Cyclooxygenase-2 Inhibition Suppresses α_vβ₆ Integrin–Dependent Oral Squamous Carcinoma Invasion

Cyclooxygenases (COX) catalyze the key step in prostanoid and thromboxane biosynthesis and are targets of nonsteroidal anti-inflammatory drugs (NSAID). Two human isoforms exist: COX-1, expressed constitutively in most mammalian cells, generates prostaglandins necessary for normal physiologic function, whereas COX-2, normally undetectable, is induced rapidly by stimuli, including cytokines, oncogenes, and tumor promoters (1, 2). Elevated COX-2 expression occurs in many carcinomas, including oral squamous carcinoma (OSCC), where it contributes to tumor progression (1–3). Transgenic female mice overexpressing COX-2, for example, have a high frequency of breast dysplasia and metastatic tumors (4), whereas genetic inactivation of COX-2 in a murine model of familial adenomatous polyposis reduced both the size and number of intestinal polyps (5).

The epithelial integrin α_vβ₆ is a receptor for fibronectin, vitronectin, tenascin, and the latency-associated peptide (LAP) of transforming growth factor-β (e.g., ref. 6). Generally found to have minimal expression in adult epithelium, α_vβ₆ is up-regulated during epithelial remodeling (e.g., wound healing) and tumorigenesis. Aberrant α_vβ₆ expression occurs in numerous carcinoma types, particularly OSCC (7–11). Moreover, increased expression of both α_vβ₆ and COX-2 have been reported in oral dysplastic epithelium, suggesting a role in tumor progression (12, 13).

NSAIDs have marked antitumor activity via COX-2 inhibition. Selective COX-2 inhibitors are chemopreventive in animal models of colon, bladder, and breast cancer (1, 2), whereas epidemiologic studies indicate that NSAIDs reduce breast, esophageal, and colon cancer mortality (1, 2). The effects of NSAIDs on tumor growth and progression are likely to be through multiple mechanisms, including indirect effects on angiogenesis and inflammation, as well as direct effects on tumor cell proliferation and apoptosis (1, 2). Particularly interesting are studies where NSAIDs inhibit tumor cell adhesion, migration, and invasion (1, 2, 14). These changes may be mediated via effects on various integrins (15, 16). For example, aspirin and indomethacin inhibit activation of the platelet integrin α_IIbβ₃ (17) whereas Dormond et al. inhibited endothelial cell migration and angiogenesis using the selective COX-2 inhibitor NS398, effects that were modulated through inhibition of the α_vβ₃-dependent activation of Rac-1 and Cdc42 (18). This may be a reciprocal effect because integrin-mediated adhesion can contribute to stabilization of COX-2 protein levels (19).

Given that both α_vβ₆ and COX-2 are expressed in OSCC and have been reported to promote invasion (20–22), and that NSAIDs may inhibit integrin function, we determined whether NSAIDs could inhibit α_vβ₆-mediated invasive activity.

Immunohistochemistry. Antibodies used were anti-cytokeratin, AE1/AE3 (1:50; DAKO, High Wycombe, United Kingdom); anti-α_vβ₆, 6.2G2 (0.5 μg/mL; Biogen Idec, Cambridge, MA); or anti-COX-2 (1:500; Cayman Chemical Co., Ann Arbor, MI). Antigen retrieval varied according to primary antibody: 0.1% α-chymotrypsin, 0.1% calcium chloride (pH 7.8) for 20 minutes at 37°C (AE1/AE3); Digest-All 3 Pepsin Solution (Invitrogen Corporation, Carlsbad, CA) for 5 minutes at 37°C (6.2G2); microwaving for 30 minutes in 0.1 mol/L citrate buffer (pH 9; COX-2). Endogenous peroxidase was neutralized with 0.45% hydrogen peroxidase in methanol for 15 minutes and primary antibodies applied in TBS (pH 7.6) for 1 hour. Anti-mouse IgG biotinylated secondary antibody (Vectastain Elite ABC Reagent, Vector Laboratories, Burlingame, CA) was applied for 30 minutes followed by peroxidase-labeled streptavidin (Vectastain Elite ABC Reagent; Vector Laboratories) for 30 minutes. Peroxidase was visualized using DAB+ (DAKO) for 7 minutes and counterstained in Mayer's hematoxylin (Sigma-Aldrich, Dorset, United Kingdom).

