Introduction

Diffuse LGG are infiltrative brain tumors which include World Health Organization (WHO) grade II astrocytomas, oligodendrogliomas, and oligoastrocytomas [32]. Surgical resection is the primary therapeutic intervention, though LGG recur and may undergo malignant progression to a higher histological grade including grade IV secondary GBM. Therefore, in patients with clinical risk factors [41, 50], postoperative adjuvant treatment is often utilized. The addition of TMZ to postoperative radiotherapy prolongs progression-free survival (PFS) and overall survival (OS) in high-risk LGG patients, and chemotherapy instead of irradiation might be as effective [3, 45] and defers the risk of late radiation-induced cognitive deterioration [14]. Moreover, postoperative TMZ or irradiation of LGG has been associated with improved quality of life, better seizure control, and longer progression-free survival [6, 29, 37, 42, 48].

Temozolomide is an alkylating agent that methylates the O 6 position of guanine. The DNA repair protein MGMT removes O 6-methyl groups induced by TMZ. Initial studies assaying DNA methylation in the MGMT gene body rather than promoter showed a direct correlation between MGMT methylation and expression [19, 40]. When the MGMT promoter is hypermethylated, however, MGMT expression is decreased and TMZ-induced DNA damage persists [12, 13]. O 6-methylguanine pairs with thymine instead of cytosine during DNA replication. MMR can recognize and repair these mismatches through MutS and MutL complexes. MSH2 and MSH6 form the MutSα complex, which identifies base–base mismatches and small insertion–deletion loops (IDLs). MSH2 and MSH3 form the MutSβ complex, which identifies large IDLs. MutS complexes directly with MutL, an MLH1/PMS2 dimer, to the site of DNA damage [20, 21]. Removal of the thymine that is base paired with O 6-methylguanine is followed by repair synthesis that reinserts thymine, leading to repeated attempts to repair the same base. This futile cycling of repair has been linked to DNA double-strand breaks and apoptosis, the apparent mechanism of TMZ-induced cytotoxicity [16].

Inactivation of the MMR pathway is a mechanism of resistance to TMZ in primary GBMs and also leads to TMZ-induced mutagenesis [8, 18, 26, 62]. In MMR-deficient cells, the base pairing of O 6-methylguanine with thymine persists, and upon DNA replication results in nucleotide transitions from guanine to adenine. TMZ-associated hypermutation has been observed in GBM [9–12], in cells treated with TMZ in vitro [8] and in unpaired post-treatment tissue samples [1, 5, 15, 16, 26]. In contrast to MMR, the impact of MGMT activity on the relative amount of cytotoxicity versus mutagenicity is much less clear. Furthermore, while MGMT methylation is associated with longer overall survival in GBM patients treated with TMZ [25], it is unclear whether this biomarker has the same prognostic value in patients with IDH1 mutated LGG [17, 53, 58].

We recently identified hypermutation in a subset of TMZ-treated recurrent GBMs that arose from IDH1-mutant astrocytic LGG [28]. Post-TMZ recurrences had a 39- to 133-fold increase in the mutation rate relative to their treatment-naïve initial LGG, more than 98 % of which are C>T/G>A mutations which are associated with TMZ-induced mutagenesis (Supplementary Table S1) [5]. TMZ-associated mutations resulted in deregulation of RB-mediated cell cycle control and hyperactivated AKT–mTOR signaling, suggesting TMZ-induced hypermutation may drive malignant progression. However, it is unclear why hypermutation developed in only six of the ten LGG treated with TMZ. To better understand the mechanism of hypermutation, here we examined the stepwise development of DNA repair deficiency and subsequent TMZ-associated hypermutation using a cohort of 34 initial LGG and their patient-matched recurrence, including 23 pairs for which exome sequencing data were available. Because TMZ-induced hypermutation in LGG was associated exclusively with GBM recurrence, this study is important for understanding and ultimately avoiding TMZ-associated hypermutation and malignant progression.

