Gli Activity Correlates with Tumor Grade in Platelet-Derived Growth Factor–Induced Gliomas

Glioma-associated oncogene homologue Gli was first identified as a gene amplified in gliomas (1). The product of this gene was later found to be a component of the sonic-hedgehog (SHH) signaling pathway (2). SHH signaling regulates multiple aspects of central nervous system (CNS) development, controlling both cell proliferation and cell differentiation (3, 4). SHH regulates proliferation of granule cell precursors in the developing cerebellum (5) and contributes to oligodendrocyte specification in the mammalian forebrain (6, 7). Furthermore, SHH is necessary and sufficient for oligodendrocyte precursor production in cortical neuroepithelial cultures (8) and drives proliferation in CNS precursor cells (9, 10).

There are several genetically accurate mouse models of gliomas that reproduce the histologic features and characteristic diffuse infiltration of these tumors (11). One such model is the RCAS/tv-a system that can recapitulate the interaction between glioma cells and the brain by driving autocrine platelet-derived growth factor (PDGF) receptor stimulation (12). This system uses an avian retroviral vector to transfer gene expression to mice genetically modified to express the receptor, tv-a. The nestin tv-a transgenic mouse (Ntv-a) expresses the avian viral receptor tv-a from the nestin promoter, allowing RCAS retroviral transduction to nestin-expressing neural stem cells (13). The relevance of PDGF signaling in human gliomagenesis is well described. PDGF and PDGF receptor are overexpressed in human glial tumor surgical samples, and increased expression is correlated with higher tumor grade (14). The dose-dependent effects of PDGF in glial tumorigenesis have been described, whereby increased PDGFB levels correspond to the formation of microvascular proliferation (MVP) and pseudopalisades, both elements of high-grade character (15). Pseudopalisades or pseudopalisading necrosis are hypercellular zones that surround necrotic foci and define high-grade gliomas. MVP is the second histologic feature that accompanies pseudopalisades and defines high-grade gliomas; it refers to endothelial proliferation within newly sprouted vessels.

There are hints of interplay between PDGF and SHH signaling in glioma development. PDGF-responsive precursors from the ventral forebrain form neurospheres, which are dependent on SHH signaling (16). Furthermore, the SHH pathway is required for cell proliferation in the mouse forebrain's subventricular zone stem cell niche (17) where gliomas have been proposed to arise (18). To the extent that gliomas parallel glial cell development, these observations raise the possibility that SHH signaling may be active in gliomas.

To determine whether Gli is activated in PDGF-induced gliomas, we generated a reporter mouse that gives an accurate readout of Gli activation. This mouse line, termed Gli-luc, expresses luciferase from a Gli responsive promoter activated by Gli, a primary target of SHH signaling. This report notes that SHH expression and Gli activity correlate with grade in PDGF-induced gliomagenesis. In low-grade gliomas, SHH-producing astrocytes were observed mainly at the periphery of the tumor, whereas in high-grade gliomas, numerous SHH-producing astrocytes were observed inside the tumors. In high-grade gliomas, tumor endothelial cells, as well as abutting astrocytes, in areas of MVP and the cells in pseudopalisades were SHH immunoreactive. Most of these SHH immunoreactive cells were not derived from the cell of origin and were low cycling. In addition, we created PDGF-induced gliomas in Ptc-lacZ mice and noted perivascular x-gal staining in high-grade gliomas with a subset of tumors having diffuse x-gal staining. We noted two populations of SHH responsive cells: Olig-2–positive tumor cells and glial fibrillary acidic protein (GFAP)–positive astrocytes that primarily reside in the perivascular niche (PVN). We observed similar immunostaining in human gliomas, where the SHH-producing cells were also found in PVN as astrocytes or endothelial cells and in pseudopalisades. In addition, we noted that Gli activity was rapidly lost in vitro when PDGF-induced glioma primary cultures were grown as a monolayer with serum but maintained when grown as neurospheres. Thus, histologically accurate in vivo models or gliomas cultured as neurospheres are imperative to studies of Gli activity in gliomas.

