|Home | About | Journals | Submit | Contact Us | Français|
Mutations in the PTEN, TP53 and RB1 pathways are obligate events in the pathogenesis of human glioblastomas. We induced various combinations of deletions in these tumor suppressors in astrocytes and neural precursors in mature mice, resulting in astrocytomas ranging from grade III to grade IV (glioblastoma). There was selection for mutation of multiple genes within a pathway, shown by somatic amplifications of genes in the PI3K or Rb pathway in tumors in which Pten or Rb deletion was an initiating event. Despite multiple mutations within PI3K and Rb pathways, elevated Mapk activation was not consistent. Gene expression profiling revealed striking similarities to subclasses of human diffuse astrocytoma. Astrocytomas were found within and outside of proliferative niches in the adult brain.
High-grade gliomas (HGGs), which are WHO grade III and IV tumors, have a uniformly poor outcome with the most aggressive form, glioblastoma, a grade IV diffuse astrocytic tumor, bearing a long term survival of less than 10% (Louis et al., 2007). Animal models that more accurately reflect HGG biology are needed to advance our understanding of this disease as well as to allow for more accurate preclinical testing of new treatment options.
The molecular pathogenesis of glioblastoma has been studied in detail, including high resolution microarray analyses to identify copy number imbalances and gene expression signatures as well as in-depth sequence analyses to identify recurrent targets of mutation. These studies concluded that glioblastoma formation requires dysregulation in three core pathways: the receptor tyrosine kinase (RTK)/phosphatidylinositol 3′-kinase (PI3K)/AKT axis, p53 signaling and RB-mediated control of cell cycle progression (TCGA, 2008; Parsons et al., 2008). Genetic aberrations affecting intermediates of all three pathways are identified in virtually all glioblastomas. For the RTK/PI3K pathway, biallelic inactivation of PTEN through mutation coupled with LOH of chromosome 10q is a common mechanism in glioblastoma to abrogate the major negative regulator restraining PI3K activation. The most common RTK target of mutation is amplification of EGFR, often coupled with intragenic deletion resulting in a constitutively activated form. P53 signaling is dysregulated frequently by bi-allelic inactivation of TP53, or in some cases by HDM2 amplification. Homozygous deletions of CDKN2A simultaneously remove p14ARF, a negative regulator of p53 signaling, as well as INK4A, a key regulator of the RB-mediated G1 checkpoint, thus impacting both p53 and RB pathways. Mutations in RB1 are also found in glioblastoma (TCGA, 2008; Parsons et al., 2008).
The effects of Pten, Rb1 or Tp53 deletion in the mouse brain have been described through the use of mouse lines expressing Cre specifically in the brain. For the most part, the phenotype obtained is cell-type and developmental-stage specific. For example, deletion of Pten in neural stem and progenitor cells resulted in expansion of the stem cell niche, a proliferative advantage in the progenitor cell population as well as migration defects causing a disruption in the laminar structure of the cortex (Gregorian et al., 2009; Groszer et al., 2001). In contrast, ablation of Pten in postmitotic cerebellar neurons resulted in cellular hypertrophy (Backman et al., 2001; Kwon et al., 2001; Marino et al., 2002). Pten-null astrocytes showed evidence of hypertrophy and proliferation (Fraser et al., 2004; Wei et al., 2006) and abnormal migration (Yue et al., 2005). In the case of Rb1, conditional ablation at E9 in the telencephalon or the whole brain resulted in cell-cycle defects characterized primarily by increased proliferation (Ferguson et al., 2002; MacPherson et al., 2003). Conditional deletion of Tp53 combined with haploinsufficiency of Nf1 and Pten, or deletion of Nf1 generated high-grade astrocytomas (Alcantara Llaguno et al., 2009; Kwon et al., 2008; Zheng et al., 2008). In contrast, an intragenic deletion of Tp53 encoding a mutant protein, induced in embryonic brain, was sufficient to cause malignant astrocytomas in mid-adulthood (Wang et al., 2009).
While the exact histogenesis of gliomas is still a matter of some controversy, there is agreement that neural progenitor cells or mature astrocytes are likely candidates (Louis et al., 2007). We recently reported the characterization of GFAP-CreER mice with inducible targeting of approximately 50% of mature astrocytes, and less than 1% of neural precursor cells located in the subventricular zone (SVZ) and subgranular zone (SGZ) of the dentate gyrus (Chow et al., 2008). Here, we used this GFAP-CreER mouse line to study the tumor suppressor function of Pten, Tp53 and Rb1 in the development of high-grade astrocytomas (HGAs) in mature brain, as this is the context in which the majority of glioblastomas arise.
