Role of Epigenomics in Bone and Cartilage Disease
ABSTRACT
Phenotypic variation in skeletal traits and diseases is the product of genetic and environmental factors. Epigenetic mechanisms include information-containing factors, other than DNA sequence, that cause stable changes in gene expression and are maintained during cell divisions. They represent a link between environmental influences, genome features, and the resulting phenotype. The main epigenetic factors are DNA methylation, posttranslational changes of histones, and higher-order chromatin structure. Sometimes non-coding RNAs, such as microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), are also included in the broad term of epigenetic factors. There is rapidly expanding experimental evidence for a role of epigenetic factors in the differentiation of bone cells and the pathogenesis of skeletal disorders, such as osteoporosis and osteoarthritis. However, different from genetic factors, epigenetic signatures are cell- and tissue-specific and can change with time. Thus, elucidating their role has particular difficulties, especially in human studies. Nevertheless, epigenomewide association studies are beginning to disclose some disease-specific patterns that help to understand skeletal cell biology and may lead to development of new epigenetic-based biomarkers, as well as new drug targets useful for treating diffuse and localized disorders. Here we provide an overview and update of recent advances on the role of epigenomics in bone and cartilage diseases. © 2019 American Society for Bone and Mineral Research.
Epigenomics as a Link Between Environment, Genotype, and Phenotype
Phenotypic variation in skeletal traits and diseases is the product of genetic and environmental factors. The extent to which genetics shapes the phenotype is different across the different skeletal conditions. Nevertheless, the genetic component of skeletal traits is large, varying from 30% (knee osteoarthritis [OA]) to 80% (bone mineral density [BMD]). This does not mean there is no effect of the environment. In fact, these DNA-sequence variants form the template upon which environmental factors can influence the phenotype, by a number of mechanisms, including epigenetic marks. Waddington coined the term epigenetics to describe the interactions between the environment and the genes leading to the development of phenotype.1 The modern definition of epigenetic mechanisms includes information-containing factors, other than DNA sequence, that cause stable changes in gene expression and are maintained during cell divisions.2
The main epigenetic factors are DNA methylation, posttranslational changes of histones, and higher-order chromatin structure. Sometimes non-coding RNAs, such as microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), are also included in the broad term of epigenetic factors. However, the exact definition of epigenetics and its components is still a matter of controversy.3 Together these different epigenetic mechanisms are key factors behind the regulation, function, and cell fate of all tissues and cells. Not surprisingly, there is a large body of evidence supporting the role of epigenetics in skeletal development, the maintenance of bone mass, and skeletal disorders. Our purpose here is to provide an overview and update of recent advances on the role of epigenomics in bone and cartilage diseases.
Epigenomic marks
DNA methylation
DNA methylation refers to the covalent addition of a methyl group to cytosines in DNA, particularly when they are part of CpG dinucleotides. In somatic cells, more than 80% CpGs are methylated, especially in repetitive sequences in intergenic regions and introns, whereas CpGs in gene promoters may be methylated or not. In general, the methylation of CpGs in gene promoters is associated with repression of gene expression, whereas the methylation of gene bodies and other regulatory regions (such as enhancers)4 has a less predictable effect. Up to 20% of the variation in DNA methylation is influenced by genetic variation, but the majority of the variation in methylation is caused by other factors, including environmental and stochastic variation. Indeed, variation in DNA methylation increases with age5 and is thought to have a role in the relation between environmental risk factors and disease risk. As all epigenetic marks, DNA methylation is dynamic; methyl groups can be added and removed from the DNA by specialized proteins, DNA-methyl-transferases (DNMTs) and ten-eleven translocation (TET) proteins, respectively (see Box 1 for an explanation of epigenomic terms).
- Chromatin: The complex of DNA and its packaging molecules. The core of the chromatin is the nucleosome, which consists of an octamer of 4 histones around which 147 bp of DNA is wrapped around.
- Chromosome conformation capture and Hi-C: Techniques used to map the spatial (3D) organization of the chromatin in the nucleus. Chromosome conformation capture (3C) quantifies the number of interactions between a given loci and the rest of the genome. In Hi-C, all genomic interaction between all genomic regions are quantified.
- CTCF: CCCTC-binding factor, highly conserved zinc finger protein involved in diverse genomic regulatory functions, including transcriptional activation/repression, insulation, imprinting, and X-chromosome inactivation, through mediating the formation of chromatin loops.
- DNA methyl-transferases (DNMTs): Family of enzymes responsible for the methylation of DNA. DNMT1 recognizes hemimethylated CpG sites on newly replicated DNA and thus it maintains the methylation pattern through cell divisions. On the other hand, DNMT3A/3B are the novo methylases, capable of converting unmethylated CpGs into methylated CpGs in double-strand DNA, which is particularly important during embryogenesis and cell differentiation.
- Epigenetics: Mechanisms causing changes in gene expression that are heritable through cell divisions and do not include modifications of DNA sequence.
- Epigenome-wide association study (EWAS): Studies of the relationship between many epigenetic marks distributed throughout the genome and phenotypic characteristics. So far, most studies aimed to analyze DNA methylation.
- Epigenomics: Usually refers to the epigenetic changes in many genes or even through the whole genome.
- Genome-wide association study (GWAS): Studies of the relationship between many genetic variants (usually hundred thousands or millions) distributed throughout the genome and phenotypic characteristics.
- Histone code: Hundreds of different posttranslational modifications (PTMs) and their combinatorial patterns form a code, the histone code. This code can give rise to a prescribed transcriptional or other genomic regulatory response, interpreted by specialized proteins that can read, write, and erase histone PTMs.
- Histone deacetylases (HDACs): Family of enzymes removing acetylation marks from histone tails. These epigenetic “erasers” are very important for modulating gene expression because histone acetylation is usually associated with active chromatin.