Twenty OSCC, 39 epithelial dysplasias (18 transforming to OSCC and 21 nontransforming; clinical follow-up range, 5-31 years), and 14 benign polyps showing fibroepithelial hyperplasia were chosen at random; stained for α_vβ₆ and COX-2; and scored according to the Quickscore method (23). The staining intensity was scored out of three (1, weak; 2, moderate; 3, strong), and the proportion of epithelial cells staining positively was scored out of four (1, <25%; 2, 25-50%; 3, 51-75%; 4, 76-100%). The score for intensity was added to the score for proportion to give a score in the range of 0 to 7 and grouped as − (score = 0), + (score = 1-3), ++ (score = 4-5), or +++ (score = 6-7).

Cell culture. Using cDNA transfection techniques, we created a panel of cell lines expressing various levels of α_vβ₆ (20, 21). H357 is an α_v-negative OSCC cell line (24) from which the V3 cell line was generated by transfection of α_v cDNA (25); V3 cells express low levels of α_vβ₆. This cell line was retrovirally infected with β₆ cDNA, creating the VB6 cell line, which has high α_vβ₆ expression. A null transfectant control cell line for the VB6 cells (C1) also was generated at this time (20, 21). BICR6, CA1, and H157 OSCC cell lines were also used: provided by Professors K. Parkinson (BICR6) and I. Mackenzie (CA1) of the Queen Mary's University, London and Professor S. Prime (H357 and H157) of the University of Bristol Dental School. OSCC cells were grown in keratinocyte growth medium (KGM; ref. 20). Human foreskin fibroblasts (supplied by Cell Services of Cancer Research UK, London Research Institute) were maintained in fibroblast growth medium (DMEM supplemented with 10% FCS) at 37°C in an humidified atmosphere.

Flow cytometry. Flow cytometry was done (20), using anti-α_vβ₆ antibody (E7P6; Chemicon International, Harrow, United Kingdom) and FITC-conjugated secondary antibody (DAKO). Negative control used secondary antibody only and was subtracted from the results. Labeled cells were scanned on a FACSCalibur cytometer (BD Biosciences, Oxford, United Kingdom) and analyzed using CellQuest software, acquiring 1 × 10⁴ events. Results show mean fluorescence (arbitrary units, log scale).

RNA interference. RNA interference (RNAi) SMART pool reagents targeting β₆, Rac-1, COX-2, or control (random) sequences were obtained from Dharmacon (Chicago, IL). Cells were seeded into six-well plates and left for 24 hours until ∼40% confluent, then transfected with 100 pmol/well of the relevant duplex pool using Oligofectamine transfection reagent (Invitrogen, Paisley, United Kingdom). Cells were used in assays after 24 to 48 hours. Cells were also lysed and used to verify protein knockdown by Western blotting analysis.

Modulation of Rac-1 activity. Cells were transfected with enhanced green fluorescent protein (EGFP)–tagged, constitutively active Rac-1 [V12Rac-1-GFP (construct made by J. Monypenny, GKT, London, United Kingdom), or vector control (pEGFP-C2; Invitrogen)]. Cells at 70% confluency in six-well dishes were transfected with 0.8 μg of DNA using Effectene transfection reagent (Qiagen Crawley, Surrey, United Kingdom). Twenty-four hours after transfection, cells were used in invasion assays and analyzed by flow cytometry to determine transfection efficiency, which typically was between 40% and 70%.

Adhesion assays. Adhesion assays were done as described (20). Before plating, ⁵¹Cr-labeled cells were incubated with anti-α_vβ₆ antibody (10D5, 1:100; Chemicon International) anti-β₁ antibody (P4C10, 1:100; Chemicon International), control antibody W6/32 (anti-MHC class I; 1:100; provided by W. Bodmer IMM, Oxford), NS398 (20 μmol/L; Cayman Chemical), SC-560 (10 μmol/L; Cayman Chemical), or indomethacin (500 μg/mL; Cayman Chemical) for 10 minutes at 4°C. Adherent cell radioactivity was expressed as a proportion of the total cell radioactivity input. Nonspecific adhesion (attachment to wells coated with bovine serum albumin) was subtracted from test samples.

