Association between MGMT Enhancer Methylation and MGMT Promoter Methylation, MGMT Protein Expression, and Overall Survival in Glioblastoma

The repair protein O6-methylguanine-DNA methyltransferase (MGMT) is regulated epigenetically, mainly by the methylation of the MGMT promoter. MGMT promoter methylation status has emerged as a prognostic and predictive biomarker for patients with newly diagnosed glioblastoma (GBM). However, a strong negative correlation between MGMT promoter methylation and MGMT protein expression cannot be applied as a rule for all GBM patients. In order to investigate if the DNA methylation status of MGMT enhancers is associated with MGMT promoter methylation, MGMT expression, and the overall survival (OS) of GBM patients, we established assays based on high-resolution melting analysis and pyrosequencing for one intragenic and three intergenic MGMT enhancers. For CpGs in an enhancer located 560 kb upstream of the MGMT promoter, we found a significant negative correlation between the methylation status and MGMT protein levels of GBM samples expressing MGMT. The methylation status of CpGs in the intragenic enhancer (hs696) was strongly negatively correlated with MGMT promoter methylation and was significantly higher in MGMT-expressing GBM samples than in MGMT-non-expressing GBM samples. Moreover, low methylation of CpGs 01–03 and CpGs 09–13 was associated with the longer OS of the GBM patients. Our findings indicate an association between MGMT enhancer methylation and MGMT promoter methylation, MGMT protein expression, and/or OS.


Introduction
Glioblastoma (GBM) accounts for approximately 45-50% of all primary malignant brain tumors in adults. GBMs have been defined as isocitrate dehydrogenase (IDH)-wildtype diffuse astrocytomas by the World Health Organization (WHO) [1]. With a median survival time of 15 months after diagnosis, prognosis of GBM is very poor. Standard therapy of newly diagnosed GBM patients consists of radiotherapy and concomitant therapy with the alkylating agent temozolomide (TMZ), followed by adjuvant treatment with TMZ [2]. TMZ causes DNA damage by the addition of methyl groups at the N7 and O6 positions of guanine and the N3 position of adenine residues [3]. However, response to TMZ critically depends on the expression of O6-methylguanine-DNA methyltransferase (MGMT), since it repairs alkylation lesions, preferentially methylation at position O6 of guanine [4].
MGMT is regulated epigenetically, mainly by the methylation status of the MGMT promoter [5]. DNA methylation is the most widely studied epigenetic modification. It plays a crucial role in gene regulation by affecting chromatin accessibility and thus gene transcription. Methylation of CpG dinucleotides (CpGs) in the promoter region commonly leads to transcriptional inactivation of genes [6]. The MGMT promoter contains a CpG island which consists of 98 CpGs and extends into the non-coding exon 1 (CpGs 74-83; NCBI Reference Sequence: NG_052673.1) [7]. Within the CpG island, a minimal promoter (88 bp, CpGs 50-62; −173 to −86 according to NCBI exon 1 coordinates) and an enhancer region (59 bp, CpGs 82-87; +40 to +98) have been identified [8]. Hypermethylation of the MGMT promoter, resulting in transcriptional silencing, has been associated with survival benefits from TMZ therapy [9]. Since this finding was confirmed in clinical trials [10][11][12], MGMT promoter methylation status has emerged as a predictive biomarker for the response to TMZ, in particular in the population of elderly patients with newly diagnosed GBM [13]. In addition, several studies have demonstrated the suitability of MGMT promoter methylation as a prognostic biomarker for GBM [14,15].
However, growing evidence suggests that a strong negative correlation between MGMT promoter methylation and MGMT protein expression does not apply to all GBM patients. There are tumors that express MGMT in spite of MGMT promoter methylation, and others that do not show MGMT expression although the promoter is unmethylated [16]. Promoter methylation is still assumed to play a major role in MGMT silencing. However, other mechanisms seem to affect the correlation between MGMT promoter methylation and MGMT protein expression as well [17], and under certain circumstances, even the overrule influence of MGMT promoter methylation.
We hypothesize that DNA methylation of MGMT enhancers is one of the factors playing a role in regulating MGMT expression in addition to MGMT promoter methylation. Enhancers are DNA regulatory elements that precisely control spatiotemporal patterns of gene expression [18]. In contrast to promoters, distal enhancers may be located upstream or downstream and even at a far distance from the transcription start site (TSS) of their target gene(s). Increasing evidence suggests that distal enhancers interact with promoters through long-range interactions via chromatin looping [19]. In general, enhancer activity may be affected by alterations in the enhancer sequence, including mutations and single nucleotide polymorphisms (SNPs), that alter chromatin accessibility and/or transcription factor binding [20]. In addition, the DNA methylation status of enhancers may have an impact on their activity. Enhancer regions are frequently depleted of CpGs and characterized by low DNA methylation levels [21]. However, by analyzing datasets for 58 cell types, Aran et al. found out that enhancers were gradually methylated and that enhancer methylation correlated stronger with gene expression than promoter methylation [22]. Very recently, Kreibich et al. proposed that DNA methylation of CpGs in enhancers plays a role in controlling the binding of transcription factors [23]. Studies indicate that aberrant enhancer methylation is a frequent event in various cancer types [24][25][26][27], including glioblastoma [28].
In the present study, we determined the methylation status of CpGs in distal MGMT enhancers in samples from 38 IDH-wildtype GBM, one IDH mutated GBM, one gliosarcoma, and for the commercial cell line T98G. We investigated if enhancer methylation is associated with MGMT promoter methylation, MGMT protein expression, and/or OS. We selected one intragenic (hs696) and two intergenic (hs737, hs699) enhancers from the VISTA Enhancer Browser [29] and an intergenic enhancer that was recently identified by Chen et al. [30]. For DNA methylation analysis of the MGMT promoter, we used a primer set from the literature [31], targeting CpGs 72-83 of the promoter. The DNA methylation status of the enhancers was determined by using methods developed in-house. We amplified the target regions by the polymerase chain reaction (PCR) and subjected PCR products to high-resolution melting (HRM) analysis and pyrosequencing (PSQ). Since HRM analysis provides information on the average methylation status across all CpGs in the PCR product, it was used for screening. In addition, HRM analysis was applied to obtain information on the occurrence of specific methylation patterns such as monoallelic methylation. The methylation status of individual CpGs was obtained by PSQ.

