Cyclooxygenase-2-Derived E Prostaglandins Down-Regulate Matrix Metalloproteinase-1 Expression in Fibroblast-Like Synoviocytes via Inhibition of Extracellular Signal-Regulated Kinase Activation ¹

By virtue of their increased numbers in pannus, their close association with cartilage, and their ability to produce and export matrix metalloproteinases (MMPs) such as MMP-1, -3, -9, and -13, fibroblast-like synoviocytes (FLSCs) have been implicated in the process of marginal cartilage erosion and joint destruction in rheumatoid arthritis (RA) (1, 2, 3, 4, 5, 6, 7, 8). Accordingly, the intracellular pathways through which FLSC activities are regulated have become a subject of increasing interest. Signaling and/or regulatory molecules implicated in the proliferative and/or degradative activities of FLSCs include the mitogen-activated protein kinases (MAPKs) c-Jun N-terminal kinase (JNK) and p38 (9, 10), the tumor suppressor p53 (11, 12), the proinflammatory transcription factor NF-κB (13), and possibly telomerase (14). However, the precise role of each of these in inflammatory joint destruction remains to be fully elucidated.

In contrast with JNK and p38, which are activated in response to inflammatory and stress signals, the MAPK extracellular signal-regulated kinase (ERK) has been most thoroughly studied in the context of growth factor stimulation and is primarily appreciated as playing roles in cell proliferation, antiapoptosis, and differentiation (15, 16, 17, 18, 19, 20, 21). However, ERK may also be activated in response to proinflammatory stimuli (22, 23). For example, Vilcek and colleagues (24, 25, 26) have demonstrated that in FS-4 fibroblasts, ERK undergoes activation in response to the cytokines IL-1β and TNF-α, both of which are critical to the pathogenesis of RA. The ability of fibroblast ERK to respond to inflammatory cytokines, as well as to signal for cell proliferation, suggests that the ERK signaling pathway might be an important regulator of FLSC functions.

Nonsteroidal anti-inflammatory drugs (NSAIDs) constitute a diverse family of agents that share an ability to inhibit cyclooxygenase (COX) activity and PG production (27, 28, 29). Nonetheless, COX inhibition may not entirely account for the actions of some NSAIDs. For instance, high concentrations of aspirin (ASA) and other salicylates have recently been shown to inhibit inducible NO synthase expression and activity (30), to inhibit NF-κB activation (31), to stimulate the activation of p38 (32), and to stimulate apoptosis, probably via p38 activation (33). These effects were COX-independent and not shared by nonsalicylate NSAIDs such as indomethacin. Our laboratory (34) and Vilcek and colleagues (25) have observed the capacity of high doses of ASA and sodium salicylate (NaS) to inhibit ERK in neutrophils (34) and FS-4 fibroblasts (25), respectively. In our studies, the effects of salicylates on ERK were COX independent and directly tied to their ability to inhibit neutrophil inflammatory responses such as stimulated adhesion.

Based on these prior observations, we were interested to know, first, whether the ERK pathway plays a prominent role in FLSC signaling and, second, whether salicylates or other NSAIDs might be effective in modifying such signaling. Our data suggest that ERK activation in FLSCs selectively mediates the extracellular expression of some, but not all, MMPs. Salicylates, via effects on ERK, may play an important, potentially disease-modifying role in the down-regulation of MMPs. In contrast, we were surprised to observe that COX inhibition enhances the expression of MMP-1 by FLSCs, via effects on PGE-mediated inhibition of ERK-dependent MMP-1 production.

