We confirm that all donors provided written, informed consent, and the study was approved by the Regional Committee for Ethics in Medical Research, Southern Norway, approval numbers 2009/742 and 2019/906. We confirm that all research was performed according to relevant guidelines and regulations. No tissues were obtained from prisoners. Articular cartilage was obtained from discarded tissue of three patients with primary OA undergoing knee replacement surgery at the Department of Orthopedic Surgery, Lovisenberg Diakonale Hospital, Oslo, Norway. Cartilage pieces were taken from a part of the surface of the femoral condyle which did not show macroscopic signs of OA. The cartilage tissue was cut and digested as previously described (Fernandes et al., 2013), then resuspended in culture medium: DMEM/F12 GlutaMAX medium (Gibco) supplemented with 10% hPLP, modified after Schallmoser et al. (Schallmoser and Strunk, 2009), 2 IU/mL heparin (Wockhardt), 100 units/mL penicillin and 100 µg/mL streptomycin (P/S) (Sigma-Aldrich), 2.5 µg/mL amphotericin B (Sigma-Aldrich), and 10 ng/mL bFGF (Gibco). The culture medium was changed every 3–4 days. Amphotericin B was discontinued after 1 week. Cells were cultured until passage 2-3.

BM aspirates were obtained from the iliac crest of three healthy donors as previously described (Shahdadfar et al., 2005) and modified after Jakobsen et al. (Jakobsen et al., 2014) In brief, BM aspirates were obtained in heparin-coated syringes, immediately diluted 1:4 with DMEM-F12 (Gibco) and centrifuged on a density gradient (Lymphoprep, Axis Shield). The mononuclear layer was washed twice and seeded into 175 cm²-culture flasks (Nunc). Cells were allowed to adhere for 72 h before the first medium change. The medium was then changed every 3–4 days. MSCs were further cultured in the same culture medium as described for the chondrocytes. Cells were cultured until passage 3.

Chondrocytes and MSC were differentiated using the same chondrogenic differentiation medium (CDM): DMEM/F12 GlutaMAX supplemented with 10 ng/mL transforming growth factor β1 (TGFβ1) (R&D systems), 500 ng/mL bone morphogenetic protein 2 (BMP2) (InductOs), 0.1 µM dexamethasone (DexaGalen, GALENpharma), 1% insulin-transferrin-sodium selenite media supplement (Sigma-Aldrich), 0.1 mM ascorbic acid 2-phosphate (Sigma-Aldrich), 1.25 mg/mL human serum albumin (Octapharma), 4.5 g/L glucose (B. Braun), 1 mM sodium pyruvate (Gibco), 40 µg/mL proline (Sigma-Aldrich) and P/S.

For cartilage disc preparation, 500,000 cells were resuspended in 150 µL of CDM and seeded in 6.5 mm polycarbonate transwell inserts in 24-well plates (Corning) as described previously (Frerker et al., 2021). The plates were centrifuged for 5 min at 200 g, and 700 µL of CDM was carefully added to the bottom wells. Inserts were transferred from 24-well plates to 6-well plates on day one of differentiation culture using custom-made adaptors, and 5 mL CDM was carefully added to each well. The small volume in the insert top was changed every second day, whereas the medium in the outer wells was changed every 4–5 days. After 10 days, cartilage discs were stripped off the transwell membrane and allowed to float freely in medium in 6-well plates, which were placed on a shaker that rotated at 65 rpm. To strip the discs from the transwell membrane, a sterile blade was used to partially cut around the membrane periphery on the bottom of the insert, trying to keep parts of the membrane attached to the insert. Then the membrane was carefully peeled off using sterile forceps, and the cartilage disc was released into the medium. Cartilage discs were harvested after 4 weeks of chondrogenic differentiation, washed with PBS, and further treated as described below. Determination of wet weight was done, removing excessive fluid by briefly holding the edge of the cartilage disc on a tissue paper before weighing on a fine scale (METTLER TOLEDO; AG204 DeltaRange).

