Celecoxib: considerations regarding its potential disease-modifying properties in osteoarthritis

Osteoarthritis (OA) is a degenerative joint disease characterized by progressive loss of articular cartilage, subchondral bone sclerosis, osteophyte formation, and synovial inflammation, causing substantial physical disability, impaired quality of life, and significant health care utilization. Traditionally, non-steroidal anti-inflammatory drugs (NSAIDs), including selective cyclooxygenase (COX)-2 inhibitors, have been used to treat pain and inflammation in OA. Besides its anti-inflammatory properties, evidence is accumulating that celecoxib, one of the selective COX-2 inhibitors, has additional disease-modifying effects. Celecoxib was shown to affect all structures involved in OA pathogenesis: cartilage, bone, and synovium. As well as COX-2 inhibition, evidence indicates that celecoxib also modulates COX-2-independent signal transduction pathways. These findings raise the question of whether celecoxib, and potentially other coxibs, is more than just an anti-inflammatory and analgesic drug. Can celecoxib be considered a disease-modifying osteoarthritic drug? In this review, these direct effects of celecoxib on cartilage, bone, and synoviocytes in OA treatment are discussed.

drug? In this review, the direct eff ects of celecoxib on cartilage, bone, and synoviocytes in OA treatment are discussed.
It is important to note that some of the eff ects described may be related to the coxib class of drugs as a whole, some may be specifi c to celecoxib, and some may result from a general COX-inhibiting eff ect. Th is review does not intend to distinguish between these but focuses on the properties of celecoxib specifi cally. Only when celecoxib has been compared to other treatments have such comparisons been taken into account. Moreover, this review does not discuss the issue of side eff ects and clinical effi cacy of celecoxib, but focuses on its potential tissue structure-modifying, mostly chondroprotective, eff ects.

Methods
Two electronic databases were searched for relevant publications: PubMed (1990 to March 2010) and EMBASE (1990to March 2010. Key words used were: celecoxib/Celebrex/SC-58635, osteoarthritis/arthrosis/OA, cartilage/chondrocytes, synovium/synovial/synovio cytes, and bone. Celecoxib studies regarding its eff ects on cartilage, bone, and synovium were selected by screening title and abstract. Publications not written in English or not containing original data were excluded. Reviews concerning subjects like the cost-eff ectiveness and cardiovascular/gastrointestinal side eff ects of celecoxib and the use of celecoxib in cancer treatment have been published and are therefore not covered in this review.

Celecoxib: direct eff ects on cartilage
In OA, chondrocytes fail to maintain the equilibrium between synthesis and degradation of the extracellular matrix, resulting in progressive disruption of the structural integrity of cartilage. Initially, chondrocytes compen sate for the enhanced catabolic processes by increasing synthesis of collagens and proteoglycans. However, as OA progresses, the increasing catabolic enzyme activity can no longer be counterbalanced [14]. IL-1β and TNF-α play key roles in the destructive process by stimulating expression and release of proteases, such as collagenases and aggrecanases, including matrix metalloproteinases (MMPs) and a disintegrin and metalloproteinase with trombospondin repeats (ADAMTS), which degrade collagen and aggrecan. Th ese pro-infl ammatory cytokines stimulate synthesis and release of nitric oxide (NO) and PGE 2 [15]. Chondrocytes from OA patients show elevated COX-2 expression, and its product PGE 2 is increased in OA cartilage [16]. Th e function of PGE 2 in OA is not exactly clear as it has both catabolic and anabolic eff ects in cartilage [17,18]. NSAIDs could potentially aff ect cartilage through their inhibition of PGE 2 production.