We generated GFAP-CreER™A (hereafter referred to as CreER); PtenloxP/loxP mice to investigate the effect of Pten loss in mature astrocytes. Cre activity was transiently induced in developmentally mature mice by tamoxifen administration after postnatal day (P) 21 (range P20–45). Recombination and inactivation of Pten were confirmed by real-time PCR (Figure S1A). Conditional deletion of Pten in mature brain did not reduce survival (Figure S1B) or disrupt brain architecture (Figures 1A–C). There was a modest increase in Gfap expression in the brains of Pten conditional knockout (cKO) mice, (Figures 1A–F, S1C, lanes 7–9), a feature also observed in primary cultures of Pten-null astrocytes (data not shown). However, we did not detect astrocyte hypertrophy or hyperproliferation (not shown).
Pten is the central negative regulator of the PI3K/Akt pathway (Chow and Baker, 2006). In rare Pten-null granule neurons of the dentate gyrus that were targeted by this CreER strain, we detected the expected increase in phosphorylation of Akt (pAkt) at S473 or phosphorylation of S6 (pS6) at S235 and S236 indicating hyperactivity of the PI3K/Akt pathway (Figure S1D–I, arrows). However, western blot analyses of cortical lysates and immunohistochemistry (IHC) did not detect any perturbations in pathway signaling induced by Pten deletion in astrocytes (Figure 1G–I, Figure S1C). Therefore, inactivation of Pten in adult astrocytes did not induce detectable activation of PI3K signaling, and accordingly did not lead to an overt phenotype or tumor predisposition.
Highly penetrant HGAs were induced in CreER; PtenloxP/loxP; Tp53loxP/loxP compound mice, hereafter termed Pten; p53 cKO mice. Cre activity was induced in adult mice as above. Pten; p53 cKO mice had a significantly reduced survival compared to Pten cKO mice, or to PtenloxP/+; p53 cKO mice (Figure 2A). 55/63 (87%) of Pten; p53 cKO brains harbored tumors within the spectrum of HGA, showing a range of histopathological features. Approximately 25% of tumors were classified as glioblastomas, on the basis that they featured scattered cells with an astrocytic phenotype, mitotic activity, cytological pleomorphism, including in some cases a substantial number of multinucleated giant cells (Figure 2B), necrosis (Figure 2C) and/or microvascular proliferation (Figure 2D). Approximately 25% of tumors were anaplastic astrocytomas with myxoid degeneration, a feature more commonly observed in pediatric gliomas than adult (Figure 2E). Half of the tumors were diffuse anaplastic astrocytomas, demonstrating infiltrating tumor cells (Figure 2F) and a focal astrocytic phenotype. In some cases, there was extensive parenchymal infiltration combined focally with the architectural and cytological features of gliomatosis (Fig. 2G). PCR analysis confirmed inactivation of the Tp53 locus (Figure S2), and Pten IHC showed that tumor cells were Pten-null (Figures 2H–J) highlighting invasion by tumor cells of perivascular spaces (Figure 2I), leptomeninges (Figure 2J) and parenchyma.
Tumors consistently showed elevated pAkt (Figure 3A), and were highly proliferative, as shown by Ki67 IHC (Figure 3B). Active Caspase-3, a marker of apoptosis, was also detectable in most tumors (Figure 3C). Immunophenotyping of the tumors revealed Gfap and S100β immunoreactive tumor cells in varying proportions (Figures 3D and not shown). Neurofilament (light chain) IHC was negative in tumor cells (Figure 3E). Nestin, a marker of neural progenitor cells (Brustle and McKay, 1995), was also expressed in tumors (Figure 3F).
The location of tumors within the CNS included all regions of the forebrain, midbrain, hindbrain and spinal cord, and did not correlate to different histopathological features. Many tumors were contiguous with regions rich in neural progenitor cells, such as the SVZ, the rostral migratory stream (RMS) and SGZ of the dentate gyrus (Figure S3 A, B). When tumors were not macroscopically evident, serial sectioning and H&E staining frequently revealed small groups of tumor cells invading through white matter tracts adjacent to zones rich in neural progenitors such as the corpus callosum and internal capsule (Figure S3C–H, arrows). These cells were highly proliferative (Figure S3F–H and insets) and Pten-null (Figure S3K arrow). They may represent early stages of tumor growth or indicate cells in transit towards distant sites. Significantly, 22% (14/63) of tumors developed in sites independent of proliferative niches including the cerebellum, spinal cord or the ventral brain stem (Figures 3G–I) in the absence of any other microscopically detectable tumor or abnormally proliferating cells in the brain including the progenitor niches, eliminating the possibility that they arose by spread from a distant established tumor mass. While it is possible that tumor-initiating cells originating from progenitor-rich regions populated these tumors, we also identified early lesions and hyperproliferative cells in the cerebellum of brains where no aberrant proliferation in progenitor niches was apparent (Figure S3I, J and inset).
