Background
Ovarian cancer (OC) is the most lethal gynecologic malignancy in industrialized countries. The overall 5-year survival rate of ovarian cancer patients is 30% to 50%, largely due to the fact that the majority of these patients are diagnosed at an advanced stage (III or IV) of disease at initial diagnosis [
1]. Substantial advances in the treatment of primary OC have occurred, but patient morbidity and mortality remain high due to metastatic dissemination [
2]. Ovarian tumor cells primarily disseminate by shedding into the peritoneal cavity where they can implant on to the mesothelium that covers the omentum and bowel surface [
3]. In order for the tumor cells to establish secondary foci and invade the underlying stroma, they need to adhere to and interact with the peritoneal mesothelial cells. This is a crucial step in OC progression and is a possible target for chemotherapeutic intervention, yet few studies have focused on this interaction. Identifying crucial factors involved in the crosstalk between the tumor cells and the mesothelial microenvironment will not only improve our understanding of the disease but will ultimately enable us to provide better patient care.
Details of the mechanisms involved in OC cell adherence to mesothelium are unclear, but the dynamics of this interaction appear to be relatively rapid, in the order of minutes [
4,
5]. Potential molecular interactions include tumor cell binding to extracellular matrix proteins such as collagens type I and IV, laminin, and fibronectin via integrins, and binding to hyaluronan expressed on the surface of human peritoneal mesothelial cells via CD44 [
5‐
7]. The secretion of proteolytic enzymes, cytokines, and growth factors by both tumor cells and mesothelium will contribute to the regulation of adherence, survival and further dissemination. Recent studies have demonstrated that c-Met overexpression is a prognostic factor in ovarian cancer and that targeting c-Met in vivo inhibits peritoneal dissemination and invasion through an α5 β1 integrin-dependent mechanism [
8].
In order to facilitate the study of ovarian tumor cell-mesothelial cell interactions, in this study we established a novel human ovarian cancer xenograft model, and a reproducible OC adherence assay for the investigation of the interaction and behavior of a highly metastatic OC cell line (MFOC3). Differential gene expression analysis was employed to identify genes with potential mechanistic influence, and one of these, interleukin-1beta (IL-1β) was found to be pivotal for increased adherence of the ovarian cell line to human mesothelial cells. Immunohistochemical evaluation of ovarian tumor tissues revealed a significant correlation between with IL-1β expression and overall survival. The derivation of this ovarian tumor xenograft model provides a powerful experimental system for the controlled evaluation of molecular mechanisms involved in ovarian carcinoma dissemination and metastasis.
Methods
Ovarian cancer cell culture
The human ovarian cancer cell line, FOC3 was kindly provided by Dr. T. Fukuda (Fukushima Medical University, Fukushima, Japan) [
9]. Ovarian cancer (OC) cell lines were maintained in Dulbecco's modified Eagle's medium (DMEM) in the presence of 10% fetal bovine serum (FBS) and 50 U penicillin-streptomycin at 37°C in a humidified atmosphere of 5% CO2.
Establishment of an intraperitoneal human ovarian cancer xenograft model
Exponentially growing cells were harvested with trypsin-EDTA, washed and resuspended in FBS-free DMEM. For subcutaneous (s.q.) inoculation, 1 × 107 FOC3 cells in 0.2 ml of DMEM/50% Matrigel (BD Bioscience, San Jose, CA) were injected s.q. into the left inguinal region of 6 week-old female severe combined immunodeficiency (SCID) female mice (Nihon Clea, Tokyo, Japan). FOC3 and MFOC3 cells were injected into 10 and 16 animals respectively. When the primary xenograft tumor reached a diameter of 1 cm, the tumor was removed, and minced with scissors aseptically, and seeded into a petri dish. After several passages, the surviving culture was designated MFOC3. A total of 1 × 107 FOC3 and MFOC3 cells in 0.5 ml PBS were injected intraperitoneally (i.p.) into SCID female mice. All mice were sacrificed 13 weeks after the injection and visible tumor, mesenterium, peritoneum, omentum and diaphragm were removed, fixed in 20% buffered formalin and processed for hematoxylin and eosin (H&E) staining. Frequency of peritoneal metastasis between FOC3 and MFOC3 were compared by the χ2 test (StatView, SAS Institute, Cary, NC, USA). p < 0.05 was considered to be statistically significant. This experiment was approved by the Institutional Animal Care and Use Committee of Fukushima Medical University (Ref# KO3129).
