Epigenetic reprogramming and chromatin accessibility in pediatric diffuse intrinsic pontine gliomas: a neural developmental disease

Abstract

Diffuse intrinsic pontine glioma (DIPG) is a rare but deadly pediatric brainstem tumor. To date, there is no effective therapy for DIPG. Transcriptomic analyses have revealed DIPGs have a distinct profile from other pediatric high-grade gliomas occurring in the cerebral hemispheres. These unique genomic characteristics coupled with the younger median age group suggest that DIPG has a developmental origin. The most frequent mutation in DIPG is a lysine to methionine (K27M) mutation that occurs on H3F3A and HIST1H3B/C, genes encoding histone variants. The K27M mutation disrupts methylation by polycomb repressive complex 2 on histone H3 at lysine 27, leading to global hypomethylation. Histone 3 lysine 27 trimethylation is an important developmental regulator controlling gene expression. This review discusses the developmental and epigenetic mechanisms driving disease progression in DIPG, as well as the profound therapeutic implications of epigenetic programming.

Key Points.

1. DIPG is an incurable pediatric brainstem tumor associated with a neural developmental origin. Thus, understanding the normal relationship of DIPG mutations with neural development is critical.

2. Since histone the H3K27M mutation is a hallmark of DIPGs, understanding the role of this oncogenic driver in tumorigenesis is essential.

3. Determining ways to rebalance the epigenome through the inhibition of epigenetic remodelers will expose unique vulnerabilities of DIPG tumors as a consequence of H3K27M.

Diffuse intrinsic pontine glioma (DIPG) is a highly aggressive pediatric brain cancer that is currently incurable. DIPG occurs in children of median age of 6–7 years and bears a bleak prognosis: a median survival of 11 months with only 10% of DIPG patients surviving 2 years from onset.¹ A DIPG diagnosis results from the culmination of acute neuropathological symptoms including cranial nerve palsies, long-tract and cerebellar signs, in addition to the characteristic radiographic abnormality in the pons.¹ The long-standing standard treatment for DIPG is focal radiation.² Radiation provides mitigation of symptoms and slows down tumor growth in the majority of DIPG patients.^3,4 Unfortunately, the efficacy of radiotherapy is limited and provides only transient symptomatic relief and meager benefits to DIPG patient survival.^4–7 Numerous clinical trials have evaluated other treatments (alone and in combination with radiation) such as cytotoxic chemotherapy, small-molecule inhibitors, and immunotherapies. However, none of these therapies have demonstrated a significant improvement in patient survival without imposing substantial adverse side effects.⁷

The location of DIPG tumors poses a challenge for diagnosis and treatment. DIPG tumors infiltrate the pons, a region of the brainstem that regulates basic functions like breathing, blood pressure, and heart rate. The crucial functions of the pons and the inability to differentiate tumor cells from normal tissue in the operating room make surgical resection impossible. Even a biopsy of the tumor was felt to be unnecessary given that the results would not change clinical management. Contemporary practice still relies on the characteristic MRI abnormality of these tumors to confirm a DIPG diagnosis: a T2 hyperintense signal abnormality occupying at least 50% of the pons.⁸ This general practice will likely continue until molecular and pathological evaluation of DIPG tumors is able to meaningfully impact the standard of care. Nevertheless, biopsy remains a significant aspect of DIPG management, as the scarcity of DIPG samples poses challenges to the characterization and development of novel therapeutic agents for this disease. Progress has been made recently, with numerous studies reporting on the safety and low morbidity of computer-aided stereotactic biopsies since their development in 1978.^9–12 There is also evidence suggesting that liquid biopsies, utilizing cerebral spinal fluid or plasma, represent a promising future strategy to diagnose and monitor DIPG via the distinct markers present on the circulating tumor DNA.¹³ While the dire need to improve DIPG treatment remains, the increase in biopsy has facilitated significant advances in determining the hallmark genetic characteristics of DIPG—namely, the 2012 discovery of H3K27M mutations in approximately 85% of DIPG tumors.^14,15 Biopsies are now being increasingly performed at academic medical centers in order to designate patients into genetically stratified clinical trials, with the hope that such targeted treatments will provide improvements in DIPG treatment.¹⁶

Pathological evaluation of DIPG describes a diffuse tumor of the pons, often infiltrating the medulla and midbrain, with otherwise typical characteristics of histopathological heterogeneous glial neoplasms.¹⁷ Historically, high-grade glioma was defined by the World Health Organization (WHO) based on histopathological characteristics such as the presence of microvascular proliferation or necrosis. However, the most recent WHO classification of central nervous system tumors, published in 2016, began incorporating molecular testing as part of its diagnostic criteria. This led the WHO to define a new neoplastic entity establishing midline gliomas (eg, brainstem, thalamic, spinal cord) harboring H3K27M mutations and diffuse growth as “diffuse midline gliomas, H3K27M-mutant.” ¹⁸ This term includes 80% of DIPGs harboring H3K27M mutations.^14,15,19 Herein, we refer to mutations of lysine 27 to methionine as H3K27M, due to the nomenclature used in the original reports. However, based on standard mutation nomenclature, it is lysine 28 that is mutated to methionine.²¹

