Recruited Cells Can Become Transformed and Overtake PDGF-Induced Murine Gliomas In Vivo during Tumor Progression

Background Gliomas are thought to form by clonal expansion from a single cell-of-origin, and progression-associated mutations to occur in its progeny cells. Glioma progression is associated with elevated growth factor signaling and loss of function of tumor suppressors Ink4a, Arf and Pten. Yet, gliomas are cellularly heterogeneous; they recruit and trap normal cells during infiltration. Methodology/Principal Findings We performed lineage tracing in a retrovirally mediated, molecularly and histologically accurate mouse model of hPDGFb-driven gliomagenesis. We were able to distinguish cells in the tumor that were derived from the cell-of-origin from those that were not. Phenotypic, tumorigenic and expression analyses were performed on both populations of these cells. Here we show that during progression of hPDGFb-induced murine gliomas, tumor suppressor loss can expand the recruited cell population not derived from the cell-of-origin within glioma microenvironment to dominate regions of the tumor, with essentially no contribution from the progeny of glioma cell-of-origin. Moreover, the recruited cells can give rise to gliomas upon transplantation and passaging, acquire polysomal expression profiles and genetic aberrations typically present in glioma cells rather than normal progenitors, aid progeny cells in glioma initiation upon transplantation, and become independent of PDGFR signaling. Conclusions/Significance These results indicate that non-cell-of-origin derived cells within glioma environment in the mouse can be corrupted to become bona fide tumor, and deviate from the generally established view of gliomagenesis.


Introduction
Gliomas are the most common primary brain tumors, characterized by the infiltration of neighboring brain structures and robust expansion during progression to a glioblastoma multiforme (GBM) [1,2]. Gliomas are frequently characterized by dysregulated signaling downstream of growth factor receptors such as EGFR, PDGFR, and IGFR, and elevated production of their corresponding ligands [3][4][5][6]. Nearly 30% of human gliomas show expression patterns that are correlated with PDGFR signaling [7], a pattern with prominent expression of OLIG2 and other genes involved in CNS development referred to as ''proneural'' [8]. PDGF ligands (A-D) are upregulated in at least a third of surgical glioma samples and human glioma cell lines [9][10][11][12][13]. The importance of PDGF signaling is underscored in genetically engineered rodent gliomas, where overproduction of human PDGFb ligand is sufficient to induce gliomagenesis in a dosedependent manner and allows to recapitulate the histologic, etiologic and pathobiologic character of the PDGF subset of human gliomas [14,15]. Additionally, infusion of PDGF into the ventricles induces proliferation of the SVZ, resulting in lesions with some characteristics of gliomas [16]. Similar to human gliomas, mouse gliomas are cellularly and molecularly heterogeneous.
Glioma progression in humans is associated with deletion of the INK4A/ARF locus and loss of PTEN expression resulting in activation of Akt [3][4][5][6]17]. The standard view of gliomagenesis is that sequential mutations occur and accumulate in cells derived from the glioma cell-of-origin. Indeed, many surgical GBM samples in patients appear clonal, with all tumor cells seemingly derived from the same cell; however, this may not necessarily mean they are derived from the cell-of-origin [18][19][20][21]. Cellular heterogeneity and reports of human gliomas comprised of several genetically unrelated clones suggest the possibility of oncogenic transformation in cells not derived from the glioma cell-of-origin [21][22][23][24][25][26]. The interconversion between human glioma subtypes upon recurrence and the existence of recurrent gliomas that lack mutations or deletions found in the original tumor further indicate the possibility for an expansion of an aggressive clone not arising from the cell-of-origin [8,27]. In fact, PDGF-induced gliomas arising in both adult and neonatal rats have been shown to contain normal stem and progenitor cells ''recruited'' into glioma mass and induced to proliferate, indicating that proliferative stem-like portions of the tumor can arise from normal progenitors. However, the precise nature and specific functional characteristics of these ''recruited'' stem or progenitor cells have not been described.
Genetic analysis of surgical samples of human gliomas merely provides retrospective static information with regards to tumor evolution; lineage tracing from the cell-of-origin cannot be done in humans. Moreover, identifying and distinguishing GBM cells from the surrounding stroma is not a trivial task -glioma cells are often defined histologically, demonstrating high mitotic indices, expression of stem or progenitor cell markers, abnormal global gene expression patterns, presence of genetic alterations, and the ability to serially transplant the disease [3,28,29]. To investigate cellular contributions and structural/functional characteristics of ''recruited'' cells in murine gliomas during tumor progression, we used RCAS/tv-a and the bacTRAP systems [30][31][32]. Defining tumor cells by histologic criteria, genetic analysis, global gene expression profiling and transplantation studies, we studied the clonality of mouse gliomas with respect to the cell-of-origin. Here we show that in murine gliomas induced by human PDGFb (hPDGFb), glioma progression can occur by expansion of the recruited cells, and that these cells unrelated to glioma cell-of-origin can be corrupted to become bona fide tumor.

