Spinal Glioma: Platelet-Derived Growth Factor B–Mediated Oncogenesis in the Spinal Cord

Human platelet-derived growth factor B (hPDGFB) has been characterized in vitro and shown to mediate numerous cellular responses including glial proliferation and differentiation. Expression of PDGFB is thought to be important in the pathogenesis of glioma and several animal models of cerebral glioma based on PDGF expression have been described. To examine whether PDGF could contribute to the pathogenesis of spinal cord glioma, we developed transgenic mice that express hPDGFB under the control of a tetracycline-responsive element (TRE/hPDGFB). These TRE/hPDGFB mice were mated with transgenic mice expressing the tetracycline transcriptional activator (tet-off), tTA, regulated by the human glial fibrillary acidic protein (GFAP) promoter and exhibiting uniquely strong promoter activity in the spinal cord. These transgenic mice (GFAP/tTA:TRE/hPDGFB) expressed hPDGFB in GFAP-expressing glia in a manner responsive to doxycycline administration. Without doxycycline, almost all GFAP/tTA:TRE/hPDGFB mice developed spinal cord neoplasms resembling human mixed oligoastrocytoma. Tumorigenesis in these animals was suppressed by doxycycline. To further examine the importance of PDGFB in mouse primary intramedullary spinal cord tumors, we also created transgenic mice expressing hPDGFB under the control of the human GFAP promoter (GFAP/hPDGFB). These GFAP/hPDGFB mice also developed spinal oligoastrocytoma. PDGFB can mediate the development of mouse spinal tumors that are histologically and pathologically indistinguishable from primary intramedullary spinal tumors of humans and may provide opportunities for both novel insights into the pathogenesis of these tumors and the development of new therapeutics.


Introduction
Glial tumors are the most common tumors of the central nervous system (CNS).Although spinal gliomas represent only a small fraction of these tumors, the morbidity that such tumors cause is dramatic, and their prognosis is poor.These tumors may occur at any location along the spine, and their clinical presentation typically reflects both their location and grade (1)(2)(3)(4)(5)(6).Little is known of the molecular pathology that underlies spinal glioma, and there are no animal models of these tumors.To establish a mouse model of spinal glioma, we were informed by the molecular pathogenesis of glioma arising within the cerebrum.Although multifactorial, the pathogenesis of these tumors often involves members of the platelet-derived growth factor (PDGF) family.
Four members of the PDGF family are known.These growth factors are disulfide-bonded dimers consisting of homodimers of the A, B, C, and D polypeptide chains or heterodimeric molecules consisting of A and B chains (7,8).There are two PDGF receptor (PDGFR) subunits, a and h, which associate as homodimeric or heterodimeric molecules following their binding of PDGF ligand dimers to form a functional, activated PDGFR (9).These PDGFmediated pathways are thought to be of great significance for the establishment of primary glial tumors of the cerebrum.Pathologic specimens obtained from human glial tumors exhibit high levels of expression of both PDGF ligands and receptors (10).
Several investigators have pursued animal models as a means to better understand the role of PDGF in the pathogenesis of primary glial tumors (11)(12)(13)(14)(15)(16).The models that have been most extensively studied thus far use a recombinant Moloney murine leukemia virus encoding PDGFB (13,15) or an acute chicken leukemia virus system that allows a recombinant RCAS virus to express PDGFB (12,16).These models have been highly informative, however, it is impossible to rule out a possible contribution for insertional mutagenesis in the development of these tumors (17).
We sought to extend work previously done in the examination of glial tumor pathogenesis by developing a model of primary, intramedullary, and astrocytic spinal tumors.We pursued a model in which the wounding of tissue by injection and the ensuing inflammation was not a confounding problem and in which the integration of a retrovirus did not raise the issue of insertional mutagenesis.

Materials and Methods
Construction of plasmids and transgenic mice production.We sought animals in which the expression of human PDGFB (hPDGFB) in cells expressing the glial fibrillary acidic protein (GFAP) was inhibited in the presence of doxycycline.We prepared a plasmid construct from the pUHD 10-3 containing tetracycline-responsive element (TRE; ref. 18) obtained from Dr. H. Bujard (Center for Molecular Biology, University of Heidelberg, Heidelberg, Germany).A 0.8 kb BamHI DNA fragment encoding hPDGFB (19) was inserted downstream of TRE (see Fig. 1).DNA from this TRE/ hPDGFB plasmid was linearized by digestion with XhoI and HindIII and fertilized eggs from C57BL/6 mice were injected.Four different TRE/ hPDGFB founder mice were identified.All experiments used a single transgenic line designated 44.These mice were bred to animals previously described (20) in which the tetracycline trans-activator protein was under the control of the GFAP promoter, GFAP/tTA (tet-off) mice, to create GFAP/ tTA:TRE/hPDGFB transgenic animals for study.
To develop transgenic mice in which hPDGFB was constitutively expressed in GFAP-expressing cells, we prepared a GFAP/hPDGFB transgene by inserting the 0.8 kb BamHI DNA fragment into the 3 ¶-end of the 2.2 kb human GFAP promoter (hgfa2; see Fig. 4).A plasmid containing the GFAP promoter (hgfa2-0) was obtained from Dr. Michael Brenner (University of Alabama, Tuscaloosa, AL; ref. 21).Five different GFAP/ hPDGFB mice were identified.The line designated C8 had the highest hPDGFB mRNA CNS expression and was used.Animals were maintained as C57BL/6.C57BL/6 and p53-null mice of B6 were purchased (The Jackson Laboratory).
