Estrogen-related receptor γ causes osteoarthritis by upregulating extracellular matrix-degrading enzymes

The estrogen-related receptor (ERR) family of orphan nuclear receptor is composed of ERRα, ERRβ, and ERRγ, which are known to regulate various isoform-specific functions under normal and pathophysiological conditions. Here, we investigate the involvement of ERRs in the pathogenesis of osteoarthritis (OA) in mice. Among ERR family members, ERRγ is markedly upregulated in cartilage from human OA patients and various mouse models of OA. Adenovirus-mediated overexpression of ERRγ in mouse knee joint or transgenic expression of ERRγ in cartilage leads to OA. ERRγ overexpression in chondrocytes directly upregulates matrix metalloproteinase (MMP)-3 and MMP13, which are known to play crucial roles in cartilage destruction in OA. In contrast, genetic ablation of Esrrg or shRNA-mediated downregulation of Esrrg in joint tissues abrogates experimental OA in mice. Our results collectively indicate that ERRγ is a novel catabolic regulator of OA pathogenesis.

O steoarthritis (OA), the most common form of arthritis, is a leading cause of disability and incurs a large socioeconomic cost 1 . OA is a whole-joint disease characterized by cartilage destruction, synovial inflammation, osteophyte formation, and subchondral bone sclerosis 2 . However, no effective disease-modifying therapy for OA has been developed to date. OA is caused by an anabolic/catabolic factor imbalance that can be induced by various etiologic risk factors and pathophysiological processes 3,4 . Important among the potential OA-causing mechanisms are mechanical stresses, including joint instability and injury, and factors that predispose toward OA, such as aging. These factors alter biochemical pathways in chondrocytes, resulting in degradation of the extracellular matrix (ECM) 5 . Among the matrix-degrading enzymes, matrix metalloproteinase 3 (MMP3), MMP13, and ADAMTS5 (a disintegrin-like and metallopeptidase with thrombospondin type 1 motif 5; aggrecanase-2) are known to play crucial roles in OA cartilage destruction [6][7][8] . Various catabolic regulators, including the proinflammatory cytokine, interleukin (IL)-1β, can upregulate MMP3, MMP13, and ADAMTS5 9 ). We previously demonstrated that hypoxia-inducible factor (HIF)-2α (encoded by Epas1), which is transcriptionally upregulated in chondrocytes by proinflammatory cytokines or mechanical stress, causes OA pathogenesis through upregulation of matrix-degrading enzymes 10,11 . We also recently demonstrated that the zinc-ZIP8-MTF1 axis-reflecting induction of the zinc importer ZIP8 (encoded by Slc39a8), zinc influx, and subsequent activation of the zinc-dependent transcription factor MTF1-acts as a crucial catabolic regulator of OA pathogenesis by upregulating matrixdegrading enzymes in joint articular chondrocytes 12,13 .
OA is currently considered as a disease associated with metabolic disorders 14 . Although this is still a matter of controversy 15 , a number of studies suggest an association of metabolic syndrome with OA pathogenesis 16 . Among the various regulatory molecules involved in metabolism, isoforms of estrogen-related receptors (ERRs) are currently known to regulate various metabolic . d-f ERR mRNA levels were quantified by qRT-PCR (n = 10 mice per group) in cartilage tissues from mice subjected to sham operation or DMM surgery (d). Cartilage destruction was scored by OARSI grading (e). Representative images (n = 12 mice per group) of safranin-O-stained and ERRγ-immunostained OA cartilage from DMM-operated mice or cartilage from sham-operated mice (f). Values are presented as means ± s.e.m. (*P < 0.01 and **P < 0.001; ns, not significant). Two-tailed t test (a,d) and Mann-Whitney U test (e). Scale bar: 50 μm processes, including metabolic disorders and energy metabolism 17,18 . The ERR family consists of the orphan nuclear receptors, ERRα (NR3B1), ERRβ (NR3B2), and ERRγ (NR3B3). Unlike other nuclear orphan receptors, ERRs are constitutively active owing to the active conformation of their ligand-binding domain, even in the absence of a ligand 19 . Because ERRs share nearly identical DNA-binding domains, their transcriptional regulatory functions are similar to each other; thus, they regulate overlapping common target genes. For instance, ERRα and ERRγ regulate energy metabolism and promote oxidative capacity in skeletal and cardiac muscle 18,20,21 . However, individual ERRs also exhibit isoform-specific functions. For example, ERRγ regulates iron homeostasis in various tissues 22,23 , ERRβ regulates embryonic stem cell pluripotency and self-renewal 24,25 , and ERRα acts as a determinant of breast cancer etiology 24,25 . The roles of ERRs in cartilage biology are beginning to be investigated, with most studies focusing on the functions of ERRα 26 . For instance, studies have shown that ERRα activates the Sox9 promoter in a rat chondrogenic cell line 27 and rat mandibular condylar chondrocytes 28 , and is required for cartilage development in zebra fish 29 . In contrast, cartilage-specific ERRγ transgenic (Tg) mice exhibit chondrodysplasia, in which repressed chondrocyte proliferation decreases the proliferative zone of the growth plate and reduces bone length 30 . It has also been reported that ERRα expression is reduced in bones and joints during collagen-induced inflammatory rheumatoid arthritis 31 . However, the functions of ERRs in regulating joint cartilage homeostasis and/or OA pathogenesis have not yet been studied, although an in vitro study by Bonnelye et al. 32 suggested a possible association of ERRα with OA pathogenesis. They demonstrated that IL-1β in human OA chondrocytes upregulates ERRα, which directly targets genes encoding MMP3 and SOX9.