Proliferation and apoptosis were investigated by western blot analysis of PCNA and Caspase-3 using monolayer cells and discs from day 1 and day 2. 0.5 × 10⁶ monolayer cells and discs made from 0.5 × 10⁶ cells were lysed in 100 µL 1x Laemmli buffer and boiled at 95°C for 10 min. Samples were stored at −20°C until analysis were performed. 20 µL of cell lysates and 10 µL of Caspase-3 control extracts (#9663; Cell Signalling Technology) were loaded onto 4%–20% gradient polyacrylamide gels (Bio-Rad). Proteins were separated by gel electrophoresis, transferred to PVDF membranes (Bio-Rad) and incubated with appropriate antibodies simultaneously. Prior to adding primary antibodies, membranes were blocked in 5% milk (skim milk powder; MERCK) in TBS-T (TBS, BioRad and Tween 20, Sigma) for 1 h at RT. Primary antibodies were anti-PCNA antibody (ab18197; Abcam; 1:1,000) (Du and Gao, 2021), Caspase-3 antibody (#9662; Cell Signalling Technology; 1:2,000) (Zhao et al., 2021b) and anti-beta actin (ab8226; Abcam; 1:1,000) (Guo et al., 2021). Secondary antibodies were peroxidase anti-mouse IgG (H + L) (PI-2000; Vector Laboratories; 1:2000) and peroxidase anti-rabbit IgG (H + L) (PI-1000; Vector Laboratories; 1:5,000). Precision Protein StrepTactin-HRP conjugate (Bio-Rad; 1:10,000) was added to the secondary antibody solution. Bands were visualized using Luminata classic and Immobilon forte Western HRP substrates (Milipore) and the myECL imager (Thermo Fisher Scientific). Detection of full-length blots is shown in the Supplementary Figures S1–S3. Proteins have been quantified using Image Lab 6.0 (Supplementary Figure S4). One sample of each condition was analysed per donor.

Cartilage discs were snap-frozen in liquid nitrogen and stored at −80°C until processing. Frozen discs were crushed in liquid nitrogen with a pestle and total RNA was isolated according to the protocol of the miRNeasy mini kit (Qiagen). cDNA synthesis and real-time RT-qPCR were performed following protocols from the manufacturer using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and TaqMan 2x universal PCR Master Mix or TaqMan™ Fast Advanced MasterMix (both Applied Biosystems). cDNA samples were probed for genes relevant for cartilage, hypertrophy and ossification using Taqman gene expression assays (Applied Biosystems). Individual samples were run as technical triplicates. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as endogenous control. Results, based on technical triplicates, are shown as expression relative to GAPDH using mean values ± SD from the 3 AC donors and the 3 MSC donors. RNA was isolated and analysed from 1 disc per donor.

Cartilage discs were embedded in Frozen Section Medium (Richard-Allan Scientific Neg50, Thermo Scientific) and frozen in dry ice-cooled isopentane. Frozen tissue blocks were stored at −80°C. The samples were cut in 9–10 µm thick sections on a CryoStar™ NX70 Cryostat (Thermo Scientific), mounted on SuperFrost Plus Adhesion slides and stored at −80°C. Sections were immediately post-fixed for 60 s in cold 95% ethanol directly before immunostaining. Sections were immunostained for the presence of ACAN (clone 4F4; Santa Cruz; at 0.1 µg/mL), COL2 (clone II-II6B3-a; DSHB; at 2.45 µg/mL), COL1 (clone EPR7785; Abcam; at 0.8 µg/mL), COL10 (clone X53 diluted 1:200; generous gift from Prof. Klaus von der Mark). Slides were incubated with primary antibodies diluted in 1.25% BSA in PBS at 4°C overnight. Negative controls were made by omitting primary antibodies (Supplementary Figure S5). The secondary antibodies, goat anti-rabbit IgG conjugated to Alexa 488 and goat anti-mouse IgG conjugated to Alexa 594 (Life Technologies), were diluted 1:400. The stained sections were mounted with Fluoroshield (Sigma), containing DAPI for nuclear staining. Samples were analyzed using an upright Nikon Eclipse E600 microscope equipped with an Olympus ColorView III camera. Images were opened in the image processing software Fiji (Schindelin et al., 2012), colour (red, green, blue) was assigned to the images, and images were merged. Thickness of cartilage discs was measured on the stained sections using Fiji. Five measures from representative areas of a section from each donor were taken. Nuclei were counted on the basis of DAPI stained sections prepared for immunofluorescence analysis. 20 equally-sized fields (200 × 200 px) were counted per donor using Fiji. Sections were prepared from 1 disc per donor.

