A scoping review of DNA methylation biomarkers for non-invasive detection of colorectal cancer at the CpG site level

Abstract

Colorectal cancer screening programs are well-established worldwide, but most applied methods are invasive, have insufficient accuracy, or have a low uptake rate. Several studies have reported DNA methylation biomarkers as promising alternatives to the established diagnostic tools. However, the current reviews on the topic lack information on the exact location of the CpG sites in the analyzed biomarker. As adjacent CpG sites may exert distinct clinical effects, precise identification of the specific CpG sites analyzed is essential for establishing clinical relevance and achieving consensus across studies. This scoping review aimed to uncover the accessibility of CpG site information in the scientific literature, map the CpG sites of the markers, and summarize the accuracy data from studies reporting on the same group of CpG sites. We systematically searched MEDLINE®, EMBASE, and Scopus, and the last search was conducted on the 13th of March 2025. A total of 6180 identified records were screened in Covidence, resulting in the inclusion of 149 articles. Four (2.7%) papers held precise CpG information. Locations of the CpG sites were obtained for additionally 109 (73.2%) papers by running a BLAST search using oligonucleotide sequences from the papers. For the remaining 36 (24.1%) papers without any information to locate the CpG sites, 26 used commercial kits. The three most frequently studied genes were SEPTIN9, SDC2, and SFRP2. We mapped the CpG sites of these to identify recurring sites across the studies. Sixteen of 22 papers analyzing SEPTIN9 had three CpG sites in common, seven of 19 papers analyzing SDC2 had consensus on a single CpG site, while seven of 13 studies reporting on SFRP2 had analyzed eight identical CpG sites. Lastly, we grouped all studies based on whether they targeted the same group of CpG sites within each gene. For genes examined in at least two papers, we summarized the reported accuracy measurements by presenting the average, minimum, and maximum values for each CpG group. This scoping review identifies a profound lack in the reporting of the CpG sites analyzed for non-invasive detection of CRC, which poses a challenge for the comparison of findings across studies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-026-02140-x.

Introduction

Colorectal cancer (CRC) is one of the few cancers with well-established screening programs worldwide. Screening programs reduce CRC mortality through early detection and treatment [1–3]. There are multiple methodologies for CRC screening, with detection of blood in stool samples using either fecal immunochemical tests (FIT) or guaiac fecal occult blood tests (gFOBT) being the most utilized [4]. A positive result of these tests would usually be followed up by a colonoscopy. The average sensitivity of FIT for CRC has been reported to be 93%, and the average specificity to be 91% [5], and for gFOBT, these values are 31% and 87%, respectively [6]. Due to the large populations screened and the suboptimal specificity for CRC for both tests, several patients receiving a positive result are false positives, and for gFOBT, the low sensitivity causes several CRC patients to be missed. Furthermore, compliance with this type of screening is highly variable between countries, causing suboptimal screening [7]. A study has found that 61% of the patients who did not accept the screening offer were willing to reconsider the screening if the fecal sampling procedure were replaced by a blood sample [8]. These challenges point out the need for more accurate and patient-friendly non-invasive screening alternatives.

New CRC screening methodologies are being developed, entailing the investigation of DNA, RNA, or protein in blood samples. Particularly, methylation of DNA has gained interest in recent years. DNA methylation is an epigenetic modification of the genome, where a methyl group is added to the cytosine base in CpG dinucleotides. It has been shown that hypermethylation of promoter regions in tumor-related genes is involved in the early development of CRC [9]. Several commercial tests are available for CRC screening using DNA methylation biomarkers [10]. One test (Epi proColon^®, Epigenomics AG, Berlin, Germany) has been FDA-approved. Epi proColon^® is used for analysis of the methylation status of septin 9 (SEPTIN9) in purified DNA. A meta-analysis from 2017 found a sensitivity of 71.1–95.6% and a specificity of 81.5–99.0% towards CRC in the second version of the test (Epi proColon^® 2.0) [11]. Other proposed DNA methylation biomarkers include SDC2, NDRG4, and VIM. These, among other biomarkers for CRC detection, have been accessed by analysis of various biopsy materials, including plasma, serum, urine, and stool [12, 13]. Analysis of these biopsies can identify circulating DNA from a tumor, as well as DNA specific to the biopsy, likely reflecting the more general biology of the individuals.

The location of methylated CpG sites in genes is essential information for the evaluation of the clinical value of a biomarker. Even neighboring CpG sites of a gene are affected differently by methylation in terms of gene expression regulation and clinical relevance. For instance, for the MLH1 gene used in the diagnosis of hereditary CRC, Lynch syndrome, it has been shown that methylation of some CpG sites has a stronger influence on gene expression than others, even though they are located within the same region of the gene [14]. Similarly, the prognostic value of MGMT promoter methylation in glioblastoma varies depending on which specific CpG sites are being analyzed, as different sites exhibit distinct associations with clinical outcomes [15].

At present, it remains unclear whether studies evaluating DNA methylation of the same genes for CRC detection target identical CpG sites, which represents a critical factor for comparing analytical accuracy and selecting the most suitable testing approach for clinical implementation. While several reviews on DNA methylation biomarkers for CRC detection exist, none summarize the evidence at a CpG site level [10, 12, 13, 16–21]. We conducted a scoping review to map the evidence on methylation biomarkers for non-invasive detection of CRC, with the objectives to (1) Calculate the proportion of studies reporting CpG sites. (2) Map the analyzed CpG sites on the genomic sequence from the most frequently studied genes. (3) Summarize sensitivity and specificity data from studies reporting on the same group of CpG sites.

Methods

Protocol

A review protocol was developed a priori in concordance with the Preferred Reporting Items for Systematic Reviews and Meta-analysis Extension for Scoping Reviews (PRISMA-ScR) [22]. The protocol can be found in Supplementary file S1.