Normal Postnatal Development of Human and Murine Pons

Our understanding of DIPG may be enhanced by studying the postnatal development of the human pons. A morphometric analysis of the human pons 0–18 years of age found that there is a rapid, six-fold increase in the volume of the pons from birth to age 5, followed by slower but continued growth throughout childhood.²² MRI and histological analysis of myelin basic protein indicate that this expansion of the pons is likely due to an increase of myelination in the basis pontis.²² Ki67 staining also revealed that the number of proliferative cells began to decline rapidly from 0–7 months followed by a further decline by age 3, and only a small population persisted throughout childhood.²² Two populations of vimentin/nestin-expressing cells were also identified—a dorsal group near the ventricular surface that persists throughout childhood, and a parenchymal population that declines by 7 months.²² Monje et al utilized 24 normal postmortem brainstem samples and analyzed neural precursor cells in the midbrain, pons, and medulla. They identified a nestin-positive population in the ventral pons that peaks first during infancy, and then peaks for a second time at age 6.²³ The results from these studies indicate that the human pons is undergoing continuous growth during childhood. It is noteworthy that two independent studies, one using pons volume as a metric and the other using density of nestin-positive cells as a metric, suggest that there is a peak growth at approximately 5 or 6 years of age, which correlates with age of incidence of DIPG. A murine study also observed a five-fold increase in the basis pontis postnatally, while the tegmentum grew four-fold.²⁴ However, this rapid growth is not due to myelination as in the human pons, but is reportedly due to proliferation.²⁴ The pons is the most proliferative brainstem region in comparison to the midbrain or medulla.²⁴ A parenchymal progenitor population of sex-determining region Y–box 2 (Sox2+) oligodendrocyte transcription factor 2 (Olig2+) was identified as the main proliferative progenitor population, and Lindquist et al also found that the majority of adult pons oligodendrocytes were found to be derived from postnatal Sox2+ progenitors.²⁴ Overall, the murine pons growth postnatally resembles human pons growth. Lineage tracing studies using markers of progenitor cell populations could help increase our understanding of the cell of origin of DIPG.

Cell of Origin in DIPG

If DIPG originates from a postnatal cell, then the cell of origin in DIPG may be the proliferative Ki67+/Olig2+ proliferative population identified by Tate et al, in the postnatal pons, or the nestin+ population identified by Monje and colleagues.^22,23 The hypothesis that DIPG arises from a postnatal precursor cell is supported by the findings of Funato et al, demonstrating that H3.3 K27M mutation, together with platelet derived growth factor receptor A (PDGFRA) activation and p53 loss, was sufficient to induce neoplastic transformation in human embryonic stem cells (ESCs) and that expression of H3.3 K27M led to a resetting of neural precursors to a more primitive stem cell state.²⁵

However, it is also possible that the cell of origin is an embryonic precursor cell. Pathania et al performed in utero electroporation to overexpress H3.3 K27M in neural progenitor cells (NPCs) in the lower rhombic lip of the developing hindbrain where some pontine nuclei are derived. They found that in combination with Trp53 loss it was sufficient to induce both focal and diffuse tumors, but expression of H3.3 K27M postnatally did not yield the same results.²⁶ Sun et al also demonstrated that the H3.3 K27M mutation was sufficient to immortalize human fetal hindbrain-derived NPCs, but not cortex-derived progenitor cells.²⁷ This finding suggested that only human fetal hindbrain-derived NPCs could become oncogenic with the introduction of the highly prevalent H3.3 K27M mutation.

In support of a developmental origin, whether postnatal or prenatal, single cell RNA sequencing of 6 H3 K27M mutated glioma patients identified that they have a developmental hierarchy, with some tumor cells having transcriptional profiles that were associated with either astrocytic differentiation, oligodendrocytic differentiation, and oligodendrocytic progenitor-like programs.²⁸This study also revealed that the largest fraction of stem-like cells in DIPGs were oligodendrocyte progenitor-like cells (OPCs), supporting the hypothesis of a precursor OPC of origin for DIPG.²⁸