Unless otherwise noted, all reagents were from Sigma-Aldrich (St. Louis, MO). Cultures of transformed rabbit FLSCs (HIG-82) (35) were from American Type Culture Collection (Manassas, VA; catalogue no. CRL-1832). Anti-phosphoERK (catalog no. sc-7383), anti-phosphoJNK (catalog no. sc-6254), anti-phosphotyrosine (catalog no. sc-508), anti-ERK-1 (catalog no. sc-93), anti- ERK2 (catalog no. sc-154), and HRP-conjugated donkey anti-mouse (catalog no. sc-2004) Abs and antisera were from Santa Cruz Biotechnology (Santa Cruz, CA). In some experiments, anti-phosphoERK antiserum was from Promega (Madison, WI; catalog no. V8031); SB203580 was also from Promega. UO126 was from Upstate Biotechnology (Lake Placid, NY). PD98059 was from Biomol (Plymouth Meeting, PA). Anti-MMP-1 (catalog no. AB806), anti-MMP-3 (catalogue no. AB812), and anti-MMP-9 (catalog no. AB805) antisera were from Chemicon International (Temecula, CA). Anti-MMP-13 antiserum (catalog no. AF511) and PGE₁ immunoassay kit were from R&D Systems (Minneapolis, MN). Precast 10% Tris-glycine/polyacrylamide gels were from Novex/Invitrogen (San Diego, CA). Nitrocellulose paper was from Schleicher & Schuell (Keene, NH). [¹²⁵I]Protein A and ECL detection kit were from Amersham (Arlington Heights, IL). Centricon centrifugal filter devices were from Millipore (Bedford, MA). Ham’s F-12 medium and FBS were from BioWhittaker (Walkersville, MD). Penicillin G sodium (10,000 U/ml)/streptomycin sulfate (10,000 μg/ml) in 0.85% NaCl was from Life Technologies (Rockville, MD). NS398 was from CalBiochem (La Jolla, CA). The selective COX-2 inhibitor SC299 was the kind gift of K. Seibert and J.J. Talley (Searle Research and Development, Skokie, IL). 6-Methoxy-2-naphthylacetic acid (6MNA) was the kind gift of Darren Rimmer and Peter G. Treagust (GlaxoSmithKline, Essex, U.K.).

Except where specified, all experiments were performed on rabbit FLSCs (35). FLSCs were grown in Ham’s F-12 medium containing 10% FBS and 1% penicillin/streptomycin, as recommended by the supplier. For experimental purposes, FLSCs were seeded into six-well plates, grown to near confluence, and then serum starved (Ham’s F-12 without FBS or penicillin/streptomycin) for 24–48 h. Protein assays of cell lysates, performed after serum starvation, confirmed that cell populations before experimentation differed by no more than 10% between wells.

Human rheumatoid FLSCs were prepared by a modification of the method of Alvaro-Gracia et al. (36). Synovium obtained from RA patients at the time of joint replacement was minced with a sterile scalpel in 90-mm plates and was treated with collagenase (1 mg/ml) in DMEM for three cycles each of 30 min at room temperature. After each collagenase exposure, the plates were gently agitated and the supernatant, containing free FLSCs, was aspirated and set aside. The total supernatant from the three incubations was pooled, washed three times in DMEM, and cultured in DMEM containing 10% FCS as well as 1% penicillin/streptomycin and l-glutamine. Cells were grown at 37°C and used in experiments between passages three and eight.

After serum starvation of FLSCs, supernatants were removed and replaced with fresh Ham’s F-12 medium for 30–60 min at 37°C/5% CO₂ before experimental treatments as described in Results and the figure legends. Except where otherwise specified, cells were exposed to inhibitors for 30 min and to stimuli for 30 min. Subsequent to experimental treatment, cells were incubated with ice-cold lysis buffer (20 mM Tris (pH 7.4), 1 mM EGTA, 2 mM sodium vanadate, 25 mM sodium fluoride, 0.5% (v/v) Triton X-100, 2 mM PMSF, 10⁵ KU/ml aprotinin, and 10 μg/ml each of chymostatin, antipain and pepstatin) for 20 min at 4°C. The adherent remnants were liberated with a rubber policeman, and the lysates were collected and supplemented with Laemmli sample buffer (37) containing 5% 2-ME. Samples were heated (100°C for 3–5 min) and analyzed by SDS-PAGE on 10% Tris-glycine gels, followed by electrophoretic transfer to nitrocellulose. Nitrocellulose papers were imaged with Ponceau S solution to confirm even loading of the gel lanes, washed three times with TBS (pH 7.4) containing 0.05% Tween 20, blocked with 3% BSA in TBST for 1 h, washed three times again, and immunoblotted with anti-phosphoERK antiserum in TBST at a dilution of 1/200 for 1 h. The blots were again washed three times with TBST, followed by 1-h incubation with [¹²⁵I]protein A and further washing and drying. The blots were imaged and quantitated using a PhosphorImager (Storm imaging system; Molecular Dynamics, Sunnyvale, CA). A similar protocol was used for determining JNK activation using an anti-phosphoJNK Ab; the image was analyzed using a secondary HRP-conjugated Ab, chemiluminescence according to the manufacturer’s instructions, and autoradiography. Duplicate samples from each experiment were treated similarly but analyzed on immunoblot using Abs that recognize both phosphorylated and unphosphorylated ERK 1 and 2 (each Ab at 1/200) or JNK to confirm that changes in levels of the phosphoenzymes were not due to changes in total enzyme expression.

