A Drosophila Model for EGFR-Ras and PI3K-Dependent Human Glioma

Gliomas, the most common malignant tumors of the nervous system, frequently harbor mutations that activate the epidermal growth factor receptor (EGFR) and phosphatidylinositol-3 kinase (PI3K) signaling pathways. To investigate the genetic basis of this disease, we developed a glioma model in Drosophila. We found that constitutive coactivation of EGFR-Ras and PI3K pathways in Drosophila glia and glial precursors gives rise to neoplastic, invasive glial cells that create transplantable tumor-like growths, mimicking human glioma. Our model represents a robust organotypic and cell-type-specific Drosophila cancer model in which malignant cells are created by mutations in signature genes and pathways thought to be driving forces in a homologous human cancer. Genetic analyses demonstrated that EGFR and PI3K initiate malignant neoplastic transformation via a combinatorial genetic network composed primarily of other pathways commonly mutated or activated in human glioma, including the Tor, Myc, G1 Cyclins-Cdks, and Rb-E2F pathways. This network acts synergistically to coordinately stimulate cell cycle entry and progression, protein translation, and inappropriate cellular growth and migration. In particular, we found that the fly orthologs of CyclinE, Cdc25, and Myc are key rate-limiting genes required for glial neoplasia. Moreover, orthologs of Sin1, Rictor, and Cdk4 are genes required only for abnormal neoplastic glial proliferation but not for glial development. These and other genes within this network may represent important therapeutic targets in human glioma.


Introduction
Malignant gliomas, neoplasms of glial cells and their precursors, are the most common tumors of the central nervous system (CNS). These tumors typically proliferate rapidly, diffusely infiltrate the brain, and resist standard chemotherapies, properties that render them largely incurable. One key to developing more effective therapies against these tumors is to understand the genetic and molecular logic underlying gliomagenesis. The most frequent genetic lesions in gliomas include mutation or amplification of the Epidermal Growth Factor Receptor (EGFR) tyrosine kinase. Glioma-associated EGFR mutant forms show constitutive kinase activity that chronically stimulates Ras signaling to drive cellular proliferation and migration [1,2]. Other common genetic lesions include loss of the lipid phosphatase PTEN, which antagonizes the phosphatidylinositol-3 kinase (PI3K) signaling pathway, and activating mutations in PIK3CA, which encodes the p110a catalytic subunit of PI3K [1,2]. Gliomas often show constitutively active Akt, a major PI3K effector [1,2]. However, EGFR-Ras or PI3K mutations alone are not sufficient to transform glial cells, rather multiple mutations that coactivate EGFR-Ras and PI3K-Akt pathways are sufficient to induce glioma [2][3][4].
Understanding the interplay of these mutations and the neurodevelopmental origins of these tumors could lead to new insights into the mechanisms of gliomagenesis. The mammalian brain contains multiple glial cell types that maintain proliferative capacities, including differentiated astrocytes, glial progenitors, and multipotent neural stem cells. EGFR-Ras and PTEN-PI3K signaling regulates many developmental processes in these cell types, particularly proliferation and self-renewal, which are also properties of glioma cells [1]. Although recent hypotheses favor that gliomas arise from multipotent stem cells, data from mouse models demonstrate that differentiated glia, glial progenitors, and stem cells can all produce gliomas in response to genetic lesions found in human gliomas [5,6]. Thus, misregulation of these genetic pathways may confer unrestricted proliferative capacities to a range of glial cell types, but how this occurs remains unclear. While many of the same effectors are utilized by EGFR-Ras and PI3K in both glial development and cancer, constitutive activation of these pathways may deploy distinct outputs, not utilized in development, that allow particular cells to escape normal physiological cues that restrain proliferation and self-renewal. The identity of such outputs remains unclear.
With these issues in mind, we developed a Drosophila glioma model to facilitate genetic analysis of glial pathogenesis. Drosophila offers many tools for precise manipulation of cell-type-specific gene expression and dissection of multigene interactions. Most human genes, including 70% of known disease genes, have functional Drosophila orthologs [7]. Among the most conserved genes are components of major signal transduction pathways, including many gliomagenic genes. Recently, Drosophila has emerged as a model system for human neurological diseases because the CNS shows remarkable evolutionary conservation in cellular composition and neurodevelopmental mechanisms [8]. Similarly, Drosophila have multiple glial cell types that require the EGFR pathway for their normal development, and these cells appear homologous to mammalian glia in terms of function, development, and gene expression [9]. These similarities between flies and humans make Drosophila an attractive system for modeling gliomas.

