Identification of an epigenetic biomarker panel with high sensitivity and specificity for colorectal cancer and adenomas

Background The presence of cancer-specific DNA methylation patterns in epithelial colorectal cells in human feces provides the prospect of a simple, non-invasive screening test for colorectal cancer and its precursor, the adenoma. This study investigates a panel of epigenetic markers for the detection of colorectal cancer and adenomas. Methods Candidate biomarkers were subjected to quantitative methylation analysis in test sets of tissue samples from colorectal cancers, adenomas, and normal colonic mucosa. All findings were verified in independent clinical validation series. A total of 523 human samples were included in the study. Receiver operating characteristic (ROC) curve analysis was used to evaluate the performance of the biomarker panel. Results Promoter hypermethylation of the genes CNRIP1, FBN1, INA, MAL, SNCA, and SPG20 was frequent in both colorectal cancers (65-94%) and adenomas (35-91%), whereas normal mucosa samples were rarely (0-5%) methylated. The combined sensitivity of at least two positives among the six markers was 94% for colorectal cancers and 93% for adenoma samples, with a specificity of 98%. The resulting areas under the ROC curve were 0.984 for cancers and 0.968 for adenomas versus normal mucosa. Conclusions The novel epigenetic marker panel shows very high sensitivity and specificity for both colorectal cancers and adenomas. Our findings suggest this biomarker panel to be highly suitable for early tumor detection.


Background
Colorectal cancer is the third most common cancer type in the US and is a major contributor to cancer-death [1]. Most cases of colorectal cancer develop from benign precursors (adenomas) during a long time interval. This provides a good opportunity for detection of colorectal cancer at an early curable stage and to screen for potentially pre-malignant adenomas [2]. Both flexible sigmoidoscopy and the Fecal Occult Blood Test (FOBT) have been tested in randomized trials and shown to reduce mortality from colorectal cancer [3]. By sigmoidoscopy adenomas may be detected and removed and thus the incidence of cancer will be reduced [4], however, this screening is invasive and cumbersome for the patient. FOBT on the other hand is non-invasive and currently the most commonly used screening test for colorectal cancer in Europe. Although the sensitivity and specificity measurements of FOBT have been substantially improved in recent years [5], they are still not optimal. FOBT is also hampered by the low sensitivity for adenomas. Therefore, during recent years, much effort has been put in the development of fecal DNA markers. A successful biomarker panel that is able to discriminate between healthy individuals and carriers of early colorectal cancer or precursor lesions has the potential of reducing both incidence and mortality of the disease.
Until today, however, no feces DNA test has achieved a satisfactory performance level compared to the screening tests mentioned above.
Aberrant DNA promoter methylation has previously been shown to be an early event in the development of colorectal cancer [6][7][8][9][10]. Several reports of DNA methylation biomarkers tested in fecal [8,[11][12][13][14][15] and blood samples [16][17][18][19] suggest the suitability of epigenetic biomarkers in early diagnostics of the disease. However, only markers which provide a high methylation frequency in samples from colorectal cancer patients and at the same time lack hypermethylation in normal mucosa are suitable for a screening test.
The present study reports on the performance (the sensitivity and specificity) of a novel epigenetic biomarker panel.

