Detection of microRNA expression

Expression levels of 365 microRNAs were evaluated with TaqMan microRNA microarray assays as previously described [21]. Validation of these results was performed using the mirVana qRT–PCR miRNA Detection Kit and qRT–PCR Primer Sets, according to the manufacturer's instructions (Ambion Inc, TX, USA). The U6 small nuclear RNA was used as an internal control.

MicroRNA Northern Blot Analysis

For microRNA Northern Blot Analysis, 10ug of RNA were separated on 12% denaturating polyacrylamide gels and transferred to GeneScreen Plus membrane (PerkinElmer, Waltham). MiRCURY LNA Probes for miR-483 and miR-22 (Exiqon, Denmark) were end-labeled with T4 polynucleotide kinase. Prehybridization of the filters was carried out in 50% formamide, 0.5% SDS, 5· SSPE, 5·Denhardt's solution and 20 mg/ml sheared, denatured salmon sperm DNA. Hybridizations were performed in the same solution at 42°C. The labeled probes were heated for 1 min at 95°C before addition to the filters in the prehybridization solution. After hybridization, the membranes were washed in 0.1 SSC, 0.1% SDS at 42°C twice for 10 min.

Reverse-Phase protein microarray analysis

Chondrocyte cell lysates were boiled for 5 min and were loaded into 384-well plates in serial dilutions (neat, 12, 14, 18, and 116) with negative control wells containing only lysis buffer. These samples were printed in duplicate onto nitrocellulose-coated glass slides (Schleicher & Schuell Bioscience, Keene, NH) using a ring-and-pin robotic arrayer (GMS 417, Affymetrix, Santa Clara, CA). The arrays were stained as previously described [22] on an autostainer (DAKO, Carpinteria, CA) using a biotinyl-linked catalyzed signal amplification system (DAKO). Specificity of each antibody was tested by western blot analysis.

Statistical analysis

All calculations were performed on a Microsoft computer, using the SPSS software (version 12.0). Correlation of between microRNA and protein expression levels with BMI was identified by correlation coefficients, calculated by Pearson rank correlation (r) and Spearman rank correlation. Statistical methods regarding the proteomic analysis are described analytically in the suppl. Methods section. Construction and statistical significance of gene networks was performed by Ingenuity pathway analysis. Statistical significant networks were considered those with p value higher than 10⁻⁵. In addition clustering of the protein data in functional groups was performed using DAVID NIH Bioinformatics Database with a p value cut-off of 10⁻⁵. Quantification of western blots was performed by standard densitometric analysis. All transfection experiments were performed in triplicate and the results were compared by student's t-test analysis.

microRNA expression correlates with BMI

Clinical characteristics of the patients and normal individuals ( Table 1 ) allowed us to study potential correlations between microRNAs expression and clinicopathological parameters. Very interestingly, we found five microRNAs to be statistically correlated with body mass index (BMI) ( Figure 1C ). miR-22 and miR-103 expression was positively correlated with BMI, while miR-25, miR-337 and miR-29a expression was inversely correlated, pointing towards the potential role of microRNAs in lipid metabolism and osteoarthritis pathogenesis.

Proteomic analysis of articular cartilage

In order to identify deregulated proteins in osteoarthritic chondrocytes and study in detail whether obesity and osteoarthritis are correlated at a molecular level, we performed proteomic analysis in the same articular cartilage samples that we performed microRNA expression analysis. Specifically reverse-phase protein arrays were constructed [23] ( Figure S1A ) and probed with antibodies to 214 proteins expressed in articular cartilage ( Table S3 ). All antibodies were tested for their quality and specificity by Western blot analysis ( Figure S1B ) Arrays were scanned and dilution curves were used to quantify relative protein expression ( Figure S1C ). We detected that 76 proteins, (48 up-regulated and 28 down-regulated) were differentially expressed between osteoarthritic and normal chondrocytes ( Figure 2A ). These results were validated by western blot analysis ( Figure 2B ). We were able to identify for the first time deregulated proteins in osteoarthritic chondrocytes that were implicated in inflammatory and lipid metabolism pathways. More specifically, we detected up-regulation of proteins involved in inflammatory pathways such as IL1B, IL6 and CCR3, while proteins (PPARA, PPARG, ACOX1) involved in lipid metabolism mechanisms, were found highly down-regulated in osteoarthritic chondrocytes ( Figure 2A, B ). In addition we detected novel proteins that were deregulated in osteoarthritis, such as SOX11, FGF23, KLF6, WWOX and GDF15.