Coactivation of EGFR-Ras and PI3K in Drosophila Glia Causes Neoplasia
Since concurrent activation of EGFR-Ras and PI3K signaling in glial precursors induces glioma in the mouse [4], we sought to create mutant phenotypes by hyperactivation of these pathways in fly glia and glial precursors. Drosophila has a single functional ortholog each for EGFR(dEGFR), Raf (dRaf), PIK3CA(dp110), PTEN(dPTEN), and Akt(dAkt), and two functional orthologs for Ras(dRas85D, dRas64B) (www.flybase.org). A diagram of the specific mutant forms of dEGFR used in our assays can be found in Figure S1. We performed glial overexpression assays with the Gal4-UAS system [10], using the repo-Gal4 driver, which gives sustained UAS-transgene expression in almost all glia, from embryogenesis through adulthood. For glial-specific RNAi, we employed UAS-dsRNA constructs [11], which we verified with phenotypic tests and/or antibody staining (see Materials and Methods). Glial morphology was visualized with membranelocalized GFP (CD8GFP) [12]. Cell number was determined with staining for Repo, a homeobox transcription factor expressed by repo-Gal4 positive glia [13].
In contrast, glial-specific activation of the EGFR-Ras pathway alone, through overexpression of dEGFR l or Raf gof , induced 5-10 fold excess glia in the larval brain and later pupal lethality ( Figure 1F and Table S1). dRas85D V12 overexpression induced approximately 5-10-fold excess glia, and these glia were smaller than wild-type or dEGFR l ;dp110 CAAX glia. ( Figure 1G). dRas85D V12 may be more potent than dEGFR l because dRas85D V12 can activate endogenous PI3K signaling [16]. Overexpression of dEGFR Elp , a classical hypermorphic mutant form of dEGFR [17], induced excess glial proliferation and neural morphogenesis defects ( Figure S2), but also caused early lethality which precluded examination of dEGFR Elp -dp110 interactions. As in mouse models, overexpression of wild-type dEGFR failed to induce excess glia [6,17], and instead retarded CNS growth ( Figure S2). Unlike dEGFR l , dEGFR WT and dEGFR Elp have functional ligand-binding domains ( Figure S1), and may cause additional defects by sequestering ligand otherwise required for normal development [17]. Glial-specific activation of the PI3K pathway alone, either by overexpression of dp110 CAAX , dp110 wild-type , dAkt, or dPTEN dsRNA gave viable animals with relatively normal brains ( Figure 1H, Figure S2, and Table S1). Therefore, coactivation of the EGFR and PI3K pathways synergize to produce much more severe phenotypes than would be expected if the effects of these pathways were additive.
In repo.dEGFR l ;dp110 CAAX brains, excess glia emerged in early larval stages and accumulated over 5-7 days. dEGFR l ;dp110 CAAX glia severely disrupt the normal cellular architecture of the larval brain ( Figure 1A and 1B and Figure 2A-C), lose normal stellate glial morphologies (Figure 2A-C), and generate multilayered aggregations of abnormal glia throughout the brain (Figure 2A-C); in these ways dEGFR l ;dp110 CAAX glia are neoplastic [18]. Like neoplastic epithelial cells, dEGFR l ;dp110 CAAX glia ectopically expressed an active form of the matrix metalloprotease dMMP1 ( Figure S3), which can confer an invasive potential [19,20], implying that abnormal dEGFR l ;dp110 CAAX glia may be invasive within the brain. Unlike neoplastic epithelia, neoplastic neural cells, such as dEGFR l ;dp110 CAAX glia, typically retain expression of genes that regulate neural cell fate, such as Repo [21,22].
Relative to controls, many dEGFR l ;dp110 CAAX glia showed BrdU incorporation, which marks S-phase cells ( Figure 2D and 2E), indicating that neoplastic glia arise from overproliferation. repo. dEGFR l ;dp110 CAAX animals also showed reduced BrdU in neuronal precursors ( Figure 2E, data not shown), demonstrating that neoplastic glia disrupt neuronal development. The cell cycle is governed by CyclinD-Cdk4 and CyclinE-Cdk2 complexes, which phosphorylate and inactivate Rb proteins, to release E2F activators to stimulate G1-

Author Summary
Malignant gliomas, tumors composed of glial cells and their precursors, are the most common and deadly human brain tumors. These tumors infiltrate the brain and proliferate rapidly, properties that render them largely incurable even with current therapies. Mutations in genes within the EGFR-Ras and PI3K signaling pathways are common in malignant gliomas, although how these genes specifically control glial pathogenesis is unclear. To investigate the genetic basis of this disease, we developed a glioma model in the fruit fly, Drosophila melanogaster. We found that constitutive coactivation of the EGFR-Ras and PI3K pathways in Drosophila glia gives rise to highly proliferative and invasive neoplastic cells that create transplantable tumor-like growths, mimicking human glioma. This represents a robust cell-type-specific Drosophila cancer model in which malignant cells are created by mutations in genetic pathways thought to be driving forces in a homologous human cancer. Genetic analyses demonstrated that EGFR-Ras and PI3K induce fly glial neoplasia through activation of a combinatorial genetic network composed, in part, of other genetic pathways also commonly mutated in human glioma. This network acts synergistically to coordinately stimulate cellular proliferation, protein translation, and inappropriate migration. Rate-limiting genes within this network may represent important therapeutic targets in human glioma.
We next examined negative regulators of the cell cycle. dEGFR l ;dp110 CAAX glia expressed Rbf1, but showed little Dap, a result we also observed in wild-type glia ( Figure S5). Dap inhibits dCyclinE-cdk2 complexes [25], and is transiently expressed in neural progenitors to promote cell cycle exit as they begin differentiation [26]. Dap overexpression completely suppressed repo. dEGFR l ;dp110 CAAX glial overgrowth ( Figure 2L), demonstrating that glial neoplasia is cell-autonomous and requires dCyclinE-dCdk2. Similarly, overexpressed Rbf1 and dCyclinE mutations also reduced repo.dEGFR l ;dp110 CAAX glial overproliferation ( Figure 2M, data not shown). The gross neural morphogenesis defects observed in repo.dEGFR l ;dp110 CAAX brains may be secondary to glial overproliferation since these defects were largely prevented by Dap or Rbf1 co-overexpression ( Figure 2J-M). In repo.dEGFR l ;dp110 CAAX animals, other mutant glia outside of the brain, such as peripheral glia, also became highly proliferative and invasive, and these defects, too, were corrected by Rbf1 or Dap overexpression (data not shown). In controls, Rbf1 or Dap overexpression in wild-type glia inhibited proliferation ( Figure S4), reducing numbers of glia by approximately half. Together, these results suggest that repo-Gal4 glia undergo at least one round of cell division, consistent with published studies [27,28], and this proliferation becomes prolonged by constitutive coactivation of EGFR and PI3K signaling.

Coactivation of EGFR and PI3K Does Not Elicit Neoplasia in All Neural Cell Types
The phenotype triggered by coactivation of EGFR and PI3K in glia is distinct from other Drosophila brain-overgrowth mutant phenotypes, which involve accretion of excess neurons or neuroblasts [21,29]. repo.dEGFR l ;dp110 CAAX cells were not transformed into neurons or neuroblasts as they lacked expression of the Elav and Miranda markers ( Figure S6). Lineage-tracing with a Su(H)-lacZ neuroblast reporter showed that excess dEGFR l ; dp110 CAAX glia did not express LacZ, and thus are not directly derived from larval neuroblasts (data not shown). Moreover, constitutive EGFR-Ras and PI3K signaling does not elicit overgrowth in all neural cell types, as assessed with defined celltype specific Gal4-drivers (Table S2). For example, dRas85D V12 overexpression in fly neurons causes defects in fate specification, patterning, and apoptosis [30,31]. Co-overexpression of dEGFR l or dRas85D V12 with dp110 CAAX in neurons (elav-Gal4, scratch-Gal4, OK107-Gal4, and Appl-Gal4) and neuroblasts/neuronal precursors (pros-Gal4, wor-Gal4, and 1407-Gal4) did not induce overgrowth ( Figure S7, data not shown), even with increased transgene expression from Gal4 amplification [32]. In fact, broad cooverexpression of dEGFR l and dp110 CAAX in neuroblasts and neuronal precursors (pros-Gal4) reduced brain size, perhaps because signaling through these pathways stimulates precocious cell cycle exit of neuronal precursors, as in the developing eye [33]. Furthermore, transient expression of dRas85D V12 or dEGFR l and dp110 CAAX in embryonic glia (gcm-Gal4) also failed to promote glial overgrowth ( Figure S7) [27], demonstrating that sustained activation of these pathways is required for glial overproliferation. Certain glial subtypes, such as oligodendrocyte-like neuropil glia (Eaat1-Gal4), some astrocyte-like cortex glia (Nrv2-Gal4), and peripheral perineurial glia (gli-Gal4) also failed to become neoplastic in response to EGFR-Ras and PI3K ( Figure S7, data not shown) [34][35][36]. Therefore, neoplastic proliferation is not a uniform cellular response to EGFR and PI3K.

