Inhibition of TET1 prevents the development of osteoarthritis and reveals the 5hmC landscape that orchestrates pathogenesis

5hmC increases in joint cartilage after PTOA induction

To identify changes in the 5hmC landscape in OA, we used the DMM model of PTOA, in mice 4 to 6 months of age (Fig. 1, A and B) (25, 26). In this model, surgical intervention causes a destabilization of the knee joint, mimicking many of the known features of human disease and producing more reliable disease initiation than spontaneous age-associated OA (29). Only male mice were used for these studies (31) because sex hormones are known to cause lower disease severity with altered dynamics in female mice. Immunofluorescence on the knee joint showed that DMM mice had twofold (P = 0.01) and fivefold (P = 0.0028) increase in global 5hmC at 6 and 8 weeks after surgery (Fig. 1, C and D, and fig. S1A). These changes are mainly attributable to DMM-induced OA because no increase in global 5hmC was observed in mice when aged in the absence of surgical induction (fig. S1B). To determine the sites of 5hmC accumulation genome-wide, we performed reduced representation hydroxymethylation profiling (RRHP) (32) to assess individual CCGG sites. Globally, we identified 43,267 differentially hydroxymethylated CCGG (dhCCGG) sites 6 weeks after surgery, with 93% of these sites gaining 5hmC during OA induction (Fig. 1E). Upon exploring the distribution of these dhCCGG sites, we noted that they were predominantly gained in gene bodies (46%), where 5hmC accumulation has been associated with gene activation, and in intergenic regions (45%) (Fig. 1F) (23). Only a small fraction of the dhCCGG sites (8%) were found in promoter regions (Fig. 1F), which we broadly defined as ±5000 base pairs from the transcription start site (TSS). It was difficult to discern the precise 5hmC changes in enhancer regions, which can be found near TSS or in intergenic regions, because no datasets have defined enhancers in mouse articular chondrocytes.

• Download high-res image
• Open in new tab
• Download Powerpoint

To monitor the effects of this 5hmC deposition on gene expression, we performed RNA-seq on cartilage microdissected from sham and DMM mouse joints 6 and 12 weeks after surgery. Eight hundred ten genes were found to be up-regulated 6 weeks after DMM surgery. Forty-six percent of these up-regulated genes showed a gain in 5hmC. Three hundred forty-seven of the activated genes gained 5hmC in gene bodies. On the other hand, only 60 genes gained 5hmC in the promoter regions, and 32 of these genes also gained 5hmC in gene bodies (Fig. 1G). In the core set of 375 activated genes, we observed an enrichment for pathways related to WNT signaling, protein kinase A activity, and inositol metabolism (Fig. 1H). The 347 activated genes that gained 5hmC in gene bodies included several extracellular matrix (ECM) remodeling enzymes (Fig. 1I).

A total of 911 genes were up-regulated 12 weeks after DMM as compared to the sham control, and 106 of these genes were already up-regulated 6 weeks after DMM, revealing the early-stage genes that persist toward the later stages of PTOA. In addition, 43% of the genes up-regulated at 12 weeks in the DMM joints had gained 5hmC at 6 weeks (fig. S1C), suggesting that 5hmC also marks genes that are important in later stages of OA. This is consistent with our studies of human OA cartilage (23, 24). Upon cross-referencing the datasets, we found that 52% of the genes that gain 5hmC in the DMM model accumulated 5hmC in the end-stage human OA chondrocytes as well (Fig. 1J). Our findings in both mouse and human samples therefore corroborate that 5hmC gain is a hallmark of OA pathogenesis.

