Background

Most colorectal cancers (CRCs) occur sporadically and progress through sequential accumulation of multiple genetic and epigenetic alterations by influencing the expression and behaviour of genes that regulate cell growth and differentiation [1,2,3]. Several crucial gene defects in sporadic CRC have been identified, and specific molecular phenotypes have been described, including chromosomal instability (CIN), microsatellite instability (MSI), and the CpG island methylator phenotype (CIMP) [1, 3,4,5]. Although somatic mutations in the POLE gene are found in 3–7% of CRCs, mutations within the proofreading (exonuclease) domain of POLE are present in only 1–2% of CRCs [6,7,8,9]. The proofreading potential of POLE is essential for ensuring replication fidelity, and its disruption by the pathogenic heterozygous mutations found in cancers leads to an ultra-hypermutated phenotype of tumours, with the highest burden of single-nucleotide variants among human cancers [6, 10, 11]. Analogous to CRC, patients with endometrial cancers harbouring pathogenic POLE exonuclease domain mutations have excellent prognosis [6, 7, 10, 12], possibly because such an extreme hypermutation event causes the enrichment of antigenic neoepitopes, which in turn stimulates a potent cytotoxic T-cell response in cancer cells [9, 13, 14].

To date, several pivotal studies have characterised POLE-mutant CRCs with their corresponding MSI status, genetic mutations, tumour lymphocyte infiltration, and clinicopathological findings, including clinical outcomes [6,7,8, 15]. However, to the best of our knowledge, no study has thus far characterised POLE-mutant CRCs in the context of epigenetic alterations. The present study aimed to evaluate the epigenetic profiles of POLE-mutant CRC and elucidate the clinicopathological features associated with key genetic and epigenetic alterations in this malignancy.

Results

Detection of pathogenic somatic POLE mutations

Of the 1,052 CRC patients, 17 had resected synchronous multiple cancers (one patient had three synchronous tumours and 16 patients had double synchronous tumours). Therefore, a total of 1,070 CRC tissues were analysed for POLE mutations. None of the patients with synchronous multiple cancers displayed POLE mutations in their cancer tissues; thus, patients with synchronous multiple cancers were analysed by the most advanced tumour lesion for further molecular studies (Fig. 1). Pathogenic proofreading POLE mutations were detected in four CRCs as recurrent variants and are known to cause an ultra-mutator phenotype and characteristic mutation spectrum (NM_006231.3: two cases were p.P286R [c.857C > G], one was p.V411L [c.1231G > C], and one was p.S459F [c.1376C > T], Fig. 2). Regarding germline mutations, the DNA of four patients was sequenced to determine whether the POLE mutations existed in the DNA extracted from their corresponding normal mucosa; however, no germline mutations were present.

Fig. 1
figure 1

The STROBE diagram of the colorectal cancer patient cohort. *Patients with synchronous multiple cancers based on the most advanced tumour lesion were subsequently confirmed pathologically for further molecular studies

Full size image
Fig. 2
figure 2

Detection of POLE proofreading mutations. Among 1,070 CRC tissues analysed, pathogenic somatic POLE proofreading mutations were detected in four CRCs (two cases were p.P286R [c.857C > G], one was p.V411L [c.1231G > C], and one was p.S459F [c.1376C > T]) and DNA from their corresponding normal mucosa displayed no germline mutations

Full size image

Evaluation of MSI status

The MSI status of the 1,052 CRC tissues was evaluated. None of the four tumours harbouring pathogenic POLE mutations displayed MSI features in the four mononucleotide repeat markers [12, 16]. Among the remaining 1,048 CRCs, 6.4% (67/1,048) displayed MSI features, while 93.3% (981/1,048) did not exhibit any MSI features and were deemed microsatellite stable (non-MSI) (Fig. 1).

