Primary mismatch repair deficient IDH-mutant astrocytoma (PMMRDIA) is a distinct type with a poor prognosis

Tumor DNA was extracted from areas with highest tumor cell content using the automated Maxwell system with the Maxwell 16 Tissue DNA Purification Kit or the Maxwell 16 FFPE Plus LEV DNA Purification Kit (Promega, Madison, USA), according to the manufacturer’s instructions. DNA extraction from EDTA–blood was done using the Maxwell RSC Blood DNA Kit (Promega). DNA concentrations were determined using the Invitrogen Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific, Waltham, USA) on a FLUOstar Omega Microplate Reader (BMG Labtech GmbH, Ortenberg, Germany).

DNA methylation profiles were generated using the Infinium HumanMethylation450 (450 k) BeadChip or Infinium MethylationEPIC (850 k) BeadChip array (Illumina, San Diego, USA) according to the manufacturer’s instructions. The data were processed as previously described [14]. T-SNE analysis was performed using the 20,000 most variable CpG sites according to standard deviation, 3000 iterations and a perplexity value of 10.

MGMT promoter methylation status was calculated from the 450 k/850 k data as described by Bady et al. [2] with some modifications; an individual confidence interval was determined for each single probe; if the cut-off value of 0.358 was located in the calculated confidence interval, the MGMT promoter methylation status was defined as being not determinable. Samples with a not determinable MGMT promoter are not included in Fig. 2c.

Copy number profiles were generated from the methylation array data using the ‘conumee’ package in R (http://​bioconductor.​org/​packages/​release/​bioc/​html/​conumee.​html) with additional baseline correction. Summary copy number plots were generated by condensing multiple copy number plots.

For 76 probes (including reference cases), next-generation sequencing was performed on a NextSeq sequencer 500 (Illumina) as described previously [42]. Libraries were enriched by hybrid capture with custom biotinylated RNA oligo pools covering exonic regions of either 130 or 171 genes, respectively. Oncoprint displays exonic and splicing indels and nonsynonymous single-nucleotide variants (SNVs) for selected genes detected in a tumor sample after subtracting low-quality calls and SNVs with a frequency of ≤ 0.001 in the 1000 genomes database (https://​www.​internationalgen​ome.​org/​). Cases 10 and 11 were sequenced using the Ion AmpliSeq™ Cancer Hotspot Panel v2. Case 14 was sequenced with an Ion Torrent NGS custom amplicon panel targeting 56 genes (both Thermo Fisher Scientific). For case 15, exome sequencing was performed. Mutational burden was calculated for samples where sequencing covered more than 0.9 Mb. It indicates the relative number of called exonic and splicing SNVs and indels in a sample per covered megabase. Low-quality calls and SNVs with a frequency of ≤ 0.001 in the 1000 genomes database were excluded. IDH-sequencing was performed as previously described [25].

For cases with sufficient material (n = 11), immunohistochemistry was conducted on 3 µm thick FFPE tissue sections mounted on StarFrost Advanced Adhesive slides (Engelbrecht, Kassel, Germany) followed by drying at 80 °C for 15 min. Immunohistochemistry was performed on a BenchMark Ultra immunostainer (Ventana Medical Systems, Tucson, USA). Slides were pretreated with Cell Conditioning Solution CC1 (Ventana Medical Systems) for 32 min at room temperature. Primary antibodies were incubated at 37 °C for 32 min, followed by Ventana standard signal amplification, UltraWash, counter-staining with one drop of hematoxylin for 4 min, and one drop of bluing reagent for 4 min. UltraView Universal DAB Detection Kit (Ventana Medical Systems) was used for visualization. Primary antibodies were diluted as followed: MutL1 (MLH1, 1:100, DAKO (Agilent), Santa Clara, USA), MutS2 (MSH2, 1:50, DAKO), MutS6 (MSH6, 1:500, DAKO), PMS2 (1:400, DAKO), ATRX (1:2000, BSB3296, BioSB, Santa Barbara, USA), IDH1 R132H (1:2, clone 1 [15]), Olig2 (1:100, Abcam, Cambridge, UK), GFAP (1:2000, Cell signalling, Promega). Stained slides were scanned on the Aperio AT2 Scanner (Aperio Technologies, Vista, USA) and digitalized using Aperio ImageScope software v12.3.2.8013.

For MSI analysis, polymerase chain reaction with fluorescently labelled oligonucleotides was performed to amplify the mononucleotide markers BAT25, BAT26, and CAT25 from tumor tissue DNA, as described previously [20]. Amplified fragments were visualized on an ABI3130xl genetic analyzer (Thermo Fisher Scientific) to detect potential length alterations of the microsatellites. Fragment sizes differing from the normal range of allelic size variation known for the amplicons encompassing BAT25 (108–110 bp), BAT26 (116–118 bp), and CAT25 (146–148 bp) were regarded as microsatellite instability if no normal tissue DNA was available as a reference. Tumors showing microsatellite instability in one or more markers were classified as microsatellite-instable.

