Modeling the Landscape of Histone-Mutant Pediatric High-Grade Gliomas: A Study in Partner Alterations

Pediatric high-grade gliomas (pHGG) are a group of lethal brain tumors that occur in children and adolescents, for which the current standard of treatment is largely palliative, demonstrating a dire need for effective, targeted therapies and new models in which to test them (1). Among the diffuse pHGGs, the majority of tumors harbor mutations in histone H3 genes: a lysine-to-methionine substitution at position 27 (K27M) in canonical histone H3.1 or its variant H3.3, or a glycine-to-arginine/valine substitution at glycine 34 (G34R//V) in H3.3 (2). These mutations exhibit unique patterns of distribution that correlate with patient ages, location of tumorigenesis along the midline or in the cerebral hemispheres, respectively, as well as specific associate partner mutations, revealing the complexity within this family of tumors (1).

Within the realm of diffuse pHGGs, patient-derived xenografts (PDX) remain the most frequently used in vivo models, largely due to the ability to reproduce the heterogeneity of the tumors from which they are derived and produce many replicates in a relatively short period of time (3). As a result, PDX models have been used extensively to investigate potential therapeutic options. However, the benefit of PDX models comes with the cost of a compromised immune system in these mice, greatly limiting the ability to investigate immunologic responses and recapitulate a more accurate tumor microenvironment. Therefore, modeling efforts have also largely been focused on the production of genetically engineered mouse models, in which specific genetic aberrations are introduced into immunocompetent mice with the goal of recapitulating tumor initiation and formation as closely as possible to patient tumors (3). In the realm of pHGGs, groups such as Oren Becher's have established models of H3.3K27M diffuse midline glioma (DMG) utilizing the RCAS/tva system, which leverages avian-derived viruses to integrate mutant DNA into specific cellular populations expressing the respective avian receptor (4). The RCAS/tva model was able to introduce multiple, specific alterations and to derive various cell lines from these in vivo models; however, the induction of these mutations is restricted temporally to postnatal time points. Alternatively, Larson and colleagues turned to transgenic knockin models, utilizing tamoxifen administration to express an H3.3K27M transgene and the two most commonly occurring alterations in p53 loss of function and constitutively active PDGFRα (5). Likewise, the sleeping beauty transposon system has been used to integrate transposable H3.3G34R DNA into the host genome, along with short hairpin RNAs against Trp53 and ATRX, two mutations co-occurring with H3.3G34R, to generate hemispheric gliomas (6). Together, these models were able to generate tumors harboring their respective histone H3 mutations and the two most common alterations as seen clinically. However, these models alone do not represent the many additional unique combinations of mutations in these histone-mutant gliomas or begin revealing their unique contributions to tumorigenesis. Therefore, unsatisfied with the current depth of available murine models of pHGG, McNicholas and colleagues took on the ambitious task of generating a compendium of murine models possessing as many relevant mutational combinations as possible to recapitulate the many subtypes of diffuse pHGGs and to investigate the roles of these partner mutations to tumorigenesis and whether these tumors were vulnerable to select therapeutics (7).

To achieve this, they refined their previously utilized in utero electroporation (IUE) technique to deliver both piggyBac transposons and CRISPR vectors harboring various combinations of mutations into specific regions of the developing mouse brain at E12.5 of gestation. An H3.3G34R construct was electroporated into the dorsal or ventral pallium of developing mice, targeting the cortex or ganglionic eminence (GE), respectively. Additionally, H3.3G34R mice were generated to harbor loss-of-function mutations in both Trp53 and ATRX (referred to as GPA mice), allowing the group to investigate the role of PDGFRα in this context via wild-type PDGFRα amplification (GPAP) or the expression of one of two PDGFRα mutations—PDGFRα^C235Y and PDGFRα^D842V (GPAC and GPAD, respectively)—leading to constitutive activation of the pathway. Interestingly, GPA mice lacking any PDGFRα alterations were unable to drive tumorigenesis, whereas all three models harboring PDGFRα alterations were able to develop penetrant tumors. GPAP tumors demonstrated the least aggressive growth and longest latency, with a median survival of 453 days, whereas PDGFRα-mutant mice significantly reduced this window to 296 and 67 days for GPAC and GPAD tumors, respectively, demonstrating PDGFRα signaling to be a key event in tumor progression. These conclusions, however, must be considered with respect to the limitations of IUE, particularly the inability to control the exact copy number of transgenes introduced and their exact location of genomic integration, which may contribute to the extended latency in some lines. The group was able to subsequently generate gliomasphere cell lines from symptomatic GPAP/GPAC/GPAD mice and investigate the sensitivity of these lines to a small set of small-molecule inhibitors targeting various pathways. Infigratinib, an inhibitor targeting FGFR1, was able to kill GPAC cells at nanomolar concentrations, whereas these cells remained largely insensitive to the PDGFRα inhibitor avapritinib. Although a surprising observation based on the suspected contribution of PDGFRα described above, these data highlight the value of investigating a broad landscape of pathways for potential vulnerabilities in these tumors. Thus, McNicholas's group generated the first GE-targeted H3.3G34R model driven by the PDGFRα^C235Y mutation, recently observed to be present in H3.3G34R patient tumors, and demonstrated a potential alternative pathway for further treatments of H3.3G34R tumors.