Samples and Cell Culturing
The sample set consisted of primary human tumor cell lines established from 40 glioma patients who underwent surgery between 2001 and 2020 at the Department of Neurosurgery, Kepler University Hospital, Neuromed Campus, as described previously [32], and the commercial GBM cell line T98G (ATCC, Manassas, VA, USA). The study was approved by the local Ethics Commission of the Faculty of Medicine at the Johannes Kepler University Linz (application number E-39-15). All patients signed a written informed consent form. Thirtyeight patients suffered from IDH-wild-type GBM (mean age at surgery 62.58 years; median 63.50 years), one patient (32 years) was diagnosed with IDH1-mutated (R132H) GBM, and one patient (43 years) was diagnosed with primary gliosarcoma (GS). Tumor-derived cell cultures were diagnosed according to the current version of the WHO classification of CNS tumors before 2022, therefore using the terms GBM (IDH wt) and GBM IDH mut. Overall survival was defined as the period between the date of surgery and death.
The cell lines were cultured in RPMI-1640, 7% fetal calf serum (FCS), and 1% glutamine without antibiotics (all Sigma-Aldrich, Schnelldorf, Germany) at 37 • C in a humidified 5% CO 2 incubator and harvested before reaching confluence between passages 2 and 6. The cell pellets were stored at −80 • C until DNA extraction. Relative MGMT protein expression, given as a ratio to MGMT overexpressing glioblastoma cell line GL80, was determined previously by Western blot analysis [33].

DNA Extraction and Bisulfite Conversion
Genomic DNA was isolated using the QIAamp DNA Blood Mini Kit (Qiagen, Hilden, Germany) following the manufacturer's protocol for cultured cells and quantified with a Qubit dsDNA BR Assay Kit using the Qubit 4 instrument (Thermo Fisher Scientific, Vienna, Austria). The DNA extracts were stored at −20 • C until PCR or bisulfite conversion. For bisulfite conversion of unmethylated cytosines, a EpiTect Fast Bisulfite Conversion Kit (Qiagen, Germany) was used according to the manufacturer's protocol. The converted DNA was quantified with a Qubit ssDNA Assay Kit (Thermo Fisher Scientific) and stored at −20 • C until PCR.
The DNA methylation status of CpGs 72-83 of the MGMT promoter was determined by using a primer set from the literature [31]. In total, ten assays targeting enhancer regions were developed in-house using PyroMark Assay Design Software 2.0.1.15 (Qiagen), with the primer sequences given in Table 1. For each primer set, the primer concentration, PCR mix as well as annealing and elongation temperature and time were optimized by using bisulfite-converted human methylated and non-methylated DNA (Zymo Research, Irvine, CA, USA) and a 50% mixture of both DNA standards. Each reaction was performed in a total volume of 20 µL, consisting of 1× PCR mix including EvaGreen HRM dye, forward and reverse primer, and 5 ng of bisulfite converted DNA. PCR was carried out using either a QuantStudio 5 instrument (Thermo Fisher Scientific) or a Rotor-Gene Q instrument with a 72-well rotor (Qiagen) for assay optimization and the Rotor-Gene Q instrument for sample analysis. The optimized PCR conditions are listed in Table S1. For all assays, the following HRM program was applied directly after final elongation: strand separation for 1 min at 95 • C, strand hybridization for 1 min at 40 • C and HRM with a ramp from 65 • C to 95 • C with 0.1 • C/hold (2 s) and gain optimization (70% before melt).
Each PCR run included bisulfite-converted human non-methylated and methylated DNA, 25%, 50%, and 75% mixtures thereof, and a no template control (2 µL nuclease-free H 2 O). All samples were analyzed in two independent PCR runs, in two replicates per PCR.