Coactivation of EGFR and PI3K Inhibits Cell Cycle Exit in Glia
Since repo.dEGFR l ;dp110 CAAX animals die in 5-7 days, we assessed the proliferative potential of mutant glia using an abdominal transplant assay, a classic test of tumorigencity in flies [37]. Brain fragments from repo.dEGFR l ;dp110 CAAX and wild-type larvae were transplanted into young adults. Wild-type transplants grew and survived over 1-6 weeks, but produced few glia ( Figure 3A and 3C). dEGFR l ;dp110 CAAX mutant glia survived and proliferated into massive tumors that filled the hosts' abdomens, often causing premature death ( Figure 3B). Tumors were composed of small glial cells with little cytoplasm ( Figure 3D-F). Tumors also contained trachea embedded throughout their mass ( Figure 3D and 3E and Video S1), suggesting that tumors stimulated growth of new trachea or enveloped existing trachea, perhaps in a process akin to tumor angiogenesis. The leading edges of the tumors harbored individual cells invading nearby repo.dEGFR l ;dp110 CAAX brains (B,C) surface glia (SG) and superficial cortex glia (brighter GFP, arrowheads) form thick multicellular aggregations throughout the brain and often adopt a spindle-shaped morphology (arrowheads). In repo.dEGFR l ;dp110 CAAX brains, deeper cortex glia (C) can still form fine projections, but more often lack these and create loose aggregates of abnormally shaped cells (asterisks). (D,E) BrdU labeling (red) of Sphase cells. The prominent replicative cells in normal brains (D) are neuroblasts ('NB') and optic lobe neural precursors ('NP'), identified by their stereotyped position and/or staining for HRP (faint blue). The prominent replicative cells in repo.dEGFR l ;dp110 CAAX brains (E) are glia ('G'), which do not stain for HRP. 2 mm optical sections. (F-I) dCyclinE (F,G) and dCyclinB (H,I) expression (red) shown alone (left panels) and with Repo (blue nuclei) and CD8GFP (green, membranes/cytoplasm) glial markers (right in each panel). 'G' denotes representative glia, 'NB', representative neuroblasts. In wild-type (F, H), glia rarely expressed dCyclinE or dCyclinB, while neuroblasts ('NB') showed predominant expression. dCyclinE expression is primarily nuclear in dEGFR l ;dp110 CAAX glia (G), visible as purple in overlay (right panel), although not all mutant glia express dCyclinE. Inset in (G) shows dCyclinE expression in polyploid glia ('ppG'). (I) dCyclinB expression in dEGFR l ;dp110 CAAX glia is primarily cytoplasmic, visible as yellow in overlay (right panel). ppG in repo.dEGFR l ;dp110 CAAX brains do not express dCyclinB, as seen by the absence of yellow or purple in overlay. tissues, such as the ovary ( Figure 3F and Video S2), which is consistent with the ectopic expression of active dMMP1 observed in dEGFR l ;dp110 CAAX glia in the larval brain ( Figure S3). However, some tissues, such as the gut, did not contain metastases, implying some degree of selective invasion. Thus, once unconstrained by the larval life cycle, dEGFR l ;dp110 CAAX glia fail to exit the cell cycle, continue to proliferate, and form highly invasive tumors, all properties of human cancer cells.

Coactivation of EGFR-Ras and PI3K Creates Invasive Neoplastic Glia
To explore the invasive potential of mutant glia, we used FLP-FRT clonal analysis, a technique in which discrete clones of mutant cells are induced in otherwise normal tissues, a situation analogous to somatic tumorigenesis. We used a heat-shock-driven FLP-recombinase to catalyze mitotic recombination between FRT-bearing chromosomes such that a daughter cell (and all of its clonal progeny) initiated expression of GFP and UAScontaining transgenes only in repo-Gal4-expressing cells [12]. Clones were induced late in development, from mitotic founder cells, and were examined in adults. We could not definitively determine if clones were derived from single cell events since our study of these clones was retrospective, but given the frequency of control clone induction, many mutant clones likely originated from single cells.
To coactivate EGFR-Ras and PI3K in glial clones, dRas85D V12 , dEGFR Elp , or dEGFR l was overexpressed within dPTEN 2/2 glia, using repo-Gal4. We observed overgrown and invasive dRas85D V12 ;dPTEN 2/2 and dEGFR Elp ;dPTEN 2/2 clones ( Figure 4E-G). dEGFR;dPTEN 2/2 clones were less affected ( Figure 4F and 4G), consistent with the larval overexpression of dRas85D V12 giving more severe growth phenotypes. Tumor-like overgrowth was only observed in dEGFR-dRas85D;dPTEN 2/2 cells, illustrating that chronic EGFR-Ras activation and PTEN loss can cause cell-autonomous over-proliferation. Cells from these double mutant clones appeared to invade the brain, typically following fiber tracts, and sometimes induced the formation of trachea ( Figure 4E). Tumor-like growths of dEGFR-dRas85D;dP-TEN 2/2 cells often penetrated deep into the brain, as exemplified by Videos S3 and S4 which show an animated 88 mm thick confocal z-stack of the dRas85D V12 ;dPTEN 2/2 clone in Figure 4E compared to a 16 mm thick z-stack of a wild-type control clone in Figure 4A. These phenotypes were reminiscent of invasion and angiogenesis in human gliomas [24]. We more commonly observed smaller dEGFR-dRas85D;dPTEN 2/2 clones composed of relatively differentiated, enlarged glia with diffusely invasive projections ( Figure 4G); these clones likely derive from glia that differentiated prior to achieving sufficient EGFR-Ras transgene expression, and are consistent with findings that not all glial subtypes become neoplastic.
Since neoplastic larval glia were concentrated in the outer anterior central brain and developing optic lobe, they may be derived from glial progenitor cells present in these regions [27,28,39]. To create clones from a discrete subpopulation of glial progenitors, we used an eyeless(ey)-promoter driven FLPrecombinase (ey-FLP), which is active in ey-expressing glial progenitors in the optic lobe [27]. Single mutant ey-FLP clones of dPTEN 2/2 , dRas85D V12 , or dEGFR Elp cells contained a modest number of excess and abnormal glia relative to wild-type controls ( Figure 4H-J and Figure S8). In contrast, the double mutant dRas85D V12 ;dPTEN 2/2 , dEGFR Elp ;dPTEN 2/2 , or dEGFR l ;dPTEN 2/2 ey-FLP clones, which emerge as approxi-mately tens of cells in 3 rd instar larval brains ( Figure 4K), became large invasive tumors composed of hundreds to thousands of cells in adults ( Figure 4L and Figure S8).

