NTRK Fusions Can Co-Occur With H3K27M Mutations and May Define Druggable Subclones Within Diffuse Midline Gliomas

Abstract

Diffuse midline gliomas (DMGs) are incurable pediatric tumors with extraordinarily limited treatment options. Decades of clinical trials combining conventional chemotherapies with radiation therapy have failed to improve these outcomes, demonstrating the need to identify and validate druggable biologic targets within this disease. NTRK1/2/3 fusions are found in a broad range of pediatric cancers, including high-grade gliomas and a subset of DMGs. Phase 1/2 studies of TRK inhibitors have demonstrated good tolerability, effective CNS penetration, and promising objective responses across all patients with TRK fusion-positive cancers, but their use has not been explored in TRK fusion-positive DMG. Here, we report 3 cases of NTRK fusions co-occurring within H3K27M-positive pontine diffuse midline gliomas. We employ a combination of single-cell and bulk transcriptome sequencing from TRK fusion-positive DMG to describe the phenotypic consequences of this co-occurring alteration. We then use ex vivo short-culture assays to evaluate the potential response to TRK inhibition in this disease. Together, these data highlight the importance of routine molecular characterization of these highly aggressive tumors and identify a small subset of patients that may benefit from currently available targeted therapies.

INTRODUCTION

Diffuse intrinsic pontine gliomas (DIPGs) are almost uniformly fatal tumors arising in the brainstem of young children. Radiation therapy, the only established standard of care, provides temporary relief, but fewer than 10% of children are alive at 2 years from time of diagnosis (1, 2). There are currently no effective systemic therapies for this disease. For years, development of new treatments has been hampered by a lack of available tumor tissue, as the eloquent neurologic location precluded routine surgical intervention due to safety concerns. However, diagnostic biopsy has now proven safe in the hands of experienced neurosurgeons (3) and resulted in new opportunities to both understand the biology of these tumors and to identify genetic alterations that may be clinically actionable (4).

Approximately 80% of DIPGs are now categorically redefined as diffuse midline gliomas, H3K27M-mutant (DMGs) according to the presence of recurrent mutations in histone H3.3 or H3.1, most commonly in the H3F3A or HIST1H3B genes, respectively (5). However, they frequently harbor other co-occurring somatic alterations (6). Comprehensive genomic characterization of pediatric high-grade gliomas (HGGs) has described neurotrophic receptor tyrosine kinase (NTRK) fusions in 10% of all nonbrainstem pediatric HGGs and 40% of nonbrainstem HGGs occurring in children younger than 3 years of age (7). This dataset also identified 2 patients (4%) with classical DIPGs harboring NTRK fusions, described in detail below.

Recurrent chromosomal translocations involving the NTRK1, NTRK2, and NTRK3 genes have been described across a range of adult and pediatric malignancies (8, 9). These fusions generate constitutively-active chimeric proteins containing the neurotrophin tyrosine kinase family of receptors (TRKA, TRKB, TRKC), which in turn mediate embryonal growth and neuronal differentiation pathways (10–12). TRK fusions are canonical drivers in rare tumors such as infantile fibrosarcoma and congenital mesoblastic nephroma (13, 14), but they have also been identified at low frequencies across a wide spectrum of solid tumor histologies (9, 15). The highly selective small-molecule TRK inhibitors larotrectinib and entrectinib have recently obtained FDA accelerated approval after demonstrating marked antitumor activity in TRK fusion-positive cancers, agnostic of tumor histology or patient age (16–18).

While efficacy of larotrectinib and entrectinib have been demonstrated in nonbrainstem pediatric HGGs (19–21), their potential role in TRK fusion-positive DMG has not been previously examined. Here, we report the clinical characteristics of 3 children with DMGs found to be TRK fusion-positive, examine the transcriptional relevance of NTRK fusions in these tumors at a single-cell level, and discuss the therapeutic implications for this actionable target in an incurable disease.