In this study, we used human OA cartilage and mouse models of OA to investigate whether ERRs are associated with OA pathogenesis. Here, we report that ERRγ is a novel catabolic regulator of OA pathogenesis. We found that, among the ERR family members, ERRγ was specifically upregulated in cartilage from human OA patients and various mouse models of OA. Our gain-of-function (adenovirus-mediated overexpression in joint tissues or Col2a1-Esrrg Tg mice) and loss-of-function (Esrrg +/mice or shRNA-mediated knockdown) approaches clearly indicated that ERRγ acts as a novel catabolic regulator of OA pathogenesis, at least in part, by the upregulating matrixdegrading enzymes, MMP3 and MMP13, in articular chondrocytes.

Results
ERRγ is upregulated in chondrocytes of OA cartilage. To explore the possible association of ERRs with OA pathogenesis, we first examined the expression levels of ERRs in OA cartilage of human patients and various mouse models. Our quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analyses revealed that ERRγ was significantly increased in OAaffected, damaged regions of human cartilage compared with undamaged regions of arthritic cartilage (Fig. 1a). Although the fold-increases in ERRα and ERRβ mRNAs were statistically significant, their expression levels were negligible relative to those of ERRγ (Fig. 1a). Consistent with the increased mRNA levels, immunostaining revealed that ERRγ protein levels were markedly elevated in chondrocytes of human OA cartilage (Fig. 1b, c). Similarly, ERRγ, but not ERRα or ERRβ, was significantly increased in cartilage from an experimental mouse model of OA induced by destabilization of the medial meniscus (DMM) surgery ( Fig. 1d-f). ERRγ was also increased in the cartilage of mice  . qRT-PCR analysis (n ≥ 8) of ERRγ mRNA levels in chondrocytes treated with IL-6 siRNA and exposed to IL-1β (middle panel). qRT-PCR analysis (n ≥ 8) of ERRγ mRNA levels in chondrocytes treated with AG490 and exposed to IL-6 (right panel). Values are presented as means ± s.e.m. (*P < 0.05, **P < 0.005, and ***P < 0.0005. One-way ANOVA) with OA caused by the overexpression of HIF-2α via intraarticular (IA) injection of an adenovirus-expressing HIF-2α (Ad-Epas1) 10,11 or by IA injection of Ad-Slc39a8 to overexpress the zinc importer, ZIP8 12,13 ( Supplementary Fig. 1a, b). However, we were unable to detect ERRα in OA cartilage of humans or mice ( Supplementary Fig. 2a, b). Additionally, IL-6, which acts as a catabolic regulator of OA cartilage destruction 33 , increased mRNA levels of ERRγ, but not ERRα or ERRβ, in primary cultured mouse articular chondrocytes ( Supplementary Fig. 2c). Because these results collectively suggested that ERRγ was associated with OA pathogenesis, we focused on the role of ERRγ in OA pathogenesis.