Methods

Sample acquisition

Patient inclusion in this cohort was dependent upon (1) an initial diagnosis of WHO grade II diffuse astrocytoma, oligodendroglioma, or oligoastrocytoma; (2) available tumor tissue from an initial tumor and a subsequent recurrence; (3) information on post-surgical treatment. A majority of the samples have been used in previous studies (Supplementary Table S2) [28, 55]. Tumor samples were fresh frozen or formalin-fixed paraffin-embedded (FFPE) tissues. Sample use was approved by the Committee on Human Research at UCSF; the Ethics Committee of the University of Tokyo; and the Medical Ethics Committees of the Dutch hospitals VU University Medical Center Amsterdam, Radboud University Medical Center Nijmegen, Isala Klinieken Zwolle and Erasmus Medical Center Rotterdam, and the Linköping University Hospital, Sweden.

DNA isolation

Genomic DNA from tumor and normal tissue samples of patients 01-38 was either extracted with the QIAGEN FFPE DNA extraction kit (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions or isolated by a standard phenol chloroform extraction as previously described [28]. FFPE blocks of initial tumor and recurrences of patients 90–302 were cut into sections of 3–5 μm thickness for pathological evaluation on hematoxylin and eosin-stained slides. For each sample, an area was delineated that contained >60 % tumor cells. The corresponding area on subsequent sections of 10 μm was used for DNA isolation extracted with the QIAGEN FFPE DNA extraction kit [55].

Bisulfite treatment, PCR, cloning and sequencing of MGMT

O 6-methylguanine-DNA methyltransferase methylation status was assessed for all patients (Fig. 1a and Fig. S1) [24]. DNA (>100 ng) was bisulfite treated for 2.5 h with the EZ DNA methylation Gold kit (Zymo Research, Irvine, California) according to the manufacturer’s instructions. Bisulfite-converted DNA was amplified by PCR using the following primers corresponding to the MGMT promoter: forward GGATATGTTGGGATAGTT and reverse ATCGTTAATAAGTCAAGCTC. Gel extraction of the amplified DNA was performed with the QIAEXII gel extraction Kit (Qiagen, Germantown, Maryland). Four to six microliters of PCR product was cloned using a pCR2.1/TOPO TA sequencing kit (Invitrogen, Carlsbad, CA, USA). Individual bacterial clones were subjected to PCR using vector-specific primers and a minimum of nine individual PCR clones were sequenced per tumor sample. Bisulfite sequence data of the MGMT promoter were analyzed with BISMA [47]. The bisulfite conversion rate was monitored in all reactions at non-CpG cytosines, which are typically unmethylated and converted. For comparison, the methylation status of the MGMT promoter in bisulfite-treated DNA was also determined in a subset of the samples by standard, non-quantitative methylation-specific PCR (MSP) [16].

Fig. 1
figure 1

Comparison of MGMT methylation in different assays and across tumor regions. a Position of the MGMT bisulfite amplicon (light blue) encompassing the 23 CpGs assessed in this study, the position of the MGMT CpG island (green), and the enhancer region encompassing CpGs 10–16 (dark blue). TSS transcription start site. b Distribution of histological subtypes and grades of the recurrent tumors in the three groups. c Comparison of the binary outcome of MSP (x axis) to MGMT methylation level of CpGs 10–16 determined by bisulfite, PCR and sequencing of 10 or more independent clones (y axis). d The degree of variation in methylation levels was determined in replicate experiments from independent aliquots of the same genomic DNA isolation (y axis) with bisulfite and sequencing of the MGMT promoter in individual samples (x axis). e MGMT methylation levels (y axis) in spatially distinct regions of individual tumors (x axis). Sample designations are the patient (p) number followed by 0.1 for initial tumor and 0.2 for the recurrent tumor

Full size image

Identification of somatic mutations and copy number aberrations in MGMT and MMR genes

