an expression signature classifies chemotherapy-resistant pediatric osteosarcoma

An Expression Signature Classifies Chemotherapy-Resistant

Pediatric Osteosarcoma

Michelle B. Mintz,1,3Rebecca Sowers,

6Kevin M. Brown,

2Sara C. Hilmer,

4BethAnne Mazza,

7

Andrew G. Huvos,8Paul A. Meyers,

7Bonnie LaFleur,

10Wendy S. McDonough,

1Michael M. Henry,

4

Keri E. Ramsey,1Cristina R. Antonescu,

8Wen Chen,

7John H. Healey,

9Aaron Daluski,

11

Michael E. Berens,1Tobey J. MacDonald,

5Richard Gorlick,

6and Dietrich A. Stephan

1

1Neurogenomics Division and 2Genetic Basis of Human Disease Division, Translational Genomics Research Institute, Phoenix, Arizona;3Department of Genetics, George Washington University; 4Research Center for Genetic Medicine and 5Center for Cancer Research,Children’s National Medical Center, Washington, District of Columbia; 6Department of Pediatrics, Children’s Hospital at Montefiore, Bronx,New York; Departments of 7Pediatrics, 8Pathology, and 9Surgery, Memorial Sloan-Kettering Cancer Center, New York, New York;10Department of Preventative Medicine, Vanderbilt University Medical Center, Nashville, Tennessee; and 11Department of Orthopedics,Cornell University Medical Center, Ithaca, New York

Abstract

Osteosarcoma is the most common malignant bone tumorin children. Osteosarcoma patients who respond poorly tochemotherapy are at a higher risk of relapse and adverseoutcome. Therefore, it was the aim of this study to identifyprognostic factors at the time of diagnosis to characterize thegenes predictive of poor survival outcome and to identifypotential novel therapeutic targets. Expression profiling of30 osteosarcoma diagnostic biopsy samples, 15 with inferiornecrosis following induction chemotherapy (Huvos I/II) and15 with superior necrosis following induction chemotherapy(Huvos III/IV), was conducted using Affymetrix U95Av2oligonucleotide microarrays. One hundred and four geneswere found to be statistically significant and highly differen-tially expressed between Huvos I/II and III/IV patients.Statistically significant genes were validated on a smallindependent cohort comprised of osteosarcoma xenografttumor samples. Markers of Huvos I/II response predominantlywere gene products involved in extracellular matrix (ECM)microenvironment remodeling and osteoclast differentiation.A striking finding was the significant decrease in osteoprote-gerin, an osteoclastogenesis inhibitory factor. Additionalgenes involved in osteoclastogenesis and bone resorption,which were statistically different, include annexin 2 , SMAD,PLA2G2A , and TGFBB1 . ECM remodeling genes include desmo-plakin , SPARCL1 , biglycan , and PECAM . Gene expression ofselect genes involved in tumor progression, ECM remodeling,and osteoclastogenesis were validated via quantitative reversetranscription-PCR in an independent cohort. We propose thatosteosarcoma tumor–driven changes in the bone microenvi-ronment contribute to the chemotherapy-resistant phenotypeand offer testable hypotheses to potentially enhance thera-peutic response. (Cancer Res 2005; 65(5): 1748-54)

Introduction

Osteosarcoma is a primary malignant tumor of the bone.Osteosarcoma accounts for f5% of childhood tumors in theUnited States and is the most common malignant bone tumor in

children (1). Over the past three decades, advances in treatmenthave been responsible for improved limb salvage, reducedmetastases, and overall higher survival rates (1). Multiagent dose-intensive chemotherapy regimens have resulted in long-termdisease-free survival rates of f60% to f76% in patients withlocalized disease. Osteosarcoma patients whose tumors respondpoorly to chemotherapy are at a higher risk of relapse and adverseoutcome. Therefore, it is imperative to identify prognostic factorsat the time of diagnosis to detect chemotherapy-resistant tumorsand to generate a modified treatment regimen.The standard therapy regimen of high-grade osteosarcoma

includes induction multiagent chemotherapy followed by surgicalresection and postoperative chemotherapy (2). Induction therapyallows for treatment of micrometastatic disease, tumor shrinkage,and decreased tumor vascularity, thus facilitating the surgicalremoval of the tumor (3). The percentage of necrotic tissuefollowing induction chemotherapy is classified using the Huvosgrading system, where the various levels of necrosis reflect theeffectiveness of the given therapy (Table 1; ref. 4). Patients with<90% tumor necrosis following induction therapy are classified asinferior responders, or Huvos grade I/II (1). To date, aside frommetastatic lesions at presentation, histologic response to chemo-therapy is the most dependable and reproducible prognosticindicator of the probability of disease-free survival (1). The degreeof necrosis following definitive surgery remains the only consistentprognostic factor in the majority of patients presenting withapparently localized disease. As treatment regimens have evolvedover time, numerous clinical trials have attempted to increase thedisease-free survival rate for poorly responding patients withintensified postoperative therapy. However, no survival benefit hasbeen convincingly shown through the administration of moreintensified therapy to poor responders. This conclusion has beenreached by several independent clinical trials (5–7). This suggeststhat there may be an innate biological difference betweenresponsive and nonresponsive tumors.Although Huvos grading is a powerful predictor of survival, the

prognostic marker is ineffective in altering the outcome ofchemoresistant tumors, as this indicator can only be determinedafter therapy has already been given. For patients who displayinferior histologic response to chemotherapy, it is crucial todetermine the biological factors that drive lack of response at thetime of diagnosis. The importance of this is twofold: First, toidentify a genetic fingerprint that will distinguish patients as goodor poor responders before therapy; second, to identify potential

Requests for reprints: Dietrich A. Stephan, Neurogenomics Division, TranslationalGenomics Research Institute, 445 North Fifth Street, Phoenix, AZ 85004. Phone: 602-343-8727; Fax: 602-334-8844; E-mail: [email protected].

