MicroRNA variants as genetic determinants of bone mass.

There are increasing reports of SNPs that can interfere with miRNA function. Such polymorphisms can occur in the miRNA targeting site of a mRNA (miR-TS-SNPs) or in the miRNA genes themselves (miR-SNPs) [ 7 - 9 ]. miR-TS-SNPs present at or near a miRNA binding site in a protein coding gene could modulate miRNA function by creating or eliminating a miRNA binding site in that target mRNA [ 10 ]. The functional impact of a miR-TS-SNP on phenotype largely depends on whether the corresponding miRNA is expressed in a particular tissue, as well as the expression of other possibly compensatory miRNAs. For example, miRs and their targets can be expressed in a tissue specific manner, and a miR-TS-SNP might only be of functional consequence if the targeted mRNA is expressed in the same tissue. These considerations could provide some basis for understanding the tissue-restricted phenotype associated with some common SNPs [ 5 ]. MicroRNA biogenesis is an intricate process; multiple steps are involved in generating a mature miRNA from its primary (pri-) miRNA transcript. These include the cleavage of pri-miRNA by DGCR8/ Drosha microprocessor complex, to yield a shorter hairpin, the precursor miRNA (pre-miRNA) [ 11 - 14 ]. Pre-miRNAs are actively exported from the nucleus to the cytoplasm and further processed by the Dicer/TRBP complex into roughly symmetric miRNA duplexes. This duplex is loaded onto the Argonaute (Ago) protein of the RNA induced silencing complex (RISC), unwound, and processed into a single stranded mature miRNA [ 15 , 16 ]. Guide strand selection from the miRNA duplex is governed by thermodynamics and the remaining passenger strand is often degraded. However, in some cases, both strands of duplex can be loaded into the RISC as functional miRNAs [ 17 , 18 ]. Thus, SNPs in miRNA genes (miR-SNPs) have the potential to alter miRNA processing, guide strand selection, or the mature miRNA, itself. Since a single miRNA could target hundreds of genes and regulate multiple pathways, a polymorphism in a miRNA gene leading to functional impairment could impact all regulatory pathways involving that miRNA. It is estimated that only approximately 10% of human miRNA genes contain SNPs; and SNPs in the region corresponding to the mature miRNA are the most rare (<1%) [ 8 , 19 ]. Thus, there appears to be considerable selective constraint on the transmittance of miR-SNPs.

In terms of trying to understand the function of mutations or polymorphisms in miRNA genes and targets, the field of cancer biology has been in the forefront [ 20 ]. Similar work in the field of skeletal biology is in its infancy. Fortunately, some miRNAs and mRNA targets important in malignancy are also important in bone or cartilage. In this review, we have included studies performed in the context of cancer and other diseases to infer the function of miRNA gene and target polymorphisms in skeletal biology. Although long non-coding RNAs (lncRNAs) that are frequently targeted by miRNAs, also contain SNPs and can impact skeletal physiology, there is relatively little known about the function of lncRNAs in bone [ 21 - 23 ]. Consequently, in reviewing the published data on sequence variants in miRNAs and miRNA targets, the following discussion will be restricted to protein coding genes and miRNAs that could have an impact on the skeleton. Moreover, many studies have associated miRNA-SNP variants with a particular phenotype, although few studies have identified a potential mechanism involved. Therefore, this review will also be restricted to work demonstrating molecular mechanisms and validated SNP function. Lastly, each miRNA has many possible regulatory interactions with targets. Modeling all regulatory circuits for the selected miRNAs presented below is beyond the scope of this review. Rather, we have limited our models to the miRNAs, targets and signaling pathways discussed in each section.