ERRγ expression is regulated by multiple signaling pathways, including alcohol-induced activation of the cannabinoid receptor, IL-6-mediated inflammation, fasting-induced glucagon receptor activation, feeding-induced insulin receptor activation, hypoxia, and endoplasmic reticulum stress 17 . In an attempt to identify regulators of ERRγ expression, we examined the expression of its upstream signaling components in primary cultured chondrocytes treated with various OA-associated catabolic regulators, including IL-1β 9 , HIF-2α 10,11 , and ZIP8 12,13 , using microarray analyses. Among the known ERRγ upstream regulators, IL-6 expression was markedly increased by IL-1β treatment and, to a lesser degree, by the overexpression of HIF-2α or ZIP8 (Supplementary Table 1). We therefore further characterized the involvement of IL-6 signaling in ERRγ expression in chondrocytes. Similar to microarray analyses, our qRT-PCR analyses indicated upregulation of ERRγ by IL-1β treatment and, to a lesser degree, by overexpression of HIF-2α or ZIP8, in primary cultured mouse articular chondrocytes (Fig. 2a). IL-1β also increased IL-6 protein levels in chondrocytes, as determined by immunofluorescence microscopy and Western blotting (Fig. 2b, c and Supplementary Fig. 9). IL-6 treatment of chondrocytes increased ERRγ mRNA levels in a concentrationand time-dependent manner (Fig. 2e). IL-6 also increased mRNA levels and secreted levels of MMP3 and MMP13 protein (Fig. 2d, e), which are encoded by target genes of IL-6 and are associated with OA cartilage destruction 33 . Additionally, IL-1β-induced upregulation of ERRγ was abrogated by knockdown of IL-6 with specific small interfering RNA (siRNA) (Fig. 2f). IL-6-mediated inflammation is known to regulate ERRγ expression via JAK/ STAT signaling 17 . Consistent with this, inhibition of JAK/STAT signaling with AG490 34 abrogated IL-6-induced ERRγ expression (Fig. 2f). These results collectively indicate that the IL-1β/IL-6 axis contributes to ERRγ expression in articular chondrocytes. This is consistent with our previous work showing that IL-6 plays an essential role in OA cartilage destruction by upregulating MMP3 and MMP13 33 .
Ectopic expression of ERRγ causes OA pathogenesis. To examine the in vivo role of ERRγ in OA pathogenesis, we ectopically expressed ERRγ in knee joint tissues of 10-week-old male mice via IA injection of an adenovirus-expressing ERRγ (Ad-Esrrg). We previously demonstrated that an adenovirus system effectively delivers genes to cartilage and other joint tissues [10][11][12][13] . Consistent with our prior findings, immunohistochemical staining indicated that ERRγ was effectively overexpressed in cartilage ( Fig. 3a), as well as the meniscus and synovium ( Supplementary  Fig. 3a). ERRγ overexpression, triggered by three weekly IA injections of Ad-Esrrg, caused synovitis, as determined by H&E staining (Fig. 3b). At this stage, safranin-O staining indicated loss of glycosaminoglycans in articular cartilage above the tidemark, without apparent development of osteophyte or subchondral bone remodeling (Fig. 3b). However, after 8 weeks, joint tissues IA-injected with Ad-Esrrg showed osteophyte development and subchondral bone sclerosis, with severe erosion of cartilage ( Fig. 3c, d). We also saw a loss of glycosaminoglycans in articular cartilage in both juvenile (8-week-old) and young adult (10-weekold) mice ( Supplementary Fig. 3b), suggesting that developmental stage is not a critical determinant of ERRγ-induced cartilage damage. Indeed, overexpression of HIF-2α, which was previously shown to cause OA cartilage destruction 10,11 , induced cartilage damage in mice at various ages from 8 to 20 weeks old (Supplementary Fig. 3c). IA injection of Ad-Esrrg caused ectopic expression of ERRγ in all joint tissues, including the synovium and cartilage (Supplementary Fig. 3a). To examine the cartilage-specific functions of ERRγ in OA pathogenesis, we generated cartilage-specific ERRγ Tg mice (Col2a1-Esrrg) using the Col2a1 enhancer and promoter [10][11][12][13]35 . Primary cultured chondrocytes from Col2a1-Esrrg Tg mice exhibited upregulation of ERRγ at both mRNA and protein levels ( Fig. 4a and Supplementary Fig. 9). ERRγ protein levels were also elevated in the cartilage and meniscus, but not synovial tissues, of Col2a1-Esrrg Tg mice ( Fig. 4b and Supplementary Fig. 4a). Because a previously generated cartilage-specific ERRγ Tg mouse was reported to exhibit chondrodysplasia, in which repressed chondrocyte proliferation decreases the proliferative zone of the growth plate and reduces bone length 30 , we characterized skeletal development in our Tg mice. Skeletal staining of E18.5 embryos revealed that skeletal structure was similar in Tg mice and wild-type (WT) littermates ( Supplementary Fig. 4b). Additionally, alcian blue staining of the metatarsal bone of 2-week-old mice revealed that the lengths of resting/proliferative and hypertrophic zones of the growth plate were similar ( Supplementary Fig. 4c). Thus, in contrast to the previous report 30 , our data collectively indicate no significant difference in skeletal development between our Tg mice and WT littermates.
