Tumor Necrosis Factor-α Promotes Malignant Pleural Effusion

A malignant pleural effusion (MPE) presents a frequent (∼500 new cases per million population per year) and dismal event in the course of various malignancies, predominantly adenocarcinomas of the lungs and other organs, signaling incurability, shortened life expectancy, and severely compromised quality of life (1–4). Pleurodesis (chemical-induced pleural fibrosis aimed at eliminating the pleural space) and chemotherapy are currently used to block the reaccumulation of a MPE; however, these therapies are nonspecific, often ineffective, and associated with significant morbidity (5, 6). Improved understanding of the pathobiology of MPE will hopefully lead to the development of effective treatment methods (7). For this, we recently developed a model of MPE in immunocompetent mice that recapitulates salient features of the human disease, including tumor-associated inflammation, angiogenesis, and vascular hyperpermeability, and is relevant for the study of its pathogenesis and potential treatment innovations (8).

The development of this and other models of cancer in immunocompetent hosts is vital, because recent evidence from human and animal studies has linked inflammation with enhanced formation, growth, and metastasis of tumors (9–15). In addition, several lines of evidence support the notion that tumor cells produce mediators that sculpt the tumor microenvironment in favor of sustained growth and metastasis (9, 10, 16, 17). Human MPE contains inflammatory cells and mediators (1, 3, 18), but their role in disease progression is uncertain (19–21).

Tumor necrosis factor-α (TNF-α) is a proinflammatory cytokine produced by a variety of cell types (22). TNF-α acts on multiple signaling pathways, including nuclear factor-κB (NF-κB), caspase, ceramide, and c-Jun NH₂-terminal kinase, to affect cell proliferation, differentiation, and death (22). In mice, host-derived TNF-α has been found to promote metastatic tumor growth depending on tumor cell activation of NF-κB (13). However, the cytokine is expressed by several human cancers, including lung cancer (22–24), raising the possibility that tumor-derived TNF-α may play a similar role. Indeed, recent work has shown that TNF-α functions as an autocrine survival signal for ovarian epithelial cancer cells (25), whereas it is known that the cytokine induces vascular permeability (26–28), an event central to the pathogenesis of MPE (1–3, 8). In a mouse model of MPE in immunocompetent mice, we have previously found sustained activation of NF-κB in pleural tumor cells, where tumor cell–specific NF-κB inhibition limited MPE progression and local TNF-α elaboration (8). Collectively, these results suggested a possible role for tumor-derived TNF-α in driving MPE progression. In these studies, we hypothesized that TNF-α promotes the formation and progression of MPE. Using the model of Lewis lung adenocarcinoma (LLC)–induced MPE in the C57BL/6 inbred mouse, we show that tumor cells produce TNF-α, which promotes MPE in an autocrine fashion.

Reagents. sTNFR:Fc (etanercept, Enbrel), a dimeric fusion protein consisting of the extracellular ligand-binding portion of the human 75 kDa TNF receptor linked to the Fc portion of human IgG1 (29), was purchased from the pharmacy; TNF-α was from Boehringer Mannheim; the proteasome inhibitor carbobenzoxy-l-leucyl-l-leucyl-l-leucinal (MG-132) was from American Peptide Co.; neutralizing anti–vascular endothelial growth factor (VEGF) antibody and recombinant mouse interleukin 6 (IL-6) were from R&D Systems; and lipopolysaccharide (LPS; serotype 055:B5) and Evans' blue were from Sigma Chemical. Kotylostatin was synthesized by ENP (30–32).

Cell line, culture, and transfection. Mouse lung adenocarcinoma (LLC) and human prostate cancer (LNCaP clone FCG), mesothelioma (MSTO-211H), and lung adenocarcinoma (A549) cell lines were purchased from the American Type Culture Collection and cultured at 37°C in 5% CO₂–95% air using DMEM 10% FCS supplemented with glutamine and 100 mg/L penicillin/streptomycin. LLC cells stably expressing a NF-κB reporter plasmid (NF-κB.GFP.LUC; pNGL) have been described elsewhere (8, 33). A549 cells were transduced with pNGL using the same methods.

