Synovium-Derived MicroRNAs Regulate Bone Pathways in Rheumatoid Arthritis

K/BxN model of RA

Preparation of serum, generation of arthritis in C57BL/6 mice, and scoring methods were as described.5 Because male mice are more susceptible to arthritis, males were used at 12 weeks of age to avoid the juvenile growth spurt and any hormonal influence on bone. Mice were injected with 200 μL of serum from K/BxN mice intraperitoneally on days 0, 2, and 7. Mice were housed in specific pathogen-free conditions and arthritic mice had full access to food and water. All animal experiments were approved by the University of Massachusetts Medical School Institutional Animal Care and Use Committee (IACUC).

Tissue RNA preparation and qPCR

Synovium and navicular bones were isolated from mice in separate experiments, and processed for RNA as described.9 Synovium from six (day 10 [D10]) and twenty (day 0 [D0]) mice represented two pools (biological replicates) at each time point (three [D10] and 10 [D0] mice per pool). Total RNA was used for iScript cDNA synthesis (Bio-Rad Laboratories, Hercules, CA, USA). Quantitative reverse transcription polymerase chain reaction (qRT-PCR) was performed on diluted cDNA samples using iScript SYBR Green RT-PCR mix (Bio-Rad Laboratories, Hercules, CA, USA). Primers were purchased from Qiagen (Valencia, CA, USA). qPCR for miRNAs was performed using QuantiMir kits (System Biosciences [SBI], Palo Alto, CA, USA) and SYBR Green PCR mix (Bio-Rad Laboratories, Hercules, CA, USA). Forward primer sequences were obtained from miRBase.20 Gene and miRNA expression was normalized to hydroxymethylbilane synthase (HMBS) and U6 (U6 spliceosomal RNA) expression, respectively. Gene expression in nonarthritic controls was stable over time, and similar variability in gene expression among individual mice was observed at each time point. Data were analyzed using the delta-delta comparative threshold cycle algorithm (2^ΔΔCt) method.21

Cell culture

Fibroblast-like synoviocytes (FLS) were isolated by incubation of three C57BL/6 mouse hindlimbs with 1 mg/mL type IV collagenase and pooled (Worthington, Reading, UK) for 2 hours. Cells were maintained in DMEM supplemented with 10% FBS (Atlanta, GA), 2 mM L-glutamine (Corning, VA), 0.1 mM nonessential-amino acids (Mediatech Ink., VA), 1× essential amino acids (Mediatech Ink., VA), 50 mM 2-ME (MP Biomedicals, OH), 10 μg/mL gentamicin 100 μg/mL (Corning, VA), penicillin/streptomycin (Corning, VA). Cells were maintained at 37°C in a humidified 10% CO₂ environment and passaged every 3 days. After six passages, cells were plated in six-well plates, and once confluent, treated with 10 ng/mL TNF (R&D Systems, Minneapolis, MN, USA), 50 ng/mL IL-17 (R&D Systems, Minneapolis, MN, USA) or 10 ng/mL IL-1 (R&D Systems, Minneapolis, MN, USA) for 72 hours. RNA was collected for miRNA expression analysis using Trizol (Thermo Fisher Scientific, MA).

Calvarial osteoblasts were isolated from neonatal mice (postnatal day 0 [P0] to postnatal day 2 [P2]) by collagenase digestion (Roche, Basel, Switzerland) and maintained as described.22 Fifty nanomoles (50 nM) miRNA and anti-miRNAs (Thermo Fisher Scientific, Waltham, MA, USA) were introduced using oligofectamine (Invitrogen, Carlsbad, CA, USA), and cells were harvested for histochemical detection of alkaline phosphatase (Sigma-Aldrich, St. Louis, MO, USA) and Von Kossa staining (Sigma-Aldrich, St. Louis, MO, USA).

Gene array, miRNA array, and bioinformatic analysis

Gene array, miRNA array, and bioinformatic analysis are provided in the Supplemental Materials and Methods. The raw data are uploaded to GEO (http://www.ncbi.nlm.nih.gov/geo/; accession number: GSE71601).

Western blotting

Protein was extracted from calvarial osteoblasts in radioimmunoprecipitation assay (RIPA) buffer. Fifty micrograms (50 μg) were loaded onto a 10% SDS-PAGE gel, transferred to polyvinylidene fluoride (PVDF) membrane and probed with antibodies against Dkk2 (Abcam, Cambridge, UK) and beta-actin (Cell Signaling Technology, Beverly, MA, USA) and detected by chemiluminescence (Thermo Fisher Scientific, Waltham, MA, USA). Protein bands were measured by Image J software (NIH, Bethesda, MD, USA; https://imagej.nih.gov/ij/).23