PSQ of PCR Products
PSQ was performed using the PyroMark Q24 Vacuum Workstation and PyroMark Q24 Advanced instrument with PyroMark Q24 Advanced Accessories, PyroMark Q24 Advanced CpG Reagents (all Qiagen), and Sepharose High-Performance beads (GE Healthcare; Thermo Fisher Scientific) according to the manufacturer s instructions.
If necessary, dispensation orders were adapted, e.g., to overcome sequencing frameshifts (Table S2). The DNA immobilization reaction was optimized in a range Cells 2023, 12, 1639 6 of 22 from 48-160 µL. The 120 µL immobilization reaction set-up contained 22.5 µL biotinylated PCR product (pool of both wells from one PCR), 1.5 µL Streptavidin Sepharose High-Performance beads (GE Healthcare,), 60 µL PyroMark Binding Buffer, and 36 µL of high-purity water (18.2 MΩ cm, ELGA PURELAB Ultra MK 2, Veolia, Celle, Germany). For the other immobilization reaction volumes, all of the components were up-or downscaled accordingly. The final immobilization reaction volume was 120 µL for assays targeting enhancers 1, 3, and 4. For assays A, B, and D of enhancer 2, the immobilization reaction volume was 160 µL, and for assay C, the immobilization reaction volume was 48 µL. DNA immobilization was performed under agitation for 10 min at 1400 rpm. The captured PCR product was denatured, washed, and the biotinylated strand was transferred into a Pyro-Mark Q24 Plate containing 20 µL of 0.375 µM sequencing primer in PyroMark Annealing Buffer. The plate was heated at 80 • C for 5 min and then transferred into the instrument, holding the PyroMark Q24 Cartridge loaded according to pre-run information provided by PyroMark Q24 Advanced software 3.0.0 (Qiagen). A representative pyrogram for each primer set is shown in Figures S1-S3.

Data Analysis and Statistics
The amplification and melting curves obtained by PCR-HRM were assessed and exported using Rotor-Gene Q Series Software 2.3.1 (Qiagen). The PSQ data were evaluated and exported with PyroMark Q24 Advanced software 3.0.0 (Qiagen). The exported data were analyzed and are presented graphically using R version 3.6.2 [37]. The R-packages used, including corrplot, ggplot2, polynom, rstatix, survival, and survminer, are listed in the Supplementary File R-packages.

Enhancer 1 Methylation
Screening the methylation status of CpGs 12-19 in enhancer 1 by HRM analysis suggested that the target region was methylated in all samples except GBM13 (Figure 4a). In addition, HRM analysis indicated that in none of the samples was the target region completely methylated. Multiple distinct melting transitions were obtained for three samples, including two GBM samples with a non-methylated MGMT promoter (GBM12, GBM24) and the commercial cell line T98G. HRM results hint at the presence of alleles with low heterogeneous methylation in addition to alleles showing mosaic methylation. The negative derivative of normalized HRM curves for GBM12 was similar to that obtained for the 25% standard, hinting at the transition of completely methylated alleles (Figure 4a).    . The PCR products of GBM12 (yellow) and T98G (blue) had two melting transitions in all four and two enhancers, respectively. DNA standards: non-methylated (~0%), methylated DNA (~100%), and 25%, 50%, and 75% mixtures thereof (dashed lines ranging from light gray (0%) to black (100%)). In the target regions of enhancer 4, the non-methylated standard turned out to be partially methylated (d). One representative measurement of two independent runs is shown. GBM01-38: IDH-wild-type GBM, GBMm01: IDH1-mutated GBM, GS01 gliosarcoma, and T98G: commercial GBM cell line.