Analysis of gene expression and gene structure.Conventional realtime PCR (RT-PCR) was performed as described (22).Primers used to assess hPDGFB mRNA were sense primer (GTCCAGGTGAGAAAGATGGAG) and antisense primer (ATAACCCTGC CCACACACTCTC).Quantitative RT-PCR (qRT-PCR) was performed with iQ SYBR Supermix (Bio-Rad laboratories) on an iQ RT-PCR detection system (Bio-Rad Laboratories).The hPDGFB shown was used for microinjection.B, ELISA assay for hPDGFBB.Individual neurosphere cultures were established from each of six embryonic brains from E16 embryos obtained from a TRE/hPDGFB +/À female bred with a male GFAP/tTA +/À in the absence of tetracycline.Cell lysate (30 Ag) from each of these six spheroid cultures was prepared in triplicate and examined by an ELISA assay to measure hPDGFBB.Expression in lysates of neurosphere cultures from GFAP/tTA;TRE/ hPDGFB embryos (tTA-TRE ), a TRE/hPDGFB embryo (TRE ), a GFAP/tTA embryo (tTA ), and a wild-type (WT ) littermate.Columns, ELISA data points normalized to the average value obtained from the cultures of the wild-type embryo (WT ) and the average normalized value (ratio) of the six cultures for each embryo; bars, SD.C, regulation of hPDGFB expression by doxycycline in vitro.Normal untreated GFAP/tTA:TRE/PDGFB mice were sacrificed at 2 months of age and spinal cords were dissected.These tissues were dissociated, and the cells were seeded into six cultures.Initially, all cells were grown to confluence, and these cultures were randomly assigned into three groups.At that time, day 0, conditioned medium from each of the cultures was examined in triplicate by ELISA assay for hPDGFBB expression and the levels of expression in these cultures from all groups were found to be indistinguishable (black column ).Immediately after the conditioned medium was collected on day 0, culture 2 (dotted column ) was re-fed with medium and cultures 4 and 6 (hatched column ) were re-fed medium containing doxycycline (2 Ag/mL).Three days later, the levels of hPDGFBB were examined in triplicate in conditioned medium from the untreated (dotted column ) and treated groups (hatched column ) by again using ELISA.D, regulation of hPDGFB RNA expression by doxycycline in vivo determined by qRT-PCR.RNA of brain and spinal cord tissues were obtained from two groups of three GFAP/tTA:TRE/hPDGFB transgenic mice at 2 mo of age that either never received doxycycline or received doxycycline during gestation and their entire lifetime.hPDGFB RNA levels were determined and the ratio of the DCT of each specimen was compared with the DCT obtained from the brain sample of the only animal receiving doxycycline that had a detectable level of PDGFBB in its brain and designated as the relative level of hPDGFB RNA.Relative levels of hPDGFB RNA (left ) and the average hPDGFB levels for individual brain and spinal cord samples; bars, SD.

Cancer Research
Cancer Res 2008; 68: (20).October 15, 2008 primers for qRT-PCR were sense primer (CATGACAAGACGGCACTGAAG-GAG) and antisense primer (ATAACCCTGCCCACACACTCTC).Samples were run in triplicate with a corresponding cyclophilin control.Northern blot analysis was performed as previously reported (22).A 32 P-labeled cDNA probe was synthesized by the Rediprime System (GE Healthcare) in which the 0.8 kb BamHI DNA fragment of hPDGFB described above was used as a template.Southern blot hybridization was performed as previously described (23).
Cell culture including the preparation of spheroid cultures from brain and tumor tissue and adherent cell cultures of spinal cordderived tissues.To prepare spheroid cultures, we used malignant tissue or brain tissue from E16.5 transgenic and control embryos.Dissected normal brain or tumor tissues were dissociated with trypsin followed by trituration and cultured in DMEM/F12 supplemented with B27 (Invitrogen), 20 ng/mL of epidermal growth factor, 100 IU/mL of penicillin, and 100 Ag/mL of streptomycin.Neurospheres were dissociated and passaged as 1:2 dilutions once a week.To prepare adherent cultures of malignant or benign CNS tissues, dissected tissue was dissociated with trypsin and trituration, plated into dishes coated with poly-D-lysine, and cultured in DMEM/10% fetal bovine serum with antibiotics.hPDGFBB levels were determined by ELISA (R&D Systems) in conditioned medium and cell lysates prepared in M-Per (Pierce) lysis buffer supplemented with 30 mmol/L of sodium fluoride, 1 mmol/L of sodium vanadate, 0.5 mmol/L of phenylmethylsulfonyl fluoride, 100 mmol/L of NaCl, 1 mmol/L of EDTA, and protease inhibitor cocktail tablets (Roche).Soft agar cloning was performed in plates containing 2 Â 10 5 cells in 0.3% agar with or without doxycycline (2 Ag/mL) for 2 weeks.Assays were evaluated in triplicate as previously described (24) colonies were counted by Quantity One software (Bio-Rad Laboratories).
Evaluation of motor function.Functional hindlimb strength was evaluated using a modified Basso mouse scale (BMS) as described in the literature (25,26).BMS is a nine-point scale in which lower numbers indicate increasing disability.BMS scores were categorized as evidence of early, mild paralysis (grade 1, BMS 9-7), moderate paralysis (grade 2, BMS 6-5), and severe paralysis (grade 3, BMS 4-0).
