Olig1/2-Expressing Intermediate Lineage Progenitors Are Predisposed to PTEN/p53-Loss–Induced Gliomagenesis and Harbor Specific Therapeutic Vulnerabilities

High-grade gliomas such as glioblastoma (GBM) are the most common aggressive brain tumors and account for more than half of mortality among central nervous system tumors (1). Current multimodal therapies for GBM, which involve surgery, radiation, and chemotherapy, have not changed significantly in several decades and are not curative (1). Based on methylation and transcriptomic profiles, human GBMs are classified into at least three subtypes: proneural, classic, and mesenchymal (2). The proneural subtype is enriched with oligodendrocyte lineage signature genes and often exhibits PDGFRA amplification, TP53 loss, and increased PI3K signaling (2–4), whereas the classic subtype is associated with astrocytic signatures and EGFR amplification, and the mesenchymal subtype is characterized by immune infiltration and NF1 and TP53 loss. Though IDH-mutated glioma generally exhibits a better disease outcome (5), GBM subtype classifications have not shown a significant prognostic value in clinical settings.

Understanding the identity of the cells of origin may impart unique phenotypic signatures and specific therapeutic vulnerabilities in brain tumors (6). Gliomas are thought to originate from the transformation of neural stem cells (NSC) that gives rise to all neural cell types. However, the cells of origin in GBM subtypes remain hotly debated. Different lineage cells, including NSCs, oligodendrocyte precursor cells (OPC), and GFAP⁺ astroglia, have been shown to serve as cells of origin for gliomagenesis following the loss of the compound tumor suppressors Nf1, Trp53, and/or Pten (7–10). NSCs have notable plasticity but are predominantly quiescent in the adult brain (11, 12). The susceptibilities of NSCs and different glial lineage progenitors to transformation due to genetic alterations are incompletely understood during gliomagenesis.

As a potent tumor suppressor, PTEN has been considered a principal negative regulator of PI3K–AKT signaling (13). Mutations in PTEN or TP53 are frequently detected in high-grade gliomas (4, 14). Mice with Trp53 and Pten deletions in neural progenitors mediated by multilineage-expressing human GFAP promoter-driven Cre (hGFAP-Cre) line are neonatal lethal; however, Pten heterozygotes combined with Trp53 ablation, induced by hGFAP-Cre, yield approximately 70% penetrance of glioma formation but with a latency of 30 to 40 weeks (8). Similarly, adult mice with Pten/Trp53 deletion mediated by inducible hGFAP-Cre^ERT expression have approximately 87% glioma penetrance over 35 weeks (15). However, these genetically engineered mouse models (GEMM) of glioma exhibit a long latency period and incomplete penetrance. It is also worth noting that prolonged Trp53 loss alone in hGFAP-Cre–expressing progenitors could lead to high-grade glioma formation in approximately 30% to 50% mice starting at 30 weeks and later stages (8, 16), which may complicate the understanding of the genetic alterations in gliomagenesis. Despite the prevalence of the genetic alterations in human gliomas, currently PTEN/p53-mutation–induced glioma GEMM models with full penetrance and short latency are lacking. In addition, given that hGFAP-Cre–driven deletions occur in multiple cell types, including radial glial cells and astrocytes, as well as oligodendrocyte lineage cells (17, 18), at present, it is unknown whether Pten/Trp53 loss in oligodendrocyte lineage cells leads to glioma formation.

Recently, single-cell profiling studies indicate that individual tumor cells exist on a continuum between cell lineages including astrocyte-like, oligodendrocyte progenitor cell–like, neural progenitor cell–like, and mesenchymal-like tumor cells (19), suggesting that the cellular plasticity of different cell lineages may contribute to tumorigenesis. OPCs are actively dividing cells (5–10%) in developing and adult brains and represent the largest pool of proliferative progenitor cells (∼ 70%) in the brain parenchyma (20). Olig1 and Olig2 mark the oligodendrocyte lineage cells during development (21, 22), and the ablation of mitotically active Olig2⁺ cells inhibits the growth of actively progressing GBM (23). Recent single-cell profiling of murine and human developing brains identified Olig1/2⁺ intermediate progenitors known as primitive OPCs (pri-OPCs) or precursor OPCs (pre-OPC; refs. 24, 25). Olig1 or Olig2 marks pri-OPCs at early developmental stages prior to OPC commitment (22, 25). In contrast to relative quiescent NSC, Olig1/2⁺ pri-OPCs are rapidly dividing progenitor cells (25) and represent a population of the transit-amplifying cells in the developing and adult brain that are susceptible to malignant transformation (23, 25, 26). However, at present, it remains elusive whether the specific states of neural cell lineages such as the oligodendrocyte lineage are differentially susceptible to distinct genetic alterations during malignant transformation.

Our analysis of The Cancer Genome Atlas (TCGA) databases reveals that comutations of PTEN and TP53 are most significantly enriched in human gliomas among potential tumor suppressors. We selectively ablated Pten and Trp53 in different progenitors to determine which cell lineage states are susceptible to malignant transformation. Surprisingly, deletion of Pten and Trp53 in Nestin-expressing neural stem cells or committed OPCs does not induce glioma formation in mice. In contrast, mice with Pten and Trp53 ablation in Olig1⁺ or Olig2⁺ progenitors developed high-grade gliomas with full penetrance and displaying the cardinal features of human proneural counterparts of this neoplasm. In the course of tumor development, we observed hyperproliferation of Olig lineage progenitors, which acquired expression of stem-like signatures during tumorigenesis. However, deleting Pten and Trp53 in later committed NG2-expressing OPCs does not lead to tumor formation. This suggests that Olig1/2-expressing transit-amplifying intermediate progenitors such as pri-OPCs are highly susceptible to Pten/Trp53 loss–induced transformation into high-grade glioma. In addition, transcriptomic and epigenomic analyses identified critical targets for the cooperative promotion of oncogenic programs including HIPPO-YAP/TAZ and PI3K/AKT signaling. Targeting YAP/TAZ and PI3K effectively inhibited tumor cell growth. Thus, our data suggest that the initiation of gliomagenesis depends not only on driver mutations but also on the developmental state along the lineage.

