Loss of mutual protection between human osteoclasts and chondrocytes in damaged joints initiates osteoclast-mediated cartilage degradation by MMPs

Osteoclasts are multinucleated, bone-resorbing cells. However, they also digest cartilage during skeletal maintenance, development and in degradative conditions including osteoarthritis, rheumatoid arthritis and primary bone sarcoma. This study explores the mechanisms behind the osteoclast–cartilage interaction. Human osteoclasts differentiated on acellular human cartilage expressed osteoclast marker genes (e.g. CTSK, MMP9) and proteins (TRAP, VNR), visibly damaged the cartilage surface and released glycosaminoglycan in a contact-dependent manner. Direct co-culture with chondrocytes during differentiation increased large osteoclast formation (p < 0.0001) except when co-cultured on dentine, when osteoclast formation was inhibited (p = 0.0002). Osteoclasts cultured on dentine inhibited basal cartilage degradation (p = 0.012). RNA-seq identified MMP8 overexpression in osteoclasts differentiated on cartilage versus dentine (8.89-fold, p = 0.0133), while MMP9 was the most highly expressed MMP. Both MMP8 and MMP9 were produced by osteoclasts in osteosarcoma tissue. This study suggests that bone-resident osteoclasts and chondrocytes exert mutually protective effects on their ‘native’ tissue. However, when osteoclasts contact non-native cartilage they cause degradation via MMPs. Understanding the role of osteoclasts in cartilage maintenance and degradation might identify new therapeutic approaches for pathologies characterized by cartilage degeneration.


Results
Osteoclasts can differentiate on cartilage and 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 (Fig. 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 (Fig. 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 (Fig. 1c). No difference in the expression of classical osteoclast marker genes (ACP5 (TRAP), ATP6V1A 21S, CA2, CLCN7, CTSK, MMP9) was evident between osteoclasts formed on dentine or on cartilage (Fig. 1d). SEM images confirmed that osteoclasts on cartilage were of similar diameter to those on dentine but were more spherical in appearance. The cartilage surface was visibly damaged around osteoclasts despite no visible resorption tracks (Fig. 1e) indicating that osteoclasts do degrade cartilage but, despite having a similar molecular profile, not in the same way as bone matrix.
Chondrocyte regulation of osteoclast differentiation and function depends on the osteoclast substrate. To quantify resorption of 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 (Fig. 2a). Osteoclasts did not release collagen from cartilage ( Supplementary Fig. S1a, b) but showed a non-significant trend to release GAG from both acellular cartilage (chondrocytes killed by freezethawing, Supplementary Fig. S1c) and cellular cartilage containing live chondrocytes (Fig. 2a). No increase in GAG release was evident when osteoclasts were cultured on dentine or plastic with the cartilage in a Transwell insert, suggesting that direct contact between osteoclasts and cartilage is required for proteoglycan degradation (Fig. 2b,c). Interestingly, basal GAG release from cellular cartilage was inhibited when distant osteoclasts (those not in direct contact with cartilage) were cultured on dentine but not plastic (Fig. 2c), suggesting a protective influence of bone-resident osteoclasts on the cartilage matrix via effects of secreted products on active chondrocytes.
Direct co-culture of CD14 + monocytes with chondrocytes increased the number of large osteoclasts (> 10 nuclei) formed on plastic (Fig. 2d), but reduced the number formed on dentine in a manner independent of direct contact (Fig. 2e). This might be partially due to a reduced RANKL:OPG expression ratio in chondrocytes cultured on dentine (Fig. 2f). This suggests that factors secreted by chondrocytes inhibit the formation and activity of bone-resident osteoclasts, but that chondrocytes can stimulate osteoclast formation at non-bone sites. www.nature.com/scientificreports/ MMPs drive osteoclast-mediated degradation of cartilage. We next sought to identify the proteinases responsible for osteoclast-driven degradation of cartilage by inhibiting key bone resorption components. 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 Fig. S2a). Neither inhibitor affected osteoclast-mediated release of GAG from cartilage. However, the pan-MMP inhibitor GM6001 significantly reduced osteoclast-mediated GAG release from acellular cartilage (Fig. 3a), while trending towards the same effect in cellular (Fig. 3b) cartilage. Some of the GM6001-mediated reduction in GAG release in cellular cartilage was due to inhibition of basal release of GAG by GM6001 in the absence of other stimuli ( Supplementary Fig. S2b), confirming that live chondrocytes release MMPs that degrade cartilage. Recombinant human tissue inhibitor of metalloproteinase 1 (TIMP1) also inhibited osteoclast-mediated release of GAG from acellular cartilage (Fig. 3c), establishing the active enzyme as a soluble MMP.