CASE PRESENTATION

Case 1

A 10-year-old girl presented to the Emergency Department for evaluation of progressive ataxia, diplopia, and left hemiparesis which had worsened over the previous 6 weeks. She had no significant medical history or preceding ill symptoms. MRI demonstrated classical pontine DMG appearance, with an expansile T2- and FLAIR hyperintense lesion centered in the pons with extension into bilateral cerebellar peduncles (Fig. 1). Per institutional standard of care, she underwent stereotactic needle biopsy via a right transcerebellar approach. Histopathology demonstrated glial fibrillary acidic protein (GFAP)-positive, H3K27M-positive pleomorphic tumor infiltrating surrounding pontine neurons (Fig. 2A–D). A clinical next-generation sequencing panel revealed a ZKSCAN1:NTRK3 fusion, which was reconfirmed on whole transcriptome sequencing (RNA-seq) performed on the biopsy specimen. Pan-TRK immunohistochemistry revealed subpopulations of TRK-positive tumor cells admixed within predominately TRK-negative tumor tissue (Fig. 2E). The patient began conventionally-fractionated radiotherapy, and family was approached regarding TRK inhibitor therapy on a compassionate use basis. However, the patient experienced rapid clinical progression despite radiation therapy, corticosteroids, and bevacizumab, and she died before any targeted therapy could be initiated.

FIGURE 1.

NTRK fusions co-occur in radiographically typical DMG. (A) Axial (left) and sagittal (right) fluid attenuation inversion recovery (FLAIR) MRI sequences demonstrate classic DMG appearance, with expansile mass seen centered in the pons with extension into cerebellar peduncles. Flow void artifact due to presence of orthodontic hardware. (B) Axial (left) and sagittal (right) postgadolinium T1 MRI sequences redemonstrate expansile pontine lesion with characteristic exclusion of contrast enhancement.

FIGURE 2.

NTRK fusions co-occur in histologically typical DMG. (A) and (B) Hematoxylin and eosin (H&E) stained representative sections show a markedly pleomorphic infiltrative tumor, tumor nuclei (arrows) are seen surrounding pontine neurons (arrowheads). (C) Tumor cells are immunopositive for GFAP. (D) Immunohistochemistry for histone 3 K27M showing strong nuclear positivity in the tumor cells. (E) Pan-TRK immunostain demonstrates populations of positive (top panel and black arrows, lower panel) and negative (white arrows, lower panel) tumor cells. Black arrowheads indicate normal neurons. Scale bars: A = 50 µm; B–E = 20 µm.

Case 2

A 5-year-old boy presented to a community hospital with cranial nerve palsies and pyramidal tract signs that had developed rapidly over one week’s time. MRI demonstrated expansile pontine mass consistent with a diagnosis of diffuse intrinsic pontine glioma. He unfortunately experienced rapid neurologic deterioration and died prior to initiation of treatment at a pediatric tertiary care center. Autopsy results revealed a diffusely infiltrative high-grade glial lesion with H3K27M mutation. Whole genome sequencing performed on a research basis showed TP53 loss and a complex BTBD1:CPEB1:NTRK3 fusion occurring in the setting of chromothripsis.

Case 3

A 6-year-old boy presented for evaluation following 2 weeks of progressive cranial nerve palsies and ataxia. Diagnostic MRI again demonstrated T2 hyperintense expansile mass centered in the pons, consistent with diffuse intrinsic pontine glioma. This patient was diagnosed based on radiographic criteria and did not undergo diagnostic biopsy. He was started on dexamethasone and underwent conventional radiation therapy. After approximately 5 months, the patient demonstrated worsening clinical symptoms and radiographic findings consistent with disease progression. He was started on conventional chemotherapy (etoposide 50 mg/m² daily and cyclophosphamide 100 mg daily) with palliative intent, and he died approximately 8 months from initial presentation. Autopsy confirmed high-grade glial histology with an H3K27M mutation, and whole genome sequencing documented a VCL:NTRK2 fusion.

MATERIALS AND METHODS

Human Tissue Samples

Frozen and formalin-fixed, paraffin embedded tumor specimens were collected from biopsy and autopsy specimens at the Children’s Hospital Colorado and were obtained in accordance with local and federal human research protection guidelines and Institutional Review Board (IRB) regulations (COMIRB No. 95-500). Informed consent was obtained for all specimens collected. ScRNAseq samples were rapidly dissociated into single cells using a mechanical process as described previously (22), viably cryopreserved and banked at <80°C for later use.