Pten^fl/fl (Jackson Stock No: 006440), Trp53^fl/fl (Jackson Stock No: 008462), Olig2tvaCre (Jackson Stock No: 011103), Olig1-Cre mice (27), Nestin-Cre, Nestin-Cre^ERT (Dr. Ryoichiro Kageyama, Kyoto University, Japan; MGI:4941327), Pdgfra-Cre^ERT (Jackson Stock No: 018280), Ng2-Cre (Jackson Stock No: 008533), and GFP reporter mice were bred to achieve indicated genotypes on a mixed C57Bl/6;129Sv background. Immunocompromised mice NGS were obtained from the Cincinnati Children's facility. All the procedures performed on the mice are approved by Institutional Animal Care and Use Committee protocols of the Cincinnati Children's Hospital Medical Center.

Tumor tissues were dissociated using Accutase (Gibco, A1110501). Cell suspension was cultured in SFM (serum-free media) composed of DMEM/F12 with Insulin-Transferrin-Selenium supplement (1×; Sigma No I3146), human-EGF (20 ng/mL; PeproTech), human-bFGF (10 ng/mL; PeproTech), and penicillin/streptomycin (1 U/mL) to establish PPO1 and PPO2 cell lines. GFP⁺ cells from Nestin-Cre;GFP mice at P7 were harvested and cultured as Nestin⁺ neural stem cells (neurosphere). Single-cell suspension was made in the neurobasal medium (Gibco) with B27 supplement (50×, Gibco) and bFGF and EGF. HEK 293T cells were used for lentivirus packaging as previously described.

Cryosections (15 μm) or vibratome sections (35–50 μm) were permeabilized and blocked-in blocking buffer (0.3% Triton X-100 and 5% normal donkey serum in PBS) for 1 hour at room temperature and overlaid with primary antibodies overnight at 4°C.

Following antibodies were used: Rabbit anti-Olig2 (Millipore; cat. #AB9610; RRID: AB_10141047 IF/IHC dilution 1:500), Mouse anti-Olig2 (Millipore; cat. #MABN50; RRID: AB_10807410 IF/IHC dilution 1:500), Rat anti-PDGFRa (BD Biosciences; cat. #558774; RRID: AB_397117 IF/IHC dilution 1:200), Goat anti-MBP (Santa Cruz Biotechnology; cat. #sc-13914; RRID: AB_648798 IHC dilution 1:200), Mouse anti-GFAP (Sigma; cat. #G3893; RRID: AB_477010 IF/IHC dilution 1:500), Rabbit anti-Ki67 (Thermo Fisher Scientific; cat. #RM-9106; RRID: AB_2335745 IF/IHC dilution 1:200), Mouse anti-BrdUrd (BD Biosciences; cat. #374580; RRID: AB_400327 IF dilution 1:200), Goat anti-GFP (Novus Biologicals; cat. #NB100-1770; RRID: AB_10128178 IF dilution 1:500), Goat anti-Sox2 (Santa Cruz Biotechnology; cat. #sc-17320; RRID: AB_2286684 IF/IHC dilution 1:200), Rabbit Anti-Phospho S6 Ribosomal Protein (Ser235/236; Cell Signaling Technology; #4858, RRID:AB_916156 IHC dilution 1:100), Rabbit Anti-Akt (Cell Signaling Technology; #9272 RRID: AB_329827 IHC dilution 1:100), Rabbit anti-Phospho Akt (Ser473; 736E11; Cell Signaling Technology; #3787 RRID: AB_331170 IHC dilution 1:100), Rabbit p44/42 MAPK (Erk1/2; Cell Signaling Antibody #9102 RRID: AB_330744 IHC dilution 1:200), Rabbit Phospho-p44/42 MAPK (Erk1/2; Thr202/Tyr204; 20G11; Cell Signaling mAb #4376 RRID: AB_331772 IHC dilution 1:400), Goat Anti-Hes5 (Santa Cruz (sc-13859; A-17; RRID: AB_2233041 IHC dilution 1:100 WB 1:500), Rabbit Anti-Yap (Cell Signaling Antibody #4912 RRID: AB_2218911WB dilution 1:1,000), Rabbit Anti-Yap/Taz (D24E4; Cell Signaling Technology; #8418 RRID: AB_10950494 WB dilution 1:1,000), Rabbit Anti-Sox10 (EPR4007-104 (ab180862; Abcam RRID: AB_2721184 IHC dilution 1:200), Rabbit Anti Sox9 (D8G8H; Cell Signaling Technology #82630 RRID: AB_2665492 IHC dilution 1:300), Rabbit Anti-HER3/ErbB3 (D22C5; Cell Signaling Technology #12708 RRID: AB_2721919 IHC dilution 1:200), Rabbit Anti-EGFR (Cell Signaling Technology #4267 RRID: AB_2246311 IHC dilution 1:200), and Mouse Anti-Actin (Clone AC-40, Sigma-Aldrich RRID: AB_476730 WB 1:1,000). Immunofluorescence images were acquired using a confocal inverted laser scanning confocal microscope (Nikon, C2+), and images are saved as raw ND2 format, which are analyzed using ImageJ. Hematoxylin and eosin and IHC images are acquired by using light microscope Nikon Eclipse and analyzed with ImageJ software.

The tumor cells (2,000 cells in total 100 μL of media) were seeded per well in a 96-well plate. For every drug concentration, 5 replicates were used. The cells were incubated with drugs for 72 hours and then assessed for loss of cell viability by the WST-1 cell proliferation assay kit (Takara Bio; cat. #MK400). It is a calorimetric assay based on mitochondrial dehydrogenase enzyme activity in live cells that cleave the tetrazolium salts provided in the reagent. The 10% (v/v) of premix was used per well and incubated for 2 hours at 37°C, and wavelength absorbance is noted at 405 nm and 650 nm by a Microplate Reader (SpectraMax M2 spectrophotometer, Molecular Devices).

