Establishing a living biobank of pediatric high-grade glioma and ependymoma suitable for cancer pharmacology

Abstract

Background

Brain tumors are the deadliest solid tumors in children and adolescents. Most of these tumors are glial in origin and exhibit strong heterogeneity, hampering the development of effective therapeutic strategies. In the past decades, patient-derived tumor organoids (PDT-O) have emerged as powerful tools for modeling tumoral cell diversity and dynamics, and they could then help define new therapeutic options for pediatric brain tumors.

Methods

Through an integrative approach based on our expertise and a careful review of the literature about glioblastoma 3D primary cultures, we set up a standardized methodological pipeline for the establishment, characterization, and biobanking of PDT-O through direct 3D in vitro culture of the deadliest pediatric glial brain tumors. To assess PDT-O fidelity and validate their preclinical relevance, we performed comprehensive histological, molecular, and drug-response analyses.

Results

Our methodological pipeline allowed the rapid and efficient generation of PDT-O recapitulating their parental tumor features, including intratumoral heterogeneity, even after several passages and cryopreservation/revival as 3D cultures. Moreover, we successfully performed preclinical test responses on these PDT-O to standard-of-care therapies and new therapeutic options. Finally, we identified ONC201 as a selective drug for pediatric glial tumor types not restricted to H3K27-altered glial tumors, as well as combination strategies to increase the therapeutic response to ONC201.

Conclusions

Hence, we describe a fast and robust process to biobank PDT-O for pediatric glial brain tumors. These PDT-O models have the potential for patient-specific modeling even after long-term expansion in vitro, and we established the proof-of-concept of their usefulness to support powerful preclinical studies.

Graphical Abstract

Graphical Abstract.

Key Points.

• Patient-derived tumor organoids (PDT-O) preserve histological, molecular, and clinical features of their parental tumors.

• ONC201 is a potential therapeutic agent for non-H3K27-altered glial tumors.

• PDT-O helped identify drug combination strategies with high clinical translational potential.

Importance of the Study.

Major differences in pediatric and adult brain tumors call for distinct therapeutic approaches. However, the testing of novel therapies remains limited in children for ethics reasons, and the reliance on faithful preclinical cancer models is therefore primordial. Recent advances in 3D culture technologies, such as patient-derived tumor organoids (PDT-O), have revolutionized adult cancer biology, albeit they are scarcely applied to childhood cancers. Hence, based on our expertise and a careful review of the literature, we designed a rapid, efficient, and standardized methodological pipeline to generate PDT-O of the deadliest pediatric glial tumors. One of the essential points to meet the definition of PDT-O is the preservation of the cytoarchitecture and molecular characteristics of the original tumor. Importantly, our PDT-O captured the histological, molecular, and clinical features of their parental tumors even after cryopreservation/revival and could rapidly be deployed to reveal therapeutic vulnerabilities and patient-specific treatment strategies.

Neoplasms of glial origin account for more than 50% of all primary brain tumors in children and adolescents and include 2 particularly deadly cancers in pediatric neuro-oncology, namely high-grade gliomas (pHGG), among which the deadliest diffuse midline gliomas (DMG), and group-A posterior fossa ependymomas (pPFA-EPN).^1,2 These tumors are intrinsically resistant to conventional therapies, and patients often relapse after an initial response.^3,4 The severity of this prognosis calls for a rethinking of preclinical approaches, using the latest methodological innovations.

The majority of pediatric DMG harbor characteristic missense mutations in histone 3-encoding genes that result in a lysine-to-methionine substitution at position 27 (H3K27M). This H3K27M-mutant protein inhibits polycomb repressive complex 2 (PRC2) activity via the sequestration of its catalytic subunit EZH2, resulting in a global decrease in H3K27 trimethylation (H3K27me3), which serves as a molecular surrogate marker.^5–7 Interestingly, decreased H3K27me3 is also a hallmark of pPFA-EPN, exhibiting an overexpression of EZH inhibitory protein (EZHIP) that similarly suppresses PRC2 activity.⁸ On the other side, pHGG restricted to hemispheric locations carries mutations at position 34 in histone variant H3.3, tyrosine kinase gene amplifications, BRAF mutations, or fusion events involving, for example, the NTRK or FGFR genes.⁹ This genetic diversity undoubtedly complexifies the definition of a common therapeutic path.

Besides genetic heterogeneity, it is now well established that tumor cells are highly heterogeneous owing to distinct transcriptional programs. In particular, Neftel et al. described 4 distinct cellular states in H3K27-wild type (wt) HGG: astrocyte (AC)-like, neural progenitor cell (NPC)-like, oligodendrocyte progenitor cell (OPC)-like, and mesenchymal (MES)-like.¹⁰ Moreover, Liu et al. revealed the developmental hierarchical model of cell states in H3K27M-mutant gliomas, with OPC-like cells at the apex, differentiating towards AC-like and oligodendrocyte (OC)-like cells.¹¹ These reminiscent developmental programs could be critical to identify new therapeutic vulnerabilities to reduce disease recurrence. Yet, these analyses remain largely descriptive, and do not allow the integration of cell population dynamics in the prediction of response to treatments, underlying the need for innovative and relevant cancer models.

