Brain tumors are thought to originate from stem/progenitor cell populations that acquire specific genetic mutations. Although current preclinical models have relevance to human pathogenesis, most do not recapitulate the histogenesis of the human disease. Recently, a large series of human gliomas and medulloblastomas were analyzed for genetic signatures of prognosis and therapeutic response. Using a mouse model system that generates three distinct types of intrinsic brain tumors, we correlated RNA and protein expression levels with human brain tumors. A combination of genetic mutations and cellular environment during tumor propagation defined the incidence and phenotype of intrinsic murine tumors. Importantly, in vitro passage of cancer stem cells uniformly promoted a glial expression profile in culture and in brain tumors. Gene expression profiling revealed that experimental gliomas corresponded to distinct subclasses of human glioblastoma, whereas experimental supratentorial primitive neuroectodermal tumors (sPNET) correspond to atypical teratoid/rhabdoid tumor (AT/RT), a rare childhood tumor. Cancer Res; 73(18); 5834–44. ©2013 AACR.
Stem and progenitor cells in the subventricular zone (SVZ) are the likely origin of many primary intrinsic brain tumors, in particular of gliomas (1–4). The most common intrinsic adult human malignant brain tumor is the glioblastoma (GBM), which is characterized by moderate-to-high mitotic activity, vascular endothelial proliferation, and/or tumor cell necrosis. Glioblastomas are diagnostically uncontroversial and there is a broad consensus about their diagnostic criteria. They can include an oligodendroglial component or show elements resembling human supratentorial primitive neuroectodermal tumors (sPNET). Other common glial tumors in adults are oligodendrogliomas, oligoastrocytomas, and astrocytomas, which often show characteristic molecular signatures, such as LOH of chromosomes 1p and 19q and an isocitrate dehydrogenase (IDH) gene mutation. However, unlike glioblastoma, the morphologic boundaries of these tumors are subject to considerable interobserver variability. Molecular tests that correlate with prognosis, for example, LOH at various genomic loci (1p/19q; Pten, NF1), IDH mutation status, O6-methylguanine DNA methyltransferase (MGMT) promoter methylation, and others, have improved the stratification of diagnostic and prognostic accuracy (5–7). Genomic and expression studies on large series of gliomas identified tumor signatures correlated with outcome and/or treatment response (8–10). Glial tumors can be experimentally generated by inactivation of tumor suppressor genes, such as p53, together with loss of Nf1, Pten, and Rb, in astrocytes, neural stem cells, and progenitor cells (1–4, 11, 12); by adenovirus-mediated cre expression in the stem/progenitor cells of the SVZ (2, 3); by constitutive cre expression in all stem/progenitor cells using GFAP-cre transgenic mice (3, 4, 12); or by tamoxifen-inducible cre expression (11). The cre expression system influences the cell population that undergoes recombination and thus the type of tumors that originate from a given combination of genetic mutations. In fact, recombination of floxed Rb, p53, and Pten alleles by tamoxifen-inducible GFAP-cre activation in parenchymal astrocytes as well as SVZ stem/progenitor cells exclusively yielded high-grade gliomas (11), while we also observed sPNET when the same combination of genetic mutations was induced by intraventricular Adeno-Cre or Adeno-GFAP-cre injection (2). Here, we have analyzed the spectrum of intrinsic brain tumors arising from the neurogenic zone of the SVZ or from cancer stem cells that were generated and propagated in vitro and introduced into the brain by orthotopic allografting. We analyzed the relationship of the combination of mutations, the morphologic tumor phenotype, and the expression profile. Finally, to validate our model system, correlations with the expression and the phenotype of their human counterparts were analyzed.
Materials and Methods
Combinations of the conditional mouse mutants RbLoxP/LoxP, p53LoxP/LoxP, and PtenLoxP/LoxP, all in a ROSA26loxP/loxP background, were used as described before (2). All mice were in a mixed background of C57 BL/6 and FVB. Further descriptions of the mouse strains and primers for genotyping are given in ref. 13, RbLoxP/LoxP and p53LoxP/LoxP mice in ref. 14, and R26RLoxP/LoxP reporter mice are described in ref. 15. Animals were kept according to institutional and UK Home Office guidelines (Project licences 70/5540 and 70/6603).
