Loss of mutually protective effects between osteoclasts and chondrocytes in damaged joints drives osteoclast-mediated cartilage degradation via matrix metalloproteinases

Osteoclasts are large multinucleated cells that resorb bone to regulate bone remodelling during skeletal maintenance and development. It is overlooked that osteoclasts also digest cartilage during this process, as well as in degradative conditions including osteoarthritis, rheumatoid arthritis and primary bone sarcomas such as giant cell tumour of bone. This study explores the poorly understood mechanisms behind the interaction between osteoclasts and cartilage. Morphologically, osteoclasts differentiated on acellular human cartilage formed multinucleated cells expressing characteristic osteoclast marker genes (e.g. CTSK, MMP9) and proteins (TRAP, VNR) that visibly damaged the cartilage surface by SEM, but without the formation of resorption pits. Osteoclasts caused increased glycosaminoglycan (GAG) release from acellular and cellular human cartilage that was dependent on direct contact with the substrate. Direct co-culture with chondrocytes during osteoclast differentiation increased the number of large osteoclasts formed. When osteoclasts were cultured on dentine, direct co-culture with chondrocytes inhibited osteoclast formation and reduced basal degradation of cartilage. This suggests a mutually protective effect on their ‘native’ tissue between bone-resident osteoclasts and chondrocytes, that is reversed when the joint structure breaks down and osteoclasts are in contact with non-native substrates. Mechanistically, osteoclast-mediated cartilage degradation was inhibited by the pan-MMP inhibitor GM6001 and by TIMP1, indicative of a role for soluble MMPs. RNA sequencing and RT-qPCR analysis identified MMP8 as overexpressed in osteoclasts differentiated on cartilage versus dentine, while MMP9 was the most highly expressed MMP on both substrates. Inhibition of either MMP8 or MMP9 by siRNA in mature osteoclasts reduced GAG release, confirming their involvement in cartilage degradation. Immunohistochemical expression of MMP8 and MMP9 was evident in osteoclasts in osteosarcoma tissue sections. Understanding and controlling the activity of osteoclasts might represent a new therapeutic approach for pathologies characterized by cartilage degeneration and presents an attractive target for further research.