Eligibility criteria

The eligibility criteria covered six topics: Report characteristics, study design, participants/populations, controls, exposure(s), and outcome(s). The inclusion and exclusion criteria can be found in Table 1.

Table 1.

Inclusion and exclusion criteria for the screening process

#	Topic	Inclusion criteria	Exclusion criteria
1	Report characteristics	Published studies in English Pre-prints Original papers, short reports, and research letters	Conference abstracts Reports without full text Editorials Letters to the editor Commentaries
2	Study design	Cohort studies Case-control studies Cross-sectional studies Diagnostic accuracy studies	Case-studies Reviews Cost-effectiveness studies Study protocols
3	Participants/ populations	Patients over the age of 18 diagnosed with CRC	“Population” of only cell lines Patients with CRC recurrence Only tumor tissue biopsy
4	Controls	Healthy controls/no evidence of disease	Cell lines or non-human reference material Normal tissue from the population defined in #3. Controls with cancer or precursors to cancer e.g., adenomas
5	Exposure(s)	DNA hyper-/hypomethylation of one or more biomarkers relevant for CRC detection Combined methylation analysis of more than one gene Pan-cancer methylation analysis with separate analysis for CRC Multi-target panels with DNA methylation data	Association studies and risk studies (e.g., association studies with EWAS or array data) and whole genome methylation analysis without analysis on specific CpG sites. E.g., exclude correlation studies of overall methylation level and CRC Predictive studies of treatment effects in CRC patients Prognostic studies of survival of CRC patients DNA hydroxymethylation
6	Outcome(s)	True positive, true negative, false positive, and false negative Sensitivity and specificity	Studies where the sensitivity and specificity for the methylation markers are combined with analysis of non-methylation biomarkers, and it is not possible to extract data from the methylation marker only Studies reporting accuracy measurements only for combined methylation biomarkers and not for individual genes

Information sources

We searched three electronic libraries: MEDLINE^® (covering from 1946 onwards) and EMBASE (covering from 1947 onwards) through the Ovid interface and Scopus through the Scopus interface. The searches were performed on the 9th of October 2023 and updated on the 13th of March 2025.

Search strategy

The search strings were made in collaboration with an experienced librarian from the University of Southern Denmark. The searches were performed by the first author. The following search string was used in MEDLINE: (colorectal neoplasms/ or colonic neoplasms/ or colorectal neoplasms, hereditary nonpolyposis/ or rectal neoplasms/ OR ((colon or colorectal or rectal or rectum or bowel or colo-rectal) adj3 (cancer* or tumor* or tumour* or neoplasm* or carcinoma* or malignanc*) or CRC).ti, ab, kf.) AND (DNA Methylation/ OR (DNA adj3 methylat*) OR DNA hypermethylation* OR DNA hypomethylation*) AND ((sensitiv: or diagnos: ).mp. or di.fs.) The search string was adapted to EMBASE and Scopus, respectively, and can be found in Supplementary file S1. We used the diagnostic accuracy filter from Kastner et al. [23]. No restrictions were put on the searches.

Selection of sources of evidence

Identified records were imported from the databases into the web-based systematic review software, Covidence (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia. Available at www.covidence.org), which automatically removed duplicates. The remaining duplicates were removed manually during the screening process by comparing administrative information and study characteristics.

The authors aligned their understanding of the eligibility criteria by pre-screening 50 records at the title/abstract level and revised the criteria and the study protocol in cases of unclarity. These 50 records were re-included in the actual screening and reassessed using the updated eligibility criteria. The change history is available in the protocol, and previous versions of the protocol can be accessed by contacting the corresponding author.

First, each record was screened by a single reviewer on the title/abstract level. In any case of doubt about eligibility, the record was moved to full-text screening. Next, full texts were retrieved. Most papers were retrieved using EndNote, while the remaining papers were identified manually using Google, PubMed, or identified via the University of Southern Denmark’s library. Finally, each full text was independently screened by two reviewers, with the first author screening all full texts and the other authors serving as the second screeners. Discrepancies were resolved through discussion. If excluded, a reason among eligibility criteria 1–6 from Table 1 was assigned.

Data charting

The first author and the last author developed a data charting form in Covidence and tested it by extracting data from five papers. The first author extracted all data items, and these were subsequently checked by the last author. Disagreement was resolved through discussion. Then, data from the remaining included articles were extracted using the form, and the extracted dataset was exported to an Excel spreadsheet, in which the data was cleansed. All gene names were checked in the National Center for Biotechnology Information’s (NCBI) Gene database, and the extracted gene name was changed to the official gene symbol provided by the HUGO Gene Nomenclature Committee (HGNC) if this was not already the listed name.

Data items

The following data items were extracted: (1) Study characteristics: First author, publication year, report characteristics, and study design. (2) Population: Country of study population, total number of individuals, number of cases and controls, and specimen type. (3) Methodologies: The laboratory method used to analyze DNA methylation, and the cut-off used for determining the samples positive for methylation (only for quantitative methods). (4) Outcome: Gene name, gene location, number of CpG sites, location of CpG sites, and diagnostic accuracy values.

Of note, in order not to limit the type of biospecimen types identified, the only criteria (see Table 1), which were made regarding biospecimen type were the exclusion criteria for the cases (’only tumor tissue biopsy’) and for the controls (‘normal tissue from the population defined in #3’).

Gene location, number of CpG sites, and location of the CpG sites were found by performing a BLAST search using the primer sequences in http://bisearch.enzim.hu/ (hereon referred to as Bisearch) if they were not directly accessible in the paper. The CpG sites could then be found using https://www.ensembl.org/ (Ensembl), which Bisearch links to. Furthermore, we grouped the studies based on the accessibility of CpG information (the scoring system was changed after the extraction) 0 = no CpG information, 1 = location of CpG sites can be accessed by a BLAST search using oligonucleotide sequences from a paper in the reference list, 2 = location of CpG sites can be accessed by a BLAST search using oligonucleotide sequences from main text/supplementary, 3 = location of CpG sites/cgID from Illumina array annotation is indicated in the main text/supplementary, but a BLAST search was needed to obtain exact location, 4 = exact location of all CpG sites/cgID reported in main text/supplementary (a BLAST search was not needed).

