Reported frequencies of MGMT promoter methylation in subgroups of gliomas vary considerably, as shown in Additional file 1: Tables S1 and S2. Various methylation detection methods and different primer sets and threshold values have been used. In the present study, we report MGMT promoter methylation frequencies in gliomas determined by qMSP (low-grade gliomas 26% and high-grade gliomas 37%) as well as by pyrosequencing (low-grade gliomas 97% and high-grade gliomas 55%). It should be noted that the CpG sites interrogated in the qMSP and pyrosequencing assays are only partially overlapping (Additional file 1: Figure S1), and caution should therefore be made when directly comparing the results. The MGMT promoter is typically reported methylated in 30-60% of glioblastomas  and in 30-90% of low-grade gliomas [12, 18]. Compared to these reports, our qMSP methylation frequencies are in the lower range whereas the pyrosequencing methylation frequencies are in the upper range.
The vast majority of previous studies of MGMT promoter methylation in gliomas have used gel based methylation-specific PCR (MSP), which is a qualitative and time-consuming method. The manual methylation scoring based on interpretation of gel band intensities will vary in stringency level between labs, which, in addition to the use of different primers, may in part explain some of the observed difference in results observed from MSP based studies. This is exemplified by two works using the same primer sets but reporting very different methylation frequencies of 23% and 44% in newly diagnosed glioblastoma samples [19, 20]. In contrast to MSP, qMSP is a quantitative, standardized, high-throughput method which is easy to perform and the results are easy to evaluate. Thus, the method is more suitable for use in routine testing. To detect methylation by amplification of methylated alleles in MSP and qMSP, all CpG sites on the same DNA strand covered by the primers have to be methylated. Compared with traditional MSP, qMSP is even more conservative as it includes a methylation-specific probe and thereby typically covers more CpG sites that all have to be methylated. This may lower the sensitivity of the assay, but more importantly, increases the specificity, as underscored by Parella et al. who analyzed MGMT promoter methylation using both MSP and qMSP assays. In glioma samples the two methods showed good concordance, whereas the results from normal brain samples demonstrated that MSP may hold a higher risk of false positive results . In the present study, we used two overlapping qMSP assays and only scored samples as methylated if they had a positive PMR value in both assays. The results from the two assays were generally overlapping. Of the 134 analyzed gliomas samples only 16 of them had conflicting methylation status from the qMSP assays. The PMR-values seemed to be somewhat higher in one of the qMSP assays. This is probably a result of the inclusion of different CpG sites in the two assays (Additional file 1: Figure S1). The conservative nature of the qMSP assay may explain why our qMSP methylation frequencies are in the lower range of previously reported MSP results [11, 12, 18]. One of the primer/probe sets used in the present study corresponds to the assay used by Parrella and coworkers [13, 21]. The second qMSP assay is modified after a recent publication by Rivera et al.  who found MGMT promoter methylation in 24% of GBM patients, which is similar to the methylation frequency found by us.
As expected, the MGMT promoter methylation frequencies as measured by pyrosequencing were found to be in the upper range compared to previous MSP based findings. In contrast to MSP and qMSP, the pyrosequencing technique is able to detect low levels of methylation because methylation in each CpG site is measured independently of the methylation status in surrounding CpG sites. Indeed, our methylation frequency in high-grade gliomas (55%) is highly concordant with the GBM methylation frequency recently determined in a pyrosequencing work by Dunn et al. (53%) .
Choice of threshold values for scoring samples as methylation positive or not may also explain some of the differences observed in reported methylation frequencies. Ideally, the threshold value should be determined using a test series of a large number of normal tissue samples as well as tumor samples. The threshold value can thereafter be chosen to give a high sensitivity (with the risk of producing false positives) or a high specificity (with the risk of failing to identify all positive cases as such). We have used high-quality DNA extracted from fresh frozen tissue for all methylation analyses. In some neurooncology centers sampling of fresh frozen tissue is not a standard procedure, hence formalin-fixed paraffin embedded (FFPE) tissue is a frequently used DNA source. All MGMT promoter fragments amplified in the present study are short (qMSPs 83-119 bp and pyrosequencing ~100 bp) and will most likely be amplifiable also in DNA extracted from FFPE tissue. Lacking access to normal brain tissue, we used four benign meningioma samples to set the threshold values for scoring samples as methylation positive. The threshold values were determined so that all the meningiomas were scored as methylation negative. These benign tumors showed little (pyrosequencing, mean methylation range 1.39-1.55%) to no (qMSP, PMR 0% in both assays) MGMT promoter methylation, resulting in low threshold values, thus supporting the assumption that meningiomas are suitable alternatives to normal tissue samples for threshold determination. However, this should be confirmed by validation studies in independent sample series. Brain tissue from surgery in epileptic patients is an alternative to the meningioma tissue for establishing cutoff values.
