Insights Gained from Modeling High-Grade Glioma in the Mouse

Abstract

High grade gliomas (HGG) are devastating primary brain tumors with universally poor prognoses. Advances toward effective treatments require improved understanding of pathogenesis and relevant model systems for preclinical testing. Mouse models for HGG provide physiologically relevant experimental systems for analysis of HGG pathogenesis. There are advantages and disadvantages to the different methodologies used to generate such models, including implantation, genetic engineering or somatic gene transfer approaches. This review highlights how mouse models have provided insights into the contribution of specific mutations to tumor initiation, progression, and phenotype, the influence of tumor microenviroment, and the analysis of cell types that can give rise to glioma. HGGs are a highly heterogeneous group of tumors, and the complexity of diverse mutations within common signaling pathways as well as the developmental and cell-type context of transformation contribute to the overall diversity of glioma phenotype. Enhanced understanding of the mutations and cell types giving rise to HGG, along with the ability to design increasingly complex mouse models that more closely approximate the process of human gliomagenesis will continue to provide improved experimental systems for dissecting mechanisms of disease pathogenesis and for preclinical testing.

Introduction

High grade gliomas (HGG) comprise a broad category of molecularly and histologically heterogeneous tumors that collectively represent the most frequent and aggressive primary brain tumors in adults. While the cell of origin remains ambiguous, these tumors are classified into astrocytomas and oligodendrogliomas based on morphological resemblance of the predominant tumor cell type to normal glial subtypes, and further stratified into four grades of increasing malignancy using criteria delineated by the World Health Organization (WHO) (Figure 1). Glioblastoma is the most malignant stage of astrocytoma (grade IV) and is characterized by nuclear atypia, brisk mitotic activity, microvascular proliferation and pseudopalisading necrosis (Louis et al., 2007).

Figure 1.

Schematic illustration of frequent genetic abnormalities and histological features associated with the malignant pathogenesis of oligodendroglial and astrocytic HGG. Multiple cell types are candidates for cell of origin, including NSC, OPC, astrocytes or oligodendrocytes. Transformed cells progress through a series of pathological grades established by the WHO, characterized by histopathological features shown in representative H&E images. Through this progression, transformed cells accumulate a characteristic pattern of mutations as they progress from lower grade oligodendroglial or astrocytic lesions to more malignant versions. The most malignant HGG, glioblastoma, can arise de novo (primary) or progress from lower grade lesions (secondary), and are histopathologically indistinguishable, yet phenotypically heterogeneous. Features of glioblastoma include necrosis (arrowhead), microvascular proliferation (arrow) and substantial cellular pleomorphism, with some tumors containing giant cells as shown in the right panel.

Glioblastoma is a heterogeneous disease with a spectrum of histopathological characteristics, and numerous genetic aberrations per tumor. Molecular subclassifications based on gene expression signatures do not correlate strongly with histological features, but instead associate with survival (Lee et al., 2008; Phillips et al., 2006) or with patterns of copy number imbalances and somatic mutations (Verhaak et al., 2010). Large-scale integrated genomic analyses of human HGG samples showed that virtually all glioblastomas contained concurrent mutations in genes from three distinct signaling pathways; the receptor tyrosine kinase(RTK)/phosphoinositide-3 kinase (PI3K) pathway, the p53 pathway, and the RB pathway, with the specific gene within each pathway targeted for mutation varying widely among tumors (Parsons et al., 2008; TCGA, 2008) (Figure 2). Glioblastoma can be further subdivided into primary glioblastoma, arising ‘de novo’ with no clinical or histological evidence of a pre-existing lower-grade lesion, and secondary glioblastoma, which progresses from lower grade lesions including grade II diffuse astrocytoma and grade III anaplastic astrocytoma. Primary and secondary glioblastoma are histologically indistinguishable with overlapping age of onset and prognoses, although secondary glioblastomas as a group arise in younger patients and are associated with a slightly longer survival (Louis et al., 2007). At the molecular level, primary and secondary glioblastoma are distinguished by hotspot mutations in isocitrate dehydrogenase 1 (IDH1), which are found at high frequency in secondary glioblastomas and grade II–III astrocytomas, but not in primary glioblastomas (Parsons et al., 2008; Yan et al., 2009).

Figure 2.

Commonly mutated pathways in glioblastoma. Large-scale integrated genomic analyses of human glioblastoma showed that virtually every tumor harbored mutations in genes encoding components of the RTK/RAS/PI3K, p53, and RB signaling pathways. The RTK/RAS/PI3K pathway transduces extracellular growth factor signals into cellular responses of growth, proliferation and survival. Binding of growth factors to the extracellular domain of receptor tyrosine kinases (RTKs) induces RTK dimerization and autophosphorylation, leading to downstream activation of the small G-protein RAS by promoting exchange of GDP for GTP (depicted by color change from white to pink), and recruitment and activation of PI3K. PI3K is a lipid kinase that phosphorylates PIP2 to generate PIP3. A cascade of effectors are activated downstream of elevated PIP3 levels, with activation of the serine/threonine kinase AKT as one of the most well-characterized components of PI3K signaling. PTEN is a lipid phosphatase that directly reverses the effects of PI3K, thus negatively regulating the pathway. RAS activity is negatively regulated by GTPase Activating Proteins (GAPs) including NF1, which facilitate the hydrolysis of RAS-GTP to RAS-GDP. In HGG, aberrant activation of the RTK/RAS/PI3K pathway is most commonly caused by amplification, overexpression or constitutive activation of RTKs, activating mutations in PI3K or its regulatory subunits, and loss of function mutations in PTEN and NF1.