Histopathology and electron microscopy.For histologic study, tissues were dissected after perfusion fixation (4% paraformaldehyde/0.1 mol/L phosphate buffer; pH 7.4) and embedded in paraffin.Some spines were treated with decalcification solution before embedding.Tissue for electron microscopy was fixed in 4% glutaraldehyde in sodium cacodylate buffer overnight, and then postfixed in 1% osmium tetroxide, and embedded.Sections were collected on 200 mesh copper grids, stained in 5% uranyl acetate/1% lead citrate, visualized on a FEI Tecnai FEG transmission electron microscope, and captured on a 1,024 Â 1,024 megapixel CCD camera.

Results
Development and characterization of GFAP/tTA:TRE/ hPDGFB transgenic mice.To evaluate PDGF in the development of mouse spinal cord tumors, we developed an animal model of spinal tumorigenesis building on two important observations: aberrant PDGF CNS expression leads to the development of cerebral glioma (13,14,16) and enhanced expression of a transgene in the spinal cord could be facilitated by a regulatory element mediating spinal cord expression.We screened mice which expressed the tetracycline-controlled trans-activator (tTa) regulated by the GFAP promoter.After identifying a mouse that had enhanced expression in the mouse spinal cord, line 78 as described by Popko and colleagues (20), we crossed this mouse to a different transgenic mouse bearing a transgene that consisted of a TRE covalently linked to the cDNA encoding hPDGFB.The partial structure of the TRE/hPDGFB transgene we used is shown in Fig. 1A and is described in Materials and Methods.We used Southern blotting hybridization analysis to evaluate genomic DNA from this mouse that was restricted with HindIII, which would not be expected to cleave within the transgene, and EcoRI, which is recognized by a single site in the transgene (Supplemental Fig. S1).Our finding of a single band larger in size than the restriction fragment used to create this mouse suggests that there is a single integration site containing a single copy of the transgene.Mice bearing this transgene are described as TRE/hPDGFB mice.
We crossed GFAP/tTA and TRE/hPDGFB to produce GFAP/ tTA:TRE/hPDGFB transgenic animals.Cell lysates of neurosphere cultures established from transgenic GFAP/tTA:TRE/hPDGFB mice were examined by an ELISA assay for hPDGFBB (Fig. 1B) and found to express readily detectable hPDGFB.Lysates of neurospheres prepared from TRE/hPDGFB, GFAP/tTA, and wild-type mice gave very low signals in this assay for hPDGFB (Fig. 1B).To evaluate this further, we established a primary culture of spinal cord tissue from adult GFAP/tTA:TRE/hPDGFB animals and examined conditioned medium in these cultures for hPDGFBB by an ELISA assay.As shown in Fig. 1C, cell cultures that did not receive doxycycline had readily detectable levels of hPDGFBB when examined at 72 hours after plating.Importantly, the production of hPDGFB was suppressed by doxycycline added to the culture medium (Fig. 1C).To examine whether hPDGFB observed in these cultures reflected hPDGFB mRNA expression in the CNS of adult animals, we prepared RNA from GFAP/tTA:TRE/hPDGFB mice.As shown in Fig. 1D, hPDGFB mRNA was detected in spinal cords, and to a much lower level in brains, of adult GFAP/tTA:TRE/hPDGFB mice, and the in vivo expression of this transgene was sensitive to the treatment of these animals with doxycycline.Differential mRNA expression was also detected in spinal and brain tissues from mice with this genotype when examined by Northern blot analysis as shown in Supplemental Fig. S2.These findings provide strong evidence that the local levels of transgene expression may contribute to the location within the CNS in which tumors arise.
Phenotype of GFAP/tTA:TRE/hPDGFB mice.GFAP/tTA:TRE/ hPDGFB mice that did not receive doxycycline had an easily recognizable phenotype.The dominant phenotypic characteristic was the development of paralysis of their forelimbs or hind legs and apparent bladder paralysis evidenced by large bladders full of urine.These two characteristics occurred in 50% of animals within the first 3 months of life, but infrequently could be first observed in much older animals.
Histopathology of GFAP/tTA:TRE/hPDGFB mice.We sacrificed more than 40 GFAP/tTA:TRE/hPDGFB transgenic mice when they developed signs of bilateral paralysis and on necropsy determined that all mice showing paralysis had spinal tumors.In approximately 8% of these animals, cerebral tumors were observed as well.Spinal cord evaluation in these animals showed evidence of spinal enlargement and hemorrhagic foci that correlated with regions of glial neoplasia (Fig. 2A1).At low-power magnification, examination of the abnormal cord stained with H&E revealed the obliteration of normal, organized cytoarchitecture with primitive tumor cells diffusely infiltrating the spinal parenchyma.
Hemorrhage was frequently seen in such examinations and rare areas suggestive of early necrosis were identified.Higher-power microscopic examination of H&E staining revealed abnormal cells with indistinct cytoplasms; small, round-to-oval and hyperchromatic nuclei with inconspicuous nucleoli; and occasional mitotic figures (Fig. 2A2 and A3).Ependymal rosettes were not identified.