In accordance with previous reports, our results show that the overall survival for patients with a methylated MGMT promoter is better than for patients with an unmethylated promoter [10, 20, 23, 24]. The observed difference at 24 months was significant based on the pyrosequencing results but only borderline significant based on the qMSP results. The log-rank test results indicate that both methylation detection methods are able to identify primary glioblastoma patients with a somewhat better prognosis. However, the survival curve differences are more distinct when using the pyrosequencing based methylation status, implying that this is the better method to use for estimating the prognosis. These results are in line with the observation that the patients with non-concordant methylation findings (unmethylated by qMSP and methylated by pyrosequencing) showed a trend towards better survival than patients with unmethylated MGMT promoter by both methods (Figure 1C). The last mentioned finding should, however, be validated in an independent sample set. Nevertheless, based on the observations done here, it could be argued that a primary glioblastoma should be regarded MGMT promoter methylated if the pyrosequencing result is positive. Because all glioblastoma patients receive TMZ as part of the Stupp regimen, the methylation status of the MGMT promoter does not change the therapeutic regime today. Nonetheless, it is a prognostic marker  of clinical interest and may be relevant for evaluation of pseudoprogression . It is also interesting that, independent of the method used, the methylation level (amount of methylation) observed in methylation positive low-grade gliomas is low compared to the level observed in methylation positive high-grade gliomas. This has not been reported previously and may in part explain the large difference in methylation frequency as assessed by qMSP and pyrosequencing in low-grade gliomas. The clinical relevance of this finding remains to be determined and the data should be validated in an independent data set. Nevertheless, the overall survival analysis, which includes GBM with a low methylation level, suggests that pyrosequencing is the better method for predicting prognosis in primary GBM patients. This may also suggest an advantage of a low methylation level in low-grade gliomas. There are not many studies reporting MGMT promoter methylation frequencies in large series of low-grade gliomas. However, two studies analyzing 68 and 185 low-grade gliomas report methylation frequencies of 93% and 81%, respectively, using the same MSP primers in a nested two-stage approach [18, 25]. These frequencies match our frequency (97%) detected by the sensitive pyrosequencing approach. However, other studies with smaller sample series report lower frequencies in the range 40-50% [26–29] when analyzed by conventional MSP.
We found a significant association between MGMT promoter methylation and reduced gene expression, regardless of methylation detection method and gene expression assay used. Based on this, one could suggest that the gene expression level might be analyzed instead of promoter methylation. However, the gene expression level in methylated and unmethylated samples shows considerable overlap (Figure 2) which may be due to lack of a linear relationship between the region analyzed for promoter methylation and gene expression. The most commonly analyzed region in the MGMT gene promoter covers 9 of the totally 97 CpG sites in the promoter, and it has been suggested that methylation in some specific CpG sites correlates better with reduction in gene expression level than analysis of the common MSP region (Additional file 1: Figure S1) . On the other hand, factors such as contamination with normal cells and loss of one MGMT allele may also influence the detected gene expression level. For a subset of the samples with gene expression data and methylation status, we had access to karyotypic and/or CGH data [16, 17] (author H-S. S. Dahlback, unpublished data). Figure 3 illustrates that the MGMT gene expression seemed to be affected by loss of the 10q26 chromosome band. Interestingly, but not unexpectedly, the loss of this chromosome band seemed to have a larger impact on gene expression in samples with a low methylation level (mean methylation by pyrosequencing < 20%) compared to highly methylated samples. However, these observations are based on results from small groups and should be tested in a larger dataset.
It is important to keep in mind that it is the MGMT protein that counteracts the effect of TMZ by removing methyl adducts at the O6-position of guanine. A recent study using human tumor cell lines derived from glioblastomas and other tumors concluded that the response to TMZ is better predicted by MGMT protein expression than by promoter methylation status . However, although cancer cell lines are useful models for the in vivo situation, findings should be validated in patient sample series, and so far immunohistochemical analyses of the MGMT protein level in human tumor samples have been inconclusive when correlated with patient outcome .