Primary culture of Mesothelial cells
Human mesothelial cells (MCs) were isolated from the peritoneum of patients who were undergoing surgery for benign conditions at Fukushima Medical University School of Medicine and Tsuboi hospital. Written informed consent was obtained from all patients as per IRB approval from the Fukushima Medical University School of Medicine, and all documentation is retained within patient files. The procedure was based on techniques described in previous articles7,9-15. Briefly, MCs were collected by scrubbing the peritoneum with a cold knife. The MCs were cultured using DMEM contained 10% FBS and 50 U penicillin-streptomycin at 37°C in a humidified atmosphere of 5% CO2. The MC's were used in experiments at passage 2-5.
Microarray analysis
Difference mRNA transcript expression between FOC3 and MFOC3 was investigated using the Human Cancer CHIP microarray, version 2.1 (Takara, Shiga, Japan) containing probes to 557 human genes reportedly associated with cancer. Total RNA was isolated from culture cells using TRIZOL (GIBCO, Grand Island, NY), and mRNA was further purified using an oligotexTM-dT30 'Super-mRNA Purification kit' (Takara). cDNA labeled with Cy3-dUTP or Cy5-dUTP was generated from 1 ug of FOC3 and MFOC3 mRNA using Takara RNA Fluorescence Labeling Core Kit (Takara), and 50 pg of lambda A was labeled as an internal control. Cy3 and Cy5-labeled probe were co-hybridized to the microarrays in hybridization solution (6 × SSC, 0.2% SDS, 5 × Denhardt's solution, carrier DNA) at 65°C over night, and the microarray slide processed as previously described (Nagata, et al, 2003). The Cy3 and Cy5 fluorescence intensities were captured with a 428 array scanner (Affymetrix, Santa Clara, CA) with separate measurement of the sample intensities for Cy3 and Cy5. The intensity of each hybridization signal was evaluated photometrically using Imagene software (BioDiscovery, Marina Del Rey, CA) and normalized to the average signal of a housekeeping gene, GAPDH. The Cy3/Cy5 ratio for each sample was calculated by averaging the spots, and the cutoff value for each expression level was calculated according to the background noise. For this cutoff, we used an expression level of above 100 where the fluctuation was less than a critical value (1.0) because other genes (those with low expression) are embedded in the background noise. Upregulated and downregulated genes in the MFOC3 cells were defined as those having a Cy3/Cy5 signal ratio of > 3.0 and < 0.50, respectively.
RT-PCR
We confirmed the differential expression of the IL-1β gene using RT-PCR. 2 ug of total RNA was reverse-transcribed using a first strand cDNA synthesis kit (Invitrogen, Carlsbad, CA). 1 μl of the first strand cDNA preparation was added to the PCR reaction. The primers for IL-1β were as described by McAlindon et. al. [
10]. Oligonucleotide IL-1β primers; 5'ATGGCAGAAGTACCTAAGCTCGC-3' and 5'-TTGACTGAAGTGGTACGTTAAACACA-3' (PCR product 802 bp). GAPDH primers; 5'-GCACAGTCAAGGCAGAGAGAAC-3' and 5'-CTTATGACCACTGTCCACGC-3' (PCR product 420 bp). PCR was performed using TAKARA™ Ex Taq (Takara). PCR for IL-1β was performed with 30 cycle of denaturation for 30 seconds at 94°C, annealing for 30 seconds at 62°C, and elongation for 1 minute at 72°C. PCR for GAPDH was performed with 20 cycle of denaturation for 30 seconds at 94°C, annealing for 30 seconds at 56°C, and elongation for 1 min at 72°C. PCR products were electropohoresed on a 1.5% agarose gel and visualized using ethidium bromide staining.