Murine gliomas contain a recruited cell population
It has been recently shown that gliomas induced in adult or neonatal rats by hPDGFb-expressing retroviruses contain stem or progenitor-like cells expressing neural markers, that are contributing to glioma mass and are induced to proliferate by glioma environment [15,33]. However, the nature and fate of these cells not derived from the glioma cell-of-origin has not been extensively studied. While these cells proliferate and express immature markers, questions as to whether they are functionally important in glioma progression, remain dependent from the glioma cell-oforigin, and whether they represent bona fide tumor cells, have not been addressed. In order to study this phenomenon of cellular contribution to glioma heterogeneity, we employed lineage tracing, molecular analysis and functional characterization of non-cell-of-origin derived cells using two different mouse models of hPDGFb-driven gliomas.
We first infected Nestin-tv-a (Ntv-a) mice with an RCAS vector expressing hPDGFb and eGFP (RCAS-hPDGFb-HA-SV40-eGFP, ''RCAS-PSG'') ( Figure 1a-i) [14,30,34]. RCAS/tv-a is a retroviral system that allows for lineage tracing from the nestinpositive glioma cell-of-origin, and the use of this specific vector labels the cell-of-origin and its progeny with eGFP expression. Consistent with the previously published results from others and us, RCAS-PSG induced high-and low-grade gliomas with typical oligodendroglioma histology, containing proliferating cells expressing olig2 and other markers of neural cell lineages (Figure 1f Figure S3a). Many but not all of these cells were derived from the cell-of-origin and expressed viral eGFP (''progeny cells''), while others appeared to not be infected with the RCAS-PSG virus (''recruited cells''), as they did not express eGFP or hPDGFb, with hPDGFb expression determined by immunostaining for hemagglutinin epitope tag included in RCAS-PSG vector ( Figure S2a-f) [14][15][16]33]. In comparison to the previously published RCAS-hPDGFb vectors, the levels of hPDGFb expression by RCAS-PSG were 3-4 fold lower (data not shown) [14]. We confirmed the correlation between eGFP expression and viral integration using RT-PCR (reverse transcription), real-time qPCR and FISH in murine glioma sections and eGFP-positive or eGFP-negative FACS-sorted glioma cell populations (Figure 1h,i; Figure S1e,f). In these tumors, histologically-defined pathognomic glioma structures (secondary structures of Scherer [35]) contained both progeny and recruited cells (Figure 1b-e; Figure S1a-d). We refer to these tumors as ''virally-induced gliomas''.
Modeling systems employing intracranial transplantation of glioma cells are frequently used in glioma research. To rule out the possibility of spontaneous proviral excision or loss of eGFP expression in the originally infected cells in the virally induced gliomas, we performed transplantations of acutely isolated nonfluorescent hPDGFb-driven mouse glioma cells derived from Ntva gliomas induced by the non-fluorescent RCAS-hPDGFb vector into the bacTRAP olig2 RP-eGFP reporter mice, expressing a fusion between ribosomal protein L10a and eGFP (RP-eGFP) under the Olig2 promoter ( Figure S3b-d) [14,31,32]. In these ''transplanted gliomas'', RP-eGFP-expressing host cells represent olig2-positive recruited cells, while transplanted glioma cells do not express eGFP. Similar to virally-induced gliomas, RP-eGFP-expressing host-derived recruited cells proliferated (stained for Ki-67), showed spindloid and/or typical oligodendroglial morphology, and contributed to the secondary structures of Scherer (Figure 1j-o). In virally-induced gliomas, olig2 progeny and olig2 recruited cells are strictly defined by the presence or absence of eGFP expression; in contrast, in transplanted gliomas, the injected non-fluorescent glioma cell population is a mixture of progeny and recruited cells. Thus, identification of recruited cells by RP-eGFP expression in this system likely underestimates the true contribution of recruited cells to the glioma mass. While virally-induced gliomas have been shown to approximate the human scenario for at least a subclass of human gliomas, the extent of similarity of transplanted gliomas to the patient tumors, as well as to other transplantation systems, has not been addressed [3]. An additional caveat of the bacTRAP system is that olig2 RP-eGFP expression in the adult murine brain labels not only progenitor cells, but also more differentiated oligodendrocytes [32].