Cartilage discs were fixed in a mixture of 2% glutaraldehyde and 0.5% paraformaldehyde in cacodylate buffer for 24 h at 4°C. Afterwards the samples were post-fixed in 2% osmium tetroxide for 2 h at 4°C and further dehydrated, infiltrated and embedded in epoxy resin (Epon). After polymerization, semi-thin sections were cut and stained with toluidine blue in order to localize the region of interest for making ultrathin sections. Epoxy blocks were then trimmed with respect to the structure of interest, and 70 nm ultrathin sections were cut on an ultramicrotome (Leica, UPC6) followed by staining with 4% uranyl acetate in 40% ethanol and Reynolds’ lead citrate. Prepared samples were examined in a transmission electron microscope (Tecnai12, FEI) (Paulsen et al., 2013). Samples were prepared based on 1 disc per donor.

The protocol used was modified after Hosseininia et al. (Hosseininia et al., 2019). Cartilage samples crushed in liquid nitrogen were mixed with 200 µL 8 M urea in 50 mM NH₄HCO₃, vortexed for 20 s and stored at 4°C for 18 h. Samples were then centrifuged at 14,000 g for 5 min and protein concentration of the supernatants was measured. 100 µg protein in 200 µL supernatant was reduced by addition of 1 µL 0.5 M DTT at 56°C for 30 min. Cysteins were alkylated (2.7 µL 550 mM IAA) before digestion into peptides by 2 µg lysylendopeptidase (Wako) for 2 h at room temperature. Samples were diluted with 50 mM NH₄HCO₃ to UREA concentration of 1 M before further digestion with 4 µg trypsin over night at 30°C. Thereafter the digested samples were passed through an ultrafiltration filter, 30 kDa (amicon) in 10 µL aliquots to remove large glycosaminoglycan-containing peptides. Samples were desalted by home-made reversed-phase chromatography using C18 microcolumns prepared by stacking three layers of C18 Empore Extraction Disks (Varian) into 200-μL pipette tips. Purified samples were diluted to a peptide concentration of 0.16 µg/µL and 3 µL was injected into the LC-MS/MS analysis. Proteins were extracted and analysed from 1 disc per donor.

Each peptide mixture was separated by Easy nLC1000 nano-LC system connected to a quadrupole—Orbitrap (QExactive Plus) mass spectrometer (ThermoElectron) for data dependent acquisition (top 10 intense peaks for MS/MS). Peptides were separated on a 50 cm EasySpray column (C18, 2 µm beads, 100 Å, 75 μm inner diameter) (Thermo) using a 120 min gradient up to an acetonitrile concentration of 30%. The resulting MS raw files were submitted to the MaxQuant software version 1.6.1.0 for protein identification and label-free quantification. Carbamidomethyl (C) was set as a fixed modification and acetyl (protein N-term), carbamyl (N-term) and oxidation (M) were set as variable modifications. First search peptide tolerance of 20 ppm and main search error 4.5 ppm were used. Trypsin without proline restriction enzyme option was used, with two miscleavages allowed. The minimal unique + razor peptides number was set to 1, and the allowed FDR was 0.01 (1%) for peptide and protein identification. Label-free quantitation was employed with default settings. The Uniprot database with ‘human’ entries (September 2018) was used for the database searches. Perseus software (ver 1.6.1.3) was used for the statistical analysis of the results. In Perseus known contaminants as provided by MaxQuant (Tyanova et al., 2016) and identified in the samples were excluded from further analysis, normalized intensities were log10 transformed and data filtered to include only those proteins that were identified in at least two out three replicates in at least one cell type. After filtering, proteins identified in only one cell type were considered to be the on/off differences. For further comparison analysis, missing values were imputed from normal distribution and t-test was performed with permutation based FDR ≤ 0.05 as the criteria.

37 proteins were analyzed by targeted mass spectrometry using parallel reaction monitoring method on an Orbitrap (QExactive HF) mass spectrometry. For each of the proteins, 1-3 unique peptides were selected based on results from data dependent acquisition analysis of the samples above. The analysis was scheduled with a limited number of peptide precursor ions monitored in 10 min retention times windows. Peptides were separated on a 25 cm EasySpray column (C18, 2 µm beads, 100 Å, 75 μm inner diameter) (Thermo) using a 90 min gradient up to an acetonitrile concentration of 30%. Raw data was submitted to MaxQuant for database search using the same settings as above. The resulting msms. txt file was submitted to Skyline (v.19.1) (MacLean et al., 2010) and was compared to the spectral library created from data dependent acquisition analysis to confirm identity. The three most intense y-ions from the parallel reaction monitoring spectra with highest intensity were used for the quantification of the peptides. Skyline is quantifying the fragment ions based on the peak areas.

