Immunogenomics of Hypermutated Glioblastoma: A Patient with Germline POLE Deficiency Treated with Checkpoint Blockade Immunotherapy

Glioblastoma is the most aggressive primary brain tumor in adults, with a median survival of 12 to 14 months following standard-of-care treatment (1). There is growing enthusiasm for treating malignant brain tumors with cancer immunotherapies due to successes in other cancers and the growing realization that the central nervous system (CNS) is not immunoprivileged, as has long been held (2). However, it is not clear what biomarkers will predict responses or whether lymphocytes will infiltrate CNS tumors following these therapies.

Work at the nexus of genomics and immunology has established the cancer immunogenomics concept in which expressed nonsilent tumor-specific mutations represent neoantigens recognizable by the immune system (3–5). Because mutational burden is the engine for neoantigen production, several groups have correlated mutational load with response to immunotherapy. Patients with greater neoantigen load exhibited improved clinical responses to checkpoint blockade immunotherapy (6–8).

Unlike carcinogen-induced tumors, glioblastomas typically harbor fewer than 100 exome-wide mutations (9), with only a subset representing candidate neoantigens. However, a subset of hypermutated glioblastomas has been described in which mutational loads can be 10- to 50-fold higher than average (10, 11). This genotype is observed in approximately 20% to 30% of recurrent glioblastomas following temozolomide treatment and has been ascribed to acquired somatic mismatch-repair deficiency (12, 13). A hypermutated genotype was also recently described in four adults with newly diagnosed glioblastomas associated with somatic POLE mutations, each carrying more than 100-fold the average number of mutations (14). However, neither the genomic landscape of an adult glioblastoma arising from a germline POLE mutation nor its recurrent tumors has been reported.

We describe an adult patient with germline POLE deficiency who developed a hypermutated glioblastoma and underwent additional resection of two metachronous spinal metastases. The patient was treated with pembrolizumab, an inhibitory monoclonal anti–programmed cell death 1 (PD-1) antibody, and had evidence of a clinical and immunologic response. Herein, we profile the genomic landscape and remodeling of the subclonal architecture of this hypermutated glioblastoma during the transition from primary tumor to metastases. Moreover, we demonstrate robust immune infiltration of these CNS tumors following pembrolizumab therapy.

A 31-year-old man with a history of colonic polyps presented to the hospital after experiencing a new-onset seizure. A brain MRI showed a 5.0 × 5.0 × 6.0 cm enhancing lesion in the left frontotemporal lobe and a T2/FLAIR hyperintense lesion in the frontal horn of the right lateral ventricle (Fig.1A). The patient underwent a left frontotemporal craniotomy for a near-gross total resection (>95%) of the left-sided lesion. Histopathologic analysis revealed a diagnosis of glioblastoma (WHO grade IV) with primitive neuroectodermal tumor (PNET) features (Fig. 1A). Immunohistochemical stains for IDH1^R132H and BRAF^V600E were negative, and O⁶-methylguanine DNA methyltransferase (MGMT) methylation was equivocal in that 10 of 17 sites were positive at one institution but assessed as indeterminate during central pathology review for clinical trial eligibility at another institution. Given the personal and family history of numerous colonic polyps, the patient underwent analysis for germline DNA mismatch–repair deficiency and tested positive for a POLE mutation encoding the L424V substitution, previously implicated in colorectal cancer susceptibility (15). Genomic profiling of the resected tumor was performed by Foundation Medicine (16), and of the 315 genes assessed, 165 (52.4%) contained mutations.

Before starting treatment, a spine MRI to evaluate for metastatic lesions was performed and showed no evidence of disease. The patient received radiotherapy to 60 Gy in 30 fractions over 41 days combined with temozolomide therapy. After chemoradiation treatment, the patient was treated with higher-dose maintenance temozolomide. Four weeks after initiating maintenance temozolomide, the patient developed difficulty walking. Repeat spinal MRI revealed an intradural/extramedullary enhancing lesion from C7-T2 (Fig. 1A). He underwent a C7-T2 laminectomy for gross total resection, and histopathologic analysis revealed glioblastoma with PNET features consistent with a “drop” metastasis (Fig. 1A). The patient's leg strength improved, and the C5-T3 region was treated with radiation to 50.4 Gy. Due to a report of clinical responses to anti–PD-1 checkpoint blockade immunotherapy in hypermutated mismatch repair–deficient colon cancers, the patient was started on pembrolizumab (8). Before treatment, a repeat spine MRI demonstrated 2 intradural enhancing foci dorsal to T7-8 concerning for disease.

