MGMT promoter methylation analysis for allocating combined CCNU/TMZ chemotherapy: Lessons learned from the CeTeG/NOA‐09 trial

The CeTeG/NOA‐09 trial showed a survival benefit for combined CCNU/TMZ therapy in MGMT‐promoter‐methylated glioblastoma patients (quantitative methylation‐specific PCR [qMSP] ratio > 2). Here, we report on the prognostic value of the MGMT promoter methylation ratio determined by qMSP and evaluate the concordance of MGMT methylation results obtained by qMSP, pyrosequencing (PSQ) or DNA methylation arrays (MGMT‐STP27). A potential association of qMSP ratio with survival was analyzed in the CeTeG/NOA‐09 trial population (n = 129; log‐rank tests, Cox regression analyses). The concordance of MGMT methylation assays (qMSP, PSQ and MGMT‐STP27) was evaluated in 76 screened patients. Patients with tumors of qMSP ratio > 4 showed superior survival compared to those with ratios 2‐4 (P = .0251, log‐rank test). In multivariate analysis, the qMSP ratio was not prognostic across the study cohort (hazard ratio [HR] = 0.88; 95% CI: 0.72‐1.08). With different cutoffs for qMSP ratio (4, 9, 12 or 25), the CCNU/TMZ benefit tended to be larger in subgroups with lower ratios (eg, for cutoff 9: HR 0.32 for lower subgroup, 0.73 for higher subgroup). The concordance rates with qMSP were 94.4% (PSQ) and 90.2% (MGMT‐STP27). Discordant results were restricted to tumors with qMSP ratios ≤4 and PSQ mean methylation rate ≤25%. Despite a shorter survival in MGMT‐promoter‐methylated patients with lower methylation according to qMSP, these patients had a benefit from combined CCNU/TMZ therapy, which even tended to be stronger than in patients with higher methylation rates. With acceptable concordance rates, decisions on CCNU/TMZ therapy may also be based on PSQ or MGMT‐STP27.

subgroups with lower ratios (eg, for cutoff 9: HR 0.32 for lower subgroup, 0.73 for higher subgroup). The concordance rates with qMSP were 94.4% (PSQ) and 90.2% (MGMT-STP27). Discordant results were restricted to tumors with qMSP ratios ≤4 and PSQ mean methylation rate ≤25%. Despite a shorter survival in MGMT-promoter-methylated patients with lower methylation according to qMSP, these patients had a benefit from combined CCNU/TMZ therapy, which even tended to be stronger than in patients with higher methylation rates. With acceptable concordance rates, decisions on CCNU/TMZ therapy may also be based on PSQ or MGMT-STP27. (IDH-wildtype glioblastomas) are the most common malignant primary brain tumors in adults. 1 With a median overall survival of about 17 months in unstratified study populations, IDH-wildtype glioblastoma patients have a dismal prognosis. 2 Standard of care treatment includes surgery followed by radiotherapy and chemotherapy with the DNA alkylating agent temozolomide (TMZ). 3 Promoter methylation of the O-6-methylguanine-DNA methyltransferase (MGMT) gene is so far the most important prognostic and predictive biomarker in IDHwildtype glioblastoma patients. 4 Patients whose tumors exhibit MGMT promoter methylation have a superior overall survival when treated with alkylating chemotherapy due to impaired DNA repair mechanisms. For this subset of glioblastoma patients, the randomized phase III multicenter CeTeG/NOA-09 trial recently showed a survival benefit when they were treated with a combination of lomustine (CCNU) and TMZ (CCNU/TMZ) instead of TMZ monotherapy. 5 Due to the encouraging results of the CeTeG/NOA-09 trial, some centers have started using combined CCNU/TMZ therapy off-study for patients with MGMT promoter-methylated glioblastoma. However, no current standard exists for MGMT promoter methylation testing in clinical routine and results from different tests may not be completely concordant. In several prospective phase II/III glioma trials 2,6-9 and also in the CeTeG/NOA-09 trial, 5 the method chosen for MGMT promoter methylation testing was a quantitative methylation-specific PCR (qMSP) assay. 10 This assay yields a methylation ratio between MGMT and β-actin (ACT-B) at a logarithmic scale after sodium bisulfite conversion of tumor DNA and amplification of methylated sequences from the MGMT-associated 5 0 CpG island. 10 Using the qMSP assay, two major issues remain for routine clinical use in decision making for or against a combined CCNU/TMZ therapy: (i) The cutoff for MGMT promoter methylation is set at an MGMT methylation ratio of 2, yet What's new?
In patients with IDH-wildtype glioblastoma, methylation of the MGMT promoter allows for improved survival after chemotherapy, due to reduced ability to repair DNA damage.
Here, the authors set out to evaluate the use of different tests for promoter methylation, with an eye toward their usefulness at allocation of chemotherapy, and on their prognostic applicability. They show that three different methods of testing methylation-quantitative methylation-specific PCR, pyrosequencing, and DNA methylation arrays-agree more than 90% of the time. Patients with lower MGMT promoter methylation had shorter survival times, but still benefited from CCNU/TMZ therapy. uncertainty is reported for ratios closely above the cutoff of 2. 10 The question remains, whether patients harboring methylated tumors with methylation ratios in the lower range (a) have shorter survival than patients with higher methylation rates and (b) have a benefit from the more aggressive combination treatment with CCNU/TMZ as compared to TMZ monotherapy. (ii) Nowadays, qMSP is not commonly used in clinical routine, while pyrosequencing (PSQ) 11 of sodium bisulfite-modified DNA seems to be the preferred method for MGMT promoter methylation testing in many European centers. Another assay that is increasingly being used to assess the MGMT promoter methylation status is based on DNA methylation profiling using the Infinium Methylation Epic 850k array (Illumina, San Diego, California) and the MGMT-STP27 algorithm. 12 In this algorithm, methylation of two CpG islands in the differentially methylated region 1 and 2 (DMR1/2) of the MGMT 5 0 CpG island respectively is determined. Here, we performed a post hoc analysis using data from the CeTeG/NOA-09 trial evaluating the prognostic and predictive value of low-level MGMT promoter methylation ratios as determined by qMSP.
In addition, we investigated the concordance of the results obtained for MGMT promoter methylation by qMSP, PSQ or MGMT-STP27.

