Calcium calmodulin kinase II activity is required for cartilage homeostasis in osteoarthritis

WNT ligands can activate several signalling cascades of pivotal importance during development and regenerative processes. Their de-regulation has been associated with the onset of different diseases. Here we investigated the role of the WNT/Calcium Calmodulin Kinase II (CaMKII) pathway in osteoarthritis. We identified Heme Oxygenase I (HMOX1) and Sox-9 as specific markers of the WNT/CaMKII signalling in articular chondrocytes through a microarray analysis. We showed that the expression of the activated form of CaMKII, phospho-CaMKII, was increased in human and murine osteoarthritis and the expression of HMOX1 was accordingly reduced, demonstrating the activation of the pathway during disease progression. To elucidate its function, we administered the CaMKII inhibitor KN93 to mice in which osteoarthritis was induced by resection of the anterior horn of the medial meniscus and of the medial collateral ligament in the knee joint. Pharmacological blockade of CaMKII exacerbated cartilage damage and bone remodelling. Finally, we showed that CaMKII inhibition in articular chondrocytes upregulated the expression of matrix remodelling enzymes alone and in combination with Interleukin 1. These results suggest an important homeostatic role of the WNT/CaMKII signalling in osteoarthritis which could be exploited in the future for therapeutic purposes.


Primary cells, cell lines and tissues.
Bovine chondrocytes were isolated from metatarsal joints obtained from a local abattoir within 24 h from death. Experiments were performed when primary chondrocytes reached 80% confluence (passage 0, P0) as previously described 18 . C3H/10T1/2 cells (ATCC) were cultured according to the manufacturer's instructions.
Murine femoral heads and patellae were explanted from 1 month-old male 129/Sv mice and processed for gene expression analysis (see below).
siRNA preparation and transfection. Small interfering RNA for mouse CAMKIIγ was prepared using the Silencer siRNA Construction Kit (Life Technologies) following manufacturer's instructions. siCaMKIIγ S: AAG GGT TTA AGG GTT TCA CTA CCT CGT GTC CTG TCT C; AS: AAA CAC GAA AAC ACG AGG TAG TGA AAC CCC CTG TCT C; scramble siRNA, S: AAT TCC GCA ATT CCG CTC TGG ACT GTA GTC CTG TCT C; AS: AAG AAC TTA AGA ACT TGA TCA ACC AGA TGC CTG TCT C. The siRNA were used at 20 nM and were transfected in C3H/10T1/2 cells using JetPRIME transfection reagent (Polyplus) as described in 18 .
RNA isolation and quantitative real time PCR. RNA isolation and PCR were performed as previously described 12 . For a complete list of primers and conditions refers to Supplementary Table 1.
Microarray analysis. Total RNA was quantified using a Nanodrop ND-1000 Spectrophotometer and assessed for quality (QC) using the Agilent 2100 Bioanalyzer Pico kit. Five nanograms of total RNA with RNA integrity numbers (RINs) higher than 9 were selected for the microarray study. A Poly-A Spiking Control (Life Technologies) was used as amplification control.
Single primer isothermal amplification (SPIA) technology was used to generate cDNA using NuGen Ovation Pico WTA System V2 kit, following manufacturer's instructions. The SPIA cDNA was subjected to QC, fragmented and biotin-labelled using the NuGen Encore Biotin Module according to the manufacturer's instructions. The processed cDNA was subjected to a further round of QC to assess fragmentation size (Agilent 2100 Bioanalyzer Nano kit; fragment size < 200 nt).
Hybridization cocktails were prepared according to Nugen's recommendations for Human Gene 2.0 ST arrays and hybridization took place at 45 °C for 16-20 h in an Affymetrix Genechip Hybridization oven 645.