It was also noted how the accuracy values were identified, i.e., whether they were extracted from the paper or calculated by the reviewer. Lastly, it was noted whether a given study held both a training (exploration) study aiming at putting forward new biomarkers, as well as a test (replication) study validating biomarkers. In such cases, data from the test study were included, unless data from the individual biomarker were only available in the training study.

In most cases, the location of the CpG sites was identified from primer sequences. However, some were identified based on the technology applied. E.g., the CpG location was identified based on the primer sequences for studies using a methylation-specific PCR (MSP) assay and identified from both primer and probe sequences in studies using a MethyLight assay, while for melt-based studies, identification was dependent on the assay’s probe or dye characteristic. For studies using dye-based assays, all CpG sites in between the primers were reported, while for studies using probes, the CpG sites under the probe were used.

Synthesis of results

The Excel sheet with all extracted data was loaded into R Studio version 4.2.2, in which all data management was conducted. First, we calculated the percentage of studies that reported CpG sites according to objective 1 of the study (i.e., ‘Calculate the proportion of studies reporting CpG sites’, see Introduction). Second, we grouped the reports based on the analyzed DNA methylation biomarkers. In this review, we use the term ‘biomarker’ to denote examined genes. Some studies included biomarkers in several genomic regions and were, therefore, grouped into several groups. For the remaining analysis, we excluded all papers without CpG site information. Third, we grouped studies that had analyzed the same biomarker and the same group of CpG sites. Each unique CpG group was assigned a letter. We created an Excel workbook with a sheet for each biomarker consisting of selected data columns: CpG group, Biomarker, gene location, number of CpG sites, location of CpG sites, sense/antisense, sensitivity, specificity, cut-off, country of population, number of CRC cases, number of controls, biopsy type, technology, accessibility of CpG sites information, and reference. Fourth, for further analysis, we included the biomarkers that had been investigated in more than one study and created histograms on a CpG site level for the biomarkers in these studies. Fifth, we created a genomic map of the CpG sites of the three most frequently reported biomarkers according to objective 2 of the study (i.e., ’map the analyzed CpG sites on the genomic sequence from the most frequently studied genes’, see Introduction). Sixth, we calculated the average sensitivity and specificity, respectively, for each CpG group to investigate objective 3 of the study (i.e., ’summarize sensitivity and specificity data from studies reporting on the same group of CpG sites’, see Introduction).

Results

Selection of studies

The workflow for the selection of studies can be seen in Fig. 1. We performed two rounds of searches, the first on October 9, 2023, and an update on March 13, 2025. We identified 1816 records in MEDLINE, 2715 in EMBASE, and 1649 in Scopus, resulting in 6180 records to be included in the screening process. A total of 2032 duplicates were removed either automatically by Covidence or manually by the reviewers, leaving 4148 records for title/abstract screening. A total of 443 studies were passed to full-text screening, where 294 studies were excluded, mainly due to “wrong report characteristics”, e.g., when the record was a conference abstract. The excluded papers with exclusion criteria from the full text screening can be found in Supplementary file S2. Finally, 149 papers were included [24–172].

Fig. 1.

Workflow of the screening process. The figure is exported and modified from www.Covidence.org

Study characteristics

Of the 149 included reports, two were research letters [111, 170], while 147 were original scientific articles. Most studies (n = 144) applied a case-control study design, two studies used a cross-sectional design [99, 170], and three studies used a type of nested case-control design [49, 96, 123]. The investigated populations were mainly from China (n = 63, 42.3%), followed by Iran (n = 17, 11.4%). For 12 papers (8.1%), the population was unknown or unclear.

The studies analyzed between 4 and 428 CRC cases (mean: 92.1, median: 60) and between 4 and 1012 controls (mean: 118, median: 50). Fifty-four studies applied a quantitative MSP (qMSP) detection technology or MethyLight, and 26 studies applied a traditional MSP assay with gel detection, which all relies on a bisulfite conversion of the sample material. Eleven studies used methylation-sensitive restriction enzyme (MSRE) for the detection of DNA methylation, and 27 studies applied other techniques. Thirty-one studies applied commercial kits.

Most of the papers used samples of either stool (n = 61, 40.9%), plasma (n = 60, 40.3%), or serum (n = 14, 9.4%), while the remaining 14 studies (9.4%) reported other types of biopsies: Four peripheral blood mononuclear cells (PBMC) [33, 71, 110, 132], three whole blood [74, 100, 131], two buffy coat [36, 92], one urine [133], and one bowel lavage fluid (BLF) [119]. One study used both stool and plasma samples [93], while two studies used stool and serum [103, 141]. All data extracted from the papers can be found in Supplementary file S3.

CpG accessibility

The proportions of reports rated 0–4 are illustrated in Fig. 2A. Thirty-six papers (24.2%) received a rating of 0 as no information on CpG sites was accessible, neither through the publication or its references, nor through conducting a BLAST search. In most of these cases, it is likely that the authors did not report this information, because they had used a commercial assay, where information on CpG sites might not be available (26 of the 36 studies without CpG information applied commercial kits). Five publications (3.4%) received a rating of 1, as it was necessary to consult one of their references to obtain information on the primers, which could then be used in a BLAST search. Eighty-eight articles (59.1%) received a rating of 2, because information on primers and probes was available in the body of the text or supplementary materials, allowing us to obtain CpG site location by using this information in a BLAST search. A rating of 3 was assigned to 16 articles (10.7%), because they reported information in the main text or supplementary materials, allowing us to find the approximate locations of the CpG sites, e.g., stating the cgID for some of their CpG sites, writing the location of their amplicon, or in other ways indicating the location of the analyzed CpG sites without giving the precise location for all their sites. A BLAST search was conducted to get the precise location of all the CpG sites analyzed. Four papers (2.7%) were given the rating 4, because they reported the precise location of all CpG sites analyzed, and a BLAST search was not needed. Of these, three reported results from analysis of single CpG sites and mentioned the Illumina cgID [86, 99, 171], while one annotated the CpG sites’ location on the chromosome [38].