Tumor suppressor loss expands the recruited population
As human gliomas progress, loss of tumor suppressor function is seen in nearly all cases [6]. It is possible that loss of tumor suppressors in the progeny or the recruited populations during glioma progression could give a preferential growth advantage to either of these cell populations. Therefore, to investigate the relative effects of tumor suppressor loss on progeny and recruited cells, we induced gliomas in Ntv-a mice with germline deletions of Ink4a, Arf and/or Pten, the latter achieved by Cre-mediated recombination of the floxed Pten alleles by RCAS-Cre [36,37]. As previously published, we found that loss of Arf and Pten, but not Ink4a, shortened tumor latency and increased incidence of murine GBMs, histologically defined by the presence of microvascular proliferation and pseudopalisading necrosis (Figure 2a; Figure  S4a,b) [3,36,37]. Of note, the contribution of the recruited cells to high-grade glioma structures was most prominent in the Ntv-a Ink4a/Arf -/-, Pten -/tumors (Figure 2b-f).
We then determined whether such robust expansion of the recruited cell population upon tumor suppressor loss is limited to modeling systems characterized by paracrine growth factor signaling, or whether it could apply to murine gliomas with cellautonomous activation of oncogenes. While mutations in RAS are not characteristic of a given human glioma subtype, we used overexpression of mutant RAS as a proxy to model human gliomas driven in a cell-autonomous manner [38]. For this purpose, we created RCAS-mRFP-SV40-Ras (RCAS-RSR) vector that expresses mutant KRASG12D and simultaneously marks the infected cells by expression of mRFP ( Figure S5a-e). While this vector was not robust at glioma initiation (,10% incidence), injections of RCAS-RSR and RCAS-Cre into Ntv-a Ink4a/Arf -/-Pten fl/fl mice could give rise to high-grade gliomas that did not express PDGFRa and contained regions of proliferating cells that did not express mRFP ( Figure S5d,e,g).
To further address the ability of the recruited cells characterized by tumor suppressor loss to contribute to high-grade glioma structures, we performed transplantation experiments using the bacTRAP system. Transplantation of non-fluorescent hPDGFbdriven glioma cells into Ink4a/Arf -/-Pten fl/fl bacTRAP olig2 RP-eGFP reporter mice resulted in formation of high-grade glioma structures comprised of both progeny and recruited cells ( Figure  S5h). Transplanted gliomas induced by hPDGFb-driven glioma cells in hosts with altered tumor suppressor function showed higher percentages of the overall recruitment than in the wild-type hosts, indicating that complete or partial tumor suppressor loss may enhance the ability of the cells to be recruited ( Figure 2g). Although gliomas induced by transplantation of non-fluorescent Pten-deleted Ras-driven murine glioma cells showed recruitment of proliferating host brain cells into the glioma mass ( Figure S5f,h), the overall amount of recruited cells into Ras-driven tumors and the number of tumors that showed large regions of recruitment was significantly less than that seen with the hPDGFb-induced gliomas (Figure 2g). Recruited cells in hPDGF-induced can overtake glioma mass during progression The above data indicates that murine glioma cells, especially those driven by PDGF signaling, recruit proliferating olig2expressing progenitors, and their contribution to high-grade glioma structures is enhanced upon homozygous loss of Ink4a/ Arf and Pten at glioma initiation. However, human gliomas undergo tumor suppressor loss stochastically as they progress. Stochastic tumor suppressor loss could in principle occur in any cell subjected to the pressures of glioma environment, regardless of its progeny or recruited status. To mimic this phenomenon, we used Ntv-a Ink4a/Arf +/-Pten +/fl mice infected with RCAS-PSG and RCAS-Cre. Given the constraints of our experimental system, formation of murine gliomas with high-grade architectural features would require loss of function of the remaining tumor suppressor allele, by either stochastic loss of heterozygosity (LOH) or epigenetic events. The rarity of such events would in turn result in the expansion of an altered cell population from a single cell, be it progeny or recruited. Although neither Ntv-a Ink4a/Arf +/-Pten +/fl nor Ntv-a Ink4a/Arf -/-Pten fl/fl mice generate spontaneous gliomas, it is likely that experimentally provided genetic hits may represent the rate limiting step for transformation and thus, transformation of recruited cells may be less common in human gliomas.
While the latency of gliomas induced in Ntv-a Ink4a/Arf +/-Pten +/fl mice by RCAS-PSG and RCAS-Cre was intermediate between the wild type and Ink4a/Arf -/-Pten fl/fl mice, the distribution of glioma grades at the point when the tumors caused mice to become moribund was similar to that of gliomas characterized by homozygous tumor suppressor loss at initiation (Figure 2a; Figure S4b). However, unlike the mixed progeny-recruited cell composition seen in gliomas null for Ink4a, Arf and Pten at tumor initiation, gliomas arising in mice initially heterozygous for these tumor suppressors frequently showed regional domination by either progeny or recruited cells (Figure 3a-d). In the most extreme cases, such glioma regions were almost fully comprised of eGFP-negative recruited cells, as verified by FACS and qPCR ( Figure 3a,b,e; Figure S5j,k). Of note, these regions dominated by recruited cells showed loss of function of the remaining wild type tumor suppressor alleles by LOH (in the case of Ink4a/Arf),or lossof expression at the protein level with or without LOH (in the case of Pten), as in human gliomas (Figure 3f; Figure S4c-e). While Ink4a/Arf loci were most frequently lost at the genetic level as determined by PCR, of 40 hPDGFb-driven gliomas arising in Ntv-a Ink4a/Arf -/-Pten fl/fl or Ntv-a Ink4a/Arf +/-Pten +/fl injected with RCAS-Cre, 28 did not express Pten; of these, 21 retained Pten allele by real-time PCR, while 7 lost it ( Figure S4c). Thus, while loss of Pten expression occurred in ,70% of all cases, loss of Pten at the DNA level occurred in 17.5% cases. There was a small but statistically significant difference with respect to latency medians depending on whether a given glioma was predominantly derived from progeny or recruited cells (two-tailed unpaired Student's t-test, p,0.006) ( Figure S5i). However, their variances largely overlapped (F test), and GBMs containing large areas with pseudopalisades predominantly comprised of recruited cells represented ,30% of all mouse GBM cases ( Figure 3g).

