DNA methylation analyses

DNA from all samples was bisulfite modified and subjected to methylation specific polymerase chain reaction (MSP) for each gene.[7,8] Two of the authors independently scored all samples and the methylation status of all positive samples was confirmed by a second, independent round of MSP. If any discrepancies appeared, a third round of analysis was performed. In line with consensus scoring procedures, we considered carcinomas with band intensities as strong as the positive control (++) as methylated [see Additional file 2] for the gene promoter in question, while the benign lesions and normal mucosa were scored as positive also when weakly methylated, i.e. (+).

For detailed MSP protocol, primer sequences, and scoring criteria see Additional file 1. Representative MSP results can be seen in Figure ​Figure11.

Open in a separate window

Figure 1

Representative methylation results in colorectal tumors and normal mucosa. Results of CDKN2A, CRABP1, HOXA9, and RUNX3 in selected samples are shown. Positive controls (POS): NB, normal blood, for the unmethylated reaction and IVD, in vitro methylated DNA, for methylated reaction. Negative controls: dH₂O. U: unmethylated alleles, M: methylated alleles. The ladder (left lane) is the EZ Load™ 100 bp Molecular Ruler (BioRad, Hercules, CA, USA).

Eleven genes, ADAMTS1, CDKN2A (encoding p16^INK4a),CRABP1, HOXA9, MAL, MGMT, MLH1, NR3C1, PTEN, RUNX3, and SCGB3A1 (encoding HIN-1), were analyzed for promoter methylation by MSP. The methylation status of ADAMTS1, CRABP1, MAL, and NR3C1 for the present series,[9,10] and the methylation status of CDKN2A, MGMT, and MLH1 for the carcinomas [11] have previously been reported.

Quantitative MSP

Primers and probes for quantitative MSP (qMSP) were designed to specifically amplify fully methylated bisulfite-converted complementary sequences of the promoter of interest. The primers and probe sequences used for the MGMT [GenBank: NM_002412] are listed in Additional file 3. To normalize for DNA input in each sample, a reference gene (ACTB [12]) was used.

Fluorescence based real-time PCR assays were carried out in a reaction volume of 20 μL, consisting of 16.6 mM ammonium sulphate; 67 mM trizma preset; 6.7 mM MgCl₂; 10 mM mercaptoethanol; 0.1% DMSO; 200 μM each of dATP, dCTP, dGTP, and dTTP; 600 nM of each primer; 0.4 μL of Rox dye; 200 nM of probe; 1 unit of platinum Taq polymerase (Invitrogen, Carlsbad, CA, USA), and 2 μl of bisulfite-modified DNA as a template. PCR was performed in separate wells for each primer/probe set and each sample was run in triplicate. Additionally, multiple water blanks were used, and as positive and negative control we used commercial methylated and unmethylated DNA (Millipore, Temecula, CA, USA). A series of dilutions of methylated DNA after bisulfite conversion were used for constructing a standard curve to quantify the amount of fully methylated alleles in each reaction. All amplifications were carried out in 96-well plates on an 7000 Sequence Detection System (Applied Biosystems, Foster City, CA, USA), at 95°C for 2 min followed by 45 cycles of 95°C for 15 s, and 60°C for 1 min.

In order to adjust for the possible various amounts of bisulfite treated DNA input in each PCR, the qMSP levels were normalized against the respective values of the internal reference gene (ACTB). The ratio thus generated constitutes an index of the percentage of input copies of DNA that are fully methylated at the primer- and probe-binding sites. The ratio was multiplied by 100 for easier tabulation (methylation level = target gene/reference gene × 100).

A given sample was considered positive for promoter hypermethylation when amplification was detected in at least 2 of the triplicates of the respective qMSP analysis. The qMSP threshold was determined by adjusting the best fit of the slope and R2, using the calibration curve.

Selection criteria for the 11 gene promoters analyzed in the present study

Some of the genes analyzed were known to be targeted through promoter methylation in cancer, including colorectal cancer (SCGB3A1, RUNX3, CDKN2A, MLH1, and MGMT). HOXA9 was a potential new methylation target in colorectal cancer. ADAMTS1, CRABP1, MAL, and NR3C1 were identified as novel epigenetically silenced target genes in colorectal cancer by our group.[9,10,13] They were selected to be tested in combination with known methylated genes in a large series of colorectal lesions to check for interdependencies. The methylation status of all included genes was compared in a series of normal mucosa from individuals without cancer with those of normal, benign and malignant tissue from the large bowel of cancer patients. Only two previous studies have compared gene methylation among the same four types of sample groups as investigated here.[14,15] The first only investigated one gene and the latter 10 genes among which only three overlapping the present selected gene list.

Gene mutation status of BRAF, KRAS and TP53

The present carcinoma series form a part of a series previously studied for genetic changes, including BRAF, KRAS and TP53.[23,24] The specific mutation status of the individual tumors included here can be found in Additional file 4.

