, & CANCER 51:300–312 (2012)

EGR1 and FOSB Expressions in Cancer Stroma Are Independent Prognostic Indicators for Epithelial Ovarian Cancer Receiving Standard Therapy

Fumio Kataoka,1 Hiroshi Tsuda,1* Tokuzo Arao,2 Sadako Nishimura,3 Hideo Tanaka,1 Hiroyuki Nomura,1 Tatsuyuki Chiyoda,1 Akira Hirasawa,1 Tomoko Akahane,1 Hiroshi Nishio,1 Kazuto Nishio,2 and Daisuke Aoki1 1Departmentof Obstetrics and Gynecology,School of Medicine,Keio University,Tokyo,Japan 2Departmentof Genome Biology,Kinki University School of Medicine,Osaka,Japan 3Departmentof Obstetrics and Gynecology,Osaka City General Hospital,Osaka,Japan

Stromal components interact with cancer cells to promote growth and metastasis. The purpose of this study was to iden- tify genes expressed in stroma, which could provide prognostic information in epithelial ovarian cancer (EOC). Seventy- four patients were included. We performed profiling and confirmed array data using RT-PCR and immuno- histochemistry. By microarray analysis, 52 candidate genes associated with progression free survival (PFS) were identified (P < 0.005). Expression of the early growth response 1 (EGR1) and FBJ murine osteosarcoma viral oncogene homolog B (FOSB) genes was further analyzed. Array data were confirmed by RT-PCR and multivariate analysis demonstrated that both EGR1 and FOSB expression in cancer stroma, and EGR1 expression in cancer are independent prognostic factors in EOC. Immunohistochemically, EGR1 is localized in cancer cells and a-smooth muscle actin positive stromal fibro- blasts. The EGR1 and FOSB expression in stromal cells and EGR1 expression in cancer cells are prognostic indicators in EOC. VC 2011 Wiley Periodicals, Inc.

INTRODUCTION ior, including invasion or metastasis and response Epithelial ovarian cancer (EOC) is a common to therapy (Bhowmick and Moses, 2005; Kim et al., cause of cancer death in women. Early diagnosis 2005; Tlsty and Coussens, 2006). However, the role of EOC is difficult because of a lack of specific of stromal cells in the development and progression symptoms so that 80% of patients are diagnosed of epithelial neoplasia has not been thoroughly at Stage III or IV. EOC is a relatively chemosen- investigated. Finak et al. used laser capture micro- sitive tumor, and 70% of advanced EOC (aEOC) dissection (LCM) to compare gene expression pro- patients treated by standard surgical procedures files of tumor stroma from 53 primary breast (bilateral adnexectomy þ hysterectomy þ greater cancers and established a new stroma-derived prog- omentectomy) with staging laparotomy and nostic predictor (SDPP) that stratifies disease out- debulking surgery, followed by postoperative come independently of standard clinical prognostic chemotherapy using a combination of platinum factors or published gene expression-based predic- and taxan, achieved a complete clinical response. tors (Finak et al., 2008). However, few studies have However, disease recurs in most patients within 2 examined the tumor–stroma interaction in clinical years after diagnosis and death is due to chemo- therapy-resistant tumor. If reliable predictive Additional Supporting Information may be found in the online markers could be established, patients who are version of this article. likely to relapse and die of disease are good can- Supported by: Grant-in-Aid for Scientific Research on Priority Areas from the Ministry of Education, Science and Culture, didates for clinical trials using new drugs. We Japan, Grant number: 20014024; A Grant-in Aid for scientific previously reported that the amplification of 8q24 Research (C) from the Ministry of Education, Science and Culture, Japan; Grant number: 19591940; Osaka City General and 20q11-13 or 17q21-q24 can predict progres- Hospital. sion-free survival (PFS) in EOC and may be new *Correspondence to: Hiroshi Tsuda, MD, PhD, Department of predictive biomarkers for aEOC (Hirasawa et al., Obstetrics and Gynecology, School of Medicine, Keio University, 35 Shinanomachi, Shinjyuku-ku, Tokyo, 160-8582, Japan. 2003; Tominaga et al., 2010). Cancer consists of Tel.: þ81-3-3353-1211 (Ext 61721). Fax: þ81-3-3353-0249. founder cancer cells and stroma, including blood E-mail: [email protected] and lymph endothelial cells, inflammatory cells, Received 20 June 2011; Accepted 27 October 2011 DOI 10.1002/gcc.21916 immunocytes, macrophages and fibroblasts. Published online 15 November 2011 in Stroma is thought to play a role in tumor behav- Wiley Online Library (wileyonlinelibrary.com).