Western blot analysis
In order to confirm the differential expression of IL-1 β in the cell lines, we performed Western blot analysis. Freshly cultured OC cells were lysed in a CellLytic™-MT reagent (Sigma, St. Louis, MO), and protein concentration of lysates were quantified by the Bradford method using Bio-Rad protein assay (Bio-Rad, Hercules, CA). Cell lysates (20 ug) were electrophoresed on BIO-RAD ready gels 4-20% Tris-HCL (Bio-Rad) and electrically transferred to a PDVF membrane (Bio-Rad). The membrane was blocked with 1xTBS, 5% nonfat dry milk, 0.05% Tween20 and incubated overnight at 4°C with 2 ug/ml mouse anti-human IL-1β antibody (R&D Systems, Abinghdon, UK), and 1 hr at room temperature with 1:2000 dilution of anti-mouse IgG conjugated HRP (Amersham Pharmacia Biotech, Tokyo, Japan). As a loading control, horseradish peroxidase-conjugated polyclonal antibody to β-actin (1:2000) (Santa Cruz, Santa Cruz, CA) was applied as described for primary antibodies above. Immunoreactive bands were visualized using ECL Western blot detection system (Amersham Pharmacia Biotech) following manufacturer's protocol.
GFP transfection of OC cell lines
Transfection was performed using SuperFect® Transfection Reagent (Qiagen, Tokyo, Japan) according to the manufacture's protocol, with minor modification. Briefly, 2 × 105 of OC cells were placed in a 35 mm dish the day prior to transfection. 2 ug of the pEGFP-N1 plasmid (Clontech, Palo Alto, CA) was mixed with 12 μl of SuperFect® reagent per dish in serum-free DMEM (final volume of 60 μl), and DNA-lipid complexes were formed by incubation for 10 min at room temperature. DNA-lipid complex was incubated with FOC3 and MFOC3 cells for 3 hours, cells were washed with PBS, and 2 ml DMEM with 10% FBS was added. Three days later, the cells were exposed to 500 ug/ml G418. The limiting dilution method was used with medium containing G418 (500 ug/ml; Promega, Madison, WI) for generation of monoclonal cell lines. Single, GFP-positive cells in 96-well plates were detected using fluorescence microscopy (Olympus, Tokyo, Japan). To confirm the tumorigenicity of GFP labeled cells, these derivative lines were inoculated into SCID mice (as described above). In the absence of any co-inoculated cytokines, GFP labeled FOC3 (0/10 animals) and MFOC3 (8/10 animals) had very similar propensity to the unlabeled cell lines with respect to peritoneal tumorigenesis.
In vitro adhesion assay
In order to compare the difference between FOC3 and MFOC3 adhesion to mesothelial cells (MCs), 1 × 105 GFP labeled FOC3 and MFOC3 cells in serum free DMEM were placed over mesothelial monolayers in an uncoated 24-well plate and incubated at 37°C for various periods. Students's t-test was performed as a test of significance (StatView, SAS Institute, Cary, NC). Data was presented as mean ± SD. p < 0.05 was considered to be statistically significant.
The effect of recombinant IL-1β on OC adhesion to mesothelium was achieved by adding different concentrations of recombinant IL-1β (R&D) in serum free DMEM incubated at 37°C for 30 minutes. To observe inhibition of IL-1β and integrin β1 on OC cell adhesion, 1 × 105 FOC3-GFP, MFOC3-GFP and FOC3-GFP containing 20 ng/ml recombinant IL-1β and neutralizing antibodies of IL-1β and β1 integrin (Chemicon, Temecula, CA) in serum free DMEM were placed over mesothelial monolayers in an uncoated 24-well plate and incubated at 37°C for 60 minutes. Normal mouse IgG (Amersham Pharmacia Biotech) was used as a negative control. Both antibodies were used at 10 ug/ml, since preliminary experiments demonstrated that this concentration produced maximum inhibition of OC cell binding to mesothelium. For all assessments, nonadherent cells were removed by washing twice with PBS after incubation. The adherent cells were fixed by 4% paraformaldehyde and counted at 100x under a fluorescence microscope. These data were analyzed by one-way analysis of variance with the Bonferroni/Dunn test (StatView). Data was presented as mean ± SD. P < 0.05 was considered to be statistically significant.
In Vivo Assessment of Ovarian Cancer Peritoneal Metastasis
To determine the effect of IL-1β and integrin β1 antibody on ovarian cancer peritoneal metastasis in vivo, 1 × 107 MFOC3 cells with anti-IL-1β (10 ug/ml) and anti- β1 integrin (10 ug/ml) in 0.5 ml PBS were injected i.p. into SCID female mouse. Normal mouse IgG (10 ug/ml) was used as a negative control. All mice were sacrificed after 13 weeks, and tumors visible on the mesenterium, peritoneum and diaphragm were monitored under low-power magnification. Frequency of peritoneal metastasis between MFOC3 and MFOC3 with anti IL-1β or anti β1 integrin were compared by the χ2 test (StatView). p < 0.05 was considered to be statistically significant.