Transwell invasion assays. Cell invasion assays were done over 72 hours using Matrigel-coated (diluted 1:2 in α-MEM) polycarbonate filters (Transwell, BD Biosciences; ref. 20). Cells invading to the lower chamber were trypsinized and counted on a Casy 1 counter (Sharfe System GmbH, Reutlingen, Germany). Results are expressed relative to invasion of VB6 control treatment (= 100). For blocking experiments, anti-α_vβ₆ antibody (10D5, 1:100 or 6.3G9, 1:100; Biogen Idec), NS398 (20 μmol/L), SC-560 (10 μmol/L), indomethacin (500 μg/mL), prostaglandin E₂ (PGE₂; 1-10 ng/mL; Cayman Chemical), PGF_2α (1-10 ng/mL; Cayman Chemical), and the thromboxane A₂ mimetic U46619 (1-10 μmol/L; CN Biosciences) were added to the cells for 30 minutes before plating and were present throughout the experiment.

PGE₂ assay. PGE₂ production was measured by a competitive enzyme immunoassay using a PGE₂ detection kit (Cayman Chemical). Cells were cultured in six-well plates and treated with NS398 for 72 hours. One-milliliter aliquots of cell-conditioned media were snap frozen in liquid nitrogen before the adherent cells were detached and counted for normalization.

Western blotting. Cells were lyzed in NP40 (COX-2) or radioimmunoprecipitation assay (Rac-1) buffer. Samples containing equal protein were electrophoresed under reducing conditions in 10% (COX-2) or 15% (Rac-1) SDS-PAGE gels. Protein was electroblotted to nitrocellulose (COX-2) or polyvinylidene difluoride (Rac-1) membranes (GE Healthcare, Buckinghamshire, United Kingdom). Blots were probed with antibodies against COX-2 (Santa Cruz Biotechnology, Santa Cruz, CA) or Rac-1 (Upstate, Lake Placid, NY). Horseradish peroxidase–conjugated anti-goat or anti-mouse (DAKO and Pierce, Rockford, IL) were used as secondary antibodies. Bound antibodies were detected with the enhanced chemiluminescence Western blotting detection kit system (GE Healthcare or Pierce). Blots were probed for HSC70 (Santa Cruz Biotechnology) as a loading control.

Determination of Rac-1 activity. The Rac-1 interactive binding (CRIB) domain of the downstream Rac-1 effector, p21-activated kinase (PAK), was used to pull down active Rac-1 from cell lysates. A construct encoding glutathione S-transferase (GST) linked to the amino acid residues 57 to 141 of the PAK-CRIB domain (GST-PAK-CRIB; provided by J. Collard, King's College, London, United Kingdom) was transformed into BL21 bacteria, and production of the fusion protein induced using 0.5 mmol/L isopropyl β-d-thiogalactopyranoside (Sigma). Bacteria were resuspended [50 mmol/L Tris-HC1 (pH 7.5), 5 mmol/L MgC1₂, 100 mmol/L NaC1], sonicated, and centrifuged. The supernatant was incubated with glutathione-coupled Sepharose 4B beads (Amersham Biosciences) for 30 minutes at 4°C. Beads were washed in resuspension buffer and stored in glycerol at −80°C. Cells were plated onto LAP-coated 10-cm cell culture dishes (0.5 μg/mL) for 24 hours following targeting of β₆ using RNAi, or treated with 50 μmol/L NS398 ± PGE₂ (2-4 ng/mL) and compared with DMSO control. Cells were lyzed on ice [50 mmol/L Tris-HC1 (pH 7.5), 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 150 mmol/L NaC1, 10 mmol/L MgC1₂, 5 μg/mL leupeptin, 5 μg/mL Pefabloc, 1 mmol/L phenylmethylsulfonyl fluoride (PMSF)]. Cleared cell lysates were incubated with the GST-PAK-CRIB beads for 40 minutes at 4°C and washed thrice [50 mmol/L Tris-HC1 (pH 7.5), 1% Triton X-100, 150 mmol/L NaC1, 10 mmol/L MgC1₂, 5 μg/mL leupeptin, 5 μg pepstatin, 5 μg/mL Pefabloc, 1 mmol/L PMSF]. Samples of lysate preincubation and postincubation with the GST-PAK-CRIB beads were probed by Western blot under reducing conditions. The proportion of active Rac-1 relative to the total amount in the cells was determined by densitometry (Scion software).

