Prediction of High Risk Ewing's Sarcoma by Gene Expression Profiling

Prediction of high risk Ewing’s sarcoma by gene expression proﬁling

Anat Ohali1, Smadar Avigad*,1,2, Rina Zaizov1,2, Ron Ophir3, Shirley Horn-Saban3, Ian J Cohen2, Isaac Meller4, Yehuda Kollender4, Josephine Issakov4 and Isaac Yaniv1,2

1Molecular Oncology, Felsenstein Medical Research Center, Rabin Medical Campus, 39 Jabotinski Street, Petah Tikva 49100, Israel and Sackler Faculty of Medicine, Tel Aviv University, POB 39040, Tel Aviv 69978, Israel; 2Pediatric Hematology Oncology, Schneider Children’s Medical Center of Israel, 14 Kaplan Street, Petah Tikva 49202, Israel and Sackler Faculty of Medicine, Tel Aviv University, POB 39040, Tel Aviv 69978, Israel; 3DNA Array Unit, Bioinformatics, Department of Biological Services, Weizmann Institute of Science, POB 26, Rehovot 76100, Israel; 4Sourasky Medical Center, 6 Weizmann Street, Tel Aviv 64239, Israel and Sackler Faculty of Medicine, Tel Aviv University, POB 39040, Tel Aviv 69978, Israel

Ewing’s sarcoma (ES) is the second most common cents and it belongs to a group of neuroectodermal primary malignant bone tumor in children and adoles- tumors known as Ewing’s Sarcoma Family of Tumors cents. Currently accepted clinical prognostic factors fail (EFT). This is an aggressive tumor with a high to classify ES patients’ risk to relapse at diagnosis. We propensity for recurrence and distant metastases (Gins- aimed to find a new strategy to distinguish between poor berg et al., 2002). All EFT share specific translocations and good prognosis ES patients already at diagnosis. We resulting in the fusion of the EWS gene on chromosome analysed the gene expression profiles of 14 primary tumor 22q12, a member of the TET family, with different ETS specimens and six metastases from ES patients, using oncogenes on different chromosomes; the most frequent oligonucleotide microarray analysis. The over-expression (B90%) is FLI1 on chromosome 11 (Burchill, 2003). of two genes was validated by quantitative PCR using the These translocations are considered distinct diagnostic LightCycler system. We identified two distinct gene features of ES tumors. An interesting finding was expression signatures distinguishing high-risk ES patients recently reported of Ewing tumors, which showed no that are likely to progress from low-risk ES patients with evidence of an EWS gene rearrangement, but instead a favorable prognosis of long-term progression-free contained translocations involving the FUS gene at survival. The microarray-based classification was superior 16p11, another member of the TET family, with the to currently used prognostic parameters. Over-expressed ETS oncogene ERG at 21q22 (Shing et al., 2003). The genes in the poor prognosis patients included genes primary site of the tumor and initial response to therapy, regulating the cell cycle and genes associated with assessed histologically as the degree of tumor necrosis invasion and metastasis, while among the downregulated following surgery, have become acceptable valid prog- genes were tumor suppressor genes and inducers of nostic factors in localized tumors. In spite of advances in apoptosis. Our results indicate the existence of a specific multimodal therapy, including combination of aggres- gene expression signature of outcome in ES already at sive chemotherapy, radiotherapy and surgery, about diagnosis, and provide a strategy to select patients who 50% of patients eventually relapse, even after 5 years would benefit from risk-adapted improved therapy. (Terrier et al., 1996). Current clinical and biological Oncogene (2004) 23, 8997–9006. doi:10.1038/sj.onc.1208060 characteristics fail to classify accurately ES patients Published online 27 September 2004 according to their clinical behavior, and it is therefore essential to search for novel reliable prognostic para- Keywords: Ewing’s sarcoma; high risk; gene expression meters, already at diagnosis. The recent development of signature; prediction; prognosis DNA microarrays provides an opportunity to take a genome wide approach to extend biological insights into the disease. Gene expression profiling using oligonucleo- ONCOGENOMICS tide high-density arrays has provided an additional tool for elucidating tumor biology as well as the potential for Introduction molecular classification of cancer (Khan et al., 2001; Nielsen et al., 2002; Yeoh et al., 2002). Ewing’s sarcoma (ES) is the second most common In this study, we used microarray analysis on primary primary malignant bone tumor in children and adoles- tumors from localized, nonmetastatic ES patients, and applied supervised classification to identify a gene *Correspondence: S Avigad, Laboratory of Molecular Oncology, expression profile that could predict risk to relapse. Pediatric Hematology Oncology, Schneider Children’s Medical Center Our results indicate the existence of a distinct gene of Israel, 14 Kaplan Street, Petah Tikva 49202, Israel; E-mail: [email protected] expression signature of outcome in ES, already at Received 9 February 2004; revised 25 May 2004; accepted 21 July 2004; diagnosis and provide a strategy to select patients who published online 27 September 2004 would benefit from risk-adapted novel therapy. Gene expression profiling of Ewing’s sarcoma A Ohali et al 8998 Results 818 genes differentially expressed in either the HR or the LR groups (t-test; Po0.01). Cluster analysis of gene expression profile In order to control false positive result as a consequence of multiple comparisons, we adjusted the The study included 14 tumor samples from localized ES P-values using false discovery rate (FDR) method patients. All samples harbored the EWS/FLI-1 chimeric (Benjamini and Hochberg, 1995). transcript. We compared the gene expression profile of Using hierarchical clustering for prognosis profile, seven tumors from patients who had tumor progression two distinct clusters could be determined: poor and between 5 months up to 5 years from diagnosis (defined good prognosis signatures (Figure 1). All of the seven as high risk – HR) with seven tumors from patients who HR and six out of the seven LR patients (86%) were were disease free for a long period of follow-up (median classified as poor and good prognosis signatures, 92 months; range 66–171) (defined as low risk – LR). respectively (Table 1). One clinically LR patient who RNA was isolated from each tumor and hybridized to was disease free for a long period of follow-up (97 Affymetrix oligonucleotide high-density arrays U95Av2. months) was classified in the poor prognosis signature We identified a subset of genes that distinguish between group. The Kaplan–Meier life table analysis indicated the two groups (HR and LR) by two steps. First, we that the patients predicted to have a good prognosis selected for 8098 genes that were expressed in one of the signature had a significantly improved progression-free groups, in at least three samples, and then we focused on survival (PFS) compared with those predicted to have a

