oncogenomics - Semantic Scholar

Oncogene (2004) 23, 2385–2400

& 2004 Nature Publishing Group All rights reserved 0950-9232/04 $25.00 www.nature.com/onc

Toru Nakamura1,2, Yoichi Furukawa1, Hidewaki Nakagawa1, Tatsuhiko Tsunoda3, Hiroaki Ohigashi4, Kohei Murata4, Osamu Ishikawa4, Kazuhisa Ohgaki5, Nobuichi Kashimura6, Masaki Miyamoto2, Satoshi Hirano2, Satoshi Kondo2, Hiroyuki Katoh2, Yusuke Nakamura*,1 and Toyomasa Katagiri1 1 Laboratory of Molecular Medicine, Human Genome Center, Institute of Medical Science, The University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan; 2Surgical Oncology, Cancer Medicine, Division of Cancer Medicine, Hokkaido University Graduate School of Medicine, North 15, West 7, Kita-ku, Sapporo, Hokkaido 060-8638, Japan; 3Laboratory for Medical Informatics, SNP Research Center, RIKEN (Institute of Physical and Chemical Research), 1-7-22 Suehiro-cho, Tsurumi-ku, Yokohama 230-0045, Japan; 4Department of Surgery, Osaka Medical Center for Cancer and Cardiovascular Diseases, 1-3-3 Nakamichi, Higashinari-ku, Osaka 537-8511, Japan; 5Department of Surgery, Kyoto Police Hospital, 14 Koyama Kitakamifusa-cho, Kita-ku, Kyoto 603-8142, Japan; 6Department of Surgery, Teine Keijinkai Hospital, 1-12-1-40 Maeda, Teine-ku, Sapporo, Hokkaido 006-8555, Japan

To characterize molecular mechanism involved in pancreatic carcinogenesis, we analysed gene-expression profiles of 18 pancreatic tumors using a cDNA microarray representing 23 040 genes. As pancreatic ductal adenocarcinomas usually contain a low proportion of cancer cells in the tumor mass, we prepared 95% pure populations of pancreatic cancer cells by means of laser microbeam microdissection, and compared their expression profiles to those of similarly purified, normal pancreatic ductal cells. We identified 260 genes that were commonly upregulated and 346 genes that were downregulated in pancreatic cancer cells. Because of the high degree of purity in the cell populations, a large proportion of genes that we detected as upregulated or downregulated in pancreatic cancers were different from those reported in previous studies. Comparison of clinicopathological parameters with the expression profiles indicated that altered expression of 76 genes was associated with lymph-node metastasis and that of 168 genes with liver metastasis. In addition, expression levels of 30 genes were related to the recurrence of disease. These genome-wide expression profiles should provide useful information for finding candidate genes whose products might serve as specific tumor markers and/or as molecular targets for treatment of patients with pancreatic cancer. Oncogene (2004) 23, 2385–2400. doi:10.1038/sj.onc.1207392 Published online 9 February 2004 Keywords: pancreatic cancer; laser microbeam microdissection; cDNA microarray; gene expression profiles

*Correspondence: Y Nakamura, Laboratory of Molecular Medicine, Human Genome Center, Institute of Medical Science, The University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan; E-mail: [email protected] Received 10 March 2003; revised 10 October 2003; accepted 11 November 2003

Introduction The mortality among patients with pancreatic cancer is worse than for any other kind of malignant tumor, with a 5-year survival rate only 4% (Greenlee et al., 2001). The poor prognosis of this malignancy reflects both the difficulty of early diagnosis and a generally poor response to current therapies (DiMagno et al., 1999; Greenlee et al., 2001). In particular, no specific tumor marker is clinically available for detection of this disease at an early and potentially curative stage, although CA19-9 has been used as tumor marker for pancreatic cancers. Surgical resection is the only possible cure at present, but cases that are surgically resectable at diagnosis account for fewer than 20% of patients with this cancer (DiMagno et al., 1999; Klinkenbijl et al., 1999). Endoscopic ultrasonography (EUS), endoscopic retrograde cholangiopancreography (ERCP), and spiral CT are available to screen individuals at risk for familial pancreatic cancer (Brentnall et al., 1999), but those approaches are not practical in terms of time and cost effectiveness to screen every asymptomatic individual. Hence, tumor markers that are sensitive and specific for pancreatic cancer must be discovered before we can see improvement in the prognosis of this devastating disease. Almost all patients at an advanced stage fail to respond to any treatment. To overcome that situation, some clinical trials have been attempting to establish therapeutic strategies on the basis of molecular technologies. Such trials have involved, for example, an MMP inhibitor, drugs designed to inhibit Ras farnesyltransferase, and antibody-based approaches (Rosenberg, 2000; Laheru et al., 2001; Hao and Rowinsky, 2002). However, so far, these experiments have achieved no remarkable effects on this disease.

ONCOGENOMICS

Genome-wide cDNA microarray analysis of gene expression profiles in pancreatic cancers using populations of tumor cells and normal ductal epithelial cells selected for purity by laser microdissection

Analysis of expression profiles in pancreatic cancers T Nakamura et al

2386

Some studies describing gene expression profiles of pancreatic cancers have discovered genes that might be candidates as diagnostic markers or therapeutic agents (Crnogorac-Jurcevic et al., 2002; Han et al., 2002; Iacobuzio-Donahue et al., 2002). However, data derived from tumor masses cannot adequately reflect expressional changes during pancreatic carcinogenesis, because a typical pancreatic ductal adenocarcinoma exists as a solid mass with a highly desmoplastic stromal reaction containing various cellular components. The proportions of neoplastic cells in such tissue samples are quite different from one case to another, ranging from 5 to 55%. Therefore, previously published results are likely to reflect heterogeneous expression profiles. Furthermore, since pancreatic adenocarcinomas are generated from normal pancreatic ductal epithelium, a cell type that accounts for less than 5% of pancreatic tissue, analyses that use normal human pancreas for control cells are inappropriate for investigating genes related to carcinogenesis and/or progression of cancer in that organ. With these issues in view, we prepared purified populations of cancer cells and normal ductal cells by means of laser microbeam microdissection (LMM), and analysed genome-wide gene expression profiles of 18 pancreatic tumors using a cDNA microarray representing 23 040 genes. Through the expression profiles, we identified the genes whose expression was significantly associated with some clinical parameters, such as lymph node metastasis, liver metastasis, and recurrence risk. These data should provide not only important information about pancreatic carcinogenesis, but should identify

candidate genes whose products might serve as diagnostic markers and/or as molecular targets for treatment of patients with pancreatic cancer.

Results Isolation of pancreatic cancer cells and normal pancreatic ductal epithelial cells using LMM To obtain precise expression profiles of pancreatic cancer cells, we employed LMM to avoid contamination of the samples by noncancerous cells. Since pancreatic cancer originates from pancreatic ductal cells, we used similarly purified populations of normal pancreatic ductal epithelial cells as controls. As the great majority of cells in the pancreas are acinar cells, we considered it inappropriate to use the entire pancreas to screen for genes associated with carcinogenesis in that organ. Figure 1 shows micrographs of representative cancers (a and b), and normal pancreatic duct (c and d) after microdissection. In these histological views of a welldifferentiated type (a) and a scirrhous type (b) of invasive ductal adenocarcinoma, the proportions of cancer cells were estimated to be about 30 and 10%, respectively. We estimated that the proportion of cancer cells in the LMM-purified samples used for our expression analysis was at least 95%. We also examined the proportion of acinar cells that were contaminating the microdissected population of normal pancreatic ductal epithelial cells serving as a universal control, by measuring the signal intensity of a

Figure 1 Microdissection of pancreatic cancer cells and normal pancreatic ductal epithelial cells. Panels a and b show cancer sections stained with hematoxylin and eosin prior to microdissection; panel a is a well-differentiated type and panel b is a scirrhous type. Panels a1 and b1 show the same sections after microdissection; panels a2 and b2 show the microdissected cancer cells captured on the collecting cap. Panel c shows a section of normal whole-pancreas tissue containing 490% acinar cells. Panel d shows microdissected normal pancreatic ductal epithelial cells Oncogene


