There are only a few studies addressing the genome-wide methylation in colorectal carcinoma [9, 31–33]. Very recently Kim et al  (Feb 2011) and Oster et al  (Mar 2011) used the same commercially available Infinium methylation 27 arrays in CRC and identified differentially methylated sites in Korean and European population respectively. Oster's study used carcinoma and normal tissue from different individuals for methylation analysis, whereas ours used paired tumor and adjacent normal tissue from the same patient. This allowed us to eliminate inter-individual variation in our methylation analysis, which may be one reason why our study detected a larger number of differentially methylated genes in carcinoma. Kim's study compared the methylation in paired samples like ours, but they looked at gene expression in a different set of individuals. While they looked for effect of methylation on gene expression, they could not find statistically significant difference in the mRNA expression level between promoter hypermethylation group and hypomethylation group, whereas we were able to calculate correlation coefficients using paired data in every gene and found several significant correlations. The only other genome-wide methylation study in CRC addressing promoter CpG loci using commercially available array, used much lower density array (only 1505 CpG loci) .
Our study in south-east Asian population suggests that, in comparison to the normal colonic mucosa, the corresponding CRC tissue shows a large number of differentially methylated loci within the CpG islands close to the transcript start site of genes, indicating the role of DNA methylation in the pathogenesis of colorectal carcinoma. The results from our study not only confirm the findings from many previous candidate gene approach based studies, but we also report a large number of novel loci that show differential methylation in CRC.
We noticed the influence of sex on genome-wide methylation that is explained by the X-inactivation process, in which one of the two copies of genes on the X-chromosome in females is silenced. A similar finding was also recently reported by Liu et al. .
Laird et al.  has recently focused on the different statistical issues for methylation data. We applied different normalization methods and found considerable overlap between the results. Use of stringent criteria for selecting differentially methylated loci and the considerable overlap between the results from different analyses, the 2-level cross validation and finally the q-PCR validation in subset suggest that we detected the truly differentially methylated loci in CRC.
Recently Irizarry et al. used a Comprehensive High-throughput Array for Relative Methylation (CHARM) assay to show that most methylation alterations in CRC occur up to 2000 bp away from the CpG islands themselves . Because of the design of the chip used in the present study, we did not have the opportunity to look at the differential methylation at loci > 1500 bp away from the TSS. However, similar to results from Irizarry et al. , we also found that the hypomethylated loci were slightly more upstream than the hypermethylated loci.
The cross-validation results are very encouraging as a potential biomarker, but we have cross-validated only in colon tissues and not in circulating plasma DNA. In the future we would like to test the markers in an independent sample set of circulating plasma or serum DNA in CRC patients and healthy individuals. Recently He et al.  selected three methylation markers from the published literature and tested the practical use of those markers in peripheral blood sample from CRC patients. They found a sensitivity of 81% and a specificity of 90%. We had the advantage of profiling a very large number of CpG loci in paired CRC and normal colonic mucosa tissue, and our 2-level cross validation suggested that the four markers could be used as biomarkers with slightly better test characteristics.
Tanaka et al.  have recently applied an analytical strategy known as structural equation modeling to understanding methylation in CRC. Using a large database of over 800 samples, the authors were able to construct causality pathways of KRAS and BRAF mutations, as well as various phenotypes, on methylation of specific genes. This strategy was not feasible for our current study because of our smaller sample size and because we had not obtained information on KRAS and BRAF mutations. Nonetheless, it will be valuable for our planned future study with an expanded cohort.
Illumina's methylation assay has been compared to other platforms by others and has shown dependable results with the correlation ranging from 0.8 to 0.9 [32, 33, 38]. We also have validated the methylation data form Infinium methylation for 12 of the highly differentially methylated genes in our study and also found similar high correlations with Methyl Profiler assay (see Additional File 1 Table-S4 and Additional File 5 Figure S4). In another study, reproducibility tests of Infinium methylation platform was reported to have correlation greater than 0.98 between technical replicates . We are aware of the fact that Illumina's Methylation27 assay detects the methylation status of on average ~2 CpG sites per gene for most genes. However, for the genes for which there were multiple CpG loci on the array (e.g. ESR1 or DAB2IP), we found all of the loci to be differentially methylated in the same direction. We also validated Illumina's platform in the top-ranking genes by methyl profiler PCR array which is (a) not dependent on bisulfite conversion and also (b) provides an overall methylation status of the target region as opposed to single loci. This paper was focused mainly to look at DML in CRC. However, we have also explored the link between chromosomal abnormalities (copy number), methylation and gene expression. Regulation of gene expression is complex and is not dependant only on methylation status or copy number status. Using integration of molecular cytogenetics, genome-wide copy number and expression microarray profiling, Camps et al have demonstrated the effect of copy number on gene expression in CRC . To our knowledge, our study is the first one to comprehensively look at the genome-wide methylation, copy number and gene expression - all three together in primary CRC tissue. In our study, expression of a small proportion of genes was found to be correlated to methylation and another small proportion was correlated to copy number changes seen in CRC. Although methylation status of many loci could not explain the functional relevance to gene expression, these promoters methylation may be used clinically as biomarkers.