Allograft models transplanted PPO2-Luciferase expressing cells (100,000 cells) were divided into 4 groups: vehicle, verteporfin, PX-866, or drug combination (verteporfin and PX-866). The mice were started on drug treatment after establishing tumor implantation verified via luciferase-based bioluminescence. The mice are followed every week with bioluminescence IVIS imaging and wight measurement. The endpoint is a loss of 20% of body weight or poor health condition.

RNAs were extracted using RNeasy Kits (Qiagen 74104). Plasmid DNAs were isolated using the QIAprep Spin Miniprep Kit (Qiagen, 27104). Genomic DNAs were extracted using the DNeasy Blood and Tissue Kit (Qiagen, 69504). The RNA-seq reads in FASTQ format are checked for quality control and adapter trimming. Gene-level expression was assessed by counting features for each gene, as defined in the NCBI RefSeq database (https://www.ncbi.nlm.nih.gov/refseq/). Raw counts were normalized as transcripts per million. The differential expression was analyzed by R using DESeq2 package using raw read as input. The heat maps are drawn by using R and AltAnalyze (http://www.altanalyze.org). Volcano plots were done with Prism (GraphPad). For gene set enrichment analysis (GSEA), normalized counts from DESeq2 were used as input in GSEA 4.0.3. Cut&Run-seq and assay for transposase-accessible chromatin using sequencing (ATAC-seq) assays were performed as described (28, 29). Briefly, trimmed reads were aligned to the reference mouse genome version mm10 with the program BOWTIE v2.3.4.1. Aligned reads were stripped of duplicate reads with the program sambamba v0.6.8. Peaks were called using the program MACS v2.1.2 using the narrow and broad peaks mode. Mutation enrichment analysis was based on TCGA (https://portal.gdc.cancer.gov/), Catalogue of Somatic Mutations in Cancer (COSMIC; https://cancer.sanger.ac.uk/cosmic/browse/tissue) and the Chinese Glioma Genome Atlas (CCGA; http://www.cgga.org.cn/).

All the analysis are done by GraphPad Prism 7. Data are shown in bar graphs as mean ± SEM. XY plot (PTEN comutation graph) and volcano plot are plotted using Prism. The statistical analysis of two samples is done with a two-tailed unpaired Student t test, and the log-rank test was used for survival analysis. Significance was set as P < 0.05, unless otherwise indicated (*, P < 0.05; **, P < 0.01; ***, P < 0.001). The data distribution is considered to be normal though not formally checked.

All high-throughput data generated in the paper are deposited in the NCBI Gene Expression Omnibus. The accession number is GSE200892.

TCGA (n = 2,334) and COSMIC data sets (n = 6,075) on human gliomas indicate that PTEN and TP53 are among the frequently mutated genes in anaplastic grade III and IV GBMs (Fig. 1A). Further mutation enrichment analysis revealed that TP53 is most significantly comutated with PTEN among the known tumor suppressors in a cohort of grade III and IV gliomas from TCGA (Fig. 1B). Similarly, the CGGA data set (n = 54 GBM) shows that PTEN and TP53 are among the most frequently mutated genes (Supplementary Fig. S1).

To investigate the effects of Pten and Trp53 loss in neural stem/progenitor cells on tumorigenesis, we deleted Pten and Trp53 floxed alleles in Nestin⁺ NSCs in a Nestin-Cre transgenic line, in which Cre is expressed in essentially all NSCs starting at E12.5 (Supplementary Fig. S2A; ref. 30). This resulted in the generation of Pten^fl/fl;Trp53^fl/fl;Nestin-Cre mice. All mice with the double mutations died during neonatal development without brain tumor formation (Supplementary Fig. S2B). To circumvent the neonatal lethality, we deleted Pten and Trp53 in postnatal Nestin⁺ NSCs in a tamoxifen-inducible Nestin-Cre^ERT2 line (31) by generating Pten^fl/fl;Trp53^fl/fl;Nestin-Cre^ERT mice. Tamoxifen (TAM) injection was carried out in the perinatal stage intraperitoneally once daily for five consecutive days (Supplementary Fig. S2C). The recombination of floxed Pten and Trp53 alleles was confirmed by PCR in the brain of TAM-treated Pten^fl/fl;Trp53^fl/fl;Nestin-Cre^ERT mice (Supplementary Fig. S2D and S2E). Intriguingly, mice with tamoxifen-induced Pten;Trp53 deletion in Nestin⁺ stem-like cells at postnatal stages failed to develop glioma, and even at 52 weeks after injection the mice showed no symptoms of brain malignancy (Supplementary Fig. S2F), though a few of the mutant mice (3/7) died due to the development of soft-tissue tumors in extremities and abdomen between the age of 31 and 40 weeks (Supplementary Fig. S2G and S2H).

To determine the effects of deletion of Pten and Trp53 floxed alleles in OPCs, another potential cell of origin for brain tumorigenesis (9), we bred floxed Pten^fl/fl;Trp53^fl/fl with a tamoxifen-inducible Cre line Pdgfra-Cre^ERT (32) to generate Pten^fl/fl;Trp53^fl/fl;Pdgfra-Cre^ERT mice (Supplementary Fig. S3A). The neonatal tamoxifen induction did not result in any mice forming brain tumors, but they developed malignant soft-tissue tumors in the axilla, flanks, extremities, and cranial nerves (Supplementary Fig. S3B–S3D). The animals were sacrificed due to high tumor load around 16 weeks after tamoxifen injection (Supplementary Fig. S3E). These results were not unexpected as Pdgfra expression is not exclusive to OPCs but is also detected in other cell types including Schwann cells, fibroblasts, and subsets of endothelial cells (33, 34). However, there were no brain tumors formed in this cohort nor did histology reveal any hyperplastic regions or other signs of tumor formation.

