Studies were carried out in 18 adult male New Zealand rabbits with a body weight of 3.0 to 3.5 kg (Granja San Bernardo, Navarra, Spain). Animal handling and experimentation were performed in accordance with Spanish regulations and the Guidelines for the Care and Use of Laboratory Animals drawn up by the National Institutes of Health (Bethesda, MD, USA). The experimental protocol was approved by the Institutional Ethics Committee. All rabbits were allowed to adapt to the facilities for one week and were then randomly separated into three groups: 1) eight healthy rabbits (healthy group), 2) three rabbits with OA (OA group), and 3) seven rabbits with chronic antigen-induced arthritis (AIA group).

AIA was induced in rabbits as previously described in detail 31 . Briefly, animals were given two intradermal injections of 4 mg ovalbumin (OVA; Sigma-Aldrich, St. Louis, MO, USA) in Freund's complete adjuvant (Difco, Detroit, MI, USA). Beginning five days after the second injection, 1 ml of OVA (5 mg/ml in 0.9% NaCl) was injected into the knee joints on a weekly basis over the following four weeks. The animals were euthanized at the end of this period to evaluate chronic damage and not acute damage (Figure 1A). Samples of the synovial membrane, articular cartilage, and subchondral bone in each knee were collected for further studies.

In order to establish a control for rheumatic disease that occurs with a lower inflammatory response and a lower subchondral bone loss than chronic arthritis, surgical OA was induced in the knees of rabbits following a previously described protocol 32 . Consequently, partial medial meniscectomy and anterior cruciate ligament (ACL) sectioning were performed. The rabbits were euthanized at the end of the eighth week (Figure 1A), and knee joint tissues were obtained for further studies.

To determine the presence of juxta-articular osteoporosis in rabbit knees, radiographic analysis and BMD measurements were taken just before the mice were euthanized. In the healthy group, these assessments were carried out at week 9 of the study.

Digital x-ray images of the knee were obtained from rabbits placed in the supine decubitus position. All radiographs were performed by the same operator using the Philips Diagnost 93 system (Philips Medical Systems, Eindhoven, The Netherlands), at 60 kV, 200 mA, 100 ms (Konica Minolta Medical film 27.9 × 35.6 cm; Casette Regius RC-110, Konica Minolta, USA). Radiographs were obtained using a vertical x-ray beam centered over the femorotibial joints and collimated from the mid-femur to the mid-tibia 33 .

To determine BMD, all rabbits underwent dual-energy x-ray absorptiometry (DXA) according to previously reported work 34 . DXA was carried out using a Hologic QDR-1000/WTM pencil-beam densitometer with a 1 mm diameter collimator on the x-ray output (Hologic Inc., Waltham, MA, USA). Briefly, BMD was measured at the left knee joint with animals placed in the supine decubitus position. Four subarticular regions, each 0.06 cm², corresponding to the medial and lateral femoral condyles and tibial plateaus, were selected. These regions were located, respectively, 1 mm above and below the joint line at the areas of maximum contact between the femoral condyles and tibial plateaus. The mean BMD values were considered representative of subchondral bone.

Tissues from synovial membranes, cartilage, and tibial subchondral bone were homogenized in liquid nitrogen, and total protein was extracted from the resulting powder by a protein extraction buffer containing 15 mM HEPES, 10% glycerol, 0.5% NP-40, 250 mM NaCl, 1 mM EDTA, 1:1000 PMSF, and a protease inhibitor cocktail (Sigma-Aldrich, St Louis, MO, USA). Protein determination was carried out as described previously 35 , and subsequently 20 μg of total protein from each tissue was resolved on 15% acrylamide-SDS gels. After transfer to polyvinylidene difluoride (PVDF) membranes (Millipore, Molsheim, France) in 48 mM Tris, 39 mM glycine, and 20% methanol at 20 V for 1 h at room temperature, membranes were blocked in 5% skimmed milk in PBS-Tween 20 for 1 h at room temperature and incubated overnight at 4°C with anti-RANKL antibodies (Peprotech, Neuilly-Sur-Seine, France) and OPG (R&D systems, Minneapolis, MN, USA) at 1/1000 dilution each. Antibody binding was detected by enhanced chemoluminescence using peroxidase-labelled secondary antibodies, and the results were expressed as arbitrary densitometric units (AU). Loading control was performed on 15% acrylamide-SDS gels by employing EZBlue' Gel Staining Reagent (Sigma-Aldrich, St Louis, MO, USA).