Compared with WT littermates, DMM-operated Col2a1-Esrrg Tg mice exhibited significantly more cartilage destruction, as   Fig. 4d). Other DMM-induced manifestations of OA, such as subchondral bone sclerosis and osteophyte formation, were also significantly enhanced in Col2a1-Esrrg Tg mice compared with WT littermates (Fig. 4c, d). We used 10-week-old mice for DMM surgery and examined OA manifestations at 8 weeks post surgery, because this technique was originally developed for use at this age of the mouse 36 and most DMM surgeries are performed at this age 37,38 ( Supplementary Fig. 5a, b). Indeed, we found no apparent differences in DMM surgery outcomes at ages between 8 and 20 weeks (Supplementary Fig. 5c). Thus, the results of our gainof-function studies collectively indicate that ERRγ is a catabolic regulator of OA pathogenesis.
The above results indicate that deletion of one allele of Esrrg is sufficient to reduce experimental OA in mice. Therefore, we further validated the function of ERRγ in OA pathogenesis by knocking down ERRγ in whole-joint tissues of mice via IA injection of adenovirus-expressing shRNA against Esrrg (Ad-shEsrrg). DMM-induced upregulation of ERRγ in the chondrocytes of cartilage tissue was effectively abrogated by IA injection of Ad-shEsrrg in DMM-operated mice (Fig. 6a). In our initial screening, we found that infection with Ad-shEsrrg effectively downregulated ERRγ expression in primary cultured mouse articular chondrocytes. Consistent with this, knockdown of cartilage ERRγ significantly suppressed all examined OA manifestations, including cartilage erosion, subchondral bone sclerosis, and osteophyte formation, in DMM-operated mice (Fig. 6b, c). These results further support the conclusion that ERRγ functions as a catabolic regulator in mouse models of OA pathogenesis caused by DMM surgery.
ERRγ upregulates MMP3 and MMP13 in articular chondrocytes. Matrix-degrading enzymes produced by chondrocytes are essential effector molecules in the destruction of cartilage during OA pathogenesis 5 . Indeed, genetic studies in mice have demonstrated that MMP3, MMP13, and ADAMTS5 play crucial roles in OA cartilage destruction [6][7][8] . Here, we examined the effects of ERRγ on the expression of these matrix-degrading enzymes in primary cultured mouse articular chondrocytes. We found that overexpression of ERRγ via infection with Ad-Esrrg significantly elevated the mRNA levels and extracellular secreted protein levels of MMP3 and MMP13, without affecting ADAMTS4 or -5 ( Fig. 7a and Supplementary Figs. 7a, b, and 9). In contrast, mRNA levels of SOX9, type II collagen (Coll-II), and aggrecan were reduced by overexpression of ERRγ, although the degree of inhibition was small ( Supplementary Fig. 7a, b). Consistent with these findings, MMP3 and MMP13 proteins were upregulated in Ad-Esrrg-injected cartilage tissue, whereas SOX9 and type II collagen protein levels were markedly reduced (Fig. 7b). We used chromatin immunoprecipitation (ChIP) assays to examine whether ERRγ directly regulates the expression levels of Mmp3 and Mmp13 in articular chondrocytes. Our nucleotide sequence analysis identified a number of putative ERR response elements (ERRE, TCAAGGTCA) 40 in the promoter regions of Mmp3 and Mmp13 (Fig. 7c). ChIP assays revealed that ERRγ was bound to ERRE #1 (-2093 to -2083) and ERRE #2 (-2489 to -2478) in the Mmp3 promoter, and to ERRE #1 (-1094 to -1291) and #2 (-1676 to -1668), but not ERRE #3 (-2389 to -2377), in the Mmp13 promoter ( Fig. 7d; Supplementary Fig. 7c). These findings suggest that Mmp3 and Mmp13 are direct target genes of ERRγ in mouse articular chondrocytes.