To avoid the systemic activation of Pdgfra-Cre^ERT and rule out the possibility that tumors would have eventually formed without early removal due to peripheral tumor formation, we performed intracranial injections of 4OH-tamoxifen into the cortices of Pten^fl/fl;Trp53^fl/fl;Pdgfra-Cre^ERT neonates at P4 (Supplementary Fig. S3F). Recombination was confirmed by the detection of GFP reporter expression at 4 days after injection (Supplementary Fig. S3G). Despite the initial apparent expansion of GFP⁺ cells after tamoxifen injection, Pten^fl/fl;Trp53^fl/fl;Pdgfra-Cre^ERT mice (0/10) failed to develop glioma in the brain even by 15 months after 4OH-tamoxifen injection (Supplementary Fig. S3H). Similarly, loss of Pten and Trp53 in Ng2-Cre–directed OPCs also did not trigger glioma formation in the brain and mice died within 4 months due to the formation of soft-tissue tumors in the upper extremity and intraperitoneal area (Supplementary Fig. S4A and S4B). Together, these observations indicate that Pten;Trp53 deletion in Nestin⁺ NSCs or PDGFRa⁺/NG2⁺ OPCs in young mice does not induce glioma formation during their lifespan. This is distinct from previous reports of glioma formation in the mice with deletion of Nf1/Trp53 or Nf1/Pten/Trp53 in Nestin⁺ NSCs or NG2⁺ OPCs (7, 9, 35), suggesting that gliomagenesis depends at least in part on different driver mutations in different neural lineage progenitors.

Given the failure to form tumors upon Pten and Trp53 deletion in NSCs and OPCs, we hypothesized that an intermediate progenitor state is susceptible to Pten/Trp53 loss–mediated gliomagenesis. We then investigated the effect of Pten and Trp53 loss in Olig-expressing intermediate progenitors prior to committed OPCs (25). Single-cell transcriptomic analysis of hGFAP-promoter–expressing glial progenitors indicated that Olig1/2-expressing pri-OPCs were the most prominently active cycling cells, representing transit-amplifying cells in the developing brain (Supplementary Fig. S5A and S5B; ref. 25). To inactivate Pten and Trp53 in Olig intermediate progenitors and their lineage cells, we utilized Olig1-Cre (27) or Olig2-tva-Cre (36) deleter lines to generate Pten^fl/fl;Trp53^fl/fl;Olig1-Cre (designated as PPO1 or Pten^fl/fl;Trp53^fl/fl;Olig2-tva-Cre mice (designated as PPO2), respectively (Fig. 1C). We confirmed the brain-specific recombination through genotyping (Supplementary Fig. S6A). Strikingly, all the double mutant PPO1 mice developed high-grade glioma with 100% penetrance and a median lifespan of 90 days (Fig. 1D). Similarly, all PPO2 mice developed neurologic symptoms within 3 months and had a median lifespan of 115 days (Fig. 1D). Mice with Pten heterozygosity and Trp53 ablation in Olig1/2-expressing cells also developed gliomas but with a longer latency and a median lifespan of approximately 240 days compared with PPO mice (Fig. 1D), indicating a dose-dependent effect of PTEN on tumorigenesis. In contrast, mice lacking Pten or Trp53 alone in Olig1/2⁺ lineage cells did not develop tumors at this stage (Fig. 1D; Supplementary Fig. S6B), suggesting a cooperative role of PTEN with p53 loss in promoting glioma tumorigenesis.

PPO1 and PPO2 mice had symptoms typical of high-grade gliomas, including loss of weight, paraplegia, and occasional seizures. Diffusely infiltrating tumors were detected in multiple noncontiguous locations in the brain regions such as the cortex (26.9%), thalamus (30.6%), midbrain (7.1%), pons (17.6%), and hypothalamus (15.2%) as well as spinal cord (2.6%; Fig. 1E and F; Supplementary Fig. S6C). Most of the tumors were diffuse in nature and occurred at noncontiguous locations, suggesting a possibility of localized tumor cell invasion or multifocal initiation.

Gliomas formed in PPO1 and PPO2 mice were microscopically similar and composed of hyperchromatic plump to spindly nucleated cells with interdigitating fascicles and a high mitotic index with bipolar to multipolar mitotic spindles (Fig. 1G). Histologic evaluations indicated that the majority of tumors resembled anaplastic glioma (WHO grade III) with some tumors characterized by pseudo-palisading necrosis suggestive of grade IV GBM (Fig. 1H). OLIG2⁺ tumor cells typically presented with evidence of invasion such as perivascular and leptomeningeal spread (Fig. 1I). Moreover, karyotype analysis of the isolated tumor cells showed gross chromosomal abnormalities, including aneuploidy and extrachromosomal DNA (Supplementary Fig. S6D), which has been implicated in human GBM pathogenesis (37).

Tumors formed in both PPO1 and PPO2 mice had an extensive expression of stem-like markers such as OLIG2 and SOX2 (23, 38) and a proliferative marker, Ki67 (Fig. 1J). The percentages of OLIG2⁺, SOX2⁺, and Ki67⁺ cells were not significantly different between PPO1 and PPO2 tumors (Fig. 1K). The tumors exhibited moderate positivity for the astrocyte marker GFAP and focal Nestin expression (Fig. 1L), suggestive of intermingling with reactive astrocytes. In addition, we observed PDGFRa⁺ OPC-like cells in tumor cells (Fig. 1L). Consistent with the loss of Pten, tumors showed activation of PI3K–Akt–mTOR signaling, manifested by strong pAKT and pS6 expression in tumor tissues (Fig. 1M). PPO tumors were essentially negative for reticulin and smooth muscle actin, the markers of sarcomatous mesenchymal differentiation (Supplementary Fig. S6E and S6F), suggesting that they are not mesenchymal-like tumors.

