PDGF-B induces a homogeneous class of oligodendrogliomas from embryonic neural progenitors


  • Irene Appolloni,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
    • Irene Appolloni and Filippo Calzolari contributed equally to this work.

  • Filippo Calzolari,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
    • Irene Appolloni and Filippo Calzolari contributed equally to this work.

  • Evelina Tutucci,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
  • Sara Caviglia,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
  • Marta Terrile,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
  • Giorgio Corte,

    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
  • Paolo Malatesta

    Corresponding author
    1. Department of Genic Transfer, National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    2. Department of Oncology and Genetics (DOBIG), University of Genoa, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    • National Institute for Cancer Research (IST), IRCCS, Largo Rosanna Benzi 10, 16132 Genoa, Italy
    Search for more papers by this author
    • Fax: +390105737405


We describe the generation of mouse gliomas following the overexpression of PDGF-B in embryonic neural progenitors. Our histopathological, immunohistochemical and genome-wide expression analyses revealed a surprising uniformity among PDGF-B induced tumors, despite they were generated by transducing a highly heterogeneous population of progenitor cells known for their ability to produce all the cell types of the central nervous system. Comparison of our microarray data with published gene expression data sets for many different murine neural cell types revealed a closest correlation between our tumor cells and oligodendrocyte progenitor cells, confirming definitively that PDGF-B-induced gliomas are pure oligodendrogliomas. Importantly, we show that this uniformity is likely due to the ability of PDGF-B overexpression to respecify competent embryonic neural precursors toward the oligodendroglial lineage, providing evidence that the transforming activity of PDGF-B is influenced by the developmental potential of the targeted cells. Interestingly, we found that PDGF-B-induced tumors harbor different proliferating cell populations. However only PDGF-B-overexpressing cells are tumorigenic, indicating that paracrine signaling from the tumor is unable to transform bystander cells. © 2008 Wiley-Liss, Inc.

Gliomas constitute a very heterogeneous group of aggressive brain tumors. They are classified mainly according to the expression of central nervous system (CNS) cell type-specific markers into astrocytomas, oligodendrogliomas, mixed oligoastrocytomas and glioblastomas. The latter is the most malignant form and is usually considered the highest grade of astrocytoma (Grade IV),1, 2 even if it has been recently proposed that some glioblastomas may instead represent high grade oligodendrogliomas.3 The diversity of gliomas is mirrored by the number of different signaling pathways shown to be involved in the generation of these tumors1, 4 and, likely, by their different population of origin.5

Alteration of PDGF-B signaling is commonly observed in human gliomas of different histophatological grades.6–8 Previous studies in mouse perinatal and adult neural progenitor/stem cells demonstrated the ability of PDGF-B to induce gliomas. Although nestin-positive neonatal neural progenitors overexpressing PDGF-B generate pure oligodendrogliomas, targeting of GFAP-expressing glial progenitors at the same developmental stage results in the generation of oligodendrogliomas as well as mixed oligoastrocytomas, pointing to a cell type-specific response to PDGF-B transforming activity.9–16 However, the extent to which the developmental plasticity of the targeted cells may influence the response to the transforming stimulus mediated by PDGF-B has not been clearly determined.

We therefore addressed this issue, by forcing PDGF-B overexpression in a population of immature precursors which is known to generate all lineages of the CNS.17–24 Much to our surprise, the spectrum of tumors generated in this system was narrower than that reported to arise from more mature and fate-restricted cell types,10, 16 and invariably consisted of tumor sharing oligodendroglial characteristics. We therefore investigated the basis of this phenomenon and found that PDGF-B overexpression in immature neural progenitor cells results in their respecification toward the oligodendroglial lineage. These data suggest that the homogeneity of the tumors induced by PDGF-B at embryonic stages is due to its fate specification activity.

Interestingly, despite its strong tumorigenic activity, PDGF-B was unable to paracrinally transform cells in the vicinity of the tumor.

Material and methods

Cell cultures

Cortical neural progenitor cultures were prepared from embryonic telencephalic explants as previously described.25

For immunohystochemistry, cells were plated onto 13-mm diameter coverslips coated with poly-D-lysine and fixed in 4% PFA 6 days after.

Fluorescent-activated cell sorting

Acutely dissociated and cultured tumor cells were sorted using a FACSAria (BD Biosciences, Inc, San Jose, CA). Purified GFP-positive and GFP-negative cell populations were visually inspected after sorting under a fluorescent microscope to quantify the contaminating fraction, which was usually <1%.