Inhibition of ERRγ blocks experimental OA in mice. Next, we examined whether ERRγ could be a therapeutic target for OA. Because IL-6, an upstream signaling component involved in regulating ERRγ expression, upregulates MMP3 and MMP13 in chondrocytes 33 , we first examined whether IL-6-induced upregulation of MMP3 and MMP13 requires the transcriptional activity of ERRγ. We found that treatment with GSK5182, an inverse agonist of ERRγ 22 , significantly inhibited IL-6-induced upregulation of MMP3 and MMP13 in primary cultured chondrocytes (Fig. 7e). To validate ERRγ as a potential therapeutic target for OA, we intraperitoneally administrated sham-and DMM-operated mice with GSK5182 (20 mg kg -1 body weight). We found that GSK5182 significantly reduced all DMM-induced manifestations of OA, including cartilage destruction, osteophyte development, and subchondral bone sclerosis (Fig. 7f). These results collectively suggest the possibility that ERRγ could serve as a therapeutic target for OA. Finally, although mRNA and protein levels of ERRα were unchanged in OA cartilage of human patients and mouse models, we examined whether overexpression of ERRα via IA injection of Ad-Esrra in joint tissues causes cartilage damage. IA injection of Ad-Esrra caused overexpression of ERRα in cartilage (Supplementary Fig. 8a). However, this overexpression of ERRα did not cause any detectable cartilage damage or synovial inflammation ( Supplementary Fig. 8b, c), clearly indicating that ERRγ, but not ERRα, functions as a catabolic regulator of OA in mice.

Discussion
The ERR isoforms, ERRα, ERRβ, and ERRγ, regulate various physiological functions, either in common or in an isoformspecific manner. However, whether ERRs regulate OA pathogenesis in vivo has not been investigated to date. Here, we demonstrated that, among ERR isoforms, ERRγ regulates OA pathogenesis in mice by causing cartilage destruction, osteophyte formation, and subchondral bone sclerosis. The results of both gain-of-function (adenovirus-mediated overexpression of ERRγ in mouse joint tissues or Col2a1-Esrrg Tg mice) and loss-offunction (Esrrg +/mice or shRNA-mediated silencing of Esrrg) experiments clearly indicate that ERRγ functions as a critical catabolic regulator of OA pathogenesis.
The role of ERRs in cartilage biology has not been extensively studied, and most existing studies have focused on ERRα 26 . Among the earliest evidence supporting a possible functional association of ERRα with cartilage development was the spatial and temporal correlation of ERRα expression with cartilage development 41 . Although a relatively large number of studies have reported on the role of ERRα in chondrogenesis and cartilage development [26][27][28] , only a few reported the possible role of   42 , which might reflect the operation of an autoregulatory feedforward mechanism in the regulation of ERRγ gene expression 43 . Similar to our observation, it has been shown that IL-1β-induced upregulation of ERRα in chondrocytes is reduced by the ERRα inverse agonist XCT790 32 . This suggests that both ERRα and ERRγ expression in chondrocytes are regulated by positive feedback mechanisms. We have previously reported amplification of HIF-2α catabolic signaling in chondrocytes during OA pathogenesis. For instance, HIF-2α signaling leading to the production of the matrix-degrading enzymes, MMP3 and MMP13 10 , is amplified by the HIF-2α targets, IL-6 32 and NAMPT (visfatin) 11 . Additionally, the NAMPT downstream target SIRT also stimulates HIF-2α transcriptional activity by stabilizing HIF-2α protein 44 . Furthermore, two essential catabolic . c The ERRE sequences in the promoter regions of Mmp3 and Mmp13 are indicated. d Quantitative ChIP assays were performed, and the data are presented as fold-changes relative to each input (n ≥ 4). e qRT-PCR (n = 11) analyses of MMP3 and MMP13 expression in chondrocytes treated with IL-6 (100 ng ml -1 ) in the presence of the indicated concentrations of GSK5182, an inverse agonist of ERRγ. f Sham-operated or DMM-operated mice were IP-injected with GSK5182 and killed 8 weeks after the operation. Representative images of safranin-O staining and scoring of OARSI grade, osteophyte size, and subchondral bone plate thickness (n = 10 mice per group). Values are presented as means ± s.e.m. (*P < 0.05, **P < 0.005, and ***P < 0.0005). One-way ANOVA (a,d,e), two-tailed t test (f), and Mann-Whitney U test (f). Scale bar: 50 μm signaling pathways in OA pathogenesis-HIF-2α and the zinc-ZIP8-MTF1 axis-reciprocally activate each other, resulting in amplification of catabolic signaling 45 . Collectively, these observations suggest that catabolic signaling processes involved in OA pathogenesis are amplified upon onset of the disease.