Sample sizes (n) and statistical tests are indicated in figure legends. Kaplan–Meier curves were created and log-rank tests calculated using Prism 8 (GraphPad Software, La Jolla, USA). Overall survival was defined as the time between first surgery and death or the last follow-up for all cases except case 17. For case 17 survival was calculated starting from diagnosis of the recurrent tumor as this was the only material available for analyses. For the Kaplan–Meier curve comparing the overall survival of patients with PMMRDIA and IDH-wt high-grade gliomas in CMMRD data from Guerrini-Rousseau et al., 2019 [24] were extracted. Data from patients with anaplastic astrocytoma or glioblastoma were included, excluding IDH-mutant tumors.

Using unsupervised t-distributed stochastic neighbor embedding (t-SNE) analysis of DNA methylation profiles of an extensive set (> 70,000) of CNS and non-CNS tumors, we identified a distinct group of 32 samples clustering closely together, neighboring but not overlapping with IDH-mutant astrocytomas (data not shown). Results of DNA methylation-based CNS tumor classification (version11b4) did not show any matching calibrated scores except for one tumor classified as “methylation class IDH glioma, subclass high-grade astrocytoma”. The histological diagnoses for tumors of this group consisted predominantly of WHO grade III and IV astrocytic gliomas (mainly glioblastoma WHO grade IV). The histological diagnoses of anaplastic oligodendroglioma WHO grade III, primitive neuroectodermal tumor (PNET) WHO grade IV and anaplastic ependymoma WHO grade III were given once each before the WHO 2016 classification update was released. This suggested that the histological appearance is heterogenous and that while an astrocytic phenotype predominates in most cases, the lineage is less obvious in some. In most cases, an IDH-mutation of the tumor was documented in the clinical records.

When restricting t-SNE analysis to IDH-mutant gliomas, this newly identified group clearly separated from all other previously described types and subtypes of IDH-mutant glioma (Fig. 1a). However, similar to conventional high-grade supratentorial IDH-mutant astrocytomas, this group displayed a reduced global hypermethylation compared to low-grade supratentorial astrocytomas and oligodendrogliomas (Supplementary Fig. 1, online resource).

To outline the clinico-pathological characteristics of this epigenetic group, a reference cohort of 185 IDH-mutant tumors representing all previously described epigenetic groups of IDH-mutant glioma was used. The age of patients with tumors of this group was significantly younger with a median age of 14 years (p < 0.0001; Fig. 1b). In a prospective registry of newly diagnosed pediatric CNS tumor patients (age 21 and younger), 33% (5/15) of patients diagnosed with IDH-mutant high-grade gliomas were part of this newly identified group (data not shown). In contrast to all other types of IDH-mutant glioma, there were more female than male patients in this group (Fig. 1c). All but one of the tumors were located supratentorial (Table 1).

Hypermethylation of the MGMT promoter is in general a typical feature of IDH-mutant gliomas, except the recently described subtype of infratentorial IDH-mutant astrocytomas [5]. Tumors of the newly identified group demonstrate the highest frequency of an unmethylated MGMT promoter among all IDH-mutant gliomas (61.3%, 19/31, Fig. 1d).

Histologically, most tumors had a poorly differentiated astrocytic appearance with little cytoplasm and sparse cellular processes reminiscent of small cell astrocytoma or glioblastoma (Fig. 2a). A single tumor exhibited perivascular pseudorosette-like arrangements of tumor cells reminiscent of ependymoma (Fig. 2b). Another tumor presented with an undifferentiated PNET-like appearance (Fig. 2c). Bizarre multinucleated giant cells (Fig. 2d) or oligoid features with perinuclear halos (Fig. 2e, f) were encountered in some tumors. Brisk mitotic activity, endothelial proliferation and necrosis were present in the majority of cases (Fig. 2g, h). Immunohistochemical loss of ATRX expression (Fig. 2k, 69.2%, 9/13) did not appear as frequent as in astrocytoma supratentorial (92%, 46/50, p = 0.05, Fisher’s exact test) and high-grade astrocytoma supratentorial (86.8%, 33/38, p = 0.154, Fisher’s exact test). However, the differences did not reach statistical significance. In all tumors, GFAP and OLIG2 positivity was present (Fig. 2h, i). Tumors showed no or scarce expression of PD-L1 in a small percentage of tumor cells (Fig. 2p–r). In summary, histology shows a broad morphological range without specific features allowing discrimination from other types of high-grade glioma.