VC 2011 Wiley Periodicals, Inc. A STROMAL BIOMARKER IN OVARIAN CANCER 301 samples of EOC. Here, we microdissected cancer Study Design stroma from aEOC by LCM, performed expression Seventy-four aEOC samples were evaluated in profiling, and analyzed the relationship between this study. We performed microarray analysis on the expression profile pattern and prognosis. microdissected stroma from 24 aEOCs and genes of interest were evaluated in an independent set of 50 samples. PFS was predicted using the MATERIALS AND METHODS results of real-time PCR analysis. Patients and Samples Subjects eligible for this study were patients Microdissection, RNA Extraction, and with histologically confirmed Stage IIc–IV aEOC Amplification of RNA (excluding mucinous and clear cell types) receiv- Microdissection was performed as described ing standard therapy. Additional inclusion criteria (Tsuda et al., 2004, 2005). In brief, frozen sec- included an Eastern Cooperative Oncology Group tions (6 lm) prepared from tumor tissue speci- performance status of 0–2. Clinical stage and his- mens were affixed to glass slides and stained by tologic grade were determined in accordance with HistogeneTM LCM Frozen Section Staining Kit the International Federation of Gynecology and (Arcturus Engineering, Mountain View, CA). Obstetrics (FIGO) systems. Undifferentiated his- Stained sections were microdissected using a Pix- tology was treated as Grade 3. Exclusion criteria Cell IIe LCM system (Arcturus Engineering). included a history of prior chemotherapy or major Tumor cells and adjacent non-tumor stromal cells surgery. All patients received standard surgery were visualized under the microscope and selec- and chemotherapy using carboplatin and pacli- tively detached by activation of the laser. Stromal taxel. Standard surgery means bilateral adnexec- tissues within 200 cells from the margin of tomy, hysterectomy, and greater omentectomy tumors were dissected in each case. Total RNA with staging laparotomy and debulking surgery. extraction was performed using the PicoPureTM Samples used for microarray analysis were from RNA Isolation Kit according to the manufac- 24 aEOC patients with a median age of 61 years turer’s instructions (Arcturus Engineering). RNA (range 41–80), 4 were Stage II, 14 Stage III, and was amplified using a modified single round T7 6 Stage IV. Three tumors were endometrioid, 17 RNA amplification protocol. In brief, total RNA serous, and 4 of undifferentiated histologic type, (600 ng) was first incubated with 1 llofT7 while 1 was histologic Grade G1, 7 G2, and 16 primer (50-GCATTAGCGGCCGCGAAATTAAT G3. Ten patients were of optimal and 14 of sub- ACGACTCACTATAGGGAGATTTTTTTTT optimal operation status. The median follow-up TTTTTTTTTVN-30, 200 ng/ll) in a total vol- period was 1,006 days (range: 339–2,283 days) ume of 50 ll for 3 min at 70C. First strand and seven patients are alive without relapse. cDNA synthesis was then performed by incubat- Fifty independent aEOC samples used for valida- ing 5 ll of primer-annealed sample and 5 llof tion by reverse transcription polymerase chain reac- first strand master mix containing 2 llof5 first tion (RT-PCR) assay were from patients with a strand buffer, 1 ll of 0.1 M Dithiothreitol (DTT), median age of 54 years (range 29–80), 7 patients 0.5 ll of Diethyl polycarbonate (DEPC) water, were Stage II, 33 Stage III, and 10 Stage IV. Eleven 0.5 ll 10 mM deoxyribonucleotide triphosphate tumors were endometrioid, 31 serous, and 8 of undif- (dNTP) mix, 0.5 ll RNase Inhibitor, and 0.5 llof ferentiated histologic type while 2 were histologic Moloney Murine Leukemia Virus Reverse Tran- Grade G1, 16 G2, and 32 G3. Twenty-seven patients scriptase (MMLV) (200 U/ll)for1hand15minat were of optimal and 23 of suboptimal operation sta- 37C. Subsequently, second strand cDNA synthesis tus. The median follow-up period was 836 days was performed by incubating the10 ll first strand (range: 191–2,885 days) and 20 patients are alive reaction with 65 ll of second master mix, which without relapse. All cases are shown in Table 1. contained 46 llDEPCwater,15ll5 second PFS was followed after the patients had strand buffer, 1.5 llof10mMdNTPmix,0.5ll received primary surgery. This study was approved of E. coli DNA Ligase (10 U/ll), 1.5 ll E. coli by the Institutional Review Board of the Osaka DNA polymerase I (10 U/ll), and 0.5 ll E. coli City General Hospital and School of Medicine, RNase H (2 U/ll), for 2 h at 16C,andthenfor Keio University and written informed consent was 15 min at 70C.Theentire75llcDNAsample obtained from all patients. Specimens obtained at was loaded onto a ChromaSpin TE-200 spin operation were immediately stored at 80C. column (BD Biosciences, San Diego, CA), which

Genes, Chromosomes & Cancer DOI 10.1002/gcc 302 KATAOKA ET AL.

TABLE 1. Clinical Background and EGR1 and FOSB Expression in All Cases

Debulking EGR1s EGR1c EGR1s EGR1c FOSBs FOSBc Clinical No. Age Stage Histology Grade status PCR PCR IHC IHC PCR PCR outcome 1 55 4 ECC 3 Optimal prog 2 52 3c SEC 3 Suboptimal prog 3 63 2c ECC 3 Optimal nprog 4 64 3c SEC 2 Suboptimal prog 5 61 3c UND 3 Optimal prog 6 49 2c SEC 2 Optimal prog 7 55 2c SEC 2 Optimal nprog 8 70 2c ECC 3 Optimal nprog 9 69 3c UND 3 Suboptimal nprog 10 72 4 UND 3 Suboptimal prog 11 51 3c UND 3 Suboptimal prog 12 51 3c SEC 2 Suboptimal prog 13 65 3c SEC 3 Suboptimal nprog 14 63 3c SEC 3 Optimal nprog 15 63 3c SEC 2 Suboptimal prog 16 78 4 SEC 2 Suboptimal prog 17 53 4 SEC 2 Optimal prog 18 41 3c SEC 3 Suboptimal prog 19 75 4 SEC 3 Optimal prog 20 71 3c SEC 3 Suboptimal prog 21 49 3c SEC 3 Suboptimal prog 22 45 3c SEC 1 Optimal nprog 23 54 3c SEC 3 Optimal nprog 24 80 4 SEC 3 Optimal prog 25 54 3c SEC 3 Optimal 5.3 3.0 20 50 10.2 3429 nprog 26 55 4 ECC 3 Optimal 142 22.1 40 10 899 330 prog 27 48 3c UND 3 Suboptimal 23.4 6.2 10 20 465 29.1 prog 28 47 3c SEC 3 Optimal 178 12.1 nd nd 4924 11.0 nprog 29 57 4 SEC 2 Suboptimal 10.3 4.0 50 60 25.8 11.0 prog 30 52 3c SEC 3 Suboptimal 183 34.2 60 40 175 62.5 prog 31 52 3c UND 3 Suboptimal 5.3 8.1 20 20 32.5 15.6 prog 32 29 3c ECC 4 Optimal 5.9 1.1 5 0 15.4 1.4 nprog 33 49 3c SEC 2 Suboptima 16.9 14.1 30 40 5.0 1.8 prog 34 54 4 ECC 3 Suboptimal 44.3 5.1 60 40 280 8.4 prog 35 43 4 ECC 3 Suboptimal 622 272 50 80 3264 3482 prog 36 66 3c ECC 1 Optimal 222 11.0 nd nd 930 41.2 prog 37 63 2c ECC 3 Optimal 9.1 1.0 30 15 20.6 5.2 nprog 38 66 3c ECC 1 Suboptimal 14.0 4.4 10 25 44.1 3.2 prog 39 64 3c SEC 2 Suboptimal 3.6 0.0 80 80 7.0 8574 prog 40 61 3c UND 3 Optimal 48.3 0.0 50 5 502 1.1 prog 41 49 2c SEC 2 Optimal 23.0 0.0 nd nd 65.2 1.0 prog 42 61 3c UND 3 Suboptimal 10.1 13.4 20 20 22.4 109 nprog 43 55 2c SEC 2 Optimal 2.6 2.1 0 0 5.6 4.5 nprog 44 70 2c ECC 3 Optimal 78.8 0.0 10 20 130 16.7 nprog 45 56 3c SEC 2 Optimal 3.6 5.2 nd nd 3.9 1.29 prog 46 54 2c SEC 3 Optimal 52.0 0.0 10 5 84.4 41.2 nprog 47 69 3c UND 3 Suboptimal 1.8 0.0 nd nd 9.1 1.0 nprog 48 72 4 UND 3 Suboptimal 26.4 7.2 40 15 75.1 82.5 prog 49 43 3c SEC 2 Optimal 8.7 0.0 20 0 10.7 4.5 nprog 50 51 3c UND 3 Suboptimal 10.5 0.0 nd nd 339 14.6 prog 51 44 2c ECC 2 Optimal 5.3 6.0 0 5 114 707 nprog 52 56 3c UND 3 Optimal 6.1 5.5 nd nd 0.0 9.0 prog 53 51 3c SEC 2 Suboptimal 7.7 4.2 30 5 30.8 13.6 prog 54 65 3c SEC 3 Suboptimal 3.2 0.0 10 20 14.0 2.8 nprog 55 63 3c SEC 3 Optimal 2.6 0.0 10 10 5.5 1.0 nprog 56 50 3c SEC 3 Optimal 1.3 8.2 nd nd nd 435 nprog 57 37 3c SEC 2 Optimal 7.5 6.2 30 10 nd 536 nprog 58 56 3c ECC 3 Optimal 2.8 0.0 nd nd 7.1 1.1 nprog 59 63 3c SEC 2 Suboptimal 39.8 8.0 nd nd 69.1 nd prog (Continued)