FLSC cultures grown to near confluence in six-well plates were serum starved (0% FBS, without penicillin/streptomycin) for 24–48 h. The supernatants were removed and replaced with fresh Ham’s F-12 and, after allowing 30–60 min for equilibration, the cultures were treated and stimulated as described in Results and the figure legends. The supernatants were recovered and concentrated by centrifugation in Centricon centrifugal filter devices (m.w. cutoff 30,000) for 35 min at 5000 rpm at 4°C. Aliquots of the concentrated supernatants were supplemented with Laemmli buffer and analyzed for the presence of MMP-1, -3, -9, and -13 by SDS-PAGE (10% Tris-glycine gels), transfer to nitrocellulose, immunoblotting with the appropriate anti-MMP Ab (1/200), and visualization via [¹²⁵I]protein A and phospor imaging.

PGE₁ in the supernatant was measured by ELISA, according to the manufacturer’s directions.

FLSCs grown to near confluence were incubated overnight ±10 mM ASA or NaS, lysed, and centrifuged as described for Western blotting (see Measurement of ERK activation). The aqueous phase of sample supernatants was transferred into fresh sample tubes for HPLC analysis using a Spherisorb S5 SAX column (Waters, Milford, MA) with a KCl gradient of 0.007–0.25 M formed over 80 min. ATP was identified by its characteristic retention time and UV absorbance spectrum. Quantitation was achieved by comparison to known standards using Millennium32 software (v3.05.01; Waters). 2-Chloroadenosine, which has a different retention time and absorbance, was used as an internal control and was added to the samples at the time of sample preparation.

Immunoblots imaged by phosphor imaging and/or autoradiography were quantitated using ImageQuant (Molecular Dynamics). Because the units of such quantitation are arbitrary and vary in magnitude between experiments, either the unstimulated or stimulated condition, as appropriate, was normalized to 100% for purposes of evaluation. The remaining conditions were then expressed relative to the normalized control values. Unless otherwise stated, all experiments were done three times in duplicate. Significant differences between the control and variable conditions in each experiment were determined by unpaired or paired Student’s t test, as appropriate, using Microsoft Excel X (Redmond, WA) for Macintosh (Apple Computer, Cupertino, CA).

Stimulation of rabbit FLSCs with IL-1β (20 ng/ml) and/or TNF-α (20 ng/ml) for 30 min resulted in increased ERK activation relative to unstimulated controls. Consistent with a prior report (38), IL-1β was a more effective stimulus, and TNF-α stimulation of ERK did not independently reach statistical significance (IL-1β stimulation of ERK 2, 150 ± 16% control, p = 0.027; TNF-α stimulation of ERK 2, 128 ± 22% control, p = 0.146) (Fig. 1, A and C). Stimulation of FLSCs simultaneously with both agents resulted in increased ERK phosphorylation relative to either alone (200 ± 33% control; p = 0.028). Both ERK 1 (p44) and ERK 2 (p42) underwent activation. Similar results were observed with human FLSCs obtained from the knees of patients undergoing joint replacement for RA (Fig. 1,B). To confirm that increases in the phosphorylated form of ERK were due to activation and not to changes in total expression of ERK protein, lysates from these experiments were analyzed for total ERK expression. Stimulation with IL-1β, TNF-α, or both did not affect total amounts of ERK (data not shown). ERK 2 was consistently more abundant than ERK 1; in some experiments, levels of activated ERK 1 were so limited as to preclude quantitative analysis. Consistent with a previous report on IL-1β-stimulated human RA cells (9), IL-1β/TNF-α-stimulated activation of ERK in rabbit FLSCs was rapid and transient, peaking at 15–30 min (Fig. 1 D; at t = 15 min, ERK 2 activity = 146% control, p = 0.029; at t = 30 min, ERK activity = 135% control, p = 0.018). Because IL-1β and TNF-α together induced more ERK stimulation than either agent alone and because in RA joints FLSCs are exposed to both of these cytokines simultaneously, subsequent experiments were conducted using IL-1β and TNF-α as costimuli.