In recent years, the transposition of the organoid technology to tumors has led to the development of patient-derived tumor organoids (PDT-O) also called tumoroids. These 3D models accurately recapitulate inter- and intratumoral heterogeneity, cell-state dynamics, and treatment response patterns of their tumors of origin,¹² as exemplified recently for adult gliomas.^13,14 This breakthrough in adult glioma modeling held great promise for its application in the pediatric field, although adaptations are necessary to take account of the biology of these tumors. Here, we present a rapid and robust strategy for pHGG and pEPN-PFA PDT-O derivation. These models can be stably cultured in 3D for months, and even years, and can be efficiently cryopreserved/reanimated. Moreover, by performing comprehensive histological, molecular, and functional analyses, we showed that PDT-O capture the clinical phenotypes of their respective parental tumors over time. We further evaluated the preclinical potential of our models by demonstrating that they can be deployed to reveal patient-specific therapeutic vulnerabilities.

Material and Methods

All Methods details can be found in supplement.

Human Specimens, Derivation, and Culture of PDT-O

Human biological samples and associated data were obtained from Hospices Civils de Lyon Biobank (CRB-HCL, BB-0033-00046). Samples were collected in the context of patient diagnosis. The Biological Resource Center (BRC) of the Centre Léon Bérard (n° BB-0033-00050) biological material collection and retention activity are declared to the Ministry of Research (DC-2008-99 and AC-2019-3426). The study had all necessary regulatory approvals, and informed consents are available for all patients. For PDT-O derivation, tissues were minced into small pieces and digested at 37 °C with gentle vortex for 30–90 min. Then, PDT-O were established in 96-well plates (20 000 cells/well) using appropriate culture medium for pHGG and pEPN-PFA-derived organoids (pHGG-O and pEPN-PFA-O) (see Supplementary Material for details). Medium was changed twice a week and PDT-O were split every 1 to 2 weeks when reaching a diameter of 800–1000 µm. The average doubling time of PDT-O was estimated by plotting growth curves and applying the formula: doubling time = ln(2)/growth-rate. All cultures were tested monthly for mycoplasma.

Histological Analyses

PDT-O were fixed, processed, and stained as described before.¹⁵ The diagnosis of each case and matching of PDT-O with their tumors of origin were confirmed by an independent histopathologist.

RNA Sequencing

Libraries were prepared from 100 to 1000 ng RNA/sample with Illumina Stranded mRNA preparation kit. Sequencing was performed on an Illumina NovaSeq6000 platform. Quality control of reads was followed by trimming of Illumina adapter sequences and mapping of the reads to the GRCh38 human with Ensembl annotations. Gene counts were computed and converted into a DESeq2 object. To remove unwanted contamination by cells from the microenvironment, we excluded genes for which the expression difference between tumor tissues and PDT-O was above the 9th decile and, to confirm normal cell transcriptome exclusion, enrichment analyses were run on these genes (Supplementary Figure S2A).

Targeted DNA Sequencing

Libraries were prepared from 200 ng gDNA/sample using the SureSelect XT low input library preparation kit. Sequencing was performed on an Illumina NovaSeq6000 platform. After quality control of reads, Unique Molecular Identifiers were extracted from reads, which were then mapped separately to the GRCh38 human genome for each sequencing lane. GATK4 Best-Practices workflow Somatic short variant discovery was used to call somatic mutations. We selected variants with expected consequences at proteic level and variant allele frequency (VAF) >5%.

Methylation Data Generation, Processing, and Visualization

DNA was hybridized to Illumina Infinium Human Methylation EPIC BeadChip (850K) arrays. Normalization was performed using functional normalization with default parameters. Filtering was performed to remove (i) poor-performing probes, (ii) probes located on sex chromosomes, (iii) probes known to have common single nucleotide polymorphisms (SNPs) at the CpG site using dropLociWithSnps function, and (iv) cross-reactive probes.

RiboMethseq

RiboMethSeq was performed using Illumina sequencing as previously described.¹⁶ Sequencing data were processed using the ribomethseq pipeline while Cscore calculation, and graphical representations were obtained using the rRMSAnalyzer R.

ddPCR

Droplet digital PCR was performed according to the manufacturer’s protocol on genomic DNA by using a QX200 automatic droplet generator and a thermal cycler with appropriate primers.

Single-Cell RNA Sequencing of PDT-O

PDT-O were dissociated, filtered, and sorted for viable cells. Then, cells were loaded on a 10× Genomic chip and run on the Chromium Controller system to target 10 000 cells per sample. Gene expression data were generated with the Chromium Single Cell 5’ assay and sequenced on the NovaSeq 6000 platform. To generate gene-barcode count matrices, raw sequencing reads were preprocessed using the “cellranger count” pipeline and human GRCh38 as reference. The rest of the analysis was conducted in R using Seurat for data processing and analysis. Clusters were identified with the Louvain algorithm; UMAPs were computed using the first 20 dimensions of the PCA. Cell cycle phase was assessed using CellCycleScoring with the default gene sets provided by Seurat. Cells were scored with the AddModuleScore function of Seurat, using the gene lists provided in Supplementary Table S5.

Western Blot

PDT-O were lysed in RIPA buffer, and 15 µg of protein lysates were used for blotting. Membranes were incubated overnight with primary antibodies at 4 °C. Samples were detected and imaged on the ChemiDoc Touch Imaging System.

Drug Assays

Drugs and doses selection.—

For all drugs, at least 4 doses were chosen based on the maximal drug concentration in patients’ plasma (Cmax), so that the range covers the Cmax concentration and at least 3 ₁₀log (Supplementary Table S6).

PDT-O seeding and treatment.—

PDT-O were dissociated and seeded at 4000 cells/well in 96-well black plates. They were allowed to form for 4 days and were then treated with appropriate dilutions of drugs. At least 3 independent tumoroids (technical replicates) from 3 different passages (biological replicates) were used except if otherwise specified.