Stereotaxic injection of adenovirus
The cre adenovirus vector was constructed and propagated as described previously (16). Viral infection of SVZ progenitors was achieved by unilateral stereotaxic injections of 109 plaque-forming units of adenovirus-expressing cre recombinase (in short Adeno-cre) with a 26 G needle attached to a 10-μL gastight Hamilton syringe (Model 1701 RN#80030) in 5 μL PBS, into the left ventricle of compound mutant mice as described and characterized in detail previously (2).
Isolation, propagation, and stereotaxic injection of tumor spheres
Neurospheres were isolated from young adult mouse brains and propagated in serum-free medium based on Dulbecco's Modified Eagle Medium (DMEM)/HamF12 (#D8437; Sigma), and supplemented with B27 (1:50; #17504-044; Invitrogen), epidermal growth factor (EGF) (20 ng/mL; # 315-09; PeproTech), and basic fibroblast growth factor (bFGF; 20 ng/mL #100-18B; PeproTech), as described previously (2). Tumor spheres were generated after one passage in vitro by infecting neurospheres with Adeno-Cre [multiplicity of infection (MOI) of ≥ 5] and passaged three to five times to produce quantities sufficient for orthotopic allografting. Five microliters of the neurosphere suspension was injected into the left striatum of adult mice (bregma; 1.5 mm lateral, 2 mm deep), with a 22 G needle attached to a 25-μL Hamilton syringe (Model 1702 RN#80230).
Histologic examination and immunostaining
Brains were fixed in 10% formalin, embedded in paraffin, cut into 3-μm sections, and stained with hematoxylin and eosin (H&E). A list of all antibodies and antisera is given in the Supplementary Methods. Immunostaining was done on Ventana Discovery automated staining machines (Ventana Medical Systems) following the manufacturer's guidelines, using horseradish peroxidase–conjugated streptavidin complex and diaminobenzidine as a chromogen.
Image capturing and analysis
Histologic slides were digitized on a LEICA SCN400 scanner (LEICA UK) at ×40 magnification and 65% image compression setting. Digital image analysis was conducted using Definiens Tissue Studio 3.6 (Definiens AG Munich, Germany). Full details of the image analysis algorithms and software settings are given in the Supplementary Methods.
Tumor sampling and use of human tissue
Mice that developed clinical signs of intracranial pressure were injected with bromodeoxyuridine (BrdUrd) 2 hours before culling. Brains were taken and assessed for tumor content, and tumors were separated from normal brain under a dissection microscope. One part was processed for formalin fixation and paraffin embedding and the other part was snap-frozen. Tumor identity and content were confirmed on histologic sections. The use of human tissue was ethically approved (Reg. Number NHNN/08/H0716/16) and the storage of human tissue was according to the Human Tissue Act (UK; Human Tissue Authority License for UCL/ION 12054s).
RNA extraction and microarray hybridization
RNA was extracted from frozen tumor fragments using TRIzol (Invitrogen). RNA analysis was conducted using Affymetrix Mouse Exon 1.0ST microarrays in the University College London/Institute of Child Health (UCL/ICH, London, United Kingdom) genomics core facility. Samples were processed following the Affymetrix GeneChip Whole Transcript Sense Target Labeling Assay Manual (P/N 701880 Rev.4). Following rRNA removal (Invitrogen; cat# K1550-02), samples were amplified, fragmented, and labeled using the GeneChip Whole Transcript Sense Target Labeling and Control Reagent Kit (Affymetrix; cat# 900652), and hybridized to the arrays. Chips were stained using the GeneChip Fluidics Station 450 and scanned with the GeneChip Scanner 3000 7G. Mouse exon array data are available in Gene Expression Omnibus (GEO) under GSE42515. sPNET data, obtained using identical methods, are available under GSE19404. For atypical teratoid/rhabdoid tumor (AT/RT) data, Illumina HT12_v3 beadchip data generated from mixed brain tumor samples and controls were exported unnormalized from GeneSpring software and imported into Partek GS 6.6 as described in ref. 17.
RNA was extracted from tissues or cells, reverse transcribed, and subjected to PCR using three p53-specific primer sets, followed by gel-extraction of the PCR products and sequencing with the appropriate primers (see Supplementary Data). Fluorescent sequence traces were analyzed, assembled, and compared with National Center for Biotechnology Information (NCBI) reference sequences using the CLC Main Workbench 6.