Developmental Role of Genes Mutated in DIPG

The developmental regulator activin A receptor type 1 (ACVR1) is a frequently mutated gene in DIPG present in approximately 24% of patients.²⁹ This mutation is typically observed in younger DIPG patients.^30–33 ACVR1 has been shown to be necessary to establish proper patterning in late gastrulation during embryogenesis for normal craniofacial and cardiac development.^34–36 The receptor is part of the transforming growth factor beta (TGF-β) family. Normally, upon bone morphogenetic protein (BMP) ligand binding, type 2 BMP receptor kinase phosphorylates ACVR1, a type 1 BMP receptor. This initiates a downstream signaling cascade through phosphorylation of the Smad transcription factors.³⁷ Canonically, ACVR1 phosphorylates Smad1/5/8.³⁸ ACVR1 is also mutated in the congenital disorder fibrodysplasia ossificans progressive (FOP).³⁹ Interestingly, cells expressing mutated ACVR1 respond to activin ligands to which wild-type ACVR1 does not normally respond.^40,41 In models of FOP, activin-ACVR1 signaling was sufficient to induce heterotypic ossification.^40,41 DIPG cells harboring mutated ACVR1 have also been shown to exhibit an aberrant response to activin A.⁴² Activin receptor signaling regulates oligodendrocyte lineage cell responses during white matter development and after injury.⁴³

During development, BMPs are tightly regulated temporally and spatially by other signaling pathways.³⁷ Postnatally, BMPs promote astrogliogenesis while inhibiting both neurogenesis and oligodendrogliogenesis.³⁷ Although a lot is known about BMP signaling in the brain, the role of BMP signaling or ACVR1 has not been studied in the early postnatal pons. Identifying the implications of ACVR1 or other BMP interacting/inhibitory pathways on neural precursor or proliferative populations could provide new directions for development of therapies. In addition to activating ACVR1 mutations, there have been amplifications in inhibitor of DNA binding 2 protein (ID2), ID3, mutations in BMP3, BMPK, and others in approximately 19% of DIPG, underscoring the importance of the BMP pathway in DIPG.⁴⁴ Hoeman et al found that in vitro expression of ACVR1 R206H upregulated mesenchymal markers and activated signaling of signal transducer and activator of transcription 3.⁴⁵ Using a replication-competent avian sarcoma-leukosis virus long terminal repeat splice acceptor (RCAS) mouse model with PDGFA, the authors also found that ACVR1 R206H accelerated gliomagenesis and cooperates with H3.1 K27M to accelerate tumor growth and malignancy.³¹ In vivo inhibition of ACVR1 with the LDN212854 inhibitor also significantly increased median survival in the RCAS model described above.⁴⁵ In addition, treatment with ACVR1 inhibitors, LDN-139189 and LDN-214117, prolonged survival in orthotropic patient-derived xenograft models of ACVR1 R206H.⁴² Thus, ACVR1 may be an important therapeutic target for the treatment of DIPG.

TP53, the gene encoding the tumor suppressor p53, is mutated in 42% of DIPG tumors. Survival and apoptosis in the developing and adult nervous systems are regulated by p53.^30,46 In the brain, p53 is preferentially expressed in progenitor cells, and in vitro experiments using neurospheres from p53 null mice suggested that TP53 regulates cell cycle progression and apoptosis.^47,48 An increased proliferation rate in neural stem cells could make them susceptible to acquiring additional mutations that lead to malignancy.

Protein phosphatase, Mg2+/Mn2+-dependent 1D (PPMI1D), mutated in 23% of DIPG cases, is expressed in the fetal mouse brain, and germline mutations of PPM1D lead to an intellectual disability syndrome, suggesting a role in neurodevelopment.³² PPM1D dephosphorylates the DNA damage response proteins p53, checkpoint kinase 2, ataxia telangiectasia mutated, and gamma–H2A histone family member X to enable reentry into the cell cycle following cellular stress.⁴⁹ PPMI1D plays a critical role in central nervous system homeostasis by regulating G2/M cell cycle progression in adult NPCs through inhibition of p53 activity.⁵⁰ Thus, PPM1D and p53 together modulate the balance between self-renewal of NPCs while minimizing genotoxic stress. In DIPG, truncating mutations of PPM1D that stabilize the protein are mutually exclusive with TP53 mutations.^32,51

DIPGs are also characterized by amplifications of genes that regulate the G1-S cell cycle progression, particularly cyclin D family members CCND1, CCND2, CCND3, and cyclin dependent kinase inhibitors CDK4 and CDK6.^32,52,53 Cyclins and cyclin dependent kinases form a complex that phosphorylates retinoblastoma protein, allowing for cell cycle progression.⁵⁴ The regulation of the cell cycle progression and length is important for neurogenesis. In vitro, neural precursor cells with dominant negative forms of CDK2, CDK4, and CDK6 undergo cell cycle arrest.⁵⁵ The overexpression of cyclin D and CDK4 in neural progenitors of developing brain shortened G1 length and inhibited neurogenesis, demonstrating that a lengthening of G1 regulates the switch from proliferation to neurogenesis.⁵⁶ Uncontrolled proliferation is a characteristic of many cancers, and there are currently 4 clinical trials testing the efficacy of CDK4/6 inhibitors that were initially developed for other cancers in DIPG (NCT02255461, NCT03387020, NCT03355794, NCT02644460). The results of these clinical trials will shed light on the effectiveness of CDK4/6 inhibitors as a single therapeutic agent in DIPG.