Figure 1 Hierarchical clustering of ES tumor samples. Illustration of the two-sided clusters dendogram, distinctly deﬁning poor prognosis vs good prognosis groups of ES patients and the differentially expressed genes. Each column represents a patient and each row represents a gene. Overexpressed genes are depicted in red and downregulated genes are depicted in blue. Tumor sample numbers are marked on the x-axis, HR – high risk, LR – low risk, M – metastaes

Oncogene Gene expression profiling of Ewing’s sarcoma A Ohali et al 8999 Table 1 Clinical data, disease course and results of molecular classification Sample Age (years) Primary site Response to therapy Relapse (months) Outcome (months) Microarray classification % necrosis prognosis group

High Risk SA3 21 Pelvis o90 Local (5) EX (7) Poor SA37 7 Cranium ND Local (29) EX (44) Poor SA38 17 Pelvis o90 Local (10) EX (18) Poor SA47 20 Pelvis >90 Cranium (61) AWD (76) Poor SA75 18 Pelvis o90 Local (27) EX (49) Poor SA78 24 Femur o90 Lung (47) EX (65) Poor SA79 12 Pelvis >90 Bone (41) EX (60) Poor

Low Risk SA2 15 Pelvis >90 — NED (103) Poor SA4 14 Chest ND — NED (92) Good SA5 13 Radius o90 — NED (66) Good SA9 13 Tibia >90 — NED (168) Good SA80 15 Pelvis >90 — NED (81) Good SA81 14 Pelvis >90 — NED (82) Good SA82 11 Tibia >90 — NED (173) Good

Metastases SA43 7 Cranium ND Local (29) EX (44) Poor SA44 27 Femur >90 Lung (61) NED (91) Good SA45 16 Femur o90 Brain (128) AWD (151) Poor SA46 16 Femur o90 Lung (67) AWD (151) Poor SA76 20 Pelvis o90 Lung (24) EX (44) Poor SA77 8 Pelvis o90 Local (37) EX (104) Good

EX ¼ expired; NED ¼ no evidence of disease; AWD ¼ alive with disease; numbers in brackets ¼ time from diagnosis; ND ¼ not done.

classify metastatic tissues to one of the prognostic groups, or as a distinct group. We assumed that the metastases gene expression profile will not differ from the one observed for the primary tumors. Indeed, the expression pattern of the primary tumor and the metastases were indistinguishable. Four metastases were identified as having a poor prognosis signature and two were classified as good prognosis signature (Figure 3).