The CDKN2A locus encodes for two separate tumor suppressor proteins, INK4A and ARF, which negatively regulate two separate signaling cascades. In response to mitogenic signals, the family of Cyclin D proteins binds to CKD4 or CDK6 and phosphorylate RB, driving the transition from the G1 phase to the S phase of the cell cycle. INK4A is a tumor suppressor that binds to and inhibits Cyclin D/CDK4/6 complexes, thus inhibiting cell cycle progression. ARF binds to HDM2 and inhibits its E3 ubiquitinase activities, resulting in p53 stabilization. p53 is a tumor suppressor involved in maintenance of genomic stability, cell cycle checkpoints, senescence and cell death. In HGG, frequent loss of function mutations occur in the tumor suppressors p53, ARF, INK4A and at lower frequency, RB. The pathways may also be disrupted by amplification of Cyclin D, CDK4 or 6, or HDM2, resulting in misregulation of cell cycle progression, apoptosis and senescence programs.

The Utility of Mouse Models

Primary glioblastomas account for more than 90% of all glioblastomas. The complexity of mutations within each tumor and the absence of precursor lesions limit the insights into disease pathogenesis that can be gained from surgical tumor samples. The development of mouse models provides the opportunity to systematically interrogate the contribution of specific pathway disruptions to gliomagenesis. Numerous strategies have been used to engineer a multitude of mouse models for glioma that recapitulate varying aspects of the histological and molecular features of human HGG. These novel experimental systems have enabled us to step beyond the correlative evidence of molecular aberrations catalogued from end-stage tumors to directly investigate the contribution of individual genetic alterations to the initiation, progression and maintenance of gliomagenesis in an in vivo setting. Mutations can be introduced into murine hosts, either alone or in combination, and can be targeted to specific cells in the brain during embryonic development, at birth, or in the adult, allowing investigation of the competence of individual cell types for transformation, and the role of developmental context and differentiation status of the target cell.

Mouse models of HGG also afford the unique opportunity to study disease progression. The majority of human patients typically present clinically with advanced tumors resulting in a paucity of information regarding the early stages of disease. By examining mouse models at early stages of spontaneous HGG development, we can gain valuable insight into the pathological progression of these tumors on a molecular level. By sequentially introducing genetic lesions to mice, we can identify mutations that are critical to the initiation of gliomagenesis, as well as those mutations that are insufficient to initiate tumor formation alone, but act cooperatively in the initiation, maintenance or progression stages of tumorigenesis. This information can be useful to guide the interpretation of human cancer genomics (Zheng et al., 2008).

Furthermore, mouse models represent a platform to gain insight into basic tumor biology, including potential supporting roles of the surrounding stroma and growth factor milieu that may support the growth and proliferation of tumor cells. Tumors generated in mice allow investigation into the properties associated with cancer stem cells, a small proportion of self-renewing stem-like cells that may hold the key to tumor recurrence following treatment. Other potential mechanisms of tumor resistance can be studied in mice, including signaling pathway interaction and feedback loops that may be enlisted upon therapeutic intervention.

Finally, cohorts of mice can ideally be used to identify therapeutic targets and serve as subjects in preclinical trials to investigate agent utility in an in vivo setting. A major advantage of mice in this context is the potential to narrow down the potential drug targets, pharmacological candidates and relevant therapeutic combinations to be tested in clinical trials. However, for the results to be transferrable to the human condition, it is preferable that the models recapitulate the molecular pathogenesis of human HGG as closely as possible. Large scale genomic studies have generated abundant data relating to the molecular alterations and expression signatures of human tumors, which can be used to interrogate the fidelity of existing models to human disease. This will ensure that the mouse models will generate the most accurate preclinical data possible in a different species. In this regard, the ideal model for preclinical drug studies should arise in an accurate orthotopic microenvironment to mimic the interactions between host and developing tumor, and display cellular and molecular heterogeneity reminiscent of the human disease. Tumors should display appropriate histopathological features, including infiltrative behaviour, and should reproduce the spectrum of molecular aberrations frequently found in human disease, progressing through similar pathological stages. For convenience, models should be highly penetrant with a short latency to tumor formation, a narrow and predictable window of disease, and possess a reliable means for monitoring tumor formation through non-invasive imaging techniques. The utility of the current mouse models in selecting effective therapeutic agents will be established in the future as findings from mouse models are translated into the design of new therapeutic approaches.

Strategies for Making Mouse Models

Numerous strategies have been employed to generate HGG in mice. These approaches vary in complexity and speed of tumor formation, and each has its advantages and limitations.

Implantation Models

Historically, drug development research has relied upon the subcutaneous engraftment of human glioblastoma cell lines to the flank of immunocompromised mice for in vivo assessment of drug efficacy. This method is not technically challenging and quickly generates easily detectable tumors. Drug response can be measured easily using callipers to quantify regression. Unfortunately, this popular method suffers from several critical disadvantages, including an inaccurate microenvironment for tumor development. The brain is protected by the blood brain barrier which has direct relevance to drug delivery and kinetics in vivo. For this reason, the orthotopic or intracranial implantation is rapidly gaining in popularity for modeling HGG (Figure 3).

Figure 3.