EGFR;PI3K Induces Glial Neoplasia via Pnt and Stg Expression
To address the function of EGFR-Ras and PI3K in glioma, we analyzed the genetic basis of glial pathogenesis in our repo. dEGFR l ;dp110 CAAX model, as this model shows robust neoplasia, similarity to human tumor genotypes, and sensitivity to dEGFR and dPTEN gene dosage (data not shown).
EGFR-Ras signaling can promote proliferation through Erk kinase-mediated induction of nuclear targets. In repo. dEGFR l ;dp110 CAAX brains, mutant glia showed high levels of nuclear, activated di-phospho-Erk relative to wild-type glia in control brains ( Figure 5A and 5B). In flies, Erk activity can induce expression of PntP1, an ETS-family transcription factor encoded by the pointed (pnt) locus [40,41]. PntP1, which is expressed in embryonic glia and is required for their normal development [40], was upregulated in dEGFR l ;dp110 CAAX glia ( Figure 5C and 5D). High levels of PntP1 can be detected normally in neuronal progenitors (data not shown), suggesting that it promotes a proliferative progenitor state. In developing eye tissue, EGFR-Ras-Erk signaling induces Pnt proteins to stimulate G2-M cell cycle progression through direct upregulation of Stg (Cdc25 ortholog) expression [41]. Glial-specific RNAi knock down of pnt reduced Stg expression and completely suppressed dEGFR l ;dp110 CAAX neoplasia ( Figure 5E-H), demonstrating that Pnt proteins are required for both Stg expression and neoplastic overproliferation in dEGFR l ;dp110 CAAX glia. Notably, in repo.pnt dsRNA ; dEGFR l ;dp110 CAAX brains, glia maintained their fate, as evidenced by repo-Gal4 and Repo expression.
Stg itself was rate limiting for glial neoplasia. Reduction of stg with a mutation or a stg dsRNA partially suppressed the repo. dEGFR l ;dp110 CAAX phenotype, whereas overexpressed Stg synergistically enhanced neoplasia ( Figure 5E and Figure S9, data not shown). In contrast, Stg overexpression alone increased glial cell numbers approximately 2-fold, and could not induce neoplasia when combined with PI3K effectors (data not shown). Thus, dEGFR l ;dp110 CAAX induces neoplasia via coordinated stimulation of G1-S entry through dCyclinE, and G2-M progression through Stg, both of which are EGFR-Ras dependent outputs [41,42].

EGFR;PI3K Glial Neoplasia Requires Akt and Tor Activation and FoxO Inactivation
We sought to determine which PI3K effectors contribute to the repo.dEGFR l ;dp110 CAAX phenotype. Genetic reduction of dAkt, a major target of PI3K signaling, with a dAkt dsRNA or a mutant allele strongly suppressed repo.dEGFR l ;dp110 CAAX glial neoplasia ( Figure 6A-C, data not shown). Therefore, Akt is necessary for the outcome of EGFR-PI3K coactivation. Many Akt effectors are implicated in glioma, and we tested orthologs of these loci in our model (Table S3).
TORC1, a complex including Tor and the Raptor regulatory protein, drives cellular growth by stimulating protein synthesis through its effectors S6 kinase and the eIF-4E translation initiation factor [47]. Akt and Erk stimulate TORC1 through phosphorylation and inactivation of the TSC1-TSC2 protein complex, which activates Rheb, and stimulates TORC1 kinase activity. We tested TORC1 function by glial-specific overexpression of dsRNAs for the single Drosophila orthologs of Raptor (dRaptor), S6-kinase (dS6K), and eIF4E (deIF4E); these all significantly reduced accumulation of dEGFR l ;dp110 CAAX mutant glial cells, but only caused mild glial hypoplasia in controls ( Figure 6I-K and Figure  S10). Co-overexpression of d4EBP, a deIF4E antagonist and dFoxO target gene, also blocked glial neoplasia ( Figure 6L). Glialspecific RNAi of dTSC1 enhanced repo.dEGFR l ;dp110 CAAX glial overgrowth ( Figure 6H). However, overexpression of dTSC1 dsRNA , dRheb, activated dS6K (dS6K act ), or deIF4E alone did not produce glial overproliferation, even though these constructs can mimic TORC1 activation ( Figure S10, data not shown) [47,49]. Thus, TORC1 activity is necessary for EGFR-PI3K-driven glial neoplasia, but is not sufficient. Moreover, neither deIF4E nor dS6K act produced glial neoplasia when cooverexpressed with dEGFR l (data not shown), illustrating that additional dTor-dependent outputs synergize with dEGFR signaling to drive neoplasia.