TET1 mediates OA-related 5hmC deposition

Oxidation of 5mC to 5hmC, 5fC, or 5caC can be catalyzed by any of the three members of the TET family (20). The short form of TET1 (TET1s) is expressed in somatic tissues and has been shown to participate in targeted binding and oxidation events (33–35). Unlike TET2 and TET3, TET1’s catalytic domain is not processive; rather, it releases 5hmC intermediates (36), putatively explaining how stable 5hmC can build up in many genomes (37), including that of differentiating chondrocytes (21). Therefore, conceptually, TET1 is the ideal candidate responsible for 5hmC deposition during OA induction despite being the lowest expressed of the three TET enzymes in the knee cartilage (fig. S2A). Immunohistochemistry of 4-month-old mouse joints confirmed the expression of TET1 in a subpopulation of cells in joint cartilage (fig. S2B), consistent with human data (24, 38).

Recent work has posited that OA is a disease of the whole joint, with multiple tissues involved, including bone and synovium. Therefore, to test whether TET1 was necessary for the 5hmC deposition during OA development, we performed DMM on a Tet1^−/− mouse model that lacked TET1 in the whole joint (39). Unexpectedly, Tet1^−/− mice showed no deterioration of the cartilage architecture 6 weeks after surgery (Fig. 2A). Grading of the DMM joints using Osteoarthritis Research Society International (OARSI) scoring (27) exhibited statistically significant lower (P < 0.0001) maximum and summit scores and decreased cartilage loss in the Tet1^−/− mice compared to the Tet1^+/+ mice (Fig. 2, B and C). Accordingly, Tet1^−/− mice had lower matrix metalloproteinase-13 (MMP-13) expression in articular chondrocytes of the joint compared to Tet1^+/+ control mice (Fig. 2, D and E). Tet1^−/− DMM mice lost 5hmC at 65% of CCGG sites compared to Tet1^+/+ DMM mice (Fig. 2F) and failed to gain 5hmC at 98% of the dhCCGGs that showed increased 5hmC in Tet1^+/+ DMM mice (Fig. 2G). Collectively, these data demonstrate that TET1 is the major TET protein responsible for the 5hmC gain upon DMM. We did not observe any significant change (P > 0.05) in Tet1 expression upon DMM induction in the RNA-seq data at 6 or 12 weeks (fig. S2C), suggesting that the increase in 5hmC deposition was due to either increased activity or increased targeting.

• Download high-res image
• Open in new tab
• Download Powerpoint

We next wanted to determine whether the protection from OA in the Tet1^−/− DMM mice would persist into later disease stages. Upon assessing joints from mice 12 weeks after surgery, which mimic late-stage human disease (29, 30), we observed that the Tet1^−/− DMM mice maintained intact cartilage and proteoglycan staining (fig. S2D), had limited loss of cartilage thickness (fig. S2E), and reduced MMP-13 production and secretion (fig. S2F). In addition, micro–computed tomography (micro-CT) analyses of the DMM joints showed the formation of osteophytes and ectopic bone in the Tet1^+/+ but not Tet1^−/−, DMM mice (Fig. 2, H and I). To complement these data, we also studied the joints in uninjured, 18-month-old Tet1^+/+ and Tet1^−/− mice to assess any spontaneous development of OA. A variable onset of OA was observed in wild-type mice, and although not statistically significant (P > 0.05), the Tet1^−/− mice trended toward being protected from OA (fig. S2, G and H). Together, these data show that TET1 is the hydroxymethylase predominantly responsible for the accumulation of pathogenic 5hmC in OA and that its loss impairs both OA initiation and progression.

TET1-mediated 5hmC controls activation of key OA pathways including WNT, metalloproteinases, and signal transducer and activator of transcription 3 signaling

We next sought to understand how the TET1-mediated 5hmC landscape in chondrocytes drives the molecular events of OA. RNA-seq analyses of the microdissected cartilage of DMM mice at 6 weeks revealed that multiple OA-related pathways, including retinoic acid receptor (RAR), WNT, and interleukin-1 (IL-1) signaling networks, failed to activate in Tet1^−/− DMM mice compared to Tet1^+/+ DMM mice (Fig. 3, A to C). To determine all ECM regulators involved, we overlapped the differentially expressed gene lists of Tet1^+/+ and Tet1^−/− with the murine matrisome database (40). We observed 16 significant (P < 0.05) gene expression changes, including metalloproteinases Adams and Adamts (Fig. 3D). Comparison of the 5hmC maps between the Tet1^+/+ and Tet1^−/− DMM mice showed a loss of 5hmC at OA-associated genes involved in key pathways like HIF-1A (hypoxia-inducible factor 1A), JAK (Janus kinase)/STAT (signal transducer and activator of transcription), and NF-κB (nuclear factor κB) (Fig. 3, E and F).