Implantation strategies to generate intracranial tumors in mice. The direct injection of human surgical samples into the brain of immunocompromised mice can generate HGG with multiple histological and molecular similarities to the human disease. Prior to intracranial grafting, human samples can be sorted based on cell surface markers, cultured under neurosphere conditions or passaged heterotypically through the flank. Alternatively, human glioblastoma cell lines can generate rapid and reproducible HGG in mice, although these cell lines may create inaccurate histological representations of human disease. To analyze the role of specific mutations, neural progenitors or astrocytes can be isolated from wild-type or genetically modified mice, and further genetically modified in vitro via viral transduction prior to intracranial implantation to assess impact on gliomagenesis.

Orthotopic xenotransplantation of well-characterized human glioblastoma cell lines is typically accomplished by injecting cells directly into the brain of immunocompromised host mice. This can generate reproducible tumors in the brain with short latency, and a narrow and predictable window of onset (Radaelli et al., 2009). However, some cultured tumor cell lines are not tumorigenic in mice, and the tumors generated from those that are tumorigenic are often poor histological representations of human HGG (Akbasak et al., 1996; de Ridder et al., 1987; Stein, 1979). Tumors tend to be well-demarcated and compact with minimal infiltration to the normal parenchyma (Finkelstein et al., 1994; Jacobs et al., 2011; Mahesparan et al., 2003; Radaelli et al., 2009). On a molecular level, cultured cell lines are exposed to different selective pressures in culture than they would be exposed to in vivo, and as a result, tend to express a different profile of genetic aberrations, including the loss of EGFR amplification and mutation (Li et al., 2008). This issue has been effectively circumvented by minimizing manipulation of the cells prior to transplantation to generate tumors that accurately replicate the parental tumor. The immediate xenotransplantation of human surgical samples can generate tumors with considerable conservation of the parental tumor in regards to histological features, invasive growth pattern, CD133+ content, and molecular profile, thus creating a supply of relevant and reproducible tumors for study (Claes et al., 2008; Giannini et al., 2005; Shu et al., 2008; Yi et al., 2011). Short term culturing of human surgical samples under neurosphere conditions prior to orthotopic implantation, or intermediate passaging as flank xenograft prior to performing heterotypic to orthotopic transplantation also maintained molecular signatures of the primary tumor (Galli et al., 2004; Giannini et al., 2005; Stockhausen et al., 2011). Spontaneously occurring or genetically induced mouse tumors can be effectively passaged intracranially as allografts. Xenotransplantation has also been used to identify populations enriched for cancer stem cells by demonstrating subpopulations of the bulk tumor that show enhanced tumorigenicity (Beier et al., 2007; Phillips et al., 2006; Rich et al., 2005).

The implantation method is further amenable to direct interrogation of the role of different cell of origin and molecular aberrations in gliomagenesis, as neurospheres and primary astrocytes can be genetically engineered in vitro prior to implantation. The major drawback to this method is the necessity for immunocompromised hosts, as the immune system is increasingly recognized to perform critical functions in shaping tumor progression and in response to therapy (Schreiber et al., 2011).

Genetically Engineered Mouse (GEM) Models

The ability to manipulate the mouse genome has made it possible to generate spontaneous tumors in immunocompetent mice. The generation of transgenic mice in which a specific promoter drives cell type-specific expression of a relevant gene of interest is a straightforward approach to determine the effect of disrupted signalling through oncogene mutation or overexpression. The simplest approach to assess the impact of tumor suppressor gene loss of function is the generation of knock-out mice in which a germline deletion compromises gene function by removing all or part of the gene of interest (Figure 4A,B) (Nagy et al., 2003). These traditional approaches may result in tumor predisposition, however, in some cases, disrupted regulation of key signalling pathways controlled by oncogenes or tumor suppressor genes can also cause profound developmental abnormalities which may result in lethality before a tumor phenotype emerges. Transgenic mice are also limited by the specificity of the promoter, which may show expression in multiple cell types and at stages in development associated with defects unrelated to the desired tumor phenotype.

Figure 4.

Germline and conditional strategies for generating genetically engineered mouse models.

A. Transgenic mice are generated by pronuclear injection of a transgenic construct into a fertilized egg, resulting in undirected integration into the germline. For tumor models, transgenes are often oncogenes driven by tissue specific promoters.

B. Germline knockout mice, to determine effects of loss of tumor suppressor gene function, are created by replacing a portion of the endogenous tumor suppressor gene with a Neo selection cassette via homologous recombination in embryonic stem (ES) cells. ES cells are injected into blastocysts to generate mice that are chimeric for the targeted allele. Offspring of chimeric mice that carry the targeted allele in the germline are selected for further breeding to generate knock-out mice that are homozygous for the targeted mutation. The result is permanent deletion of the tumor suppressor gene and loss of its associated functions throughout all developmental stages.

C. Cre-mediated activation of oncogenes allows temporal control of expression. A strong transcriptional STOP sequence flanked by loxP sites prevents expression of the oncogene in the absence of Cre recombinase. Gene expression is induced by Cre recombinase which catalyzes the deletion of the DNA between the loxP sites, removing the STOP sequence and allowing expression of the activated oncogene.

D. Conditional knock-out mice are designed to delete the gene of interest only after exposure to Cre recombinase. loxP sites are introduced into intronic sequences flanking exons that encode key activities of the tumor suppressor gene using homologous recombination in ES cells similar to (B). In conditional knock-out mice, loxP sites do not interfere with normal gene expression, so expression of the targeted gene is the same as the wild-type allele in the absence of cre recombinase. Selective expression of Cre recombinase mediates deletion of the sequences between the loxP sites, resulting in conditional knock-out of the gene of interest.