Immunohistochemical examination facilitated the identification of these neoplasms as high-grade gliomas.Tumor tissues were stained with antibodies that recognize Olig2, a marker of oligodendrocytes; GFAP, a marker of astrocytes; the PDGF receptor-a (PDGFRa); and NG2 (chondroitin sulfate proteoglycan 4), a marker of oligodendrocyte precursors (Fig. 2B).We detected diffuse immunoreactivity of tumor cell nuclei against Olig2, although GFAP immunostaining was more variable with some tumor cells displaying staining and others not.GFAP is a widely used marker of astrocytic differentiation in the examination of primary glial tumors and its reactivity in these tumor cells indicates an astrocytic component in these tumors.Olig2 can be expressed in tumors of both the astrocytic and oligodendroglial lineage but strongly suggests oligodendroglial differentiation (27).The expression of NG2, a marker of bipotential O2A progenitors (28,29), provides further evidence that these tumors are best described as mixed glioma with both astrocytic and oligodendroglial components.The WHO grading scheme recognizes these as oligoastrocytomas (WHO grade 2) and anaplastic oligoastrocytomas (WHO grade 3).Grade 4 glial neoplasms with mixed astrocytoma and oligodendroglioma features also exist and are designated by the WHO as glioblastoma multiforme (30,31).We conducted a complete pathologic analysis of 14 consecutive tumors and found among these 11 grade 3 oligoastrocytoma and 3 glioblastoma multiforme with evidence of both oligodendroglial and astrocytic differentiation (Supplemental Table S1).The expression of PDGFRa is consistent with a pathologic role for hPDGFB mitogenic signal being important for tumorigenesis (Fig. 2B).The proliferative index varied somewhat from animal to animal and within individual tumor areas but was estimated to be f13% in grade 3 tumors, although we could identify indices of 25% in the most proliferative areas (Fig. 2B).These findings indicated that the spinal tumors that develop in GFAP/tTA:TRE/ hPDGFB mice were similar to spinal gliomas seen in humans, which can range from low-grade to high-grade and can exhibit evidence of either oligodendroglial or astrocytic features or both.The pattern of marker expression observed suggests that they are most similar to mixed gliomas seen clinically.Ultrastructural analysis of these spinal cord tumors revealed collections of primitive tumor cells with variable ultrastructural features (Fig. 2C1 and C2).In many areas, the tumor cells were small with minimal cytoplasm.Commonly, they were minimally cohesive within an edematous background filled with small processes and few intercellular junctions and had round to elongate nuclei with variable nuclear contours and chromatin patterns.In some collections of tumor cells, the cytoplasm showed intermediate filaments consistent with GFAP filaments found in astrocytic neoplasms.However, many tumor cells did not have filaments and displayed primitive, nonspecific ultrastructural features.Importantly, cell lines established from three different GFAP/tTA:TRE/ hPDGFB mice were cloned in soft agar, and that cloning ability is sensitive to treatment with doxycycline (Supplemental Fig. S3).
Effect of hPDGFB expression on tumor initiation.We found that the development of spinal tumors was a highly penetrant characteristic of GFAP/tTA:TRE/hPDGFB animals.Although some animals died of unknown cause(s) in the first weeks of life, of the animals that were genotyped at 2 weeks of life, the onset of characteristic signs of disease occurred at a median age of 11 weeks in animals that were never exposed to doxycycline (Fig. 3A).When animals with forelimb or hindlimb paralysis were sacrificed, highly invasive spinal tumors were invariably found.If mothers of GFAP/ tTA:TRE/hPDGFB animals received drinking water containing doxycycline only throughout pregnancy and during weaning (approximately 21 days of age), but were then fed water without doxycycline, the onset of paralysis, our surrogate marker for tumorigenesis, was greatly delayed (median onset, 22 weeks; Fig. 3A).Animals in this group invariably were also found to have spinal tumors when examined after the onset of symptoms.If the mothers of GFAP/tTA:TRE/hPDGFB animals were maintained on drinking water containing doxycycline throughout pregnancy and these offspring were maintained throughout their life on water containing doxycycline, we never observed either the development of neurologic symptoms or tumors (Fig. 3A).These findings suggest that the expression of hPDGFB is regulated in vivo by doxycycline in these GFAP/tTA:TRE/hPDGFB transgenic mice.To evaluate this further, we established spheroidal and monolayer cultures from tumors arising in untreated adult GFAP/tTA:TRE/hPDGFB animals and examined hPDGFBB by an ELISA assay.As shown in Fig. 3B, we detected high levels of hPDGFBB in those tumor cells, and importantly, the production of hPDGFBB was suppressed by doxycycline added to the culture medium.Development and characterization of mice overexpressing hPDGFB regulated by the human GFAP promoter.Our examination of GFAP/tTA:TRE/hPDGFB transgenic mice indicated that expression of hPDGFB in vivo greatly enhances the predisposition of mice to develop spinal cord mixed oligoastrocytoma.We sought to examine this observation further by developing a transgenic animal in which PDGFB was constitutively expressed from the GFAP promoter.Such a mouse would provide strong, independent evidence of the importance of PDGFB in the pathogenesis of spinal tumors and facilitate the study of spinal glioma.We therefore prepared the molecular construct of GFAP/ hPDGFB shown in Fig. 4A and developed a line of GFAP/hPDGFB transgenic animals, C8, as described in Materials and Methods.We used Southern blotting hybridization analysis to evaluate genomic DNA from this mouse that was restricted with HindIII, which would not be expected to cleave within the transgene, and XhoI, which is recognized by a single site in the transgene (Supplemental Fig. S1).Our finding of a single intense band equal in size to the EcoRI fragment used to create this mouse suggests that there are multiple copies of tandemly integrated transgene.