PD98059 (39) and UO126 (40) are specific inhibitors of the proximal ERK activator MEK. As shown in Fig. 2 A, incubation of FLSCs with PD98059 (100 μM) before stimulation with IL-1β and TNF-α resulted in 56 ± 7% (p = 0.002) inhibition of maximally stimulated ERK activation levels. Consistent with previous reports in other cell types (40), UO126 (10 μM) was a more effective inhibitor of ERK, producing 80 ± 14% (p = 0.005) inhibition. In separate experiments, PD98059 and UO126 also inhibited baseline levels of ERK activity in the absence of stimulation, indicating that, even under conditions of serum starvation, these rabbit FLSCs are not completely quiescent vis-a-vis ERK activity (data not shown). At concentrations ≤10 μM, SB203580 is a specific inhibitor of another MAPK, p38 (41). In contrast with PD98059 and UO126, SB203580 (10 μM) failed to inhibit ERK activation stimulated by IL-1β/TNF-α (17 ± 8% enhancement of IL-1β/TNF-α stimulation, p = 0.063).

FLSC production of MMPs, including MMP-1 and MMP-13, has been implicated in cartilage destruction in rheumatoid and other inflammatory forms of arthritis (2, 42). As shown in Fig. 2 B, FLSCs constitutively expressed both of these MMPs, and stimulation of FLSCs with IL-1β/TNF-α for 18 h resulted in up-regulation of extracellular levels of both MMP-1 and -13 (341 ± 45% (p = 0.0001) and 1182 ± 642% (p = 0.084) control, respectively). Incubation of FLSCs with PD98059 (100 μM) or UO126 (10 μM) but not SB203580 (10 μM) inhibited the extracellular expression of MMP-1 in response to IL-1β/TNF-α. As in the case of ERK, UO126 (55 ± 17% inhibition, p = 0.039) was a more potent inhibitor of MMP-1 release than was PD98059 (25 ± 12% inhibition, p = 0.023). PD98059 also inhibited MMP-1 release from human RA FLSCs stimulated with IL-1β/TNF-α (40 ± 6% inhibition, p = 0.01, n = 3). PD98059 and UO126, but not SB203580, also inhibited MMP-3 and -9 release from FLSCs (data not shown). In contrast, PD98059 and UO126 had little or no effect on extracellular MMP-13 levels (18 ± 4% (p = 0.004) and 16 ± 9% (p = 0.112) inhibition, respectively). SB203580 also failed to inhibit MMP-13. These data suggest that the release of various MMPs is differentially regulated and that extracellular expression of MMP-1, -3, and -9 in response to IL-1β/TNF-α is regulated by ERK, but not p38. In contrast, MMP-13 regulation by IL-1β/TNF-α appears to be relatively independent of both ERK and p38.

To further investigate the role of ERK in regulating MMP-1, we tested the effects of epidermal growth factor (EGF), a canonical activator of ERK, on FLSCs. Stimulation of FLSCs with EGF resulted in ERK activation and preincubation with UO126, but not SB203580, inhibited ERK activation (Fig. 2,C, top). Stimulation of FLSCs with EGF also resulted in increased MMP-1 release, which was inhibited by UO126, but not SB203580 (Fig. 2,C, middle). These data provide additional evidence for a role for ERK, but not p38, in the regulation of MMP-1 release from cells. In contrast, EGF stimulation of MMP-13 release was not inhibited by either UO126 or SB203580 (Fig. 2 C, bottom).

We have previously demonstrated that millimolar (clinically anti-inflammatory) but not micromolar (analgesic, COX-inhibiting) concentrations of ASA inhibit ERK activation in human neutrophils (34). As shown in Fig. 3 A, 10 mM ASA also inhibited IL-1β/TNF-α stimulation of ERK 1 (24 ± 6% inhibition, p = 0.020) and ERK 2 (35 ± 10% inhibition, p = 0.005) in FLSCs. To determine whether the ability of ASA to inhibit ERK was shared by other NSAIDs (i.e., whether it was COX-dependent), we tested the effects of nonselective COX inhibitors on ERK activation by IL-1β plus TNF-α. Exposure of FLSCs to indomethacin for 30 min at concentrations sufficient to inhibit both COX-1 and -2 (10 μM) had little or no effect on ERK activation. Similarly, 30-min exposure to levels of ibuprofen (100 μM) or 6MNA (50 μM) sufficient to inhibit both COX-1 and -2 (43) had no significant effect on ERK. Thus, stimulation of ERK via IL-1β/TNF-α depends upon neither the rapid activation of COX nor the stimulated production of PGs.