Retroviral vectors and transduction procedures

The cDNA of mouse PDGF-B, derived from the RCAS-pBIG plasmid (kindly provided by Dr. E. Holland, Memorial Sloan-Kettering Cancer Center, NY), was inserted into the SalI site of the pCEG retroviral vector (kindly provided by Gordon Fishell, The Skirball Institute of Biomolecular Medicine, NY), generating a PDGF-B/IRES/GFP construct. Replication-defective retroviral supernatants were prepared by transiently transfecting plasmids into Phoenix packaging cells as described in Ref.26. Retroviral vectors were used on cultured E14 cells as previously described in Ref.27 and in vivo as described in Ref.25. All the PDGF-B-overexpression experiments, in vitro and in vivo, were carried on using the IRES-containing retroviral vector, in which the coding sequence for GFP was downstream the IRES.

Animal procedures

Mice were handled in agreement with guidelines conforming to the Italian current regulations regarding the protection of animals used for scientific purposes (D.lvo 27/01/1992, n. 116). Procedures were specifically approved by the Ethical Committee for animal experimentation (CSEA) of the National Institute of Cancer Research and by the Italian Ministry of Health. All experiments have been performed on the C57/Bl6 mouse strain.

In utero intraventricular injection where performed as described elsewhere.17 Animals injected at embryonic stages were let develop to term and after birth they were monitored for the appearance of symptoms indicating the presence of brain tumors. At first signs of symptoms, the animals were sacrificed and perfused with 4% PFA. The brains were photographed on a transilluminator and/or cryoprotected in 20% sucrose and sectioned with a Leica CM1100 cryostat (Leica, Wetzlar, Germany).

Tumor cells were injected in deeply anesthetized adult animals mounted on a stereotaxic apparatus. Up to 5 μl of cell suspension for each mouse, containing from 50 to 50,000 cells, were injected using a Hamilton Syringe (Bregma coordinates: AP 1.0 mm, L 1.5 mm left and 2.5 mm below the skull surface). Reabsorbable suture was used before awaken the animals. Animals were then monitored daily for the onset of neurological symptoms.


Immumostainings on brain sections or cultured cells were performed using the following antibodies: mouse monoclonal antibodies against nestin (1:250, BD Pharmingen, San Diego, CA); GFAP (1:200, Sigma, St. Louis, MO), APC (1:100, CC1, Calbiochem, San Diego, CA), rabbit polyclonal antibodies against Olig2 (1:200, Sigma), Ng2 (1:300, Chemicon, Millipore, Billerica, MA), chicken polyclonal antisera against GFP (1:500, Abcam, Cambridge, UK); rat monoclonal antibodies against PDGFRα (1:100 BD Pharmingen), Ki67 (1:25, tec3 clone, Dako, Glostrup, Denmark). Binding of primary antibodies was revealed with appropriate secondary FITC- and TRITC-conjugated antibodies (1:50, Immucor, Norcross, GA), or biotinylated secondary antibodies (1:50, Dako) which were revealed with streptavidin-conjugated Alexa 488 (1:500, Molecular Probes, Carlsbad, CA). Nuclei were stained through 5 min incubation in DAPI solution (1 μg/ml, Sigma).

Immunostaining were examined with a NIKON Eclipse 800i, Tokyo, Japan. Images were acquired with NIKON Digitalsight DS-M5c camera and analyzed with ImageJ (Rasband, W.S., ImageJ, US National Institutes of Health, Bethesda, MD, http://rsb.info.nih.gov/ij/, 1997–2007). In our analyses, we classified a GFP-negative cell as “in the close vicinity” of a GFP-positive 1 when their relative distance is smaller than 5 cell diameters. In the same analyses we used as negative control the GFP-negative cells residing in noninfiltrated regions, when possible, in controlateral areas. At least 3 independent tumors were analyzed for each marker and a total of more than 500 GFP-positive (or GFP-negative, for the control regions) cells from at least 3 nonconsecutive sections were counted for each brain.

Data analysis on brain sections and cultured cells

The mean and standard errors were calculated from different experiments: “number of experiments” was adopted to denote the total number of independent experiments; “number of cells” indicate the total number of cells analyzed; “n” denotes the number of animal used for in vivo experiments. The threshold for statistical significance, which was determined with a 2-tailed Student's t-test, was considered as p < 0.05.