2387

gene (AMY1A) that is expressed exclusively in acinar cells. When we investigated the signal intensity of this gene in whole pancreatic tissue, where 490% of the cells are acinar cells, the ratio of the average signal intensity of the pancreatic amylase gene to that of beta actin was approximately 96.7; the ratio in microdissected normal pancreatic ductal epithelial cells was approximately 0.28 (see Contamination percentage section in Materials and methods). Therefore, we estimated the average proportion of contaminating acinar cells in the populations of control cells to be 0.29% after microdissection. Filtering of data A gene expression analysis of 18 pancreatic cancers on a cDNA microarray representing approximately 23 000 genes identified 260 genes that were commonly upregulated more than fivefold over their levels in normal pancreatic ductal epithelial cells (Supplemental Table 1; also see Materials and methods). In particular, 167 of them were expressed at a level more than 10-fold higher than in normal ductal cells. Among the 260 upregulated genes, the biological functions of 197 were already known to some extent. Of them, interferoninduced transmembrane protein 1 (IFITM1), plasminogen activator, urokinase (PLAU), prostate stem cell antigen (PSCA), S100 calcium binding protein P (S100P), RNA binding motif single-stranded interacting protein 1 (RBMS1), and baculoviral IAP repeatcontaining 5 (BIRC5) had already been reported as overexpressed in pancreatic cancers (Han et al., 2002; Iacobuzio-Donahue et al., 2002). The list of upregulated elements included genes encoding transcriptional factors and proteins involved in the signal transduction pathway, in the cell cycle, and in cell adhesion (Table 1). On the other hand, we identified 346 genes whose expression ratio was reduced to less than 0.2 in pancreatic cancer cells (Supplemental Table 2). Among them, 212 had been functionally characterized to some extent. They included AXIN1 upregulated 1 (AXUD1), deleted in liver cancer 1 (DLC1), growth arrest and DNA-damage-inducible, beta (GADD45B), and P53-dependent damage-inducible nuclear protein 1 (p53DINP1), all of which have been implicated in growth suppression (Yuan et al., 1998; Satoh et al., 2000; Ishiguro et al., 2001; Okamura et al., 2001). Verification of selected genes by semiquantitative RT–PCR To validate the expression data obtained by microarray analysis, we performed semi-quantitative RT–PCR experiments for 12 genes that were strongly overexpressed in almost all informative cases: ATP1B3, ARHGDIB, APP, BIRC5, CDH3, EphA4, GYS1, KPNB2, RBMS1, REGIV, S100P, and VANGL1. The results of the RT–PCR analysis were highly concordant with microarray data in the great majority of the tested cases (Figure 2).

Identification of genes correlated with clinicopathological features Lymph node metastasis and liver metastasis In order to investigate relations between gene expression profiles and clinicopathological parameters, we searched genes that were possibly associated with lymph-node metastasis and liver metastasis, which are important determining factors of patients’ prognosis. We first examined the expression profiles and the status of lymph-node metastasis using nine lymph-node-positive and four node-negative cases, and identified 76 genes that were associated with lymph-node status by a random permutation (P-value o0.05) (Table 2). Of these, 35 genes were relatively upregulated, and 41 genes were downregulated in node-positive tumors (Figure 3). In addition, we compared expression profiles of five cases with predominant recurrence in liver with those of six cases with metastasis to other sites (local, peritoneal, and chest). We identified 168 genes that showed altered expression patterns uniquely in cases that had liver metastasis (Table 3), and 60 of them were relatively upregulated in tumors (Figure 4). These genes included some key factors that had been proposed to play crucial roles in tumor cell proliferation, invasion, and metastasis: integrin, beta 4 (ITGB4) (Shaw et al., 1997), colony stimulating factor 1 (CSF1) (Chambers et al., 1997), basigin (BSG) (Guo et al., 2000), and kinesin-like 6 (KNSL6) (Scanlan et al., 2002). Hierarchical clustering analysis using these identified gene sets was also able to clearly classify the groups with regard to lymph-node status or those with liver metastasis (Figures 3 and 4). Prognosis To further investigate genes that might be associated with prognosis, we compared expression profiles of seven cases who had recurrence within 12 months after surgery (disease-free interval o12 months; median 6.4 months) with those of six cases who had 412 months of disease-free interval (median 17.0 months). As shown in Figure 5a, we identified 84 genes that were expressed differently between these two groups using a random permutation method (Po0.05). In attempt on establishment of a predictive scoring system using gene expression pattern for recurrence after surgery, we rank ordered the above prognostic 84 candidate genes on the basis of the magnitude of their permutation P-values (Table 4) and calculated the prediction score by the leave-one-out test for crossvalidation using top 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, and 84 genes on the rank-ordered list. To determine the number of discriminating genes giving the best separation of the two groups, we calculated a classification score (CS) for each gene set (Figure 5b). As shown in Figure 5c, the best separation was obtained when we used the top 30 genes in our candidate list for scores calculation. Oncogene


2388 Table 1 Representative upregulated genes with known function in pancreatic cancers GenBank ID

Symbol

Gene name

Genes involved in signal transduction pathway AA916826 APP L20688 ARHGDIB L36645 EPHA4 J03260 GNAZ AA574178 KAI1 AA458825 MTIF2 M80482 PACE4 M16750 PIM1 X62006 PTB AA346311 RAI3 S45545 RCV1 D38583 S100A11 AA308062 S100P AF029082 SFN J03040 SPARC

Amyloid beta (A4) precursor protein (protease nexin-II, Alzheimer’s disease) Rho GDP dissociation inhibitor (GDI) beta EphA4 Guanine nucleotide-binding protein (G protein), alpha z polypeptide Kangai 1 Mitochondrial translational initiation factor 2 Paired basic amino-acid cleaving system 4 Pim oncogene Polypyrimidine tract-binding protein (heterogeneous nuclear ribonucleoprotein I) Retinoic acid-induced 3 Recoverin S100 calcium-binding protein A11 (calgizzarin) S100 calcium-binding protein P Stratifin Secreted protein, acidic, cysteine-rich (osteonectin)

Transcriptional factors AF047002 AA905901 L16783 AA418167 X92518 M16937 U24576 X13293 X64652 M81601

Transcriptional coactivator Cofactor required for Sp1 transcriptional activation, subunit 3 (130 kDa) Forkhead box M1 GATA-binding protein 3 High-mobility group (nonhistone chromosomal) protein isoform I-C Homeo box B7 LIM domain only 4 v-myb avian myeloblastosis viral oncogene homolog-like 2 RNA-binding motif, single-stranded interacting protein 1 Transcription elongation factor A (SII), 1

ALY CRSP3 FOXM1 GATA3 HMGIC HOXB7 LMO4 MYBL2 RBMS1 TCEA1

Cell adhesion and cytoskeleton AF006086 ARPC3 AA557142 CD2AP X63629 CDH3 AA977821 COL1A1 J03464 COL1A2 X14420 COL3A1 AI140851 COL6A1 U16306 CSPG2 X02761 FN1 M15395 ITGB2 J00269 KRT6A L11370 PCDH1 AA234962 PKP3

Actin-related protein 2/3 complex, subunit 3 (21 kDa) CD2-associated protein Cadherin 3, type 1, P-cadherin (placental) Collagen, type I, alpha 1 Collagen, type I, alpha 2 Collagen, type III, alpha 1 (Ehlers–Danlos syndrome type IV, autosomal dominant) Collagen, type VI, alpha 1 Chondroitin sulfate proteoglycan 2 (versican) Fibronectin 1 Integrin, beta 2 Keratin 6A Protocadherin 1 (cadherin-like 1) Plakophilin 3

Cell cycle X54941 L16783 U63743 AF044588 K02581

CDC28 protein kinase 1 Forkhead box M1 Kinesin-like 6 (mitotic centromere-associated kinesin) Protein regulator of cytokinesis 1 Thymidine kinase 1, soluble

CKS1 FOXM1 KNSL6 PRC1 TK1

Discussion The cDNA microarray is a powerful tool for identifying genes that may be applicable for development of novel diagnostic markers and molecular targets for therapeutic purposes (Ishiguro et al., 2002; Yagyu et al., 2002). A few groups already have reported gene expression profiles of pancreatic cancer (Crnogorac-Jurcevic et al., 2002; Han et al., 2002; Iacobuzio-Donahue et al., 2002; Logsdon et al., 2003), but the results reported here are quite different from the others, probably for the reasons described below. First, a microarray analysis using clinical samples is not easy because various cellular components contamOncogene

inate normal as well as cancer tissues. Pancreatic ductal adenocarcinomas are good examples in this respect because their highly desmoplastic stromal reactions result in a low proportion of cancer cells in the tumor mass. Furthermore, in the normal pancreas, acinar cells and islets account for more than 95% of this organ, and epithelial cells in the pancreatic duct, from which the carcinoma originates, correspond to a very small percentage of the total (Bockman et al., 1983; Hruban et al., 2000). Therefore, an analysis of gene expression profiles using bulk tissues from cancerous and normal whole pancreas is significantly influenced by the particular mixture of cells in the tissues examined; proportional differences of acinar cells, islet cells,


2389

Figure 2 Confirmation of reliability of the microarray data. Expression of 12 representative upregulated genes and alpha tubulin was examined by semi-quantitative RT–PCR using cDNA prepared from amplified RNA. The signals shown here correspond well to the data obtained on the microarray

Table 2 Rank Permutation P-value GenBank ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

5.73E05 0.0018 0.0021 0.0022 0.0037 0.0037 0.0041 0.0074 0.0077 0.0086 0.0099 0.0104 0.0106 0.0115 0.0116 0.0117 0.0128 0.0129 0.0137 0.0145 0.0157 0.0158 0.0159 0.0170 0.0176 0.0180 0.0195 0.0209 0.0217 0.0225 0.0226 0.0235 0.0236 0.0244 0.0245 0.0248