Open in a separate window

Figure 2

Reverse phase protein arrays in osteoarthritic and normal cartilage tissues.

(A) Differentially expressed proteins between osteoarthritic and normal chondrocytes. Up-regulated are shown with red color, while down-regulated with green color. (B) Representative western blot analysis in protein extracts from five osteoarthritic tissues in comparison with normal cartilage. (C) Sub-cellular localization of differentially expressed proteins. (D) Functional clustering analysis of differentially expressed proteins (using DAVID NIH Bioinformatic Database). (E) Correlation coefficient wheel between protein expression levels of differentially expressed proteins in osteoarthritic vs normal chondrocytes and body mass index (BMI). We identified 3 protein groups, which showed statistically significant correlations between protein expression and BMI, according to the coefficient correlation index (r²). More specifically, the first group with the highest correlation (r²>0.900) consisted of PPARA, BMP7, IL1B, LEP (leptin) and SREBP1 proteins. The second group (0.600>r²>0.900) consisted of ITGA5, ADIPOQ (adiponectin), FGF23, MMP13, RETN (resistin) and SOX9. The third group (0.400>r²>0.600) consisted of 3 proteins (HADHA, ADAMTS5, PPARG) which had low degree of correlation with BMI. The rest of the proteins were not correlated with BMI.

Around half of the deregulated proteins in osteoarthritic chondrocytes were located in the extracellular space (50.67%), while the rest had cytoplasmic (20%), plasma membrane (16%) and nuclear (13.33%) localization ( Figure 2C ). Functional clustering analysis categorized the differentially expressed proteins in nine statistically significant pathways. Specifically, 45.33% of the deregulated proteins were involved in cartilage homeostasis pathways ( Figure 2D ), while 18.67% and 9.33% were involved in lipid metabolism and inflammation pathways, respectively.

Metabolism-related proteins correlate with BMI

Furthermore, we tried to correlate the expression levels of differentially expressed proteins in osteoarthritic cartilage with clinicopathological parameters, such as Body Mass Index. A recent study suggested that BMI is a significant risk factor for knee osteoarthritis leading to arthroplasty, speculating that biomechanics and metabolic factors associated with adipose tissue contribute to this phenotype [5]. We identified that PPARA, BMP7, IL1B, LEP, ITGA5 and SREBP1 protein levels in osteoarthritic chondrocytes were highly correlated with BMI ( Figure 2E ), suggesting the potential role of metabolic-related proteins in the development of osteoarthritis.

Detection of microRNA gene targets

Since microRNAs exert their biological functions through suppression of target genes, it is important to identify microRNA-target pairs. As the available bioinformatic algorithms predict hundreds of microRNA-gene target pairs, it is evident that experimental data are needed for verification of these pairs. However, as it is known that several microRNAs target gene expression only at the protein and not at the mRNA level, the integration of microRNA along with protein data sets could be considered more effective for microRNA-gene target verification. Recently Huang et al., used microRNA with cDNA expression profiling data to identify human microRNA targets using Bayesian data analysis algorithm [16]. In our study, we identified microRNA-gene target pairs by matching microRNA and protein data. Subsequently, we filtered these data through three different selection criteria ( Figure S2 ) and revealed 17 microRNA-gene target pairs implicated in osteoarthritis pathogenesis ( Table S4 ). More specifically, we found microRNA-gene target pairs potentially involved in cartilage homeostasis and structure (miR-377-CART1, miR-140-ADAMTS5, miR-483-ACAN, miR-23b-CRTAP, miR-16-TPM2, miR-223-GDF5, miR-509-SOX9, miR-26a-ASPN), in biomechanic pathways (miR-25-ITGA5), in apoptotic mechanisms (miR-373-CASP6, miR-210-CASP10) and in lipid metabolism pathways (miR-22-PPARA, miR-22-BMP7, miR-103-ACOX1, miR-337-RETN, miR-29a-LEP). Several of these target genes such as ADAMTS5, GDF5 and LEP have been previously correlated with osteoarthritis [3], [24], [25].