I2005 American Association for Cancer Research.

Cancer Res 2005; 65: (5). March 1, 2005 1748 www.aacrjournals.org

Research Article

target candidates to customize induction therapy. Therefore, it wasthe aim of this study to compare the gene expression profiles ofbiopsy samples obtained from responsive and nonresponsivepatients with osteosarcoma to characterize the genes predictiveof poor survival outcome and to identify potential noveltherapeutic targets. The results of this study show that resistanttumors may have an increased ability to express osteoclastogenesis,tumor progression, and extracellular matrix (ECM) remodelinggenes. We have validated these findings on a limited number ofindependent tumor samples and propose bisphosphonate ana-logues as potential adjuvant therapies in patients with chemother-apy-resistant osteosarcoma at the time of induction therapy.

Materials and Methods

Patient Samples. All tumor samples utilized in this study were obtained

from patients treated with the same standard induction therapy of 10 weeks

cisplatin, doxorubicin, and high-dose methotrexate. Thirty flash-frozen

osteosarcoma specimens, including 15 Huvos grade I/II and 15 Huvos gradeIII/IV osteosarcoma tumors, were obtained from Memorial Sloan-Kettering

Cancer Center (New York, NY) with informed consent from patients/

guardians in accordance with a biology study approved by Memorial

Hospital Institutional Review Board and stored at �80jC until ready for usein the gene expression profiling studies. Samples from each primary tumor

were acquired at the time of the initial diagnostic biopsy. Three additional

samples that had been maintained as xenografts in severe combinedimmunodeficient mice were obtained from Memorial Sloan-Kettering

Cancer Center and used for gene expression profiling. The tumors were

prepared using the following protocol: Poor histologically responsive

primary tumors were isolated and grown in culture to confluence;thereafter, cells were harvested and injected into severe combined

immunodeficient mice. The resulting xenograft tumors were isolated and

stored at �80jC until ready for use in the gene expression profiling studies.

Quantitative reverse transcription-PCR (RT-PCR) analysis was conducted onsix independent osteosarcoma biopsy specimens (three Huvos grade I/II

and three Huvos grade III/IV osteosarcoma tumors) obtained at Memorial

Sloan-Kettering Cancer Center and stored at �80jC until ready for use. The

Children’s National Medical Center Institutional Review Board approved theconduct of the laboratory analyses.

Gene Expression Profiling. Total RNA was isolated from tissues using

TRIzol reagent (Invitrogen, Carlsbad, CA) as described previously (8). Five

to twenty-five micrograms total RNA were obtained from each tumor tissue

sample. Five micrograms extracted RNA from each sample were synthesized

into double-stranded cDNA using the SuperScript choice system

(Life Technologies, Inc., Carlsbad, CA) using an oligo(dT) primer containing

a T7 RNA polymerase promoter (Genset, San Diego, CA). The double-

stranded cDNA was purified by phenol-chloroform extraction and in vitro

transcribed using the ENZO Bioarray RNA transcript labeling kit

(Affymetrix, Santa Clara, CA). The biotin-labeled cRNA was purified using

the RNeasy kit (Qiagen, Valencia, CA) and then fragmented to f200 bp

through alkaline treatment [200 mmol/L Tris-acetate (pH 8.2), 500 mmol/L

potassium acetate, and 150 mmol/L magnesium acetate]. Each cRNA

sample was hybridized to an Affymetrix U95Av2 GeneChip Array containing

12,625 human transcripts. cRNA hybridization was detected using

a streptavidin-conjugated fluor and confocal laser scanning according to

manufacturer’s recommendations (Affymetrix). Data was extracted using

Microarray Suite (MAS) version 5.0 (Affymetrix) and linearly scaled to

achieve an average intensity of 800 across each chip. Data from a chip was

discarded when the following control criteria was not met: if the ratio of

the average intensities of the 3V to 5V ends of the human glyceraldehyde-

3-phosphate dehydrogenase gene exceeded 3; if the linear scaling factor

exceeded 4; or if the percent ‘‘present’’ calls across an array were <30%.Statistical Methods and Selection of Differentially Expressed Genes.

