Pyrosequencing versus methylation-specific PCR for assessment of MGMT methylation in tumor and blood samples of glioblastoma patients

Circulating biomarkers in blood may provide an interesting alternative to risky tissue biopsies in the diagnosis and follow-up of glioblastoma patients. We have assessed MGMT methylation status in blood and tissue samples from unresected glioblastoma patients who had been included in the randomized GENOM-009 trial. Paired blood and tissue samples were assessed by methylation-specific PCR (MSP) and pyrosequencing (PYR). After establishing the minimum PYR cut-off that could yield a significant difference in overall survival, we assessed the sensitivity, specificity, positive predictive value and negative predictive value (NPV) of the analyses. Methylation could be detected in cfDNA by both MSP and PYR but with low concordance with results in tissue. Sensitivity was low for both methods (31% and 38%, respectively), while specificity was higher for MSP in blood than for PYR in plasma (96% vs 76%) and NPV was similar (56 vs 57%). Concordance of results in tissue by MSP and PYR was 84.3% (P < 0.001) and correlated with outcome. We conclude that detection of cfDNA in the blood of glioblastoma patients can be an alternative when tumor tissue is not available but methods for the detection of cfDNA in blood must improve before it can replace analysis in tumor tissue.

MGMT methylation and prognosis. MGMT   outcome, while PYR-plasma results correlated with PFS (p = 0.002) and OS (p = 0.007) but not with clinical benefit (Table 5 and Figs 2 and 3).
Since four tumor samples did not amplify correctly for analysis by MSP but were able to be rescued for PYR, we speculated that PYR-tumor may yield more informative results than MSP-tumor. However, when we compared OS in the group of patients identified as MET by either method with those identified as UNMET by both methods, results were not superior to those obtained with either method alone (Table 5).

Discussion
MGMT methylation status is a well-known predictive and prognostic factor in glioblastoma, and its assessment at the time of diagnosis is an important factor both for clinical trials and for deciding on the optimal treatment strategy. In the present study, we have assessed MGMT methylation in tumor tissue and in blood by MSP and PYR and compared the reliability of the different analyses. Although results by MSP-tumor and PYR-tumor were not completely identical, concordance was high (84.3%; p < 0.001) and both methods provided reliable results. In contrast, assessment in blood was feasible but less reliable, with a high percentage of false negatives in both PYR-plasma and MSP-blood and a lower level of concordance with the results in tumor tissue.
Although several methods are currently available to assess MGMT methylation, there is as yet no agreement about which test should be considered the "gold standard" 8,30 . Comparative studies often lack a previous study setting the optimal cut-off point for PYR related to clinical benefit 31,32 . To further complicate the comparison of results, each method can interrogate different CpG sites in the MGMT promoter region, and there is still no consensus on how many CpG sites should be explored, which are most highly correlated with prognosis, and whether it is better to select them consecutively or randomly. In the PYR analyses in the present study, we explored CpG sites 74-78, which have a good correlation with the prognosis of patients newly diagnosed with glioblastoma [30][31][32][33][34] and which overlap almost exactly with those interrogated by MSP 30,35 .
In our clinical practice, we routinely use MSP to analyze MGMT methylation because it is a well-known technique, it is inexpensive when testing only a few samples, it has demonstrated sensitivity, and its results have been associated with outcome in clinical trials. However, MSP is a not an automatized method, making it difficult to standardize, and results may be influenced by tumor heterogeneity and/or a subjective interpretation. Conversely, PYR has been standardized and by giving a quantitative methylation percentage for each analyzed CpG, it is not subject to individual interpretation of results once the cut-off value has been defined. Different cut-off points have been recommended for PYR, with little consensus on the optimal point 32,36,37 , which can vary according to the CpG sites analyzed and which will ultimately depend on its predictive capacity. Reported cut-off values range from 2.7% to 35% and the number of CpG sites analyzed range from four to 62 38 . In the present study, we first defined the cut-off as the one that identified the maximal differences in OS in our patient population. In fact, however, differences in OS started to be seen at lower levels of methylation. Since the clinical objective behind determining MGMT methylation status is to identify those patients most likely to benefit from treatment with TMZ, we then identified the minimum cut-off that identified any differences in OS. This is in line with a recent pooled analysis of 4041 patients from four clinical trials, in whom MGMT methylation was analyzed by quantitative MSP. Lower MGMT methylation conferred some sensitivity to TMZ, leading the authors to recommend    Table 2. Percentage of cytosine methylation in five CpG sites in tumor, plasma, and serum samples by pyrosequencing (PYR). SD, standard deviation; NC, not calculated. a The optimal cut-off was able to identify maximal differences in OS. The minimum cut-off was able to identify any significant differences in OS. b p-value for differences in overall survival. c All values for normal tissue were below the minimum cut-off for PYR-tumor in glioblastoma tissue. d The minimum cut-off for PYR-serum was not calculated because the cut-off was below the median value.  www.nature.com/scientificreports www.nature.com/scientificreports/ brain, lymphocyte, and colon tissue samples, although all values were under the minimum cut-off identified for glioblastoma tissue.