Further refinement of genetic modification techniques allows for the conditional targeting of genetic aberrations to particular tissues, and to cell types within that tissue. This is most frequently accomplished using the cre-lox system. Cre (causes recombination) recombinase mediates deletion of all sequences between two loxP (locus of crossover in P1) sequences, resulting in the permanent deletion of the intervening DNA, with a single loxP site remaining. Conditional knock-out mice are generated by inserting loxP sites in the introns surrounding exons encoding a critical function, and conditional transgenic mice are often generated by the presence of a strong transcriptional stop sequence flanked by loxP sites. Thus, tissue or cell-type specific expression of Cre recombinase may be used to generate conditional loss of function of a tumor suppressor gene, or activation of an oncogene in specific cells (Figure 4C, D) (Orban et al., 1992; Sauer and Henderson, 1988). The most frequently employed Cre transgenic animals in glioma modeling use the Nestin promoter to direct expression to the neural progenitor cell compartment, and GFAP to direct expression of cre to mature astrocytes and a subpopulation of progenitors in the SVZ. Temporal regulation of Cre recombinase activity has been achieved through the use of inducible Cre recombinase, which is fused to a mutated estrogen receptor ligand binding domain (cre-ER). Tamoxifen administration allows the cre-ER fusion protein to translocate to the nucleus and mediate recombination (Feil et al., 1996). This provides temporal control of gene deletion or oncogene activation, allowing investigation of genetic lesions at different stages of development. Precisely timed and targeted genetic lesions can effectively avoid the incidence of developmental abnormalities or non-CNS tumors. Through complex breeding schemes, multiple mutations can be combined in the same cell type. Multiple GEM models develop HGG with histological and genetic similarities to the human condition, and the development of non-invasive imaging techniques (MRI, microPET-CT and bioluminescence) allows the early detection and monitoring of tumor growth, and makes spontaneous glioma models more tractable. However, GEM models are criticized for their lengthy and costly development and maintenance, and prolonged latency to tumor formation.

Somatic Gene Transfer

Most GEM models introduce mutations to an entire tissue, encompassing a large number of cells. Human cancer is more likely a stochastic event, with transforming mutations arising in single cells and imparting a selective growth advantage which results in clonal expansion. To effectively mimic this mode of initiation, several groups have developed models of somatic gene transfer, employing viruses to deliver activated oncogenes or cre recombinase to a limited number of cells (Figure 5).

Figure 5.

Schematic illustration of somatic gene transfer using viral transduction. Somatic gene transfer can be accomplished by the direct injection of viral particles or virus producing cells into the mouse brain. The RCAS/tv-a system requires the engineering of a recipient mouse to express the tv-a receptor, depicted by the red Y-shaped cell surface receptor. When this receptor is expressed under a tissue specific promoter, it enables specificity of viral transduction directed only to particular cells expressing the receptor. Other types of viruses infect mammalian cells without the provision of a receptor, though viruses differ in their ability to transduce dividing and non-dividing cells.

The most commonly used viral delivery method is the RCAS/tv-a system. This strategy employs an avian retrovirus RCAS (replication-competent avian sarcoma-leukosis virus long terminal repeat with splice acceptor), which can be engineered to encode exogenous DNA sequences, most commonly those for activated oncogenes. RCAS virus binds exclusively to a receptor known as tv-a. Since mammalian cells do not express tv-a, they are not susceptible to infection unless modified. Three transgenic mouse lines have been engineered to express tv-a in different cell types in brain. Ntv-a mice express tv-a receptor only in Nestin positive cells of the progenitor compartment, Gtv-a mice express the tv-a receptor in Gfap-expressing cells, which include mature astrocytes as well as a subset of progenitors in the SVZ, and Ctv-a mice express the tv-a receptor in CNPase positive oligodendrocyte progenitors (Holland et al., 1998; Holland and Varmus, 1998; Lindberg et al., 2009). By injecting RCAS virus or virus-producing chicken fibroblasts into the brain of tv-a transgenic mice, activated oncogenes can be delivered to a small number of a specific cell type, which allows convenient investigation into susceptibility of particular cell types to transformation by selected mutations. A key advantage to this system is the ability to transduce the same cell with multiple viruses simultaneously, allowing the interrogation of cooperative mutations, as well as the inclusion of a virus to allow bioluminescent tracking of tumor progression. Furthermore, the transgenic mouse can be bred with other GEMs expressing floxed alleles, and a virus encoding cre can create additional mutations to investigate mutations in the context of a tumor suppressor-deficient background.

Historically, the RCAS viruses have been delivered to the cortex of neonatal mice, as they infect only dividing cells and thus efficiency of infection decreases significantly with age. This system has recently been shown to generate tumors in the adult brain (Hambardzumyan et al., 2009), a critical advance that allows assessment of glioblastoma development in the setting in which it most commonly occurs, as well as allowing comparison between developmental stages of target cells, and different regions of the adult brain. Limitations of this system include trauma and inflammation associated with injection, recruiting cells to the area and eliciting an immune response that may complicate interpretation. Also, analysis must consider that multiple and sequential infections are likely to generate a heterogeneous population of cells, complicating tumor analysis. Finally, comparisons of target cell susceptibility to transformation may be influenced by technical issues. For example, stability of the tv-a gene product has not been analyzed and may differ between Ntv-a and Gtv-a mice as a function of development (Fisher et al., 1999). Alternatively, the tv-a transgene may be expressed in rare uncharacterized cells due to unexpected expression of the transgene promoter in other cell types.