To further determine the susceptibility of Olig2⁺ progenitor cells to form tumors at the postnatal stage, we utilized temporally regulated Olig2-Cre^ERT2 to ablate Pten and Trp53. Pten^fl/fl;Trp53^fl/fl;Olig2-Cre^ERT2 animals were administered with TAM at the early postnatal period from P2 to P7 daily. We found that 100% of TAM-treated mice (5 out of 5 mice) developed high-grade glioma with grade III or IV histology in young adulthood (Supplementary Fig. S7A–S7D). Tumors show high OLIG2 and Nestin positivity (Supplementary Fig. S7E and S7F), suggesting that Olig2⁺ intermediate lineage precursors are highly susceptible to Pten/p53-loss–induced gliomagenesis at the early postnatal stage.

To determine whether pri-OPCs show similar susceptibility to other genetic alterations for transformation, we examined the effect of Trp53 and Nf1 mutations, which can lead to the formation of malignant astrocytoma (39). To knockout Trp53 and Nf1 in pri-OPCs, we generated Nf1^fl/fl; Trp53^fl/fl; Olig1-Cre mice (Supplementary Fig. S8A). These mice also developed multifocal high-grade gliomas with a median survival of 119 days (Supplementary Fig. S8B). Gliomas in these mice had a morphology similar to those in PPO1 and PPO2 mice and also expressed SOX2, OLIG2, pERK, and pAKT (Supplementary Fig. S8C and S8D). These data suggest that Olig1/2⁺ transit-amplifying pri-OPCs are prone to form gliomas in response to the loss of function of tumor suppressors, including Pten/Trp53 or Nf1/Trp53.

Tumor cells isolated from the gliomas in PPO1 and PPO2 mice were able to form tumor spheres when cultured in SFM (Fig. 2A). The number and size of spheres were higher than those derived from mouse NSCs (Fig. 2A–C), suggesting that the growth of tumor cells was faster than that of normal NSCs. Almost 80% of seeded single cells isolated from the gliomas formed tumorspheres as compared with 10% of normal NSCs as shown by serial dilution in the tumor-sphere assay (Fig. 2B), suggesting that tumor cells retain their stemness. To access the stem cell frequency in PPO1 and PPO2 cells, we performed an extreme liming dilution assay (40). The stem cell frequency (average number of cells representing one active stem cell) in PPO1 and PPO2 was 3.10 and 3.04, respectively, suggesting no discernible difference in self-renewal potential between PPO1 and PPO2 cells (Fig. 2D).

To evaluate the tumorigenicity of PPO1 and PPO2 neoplastic cells, primary neoplastic cells were transplanted into secondary immunocompromised recipients. Tumors formed with 100% penetrance (Fig. 2E), and mice with secondary infiltrative tumors had a median lifespan of 18.4 and 16.4 weeks for PPO1 and PPO2, respectively (Fig. 2F). Tumors in secondary recipients were hypercellular and showed a high frequency of mitosis (Fig. 2G). Allografted tumors resembled parent tumors microscopically (Fig. 2E–G), and so did the expression pattern for OLIG2, SOX10, and pS6 as well as Nestin and GFAP (Fig. 2H). Moreover, transplantation of PPO2 cells isolated from mice with a mixed background into immunocompetent C57BL/6 mice resulted in the formation of the secondary glioma in one out of four allografts (Supplementary Fig. S9A and S9B). These data indicate that primary neoplastic cells from PPO1/2 mice are highly tumorigenic.

To determine whether Pten/Trp53 ablation alters OPC proliferation during development prior to glioma formation, we performed BrdUrd pulse labeling in the brains of wild-type controls, single-knockout Pten ^fl/fl or Trp53 ^fl/fl mediated by Olig1-Cre, and PPO1 mice at P14. We detected a modest increase of BrdUrd⁺/OLIG2⁺ cells in the cortex of Pten and Trp53 single-knockout mice in contrast to a substantial increase in the cortex of PPO1 mice compared with controls (Fig. 3A and B). Similarly, in PPO2 mice, we observed an increase in BrdUrd⁺ proliferating cells in the cortex and thalamus locations (Fig. 3C and D). These data suggest an increase of OPC proliferation in mice with deletions of both Pten and Trp53 compared with wild-type and single knockout mice. Intriguingly, we detected higher frequencies of PDGFRa⁺ cells that colabeled with the stem cell marker SOX2 and Nestin in the cortex of Pten/Trp53 knockout mice compared with controls at P25 (Fig. 3E–G), suggesting that a subset of PDGFRa⁺ OPC-like cells have acquired stem-like properties in PPO1 and PPO2 mice. Taken together, our results suggest that Pten/Trp53 deletion in Olig1/2⁺ progenitors enhances OPC proliferation and their acquisition of stem-like characteristics at preneoplastic stages.

When compared with human gliomas, the gene expression signature of Pten- and Trp53-deficient gliomas is much closer to the proneural subtype of human glioblastoma (Fig. 4A and B). A recent study classified murine gliomas as type 1 tumors originating from NSCs with high expression of EGFR and SOX9, and type 2 tumors originating from OPCs with high SOX10 and ERBB3 expression (41). To determine whether tumors in PPO1 and PPO2 mice are type 1 or type 2 gliomas, we performed transcriptomic profiling and compared our data with published type 1 and type 2 tumor profiles (41). Principal component analysis indicated that the profiles of PPO2 and PPO1 clustered together but were separated from those of type 1 or 2 tumors (Fig. 4C), suggesting that PPO2 and PPO1 tumors have unique gene expression profiles distinct from those of type 1 or 2 murine gliomas.

It has been suggested that type 1 tumors are predominantly located in dorsal brain regions and that type 2 tumors are present predominantly in ventral brain regions (41). In contrast, we observed multifocal tumors in both dorsal and ventral brain regions of PPO1 and PPO2 mice. In the tumor lesions located in dorsal and ventral brain regions in these mice, both type 1 and 2 markers such as SOX9 and SOX10, respectively, were detected (Fig. 4D and E). Although SOX10⁺ cells appeared more abundant in PPO1 gliomas than in PPO2 tumors (Fig. 4D and E), and PPO2 tumors appeared to have a higher density of SOX9⁺ cells than PPO1 tumors (Fig. 4F and G), we did not observe definite differential dorsal-ventral distributions of SOX9 and SOX10. Expression of another type 1 marker, EGFR, was detected sparsely to moderately in both PPO1 and PPO2 tumors. Intriguingly, cells expressing the type 2 marker ERBB3 were present in low amounts in PPO1 tumors in both dorsal and ventral regions and in dorsal regions of PPO2 tumors but were essentially absent from the ventral brain region of PPO2 tumors. Similar distributions of type 1 and 2 marker–expressing cells were also observed in the Nf1^fl/fl;Trp53^fl/fl;Olig1-Cre mice compared with PPO1 mice (Supplementary Fig. S8D). Thus, tumors in PPO1 and PPO2 mice derived from the Olig1/2 lineage may represent a unique tumor subtype distinct from type 1 and 2 tumors.