Array comparative genomic hybridization (aCGH) analysis revealed that most tumors carried multiple focal amplifications [copy number (CN) > 3] and deletions (CN < 1.5), large scale copy number gains and losses, and more complex patterns of chromosomal copy number alterations (CNAs). The most frequent focal amplification, present in 24% (12/50) of tumors (10 of these with CN > 3), was localized to chromosome (Chr) 6A2 and included Met (Figure 4A), which encodes an RTK amplified in human glioblastomas, as well as other genes, some with association to HGG (Table S1). Additional focal amplifications involved Egfr and platelet-derived growth factor receptor, alpha (Pdgfra), encoding two RTKs that have been strongly implicated in the pathogenesis of HGG (Furnari et al., 2007). The region including Egfr was focally amplified in four tumors (Figure S4A) while the majority of Chr11 containing Egfr was amplified to a high degree in an additional tumor. Reverse-transcription PCR (RT-PCR) of Egfr transcripts from tumors with or without amplification did not detect truncations analogous to the EGFR vIII mutated variant found in some human glioblastomas (not shown). The Pdgfra gene was contained within regions of high-level focal amplification in two tumors, and regions of focal gain (CN ranging from 2.3–3) in an additional five tumors (Figure S4B). All except one of the Pdgfra amplicons included the gene encoding the RTK Kit. Large-scale gains of chromosome 5 including Pdgfra were found in an additional nine tumors. Most of the amplifications and gains of the three RTK genes described were mutually exclusive although three tumors had concurrent focal amplifications of Met and another RTK. Thus 40% (20/50) of Pten; p53 cKO tumors contained somatic focal CNAs in these three RTK genes.
Gene amplification was confirmed by fluorescence in-situ hybridization (FISH) (Figure 4A, S4A, B). Real-time RT-PCR showed that focal amplification was consistently associated with high expression levels of the targeted RTK (Figure S4C). Several HGAs overexpressed RTKs, most frequently Pdgfra, in the absence of a detectable gene amplification event, as has been observed in human astrocytomas (Furnari et al., 2007). Western blot analysis showed that Met (lanes 7–9) and Egfr (lanes 3, 5) overexpression in HGG was correlated with gene amplification whereas Pdgfr-α was overexpressed in the majority of HGA samples and did not correlate with CN (Figure 4B). As expected for Pten-deficient tumors, levels of pAkt were elevated compared to normal cortex (Figure 3A, ,4B).4B). Interestingly, levels of pMapk T202/Y204 were not consistently altered among tumor samples, suggesting that steady state signaling through the Ras/Mapk pathway was not necessarily elevated in HGA. There was no consistent correlation between RTK overexpression and tumor histology, or with levels of pMapk T202/Y204. Levels of pS6 S235/236 were more variable and tended to correlate with total S6 protein levels (Figure 4B).
To model the concurrent dysregulation of the RTK/PI3K/AKT, p53 and RB pathways found in the majority of HGG (Cerami et al., 2010; TCGA, 2008; Parsons et al., 2008), we generated CreER; PtenloxP/loxP; Tp53loxP/loxP; Rb1loxP/loxP compound mice (hereafter termed triple cKO mice) and induced Cre activity as above. Deletion of Rb1 in addition to Pten and Tp53 significantly reduced the latency of tumor development and time to morbidity from a median of 211 days to 159 days (Figure 5A). 85% (26/31) of triple cKO brains contained HGAs encompassing the same histological variation observed in Pten; p53 cKO mice (Figure 5B and not shown). Deletion of Pten in tumors was confirmed by IHC (Figure 5C) and deletion of Tp53 and Rb1 was demonstrated by PCR (Figure S5A, B). Triple cKO and Pten; p53 cKO tumors showed similar high proliferation rates (Figure 5D and S5C), invasive properties and immunophenotypes (data not shown), and pathway activation, with consistently elevated pAkt at S473 and T308 and variable pS6 S235/236 and p-Mapk (Figure 6).