Function of Histone Variants and Other Chromatin Regulators in Development

The most frequent mutation in DIPG is a K27M mutation that occurs on genes encoding histone H3 variants H3.3 and H3.1/2.^14,15,19 Histones are proteins that package DNA into compact units called nucleosomes, an octamer comprising 4 core histones (H2A, H2B, H3, and H4) and approximately 147 base pairs of DNA.⁵⁷ Nucleosome cores are connected by a strand of linker DNA that is stabilized by the recognition and binding of histone H1.⁵⁷ Histone H3 is posttranslationally modified (methylation, acetylation, ubiquitiniation, sumoylation, etc.) on the N-terminal tail to regulate transcription, DNA replication, and repair. H3.1 and H3.2 are canonical histones encoded by multiple genes which are “replication coupled,” meaning that they are synthesized and deposited on DNA during S-phase (Table 1).⁵⁷ The histone variant H3.3 is a non-canonical histone (Table 1). Non-canonical histones, in contrast, are usually encoded by one or two genes, and are expressed throughout the cell cycle.⁵⁷

Table 1.

Summary of human histone H3 variants

Histone	Gene	Deposition	Oncohistone in DIPG
H3.1	HIST1H3A	Canonical
	HIST1H3B		H3.1K27M
	HIST1H3C		H3.1K27M
	HIST1H3D
	HIST1H3E
	HIST1H3F
	HIST1H3G
	HIST1H3H
	HIST1H3I
	HIST1H3J
H3.2	HIST2H3A	Canonical
	HIST2H3B
	HIST2H3C		H3.2K27M
	HIST2H3D
H3.3	H3F3A	Non-canonical	H3.3K27M
	H3F3B
	H3F3C

Histone variants have distinct functions—for example, H3.3. is typically deposited on transcriptionally active genes and generally harbors posttranslational modifications associated with activation. Meanwhile H3.1 and H3.2 are associated with inactivation modifications.⁵⁸ An analysis of histone H3 gene expression in silico during human brain development found that genes encoding the H3.3 variant, including H3F3A, were present at all developmental stages; however, their expression did gradually decrease across developmental stages.⁵⁹ In contrast, the genes encoding the H3.1 variant, including HIST1H3B, are silenced at early developmental stages.⁵⁹ One important caveat that the authors noted was that H3.1 transcripts lack polyadenylation and thus are potentially underrepresented in poly(A)-derived cDNA libraries and RNA-seq datasets used in the study.⁵⁹

Polycomb repressive complex 2 (PRC2), is a Polycomb group (PcG) protein, which is composed of 3 core subunits—enhancer of zeste homolog 1 (EZH1) or EZH2, embryonic ectoderm development (EED), and suppressor of zeste 12 homolog (SUZ12)—that mediate the repression of developmental genes.⁶⁰ EZH1 and EZH2 contain a conserved catalytic Su(var)3-9/enhancer-of-zeste/trithorax (SET) domain found in many lysine methyl transferases, which catalyzes the di-, and trimethylation of histone 3 lysine 27 (H3K27me2/3).⁶¹ H3K27me3 is a repressive mark that is associated with inactive gene promoters.⁶⁰ The trithorax group of proteins, on the other hand, are antagonistic regulators of PcG proteins and include the SET1 family (SET1A, SET1B, MLL1, MLL2, MLL3, and MLL4 in humans), which catalyze methylation of histone 3 lysine 4.⁶² Trimethylation of histone 3 lysine 4 (H3K4me3) is associated with active chromatin.⁶² In DIPG cells, the genomic distribution of H3K4me3 methylation is largely unaltered compared with neural stem cells.⁶³

Many developmental regulatory genes are regulated by a “bivalent” state in which promoters have both H3K27me3 and H3K4me3 modifications that enable rapid expression upon removal of the repressive mark, a process that is mediated in part by Jmjd3, an H3K27me3 demethylase.⁶⁴ In the central nervous system, PcG proteins mediate the transition from neurogenesis to astrogliogenesis by repressing neurogenic genes and facilitating the expression of astrogenic genes such as glial fibrillary acidic protein. Polycomb group protein activity is also thought to be involved in the transition to oligodendrogenesis, where it may play a role in the repression of distal-less homeobox 2 (Dlx2), a gene that negatively regulates OPC formation.^60,65

Histone acetylation plays a critical role in gene expression by modification of the levels of chromatin compaction. Acetylated lysine residues neutralize the positive charge of the histones, impacting DNA–protein interactions and inducing chromatin de-condensation associated with transcriptional activation.⁶⁶ The dynamics of histone acetylation are regulated by histone acetyltransferases (HATs) and histone deacetylases (HDACs), which have key functions in active and inactive genes, respectively.⁶⁷ Histone lysine acetylation is recognized by bromodomain and extra-terminal (BET) proteins (BrD2, BrD3, and BrD4) which bind acetyl-lysines and recruit chromatin factors and transcriptional machinery.⁶⁸