Fig. 2.

Biomarkers investigated in the included studies and their CpG information. A CpG information in the studies, 0 = no CpG information, 1 = location of CpG sites can be accessed by a BLAST search using sequences reported in paper from the reference list, 2 = location of CpG sites can be accessed by a BLAST search using sequences from the main text/supplementary, 3 = location of CpG sites/cgID is indicated in the main text/supplementary, but a BLAST search was needed to obtain exact location, 4 = exact location of all CpG sites/cgID reported in the main text/supplementary (a BLAST search was not needed). B Word cloud of all biomarkers investigated in the included reports. Word size is correlated with the frequency of the biomarkers. C The top 12 studied biomarkers grouped by CpG information accessibility yes/no (yes = group 1–4, no = group 0)

In total, we conducted a BLAST search using oligonucleotide sequences from 109 reports. However, for 11 papers, we failed to obtain a result from the BLAST search on one or more of the investigated biomarkers. All extracted oligonucleotide sequences pasted into BiSearch can be found in Supplementary file S4.

DNA methylation biomarkers for CRC

A total of 107 biomarkers were identified in the included reports. The investigated biomarkers and their relative frequencies are illustrated in the word cloud in Fig. 2B. The 12 most frequently analyzed biomarkers were (in descending order) SEPTIN9, SDC2, SFRP2, NDRG4, TFPI2, CDKN2A, VIM, SFRP1, MGMT, IKZF1, APC, and ALX4, where the latter five were equally frequently reported (Fig. 2C). For SEPTIN9, the CpG information was accessible for 22 of 43 papers. We used the official gene symbols from NCBI to make the gene names uniform, but for CYCD2, YHL, and HIC (not specified if it is HIC1 and HIC2) [106], it was not possible to find a match in the gene database. This was also the case with relation to MINT31 [92]. One paper investigated three ultra-conserved regions: Uc160, Uc283, and Uc346, which are not a part of the NCBI’s gene database [76].

Mapping of CpG sites

For the remaining analysis, we excluded all papers where information on CpG sites was not accessible in any way (rating = 0), leaving 113 papers for further investigation. An Excel workbook containing the extracted information on each of the 98 identified biomarkers with CpG site information can be found in Supplementary file S5. The data for SEPTIN9, SDC2, and SFRP2 are presented in Tables 2, 3 and 4. Forty biomarkers had been investigated in at least two independent studies. A histogram for the CpG site frequency in each of these 40 biomarkers can be found in Supplementary file S6. These graphs illustrate the locations of CpG sites and how frequently each CpG site had been investigated, along with references to the studies investigating them. The three most frequently analyzed biomarkers were SEPTIN9, SDC2, and SFRP2, and maps of their investigated CpG sites’ locations on the genomic sequence are illustrated in Figs. 3, 4 and 5. SEPTIN9 had the most frequently investigated CpG sites with the locations chr17:77,373,502; 77,373,509; 77,373,511 (GRCh38/hg38). Sixteen papers covered these locations (see Fig. 3). One study [63] investigated a region of SEPTIN9 located plus 28,752 base pairs (bp) away from the regions studied by the other SEPTIN9-grouped papers.

Table 2.

Details on SEPTIN9 CpG groups with accuracy data (chromosome 17)

CpG group	N CpG	Location of CpG sites	Strand	Sensitivity	Specificity	Cut-off	Country of population	N CRC	N Controls	Biopsy type	Technology	CpG info	Ref.
A	10	77,373,423	+	0.08	1	> 265.2606 relative methylation	Portugal	100	136	Plasma	qMSP probe	2	[50]
		77,373,425
		77,373,434
		77,373,502
		77,373,509
		77,373,511
		77,373,518
		77,373,520
		77,373,541
		77,373,548		0.1111	1	> 8.973 relative methylation level	Portugal	72	103	Plasma	qMSP probe	2	[113]
B	4	77,373,423	+	0.6667	0.8667	> 1.56 relative methylation level	Iran	30	15	Plasma	qMSP dye	2	[130]
		77,373,425
		77,373,434
		77,373,548
C	6	77,373,480	+	0.5	1		Commercial samples	20	20	Plasma	HeavyMethyl /MR-SNuPE	2	[143]
		77,373,502
		77,373,509
		77,373,511
		77,373,518
		77,373,520
D	3	77,373,502	+	0.72 (0.56)	0.86 (0.95)	1/3 replicates positive (2/3 replicates positive)	Unknown	90	155	Plasma	qMSP probe	2	[52]
				0.5	0.9	> 2.5% methylation	Brazil	38	43	Plasma	MS-ddPCR probe	3	[85]
				0.591	0.955	1/3 replicates positive	Australia, USA	44	44	Plasma	HeavyMethyl	3	[108]
				0.9	0.883	1/3 replicates positive Ct < 45	USA, Russia	50	94	Plasma	qMSP probe	2	[150]
		77,373,509		0.711	0.759	dCt ≤ 7.49	China	45	37	Plasma	qMSP probe	2	[154]
		77,373,511		0.516	1	dCt ≤ 4.02	China	31	19	Plasma	qMSP probe	2	[155]
E	5	77,373,502	+	0.7391	0.8		Vietnam	46	50	Plasma	semi-nested MS-HRM dye	2	[54]
		77,373,509
		77,373,511
		77,373,518
		77,373,520
F	6	77,373,502	-	0.6122	0.9842	2/3 replicates positive	China	98	253	Plasma	Epi proColon 2.0	1	[56]
		77,373,509		0.58	0.9	2/3 replicates positive	Unknown	126	183	Plasma	Heavy Methyl	2	[59]
		77,373,511		0.52	0.95	> 0.011 mg/L methylated DNA	Unknown	133	179	Plasma	qMSP probe	2	[95]
		77,373,518		0.73	0.945	Ct ≤ 41	China	63	494	Plasma	Epi proColon 2.0	2	[139]
		77,373,520		0.40 (0.40)	0.955 (0.955)	2/3 replicates positive (1/3 replicates positive)	Germany	5	22	Plasma	Heavy MethyLight and MethyLight	2	[142]
		77,373,541		0.766	0.959	Ct < 41	China	291	295	Plasma	qMSP probe	2	[151]
G	2	77,373,518	+	0.9	0.658		Russia	38	30	Plasma	qDMA-HP	4	[38]
		77,373,520
H	1	77,373,541	+	0.967	0.526		Russia	38	30	Plasma	qDMA-HP	4	[38]
I	8	77,402,301	+	0.7473	0.9647	≥ 4% methylation	China	182	170	Plasma	MethyLight PCR	2	[63]
		77,402,326
		77,402,357
		77,402,362
		77,402,367
		77,402,372
		77,402,387
		77,402,407