Proteoglycan turnover
Celecoxib dose-dependently inhibits glycosaminoglycan release and stimulates proteoglycan synthesis in healthy human articular cartilage explants when exposed to peripheral blood mononuclear cells from rheumatoid arthritis patients or IL-1β and TNF-α [19]. Th e fact that the decreased proteoglycan synthesis induced by IL-1β and TNF-α is reversed by celecoxib indicates that this drug can also exert its eff ects directly on activated cartilage. Furthermore, in OA cartilage explants, celecoxib stimulated proteoglycan synthesis and retention of newly formed proteoglycans [20][21][22]. Th e non-selective COX inhibitors diclofenac and naproxen did not aff ect proteoglycan turnover in OA cartilage, and indomethacin and an experimental COX-1 selective inhibitor (SC-560) had adverse eff ects [20,21]. Th is diff erence in NSAID eff ects supports COX-2 involvement in catabolic activity regulation in cartilage, whereas COX-1 activity might have a more physiological role in chondrocytes.
No eff ect of celecoxib on proteoglycan turnover was observed in healthy cartilage [19,22]. Th is is in contrast to the protective in vitro eff ect of celecoxib on end-stage OA cartilage obtained at joint replacement surgery. For the treatment of OA in clinical practice, it would be benefi cial if celecoxib could infl uence proteoglycan turnover in earlier stages of disease. It was shown that in both degenerated (pre-clinical) and late-stage OA cartilage, celecoxib not only stimulated proteoglycan synthesis and retention of newly formed proteoglycans, but also had favorable eff ects on proteoglycan content. Importantly, proteoglycan content in degenerated cartilage norma l ized in vitro during celecoxib treatment, suggesting celecoxib treatment in the early stages of OA could slow down or even reverse the destructive process [22].
Whereas the in vitro eff ects of celecoxib on OA cartilage are benefi cial, results obtained with isolated chondrocytes are not consistent. In a mechanically stretched monolayer of chondrocytes, celecoxib had a positive eff ect on aggrecan expression and reduced the release of chondroitin sulfate [23]. In contrast, celecoxib had no positive eff ect on proteoglycan turnover of osteoarthritic chondrocytes cultured in alginate beads [24], of a monolayer of chondrocytes [25], nor in an in vitro model of post-traumatic OA [26]. Th is variation in the eff ects of celecoxib could potentially be due to diff erences in chondrocyte culture models, whereas cartilage explants probably better refl ect the in vivo situation.

Prostaglandin E 2 -induced catabolism
A possible way in which celecoxib exerts its eff ect on proteo glycan turnover is inhibition of PGE 2 production. PGE 2 is highly expressed in OA cartilage and studies indicate a pivotal role for PGE 2 in OA cartilage metabolism [27]. Expression of PGE 2 and COX-2 in OA cartilage is strongly inhibited by celecoxib [10,21,22,28,29]. PGE 2 enhances IL-1β-/TNF-α-induced proteo glycan release, resulting in decreased proteoglycan content in cartilage explants [28]. Th e eff ect of PGE 2 on the synthesis of proteoglycans remains controversial; in OA cartilage, proteoglycan synthesis is inhibited by PGE 2 [17], whereas PGE 2 does not aff ect proteoglycan synthesis rate in healthy cartilage [28]. Th is discrepancy could be due to diff erences in expression levels of individual members of the EP receptor family (EP1 to EP4) through which PGE 2 exerts its eff ects. EP4 has been implicated in mediating catabolic eff ects because it is highly expressed in OA cartilage [17]. IL-1-induced expression of EP4 in cultured OA chondrocytes is decreased by celecoxib [29], but not consistently [17]. Th e overall negative eff ect of PGE 2 on proteoglycan turnover in cartilage might be mediated through the EP4 receptor ( Figure 1). PGE 2 inhibits collagen synthesis and stimulates expression of MMP and ADAMTS-5, proteolytic enzymes involved in the degradation of collagens and proteo glycans [17,30,31]. Th eoretically, celecoxib could also prevent cartilage destruction by inhibiting induction of MMP expression in OA cartilage. Both inhibitory and stimulatory eff ects of celecoxib on IL-1-induced expres sion of MMP-13 in OA chondrocytes have been reported [10,17]. Also, there is no agreement on the eff ect of  (1). Subsequently, the prostanoid receptor EP4 is up-regulated via a COX-2-dependent mechanism (2). Increased COX-2 activity results in large concentrations of prostaglandin E 2 (PGE 2 ) (3). PGE 2 exerts its eff ects through the prostanoid receptor EP4 (4), resulting in the increased expression of matrix metalloproteinases (MMPs) and a disintegrin and metalloproteinase with thrombospondin repeats (ADAMTS)-5. Furthermore, PGE 2 augments the release of newly formed proteoglycans from cartilage and reduces the synthesis of proteoglycans (5). IL-1β and TNF-α also activate the transcription factors NF-κB and JNK (6), which stimulate the expression of inducible nitric oxide synthase (iNOS) (7), resulting in the formation of nitric oxide (NO) (8). NO has a potential role in inducing chondrocyte apoptosis, inhibiting proteoglycan synthesis and stimulating MMP activity (9). Together, the eff ects of NO and PGE 2 result in cartilage degeneration. Celecoxib prevents the negative eff ects of PGE 2 and NO on cartilage destruction by inhibiting both COX-2 and NF-κB/JNK, thereby potentially slowing cartilage degradation in osteoarthritis. celecoxib on MMP-1 expression in cartilage [10,25,32]. Celecoxib reverses IL-1β-induced ADAMTS-5 expres sion in OA cartilage explants [17]. As such, it could prevent enhanced proteoglycan turnover in OA by aff ecting both MMP and ADAMTS-5 expression. But our understanding of the infl uence of celecoxib on PGE 2 -induced cartilage catabolism is clearly far from complete and it would be worthwhile to explore this role in more detail.