Statistical and data analysis
Statistics internal to R, gene set enrichment analysis (GSEA), and Partek applications were used for microarray analysis. Array data were obtained from own hybridization experiments or downloaded from GEO and other locations. Subsequent data analyses, including hierarchical clustering and principal component analysis (PCA) and generation of tumor-specific gene sets, were conducted with Partek GS workflows with GeneChip Robust Multiarray Averaging (GC-RMA) quantile normalization. In cases where expression differences were extreme, that is, when comparing in vitro and in vivo samples, quantile normalization was not used. Pearson correlation plots and tab-delimited annotated expression sets for GSEA were generated using Affymetrix Expression Console with GC-RMA quantile normalization. Unless otherwise stated, all P values and false discovery rates (FDR) are multiple-testing corrected according to Benjamini and Hochberg (18). GSEA was conducted using software downloaded from the Broad Institute (Cambridge, MA; http://www.broadinstitute.org/gsea/index.jsp; refs. 19, 20). Single sample GSEA (ssGSEA; ref. 21) was conducted using the ssGSEA projection module at the Broad Institute (http://www.broadinstitute.org/cancer/software/genepattern; ref. 22).
Experimental primary CNS tumors correspond to known human tumor entities and express stem and progenitor cell markers
To generate primary brain tumors, we injected Adeno-cre into the lateral ventricle of mice carrying “floxed” double-mutant (in short: p53/Pten, p53/Rb, and Pten/Rb) or triple-mutant (p53/Pten/Rb) alleles. The lineage relationships to neural stem/progenitors were studied with a panel of markers against Glial Fibrillary Acidic Protein (GFAP; B-type stem cell; astrocyte; ref. 23), platelet-derived growth factor receptor-α (PDGFRα; radial glia-like cell; B-Stem cell; ref. 24), Nestin (radial glia and transient amplifying cells, C-type cell; ref. 25), Sox2 (26), Olig2 (27), and Doublecortin (transient amplifying and neuroblast, C; A-cell; ref. 28). CD15 (29), MAP2, Synaptophysin, and NeuN are expressed in maturing neurons. p53/Pten mice developed a spectrum of gliomas, including anaplastic astrocytomas, oligodendrogliomas, and, most commonly, anaplastic oligoastrocytomas, but no glioblastoma with palisading necrosis or microvascular proliferation. These gliomas expressed the stemness markers GFAP, PDGFRα, Nestin, and the transient amplifying markers Sox2, Olig2, Doublecortin, and CD15, whereas markers of neuronal differentiation, such as synaptophysin and NeuN, were not expressed (Fig. 1A). This expression pattern is similar to that of human anaplastic oligoastrocytomas (Fig. 1D). The quantification of marker expression is summarized in Fig. 2. Additional deletion of the Rb gene (p53/Pten/Rb; Supplementary Fig. S1) or deletion of p53/Rb only (Supplementary Fig. S2) causes predominantly poorly differentiated, well-circumscribed neoplasms, with a morphologic similarity to human sPNET. sPNETs in p53/Rb (Fig. 2A and Supplementary Fig. S2A) or in p53/Pten/Rb mutants (Fig. 2A and Supplementary Fig. S1A) were negative for stemness (GFAP, PDGFRα, and Nestin) or transient amplifying markers (Sox2, Olig2, and Doublecortin), in keeping with their primitive neuroectodermal phenotype. CD15 was expressed in a subset of these tumor cells, in keeping with previous reports in medulloblastoma (29), whereas MAP2 and synaptophysin were consistently strongly expressed in all sPNETs (Supplementary Figs. S1A and S2B), in keeping with the profile of human counterparts (Supplementary Fig. S2D). Mice with the underlying combination of mutations Rb and Pten developed no tumors (0 tumors in 20 mice after up to 440 days; Supplementary Figs. S3 and S4).
