Epigenetic silencing of genes, mostly mediated by aberrant DNA methylation, is a mechanism of gene inactivation in patients with colorectal cancer . Among the loci that can undergo aberrant methylation in colorectal cancer, one subgroup appears to become aberrantly methylated as a specific group [2, 4], a phenomenon called the CpG island methylator phenotype (CIMP). Results in the literature on CIMP status as a prognostic factor for colon cancers remain controversial. These conflicting results could result from differences in patient cohorts, samples (the use of frozen or FFPE tissues), analytical techniques, methylation markers (different genes or the same genes but screened at different sites), primer sequences and criteria for CIMP. In our study, we chose to focus on the impact of the pre-analytical phase. A variety of assays to measure DNA methylation have been developed for FFPE and frozen tissues many of which rely on the bisulfite conversion of unmethylated cytosines from tumour tissue into uracils. However, the efficacy of sodium bisulfite treatment and the measurement of methylation levels in FFPE samples in one hand and in frozen samples on the other hand have never really been compared and evaluated.
In this study, we assessed the quality of bisulfite conversion as well as methylation levels for LINE-1, MLH1 and MGMT markers by pyrosequencing from 40 pairs of FFPE and frozen samples. Pyrosequencing is particularly useful because it provides resolution at the individual nucleotide level and includes a conversion control for each analysis. Assays targeting CpG islands of LINE-1, MLH1 and MGMT were chosen since aberrant methylation of this retro-transposon and these genes is implicated in colon cancers. LINE-1 (long interspersed nucleotide element-1) is a retro-transposable element of DNA that is present in 15% of the human genome. It is a surrogate marker of genome-wide DNA methylation [18, 22, 23]. It is frequently hypomethylated in human cancers [24, 25]. In contrast, the two genes MLH1 (the human homolog of the E. coli DNA mismatch repair gene mutL) and MGMT (O6-Methylguanine DNA methyltransferase) are hypermethylated in a number of different cancers, including colorectal cancer [26–28].
We first demonstrated that bisulfite-to-bisulfite standard deviations of methylation levels (mean 1.3%) and PCR/pyrosequencing run-to-run standard deviations of methylation levels (mean 0.9%) were low and acceptable. We chose to carry out the assays on frozen tissues to avoid the potential impact of DNA degradation from FFPE tissues. The results obtained for LINE -1 can be extrapolated to any other validated methylated markers. These assays allowed us to consider that our process of bisulfite conversion and PCR/pyrosequencing is reproducible and to establish an arbitrary threshold of 6.0% beyond which the difference in methylation value was not due to variability in the measurements.
To assess the quality of sodium bisulfite conversion and the subsequent PCR/pyrosequencing assays, using the LINE-1 marker, we evaluated conversion of 40 FFPE DNAs extracted using a routine method. The results were neither satisfactory nor reliable since only 15 cases showed valid control of conversion after the first analysis and 22 cases (55.0%) after combining the two analyses. We consequently decided to use a dedicated kit to re-extract DNA from FFPE tissues and to extend the assays to the two other markers. All the samples were evaluated using the LINE-1 and MGMT markers: only 27 cases (67.5%) showed good control of bisulfite conversion with the LINE-1 marker and 22 cases (55.0%) with the MGMT marker in the first assay. In contrast, with the MLH1 marker, in 11 cases amplification failed, possibly because of the length of the MLH1 amplicon (181 pb). Among the 29 amplifiable samples, 23 (57.5% of the 40 cases) showed valid control of conversion in the first analysis. Furthermore, the results were not the same in the second analysis. It is thus clear that the extraction phase should not be conducted using a routine process and must comprise a supplementary step of heating to 90°C for 1 h to improve bisulfite conversion. Even with this dedicated method, the rate of satisfactorily converted samples was much lower for FFPE than for frozen samples. In contrast, the proportion of satisfactorily converted samples in the first assay on frozen samples was 95.0% with LINE-1, 97.4% with MLH1 and 87.2% with MGMT. The results were reliable and similar after the second analysis. Furthermore, we did not encounter the problem of failed amplification (due to amplicon length) found with the MLH1 marker on FFPE tissues. DNA derived from FFPE is an extremely valuable source of material for retrospective studies, but is often highly degraded. When PCR-based methods are used to study DNA methylation changes, it is necessary to modify the DNA with sodium bisulfite to preserve the DNA methylation information of the original template, and this treatment may further damage the DNA. Fragmentation of FFPE DNA is a real drawback, particularly in DNA methylation studies based on methylation-independent PCR: the design of the primers must be conducted with the highest score, primers have to hybridize out of the CpG islands, and amplicon length has to be limited. It is highly required to respect these constraints when using pyrosequencing which is the only technique allowing a real quantification of methylation.
