CANCER STEM CELLS

Gliosarcoma Stem Cells Undergo Glial and Mesenchymal

Differentiation In Vivo

ANA C. DECARVALHO,aKEVIN NELSON,

aNANCY LEMKE,

aNORMAN L. LEHMAN,

bALI S. ARBAB,

cSTEVEN KALKANIS,

a

TOM MIKKELSENa,d

aDepartments of Neurosurgery; bPathology; cRadiology; and dNeurology, Henry Ford Hospital, Detroit, Michigan,

USA

Key Words. Gliosarcoma • Glioblastoma • Neurospheres • Cancer stem cells • Mesenchymal differentiation

ABSTRACT

Cancer stem cells (CSCs) are characterized by their self-renewing potential and by their ability to differentiateand phenocopy the original tumor in orthotopic xeno-grafts. Long-term propagation of glioblastoma (GBM)cells in serum-containing medium results in loss of theCSCs and outgrowth of cells genetically and biologicallydivergent from the parental tumors. In contrast, the useof a neurosphere assay, a serum-free culture for selection,and propagation of central nervous system-derived stemcells allows the selection of a subpopulation containingCSCs. Gliosarcoma (GS), a morphological variant com-prising approximately 2% of GBMs, present a biphasicgrowth pattern, composed of glial and metaplastic mesen-chymal components. To assess whether the neurosphereassay would allow the amplification of a subpopulation ofcells with ‘‘gliosarcoma stem cell’’ properties, capable ofpropagating both components of this malignancy, we have

generated neurospheres and serum cultures from primaryGS and GBM surgical specimens. Neurosphere culturesfrom GBM and GS samples expressed neural stem cellmarkers Sox2, Musashi1, and Nestin. In contrast to theGBM neurosphere lines, the GS neurospheres were nega-tive for the stem cell marker CD133. All neurosphere linesgenerated high-grade invasive orthotopic tumor xeno-grafts, with histological features strikingly similar to theparental tumors, demonstrating that these cultures indeedare enriched in CSCs. Remarkably, low-passage GS serumcultures retained the expression of stem cell markers, theability to form neurospheres, and tumorigenicity. The GSexperimental tumors phenocopied the parental tumor,exhibiting biphasic glial and mesenchymal components,constituting a clinically relevant model to investigate mes-enchymal differentiation in GBMs. STEM CELLS 2009; 28:181–190

Disclosure of potential conflicts of interest is found at the end of this article.

INTRODUCTION

Glioblastoma (GBM), classified as grade IV astrocytoma bythe World Health Organization (WHO), is the most aggressiveand frequent type of glial neoplasia. Glioblastomas are hetero-geneous tumors and are refractive to therapy. Recent globalgene expression analysis defined a subtype of glioblastomaassociated with a poor prognosis characterized by aberrantexpression of mesenchymal genes [1, 2]. Another example ofa prominent mesenchymal phenotype in a central nervous sys-tem tumor is found in a morphological variant of GBM,termed gliosarcoma (GS), comprising approximately 2% ofGBMs [3]. GS presents a biphasic growth pattern, composedof glial and sarcomatous components, typically identified byglial fibrillary acidic protein (GFAP) immunohistochemicallabeling and reticulin staining of the dense extracellular ma-trix, respectively [4]. Gliosarcoma occurs de novo, as a pri-mary tumor, or develops after radiotherapy of glioblastomas

[5, 6]. Gliosarcoma is indistinguishable from primary glio-blastoma in relation to median patient age, cerebral hemi-sphere location, standard treatment, and clinical outcomes [5,7–9]. The mesenchymal component of GSs can present differ-entiation along fibroblastic, osteogenic, chondrogenic, adipo-genic, smooth, and skeletal muscle lineages, according tomorphological appearance and expression of lineage-specificmarkers [4]. The sarcomatous component of gliosarcomasshow signs of malignant transformation, such as nuclearatypia, mitotic activity, and necrosis [4]. Importantly, identicalgenetic abnormalities have been identified in both glial andmesenchymal components, suggesting the occurrence of ametaplastic mesenchymal differentiation from a common ori-gin, during the progression of the astrocytic tumor [7, 10–13].The molecular mechanisms directing the mesenchymal differ-entiation in glioblastomas, whether in the mesenchymalsubgroup of GBMs or in the biphasic GS subtype, and thesignificance of this plasticity for tumor progression andgrowth are poorly understood.

Author contributions: A.C.dC.: conception and design, data collection, analysis and interpretation, manuscript writing, final approval ofmanuscript; K.N., A.S.A., and N.L.: collection, assembly and analysis of data; N.L.L.: data analysis and interpretation, manuscriptwriting, final approval of manuscript; S.K.: financial support, provision of patient material, final approval of manuscript; T.M.:conception, financial support, provision of patient material, final approval of manuscript.

Correspondence: Ana C. deCarvalho, Ph.D., Henry Ford Hospital, Department of Neurosurgery, Hermelin Brain Tumor Center, E&R3052, 2799 West Grand Boulevard, Detroit, Michigan 48202-2689, USA. Telephone: 313-916-3483; Fax: 313-916-9855; e-mail: ana@neuro.hfh.edu Received August 17, 2009; accepted for publication November 9, 2009; first published online in STEM CELLS EXPRESS

November 23, 2009. VC AlphaMed Press 1066-5099/2009/$30.00/0 doi: 10.1002/stem.264

STEM CELLS 2010;28:181–190 www.StemCells.com

The remarkable heterogeneity of GBMs is also observedin the range of differentiation states of the neoplastic cells.Along with other solid and hematopoietic malignancies, it hasbeen hypothesized that a subpopulation of GBM tumor cellspresent cancer stem cell (CSC) properties. These propertiesinclude the ability to self-renew and differentiate into cellscomposing the bulk of the tumor and are thus implicated insustaining tumor growth and in post-treatment recurrence[14]. Long-term propagation of glioma cells in serum-contain-ing medium results in loss of the CSC population and out-growth of cells genetically and biologically divergent fromtheir parental tumors [15]. In contrast, culturing dissociatedglioblastoma surgical specimens in a serum-free culture me-dium, developed for propagation of neural stem cell (NSC),as floating multicellular spheroids, known as neurospheres[16], allows for the selection and expansion of a subpopula-tion of cells enriched in CSCs carrying patient-specific molec-ular fingerprints [15, 17–19]. These CSC-enriched subpopula-tions are characterized by long-term self-renewal in vitro, bythe expression of NSC markers, such as Sox2, Nestin, andCD133, and by the ability to differentiate into cells that formthe bulk of the tumor, recapitulating the parental tumor inorthotopic xenografts [17–19]. Long-term culture in serum-con-taining medium leads to differentiation of CSC accompaniedby loss of NSC markers Sox2, CD133, and Nestin [15]. How-ever, the recovery of cells with CSC properties from long-termserum-cultured human [20] and rat [21] GBM cell lines, andalso from a rat gliosarcoma cell line [22], has been reported.Thus, neurosphere cultures from GBM surgical specimens as astable source of cells carrying patient-specific molecular finger-prints have become an asset for preclinical studies. Here weinvestigate how culture conditions for GBM and GS tumor cellsaffect the potential for recapitulating the mesenchymal differen-tiation of the founder tumor in orthotopic xenografts. Given theproper stimulus, adult mouse neural stem cells differentiatealong mesenchymal lines in vitro and in vivo [23]. Likewise,modulation of mesenchymal differentiation of GBM stem cellsby the microenvironment would be expected. A recent studyshows that GBM neurospheres underwent chondroblastic andosteoblastic differentiation in subcutaneous tumor xenografts,but not in orthotopic xenografts [24]. Here we show mesenchy-mal differentiation in immunocompromised rat brain by cul-tured GS cells, recapitulating the biphasic parental tumor phe-notype. Furthermore, early passage GS1 serum-cultured cellsretained the expression of NSC markers, the ability to formneurospheres when transferred to serum-free medium, and reca-pitulated the parental tumor in rats, in contrast to the propertiesobserved for the GBM serum-cultured cells.