AA747290 AA641744 AI188196 AI222007 AA192445 D16480 AF015767 AW069055 AI365733 AF017418 D49742 M37400 Z11502 AF024714 AU155489 D32050 AW779142 AA487669 U42376 AA601564 AI042204 D14662 BF059178 U70063 AA091553 L12350 AA324335 U68019 AI248620 U24183 AI626007 AI261382 AA193416 AA911109 AF070616 AF046024

fibroblasts, and inflammatory cells can mask significant increases or decreases in the expression of genes involved in pancreatic carcinogenesis. Hence, in this study, we used an LMM system to purify as much as possible the populations of cancerous and normal epithelial cells obtained from surgical specimens (Gjerdrum et al., 2001; Kitahara et al., 2001). Because it is possible to microdissect even a single cell with LMM, this technology is crucial for achieving an effective microarray analysis of pancreatic cancer specimens (Figure 1). To evaluate the purity of microdissected cell populations, we analysed expression of the AMY1A gene, which is expressed specifically in acinar cells. After the dissection procedure, the proportion of contaminating acinar cells among the normal pancreatic ductal epithelial cells was estimated to be smaller than 0.29% (see Materials and methods). In addition to AMY1A, we examined expression levels of other genes that are highly expressed in acinar cells, for example, elastase 1, trypsin 1, and pancreatic lipase, and obtained similar results (data not shown); the purity of cell populations subjected to the LMM technique could therefore be as high as 99.7%. Second, the quality of RNA extracted from clinical tissue, particularly from pancreas, is extremely important. The pancreas is rich in RNase, and degradation of

List of 76 candidate genes for lymph node metastasis Symbol

RPS15A RPA2 USP22 TMEPAI HADHA BRE FLJ10773 MEIS2 HABP2 GOT1 ANXA13 AIM2 MMP7 AARS HUMAGCGB GSTM1 LY6E DLG5 FLJ12895 KIAA0106 NONO ASAH UBE2H THBS2 ERF MADH3 AP3D1 PFKM NTRK1 SH120 FLJ20254 HPCAL1 UBE1C

Gene name Ribosomal protein S15a Replication protein A2 (32 kDa) Ubiquitin-specific protease 22 ESTs Transmembrane, prostate androgen-induced RNA Hydroxyacy dehydrogenase, subunit A Brain and reproductive organ expressed (TNFRSF1A modulator) Likely ortholog of mouse NPC derived proline-rich protein 1 ESTs Meis (mouse) homolog 2 Hyaluronan-binding protein 2 Glutamic-oxaloacetic transaminase 1 Annexin A13 Absent in melanoma 2 Matrix metalloproteinase 7 (matrilysin, uterine) Alanyl-tRNA synthetase Chromosome 3p21.1 gene sequence Glutathione S-transferase M1 Lymphocyte antigen 6 complex, locus E Discs, large (Drosophila) homolog 5 Hypothetical protein FLJ12895 Antioxidant protein 2 Non-POU-domain-containing, octamer binding N-acylsphingosine amidohydrolase (acid ceramidase) Ubiquitin-conjugating enzyme E2H (homologous to yeast UBC8) Thrombospondin 2 Ets2 repressor factor MAD (mothers against decapentaplegic, Drosophila) homolog 3 Adaptor-related protein complex 3, delta 1 subunit Phosphofructokinase, muscle Neurotrophic tyrosine kinase, receptor, type 1 Putative G-protein-coupled receptor ESTs Hypothetical protein FLJ20254 Hippocalcin-like 1 Ubiquitin-activating enzyme E1C (homologous to yeast UBA3) Oncogene


2390 Table 2 Continued Rank Permutation P-value GenBank ID 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76

0.0260 0.0264 0.0267 0.0274 0.0285 0.0291 0.0294 0.0299 0.0308 0.0315 0.0324 0.0330 0.0341 0.0342 0.0353 0.0365 0.0366 0.0368 0.0374 0.0376 0.0387 0.0390 0.0392 0.0401 0.0402 0.0406 0.0411 0.0412 0.0413 0.0418 0.0425 0.0437 0.0438 0.0446 0.0455 0.0467 0.0475 0.0478 0.0485 0.0498

AI299911 X07979 AI143127 AA412250 D45131 W45244 AI245516 AB010427 AA907519 H20386 AA371593 D42041 L31581 AI300002 AI338165 AA922357 AI312689 U07424 AI248327 NM_006077 AF055022 AF131847 M37435 AA676585 U85658 AB011090 U34683 L41351 X52186 R52161 U93867 Z11531 U23028 AI336230 AA676322 AI268861 U73036 AI339006 AI097058 L36151

Symbol

Protein phosphatase 3 (formerly 2B), catalytic subunit, alpha isoform Integrin, beta 1 Dynactin 4 PYGB Phosphorylase, glycogen; brain BSG Basigin C3 Complement component 3 EST WDR1 WD repeat domain 1 C3ORF4 Chromosome 3 open reading frame 4 MYG1 MYG1 protein GCN1L1 GCN1 (general control of amino-acid synthesis 1, yeast)-like 1 KIAA0088 KIAA0088 protein CCR7 Chemokine (C-C motif) receptor 7 CCNI Cyclin I HEF1 Enhancer of filamentation 1 (cas-like docking; Crk-associated substrate related) DKFZp586A0618 HE1 Epididymal secretory protein (19.5 kDa) FARSL Phenylalanine-tRNA synthetase-like FLJ22233 CBARA1 Calcium-binding atopy-related autoantigen 1 DKFZP727M231 DKFZP727M231 protein MRG15 MORF-related gene 15 CSF1 Colony-stimulating factor 1 (macrophage) NPM1 Nucleophosmin (nucleolar phosphoprotein B23, numatrin) TFAP2C Transcription factor AP-2 gamma (activating enhancer-binding protein 2 gamma) KIAA0518 Max-interacting protein GSS Glutathione synthetase PRSS8 Protease, serine, 8 (prostasin) ITGB4 Integrin, beta 4 DKFZp434A2410 RPC62 Polymerase (RNA) III (DNA directed) (62 kDa) EEF1G Eucaryotic translation elongation factor 1 gamma EIF2B5 Eucaryotic translation initiation factor 2B, subunit 5 (epsilon, 82 kDa) RPS8 Ribosomal protein S8 MTF1 Metal regulatory transcription factor 1 EST IRF7 Interferon regulatory factor 7 DKFZp586L1121 FLJ23538 PIK4CA Phosphatidylinositol 4-kinase, catalytic, alpha polypeptide PPP3CA ITGB1

RNA occurs very rapidly. We examined the quality of the RNA extracted from each specimen by visualizing the 28S and 18S ribosomal RNAs after electrophoresis on denaturing agarose gels, and selected samples for which we clearly observed bands corresponding to both ribosomal RNAs. Eventually, we selected 18 pancreatic cancers (32%) among 56 surgical resections, because pancreatic RNA from the other 38 patients was of poor quality (data not shown). Careful purification of cancer cells and normal epithelial ductal cells, subsequent RNA isolation, and cDNA microarray analysis identified 260 genes whose expression was commonly upregulated according to the criteria described in Materials and methods section. As expected, in view of differences between the materials used in this study as opposed to previous reports, such as the quality of cancer cells and the number of genes on the microarray, most of our data differed from those reported previously (Crnogorac-Jurcevic et al., 2002; Iacobuzio-Donahue et al., 2002). However, we are confident that our own study is the first to provide Oncogene

Gene name

precise and genome-wide gene expression profiles of pancreatic cancers on a large scale. We identified 260 genes that were overexpressed to at least a fivefold greater extent in cancer cells than in normal pancreatic epithelial cells (Supplemental Table 1). Such genes could have immediate diagnostic potential, and those known to be critical for tumor growth also would have therapeutic potential. Some, such as regenerating gene type IV (REGIV), ephrin and vang (van gogh, Drosophila)-like 1 (VANGL1), might well serve as molecular targets for new therapeutic agents. REGIV was overexpressed in all of the informative pancreatic cancer cases we examined, and its overexpression was confirmed in seven of the 12 pancreatic cancers examined by semi-quantitative RT–PCR. Since the REGIV gene product is suspected to be a secretory protein on the basis of its amino-acid sequences, and in fact its secretion has been detected in the culture medium of HT29-5M12 cells (Violette et al., 2003), it might represent a candidate tumor marker.