Recruited regions of hPDGF-induced gliomas can become independent of PDGF signaling
We then characterized tumors with regions dominated by recruited cells for expression of growth factor receptors commonly activated in human gliomas, including PDGFR, EGFR and IGFR [3][4][5]. In gliomas where both alleles of Ink4a, Arf, Pten were deleted at tumor initiation, and in regions of gliomas dominated by progeny in tumors with initially heterozygous deletion of these tumor suppressors, both progeny cells and recruited cells expressed PDGFRa, with ,5% tumors co-expressing EGFR ( Figure S6b-d). The primary mechanism for EGFR and IGFR expression in recruited regions is unclear; however, low-level amplification of these genes was seen in some cases. FISH analysis for mouse EGFR of eleven murine gliomas containing large regions of recruited cells with high expression of EGFR and/or IGFR showed .2 copies of EGFR in 5 of 11 cases, while 1 of 11 showed .2 copies of IGFR ( Figure S6e In hPDGFb-induced gliomas, pharmacologic blockade by the PDGFR/VEGFR inhibitor PTK787 that crosses the blood-brain barrier results in cell cycle arrest of the tumor cells; thus, these gliomas retain their dependence on PDGFR signaling [39]. Since gliomas with large areas of recruited cells did not express PDGFRa, we investigated whether recruited cells became independent of hPDGFb signaling during glioma progression. We treated ten Ntv-a Ink4a/Arf +/-Pten +/fl glioma-bearing mice with PTK787 for a week at the onset of symptoms; eight mice survived the full course of treatment. Of these, three mice had gliomas that expressed PDGFRa and eGFP in high-grade tumor areas, concomitant with reduced proliferation rates (p,0.0001) ( Figure 4f). Gliomas in the remaining five mice contained .85% recruited cells in high-grade tumor areas, showing proliferation rates statistically similar to the untreated tumors with large areas of recruitment ( Figure 4f; data not shown). Thus, regions dominated by recruited cells became independent of PDGFR signaling during glioma progression, and the response to PTK787 treatment was dependent on the cellular composition of a given glioma. Although PTK787 also targets VEGFR, the conclusion that recruited glioma regions are independent of PDGFR activity is not confounded by this fact, since our readout is the relative lack of cell cycle arrest in the recruited glioma regions.

Recruited cells initiate gliomas upon transplantation and aid glioma formation by progeny cells
Although glioma patients do not die from glioma cells transplanted into them but rather from the massive expansion/ infiltration by the glioma cells, tumor cells in rodent systems are most stringently defined by their ability to form secondary tumors upon transplantation into recipient mice [40][41][42][43][44]. We isolated either progeny or recruited cells based on their expression of viral eGFP by FACS from Ntv-a Ink4a/Arf -/-Pten fl/fl or Ink4a/Arf +/-Pten +/fl mice injected with RCAS-PSG and RCAS-Cre (Figure 5a Table S1). In addition to EGFR expression and unlike gliomas with the typical oligodendroglial morphology induced by transplantation of progeny cells, these tumors showed regionally high expression of nestin, GFAP, CD44, vimentin and VEGFR, while expression of olig2 was variable in some tumor areas and absent in most ( Figure S8) [8]. Thus, the recruited cells could form gliomas upon transplantation in some circumstances.
CD133, a pentaspan transmembrane protein and a hematopoietic stem cell marker, was reported to label the most proliferative and robustly tumorigenic cells in human gliomas, although this has been questioned in different contexts [40,42,44]. We addressed tumorigenic properties of CD133-positive cells in the context of virally induced hPDGFb gliomas. Purified doublesorted CD133-positive cells derived from Ntv-a Ink4a/Arf -/-Pten fl/fl or Ntv-a Ink4a/Arf +/-Pten +/fl mice injected with RCAS-PSG and RCAS-Cre were not derived from progeny cells and were all recruited, as confirmed by RT-PCR, qPCR and FISH on FACSsorted cells (Figure 6a Figure 6h), transplantation of purified CD133-positive recruited eGFP-negative cells from Ntv-a Ink4a/Arf -/-Pten fl/fl or Ntv-a Ink4a/ Arf +/-Pten +/fl mice did not result in glioma formation, suggesting that in hPDGFb-driven murine gliomas, tumorigenicity lies in the CD133-negative recruited cell subset (Figure 6g).
