If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
IRMB, University of Montpellier, INSERM, Montpellier, FranceClinical Immunology and Osteoarticular Diseases Therapeutic Unit, Hôpital Lapeyronie, Montpellier, France
Sorbonne Université, INSERM, Centre de Recherche Saint-Antoine, Paris, FranceLaboratoire de Pharmacologie Clinique et Toxicologie, Centre Hospitalier Régional Universitaire de Nancy, Vandœuvre Les Nancy, France
Sorbonne Université, INSERM, Centre de Recherche Saint-Antoine, Paris, FranceDepartment of Rheumatology, AP-HP Saint- Antoine Hospital, DMU 3iD, Paris, France
Inserm, UMR 1229, RMeS, Regenerative Medicine and Skeleton, Université de Nantes, ONIRIS, Nantes, FranceUniversité de Nantes, UFR Odontologie, Nantes, France
INSERM UMR-S 1124, Toxicité Environnementale, Cibles Thérapeutiques, Signalisation Cellulaire et Biomarqueurs, Faculté des Sciences Fondamentales et Biomédicales, Université de Paris, Sorbonne Paris Cité, Paris, FranceAP-HP, Groupe Hospitalier AP-HP, Centre-Université de Paris, Hôpital Cochin, Service de Rééducation et de Réadaptation de l’Appareil Locomoteur et des Pathologies du Rachis, Paris, FranceUniversité de Paris, Faculté de Santé, UFR Médecine Paris Descartes, Paris, France
IRMB, University of Montpellier, INSERM, Montpellier, FranceClinical Immunology and Osteoarticular Diseases Therapeutic Unit, Hôpital Lapeyronie, Montpellier, France
Inserm, UMR 1229, RMeS, Regenerative Medicine and Skeleton, Université de Nantes, ONIRIS, Nantes, FranceUniversité de Nantes, UFR Odontologie, Nantes, FranceCHU Nantes, PHU4 OTONN, Nantes, France
Sorbonne Université, INSERM, Centre de Recherche Saint-Antoine, Paris, FranceDepartment of Rheumatology, AP-HP Saint- Antoine Hospital, DMU 3iD, Paris, France
TGFβ is a key player in cartilage homeostasis and OA pathology. However, few data are available on the role of TGFβ signalling in the different OA phenotypes. Here, we analysed the TGFβ pathway by transcriptomic analysis in six mouse models of OA.
Method
We have brought together seven expert laboratories in OA pathophysiology and, used inter-laboratories standard operating procedures and quality controls to increase experimental reproducibility and decrease bias. As none of the available OA models covers the complexity and heterogeneity of the human disease, we used six different murine models of knee OA: from post-traumatic/mechanical models (meniscectomy (MNX), MNX and hypergravity (HG-MNX), MNX and high fat diet (HF-MNX), MNX and seipin knock-out (SP-MNX)) to aging-related OA and inflammatory OA (collagenase-induced OA (CIOA)). Four controls (MNX-sham, young, SP-sham, CIOA-sham) were added. OsteoArthritis Research Society International (OARSI)-based scoring of femoral condyles and ribonucleic acid (RNA) extraction from tibial plateau samples were done by single operators as well as the transcriptomic analysis of the TGFβ family pathway by Custom TaqMan® Array Microfluidic Cards.
Results
The transcriptomic analysis revealed specific gene signatures in each of the six models; however, no gene was deregulated in all six OA models. Of interest, we found that the combinatorial Gdf5-Cd36-Ltbp4 signature might discriminate distinct subgroups of OA: Cd36 upregulation is a hallmark of MNX-related OA while Gdf5 and Ltbp4 upregulation is related to MNX-induced OA and CIOA.
Conclusion
These findings stress the OA animal model heterogeneity and the need of caution when extrapolating results from one model to another.
Although osteoarthritis (OA) is the most prevalent joint disease worldwide, there is still not a single disease-modifying OA drug on the market. The current treatment options usually result in poorly predictable outcomes due to the high interpatient variability in OA clinical and structural features. Indeed, some studies have reported OA phenotype heterogeneity among patients
. This will help to improve clinical research and to develop targeted treatments and prevention strategies based on a phenotype-guided approach. The advantage of searching for targets based on differences between risk factors is the simplicity then of selecting patients for future personalized medicine.
