Expression and regulation of tumor necrosis factor α in normal and malignant ovarian epithelium

Several inflammatory conditions, such as pelvic inflammatory disease and endometriosis, are associated with an increased risk of ovarian cancer (1). The links between inflammation and cancer are increasingly recognized; various mechanisms involving an array of mediators, including cytokines, chemokines, growth factors, and free radicals, have been described (2, 3).

In situ hybridization studies correlated expression of tumor necrosis factor α (TNF-α) by the tumor epithelium and infiltrating macrophages, with increasing tumor grade in serous ovarian carcinoma (4). Others have shown that TNF-α mRNA is expressed by ovarian carcinoma cells and may be induced in normal ovarian surface epithelial cells, promoting growth in both cell types (5, 6). Previous studies have also indicated that ovarian surface epithelial cells can secrete cytokines such as interleukin (IL)-1, IL-6, and M-CSF, with a tendency for greater production in vitro compared with malignant cells (7). However, this contrasts with the frequently elevated levels of these cytokines in the serum and ascites of patients with ovarian cancer (8–12), suggesting that inflammatory cells and/or chronic activation of the tumor cells may be clinically of greater significance.

Here, we explored the hypothesis that TNF-α may be an important regulator of the proinflammatory milieu in ovarian cancer. We assessed the autocrine regulation of TNF-α in malignant (both ovarian carcinoma cell lines and primary ovarian tumor cells) and normal ovarian surface epithelial cells and its ability to induce a variety of cytokines that are present in ovarian epithelium. Our studies suggest that TNF-α within ovarian tumor epithelium is subject to complex regulation and that blocking its activity may, at least in part, suppress the inflammatory cytokines associated with this disease.

Ovarian carcinoma cell lines [IGROV-1 indirectly from Bernard et al. (13); OVCAR-3 from Hamilton et al. (14); PEO1 from Langdon et al. (15); and SKOV-3 from the American Tissue Culture Collection (Rockville, MD)] were grown in a humidified atmosphere at 37°C and 5% CO₂ in either endotoxin-free RPMI medium (IGROV-1, OVCAR-3, and PEO1 cell lines) or endotoxin-free DMEM medium (SKOV-3 cell line) supplemented with 10% fetal bovine serum (Sigma-Aldrich Ltd., Poole, United Kingdom). Insulin (2.5 μg/mL) was added to the medium of OVCAR-3 and PEO1 cell lines. TNF-α (20 ng/mL) was added to 70% confluent cells and analysis was done at 0, 0.5, 1, 3, 6, and 24 hours after addition of the cytokine. The TNF-α concentration was selected from dose-response studies for TNF-α mRNA induction (data not shown). Unless indicated, TNF-α refers specifically to human TNF-α whereas mTNF-α is used to denote the murine form (also used at 20 ng/mL). Actinomycin D experiments were done using a dose of 5 μg/mL. Infliximab (10 μg/mL) and the immunoglobulin G1 isotype control (10 μg/mL) were a kind gift from Centocor (Malvern, PA).

IGROV-1 cells were transfected with SUPER RNAi plasmids for TNF-α and isolated according to the protocols described (16). Four micrograms of plasmid DNA in 250 μL serum-free Opti-MEM medium (Life Technologies, Inc., Gaithersburg, MD) were mixed with 10 μL Lipofectamine 2000 (Invitrogen, Paisley, United Kingdom) in 250 μL Opti-MEM and incubated for 20 minutes at room temperature. Serum-free medium (1.5 mL) was mixed with the Lipofectamine 2000/DNA solution and added to the IGROV-1 cells, which were then incubated for 4 hours at 37°C before the medium was replaced with fresh medium supplemented with 10% FCS. Antibiotic selection for stable cell lines started after 48 to 72 hours; cells were selected for SUPER RNAi plasmid expressing cells in 4 μg/mL puromycin (Sigma) for 30 days.

The study was approved by the East London and City Health Authority research ethics committee and informed consent was obtained from patients attending the gynecologic oncology unit at St. Bartholomew's Hospital (London, United Kingdom). Normal ovarian surface epithelial cells were obtained by scraping the ovarian surface at the time of laparotomy and cultured in a humidified atmosphere at 37°C and 5% CO₂ in endotoxin-free MCDB105/M199 medium (Sigma) supplemented with 15% fetal bovine serum (17). Primary ovarian cancer cells from ascites were obtained from patients undergoing surgery for ovarian cancer and cultured in RPMI medium (as above). Malignant and normal ovarian surface epithelial cells were identified on the basis of microscopic appearance, positive staining for cytokeratin and negative staining for factor VIII. TNF-α (or mTNF-α) was added to 70% confluent cells and analysis was done at intervals of 0, 1, 3, 6, and 24 hours after addition of the cytokine.