Other viruses used to similarly generate glioma models include adenoviruses and lentiviruses. The advantage of these viruses is that they transduce both dividing and non-dividing cells and are thus efficient for use in the adult brain. However, it is not possible to determine which cell type is the cell of origin of the ensuing tumor. These viruses are also capable of encoding DNA sequences of considerably larger size (> 10kb) than RCAS, which is limited to 2.6kb transcripts (Holland et al., 1998).

In a related approach, the sleeping beauty (SB) transposase system has been used to generate HGG using non-viral transfection (Wiesner et al., 2009). SB, delivered to the brain as plasmid DNA, mediates the chromosomal integration of plasmid DNA concurrently delivered to the lateral ventricle of newborn mice. Transposons up to 10kb can be integrated using the SB system, and this model provides a mutagenesis platform for rapid validation of proposed driver mutations, as well as forward genetic screens to identify novel genes involved in gliomagenesis. Limitations include ambiguity regarding the cell type transformed, as well as the clonality of the resulting tumor.

Lessons Learned from Mouse Models of HGG

Powerful oncogenes that impact multiple signaling pathways are capable of driving gliomagenesis in mice

Gliomagenesis can be initiated in mice by the introduction of oncogenes. Successful driving oncogenes activate multiple signaling pathways and/or impair tumor suppression mechanisms. Transgenic expression of the SV40 T Antigen alone (Danks et al., 1995) or in combination (Wiesner et al., 2009), inactivates both p53 and Rb signaling pathways, two of the most frequent events in the pathogenesis of HGG, and results in rapid astrocytoma formation. Similarly, a truncated version of SV40 T Antigen, T121, inactivates the Rb family of proteins, and successfully generates HGG in mice (Xiao et al., 2002; Xiao et al., 2005). Transgenic overexpression of oncogenic v-src also promotes gliomagenesis in mice. V-src is a constitutively active kinase that activates a wide variety of signaling pathways in the cell which promote growth and survival (Smilowitz et al., 2007; Weissenberger et al., 1997). The additional loss of p53 or Rb did not enhance gliomagenesis in this model suggesting that these tumor suppression pathways are also impaired by the v-src transgene (Maddalena et al., 1999), though it is difficult to discern specifically which pathway alterations are necessary or sufficient for the generation of tumors in this model.

Alternatively, HGG can be initiated via activation of specific signalling pathways known to be activated in human HGG. Human HGG frequently demonstrate robust activation of the Ras signalling cascade. Transgenic over-expression of activated p21-ras (V¹²Ha-Ras) under the control of the GFAP promoter resulted in HGG, where malignancy of these astrocytomas was HaRas dose-dependent; hemizygous animals develop predominantly grade II, while homozygous mice develop grade III astrocytomas (Ding et al., 2001). A germline transgenic mouse with single copy integration (RasB8) was selected for further study of the molecular and pathological progression of spontaneous Ras-driven astrocytoma formation (Shannon et al., 2005). The progression of tumor formation in this model is remininscent of secondary glioblastoma. The prolonged latency preceding tumor formation was associated with the accumulation of additional mutations, which included alterations associated with human gliomagenesis such as p53 mutation, as well as decreased expression of Pten and p16Ink4a (Gutmann et al., 2002).

The majority of mutations are individually insufficient for gliomagenesis but require cooperating mutations to recapitulate the histological features of HGG

Oncogenes with the ability to affect multiple signalling pathways initiate tumors efficiently, but more specific pathway activation models suggest a requirement for additional pathway alterations that must accumulate over time to effectively promote tumorigenesis. As human HGG samples possess a wide spectrum of mutations, mouse models can be used to dissect the importance of particular secondary mutations for cooperativity in gliomagenesis. Ras activation has been modelled by multiple groups, using different strategies, but each study has confirmed that activated Ras alone is in general insufficient to initiate gliomagenesis. However, activated Ras and downstream Ras pathway components can cooperate with other mutations frequently found in human HGG, including oncogenes such as activated Akt or c-myc over-expression, or tumor suppressor deletions such as Ink4a/Arf, to generate glioblastoma in mice (de Vries et al., 2010; Lassman et al., 2004; Lyustikman et al., 2008; Marumoto et al., 2009; Robinson et al., 2010; Robinson et al., 2011; Uhrbom et al., 2002), although the phenotype of the resulting tumor may depend on the combinations of mutations used to initiate its formation (Uhrbom et al., 2002).

These models collectively support the hypothesis that constitutive activation of Ras via RTK amplification/overexpression is sufficient to drive gliomagenesis when combined with other frequently reported mutations. However, Ras mutations are rare in human glioblastoma, and therefore may not generate ideal models for purposes such as preclinical testing of targeted inhibitors. Models guided by human genomics may be more likely to accurately recapitulate human tumors. Several groups used signature mutations of human glioblastoma, specifically RTK amplifications, as starting point to generate models of HGG. EGFR amplification/overexpression and constitutively active mutation are common genetic lesions in human glioblastoma, and are frequently accompanied by Ink4a/Arf deletion which impairs both the p53 and Rb signaling pathways. Constitutively active EGFR and PDGFRα overexpression are individually insufficient for tumorigenesis, but each can cooperate with Ink4a/Arf deletion to induce formation of glioma-like lesions (Holland et al., 1998; Liu et al., 2011b; Zhu et al., 2009). These models provide experimental evidence of the strong selective advantage imparted by disrupting PI3K, Rb and p53 signaling pathways, consistent with the mutation analysis of human tumors shown later (Parsons et al., 2008; TCGA, 2008).