Selection of epigenetic markers analyzed in the present study
From an epigenomic screen of colon cancer in vitro models we have previously identified a number of genes responding to 5-aza-2'deoxycytidine treatment [20]. In the present study, thirteen of these candidates were analyzed in 20 colon cancer cell lines in order to identify the most suitable DNA methylation markers for colorectal cancer (Table 1, Figure 1). From this analysis we selected CNRIP1, FBN1, INA, and SNCA for detailed studies in clinical sample series using quantitative MSP (qMSP) assays. Additionally, a qMSP assay was designed and applied for analysis of the MAL gene promoter, previously reported by us as a biomarker for early detection of colorectal tumors by qualitative MSP [21,22]. Finally, the SPG20 biomarker recently reported with a sensitivity of 89% and 78% in colorectal cancer and adenomas, respectively and a specificity of 99% [9] was included for evaluation of a combined biomarker panel performance.
Three MSI cell lines (HCT15, RKO, and SW48) and three MSS cell lines (HT29, LS1034, and SW480) were treated with a) 1 μM of 5-aza-2'deoxycytidine (Sigma-Aldrich, St. Louis, MO, USA) for 72 h, b) 0.5 μM of trichostatin A (Sigma-Aldrich) for 12 h, and c) a combination of both drugs (1 μM 5-aza-2'deoxycytidine for 72 h, 0.5 μM trichostatin A added the last 12 h). The medium was exchanged daily. Twenty-nine additional cell lines from various cancer tissues other than colon cancer were included (described in Additional file 1). All cell lines were harvested before reaching confluence. DNA was extracted from the cancer cell lines by using a standard phenol-chloroform procedure [24], and total RNA was isolated using Trizol (Invitrogen, Carlsbad, CA).

Colorectal cancer and adenoma samples
The colorectal cancer test set comprised DNA from 79 fresh-frozen cases from 78 patients derived from a prospective series from seven hospitals in South-East Norway during 1987-1989 [25]. Twenty-eight tumors were MSI, and 51 were MSS [26,27]. The median patient age at diagnosis was 71 years (range 33-92 years). One patient fulfilled the criteria for hereditary non-polyposis colorectal cancer (HNPCC) [26]. The colorectal cancer validation set consisted of an independent cohort of 105 cases obtained from a prospective series of fresh-frozen colorectal cancer samples at the Department of Surgery, at Oslo University Hospital -Aker Hospital in the period of 2005-2007. Twenty-two out of 105 (21%) cancers were MSI-high, whereas the remaining 83 (79%) samples were of the MSS or MSI-low phenotype. The median patient age at surgery was 71 years (range 29-93 years).
A series of 61 adenomas was obtained from 50 individuals attending a population-based colonoscopy screening study (Telemark, Norway) [28]. The median age at adenoma removal was 67 years (range 62-72 years) and the median adenoma size was 8 mm (range 5-50 mm). Two of the adenomas were MSI whereas the remaining 59 were MSS [20]. The validation set comprised 51 adenomas from 46 individuals attending another screening study [29]. The median age at adenoma removal was 58 years (range 50-64 years) and the median adenoma size was 11 mm (range 4-40 mm).

Normal colorectal tissue samples
Two cohorts of normal colorectal mucosa samples were analyzed. The test set consisted of 51 samples from 48 deceased colorectal cancer-free individuals collected at the Institute of Forensic Medicine, University of Oslo. The median age was 55 years (range 22-86 years). The validation set consisted of rectal mucosa biopsies from 59 individuals attending the population-based colonoscopy screening study mentioned above, harboring neither colorectal adenomas nor cancers [28]. The median age was 67 years (range 63-72 years). Also, 105 normal colorectal mucosa samples taken from the resection margin of the colorectal cancer validation series were included in the present study. Median age for these was 71 years.

Bisulfite Treatment
DNA (1.3 μg) from each tissue sample was bisulfite treated using the EpiTect bisulfite kit (Qiagen) according to the manufacturer's protocol. The desulfonation and washing steps were performed using a QiaCube (Qiagen).