| Quantitative methylation-specific PCR
qMSP-based MGMT promoter methylation testing was prospectively conducted within the CeTeG/NOA-09 trial in central laboratories of MDxHealth (Herstal, Belgium) as reported. 10 In short, DNA was extracted from representative Formalin-fixed Paraffine-embedded (FFPE) sections and sodium bisulfite conversion was performed, followed by real-time MSP using primers for the methylated sequence. The results were normalized to ACTB as reference gene.
For calculation of the MGMT methylation ratio, log2 (1000*mMGMT/ ACTB) was used. In a dichotomized manner, a log2 value of above 2 was considered as MGMT promoter-methylated and below 2 as MGMT promoter-unmethylated. In case of a copy number below 20 for β-actin/ACTB, the result was considered as invalid.

| DNA pyrosequencing
PSQ was performed as reported using modified oligonucleotide primer sequences. 15 Briefly, DNA was extracted from representative FFPE tissue samples. Extracted tumor DNA was treated with sodium bisulfite using the MethylEdge Bisulfite Conversion System (Promega GmbH, Mannheim, Germany). MGMT promoter-methylated and promoter-unmethylated DNA control samples as well as a no template DNA control were run with each experiment. The following oligonucleotide primers were used for amplifying a 99-bp genomic fragment from the MGMT-associated 5 0 -CpG island: MGMT_PSQ_F1 Pyrosequencing was carried out on a PyroMark Q24 system (Qiagen, Hilden, Germany). As sequencing primer, we used MGMT_PSQ_S1:

| Statistical analysis
For survival analysis, all patients of the modified-intent-to-treat (mITT) population as the target population for the primary analysis of the CeTeG/NOA-09 trial (n = 129 5 ) were considered. For multivariate analysis, Recursive Partitioning Analysis (RPA) class, study center and IDH mutation status were considered as covariates. To further explore the correlation between qMSP and overall survival, we analyzed significance levels for dichotomized MGMT promoter methylation via qMSP in a Cox-regression model with RPA class and study centers as co-variates using varying cutoffs. Tests were conducted using the SAS software, R version 3.5.2 (SPSS Inc, Chicago, Illinois).
Survival curves were generated using Kaplan-Meier plots and a logrank test (GraphPad 8.0 Software, La Jolla/San Diego, California).
For calculation of the PSQ and MGMT-STP27 probability of missing patients who would have been included in the CeTeG/NOA-09 trial according to qMSP and, on the other hand, the probability of treating patients with experimental CeTeG/NOA-09 protocol who would have been excluded from the CeTeG/NOA-09 trial based on qMSP, the Bayes' theorem was used. As an example, we show the formula for calculation of the risk considering a sample as methylated by PSQ, but nonmethylated by qMSP.
Correlation of qMSP and PSQ was performed using nonparametric Spearman's rank correlation coefficient (GraphPad 8.0 Software).