MMP8 and MMP9 drive osteoclast-mediated release of GAG from cartilage. To identify specific MMPs responsible for osteoclast-mediated cartilage degradation, as well as other contributory matrix-degrading genes, we used RNAseq to compare the expression profile of osteoclasts differentiated on cell culture plastic, acellular cartilage and dentine. Osteoclasts on dentine clustered away from the other substrates, suggesting distinct gene expression profiles between bone-resident osteoclasts and those on 'non-bone' substrates (Fig. 4a). www.nature.com/scientificreports/ MMP8 showed the greatest fold upregulation in osteoclasts on cartilage versus dentine (8.89-fold, p = 0.0133) (Fig. 4b, Supplementary Fig. S3, Table 1). No other top-regulated gene had functions suggestive of an involve-  www.nature.com/scientificreports/ ment in cartilage degradation. We therefore explored expression of other members of the large sub-family of soluble MMPs. MMP1, MMP3, MMP13 and ADAMTS1 were upregulated in osteoclasts differentiated on cartilage whereas MMP12 was downregulated. MMP9 mRNA was consistently expressed at a higher level than other MMPs (Fig. 4c). Due to high levels of variability in the expression of these MMPs in osteoclasts on cartilage (which were therefore not picked up as differentially expressed by RNAseq), we focussed our attention on MMP8 and MMP9. Gelatin zymography confirmed production of active MMP8 and MMP9 by osteoclasts cultured on cartilage (Fig. 4d), running at approximately 65 kDa (pro-MMP8 75 kDa) and 82 kDa (pro-MMP9 92 kDa) respectively as indicated by the bands in the positive control lanes. Isoform-specific siRNA-mediated knockdown of MMP8 or MMP9 ( Supplementary Fig. S4a) reduced osteoclast-mediated GAG release from cartilage explants by 39% and 28% respectively (Fig. 4e), although this trend was not significant. GAG release from cartilage alone was unaffected by siRNA ( Supplementary Fig. S4b). Immunohistochemistry of human OA tissue confirmed strong expression of MMP8 and MMP9 in osteoclasts located between the articular cartilage and subchondral bone (Fig. 4f). Chondrocytes surrounded by the cartilage ECM expressed lower levels of both MMPs. www.nature.com/scientificreports/

Discussion
Our data suggests MMP8 and MMP9 as contributory MMPs that drive osteoclast-mediated cartilage degradation and suggests a role for osteoclast-chondrocyte interactions in maintaining joint function. We propose that, in the healthy joint, chondrocytes and bone-resident osteoclasts exert distant effects on the formation and activity of the other cell type to maintain joint integrity. When joint architecture is compromised in conditions such as RA and OA, chondrocytes enhance osteoclast formation within cartilage which drives pathological degradation of the cartilage matrix.
Osteoclasts differentiated on cartilage explants formed large multinucleated cells with the immunophenotype and gene expression profile of osteoclasts on dentine. However, osteoclasts on cartilage had a more rounded three-dimensional profile and did not form the F-actin rings characteristic of sealing zone formation. Human osteoclasts on cartilage in RA also form a less visible ruffled border without actin rings 22 . It may be that minerals are necessary to induce sealing zone formation. Murine osteoclasts on glass coverslips half-coated with apatite only form a sealing zone on the mineralized surface 23 . 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 24 . 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 for osteoclasts to degrade unmineralised cartilage.
Despite the absence of resorption pits, cartilage matrix degradation was evident by SEM. Crumbling of cartilage matrix and fragmentation of collagen fibrils adjacent to osteoclasts is also evident in the mandibles of rat foetuses 4 . We did not observe collagen degradation from cartilage by human osteoclasts in vitro; this may be an issue with the sensitivity of the assay used, a species-specific difference or due to additional osteoclast effects present in the in vivo situation. Cartilage matrix degradation was confirmed by the increased release of cartilage matrix GAG in the presence of osteoclasts, as we described previously 18 .