Currently, OA research relies on the use of various animal models (mainly mice and rats, and more rarely large animals) that mimic mechanical, metabolic or inflammatory OA. However, none of these models covers the complexity and heterogeneity of the human disease but different models likely reflect the heterogeneity of human OA. Moreover, it is difficult to compare the results of different experimental studies due to the heterogeneity of animal backgrounds and experimental protocols. Several studies have analysed global gene expression in OA samples using ribonucleic acid (RNA)-seq
and have generated huge amounts of datasets. However, only small subsets of data are validated and large amounts of data are commonly not investigated. To try to tackle some of these limitations, seven French academic laboratories experts in OA animal models formed a Research on OsteoArthritis Disease (ROAD) consortium to centralize many experimental steps and to put in place standard operating procedures (SOP) in order to minimize bias and increase reproducibility. The first objective of the ROAD consortium was to investigate the TGFβ pathway in various OA phenotypes. Indeed, recent findings have shown that TGFβ is a central player in cartilage homeostasis and OA pathology
. However, few data are available on the pathophysiological role of TGFβ family members in the different OA phenotypes. Therefore, the consortium analysed the TGFβ pathway by transcriptomic analysis in six murine models of knee OA that reproduce the main phenotypes of the human disease: surgical meniscectomy (MNX) to mimic mechanical or post-traumatic OA, hypergravity and MNX (HG-MNX) to mimic overweight-induced mechanical OA, High Fat (HF) diet and MNX (HF-MNX) to mimic obesity-induced OA, seipin knock-out and MNX (SP-MNX) to mimic metabolic syndrome-induced OA, aging to mimic age-related OA, and collagenase-induced OA (CIOA) to mimic inflammatory OA.
Material and methods
Animal models
Animal models and controls (ten mice/group) were generated using C57BL/6JR6 males that are known to display more severe and reproducible disease
and were supplied by the same company (Janvier Labs, France). Bscl2−/− mice (SP-MNX and SP-sham controls; C57BL/6 J background) were from Commissariat à l'Energie Atomique et aux Energies Renouvelables (CEA) (Direction des Sciences du Vivant/Genoscope/LABGEM). Six animals per group were calculated to be required to demonstrate significance at the 5% level with a power of 80% using the G∗power software but 10 animals were included to have six animals with an OA score ≥3 at the end of the experiment. MNX was performed in one joint of 10 weeks old mice by the use of partial meniscectomy as described
and done by a single trained operator in all laboratories. All animal procedures were approved by the local institutions' animal welfare committees and were performed in accordance with the European guidelines for the care and use of laboratory animals (2010/63/UE). Surgery and euthanasia were performed after anaesthesia with isoflurane gas, and all efforts were made to minimize suffering. Mice were housed in solid bottomed plastic cages in quiet rooms at 22° ± 1°C, 60% controlled humidity, and 12 h/12 h light/dark cycle. Animals were used after an adaptation period of 7 days and had free access to tap water and standard pelleted chow (except the HF model). Mice were sacrificed at week 6 after OA induction to have a comparable disease time induction although we were aware that OA severity can vary according to the model.
•
Joint instability model
MNX was selected as the reference model of joint instability-related OA
. Knee joint instability was induced surgically in the right knee by medial partial meniscectomy. Surgery was performed under a binocular magnifier (X15) using a Sharpoint microsurgical stab knife. Mice were placed in dorsal position, knee flexed and right foot taped. After skin incision, the medial femoro-tibial ligament was cut, a short incision of the medial side of quadriceps muscle was performed, the knee capsule was cleaved and the patella was sub-luxated laterally. After section of the meniscotibial ligament, the medial meniscus was gently pulled out and ¾ of its anterior horn removed. Then, the patella was replaced, the quadriceps muscle and the skin plan sutured. Control animals underwent sham surgery (ligament visualization but not dissection).