The identification of MMR pathway alterations was limited to those for which sufficient tumor DNA and matched normal DNA was available for exome sequencing. The mutational and copy number status of MGMT as well as the key MMR pathway genes MLH1, MLH3, MSH2, MSH3, MSH5, MSH6, PMS1, and PMS2 [27] were assessed from the exome sequencing data. For patients 01–24, somatic mutations and copy number aberrations in genes of interest were identified as previously described [28]. Nine new exome sequencing datasets were also generated for this study using the Agilent SureSelect Target Enrichment System Protocol (Version 1.0 September 2009) with the SureSelect Human All Exon 50 Mb kit (Agilent Technologies) according to the manufacturer’s instructions. Paired-end reads of 76 or 100 bp in length were generated from Illumina HiSeq 2000 or 2500 instrumentation. Paired-end sequencing data from exome capture libraries were aligned to the reference human genome (build hg19) with the Burrows–Wheeler Aligner (BWA) 0.5.10 [31]. Single-nucleotide variants (SNVs) were detected with MuTect [11], and indels were detected with Pindel [61], followed by custom filters to remove false positives [28]. All candidate mutations were subsequently validated with PCR amplification of the target region from tumor and matched normal genomic DNA followed by conventional Sanger sequencing. Copy number segmentation was performed with an adaptation of circular binary segmentation (CBS) [28, 56]. We identified germline heterozygous SNPs from the matched normal exome of each patient tumor using the UnifiedGenotyper [34]. From only those SNPs present in dbSNP (Build ID: 132) (http://www.ncbi.nlm.nih.gov/SNP/) and with a coverage level of 10 or more reads, we calculated their minor allele frequency in all exomes of each patient (initial tumor, recurrence, and patient-matched normal) and used these to infer genomic regions with loss of heterozygosity (LOH). Regions of LOH were then correlated with DNA copy number alterations. For patients 171–296 copy number aberrations at genes of interest were identified from low-coverage whole-genome sequencing as previously described [49, 55]. Copy number segmentation was performed by CBS [28] and gains and losses were identified using CGHcall [54].

Significance tests

For differences in methylation, the initial comparison between subgroups was done using the Kruskal–Wallis test (the nonparametric alternative to ANOVA), followed by subsequent post hoc testing using the Wilcoxon rank-sum or signed-rank test (on data from tumor pairs) for two group analyses. P values below 0.05 were considered statistically significant.

Results

Histological features and disease course of the cohort

Clinical and genomic data for many of the paired initial and recurrent tumor samples used in this study have been described previously [28, 55]. In total, we studied 87 samples from 34 LGG patients. Samples from spatially distinct regions of the tumor were available for six surgeries. Ten of the 34 patients received adjuvant TMZ. Accordingly, we divided the cohort into three groups based on the clinical and exome sequencing data; patients with recurrent tumors displaying a TMZ-associated hypermutator phenotype (TMZ-HM, n = 6), TMZ-treated patients without a TMZ-associated hypermutator phenotype (TMZ-non-HM, n = 4) and patients not treated with TMZ (non-TMZ, n = 24) (Supplementary Table S2). While the overall size of the cohort is relatively modest, very few studies have reported genomic and epigenomic evolution in similarly sized cohorts of paired initial LGG and recurrent tumors [57]. In the TMZ-HM group, all six tumors recurred with GBM histology. Relative to the TMZ-HM group, the number of recurrent tumors with GBM histology was variable in the TMZ-non-HM group and non-TMZ group, (ANOVA, p value 0.013) (Fig. 1b).

MGMT methylation level is similar across assays, technical replicates and spatially distinct samples

Methyl-specific PCR (MSP) is used in clinical tests for MGMT methylation status and yields a low-resolution, non-quantitative binary call of methylated or unmethylated. However, MGMT methylation levels in tumors span a full range from unmethylated to fully methylated at each CpG. We therefore compared MSP to a more quantitative, single-CpG resolution method involving bisulfite treatment, PCR, cloning and sequencing in 18 samples. The genomic region assessed by the bisulfite sequencing approach assesses methylation level at each of 23 CpG sites in the MGMT promoter and enhancer including the region covered by the MSP assay which spans CpG sites 10–16 [2, 33], a previously reported differentially methylated region 2 (DMR2) covered by CpGs 3–20 [33], and CpG site 13 which has prognostic value in GBMs [2] (Fig. 1a). In eight samples, classified as MGMT unmethylated by MSP, median methylation level was 25.4 % (range 1.6–28.6 %), and in ten samples that were methylated according to MSP, median methylation level was 37.4 % (range 6.5–98.6 %) (Fig. 1c). To test the reproducibility of bisulfite sequencing approach, experiments of seven samples were repeated on independent aliquots from the same genomic DNA isolation. Very little variation in methylation level was observed between replicate experiments (Fig. 1d). An analysis of spatially distinct regions in samples obtained from the same surgery revealed that there was also very little variation in MGMT methylation levels within a tumor at a given time point (Fig. 1e). This limited intratumoral heterogeneity of MGMT methylation level provided evidence that the result of a single sample was likely to be representative for LGG, as previously shown in GBM [15, 21, 23].