In addition, the total amount of RANKL resulting from the sum of cartilage, synovium, and subchondral bone expressions was considered as the RANKL global expression in joint tissues. Thus, RANKL global expression was estimated as 100% and the contribution of each joint tissue was estimated in partial percentages.

After sacrifice, femur sections were fixed in 4% paraformaldehyde and further decalcified for 4 weeks in a solution made up of 10% formic acid plus 5% paraformaldehyde. The decalcified knee joints were cleaved in a sagittal plane along the central portion of the articular surface of each medial femoral condyle corresponding to the weight-bearing area, and were then embedded in paraffin block. Sections of 4 μm were stained with safranin-O Fast Green to assess pathological changes in cartilage. These samples were evaluated using a modified version of Mankin's grading score system. The Mankin score has been used by other authors to analyze cartilage damage in arthritic samples 36 . Thus, partial scores evaluating structure abnormalities, cellularity, and matrix staining (tidemark integrity was omitted) were calculated and further combined to give a maximum total score of 21 for each histological section. Each cartilage sample was evaluated twice by two experienced observers, and the mean of two reading scores was used in all statistical analyses. Samples were presented to observers in random order. The observers were blinded with respect to each other's reading, and rabbit group. Assessment of cartilage histology was performed at the weight-bearing surface of the medial femoral condyle because it shows the earliest and most severe histological abnormalities 37 .

Other sections were stained with Alcian blue-PAS and photographed at 4× magnification to evaluate subchondral bone plate thickness. The subchondral bone plate thickness was assessed in five different regions of the femoral condyle, including the pannus-bone interface (regions I and V), subchondral bone under the weight-bearing femoral surface (regions III and IV; Figure 2A), and subchondral bone close to the anterior synovial-bone interface (region II; Figure 2B). A straight line linking both pannus-bone interfaces was divided into five equal fragments to delineate the regions of interest. Then, two tangential lines were drawn: the first at the intersection between articular cartilage and the initial cortical layer of the subchondral bone, and the second at the point where the subchondral bone ends. The distance between these two lines was taken to define the subchondral bone plate thickness in the five regions (Figure 2D).

Paraffin-embedded femur sections of 4 μm were prepared to detect the distribution pattern of cells expressing RANKL and OPG. The sections were incubated with antibodies against OPG (R&D Systems, Minneapolis, MN, USA) and RANKL (Santa Cruz Biotech, Heidelberg, Germany). The antibodies were detected with a biotinylated donkey anti-goat immunoglobulin G (IgG) and visualized with horseradish peroxidase/ABComplex using 3,3'-diaminobenzidine tetrahydrochloride as the chromogen (Dako, Camarillo, CA, USA). The tissues were counterstained and mounted in DPX (VWR International Ltd, Poole, England). The negative controls involved incubation with an IgG isotype.

Following procedures approved by the local Ethical Committee, chondrocytes were isolated from the joint cartilage of OA patients who had undergone knee replacement surgery in the Orthopedic Surgery Department of Fundación Jiménez Díaz. Written informed consent was obtained from all patients. The chondrocytes were obtained after sequential digestion with trypsine (Lonza Group Ltd, Basel, Switzerland) for 15 minutes and collagenase type IV (Sigma-Aldrich, St Louis, MO, USA; 1 g/l) for 6 h, both at 37°C. The chondrocytes were grown to confluence in DMEM (Lonza Group Ltd, Basel, Switzerland) supplemented with 10% fetal calf serum, 60 U/ml penicillin, 60 μg/ml streptomycin, and 2 mmol/l glutamine at 37°C in the presence of 5% CO₂. Simultaneously, peripheral blood mononuclear cells (PBMCs) were isolated from healthy donors by Ficoll gradient cell separation (Histopaque - 1077; Sigma-Aldrich, St. Louis, MO, USA). PBMCs were cultured for 24 h at a density of 2 million cells/well in 6 well-plates in α-MEM supplemented with 10% fetal calf serum, 60 U/ml penicillin, 60 μg/ml streptomycin, 2 mmol/l glutamine, and 40 ng/ml of M-CSF (Sigma-Aldrich, St. Louis, MO, USA) at 37°C in the presence of 5% CO_2.