This study suggests contributory MMPs responsible for cartilage degradation by osteoclasts. Osteoclasts cultured on cartilage secreted active MMP8 and MMP9 and siRNA-mediated inhibition of either MMP showed a trend towards reduced GAG release from cartilage by osteoclasts. Lovfall et al. 21 also described how inhibition of osteoclast bone-resorption pathways affects their ability to degrade cartilage. In agreement with our study, the pan-MMP inhibitor GM6001 inhibited osteoclast-mediated degradation of cartilage, whereas other inhibitors of osteoclast-mediated bone resorption (cathepsin K inhibitors, vacuolar-type H + -ATPase (V-ATPase) inhibitors) had no effect 21,25 . Lack of effect of V-ATPase inhibitors on osteoclast-mediated cartilage degradation supports the hypothesis that acidification is not necessary for osteoclasts to degrade cartilage. Support for this hypothesis also comes from the bone literature; MMP inhibitors only reduce osteoclast-mediated release of collagen from bone when the bone is decalcified 26 .
Expression of MMP8 and MMP9 was observed in osteoclasts in contact with cartilage in OA tissue, as we have previously described for MMP9 18 . Osteoclasts invading the articular cartilage in knee OA also express MMP1, MMP3 and MMP13 8 . We also found these to be upregulated in osteoclasts differentiated on cartilage, suggesting that they might contribute to osteoclast-mediated cartilage degradation and indicating an intriguing avenue for future investigations.
MMP8 was investigated based on its robust overexpression in osteoclasts cultured on cartilage versus dentine. To our knowledge, the gene expression profile of osteoclasts has not previously been compared when they are in contact with their two primary natural substrates. Most osteoclast studies are performed on tissue culture plastic or glass 27 , failing to address the critical role of the substrate in osteoclast differentiation, polarisation and activation. Publications that do study effects of bone on osteoclastogenesis perform comparisons with osteoclasts cultured on synthetic cell culture plastic 28,29 . 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) www.nature.com/scientificreports/ differentiated into osteoclasts on devitalised mouse bone versus plastic 29 . However, differential expression of nonclassical osteoclast genes could explain the different cytoskeletal arrangements. Crotti et al. found that annexin A8 expression increases in murine BMM-derived osteoclasts cultured on bone versus plastic, which regulates cytoskeletal reorganization and F-actin ring formation in osteoclasts on a mineralised matrix 28 . 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 mineralised bone substrate and those in non-bone, unmineralised 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 cellular, but not acellular, cartilage matrix, suggestive of effects of an unidentified osteoclast secreted product(s) on chondrocytes. Similarly, chondrocytes inhibited distant osteoclast formation on bone. However, when osteoclasts were cultured on plastic (a 'non-bone' substrate) they did not exert a protective effect on cartilage matrix and chondrocytes promoted their formation. This suggests a potential homeostatic relationship between osteoclasts and chondrocytes that maintains the integrity of the normal joint but drives matrix degradation in disease conditions where joint integrity has been lost.
Insight into this homeostatic relationship can be obtained from OA. As disease progresses, increased diffusion of small molecules between bone and cartilage allows direct crosstalk between cells via mechanisms including penetration of blood vessels and microcracks into the calcified cartilage and increased hydraulic conductance of the articular cartilage and subchondral bone plate [30][31][32] . Pathways implicated in the crosstalk include the TGF-β/ Smad, Wnt/β-catenin, RANK/RANKL/OPG, and MAPK pathways 33 . The RANK/RANKL/OPG pathway is highly relevant to osteoclasts. RANKL expression increases in human OA cartilage compared with normal cartilage, largely due to its expression by hypertrophic OA chondrocytes 34 . In equine OA, the relationship between articular cartilage RANKL expression and osteoclast density is stronger than in the subchondral bone and correlates with microcrack formation, suggesting that cartilage RANKL recruits osteoclasts which drive cartilage degradation 35 . In C57BL/6J mice with surgically-induced OA, exogenous OPG, an osteoclast inhibitor and decoy receptor for RANKL, protects articular cartilage from OA progression and prevents chondrocyte apoptosis 15 . Further evidence that osteoclasts and chondrocytes cooperate to drive OA comes from experiments using osteoclast-targeting bisphosphonates. Alendronate, which inhibits osteoclast-mediated bone resorption, is chondroprotective in murine models of OA, inhibiting vascular invasion of calcified cartilage and increasing cartilage thickness 13,36 .