The growth factor–rich environment of tumor spheres modulates the tumor phenotype
SVZ-derived stem/progenitor cells isolated from double- and triple-mutant mice were recombined in vitro to form tumor spheres and were subsequently allografted to generate tumor grafts. To understand the role of the growth factor–enriched medium on the modulation of the tumor phenotype, we analyzed the profiles of tumor spheres and tumor grafts. Unlike the distinctive genotype–phenotype correlation of SVZ-derived primary tumors, spheres of all four genotypes [i.e., p53/Pten (Fig. 1B), p53/Pten/Rb (Supplementary Fig. S1B), p53/Rb (Supplementary Fig. S2B), and Pten/Rb (Supplementary Fig. S3B)], as well as nonrecombined controls, expressed GFAP, PDGFRα, Nestin, Olig2, Sox2, and Doublecortin at high levels, whereas markers of mature neurons, synaptophysin, and NeuN were only rarely expressed. This indicates that in vitro–recombined tumor spheres favor an expansion of progenitor cells akin to stem/transient amplifying cells (Fig. 2A) and suggests that the culture of spheres in growth factor–enriched medium overrides the effect of the genetic mutations (Fig. 2A), possibly by the preferential growth of a particular cell type. Grafted p53/Pten spheres recapitulate the primary tumor phenotype, forming infiltrative oligoastrocytomas with an expression profile indistinguishable from primary p53/Pten gliomas and similar to that of tumor spheres (Fig. 1A and C). p53/Rb and triple-mutant p53/Pten/Rb tumor grafts show dual differentiation, with a stem/progenitor cell profile (Sox+, Olig2+, Dcx+, and GFAP+; Supplementary Fig. S1C and S1D) adjacent to primitive neuroectodermal areas with little glial differentiation but strong MAP2 and synaptophysin and moderate NeuN expression (Supplementary Fig. S1D). Importantly, in vitro propagation allowed growth of tumors from Pten/Rb tumor spheres [n = 12 of 17 mice (70%); Supplementary Figs. S3C and S3D and S4], with a morphology resembling human glioblastoma with pseudopalisading necrosis, vascular endothelial proliferations, and expression of stem/progenitor and glioma markers (GFAP+, Nestin+, Doublecortin+, Sox2+, Olig2+, and MAP2+; Supplementary Fig. S3C and S3D). This genotype is distinguished from the other tumors by the presence of intact p53, confirmed by sequencing (Supplementary Fig. S8), suggesting that p53 may exert a tumor-suppressive function in a Pten/Rb–mutant background in the context of an intact SVZ, which can be overcome in an in vitro environment with growth factor–rich conditions. Our data suggest that in vitro passage may select for a cell population with stem/progenitor properties, which continues to be prevalent in tumor cells expanding in a mature central nervous system (CNS) environment.
mRNA Expression profiling of primary brain tumors, in vitro tumor spheres, and tumor grafts suggests a critical role of in vitro culture on tumor phenotypes
The predominance of a glial phenotype of grafts derived from tumor spheres prompted us to compare the transcriptome of primary tumors, in vitro growing tumor spheres, and tumor grafts using the Affymetrix Mouse Exon 1.0ST microarrays (Fig. 3A). We initially compared the global expression profiles of all samples with each other, including publicly available control samples including normal brain that had been hybridized to the same array platform (Fig. 3A). A Pearson correlation plot shows a marked segregation of expression profiles of cultured spheres compared with any of the in vivo tumors (Fig. 3A), indicating that cultured cells express genes that are very different from experimental tumors and controls, but shows striking similarity between all cell cultures independent of the underlying genetic mutation, with differences that did not exceed an adjusted P value/FDR of 0.05. Further comparison within the profiles of individual cell cultures showed that cultures and tumors showed more than 50% of genes were expressed differentially (FDR < 0.05; Fig. 3B). We analyzed the top 2,000+ changed genes (442 “upregulated” in vivo; 1,689 “downregulated” in vivo; Supplementary Table S1) for enrichment in Gene Ontology (GO) terms using the online tool Database for Annotation, Visualization and Integrated Discovery (DAVID; ref. 30) and found that the most significantly enriched GO terms belonged to cell culture–related pathways, that is, growth factor–induced cell-cycle genes and oxidative stress. This indicates that differences in oxygen concentration and abundance of growth factors in the cultures were the factors most likely to cause this differential gene expression, overriding any mutation-specific effect. A saturation of these pathways in cultured cells could explain why the cultures did not express any mutation-specific signatures to achieve genome-wide significance when compared with each other, regardless of the combination of genetic mutations. It is also unlikely that host-derived vascular endothelium or stromal cells significantly contribute to these differences between tumors and spheres, as we have previously shown that tumor preparations are largely devoid of contaminating cells (2). Comparison of primary and secondary tumors shows a close match with a genome-wide difference of less than 10% across all genotypes (Fig. 3B). Even the difference within matched genotypes was reduced to below 1% to 3% depending on the genotype. This highlights the importance of in vitro cultures in selecting a specific cell type or by modulating expression in a given cell population.