Real-time PCR

Genomic DNA from sorted GFP-positive and GFP-negative cells was obtained by lysating the cells in 100 mM Tris/Cl pH 8.5, 5 mM EDTA, 0.2% SDS, 200 mM NaCl, 100 μg/ml proteinase K at 56°C and precipitated with isopropanol. In other cases, RNAs were extracted from sorted cells with TRIzol reagent (Invitrogen, Carlsbad, CA), according to the manufacturer guidelines. cDNAs were then obtained from 500 ng of RNA using the iScript (Bio-Rad Laboratories, Hercules, CA) retrotranscription kit. Real-time PCR was performed on 10 ng of genomic sample or 1/100 of the retrotranscription reaction using iQ SyBr green supermix (Bio-Rad Laboratories). The presence of the proviral insert was determined using a pair of primers specific for GFP amplification, whereas the amplification level of the Fyn gene (MGI:95602) was used as reference for data normalization since it has no pseudogenes. The sequences of the primers are available on request.

Microarray analysis

Sorted cultures of GFP-positive cells from 3 independent tumors were harvested with Trizol (Invitrogen) and used for further processing. RNA extraction and microarray hybridization procedure were performed at the “Consorzio Genopolis” (University of Milano-Bicocca, Department of Biotechnology and Bioscience) using the Affymetrix GeneChip Mouse Genome 430A 2.0 Array (Santa Clara, CA). Files containing the raw data from the study published by Cahoy et al.,28 using the Affymetrix GeneChip Mouse Genome 430 2.0 Array, were retrieved at the NCBI GEO DataSets online repository (GSE9566 record). Data were analyzed using the R software and BioConductor version 2.2.29 Only the probesets common to both platforms were used for the comparison. Expression values were extracted from both our and public raw data files using the RMA method built in the affy 1.18.2 package without prior normalization and collected in a single data set. Probe level quantile normalization method provided by the affyPLM 1.16.0 package was then used to normalize the dataset. The probesets where then ranked based on the variance of their expression level across the different cell types and the uppermost 5% (1135 probesets with highest variance) were selected and used to produce a hierarchical clustering of the samples. Hierarchical clustering was obtained using hclust method with complete linkage and euclidean distance in R.

Array validation was performed by real time RT-PCR on a pool of 16 genes randomly selected among those showing a detectable expression level in our arrays (gene names and primers are available on request). Log2 scaled expression values from array and real time PCR were compared as follows. A matrix of the relative differences between the expression level of each gene of the pool and the level of every other gene of the pool was calculated for both real time PCR and microarray data. Differences smaller than 1 were filtered out and the remaining differences were assayed for concordance of sign, and the percentage of concordant comparisons was calculated obtaining a percentage of 83%.


The overexpression of PDGF-B in embryonic neural precursors efficiently induces pure oligodendrogliomas

We injected replication-deficient retroviruses expressing a PDGF-B/IRES/GFP cassette into the lateral telencephalic ventricles of mouse embryos at mid neurogenesis (embryonic day 14; E14). At this stage, precursor cells (radial glia) that will give rise to all lineages of the future telencephalon are actively dividing and can be representatively infected by retroviruses as shown by a number of previous observations from multiple laboratories, including our own17–24 (see Supp. Info. Fig. 1A). These observations firmly established that MoMLV-derived retroviral vectors can be successfully used to obtain broad and unbiased labeling of precursors of all the possible cell lineages in the rodent CNS, including adult and neonatal stem cells (see Supp. Info. Fig. 1B–S).25, 30, 31 Notably, the latter are the cells targeted by other authors to investigate the in vivo transforming potential of PDGF-B.10, 12, 32 Our analysis of 4 independent brains, 2 days after the injection with our retroviral supernatant (typically 106 colony forming units), showed that an average of 3500 ± 1000 infected precursor cells were present in each injected brain (data not shown).

Figure 1.

PDGF-B induces brain tumors from embryonic neural progenitors. (a) Survival curve of animal injected with PDGF-B retroviral vectors inducing gliomas (green line) and of the control pool (15 animals injected with a control vector, red line). (b) Merged panel of immunofluorescence microphotographs of a coronal section from a PDGF-B-induced glioma stained with anti-GFP antibody in green. (b′, b″) Magnified insets from (b) showing cross sections of aberrant blood vessels. The contour in (b) outlines the brain section, whereas the dashed contours define the boundary of the main tumor mass. (c, d) Coronal sections of a primary high-grade tumor stained with anti-GFP antibody in green (c)or with hematoxylin and eosin (d). The dashed line in (c) and (d) define the contour of a pseudopalisade. Fluorescence and bright field microphotographs were taken from the same sections after each staining step. DAPI is in blue in (b′) and (b″). Scale bars: 1 mm (b), 100 μm (b′–d).