2391

Figure 3 Dendrogram of two-dimensional hierarchical clustering analysis using 76 genes selected by a random permutation test that compared expression profiles of nine lymph-node-positive cases with those of four lymph-node-negative cases. In the vertical axis, 35 genes were clustered in the upper branch, indicating relatively high levels of expression in lymph-node-positive cases

Table 3 List of 168 candidate genes for liver metastasis Rank Permutation P-value GenBank ID Symbol 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

12

1.14 10 3.72 1010 4.08 107 1.21 106 1.54 106 2.98 106 9.90 106 2.72 105 5.27 105 7.45 105 0.0002 0.0003 0.0004 0.0008 0.0011 0.0013 0.0015 0.0018 0.0020 0.0020 0.0020 0.0023 0.0024 0.0027

U63743 S74678 U12707 D56784 U31383 H06970 AF038954 W19984 AA282650 U16738 AA614311 AF006088 AF007871 AA904028 D21090 AA910279 T69711 AA226073 AI338282 AA583455 AA731151 U14575 AA843756 M81637

KNSL6 HNRPK WAS DEK GNG10 STK24 ATP6J DREV1 SAC1 RPL14 VCP ARPC5 DYT1 PAPPA RAD23B STAU ITM2C TIGA1 RNF7 KIAA1085 PPP1R8 ID2 GCL

Gene name Kinesin-like 6 (mitotic centromere-associated kinesin) Heterogeneous nuclear ribonucleoprotein K Wiskott–Aldrich syndrome (eczema–thrombocytopenia) DEK oncogene (DNA binding) Guanine nucleotide-binding protein 10 Serine/threonine kinase 24 (Ste20, yeast homolog) ATPase, H+ transporting, lysosomal (vacuolar proton pump), member J CGI-81 protein Suppressor of actin 1 Ribosomal protein L14 Valosin-containing protein Actin-related protein 2/3 complex, subunit 5 (16 kDa) Dystonia 1, torsion (autosomal dominant; torsin A) Pregnancy-associated plasma protein A RAD23 (S. cerevisiae) homolog B Staufen (Drosophila, RNA-binding protein) EST Integral membrane protein 2C Homo sapiens mRNA; cDNA DKFZp566L203 (from clone DKFZp566L203) Ring-finger protein 7 KIAA1085 protein Protein phosphatase 1, regulatory (inhibitor) subunit 8 Inhibitor of DNA binding 2, dominant negative helix–loop–helix protein Grancalcin Oncogene


2392 Table 3 Continued Rank Permutation P-value GenBank ID Symbol 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 Oncogene

0.0033 0.0037 0.0045 0.0048 0.0049 0.0050 0.0054 0.0054 0.0057 0.0058 0.0062 0.0070 0.0070 0.0071 0.0074 0.0075 0.0079 0.0082 0.0083 0.0086 0.0092 0.0097 0.0103 0.0103 0.0104 0.0105 0.0107 0.0108 0.0108 0.0110 0.0114 0.0114 0.0120 0.0122 0.0122 0.0123 0.0124 0.0127 0.0135 0.0142 0.0142 0.0143 0.0143 0.0145 0.0152 0.0154 0.0160 0.0161 0.0161 0.0162 0.0167 0.0171 0.0180 0.0182 0.0182 0.0210 0.0211 0.0216 0.0221 0.0223 0.0224 0.0225 0.0226 0.0237 0.0241 0.0247 0.0250 0.0253 0.0254 0.0258

L37368 AK000403 AF076483 D13315 U66818 X56351 L08424 X15187 U33286 AA747290 AI148832 S65738 X53586 AA447852 U31906 AA664213 AI338165 W74416 L13939 AI344213 AI125978 H96478 X74929 U46570 U21242 U92459 AA078295 W95089 D55654 AA084871 AF072860 M26252 AF042081 D63881 AA195740 AI280555 M36341 C06051 D28473 R23830 U51166 U09278 AA989386 AA128470 U01184 M77698 AI272932 U45879 Z35491 AF016507 X89478 X06323 NM_016401 M29065 AW245101 AA431846 E02628 U47025 AI349804 X99584 Z21507 D13630 U24223 AA315729 AA401318 AA524350 AB004857 AA379042 U38320 AW779142

RNPS1 FLJ20396 PGLYRP GLO1 UBE2I ALAS1 ASCL1 TRA1 CSE1L RPS15A KIAA1209 ADF ITGA6 PC326 GOLGA4 DKC1 HEF1 LOC51126 AP1B1 CCS SNX2 KRT8 TTC1 GTF2A2 GRM8 HSPC033 MDH1 YKT6 PRKRA PKM2 SH3BGRL KIAA0160 KIAA0860 ARF4 JAK1 IARS TDG FAP DSP FLII YY1 BAG5 BIRC2 BAG1 CTBP2 HRB MRPL3 HSPC138 HNRPA2B1 E2IG3 LOC51187 PYGB SMT3H1 EEF1D KIAA0005 PCBP1 DKFZP566D193 LOC51719 SLC11A2 PUM2 MMP19 HUMAGCGB

Gene name RNA-binding protein S1, serine-rich domain Hypothetical protein FLJ20396 Peptidoglycan recognition protein Glyoxalase I Ubiquitin-conjugating enzyme E2I (homologous to yeast UBC9) Aminolevulinate, delta-, synthase 1 Achaete-scute complex (Drosophila) homolog-like 1 Tumor rejection antigen (gp96) 1 Chromosome segregation 1 (yeast homolog)-like Ribosomal protein S15a KIAA1209 protein Destrin (actin depolymerizing factor) Integrin, alpha 6 PC326 protein Golgi autoantigen, golgin subfamily a, 4 Dyskeratosis congenita 1, dyskerin Enhancer of filamentation 1 (cas-like docking; Crk-associated substrate related) N-terminal acetyltransferase complex ard1subunit Adaptor-related protein complex 1, beta 1 subunit Copper chaperone for superoxide dismutase Sorting nexin 2 EST Keratin 8 Tetratricopeptide repeat domain 1 General transcription factor IIA, 2 (12 kDa subunit) Glutamate receptor, metabotropic 8 ESTs HSPC033 protein Malate dehydrogenase 1, NAD (soluble) SNARE protein Protein kinase, interferon-inducible double-stranded RNA-dependent activator Pyruvate kinase, muscle SH3 domain-binding glutamic acid-rich protein like KIAA0160 protein Homo sapiens mRNA full-length insert cDNA clone EUROIMAGE 41832 KIAA0860 protein ADP-ribosylation factor 4 Janus kinase 1 (a protein tyrosine kinase) Isoleucine-tRNA synthetase ESTs Thymine-DNA glycosylase Fibroblast activation protein, alpha EST Desmoplakin (DPI, DPII) Flightless I (Drosophila) homolog YY1 transcription factor BCL2-associated athanogene 5 Baculoviral IAP repeat-containing 2 BCL2-associated athanogene C-terminal binding protein 2 HIV Rev-binding protein Mitochondrial ribosomal protein L3 Hypothetical protein Heterogeneous nuclear ribonucleoprotein A2/B1 Putative nucleotide-binding protein, estradiol-induced 60S ribosomal protein L30 isolog Polypeptide chain elongation factor 1 alpha Phosphorylase, glycogen; brain EST SMT3 (suppressor of mif two 3, yeast) homolog 1 Eucaryotic translation elongation factor 1 delta KIAA0005 gene product poly(rC)-binding protein 1 FLJ23197 DKFZP566D193 protein MO25 protein Solute carrier family 11, member 2 Pumilio (Drosophila) homolog 2 Matrix metalloproteinase 19 Chromosome 3p21.1 gene sequence


2393 Table 3 Rank Permutation P-value GenBank ID Symbol 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164

0.0262 0.0270 0.0271 0.0272 0.0274 0.0276 0.0282 0.0282 0.0289 0.0290 0.0290 0.0290 0.0291 0.0293 0.0301 0.0304 0.0313 0.0313 0.0314 0.0315 0.0317 0.0319 0.0320 0.0321 0.0327 0.0329 0.0330 0.0331 0.0337 0.0342 0.0342 0.0346 0.0348 0.0352 0.0363 0.0364 0.0366 0.0367 0.0367 0.0376 0.0378 0.0380 0.0384 0.0389 0.0395 0.0395 0.0403 0.0409 0.0417 0.0419 0.0419 0.0423 0.0423 0.0427 0.0429 0.0433 0.0436 0.0443 0.0448 0.0449 0.0450 0.0458 0.0462 0.0466 0.0469 0.0471 0.0471 0.0472 0.0476 0.0482

AA233644 R39044 M58458 H89110 U47077 L40401 AI365683 AA236252 D50683 AF039690 L19067 U48734 M61199 U56637 AA514818 M22324 AA921921 N45298 X76104 X97630 AJ002308 D14812 AA447019 AA357508 U96915 D10522 N46856 D26125 AI085802 M98252 H48649 AI139231 AI289407 U54831 AA281115 AA249454 U89278 N41902 M24398 AA432312 AF006516 AA706503 L41668 N95414 M20472 D78298 AI078833 X89602 M91029 X73478 U09953 U44772 AI189477 AA973853 U81504 D30612 AA506972 AA404724 AA634090 AB001451 U83463 AL120683 H20386 AA477862 AF075590 AI092703 AA748421 AA639771 AF052113 AF007216

PPP1CC RPS4X PRKDC ZAP128 ASH2L TGFBR2 SDCCAG8 RELA ACTN4 SSFA2 CAPZA1 KIAA0068 ANPEP KIAA0414 ARHGEF12 DAPK1 EMK1 SYNGR2 KIAA0026 MAN1B1 SAP18 MACS AKR1C4 CAV2 PLOD FGG FBL ZNF207 TOP2B UBQLN1 EDR2 CLTH PTMS TSPYL SSH3BP1 EEF1A1 GALE CLTA VLCAD TAX1BP1 HSRTSBETA AMPD2 PPP2R4 RPL9 PPT1 IDH2 AP3B1 ZNF282 KIAA0668 GPRK7 HNRPA1 SLI SDCBP LASS2 MYG1 KIAA0974 BZRP FBXW1B TFR2 MMP12 Rab14 SLC4A4