- Long non-coding RNAs (lncRNAs): ncRNAs with more than 200 nucleotides that regulate gene expression and interact with other epigenetic mechanisms.
- Methylation quantitative trait loci (meQTL): A DNA locus (usually a single-nucleotide polymorphism [SNP]) that is associated with DNA-methylation levels from a certain CpG.
- MicroRNAs (miRNAs): Small ncRNAs, 18 to 25 nucleotides long. One known function (of the many that are known) of miRNA's target mRNAs, which interferes with protein translation.
- Non-coding RNAs (ncRNAs): RNAs that do not code proteins but have regulatory roles on chromatin structure, gene expression, or translation. There are multiple types, with different sizes and functions.
- Quantitative trait loci (QTL): A DNA locus that is associated with a particular quantitative phenotypic trait.
- Ten–eleven translocation (TET): Family of proteins involved in the demethylation of CpGs.
- Topologically associated domain (TAD): Chromatin conformation capture techniques, such as Hi-C, showed that the genome was divided in compartments that interact more frequently with themselves than the rest of the genome. Regulatory regions, such as enhancers, usually contact genes located within the same TAD as the regulatory region but not outside of their TAD.
Chromatin structure
The spatial organization of the DNA itself in the cell nucleus, the chromatin structure, is also important for the functional read-out of the genome. Histones are critical components of the chromatin (Fig. 1). In fact, DNA-bound histones play major roles in the regulation of gene transcription. Posttranslational modifications (PTM) of specific amino acids in the N-terminal tail of histones, such as methylation, phosphorylation, acetylation, and ubiquitylation, remodel the shape of the chromatin. This, in turn, alters the DNA's accessibility for proteins involved in the transcription machinery, thereby regulating gene expression.6 Histone PTMs are dynamic and a number of enzymes are able to add or remove histone marks. For example, acetyl groups can be added by histone acetylases (HATs) and removed by histone deacetylases (HDACs).
Next to histone PTMs, also the spatial organization of the chromatin itself in the nucleus can modulate gene expression. Chromosome-conformation capture techniques have shown that the genome is divided into so-called topological associated domains (TADs), which are large (megabase scale) compartments of the genome. These regions interact more frequently with themselves than the rest of the genome and enhancers usually contact genes located within these TADs but not outside.7 Distant enhancers and their target gene promoters are brought into contact with each other using the formation of so-called “DNA-loops,” mediated by, for example, CTCF and cohesins (Fig. 1).
Non-coding RNAs
Besides chromatin-related marks and structure, non-coding RNAs (ncRNAs) are also frequently included among the mechanisms of epigenetic control.8 They are classified as small RNAs (<200 nucleotides) and long RNAs (>200 nucleotides). The best-known subset of small RNAs are microRNAs (miRNAs, 18 to 25 nucleotides), which inhibit protein synthesis by binding to the 3′-untranslated region of target mRNAs. Long non-coding RNAs (lncRNAs) modulate the activity of both nearby genes and distant genes by a variety of mechanisms. For instance, they often serve as scaffolds for transcription factors and other molecules involved in initiation of transcription, including repressive chromatin modifiers such as polycomb repressive complex proteins (PRC1 and PRC2) or activating chromatin modifiers.9 Some lncRNAS are mainly located in the cytosol, where they target mRNAs and downregulate protein translation. Interestingly, they may also act as decoys for miRNAs, thus preventing the inhibitory effect of the binding of miRNAs to their target mRNAs.10
Epigenomic interplay
It is worth emphasizing that epigenetic mechanisms often act in concert by interacting with each other. For example, MeCP2, a protein recognizing methylated CpGs, promotes the activity of HDACs. On the other hand, some histone marks modulate the binding of DNMTs and subsequently DNA methylation. The methylation of promoters regulates the transcriptional activity not only of protein-coding genes but also of miRNAs and other non-coding RNAs. In turn, miRNAs contribute to modulating the synthesis of DNMTs and histone-modifying enzymes. lncRNAs also influence the activity of genes encoding chromatin-modifying enzymes and miRNAs.10 Although the sequence of molecular steps is still unclear, there is evidence for the notion that DNA, RNA, and histone proteins, along with their modifications, act in a concerted fashion to bring about chromatin states that are important for dictating genomic functions8 (Fig. 1).
Epigenetic programming during development
From the moment of conception to adulthood, the environment shapes the phenotypic output. It is thought that there are certain “high sensitive” windows especially during development that have major influence on the epigenome.11 The developmental origins of health and disease concept suggests that poor developmental experience can increase the risk of non-communicable diseases in later life, including cardiovascular, metabolic, neurological, and skeletal disorders.12 A variety of mechanisms, including DNA methylation and other long-lasting epigenetic marks, may mediate the influence of the environment on the developing organism.13, 14 A few studies have explored the role of developmental factors in skeletal disorders. In a systematic review of the literature, a positive association between birth weight and bone mass was clear among children, unclear among adolescents, and weak among adults. The effect was stronger on bone mineral content (BMC) than on BMD regardless of age.15 This suggests that intrauterine growth is more closely related to bone size than to bone density and that the effect tends to be mitigated by postnatal influences. It seems that early life exposures are important for determining peak bone mass, which may be a reflection of the combined influence of intrauterine and early postnatal environmental exposures.
Maternal nutrition and specifically the maternal vitamin D status may be a critical factor for an adequate intrauterine growth rate,16 but studies have shown conflicting results.17 Rather surprisingly, in the Rotterdam cohort, severe maternal 25(OH)D deficiency (<25 nmol/L) during mid-pregnancy was associated with higher offspring BMC and bone area at 6 years of age, while no associations were found between maternal vitamin D status and offspring BMD.18 In experimental animals, vitamin D status has a transgenerational effect on the methylation of multiple genes.19 However, human studies about the relationship between maternal vitamin D levels and DNA methylation in offspring have given controversial results.16 Therefore, the actual relevance of maternal vitamin D on DNA methylation and the bone mass of the offspring is still unclear.
