Epigenetic therapies for osteoarthritis

Abstract

Osteoarthritis (OA) is an age-associated disease characterized by chronic joint pain resulting from degradation of articular cartilage, inflammation of the synovial lining, and changes to the subchondral bone. Despite the wide prevalence, no Food and Drug Administration (FDA) approved disease modifying drugs exist. Recent evidence has demonstrated that epigenetic dysregulation of multiple molecular pathways underlies OA pathogenesis, providing a new mechanistic and therapeutic axis with the advantage of targeting multiple deregulated pathways simultaneously. In this review, we focus on the epigenetic regulators that have been implicated in OA, their individual roles, and potential crosstalk. Finally, we discuss the pharmacological molecules that can modulate their activities and discuss the potential advantages and challenges associated with epigenome-based therapeutics for OA.

Lessons from previous efforts to find drugs for OA

Osteoarthritis (OA) is the most common disease of the joint, affecting an estimated 242 million adults worldwide [1], representing the largest cause of age-related disability. Despite these staggering statistics, Food and Drug Administration (FDA) approved therapies for OA remain limited and there are no disease modifying osteoarthritis drugs (DMOADs) that could prevent or restrain development of OA. Current standard-of-care includes pain amelioration and eventual total joint replacement. This underscores the medical need for a DMOAD, which should ideally alleviate the pain associated with the disease and prevent further joint degradation [2].

Over the past decades, many therapeutic strategies have attempted targeting molecular pathways that are deregulated in OA. These include targeting catabolic enzymes responsible for degradation of articular cartilage i.e. metalloproteinase and aggrecanase inhibitors, using bisphosphonates and anti-resorptive drugs to stabilize bone homeostasis, and targeting inflammatory pathways through inducible nitric oxide synthase (iNOS) and NFKB inhibitors [2]. These drugs, however, have had limited pre-clinical and clinical success [2]. In part, developing effective DMOADs is hindered by the complex etiology of the disease, whose drivers include genetic predisposition, metabolism, trauma, inflammation, biomechanics, aging, and epigenetics (see Glossary) [3]. This, in essence, results in a heterogeneous group of patients who are all classified as having one disease based on radiological evidence. Additionally, OA is a whole joint disease [4] (Box 1) therefore, therapeutic strategies need to be able to target a wide-range of deregulated gene programs in cartilage, synovium, and bone, increasing the complexity of the challenge. Here, we focus on the enzymes that catalyze the writing, reading, and erasure of the epigenetic changes found in OA, as they represent a potential new class of candidates for DMOADs. These regulators can have a widespread effect on multiple gene programs at the same time, in essence having the potential to reset or reprogram the aberrant state of the cell, in a way that targeting single pathways cannot.

Box 1: Changes that occur in the osteoarthritic joint.

The healthy knee joint is composed of three tissues: cartilage, bone and the synovium (Box 1, Figure I). Cartilage is an avascular and acellular tissue composed of the resident cells, chondrocytes and the surrounding extracellular matrix (ECM) that they produce, enriched in collagen type II and aggrecan. This ECM deposition is zonal, with the chondrocytes in each zone (superficial, middle and deep) secreting different ECM proteins [75]. Due to their limited contact with the vasculature, chondrocytes receive the majority of their microenvironmental cues from the synovial fluid, which bathes the superficial zone and is secreted by the synovial fibroblasts of the synovium [76]. During the onset of osteoarthritis (OA), several changes occur to the joint (Box 1, Figure I). The synovium becomes inflamed and begins to secrete several pro-inflammatory cytokines including interleukin 1 beta (IL1β) and Tumor Necrosis Factor alpha (TNFA). This in turn activates inflammatory signaling in the chondrocytes, which secrete catabolic enzymes that degrade the ECM, leading to cartilage erosion. There are also changes to the underlying subchondral bone, including formation of cysts and bone spurs (Box 1, Figure I).

Box 1, Figure I: Pathogenic changes that occur to the joint in osteoarthritis.

(A) The healthy joint in homeostasis showing its different components: the cartilage, synovium and the bone, as well as the zonal organization. (B) In osteoarthritis, this balance is altered in several ways, including inflammation of the synovium (synovitis), which leads to increased secretion of IL1β and other pro-inflammatory molecules that trigger degradation of the extracellular matrix (ECM) in the chondrocytes and cartilage erosion. The underlying subchondral bone is further changed leading to formation of cysts and bone spurs.