The candidate gene list obtained from MAS 5.0–extracted data was

selected by eliminating genes that were not present in at least onesample. Statistically significant genes were then selected using a

permutational t test (P < 0.05). The permutation test was done asdescribed previously to determine the genes that showed significant

differences in the mean intensities between the two classes (8). Highly

differentially expressed genes were identified by selecting genes that met

an average fold-change criterion of >2.0. MAS 5.0 normalized data wasadditionally inserted into Significance Analysis of Microarrays version 1.21

add-in to Microsoft Excel (www-stat.stanford.edu/~tibs/SAM/index) to

determine the false discovery rate of significant genes. Supervised

clustering of MAS 5.0 normalized genes was conducted using dChipsoftware, where the distance between two genes is defined as 1 � r,

where r is the Pearson correlation coefficient between the standardized

expression values (Wong Lab, Department of Biostatistics, Harvard Schoolof Public Health, Boston, MA). Gene ontology annotations for genes

represented on the Affymetrix U95Av2 chip were compiled from the

NetAffx Analysis Center (www.affymetrix.com/analysis/index.affx).

Reverse Transcription and TaqMan Quantitative PCR Analysis.All reagents and primers used for reverse transcription and PCR

amplification were obtained from Invitrogen. Total RNA was reverse-

transcribed using a RT-PCR reaction mix consisting of oligo (dT; 0.5 Ag/AL),random hexamers (50 ng/AL), 10� cDNA synthesis buffer, 25 mmol/LMgCl2, 10 mmol/L deoxynucleotide triphosphate, 0.1 mol/L dithiothreitol,

RNaseOUT (40 units/AL), and SuperScript II RT (200 units/AL). The

samples were incubated at 25jC for 10 minutes, followed by a 50-minute

incubation at 42jC. The reaction was terminated at 70jC for 20 minutesfollowed by a 20-minute incubation at 37jC with E. coli RNase H

(2 units/AL).TaqMan PCR primers were designed using Invitrogen LUX (Light Upon

eXtension) fluorogenic designer software and labeled with 6-carboxy-

fluorescein phosphoramidite (FAM). 18S rRNA predesigned primers

were purchased and used as an internal control. Primer sequences for

annexin 2 were ( forward) 5V-CCTGAGCGTCCAGAAATGG[FAM]G-3Vand (reverse) 5V-GGACTGTTATTCGCAAGCTGGTT-3V; for biglycan,

( forward) 5V-CACGGCCTCTTCAACAACCCCG[FAM]G-3V and (reverse)

5V-TCAGTGACGCAGCGGAAAGT-3V; for desmoplankin, ( forward) 5V-CAC-CAGCCATGATCTGTTTCCACTGG[FAM]G-3V and (reverse) 5V-GCC-TCAGAACTCATTCTCCAAA-3V; for osteoprotegerin (OPG), ( forward)

5V-CACTGAGGCACTTGAGGCTTTCAG[FAM]G-3V and (reverse) 5V-TGG-AGGAGATGTTAGTCCTTTCTC-3V; for phospholipase A2 (PLA2G2A),( forward) 5V-GAACTTGCCCTCCCTACCCTAACCAAG5TC-3V and (reverse)

5V-GAAGGCTGGAAATCTGCTGGA-3V; for plasminogen activator inhibitor

type 1, ( forward) 5V-CACGCGAGGAAGAAATGTCAGATGCG[FAM]G-3V and

Table 1. Histologic grading of the effect of inductivechemotherapy on primary osteosarcoma

Huvos

grade

Percent

necrosis (%)

Histologic effect

I 0-50 Little or no effect identified

II 51-90 Areas of acellular tumor osteoid,

necrotic, or fibrotic materialattributable to the effect of

chemotherapy, with other areas

of histologically viable tumor

III 91-99 Predominant areas of acellular osteoid,necrotic, or fibrotic material

attributable to the effect of

chemotherapy, with only scatteredfoci of histologically viable tumor

cells identified

IV 100 No histologic evidence of viable tumor

identified within the entire specimen

NOTE: Adapted from ref. 4.

Chemotherapy Resistance in Osteosarcoma

www.aacrjournals.org 1749 Cancer Res 2005; 65: (5). March 1, 2005

(reverse) 5V-CAAGCACTCAAGGGCAAGGA-3V; for secreted protein acidicand rich in cysteine (SPARC)-like 1, ( forward) 5V-GATCAAGTCCTCCAA-CAAAACCCCTTGA[FAM]C-3V and (reverse) 5V-TCCAGTCTGCATTTAG-TAGCGAAT-3V.

A TaqMan ABI 7700 (Applied Biosystems, Foster City, CA) was used forcDNA quantification of independent Huvos I/II (n = 3) and Huvos III/IV

(n = 3) tumor samples. Briefly, 50 ng sample cDNA were added to a PCR

reaction mix consisting of 2� Platinum Quantitative PCR SuperMix-UDG

[60 units/mL Platinum Taq polymerase, 40 mmol/L Tris-HCL (pH 8.4), 100mmol/L KCL, 6 mmol/L MgCl2, 400 Amol/L dGTP, 400 Amol/L dATP, 400

Amol/L dCTP, 800 Amol/L dUTP, 40 units/mL UDG], 1 Amol/L 5,6-carboxy-

X-rhodamine reference dye, forward FAM-labeled LUX primer (10 Amol/L),

and reverse unlabeled primer (10 Amol/L). Samples were amplified intriplicate by using the following thermocycling conditions: 50jC for 2

minutes, 95jC for 2 minutes, followed by 45 cycles of amplification at 95jCfor 15 seconds, 55jC for 30 seconds, and 72jC for 30 seconds. cDNAquantification was directly related to fluorescence of the FAM fluorophore

during 45 cycles of amplification. FAM fluorescence is detected when the

primer becomes incorporated into double-stranded PCR product, which

causes the fluorophore to dequench and results in a significant increase influorescent signal. Estimation of mRNA levels were quantified relative to

18S rRNA to compensate for variations in sample quantity and reverse

transcription efficiency. Sample mRNA levels were measured using the

DDCT method as described in the ABI Prism 7700 Sequence DetectionSystem manual (Applied Biosystems).

