Celecoxib is the only nonsteroidal anti-inflammatory drug to inhibit bone progression in spondyloarthritis

Abstract

Spondyloarthritis (SpA) is a chronic inflammatory disease that leads to ankylosis of the axial skeleton. Celecoxib (cyclooxygenase–2 inhibitor, COX-2i) inhibited radiographic progression in a clinical study of SpA, but in the following study, diclofenac (COX-2 non-selective) failed to show that inhibition. Our study aimed to investigate whether nonsteroidal anti-inflammatory drugs (NSAIDs) inhibited bone progression in SpA, and whether celecoxib had a unique function (independent of the COX-inhibitor), compared with the other NSAIDs. We investigated the efficacy of various NSAIDs in curdlan-injected SKG mice (SKGc), an animal model of SpA, analyzed by bone micro-CT and immunohistochemistry. We also tested the effect of NSAIDs on osteoblast (OB) differentiation and bone mineralization in primary bone-derived cells (BdCs) from mice, and in ankylosing spondylitis (AS) patients and human osteosarcoma cell line (SaOS2). Celecoxib significantly inhibited clinical arthritis and bone progression in the joints of SKGc, but not etoricoxib (another COX-2i), nor naproxen (COX-2 non-selective). Both DM–celecoxib, not inhibiting COX-2, and celecoxib, inhibited OB differentiation and bone mineralization in the BdCs of mice and AS patients, and in SaOS2, but etoricoxib or naproxen did not. The in silico study indicated that celecoxib and 2,5–dimethyl–celecoxib (DM–celecoxib) would bind to cadherin–11 (CDH11) with higher affinity than etoricoxib and naproxen. Celecoxib suppressed CDH11-mediated β–catenin signaling in the joints of SKGc, primary mice cells, and SaOS2 cells. Of the NSAIDs, only celecoxib inhibited bone progression in SKGc and OB differentiation and bone mineralization in the BdCs of mice and AS patients via CDH11/WNT signaling, independent of the COX-2 inhibition.

INTRODUCTION

Spondyloarthritis (SpA) is a chronic inflammatory rheumatic disease that is characterized by inflammatory back pain due to sacroiliitis and spondylitis, the formation of syndesmophytes leading to ankylosis, and frequently associated with peripheral arthritis, enthesitis, acute anterior uveitis, and inflammatory bowel diseases (1). Based on the treatment recommendations issued by the American College of Rheumatology (ACR) and the Assessment of SpondyloArthritis International Society, nonsteroidal anti-inflammatory drugs (NSAIDs) are first-line, and biologics inhibiting tumor necrosis factor alpha (TNF–α) or interleukin– 17 (IL-17) are used for patients refractory to NSAID treatment. However, unmet needs remain, because TNF–α inhibitors can inhibit abnormal bone formation only when used in the early phase of SpA, while IL-17 inhibitors may induce or exacerbate bowel inflammation (2).

The mechanism of new bone formation in SpA has yet to be fully understood. Implicating genetic mechanisms is logical, considering their major role in susceptibility to SpA. Recent genome-wide association studies have identified single-nucleotide polymorphisms in the prostaglandin (PG) receptor EP4 (PTGER4) gene as risk alleles for SpA (3, 4). PTGER4 encodes the EP4 receptor of PGE₂, and signaling through this receptor has been linked to increased osteoblastic and osteoclastic activity (5, 6). One study (7) has reported that PTGER4 expression contributes to pathogenic Th17 cell accumulation, and is associated with high disease activity in SpA.

NSAIDs are also prescribed for managing pain and swelling in various arthritic diseases and controlling postoperative pain by inhibiting PG synthesis (8). Regarding the effects of NSAIDs on the bone, cyclooxygenase–2 (COX-2) inhibitors (COX-2i), but not cyclooxygenase-1 inhibitors (COX-1i), have been reported to prevent bone formation and fracture healing in vivo (9-12). A clinical study (13) on NSAIDs for SpA indicated that continuous celecoxib administration inhibited radiographic progression. However, one trial of diclofenac indicated no difference in progression (14). Celecoxib is a COX-2i with which cartilage loss was suppressed in osteoarthritis, unlike other NSAID treatments (15). In addition, celecoxib reduced the expression of inflammatory substances and growth factors required for endochondral ossification (9, 10). Two in vitro studies using cell lines demonstrated that celecoxib and 2,5–dimethyl–celecoxib (DM–celecoxib) inhibited osteoblast (OB) differentiation (11, 12); however, the exact mechanism remains unclear. Our study aimed to investigate whether NSAIDs affected abnormal bone change in SpA; and if so, whether there would be a difference in efficacy between COX-2i and COX-2 non-selective inhibitors, or whether celecoxib played a unique role, compared with the other NSAIDs.