We further addressed whether the CD133-positive subset of recruited cells showed functional importance for initiation or progression of transplanted hPDGFb-induced murine gliomas by cell-mixing experiments. While purified CD133-positive recruited cells could not give rise to transplanted gliomas at 10,000 cells (0/ 24 cases), and progeny cells did not robustly form tumors with cell numbers less than 3,000 (3/20 cases), addition of 10,000 CD133positive recruited cells to 1,000 eGFP-positive progeny resulted in robust gliomagenesis (Figure 6i). Thus, CD133-positive recruited cells enhanced the ability of the progeny cells to initiate tumors upon transplantation, with small numbers of CD133-positive recruited cells sufficient to promote induction of transplanted tumors, the final mass of which was predominantly derived from the recruited cells (data not shown).

Recruited cell gene expression resembles tumor cells
An unbiased way to define a cell population lies in identification of its gene expression signature and subsequent comparison to gene expression signatures of known normal or cancer cells to define its position on the axis of tumorigenesis. To quantify similarities and differences in the expression profiles of recruited cells and tumor cells using microarray analysis, we used the bacTRAP technology that allows immunoprecipitation of polysomes from specific cell types in vivo ( Figure S3a-d) [31,32]. We compared polysomal expression profiles of hPDGFb-driven Ntv-a Ink4a/Arf +/-olig2 mouse glioma cells representing a histologically defined glioma population, recruited olig2 cells derived from Ink4a/Arf +/-olig2 RP-eGFP mice transplanted with non-fluorescent glioma cells, and normal adult cortical olig2 mouse progenitors, using Affymetrix 430 mouse 2.0 chips and Genespring GX10 software [31,32]. Translational profiles of recruited olig2 cells and glioma olig2 cells were similar to each other and clearly distinct from the normal olig2 progenitors, majority of the differences accounted for by the ''statistically significantly changed'' set of mRNAs (ANOVA, p,0.05), and sample clustering reduced upon removal of the ANOVA-tested mRNA set (Figure 7a-d; Figure  S10a-c,e). Unsupervised hierarchical mRNA clustering likewise indicated that recruited olig2 cells clustered more closely to glioma olig2 cells (Figure 7e). The caveat is that olig2 expression in the adult normal brain of bacTRAP olig2 RP-eGFP mice labels both progenitors and mature oligodendrocytes, while recruited and tumor olig2 cells are almost entirely progenitors; therefore, some of the differences seen in the comparative analysis of normal, tumor and recruited olig2 cells may in part result from the shift in relative abundance of different progenitor and mature olig2 cell populations.
Pair-wise comparisons indicated that out of 2,593 mRNAs differentially present on the polysomes, 2,310 mRNAs were differentially present in glioma vs. normal olig2 cells and 2,262 mRNAs in recruited vs. normal olig2 cells (one-way ANOVA, p,0.05, SNK post-hoc, FDR correction); of note, 86% (1,979 mRNAs) of these differentially represented mRNAs were the same (Figure 7d). Ingenuity analysis of this polysomal set revealed mRNAs implicated in genetic disorders, neurological disease and cancer, regulating cell death, morphology and the cell cycle (data not shown). Of 500 mRNAs most different between glioma and normal olig2 cells, 490 mRNAs were changed in the same direction but smaller magnitude in recruited olig2 cells, 440 of which were at least two-fold different, and 183 were five-fold different or more ( Figure S10d). From the above data, the expression profile of recruited cells closely resembled that of tumor cells, placing the two closely together in the continuum of change towards the ''malignant state'', and further confirming that recruited olig2 expressing cells are tumor. While only ,300 mRNAs were the same between normal and either glioma or recruited olig2 cells, the number of mRNAs that were the same between glioma and recruited olig2 cells was at least three-fold more, amounting to 1,047 (Figure 7d). Many of these mRNAs were classified into gene categories listed above by Ingenuity (data not shown).