Animal model. Wild-type (tnfa+/+) and TNF-α knockout (tnfa−/−) mice (C57BL/6 background; ref. 34) were purchased from Biomedical Sciences Research Center “Alexander Fleming” Vari, Greece, and inbred at the Animal Care facilities of Department of Critical Care and Pulmonary Service, General Hospital “Evangelismos”, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece, and Division of Allergy, Pulmonary, and Critical Care Medicine, Vanderbilt University School of Medicine, Nashville, TN. Animal experiments were approved by the Veterinary Administration Bureau of the Prefecture of Athens, Greece, and the Institutional Animal Care and Use Committee at Vanderbilt University. Mice used for experiments were sex, weight (19–24 g), and age (8–10 weeks) matched. Intrapleural injections of wild-type or pNGL LLC cells (1.5 × 10⁵/50 μL PBS/mouse) in tnfa+/+ and tnfa−/− mice (day 0), as well as sacrifice and specimen collection (day 14), were done as described previously (8). For survival studies, mice were observed daily and sacrificed when moribund (mice killed at day 14 for other analyses were censored).

In vivo TNF-α neutralization. For in vivo TNF-α neutralization, mice received i.p. 300 μg sTNFR:Fc/100 μL vehicle (control, 100 μL vehicle) on days 3, 5, 7, 9, 11, and 13 after LLC cells, a dosing reported to effectively neutralize TNF-α (35). In a separate dose-response experiment, mice received 0, 10, 100, or 300 μg sTNFR:Fc/100 μL vehicle on days 3, 5, 7, 9, 11, and 13 after LLC cells.

Pleural tumor enumeration and processing. Pleural tumors were enumerated by three independent and blinded readers under a dissecting microscope, and the average number was used for analyses. Parietal pleural tumors were dissected, snap-frozen in liquid nitrogen, and stored at −80°C. Tumor tissue was suspended in passive lysis buffer (volume 1 mL; Promega), and cytoplasmic protein extracts were collected after Dounce homogenization and centrifugation (16,000 × g; 5 min). All measurements in tumor extracts were corrected for protein content.

In vivo vascular permeability assays. To determine pleural vascular permeability in mice bearing MPEs, the animals received 200 μL of 4 mg/mL Evans' blue (total dose 0.8 mg) i.v. on day 14 after LLC cells, and were killed 1 h later. Pleural fluid Evans' blue levels were determined as described previously (8). To evaluate vascular permeability induced by mediators contained in mouse MPEs, we used a modified Miles vascular permeability assay (7, 27, 36, 37). MPE supernatants from tnfa+/+ mice or PBS (volume 50 μL) were injected at different spots of the shaved dorsal skin of C57BL/6 mice. Five spots on each mouse received premixed PBS, MPE + PBS, MPE + sTNFR:Fc, MPE + neutralizing anti-VEGF antibody, or MPE + sTNFR:Fc + anti-VEGF antibody. sTNFR:Fc and anti-VEGF were added at isomolar ratios to TNF-α and VEGF, respectively, after measurement of cytokine levels in MPE. The mice received 200 μL of 4 mg/mL Evans' blue (total dose 0.8 mg) i.v. immediately after dermal injections and were killed 30 min later. The skin was removed, the test sites were photographed, and the area of Evans' blue extravasation was determined using ImageJ freeware (Rasband 1997–2006, available online⁹

Bioluminescence imaging. In vivo bioluminescence imaging was done at day 10 after pNGL LLC cells, before and 6 and 24 h after a single i.p. dose of sTNFR:Fc or vehicle, using the Xenogen IVIS cooled charged coupled device (Xenogen). Data were analyzed using Living Image v.2.50 (Xenogen) and IgorPro (Wavemetrics), as described previously (8, 33, 38).

Cytology. Fifty thousand pleural fluid cells were used for cytocentrifugal specimen (cytospin) preparation. The slides were air dried, fixed in methanol for 10 s, and stained with May-Gruenwald-Giemsa. Distinct cell types were enumerated as a percentage of 400 cells on the slide.

