Prognostic Value of Three Different Methods of MGMT Promoter Methylation Analysis in a Prospective Trial on Newly Diagnosed Glioblastoma
et al. (2012) Prognostic Value of Three Different Methods of MGMT Promoter Methylation
Analysis in a Prospective Trial on Newly Diagnosed Glioblastoma. PLoS ONE 7(3): e33449. doi:10.1371/journal.pone.0033449
Prognostic Value of Three Different Methods of MGMT Promoter Methylation Analysis in a Prospective Trial on Newly Diagnosed Glioblastoma
Arne Christians 0
Christian Hartmann 0
Axel Benner 0
Jochen Meyer 0
Andreas von Deimling 0
Michael Weller 0
Wolfgang Wick 0
Markus Weiler 0
Javier S. Castresana, University of Navarra, Spain
0 1 Clinical Cooperation Unit Neuropathology, German Cancer Research Center (DKFZ) , Heidelberg, Germany , 2 Clinical Cooperation Unit Neurooncology, German Cancer Research Center (DKFZ) , Heidelberg, Germany , 3 Division of Biostatistics, German Cancer Research Center (DKFZ) , Heidelberg, Germany , 4 Department of Neuropathology, Institute of Pathology, Heidelberg University Hospital , Heidelberg, Germany , 5 Department of Neurooncology at the National Center for Tumour Diseases, Heidelberg University Hospital , Heidelberg, Germany , 6 Department of Neurology, University Hospital Zurich , Zurich , Switzerland , 7 Department of General Neurology, Hertie Institute for Clinical Brain Research, University of Tu bingen , Tu bingen, Germany , 8 Department of Neuropathology, Institute of Pathology, Hannover Medical School (MHH) , Hannover , Germany
Hypermethylation in the promoter region of the MGMT gene encoding the DNA repair protein O6-methylguanine-DNA methyltransferase is among the most important prognostic factors for patients with glioblastoma and predicts response to treatment with alkylating agents like temozolomide. Hence, the MGMT status is widely determined in most clinical trials and frequently requested in routine diagnostics of glioblastoma. Since various different techniques are available for MGMT promoter methylation analysis, a generally accepted consensus as to the most suitable diagnostic method remains an unmet need. Here, we assessed methylation-specific polymerase chain reaction (MSP) as a qualitative and semi-quantitative method, pyrosequencing (PSQ) as a quantitative method, and methylation-specific multiplex ligation-dependent probe amplification (MS-MLPA) as a semi-quantitative method in a series of 35 formalin-fixed, paraffin-embedded glioblastoma tissues derived from patients treated in a prospective clinical phase II trial that tested up-front chemoradiotherapy with dose-intensified temozolomide (UKT-05). Our goal was to determine which of these three diagnostic methods provides the most accurate prediction of progression-free survival (PFS). The MGMT promoter methylation status was assessable by each method in almost all cases (n = 33/35 for MSP; n = 35/35 for PSQ; n = 34/35 for MS-MLPA). We were able to calculate significant cut-points for the continuous methylation signals at each CpG site analysed by PSQ (range, 11.5 to 44.9%) and at one CpG site assessed by MS-MLPA (3.6%) indicating that a dichotomisation of continuous methylation data as a prerequisite for comparative survival analyses is feasible. Our results show that, unlike MS-MLPA, MSP and PSQ provide a significant improvement of predicting PFS compared with established clinical prognostic factors alone (likelihood ratio tests: p,0.001). Conclusively, taking into consideration prognostic value, cost effectiveness and ease of use, we recommend pyrosequencing for analyses of MGMT promoter methylation in high-throughput settings and MSP for clinical routine diagnostics with low sample numbers.
Funding: The UKT-05 trial was supported by the University of Tu bingen Medical Center, Tu bingen, Germany. The present study was supported by the Hertie
Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: M. Weller has been a consultant to, and has received honoraria from MSD (former Schering-Plough, one company which manufactures
temozolomide) and Oncomethylome Sciences. W. Wick has been a consultant to, and has received honoraria from MSD, Roche and Eli Lilly. All other authors have
declared no conflicts of interest. This does not alter the authors adherence to all the PLoS ONE policies on sharing data and materials.
Assessment of the methylation status of the O6-methylguanine-DNA
methyltransferase (MGMT) gene promoter in malignant glioma has
become one of the most requested molecular assays in clinical
neuro-oncology. Since the landmark study by Hegi et al. 
numerous clinical trials in glioblastoma have confirmed that
hypermethylation of the MGMT promoter serves as a strong
prognostic factor for progression-free survival (PFS) and overall
survival (OS) [1,2,3,4,5,6]. The MGMT gene encodes a
ubiquitously expressed suicide DNA repair enzyme that counteracts the
normally lethal effects of alkylating agents by removing alkyl
adducts from the O6-position of guanine . O6-alkylated guanine
causes base mispairing and double-strand breaks, thus inducing
apoptosis and cell death . Due to this DNA repair activity, the
MGMT protein is believed to provide resistance against cytotoxic
effects of alkylating agents . This therapeutically disadvantageous
protective effect is thought not to be present when MGMT is
epigenetically silenced through promoter methylation as observed in
many human cancers including glioblastoma, thus rendering cells
more sensitive to alkylating drugs. Due to the prognostic and
predictive role of the MGMT promoter status for patients suffering
from malignant glioma, promoter methylation of this gene is
commonly assessed both in clinical trials and routine diagnostics .