Dose response, absolute IC50/AUC determination, and drug combination assay.—

Viability was measured after 72 h of treatment using a CellTiter-Glo Assay in accordance with the manufacturer’s instructions. Data were normalized with negative (DMSO only) and positive (gambogic acid) control wells. Area under the curves (AUCs), dose–response curves, absolute IC₅₀, box-plots, and heatmaps were obtained using Prism (GraphPad) and R (v.4.2.0). For drug combination assays, analysis and plot visualization were performed with the Combenefit software using the Loewe mathematical model and the matrix representation.^17,18

Phenotypic combination assay.—

PDT-O were stained using the LIVE/DEAD kit in accordance with the manufacturer’s instructions, and were imaged using EVOS M7000 Imaging System. LIVE/DEAD stainings were quantified by measuring the surface area of live and dead cells using ImageJ software.

In Vivo Evaluation of ONC201 Efficacy

Orthotopic engraftment of dHGG9-O into immunodeficient mice was carried out as previously described.¹⁹ Mice were allowed to recover for 10 days before intraperitoneal injection of PBS (vehicle) or ONC201 at 25 mg/kg twice a week, for 4 weeks. After animal sacrifice, brains were dissected and fixed, and immunohistochemistry stainings on brain sections were quantified using HALO AI software.

Statistical Analyses

Experiments were performed using at least 3 technical replicates. Data variation was expressed as mean ± SD. Statistical analyses were performed using R for -omics studies, GraphPad Prism for IC₅₀/AUC determination, phenotypic combination assays, and in vivo experiment, and Combenefit software for drug combination assays.

Results

3D pHGG-Derived Organoids Retain the Characteristics of Their Parental Tumors

Glioma in vitro culture models include 2D cell lines, neurospheres, and matrigel-embedded tumoroids.²⁰ They have contributed tremendously to our current understanding of biological/molecular mechanisms underlying pediatric brain tumors’ pathogenesis. However, they have some limitations in a perspective of precision/personalized medicine including (i) the use of serum and/or extracellular matrix, which are not favorable to maintain tumoral cells’ heterogeneity and proliferation, respectively,^13,21–23 and/or (ii) the lack of systematic comparison to the parental tumors, thus preventing the evaluation of the degree to which these in vitro systems recapitulate the characteristics of their tumors of origin, especially over time.

Hence, we designed a rapid and reliable integrated process to generate, expand, and characterize in depth 3D pHGG-O from fresh tissues to reach applicability for further functional precision/personalized medicine assays (Supplementary Figure S1A). By using (i) a specific enzyme mixture with low toxicity to ensure high cell viability and yield from very small tissue fragments such as fine-needle biopsies, (ii) a rationalized and standardized culture medium defined thanks to our expertise and careful review of the literature about in vitro glioblastoma primary cultures, neurospheres, and tumoroids (Supplementary Table S1),^13,21–23 and (iii) a size-adjusted expansion procedure, we obtained 3D PDT-O from various primary and relapsed pHGG histological subtypes and tumor grades (WHO grades 3–4) (Supplementary Table S2), as early as 3 to 7 days post culture initiation. pHGG-O were split every 7 to 10 days (800 µm diameter), at a 1:2 to 1:4 ratio (doubling time ∼75 h, Supplementary Figure S1B), for at least 1 year. The 3D models obtained were biobanked and reanimated for further molecular, cellular, and preclinical analyses, with a success rate of 91% (N = 10/11 tumors) (Supplementary Tables S1 and S2).

One of the essential points to meet the definition of PDT-O is the preservation of the cytoarchitecture and molecular characteristics of the original tumor sample. Then, we first characterized the pHGG-O at histological level after 2 to 3 months in culture or after cryopreservation/revival (Figure 1, Supplementary Figure S1C). This phenotypic analysis revealed that pHGG-O preserved the histological features of their primary tumor, including the expression of markers of proliferation and differentiation observed in their parental tumors, as assessed by Ki67, GFAP, OLIG2, SOXs, and Vimentin stainings (Figure 1). This remained true even after several months in culture, or following cryopreservation/revival, indicating their high relevance for subsequent preclinical testing (Supplementary Figure S1C).

Figure 1.

pHGG-derived tumor organoids preserve histologic features of parental tumor subtypes. Representative hematoxylin-phloxine-saffron (HPS) and immunohistochemistry images of dHGG1-O, dHGG1R-O, DMG1-O, and their tumors of origin. Key clinical marker proteins routinely used for pHGG diagnosis, that is, Ki67, GFAP, OLIG2, SOX2, SOX10, and VIM, were stained. Abbreviations: -O, tumor organoid; -T, tumor. Black scale bar: 100 µm; red scale bar: 25 µm, identical for all images. For tissues, one piece of 0.5 cm² was used for this characterization; for PDT-O, 8 to 18 spheres were fixed between passages P3 and P6, depending on the model.