• Download high-res image
• Open in new tab
• Download Powerpoint

To determine which of these genes were direct targets of TET1, we overlapped genes losing both 5hmC and gene expression in the Tet1^−/− DMM joints (Fig. 3G), resulting in 313 targets. Of these, 59% gained 5hmC during Tet1^+/+ OA induction. Upon assembling a protein interaction network using STRING (41), these direct TET1 target genes were found to be responsible for controlling several important nodes in the homeostasis of cartilage, including mTOR signaling, WNT signaling, apoptosis, and chemokine signaling (Fig. 3H). Some members of the mitogen-activated protein kinase (MAPK) family, including Mapk1 (extracellular signal–regulated kinase 2) and Mapk14 (p38), were found to be down-regulated in Tet1^−/− DMM cartilage (Fig. 3I). In addition, several other kinases, including Oxsr1, which activates pathways related to oxidative stress, tumor necrosis factor (TNF) signaling, and apoptosis, were direct target genes (table S1).

Among the 313 direct target genes, several were transcription factors (TFs) whose down-regulation might explain the TET1-dependent gene expression changes that are independent of 5hmC. These included Cxxc5, Gata2, Klf3, Kl9, Tcf3, and Yy1 (table S1). Among these, YY1 has been associated with rheumatoid arthritis and is known to up-regulate the expression of IL-6 and inhibit the expression of IL-8 (42, 43). In addition, it acts as an activator for Gdf5 (44) and inhibitor of miR-10a, whose inhibition leads to increased NF-κB activation (45). Several chromatin regulators were also down-regulated, including Setd8, Kdm2b, and Prdm16 (table S1).

To determine which TFs might be sensitive to TET1-mediated 5hmC deposition, we used Enrichr (46) to analyze expression data from knockout studies of known TFs. In the top 30 enriched sets, we found TFs with known roles in OA, such as Creb1, Gata1, Stat3, and Yap1 (Fig. 3J), that can act upstream of the 313 direct TET1 targets. Previous work on the interaction between TET1 and STAT3 suggests a model in which TET1-mediated 5hmC deposition allows for TF binding at its target sites and subsequent target activation (47). These TFs may also represent the factors that facilitate the binding of TET1 to their target genes during OA initiation.

TET1 is also able to interact with the polycomb repressive complex 2 and act as a repressor (48). To determine whether the loss of this repressor activity led to the up-regulation of any chondroprotective genes in the Tet1^−/− mice in a 5hmC-independent manner, we focused on targets that were up-regulated in the Tet1^−/− sham compared to Tet1^+/+ control sham mice. Pathway analyses on the 263 up-regulated genes showed an enrichment for negative regulation of Toll-like receptor 4 signaling and protein catabolic processes. Sham Tet1^−/− mice also had a 20-fold up-regulation of Dot1l (P = 0.0004), which was previously shown to be chondroprotective in OA by repressing sirtuin 1 (SIRT1)–induced expression of the WNT pathway (49). DNA methyltransferase 3b (Dnmt3b) was also 20-fold up-regulated (P = 0.0004), whose gain of function has been shown to be chondroprotective through altering metabolism and mTOR signaling (9), further underscoring the broad effects of TET1 on multiple aspects of cartilage regulation. The genes up-regulated in Tet1^−/− sham mice were not enriched for any ECM components, as tested using a curated matrisome list (40). After DMM induction, however, we observed that Tet1^−/− mice up-regulated pathways related to N-glycan biosynthesis (Dolpp1) and sialic acid biosynthesis that are involved in the modification of ECM proteins, along with chondroprotective genes like Smad4, Pparg, and Timp1 (Fig. 3K). Therefore, although the Tet1^−/− mice do not up-regulate ECM genes directly, they might preserve cartilage integrity by decreasing metalloproteinase activity and through glycan metabolism.