Several diagnostic methods are available for promoter
methylation analysis: The most commonly used technique is
methylationspecific polymerase chain reaction (MSP) , a non-quantitative
method established for MGMT by Esteller et al. . Other
techniques of methylation analysis include real-time quantitative
MSP (RT-MSP) , methylation-specific multiplex
ligationdependent probe amplification (MS-MLPA) , bisulfite
sequencing , combined bisulfite restriction analysis (COBRA)
, pyrosequencing (PSQ) , SIRPH (SNuPE ion pair-reverse
phase high-performance liquid chromatography) , and others.
Alternative methods of determining the MGMT status of a tumour
include quantification of mRNA expression by quantitative reverse
transcription polymerase chain reaction (qRT-PCR) , protein
detection by immunohistochemistry (IHC) [19,20], and assessment
of MGMT activity , rather than promoter methylation
analysis. Despite this variety of available techniques, a generally
accepted consensus as to the most suitable method of assessing
MGMT promoter methylation in glioma tissues has not been
achieved so far, neither for the requirements of large clinical trials
nor for routine diagnostics [10,22].
In the present study, we analysed formalin-fixed,
paraffinembedded (FFPE) tumour specimens derived from patients with
newly diagnosed glioblastoma who were treated up-front
according to a dose-intensified TMZ-containing chemoradiotherapy
protocol within a prospective clinical phase II trial (UKT-05) .
We compared three different assays for MGMT promoter
methylation analysis and included MSP as a qualitative and
potentially semi-quantitative, PSQ as a quantitative and
MSMLPA as a semi-quantitative method to determine which of these
methods would predict clinical outcome most reliably.
Analyses of MGMT promoter methylation by MSP, PSQ
The main goal of this study was to compare three different
methods of MGMT promoter methylation assessment with regard
to their respective value of predicting clinical outcome. To this
end, we investigated methylation of the MGMT promoter by MSP,
PSQ and MS-MLPA in 35 FFPE glioblastoma tissues derived
from patients treated with dose-intensified TMZ in a prospective
clinical phase II trial (UKT-05) . A schematic overview of the
MGMT promoter region including highlighted CpG sites
addressed by each of the three diagnostic methods is given in
Figure 1. Methylation data were successfully obtained in 33 of 35
(MSP), 35 of 35 (PSQ) and 34 of 35 (MS-MLPA) tumour
specimens, respectively. The results of the three different
diagnostic methods are listed in Table 1. For MSP, 14 of 33
analysed tumours (42%) were methylation-positive. As an
alternative approach to evaluate the MSP data in a
semiquantitative way, the ratio of methylation was calculated for each
specimen by comparing the intensities of methylated (M) and
unmethylated (U) MSP bands. 7 of 33 analysed tumours displayed
an M/U ratio.1 and were therefore assessed as strongly
Methylation +, n (%)
Methylation 2, n (%)
M/U ratio.1, n (%)
M/U ratio,1, n (%)
M/U ratio = 0, n (%)
Successful analysis, n (%)
33/35 (94) 35/35 (100)
Mean methylation .10%, n (%) N/A
Mean methylation .30%, n (%) N/A
For PSQ and MS-MLPA, the mean methylation signal averaged over all CpGs
addressed in each method is indicated. N/A, not applicable.
methylated, whereas for 11 of 33 tumours, an M/U ratio between
0 and 1 was calculated, indicating only weak MGMT promoter
methylation. In 15 tumours, the M/U ratio was 0 (no M primer
MSP product detectable), indicating an unmethylated MGMT
promoter status. For PSQ, single methylation signals ranged from
0% to 100%. Averaged over all five CpGs, 15 of 35 tissues (43%)
showed a mean methylation signal above 10%. Of these, 12
specimens (34%) showed a mean methylation signal above 30%.
For MS-MLPA, methylation signals ranged between 225% and
+76%. Averaged over all three CpG sites, 13/34 tissues (38%)
displayed a mean methylation signal above 10%, and of these, 3
specimens (9%) above 30%.
Cut-point testing and survival analyses
To allow for a direct comparison with the qualitative
methylation data obtained by MSP in respect to survival, the
continuous (semi-)quantitative methylation signals from PSQ and
MS-MLPA were converted into binary methylation data. To this
end, we used maximally selected log-rank statistics to test for
cutpoints in the continuous CpG methylation data obtained by PSQ
and MS-MLPA and to estimate the corresponding cut-off value in
case of significance. With respect to PFS, we identified significant
cut-points for the quantitative methylation signals of all five CpG
sites addressed by PSQ, and for MS-MLPA site 1. With respect to
OS, cut-point estimations yielded a significant value only for PSQ
CpG site 5 but for none of the interrogated MS-MLPA sites
(Table 2). Notably, the cut-points for the PSQ CpG sites 1, 3 and 4
were in the range of 11.5%14.0%. This correlates well with our
observation that in normal brain tissue, the degree of MGMT
promoter methylation mostly ranges between 0% and 10% (data
not shown). For PSQ CpG 2 (PFS, 31.5%) and CpG 5 (PFS,
44.9%; OS, 44.3%) higher cut-points were calculated.
