In this study we evaluated CDKN2A methylation using pyrosequencing on a large cohort of 422 patients with colorectal cancer. Using both primary tumors and matched non-neoplastic tissues, we analyzed individual CpG sites and different amplification conditions. Additionally, we highlight the negative and independent prognostic effect of CDKN2A methylation on prognosis in patients with colorectal cancer using pyrosequencing.
Our results show acceptable levels (<10%) of background CDKN2A methylation with ≥35 PCR cycles; lowest levels were reached at 45 cycles. The predicted and observed degrees of methylation found after serial dilutions with methylated and non-methylated DNA show a tight correlation at these amplification conditions, while even low levels could be accurately detected using 45 PCR cycles. A high number of amplification cycles is often necessary for pyrosequencing. Since both biotinylated template strands and unincorporated biotinylated primers can be captured on the streptavidin-coated beads, a high number of PCR cycles ensures that the biotinylated primer does not itself act as an additional sequencing primer thereby interfering with the subsequent sequencing reaction .
Although the average percentage of methylation in non-neoplastic tissues was <10% across all CpG sites, in some cases it could reach >90%. This non-negligible methylation pattern suggests that the normal corresponding mucosa must be used as a control in the assessment of CDKN2A hypermethylation in colorectal cancers. In fact, nearly 10% of all cases showed greater methylation in the adjacent non-neoplastic regions than in the carcinoma. Not only has age-dependent methylation been recognized as an important physiological process but recent studies show further that CpG island methylation in normal colorectal mucosa may also be related to ethnicity, tumor location and intake of supplements such as folic acid [5, 8]. This indicates that methylation patterns in normal colorectal mucosa cannot be ignored supporting our decision to use normal tissue as a control in this study.
Although several different options were considered for setting a threshold value for methylation positivity, we chose to consider cases with at least 20% methylation difference between tumor and normal tissue as methylation positive. The number of positive cases, namely 20% of patients, falls within the range previously described . Since background methylation in both tumor and normal tissues may reach 10%, setting a cut-off at 20% ensures that only methylated cases be assigned as positive. In addition, since we included cases enriched for >70% tumor content, it is possible that a few samples containing sufficient non-neoplastic tissue may be misclassified as methylation negative, i.e., false-negatives.
Several study groups have addressed the issue of threshold values for methylation positivity using pyrosequencing. Vasiljevic and colleagues found an optimal cut-off of 35% for methylation in prostate cancers using data resampling and statistical methods . Several groups have assigned positivity to cases with a methylation density >15% [10–13]. In lymphoma, cut-offs for CDKN2A methylation positivity were based on receiver operating characteristics (ROC) curves and compared to the median and mean methylation levels ultimately categorized as negative, low, intermediate and high when <5%, 5-25%, 25-40% and >40% methylation was found, respectively . Others have used the mean and standard deviation as a basis for cut-off value determination for CIMP-related markers . These methods have advantages and drawbacks. Methods based on the mean and SD may be suboptimal since the presence of outliers, as was seen in our study, may have a considerable impact on skewing the distribution of the methylation data in normal tissues in particular. Cut-off scores derived after the analyses of entire cohorts may be disadvantageous in that they are not generalizable to other datasets. A cut-off score derived from ROC curve analysis is often advantageous as it may have the most clinically relevant value for a specific endpoint of interest, such as survival. It does nonetheless test the entire range of possible methylation values including those that may be irrelevant. Applying ROC curve analysis to our data here, we found an ‘optimal’ difference of 5% to be sufficient. This value, although statistically optimal is less compatible with the biological relevance.
In our series, CDKN2A methylation positivity correlated with more frequent right-sided disease, mucinous histology, tumor grade as well as with MSI, BRAF mutation and with KRAS mutation in the MSI setting only. This finding is in line with other studies showing the association of methylation with “classical” features of MSI and high-level CIMP . Poorer prognosis is found in patients with CDKN2A methylation positive tumors regardless of BRAF gene status. Additionally, unfavorable survival time was again particularly observed in patients with MSS disease. These results are in line with findings from Kim et al. using pyrosequencing (n = 131), Mitomi and colleagues using q-MSP (n = 151), Liang et al. using MSP (n = 84) cases and Maeda and coworkers using q-MSP (n = 90) showing either a negative prognostic effect of CDKN2A methylation in univariate and/or multivariate analysis [16–19].