Next, we performed extensive multiomics characterization to assess the molecular fidelity of PDT-O to their tumors of origin over time. Pearson’s correlation heatmaps established from transcriptomic profiling unveiled that pHGG-O models were grouped with their respective tumors of origin even after long-term culture and cryopreservation/revival (Figure 2A, Supplementary Figure S2A, Supplementary Table S3). Targeted DNA sequencing and VAF confirmed the broad concordance between mutational patterns in pHGG-O and their respective parental tumor samples, regardless of culture time (Figure 2B, Supplementary Table S4). This observation was still true after filtering for deleterious cancer-related somatic variants only, indicating that pHGG-O retained major clonal and subclonal populations even after cryopreservation (Supplementary Figure S2B and C, Supplementary Table S4). For example, and as expected from the pHGG mutational spectrum and patient clinical reports, all DMG-O lines harbored H3.3 mutations, and ndHGG3-O, dHGG4-O, DMG2-O, and DMG4-O cultures are characterized by patient-specific missense mutations in TP53 (Supplementary Figure S2B, Supplementary Tables S2 and S4). Moreover, we found a prototypical missense mutation in BRAF (V600E) in the TP53-mutated pleiomorphic xanthoastrocytoma ndHGG3 tissue/tumoroid pair (Supplementary Figure S2C, Supplementary Table S4). Last and consistently with recent reports,²⁴ the only DMG patient with an overall survival superior to 36 months was TP53-wt and presented a specific mutation in the PTPN11 gene of the MAPK pathway, conserved in the paired tumoroid line (Supplementary Table S4). PDT-O also preserve the DNA methylation profiles of their corresponding tissues even after cryopreservation/revival, as shown on the UMAP, with an independent clustering notably of DMG and ndHGG3 BRAF V600E samples (Supplementary Figure S2D). Finally, using the RiboMethSeq approach,¹⁶ we showed that pHGG-O conserved the rRNA 2’Ome pattern of their parental tumors (Supplementary Figure S2E). Moreover, we observed that the rRNA 2’Ome 18S_Am576 site is hypermethylated in H3K27-wt but not in H3K27M-mutant tumors (Supplementary Figure S2F). Thus, even though the limited number of cases tested prevent us from drawing definitive conclusion, rRNA 2’Ome could become an additional marker to discriminate between different pHGG molecular subtypes.¹⁶

Figure 2.

pHGG-derived tumor organoids preserve multiomics features of parental tumor subtypes. (A) Heatmap of Pearson correlation coefficients between all samples based on normalized transcriptomic data (gene expression). The correlation coefficients were calculated on all genes, except for the 10% with the highest average expression in tissue, to exclude the immune and healthy components due to normal cell contamination. The hierarchical clustering was based on the Euclidean distance. (B) Heatmap of Pearson correlation coefficients between all samples based on genomic data. Correlation coefficients were calculated using the variant allele frequency, and the hierarchical clustering was based on the Euclidean distance. (A, B) Abbreviations: -O, tumor organoid; -T, tumor. Tissue samples are annotated “1” or “2” as an indication of the number of biological replicates and PDT-O samples are annotated with passage number (P). PDT-O were further annotated with “*” when submitted to a cryopreservation/thawing cycle.

Collectively, these data demonstrate that our organoid-based approach provides a simple and efficient workflow for the generation and biobanking of PDT-O established and propagated in 3D directly from pediatric glioma tissues. These models can be stably grown for months to years and cryopreserved, while fully preserving the molecular profiles of their original tumors. Thus, they could overcome the limitation of scarcity/small size of samples in pediatric oncology and help apprehend, at different scales, new therapeutic vulnerabilities.

pHGG-O Maintain pHGG Tumoral Cell-Type Heterogeneity

Intratumoral heterogeneity is widely considered a key driver of resistance to treatment.²⁵ Preservation of the recently described pHGG tumor hierarchy^10,11 in PDT-O is thus a prerequisite for any prospective preclinical model. To investigate the functional identity and hierarchical relationships of HGG tumor cells in these models, we performed droplet-based scRNA-seq of one H3K27M-mutant and one H3K27-wt PDT-O and we compared these data with publicly available data from biopsies.^10,11,26 First, using an unsupervised Leiden algorithm and UMAP, we identified different cell clusters in our models, reflecting the diversity of cell states observed in patients (Supplementary Figure S3A, Supplementary Table S5).^10,11 We first focused on cell-cycle stages and observed cycling and quiescent tumor cell populations in both PDT-O, as shown on UMAPs presenting dedicated module scores^27,28 and the expression of the proliferation marker MKI67 (Supplementary Figure S3B–D). Interestingly, as described in pHGG tumors,^10,29 some of these quiescent cells are characterized by a glial cancer stem cell signature²⁷ (Supplementary Figure S3E).

Moreover, in line with the differentiation trajectories recently reported in primary tumors,¹¹ we identified neoplastic populations of OPC-, OC-, AC, and MES-like cells in DMG1-O (Figure 3A, Supplementary Figure S3F, Supplementary Table S5). Consistently with tumors profiling, OPC-like cells were further resolved into 3 subpopulations, (i) OPC-like-1/OC-like cells defined by high expression of PDGFRA and OLIG1, (ii) OPC-like-3 cells depicted by a higher expression of marker genes linked to more immature oligodendrocyte precursors of the developing brain (eg, DLL1), and (iii) OPC-like-2 cells highly expressing genes encoding ribosomal proteins such as RPL17¹¹ (Supplementary Figure S3F and G).

Figure 3.

pHGG-O preserve intratumoral heterogeneity and cellular hierarchies of gliomas. (A) UMAPs of DMG1-O depicting module scores of OPC- (shared and variable), OC-, AC-, and MES-like cell programs,^11,26 color-coded with a blue (low score) to red (high score) gradient. (B) UMAPs of dHGG1R-O depicting module scores of NPC1/2-, OPC-, AC-, and MES2-like cell programs,¹⁰ color-coded with a blue (low score) to red (high score) gradient.

Similarly, dHGG1R-O exhibited multiple populations of neoplastic cells with distinct transcriptomic features, including the classical AC-, OPC-, and NPC-like cells¹⁰ (Figure 3B, Supplementary Figure S3H, Supplementary Table S5). Moreover, the NPC-like cells can be subdivided into 2 subprograms also described in tumor tissues, termed NPC1- and NPC2-like cells, and distinguished, respectively, by the inclusion of OPC-related genes (eg, OLIG1) versus neuronal lineage genes (eg, TAGLN3)¹¹ (Figure 3B, Supplementary Figure S3H and I). Finally, we identified a mesenchymal tumor cell population also described in patients (MES2-like), and distinguished by an elevated expression of mesenchymal- (eg, VIM), hypoxia-response- (eg, HILPDA), and glycolytic- (eg, ENO2) related genes¹⁰ (Figure 3B, Supplementary Figure S3H and I).