Exosome collection from FLS culture media

Exosomes were isolated from FLS culture media. Cells were cultured in exosome-depleted FBS (System Biosciences [SBI], Palo Alto, CA, USA) medium with or without 10 ng/mL TNF for 3 days. For the isolation of exosomes, 50 mL of culture medium was centrifuged at 3000g for 15 min to remove cells and cell debris. The resulting supernatant was concentrated using the Amicon Ultra-15 Centrifugal Filter Unit with Ultracel-100 membrane (Millipore, Billerica, MA, USA). Exosomes were precipitated using ExoQuick-TC reagent (System Biosciences [SBI], Palo Alto, CA, USA) as described by the manufacturer. The exosome pellet was resuspended in PBS. Protein volume of exosomes was measured by Bradford assay (Bio-Rad Laboratories, Hercules, CA, USA). RNA was collected from 5 μg of exosomes using the RNeasy Micro Kit (Qiagen, Hilden, Germany).

3′ Untranslated region reporter luciferase assay

The 3′ untranslated region (3′UTR) reporter plasmid for Dkk2 was purchased from GeneCopoeia (Rockville, MD, USA). The 1074-bp fragment of the Dkk2 3′UTR region that contains the miR-221-3p binding site (TGTAGCA) was cloned into the pEZX-MT06 vector. The binding site is deleted in the mutant Dkk2 3′UTR reporter plasmid. Transfection of MC3T3 cells was performed in 96-well plates. The 3′UTR reporters are co-transfected with 25 nM miR-221-3p, or negative control (Thermo Fisher Scientific, Waltham, MA, USA) using Lipofectamine (Invitrogen, Carlsbad, CA, USA). The transfected cells were incubated for 24 hours to determine firefly luciferase activity and renilla luciferase activity using Dual-Glo luciferase assay system (Promega, San Luis Obispo, CA, USA). The experiment was repeated three times.

Statistical analysis

For bioinformatic analysis, principal component analysis (PCA) and ANOVA methods were used (see Supplemental Materials and Methods). Biological replicate average, fold change, and ANOVA p value were calculated between groups in Spotfire DecisionSite. Statistical significance was calculated using t tests for qPCR data. Data are presented as means ± SD or SE (for qPCR data from bone samples from six individual mice).

Gene and miRNA expression profiles in synovium

K/BxN STA replicates many features of chronic RA in humans in a synchronized manner.24 Therefore, to examine the possibility that synovium-derived miRNAs regulate skeletal pathways in RA, we analyzed expression of both miRNAs and mRNAs in whole synovial tissues from the STA model.5 We injected serum from K/BxN mice into 33 wild-type C57BL/6 mice. Thirty mice (91%) developed arthritis in paws and ankle joints. Peak inflammation occurred at approximately 10 days (D10) after initial serum injection (Fig. 1A). Expression levels at D10 were compared with those in nonarthritic mice (D0, uninjected). Gene array analysis confirmed upregulation at D10 of mRNA for the expected proinflammatory cytokines TNF and IL-1β and genes that inhibit Wnt signaling (eg, Sfrp-1, Sfrp-2), as well as downregulation of the Wnt inhibitor Dkk2, which augments osteoblast mineralization, confirming previous studies (Supplemental Fig.  1A).5

High-throughput TaqMan-based qPCR25 identified 168 miRNAs that passed quality control (QC) criteria (Supplemental Fig.  1B). We found few significant changes in miRNA expression between D10 and D21, when inflammation was resolving (data not shown). This may be because inflammation was not completely resolved at the D21 time point. We therefore pursued expression analysis between nonarthritic synovium and synovium at peak inflammation. Consistent with other RA models, miRNAs that regulate disease pathogenesis were identified in synovium at peak inflammation, including miR-23b-3p (downregulated 10-fold) and miR-221-3p (upregulated eightfold).15, 26 Having confirmed that miRNAs related to inflammation in our model were similar to previous reports, we focused on those miRNAs with differential expression during inflammation (D0 versus D10) that could affect bone.

Synovium-derived miRNAs regulate skeletal pathways

To analyze potential pathways regulated by synovial miRNAs, we compared gene and miRNA array results. Of the 168 miRNAs passing QC analysis, 72 miRNAs were significantly downregulated and 59 upregulated (≥1.5 fold) between nonarthritic and arthritic synovium. We compared the predicted 6756 target genes of the 72 downregulated miRNAs with the 929 upregulated genes identified in the gene array analysis (≥1.5 fold) (Fig. 1B). We similarly compared the predicted 6132 target genes of the 59 upregulated miRNAs with the 810 downregulated genes (≥1.5 fold) (Fig. 1C). We found 536 upregulated and 417 downregulated genes to be in the overlap of each respective comparative analysis (Venn diagram). To identify the biological pathways to which these genes belong, we performed ontology analysis using DAVID v6.7 (Fig. 1D, E).27 Interestingly, pathways identified in the upregulated gene group showed significant enrichment of skeletal pathways, including “Role of Osteoblasts, Osteoclasts, and Chondrocytes in Rheumatoid Arthritis” (p = 9.1E–06), “Wnt/β-catenin signaling” (p = 1.0E–04), and “RANK Signaling in Osteoclasts” (p = 6.8E–04) (Fig. 1D). These pathways included inhibitors of Wnt signaling, such as Dkk2, Dkk3, Sfrp-1, and Sost (Supplemental Table 1).