To understand the process of tumorigenesis in the mice lacking Pten and Trp53, we performed RNA-seq transcriptional profiling of the cortices from control and PPO1 mice at P53, P68, P91, P120, and P131. Principal component dimensionality reduction analysis indicated that the gene expression clusters of tumor tissues at early stages (P53 and P68) segregated from those at late stages at P91, 120, and P131 (Fig. 5A), suggesting that they represented different phases during glioma progression. Tumor purity defined by the ESTIMATE algorithm (42) indicated that the proportion of tumor cells, immune and stromal components in the isolated tumor tissues was similar over time from P68 to P131 except at the early-stage P53 (Fig. 5B). Similarly, deconvolution analysis by Xcell (43) also indicated that later-stage tumors have a higher proportion of immune cells including M2 macrophages, T effector memory cells, and activated and immature dendritic cells compared with early-stage tumors (Supplementary Fig. S10).

The hierarchical clustering analysis revealed a gradual change in the expression of genes and pathways over the course of tumorigenesis (Fig. 5C). Gene ontology analysis indicated that there is a progressive increase in the pathways associated with MAPK, PI3K–AKT, and stem cell differentiation during tumor progression, whereas the genes associated with cell-cycle checkpoint signaling, RB tumor suppressor signaling, and apoptosis were downregulated (Fig. 5C and D). There was also a gradual increase in the expression of oncogenic signaling (Myc, Pdgfra) and Notch signaling (Hes5, Notch1), which is highly active in glioblastoma stem cells and maintains stem-like properties (44), in later phases of tumor progression (Fig. 5E and F). In a similar vein, we detected upregulation of a NOTCH effector HES5 expression in tumor nodules compared with surrounding areas (Supplementary Fig. S11). In addition, GSEA indicated enrichment of the HIPPO-YAP conserved signature and upregulation of TAZ target genes over time (Fig. 5G). Activation of HIPPO-YAP/TAZ signaling has been shown to regulate stemness and cell plasticity in GBM (45). Thus, our serial time point analysis indicates a gradual upregulation of oncogenesis and stemness-related pathways, including HIPPO-YAP/TAZ signaling, during gliomagenesis.

In order to better characterize tumor cells, we isolated and established tumor cell lines from PPO1 and PPO2 tumors and performed a gene expression analysis. Unbiased principal component analysis (PCA) showed that PPO1 and PPO2 tumor cells appeared to cluster together in closer proximity to mouse NSCs than to OPCs (Fig. 6A; ref. 46), suggesting that tumor cells have a stem-like state. Analysis of differentially expressed genes revealed overrepresentation of the PI3K–AKT, MAPK, and Hippo signaling pathways and increased autophagy and extracellular matrix interactions in PPO1 and PPO2 tumor cells compared with NSCs (Fig. 6B). In Pten- and Trp53-deficient PPO1 and PPO2, there was a marked upregulation of HIPPO-YAP signaling effectors (Yap1, Tead2, Ndrg1, Serpine1, and Wwtr1), the genes involved in cell-cycle regulation (Ccnd1, Lamb1, and Rgs16), and PI3K–AKT pathway genes (Akt1 and Ndrg4) compared with OPCs or NSCs (Fig. 6C and D; Supplementary Fig. S12A and S12B). GSEA indicated enrichment of the pathways associated with AKT, YAP, and cell proliferation signaling (e.g., E2F) in PPO2 tumor cell lines compared with committed OPCs and NSCs (Fig. 6E and F). Thus, our data indicate that glioma cells derived from PPO1 and PPO2 acquire stem-like characteristics with elevated HIPPO and PI3K–AKT signaling, which may promote the tumorigenic process.

To determine cis-regulatory landscape profiles in tumor cells with Pten/p53 loss, we performed a CUT&RUN assay for the enhancer mark H3K27ac (47, 48) together with an ATAC-seq to map genome-wide “open” accessible chromatin (49). The distribution of H3K27ac-marked enhancers in PPO1 and PPO2 tumor cells differed from that in mouse OPCs in regulatory regions both proximal and distal to transcription start sites (Fig. 7A). H3K27ac- marked enhancer peaks were essentially overlapped in PPO1/PPO2 tumor cells (Fig. 7B). There were approximately 7,268 H3K27ac peaks unique in PPO1/PPO2 tumor cells compared with OPCs (Fig. 7C). Intersection of the distinct H3K27ac-enriched gene loci in both PPO1/PPO2 tumor cells with the differentially expressed genes compared with OPCs yielded approximately 1,203 H3K27ac-enriched gene loci that were upregulated in the tumor cells (Fig. 7D). Gene ontology analysis of the upregulated genes revealed pathways associated with oncogenic processes and stemness including focal adhesion, HIPPO signaling, glycosaminoglycan biosynthesis, PI3K–AKT signaling and extracellular matrix interaction (Fig. 7E), which are associated with human GBM (50, 51).

Enhancers with open chromatin structure have been shown to be enriched for transcription factor (52) binding sites, which serve as active regulatory elements for transcription (53). To identify TF motifs in the active regulatory elements, we performed motif enrichment analysis of the enhancers that were unique in PPO1/PPO2 tumor cells by using the motif discovery program Meme-Chip (54). We found that NF-kB, E2F1, MYCN, MYC, and TEAD1 motifs were highly enriched in the H3K27ac-marked enhancer elements in tumor cells (Fig. 7F), suggesting that members of these TF families may actively participate in the transcriptional regulation of glioma tumorigenic processes.