The arrays were washed and stained using wash protocol FS450_0002 as recommended by Affymetrix on the GeneChip Fluidics station 450. The arrays were scanned using the Affymetrix GeneChip Scanner. CEL files were QC checked using the Expression Console software package (Affymetrix, Thermo Fisher) by using standard metrics and guidelines for the Affymetrix microarray system. Data were normalised together using the Robust Multi-array Average (RMA) sketch algorithm. The processing of all sample files was performed using R Programming Language (R). In brief, raw data were inputted using the 'read.celfiles' function of the ' oligo' R package 19 , with data then undergoing Robust Multiarray Average (RMA) normalisation using the 'rma' function. For cross-sample comparisons, the study design model was first created by identifying sample groupings based on KN92, KN92-WNT3a, and KN93-WNT3a status.
Raw and normalized microarray data were submitted to the Gene Expression Omnibus database at the National Center for Biotechnology Information (accession number GSE162076). Immunohistochemical analysis. CaMKII isoforms and phosphoCaMKII (pCaMKII) were detected by indirect immunofluorescence as described in 12 . For a list of retrieval methods, antibodies and dilutions please refer to Supplementary Table 2. All images were acquired using identical settings on an Olympus BX61 microscope and consistently modified for best rendering using Adobe Photoshop. The fluorescent and dapi images of the same field were superimposed, or the fluorescent and Nomarski images of the same field were inverted and superimposed (subtraction).
Instability-induced OA in mice. All animal procedures were approved by the UK Home Office (PPL 70-7986). All the experimental procedures were performed in accordance to relevant guidelines and regulations.
The study was carried out in compliance with ARRIVE guidelines. Mice were kept in an approved animal care facility, were housed 4 or 5 per cage in standard cages and fed ad libitum. Ten week old, male C57BL/6 mice (Charles River) were subjected to resection of the medial collateral ligament and of the anterior horn of the medial meniscus (menisco-ligament injury-MLI) as previously described 20 . In each cage, the animals were randomized to receive either PBS or KN93 (see Supplementary Table 3). The contralateral knee underwent sham surgery (arthrotomy but no damage to ligaments or menisci). The joint capsule was sutured with Vicryl 6-0 and the skin with Ethylon 5-0 sutures with atraumatic needles (Ethicon).
Pharmacological blockade of CaMKII. Four weeks after surgery the mice were randomized to receive either the water-soluble version of the CaMKII inhibitor KN93 or PBS as control (15 animals per group). For the first 3 days KN93 (10 µmol/kg/day) or PBS were administered by intraperitoneal injection. Subsequently, ALZET osmotic minipumps (Charles River Laboratories) were inserted subcutaneously on the back of the mice allowing a continuous release of KN93 (5 µmol/kg/day) or PBS for additional 28 days. The pumps were changed once during this time. The animals were then killed by cervical dislocation.
Histology and OA scoring. Knee joints were dissected, the majority of the soft tissue was removed, and the joints were fixed in 70% ethanol following which they were decalcified in formic acid, embedded in paraffin and sectioned as described in 7 . The sections were stained with Safranin O (SO) (0.1%, pH 4). Images were taken using the same settings on an Olympus BX61 microscope and consistently modified for best rendering using GIMP software. The extent of cartilage degradation was scored by two blinded, independent investigators using the Osteoarthritis Research Society International (OARSI) scoring system 21 . At least 5 sections/knee were scored.
Histomorphometry. Proteoglycan loss, osteophyte size and differentiation, and subchondral bone thickness were measured by histomorphometry using ImageJ 22 . The mean values from at least 3, but on average 5 sections from each knee (1 section every 50 µm, spanning the entire depth of the joint) were calculated and used for statistical analysis.
Proteoglycan loss was analysed by densitometry of the articular cartilage as previously described 7 . In brief, images were rotated so the cartilage-bone junction in the middle of the plateau was horizontal. This is important so that non-cartilage staining in the bone marrow spaces can be eliminated. The tibial plateau was selected to include the growth plate, any osteophytes and up to the end of the cartilage; and selections from every section were placed on one canvas, so they were perfectly aligned horizontally. To isolate the metachromatic Safranin O staining from the background and most of the orthochromatic staining (subchondral bone), the RGB image was transformed into an HSB stack. On the 'Saturation' slice, a rectangle 600 microns wide and the full height of the canvas was placed over the first section to include the tibial plateau, making sure to exclude any osteophytes and the "bulgy" part near the intercondylar notch. Ctrl + 1 was pressed, and then the rectangle was moved laterally to the next section and Ctrl + 2 was pressed. This was repeated until all sections were selected and then Ctrl + 3 plotted the density profile of each section. The first peak is the articular cartilage. A horizontal line was drawn to cut residual background if any, and a lateral line to limit the articular cartilage peak and eliminate any staining of bone marrow spaces. Using the wand tool to select the inside of the density profile returned the area. These results were copied from the measurements box to a statistics programme for further analysis.