MiRNA was isolated together with RNA (based on 1 disc per donor) using the miRNeasy mini kit (Qiagen) according to the manufacturer’s instructions. Library preparation was performed using the QIAseq miRNA Library Kit (Qiagen) according to manufacturer’s instructions, using 200 ng/sample input and 15 cycles of PCR amplification. Pooled libraries were sequenced on a single run of a NextSeq 500 (Illumina) with 75 cycle high-output reagents. These procedures were performed simultaneously for all samples.

Raw data was trimmed using FAIRPype, an in-house Python pipeline that performs the analysis as a series of steps. Reads were filtered for low quality reads, adapter trimmed, and collapsed to fasta format to reduce mapping time. Reads were mapped using the Bowtie (Langmead et al., 2009) mapping tool version 1.3.0. As the study was focusing on miRNAs and isomiRs, a reference index for read mapping was built based on the 1918 hairpin sequences present in v22.1 of miRBase. Overlapping entries (i.e., entries sharing identical or overlapping sequences with other entries) were removed using a Python script to produce a final set of 1743 unique hairpin sequences (Zhong X. et al., 2019). A Bowtie index was generated using the bowtie-index program and mapping was performed with parameters -v 2. IsomiR read counting was performed using the small RNA analysis pipeline step ‘ParseSAMForMiRNAs’ with the following conditions: 1) the ‘bleed’ was parameter set to 2. i.e., any read that started and ended within ± 2 nt of the specified start and stop position of a miRNA location specified in miRBase v21.1 was retained, 2) isomiRs sharing the same seed region were counted together, as these are likely to produce similar strong repression on the targets. Estimation of differentially expressed features was performed using EdgeR with the classic Single Factor analysis. Full details and links to the software are provided in the Supplementary Material.

Target Prediction was performed using the miRAW (Pla et al., 2018) software package for a subset of predicted up- and downregulated miRNAs, with the miRAWwrapper Python script to generate the shell scripts for batch execution. The Pita model was selected and both canonical (seed region pairing) and non-canonical (extended seed region pairing) were retained. For 3′UTR targets, Ensembl reference annotation Homo_sapiens.GRCh38.102 was used. For genes with multiple annotated 3′UTRs, these were filtered to retain only 1) TSL1 or TSL2 support and (ii) the longest 3′UTR to give a final target set of 19,362 3′UTRs. As all miRNA target prediction tools have sub-optimal performance (i.e., predictions contain many False Positives and False Negatives) a Python script filterMiRAWpredictions.py was used to filter the prediction sets. In this step, predicted targets with an estimated Mean Free Energy (MFE) between the miRNA and the mRNA < −15 or a target prediction probability <0.99995 were removed. Full details and links to the software are provided in the Supplementary Material.

MiRNA was isolated as described above (based on 1 disc per donor). cDNA synthesis and qRT-PCR were performed following protocols from the manufacturer using the Taqman MicroRNA Reverse Transcription Kit (Thermo Fisher Scientific). 10 ng miRNA in a total volume of 15 µL was reverse transcribed into cDNA. All samples were run in technical triplicates for RT-qPCR with each replicate containing 1 µL cDNA in a total volume of 15 µL. The thermocycling parameters were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. U6 was used as endogenous control. Results are shown as expression relative to U6 using mean values ± SD from the 3 AC donors and the three MSC donors.

For determination of wet weight, thickness measurement, nuclei count and RT-qPCR data, differences between groups were analysed using two-tailed Student’s t-test. p ≤ 0.05 was considered statistically significant. Results are presented as dot plots with mean and SD. For analysis of proteomics data, Student’s t-test with permutation-based FDR (FDR ≤ 0.05) for full proteomics and with p ≤ 0.05 for targeted proteomics were used to identify differently expressed proteins. All experiments were performed with N = 3 for each experimental group and degree of freedom (DF) = 4. For miRNA analysis, EdgeR-based single factor analysis was used for differential expression with correction for multiple testing with a false discovery rate (FDR ≤ 0.05).