Myc and Max Drive EGFR;PI3K Glial Neoplasia via CyclinD-Cdk4
dTor coordinates increased translation, mediated by dS6K and deIF4E, with expression of cell cycle regulators and ribosomal components, through dMyc, the single Drosophila ortholog of the Myc bHLH transcription factors [49]. Within developing epithelial tissues, dMyc is required for TORC1-dependent growth and can substitute for dTor activity [49,50]. Myc protein levels can also be posttranslationally upregulated by EGFR-Ras-Raf signaling [16,51]. Thus, we suspected that dMyc might mediate signal integration between EGFR-Ras and PI3K. dEGFR l ;dp110 CAAX glia showed high levels of nuclear dMyc compared to wild-type glia ( Figure 7A and 7B). dMyc was also highly expressed in wild-type neuroblasts ( Figure 7A), suggesting that dMyc promotes a proliferative progenitor state [21]. Genetic reduction of dMyc with a dsRNA or a single loss-of-function allele strongly suppressed dEGFR l ;dp110 CAAX glial neoplasia ( Figure 7D, data not shown). In fact, some dMyc +/2 ; repo. dEGFR l ;dp110 CAAX animals were rescued to viability, indicating that dMyc is an essential rate-limiting output of EGFR-PI3K coactivation.

Myc Overexpression or Rb Loss Synergizes with EGFR
dMyc is necessary for glial neoplasia, but is not sufficient when overexpressed alone ( Figure 8C). dMyc-overexpressing glia showed polyploidy ( Figure 8C, data not shown), indicating that these cells undergo DNA replication without mitosis, but require additional signals for cell cycle progression. In contrast, cooverexpression of dMyc with dEGFR l produced a phenotype on par with that of repo.dEGFR l ;dp110 CAAX ( Figure 8D), indicating that dMyc overexpression can substitute for PI3K activation and promote neoplasia when combined with EGFR signaling.
Given that the dMyc targets dCyclinD-dCdk4 are required for dEGFR l ;dp110 CAAX neoplasia, we tested whether dCyclinD-dCdk4 overexpression could cooperate with dEGFR l . Additionally, we tested loss of Rbf1, the only known dCdk4 substrate that controls proliferation [23]. repo.dCyclinD;dCdk4;dEGFR l and repo.Rbf1 dsRNA ;dEGFR l animals showed glial overgrowth, but did not accumulate as many cells as repo.dEGFR l ;dp110 CAAX animals ( Figure 8E and 8F). Thus, glia likely require additional dp110 or dMyc effectors to undergo full neoplastic proliferation. Other known PI3K and dMyc-dMax target genes that promote proliferation include ribosomal proteins and translation regulators [49,52], such as eIF4E, which is highly expressed in dEGFR l ;dp110 CAAX glia and required for neoplasia ( Figure 6 and Figure S11).

Discussion
We show that constitutive coactivation of EGFR-Ras and PI3K signaling in Drosophila glia and glial precursors gives rise to neoplastic, invasive cells that create transplantable tumor-like growths, mimicking human glioma, and mirroring mouse glioma models. This represents a robust organotypic and cell-type specific Drosophila cancer model in which malignant cells are created by mutations in the signature genes and pathways thought to be driving forces in a homologous human cancer. This was not necessarily an expected result since fly and human glia show many biological differences despite displaying important similarities [9,55,56]. Through genetic analysis of our model, we identified crucial downstream effectors of EGFR and PI3K signaling, many of which are mutated and/or activated in human glioma. These effectors act in a combinatorial network to coordinately stimulate cell cycle entry and progression, block cell cycle exit, and promote inappropriate cellular growth and migration ( Figure 8J). Pathways within this network, while interdependent, act synergistically, rather than additively. Thus, Drosophila shows evolutionary conservation of oncogene cooperation.
At least four pathway circuits are necessary for glial neoplasia initiated by EGFR and PI3K signaling, including dRas and dMyc circuits, which induce dCyclinE and dCyclinD to drive cell cycle entry, a Pnt circuit, which induces Stg to promote cell cycle progression, and a Tor-eIF4E-S6K pathway, which provides protein translation necessary for proliferation and growth ( Figure 8J). When activated individually, these pathways fail to elicit glial neoplasia, implying a requirement for coordinated stimulation of multiple effectors and inactivation of negative regulators. Orthologs for many of the genes within these pathways, such as dRictor, are implicated in human glioma, although specific roles for some, such as ETS transcription factors, have not been defined despite their expression in glioma [2,57,58]. While many of these genes are known EGFR and PI3K pathway components, we did not necessarily expect them to be required for EGFR and PI3K dependent glial neoplasia. Indeed, we have tested many other pathway components and outputs, such as Jun kinase, that did not significantly suppress repo.dEGFR l ;dp110 CAAX phenotypes upon reduced function (unpublished data).
Coactivation of EGFR and PI3K signaling upregulates dMyc, which is necessary for glial neoplasia. This is consistent with findings that, in flies and mammals, EGFR-Ras, PI3K, and Tor signaling upregulate Myc protein levels [16,49,51,59,60]. Myc oncogenes are well-known to cooperate with RTK-Ras signaling to drive neoplastic transformation [51], and we demonstrate that this property of Myc is conserved in flies. We also observed sensitivity to reduced Myc gene dosage in our glioma model, which has also been recently documented in a mouse model of PTEN-dependent glioma [61]. c-myc is commonly amplified in gliomas [62], implying that Myc is rate limiting, and c-myc amplification may be selected for this reason. D-cyclins, established Myc target genes, and Cdk4 are also commonly amplified and/or overexpressed in gliomas [1,51]. We observed dMycdependent dCyclinD overexpression, and a requirement for dCyclinD-dCdk4 in repo.dEGFR l ;dp110 CAAX neoplasia, although dCdk4 itself is not required for normal glial proliferation. Together with our analysis of TORC2, this illustrates that oncogenic EGFR-PI3K co-opts effectors that do not control normal glial development. Similarly, cdk4 2/2 mutant mice show normal proliferation in many tissues, but are resistant to ErbB-2-driven breast cancers [63,64]. Our data argue that Cdk4 activity is a key tumor-specific rate-limiting output of EGFR and PI3K signaling in glioma as well.
In contrast to glia, coactivation of EGFR-Ras and PI3K in neuroblasts, which are fly neural stem cells, does not promote unchecked proliferation, despite the fact that neuroblasts express dMyc and are capable of undergoing neoplastic transformation in response to other genetic mutations [21]. Thus, in Drosophila, neither a neural stem cell fate nor Myc activity confer competence to undergo EGFR-PI3K neoplastic transformation. Rather, our results suggest that neoplastic cells arise from committed glial progenitors: dEGFR-dRas85D;dPTEN 2/2 clones derived from progenitor cells produce large tumors, and anaplastic cells in repo.dEGFR l ;dp110 CAAX brains are concentrated in regions enriched for glial progenitors. Notably, regulated developmental signaling through the EGFR pathway promotes proliferation of normal Repo-expressing glial progenitors [27], and our results show that constitutive EGFR and PI3K signaling prolongs this proliferative progenitor state. Further studies of Drosophila glial progenitors and glioma-like cells may illuminate the cellular origins of human gliomas, which are thought to arise from progenitor-like glial cells. Moreover, our results argue that celltype specific factors govern glial neoplasia. One such factor may be Dap, the single p21/p27 ortholog, which is normally expressed in only 5% of all glia ( Figure S4). Perhaps glial progenitors do not express Dap, whereas neuronal progenitors do [26], and this underlies susceptibility to transformation by EGFR-Ras and PI3K. Dap is highly regulated in a cell-type specific manner [26], and studies of Dap regulation in glia may further illuminate the genetic Figure 8. Myc overexpression or Rb loss synergizes with EGFR. (A-G) 2 mm optical sections of larval brain hemispheres from late 3 rd instar larvae, displayed at the same scale. 20 mm scale bars. Frontal sections, midway through brains. Anterior up; midline to left. Glial cell nuclei are labeled with Repo (red); glial cell bodies and membranes are labeled with CD8GFP (green) driven by repo-Gal4; HRP (blue) reveals neuropil at high intensity and neuronal cell bodies at low intensity. repo.dMyc brains (C) contain many glia with enlarged nuclei (arrows). repo.dEGFR l ;dMyc (D) brains have excess glia (red) throughout the brain, and these glia have very compact cell bodies (green) relative to wild-type glia (B). repo.dEGFR l ;dCyclinD;dCdk4 (E) and repo.dEGFR l ;Rbf1 dsRNA (F) brains display increased numbers of glia (red) relative to wild-type (B), but show a less severe glial phenotype and a more normal neuronal compartment (blue) than either repo.dEGFR l ;dp110 CAAX (A) or repo.dEGFR l ;dMyc (D). The combination of repo.Rbf1 dsRNA ;dEGFR l ;dp110 CAAX (G) enhances neoplasia, leading to a dramatic increase in aberrant glia. (H,I) dCyclinE expression (red) in repo.dEGFR l ;dp110 CAAX glia (H) and in repo.dEGFR l ;dp110 CAAX ;Rbf1 dsRNA glia (I), alone and overlaid with the Repo glial marker (blue, right panels). dCyclinE is more strongly and more broadly expressed in neoplastic glia upon Rbf1 loss, as evidenced by the increased number and intensity of dCyclinE-expressing glial nuclei in ( origins of glioma, especially given that lack of p21 expression may underlie the tumorigenic response of mammalian glial progenitors to constitutively active EGFR [65].
While EGFR-Ras and PI3K are commonly upregulated in gliomas and experimental models demonstrate that these pathways are required for tumorigenesis, therapies that target EGFR and PI3K signaling have proven disappointing. This discrepancy between clinical and experimental data has many possible explanations. For example, recent studies have demonstrated that EGFR inhibitors are attenuated by particular mutations found in glioma cells, such as PTEN loss or RTK co-amplification [2]. Addressing these and other possibilities remains a challenge that dictates a need for new experimental models. The results presented here establish Drosophila as a viable model system for the study of glioma, offering a complex organismal system for rapidly identifying and evaluating therapeutic targets using genetic approaches. Such a system may be especially useful for distinguishing those genetic mutations and pathways that drive tumorigenesis from the large number of genes that show mutations and altered expression in glioblastomas uncovered by recent genomic analyses of patient samples [66,67]. Our studies have already identified key rate-limiting genes, such as dCyclinE, Stg, and dMyc, and genes only required for abnormal neoplastic glial proliferation, such as dSin1, dRictor, and dCdk4, which may represent important therapeutic targets in human gliomas.