Continued Gene name

Protein phosphatase 1, catalytic subunit, gamma isoform Homo sapiens clone 25194 mRNA sequence Ribosomal protein S4, X-linked ESTs Protein kinase, DNA-activated, catalytic polypeptide Peroxisomal long-chain acyl-coA thioesterase; putative protein Homo sapiens PAC clone RP4-751H13 from 7q35–qter ash2 (absent, small, or homeotic, Drosophila, homolog)-like Transforming growth factor, beta receptor II (70–80 kDa) Serologically defined colon cancer antigen 8 v-rel avian reticuloendotheliosis viral oncogene homolog A Actinin, alpha 4 Sperm-specific antigen 2 Capping protein (actin filament) muscle Z-line, alpha 1 KIAA0068 protein Alanyl (membrane) aminopeptidase KIAA0414 protein Rho guanine exchange factor (GEF) 12 Death-associated protein kinase 1 ELKL motif kinase Synaptogyrin 2 MORF-related gene X Mannosidase, alpha, class 1B, member 1 Homo sapiens clone 24711 mRNA sequence sin3-associated polypeptide, 18 kDa Myristoylated alanine-rich protein kinase C substrate (MARCKS, 80K-L) Homo sapiens cDNA: FLJ23091 fis, clone LNG07220 Aldo-keto reductase family 1, member C4 Caveolin 2 Procollagen-lysine, 2-oxoglutarate 5-dioxygenase Fibrinogen, gamma polypeptide Fibrillarin Zinc-finger protein 207 Topoisomerase (DNA) II beta (180 kDa) Ubiquilin 1 ESTs, weakly similar to KIAA0227 [H.sapiens] Early development regulator 2 (homolog of polyhomeotic 2) Clathrin assembly lymphoid-myeloid leukemia gene Parathymosin TSPY-like Spectrin SH3 domain-binding protein 1 Eucaryotic translation elongation factor 1 alpha 1 Galactose-4-epimerase, UDPESTs Clathrin, light polypeptide (Lca) Very-long-chain acyl-CoA dehydrogenase Tax1 (human T-cell leukemia virus type I)-binding protein 1 rTS beta protein Adenosine monophosphate deaminase 2 (isoform L) Protein phosphatase 2A, regulatory subunit B0 (PR 53) Ribosomal protein L9 Palmitoyl-protein thioesterase 1 Isocitrate dehydrogenase 2 (NADP+), mitochondrial Homo sapiens cDNA FLJ20532 fis, clone KAT10877 Adaptor-related protein complex 3, beta 1 subunit Zinc-finger protein 282 KIAA0668 protein G-protein-coupled receptor kinase 7 Heterogeneous nuclear ribonucleoprotein A1 Neuronal Shc adaptor homolog Syndecan-binding protein (syntenin) LAG1 longevity assurance homolog 2 (S. cerevisiae) MYG1 protein KIAA0974 protein Benzodiazapine receptor (peripheral) f-box and WD-40 domain protein 1B Transferrin receptor 2 Matrix metalloproteinase 12 (macrophage elastase) GTPase Rab14 Solute carrier family 4, sodium bicarbonate cotransporter, member 4 Oncogene


2394 Table 3 Continued Rank Permutation P-value GenBank ID Symbol 165 166 167 168

0.0485 0.0489 0.0492 0.0494

AA809819 AI218495 N80334 AA847660

Gene name

Cellular repressor of E1A-stimulated genes ESTs, moderately similar to integral inner nuclear membrane protein MAN1 DKFZP586O0223 Hypothetical protein HEXA hexosaminidase A (alpha polypeptide)

CREG

Figure 4 Dendrogram of two-dimensional hierarchical clustering analysis using 168 genes selected by a random permutation test that compared expression profiles of five liver-metastasis-positive cases with those of six negative cases. In the vertical axis, 60 genes were clustered in the upper branch, which was more highly expressed in liver-metastasis-positive cases

VANGL1 was overexpressed in all of the informative pancreatic cancers according to our microarray data, and its high expression was also confirmed by semiquantitative RT–PCR. The VANGL1 product, which contains four putative transmembrane domains, was expressed specifically in testis and ovary among 29 normal tissues examined elsewhere (Yagyu et al., 2002). This gene is also highly and frequently transactivated in hepatocellular carcinomas. Since enforced reduction of its expression has induced apoptosis in hepatocellular carcinomas (Yagyu et al., 2002), the gene product is likely to be another good candidate for development of novel anticancer drugs. Among the numerous other Oncogene

genes that were highly overexpressed in the pancreatic cancers we examined, those whose products are putative membranous or secreted proteins are all potential targets for novel anticancer drugs or for serological diagnostic markers to aid early detection. We also identified 346 genes that were significantly downregulated in pancreatic cancer cells (Supplemental Table 2), but the functions of 135 of them are unknown at present. Since the pancreatic duct epithelial cell is the progenitor of pancreatic ductal cancer, our data were obtained by comparing purified populations of cancer cells with origin cells. Therefore, the set of genes listed here reflects specific downregulation during


2395

Figure 5 (a) Result of a two-dimensional hierarchical clustering analysis using 84 genes selected by a random permutation test that compared expression profiles of seven early-recurrent cases (within 12 months after surgery) with those of six late-recurrent cases (over 12 months after surgery). In the vertical axis, 84 genes were clustered in different branches according to similarity in relative expression ratios. (b) Optimization of the number of discriminating genes. The classification score (CS) was calculated by using the prediction score of early-recurrent case (PSr) and late-recurrent case (PSn) in each gene set, as follows:

CS ¼ ðmPSr mPSn Þ=ðsPSr þ sPSn Þ A larger value of CS indicates better separation of the two groups by the predictive scoring system. (c) Different prediction scores appear when the number of discriminating genes is changed. Red diamonds represent early-recurrent cases; blue diamonds denote laterecurrent cases

transformation from normal epithelial cells to cancer cells. Although downregulation of genes can be either the cause or the result, some of the genes listed here are likely to encode tumor suppressors. Some known suppressor genes for pancreatic cancer, such as SMAD4, TP53, INK4A, and BRCA2 (Rozenblum et al., 1997; Goggins et al., 2000), do not appear on our list of downregulated genes, but others known to be involved in tumor suppression or apoptosis, such as AXIN1 upregulated 1 (AXUD1), deleted in liver cancer 1 (DLC1), growth arrest and DNA-damage-inducible, beta (GADD45B), p53-dependent damage-inducible nuclear protein 1 (p53DINP1), are included. AXUD1, a nuclear protein, is induced in response to elevation of axin; the latter is a key mediator of the Wnt signaling pathway and is important for axis formation during early development. As dysfunction or down-

regulation of the Wnt signaling pathway is observed often in human tumors (Satoh et al., 2000; Ishiguro et al., 2001), this gene product may have a tumor suppressor function; our data imply that downregulation of AXUD1 might lead to downregulation of this signaling pathway and then to pancreatic carcinogenesis. Deleted in liver cancer 1 (DLC1) is a candidate tumor suppressor gene for human liver cancer as well as for prostate, lung, colorectal, and breast cancers. DLC1 is highly similar in sequence to rat p122 RhoGap, which negatively regulates Rho GTPases. Hence, downregulation of DLC1 is considered to result in constitutive activation of the Rho-Rho-kinase pathway and to lead to oncogenic malignant transformation (Yuan et al., 1998; Ng et al., 2000). These observations suggest that pancreatic carcinogenesis involves complicated and diverse pathways. Oncogene


2396 Table 4 Rank

List of 84 candidate genes for prognosis

Permutation P-value

GenBank ID

Symbol

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

5.09E06 3.76E05 0.0001 0.0002 0.0009 0.0010 0.0016 0.0016 0.0017 0.0018 0.0022 0.0026 0.0032 0.0034 0.0035 0.0041 0.0043 0.0053 0.0064 0.0070 0.0086 0.0104 0.0114 0.0118 0.0121 0.0140 0.0143 0.0149 0.0158 0.0172 0.0191 0.0191 0.0198 0.0201 0.0206 0.0222 0.0231 0.0242 0.0243

AF049884 NM_006077 AA700379 AI340331 Z11531 AA459167 AI014395 AW157203 M94083 AI123363 X53777 U16798 X76013 M22382 AF075590 L38995 AA150867 L76687 T70782 H89783 D83782 AI018632 AA531437 M75126 AA936173 AA488766 M60922 AI075048 AL031668 D26600 L19711 AI357601 AI148194 U51586 X57398 AF004430 M17886 AI279562 L14778

ARGBP2 CBARA1 MTMR1 HT010 EEF1G NPD002 YME1L1 LCAT CCT6A RPL23A RPL17 ATP1A1 QARS HSPD1 BZRP TUFM TIMM9 GRB14 FLJ10803 SERPINA4 SCAP LAMP1 MLLT4 HK1 RPS11 SYNGR2 FLOT2 CTSB RALY PSMB4 DAG1 RPL37A