Whether related to parental influences or not, a few studies reported an association of the methylation of some genes (such as eNOS, RXRA, and CDKN2A) in cord blood and childhood bone mass.20-22 However, those results have not been confirmed in other cohorts yet. Although less studied than the relationship between early life experiences and osteoporosis, some data support a developmental component in OA. For example, exposure to Chinese famine during childhood has been associated with arthritis (including both OA and inflammatory arthritis) in later life.23 Similarly, in a British study, lower weight at birth and year 1 was associated with higher rates of OA.24 Weight and body length differences, which have a clear developmental component, may explain, at least in part, those associations. Another example is finger length pattern, which is thought to be an indicator of prenatal androgen exposure. Type 3 finger length pattern (longer fourth digit than second digit) has been associated with having symptomatic knee OA and chronic pain,25 which might be explained by an influence of embryogenic sex hormone exposure on brain development.26
It is thought that epigenetic programming plays a role in all of these associations, but the exact role of epigenetic factors in the relation between prenatal exposures and skeletal diseases has not been elucidated yet.
Epigenomic plasticity in adult life
In addition to the developmental epigenetic programming, there is an enormous epigenomic plasticity in adult life. The variance in epigenetic marks increases with age, which is thought to reflect the response to environmental exposures in such a way that they modulate the expression of genes. However, studies in highly inbred rodent lines highlighted that a part of the phenotypic variation could not be attributed to environmental exposures.27 Also, monozygotic twin studies examining discordances have shown that part of the phenotypic variation is attributed to so-called “stochastic” variation, possibly caused by “molecular noise” due to imperfect control of the molecular interactions in the cell.28 Stochasticity, or random variation, is thought to have a large impact on disease susceptibility.29, 30
DNA Methylation and Skeletal Disorders
DNA methylation and the differentiation of skeletal cells
Osteoclast precursors derive from hematopoietic stem cells, whereas the bone- and cartilage-forming cells, osteoblasts and chondrocytes, derive from mesenchymal stem cells (MSCs). Pluripotent MSCs can also differentiate into other cell types, such as adipocytes and myocytes. The differentiation of MSCs toward the osteoblastic lineage is induced by the master transcription factors RUNX2 and osterix and is stimulated by ligands of the Wnt and BMP pathways.31 As it happens in other tissues, epigenetic factors play critical roles in determining the fate of MSCs. Specifically, genes that are characteristic of the osteoblast-osteocyte lineage (such as alkaline phosphatase, sclerostin, RANKL, osteoprotegerin, etc.) tend to undergo demethylation and de-repression during the differentiation of MSCs.32-34 In line with this concept, the demethylating agents 5-azacytidine and 5-deoxy-azacytidine improve the osteogenic differentiation of MSCs.35, 36 The ability of MSCs to proliferate and differentiate may decrease with aging, a phenomenon that may be explained, at least partially, by changes in the methylation and hydroxymethylation of DNA.37-39
The differentiation of osteoclast precursors is associated with marked changes in their DNA methylation signature, with hypermethylation and hypomethylation occurring in a variety of gene categories. In this process, PU.1 may play an important role, by recruiting DNMT3B to hypermethylated promoters, and TET2, which converts 5-methylcytosine to 5-hydroxymethylcytosine, to genes that become demethylated.40 Another DNA methyltransferase, DNMT3A, is essential to methylate and repress anti-osteoclastogenic genes, thus allowing osteoclast differentiation to continue.41 On the other hand, by influencing the expression of RANKL and OPG in cells of the osteoblast-osteocyte lineage, DNA methylation indirectly contributes to regulating osteoclastogenesis.33
Although cartilage does not undergo a remodeling process as bone does, epigenetic mechanisms also contribute to regulation of chondrogenesis and cartilage maintenance. DNA methylation modulates the expression of genes involved in cartilage homeostasis, such as GDF5, SOX9, and MMP13.42
A number of studies have explored the relation between genomewide methylation patterns and skeletal disease status in humans. These so-called epigenomewide association studies (EWAS) have a hypothesis-free approach that have the potential to find novel genes and/or pathways involved in skeletal disease. Because epigenetic marks are tissue specific, it makes sense to perform these studies in the tissue of interest. For osteoporosis, this would be bone, but for osteoarthritis, the tissue of interest is less straightforward because there are many tissues involved, including cartilage, bone, and synovial tissue. In addition, epigenetic marks in the circulation (blood) could be indicative of systemic mechanisms playing a role in disease and could also be easily accessible biomarkers for the disease. An overview of published studies in the field of osteoarthritis and osteoporosis is given in Table 1.