Thus, pHGG-O recapitulate the recently described intratumoral heterogeneity of their parental tumors and could then help identifying cell-state-specific targetable vulnerabilities and designing drug combinations.

pHGG-O Drug Response is Correlated With Patient Clinical Response

The predictive value of PDT-O has been reported in cancer therapy and paves the way to personalized medicine approaches in several malignancies.^30,31 Following the phenotypic and molecular validation of our models, we performed dose titration assays to examine how well they mirrored drug responses in patients. For example, patient dHGG1 relapsed after 8 months of Larotrectinib, a first-in-class TRK inhibitor³² and we successfully established PDT-O for this patient before (dHGG1-O) and after (dHGG1R-O) relapse (Figure 4A, Supplementary Table S2). While dHGG1-O responded with remarkable sensitivity to Larotrectinib, as did the patient, who presented a dramatic decrease in the tumor size after 108 days of treatment,³² dHGG1R-O displayed an 80 times higher IC₅₀ (Figure 4B). Consistently, a missense mutation in NTRK2 (G639R) was found in the Larotrectinib-treated dHGG1 tissue/tumoroid pair at relapse (dHGG1R), which is absent in the initial tissue/tumoroid pair (dHGG1) (Supplementary Figure S2C).

Figure 4.

PDT-O are predictive of clinical responses in corresponding patients. (A) Timelines of patient clinical history. Treatment regimens of hemispheric H3K27-wt pHGG (left) and DMG (right) patients with major molecular alterations, status (alive/dead), timing of PDT-O derivation, and overall survival from diagnosis to last known status (September 2024). Abbreviations: -O, tumor organoid; -T, tumor; TMZ, Temozolomide; RT, radiotherapy; MTX, Methotrexate; PARPi, PARP inhibitor; ATRi, ATR inhibitor; mth, month; yr: year. (B–D) Drug dose–response curves. Representative results of 3 independent experiments. Red-dotted lines represent the maximal plasmatic concentrations (Cmax, in ₁₀log nM or µM). Abbreviations: -O, tumor organoid; -T, tumor. (B) NTRK-inhibitor Larotrectinib (from 0.57 to 18 750 nM) in dHGG1-O and dHGG1R-O. (C) mTOR inhibitor Everolimus (from 0.0001 to 1 µM) in DMG1-O, DMG2-O, and DMG3-O. (D) ONC201 (from 0.01 to 100 µM) in DMG4-O and DMG4R-O.

We next assessed the dose–response profiles of DMG-O lines derived from patients treated with the mTOR inhibitor, Everolimus. Overall, PDT-O lines showed a very low sensitivity to Everolimus that correlates with patients’ clinical responses (Figure 4C). Yet, we observed that the DMG1-O line, derived from the tumor of the only patient who exhibited partial disease stabilization following mTOR inhibitor administration, responded significantly more to Everolimus than DMG2- and DMG3-O lines (Figure 4A and C, Supplementary Figure S4A).

Similarly, the DMG4 patient was first treated with radiotherapy and ONC201, but the tumor continued growing under treatment. We derived PDT-O lines at diagnosis (DMG4-O) and during the ONC201 maintenance therapy (DMG4R-O) for this patient. Interestingly, those 2 PDT-O lines were overall resistant to ONC201, with >40% of tumor cells alive after a treatment 10 times higher than the maximal plasma concentration (Cmax, see Methods) (Figure 4D).

Then, our data suggest that pHGG-O could become useful predictive biomarkers for treatment response in pediatric patients.

PDT-O Allow the Identification of Patient-Specific Drug Sensitivities, Including ONC201 as a Potential Therapeutic Agent for H3K27-altered and -wt High-Grade Glial Tumors

Considering the potential predictive value of pHGG-O, we decided to perform drug testing on our models to evaluate their ability to identify patient-specific sensitivities. We identified 13 drugs with a strong potential for direct pHGG clinical translation (Supplementary Figure S4B and C, Supplementary Table S6). These included 12 FDA-approved antineoplastic drugs, and JQ1, as the archetypal BET inhibitor.³³ Using this drug library, we performed a cell viability assay on 5 H3K27-wt pHGG-O and 4 DMG-O lines. The landscape of treatment responses across the different PDT-O lines and drugs is shown in Figure 5A and Supplementary Figure S4D. We found sensitivity to a limited number of drugs within the context of clinically achievable serum levels (IC₅₀ < Cmax), with 5/13 compounds demonstrating activity against more than half of the PDT-O lines (Figure 5B). Interestingly, we observed patient-specific drug sensitivities even among a given subtype. For example, DMG1-O appeared globally more sensitive to our panel of drugs than other DMG-O lines, whereas ndHGG3-O, derived from a pleiomorphic xanthoastrocytoma resection, was the most resistant PDT-O line (Figure 5A, Supplementary Figure S4D). Interestingly, DMG tumors harboring PIK3CA mutations are described to be more sensitive to ONC201³⁴ and DMG1-O was indeed both the only PDT-O line bearing this mutation and the most sensitive to ONC201 (Figure 5A, Supplementary Figure S4D, Supplementary Table S4). Unexpectedly, ONC201 showed some beneficial effects in all models, with a strong cytotoxic effect also in some H3K27-wt pHGG-O, being then irrespective of the H3K27M-status of the PDT-O lines (Figure 5A–C, Supplementary Figure S4D, Supplementary Table S2). Moreover, in mice orthotopically xenografted with the H3K27-wt dHGG9-O, we observed a decrease in Ki67 staining in ONC201-treated animals (Supplementary Figure S5A and B), as described by others for H3K27M-mutant DMG.³⁵

Figure 5.