Regulators of Cytosine Modifications

Cytosines (C) in DNA can be modified by the addition of a methyl group on the 5’ carbon (5mC). Multiple studies have mapped global DNA methylation changes associated with OA (reviewed in [5]). DNA methylation reactions are catalyzed by the DNA methyltransferase family of enzymes (DNMT1, DNMT3A and DNMT3B, Figure 1). DNMT1 is a maintenance methyltransferase, which is responsible for copying the methyl group onto Cs of a hemi-methylated DNA strand after replication. During this maintenance of methylated DNA, DNMT1 is recruited to the hemi-methylated DNA by another protein, Ubiquitin-like containing PHD and RING finger domains 1 (UHRF1) through a cooperative binding of the histone modifications and hemi-methylated DNA (Figure 1) [6]. Although a role for UHRF1 in OA has not been tested, a recent study demonstrated that loss of Uhrf1 leads to chondrodysplasia in mice, thereby implicating its role in skeletal development [7].

Figure 1: Cytosine Modifications.

Cytosine modification cycle and the enzymes responsible for each modification are shown. An unmethylated cytosine is first methylated by one of the DNMT family of enzymes (DNMT1, DNMT3A, DNMT3B), and can then be successively oxidized by the TET family (TET1/2/3) of enzymes. 5fC and 5caC can be acted upon by TDG and the base excision repair (BER) pathway. Currently available inhibitors for the DNMTs (5-Aza, Decitabine and Zebulorine) as well as for the TETs (2-HG and TiP1) are noted. Enzymes colored in red (DNMT3A and TET1) have been found of accelerate OA pathogenesis while enzymes colored in green (DNMT3B) have been found to be protective against OA pathogenesis. Abbreviations used: 5mc: 5-methylcytosine, 5hmC: 5-hydroxymethylcystoine, 5fC: 5-fluorocytosine, 5caC: 5-carboxylcytosine, 2-HG: 2-hydroxyglutarate, UHRF1: ubiquitin-like, PHD and ring finger-containing 1, DNMT: DNA methyltransferase, TDG: thymidine DNA glycosylase, Ten-eleven Translocation (TET).

DNMT3A and 3B are de novo methyltransferases, which are able to catalyze the addition of a methyl group in a new location. As such, these two enzymes are good candidates for altering the DNA methylation landscapes in OA. DNMT3A was found to be up-regulated in a subset of OA patients (n=71) compared to normal controls (n=32), and it was observed that knockdown of DNMT3A decreases the catabolic effects of interleukin 1 beta (IL1β) treatment on the extracellular matrix (ECM) [8]. The opposite is true of DNMT3B, whose expression levels were decreased in both a spontaneous mouse model of OA and in human OA samples (n=71 OA, n=11 healthy), as well as in healthy chondrocytes (cartilage cells) after treatment with IL1β in vitro [9]. The same group demonstrated that mice lacking DNMT3B showed an increased susceptibility to spontaneous OA, while increasing Dnmt3b expression protected mice from surgically induced OA. Mechanistically, this was in part due to increased mitochondrial metabolism in chondrocytes upon genetic loss of Dnmt3b [9]. DNTM3B represses 4-aminobutyrate aminotransferase (Abat), an enzyme that metabolizes a key intermediate of the tricarboxylic acid (TCA) cycle by methylating its promoter [9]. Increased Abat expression leads to an elevation in catabolic programs, including the expression of hypertrophic genes Mmp13 and Runx2 [10]. Taken together, these studies suggest that DNMT3A and 3B have different targets in chondrocytes. Development of a specific, non-overlapping inhibitor of DNMT3A, which does not affect DNMT3B, is therefore critical. In addition, the dynamics and timing of when to administer such an inhibition during disease remain to be determined in order to identify a putative therapeutic window.

5mC can be further oxidized by the ten-eleven-translocation (TET) family of enzymes (TET1, 2, 3), producing 5-hydroxymethylation (5hmC), 5-formylcytosine (5fC), and 5-carboxylcytosine (5caC) (Figure 1) [11]. 5fC and 5caC can be acted on by the enzymes in the base-excision repair (BER) pathway, specifically thymine DNA glycosylase (TDG), to restore an unmethylated cytosine [11]. This raised the important question of whether these cytosine modifications are merely intermediates in the DNA demethylation cycle or are epigenetic marks in their own right. Multiple studies have shown that 5hmC is stably accumulated, and may function as an epigenetic mark. Our previous studies on in vitro differentiated chondrocytes [12], as well as studies in other tissues, have shown that 5hmC accumulates in differentiating cells in a lineage-specific manner and is associated with lineage-specfic gene activation, including Col2a1 and Acan in chondrocytes [13]. We have also demonstrated that 5hmC accumulates in gene bodies of end-stage human OA chondrocytes (n=8 healthy, n=13 OA), marking multiple OA-associated genes that are highly upregulated in OA cartilage, including Mmp3, Mmp9, Mmp13 and Adamts5 [14,15].