Bromodomain proteins play an important role in the regulation of chromatin accessibility and rearrangement. Chromatin condensation during differentiation renders DNA inaccessible to transcription factors and requires the recruitment of chromatin remodeling factors to induce changes in gene expression.⁶⁹ BET proteins have 2 bromodomains, BrD1 and BrD2. Interestingly, a study in oligodendrocyte lineage progenitor cells found that selective inhibition of BrD1 enhanced differentiation, while inhibition of both BrDs inhibited differentiation, underlying the importance of histone acetylation and BET proteins in oligodendrocyte progenitor differentiation.⁷⁰ In NPCs, bromodomains are inhibited by JQ1, a small molecule that binds competitively to acetyl-lysine recognition motifs. BET bromodomain inhibition by JQ1 was found to promote neurogenesis and inhibit cell cycle progression and gliogenesis.^71,72

Histone Mutations in DIPG

H3K27M functions as a dominant negative mutation that dramatically reduces global levels of H3K27me3 in DIPG cells and tissue even though H3K27M protein makes up only a small percentage of total H3 (3–17% in DIPG cells as measured by mass spectrometry).^63,73,74 Several hypotheses have been suggested to explain the mechanism of global reduction of H3K27me3 by the H3K27M mutation. Lewis et al observed that synthetic H3K27M peptides interact with the EZH2 active site.⁷³ In support of this, co-immunoprecipitation by experiments of wild-type or H3.3 K27M containing mono-nucleosomes revealed enrichment of EZH2 and SUZ12 proteins in mutated nucleosomes.⁷⁴ Bender et al also demonstrated that the H3.3 K27M mutation decreased EZH2 methyltranferase enzymatic activity.⁷⁴ Additionally, a biochemical study demonstrated that EZH2 recognizes the histone H3 tail.⁷⁵ The authors also showed that inhibition of EZH2 catalytic activity by H3K27M could be alleviated through enhanced acetylation of surrounding chromatin regions.⁷⁵ Consistent with this, a crystal structure of the human PRC2 complex with the H3K27M peptide demonstrated that the mutant peptide binds to the active site of the SET domain. Furthermore, the binding of EED to H3K27me3, present as a result of an existing repressive mark, leads to the stabilization of the SET domain. This enhances the binding of PRC2 with the H3K27M peptide.⁷⁶ The resulting theories suggest that H3K27M binds to the EZH2 component of the PRC2 complex, to inactivate or sequester the complex, thus preventing it from carrying it out di- and trimethylation on wild-type nucleosomes (Fig. 1A).

Fig. 1.

Mechanisms by which H3K27M leads to defective PRC2 activity and global hypomethylation. (A) H3K27M mutant histone inhibits (shown by curved bracket) the catalytic activity of EZH2 to deposit the H3K27me3 mark (shown by red arrow). (B) PRC2 is sequestered at poised enhancers by binding to H3K27M, which restrains the available PRC2 pool from binding to its strong affinity sites. (C) H3K27M interferes (shown by curved bracket) with the spread of the H3K27me3 mark by PRC2 (shown by black arrows) leading to global hypomethylation and transcriptional de-repression.

However, Piunti et al reported that EZH2 and SUZ12 are largely excluded from chromatin containing H3.3 K27M. Instead mutant K27M proteins are found in regions that contain H3K27 acetylation (ac), which many groups had reported to be increased in H3.3 K27M mutant cells.^73,77,78 Subsequently, they found that PRC2 activity is required for DIPG cell maintenance and growth through potentially dependent or independent repression of p16, a PRC2 target gene.⁷⁷ Indeed, others have observed locus-specific enrichment of H3K27me3 in H3K27M-expressing cells, and hypothesize that H3K27M promotes tumorigenesis through repression of tumor suppressors.⁶³ In support of this, using a genetic mouse model, Cordero et al found that the H3.3 K27M mutation in the context of PDGF signaling promoted gliomagenesis through repression of the p16 tumor suppressor.⁷⁹ Additionally, a mouse model generated by neonatal induction of H3.3 K27M, PDGFRA, and Trp53 demonstrated that H3.3 K27M accelerated tumor development and that genes enriched in H3.3 K27M tumors were associated with bivalently regulated neural developmental genes.⁸⁰ Silveira et al observed that knockdown of H3F3A, the gene encoding the H3K27M mutation, restored H3K27me3 levels.⁸¹ The restoration of this repressive mark resulted in a promoter shift from active to bivalent state in 32–48% of genes upregulated by K27M.⁸¹ Similarly, a study using ESCs expressing the H3.3 K27me3 mutation found that EZH2 (a component of PRC2) is sequestered to poised enhancers, which was unique to H3K27M-expressing cells compared with wild-type H3.3 ESCs. The authors propose that the limited availability of PRC2 to bind its strong affinity sites across the genome leads to global hypomethylation (Fig. 1B). The authors also found that H3K27M is enriched at highly transcribed genes, and is low within regions that are enriched in H3K27me3 peaks, thereby allowing for H3K27me3 silencing at those sites. Recently, Harutyunyan et al have reported that the global deposition of H3K27me2/3 is reduced in H3.3 K27M cells and that removal of H3K27M could rescue this effect.⁸² These findings suggest that neither the recruitment of PRC2 nor its ability to deposit H3K27me3 in the local proximity of unmethylated cytosine-phosphate-guanine islands is affected by H3K27M. Instead the ability of PRC2 to facilitate the spread of the H3K27me3 mark is inhibited by H3K27M, represented by the reduced distribution of H3K27me3 at these sites (Fig. 1C).⁸² Interestingly, posterior fossa type A ependymomas with high expression of enhancer of zeste homolog inhibitory protein (EZHIP) are also characterized by global loss of H3K27me3 and retention of this mark at cytosine-phosphate-guanine islands.^83–85 Several groups have recently reported that a conserved sequence in EZHIP functions as an endogenous H3K27M analog that inhibits PRC2 activity.^85–87 Jain et al propose a model where EZHIP blocks H3K27me3 spreading by inhibiting allosterically activated PRC2.⁸⁵ Thus, K27M mutation and EZHIP appear to deregulate PRC2 activity by similar mechanisms.