*Two studies [84, 87] did contain CpG information, but we did not obtain a result the BLAST search. qMSP: quantitative methylation-specific PCR, MR-SNuPE: methylation-restricted single nucleotide primer extension, MS-ddPCR: methylation-specific droplet digital PCR, MS-HRM: methylation-sensitive High-Resolution Melt, qDMA-HP: quantitative DNA Melting Analysis with hybridization Probes

Table 3.

Details on SDC2 CpG groups with accuracy data (chromosome 8)

CpG group	N CpG	Location of CpG sites	Strand	Sensitivity	Specificity	Cut-off	Country of population	N CRC	N Controls	Biopsy type	Technology	CpG info	Ref.
A	7	96,493,747	Not BS	0.8154	0.6923	> 20% methylation	Iran	65	65	Whole blood	MethyQESD	2	[131]
		96,493,752
		96,493,769
		96,493,802
		96,493,836
		96,493,845
		96,493,847
B	10	96,493,775	Not BS	0.6444	0.9603	Unclear	China	180	629	Stool	MSRE qMSP probe	2	[87]
		96,493,777
		96,493,800
		96,493,802
		96,493,834
		96,493,836
		96,493,838
		96,493,843
		96,493,845
		96,493,847
C	5	96,493,937	+	0.4194	0.8235	Ct < 40	China	75	171	Plasma	qMSP probe	2	[84]
		96,493,950
		96,493,952
		96,494,018
		96,494,023
D	7	96,494,023	+	0.872	1	Unclear	Unknown	47	37	Plasma	Nested MethyLight PCR	3	[34]
		96,494,094
		96,494,109
		96,494,113
		96,494,151
		96,494,157
		96,494,162
E	8	96,494,053	+	0.9	0.909	Ct < 40	South Korea	50	22	Stool	LTE-qMSP	3	[114]
		96,494,064
		96,494,069
		96,494,077
		96,494,081
		96,494,132
		96,494,140
		96,494,151		0.8	0.889	Ct < 40 and 1/2 replicates positive	South Korea	10	54	BLF	Nested qMSP probe	2	[119]
F	4	96,494,109	+	0.739	0.916	Ct < 39	China	138	28	Stool	qMSP probe	2	[97]
		96,494,113
		96,494,119
		96,494,126		0.8114	0.9524	Ct < 39	China	403	210	Stool	qMSP probe	2	[163]
G	9	96,494,109	+	0.811	0.933	Unclear	China	196	179	Stool	qMSP	2	[112]
		96,494,113
		96,494,119
		96,494,126
		96,494,132
		96,494,140
		96,494,151
		96,494,157
		96,494,162
H	7	96,494,109	+	0.8053	0.9905	Ct < 28.24	China	113	105	Stool	qMSP dye	2	[158]
		96,494,113
		96,494,119
		96,494,140
		96,494,151
		96,494,157
		96,494,162
I	8	96,494,140	+	0.7156	1	Ct < 38	China	320	300	Stool	iColocomf	2	[149]
		96,494,151
		96,494,157
		96,494,162
		96,494,220
		96,494,316
		96,494,321
		96,494,324		0.77	0.981	Ct < 38	China	61	53	Stool	qMSP probe	2	[165]
J	8	96,494,175	-	0.819	0.99	Unclear	China	105	100	Stool	qMSP probe	2	[80]
		96,494,181
		96,494,187
		96,494,189
		96,494,196
		96,494,220
		96,494,243
		96,494,257
K	1	96,494,187	Not BS	0.912	0.957	Unclear	China	205	161	Stool	mSTEM-PCR	4	[171]
L	6	96,494,621	-	0.591	0.841	1/3 replicates positive	Australia, USA	44	44	Plasma	qMSP probe + dye	3	[108]
		96,494,624
		96,494,636
		96,494,697
		96,494,705
		96,494,716

*Three studies [102, 154, 155] did contain CpG information, but we did not obtain a result from the BLAST search. Not BS: Not bisulfite converted DNA. MethyQESD: methylation quantification endonuclease-resistant DNA, MSRE: Methylation sensitive restriction enzyme, qMSP: quantitative methylation-specific PCR, LTE-qMSP: quantitative methylation-specific PCR coupled with linear target enrichment, BLF: bowel lavage fluid, mSTEM-PCR: methylation specific terminal-mediated polymerase chain reaction

Table 4.