Genetic loss as well as pharmacological inhibition of TET1 with 2-hydroxyglutarate (2-HG, discussed below), inhibits OA in a surgical mouse model of OA where the medial meniscus is destabilized to initiate OA pathology [16]. Pathway analysis demonstrated that TET1 is involved in several aspects of cartilage homeostasis, including mTOR and WNT signaling, and the balance of anabolic versus catabolic factors in the ECM [16]. Both genetic and pharmacological inhibition of TET1 with 2-HG modulated inflammation in end stage human OA chondrocytes, mirroring the mouse studies [16]. Another study revealed that Twist Basic Helix-Loop-Helix Transcription Factor 1 (TWIST1), a transcription factor that is elevated in OA, upregulated TET1 expression and 5hmC in chondrocytes, leading to an increase in matrix metalloproteinase-3 (MMP3) expression [17]. In contrast to these findings that TET1 is increased in OA-like conditions, another group found that treatment of OA chondrocytes with IL1β decreased 5hmC and TET1 expression [18], suggesting a potential negative feedback loop. Further studies are needed to understand this link. While TET1 appears to be a major regulator in OA, the contributions of the other paralogues (TET2 and TET3) that are also expressed in OA cartilage, remain to be elucidated.

Small molecule inhibitors for DNMTs and TETs

Azacytidine, and its derivative decitabine, are nucleoside mimics that act as DNMT1 inhibitors during DNA replication (Figure 1) and cause global hypomethylation. Both have been FDA approved for the treatment of certain blood cancers, where they have been extensively studied [19]. In OA, decitabine treatment of patient chondrocytes has shown the potential to modulate gene expression [20]. Decitabine may also be useful for correcting allelic imbalance by demethylating the C allele [22] in patients with C/T single nucleotide polymorphisms (SNPs) in the gene Growth Differentiation Factor 5 (GDF5), whose loss of expression as been associated with increase risk for OA [21]. However, no studies have systematically looked at the use of the broad DNMT inhibitors to modulate OA.

To date, TET activity in vivo has only been inhibited by the oncometabolite 2-hydroxyglutarate (2-HG), that acts as a competitive inhibitor of alpha-ketoglutarate (a-KG) [23,24]. 2-HG inhibits all three TET paralogues, in addition to other a-KG dependent enzymes including histone demethylases [23,24]. We have recently demonstrated that intra-articular injections of 2-HG ameliorate OA pathology [16]. Efforts are underway to optimize cytosine based molecules that can selectively target the TET enzymes over the DNMTs [25] and TET1 over the other paralogues [26].

Regulators of Histone Modifications

Similar to DNA, histones can also be chemically modified by a series of writer and eraser proteins. Below we discuss the different histone modifying enzymes implicated in OA.

DOT1L - H3K79 methyltransferase

Disruptor of telomeric silencing 1-like (DOT1L) is a H3K79 histone methyltransferase, which is responsible for mono, di or trimethylation of the H3K79, using s-adenosylmethionine (SAM) as a cofactor [27] (Figure 2A). The H3K79 modifications associated with gene activation are considered among the most stable modifications, with the mono form having a half-life of 1.1 days, and the di- form having a half-life of 3.6 days [28]. In comparison, histone acetylation or phosphorylation marks have a half-life in the scale of seconds to minutes [28]. Many genome-wide association studies (GWAS) have identified single nucleotide polymorphisms (SNPs) in DOT1L that are related to cartilage thickness and predisposition to OA [29]. During skeletal development, DOT1L has been shown to regulate WNT signaling, by preventing the hyperactivation of target genes of another epigenetic regulator, sirtuin-1 (SIRT1) [30]. DOT1L deficient mice are more susceptible to both post-traumatic and spontaneous OA development, suggesting a protective function for this H3K79 methyltransferase during OA pathogenesis [31]. Postnatal loss of DOT1L causes disruption in cartilage homeostasis [30] along with an increase in ectopic bone formation and increased expression of Col10a1 [32]. These studies provide a putative mechanistic basis for the detrimental role of DOT1L loss during OA development. This suggests that for patients with decreased DOT1L activity or a disease-associated SNP in DOT1L, a drug treatment to increase DOT1L expression or activity may be beneficial. However, no known pharmacological agents that can activate DOT1L activity or expression currently exist.

Figure 2: Histone Modifications.