To test which cell type is more suitable for production of cartilage in vitro, ACs and MSCs were expanded in vitro under identical culture conditions, and then allowed to form discs through differentiation in the presence of identical chondrogenic differentiation cocktails. The culture conditions and differentiation protocol used are, in our hands, the best available for both cell types. The 3D cultures were maintained for 4 weeks (Figure 1). While appearance, such as colour and surface, was similar (Figure 2A), ACs produced on average 2-fold thicker discs compared with MSCs (Figures 2A, B, p ≤ 0.05). Consequently, AC discs showed on average 2.5-fold higher wet weight compared with MSC discs (Figure 2C, p ≤ 0.05).

The production of thicker and heavier discs by ACs could be due to more cells producing ECM or each cell producing more ECM. Counting cell nuclei per defined area revealed that the density of nuclei in AC discs was higher than in MSC discs (Figures 2D, E) (p ≤ 0.05). Thus, although the two populations were seeded on the semipermeable membranes at the same density, at 4 weeks the AC discs contained many more cells. This means that either the ACs proliferated faster than MSCs, or that more MSCs died following establishment in 3D culture. Western blot analysis presented in Figure 2F shows that the band representing the proliferation marker proliferating cell nuclear antigen (PCNA) was much stronger in monolayer AC cultures than in monolayer MSC cultures, and that this difference persisted, with decreasing band strength, through day 2. Further, membranes were tested for caspase-3, which promotes apoptosis when activated by proteolytic processing into p17 and p12 fragments. Uncleaved caspase-3 does not promote apoptosis. The data revealed stronger caspase-3 expression in MSCs compared to ACs. However, neither ACs nor MSCs showed evidence of activated, cleaved caspase-3 (Figure 2F).

These findings indicate that discs made from ACs were thicker because ECM was produced from more cells. There were more cells most likely because ACs proliferated faster after being seeded for disc culture than the MSCs. Neither cell population showed evidence of apoptosis.

Having determined quantitative differences in cartilage disc production between the two cell populations, we proceeded to investigate qualitative differences between the discs. Immunofluorescence analysis of AC and MSC discs showed similar staining intensities for key chondrogenic molecules type II collagen (COL2), COL1 and aggrecan (ACAN) in discs from both cell types, whereas COL10 expression was clearly stronger in MSC discs. In AC discs, scattered cells close to the edges showed intracellular staining of COL10, while MSC discs showed intracellular as well as diffuse and speckled extracellular staining of COL10 (Figure 3A). Examination of AC and MSC discs using transmission electron microscopy (TEM) showed fibrils of relatively uniform thickness, but with no obvious difference between AC and MSC discs (Figure 3B). Toluidine blue staining (Supplementary Figure S6) indicated an increased concentration of proteoglycans/glycosaminoglycans in the periphery of discs.

As COL10 is a marker for hypertrophic cartilage, we proceeded to compare the two types of discs for the expression of genes associated with normal and hypertrophic cartilage and bone formation by RT-qPCR (Figure 4). There were no differences in the expression of hyaline cartilage components COL2A1 and ACAN between discs made from ACs and MSCs, while COL1A1 levels were significantly higher in MSC discs. As suggested by the immunofluorescence microscopy images, MSC discs revealed significantly higher COL10A1 expression levels than AC discs. The COL2A1/COL1A1 ratio in AC discs was 15.4 and in MSC discs 4.6, while the COL2A1/COL10A1 ratios were 61.1 and 4.7, respectively. Expression levels of cartilage components CILP2 and CD44 were significantly increased in discs made from ACs (p ≤ 0.05), whereas levels of COMP, PRG4, OGN and OMD were higher, but the differences were not statistically significant. Known drivers of hypertrophy and bone formation such as BGLAP, IBSP, SSP1 and TMEM119 were significantly higher in discs made from MSCs (p ≤ 0.05), while RUNX2, IHH, ALPL, SP7 were higher, but not significantly increased in MSCs (Figure 4). Additional gene expression data is shown in Supplementary Figure S7.

These results suggest that there are neither obvious differences in the major hyaline cartilage components COL2 and ACAN, nor in matrix morphology. However, MSC discs showed increased expression of cartilage hypertrophy and ossification markers.