•
Hypergravity model
Hypergravity mimics the overweight-associated mechanical strain on joints without metabolism dysregulation. In mice with MNX, hypergravity induces large OA lesions that are not observed without surgical induction
. MNX was performed on the right knee, and mice were put back in their box for 48 h. Then, cages were transferred in the gondolas of the centrifuge (COMAT Aérospace, Flourens, France) to maintain a permanent level of hypergravity
. This device has four 1.4 m-long arms that hold at their distant end a mobile octagonal gondola (56.2 × 52.0 × 59.2 cm). All gondolas are equipped with an infra-red video surveillance system to monitor the animals' condition and food/water stocks. In the centrifuge, temperature and light conditions were identical to that of control cages. At the start of centrifugation, acceleration was smoothly and gradually increased over a period of 40 s. The final acceleration was 2 g (29.6 rpm), and animals were kept at 2 g for 6 weeks. Animals were provided with enough food and water for 4 weeks. Then, the centrifuge was transiently stopped to allow litter change, animal weighing, and chow and water supply refilling. Control mice with MNX were not exposed to hypergravity.
•
Metabolic disorder model
Seipin (SP) knock-out mice are representative of metabolism disorder, which is a feature associated with OA
. Bscl2 deficiency in mice recapitulates the main features of the phenotype of patients with Berardinelli-Seip Congenital Lipodystrophy (BSCL), including almost complete absence of adipose tissue, hyperglycaemia, hyperinsulinemia, and insulin resistance. MNX and Sham surgery were performed in 10 week-old Bscl2−/− mice.
•
High Fat Diet model
The HF diet model reproduces the effect of obesity and dysregulated metabolism on OA onset
. At the age of 6 weeks, mice were fed with High Fat Diet (HFD, 60% of calories from fat, Ssniff, EF D12492 (II) mod. Soest, Germany) that was provided ad libitum for 10 weeks with the chow changed twice per week. A number of mice 20% higher than the final group size was included to ensure statistical power of the experimentation. MNX was performed at the age of 10 weeks. In absence of surgical induction, mice did not develop spontaneous lesions of OA. The average weekly weight gain ranged from 1 g to 1.5 g, leading to a final weight gain of 73% (mean: 14.6 g) associated with insulin resistance (HOMA-IR: +246%). Considering the large variability generally observed in the final body weight and fat mass, only animals with a final weight gain higher than 70% were analysed.
•
Collagenase-induced OA model
The collagenase-induced model (CIOA) is characterized by low grade inflammation of the synovial membrane, leading to OA lesions
. A solution of 1 U/5 μL type VII collagenase from Clostridium histolyticum (Sigma–Aldrich) was prepared in saline solution. At day 0, a small skin incision was performed on top of the patellar tendon. The knee was bended and the collagenase solution (5 μL) was injected in the intra-articular space using a 10 μL syringe (Hamilton) with a 25 gauge needle. On day 2, a second collagenase injection was performed according to the same procedure. 6 weeks later, animals were sacrificed. Control animals were injected with saline solution.
. Mice were housed with free access to food and water and euthanized at the age of 24 months. Control young mice were kept in the animal facility and euthanized at the age of 16 weeks.
Sample preparation for histology and messenger ribonucleic acid (mRNA) extraction
After sacrifice, femora and tibiae from 10 knee joints (one joint/mouse) per model were dissected. Skin and muscles were removed and the knee joint was isolated by sectioning the distal extremity of tibiae and proximal part of the femurs. The tibial plateau was isolated from bone at the growth plate interface, by cutting 2–3 mm beneath the cartilage surface. The remaining soft tissues (meniscus, ligaments and synovium) were removed. The tibial plateau was immediately placed in 1 mL of TRIzol® Reagent (Life Technologies), snap-frozen in liquid nitrogen, and stored at −80°C till RNA extraction. After isolation, femoral condyles were fixed in 4% paraformaldehyde for 36 h, and then decalcified in 0.5 M ethylenediaminetetraacetic (EDTA) at room temperature for 15 days.
Histology
After dehydration in a graded series of alcohol, femoral condyles were embedded in paraffin at 60°C in a tissue processor. On average, 30 serial sagittal sections of 5 μm were cut, and three were chosen at the upper, medium and lower levels every 50 μm from cartilage surface. OA scoring was performed after Safranin O-Fast Green staining, according to the OsteoArthritis Research Society International (OARSI) recommendations
. For each animal, the OA score was the highest score obtained at one of the three levels. For each model, all sections were blindly scored by the same three readers.