Nitric oxide, NF-κB, and chondrocyte apoptosis
NO plays an important role in cartilage destruction in OA -for example, by inhibiting matrix synthesis, activating MMPs, and inducing chondrocyte apoptosis [33][34][35]. Because NO is an attractive target in OA treatment, several studies have addressed the question of whether celecoxib infl uences NO production, although little agree ment has been reached. Several studies found inhibi tory eff ects of celecoxib on NO production in chondro cytes [25,32,36], whereas others did not [28,37]. Th ese contradictory eff ects are potentially due to diff erences in culture models, treatment duration, and celecoxib concentration used.
In articular chondrocytes, NO production is regulated by NF-κB, JunNH 2 -terminal kinase (JNK) and p38 [32,38]. Celecoxib was shown to suppress NO production by inactivating JNK and NF-κB [32]. An inhibitory eff ect of celecoxib on NF-κB signaling in OA chondrocytes was reported previously [10]. NF-κB has an essential role in OA pathogenesis, being involved in cytokine stimulation, MMP and ADAMTS expression, and diminished secretion of extracellular matrix proteins by chondrocytes. Inhibition of NF-κB could potentially be benefi cial in OA treatment. Interestingly, it was reported that celecoxib reduces expression of IL-1 [37] and IL-6 [24], both infl am matory cytokines involved in OA pathogenesis [39]. It is currently unknown how celecoxib mediates its eff ects on cytokine expression and NF-κB activity.
Celecoxib induced apoptosis in a dose-dependent manner in chondrocytes derived from cartilage from patients with OA [25], although reduced apoptosis via COX inhibition by celecoxib has also been reported [26].
In general, celecoxib has favorable eff ects on cartilage destruction in vitro, thereby theoretically slowing down disease progress in vivo ( Figure 1).

Inhibition of signal transduction and pro-infl ammatory mediators
Although originally viewed as a non-infl ammatory arthro pathy, a pivotal role of synovial infl ammation in OA progression is now recognized. Imaging studies have shown synovium changes in early and late OA [40]. Histologically, synovium from OA patients shows hyperplasia, increased lining layer thickness, blood vessel for ma tion and mononuclear cell infi ltration, mainly consist ing of macrophage like cells. IL-1β and TNF-α levels are increased in OA synoviocytes, potentially contributing to disease progression by activating chondrocytes and synovial fi broblasts [41,42]. Enhanced PGE 2 and COX-2 expression in synovial fl uid and synovial membrane have been observed [43,44]. Several eff ects of celecoxib on synovium, with a focus on fi broblasts, have been des cribed. Celecoxib reversed IL-1βinduced PGE 2 and COX-2 protein expression in synovial fi broblasts. Further more, celecoxib inhibited IL-1βinduced activa tion of NF-κB in synovial fi broblasts from OA patients [10]. NF-κB induces expression of large numbers of infl ammatory mediators and plays a major role in the initiation and maintenance of synovitis, synovial hyperplasia, and inhibition of synovial apoptosis in rheumatoid arthritis. Although less is known concerning the function of NF-κB in osteoarthritic synovium, it is clear that celecoxib could reduce expression of various infl amma tory mediators by downregulation of NF-κB [45].

