Highlights from the Literature

Pan-mutant IDH1 Inhibitor BAY 1436032 for Effective Treatment of IDH1 Mutant Astrocytoma In vivo

Mutations of the isocitrate dehydrogenase gene 1 are frequent in diffuse glioma and are also found in some cases of acute myeloid leukemia, chondrosarcoma, and intrahepatic cholangiocarcinoma. IDH mutations affect all tumor cells in many cases of glioma and are not found in non-neoplastic cells, thus making them highly attractive targets for experimental therapies. Several pre-clinical and early-phase clinical studies are currently investigating various approaches of IDH-targeted treatment such as vaccination strategies and IDH inhibitors. Pusch et al report the successful development of a novel compound, BAY1436032, that inhibits IDH after oral administration. Several features of BAY1436032 make it very interesting for further development in the clinical setting. First, BAY1436032 showed not only a significant effect on tumors carrying the most frequent IDH1 mutation, IDH1-R132H, but also on four less frequent IDH mutations. Furthermore, and in line with previous reports on other IDH inhibitors, BAY1436032 induced differentiation of glioma cell cultures. In two independent experiments the authors document prolonged survival times of mice treated with BAY1436032 in comparison to control animals. A phase I study (NCT02746081) has been initiated and will determine the safety profile of BAY1436032 in humans. It is noteworthy that Pusch et al observed an off-target effect of BAY1436032 on angiotensin 2 (AT2) in their in vitro experiments, and it will be interesting to analyze whether this translates into a clinically relevant influence on the blood pressure of patients exposed to this drug. Overall, the data presented by Pusch et al are very promising and raise the hope for a role of IDH inhibitors in the treatment of patients with IDH-mutant diffuse glioma.

Epigenetic Targeting of H3K27-mutant Diffuse Intrinsic Pontine Gliomas

Diffuse intrinsic pontine glioma (DIPG) is a highly lethal pediatric tumor for which new therapeutic options are needed. A heterozygous missense mutation of histone H3 at the lysine 27 residue (H3K27M) is found in the majority of DIPG cases, and the mutant protein results in epigenetic consequences, since the lysine 27 residue can normally be trimethylated (H3K27me3, by PRC2) or acetylated (H3K27ac, by p300/CBP), with downstream repressed or activated transcription, respectively. Two recent studies, published in the same issue of Nature Medicine, investigate the role of H3K27M mutant protein in DIPG and describe distinct approaches to targeting this aberrant epigenetic pathway.

The Mohammad et al paper² focuses on chromatin modifier EZH2, a component of PRC2. They show that H3K27M enhances tumorigenesis in mice, in a manner that requires EZH2 activity. Whereas the level of H3K27me3 decreased globally in their mouse model of DIPG, many regions of the genome harbored H3K27me3, and where present, PRC2 was confined to these regions, localized to CpG islands that normally have high H3K27me3. Interestingly, they found that EZH2 activity remains locally at the Ink4a promoter in H3K27M cells, resulting in repression of the Ink4a promoter and the maintenance of cell proliferation. Blocking EZH2 activity using a small-molecule inhibitor reactivated p16INK4A expression and growth arrest, and further showed that the effect of EZH2 inhibition on cell proliferation is dependent on the restoration of p16INK4A expression.

The companion paper, by Piunti et al,³ uses a different approach. By first mapping mutant H3K27M on chromatin in DIPG cells, they show that H3K27M co-localizes with H3K27ac and RNA polymerase II (pol II), indicating sites of active transcription. They find that the transcriptional bromodomain activators BRD2 and BRD4 (BET proteins that read acetylated lysines on chromatin) co-localized with most of the sites occupied by H3K27M. Notably, PRC2 and H3K27me3 were absent from these regions, suggesting that H3K27M competes with H3K27me3, with exclusion of PRC2 from chromatin. The absence of the repressive mark permits acetylation of H3K27 with subsequent transcriptional activity. Treatment with bromodomain (BET) inhibitors caused decreased proliferation/increased differentiation of DIPG cells in vitro and prolonged survival of mice with DIPG tumors. Genes that were downregulated by BET inhibition were correlated with genes that were found to be occupied by H3K27M-BRD2/4.

Together, these papers shed substantial light on the epigenetic mechanisms by which the H3K27M mutant protein mediates gliomagenesis in DIPG and provide robust pre-clinical evidence of possible therapeutic strategies. Notably, BET and EZH2 inhibitors are currently in clinical development, and given the functional connection of H3K27M and H3K27ac, coordinate histone deacetylase (HDAC) inhibition could also play an important role. Overall, the findings pave the way for clinical trial strategies in DIPG, using BET, EZH2 or HDAC inhibition, either individually or in combination.

Transcriptional Dependencies in Diffuse Intrinsic Pontine Glioma

Diffuse intrinsic pontine gliomas (DIPG) are a universally fatal disease that has been challenging to study, but the discovery that these tumors harbor a mutation in histone-3 (H3K27M) has highlighted the importance of transcription dysregulation in the pathogenesis of these tumors. Based on this information, a trial of panobinostat, a histone deacetylase inhibitor, in DIPG patients is currently underway. Resistance to panobinostat has developed, however, so Monje and colleagues have extended their prior work in DIPG⁴ to try to identify rational combination therapies.