Generation of transgenic mice. pGL3B/8GliBS-luc plasmid (American Type Culture Collection) includes eight Gli DNA-binding consensus sequences (′GAACACCCA′) upstream of the chicken lens crystalline promoter, which is upstream of the firefly luciferase gene. The 3′ end contains an SV40 polyadenylation sequence. The plasmid was digested with KpnI and AfeI to remove the transgene (2.7 kb in size). We used AseI to digest the plasmid backbone as plasmid and insert were similar in size. Chimeric founder mice were generated by pronuclear microinjection of the linearized Gli-luc construct into fertilized FVB oocytes. Chimeric founder mice were genotyped via PCR using the following primers: 5′CGGGCGCGGTCGGTAAAGT′3 and 5′AACAACAACGGCGGCGGGAAGT′3. We identified nine chimeric founders with germline transmission. Only one chimeric founder (10) showed tight regulation of expression in vitro (Granule cell preparation).

Imaging of organs. At postnatal days 1, 2, 5, 7, 9, 11, 14, 20, and 44, mice were injected with 75 mg/kg of luciferin (30 mg/mL in water, i.p.), imaged in the IVIS for 1 min (Xenogen) to confirm positive Gli-luc expression, and then sacrificed. Intestine, testes, brain, lungs, stomach, kidney, spleen, pancreas, bladder, liver, and heart were dissected, placed in 0.4 mg/mL luciferin in PBS, and imaged for 30 s.

Granule cell preparation. Cerebella from Gli-luc mice at 4 to 5 postnatal days were prepared as described (19). Three experimental conditions were evaluated: media alone, SHH, and SHH + cyclopamine (gift from Dr. Gardner, U.S. Department of Agriculture). Plates were imaged using IVIS 24 h later.

Generation of mouse brain tumors. Double-transgenic neonatal Gli-luc;Ntv-a mice were injected intracranially (cortical injections) with 1 μL of DF-1 cells producing RCAS-PDGFB retrovirus to generate gliomas, as described previously (15). Mice were monitored carefully for symptoms of tumor development (hydrocephalus, lethargy, head tilt). All injected mice were routinely screened with bioluminescent imaging (BLI), and image-positive mice were sacrificed. PDGF induced gliomas in double transgenic Ptc^+/−;Ntv-a mice were generated in a similar fashion. X-gal staining was done as per standard protocol.

BLI of Gli-luc mice. Mice are anesthetized with 3% isolflurane before retroorbital injection with 75 mg/kg body weight luciferin (Xenogen). One minute after injection of the luciferin, images are acquired for 4 min with the IVIS (Xenogen). A photographic image was taken onto which the pseudocolor image representing the spatial distribution of photon count is projected. We defined a circular region between the ears and used it as a standard in all experiments. From this region, photon counts are compared between different mice.

Tissue microarray construction. Normal brain and brain tumor tissues were fixed with formalin and embedded in paraffin. Five-micron sections stained with H&E were obtained to identify viable representative areas of the specimen. After Institutional Review Board approval, defined areas core biopsies, mainly enriching for viable representative tumor areas, were taken with a precision instrument (Beecher Instruments). Tissue cores with a diameter of 0.6 mm from each specimen were punched and arrayed in triplicate on a recipient paraffin block. Sections (5 μm) of these tissue microarray (TMA) blocks were cut and placed on charged poly-lysine–coated slides and were used for immunohistochemical analysis.

Generation of gliomas using PDGF-eGFP. RCAS-PDGFB-SV40-eGFP (PDGF/GFP) vector clones were derived from the RCAS-PDGFB vector containing human PDGFB with partially deleted 5′ untranslated region by direct cloning of the SV40-eGFP construct. For tumor induction, neonatal Ntv-a pups were infected with PDGF/GFP by intracranial injection. Mice were observed daily for the appearance of brain tumor symptoms. Upon the appearance of brain tumor symptoms, mice were anesthetized by an i.p. injection of ketamine (100 mg/kg)/xylazine (10 mg/kg) cocktail and underwent transcardiac perfusion with 10 mL ice-cold 0.01% heparin/0.9% sodium chloride followed by 10 mL ice-cold 4% paraformaldehyde. Brain tissue was extracted and postfixed for 30 min in ice-cold 4% paraformaldehyde, transferred into 30% sucrose at 4°C for cryoprotection, embedded using the ornithine carbamyl transferase compound, frozen on dry ice, and stored at −80°C until sectioned.