Subsequently, confluent chondrocytes were re-suspended in α-MEM supplemented with 10% fetal calf serum and co-cultured with PBMCs at a density of 90,000 cells/well for 21 days in the presence of 40 ng/ml M-CSF (Sigma-Aldrich, St. Louis, MO, USA) and where indicated, 10^-6 M PGE₂ (Cayman Chemical, Ann Arbor, MI, USA). Two osteoclastogenic inhibitors, anti-RANKL antibody (Peprotech, Neuilly-Sur-Seine, France) and OPG recombinant protein (Peprotech, Neuilly-Sur-Seine, France) were used to block osteoclast formation in the presence of 10^-6 M PGE₂ in co-cultures. Both the medium and differentiation factors were changed every 48 h. After 21 days of differentiation, osteoclast differentiation was evaluated by staining for tartrate-resistant acid phosphatase (TRAP) using a commercial kit (Sigma-Aldrich, St. Louis, MO, USA). Osteoclasts were identified by the presence of ≥ 3 nuclei and purple color.

All statistical analyses were performed using SPSS version 17.0 software for Windows (SPSS, Chicago, IL, USA), and results were expressed as the mean ± standard deviation (SD). The data from multiple groups were compared using a Kruskal-Wallis nonparametric test, and a pairwise comparison using the Mann-Whitney test was applied when overall differences were identified. Spearman correlation coefficients were calculated. P values less than 0.05 were considered significant.

Radiographic analysis showed that rabbits with AIA had juxta-articular bone loss and symmetrical joint space narrowing in the affected knees. However, knees of OA rabbits had medial joint space narrowing (Figure 1B). The low densities observed on radiographs were confirmed by analyzing subchondral BMD in the knee bones (Figure 1C). In fact, rabbits with AIA had significantly lower BMD than healthy ones (0.27 ± 0.04 vs. 0.55 ± 0.05 g/cm², P = 0.029). Furthermore, rabbits with OA also had a stronger tendency for lower BMD than healthy ones, although the difference was not statistically significant (0.43 ± 0.06 vs. 0.55 ± 0.05 g/cm², P = 0.057). We also observed that BMD was lower in rabbits with AIA than in those with OA, though with limited statistical significance.

Juxta-articular bone loss observed in AIA rabbits was confirmed in histological studies. Indeed, thinner and discontinuous bone trabeculae are seen in AIA rabbits when compared with healthy ones (Figure 2A). To test if that bone loss was higher in the zones of bone invaded by pannus, the subchondral bone plate thickness was evaluated in weight-bearing regions and in bone-pannus interfaces. Our results showed a greater thickness in weight-bearing regions than in bone-pannus interfaces in the healthy group, whereas, in the AIA group, subchondral bone plate thickness loss was homogeneous along the condylar surface (Figure 2C). In addition, while all AIA rabbits had significant pannus associated with high osteoclastogenic activity, none of OA animals presented with pannus or bone erosions. Furthermore, pannus and bone erosions were only seen in the synovial membrane-cartilage interface of AIA animals, but not throughout the remainder of the osteochondral junction, as shown in Figure 2A and 2B.

Cartilage damage in the knees of our experimental models was evaluated using a modified histopathological Mankin grading score 34 . This assessment was performed by two blinded observers, and a high correlation between both evaluations was obtained (r = 0.95). Cartilage samples from AIA and OA rabbits showed total Mankin scores that were significantly higher than those from healthy animals (AIA, 12.21 ± 3.15 vs. healthy 1.56 ± 2.49, P < 0.001; OA, 8.33 ± 2.36 vs. healthy 1.56 ± 2.49, P = 0.036), as seen in Figure 3. Furthermore, cartilage damage was inversely related to subchondral BMD in all rabbits (r = -0.829, P = 0.021; see Figure S1 in Additional file 1).