Chromatin Regulators in DIPG

Other genes related to the organization of the chromatin architecture were found to be frequently mutated in DIPG.^14,32,44 In this regard, the alpha thalassemia/mental retardation syndrome X-linked gene (ATRX), encoding a chromatin remodeling protein belonging to the SWItch/sucrose nonfermentable family, has been found to be mutated in different brain tumor types, including adult low-grade glioma, pediatric high-grade glioma, and DIPG.^88–90 Mutations in the death domain associated gene (DAXX), which encodes a protein that forms a complex with ATRX, have been found in these tumors, but to a lesser extent.^14,91 ATRX and DAXX proteins form a complex necessary for the deposition of the histone variant H3.3 at telomeres, pericentric heterochromatin, and other DNA-repetitive regions.^88,89 Telomeres protect chromosome ends and shorten as cells replicate, and eventually this leads to permanent cell cycle arrest known as replicative senescence. Cancer cells can overcome this by activating telomerase or the alternative lengthening of telomeres, and ATRX mutations have been linked to an alternative lengthening of telomeres and genomic instability.^{88,89,92–94}

In DIPG tumors, ATRX/DAXX mutations are not common, occurring in 10% of these midline gliomas, and they tend to be found in older children.¹⁹ In one study, ATRX loss was observed in 24% of H3.3 K27M tumors, and was not observed in H3.1K27M tumors.⁹⁵ However, the specific consequences of ATRX/DAXX mutations and molecular mechanisms explaining its co-occurrence with H3.3 K27M mutations have not been elucidated.

Implications of Epigenetic Reprogramming on Therapeutic Strategies

The prevalence of oncohistone mutations, or missense mutations occurring in histones, in DIPG has led to the investigation of drugs targeting epigenetic regulators, including inhibitors of histone modifications and transcriptional effectors.⁹⁶ A study that analyzed 83 drugs in 16 DIPG patient-derived cell cultures identified panobinostat as a potential therapeutic target for DIPG.⁹⁷ Panobinostat, a nonspecific HDAC inhibitor approved by the FDA for the treatment of multiple myeloma, was effective in vitro and in DIPG xenograft models.^97,98 A different study using xenograft and genetically engineered DIPG mouse models demonstrated that panobinostat reduced tumor cell proliferation and increased H3K27 acetylation in the tumor. However, prolonged intraperitoneal administration of the drug generated toxicity, and well-tolerated doses failed to improve median survival in these models.⁹⁹ In adults, common toxicities associated with panobinostat for the treatment of multiple myeloma included gastrointestinal (diarrhea, nausea, and vomiting) and hematological effects (thrombocytopenia).¹⁰⁰ Ongoing clinical trials using panobinostat in pediatric populations (NCT02717455, NCT03632317) will help shed light on how well tolerated this drug will be. Another concern with panobinostat is that it has poor blood–brain barrier (BBB) permeability. However, convection-enhanced delivery (CED) has emerged as a promising technique for delivery of panobinostat to bypass the BBB and toxicities associated with oral delivery. A study using small and large animal models shows that CED allows water-soluble panobinostat penetration in normal brain without toxicity and could be used in DIPG clinical trials.¹⁰¹ A clinical trial testing the safety of a panobinostat nanoparticle formulation given by CED is also currently under way (NCT03566199). Results from the first clinical trial to test the safety of CED delivery in the brainstem suggest that CED in the brainstem is clinically feasible and safe.¹⁰²