Genes, Chromosomes & Cancer DOI 10.1002/gcc A STROMAL BIOMARKER IN OVARIAN CANCER 303

TABLE 1. Clinical Background and EGR1 and FOSB Expression in All Cases (Continued) Debulking EGR1s EGR1c EGR1s EGR1c FOSBs FOSBc Clinical No. Age Stage Histology Grade status PCR PCR IHC IHC PCR PCR outcome 60 78 4 SEC 2 Suboptimal 3.6 2.1 nd nd 42.9 6.4 prog 61 53 4 SEC 2 Optimal 4.8 6.1 nd nd 2.5 nd prog 62 41 3c SEC 3 Suboptimal 12.4 1.0 nd nd 257 23.7 prog 63 75 4 SEC 3 Optimal 12.3 0.0 nd nd 4.0 3.4 prog 64 71 3c SEC 3 Suboptimal 32.8 237 nd nd 417 4525 prog 65 49 3c SEC 3 Suboptimal 23.0 9.2 nd nd 40.4 20.6 prog 66 52 3c SEC 2 Optimal 4.4 1.3 nd nd 1.1 1.0 nprog 67 71 3c SEC 3 Suboptimal 74.6 4.0 nd nd 2603 308 prog 68 45 3c SEC 2 Optimal 10.0 4.0 nd nd nd 17.9 nprog 69 54 3c SEC 3 Optimal 2.0 0.0 nd nd 4.8 0.0 nprog 70 65 4 SEC 3 Suboptimal 4.6 0.0 nd nd 9.1 1.2 nprog 71 80 4 SEC 3 Optimal 278 10.5 nd nd 557 nd prog 72 48 2c SEC 3 Optimal 4.9 6.1 nd nd 2.8 nd prog 73 49 3c ECC 3 Optimal 2.2 0.0 nd nd 1.0 nd nprog 74 56 3c SEC 3 Optimal 89.2 9.1 nd nd 454 nd prog

No. 1–24: reference sets, no. 25–74: test sets, SEC: serous type, ECC: endometrioid type, UND: undifferentiated type, EGR1s: stromal EGR1 expression, EGR1c: cancer EGR1 expression, FOSBs: stromal FOSB expression, FOSBc: cancer FOSB expression, IHC: immunohistochemistry, prog: progressive disease, nprog: non-progressive disease. was centrifuged for 5 min at 2,900 rpm (700 g)in verted to numerical data using the GeneChip an Eppendorf centrifuge. Purified cDNA was col- Operating Software, Ver.1 (Affymetrix). Candi- lected, lyophilized, dissolved in 8 llofRNase-free date was analyzed using Kyoto water, and incubated at 70C for 10 min. In vitro encyclopedia of genes and genomes (KEGG). transcription was subsequently performed by incu- bating the 8 ll post-lyophilization cDNA product with 12.2 llofmastermixcontaining2llof10 Real-Time Quantitative RT-PCR T7 reaction buffer, 6 llof25mMrNTPMix,2 The method has been described previously ml of 100 mM DTT, 0.2 ll of RNase inhibitor (Tsuda et al., 2004, 2005). In brief, mRNA copy (40 U/ml), and 2 ll of T7 RNA polymerase for 3 numbers were validated for FOSB and EGR1 hat37C. The amplified RNA was purified on an gene expression based on fold-change and gene RNeasy mini column (Qiagen, Valencia, CA) as function. All results were normalized to the per the manufacturer’s protocol. The purified amount of glyceraldehyde 3 phosphate dehydro- amplified RNA was quantified with RiboGreen genase (GAPDH, NM_002046). RNA was con- RNA Quantitation Reagent (Molecular Probes, verted to cDNA using a GeneAmp RNA PCR Eugene, OR). Core kit (Applied Biosystems, Foster City, CA). The cDNAs were quantified using the Power SYBR Green PCR Master Mix (Applied Biosys- Oligonucleotide Microarray Study and Analysis of tems) and 7900HT fast real-time PCR system Gene Ontology (Applied Biosystems) and reported relative to the The microarray procedure was performed GAPDH expression levels. PCR conditions were according to Affymetrix protocols (Santa Clara, as follows: one cycle of denaturation at 95C for CA). In brief, the total RNA extracted from the 10 min, followed by 40 cycles at 95C for 15 sec tumor stromal samples was checked for quality and 60C for 60 sec. To amplify the target genes, using an Agilent 2100 Bioanalyzer (Agilent Tech- the following primers were purchased from nologies, Waldbronn, Germany) and cRNA was Takara (Yotsukaichi, Japan): FOSB-FW: 5-ctg cca 0 synthesized using the GeneChipVR 3 -amplifica- atg ctc cag ctg tc-3, RW: 5-act cgc acc cag aat tgt tion reagents one-cycle cDNA synthesis kit (Affy- caa ag-3, EGR1-FW: 5-gta cag tgt ctg tgc cat gga metrix). Labeled cRNAs were purified and used ttt c-3, RW: 5-gag gat cac cat tgg ttt gct tg-3, for construction of probes. Hybridization was GAPDH-FW: 5-gca ccg tca agg ctg aga ac-3, RW: performed using the Affymetrix GeneChip 5-atg gtg gtg aag acg cca gt-3. Cases were classi- HG-U133 Plus2.0 array for 16 h at 45C. fied as exhibiting high or low expression based Signal intensities were measured using a on the median mRNA copy number of the EGR1 GeneChipVR Scanner3000 (Affymetrix) and con- and FOSB gene.