The only study to date describing a role for ERRγ in cartilage is one by Cardelli et al. 30 , who reported that cartilage-specific overexpression of ERRγ results in chondrodysplasia and reduced chondrocyte proliferation. These authors found a reduction in the length of the femur and tibia in Tg mice, with no differences in intramembranous ossification. They additionally found that the overexpression of ERRγ impaired chondrocyte proliferation, leading to a decrease in the size of the proliferative zone in the growth plate 30 . However, our Tg mice exhibited no marked differences in the whole skeleton of embryos (E18.5) and length of resting/proliferative and hypertrophic zones in 2-week-old mice. Although it is not clear what accounts for these differences, they may be attributable to differences in expression systems 46 and expression levels of target genes. We previously reported that cartilage-specific Dickkopf-1 Tg mice, generated using the same system as the current Tg mice, exhibit reduced experimental OA 47 without differences in cartilage or bone development 46 . We also showed that cartilage-specific Tg mice generated by our system, including those overexpressing HIF-2α 10 , NAMPT 11 , BATF 35 , or ZIP8 12 , exhibited enhanced experimental OA without detectable differences in skeletal development. In addition to the difference in expression system, we also found that the expression level of ERRγ protein in our Tg mice was less than that in mice generated by Cardelli et al. 30 . We postulate that these differences in the expression levels of ERRγ contribute to the observed differences. Nevertheless, the enhanced expression of matrixdegrading enzymes in cartilage of Tg mice compared with WT mice appears to be responsible for the enhanced DMM-induced OA pathogenesis in Tg mice.
The prevalence of OA is higher among women than men, and this prevalence increases after menopause. Various studies using animal models indicate that estrogen and estrogen signaling are associated with OA pathogenesis 48 . For instance, mice lacking estrogen receptor-α develop more and larger osteophytes and a thinner lateral subchondral plate 49 . A lack of estrogens also increases subchondral bone remodeling 50 . However, although the term "estrogen-related receptor (ERR)" reflects the molecular origin of the three family members, ERRs are functionally different from estrogen receptors. For instance, ERRs do not bind to natural estrogens, nor are they directly involved in classic estrogen-signaling pathways or biological processes 18,24 . Furthermore, ERRs and estrogen receptors display strict binding site specificities 51 , with ERRs exhibiting a marked preference for the sequence TCAAGGTCA, an ERR response element (ERRE). Furthermore, unlike estrogen receptors, ERRs do not require ligand binding for their transcriptional activity 18,24 . Therefore, the catabolic functions of ERRγ in OA pathogenesis might not be related to estrogen, even though estrogen is known to be associated with OA pathogenesis 48 .
Cartilage destruction is one of the most prominent characteristics of OA and is primarily caused by the upregulation of matrix-degrading enzymes, such as MMP3, MMP13, and ADAMTS5 5-8 . These matrix-degrading enzymes originate from chondrocytes of cartilage tissue or from synovial cells, such as fibroblast-like synoviocytes (FLS) 5 . Our approaches using Esrrg +/mice, shRNA-mediated silencing, and adenovirus-mediated overexpression affected ERRγ expression in all joint tissues, limiting our ability to identify specific joint tissues associated with OA pathogenesis. We observed, however, that ERRγ was upregulated in the cartilage tissue, but not the synovial tissue, of DMM-operated mice, and that ERRγ caused upregulation of MMP3 and MMP13 in primary cultured articular chondrocytes. These results strongly support the interpretation that ERRγ causes OA pathogenesis by upregulating MMP3 and MMP13 in articular chondrocytes. We also found that ERRγ downregulated SOX9, type II collagen, and aggrecan, although the degree of downregulation was small. Interestingly, compared with in vitro effects of ERRγ overexpression, ERRγ overexpression via IA injection of Ad-Esrrg in cartilage tissue caused marked decreases in the protein expression levels of SOX9 and type II collagen. The decrease in type II collagen might reflect its MMP-mediated degradation in ERRγ-overexpressing cartilage tissue. Thus, it may be possible that the reduction in SOX9 expression in Ad-Esrrginjected cartilage is attributable to both direct and indirect actions of ERRγ. In addition to cartilage destruction, OA manifestations include osteophyte formation and subchondral bone sclerosis, reflecting the fact that OA is a whole-joint disease affecting all tissues of the joint, including cartilage, synovial tissues, and subchondral bone 2 . However, the relative contributions, relationships, and key tissues involved in each event that occurs during OA pathogenesis have not been elucidated 2,3 . For instance, it is not yet clear whether subchondral bone sclerosis is a driving force for cartilage breakdown or a consequence of articular cartilage catabolism. Therefore, the regulatory mechanisms of ERRγ in osteophyte formation and subchondral bone sclerosis remain to be elucidated.