Enhancer 2 Methylation
In total, 19 CpGs of enhancer 2 were targeted. However, four assays (A-D) had to be applied for determining the methylation levels of CpGs 05-08, CpGs 11-18, CpGs 24-27, and CpGs 37-39, respectively. Since amplification of the target regions turned out to be challenging, a special PCR mix that is not optimized for HRM applications had to be used. Due to the limited sample amount, samples GBM24, GBM36, and GBM37 were omitted from the analysis of CpGs 24-27 and CpGs 37-39.
GBM12 was the only sample resulting in two melting transitions, originating from non-methylated and completely methylated alleles, in all four target regions of enhancer . The PCR products of GBM12 (yellow) and T98G (blue) had two melting transitions in all four and two enhancers, respectively. DNA standards: non-methylated (~0%), methylated DNA (~100%), and 25%, 50%, and 75% mixtures thereof (dashed lines ranging from light gray (0%) to black (100%)). In the target regions of enhancer 4, the non-methylated standard turned out to be partially methylated (d). One representative measurement of two independent runs is shown. GBM01-38: IDH-wild-type GBM, GBMm01: IDH1-mutated GBM, GS01 gliosarcoma, and T98G: commercial GBM cell line.

Enhancer 2 Methylation
In total, 19 CpGs of enhancer 2 were targeted. However, four assays (A-D) had to be applied for determining the methylation levels of CpGs 05-08, CpGs 11-18, CpGs 24-27, and CpGs 37-39, respectively. Since amplification of the target regions turned out to be challenging, a special PCR mix that is not optimized for HRM applications had to be used. Due to the limited sample amount, samples GBM24, GBM36, and GBM37 were omitted from the analysis of CpGs 24-27 and CpGs 37-39.
GBM12 was the only sample resulting in two melting transitions, originating from non-methylated and completely methylated alleles, in all four target regions of enhancer 2 (Figure 4b shown for assay B). The HRM curves obtained for CpGs 05-08, CpGs 11-18, and CpGs 37-39 overlapped with those for the 50% standard, and the HRM curves for CpGs 24-27 overlapped with that of the 25% standard. In addition, the HRM data suggested CpGs 24-27 to be unmethylated in 19 GBM samples and T98G; CpGs 37-39 to be unmethylated in GBM02 and GBM09; CpGs 11-18 to be completely methylated in GBM03, GBM25, and GBM38; and CpGs 37-39 to be completely methylated in GBM03, GBM26, and GBM38.

Enhancer 3 Methylation
The assay allowed for DNA methylation analysis of eight CpGs (CpGs 15-22) of enhancer 3. According to the HRM analysis, GBM28 was the only sample in which the target region was unmethylated (Figure 4c). For the PCR products of five GBM samples, multiple distinct melting transitions were obtained. GBM05-06, GBM11, and GBM34 seemed to contain alleles showing low heterogeneous methylation in addition to alleles with higher mosaic methylation (Figure 4c). The melting curve obtained for GBM12 was identical to that of the 50% standard, indicating the presence of non-methylated and completely methylated alleles in a ratio of 50:50 (m/m).

Association between the MGMT Promoter and/or Enhancer Methylation
Next, we searched for correlations between the methylation status of CpGs located in the same but also in different regulatory elements (promoter/enhancer, two different enhancers). Correlation analysis was performed by including all IDH-wildtype GBM samples (GBM01-38) (Figure 5a), but we also stratified the samples by their MGMT promoter methylation status (Figure 5b,c).
CpGs targeted in enhancer 4, CpG 03 and CpG 09 showed the strongest negative correlations with the CpGs in the MGMT promoter. In general, the CpGs targeted in the other enhancers did not correlate with the CpGs in the promoter.
We also found positive correlations between the methylation levels of the CpGs located in different enhancers. The strongest correlations were obtained between the methylation levels of the CpGs located in enhancer 1 and enhancer 3 for samples with an unmethylated MGMT promoter (Figure 5b). The strongest positive correlations were found between the methylation status of CpGs in the MGMT promoter when all IDH-wild-type GBM samples were included (Figure 5a). Positive correlations, albeit generally weaker, were also obtained for samples with methylated MGMT promoters (Figure 5c).
Strong positive correlations were also found between the methylation status of CpGs located in enhancer 1 ( Figure 5). For samples with an unmethylated MGMT promoter, strong correlations were found between all individual CpGs (Figure 5b). In the case of promoter-methylated samples, the correlations between CpGs 12-16 were stronger than the correlations between the other CpGs targeted in enhancer 1 (Figure 5c). For enhancer 2, strong positive correlations were predominantly found between the methylation levels of CpGs 15-18, as well as between those of CpGs 37-39, independent of whether the samples stratified by their MGMT promoter methylation status (Figure 5b,c) or not (Figure 5a). In general, the methylation levels of CpGs targeted in enhancer 3 were also positively correlated with each other. The correlations were slightly stronger in samples with unmethylated MGMT promoters (Figure 5b) compared to those with methylated MGMT promoters (Figure 5c). The methylation levels of the CpGs located in enhancer 4 also showed positive correlations with each other. The highest number of correlations was found when all IDH-wild-type GBM samples were included in the correlation analysis ( Figure 5a); the lowest number of correlations was found when the correlation analysis was restricted to samples with a methylated MGMT promoter (Figure 5c).
By searching for correlations between the methylation levels of CpGs located in different regulatory elements, we found negative correlations between almost all CpGs in the MGMT promoter and almost all CpGs in enhancer 4, but only when the samples were not stratified by their MGMT promoter methylation status (Figure 5a). Among all of the CpGs targeted in enhancer 4, CpG 03 and CpG 09 showed the strongest negative correlations with the CpGs in the MGMT promoter. In general, the CpGs targeted in the other enhancers did not correlate with the CpGs in the promoter.
We also found positive correlations between the methylation levels of the CpGs located in different enhancers. The strongest correlations were obtained between the methylation levels of the CpGs located in enhancer 1 and enhancer 3 for samples with an unmethylated MGMT promoter (Figure 5b).