DIPG tumors harboring K27M display decreased global methylation levels in H3K27, inducing an aberrant gene expression associated with tumor malignancy. Therefore, the reversibility of the H3K27 hypomethylation phenotype could mitigate some effects of the epigenetic reprogramming observed in DIPG tumors and is a potential therapeutic approach for this pathology.¹⁰³ This strategy can be implemented by enhancing the activity of methyltransferases or inhibiting the activity of histone K27 demethylases. As described by Hashizume et al, GSKJ4, an inhibitor of Jumonji-domain demethylase JMJD3,^104–106 was tested in brainstem glioma cell lines harboring H3K27M and in wild-type H3.3 and H3.3 G34V glioma cells. GSKJ4 treatment increased levels of K27me2 and K27me3, specifically in cells expressing H3K27M (Fig. 2A).¹⁰⁶ In addition, GSKJ4 decreased cell viability and colony formation in H3K27M DIPG cells and induced apoptosis. In vivo administration of GSKJ4 increased median survival in subcutaneous and orthotopic xenografted H3K27M brainstem tumor models, exhibiting enhanced transport of the compound across the BBB.¹⁰⁶ These results indicate that modulation of H3K27 methylation is a valid therapeutic approach for DIPG tumors. In addition, recent work shows that treatment with GSKJ4 enhanced radiosensitivity of H3K27M DIPG cultures and increased the genotoxic activity of radiation therapy in an orthotopic xenografted H3K27M brainstem tumor model.¹⁰⁷ The authors demonstrate that GSKJ4 reduced expression of DNA double-strand break repair genes, such as PCNA and XRCC1, and DNA accessibility in K27M DIPG cultures.¹⁰⁷ These results provide a rationale for testing a combination therapy of GSKJ4 and radiation.¹⁰⁷

Fig. 2.

Therapeutic strategies to modify aberrant epigenetic and transcriptional dysregulation induced by H3K27M mutation in DIPG. (A) H3K27M mutation inhibits PRC2 from maintaining H3K27me3 marks (green circles) at appropriate binding sites. The recruitment of histone demethylase Jumonji domain containing-3 (JMJD3) to remove H3K27 methyl marks and H3K4-methyltransferase mixed-lineage leukemia protein 2/3 (MLL) to add methyl groups to H3K4 (blue circles) allows gene activation. GSKJ4 blocks JMJD3 from removing H3K27me3. The arrow allows PRC2 to reestablish the repressive mark H3K27me3 at aberrantly active gene regions induced by H3K27M. (B) DIPG cells harboring H3K27M mutations display increased H3K27me3 at the CDKN2A locus associated with decreased p16 expression. Small-molecule inhibitors against EZH2 (GSK343, EPZ6438) can reduce tumor cell proliferation through a mechanism dependent upon the restored tumor suppressor function of p16. (C) CDK7 acts as a regulator of transcriptional initiation by phosphorylating serine 5 on the carboxyl terminal domain of RNA polymerase II, which allows RNA polymerase to leave the gene promoter regions. DIPG cell lines treated with THZ1; a CDK7 inhibitor, displayed retention of RNA polymerase II within the gene body suggesting slower rates of elongation, which hinders transcription. BRrD4 associates with lineage-determining transcription factors and binds to acetyl residues on histones to recruit the transcription machinery to super enhancers. In a mouse model of DIPG, JQ1 a BrD4 inhibitor could induce neuron-like differentiation and delay tumor progression.

EZH2 is a regulator of H3K27 methylation and another potential therapeutic target in DIPG. EZH2 is a catalytic subunit of PRC2, responsible of H3K27 trimethylation, and its overexpression is associated with poor outcomes in solid tumors.¹⁰⁸ Recently, in DIPG cultures obtained from patients, Mohammad et al¹⁰⁹ demonstrated that inhibition of EZH2 using 2 different inhibitors, SK343 and EPZ6438, reduced cell proliferation by inducing expression of p16INK4a tumor suppressor. This is in agreement with results documented by Piunti et al (Fig. 2B), who also show lower cell proliferation after EZH2 inhibition but without participation of p16.⁷⁷ These advances suggest that EZH2 inhibition can be a novel strategy to treat DIPG patients.