Results
There was some discordance between MSP-tumor and PYR-tumor results. Eight cases identified as UNMET by MSP were identified as MET by PYR. This could be due to the fact that PYR can detect partial methylation of a given CpG that could be labelled UNMET by MSP 30 . In addition, three cases identified as UNMET by PYR were classified as MET by MSP, perhaps due to an insufficient bisulfite conversion 43 . Moreover, four cases deemed non-evaluable by MSP were evaluable by PYR, probably because less DNA is required to run PYR. These discordances led us to postulate that the two tests could complement each other, so we analyzed differences in OS between patients identified as MET by either test and those identified as UNMET by both tests. However, we found that results were similar to those obtained by each method alone. Therefore, we can conclude that both methods can be reliably used to identify patients as MET or UNMET and we recommend performing PYR after MSP only if MSP does not give an informative result and quality controls fail.
The shedding of cfDNA from tumors to fluids has been known for many years 44 , and multiple studies have shown that tumor-associated alterations can be detected in different fluids at the protein, DNA, and RNA levels using diverse methods, making these liquid biopsies an easier and less invasive way to obtain information on the tumor. Brain tumors have not been an exception to the search for circulating biomarkers 17,26,45 . In the present study, we were able to detect methylation in cfDNA by both MSP-blood and PYR-plasma, but when we examined the capacity of MSP-blood and PYR-plasma to predict MGMT methylation status by the matched tumor assessments, concordance was relatively low, with a sensitivity of 30-40%. The specificity and PPV of PYR-plasma was higher than that of MSP-blood, but false negatives occurred in around 56% of cases by both methods. MGMT methylation in cfDNA has been studied but data on sensitivity, specificity, PPV and NPV are seldom reported 27,46,47 .
The low concordance between results obtained in tissue and plasma could be due to several factors. Firstly, glioblastoma tumors may not shed DNA into the circulation due to the special structure of the blood-brain barrier (BBB) and the blood-brain tumor barrier (BBTB) [48][49][50] . Even though cfDNA can be detected in the blood of glioblastoma patients 51 , the release of DNA to body fluids has been related to tumor size and aggressiveness 21 . All our patients had enhancing measurable disease on the MRI (Table 1) at the time of blood sample extraction and enrollment in the trial. As contrast enhancement lesions on MRI are due to disruption of the BBB, this suggests that these tumors would logically shed DNA, which would not be prevented by a disrupted BBB. Secondly, even though cfDNA is abundant in plasma, only 0.01-10% comes from tumors, while the rest is released from bone marrow (80-90%), or skin and the gastrointestinal tract (5-10%) 21 . Moreover, cfDNA in fluids is found as small fragments (180-200 bp) that may not contain the regions subjected to analyses of MGMT methylation. On the other hand, cfDNA can also be released to CSF and it seems logical that glioblastoma cfDNA would be more abundant in CSF, simply because CSF is in close contact with tumor cells in CNS tumors 52 or because cfDNA could be released through areas without the BBB, such as the circumventricular organs and the choroid plexus 53,54 . In fact, diverse mutations detected in CSF could potentially be used for the diagnosis and follow-up of glioma 52,[55][56][57][58][59] . Several studies have compared results in serum and CSF and found that sensitivity in CSF seems to be higher than in serum 47,55 . However, increased intracranial pressure can be a limitation for the routine assessment of CSF in patients with intracranial hypertension because lumbar puncture carries the risk of cerebral herniation in these cases 60 . Finally, the detection of cfDNA is also limited by the sensitivity of the technique, as semi-quantitative PCR detects only around 1% of cfDNA. Nevertheless, if more sensitive methods for detecting cfDNA in fluids are developed, these limitations could be minimized in the near future 61 .  www.nature.com/scientificreports www.nature.com/scientificreports/ In summary, both PYR and MSP are reliable methods for detecting MGMT methylation in tumor tissue and can be useful for identifying patients likely to benefit from TMZ. MSP can be recommended for use with small numbers of patients, while PYR is more efficient in large numbers of cases and, moreover, can serve as a useful back-up when MSP results are inconclusive 32,62 . In contrast, both methods are imperfect for analysis in blood.
The main limitation of our study is the low number of paired tissue-blood samples that could be compared. Nevertheless, we are convinced that the identification of reliable circulating biomarkers can lead to major changes in our approach to brain tumors, making it essential to continue to search for circulating biomarkers with high sensitivity, specificity, PPV, and NPV. We therefore recommend that biomarker analysis be incorporated into large clinical trials.  www.nature.com/scientificreports www.nature.com/scientificreports/ Methods Patients and samples. Patients proceeded from the GENOM 009 trial 29 , in which unresected glioblastoma patients were randomized to add or not bevacizumab to two cycles of TMZ before radiation therapy with concomitant and adjuvant TMZ. Overall response rate, PFS and OS were primary endpoints of the trial. Results showed a non-significant trend towards improvement in PFS and OS for those patients treated with bevacizumab. A secondary endpoint was the study of MGMT methylation status in blood as compared with tissue in these patients. Before patient inclusion, formalin-fixed paraffin-embedded (FFPE) tumor tissue blocks and paired blood samples were obtained from patients and sent to our center for assessing the methylation status of MGMT.