Association of the MGMT Promoter and Enhancer Methylation with MGMT Protein Expression
MGMT protein expression was only detected for samples with an unmethylated MGMT promoter, as well as for the commercial cell line T98G (Figure 3). GBM01 was the only sample that did not express MGMT although the MGMT promoter was unmethylated ( Figure 3). The enhancer methylation levels for GBM01 were similar to those determined for samples with an unmethylated MGMT promoter that did express the MGMT protein. We therefore assume, that MGMT expression in GBM01 was not impaired by DNA methylation and excluded GBM01 from further statistical analyses. In general, statistical analyses were restricted to IDH-wild-type GBM patients (GBM02-38), because for IDH-mutated GBM patients, gliosarcoma patients, and commercial GBM cell lines, only one representative of each was available.
In the MGMT-non-expressing GBM samples, MGMT promoter methylation was significantly higher than in MGMT-expressing GBM samples (Figure 6a). Significant differences were found for all individual CpGs as well as for the mean methylation of CpGs 72-83 (p < 0.001). In contrast, the methylation levels of the CpGs targeted in enhancer 3 ( Figure 6b) and enhancer 4 ( Figure 6c) were higher in the MGMT-expressing GBM samples than in the non-expressing GBM samples. For enhancer 3, we found significant differences for four individual CpGs (CpG 17, 18, 21, and 22) and for the mean methylation of all CpGs targeted (p ≤ 0.042). In the case of enhancer 4, significant differences (p < 0.001) were obtained for the mean methylation of the CpGs targeted by one and the same assay (assay A-D, respectively). In addition, each individual CpG of enhancer 4 was significantly more highly  The MGMT-expressing and -non-expressing samples did not differ significantly in their enhancer 1 and enhancer 2 methylation levels. Significant differences were neither found for the mean methylation of the regions targeted by one and the same assay nor for individual CpGs (Figure 6d

Association between the MGMT Promoter and Enhancer Methylation and Overall Survival
The MGMT promoter was found to be either unmethylated (methylation status of all CpGs ≤ LLOQ) or methylated (mean methylation ≥ 9.6%). Among the 37 IDH-wildtype GBM patients, OS was significantly higher for patients with a methylated MGMT promoter than for those with an unmethylated MGMT promoter (Figure 7a). For patients with a methylated MGMT promoter, the methylation levels of CpGs 75 (r = 0.49, p = 0.039), 78 (r = 0.56, p = 0.016), and 80 (r = 0.57, p = 0.014) were positively correlated with OS.
The methylation levels of the CpGs in enhancer 4 did not significantly correlate with OS, neither for the samples with a methylation status < 55% or a methylation status ≥ 55% (p ≥ 0.068). However, OS prediction based on CpGs 01-03 (assay A), CpGs 09-13 (assay C), or CpGs 01-03, CpGs 07-08 and CpGs 09-13 (assays A, B, and C) was more precise for patients with lower OS than prediction based on the promoter region (CpGs 72-83) (Figure 7a,b). This also holds true when higher cut-offs (10-25%) were used for promoter methylation. The methylation status of CpGs 19-22 (assay D) did not allow for distinguishing between patients with shorter and longer OS (p ≥ 0.196).
The methylation status of the CpGs targeted in enhancers 1-3 did not allow for OS prediction using cut-offs ranging from 5-90% for enhancer methylation (Figure 7c-e, p ≥ 0.289). The methylation status did not correlate with OS nor when the GBM patients were stratified by their MGMT promoter methylation status or not (p ≥ 0.053).