MATERIALS AND METHODS

Primary Tumor Cell Culture

Resected brain tumors were collected at Henry Ford Hospital(Detroit, MI) with written consent from patients in accordancewith institutional guidelines and graded pathologically accordingto the WHO criteria. Tumors were processed immediately aftersurgery, essentially as described [17, 19]. Briefly, tumors werecut into 1-mm3 pieces, washed three times in Dulbecco’s modi-fied Eagle’s medium (DMEM)/F-12 serum-free medium (Invitro-gen Life Technologies, Carlsbad, CA, http://www.invitrogen.com)and dissociated in cell dissociation buffer (Cognate BioServices,Baltimore, MD, http://www.cognatebioservices.com), trituratedmechanically with fire-narrowed Pasteur pipettes, and filtratedthrough a 40-lm cell strainer (Becton, Dickinson and Company,Franklin Lakes, NJ, http://www.bd.com). Red blood cells wereremoved by density centrifugation in Lympholyte-M (Cedar Lane

Laboratories, Burlington, NC http://www.cedarlanelabs.com).Nucleated cells were washed several times in DMEM/F-12. Forregular monolayer cultures, cells were cultured in DMEM supple-mented with 10% fetal bovine serum (FBS) (HyClone, Logan, UT,http://www.hyclone.com), 50 U/ml penicillin G sodium, and 50 lg/ml streptomycin sulfate. For neurosphere cultures, cells were platedin T-25 flasks at a density of 2 � 104 cells per milliliter in neuro-sphere medium (NM): DMEM/F12 medium containing 2 mM L-glutamine and supplemented with N-2 (Gibco Invitrogen Cell Cul-ture, Gibco, Grand Island, NY, http://www.invitrogen.com), 0.05%bovine serum albumin (BSA), 25 lg/ml gentamicin, 50 U/ml peni-cillin G sodium, 50 lg/ml streptomycin sulfate, 20 ng/ml EGF,and 20 ng/ml bFGF (Peprotech, Rocky Hill, NJ, http://www.pepro-tech.com). After 1–3 weeks, multicellular floating neurospheresformed and were dissociated in Mg2þ-Ca2þ-free phosphate-buf-fered saline (PBS) and replated in NM medium for the formationof secondary spheroids. The glioblastoma neurospheres used forthis work were below passage 10.

Neurosphere Assay for Differentiated GlioblastomaCells and Three-Dimensional Cell Cultures

Neurosphere Assay for Differentiated GBM or GSMonolayer Cultures. Dissociated tumors were first cultured in10% FBS DMEM and passaged upon confluency. After 3–20 pas-sages, as indicated, monolayer cells were harvested by Trypsin(Invitrogen) treatment, washed several times in serum-freeDMEM/F-12 medium, and plated at a density of 2 � 104 cellsper milliliter in NM in regular tissue culture-treated flasks. Neu-rosphere formation was monitored for up to 4 weeks.

To evaluate the differentiation potential, GS1 neurosphereswere incubated in NM medium without EGF and bFGF for 1week in a regular tissue culture-treated flask. In parallel, neuro-spheres were dissociated and plated in 10% FBS DMEM in ultra-low attachment plates (Corning, Corning, NY, http://www.cor-ning.com), to preserve the 3D architecture, and incubated for 1week under standard conditions. After 1 week in regular NM,NM without growth factors or serum medium, multicellular sphe-roids were harvested, fixed in formalin for 10 minutes, and paraf-fin-embedded for immunohistochemistry (IHC).

Experimental Orthotopic Tumor

Following IACUC guidelines in an institutionally approved ani-mal use protocol, dissociated GBM neurosphere or serum-cul-tured cells were inoculated intracranially in nude rats (RNU/RNU). Animals were anesthetized and immobilized in a small-animal stereotactic device (David Kopf Instruments, Tujunga,CA, http://www.kopfinstruments.com). A Hamilton syringe wasused to inject between 4 � 104 and 4 � 105 cells, as indicated,through a hole 3 mm to the right of the bregma, at a depth of 2.5mm, at a rate of 0.5 lL/10 s. The surgical zone was flushed withsterile saline, the hole sealed with bone wax, and the skin overthe injection site sutured. Animals were monitored daily and sac-rificed between 9 and 14 weeks after implantation.

Histology and Immunohistochemistry

Sections of formalin-fixed, paraffin-embedded human glioma sur-gical samples, tumor xenografts, or multicellular spheroids weredeparaffinized with xylene and rehydrated through graded alcoholin PBS. Antigens were unmasked by 10 minutes of incubation inboiling citrate buffer. Primary antibodies used were as follows:mouse anti-human Nestin (Chemicon International/ Millipore,Billerica, MA, http://www.millipore.com), mouse anti-humanGFAP (Biocare Medical, Concord, CA, http://www.biocare.net),CD133 (Miltenyi Biotec, Auburn, CA, http://www.miltenyibiotec.com), mouse anti-human Vimentin (clone CM048) (Biocare Med-ical), and mouse anti-human smooth muscle actin (a-SMA) cloneHHF35 (DakoCytomation, Carpinteria, CA, http://www.dakocyto-mation.com). After primary antibody incubation, slides werewashed several times with PBS and incubated with appropriatebiotinylated secondary antibody (Vector Laboratories,

182 Gliosarcoma Model Using Cancer Stem Cells

Burlingame, CA, http://www.vectorlabs.com). After washes inPBS, sections were treated with streptavidin-peroxidase complexand incubated in diaminobenzidine tetrachloride (DAB), 3-amino-9-ethylcarbazole (AECþ), or blue substrate-chromogen solutions(DakoCytomation). Sections were counterstained with Mayer’shematoxylin or nuclear fast red and mounted using Vctamount(Vector Laboratories). For antigen specificity controls, preimmuneserum replaced the primary antibody. Reticulin was stained withan automated reticulin stain kit (DakoCytomation). Images of thelabeled sections were captured with an Olympus IX50 micro-scope, equipped with a SPOT Insight 4 camera.