After birth, the injected mice were monitored and sacrificed as soon as they showed symptoms of neurological distress. All the injected animals (n = 57; Fig. 1a) developed neurological symptoms within 200 days after birth. In contrast no mice ever developed gliomas when injected with a control virus (n = 15). Although a percentage of injected animals died within 40 days as a consequence of a severe hydrocephalic condition (data not shown), the analysis of the brains explanted from all the animals that developed symptoms after 70 days, invariably showed large masses of GFP positive cells, revealing a neoplastic proliferation of cells that had integrated the PDGF-B-expressing retrovirus (Fig. 1b). Smaller masses of infiltrating GFP-positive cells were also present in all the animals developing hydrocephalus (data not shown). In this work, however, we focused our analysis on later-arising gliomas, referring the reader to an additional article33 for further discussion on the early-onset tumors. Nearly all the tumors causing symptoms after 70 days harbored widely infiltrating GFP-positive cells (20 of 22). Tumors showed signs of necrosis, hemorrhage and blood vessel remodeling (Figs. 1b1b″, data not shown) and cells were often arranged into pseudopalisades around necrotic foci (Figs. 1c and 1d). These tumors, therefore, closely resembled the most malignant form of glioma, the glioblastoma. PDGF-B-induced tumors were all very similar, and transduced cells did not express either neuronal or astroglial markers, lacking both NeuN and GFAP (Figs. 2a and 2b). Transduced cells expressed the proliferative marker Ki67 (Fig. 2c) and progenitor/stem cells markers such as the intermediate filament nestin (Fig. 2d; Table I), highlighting their immature condition.34 They widely and consistently also expressed the oligodendroglial markers NG2,35 Olig2,36 PDGFRα28 and the form of the adenomatous polyposis coli (APC) gene product recognized by the CC1 antibody37 (Figs. 2e2h; Table I).

Figure 2.

PDGF-B-induced tumors resemble oligodendrogliomas. (ah) Immunofluorescence stainings of coronal tumor sections with anti-GFP antibody in green, DAPI for nuclear staining in blue and the antibody for the indicated antigen in red. Arrows point to examples of doubly positive cells. Scale bar: 50 μm.

Table I. Tumor Cells Express Oligodendrocyte Progenitor Markers
Marker% Marker positive cells
  1. The percentage of positivity to the indicated molecular markers ± the standard error is indicated. In brackets are indicated the number of independent tumors analyzed and the total number of cells counted.

Nestin79 ± 5 (3, 2262)
Ng286 ± 4 (3, 1913)
Olig283 ± 6 (3, 2327)
APC87 ± 11 (2, 636)

This analysis showed that the tumors generated by the embryonic transduction of PDGF-B are comprised of cells with unambiguous traits of the oligodendroglial lineage, suggesting they all represent pure oligodendrogliomas.

Microarray analysis

To further characterize the tumors induced by PDGF-B, we performed a microarray-based gene expression analysis on 3 independent cultures established from PDGF-B-induced gliomas. Microarray gene expression data were validated by real-time PCR as described in Material and methods. The entire data set and relative row data are available on NCBI GEO DataSets online repository with the accession number GSE12836. Given the uniformity observed among all the tumors analyzed by immunohistochemistry, we first assessed the variability among the gene expression profiles of the 3 independent PDGF-B-induced gliomas, finding that, despite they were originated by transducing a heterogeneous population of progenitor cells, their gene expression profiles were strikingly similar (Fig. 3a), indicating that they represent a very coherent group of tumors. Pairwise comparisons displayed a remarkably high correlation coefficient (on average R2 = 0.95 ± 0.01). This suggests that either PDGF-B overexpression acted instructively to drive the acquisition of an oligodendroglial identity in neural progenitors or selectively transformed a very specific precursor subpopulation. As a further step to analyze these gliomas, their gene expression profiles were compared with those generated by Cahoy et al.28 using a compatible microarray platform. In this study, the authors analyzed a broad and very diverse set of acutely purified mouse neural cell types, including different maturation steps of all the major neural cell types of the CNS. By comparing our PDGF-B-induced gliomas to this set of neural cell types, we aimed to perform a biologically meaningful and comprehensive analysis to address more precisely the issue of the identity of these tumors. Remarkably, in agreement with our immunohistochemical analysis, the gene expression profiles of PDGF-B-induced gliomas displayed the closest similarity to that of oligodendrocyte progenitor cells (OPCs). In hierarchical clustering analyses, PDGF-B-induced tumors grouped together with OPCs, in a cluster closely associated with the immature and myelinating oligodendrocytes, giving rise to an oligodendroglial lineage supercluster (Fig. 3b). Outside this major group, neurons at different maturation stages were associated into a well-nested neuronal assemblage and, outside this oligodendrocyte-neuron clustering, astrocytes formed the third major branch. Our findings therefore strongly support the view that the tumors induced by PDGF-B from neural progenitor cells are composed of immature cells with a clear oligodendroglial gene expression signature. Notably, in support of the validity of our approach, the topology of the tree (to the obvious exclusion of the presence of our samples) was identical to that obtained by Cahoy et al., showing a good degree of comparability between the 2 data sets.