| qMSP-based MGMT promoter methylation ratio: analysis of its prognostic potential
In the CeTeG/NOA-09 trial, tumors of patients of the mITT population (n = 129) exhibited a median MGMT promoter methylation ratio of 31.6 (range: 2.5-526.9) as determined by qMSP. To unravel a general prognostic relevance of qMSP ratio, we serially varied the cutoff point for qMSP (dichotomized variable) for the Cox analysis of OS throughout the range of qMSP ratios in the entire mITT population irrespective of treatment arm ( Figure 1). Patients whose tumors had a methylation ratio greater than 4, 9 or 12 showed a significantly superior overall survival, when compared to patients whose tumors had lower MGMT methylation levels (ratios 2-4, 2-9 or 2-12, Figure 2A-D).
We saw a nonsignificant trend toward inferior survival for patients with methylation levels 2 to 25 compared to patients with ratios >25 ( Figure 2D). Beyond a qMSP ratio cut-off value of 25, no association of higher ratios with OS was found in this dichotomized analysis were seen in subgroups separated by different cutoffs (Table S1) (Table 1) and confirmed the effect of the treatment arm as previously reported. 19 Moreover, we did not observe a correlation of MGMT promoter methylation ratio with number of chemotherapy courses (CCNU/TMZ for the experimental and TMZ for the standard arm, Figure S2).
We also studied the treatment arm effect of CCNU/TMZ vs TMZ monotherapy separately in different subgroups defined by the qMSP cutoff ratios 4, 9, 12 and 25 (chosen based on the results of the dichotomized analyses of the mITT population, Figure 1). As shown in Table 2, experimental treatment with CCNU/TMZ showed hazard ratios substantially below 1 in each subgroup, yet significance was mostly not reached in any of the subgroups potentially due to small patient numbers. Interestingly, there was a tendency to lower HRs and thus a more pronounced effect of CCNU/TMZ therapy in subgroups with lower qMSP ratios (2-4; 2-9; 2-12; 2-25; Table 2).

| Concordance between PSQ and qMSP results
We compared the results obtained for the MGMT promoter methylation status by PSQ and qMSP in a subset of patients screened for the CeTeG/NOA-09 trial (n = 76; Table 3; Patient populations studied are outlined in Figure S3). qMSP showed evidence for MGMT promoter CpGs of 12%, 18% and 25%). One tumor was found to be methylated by qMSP (ratio 2.4) but unmethylated by PSQ (mean methylation percentage across the seven CpGs of 2%; Table 3). methylated allele frequency of more than 25% (P = .009, log-rank test; Figure 4A). In an analysis restricted to IDH-wild-type tumors (n = 37, 28 with PSQ methylated allele frequency above 25% and nine with PSQ methylated allele frequency of 8%-25%), median survival was 17.6 months in patients with mean methylation percentages between 8% and 25% and 32.4 months with mean methylation percentages above 25% (P = .04, log-rank test). An additional analysis for the effect of the treatment arm separately for the low and the high PSQ methylation rate subgroup (in analogy to  the concordance between the three methods was complete. In our dichotomized survival analyses (Figures 2 and 4), we found that patients whose tumors exhibited MGMT promoter methylation below certain cutoffs (qMSP ratios up to 25; PSQ mean methylation 25%) had a worse prognosis as compared to patients with tumors above these cutoffs. This was not further supported by the Cox regression analysis with qMSP as a continuous variable, where we found that the qMSP-based MGMT promoter methylation ratio in general was not an independent prognostic factor ( Table 1)  In conclusion, MGMT-promoter-methylated glioblastoma patients whose tumors show a low methylation ratio as defined by qMSP  or a low MGMT promoter methylation as defined by PSQ (8%-25%) tend to have an inferior overall survival, but seem to benefit from CCNU/TMZ treatment nonetheless. Based on concordance rates found for qMSP with PSQ and MGMT-STP27, we conclude that a decision for CCNU/TMZ treatment in patients with MGMT promotermethylated glioblastoma could be based on MGMT promoter methyla-