To get a more detailed impression of the differences between discs made from ACs and MSCs we performed high-resolution mass spectrometry-based proteomics analyses on discs differentiated for 4 weeks. More than 3,400 proteins were identified by mass spectrometry data analysis. As a measure for protein abundance, intensity Based Absolute Quantification (iBAQ) (Tyanova et al., 2016) (Supplementary Table S1) showed that most of the highest expressed proteins of one disc type were also among the highest expressed proteins in the other disc type. Of the proteins with aberrant ranking, tenascin (TNC) and cartilage oligomeric matrix protein (COMP) were ranked higher in AC discs, while protein S100-P (S100P) and inorganic pyrophosphatase (PPA1) were ranked higher in the MSC discs.

We investigated differential expression of proteins using full proteome analysis and targeted proteomics (Figure 5; Supplementary Table S2). First, using label-free quantification (LFQ) (Tyanova et al., 2016) for a comparison of normalised intensity values between samples, we found proteins that were expressed in discs made from one cell population, but not in the other population (on/off proteins). This analysis identified 111 differentially expressed proteins. Further, applying the t-test with permutation-based FDR (≤ 0.05, minimum 2-fold difference in expression) as selection criteria we identified 45 significantly differentially expressed proteins. Out of these, 13 proteins were shared with “on/off proteins.” Lastly, we performed targeted mass spectrometry analysis, which has better quantification accuracy than label-free analysis, for some of the proteins that were significantly different in the LFQ analysis based on p-value<0.05 only. Selected candidates for this analysis were proteins known to be involved in chondrogenesis, cell proliferation, OA and hypertrophy/ossification processes. Some additional proteins that were not significant based on p-value<0.05, but were of particular interest for articular cartilage, namely COL10 and lubricin (PRG4), were also chosen for analysis by targeted proteomics. Altogether 37 proteins were analysed by targeted mass spectrometry, of which 14 proteins were found to be significantly differentially expressed (p ≤ 0.05). Out of these, one protein is shared with “on/off proteins.” Thus, altogether 156 proteins were differentially expressed, 72 had higher expression in ACs while 84 were expressed at higher levels in MSCs (Supplementary Table S2).

The differentially expressed proteins were uploaded into the analysis program Enrichr (https://maayanlab.cloud/Enrichr/). Here a number of gene ontology (GO) and pathway analyses were applied (Figure 6; Supplementary Table S3). Many proteins upregulated in AC discs were implicated in the organization of the ECM, both collagen fibrils and microfibrils. Of these, gremlin 1 (GREM1) has been found to identify a stem cell with bone, cartilage and stromal potential (Worthley et al., 2015). Furthermore, GREM1 is known to inhibit BMP-2 induced differentiation of osteoblasts and to suppress the production of the hypertrophic marker COL10 (Kišonaitė et al., 2016; Díaz-Payno et al., 2020). Highly differentially expressed mimecan (OGN), dermatopontin (DPT), TNC, COL7A1 and COL16A1, matrix Gla protein (MGP) and lysyl oxidase-like 1 and 2 (LOXL1 and LOXL2) are cartilage ECM constituents, and many of these are associated with collagen cross-linking (Tallheden et al., 2004; Deckx et al., 2016; Mortensen and Karsdal, 2016; Sand and Karsdal, 2016; Lin et al., 2020). Also, highly differentially expressed CD44 is the chondrocyte receptor for hyaluronic acid (HA) (Ishida et al., 1997). The CD44/HA interaction promotes chondrocyte proliferation and is essential for the synthesis of the proteoglycan component of the ECM (Chow et al., 1998). A number of proteins were also involved in the control of the cell cycle. One of these was PCNA (BioPlanet pathway analysis) (Supplementary Table S2; Figure 7C), which was also shown by Western blot to be expressed at higher levels in the very early days of 3D culture (Figure 2F).

The analysis revealed a number of proteins expressed at higher levels in MSC discs that have been reported to be associated with cartilage hypertrophy and bone formation. Beta-catenin (CTNNB1), found under “metabolism” in the pathway analysis, blocks the development of stable chondrocytes from mesenchymal progenitors in favour of pre-osteoblasts (Galea et al., 2021). Transmembrane protein 119 (TMEM119) (Hisa et al., 2011; Tanaka et al., 2012) and fibrillin-2 (FBN2) (Nistala et al., 2010) are important for osteoblast differentiation and bone mineralization. Bone sialoprotein 2 (IBSP) (Ganss et al., 1999; Malaval et al., 2008) and osteopontin (SPP1) (Hunter, 2013; Si et al., 2020) are major non-collagenous structural proteins of bone matrix that bind tightly to hydroxyapatite. Phosphoethanolamine/phosphocholine phosphatase 1 (PHOSPHO1) (Roberts et al., 2004; Roberts et al., 2005) is involved in the generation of inorganic phosphate for bone mineralization, while polypeptide N-acetylgalactosaminyltransferase 3 (GALNT3) (Chefetz et al., 2009) plays a central role in phosphate homeostasis. Alkaline phosphatase (ALPL) (Sharma et al., 2014; Vimalraj, 2020), a key molecule for skeletal mineralization, was expressed more than three times higher in MSC discs (Supplementary Table S1), but this difference was not found to be significant. Finally upregulated protein COL10A1 is a canonical marker for hypertrophic chondrocytes, and it has been strongly argued that Col10a1+ hypertrophic chondrocytes may become osteoblasts and osteocytes during endochondral bone formation (Yang et al., 2014).