Immunohistochemistry

Fresh biopsy specimens were fixed in 10% neutral-buffered formalin for at least 6 hours. Following fixation, tissues received standard histologic processing in a Clinical Laboratory Improvement Amendments (CLIA)-certified clinical laboratory. Five-millimeter-thick sections from paraffin-embedded tissues were stained with hematoxylin and eosin and were further processed for immunohistochemical staining with GFAP (1:2500 dilution; Dako, Glostrup, Denmark), H3K27M (Millipore catalogue No. ABE419, dilution), Pan TRK (Roche-Ventana, prediluted).

Short-Term Patient-Derived Primary Culture

Tumor samples were collected in the autopsy suite and manually disaggregated prior to freezing in liquid nitrogen. After thawing, cells were sorted for viability using propidium iodide at the University of Colorado Flow Cytometry Core. DMG cells were cultured in tumor stem media consisting of Neurobasal(-A) (Invitrogen, Carlsbad, CA), B27(-A) (Invitrogen), human-basic FGF (20 ng/mL; Shenandoah Biotech, Warwick, PA), human-EGF (20 ng/mL; Shenandoah Biotech), human PDGF-AB (20 ng/mL; Shenandoah Biotech), and heparin (10 ng/mL).

Ex Vivo Drug Treatment

Larotrectinib was obtained commercially from Selleckchem (Houston TX, Cat No. S7960). Entrectinib was purchased from MedChemExpress (Monmouth Junction NJ, HY-12678). Both were dissolved in DMSO per vendor instructions and diluted in media as described above. DMG cells were seeded in 96 well plate at a density of 15 000–50 000 cells per well. Control DMG and NTRK DMG cells were treated at a dose of 100 nM/day for 3 days. Equivalent volume of DMSO was used for control. After an additional 48 hours, cell viability was assessed by manual microscopy performed by 2 independent observers following trypan blue staining. Ex vivo experiments were performed in minimum n = 2 biological replicates. Data displayed are as mean ± SEM. Statistical analysis was performed in GraphPad Prism 8.0 (GraphPad, La Jolla, CA) using two-tailed unpaired Student t-test. Statistical significance was established as p < 0.05. Statistical analysis was performed prior to normalization for display purposes.

Transcriptome Sequencing (RNA-Seq)

Ribonucleic acid was isolated using a Qiagen miRNAeasy kit (Valencia, CA) and measured on an Agilent Bioanalyzer (Agilent Technologies, Santa Clara, CA). Illumina Novaseq 6000 libraries were prepared and sequenced by the Genomics and Microarray Core Facility at the University of Colorado Anschutz Medical Campus. High-quality base calls at Q30 ≥ 80% were obtained with ∼40 M paired-end reads. Sequenced 150 bp pair-end reads were mapped to the human genome (GRCh38) by STAR 2.4.0.1, read counts were calculated by R Bioconductor package GenomicAlignments 1.18.1, and differential expression was analyzed with DESeq2 1.22.2 in R. RNA extraction and sequencing for SJHGG004 and SJHGG009 were performed as previously described (7). Further analysis by gene set enrichment analysis (GSEA) was performed in GSEA v2.1.0 software with 1000 data permutations (23). Genes were considered significantly differentially expressed if meeting fold change </> 2 and p-adj <0.05. Normalized enrichment scores and estimated false discovery rate as defined by GSEA are listed in figure panels. Gene Ontology Metascape analysis was performed using Metascape software (http://metascape.org/gp/index.html) using default settings (24).

Single-Cell RNAseq

For scRNAseq, samples were thawed and flow sorted (Astrios EQ, Beckman Coulter, Indianapolis, IN) to obtain viable single cells based on propidium iodide exclusion. With the study goal of performing scRNAseq on 2000 cells per sample, we utilized a Chromium Controller in combination with Chromium Single Cell V2 Chemistry Library Kits, Gel Bead & Multiplex Kit, and Chip Kit (10× Genomics, Pleasanton, CA). This approach involves the isolation of single cells into microfluidic droplets containing oligonucleotide-covered gel beads that capture and barcode the transcripts. Transcripts were converted to cDNA, barcoded, and sequenced on Illumina HiSeq4000 and NovaSeq6000 sequencers to obtain approximately 50 000 reads per cell.