Details on SFRP2 CpG groups with accuracy data (chromosome 4)

CpG group	N CpG	Location of CpG sites	Strand	Sensitivity	Specificity	Cut-off	Country of population	N CRC	N Controls	Biopsy type	Technology	CpG info	Ref.
A	1	153,781,309	Not BS	0.859	0.957	Unclear	China	205	161	Stool	mSTEM-PCR	4	[171]
B	8	153,788,918	-	0.6	0.92		Iran	25	25	Stool	MSP	2	[31]
		153,788,927
		153,788,936		0.6	1		Korea	30	31	Stool	MSP	2	[42]
		153,788,941		0.942	0.952		China	52	24	Stool	MSP	2	[67]
		153,789,029		0.942	0.958		China	52	24	Stool	MSP	2	[68]
		153,789,031		0.84 (0.669)	0.933 (1.00)		China	169	30	Stool (serum)	MSP	3	[141]
		153,789,039		0.87	0.933	Unclear	China	69	30	Stool	MethyLight PCR dye	2	[147]
		153,789,049		0.563	1		China	48	30	Stool	MSP	2	[161]
C	4	153,788,936	-	0.611	0.863		Austria	18	22	Stool	MSRH probe	2	[77]
		153,788,941
		153,788,964
		153,788,967
D	7	153,788,976	Not BS	0.5889	0.9682	Unclear	China	180	629	Stool	MSRE qMSP probe	2	[87]
		153,788,978
		153,789,005
		153,789,007
		153,789,027
		153,789,029
		153,789,031
E	10	153,789,266	+	0.638	0.973	Unclear	Unknown	47	37	Plasma	Nested MethyLight PCR	3	[34]
		153,789,269
		153,789,273
		153,789,277
		153,789,347
		153,789,349
		153,789,356
		153,789,371
		153,789,383
		153,789,385
F	4	153,789,674	-	0.5714	0.9		China	56	40	Stool	MSP	2	[98]
		153,789,692
		153,789,809
		153,789,820		0.5439	0.7234		China	57	47	Plasma	MSP	2	[166]

Not BS: mSTEM-PCR: methylation specific terminal-mediated polymerase chain reaction, MSP: methylation-specific PCR, MSRH: methylation-specific reverse hybridization

Fig. 3.

Map of the analyzed CpG sites in SEPTIN9 on chromosome 17. The first part shows chr17:77,373,423 − 77,373,549, and the last part shows chr17:77,402,301 − 77,402,408. All locations are according to GRCh38/hg38. CpG sites are marked in bold, and the CpG sites analyzed in the individual studies are marked by a color bar, with the studies analyzing the same group of CpG sites marked with the same color

Fig. 4.

Map of the analyzed CpG sites in SDC2 on chromosome 8. The first part shows location chr8:96,493,747 − 96,494,325, and the last part shows chr8:96,494,621 − 96,494,717. All locations are according to GRCh38/hg38. CpG sites are marked in bold, and the CpG sites analyzed in the individual studies are marked by a color bar, with the studies analyzing the same group of CpG sites marked with the same color

Fig. 5.

Map of the analyzed CpG sites in SFRP2 on chromosome 4. The first part shows chr4:153,781,309 − 153,781,310, the second is chr4:153,788,918 − 153,789,050, the third is chr4:153,789,266 − 153,789,386, and the last part shows chr4:15,789,674 − 153,789,821. All locations are according to GRCh38/hg38. CpG sites are marked in bold, and the CpG sites analyzed in the individual studies are marked by a color bar, with the studies analyzing the same group of CpG sites marked with the same color

CpG sites in the SDC2 biomarker were also frequently studied. Seven studies included an investigation on CpG site chr8:96,494,151, while six papers reported on chr8:96,494,140 (Fig. 4). The analyses regarded CpG sites within a region of 969 bp (chr8:96,493,747 − 96,494,716).

CpG sites in the SFRP2 biomarker were similarly frequently analyzed. Four CpG sites, chr4:153,788,936; 153,788,941; 153,789,029; 153,789,031, were each investigated in eight reports. However, only seven papers included all four sites. Furthermore, these reports had an overlap in additionally four CpG sites and were thereby analyzing the same group of eight CpG sites. All papers reported on CpG sites in a region of 902 bp (chr4:153,788,918 − 153,789,820), except one paper [171] analyzing one CpG site minus 7608 bp from the other papers. The mapped CpG sites of SFRP2 are shown in Fig. 5.

Overlap was also found concerning NDRG4. Five of eight papers included the CpG sites chr16:58,463,534 and 58,463,536. All six papers analyzing CpG sites in CDKN2A included chr9:21,974,757; 21,974,763; 21,974,768; 21,974,770; 21,974,774 in their analyses. For TFPI2, three of six reports covered the same CpG sites located at chr7:93,890,700; 93,890,712; 93,890,716; 93,890,762; 93,890,767; 93,890,775, and all analyzed CpG sites were in the region chr7:93,890,158 − 93,890,871 (713 bp). Four of six papers included the same CpG sites in the VIM gene, and one paper [98] gave a result from the BLAST search on chromosome 6, but the VIM gene is located on chromosome 10. For ALX4, three of the papers investigated CpG sites in the region chr11:44,304,860 − 44,304,966, and one paper [63] analyzed a region + 4,042 bp from the other papers. All four papers investigating IKZF1 used CpG sites in a region spanning 92 bp (chr7:50,304,273 − 50,304,365), and these papers had five sites in common. For MGMT, five CpG sites were analyzed in the same three papers, although a total of 32 sites were investigated by the five papers reporting CpG information on this gene.