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69

0.0251 0.0253 0.0255 0.0256 0.0257 0.0262 0.0270 0.0294 0.0296 0.0308 0.0309 0.0313 0.0314 0.0330 0.0331 0.0331 0.0336 0.0343 0.0347 0.0350 0.0366 0.0373 0.0374 0.0384 0.0397 0.0401 0.0405 0.0407 0.0409 0.0412

AA156481 AA083406 M11717 AF015767 AF000984 X17206 W45522 X06323 X83218 AI305234 AI246699 W24533 AA029875 AA504081 AA778572 D11999 D32050 D63997 AI366139 R64726 M61715 U46191 AI090753 AA487669 AI289991 AI131289 AA345061 AI299327 AA922716 AA845165

Oncogene

SIAHBP1 PM5 TPD52L2 RPLP1 KIAA0469 PPP3CA RPL13A EIF3S8 HSPA1A BRE DBY RPS2 LOC51189 MRPL3 ATP5O CATX-8 GRB10 CASP4 CSH2 HSPC164 GLS AARS GOLGA3 MAC30 WARS RAGE SHMT2 GSTM1 DKFZP761C169 RPLP2 KIAA0903 PRKACB PRSS1

Gene name Arg/Abl-interacting protein ArgBP2 Calcium-binding atopy-related autoantigen 1 Myotubularin related protein 1 Uncharacterized hypothalamus protein HT010 Eucaryotic translation elongation factor 1 gamma NPD002 protein YME1 (S.cerevisiae)-like 1 Lecithin-cholesterol acyltransferase Chaperonin-containing TCP1, subunit 6A (zeta 1) Ribosomal protein L23a Ribosomal protein L17 ATPase, Na+/K+ transporting, alpha 1 polypeptide Glutaminyl-tRNA synthetase Heat shock 60 kDa protein 1 (chaperonin) Benzodiazapine receptor (peripheral) Tu translation elongation factor, mitochondrial Translocase of inner mitochondrial membrane 9 (yeast) homolog Growth factor receptor-bound protein 14 Hypothetical protein FLJ10803 Serine (or cysteine) proteinase inhibitor, clade A , member 4 srebp cleavage-activating protein Lysosomal-associated membrane protein 1 Myeloid/lymphoid or mixed-lineage leukemia translocated to, 4 Hexokinase 1 Ribosomal protein S11 Synaptogyrin 2 Flotillin 2 Cathepsin B RNA-binding protein (autoantigenic) Proteasome (prosome, macropain) subunit, beta type, 4 Dystroglycan 1 (dystrophin-associated glycoprotein 1) Ribosomal protein L37a Novel human gene mapping to chomosome 22 siah-binding protein 1 pM5 protein Tumor protein D52-like 2 Ribosomal protein, large, P1 KIAA0469 gene product Protein phosphatase 3 (formerly 2B), catalytic subunit, alpha isoform (calcineurin A alpha) Ribosomal protein L13a Eucaryotic translation initiation factor 3, subunit 8 (110 kDa) Heat shock 70 kDa protein 1A Brain and reproductive organ-expressed (TNFRSF1A modulator) DEAD/H (Asp-Glu-Ala-Asp/His) box polypeptide, Y chromosome Ribosomal protein S2 ATPase inhibitor precursor Mitochondrial ribosomal protein L3 ATP synthase, H+ transporting, mitochondrial F1 complex, O subunit ESTs CATX-8 protein Growth factor receptor-bound protein 10 Caspase 4, apoptosis-related cysteine protease Chorionic somatomammotropin hormone 2 Hypothetical protein Glutaminase Alanyl-tRNA synthetase Golgi autoantigen, golgin subfamily a, 3 Hypothetical protein Homo sapiens cDNA: FLJ23591 fis, clone LNG14729 Tryptophanyl-tRNA synthetase Renal tumor antigen Serine hydroxymethyltransferase 2 (mitochondrial) Glutathione S-transferase M1 Hypothetical protein DKFZp761C169 Ribosomal protein, large P2 KIAA0903 protein ESTs Protein kinase, cAMP-dependent, catalytic, beta Protease, serine, 1 (trypsin 1)


2397 Table 4 Continued Rank

Permutation P-value

GenBank ID

Symbol

70 71 72 73 74 75 76 77 78 79 80

0.0417 0.0424 0.0428 0.0433 0.0434 0.0435 0.0439 0.0463 0.0464 0.0467 0.0471

AA877534 AA255699 H73961 D87666 C01335 AI075943 D87989 Z26876 AI080640 X04588 D30949

GPRC5C

81 82 83 84

0.0484 0.0485 0.0486 0.0499

Z11559 D86956 L13740 AA320379

ACO1 HSP105B NR4A1 POH1

ARPC3 GPI SENP2 UGTREL1 RPL38 AGR2 TPM3

Pancreatic cancer is characterized by very aggressive progression and rapid recurrence after surgical treatment. It has been reported that the cumulative 1-, 3-, and 5-year disease-free survival rates were 66, 7, and 3% respectively, and median disease-free survival time was only 8 months (Sperti et al., 1997). Most common recurrent sites are the local region and the liver, and distant metastases appear in the peritoneal cavity. However, since the relationships between tumor characteristics and the recurrence patterns are still little understood, we compared the expression profiles to lymph node status or liver metastasis. We identified 76 genes that might be associated with lymph-node status, and 168 genes with liver metastasis. These genes included some key molecules whose possible roles in tumor progression had been reported previously: ITGB4 and BSG were upregulated in lymph-node-positive cases, and KNSL6 and KRT8 were relatively upregulated in liver metastasis cases. ITGB4 was reported to promote carcinoma invasion through a preferential and localized targeting of phosphoinositide-3 OH kinase activity (Shaw et al., 1997), supporting the possible involvement of ITGB4 in lymph node metastasis. KNSL6, a member of the kinesin family of motor proteins, is known to be involved in chromosome segregation during mitosis (Maney et al., 1998). The transcript of KNSL6 was highly expressed in colon cancer, and was identified as cancer antigens associated with a cancer-related serum IgG response (Scanlan et al., 2002). Thus, this antigen could be a biological marker for diagnosis and for the monitoring of recurrence site. In addition, we identified 84 genes possibly associated with tumor recurrence of pancreatic cancers. Expression levels of a subset of 30 genes selected from these 84 genes would be useful for predicting the disease-free interval after surgical operation (Figure 5). These results might be useful for selection of patients for active adjuvant therapy although larger-scale study will be required to further evaluate our prediction system. Cancer therapies directed at specific molecular alterations that occur in cancer cells have been validated through clinical development and regulatory approval of anticancer drugs such as trastuzumab (Herceptin) for

Gene name G-protein-coupled receptor, family C, group 5, member C Human DNA sequence from clone RP3-324O17 on chromosome 20 Actin-related protein 2/3 complex, subunit 3 (21 kDa) Glucose phosphate isomerase ESTs, weakly similar to FLDED [H. sapiens] Sentrin-specific protease UDP-galactose transporter related Ribosomal protein L38 Anterior gradient 2 (Xenepus laevis) homolog 2.5 kb mRNA for cytoskeletal tropomyosin TM30 Homo sapiens cDNA FLJ12750 fis, clone NT2RP2001168, weakly similar to verprolin Aconitase 1, soluble Heat shock 105 kDa Nuclear receptor subfamily 4, group A, member 1 26S proteasome-associated pad1 homolog

the treatment of advanced breast cancer, imatinib methylate (Gleevec) for chronic myeloid leukemia, gefitinib (Iressa) for non-small cell lung cancer (NSCLC), and rituximab (anti-CD20mAb) for B-cell lymphoma and mantle cell lymphoma (Fang et al., 2000; Ciardiello and Tortora, 2001; Slamon et al., 2001; Rehwald et al., 2003). These drugs are clinically effective and better tolerated than traditional anticancer agents because they target only transformed cells. Hence, such drugs not only improve survival and quality of life for cancer patients, but also validate the concept of molecularly targeted cancer therapy. Furthermore, targeted drugs can enhance the efficacy of standard chemotherapy when used in combination with it (Gianni, 2002; Klejman et al., 2002). Therefore, future cancer treatments will probably involve combining conventional drugs with target-specific agents aimed at different characteristics of tumor cells such as angiogenesis and invasiveness. Despite the success of the anticancer drugs mentioned above, no specific molecular-targeted drugs or tumor markers except CA19-9 have yet been developed for diagnosis or treatment of patients with pancreatic cancer. The poor prognosis of this disease is due to both the difficulty of diagnosis at an early stage and a generally poor response to current therapeutic methods; at present, no effective treatment is available for patients at an advanced stage. Hence, sensitive and specific tumor markers for pancreatic cancer, as well as new therapeutic approaches, are urgently required. The extensive list reported here of genes that are up- or downregulated in pancreatic cancers should provide useful information for a better understanding of the precise mechanisms of tumorigenesis in this vital organ, and for identifying molecules to serve as molecular targets for early diagnosis and treatment of pancreatic cancer. Materials and methods cDNA microarray We fabricated a genome-wide cDNA microarray with 23 040 cDNAs selected from the UniGene database (build #131) of Oncogene