RNA Preparation and Quantitative ReverseTranscription Polymerase Chain Reaction(qRT-PCR)

Total RNA was isolated from cultured cells using an RNeasyMiniprep kit (Qiagen, Valencia, CA, http://www1.qiagen.com).cDNA was prepared from 1 lg of DNAseI-treated RNA usingSuperscript III and oligo dT (Invitrogen Life Technologies). Rela-tive quantification of gene expression was performed by real-timePCR using SybrGreen and ABIPrism 7000 sequence detectionsystem (Applied BioSystems, Foster City, CA, http://www.appliedbiosystems.com), according to the manufacturer’s instruc-tions. PCR conditions were as follows: 10 minutes at 95�C fol-lowed by 40 cycles of 15 seconds at 95�C and 1 minute at 60�C.Dissociation curves and agarose gel electrophoresis were used toverify primer specificity. b-Actin was used as internal referenceand relative mRNA levels were quantified by the 2(�DDCt) method[25]. Primer sets used were as follows: CD105 sense 50-ACAAGTCTTGCAGAAACAGTCC-30, antisense 50-GACCTGGCTAGTGGTATATGTCA; Sox2 sense 50-TGGACAGTTACGCGCACAT-30, anti-sense 50-CGAGTAGGACATGCTGTAGGT30;Msi1 sense 50-TTCGGGTTTGTCACGTTTGAG-30, anti-sense 50-GGCCTGTATAACTCCGGCTG-30; nestin sense 50-ATCGCTCAGGTCCTGGAAGG-30, anti-sense 50-AAGCTGAGGGAAGTCTTGGAG-30; b-actin sense 50-CCGACAGGATGCAGAAGGAG-30, anti-sense 50-CATCTGCTGGAAGGTGGACA-30.

Protein Sample Preparation and Western BlotAnalysis

Tumor Tissue. Fresh tumor samples were snap frozen in liquidnitrogen. Frozen tissue was minced before being transferred tolysis buffer. The GS1 patient presented local progressive disease88 days after the first surgery, with a pathology essentially identi-cal to the primary tumor. Tissue from this recurrence (GS1-TR)was used for the Western blot analysis, because of limited avail-ability of the primary tumor.

Cell Lysates. Cell monolayers were washed three times withPBS and lysed with cell lysis buffer (150 mM NaCl, 1% NonidetP-40, 0.5% deoxycholate, 0.5% Triton X-100, 0.5 mM EDTA,0.1% SDS, 50 mM Tris-HCl, pH 7.6), containing protease andphosphatase inhibitor cocktails (Calbiochem, San Diego, http://www.emdbiosciences.com) supplemented with 17.4 lg/ml PMSF(Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com). Pro-teins in the lysates were quantified by a modified Bradford Pro-tein Assay (Bio-Rad, Hercules, CA, http://www.bio-rad.com),denatured in SDS-gel loading buffer and frozen at �80�C.

Immunoblot Analysis. Thirty micrograms of total protein fromcell lysates were separated by SDS-PAGE. Proteins were electro-phoretically transferred to nitrocellulose membranes (InvitrogenLife Technologies) and probed with the following antibodies:Sox2 (Chemicon), anti-human specific Nestin (Chemicon), anti-human specific GFAP (Biocare Medical), CD133 (Cell SignalingTechnology, Beverly, MA, http://www.cellsignal.com), and b-actin (Sigma-Aldrich), followed by secondary peroxidase conju-gated antibodies (Jackson Immunoresearch Laboratories, WestGrove, PA, http://www.jacksonimmuno.com). Protein bands were

visualized using SuperSignal West Pico Chemoluminescent Sub-strate (Pierce, Rockford, IL, http://www.piercenet.com).

In Vivo MRI Imaging Studies

Animals underwent in vivo magnetic resonance imaging (MRI) atdifferent time points after implantation of neurospheres or mono-layer GBM cells. MRI was obtained with a 3 Tesla clinical system(Signa Excite; GE Healthcare, http://www.gehealthcare.com)using 50 mm diameter � 108 mm RF rung length small-animalimaging coil (Litzcage small-animal imaging system; Doty Scien-tific Inc., Columbia, SC, http://www.dotynmr.com). Animals wereanesthetized, positioned using a triplanar FLASH sequence, andMR studies were performed using three-dimensional (3D) FIESTAsequence before and after gadolinium-based contrast enhance-ment. The following parameters were used to acquire MRI: 3D FI-ESTA-c MR images were obtained with TR ¼ 12 ms, TE ¼ 2.8ms, 40� flip angle, 60 � 50 FOV, 320 � 256 matrix, 32–40 sliceswith a slice thickness of 0.5 mm and NEX ¼ 4. To delineate themargin of the tumor properly, gadolinium contrast agent wasinjected intravenously and after contrast 3D FIESTA images wereacquired using the above-mentioned image parameters.

Molecular Characterization

TP53 Mutations. cDNA was obtained from total RNAextracted from tumor samples, neurosphere cultures, and xeno-graft tumor, as described above. The p53 region corresponding tonucleotides 678–1176 (exons 5–8) of the human TP53 mRNA(NM_000546) was amplified by PCR and sequenced (DNASequencing Facility, Wayne State University, Detroit, MI), usingDNA oligo sequences for amplification and sequencing describedelsewhere [26].

Loss of Heterozygosity. Allelic deletions in the phosphataseand tensin homologue (PTEN) region of chromosome 10 weredetected using the microsatellite markers D10S215, D10S574,D10S187, and D10S217, as described [27].

RESULTS

Plasticity of Gliosarcoma Serum-Cultured Cells

Serum-cultured glioma cells undergo a phenotypic alterationover time termed ‘‘mesenchymal drift’’, the tendency to loseglial and acquire mesenchymal characteristics [28, 29]. Cul-turing GBM surgical specimens in serum-free NM, however,allows the amplification of a cell subpopulation with neuralstem/progenitor cell characteristics, namely, long-term self-renewal ability, and the potential to differentiate and to reca-pitulate the original tumor in orthotopic xenografts [15].Because the GBM variant gliosarcoma presents a metaplasticmesenchymal differentiation morphologically distinct fromthe astrocytic component, we tested how culture conditionswould affect the amplification of cells with gliosarcoma stemcell characteristics, capable of recapitulating the biphasicnature of the tumor in vivo. Surgical GBM and GS tumorspecimens were cultured immediately after surgical resectionin NM supplemented with EGF and bFGF, or in medium sup-plemented with 10% FBS. First, we tested the recovery ofcells with progenitor/stem cell characteristics from early pas-sage (P3) serum cultures derived from 12 primary GBM and1 GS surgical specimens, using the neurosphere assay andmonitoring neurosphere formation for several weeks (asdetailed in Materials and Methods). Whether a cell subpopula-tion capable of dedifferentiating to a progenitor/stem cell phe-notype in vitro can be recovered from serum-cultured humanglioblastoma cells is an unsettled matter [15, 20]. Passage 3gliosarcoma (GS1) cells readily formed self-adherent multicel-lular spheroids when transferred from 10% FBS growth