Introduction
Osteoclasts are large multinucleated cells that resorb bone. They are formed by the fusion of CD14+ monocytes and regulate skeletal development and maintenance via homeostatic interactions with bone-forming osteoblasts.
It is often overlooked that osteoclasts also digest cartilage. During skeletal development, endochondral ossification replaces cartilage templates of long bone with mineralized bone following osteoclast-mediated digestion of the calcified cartilage matrix secreted by hypertrophic chondrocytes [1][2][3]. Osteoclasts also degrade the calcified cartilaginous callus during bone fracture healing. In a murine model, mice deficient in osteoprotegerin (OPG), a negative regulator of osteoclasts, degrade the cartilaginous callus faster, with increased osteoclast numbers and faster union of the fractured bone [4].
Dysregulation of the balance between osteoclasts and osteoblasts towards relative osteoclast overactivation results in pathological osteolysis in conditions from osteoporosis and bone cancer to rheumatoid arthritis (RA). RA, a connective tissue disease of the joints, is also characterised by cartilage destruction. Inflammatory cytokines activate fibroblast-like synoviocytes (FLS) and T cells to produce osteoclastogenic macrophage colony stimulating factor (M-CSF) and receptor activator of nuclear factor kappa B ligand (RANKL), promoting the fusion of monocytes to form multinucleated osteoclasts. These osteoclasts drive joint destruction via physical contact with bone and both nonmineralised and calcified cartilage [5,6]. Osteoclasts also invade articular cartilage in human knee osteoarthritis (OA) [7]; OA being the most common disease of cartilage loss.
Inflammatory cytokines such as TNF potentiate the osteoclastogenic effect of RANKL [8], which could explain how TNF-blocking agents protect against bone and cartilage loss in patients with RA [9]. In human TNF-transgenic mice, which develop destructive arthritis associated with enhanced formation of osteoclasts, both an anti-TNF antibody and the osteoclast-inhibiting bisphosphonate zoledronate block bone erosion and cartilage damage and reduce osteoclast formation within the inflamed synovium [10]. Denosumab, a monoclonal anti-RANKL antibody, inhibits osteoclast formation, suppresses bone resorption and causes a reduction in the cartilage turnover marker urine CTX-II/creatinine in patients with RA [11]. Osteoclast-inhibiting bisphosphonates [12,13] and OPG [14] are also chondroprotective in murine models of OA.
Mature osteoclasts have a morphology highly specialised for bone-resorption [15]. Resorption is preceded by osteoclast attachment to the bone matrix, mediated by integrins such as the vitronectin receptor (VNR, v3 integrin). This is followed by polarization and cytoskeletal reorganization; Factin-rich podosomes form a dynamic actin ring which, alongside the integrins, forms a sealing zone that isolates the highly folded bone-facing ruffled border membrane from the extracellular environment [15]. Ultrastructural features of osteoclasts resorbing calcified cartilage exhibit many features common to bone-resident osteoclasts including abundant mitochondria, vacuolation, lysosomes and deep infoldings at points of contact with the calcified matrix [3].
Osteoclasts on bone and those on cartilage (sometimes termed 'chondroclasts' [16]) also have similar basic molecular characteristics. Multinucleated cells form within the inflammatory synovium in RA but only express the calcitonin receptor, a marker of fully differentiated osteoclasts, when they come into direct contact with calcified cartilage or subchondral bone, suggesting that contact with either form of mineralised tissue can direct the final stages of differentiation [17]. Multinucleated cells digesting cartilage in RA, OA and the primary bone tumour giant cell tumour of bone (GCTB) express an osteoclast phenotype (CD14−, HLA-DR−, CD45+, CD68+ CD51+, TRAP+, cathepsin K+ and MMP9+) [18].
However, the mechanism of osteoclast formation on and degradation of cartilage is poorly understood. Osteoblasts produce the necessary cytokines for homing osteoclastogenesis to bone, but chondrocytes can also produce these factors [19,20]. CD14+ monocytes differentiated on healthy human articular cartilage express immunophenotypic markers characteristic of osteoclasts (CD68+, CD14− and CD51+) and degrade cartilage via release of glycosaminoglycans (GAG). Primary human osteoclasts from GCTB and pigmented villonodular synovitis (PVNS) tissue can also release GAG from cartilage [18]. Maintenance of articular cartilage is a balance between anabolic and catabolic pathways with degradation mediated in large part by the matrix metalloproteinases (MMPs), which can also be produced by osteoclasts [7,21].
Despite evidence of a physical interaction between osteoclasts and cartilage in RA, OA and GCTB the mechanism(s) by which osteoclasts degrade cartilage remain unknown. The aim of this study was to enhance our understanding of the interaction between osteoclasts and cartilage; to ascertain whether osteoclasts degrade cartilage using the same cellular machinery as for bone resorption and whether chondrocytes affect this process. If osteoclasts contribute to driving cartilage degradation, understanding and controlling their activity would represent an important perspective for diseases characterized by degeneration of cartilage.

Ethics
Use of leucocyte cones for osteoclast differentiation was approved by the London-Fulham Research Ethics Committee (11/H0711/7). Archival pathological bone and joint specimens were obtained from the Nuffield Orthopaedic Centre, Oxford, UK. Cartilage was obtained at the time of surgery from patients undergoing total knee arthroplasty for OA at the Nuffield Orthopaedic Centre. Samples

Osteoclast differentiation
Peripheral blood mononuclear cells were isolated from human leucocyte cones (NHS Blood and Transplant, Bristol, UK) by density gradient centrifugation. CD14+ monocytes were positively selected using MACS CD14+ microbeads (Miltenyi Biotech, Surrey, UK) and seeded at 0.25 x 10 6 cells/well into 96-well plates containing dentine discs, at 1 x 10 6 cells/well into 48-well plates containing cartilage pieces, or at 1 x 10 6 cells/well into 24-well plates. After overnight adhesion, dentine discs and cartilage pieces were transferred to new wells.

Preparation of dentine and cartilage substrate
Dentine (elephant ivory discs; HM Revenue & Customs, Heathrow Airport, UK) was prepared by cutting 250 μm transverse wafers using a low-speed saw and diamond-edged blade (Buehler, Coventry, UK), out of which 4 mm diameter discs were punched. Cartilage tissue was washed to remove blood and debris and cut to form approximately 6x6 mm squares. To generate acellular cartilage, explants were stored at -80˚C then thawed and snap-frozen in liquid nitrogen to ensure killing of the chondrocytes. Cellular cartilage pieces (containing live chondrocytes) were maintained in MEM for 4 days prior to experimental use. Alamar Blue was used to confirm that chondrocytes in cellular cartilage were metabolically active compared to those in acellular cartilage pieces.