Qualitative methylation-specific polymerase chain reaction (MSP)
Qualitative methylation-specific polymerase chain reaction (MSP) primers were designed using Methyl Primer Express v1.0 (Applied Biosystems, Foster City, CA, USA) according to the following criteria: primers amplified a region within 300 bases of the annotated transcription start site (UCSC Genome Browser [30]), the maximum fragment length was 200 bp, each primer contained a minimum of two CpG sites, all primer pairs contained a minimum of five Cs in non-CpG sites, and all forward primers had a 3'-proximal CpG site. Primers were purchased from MedProbe (MedProbe, Oslo, Norway) and their sequences are listed in Additional file 1, Table S1, along with the product fragment lengths, primer locations, MgCl 2 reaction concentration, and PCR annealing temperature. The MSP templates were amplified using the HotStarTaq DNA polymerase (Qiagen). Human placental DNA (Sigma-Aldrich) treated in vitro with Sss1 methyltransferase (New England Biolabs, Ipswich, MA, USA) was used as a positive control for the methylated MSP reaction, whereas DNA from normal lymphocytes was used as a positive control for the unmethylated reaction. In both reactions we used water as a negative control. PCR products were separated by electrophoresis using 2% agarose and visualized by UV irradiation using a Gene Genius (Syngene, Frederick, MD, USA). All results were confirmed by a second independent round of MSP and scored independently by two authors (SAD and GEL). A third independent MSP was done for samples with diverging results or discrepant scoring from the two initial runs. For each gene, the MSP controls were sequenced in order to verify the identity of the amplified product.
Quantitative methylation-specific polymerase chain reaction (qMSP) CNRIP1, FBN1, INA, MAL, and SNCA promoter methylation were analyzed by qMSP in the test and validation sets as well as in stool samples. SPG20 has previously been reported [9]. Primers and probes were designed using Primer express v3.0 (Applied Biosystems) and purchased from MedProbe (MedProbe, Oslo, Norway) and Applied Biosystems, respectively. Sequences are listed in Additional file 1, Table S2. Probes were labelled by 6-FAM and a minor groove binder non-fluorescent quencher. All genes were amplified and normalized for DNA input using the ALU-C4 sequence [31]. The qMSP was carried out in triplicates in 384-well plates using a 20 μl reaction volume including 0.9 μM each of forward and reverse primers, 0.2 μM probe, 30 ng bisulfite treated template (tissue samples) or 1 ul bisulfite treated template (stool sample), and 1 × TaqMan Universal PCR master mix NoAmpErase UNG (including Ampli-Taq Gold DNA polymerase and passive reference; ROX). Fragments were amplified at 95°C for 10 min, then 45 cycles of 95°C for 15 sec followed by 60°C for 1 min using the 7900HT Sequence Detection System (Applied Biosystems). The median value was used for data analysis. Bisulfite-converted completely methylated DNA (CpGenome Universal Methylated DNA; Millipore Billerica, MA, USA) served as a positive control for the qMSP reaction and 1:5 serial dilutions (32.5 -0.052 ng) were used to generate a standard curve for quantification. Additionally, all plates contained multiple water blanks, bisulfite modified DNA from normal lymphocytes as well as unmodified DNA as a negative control.
For all samples, amplification after cycle 35 was censored, in accordance with the protocol from the manufacturer (Applied Biosystems). The qMSP results were calculated as percent of methylated reference (PMR) in accordance with a previous report [32]. In brief, the median GENE:ALU ratio of a sample was divided by the median GENE:ALU ratio of the positive control (CpGenome Universally Methylated DNA) and multiplied by 100. To ensure high specificity for tissue samples, the percentile of the highest PMR value across all genes and all normal mucosa samples in the test set was used to set a fixed threshold (at PMR = 7) for scoring positive methylation, regardless of the gene in question. This scoring threshold was used in analyses of all tissue samples, including normal mucosa, adenomas and carcinomas in both the test and validation sets. Three samples had outlier PMR values (CNRIP1, 12.41; FBN1, 12.00; and MAL, 12.22) that were excluded when the fixed threshold was set. The validation sets and the normal mucosa samples matching the CRC validation set were analyzed in a blinded manner.
Direct bisulfite DNA sequencing DNA bisulfite primers were designed using Methyl Primer Express v1.0 (Applied Biosystems) according to the following criteria: sequences covered the area amplified by the MSP primers of the respective gene promoter including the transcription start point, maximum fragment length was 450 bp, preferably none and maximum two CpG sites were included in each primer, and when possible, repetitive sequences of more than eight bases were avoided. Primer sequences, product fragment lengths, primer locations, MgCl 2 reaction concentration, and PCR annealing temperatures are listed in Additional file 1, Table S1. A representative promoter region of CNRIP1, FBN1, INA, and SNCA was subjected to direct bisulfite sequencing in 20 colon cancer cell lines, as previously described [22]. The approximate amount of methyl cytosine of each CpG site was calculated by comparing the peak height of the cytosine signal with the sum of the cytosine and thymine peak height signals [33]. CpG sites with ratios ranging from 0-0.20 were classified as unmethylated, CpG sites within the range 0.21-0.80 were classified as partially methylated, and CpG sites ranging from 0.81-1.0 were classified as hypermethylated.
RNA isolation, cDNA preparation, and real-time quantitative gene expression analysis Total RNA was extracted from cancer cell lines (n = 47), colorectal cancers (CRC test set; n = 17), and normal colorectal tissue (taken from the resection margin of the CRC test set; n = 3) samples using Trizol (Invitrogen, Carlsbad, CA, USA). RNA from 16 additional colorectal cancer samples (CRC test set) was extracted using AllPrep DNA/RNA Mini Kit (Qiagen). The RNA quality was measured using a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and the concentration was determined using ND-1000 Nanodrop (Nano-Drop Technologies, Wilmington, DE, USA).
Total RNA was converted to cDNA using the High-Capacity cDNA Archive kit (Applied Biosystems), including random primers. The cDNA of CNRIP1 (Hs00384403_m1), FBN1 (Hs00171191_m1), INA (Hs00190771_m1), SNCA (Hs00240906_m1), and the endogenous controls ACTB (Hs99999903_m1) and GUSB (Hs99999908_m1) was amplified separately in 384 well plates according to the manufacturers' protocol (Applied Biosystems), and the resulting quantitative gene expression measurements were registered by the 7900HT Sequence Detection System (Applied Biosystems). Samples were analyzed in triplicates, and the median value was used for data analysis. The human universal reference RNA (containing a mixture of total RNA from ten different cell lines; Agilent) was used to generate a standard curve, and the resulting quantitative expression levels of CNRIP1, FBN1, INA, and SNCA were normalized against the mean value of the two endogenous controls.

