Frzb^-/- mice were generated in our research group 7 and back-crossed into the C57Bl/6J background for over 10 generations. Genotypes were determined as described 7. Six-week-old male Frzb^-/- and wild-type mice were sacrificed by cervical dislocation. The articular cartilage and subchondral bone from the tibial plateau of the knee joint of the hind limb was carefully dissected in one piece at the growth plate region using micro-dissection forceps, a procedure easy to perform at this age when the growth plate is not yet closed. The tissues were immediately snap frozen in liquid nitrogen and stored at -80°C until further processing or used for histology. Animal procedures were approved by the Ethical Committee for Animal Research, KULeuven.

Per microarray, articular cartilage and subchondral bone from a single joint were used. Samples were homogenised using the Fastprep-24 tissue-homogeniser (MP Biomedicals, Solon, OH, USA) in lysing matrix A tubes and RLT lysis buffer (RNeasy Fibrous Tissue kit (Qiagen, Chatsworth, CA, USA)). Samples were kept under pre-cooled conditions using the CryoPrep Adaptor. RNA was isolated with the RNeasy Fibrous Tissue kit (Qiagen) with proteinase K and deoxyribonuclease (DNaseI) treatment. RNA concentration and purity were assessed with a NanoDrop Spectrophotometer (NanoDrop Technologies, Centreville, DE, USA) and integrity was determined using RNA nanochips and the Agilent 2100 Bio-analyzer (Agilent Technologies, Diegem, Belgium). Only non-degraded RNA without impurities (RNA integrity number > 7.7), was considered for microarray analysis.

Transcriptional profiles of three Frzb^-/- and three wild-type samples were analyzed by the VIB Microarray Facility 19. Per sample, 2 μg of total RNA spiked with bacterial RNA transcript positive controls (Affymetrix, Santa Clara, CA, USA) was converted to double stranded cDNA. Subsequently, the sample was converted and amplified to antisense cRNA and labeled with biotin. A mixture of purified and fragmented biotinylated cRNA and hybridisation controls (Affymetrix) was hybridised on Affymetrix GeneChip Mouse Genome 430-2.0 arrays followed by staining and washing in a GeneChip fluidics station 450 (Affymetrix). To assess the raw probe signal intensities, chips were scanned using a GeneChip scanner 3000 (Affymetrix). Microarray data have been deposited in the Gene Expression Omnibus (GEO) 20 and are accessible through Gene Expression Omnibus accession number GSE33656.

Proteins were isolated from the dissected articular cartilage and subchondral bone pieces using cell extraction buffer (Invitrogen, Merelbeke, Belgium) supplemented with 1 mM phenylmethanesulfonyl (Sigma-Aldrich, Bornem, Belgium) and 5% protease inhibitor cocktail (Sigma-Aldrich) using the Fastprep-24 tissue homogeniser (MP Biomedicals). A total of 20 μg of each sample was denatured and separated on a 4 to 12% polyacrylamide Bis-Tris gel (Invitrogen) by electrophoresis using NuPage MES SDS Running buffer (Invitrogen). Proteins were transferred to a PVDF (polyvinylidene difluoride) membrane (Millipore, Brussels, Belgium). Non-specific binding sites were blocked using 5% blottoB (Santa Cruz Biotechnology, Santa Cruz, CA, USA) in Tris-buffered saline with 0.1% Tween (TBS/T) for one hour at room temperature. Blots were probed overnight at 4°C with the following antibodies: 1/500 dephospho-β-catenin (CTNNB1) sheep antibody (Genway Biotech, San Diego, CA, USA), 1/1,000 phospho-Smad1(Ser463/465)/Smad5(Ser463/465)/Smad8(Ser426/428) rabbit antibody (Cell Signaling Technology, Danvers, MA, USA), 1/500 anti-SFRP1 rabbit antibody (Abcam, Cambridge, UK), 1/500 mouse DKK2 affinity purified polyclonal goat antibody, 1/1,000 mouse SFRP2 affinity purified polyclonal goat antibody (both from R&D Systems, Minneapolis, MN, USA) or 1/4,000 anti-GAPDH mouse monoclonal 6C5 (Ambion, Applied Biosystems) in 5% bovine serum albumin in TBS/T with 0.02% sodiumazide. Horseradish peroxidase-conjugated donkey anti-sheep (1/5,000), mouse anti-rabbit (light chain specific) (1/5,000), donkey anti-goat (1/5,000) and goat anti-mouse (1/50,000) polyclonal antibodies (Jackson ImmunoResearch Laboratories, West Grove, PA, USA) in 5% blottoB in TBS/T were used as secondary antibodies. Blots were visualised using Western Lightning Chemiluminescent Substrate (Perkin Elmer Life and Analytical Sciences, Inc., Waltham, MA, USA) for dephospho-β-catenin, DKK2, SFRP2, SFRP1 and GAPDH or SuperSignal West Femto Maximum Sensitivity Substrate (Pierce, Thermo Scientific, Rockford, IL, USA) for phosphorylated Smad. Densitometry analysis was performed with ImageJ Software (NIH Image, National Institutes of Health, Bethesda, MD, USA 21).