RNA isolation
Tibial plateau samples were prepared in each consortium laboratory and then shipped for centralized RNA extraction that was performed by crushing thawed samples with ceramic beads (Precellys® Lysing kit CK28R), using a Precellys® 24 tissue homogenizer equipped with the Cryolis® cooling unit (Bertin Technologies). Samples underwent three successive lysis cycles at 6500 rpm for 15 s, spaced by a 5 min lag phase at 4°C, before addition of 200 μL chloroform. After incubation at room temperature for 3 min, the aqueous phase was recovered, 600 μL of 70% ethanol was added, and the solution was transferred to an RNeasy® spin column (Qiagen) and the next steps were performed according to the supplier's recommendations. Total RNA was quantified with a Nanodrop® instrument and aliquots were frozen at - 80°C. RNA integrity was confirmed with the Agilent® RNA 6000 kit on an Agilent Bioanalyzer 2100®.
Transcriptomic analysis
Transcriptomic analysis was performed on Custom TaqMan® Array Microfluidic Cards (TAC) that were designed to perform 384 real-time polymerase chain reaction (PCR) reactions on a ViiA™ seven Fast Real-Time PCR System (Applied Biosystems®). Custom TAC were designed for the analysis of TGFβ family members (Table I). Reverse transcription was performed using 250 ng of total RNA and the High Capacity cDNA Reverse Transcription Kit (Life Technologies). Quantitative PCR was done using complementary deoxyribonucleic acid (cDNA) (150 ng) mixed with TaqMan Fast Advanced Master Mix (Life Technologies). After 40 cycles of amplification (95°C for 20 s and then 95°C for 1 s and 60°C for 20 s), data were analysed with the Applied Biosystems® Relative Quantification Analysis Module. Amplification curves for each target were individually checked and baselines adjusted, when necessary, to determine the cycle threshold (CT) values. Gene expression was normalized to the mean CT value of four housekeeping genes (Gusb, Hprt, Rps9, Ppia) and expressed as relative gene expression using the 2−ΔCT formula or as a fold change (FC) expression using the 2−ΔΔCT formula.
Table IList of genes and identification (ID) in the TaqMan® Gene Expression Assay
Unsupervised two-dimensional hierarchical clustering was generated with mean-centred relative expression values (2−ΔCt) of 91 genes per sample using XLStat software. Distances between samples were calculated based on the ΔCT values using Pearson's Correlation and average linkage method. The vertical height of the dendogram shows the Euclidean distances between samples. The two-dimensional scatter plot of Principal Component Analysis (PCA) was performed using XLStat and represents the expression pattern of (2−ΔCt) sample values of the ten subgroups. When plotting the sample data points, F1 (PCA Component 1 (32.85% variance)) was used as the x-axis and F2 (PCA Component 2 (12.69% variance)) as the y-axis. Data did not assume a Gaussian distribution and were considered unpaired. The statistical analysis was performed between two groups for each OA model vs its respective control (MNX vs MNX-sham, CIOA vs CIOA-sham, Aged vs Young, SP-MNX vs MNX, HG-MNX vs MNX, and HF-MNX vs MNX) using the Mann–Whitney test and GraphPad 7 (San Diego, CA, USA). Data were expressed as relative expression (2−ΔCt) or as fold change (FC of gene expression in one OA sample as compared to its respective control normalized to 1) and represented as median with interquartile range. Differences were considered significant at p < 0.05 and p < 0.01.
Results
Defining the standard operating procedures
One important feature in the study design was to define the SOP after the harmonization of the experimental protocols (from animal models to transcriptomic analysis) in three consensus meetings of the ROAD consortium. A study workflow was designed (Fig. 1). At each step, the analysis technique was performed in a single laboratory by the same operator to avoid experimental bias. The centralized OA scoring of the different models and controls [Fig. 2(A)] showed that OA scores were significantly higher in all models (≥3 on a scale of 0–5) than in their respective control (score ≤2), although variability in control samples was observed [Fig. 2(B)]. The concentration of total RNA isolated from tibial plateau samples was not homogeneous among samples, and was significantly higher in the aged, HG-MNX and HF-MNX models than in their controls (young and MNX mice, respectively) [Fig. 2(C)]. The RNA Integrity Number (RIN) score, which estimates RNA quality and integrity, was heterogeneous among samples, with significantly lower scores in samples from the aged and CIOA animals than from their controls (young and CIOA-sham) [Fig. 2(D)]. Among all samples, six out of the ten samples per group that met the criteria of selection were analysed by TAC. The mean CT values for the housekeeping genes were significantly higher in the CIOA, HG-MNX and HF-MNX samples than in their controls [Fig. 2(E)]. However, the mean CT values for the housekeeping genes were positively correlated with the mean CT values for all genes [Fig. 2(F)]. This indicated that the lower expression of housekeeping genes in some samples could be attributed to a lower amount of cDNA loaded in the TAC and not to a differential regulation of the housekeeping genes. We also detected the expression of genes specific for cartilage (type II collagen, aggrecan) or bone (Runx2, Sp7) in all OA models (data not shown), indicating that both tissues were represented in our samples.