Immunohistochemical (IHC) analysis of MMR expression in CRCs with MSI

The expression status of the four DNA mismatch repair (MMR) proteins (MLH1, MSH2, PMS2, and MSH6) was confirmed in all 67 MSI CRC tissues. By IHC analysis, 11.9% (8/67) were classified as MMR-proficient (pMMR), and the remaining 86.8% of MSI CRCs (59/67) were confirmed to be MMR-deficient (dMMR). Of the 59 dMMR tumours, 79.7% (47/59) exhibited both MLH1 and PMS2 deficiency (dMLH1), 15.3% (9/59) exhibited both MSH2 and MSH6-deficiency (dMSH2), 5.1% (3/59) exhibited MSH6-deficiency alone (dMSH6), and none of the tumours displayed PMS2-deficiency on their own (dPMS2). The precise status of IHC staining and MSI markers in MSI tumours is shown in Additional file 1: Table 1.

Classification of four CRC subtypes according to POLE mutations, MSI, and methylation status in the MLH1 promoter region

Sporadic MSI tumours are primarily caused by inactivation of the MLH1 gene, which has a large CpG island within its promoter region that divides it into at least two discrete regions of methylation (the AB and C regions in this study, Fig. 3a) [17,18,19,20]. Inactivation of MLH1 was observed when CpG methylation spread through the AB to the C region.

Fig. 3
figure 3

Methylation analysis of the promoter region in the MLH1 gene. a Schematic depiction of two regions (AB and C region) of the MLH1 promoter for methylation, and results of a panel of representative fluorescent bisulphite PCR following restriction enzyme analysis. AB and C region in this study link to A plus B and C plus D region, defined by Deng and colleagues [18], respectivily.  Methylated samples had the new fragment cleaved by the restriction enzyme (black triangles). White triangles represent unmethylated alleles. M and U denote methylated and unmethylated, respectively. A grey bar denotes an untranslated exon. b The frequencies of MLH1 promoter methylation status among the POLE, MSI, and non-MSI groups. Extensive methylation is defined as 5.0% or more methylation on both AB and C regions. Partial methylation is defined as 5.0% or more methylation on AB region only. c The frequencies of MLH1 promoter methylation status according to MMR protein expression status. dMLH1, dMSH2, and dMSH6 denote deficiency of MLH1, MSH2, and MSH6, respectively. pMMR denotes proficiency of all MMR proteins. d The final classification of 1,013 CRCs according to POLE mutation, MSI, and MLH1 promoter methylation status. MSI-M denotes MSI tumours with extensive MLH1 methylation. MSI-U denotes MSI tumours without extensive MLH1 methylation. The number in each column denotes the number of samples classified in each category divided by MLH1 methylation status

Full size image

In this cohort, no methylation in the AB or C region (unmethylated) was observed in 50.0% (2/4) of POLE-mutant CRCs, 38.5% (25/65) of MSI CRCs, and 85.1% (803/944) of non-MSI CRCs. Partial methylation in MLH1 (i.e. affecting the AB region only) was observed in 50.0% (2/4) of POLE-mutant CRCs, 16.9% (11/65) of MSI CRCs, and 14.9% (141/944) of non-MSI CRCs. Extensive methylation (i.e. affecting both the AB and C regions) was observed in 44.6% (29/65) of MSI CRCs (P < 0.0001, Fig. 3b) and 64.4% (29/45) of dMLH1 tumours, which were suspected to be sporadic MSI tumours. In contrast, no extensive methylation was observed in either dMSH2 or dMSH6 cancers, including in patients with Lynch syndrome (Fig. 3c). Therefore, in this study, we classified MSI tumours according to their methylation status in the MLH1 promoter; 2.9% (29/1,013) tumours with extensive MLH1 methylation were categorised as MSI with MLH1 methylation (MSI-M) and 3.6% (36/1,013) tumours without extensive MLH1 methylation as MSI with unmethylated MLH1 (MSI-U).

Finally, we classified 1,013 CRC patients into four subgroups: non-MSI CRC patients with POLE mutations (POLE group, 0.4%, n = 4), MSI CRC patients with extensive MLH1 methylation (MSI-M group, 2.9%, n = 29), MSI CRC patients with unmethylated MLH1 (MSI-U group, 3.6%, N = 36), and non-MSI CRC patients (non-MSI group, 93.2%, n = 944, Fig. 3d).