Similarly, ATAC-seq analysis showed more robust open chromatin sites in PPO1 and PPO2 tumor cells than mouse OPCs (Fig. 7G). The dimensionality reduction analysis shows clustering ATAC-seq peaks of PPO1 and PPO2 cells compared with OPCs (Supplementary Fig. S13). The genes associated with differential open chromatin sites were positively correlated to transcriptomic changes in PPO1/PPO2 tumor cells compared with OPCs (Fig. 7H), suggesting that the gene expression change in PPO tumor cells is likely due to the alteration of gene regulatory elements, not the downstream secondary responses. The gene loci associated with enhancers within open chromatin sites included those pertinent to HIPPO-YAP signaling (Cyr61, Birc5, and Tead3) and the PI3K–AKT–mTOR axis (Akt1s1, Pik3r1, and Etv4; Fig. 7I and J). Moreover, chromatin accessibilities in the enhancers of stemness-associated genes (Olig2, Sox2, Nestin, and Ascl) and cell-cycle–related genes Cdkn2a and Cdk1 were increased in tumor cells (Fig. 7J). Consistently, expression of these tumorigenesis-associated genes was upregulated in tumor cells compared with OPCs. Together, our analyses of active regulatory elements indicate that PTEN/p53-loss–induced PPO tumor cells activate the oncogenic signaling including HIPPO/TEAD and PI3–AKT pathways, and the stemness program for promoting gliomagenesis.

To assess the HIPPO-YAP/TAZ signaling dependencies in PPO1/PPO2 tumor cells, we used small interfering RNAs (28) to deplete Taz and Yap. qPCR analysis confirmed significant reductions in Taz and Yap upon treatment of PPO1/PPO2 cells with siRNAs (Fig. 8A). Depletion of both TAZ and YAP drastically reduced the percentage of proliferative tumor cells as assayed by BrdUrd (Fig. 8B and C) incorporation and the proliferation marker Ki67 (Fig. 8B and C), while also inducing cell apoptosis as shown by increased cleaved caspase-3 expression and loss of stem cell marker SOX2 (Fig. 8D and E). These results suggest that the growth of PPO1 and PPO2 tumor cells depends on TAZ and YAP.

The activation of YAP/TAZ and PI3K signaling in PPO1/PPO2 tumor cells suggests that these two pathways could be targeted pharmacologically to curb tumor growth. We therefore examined whether cotargeting HIPPO-YAP/TAZ and PI3K signaling inhibited tumor cell proliferation. Inhibition of individual pathways with a PI3K inhibitor, PX-866, which targets PI3K activity (55), or with verteporfin, which inhibits YAP/TAZ signaling activity (56), blocked cell proliferation to some extent (Fig. 8F), whereas normal OPCs were not sensitive to PX-866 or verteporfin (Supplementary Fig. S14A and S14B). The combined inhibition of HIPPO-YAP/TAZ and PI3K signaling attenuated the proliferation of PPO1 and PPO2 tumor cells to a significantly greater extent than either drug alone (Fig. 8G). To determine if the effects of the two drugs on cell growth were synergistic, we applied the Bliss model to calculate synergy scores (57) and classified the Bliss score > 10 as synergistic (57). The combined treatment with PX-866 and verteporfin elicited synergistic effects on net cell growth in PPO1/2 tumor cells across a range of concentrations (Fig. 8G and H). In contract, the growth of PPO1 or PPO2 tumor cells was not altered by treatment with dasatinib or irbinitinib (Fig. 8J), which are inhibitors of the growth of type 2 glioma (41). This is consistent with our findings that PPO1 or PPO2 tumors are distinct from type 2 glioma. To further determine the treatment effect on glioma growth in vivo, we treated the allografts bearing PPO2 tumor cells with PX-866, verteporfin, or in combination. As shown Fig. 8I, we find that the combination of PI3K and HIPPO-YAP/TAZ inhibition significantly increases survival compared with treatment with individual inhibitors alone. Together, our data suggest that activation of HIPPO-YAP/TAZ signaling and PI3K signaling is an important driver that promotes the growth of tumors with Pten and Trp53 loss.

Aggressive GBMs are highly heterogeneous with distinct genetic alterations and phenotypic characteristics (4, 19). Additional layers of complexity originate from differences in cells of origin and the tumor microenvironment (58). Understanding the genetic and functional heterogeneities is critical for devising effective therapeutic interventions. Distinct genetic alterations observed in GBM subtypes appear to be associated with specific cell lineages, including NSCs, astrocytes, and oligodendrocyte lineage cells (4, 19). PTEN and TP53 mutations are among the most prevalent genomic alterations in brain tumors. We further show that comutations of PTEN and TP53 are most significantly enriched in human high-grade gliomas among tumor suppressor gene alterations. Previous studies showed that mice with Pten/Trp53 mutations mediated by human GFAP-promoter–expressing cells (8, 15) develop gliomas but with a long latency and incomplete penetrance. Because the hGFAP promoter has been shown to direct Cre expression in multiple cell lineages including radial glia/NSCs, astrocytes, and OPCs (17, 18), identities of tumor-initiating cells in those models remain unknown. In addition, at present, it is not clear whether certain states of progenitors in the oligodendrocyte lineage are susceptible to the malignant transformation caused by these genetic alterations.

In this study, by using multiple cell-state–specific in vivo ablation strategies, surprisingly, we find that mice with Pten and Trp53 ablation in NSCs (nestin⁺) and OPCs (PDGFRa⁺ or NG2⁺), two major cells of origin of glioma (10, 59), do not develop gliomas, instead they develop nonglioma soft-tissue tumors in peripheral nerves in these mice. In contrast, Pten and Trp53 deletions in Olig⁺ lineage cells, including transit-amplifying intermediate pri-OPC states between NSCs and OPCs in the developmental continuum of the oligodendrocyte lineage, induce high-grade glioma formation with 100% penetrance and short latency (2–3 months). High-grade gliomas caused by the loss of PTEN/p53 in Olig lineage cells mimic the histologic, molecular, and pathologic characteristics of human proneural GBM. Similarly, the ablation of Nf1 and Trp53 in the Olig⁺ lineage cells also resulted in glioma formation with full penetrance. These observations suggest that Olig+ transit-amplifying intermediate progenitor states are uniquely vulnerable to malignant transformation caused by tumor suppressors PTEN/p53 or NF1/p53 loss of functions.