Ethics
According to National legislation all samples belong to approved research biobanks and approvals are given by the Regional Ethics Committee (S-09282c2009/4958 biobank 2781;S95151).

Statistical analysis
For statistical analyses, SPSS 16.0 (SPSS, Chicago, IL, USA) was used. Pearson's chi-square and Fisher's exact tests were used for categorical variables. Student T-test and Mann-Whitney U test were used to investigate potential associations between tumor DNA methylation and patient age and polyp size. Student T-test was also used to examine the relationship of aberrant promoter methylation to gene expression in tissue samples and cancer cell lines. All P values derive from two-tailed tests. Receiver Operating Characteristics (ROC) curves for individual biomarkers were generated using percentage methylated reference (PMR) values and tissue type (cancer or adenoma and normal) as input. For evaluation of the combined biomarker panel the sum of PMR values from CNRIP1, FBN1, INA, MAL, SNCA, and SPG20 was used.

Identification of the most suitable biomarkers
In the evaluated 20 colon cancer cell lines, all but one (UBE3A) gene promoters were hypermethylated with frequencies ranging from 20% (LEF1 and MEF2C) to 100% (CNRIP1; Figure 1, and Additional file 2, Figure  S1). Six genes (BEX1, C3orf14, CNRIP1, FBN1, INA, and SNCA) were methylated in > 80% of the cell lines, representing promising biomarkers in terms of sensitivity. These were subjected to further detailed analyses in parts of the colorectal cancer (n = 51) and normal tissue (n = 21) test sets using qualitative MSP analysis. As shown in Figure 2, for all genes the frequency of promoter methylation was significantly higher in cancers compared to the normal tissue samples. Methylation in colorectal cancers ranged from 55% (C3orf14) to 96% (BEX1; median 84%), while methylation in normal tissue was between 0% (CNRIP1, FBN1, INA) and 45% (BEX1; median 5%), P < 0.0001 to P < 0.02. To increase the likelihood of identifying biomarkers with tumor specific methylation, only gene promoters with methylation frequencies equal to or lower than 10% in normal samples (CNRIP1, FBN1, INA, and SNCA) were subjected to further analyses.
Quantitative DNA methylation analyses in test sets: cancers, adenomas, and normal colorectal samples CNRIP1, FBN1, INA, MAL, and SNCA promoter methylation was analyzed quantitatively (qMSP) in the colorectal cancer (n = 74; median age 71 years) and normal mucosa (n = 51; median age 55 years) test sets. We found overall methylation percentages of 99, 81, 66, 92, and 73 in colorectal cancers and 2, 2, 0, 2, and 0 in normal mucosa, respectively ( Table 2). Quantitative data for SPG20 [9] was included in a biomarker panel evaluation. Co-methylation, here defined as simultaneous hypermethylation of two or more of the six gene promoters (CNRIP1, FBN1, INA, MAL, SNCA, and SPG20), was found in 99% of the colorectal cancer test set samples and 2% of the normal mucosa samples. ROC curves of the six biomarkers combined resulted in an area under the ROC curve (AUC) value of 0.999 (Additional file 1, Table S3).

DNA methylation analyses in validation series
The frequencies of methylation in the validation series were comparable with the findings in the test set (Table  2). Ninety-five out of 105 colorectal cancers (90%; median age 71 years) and 49 out of 51 adenomas (96%; median age 58 years) were hypermethylated in at least two of the six analyzed markers, in contrast to one of the 59 normal mucosa samples (2%; median age 67 years). The distribution of PMR values in the normal mucosa samples, the adenomas, and cancers is illustrated in Figure 3. The AUC Figure 2 Summary of promoter methylation status in test sets of colorectal cancer and normal colorectal tissue. Red color indicates methylated sample, green color indicates unmethylated sample, and grey color indicates samples that were not successfully amplified and thereby not scorable. Gene promoters in the upper panel (CNRIP1, FBN1, INA and SNCA) were methylated in 10% or less of the normal tissue samples tested and represent biomarkers with potentially high specificities, and thus suitable for a future diagnostic test. Gene promoters in the lower panel (C3orf14 and BEX1) were methylated more frequently than 10% in normal tissue samples, which might limit the specificity in a test situation, and were excluded from further analyses.  Table S3).

Associations of genetic and clinico-pathologic data with tumor methylation of individual genes
Across the test (n = 74) and validation (n = 105) series, 33 out of 179 (18%) colorectal cancers harbored a BRAF mutation in exon 15. In 23 out of 48 colorectal cancer cases with microsatellite instability (MSI-high; 48%), a BRAF mutation in exon 15 was present, compared to only 10 out of 131 (8%) of the MSS/MSI-low tumors (P = 8.9E-9).
Methylation of the individual genes was more common among proximal, MSI-high, and BRAF mutated cancers, although not statistically significant for all comparisons. DNA methylation of each gene was equally frequent in colorectal cancers of all stages (I-IV) as well as in tumors from female and male patients ( Table 3). The age of patients with an INA methylation positive cancer was slightly higher (mean 72 years) than that of patients with an INA methylation negative cancer (mean 65 years; P = 1.4E-4; T-test). However, all normal colorectal samples tested (n = 215) contained unmethylated INA promoters, ruling out age-specific methylation.
Among the adenomas, INA and FBN1 methylation was more frequently found in large (10 mm or larger) than in small adenomas (smaller than 10 mm in diameter; P = 0.012 and P = 0.047, respectively; T-test).
Additional detailed information about clinical associations is presented in Table 4.
Associations of genetic and clinico-pathologic tumor sample data with a combined biomarker panel Co-methylation of two out of six genes (CNRIP1, FBN1, INA, MAL, SNCA, and SPG20) was not associated with patient gender, and neither with cancer stage, proximal or distal location, nor BRAF and MSI status (Table 3). A Mann-Whitney U analysis showed a non-significant trend (P = 0.067) towards higher age among colorectal cancer patients positive for the biomarker panel compared with negative patients, but this was not confirmed in the other sample series (adenomas, normal mucosa, and normal mucosa from cancer patients). All colorectal cancers that harbored BRAF mutation as well as the vast majority (98%) of cancers with the MSI phenotype were biomarker panel positive, in line with the CIMP concept [34]. The biomarker panel was additionally positive in 111/121 (92%) cancers harboring the MSS/ MSI-low phenotypes and wild-type BRAF (Figure 4). Cancers located in the colon, and particularly the distal colon, showed somewhat more frequent co-methylation than did cancers located in the rectum (P = 0.011). The combined biomarker panel reached an AUC value of 0.984 across the colorectal cancer validation and test sets (P = 1.9E-43; Figure 5; Additional file 1, Table S3).
The biomarker panel was also positive in adenomas independent of clinico-pathological characteristics as there was  (Table 4). For adenomas, the AUC value of the combined test and validation sets was 0.968 (P = 2.6E-33; Figure 5; Additional file 1, Table S3).