Genes, Chromosomes & Cancer DOI 10.1002/gcc 304 KATAOKA ET AL.

Immunohistochemical Analysis tion; Molecular Probes–Invitrogen) for nucleic EGR1 protein expression was detected immuno- acid staining. histochemically using a rabbit anti-EGR1 antibody (15F7) (1:100 dilution; Cell Signaling Technology, Statistical Analysis Danvers, MA). In 25 of 50 independent EOC Microarray analysis was performed using BRB cases, formalin-fixed paraffin-embedded specimens Array Tools software ver. 3.3.0 (http://linus.nci.- were available for analysis. Histological sections nih.gov/BRB-ArrayTools.html) developed by Dr. l (4 m) were affixed to glass slides, dewaxed, and Richard Simon and Amy Peng. In brief, a log rehydrated. The sections were incubated in 3% base 2 transformation was applied to the raw hydrogen peroxide for 10 min at room temperature microarray data, and global normalization was to quench endogeneous peroxidase activity, and used to calculate the median over the entire incubated with the anti-EGR1 antibody at 4 C array. Genes were excluded if the percentage of overnight. After rinsing, the sections were incu- data missing or filtered out exceeded 20%. Genes þ bated for 30 min with rabbit EnVision Peroxi- that passed the filtering criteria were then consid- dase (DAKO, CA). Peroxidase activity was ered for further analysis. We computed a statisti- visualized by applying diaminobenzidine chromo- cal significance level (P < 0.01) and fold change gen containing 0.05% hydrogen peroxide for 90 sec >1.5 or <0.67 for each gene based on a univari- at room temperature, and the sections were coun- ate proportional hazard model. terstained with hematoxylin. Slides were observed The mRNA copy numbers were validated for by two independent pathologists, who were the FOSB and EGR1 genes using real time PCR blinded to the clinical background of the patients. in 50 independent cases of aEOC. PFS was We calculated labeling index in both cancer and calculated by the Kaplan–Meier method and stromal cells. log rank test. Univariate and multivariate Cox’s proportional hazard tests were applied to identify Immunolocalization of EGR1 in Cancer and variables associated with PFS. A P values of Stromal Tissues <0.05 was considered to denote significance (SAS We analyzed the immunolocalization of EGR1 software ver. 9.1.3; SAS Institute Inc., Cary, NC). protein in six cases. Tissues were dissected and frozen in optimum cutting temperature (OCT) RESULTS embedding medium (Sakura Finetec, Torrance, CA) using dry-ice-cooled isopentane and sec- Identification of 52 Candidate Disease tioned at 8 lm on a Leica cryostat (Leica, Wetzlar, Progression-Related Genes by Microarray Analysis Germany). Cryosections were fixed with phos- To identify candidate disease progression- phate-buffered 4% formaldehyde for 10 min at related genes from 54,675 transcripts, microarray room temperature and washed in phosphate-buf- analysis was performed on a training set of 24 fered saline (PBS). Non-specific binding sites samples. A total of 18,563 genes passed the filter- were blocked with 5% goat serum albumin (GSA) ing criteria and were further analyzed. Fifty-two in PBS for 1 h at room temperature and then incu- genes were significantly correlated with disease bated in primary and secondary antibodies progression, with a P-value of <0.01 and fold (diluted in PBS with 2.5% GSA) for 2 h at room change >1.5 or <0.67 (Fig. 1). Thirty-seven temperature. Cryosections were washed thor- genes were overexpressed and 15 underexpressed oughly in PBS before and after each incubation in progressive compared to non-progressive dis- step. Specimens were examined with an LSM 510 ease cases. The top 10 overexpressed and under- META microscope (Carl Zeiss). We used the fol- expressed genes are shown in Table 2. lowing primary antibodies: rabbit anti-EGR1 anti- The ontology of these 52 candidate genes was body (15F7) (1:100 dilution; Cell Signaling analyzed using KEGG (Supporting Information Technology, USA) and mouse anti-alpha smooth Table 1). When the genes were classified by bio- muscle actin antibody (1:400 dilution; abcam 1A4 logical process, five genes involved in regulation of (ab7817)). As secondary antibodies, we used transcription, five further transcription-related chicken anti-Rabbit Alexa 488 conjugated IgG (5 genes, three transport-related genes, and three pro- lg/ml), goat anti-mouse Alexa 546-conjugated IgG teolysis-related genes were detected. Classification (5 lg/ml) from Molecular Probes–Invitrogen by cellular component identified 15 nuclear genes, (Carlsbad, CA) and To-PRO-3 iodide (1:400 dilu- 7 cytoplasmic genes, 7 membrane genes, and

Genes, Chromosomes & Cancer DOI 10.1002/gcc A STROMAL BIOMARKER IN OVARIAN CANCER 305

Figure 1. Fifty two differentially expressed genes in progressive and non-progressive aEOC cases. Red spots indicate non-progressive cases.

4 integral membrane genes. Classification by mo- performed on the microdissected stromal tissue of lecular function identified 6 protein binding genes 50 independent EOC samples. Cases were classi- and 2 DNA binding genes. These data are sum- fied as exhibiting high or low expression based marized in Supporting Information Tables 2 and 3. on the median mRNA copy number of each gene. FOSB gene expression in cancer stroma was significantly inversely correlated with PFS in FOSB EGR1 Relationship between or Gene aEOC, (median survival for high-expressing Expression and PFS tumors 793 days (95% CI: 686–900 days) vs. Among the 52 candidate genes, we focused on 1,263 days (95% CI: 321–2,205 days), P ¼ 0.01 the FOSB and EGR1 genes based on fold change for low-expressing tumors). However, its expres- and their known functions. Real time RT-PCR sion in cancer cells was not correlated with PFS analysis of FOSB and EGR1 gene expression was in aEOC (Figs. 2a and 2b). EGR1 gene

Genes, Chromosomes & Cancer DOI 10.1002/gcc 306 KATAOKA ET AL.