Summary of the accuracy data

The accuracy data for the 40 biomarkers investigated in at least two papers are summarized in Supplementary File S7, where the studies analyzing the same group of CpG sites are grouped under the same letter, and average, minimum, and maximum sensitivity and specificity values are presented for each CpG group. Of the 40 biomarkers, 17 held at least one CpG group, which had been investigated in at least two studies, while for the remaining 23 biomarkers, the CpG groups investigated had only been examined in a single study.

Sensitivities of CpG site groups in SEPTIN9 ranged from 0.08 to 0.97, with specificities ranging from 0.53 to 1.00 (Table 2 and Supplementary S7). Low sensitivity (0.08 and 0.11) but high specificity (1.00) was observed for CpG group A. The most frequently studied CpG groups, D and F had an average sensitivity of 0.64 (range: 0.50–0.90) and 0.49 (range: 0.40–0.77), respectively, with the average specificity being 0.90 (range: 0.76-1.00) and 0.95 (range: 0.90–0.98), respectively. A sensitivity of 0.74 and a specificity of 0.96 were found by [63] (group I), which analyzed a different region of the SEPTIN9 gene compared to the other studies. SEPTIN9 studies utilized plasma as the biopsy and variations of qPCR technologies on bisulfite-treated DNA. The highest sensitivity (0.97) was found in CpG group H, consisting of a single CpG site analyzed by one study [38], but this study also found the lowest specificity (0.53).

Sensitivities for groups of CpG sites in SDC2 ranged from 0.42 to 0.91, and specificities ranged from 0.69 to 1.00 (Table 3 and Supplementary S7). The studies used various biopsies for the analyses, covering stool, plasma, whole blood, and BLF, but the majority used stool. The highest sensitivity (0.91) was found for CpG group K, in which a single CpG site was investigated by one study [171]. This study also found a high specificity of 0.96.

For SFRP2, the sensitivity range was 0.54–0.94, with specificity values ranging from 0.72 to 1.00 (Table 4 and Supplementary S7). CpG group B was the most studied group with an average sensitivity of 0.75 (range: 0.56–0.94) and an average specificity of 0.96 (range: 0.92-1.00), which shows that this group of CpG sites has a high specificity but variance in sensitivity. All the studies on group B used stool as the biopsy, except for [141], which also included serum in their analysis. Further, the studies used the same technology (MSP), except for [147], which used MethyLight PCR. The highest average sensitivity (0.86) was found in CpG group A, which only contained a single CpG site that had been investigated in one study [171]. This study also found a high specificity of 0.96.

Discussion

This scoping review gives a broad overview of the existing evidence on DNA methylation biomarkers for CRC detection. Among 149 included reports, less than 3% reported the exact information on the CpG site location of the studied biomarker. A total of 73% provided information allowing the identification of the genomic position of the CpG sites by BLAST analysis. The three most frequently studied biomarkers were SEPTIN9, SDC2, and SFRP2, and we mapped the CpG sites on the corresponding chromosomes. Interestingly, the highest average sensitivities for a CpG group for SEPTIN9, SDC2, and SFRP2 (0.97, 0.91, and 0.86, respectively) were related to the study of a single CpG site. The corresponding specificities were 0.53, 0.96, and 0.96, respectively. However, these findings were only concluded based on single studies. In general, our summary showed large variation in sensitivity and specificity values across studies investigating the same group of CpG sites.

It is a concerning finding that the specific CpG sites are not reported directly for most of the studies. This challenges the comparison of findings from research apparently investigating the same biomarker. A similar lack of precision in reporting would, to the best of our experience, not be accepted for studies of genetic variation, where a given genetic variant is defined by its base pair position. Moreover, some genes have several alias symbols, which adds another layer to the challenge. We found that some genetic annotations could not be identified in NCBI’s gene database, which stresses the importance of reporting the specific location of the analyzed CpG sites. A review focused on the general reporting quality according to the Standards for Reporting Diagnostic accuracy studies (STARD) guidelines and concluded that there was a lack of information when reporting the findings for the CRC methylation biomarkers [173]. However, the review did not cover the quality of the reporting of the investigated CpG sites. With this scoping review, we are adding another area where improvement is needed in the reporting quality of studies investigating DNA methylation biomarkers. For most of the identified articles in our review, information was available to conduct a successful BLAST search of primers and/or probe sequences in BiSearch. However, this procedure requires sufficient technical understanding of the underlying laboratory technology. For instance, when making a BLAST search using bisulfite-converted primers, they must be longer than 15 bp. When shorter chains are reported, several nucleotides must be assumed, which is time-consuming.

We found that the biomarkers SEPTIN9, SDC2, and SRFP2 were most frequently studied, similar to a systematic review from 2022, concluding these three markers to be the most frequently investigated in non-tissue biopsies [173]. Additionally, we observed that SEPTIN9 and SDC2 are utilized in commercially available screening tests. Currently, several commercial kits include SEPTIN9 as a biomarker, i.e., Epi proColon and Epi proColon 2.0 CE (Epigenomics AG, Berlin, Germany), the ColoDefense test (Suzhou Versa-Bio Technologies Co. Ltd.), the Septin 9 Gene Methylation Detection Kit (Beijing BioChain co., ltd., China), and the SensiColon (BioChain (Beijing) Science and Technology, Inc.). In addition to SEPTIN9, the ColoDefense test also tests for SDC2. Further, SDC2 was included as a biomarker in the ColoCaller test (Apexbio, Suzhou, China), the SpecColon test (Suzhou VersaBio Technologies Co. Ltd.), iColocomf (Wuhan Ammunition Life-Tech Company, Ltd.), and a methylation detection kit for human SDC2 gene (Creative Biosciences Co. Ltd., Guangzhou, China).