Increase in MGMT methylation level associated with temozolomide-induced hypermutation

MGMT methylation levels varied widely between patients. Across the whole cohort, the median methylation level of initial tumors was 29.7 % (range 6.1–70.9 %) and the median methylation level of recurrent tumors was also 29.7 % (range 3.9–79.1 %), (p value 0.49) (Supplementary Material). Between the three subgroups, median methylation levels in initial tumors were not significantly different (TMZ-HM 38.3 % vs. TMZ-non-HM 17.8 %, p value 0.15; TMZ-HM vs. non-TMZ 33 %, p value 0.33), while overall methylation level in recurrent tumors of the TMZ-HM group (median 55 %) was significantly higher compared to recurrent tumors of the TMZ-non-HM (median 20 %) and non-TMZ (median 28 %) groups (p value 0.013 and 0.03, respectively) (Fig. 2). Patient 24 was not included in this analysis, as methylation data were confounded by low tumor purity in the initial and first recurrent tumor, complicating the interpretation.

Fig. 2
figure 2

Significantly elevated MGMT methylation in TMZ-associated hypermutated recurrent tumors that arise from LGG. MGMT methylation levels in initial (green) and recurrent (yellow) tumors of three patient subgroups. non-TMZ patients not treated with TMZ, TMZ-non-HM patients treated with TMZ without a hypermutated recurrent tumor, TMZ-HM patients treated with TMZ with a hypermutated recurrent tumor

Full size image

We explored how methylation levels changed over time by performing paired analysis in initial and recurrent tumors of the three subgroups TMZ-HM, TMZ-non-HM, non-TMZ. The change in methylation level from initial to recurrence in the TMZ-HM group was non-significant but showed a trend (p value 0.063). This is supported by the consistent increase in methylation level in this subgroup. This pattern was significantly different from the variable patterns of change over time in the TMZ-non-HM and non-TMZ groups (TMZ-HM vs. TMZ-non-HM p value 0.050; TMZ-HM vs. non-TMZ p value 0.005) (Fig. 2). Eleven of 23 individual CpGs were significantly more methylated in the recurrent tumors of the subgroups TMZ-HM vs. TMZ-non-HM (CpGs 1–6, 8–10, 12 and 13, p values 0.012–0.044).

Evolution of DNA repair deficiency in the TMZ-HM group

In the initial tumors of the TMZ-HM group (n = 6), no MMR gene mutations were detected, however, DNA repair may have been impaired by the heterozygous loss of MGMT in patient 01 and MLH1 in patient 10. In contrast, five of the six TMZ-HM GBMs contained a TMZ-associated mutation in one of the MMR genes, concurrent with deletion of the other allele or deletion encompassing another MMR gene or MGMT (Fig. 3a, b, Fig. S2). We also identified a clonal TMZ-associated mutation in MGMT of unknown significance in the recurrent tumor of patient 18. In the initial and recurrent tumors of the TMZ-non-HM (n = 4) patients, MMR pathway genes were intact, but heterozygous loss of MGMT was detected in the initial tumor of patient 11 and the recurrent tumor of patient 17. Interestingly, the recurrent tumor of patient 11 grew out from an earlier cell that retained both alleles of MGMT, while the recurrent tumor of patient 17 had decreased levels of DNA methylation at MGMT (initial 28.3 %, recurrence 12.7 %), indicating that in both cases MGMT levels may not have been impaired during TMZ treatment and recurrence. In the non-TMZ group (n = 24), mutational analysis was performed in patients of which sufficient DNA from matched normal and initial and recurrent tumor was available (n = 7). In these seven cases, only one MMR mutation was detected, an MSH3 mutation in the initial tumor but not in the recurrent tumor of patient 07. DNA copy number status of MGMT and MMR genes was available for 21 of 24 patients in the non-TMZ subgroup. Genomic loss affecting the MGMT region was detected in five initial and seven recurrent tumors of the non-TMZ subgroup. Similarly, deletion encompassing an MMR gene was shared between the initial and recurrent tumors of five cases, while four patients acquired a deletion encompassing a MMR gene at recurrence (Supplementary Table S3). As these patients did not receive TMZ, it is not known how TMZ treatment may have affected MGMT methylation levels at recurrence, or if the identified genetic alterations to MGMT and MMR genes may indicate a susceptibility to hypermutation.