TABLE 2. Top 10 Overexpressed and Underexpressed Genes Selected from 52 Candidate PFS-Related Genes Identified by Microarray Analysis

Fold No. Gene name Symbol change Gene ontology 1 FBJ murine osteosarcoma viral FOSB 19q13.32 5.3 Multicellular organismal development, oncogene homolog B negative regulation of transcription from RNA polymerase II promoter, binding 2 Early growth response 3 EGR3 8p23-p21 2.5 Circadian rhythm, muscle organ development, transcription factor activity 3 Early growth response 1 EGR1 5q31.1 2.4 Positive regulation of transcription, transcription factor activity 4 Sema domain, immunoglobulin SEMA4B 15q25 2.3 Cell differentiation, multicellular domain, transmembrane organismal development, nervous domain and short cytoplasmic system development domain, 4B 5 Transcribed locus - - 2.3 - 6 Dickkopf homolog 2 (Xenopus DKK2 4q25 2.2 Multicellular organismal development, laevis) negative regulation of Wnt signaling pathway 7 Regulatory factor X, 2 (influences RFX2 19p13.3-p13.2 2.1 DNA binding, transcription regulator HLA class II expression) activity, regulation of transcription, DNA-dependent 8 Carboxypeptidase X member 1 CPXM1 20p13-p12.3 2.1 Cell adhesion, proteolysis 9 Chromosome 6 open reading C6orf189 6q22.2 1.9 - frame 189 10 Leukotriene B4 12- LTB4DH 9q31.3 1.9 Leukotriene metabolic process, hydroxydehydrogenase oxidation reduction 1 Major histocompatibility complex, HLA-DQA1 /// 06p21.3 0.3 Antigen processing and presentation class II, DQ alpha 1 /// major HLA-DQA2 /// of peptide or polysaccharide histocompatibility complex, LOC650946 antigen via MHC class II, immune class II, DQ alpha 2 /// response similar to HLA class II histocompatibility antigen, DQ(1) alpha chain precursor (DC-4 alpha chain) 2 Major histocompatibility complex, HLA-DRB6 6p21.3 0.4 - class II, DR beta 6 (pseudogene) 3 Mannose receptor, C type 1 /// MRC1 /// MRC1L1 10p12.33 0.5 - mannose receptor, C type 1-like 1 4 GTPase, IMAP family member 1 GIMAP1 7q36.1 0.5 Receptor-mediated endocytosis /// endocytosis GTP binding 5 Phosphoinositide-binding protein PIP3-E 6q25.2 0.5 Oxygen transport, response to oxidative stress 6 Chromosome 17 open reading C17orf27 17q25.3 0.5 Metal ion binding, nucleoside-triphos- frame 27 phatase activity, nucleotide binding, protein binding, zinc ion binding 7 Solute carrier family 16,member 6 SLC16A6 17q24.2 0.5 Monocarboxylic acid transport, transport 8 Chromosome 6 open reading C6orf211 6q25.1 0.6 Protein binding frame 211 9 Tumor necrosis factor receptor TNFRSF11A 18q22.1 0.6 Cell–cell signaling, positive regulation superfamily, member 11a, of cell proliferation, signal NFKB transduction, sensory perception of sound 10 Zinc finger and BTB domain ZBTB1 14q23.3 0.6 Regulation of transcription containing 1 (DNA-dependent), transcription

Genes, Chromosomes & Cancer DOI 10.1002/gcc A STROMAL BIOMARKER IN OVARIAN CANCER 307

Figure 2. The relationship between FOSB and EGR1 expression aEOC. Median survival with high EGR1 expression 801 days (95% CI: and PFS. A: FOSB gene expression in cancer stroma is significantly 675–927 days) vs. 1,263 days with low expression (95% CI: 990–1536 correlated with PFS in aEOC. Median survival with high FOSB expres- days) (P ¼ 0.016). D: EGR1 gene expression of cancer cells is signifi- sion 793 days (95% CI: 686–900 days) vs. 1263 days with low expres- cantly correlated with PFS in aEOC. Median survival with high EGR1 sion (95% CI: 321–2,204 days) (P ¼ 0.01). B: FOSB gene expression expression 1,036 days (95% CI: 809–1,263 days) vs. 2,020 days with of cancer cells is not correlated with PFS (P ¼ 0.188). C: EGR1 gene low expression (95% CI: 1,602–2,438 days). expression in cancer stroma is significantly correlated with PFS in

TABLE 3. Univariate and Multivariate Analyses of the Effect of Various Prognostic Factors on PFS (Stroma) and PFS (Cancer Cell)

Univariate Multivariate

Variablea Hazard ratiob 95% CI P Hazard ratiob 95% CI P PFS (stroma) EGR1 2.614 1.157–5.908 0.021 2.559 1.093–5.993 0.03 Age 1.454 0.678–3.077 0.327 1.671 0.754–3.704 0.206 Histologic grade 1.423 0.660–3.070 0.368 1.719 0.738–4.002 0.209 Debulking status 0.363 0.168–0.785 0.01 0.395 0.178–0.874 0.022 PFS (cancer cell) EGR1 3.159 1.397–7.142 0.006 2.598 1.131–5.968 0.024 Age 1.160 0.557–2.418 0.692 0.793 0.348–1.806 0.58 Histologic grade 1.512 0.7–3.266 0.293 1.115 0.508–2.451 0.786 Debulking status 0.218 0.094–0.507 0.001 0.223 0.086–0.578 0.002 aEGR1: overexpression vs. normal-expression in cancer cells, Age: >median vs.

Genes, Chromosomes & Cancer DOI 10.1002/gcc 308 KATAOKA ET AL.

TABLE 4. Univariate and Multivariate Analysis of the Effect of Various Prognostic Factors on PFS (stroma)

Univariate Multivariate Variablea Hazard ratiob 95% CI P Hazard ratiob 95% CI P FOSB 3.254 1.452–7.239 0.004 3.433 1.457–8.088 0.005 Age 1.778 0.824–3.837 0.294 1.867 0.838–4.162 0.127 Histologic grade 1.279 0.594–2.751 0.529 1.272 0.573–.826 0.554 Debulking status 2.809 1.300–.098 0.009 3.125 1.368–.143 0.007 aFOSB: overexpression vs. normal-expression in cancer stroma, Age: >median vs.