Results

Expression Profiling of Osteosarcoma. Expression profilingwas conducted on 30 independent osteosarcoma biopsy samples.Fifteen of these samples were obtained from patients with aninferior response to chemotherapy (Huvos I/II), whereas theremaining 15 samples were obtained from patients with a favorableresponse to chemotherapy (Huvos III/IV). Grading of these biopsysamples was assigned after induction therapy by observing thepercentage of necrosis visible after preoperative chemotherapy.Expression profiling was conducted using Affymetrix U95Av2

chips comprised of 12,625 probe sets. The 12,625 probe sets werenormalized via a linear scaling on MAS 5.0 and cropped forsamples displaying at least one present call across the 30osteosarcoma tumor set, resulting in a gene list of 9,030 genes.The 9,030 genes were analyzed using the Significance Analysis ofMicroarrays add-in to Microsoft Excel to ascertain the authenticityof genes deemed significant. One hundred and twelve significantgenes were selected using a y of 0.4248. The false discovery rateobtained for these 112 genes was 17.8%, indicating a highprobability that these genes were selected due to statisticalsignificance and not due to chance alone.The 9,030 genes exhibiting at least one present call were

further cropped for genes differentially expressed between HuvosI/II and III/IV tumors through the use of a permutational t testand differential fold change cutoff. Nine hundred and ten geneswere differentially expressed (P < 0.05). To further select forgenes with the largest average fold changes, we cropped the 910differentially expressed genes by first selecting genes with anaverage fold change >2.0, and then selecting genes with anaverage expression of at least 500 in at least one class (Fig. 1).Of the 104 differentiated genes that met these criteria, 63 (60.5%)were up-regulated in poorly responsive tumors and 41 (39.5%)were down-regulated in poorly responsive tumors (Fig. 1). The104 significant genes obtained from permutational t testing andfold-change restrictions displayed absolute concordance with the112 significant genes selected by Significance Analysis ofMicroarrays software.

The 910 significant genes were examined to determine theoverarching pathways involved in poor histologic response.Confirmation of the identified pathways was conducted byvalidating select genes obtained from the 104 highly differentiallyexpressed genes.Classification of Independent Tumor Samples. Xenograft

samples were derived from aggressively growing, poor chemo-responsive primary tumor cells grown in severe combinedimmunodeficient mice. Xenograft samples derived from primaryosteosarcoma tumor cells were expression-profiled using Affyme-trix U95Av2 Gene Chips. Independent validation of the 104 highlydifferentially expressed and statistically significant genes wasconducted on this separate sample cohort using hierarchicalclustering (Fig. 2). The xenograft specimens clearly clustered withthe poorly responsive tumors using this unsupervised method.RT-PCR Validation of Microarray Expression Results. RT-

PCR validation was conducted on six independent osteosarcomatumor samples (Table 2). Genes were selected for mRNAconfirmation based on biologically and statistically significantgenes through MAS 5.0 normalization methods. Quantitative RT-PCR was conducted on desmoplakin, OPG, plasminogen activatorinhibitor type 1, biglycan, annexin A2, SPARC-like 1, and PLA2G2A.Fold changes obtained from quantitative RT-PCR confirm themicroarray expression fold changes (Table 2).

Discussion

Microarray technology has provided the means for studying themolecular basis of human disease by examining thousands of genessimultaneously. Microarray technology holds particular promise inthe study of the innate chemotherapy resistance in osteosarcomapatients, where presently there is an inability to salvage poorlyresponsive patients once induction therapy has been given. It isimperative to determine the biological factors that drive chemo-therapy resistance in osteosarcoma patients with adverse histologicresponse to chemotherapy to predict survival outcome and toidentify potential novel therapeutic targets at diagnosis. This is thefirst study to subclassify nonresponders, to identify new therapeu-tic targets for nonresponders, and to suggest chemotherapeuticopportunities for further testing.We expression-profiled 15 independent Huvos I/II and 15 Huvos

III/IV osteosarcoma biopsy samples and identified 910 statisticallysignificant genes and 104 highly differentially expressed significantgenes. Permutational analysis of this data set suggests a low falsediscovery rate, giving credence and biological significance to thepoor-responsive gene expression signature. Gene expressionvalidation was conducted through the unsupervised classificationof independent therapy-resistant tumor xenografts and quantita-tive RT-PCR in an independent set of osteosarcomas. We proposethat a poor-prognosis osteosarcoma expression signature involvesa drive toward osteoclastogenesis, ECM remodeling, and tumorprogression. We further speculate that functionally diverse genescooperatively promote an altered ECM environment to promotetumor survival (Fig. 3).Confirmation of Differential Expression between Classes on

