Introduction
Optimal treatment strategy of patients with high-risk low-grade glioma WHO grade II remains controversial [
45,
46]. The prognosis of patients varies greatly depending on clinical factors (tumor size, patient’s age) and the molecular subtype, oligodendroglioma: mutant for isocitrate dehydrogenase 1 or 2 (
IDH1 or
2; IDHmt) with co-deletion of the chromosomal arms 1p and 19q (codel); or astrocytoma: with (IDHmt), or without IDH mutation (IDH wild type) [
7,
18,
36,
48]. Recent results of the Radiation Therapy Oncology Group trial (RTOG) 9802 phase III trial suggest that early administration of adjuvant chemotherapy [PCV, procarbazine, lomustine (CCNU), and vincristine] following radiotherapy improves overall survival compared to RT alone [
8]. Unfortunately, detailed molecular tumor characteristics are not available for the RTOG trial, key information for adequate tumor classification according to the updated WHO classification 2016 [
35]. The European Organisation for Research and Treatment of Cancer (EORTC) randomized phase III trial (EORTC 22033) was prospectively designed to compare two treatment modalities, and to identify putative prognostic and predictive molecular markers. Initial clinical results have recently been reported [
6]. There was no difference in progression-free survival for patients treated initially with radiotherapy alone or with dose-dense temozolomide [
6]. Molecular subgroup analysis according to WHO classification 2016 showed no difference in outcome between the subpopulation of patients with IDHmt codeleted and IDHmt non-codeleted tumors when treated with radiotherapy. In contrast, in the TMZ-treatment arm, patients with IDHmt codeleted tumors did better than the IDHmt and but non-codeleted subgroup. This implies a molecular difference rendering IDHmt codeleted tumors more sensitive to TMZ. This supports previous studies that reported improved responsiveness of 1p/19q co-deleted tumors to therapy; however, at the time IDH mutations were not yet discovered [
23].
For individual treatment decisions and risk-adapted therapeutic strategies, it is it is crucial to identify the mechanisms and associated molecular factors predicting benefit from the distinct treatment modalities. For very poor prognosis patients an aggressive and combined modality treatment approach may be warranted, while for patients with a favorable prognosis and long-term survival avoidance of long-term neurocognitive toxicities [
14] is of importance.
The majority of low-grade glioma harbor an
IDH1 or 2 mutation, which is associated with a glioma CpG island methylator phenotype (G-CIMP) [
16,
39]. This implies that a large number of genes are epigenetically inactivated by promoter methylation, impacting cancer-relevant pathways and potentially modulating treatment response. Given that the patients in EORTC 22033 were randomized to distinct genotoxic treatments, we hypothesized that variations in DNA repair proficiency, associated with G-CIMP linked aberrant promoter methylation, may explain differences in outcome in the two treatment arms. With a focus on IDHmt LGG, we set out to assess the DNA methylome of DNA damage response (DDR) genes that includes
MGMT, to find predictive factors for treatment. We further aimed at uncovering potential DNA repair vulnerabilities that may be exploited as the “Achilles’ heel” of the tumors, actionable by novel treatment approaches [
21].
Discussion
Here we set out to investigate whether epigenetic inactivation of DDR genes affected the patients’ benefit (PFS, time lapse from initial surgery to treatment failure upon randomization to RT or TMZ therapy) from genotoxic treatment with either RT or TMZ in EORTC 22033. Furthermore, we aimed at uncovering potential DDR pathway vulnerabilities as potential drug targets. Through multidimensional analyses of the molecular LGG dataset of TCGA, we identified 24 candidate DDR genes that are functionally downregulated by aberrant promoter methylation. Importantly, promoter methylation of four of these DDR genes was predictive for benefit (PFS) from either alkylating agent chemotherapy (TMZ) or RT in EORTC 22033.
The most prominent functionally methylated gene was
MGMT that is known as predictive factor for benefit from TMZ in GBM [
22]. Most IDHmt LGG have a methylated
MGMT gene promoter [
4‐
6,
18]. In line, in this study all IDHmt 1p/19q codel LGG, and almost 90% of the IDHmt non-codel glioma were
MGMT methylated [
6]. Due to this nested relationship, the mechanistic impact of
MGMT promoter methylation on treatment-related outcome remains unclear. The two
MGMT CpG probes identified in this study as predictive, are also the ones selected in the MGMT-STP27 classifier to calculate the score, and both are predictive for benefit from treatment with alkylating agents in GBM [
5]. Most importantly, in this study an increased
MGMT methylation score (~ extent of
MGMT methylation) was predictive for benefit under TMZ treatment, but not under RT. The
MGMT score was significantly higher in codeleted vs non-codeleted IDHmt tumors. This is in agreement with the clinical observation of a worse PFS of IDHmt non-codeleted patients when treated with TMZ, while there was no difference when treated with radiotherapy [
6].