Although tumor and recruited olig2 cells were predominantly similar, the number of differentially recruited mRNAs with over 2fold change between tumor and recruited olig2 populations amounted to 665, and those with over 5-fold change amounted to 132. mRNAs specifically present on polysomes of olig2 recruited cells coded for proteins in Hedgehog, Wnt, Ras, Rho/ Rac and PLC pathways. While PDGFRa and TGFb mRNAs were lower in recruited olig2 cells than tumor cells, mRNAs coding for EGFR, TGFa, TNF, LIF and FGFR1/2 were higher in recruited cells; prominin1 (CD133) mRNA was 5.3 fold higher in the recruited than tumor olig2 cells. mRNA populations corresponding to genes defining oligodendrocytic character were reduced in olig2 recruited cells, while mRNAs associated with astroglial and endothelial characteristics were upregulated in the recruited population. In addition, 41 mRNAs were increased on recruited cell polysomes by 10-fold or more, as compared to the tumor olig2 cells and differentially from normal olig2 cells (Figure 7f). Many of these mRNAs were gradually reduced in transition from normal to recruited and to tumor states, correlating with loss of oligodendroglial character (Aspa, Mobp, Tfr, Mog), evasion of cell death (Mal, Ndrg1, Gjb6, Gpr37), maintenance of undifferentiated phenotype (Mal, Ndrg1, Efhd1, Kndc1, Ermn), cell-cell and cell-matrix interactions (Thbs3, Tppp3, Itih3), and cell signaling (Ppp1r14a, Gja1). Cnn2 mRNA was specifically reduced on the polysomes of recruited cells as compared to both normal and tumor olig2 cells, while Nupr1, Aldh1l1, Lcn2, Papss2, and Igf2bp2 mRNAs were specifically increased in this population (Figure 7f).

Discussion
RCAS/tv-a system allows to closely model the biology and histopathology of human oligodendrogliomas; although we have not performed direct comparisons between our mouse model and human glioma classifications, in many ways, hPDGFb-driven murine gliomas may mimic the proneural subclass of human GBMs [6][7][8]. Our data suggests that olig2-expressing cells recruited in PDGF-induced murine gliomas can have similar morphologic, proliferative and functional characteristics as olig2expressing tumor cells derived from the cell-of-origin, with the caveat that a part of the similarity between the recruited and tumor olig2 cells may stem from their proliferative behavior. Extensive similarities between polysome-associated transcriptome of recruited versus glioma olig2 cells call for further evaluation of the precise definitions and criteria applied to terms ''tumor'' and ''normal'', and indicate similar gene expression character of these cells. Furthermore, tumor suppressor loss and acquisition of mutations typically found in human gliomas allow these murine recruited cells to occupy large areas of the tumor and at times become the predominant cellular component of the glioma mass, completely independent of the cell-of-origin. Thus, the clonal expansion that overtakes glioma bulk during tumor progression need not be derived from glioma cell-of-origin. Transformation is the process of self-autonomous acquisition of the sufficient oncogenic alterations that change a normal cell into the tumor cell. Unlike epithelial cancers, the ''normal'' or stromal component of a glioma is composed of cells of the same lineage as the tumor cells. The inherent implication of this observation is that extracellular factors that promote glioma tumor cell growth may similarly affect glioma stroma. In our glioma model, initially normal recruited cells not derived from the glioma-initiating cellof-origin can be stimulated by hyperproduction of a growth factor receptor ligand, induced to proliferate, and driven to acquire various genetic aberrations and aberrant expression profiles as gliomas progress, presumably because of the tumor microenvironment created by the progeny of the cell-of-origin. The process of acquisition and selection for such genetic and/or epigenetic alterations due to the pressures of the glioma microenvironment may be better termed ''corruption''. Although it is not known how frequently recruitment and subsequent corruption occurs in human PDGF-driven gliomas, gliomas in general or in other tumor types, our data indicates that this process may not be limited to hPDGFb-induced murine gliomas. As such, human gliomas that appear to be comprised of two or more genetically unrelated clones, or recurrent patient gliomas that lack the genetic alterations present in the originally resected tumors subjected to therapeutic stress, may be examples of these processes. Although similar processes have not been described in vivo, an in vitro line of evidence from the colon cancer cell lines suggested that one of the mechanisms of how cancer cells bypass senescence may be related to their potential for continuous ''clonal diversification'', even within the what is perceived a clonal population of cancer cells, not affected by native stromal environment [45].
The process of corruption would thus occur if the tumor microenvironment initially drive proliferation and induce the stem-like character in both progeny of the cell-of-origin and the recruited cells. In hPDGFb-driven gliomas there is evidence for both -tumor cells secrete proliferative growth factors, inflammatory cells secrete cytokines, and endothelial and inflammatory cells secrete nitric oxide that activates the Notch signaling pathway, promoting the stem-like character [46]. Such environment affects cells within the tumor regardless of whether they are derived from the glioma-initiating cell-of-origin or not. This notion is further supported by the analysis of expression array data using the bacTRAP olig2 RP-eGFP reporter system and immunostaining, which suggest that recruited olig2 cells can acquire expression of GFAP and vimentin and increase the expression of nestin as compared to the tumor olig2 cells, triple co-expression of which is considered to be typical for the bona fide neural stem cells.