Validation of promoter methylation status by direct bisulfite sequencing
Methylation status of representative samples was confirmed by direct bisulfite sequencing using primers that flank the MSP and qMSP regions of CNRIP1, FBN1, INA, and SNCA ( Figure 6).

Promoter methylation in relation to gene expression
In order to examine the relationship of aberrant promoter methylation to gene expression, cancer cell lines, colorectal cancers, and normal mucosa samples were subjected to quantitative real time analysis. The methylation status of CNRIP1, FBN1, INA, and SNCA across all cancer cell lines (n = 49) can be seen in Additional file1, Figure S1. The level of mRNA expression of CNRIP1, FBN1, INA, and SNCA was strongly associated with promoter methylation status in cancer cell lines (P = 0.037, P = 0.017, P = 0.006, and P = 0.001, respectively; Figure 7A). Analysis of tissue samples supported that DNA promoter methylation was significantly associated with reduced gene expression for CNRIP1, INA, and SNCA (Additional file 1, Table S4). The mean mRNA expression of FBN1 in methylated samples was lower than in unmethylated samples (0.4 versus 0.8), although this was not statistically significant (P = 0.125) An association between promoter DNA methylation and gene expression was further confirmed by measurements of mRNA levels in colon cancer cell lines treated with epigenetic drugs (both 5-aza-2'deoxycytidine and trichostatin A, alone and in combination). Before drug treatment all colon cancer cell lines tested harbored promoter methylation of CNRIP1, INA, FBN1, and SNCA accompanied by little or no expression of the same genes. With the exception of FBN1, the mRNA levels of all genes subjected to real-time PCR analysis were up-regulated in a least four of the six treated cell lines ( Figure 7B). The combined results from real-time analyses using both in vivo as well as in vitro samples thereby suggest that CNRIP1, INA, and SNCA gene expression might be subjected to epigenetic regulation.