Cytokine determinations. Mouse and human TNF-α (detection limits 5.1 and 0.12 pg/mL, respectively) and VEGF (detection limits 3.0 and 7.8 pg/mL, respectively) were determined using commercial ELISA kits (R&D). Alternatively, mouse TNF-α, IFN-γ, monocyte chemoattractant protein-1 (MCP-1), and IL-6, IL-10, and IL-12p70 were measured using a cytometric bead array (BD; detection limits 7.3, 2.5, 52.7, 5.0, 17.5, and 10.7 pg/mL, respectively; ref. 39).

Histology. Mouse lungs and attached tumors, liver, spleen, heart, and kidney were fixed in 10% neutrally buffered formalin (24 h) and 70% ethanol (3 days). Tumors were dissected and embedded in paraffin. Five-micrometer-thick sections were cut, mounted on glass slides, and stained with H&E. Alternatively, tissue sections were immunostained using antibodies for proliferating cell nuclear antigen (PCNA; Santa Cruz Biotechnology), terminal deoxyribonucleotide transferase–mediated nick-end labeling (TUNEL; Roche Molecular Biochemicals), and factor VIII–related antigen (Invitrogen), as described previously (40–42). PCNA labeling was quantified by two blinded readers by counting the percentage of labeled cells in respect to total cells in 10 different, nonoverlapping, high-power fields (Å, 1,000) from each tumor (41). TUNEL was quantified by counting the number of positive cells in 10 different, nonoverlapping, high-power fields (Å, 600) from each tumor (42). Factor VIII–related antigen labeling was quantified by the “hotspot” method (40). Briefly, the whole tumor area was scanned at low power (Å, 40) to detect areas of enhanced angiogenesis (hotspots). Subsequently, five different, nonoverlapping, high-power fields (Å, 600) within hotspots from each tumor were assessed by two blinded readers for the number of individual cells or clusters of cells (counted as a unit) exhibiting immunoreactivity. Results were averaged for each tumor, and then for each mouse.

Adenoviral vectors and transient LLC cell infection. Adenoviral vectors expressing a dominant inhibitor of NF-κB (Ad-IκBα-DN), and the active NF-κB subunit RelA/p65 (Ad-RelA), inhibiting and activating NF-κB, respectively, have been previously reported (38, 43). LLC cells were infected at a multiplicity of infection of 500 for 24 h (conditions giving optimal results during titration studies).

Biochemical and cellular assays. Protein was determined using the Bio-Rad protein assay. A 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt (MTS) assay was used to assess cell viability (Promega). For cell experiments, LLC, A549, MTSO-211H, and LnCaP clone FCG cells were plated at equal densities in 12- or 96-well culture dishes. Indicated treatments were applied when cells were 20% to 30% confluent. All cell experiments were done in triplicate.

Luciferase assay. To determine luciferase bioactivity in tumor tissue and whole-cell lysates, tumor tissue was homogenized as detailed above and cells cultured in 12-well culture dishes were lysed in 100 μL passive lysis buffer (Promega). Twenty-microliter samples were added to 100 μL luciferase assay system (Promega) and photon emission was immediately determined using a “Junior” luminometer (EG&G Berthold). Measurements in whole-cell lysates are given as NF-κB activity ratio, after correction for control cells in neighboring wells. Measurements in tumor tissue are given as relative luciferase units per microgram of protein, as described previously (38, 43).

Human MPE. MPEs were obtained from the initial diagnostic thoracenteses done in 36 consecutive patients with MPE who were treated at Department of Critical Care and Pulmonary Services, General Hospital “Evangelismos”, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece, and 2^nd Pulmonary Department, “Attikon” University Hospital, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece, from May 2006 till January 2007. The study was approved by the Ethics Committees of both institutions and all patients gave written informed consent. The radiologic size of MPE was assessed using a semiquantitative scale from 0 to 5: 0, no detectable effusion; 1, blunting of costophrenic angle; 2, effusion occupying 0% to 25% of hemithorax; 3, effusion occupying 25% to 50% of hemithorax; 4, effusion occupying 50% to 75% of hemithorax; 5, effusion occupying 75% to 100% of hemithorax. Human MPEs were handled as mouse MPEs (see above).