Discussion
We determined the methylation status of 61 CpGs, with 12 CpGs being located in the MGMT promoter, and 49 CpGs being located in enhancers that had already been associated with the MGMT gene. Primer sequences for targeting 12 (CpGs 72-83) out of 98 CpGs in the MGMT promoter were taken from the literature [31]. CpGs 82-83 belong to a 59 bp intrapromoter enhancer. The methods for DNA methylation analysis of the enhancers were developed in-house. Enhancers 1-3 are intergenic enhancers, located upstream of the MGMT gene. Enhancer 4 is an intragenic enhancer, located in the MGMT intron 2. The assay for enhancer 1 allowed for determination of the DNA methylation status of 8 (CpGs [12][13][14][15][16][17][18][19]  Our strategy was to amplify the target regions by PCR and to subject the PCR products to HRM analysis and subsequently to PSQ. The potential of analyzing the PCR products after HRM directly by PSQ has already been demonstrated not only for DNA methylation analysis [40], but also for SNP genotyping [41]. By providing information on the

Discussion
We determined the methylation status of 61 CpGs, with 12 CpGs being located in the MGMT promoter, and 49 CpGs being located in enhancers that had already been associated with the MGMT gene. Primer sequences for targeting 12 (CpGs 72-83) out of 98 CpGs in the MGMT promoter were taken from the literature [31]. CpGs 82-83 belong to a 59 bp intrapromoter enhancer. The methods for DNA methylation analysis of the enhancers were developed in-house. Enhancers 1-3 are intergenic enhancers, located upstream of the MGMT gene. Enhancer 4 is an intragenic enhancer, located in the MGMT intron 2. Our strategy was to amplify the target regions by PCR and to subject the PCR products to HRM analysis and subsequently to PSQ. The potential of analyzing the PCR products after HRM directly by PSQ has already been demonstrated not only for DNA methylation analysis [40], but also for SNP genotyping [41]. By providing information on the average methylation status across all CpGs in the PCR product, HRM was primarily applied for screening for non-methylated and completely methylated samples. In addition, the negative derivative of the normalized HRM curves yielded information on the occurrence of specific methylation patterns such as monoallelic methylation. In the case of monoallelic methylation, two sharp melting transitions are obtained, as for standard mixtures consisting of non-methylated and completely methylated DNA strands. Information on monoallelic methylation is of interest because it is a key mechanism of monoallelic expression, with the unmethylated allele commonly being expressed and the methylated one being silenced [42]. In contrast to HRM analysis, PSQ does not provide information on monoallelic methylation, with the methylation status for individual CpGs being the average methylation across all alleles in the sample.
We found monoallelic methylation of the MGMT promoter for the commercial GBM cell line T98G. The negative derivative of normalized HRM curves overlapped with that for the 50% standard, consisting of non-methylated and completely methylated DNA strands in a ratio of 50:50 (m/m). Our finding that the MGMT promoter shows monoallelic methylation explains why MGMT is expressed in T98G although the promoter is highly methylated. However, under certain circumstances, protein expression may be impaired in spite of the presence of an unmethylated allele. By investigating the methylation patterns of the MGMT promoter in GBM samples, Kristensen et al. observed monoallelic methylation in a number of samples [43]. Among them, several did express MGMT, as expected. However, some of these samples did not express MGMT, most probably due to hemizygous deletion of the MGMT locus, a frequent event in GBM [43].
For T98G, we found monoallelic methylation not only for the MGMT promoter but also for the enhancer 1 and enhancer 4 regions. In the case of sample GBM12, we detected monoallelic methylation for CpGs 05-08, CpGs 11-18, CpGs 24-27, and CpGs 37-39 of enhancer 2 and CpGs 01-03, CpGs 07-13, and CpGs 19-22 of enhancer 4. In contrast with T98G, the MGMT promoter was unmethylated in GBM12. GBM12 was the only sample for which multiple sharp melting transitions were obtained for each of the four enhancers. GBM12 showed the third highest MGMT expression of all samples analyzed.
Several studies have already been performed aimed at identifying individual CpGs in the MGMT promoter that are most applicable as potential biomarkers for the overall survival of GBM patients [7]. By determining the methylation status of 60 CpGs in 54 GBM samples, Everhard et al. found hypermethylation of six single CpGs (CpGs 27, 32, 73, 75, 79, and 80) and two CpG regions (CpGs 32-33 and CpGs 72-83) to be associated with low MGMT gene expression [44]. By using the HM-450K BeadChip, Bady et al. analyzed 22,25,31,[60][61][62]64,70,84, and 97 to identify associations with gene silencing and patient survival [45]. The highest negative correlation between methylation and gene expression and the strongest association with OS were found for CpGs 22, 25, 31, and 84. Growing evidence suggests that in particular CpGs in the intrapromoter enhancer are of regulatory and/or clinical relevance [46][47][48][49][50][51]. By using primer sequences from the literature [31], we targeted CpGs 72-83, with CpGs 82-83 being located in the intrapromoter enhancer. In 19 GBM samples and the IDH mutated sample GBMm01, the CpGs were unmethylated, in 19 GBM samples, the gliosarcoma sample GS01, and in T98G, they were methylated. MGMT promoter methylation has already been reported for gliosarcoma, but it seems to occur with a lower frequency than in GBM [52]. In most samples with a methylated MGMT promoter, we found the twelve CpGs rather heterogeneously methylated. Only four samples showed a significant difference between the methylation levels of CpGs 82-83 in the intrapromoter enhancer and those of CpGs 72-81, with CpGs 82-83 being more highly methylated. The methylation levels of CpGs 72-83 correlated strongly positively with each other, highlighting the relevance of DNA methylation in the promoter region. MGMT protein expression was only detected for samples with an unmethylated MGMT promoter (and for T98G, as described above). In samples that did not express the MGMT protein, the methylation levels were quite diverse, ranging from low to high methylation, indicating that even a low DNA methylation status of the promoter is sufficient to suppress MGMT expression.