Following two doses of pembrolizumab over 3.4 weeks, the patient again developed difficulty walking. Spinal MRI demonstrated increased size of the enhancing T7-8 lesion, causing local compression. The patient underwent a T7-8 laminectomy for tumor resection, and histopathologic analysis confirmed a glioblastoma metastasis but with significant inflammatory changes. The patient continued on pembrolizumab and underwent radiotherapy to the T6-L4 region treating to 45 Gy. The patient's strength improved to the point that he was walking. Four months after starting pembrolizumab, a nonenhancing left cerebellopontine angle lesion and several small nonenhancing intraventricular lesions were being monitored.

To understand the genomic landscape of glioblastoma arising in the setting of germline POLE mutation, we performed DNA and RNA sequencing (RNA-seq) on the patient's pretreatment frontotemporal glioblastoma and both spinal metastases. All tumors were hypermutated, containing 17,276 to 20,045 mutations, over half of which were nonsynonymous (Fig. 1B; Supplementary Table S1), and all tumors carried a nearly identical spectrum of mutations, with the majority being C>G alterations (Fig. 1C; Supplementary Fig. S1). Few copy-number alterations (CNA) were observed, and gene fusions were predominantly found in the cervical metastasis (Supplementary Fig. S2; Supplementary Tables S2–S3). No CNAs or fusions were shared among all three samples, suggesting that the few events identified are subclonal. As the patient's metastases developed after temozolomide treatment, we examined RNA-seq data to determine whether MGMT, which drives temozolomide resistance, was more highly expressed in post-temozolomide lesions (Supplementary Table S4). Both metastases harbored nearly 4-fold higher MGMT expression compared with the primary tumor, consistent with enrichment of temozolomide-resistant subclones following chemotherapy (Fig. 1D). Using previously described molecular subtypes for glioblastoma (17), transcriptome data categorized the primary tumor as most representative of the classic subtype, whereas the metastases most represented the mesenchymal subtype (Supplementary Fig. S3).

Given the substantial mutational burden, specific driver mutations were difficult to identify. However, we identified 62 DNA repair–related genes containing nonsilent somatic mutations within the founder clone (Supplementary Table S5). These included ATM (R457*), MSH5 (P581H), and TP53 (R175C), which are classified as pathogenic in the ClinVar database (18), as well as 34 other gene variants with supporting functional implications from ClinVar, Catalogue of Somatic Mutations in Cancer (COSMIC), SIFT, or PolyPhen (19–21). There were two additional POLE missense somatic mutations in the founder clone: R793C, predicted to be damaging and previously seen in a stomach carcinoma (22), and V1002A, which is of unknown significance. Mutational signature decomposition links the founder clone to a poorly characterized signature associated with deficient DNA-repair processes, but not specifically with POLE-induced mutations (Supplementary Figs. S1 and S3).

Because glioblastomas are heterogeneous (23), we examined the subclonal architecture of the primary brain tumor and the two spinal metastases to determine how it changed over the course of two different treatments. Clonal analysis revealed four distinct mutation clusters in the treatment-naïve brain tumor, of which one was the founder clone (Fig. 2A–C; Supplementary Table S6). Comparatively, in the first spinal metastasis, which emerged following concurrent radiation and temozolomide therapy, cluster 4 was absent, and cluster 3 became the dominant clone. We also identified new subclonal populations (clusters 5–7) that became apparent in this tumor (Fig. 2A–C). The second metastasis (after 3 weeks of treatment with pembrolizumab) lacked clusters 2 and 3, suggesting that it was descended from the primary brain lesion and not from the first metastasis. This tumor also harbored two unique subclones (clusters 8 and 9). By examining each subclone's mutation spectrum individually, we found that all subclones have similar mutational signatures that are associated with DNA-repair defects (Supplementary Fig. S3) but have not been otherwise well characterized (24). Cluster 9, an emergent subclone in the second metastasis, also contains a signature suggesting that around 30% of its mutations may have been temozolomide induced.