Fig. 1Study workflow. Samples were collected from six OA models and their respective controls: OA induced by Destabilization of the Medial Meniscus (MNX) vs sham, OA associated with age (Aging) vs young animals, inflammation (Collagenase-induced OA; CIOA) vs sham, obesity [High Fat (HF) vs Normal Diet (ND)] overweight [Hypergravity (HG-MNX) vs MNX] (these models were generated in C57BL/6JRj male mice), and OA associated with metabolic syndrome (Seipin knock-out; SP-MNX vs SP-sham). At the experiment end, knee joints were harvested; femoral condyles and tibial plateaus were prepared for histological analyses and RNA isolation, respectively. OA severity was scored after Safranin-O-Fast Green staining using the OARSI grading system. The expression of genes involved in TGF-β signalling was analysed using custom-made Taqman™ Array Cards.
Fig. 2Sample quality controls. (A) Sagittal views of femoral condyles from CIOA-sham (left) and CIOA (right) mice as representative of control and OA cartilage (OA score = 1 and 4, respectively). (B) Distribution of OARSI scores for cartilage destruction in histological sections of femoral condyles. Results are presented as median with interquartile range (n = 6). (C) Total RNA concentration (ng/μL) after extraction from tibial plateau samples of OA models and controls. (D) RNA Integrity Number (RIN) for total RNA extracted from tibial plateau samples of OA models and controls. (E) Mean CT values for the housekeeping genes Ppia, Hprt, Gusb, and Rps9 obtained using Taqman® Array Cards. Data are represented as median with interquartile range; ∗p < 0.05, ∗∗p < 0.01 (Mann–Whitney test). (F) Correlation between the mean CT value of the four housekeeping genes and the mean CT value of all genes. Each dot represents a sample (n = 59): Pearson's r = 0.8471, p < 0.001.
TGFβ signatures according to the experimental OA phenotypes
Hierarchical clustering and average linkage clustering of the mRNA expression data in the 10 groups of mice (6 OA models and four controls) revealed marked differences among groups [Fig. 3(A)]. Three main subgroups could be detected: a cluster that included samples from mice with Aging-, HF-MNX-, HG-MNX-related OA; a cluster that included mainly samples from SP-sham, SP-MNX mice; and a cluster of samples from MNX, CIOA, and CIOA-sham. Sham samples did not cluster together, even though most of them are distributed in the last group with the exception of SP-sham, which is closer to SP-MNX. PCA revealed distinct transcriptional profiles among groups that allowed gathering them in three distinct clusters [Fig. 3(B)]. One (HG-MNX and HF-MNX) was clearly separated from the other two clusters that included 1) CIOA, MNX and young animals, and 2) control groups (sham, SP-sham, CIOA-sham). Conversely, old and SP-MNX animals were set apart from the others.
Fig. 3Global gene expression analysis. (A) Unsupervised hierarchical clustering (Pearson correlation, average linkage) of the six samples for each OA model and for each of the four controls. (B) Principal component analysis of the same data as in A. Squares represent the centroid of each group (n = 6 mice per group). (C) Number of significantly deregulated genes in the six OA murine models compared with their controls. (D) Venn diagram showing the number of significantly deregulated genes identified in each OA model in a set of 91 targets (Table I).