In contrast to Nf1/p53 loss-induced glioma from both NSCs and OPCs (9, 35), the loss of Pten and Trp53 in NSCs and OPCs is not sufficient to induce glioma within 6 months, although we could not exclude the possibility of tumor formation with a prolonged loss of Pten and Trp53. Thus, our observations suggest that genetic alterations dictate the tumorigenic potential of neural lineage progenitors or tumor-initiating cells, which have distinct mutational vulnerabilities. Given that both PTEN/p53 and NF1/p53 loss of functions induce glioma formation in Olig1/2⁺ intermediate lineage progenitors, the presence of abundant Olig2-expressing pri-OPCs in human and mouse models of proneural GBMs (23, 25, 38) suggests a therapeutic vulnerability in GBMs such as proneural subtype tumors by targeting the transit-amplifying pri-OPC population.

A recent study classified murine gliomas carrying Nf1, Pten, and Trp53 triple mutations, which rarely occur in an individual, that originate from NSCs and OPCs as type 1 and type 2 tumors, respectively (41), exhibiting disparate pharmacologic vulnerabilities. Type 2 tumors, which are mainly localized in the ventral forebrain, are sensitive to HER2 kinase inhibitors such as dasatinib and irbinitinib, whereas type 1 tumors, which are restricted to dorsal locations, are not (35). In contrast, we found that PPO1 and PPO2 tumors form multifocal tumor lesions in both dorsal and ventral forebrain, consistent with broad expression patterns of Olig1 or Olig2 in the brain and tumor tissues. Notably, both type 1 (e.g., SOX9) and type 2 markers (e.g., SOX10) were detected in PPO1 and PPO2 tumors, although other type 1 (e.g., EGFR) and type 2 (e.g., ERBB3) markers were not or weakly expressed. Furthermore, PPO1 and PPO2 tumors are not sensitive to type 2 kinase inhibitors. Thus, PPO1 and PPO2 tumors derived from PTEN/p53 loss of functions in the Olig1/2-intermediate progenitors appear to encompass expression of a set of type 1 and type 2 markers and may represent a distinct tumor subtype or a potential type 3 tumor. This suggests that different tumor-initiating lineage cells may result in distinct tumor phenotypes and drug sensitivities, indicating a complexity of glioma subtypes in the mouse brain. In addition, PPO1/2 tumors formed in dorsal and ventral brains do not show distinct type 1 and type 2 markers, respectively, suggesting a lack of positional identity in PPO glioma models.

The transcriptomic genomic profiling of PPO tumors at different stages revealed activation of oncogenic pathways, including MYC, PI3K–AKT, MAPK, NOTCH, and PDGFRa signaling, and cell-cycle regulators, at initial tumorigenic stages. Strikingly, HIPPO-YAP/TAZ signaling was substantially upregulated during late stages. HIPPO-YAP/TAZ upregulation maintains the stemness of glioma stem-like cells and is associated with aggressive GBM subtypes and poor survival (60). Our data show that the depletion of YAP/TAZ substantially reduced the growth of PPO1/PPO2 tumor cells, which is in keeping with the critical requirement of YAP/TAZ for tumor initiation and maintenance in animal models of GBM (45). Thus, HIPPO-YAP/TAZ signaling hyperactivation at later phases of tumorigenesis likely endows PPO1/2 cells with tumor stem cell–like attributes, enabling the formation of high-grade gliomas.

HIPPO-YAP/TAZ signaling has been shown to regulate signaling mediated by insulin/IGF, MAPK, G-protein–coupled receptors, and WNT signaling pathways, which are associated with tumor cell growth (61, 62), and may serve as a nexus to integrate diverse tumorigenic signals to promote oncogenic signaling during glioma progression. Upregulation of YAP/TAZ signaling in our animal models of GBM and in different human GBM subtypes (45) pinpoints YAP/TAZ as potential candidates for targeted therapy in GBM. We showed that blocking of YAP/TAZ activity using FDA-approved verteporfin (63) inhibits the growth of PPO1 and PPO2 tumor cells. Importantly, we found that combinatorial treatment with the YAP/TAZ inhibitor verteporfin and the PI3K inhibitor PX-866, which is currently being evaluated in clinical trials for GBM and other cancer types (NCT01259869 and NCT01331083), synergistically decreases the proliferation of PPO tumor cells at reduced doses in culture compared with treatments with single agents. Our data suggest that the combined targeting of YAP/TAZ activity and PI3K signaling in GBMs carrying PTEN/p53 mutations might improve treatment efficacy. Thus, our PTEN/p53-loss–induced high-grade glioma models with accelerated kinetics of tumor occurrence may serve as a platform to dissect the mechanisms of glioma initiation and progression, microenvironment roles, and therefore identify better diagnostic and therapeutic strategies.

No disclosures were reported.

R. Verma: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. X. Chen: Conceptualization, investigation, visualization. D. Xin: Investigation. Z. Luo: Investigation. S. Ogurek: Investigation. M. Xin: Supervision. R. Rao: Writing–review and editing. K. Berry: Writing–review and editing. Q.R. Lu: Conceptualization, resources, supervision, funding acquisition, methodology, writing–review and editing.

This study was supported in part by grants from the CancerFree Kids Foundation, Pray-Hope-Believe Foundation, TeamConnor Childhood Cancer Foundation, and CureStartsNow Foundation to Q.R. Lu. The authors thank A. Esshaghi Bayat for his technical support, L. Parada for the Nf1 floxed mice, and E. Hurlock for critical comments.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.