This study was limited by the high natural inter-individual variation in both the resorption / digestion capacity of primary human osteoclasts and in the basal rate of cartilage degradation between donors (Fig. 2a). Additionally, the current study focussed on a restricted number of candidate genes that drive osteoclast-mediated degradation of cartilage. Future consideration of more genes-of-interest from our RNAseq and RT-PCR data will provide further insight into the mechanisms driving both contact-dependent and contact-independent effects of osteoclasts on cartilage and chondrocytes, as well as shedding light on why the collagen network in cartilage is not affected by osteoclasts. Beyond this, experiments manipulating expression of these genes in relevant in vivo models will provide an interesting and important next step enabling more detailed dissection of the mechanism(s) described.
In summary, our data demonstrates that osteoclasts can perform proteolysis-driven cartilage degradation via MMPs and that MMP8 and MMP9 might be involved in this process. It provides insight into the complexity of the interactions between bone and cartilage; specifically, how interactions between osteoclasts and chondrocytes depend on whether osteoclasts reside on bone or a non-bone substrate. This data is relevant to future research into a variety of pathological joint conditions where understanding and controlling the activity of osteoclasts presents an attractive option for targeting cartilage degradation.

Preparation of dentine and cartilage. Dentine (elephant ivory; 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) then punching out 4 mm diameter discs. Cartilage tissue was washed and cut into approximately 6 × 6 mm squares. Cellular cartilage pieces (containing live chondrocytes) were maintained in αMEM for 4 days prior to experimental use. To generate acellular cartilage, explants were stored at − 80 °C, thawed, then snap-frozen in liquid nitrogen to kill the chondrocytes. Despite the histological presence of cell remnants in some cartilage lacunae, fluorescence (Ex 560 / Em 590 nm) read after 4 h incubation with 10% Alamar Blue at 37 °C confirmed that chondrocytes in cellular cartilage were metabolically active compared to those in acellular cartilage pieces. www.nature.com/scientificreports/ Histology and immunohistochemistry. Osteoclasts cultured on dentine or cartilage were fixed in 4% formalin. Dentine discs were decalcified in 0.5 M EDTA prior to paraffin-embedding. H&E staining was performed on transverse 5 µm sections. Antigen retrieval of deparaffinised OA tissue sections was performed 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 4 °C and staining was visualised with 3,3′-diaminobenzidine (DAB). Image acquisition was performed using a Zeiss AxioImager MI microscope, AxioCam HRC camera and AxioVision software. Osteoclasts in tissue sections were considered as large, multinucleated cells containing ≥ 3 nuclei; a widely accepted identification criterion 18,37,38 . Characterisation of osteoclasts. Tartrate-resistant acid phosphatase (TRAP) staining of formalin-fixed osteoclasts used naphthol AS-BI phosphate as a substrate, reacting the product with fast violet B salt at 37˚C for 3 h. TRAP-positive multinucleated cells containing 3 ≥ nuclei were considered osteoclasts. Vitronectin receptor (VNR, CD51/61, αVβ3-integrin) expression was determined by immunohistochemistry using a murine monoclonal antibody targeting CD51/61 (MCA757G; AbD Serotec, Oxford, UK) and visualised with DAB. Immunofluorescent staining of F-actin filaments was performed on formalin-fixed cells for 30 min using TRITCconjugated phalloidin (Sigma-Aldrich, Poole, UK), mounted with fluoroshield containing DAPI (Sigma)and visualised at 450-480 nm. Image acquisition was performed using a Zeiss AxioImager MI microscope, Axi-oCam HRC camera and AxioVision software. Quantification was performed using ImageJ software (National Institutes of Health, Bethesda, USA) to count the average number of osteoclasts per field of view. Fields of view were randomly selected using pre-defined and consistent locations within each well, counting 4 fields of view at 4 × magnification per well with triplicate wells per experimental condition. To avoid observer bias, counting was performed while blinded to the image identity.  Table 1. In-house primers were validated by gradient PCR to determine the optimum annealing temperature, then by RT-PCR melt curve analysis to check for primer dimers and a single amplification product. Validated primers produced a single, sharply defined band of the correct size on an agarose gel). Genes were not considered to be expressed at an average cycle threshold (Ct) value over 34. Expression of individual mRNAs was calculated relative to expression of β-actin (ACTB) using the (2-ΔCt) method.