(A) The regulators of histone methylation implicated in OA along with currently available inhibitors or activators are shown. (B) The regulators of histone acetylation implicated in OA along with currently available inhibitors or activators are shown. Enzymes shown in green (SIRT1) have been suggested to be protective factors against OA pathogenesis. Abbreviations used: H3K79: histone 3, lysine 79, me: methylation, DOT1L: Disruptor of telomeric silencing 1-like, PRC2: Polycomb repressive complex 2, EZH2: enhancer of zeste homolog 2, JMJD3: jumonji domain-containing protein D3, UTX: ubiquitously transcribed tetratricopeptide repeat X, HDACs: histone deacetylases, SIRT1: sirtuin 1, HAT: histone acetyltransferase.

Polycomb group proteins

The polycomb group proteins constitute a family of chromatin associated repressors with polycomb repressive complexes (PRC) 1 and 2 having been the most widely studied (the molecular composition of each is reviewed in [33]). Compared to PRC2, little work has been done on PRC1 in OA. Overexpression of one of its component proteins, chromobox 4(CBX4), via lentiviral injection into the joint prevented cartilage degradation in a surgical model of OA by decreasing cellular senescence [34]. PRC2 is involved in the silencing of gene expression via mono, di- and tri- methylation of H3K27 (Figure 2A). PRC2 is made up of several protein subunits that vary depending on the specific cell and tissue type as well as cellular context [35]. The methyltransferase activity in the PRC2 core complex is attributed to enhancer of zeste homologue 2 (EZH2), with the other non-enzymatic subunits being embryonic ectoderm development (EED) and suppressor of Zeste 12 homolog (SUZ12) [33,35]. Conditional knockout of Eed in chondrocytes showed a growth defect with decreased chondrocyte proliferation and accelerated hypertrophy [36]. Mechanistically, Eed deficiency led to overactive WNT signaling [36]. A separate group that performed Ezh2 loss-of-function (LOF) in mesenchymal stem cells (MSCs), using a mesenchyme specific knockout model, found similar results including mice with skeletal abnormalities and shortened forelimbs [37]. Loss of EZH2 in early mesenchymal cells or immature chondrocytes impaired their ability to up-regulate Col10a1 expression robustly, thereby affecting timely hypertrophy. In a follow up study, the same group further revealed that an EZH2 loss after the skeletal cells had become lineage committed did not lead to any skeletal defects [38], demonstrating a stage-specific role for EZH2 during chondrocyte development.

The role of EZH2 in OA, however, is unclear. One study has reported that EZH2 expression is increased in OA cartilage (n=8) and that it drives the expression of hypertrophic genes (MMP13, type X collagen) upon treatment with IL1β in cultured cells [39]. The idea that EZH2 is involved in hypertrophic gene regulation is consistent with its role during skeletal development. Mechanistically, EZH2 was found to mediate silencing of the WNT inhibitor, secreted frizzled-related protein 1 (SFRP-1), in OA chondrocytes [39]. Intra-articular injections of a pharmacological inhibitor of EZH2, EPZ005687, inhibited OA development in mice where OA was induced by an anterior cruciate ligament transection (ACLT) [39]. In stark contrast, another group demonstrated that EZH2 is increased during a medial meniscectomy (MMx) induced OA (surgical model) in mice and is beneficial for the wound healing response in chondrocytes [40]. A cartilage specific loss of EZH2 aggravated OA progression in the MMx mice model, thereby this study concluded that EZH2 is protective during OA pathogenesis (Figure 2A) [40]. The different results could be attributed to the two different ways that EZH2 was inhibited, namely the intra-articular pharmacological treatment in the first study [39], which would inhibit EZH2 in the whole joint affecting the cartilage, synovium, fat pad and potentially the subchondral bone, in contrast to genetic loss of EZH2 in cartilage alone in the second study [40]. Another interesting observation in this latter study was that EZH2 expression was found to be variable upon expanding the cohort of patients, with only 26 out of 66 OA cartilage samples showing an increased EZH2 expression in RNA-seq analyses [40]. This study therefore highlights the potential of epigenetic regulators like EZH2 inhibitors as DMOADs, but also cautions that patient heterogeneity, and the specific joint tissues targeted, needs to be taken into consideration for these drug candidates. Moreover, further research is needed on identification of other PRC2 components that might play a role in OA. Besides EPZ005687, multiple other EZH2 inhibitors have been developed and tested for cancer treatments (reviewed in [41]) that can be further evaluated for their effects in OA.

KDM6 demethylases: JMJD3 and UTX

While EZH2 is the writer that adds methyl marks, the erasers that remove di- and tri-methyl groups from H3K27 are Jumonji domain containing-3 (JMJD3), also known as KDM6B demethylase and the ubiquitously transcribed X-chromosome tetratricopeptide repeat protein (UTX) or KDM6A demethylase (Figure 2A). During skeletal development, JMJD3 is highly expressed during endochondral ossification and associates with RUNX2 to facilitate activation of its target genes [42].