Organotypic culture. Organotypic cultures with an air-tissue interface were prepared as described (26). Gels comprised a 50:50 mixture of Matrigel (Becton Dickinson, Oxford, United Kingdom) and type I collagen (Upstate) containing 5 × 10⁵/mL human foreskin fibroblasts, to which were added 5 × 10⁵ OSCC cells. For inhibition studies, NS398 (20 μmol/L) with or without PGE₂ (1-4 ng/mL) was added to the KGM. The medium was changed every 2 days. After 6 or 14 days, the gels were bisected, fixed in formal-saline, and processed to paraffin. Four-micrometer sections were immunostained with the pan-cytokeratin antibody AE1/AE3 (DAKO).

Quantitative analysis. Invasion in organotypic culture was quantified as described (26). Immunostained sections were viewed at ×100 magnification and digitally photographed. The digital image was converted to greyscale (Optilab 2.6.1, Graftek Imaging, Inc., Austin, TX), and an “invasion index” was calculated from this image. This gave a quantitative value for tumor invasion (the product of the average depth of invasion, the number of invading particles, and the area of the invading particles; ref. 26). Three 4-μm sections spaced at 150 μm were analyzed for each in vitro gel.

Transplantation of organotypic cultures onto nude mice. All animal experiments were done according to the guidelines of the local ethics committee and were as specified in the project license. Organotypic cultures were grown in vitro for 7 days before being transplanted into 10-week-old athymic nude (nu/nu) mice as described previously (26). Seven days after transplant, test animals were given 5 mg/kg NS398 i.p. every 72 hours, whereas control animals received vehicle (1:100 DMSO in PBS) only. At 6 weeks, animals were killed, and the transplants were dissected en bloc and further processed for histology. Five 4-μm sections spaced at 150 μm were analyzed for each in vivo gel. Sections where the tumor epithelium deviated significantly from the horizontal were excluded from analysis.

Statistical analysis. Data are expressed as the mean ± SD of a given number of observations. Where appropriate, one-way ANOVA was used to compare multiple groups. Comparisons between groups were by Fisher's PLSD. P < 0.05 was considered significant. Figures show representative examples of independent repeats with error bars representing SD unless stated otherwise.

α_vβ₆ and COX-2 expression in hyperplastic and dysplastic epithelium and in OSCC. We examined the expression and distribution of α_vβ₆ and COX-2 expression in hyperplastic and dysplastic epithelium and OSCC by immunochemistry. Results are summarized in Table 1. Of 39 dysplastic lesions examined, moderate (++) to strong (+++) expression of α_vβ₆ and COX-2 was present in 21% and 31%, respectively. Neither protein preferentially identified lesions, which subsequently transformed to OSCC, and similar staining sometimes was found in hyperplastic epithelium. In contrast, α_vβ₆ and COX-2 were both up-regulated in established OSCC, with moderate to strong expression in 85% and 80% of tumors, respectively. Staining for both proteins often was more intense at the periphery of invading tumor islands (Fig. 1).

COX-2 inhibition suppresses α_vβ₆-dependent cell invasion in Transwell assays. From the H357 α_v-negative OSCC line, a high α_vβ₆ expressing line (VB6) was generated using cDNA transfection and retroviral infection (20, 21). A null transfectant line (C1), expressing low levels of α_vβ₆, also was created (Fig. 2B; ref. 20). As shown previously, VB6 cells are significantly more invasive than C1 control cells, and the increased invasion is α_vβ₆ dependent (Fig. 2A; ref. 20).

The cell lines were examined for COX-2 expression by Western blotting. All three lines expressed the 72-kDa COX-2 protein at similar levels (Supplementary Fig. S1A). An additional band (∼65 kDa), corresponding in size with unglycosylated COX-2 (3), was apparent, possibly reflecting synthesis of new COX-2 protein. Inhibition of α_vβ₆ with blocking antibodies did not effect COX-2 expression in VB6 cells (data not shown).