To determine whether a specific gene signature could be associated with the different OA phenotypes, the gene expression profile of each OA group was compared with that of its control: MNX vs MNX-sham, CIOA vs CIOA-sham, Aged vs Young, SP-MNX vs MNX, HG-MNX vs MNX, and HF-MNX vs MNX. The number of significantly deregulated genes was similar in the MNX, SP-MNX, HG-MNX and HF-MNX groups (around 30 genes) [Fig. 3(C)]. Conversely, 15 and 47 genes were deregulated in the samples from CIOA and Aged animals, respectively. We identified genes that were common to two or more groups and a gene signature that was specific for each OA model (see Venn diagram in Fig. 3(D) and Table II). Importantly, no gene was deregulated in all six OA models.
Table IIList of genes specific to each murine model of OA. Modulation of gene expression in each model is expressed as a fold change related to the controls (MNX vs MNX-sham; CIOA vs CIOA-sham; Aging vs young; SP-, HG-, HF-MNX vs MNX)
To further analyse the specific gene signatures, we visualized the genes that were significantly dysregulated (FC >1.5) in each OA model using Volcano plots. In the MNX model, gene expression profiling revealed that all 30 modulated genes were upregulated compared with control [Fig. 4(A)]. In the CIOA model, 14 of the 15 deregulated genes were significantly upregulated [Fig. 4(B)]. Conversely, in the Aging- and SP-MNX-related OA, most genes were downregulated (44/47 and 25/28 genes, respectively) [Fig. 4(C)–(D)]. Finally, in the HG-MNX and HF-MNX models, 70% and 71% of genes were upregulated [Fig. 4(E)–(F)]. Only four genes were differentially regulated between these models: Smurf2 and Id2 were upregulated, Tgfbrap1 and Lefty were downregulated only in the HF-MNX model. Altogether, our data revealed that many TGFβ family members were deregulated in the different OA subtypes, supporting the key role of the TGFβ pathway, whatever the OA risk factor.
Fig. 4Analysis of differentially expressed genes in the OA murine models. Results are shown for DMM (A), CIOA (B), Aging (C), SP-DMM (D), HG-DMM (E) and HF-DMM (F). Volcano plots (left panels) show the up- and downregulated genes in each OA model vs its control. For each plot, the x-axis represents the log 2-fold change (FC), and the y-axis represents the log 10 P-values. Genes with an exact P-value <0.05 were considered as differentially expressed. Scatter plots (right panels) show the expression FC of significantly deregulated genes (p < 0.05) in the six OA models compared with their control group. Results are expressed as the median with interquartile range (n = 6).
A Gdf5, Ltbp4, Cd36 combinatorial gene signature for OA
Then, we split the six OA models in two groups. The first group included the OA models related to obesity or fat metabolism (SP-MNX, HG-MNX, and HF-MNX) and/or MNX. The number of shared and specific genes is shown in the Venn diagram [Fig. 5(A)]. Most of the modulated genes were common to two or three models, and few genes were specific to each model. However, only Cd36 was deregulated in all four models. The second group included MNX and the two other most common OA models: inflammation (CIOA) and Aging [Fig. 5(B)]. Approximately 50% of all deregulated genes were specific to each model and only two genes were deregulated in all three models: Gdf5 and Ltbp4. Analysis of these three genes in all models and their respective controls showed that Cd36 was significantly upregulated in MNX, SP-MNX, HG-MNX and HF-MNX samples [Fig. 5(C)]. Gdf5 was significantly upregulated in the MNX and CIOA models and significantly downregulated in the Aging model. Ltbp4 was significantly upregulated in all models, but for the Aging model where it was significantly downregulated. These data suggest that Cd36 upregulation is a hallmark of trauma-related OA, while the deregulation of Gdf5 and Ltbp4 is related to different OA stimuli.
Fig. 5Genes significantly deregulated in the various OA models. (A) Venn diagrams showing the number of differentially expressed genes identified in the four OA models related to overweight or metabolism disorder (MNX, SP-MNX, HG-MNX, HF-MNX), and (B) in the three most common OA models (MNX, CIOA, Aged), among a set of 92 targets (Table I). (C) Relative gene expression of Cd36, Gdf5, Ltbp4 in the four controls and six OA models are expressed as the median with interquartile range (n = 6/group); ∗p < 0.05, ∗∗p < 0.01 (Mann–Whitney test).
The first objective of the ROAD consortium was to identify specific gene signatures for the main OA clinical phenotypes using six relevant murine models by focusing on the transcriptomic analysis of the TGFβ pathway. Although this pathway has been extensively studied in some OA murine models
Inducible chondrocyte-specific overexpression of BMP2 in young mice results in severe aggravation of osteophyte formation in experimental OA without altering cartilage damage.