As expected, we also found inflammation-related pathways, including “Role of Macrophages, Fibroblasts, and Endothelial Cells in Rheumatoid Arthritis” (p = 2.8E–04), “VEGF Signaling” (p = 9.1E–04), and “Leukocyte Extravasation Signaling” (p = 1.2E–04), among others (Fig. 1D, E). The most enriched pathway was “Axonal Guidance Signaling” (p = 1.1E–07) (Fig. 1D). Notably, some genes in this pathway, such as ADAMTS7,28 HHIP,29 Sema3A,30 and MMP9,31 are also important regulators of skeletal development and bone formation (Supplemental Table 1). These results support the hypothesis that synovium-derived miRNAs not only regulate inflammation but also regulate skeletal pathways in RA.

Using statistical analyses, we constricted our definition of significant changes to determine the most critical candidate miRNAs regulating bone. Of the 168 miRNAs that passed QC analysis, 22 exhibited a greater than fourfold change (D10 versus nonarthritic) with an ANOVA p value <0.01 (Fig. 2A). Some of these miRNAs are known to play a role in inflammation. For example, miR-224 (current ID: miR-224-5p) is upregulated and miR-145 (current ID: miR-145a-5p) downregulated in T cells from systemic lupus erythematosus patients32 and miR-16 (current ID: miR-16-5p) is decreased in serum from early stage RA patients.33 We also found that miR-224-5p is upregulated and miR-145a-5p and miR-16-5p are downregulated at peak inflammation (Fig. 2A).

We performed PCA on the 22 candidate miRNAs to identify similarities and determine functional miRNA clusters affecting biological processes. Candidate miRNAs segregate into three clusters based on expression changes (Fig. 2B). Target genes of miRNAs in each cluster are enriched in distinct skeletogenesis pathways (Table 1). miRNAs in cluster I are downregulated and their target genes are enriched in cartilage matrix pathways and osteogenic signaling, including JAK/STAT,34 IGF-1,35 and HIF1α signaling.36 miRNAs in cluster II are downregulated and their target genes are enriched in Wnt/β-catenin signaling and cell cycle–related pathways. miRNAs in cluster III are upregulated and their target genes are enriched in pathways regulating bone metabolism, such as glucocorticoid receptor37 and eNOS signaling.38 Within these 22 miRNAs, several are identified that have been reported to regulate skeletal pathways. For example, miR-133a-3p and miR-204-5p (cluster II) inhibit osteoblast differentiation by suppressing expression of Runx2.39, 40 miR-145a-5p (cluster II) inhibits differentiation of osteoblasts and chondrocytes by suppressing Osx and Sox9.41, 42 Downregulation of these miRNAs may contribute to maintaining chondrocyte and osteoblast progenitors during bone erosion as a compensatory mechanism. Thus, the most highly regulated miRNAs in inflamed synovium are commonly involved in regulation of skeletal pathways.

Downregulated miRNAs are predicted to target inhibitors of skeletal pathways

We and others have previously observed expression of mRNA and protein for the Wnt signaling inhibitors, Sfrp-1 and Dkk1 in bones at erosion sites in models of RA.6, 9 Although Runx2-expressing immature osteoblasts were present at bone erosion sites, maturation of these osteoblasts is not frequently observed.9 Here we examined expression of Wnt and BMP signaling components using RNA prepared from navicular bone, a reproducible site of erosion in STA, and found mRNA expression of Gsk3β, Sfrp-2, and Tob1, as well as an early osteoblast marker, Osx, to be upregulated in bones at erosion sites during peak inflammation (Supplemental Fig.  2). Upregulation of Gsk3β, Sfrp-2, and Tob1 would be predicted to inhibit osteoblast differentiation, whereas upregulation of Osx would promote early osteoblast differentiation. These results suggest that BMP signaling, Wnt signaling, and maturation of osteoblasts are suppressed in bone at erosion sites. Upregulation of Osx mRNA expression may provide a counter-regulatory mechanism, or may be due to an inhibition of osteoblast maturation at this site. To identify candidate miRNAs that may regulate skeletal pathways at erosion sites, we searched for miRNAs that are predicted to target components of major skeletal pathways using TargetScan v6.2. Among the 22 candidate miRNAs showing a greater than fourfold change in expression, we identified 12 miRNAs predicted to target components of major skeletal pathways (Fig. 2C). These miRNAs have known roles in inflammation,26, 32, 43-46 proliferation,47 apoptosis,48 or musculoskeletal systems39-42, 49 (Fig. 2C). Notably, nine downregulated miRNAs in clusters I and II target numerous inhibitors of Wnt and BMP signaling, including Dkk3, GSK3β, Sfrp-1, Sfrp-2, Smad7, Tob1, and Tob2,50 suggesting that decreased expression of these miRNAs in synovium may contribute to the induction of BMP and Wnt signaling inhibitors and limit bone formation at erosion sites.