RESULTS

Celecoxib inhibited the development of clinical arthritis and abnormal bone changes in SKGc mice, but not the other NSAIDs

First, we investigated the efficacy of various NSAIDs in treating arthritis in SKGc mice with celecoxib, etoricoxib, and naproxen. The etanercept injection was used as a comparator for the NSAIDs. To monitor the severity of arthritis, morphology scores (MS) were measured twice weekly, while CT scores and myeloperoxidase (MPO) intensities were measured on day (d) 56 (Fig. 1A). The MS was significantly lower in the celecoxib treatment group than in the saline treatment group or the naproxen treatment group (Fig. 1B, C). Additionally, micro-CT revealed that bone mineral density (BMD) in the trabecular bone was higher in all NSAID-treated groups, than in the saline-treated group (Fig. 1D, E).

Fig. 1.

Celecoxib inhibits the development of arthritis and bone formation in SKGc mice. (A) NSAIDs were administered to SKG mice for 8 weeks after the injection of curdlan (3 mg/kg), and the MS was measured twice per week for up to 8 weeks. (B) The MSs of mice treated with various NSAIDs and etanercept. (C) MPO activity, measured using in vivo imaging system (IVIS) at the end of 8 weeks. (D) Representative images of the micro-CT of trabecular bones in the femur of mice at the end of 8 weeks. (E) The BMD findings of mice treated with various NSAIDs and etanercept. (F) Representative images of ankle joints in SKG mice obtained with micro-CT at the end of 8 weeks. The red circles indicate abnormal bone formation. (G) Analysis of CT scores for bone erosion and formation (n = 5 for each group). (H) Representative ankle joint tissues stained with H&E at the end of 8 weeks. (I) The mRNA expression levels of ALP, OCN, OPN, and osterix relative to 18S rRNA in the ankle joint tissues of mice (n = 5 for each group). (J) Representative images of ankle joint tissues stained with antibodies against OCN and OPN. Values are presented as the mean ± SD; *P < 0.05 and **P < 0.01, based on the Mann–Whitney U test.

Next, we tested whether NSAIDs affected abnormal bone erosion and formation in the ankle joints of SKGc mice, and whether there was any difference in efficacy among those drugs. Only celecoxib, but not the other NSAIDs, significantly inhibited bone erosion and formation, compared with the control group (Fig. 1F, G). In contrast, naproxen significantly increased bone erosion, compared with the control group. In addition, the inhibition of bone erosion and formation with celecoxib was observed in the joint tissues of these mice (Fig. 1H). The expression of genes associated with bone formation (i.e., alkaline phosphatase (ALP), osteocalcin (OCN), osteopontin (OPN), and osterix) was significantly decreased in the celecoxib group and naproxen treatment group, compared with that in the control group (Fig. 1I). The protein levels of OCN and OPN in mice ankle tissues decreased in the celecoxib treatment group, compared with those in the control group, even though the difference was not statistically significant (Fig. 1J).

Celecoxib inhibited osteoblast differentiation and bone mineralization in mice primary cell culture and primary BdCs from SpA patients

To determine whether celecoxib had a direct effect on bone cells, we set up an OB cell culture system in which various concentrations of NSAIDs were tested: celecoxib at (20 and 50) μM (C20 and C50, respectively); etoricoxib at (10 and 20) μM (E10 and E20, respectively); and naproxen at (100 and 200) μM (N100 and N200, respectively). In addition, to analyze whether the efficacy of celecoxib was due to mechanisms other than inhibiting COX-2, 20 μM DM–celecoxib (DM20), which does not inhibit COX-2, was used as well. Celecoxib, DM–celecoxib, and etoricoxib, but not naproxen, significantly inhibited ALP activity on d 7 (Fig. 2A, B). However, etoricoxib lost its inhibitory effect on ALP activity after d 7, whereas celecoxib and DM–celecoxib maintained their inhibitory effects until d 21. On d 21, the optical density (OD) of ARS was significantly decreased in cells treated with celecoxib or DM–celecoxib, but not with etoricoxib or naproxen. The OCN expression levels were significantly lower in celecoxib- and DM–celecoxib-treated cells, than in etoricoxib- and naproxen-treated cells (Fig. 2C).