Array CGH showed a significant difference in the selective pressure for amplifications with no focal amplifications of the RTKs Met and Egfr in triple cKO tumors compared to 30% of Pten; p53 cKO tumors showing focal amplification of one or both of these genes (P=0.003). Large-scale gains of the chromosomes containing these genes were also not consistently associated with RTK overexpression in triple cKO tumors (Figure 6). However, the frequency of Pdgfra amplification was very similar between Pten; p53 cKO tumors and triple cKO tumors (14% in both, P=1), and both sets of tumors showed frequent overexpression of Pdgfra even in the absence of focal amplification (Figures 4 and and6).6). Thus, Rb1 inactivation further cooperated with Pten and Tp53 deletions to generate HGAs with similar histopathological and biochemical signatures, but different selective pressure for RTK amplifications.
In addition to genes encoding the RTKs mentioned previously, other notable genes contained within regions of CNA included amplification of Cdk4, Cdk6, Ccnd1, Ccnd2, and Ccnd3, which directly regulate Rb activity at the G1/S cell cycle checkpoint, as well as loss of Dcc and Smad4. Expression changes in all of these genes have been implicated in glioma biology (Buschges et al., 1999; Kjellman et al., 2000; TCGA, 2008; Reyes-Mugica et al., 1997). 12 of 50 Pten; p53 cKO tumors (24%) had an amplification or deletion in at least one gene that would be expected to promote progression through the cell cycle (Table S2). Analysis of Rb-null tumors showed that three of 21 tumors (14%) also had focal mutations in key cell cycle regulatory genes (P=0.5), although these did not involve the upstream regulators of Rb (Table S2). We observed large deletions of Chr14 that included the Rb1 locus in four of 50 Pten; p53 cKO tumors (8%), however similar deletions of Chr14 were also detected in six of 21 triple cKO tumors (29%) suggesting that Rb1 was not the critical gene targeted by these deletions. Finally, Mycn, which has been proposed to promote cell cycle progression in the context of tumorigenesis (Pession and Tonelli, 2005), was amplified or gained in five (10%) Pten; p53 and four (19%) triple cKO tumors (Figure S6A). Thus the data support the notion that dysregulation of cell-cycle regulation through additional mutations provides additional selective advantages above that provided by Rb1 loss alone.
Overall, large-scale CNAs were well conserved between the two models (Figure 7). The region of Chr12 that is deleted in 54% of Pten; p53 and 52% of triple cKO tumors shares synteny with human Chr14q which is lost in ~25% of glioblastomas. Similarly, partial loss of Chr14, occurring in 20% of Pten; p53 and 43% of triple cKO tumors (P=0.08) is syntenic to both human Chr14q and 13q which undergo LOH in ~30% of glioblastomas. In fact, with the exception of Chr10q (which contains the PTEN locus), Chr14q and 13q losses are among the most frequent in glioblastoma (TCGA, 2008; Parsons et al., 2008). Other significant large CNAs occurring in both models include loss of Chr9 and Chr18 and gain of Chr3 whereas partial deletion of Chr16 was more common in Pten; p53 cKO tumors (P=0.003). One of the more frequent focal deletions noted in triple but not in Pten; p53 cKO tumors involves Lrp1b which was lost in 43% (9/21) of tumors (Figure S6B), including six intragenic deletions. LRP1B deletions, including intragenic deletions, have been reported in glioblastoma (Parsons et al., 2008; Paugh et al., 2010; Yin et al., 2009; Zarghooni et al., 2010) as well as other tumor types (Cengiz et al., 2007; Langbein et al., 2002; Nagayama et al., 2007; Nakagawa et al., 2006; Sonoda et al., 2004). Taken together, the genomic landscape of tumors arising from these murine HGA models are similar to each other and more importantly, resemble human glioblastomas.