Transgenic GFAP/hPDGFB mice from the C8 line showed distinctive phenotypes.Although they appeared normal at birth when compared with wild-type littermates, they had a slightly smaller body size (data not shown).An unsteadiness of the hindlimbs became apparent at f2 to 3 months of age in many transgenic animals.We monitored 123 GFAP/hPDGFB transgenic mice throughout their lifetime, as a means to show the time course of progressive symptom development and described these findings for all 123 transgenic mice in Fig. 4B.Although typically born without evidence of paralysis, C8 GFAP/hPDGFB mice developed slow, progressive deterioration of motor function in their hind legs as they aged.Typically, the onset of hindlimb paralysis occurred at f2 months of age, but some animals clearly developed such symptoms earlier and rarely, considerably later.Generally, the unsteadiness progressed more rapidly in females than in male animals.The bony structure and musculature of the hindlimbs in these animals appeared normal.As animals progressed, they typically developed urinary incontinence with large bladders that leaked urine continuously.In the most severe cases, hindlimbs become completely paralyzed before the 5th month of life.When obvious symptoms were observed, animals were sacrificed.Forelimb motor function was typically spared, although animals did sometimes develop forelimb unsteadiness as well as occasional hydrocephalus and seizures.Invariably, animals that developed paralysis had autopsy evidence of spinal neoplasms (see below).When we used the development of grade 3 symptomatology as a surrogate for tumorigenesis, we found that this phenotype was highly penetrant with almost 65% of the animals developing evidence of a spinal tumor over the course of the 300 days we observed them.The median time of tumor development was f24 weeks (Fig. 4C).
In tumor models of primary CNS malignancies, p53 inactivation has been described as increasing penetrance and decreasing tumor latency (32).To seek evidence for a possible contribution of p53 inactivation to spinal cord tumorigenesis and to provide additional evidence that the paralytic symptoms we observed could be attributed to tumor development and that paralysis was a reasonable surrogate for tumor development, we crossed C8 GFAP/hPDGFB mice with mice lacking p53 (33).We monitored the development of paralysis in GFAP/hPDGFB:TP53 +/À mice and compared the time course of symptom development to that observed in wild-type and GFAP/hPDGFB transgenic C8 mice.The apparent median time to the development of severe symptoms was f18 weeks, and we found that these animals invariably had a spinal tumor.The time course of grade 3 symptom onset was plotted on a Kaplan-Meier curve (Fig. 4C).A statistically significant difference in symptom onset was found between these groups (P < 0.01).Importantly, animals that developed paralysis and were subsequently autopsied had intramedullary astroglial spinal cord tumors, although we never found such a tumor in an animal without symptoms.Although rare cerebral tumors were discovered (<5% of animals), we could not ascertain either a different A, Kaplan Meier survival analysis (limb paralysis was taken as a surrogate for death) of GFAP/tTA:TRE/hPDGFB mice.Three groups of transgenic mice with or without doxycycline in their drinking water were monitored: 15 consecutive offspring of the GFAP/tTA:TRE/hPDGFB transgenic genotype whose parents received doxycycline (1 mg/mL) at the time of mating and continued to receive doxycycline in their drinking water throughout life were monitored postnatally for the development of symptoms (dotted line, n = 15).In 17 other GFAP/ tTA:TRE/hPDGFB transgenic animals whose parents were similarly treated, doxycycline was terminated after weaning (dashed line, n = 17).These animals had a delayed onset of symptoms, but eventually all died of their tumors.Additionally, 45 consecutive transgenic mice that did not receive doxycycline and whose parents were never given doxycycline were also evaluated for the development of paralysis (solid line, n = 45).A significant difference was easily observed between the group that continuously received doxycycline in the drinking water throughout life from the time of conception and animals that never received doxycycline (P < 0.001).Animals that received doxycycline until weaning also had a significant different time course of tumor development compared with animals that never received doxycycline (P < 0.01).B, in vitro regulation of hPDGFB expression in tumor cells by doxycycline.Tumor cells obtained from spinal tumors arising in three different GFAP/tTA:TRE/hPDGFB transgenic mice were grown as monolayer or spheroid cultures in the presence (gray columns ) or absence (black columns ) of doxycycline (2 Ag/mL).At day 5 of culture, 30 Ag of cell lysate from each culture was examined in triplicate for hPDGFBB expression by ELISA assay.Columns, mean of triplicate cultures; bars, SD. spectrum of tumor locations nor variability in the histopathologic features of comparable grade tumors developing in GFAP/ hPDGFB/TP53 +/À and GFAP/hPDGFB animals.