To get a deeper insight into the molecular characteristics of this group, we analyzed next-generation sequencing results from 17 cases. Seven cases were analyzed using the Heidelberg 130 gene panel, six cases were sequenced using the Heidelberg 171 gene panel, two cases were analyzed with a 50 gene panel, one case was analyzed using a 56 gene panel and for another case, whole exome sequencing data were available. All cases harbored IDH1 mutations, 90% of which were IDH1-R132H similar to conventional supratentorial IDH-mutant astrocytomas (Fig. 3a). Inactivating stop-gain or frameshift mutations in one of the MMR genes MutL homolog 1 (MLH1), MutS homolog 2 (MSH2), and MutS homolog 6 (MSH6) were detected in 11/17 tumors (Fig. 3b), and where available (n = 6), these mutations could all be identified as germline in origin. For cases 10, 11 and 14 MMR genes were not or only partially covered by the panel sequencing. However, clinical records of case no. 10 describe a personal and a family history of colorectal cancer. Also, case no. 14 and 15 were diagnosed with Lynch syndrome with a known deleterious germline MSH2 mutation and in case no. 8 and 11 CMMRD was known. For case number 17, paraffin blocks were available, and the tumor showed a tumor cell specific loss of MSH6 expression. This was the only tumor of the series which was available at recurrence only, precluding formal proving of primary MMRD. However, clinical documentation of a primary lesion at young age, immuno-histochemical MSH6 deficiency of the tumor cells and a DNA methylation profile indistinguishable from other samples of this group are still compatible with a Lynch-associated case. Of note, detection of mutations in PMS2 by panel-sequencing from FFPE derived DNA is hampered by multiple pseudogenes reducing sensitivity [27]. Taken together, these data suggested that likely all tumors of this group occurred in association to an MMR-deficiency syndrome. Indeed, all but one tumor showed loss of expression of at least one MMR protein (88.8%, 8/9, Fig. 2i–o). The single tumor with retained expression of MMR proteins occurred in a child with known CMMRD, likely due to a missense mutation with retained protein expression (case no. 11).

Considering the cut-off for hypermutation at 10 or more somatic mutations per Mb [12], 58.8% (10/17) of primary MMR-deficient IDH-mutant astrocytomas were hypermutant (Fig. 3c). Microsatellite instability is frequently observed in other MMR-deficient solid tumors [38]. More than half of the primary mismatch repair-deficient IDH-mutant astrocytomas tested were microsatellite instable, which was detected in hypermutant tumors only (Table 1).

Variants in MMR genes in the supratentorial IDH-mutant reference cohort were only detected in pretreated, recurrent tumors except for one case, where the MSH6 variant was found in a primary supratentorial astrocytoma. However, the pathogenic relevance is unclear since this variant was not found in ClinVar or COSMIC and the tumor was not hypermutant (Table 2). If available, we panel-sequenced the respective primary tumors of the recurrent MMR-mutant tumors and found these to be MMR-wildtype and non-hypermutant consistent with a secondary, therapy-associated MMR-deficient hypermutant status. Importantly, these secondary MMR-deficient IDH-mutant gliomas do not display an aberrant DNA methylation profile compared to MMR-proficient conventional IDH-mutant astrocytomas. In restricted t-SNE analyses, primary and secondary MMR-deficient IDH-mutant astrocytomas were completely separated (Fig. 1a). We concluded that epigenetic profiles associate with MMR-deficiency syndromes but not with MMR deficiency per se. Considering these findings, we suggest to designate this novel group as “Primary Mismatch Repair Deficient IDH-mutant Astrocytoma” (PMMRDIA).

Next, we analyzed which genes are recurrently altered in PMMRDIA beside IDH1 and MMR genes by point mutations or small insertions/deletions and which chromosomal copy number alterations are present. Somatic mutations in TP53 were present in all but one PMMRDIA (Fig. 3b). Bi-allelic inactivation of TP53 was evident in the majority of cases, either by the presence of two pathogenic missense mutations or the combination of a pathogenic point mutation and monosomy of chromosome 17p (58.8%, 10/17). ATRX variants occurred in the vast majority of cases (80%, 12/15) most of which were clearly pathogenic. Of note, several cases showed ATRX missense variants of unknown significance (20%, 3/15). Considering a possibly pathogenic relevance of the variants, this may point to a higher frequency of ATRX functional inactivation than estimated by immuno-histochemical loss. Consistent with an astrocytic genomic profile, no promoter TERT mutation was detected and no 1p/19q co-deletion was present. While inactivation of TP53 is typical for the complete spectrum of IDH-mutant astrocytomas, alterations of the retinoblastoma tumor-suppressor gene (RB1) or related pathway components have been implicated in the malignant progression of IDH-mutant astrocytomas. Among the different RB1-pathway alterations recognized, the homozygous deletion of 9p including the CDKN2A/B locus has the strongest association with poor overall survival in conventional IDH-mutant supratentorial astrocytomas [7]. In PMMRDIA, the RB1 gene was affected by point mutations more frequently (23.5% vs. 4.6%) whereas deletions of CDKN2A/B were present less frequently (35% vs. 70%) compared to the reference cohort of high-grade supratentorial IDH-mutant astrocytomas (Fig. 4).