The ability of 10 mM ASA to inhibit ERK might have been due to general toxic salicylate effects. However, overnight exposures to ASA or NaS did not result in significant increases in cell death as measured by lactate dehydrogenase release (Fig. 4,A). Ten millimolar ASA or NaS also did not significantly impair the ability of FLSCs to exclude trypan blue (Fig. 4,A), suggesting both that the cells were viable and that their metabolic functioning remained intact. As shown in Fig. 4,B, neither ASA nor NaS resulted in depletion of cellular ATP levels (p = 0.727 for control vs ASA; p = 0.58 for control v NaS). Moreover, neither ASA nor NaS markedly altered the overall tyrosine phoshporylation of cellular proteins (Fig. 4,C). Finally, neither ASA nor NaS inhibited the phosphoactivation of another MAPK, JNK (Fig. 4 D).

Because millimolar concentrations of ASA inhibited ERK, we reasoned that they should also inhibit extracellular MMP-1 expression from FLSCs. As shown in Fig. 3,B, pre-exposure of FLSCs to 10 mM ASA inhibited IL-1β/TNF-α stimulation of MMP-1 release (29 ± 9% inhibition, p = 0.029). To confirm that this effect of ASA was not due to inhibition of COX, we tested the effects of indomethacin, ibuprofen, and 6MNA. Strikingly, these agents did not merely fail to inhibit MMP-1 release, but they enhanced the release of MMP-1 from these stimulated cells (indomethacin, 140 ± 18% (p = 0.088); ibuprofen, 147 ± 14% (p = 0.045); 6MNA, 191 ± 23% (p = 0.014); above IL-1β/TNF-α-stimulated levels). Thus, millimolar salicylates inhibit, but COX inhibitors enhance, MMP-1 release. To examine this observation in more detail, we performed dose-response experiments comparing ASA (potent COX inhibitor, IC₅₀ ≈ 1.7 μM for COX-1, 7.5 μM for COX-2 (43)) with NaS (weak COX inhibitor, IC₅₀ ≈ 5 mM for COX-1, 500 mM for COX-2 (43)). As shown in Fig. 5,A, micromolar concentrations of ASA had no effect on ERK, but inhibited COX activity measured by extracellular PGE₁ expression and resulted in increased MMP-1 release from IL-1β/TNF-α-stimulated cells (32 ± 15% increase in MMP-1 above IL-1β/TNF-α at 10 μM ASA concentrations (p = 0.048)). In contrast, micromolar concentrations of NaS did not affect either COX or MMP-1 (Fig. 5,B). At millimolar concentrations, NaS reproduced ASA’s ability to inhibit both ERK and MMP-1 release. The IC₅₀ for both ASA and NaS, for both ERK 1/2 and MMP-1 inhibition, were ≥1 mM. We also tested the effects of micromolar and millimolar ASA on ERK activation and MMP-1 expression in response to EGF. As shown in Fig. 5,C (top), stimulation of ERK by EGF was unaffected by 10 μM ASA, but was inhibited by 10 mM ASA. Consistent with the effects seen in IL-1β/TNF-α-stimulated cells, 10 μM ASA enhanced but 10 mM ASA inhibited EGF-stimulated MMP-1 expression (Fig. 5 C, bottom). Thus, both acetylated and nonacetylated salicylates, at concentrations sufficient to inhibit ERK, inhibit MMP-1 production. However, ASA differs from NaS in that at lower concentrations that do not inhibit ERK, ASA inhibits COX and enhances MMP-1 production.

To determine whether the enhancement of MMP-1 release by nonselective COX inhibitors was COX-2-related, we tested the effects of three selective COX-2 inhibitors (NS398 (0.5 μM), celecoxib (0.5 μM), and SC299 (0.5 μM), all at concentrations selected to ensure COX-2 selectivity (43, 44, 45)) on both ERK activation and MMP-1 production. Brief (30-min) exposure to any of the COX-2 inhibitors had little or no effect on IL-1β/TNF-α stimulation of ERK (Fig. 6,A). However, as in the case of nonselective COX inhibitors, exposure of FLSCs to COX-2 inhibitors before overnight stimulation resulted in enhancement of MMP-1 release (NS398, 73 ± 27% increase (p = 0.027); for celecoxib, 96 ± 27% increase (p = 0.034); for SC299, 82 ± 20% increase (p = 0.029)) (Fig. 6 B). Immunoblotting confirmed that COX-2 was constitutively present in these cells and underwent up-regulation in response to IL-1β and/or TNF-α (data not shown). These data suggest that COX-2 activity in particular down-regulates extracellular MMP-1 expression, presumably through the expression of a common set of biologically active PGs.