Subclassification of differentially expressed genes Additionally, we reordered the genes into two major clusters by performing hierarchical clustering of all signature genes. The two major groups correspond to overexpressed in the poor prognosis group and downregulated in the good prognosis group, and vice versa. These clusters could be further subdivided into six Figure 2 Kaplan–Meier PFS analysis presents a significant subclusters, corresponding to the variability of genes correlation between poor prognosis vs good prognosis patients among the poor vs favorable prognosis signature according to the microarray classification, and outcome patients, which was more considerable in the good prognosis group (Figure 4). Downregulated genes in the poor prognosis patients (three top subclusters) included poor prognosis signature (100 vs 12.5%, respectively) tumor suppressor genes like FHIT, NEURL, LLGL1; (Figure 2, P ¼ 0.002). inducers of apoptosis like TGFB1, TNFRSF12, CASP10; DNA repair genes: IGHMBP2, XRCC2; Unsupervised analysis of metastases immune response genes: IL-2, IL1RL1, HLA-DOB; genes involved in cytoskeleton organization like We further tested six metastases, obtained from patients MYO1C and COL6A1; cell adhesion molecules: CDH- with localized disease who had tumor progression, using 2, ITGA2B, SCAM-1, ADAM15, ADAM19, ISLR; the the unsupervised learning methodology, whether the estrogen receptor (ESR2), a signal transduction mole- poor and good prognosis signature set of genes can cule, and others (Table 2).

Oncogene Gene expression proﬁling of Ewing’s sarcoma A Ohali et al 9000

Figure 3 Gene expression proﬁle of all ES samples including primary tumors (numbers in black) and metastases samples (numbers in red), HR – high risk, LR – low risk, M – metastaes. The expression pattern of the primary tumors and the metastases were indistinguishable

Among the overexpressed genes in the poor prognosis Examination of the MTA1 protein suggests that it is a patients (subclusters 4–6) were known markers of ES, histone deacetylase and may serve multiple functions in like EWS breakpoint region 1 and beta 2 microglobulin, cellular signaling, chromosome remodeling and tran- genes regulating the cell cycle like CDK2, E2F, RAF and scription processes that are important in the progres- MAPKs, and genes associated with invasion and sion, invasion and growth of metastatic cells (Nicolson metastasis like cadherin-11 and MTA1 (Table 3). et al., 2003). To validate the microarray data, these two over- Validation of overexpressed genes by RQ–PCR expressed genes were analysed in further detail using reverse transcriptase–quantitative real-time PCR (RQ– Two genes that were significantly overexpressed in the PCR). Microarray-based expression and RQ–PCR-based poor prognosis signature group (Po0.01) attracted a expression data correlated significantly (Figure 5a and b). particular attention; both are associated with invasion The mean log expression value of the poor prognosis and metastasis. The first one is cadherin-11 (OB-cadherin), signature group is significantly higher than that of the a homophilic calcium-dependent cell adhesion molecule, good prognosis signature group for both genes, cadherin- and the second is MTA1, tumor metastasis-associated 11 and MTA1, P ¼ 0.024 and P ¼ 0.003, respectively. gene. Cadherins modulate calcium ion-dependent cell–cell adhesion and are important in cell aggregation, migration and sorting (Takeichi, 1991). Defective cell–cell and cell– Discussion matrix adhesion are among the hallmarks of cancer. The MTA1 gene is a novel, highly conserved gene that In this study, we report the determination of high- encodes a nuclear protein product (Toh et al., 1994). risk ES patients and prediction of outcome using

Oncogene Gene expression proﬁling of Ewing’s sarcoma A Ohali et al 9001

Figure 4 The two major gene clusters and the six subclusters formed on the basis of hierarchical clustering of all signature genes. The two gene clusters consist of differentially expressed genes: overexpressed in the poor prognosis group and downregulated in the good prognosis group, and vice versa. Representative genes of each subcluster are shown oligonucleotide high-density array gene expression data agreement with the observations reported recently for with a supervised learning method. We identified gene breast cancer (van’t Veer et al., 2002) indicating that the expression signatures that distinguished HR ES patients ability to metastasize to distant sites is an early and that are likely to progress up to 5 years, from LR inherent genetic property – already present at diagnosis. patients with a favorable prognosis of long-term PFS, The report by Ramaswamy et al. (2003) further between 5 and 14 years. The microarray-based classifi- strengthens this hypothesis. They have detected a cation of distinct risk groups was superior to currently molecular program of metastasis that is shared by used important prognostic parameters (Table 1). Our multiple solid tumor types, and are present in the results indicate that ES outcome can already be derived primary tumor. These findings argue against the widely from the gene expression profile of the primary tumor, accepted previous theory that metastatic potential is early at diagnosis. acquired relatively late during multistep tumorigenesis Our results are compatible with recent reports (Bernards and Weinberg, 2002). indicating the ability to predict outcome, based on gene One of the most common and important mechanisms expression profile of the malignancy at diagnosis (Shipp in the transformation of normal cells into malignant et al., 2002; van de Vijver et al., 2002; Yeoh et al., 2002). cells is the inactivation of one or more tumor suppressor In pediatric acute lymphoblastic leukemia, the gene genes. expression profile identified prognostic leukemia sub- The fragile histidine triad (FHIT) gene is a tumor types, and identified patients who would eventually fail suppressor gene that belongs to the histidine triad family treatment (Yeoh et al., 2002). A ‘poor prognosis’ of nucleoside-binding proteins. Numerous studies have signature was identified in breast cancer patients with indicated that FHIT gene expression is often altered in a short interval to develop distant metastases (van de tumor cells from many malignancies, and successful Vijver et al., 2002), and the 5-year overall survival rates FHIT gene therapy was performed in mouse models, in differed between the two groups of diffuse large B-cell lung carcinoma and in tumor cell cultures (Fouts et al., lymphoma patients (Shipp et al., 2002). Our data is in 2003). In this study, we report for the first time the