In DIPG, H3K27M associates with increased H3K27ac, and H3K27ac co-localizes with BrD4 in the nucleosomes of DIPG cells harboring H3K27M.⁷⁷ In these cells, inhibition of BrD4 with JQ1 decreased cell proliferation and induced a neuron-like phenotype in correlation with transcriptional upregulation of CDKN1A (a cell cycle arrest marker), TUBB3, and MAP2 (differentiated neuronal markers).⁷⁷ In vivo, using a xenograft mouse model of DIPG-K27M, treatment with JQ1 for 10 days reduced tumor size and enhanced animal survival.⁷⁷ The impact of BET inhibition was also confirmed using I-BET151, another BET inhibitor, which decreased DIPG tumor growth.⁷⁷ The results of this study point to BET inhibition as a promising strategy for treatment of DIPG. Another study combined BET inhibition using JQ1 and EZH2 inhibition by EPZ6438 inhibitor¹¹⁰ and demonstrated that administration of both inhibitors of epigenetic regulators have stronger antitumor effects compared with inhibitors alone. In addition, Nagaraja et al¹¹¹ used a DIPG animal model to demonstrate a modest benefit when combining panobinostat with JQ1 or THZ1, a covalent CDK7 inhibitor which alters RNA polymerase dynamics.¹¹² THZ1 disrupted transcription and decreased cell viability of DIPG cells and tumor growth in DIPG animal models (Fig. 2C).¹¹¹ As with panobinostat, DIPG-K27M cells also exhibited resistance to JQ1; however, the combination with THZ1 strengthened the antitumor activity of panobinostat and JQ1. This work suggests that the inhibition of CDK7 can evade drug resistance observed in DIPG cells in response to HDAC or BET protein inhibitors, proposing a novel therapeutic strategy by disrupting RNA polymerase II activity.

It has also been reported that primary human DIPG cells bearing the H3K27M mutation display increased deposition of H3K27ac within intergenic regions encoding endogenous retroviruses (ERVs) compared with isogenic H3K27M knockout counterparts.¹¹³ This finding was validated in a syngeneic H3K27M mouse model, where aberrant transcription of murine repeat elements was observed.^26,113 Genomic silencing of repetitive elements is normally tightly regulated by DNA methylation and by H3K9 and H3K27 trimethylation.¹¹⁴ A DNA demethylating agent (5-azacytidine) and pan-HDAC inhibitor (panobinostat) further promote the transcriptional de-repression of ERVs by impairing the maintenance of DNA methylation or elevating H3K27ac, respectively.¹¹³ Low doses of these agents in combination had an additive cytotoxic effect on H3K27M cells in vitro and extended survival in vivo. ERVs form double-stranded RNA, which trigger innate immune responses within the cell similar to a viral infection (viral mimicry).¹¹⁵ The recognition of double-stranded RNA by DNA pattern receptors (RIG-1 and MDA5) and mitochondrial signaling protein induces an interferon type 1 signaling and promotes tumor cell death.¹¹⁵ Epigenetic modulators present an effective therapeutic strategy to improve antitumor response in H3K27M high-grade glioma.

Conclusions

Alterations in developmental and epigenetic mechanisms drive disease progression in DIPG. It is likely that DIPGs arise from an NPC that accumulates characteristic mutations. It is thought that the histone H3K27M mutation serves as the initial event, making the transformed cells susceptible to acquiring TP53, ACVR1, and additional mutations.^116,117 The H3K27M mutation alters chromatin organization by impairing PRC2-mediated deposition of H3K27me3 either directly by catalytically inactivating PRC2, restricting the spread of the H3K27me3 mark or sequestering PRC2 at sites it is usually not found. Several labs have characterized H3K27M-induced repression of p16, where the cyclin-dependent kinase inhibitor 2A (CDKN2A) promoter retains H3K27me3 despite the presence of the oncohistone. However, to date, no work has investigated additional sites that gain H3K27me3 enrichment and their potential contribution to gliomagenesis in H3K27M mutated DIPGs. It has also been shown through isogenic cell lines that the chromatin remodeling induced by H3K27M can be reversible upon the removal of the mutation. Future studies utilizing these isogenic cell lines following epigenetic modifiers’ treatment will enable us to further discern unique susceptibilities as a consequence of H3K27M. It will also be necessary to demonstrate the in vivo efficacy of epigenetic modulation using various animal models that express the precise genetic lesions found in DIPG patients. In addition, it will be necessary to demonstrate an efficient delivery modality that guarantees sufficient drug penetration within the brain tumor tissue with the goal of providing antitumor effects that rebalance the tumor epigenome.

Funding

This work was supported by National Institutes of Health/National Institute of Neurological Disorders & Stroke (NIH/NINDS) Grants R37-NS094804, R01-NS105556, R21-NS107894 to M.G.C.; NIH/NINDS Grants R01-NS076991, R01-NS082311, and R01-NS096756 to P.R.L.; NIH/NIBIB R01-EB022563 and NCI/UO1-CA-224160 to M.G.C. and P.R.L.; the Department of Neurosurgery; Leah’s Happy Hearts Foundation, ChadTough Foundation, Pediatric Brain Tumor Foundation, and Smiles for Sophie Forever Foundation to M.G.C. and P.R.L. RNA Biomedicine Grant F046166 to M.G.C. NIH/NINDS-F31NS103500 and Rackham Pre-doctoral Fellowship to F.M.M; S.C. is supported by NIH/NCI-T32-CA009676. OJB is supported by R01-CA197313, American Cancer Society RSG-16-218-01-TBG, the Rory David Deutsch Foundation, Max Cure Foundation, Cancer Smashers, and Madox’s Warriors.

Conflict of interest statement. The authors declare that there is no conflict of interest regarding the publication of this article.