ATDC5 cells were cultured in maintenance medium (1:1 Dulbecco's modified Eagle's medium (DMEM):Ham's F-12 mix (Gibco Life Technologies, Gent, Belgium), 1% antibiotic-antimycotic (AB) (Gibco), 5% fetal bovine serum (FBS) (Gibco) containing 10 μg/ml human transferrin and 30 mM sodiumselenite (Sigma-Aldrich) and maintained in a humidified atmosphere of 5% CO₂ and 95% O₂ at 37°C.

In FRZB overexpression experiments, ATDC5 cells were transfected with control pcDNA3.1+ (Invitrogen) or the pcDNA3.1-full length FRZB construct (pfrzb 17) using lipid-based agent Fugene HD (Roche Diagnostics, Vilvoorde, Belgium). After 24 hours, selection with 1 mg/ml geneticin (Gibco) was initiated. Selection medium was renewed every day for 14 days. Antibiotic resistant cells were dilution-cloned.

In Frzb knock-down experiments, ATDC5 cells were transfected with control pGIPZ-non-silencing shRNAmir (Open Biosystems, Thermo Scientific IT IS Open Biosystems, Thermo Scientific, Lafayette, CO, USA) or with a pGIPZ-shRNAmir directed against Frzb (Open Biosystems) using lipo-polymeric agent Arrest-In (Open Biosystems). After 24 hours, selection with 0.5 μg/ml puromycin was initiated. Selection medium was renewed every day for seven days. Antibiotic resistant cells were dilution-cloned.

Stably-transfected ATDC5 cells were grown in micro-masses to undergo chondrogenesis. Three drops cell suspension (2 × 10⁵ cells) were placed in a single well of a standard 12-well culture plate. The cells were allowed to adhere for two hours at 37°C, then 1 ml maintenance medium was added to each well. Geneticin or puromycin pressure was maintained during chondrogenesis.

Micro-masses were cultured in the maintenance medium containing an ITS premix (10 μg/ml insulin, 5 μg/ml human transferrin and 30 mM sodiumselenite) (Gibco) and 5 μg/ml human transferrin for two weeks. The mineralization phase was induced using α-MEM medium (Gibco) containing 5% fetal bovine serum (Gibco), ITS premix, 5 μg/ml human transferrin and 7 mM beta-glycerolphosphate (Sigma-Aldrich) from Day 14 until Day 21. Each condition was performed in triplicate. Total RNA from micro-masses was isolated after 7, 14 or 21 days in culture using the Nucleospin RNA II kit (Macherey-Nagel, Düren, Germany).

Protein extraction of the micro-masses stably overexpressing FRZB or controls after seven days was performed using cell extraction buffer supplemented with 1 mM phenylmethanesulfonyl and 5% protease inhibitor cocktail, followed by quantification using the Pierce BCA Protein Assay kit (Thermo Scientific).

