivity and 97% specificity for CRC when investigating a panel combining ITGA4, SFRP2, and p16 [70].
The N-Myc downstream-regulated gene 4 (NDRG4) plays a role in cell growth and differentiation and is a putative tumor suppressor shown to be downregulated in colorectal cancer [71]. In the present systematic review, methylation of NDRG4 had 76% sensitivity for CRC in stool and 73% in urine [34,72]. No patients with adenomas and no healthy controls were included; hence, no sensitivity for adenoma and no specificity were reported [34,72]. A study on cellfree DNA in plasma found methylation of NDRG4 in only 18 of 193 patients with CRC (sensitivity 9%) but in none of the healthy controls (specificity 100%) [62]. In two sets of samples from different individuals, NDRG4 methylation analysis in stool found the sensitivity for CRC to be 53 to 61% and the specificity to be 93 to 100% [71]. Methylation was detected in 70 to 86% of CRC tissues as compared with 4% in noncancerous colon mucosa [71]. Despite NDRG4 being included in the Cologuard® test, there are only few studies addressing its role as a potential biomarker.
The oncostatin M receptor gene (OSMR) encodes a subunit of both the oncostatin M (OSM) receptor type II and the interleukin-31 receptor and transduce signals with pro-or anti-proliferative functions. OSMR promoter methylation has been identified in tissue and stool samples from patients with CRC. Two of the included studies confirmed that OSMR methylation is a common epigenetic event in colorectal cancer [35,38], but one study also revealed low concordance (48%) and specificity (33%) when comparing matched plasma and tumor tissue samples [38].
The protein phosphatase 1 regulatory subunit 3C (PPP1R3C) modulates glycogen metabolism, and methylation of the PPP1R3C gene has been proposed to play a critical role in colorectal cancer. One of the studies included in this review found that methylated PPP1R3C had 92% sensitivity for the detection of stage I CRC [32]. However, the sample size of this study was small and its value as a potential biomarker, especially for early-stage CRC cancer, needs to be further investigated.
SEPT9 encodes a GTP-binding protein involved in cell proliferation and migration, cytokinesis, and angiogenesis [1]. Methylation of SEPT9 is the most extensively studied ns, not specified; QuARTS, quantitative allele-specific real-time target and signal amplification; MSP, methylation-specific PCR; RRBS, reduced representation bisulfite sequencing; qMSP, quantitative methylation-specific PCR; MS-MCA, methylation-specific melting curve analysis single methylated marker for colorectal cancer. There are two commercialized assays already in clinical use as blood tests. These are ColoVantage® (sensitivity 90%) [73] and Epi proColon 2.0 (sensitivity 66-81% and specificity 96-99%) [74,75]. However, the sensitivity in the blood for advanced adenomas has been < 10% in several studies [76][77][78]. In the present systematic review, the most well-designed studies generally found a good correlation between the methylation status of SEPT9 in plasma and tissue for CRC patients and also for healthy controls [35,36,79,80]. In patients with adenomas the impression is that even though patients have methylated SEPT9 in tissue, only one-third of them have detectable methylated SEPT9 in plasma [37]. The reason for this discrepancy remains unknown. The secreted frizzled-related protein 1 and 2 genes (SFRP1 and SFRP2) encode antagonists of the Wnt signaling pathway, acting as tumor suppressors. Altered methylation of these genes has been reported in colorectal cancer tissue [81].
Aberrant promoter methylation of SFRP2 is associated with poor survival of colorectal cancer [39,82]. Two well-designed studies found the sensitivity of SFRP2 for CRC to be above 85% in stool and ranging from 46 to 62% for adenoma, the specificity was above 90%, and the findings corresponded well with the findings in tissue for the same patients [39,40]. In studies of only one type of material, Rasmussen et al. found that the sensitivity of SFRP2 for CRC to be 21% for CRC and the specificity was 82% in plasma [62]. In stool only studies, sensitivity for CRC was 63-94%, sensitivity for adenomas was 46-92%, and specificity was 32-100% [83][84][85][86][87]. As mentioned previously, the combination of SFRP2 with other markers, such as VIM, RASSF1A, TMEFF2, MGMT, ITGA4, and p16, has provided promising results [70,72,85,87,88].