Figure 3. Correlation of EGR1 and FOSB expression between cancer and stroma. A: EGR1 gene expression between cancer and stroma was weakly correlated (r ¼ 0.64, P < 0.01). B: FOSB gene expression between cancer and stroma was not correlated. C: EGR1 protein expression of both cancer and stroma was higher in progressive cases than it in non-progressive cases. D: Typical image of EGR1 protein expression (400). expression in cancer stroma was significantly mal or cancer EGR1 expression alone. In 31 inversely correlated with PFS in aEOC (median serous cases, EGR1 stromal and cancer cells survival for high-expressing tumors 801 days expression tended to be related with and PFS, (95% CI: 675–927 days) vs. 1,263 days for low however, it was not significant (data not shown). expressing tumors (95% CI: 990–1,536 days), P ¼ In addition, we performed univariate and multi- 0.016). EGR1 gene expression in cancer cells was variate Cox proportional hazard tests to identify also inversely correlated with PFS (median sur- variables associated with PFS including age, his- vival for high expressing tumors 1,036 days (95% tologic grade, debulking status, and FOSB or CI: 809–1,263 days) vs. 2,020 days for low EGR1 gene mRNA copy number of cancer cells expressing tumors (95% CI: 1,602–2,438 days), P or cancer stroma. This analysis showed that ¼ 0.016) (Figs. 2c and 2d). Combined stromal EGR1 gene expression in both cancer cells and and cancer EGR1 gene expression was not a bet- cancer stroma are independent prognostic factors ter prognostic predictor compared to either stro- in aEOC (Table 3). FOSB expression in cancer

Genes, Chromosomes & Cancer DOI 10.1002/gcc A STROMAL BIOMARKER IN OVARIAN CANCER 309

TABLE 5. Immunolocalization of EGR1 Protein in 6 Cases

No. Age ST HIS GR EGR1 cancer EGR1 stroma 1 63 4 SC 3 þþ 2 61 1cb SC 2 þþ 3343cSC1 þ 4532aSC1 þþ 5593cSC2 þþ 6533cSC1 þþ

ST: FIGO stage, HIS: histology, EC: endometrioid adenocarcinoma, SC: serous cystadenocarcinoma, UD: undifferentiated carcinoma, GR: Histologic grade.

Figure 4. Immunolocalization of EGR1. Orange color aSMA, green in cancer stroma (2003). B: Case 4, EGR1 expression was detected color EGR1, and blue nucleus. EGR1 expression is localized in cancer in both cancer cells and aSMA positive cells in cancer stroma cells and aSMA positive spindle shaped stromal cells around the bor- (4003). C, D: Case 5, EGR1 expression was detected in both cancer der between cancer cells and cancer stroma. A: Case 4, EGR1 cells and aSMA positive cells in cancer stroma (4003). expression was detected in both cancer cells and aSMA positive cells stroma is also an independent prognostic factor, higher than in 12 non-progressive cases (40.8 (95% but its expression in cancer cells is not (Table 4). CI: 28.32–53.22) vs. 13.8 (95% CI: 7.24–20.26), P EGR1 gene expression in cancer stroma and ¼ 0.001) (Fig. 3c). EGR1 protein expression cancer cells was weakly correlated, however, between cancer cells and cancer stroma was corre- FOSB gene expression in stromal and cancer cells lated (r ¼ 0.659, P < 0.001). A typical image of was not correlated (Figs. 3a and 3b). Further- EGR1 protein expression is shown in Figure 3d. more, we evaluated EGR1 protein expression in All data are shown in Table 1. 25 cases. Mean labeling index of EGR1 of cancer in 13 progressive cases was significantly higher than in 12 non-progressive cases (33.9 (95% Immunolocalization of EGR1 Protein CI: 18.14–49.55) vs. 12.9 (95% CI: 3.99–21.9), We analyzed the immunolocalization of EGR1 P ¼ 0.02) (Fig. 3c). Mean labeling index of EGR1 protein in six cases. The results are shown in in stroma in 13 progressive cases was significantly Table 5. EGR1 protein was detected in cancer