As response to checkpoint blockade immunotherapy is likely influenced by the presence of neoantigens (6, 8, 25), we applied a cancer immunogenomics approach to define the expressed neoantigen landscape across all tumor samples (3, 26; Fig. 3A; Supplementary Table S7). By incorporating expression data, the number of candidate neoantigens was reduced substantially. As such, we identified between 2,040 and 3,254 high-quality neoantigenic mutations per sample, with the first metastasis carrying the highest burden (Fig. 3A; Supplementary Table S7). Interestingly, the founder clone contained 1,245 expressed candidate neoantigens representing a shared population of potentially immunodominant epitopes (Fig. 3A).

Following progression while on standard adjuvant chemotherapy, the patient was transitioned to pembrolizumab treatment because responses to checkpoint blockade immunotherapies have been correlated to tumors with hypermutated genomes (8). Of note, prior to pembrolizumab administration, the right frontal horn lesion included in the prior radiation plan had also progressed (Fig. 3B, blue arrow). There was increased T2/FLAIR signal, suggestive of edema and inflammation, noted in this location (Fig. 3B, white arrow), and persistent enhancement in the left temporal lobe near the initial resection bed, indicating residual disease (Fig. 3B, red arrow). Reirradiation to this site and the entire neuraxis was considered, but the patient elected to first pursue systemic treatment with pembrolizumab. After 3 weeks of pembrolizumab, there was decreased enhancement of the right frontal lesion but a perilesional increase in the T2/FLAIR signal that included the frontal horn ependyma (Fig. 3B, middle). After 13 weeks of pembrolizumab, the right frontal lesion decreased further in size, and no enhancement was observed adjacent to the left-sided resection bed. There was also a corresponding decrease in T2/FLAIR hyperintensity in the right frontal horn compared with previous studies (Fig. 3B, right).

We next tested the hypothesis that pembrolizumab treatment would lead to increased cytolytic tumor-infiltrating lymphocytes in CNS tumors, using two approaches. First, we assessed the expression of key immune genes within the three tumor samples by RNA-seq. Expression of CD3, CD8, granzyme A (GZMA), and perforin (PRF1) was significantly elevated in CNS metastatic tissue following pembrolizumab treatment, and levels of PD-1 (PDCD1), PD-L1 (CD274), and IFNγ (IFNG) were also increased (Fig. 3C; Supplementary Fig. S4). Second, immunohistochemistry was performed on the resected metastases taken before and 3 weeks after pembrolizumab treatment. We identified prominent areas of brisk CD3⁺, CD4⁺, and CD8⁺ T-cell infiltrate in the post-pembrolizumab lesion (Fig. 3D). Together, these data show that hypermutated tumors harbored abundant high affinity, expressed neoantigens, and that checkpoint blockade in this setting was associated with objective radiographic responses and brisk cytolytic lymphocyte infiltration into CNS tumors.

Hypermutated tumors with POLE germline mutations have been described in patients with colorectal cancer (15), and somatic alterations in POLE were recently reported in subsets of patients with glioblastomas and endometrial cancers (14, 27). To our knowledge, our report represents a description of only the second reported adult patient with glioblastoma in the setting of germline POLE deficiency (28). POLE encodes the DNA polymerase ϵ catalytic subunit, which synthesizes and proofreads leading strand DNA. In this report, the germline POLE alteration was likely permissive in creating an unstable genomic context that facilitated additional abundant somatic DNA-repair machinery alterations. However, the level of hypermutation makes it difficult to clearly assign causality to mutated POLE alone, based on the observed mutational signatures.

The relationship between neoantigen burden and checkpoint blockade therapy response has been described in other tumor types but has not been reported extensively in the setting of glioblastoma. Patients with colorectal cancer harboring high mutational burdens due to mismatch-repair deficiency demonstrated improved clinical responses compared with patients with lower mutational loads (8), a correlation also described in other solid tumors (6, 29). Also, a recent study reported clinical benefit from nivolumab treatment in 2 pediatric patients with hypermutated gliomas (30). Herein, the sequential resection of glioblastoma metastases before and after pembrolizumab treatment provided an opportunity to observe posttreatment lymphocyte infiltration in a CNS tumor, which has not been observed previously following checkpoint blockade. Although it is important to broaden this observation, these data suggest that CNS malignancies may be appropriate targets for checkpoint blockade treatment in defined scenarios and that lymphocytes may infiltrate across the blood–brain/blood–cerebrospinal fluid barriers.