Discussion
We have identified four genes, CNRIP1, FBN1, INA, and SNCA that were frequently hypermethylated in colorectal adenomas and cancers. The methylation was highly tumor-specific, as only a minority of normal mucosa samples harbored promoter hypermethylation Figure 3 Percent methylated reference (PMR) values of all biomarkers in combined test and validation sets of normal mucosa, adenomas, and colorectal cancer. Note that for SPG20 two outliers (PMR > 150) are excluded from the graph. The SPG20 promoter methylation status has previously been published in the same sample series [9]. of the same genes, fulfilling the criteria for cancer-specific methylation (type C) defined by Toyota and coworkers [34]. These genes in combination with the previously reported MAL [21,22] and SPG20 [9] resulted in a biomarker panel with a sensitivity of 94% for colorectal cancers and 93% for adenomas, and a specificity of 98%. A high specificity will increase the positive predictive value in screening and thereby limit the number of false positives and subsequent unnecessary colonoscopies.
The use of genome-wide experimental approaches for identification of novel target genes for tumor-specific DNA methylation has provided long candidate gene lists [35][36][37][38], and a substantial number of genes has been analyzed in detail, i.e. [36,39,40]. In spite of this, only few genes, including GALR2, ITGA4, NTRK2, OSMR, SFRP1, SFRP2, SLC16A12, TUBG2, MAL and NMDAR2A [8,14,21,22,36,[41][42][43] (reviewed in [44] and [15]) have been found to harbor promoter methylation in more than 80% of colorectal cancers analyzed, and simultaneously in less than 10% of normal mucosa samples, opening up the possibility of achieving high sensitivities and specificities in a future test. Up to now, only a few of the markers mentioned above have been analyzed in non-invasive sample material [13,14,42,[45][46][47]. Additionally, the promising biomarkers SEPT9 [48] and VIM [12] have been analyzed in blood/serum samples and stool samples, respectively and are the only markers included in currently available non-invasive tests. Vimentin is the target gene of the ColoSure™ test, which has a company-reported sensitivity and specificity range of 72-77% and 83-94%,  I  27  3  23  7  19  11  26  4  20  10  27  3   II  76  3  60  19  54  25  72  7  52  27  75  4   III  45  4  42  7  31  18  44  5  33  16  45  4   IV  21  0  17  4  13  8  20  1  14  7  respectively. SEPT9 is the target gene of the plasma based Epi ProColon early detection assay which can detect 68%-72% of colorectal cancers with a specificity of 89%-93% [48]. Despite that the first test analyzes stool samples and the second test is blood based, the sensitivity and specificity measurements are quite comparable, both leaving room for improvements. This is underscored by an independent report finding vimentin methylation in only 72% of colorectal cancers and in as much as 11% of normal mucosa samples [39]. In the present study, four of the biomarkers (CNRIP1, FBN1, MAL, and SPG20) were more frequently methylated in colorectal cancers than previously reported for the vimentin gene, and additionally less frequently methylated in normal mucosa samples, underscoring the suitability of these  markers for diagnostic use. The most important results in the present study is, however, the high performance of the combined biomarker panel and the robustness such a panel offers in a future test compared to single markers.