Significant cut-point values were then used to dichotomise the
continuous methylation data from PSQ and MS-MLPA and
generate Kaplan-Meier plots based on the binary covariates.
Figure 2AC provides plots of the Kaplan-Meier survival curve
estimates for both qualitative MSP results and dichotomised
methylation data from all five PSQ CpG sites and all three
MSMLPA sites. Of note, PSQ variables and particularly the resulting
binary PSQ covariates were highly significant prognostic factors,
whereas for MS-MLPA weaker associations were obtained.
Applying semi-quantitative assessment of MSP based on M/U
ratios, PFS was only significantly prolonged in patients suffering
from tumours with strong promoter methylation (M/U ratio.1)
when compared with all other patients (p = 0.008, log-rank test).
However, tumour samples with an M/U ratio.1 were not
significantly higher correlated with prolonged survival when
specifically compared to tumours with few methylation, i.e. with
an M/U ratio between 0 and 1 (PFS, p = 0.06; OS, p = 0.22,
logrank test). Moreover, semi-quantitative MSP using M/U ratios
was not significantly associated with OS (p = 0.22, log-rank test).
As opposed to this, conventionally applied qualitative MSP
assessment (+/2) was superior in that it was significantly related
to both survival endpoints, PFS (p,0.001; log-rank test) and OS
(p = 0.04, log-rank test).
Likelihood ratio tests and prediction errors
We determined the additional predictive value gained by the
results of each methylation measurement relative to major
therapy-independent clinical covariates for newly diagnosed
glioblastoma, i.e., age, gender, Karnofsky performance status,
extent of resection, using likelihood ratio tests (Table 3). MSP and
PSQ improved prediction of PFS to a similar extent (p,0.001 for
MSP and continuous and dichotomised PSQ data) while
MSMLPA did not provide a significant gain of additional information
(p = 0.02). With regard to OS, only MSP delivered a significant
gain of prediction relative to clinical data alone (p = 0.001) while
the continuous data for both PSQ and MS-MLPA did not provide
such an improvement (Table 3). Of note, it must be conceded that
the likelihood ratio test results are overoptimistic as cut-point
estimation was performed on the same data set. To this end,
Table 3 additionally lists the cross-validated measures of prediction
error and R2 that take this problem into account and hence show
less optimistic values.
To assess the predictive accuracy of the three diagnostic
methods relative to the above-stated clinical prognostic factors, we
computed cumulative prediction error curves over 18 months
follow-up time (which was near to the median follow-up of 21.7
months) and a time-dependent R2-like measure using the
KaplanMeier model as a reference (Figure 3). With regard to PFS, we
found that Kaplan-Meier estimates improved the survival
predictability relative to sole clinical covariates, and both MSP
and PSQ (both original and dichotomised data), in combination
with the clinical covariates, tended to improve this effect to a
similar degree. However, MS-MLPA plus clinical data hardly
reduced the prediction error of the Kaplan-Meier reference
Figure 2. Method-dependent Kaplan-Meier estimates of PFS. PFS data for binary methylation covariates were obtained from (A) MSP, (B) PSQ
CpG sites 1 to 5 and (C) MS-MLPA sites 1 to 3 based on missing data imputation (n = 35) for each diagnostic method applied. Respective cut-points
allowing the conversion from continuous to binary methylation data in (B) and (C) are indicated in Table 2. Of note, Kaplan-Meier curves for MS-MLPA
sites 2 and 3 were computed on grounds of insignificant cut-points (see Table 2). Respective p values of the maximally selected log-rank test and
respective numbers (percentages) of tumours assigned to each group, methylated or unmethylated, are given in each plot. Black curves,
unmethylated; red curves, methylated.
(Figure 3A) reflecting the relative inferiority of this method as
proved by the likelihood ratio tests (Table 3). With respect to OS,
neither PSQ nor MS-MLPA data additionally decreased the
prediction error of the Kaplan-Meier reference to a relevant
extent, whereas MSP demonstrated a clear-cut improvement of
OS prediction (Figure 3B).
Due to its high prognostic and predictive relevance, assessment
of the MGMT status has become state-of-the-art in current and
planned clinical trials in glioma as a prognosticator, to stratify
patients or even to limit trial entry accordingly. Moreover, it is
CF + PSQ (CpG 15)
CF + dPSQ (CpG 15)
CF + MS-MLPA (site 13)
CF + dMS-MLPA (site 13)
For LR, the clinical prognostic factors were used as a reference (Ref.). For R2, Kaplan-Meier was used as a reference (Ref.). MSP, methylation-specific polymerase chain
reaction; dPSQ (CpG 15), dichotomised pyrosequencing data for CpG 15; dMS-MLPA, dichotomised methylation-specific multiplex ligation-dependent probe
amplification data for site 13. N/A, not applicable.
frequently requested in routine diagnostics as a prognostic tool.
The ongoing debate among neuropathologists about whether or not
MGMT testing should be incorporated into the next revision of the
WHO classification scheme for central nervous system tumours also
points at an emerging diagnostic role for MGMT. Meanwhile, an
internationally accepted consensus as to the most appropriate
diagnostic instrument for MGMT testing is still unmet .