Fly Stocks, Genetics, and Culture Conditions
Flies were cultured at 25uC. All genotypes were established by standard genetics. To assess larval brain overgrowth phenotypes, embryos were collected for 6-24 hrs, grown for 120-140 hrs, and wandering 3 rd instar larvae were selected for dissection.
Stocks were obtained from the Bloomington Stock Center unless otherwise noted. Other than UAS-PTEN dsRNA lines from Bloomington, all UAS-dsRNA lines were obtained from the VDRC stock center [11]. The following stocks were obtained from other investigators: UAS-dEGFR l (T. Schubach), UAS-dEGFR Elp , UAS-

Abdominal Transplants
Larval brains were dissected into sterile PBS, washed, and cut into fragments. Abdominal incisions were made in virgin female hosts and single brain fragments were inserted. Hosts were cultured for 1-6 weeks, dissected and fixed in 4% paraformaldehyde, incubated in 10% sucrose and embedded in O.C.T. Thick 50 mm sections were stained as described below.

Clonal Analysis
For hs-FLP clones, genotypes are indicated in figure legends. Flies were initially grown at 18uC or 20uC to minimize spontaneous clones, which occurred at a low frequency during late larval-pupal stages. 3 rd instar larvae, 0-48 hr pupae, or 0-2 day old young adults were treated with heat shock to induce clones and subsequently cultured at 25uC for 1-4 weeks. For ey-FLP clones, flies were cultured at 25uC continuously.