Cut-off values for stratifying the GBM samples by their MGMT promoter methylation status are critically discussed in the literature [17,53]. For MGMT promoter methylation levels determined by PSQ, the cut-off value is most commonly set at 8% or 9% [54][55][56]. By using a cut-off of 8%, OS was significantly higher for patients with a methylated MGMT promoter than for those with an unmethylated MGMT promoter which is in line with previous studies [14,15]. GBM01 was the only sample that did not express MGMT although the MGMT promoter was unmethylated. With 7.4 months, the OS of patient GBM01 was rather short, corresponding to that of other patients with an unmethylated MGMT promoter that did express MGMT.
With respect to the enhancers investigated, we obtained the most interesting results for intergenic enhancer 2, located 560 kb upstream of the MGMT promoter, and intragenic enhancer 4, located in intron 2. Chen et al., who identified enhancer 2, found that activation of the enhancer positively correlated with MGMT expression [30]. By activating the enhancer in cell lines with low MGMT expression, MGMT expression increased, whereas deletion of the enhancer region in cell lines with high MGMT expression resulted in a drastic reduction in MGMT expression. By investigating the enhancer region in more detail, Chen et al. distinguished between two regions: "Del 1" at the 5' end of the enhancer, and "Del 2" at the 3' end and thus closer to the MGMT promoter. Deletion of "Del 1" resulted in a dramatic reduction in MGMT expression, whereas deletion of "Del 2" did not lead to a significant change in MGMT expression. When we developed assays for DNA methylation analysis of enhancer 2, we attempted to target both "Del 1", containing CpGs 01-25, and "Del 2", containing CpGs 26-46 of the enhancer. Finally, we established four assays (A-D), targeting 19 CpGs in total, with CpGs 05-08 (assay A) and CpGs 11-18 (assay B) being part of "Del 1", CpGs 37-39 (assay D) being part of "Del 2", and CpGs 24-27 (assay C) being located at the border of "Del 1" and "Del 2". CpGs 05-08, CpGs 11-14, and CpGs 37-39 showed similar methylation levels in the glioma samples investigated. However, CpGs 15-18 were significantly more highly methylated, and CpGs 24-27 were significantly less methylated. In most glioma samples, the CpGs in enhancer 2 were methylated, with the exception of CpGs 24-27 in nine samples and CpGs 37-39 in one sample(s). The methylation levels of CpGs 15-18 strongly correlated with each other, and a strong positive correlation was also found for CpGs 37-39, independent of whether the samples were stratified by their MGMT promoter methylation status or not. We did not find significant differences in enhancer 2 methylation levels between the MGMT-expressing and -non-expressing samples, neither for the mean methylation of the target regions nor for the individual CpGs. However, we found a significant negative correlation between the mean methylation of all CpGs targeted in enhancer 2 and the MGMT protein levels for the GBM samples expressing MGMT. The mean methylation of CpGs 5-8, 11-14, and 37-39 was also negatively correlated with MGMT expression levels. A significant negative correlation was also found for individual CpGs 05, 06, 07, 08, 11, 12, 13, 14, 37, and 39. Our results suggest that the DNA methylation levels of CpGs in both enhancer 2 regions, "Del 1" and "Del 2" [30], have an impact on enhancer activity and consequently on MGMT protein expression, whereas the methylation status of CpGs at the border of the two regions does not seem to play a role.
Application of four assays (A-D) developed in-house for the analysis of CpGs 01-03, CpGs 07-08, CpGs 09-13, and CpGs 19-22, respectively, in intragenic enhancer 4 (hs696) revealed that CpGs 19-22 were significantly more highly methylated in most samples than CpGs 01-03, CpGs 07-08, and CpGs 09-13 (p < 0.001). Interestingly, several samples exclusively with a methylated MGMT promoter showed low methylation (mean methylation < 25.0%) of CpGs 01-03, CpGs 07-08, and CpG 09-13, whereas high methylation (mean methylation > 75.0%) was found exclusively for samples with a non-methylated MGMT promoter. The methylation levels of the CpGs located in enhancer 4 correlated positively with each other. Most correlations were found when the samples were not stratified by their MGMT promoter methylation status. In addition, when the samples were not stratified, the methylation levels of almost all CpGs in enhancer 4 negatively correlated with the methylation levels of almost all CpGs in the MGMT promoter. The strongest negative correlations were found for CpG 03 and CpG 09 in enhancer 4. The methylation levels of enhancer 4 were significantly higher in the MGMT-protein-expressing GBM samples than in the -non-expressing GBM samples. This holds true not only for the mean methylation of the CpGs targeted by one and the same assay (assay A-D) but also for all individual CpGs. All our findings suggest that DNA methylation of enhancer 4 is associated with MGMT promoter methylation and MGMT protein expression. In addition, we found DNA methylation of CpGs 01-03 and CpGs 09-13 to be associated with OS. Patients in which these CpGs were methylated to a lower degree had longer OS than patients with a higher methylation status. Moreover, for patients with shorter OS, prediction based on the methylation status of enhancer 4 was more precise than prediction based on the promoter methylation status.
By applying in-house assays for intergenic enhancers 1 (hs737) and 3 (hs699), we found strong positive correlations between CpGs located in the respective enhancers. In the case of both enhancers, correlations were stronger for samples with an unmethylated MGMT promoter. For samples with an unmethylated MGMT promoter, we also found strong positive correlations between the methylation levels of CpGs located in enhancer 1 and those located in enhancer 3. Methylation levels of four individual CpGs (CpGs 17, 18, 21, and 22) in enhancer 3 and the mean of all CpGs targeted were significantly higher in the MGMT-expressing GBM samples than in the -non-expressing GBM samples; for the CpGs in enhancer 1, we did not find a difference between the MGMT-expressing and -nonexpressing samples. Neither for enhancer 1 nor for enhancer 3 did we find the methylation status to be associated with the OS of the patients.