In this study, we sought to clarify which of three frequently
applied techniques of assessing MGMT promoter methylation,
MSP, PSQ and MS-MLPA, would predict clinical outcome most
accurately. We chose a well-characterized, uniform patient cohort
newly diagnosed with glioblastoma that received upfront
chemoradiotherapy with dose-intensified temozolomide within the
UKT05 trial . We used PFS as the clinical outcome measure in order
to avoid a potential bias caused by differences in post-progression
treatments. Moreover, this endpoint reflected the primary
endpoint in the original UKT-05 trial  and was recently
validated by a study that proposed six-month PFS as an alternative
primary efficacy endpoint to OS in newly diagnosed glioblastoma
patients receiving temozolomide .
Although done on a comparatively low sample number, our
study features the following distinctive methodological strengths: i)
all tissues examined stemmed from a prospective multicentre trial;
ii) all tissues were exclusively obtained from open tumour
resections (either complete or partial) to avoid a tissue
contamination bias and gate out the lacking therapeutic and prognostic
effect of sole biopsies from our analysis; iii) all three diagnostic
methods applied were performed at one neuropathological
institute; iv) all three techniques were performed on FFPE tissues
closely resembling the clinical routine situation.
The major findings of our study are: i) all three techniques tested
are feasible on FFPE tissue; ii) a statistical dichotomisation of
continuous methylation data obtained by PSQ is feasible within a
given cohort and allows condensation to qualitative +/2 results as
a prerequisite for comparative survival analyses; iii)
MS-MLPAderived methylation data are correlated to a weaker degree than
PSQ data, and the estimation of a valid cut-off value is probably
more difficult with MS-MLPA than with PSQ, especially when
smaller cohorts such as in clinical phase II trials are evaluated; iv)
both MSP and PSQ are superior to MS-MLPA in predicting
The MGMT gene located on chromosome 10q26 has five exons
and a CpG-rich island of 762 bp with 98 CpG dinucleotides
encompassing the first exon and large parts of the promoter
(Figure 1). Despite a few studies that conducted correlative
analyses between MGMT promoter methylation at individual CpG
sites and gene expression [24,25,26,27], it still remains an open
question which or how many CpGs in the MGMT promoter CpG
island (CGI) have a major impact on expression and best reflect
response to treatment and survival.
Using comprehensive pyrosequencing of the entire MGMT
CGI, mRNA expression analyses and luciferase reporter assays, a
recent contribution to this issue identified a distinct region within
MGMT exon 1 (spanning CpG73-90) and particularly four specific
CpG sites (CpG 83, 86, 87 and 89) to be most critical in the
transcriptional control of MGMT and thus recommendable for
MGMT testing . This region harbours the annealing sites of
the most commonly used MSP primers (forward M primer,
CpG76-80; forward U primer, CpG75-80; reverse M and U
primers, CpG84-87) that were also applied in our study, as well as
the five adjacent CpGs that we interrogated by PSQ (CpG74-78)
(Figure 1). The fact that MSP and PSQ resulted in similar
prediction of PFS may be a consequence of this partial overlap in
analysed CpGs. Unlike MSP and PSQ, the three CpGs addressed
by MS-MLPA are widespread throughout most of the MGMT
CGI (CpG 9, 23 and 81) and two of them (CpG 9 and 23) are even
located in the upstream promoter region, distant from the sites
that were interrogated by MSP and PSQ (Figure 1). This might
explain the weaker correlation of CpG methylation of the
MSMLPA data and their inferior value in predicting PFS in our study.
Recently published studies correlated the MGMT status with
clinical survival data: Based on promoter-wide methylation
analyses of snap-frozen glioblastoma tissues using quantitative
bisulfite sequencing and correlations with mRNA expression,
protein expression and PFS, Shah et al. proposed a new
classification scheme using methylation data from three different
regions of the entire MGMT promoter, and provided confirmative
data achieved by MS-MLPA. This approach seems promising in
that it accounts for whole promoter-wide methylation patterns and
integrates methylation, expression and survival data .
In a prospective study examining 63 patients diagnosed with
malignant glioma, Kreth et al. identified MGMT mRNA
expression as a predictor of clinical outcome independent from
MGMT promoter methylation underscoring the necessity of
approaching MGMT biology more comprehensively and also
elucidating methylation-independent mechanisms that may
regulate MGMT expression . Assessing the specimens of the
UKT05 trial for MGMT mRNA expression would have been tempting.
However, only FFPE tissues were available from different sources
that do not allow sufficient RNA extraction.
The value of PSQ, as shown in our analyses, is supported by
Karayan-Tapon et al. who evaluated MGMT promoter
methylation assessment by MSP, semi-quantitative MSP, PSQ, qRT-PCR,
and IHC for their value of predicting OS in glioblastoma patients.
In this study, PSQ reached the highest predictive value,
particularly at CpG site 4 .
MSP has evolved as the gold standard for methylation
analysis of the MGMT gene promoter. It is the easiest to perform
and least expensive of the three methods and does not require any
special equipment or consumables aside from what is present in
most medical laboratories anyway. MSP is often regarded to be
non-optimal for some settings, especially when performed on low
quality DNA extracted from FFPE tissue [10,26]. Irregular mosaic
methylation patterns and incomplete bisulfite conversion may lead
to mispriming and lower sensitivity and specificity [10,26,31,32].