Oncogene Gene expression proﬁling of Ewing’s sarcoma A Ohali et al 9002 Table 2 Selected down-regulated genes in the microarray-based poor prognosis signature ES patients Gene symbol Gene name GeneBank ID P-value

Tumor suppressor genes NEURL Neuralized-like AF029729 0.008745 FHIT Fragile histidine triad gene U46922 0.002909 LLGL1 Lethal giant larvae homolog 1 X86371 0.001259 NF1 Neuroﬁbromin 1 (neuroﬁbromatosis) D12625 0.008977

Induction of apoptosis TNFRSF12 Tumor necrosis factor receptor U83598 0.001977 Superfamily member TGFB1 Transforming growth factor, b1 X02812 0.004213 CASP10 Caspase 10, apoptosis-related cysteine protease U60519 0.005822 TP63 Tumor protein TP63 Y16961 0.001205

DNA repair IGHMBP2 Immunoglobulin m-binding protein 2 L14754 0.003300 XRCC2 X-ray repair complementing defective repair in Chinese hamster cells Y08837 0.001514 ERCC2 Excision repair cross-complementing AA079018 0.001977

Histogenesis and organogenesis PAX8 Paired box gene 8 X69699 0.005737 PAX6 Paired box gene 6 M93650 0.003006 COL6A1 Collagen, type VI, alpha 1 AA885106 0.001771

Cytoskeleton and muscle development BMP10 Bone morphogenic protein 10 AF101441 0.009646 MYO1CMyosin cytoskeleton 1C,actin X98507 0.003714 ARCActivity regulated cytoskeleton-associated protein D87468 0.009500

Neurogenesis NEUROD2 Neurogenic differentiation 2 AB021742 0.001542 NTRK2 Neurotrophic tyrosine-kinase U12140 0.007405 Receptor type 2 GRIN2A Glutamate receptor, ionotropic U09002 0.001287 N-methyl-D-aspartate 2A SIM2 Single-minded homolog 2 U80456 0.006273

Signal transduction ESR2 Estrogen receptor 2 (ER b) X99101 0.006273 ING1L Inhibitor of growth family 1-like AI186701 0.008265

downregulation of FHIT in ES, and its differential repressor of estrogen receptor activity (REA), in the expression between poor and good prognosis patients. poor prognosis patients. These observations might These findings may propose FHIT as a potential target imply of an involvement of the estrogen receptor for therapeutic intervention. pathway in the tumorigenesis of ES, similar to breast The EWS-FLI1 oncogene was found to repress cancer. expression of TGFbRII and may account for decreased Cell adhesion molecules are thought to play a TGF-b responsiveness. Moreover, other EWS fusion significant role not only in maintaining tissue architec- genes, such as EWS-ERG and EWS-ETV1, also repress ture, but also in tumor progression, which includes TGF-bRII expression (Im et al., 2000). This data along change in morphology, invasion and metastasis. Dis- with the fact that TGF-bI was downregulated in the ruption of the cadherin–catenin complex and the loss of poor prognosis patients and significantly differentiated the E-cadherin expression has been demonstrated in between the two groups, may suggest that inactivation carcinomas arising in several tissues including prostate of TGF-b may be an important step in ES tumorigenesis (Bussemakers et al., 2000), gastric (Shibata et al., 1996) and associated with a more aggressive disease. and breast carcinomas (Pishvaian et al., 1999), and has Estrogen receptor b (ER b) expression was observed been correlated with various pathologic and clinical to be significantly decreased in breast cancer and features, such as tumor differentiation, proliferation and metastatic lymph node tissues compared with normal a poor patient prognosis (Hajra and Fearon, 2002). mammary and benign breast tumors (Park et al., 2003). N-cadherin and cadherin-11 are expressed during An inverse relationship was found between ER b embryonic development in mesenchymal cells and are mRNA level and both histologic grade and progesterone believed to participate in chondro-osteogenic cell con- receptor expression. The same results were presented in densation in somites and limb buds (Oberlender and our study, significantly decreased expression of ER b Tuan, 1994). The distribution of these two cadherins and overexpression of progesterone receptor and of a largely overlaps during embryogenesis, although cadher-