deCarvalho, Nelson, Lemke et al. 183

www.StemCells.com

medium into serum-free NM (‘‘GS1 P3’’; Fig. 1A). The prop-erty of yielding neurospheres in vitro was lost after 15 pas-sages (‘‘GS1 P16’’; Fig. 1A). None of the serum-culturedGBM samples tested exhibited the neural stem cell growthphenotype in the neurosphere assay, even when kept in NMfor 4 weeks, as shown for GBM2 (Fig. 1A), in agreementwith previous reports [15]. The GS1 neurospheres were disso-ciated and replated for de novo neurosphere formation andpassaged for over 18 months (not shown), an indication oflong-term self-renewal potential. Because the property ofdedifferentiating into long-term self-renewing neurosphereswas exclusive of the low-passage GS1 FBS, the neural stemcell markers expression in this population was comparedagainst the higher GS1 FBS passages and GBM2 FBS-P3,which did not dedifferentiate in vitro (Fig. 1A). GBM2 andGBM3 neurospheres were obtained from the dissociated tumorbiopsies. All tumor neurospheres lines in this study were belowpassage 10, although self-renewal, tumorigenicity, and NSCgene expression in higher passages cells maintained in NMremained stable (not shown), in agreement with previous work[15]. The NSC markers Sox2, Musashi1 (Msi1), and NestinmRNA levels, determined by qRT-PCR, were downregulatedin GBM2 FBS-P3 cells in comparison to GBM2 neurospheres(Fig. 1B). In contrast, the mesenchymal marker endoglin(CD105), expressed in a subgroup of GBMs [30], is upregu-lated in GBM2 FBS-P3 in relation to the neurospheres (Fig.1B). These results are in agreement with previously reportedloss of stem cell marker expression in GBM cells cultured inthe presence of serum [15]. Conversely, GS1 FBS-P3 retainedthe expression of Sox2, Msi1, and Nestin (Fig. 1B), suggestingthat the ability to grow as neuroprogenitor/stem cells in NMdescribed above (Fig. 1A) is accompanied by the maintenanceof the expression of neural stem cell markers, typically down-regulated in differentiated serum-cultured GBM cells. Accord-ingly, the decreased developmental plasticity observed inhigher passages GS1 FBS cells (Fig. 1A) was accompanied bytranscriptional downregulation of the NSC markers Sox2,Msi1, and Nestin, whereas the mesenchymal marker CD105was upregulated (P16 and P20; Fig. 1B).

Sox2 protein was expressed in the GBM2 and GBM3tumors and neurosphere cultures and was not detected in thedifferentiated serum-cultured monolayer cells (Fig. 2A). How-ever, Sox2 protein expression was observed in GS1 FBS-P3cells, but downregulated in GS1 FBS-P20 (Fig. 2A), in agree-ment with the results above (Fig. 1B). The neural and hemato-poietic stem/progenitor cell marker CD133 has been used tosort a subpopulation of GBM cells with CSC properties [19].Recently, it has been shown that CD133 expression is not arequired property defining GBM CSC [31]. CD133 wasexpressed in the tumor specimens and neurosphere culturesfrom GBM2 and GBM3, whereas no expression was observedfor the differentiated cells (Fig. 2A). No CD133 expression wasdetected for the GS1 neurosphere or serum cultures (Fig. 2A). Ithas been reported that CD133 negative glioblastoma cells cangive rise to CD133 positive cells in orthotopic xenografts [32].The orthotopic xenografts generated by the CD133 negativeGS1 neurospheres did not contain CD133 positive cells (Fig.2B). As expected, CD133 positive GBM3 neurospheres gaverise to xenografts containing CD133 positive cells (Fig. 2B).Thus, CD133 expression in the tumor xenografts and neuro-spheres correlated with the CD133 status in the patient biopsy.

Glioblastoma and Gliosarcoma Neurospheres FormHigh-Grade Tumors in Rats

A defining property of CSCs is the ability to phenocopy theoriginal tumor in orthotopic xenografts. We tested whether

the in vivo tumorigenic potential of the GS1 neurospheres,obtained from dedifferentiation of passage 3 monolayer cells(Fig. 1A) would be comparable to the GBM2 and GBM3 neu-rospheres lines, derived directly from dissociated tumor biop-sies. Neurosphere cells were dissociated and implanted intothe brains of immunodeficient rats. Tumors were establishedin 100% of animals implanted with the three neurospherelines (Table 1). Animals were sacrificed 14 weeks after im-plantation and coronal brain sections were stained with eosinand hematoxylin (H&E). All three neurosphere lines initiated

Figure 1. Neural stem cell properties of serum GS and GBM cul-tures. (A): Neurosphere assay: low (P3) or higher (P16) passages GS1monolayer cells cultured in 10% FBS Dulbecco’s modified Eagle’smedium (‘‘day 0’’) were harvested and seeded in serum-free NM.Neurosphere formation was monitored for 4 weeks. Phase contrastimages shown for day 1 and day 14 are representative of three inde-pendent experiments. GBM2 P3 monolayer cells did not form neuro-spheres and this result is representative of observations made for 12GBM samples tested. (B): Expression of neural stem cell (Sox2,Msi1, and Nestin) and mesenchymal cell (CD105) markers in GBM2and GS1 serum cultures of the indicated passage were compared byqRT-PCR. Values are expressed as fold increase in relation to mRNAlevels in the corresponding neurosphere cultures. Results representmean 6 SE, for n ¼ 3 samples. Abbreviations: FBS, fetal bovine se-rum; GBM, glioblastoma; GS, gliosarcoma; NM, neurosphere me-dium; NP, neurosphere; qRT-PCR, quantitative reverse-transcriptionpolymerase chain reaction.

184 Gliosarcoma Model Using Cancer Stem Cells

high-grade invasive tumors (Fig. 3). Histological comparisonwith corresponding founder tumor biopsies revealed a remark-able similarity regarding cellular density and morphology(Fig. 3A, 3B, 3D, 3E, 3G, 3H). Pseudopalisading necrosiswas observed in the GS1 patient biopsy (Fig. 3A) and xeno-grafts (Fig. 3B). Clinical information is shown in Figure 3J.Genetic abnormalities frequently associated with GBM, suchas TP53 mutations and chromosome 10q losses, were eval-uated in the parental tumors, tumor neurospheres, and derivedxenografts. LOH on chromosome 10q was observed for allparental tumors, neurospheres, and xenografts (Fig. 3J). PTENprotein was not detected in any of the glioblastoma or gliosar-coma neurosphere lines (not shown). TP53 mutational statusin the tumor neurospheres and derived xenografts was identi-cal to the founder tumor (Fig. 3J). These results confirm theneoplastic origin of the neurosphere cultures and the preserva-tion of the patient tumor genetic abnormalities.