MicroRNAs regulate skeletal development and homeostasis by targeting multiple genes and pathways [ 24 - 26 ]. Thus, SNPs that alter expression and/or function of a single miRNA could represent variants with larger effects. SNPs that compromise miRNA expression levels can occur in the miRNA gene promoter (miR-P-SNPs), or in the body of the miRNA gene (miR-SNP) ( Figures 1 and 2 ). miR-P-SNPs modulate miRNA expression by affecting transcription factor binding sites in one of the following ways: (1) abolishing an existing transcription factor binding site (TFBS) or (2) creating a new TFBS or (3) altering the binding affinity of TFBS [ 27 ]. Reportedly, databases using bioinformatics to predict TFBS in miRNA promoters (i.e. miRGen 2.0) and the potential for a SNP to affect transcriptional regulation of a miRNA gene (i.e. dPORE-miRNA) have estimated that ~20,000 SNPs prevail in human miRNA promoters [ 28 , 29 ]. Despite the elaborate predictions, however, information on the impact of miR-P-SNPs on disease risk is limited and the mechanism by which miR-P-SNPs affect miRNA gene regulation is known for only a few cases. In contrast, miR-SNPs affect miRNA processing by altering the RNA secondary structure and decreasing the stability of either pri- or pre-miRNA. Alternatively, SNPs can lead to increased stability of the passenger strand, which could then function as another guide strand ( Figure 2A and B ). In this case, both guide and passenger strands will be loaded into the RISC, to target a distinct set of mRNAs [ 7 ]. Lastly, SNPs that compromise miRNA function most commonly occur in the seed-binding region of the miRNA ( Figure 2C ). In this section, we will highlight SNPs in miRNAs that are important in the skeleton, and how these SNPs modulate miRNA transcription, processing or function. The human miR-146 family consists of 2 distinct isoforms, miR-146a and -146b, which are expressed from human chromosomes 5q33.3 and 10q24, respectively [ 30 ]. miR-146a is expressed in articular cartilage and femur, as well as in non-skeletal tissues including prostate, liver, breast, and hematopoietic cells [ 31 - 36 ]. An NF-κB binding motif near the transcription start site of miR-146a has been recognized, and factors including TNF-α, lipopolysaccharide or IL-1β (interleukin-1β) can induce miR-146a expression [ 34 , 37 ]. In turn, miR-146a limits the inflammatory response by targeting TRAF6 (TNF receptor associated factor 6) and IRAK1 (IL1 receptor associated kinase 1), which play a role in activating NF-κB signaling [ 34 ]. In osteoclasts, TRAF6 and IRAK1 mediate IL-1β induced activation of NF-κB signaling, to promote osteoclast survival and activity [ 38 , 39 ]. miR-146a has been linked to the pathogenesis of osteoarthritis (OA), a disease characterized by the progressive degeneration of articular cartilage, primarily in response to mechanical injury [ 37 , 40 , 41 ]. Studies have reported high levels of miR-146a in synovial fluid during the early stages of OA, whereas in the later stages miR-146a levels decline [ 40 - 42 ]. miR-146a can target Smad4, which might suppress TGF - β mediated chondrocyte survival [ 40 , 41 ]. In the later stage of OA, lower levels of miR-146a may contribute to increased TRAF6 and IRAK1 levels, and IL-1β mediated NF-κB signaling. NF-κB signaling promotes the expression of cartilage-degrading enzymes, including MMP13 and ADAMTS5, while it suppresses expression of cartilage matrix such as aggrecan and type II collagen [ 37 ] ( Figure 3 ). miR-146a also plays a crucial role in regulating bone remodeling. It functions as a positive regulator in osteoblasts, by driving commitment of mesenchymal stem cells (MSCs) to an osteogenic rather than a chondrogenic fate [ 43 , 44 ]. By targeting Smad2/3, miR-146a regulates TGF-β signaling and indirectly downregulates expression of chondrocytic marker Sox9, and upregulates Runx2 [ 37 , 40 , 45 ]. In osteoclasts, miR-146a levels increase during differentiation, and miR-146a has been shown to inhibit both TNF-α and RANKL induced osteoclast differentiation by mechanisms likely to include the targeting of TRAF6 and IRAK1, as well as other mRNAs [ 46 , 47 ]. The miR-146a gene contains SNPs in the promoter region as well as in the pre-miRNA. miR-P-SNP rs57095329 is located within 1 kb of the miR-146a transcription start site and was first identified as strongly associated with risk of systemic lupus erythromatosis (SLE), a chronic autoimmune disorder affecting multiple organ systems [ 27 ]. This disease association was later validated in a meta-analysis of GWAS that included cohorts of European and Asian descent [ 48 ]. The studies in SLE linked the risk-associated minor allele variant rs57095329-G (A>G, minor allele frequency (MAF) 0.14 in 1000 Genomes Project) with reduced miR-146a promoter activity, leading to low mature miR-146a levels. The risk allele was shown to reduce the binding affinity of Ets1 to the miR-146a promoter, providing a potential mechanism for decreased miR-146a levels [ 27 ]. In bone, Ets1 is also