Proteolysis
Among the downstream factors of NF-κB are MMPs, which play a crucial role in cartilage degradation in OA. Both MMP-1 and MMP-13 levels are enhanced in OA; MMP-1 is predominantly released by synovial cells, and MMP-13 is highly expressed by chondrocytes [46]. MMP-2 and MMP-9 are also elevated in the osteoarthritic joint. MMP-2 expression is regulated by COX-2. Several NSAIDs, including celecoxib, inhibit MMP-2 secretion in OA synovial fi broblast cultures [47]. Furthermore, celecoxib can decrease the expression of MMP-9 and urokinase-type plasminogen activator (u-PA) and its inhibitor PAI [47]. Alterations in u-PA and PAI expression have been found in osteoarthritic tissue and contribute to a disturbed proteolytic balance [48].
It was shown that celecoxib, but no other selective COX-2 inhibitors, enhances MMP-1 and MMP-13 protein expression in IL-1β-stimulated synoviocytes [10,49]. Th is observation does not corroborate the inhibitory eff ect of celecoxib on MMP-1 expression in rheumatoid arthritis synoviocytes [50]. Th is discrepancy could be due to diff erent concentrations used, celecoxib being stimulatory at low concentrations (0.5 to 1 μM) and inhibitory at higher concentrations (5 to 10 μM). Evidently, a stimulatory eff ect of celecoxib on synovial MMP-1 and MMP-13 expression could be detrimental in OA treatment [10]. In conclusion, celecoxib infl uences the balance of proteolytic enzymes in OA synovium, and although this appears to be generally benefi cial (reducing expression of MMP-2, MMP-9, and uPA), adverse eff ects have been reported as well (increas ing expression of MMP-1 and MMP-13).

Apoptosis
Recently, it was shown that celecoxib dose-dependently inhibits proliferation and induces apoptosis in synovial fi broblasts obtained from OA patients [51,52]. Th is is in agreement with fi ndings in rheumatoid arthritis [53][54][55]. Remarkably, various other COX-2 selective inhibitors, including nimesulide and rofecoxib, did not induce apoptosis of synovial fi broblasts, indicating that celecoxib stimulates apoptosis in a COX-2-independent manner [51]. In cancer cells celecoxib has been shown to modulate apoptosis pathways by inhibiting anti-apoptotic proteins, elevating Ca 2+ concentration and altering NF-kB signaling (reviewed in [56]). Although the exact proapoptotic mechanism of celecoxib in synovial tissue remains to be established, it is evident that antiproliferative and pro-apoptotic eff ects of celecoxib on synovium are benefi cial in reducing synovial hyperplasia and potentially slow down synovitis-mediated OA disease progress.
Taken together, celecoxib modulates several pathogenic mechanisms of synovial cells that are not always aff ected by other NSAIDs, suggesting that celecoxib may have additional, COX-2-independent value in the treatment of OA (Figure 2).

Subchondral bone in osteoarthritis
Subchondral bone sclerosis and osteophyte formation are radiographic hallmarks of end-stage OA. Several studies suggest that bone remodeling in OA is biphasic: an early decrease in trabecular bone formation, followed by an increase in subchondral bone density and stiff ness [57,58]. Th e initial thinning of the subchondral plate coincides with changes in articular cartilage, suggesting a pivotal role for the cartilage and subchondral bone interaction in OA progression. In established OA, the increased subchondral bone stiff ness probably contributes to further cartilage degeneration [59].

Osteoclastogenesis
Osteoclasts play a pivotal role in the destruction of subchondral bone [4,14,59]. Osteoclastogenesis and activa tion of mature osteoclasts are critically regulated by the receptor activator of NF-κB ligand (RANKL). RANKL mediates its function by binding to its cell-surface receptor RANK on osteoclast precursor cells and osteoclasts, thus stimulating diff erentiation and activation of osteoclasts. It is mainly expressed by osteoblasts and stromal cells, where expression of RANKL is In osteoarthritic synovium, increased levels of IL-1β and TNF-α stimulate the expression of cyclooxygenase (COX)-2 and the ensuing production of prostaglandin E 2 (PGE 2 ). PGE 2 augments the expression of proteolytic enzymes, including matrix metalloproteinases (MMPs) and urokinase-type plasminogen activator (uPA), thereby contributing to the destruction of the articular joint. Celecoxib has a benefi cial eff ect on synovium through its inhibition of COX-2 (1). In addition, celecoxib inhibits IL-1β-induced activation of NF-κB, and hence diminishes the expression of various infl ammatory mediators (2). A third potential favorable eff ect of celecoxib on synovium is its induction of apoptosis in synovial fi broblasts (3). This would theoretically contribute to decreased synovial hyperplasia and reduced expression of infl ammatory mediators. PAI, plasminogen activator inhibitor. COX-2-dependent [60]. During infl ammation RANKL is also produced by T lymphocytes and fi broblast-like synovio cytes. Osteoprotegerin (OPG), a soluble decoy receptor for RANKL, can prevent the biological eff ects of RANKL, and the ratio between OPG and RANKL determines whether the balance is in favor of bone resorption or bone formation [61,62]. Interestingly, two osteoblast sub populations were identifi ed in OA, one with a low OPG/RANKL ratio that favors bone resorption, and one with a high OPG/RANKL ratio that promotes bone formation [61,63]. Inhibition of COX-2 by NSAIDs diminishes RANKL production by osteoblasts, and since RANKL is an important inducer of osteoclastogenesis, celecoxib inhibited osteoclast diff erentiation in co-cultures of osteo blasts and bone marrow-derived cells [12,64]. Besides aff ecting osteoclastogenesis indirectly through its eff ect on osteoblasts, celecoxib also directly infl uenced osteo clast precursor cells by inhibiting COX-2 expression. Adding celecoxib to bone marrow-derived monocyte/ macrophage cells, in the absence of stromal cells, suppresses RANKL-induced osteoclast diff erentiation [65,66]. Th is celecoxib eff ect was reversed by PGE 2 , indicat ing that RANKL-induced COX-2 and PGE 2 expression in osteoclast precursors is critically involved in osteoclastogenesis [65] (Figure 3).