Clinical and genetic features of CRC patients with POLE mutations

Table 1 illustrates in detail associations between the four CRC subgroups and their clinical and genetic features. We observed that a significant proportion of tumours classified within the POLE group occurred in younger patients with a mean age of 52.5 years. Additionally, all POLE (4/4) and 93.1% (27/29) of MSI-M were proximally located, in contrast to 36.1% (13/36) of MSI-U and 29.9% (282/944) of non-MSI. All patients with POLE mutations were diagnosed at an earlier stage (4 out of 4 at stages I–II), similar to 72.4% (21/29) of MSI-M and 77.8% (28/36) MSI-U, compared to 46.6% (440/944) of non-MSI (P < 0.0001).

Table 1 Association between the clinicopathological features of CRC patients stratified by POLE mutation, MSI, and MLH1 methylation status
Full size table

BRAF and KRAS mutations were not observed in POLE-mutant tumours. In contrast, BRAF mutations were observed in 69.0% (20/29) of MSI-M, 11.1% (4/36) of MSI-U, and 3.7% (35/944) of non-MSI, whereas KRAS mutations were observed in 22.2% (8/36) of MSI-U and 32.5% (307/944) of non-MSI (P < 0.0001).

Epigenetic features of CRC patients with POLE mutations

To better understand the differences between tumours with POLE mutations and the other subtypes, we evaluated the methylation status of discrete regions within the promoter of the MGMT and SFRP2 genes.

Concerning MGMT methylation status, lack of methylation in the minimal promoter (Mp) and enhancer (Eh) regions (defined as unmethylated) was observed in 100% (4/4) of POLE, 37.9% (11/29) of MSI-M, 75.0% (27/36) of MSI-U, and 76.1% (718/944) of non-MSI. Partial methylation in MGMT (i.e. affecting either MP or Eh) was observed in none of the POLE cases, but in 3.5% (1/29) of MSI-M, 13.9% (5/36) of MSI-U, and 6.4% (60/944) of patients without MSI. Extensive methylation of MGMT (i.e. affecting both Mp and Eh) was observed in none of the POLE, 58.6% (17/29) of MSI-M, 11.1% (4/36) MSI-U, and 17.6% (166/944) non-MSI (P < 0.0001, Fig. 4a, b).

Fig. 4
figure 4

Methylation analysis of the promoter region in the MGMT and SFRP2 gene. a Schematic depiction of two regions (Mp and Eh region) of the MGMT promoter for methylation, and results of a panel of representative fluorescent bisulphite PCR following restriction enzyme analysis. Methylated samples had the new fragment cleaved by the restriction enzyme (black triangles). White triangles represent unmethylated alleles. M and U denote methylated and unmethylated, respectively. A grey bar denotes an untranslated exon. b The frequencies of MGMT promoter methylation status among the POLE, MSI-M, MSI-U, and non-MSI groups. Extensive methylation is defined as 5.0% or more methylation on both Mp and Eh regions. Partial methylation is defined as 5.0% or more methylation on either Eh or Mp region. c Schematic depiction of two regions (R1 and R2 region) of the SFRP2 promoter for methylation, and results of a panel of representative fluorescent bisulphite PCR following restriction enzyme analysis. Methylated samples had the new fragment cleaved by the restriction enzyme (black triangles). White triangles represent unmethylated alleles. M and U denote methylated and unmethylated, respectively. A black bar denotes translated exon. d The frequencies of SFRP2 promoter methylation status among the POLE, MSI-M, MSI-U, and non-MSI groups. Extensive methylation is defined as 5.0% or more methylation on both R1 and R2 region. Partial methylation is defined as 5.0% or more methylation on either R1 or R2 region. The number in each column denotes the number of samples classified in each category divided by MGMT or SFRP2 methylation status