| Skeletal trait | Nr and type of samples | Tissue/cell | Adjustments | Technique | Results | Pathway enrichment | Reference |
|---|---|---|---|---|---|---|---|
| Knee/hip OA | 24 intact vs damaged | Cartilage | No | 450K array | 550 DMS | SMAD3, angiogenesis, inflammation | 46 |
| Knee/hip OA | 31 intact vs damaged | Cartilage | Sex, age, BMI, disease status | 450K array | 6272 DMS, 357 DMR | Skeletal development | 45 |
| Knee/hip OA | Intact vs damaged 12 discovery, 26 replication | Cartilage | No | 450K array | 9896 DMS, 271 DMR | Skeletal system development | 43 |
| Knee and hip OA and neocartilage | 12 cartilage vs 8 neocartilage from MSCs | Cartilage | No | 450K array | 5706 DMS | Developmental and transcription regulation processes | 51 |
| Knee OA | 5 intact vs damaged | Cartilage | No | Agilent Promoter array | 1214 promoters | Development, TGFbeta-, MAPK, hedgehog | 47 |
| Knee OA | 10 intact vs damaged | Cartilage | No | RRBS | >1000 DMRs | Embryogenesis and skeletal development | 44 |
| Knee OA progression | 12 early vs interm; early vs late | Cartilage Subchondral bone | No | 450K array | 0;519 DMS 72;397 DMS | Skeletal system development Morphogenesis, skeletal development, HOX | 142 |
| Knee OA and controls | 18 controls vs 23 OA | Cartilage | No | 27K array | 91 DMS | Inflammation, transcription activity, phosphorylation, MAPK | 50 |
| Knee OA and controls | 11 controls vs 12 OA | Cartilage | Sex, BMI (age) | 450K array | 929 DMS (0 after age adjustment) | Integrin, Wnt, FGF | 143 |
| Knee OA and controls | 4 TKR vs 4 controls | Chondrocytes | No | MeDIP seq | 70,591 DhMR | Many, including Wnt, bone remodeling, inflammation | 144 |
| Hip OA | 9 osteophytes vs cartilage | Chondrocytes | No | 450K array | 3161 DMR | Extracellular matrix, skeletal system development, endodermal cell differentiation | 145 |
| Hip fracture and OA | 22 fracture vs 17 OA | Bone marrow MSCs | Age | 450K array | 9038 DMS | Only enhancer CpGs: stem-cell development, osteoblast differentiation, Wnt pathway | 58 |
| Hip fracture and OA | 7 hip fracture vs 11 OA | Cartilage | No | 450K array | 103 DMS | Embryonal skeletal development, HOX | 48 |
| Hip fracture and OA | 27 fracture vs 26 OA | Bone | Age | 27K array | 241 DMS | Neuron differentiation, glycoprotein, skeleton, Homeobox | 57 |
| OP | 40 normal BMD vs 26 OP | Bone Blood | No Age, sex, cell counts, batch | 450K array | 63 13 | Not reported | 59 |
| OP | 22 normal BMD vs 12 early OP/10 advanced OP | Blood | No | 450K array | 1233 | Not reported | 61 |
| BMD | 4614 discovery; 901 replication | Blood | Age, weight, sex, smoking, family structure, cell counts, batch | 450K array | 0 | NA | 146 |
- OA = osteoarthritis; DMS = differentially methylated site; BMI = body mass index; DMR = differentially methylated region; DhMR = differentially hydroxymethylated region; RRBS = reduced representative bisulphite sequencing; OP = osteoporosis; BMD = bone mineral density.
Epigenomewide studies in osteoarthritis
A relatively large number of EWAS have been done focusing on the differences in the methylome of diseased versus preserved cartilage,43-47 cartilage from hip fracture,48, 49 controls,48, 50 or neocartilage engineered from MSCs51 (Table 1). Although the exact genes that are identified in the studies can differ, the identified pathways that emerge are robust. They involve skeletal development and morphogenesis and known signaling pathways (such as the TGFbeta, Wnt, HOX) involved in skeletal development. Also, inflammation seems to be a pathway that is identified by multiple studies and two studies have also shown the possibility of clustering a subset of patients characterized by methylation differences in (or near) inflammatory genes.50, 52 It is interesting to note that genomewide methylation studies consistently found enrichment of differentially methylated CpGs in enhancer regions.53 This suggests that modifications of DNA methylation marks may be more important in distant regulatory regions than in proximal promoters. This observation is consistent with other EWAS of complex diseases, where most of the associations found have been highly enriched in enhancer regions. Good annotation and interpretation of these enhancers are therefore of great importance. Methylation patterns can also be used to calculate a so-called “epigenetic age,” which is thought to reflect biological aging and has been linked to a whole range of age-related diseases and time of death.54, 55 One study examined the relation between this “epigenetic clock” and osteoarthritis and observed accelerated aging in OA cartilage.56
Epigenomewide studies in osteoporosis
The number of EWAS studies that examine osteoporosis as an endpoint is much lower than the cartilage studies described above. The first study examined femoral head bone tissue from hip fracture and OA patients using a limited set of methylation sites, and showed among a number of pathways, differentially methylated regions in the family of HOX-genes.57 A second, more recent study, examined MSCs in fracture versus OA patients and identified a number of differentially methylated genes in stem cell and osteoblast differentiation pathways.58 In another study using bone biopsies of women with low (osteoporotic) or normal BMD, 63 differentially methylated CpGs were found at a lenient false discovery rate (FDR <0.1).59 Because bone is a multicellular tissue, methylation differences could also reflect the changes in cell-type proportions, which was not accounted for in the published studies.
Methylation signatures in the circulation can potentially be powerful biomarkers for disease. A large EWAS examining methylation patterns of circulating leukocytes was performed in a collaborative study examining the relationship with BMD in a total of 5515 participants.60 However, this did not result in robustly associated CpGs, suggesting that blood might not be the correct tissue to study the epigenomic features in relation to bone phenotypes. In contrast, another small study examining 22 controls versus 22 osteoporotic women (defined by low BMD) identified a large number of differentially methylated sites.61 Importantly, this study lacked replication and adjustment of several important possible confounders, such as cell counts and BMI.