Raw sequencing reads were demultiplexed, mapped to the human reference genome (build GRCh38) and gene-expression matrices were generated using CellRanger (version 3.0.1). The resulting count matrices were further filtered in Seurat 2.1.0 (https://satijalab.org/seurat/) to remove cell barcodes with less than 250 genes, more than 20% of UMIs derived from mitochondrial genes, or more than 7000 UMIs (to exclude putative doublets) (25). We applied Harmony alignment to correct for intersample variability due to experimental or sequencing batch effects (26). After normalization, these cells were clustered using the Seurat workflow based on dimensionality reduction by principle component analysis using the most variable genes and visualized using Uniform Manifold Approximation and Projection (UMAP).

Chromosomal CNVs of single cells from scRNAseq were inferred on the basis of average relative expression in variable genomic windows using InferCNV (https://github.com/broadinstitute/inferCNV) (27). Cells classified as nonneoplastic were used to define a baseline of normal karyotype such that their average copy number value was subtracted from all cells.

RESULTS

NTRK Fusions May Alter Cellular Architecture Within DMG Intratumoral Subpopulations

In order to interrogate both the inter- and intratumoral effects of the NTRK fusion at the level of individual cell architecture, we performed single-cell transcriptome sequencing (scRNAseq) on our sentinel NTRK DMG and 2 H3K27M-positive, non-TRK-fused DMG (H3K27M-positive/NTRK-negative, denoted control DMG). Autopsy samples from each patient were rapidly collected, dissociated to single cells, and viably cryopreserved, an approach which controls for experimental and batching variance while preserving scRNAseq analytic yield (28). Cell profiling on a Chromium 10× Genomics platform yielded 1568 cells that passed viability and quality controls (n = 844 cells control DMG No. 1, n = 315 cells control DMG No. 2, n = 409 cells NTRK DMG). Cell profile cluster variability by patient of origin that has been previously reported in scRNAseq analyses of DMG and other high-grade gliomas (29, 30) largely resolved following Harmony alignment (26), highlighting the consistency and reproducibility of this workflow (Fig. 3A and Supplementary Data Fig. S1).

FIGURE 3.

NTRK fusions may alter cellular architecture within DMG intratumoral subpopulations. (A) UMAP projection of all single cells from NTRK DMG and non-NTRK DMG control (n = 844 cells control DMG No. 1, n = 315 cells control DMG No. 2, n = 409 cells NTRK DMG). (B) Distinct cell classes with characteristic expression overlaid on UMAP space. Insert reflects neoplastic subpopulations with top classifier genes overlaid. (C) Single-cell inferred copy number variation profiles of immune control (top) and neoplastic (bottom) cells. Color code on left corresponds with UMAP cell-type-specific cluster. (D) Expression of STMN2 overlaid on UMAP space (purple = high, grey = low) (left) with proportion of sample of origin represented in STMN2-high subcluster (right).

To identify and focus on neoplastic cells, we combined 2 parallel approaches. First, we profiled individual cell clusters according to marker gene expression for established cell states (ie immune cells, cycling/mitotic cells) (Supplementary Data Table S1, see Materials and Methods). This identified clear myeloid and lymphoid populations in common across samples, as well as putative neoplastic clusters (Fig. 3B). This finding is consistent with prior studies which report greater variability amongst malignant subpopulations than in nonneoplastic cells when compared across patients (30–32). We next inferred copy number variations for each chromosomal region within the neoplastic populations with respect to average expression in nonneoplastic controls. This identified large gains and losses within most neoplastic cells from both control and NTRK DMG samples (Fig. 3C). Together, this parallel approach yielded a concordant classification of neoplastic subpopulation clustering within the dataset.

Finally, we examined the neoplastic subpopulations both shared between and unique to the NTRK and control DMG samples. The DMG clustering broadly recapitulated the cellular architecture of an oligodendrocyte precursor cell-derived tumor as has been previously described (29). Subpopulations of oligodendrocyte-like (PLP1, ERMN, and MBP), astrocyte-like (FABP7, GFAP), and mitotic/cycling (TOP2A, CENPF) cells were observed within each patient sample (Fig. 3B and Supplementary Data Figs. S1–S3). Interestingly, the NTRK DMG malignant cells harbored an additional subpopulation defined by neuronal-lineage markers such as STMN2 and ROBO3 that was largely unique to this patient sample (Fig. 3D and Supplementary Data Fig. S4). Neuronal lineage programs characterized by high stathmin expression (STMN1, STMN2, STMN4) have been described in H3 wild-type glioblastomas (30), but they have not been previously observed within H3K27M-mutant DMG architecture (29). Functional limitations in the Chromium platform’s 3′ sequencing approach preclude identification of the fusion transcript within this subpopulation. Importantly, observations from a single patient tumor do not allow for broader extrapolations or conclusions regarding a tumor phenotype. However, this shift in prevailing transcriptional programs echoes known NTRK signaling upstream of diverse neuronal developmental functions (9, 12, 33), and it introduces the possibility that the co-occurring presence of this fusion may significantly alter cell state architecture within these tumors.