, it is quite impossible to compare these results from independent laboratories due to potential biases that may influence gene expression, such as mouse genetic background, age, sex, housing conditions, histopathological scoring subjectivity and inter-investigator variability. Here, we wanted to limit these potential biases by defining SOPs and by centralizing each step of data acquisition and processing, thereby minimizing the risks of failure to identify relevant targets
. The resulting data allowed the accurate comparative analysis of six models using their respective controls.
The main finding of our transcriptomic analysis is the unexpected lack of deregulated genes common to all murine models of OA, although many TGFβ family members were deregulated pointing out the critical role played by the TGFβ pathway in OA
. This might reflect the heterogeneity of responses to the different stimuli leading to similar symptoms, as observed in patients with OA. Differences in the expression pattern between the different models likely relate to the peculiarities and distinct natures of the models. We are also aware that transcriptional regulation of genes may not be reflected at the protein level. Analysis of these differences at the protein level were beyond the scope of the present study but likely warrants further studies. Some genes, such as type II collagen, may be differently regulated depending on the OA model suggesting possible different timings or mechanisms of regulation that warrant further investigation. Heterogeneity may also be emphasized by the individual responses within the same model, thus highlighting the interest of classifying OA phenotypes using relevant biomarkers in the clinic. Heterogeneity might also reflect different stages of OA in the different models but this is unlikely since the OA scores are similar in all models. The absence of a common signature could also be due to the late time point (6 weeks after OA induction and 24 months of age for old mice) chosen for the transcriptomic analysis when the gene expression profile might reflect an adaptive response. However, this time point is relevant for patients in whom OA is generally diagnosed long after disease initiation.
Another important finding is the identification of the combinatorial Gdf5-Cd36-Ltbp4 signature that might discriminate distinct subgroups of OA phenotypes. Indeed, Cd36 was upregulated in all mice that underwent surgical MNX. CD36 is a membrane-bound protein and the receptor of thrombospondin-1, fatty acid translocase (FAT), platelet glycoprotein 4 (PG4) and scavenger receptor class B member 3 (SCARB3). It is expressed in adipocytes and mesenchymal stromal cells isolated from fat tissue, and its expression level correlates with poor differentiation into the chondrogenic lineage
Differences in surface marker expression and chondrogenic potential among various tissue-derived mesenchymal cells from elderly patients with osteoarthritis.
. CD36 expression is increased at sites of cartilage injury and co-localizes with developing hypertrophic chondrocytes and the aggrecan NITEGE neo-epitope
The pattern recognition receptor CD36 is a chondrocyte hypertrophy marker associated with suppression of catabolic responses and promotion of repair responses to inflammatory stimuli.
. In patients with OA, CD36 expression has been significantly associated with the presence of osteophytes, of joint space narrowing, and higher Kellgren–Lawrence score
. Moreover, in chondrocytes from patients with OA, expression of thrombospondin 1 (a CD36 ligand) is strongly decreased concomitantly with the increase in CD36 expression
. More recently, the anti-inflammatory and analgesic effects of serum albumin in patients with knee OA was related to inhibition of CD36 in synoviocytes, macrophages and chondrocytes
. In addition, our study suggests that CD36 might be a specific biomarker of post-traumatic OA. CD36 expression should be thoroughly investigated in cartilage and bone samples from patients with different OA phenotypes.
We also found that Gdf5 expression was deregulated in three of the six OA models under study. It was previously shown that a loss-of-function GDF5 gene mutation results in joint fusions, and a single-nucleotide polymorphism is associated with higher susceptibility to OA
. Our data strongly suggest that GDF5 expression is a biomarker of painful OA phenotypes, as also suggested by genomic studies in humans.
Finally, we found that Ltbp4 was deregulated in all six OA models [Fig. 5(C)], although it was not identified as a deregulated gene common to all models in the statistical analysis [Fig. 3(D)]. In the bioinformatic analysis, SP-MNX samples were compared with MNX samples (Fig. 4) to investigate the impact of the genetic background on OA. Conversely, in the data presented in Figure 5(C), all groups were analysed independently of their control. LTBP4 is a key molecule required for the stability of the TGFβ receptor (TGFβR) complex via interaction with TGFβR2, thereby preventing its endocytosis and lysosomal degradation
. However, LTBP4 has not been associated with cartilage or OA and unlike its paralogues, LTBP4 is not regulated during chondrogenic differentiation of mesenchymal stromal cells
In-vitro analysis of the expression of TGFbeta -superfamily-members during chondrogenic differentiation of mesenchymal stem cells and chondrocytes during dedifferentiation in cell culture.