Statistical analysis. All values represent mean ± SE. Survival is given as median (confidence interval) and human MPE size is given as median ± interquartile range. Student's t test or one-way ANOVA with least significant difference post hoc test was used to test for differences in the means between two or multiple groups, respectively. Kaplan-Meier analysis with log-rank test between groups was used for survival studies. Correlation was done using Spearman's correlation analysis, and TNF-α in human MPE categorized according to size was compared using Kruskal-Wallis analysis and the Mann-Whitney U test. All P values are two-tailed; P values <0.05 were considered significant. All statistical analyses were done using the Statistical Package for the Social Sciences v.13.0.0 (SPSS).

Tumor-derived TNF-α promotes MPE. To determine the effects of TNF-α on MPE progression, tnfa+/+ and tnfa−/− mice (C57BL/6 background) received intrapleural LLC cells and were treated with 300 μg (∼12 mg/kg) sTNFR:Fc (etanercept) or vehicle i.p. on days 3, 5, 7, 9, 11, and 13 after LLC cells. At day 14, all mice had a MPE and pleural tumors. However, although the mean MPE volume and the mean number of pleural tumors were not different between tnfa+/+ and tnfa−/− mice, they were significantly reduced by sTNFR:Fc treatment of both tnfa+/+ and tnfa−/− mice (Fig. 1A). Lower doses of sTNFR:Fc (10 μg), comparable with the doses used in humans with rheumatoid arthritis (∼0.4 mg/kg), were also effective in reducing tumor burden in the MPE model (Fig. 1B). However, maximal effects with no toxicities were observed with the highest dose (300 μg). Tumor cells produced TNF-α in vivo, because the cytokine was detected in MPE from tnfa−/− mice (Fig. 1C), and in vitro, because TNF-α was found in medium conditioned by confluent LLC cells (48 h: 9 ± 1 pg/mL; 72 h: 16 ± 1 pg/mL; n = 3/time point). Because TNF-α neutralization retarded MPE progression, we sought to determine whether this treatment would affect survival. Indeed, sTNFR:Fc-treated tnfa+/+ and tnfa−/− mice survived significantly longer after intrapleural LLC cells, compared with vehicle-treated mice of the same genotype (Fig. 1D). These results indicated that tumor-derived TNF-α promotes intrapleural fluid accumulation and tumor dissemination and tumor-related death in the mouse model of MPE.

TNF-α promotes MPE-associated vascular permeability. We next assessed pleural vascular permeability, a key mechanism of MPE formation, in tnfa+/+ and tnfa−/− mice treated with sTNFR:Fc or vehicle. In a first line of experiments, MPE-bearing mice received i.v. Evans' blue 1 h before sacrifice (day 14), with subsequent measurement of the dye in MPE (8). Although MPE levels of the dye were not different between tnfa−/− and tnfa+/+ mice, they were significantly reduced in sTNFR:Fc-treated tnfa+/+ and tnfa−/− mice, indicating induction of pleural vascular permeability by tumor-derived TNF-α (Fig. 2A). Hence, we postulated that TNF-α contained in MPE directly contributes to vascular hyperpermeability in addition to VEGF, an established permeability enhancer (36, 37). We tested this hypothesis using a skin vascular permeability (Miles) assay (7, 27, 36, 37). For this, individual mice received intradermal MPE, with and without prior neutralization of TNF-α and/or VEGF contained in the MPE samples (Figs. 2B and 3A). This line of experiments supported a vascular hyperpermeability–inducing role for TNF-α, possibly a mechanism of TNF-α–induced pleural fluid accumulation. As shown here and below, this effect of TNF-α may be both indirect (via induction of VEGF) and direct, in addition to the effect of VEGF.

TNF-α does not significantly contribute to tumor-associated inflammation. We next assessed pleural and systemic inflammatory variables in MPE-bearing tnfa+/+ and tnfa−/− mice treated with sTNFR:Fc or vehicle. With the exception of TNF-α (Fig. 1C), we observed no significant differences in pleural fluid and blood inflammatory cells and mediators (IFN-γ, MCP-1, IL-6, IL-10, and IL-12p70) between the experimental groups (data not shown). These results suggested that TNF-α does not function to alter the inflammatory response associated with MPE.