Osteoclast activity
Besides inhibiting osteoclast diff erentiation, celecoxib is able to almost completely inhibit the activity of human osteoclasts [66]. Slightly lesser eff ects were observed with indomethacin, and no eff ects were seen with a selective COX-1 inhibitor, suggesting a COX-2-dependent pathway is involved [66]. However, other mechanisms might be involved in inhibiting osteoclast activity as well. Celecoxib, as well as other sulfonamide-type COX-2 inhibi tors, contain an aryl sulfonamide moiety that inhibits carbonic anhydrase II [67]. Abundantly expressed on the inner surface of osteoclasts, carbonic anhydrase II catalyzes conversion of CO 2 and H 2 O into bicarbonate and H + . Acidifi cation in the resorption pit is required for dissolution of the inorganic matrix of bone [68]. Treatment with celecoxib reduced carbonic anhydrase activity and thereby inhibited osteoclast activity, an eff ect not observed for COX-inhibitors without this sulfonamide moiety [12].

Osteoprotegerin/RANKL in chondrocytes
Recently, it was found that human chondrocytes express OPG, RANKL and RANK [61,69]. Interestingly, the OPG/RANKL ratio is signifi cantly lower in OA chondrocytes compared to healthy chondrocytes [70]. Th is shift in OPG/RANKL ratio is mediated by PGE 2 [69,71], and inhibition of PGE 2 production by celecoxib resulted in a higher OPG/RANKL ratio [71,72]. It was shown that RANKL produced by chondrocytes can stimulate osteoclasto genesis [73,74] and, furthermore, as a chemoattractant for peripheral blood monocytes, it could attract osteoclast precursor cells to the joint [75]. Inhibition of chondrocyte RANKL expression by celecoxib might thus prevent subchondral bone loss (Figure 3).