PDT-O are eligible for screenings for a personalized selection of effective drugs. (A) Heatmap of the AUCs calculated from N = 3 independent experiments for each drug at the concentrations presented in Supplementary Table S6, for all H3K27-wt pHGG-O and DMG-O models. AUCs are color-coded with a blue (high AUC, low drug efficiency) to red (low AUC, high drug efficiency) gradient. (B) Drug screen “hits” within the context of clinically achievable plasmatic drug levels. Box-plot of IC₅₀ absolute values (in ₁₀log µM) calculated from N = 3 independent experiments for all drugs in each PDT-O line, including the number of samples reaching 50% of viability in the range of tested concentrations. Red lines represent the maximal plasmatic concentrations (Cmax, in ₁₀log µM). (C) Scatter plot of the AUCs for ONC201 in all pHGG-O and DMG-O models based on H3K27M status: H3K27-wild-type (H3K27-wt, blue) or H3K27M-mutant (H3K27M-mut, red). P-value was determined using a Mann–Whitney test (2-tailed). (D) Representative hematoxylin-phloxine-saffron (HPS), hematoxylin-eosin (HE), and immunohistochemistry images of EPN1-O, EPN2-O, and their tumors of origin (EPN1-T, EPN2-T). Key clinical marker proteins routinely used for pEPN-PFA diagnosis, that is, GFAP, EMA, and OLIG2, were stained. Black scale bar: 300 µm; red scale bar: 50 µm. For tissues, one piece of 0.5 cm² was used for this characterization; for PDT-O, 4 to 5 spheres were fixed at passages P36* (submitted to a cryopreservation/thawing cycle; EPN1-O) and P4 (EPN2-O). (E) Heatmap of the AUCs calculated from N = 3 independent experiments for each drug at the concentrations presented in Supplementary Table S6, for all pPFA-EPN-O samples. AUCs are color-coded with a blue (high AUC, low drug efficiency) to red (low AUC, high drug efficiency) gradient.

We herein wondered whether ONC201 could also be a treatment option for patients with pPFA-EPN, which present similar epigenetic remodeling without H3K27M mutation.³⁶ EPN are difficult to propagate in vitro as 3D PDT-O models, as they present a very limited capacity to expand.¹⁹ Following the same methodological pipeline than for pHGG-O, we set up a robust protocol that allows the long-term expansion and efficient cryopreservation/revival of pPFA-EPN-derived organoids (EPN-O), notably by taking into account their glucose addiction³⁷ (Supplementary Figure S4E). EPN-O were obtained as early as 1 to 3 days post culture-initiation, reached their maximum size after the initial 3–5 weeks of culture, and were then routinely split every 12–15 days (800 µm diameter) at a 1:2 ratio (doubling time ∼140 h, Supplementary Figure S1B), for more than 1 year. The 3D models were biobanked and reanimated for further characterization and preclinical analyses (Figure 5D and E, Supplementary Figure S4F and G), with a success rate of 100% for not-treated fresh and DMSO-frozen resections (N = 2/2 tumors) (Supplementary Tables S1 and S2). EPN-O preserve (i) the histology of their tumors of origin, including the expression of differentiation markers (GFAP, EMA, and OLIG2) and (ii) the loss of histone mark, even after a long time in culture or growth following cryopreservation (Figure 5D, Supplementary Figure S4F). We then performed a drug sensitivity testing using the same 13-drug library, except for Ceralasertib that was swapped for Vincristine, a chemotherapeutic agent used for pPFA-EPN patients. The epidrug Panobinostat showed the strongest cytotoxic effect at Cmax but ONC201 was also one of the most efficient drugs of the library (Figure 5E, Supplementary Figure S4G).

PDT-O models preserve the overall resistance to treatment observed in patients, but they highlighted ONC201 as a selective drug for a broad range of pediatric glial tumor types not restricted to H3K27M-mutant glioma.

pHGG-O Can Be Used to Identify Relevant Combination Treatments

The complex and heterogeneous nature of brain tumors renders monotherapeutic approaches unlikely to promote long-term survival,³⁸ as exemplified with the transient benefits of ONC201.^35,39 We thus used a PDT-O-based approach to identify synergistic combinations to further enhance the effects of ONC201 in pediatric glial tumors.

We first examined the synergistic antitumor effects of ONC201 and the MEK inhibitor Trametinib. We selected 3 DMG-O and 2 H3K27-wt pHGG-O with an intermediate to low sensitivity to both drugs (Figure 5A, Supplementary Figure S4D), and treated them with increasing concentrations of ONC201 or Trametinib, or both. Analyses based on Combenefit software and the Loewe additivity model^17,18 revealed drug additivity in DMG-O and H3K27-wt pHGG-O models even at physiological concentrations (Figure 6A and B). Using a viability/cytotoxicity phenotypic assay, we confirmed that a strong synergistic effect occurred also in models sensitive to ONC201 (Figure 6C and D). We used the same methodologies to assess the ONC201/PTC-596 combination effects on 3 H3K27-wt pHGG-O and 4 DMG-O and also emphasized drug additivity and synergy of both drugs (Supplementary Figure S6A–D). Interestingly, when comparing these synergistic effects with the one already described between ONC201/Paxalisib by others,³⁴ we observed that the combination ONC201/Trametinib was more potent for the infant-type hemispheric glioma (dHGG1/1R-O) and also for one of the DMG patients (DMG2-O) (Figure 6C and D). The DMG2-O line was also more sensitive to ONC201/PTC-596 than ONC201/Paxalisib (Supplementary Figure S6C), and ONC201 efficacy was potentiated by the PTC-596 in the pleiomorphic xanthoastrocytoma super-resistant PDT-O line (ndHGG3) (Supplementary Figure S6D).