The fact that single activated oncogenes or mutated/amplified RTK are insufficient to promote gliomagenesis unless combined with additional oncogene activation or tumor suppressor gene loss confirms that multiple cooperative mutations are required to generate HGG. Many studies strive to identify molecular subgroups of HGG based on mutations and/or expression signatures that may be used to stratify patients for rational use of targeted therapies. Therefore, there is a need to develop multiple models to test the role of particular combinations of mutations and the utility of specific molecular markers to accurately represent the underlying processes driving tumor growth, maintenance and progression. It is essential to carefully evaluate such models to ensure relevance and fidelity to the human condition.

Some mutations found with high frequency in human HGG play a role in progression rather than initiation of HGG

Many mutations may contribute to gliomagenesis through cooperative effects with other mutations, influencing malignant progression or maintenance of the tumor rather than initiation. These relationships are not possible to determine from analysis of human primary glioblastomas, which arise without progression from lower grade lesions. Mouse models allow the introduction of additional mutations to established models to determine cooperativity in terms of enhanced malignancy, increased penetrance or shortened latency to tumor formation. Pten loss is insufficient for tumorigenesis, even when combined with Rb loss (Chow et al., 2011). However, it has been demonstrated to accelerate tumor formation in multiple models, suggesting a role for Pten in tumor progression (Kwon et al., 2008; Wei et al., 2006; Xiao et al., 2002; Xiao et al., 2005). p53 deletion alone is insufficient for highly penetrant gliomagenesis, but p53 mutations are detected frequently in low grade human astrocytomas, and may create a plasticity within the cell that allows the accumulation of cooperative alterations in RB and RTK pathways for transformation. Combined conditional deletion of p53 with Rb, Pten, and/or Nf1 results in highly penetrant high-grade astrocytomas (Alcantara Llaguno et al., 2009; Chow et al., 2011; Kwon et al., 2008; Zheng et al., 2008). P53 and Ink4a/Arf loss of function both shorten the latency to tumor formation in multiple models, and loss is frequently associated with additional enhancement of malignancy and penetrance (Dai et al., 2001; de Vries et al., 2010; Hesselager et al., 2003; Marumoto et al., 2009; Squatrito et al., 2010; Tchougounova et al., 2007; Uhrbom et al., 2005; Weiss et al., 2003). Surprisingly, an intragenic deletion of p53 induced in GFAP-expressing cells induced astrocytomas with high frequency even in the absence of other engineered mutations, suggesting a dominant effect of this mutation compared to complete loss of p53 function (Wang et al., 2009).

Bidirectional communication between tumor and microenvironment influences the pathogenesis of HGG

The complex and dynamic interplay of communication between the tumor and the surrounding non-neoplastic tissues also influences the development of HGG. The tumor microenvironment supplies instructive cues to direct tumor behavior via the secretion of paracrine factors, including cytokines, extracellular matrix proteins and soluble factors (Hoelzinger et al., 2007; van Kempen et al., 2003). Mouse models provide a closed experimental system in which microenvironmental factors can be manipulated to elucidate their role in gliomagenesis. This includes changes in the genetic background of the surrounding stromal compartment, and the milieu of available paracrine growth factors.

The role of the stroma in instructing gliomagenesis has been extensively investigated within the context of Neurofibromatosis 1 (NF1) (Bajenaru et al., 2001; McGillicuddy et al., 2009; Pong and Gutmann, 2011). NF1 is a genetic disorder in which patients inherit one mutant allele of NF1, and an increased susceptibility to tumors including gliomas. Mice carrying combined germline heterozygous loss of Nf1 and p53 developed low to intermediate grade astrocytomas (Reilly et al., 2000), and germline deletion of p53 combined with neural-specific deletion of Nf1 induced high-grade astrocytomas with high penetrance (Zhu et al., 2005). Mouse models also revealed that Nf1 loss plays important roles in the tumor microenvironment. Deletion of both alleles of Nf1 targeted specifically to astrocytes failed to induce gliomas in the context of wild-type stroma (Bajenaru et al., 2002). However, in the context of an Nf1+/− germline, targeted deletion of the second Nf1 allele in astrocytes induced optic pathway gliomas (Bajenaru et al., 2003). Nf1+/− microglia in the tumor microenvironment provide critical cues to promote optic pathway glioma formation and growth (Daginakatte and Gutmann, 2007).

The tumor can also impact the microenvironment. Frequent alterations in PDGF/PDGFRα signalling pathway occur in human glioblastoma, suggesting that an autocrine and/or paracrine loop may promote gliomagenesis (Shih and Holland, 2006). Several models have successfully exploited this aberrant pathway to generate HGG with fidelity to the human disease. The introduction of aberrant PDGF signalling to neural progenitor cells can generate oligodendrogliomas (Calzolari and Malatesta, 2010; Dai et al., 2001), and astrocytomas of the proneural subtype (Becher et al., 2010; Hambardzumyan et al., 2009; Hede et al., 2009; Hesselager et al., 2003; Lei et al., 2011; Pitter et al., 2011). PDGF is a secreted factor that is often overexpressed by HGG, and can have paracrine effects on surrounding non-neoplastic tissues. Injection of PDGF to the brain recapitulates PDGF overexpression, and impacts a wide range of non-neoplastic cells, recruiting distant cells to co-opt for transformation and participation in HGG pathogenesis, contributing to the heterogeneity of the tumor histology (Assanah et al., 2006; Fomchenko et al., 2011).

