Characterizing temporal genomic heterogeneity in pediatric high-grade gliomas

Genetic and epigenetic molecular profiling techniques have revolutionized our understanding of the etiology and biology of pediatric high-grade gliomas (pHGGs) (reviewed in [20]). Unfortunately, this has not yet led to an improvement in outcome for children with this disease [40] despite the use of agents that target pathways identified through these biological advances. Novel agents for the treatment of pHGGs are first tested in the relapse setting, and target genomic alterations typically present in therapy-naïve diagnostic tumor samples or models. However, there is limited data on the relevance of genomic aberrations at diagnosis on disease progression after multimodal therapy, making the effectiveness of this approach questionable. An improved understanding of temporal and therapy-driven evolution of recurrent pHGGs is therefore needed, especially in the context of hemispheric HGGs that show increased genetic heterogeneity [5, 12, 13, 19, 37, 50, 51].

Clonal evolution is a dynamic process that has been reported in many cancer types [3, 28, 39, 48], even without exposure to therapy [11]. Morrissy et al., have recently demonstrated poor overlap in genetic events between primary and post-treatment medulloblastoma both in murine models and human samples [28]. This included a marked divergence in actionable genes between diagnosis and recurrence, despite conservation of molecular subgroup affiliation [28, 36, 47]. Whole exome sequencing (WES) of 23 initial and recurrent gliomas in adults by Johnson et al., revealed variable genetic relatedness across pairs; in 10 cases, most mutations from diagnosis were not conserved in the recurrent sample, including the BRAF V600E hotspot mutation [19]. In adult glioblastoma multiforme (GBM), a longitudinal study of the genetic landscape of 114 untreated and recurrent paired tumors revealed a switch in expression-based subtypes in 63% of cases. Enrichment of a hypermutated phenotype in recurrent disease exposed to temozolomide (TMZ) was also identified, suggesting the occurrence of therapy-induced mutagenesis [45]. Moreover, an analysis of tumor phylogeny revealed that dominant clones at recurrence were infrequently direct descendants of dominant clones from diagnosis [45]. We have previously shown that disease-defining somatic mutations in oncohistones [K27M in Histone 3 (H3) variants (H3F3A, HIST1H3B)] are spatially stable in diffuse intrinsic pontine glioma (DIPG), and co-occur with highly conserved partners throughout geographically distinct tumor sites [18, 30]. However, limited data on disease recurrence are available for supratentorial pHGGs. This is of major therapeutic interest as hemispheric pHGGs show more genetic variability at diagnosis than midline tumors, the vast majority of which are defined by H3K27M mutations (> 90%) [14, 51]. In the current study, we characterize the temporal genomic heterogeneity in pHGGs by assessing the mutational profile and methylome of paired primary and recurrent tumors with emphasis on supratentorial pHGGs.

In this work, we performed whole exome sequencing on 16 primary and recurrent pHGG pairs including two pHGGs from patients with germline NF1 mutations, and provide insight into the temporal genomic evolution of these tumors. A direct comparison of the mutational landscape of paired samples reveals that oncogenic driver mutations are typically conserved. The identification of these mutations in both the primary and recurrent tumors suggests that these mutations are early initiating events in tumorigenesis, are stable, and unaffected by treatment. This is in contrast to adult GBM where cancer driver mutations can be subclonal in the primary and recurrent tumors [19].

In our dataset, 10 of 16 patients were treated with TMZ, and despite a trend towards an increase in the number of mutations at recurrence, there was no statistically significant increase in mutational burden. This is in contrast to adult GBM, where an increase in mutational burden is observed with TMZ treatment [19]. Although our sample size is small, our findings may reflect a different biological process in response to TMZ in children compared to adults and warrants further evaluation. We observed one H3/IDH1 wildtype primary tumor (HGG11) with an increased number of somatic mutations compared to other primary tumors, and was identified to harbor a germline MLH1 splice missense mutation. Immunohistochemical analysis did not show loss of the MMR proteins, however, we hypothesize that the missense splice mutation likely resulted in the translation of a dysfunctional MLH1 protein product to cause mismatch repair deficiency (MMRD) and hypermutation. After treatment with radiation and TMZ, this tumor acquired an increased number of somatic mutations compared to the primary tumor, suggesting that treatment further exacerbated the hypermutated phenotype. Several controversial and contradictory studies have variably reported the presence of microsatellite instability which results in mismatch repair deficiency in pediatric HGG and adults [10, 44], highlighting the need for further studies. Future genetic testing for MMRD in pediatric HGG patients could steer treatment towards immunotherapy, as immune checkpoint blockade has shown clinical benefits in MMRD colorectal cancers as well as children with high-grade glioma [4, 23].

Similar to findings in adult IDH1-mutant gliomas [19], we identify heterogeneous ATRX alterations among IDH1 mutant pHGG tumor pairs. While IDH1 mutant tumors are more common in adult GBM and occur in up to 98% of secondary GBMs, they make up less than 10% of all pediatric HGGs [2, 52]. In contrast to IDH1-mutant gliomas, ATRX mutations associated with H3G34V, ZMYND11, EP300, or BRAF V600E were stable across the disease course in our study. Additionally, the BRAF V600E mutation was present in both primary and relapse samples in two children in our study which is in contrast to adult studies where it was identified either at diagnosis or at recurrence [19].