These results have several clinical implications. First, diagnostic efforts should be made to identify tumors with highly elevated mutational loads. Extensive genomic characterization of this patient's tumor was critical in establishing the mutational burden, underscoring the importance of assaying a sufficient number of genes in clinical workflows to identify the hypermutated state. Second, although hypermutation in newly diagnosed glioblastomas is rare, it is more common in recurrent glioblastomas after temozolomide treatment, in which the incidence may approach 20% to 30% (10, 12, 31). Thus, recurrent tumors could be sampled and tested for a hypermutated genotype, which may confer increased sensitivity to checkpoint blockade treatment. Prospective multi-institutional clinical trials are needed to validate this possibility and determine whether temozolomide-associated hypermutation will be as favorable a biomarker as hypermutation due to heritable or somatic mismatch-repair deficiency alone. Third, although this patient's hypermutated condition was not discovered until after chemoradiation treatment, early identification in other patients with glioblastoma prior to concurrent temozolomide and radiation treatment could lead to trials that assess the replacement of alkylating agents with checkpoint blockade for this select subset of patients. This patient's disease dissemination was likely due to both the aggressive behavior exhibited by a tumor with PNET components and the temozolomide resistance that developed once DNA mismatch-repair mutations were enriched (12, 13). Finally, clonal analysis of the expressed neoantigen landscape revealed that identification of the founder clone (7), which contains mutations present in all cells, may contribute to personalized cancer vaccine strategies that can broadly target heterogeneous cancers or those with spatially distinct sites of metastatic disease.

The patient provided written informed consent for tissue banking and genomic characterization to protocol #201111001, approved by the Washington University School of Medicine Institutional Review Board. DNA exome and RNA sequencing was performed, resulting in over 86× coverage of the normal and over 200× coverage of each tumor. Sequence data were aligned to reference sequence build GRCh37-lite-build37, then variant detection was performed using an ensemble approach followed by statistical and heuristic filtering. Copy-number aberrations were detected using VarScan. RNA-seq data were aligned to reference sequence build GRCh37-lite-build37 with Tophat, and expression levels were quantified using Cufflinks. Gene fusions were inferred jointly from RNA and DNA using INTEGRATE. The SciClone algorithm (32) was used for clonal inference, followed by reconstruction of tumor phylogeny using clonEvol (33). Mutational signatures were inferred using the deconstructSigs package for R.

Neoantigen prediction was performed using the pVacSeq pipeline (26), incorporating 5 MHC-binding prediction algorithms and requiring a minimum IC₅₀ score of 500 nmol/L.

Additional methods and detailed version and parameter information are available in the Supplementary Appendix.

I.G. Dorward reports receiving a commercial research grant from Stryker and is a consultant/advisory board member for Medtronic, Stryker, DePuy, and Advance Medical. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.M. Johanns, C.A. Miller, R. Uppaluri, G. Ansstas, E.R. Mardis, G.P. Dunn

Development of methodology: G. Ansstas, G.P. Dunn

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.M. Johanns, I.G. Dorward, C. Tsien, E. Chang, A. Perry, C. Ferguson, R.E. Schmidt, S. Dahiya, G. Ansstas, E.R. Mardis, G.P. Dunn

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.M. Johanns, C.A. Miller, R. Uppaluri, R.E. Schmidt, S. Dahiya, G. Ansstas, E.R. Mardis, G.P. Dunn

Writing, review, and/or revision of the manuscript: T.M. Johanns, C.A. Miller, C. Tsien, A. Perry, R. Uppaluri, C. Ferguson, S. Dahiya, G. Ansstas, E.R. Mardis, G.P. Dunn

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Chang, R.E. Schmidt, S. Dahiya, E.R. Mardis

Study supervision: G. Ansstas, E.R. Mardis, G.P. Dunn

This work was supported by NIH grant K08NS092912 (G.P. Dunn), the Physician-Scientist Training Program at Washington University School of Medicine (T.M. Johanns), and an endowment to The Elizabeth H. and James S. McDonnell III Genome Institute from the McDonnell family.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.