Histology and immunohistochemistry
Osteoclasts cultured on dentine or cartilage were fixed in 4% formalin. Dentines were decalcified in 0.5M EDTA prior to paraffin embedding. H&E staining was performed on transverse 5 m sections.
For immunohistochemistry, antigen retrieval of deparaffinised OA tissue sections was performed by immersion in hot citric acid solution. Sections were exposed to primary rabbit monoclonal antibodies against MMP-8 (ab81286, 1:1000, Abcam, Cambridge, UK) or MMP-9 (ab73734, 1:1000, Abcam) overnight at 4C or a PBS control and staining was visualised with DAB. Image acquisition was performed using a Zeiss AxioImager MI microscope, AxioCam HRC camera and AxioVision software.

Characterisation of osteoclasts
Tartrate-resistant acid phosphatase (TRAP) staining of osteoclasts was performed on formalin-fixed cells using naphthol AS-BI phosphate as a substrate and reacting the product with fast violet B salt at  Table 1). A gene was not considered to be expressed when the average cycle threshold (Ct) value per condition was higher than 34. The expression of individual mRNAs was calculated relative to expression of -actin (ACTB) using the (2-ΔCt) method.

Scanning Electron Microscopy
Dentine slices and cartilage pieces were fixed in 4% formalin, sputter-coated with gold using the SC7620 Mini Sputter Coater System (Quorum Technologies, Lewes, UK) and imaged using a Philips XL 30/SEM with field emission gun.

Isolation of chondrocytes
Chondrocytes were isolated from cartilage tissue by overnight digestion with collagenase (1 mg/ml) then passed through a 70 μm cell strainer (Falcon Becton Dickinson, Oxford, UK). Chondrocytes were grown to 80% confluence and used for experiments up to passage 3.

Co-culture experiments
Direct co-culture was performed with CD14+ monocytes (1 x 10 6 / well) and chondrocytes (4 x 10 4 / well) seeded in 24-well plates in α-MEM. For Transwell experiments, chondrocytes (4 x 10 4 ) were seeded onto 0.4 μm filter cell culture inserts. After 24 h the Transwells were moved into 24-well plates in which CD14+ monocytes (1 x 10 6 / well) had been seeded on cell culture plastic or dentine discs in the lower chamber. For Transwell cartilage experiments, cartilage was placed in the cell culture insert, with monocytes in the lower chamber. Cultures were treated for 9-10 days with M-CSF and RANKL as for osteoclast differentiation. A computational pipeline was written calling scripts from the CGAT toolkit https://github.com/cgatdevelopers/cgat-flow) [24,25]. Sequencing reads were de-multiplexed based on the sample index and aligned to the human genome assembly version 38 (GRCh38) using the STAR (Spliced Transcripts Alignment to a Reference) aligner [26] At least 14 million aligned reads were obtained per sample.
Reads were mapped to genes using featureCounts v1.4.6 (part of the subreads package), in which only uniquely mapped reads were counted to genes. Differential expression analysis was performed using DESeq2 [27] between three groups: osteoclasts cultured on cell culture plastic, dentine or cellular cartilage.

Gelatin zymography
Zymography was used to assess the activity of different MMP isoenzymes. Proteins are separated in sodium dodecyl sulphate (SDS). After 9 days of osteoclast differentiation on acellular cartilage, the media was changed to FBS-free media with M-CSF and RANKL for 24 h.

Osteoclasts can differentiate on cartilage and visibly degrade the cartilage matrix
Multinucleated cells morphologically characteristic of osteoclasts were observed adjacent to visibly eroded mineralised and non-mineralised cartilage in cases of OA, RA and GCTB (Figure 1a). Human osteoclasts differentiated on dentine produced clear resorption pits beneath the cells, visible by both light microscopy and SEM, associated with formation of an F-actin ring (Figure 1b). In contrast, and despite formation of large multinucleated cells expressing the osteoclast markers TRAP and VNR, human osteoclasts differentiated on unmineralised acellular articular cartilage did not form an F-actin ring and did not perform visible erosion by light microscopy (Figure 1c). No difference in the expression of classical osteoclast marker genes (ACP5 (TRAP), ATP6V1A 21S, CA2, CLCN7, CSTK, MMP9) was evident between osteoclasts formed on dentine or on cartilage (Figure 1d). SEM images confirmed that osteoclasts on cartilage were of similar size to those on dentine but were more spherical in appearance. The cartilage surface was visibly damaged around osteoclasts despite no visible resorption tracks (Figure 1e) indicating that osteoclasts do degrade cartilage but, despite having a similar molecular profile, not in the same way as bone matrix.