Oncogene Gene expression proﬁling of Ewing’s sarcoma A Ohali et al 9003 Table 3 Selected over-expressed genes in the microarray-based poor prognosis signature ES patients Gene symbol Gene name GeneBank ID P-value

Cell cycle control CDK2 Cyclin-dependent kinase 2 M68520 0.001542 ID1 Inhibitor of DNA binding 1 X77956 0.006313 CREB2 cAMP responsive element binding 2 AF039081 0.008076 RAF 1 v-raf-1 murine leukemia viral oncogene X06409 0.007728 DMTF1 Cyclin D-binding myb-like transcription factor 1 AF052102 0.002852 E2F3 E2F transcription factor 3 D38550 0.009677 E2F5 E2F transcription factor 5, p130 U31556 0.002985

Signal transduction MAPK9 Mitogen-activated protein kinase 9 U09759 0.001344 MKNK1 MAP kinase-interacting serine/threonine AB000409 0.001259 Kinase 1 AXL AXL receptor protein tyrosine kinase M76125 0.003556 PDAP1 PDGFA-assiciated protein 1 U41745 0.001839 PTK7 PTK7 protein tyrosine kinase 7 U33635

Transcription factors ZNF175 Zinc-ﬁnger protein 175 D50419 0.006313 EP300 E1A-binding protein p300 U01877 0.002852 ZFP36L2 Zinc-ﬁnger protein 36L2 U07802 0.008936 TAF6 TAF6 RNA polymerase II, TATA box L25444 0.001205 Binding protein TCFL4 Transcription factor-like 4 AW00599 0.002399 EIF4A2 Eucaryotic translation initiation factor 4A D30655 0.005664

Members of RAS oncogene family RAB2 RAB2, member of RAS oncogene family M28213 0.006313 RAB1A RAB1A, member of RAS oncogene M28209 0.009319 RHEB2 RAS homolog , enriched in brain 2 D78132 0.009826

Cell adhesion CDH11 Cadherin11, type 2, OB-cadherin D21255 0.007121 ITGAE Integrin a E (antigen CD103) L25851 0.002714 ITGB2 Integrin b 2 (antigen CD18) M15395 0.006331

Invasion and metastasis MTA1 Metastasis-associated 1 U35113 0.009341 HBXIP Hepatitis B virus X interacting protein AF029890 0.002852

Figure 5 Correlation between expression of the cadherin-11 and the MTA1 genes by microarray analysis and by real-time PCR. (a) Expression mean log value of cadherin-11 in poor prognosis patients was signiﬁcantly higher than the expression mean log value in good prognosis patients by both analyses. (b) Gene expression pattern in the poor and good prognosis patients was also signiﬁcantly correlated by both analyses, for the MTA1 gene