Full size image

In contrast to MGMT methylation status, lack of methylation in both region-1 (R1) and region-2 (R2) in SFRP2 (defined as unmethylated) was not observed in POLE or MSI-M, but in 13.9% (5/36) of MSI-U and 15.7% (148/944) of non-MSI. Partial methylation in SFRP2 (i.e. affecting either R1 or R2) was observed in none of the POLE, but in 3.5% (1/29) of MSI-M, 22.2% (8/36) of MSI-U, and 20.6% (194 of 944) of non-MSI patients. Extensive methylation of SFRP2 (i.e. affecting both R1 and R2) was observed in all of the POLE, 96.6% (28/29) of MSI-M, 63.9% (23/36) of MSI-U, and 63.8% (602/944) of non-MSI (P = 0.0158, Fig. 4c, d).

Using MLH1, MGMT, and SFRP2 methylation status, we calculated the mean methylation score for each subgroup. When methylation data were analysed using the discrete regions in the promoter of the three genes, the mean methylation score was significantly higher in MSI-M, while the mean methylation score in POLE was the same as that in MSI-U and non-MSI (MSI-M: 5.2; POLE: 2.5; MSI-U: 2.2; non-MSI: 2.0, Fig. 5a).

Fig. 5
figure 5

Features of POLE-mutant CRCs. a Mean methylation score in various subgroups of CRCs categorised by POLE mutation, MSI, and MLH1 promoter methylation status. The mean methylation score in each subset was calculated based on the MLH1, MGMT, and SFRP2 gene (six loci). In the red box plot diagrams, the horizontal line within each box represents the median; the limits of each box are the interquartile ranges, the whiskers are the maximum and minimum values, and the green horizontal bar within each box depicts the mean value. The numbers under the green horizontal bar denote the mean methylation score. The P values above the square panels are statistical differences among any two individual groups calculated by the Steel–Dwass test. b Representative immunohistochemical images for the cytotoxic T-cell marker CD8 (brown) in POLE-mutant CRC samples. c Association among age at diagnosis, MLH1-AB region methylation status, and the number of CD8-positive ( +) cells per HPF in POLE-mutant CRCs. The P value is calculated by the Kruskal–Wallis test. Red circles denote patients with POLE proofreading mutations. HPF; high-power field (× 400)

Full size image

Finally, Table 2 presents detailed information on the four POLE-mutant CRC patients. Somatic POLE mutations in CRCs are associated with enhanced tumour immunogenicity and increased CD8-positive lymphocytic infiltration [7, 13, 14, 21]. We examined CD8-positive infiltration in the four POLE-mutant CRC tissues (Fig. 5b). The number of CD8-positive cells per high-power field (HPF, × 400) was increased in POLE-mutant patients over 60 years of age at diagnosis, who possessed MLH1-AB region methylation (Table 2). For reference, we examined the association between age at diagnosis, MLH1-AB region methylation status, and the number of CD8-positive cells per HPF (Fig. 5c). The mean age at diagnosis of patients with unmethylation in the MLH1-AB region was significantly younger than that of patients with methylation in the MLH1-AB region (65.5 years old [range, 21–95] vs. 68.7 years old [range, 37–89] years, P = 0.001). Although tumour-infiltrating lymphocytes were likely to be more common in older POLE-mutant patients and/or with methylation in the MLH1-AB region, the analysed number of patients was too small to find associations.

Table 2 Detailed information of the four CRC patients with POLE proofreading mutations
Full size table

Discussion

In the present study, we report a biologically distinct subset of CRCs with pathogenic somatic POLE mutations, especially those with simultaneous epigenetic alterations. As previously reported, pathogenic somatic POLE mutations are usually found in the early stages of CRC in the right side of the colon and relatively younger patients with CRC.

CIN, MSI, and CIMP constitute the three major mechanisms of genomic or epigenetic instability in CRC [1, 2, 5]. Experimental evidence has consistently supported the presence of CIMP in a subset of CRCs and correlates with the presence of BRAF V600E mutations [1, 22, 23]. In terms of CIMP development, it is debatable whether BRAF mutations can directly induce CIMP [24, 25]. Recently, Tao et al. [4] reported that the BRAF V600E mutation does not directly cause CIMP, and thus, aging-like acquisition of DNA methylation may favour the survival of cells with such mutations in the BRAF gene through suppression of senescence and activation of stem cell pathways.