Fig. 2.

Celecoxib and DM–celecoxib decrease osteogenic differentiation and bone mineralization. Various concentrations of NSAIDs were administered to mice primary osteoprogenitor cell culture system: celecoxib at (20 and 50) μM (C20 and C50, respectively), DM–celecoxib at 20 μM (DM20), etoricoxib at (10 and 20) μM (E10 and E20, respectively), and naproxen at (100 and 200) μM (N100 and N200, respectively). (A) ALP and ARS staining results on d (7 and 21), and (B) the analysis of ALP and ARS activity. (C) The mRNA levels of OCN on d (7 and 21). (D) The same concentrations and types of NSAIDs were applied to SaOS2 cells for (3 or 14) d. ALP and ARS staining results on d (3 and 14), and (E) the analysis of ALP and ARS activity. (F, G) NSAIDs were applied to mouse OBs from d (8 to 21). (F) OBs stained with ALP and ARS on d 21, and (G) the ALP and ARS staining activity findings. (H) Various concentrations of NSAIDs were administered to BdCs from SpA patients. ALP and ARS staining findings on d (3 and 14). (I) Analysis of the activity of ALP. Values are presented as the mean ± SD; *P < 0.05 and **P < 0.01, based on the Mann–Whitney U test.

Celecoxib and DM–celecoxib significantly decreased the ALP activity and the OD values of ARS in SaOS2 cells, whereas etoricoxib and naproxen did not (Fig. 2D, E). To determine whether celecoxib directly inhibited bone mineralization, NSAIDs were administered only during the bone mineralization phase from d (8 to 21) after the completion of OB differentiation. The ODs for ARS were significantly suppressed only in the C50 group, whereas ALP activity was similar in all treatment groups (Fig. 2F, G). In addition, we observed that a higher dose of celecoxib significantly inhibited the OB differentiation and bone mineralization of bone-derived cells from SpA patients (Fig. 2H, I).

Celecoxib and DM–celecoxib bind to CDH11 with higher affinity than etoricoxib and naproxen

A previous study (16) demonstrated that celecoxib and DM–celecoxib displayed structural potential to bind to cadherin–11 (CDH11), using a new proteochemometric computational drug repurposing method. Structural analysis of the homodimeric interface of the EC1 domain of human CDH11 revealed two patches: A and B (Fig. 3A, left). Patch–A consisted of hydrogen-bonding and hydrophobic interactions, whereas patch–B largely consisted of hydrophobic interactions (Fig. 3A, right). In silico docking simulation of four NSAIDs showed that celecoxib and DM–celecoxib potentially bind to the EC1 domain of CDH11 with higher affinity than naproxen and etoricoxib (Fig. 3B). The mean binding energies of the celecoxib and DM–celecoxib docking models were approximately (–6.38 and –6.53) kcal/mol, respectively, and were not significantly different. Binding affinities were lower with etoricoxib and naproxen than with celecoxib and DM–celecoxib; however, the mean binding energies were (–5.75 and –5.53) kcal/mol, respectively. Interestingly, blind docking revealed that all NSAIDs bound to patch–A of the EC1 domain in the top four models.

Fig. 3.

Structural modeling of celecoxib and DM–celecoxib binding to CDH11. (A) The EC1 homodimer interface of CDH11 (PDB: 2A4C). Patch–A is a hydrogen-bonding concave surface that binds two W residues from the partner EC1 monomer. Patch–B is a hydrophobic interaction pocket. (B) The structure and binding affinity prediction for CDH11-NSAID complexes. The top four binding models are represented. ns, not significant. Values are presented as the mean ± SD; **P < 0.01 and ***P < 0.001, based on the Mann–Whitney U test.