To validate expression of these 12 miRNAs in synovial fibroblasts, we isolated FLSs from wild-type mice, stimulated with proinflammatory cytokines including TNF, IL-1, or IL-17 and examined expression of candidate miRNAs. We confirmed that miR-486-5p, miR-708-5p, miR-186-5p, and miR-145a-5p are downregulated following stimulation with TNF treatment (Fig. 3). We also found that IL-1 suppresses expression of miR-486-5p, miR-708-5p, and miR-186-5p, and IL-17 suppresses expression of miR-186-5p. Expression of miR-133a-3p, miR-204-5p, and miR-182-5p is suppressed only by TNF (data not shown). We validated expression of these miRNAs (miR-708-5p, miR-186-5p, miR-486-5p, miR-145a-5p) by qPCR in whole synovial tissues at D10 compared to nonarthritic tissues using the same synovial samples used for array experiments. Taken together, we found numerous FLS-derived miRNAs that are predicted to target inhibitors of BMP and Wnt signaling and that are downregulated by inflammatory cytokines. These miRNAs may negatively regulate bone formation at erosion sites.

Overexpression of miR-221-3p suppresses differentiation of primary osteoblasts

miR-224-5p, miR-221-3p, and miR-28-5p are among the most highly upregulated miRNAs in cluster III (Fig. 2B) that target components of the BMP and Wnt pathways. (Fig. 2C). We confirmed that miR-221-3p is induced in TNF-treated FLS (Fig. 4A) and in inflamed synovium (Fig. 4B)26. miR-224-5p and miR-28-5p did not show relevant expression patterns in FLS and may be regulated in other synovial cell types such as hematopoietic or endothelial cells, or by other cytokines. miR-221-3p was also reported to be induced in a TNF-driven model of arthritis and in FLS from RA patients.26 We therefore focused on miR-221-3p and questioned whether miR-221-3p might affect skeletal pathways at the erosion site. Since miRNAs can be exported from one cell type and taken up to regulate gene expression in a distant cell type, we examined expression of miR-221-3p in FLS-derived exosomes. We found that with treatment of FLS cells with TNF (10 ng/mL), miR-221-3p is significantly upregulated in FLS-derived exosomes (Fig. 4C). This result suggests that secretion of miR-221-3p into exosomes from synovium is also upregulated by inflammation and that synovium-derived miR-221-3p may regulate bone formation at erosion sites. Ontology analysis implicated miR-221-3p as targeting the Wnt pathway26; however, Wnt reporter activity was not affected by miR-221-3p in HEK 293T cells.26 Using TargetScan v6.2, miR-221-3p is predicted to target bone anabolic genes including Runx2, TCF4, Esr1,51 and Dkk2.10 We therefore transfected miR-221-3p into mouse primary calvarial osteoblasts. This inhibited differentiation and mineralization, as indicated by decreased staining for alkaline phosphatase (Alp) and mineralization (von Kossa) (Fig. 4D). We did not find significant change in osteoblast differentiation by anti-miR-221-3p (qPCR data for osteoblast markers not shown). This may indicate that only excess miR-221-3p affects osteoblast differentiation but not loss of miR-221-3p expression. We further observed a significant decrease in mRNA expression of Alp and a mature osteoblast marker, Osteocalcin (Ocn) by miR-221-3p (Fig. 4E). mRNA expression of Tcf-1, a readout of Wnt signaling,52 was also suppressed by miR-221-3p (Fig. 4E), confirming that miR-221-3p inhibits osteoblast differentiation. Among the predicted pro-osteoblastic target genes of miR-221-3p, we found the protein level of Dkk2 was suppressed in the transfected osteoblasts (Fig. 4F). miR-221-3p is known to bind the 3′UTR region of Dkk2 mRNA in 293T cells.53 We confirmed the binding of miR-221-3p to the 3′UTR region of Dkk2 mRNA in the osteoblastic cell line MC3T3 (Fig. 4G), supporting Dkk2 as a target gene of miR-221-3p. These results suggest that upregulation of synovium-derived miR-221-3p may suppress expression of Dkk2 in osteoblasts, and subsequent osteoblast maturation, and may, at least in part, contribute to the inhibition of bone formation in RA erosions (Fig. 4H).