Clinically, most sporadic MSI CRCs arise from a CIMP background caused by the epigenetic silencing of the MLH1 gene, which justifies our reason for selecting methylation status of the MLH1 gene at the beginning as a critical biomarker for classifying CRCs [1, 26]. CIN and CIMP have been proposed to represent two major mechanisms of genomic and epigenetic instability in CRC, and up to 50% of CRCs may be CIMP positive [1]. Indeed, CIMP determinations using CIMP-related markers have consistently identified clusters of CRCs with MSI and BRAF V600E mutations, but rarely KRAS mutations [27, 28]. However, when additional methylation loci are investigated, additional subsets of CRCs have been identified with extensive methylation; these tumours are non-MSI and are associated with mutations in the KRAS gene [1, 17, 29]. BRAF and KRAS gene products function in the same MAP-kinase signalling pathway, and activating mutations in these genes occur mutually exclusive [30, 31]. Interestingly, before the discovery of BRAF mutations, KRAS mutations have been proposed as a possible cause of aberrant methylation. Fibroblasts transformed by fos or ras show upregulated DNA methyltransferase expression and consequent global hypermethylation [32]. Indeed, as previously reported, most non-conventional CIMP markers are more likely to gain methylation in CRCs with KRAS mutations [26]. Therefore, we hypothesised that CIMP in CRC would result from activating mutations in either BRAF or KRAS. From this point of view, it might be sufficient to summarise the features of methylation by using the MLH1, MGMT, and SFRP2 genes.

In the case of KRAS mutations, promoter methylation within the MGMT gene, which encodes a DNA repair gene that removes pro-mutagenic O6-methylguanine residues from DNA, is associated with KRAS-mutant CRC [17, 26, 33]. Thus, in this study, we examined MGMT methylation status to clarify the epigenetic features of POLE-mutant CRCs. Interestingly, although half of the POLE-mutant CRCs showed partial methylation in the MLH1 promoter, none of the POLE-mutant CRCs displayed any methylation in the discrete promoter regions of the MGMT gene.

In addition to MLH1 and MGMT, we evaluated SFRP2 methylation (partial or extensive) because most CRCs possess extensive methylation in their promoter regions [34,35,36]. An interesting feature of SFRP2 methylation is that extensive methylation is rare in adenomatous polyps, while it is common in CRCs [35, 36]. Additionally, extensive SFRP2 methylation is more frequently observed in CRCs with KRAS mutations than in those with BRAF V600E mutations and those with wild-type KRAS or BRAF mutations [26, 35]. Thus, although extensive methylation in MGMT and SFRP2 has a similar feature in terms of association with CRC with KRAS mutations, all POLE-mutant tumours demonstrated extensive methylation in SFRP2, but no methylation in MGMT.

Temko and colleagues demonstrated that acquisition of POLE mutations induces a distinct pattern of mutations in cancer driver genes, a substantially increased mutation burden, and an enhanced immune response that is detectable even in precancerous lesions [14]. Similar to this distinct genetic mutation pattern, although the mean methylation score in POLE-mutant tumours was similar to that in the MSI-U and non-MSI groups, POLE mutations may also cause a distinct pattern of epigenetic alterations in cancer-associated genes.

POLE-mutant CRCs have been reported to have a favourable prognosis, as noted for early-stage dMMR tumours [7]. The four POLE-mutant CRC patients in this study were diagnosed at stages I to II, and there was no recurrence within five years after surgical resection. IHC analysis showed that the number of CD8-positive cytotoxic tumour-infiltrating lymphocytes in POLE-mutant CRCs significantly increased in POLE-wild-type CRCs, especially in mismatch repair protein-proficient CRCs [7, 14]. We also examined the number of CD8-positive cells and compared them to the MLH1-AB region methylation status and age at diagnosis. In this study, CD8-positive cells were likely to be more common in older POLE-mutant patients and/or with methylation in the MLH1-AB region, but the number of analysed POLE-mutant patients was too small to reach statistical significance.