Tumor-derived TNF-α promotes NF-κB activation and TNF-α expression in pleural tumors. We have previously shown that tumor cell NF-κB activation promotes MPE in the mouse model (8). Because TNF-α activates NF-κB (22), and TNF-α neutralization has been shown to inhibit NF-κB in nontumor tissue (28, 44), we investigated whether TNF-α induced NF-κB in tumor cells in vivo. For this, we established MPEs in tnfa+/+ and tnfa−/− mice using LLC cells stably expressing a NF-κB luciferase reporter (pNGL; refs. 8, 33). Pleural tumor tissue from tnfa+/+ and tnfa−/− mice showed similar levels of NF-κB–driven luciferase activity. However, treatment with sTNFR:Fc significantly inhibited tumor cell NF-κB in both tnfa+/+ and tnfa−/− mice (Fig. 4,A, left). To further corroborate this effect, tnfa+/+ mice were serially imaged for bioluminescence at day 10 after intrapleural pNGL LLC cells. Imaging was done before and after a single i.p. dose of sTNFR:Fc or vehicle. Compared with vehicle, sTNFR:Fc inhibited pleural tumor cell NF-κB at 6 h after injection (Figs. 3B and 4,A, middle). In line with reduced NF-κB activation, sTNFR:Fc treatment decreased TNF-α expression in pleural tumors (Fig. 4 A, right). These results collectively indicated that tumor-derived TNF-α activates NF-κB and promotes its own expression in pleural tumors.

Tumor-derived TNF-α sustains pleural tumor cell survival. As NF-κB activation promotes the survival and inhibits the apoptosis of tumor cells (22, 45), we next assessed proliferation and apoptosis of tumor cells in vivo using PCNA immune labeling and TUNEL of pleural tumor tissue obtained from our four experimental groups. Although tumor cell proliferation was not different between groups (data not shown), tumor cell apoptosis was significantly increased in pleural tumors from tnfa+/+ and tnfa−/− mice treated with sTNFR:Fc, compared with vehicle-treated mice (Figs. 3C and 4B). In vitro, TNF-α, sTNFR:Fc, or both, had no effect on tumor cell viability determined by a MTS assay (data not shown). These results indicated that tumor-derived TNF-α functions to promote the survival of pleural tumor cells, an effect restricted to the in vivo setting.

TNF-α induces NF-κB and TNF-α in tumor cells in vitro. Based on these findings, we postulated positive feedback of TNF-α expression, NF-κB activation, and further promotion of TNF-α expression in tumor (LLC) cells. Indeed, NF-κB–dependent luciferase activity in pNGL LLC cells in vitro was significantly increased by TNF-α (Supplementary Fig. S1A), whereas it was not affected by LPS and IL-6 (data not shown). These treatments had no effect on LLC cell viability (data not shown). TNF-α–induced NF-κB activation in LLC cells was dose dependently blocked by sTNFR:Fc (Fig. 4C). We next tested whether LLC cell NF-κB activation, in turn, contributed to TNF-α expression to provide a positive loop. To this end, we transiently infected pNGL LLC cells with control (Ad-GFP) and NF-κB activator (Ad-RelA) and inhibitor (Ad-IκBα-DN) adenoviral vectors (38, 43). Ad-RelA and Ad-IκBα-DN infection, respectively, activated and inhibited NF-κB in LLC cells (Supplementary Fig. S1B), without changes in cell viability (data not shown). In addition, TNF-α was induced in a time-dependent manner specifically in Ad-RelA–infected LLC cells (Supplementary Fig. S1C). We next examined whether TNF-α can induce its own expression in LLC cells. For this, LLC cells were briefly exposed to TNF-α, washed, and de novo TNF-α secretion was assessed. Pretreatment of LLC cells with TNF-α resulted in time-dependent enhancement of TNF-α elaboration, compared with PBS (Fig. 4D). It was impossible to test directly whether TNF-α–induced TNF-α expression by LLC cells was dependent on NF-κB, as inhibition of NF-κB via Ad-IκBα-DN followed by TNF-α exposure led to cell death (data not shown), as previously described (45). Hence, TNF-α potently and specifically activates NF-κB and stimulates TNF-α gene expression in LLC cells.