Oncogene Gene expression profiling of Ewing’s sarcoma A Ohali et al 9004 in-11 is particularly abundant in areas undergoing favorable vs unfavorable ES patients, regardless of chondrogenesis. Cadherin-11 expression is downregu- clinical parameters. This study should be the basis for an lated in differentiated structures (Shin et al., 2000). extended investigation of ES tumors, which may lead to Cadherin-11 was highly expressed in embryonal and the development of an important prognostic tool. alveolar rhabdomyosarcoma, while it was downregu- Our data imply that classification of patients into lated in normal skeletal muscle. This data suggest that high- and low-risk subgroups may be useful in selecting cadherin-11 might be involved in myogenesis and that ES patients who would benefit from early intensive or rhabdomyosarcoma may re-express or fail to down- reduction of systemic adjuvant treatment, reducing regulate cadherin-11 (Markus et al., 1999). Similarly, in overtreatment associated with undesirable side effects Wilms’ tumors, there was a strong expression of and augment undertreatment. Furthermore, genes that cadherin-11, and in several cases it was inversely are overexpressed in tumors with a poor prognosis correlated with the expression of E-cadherin (Schulz profile are potential targets for the development of new et al., 2000). rational cancer therapy. In our microarray-based analysis, overexpression of cadherin-11 on one hand, and downregulation of N- cadherin, on the other hand, discriminated between HR Materials and methods and LR ES patients. The overexpression of cadherin-11 was validated by RQ–PCR. Based on our results and on Patient samples the observations reported for rhabdomyosarcoma and A total of 14 primary tumor specimens and six metastases were Wilms’ tumor, we propose that in tumors originated obtained from 18 ES patients with nonmetastatic disease. Of from mesenchymal cells, N-cadherin functions as the E- one patient, both primary and recurrent tumors were analysed cadherin in epithelial cells, and therefore the down- (SA37 and SA43), and two metastases were taken from regulation of N-cadherin and overexpression of cadher- another patient, six years apart (SA45 and SA46). All patients in-11 facilitate the invasive properties of sarcoma cells. were admitted to the Pediatric Hematology Oncology Depart- It was recently found, by Mahoney et al. (2002) that ment at Schneider Children’s Medical Center. Informed MTA1 gene expression is associated with migration and consent was obtained from the patients or their guardians, invasion and it is involved in the metastatic process. The and the local and National Ethics Committees approved the MTA1 gene was found to be overexpressed in a variety research project. All patients were treated with a combination of aggressive chemotherapy, radiotherapy and surgery. The of cancerous tissues including breast, esophageal, color- median age at diagnosis was 15 years (range 7–27). Five ectal, gastric and pancreatic carcinomas (Toh et al., patients were female and 13 were male subjects. Response to 1997, 1999; Iguchi et al., 2000; Nawa et al., 2000a). The therapy was defined by histopathological response and MTA1 protein is likely a nuclear regulatory protein, and assessed by percentage of tumor necrosis at the time of surgery it might interact with specific genes involved in cellular (limb salvage procedure) following neoadjuvant chemotherapy regulation (Nicolson et al., 2003). Antisense oligonu- and radiotherapy. The median follow-up was 72.5 months cleotide treatment of breast cancer cell lines that showed (range 7–171). Tumors were snap-frozen in liquid nitrogen high levels of expression of the MTA1 gene inhibited the immediately after surgery and stored at À801Cuntil use. cell growth and in vitro invasion (Nawa et al., 2000b). Therefore, MTA1 gene might be used as a therapeutic Microarray hybridization target. It was also reported that MTA1 represses ER Total RNA (10 mg) was extracted from each tumor using Tri transcription by recruiting histone deacetylases to the Reagent (Molecular Research Center, Inc. Cincinnati, OH, estrogen receptor element (ERE)-containing target gene USA). Double-stranded cDNA was generated using the chromatin in breast cancer cells (Mishra et al., 2003). SuperScript Choice System (Gibco Brl, Rockville, MD, Here we report, for the first time, the overexpression of USA), using an oligo(dT)24 primer containing a T7 promoter the MTA1 gene in ES. Furthermore, it was one of the site at the 30 end (Genset, La Jolla, CA, USA) and purified via genes that distinguished between HR and LR patients a phenol–chloroform extraction followed by an ethanol and its overexpression was also confirmed by RQ–PCR. precipitation. Purified cDNA was used as a template for in vitro transcription (IVT), which was performed with T7 RNA The process of invasion and metastasis is a defined polymerase and biotin-labeled ribonucleotides, using the phenotype of a malignant neoplasm and the principal ENZO BioArray High Yield RNA Transcript Labeling Kit cause of cancer-treatment failure. These findings sup- (Enzo Diagnostics, New York, NY, USA) and purified over port the emerging notion that the clinical outcome of RNeasy mini columns (Qiagen, Valencia, CA, USA) according individuals with cancer can be predicted using the gene to the manufacturer’s instructions, and the labeled cRNA was expression profiles of primary tumors (Alizadeh et al., fragmented in fragmentation buffer. A mixture of four control 2000; Van De Vijver et al., 2002). Thus, it was not bacterial and phage cRNA was included to serve as an internal surprising that 67% of the metastases were classified in control for hybridization efficiency. the poor prognosis signature group, indistinguishable In total, 12 mg cRNA of each sample were hybridized to a from the primary tumors, since the ‘metastasis profile’ is Genechip Human Genome U95Av2 array (Affymetrix, Santa Clara, CA, USA). After hybridization, each array was washed present already at diagnosis. according to procedures developed by the manufacturer We are aware that our report consists of a small (Affymetrix), and stained with streptavidin–phycoerythrin sample size, even so, the highly significant results conjugate (Molecular Probes, Eugene, OR, USA). The distinguishing the two clinical prognostic groups are hybridization signal was amplified by using biotinylated remarkable. The microarray analysis could distinguish antistreptavidin antibodies (Vector Laboratories, Burlingame,