We found that the reported sensitivity values of SEPTIN9 ranged from 0.08 to 0.97, with specificities ranging from 0.53 to 1.00. A meta-analysis from 2020 found the pooled sensitivity of SEPTIN9 to be 0.69 and the pooled specificity to be 0.92 [174]. However, these pooled accuracy measurements do not take the location of CpG sites into account. We observed high variation in sensitivity values for all summarized CpG site groups across the most frequently reported biomarkers. Specificity values were also highly variable for CpG site groups in the biomarkers SEPTIN9 and ALX4, and lower variation was seen in specificity values for CpG site groups of TFPI2 and CDKN2A, which were all above 0.90. The variation could be due to differences in detection technologies, biopsy types, and cutoffs. Four studies [67, 68, 141, 161] investigated the same CpG sites in SFRP2 utilizing stool from a Chinese population and applying MSP as technology. Despite these similarities, high variation in sensitivity values persisted, ranging from 0.56 to 0.94, which indicates that additional factors influence the sensitivity of the biomarkers. An influencing factor could be the stage of CRC, which has been shown to affect sensitivity [175]. The technical handling of the biological samples before methylation analysis has also been shown to influence the methylation degree [176, 177]. Overall, the identified accuracy values reflect a lack of consensus when considering the specific CpG site locations and strongly reflect the need for future studies to include this information.

In the present study, we only included papers written in English, and in total, we searched three databases for articles, which are both limitations to this scoping review, as additional relevant literature might exist. A single-screening approach was used for the title and abstract screening due to time constraints and a limited number of reviewers, which may have led to the omission of relevant studies. However, the reviewers adopted an inclusive strategy by advancing any article to full-text screening in cases of uncertainty, and before the actual screening, all reviewers participated in a pilot screening to align their understanding of the eligibility criteria. During the full-text phase, all studies were independently assessed by two reviewers. Another limitation is that we did not make a systematic bias or quality assessment of the included papers, as our goal was simply to give an overview of the literature with a focus on CpG position, an approach that has not been taken before. We did not extract information regarding the stage of disease, which would have been beneficial to identify the CpG sites with the highest clinical relevance for the non-invasive detection of early-stage CRC. Nevertheless, considering the limited number of studies identified per biomarker in the present study (i.e., Tables 2, 3 and 4), to stratify per disease stage would potentially have led to even fewer studies per biomarker, as not all studies included information on disease stage groups. This hence points to the need for proper CpG reporting in order to be able to include all published studies in such assessment. Lastly, all CpG sites identified from the BLAST searches were manually extracted, which potentially could lead to typing errors. It might have been beneficial to investigate whether this process could have been automated to a larger degree. However, we do believe that the overall findings of this scoping review, i.e., the lack of detailed information in the published papers, are not affected by these limitations. Future studies including such relevant information will enable a more meaningful assessment of bias and will be less biased towards the BLAST methods.

Conclusions

This scoping review summarizes for the first time the evidence on the accuracy of specific CpG sites of biomarkers investigated in the detection of CRC. It highlights the importance of reporting CpG sites in research investigating the potential of DNA methylation. Further research focusing on specific CpG sites is needed to investigate the clinical relevance of each biomarker and its CpG sites.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BLF

bowel lavage fluid

BS

bisulfite converted

BS-pyrosequencing

bisulfite pyrosequencing

CRC

Colorectal cancer

ddPCR

droplet digital PCR

DREAMing

digital restriction enzyme analysis of methylation

FIT

fecal immunochemical tests

gFOBT

guaiac fecal occult blood tests

HGNC

HUGO Gene Nomenclature Committee

IP-MSP

immunoprecipitation with anti-histone antibody methylation-specific PCR

LTE-qMSP

quantitative methylation-specific PCR coupled with linear target enrichment

MethyQESD

methylation-quantification of endonuclease-resistant DNA

MR-SNuPE

methylation-restricted single nucleotide primer extension

MS-ddPCR

methylation-specific droplet digital PCR

MS-HRM

methylation-sensitive High-Resolution Melt

MS-MCA

methylation-specific melting curve analysis

MSP

methylation-specific PCR

MSP-SSCP

methylation-specific polymerase chain reaction (PCR)-single-strand conformation polymorphism

MSRE

methylation-sensitive restriction enzyme

MSRH

methylation-specific reverse hybridization

mSTEM-PCR

methylation specific terminal-mediated polymerase chain reaction

NCBI

National Center for Biotechnology Information

PBMC

peripheral blood mononuclear cell

PMR

percentage of methylated reference

PRISMA-ScR

Systematic Reviews and Meta-analysis Extension for Scoping Reviews

qDMA-HP

quantitative DNA Melting Analysis with hybridization Probes

qMSP

quantitative methylation-specific PCR

STARD

Standards for Reporting Diagnostic accuracy studies

Author contributions

K.B. has drafted the manuscript with guidance from M.S. M.S. is the guarantor of the overall scientific quality of the manuscript. All authors except R.P. were involved in the screening process. W.P. participated in the title/abstract screening, L.P. participated in the title/abstract screening and in the full-text screening. M.S. and K.B. were involved in all stages of the screening and extraction. R.P. contributed to the conceptualization of the scoping review. All authors have read the protocol and manuscript, provided feedback, and approved the final version.

Funding

K.B. was employed as an industrial PhD student by PentaBase A/S in the period of conducting the review and enrolled as an industrial PhD student at the University of Southern Denmark (SDU). The PhD was funded by the Innovation Fund Denmark’s Industrial PhD program [grant number 1044-00027B]. M.S. is the university supervisor of K.B. K.B.’s salary was covered by PentaBase A/S, and M.S. and W.P.’s salaries were covered by the University of Southern Denmark.

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

K.B. and R.P. were employed at PentaBase A/S during the period when this review was conducted. PentaBase produces and sells products for in-vitro cancer diagnostics, including products used for the analysis of DNA methylation. R.P. is a co-owner of PentaBase A/S. M.S., L.G., and W.P. declare that they have no known conflicts of interest.