Fig. 3
figure 3

Sequential acquisition of DNA repair deficiency exclusively in LGG patients treated with TMZ and had a hypermutated recurrence. a Exome-derived copy number status and germline variant allele frequency across chromosome 2 encompassing MSH2 and MSH6 in the initial and recurrent tumor of patient 18. b Methylation level of MGMT, mutations and copy number status of MGMT and MMR-related genes in initial LGG and paired recurrent tumors. The panel includes patients for whom exome sequencing data were available, Asterisk indicates samples with a proportion of tumor cells lower than 50 %

Full size image

Discussion

We compared spontaneous and treatment-associated evolution of DNA repair deficiency in a cohort of 34 initial LGG and their patient-matched recurrences. Our data suggest MGMT and MMR-mediated DNA repair may be compromised by sequential and coincident loss of heterozygosity, methylation [44], and TMZ-associated mutation, although repair activity could not be tested directly. Considering prior studies of TMZ-treated GBM patients [1, 8, 26, 62] and cells treated with TMZ in vitro [5], our results suggest that TMZ-induced hypermutation is the consequence of a TMZ resistance mechanism in LGG. This putative mechanism is not fully understood, but may be induced directly by the mutagenic action of TMZ on DNA repair genes, in combination with pre-existing and concurrent copy number alterations in cells with a higher level of MGMT methylation. The resistance mechanism appears to involve a switch from toxicity to tolerance of TMZ-induced DNA damage. The sequential acquisition of genetic and epigenetic change in MGMT and MMR genes in the TMZ-HM group differs notably from the patterns in patients who did not received TMZ, and in TMZ-treated but not hypermutated patients.

We observed a consistent increase over time in MGMT methylation level, which was not detected in LGG patients without a TMZ-associated hypermutator phenotype. The apparent positive selection of MGMT hypermethylated cells and a corresponding decrease in MGMT expression may predispose a cell to persistent O 6-methylguanine lesions and acquisition of MMR gene mutation, enabling hypermutation from subsequent rounds of TMZ treatment (Fig. 4). MGMT activity also may be decreased by TMZ treatment itself, as the MGMT protein is not regenerated following repair [52].

Fig. 4
figure 4

A working model of the effect of an impaired MMR system on clonal outgrowth of MGMT methylated cells during acquisition of the TMZ-associated hypermutator phenotype and TMZ-associated malignant progression. Left: when MMR is intact, TMZ treatment induces cell death in MGMT methylated cells. The histology of the recurrent tumor is variable. Right: when MMR is deficient, TMZ treatment fails to induce cell death and MGMT methylated tumor cells may expand, become hypermutated, and undergo malignant progression to GBM

Full size image

Other studies have addressed temporal changes of MGMT methylation in smaller cohorts of grade II astrocytomas [22, 30, 35] and GBMs [7, 9, 38] with MSP only, and without mutational and copy number analysis. In the present study, bisulfite sequencing of the MGMT promoter in 34 paired initial and recurrent tumors enabled detailed, quantitative analysis of temporal evolution in individual patients. Given the sample size, a meaningful comparison of MGMT methylation change was not possible for GBM recurrences that were HM (n = 6) versus GBM recurrences in the TMZ-non-HM subgroup (n = 1). We observed a distinct pattern of increased MGMT methylation level between initial and patient-matched, TMZ-treated and hypermutated recurrences.