GS1 Neurospheres Undergo AstrocyticDifferentiation In Vitro

Neural stem cells cultured in medium supplemented withFBS, or upon withdrawal of growth factors from the neuro-

sphere medium, differentiate into GFAP-expressing astrocytes[16, 33]. To characterize the GS1 neurosphere differentiationpotential in vitro, neurospheres with 30- to 50-lm diameterwere divided into three groups in quadruplicates and culturedfor 1 week in NM, in NM without growth factors, or in 10%FBS. Serum-cultured cells were maintained in ultra-low-attachment plates to preserve the 3D architecture, which isknown to affect cell signaling and gene expression [34]. Sphe-roids were collected and the expression of markers for cellproliferation (Ki-67), stem/progenitor cells (Nestin), and astro-cytes (GFAP) were analyzed by immunocytochemistry (Fig.4A). Ki-67 and GFAP positive cells were counted in four400� images, and results are expressed as percentage of totalnuclei (Fig. 4A). GS1 neurospheres cultured in NM mediumwere highly proliferative and exhibited widespread expressionof Nestin and no expression of the astrocytic marker GFAP(Fig. 4A). The percentage of GFAP-expressing cells signifi-cantly increased and the percentage of proliferating cellsdecreased upon growth factor withdrawal or in the presenceof serum (Fig. 4A). The intermediate filament protein Nestinis expressed in neural stem and precursor cells [35], and alsoin a subpopulation of mesenchymal stem cells [36]. Wide-spread Nestin expression was observed in GS1 neurospheres,and expression was retained upon growth factor withdrawal,or in the serum-cultured spheroids, albeit at a lower intensity(Fig. 4A). In vivo expression pattern of Nestin, along with thecell proliferation marker Ki-67, was assessed by immunohis-tochemistry in xenograft tumors 9 weeks after GS1 neuro-sphere implantation, using human-specific antibodies. Ubiqui-tous expression of Nestin and Ki-67 was observed in thetumor core and in invasive cells (Fig. 4B). Few cells positivefor a-SMA and sporadic reticulin staining were observed onlyin the GS1 neurospheres cultured for 1 week under differen-tiation stimuli (supporting information Fig. 1), suggesting invitro mesenchymal differentiation potential. Vimentin expres-sion was observed for neurospheres growing in NM and underdifferentiation conditions (supporting information Fig. 1).

Serum-Cultured Gliosarcoma CellsAre Tumorigenic

It has been shown that patient-derived serum-cultured glio-blastoma cells are not tumorigenic at early passages [15].We compared the tumorigenic potential of neurosphere andserum monolayer cells to assess whether the expression ofNSC markers in the low-passage GS1 serum cultures, alongwith the ability to form neurospheres when transferred toNM (Fig. 1A), correlates with tumorigenicity in vivo. Thetake rate for nude rats implanted with 2–4 � 105 GS1,GBM2, and GBM3 neurosphere cells was 100% (Table 1).The take rate remained at 100% when the number of neuro-sphere cells implanted was reduced to 4 � 104 (Table 1).Low-passage GBM2 serum-cultured cells were not tumori-genic (Table 1), as expected from previous reports [15]. Incontrast, the 100% take rate was observed for GS1 early pas-sage serum-cultured cells, when 4 � 105 or 4 � 104 cells

Figure 2. Sox2 and CD133 protein expression in GS1 and GBMsamples. (A): Sox2 and CD133 protein expression in GBM2 andGBM3 tumors, GS1 tumor recurrence, derived NPs, and serum cul-tures (FBS) was assessed by immunoblot; b-actin was used as loadingcontrol. (B): CD133 expression (blue) in GBM3 and GS1 xenograftswas analyzed by immunohistochemistry of coronal brain sections;nuclei were counterstained with nuclear fast red. Scale, 20 lm.Abbreviations: GBM, glioblastoma; GS, gliosarcoma; NP, neuro-sphere; T, tumor; TR, tumor recurrence.

Table 1. Percentage of rats developing tumors 3 months after intracranial implant

Neurospheres Serum monolayers

2–4 � 105 cells 4 � 104 cells 4 � 105 cells 4 � 104 cells

GS1 100% (11 /11) 100% (4/4) 100% (8/8) 100% (4/4)GBM2 100% (4/4) 100% (4/4) 0% (0/4) NDGBM3 100% (3/3) 100% (6/6) ND* ND*

*Dissociated tumor cells did not grow well in serum-containing medium after passage 2.

deCarvalho, Nelson, Lemke et al. 185

www.StemCells.com

were implanted (Table 1). Histological comparison of repre-sentative xenografts derived from early passage serum (FBS-P3) or neurosphere (NP) GS1 cells are shown in Figure 5A.At 9 weeks after implantation, the neurosphere tumors aremore disperse diffusely along the white matter tract, fol-lowed by rapid growth, resulting in a high-grade tumor withnecrotic areas at 11 weeks (Fig. 5A). By comparison, theFBS-P3 tumors are larger and more circumscribed at 9weeks, but also grow into large necrotic tumors (Fig. 5A).Lee et al. observed that GBM cells cultured in serumundergo genomic rearrangements after passage 10, eventuallyregaining the ability to form tumor xenografts in some cases[15]. Such tumors are genetically and morphologically dis-similar to the original tumor, in contrast to xenografts gener-ated from neurospheres [15]. Rats implanted with GS1 FBS-P20 cells also developed tumors (Fig. 5A). To evaluate theGS1 CSCs long-term self-renewal in vivo, we compared thetumorigenicity of neurospheres derived from excised xeno-graft tumors (GS1-XP1) with the original GS1 neurospheres

cells. Nude rats were implanted with 4 � 104 GS1-XP1 orGS1 cells, four animals each, and sacrificed after 11 weeks.All animals developed tumors, with remarkable morphologi-cal similarities between the two groups (Fig. 5B). Tumorgrowth monitoring by MRI showed that rats implanted withGS1 FBS-P3 cells developed detectable tumors earlier,whereas a longer lag time followed by a rapid growth wasobserved for NP tumors (Fig. 5C), consistent with the histo-logical observations (Fig. 5A).

Gliosarcoma Phenotype in Xenografts Generated byNeurospheres and Low-Passage Serum Cultures

We then asked whether the GS1 cultured cells would retainthe gliosarcoma multilineage differentiation potential in vivo.The presence of glial and mesenchymal components in theGS1 xenografts was assessed by reticulin staining and IHCand compared with the parental tumor. Representativeimages of at least three samples per group for the GS1

Figure 3. Orthotopic xeno-grafts generated by GS1 andGBM neurospheres phenocopythe parental tumors. Tumorbiopsy samples and rat brainsimplanted with the respectiveneurosphere cultures were for-malin-fixed and paraffin-embed-ded, sectioned, and stained withH&E. Representative images areshown for GS1 biopsy (A) andcorresponding xenograft (B,C),where arrows point to pseudopa-lisading necrosis. Representativeimages are shown for primaryGBM samples GBM2 (D) andcorresponding xenografts (E,F),and GBM3 (G) and correspond-ing xenografts (H,I). ‘‘T’’ indi-cates tumor in the 10�magnification images (C,F,I).Scale, 80 lm (A,B,D,E,G,H).(J): Patient information and mo-lecular analysis of the tumorbiopsies. Abbreviations: GBM,glioblastoma; GS, gliosarcoma;T, tumor.