Conclusions

Although we analysed over 1,000 CRC samples, CRCs with POLE proofreading mutations were found in only four tumours. Therefore, this rarity suggests that our results should be interpreted with caution. We conclude that CRC patients with POLE mutations are rare, such mutations are observed in younger individuals, lesions are often located within the right colon, diagnosis occurs at an earlier stage, and distinct epigenetic alterations might be associated with CD8 cell infiltration.

Methods

Aim of the study

This study aimed to clarify the incidence of POLE proofreading mutations in a large cohort of CRC patients in Japan and to evaluate the epigenetic profiles of POLE-mutant CRC and elucidate the clinicopathological features associated with key genetic and epigenetic alterations in this malignancy.

Study participants and sample collection

A cohort of 1,052 CRC patients underwent surgical resection between 1998 and 2017 at the Okayama University Hospital, Japan. Tumour specimens and corresponding normal mucosa samples were collected according to institutional review board (IRB)-approved protocols (genome 270 and genome 271 at Okayama University; 3196–1 and 3239 at Kawasaki Medical School).

Extraction of DNA and bisulphite conversion

Genomic DNA was extracted from fresh-frozen samples using a QIAamp DNA Mini Kit (Qiagen NV, Hilden, Netherlands). The extracted DNA was quantified using a Qubit 4 fluorometer with a Qubit dsDNA BR assay kit (Thermo Fisher Scientific, Cleveland, OH, USA). After determining the DNA quality, an EZ DNA methylation kit (Zymo Research, USA) was used for bisulfite conversion of the normalised samples.

Detection of pathogenic POLE proofreading mutations

Pathogenic POLE hotspot mutations in the proofreading domain (exons 9, 13, and 14) were evaluated by Sanger sequencing. The primer sequences used are listed in Additional file 2: Table 2. PCR products were purified using a QIAquick PCR purification kit (Qiagen) and directly sequenced using an ABI PRISM® 3100-Avant Genetic Analyser (Applied Biosystems) and a SeqStudio Genetic Analyser (Thermo Fisher Scientific).

MSI analysis

MSI status was analysed in all CRC tissues using four mononucleotide repeat markers (BAT26, NR21, NR27, and CAT25), as described previously [12, 16]. When at least one or more mononucleotide repeat markers displayed MSI, tumours were defined as having an MSI phenotype [37]. However, tumours without MSI in the four mononucleotide repeat markers were defined as having a non-MSI phenotype, as described in our previous studies [12, 16].

MMR protein and CD8 immunohistochemistry

We employed immunohistochemistry to examine the MMR protein expression of MLH1, MSH2, PMS2, and MSH6 in primary tumour tissues of CRC specimens that showed the MSI phenotype. Staining was performed manually using FFPE specimens. Thin (5 µm) sections of representative blocks were deparaffinised and dehydrated using gradient solvents. Following antigen retrieval in citrate buffer (pH 6.0), endogenous peroxidase was blocked with 3% H2O2. Thereafter, the slides were incubated overnight in the presence of purified mouse monoclonal antibodies against MLH1 (clone G168-15, BD Pharmingen, San Diego, CA; dilution 1:50), MSH2 (clone G219-1129, BD Pharmingen, San Diego, CA; dilution 1:200), PMS2 (clone A16-4, BD Pharmingen San Diego, CA; dilution 1:200), and MSH6 protein (clone 44/MSH6, BD Pharmingen San Diego, CA; dilution 1:100). Further incubation was performed with a secondary antibody and the avidin–biotin–peroxidase complex (Vector Laboratories, Burlingame, CA, USA), followed by incubation with biotinyl tyramide and streptavidin–peroxidase. Diaminobenzidine was used as a chromogen, and haematoxylin was used as a nuclear counterstain. Tumour cells were scored negative for MMR protein expression only if the epithelial cells within the tumour tissue lacked nuclear staining, while the surrounding stromal cells showed positive staining. Samples showing proficiency in all MMR protein expressions were defined as pMMR, and samples showing deficiency in at least one of the four MMR proteins were defined as dMMR. When a tumour showed neither MLH1 nor PMS2 with staining, the tumour was classified as MLH1-deficient (dMLH1); when a tumour showed neither MSH2 nor MSH6 with staining, the tumour was classified as MSH2-deficient (dMSH2); when a tumour showed negative staining only for PMS2 but was positive for MLH1, the tumour was classified as PMS2-deficient (dPMS2), and when a tumour showed negative staining only for MSH6 but was positive for MSH2, the tumour was classified as MSH6-deficient (dMSH6).