To confirm that COX-2 down-regulates MMP-1 through production of PGs, we tested the effects of E- and F-series PGs on extracellular MMP-1 expression. As shown in Fig. 7,A, incubation of FLSCs with either PGE₁ (1 μM) or PGE₂ (1 μM) before stimulation with IL-1β/TNF-α resulted in significant inhibition of MMP-1 release. Consistent with prior reports in FLSCs stimulated with phorbol myristate acetate (46, 47), PGE₁ appeared to be a more efficient inhibitor of MMP-1 expression (PGE₁, 38 ± 4% inhibition (p = 0.0002); PGE₂, 29 ± 6% inhibition (p = 0.035)). Higher concentrations of PGE₁ (10 μM) were more efficacious at inhibiting stimulation of extracellular MMP-1 expression (60% inhibition; data not shown). In contrast, neither 1 μM PGF_2α nor 10 μM PGF_2α (data not shown) inhibited stimulated MMP-1 expression (1 μM PGF_2α, 3 ± 12% inhibition, p = 0.326). Thus, the effect of PGEs on MMP-1 release appeared to be class specific and not a general lipid or PG effect. To further relate the effects of PGs to the effects of COX inhibition, we tested the capacity of PGE₁ to reverse the effects of the selective COX-2 inhibitor NS398 on extracellular MMP-1 expression. As shown in Fig. 7 B, incubation of FLSCs in the simultaneous presence of PGE₁ and NS398 abrogated the ability of NS398 to enhance MMP-1 production, suggesting that enhancement of extracellular MMP-1 expression by COX inhibitors occurs through the inhibition of PGE production.

Although brief exposure to COX inhibitors had no direct effect on IL-1β/TNF-α-stimulated ERK activation, we and others have previously demonstrated that E-series PGs inhibit ERK activation in a number of other cell types. Accordingly, we tested the effects of PGE₁, PGE₂, and PGF_2α (each 1 μM) on IL-1β/TNF-α stimulation of ERK. As shown in Fig. 7,C, PGE₁ and PGE₂, but not PGF_2α, inhibited cytokine-stimulated ERK activation in FLSCs. As in the case of MMP-1, PGE₁ was a more potent inhibitor of ERK than was PGE₂ (PGE₁, 47 ± 12% inhibition (p = 0.015); PGE₂, 29 ± 6% inhibition (p = 0.008). Thus, whereas COXs do not participate in the direct stimulation of ERK activation by cytokines, the presence of specific PG products of COXs may serve to down-regulate both ERK activation and the extracellular expression of MMP-1. Consistent with these observations, both PD98059 and PGE₁ inhibited ERK activation in human RA FLSCs (Fig. 7 D).

The ability of PGEs to down-regulate extracellular MMP-1 expression, together with the constitutive presence COX-2 in FLSCs, suggested that even in cells that are not exogenously stimulated, PGE production might contribute to autacoid control of baseline MMP-1 expression levels. To test this hypothesis, FLSCs were incubated in the presence of various selective and nonselective COX inhibitors overnight in the absence of cytokine stimulation, and the supernatants were assayed for the presence of MMP-1. As shown in Fig. 8 A, incubation of FLSCs with 10 μM NaS (weak COX inhibitor) had no effect on baseline MMP-1 levels in the extracellular fluid (111 ± 7% control, p = 0.087). However, 100 μM ibuprofen and 0.5 μM SC299 each caused an increase in extracellular MMP-1 (ibuprofen, 177 ± 21% control (p = 0.003); SC299, 230 ± 32% control (p = 0.005)) in the absence of additional stimuli. NS398 was also tested and increased extracellular MMP-1 expression (132 ± 10% control, p = 0.011; data not shown). Thus, FLSCs in culture appear to maintain a program of baseline MMP-1 production that is tonically down-regulated by PGEs.