There are some notable differences of
MGMT methylation in LGG compared to GBM. In IDHmt LGG the
MGMT methylation score summarizes methylation of two
MGMT alleles, while GBM harbor frequent loss of one allele (> 80%; CHR 10q26) [
4], and the methylation of the retained allele informs on inactivation of the
MGMT gene. Thus, in presence of two alleles, the detection of methylation may indicate methylation of both alleles, or complete methylation of only one allele, leaving the gene on the remaining allele functional. Thus, a high
MGMT methylation score increases the probability of inactivating both
MGMT alleles.
The relevance, and mechanistic implications of epigenetic silencing, of the other three genes identified as predictive for benefit from RT, are less clear. Methylation of three CpGs in the
MLH3 promoter appear predictive for benefit from RT. MLH3 is usually associated with MMR, although it seems to play only a minor role in this process, and may not actually be required [
9,
26]. Germline mutations in
MLH3 have been associated with the Lynch syndrome, although not fulfilling the Amsterdam I criteria, and with an unclear clinical role [
42]. Frequent
MLH3 methylation in IDHmt LGG has been reported previously [
32]; MLH3 may be involved in other cancer-relevant processes [
9].
For the other two genes only one CpG each was associated with outcome and treatment. RAD21 is involved in DSB repair, and haploinsufficiency has been reported to enhance radiosensitivity in a mouse model [
51], in line with a potential predictive value for RT. SMC4 is part of the condensin complexes I and II that are essential for chromosome assembly and segregation, and is involved DSB repair [
34]. Genes encoding members of the cohesion complex are targeted by deletion or mutation in 16% of LGG/GBM, suggestive of a glioma genesis relevant role of the pathway [
11].
The search for biomarkers identifying potential vulnerabilities in repair pathways yielded a number of candidates. Twenty-four functionally methylated DDR genes were confirmed in three independent LGG datasets, and some may open novel treatment options. Most interestingly, recent encouraging results from clinical trials have raised the interest in targeting DNA repair vulnerabilities in cancer. Responses to the PARP inhibitor olaparib have been observed in metastatic prostate cancer with defects of repair genes (mutations, deletions) [
37]. In breast cancer inactivation of
BRCA1 and/or
2 by promoter methylation is currently considered for treatment with the same PARP inhibitor in a trial (clinicaltrial.gov, NCT03205761). Veliparib is currently undergoing phase 3 evaluation in GBM in a randomized Alliance trial (NCT02152982). Along the same lines, PARP inhibitors may be useful for LGG patients with promoter methylation of
MLH3, or
XRCC1 that when defective have been reported to render cells sensitive to PARP inhibition, and are therefore discussed as being potentially amenable to PARP inhibitor treatments [
37,
38].
Of interest,
XRCC1 is located on CHR 19q, hence haploinsufficient in the IDHmt codeleted tumors. Consequently, detection of promoter methylation informs on the silencing of the remaining allele. Furthermore, based on predictions for druggability of DDR genes (see recent reviews by Pearl et al. [
41] and Stover et al. [
47]), five of our functionally methylated DDR genes (
RPA2 SMARCA5, SMC4, XRCC1, mTOR) may warrant further investigations. Targeting of mTOR is currently under investigation in a trial for LGG with or without TMZ treatment (NCT02023905).
The current efforts to promote the HM-450K methylation platform (or the more recent version, EPIC) as a diagnostic tool for classification of brain tumors [
10,
17], will make the here identified markers potentially evaluable in routine diagnostics and may allow validation of the presented results. This diagnostic tool integrates the
MGMT methylation score as part of the MGMT-STP27 classifier [
10]. However, most quantitative assays for
MGMT methylation analysis, such as methylation-specific pyrosequencing should be amenable to determine the extent of methylation. We previously showed good concordance between the HM-450K based MGMT-STP27 classifier and pyrosequencing using the respective cut-offs for glioblastoma [
5].
We have focused our investigations to epigenetically silenced DDR genes, since promoter methylation seems to be quite stable in tumors and may not easily change under treatment [
19,
28]. While mutations in DDR genes are exceedingly rare in chemo-naive LGG, with the exception of
TERTp and
ATRX [
48], there are other mechanisms attenuating relevant DNA repair systems. Of particular interest for lower grade glioma is the recently described inhibitory effect of the oncometabolite 2-hydroxy glutarate (2HG) on the α-ketoglutarate-dependent ALKBH repair enzymes that are involved in direct DNA repair, including TMZ induced lesions [
24,
49]. 2HG, produced by the neomorphic function of the IDH1 and 2 mutants is accumulated to high concentrations in the respective tumors [
13] and may confer sensitivity to alkylating agent chemotherapy and protracted natural history.
In conclusion, our analyses suggest that a high MGMT methylation score predicts PFS in TMZ-treated patients with IDHmt tumors regardless of the 1p/19q status. This information may guide clinical decision-making for individual patients, and in particular when considering deferring radiotherapy in patients with a better prognosis aiming at avoiding or delaying potential neurocognitive toxicity.
The limitation of the present retrospective study is the low statistical power and the current lack of a comparable validation set. Data for OS are not available yet. However, the development of predictive markers to allow treatment de-escalation are important in the future, personalizing treatment strategy (single vs combined modality, choice of modality, novel targets) needs to be risk-adapted and is of particular importance in good-prognosis disease.
Compliance with ethical standards