However, in our setting the assay was successful in a high
percentage (94%). Other studies have reported a much higher
failure rate of the MSP assay when FFPE tissue was used [1,31]. In
our setting, it turned out to be similarly good as PSQ as to the
prediction of PFS and the only of the three assays that significantly
improved prediction of OS. Although the MSP assay delivers,
when functional, mostly easy-to-interpret results ready to assist
clinical decision-making, its major downside consists in its inability
to detect heterogeneous patterns of methylation. The use of the
MSP technique as a semi-quantitative assay through the
comparison of the relative intensities of M and U primer-specific
MSP bands did not improve on the prognostic value of the
conventional qualitative assessment of MSP. We conclude from
this that MSP should be used as a purely qualitative assay that is
less appropriate to allow (semi-)quantitative interpretations.
PSQ overcomes this problem as it provides quantitative
information on the extent of methylation at each individual
CpG site with high sensitivity and specificity . However, the
interpretation of PSQ data is limited by a lack of consensus
concerning a biologically relevant threshold that allows conversion
of the original continuous data into a clinically practical binary
code, i.e., methylated or unmethylated. The results of our study
indeed demonstrate that, within a given cohort of samples, such a
dichotomisation of PSQ data is statistically feasible and allows
comparative survival analyses leading to highly significant results.
However, due to the high costs of the required equipment, PSQ is
not widely used in clinical diagnostics when single samples are
subject to analysis .
MS-MLPA is a semi-quantitative method that has the
advantage of omitting a DNA-modifying bisulfite treatment step,
thus avoiding additional damage to the sample DNA. Besides its
requirement for special equipment and expensive reagents,
MSMLPA is limited by its dependence on the presence of HhaI
restriction sites, and only one of the three MGMT CpG sites
suitable for MS-MLPA is located within the region that is
commonly analysed by MSP or PSQ (Figure 1). Furthermore,
similar to PSQ, MS-MLPA data need an algorithm of conversion
into a +/2 code. In our study, such a dichotomisation was possible
only for MS-MLPA site 1. However, MS-MLPA data did not
improve prediction of survival significantly (Table 3, Figure 3).
Conclusively, taking into consideration the ability to predict
clinical outcome, cost effectiveness and ease of use,
pyrosequencing seems to be most suitable for methylation analysis in a
highthroughput setting (e.g., for the evaluation of the MGMT status of
many specimens in larger clinical trials) while MSP seems to be
more convenient in clinical routine diagnostics when low numbers
of specimens need to be examined at a time.
Materials and Methods
UKT-05 was designed as a prospective clinical phase II trial
that included 41 adult patients (median Karnofsky performance
status: 90%; median age: 56 years) who were newly diagnosed with
glioblastoma and treated up-front according to a dose-intensified
TMZ-containing chemoradiotherapy protocol. The ethics
committee at the University of Tu bingen (Tu bingen, Germany)
approved the trial (253/2004). All patients gave written informed
consent prior to study entry. MGMT analyses are specifically
mentioned in the informed consent. All patients also consented to
this translational research to be performed. TMZ was
administered orally before and after radiation therapy in a weekly
alternating schedule starting at 150 mg/m2 on days 17 of 14
daycycles (1 week on/1 week off), with individual dose adjustments
of TMZ in 25 mg-steps according to weekly haemograms.
Standard involved-field radiotherapy was delivered in daily single
fractions of 1.8 to 2.0 Gy, 5 days per week. During radiotherapy,
low-dose TMZ was given concomitantly at 50 mg/m2. In
addition, maintenance indomethacin was orally administered at
25 mg twice daily throughout the entire treatment without
individual dose adjustments. PFS was defined as the time interval
between the day of surgery and tumour progression on magnetic
resonance imaging according to the criteria of MacDonald et al.
 and/or clinical progression. OS was defined as the time
interval between the day of surgery and death. A more detailed
description of this trial is given in . The tumour specimens of 35
patients (85%) were assessable for MGMT promoter methylation
analysis. All tumour specimens analysed in the present study were
obtained by either complete (n = 17/35; 49%) or partial (n = 18/
35; 51%) debulking surgery.
DNA extraction and bisulfite treatment
To ensure high tumour DNA content, FFPE tissue sections were
stained with H&E and histologically examined by an experienced
neuropathologist (C. Hartmann). Sections showing a tumour cell
content of more than 80% were directly subjected to DNA
extraction, while on sections with adjacent non-neoplastic tissue,
the tumour portion was microdissected and further processed.
Extraction of genomic DNA was performed using the QIAamp
DNA Mini Kit (Qiagen, Hilden, Germany) and quantified with a
NanoDrop ND-1000 (PeqLab, Erlangen, Germany). Five hundred
nanograms of extracted DNA as well as CpGenome Universal
Methylated DNA (Chemicon International, Temecula, CA) and
CpGenome Universal Unmethylated DNA (Chemicon
International) as controls were subjected to bisulfite treatment using the
EpiTect Bisulfite Kit (Qiagen). The bisulfite-treated DNA was
used for MSP and PSQ, while MS-MLPA was performed with
untreated genomic DNA. The efficiency of the bisulfite conversion
was checked by analysing the control DNA by pyrosequencing.