Here, we demonstrated that intraperitoneal administration of GSK5182 reduces DMM-induced OA pathogenesis, although it remains to be clarified whether this is attributable to inhibition of ERRγ in the tissues, since intraperitoneal administration might affect ERRγ in the whole body. However, although cartilage is an avascular tissue, OA is not merely a local disease confined to the joint, and many soluble mediators that regulate OA pathogenesis are found in circulating blood 52 . Additionally, the human pathology may differ in certain respects from that in mouse models. However, we found that expression patterns of ERRγ as well as ERRα and ERRβ were similar in OA cartilage of human patients and various mouse models. Our results are the first to demonstrate that ERRγ acts as a catabolic regulator of cartilage degeneration and OA pathogenesis, and collectively support the idea that ERRγ could be a therapeutic target for OA.

Methods
Human OA cartilage and experimental OA in mice. Human OA cartilage was sourced from individuals undergoing arthroplasty [10][11][12][13] . The Institutional Review Board of Wonkwang University Hospital approved the use of these materials, and all participants provided written informed consent before the operative procedure. C57BL/6 J mice were used for the experimental OA studies. Esrrg +/mice (B6.129P2-Esrrg tm1Dgen /Mmnc) were purchased from the Mutant Mouse Regional Resource Center (MMRRC). Col2a1-Esrrg Tg mice were generated using the Col2a1 promoter and enhancer, as previously described [10][11][12]45 . All experiments were approved by the Gwangju Institute of Science and Technology Animal Care and Use Committee. Experimental OA was induced in 10-week-old male mice by DMM surgery, or by IA injection (once weekly for 3 weeks) of adenovirus (1 × 10 9 plaque-forming units [PFUs] in a total volume of 10 μl) expressing ERRγ (Ad-Esrrg), ERRα (Ad-Esrra), ZIP8 (Ad-Slc39a8), or HIF-2α (Ad-Epas1) [10][11][12][13]29 . Where indicated, male mice of various ages (8,12,16, and 20 weeks old) were used for experimental OA studies. Mice were killed at 8 weeks after DMM surgery or at 3 or 8 weeks after the first IA injection, and subjected to histological and biochemical analyses. Intraperitoneal injection of GSK5182 (20 mg kg -1 body weight) was performed every 2 days for 7 weeks, beginning 1 week after DMM surgery.
Primary culture of articular chondrocytes. Femoral condyles and tibial plateaus were obtained from mice, and chondrocytes were isolated from cartilage tissue by digestion with 0.2% collagenase [10][11][12][13]54 . The cells were maintained as a monolayer in Dulbecco's modified Eagle's medium (DMEM; Gibco) supplemented with 10% fetal bovine serum and antibiotics (penicillin G and streptomycin). On culture day 2, cells were infected with 800 MOI (multiplicity of infection) of empty adenovirus (Ad-C); infected with Ad-Esrrg, Ad-Epas1, or Ad-Slc39a8 at various MOIs; or treated with the inflammatory cytokines, IL-1β (1 ng ml -1 ) or IL-6 (100 ng ml -1 ) for the indicated time periods, unless otherwise indicated in each experiment.
Infection and IA injection of adenovirus. The adenoviruses expressing mouse ERRγ (Ad-Esrrg), Ad-shEsrrg, and ERRα (Ad-Esrra) were as previously described 22 . The adenoviruses expressing mouse HIF-2α (Ad-Epas1) and ZIP8 (Ad-Slc39a8) were purchased from Vector Biolabs 10,12 . Mouse articular chondrocytes were cultured for 2 days, infected with the indicated MOI of adenovirus for 2 h, and cultured in the presence or absence of pharmacological agents for an additional 36 h. For IA injection of adenovirus, Ad-Esrrg, Ad-shEsrrg, Ad-Esrra, or Ad-C (1 × 10 9 PFUs in a total volume of 10 μl) was injected into the knee joints of mice once per week for 3 weeks.