We assessed the effects of NSAIDs on α_vβ₆-dependent OSCC Transwell invasion using SC-560 (COX-1 inhibitor), NS398 (COX-2 inhibitor), and indomethacin (nonselective COX inhibitor). Over 16 experiments, invasion of the high α_vβ₆-expressing VB6 cells was inhibited significantly using NS398 (Fig. 2A; P < 0.00001). Indomethacin produced similar results (Fig. 2A; P < 0.0001). COX-2 inhibition in VB6 cells reduced invasion to control C1 levels, comparable with inhibiting α_vβ₆ with blocking antibodies. NS398 combined with α_vβ₆ blocking antibodies did not further reduce VB6 invasion. Because VB6 invasion of Matrigel is α_vβ₆ dependent (Fig. 2A; ref. 20), COX-2 seems to promote α_vβ₆-dependent invasive activity. Inhibition of COX-1 did not affect VB6 invasion (P < 0.19). C1 control cells invaded poorly and were unaffected by these treatments (Fig. 2A). H357 cells lack α_v integrins and invade Matrigel using α₃β₁- and α₆-dependent mechanisms (data not shown). Inhibition of COX-2 (NS398 or indomethacin) had no significant effect on H357 invasion (P < 0.424 and P < 0.331, respectively; Fig. 2A). Inhibition of VB6 but not H357 cells by NS398 indicates the effects are α_vβ₆ specific.

The invasion assay was repeated using higher concentrations of NS398 (titrated at 0.5, 5, and 50 μmol/L). VB6 invasion inhibition was evident at 0.5 μmol/L (P = 0.014) with maximum inhibition at 5 μmol/L (P < 0.002; Fig. 2A). No effect on H357 or C1 invasion was seen even at 50 μmol/L NS398.

To confirm that suppression of VB6 invasion was not due to growth inhibition. Cells were grown on Matrigel gels (diluted 1:2 in α-MEM) for 72 hours ± NS398, extracted using cell dispersal solution (BD Biosciences), and counted (Casy 1 counter; Sharfe System). No inhibition of growth occurred in H357, VB6, or C1 cells (data not shown: P < 0.448, P < 0.306, and P < 0.581, respectively). Similarly, NS398 had no effect on proliferation of cells on uncoated plastic (data not shown).

Because VB6 cells have been genetically manipulated to express high levels of α_vβ₆, invasion assays were carried out using three OSCC cell lines (BICR6, CA1, and H157), which express high levels of endogenous α_vβ₆ and show α_vβ₆-dependent invasion in Transwell assays (Fig. 2B). Invasion of all three cell lines was inhibited by NS398 (Fig. 2B; BICR6, P < 0.0001; CA1, P < 0.0001; H157, P < 0.011).

NSAIDS have been reported to have antitumor effects through both COX-2-dependent and COX-2-independent mechanisms (1, 2). To test whether suppression of α_vβ₆-dependent invasion by NS398 was modulated specifically through inhibition of COX-2, the invasion assays were repeated following RNAi knockdown of COX-2 protein. Forty-eight hours after RNAi treatment, levels of COX-2 protein were reduced in H357, VB6, and C1 cells by 50%, 50%, and 69%, respectively; the (presumed) 65-kDa unglycosylated form had disappeared completely (Supplementary Fig. S1B). Suppressing COX-2 protein reduced invasiveness of VB6 cells to control C1 levels but had no effect on H357 or C1 invasion (Supplementary Fig. S1B; P < 0.003, P < 0.86, and P < 0.89, respectively).

PGE₂ restores α_vβ₆-dependent invasion. To determine whether COX-2 modulated α_vβ₆-dependent invasion through a metabolic product of COX, invasion assays were repeated, but adding NS398 with PGE₂, PGF_2α, or U46619 (a thromboxane A₂ mimetic). PGE₂ at 2 to 4 ng/mL completely abrogated the anti-invasive activity of NS398 (Fig. 2C; P < 0.0017), whereas PGF_2α and U46610 had no effect (Fig. 2C; P < 0.444 and P < 0.276, respectively). PGE₂ also seemed to produce some restoration of invasion in the presence of α_vβ₆ blocking antibodies, although this did not reach significance (P = 0.09; data not shown).

Interestingly, the dose response of VB6 cells to PGE₂ produced a bell-shaped curve, with restoration of invasion between a narrow spectrum of concentrations (2-4 ng/mL) but not at higher concentrations (Fig. 2C). Indeed, in the absence of NS398, PGE₂ at concentrations of ≥10 ng/mL inhibited invasion of all three cell lines (data not shown). Similar biphasic response curves have been described previously, and it is suggested that this may result from prostaglandins interacting nonspecifically with other classes of prostanoid receptors (27).