RT-PCR.
Scanning electron microscopy. Dentine slices and cartilage pieces were fixed in 4% formalin, sputtercoated 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.
GAG and collagen release. The 1,

Isolation of chondrocytes.
Chondrocytes were isolated from cartilage tissue by overnight digestion with collagenase (1 mg/ml) and 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 × 10 6 /well) and chondrocytes (4 × 10 4 /well) in 24-well plates. For Transwell experiments, chondrocytes were seeded onto 0.4 μm filter cell culture inserts for 24 h then moved into wells in which CD14 + monocytes 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 and monocytes in the lower chamber. Cultures were treated for 9-10 days with M-CSF and RANKL as for osteoclast differentiation.
RNAseq. RNA  www.nature.com/scientificreports/ and an Agilent 4200 tapestation. Single-indexed and multiplexed samples were run on an Illumina Next Seq 500 sequencer using a NextSeq 500 v2 kit (FC-404-2005; Illumina, Can Diago, CA) for paired-end sequencing. A computational pipeline was written calling scripts from the CGAT toolkit (https:// github. com/ cgat-devel opers/ cgat-flow) 39,40 . 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 41 . 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 42 between three groups: osteoclasts cultured on cell culture plastic, dentine or cellular cartilage.
Gelatin zymography. On day 9 of differentiation, media was changed to FBS-free media containing M-CSF and RANKL. Conditioned media was collected after 24 h, centrifuged, mixed with 2 volumes of Zymogram sample buffer (Bio-Rad) and loaded onto a 10% acrylamide gel containing co-polymerised gelatin (ZY00100BOX; ThermoFisher Scientific). Active recombinant human MMP8 and MMP9 (R&D systems) were used as positive controls. Gels were washed in renaturation buffer (BioRad) and then in development buffer (50 mM Tris, 10 mM CaCl2, 50 mM NaCl, 0.05% Brij 35, pH 7.6) to activate MMPs. Gels were processed for silver staining and scanned using an Epson V700 Photo scanner and Epsonscan software.
MMP8 and MMP9 siRNA. Osteoclasts were differentiated for 7 days before transfection with 25 nM of two separate siRNAs targeting MMP8 (105332, 112911; Life Technologies), MMP9 (104072, 113182; Life Technologies) or a non-specific siRNA control using lipofectamine RNAiMax. Duplexes were removed after 6 h and replaced with osteoclast differentiation medium.
Statistics. Statistical analysis was performed in GraphPad Prism 8 (GraphPad Software, California, USA). D' Agistono Pearson or Shapiro-Wilk were used to test for normality, depending on the sample size. Statistical significance of the unpaired conditions was determined using one-way ANOVA with Tukey's multiple comparison, Kruskal-Wallis ANOVA with Dunn's multiple comparison, Mann-Whitney test or t-test as appropriate. The False Discovery Rate (FDR) method was used when calculating multiple Mann-Whitney tests, with the two-stage Benjamini, Krieger, & Yekutieli procedure for controlling the FDR ie correcting for multiple testing. Results were considered significant at p < 0.05. All figures include at least four independent experiments performed with osteoclasts from different donors (n ≥ 4, each n being the average of 3 technical replicates). Each donor is represented by an individual point on a graph with an indication of either mean ± SD (normally distributed data) or the median (data that is not normally distributed).
Ethics. Use of leucocyte cones 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 were obtained via the Oxford Musculoskeletal Biobank and collected with informed donor consent in full compliance with national and institutional ethical requirements, the United Kingdom Human Tissue Act and the Declaration of Helsinki (HTA Licence 12217, Oxford REC C 09/H0606/11+5, London-Fulham Research Ethics Committee 07/H0706/81).

Data availability
The datasets generated and analysed during the current study are available in the GEO repository (GSE166535) or can be obtained from the authors by reasonable request.