2398 the National Center for Biotechnology Information (NCBI). This microarray system was constructed essentially as described previously (Ono et al., 2000). Briefly, the cDNAs were amplified by RT–PCR using poly (A) þ RNAs isolated from various human organs as templates; the lengths of the amplicons ranged from 200 to 1100 bp, without any repetitive or poly (A) sequences. Tissue samples and microdissection Tissue samples from pancreatic cancers (n ¼ 18) and from normal pancreas (n ¼ 7) were obtained from surgical specimens, concerning which all patients had given informed consent. All cancer tissues had been confirmed histologically as invasive ductal carcinomas by the pathologist; clinical information was obtained from medical records (four female and 14 male patients; median age 65.0 with range 46–77 years). Clinical stage was judged according to the UICC TNM classification. Since almost all pancreatic ductal cells from corresponding blocks of normal tissue showed dysplastic changes, mostly because of downstream obstruction of the duct, ductal cells from only four of the 18 patients were suitable for use as normal controls. Hence, we obtained normal pancreatic ductal cells from three patients who had undergone pancreatoduodenectomy for cholangiocarcinoma, duodenal leiomyosarcoma, or ampullary tumor. These normal pancreatic duct cells were observed as a normal epithelium pathologically, and they were not dysplasia. All specimens were harvested immediately after surgical resection and were embedded in TissueTek OCT medium (Sakura, Tokyo, Japan) before storage at 801C. The frozen tissues were cut into 8 mm sections using a cryostat (Sakura, Tokyo, Japan) and then stained with hematoxylin and eosin for histological examination. Pancreatic carcinoma cells and normal pancreatic ductal epithelial cells were isolated selectively using the EZ cut system with a pulsed ultraviolet narrow beam-focus laser (SL Microtest GmbH, Germany) in accordance with the manufacturer’s protocols. After microdissection, the seven populations of normal ductal epithelial cells were mixed to make a ‘universal control’ for all 18 cancer samples. RNA extraction, T7-based RNA amplification, and hybridization Total RNAs were extracted from each sample of lasermicrodissected cells into 350 ml of RLT lysis buffer (QIAGEN, Hilden, Germany). The extracted RNAs were treated for 30 min at room temperature with 30 U of DNase I (Roche, Basel, Switzerland) in the presence of 1 U of RNase inhibitor (TOYOBO, Osaka, Japan) to remove any contaminating genomic DNA. After inactivation at 701C for 10 min, the RNAs were purified with an RNeasy Mini Kit (QIAGEN) according to the manufacturer’s recommendations. All of the DNase I-treated RNAs were subjected to T7-based RNA amplification; two rounds of amplification yielded 50–100 mg of aRNA from each sample. Then 2.5 mg aliquots of aRNA from cancer cells or normal pancreatic ductal epithelial cells were labeled by reverse transcription with Cy5-dCTP or Cy3dCTP (Amersham Biosciences), respectively, as described previously (Ono et al., 2000). Hybridization, washing, and scanning were also carried out according to methods described previously (Ono et al., 2000). Data analysis Signal intensities of Cy3 and Cy5 from the 23 040 spots were quantified and analysed by substituting backgrounds, using ArrayVision software (Imaging Research, Inc., St Catharines, Oncogene

Ontario, Canada). Subsequently, the fluorescent intensities of Cy5 (tumor) and Cy3 (control) for each target spot were adjusted so that the mean Cy3/Cy5 ratio of 52 housekeeping genes on the array was equal to one. Because data derived from low signal intensities are less reliable, we determined a cutoff value on each slide as described previously (Ono et al., 2000) and excluded genes from further analysis when both Cy3 and Cy5 dyes yielded signal intensities lower than the cutoff (Saito-Hisaminato et al., 2002). For other genes, we calculated the Cy5/Cy3 ratio using the raw data of each sample. We identified up- or downregulated genes common to pancreatic cancer according to the following criteria: (1) genes for which we were able to obtain expression data in more than 50% (at least nine of the 18 cases) of the cancers examined; and (2) genes whose expression ratio was more than 5.0 in pancreatic cancer cells (defined as upregulated genes) or genes whose expression ratio was less than 0.2 (defined as downregulated genes) in more than 50% of informative cases. Additionally, we further selected genes according to the following criteria: (1) genes for which we were able to obtain expression data in six, seven, or eight cases; and (2) genes whose expression ratio was more than 5.0 in all of the informative cases. Calculation of contamination percentage Pancreatic amylase (AMY1A), an enzyme expressed exclusively in pancreatic acinar cells, was used to evaluate the proportion of acinar cells present in the population of microdissected normal pancreatic ductal epithelial cells. Each intensity was normalized to the intensity of the beta actin gene (ACTB) as follows: (Ratio A) the AMY1A/ACTB intensity ratio in whole pancreas (where most of the cells correspond to acinar cells); the signal intensity of poly (A) þ RNA isolated from normal whole pancreas pooled from nine individuals (Clontech) was 96.74. (Ratio B) the AMY1A/ACTB intensity ratio in microdissected normal ductal epithelial cells ¼ 0.28 (the average of signal intensities from 18 hybridizations using a mixture of normal pancreatic ductal cells from seven individuals as a universal control). The contamination percentage was calculated as (Ratio B)/ (Ratio A) 100 ¼ 0.29%. Identification of genes responsible for clinicopathological data Genes associated with clinicopathological features, such as lymph node positive (r) and negative (n), liver metastasis positive (r) and negative (n), and early recurrence (r) and late recurrence (n), were chosen according to following two criteria: (i) signal intensities are higher than the cutoff value in at least 80% of the cases; and (ii) |MedrMedn|X0.5, where Med indicates the median derived from log-transformed relative expression ratios in two groups. Genes were selected as candidates when they met the criteria with a permutation P-value of smaller than 0.05 in each clinicopathological status. First, we applied a random permutation test to identify genes that were expressed differently in the two groups. The mean (m) and standard deviation (s) were calculated from the log-transformed relative expression ratios of each gene in node-positive (r) and node-negative (n) cases, liver-metastasispositive (r) and -negative (n), and early recurrence (r) and late recurrence (n), respectively. A discrimination score (DS) for each gene was defined as follows: DS ¼ ðmr mn Þ=ðsr þ sn Þ


2399 Table 5 Primer sequences for semi-quantitative RT–PCR experiments GenBank ID U51478 L20688 AA916826 X63629 X64652 J04501 U70322 AA308062 U75285 AA316525 L36645 AA789332 AF141347

Symbol

Forward primer

Reverse primer

ATP1B3 ARHGDIB APP CDH3 RBMS1 GYS1 KPNB2 S100P BIRC5 REGIV EPHA4 VANGL1 TUBA3

50 -CAGTGTACAGTCGCCAGATAG-30 50 -CTCCCTCTGATCCTCCATCAG-30 50 -CTGCTGGTCTTCAATTACCAAG-30 50 -ACCTTCTTAGGCCTCCTGGTG-30 50 -CTGTCGAGACGTCTAATGACC-30 50 -TGCCCACTGTGAAACCACTAG-30 50 -TCTTGGAGACTATAAGGGAGCC-30 50 -GCATGATCATAGACGTCTTTTCC-30 50 -CTCCCTCAGAAAAAGGCAGTG-30 50 -CCAGTAGTGGCTTCTAGCTC-30 50 -CACAGCTGCTGGTTATACCAC-30 50 -GGCCTTCTGCATCACCAACG-30 50 -CTTGGGTCTGTAACAAAGCATTC-30

50 -TCCTCACATACAGAACTTCTCCAC-30 50 -TCTTGTTCTCTTGTGTCGTTTACAG-30 50 -CTCATCCCCTTATATTGCCACTT-30 50 -TACACGATTGTCCTCACCCTTC-30 50 -TTACTAAAATAAACCTGTTCGGGGG-30 50 -CATCTCATCTCCGGACACACT-30 50 -TTTTGCTTCTTCACATCCACTG-30 50 -GATGAACTCACTGAAGTCCACCT-30 50 -GAAGCTGTAACAATCCACCCTG-30 50 -GAAAAACAAGCAGGAGTTGAGTG-30 50 -CTTTAATTTCAGAGGGCGAAGAC-30 50 -ATGTCTCAGACTGTAAGCGAAGG-30 50 -AAGGATTATGAGGAGGTTGGTGT-30

GenBank ID and gene symbols were retrieved from the Unigene Databases (build #131)

We carried out permutation tests to estimate the ability of individual genes to distinguish between two groups; samples were randomly permutated between the two classes 10 000 times. Since the DS data set of each gene showed a normal distribution, we calculated a P-value for the user-defined grouping (Golub et al., 1999). For this analysis, we applied the expression data of 13 cases consisting of four lymphnode-positive and nine-negative cases, those of 11 cases consisting of five liver-metastasis-positive and six -negative cases, and those of 13 cases consisting of seven early-recurrent cases and six late-recurrent cases. For these, analyses were performed by using only Stage IV cases according to UICC TNM classification. Calculation of prediction score We further calculated the prediction score of recurrence according to procedures described previously (Golub et al., 1999). Each gene (gi) votes for either early-recurrent cases or late-recurrent cases depending on whether the expression level (xi) in the sample is closer to the mean expression level of earlyrecurrent cases or late-recurrent cases in reference samples. The magnitude of the vote (vi) reflects the deviation of the expression level in the sample from the average of the two classes:

cases (PSn) in each gene set, as follows: CS ¼ ðmPSr mPSn Þ=ðsPSr þ sPSn Þ A larger value of CS indicates better separation of the two groups by the predictive scoring system. For the leave-one-out test, one sample is withheld, the permutation P-value and mean expression levels are calculated using remaining samples, and the class of the withheld sample is subsequently evaluated by calculating its prediction score. We repeated this procedure for each of the 13 samples. Semi-quantitative RT–PCR We selected 12 highly upregulated genes and examined their expression levels by means of semi-quantitative RT–PCR experiments. A 3 mg aliquot of aRNA from each sample was reverse transcribed to single-stranded cDNAs using random primer (Roche) and Superscript II (Invitrogen). Each cDNA mixture was diluted for subsequent PCR amplification with the same primer sets that were prepared for the target DNA- or alpha tubulin-specific reactions (Table 5). Expression of alpha tubulin served as an internal control. PCR reactions were optimized for the number of cycles to ensure product intensity within the linear phase of amplification.