Non-coding RNAs in Skeletal Disorders
Among ncRNAs, miRNAs have been most extensively studied in relation to bone and cartilage diseases. More than 35 miRNAs modulate the differentiation of osteoblast precursors in vitro, through various mechanisms, including targeting master regulators such as RUNX2 (see recent reviews).62-64 A few of them have demonstrated effects on bone in animal models in vivo and may be involved in osteoclast-osteoblast communication. For example, osteoclast-derived exosomal miR-214-3p inhibits osteoblast activity in vitro and reduces bone formation in vivo.65 The miR-34 family and miR-214 also tend to have a negative influence on bone formation.66 On the other hand, miR-2861 and miR-29a stimulate bone formation. Interestingly, they target HDACs, thus illustrating the interactions between different layers of epigenetic marks.67 As with the osteoblastic lineage, several miRNAs influence osteoclast differentiation in vitro. Some of them have been validated in vivo. For instance, miR-503, which targets RANKL, and miR-34a inhibit bone resorption in animal models, whereas miR-148a tends to stimulate resorption.62, 64, 68 Some miRNAs are abundant in cartilage and appear to be important for the regulation of metalloproteases, toll-like receptor signaling, and other genes involved in catabolic pathways.69 A recent review of 57 studies about miRNA expression in cartilage revealed 46 differentially expressed miRNAs in OA, which were involved in autophagy, chondrocyte homeostasis, and degradation of the extracellular matrix.70 For instance, miR-140 is involved in the pathogenesis of OA by regulating, at least in part, MMP13 and ADAMTS5. miR-140 is downregulated in OA cartilage and the intra-articular injection of miRNA-140 alleviates OA progression in rats.71 The potential role of other miRNAs in OA pathogenesis has been recently reviewed.42, 72 In a few cases, their effects have been validated by gain-of-function or loss-of-function experiments in vivo (Table 2).
| Positive effects on bone formation or bone mass (ref.) | Negative effects on bone formation or bone mass (ref.) | Amelioration of osteoarthritis (ref.) | Aggravation of osteoarthritis (ref.) |
|---|---|---|---|
| miR-135147 | miR-103148 | miR-140-5p149 | miR-101150 |
| miR-141151 | miR-125b152 | miR-142-3p153 | miR-181a-5p138 |
| miR-145a154 | miR-133155 | miR-210-5p156 | miR-221157 |
| miR-148158 | miR-138-5p159 | miR-370160 | |
| miR-199a-5p161 | miR-14076 | miR-373160 | |
| miR-21a74 | miR-145a162 | miR-98163 | |
| miR-216a164 | miR-146165 | ||
| miR-26aa166 | miR-148167 | ||
| miR-29b-3p168 | miR-208169 | ||
| miR-33577 | miR-21a170 | ||
| miR-34a-5p171 | miR-214-3p66 | ||
| miR-503172 | miR-22275 | ||
| miR-286167 | miR-23b173 | ||
| miR-375174 | miR-26aa175 | ||
| miR-31-5p78 | |||
| miR-34176 | |||
| miR-383177 | |||
| miR-495178 | |||
| miR-503-5p179 | |||
| miR-542-3p180 | |||
| miR-92a181 |
- a Some miRNAs have shown contradictory effects in different models.
Use of miRNAs to diagnose and treat disease
The translational potential of miRNAs is large, since they can potentially be used to directly treat disease. A number of miRNAs enhance the osteogenic differentiation of MSCs; hence, they have been pointed out as potential therapies to promote bone regeneration. For example, scaffolds and particles loaded with analogs of miR-21 or miR-148 potentiate bone regeneration in experimental models, at least in part, by interacting with the RUNX2 pathway.73, 74 On the other hand, anti-sense oligonucleotides and “sponges” inhibiting miR-22, miR-29, miR-31, miR-133, miR-138, or miR-214 have a stimulatory effect on osteogenesis and may improve fracture healing.62, 75, 76 Also, miRNA-engineered MSCs may find a role in bone regeneration.77, 78 miRNA-based therapies have also been explored in joint disorders. In this view, the intra-articular injection of lentiviruses expressing miR-140 and miR-210 ameliorated joint disease in experimental models of OA.79
Many cells can secrete miRNAs (mainly via exosomes) that are detectable in the synovial fluid and the circulation. Hence, several miRNAs have been suggested as potential biomarkers for the diagnosis or follow-up of skeletal disorders (see recent reviews).72, 80-82 However, in general, the identity of regulated miRNAs varies widely across the studies and sometimes conflicting evidence is found. In addition, given the absence of replication across study outcomes and the small sample sizes, these studies should be considered as exploratory and further replication and validation is warranted. It is also to mention that circulating levels of miRNAs reflect processes in the whole body, most notably those of blood cells, and therefore miRNA signatures could be influenced by co-morbid conditions and other circumstances (such as aging in general, low-grade inflammation, etc.).80 Thus, published data are promising, but more work is needed before miRNAs can serve as robust diagnostic and prognostic tools in the clinic.
Histones and Chromatin Structure in Skeletal Disorders
Evidence is slowly being accumulated for a role of histone PTMs in chondrocyte and osteoblast differentiation, skeletal development, and the pathogenesis of OA and osteoporosis.83-88 There is a well-established role for HATs and HDACs in chondrogenesis, involving the regulation and function of Sox9 and its downstream targets. Sox9 is an essential regulator of chondrocyte differentiation and homeostasis.89 The HDAC KDM4B mediates Sox9 activation by removing a repressive histone PTM (namely, methylation of lysines 9 in histone 3; H3K9me3) from the Sox9 promoter region, which in turn triggers TGF-β-mediated chondrogenesis.90 Regulation of downstream targets of Sox9 is also dependent of histone remodelers. The HDAC Sirt1 binds to the enhancer and promoter region of COL2A1, which in turn recruits the HATs p300/CBP.91 These form a complex with Sox9 at the COL2A1 promoter. By relaxing the chromatin structure near the COL2A1 promoter by P300/CBP through their HAT activity, Sox9 is able to initiate COL2A1 expression.92, 93 Recently, the HDAC KDM6B has also been suggested to regulate COL2A1 expression, possibly also by direct interaction with the promoter of COL2A1.94 Conversely, the HDAC ELF3 suppresses COL2A1 transcription by inhibiting the HAT activity of the Sox9/CBP complex.95, 96
Many other histone remodelers are being implicated in osteoblast and chondrocyte development.83, 97 Recently, KDM6B, a HDAC, has been shown to be significantly increased in expression during cartilage development.94 Also, identified through genomewide association studies (GWAS), DOT1L, a histone lysine methyltransferase, might be involved in chondrocyte differentiation and homeostasis.98, 99 The proteins that “read” the histone PTMs are also important for cell differentiation. Proteins from the bromodomain and extra-terminal domain (BET) family recognize and bind to acetylated lysine on histones, forming a scaffold for protein complexes involved in gene transcription.100, 101 The inhibition of BET proteins leads to a suppression of osteoclast differentiation and activity in vitro.100 Because of these seemingly key roles, HDACs, HAT, and BET have been suggested as novel therapeutic targets in a range of skeletal diseases.97, 102 Thus, JQ1 and other inhibitors of BET proteins ameliorate bone loss in ovariectomized mice and several preclinical models of inflammatory disorders, such as arthritis and periodontitis.102, 103
However, it is important to keep in mind that most histone remodelers and readers also fulfill essential roles in other tissues and cell types.104 For example, DOT1L is not only essential for chondrogenesis and cartilage homeostasis99 but also for telomere silencing, meiotic checkpoint control, and DNA damage response.105, 106 Murine knockout models of DOT1L show multiple developmental abnormalities, not only restricted to skeletal abnormalities.106 Similar observations can be made for most histone remodelers associated with skeletal development and disease, such as the sirtuins (SirT1),107 KDM6B,108 and BET proteins.109 Because histone remodelers and readers are involved in a plethora of cellular processes in diverse cell types,110 globally administered therapeutic targeting may produce off-target and side effects, illustrating the need to examine such histone remodelers and readers in careful detail.