Gene expression profiles were analyzed for 24 Pten; p53 and 14 triple cKO HGAs. Unsupervised hierarchical clustering segregated the tumors into three distinct subgroups (Figure 8A; hierarchical cluster (HC) 1 to HC3). Tumors from each cKO background were present in all three subgroups, further demonstrating that similar HGAs arise despite differences in the number of initiating mutations. Grade III anaplastic astrocytomas or pilomyxoid tumors did not cluster with specific subgroups, however glioblastomas were exclusively found in HC2 and 3, not in HC1 (P=0.0001). Notably, 38% (11/29) of HC1 and HC2 tumors, and none of the HC3 tumors had focal amplifications in RTK genes (P=0.04, Figure 8A). Several studies have identified gene expression subgroups in human HGG. Using gene set enrichment analysis (GSEA), we demonstrated highly significant similarity between the expression signatures of the mouse HC1, HC2 and HC3 subgroups with the human HGG expression subgroups that were identified by comparing prognosis, termed Proneural, Proliferative and Mesenchymal, respectively (Phillips et al., 2006) (Figure 8B and Table S3). The Cancer Genome Atlas Project (TCGA) identified four expression subgroups that correlate to some extent with specific combinations of mutations (Verhaak et al., 2010), and there was also significant similarity between signatures in mouse HC1, with the TCGA Proneural and Neural subgroups, HC2 with the Proneural subgroup and HC3 with the Mesenchymal subgroup (Table S3). Single sample GSEA was also used to demonstrate the correlation of each individual mouse tumor gene expression profile with the gene sets for each human subgroup (Figure 8C). Interestingly, 78% (7/9) of the mouse HC3 tumors were exophytic tumors arising from the basal hypothalamus or the ventral pons (Figure 3I, ,8A).8A). To determine if this location was strongly associated with a specific expression signature in human tumors from similar locations, we analyzed the gene expression signatures of six human exophytic gliomas arising from the brainstem along with published data from pediatric high-grade glioma (Paugh et al., 2010) and adult glioblastoma (Lee et al., 2008). Single sample GSEA analysis showed heterogeneity among the human exophytic brainstem gliomas, with 33% (2/6) showing significant similarity to the mesenchymal subclass, while 67% (4/6) showed the most significant similarity to the Proneural subclass (Figure S7). Similarly, we found gene expression signature heterogeneity among 31 Grade IV human diffuse intrinsic pontine gliomas, with a similar proportion showing significant similarity to the mesenchymal subgroup (our unpublished data). Thus the mesenchymal signature found in the majority of mouse exophytic tumors arising from the brainstem or basal hypothalmus was also found in a subset of human gliomas arising from the brainstem.
To examine the cooperativity between tumor suppressors in the triple cKO mice, Pten; Rb1 cKO and p53; Rb1 cKO mice were also evaluated. No brain tumors were detected in 38 Pten; Rb1 cKO brains that were serially sectioned in their entirety, although eight had pituitary tumors, which have been noted in Rb1+/− mice. 13 of 19 (68%) Rb1; p53 cKO mice developed brain tumors with prolonged latency compared to Pten; p53 cKO mice (Figure S5D). Two of these mice had tumors of morphology consistent with PNET, although they did not show robust synaptophysin staining. An additional three mice had tumors within the olfactory bulb with strong resemblance to human olfactory neuroblastoma. Ten of 13 mice with brain tumors had HGA with histopathological features comparable to tumors from the other models (Figure S5E–G and not shown). All but one of the tumors maintained Pten expression (Figure S5H–J). However, pAkt expression was strong throughout the tumor in seven of 13 tumors (54%) (Figure S5K, L), and patchy or lower level staining in an additional three tumors, indicating that the PI3K/AKT pathway is engaged in the majority of Rb1; p53 cKO tumors.
Induced gene deletion in the adult brain reveals selective cooperativity among key tumor suppressors in the developmental context where the majority of HGA arise. The high penetrance of HGA in mice with induced combined mutation of Pten, Tp53 and Rb1 shows that the same core pathways drive gliomagenesis in mouse and human. The mutation spectrum in human glioblastomas has been extensively studied. More than 90% of human glioblastomas are primary glioblastomas that arise de novo without clinical or histological evidence of progression from a pre-existing lesion (Louis et al., 2007). In the absence of clear progression, there is no unambiguous indication whether specific frequent genetic mutations arise preferentially as early or late events. This timing is implied by the analysis of tumor progression for the less common secondary glioblastomas that arise as a consequence of malignant transformation from low grade diffuse astrocytomas. However, there are distinct differences in the molecular pathogenesis of primary and secondary glioblastomas. Most significantly, IDH1 mutations occur in more than 70% of secondary glioblastomas and the lower grade lesions from which they arise, but are rare in primary glioblastomas (Yan et al., 2009). This suggests that primary glioblastomas lack a common initiating mutation found in secondary glioblastoma, and may be driven by alternative initiating events. Thus, the role of common mutations in primary glioblastoma to act as initiating or late events remains unclear.