Vi ¼ jxi ðmr þ mn Þ=2j We summed the votes to obtain total votes for the earlyrecurrent cases (Vr) and late-recurrent cases (Vn), and calculated PS values as: PS ¼ ððVr Vn Þ=ðVr þ Vn ÞÞ100 reflecting the margin of victory in the direction of either earlyrecurrent cases or late-recurrent cases. PS values range from 100 to 100; a higher absolute value of PS reflects a stronger prediction. Evaluation of classification and leave-one-out test We calculated the classification score (CS) by using the prediction score of early-recurrent (PSr) and late-recurrent

Acknowledgements We thank Hiroko Bando, Noriko Sudo, Kie Naito, Saori Osawa, and Miwako Ando for fabrication of the cDNA microarray; Emi Ichihashi for analysis of the data; Tae Makino for preparation of samples by cryostat; Drs Minoru Takada, Katsuhiko Murakawa, Eiji Tamoto, and Manabu Shindo for preparation of surgical specimens; Drs Tatsuya Kato, Toshihiko Nishidate, and Keisuke Taniuchi for experiment of LMM and semi-quantitative RT–PCR; Dr Naoki Oyaizu for pathological advice; and Drs Hitoshi Zembutsu, Takefumi Kikuchi, Satoshi Nagayama, and Soji Kakiuchi for helpful discussions. This work was supported in part by Research for the Future Program Grant #00L01402 from the Japan Society for the Promotion of Science.

References Bockman DE, Boydston WR Gastroenterology, 85, 55–61.

and

Parsa

I.

(1983).

Brentnall TA, Bronner MP, Byrd DR, Haggitt RC and Kimmey MB. (1999). Ann. Intern. Med., 131, 247–255. Oncogene


2400 Chambers SK, Kacinski BM, Ivins CM and Carcangiu ML. (1997). Clin. Cancer Res., 3, 999–1007. Ciardiello F and Tortora G. (2001). Clin. Cancer Res., 7, 2958–2970. Crnogorac-Jurcevic T, Efthimiou E, Nielsen T, Loader J, Terris B, Stamp G, Baron A, Scarpa A and Lemoine NR. (2002). Oncogene, 21, 4587–4594. DiMagno EP, Reber HA and Tempero MA. (1999). Gastroenterology, 117, 1464–1484. Fang G, Kim CN, Perkins CL, Ramadevi N, Winton E, Wittmann S and Bhalla KN. (2000). Blood, 96, 2246–2253. Gianni L. (2002). Oncology, 63 (Suppl 1), 47–56. Gjerdrum LM, Lielpetere I, Rasmussen LM, Bendix K and Hamilton-Dutoit S. (2001). J. Mol. Diagn., 3, 105–110. Goggins M, Hruban RH and Kern SE. (2000). Am. J. Pathol., 156, 1767–1771. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, Bloomeld CD and Lander ES. (1999). Science, 286, 531–537. Greenlee RT, Hill-tHarmon MB, Murray T and Thun M. (2001). CA Cancer J. Clin., 51, 15–36. Guo H, Li R, Zucker S and Toole BP. (2000). Cancer Res., 60, 888–891. Han H, Bearss DJ, Browne LW, Calaluce R, Nagle RB and Von Hoff DD. (2002). Cancer Res., 62, 2890–2896. Hao D and Rowinsky EK. (2002). Cancer Invest., 20, 387–404. Hruban RH, Wilentz RE and Kern SE. (2000). Am. J. Pathol., 156, 1821–1825. Iacobuzio-Donahue CA, Maitra A, Shen-Ong GL, van Heek T, Ashfaq R, Meyer R, Walter K, Berg K, Hollingsworth MA, Cameron JL, Yeo CJ, Kern SE, Goggins M and Hruban RH. (2002). Am. J. Pathol., 160, 1239–1249. Ishiguro H, Shimokawa T, Tsunoda T, Tanaka T, Fujii Y, Nakamura Y and Furukawa Y. (2002). Oncogene, 21, 6387–6394. Ishiguro H, Tsunoda T, Tanaka T, Fujii Y, Nakamura Y and Furukawa Y. (2001). Oncogene, 20, 5062–5066. Kitahara O, Furukawa Y, Tanaka T, Kihara C, Ono K, Yanagawa R, Nita ME, Takagi T, Nakamura Y and Tsunoda T. (2001). Cancer Res., 61, 3544–3549. Klejman A, Rushen L, Morrione A, Slupianek A and Skorski T. (2002). Oncogene, 21, 5868–5876. Klinkenbijl JH, Jeekel J, Sahmoud T, van Pel R, Couvreur ML, Veenhof CH, Arnaud JP, Gonzalez DG, de Wit LT, Hennipman A and Wils J. (1999). Ann. Surg., 230, 776–784.

Laheru D, Biedrzycki B and Jaffee EM. (2001). Cancer J., 7, 324–337. Logsdon CD, Simeone DM, Binkley C, Arumugam T, Greenson JK, Giordano TJ, Misek DE and Hanash S. (2003). Cancer Res., 63, 2649–2657. Maney T, Hunter AW, Wagenbach M and Wordeman L. (1998). J. Cell Biol., 142, 787–801. Ng IO, Liang ZD, Cao L and Lee TK. (2000). Cancer Res., 60, 6581–6584. Okamura S, Arakawa H, Tanaka T, Nakanishi H, Ng CC, Taya Y, Monden M and Nakamura Y. (2001). Mol. Cell, 8, 85–94. Ono K, Tanaka T, Tsunoda T, Kitahara O, Kihara C, Okamoto A, Ochiai K, Takagi T and Nakamura Y. (2000). Cancer Res., 60, 5007–5011. Rehwald U, Schulz H, Reiser M, Sieber M, Staak JO, Morschhauser F, Driessen C, Rudiger T, Muller-Hermelink K, Diehl V and Engert A. (2003). Blood, 101, 420–424. Rosenberg L. (2000). Drugs, 59, 1071–1089. Rozenblum E, Schutte M, Goggins M, Hahn SA, Panzer S, Zahurak M, Goodman SN, Sohn TA, Hruban RH, Yeo CJ and Kern SE. (1997). Cancer Res., 57, 1731–1734. Saito-Hisaminato A, Katagiri T, Kakiuchi S, Nakamura T, Tsunoda T and Nakamura Y. (2002). DNA Res., 9, 35–45. Satoh S, Daigo Y, Furukawa Y, Kato T, Miwa N, Nishiwaki T, Kawasoe T, Ishiguro H, Fujita M, Tokino T, Sasaki Y, Imaoka S, Murata M, Shimano T, Yamaoka Y and Nakamura Y. (2000). Nat. Genet., 24, 245–250. Scanlan MJ, Welt S, Gordon CM, Chen YT, Gure AO, Stockert E, Jungbluth AA, Ritter G, Jager D, Jager E, Knuth A and Old LJ. (2002). Cancer Res., 62, 4041–4047. Shaw LM, Rabinovitz I, Wang HH, Toker A and Mercurio AM. (1997). Cell, 91, 949–960. Slamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Bajamonde A, Fleming T, Eiermann W, Wolter J, Pegram M, Baselga J and Norton L. (2001). N. Engl. J. Med., 344, 783–792. Sperti C, Pasquali C, Piccoli A and Pedrazzoli S. (1997). World J.Surg., 21, 195–200. Violette S, Festor E, Pandrea-Vasile I, Mitchell V, Adida C, Dussaulx E, Lacorte JM, Chambaz J, Lacasa M and Lesuffleur T. (2003). Int. J. Cancer, 103, 185–193. Yagyu R, Hamamoto R, Furukawa Y, Okabe H, Yamamura T and Nakamura Y. (2002). Int. J. Oncol., 20, 1173–1178. Yuan BZ, Miller MJ, Keck CL, Zimonjic DB, Thorgeirsson SS and Popescu NC. (1998). Cancer Res., 58, 2196–2199.

Supplementary information accompanies the paper on Oncogene website (http://www.nature.com/onc).

Oncogene