Figure 3.

Gene expression analysis identifies PDGF-B-induced gliomas as pure oligodendrogliomas. (a) Scatter plots for pairwise comparisons between the normalized probe expression values for 3 independent PDGF-B-induced gliomas (T41F, T118D and T140C). Squared R correlation coefficients for each comparison are indicated above the plots. (b) Dendrogram representation of the results of the hierarchical clustering analysis between the gene expression profiles of our 3 PDGF-B-induced gliomas and of 27 primary neural cell populations from a previous article.28 Nontumor samples are color-coded as in the article mentioned before, for purely comparative purposes. Tumor samples branches and names are in bold.

Altogether, our data indicate that PDGF-B-induced tumors represent pure oligodendrogliomas, and support the view that some high-grade human gliomas with oligodendroglial features previously classified as glioblastoma may indeed represent “Grade IV oligodendrogliomas.”3

PDGF-B overexpression induces embryonic progenitor cells to acquire an oligodendroglial identity

We have observed that purely oligodendroglial tumors arise from the overexpression of PDGF-B into the heterogeneous precursor cell population that gives origin to all the lineages of the CNS. This observation is intriguing, since other authors formerly showed that the retroviral transduction of PDGF-B in the neonatal mouse brain can cause a more diverse set of gliomas, including GFAP-expressing oligoastrocytomas, depending on the identity of the precursor cells targeted by the retrovirus.10 We therefore asked whether these apparently conflicting results could be due to the ability of PDGF-B to shift the cell fate of immature embryonic progenitors toward the oligodendroglial lineage.

To answer this question, we performed an in vitro retroviral cell lineage-tracing analysis on E14 neural progenitor cells, by transducing them with an extremely low titer PDGF-B-expressing virus and analyzing them after 7 days. In this well-established experimental paradigm,25, 27, 38–40 it is possible to follow the entire progeny of each transduced precursor, since it remains associated in a well-defined cluster (clone). Since the clones are defined on the basis of their expression of GFP (and PDGF-B), only cells which actually produce PDGF-B are monitored. This experimental design is required to focus the analysis on the progeny of single precursors whose composition, in turn, reflects the influence of PDGF-B expression on commitment/potentiality of a progenitor cell. The pattern of expression of the transmembrane proteoglycan Ng2 would allow to distinguish between the 2 possible alternative effects of PDGF-B: an induction of a selective proliferative expansion of already committed OPCs or a bias on the fate of multipotent progenitors, forcing them to acquire an OPC identity. In this experimental system, the “selective proliferative expansion” scenario predicts an increase in the size (i.e., number of cells) of the clones containing Ng2-positive cells, without any change in the abundance of these clones. On the contrary, cell fate respecification would result in an increase in the frequency of the clones containing at least 1 Ng2-positive cell, due to a de novo appearance of Ng2-positive cells in clones that would otherwise contain none if unstimulated (Fig. 4a).

Figure 4.