Hierarchical clustering of global expression profiles corresponds to tumor morphology and suggests common patterns of experimental high-grade gliomas, overriding initiating mutations
Given the predominance of glial signatures upon in vitro propagation, we next analyzed whether histologic appearance or tumor-initiating genotype correlated better with the transcriptome. We analyzed oligoastrocytomas, sPNETs, and the “dual-differentiated” tumors by hierarchical clustering our samples and control brains (Fig. 3C). Normal brain (Fig. 3C, NB, right column) differed markedly from all tumor samples, and spheres of all genotypes (column 4 in Fig. 3C, orange box) clustered together as assessed by the Pearson's analysis (Fig. 3A and B). The yellow box within the sphere cluster highlights nonrecombined control spheres, confirming that the cell culture environment overrides any other differences in the transcriptome (Fig. 3C). Comparison of the histologic tumor entities by unsupervised hierarchical clustering segregates sPNETs away from glial tumors, and segregates the latter into oligoastrocytomas and glioblastomas. Importantly, the three p53/Rb grafts fall within the glioma cluster (top of Fig. 3C, large black box), showing that their morphologic “glioma” phenotype corresponds to an expression pattern akin to p53/Pten gliomas rather than to the genetically matched, but morphologically distinct p53/Rb sPNET. Strikingly, a similar expression pattern was also found in a rare SVZ-derived p53/Rb glioma (top of Fig. 3C, small black box). PCA on the 36 arrays (Fig. 3D) shows segregation into a cluster with tumor spheres (Fig. 3, orange spheres), and solid tumors (purple, red, and blue spheres). Separate analysis of primary and grafted tumors clustered them according to their histologic appearance, overriding their different genotypes. Tumors with dual differentiation (all grafted) clustered closer to the gliomas (Fig. 3E). These PCAs (Fig. 3D and E) underpin the effects of in vitro propagation, which overrides the influence of the primary genetic mutation, and further indicate that the histologic morphology more closely corresponds to the global expression pattern of the tumor than its initiating genotype, as also suggested from human tumors, where, for example, the 1p/19q status of oligodendrogliomas or the IDH mutation status of diffuse astrocytomas does not impact on their histologic appearance.
The expression profiles of experimental gliomas resemble those of a subset of human glioblastoma
Next, we determined whether the distinct experimental tumor types corresponded to a glioma subclass of published gene datasets. Using GSEA (19), we compared the expression profiles of our experimental tumors with the four glioblastoma subtypes (9) of The Cancer Genome Atlas (TCGA) and three glioma subtypes identified by Phillips and colleagues (Fig. 4A; ref. 8). We ensured maximal reproducibility by using a previously published expression dataset (kindly provided by C. Qu and S. Baker, St. Jude Children's Research Hospital, Memphis, TN; ref. 11). Comparison of “oligoastrocytomas” to the remaining tumor phenotypes (i.e., PNET+GBM; labeled “The Rest” in Fig. 4A), showed that experimental oligoastrocytomas did not significantly resemble any of the human glioblastoma subtypes (all FDR > 0.25; Fig. 4A, column 1), but there was a statistically nonsignificant resemblance to the Phillips proneural and TCGA classical glioblastoma. Instead, PNET+GBM showed a significant correlation to the Phillips proliferative subtype (FDR, 0.08; see later). Glioblastoma (orange headers in Fig. 4A) are more similar to oligoastrocytomas than to sPNET. The “dual-differentiation” of grafted p53/Pten/Rb neurospheres is also evident from their heatmap profile (yellow box in Fig. 4A, column 1). By grouping the tumors by genotype (Fig. 4A, column 2), p53/Pten tumors significantly resemble TCGA classical, “Phillips” proneural (FDR, 0.11), and TCGA neural (FDR, 0.12). Finally, by comparing murine sPNET with the other tumors, they show a significant correlation of murine sPNET to the Phillips proliferative subtype (FDR 0.06; Fig. 4A, column 3). Again, the heatmap shows that all grafted samples, including murine glioblastoma, show marked resemblance to the expression profile of the p53/Pten–mutant tumors (oligoastrocytomas). There was no obvious change in the gene order compared with the analysis according to the histologic grouping conducted earlier (Supplementary Table S3).