Using mouse models, we induced mutations in specific key tumor suppressors and evaluated which mutations cooperated to cause glioma, and which additional mutations provided a selective advantage. In our models, none of these tumor suppressor deletions alone induced astrocytomas with high frequency, although Tp53 deletion induced late onset, low frequency astrocytomas similar to a previous study (Zheng et al., 2008) (data not shown). The greatest cooperativity and earliest tumor onset occurred with simultaneous mutation of all three tumor suppressors, consistent with concurrent mutations in these three pathways occurring in virtually all human glioblastomas. Combinations of mutations that included Tp53 induced astrocytomas efficiently. This supports a role for Tp53 inactivation in astrocytoma initiation and is consistent with the high frequency of TP53 mutations and low frequency of RB1 and PTEN mutations in Grade II human astrocytomas (Louis et al., 2007; Wen and Kesari, 2008). The significantly earlier onset of tumors arising from Pten; p53 deletion compared to Rb1; p53 deletion demonstrated that Pten loss cooperated more effectively than Rb1 deletion with p53 mutation. Importantly, we failed to detect any brain tumors arising from mice with targeted co-deletions in Rb1 and Pten. This doesn’t necessarily indicate that these two pathways fail to cooperate in gliomas, but rather indicates that the specific mutation of Pten and Rb1 do not effectively cooperate in the absence of other mutations. In another study, expression of a truncated SV40 T antigen that inactivates all Rb family members induced mouse astrocytomas and cooperated with Pten loss (Xiao et al., 2005). This suggests that inactivation of all Rb family members may be required to initiate gliomagenesis, or that the truncated T antigen exerted additional Rb-independent effects.
PTEN functions primarily by regulating RTK/PI3K/AKT signaling through its lipid phosphatase activity. Loss of Pten in several cell types within the brain during development, including neurons, astrocytes and neural progenitor cells, has been shown to result in Akt activation and proliferation but not tumorigenesis (Fraser et al., 2004; Gregorian et al., 2009; Groszer et al., 2001; Kwon et al., 2001). We found that Pten loss in the mature astrocyte induced elevated Gfap expression, but did not demonstrate features of severe astrocytic gliosis such as hypertrophy, hyperproliferation and abnormal multinucleated astrocytes. Thus, Pten loss may disrupt cellular homeostasis enough to be detected as a cellular stress inducing a low-level astrogliosis response, but it is insufficient to drive proliferation, consistent with the inability to initiate gliomagenesis in the absence of other mutations. Importantly, Pten loss did not substantially stimulate PI3K signaling suggesting that in this context, basal signaling through the RTK/PI3K axis is not sufficient to elevate pathway activity, even in the absence of negative regulation. This is consistent with the cell-type specific feedback regulation that inhibits PI3K signaling in glia, but not in neurons (Chalhoub et al., 2009). Therefore, the net outcome of mutations that deregulate PI3K signaling depends on the cellular and developmental context.
To precisely define a cell of origin for a tumor in a model system, it is necessary to use highly selective methods to induce gene mutations only in specific populations of cells. Using this approach, several recent reports identified neural progenitor cells as the likely cell of origin for HGAs in mouse models and showed that cells outside of the proliferative niche failed to form HGAs with combinations of deletion in Pten, Tp53, Nf1 and Rb1 (Alcantara Llaguno et al., 2009; Holland et al., 2000; Jacques et al., 2010; Wang et al., 2009). In contrast, the goal of this study was to induce widespread tumor suppressor deletion in astrocytes and progenitor cells in the adult brain to test the ability of these mutations to induce gliomas in diverse brain regions. Using GFAP-CreER mice, we induced Cre activity in a much larger population of cells outside of proliferative niches than within the neurogenic zones of adult brain. We found HGAs both within and outside of proliferative niches, and heterogeneity in tumor histopathology was independent of location. Human glioblastomas also often occur in regions not associated with neurogenesis. Our analysis may underestimate the frequency of tumors arising outside of the neurogenic niche, as we evaluated tumors at the time of morbidity, where large tumors that may have arisen in surrounding regions and grown into the neurogenic niche were scored as associated with proliferative zones. The higher frequency of tumors arising in association with proliferative niches is consistent with progenitor/stem cells having a greater propensity for malignant transformation. Indeed, despite much lower Cre activity in neurogenic niches, tumors arose predominantly in areas contiguous with or adjacent to neural stem cell niches in the brain, and we also detected early invasion of proliferative tumor cells through white matter tracts in these regions (Figure S3C–H). Nevertheless, macroscopic HGAs and microscopic proliferative lesions were also found in areas isolated from canonical neural progenitors in mice without detectable hyperplasia or abnormalities in proliferative zones (Figure 3G–I, S3I, J). While the precise cell of origin remains to be determined, these tumors may arise from mature astrocytes (Dufour et al., 2009), locally resident progenitor cells (Laywell et al., 2000; Lee et al., 2005), or progenitor cells that escaped from proliferative niches to populate distant sites.