Some ATDC5 micro-masses were fixed in 95% ice-cold methanol for staining. For Picrosirius Red, micro-masses were stained for one hour in Picrosirius Red (0.1% Direct Red 80 (Sigma-Aldrich) in a saturated aqueous solution of picric acid), washed three times with 0.5% acetic acid in water and air-dried. For Safranin O, micro-masses were stained for one hour in Safranin O (1% alcoholic solution (Klinipath, Olen, Belgium)), washed three times with water and air-dried. Quantification of the staining was performed by dissolving the micro-masses with 1 M NaOH (for Picrosirius Red) or 6M Guanidine-HCl (for Safranin O) (both from Sigma-Aldrich) and by measuring the absorbance at 540 and 512 nm respectively with the Infinite M200 (Tecan, Männedorf, Switzerland).

Complementary DNA was synthesised from 1 μg of RNA isolated from tibia articular cartilage and subchondral bone pieces or ATDC5 cell micro-masses using the RevertAid H minus First Strand cDNA synthesis kit (Fermentas GmbH, St-Leon-Rot, Germany). TaqMan gene expression assays (Applied Biosystems, Carlsbad, CA, USA) or the SYBRgreen master mix system (Fermentas) were used to verify differential expression of Frzb (Mm00441378_m1), Sfrp1 (Mm00489161_m1), Sfrp2 (Mm0485986_m1), Dkk2 (Mm00445025_m1), aggrecan (forward 5'-GCTGCAGTGATCTCAGAAGAAG-3', reverse 3'-GATGGTGAGGGAAGACCCTA-5'), Col3a1 (forward 5'-TTATTCTCCCCAATTCGACTCA-3', reverse 3'-AGATCCAGGATGTCCAGAAGAA-5'), Col5a1 (forward 5'-CGGATGTTGCCTACCGAGT, reverse 3'-ACGGTTGTCAGGATGGAGAA-5') and Col2a1 (Mm01309565_m1) (forward 5'-CCAGGATGCCCGAAAATTAG-3', reverse 3'-TTCTCCCTTGTCACCACGAT-5'). For TaqMan assays analysis was performed using the PerfeCTa qPCR FastMix UNG (Quanta Biosciences, Gaithersburg, MD, USA) using the following conditions: 1 minute at 95°C, 40 cycles of 3 seconds of denaturation at 95°C, followed by 20 seconds of annealing-extension at 60°C. All experiments were performed in duplicate. For SYBRgreen, quantitative analysis was performed as follows: 10 minutes at 95°C, 40 cycles of 15 seconds of denaturation at 95°C, followed by 60 seconds of annealing-extension at 60°C. Melting curve analysis was performed to ensure amplification of a specific product. The Corbett Rotor-Gene 6000 (Corbett Research, Westburg, Leusden, The Netherlands) was used for both systems. Results are expressed using the comparative threshold method 22 and were normalised to housekeeping gene Hprt (hypoxanthine guanine phosphoribosyl transferase) (Mm00446968_m1 or forward 5'-TGCTGACCTGCTGGATTACA-3', reverse 3'-TATGTCCCCCGTTGACTGAT-5').