Recognizing that transcription is mediated by RNA polymerase II (RNAPII), which complexes with bromodomain-containing protein 4 (BRD4) and cyclin-dependent kinase 7 (CDK7), Nagaraja et al targeted BRD4 with the drug JQ1 and CDK7 with THZ1.⁵ In DIPG cultures, they found that JQ1 appeared to be cytostatic but did have a synergistic impact on DIPG cell viability when combined with panobinostat. Cells that were resistant to panobinostat were also resistant to JQ1, and further transcriptome analysis to tease out the reason for this demonstrated that similar genes were targeted by both JQ1 and panobinostat. THZ1 also demonstrated synergism when combined with panobinostat, but panobinostat-resistant cells retained their sensitivity to THZ1. Transcriptome analysis revealed that distinct genes were targeted by panobinostat and THZ1, likely explaining the preservation of efficacy with THZ1. Unfortunately, both JQ1 and THZ1 have poor brain penetration, so drugs with better brain penetration need to be developed to capitalize on these promising combination strategies.

The transcriptome analysis performed to better understand why targeting BRD4 and CDK7 was effective in suppressing DIPG cell viability revealed that super enhancers, clusters of enhancers that drive transcription, were preferentially disrupted by JQ1 and THZ1. Therefore, the authors further evaluated super enhancers and the chromatin landscape in DIPG and found that oligodendroglia lineage genes (e.g. OLIG1, OLIG2), and, in particular, early precursor genes (e.g. SOX2) were associated with super enhancers in DIPG—findings that suggest an early oligodendroglia precursor may be the cell of origin for DIPG.⁵

The super enhancer analysis also identified associations with genes involved in neurodevelopement/neuronal communication and K⁺ channel regulation in DIPG. The EPH-ephrin pathway is important in migration and invasion, and when the authors blocked this pathway in culture, there was a decrease in cell invasion, suggesting another therapeutic target.⁵ Blocking K⁺ channels also led to a decrease in cell viability. Thus, in a very comprehensive, thorough piece of work, Nagaraja et al show several possible paths forward for combination strategies to treat this challenging pediatric cancer.

In vivo Detection of EGFRvIII in Glioblastoma via Perfusion MRI Signature

Identifying imaging correlates of molecular subtypes of glioma can improve understanding of tumor biology and provide a non-invasive method to globally and repeatedly surveil a tumor during the course of treatment. Prior work has identified a number of associations between imaging phenotypes and gene expression and other molecular classifiers of glioma that impact prognosis and potentially treatment susceptibility.

Recently, Bakas et al⁶ analyzed the relationship between MRI measurements of tumor perfusion and the expression of the epidermal growth factor receptor variant III (EGFRvIII) in glioblastoma (GBM). The EGFRvIII mutation has been previously shown to be associated with shorter survival and increased tumoral perfusion.⁷ In the current work the authors analyzed perfusion not in the enhancing tumor mass, but in the areas of T2-weighted signal change around the tumor, which likely represent a combination of edema and infiltrative tumor cells. Specifically, they measured perfusion using dynamic susceptibility contrast MRI immediately adjacent to, and distant from (but still within the area of T2 signal abnormality), the enhancing tumor margin. Using these data they were able to construct a heterogeneity index (essentially, a measure of the difference in perfusion close to and far from the tumor margin), first in a discovery cohort of 64 GBM patients and then in an independent test cohort of 78 GBM patients. A heterogeneity index of 0 indicates similar perfusion near and far from the tumor, whereas an index of 1 indicates significant difference (higher at the tumor margin than more distantly). For the combined cohort (n = 142, 42 EGFRvIII⁺), the heterogeneity index for EGFRvIII⁺ was 0.096 compared with 0.28 for EGFRvIII^- tumors, with a p value of 4.0 × 10⁻¹⁰. Thus, they were able to achieve an accuracy of nearly 90% in distinguishing tumors with versus tumors without the EGFRvIII mutation, a gain in accuracy of approximately 10% compared with prior work at the same institution using perfusion measurements from the enhancing tumor itself.⁷ The authors hypothesize that because EGFRvIII⁺ tumors are highly infiltrative and migratory, they exhibit less variation of perfusion values within the peritumoral T2-abnormal zone. Conversely, the EGFRvIII^- tumors potentially have a more confined zone of infiltration and vascularization, reflected in a steep gradient of perfusion values.

Future directions include the possibility of following this index as patients are treated with EGFRvIII-targeted or other therapies, as tumors are known to modulate EGFR levels through gene amplification and also may lose or down-regulate EGFRvIII expression. Since the heterogeneity index is based on within-patient rather than population measurements, the ability of this metric to be translated to a clinical setting for individualizing patient therapy is highly promising.

IDH-mutant Astrocytoma Versus Oligodendroglioma: What is the Biological Basis of this Distinction?