NTRK Fusions Are Associated With Neuronal Transcriptional Programs in DMG

To evaluate the downstream transcriptional consequences and predicted functional relevance of these NTRK fusions in DMG, we compared bulk RNA-seq data from 2 primary patient samples of NTRK DMG (Cases 1 and 3) in comparison to 3 DMG controls. This identified 495 differentially expressed genes with fold change </> 2 and p-adj <0.05 (348 genes up, 147 genes down) (Fig. 4A and Supplementary Data Table S2A). Unsupervised principle component analysis of all expressed genes by RNA-seq from primary patient samples revealed a divergence in NTRK DMG expression signatures with respect to a relatively conserved control DMG clustering (Fig. 4B). Gene set enrichment analysis (GSEA) of differentially-expressed genes demonstrated a modest enrichment in transcriptional programs associated with neuronal differentiation with a corresponding shift away from defined mesenchymal or astroglial signatures (Fig. 4B). Similarly, gene ontology analysis revealed significant enrichment in functional networks associated with neuronal morphogenesis or synaptic signaling organization (Fig. 4C and Supplementary Data Table S2B and Fig. S5). These expression analyses resemble the cellular architecture suggested by our scRNAseq data, and they are consistent with previous reports which describe TRK-containing chimeric proteins as driving constitutively active, aberrant downstream signaling cascades (34, 35). Given the limited number patient samples available, however, we cannot exclude the possibility that the observed expression changes reflect, at least in part, interpatient variability as opposed to conserved TRK-dependent effects. Likewise, observed interpatient variability precludes robust characterization of an NTRK-dependent DMG phenotype. Taken together, however, these data do suggest that NTRK fusions are functionally active and could represent a relevant therapeutic target within the DMG cells that harbor them.

FIGURE 4.

NTRK fusions generate transcriptionally-active therapeutic targets in DMG. (A) Unsupervised hierarchical clustering of 495 differentially expressed genes from NTRK DMG (n = 2) vs non-TRK DMG controls (n = 3) (log fold change </> 2 and p-adj <0.05). (B) Principle component analysis (PCA) and representative GSEA plots demonstrate significant divergence in neuronal vs astroglial or mesenchymal gene sets. (C) Gene Ontology Metascape analysis demonstrates functional enrichment in neurotrophin-mediated signaling programs. Each node denotes an enriched term, and networked clusters are indicated by different colors. (D) Schema reflecting proof-of-concept ex vivo testing. (E) Relative cell numbers following 100 nM entrectinib treatment (two-tailed t-test, control DMG No. 1 min. n = 2 p = 0.30, control DMG No. 2 min. n = 3 p = 0.18, NTRK DMG min. n = 4 p < 0.0001).

NTRK Fusion Status Correlates With Ex Vivo Response to TRK Inhibitors

Finally, we sought to determine whether NTRK DMG cells would respond to TRK inhibition ex vivo. Reported IC₅₀ values for TRK inhibitors in other NTRK-driven, non-DMG patient derived culture models have ranged from <10 nM to approximately 100 nM (36–39). We took primary patient tumor from our sentinel NTRK DMG, which was collected immediately postmortem and disaggregated to single cells. Flow-sorted viable cells were established in short-term culture, and cells were then treated with either larotrectinib or entrectinib at 100 nM per day for 3 days (Fig. 4D). A decrease in NTRK DMG live cell number was observed with TRK inhibitor treatment in comparison to DMSO control (larotrectinib mean 59% of control, two-tailed t-test p = 0.002, entrectinib mean 47% of control p < 0.0001), an effect not seen in control DMG primary patient comparators (control DMG No. 1 larotrectinib p = 0.80, entrectinib p = 0.30, control DMG No. 2 entrectinib p = 0.18) (Fig. 4E and Supplementary Data Fig. S6) or patient-derived cell cultures (Supplementary Data Fig. S6).