A study in OA chondrocytes reported an increased expression of JMJD3 but not UTX, and an inhibition with GSK-J4, an inhibitor of both JMJD3 and UTX, prevented the expression of MMP13 and Col10a1 [43] (Figure 2A). This pharmacological effect was recapitulated by short interfering RNA (siRNA)-mediated knockdown of JMJD3 but not UTX [43]. It is important to note here that although GSK-J4 was originally reported to be specific to the KDM6 family of demethylases [44], it was independently reported to have weaker activity against KDM5B and KDM5C as well [45]. Moreover, treatment of normal human articular chondrocytes with IL1β caused an upregulation of JMJD3 but not UTX, providing a putative mechanism for JMJD3 induction during OA [43]. The same study also reported that treatment of MSCs with GSK-J4 inhibited differentiation [43]. Mapping global targets of the JMJD3 and UTX demethylases in OA cartilage and synovium can provide us better insights into the putative candidature of GSK-J4 and related molecules for DMOADs.

Histone deacetylases (HDACs)

Histone deacetylases (HDACs) remove acetyl groups from histones, allowing DNA to compact more tightly with the histones, thereby decreasing accessibility and gene expression [46] (Figure 2B). They consist of two families: (a) classic HDACs (class I, II and IV) that have a zinc-dependent active site and include HDACs 1 to 11, and (b) class III HDAC enzymes, which require NAD+ as a co-factor and are known as sirtuins (SIRT1 to 7) [46].

Classic HDACs

Among the classic HDACS, HDAC3 has been characterized during skeletal development and regeneration. Loss of HDAC3 hinders commitment of mesenchymal progenitors [47], and changes lipid storage and glucocorticoid metabolism in osteochondroprogenitors (cells which can give rise to either cartilage or bone, depending on the cues) [48]. Its loss in chondrocytes increases the expression of ECM catabolic genes and decreases the expression of ECM anabolic genes, thereby altering endochondral ossification [49].

Multiple HDACs have been associated with OA pathology. A focused screen for the expression of HDACs 1 to 7 in healthy (n= 6) and OA patients (n= 10) found HDAC7 to be significantly upregulated in OA cartilage tissue, with increased expression specifically in the middle and deep zones of cartilage (Figure 2B) [50]. Knockdown of HDAC7 decreased MMP13 expression, although treatment with a broad HDAC inhibitor, Trichostatin A (TSA) failed to do the same [50]. Since HDACs normally function to repress gene expression, it is surprising that HDAC7 inhibition led to decreased expression of MMP13, suggesting that this is likely an indirect effect of HDAC7 silencing a negative regulator of MMP13. Another study reported an age dependent decrease in the protein levels of HDAC4 (n=6 in normal, aged and OA cartilage). Overexpression of HDAC4 decreased expression of RUNX2 and MMP13 and could partially block the effects of IL1β treatment on OA chondrocytes [51]. This report was countered by another group who reported the opposite trend: a loss of HDAC4 in OA cartilage (n=12). Their experiments also showed that HDAC4 downregulation in SW1353 chondrosarcoma cells prevents the effects of IL1β [52]. Since small patient cohorts were used for these initial studies, the contradicting trends regarding the expression levels could be explained by presence of patient heterogeneity.

Many HDAC inhibitors (HDACi) have been developed ranging in their specificity for one particular HDAC to a set of HDACs. ACY-1215 (HDAC6i) [53] and RGFP966 (HDAC3i) [54] have been tested in cell culture and mouse models of OA and have reported beneficial effects. However, the majority of OA studies use two pan-HDACi, TSA, mentioned above, and Vorinostat [55] (Figure 2B). A recent review has even focused on the potential use of HDACi for OA [56].

Sirtuin1

Sirtuin 1(SIRT1) is a class III HDAC, with a broad protein deacetylase activity that extends beyond histones as well. Some of its prominent non-histone targets include the FoxO family, HEY1 and HEY2, PPARa, and NF-kB [57]. Additionally, in chondrocytes, SIRT1 overexpression after IL1β treatment has been shown to repress lymphoid enhancer-binding factor 1 (LEF1), attenuating MMP13 expression [58]. As a transcriptional regulator, SIRT1 has been found to regulate multiple pathways related to glucose and metabolic homeostasis and functions across a broad range of tissues [59]. SIRT1, in both its epigenetic and non-epigenetic functions, plays a role in ECM homeostasis during skeletal development. Mice carrying the SIRT1^tm2.1Mcby (SIRT1^y/y) allele, which ablates SIRT1 enzymatic activity, were smaller than their wildtype counterparts, at postnatal day 2, and were observed to have less ECM deposition in the adult skeletons [60]. This was further confirmed by another study where EX527, a SIRT1 inhibitor, was used in the growth plate [61]. Mice lacking SIRT1 have premature, OA-like cartilage degradation, even at an early age [60] and this is exacerbated by injury [62]. Increased expression of MMP13, type X collagen and apoptotic markers were observed in the SIRT^−/− mice, in addition to an increased acetylation of the NFkb factor p65, explaining the acceleration of OA [62].