Multiple cell types within the glial lineage can serve as the cell of origin for HGG

Mouse models have provided compelling evidence that more than one type of neural cell may be capable of transformation into HGG. Neural stem cells (NSC) have long been hypothesized to be the target of transformation in HGG because they possess properties of self-renewal and multipotentiality, and proliferate throughout adulthood making them susceptible to the accumulation of transforming mutations. The SVZ is comprised of multiple neural progenitor cell types, which could explain some of the heterogeneity of HGG (Canoll and Goldman, 2008). Human HGG are often located in the cerebral hemispheres, not contiguous with proliferative niches in the brain. However, the aberrant migration of tumor suppressor-deficient neural progenitors suggests that mutated NSC may escape from the proliferative niche and account for the appearance of tumors in other brain regions (Kwon et al., 2008)(Alcantara Llaguno et al., 2009). HGGs often express NSC markers including nestin and Sox2, suggesting that they share some molecular signatures of NSC, although this may reflect disrupted differentiation rather than the cell of origin. Multiple models have shown that targeted mutations in neural progenitor cells can induce gliomas efficiently. Very selective targeting of mutations to specific cell populations has been performed to identify susceptible brain regions and cell types. Mutations introduced to the cortex are less efficiently transforming than those directed to the hippocampus or SVZ, regions that retain neural progenitor cells into adulthood, suggesting a more primitive cell of origin for gliomas (Alcantara Llaguno et al., 2009; Hambardzumyan et al., 2009; Marumoto et al., 2009). The limitation of such studies is that very selective targeting of mutations to a relatively small number of cells of a specific phenotype is used to define the cell of origin. However, this approach may lack the sensitivity to detect transformation events of lower frequency.

Multiple lines of investigation indicate that more differentiated cells in the glial lineage may also give rise to glioma. Targeted expression of PDGF-B to oligodendrocyte progenitor cells (OPCs) induced oligodendroglioma in the mouse (Lindberg et al., 2009; Persson et al., 2010). Oligodendrogliomas driven by expression of v-erbB, the viral oncogenic form of EGFR, showed expression signatures and differentiation potential similar to OPCs rather than NSCs. Interestingly, tumor cells expressing markers of OPCs showed a greater capacity for tumor initiation upon transplantation into a host brain compared to NSCs. Oligodendrogliomas in this mouse model, and in humans, appear to arise in association with white matter tracts (Persson et al., 2010). Evidence is mounting that OPCs may be a target for astrocytoma formation as well. Retroviral injections directed to the white matter of adult mice demonstrated that PDGF stimulation cooperated with the deletion of Pten and p53 to transform white matter progenitors, and provided compelling evidence that the OPC was the cell of origin for the proneural glioblastomas generated in this model (Lei et al., 2011). Finally, a recent report used a mosaic analysis of double markers technique to determine that despite introduction of genetic lesions to NSC, only the progeny OPCs developed a proliferative growth advantage. Introduction of lesions to OPCs directly resulted in tumors similar to the proneural subgroup (Liu et al., 2011a). In spontaneous models of astrocytoma induced by combined deletion of p53 and Nf1, or Pten and p53, the earliest evidence of aberrant hyperproliferative lesions were often detected in the white matter tracts of the corpus callosum in mice analyzed at early time points prior to astrocytoma formation (Chow et al., 2011; Wang et al., 2009).

There is also evidence that astrocytes can give rise to astrocytomas. The efficiency of this event likely depends on the specific mutations used to drive tumorigenesis. Expression of a mutant EGFR was equally transforming in isolated astrocytes or NSCs from Ink4A/Arf-null mice, which generated similar tumors following intracranial implantation (Bachoo et al., 2002) Expression of EGFRvIII also cooperated with deletion of p53 and Pten to render astrocytes tumorigenic (Endersby et al, 2011). In contrast, combined deletion of Pten, p53 and Rb only induced tumorigenic growth of NSCs, but not similarly engineered astrocytes (Jacques et al., 2010). This result may reflect the absence of a driving activated oncogenic mutation. Taken together, these studies suggest that the relative susceptibility of different neural cell types to transformation likely varies depending on the specific combinations of mutations.

Given the abundant recent evidence that differentiated cells can be reprogrammed into induced pluripotent cells, or into different cell fates by expression of a small number of transcription factors (Chambers and Studer, 2011; Wu and Hochedlinger, 2011), it is not surprising that multiple neural cell types can give rise to glioma depending on the specific mutations driving tumorigenesis. While it would appear that NPCs, astrocytes and OPCs may all be capable of serving as the cell of origin for HGG, it remains unclear what role dedifferentiation may play, and what ultimately dictates the tumor phenotype of oligodendroglioma versus astrocytoma. Mouse models provide powerful experimental tools to systematically address these issues.