DNA promoter methylation in normal mucosa, adenomas, and carcinomas

The results of the MSP analyses of all samples and each gene are summarized in Figure ​Figure2,2, Table ​Table1,1, and Additional file 5. The mean number of genes methylated per sample was 0.4 for the N1 group, 1.2 for N2, 2.2 for adenomas, and 3.9 for carcinomas, and was significantly different among the groups using Kruskal-Wallis test; P < 0.0001 (mean rank N1, 10.2; N2, 17.3; adenomas, 23.1; and carcinomas, 31.4). Overall, 6/21 (29%) of the N1 samples, 9/18 (50%) of the N2 samples, 52/63 (83%) of the adenomas, and 48/52 (92%) of the carcinomas, were methylated in one or more of the eleven analyzed genes.

Open in a separate window

Figure 2

Methylation profiles of normal mucosa, adenomas, and carcinomas. Eleven genes were analyzed by MSP. Upper panel: non-cancerous lesions; lower panel: carcinomas stratified according to MSI-status. X-axis, the analyzed genes; Y-axis, the percentage of methylated samples. N1: normal colon samples from cancer-free individuals; N2: normal colon samples from cancer patients; MSI: microsatellite instability; MSS: microsatellite stability.

Statistically significant differences in methylation frequencies among sample groups were also evident at the single gene level. ADAMTS1, CDKN2A, CRABP1, MLH1, NR3C1, RUNX3, and SCGB3A1 showed increasing methylation frequencies from adenomas to carcinomas, while HOXA9, MAL, and MGMT displayed overall equal methylation frequencies in all tumor subgroups. PTEN was unmethylated in carcinomas, and was thus not investigated in adenomas or included in the figures, tables (except Additional file 4) or statistics.

The more frequent promoter hypermethylation found among N2 samples compared with N1 samples was apparent both for the total number of methylated genes, and at the individual gene level (MGMT, P = 0.055). The reliability of our MSP scorings was tested by quantitative MSP analysis of one example gene performed in a blinded manner in another lab. The results were in perfect concordance [see Additional file 6].

No difference was seen in methylation frequencies between N2 samples with corresponding MSI-positive carcinomas (n = 6) and those with corresponding MSS carcinomas (n = 12).

Overall, gene methylation frequencies were higher among MSI than among MSS carcinomas, and were statistically significant for CRABP1, MLH1, NR3C1, RUNX3, and SCGB3A1 (P ≤ 0.0001, P ≤ 0.0001, P = 0.001, P ≤ 0.0001, and P = 0.03, respectively). Methylation of these genes showed a strong association to proximal carcinoma location, demonstrating the close connection between high methylation levels, proximal location and MSI (P ≤0.0001, P ≤ 0.0001, P ≤ 0.001, P = 0.001, and P = 0.04, respectively). Association to site was also seen for HOXA9 in N1 samples (P = 0.04). HOXA9 was also more frequently methylated among non-cancerous normal mucosa (n = 20) from older patients compared to younger patients, indicating age-specific methylation (P = 0.025). However, this was not confirmed among the larger group of carcinomas (n = 52).

Interdependence among hypermethylated genes

From bivariate correlation analysis [see Additional file 7], methylation of MLH1 was correlated with methylation of CRABP1 (correlation coefficient 0.51; P = 5 × 10^-11), NR3C1 (correlation coefficient 0.72; P = 1 × 10^-25) and RUNX3 (correlation coefficient 0.57; P = 6 × 10^-14). Methylation of RUNX3 itself was strongly correlated to methylation of both NR3C1 (correlation coefficient 0.75; P = 5 × 10^-28) and CRABP1 (correlation coefficient 0.67; P = 3 × 10^-20). Methylation of NR3C1 and CRABP1 was also correlated (correlation coefficient 0.59; P = 4 × 10^-15), as well as ADAMTS1 and MAL (correlation coefficient 0.53; P = 2 × 10^-12).

Hierarchical clustering of samples according to gene methylation status showed that MLH1 and NR3C1 were most closely related, followed by RUNX3 and CRABP1. In contrast, HOXA9 and MGMT displayed methylation patterns independent from each other and the other genes (Figure ​(Figure33).

Open in a separate window

Figure 3

Methylation HeatMap. Hierarchical clustering reveals that methylation of NR3C1 and RUNX3 are most closely related, followed by MLH1 and CRABP1. Methylation of MGMT and HOXA9 are most independent both from each other and from rest of the set. The genes are presented in columns, while the samples are presented in rows. Black, unmethylated; red, methylated; and grey, missing values.