Several studies have assessed the expression levels of SIRT1 in human OA cartilage compared to normal controls. A study with 38 patient samples saw a decrease in SIRT1 correlated with increased disease severity [63], in corroboration with previous smaller studies by other groups [64,65]. Loss of SIRT1 in human articular chondrocytes also induced OA-like gene expression profiles [64]. Collectively, this and the developmental studies suggest that SIRT1 is protective against OA. Another study assessed the activity of SIRT1 in peripheral blood mononuclear cells of OA and normal controls but found no differences, demonstrating that the changes are cartilage-specific and likely not reflected in other tissues [66]. Intriguingly though, treatment of these cells with a SIRT1 activator increased IL-6 secretion levels only for OA patients [66] suggesting that epigenetic differences between normal and OA patients likely exist and could potentially be detected in blood.

The observations that the levels of SIRT1 decrease in damaged cartilage, as well as the acceleration of OA in KO mice suggest that SIRT1 activators may be putative therapeutics for OA [67]. One such SIRT1 activator is resveratrol. Treatment of OA chondrocytes with resveratrol has been validated to increase SIRT1 expression by 2.6 fold, and decrease apoptosis [68]. Resveratrol treatment has also been shown to ameliorate pathology in a rat model of surgically induced OA [69]. Recently, a more potent activator of SIRT1, SRT1720, was developed, which increases SIRT1 activity by 8.7-fold (Figure 2B) [70]. Injection of SRT1720 after surgical induction of OA in mice decreased synovial inflammation and progression of OA [71]. However, it is important to keep in mind that all of these studies suffer from the caveat that they do not validate that the effects of SIRT1 activation are due to histone deacetylation in the promoters of its target genes. It is therefore difficult to rule out that the SIRT1 effects are non-epigenetic i.e. through affecting deacetylation of other key proteins.

Concluding Remarks and Future Perspectives

The epigenome presents a powerful new way to understand and target deregulated pathways in OA. As we have outlined in the review, there is a disease-associated redistribution of epigentic regulators in OA, which leads to aberrant gene expression (Figure 3). Utilizing epigenetic regulators (for example DNMT3B, TET1 or SIRT1) as pharmacological candidates for DMOADs is therefore attractive, as multiple sets of deregulated genes could potentially be corrected at once. However, a major challenge with modulating the epigenetic regulators directly is ensuring specificity and preventing off-target effects (Figure 3C, OA therapy 1). Thus far, drugs targeting epigenetic regulators have primarily been used as cancer therapeutics (reviewed in [72]), and although they have seen wide efficacy, the bottleneck has been that these regulators also affect unintended genes. For example, while deactivating catabolic genes like the metalloproteinase that are aberrantly expressed in OA is a desirable effect, deactivating anabolic genes like ECM genes would be undesirable (Figure 3C, OA therapy 1). This problem arises because many of the epigenetic regulators have no inbuilt targeting mechanism toward any given set of genes. Therefore, the target genes are mainly being driven by the available DNA binding cofactors and hence are largely context dependent (see Box 2 for examples, and Figure 3A versus 3B). Using modulators against these precise ‘binding cofactors’ in the diseased cells could provide the desired specificity and fine tuning to an epigenetic drug (Figure 3C, OA therapy 2). Future studies should therefore focus on better understanding and identifying these precise targeting co-factors for the epigenetic enzymes like DNMTs, TETs and HDACs in OA.

Figure 3: Osteoarthritis redistributes epigenetic regulators.