Flow cytometry
Analysis of adhesion molecules in MCs was detected using Flow cytometry. At confluency, MCs were washed twice with fresh culture media and then incubated with 20 ng/ml recombinant IL-1β in serum free RPMI 1640 for 30 minutes. MCs were harvested using trypsin (0.05%) and EDTA (0.02%) and resuspended in PBS with 2% FBS. 100,000 cells treated with primary monoclonal antibodies for integrin α1 (Chemicon), integrin α2 (SouthernBiotec, Birmingham, AL), integrin α5 (Cymbus Biotechnology, Hampshire), integrin α6 (Chemicon), integrin αV (Chemicon) and CD44 (SouthernBiotec) at a dilution of 1:15, and integrin β1 (Chemicon) at a dilution 1:100, at 4°C for 30 minutes. Cells were washed twice in PBS with 2% FBS, and labeled with FITC-conjugated rabbit-anti-mouse Ig (Dako, Tokyo, Japan) at 4°C for 30 minutes. After two additional washes, the cells were analyzed using a flow cytometry (BD Bioscience, San Jose, CA). Statistical analysis was performed using the Students's t-test for paired data (StatView). Fluorescence intensities are presented as median ± SD absolute deviation in relation to control, which were set as 100%. p < 0.05 was considered to be statistically significant.
Patient Population
Patients with a diagnosis of early stage (stages I and II) ovarian carcinoma were identified at Fukushima Medical University School of Medicine and Tsuboi hospital during a 10 year period from 1995 to 2004. The clinical stage was determined using FIGO criteria. A total of 96 patients had primary surgical cytoreductive surgery with no exposure to prior chemotherapy. The median age of the patients was 50.7 years (range 12-86). All patients gave informed consent before their inclusion in this study. Association between IL-1β expression and age, histology and ascites were assessed using χ2 test (StatView). p < 0.05 was considered to be statistically significant.
Immunohistochemistry
Paraffin-embedded tissue sections were deparaffinized, and endogenous peroxidases were inactivated by a 15-min treatment with methanol containing 3% hydrogen peroxide. Sections were subjected to heat-induced antigen retrieval for 15 min at 100°C in 10 mM citrate buffer before addition of the primary antibody. After nonspecific binding was blocked with goat serum for 30 min, all slides were incubated overnight at 4°C for overnight with a 1:50 dilution of rabbit antihuman IL-1β antibody (Santa Cruz Biotechnology, Santa Cruz, CA). After incubation, an avidin-biotin-peroxidase reaction was performed using Histofine SAB-PO(R) kit (Nichirei, Tokyo, Japan) following manufacturer's protocol. The sections were subjected to diaminobenzidine for 10 min and counterstained with Mayer's hematoxylin.
IL-1β expression was estimated in a combined scoring system based on both the percentage of positive cells (negative = 0, < 10% = 1, 11-50% = 2, and 51-80% = 3, > 81% = 4) and the staining intensity (negative = 0, weak = 1, moderate = 2, and strong = 3). The final immunoreaction score (IRS) was obtained by multiplication of the % value of positive cells by the value for staining intensity. The IRS can range from 0 to 12, and the cases of 4 to 12 and 0 to 3 were defined as high and low expression of IL-1β respectively. Survival curves were calculated using the Kaplan-Meier method, and differences in survival were analyzed using the log-rank test (SPSS 12.0, SPSS Co., Tokyo, Japan). In all analyses, statistical significance was defined as a probability value less than 0.05.
Discussion
Despite improvements in surgery and treatment of ovarian cancer, the long-term survival of patients with this disease is limited due to the intraperitoneal (i.p.) spread of tumor cells from the primary tumor mass. A crucial step in the establishment of secondary foci is the adherence of disseminated tumor cells to the mesothelial surfaces of the peritoneum. Thus, a potentially efficacious adjuvant therapy approach could involve the i.p. administration of agents that target tumor cell and mesothelial cell interactions. To better understand the pivotal processes, we developed an in vitro cell adhesion model in order to identify key molecules and to test their effect on phenotype. Findings from the model were subsequently confirmed in human tissue specimens.