Comparing methylation profiles of normal mucosa, adenomas and carcinomas of the large bowel

The identified methylation profiles of normal colorectal tissues, adenomas, and carcinomas demonstrated a stepwise increase in CpG island promoter methylation towards malignancy, indicating that their inactivation plays a role in the progression of the tumor. This was evident both for widespread methylation and at the single gene level (increasing frequencies of methylation from benign to malignant stages) with the exception of HOXA9, MAL, and MGMT. The lack of increase in methylation frequencies between non-malignant adenomas and carcinomas for these three genes may suggest that they are more important in the initiation of cancer, rather than in progression. These genes in addition to ADAMTS1 were also hypermethylated in comparable frequencies among MSS and MSI carcinomas. These observations, and the fact that the separation of the MSI- and MSS-pathway is thought to occur early in colorectal tumorigenesis suggest that alterations of the four genes represent early events. ADAMTS1 is believed to be an inhibitor of both angiogenesis and endothelial proliferation,[32] features commonly activated in cancer, as a tumor must turn on angiogenesis in order to grow larger than 1–2 mm³[33]. Members of the HOX gene family are shown to be commonly altered in several cancers, and to the best of our knowledge, this is the first report of HOXA9 methylation in colorectal neoplasms. HOXA9 methylation has received increasing interest in recent time as it is included in the HOXA-cluster which harbors methylation over a larger area than just a single promoter, indicating that methylation may mimic genetic micro-deletions and turn off a cluster of genes rather than just one at the time, i.e. yet another example of long range epigenetic silencing. [34-36]. MAL is involved in T-cell differentiation, especially in the late or intermediate stages.[37] It is also involved in polarization of epithelial cells caused by apical transport of lipids and proteins. Loss of cell polarity is often seen in neoplastic transformation.[38] For MGMT the early involvement is further supported by the fact that promoter methylation has previously been identified in aberrant crypt foci.[39]

Our data do not suggest that any of the markers included here were methylated in an age dependent manner. Of the 11 analyzed genes, six were unmethylated in all normal samples from non-affected individuals, excluding them as age-specific methylation targets. For two genes (SCGB3A1 and MAL) only one of 21 samples was methylated. Although the sample in question was from an older individual (75 years), the resulting overall methylation frequency was only 5%. This is in strong contrast to the frequent reported age-specific methylation of the N33 gene, which shows approximately 46% methylation among normal samples in general and 58% methylation in normal samples from individuals over 60 years.[40]HOXA9 is the only gene in the present study harboring "frequent" promoter methylation in normal samples (19% overall, and 43% for individuals of 60 years or older). Binary regression analysis resulted in a significant P value, however, when using the same statistical analysis in the tumor sample series age dependence could not be confirmed. Both technical and biological aspects influence the interpretation of DNA promoter methylation analyses.

The importance of primer design is emphasized in the PTEN assay. Promoter hypermethylation of PTEN has been frequently reported in various tumor types, including CRC. [16-19] However, the majority of MSP primer sets used have failed to discriminate between PTEN and its frequently methylated pseudogene, leading to a high rate of false positives.[20] In the present study, we used MSP primers specifically designed to amplify the protein-encoding PTEN gene,[21] and showed that PTEN was not subject to promoter hypermethylation in colorectal carcinomas. A novel study confirms that methylation of PTEN is an unusual event in colorectal cancer as a whole.[22]

Interdependence among hypermethylated genes and widespread methylation

The hierarchical clustering analysis of gene promoter methylation status in normal, benign, and malignant samples confirmed that the distribution of HOXA9 and MGMT methylation frequencies across sample groups differed from the other genes. Overall, methylation of NR3C1 and RUNX3 had the highest correlation (Figure ​(Figure33 and Additional file 7), in addition to MLH1, which was also closely related to NR3C1 and RUNX3. Furthermore, the present study confirmed that hypermethylation of MLH1 was characteristic of right-sided sporadic colon tumors with MSI.[41] The lack of MLH1 hypermethylation in adenomas analyzed in the present study supports the theory that CIMP and MSI-tumors arise from sessile serrated polyps rather than from adenomas.[42]NR3C1, RUNX3, CRABP1, and SCGB3A1 were also shown to have the same characteristics as MLH1, supporting the hypothesis that DNA methylation plays a more prominent role in proximal than in distal carcinogenesis. CRABP1, MLH1, NR3C1, and RUNX3 have recently been shown to belong to a panel of epigenetically regulated genes which best discriminate between CIMP-positive and CIMP-negative tumors, a phenotype strongly related with MSI status.[43]

We found that the MSI positive samples with V600E BRAF mutations were accompanied by promoter hypermethylation of several genes, in agreement with the CIMP phenotype (Figure ​(Figure4).4). Furthermore, we also confirmed that MSS tumors with TP53 mutations had less overall methylation, and thus in agreement with a CIMP negative phenotype. KRAS mutations were evenly distributed between MSI and MSS samples but seemingly the KRAS/MSI samples had more methylation than KRAS/MSS samples. Interestingly, three MSS samples had BRAF mutations, and all differed from the V600E mutation found among the MSI tumors.