Rib and sternum chondrocytes were isolated from three six-week-old wild-type and three Frzb^-/- mice, as described with minor modifications 23. The sternum was longitudinally cut, followed by complete removal of the ventral part of the ribcage. The ribcage was washed three times in Dulbecco's phosphate buffered saline (DPBS) (Lonza, Verviers, Belgium) with 1% AB (Gibco). Soft tissues were digested in 3 mg/ml collagenase D (Roche Diagnostics) in medium (DMEM, 1% AB and 1% sodium pyruvate (Invitrogen)) for 1 h standing upright in a collection tube in humidified atmosphere of 5% CO₂ and 95% O₂ at 37°C, followed by rotation for a further 1.5 h. Soft tissues were carefully removed, followed by further digestion in fresh 3 mg/ml collagenase D in medium when the soft tissues kept adhering. After washing twice in DPBS with 1% AB, cartilage was digested using 1 mg/ml collagenase D in medium overnight in a petri dish in the incubator. The medium containing chondrocytes was transferred to a collection tube. The bones were rinsed with complete growth medium (10% FBS (Gibco)) and this was also transferred to the collection tube. After centrifugation, cells were resuspended in 4 ml complete growth medium, plated on a T25 plate (Greiner Bio-One, Frickenhausen, Germany) and grown until confluent. The medium was changed every two days. For the proliferation assay, chondrocytes from three Frzb^-/- and three wild-type mice were plated at different cell densities (500, 2,000 or 4,000 cells/well) in triplicate on fluorescence compatible 96-well flat bottom plates (μClear-plate, black, 98-well, Greiner Bio-one). Fluorescence was measured 24 h and 1 week after plating using the CyQuant NF Cell proliferation kit (Molecular Probes, Invitrogen) and the Wallac Victor 1420 Multilabel counter (Perkin Elmer) at an excitation wavelength of 485 nm and emission of 535 nm. The difference in fluorescence between the two time points (24 h and 1 week) was calculated and considered the amount of proliferation in that time window. A different plate was used for each time point.

The quality of hybridization and data acquisition was assessed by RNA-degradation plots, histograms of the perfect match values distribution and quality control graphs. Data were pre-processed by removal of the hybridisation, labeling control and absent probe sets, followed by a log2 transformation and normalisation of the results to obtain the Robust Multiarray Averaging (RMA) algorithm defined expression values and the Microarray Analysis Suite (MAS) 5.0 software detection calls. Significant differences in gene expression were defined using a modified t-test by the limma package from Bioconductor 24 followed by Benjamini-Hochberg multiple testing correction. For further analysis, we used the PANTHER 25, DAVID 26 and GSEA 27 tools 282930313233.

PANTHER uses pathways compiled by experts and determines the representation of a specific pathway on the selected gene list by applying a binomial statistic to which we applied an additional false discovery rate (FDR) test. Only pathways that included at least 15 annotated genes were taken into consideration. With DAVID we interrogated representation in KEGG 34 and Biocarta pathways 35. It uses a modified Fisher's exact test and applies a Benjamini-Hochberg multiple testing correction. The GSEA system uses all data in the microarray analysis in a ranked list and compares a maximal enrichment score to a series of 1,000 random permutations resulting in nominal P-values and FDR q-values. For GSEA analysis, the KEGG curated pathway set, the miRNA motif and transcription factor motif gene sets were used applying 1,000 permutations defined by the gene set. A weighed enrichment statistic using log2-ratio of classes was applied. A stringent limit with a nominal P-value < 0.001 and a FDR q-value < 0.01 was applied. In addition, we compiled a list of WNT target genes based on the WNT homepage 36 (see Additional file 1) and used a Yates corrected Chi-square test to compare our selected gene lists with the reference list. Other datasets were analyzed using a Mann-Whitney test for unpaired samples.

In silico promoter analysis of the Col3a1, Col5a1 and Col5a3 genes was performed using the TFSearch 37 and ALIBABA 38 online software, based on the TRANSFAC algorithm. Stringent criteria were applied so that only the responsive elements with a high homology to the consensus sequence matched our search (> 90%). Additionally, TCF/LEF responsive elements, specific transcription factors associated with WNT signaling, were investigated using the different consensus sequences as previously identified 39.