known to interact with CbfA1/Runx2, a quintessential transcription factor for osteogenesis [ 49 ]. In cells from a patient displaying Runx2 haploinsufficiency, levels of miR-146a were depressed, further suggesting a connection between Runx2 and miR-146a [ 50 ]. Although other SNPs in the promoter region of miR-146a have been recognized (rs17057381, rs73318382, and rs6864584), their function has not yet been ascertained. Interestingly, miR-146a has been also implicated in the pathogenesis of rheumatoid arthritis, another autoimmune disease affecting skeletal tissues [ 47 , 51 - 54 ]. In rheumatoid arthritis, joint destruction and localized bone loss are caused by chronic inflammation; and miR-146a has been shown to suppress osteoclastogenesis [ 47 , 52 ]. Although genetic determinants play a role in susceptibility to rheumatoid arthritis, an association between the SLE-linked miR-P-SNP rs57095329 and rheumatoid arthritis has not yet been examined. A second functional SNP in miR-146a occurs in the pre-miRNA, SNP rs2910164. This common SNP (G>C, MAF 0.458 in CSAgilent Project) was mapped to the passenger strand in the stem of pre-miR-146a. This SNP was associated with risk of many cancer types, including papillary thyroid carcinoma, prostate cancer, breast cancer, gastric cancer, ovarian cancer, esophageal squamous cell carcinoma, and hepatocellular carcinoma [ 30 , 33 , 55 - 61 ]. The mechanism for SNP rs2910164 mediated alterations in mature miR-146a levels has been examined in studies of papillary thyroid carcinoma and nasopharyngeal carcinoma. GC heterozygosity was linked to high papillary thyroid carcinoma risk, while CC homozygosity was associated with increased nasopharyngeal carcinoma risk [ 30 , 57 , 61 ]. The risk-associated C allele resulted in lower miR-146a levels, due to impaired Drosha mediated processing of the pre-miRNA [ 57 ]. This led to decreased regulation of miR-146a target genes, TRAF6 and IRAK1. Apart from impairing Drosha processing, this SNP was also suggested to stabilize the passenger strand. In addition to the leading strand, papillary thyroid carcinoma patients with GC heterozygosity were shown to generate two passenger strands, −146a*G and −146a*C [ 55 ]. Bioinformatics suggest that these alternate passenger strands target different mRNAs, which might contribute to the risk of papillary thyroid carcinoma. Although rs2910164 was previously linked to rheumatoid arthritis, a later meta-analysis disproved this association [ 53 , 54 ]. These seminal studies of miR-146a polymorphisms elegantly demonstrate the impact of miRNA-SNPs on miRNA transcription, processing and targeting. Given the role of miR-146a in regulating inflammation, autoimmunity, and skeletal homeostasis, it will be important to examine the function of these and other miR-146a SNPs, and determine whether they might be associated with skeletal diseases. The human miR-125 family is comprised of three homologs: miR-125b1, -125b2 and -125a. Both miR-125b1 (on chromosome 11q23) and -125b2 (on chromosome 21q21) genes produce identical mature miR-125b. The miR-125a gene is located on chromosome 19q13.41, within a cluster also containing miR-99b and let-7e [ 7 , 62 ]. During RANKL-stimulated osteoclastogenesis in vitro, the expression of miR-125a is increased in both human and mouse cells [ 46 , 63 ]. Upon RANKL stimulation, the p65 subunit of NF-κB can bind near the transcription start site of the miR-125a gene cluster, inducing its expression. In turn, miR-125a positively regulates NF-κB signaling by directly targeting TNFAIP3, a ubiquitin editing enzyme [ 63 ]. Remarkably, in B-cell lymphoma, ectopic expression of miR-125a was shown to suppress TNFAIP3 mediated ubiquitination of TRAF2 and IκBα, and lead to elevated NF-κB signaling [ 64 ]. Thus, impairment in miR-125a levels due to pri-miR-SNPs could modulate osteoclastogenesis and disrupt the balance in bone remodeling ( Figure 4 ). The other member of miR-125 family, miR-125b, acts as a negative regulator of osteoblastic differentiation, by targeting the Cbf (core-binding factor) family members Cbfβ and Runx2 (Cbfα1) [ 65 ]. miR-125b levels plummet in response to BMP 2/4 induced osteoblast differentiation [ 65 , 66 ]. Congruent with its negative role, increased miR-125b levels were detected in serum and bone samples of osteoporotic patients [ 67 ]. In cartilage, miR-125b was shown to confer protection against OA by directly targeting the cartilage degrading enzyme aggrecanase-1 (ADAMTS-4). Reportedly, IL-1β suppresses miR-125b levels and upregulates aggrecanase-1 expression in OA chondrocytes [ 68 ]. Mechanistic studies examining the effect of a pri-miRNA SNP on processing have been described for miR-125a. SNP rs12975333 (G>T), located in the stem region of pri-mi125a, was shown to modulate miRNA expression and activity. The minor variant, rs12975333-T, decreases binding of pri-miR-125a to DGCR8, a component of the miRNA biogenesis machinery. This causes defective pri- to pre-miR processing and low mature miR-125a levels; and