Down-regulation of inflammatory pathways, such as PPARγ (peroxisome proliferator–activated receptor γ) and IL-12 signaling, was maintained in the 12-week-old Tet1^−/− DMM mice (fig. S3, A and B). Other target genes related to ECM degradation, cell cycle, and cytoskeletal rearrangement (Adamtsl4, Nek1, and Wipf1) were also persistently down-regulated in the Tet1^−/− joints 12 weeks after DMM induction. Together, these data collectively highlight the multiple TET1-regulated pathways (Fig. 3K), demonstrating that TET1 is a key epigenetic regulator of OA.

Loss of Tet1 protects cartilage from OA without impairing the systemic response to inflammation

Our analyses of 5hmC deposition and gene expression provided an overview of the cartilage intrinsic mechanisms of protection in the Tet1^−/− mice. Next, we asked whether Tet1 loss predominantly affected cartilage or also caused a systemic dampening of low-grade inflammation (50). We first assessed synovial inflammation by (i) the extent of immune filtration in the synovium and (ii) the amount of inflammatory cytokines in the serum of the DMM mice. DMM joints were immunostained for F4/80, a marker of monocyte and macrophage immune cell lineages, 6 weeks after surgery. We noted that immune cell infiltration in the synovial lining was not affected because we observed twofold more infiltration of F4/80-positive immune cells in the Tet1^−/− joints than in the Tet1^+/+ DMM joints (Fig. 4, A and B). We also observed increased expression of Cxcr4 and Lcp2, two markers of B cells, in Tet1^−/− joints in the RNA-seq dataset. Thus, we ruled out a potentially lowered immune infiltration as the mechanism of protection from OA in the in Tet1^−/− mice.

• Download high-res image
• Open in new tab
• Download Powerpoint

To assess the low-grade inflammation typically observed in osteoarthritic mice, we characterized serum from mice 6 and 12 weeks after surgery using a 39-plex Luminex cytokine panel. In Tet1^+/+ DMM mice, we detected an up-regulation of LEPTIN, CCL7 (C-C motif chemokine ligand 7), and EOTAXIN over the course of disease progression. When compared to their own baseline, Tet1^−/− mice also had increased amount of LEPTIN (at 6 and 12 weeks), CCL7 (at 12 weeks), and EOTAXIN (at 6 and 12 weeks); however, the absolute quantity was lower than Tet1^+/+ controls (Fig. 4C). These data ruled out a potentially heightened inflammatory microenvironment due to increased infiltration of F4/80⁺ immune cells in the synovium. However, the decrease in OA-associated cytokines LEPTIN, CCL7, and EOTAXIN (51–53) suggested that loss of TET1 partially modulated inflammation in the joint microenvironment. To exclude any systemic immune defects in the Tet1^−/− mice, as observed in the Tet2^−/− mice (54), we challenged mice with an acute infection by intraperitoneal lipopolysaccharide (LPS) injection. Both Tet1^−/− and Tet1^+/+ mice mounted a comparable, robust response against LPS (Fig. 4D and fig. S4A). In particular, CCL7 and EOTAXIN were up-regulated to a similar extent in both Tet1^−/− and Tet1^+/+ mice.