By 'conditioning' FOC3 cells through immune-compromised mice [
12,
13] we derived the MFOC3 sub-line by recovery from an advanced FOC3 xenograft tumor formed in SCID mice. Unlike FOC3 cells, inoculation of the MFOC3 subline resulted in multiple tumors across the peritoneal cavity in multiple experiments. These isogeneic cell line populations constitute a valuable model for the elucidation of mechanisms involved in peritoneal metastasis of ovarian cancer cells. It is to be noted that our molecular profiling was restricted to 557 genes and so cannot be considered comprehensive, but of particular interest from our differential gene expression analyses was the confirmed increased expression of interleukin-1beta (IL-1β) in metastatic MFOC3 cells. IL-1β has previously been implicated in peritoneal mesothelium integrity [
14] and in cancer cell dissemination [
11,
15]. Development of an
in vitro cell adhesion assay using human mesothelial cells (MCs) enabled the testing of a role for IL-1β in OC/MC cell adhesion. Previous studies, have performed similar
in vitro adhesion assays using isolated human mesothelial cells [
16,
17], but here we monitored intact adherent cells accurately by tracking GFP-labeled FOC3 and MFOC3 cells. In these assays, MFOC3 cells were observed to adhere up to 3 times more avidly than the parental FOC3 cell population, and a dose-dependent increase in FOC3 adhesion after preincubation of MCs with IL-1β was observed. The specificity of this effect was confirmed by neutralization with an IL-1β antibody and reduction of MFOC3 adhesion to MCs in a dose-dependent manner. These findings indicate that IL-1β plays a pivotal role in the adhesion of FOC3 cells to MCs. A role for IL-1β in cell adhesion has been reported in different model systems. For example, IL-1β promoted the adherence of human colorectal carcinoma cell lines and melanoma cell lines to endothelial cells in a dose-dependent manner [
18,
19].
To monitor the effects of IL-1β treatment on MCs we measured the expression of a panel of adhesion molecules. No differences were observed in the MC expression of CD44, or 5 of 6 integrins monitored by flow cytometry, but an increase in the expression of MC β1 integrin was evident. Inclusion of a neutralizing anti- β1 integrin antibody in our in vitro cell adhesion assay significantly inhibited MFOC3 cell adhesion to MCs in a dose-dependent fashion. In combination, these analyses show that the increased adhesion of MFOC3 to MCs is influenced by the release of IL-1β released from the ovarian tumor cells, and one consequence of this is the induction of cell surface β1 integrin on MCs. The interference of tumor cell adhesion with neutralizing antibodies was also tested in vivo. MFOC3 metastasis rates of ~80% were not significantly changed in mice treated with anti- IL-1β antibodies, explained by high rates of antibody clearance or stability in vivo, or sequestration of antibodies by murine cytokines or other factors, however, peritoneal metastasis was observed in only 20% of mice when treated with anti- β1 integrin antibodies. Coupled with the presence of ligands laminin and fibronectin on FOC3 and MFOC3 cells, this confirms a role for the β1 integrin subunit in MFOC3 metastasis to the peritoneal cavity.
In order to link the model findings to events in human disease, we investigated the expression of IL-1β in archival tumor tissues obtained from a cohort of 96 early-stage ovarian cancer cases. Immunohistochemical analyses revealed no significant association between histological type and IL-1β expression level, but the IL-1β expression rate was greater in clear and mucinous tumors than that observed in serous and endometrioid tumors, and patients with ascites were more likely to be positive for IL1β. Most importantly, overall survival analysis revealed that 29% of patients whose tumors expressed IL-1β died from the disease within a 5-year follow-up period. In contrast, only 1.6% of patients whose tumors did not express IL-1β died within this same period. These results suggest IL-1β might be associated with tumors of a more aggressive phenotype, and interference with its production or action may be a promising target for ovarian cancer therapy.