A number of proteins which affect cell growth were upregulated in MSC discs. For instance Sestrin 2 (SESN2) and SH3 domain binding protein 4 (SH3BP4) are both known to negatively regulate cell proliferation through inhibition of the mammalian target of rapamycin complex 1 (Kim et al., 2012; Luo et al., 2018), while disco-interacting protein 2 homolog B (DIP2B) may epigenetically regulate cell proliferation through DNA methylation (Adlat et al., 2020). A number of upregulated proteins in MSC discs were also associated with mitochondrial protein synthesis.

In conclusion, both types of discs contain high levels of proteins known to be essential in normal hyaline cartilage. However, MSC discs contained more proteins known to be associated with bone formation and negative control of the cell cycle.

Next, we performed small RNA sequencing to investigate if differential expression of miRNAs and associated isomiRs was likely to contribute to differences in cartilage composition (Supplementary Figure S8). 352 miRNAs including isomiRs were detected at greater than 10 reads per million. Out of these, 76 miRNAs were differentially expressed between the two types of discs (fold change ≥2, FDR ≤ 0.05, Supplementary Table S4). 36 miRNAs were at higher levels in ACs and 40 at higher levels in MSC discs. Of all differentially expressed miRNA sequences 40 were 3p strands, 36 were 5p strands and 24 of the 76 sequences were isomiRs. Several of the canonical miRNAs upregulated in ACs, such as miR-675-3p and 5p, miR-335-5p, miR-582-5p and 3p, miR-9-5p, miR-136-3p and miR 495-3p are known to be involved in cartilage biogenesis and disease (Tomé et al., 2011; Díaz-Prado et al., 2012; Chen et al., 2017; Chen et al., 2018; Zhong G. et al., 2019; Chen et al., 2019; Seidl and Murphy, 2019; Wang et al., 2019; Chen et al., 2020; Shen et al., 2020; Zhang et al., 2021). Of the highest differentially expressed canonical miRNAs in MSCs miR-10a-5p, miR-181b, miR-181a, miR-135b-5p and miR-218-5p are known to be involved in osteogenesis, cartilage hypertrophy and degeneration and OA (Bhushan et al., 2013; Song et al., 2013; Gabler et al., 2015; Nakamura et al., 2016; Lu et al., 2017; Ma et al., 2019; Yin et al., 2019; Li et al., 2021). Hardly any of the differentially expressed isomiRs shown in Supplementary Table S4 have known functionality related to cartilage, bone or skeletal diseases.

To further dissect the role played by the differentially expressed miRNAs we performed target mRNA prediction analysis for all the differentially expressed miRNA sequences, and present results for the top 10 differentially expressed miRNAs/isomiRs in both disc types compared with the proteins that were differently expressed by the AC and MSC discs. Figure 7A shows that 46 of the 72 (64%) proteins that had lower expression in MSCs (presented as higher in ACs) were predicted targets of one or more of the top 10 miRNAs that were more highly expressed in MSC discs. Figure 7B shows that 64 of the 84 (76%) proteins that were expressed at lower levels in ACs (thus presented as higher in MSCs) were predicted targets of one or more of the top 10 miRNAs that were higher in AC discs. For additional information, the 10 most differentially expressed proteins from the different approaches of proteomic analysis (based on Supplementary Table S2) and the 10 most differentially expressed miRNAs (based on Supplementary Table S4) are presented for each cell type in Figures 7C, D. Some of these miRNAs were validated using RT-qPCR (Figure 8). These observations strongly suggest that miRNAs contribute to the differences in cartilage phenotypes found in the two disc types.