Osteophytes consistently developed only on the medial tibial compartment of the operated joints. On individual sections, osteophyte size was determined by selecting the osteophyte with the polygon tool, and ensuring a global scale had been set, measuring the area. Osteophyte differentiation was assessed by comparing the intensity of SO staining with the staining of the growth plate, which was set as the threshold for comparison in Image J and defined as the staining for fully differentiated cartilage. The percentage of the thresholded area in each individual section was then normalised for the area of the osteophyte. The points in the graph represent the average of the measurements for multiple sections (at least 3/mouse).
Subchondral bone thickness was determined by selecting the subchondral area underneath the load-bearing area of the articular cartilage of the tibiae as described by Botter et al. 23  Micro-CT analysis, Tibiae epiphyseal volume measurement. The reconstructed scan was rotated and resliced using ImageJ 22 in a coronal orientation. Using CTAn software (Skyscan), the tibial epiphysis was reconstructed as a volume of interest (VOI) by manually selecting the epiphysis on every 10th slice for the entire stack of images. A threshold of 90 density units was set to distinguish mineralised from non-mineralised tissue within the trabeculae.
Statistical analysis. Parametric data were compared with the t-test or Anova with Tukey-Post test for multiple comparisons. When possible, data transformation was applied to satisfy the assumptions of parametric tests as described 24 . Non-parametric data were analysed using the Wilcoxon-Mann-Whitney test. p values < 0.05 were considered significant: *p < 0.05; **p < 0.01; ***p < 0.001. For the analysis of the microarray experiment, individual comparisons within the study design model were conducted by fitting a linear model independently for each probe, with group as the y variable, using 'lmfit' ('limma' R package) 25 . The linear fit for each comparison was subsequently modified using the empirical Bayes (' eBayes') approach, which aims to bring the probe-wise variances across samples to common values, resulting in modified t-statistics, F-statistic, and log odds differ- www.nature.com/scientificreports/ ential expression ratios. Finally, for each comparison, log 2 fold-change (logFC), P value, and corrected P value (false discovery rate, FDR) was output. A cut-off of 10% FDR was used. Volcano plots were generated using 'EnhancedVolcano' ('EnhancedVolcano' R package) 26 .

Results
Transcriptional signature of the WNT3A/CaMKII pathway in the articular chondrocytes. We reported that WNT3A can activate both the WNT/beta-catenin and the WNT/CaMKII pathways in articular chondrocytes with distinct, dose-dependent biological outcomes 12 . This implies that the modulation of some transcriptional targets of WNT3A will be due to the activation of the WNT/beta-catenin-dependent pathway and others due to the activation of the CaMKII pathway. To elucidate the specific transcriptional signature of the CaMKII pathway activation by WNT3A, we performed an expression microarray analysis of human articular chondrocytes treated with WNT3A in presence of KN93 or of its inactive control compound KN92 (Fig. 1ac). We proposed that genes that are activated in a WNT/CaMKII-dependent manner will be upregulated by WNT3A, but only in the absence of the CaMKII inhibitor KN93. This analysis revealed that 44 genes were modulated by WNT3A in the absence of CaMKII inhibitor (Fig. 1d), but only 3 were modulated by WNT3A in a CaMKII-dependent manner (Fig. 1d). The 3 genes modulated by WNT3A in a CaMKII-dependent manner included SRY-Box Transcription Factor 9 (SOX9) (upregulated), the uncharacterized transcript RP11-474G23.3 (downregulated) and Heme-Oxygenase 1 (HMOX1) (downregulated) mRNA. The upregulation of SOX9 confirmed our previous results 12 . The downregulation of HMOX1 mRNA in response to WNT/CaMKII activation was confirmed at qPCR level in human articular chondrocytes (Fig. 1e). Consistent with the requirement for CaMKII activation, KN93 effectively rescued the downregulation of HMOX1 induced by WNT-3A and significantly upregulated its expression of this gene without exogenous WNT3A.