PDGF-B overexpression drives neural progenitors to develop as OPCs. (a) Schematic representation of the hypothesized alternative outcomes of the clonal overexpression of PDGF-B on cultured neural progenitor cells (lower panel), compared with the control (upper panel). (b, c) Immunofluorescence microphotographs of E-14 cortical neural progenitor cell cultures stained with anti-GFP (green) and anti-Ng2 (red) antibodies. Cells were transduced with either a control (b) or PDGF-B-expressing virus (c). (df) Histograms show the effects of PDGF-B overexpression on cortical cultures. Percentage of Ng2-positive cells among the transduced cells (d), frequency of clones containing at least 1 Ng2-positive cell (Ng2-clones; e) and percentage of Ng2-positive cells per Ng2-clone (f). **p < 0.01 in a 2-tailed Student's t-test. Scale bar: 25 μm.

Our analysis revealed that the cultures transduced with the PDGF-B-expressing virus contained a significantly higher amount of Ng2-immunoreactive cells (15 ± 0.8%; number of experiments = 2; number of cells = 942) compared with the control (2.4 ± 0.4% number of experiments = 2; number of cells = 1641; t-test < 0.01; Figs. 4b4d). Most importantly, this increase was associated with a doubling of the percentage of clones containing at least 1 Ng2-positive cell (“Ng2-clones”) from 3 ± 1% to 7 ± 1% (t-test, p < 0.01; Fig. 4e), rather than to a selective expansion of the size of the Ng2-clones. The average size of the Ng2-clones was indeed not significantly different between PDGF-B-overexpressing and control-transduced cultures (12 ± 2.3 and 7.8 ± 1.3 cells/clone, respectively; p = 0.27), suggesting that, in the short term, PDGF-B did not selectively affect the proliferative activity of these cells. Notably, the percentage of Ng2-positive cells within the Ng2-clones (prevalence) was not changed (Fig. 4f), as would instead be predicted if PDGF-B specifically promoted the proliferation of these cells. This finding further reinforces the notion that the increased abundance of Ng2-positive cells was not due to the selective expansion of already Ng2-positive clones by PDGF-B. Finally, the total number of transduced clones did not significantly differ between PDGF-B- and control-transduced cultures (data not shown), suggesting that cell survival was not a significant factor in these experiments. Altogether, the results of these experiments show that, in the short term in vitro analysis, overexpression of PDGF-B affected E14 precursors cell lineage without significantly acting on their proliferation. Rather, PDGF-B induced a de-novo expression of Ng2 in the progeny of precursors that otherwise would not have generated Ng2 positive cells. The effects of PDGF-B on differentiation are therefore independent from the effects on proliferation, which surely occur in the long term in vivo, as indicated by the appearance of hyperplasiae and tumors. In interpreting these data, it is important to consider that, at this developmental stage, most embryonic neural progenitors are already strongly committed to a specific lineage, with as much as about 50% unipotent neuron-generating progenitors. These cells differentiate into postmitotic neurons in about 24 hr after plating (data not shown) and are therefore likely unaffected by PDGF-B overexpression.25 The relative increase in the frequency of the Ng2-clones we observed is therefore more noteworthy, as the proportion of immature progenitors which can be potentially affected by PDGF-B is only a fraction of the total number of progenitors.

These in vitro observations suggest that the uniformity in tumor subtypes generated following the transduction of PDGF-B in embryos, could be due to the ability of PDGF-B to respecify embryonic precursor cells toward the oligodendroglial lineage.

PDGF-B-induced tumors comprise a proliferating nontumorigenic glial progenitor population

On close inspection, PDGF-B-induced tumors were found to harbor proliferating (Ki67-positive) GFP-negative cells in the close vicinity of the GFP-positive PDGF-B-overexpressing cells (Fig. 5a). Interestingly, the GFP-negative population contained a percentage of actively proliferating cells (11 ± 2.9%) very similar to that found in the GFP-positive population (15 ± 2.3%, Fig. 5b). Since GFP-negative cells represented a large proportion of the tumors (32 ± 12%), their contribution to the growing fraction of the tumors was significant. Further analysis of the GFP-negative population found within gliomas revealed that it was very similar to the GFP-positive 1, displaying an OPC-like phenotype, as shown by the expression of Olig2, NG2, PDGFRα and APC (Figs. 5c5e, Table II, data not shown).

Figure 5.