. Like Gdf5, Ltbp4 expression was decreased in old mice and not upregulated as observed in the murine models of induced OA. This suggests that spontaneous aging-related OA might involve different mechanisms.
In conclusion, the originality of the present study was to rely on relevant murine models of OA to understand the complexity of OA phenotypes in humans through investigation of the TGFβ pathway and based on rigorous SOPs. We did not identify a unique gene signature common to all six OA phenotypes. This highlights the huge heterogeneity of the animal models and the need of caution when extrapolating results from one model to another. But this also highlights that the diversity of the mouse models likely reflects the heterogeneity in human OA. Further studies are needed to validate these potential signatures.
Author contributions
All authors were involved in revising critically the manuscript and approved the final version. MM: Data analysis, manuscript writing; DN: Experiment design, data analysis, manuscript writing; HKE, DM, MR, EH, XH, DC, MCS, CJa, JYJ, MHLP, PR, JS, CV: Experimental work; FR, CJo, JG, FB: Experiment design, manuscript writing.
Conflict of interest
The authors declare that they have no competing interests.
Role of the funding source
Authors would like to thank the Fondation Arthritis that sponsored the network called ROAD (Research on OsteoArthritis Diseases) that included the seven academic French laboratories involved in this study. The sponsor had no role in the study design or in the collection, analysis or interpretation of the data. DM was granted a contrat d'interface by the Centre Hospitalier Régional Universitaire of Nancy.
Acknowledgments
Authors acknowledge Sandy Van Eegher (ROAD consortium) for her helpful contribution to ROAD, Laure Sudre and Audrey Pigenet (Centre de Recherche Saint-Antoine) for histology and Meriem Koufany (iMPOA) for her expert technical assistance. Thanks to the UTE Platform (SFR François Bonamy, FED 4203/Inserm UMS 016/CNRS 3556) who provided daily care to the animals, and to J. Lesoeur and M. Dutilleul from the SC3M platform (INSERM -U1229 RMeS, SFR François Bonamy, FED 4203/Inserm UMS 016/CNRS 3556, CHU Nantes) for histology in the age-related OA model. Authors also thank the “Réseau d’Histologie Expérimentale de Montpellier” histology facility for tissue processing and the “SMARTY platform and Network of Animal facilities of Montpellier”.
Appendix A. Supplementary data
The following is the Supplementary data to this article:
Supplementary FigRepresentative photographs of sagittal histological sections of femoral condyles from the different OA models.
Inducible chondrocyte-specific overexpression of BMP2 in young mice results in severe aggravation of osteophyte formation in experimental OA without altering cartilage damage.
Differences in surface marker expression and chondrogenic potential among various tissue-derived mesenchymal cells from elderly patients with osteoarthritis.
The pattern recognition receptor CD36 is a chondrocyte hypertrophy marker associated with suppression of catabolic responses and promotion of repair responses to inflammatory stimuli.
In-vitro analysis of the expression of TGFbeta -superfamily-members during chondrogenic differentiation of mesenchymal stem cells and chondrocytes during dedifferentiation in cell culture.
☆A collaborative study between seven expert laboratories in osteoarthritis pathophysiology rose the challenge to identify common gene signatures in six murine models of osteoarthritis representative of clinical phenotypes in osteoarthritic patients using standard operating procedures and centralized analyses.
☆☆The main findings of the study are the absence of one common deregulated gene in all six osteoarthritis models but the identification of a combinatorial Gdf5-Cd36-Ltbp4 signature that might discriminate distinct subgroups of OA.
31059-1&id=gr1.jpg){kind=link}
31059-1&id=gr2.jpg){kind=link}
31059-1&id=gr3.jpg){kind=link}
31059-1&id=gr4.jpg){kind=link}
31059-1&id=gr5.jpg){kind=link}
31059-1&id=figs1.jpg){kind=link}