Figure 6.

Trametinib is synergistic with ONC201 across pHGG-O models. (A, B) Quantification of ONC201 and Trametinib combination effects in terms of synergy, additivity, and/or antagonism in selected PDT-O lines. Combination effect (synergy, additivity, or antagonism) between both drugs was determined on 3 independent tumoroids using Combenefit software and Loewe synergy and antagonism analysis. Scores < −5 indicate an antagonist effect, scores between −5 and 5 indicate an additive effect, scores between 5 and 10 indicate a low synergistic effect, and scores >10 indicate a synergistic effect. *P-value < .05, **P-value < .001, ***P-value < .0001. (A) DMG-O models (DMG2-O, DMG3-O, and DMG4-O) and (B) H3K27-wt pHGG-O models (dHGG1-O and ndHGG3-O). (C, D) Quantification of live/dead immunofluorescence stainings of selected PDT-O lines, treated with DMSO only (CTRL), ONC201 (10 µM), Trametinib (TRAM; 20 nM), ONC201 and Trametinib (ONC201 + TRAM; 10 µM and 20 nM, respectively) or ONC201 and Paxalisib (ONC201 + PAXA; 10 µM and 0.5 µM, respectively). The ONC201 and Paxalisib combination was used as a synergy control, as demonstrated by Jackson et al.³⁴ Quantification was performed by measuring the ratio of the pixel surface area of dead cells to the one of live cells using the ImageJ software on 3 to 5 independent tumoroids per condition. P-values were determined using Mann–Whitney tests (2-tailed). (C) DMG-O models (DMG1-O, DMG2-O, and DMG3-O) were treated for 72 h (DMG1-O and DMG3-O) or 96 h (DMG2-O). (D) H3K27-wt pHGG-O (dHGG1-O and dHGG1R-O) models were treated for 72 h.

Together, these results demonstrate that ONC201 in combination with Trametinib or PTC-596 synergistically inhibits the growth of pHGG-O regardless of their H3K27M-status and of the efficacy of ONC201 alone.

Discussion

Pediatric brain tumors definitely lack new efficient curative treatment options. Here, we report an efficient and standardized pipeline to rapidly generate and characterize in depth PDT-O from different types of the deadliest pediatric glial tumors. By performing comprehensive histological and molecular analyses, we showed that these models retain key histological and molecular features of their corresponding parental tumors, and the wide range of pHGG cell states found in the literature.^10,11 Moreover, we demonstrated that our models are both expandable in 3D for months to years and stable over time even after cryopreservation, allowing for the establishment of a robust close-to-native living tumor biocollection for precision medicine. Along that line, even though they need further validation to apprehend pHGG heterogeneity, our data suggest that pHGG-O could become useful predictive biomarkers for treatment response in pediatric patients.

Considering the strengths/weaknesses of each model is critical to selecting the most appropriate one for a particular investigation. Approaches to drug testing often include patient-derived xenografts (PDX) and cell lines, and/or PDT-O models when available. Moreover, to circumvent the lengthy process of model development, scalable platforms exploiting fresh uncultured patient cells for ex vivo personalized drug testing have been proposed.^40,41 Although these platforms meet the challenges of rapid/efficient generation of patient-specific drug sensitivity profiles, they consume the entire biological sample without expansion of the tumoral cells, which limits the overall number of drugs/combinations tested and prevents the constitution of biocollections for research purposes. It also hinders the precise and detailed characterization of the molecular and functional characteristics of cells resistant to treatment. Thus, given their high derivation and expansion efficiency and their potential for drug testing, PDT-O represent an advantageous compromise, combining the possibility of rapid evaluation of patient-specific therapeutic combinations without undermining the potential for future research and preclinical evaluation of drug discovery.

Meanwhile, PDT-O systems are highly complementary to traditional in vivo systems. Transgenic murine models are vital to understand how oncogenic mutations contribute to tumor initiation/progression; their foremost limitation is their inability to recapitulate the complex genetic and phenotypic heterogeneity/cellular hierarchy observed in patient tumors. On the other side, PDX are currently the gold standard for pHGG modeling because they accurately recapitulate patient tumors including histological markers and invasiveness.⁴² Beyond institutional and societal demands to limit the use of animal models, the major drawbacks of PDX are the time, cost, and scalability, as well as the recent evidence suggesting that extended PDX culture favors mouse-specific tumor evolution.⁴³ In that context, PDT-O with their high ease of handling and potential for patient-specific modeling, emerge as advantageous and robust research tools. In the future, comparing PDT-O to PDX and PDX-derived organoid chemograms, and correlating the data to patient response could help further define PDT-O’s potential for precision/personalized medicine.

Finally, by recapitulating tumoral cell characteristics including in their microenvironment thanks to their integration in sophisticated co-culture systems, PDT-O offer a possibility to spawn new drug discovery efforts. Indeed, PDT-O are also amenable to co-culturing with normal brain-organoids and immune cells in various extracellular matrices, such as the GLICO model that combines healthy human cerebral organoids with patient-derived glioma stem cells,⁴⁴ or the recently described 3D-pediatric glioma avatar embedding low-passage primary glioma cells in Matrigel,²³ which could facilitate addressing questions regarding the role of a complex microenvironment on pediatric brain tumors biology.