Cartilage
In vitro experiments have shown a cartilage-sparing eff ect of celecoxib in OA cartilage; however, in vivo data, from either human or animals, are scarce. Contrary to its positive eff ects on cartilage degeneration in vitro, no chondroprotective eff ect of celecoxib in the canine groove model of OA was observed [76]. Although PGE 2 levels in the joint were inhibited, celecoxib did not improve cartilage histopathology or proteoglycan turnover. Th is lack of chondroprotective eff ect might have been due to increased loading of the joint in the celecoxib-treated group compared to the placebo-treated group, where no analgesics were given [76]. Conversely, celecoxib was shown to reduce cartilage damage in collagen-induced osteoarthritis in rabbits; histopathological evaluation showed less cartilage erosion, reduced cartilage fi brillation and decreased loss of chondrocytes. Proteoglycan content, determined by Safranin-O staining intensity, was higher than in the placebo-treated group [77]. Next to the direct eff ects of celecoxib, the anti-infl ammatory eff ects of celecoxib may have caused this chondroprotective eff ect as the model depends on infl ammation and the number of infl ammatory cells and the PGE 2 concentration in synovial fl uid was signifi cantly reduced by celecoxib. Few studies have described the in vivo eff ects of celecoxib on cartilage destruction in OA patients [37,44,[78][79][80]. However, these studies generally have limitations with respect to their small size and short duration. A way to study drug eff ects is to treat patients with severe knee OA waiting for joint replacement surgery and analyze the cartilage ex vivo. In this manner, a benefi cial eff ect of celecoxib on cartilage degradation after 4 weeks of treatment was observed [78]. Although no diff erences in the histopathological Mankin score were observed, proteoglycan synthesis rate and retention of newly formed proteoglycans was signifi cantly increased in celecoxib-treated OA patients compared to indomethacintreated or untreated patients. Th e expression of key players in the destructive process, NO and PGE 2 , was inhibited by both celecoxib and indomethacin [78]. Hence, diff erences in cartilage proteoglycan turnover between celecoxib-and indomethacin-treated patients could result from specifi c eff ects of indomethacininduced COX-1 inhibition on cartilage [20,21], or from COX-2-independent actions of celecoxib [13]. Using a similar approach, long-term (3 months) eff ects of celecoxib and aceclofenac were studied in OA patients [37]. It was demonstrated that expression of COX-2, microsomal prostaglandin E synthase-1 (mPGES-1) and inducible NO synthase, an enzyme involved in NO generation, was strongly reduced in both celecoxib-and aceclofenac-treated patients. Only celecoxib was shown to inhibit expression of the PGE 2 receptors EP2 and EP4, as well as TNF-α and IL-1β, in articular cartilage. A positive correlation exists between TNF-α/IL-1β levels and cartilage damage [81], suggesting a chondroprotective eff ect of celecoxib in vivo.
Th e eff ects of celecoxib treatment on disease progression are more ambiguous [79,80,82]. In an observational study, conventional NSAID use was associated with enhanced cartilage destruction compared to selective COX-2 inhibitors. Furthermore, the COX-2 inhibitors rofecoxib and celecoxib showed benefi cial eff ects on tibial cartilage defects in knee OA compared to no medication [82]. Recently, the eff ect of celecoxib treatment (200 mg daily, 12 months) on cartilage volume loss was studied compared to a historical cohort of patients receiving standard care [79]. Using quantitative magnetic resonance imaging, no protective celecoxib eff ect on knee cartilage was found. Only one randomized controlled trial has addressed the eff ects of celecoxib on cartilage degeneration [80]. Patients who met radiographic criteria grade 2 and 3 (Kellgren and Lawrence) were blinded and given celecoxib, chondroitin sulfate, glucosamine or placebo. Unexpectedly, no diff erences in joint space narrowing (measured radiographically) or  (2), leading to production of receptor activator of NF-κB ligand (RANKL). RANKL stimulates the diff erentiation of osteoclast precursor cells into quiescent osteoclasts (3). Furthermore, it induces the expression of COX-2 and prostaglandin E 2 (PGE 2 ) in quiescent osteoclasts, and subsequently PGE 2 activates osteoclasts in both an autocrine and paracrine manner (4). Celecoxib inhibits the COX-2-dependent RANKL production by chondrocytes and osteoblasts, thereby avoiding osteoclastogenesis and osteoblast activity. Furthermore, celecoxib directly inhibits the diff erentiation of precursor cells, independent of RANKL production by stromal cells. Celecoxib can also directly aff ect the osteoclasts themselves by decreasing RANKL-induced PGE 2 expression and inhibiting carbonic anhydrase. Decreased carbonic anhydrase activity will diminish acidifi cation of the resorption pit, and hence decrease osteoclast activity (5). disease progression between celecoxib-and placebotreated groups were observed after 2 years follow-up [80]. Less than anticipated loss of joint space width in the placebo-treated group hampered the study and prevented a strong conclusion. Moreover, the results found in these studies were obtained in an un-controlled trial set-up and, as such, could be aff ected by the selection of patients. Also, the numbers of patients used in most studies is rather limited. Figure 4 summarizes the suggested in vivo eff ects of celecoxib. Th e benefi cial in vitro eff ects and the somewhat controversial in vivo eff ects on cartilage, mostly based upon weak evidence, clearly indicate the requirement for properly designed randomized controlled trials on the potential disease-modifying osteoarthritic drug eff ects of celecoxib.