The spastic paraplegia 20 gene (SPG20) encodes Spartin, a multifunctional protein found to be involved in intracellular epidermal growth factor receptor trafficking [89]. In the Table 3 Case-by-case relationship for single methylation markers in > 1 type of material from the same individual (other= blood (bold) or stool (italics) or urine (bold-italics))  [44] k, Cohen's kappa; SE, standard error of k present systematic review, we identified two studies investigating this biomarker in more than one type of material, although none of these studies included adenomas. Rezvani et al. found methylated SPG20 in 94% of CRC tissues and correspondingly in 81% of the plasma samples from these patients [41]. The specificity in tissue samples was 99% and in plasma samples 97% [41]. Zhang et al. found methylated SPG20 in 85% of CRC tissues and correspondingly in 80% of stool samples from these patients [42]. No tissue samples from healthy controls were investigated, but the specificity in the stool was 100% [42]. In a study carried out in plasma only, the sensitivity of SPG20 was 16% for CRC and the specificity was 88%, whereas the sensitivity for adenomas was not investigated [62]. The tissue factor pathway inhibitor 2 gene (TFPI2) encodes a serine protease inhibitor that decreases the activities of several enzymes, thereby protecting the extracellular matrix from degradation and inhibiting in vitro colony formation and proliferation [68]. Loss of TFPI2 function has been associated with pro-invasiveness and methylation of TFPI2 is considered an independent prognostic factor for CRC as it has been associated with later stages of carcinogenesis and advanced colorectal cancer [68,90]. TFPI2 has been demonstrated to be completely unmethylated in tissue from healthy controls, whereas methylation frequency increases with progression into inflamed colon tissue or CRC [68]. In the present review, two studies investigating TFPI2 methylation in stool and corresponding tissues found methylation of TFPI2 in CRC tissues in 89% and 99% of the samples, respectively, and the sensitivity in stool was 80% and 76%, respectively [33,43]. Specificity for CRC in stool was 100% and 93%, respectively [33,43]. Glöckner et al. also found methylated TFPI2 in three of the tissue samples from healthy controls (6%) and in 55 of 56 adenoma samples (98%) [43]. However, the sensitivity of TFPI2 methylation in adenoma stool samples was only 21% [43]. In a study of plasma, the CRC sensitivity of TFPI2 methylation was only 7%, no adenomas were included, and the specificity was 98% [62]. In a study of stool, the sensitivity for CRC was 68%, the specificity was 100%, whereas the sensitivity for adenoma was 35% [91,92].
VIM encodes a protein of the cytoskeleton. VIM is considered a biomarker of mesenchymal-derived cells and cancer cells undergoing epithelial-mesenchymal transition (EMT) during invasion and metastasis, and promoter methylation of VIM has been documented in CRC [1]. ColoSure® is a fecal single-marker test used in combination with colonoscopy. The test detects methylated VIM as a marker for CRC (sensitivi-ty38-88%) [1,93]. In paired samples from patients with CRC and/or adenomas, the share with detectable methylated VIM was 33% in stool, 4% in serum, and 8% in urine and the specificity was 100% in all materials [94]. In one study, where 83% of CRC cases had methylated VIM in tissue samples, 55% also had methylated VIM in stool [95]. In another study, where 44% of CRC cases had methylated VIM in tissue, 40% also had methylated VIM in stool [33]. Gerecke et al. also found that 72% of patients with adenomas had methylated VIM in adenoma tissue [33]. In Song et al.'s study, 85% of CRC cases had methylated VIM in tissue samples and 75% also had methylated VIM in urine [44]. When combining analysis of SFRP2 and VIM, methylation in at least one of these was found in 92% of CRC and 94% of adenoma tissues and correspondingly in 89% and 85% of stool samples [72]. VIM methylation is extensively studied in unpaired samples. The sensitivity for CRC detection by VIM methylation in the blood is 18-59% and the specificity is 63-93% [62,96,97]. The sensitivity for CRC in stool is 38-85%, and the sensitivity and specificity for adenoma are 75-90% and 82-100 % respectively [98][99][100][101][102][103].
One plasma panel consisting of the methylation markers APC, MGMT, RASSF2A, and WIF1 was identified by the present systematic review as promising, by reporting 86% sensitivity and 92% specificity for CRC [45]. Methylation in corresponding fresh-frozen tissue samples was found to be 27% for APC (specificity 97%), 39% for MGMT (specificity 96%), 58% for RASSF2A (specificity 100%), and 74% for WIF1 (specificity 98%) [45]. Methylation of APC leads to activation of growth-promoting genes, methylation of MGMT impairs the elimination of DNA alkylation adducts which may lead to mutation, methylation of the tumor suppressor RASSF2 alters the regulation of the cytoskeleton and apoptosis, and methylation of WIF1 results in reduced blocking of the Wnt signaling pathway [45].
We also identified two stool panels as the most promising panels tested in more than one type of material from the same individual. One was the panel consisting of methylated BMP3/NDRG4/VIM/TFPI2/mutant KRAS, β-actin, and Hblevel (and later refined and commercialized as Cologuard®), which in stool detected CRC with 87% sensitivity, adenoma with 82% sensitivity, and demonstrated a specificity of 93% [46,48,66]. Corresponding plasma samples were tested for methylated SEPT9 and the reported sensitivity for CRC was 60%, adenoma 14%, and the specificity was 73% [48]. The second stool panel identified was one consisting of RARB2, p16, MGMT, and APC. This panel detected CRC with 75% sensitivity, adenomas with 60% sensitivity, and the specificity was 100% [49]. To our knowledge, no other research groups have reproduced these findings and the findings have not led to a commercialized test.