The chromatin structure itself can also modulate gene regulation and expression. For example, disruption of the TAD structure near certain genes can cause congenital skeletal disorders. Depending on the type and size of the of TAD disruption, brachydactyly, syndactyly, and polydactyly may be caused by changes in enhancer-promoter regulation in the WNT6/IHH/EPHA4/PAX3 locus.111 Further, the deletion of a TAD boundary as a disease mechanism has also been proposed for Liebenberg syndrome, a rare disorder where the arms of the patient acquire morphological characteristics similar to those of the legs.112 For a comprehensive review regarding the disruption of TADs and skeletal disorders, see Lupiáñez and colleagues.7
Up to now, only small-scale data are available on histone modifications and 3D chromatin structure of chondrocytes and bone. This may reflect the novelty of these data, the costs, material amounts, and skills needed to perform such experiments and analysis. However, recent large-scale efforts from the ROADMAP consortium113 and the encyclopedia of DNA elements (ENCODE)114 have built genomewide maps of several histone modifications and chromatin conformations in multiple human cells and tissues, including bone and cartilage.4 Those data are freely accessible (Table 3) and can be used as an epigenomic reference map for the locations of regulatory elements in osteoblasts and chondrogenic cells. To our knowledge, no chromatin conformation capture data of chondrocyte or osteoblast is available. However, TADs have been shown to be stable across cells, tissues, and even species, which highlights biological relevance and suggests that they function as a general 3D framework to determine domains of possible interaction partners. Thus, chromatin conformation maps are a valuable resource to identify potential causal variants and possible causal genes in GWAS. TAD boundaries can be used to limit possible causal genes, and reference epigenome maps can help to identify active genes and variants in cell-specific active regulatory elements.
| Project | Cell | Type | Source | Experiment | Data | Combined data | Reference |
|---|---|---|---|---|---|---|---|
| ENCODE | Chondrocyte | Primary cells | Knee articular chondrocyte | RNA-seq | Small RNA expression Total RNA expression | 114 | |
| Osteoblast | Primary cells | ??? | RNA-seq RNA-microarray CAGE RRBS DNase-seq ChIP-Seq | Small RNA expression Total RNA expression Transcriptional start sites Methylation DNA accessibility Histone marks: H2AFZ, H3K27ac, H3K27me3, H3K36me3, H3K4me1, H3K4me2, H3K4me3, H3K79me2, H3K9me3, H4K20me1 DNA binding proteins: CTCF, EP300 | Full reference epigenome by ROADMAP: E129 | 113 4 | |
| ROADMAP | Chondrocyte | Cultured cells | Mesenchy mal stem cell derived | ChIP-Seq CAGE | Transcriptional start sites Histone marks: H3K27ac, H#K27me3, H3K36me3, H3K4me1, H3K4me3, H3K9ac, H3K9me3 | Full reference epigenome by ROADMAP: E049 FANTOM5: active enhancers | 113 4 182 |
- ENCODE = encyclopedia of DNA elements; ROADMAP = Roadmap Epigenomics Mapping Consortium; RNA-seq = RNA sequencing; CAGE = cap analysis gene expression; RRBS = reduced representation bisulfite sequencing; DNase-seq = DNase-I hypersensitive sites sequencing; ChIP-seq = chromatin immunoprecipitation sequencing.