Oncogene Gene expression profiling of Ewing’s sarcoma A Ohali et al 9005 CA, USA), followed by restaining with streptavidin–phycoer- Quantitative real-time PCR (RQ–PCR) The microarray- ythrin. The expression value for each gene was determined by derived expression data was evaluated for the cadherin-11 calculating the average differences of the probe pairs in use for and MTA1 genes using quantitative PCR by the LightCycler that gene. system (Roche Diagnostics, Mannheim, Germany). In all, 3 mg Two samples were analysed in duplicate and results were of total RNA was processed to cDNA by the Reverse reproducible. Transcription System (Promega Corporation, Madison, WI, USA) according to the manufacturer’s protocol. The cDNA Data analysis was purified with GFX PCR DNA and Gel Band Purification kit (Amersham Biosciences, Piscataway, NJ, USA). Normalization and filtering The microarray results were Double-stranded cDNA (5 ml) was amplified in a 20 ml analysed using the GeneSpring Softwares. Normalization was reaction containing 4 mM MgCL2, 10 mM of each primer and performed by setting expression values lower than zero to zero LightCycler – FastStart DNA Master SYBR Green I mix and then each measurement was divided by the median of all (Roche Diagnostics). Gene expression of the house-keeping measurements in that sample. gene porphobilinogen deaminase (PBGD) was used as a In order to filter out genes that are not expressed in any of control. The mean log expression of poor prognosis samples the groups, we used Affymetrix absolute call (MAS 4.0: P, M – was compared with the mean log expression of good prognosis expressed genes, A – not expressed). Genes that were expressed samples by Student’s t-test. The primers used for Cadherin-11 in one group were defined as genes expressed in at least three gene amplification were: sense 50-AGAGGCCTACATTCT samples. GAACG-30 and antisense 50-TTCTTTCTTTTGCCTTCTC AGG-30, and for the MTA1 gene amplification: sense 50-AG Selecting for differentially expressed genes A Student’s t-test CTACGAGCAGCACAACGGGGT-30 and antisense 50-CAC was applied for each gene, and genes with an adjusted P-value GCTTGGTTTCCGAGGAT-30. less then 0.01 were selected as differentially expressed genes. P- All reactions were performed in duplicate. Quantitative values were corrected to reduce false positive using Benjamini analysis was performed using the LightCycler Software. The and Hochberg False Discovery Rate (Benjamini and Hoch- specificity of the PCR products was determined with the berg, 1995). LightCycler Software’s melting-curve analysis feature.

Hierarchical clustering A divisive hierarchical clustering was performed as described by Eisen et al. (1998), using centered Acknowledgements correlation as measurment distance. This work was supported by the Josefina Maus and Gabriela Cesarman Maus Chair for Pediatric Hematology Oncology PFS analysis The Kaplan–Meier PFS analysis, using the log (RZ). This work was performed in partial fulfillment of the rank test, was performed in order to correlate the microarray requirements for the PhD degree of Anat Ohali, Sackler classification results with patients’ clinical outcome. School of Medicine, Tel-Aviv University, Israel.

References

Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Im Y-H, Kim HT, Lee C, Poulin D, Welford S, Sorensen Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, Powell PHB, Denny CT and Kim S-J. (2000). Cancer Res., 60, JI, Yang L, Marti GE, Moore T, Hudson J, Lu L, Lewis DB, 1536–1540. Tibshirani R, Sherlock G, Chan WC, Greiner TC, Weisen- Khan J, Wei JS, Ringner M, Saal LH, Ladanyi M, burger DD, Armitage JO, Warnke R, Levy R, Wilson W, Westermann F, Berthold F, Schwab M, Antonescu Grever MR, Byrd JC, Botstein D, Brown PO and Staudt CR, Peterson C and Meltzer PS. (2001). Nat. Med., 7, LM. (2000). Nature, 403, 503–511. 673–679. Benjamini Y and Hochberg Y. (1995). J. Roy. Stat. Soc B., 57, Mahoney MG, Simpson A, Jost M, Noe´ M, Kari C, Pepe D, 289–300. Choi YW, Uitto J and Rodeck U. (2002). Oncogene, 21, Bernards R and Weinberg RA. (2002). Nature, 418, 823. 2161–2170. Burchill SA. (2003). J. Clin. Pathol., 56, 96–102. Markus MA, Reichmuth C, Atkinson MJ, Reich U, Hoff- Bussemakers MJ, Van Bokhoven A, Tomita K, Jansen CF and mann I, Balling R, Anderer U and Ho¨ ﬂer H. (1999). Schalken JA. (2000). Int. J. Cancer, 85, 446–450. J. Pathol., 187, 164–172. Eisen MB, Spellman PT, Brown PO and Botstein D. (1998). Mishra SK, Mazumdar A, Vadlamudi RK, Li F, Wang RA, Proc. Natl. Acad. Sci. USA, 95, 14863–14868. Yu W, Jordan VC, Santen RJ and Kumar R. (2003). J. Biol. Fouts RL, Sandusky GE, Zhang S, Eckert GJ, Koch MO, Chem., 278, 19209–19219. Ulbright TM, Eble JN and Cheng L. (2003). Cancer, 97, Nawa A, Nishimori K, Lin P, Maki Y, Moue K, Sawada H, 1447–1452. Toh Y, Fumitaka K and Nicolson GL. (2000a). J. Cell. Ginsberg JP, Woo SY, Johnson ME, Hick MJ and Horowitz Biochem., 79, 202–212. ME. (2002). Ewing sarcoma family of tumors: Ewing’s Nawa A, Sawada H and Toh Y. (2000b). Int. J. Med. Biol. sarcoma of bone and soft tissue and the peripheral primitive Environ., 28, 33–39. neuroectodermal tumors. Principles and Practice of Pediatric Nicolson GL, Nawa A, Toh Y, Taniguchi S, Nishimori K and Oncology, 4th ed. Pizzo PA and Poplack DG (eds), Moustafa A. (2003). Clin. Exp. Metast., 20, 19–24. Lippincott-Williams and Wilkins publishers: Philadelphia, Nielsen TO, West RB, Linn SC, Alter O, Knowling MA, Pennsylvania, pp 973–1016. O’Connell JX, Zhu S, Fero M, Sherlock G, Pollack JR, Hajra KM and Fearon ER. (2002). Genes Chromos. Cancer, Brown PO, Botstein D and van de Rijn M. (2002). Lancet, 34, 255–268. 359, 1301–1307. Iguchi H, Imura G, Toh Y and Ogata Y. (2000). Int. J. Oncol., Oberlender SA and Tuan RS. (1994). Cell. Adhes. Commun., 2, 16, 1211–1214. 521–537.