Methylation-specific polymerase chain reaction (MSP)
The two primer sets established by Esteller et al. for MSP of
MGMT  were 59-TTTCGACGTTCGTAGGTTTTCGC-39
(forward primer) and 59-GCACTCTTCCGAAAACGAAACG-39
(reverse primer) for methylated template detection (M primers,
product length 81 bp; Figure 1, purple bars) and
59-AACTCCACACTCTTCCAAAAACAAAACA39 (reverse primer) for unmethylated template detection (U
primers, product length 93 bp; Figure 1, mint bars). The PCR
was performed in a total volume of 20 ml containing 10 ml
HotStarTaq Mix (Qiagen), 1 ml of the respective forward and
reverse primer (10 pmol), 6 ml high purity water and 2 ml
bisulfitetreated template DNA. The PCR programme was 95uC for
15 min, then 35 cycles of 95uC for 50 s, 59uC for 50 s and 72uC
for 50 s, followed by a final step at 72uC for 10 min. PCR
reactions with CpGenome Universal Methylated DNA (Chemicon
International), with CpGenome Universal Unmethylated DNA
Vial A (Chemicon International), and without any DNA
(nontemplate control) were included as controls. PCR products were
separated on a 2% agarose gel. For a qualitative assessment, a
visible M primer band indicated a positive methylation status,
whereas absence of an M primer MSP product was evaluated as a
negative methylation status of the respective tumour specimen. For
an alternative semi-quantitative approach, images of the agarose
gels were analysed with ImageJ software (National Institutes of
Health, Bethesda, MD; http://rsb.info.nih.gov/ij). For each
specimen, the optical band intensities of the corresponding M
primer and U primer MSP products were quantified and corrected
against the background of the gel. Tumour specimens with an M/
U ratio.1 were assessed as strongly methylated, whereas an M/U
ratio between 0 and 1 indicated weak promoter methylation.
Tumour specimens with an M/U ratio = 0 (no M primer MSP
product) were assessed as unmethylated.
PSQ was performed on a PSQ 96ID system (Qiagen) with a set of
primers provided with the PyroMark MGMT Kit (Qiagen). The
primer set covers a region of the MGMT promoter located at the
start of the first exon (Figure 1, dark blue box), which is adjacent to
the region that is covered by the MSP primers. The PCR was
performed in a volume of 40 ml containing 20 ml HotStar Taq Mix
(Qiagen), 1 ml of each PCR primer (10 pmol), 8 ml high purity water
and 10 ml of bisulfite-treated template DNA. The PCR cycling
programme for both primer sets was composed of an initial
activation step at 95uC for 15 min, followed by 40 cycles of
denaturation at 95uC for 30 s, annealing at 53uC for 30 s and
elongation at 72uC for 30 s. The programme was finished by a final
elongation step at 72uC for 10 min. PCR products were visualized
by gel electrophoresis, and 30 ml were subjected to the PSQ sample
preparation process. DNA was mixed with streptavidin-coated
sepharose beads, followed by strand separation and washing
utilising the vacuum prep tool (Qiagen). The single-stranded
DNA bound to the sepharose beads was mixed with 40 ml of
0.4 mM sequencing primer solution, heated to 80uC for 60 s and
then cooled down to room temperature for annealing. For the
sequencing reaction PyroMark Gold reagents were used (Qiagen).
The sequencing results were analysed using the PSQ PyroMark
software (Qiagen). As controls, CpGenome Universal Methylated
DNA (positive methylation control; Chemicon International) and
DNA from FFPE non-tumourous brain tissue (negative methylation
control) were included in the assay, as well as a reaction without any
template DNA (non-template control). Pyrograms of the control
DNA were analysed to confirm complete bisulfite conversion. All
tumour and control specimens were measured in triplicates.
Methylation-specific multiplex ligation-dependent probe
Two hundred nanograms of non-bisulfite treated DNA were
subjected to the MS-MLPA procedure using the ME011 kit (MRC
Holland, Amsterdam, Netherlands). Three GCGC Hhal sites
within the MGMT promoter that are addressed by this kit are
depicted in Figure 1 (light blue boxes). The amplification products
were separated by capillary electrophoresis on an ABI 3100
Genomic Analyser (Applied Biosystems, Foster City, CA). As
reference samples, untreated DNA from FFPE non-tumourous
brain tissue was used. The data analysis was performed with
The prognostic value of MGMT methylation data obtained by
MSP, PSQ or MS-MLPA was assessed with respect to PFS and
OS. PFS was used as the primary read-out as it provided the most
valid information on the biological activity of dose-intensified
TMZ tested upfront in UKT-05 without being biased by
MGMTindependent effects of second-line treatments. Survival analysis on
an intention-to-treat basis in UKT-05 required initiation of the
trial treatment. Survival curve estimation was done using the
Kaplan-Meier method . Analysis of the original continuous
methylation data was performed using Cox proportional hazards
regression models. Missing values were imputed by single
imputation applying predictive mean matching using the R
package Hmisc, version 3.83. Maximally selected log-rank
statistics were used to test for cut-points in continuous CpG
methylation data obtained by PSQ and MS-MLPA and, if
significant, to estimate the corresponding cut-off value . This
was done using R package coin, version 1.018. Methylation data
obtained by PSQ reflect the median of triplicates for each CpG
site addressed. P,0.01 was considered to indicate statistical
For correlation analysis, partial correlations were computed
using the estimation procedure described by Schafer and
Strimmer  and implemented in the R package GeneNet,
version 1.2.4. To assess the additional prognostic value of
methylation measures beyond clinical factors, models with and
without MSP, PSQ, or MS-MLPA data were compared by
likelihood ratio tests. The model including the major
therapyindependent clinical prognostic factors, age, gender, Karnofsky
performance status and extent of resection was compared with the
models including dichotomised methylation factors obtained by
PSQ and MS-MLPA. To assess the predictive accuracy of models
including methylation data, the cumulating prediction error curves
over 18 months follow-up time (which was near to the median
follow-up of 21.7 months) and a time-dependent R2-like measure
were computed using 10-fold cross-validation taking the estimation
of cut-points into account . All statistical analyses were
computed using the statistical software environment R, version
Conceived and designed the experiments: AC CH AvD M. Weiler.