Genes, Chromosomes & Cancer DOI 10.1002/gcc 310 KATAOKA ET AL. tissue in 5 of 6 cases and in alpha smooth muscle transcription and none to the immune response active (aSMA) positive stromal cells in all six or signal transduction. However, Finak’s SDPP cases. aSMA and EGR1 positive cells were seen gene set included three (12%) immune response at the border between cancer cells and stromal related genes and four (15%) signal transduction cells and were spindle shaped. A typical appear- related genes. EOC cells are believed to originate ance is shown in Figures 4a–4d. from the cells covering the ovarian surface and disseminate directly to the abdominal cavity (Naora and Montell, 2005). In contrast, breast DISCUSSION cancer cells spread by invasion of surrounding Cancer stroma is thought to play an important stromal tissue. These findings suggest that the role in tumor behavior including invasion or me- role of cancer stroma may differ among cancer tastasis and response to therapy (Bhowmick and types. Finally, we suggest that gene ontology Moses, 2005; Kim et al., 2005; Tlsty and Cous- analysis may be useful in evaluating the role of sens, 2006). Dakhova et al. demonstrated that for- cancer stroma among different cancers. mation of reactive stroma in prostate cancer is a Among the 52 candidate genes, we focused on dynamic process characterized by significant alter- FOSB and EGR1 based on fold change and their ations in growth factor and signal transduction known functions. The FOSB gene was reported pathways and formation of new structures, includ- to be a proto-oncogene. Milde-Langosch et al. ing nerves and axons (Dakhova et al., 2009). demonstrated that MCF7 cell line invasion was Gregg et al. examined the patterns of differential enhanced by FOSB and FOSB levels were signif- expression between cancer cells and cancer icantly associated with MMP1 overexpression stroma in prostate cancer using microdissection (Milde-Langosch et al., 2004). Recently, it was and expression profile, and demonstrated that reported that neurobehavioral stress leads to elevated expression of zinc finger transcription FOSB-driven increases in IL8 in ovarian cancer factors (WT1, EGR1, and GATA2) and growth cell lines, which is associated with increased tu- factor receptors (IGF-1R and FGF-R3) was mor growth and metastasis (Shahzad et al., 2010). observed in the cancer cells, while expression of In our study, FOSB gene overexpression in can- chemokines (CCL5 and CXCL13) and growth cer stroma is an independent prognostic factor; factor ligands (IGF-1, FGF-2, and IGFBP3) was however, its expression in cancer cells is not found in stroma (Gregg et al., 2010). Finak et al. related to prognosis of aEOC. In contrast, EGR1 performed gene expression profiling of microdis- gene expression in both cancer cells and cancer sected tumor stroma from 53 primary breast can- stroma were related with PFS, and multivariate cers and established a new stroma-derived analysis demonstrated that EGR1 gene expression prognostic predictor (SDPP) consisting of 26 in both cancer cells and cancer stroma are inde- genes (Finak et al., 2008). In addition, they dem- pendent prognostic factors in aEOC. Further- onstrated that tumor stroma samples from good- more, we evaluated EGR1 protein expression in outcome patients overexpress a distinct set of 25 cases. Mean labeling index of EGR1 of both immune-related genes, including T cell and NK cancer and stroma in 13 progressive cases was sig- cell markers (GZMA, CD52, CD247, and CD8A). nificantly higher than in 12 non-progressive cases. In contrast, stroma samples from poor-outcome EGR1 is a multifunctional transcription factor patients show markers of increased hypoxia and that is considered to be a tumor suppressor gene angiogenic response, including the matrix metal- because it directly regulates TP53, PTEN, and lopeptidases ADM, MMP1, and SPP1. In the pres- TGFb1 and acts as suppressor gene in esophageal ent study, stroma samples from poor-outcome cancer, non-small cell lung cancer, and neuroblas- patients overexpress the oncogene FOSB, several toma (Wu et al., 2004; Ferraro et al., 2005). In transcription related (EGR1, EGR3, and RFX2) contrast, EGR1 has also oncogenic function in and proteolysis related genes (CPXM1, IMMP1L) gastric cancer and prostate cancer (Kobayashi but in contrast, underexpress human leukocyte et al., 2002; Adamson et al., 2003; Virolle et al., antigen (HLA) related genes (HLA-DQA1/ 2003). In hypoxia, EGR1 was reported to up-reg- HLADQ2, HLA-DR6) and the apoptosis related ulate HIF-1 alpha (Sperandio et al., 2009). Many gene CASP1. There was no overlap in genes other putative EGR1 target genes related to can- between Finak’s report and our study. Gene on- cer have been identified, including cyclin D, tology analysis of 52 candidate genes demon- EGFR, FGF, IGF-I, thymidine kinase, PDGF-A, strated that 11 of 52 (21%) genes are related to Bcl2, CD44, , PTEN, TNFa, and VEGF.

Genes, Chromosomes & Cancer DOI 10.1002/gcc A STROMAL BIOMARKER IN OVARIAN CANCER 311 Saegusa et al. demonstrated that transfection of cells were derived from the inoculated tumor EGR1 up-regulated TCF4 or p300 leading to nu- cells. Functional studies of EGR1 in both cancer clear accumulation of beta-catenin (Saegusa et al., cells and stromal cells are essential to solve these 2008). Fahmy et al. demonstrated that EGR1 problems. supports FGF2-dependent angiogenesis during In conclusion, EGR1 and FOSB gene expres- neo-vascularization and tumor growth (Fahmy sions in fibroblasts of cancer stroma are independ- et al., 2003). Barbolina et al. put forward a model ent prognostic indicator in aEOC. for intraperitoneal metastasis in ovarian cancer, wherein collagen adhesion and clustering of colla- ACKNOWLEDGMENTS gen binding integrins of cancer cells induces The authors are grateful to Asami Nagata, expression of EGR1, resulting in transcriptional Kanako Matsumoto, and Nozomi Tsuji for their activation of the membrane type 1 matrix metal- technical assistance. loproteinase (MT1-MMP) promoter and subse- quent MT1-MMP catalyzed collagen invasion (Barbolina et al., 2007). In our study, EGR1 expression not only in stro- REFERENCES mal cells but also in cancer cells was related to Adamson E, de Belle I, Mittal S, Wang Y, Hayakawa J, Korkmaz prognosis in aEOC and in agreement with this, K, O’Hagan D, McClelland M, Mercola D. 2003. Egr1 signaling in prostate cancer. Cancer Biol Ther 2:617–622. EGR1 protein was found in the nuclei of cancer Barbolina MV, Adley BP, Ariztia EV, Liu Y, Stack MS. 2007. cells as well as aSMA positive fibroblasts in can- Microenvironmental regulation of membrane type 1 matrix metalloproteinase activity in ovarian carcinoma cells via colla- cer stroma. These EGR1 protein expressing fibro- gen-induced expression. J Biol Chem 282:4924–4931. blasts with aSMA expression were located around Bhowmick NA, Moses HL. 2005. Tumor-stroma interactions. Curr Opin Genet Dev 15:97–101. the border between cancer cells and cancer Dakhova O, Ozen M, Creighton CJ, Li R, Ayala G, Rowley D, stroma. Fibroblasts are associated with cancer Ittmann M. 2009. Global gene expression analysis of reactive stroma in prostate cancer. Clin Cancer Res 15:3979–3989. cells at all stages of cancer progression and under Fahmy RG, Dass CR, Sun LQ, Chesterman CN, Khachigian LM. these circumstances called cancer associated 2003. Transcription factor egr-1 supports fgf-dependent angio- genesis during neovascularization and tumor growth. Nat Med fibroblasts (CAF) (Kalluri and Zeisberg, 2006). 9:1026–1032. Hence, these EGR1 protein expressing fibroblasts Ferraro B, Bepler G, Sharma S, Cantor A, Haura EB. 2005. Egr1 a predicts pten and survival in patients with non-small-cell lung with SMA expression might be CAF. In the cancer. J Clin Oncol 23:1921–1926. light of these findings, we speculate as follows: Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H, Chen H, Omeroglu G, Meterissian S, Omeroglu A, Hallett (1) EGR1 produced by cancer cells acts in an M, Park M. 2008. Stromal gene expression predicts clinical out- autocrine manner as well as stimulating stromal come in breast cancer. Nat Med 14:518–527. Gregg JL, Brown KE, Mintz EM, Piontkivska H, Fraizer GC. fibroblasts (CAF) to produce EGR1, via its tran- 2010. Analysis of gene expression in prostate cancer epithelial scriptional activity. This in turn may further stim- and interstitial stromal cells using laser capture microdissection. BMC cancer 10:165. ulate cancer cells in a paracrine manner, via its Hirasawa A, Saito-Ohara F, Inoue J, Aoki D, Susumu N, transcriptional activity. (2) EGR1 transcription Yokoyama T, Nozawa S, Inazawa J, Imoto I. 2003. Association of 17q21–q24 gain in ovarian clear cell adenocarcinomas with factor is reported to be induced by multiple poor prognosis and identification of ppm1d and appbp2 as likely extracellular agonists (such as growth factors and amplification targets. Clin Cancer Res 9:1995–2004. Kalluri R, Zeisberg M. 2006. Fibroblasts in cancer. Nat Rev Can- cytokines) or environmental stress (such as cer 6:392–401. hypoxia, fluid shear stress, and vascular injury) Kato K, Takao T, Kuboyama A, Tanaka Y, Ohgami T, Yamaguchi S, Adachi S, Yoneda T, Ueoka Y, Hayashi S, Asanoma K, Wake (Khachigian et al., 1996; Nishi et al., 2002; N. 2010. Endometrial cancer side-population cells show promi- Maegawa et al., 2009). EGR1 expression in both nent migration and have a potential to differentiate into the mesenchymal cell lineage. Am J Pathol 176:381–392. cancer stroma (CAF) and cancer cells may be Khachigian LM, Lindner V, Williams AJ, Collins T. 1996. Egr-1- induced by extracellular cues from the cancer mi- induced endothelial gene expression: A common theme in vas- cular injury. Science 271:1427–1431. lieu, such as hypoxia, which are common to both Kim JB, Stein R, O’Hare MJ. 2005. Tumour-stromal interactions cancer and stromal cells. (3)The fibroblasts in breast cancer: The role of stroma in tumourigenesis. Tumour Biol 26:173–185. around cancer cells may originate from cancer Kobayashi D, Yamada M, Kamagata C, Kaneko R, Tsuji N, Naka- cells by epithelial mesenchymal transition (EMT) mura M, Yagihashi A, Watanabe N. 2002. Overexpression of early growth response-1 as a metastasis-regulatory factor in gas- of cancer cells. Recently, Kato isolated and char- tric cancer. Anticancer Res 22:3963–3970. acterized side-population (SP) cells in human en- Maegawa M, Arao T, Yokote H, Matsumoto K, Kudo K, Tanaka K, Kaneda H, Fujita Y, Ito F, Nishio K. 2009. Egfr mutation dometrial cancer (Kato et al., 2010). They up-regulates egr1 expression through the erk pathway. Anti- demonstrated that SP cells formed large, invasive cancer Res 29:1111–1117. Milde-Langosch K, Roder H, Andritzky B, Aslan B, Hemminger tumors, which were composed of both tumor cells G, Brinkmann A, Bamberger CM, Loning T, Bamberger AM. and stromal-like cells and that the stromal-like 2004. The role of the ap-1 transcription factors c-fos, , fra-1