Proliferating GFP-negative cells found within PDGF-B-induced tumors may represent recruited glial progenitors, but lack tumorigenic potential. (a, ce) Immunofluorescence stainings of tumor sections with anti-GFP antibody in green, DAPI for nuclear staining in blue and the antibody for the indicated antigen in red. (b) Histogram shows the percentage of proliferating (Ki67-positive) cells among the GFP-negative (red bar) and GFP-positive (green bar) population within the tumor mass (percentages represent the averages of values from 3 distinct tumors; >500 GFP-positive cells counted for each tumor). (f) Survival curves for animals receiving GFP-positive (green line) and GFP-negative (blue line) FACS-purified cells from acutely dissociated gliomas. Asterisk marks a tumor derived from contaminating GFP-positive cells, as shown in the inset image obtained by merging bright-field and fluorescent photographs. Arrows indicate examples of GFP-negative cells expressing the indicated markers. Scale bar: 50 μm.

Table II. Neural Marker Expression in GFP-Negative Cells Close to the Tumor Mass Compared with Control Region
MarkerPercentage among GFP-negative cells in
GFP-positive cells vicinityControl areas
  1. The percentage of positivity to the indicated molecular markers ± the standard error is indicated both for the GFP-negative cells near to the GFP-positive masses and for control regions. Data were collected from three independent experiments. In brackets are indicated the total number of cells counted.

Olig246 ± 6 (2327)6 ± 2 (1898)
Ng261 ± 7 (1257)16 ± 4 (1168)

These observations suggested that the PDGF-B secreted by transduced cells could also act paracrinally to recruit neighboring cells to the growing tumor mass, possibly resulting in their transformation. This possibility was also supported by the recent observation that gliomas induced by the overexpression of PDGF-B in adult white matter glial progenitors, can paracrinally recruit glial progenitors to the tumor.14 However, the presence of a GFP-negative fraction in the tumor may also be explained by the silencing of the PDGF-B/GFP-expressing provirus. We indeed observed this phenomenon in vitro, when culturing PDGF-B-overexpressing tumor cells,33 and it may also occur in vivo. To clarify this issue, we dissociated a PDGF-B-induced tumor and we FAC-sorted the GFP-positive and GFP-negative cells (Supp. Info. Fig. 2A–B). The GFP-negative population represented one third (33%) of the population gated by morphological features (i.e., FSC/SSC), in very good agreement with the observed proportion of GFP-negative cells within the tumor sections (32 ± 12%, see earlier section). A real time quantitative PCR analysis on the genome of the 2 sorted populations revealed that the amount of integrated PDGF-B/GFP provirus in the GFP-negative cells was about 9% of that found within the same number of GFP-positive cells (Supp. Info. Fig. 2D). This percentage was well above the contamination of GFP-positive cells in the GFP-negative sorted population, which was lower than 0.3% (Supp. Info. Fig. 2C), indicating that a small, yet detectable, fraction of the GFP-negative population was represented by silenced cells. The remaining 91% of GFP-negative cells, accounting for about 30% of the original unsorted population, were therefore not transduced. Some of these cells may represent a contamination from the surroundings of the tumor. Nevertheless, this cannot certainly be the case for the totality of the GFP-negative-untransduced cells, especially considering that only the inner part of the tumor masses was dissected. This, in turn, implies that a relevant fraction of the GFP-negative population was constituted by recruited cells.

The presence of recruited glial progenitors within PDGF-B-induced gliomas, prompted us to test the possibility that these cells had acquired a tumorigenic potential as a result of the exposure to the tumor microenvironment. Sorted GFP-positive and GFP-negative fractions were separately reinjected intracranially in adult mice. Remarkably, these experiments revealed a dramatic difference between the 2 cell populations. Although the majority of the brains receiving GFP-positive cells from high-grade tumors developed gliomas (60% of cases, n = 10 from 4 independent tumors), none of the animals receiving GFP-negative cells (n = 9 from 4 independent tumors) developed a tumor. The only tumor that developed in animals receiving GFP-negative cells was a GFP-positive glioma, clearly resulting from the contamination affecting sorting procedures (Fig. 5f). This observation showed that even few tens of GFP-positive cells could be enough to reestablish a tumor as we previously demonstrated.33

Notably, GFP-positive PDGF-B-overexpressing cells were also able to be serially propagated as tumors in vivo, following dissociation and reinjection of the secondary tumors obtained from the primary reinjection experiments (data not shown). This observation demonstrates the presence, within the primary tumors, of cells with long-term tumor-propagating potential, a feature typically associated with highly malignant tumor initiating cells, highlighting the fully transformed nature of the PDGF-B-induced gliomas. Altogether, these findings provide evidence that, despite the strong cell-autonomous transforming potential displayed by PDGF-B, the gliomas it induces do not exert a significant transforming action on the surrounding brain tissue components.