DNase-treated total RNA was reverse transcribed using the Ready-To-Go kit (Amersham Pharmacia Biotech, Piscataway, NJ). cDNA (1 μL) was amplified by PCR using primers for TNF-α, TNFRI, and TNFRII (Clontech, Cowley, Oxford, United Kingdom) with β-actin primer as a positive control.

DNase-treated RNA was reverse transcribed with Moloney murine leukemia virus reverse transcriptase (Promega, Southampton, United Kingdom) according to the instructions of the manufacturer. Multiplex real-time reverse transcription-PCR (RT-PCR) analyses were done using TNF-α, IL-1α, IL-6, CCL2, CXCL8, M-CSF (all FAM), and 18S rRNA (VIC) primers and probes with the ABI Prism 7700 Sequence Detection System instrument and software (PE Applied Biosystems, Warrington, United Kingdom). Two microliters of cDNA were used per 25 μL reaction with cycling conditions as follows: 50°C for 2 minutes, 95°C for 10 minutes, and then 40 cycles of 15 seconds at 95°C and 1 minute at 60°C. Each sample was analyzed in duplicate or triplicate and normalized (ΔC_t) to 18S by removing the cycle threshold (C_t) value of 18S from the C_t value of the gene under investigation. The ΔC_t for the control was subtracted from the ΔC_t for TNF-α stimulation (i.e., ΔΔC_t) and the fold difference was calculated as 2^−ΔΔC_t.

[α-³²P]dCTP-labeled TNF-α and β-actin cDNA probes were obtained from Dr. S.C. Robinson (Translational Oncology, QMUL, London, United Kingdom; TNF-α and β-actin), prepared and used for Northern blot analysis as previously described (18).

The Cancer profiling array II (Clontech) comprises cDNA from tumor and corresponding whole normal ovarian tissue from individual patients spotted on a nylon membrane. The array was hybridized with a specific radiolabeled TNF-α probe according to the instructions of the manufacturer. Hybridization signals were detected by phosphorimaging and analyzed using ImageQuant Version 1.11. The membrane was stripped and reprobed with a ubiquitin probe to normalize each signal.

R&D ELISA kits (TNF-α, TNFRI, TNFRII, IL-1α, IL-6, CCL2, and M-CSF) and Biosource ELISA kits (Biosource Europe, Nivelles, Belgium; CXCL8 and TNF-α) were used to measure the cytokine content of cell supernatants. ELISAs were carried out according to the instructions of the manufacturer.

Monoclonal antibodies against TNFRI and TNFRII (R&D Systems, Abingdon, United Kingdom) and isotype-matched control were used. Cells were resuspended in PBS supplemented with 1% heat-inactivated FCS and 0.01% NaN₃. Antibodies diluted in this buffer were used between 2 and 20 μg/mL and incubated for 30 minutes on ice. Cells were washed and 10,000 cells were analyzed by flow cytometry on a FACScan flow cytometer using CellQuest software (BD PharMingen, Oxford, United Kingdom).

NIH Image v.1.61 software was used to analyze Western blots by densitometry and ImageQuant software (Molecular dynamics) v1.11 was used to analyze phosphorimaging data (Storm scanner) obtained from the cancer profiling array. InStat v2.01 software was used to test results for statistical significance (Student's t test and Bonferroni test).

We screened for the expression of TNF-α in matched normal whole ovary and ovarian tumor tissue by hybridizing a radiolabeled TNF-α probe to a cDNA expression array. Levels of TNF-α mRNA were higher in malignant ovarian tumors compared with the corresponding normal tissue (Fig. 1). Increased TNF-α mRNA was present in a variety of tumor subtypes, including adenocarcinoma (samples 3, 5, 7, and 8), clear cell adenocarcinoma (sample 4), and serous papillary carcinoma (sample 9).