Histopathology of GFAP/hPDGFB mice.Spinal cord tissue was examined grossly and histologically in more than 30 animals from which grade 3 symptomatology developed.Although the brains of C8 GFAP/hPDGFB animals occasionally appeared enlarged, and upon examination had evidence of hydrocephalus, only very rarely did they harbor an infiltrating glial proliferation (data not shown); however, all spinal cords examined displayed a significantly altered gross appearance histologically recognizable as an infiltrative glioma.The most common gross pathologic findings were translucent cord tissue, distortion of the normal cord shape, and hemorrhage (Fig. 5A1).Histologic examination of grossly pathologic spinal cord tissues revealed disruption of the normal spinal architecture by extraparenchymal infiltrates and infiltrative small cells with indistinct cell borders and fibrillary processes within the parenchymal tissues at the sites of pallor and distortion (Fig. 5A2).Dorsal and ventral entry route zones were often found to be involved (data not shown).Occasionally, collections of proliferating cells were found in subarachnoid regions, but these did not extend into peripheral nerves or dorsal root ganglia.Microcyst formation was also often observed.In some cases, the tumor cell collections had a recognizable border with the parenchyma, whereas in most cases, there was a more infiltrative pattern.Histologic examination (Fig. 5B) of this tissue revealed nuclei from proliferating cells that were mildly pleomorphic with small, round to oval contours.No rosettes characteristic of ependymoma were identified.Immunohistochemical analysis with antibodies against OLIG2, GFAP, and NG2 revealed expression of markers, which typically characterize either oligodendroglial or astrocytic differentiation (Fig. 5B).Cytoplasmic immunostaining with antibodies against S-100h consistently stained tumor cells, although nuclear staining against S-100h was less consistent (data not shown).The expression of the PDGFRa is consistent with a pathologic role for hPDGFB mitogenic signal being important for tumorigenesis (Fig. 5B).Ultrastructural evaluation of a spinal tumor from C8 GFAP/ hPDGFB revealed collections of primitive polymorphic cells within A, diagram of a GFAP/hPDGFB transgene construct.The 0.8 kb Bam HI fragment of the complete hPDGFB cDNA (gray box ) was inserted at the 3 ¶-end of a 2.2 kb genomic fragment containing the human GFAP promoter (hgfa2 ) in the vector hgfa2-0 (21).A segment of the mouse protamine-1 gene (mP1 ) that provides an intron and a polyadenylation signal is shown (black diagonally hatched box ).The Eco RI DNA fragment of this GFAP/hPDGFB plasmid was used for microinjection.B, GFAP/hPDGFB mice developed slow, progressive deterioration of motor function in their hind legs as they aged.Horizontal bars, neurologic grading of motor function observed in individual animals as described in Materials and Methods.Normal or slightly impaired, grade 1, motor function (dark grey ); grade 2 (light grey) and grade 3 symptoms (black ).C, effect of p53 deletion on neurologic deterioration in C8 mice.The time of onset of grade 3 toxicity in C8 GFAP/hPDGFB mice with different genetic backgrounds is plotted on a Kaplan-Meier curve.C8 GFAP/hPDGFB transgenic mice that are wild-type (n = 123) or heterozygous for p53 deletion (n = 14) and nontransgenic wild-type (n = 40).Statistically significant differences were found between each group (P < 0.01).
an edematous or cystic background (Fig. 5C).Other ultrastructural features included multiple interdigitating processes and very few cell junctions.Nuclear shape varied from elongated to round with irregular contours and chromatin patterns.These features are indistinguishable from those seen in tumors from the GFAP/ tTA:TRE/hPDGFB mice (Fig. 2C).Among 19 consecutive tumors that we pathologically examined in detail and graded according to the WHO criteria, 8 were best described as grade 2 oligoastrocytoma, 10 as grade 3, and 1 as glioblastoma multiforme indicating that the tumors arising in these animals include more of lower grade than those arising in the GFAP/tTA:TRE/hPDGFB mice (Supplemental Table S1).The average proliferative index of these grade 3 oligoastrocytomas was approximately 10, quite similar to the index observed for the grade 3 tumors arising in GFAP/ tTA:TRE/hPDGFB mice (Supplemental Table S1).These findings are consistent with our ability to derive cell lines from the spinal tumors arising in GFAP/hPDGFB animals, and the efficient growth of these cells in soft agar (data not shown).
Expression of hPDGFB in tissue from C8 GFAP/hPDGFB transgenic mice.To characterize hPDGFB expression in C8 GFAP/ hPDGFB transgenic mice, we produced neurosphere cultures and examined them for hPDGFBB expression using an ELISA assay.We found that neurospheres from E16 transgene-positive embryos expressed high levels of hPDGFBB (Fig. 6A).We also examined CNS tissues of such animals and identified high levels of mRNA encoding hPDGFB in this C8 line, but not in other transgenic lines, D3 and E5 (Fig. 6B).To visualize mRNA in brain required a long exposure of the Northern blot of brain tissue shown in Fig. 6B.Indeed, expression of hPDGFB was either not detectable or only barely detectable by RT-PCR in tissues from any transgenic line other than C8 (Fig. 6B; data not shown), and these lines showed no recognizable phenotype (data not shown).To further quantify the levels of transgene expression in the CNS of these animals, we examined RNA from the brain and spinal cord of individual GFAP/ hPDGFB mice (Fig. 6C).We used RT-PCR analysis and determined that in four consecutive asymptomatic animals without detectable tumors, the level of transgene expression in the spinal cord averaged approximately 11-fold higher than the levels of transgene expression in the brain of the same animals.

Discussion
To pursue the role of PDGF signaling in the pathogenesis of glial tumors arising in the spinal cord, we developed transgenic mice aberrantly expressing hPDGFB in the spinal cord.Others have explored the in vivo activity of PDGF ligands in transgenic animals, targeting the transgene to other tissues (34)(35)(36).Two independent models were developed.In the first model, we used a tetracyclineresponsive expression system to express hPDGFB from the GFAP promoter in a manner that could be repressed by placing doxycycline in the drinking water (GFAP/tTA:TRE/hPDGFB mice).This model provided evidence of the importance of PDGFB in the initiation of glial tumors arising in the spine.In a second model (GFAP/hPDGFB mice), we used the human GFAP promoter to drive hPDGFB mRNA expression and selected a mouse for further evaluation that developed spinal tumors.Although tumors from GFAP/tTA:TRE/hPDGFB transgenic mice without suppression of PDGFB by doxycycline had more obvious oligodendroglial characteristics and tumors from GFAP/hPDGFB mice had somewhat more dominant astrocytic features, these tumors share an indistinguishable spectrum of histologic characteristics.This tumor spectrum closely resembles tumor types previously identified in mouse models based on the injection of recombinant retroviruses which express PDGFB (13,14,16).We also observed that whereas the onset of tumorigenesis in the two models is clearly very early in life, the progressive growth of tumors arising in the GFAP/hPDGFB mice might be slower than those arising in the GFAP/tTA:TRE/ hPDGFB transgenic mice.