Celecoxib suppressed CDH11-mediated β–catenin signaling

Celecoxib and DM–celecoxib had the potential to bind to CDH11; therefore, we investigated whether celecoxib affected CDH11 expression levels. The CDH11 expression level was decreased in the ankle joints of the celecoxib treatment group (Fig. 4A). Next, we measured the mRNA expression levels of CDH11 in the ankle joints of mice, which were significantly higher in the SKGc group than in the control group (Fig. 4B). However, no differences existed between celecoxib and the other NSAIDs.

Fig. 4.

Celecoxib suppresses CDH11-mediated β–catenin signaling. (A) Immunohistochemical staining of CDH11 in the ankle joints of SKGc mice treated with various NSAIDs. Scale bar = 100 μm. Relative expression of the CDH11 mRNA level: (B) in the ankle joints of SKGc mice, and (C) in primary mice OBs. (D) Immunoblot analysis of CDH11, β–catenin, TCF−4, and GAPDH in mouse OBs during OB differentiation and mineralization. (E) Quantitation of the result from (D) (n = 3 for each group). Values are presented as the mean ± SD; *P < 0.05, and **P < 0.01, based on the Mann–Whitney U test. Confocal images of immunocytochemical staining: (F) for CDH11 (top left, green) and F–actin (top right, red), and (G) for β–catenin (top left, green) and CDH11 (top right, red) in SaOS2 cells (n = 3 for each group). Scale bar = 50 μm (F), and 20 μm (G). The nuclei were counterstained with DAPI (blue). Arrows in E indicate colocalization on the cell surface.

In mouse OBs, the mRNA expression of CDH11 tended to be lower in celecoxib-treated cells, than in etoricoxib and naproxen-treated cells (Fig. 4C). In addition, western blot showed that the expression level of CDH11 increased gradually up to d 7 just before mineralization started, and then decreased in controls and naproxen-treated cells, as previously reported. Interestingly, celecoxib-treated cells showed that the expression of CDH11 continuously increased up to d 14, probably by inhibiting OB differentiation (Fig. 4D, E). The protein levels of transcription factor 4 (TCF-4), a transcription factor that binds to β–catenin, decreased significantly in celecoxib-treated OBs (Fig. 4D, Supplementary Fig. 1).

In SaOS2 cells, the CDH11 expression level on the cell surface decreased with celecoxib and DM–celecoxib, but not with etoricoxib and naproxen treatment (Fig. 4F). We then investigated whether celecoxib affects the binding between CDH11 and β–catenin. Immunofluorescence staining of CDH11 and β–catenin showed that these two proteins colocalized on the cell surface, which when cells were treated with etoricoxib or naproxen, was maintained. However, when cells were treated with celecoxib or DM–celecoxib, these proteins did not colocalize (Fig. 4G). Consistent with these findings, experiments using in vitro binding assay yielded similar results (Supplementary Fig. 2).

DISCUSSION

In this study, we demonstrated that celecoxib, but not the other NSAIDs, inhibited clinical arthritis and aberrant bone formation in animal model of SpA (Fig. 1). NSAIDs, particularly COX-2i, have been reported to prevent bone formation and fracture healing in vivo (12, 17).

We found that celecoxib inhibited abnormal bone formation through mechanisms other than inhibiting PG production, and independent of its COX-2 inhibitory activity. This is supported by our findings that etoricoxib, another type of COX-2i, did not prevent abnormal bone formation (Fig. 1F, G), whereas DM–celecoxib, a close analog of celecoxib that does not have COX-inhibitory activity (11, 12), still inhibited OB differentiation (Fig. 2).

Two studies (12, 17) have shown that celecoxib negatively affects bone formation; however, the exact mechanism remains unknown. Here, the in silico study suggests that celecoxib and DM–celecoxib bind to CDH11 with higher affinity than etoricoxib and naproxen (Fig. 3B). Additionally, celecoxib decreased the expression of CDH11 (Fig. 4A-F).