Last, we challenged Tet1^−/− and Tet1^+/+ mice with IL-1β injections to test the cartilage response to direct cytokine stimulation in the joint. Mouse knee joints were injected every other day with IL-1β over a course of 6 days (Fig. 4E). As previously described (55, 56), wild-type animals exhibited loss of proteoglycan content in cartilage after injections as observed by Safranin O staining (Fig. 4, F and H). In contrast, proteoglycan staining in Tet1^−/− cartilage was still high (Fig. 4, F and H). In line with these findings, MMP-13 secretion, a target of IL-1β (57), was ~25% lower in Tet1^−/− joints compared to wild type (Fig. 4, G and I). These data were corroborated by in vitro studies using differentiated ATDC5 cells, a mouse chondrogenic cell line (58). We observed that ~75% Tet1 knockdown through short hairpin RNA (shRNA) inhibited the up-regulation of Mmp3, Mmp13, and IL-6 in response to IL-1β stimulation (fig. S4B). Collectively, these data demonstrate that Tet1^−/− cartilage is intrinsically protected from the effects of the proinflammatory cytokine milieu in the joint.

As a corollary, we wanted to test whether TET1 was sufficient to activate cytokine-induced gene expression changes in healthy chondrocytes. For this, we used chondrocytes isolated from three normal human donors with no history of OA. We found no difference in the amount of MMP3 or CCL2 in Tet1-overexpressing chondrocytes (fig. S5). Likewise, both control and Tet1-overexpressing cells had the same magnitude of MMP3 and CCL2 up-regulation after treatment with IL-1β or with a combination of recombinant TNFα and oncostatin M (fig. S5). This suggests that although Tet1 is necessary for the expression of these OA-associated genes, it is not sufficient.

Pharmacological inhibition of TET activity prevents OA

Next, we wanted to directly test the therapeutic potential of TET1 inhibition in OA. As a proof-of-principle experiment, we performed shRNA-mediated Tet1 knockdown on three samples from patients with OA, obtaining a 40 to 70% knockdown and a subsequent 80 to 90% reduction in 5hmC (fig. S6A and Fig. 5A). Consistent with the mouse data, we found that the silencing of TET1 decreased MMP3 and MMP13 expression (fig. S6B). To assess any low-grade inflammation in the OA cells, we used a cytokine array and observed that Tet1 knockdown significantly (P < 0.001) decreased concentration of IL-6, IL-8, vascular endothelial growth factor (VEGF), vascular cell adhesion molecule 1 (VCAM1), plasminogen activator inhibitor 1 (PAI1), and CCL2 (Fig. 5, B and C). The plurality of pathways altered by the loss of Tet1 validated our strategy in targeting an upstream epigenetic regulator rather than perturbing each pathway separately.

• Download high-res image
• Open in new tab
• Download Powerpoint

Next, we tested the feasibility of modulating TET1 activity pharmacologically using a known TET inhibitor, 2-HG (59). Treatment of chondrocytes from three patients with OA with 2-HG resulted in a 25% decrease in 5hmC (Fig. 5D). Although 2-HG is a broad inhibitor affecting other demethylases besides TET (60), we observed effects similar to Tet1 knockdown on MMP3 and MMP13 expression and proinflammatory cytokine secretion (Fig. 5, E and F, and fig. S6C) despite some expected variability between patients.

Last, we asked whether early intervention with the 2-HG inhibitor could prevent the onset of OA in vivo. Tet1^+/+ control mice were injected with 2-HG twice a week for 5 weeks after DMM surgery (Fig. 5G). Periodic administration of the inhibitor in the joints was sufficient to reduce cartilage degradation (Fig. 5, H and I), along with decreasing the secretion of MMP-13 (Fig. 5J) and reducing 5hmC (Fig. 5, K and L), akin to the genetic loss of TET1. This demonstrates that 2-HG is active within the joint environment and can alter the pathological expression of 5hmC observed during OA onset. Furthermore, we tested the expression of three direct TET1 targets (Oxsr1, Igfbp7, and Adamtsl4) identified in Tet1^−/− mice in inhibitor-treated joints by quantitative polymerase chain reaction and found their expression down-regulated (Fig. 5M). Collectively, these data show that pharmacological inhibition of TET mirrors the genetic loss of TET1, further confirming our hypothesis that TET inhibition would be therapeutic for OA.