TNF-α induces VEGF and new vessel formation in pleural tumors. Tumor-derived VEGF has been shown to contribute significantly to the development of MPE (36, 37), and high levels of VEGF are present in mouse MPE (8). Because TNF-α can promote VEGF expression by nonmalignant cells (26, 27), we sought to determine the expression of this cytokine in pleural tumors from tnfa−/− and tnfa+/+ mice treated with sTNFR:Fc or vehicle. Compared with vehicle-treated mice, VEGF expression in tumor tissue from sTNFR:Fc-treated tnfa+/+ and tnfa−/− mice was significantly reduced (Fig. 5A). In addition, we observed an unequivocal reduction of newly formed vessels in tumor tissue from tnfa+/+ and tnfa−/− mice treated with sTNFR:Fc (Figs. 3D and 5B). These results clearly indicated that tumor-derived TNF-α promotes VEGF expression and angiogenesis in pleural tumors of the MPE model system.

TNF-α induces VEGF in tumor cells in vitro in a neutral sphingomyelinase–dependent manner. Based on the above findings, we hypothesized that TNF-α can contribute to VEGF expression by tumor cells. Indeed, LLC cells in vitro produced VEGF in a time-dependent manner further inducible by TNF-α (Fig. 5C). On the contrary, exposure of LLC cells to high doses of LPS or IL-6 had no effect on VEGF expression (data not shown). In addition, modulation of NF-κB in LLC cells using Ad-RelA and Ad-IκBα-DN did not affect VEGF elaboration by LLC cells (Supplementary Fig. S1D). These results collectively indicated that TNF-α potently and specifically promotes the expression of VEGF by LLC cells in a NF-κB–independent manner. Recent work has shown that the ceramide pathway and especially neutral sphingomyelinase, an enzyme generating intracellular ceramide from membrane lipids, contributes to proinflammatory gene expression in epithelial and endothelial cells (30). To test whether TNF-α–induced VEGF expression by LLC cells is mediated by neutral sphingomyelinase, we used a novel and specific small-molecule inhibitor of the enzyme, the scyphostatin analogue kotylostatin, which nonreversibly and completely inhibits the enzyme at a concentration of 5 μmol/L (31, 32). Although concentrations of 5 μmol/L of the compound had no effect on LLC cell survival and NF-κB activity (data not shown), they completely abrogated TNF-α–induced VEGF expression (Fig. 5D). These data indicated that TNF-α–induced VEGF expression by LLC cells is mediated by neutral sphingomyelinase and the ceramide pathway.

TNF-α promotes NF-κB activation and induces TNF-α and VEGF in human cancer cell lines. To expand the scope of our findings, we stably transduced a human lung adenocarcinoma cell line (A549) with the pNGL reporter. In A549 cells, NF-κB–dependent luciferase activity was induced by TNF-α (Fig. 6A). In addition to LLC cells, TNF-α enhanced the expression of TNF-α and VEGF in human prostate (LNCaP clone FCG), mesothelioma (MSTO-211H), and lung adenocarcinoma (A549) cell lines (Fig. 6B). As these types of tumors are associated with MPE in humans, these data collectively suggested that TNF-α–induced activation of NF-κB and expression of TNF-α and VEGF may contribute to the pathogenesis of human MPE.

TNF-α content of human MPE is correlated with VEGF levels and effusion size. To further seek whether our preclinical findings may apply to human MPE, we determined the levels of TNF-α and VEGF in 36 human MPEs. The primary tumors were lung cancer in 24 (non–small cell in 21; small cell in 2; large cell neuroendocrine in 1); mesothelioma in 4; breast cancer in 3; and colon, gastric, ovarian, thyroid cancer, and Hodgkin's lymphoma in 1 each. We found a significant correlation between TNF-α and VEGF levels in MPE from these patients (Fig. 6C). In addition, larger MPEs on a semiquantitative radiologic scale had higher TNF-α content, suggesting a link between TNF-α and MPE size (Fig. 6D).