Ethics Statement
All animal experiments were conducted using protocols approved by the Institutional Animal Care and Use Committees of Memorial Sloan-Kettering Cancer Center. The approved protocols are: 00-11-189(MSKCC, last approved 3/15/2010).
Tissue collection and processing. Experiments involving mice were performed in accordance with the regulations set by the MSKCC IACUC committee. Upon the appearance of brain tumor symptoms (e.g. poor grooming, lethargy, macrocephaly, hydrocephaly, hemiparesis, weight loss), mice were anesthetized by an intraperitoneal injection of 1 mg/kg Nembutal solution or ketamine (150 mg/kg)/xylazine (15 mg/kg) cocktail and underwent transcardiac perfusion with 10 ml ice-cold heparin normal saline, followed by 10 ml of ice-cold 4% paraformaldehyde. Brain tissue was extracted and post-fixed for 30 minutes in ice-cold 4% paraformaldehyde, transferred into 30% sucrose at 4uC for cryoprotection, embedded using the OCT compound, frozen on dry ice and stored at 280uC until sectioned. A small subset of animals were sacrificed using CO 2 , their brain tissue extracted and fixed for 3 days in 4% paraformaldehyde at 4uC, and subsequently processed using standard paraffin embedding and processing techniques.
bacTRAP technology. Normal olig2 progenitors were collected from three replicates of pooled cortices of three olig2 RP-eGFP bacTRAP reporter mice. Mouse gliomas induced by injection of a non-fluorescent RCAS-hPDGFb [14] or by transplantation of murine glioma cells derived from Ntv-a gliomas induced by the non-fluorescent RCAS-hPDGFb were collected from Ntv-a olig2 RP-eGFP bacTRAP reporter mice and immunoprecipitated; each tumor was processed as a separate sample. Briefly, mouse tissues were collected into ice-cold cyclohexamide-containing buffer, homogenized, cells were lysed in NP-40 and DHPC-containing buffer, centrifuged at 20,000 g for 15 min, supernatant incubated with anti-eGFP-conjugated protein G beads 30 min at 4C, washed and RNA collected using Trizol reagent as per manufacturer's instructions [31,32]. RNA was purified, concentrated using Qiagen RNeasy kit, its quality confirmed by Agilent Bioanalyzer, 15 ng per sample amplified using Affymetrix Two-Cycle Amplification kit and hybridized to Affymetrix 430 mouse 2.0 chips as described in [31,32].
Microarray analysis. Data was analyzed using Genespring GX10 software, as described in the text and in [31,32]. Briefly, samples were normalized using GCRMA, filtered to remove probe sets with low intensity, and analyzed using ANOVA (p,0.05) with SNK post-hoc and Benjamini and Hochberg FDR multiple testing correction. Probe quality, Pearson's correlation, hierarchical clustering, heatmaps, gene profile plots were generated using Genespring GX10 software. Pathway analysis was performed using Ingenuity as per manufacturer's instructions. The mRNA microarray data generated by this study are available through the NCBI Gene Expression Omnibus (GEO) (accession number GSE30016, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE30016).
Cell Sorting and transplantation of murine glioma cells. Glioma-bearing brains were extracted from anesthetized mice, gliomas excised from surrounding brain parenchyma, dissociated using a mix of collagenases as previously described, and sorted using MoFlo FACS sorting. Various numbers of cells (1,000-500,000) resuspended in sterile 1X PBS were injected into brain parenchyma of neonatal immunonaive mice of various genetic backgrounds, or stereotactically transplanted into anesthetized adult SCID mice.
PCR on tumor sections. Mouse glioma sections were scraped from slides and lyzed with 0.05% SDS buffer. Genomic DNA was extracted using phenol-chloroform according to standard techniques, and precipitated at 220uC overnight. Prior to PCR for hPDGFb, DNA was heated at 56uC for 2 hours or for 5 min at 100uC, concentration measured, and equal amounts of DNA added to PCR reactions. PCR reactions for human PDGFB, Pten, and the Ink4a/Arf locus were performed using previously published primers. Control GAPDH reactions were performed in the same PCR mix. Data was obtained using GelDoc Software.
RT-PCR on FACS-sorted cells. Total RNA from sorted cells was extracted using the Qiagen RNeasy kit and resuspended in 30-50 ul of DEPC-treated water. About 5 ug of total RNA was reverse-transcribed using the SuperScriptIII Reverse strand synthesis kit from Invitrogen. Quantitative RT-PCR was performed using primers for human PDGFb and mouse actin as a control.