Isolated viability assays from a single patient sample carry several limitations. First, questions remain whether the residual viable NTRK DMG cells could be overcome through a more optimal dosing regimen, whether these reflect a TRK inhibitor resistant population, or whether they represent non-NTRK-containing subpopulation outgrowth. Additionally, viable sample material was insufficient to assess pharmacodynamic or functional consequences of TRK inhibition in these cells. Further clarity could be provided to these dynamics with identification and future studies of additional patient samples. However, this data lends support to the possibility that TRK fusion-positive DMG may likewise exhibit some response to TRK inhibition in clinical practice.

DISCUSSION

The dismal outlook for DMG has unfortunately proven recalcitrant to decades of basic, translational, and clinical research efforts. Greater biologic understanding and improved pharmacologic targeting of oncogenic drivers will be required to provide benefit across the full spectrum of patients diagnosed with this disease. However, increasingly widespread adoption of routine molecular characterization of pediatric brain tumors has enabled the identification of rare but clinically relevant mutational events. As the breadth of therapeutically targetable or prognostically informative genomic alterations continues to expand, so too does the impetus to incorporate these evaluations on a routine clinical basis. Increasingly detailed molecular profiling has been rapidly advancing the risk stratification and clinical management of other pediatric brain tumors, most notably for low-grade gliomas and medulloblastomas (40–43). In contrast, the historical concerns surrounding the safety of stereotactic sampling in pontine DMG has hampered adoption of diagnostic biopsies as standard of care at many institutions. In light of growing data supporting the safety of this procedure in the hands of an experienced neurosurgeon (3, 44), it is imperative that these diagnostic assessments be more routinely offered not simply to expand tissue availability for ongoing research, but to capture the patients for whom genomic data may be clinically informative.

The landscape of preclinical target discovery has provided fertile ground for the expanded implementation of precision medicine trials in pediatric oncology (45, 46). While variable in their successes, targeted agents showing selective, on-target inhibition promise to be transformative in their clinical implementation. The phase 1–2 studies of larotrectinib and entrectinib across TRK fusion-positive cancers demonstrated remarkable antitumor effect across a diverse range of pediatric and adult malignancies (16–18). While not enrolled on the original published phase 1–2 studies, the prevalence of NTRK fusions in pediatric HGGs makes this a promising class of agents for children with brain tumors harboring these fusions. Pharmacokinetic studies from the pediatric phase 1 trial documented meaningful larotrectinib concentrations in patient CSF sampling (16), and case reports have now described the response of a TRK fusion-positive, nonbrainstem pediatric HGG to TRK inhibitor therapy (19). Early data from the current entrectinib phase 1/1b trial have likewise shown similarly promising results, with 1 CR and at least 3 PR reported amongst the 6 pediatric HGGs enrolled on study (47).

Our observations presented here are significantly limited by the small number of individual tumors available for detailed analysis, particularly as this makes it difficult to control for intrapatient heterogeneity. Furthermore, given our current understanding of H3K27M-mutant DMG biology, it is unlikely that TRK inhibitor therapy alone, if effective, would offer a curative potential. The heterogeneity we observe in our sentinel case through immunohistochemical staining, scRNA-seq subpopulations, and ex vivo therapy response may collectively suggest that NTRK fusions mark only a subclone within the larger neoplastic cell population. Comparable characterization of additional patient samples would be necessary to clarify the extent of TRK fusion-positive subclones that may be observed within DMG tumor cell populations. However, the responses observed in clinical trials of other TRK fusion-positive malignancies combined with the ex vivo data presented here would together support that some selective measure of objective clinical response may be achievable, even if only within these TRK fusion-positive subclones. In a disease with no effective medical therapies, any survival or quality of life benefit would prove transformative, even if available to only a small subset of patients.

 

Sequencing through the University of Colorado Genomics Shared Resource is supported in part though the Cancer Center Support (Grant NCI P30 CA046934). Other support is provided by grants through the Morgan Adams Foundation, the Luke Morin Foundation, and the NCI (P01 CA096832 to S.J.B.).

The authors have no duality or conflicts of interest to declare.

Supplementary Data can be found at https://academic.oup.com/jnen.