We used the same human 2.2 kb GFAP promoter in both models.Other laboratories have characterized the specificity of this GFAP promoter in transgenic mice driving several reporter genes (21,37,38).These studies identified differences in both the temporal and tissue specificity of transgene expression, presumably due to the variable integration sites.Here, we report a line of GFAP/hPDGFB mice that primarily developed spinal CNS malignancies.Also, both transgenic models we have presented had retinal alterations (data not shown), a finding consistent with the work of others (36,39).The pattern of tumorigenesis, and our characterization of high levels of hPDGF expression in spinal tissue, suggest that differential transgene expression underlies the phenotypes we observed.
The cell of origin of high-grade glial tumors is not known (40), and there is an ever increasingly greater sensitivity to the difficulty of understanding the lineage in which such tumors arise or the normal cell lineage which they mimic (41).The convergence of these characteristics in higher-grade glial tumors is reflected in the fact that both WHO grade 4 oligodendroglioma and WHO grade 4 astrocytomas are designated as glioblastoma multiforme (30,31).Similarly, efforts to identify definitive molecular signatures using genomic techniques such as expression analysis by microarrays (42) to distinguish between tumors of the oligodendroglial and astrocytic lineage have been inconclusive.Although deletions of 1p and 19q are more common in oligodendroglial tumors (43), and alterations such as PTEN deletion and epidermal growth factor receptor overexpression are more common in high-grade astrocytoma or GBM, these genetic alterations are not exclusive to these lineages (44)(45)(46).Also, lineage-specific markers characterized in normal tissues are not expressed in as restrictive a manner in glioma.For example, whereas OLIG1 and OLIG2, important markers of the oligodendroglial lineage, are not expressed in normal, adult, mature astrocytes (47), OLIG1 and OLIG2 expression is detectable in diffuse astrocytomas including most glioblastoma (27,48,49).Also, the oligodendroglial progenitor marker NG2 and oligodendrocyte differentiation factor, Sox10, are expressed in astrocytoma (28,50).Clearly, our findings and the findings of others (13,14,16) of mixed lineage tumors in models of CNS glioma Figure 6.In vitro expression of hPDGFB in C8 GFAP/hPDGFB mice.A, individual spheroid cultures were established from each of six E16 C8 GFAP/hPDGFB embryos (lanes 1-6) and three wild-type littermates (lanes 7-9 ) as described in Materials and Methods, and hPDGFBB expression was evaluated by ELISA assays of cell lysates (30 Ag).Each lysate was evaluated in triplicate and expression was normalized to the average value of hPDGFBB expression in the three wild-type neurosphere cultures.Columns, averages of the three cultures; bars, SD.This is indicated as the relative ratio of PDGFBB.B, hPDGFB mRNA expression in the CNS of C8 GFAP/hPDGFB transgenic mice.Northern blot analysis using an hPDGFB cDNA probe detected the 1.2 kb transcript produced from the GFAP/hPDGFB transgene in the brain of a representative transgenic C8 mouse (black arrow ), but not in the brains from other transgenic mouse lines (D3, E5) or a wild-type mouse (WT ).A transcript of the same size was detected in the spinal cord of the C8 mouse (white arrow ), but not in the spinal cord of a wild-type mouse (WT ) or a human glioma cell line (C.L. ).These same membranes stained with methylene blue are shown below each Northern blot.C, in vivo expression of hPDGFB RNA in GFAP/hPDGFB mice determined by qRT-PCR.RNAs from brain and spinal cord of four 2-month-old asymptomatic GFAP/hPDGFB mice without evidence of tumor were evaluated.hPDGFB RNA levels were determined and the ratio of the DCT of each specimen was compared with the average DCT of the four brain specimens.The levels of hPDGFB RNA are shown as the average of the four specimens examined (right ).*, P < 0.01, statistical significance was evaluated using a Student's t test.
are consistent with the gene expression patterns observed in clinical specimens of high-grade glioma.
The use of transgenic mice to model the role of PDGFB in gliomagenesis offers distinct advantages over strategies reported to date.Retroviral systems are compromised by rare insertional mutations (17).Also, the delivery of recombinant viruses is difficult in locations such as the spinal cord.Successful viral transgene strategies lead to the acute expression of very high levels of PDGFB, whereas the presumed chronic expression of lower levels of the transgene in the models we described seems more likely to mimic the pathogenesis of human tumors.The availability of a tetracycline-regulatable model opens the possibility for studies regarding the precise role of PDGFB in established tumors and may facilitate the examination of inhibitors of PDGF signaling for the future therapeutic management of these tumors.

Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.

Figure 1 .