The ability of COX inhibitors to enhance extracellular MMP-1 expression in the absence of cytokines permitted us to examine in more detail their mechanism of MMP-1 enhancement. As shown in Fig. 8,B (top), incubation of FLSCs with PGE₁ but not PGF_2α attenuated the ability of ibuprofen to enhance MMP-1 release. UO126 but not SB203580 also blocked ibuprofen enhancement MMP-1 production. Similarly, both PGE₁ and UO126 reversed the enhancement of MMP-1 production stimulated by SC299 (Fig. 8,B, bottom). Taken together, these data suggest that nonselective and COX-2-selective COX inhibitors enhance MMP-1 production via reduction of PGE levels and a consequent increase in ERK activity. Such an interpretation predicts that, in contrast with brief exposures, exposure of FLSCs to COX inhibitors for durations sufficient to reduce endogenous PGE levels should result in increased activity of ERK. As shown in Fig. 8 C, extended incubation of FLSCs with 100 μM ibuprofen did indeed result in increased ERK activity, with peak activity at 6 h. Moreover, the increase in ERK activity at 6 h corresponded precisely with the earliest detectable decline in extracellular PGE₁ concentrations (13 ± 4% inhibition, p = 0.048).

FLSCs are the most numerous cells in rheumatoid synovium and, via their ability to proliferate and produce cytokines, PGs, and metalloproteinases, are implicated in cartilage destruction in RA. Pharmacologic agents that regulate FLSC signaling and metalloproteinase expression thus provide insight into synovial biology and possible future therapies for RA. Our studies indicate that both E-series PGs and salicylates inhibit FLSC production of MMP-1 through inhibition of ERK. In contrast, COX inhibitors enhance MMP-1 production via depletion of E PGs and consequent enhancement of ERK activation.

We first examined the role of ERK in IL-1β/TNF-α-dependent FLSC signaling. ERK was activated in response to IL-1β/TNF-α, and the selective MEK inhibitors PD98059 and UO126 inhibited ERK activation. Moreover, MMP-1 but not MMP-13 production was inhibited as a result of ERK pathway inhibition. ERK also regulated the expression of MMP-3 and -9, a fact of potential clinical importance, because MMP-3 cleaves and activates MMP-1 both in vitro and in vivo (48). Thus, ERK activation in FLSCs selectively regulates the cytokine-stimulated, extracellular expression of some but not all MMPs. To our knowledge, ours is the first demonstration that ERK regulates FLSC expression of MMP-1 protein in response to cytokines found in the rheumatoid joint. However, our data are consistent with reports by Han et al. (9, 10) that PD98059 inhibits TNF-α-stimulated increases in MMP-1 message in FLSCs and by Sun et al. (49) that induction of the FLSC MMP-1 promoter by calcium phosphate crystals is ERK-dependent. In chondrosarcoma cells, Mengshol et al. (50) have shown that PD98059 inhibits TNF-α-stimulated expression of MMP-1 but not MMP-13. One way in which our data vary from these other authors is in our concurrent use of IL-1β and TNF-α as costimuli. However, we also observed similar effects in experiments in which IL-1β or TNF-α was used separately (data not shown). Moreover, EGF stimulation of MMP-1 but not MMP-13 expression was also ERK dependent, suggesting that ERK regulates MMP-1 release in response to a variety of different stimuli.

We have previously demonstrated that millimolar but not micromolar concentrations of salicylates inhibit ERK in human neutrophils (34), and Schwenger et al. (25) have documented a similar effect in FS-4 fibroblasts stimulated with TNF-α. We now confirm that millimolar concentrations of ASA and NaS inhibit ERK activation in FLSCs. It is unlikely that the effects of salicylates on FLSC ERK were due to COX inhibition because micromolar concentrations of ASA did not inhibit ERK, despite their ability to inhibit COX both historically (43) and in our experiments. Moreover, neither nonsalicylate NSAIDs (ibuprofen, indomethacin, and 6MNA) nor selective COX-2 inhibitors (NS398, celecoxib, and SC299) inhibited ERK activation. Indeed, if ASA were to inhibit ERK via inhibition of COX, then PGs would be expected to stimulate ERK activation. In contrast, we observed that PGEs inhibited FLSC ERK activation. The mechanism of this inhibition was not determined, but probably relates to the ability of E-prostanoid₂ receptors to activate adenyl cyclase, leading to cAMP accumulation, protein kinase A (PKA) activation, and ERK pathway inhibition at the level of Raf-1 interaction with 14-3-3 proteins (51, 52, 53). Consistent with such a model, E-series PGs inhibit ERK in neutrophils in a cAMP- and PKA-dependent fashion (23, 54, 55). In FLSCs, we have observed that the cell-permeable cAMP analog dibutyryl cAMP inhibited IL-1β/TNF-α-stimulated ERK activation (M.H.P. and P.B.R., unpublished observation).