Immunohistochemistry and Imaging
Larval tissue was fixed for 30-50 minutes in 16PBS 4% paraformaldehyde. Adult brains were fixed for 1-2 hr in 16PBS 4% paraformaldehyde or in PLP with 2% paraformaldehyde. For BrdU labeling, larvae were cultured in food with 1 mg/ml BrdU for 4-6 hrs, and fixed larval brains were treated with 2 N HCl for 30 minutes followed by DNase for 1 hr. Stains were performed in 16PBS 10% BSA with 0.3% Triton-X100 for larval brains and 0.5% Triton-X100 for adult samples.
The following antibodies were obtained from the Developmental Studies Hybridoma Bank and diluted 1:5-1:10: 8D12 anti-Repo, anti-dMMP1, anti-dCyclinB, anti-Elav, and 40-1a anti-lacZ. Larval and/or adult brains were also stained with rabbit anti-Repo (G. Technau, 1:500), rat anti-dCyclinE (H. Richardson Brains were imaged as whole mounts on a Zeiss LSM 510 confocal system. Images were analyzed in Zeiss LSM Image Browser and processed in Photoshop CS3. For experiments in which protein levels were compared between genotypes, all sample preparation, histochemistry, imaging, and image processing was performed in parallel in the same manner. Figure S1 Diagram of Drosophila EGFR mutant forms. Proteins are shown as horizontal bars along which functional domains are indicated, and the locations of alterations in mutant forms are noted. Wild-type human EGFR and Drosophila EGFR (dEGFR) show extensive conservation, with 55% homology in the kinase domain and 41% homology in the ligand-binding portion of the extracellular domain, which includes extensive cysteine repeats. The signal peptide is labeled in yellow. Following the signal peptide, the entire cytoplasmic domain is replaced with the lambda dimerization domain in dEGFR l , which causes constitutive activation. Both human EGFR and dEGFR show extensive cysteine repeats in the extracellular domain, indicated in blue. The tyrosine kinase domain is labeled in magenta. The Elp mutant form of dEGFR contains an A887T substitution in the N-lobe of the kinase domain, which causes constitutive activation. Repo (red). CD8GFP (green), driven by the repo-Gal4 driver, labels glial cell bodies and membranes. An anti-HRP counterstain (blue) reveals neuropil (neuronal fiber tracts) at high intensity and some cell bodies of neurons and neuronal precursors at low intensity, and this varies slightly according to exact plane of section and mutant phenotype. repo.dEGFRl;dp110CAAX (A), repo. dRas85DV12;dPTENdsRNA (B), and repo.dRafgof;dp110CAAX (C) brains show increased numbers of glia relative to wild-type (D), repo.dRas85DV12 alone (E), or repo.dPTENdsRNA alone (F). Compared to wild-type (D), repo.dEGFRwild-type (H) brains show reduced neurons (HRP, low intensity blue), which renders glia more densely packed and the entire brain smaller than normal. However, repo.dEGFRwild-type brains (H) do not show a substantial increase the number of glia compared to wild-type (D). Brains of repo.dEGFRElp 2nd instar larvae (I) show substantial neuron loss, brain malformation, and excess glia for that stage of In wild-type brains, neuronal fibers and neurons (blue) are enveloped by glial processes (green) to give the brain a honeycombed appearance. In repo.dEGFRl;dp110CAAX brains, neurons ('N') lack GFP or Repo expression, in contrast to glia ('G'). (C-D) Elav expression (red) in wild-type (C) and repo. dEGFRl;dp110CAAX brains (D), shown alone (left panels) and overlaid (right panels) with the CD8GFP (green) glial marker and HRP (blue) neuronal marker. In both genotypes, Elav is highly expressed in neurons ('N') but absent from glia (green cell bodies indicated by white arrows). (E-F) Miranda expression (red) in wildtype (E) and repo.dEGFRl;dp110CAAX (F) brains, shown alone (left panels) and overlaid (right panels) with CD8GFP (green) glial marker and HRP (blue) neuronal marker. In both genotypes, Miranda is highly expressed in neuroblasts ('NB'), which are identified by their characteristic positions and large cell bodies, but is absent from glia (green cell bodies, indicated by white arrows). in glial cell nuclei in all genotypes. An HRP counter-stain (blue) reveals neuropil at high intensity and cell bodies of neurons and some neuronal precursors at low intensity, and this varies slightly according to variance in exact section plane. (A-D) Wild-type (A), dEGFR l ; dp110 CAAX overexpressed using the 1407-Gal4 neuroblast driver (B), dEGFR l ; dp110 CAAX overexpressed using the pros-Gal4 neural driver (C), and dEGFR l ;dp110 CAAX overexpressed using the gcm-Gal4 embryonic glial driver (D). In all cases, dEGFR l ;dp110 CAAX overexpression did not induce neoplastic overgrowth of glial or neuronal cell types, as determined by brain size and cell-type specific stains. (E,F) Cytoplasmic GFP (green), driven by the Nrv2-Gal4 driver, labels cell bodies and cytoplasmic processes, as seen in wild-type (E). dEGFR l ;dp110 CAAX overexpressed using the Nrv2-Gal4 glial driver (B). Nrv2-Gal4 is expressed by post-mitotic cortex glia, which proliferate somewhat in response to dEGFR l ;dp110 CAAX to cause slight brain enlargement, but do not become neoplastic like repo.dEGFRl;dp110CAAX glia.  Figure S9 Stg overexpression exacerbates EGFR-PI3K glial neoplasia. 2 mm optical sections of larval brain hemispheres from late 3rd instar larvae, displayed at the same scale. 20 mm scale bars. Frontal sections, midway through brains. Anterior up; midline to left. Glial cell nuclei are labeled with Repo (red); glial cell bodies and membranes are labeled with CD8GFP (green) driven by repo-Gal4. An HRP counter-stain (blue) reveals neuropil at high intensity and neuronal cell bodies at low intensity. HRP stains varied between the two samples according to effects of mutant glia and slight variance in section plane. Stg cooverexpression with dEGFR l ;dp110 CAAX (B) enhances neoplasia compared to dEGFR l ;dp110 CAAX (A), leading to a dramatic increase in aberrant glia and yielding a phenotype similar to that of repo.Rbf1dsRNA;dEGFRl;dp110CAAX (see Figure 8G). Geno-types: (A) UAS-dEGFRl UAS-dp110CAAX/+; repo-Gal4 UAS-CD8GFP/+ (B) UAS-dEGFRl UAS-dp110CAAX/+; UAS-stg/+; repo-Gal4 UAS-CD8GFP/+. Found at: doi:10.1371/journal.pgen.1000374.s009 (6.27 MB TIF) Figure S10 Knock-down and/or overexpression of dTor and dTor effectors. 