Exposure of cells to a mutagen such as TMZ will result in a different set of mutations in each cell within the population. However, our detection of somatic mutations in MMR-related genes in hypermutated DNA derived from bulk samples strongly suggests that these recurrences are derived from clonal expansion from a very small number of hypermutated cells [26]. The mutations are predominantly C>T/G>A transitions, the type of mutation known to be induced by TMZ treatment. Literature on the functionality of these mutations varies, for example somatic mutations MLH1 P648L and P640S in the TMZ-treated recurrent tumors of patients 1 and 10 also occur in the germline of families with hereditary nonpolyposis colon cancer, significantly affect MLH1 protein function, and are predicted to be pathogenic [10, 20, 43, 51]. The splice site mutation in MSH3 of patient 18 leads to a single-nucleotide shift of the splice acceptor site, resulting in an out-of-frame transcript and premature truncation of the protein. However, the role of MSH3 in cancer is less clearly defined [39]. In MMR-deficient cells, futile cycling of MMR repair does not occur, enabling this type of mutation. The C>T/G>A mutation also occurs spontaneously, however, the extreme number of new mutations, the strong bias towards C>T/G>A versus other mutations, and the occurrence of hypermutation after TMZ treatment but not in patient-matched pre-treatment samples suggests TMZ is the predominant source. The proportion of tumors developing a hypermutation profile after TMZ treatment in our series is 60 % (6 out of 10). This cohort and others [1, 26] are too small to determine the actual incidence of hypermutated diffuse gliomas after alkylating agent chemotherapy.

Biomarkers of susceptibility to TMZ-associated hypermutation could have significant clinical value. Rare germline and somatic MSH6 mutations that might affect how cells respond to TMZ have been detected in patients with untreated anaplastic oligodendrogliomas and GBMs [36, 46]. Within our small cohort, we found that loss of heterozygosity spanning MMR genes was unique to the TMZ-HM group relative to TMZ-non-HM group. In three TMZ-HM patients, the initial tumor showed deletion of MGMT or an MMR gene. The copy number data of two of the other initial tumors from the TMZ-HM group were ambiguous. In a study of MMR protein expression assessed by immunohistochemistry decreased expression was observed for MLH1, MSH2 and MSH6 in a subset of initial low- and high-grade astrocytomas, but DNA copy number status and paired recurrences were not assessed [46]. A larger cohort of paired samples will be needed to determine if loss of heterozygosity of MGMT and/or MMR genes in initial tumors has predictive or prognostic value. Contrary to primary GBM, where copy number loss of the entire chromosome 10 is a frequent event [4], we observed variability in the size of the region lost in initial LGG and secondary GBM in our cohorts [55]. MGMT hypermethylation and corresponding impaired MGMT activity prior to TMZ treatment could also be a predisposing factor, but we did not detect statistically significant differences when analyzing MGMT methylation level alone between the initial tumors of the TMZ-HM and TMZ-non-HM subgroups. An alternative hypothesis is that, because TMZ-HM tumors appear to derive from a very limited number of cells, MGMT methylation in a small number of cells in the initial tumor may allow positive selection and hypermutation. Other studies with variable designs, and predominantly examining HGGs, were also unable to identify a correlation between MGMT methylation and MMR status [18, 36]. Similar to GBM [15, 21, 23], variation in MGMT methylation levels among multiple regions of the initial LGG of our patients was negligible, suggesting that single samples of the initial and recurrent tumor may be sufficient to elucidate temporal patterns. However, because the TMZ-HM group had an increased level of MGMT methylation relative to more variable patterns in the other groups, recurrences in the TMZ-HM group may exhibit greater intratumoral heterogeneity if the initial tumor resection was incomplete and sampling at recurrence included hypermutated and non-hypermutated regions.

The results presented here and in prior studies [1, 5], along with the well-established mechanisms of DNA repair by MMR and MGMT, further suggest that compromised DNA repair contributes to the onset of hypermutation and subsequent malignant transformation. Taken together, the data suggest a working model in which a hypermutated tumor arises through clonal expansion of cells with high levels of MGMT methylation, pre-existing loss of heterozygosity of a key MMR gene and/or MGMT, and TMZ-associated mutation in MMR genes. Tumor tissue and clinical data from LGG patients participating in clinical trials with TMZ treatment will be required to follow-up these initial findings [59, 60] and to assess the clinical relevance of the TMZ-associated hypermutator phenotype.

Data availability

Whole exome sequence data are uploaded to the European Genome-phenome Archive (EGA) for patients 1–23 (accession number EGAS00001000579), and shallow whole-genome sequencing data of patients 90–296 (EGAS00001000643). Data of patient 24 was deposited to the Japanese Genotype–phenotype Archive under accession number JGAS00000000004.