(A) A schematic of a scenario in healthy cartilage, in which an epigenetic regulator (regulator 1) is targeted by a DNA binding factor, i.e a transcription factor or lncRNA (see Box 2), and deposits chemical modifications associated with gene activation on genes related to cartilage homeostasis (blue circles on gene 1). Repressive marks (shown as grey circles) keep catabolic genes (shown as genes 2 and 3) repressed. (B) A schematic of a scenario in OA cartilage. The diseased environment causes new DNA targeting molecules to be expressed (orange hexagon), which redistributes the way that regulator 1 is targeted. Compared to (a), regulator 1 now also activates catabolic genes 2 and 3, which should not be expressed under homeostasis. (C) Schematics address how epigenetic therapies might target the altered OA epigenome. Three possible therapeutic axes exist: (i) OA therapy 1- decreasing regulator 1 activity with an inhibitor (shown as red star). This inhibitor has both intended and unintended consequences. It will decrease the pathogenic overexpression of gene 2 and gene 3, but will also decrease the expression of the anabolic gene 1. We refer to this as an “off-target” effect. (ii) OA-therapy 2- In the second approach, the OA-specific targeting factor (orange hexagon) by inhibiting by a specific drug (shown as red circle). This avoids the problem of turning off gene 1, while decreasing expression of gene 2 and 3. (iii) OA-therapy 3- Finally, using genome-engineering tools like CRISPR/Cas9, epigenetic regulators can be targeted to a particular site. For example, here a regulator placing a silencing mark (regulator 2) is targeted to pathogenically expressed gene 3. However, while this avoids silencing gene 1, it does not silence gene 2, which is also pathogenic. To date, only strategy (i), OA therapy 1, has been attempted in pre-clinical OA models.

Box 2: Guiding the Epigenetic Regulators.

Most epigenetic regulators do not have intrinsic DNA target sequences, although many can independently bind chromatin. Rather, the set of genes targeted by an epigenetic regulator in a given cell type is determined by the interactions of that regulator with targeting co-factors (Figure 3A). Epigenetic regulators can be targeted via interactions with transcription factors (TFs) or long-noncoding RNAs (lncRNAs). For example, with respect to DNMTs or TETs, TFs with CpG dinucleotides in their binding motifs recruit the enzymes to modifiable cytosines, facilitating methylation, hydroxymethylation or demethylation. Several TFs with known roles in OA [77], bind to motifs with cytosine-guanine dinucleotides (CpGs) CGs.

For example, Early growth response protein 1 (ERG1), a TF implicated in OA and inflammatory responses, is induced by IL1β and NFKB signaling in chondrocytes [78–80]. During post-natal brain development, ERG1 is known to recruit TET1 to target genes for demethylation and gene activation [81]. Although an ERG1-TET1 interaction has not been validated in OA chondrocytes, the expression of ERG1 is upregulated upon chondrogenesis [82]. ERG1 expression in OA is contested, as it was found to be upregulated (n=9 OA, n=9 normal) in one study [78], while a smaller cohort study found it to be downregulated (n=3 OA, n=2 normal) [83]. Such variability likely reflects patient heterogeneity, and thus only a subset of patients may have ERG1 targeted TET1 gene activation. Another potential TET1 interactor in OA chondrocytes is Hypoxia Inducible Factor 1 Subunit Alpha (HIF1A). TET1 and HIF1A have been shown to interact in chronic stress conditions in the brain [84], leading to a higher rate of 5hmC deposition at HIF1A target genes. In the context of chondrocytes, HIF1A appears to be elevated in OA, mediating apoptosis and autophagy [85].

A similar paradigm may be true for lncRNAs. One of the first examples of interaction of an epigenetic regulator and lncRNA was that between the lncRNA, HOTAIR and PRC2 [86]. HOTAIR’s aberrant expression in epithelial cancer cells altered PRC2 occupancy in cancer cells [87]. Although direct interactions between lncRNAs and epigenetic regulators have not yet been characterized in OA, several studies have demonstrated a plethora of lncRNAs that are altered in OA, thus it is plausible that their aberrant expression would cause a redistribution of the epigenetic regulators.

The recent advances in gene editing have created another possibility for epigenome engineering, where synthethic modules consisting of epigenetic regulators fused to a genome targeting system, such as the CRISPR-CAS9, can be generated. These synthethic modules can then provide precise gene or locus targeting for an epigenetic regulator (Figure 3C, OA therapy 3). Such a molecular engineering strategy has been utilized for DNMT3A in both direct [73] and modular forms [74] and can potentially be applied to OA.

Another pertinent challenge with epigenetic therapies is drug delivery. Given the broad function of the epigenetic regulators, a systemic delivery of the therapeutic can result in toxicity and unintended effects in tissues besides the target tissue [65]. The systemic toxicity can be avoided in OA therapeutics by intra-articular delivery, as is already established for corticosteroids, hyaluronic acid and platelet-rich plasma injections [2]. However, it is important to understand if drug effects would be benefical to the cartilage, synovium and subchondral bone. Such intra-articular injections may also benefit from various biomaterials-based delivery platforms for enhanced stability and release kinetics of the drugs.

Although multiple epigenetic regulators are now identified as drivers or dampeners of OA pathogenesis, several important outstanding questions remain (see Outstanding Questions). A critical question is that of timing, namely understanding how early in disease progression are these epigenetic alterations occur. For example, one hypothesis is that changes to the epigenetic landscape occur early in OA, before the onset of symptoms. This would directly effect the timing of the therapy, and future studies would need to test if the epigenome can be rewired at end-stage of the disease.