186 Gliosarcoma Model Using Cancer Stem Cells

biopsy (Fig. 6A–6E), GS1 NP xenografts (Fig. 6F, 6G), GS1FBS-P3 xenografts (Fig. 6H–6L), GS1 FBS-P20 xenografts(Fig. 6P–6T), and GBM3 NP xenografts (Fig. 6U–6Y), areshown. The astrocytic and the sarcomatous components ofgliosarcomas can be geographically distinct or intermingledand are typically identified by GFAP and reticulin staining,respectively [4]. Both the sarcomatous component andGFAP positive cells observed in the GS1 biopsy (Fig. 6A,6B) were present in the GS1 NP xenografts (Fig. 6F, 6G).The GS1 FBS-P3 xenografts presented extensive areas posi-tive for reticulin staining, and a GFAP expression patternsimilar to the patient biopsy (Fig. 6K, 6L). The GS1 paren-tal tumor and all GS1 xenografts were positive for the inter-mediate filament Vimentin, found in both mesenchymal andneural progenitor cells [33] (Fig. 6D, 6I, 6N, 6S). Nestinwas widely expressed in the GS1 NP xenografts (Figs. 6H,4B) and expressed in discrete regions of the GS1 FBS-P3xenografts (Fig. 6M), in a pattern similar to the GS1 patientbiopsy (Fig. 6C). Reticulin and GFAP weakly positiveregions were still observed in the GS1 FBS-P20 xenografts(Fig. 6P, 6Q). Vimentin expression was maintained in GS1FBS-P20 tumors (Fig. 6S), whereas Nestin was downregu-lated in the FBS-P20 xenografts (Fig. 6R), in agreementwith the downregulation of the Nestin transcript in vitro inGS1 FBS-P20 cells (Fig. 1B). Gliosarcomas frequently pres-ent gain of additional mesenchymal lineage markers. To fur-ther characterize the sarcomatous component of GS1 biopsyand xenografts, markers of muscle, epithelial, and neuroecto-dermal lineages were analyzed by immunohistochemistry.Expression of a-SMA was observed in the patient biopsyand in discrete regions of all GS1 xenografts (Fig. 6E, 6J,6O, 6T). Smooth muscle differentiation in gliosarcoma hasbeen previously reported [5, 37]. The GS1 specimen andxenografts were negative for the epithelial markers epithelialmembrane antigen (EMA) and cytokeratin and the neuroec-todermal marker synaptophysin (not shown). The GBM3

xenograft was positive for GFAP and Nestin (Fig. 6V, 6W),negative for the mesenchymal markers reticulin (Fig. 6U)and a-SMA (Fig. 6Y), and weakly positive for Vimentin(Fig. 6X). The lack of reticulin staining and a-SMA positiveneoplastic cells shown for GBM3 is representative of obser-vations for xenografts derived from four other primary GBMneurosphere lines, whereas variable levels of Vimentinexpression were observed for all GBM xenografts and biop-sies tested (not sown). These results demonstrate that GScells cultured in neurosphere medium can recapitulate notonly the astrocytic component of the tumor but also themesenchymal differentiation in vivo. The differentiated low-passage GS1 serum-cultured cells also retained the expres-sion of neural stem cell markers and phenocopied the paren-tal tumor in rats.

DISCUSSION

Cell culture medium containing FBS causes irreversible dif-ferentiation of NSCs [16]. Likewise, it has been shown thatserum-cultured primary glioma cells undergo changes that can-not be subsequently reversed following transition to neuro-sphere medium, for example, loss of NSC markers Sox2,CD133, and Nestin [15]. However, the recovery of cells withCSC properties from long-term serum-cultured human GBMcell lines [20] and rat gliosarcoma 9L cells has been reported[22]. Here we show that low-passage GS1 serum-cultured cellsretained the expression of NSC markers, the neural stem/pro-genitor cell property of growing as floating multicellular sphe-roids in NM, and the ability to recapitulate the parental tumorin experimental models. The latter are attributes defining CSCs.In contrast, no CSCs were recovered from the low-passageGBM serum cultures tested. It has been suggested that in vitroneurosphere formation correlates with clinical outcome for

Figure 4. Differentiation of GS1 neuro-spheres in vitro and ubiquitous expressionof Nestin in vitro and in vivo. (A): GS1neurospheres cultured in neurosphere me-dium, for 7 days in the absence of growthfactors, or for 7 days in 10% FBS mediumwere collected, fixed, and embedded in par-affin. Representative images of 6-lm-thicksections stained for the astrocytic markerGFAP, the proliferation marker Ki-67, andNestin are shown. Nuclei were counter-stained with hematoxylin. Ki-67 and GFAPpositive cells were counted in four 400�images and results are expressed as meanpercentage of positive cells þ SE. (B):Representative image showing Nestin andKi-67 expression in the GS1 tumor xeno-graft core and invasive cells, assessed byimmunohistochemistry of coronal sections9 weeks after GS1 neurospheres implant(magnification, 10� and 200�). Abbrevia-tions: C, core; FBS, fetal bovine serum;GF, growth factors; GFAP, glial fibrillaryacidic protein; GS, gliosarcoma; I, inva-sive; NM, neurosphere medium.

deCarvalho, Nelson, Lemke et al. 187

www.StemCells.com

gliomas [38]. The tumor neurospheres used for this study wereobtained from patients that presented a wide range of progres-sion-free survival, 88, 196, and 648 days for GS1, GBM2, andGBM3, respectively (Fig. 3J). Whether the poor clinical out-come for the GS1 case is associated with the increased plastic-ity observed in low-passage serum culture remains to be veri-fied by testing a larger number of samples.

The neural stem cell markers Sox2, Msi1, and Nestinare expressed in high-grade gliomas [39], and their expressionwas preserved in the GBM and GS neurosphere culturesreported here. CD133, a hematopoietic and neural stem cellmarker, is also expressed in high-grade gliomas [39] and hasbeen proposed as a marker for GBM CSC [40]. In contrast tothe GBM neurosphere lines, the GS1 was CD133 negative.Recently, it was shown that GBM CSCs negative for CD133are tumorigenic [31, 41]. Our results support the concept thatalthough frequently expressed in GBM neurospheres, CD133expression is not a requirement to define CSCs. Sox2, a mas-ter regulator of embryonic stem cells and neurogenesis [42],is upregulated in glioblastomas in comparison to nontumorbrain tissues [30, 43]. It also has roles in tumorigenesis andin the proliferation of glioblastoma CSC population [44]. Wespeculate that Sox2 expression in low-passage serum GS1

could participate in the regulation of the singular plasticityand tumorigenicity observed for this cell population.