Interpretation of Epigenetic Studies: Problems and Tools
Cause or effect: confounding and reverse causation
Interpretation of epigenetic studies is not trivial because many factors can influence the association between the phenotype and epigenetic features. In contrast to DNA-sequence variants, epigenomic features are dynamic, meaning that they can be highly tissue specific (especially enhancers), dependent on environmental stimuli and developmental stage. This makes epigenetic analysis vulnerable for classical epidemiological pitfalls such as confounding and reverse causation. Besides the previously mentioned cellular heterogeneity as a potential confounder in case of multicellular tissues, such as bone or synovial tissue, a major issue is reverse causation. In cross-sectional studies, one can never be sure whether the epigenomic features cause the phenotype or vice versa. With respect to the epigenetic studies performed in cartilage and bone, this problem is also realistic. The massive epigenomic deregulation apparent in degraded cartilage can be a consequence of a process initiated by an entirely different cause of the disease. It is possible to use known genetic association and the concept of mendelian randomization to investigate the potential causal relationships between DNA methylation and the phenotype of interest.115 In these kinds of studies, single nucleotide polymorphisms (SNPs) known to influence the CpG site are used as genetic instruments to test for causation. Similarly, SNPs associated with the trait/disease of interest can then be used to test whether they are associated with methylation levels at the same CpG site. In this way, the direction of cause can be disentangled (Fig. 2). This approach has recently been used to show that DNA methylation differences associated with BMI are predominantly a consequence of adiposity, rather than a cause.116 In addition, this methodology has also been used to suggest that several known risk factors do not show a causal effect on bone fracture, except for BMD.117 Similarly, high BMI was shown to cause OA, but other known clinical risk factors did not have a causal relationship with OA.118 Typically, these mendelian randomization studies need genetic data with large sample sizes, which is increasingly available for both osteoporosis (GEFOS consortium) and osteoarthritis (Genetics of OA-consortium). However, sample sizes for the methylation studies in target tissue are typically small (Table 1), and therefore collaboration and meta-analysis across the different data sets are needed. This is increasingly recognized in the field.119
Integrating different omic levels
Interpretation of epigenomic data is complicated by the fact that the function of the genome is not fully known. It has become apparent that gene expression can be regulated by genomic features (enhancers/insulators) far away from the gene.120 In fact, recent studies in cancer show that variation in methylation in distal enhancers account for a larger part in the regulation of expression, then promoter-methylation variation.121 Similarly, den Hollander and colleagues observed that only 10% of the differentially methylated regions in cartilage is associated with differential expression of the nearest gene.45 Integrating epigenomic data with data from other molecular layers, such as RNA expression and/or protein expression, can help to identify the function of the epigenomic features.
Epigenetics as mechanism for genetic etiology of disease
Of all the genetic loci identified by GWAS for complex diseases (such as OA and osteoporosis), the majority does not affect the protein coding but is instead thought to affect gene expression regulation. Epigenetics may mediate genetic risk, meaning that the methylation state of a specific locus is driven by a nearby genetic variant(s), also called a methylation quantitative trait locus (meQTL). In this way, integration of epigenomic and genetic data can be used to elucidate the function and biological pathway underlying the genetic association. Large-scale studies in blood have shown that meQTLs are widespread (>30% of all CpGs have a meQTL), but that only 10% of the meQTLs also associate with expression levels of nearby genes.122 A number of studies have examined the relation between previously identified genetic loci for BMD and/or OA and methylation (and/or gene expression) in the target tissue. These studies identified a number of meQTLs specific for cartilage,123-125 suggesting that methylation mediates the association between the genetic locus and disease. Notably, absence of a meQTL in these studies does not mean that methylation does not play a role, since methylation is dynamic and context-dependent and can be dependent on a specific stimulus or developmental stage.
Annotating epigenomic features: available information regarding skeletal tissues
Recent large-scale efforts have provided new understanding in the function of epigenetic modifications, and efforts are ongoing to map histone modifications, transcription factor binding, and 3D chromatin structure of multiple cell types and tissues.5, 43, 44 The data generated by these large-scale efforts are publicly available and contain DNA annotation on multiple levels: histone modification, binding of transcription factors and other DNA binding proteins, methylation, gene expression, DNA accessibility, and chromatin conformation. Some also contain epigenetic information on cartilage and bone tissues; most notably are the “reference” epigenomes for osteoblasts and chondrocytes generated by the ROADMAP consortium (Table 3). The reference epigenomes construct an annotation of functional elements, ie, enhancers, promoters, etc., of the DNA per cell type,4 which has already been proven to be a valuable resource for skeletal GWAS and EWAS, to finemap causal SNPs, CpGs, and genes, and also to provide for a hypothesis on the mechanism underlying GWAS/EWAS findings for skeletal outcomes.126, 127
The interesting SUPT3H-RUNX2 locus provides a good example on how different levels of molecular data can be combined to build a model for the regulatory mechanisms involved in the RUNX2 locus, a master regulator of osteoblast and chondrocyte regulation (Fig. 3). This locus encompasses a region of roughly ∼700 kb, where multiple genetic association signals have been identified for OA,127 BMD,128 OPLL (ossification of the posterior longitudinal ligament of the spine), cartilage thickness (measured as the minimum joint space width),127 height,129 and facial morphology.130
Interestingly, these GWAS signals are independent from each other.127 Data from the ROADMAP consortium shows that several of the GWAS signals are located in different (potential) enhancers in cartilage and osteoblast cells. This suggests that the genetic association with various skeletal phenotypes might arise due to a difference in control of RUNX2 expression. Recent studies have shown that regulation of RUNX2 is tightly regulated by multiple epigenetic mechanisms, such as miR-204/211 and other miRNAs,131, 132 HDAC,133 and DNA methylation.123, 134 Also, some of the genetic association signals near RUNX2 exert effects on epigenetics. For the osteoarthritis-associated SNP, rs10948172, it has been shown to operate as a methylation quantitative trait loci (mQTL) for 4 CpG sites in the RUNX2 locus.123, 134 Nevertheless, the GWAS signals are positioned at a relative large distance from RUNX2, and chromatin interaction from regulatory region to gene promoter is needed. Using TAD localizations from the 3D genome browser, we notice that all skeletal genetic RUNX2 signals are within one TAD. This makes physical interaction of enhancers—promoters through chromatin “looping” likely to occur within this region. Such loops are mostly regulated by several DNA binding proteins such as cohesin and/or CTCF.135 Using data from the ENCODE database, several CTCF binding locations in osteoblast primary cells can be identified near the RUNX2 P1 and P2 promoter regions (Fig. 3). One can imagine that depending on the developmental stage, the different enhancer regions are actively involved in regulation of expression, and different chromatin loops are established with the P1 and/or P2 promoter of RUNX2. The RUNX2 regulatory region carries all elements of a so-called super-enhancer region, which is a genomic region comprising multiple enhancers. These super-enhancers are typically identified near genes important for cell-identity/master transcription regulation genes, which require detailed regulation of expression, such as is the case for RUNX2.
What Will the Future Hold?