In addition, we obtained 21 samples of non-tumor tissue proceeding from normal brain (6), lymphocytes (5), and colon (10). MGMT methylation status in these samples was analyzed by PYR.
Written informed consent was obtained from all patients before registration, both for participation in the trial and for molecular tests. The study was approved by the Ethics Committee of the Hospital Germans Trias i Pujol, Badalona, Spain and by the Ethics Committees of each of the participating hospitals. All research was performed in accordance with relevant guidelines/regulations. DNA extraction from tissue and serum/plasma samples. Genomic DNA was extracted from FFPE tissue after macrodissection was performed to assure more than 80% of tumor cells. The material was deparaffinated and DNA obtained using the QIAamp ® DNA micro kit (Qiagen, Hilden, Germany). Venous blood (10 mL) was drawn from each patient into Vacutainer tubes containing SST (serum-separating tube) gel and clot activator (Becton Dickinson, NJ, USA), and for plasma in Vacutainer tubes containing k2E (EDTA). Serum and plasma were isolated after centrifugation at 2,500 rpm for 10 min at room temperature and were stored at −20 °C until analysis. The QIAmp ® Blood Mini-Kit (Qiagen) was used to obtain DNA according to the manufacturer's instructions.
DNA bisulfite conversion. DNA methylation in CpG island of MGMT (Genbank accession number NG_052673.1) was determined using two different methods, MSP and PYR. The first step in both methods was MGMT methylation assessment by PYR. Bisulfite-converted DNA (2 μl) was used to amplify the MGMT promoter with the primers provided by the Pyromark ® Q24 CpG MGMT kit. A PCR was set up with the reactives and conditions given by the PyroMark ® PCR kit. The same CpG sites (74 to 78) analyzed by MSP were analyzed by PYR. Briefly, PCR conditions for tissue DNA were 5 min 95 °C, followed by 45 cycles of (20′′ 95 °C, 30′′ 53 °C, 20′′ 72 °C) and 5 min at 72 °C. In plasma and serum, the number of cycles was increased to 50. PCR products (10 μl) were attached to Streptavidina Sepharose perls (GE Healthcare, Buckinghamshire, UK) and the template strands were purified in the Pyromark Q24 Vaccum Workstation. The purified templates were incubated for 20 min with the sequencing primer provided by the kit and run in the instrument Qiagen Pyromark Q24 System with the reactives of Pyromark Gold Q24 Reagents. All the steps were followed according to the Pyromark Q24 user manual. Sequencing conditions and analysis of results were performed with the Pyromark Q24 sofware 2.0.8. Commercial unmethylated DNA was used to calculate baseline (n = 13).
In case of discordant results between PYR and MSP, MSP was repeated to confirm the discordance. www.nature.com/scientificreports www.nature.com/scientificreports/ Determination of cut-off points for PYR-tumor, PYR-plasma, and PYR-serum results. PYR results were delivered as a percentage of cytosine methylation for each CpG site assessed in tissue, plasma, and serum. We created a new variable using the overall mean of all five CpG sites (Table 2). We calculated the cut-off that yielded maximal differences in OS for each CpG and for the mean (optimal cut-off point) for PYR-tumor, PYR-plasma, and PYR-serum. However, despite having identified the optimal cut-off point for maximal differences in OS, we reasoned that many patients with MGMT methylation status below that cut-off could still benefit from treatment because of a certain degree of methylation. We therefore identified the minimum cut-off point for the mean of all five CpG sites that yielded a significant difference in OS. This cut-off was used for further analyses. If the value was higher than the cut-off, a sample was classified as MET; if it was equal to or lower than the cut-off, a sample was classified as UNMET. All cut-off points were calculated with the Maximally Selected Rank Statistics (Maxstat package of R, version 1.1.442) and confirmed with the Kaplan-Meier method and log-rank test.
Statistical analyses. The epiR package of R was used to calculate the sensitivity, specificity, PPV, and NPV of the different analyses to predict MGMT methylation status as assessed by other analyses. Since in this case, none of the analyses could be considered the established gold standard against which to compare the others, we performed bilateral comparisons of PYR-tumor vs MSP-tumor, MSP-tumor vs PYR-tumor, PYR-plasma vs PYR-tumor, and MSP-blood vs MSP-tumor.
Categorical variables were compared with the χ 2 test or the Fisher's exact test. All patients who started treatment were included in the analyses. Response to neoadjuvant therapy was evaluated by RANO criteria, after two cycles of therapy and before radiation therapy. For the purpose of analysis, responses were grouped as clinical benefit (stable disease, partial response, or complete response) and no clinical benefit (progression) 64 . PFS was defined as the time from inclusion to the first documented progression or death from any cause, while OS was defined as the time from inclusion to death from any cause. Patients who were still progression-free or alive at the date of last contact were censored. Median PFS and OS were calculated with the Kaplan-Meier method and compared using the log-rank test. All statistical tests were two-sided and significance was set at 0.05. These analyses were performed with SPSS v24.0 (IBM).

Data Availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.