Recently, Jacques et al. reported that tumors arising from p53; Rb1 or Pten; p53; Rb1 co-deletions induced in subventricular zone cells, presumably neural stem cells and/or progenitor cells, were almost exclusively of the primitive neuroectodermal (PNET) phenotype whereas Pten; p53 co-deletion resulted in gliomas (Jacques et al., 2010). In contrast, all of these combinations of tumor suppressor inactivation induced in adult mice by GFAP-creER gave rise to HGAs, consistent with concurrent mutation of these pathways in human HGAs. None of the tumors arising in triple cKO mice was a typical PNET on the basis of histopathological features, lack of synaptophysin expression and the consistent expression of Gfap (data not shown). The discrepancies between tumor types obtained could be explained by differences in the specific cells targeted by adeno-Cre virus injections into the ventricle compared to GFAP-CreER, the age of the mice at the time of tumor suppressor inactivation, inflammation and wound healing secondary to the intracranial injection procedure, differences in genetic background or differences in the Pten floxed strain used in the two studies.
Genome-wide analysis of CNAs in mouse HGAs revealed strong selective pressure in Pten; p53 cKO tumors for focal amplifications in the RTKs Met, Egfr and Pdgfra, genes that are also well-known amplification targets in human glioblastoma (Furnari et al., 2007). These somatic amplifications in upstream components of PI3K signaling strongly suggest that Pten deletion alone was insufficient to activate the pathway to the levels required to drive gliomagenesis in this model and/or that pathways downstream of RTKs other than PI3K/Akt are required for HGA growth. While RTKs are known to signal through both the Mapk and PI3K pathways, in the mouse HGAs, RTK amplifications were associated with consistently high PI3K signaling, but not consistently elevated Mapk signaling. This demonstrates that the downstream consequences of RTK activation vary in gliomas. Somatic mutations also demonstrated that amplifications of different RTKs were not equivalent. Pdgfra amplification and overexpression was very similar between Pten; p53 cKO and triple cKO tumors, however there were no amplifications of Met or Egfr in triple cKO tumors, where G1 cell cycle checkpoint control was compromised by Rb1 deletion. This implies that the initiating mutations may strongly influence the selective advantage of hyperactivation of specific RTKs. The more uniform involvement of Pdgfra indicates that it may play a unique role relative to other RTKs. Further investigation is needed to delineate the mechanisms driving different selective pressures for RTK amplifications in Pten; p53 cKO and triple cKO tumors. Akt phosphorylation was also consistently elevated in Rb1; p53 cKO tumors, showing a selective pressure for activation of PI3K signaling regardless of whether tumors were initiated by a mutation in this pathway.
Dysregulation of RB signaling leading to G1/S progression appears to be a critical event in gliomagenesis. This pathway was also targeted by somatic amplifications, with Pten; p53 cKO tumors showing amplification of Rb regulators as well as other cell cycle genes and triple cKO tumors showing amplification of pathway components downstream of Rb (Table S2). Again, we noted alterations in multiple pathway members in the same tumor suggesting that pathway mutations cooperate to provide a greater selective advantage.
The murine HGAs reported here are relevant to human disease, reflecting a spectrum of tumor histology and molecular features. In addition to focal amplifications and deletions of known glioma genes, large scale chromosomal losses in regions syntenic to those lost in human HGA were detected. Gene expression signatures from gliomas arising in Pten; p53 or triple cKO mice fell into three subgroups showing significant similarity to the three expression subclasses of human high grade glioma associated with tumor prognosis (Phillips et al., 2006). Importantly, tumor subgroups did not correlate with initiating mutations or tumor histology (Figure 8A). In humans, the mesenchymal subgroup is associated with the worse prognosis (Phillips et al., 2006). Unexpectedly, 78% (7/9) of the mouse tumors with greatest similarity to the mesenchymal subgroup arose as exophytic tumors from the ventral base of the brain or the pons (Figure 8). Conversely, 78% (7/9) of tumors from this location had a mesenchymal gene signature compared to 7% (2/29) of HGAs in all other locations of the central nervous system (p=0.00009), suggesting that the microenvironment or specific cells targeted in this region of brain contributed to tumor phenotype. Taken together, the somatically acquired genomic imbalances and the gene expression signatures of HGAs in Pten; p53 cKO and triple cKO mice indicate that similar selective pressures drive gliomagenesis and generate similar tumors in humans and these mouse models, suggesting that they should be valuable for preclinical testing.