We previously showed that the WNT/beta-catenin pathway antagonizes the WNT/CaMKII pathway in chondrocytes and that, therefore, its suppression using DKK1 results in hyper-activation of the CaMKII-dependent targets 12 . In keeping with this model, treatment of chondrocytes with DKK1 alone decreased the expression of HMOX1 (Fig. 1f) and did not rescue the downregulation of HMOX1 induced by WNT3A, confirming that www.nature.com/scientificreports/ HMOX1 is a negative target of the WNT3A/CaMKII pathway but not of the beta-catenin-dependent WNT pathway (Fig. 1f).

Activated CaMKII and HMOX1 expression are inversely correlated in osteoarthritis.
We next sought to confirm our in vitro findings in human cartilage and human OA pathology. We mined transcriptomics data in publicly available OA cartilage databases 27 . CaMKIIγ and -δ RNA were the most expressed CaMKII isoforms at mRNA levels in normal as well as in osteoarthritic cartilage (Fig. 2c). A similar expression pattern was detected in a previously reported expression microarray in human cartilage explants 6 ( Fig. 2a) and in cultured human chondrocytes through in vitro culture (Fig. 2b). Confirming a previous report 28 , CaMKII phosphorylation was increased in osteoarthritic cartilage as assessed with immunostaining with an antibody directed to the T287 residue (Fig. 2d). In keeping with our in vitro data showing that HMOX1 is a negative CaMKII target, HMOX1 expression decreased in cartilage isolated from OA patients (Fig. 2e). These results therefore suggest that the WNT/CaMKII pathway is activated in osteoarthritic cartilage. Whilst our data show a correlation between CaMKII activation and HMOX1 downregulation, they do not establish causality in vivo and other factors may contribute to HMOX1 suppression in OA.
CaMKIIγ and -δ were the most expressed isoforms in cartilage isolated from murine femoral heads and patellae (Fig. 3a,b), as previously seen in human cartilage. However, the different CaMKII isoforms displayed a tissue specific-expression pattern in the murine joint. CaMKIIγ was expressed in the most superficial layer of the articular cartilage, with some limited expression in the menisci (Fig. 3h-j). CaMKIIβ and -δ were also detectable in the surrounding soft tissues and in the bone marrow (Fig. 3c-g and k-o). To further validate a direct dependence of HMOX1 downstream the activation of CaMKII, we silenced CaMKIIγ in the murine fibroblast cell line C3H/10T1/2 by siRNA (Fig. 3p,q). Silencing CaMKIIγ did not result in compensation from the other isoforms (Fig. 3p) and was sufficient to significantly upregulate HMOX1 mRNA in these cells (Fig. 3q).

Pharmacological inhibition of CaMKII promotes cartilage degeneration and subchondral bone remodelling in a murine model of OA.
To investigate if CaMKII inhibition affects the outcome of OA, we induced OA in adult mice by meniscus-ligament injury (MLI) 20,29 and 4 weeks later we inhibited CaMKII by systemic administration of the CaMKII inhibitor KN93 or PBS as control for additional 4 weeks (Fig. 4a). The weight and general health status of the animals were monitored throughout the entire length of the stimulation and no significant difference between vehicle-treated vs inhibitor-treated animals was noted (Supplementary Table 3). CaMKII phosphorylation was effectively inhibited in KN93-treated animals (Supplementary Figure 1). www.nature.com/scientificreports/ Animals receiving KN93 developed more severe OA, as quantified using the OARSI scoring system (Fig. 4b,c) and a corresponding decrease in proteoglycan content (Fig. 4d,e), measured by histomorphometry. A statistically significant decrease in cell density was detected in the articular cartilage of KN93-treated animals (Supplementary Figure 2). Bone changes were assessed by microCt. As expected, BV/TV of the entire tibial epiphysis increased in the MLI-operated compared to the sham-operated knee, but no difference was observed between treatment groups (Fig. 5a,b). Subchondral bone thickness, however, was significantly less increased in KN93-treated mice (Fig. 5c). Osteophyte size and bone vs cartilage ratio was not affected by KN93 treatment (Fig. 5c).