Independent Tumor Samples. We have generated a geneticfingerprint of chemotherapy-resistant tumors (Fig. 1). A selection ofthese genes involved in osteoclastogenesis and ECM remodelingwas validated via quantitative RT-PCR on an independent tumorsample set (Huvos I/II n = 3; Huvos III/IV n = 3). Quantitative RT-PCR validation was conducted on desmoplakin, OPG, plasminogenactivator inhibitor type 1, biglycan, annexin 2, PLA2G2A, and

Cancer Research


Figure 1. Chemotherapy resistance genes in osteosarcoma. One hundred and four highly differentiating genes were identified through statistical significance andfold change criteria in 30 individual osteosarcoma microarray samples. These genes differentiating poor and good histologic response to chemotherapy are representedin this cluster gram. Levels of expression are represented on a scale from the lowest expression (dark green) to the highest expression (bright red). Fold differencesare represented in the last column. A negative fold difference indicates a decreasing fold change between Huvos I/II versus Huvos III/IV tumors, whereas a positivefold difference indicates an increasing fold change.



SPARC-like 1. The microarray expression differences of these geneswere confirmed in this independent sample set (Table 2).This quantitative RT-PCR validation verifies the proposed dysre-gulation of osteoclastogenesis and ECM modulating genes in poor-prognosis samples.To further confirm that the significant genes identified in the

microarray data were indeed biologically significant, we investi-gated whether this gene list (Fig. 1) was able to segregateosteosarcoma tumors characterized by aggressive growth behaviorand poor therapeutic response. Xenograft samples grown fromprimary aggressive and chemotherapy-resistant osteosarcomatumors were expression-profiled on Affymetrix U95Av2 GeneChips.Using an unsupervised strategy, xenograft samples clearlyclustered with Huvos I/II samples (Fig. 2), concluding that genesidentified from osteosarcoma tumors in this preliminary study candefinitively classify Huvos I/II osteosarcoma tumors. We recognizethat the validation cohort was limited but the excellentclassification ability of this signature using unsupervised analysiscoupled with the RT-PCR validation data illustrates a robustsignature.Poor-Prognosis Tumors Are Associated with Osteoclasto-

genesis and Bone Resorption. Statistically significant genes arecharacterized by a prevailing pattern of genes involved inosteoclast differentiation and bone resorption. The dramatic foldchange decrease of �5.44 of OPG is a striking implication of theinvolvement of osteoclast promotion in poorly responsive osteo-sarcoma tumors. OPG is a secreted glycoprotein that regulates

bone resorption through the inhibition of osteoclast differentiation(9). A decrease in OPG has been implicated in the malignant andbone-metastasizing phenotype in breast carcinoma (10, 11) and inmyeloma-induced osteolysis (12).The tumor-mediated bone remodeling cycle arises from the

interaction between the tumor and bone microenvironment. Inmetastatic breast cancer, tumor cells mediate osteoclast activitythrough the inhibition of OPG and expression of parathyroid-related peptide. The inhibition of OPG allows for the differentiationof osteoclast multinucleated cells through the interaction ofreceptor activator of nuclear factor nB and its ligand. Osteoclasticactivity liberates cytokines and growth factors embedded in thebone matrix. These growth factors, such as TGF-h, stimulate thetumor cell and facilitate tumor gene expression through SMAD andmitogen-activated protein kinase pathways, resulting in the cyclicmediation of bone resorption (13).Multiple genes involved in the osteoclast differentiation signaling

pathway were identified among the 910 statistically significantgenes. An increased expression of TGF-h1 (P = 0.03; fold increase,1.6), MAPK (P = 0.02; fold increase, 1.4), and SMAD (P = 0.005; foldincrease, 1.6) genes were observed in Huvos I/II tumor samples(Fig. 3; ref. 14). The mitogen-activated protein kinase pathway hasbeen shown to be a major component of SMAD signaling by TGF-hin osteolytic bone metastases in breast cancer (14).Additionally, expression of annexin 2 (P = 0.049) and PLA2G2A

(P = 0.025), genes indirectly involved in bone resorption, were up-regulated in Huvos I/II samples. Annexin 2 is a Ca2+-dependentphospholipid and membrane-binding protein that enhancesosteoclast formation and bone resorption (15). Additionally,PLA2G2A has been implicated in osteoclastic bone resorption(16). Increased levels of PLA2G2A have additionally beenassociated with various tumors, such as prostate cancer, pancreaticcarcinoma, small bowel adenocarcinomas, and colorectal carcinoma(17). More importantly, PLA2G2A has proven to play a crucial rolein the production of prostaglandins, potent regulators of boneformation, and resorption. Prostaglandin E2 has been proven tostimulate bone resorption in organ culture and mediate resorptionof mouse calvaria in response to cytokines and growth factors, suchas TGF-h (18). Prostaglandin is intricately tied to OPG and receptoractivator of nuclear factor nB ligand expression and has shown tosuppress OPG mRNA expression to stimulate bone resorption(19, 20). Additionally, osteolytic metastases of breast cancer havebeen shown to be caused by a prostaglandin E2–dependentmechanism (19).

Figure 2. Unsupervised hierarchical clustering of osteosarcoma xenograftsamples was conducted on the 104 probe sets found to be biologicallysignificant in the 30 osteosarcoma biopsy tumor sample set.