2 mm optical sections of larval brain hemispheres from wandering 3rd instar larvae, approximately 130hr AED, all displayed at the same scale. 20 mm scale bars. Anterior up; midline to the left. Frontal sections; midway through brains. Repo (red) marks glial cell nuclei; CD8GFP (green), driven by repo-Gal4, labels glial cell bodies and membranes. An HRP-counter stain (blue) reveals neuropil (neuronal fiber tracts) at high intensity and some cell bodies of neurons and neuronal precursors at low intensity, and this varies slightly according to exact plane of section, brain orientation, and mutant phenotype.  Figure S11 Validation of dsRNA constructs. 2 mm optical sections of larval brain hemispheres from wandering 3rd instar larvae. Frontal sections. Anterior up, midline to the left. Each individual staining pattern is shown alone (left panels) and with overlaid glial or neuronal markers (right panels). Repo (blue) in (A-J) marks glial cell nuclei. Glial cell bodies and membranes are labeled with CD8GFP (green) driven by repo-Gal4. In (K,L) an HRP counter-stain (blue) reveals neurons. Red marks histochemical stains for each indicated protein in repo.dEGFRl;dp110CAAX (A,C,E,G,I,K) and repo.dEGFRl;dp110CAAX with each indicated dsRNA construct (B,D,F,H,J,L). Each dsRNA construct was expressed with repo-Gal4 to yield glial-specific knock-down, which left protein expression in surrounding neuronal tissue intact. For the nuclear protein PntP1, knock-down was verified by the lack of glial-specific staining in the presence of the pntdsRNA (B), although PntP1 is present within neighboring neurons ('N'). For dAkt, deIF4E, and dS6K, gene knock-down was confirmed using antibodies for total protein. Glial-specific reduction in gene expression by dAktdsRNA and deIF4EdsRNA is highlighted by white outlines in (D) and (F). dAkt protein is higher in neuronal tissue ('N') in both repo.dEGFRl;dp110CAAX (C) and repo.dAktdsRNA;-dEGFRl;dp110CAAX (D). deIF4E is high in dEGFR l ;dp110 CAAX glia (E) and low in neurons (E,F). repo.deIF4EdsRNA;-dEGFRl;dp110CAAX brains (F) have reduced glial eIF4E. Reduced glial dS6K protein, which is diffusely cytoplasmic, is observed in repo.dS6KdsRNA;dEGFRl;dp110CAAX brains (H) compared to repo.dEGFRl;dp110CAAX (G). Reduced glial dMyc expression in repo.dMycdsRNA;dEGFRl;dp110CAAX brains (J) is noted by white outlines, and the absence of purple nuclei in overlay (J, right panel), relative to repo.dEGFRl;dp110CAAX (I, right panel). dMyc is highly expressed in neuroblasts ('NB') in both samples (I, J), which do not express the dMycdsRNA. Glial Rbf1 protein is absent in repo.dRbf1dsRNA;dEGFRl;dp110CAAX brains (L) in contrast to neighboring neurons ('N') in ,repo.dEGFRl;dp110CAAX brains (K). Genotypes: (A,C,E,G,I,K) UAS-dEGFR, l UAS-dp110,CAAX/ +; UAS-CD8GFP/+; repo-Gal4/+ (B) UAS-dEGFRl UAS-dp110CAAX/+; repo-Gal4 UAS-CD8GFP/UAS-pntdsRNA (D) UAS-dEGFRl UAS-dp110CAAX/+; repo-Gal4 UAS-CD8GFP/UAS-dAktdsRNA (F) UAS-dEGFRl UAS-dp110CAAX/+; repo-Gal4 UAS-CD8GFP/UAS-deIF4EdsRNA (H) UAS-dEGFRl UAS-dp110CAAX/ UAS-dS6KdsRNA; repo-Gal4 UAS-CD8GFP/+ (J) UAS-dEGFRl UAS-dp110CAAX/+; UAS-dMycdsRNA/+; repo-Gal4 UAS-CD8GFP/+ (L) UAS-dEGFRl UAS-dp110CAAX/+; repo-Gal4 UAS-CD8GFP/UAS-Rbf1dsRNA.    Video S1 An animated 7.5 mm thick confocal z-stack of dEGFR l ; dp110 CAAX tumor cells derived from transplanted larval glia, pictured in Figure 3D. The depth of each frame is noted in the upper left-hand corner. 20 mm scale bar for x/y-axis. Transplanted mutant glia are labeled with membrane bound CD8GFP (green) and the Repo nuclear protein (blue). Actin staining (red, phalloidin) reveals abdominal anatomy of the host. Asterisks indicate trachea embedded in the pictured dEGFR l ;dp110 CAAX tumor, visible as hollow actin-positive (red) tubules running through tissues. Genotypes: Hosts were w1118 virgin females. Transplanted glia were UAS-dEGFRl UAS-dp110CAAX/+; UAS-CD8GFP/+; repo-Gal4/+. Video S2 An animated 27 mm thick confocal z-stack of dEGFR l ; dp110 CAAX tumor cells derived from transplanted larval glia, pictured in Figure 3F. The depth of each frame is noted in the upper left-hand corner. 20 mm scale bar for x/y-axis. Transplanted mutant glia are labeled with membrane bound CD8GFP (green) and the Repo nuclear protein (blue). Actin staining (red, phalloidin) reveals abdominal anatomy of the host. Arrowheads in indicate dEGFR l ;dp110 CAAX glial cells invading an ovary, distinguished by its characteristic actin staining (bright red). An asterisk indicates trachea embedded in this dEGFR l ;dp110 CAAX tumor, visible as a hollow actin-positive (red) tubule running through the tissue. Genotypes: Hosts were w1118 virgin females. Transplanted glia were UAS-dEGFRl UAS-dp110CAAX/+; UAS-CD8GFP/+; repo-Gal4/+. Video S3 An animated 88 mm thick confocal z-stack of a hs-FLP/FRT clone of dPTEN2/2;dRas85DV12 cells pictured in Figure 4E. The depth of each frame is noted in the upper left-hand corner. 20 mm scale bar for x/y-axis. Approximately half of the affected brain is displayed, with the midline to the left. The dPTEN2/2;dRas85DV12 mutant cells are labeled with membrane-bound CD8GFP (green) and penetrate through the entire depth of the brain, some of which is visible as red background. These dPTEN2/2;dRas85DV12 cells appear to originate in the optic lobe (right) and extend variable tendrils into the central brain (middle, left). Genotype: hs-flp1/+; FRT40A tubGal80/FRT40A PTEN2L117; repo-Gal4 UAS-CD8GFP/UAS-dRas85DV12. Video S4 An animated 16 mm thick confocal z-stack of a hs-FLP/FRT clone of wild-type cells pictured in Figure 4A. The depth of each frame is noted in the upper left-hand corner. 20 mm scale bar for x/y-axis. Midline to the left. This clone of normal glia is labeled with membrane-bound CD8GFP (green) and shows cells in a cell-body rich region of the central brain adjacent to the optic lobe. The clone is composed of 2 cells with extensive cytoplasmic projections. Genotype: hs-flp1/+;FRTG13 tubGal80/FRTG13 UAS-CD8GFP; repo-Gal4/+.