CaMKII inhibition enhanced the capacity of IL1B to upregulate catabolic enzymes in articular chondrocytes. To gain insight into the mechanisms leading to increased cartilage degeneration in mice treated with KN93, we treated monolayer cultures of primary bovine articular chondrocytes with IL1B, a potent driver of extracellular matrix degradation 5 . CaMKII inhibition with KN93 (Fig. 6a,c,e,g) or AIP (Fig. 6b,d,f,h) induced the upregulation of the matrix-degrading enzymes MMP3 and ADAMTS5 alone and in synergy with IL1B.

Discussion
Deregulation of Wnt signalling is associated with OA in in patients 30 and in animal models 8,9,31 .
We previously showed that WNT3A could simultaneously activate both the beta-catenin dependent and the CaMKII-dependent pathways in chondrocytes, resulting in distinct transcriptional and biological outcomes 12 .
In this study we established the transcriptional signature of WNT/CaMKII signalling and identified HMOX1 as a specific transcriptional target of this branch of the Wnt pathway (see schematic representation in Fig. 7). We showed that CaMKII phosphorylation was increased in human and murine OA, while the expression of HMOX1 was decreased. CaMKII blockade increased cartilage breakdown in an instability-induced OA model in mice. CaMKII inhibition increased the capacity of IL1B to upregulate catabolic enzymes in chondrocytes.
The microarray analysis also confirmed SOX9 mRNA upregulation in response to CaMKII blockade, as we previously showed by qPCR 12 . The mechanism by which CaMKII blockade results in SOX9 upregulation and, simultaneously, in upregulation of cartilage-degrading enzymes is unknown. It could be speculated that www.nature.com/scientificreports/ SOX9 upregulation occurs as a homeostatic response to the increased levels of extracellular matrix degradation. HMOX1 is a cytoprotective enzyme exerting antioxidant effects and chondroprotective effects in vitro [32][33][34] .
The three CAMKII isoforms expressed in the joint had distinct expression patterns. CaMKIIγ was selectively expressed at the articular interface in the cartilage and in the menisci, while -CaMKIIβ and CaMKIIδ had a broader expression pattern. Unfortunately, because of their high similarity across the phosphorylation site, no antibody specifically detecting the phosphorylated form of the different isoforms is available, making impossible to determine if only one isoform or all of them are activated in OA. Interestingly, however, we demonstrated that downregulation of CaMKIIγ alone is sufficient to mimic the effect of KN93 in inducing the upregulation of HMOX1, suggesting that this isoform might be downstream the activation of the WNT/CaMKII in the adult articular cartilage. The understanding of the differences in the activity of the four isoforms is going to be important for targeting CaMKII for therapeutic purposes.
In our in vivo experiments, CaMKII inhibition affected subchondral bone thickness. During embryonic development CaMKII is a modulator of chondrocyte hypertrophic differentiation during development 35,36 and therefore we cannot exclude that altered skeletal morphogenesis may have also contributed to the worsening of the chondropathy.
CaMKII appears to modulate the final outcomes of major homeostatic effects in cartilage homeostasis in a highly context-dependent manner both embryonic development and in adulthood 12,28,35,36 . Saitta et al. also recently reported that CaMKII blockade reduced the capacity of bone morphogenetic proteins (BMPs) to upregulate cartilage phenotypic markers 37 . Calcification and PI turnover play a crucial role in OA progression 38 . It has been reported that the calcium channel TRPM7 links Ca 2+ fluctuations to CaMKII activation and downstream signalling in growth plate chondrocytes during skeletal development 39 . It is possible that a similar mechanism may be responsible for CaMKII activation in OA articular cartilage. The identification of the calcium channels responsible for CaMKII activation in OA may offer a therapeutic opportunity. Therefore, understanding the biology of CaMKII activation is going to be key to successfully harness such pathways in musculoskeletal medicine.