TNF-α, a cytokine expressed by various benign and malignant cells, has been detected in the microenvironment of solid tumors (13, 23–25) and in the pleural cavity of patients with MPE (46). Although the cytokine was named after its ability to induce malignant cell death (47), recent studies on models of solid tumors have shown that TNF-α produced by either tumor (13) or host (25) cells promotes cancer growth and metastasis. However, whether TNF-α is an inducer or an antagonist of the process underlying pleural fluid accumulation in patients with malignant effusions remains unclear. The present study examined for the first time the role of this cytokine in MPE pathogenesis. To assess the effects of TNF-α of host and tumor origin, we used tnfa−/− mice (whose host cells cannot produce the cytokine) and in vivo neutralization of all extracellular (both host and tumor cell–derived) TNF-α. Our results show that tumor cells produce TNF-α, which promotes: (a) pleural fluid accumulation and tumor dissemination; (b) cancer-related death; (c) pleural vascular hyperpermeability; (d) tumor cell NF-κB activation and further TNF-α expression; and (e) tumor cell VEGF expression and angiogenesis. Importantly, TNF-α–induced TNF-α and VEGF expression are mediated, respectively, via NF-κB–dependent and NF-κB–independent pathways and, in addition to mouse, also occur in human cancer cell lines. In patients with MPE, high pleural fluid TNF-α levels were associated with increased VEGF levels and increased size of the effusion. These studies expand our previous observations that NF-κB activation in lung adenocarcinoma cells exerts a positive effect on MPE progression and local TNF-α production in the mouse pleural cavity. Our data imply that therapeutic targeting of TNF-α may halt the progression of the disease.

How can TNF-α promote MPE? Could it be due to direct (enhancement of tumor cell survival) or indirect (mediator secretion, angiogenesis) effects? Our data indicate that both may be true: On the one hand, TNF-α promoted tumor cell survival in vivo via limiting apoptosis. This effect of the cytokine was linked to enhanced activation of NF-κB, a known antiapoptotic transcription factor (45). Thus, our data indicate that TNF-α may act as a survival signal for tumor cells in vivo. Why this in vivo function of the cytokine was not recapitulated in vitro is unclear. One may postulate that this effect of TNF-α may require the presence of other mediators/cells that are absent in culture conditions. On the other hand, we clearly show that TNF-α exerts indirect effects that support tumor growth and malignant effusion formation. In this regard, TNF-α (a) induces VEGF and angiogenesis in pleural tumors; (b) induces VEGF in mouse and human cancer cells in a neutral sphingomyelinase–dependent manner; and (c) induces vascular permeability both directly and via induction of VEGF. It is worth noting that in addition to tumor cells, TNF-α up-regulates the expression of angiogenic cytokines, including VEGF, by endothelial, immune, and mesenchymal cells (26, 27), potentially leading to additional proangiogenic and propermeability effects in the pleural cavity of patients with MPE. Our findings not only indicate for the first time that TNF-α is an important inducer of pleural vascular hyperpermeability but also suggest that TNF-α contained in the pleural cavity is as potent as VEGF in inducing fluid extravasation (see Figs. 2B and 3A), a function central to the formation of a malignant effusion. Taken together, TNF-α promotes MPE formation and intrapleural cancer dissemination by limiting tumor cell apoptosis and by inducing tumor angiogenesis and pleural vascular hyperpermeability, effects exerted both directly on the tumor cell via enhanced activation of NF-κB and indirectly on the host milieu via the induction of VEGF expression by tumor cells.

In addition to TNF-α, tumors produce several other proinflammatory mediators, also detected in human MPE, which potentially participate in its pathogenesis (16, 17). Among them, IL-6 has been recently shown to induce signal transducers and activators of transcription 3–mediated VEGF expression, thereby promoting MPE and tumor growth in immunodeficient mice (7). In our hands, IL-6 had no effect on growth or VEGF expression in LLC cells. However, our findings do not rule out a possible contribution of IL-6 to the formation and progression of MPE in the LLC-C57BL/6 model, which is nevertheless overridden by an independent effect of TNF-α. Collectively, these different lines of evidence support the concept of tumor and MPE promotion by inflammatory mediators secreted by the tumor cell and raise the possibility that the relative significance of any individual mediator is dependent on the biological properties of the cancer cells infiltrating the pleural tissues.