Gene networks in osteoarthritis

It is becoming increasingly clear that most proteins interact in complex cellular networks, the properties of which might be altered in osteoarthritic compared to normal chondrocytes. Therefore, global strategies aimed at modeling the functional interrelationships between microRNA and proteins, as complex interdependent networks, are required. We integrated microRNA and protein data sets in order to generate a model of macromolecular network that is perturbed in osteoarthritis using Ingenuity program analysis [26]. The resulting network contained 11 microRNAs, 58 proteins and 414 potential functional associations ( Figure 3A ). We were able to detect three sub-networks representing key functional units that make up the co-expression network. A metabolism-related, an inflammation and a cartilage homeostasis sub-network were found to be interrelated contributing all together to cartilage destruction and osteoarthritis development ( Figure S3 ).

Open in a separate window

Figure 3

Interactome network in osteoarthritis.

(A) Construction of an interactome network (p=10-⁴⁹) by integrating microRNA and proteomic data using Ingenuity Pathway Analysis (IPA) (more information in suppl. methods). The p value indicates the likelihood of focus genes to belong to a network versus those obtained by chance. Around half of the microRNAs (mir-337, miR-29a, miR-22, miR-103-1, miR-25) regulate genes involved in the metabolic pathways. (B) Correlation coefficients between microRNA expression levels and their gene targets protein levels in osteoarthritic and normal chondrocytes. (C) Predicted duplex formation between PPARA and BMP7 3′UTR with miR-22. (D) BMP7 and PPARA protein levels after miR-22 or inhibitor of miR-22 (as-miR-22) treatment (50 nM) for 48 h in normal and osteoarthritic chondrocytes, respectively.

miR-22 correlates with PPARA and BMP7 protein expression

The functional significance of our predicted “interactome” network was tested by experimental validation. Our finding that specific microRNAs and proteins were related to BMI, focused our interest in identifying functional microRNA-gene target pairs relating obesity with osteoarthritis pathogenesis mechanisms. Epidemiological studies have shown that the risk for knee osteoarthritis is increased by 36% for every 2 units of BMI (5 kg) of weight gain [27]. At first we tried to identify which microRNA-gene target interactions have biological significance (inverse correlation in expression levels) in our network. According to our previous combined in silico and expression data analysis we identified 5 microRNAs (miR-22, miR-103, miR-337, miR-25, miR-29a) and their 6 targets (PPARA, BMP7, ACOX1, RETN, ITGA5, LEP) that were highly correlated with BMI ( Figure 1C , Figure 2E ). Correlation of microRNA-gene target expression levels revealed that miR-22 was highly inversely correlated with PPARA (r²=0.919) and BMP7 (r²=0.816). ( Figure 3B ), while we detected low correlation between miR-29a with LEP (r²=0.491), miR-25 and ITGA5 (r²=0.456), miR-337 and RETN (r²=0.385) and miR-103 and ACOX1 (r²=0.252). The above results suggested the potential functional relationship between miR-22, PPARA and BMP7.

miR-22 regulates PPARA and BMP7 in normal and osteoarthritic chondrocytes

In order to detect whether PPARA and BMP7 were direct targets of miR-22, we performed luciferase assay. We found that BMP7-encoded mRNA contains a 3′UTR element that is partially complementary to miR-22 ( Figure 3C ) and luciferase assay showed that BMP-7 is a direct target of miR-22 (69% reduction, p<0.001, Figure S4A ). Furthermore, miR-22 overexpression in chondrocytes reduced the activity of a luciferase reporter gene fused to the PPARA 3′UTR (52% reduction, p<0.001, Figure S4B ). Evaluation of BMP7 and PPARA mRNA expression levels after miR-22 expression revealed that only BMP7 mRNA levels were significantly down-regulated (p<0.001), while PPARA were not (p=0.492) ( Figure S5 ). Western blot analysis revealed that miR-22 regulates both PPARA and BMP7 protein expression levels in normal and osteoarthritic chondrocytes ( Figure 3D ). Overexpression of miR-22 inhibited BMP-7 (76%) and PPARA (93%) protein expression in normal chondrocytes. Subsequently, inhibition of miR-22 in osteoarthritic chondrocytes by antisense miR-22 treatment, highly up-regulated BMP-7 (8.35 fold) and PPARA (12.55 fold) expression, suggesting that miR-22 is a strong regulator of BMP-7 and PPARA proteins. All above data suggest that miR-22 regulates BMP-7 at the mRNA level and PPARA at the protein level, revealing thus the advantage of using proteomic instead of cDNA microarray data for detecting microRNA gene targets.