Our overall impression is that the quality of the included studies was variable. With reference to QUADAS; many studies do not clearly classify the patients with colorectal cancer (some also include patients with adenomas and polyps), many have a sub-optimal or mixed "control group" (self-declared healthy controls or the control group including inflammatory bowel disease, colitis, polyps), many do not report blinding (i.e., it is not known whether the researcher who performed the methylation analysis knew whether the patient was a healthy control or a cancer patient), and many studies fail to report important findings ("not specified" in Supplementary Tables 1  and 2). To determine the strength of each methylated marker, it would be useful to aggregate the findings from single studies into a meta-analysis. Currently, this is not possible due to inconsistency in the reporting. To ensure comparability and future aggregation of findings, we suggest that future studies report their findings according to Table 4.
During the present review, we found that many of the studies claiming to report methylation in so-called paired samples do not analyze corresponding materials from the same individual. In addition, few of the studies reported the absolute numbers needed to calculate sensitivity and sensitivity. For the studies reporting the absolute numbers of methylation status in tissue and corresponding other materials from single individuals, we performed a case-by-case analysis ( Table 3). The correspondence of methylation status in different materials from the same individual was promising for CDH4, ERCC1, p16 INK4a , and SPG20. However, for most of the methylation markers and in most of the corresponding materials investigated (16 of 27), only a marginal reproducibility of methylation status was found. This inconsistency between "liquid biopsies" and tissue biopsies is confusing and needs to be addressed by analyzing methylation markers in several different materials from the same individual. And, this needs to be done both for patients with CRC, patients with adenoma, and healthy controls.
In the publications reviewed here, we rediscover some of the known technical challenges to the practical and clinical application of methylated cell-free DNA markers [58]. To date, there are still no standards agreed upon for analysis of methylated cell-free DNA and among the included publications, there is considerable variation regarding the methods used to extract and quantify cell-free DNA. The methods for sample processing and analysis also vary a lot, and many of the studies do not address or report upon pre-analytical and analytical issues. A study on optimization of the yield and utility of circulating cell-free DNA showed that time from blood sampling to plasma separation was extremely significant, whereas the temperature of the blood sample before plasma separation (kept on ice or at room temperature) did not affect the amount of cell-free DNA significantly [104]. One report has stated that DNA levels in plasma stored at − 80°C declines by 30% per year [105]. The cell-free DNA methylation pattern may be affected by demographic and lifestyle characteristics (i.e., age, race, gender, smoking, alcohol consumption), diet (i.e., folate, green tea, phytoestrogen), environmental exposures (i.e., arsenic, cadmium), and disease status [106]. Only a few of the studies take such factors into account when analyzing and reporting their results. The small sample sizes, selective and incomplete reporting, and lack of independent validation of promising biomarkers are factors hampering the implementation of potential biomarkers into clinical use [107]. The use of sensitive technologies, unbiased optimization, and standardization of commonly used methods is utterly important to enable validation of the findings from promising biomarker discovery studies [108].

Conclusion
The identified studies analyzing methylated DNA markers in paired samples from the same individuals generally included few patients, reported only some of the variables of interest and investigated only some of the markers of interest. Hence, a meta-analysis was not possible at this point. Larger, prospective studies need to be designed to overcome the challenge of discordant detection of markers in tissue and liquid biopsies. To improve comparability and to facilitate aggregation of the evidence base regarding possible DNA methylation markers Single marker Y n n n n Panel of markers XY n n n n N and n, absolute numbers; CI, 95% confidence interval with upper and lower limits *Specification of the type of other material analyzed (in addition to tissue). Examples: stool, plasma, serum, urine **Specification of the type of "control group." Examples: colonoscopy-verified healthy controls, self-declared healthy controls, age-matched colonoscopy-verified healthy controls, etc.
***Specification of the method used. Examples: MSP, methylation-specific PCR; qMSP, quantitative methylation-specific PCR; MA, microarray analysis; MS-MCA, methylation-specific melting curve analysis; MSRE-PCR, methylation-sensitive restriction enzyme and multiplex PCR; SEQ, sequencing; RRBS, reduced representation bisulfite sequencing of colorectal cancer, we suggest that this type of study use a standardized format when reporting their findings.
Code availability Not applicable.
Authors' contributions E.A.L., S.S.B., and F.S. collected, analyzed, and interpreted the data together. E.A.L. wrote the first draft of the manuscript. All authors revised the manuscript and approved the final version of it.
Funding Open Access funding provided by NTNU Norwegian University of Science and Technology (incl St. Olavs Hospital -Trondheim University Hospital).
Availability of data and material Not applicable.

Compliance with ethical standards
Conflict of interest The authors declare that they have no conflicts of interest.
Ethics approval Not applicable.
Consent to participate Not applicable.
Consent for publication All authors gave their consent to publication.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.