An important issue in classification of lower-grade diffuse glioma is the distinction between astrocytoma and oligodendroglioma. These tumors were named according to their presumed cell of origin, based on morphologic appearance, but specific protein markers to distinguish these entities have proved elusive. Whereas markers such as IDH, ATRX, and 1p/19q provide useful information, much effort and training goes into distinguishing these WHO-defined entities.

A recent manuscript by Venteicher et al sheds light on this distinction. The authors performed RNAseq on more than 9800 cells from 10 IDH-mutant astrocytomas (IDH-A) and more than 4300 cells from 6 IDH-mutant oligodendrogliomas (IDH-O). In tandem, they compared differences between astrocytoma and oligodendroglioma bulk tumors using TCGA data. Interestingly, they found that only a subset of the genes differentially expressed in the bulk tumors were also differentially expressed when comparing the single-cell RNAseq data with the tumor cells. The remaining genes could be attributed to a component of non-malignant cells, with macrophage-specific genes present in astrocytomas and neuron-specific genes in oligodendrogliomas. Of note, there were only limited differences in the expression of astrocyte-specific and oligodendrocyte-specific genes between IDH-A and IDH-O present in their data, further supporting the hypothesis that the respective cells of origin of these tumors are not as distinct as predicted. The changes in these genes that were differentially expressed by the tumor cells between IDH-A and IDH-O cells were seen primarily in the well-differentiated tumor cells and were absent in the undifferentiated cancer cells.

Overall, these findings suggest that cells within the tumor microenvironment (TME) vastly influence the bulk tumor profile and underlie the difference commonly observed in genomic profiling studies and in establishing the specific tumor subclasses. The authors went on to examine gene expression differences as a function of tumor grade. Comparing IDH-mutant tumors ranging from grade II to IV, they found that high-grade lesions show larger pools of undifferentiated cells, changes in differentiation programs, and increased infiltration by macrophages over resident microglia. These findings emphasize the critical role of the TME and its composition in response to treatment and design of new strategies that would specifically target cells within the TME. The authors hypothesize that differences in the composition of the TME may be driven by genetic factors such as TP53 mutations found in diffuse astrocytomas that may affect immune regulatory pathways and perhaps lead to microglial recruitment in tumors with this genetic background. Most importantly, this study highlights shared glial lineages that suggest a common progenitor for these two IDH-mutant gliomas and questions the hypothesis of distinct glial lineages for IDH-mutant astrocytoma and oligodendroglioma.

Brain Tumor Heterogeneity Dictates the Number of Biopsies Required to Plan Targeted Therapies

The current molecular classification of glioblastomas (GBMs) and medulloblastomas (MBs) uses gene expression profiling to assign tumors into subtypes with distinct transcriptional signatures. However, regional⁹ and single-cell¹⁰ sequencing of GBM have shown that even when a tumor is classified as a particular subtype, it contains clones that may have considerably different gene expression from the assigned subtype. This intra-tumoral heterogeneity is a major obstacle for targeted therapies directed against a single mutation or gene product, because clones lacking the target will be resistant to treatment and a source of recurrence.

To provide a comprehensive view of this heterogeneity, Morrissy and collaborators have performed an exhaustive transcriptomic and genomic profiling of multiregion biopsies from both GBM and MB.¹¹ They observed that at least 20% of the regional biopsies from a given GBM (and as high as 60%) had transcriptional signatures different from the overall subtype of the tumor. In a remarkable contrast, almost all of the biopsies from each MB were transcriptionally similar and matched the predominant tumor subtype, suggesting that a single biopsy is sufficient to identify conserved gene expression patterns in MB but not in GBM. Strikingly, the regional homogeneity of MBs at the transcriptional level was not matched by a homogeneous profile at the DNA level: when the authors analyzed copy number alterations and somatic mutations they found that both GBMs and MBs had similar regional heterogeneity. Some tumors were homogeneous and their regional biopsies contained the same genomic alterations, whereas others were very heterogeneous and exhibited clonal and subclonal mutations present only in one or few biopsies. Using phylogenetic algorithms the authors reconstructed the lineage of clones within each tumor and concluded that genomic alterations detected in single biopsies from heterogeneous tumors (both GBMs and MBs) may not represent alterations conserved across the tumor, including mutations in known driver genes.

Therefore, a major take-home message of this work is the demonstration that single regional biopsies may be insufficient to provide genomic information that can be generalized to the whole tumor to direct targeted therapies. Tumors that are genomically heterogeneous (as is the case in the majority of GBMs) may require at least four spatially different biopsies to identify therapeutically actionable mutations that are retained across all regions of the tumor. A minimum of two biopsies may provide sufficient information only to determine if a tumor is genomically heterogeneous, thus requiring further biopsies to reconstruct its genomic variability and to direct targeted therapies. Accordingly, Morrissy et al conclude that the relative homogeneity of the transcriptome, at least in MB, may offer better therapeutic alternatives against targets that may not be mutational drivers but are similarly expressed by different clones across a tumor (for example, targets for immunotherapy).