OBs are highly specialized cells that produce several extracellular matrix proteins and contribute to tissue calcification. CDH11 is expressed in cells of the bone and cartilage, including OBs; and during OB differentiation, its expression is maintained (18). The protein distribution of CDH11 in mesenchymal tissues has been previously described (19), and is related to OB differentiation (20, 21). CDH11 contains extracellular (EC) domains with five repeat sequences (EC1-EC5), one transmembrane region, and one cytoplasmic domain. CDH11 binds to the same type of CDH11 on opposing cells, thereby allowing cell–cell adhesion. CDH11 is anchored to the actin cytoskeleton by binding to β–catenin in a complex that organizes the adherens junction (22). Moreover, the expression of CDH11 has been reported to correlate positively with that of WNT signaling in breast cancer (23). CDH11 and β–catenin were separated in SaOS2 cells treated with celecoxib (Fig. 4G), and the degradation of TCF-4 binding to β–catenin was induced in celecoxib-treated OBs (24, 25). Our findings suggest that celecoxib inhibits OB differentiation via CDH11/WNT signaling.

Our study had several limitations. First, the in silico experiment indicated that celecoxib would bind to CDH11; therefore, further detailed studies of this binding of celecoxib to CDH11 are needed. Second, the precise mechanisms through which celecoxib–CDH11 binding inhibits OB differentiation and mineralization remain to be clarified. Although we found that the TCF-4 expression was reduced in celecoxib-treated OBs (Fig. 4D, Supplementary Fig. 1), the mechanisms by which celecoxib changed the CDH11 and WNT molecules interaction and the WNT signaling during OB differentiation have yet to be elucidated. Finally, we discovered that the administration of recombinant CDH11 suppressed OB mineralization (Supplementary Fig. 3). Further experiments are warranted to compare the inhibiting efficiency between celecoxib and recombinant CDH11.

In conclusion, celecoxib, but not the other NSAIDs, inhibited aberrant bone formation, independent of the COX-inhibitory effect in an animal model of SpA and primary mice cells, SaOS2 cells, and BdCs from AS patients, probably via CDH11/WNT signaling. Further basic and clinical studies are warranted to choose the optimal NSAID in the treatment of SpA.

MATERIALS AND METHODS

Experimental animal model

Female SKG mice with a BALB/c background were purchased from Clea Japan and maintained in a specific pathogen-free facility. Severe arthritis was induced in SKG mice aged 11 weeks by injecting a suspension of curdlan (Wako) intraperitoneally (i.p.) at 3 mg/kg (SKGc). For the experiments, the animals were divided into six groups: negative control group (n = 5 mice), vehicle group (saline, n = 15), and four active treatment groups [celecoxib (n = 15); naproxen (n = 10); etoricoxib (n = 10); and etanercept (n = 15)]. For the treatment groups, 30 mg/kg celecoxib (Pfizer Inc.), 100 mg/kg naproxen (Chong Kun Dang), 20 mg/kg etoricoxib (Merck), or saline was administered to the mice for 5 days each week from day 0 to day 56. Etanercept (Pfizer Inc.) was administered at 4 mg/kg i.p. twice weekly for 56 days. All experiments were approved by the Institutional Animal Care and Use Committee of Chungnam National University Hospital (Daejeon, South Korea; approval no. CNUH-019A0025) and were conducted in accordance with the Laboratory Animals Welfare Act and Guide for the Care and Use of Laboratory Animals.

Human primary bone-derived cells (BdCs)

Primary BdCs derived from vertebral bone tissue obtained during spinal surgery in SpA patients were provided by Dr. Tae-Hwan Kim (Department of Rheumatology, Hanyang University Hospital for Rheumatic Diseases, Seoul, South Korea). The cells were isolated and cultured, as previously described (26, 27). This study was approved by the Institutional Review Boards of Hanyang University Guri Hospital (Guri, South Korea; IRB file No. 2014-05-002) and Hanyang University Seoul Hospital (Seoul, South Korea; IRB file No. 2014-05-001). It was conducted in accordance with the principles of the Declaration of Helsinki. All patients provided written informed consent, and their data were anonymized and de-identified.

Statistical analysis

All values are expressed as mean ± the standard deviation. Significant differences between groups were determined using a nonparametric statistical method [i.e., two-tailed Mann–Whitney U test (ver. 18.0. Chicago, IL)]. Differences were statistically significant at a P-value of < 0.05.

Additional details

Clinical scoring, reagents and more detailed methods of histology, immunohistochemistry, cell culture, differentiation of osteoblasts, assessment of osteoblast differentiation and mineralization, RT-qPCR, immunoblot analysis, immunofluorescence staining, and confocal microscopy are described in the Supplementary Information.