The most frequent genetic alterations in gliosarcomas aregains on chromosome 7 and losses on chromosome 10,which include the PTEN locus [13]. Similarly, these geneticalterations also characterize GBMs belonging to the mesen-chymal subgroup, according to molecular profiling of glio-blastomas [1]. The GS1 tumor belongs to the mesenchymalsubgroup, characterized by poor prognosis (Ken Aldape, per-sonal communication). Whether all gliosarcomas belong tothe mesenchymal subgroup is currently unknown. In vitrotransdifferentiation potential of neural stem cells along mes-enchymal lineages has been observed [45]. The ability ofestablished serum-cultured glioblastoma cell lines to transdif-ferentiate along mesenchymal lineages in vitro [30], and the‘‘mesenchymal drift’’ observed as a result of long-termserum culture of glioblastomas [28, 29], have long beenknown and are generally considered a consequence ofgenetic aberrations acquired under these conditions [15].Recently, it has been reported that GBM neurospheres candifferentiate along mesenchymal lineages in vitro when theproper stimuli are present [24]. Glioblastomas with chondro-blastic and osteoblastic differentiation was observed in

Figure 5. Orthotopic tumorgrowth and morphology in nuderats implanted with GS1 cells.(A): Rats implanted with anequal number of GS1 cells fromNP cultures, low-passage serum(FBS-P3), and higher passageserum (FBS-P20) were sacrificedat 9 and 11 weeks, and tumorswere monitored by histology(H&E). (B): Similar tumorswere observed when 4 � 104

NP cells derived from surgicalspecimens (GS1) or from a GS1tumor xenograft (GS1-XP1)were implanted in nude rats,demonstrating long-term self-renewal of the GS1 cancer stemcells by serial transplantation.(C): Magnetic resonance imageswere obtained at the indicatedtime points for a rat implantedwith GS1 NPs and a ratimplanted with low-passage se-rum-cultured GS1 cells (FBS-P3). Axial plane images areshown; arrowhead indicates theinjection site for the 7-weeksNP rat, and arrows point to thetumor in the other images.Abbreviations: FBS, fetal bovineserum; GS, gliosarcoma; NP,neurosphere; T, tumor.

188 Gliosarcoma Model Using Cancer Stem Cells

subcutaneous tumor xenografts from two glioblastoma CSC/neu-rosphere lines, but not in orthotopic xenografts [24]. Here weshow that high-grade invasive experimental tumors were initi-ated by all glioblastoma neurosphere lines, as expected from apopulation containing CSCs. Furthermore, our results indicatethat GS1 neurospheres and low-passage serum cultures exhib-ited ‘‘gliosarcoma stem cell’’ properties, as metaplastic mesen-chymal differentiation in orthotopic tumors was observed specif-ically for the GS1 cells and not for the GBM neurospheres.

Analysis of additional GS samples will be necessary to clar-ify whether the in vitro phenotypic plasticity and in vivo reca-pitulation of the biphasic tumor observed for the GS1 model aregeneral characteristics of this glioblastoma variant. Mechanismsleading to mesenchymal differentiation in gliosarcomas, its path-ological significance, or whether the same mechanisms are re-sponsible for the mesenchymal phenotype observed for a sub-group of GBMs [1] are not currently understood. The humangliosarcoma model presented here will be useful in future inves-tigations addressing these important questions.

ACKNOWLEDGMENTS

The authors would like to thank Mr. Ali Rad for assistancewith the MRI images, Mr. Enoch Carlton for histology work,Ms. Kathleen Roszka for expert immunohistochemistry work,and Ms. Laura Hasselbach for LOH analysis. Drs. JacquesBaudier and Laurent Balenci provided helpful advice at theearlier stages of this work. This work was supported in partby funds from the Hermelin Brain Tumor Center, Henry FordHospital, to S.K. and T.M. and by the NIH grantR01CA122031 to A.S.A.

DISCLOSURE OF POTENTIAL CONFLICTS

OF INTEREST

The authors indicate no potential conflicts of interest.

REFERENCES

1 Phillips HS, Kharbanda S, Chen R et al. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progres-sion, and resemble stages in neurogenesis. Cancer Cell 2006;9:157–173.

2 Freije WA, Castro-Vargas FE, Fang Z et al. Gene expression profilingof gliomas strongly predicts survival. Cancer Res 2004;64:6503– 6510.

3 Meis JM, Martz KL, Nelson JS. Mixed glioblastoma multiforme andsarcoma. A clinicopathologic study of 26 radiation therapy oncologygroup cases. Cancer 1991;67:2342–2349.

4 Louis DN, Ohgaki H, Wiestler OD, Cavenee WK, eds. WHO Classifi-cation of Tumours of the Central Nervous System. 4th ed. Lyon,France: IARC, 2007.

5 Perry JR, Ang LC, Bilbao JM et al. Clinicopathologic features of primaryand postirradiation cerebral gliosarcoma. Cancer 1995;75:2910–2918.

Figure 6. Gliosarcoma phenotype is maintained in the GS1 xenografts. Expression of neural (GFAP and Nestin), mesenchymal (reticulin anda-SMA), and a marker for both lineages (Vimentin) in founder GS1 tumor (A–E), and in xenografts generated by neurospheres [NP, (F–J)],low-passage [FBS-P3; (K–O)], and high-passage [FBS-P20; (P–T)] serum-cultured cells. Reticulin staining was observed for all GS1 samples(A,F,K,P). GBM3 xenograft was included as a control (U–Y) and did not stain for the mesenchymal markers (U,Y). Arrows in (E), (J), (O),and (T) point to individual GS1 cells expressing a-SMA. Bar, 20 lm (C–E,H–J,M–O,R–T,W–Y) and 40 lm (A,B,F,G,K,L,P,Q,U,V).Abbreviations: FBS, fetal bovine serum; GBM, glioblastoma; GFAP, glial fibrillary acidic protein; GS, gliosarcoma; NP, neurosphere; r, reticulinstaining.

deCarvalho, Nelson, Lemke et al. 189

www.StemCells.com

6 Lieberman KA, Fuller CE, Caruso RD et al. Postradiation gliosar-coma with osteosarcomatous components. Neuroradiology 2001;43:555–558.

7 Reis RM, Konu-Lebleblicioglu D, Lopes JM et al. Genetic profile ofgliosarcomas. Am J Pathol 2000;156:425–432.

8 Lutterbach J, Guttenberger R, Pagenstecher A. Gliosarcoma: a clinicalstudy. Radiother Oncol 2001;61:57–64.

9 Galanis E, Buckner JC, Dinapoli RP et al. Clinical outcome of glio-sarcoma compared with glioblastoma multiforme: North Central Can-cer Treatment Group results. J Neurosurg 1998;89:425–430.