Generation of Gli reporter mouse. The use of BLI is an established method that allows for sequential imaging of a signaling pathway in vivo in real time (20). To image Gli activity in vivo, we created a transgenic reporter mouse using a Gli-luciferase construct designed by Sasaki. The construct contains eight Gli DNA binding sites in front of the chicken lens crystalline promoter that is upstream of the firefly luciferase gene. In vitro, this construct is activated by both Gli1 and Gli2, but not by Gli3 (21). The transgene is ubiquitous, and regions of Gli activity can be visualized after injection of luciferase substrate. As Gli is constitutively active in normal skin, there is background luciferase expression throughout the entire mouse in the raw image, which can be adjusted to a threshold image by adjusting the minimum and maximum bars (Supplementary Fig. S1). As fur absorbs some of the light, there is increased light in hairless areas, such as the ears, nose, and tail.

Validation of reporter expression specificity. We investigated the specificity of the Gli-luc reporter mouse by imaging various organs during development to determine whether luciferase expression correlates with the published literature for Hedgehog/Gli signaling. We injected Gli-luc transgenic pups with 75 mg/kg of luciferin, sacrificed the animals 1 minute later, removed the organs, placed them in six-well plates containing 0.4 mg/mL luciferin in PBS (heretofore called PBS/luciferin), and imaged for 30 seconds in IVIS (Xenogen). Figure 1A shows a graph of luciferase expression in the intestine, testes, brain, lung, stomach, kidney, spleen, pancreas, bladder, liver, and heart at postnatal days 1, 2, 5, 7, 9, 11, 14, 20, and 44. Individual graphs of each organ with SEs can be seen in Supplementary Fig. S2. Luciferase expression is not normalized for organ size. Nonetheless, most organs, such as brain, kidney, stomach, spleen, pancreas, lung, heart, and liver, initially have high luciferase expression levels, which decrease once the mouse reaches adulthood. Organs, such as the liver, have relatively low expression during development in contrast to other organs, including brain or kidney. The intestines and the skin, both self-renewing organs, maintain high expression into adulthood. The testes, which mature when spermatogenesis is initiated, have high expression after male mice reach adulthood. These data fit well with previously published data for Gli activity in vivo (22–24).

Next, we closely examined brain development. Figure 1B shows images of the brain immersed in PBS/luciferin and imaged using the IVIS 100 at various postnatal days. During postnatal days 0 to 7, there is a high luciferase activity in both cortex and cerebellum, followed by a decrease in luciferase expression in the cortex, whereas expression in the cerebellum is maintained during postnatal days 7 to 14. At postnatal day 20, there is little detectable luciferase activity in either cortex or cerebellum. These data are consistent with the known SHH activation during oligodendrocyte development in the first week of life and SHH pathway activation in the developing cerebellum up to postnatal day 14. Thus, the Gli-luc reporter seems to be regulated appropriately during CNS development. The low-light production in adult mice may not indicate a complete lack of Gli activity. Therefore, to investigate whether a low level of Gli activity might exist in the stem cell regions of the adult brain, we injected adult Gli-luc mice with 75 mg/kg of luciferin, sacrificed the animals 1 minute later, removed the brain, sliced into 2-mm slices, placed slices in PBS/luciferin, and imaged for 5 min in IVIS. We noted luciferase expression in the hippocampus and brain stem (Fig. 1C). These are regions where Gli activity has been shown to remain active in adult rodents (25–27).

To confirm that the Gli-luc reporter is regulated by SHH through Smoothened (Smo), we tested the effects of Smo inhibition in vitro. During the first 2 weeks postnatally, cerebellar granule cells proliferate in response to SHH activation. We prepared cultures of these cells and stimulated them with SHH, with SHH + cyclopamine, or with media alone. Figure 1D shows that the luciferase light production increases in response to SHH and is blocked in response to 2.4 μmol/L cyclopamine in granule cells, thus demonstrating dependence on smoothened function. We conducted parallel studies and stained cyclopamine-treated cells with trypan blue. There was no difference in the percentage of cells that excluded trypan blue (>90% viability in controls and cyclopamine-treated cells). Thus, the reduction in light in our cyclopamine-treated EGL cells is due to down-regulation of the SHH pathway rather than nonspecific cell toxicity.