Table 2. Quantitative RT-PCR confirmation of transcriptional changes of six genes differentially expressed in an independentset of chemotherapy-responsive and nonresponsive osteosarcoma tumors

Affymetrix

probe name

GenBank

identifier

Gene name Microarray Quantitative RT-PCR

Average fold change:Huvos I/II versus III/IV

P Fold change (mean F SD):Huvos I/II versus III/IV

36133_at AL031058 Desmoplakin �8.87 8.680 � 10�3 �4.1 F 1.2237611_at AB008822 Osteoprotegerin �5.44 3.126 � 10�2 �5.47 F 2.66

38125_at M14083 PAI-1 �2.69 4.084 � 10�2 �2.97 F 2.75

38126_at J04599 Biglycan 2.15 2.516 � 10�2 2.12 F .98

769_s_at D00017 Annexin A2 2.17 4.996 � 10�2 1.52 F 1.0136627_at X86693 SPARCL1 7.65 9.80 � 10�4 6.29 F 2.97

37017_at M22430 PLA2G2A 10.58 2.54 � 10�2 7.21 F 1.82

Cancer Research


Given these data implicating osteoclasts in tumor growth andtherapy resistance, drugs inhibiting osteoclast formation andfunction may represent a possible novel adjunct therapy forosteosarcoma. Bisphosphonate compounds inhibit osteoclastdifferentiation and bone resorption by targeting gene transcriptionand enzyme metabolism in mature osteoclasts, leading toosteoclast apoptosis and loss of bone-resorptive activity (21).Bisphosphonates have been shown to inhibit osteosarcoma cell

lines and tumor growth. Aminobisphosphonate pamidronate hasbeen implicated as a potent inhibitor of osteosarcoma cell linegrowth. The administration of pamidronate was shown to reducecell viability and proliferation in seven osteosarcoma cell lines,namely HOS, MG-63, OST, SaOS-2, SJSA-1, U2OS, and ZK-58.Observed reduction of cell viability occurred in a dose-dependentmanner (22). Additional studies conducted with the administrationof the aminobisphosphonate pamidronate to the UMR 106-01clonal rat osteosarcoma cell line showed a decrease in osteosar-coma cell growth, an induction of apoptosis, and an alteredexpression of osteoclast-regulating factors (23). The bisphospho-nate, zoledronic acid, has additionally shown to induce apoptosis inthe human osteosarcoma cell lines HOS, BTK-143, MG-63, SJSA-1,G-292, and SAOs2 (24).Bisphosphonates are approved by Food and Drug Administra-

tion to treat bone loss associated by osteoporosis, Paget’s disease,and osteolytic metastases of breast cancer. The potential forbisphosphonates merits further evaluation as an effective tool todecrease osteosarcoma cell growth, both for its proven ability todecrease osteosarcoma cell growth in vitro and its minimal sideeffects (22).Poor-Prognosis Tumors Are Associated with ECMRemodeling.

In addition to osteoclast-stimulating genes, ECM modulatingproteins were differentially expressed between Huvos I/II and III/IV tumors. Huvos I/II tumors display a marked increase of genesinvolved in cell migration, proliferation, and remodeling. Forinstance, thrombospondin 4 (P = 0.04) is an extracelluar bindingprotein involved in cell proliferation and migration (25). Plateletendothelial cell adhesion molecule (P = 0.027) and perlecan

(P = 0.0072) are both implicated in angiogenesis pathways as wellas tumor growth promotion (26, 27). Biglycan (P = 0.025) isinvolved in ECM remodeling through connective tissue metabolism(28). SPARC-like 1 (P = 0.001) is additionally involved in remodelingcellular architecture through the rounding of adherent cells (29).Poorly responsive tumors display a decrease in genes involved

with cytoskeleton stability and homeostasis. For instance, keratin19 (P = 0.015) and desmoplakin (P = 0.008) function in main-taining cytoskeletal architecture and cell stability (30, 31). Poorlyresponsive tumors additionally display a decrease in plasminogenactivator inhibitor type 1 (P = 0.04), a protein that functions in theinhibition of cell migration and the invasive phenotype of cancercell lines (32). A decrease in these genes strongly implies anincrease in motility in poorly responsive osteosarcoma tumors.Poor-Prognosis Tumors Are Associated with Tumor Pro-

gression and Apoptosis Resistance. Analysis of the 104 highlydifferentially expressed genes revealed functionally diverse genesinvolved in tumor apoptosis resistance and tumor progression inpoorly responsive osteosarcoma tumor samples. Poorly responsivetumors show an up-regulation of two metallothionein compounds,namely metallothionein IG (P = 0.045) and metallothionein 1L(P = 0.007). Metallothionein, a heavy metal–binding protein classcharacterized by high cysteine content, is associated withchemotherapy resistance. Metallothioneins have been found toplay a physiologic role in metabolizing and eliminating cytotoxicheavy metals found in chemotherapy agents commonly used totreat osteosarcoma, such as cisplatin (33).Additionally, BAX and ICAM2 genes were differentially expressed

in Huvos I/II tumors. A decrease in the proapoptotic BAX gene(P = 0.002) is observed in Huvos I/II tumor samples. ICAM2(P = 0.029) gene was increased in Huvos I/II tumors. ICAM2 hasbeen shown to mediate a block in apoptosis through thephosphatidyl inositol-3-kinase pathway (34).A number of oncogenic factors were increased in Huvos I/II

tumors. RAB4B (P = 0.016), a member of the RAS oncogene family,was highly increased in poor-prognosis samples ( fold increase of5.7). Additionally, pleiotrophin (P = 0.024) expression was increased