Figure 1.Development and characterization of GFAP/tTA:TRE/hPDGFB mice.A, diagram of the TRE/hPDGFB transgene DNA used for injection.A 0.8 kb Bam HI DNA fragment containing the complete hPDGFB coding sequence (gray box ) was inserted at the 3 ¶-end of the TRE in plasmid pUHD10-3 to create a TRE/ hPDGFB DNA insert.The SV40 poly(A) signal sequence (black, diagonally hatched box ) is at the 3 ¶-end of the construct.The Xho I and Hin dIII DNA fragmentshown was used for microinjection.B, ELISA assay for hPDGFBB.Individual neurosphere cultures were established from each of six embryonic brains from E16 embryos obtained from a TRE/hPDGFB +/À female bred with a male GFAP/tTA +/À in the absence of tetracycline.Cell lysate (30 Ag) from each of these six spheroid cultures was prepared in triplicate and examined by an ELISA assay to measure hPDGFBB.Expression in lysates of neurosphere cultures from GFAP/tTA;TRE/ hPDGFB embryos (tTA-TRE ), a TRE/hPDGFB embryo (TRE ), a GFAP/tTA embryo (tTA ), and a wild-type (WT ) littermate.Columns, ELISA data points normalized to the average value obtained from the cultures of the wild-type embryo (WT ) and the average normalized value (ratio) of the six cultures for each embryo; bars, SD.C, regulation of hPDGFB expression by doxycycline in vitro.Normal untreated GFAP/tTA:TRE/PDGFB mice were sacrificed at 2 months of age and spinal cords were dissected.These tissues were dissociated, and the cells were seeded into six cultures.Initially, all cells were grown to confluence, and these cultures were randomly assigned into three groups.At that time, day 0, conditioned medium from each of the cultures was examined in triplicate by ELISA assay for hPDGFBB expression and the levels of expression in these cultures from all groups were found to be indistinguishable (black column ).Immediately after the conditioned medium was collected on day 0, culture 2 (dotted column ) was re-fed with medium and cultures 4 and 6 (hatched column ) were re-fed medium containing doxycycline (2 Ag/mL).Three days later, the levels of hPDGFBB were examined in triplicate in conditioned medium from the untreated (dotted column ) and treated groups (hatched column ) by again using ELISA.D, regulation of hPDGFB RNA expression by doxycycline in vivo determined by qRT-PCR.RNA of brain and spinal cord tissues were obtained from two groups of three GFAP/tTA:TRE/hPDGFB transgenic mice at 2 mo of age that either never received doxycycline or received doxycycline during gestation and their entire lifetime.hPDGFB RNA levels were determined and the ratio of the DCT of each specimen was compared with the DCT obtained from the brain sample of the only animal receiving doxycycline that had a detectable level of PDGFBB in its brain and designated as the relative level of hPDGFB RNA.Relative levels of hPDGFB RNA (left ) and the average hPDGFB levels for individual brain and spinal cord samples; bars, SD.

Figure 3 .
Figure 3.In vivo and in vitro regulation of PDGFB expression by doxycycline.A, Kaplan Meier survival analysis (limb paralysis was taken as a surrogate for death) of GFAP/tTA:TRE/hPDGFB mice.Three groups of transgenic mice with or without doxycycline in their drinking water were monitored: 15 consecutive offspring of the GFAP/tTA:TRE/hPDGFB transgenic genotype whose parents received doxycycline (1 mg/mL) at the time of mating and continued to receive doxycycline in their drinking water throughout life were monitored postnatally for the development of symptoms (dotted line, n = 15).In 17 other GFAP/ tTA:TRE/hPDGFB transgenic animals whose parents were similarly treated, doxycycline was terminated after weaning (dashed line, n = 17).These animals had a delayed onset of symptoms, but eventually all died of their tumors.Additionally, 45 consecutive transgenic mice that did not receive doxycycline and whose parents were never given doxycycline were also evaluated for the development of paralysis (solid line, n = 45).A significant difference was easily observed between the group that continuously received doxycycline in the drinking water throughout life from the time of conception and animals that never received doxycycline (P < 0.001).Animals that received doxycycline until weaning also had a significant different time course of tumor development compared with animals that never received doxycycline (P < 0.01).B, in vitro regulation of hPDGFB expression in tumor cells by doxycycline.Tumor cells obtained from spinal tumors arising in three different GFAP/tTA:TRE/hPDGFB transgenic mice were grown as monolayer or spheroid cultures in the presence (gray columns ) or absence (black columns ) of doxycycline (2 Ag/mL).At day 5 of culture, 30 Ag of cell lysate from each culture was examined in triplicate for hPDGFBB expression by ELISA assay.Columns, mean of triplicate cultures; bars, SD.

Figure 4 .
Figure 4. Development and characterization of GFAP/hPDGFB mice.A, diagram of a GFAP/hPDGFB transgene construct.The 0.8 kb Bam HI fragment of the complete hPDGFB cDNA (gray box ) was inserted at the 3 ¶-end of a 2.2 kb genomic fragment containing the human GFAP promoter (hgfa2 ) in the vector hgfa2-0(21).A segment of the mouse protamine-1 gene (mP1 ) that provides an intron and a polyadenylation signal is shown (black diagonally hatched box ).The Eco RI DNA fragment of this GFAP/hPDGFB plasmid was used for microinjection.B, GFAP/hPDGFB mice developed slow, progressive deterioration of motor function in their hind legs as they aged.Horizontal bars, neurologic grading of motor function observed in individual animals as described in Materials and Methods.Normal or slightly impaired, grade 1, motor function (dark grey ); grade 2 (light grey) and grade 3 symptoms (black ).C, effect of p53 deletion on neurologic deterioration in C8 mice.The time of onset of grade 3 toxicity in C8 GFAP/hPDGFB mice with different genetic backgrounds is plotted on a Kaplan-Meier curve.C8 GFAP/hPDGFB transgenic mice that are wild-type (n = 123) or heterozygous for p53 deletion (n = 14) and nontransgenic wild-type (n = 40).Statistically significant differences were found between each group (P < 0.01).