Activating point mutations in platelet-derived growth factor receptor alpha (PDGFRA) were present in 6/17 (35.3%) of PMMRDIA, much more often than in supratentorial, high-grade IDH-mutant astrocytomas (9.1%). Overall, there were fewer chromosomal copy number alterations in PMMRDIA compared to supratentorial high-grade IDH-mutant astrocytomas (4.5 × 10⁸ vs 8.7 × 10⁸, p < 0.0001, unpaired t-test). However, segmental amplifications of 4q including PDGFRA, often associated with a concomitant loss of larger parts of 4q, were observed with a similar frequency in PMMRDIA (30%) and conventional supratentorial IDH-mutant high-grade astrocytomas (35%, Fig. 4). Of note, PDGFRA amplifications and/or activating point mutations were present in about half of all PMMRDIA. Besides one case where a variant of unknown significance was detected, activating mutations in the phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) were found in 5/17 (29.4%) of PMMRDIA, a higher rate as in conventional supratentorial IDH-mutant high-grade astrocytomas (13,6%, 3/22, Table 2). Additional recurrently mutated genes were NF1 (3/17), PTPN11 (2/17) and PTEN (2/17) all of which are coding for important regulators of RAS/PI3K/AKT signaling ([34, 40, 52], Fig. 3d). Segmental amplifications of oncogenes included CDK4 (2/17), CDK6 (2/17), MDM4 (1/17), CCND2 (1/17), MET (1/17) and EGFR (1/17). Taken together, these data suggest that the vast majority of PMMRDIA harbor genomic alterations leading to increased oncogenic RAS/PI3K/AKT signaling in addition to a combined inactivation of TP53 and RB1 tumor-suppressor pathways.

Given the well-known favorable outcome of diffuse IDH-mutant gliomas compared to malignant diffuse IDH-wt gliomas, on the one hand, and the known temozolomide-resistant phenotype of MMR-deficient cells, on the other hand, we addressed the question whether the clinical outcome of PMMRDIA significantly differs from that of other IDH-mutant gliomas. Overall survival (OS) data were available for 19 patients. When comparing Kaplan–Meier survival curves, it was striking that PMMRDIA exhibit by far the worse clinical outcome among all IDH-mutant gliomas (Fig. 5). Mean overall survival was only 15 months, compared to 168.4 months for astrocytoma supratentorial (p < 0.0001), 85.2 months for high-grade astrocytoma supratentorial (p < 0.0001), 76.9 months for astrocytoma infratentorial (p < 0.0001) and not defined for oligodendroglioma (p < 0.0001, Fig. 5) which is known to have the best prognosis among diffuse gliomas. Compared to data from the literature, this very poor OS is well in the range of adult-type glioblastoma, IDH-wildtype [1] and clearly worse than that for glioblastoma, IDH-mutant [31]. There is no difference between overall survival of PMMRDIA patients compared to CMMRD patients with IDH-wildtype high-grade gliomas published by the European C4CMMRD consortium (Supplementary Fig. 2, online resource, [24]). All except one of the 19 patients with PMMRDIA died within 26 months, including the single patient of the cohort with a WHO grade II tumor which had an OS of only 12 months. Most patients were treated with radio-chemotherapy including TMZ or CCNU (Supplementary Table 1, online resource). Patient 7, 13 and 14 that were diagnosed with an MMR-deficiency syndrome were treated with immune checkpoint inhibitors with limited effect. These patients did not have any further known cancer manifestations. Importantly, patients in our cohort presenting with multiple different tumors due to germline MMR defects (e.g. colon carcinoma or non-Hodgkin-lymphoma) deceased caused by complications of their brain tumor. Given the role of MGMT promoter methylation as predictive biomarker for TMZ response in glioblastoma IDH-wt, we compared the patient’s outcome for tumors with and without MGMT promoter methylation but did not find a difference (Supplementary Fig. 3a, online resource). A WHO grade IV associates with a significantly worse survival in conventional IDH-mutant astrocytomas. However, Kaplan–Meier analyses of PMMRDIA of WHO grade IV versus WHO grade II/III tumors did not show a significant difference in OS (Supplementary Fig. 3b, online resource). There was, however, an association between homozygous deletion of CDKN2A/B and shorter OS in PMMRDIA (Supplementary Fig. 3c, online resource).