The expression profiles of experimental gliomas are dissimilar to human oligodendrogliomas
As our p53/Pten–mutant gliomas resemble a spectrum from astrocytomas to oligodendrogliomas, we compared their expression profile to human oligodendrogliomas, using Partek quantile normalized and RMA background-corrected array data from GSE9385 (31). Expression profiles of oligodendroglioma were compared with those of control brain and other glioma samples within the dataset and an expression set comprising genes with at least a 2-fold expression difference in (human) oligodendroglioma versus control brain or “other glioma” were identified (Supplementary Table S4). GSEA analysis containing 454 samples from the Rembrandt project (32) showed that the oligodendroglial gene set shows a highly significant correlation with the oligodendroglial samples (FDR, 0.01; Supplementary Fig. S5), whereas our experimental murine gliomas did not show significant overlap with the oligodendroglial expression set.
The expression profiles of experimental sPNETs resemble human AT/RT but not medulloblastoma or sPNET
Our experimental supratentorial PNET appear histologically similar to human sPNET, a heterogeneous group of aggressive supratentorial tumors of children and adolescents, composed of undifferentiated neuroepithelial cells, which may express synaptophysin (33) and NeuN and CD15 (29). We compared our experimental sPNET with experimental medulloblastomas arising from Ptch+/− external granule layer (EGL) neuron precursor cells (GSE17702; ref. 34) by hierarchical clustering (Fig. 4B). Neither our cultures (Fig. 4B, black box), nor our sPNET (Fig. 4B, green box) resembled experimental medulloblastomas or granule neuron precursor cells (Fig. 4B, purple boxes; ref. 34), indicating that experimental sPNET are distinct from morphologically similar experimental medulloblastoma. Next, we compared the murine sPNET with a small set of histologically similar human sPNET (Supplementary Fig. S2D) prepared on the Affymetrix HG-U133+2 arrays encompassing 10 human sPNET, 1 pineoblastoma, and 1 atypical AT/RT. A sPNET-specific gene set was derived by comparing the human sPNET with datasets (GSE21687) of ependymomas (35) and GSE15824 (gliomas including glioblastoma and normal brain; ref. 36). Differentially expressed genes with more than 2-fold expression difference were identified, yielding a sPNET-specific expression set containing 1,143 genes (Supplementary Table S2). The gene set was successfully tested by conducting a PCA and hierarchical clustering of ependymal samples, gliomas, and our samples (Supplementary Fig. S6A and S6B). A recent study (37) analyzed 51 pediatric sPNET expression profiles and found three clusters, one of which corresponded closely to their previously published sPNET with amplified C19MC locus (38). The other two groups (“oligoneural,” “group 2” and “mesenchymal,” “group 3”) overlapped with each other. We derived three gene sets from the study, encompassing fewer than 15 genes each and used these for a GSEA of our sPNET. The experimental sPNET correlated best (FDR, 0.15) with the mesenchymal subclass (group 3) of sPNET. All other comparisons were nonsignificant.
One further form of primitive brain tumors is the AT/RT. We collected an independent dataset of tumors that had been profiled on Illumina HT12_v3 beadchip arrays, including 5 AT/RT, control brain samples, 17 pilocytic astrocytomas, and 15 gliomas (17). An AT/RT-specific set of 345 genes was derived and GSEA on all experimental tumors or selectively on primary tumors was conducted using sPNET (Supplementary Fig. S6) and AT/RT (Supplementary Fig. S7) expression sets. Tumor-type specific gene sets were derived by contrasting each tumor type against “other brain tumors” and against “normal tissue” array samples. There was a strong correlation (FDR, 0.04) of experimental sPNET with AT/RT (Supplementary Fig. S5C), which was reduced when tumor grafts were included (FDR, 0.09), whereas no correlation with sPNET was seen. The tumor suppressor SMARCB1/INI1, a gene specifically downregulated in AT/RT, is underexpressed in all human AT/RT samples as well as in human sPNET, compared with controls, whereas human gliomas tend to overexpress INI1. Instead, experimental sPNET express normal levels of this transcript, probably due to a different molecular pathogenesis of these entities.