We determined the contribution of the ovarian epithelium to increased TNF-α mRNA expression by comparing the levels of TNF-α mRNA in unstimulated ovarian carcinoma cell lines, primary ascitic ovarian tumor cells, and normal ovarian surface epithelial cells. TNF-α mRNA expression was detected by conventional RT-PCR in four ovarian cancer cell lines but not in normal ovarian surface epithelial cells (data not shown). However, the more sensitive technique of quantitative real-time RT-PCR confirmed expression of very low levels of TNF-α mRNA in normal ovarian surface epithelial cells with up to 1,000-fold more TNF-α mRNA in the ovarian carcinoma cell lines (Fig. 2A). Intermediate levels of TNF-α mRNA expression were observed in unstimulated primary ascitic ovarian tumor cells [i.e., ovarian carcinoma cell lines (IGROV-1 > OVCAR-3 > PEO1 > SKOV-3) > primary ovarian tumor cells > normal ovarian surface epithelial cells (P < 0.001).

Conventional RT-PCR revealed expression of TNFRI, but not TNFRII, in the ovarian carcinoma cell lines and normal ovarian surface epithelial cells, indicating that TNF-α signals via the former receptor (Fig. 2B). Similarly, TNFRI, but not TNFRII, was expressed in the primary ascitic tumor cells; the only exception was AS1, which derived from a patient with a mixed mesodermal tumor and expressed both receptors subtypes. Fluorescence-activated cell sorting analysis of the ovarian cancer cell line IGROV-1 and normal ovarian surface epithelial cells confirmed the presence of TNFRI, but not TNFRII, at the cell surface (Fig. 2C). Similarly, only soluble TNFRI was detected in the cell supernatant of IGROV-1 ovarian cancer cells by ELISA (range, 47–161 pg/mL).

Based on the elevated levels of TNF-α mRNA in malignant epithelium and the presence of TNFRI in tumor and normal ovarian surface epithelial cells, we proceeded to compare the expression of TNF-α mRNA in both cell types following stimulation with 20 ng/mL of TNF-α (a dose selected on the basis of efficient induction of TNF-α mRNA at 1 and 24 hours in the IGROV-1 cell line; data not shown). Experiments were initially done using Northern blot analysis (Fig. 3A and B) followed by more extensive studies with quantitative real-time RT-PCR (Fig. 3C–E). Both malignant and normal ovarian surface epithelial cells displayed maximal up-regulation of TNF-α mRNA by 1 hour. TNF-α mRNA expression remained elevated over the 24-hour period of TNF-α stimulation (compared with control) and declined slowly in the majority of the cancer cell lines (Fig. 3C). The SKOV-3 cell line was an exception, revealing a rapid decline to baseline of the mRNA transcript, showing similar TNF-α mRNA kinetics to the normal ovarian surface epithelial cells between 6 and 24 hours (Fig. 3E). The limited number of experiments done in the primary tumor cells (Fig. 3D) revealed mRNA expression kinetics halfway between the cell lines and normal ovarian surface epithelial cells.

The differential TNF-α mRNA expression profile observed in the tumor cells may be due to enhanced mRNA stability. Actinomycin D chase experiments were done comparing the stability of TNF-α mRNA in IGROV-1 and OVCAR-3 cell lines with that in the SKOV-3 cell line. Cells were treated with 20 ng/mL of TNF-α for either 1 or 24 hours before the addition of actinomycin D. IGROV-1 and, to a lesser extent, OVCAR-3 displayed significantly increased TNF-α mRNA stability after 24 hours compared with 1 hour of TNF-α treatment (Fig. 4A and B). In contrast, there was no evidence for TNF-α mRNA stability in the SKOV-3 cell line (Fig. 4C), consistent with the abrupt decline in TNF-α mRNA expression observed between 6 and 24 hours. Thus, the regulation of TNF-α mRNA expression in ovarian tumor cells by TNF-α occurs both at the transcriptional and posttranscriptional levels.

TNF-α Protein Production Is Confined to Malignant Ovarian Epithelial Cells. We reasoned that TNF-α mRNA stability may correlate with the production of TNF-α protein. Previously such an assessment has proved difficult due to the excess TNF-α used for stimulation. It is known, however, that mTNF-α binds human TNFRI with identical affinity to human TNF-α but does not cross-react in a specific TNF-α ELISA kit. Hence, we stimulated human cells with mTNF-α and measured human TNF-α production. Both IGROV-1 and OVCAR-3 ovarian cancer cells released human TNF-α into the cell supernatant in response to mTNF-α, with maximal levels detected by 24 hours (Fig. 5A and B). PEO1 cells also made TNF-α constitutively (Fig. 5C) whereas no TNF-α was detected in SKOV-3 cell supernatant (n = 6; data not shown; Fig. 5D). One of three primary ascitic tumor isolates (AS4) revealed constitutive TNF-α production. Induction of human TNF-α protein was only observed in one of three primary ascitic tumor isolates (AS1) following mTNF-α stimulation. In contrast, TNF-α was not detected when cultures of normal ovarian surface epithelial cells were treated with mTNF-α (n = 6; data not shown).