Advances in understanding the epigenomic mechanisms have greatly expanded our knowledge about the molecular mechanisms involved in the differentiation and activity of cells responsible for skeletal homeostasis, which are central players in the pathogenesis of disorders such as osteoporosis and OA. Thus, we can anticipate that epigenetic mechanisms must also play an important role in these disorders and may become the foundation for new therapies. However, much more research is needed in order to use epigenetic marks as biomarkers of disease risk or progression and to introduce epigenetic-based therapies into the clinic.
To date, association studies aimed to elucidate the association of epigenetic signals with bone phenotypes have had limited reproducibility, particularly regarding DNA methylation marks. Similarly, some miRNA signatures have been produced in a few small-size studies, but replication in larger studies is still pending. Of course, this is critical before introducing the analysis of epigenetic markers as tools for diagnosis or prognosis of disease. It will likely require large collaborative, epigenomewide studies.
A major issue regarding therapy aimed at modulating epigenetic mechanisms is related to lack of specificity and undesired effects. Drugs interfering with DNA methylation are already being used to treat some bone marrow disorders, but their widespread effects make them rather unsuitable for non-neoplastic disorders. Thus, CRISPR-based methods and other procedures to induce targeted DNA methylation changes to specific loci in somatic cells may greatly help to first delineate the effects of methylation marks and then to epigenetically modify the activity of specific genes.136 SAHA-PIPs are a novel class of histone modifiers made by conjugating selective DNA binding pyrrole-imidazole polyamides (PIPs) with the histone deacetylase inhibitor SAHA. They show some selectivity and modulate the transcription of certain clusters of genes.137
Localized skeletal disorders, such as some forms of OA and delayed-union fractures may be amenable to local therapies, such as agonists and antagonists of specific miRNAs or genetically/epigenetically engineered MSCs. The local nature of the disease and the therapy may help to avoid generalized undesired effects. In this line, antisense oligonucleotides of miRNA 181a-5p attenuate cartilage destruction when injected into the knees of rats and mice with experimentally induced OA.138
The task of revealing the role of epigenomic variants and pathways is even more difficult because, unlike the genome, the epigenome is cell- and tissue-specific and may change over time. Thus, studies may need to be done using skeletal samples, which poses obvious difficulties for human studies and particularly for those requiring sampling at multiple time points. For studies on developmental processes for bone and cartilage in humans, model systems such as human stem cells and IPSCs might prove a good alternative. In addition, organ on a chip technology is promising for studying early disease processes. In this regard, recent findings that suggest that some molecules circulating in blood may be used as biomarkers of the status of solid tissues, if confirmed, may facilitate using epigenetic elements as biomarkers. In the skeletal field, miRNAs have been mostly studied from this perspective. However, methylation marks in circulating cell-free DNA are also being actively explored in cancer and other disorders.
Technical advances to facilitate high-throughput analysis of epigenomic marks at decreasing costs are emerging and will facilitate larger-scale applications. Sequencing costs are continuously going down, making it possible to generate increasing amounts of data. The bottleneck for these “big data” studies is the data analysis, which in general requires the same costs as generation of the raw data itself. An emerging technique is single-molecule sequencing, with Pac Bio's single-molecule real-time sequencing (SMRT) and Oxford Nanopore's (ON) nanopore sequencing as the most prominent players in the field. SMRT and ON are able to directly sequence native DNA or RNA, making it possible to directly measure chemical groups attached to the nucleic acid sequence (such as methylation), avoiding the bias of other techniques due to the necessary amplification and/or bisulphite treatment. Although single-molecule sequencing is still in a developmental stage, more and more applications are being developed, indicating that the technique is almost ready for wider-scale applications.
Another important technical advance is single-cell epigenomics, which is developing rapidly. Single-cell “omic” technologies have emerged as powerful tools to explore cellular heterogeneity at individual cell resolution. Previous scientific knowledge in cell biology is largely limited to data generated by bulk profiling methods, which only provide averaged read-outs that generally mask cellular heterogeneity. Single-cell technology has recently led to the identification of a self-renewing skeletal stem cell that generates progenitors of bone, cartilage, and stroma but not fat.139 In addition, the first study applying single-cell RNAseq on OA cartilage identified seven chondrocyte subpopulations in cartilage.140
Because of the high sensitivity of single-cell genomics, attention must be put into experimental setup and execution of this technology. Careful handling and processing of cells is critical to preserve the native epigenetic and expression profile; only then meaningful analysis and conclusions can be drawn. This is especially important for the study of skeletal cells because their natural habitat is normally deeply embedded in the extracellular matrix and getting a single-cell solution from cartilage or bone is not trivial.
A new field of epigenetic regulation is epi-transcriptomics.141 These are RNA modifications that represent a novel layer of regulation of gene expression. Early reports show the existence of many different dynamic RNA modifications. The role of these RNA modifications in regulating gene expression is largely unexplored, but recent evidence suggests that these RNA modifications are key switches for RNA metabolism.
It took more than 1,500 years to evolve from Ptolomeo's map to good world maps. The map of the human epigenome is probably much more complex. Fortunately, technical advances allow us to be optimistic and not expect such a long time for the final drawing. However, building the maps of the human epigenome and specifically of human skeletal cells, as well as the tracks to be travelled to get into the core mechanisms driving the move from healthy bones and joints toward disease, is still a formidable work. Intensive collaboration of multiple research groups working together is a must to accomplish such a daunting task. Previous experience in the genetic field shows that it may be difficult to do, but it is feasible. A new world of hope for millions of patients is ahead.
Disclosures
All authors state that they have no conflicts of interest.
Acknowledgments
Authors’ studies were supported by grants from Instituto de Salud Carlos III to JAR (PI16/915; which may be co-funded by EU funds) and Reuma Nederland (16-1-404).
Authors’ roles: JBJM, CGB, LLD and JAR contributed to conception, writing and revision of the manuscript.