H3/IDH1 wildtype pHGGs have previously been shown to be a diverse group of tumors with mutations in many cancer pathways [35, 37, 51], but have not been directly linked to any particular epigenetic driver as is the case with H3 and IDH1 mutant tumors. Our data reflect the heterogeneity of tumors in the H3/IDH1 wildtype group while also identifying two novel pHGG epigenetic cancer drivers (ZMYND11 and EP300) in this group. ZMYND11 has recently been described as an epigenetic regulator that specifically interacts with H3K36me3 to regulate transcription. Wen et al. have reported that H3 G34R/V mutations impair binding of ZMYND11 to an H3.3K36me3 peptide, suggesting that H3.3 G34R/V and ZMYND11 mutations alter H3K36me3 levels in similar fashions [49]. To the best of our knowledge, ZMYND11 mutations have not been previously described in pHGGs. The tumor harboring this mutation (HGG9) was located in the right parietal lobe and carried partner mutations in ATRX and TP53, further supporting its similarity to hemispheric H3.3 G34R/V mutated tumors. In addition, inactivating mutations identified in the HAT gene EP300 have been implicated in a wide array of cancer types including diffuse large B cell lymphoma [34], head and neck, esophageal, colorectal, medulloblastoma and non-small cell lung carcinoma [7, 15]. We also report a specific EP300 hotspot D1399N mutation (HGG8) which has not been previously identified in HGGs. Structural analysis of EP300 has shown that the D1399 residue has effects on the conformation of the HAT domain, specifically the L1 loop [25]. This is also an inactivating mutation which abolishes autoacetylation required for HAT activity, thus affecting post-translational modification of K27 on H3 variants [8]. Interestingly, EP300 D1399Y mutations alter its interaction with transcription factor AP-2alpha indirectly leading to the transactivation of Myc [16]. Moreover, the tumor harboring the EP300 mutation was located in the thalamus which is a neuroanatomical structure in the brain midline where the majority of HGGs harbor H3K27M mutations. This novel epigenetic mutation may reproduce some of the effects of K27M in a wildtype H3K27 tumor. In our study, the tumor with the EP300 D1399N mutation had increased Myc expression (data not shown), suggesting that this particular EP300 mutation may also play a role in Myc-related oncogenesis similar to K27M mutagenesis. Although interesting, these findings need further testing and functional validation in relevant disease models. The two HGGs from patients with germline NF1 did not show a high mutational burden at diagnosis or at recurrence, and no clear associated driver mutation. Interestingly, a tendency towards increased copy number alteration was observed in both pairs at recurrence. These findings also need further validation on a larger sample set.

Somatic mutations in RTKs are common in adult GBM [5, 6] and are generally found at low frequencies in pHGGs [41]. Similar to our previous report [41], the H3/IDH1 wildtype group in this study seemed enriched with RTK mutations (5/7, 71%). One striking finding in this molecular group was the discovery of EGFR missense mutations in the primary occurrence of HGG10 (T790M and E709A), which were lost in the recurrence. A shared EGFR R222C missense mutation was present in both the primary and recurrent tumors, indicating that alteration of the RTK pathway is nonetheless conserved in the recurrent tumor. The EGFR T790M mutation has been implicated in acquired resistance to most EGFR tyrosine kinase inhibitors [21, 27]. This may, in part, explain tumor progression in this patient despite treatment with lapatinib (Novartis, East Hanover, NJ), and highlights the importance of identifying resistance-promoting mutations in the clinical setting. We also identified three tumors with targetable RTK lesions (PDGFRA, ERBB2, ERBB4) that were exclusive to the recurrent tumor (HGG5, HG8, HGG11), indicating that genomic data from tumor tissue at recurrence may provide better guidance for therapeutic choices. Conversely, one case harbored a low level subclonal PIK3CA mutation that was discovered by a clinical genomics panel in the primary tumor, but was not identified by WES in different primary tumor blocks from the same case, nor in multiple samplings of the recurrent tumor. Excluding the subclonal nature of this mutation, and confirming its maintenance at recurrence are important therapeutic considerations before embarking on targeted treatment, especially with single agents such as rapamycin used in this patient.

In conclusion, this study further highlights the molecular distinction between pediatric and adult HGGs, especially in therapy-induced tumor evolution. We show that genes with driver mutations (H3, TP53, PPMID, ZMYND11, EP300) as well as some targetable mutations (e.g. IDH1, BRAF V600E) are conserved. Importantly, we demonstrate that some actionable mutations are unstable (PI3K, EGFR), indicating that re-biopsy is warranted in order to optimize personalized therapy. The presence of subclonal targetable alterations concurrently with driver mutations supports the use of combination therapy approaches to address disease biology and evolution with the aim of improving patient outcomes.