Infliximab Down-Regulates TNF-α mRNA in Ovarian Cancer Cells. The TNF-α neutralizing antibody infliximab has been shown to block TNF-α activity in several inflammatory conditions, advancing the clinical management of patients with rheumatoid arthritis and Crohn's disease. Several mechanisms of action have been postulated in addition to neutralization of soluble TNF-α, including antibody- and complement-dependent cellular cytotoxicity (reviewed in ref. 19). In view of the increased expression of TNF-α in ovarian cancer, we treated the two cell lines that release TNF-α in an autocrine fashion, IGROV-1 and OVCAR-3, with infliximab and assessed the effect on TNF-α mRNA. TNF-α mRNA decreased by 30% to 40% in IGROV-1 and OVCAR-3 following 3 and 6 hours of incubation, respectively, with 10 μg/mL of infliximab (Fig. 6).

Next, we asked whether the abnormal expression and regulation of TNF-α in malignant cells was, in part, responsible for modulating the proinflammatory cytokine milieu in ovarian cancer. First, we evaluated the constitutive and TNF-α-inducible levels of IL-1α, IL-1β, IL-6, CCL2, CXCL8, and M-CSF mRNA in the IGROV-1 cell line using quantitative real-time RT-PCR. All these cytokines were constitutively expressed and TNF-α-inducible in the IGROV-1 cell line (Fig. 7A). In particular, TNF-α induced a sustained increase in IL-6 and CCL2 mRNA with maximal levels by 24 hours, contrasting with a transient increase in the expression of the remaining cytokines (maximal by 1–3 hours). We then screened the remaining cell lines, primary tumor cells and normal ovarian surface epithelial cells, selecting the time point that showed maximal cytokine mRNA induction in the IGROV-1 cell line by TNF-α. All the cell types, except for the SKOV-3 cell line, expressed similar levels of constitutive cytokine mRNA with TNF-α inducing further up-regulation of proinflammatory cytokine mRNA (data not shown). Of note, TNF-α differentially modulated cytokine mRNA levels with de novo transcription required for IL-6 induction and enhanced mRNA stability, a feature of CCL2 induction in the IGROV-1 and OVCAR-3 cancer cell lines (Fig. 7B).

We confirmed both constitutive and TNF-α-inducible levels of IL-6, CXCL8, CCL2, and M-CSF proteins in the supernatant of malignant and normal ovarian surface epithelial cells (Table 1). No TNF-α-inducible IL-6 or CCL2 protein was detected in the SKOV-3 cell line, confirming the earlier findings of weak or absent mRNA induction in response to TNF-α. Constitutive IL-1α was absent but could be induced by TNF-α in both malignant and normal ovarian surface epithelial cells. Although IL-1β protein was detected in the supernatant of normal ovarian surface epithelial cells, it was absent in the supernatant of malignant epithelial cells, consistent with published data (20).

We sought to understand further the role of TNF-α in the regulation of the ovarian cancer cytokine network by treating IGROV-1 and OVCAR-3 cells with infliximab. We revealed a 10% to 15% reduction in constitutive CCL2 (P < 0.001) and IL-6 (P < 0.01) protein in the supernatant of IGROV-1 and OVCAR-3 cells exposed to infliximab for 24 hours (Fig. 8A and B). Infliximab had no effect on the levels of IL-1α, CXCL8, or M-CSF (data not shown). On the other hand, RNA interference directed against TNF-α largely abrogated IL-6 production by IGROV-1 cells (Fig. 8C), suggesting a critical role for TNF-α in maintaining endogenous levels of this cytokine, known to be linked to poor prognosis in ovarian cancer (21).

Chronic inflammation is recognized as an important component of the tumor-stromal interaction and is increasingly considered to be a valid target for modulation (2, 3). Persistent inflammation triggered by TNF-α may predispose to DNA damage, angiogenesis, invasion and metastasis, and local immunosuppression (22). Animal models investigating the link between inflammation and tumorigenesis have emphasized a central role for tumor epithelial TNF-α in this process. For instance, TNF-α is involved in the progression of chemically induced skin cancer with TNF-α^−/− mice displaying a 10-fold reduction in papilloma growth (23). Similarly, our laboratory has shown a modest antitumor effect of a TNF-α neutralizing antibody in models of skin and breast tumorigenesis (24). Pikarsky et al. (25) have recently shown that neutralizing stromal TNF-α is also protective in an animal model of inflammation-associated liver cancer. In the present study, we have compared the expression and regulation of TNF-α in normal and malignant ovarian epithelium based on previous work linking TNF-α expression with ovarian tumor progression (4). We confirmed that TNF-α is overexpressed in ovarian tumor epithelium and we also provide evidence for dysregulated TNF-α activity with the promotion of a proinflammatory tumor microenvironment.