Figure 3. Multigenic mediation of poor histologicresponse to chemotherapy. This schematic modeldepicts a hypothesis underlying chemotherapyresistance in osteosarcoma as mediated byseveral genes differentially expressed betweenHuvos I/II and III/IV tumors. Poor-prognosistumors might progress and develop a malignantphenotype through the increased expression ofoncogenic factors RAB4B, myc , and pleiotrophinand a decreased expression of proapoptotic BAX.The elevated expression of migration and invasiongenes could cooperatively function to alter ECMmicroenvironment. The decrease in OPG andincrease in bone resorption genes could mediatea drive toward osteoclastogenesis. Thereafter,the destruction of bone matrix might release storedgrowth factors, such as TGF-h, which establisha positive feedback toward tumor progressionand ECM remodeling. This model representsa starting point for further hypothesis testingto abrogate the tumor in resistant cases.



in poor-prognosis samples. Pleiotrophin has been implicated in themalignant phenotype of breast and pancreatic cancer (35–37).Additionally, the v-myc oncogene (P = 0.034) and the myc-associated zinc factor (P = 0.006) were increased in Huvos I/IIsamples by fold increases of 2.43 and 1.82, respectively. Ribosomalprotein S2 (P = 0.04), s23 (P = 0.008), L12 (P = 0.014), s27 (P = 0.008),L31 (P = 0.011), L28 (P = 0.008), and s14 (P = 0.007) were up-regulated in poor-prognosis osteosarcoma samples. These ribo-somal proteins further implicate the myc oncogene, as theseproteins are regulated by myc expression.

Summary

This study has defined a set of 104 genes that classify poorhistologic response to chemotherapy in osteosarcoma tumors atthe time of diagnosis and validate these findings in a smallindependent cohort of primary and xenograft tumors. Whilespeculative, we hypothesize that chemotherapy-resistant osteo-sarcoma tumors mediate tumor growth and survival through

altered proteolytic mechanisms that function in a multigenicfashion to mediate osteoclast activation, tumor survival, andmodified ECM environment. Although further investigation isrequired to validate this proposed model, the defined differen-tially expressed genes raise the possibility that the use ofbisphosphonate analogues should be considered a testablehypothesis as potential therapeutic interventions to suppressthe bone remodeling and tumor osteolysis involved in osteosar-coma chemotherapy resistance.

Acknowledgments

Received 7/28/2004; revised 11/15/2004; accepted 12/7/2004.Grant support: National Cancer Institute grant 7 R21 CA095618-02 (D.A. Stephan).The costs of publication of this article were defrayed in part by the payment of page

charges. This article must therefore be hereby marked advertisement in accordancewith 18 U.S.C. Section 1734 solely to indicate this fact.

We thank the families and patients who participated in this study, and Jill Lemnaand Theresa Teslovich for their assistance in project support and manuscriptpreparation.

References1. Meyers PA, Gorlick R. Osteosarcoma. Pediatr ClinNorth Am 1997;44:973–89.

2. Rosen G, Caparros B, Huvos AG, et al. Preoperativechemotherapy for osteogenic sarcoma: selection ofpostoperative adjuvant chemotherapy based on theresponse of the primary tumor to preoperativechemotherapy. Cancer 1982;49:1221–30.

3. Meyers PA, Heller G, Healey J, et al. Chemotherapyfor nonmetastatic osteogenic sarcoma: the Memo-rial Sloan-Kettering experience. J Clin Oncol 1992;10:5–15.

4. Rosen G, Marcove RC, Huvos AG, et al. Primaryosteogenic sarcoma: eight-year experience with adju-vant chemotherapy. J Cancer Res Clin Oncol 1983;106:55–67.

5. Bacci G, Forni C, Ferrari S, et al. Neoadjuvantchemotherapy for osteosarcoma of the extremity:intensification of preoperative treatment does notincrease the rate of good histologic response to theprimary tumor or improve the final outcome. J PediatrHematol Oncol 2003;25:845–53.

6. Gorlick R, Meyers PA. Osteosarcoma necrosis follow-ing chemotherapy: innate biology versus treatment-specific. J Pediatr Hematol Oncol 2003;25:840–1.

7. Smeland S, Muller C, Alvegard TA, et al. ScandinavianSarcoma Group Osteosarcoma Study SSG VIII: prog-nostic factors for outcome and the role of replacementsalvage chemotherapy for poor histological responders.Eur J Cancer 2003;39:488–94.

8. MacDonald TJ, Brown KM, LaFleur B, et al. Expressionprofiling of medulloblastoma: PDGFRA and the RAS/MAPK pathway as therapeutic targets for metastaticdisease. Nat Genet 2001;29:143–52.

9. Simonet WS, Lacey DL, Dunstan CR, et al. Osteopro-tegerin: a novel secreted protein involved in theregulation of bone density. Cell 1997;89:309–19.

10. Chirgwin JM, Guise TA. Molecular mechanisms oftumor-bone interactions in osteolytic metastases. CritRev Eukaryot Gene Expr 2000;10:159–78.

11. Kang Y, Siegel PM, Shu W, et al. A multigenicprogram mediating breast cancer metastasis to bone.Cancer Cell 2003;3:537–49.