Mice were maintained on a predominantly FVB/NJ background with contributions from 129/SV and C57Bl6. The GFAP-CreER™A (FVB/NJ background) and floxed Pten (back-crossed six generations to FVB/NJ) mouse lines have been previously described (Chow et al., 2008; Suzuki et al., 2001) and were crossed to generate GFAP-CreER; PtenloxP/loxP mice. Controls were PtenloxP/loxP mice injected with tamoxifen or CreER; PtenloxP/loxP mice injected with vehicle. Floxed Tp53 and Rb1 mice were obtained from the Mouse Models of Human Cancer Consortium repository (Jonkers et al., 2001; Marino et al., 2000). All combinations of compound floxed mice were generated by crossing GFAP-CreER; PtenloxP/loxP with Tp53; Rb1 double floxed mice. Tamoxifen (Sigma-Aldrich, St. Louis. MO) was dissolved in corn oil (Sigma-Aldrich) at a concentration of 20 mg/ml at 37°C, filter sterilized and stored for up to 7 days at 4°C in the dark. A 27 g needle tuberculin syringe (Becton Dickinson, Franklin Lakes, NJ) was used for intraperitoneal injections. Tamoxifen (9 mg/40g body weight) was administered daily for three days after P28 (range P25–49). There was no significant difference associated with the age of induction in the time from tamoxifen administration to tumor onset, or in the histology of the resulting tumors. Procedures for all mouse experiments were reviewed and approved by the Animal Care and Use Committee at St. Jude Children’s Research Hospital, and are in compliance with national and institutional guidelines. Mice were monitored daily and euthanized following NIH guidelines.
Mice were perfused with 1 x phosphate-buffered saline (PBS) to exsanguinate tissues. Following dissection, half the brain containing a portion of the tumor, if evident, was fixed by immersion in 4% paraformaldehyde (PFA) in 1 x PBS at 4°C overnight. If no tumor could be visually identified, the brain was divided sagittally and both halves fixed in this manner. The tissue was processed, embedded in paraffin and 5 μm sections were cut. If the presence of a tumor was evident, several slides spanning the tumor were stained with H&E; if a tumor was not grossly apparent, every fifth slide was stained spanning the entire brain. All slides were examined by L.M.L.C. and sections containing tumor or other anomalies were viewed by S.J.B. and D.W.E., a clinical neuropathologist. The second half of the tumor-containing brains was further dissected to separate tumor tissue from normal cortex. Several tubes of each if possible were immediately flash-frozen on dry ice pellets for subsequent nucleic acid and/or protein preparations.
Tumor diagnoses were assigned by D.W.E., a neuropathologist. Microarray analyses were performed using Roswell Park Cancer Institute (RPCI) 6.5K mouse BAC array, and Agilent mouse 244K Oligo array for array CGH, and Affymetrix Mouse Genome 430 2.0 chip for expression profiles.
Snap-frozen surgical samples of Grade II exophytic brainstem gliomas or autopsy samples of normal brainstem were obtained from the St. Jude tumor bank. Histopathology was reviewed by D.W.E. These samples were collected and experiments performed with the informed consent of patients and/or their legal guardians under the approval of the St. Jude Children’s Research Hospital Institutional Review Board.
For categorical statistical analysis, we used two-sided Fisher’s exact test.
We thank M. Dyer for providing mice, J.M. Lahti and V.A. Valentine from the Cancer Center Core Cytogenetics Laboratory for FISH analyses, the Hartwell Center for array CGH and gene expression microarray labeling, hybridization and data acquisition, M. Wang for help with aCGH data analysis, and J. Mitchell and K. Cox for mouse genotyping, and A. Gajjar for support. L.M.L.C. was a recipient of the Jean-François St.-Denis Fellowship in Cancer Research from the Canadian Institutes of Health Research. This work was supported by grants to S.J.B. from the National Institutes of Health (CA135554 and CA096832), and by the Ryan McGee Foundation and ALSAC.
Array data is deposited in Gene Expression Omnibus (GEO) under accession number GSE22927.
The Supplemental Data include seven figures and three tables along with Supplemental Experimental Procedures including details of Immunophenotyping, Western blotting, real-time PCR and FISH, and array data analyses.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.