PGE₂ levels were measured by ELISA (Cayman Chemicals); the cell lines produced similar levels of PGE₂. NS398 treatment or RNAi knockdown of COX-2 inhibited prostaglandin production by >95%, suggesting that synthesis of the prostaglandin was mediated largely through COX-2 rather than COX-1 (Fig. 2C).

COX-2 inhibition suppresses α_vβ₆-dependent invasion in organotypic culture. NS398 consistently inhibited the α_vβ₆-dependent invasion of VB6 cells in Transwell assays. Such assays, however, oversimplify the invasive process because stromal cells interact dynamically with tumor cells in vivo to promote invasion (28). We, therefore, used organotypic cultures containing fibroblasts to measure invasion (26).

We transfected VB6 cells with β₆ RNAi and measured invasion. Because RNAi effects are transient, organotypic cultures were harvested after 6 days (7 days after transfection). Sections were analyzed to generate an invasion index, as described (26). Seven days after transfection, RNAi knockdown of β₆ was ∼50% (3 days, 91%; 5 days, 54%; data not shown), and VB6 invasion was inhibited by >95% (Fig. 3A and D). As in Transwell assays, the high α_vβ₆-expressing VB6 cells were significantly more invasive than C1 control cells (P < 0.0002) but invaded at similar levels to the invasive, α_v-negative H357 cells (Fig. 3B-D; P < 0.07). NS398 treatment reduced VB6 invasion significantly by ∼90% over three experiments (Fig. 3B and D; P < 0.0001). Similar to Transwell assays, addition of 2 ng/mL PGE₂ completely abolished NS398-induced inhibition of invasion (Fig. 3B and D). C1 cells, expressing low levels of α_vβ₆, invade organotypic cultures poorly (Fig. 3B and D), but this limited invasion was still inhibited by NS398 (Fig. 3D). NS398 had no effect on the invasion of α_v-negative H357 cells (Fig. 3C and D; P < 0.572).

COX-2 inhibition interferes with α_vβ₆ post-receptor events. COX-2 inhibition might modulate α_vβ₆ function by reducing cell surface expression, decreasing ligand binding affinity, and/or interfering with α_vβ₆ signaling. Flow cytometry revealed that there was no effect on surface levels of α_vβ₆ in VB6 cells following a 72-hour incubation with NS398 (Supplementary Fig. S2A). We next examined the effect of COX-2 inhibition on α_vβ₆ ligand binding. We have shown previously that adhesion to the LAP in the cell lines is modulated solely through α_vβ₆: VB6 cells adhere well; C1 cells adhere poorly; and α_vβ₆-negative H357 cells do not adhere to this substrate (29). Inhibition of α_vβ₆ with 10D5 antibody inhibited adhesion completely (Supplementary Fig. S2B), whereas SC-560, NS398, indomethacin, or inhibitory antibodies against β₁ integrin had no effect (P < 1.00, P < 0.60, P < 0.65, and P < 1.00, respectively). These data suggest that COX-2 does not modulate α_vβ₆ expression or activity.

α_vβ₆ ligand binding activates Rac-1. The role of small GTPase Rac-1 in regulating cell migration and cell spreading following integrin-binding are well described (30–32). In particular, several lines of evidence indicate that Rac-1 may play a critical role in a number of aspects of tumor development (31–34), and Rac-1 has also been shown to modulate tumor cell invasion (32–34). Therefore, we examined the effect of α_vβ₆ ligand binding on Rac-1 activation in VB6 cells using a pulldown assay. Compared with cells in suspension, binding of VB6 cells to LAP resulted in Rac-1 activation (Fig. 4A,, i, lanes 3 and 4). Rac-1 activation increased over time and was proportionally greater after 24 hours than 4 hours (Fig. 4A,, i, lane 6). This effect was α_vβ₆ dependent: C1 control cells on LAP showed reduced Rac-1 activation compared with VB6 cells (Fig. 4A,, ii, quantified in Fig. 4A,, v; 61% reduction; P ≤ 0.0001). Rac-1 activation in VB6 cells was inhibited by 69% following β₆ knockdown with RNAi (Fig. 4A,, iii, quantified in Fig. 4A , v; P ≤ 0.0001).