We have found that PDGF-B overexpression in telencephalic mouse neural progenitor cells is highly efficient at inducing tumors. By performing immunohistochemical and genome-wide gene expression analyses, we demonstrated that PDGF-B-induced tumors are a very uniform class of gliomas. All these tumors share unambiguous markers of the oligodendroglial lineage, even when displaying histopathological traits typical of glioblastoma, such as extended hemorrhage and necrosis, pseudopalizades and widespread angiogenesis. Comparative gene expression analysis revealed that high-grade PDGF-B-induced gliomas most closely resemble OPCs, reinforcing our view that these tumors represent high-grade oligodendrogliomas. This interpretation also gains support from, and provides evidence for, the notion that some of the human tumors named “glioblastomas with oligodendroglioma component” in the last revision of the WHO classificatory scheme could indeed be considered as “Grade IV oligodendroglioma.”3

Interestingly, such a homogeneous class of tumors was generated following the infection of a highly heterogeneous population of immature neural progenitor cells belonging to all possible lineages of the CNS.30 This result is intriguing, considering that PDGF-B overexpression can induce different gliomas in a cell type-specific manner when targeting different subpopulations of more mature neonatal neural progenitors.10 Furthermore, most if not all neonatal neural progenitors, likely comprising the populations targeted by PDGF-B overexpression in previous studies, have been shown to derive from the embryonic neural progenitors targeted by our approach.30, 31 A possible reason for this result may be found in the greater plasticity of embryonic precursor cells, which may be more easily committed or respecified toward the oligodendroglial lineage. The results of our in vitro clonal analyses indeed show that the overexpression of PDGF-B increases the proportion of oligodendrocyte progenitors by driving immature cortical precursor cells toward the oligodendroglial lineage rather than by selectively stimulating the proliferation of already committed oligodendrocyte precursors. This represents a novel view to our knowledge, since PDGF signaling has classically been regarded as mitogenic for members of the oligodendroglial lineage.41, 42 We provide evidence that the oligodendroglial identity and uniformity of the PDGF-B-induced gliomas result from an early induction process driven by PDGF-B on immature neural progenitors, rather than from the specific targeting of a responsive glial progenitor population. Such an instructive effect of PDGF-B on immature neural progenitors is in agreement with evidence provided by Hu and colleagues suggesting a role for PDGF-A in driving the acquisition of oligodendroglial features by embryonic neural progenitors. Our data are also well in agreement with the observed ability of PDGF-A to induce gliomas and to force oligodendrocyte formation while reducing interneuron production from subventricular zone adult neural stem cells.15, 43 Altogether, these data suggest that the most immature neural progenitor/stem cells at different ontogenetic stages respond to PDGF signaling by acquiring an oligodendroglial identity. The PDGF-B fate-influencing activity we have observed seems therefore to contribute to the gliomagenic process by driving the acquisition of specific cell-lineage features. These observations are modifying our view of brain tumorigenesis, from an ill-defined process whereby different oncogenic stimuli induce different tumors in an unpredictable manner, to a scenario in which the differentiation potential and developmental plasticity of the targeted cells actively shape the “gliomagenic landscape,” determining the available transformation pathways.

Other relevant issues in tumor biology are represented by the high heterogeneity of primary tumors and the interactions between tumors and components of the surrounding tissue environment. Interestingly, we have observed that PDGF-B-induced gliomas harbor abundant GFP-negative proliferating cells with glial progenitor-like features, closely resembling the GFP-positive population. We demonstrated that these cells, which are found only in the close vicinity of the PDGF-B-overexpressing cells, mainly represent resident cells recruited by the tumor, significantly contributing to the proliferation of the tumor. Interestingly, we have found that the 2 cell populations, despite their high degree of similarity, deeply differ in their tumorigenic potential, since only the GFP-positive population shows the ability to propagate as tumors after in vivo intracranial reinjections. This demonstrates that neither PDGF-B nor any factor secreted by the PDGF-B-induced gliomas is able to paracrinally transform cellular components of the surrounding brain tissue.

The presence of an easily distinguishable, nontumorigenic, proliferating cell population within these PDGF-B-induced gliomas makes this model well suited for studying the poorly understood relationships between the different cellular components of glial tumors.


We thank Dr. Ennio Albanesi for FacSorting. Fondazione Italiana per la lotta al Neuroblastoma funded Ph.D. scholarship to F.C.