12. Barille-Nion S, Bataille R. New insights in myeloma-induced osteolysis. Leuk Lymphoma 2003;44:1463–7.

13. Guise TA. Molecular mechanisms of osteolytic bonemetastases. Cancer 2000;88:2892–8.

14. Kakonen SM, Selander KS, Chirgwin JM, et al.Transforming growth factor-h stimulates parathy-roid hormone-related protein and osteolytic metas-tases via Smad and mitogen-activated proteinkinase signaling pathways. J Biol Chem 2002;277:24571–8.

15. Takahashi S, Reddy SV, Chirgwin JM, et al.Cloning and identification of annexin II as anautocrine/paracrine factor that increases osteoclastformation and bone resorption. J Biol Chem 1994;269:28696–701.

16. Miyaura C, Inada M, Matsumoto C, et al. An essentialrole of cytosolic phospholipase A2a in prostaglandin E2-mediated bone resorption associated with inflamma-tion. J Exp Med 2003;197:1303–10.

17. Laye JP, Gill JH. Phospholipase A2 expression intumours: a target for therapeutic intervention? DrugDiscov Today 2003;8:710–6.

18. Krieger NS, Bushinsky DA, Frick KK. Cellularmechanisms of bone resorption induced by metabolicacidosis. Semin Dial 2003;16:463–6.

19. Ohshiba T, Miyaura C, Ito A. Role of prostaglandin Eproduced by osteoblasts in osteolysis due to bonemetastasis. Biochem Biophys Res Commun 2003;300:957–64.

20. Okada Y, Pilbeam C, Raisz L, Tanaka Y. Role ofcyclooxygenase-2 in bone resorption. J Uoeh 2003;25:185–95.

21. Brown JE, Neville-Webbe H, Coleman RE. The role ofbisphosphonates in breast and prostate cancers. EndocrRelat Cancer 2004;11:207–24.

22. Sonnemann J, Eckervogt V, Truckenbrod B, Boos J,Winkelmann W, van Valen F. The bisphosphonatepamidronate is a potent inhibitor of human osteosar-coma cell growth in vitro . Anticancer Drugs2001;12:459–65.

23. Mackie PS, Fisher JL, Zhou H, Choong PF. Bisphosph-onates regulate cell growth and gene expression in theUMR 106-01 clonal rat osteosarcoma cell line. Br JCancer 2001;84:951–8.

24. Evdokiou A, Labrinidis A, Bouralexis S, Hay S,Findlay DM. Induction of cell death of humanosteogenic sarcoma cells by zoledronic acid resemblesanoikis. Bone 2003;33:216–28.

25. Lawler J, McHenry K, Duquette M, Derick L.Characterization of human thrombospondin-4. J BiolChem 1995;270:2809–14.

26. Newman PJ, Albelda SM. Cellular and molecularaspects of PECAM-1. Nouv Rev Fr Hematol 1992;34:S9–13.

27. Sharma B, Handler M, Eichstetter I, et al. Antisensetargeting of perlecan blocks tumor growth andangiogenesis in vivo . J Clin Invest 1998;102:1599–608.

28. Fisher LW, Heegaard AM, Vetter U, et al. Humanbiglycan gene. Putative promoter, intron-exon junc-tions, and chromosomal localization. J Biol Chem 1991;266:14371–7.

29. Girard JP, Springer TA. Cloning from purified highendothelial venule cells of hevin, a close relative of theantiadhesive extracellular matrix protein SPARC. Im-munity 1995;2:113–23.

30. Ruhrberg C, Watt FM. The plakin family: versatileorganizers of cytoskeletal architecture. Curr Opin GenetDev 1997;7:392–7.

31. Bader BL, Magin TM, Hatzfeld M, Franke WW.Amino acid sequence and gene organization ofcytokeratin no. 19, an exceptional tail-less intermediatefilament protein. Embo J 1986;5:1865–75.

32. Whitley BR, Palmieri D, Twerdi CD, Church FC.Expression of active plasminogen activator inhibitor-1 reduces cell migration and invasion in breast andgynecological cancer cells. Exp Cell Res 2004;296:151–62.

33. Endresen L, Rugstad HE. Protective function ofmetallothionein against certain anticancer agents.Experientia Suppl 1987;52:595–602.

34. Perez OD, Kinoshita S, Hitoshi Y, et al. Activation ofthe PKB/AKT pathway by ICAM-2. Immunity2001;16:51–65.

35. Zhang N, Yeh HJ, Zhong R, Li YS, Deuel TF. Adominant-negative pleiotrophin mutant introduced byhomologous recombination leads to germ-cell apoptosisin male mice. Proc Natl Acad Sci U S A 1999;96:6734–8.

36. Souttou B, Juhl H, Hackenbruck J, et al. Relationshipbetween serum concentrations of the growth factorpleiotrophin and pleiotrophin-positive tumors. J NatlCancer Inst 1998;90:1468–73.

37. Weber D, Klomp HJ, Czubayko F, Wellstein A, Juhl H.Pleiotrophin can be rate-limiting for pancreatic cancercell growth. Cancer Res 2000;60:5284–8.

Cancer Research


an expression signature classifies chemotherapy-resistant pediatric osteosarcoma

Documents