Regulation of osteoclast differentiation and function by chondrocytes is dependent on the osteoclast substrate
To quantify resorption of the different substrates, we measured osteoclast-mediated release of collagen and GAG. Generation of active osteoclasts was confirmed by release of collagen from dentine during resorption pit formation (Figure 2a (Figure 2a). Increased GAG release was not evident when osteoclasts were cultured on dentine or plastic with cartilage in a Transwell insert, suggesting that direct contact between osteoclasts and cartilage is essential for proteoglycan degradation (Figure 2b, c).
Interestingly, basal GAG release from cellular cartilage was inhibited when distant osteoclasts (i.e. those not in direct contact with cartilage) were cultured on dentine but not plastic (Figure 2c (Figure 2d), but reduced the number formed on dentine in a manner independent of direct contact (Figure 2e). This might be partially due to a reduced RANKL:OPG expression ratio in chondrocytes cultured on dentine (Figure 2f). This suggests that factors secreted by chondrocytes inhibit the formation and activity of bone-resident osteoclasts, but that chondrocytes can stimulate osteoclast formation on non-bone sites.

MMPs drive osteoclast-mediated degradation of cartilage
We next sought to identify the proteinases responsible for osteoclast-driven degradation of cartilage by inhibiting key components of bone resorption. The efficacy of bafilomycin and E64 to inhibit acidification of the resorption lacunae and cathepsin K activity respectively was confirmed in osteoclasts cultured on dentine (Supplementary Figure 2a). Neither inhibitor affected osteoclastmediated release of GAG from cartilage. However, the pan-MMP inhibitor GM6001 significantly ruffled border without actin rings [28]. It may be that minerals are necessary to induce formation of the sealing zone. Murine osteoclasts cultured on glass coverslips half-coated with apatite only form a sealing zone on the mineralized surface [29]. Without a sealing zone, the ruffled border cannot form, and no resorption pits will be produced. Rabbit osteoclasts form resorption pits on human femoral cortical bone but not on demineralised bone, again suggesting that osteoclasts do not polarise in the absence of minerals [30]. The absence of a ruffled border and sealing zone in osteoclasts on cartilage is probably due to differences in its mechanical, physical and chemical properties compared to dentine and suggests that acidification is not necessary to degrade cartilage, as supported by the lack of effect of bafilomycin on osteoclast-mediated release of GAG.
Despite the absence of resorption pits, clear evidence of cartilage matrix degradation was evident by SEM. Crumbling of cartilage matrix and fragmentation of collagen fibrils adjacent to osteoclasts on cartilage is also evident in the mandibles of rat foetuses [3]. Matrix degradation was confirmed by the increased release of cartilage matrix GAG in the presence of osteoclasts, as we described previously [18]. This was observed despite the high level of natural inter-individual variation in both the resorption / digestion capacity of primary human osteoclasts and the basal rate of cartilage degradation between cartilage donors (Figure 2a). This is the first study to identify the specific MMPs responsible for cartilage degradation by osteoclasts. Osteoclasts cultured on cartilage secreted active forms of both MMP8 and MMP9 and inhibition of expression of either MMP with siRNA inhibited the degradation of proteoglycans from cartilage by osteoclasts. Only Lovfall et al [21] have previously described how inhibition of pathways important for osteoclast-mediated resorption of bone affects their ability to degrade cartilage. In agreement with our study, the pan-MMP inhibitor GM6001 inhibited osteoclast-mediated degradation of cartilage, whereas inhibitors of osteoclast-mediated bone resorption (cathepsin K inhibitors, vacuolar-type H+-ATPase (V-ATPase) inhibitors) had no effect [21,31]. The lack of effect of V-ATPase inhibitors on cartilage degradation by osteoclasts supports the hypothesis derived from microscopy studies that acidification is not necessary for osteoclasts to degrade cartilage. Support for our data also comes from work on bone; MMP inhibitors only reduce osteoclast-mediated release of collagen from bone when the bone is decalcified [32]. Immunohistochemistry demonstrated expression of MMP8 and MMP9 in osteoclasts in contact with cartilage in OA patient tissue, as we have previously described for MMP9 [18]. Osteoclasts invading the articular cartilage in knee OA also express MMP1, MMP3 and MMP13 [7], suggesting that other MMPs might also contribute to osteoclast-mediated cartilage degradation.