COX-2 inhibition prevents α_vβ₆-dependent Rac-1 activation. We next examined whether COX-2 activity was required for α_vβ₆-dependent Rac-1 activation. Rac-1 activation in VB6 cells was reduced in the presence of NS398 by 72% (Fig. 4A,, iv, quantified in Fig. 5A,, v; P = 0.0057). This effect was eliminated if NS398 was added combined with PGE₂ (2-4 ng/mL; Fig. 5A,, iv, quantified in Fig. 4A , v). Rac-1 activation in VB6 cells treated with NS398 and PGE₂ (4 ng/mL) did not differ significantly from control cells (P = 0.146). This confirmed that PGE₂ was required for α_vβ₆-dependent Rac-1 activation.

Constitutively active Rac-1 restores α_vβ₆-dependent invasion in the presence of NS398. To confirm that the inhibition of Rac-1 activation by NS398 was responsible for the inhibition of VB6 invasion, we expressed constitutively active Rac-1 in VB6 cells using transient transfection experiments. Over four experiments, cells transfected with V12-Rac-1 in the presence of NS398 had their invasive activity restored to control levels (Fig. 4B; no inhibition of invasion; P < 0.88) but not when transfected with empty vector controls (Fig. 4B; invasion inhibited; P < 0.0001). Additionally, in the presence of NS398, those VB6 cells that had invaded were found to be enriched for V12 Rac-1-GFP (by 42%; data not shown).

Rac-1 knockdown suppresses α_vβ₆-dependent invasion. Additional experiments using RNAi to inhibit Rac-1 confirmed that α_vβ₆-dependent invasion was mediated through Rac-1. RNAi treatment of VB6 cells reduced Rac-1 protein expression after 48 hours by 96% (Fig. 4C; 3 days, 93%; 5 days, 82%; 7 days, 78%; data not shown). Over five experiments, Rac-1 knockdown inhibited VB6 Transwell invasion by ∼50% (Fig. 4C; P < 0.0001). Rac-1 knockdown also inhibited significantly invasion of VB6 cells in organotypic culture (Fig. 4D; P < 0.01), an effect similar in magnitude to direct inhibition of α_vβ₆ or treatment with NS398 (Fig. 3A and B, respectively).

COX-2 inhibition suppresses α_vβ₆-dependent invasion in vivo. Organotypic cultures can be transplanted onto the back muscle fascia of nude mice where they grow and invade into the host tissue (26). Using this method, VB6 cells produced significantly more invasion than C1 control cells (Fig. 5A and C). Transplants were established for 7 days before commencing treatment, when animals received NS398 (5 mg/kg i.p.) or DMSO every 72 hours for 5 weeks. Tumor take was 100%.

NS398 inhibited the invasion of VB6 cells in vivo by ∼70% (P < 0.009; Fig. 5A and C), suppression similar to that seen in Transwell assays and in vitro organotypic culture (Figs. 2 and 3B). There was no significant effect of NS398 treatment on the α_v-negative H357 cells (P < 0.933; Fig. 5B and C). Because C1 cells invade poorly in vivo, the effect of NS398 on this line was not tested.

Expression of α_vβ₆ occurs in many carcinoma types (6–10, 35–37) but is found most consistently and at its highest levels in OSCC (7, 37), where experimental studies indicate that it promotes invasion (20, 21, 38, 39). Similarly, high COX-2 expression is also found in OSCC, where it seems to be invasion promoting (22, 40, 41). Our results suggest a novel link among α_vβ₆, COX-2, and Rac-1 by showing that the selective COX-2 inhibitor NS398 suppresses α_vβ₆-dependent, Rac-1-mediated invasion.

Elevated levels of both α_vβ₆ and COX-2 have (separately) been reported in oral dysplasia (12, 13, 40), where it is suggested that expression may identify lesions at risk of malignant transformation. We examined 39 dysplasias from patients with a minimum of 5 years of clinical follow-up. We found that neither α_vβ₆ nor COX-2 expression identified high-risk epithelial dysplasias. Thus, 20% to 30% of dysplasias were positive for one or both proteins. There were positive nontransforming lesions and, conversely, negative lesions that subsequently transformed. Additionally, focal basal expression of α_vβ₆ and COX-2 was present in several hyperplastic lesions. Although, generally, there was increased expression of α_vβ₆ and COX-2 in dysplasias compared with hyperplastic epithelium, this was much lower than expression in OSCC; most OSCC showed strong expression of both proteins with expression particularly prominent at the periphery of invading tumor islands. These data suggest that the invasion-promoting effects of α_vβ₆ and COX-2 primarily may occur in established malignancy. It is possible, however, that negative dysplasias subsequently may express the proteins at a later disease stage, before transformation.