Oncogene Gene expression proﬁling of Ewing’s sarcoma A Ohali et al 9006 Park BW, Kim KS, Heo MK, Ko SS, Hong SW, Yang WI, Terrier P, Llombart-Bosch A and Contesso G. (1996). Semin. Kim JH, Kim GE and Lee KS. (2003). Breast Cancer Res. Diagn. Pathol., 13, 250–257. Treat., 80, 79–85. Toh Y, Kuwano M, Mori M, Nicolson GL and Sugimachi K. Pishvaian MJ, Feltes CM, Thompson P, Bussemakers MJ, (1999). Br. J. Cancer, 79, 1723–1726. Schalken JA and Byers SW. (1999). Cancer Res., 59, Toh Y, Oki E, Oda S, Tokunaga E, Ohno S, Maehara Y, 947–952. Nicolson GL and Sugimachi K. (1997). Int. J. Cancer, 74, Ramaswamy S, Ross KN, Lander ES and Golub TR. (2003). 459–463. Nat. Genet., 33, 49–54 , 2003. Toh Y, Pencil SD and Nicolson GL. (1994). J. Biol. Chem., Schulz S, Becker K-F, Braungart E, Reichmuth C, Klamt B, 269, 22958–22963. Becker I, Atkinson M, Gessler M and Ho¨ ﬂer H. (2000). van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AAM, J. Pathol., 191, 162–169. Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton Shibata T, Ochiai A, Gotoh M, Machinami R and Hirohashi MJ, Parrish M, Atsma D, Witteveen A, Glas A, Delahaye L, S. (1996). Cancer Lett., 99, 147–153. van der Velde T, Bartelink H, Rodenhuis S, Rutgers ET, Shin CS, Lecanda F, Sheikh S, Weitzmann L, Cheng S-L and Friend SH and Bernards R. (2002). N. Engl. J. Med., 347, Civitelli R. (2000). J. Cell. Biochem., 78, 566–577. 1999–2009. Shing DC, McMullan DJ, Roberts P, Smith K, Chin S-F, van˜ ’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AAM, Nicholson J, Tillman RM, Ramani P, Cullinane C and Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen Coleman N. (2003). Cancer Res., 63, 4568–4576. AT, Schreiber GJ, Kerkhoven RM, Roberts C, Linsley PS, Shipp MA, Ross KN, Tamayo P, Weng AP, Kutok JL, Aguiar Bernards R and Friend SH. (2002). Nature, 415, 530–536. RCT, Gaasenbeek M, Angelo M, Reich M, Pinkus GS, Ray Yeoh E-J, Ross ME, Shurtleff SA, Williams WK, Patel D, TS, Koval MA, Last KW, Norton A, Lister TA, Mesirov J, Mahfouz R, Behm FG, Raimondi SC, Relling MV, Patel A, Neuberg DS, Lander ES, Aster JCand Golub TR. (2002). Cheng C, Campana D, Wilkins D, Zhou X, Li J, Liu H, Pui Nat. Med., 8, 68–74. C-H, Evans WE, Naeve C, Wong L and Downing JR. Takeichi M. (1991). Science, 251, 1451–1455. (2002). Cancer Cell, 1, 133–143.

Oncogene