Synovium
Celecoxib has been shown to reduce synovitis, leukocyte infi ltration and synovial hyperplasia in diff erent arthritis animal models [83][84][85]. In the synovium of severe knee OA patients, inhibitory eff ects of celecoxib on IL-1β and TNF-α expression have been demonstrated [44,78]. Further more, celecoxib reduced IL-6 concentrations in the synovial fl uid of patients with moderately severe OA after 2 weeks of treatment [86]. Interestingly, aceclofenac and indomethacin had no or only moderate eff ects on cytokine expression in these studies [44,78].
Reduction of pro-infl ammatory cytokines in synovial fl uid by celecoxib could be the result of decreased production by chondrocytes, as has been shown in vitro [24]. However, synovial macrophages are also an important source of pro-infl ammatory cytokines [42]. Ex vivo analysis of OA synovium after in vivo celecoxib treatment showed a signifi cant reduction in synovial macrophage numbers, which was not observed for aceclofenac [44]. Th is macrophage depletion might be due to enhanced apoptosis in response to celecoxib, which has a proapoptotic eff ect on synoviocytes and macrophages [51,53,55]. Decreasing macrophage numbers would result in lower pro-infl ammatory mediator levels in synovial fl uid. Only one study has addressed the infl uence of celecoxib on MMP activity in synovial tissue; despite controversial results on MMP activity in synoviocytes in vitro, no celecoxib eff ect on MMP activity was demonstrated in vivo [78].
In conclusion, under certain conditions pro-infl ammatory cytokines play a crucial role in OA pathogenesis by inhibiting proteoglycan synthesis, inducing chondrocyte apoptosis and activating other cells. Preventing enhanced production of these infl ammatory mediators by celecoxib will likely slow disease processes. Several lines of evidence indicate that synovial changes can be among the fi rst to occur in OA (reviewed in [87]), suggesting early treatment could slow or maybe prevent joint damage. As little research has focused on the eff ects of celecoxib on synovial tissue, further research should elucidate the eff ects of celecoxib in disease progression.

Bone
Various studies have shown a benefi cial eff ect of celecoxib on bone in vivo [12,[88][89][90]. Celecoxib, but not other NSAIDs, reduced bone mineral density loss [12,88,90] and enhanced trabecular bone volume in adjuvant-and collagen-induced arthritis in rats [88][89][90]. Th e increased trabecular bone volume correlated with reduced serum type I collagen C-telopeptide, a bone resorption marker representing osteoclast activity [12,88], and other bone resorption parameters [89]. Whereas celecoxib did not aff ect bone formation, it suppressed osteoclast numbers in tibia of arthritic animals [89,90]. Th ese celecoxib eff ects were partly mediated by RANKL, as celecoxib decreased expression of RANKL in synovial tissue, bone marrow cells and cartilage in vivo [71,89]. As shown in vitro, celecoxib inhibited both osteoclastogenesis and osteoclast activation, thereby directly diminishing bone destruction. Despite celecoxib being used for treatment of OA for many years, no eff ects of it on serum markers of bone resorption and formation or on structural changes in bone have been reported. As celecoxib has benefi cial eff ects on bone resorption in vitro and in vivo in animal models, it would be interesting to explore these eff ects on bone metabolism in OA patients in more detail.

Conclusion
Despite celecoxib being approved for OA treatment for over a decade, few studies have addressed the diseasemodifying properties of this selective COX-2 inhibitor, specifi cally in vivo. Th is review does not address the clinical risk and side eff ects related to the clinical benefi ts of celecoxib but focuses on the disease-modifying properties of this compound. However, the increased risk of myocardial infarction and worsening of high blood pressure can not be ignored when prescribing celecoxib. Th ese issues have been extensively described in other reviews and are still under discussion at present. Also, it is not the intention of this review to compare in a systematic way the disease-modifying eff ects of celecoxib with other coxibs and conventional NSAIDs. As such, all eff ects described might be partly class-specifi c and partly celecoxib-specifi c. Nonetheless, celecoxib's chondro protec tive eff ects -prevention of synovial hyperplasia, and inhibition of bone destruction in vitro and in vivo specifi cally in animal models -suggest that it and maybe other coxibs could potentially slow OA disease progression in humans. At present, however, good quality randomized controlled trials examining the diseasemodifying eff ects of celecoxib are lacking. Future studies should elucidate the actual role of celecoxib and other selective coxibs as disease-modifying osteoarthritic drugs.

Competing interests
MZ, JAGR, FPJGL and SM declare that they have no competing interests. The work of TNB is supported by an unrestricted grant from Pfi zer. JWJB received a consultancy fee from Pfi zer (<US$10,000).