Performed the experiments: AC JM. Analyzed the data: AC CH AB M.
Weiler. Contributed reagents/materials/analysis tools: AVD M. Weiler
WW. Wrote the paper: AC CH AB AvD M. Weller WW M. Weiler.
Hegi ME , Diserens AC , Gorlia T , Hamou MF , de Tribolet N , et al. ( 2005 ) MGMT gene silencing and benefit from temozolomide in glioblastoma.
N Engl J Med 352 : 997 - 1003 .
( 2000 ) Inactivation of the DNA-repair gene MGMT and the clinical response of gliomas to alkylating agents . N Engl J Med 343 : 1350 - 1354 .
Paz MF , Yaya-Tur R , Rojas-Marcos I , Reynes G , Pollan M , et al. ( 2004 ) CpG island hypermethylation of the DNA repair enzyme methyltransferase predicts response to temozolomide in primary gliomas . Clin Cancer Res 10 : 4933 - 4938 .
Hegi ME , Liu L , Herman JG , Stupp R , Wick W , et al. ( 2008 ) Correlation of O6- methylguanine methyltransferase (MGMT) promoter methylation with clinical outcomes in glioblastoma and clinical strategies to modulate MGMT activity .
J Clin Oncol 26 : 4189 - 4199 .
Weiler M , Hartmann C , Wiewrodt D , Herrlinger U , Gorlia T , et al. ( 2010 ) Chemoradiotherapy of newly diagnosed glioblastoma with intensified temozolomide . Int J Radiat Oncol Biol Phys 77 : 670 - 676 .
6. Weller M , Felsberg J , Hartmann C , Berger H , Steinbach JP , et al. ( 2009 ) Molecular predictors of progression-free and overall survival in patients with newly diagnosed glioblastoma: a prospective translational study of the German Glioma Network . J Clin Oncol 27 : 5743 - 5750 .
7. Pegg AE ( 2000 ) Repair of O(6)-alkylguanine by alkyltransferases . Mutat Res 462 : 83 - 100 .
8. Karran P , Bignami M ( 1994 ) DNA damage tolerance, mismatch repair and genome instability . Bioessays 16 : 833 - 839 .
9. Gerson SL ( 2004 ) MGMT: its role in cancer aetiology and cancer therapeutics . Nat Rev Cancer 4 : 296 - 307 .
10. Preusser M ( 2009 ) MGMT analysis at DNA, RNA and protein levels in glioblastoma tissue . Histol Histopathol 24 : 511 - 518 .
11. Herman JG , Graff JR , Myohanen S , Nelkin BD , Baylin SB ( 1996 ) Methylationspecific PCR: a novel PCR assay for methylation status of CpG islands . Proc Natl Acad Sci U S A 93 : 9821 - 9826 .
12. Vlassenbroeck I , Califice S , Diserens AC , Migliavacca E , Straub J , et al. ( 2008 ) Validation of real-time methylation-specific PCR to determine O6-methylguanine-DNA methyltransferase gene promoter methylation in glioma . J Mol Diagn 10 : 332 - 337 .
13. Nygren AO , Ameziane N , Duarte HM , Vijzelaar RN , Waisfisz Q , et al. ( 2005 ) Methylation-specific MLPA (MS-MLPA): simultaneous detection of CpG methylation and copy number changes of up to 40 sequences . Nucleic Acids Res 33 : e128 .
14. Frommer M , McDonald LE , Millar DS , Collis CM , Watt F , et al. ( 1992 ) A genomic sequencing protocol that yields a positive display of 5-methylcytosine residues in individual DNA strands . Proc Natl Acad Sci U S A 89 : 1827 - 1831 .
15. Xiong Z , Laird PW ( 1997 ) COBRA: a sensitive and quantitative DNA methylation assay . Nucleic Acids Res 25 : 2532 - 2534 .
16. Ronaghi M , Uhlen M , Nyren P ( 1998 ) A sequencing method based on real-time pyrophosphate . Science 281 : 363 - 365 .
17. El-Maarri O , Herbiniaux U , Walter J , Oldenburg J ( 2002 ) A rapid, quantitative, non-radioactive bisulfite-SNuPE- IP RP HPLC assay for methylation analysis at specific CpG sites . Nucleic Acids Res 30 : e25 .