10 Meis JM, Ho KL, Nelson JS. Gliosarcoma: a histologic and immuno-histochemical reaffirmation. Mod Pathol 1990;3:19–24.

11 Boerman RH, Anderl K, Herath J et al. The glial and mesenchymalelements of gliosarcomas share similar genetic alterations. J Neuropa-thol Exp Neurol 1996;55:973–981.

12 Biernat W, Aguzzi A, Sure U et al. Identical mutations of the p53tumor suppressor gene in the gliomatous and the sarcomatous compo-nents of gliosarcomas suggest a common origin from glial cells.J Neuropathol Exp Neurol 1995;54:651–656.

13 Actor B, Cobbers JM, Buschges R et al. Comprehensive analysis ofgenomic alterations in gliosarcoma and its two tissue components.Genes Chromosomes Cancer 2002;34:416–427.

14 Vescovi AL, Galli R, Reynolds BA. Brain tumour stem cells. Nat RevCancer 2006;6:425–436.

15 Lee J, Kotliarova S, Kotliarov Y et al. Tumor stem cells derived fromglioblastomas cultured in bFGF and EGF more closely mirror the phe-notype and genotype of primary tumors than do serum-cultured celllines. Cancer Cell 2006;9:391–403.

16 Reynolds BA, Weiss S. Generation of neurons and astrocytes fromisolated cells of the adult mammalian central nervous system. Science1992;255:1707–1710.

17 Galli R, Binda E, Orfanelli U et al. Isolation and characterization oftumorigenic, stem-like neural precursors from human glioblastoma.Cancer Res 2004;64:7011–7021.

18 Ignatova TN, Kukekov VG, Laywell ED et al. Human cortical glialtumors contain neural stem-like cells expressing astroglial and neuro-nal markers in vitro. Glia 2002;39:193–206.

19 Singh SK, Clarke ID, Terasaki M et al. Identification of a cancer stemcell in human brain tumors. Cancer Res 2003;63:5821–5828.

20 Yuan X, Curtin J, Xiong Y et al. Isolation of cancer stem cells fromadult glioblastoma multiforme. Oncogene 2004;23:9392–9400.

21 Kondo T, Setoguchi T, Taga T. Persistence of a small subpopulationof cancer stem-like cells in the C6 glioma cell line. Proc Natl AcadSci USA 2004;101:781–786.

22 Ghods AJ, Irvin D, Liu G et al. Spheres isolated from 9L gliosarcomarat cell line possess chemoresistant and aggressive cancer stem-likecells. Stem Cells 2007;25:1645–1653.

23 Clarke DL, Johansson CB, Wilbertz J et al. Generalized potential ofadult neural stem cells. Science 2000;288:1660–1663.

24 Ricci-Vitiani L, Pallini R, Larocca LM et al. Mesenchymal differen-tiation of glioblastoma stem cells. Cell Death Differ 2008;15:1491–1498.

25 Livak KJ, Schmittgen TD. Analysis of relative gene expression datausing real-time quantitative PCR and the 2(�DDC(T)) method. Meth-ods 2001;25:402–408.

26 Szybka M, Zawlik I, Kulczycka D et al. Elimination of wild-type P53mRNA in glioblastomas showing heterozygous mutations of P53. Br JCancer 2008;98:1431–1433.

27 Mihaila D, Jankowski M, Gutierrez JA et al. Meningiomas: loss ofheterozygosity on chromosome 10 and marker-specific correlationswith grade, recurrence, and survival. Clin Cancer Res 2003;9:4443–4451.

28 McKeever PE, Varani J, Papadopoulos SM et al. Products of cellsfrom gliomas: IX. Evidence that two fundamentally different mecha-nisms change extracellular matrix expression by gliomas. J Neuroon-col 1995;24:267–280.

29 McKeever PE, Davenport RD, Shakui P. Patterns of antigenic expres-sion of human glioma cells. Crit Rev Neurobiol 1991;6:119–147.

30 Tso CL, Shintaku P, Chen J et al. Primary glioblastomas express mes-enchymal stem-like properties. Mol Cancer Res 2006;4:607–619.

31 Joo KM, Kim SY, Jin X et al. Clinical and biological implications ofCD133-positive and CD133-negative cells in glioblastomas. LabInvest 2008;88:808–815.

32 Wang J, Sakariassen PO, Tsinkalovsky O et al. CD133 negative gli-oma cells form tumors in nude rats and give rise to CD133 positivecells. Int J Cancer 2008;122:761–768.

33 Gage FH, Ray J, Fisher LJ. Isolation, characterization, and use ofstem cells from the CNS. Ann Rev Neurosci 1995;18:159–192.

34 Ghosh S, Spagnoli GC, Martin I et al. Three-dimensional culture of mela-noma cells profoundly affects gene expression profile: a high density oligo-nucleotide array study. J Cell Physiol 2005;204:522–531.

35 Lendahl U, Zimmerman LB, McKay RD. CNS stem cells express anew class of intermediate filament protein. Cell 1990;60:585–595.

36 Wislet-Gendebien S, Leprince P, Moonen G et al. Regulation of neu-ral markers nestin and GFAP expression by cultivated bone marrowstromal cells. J Cell Sci 2003;116:3295–3302.

37 Haddad SF, Moore SA, Schelper RL et al. Smooth muscle can com-prise the sarcomatous component of gliosarcomas. J Neuropathol ExpNeurol 1992;51:493–498.

38 Laks DR, Masterman-Smith M, Visnyei K et al. Neurosphere forma-tion is an independent predictor of clinical outcome in malignant gli-oma. Stem Cells 2009;27:980–987.

39 Ma YH, Mentlein R, Knerlich F et al. Expression of stem cell markersin human astrocytomas of different WHO grades. J Neurooncol 2008;86:31–45.

40 Singh SK, Hawkins C, Clarke ID et al. Identification of human braintumour initiating cells. Nature 2004;432:396–401.

41 Beier D, Hau P, Proescholdt M et al. CD133(þ) and CD133(�) glio-blastoma-derived cancer stem cells show differential growth character-istics and molecular profiles. Cancer Res 2007;67:4010–4015.

42 Episkopou V. SOX2 functions in adult neural stem cells. Trends Neu-rosci 2005;28:219–221.

43 Schmitz M, Temme A, Senner V et al. Identification of SOX2 as anovel glioma-associated antigen and potential target for T cell-basedimmunotherapy. Br J Cancer 2007;96:1293–1301.

44 Gangemi RM, Griffero F, Marubbi D et al. SOX2 silencing in glio-blastoma tumor initiating cells causes stop of proliferation and loss oftumorigenicity. Stem Cells 2009;27:40–48.

45 Galli R, Gritti A, Bonfanti L et al. Neural stem cells: an overview.Circ Res 2003;92:598–608.

See www.StemCells.com for supporting information available online.

190 Gliosarcoma Model Using Cancer Stem Cells