Another interesting aspect of our findings is that the functionally important fraction of TNF-α seems to be produced by the tumor cells and not the host. Because the cytokine has been detected in the tumor microenvironment including MPE (8, 13, 23–25, 46), and can be produced by both tumor and host cells, it is not surprising that the question whether the tumor- or host-derived cytokine plays the most crucial role has been raised by different groups. Luo et al. (13) have discovered that host-derived TNF-α is important in the promotion of tumor growth. The findings of a more recent study on ovarian cancer (25) are in concert with ours, indicating that the tumor-derived cytokine exerts the biological effects on the progression of i.p. growth of the malignancy. The reason for the discrepancies between the different studies is unclear. It is reasonable to speculate that they may be ascribed to both anatomic and biological differences between the various models. In the mouse model of MPE, the primary role of tumor versus host-derived TNF-α in driving MPE progression may have several explanations: First, anatomic sequestration of the tumor-derived cytokine in pleural tumors may lead to enhanced functions of locally produced over in-streaming TNF-α (compartment effect). Second, minimal amounts of the cytokine may be sufficient to elicit biological responses, rendering excess TNF-α (i.e., that produced by the host cells) unnecessary (threshold effect). Third, local TNF-α signaling in pleural tumors by membrane-bound cytokine may override the effects of soluble TNF-α (paracrine effect). Along with the above, it should be also acknowledged that our observations were made with a TNF-α–producing cell line (LLC). Hence, in MPE, secondary to tumors characterized by minimal or no production of TNF-α, the host-derived cytokine may be required for the tumor-promoting effects.

Whatever the cellular source of the functionally important fraction of TNF-α, an inhibitor capable to block the activity of TNF-α of any origin inhibited MPE and prolonged survival in this preclinical study. It is therefore tempting to consider potential clinical implications of the present study given that MPE is a common and devastating clinical condition and current therapeutic interventions are not ideal. Thus, novel, effective, more convenient, and safe therapies are needed. In this regard, proinflammatory, angiogenic, and hyperpermeability mediators that promote the formation of a MPE may be therapeutically targeted to halt pleural fluid accumulation, relieve breathlessness, and avoid the adverse effects associated with the contemporary practice of pleurodesis (5). Our work indicates that inhibition of TNF-α signaling may provide a novel means of treating malignant pleural disease, similar to ongoing trials of TNF-α neutralization for solid tumors (48, 49). Skepticism about possible protumor effects of future anti–TNF-α therapy for MPE is most likely unsubstantiated, given the short survival expected by this patient group (1–4) and the lack of evidence of tumor progression in large cohorts of patients treated with anti–TNF-α agents over long periods of time (50). However, patients with MPE are often debilitated, and a major proximate cause of death in this group is infection. TNF-α neutralization in the setting of local or systemic infection can have devastating consequences on host defense, limiting the clinical applicability of these findings.

The dose of etanercept used in these studies (300 μg/animal ∼12 mg/kg) was almost 30 times higher than the dose administered to humans with rheumatoid arthritis. Our studies and the literature (35) indicate that this dosing effectively blocks TNF-α in mice without apparent toxicities. A much lower dose (10 μg/animal ∼0.4 mg/kg), comparable with the one used in humans, was also effective in reducing intrapleural tumor dissemination and pleural fluid accumulation in the mouse model of LLC-induced MPE. The optimal dosing of etanercept for the treatment of MPE in humans remains to be determined by future clinical studies.

In conclusion, we identified tumor-derived TNF-α as an important effector of MPE formation and progression. In addition to tracing the effects of TNF-α on MPE-associated angiogenesis and vascular hyperpermeability, these studies provide preclinical evidence for the efficacy of TNF-α blockade against MPE. Our observations may serve to generate a hypothesis that is probably worthy to be examined in future clinical trials.

Grant support: “Thorax” Foundation (Athens, Greece), NIH grants HL66196 and HL61419 (T.S. Blackwell), and research grant by Wyeth Hellas SA (G.T. Stathopoulos and I. Kalomenidis).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

G.T. Stathopoulos and I. Kalomenidis received funding (30,000) from Wyeth Hellas SA, the marketer of etanercept in Greece. Wyeth had no involvement in study design, collection, analysis, and interpretation of data; writing of the report; and in the decision to submit the report for publication.

We thank Dr. Anna Marantidou, Zoe Kollia, and Aggeliki Apostolopoulou for professional veterinarian, animal care, and editorial assistance, respectively.