18. Tanaka S , Kobayashi I , Utsuki S , Oka H , Fujii K , et al. ( 2003 ) O6- methylguanine-DNA methyltranspherase gene expression in gliomas by means of real-time quantitative RT-PCR and clinical response to nitrosoureas . Int J Cancer 103 : 67 - 72 .
19. Anda T , Shabani HK , Tsunoda K , Tokunaga Y , Kaminogo M , et al. ( 2003 ) Relationship between expression of O6-methylguanine-DNA methyltransferase, glutathione-S-transferase pi in glioblastoma and the survival of the patients treated with nimustine hydrochloride: an immunohistochemical analysis . Neurol Res 25 : 241 - 248 .
20. Capper D , Mittelbronn M , Meyermann R , Schittenhelm J ( 2008 ) Pitfalls in the assessment of MGMT expression and in its correlation with survival in diffuse astrocytomas: proposal of a feasible immunohistochemical approach . Acta Neuropathol 115 : 249 - 259 .
21. Preuss I , Eberhagen I , Haas S , Eibl RH , Kaufmann M , et al. ( 1995 ) O6- methylguanine-DNA methyltransferase activity in breast and brain tumors . Int J Cancer 61 : 321 - 326 .
22. Weller M , Stupp R , Reifenberger G , Brandes AA , van den Bent MJ , et al. ( 2010 ) MGMT promoter methylation in malignant gliomas: ready for personalized medicine ? Nat Rev Neurol 6 : 39 - 51 .
23. Polley MY , Lamborn KR , Chang SM , Butowski N , Clarke JL , et al. ( 2010 ) Sixmonth progression-free survival as an alternative primary efficacy endpoint to overall survival in newly diagnosed glioblastoma patients receiving temozolomide . Neuro Oncol 12 : 274 - 282 .
24. Watts GS , Pieper RO , Costello JF , Peng YM , Dalton WS , et al. ( 1997 ) Methylation of discrete regions of the O6-methylguanine DNA methyltransferase (MGMT) CpG island is associated with heterochromatinization of the MGMT transcription start site and silencing of the gene . Mol Cell Biol 17 : 5612 - 5619 .
25. Nakagawachi T , Soejima H , Urano T , Zhao W , Higashimoto K , et al. ( 2003 ) Silencing effect of CpG island hypermethylation and histone modifications on O6-methylguanine-DNA methyltransferase (MGMT) gene expression in human cancer . Oncogene 22 : 8835 - 8844 .
26. Mikeska T , Bock C , El-Maarri O , Hubner A , Ehrentraut D , et al. ( 2007 ) Optimization of quantitative MGMT promoter methylation analysis using pyrosequencing and combined bisulfite restriction analysis . J Mol Diagn 9 : 368 - 381 .
27. Everhard S , Tost J , El Abdalaoui H , Criniere E , Busato F , et al. ( 2009 ) Identification of regions correlating MGMT promoter methylation and gene expression in glioblastomas . Neuro Oncol 11 : 348 - 356 .
28. Malley DS , Hamoudi RA , Kocialkowski S , Pearson DM , Collins VP , et al. ( 2011 ) A distinct region of the MGMT CpG island critical for transcriptional regulation is preferentially methylated in glioblastoma cells and xenografts . Acta Neuropathol 121 : 651 - 661 .
29. Shah N , Lin B , Sibenaller Z , Ryken T , Lee H , et al. ( 2011 ) Comprehensive analysis of MGMT promoter methylation: correlation with MGMT expression and clinical response in GBM . PLoS One 6 : e16146 .
30. Kreth S , Thon N , Eigenbrod S , Lutz J , Ledderose C , et al. ( 2011 ) Omethylguanine-DNA methyltransferase (MGMT) mRNA expression predicts outcome in malignant glioma independent of MGMT promoter methylation . PLoS One 6 : e17156 .
31. Karayan-Tapon L , Quillien V , Guilhot J , Wager M , Fromont G , et al. ( 2010 ) Prognostic value of O6-methylguanine-DNA methyltransferase status in glioblastoma patients, assessed by five different methods . J Neurooncol 97 : 311 - 322 .
32. Shaw RJ , Akufo-Tetteh EK , Risk JM , Field JK , Liloglou T ( 2006 ) Methylation enrichment pyrosequencing: combining the specificity of MSP with validation by pyrosequencing . Nucleic Acids Res 34 : e78 .
33. Macdonald DR , Cascino TL , Schold SC , Jr., Cairncross JG ( 1990 ) Response criteria for phase II studies of supratentorial malignant glioma . J Clin Oncol 8 : 1277 - 1280 .
34. Kaplan E , Meier P ( 1958 ) Non-parametric estimation from incomplete observations . J Am Stat Assoc 53 : 457 - 481 .
35. Hothorn T , Lausen B ( 2003 ) On the Exact Distribution of Maximally Selected Rank Statistics . Comput Stat Data Anal 43 : 121 - 137 .
36. Schafer J , Strimmer K ( 2005 ) A shrinkage approach to large-scale covariance matrix estimation and implications for functional genomics . Stat Appl Genet Mol Biol 4: Article 32.
37. Gerds TA , Schumacher M ( 2007 ) Efron-type measures of prediction error for survival analysis . Biometrics 63 : 1283 - 1287 .
38. R Development Core Team ( 2011 ) R: A language and environment for statistical computing . Vienna, Austria.