Bioluminescence imaging of PDGF-induced gliomagenesis. To determine whether PDGF-induced gliomas have Gli activity, Ntv-a;Gli-luc mice were infected with an RCAS-PDGFB virus that is known to induce gliomas in Ntv-a mice by injection of virus into the cortex of neonatal mice. Double transgenic mice infected with RCAS-PDGFB were monitored weekly after 4 weeks of age for evidence of increased luciferase activity over the cranium by BLI (Fig. 2A). We identified 30 out of 42 mice injected with RCAS-PDGFB with elevated intracranial light relative to controls (Fig. 2B). We confirmed that the additional light was due to the presence of a tumor by sacrificing mice that emitted additional light, placing brain in PBS/luciferin, imaging (Fig. 2A), and verifying the presence of gliomas in these mice by histologic analysis (Fig. 2C). Thirty tumors, comprising a mixture of all grades, were analyzed in this manner—in all cases imaging indicated Gli activity throughout the tumor. Twelve of 42 mice injected with RCAS-PDGFB did not have increased light over the cranium (≤200,000 photons/s in FVB mice). We sacrificed these mice at 12 weeks, analyzed the brains by histologic analysis, and found that these 12 mice did not have tumors. Thus, PDGF-induced gliomagenesis activates Gli, and the Gli-luc mouse is useful for imaging Gli activity in gliomas in vivo. We also noted that high-grade gliomas with MVP and pseudopalisades had the highest light production. High-grade gliomas with MVP alone had the second highest light production. Gliomas with the lowest light production had neither MVP nor pseudopalisades. These differences were statistically significant (Fig. 2D). In contrast, the difference between low-grade gliomas and control mice was not statistically significant.

Astrocytes express SHH in both low-grade and high-grade gliomas. We next investigated the mechanism of Gli activation in PDGF-induced gliomagenesis. Several signaling pathways are known to activate Gli; the best-known pathway is SHH ligand binding to Ptc (28). We therefore immunoblotted three PDGF-induced gliomas for the presence of SHH and observed that SHH is overexpressed relative to normal cortex and at similar level to the developing cerebellum at postnatal day 4 when SHH signaling is at its maximum level (Supplementary Fig. S3A). We performed in situ hybridization for SHH in these tumors and found expression of the RNA concentrated around MVP and pseudopalisades (Supplementary Fig. S3D). To further characterize which cells within the tumor produce SHH, we used immunofluorescence. In low-grade gliomas, we noted very few SHH immunoreactive cells within the tumor but numerous SHH immunoreactive cells in the periphery of the tumor (Fig. 3A). In high-grade gliomas with MVP, there were increased numbers of SHH immunoreactive cells in areas of PVN, as well as brighter SHH staining throughout both the tumor and the periphery of the tumor (Fig. 3B). In high-grade gliomas with MVP and pseudopalisading structures, in addition to the dramatic increase of SHH immunoreactive cells and brighter SHH staining, we noted a second population of SHH-positive cells located mainly along the rim and the inside of pseudopalisading structures (Fig. 3C). Overall, there was a correlation between glioma grade and SHH expression in that high-grade gliomas showed stronger and more widespread SHH expression relative to low-grade gliomas. Both SHH RNA and protein associate with structures that define high-grade gliomas. Consistent with observations by others, we were not able to generate gliomas by infecting nestin-expressing neural stem cells in the forebrain of Ntv-a neonatal pups using RCAS-SHH, suggesting the SHH is not sufficient to initiate gliomagenesis (data not shown; ref. 29).

Characterization of SHH-expressing cells in high-grade gliomas. A large fraction of the SHH-expressing cells had astrocytic morphology and immunostained for expression of GFAP and S100β (Fig. 4A and B and Supplementary Fig. S4). We noted that these astrocytic SHH immunoreactive cells mainly reside in PVN and colocalize but are mainly nonoverlapping with nestin immunoreactive cells (Fig. 4D) and triple immunofluorescence (Supplementary Fig. S5A).