Skeletal staining and histological analysis. Skeletons of whole-mouse embryos were stained with alcian blue and alizarin red, as described previously [10][11][12][13]46 . Briefly, whole embryos were skinned, eviscerated, fixed with 95% ethanol for 4 days, and immersed in acetone for 3 days. Samples were stained with one volume of 0.3% alcian blue 8GX in 70% ethanol, one volume of 0.1% alizarin red S in 95% ethanol, one volume of 100% acetic acid, and 17 volumes of 100% ethanol. Bone histology was analyzed as described previously 46 . Briefly, metatarsals of 2-week-old mice were fixed in 4% paraformaldehyde, decalcified in 0.5 M EDTA (pH 7.4) for 7 days at 4°C, and embedded in paraffin. The deparaffinized sections were stained with 1% alcian blue in 0.1 N HCl for 10 minutes, and washed with 0.1 N HCl for 5 minutes. The lengths of the resting/proliferative and hypertrophic zones were measured using a microscope image-analysis program (AxioVision).
Conventional RT-PCR and qRT-PCR. Total RNA was extracted from primary cultured chondrocytes using the TRI reagent (Molecular Research Center, Inc.). For cartilage tissues, mouse knee joints were scraped with a blade after DMM surgery or sham operation, and RNA was isolated using the TRI reagent [10][11][12][13] . Total RNA was reverse transcribed, and the resulting cDNA was PCR-amplified using the PCR primers and experimental conditions summarized in Supplementary  Table 2. qRT-PCR was performed using an iCycler thermal cycler (Bio-Rad) and SYBR premixExTaq reagents (TaKaRa Bio). For each target gene, transcript levels were normalized to that of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and expressed as fold-change relative to the indicated control.
ChIP assay. ChIP assays were performed using a Pierce Agarose ChIP Kit (Thermo Scientific) 35 , as described by the manufacturer. Briefly, 70% confluent mouse articular chondrocytes were infected with Ad-Esrrg at an MOI of 800 for 36 h. The primers for ChIP assays were designed to amplify two different ERREcontaining regions of the Mmp3 promoter and three regions of the Mmp13 promoter (Supplementary Table 3). DNA and proteins were cross-linked by incubating cells with 1% formaldehyde for 10 minutes at room temperature. Excess formaldehyde was quenched by incubating with glycine for 5 minutes. Cells were lysed and nuclei were digested using micrococcal nuclease. Sheared chromatin was diluted and immunoprecipitated with 2 μg of anti-ERRγ or control IgG antibody. DNA-protein complexes were eluted from protein A/G agarose beads using a spin column, and then reverse cross-linked by incubation with NaCl at 65°C. The relative binding of ERRγ to the ERRE regions of Mmp3 and Mmp13 promoters was analyzed by PCR amplification using a Mastercycler thermal cycler. For quantitative ChIP assays, qRT-PCR was performed using an iCycler thermal cycler (Bio-Rad) and SYBR premixExTaq reagents (TaKaRa Bio). The data represent foldchanges relative to each input.
Microarray analyses. Total RNA was extracted from mouse articular chondrocytes using a Purelink RNA mini kit (Ambion). The concentration, purity, and integrity of the extracted RNA were determined by spectrophotometry. Four replicates for each cell type were isolated and processed. RNA from mouse chondrocytes was analyzed using Affymetrix Gene Chip arrays (Affymetrix Gen-eChip Mouse Gene 2.0 ST Array) in accordance with the Affymetrix protocol (Macrogen Inc., Seoul, Korea). Probe signals in the raw data were normalized by the RMA (robust multiarray average) for each separate data set (IL-1β treatment or injection of Ad-C, Ad-Epas1, or Ad-Slc39a8). Normalization was performed using R v.3.3.2 with Affy package v.1.52.0. To find DEGs (differentially expressed genes), we performed the Student's t-test followed by the Benjamini-Hochberg multiple hypothesis test for each group using python v.3.4.3 and StatsModels library v.0.8.0. The cutoff values for DEG identification were adjusted such that the P-value was less than 0.05 (FDR < 0.05) and the absolute value of log fold change was greater than 1 (|LogFC| > 1).
Statistical analysis. All statistical analyses were performed using IBM SPSS Statistics 21 software. For cell-based in vitro studies, two-tailed Student's t-tests with unequal sample sizes and variances, and two-way analysis of variance (ANOVA) with post hoc tests (LSD) were used for pairwise comparisons and multicomparisons, respectively. Data collected from mouse experiments were analyzed using the nonparametric Mann-Whitney U test. The n-value indicates the number of independent experiments or mice. Significance was accepted at the 0.05 level of probability (P < 0.05).

Data availability. Microarray data have been deposited in Gene Expression
Omnibus with the accession codes GSE104794 (for HIF-2α), GSE104795 (for ZIP8), and GSE104793 (for IL-1β). All other data supporting the findings of this study are available within the paper and its supplementary information files.