Conclusions
We have provided DNA methylation levels for four MGMT enhancers that have not been subjected to DNA methylation analysis before. Our findings indicate that the methylation levels of the CpGs in an enhancer located in MGMT intron 2 (enhancer 4, hs696) is significantly negatively correlated with the methylation levels of the CpGs in the MGMT promoter. CpGs of enhancer 4 were significantly more highly methylated in the MGMTexpressing samples compared to the -non-expressing samples. Moreover, low methylation of CpGs 01-03 and CpGs 09-13 of enhancer 4 turned out to be favorable for the OS of GBM patients. For the CpGs in enhancer 2, located 560 kb upstream of the MGMT promoter, we found a significantly negative correlation between methylation status and MGMT protein levels for GBM samples with an unmethylated MGMT promoter, expressing the MGMT protein. The methylation levels of four CpGs (CpG 17, 18, 21, and 22) of enhancer 3 (hs699), located upstream of the MGMT promoter and downstream of enhancer 2, were significantly higher in the MGMT-expressing GBM samples than in the -non-expressing GBM samples. The methylation status of enhancer 1 (hs737), located upstream and at a greatest distance from the MGMT promoter, was neither associated with MGMT promoter methylation or MGMT expression nor with the OS of GBM patients.
Our findings suggest that enhancer methylation contributes to MGMT regulation and is a potential prognosticator for GBM patient survival. However, it remains to be elucidated if DNA methylation of the four enhancers targeted is associated with SNP genotypes, e.g., MGMT rs16906252, and/or clinicopathological characteristics, including the proliferation marker Ki-67, progression-free survival, and the response to TMZ in GBM.