In this study, we have analyzed the relationship of primary murine intrinsic brain tumors, anaplastic oligoastrocytomas, glioblastoma, and primitive neuroectodermal tumors with their human counterparts. We have identified similarities between the transcriptomes of distinct human brain tumor entities, specifically glioblastomas of the TCGA classical subtype, and oligodendroglial tumors on the one hand, and Phillips proliferative gliomas and AT/RT on the other hand. Our key findings are (i) that the genotype and phenotype can be strongly modified in vitro by growth factors, characterized by a shift toward a stem/transient amplifying phenotype in vitro; (ii) the correlation of experimental gliomas with the TCGA classical subtype of human glioblastomas (Fig. 4A); (iii) an experimental correlation of human glioblastomas with PNET component (Supplementary Fig. S1E and S1F); and (iv) a distinct expression profile shared by experimental primitive neuroectodermal tumors and human AT/RT (Supplementary Fig. S2A and S2D and Fig. 4C).
As reported previously, there is a strong correlation between the initial genetic mutation and the tumor phenotype. The different efficacies of tumorigenesis (Supplementary Fig. S4) can be explained by a differential susceptibility of the targeted SVZ stem/progenitor population by growth-promoting signals. p53/Rb inactivation in the SVZ is relatively ineffective to generate tumors, whereas it is very effective when occurring during granule cell amplification in the developing cerebellum (14). Other cell types, for example, cortical astrocytes and neurones, are not susceptible to neoplastic transformation at all (2). Identical primary mutations in the SVZ stem cell compartment can give rise to a spectrum of gliomas, ranging from anaplastic astrocytic to oligodendroglial tumors. There is recent evidence that a defined mutation can give rise to a spectrum of gliomas also in humans, often causing diagnostic ambiguity (39, 40). Tumors caused by a mutation in the IDH gene encompass a spectrum from astrocytomas to oligodendrogliomas (41). At the same time, there is increasing evidence that a morphologic subclassification of gliomas lacks clinical or prognostic relevance (42), and there is growing support for an approach combining morphologic assessment and molecular profiling of gliomas (31).
In vitro cell cultures derived from primary tumors remain a mainstay in human brain tumor studies and are used to study signaling pathways or drug responses (43, 44). To generate xenografts, primary tumors are commonly subjected to a number of in vitro passages, which have been shown to render them to a potential selection bias (45), whereas for example, MGMT methylation was found to be similar in primary tumors and glioma spheres (46). We compared the transcriptome and protein expression in corresponding genotypes in genetically defined cells, and a controlled and reproducible environment. Although primary tumors show a robust relationship between initiating mutations and tumor phenotype, propagation of CNS stem/progenitor cells in growth factor–enriched conditions overrides genetic mutations and promotes a glial signature (Figs. 2 and 3C and D). These data can be interpreted as follows: (i) in vitro cultured cells are likely to be saturated by growth factors essential for propagation of these cells; (ii) genomic alterations are not predictive for in vitro transcriptome expression profiles; and (iii) the in vitro conditions are selective for a subgroup of tumor stem cells that express a stem/progenitor transcriptome, and share features with gliomas (Figs. 1 and 2A). Expansion of in vitro recombined cancer progenitor cells in an environment similar to that of primary brain tumors partly reverts this phenotype, but still with a significant shift toward a stem/progenitor or glioma phenotype (Figs. 2 and 3B and D), possibly by a reversal to a mature CNS environment where different populations of cancer stem cells have a growth advantage. An important finding is the inability to generate brain tumors by inactivating Pten and Rb alone in SVZ progenitor cells. This is a contradiction to previous studies (47, 48) where gliomas were generated by inactivation of Rb family members in mice expressing GFAP-T121, a truncated SV40 Large T antigen under the control of the GFAP promoter in a cre-inducible fashion and where additional Pten but not p53 alleles (47) were deleted. The discrepancy with our findings can be explained by an activation of the Rb pathway in a wider range of cells or by inactivation of several Rb family members in the GFAP-T121 model (47–49). The vast majority of studies with genetically modified mice showed that p53 suppression is required for brain tumor initiation and propagation (1, 3, 4, 11, 12, 50). No p53 mutations are detectable in Pten/Rb cells or grafted tumors (Supplementary Fig. S8). Pten/Rb tumor grafts strongly express olig2, offering a possible explanation for a selective downregulation of p53 function in the absence of a genetic mutation (51).