OUTSTANDING QUESTIONS.

• How early in OA progression are epigenetic changes detectable and what is the therapeutic window?

• Could broad DNMT/TET/HDAC inhibitors be used for treating OA?

• What are the distinct targets of DNMT3A, DNMT3B, TET 1 and HDACs in osteoarthritic cartilage and other joint tissues?

• How are epigenetic regulators targeted in OA?

• Could the targeting molecules (TFs or lncRNAs) be used as therapeutic targets?

• Can OA patients be subtyped by their epigenetic profiles or by deregulated epigenetic regulators?

• Are readily accessible tissues, such as blood, amenable to epigenetic profiling and subtyping of OA patients?

Another problem that is common to OA drug development is the heterogeneity among patients, which may be especially accentuated for epigenetic changes., as both the activity of the regulator or the targeting factors may be different between patients. However, epigenetic profiling of OA patients could also provide a unique opportunity to subtype patients by their epigenetic profiles, thereby gaining further insights into the disease. Future studies should test if readily accessible tissues, such as blood, might be amenable to epigenetic profiling and subtyping of OA patients. In summary, epigenome-based profiling and drug therapies hold a lot of untapped promise for OA therapeutics.

HIGHLIGHTS.

• Despite its prevalence, osteoarthritis (OA) has no clinically approved disease-modifying drug. Numerous drug development efforts focused on single molecules or pathways have failed, signifying the need for multiple gene/protein pathway correction.

• Several epigenetic regulators that affect large gene networks have been demonstrated to play a role in OA pathogenesis.

• Targeting these epigenetic regulators as disease modifying OA drugs (DMOADs) holds the potential to reset the aberrant epigenetic landscape found in OA tissues and rewire gene networks.

• Identifying targeting co-factors can enhance specificity of these new epigenetic drugs.

GLOSSARY

Base Excision Repair (BER) pathway

BER is the primary DNA repair pathway in mammals. It is responsible for removing small base lesions, often derived from oxidation, alkylation or other events. The process is started by a glycosylate that recognizes and removes the damaged base. In the context of DNA demethylation, is pathway is utilized first by the targeted oxidation of the base by the TET enzymes, and then later this base is acted on by Thymine-DNA glycosylase (TDG)

Endochondral ossification

Endochondral ossification is the developmental process by which long bones are formed. Cartilaginous tissue, formed by the condensation of mesenchymal stem cells, first lays down template for the developing bone. The chondrocytes go through a variety of stages of maturation, including proliferation and maturation into hypertrophic chondrocytes. At this final stage, the cells undergo apoptosis, leaving room for the invasion of the template by osteoblasts to form the final calcified bone

Epigenetics

Epigenetics is broadly defined as the changes “on top of” (epi) the genome that influence the transcription of genes. While traditionally this has implied chemical changes to DNA or histones, this definition has been broadened to include chromatin folding and organization as well as different coding and non-coding RNAs that can interact with DNA and influence gene expression

Gene Body

The gene body is defined as the entire gene from the transcriptional start site (TSS) to the transcriptional end site. This includes both the exons and introns contained within the gene

Histones

Histones are the proteins around which DNA is wrapped to form nucleosomes. Histones can be chemically modified in a variety of ways including methylation, acetylation and phosphorylation. Depending on the particular histone modification, its positioning and combination with other marks, the transcriptional machinery can be either recruited or excluded at these chromatin sites

Histone methyltransferases

Histone methyltransferases are responsible for the addition of a methyl group to lysine residues in histones. Depending on the context, this mark can either be activating (H3K79me/me2/me3) or repressive (H3K27me2/me3). Conversely, histone demethylases remove these methylation marks from the target histones

Hypertrophy

Within the contexts of endochondral ossification, this refers to the process by which columnar chondrocytes mature and begin to produce different types of extracellular matrix proteins such as type X collagen. In addition, they begin to undergo apoptosis to make way for the new bone. These changes are controlled, in part, by the transcription factor RUNX2 and by WNT signaling. While a normal part of skeletal development, chondrocyte hypertrophy can also occur in OA, in which chondrocytes, which should normally make type II collagen, switch their fate. These changes are associated with pathology and change the mechanical properties of articular cartilage

Mouse model of OA

Several types of mouse OA models exist. In genetic models, a mutation in the mouse genome increases the rate of spontaneous OA, modeling human predisposition to the disease. Other models use surgical intervention to destabilize the joint, including destabilization of the medial meniscus (DMM), tearing of the anterior cruciate ligament (ACLT), or medial meniscectomy (MMx). These models generally represent post-traumatic OA