Metabolic, inflammatory and cartilage homeostasis networks are inter-related

Subsequently, we proceeded by investigating how these miR-22 target gene pairs are correlated with the rest of the proteins present in the “interactome” network. Specifically, PPARA belongs to the metabolism sub-network, which is connected with the inflammation sub-network through IL1B. A recent report suggested that PPARA is a receptor involved in inflammatory processes [28] and was recently found down-regulated in osteoarthritic cartilage [29]. In our study, the inflammatory sub-network having IL1B and IL6 as central nodes is connected with MMP13, which is central node of the cartilage structure sub-network. To test this hypothesis predicted by the gene network of IL1B-MMP13 interaction, we treated normal and osteoarthritic chondrocytes with IL1B and examined MMP13 expression. We found that IL1B up-regulated MMP13 expression both at mRNA and protein levels ( Figure 4A, B ), verifying the correlation that we had previously described between IL1B and MMP13 expression levels in clinical samples [30]. In order to understand how IL1B affects not just MMP13 but cartilage homeostasis pathways, we over-expressed IL1B in normal chondrocytes and monitored the expression of the proteins involved in the cartilage network. IL1B over-expression pertubated the cartilage homeostasis sub-network by activation of metalloproteinases and aggrecanases and down-regulation of cartilage structural proteins. ( Figure 4C ), connecting thus inflammation and cartilage homeostasis sub-networks with osteoarthritis development.

Open in a separate window

Figure 4

IL1B regulates important components of cartilage homeostasis network.

(A) Treatment with IL1B (10 ng/ml) induces MMP-13 mRNA levels assessed by Real-time PCR analysis in normal and osteoarthritic chondrocytes. (B) ELISA assay detecting MMP-13 levels after IL-1b treatment of osteoarthritic chondrocytes. (C) Pertubation of IL1B affects important gene network components in normal chondrocytes. Treatment of normal chondrocytes with IL1B (10ng/ml) for 48 h affects the protein expression of cartilage structure related genes (red color shows up-regulation while green color down-regulation of protein expression). This experiment was performed in quadruplicate. Specifically there is activation of metalloproteinases 3 and 13 (MMP3, MMP13) and aggrecanases (ADAMTS4, ADAMTS5) leading to down-regulation of the cartilage structural proteins (ACAN, SPARC, COMP, TPM2, MATN3, COL2A1). In addition asporin (ASPN) is up-regulated which has been shown to inhibit TGF-beta and aggrecan synthesis (look ref 11). Furthermore SOX9, an important trascription factor implicated in chondrogenesis is highly down-regulated.

PPARA and BMP7 regulate IL1B and MMP13 expression in chondrocytes

In order to delineate the PPARA-IL1B-MMP13 potential pathway we followed an RNA interference strategy. More specifically, siRNA inhibition of PPARA increased IL1B (2.5 fold) and MMP13 (3.5 fold) expression levels ( Figure 5A ). IL1B was found to be highly induced 24 h after siRNA PPARA treatment, while MMP13 was highly induced 48 h after siRNA treatment, suggesting a sequential activation of MMP13 by IL1B.

Open in a separate window

Figure 5

PPARA and BMP7 signaling pathways in chondrocytes.

(A) Assessment of IL1B and MMP13 mRNA levels after down-regulation of PPARA, 24 and 48 h after siRNA liposomal treatment into chondrocytes. (B) IL1B and MMP13 expression 24 and 48 h after BMP7 siRNA treatment. (C) Evaluation and correlation of BMP7 and ACAN (aggrecan) mRNA levels in normal and osteoarthritic chondrocytes assessed by real-time PCR analysis. (D) Aggrecan expression levels 48 h after BMP7 inhibition of expression using siRNA transferred by liposomes into normal chondrocytes. All experiments have been performed in triplicate.

The second target of miR-22, BMP7, is frequently down-regulated in osteoarthritic cartilage, while BMP7 overexpression induces cartilage formation in vitro and in vivo [31]. It has been shown that IL1B and ACAN (aggrecan) [32] levels are regulated by BMP7 in osteoarthritic cartilage. These correlations were present in the cartilage homeostasis sub-network and in addition we found that siRNA against BMP7 resulted in increased IL1B and MMP13 (2 fold) expression in normal chondrocytes ( Figure 5B ). Furthermore, we detected high correlation between BMP7 and ACAN mRNA levels ( Figure 5C ) and found that BMP7 siRNA down-regulation blocked effectively ACAN expression ( Figure 5D ).