VB6 cells, expressing high levels of α_vβ₆, are significantly more invasive than C1 control cells (this work and refs. 20, 21). Interestingly, α_v-negative H357 cells invade as well as VB6 by using α₃β₁ integrin– and α₆ integrin–dependent mechanisms.⁵

As well as having direct effects on invasion (4, 41–43), NSAIDs may inhibit tumor proliferation and promote apoptosis (2). However, NS398 did not affect growth of our cells at the concentrations used in this study, suggesting that effects of NS398 seem to be via a direct inhibition of α_vβ₆ activity.

Although NSAIDs have both COX-2-dependent and COX-2-independent antitumor effects (1), we show that inhibition of VB6 invasion was COX-2 activity dependent; α_vβ₆-dependent invasion was suppressed using COX-2 RNAi and restored, in the presence of NS398, by exogenous PGE₂. Recently, we developed a quantitative invasion assay using organotypic cultures (26). High α_vβ₆-expressing VB6 cells are significantly more invasive than C1 control cells in such cultures, and invasion is α_vβ₆ dependent. NS398 markedly inhibited VB6 cell invasion by >90%. As with Transwell assays, addition of exogenous PGE₂ (at 2 ng/mL) restored VB6 invasion in the presence of the COX-2 inhibitor. However, NS398 had no effect on H357 invasion.

Although COX-2 and α_vβ₆ expression correlate in OSCC in vivo, we found no direct evidence that either regulates expression of the other. Antibody blockade of α_vβ₆ did not affect levels of COX-2 protein in VB6 cells, nor did COX-2 inhibition affect α_vβ₆ levels or ligand-binding affinity. Rather, NS398 affected α_vβ₆ post-receptor events.

Based on previous observations (18), we examined the role of Rac-1 in the COX-2/α_vβ₆–dependent invasion. The small GTPase Rac-1 regulates cell migration and spreading following integrin binding (30) and may also modulate tumor cell invasion (31, 32). Unlike Ras, however, mutation of Rac-1 has not been found in human cancer: activation probably occurring through aberrant, or overactive, signaling pathways (33, 34). Rac-1 was activated by α_vβ₆ ligand binding in VB6 cells, and this activation was not transient but was maintained, indeed increased, over a 24-hour time period. It is possible that pathologic up-regulation of α_vβ₆ results in overactive Rac-1 signaling, leading to prolonged Rac-1 activation. α_vβ₆-dependent Rac-1 activation was COX-2 dependent because it was inhibited by NS398 and restored on addition of PGE₂. Functional studies confirmed that α_vβ₆-dependent invasion was mediated through Rac-1. Thus, transfection with constitutively active Rac-1 restored VB6 invasion in the presence of NS398. Additionally, RNAi-mediated Rac-1 knockdown inhibited VB6 invasion to levels comparable with NS398 treatment or α_vβ₆ blockade, both in Transwells and in organotypic cultures.

In summary, our data suggest that neither α_vβ₆ nor COX-2 expression predicts malignant transformation of oral dysplasia. However, there is strong coexpression of both proteins in established OSCC, in accord with our proposal that these molecules might interact in promoting invasion. We show, for the first time, that α_vβ₆-dependent invasion in vitro and in vivo by OSCC cells requires COX-2-dependent activation of Rac-1, via up-regulation of PGE₂; abrogating COX-2, thus, blocks α_vβ₆-dependent invasion.

Worldwide, OSCC represents about 5.5% of all malignancies (44). Most patients are treated by surgery, and >50% of patients die within 5 years (44). New types of therapy obviously are required, and both COX-2 and α_vβ₆ have been suggested as possible therapeutic targets. Our data suggest that NSAIDs may inhibit the pro-invasive effect of both proteins. The significant morbidity and mortality rates of OSCC are such that the therapeutic benefit of these drugs should be considered, particularly in α_vβ₆-positive tumors.

Grant support: Cancer Research UK, The Health Foundation, U.K. Medical Research Council, DebRA, AICR, and University of London central research fund.

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.