As expected, the identified methylation markers were frequently detected in MSI positive colorectal cancers containing BRAF mutations, features that are consistent with the CIMP positive phenotype [34]. However, due to the broad coverage of the present biomarker panel,  Table S3. The ROC curves for the SPG20 biomarker has been published previously in the same sample series [9]. not only cancers harboring these features were biomarker positive, but also the majority of MSS tumors with wild-type BRAF. It is essential that a diagnostic biomarker panel adequately "represents" the entire large bowel, in order to detect all pheno-and geno-typically different colorectal tumors. The present biomarker panel is positive in tumors independent of the patient's gender and age, as well as of tumor stage and location. Even though tumors of all clinico-pathological subtypes were represented among the biomarker panel methylation positive samples, the very few negative colorectal cancers were generally of the MSS phenotype and were BRAF wildtype, compatible with CIMP negative tumors [34]. However, it should be noted that in the present study we have used the test sets to establish the most optimal threshold for each biomarker assay (resulting in the highest specificity). From these values, we chose the highest and most conservative scoring threshold (percentage methylated sample -PMR value) and applied this to all biomarkers. Alternatively, by setting the threshold for each individual gene, the sensitivity could be increased without necessarily affecting the specificity.
In spite of several genome-wide epigenetic studies, only a few reports include DNA methylation data and potential subsequent epigenetic silencing of CNRIP1, FBN1, INA, and SNCA in cancer [10,[49][50][51][52][53]. Due to the exceptionally high methylation frequencies of these genes in colorectal tumors they are suitable as biomarkers, but the potential roles of the encoded proteins in tumor genesis remain unknown. Interestingly, mutations in FBN1, which is a member of the fibrillin family, are associated with the Marfan syndrome [49]. FBN1 methylation has previously been identified in prostate cancer cell lines [50] and the gene has also been shown to be epigenetically silenced in tumor endothelial cells [51]. Functional validation by RNA interference in endothelial cells pinpointed FBN1 as a negative regulator of cell growth and angiogenesis [51]. The expression of the synuclein alpha gene, SNCA, was recently shown to be regulated by methylation and decreased in the brain tissue of patients with Parkinsons's disease [52]. The gene has also been reported methylated in 38% of breast cancer tumors [53]. Interestingly, the majority of adjacent normal tissue samples were also found to be methylated, indicating a potential field defect that could help pinpointing geographical zones of increased breast cancer risk [53]. Such a field defect was not seen in the present study for SNCA, where all normal mucosa samples taken in distance from the colorectal cancers were unmethylated. In contrast, we observed a 40% promoter methylation frequency of CNRIP1 in the same sample group. The majority of normal mucosa samples from colorectal cancer free individuals (95%) were unmethylated in the CNRIP1 gene promoter. The methylation positive normal mucosa cells adjacent to the methylated cancer could potentially increase the detection of this aberration in stool samples, by increasing the total number of methylation positive cells shed into the lumen. In 2006 we identified CNRIP1 as a promoter methylation target in colorectal cancer by treatment of cell lines with 5-aza 2'deoxycytidine followed by expression microarray analyses [20,54]. By applying DNA methylation microarray analysis and subsequent methylationsensitive high resolution melting analysis colorectal tumor specific promoter methylation of CNRIP1 was recently confirmed [10]. Most of the adenomas (n = 12) and carcinomas (n = 55) analyzed were methylated [10] in agreement with the present report, underscoring the potential of CNRIP1 as a novel marker for early detection of colorectal tumors. Here we found that the DNA methylation of CNRIP1, FBN1, INA, and SNCA was associated with reduced or lost gene expression in cell lines, indicating that they might harbor a tumor suppressor function. Down-regulation in colorectal cancers compared with normal mucosa was also seen for all genes from in silico analyses, including MAL and SPG20, using the GeneSapiens website, which contains Affymetrix gene expression array experiments representing more than 500 colorectal cancers and 23 normal colon samples [55].