Our results obtained using matched normal tumor tissue show that TNF-α mRNA is increased in ovarian cancer. This was attributed in part due to significantly increased levels of constitutive TNF-α mRNA within the ovarian tumor epithelium as compared with normal ovarian surface epithelial cells. Furthermore, our studies of TNF-α mRNA decay indicate that posttranscriptional regulation is an important mechanism underlying ovarian tumor cell TNF-α production. In renal cancer, the von Hippel-Lindau mutation (also described in clear cell carcinoma of the ovary) is causally linked to increased TNF-α expression via derepression of mRNA translation (26, 27). TNF-α also regulated the stability of TNF-α mRNA as well as CXCL8 and vascular endothelial growth factor mRNA in glioma cells (28). The mechanism involved HuR, an RNA-binding protein that binds to the AU-rich sequence in TNF-α mRNA (29). As HuR is up-regulated in ovarian cancer (30, 31) and responds to stimulation by TNF-α (32), this pathway may be involved in stabilizing TNF-α mRNA in ovarian cancer.

The high levels of TNF-α mRNA observed in ovarian tumor cells were partly suppressed by infliximab. This may be explained either by a direct effect of infliximab on blocking the autocrine TNF-α loop (i.e., simple sequestration of soluble TNF-α) or the triggering of inhibitory reverse signaling by binding to transmembrane TNF-α (33). The latter is supported by recent work showing that infliximab can bind to the surface of ovarian tumor cells.⁴

Besides regulating TNF-α expression, TNF-α also stimulated the differential release of other proinflammatory cytokines in malignant and normal ovarian surface epithelial cells. IL-6 is increased in the serum and ascites of patients with malignant ovarian disease compared with patients with benign ovarian disease, and indicates a poor prognosis (21). The chemokine CCL2 has been implicated in the recruitment of tumor-associated macrophages in ovarian cancer (34). Infliximab produced a modest reduction of tumor-derived IL-6 and CCL2, with TNF-α RNAi indicating that the former is largely dependent on endogenous TNF-α. M-CSF, a potent macrophage chemoattractant was also induced by TNF-α and is aberrantly expression in gynecological tumors (12). Recent work in breast cancer has shown a critical requirement for M-CSF in the modulation of the macrophage content of developing tumors (35).

Currently, clinical trials aimed at inactivating TNF-α in solid tumors are under evaluation. A recent phase I study of the specific TNF-α antagonist etanercept (TNFRII:Fc fusion protein) in ovarian cancer provided biochemical (CA125) and radiological evidence for prolonged disease stabilization in 6/30 patients with advanced disease (36). Maisey et al. (37) documented a partial response in 3 of 19 patients with renal cell cancer treated with single agent infliximab although the mechanism at present is unclear. Studies of infliximab are also ongoing in patients with ovarian cancer and should enable delineation of the role of TNF-α within the ovarian cytokine network. Of note, the cancer profiling array also revealed increased TNF-α mRNA expression in lung, breast, cervix, melanoma, and thyroid tumors (data not shown). Lastly, in view of the “double-edged” nature of the inflammatory and immune response in cancer, it will be important to continue careful monitoring of patients enrolled on such studies. Although TNF-α may trigger a proinflammatory protumorigenic milieu, an acute inflammatory response such as that observed with immune manipulation (38) or with hyperinduction of TNF-α may produce an antitumor effect (39).

In conclusion, we have identified a role for TNF-α in the transcriptional and posttranscriptional regulation of endogenous TNF-α in ovarian cancer. Increased understanding of the posttranscriptional regulation of TNF-α, in particular, may lead to more potent drugs that target TNF-α expression. We also showed that TNF-α differentially regulates various proinflammatory cytokines expressed in ovarian tumor epithelium and propose that TNF-α may be considered a key regulator of this cytokine pattern in vivo. Finally, TNF-α antagonists may provide novel therapeutic approaches by targeting the inflammation associated with the tumor microenvironment of ovarian cancer.