Our western blot results confirmed that both RANKL and OPG are expressed in healthy cartilage (Figure 4A, B). RANKL and OPG expression in AIA and OA cartilage was also analyzed. The expression of both RANKL and OPG was higher in cartilage from AIA rabbits than in cartilage from OA and healthy rabbits (RANKL in AIA, 876.80 ± 149.52 vs. OA 204.92 ± 52.47, P = 0.017; vs. healthy 64.15 ± 22.26, P < 0.001 (Figure 4A); OPG in AIA 153.79 ± 54.63 vs. OA 53.17 ± 4.61, P = 0.017 vs. healthy 35.63 ± 18.22 AU, P = 0.0001 (Figure 4B). Moreover, the RANKL/OPG ratio showed an increase in AIA cartilage when compared to healthy cartilage (6.20 ± 2.13 vs. 3.85 ± 0.90, P = 0.009; Figure 4C). An increased ratio, however, was not found in synovium (5.21 ± 1.02 vs. 5.14 ± 2.90, P = 0.613; Figure 4C) or subchondral bone (6.56 ± 1.59 vs. 5.88 ± 4.18, P = 0.397) of AIA rabbits when compared with healthy ones. In addition, the RANKL/OPG ratio in articular cartilage was not different from that in synovium in the AIA group (6.21 ± 2.14 vs. 5.21 ± 1.02, P = 0.902). RANKL expression and the RANKL/OPG ratio in articular cartilage were inversely related to subchondral BMD (RANKL: r = -0.891, P < 0.001; RANKL/OPG: r = -0.736, P = 0.01; see Figures S2 and S3 in Additional file 1).

We also analyzed the relative contribution of each tissue to the articular global RANKL production, and we observed that 30% of the total RANKL expressed in an arthritic knee is produced by the inflamed cartilage, a tissue with a poor cellular content compared to synovium or subchondral bone (data not shown).

Immunohistochemical detection confirmed western blot results, which showed a higher expression of RANKL (Figures 5C, D, H and Additional file 2, Figure 1) and OPG (Figure 5G), throughout the articular cartilage in AIA rabbits. In healthy cartilage, the RANKL expression was intracellular and was higher in both superficial and middle zones than in deep cartilage. In the calcified cartilage of these healthy rabbits, a mild expression of RANKL was again appreciable, and most RANKL+ chondrocytes were located around mesenchymal structures (Figure 5A). The same RANKL distribution pattern, though with a higher intensity, was gradually observed in the articular cartilage from OA (Figure 5B) and AIA rabbits (Figures 5C, D, H; see Figure S1 in Additional file 2). In the AIA rabbits, we observed that higher cartilage damage corresponded with a higher RANKL expression. RANKL expression was predominantly intra-cellular. However, a clear extracellular expression of RANKL was also detected in AIA rabbits (Figure 5D), especially near the vessel in the calcified cartilage (Figure 5H). Regarding OPG expression in healthy samples, it was only observed within chondrocytes from the deep zone (Figure 5E). The localization was almost the same for OA (Figure 5F) and AIA rabbits (Figure 5G), although we observed an increase in the intensity of the staining.

Previous studies by our group have demonstrated that PGE₂ induces RANKL expression by chondrocytes in vitro 28 . Accordingly, and in light of the increased activity of PGE₂ in RA, we stimulated chondrocytes with PGE₂ to replicate in vitro micro-environmental conditions occurring in the RA joint. Six experimental co-culture conditions were developed: 1) co-culture of PBMCs with chondrocytes in the presence of M-CSF; 2) co-culture of PBMCs with chondrocytes in the presence of M-CSF and PGE₂; 3) co-culture of PBMCs with chondrocytes in the presence of M-CSF, PGE₂ and anti-RANKL antibody; 4) co-culture of PBMCs with chondrocytes in the presence of M-CSF, PGE₂ and OPG recombinant protein; 5) culture of PBMCs in the presence of M-CSF and PGE₂, and 6) culture of PBMCs in the presence of M-CSF and RANKL. Our results demonstrated that basal RANKL production by chondrocytes was not enough to induce osteoclastogenesis from monocytes in vitro (Figure 6A). However, RANKL produced by PGE₂-stimulated chondrocytes induced differentiation of monocytes into osteoclasts (Figure 6B and 6C). In addition, we have observed that both RANKL inhibitors blocked osteoclastogenesis as no TRAP+ cells were seen in these two culture conditions (data not shown).