We observed a second population of SHH immunoreactive cells that were negative for GFAP (Fig. 4A and C) and S100β (data not shown) and were located in rim of pseudopalisades. These cells were nestin negative (data not shown). Both populations of cells have a low mitotic index as shown by SHH and proliferating cell nuclear antigen (PCNA) double immunofluorescence (only the astrocytic SHH population is illustrated in Fig. 5A and B). We confirmed a low mitotic index in these cells with both a 24-hour pulse with BrdUrd followed by double immunofluorescence for SHH and BrdUrd, as well as by double immunofluorescence for Ki67 and SHH (data not shown). A third population of SHH immunoreactive cells were GFAP negative, located adjacent to the lumen of blood vessels with elongated cell body and a flattened nuclei, and stained positive for CD34 (Supplementary Fig. S8C). In contrast, normal blood vessels in the adult murine brain did not express SHH. As the SHH immunoreactive cells had a low proliferative rate, we investigated whether the SHH immunoreactive cells were progeny derived from the original infected cell or part of the brain stroma.

Most SHH-expressing cells are not derived from the tumor cell of origin. We infected nestin positive neural stem cells in Ntv-a mice with an RCAS-PDGFB-SV40-eGFP (PDGF/eGFP), whereby PDGFB is expressed from the retroviral promoter and eGFP is expressed from SV40 promoter. The tagged PDGF protein allows for lineage tracing of tumor cell of origin. Ntv-a mice generated with this PDGFB/eGFP vector develop gliomas in a similar fashion to RCAS-PDGF. Tumor bearing mice were sacrificed and cryosections were prepared and immunostained for SHH. We observed that SHH immunoreactive cells were mostly GFP negative. With the exception of a few SHH and GFP double-positive cells, SHH immunoreactive cells were GFP negative (Fig. 5C and D).

Localization of SHH immunoreactive astrocytic cells in the perivascular niche. Whereas in most glial tumors MVP is a sign of malignant progression, its exact cellular composition is not well defined. Historically, MVP has been considered to be proliferation of endothelial cells, but recent studies show that pericytes or vascular smooth muscle cells (vSMC) also play a key role (30). Additionally, elevated PDGFB levels in PDGF-induced gliomas mediate vascular smooth muscle recruitment that supports tumor angiogenesis (15). Using confocal microscopy and triple immunofluorescence for nestin, SHH, and pS6RP, we observed that astrocytic SHH immunoreactive cells localize adjacent to nestin immunoreactive cells and a subset of these cells coexpress nestin (Supplementary Fig. S5A). Using confocal microscopy and triple immunofluorescence for SHH, smooth muscle actin (SMA), and pS6RP, as well as for SHH, SMA, and GFAP, we observed that SHH immunoreactive astrocytic cells are located on the inside of vessels, whereas vSMC (SMA positive and pS6RP positive) are located on the outside of vessels (Supplementary Fig. S5B and C).

SHH-responsive cells in PDGF-induced gliomas are Olig-2–positive tumor cells and GFAP-positive astrocytes. As Ptc is a downstream target of Gli, we used a second reporter mouse, Ptc-lacZ mice, to identify the Gli-expressing cells. This mouse strain is a knock-in of the lacZ gene into the ptc locus (31). We crossed Ntv-a mice with Ptc-lacZ mice and generated gliomas by infecting nestin-expressing cells with RCAS-PDGFB, as described above. Cryosections of gliomas generated in these double transgenic mice were stained for β-galactosidase activity using X-gal. Whereas low-grade gliomas had minimal staining, high-grade tumors show perivascular staining with a subset of high-grade tumors having both perivascular staining and diffuse X-gal staining throughout large areas of the tumor, areas where most of the tumor cells were positive (Supplementary Fig. S6A). To further characterize these SHH-responsive tumors cells, we immunostained for Olig-2, a marker for cells of the oligodendroglial lineage, and for neural stem cells (32). We immunostained sections of these diffusely positive tumors for both X-gal and Olig-2 and noted significant overlap in most of the cells (Supplementary Fig. S7). In addition, we noted intensely staining X-gal–positive cells with larger nuclei that cluster in the PVN as well as throughout the tumor. Double staining with GFAP confirmed that these cells are astrocytes (Supplementary Fig. S6B–D). In summary, a subset of high-grade PDGF-induced gliomas had large areas of Ptc expression throughout the tumor bulk with astrocytes, preferentially residing in PVN, having the most Ptc expression. This suggests diffuse Gli activity throughout the tumor and potentially autocrine hedgehog signaling in astrocytes.