A number of studies have focused on identifying targets that may prove clinically useful in the management of ovarian cancer patients. Targets under development include TRAIL, BCL-2 family, EGFR and VEGF, and clinical trials evaluating agents that inhibit VEGF and EGFR signaling pathways in ovarian cancer have been performed [
20,
21]. Interestingly, there is a link between IL1b and VEGF. IL-1β triggers the release of pre-existing, sequestered VEGF from the extracellular matrix, that then homes to the endothelial compartment where it induces endothelial cell proliferation [
11]. So IL-1β is upstream of VEGF in the angiogenic pathway and therefore plays a key role in the expansion of the vascular compartment during tumor progression. VEGF actions in enhancing ovarian tumor metastasis are also thought to be mediated through crosstalk with integrin cell adhesion receptors [
22]. Previous reports have shown a role for β1 integrin in the adhesion of OCs to mesothelia, but the hypothesis tested has invariably been that it is the integrins on the OC cell surface that mediate the interaction with mesothelial cells [
5,
15]. Our study shows that IL-1β released from the tumor cells induces β1 integrins on the MCs and that axis mediates cell adhesion between the two cell types. Thus, the IL-1β/integrin axis may not only advance initial cell adhesion of ovarian tumor cells to the peritoneum, but trigger a cascade of events that favor the establishment of dispersed secondary tumors. Perturbation of this axis at any stage of disease may have significant benefit.
The peritoneal mesothelium is a highly specialized layer of epithelial cells that covers the entire surface of the abdominal cavity. It serves as a protective anatomical barrier, and as a low friction, non-adhesive interface allowing the movement of abdominal organs. Cytokines present in inflammatory or malignant peritoneal effusions may significantly affect the function and the structure of the peritoneal mesothelial cell lining [
23]. The cytokine milieu can induce an 'activated' mesothelium which, in turn, results in the secretion of a variety of factors, including IL-6, IL-8, VEGF [
24,
25], and IL-1β [
26,
27]. In the case where the tumor cell secretes its own IL-1β, then the phenomenon can occur without injury, or in an enhanced way at sites already undergoing remodeling.
It is likely that IL-1B plays multiple roles in progression and/or metastasis in ovarian cancer. It has been demonstrated that IL-1β can cause major morphological changes and loss of mesothelium integrity, characterized by cellular retraction and wide-spread exposure of the submesothelial ECM [
14]. The disruption of the mesothelial barrier may be important for the progression of neoplastic disease [
28]. Once ovarian carcinoma cells present in the peritoneal fluid adhere to mesothelial cells, the cancer cells may migrate through this layer, invade the local organs, and spread to distant sites. This idea is supported by an
in vivo model study in which damage to the mesothelium caused by a pneumoperitoneum resulted in peritoneal tumor spread and a significant decrease in animal survival [
29]. Other groups have studied the effect of peritoneal wounding on tumor cell adhesion. The adhesion of CC531, Caco2 and HT29 colon carcinoma cell lines to an autologous monolayer of rat mesothelial cells [
30,
31] pre-incubated with factors released after surgical trauma (IL-1β and EGF) resulted in at least 60% more cell adhesion in a dose-dependent manner. Additional blocking experiments with anti-IL-1β resulted in significant inhibition of the tumor cell adhesion in the model, but they did not test the effect
in vivo. Both IL-1β and TNF has been shown to regulate ovarian cancer in nude mouse xenograft models [
32]. Intraperitoneal recombinant IL-1β administration led to a dose dependent replacement of peritoneal ascitic tumour with solid tumours attached to the peritoneum, and high doses had an anti-tumor effect. Recombinant human TNF also promoted tumor implantation in the same xenograft models, but had no anti-tumor effect. There has been literature reporting an association between inherited polymorphisms in the IL-1β gene and various cancers, including cervical, bladder and breast, but the results remain ambiguous [
33,
34].
Thus, anti- IL-1β therapy may act on several levels; reduce tumor cell adhesion, maintain peritoneal integrity, and ameliorate peritoneal tumor cell induced inflammatory responses that may exacerbate adhesion and establishment of secondary tumors. Furthermore, because the frequency of IL-1β expression in human ovarian cancer specimens was strongly associated with overall survival, it is possible that IL-1β may also serve as a biomarker for prognosis or as a monitoring biomarker for therapeutic response. Additional tumor samples need to be evaluated to investigate the value of IL-1β as a biomarker with clinical utility.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
TW participated in all aspects of the study, from design to clinical and laboratory performance, and manuscript writing. TH established MFOC3 and participated in the cDNA microarray analyses. TS aided in conception, design, and experimentation, particularly immunohistochemistry. SS performed cell culture and adhesion assays. HM performed cell culture and immunoassays. SG participated in design, data analysis and drafting of the manuscript. YM, HY and KF participated in design and analysis of clinical data. All authors have read and approved the manuscript.