In our study, FFPE DNAs (n°30 and 36), showed invalid control of conversion with MLH1 and good control with LINE-1 and MGMT. In addition, none of the FFPE DNAs showing invalid control of bisulfite conversion with LINE-1 (10 cases) showed invalid conversion with MLH1, while in three cases (cases n°3, 21 and 37) the results coincided with those of MGMT. As the efficacy of bisulfite conversion appeared to be heterogeneous all along the DNA and variable according to the analysed marker, we think that percentages of bisulfite conversion differ from one cytosine to another due to residual cross-linkings (even with the 90°C heating step) leading to non-reproducible results on FFPE samples with the three markers.
Thus, even when extraction methods dedicated to FFPE tissue were used, problems occurred with bisulfite conversion. The importance of successful bisulfite conversion was underlined by a panel of experts who reported that incomplete conversion of DNA was the major cause of false-positive results in methylation analysis . We chose not to evaluate the results of the methylation levels obtained from FFPE DNA extracted using the classical method because the number of unsatisfactory conversions was too high, but it was clear that poor bisulfite conversion led to overestimation of methylation (Additional file 3: Figure 1). In order to compare the methylation levels of cryo-preserved and FFPE samples (DNA extraction with the dedicated kit), we established an arbitrary threshold (6.0%) beyond which differences in the methylation value were not due to variability in the measurements. We demonstrated that with LINE-1 eight pairs (27.6%), with MLH1, four pairs (15.4%), and with MGMT, eight pairs (25.8%) displayed significant differences in the methylation level. These deviations in methylation levels between matched FFPE and cryo-preserved samples cannot be due to differences in tumour cellularity as we checked that they were slight. In a previous work (data not published) we evaluated the variations in methylation levels induced by variations in tumor cellularity using the LINE-1 marker, and we observed a maximal deviation of 3.6% of methylation for a cellularity difference of around 20% from the same sample. This argument is supported by the study published by Irahara et al.  who compared average methylation values for LINE-1 in macrodissected colon cancers with those for matched Laser Capture Microdissection specimens providing a pure collection of tumor cells. They found no substantial effects of contaminating normal cells on LINE-1 methylation.
We found no pairs that showed a difference in methylation level with all three markers and we were unable to establish a trend for the differences in methylation levels in FFPE versus frozen samples. Clearly these deviations in methylation levels differentially impact according to the analyzed marker. For MLH1 and MGMT markers, a sample is considered as methylated whatever it displayed 20% or 50% of methylation. But for the LINE-1 marker, several subgroups of methylation can be distinguished and variations of methylation shown here can be critical for the creation of these subgroups.
Our study is the first to compare the results obtained for DNA extracted from FFPE and frozen tissues to assess the feasibility of DNA methylation analysis using pyrosequencing. Some authors maintain that sodium bisulfite treatment is sufficiently precise and shows good reproducibility on FFPE samples leading to reliable assessment of methylation levels [19, 31]. Nevertheless, these authors did not perform comparative studies with frozen material and used techniques such as Methylight (real time PCR), SMART-MSP (Sensitive Melting Analysis after Real Time PCR - methylation-specific polymerase chain reaction), MS-HRM (Methylation Sensitive- High Resolution Melting), without a built-in measure to check the completeness of bisulfite conversion (conversion control). Furthermore, to confirm that FFPE tissue can be effectively used for high-throughput DNA methylation analysis, Ogino et al.  used the alternative method of protein expression by immunohistochemistry, which is a surrogate indicator of DNA methylation.
Looking ahead, there is another source of variability that should be investigated: bisulfite conversion methods which vary according to laboratories. This point reinforces the need for standardization in this domain.