Scaling up this experimental strategy, McNicholas and colleagues proceeded to tackle the different subtypes of DMGs possessing H3K27M mutations in combination with additional partners. Targeting the lower rhombic lip at E12.5, IUE was again utilized with the piggyBac transposon system as before to deliver a series of mutational combinations seen in patients. H3.1 and H3.3K27M constructs were delivered into embryos, along with various partner alterations in p53, PDGFRα, PIK3CA, PPM1D, ATRX, NF1, and FGFR1. Four models—KPP (H3.3K27M, P53^LOF, and PDGFRα^WT—overexpression), KPPMPIK (H3.3K27M, PPM1D^ΔC, and PIK3CA^E545K), H3.1ACVPIK (H3.1K27M, ACVR1^G328V, and PIK3CA^E545K), and KNF (H3.3K27M, NF1^LOF and FGFR1^N457K), representing some of the most commonly mutated genes in histone-mutant DMGs—were all able to induce brainstem penetrant tumors that show diffuse infiltration, and most showed relatively short median survival (KNF: 28 days, KPPMPIK: 42 days, H3.1ACVPIK: 65 days). Interestingly, KPP mice demonstrate a novel approach to this combination, relying on the amplification of wild-type PDGFRα to help drive tumorigenesis rather than a constitutively active mutant but at the expense of prolonged survival with a median of 230 days. This longer latency could, however, also be a result of the IUE technique for transposon-based integration in which the exact copy number and site of integration of transgenes introduced are not well controlled. Therefore, the overall survival and subsequent experiments must be considered with this limitation in mind, particularly with respect to the number of replicates for each mouse line, as individual mice within the same line may differ greatly in their tumorigenic capacity based on variable levels of expression and potential interference of endogenous genes. KNF mice represent a small subset of H3K27M DMGs that have not been well characterized. Here, tumors produced by the KNF combination exclusively disseminate into the thoracic spine while showing little to no invasion into highly myelinated cranial nerves compared with the other three models, emphasizing the importance of specific partner alterations driving tumorigenesis and the unique phenotypes they present.

The utility of McNicholas's study is emphasized in the ability to culture these tumors and subsequently engraft them syngeneically into immunocompetent C57BL/6J mice, enabling future studies to investigate the interactions of tumor and immune cells in these mice. Additionally, proliferating these tumors in vitro opens the investigation of potential therapeutic vulnerabilities unique to each mutational combination. Highlighted in this study, both mouse- and patient-derived lines possessing amplified PDGFRα or PPM1D^ΔC/PIK3CA^E545K alterations were most sensitive to a combinatorial approach of trametinib and alpelisib administration, targeting the MEK and PIK3CA pathways, respectively. This combination was also able to reduce tumor dissemination in KPP allograft models, combating the highly invasive nature of these tumors. The work conducted on both the H3G34R and H3K27M cell models in this study highlights not only the complexity of pHGG tumorigenesis based on their associate mutations but stands to broaden the scope of therapeutic development beyond the histone mutants themselves.

Altogether, McNicholas and colleagues were able to generate a compendium of in vivo models representing various subtypes of histone-mutant pHGGs, emphasizing the importance of partner alterations in tumor progression, and have provided the pHGG community with a basis for the next generation of tumor models.