We were able to dissect the subchondral bone and articular cartilage in one piece (Figure 1A). The heatmap of the RMA expression values from the microarray analysis showed clustering of the transcriptomes into groups formed by the three wild-type and two out of three Frzb^-/- mice, respectively (Figure 1B). The third presumed Frzb^-/- mouse clustered with the wild-types and was subsequently identified by re-genotyping as a heterozygous animal. This sample was not used in the analysis. A total of 697 probe sets out of 30,590 that had a "present" detection call were significantly up-regulated in the Frzb^-/- samples and 1,524 were significantly down-regulated as compared to the wild-type mice (defined by a P-value < 0.01 after Benjamini-Hochberg correction and |log2|-ratio > 1). Cartilage-specific and bone-specific genes were found in the highest percentiles of expressed genes in the microarray analysis, whereas genes specifically related to T cells, B cells and platelets were found in lower percentiles; possibly from RNA originating from the subchondral bone marrow (Figure 1C).

Using the PANTHER resource, 493 mapped genes were identified as up-regulated and 905 mapped genes were identified as down-regulated in Frzb^-/- mice. The 25 genes with the largest fold-difference between Frzb^-/- and wild-type mice are presented in Table 1. A complete list of all regulated genes and fold differences can be found in the additional materials (see Additional file 2).

Different bioinformatics tools were used for analysis of the large dataset with emphasis on the identification of pathways differentially regulated between the Frzb^-/- and wild-type mice. The PANTHER pathway analysis is shown in Table 2. Among the up-regulated pathways the ECM-associated integrin pathway, the cadherin pathway, as well as WNT signaling, were most striking from a biological perspective. Down-regulated pathways pointed towards inflammation and immune cascades, the cell cycle, p53 activation and again integrins (Table 2). Associations of the differentially regulated gene set using databases defining "biological processes" as analysed by PANTHER are shown in the additional materials (see Additional file 3).

We also applied the DAVID bioinformatics tools specifically interrogating gene representation in KEGG and Biocarta databases. Again, pathways associated with WNT signaling, cell adhesion and ECM interactions were most prominent among the up-regulated gene sets and appeared relevant from a biological perspective (see Additional file 4). Members of transforming growth factor-beta (TGFβ) superfamily signaling, including bone morphogenetic proteins (BMPs), were also up-regulated. Pathways among the down-regulated gene list were again linked to p53 signaling and the cell cycle, and to different systems associated with immunity and inflammation. The GSEA analysis further confirmed positive associations between Frzb^-/- mice and ECM interactions as well as negative associations with the cell cycle (see Additional file 5). No miRNAs were associated with the Frzb^-/- or wild-type phenotype using the stringent limit (nominal P-value < 0.001 and FDR q-value < 0.01). Only miRNA-147 had a nominal P-value < 0.001 and a FDR q-value < 0.25 (0.17). This miRNA has been associated with WNT and ECM pathways 40 (Table 3). In the transcription factor analysis, motifs associated with Foxd1, Znf238 and Pbx1 had nominal P-values < 0.001 and FDR q-values < 0.05. Foxd1 has been suggested as a WNT target gene in the developing chick retina 41 (Table 3). In addition, two motifs without specific transcription factor association were also enriched with P-values < 0.001 and FDR q-values < 0.05 (Table 3). Genes overexpressed in the wild-type mice compared to the Frzb^-/- mice were associated with different members of the E2F family of transcription factors applying the stringent criteria. E2F1 has been negatively associated with WNT signaling 42.

We focused on a detailed analysis of changes in the WNT, the integrin/cadherin/ECM and the cell cycle pathways. Many genes mapped in the down-regulated inflammation-associated signaling systems were specifically linked to immune cell populations present in the bone marrow and were not further taken into account for this study.

The WNT pathway gene set demonstrated up-regulation of different extracellullar WNT antagonists in the Frzb^-/- mice as compared to wild-types. These genes belonged to the SFRP/FRZB-family, to the DKK family and to a group of intracellular WNT pathway modulators (Table 4 - compiled from PANTHER and DAVID analysis). Different frizzled (FZD) receptors were up-regulated and there was evidence for activation of both canonical and non-canonical signaling with increased expression of target genes, such as Rspo2, Wisp2, Sox17, Tbl1x and Acta2, and of intracellular messenger molecules Nfatc2 and 4 that are activated in the calcium-dependent WNT pathway (Table 4).