Promoter DNA methylation and associated silencing is a frequent and early event in colorectal carcinogenesis 
. Some of the genes affected, such as MLH1, p16 and p14, clearly contribute physiologically to the neoplastic phenotype 
. The occurrence of liver metastasis leads to a poor clinical outcome in CRCs, and here we sought to determine the possible involvement of DNA methylation in the process. Generally, we found that methylation does not increase with increasing stage, confirming that it is an early event. Importantly, we did find substantial drift in methylation patterns in liver metastases compared to primary tumors, but the patterns at loci examined appeared more consistent with random flux rather than selection for specific genes.
When we looked at the differences in methylation between primary tumors with and without liver metastases, methylation levels of p14, TIMP3 and HPP1 progressively decreased from early-stage to late-stage disease. We have previously found that methylation of p14 and TIMP3 is the markers for predicting CIMP1 
. Thus, this consistent decrease of methylation in CRCs with liver metastasis likely represents the generally good prognosis of CIMP1 cancers which rarely progress to advanced disease 
. Depletion of TIMPs has been reported to abrogate normal apoptotic programs, enhance primary tumor growth and angiogenesis, invasiveness, and metastasis and possibly contribute to all stages of malignant progression including metastasis 
. Our data are not consistent with a major role for TIMP3 in CRC metastasis. It is possible that other members of the TIMP family such as TIMP1 and TIMP2 might be more important for the liver metastatic process in CRCs 
Overall, we quantitatively compared the methylation status of 21 genes (13 candidates and 8 from the microarrays) between paired primary and liver metastasis lesions. Of these, only MGMT methylation was consistently higher in the liver metastases than primary tumors. Of 16 pairs studied, five (31%) showed significantly higher MGMT methylation at the metastatic site. Of these five tumor pairs, four pairs demonstrated MGMT methylation at both sites (primary and liver metastatic tumors) with an increase in methylation density. Increased density of methylation could be explained by multiple different factors – increased proportion of methylated cells, switch from monoallelic to biallelic methylation or even differences in the degree of normal cell contamination of the tumor samples. Our data do not allow us to distinguish these possibilities and a larger series with more detailed analysis is needed to confirm our results and address the issue.
MGMT protein stoichiometrically repairs O6
-alkylG-DNA adducts 
. Inactivation of MGMT by promoter-methylation can lead to G to A transition mutations in several genes, including KRAS 
. Thus, MGMT methylation could be associated with the metastatic process by increasing the rate of mutations. However, this has not yet been convincingly demonstrated in CRCs. Park et al. have reported that MGMT methylation in patients with gastric carcinoma is significantly associated with lymph-node metastasis, tumor stage and disease free survival 
. However, another study showed significant association between MGMT methylation and improved overall survival in diffuse large B-cell lymphoma 
. Thus, the relationship between MGMT methylation and metastasis or tumor prognosis might be tissue specific, or possibly coincidental.
Our genome-wide analysis of hypermethylated genes at the liver metastatic tumor revealed that 7.4% (range, 4.7% to 10.9%) of the genes showed hypermethylation in the metastatic tumors and 1.3% was commonly hypermethylated among three patients. These numbers are quite large at face value, but when we validated the data by bisulfite-pyrosequencing, a change in methylation density was the explanation in most cases. One additional clue to explain this finding came from an analysis of resection time differences between the primary and metastatic lesions. Thus, the percentage of hypermethylated genes at liver metastasis was significantly higher in metachronous metastasis than in synchronous metastasis. In one patient, the time between surgery for the primary tumor and the liver metastasis was 46 months and 10.9% of genes analyzed using MCAM showed differential hypermethylation at the liver metastatic tumor. MCAM data in a patient with synchronous metastasis revealed 4.7% differential hypermethylated genes. Given that population doubling (reflected by patient age) is a prime determinant of methylation in normal and neoplastic colon, 
our data could be explained by continued accumulation of methylation at the metastatic site. Overall, looking at methylation frequency, we find few differences between primary tumors and liver metastases, suggesting that aberrant DNA methylation is a very early event and that tumor cells acquire methylation changes before progression to liver metastasis. We cannot exclude the possibility that a few rare genes are highly selected for during the process of metastasis, but discovering these will require whole-genome methylation analysis technology that is more quantitative than what is currently available.
In summary, our results indicate that methylation frequency between primary tumors and matched liver metastasis is similar, suggesting that tumor cells acquire methylation changes before progression to liver metastasis. While we cannot rule out rare consistent changes, it appears that DNA methylation frequency is very stable over time in CRC.