Mutations in common pathways are capable of driving diverse glioma phenotypes

Although primary glioblastoma is the most common human glioma, generating a mouse model of 100% penetrant grade IV glioblastoma has not been easy to accomplish. The majority of models created to date generate a mixture of grade III and grade IV tumors. Combinations of mutations known to occur in human glioblastoma sometimes generate tumors of different phenotypes when modeled in the mouse. For example, activated EGFR efficiently drives astrocytoma formation (Bachoo et al., 2002; Endersby et al., 2011; Wei et al., 2006) or oligodendroglioma (Weiss et al., 2003; Ding et al., 2003). In some cases, similar combined mutations and cell types are targeted, implying critical roles for the timing and sequence of mutations. The combined deletion of Pten and p53 with or without concomitant Rb deletion in the mouse brain efficiently generated HGG with a long latency to tumor formation. The resulting tumors acquired somatic mutations similar to the spectrum of copy number alterations noted in human HGG samples, indicating that there are similar selective pressures driving gliomagenesis in mouse and man. Heterogeneity among the resulting tumors showed that the expression signatures in the mouse model recapitulated the three expression subgroups found in humans (Phillips et al., 2006), demonstrating that expression subgroups do not appear to be linked to histological features or driving mutations. Analysis of the somatic mutations acquired in mouse gliomas arising from combined deletion of Pten, p53 and Rb also showed that there is a strong selective pressure to de-regulate these signalling pathways by simultaneous mutation of multiple genes within the same pathway (Chow et al., 2011). This highlights the fact that the molecular pathogenesis underlying gliomas can not be fully understood by grouping together tumors carrying mutations from a common pathway, as there are independent selective advantages of mutations in different components of the same pathway that contribute to gliomagenesis. The complex combinations of specific mutations likely play an important role in the final tumor phenotype and contribute to the substantial heterogeneity associated with gliomas.

The specific method used to generate glioma models can produce dramatically different results. The combined deletion of Pten and p53 consistently generates HGG in multiple models (Chow et al., 2011; Jacques et al., 2010; Zheng et al., 2008). The additional deletion of Rb mediated by GFAP-cre, generated high-grade astrocytoma with shortened latency, including a majority of tumors that were contiguous with the SVZ (Chow et al., 2011). Unexpectedly, the same combined deletion of Pten, p53 and Rb induced via adeno-cre injection to the SVZ exclusively generated primitive neuroectodermal tumors (PNETs) instead of astrocytomas (Jacques et al., 2010). Although both approaches appeared to target GFAP+ cells within the SVZ, there were likely critical differences in the specific subtypes of cells targeted in the different studies, resulting in a completely different tumor phenotype.

Challenges and Future Directions

Mouse models can be valuable tools in the study of HGG pathogenesis and treatment, but there are some limitations to their modeling capabilities. Models are simply approximations of disease. A mouse is not a man and some inherent species differences cannot be overcome, including potentially critical differences in metabolism, lifespan, telomerase length and complexity which likely influence gliomagenesis (Atai et al., 2011; Rangarajan and Weinberg, 2003).

Generating and defining “ideal” models of HGG remains a challenge. Increased understanding of the genetic and environmental factors driving gliomagenesis provides impetus to improve mouse models to generate the most accurate representation of human disease. To match the complexity of human tumors, researchers need to increase the complexity of the models. Mutations are likely accumulated stochastically, not simultaneously, and the experimental introduction of mutations in a stepwise manner, combining the use of GEMs with somatic gene transfer mechanisms may replicate the apparent stepwise accumulation of mutations that accompanies malignant progression. This strategy may further inform which mutations are critical to gliomagenesis and which ones allow the accumulation of additional mutations that drive transformation.

Mouse models are purported to be useful for testing preclinical therapeutics to narrow the candidates to proceed to clinical trials. Once essential alterations are defined, a wider adaptation of inducible, conditional genetic modifications, such as the tet-on/tet-off system, in which gene expression can be regulated by the provision of doxycycline, would allow researchers to express an oncogene to generate HGG, and then silence its expression to provide proof of principle that direct inhibition of that target will result in regression. Similarly regulated shRNA constructs could remove and replace tumor suppressor genes to investigate the requirement of loss for tumor maintenance, and the impact of replacement. However, this approach is dependent on the relevance of the molecular alterations driving gliomagenesis in each model.

The success of a mouse model depends on the goal of the model. The current standard measures how well the model recapitulates the spectrum of histopathological features of human HGG. If the goal is to determine the impact of certain genes on the extent of infiltration or the promotion of microvascular proliferation, then models that fulfill histopathological requirements may be quite successful. However, if the goal is to perform preclinical trials to predict efficacy in human tumors, the driving mutations and underlying molecular profile is likely of equal or greater importance, as the pathway targeted for tumorigenesis will dictate the response to therapeutic approaches with newly developed selective pathway inhibitors.

Abbreviations

Cre

causes recombination

EGFR

Epidermal growth factor receptor

GEM

genetically engineered mouse

GFAP

glial fibrillary acidic protein

HGG

High grade glioma

IDH1

isocitrate dehydrogenase

LoxP

locus of crossover in P1

NF1

Neurofibromatosis-1

NSC

neural stem cell

OPC

oligodendrocyte progenitor cells

PDGFRα

platelet derived growth factor receptor alpha

PI3K

phosphoinositide-3 kinase

RCAS

replication-competent avian sarcoma-leukosis virus long terminal repeat with splice acceptor

RTK

receptor tyrosine kinase

SB

sleeping beauty transposase

SVZ

subventricular zone

WHO

World Health Organization