Genes, Chromosomes & Cancer DOI 10.1002/gcc 312 KATAOKA ET AL.

and fra-2 in the invasion process of mammary carcinomas. Tominaga E, Tsuda H, Arao T, Nishimura S, Takano M, Kataoka Breast Cancer Res Treat 86:139–152. F, Nomura H, Hirasawa A, Aoki D, Nishio K. 2010. Amplifica- Naora H, Montell DJ. 2005. Ovarian cancer metastasis: Integrating tion of gnas may be an independent, qualitative, and reproduci- insights from disparate model organisms. Nat Rev Cancer 5:355–366. ble biomarker to predict progression-free survival in epithelial Nishi H, Nishi KH, Johnson AC. 2002. Early growth response-1 ovarian cancer. Gynecol Oncol 118:160–166. gene mediates up-regulation of epidermal growth factor recep- Tsuda H, Birrer MJ, Ito YM, Ohashi Y, Lin M, Lee C, Wong tor expression during hypoxia. Cancer Res 2002;62:827–834. WH, Rao PH, Lau CC, Berkowitz RS, Wong KK, Mok SC. Saegusa M, Hashimura M, Kuwata T, Hamano M, Watanabe J, 2004. Identification of DNA copy number changes in microdis- Kawaguchi M, Okayasu I. 2008. Transcription factor egr1 acts sected serous ovarian cancer tissue using a cdna microarray plat- as an upstream regulator of beta-catenin signalling through up- form. Cancer Genet Cytogenet 155:97–107. regulation of and p300 expression during trans-differentia- Tsuda H, Ito YM, Ohashi Y, Wong KK, Hashiguchi Y, Welch tion of endometrial carcinoma cells. J Pathol 216:521–532. WR, Berkowitz RS, Birrer MJ, Mok SC. 2005. Identification of Shahzad MM, Arevalo JM, Armaiz-Pena GN, Lu C, Stone RL, overexpression and amplification of abcf2 in clear cell ovarian Moreno-Smith M, Nishimura M, Lee JW, Jennings NB, Botts- adenocarcinomas by cdna microarray analyses. Clin Cancer Res ford-Miller J, Vivas-Mejia P, Lutgendorf SK, Lopez-Berestein 11:6880–6888. G, Bar-Eli M, Cole SW, Sood AK. 2010. Stress effects on fosb- Virolle T, Krones-Herzig A, Baron V, De Gregorio G, Adamson and interleukin-8 (il8)-driven ovarian cancer growth and metas- ED, Mercola D. 2003. Egr1 promotes growth and survival of tasis. J Biol Chem 285:35462–35470. prostate cancer cells. Identification of novel egr1 target genes. J Sperandio S, Fortin J, Sasik R, Robitaille L, Corbeil J, de Belle I. Biol Chem 278:11802–11810. 2009. The transcription factor egr1 regulates the hif-1alpha Wu MY, Zhuang CX, Yang HX, Liang YR. 2004. Expression of gene during hypoxia. Mol Carcinog 48:38–44. egr-1, c-fos and cyclin d1 in esophageal cancer and its precur- Tlsty TD, Coussens LM. 2006. Tumor stroma and regulation of sors: An immunohistochemical and in situ hybridization study. cancer development. Annu Rev Pathol 1:119–150. World J Gastroenterol 10:476–480.

Genes, Chromosomes & Cancer DOI 10.1002/gcc