A solid preclinical pipeline compatible with the expectations of precision/personalized medicine could therefore be based on amplification of tumor material via the PDT-O models, and on their use in the first line to select drug combinations capable of crossing the blood-brain barrier (BBB) and presenting the greatest antitumor efficacy depending on the patient; these PDT-O could then be grafted in mouse models, to validate the optimal therapeutic regimen and precise the physiologically relevant drug doses as recently exemplified for ONC201.³⁹

Because they finely reproduce their tumors of origin, PDT-O can also be used to understand and anticipate the genetic and nongenetic mechanisms involved in treatment resistance. For example, TRK fusion-positive cancers can develop resistance to TRK inhibition, through NTRK kinase domain mutations.⁴⁵ In line with this, we found a missense mutation in NTRK2 (G639R) in the Larotrectinib-resistant dHGG1R tissue, which was preserved in the corresponding PDT-O. We observed that 60%–80% of the tumoral cells present neither the G639R mutation in dHGG1R-O nor other known mutations of resistance to Larotrectinib, suggesting additional nongenetic mechanisms of resistance and placing our PDT-O models as a powerful complementary tool to classical molecular analysis used to orientate treatment in clinic. Conversely, PDT-O can be used to identify therapeutic vulnerabilities. Thus, we demonstrated that ONC201 inhibits the growth and induces cell death of glial brain tumors regardless of their loss of H3K27me3.

Although promising, ONC201 benefits remain time-limited in clinic.³⁴ Here, we identified 2 drug combinations with high clinical translation potential, involving the concomitant use of ONC201 with Trametinib or PTC-596. Lim et al. also recently identified such a synergy between ONC201 and Trametinib in breast cancer models, resulting from mitochondrial apoptosis induction.⁴⁶ PTC-596 is a small compound able to cross the BBB and to induce BMI1 proteosomal degradation.^47,48 Interestingly, BMI1 is overexpressed in ~50% of pediatric brain tumors including pHGG and EPN, and its silencing is sufficient to eliminate the tumor-forming capacity of pHGG stem cells.⁴⁹ Moreover, BMI1 and the PI3K/Akt pathway are described to be coactivated in a substantial fraction of HGG tumors, with Akt pathway enhancing BMI1 oncogenic potential.⁵⁰ Interestingly, Dun and colleagues recently highlighted the utility of a combination treatment strategy using ONC201 and the PI3K/Akt inhibitor Paxalisib,³⁴ suggesting that BMI1-PI3K/Akt could constitute an oncogenic node worth targeting to improve ONC201 efficiency.

Thus, because it allows to derive PDT-O from a wide range of pediatric glial brain tumors within few days, while finely preserving the characteristics of their parental tumor over time, the methodology presented here constitutes a timely and standardized platform for envisioning basic and clinical research based on the optimization of therapeutic combinations and the transition to precise medicine strategies, an essential paradigm shift to improve the therapeutic management of pHGG and pPFA-EPN.

Supplementary material

Supplementary material is available online at Neuro-Oncology (https://academic.oup.com/neuro-oncology).

Conflict of interest statement

The authors declare no competing interests.

Funding

This research was supported by the Foundation ARCECI innovation award (USA) awarded to L.B., the SFCE (Société Française de Lutte contre les Cancers et les Leucémies de l’Enfant et de l’Adolescent), the CLARA (Cancéropôle Lyon Auvergne Rhône-Alpes), the Ligue Contre le Cancer (AAPEAC2022.LCC), the DevWeCan2 LabEx (ANR-10-LABX-0061), the Convergence Institute Plascan (ANR-17-CONV-0002), MSD avenir (ERiCAN) and charities: “Mia Neri Foundation,” “Imagine for Margo,” “Nathan Graine de soleil,” “Wonder Augustine,” “Ligue Départementale contre le Cancer de Haute-Savoie” and “Les Etoiles Filantes.” A.T. received financial support from the Ligue Nationale Contre le Cancer and Fondation ARC. F.B. received financial support from “Mia Neri Foundation.” S.M. is supported by a private donation from T. family.

Authorship statement

C.D. and A.T.: conceptualization, methodology, data curation, investigation, formal analysis, validation, visualization, and writing—review & editing. F.B., S.M., T.D., H.P., and C.S.: methodology, data curation, software, investigation, formal analysis, validation, and visualization. A.C., M.H., L.C., and C.Bo.: methodology, investigation, formal analysis, validation, and visualization. L.T.: resources, supervision, and validation. M.G., C.C., C.Be., J.E., J.T., N.G., V.A., and M.Car.: methodology, data curation, investigation, and validation. V.M.: resources, formal analysis, validation, and writing—review & editing. I.R. and M.Ba.: resources and data curation. M.Bi., A.D, M.C.-B., and J.-Y.B.: writing—review & editing. P.-A.B., A.S., and F.D.R.: resources. A.V. and D.M.: resources, formal analysis, and validation. M.Cas. and P.L.: conceptualization, writing—review & editing, and funding acquisition. L.B.: conceptualization, supervision, methodology, data curation, investigation, formal analysis, validation, visualization, writing—original draft, project administration, and funding acquisition.

Data availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Laura Broutier (laura.broutier@lyon.unicancer.fr). All -omic normalized data sets and data analyses’ codes will be made available via the open-access Share4Kids platform. Within the limits of our stocks, biobanked PDT-O generated in this study will be made available for collaborative work upon request following approval by an internal review board and completion of a Materials Transfer Agreement.