IHC analysis for CD8 was performed in the four POLE-mutant tumours, as described previously [13]. The number of CD8-positive cells in the epithelial and stromal regions was quantified. The CD8 count per case was evaluated in a high-power field (HPF, × 400).

DNA methylation detection within discrete regions of the MLH1, MGMT, and SFRP2 gene promoters

In addition to MLH1, we examined the discrete regions of MGMT and SFRP2 promoters that affect these expressions or other critical features in tumorigenesis to clarify the epigenetic features in tumours with POLE mutations. Like MLH1, the MGMT gene is inactivated as a consequence of extensive methylation in its promoter region (dense methylation through minimal promoter [Mp] region to enhancer [Eh] region) [17, 26, 38, 39]. Extensive methylation in MGMT is significantly associated with MGMT protein downregulation and an increased burden of KRAS mutations in CRC patients [17, 33, 38]. Consistent with this epigenetic feature, dense methylation of regions 1 and 2 within the SFRP2 gene promoter is a definitive feature of advanced CRCs [4, 34,35,36].

To quantify the population of methylated alleles of the MLH1, MGMT, and SFRP2 promoters in each sample, a modified combined bisulfite restriction analysis (COBRA) with fluorescence dyes was performed to quantitatively measure the methylation density [12, 36, 38]. The primer sequences and restriction enzymes used are listed in Additional file 2: Table 2. PCR products digested with HhaI, RsaI, or Bst UI (New England BioLabs, Ipswich, MA, USA) were loaded simultaneously onto a SeqStudio Genetic Analyser (Thermo Fisher Scientific). Signals from individual PCR products were distinguished by the unique fluorescent PCR signal from each target and their fragment length, and the data were analysed using GeneMapper software 5 (Applied Biosystems, Foster City, CA, USA). In this study, the percentages of methylated CpG sites (digested by restriction enzymes) were calculated by determining the ratios between the restriction enzyme-cleaved PCR products and the total amount of PCR product in each locus, which was defined as the percentage of methylated CpG sites at 5.0% or more.

In our cohort of 1,052 CRCs, due to technical challenges in performing fluorescence bisulfite PCRs for the MLH1, MGMT, or SFRP2 genes, two tumours that were otherwise characterised as dMLH1 by IHC analysis and 37 tumours without MSI were excluded from further analysis (Fig. 1).

Detection of BRAF and KRAS mutations

Sanger sequencing was performed to confirm mutations in KRAS exon 2 and BRAF exon 15 (including codon 600), as described previously [12, 26]. PCR products were purified using a QIAquick PCR purification kit (Qiagen) and directly sequenced using an ABI PRISM® 3100-Avant Genetic Analyser (Applied Biosystems) and a SeqStudio Genetic Analyser (Thermo Fisher Scientific).

Statistical analysis

All statistical analyses were performed using JMP Pro software (version 14.3; SAS Institute, Inc., Cary, NC, USA). Methylation levels in the MLH1, MGMT, and SFRP2 promoters were analysed as both continuous and categorical variables (methylation level ≥ 5%; unmethylated: methylation level < 5%).

Categorical variables were compared using the Chi-squared test. The pair-wise comparisons for each subgroup were performed using a nonparametric multiple comparison method using the Steel–Dwass test. The association between age at dignosis and MLH1-AB region methylation status was evaluated by the Kruskal-Wallis test. All reported P values were two-sided, and statistical significance was set at P < 0.05.