Immunohistochemistry. Cryogenically processed murine glioma tissue was sectioned at 7 microns, washed with PBS, blocked with 10% serum in PBS using 0-0.3% TX-100 for polyclonal rabbit/goat or MOM kit for monoclonal mouse and rat antibodies, and incubated with primary antibodies for 1 hour at room temperature or overnight at 4C. The primary antibodies included PDGFRa and PTEN from Cell Signaling; olig2, Ki67, hNA and EGFR from Abcam; NeuN from Chemicon; HA from Santa Cruz; and eGFP from Molecular Probes. Secondary fluorochrome-conjugated antibodies (Alexa555, Alexa488, and Alexa647) were applied for 1 hr at room temperature, slides washed several times, counterstained with DAPI (1:3000 of a 1 mg/ml 1X PBS solution), mounted in 70% glycerol, coverslipped and photographed with inverted fluorescent or confocal Leica microscopes.
Cell Sorting and transplantation of murine glioma cells. Glioma-bearing brains were extracted from anesthetized mice, gliomas excised from surrounding brain parenchyma, dissociated using a mix of collagenases as previously described, and sorted using DAKO-Cytomation MoFlo FACS sorters for eGFP expression. Samples were gated to exclude debree, dead cells and doublets. Various numbers of cells (1,000-500,000) resuspended in sterile 1X PBS were injected into the brain parenchyma of neonatal immunonaive mice of various Ntv-a genetic backgrounds, or stereotactically transplanted into anesthetized adult SCID mice, using Hamilton syringe. CD133 (eBioscience) staining of acutely dissociated murine glioma cells was done as per manufacturer's instructions; CD133 antibody was used at a 10-fold lower concentration based on titration. Single-or double-sorted CD133-positive cells were transplanted as described above. Cell-mixing experiments were performed using doublesorted eGFP-positive progeny and double-sorted CD133-positive recruited cells, which were mixed just prior to transplantation, and injected using Hamilton syringe.
Fluorescent in-situ hybridization. RCAS DNA was fragmented using nick translation kit, labeled with biotin, and digested to 500 bp-2 kb probe size. Resulting probes were denatured at 80uC and applied to tumor sections blocked with salmon sperm DNA for an overnight incubation at 37uC. Slides were washed with 2X SSC-50% formamide solution at 45uC, signal amplified according to standard techniques using TSA kit, and revealed with Alexa488 and Alexa555. Slides were counterstained with DAPI (1:3000 of 1 mg/ml PBS solution), mounted, coverslipped and photographed using inverted fluorescent or confocal Leica microscopes.
Statistical analysis. Kaplan-Meyer curves for mouse glioma latency were made using Prism4 software and analyzed with standard log rank test, or using R. The p value was determined to be less than 0.001 for survival curves of Ntv-a wild-type vs. Ntv-a Arf -/mice injected with RCAS-PSG vs. Ntv-a Ink4a/Arf -/-Pten fl/fl mice injected with RCAS-PSG and RCAS-Cre. Survival curves for Ntv-a wild-type vs. Ntv-a Ink4a -/-, or for Ntv-a Ink4a/Arf -/vs. Ntv-a Arf -/-, were not statistically different from each other. Statistical analysis of other data, including percentages of eGFPpositive cells in different murine Ntv-a backgrounds and percentages of Ki67-positive cells in PTK787-treated murine gliomas, was performed using unpaired two-tailed t-tests and Ftests in Prism4.  Figure S6 hPDGFb-induced Ntv-a gliomas show regional expression of growth factor receptors important in human glioma biology, expression of which in recruited cells may be associated with low level amplifications in some cases. (a-d) Ntv-a gliomas induced in Ntv-a Ink4a/ Arf +/-Pten +/fl mice injected with RCAS-PSG and RCAS-Cre stained with IGFR (red), EGFR (pink) (a,d) or PDGFRa (red) (b,c). Glioma regions predominantly derived from progeny cells (a,b) express PDGFRa; expression of EGFR and IGFR is limited to perivascular areas. (c,d) Adjacent sections stained with PDGFRa and EGFR (red) show regional expression of growth factor receptors; eGFP omitted for easier view. (e) FISH for IGFR (green) and EGFR (red) on Ntv-a GBMs with large areas of recruitment and high EGFR expression. Panels show a neuron, EGFR-expressing non-amplified GBM with extensive recruitment, and EGFR-expressing GBM with EGFR amplification in the recruited cells. (f) Summary of mEGFR FISH performed on EGFR-expressing GBMs with large areas of recruitment. Gliomas containing more than 15% of cells with $3 mEGFR signals are marked with an asterisk (amplified). (g,h) real-time qPCR on DNA extracted from sections of mouse Ntv-a GBMs containing large areas of recruitment (blue squares) highly expressing EGFR. Primers for mEGFR kinase domain (g) or intron region of mEGFR (h) were designed by Roche; graphs show fold increase over normal EGFR copy number (Ø); gliomas with large areas composed of progeny cells (green squares) were used to establish range of variability of primer noise. Gliomas marked with asterisks were considered amplified, and gliomas marked with arrows appeared amplified by both primer pairs, potentially representing tumors with larger chromosomal amplifications.