Human glioblastomas multiforme and oligodendrogliomas also overexpress SHH. We determined whether SHH expression is conserved between mouse and human gliomas by performing a SHH Western blot on human glioblastoma multiforme (GBM) and oligodendroglioma lysates and found that most (16 of 20) GBMs overexpressed SHH relative to normal cortex (temporal lobe) and five of six oligodendrogliomas overexpressed SHH relative to temporal lobe (Fig. 6A). We obtained several human surgical specimens of high-grade gliomas for immunostaining for SHH and noted that SHH immunoreactive cells in human gliomas are similar to SHH-producing cells in murine gliomas (Fig. 6B). As in the murine tumors, astrocytes, endothelial cells, and cells residing in pseudopalisades are sources of SHH in human gliomas. The low proliferation rate of SHH immunoreactive astrocytic cells is evident in human gliomas by double immunofluorescence for SHH and Ki67 (Supplementary Fig. S8A). We confirmed that endothelial cells are SHH immunoreactive in human gliomas by double immunofluorescence for SHH and von Willebrand factor, an endothelial cell marker (Supplementary Fig. S8B).

We then immunostained an independent brain tumor TMA for SHH to confirm the Western blot data. The scoring was rated from 0 to 4, where 0 denotes absence of staining and 4 denotes 100% of the cells positive for SHH expression. Thirty-one of 35 or 89% of gliomas rated at least a 1 or higher. Five of five astrocytomas stained 2 or higher, 9 of 13 anaplastic oligodendrogliomas stained 1 or higher, 8 of 8 GBMs stained 1 or higher, 9 of 9 glial regions of gliosarcomas (GSA) stained 1 or higher, whereas the sarcomatous parts of GSAs were mostly negative (Fig. 6C). Therefore, SHH expression is common across many subgroups of human gliomas.

Human glioma cell lines do not have Gli activity. Although Gli was originally described in gliomas (1), clear activation of this pathway in glioma cell lines has not been reported. Therefore, we examined commonly used human glioma cell lines to determine if Gli activity is intact in vitro. First, we performed a Western blot on U87MG, T98G, U373MG, and U251 cell lines for SHH and found that only U251 cells expressed SHH (Supplementary Fig. S9A). We then assessed whether the Gli is active in those cell lines by transiently transfecting the cell lines with the same Gli-luc construct used to generate the transgenic mouse and determined whether cell lines could respond to SHH stimulation. Although all four cell lines were able to respond to a cotransfection with a construct expressing activated Gli used as a positive control, none of the human cell lines responded to exogenous SHH ligand in low or normal serum concentrations (Supplementary Fig. S9A). These results indicate that commonly used glioma cell lines that are used for in vitro and xenograft studies have lost Gli activity.

Gli activity in PDGF-induced gliomas is down-regulated in vitro in serum. It has been shown that medulloblastoma cells lose Gli activity when placed in culture (33). We were interested to determine whether Gli activity in gliomas is similarly dependent upon an intact tumor microenvironment or differentiation status. Therefore, we cultured PDGF-induced gliomas from a Gli-luc background. Cells were cultured in DMEM with 10% FCS on poly-l-ornithine–coated plates and imaged daily. We noted that luciferase expression decreased steadily over the first few days in culture and was completely absent after two passages (Supplementary Fig. S9B and C). We then asked whether these tumor cells could respond to SHH in vitro. Tumor cells at 70% confluency were incubated with SHH at 3 μg/mL (in 0.5% FCS media) for 24 hours and imaged for 5 minutes after the addition of luciferin. At this point, there was no detectable luciferase activity indicating SHH unresponsiveness in three independent primary cell cultures derived from PDGF/Gli tumors (Supplementary Fig. S9D) The light II 3T3 cells (3T3 cells stably transfected with same SHH responsive Gli-luc containing plasmid) were used as a positive control for Gli activity.