Conclusions
In the present study we identified and validated four novel methylated genes with high sensitivity and specificity for both colorectal cancers and adenomas. A combination of these and two of our previously identified biomarkers (MAL [21,22] and SPG20 [9]) provided an excellent biomarker panel. Early detection of colorectal cancer, at a stage where it is localized and curable will contribute substantially to reduce mortality due to the disease. Moreover, the present biomarker panel has been shown to also be positive in premalignant adenomas. Therefore, a test using this panel may also provide non-invasive detection of lesions prior to malignancy, thereby potentially reducing the incidence of colorectal cancer. The advantage of a marker panel compared to single marker analysis is obviously the higher sensitivity and specificity but also increased robustness. A single marker will be more prone to biological and/or technical failure than a combined panel. Finally, the benefit of a non-invasive epigenetic test compared to a genetic test is the simplicity of performance, which is expected to lower the costs, hence supporting it as appropriate for screening purposes. The presented biomarker panel seems highly suitable for development of a non-invasive test for early detection of colorectal tumors.

Additional material
Additional file 1: Supplemental tables.
Additional file 2: Figure S1. Summary of methylation-specific polymerase chain reaction (MSP) results in cancer cell lines. Red color indicates methylated sample, orange color indicates the presence of both methylated and unmethylated bands, green color indicates unmethylated sample, and grey color indicates samples that were not successfully amplified/analyzed. The upper part of the figure illustrates the results from colon cancer cell lines (n = 20) and the lower part of the figure from cell lines originating from various cancer tissues (n = 29).