Consistent with their effects on ERK, millimolar concentrations of salicylates also inhibited MMP-1 expression. The inhibitory effects of salicylates on MMP-1 were unrelated to COX inhibition because 1) the doses of ASA required exceeded those required to inhibit COX and 2) the ability of salicylates to inhibit MMP-1 was not duplicated by nonsalicylate NSAIDs. On the contrary, COX-inhibiting concentrations of ASA as well as ibuprofen, indomethacin, and 6MNA enhanced rather than inhibited IL-1β/TNF-α-stimulated MMP-1 release. This phenomenon was COX-2 dependent because selective COX-2 inhibitors also enhanced MMP-1 expression. These observations are consistent with reports by Dayer et al. (56) that indomethacin stimulated collagenase production in rheumatoid FLSCs and by He et al. (57) that NS-398 induced MMP-1 synthesis in synovial explants from patients with osteoarthritis. Therefore, we hypothesized that production of one or more PGs by COX enzymes suppresses cytokine-stimulated MMP-1 expression by FLSCs. In fact, we observed that exogenous PGEs, but not PGF_2α, inhibited IL-1β/TNF-α-stimulated MMP-1 expression, and PGE₁ reversed the effects of NS398 on MMP-1. Consistent with these observations, both DiBattista et al. (58, 59) and Bockman and colleagues (46, 47) have reported that PGEs suppressed the expression of MMP-1 in a cAMP- and PKA-dependent fashion. Taken together, these data suggest that the mechanism through which COX inhibitors enhance MMP-1 levels depends on their ability to block the production of E-series PGs. Indeed, the ability of COX inhibitors to induce MMP-1 expression in the absence of additional stimuli suggests that baseline PG production exerts a tonic inhibitory effect on the cellular production of MMPs.

As noted above, brief exposure of FLSCs to COX inhibitors had no effect on ERK activation, indicating that rapid, stimulated production of PGs does not mediate ERK activity. Nonetheless, the ability of exogenous E PGs to inhibit both ERK and MMP-1 suggested that COX-2 might regulate MMP-1 by synthesizing endogenous PGEs that inhibit ERK. Consistent with this hypothesis, the enhancing effects of both ibuprofen and SC299 on MMP-1 in the absence of cytokines could be abrogated by either PGE supplementation or pharmacologic ERK inhibition. Moreover, in contrast with the lack of effect of brief exposures to COX inhibitors (which did not affect pre-existing PGE concentrations), extended exposures to ibuprofen of sufficient duration to cause observable decrements in extracellular PGE levels transiently stimulated ERK activity. Thus, whereas millimolar salicylates appear to inhibit MMP-1 expression through effects on ERK, COX inhibitors appear to enhance MMP-1 expression through the very same pathway. In Fig. 9, we have provided a best-fit model for this counter-regulation. Because cytokines stimulate ERK and also up-regulate COX-2 expression, which may then down-regulate ERK via PGE production, this model suggests a possible system of feedback inhibition.

The ability of COX inhibitors to enhance both ERK and MMP-1 in vitro raises questions about whether a similar, clinically relevant effect may occur in vivo. Our data do not permit such an extrapolation in the absence of additional studies, although evidence for a possible antiarthritic effect of PGEs may be found in the reported ability of these prostanoids to suppress adjuvant arthritis in rats and collagen-induced arthritis in mice (60, 61). Similarly, the clinical relevance of ERK inhibition by salicylates or other agents remains to be determined. It is worth noting, however, that recent studies by Pelletier et al. (62) demonstrate that, in a rabbit model of osteoarthritis, oral administration of a selective MEK inhibitor results in ERK inhibition and a reduction of synovial inflammation, as well as a reduction in MMP-1 expression and reduced structural damage. Thus, our observations suggest the importance of ERK in the regulation of MMPs in FLSCs, point to a novel anti-inflammatory effect of salicylates, and suggest that cross-talk between COX pathways and MAPKs may play a central role in the regulation of the FLSC phenotype. Our studies also underline the complexity of cellular responses to PGs and suggest that the net response of articular tissues to COX inhibition is likely to result from a multiplicity of individual effects.

We thank Madeline Rios for her excellent editorial and administrative support, Robert M. Clancy and Clifton Bingham III for helpful discussions, and Nada Marjanovic and Avani Desai for technical assistance.