Gliomas cultured as neurospheres maintain Gli activity. As neurosphere culture conditions are thought to maintain stem cells in an undifferentiated state in vitro, we isolated PDGF-induced gliomas in Gli-Luc background and cultured tumor cells as neurospheres with epidermal growth factor and basic fibroblast growth factor. Primary neurospheres were developed in 5 days and dissociated to single cells every 5 days. We imaged the neurospheres before dissociation every 5 days and observed that neurospheres maintain Gli activity for several passages. We compared the level of luciferase expression per cell of glioma cells cultured in neurospheres versus serum after two passages and noted that glioma cells cultured as neurospheres have >10-fold higher levels of luciferase expression per cell than glioma cells grown in serum (Supplementary Fig. S9C). Thus, glioma cells maintain Gli activity in vitro when cultured as neurospheres but not when grown in serum.

Whereas the SHH pathway clearly drives oncogenesis in medulloblastomas and basal cell carcinomas through pathway-activating mutations (31), the significance of Gli activity in gliomagenesis is not clear. Several studies since the original description of the Gli gene have shown that Gli is amplified in a small subset of gliomas (34–36). It has also been shown that mRNAs encoding elements of the SHH pathway are expressed in gliomas, and cyclopamine inhibits proliferation of glioma cell lines and gliomaspheres in vitro, as well as in orthotopic xenografts (37–40). However, other studies using glioma cell lines have suggested that the SHH pathway plays a relatively minor role in gliomagenesis (41, 42), and one study suggested that the SHH pathway inversely correlates with histologic malignancy (43). One potential reason for the conflicting data on the SHH pathway in gliomas could be explained in part by our demonstration that an intact tumor microenvironment or neurospheres conditions are required to maintain Gli activity, as it is down-regulated in vitro when gliomas are cultured as adherent cells with serum.

We show a SHH-producing microenvironment in the perivascular niche of gliomas. Endothelial cells and astrocytes are two populations of SHH-producing cells found in regions of MVP. These observations fit with the published literature demonstrating that hedgehog signaling is involved in vascular development, as well as postnatal angiogenesis (44–46). In addition, SHH immunoreactive astrocytes have been identified in models of CNS injury, such as an experimental autoimmune encephalitis model (47). Our observation that SHH-immunoreactive astrocytes reside in PVNs in close association with nestin-expressing tumor cells in gliomas is consistent with a recent report that suggest that cancer stem cells in brain tumors reside in a vascular niche. Endothelial cells have been suggested to secrete factors that maintain the self-renewal of brain tumor stem cells in culture (48), whereas others have shown that endothelial cells stimulate the self-renewal and expand neurogenesis of normal neural stem cells (49). In our study, we show that these SHH immunoreactive cells are not derived from the cell of origin, which suggest that these cells are reactive astrocytes or neural progenitor cells recruited to the tumor. Given the role of SHH pathway in normal neural stem cell self-renewal (17, 26), it is likely that the SHH-producing microenvironment in the PVN contributes to the glioma stem cell niche.

This study emphasizes the importance of in vivo models which allow one to unravel contributions of distinct cell subtypes in a tumor, i.e., SHH-producing cells in our PDGF-induced gliomas were astrocytes, endothelial cells, and pseudopalisade cells; these were mostly not tumor cells but part of normal stroma. Although Gli and SHH correlate with grade, other signaling pathways that also correlate with grade may be contributing to Gli activity, such as AKT, RAS, and transforming growth factor β (28, 50). It remains to be determined whether smoothened inhibitors, as either single agents or combined with other therapeutic modalities, will be useful in the treatment of gliomas.

Grant support: NIH grants R01 CA 5R01CA099489-1, R01 5R01CA100688-2, and R01CA/NS96582-01 and Kirby Foundation (E.C. Holland) and Charles Trobman Foundation, Witmer Foundation, Perry's Promise Fund and MSKCC K12 award (O.J. Becher).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Edward Nerio, Jim Finney, and Maria Salpietro for technical assistance; Dr. Katia Manova for assistance with histologic analysis; Dr. Kathryn Anderson for in situ probes; Dr. Dan Fults for RCAS-SHH; Dr. David Walterhouse for activated Gli construct; and Dale Gardner (U.S. Department of Agriculture) for cyclopamine.