MMP8 was investigated on the basis of its overexpression in osteoclasts cultured on cartilage versus dentine. This is the first time that the gene expression profile of osteoclasts has been compared when they are in contact with their two primary natural substrates. Most osteoclast studies are performed on tissue culture plastic or glass [33] and fail to address the critical role of the substrate in osteoclast differentiation, polarisation and activation. The few publications to study effects of the bone substrate on osteoclastogenesis have performed comparisons with osteoclasts cultured on synthetic cell culture plastic [34,35]. We found no difference in the expression of classical osteoclast genes on a bone versus cartilage substrate, an observation mirrored in murine bone marrow macrophages (BMM) differentiated into osteoclasts on devitalised mouse bone versus plastic [35].
Differential expression of non-classical osteoclast genes could explain the different cytoskeletal arrangements observed on bone versus cartilage. Crotti et al found that expression of annexin A8 increases in murine BMM-derived osteoclasts cultured on bone versus plastic, which regulates cytoskeletal reorganization and formation of the F-actin ring in osteoclasts generated on a mineralised matrix [34]. Principal components analysis revealed osteoclasts on plastic to cluster, to some extent, with osteoclasts on cartilage and away from osteoclasts on bone, suggestive of a broad distinction between the gene expression profile of osteoclasts on their native bone substrate and those in non-bone environments.
The distinction between bone-resident and non-bone-resident osteoclasts becomes evident when considering interactions between chondrocytes and osteoclasts. Bone-resident osteoclasts exerted a protective effect on the cellular, but not acellular, cartilage matrix, suggestive of distant effects of an unidentified osteoclast secreted product(s) on active chondrocytes. Similarly, chondrocytes inhibited the number of distant osteoclasts formed on bone. However, when osteoclasts were cultured on plastic (a 'non-bone' substrate) these osteoclasts exerted no indirect, protective effect on cartilage matrix and chondrocytes promoted their formation. This suggests a potential homeostatic relationship between osteoclasts and chondrocytes that maintains joint integrity in the normal joint but drives matrix degradation in disease conditions where joint integrity has been lost.
Some insight into this homeostatic relationship can be obtained from OA. As disease progresses in murine models of OA, increased diffusion of small molecules between bone and cartilage allows direct crosstalk between the cell types due to mechanisms including the penetration of blood vessels and microcracks into the calcified cartilage and increased hydraulic conductance of the articular cartilage and subchondral bone plate [36][37][38]. Pathways implicated in the general crosstalk between cartilage and bone in OA include the TGF-β/Smad, Wnt/β-catenin, RANK/RANKL/OPG, and MAPK pathways [39]. Considering effects on osteoclasts, the RANK/RANKL/OPG pathway is potentially most interesting. RANKL expression is increased in human OA cartilage compared with normal cartilage, largely due to its expression by hypertrophic OA chondrocytes [40]. In equine OA, the relationship between RANKL expression in the articular cartilage and osteoclast density is stronger than in the subchondral bone and correlates with the number of microcracks, suggesting that cartilage RANKL recruits osteoclasts which contribute to cartilage degradation [41]. In C57BL/6J mice with surgically induced OA, exogenous OPG, a natural decoy receptor for RANKL and osteoclast inhibitor, protected articular cartilage from the progression of OA and also prevented chondrocyte apoptosis [14]. Further evidence that osteoclasts and chondrocytes cooperate to drive OA comes from experiments using osteoclast-targeting bisphosphonates. Alendronate, primarily used to target osteoclast-mediated bone resorption, is chondroprotective in murine models of OA, inhibiting vascular invasion of calcified cartilage and increasing cartilage thickness [12,42].
In summary, our data demonstrates that osteoclasts can perform proteolysis-driven cartilage degradation and that MMP8 and MMP9 are involved in this process. It also provides new insights into the complexity of the interactions between bone and cartilage; specifically, regarding how interactions between osteoclasts and chondrocytes depend on whether osteoclasts reside on bone or a non-bone substrate. This data is relevant to a variety of pathological joint conditions where understanding and controlling the activity of osteoclasts presents an attractive option for targeting cartilage degradation.