The genome-wide expression data confirm that experimental gliomas resemble human glioblastoma subtypes. Two large studies (8, 9) used overlapping terminology for distinct profiles. Thus, clustering of human glioblastomas using either gene set will often place a given glioblastoma in a cluster with a different name. Thus, the murine glioma fall in the “Classical” cluster as defined by Verhaak and colleagues (9), while also resembling the “Mesenchymal” cluster as defined by Phillips (8). A morphologically distinct glioma phenotype with a biphasic pattern with glial (GFAP+, Nestin+, PDGFRa+, Doublecortin+, Sox2+, and Olig2+) and separate, PNET-like component with synaptophysin and NeuN expression was observed in p53/Rb and p53/Pten/Rb secondary tumors, suggesting dual differentiation with glial and PNET components (Supplementary Fig. S1C and S1D). This heterogeneity can be explained by the grafting of tumor spheres that share an initial mutation and then diversified during separate clonal expansions in vitro. Such PNET-like components in the context of glial tumors have been increasingly recognized in human high-grade gliomas, in particular those that arise through progression from lower-grade astrocytomas (52) and could be confirmed in a series of 5 GBM–PNET from our archive (Supplementary Fig. S1E and S1F). In this context, our model system supports the notion of the PNET component as part of an expansion of a dedifferentiated subclone in a secondary glioblastoma.
Although there is little resemblance of experimental and human sPNET expression profiles, there are significant similarities to human AT/RT (Supplementary Fig. S7), even though the experimental tumors do not replicate the loss of SMARC/INI1. Thus, our model suggests that both primitive primary CNS tumors and gliomas are likely to originate from the same stem/progenitor compartment in the CNS. A recent study of profiling supratentorial childhood sPNET (37) reveals three subgroups, group 1 being enriched for markers of embryonic or neural stem cells, group 2 tumors showing an upregulation of markers of “oligoneural” differentiation (53), and group 3 tumors showing upregulation of epithelial and mesenchymal differentiation genes. In this context, it is important to note that despite a number of parallels between our model system and human gliomas, there are a number of limitations in this and similar other model systems. To date, there are no robust data to suggest from which cell type and, importantly, from how many cells a human brain tumor arises. Instead, all genetically engineered mice (GEM) have in common that they require a “critical mass” of recombined cells, usually with initial mutations of multiple genes, to develop tumors. Despite significant advances in refining the population of cells, there are also limitations to the specificity of targeting stem progenitor cells in GEM (54, 55). Because of the simultaneous recombination of cells with identical mutations, GEM may not reflect all aspects of the biology of human gliomas, which are known to show significant intratumoral heterogeneity.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: N.V. Henriquez, T.S. Jacques, R. Grundy, S. Brandner
Development of methodology: N.V. Henriquez, D. Sheer, S. Brandner
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.V. Henriquez, T. Forshew, R. Tatevossian, H. Rogers, T.S. Jacques, P.G. Reitboeck, K. Pearce, D. Sheer, R. Grundy, S. Brandner
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.V. Henriquez, M. Ellis, P.G. Reitboeck, S. Brandner
Writing, review, and/or revision of the manuscript: N.V. Henriquez, T. Forshew, R. Tatevossian, M. Ellis, H. Rogers, P.G. Reitboeck, D. Sheer, R. Grundy, S. Brandner
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.V. Henriquez, T. Forshew, R. Tatevossian, M. Ellis, A. Richard-Loendt, D. Sheer
Study supervision: S. Brandner
This study was supported by funding from The Brain Tumour Charity (UK; formerly Samantha Dickson Brain Tumour Trust), The Brain Research Trust UK, and Ali's Dream Charitable Foundation. N.V. Henriquez was a Brain Research Trust Senior Research Fellow.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The authors thank the UCL IQPath Histology facility for excellent histologic assistance and for digitizing histologic specimens. We thank all patients and their families for their generous support of our research by donating brain tumor tissue. The authors also thank Dr. Silvia Marino (Queen Mary University of London) for her helpful comments on the article.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received May 3, 2013.
- Revision received June 17, 2013.
- Accepted July 18, 2013.
- ©2013 American Association for Cancer Research.