contributes to the resulting increase in levels of miR-125a targets observed in rs12975333-T expressing cells [ 7 ]. In addition to affecting processing, rs12975333 is retained by the mature miRNA and was shown to impair miR-125a mediated translational repression. Specifically, the T variant of rs12975333 impairs the seed-binding region of miR-125a, and reduces repression of the miR-125a target gene, Lin-28 [ 7 ]. Lin-28 is a RNA binding protein that acts as a translational enhancer of Igf2 [ 69 ]. SNPs that reside in the seed-binding of mature miRNAs are quite rare, which might reflect their profound impact on cell physiology [ 8 , 9 ]. Although 6 polymorphisms in pri-miR-125a (rs41275794, rs12975333, rs12976445, rs10404453, rs78758318, rs143525573) have been reported, so far only rs12975333, described above, has been shown to be functional. Given the function of the miR-125 family in both the osteoblast and osteoclast lineage, it is of interest to determine whether these other SNPs display any association with BMD. The miR-196 family consists of 3 members, miR-196a1, -196a2, and -196b that are embedded in the highly conserved clusters of Hox (homeobox) genes. Located on chromosome 17, the miR-196a1 gene is lies within the Hoxb cluster, miR-196a2 lies on chromosome 12 within the Hoxc cluster, and miR-196b gene lies within the Hoxa cluster on chromosome 7. miR-196 isoforms show a high level of identity. Both miR-196a-1 and miR-196a-2 genes transcribe the same mature miR-196a sequence, while the mature miR-196b differs by a single base from miR-196a. This sequence similarity suggests that they may target the same panel of mRNAs. Hox genes regulate skeletal patterning and limb development by coordinating morphogen signaling pathways including those for retinoic acid, BMP and TGF-β [ 70 , 71 ]. By regulating the Hox genes (Hoxb8, Hoxc8, Hoxd8, Hoxa7), miR-196 family members play a crucial role in development of the axial skeleton, fine-tuning levels of HOX proteins in a temporal and spacial manner [ 70 - 76 ]. miR-196a promotes commitment of mesenchymal stem cells (MSCs) towards the osteoblast lineage by targeting Hoxc8 [ 77 , 78 ]. In osteoblasts, Hoxc8 acts as a negative regulator of differentiation and transcriptionally suppresses osteopontin [ 72 , 73 , 76 ]. In turn, BMP signaling represses Hoxc8 activity through SMAD1 interaction with Hoxc8, relieving repression of osteopontin transcription. In addition to Hoxc8, miR-196a targets Hoxb7 in adult MSCs [ 79 ]. Over-expression of Hoxb7 enhances the proliferative and osteogenic capacity of MSCs. In aging MSCs, decreased osteogenesis due to low Hoxb7 was associated with increased levels of miR-196a. Two miR-196a2 promoter region SNPs (rs35010275 and rs12304647) and one SNP in the pre-miRNA (rs11614913) have been studied in gastric cancer cohorts [ 80 , 81 ]. Of these, miR-P-SNP rs35010275 (G>C; MAF 0.28 in 1000 Genomes Project) mapped within 1 kb of the miR-196a2 transcription start site and was associated with gastric cancer risk in a Chinese population [ 80 ]. High levels of miR-196a detected in these patients were linked to the risk-associated G variant. Functional analyses indicated that rs35010275 modulated the interaction of nuclear proteins with DNA, and promoter activity. Although the transcription factor complexes interacting with the SNP rs35010275 region were not identified, this study demonstrates that the allele-specific differences in promoter region of a miRNA gene can alter promoter activity and modulate miRNA expression [ 80 ]. Another variant in the miR-196a2 precursor rs11614913 (C>T MAF 0.334 in 1000 Genomes Project) was associated with breast cancer risk [ 82 ]. Specifically, the T-allele was associated with decreased cancer risk and low miR-196a levels. Functional analysis revealed that the T variant leads to decreased mature miR-196a, without affecting the pre-miR-196a2 levels, thereby indicating a miRNA processing defect [ 82 ]. Since Hox genes play a critical role in bone and cartilage, these common miR-196a variants could have an impact on skeletal phenotype. The human miR-149 gene has been mapped to intron 1 of glypican1 (GPC1), an integral membrane proteoglycan. This highly conserved miRNA is expressed independently of its host gene in a wide variety of non-skeletal tissues [ 83 ]. In the skeleton, expression of miR-149 has only been reported in chondrocytes, where it is upregulated by the inflammatory cytokines IL-1β, IL-6 and TNF-α. Furthermore, miR-149 can directly target these inflammatory cytokines in a feedback loop. In OA, dramatically decreased miR-149 expression has been suggested to exacerbate the inflammatory response [ 84 , 85 ]. So far, two SNPs have been mapped to pre-miR-149 at rs2292832 (C>T MAF 0.387 in 1000 Genomes Project) and rs71428439 (A>G MAF 0.144 in 1000 Genomes Project). Although rs2292832 was extensively studied in the context of cancer risk, newer meta-analyses do not suggest an association between this SNP and cancer, and functional data on rs2292832 are lacking [ 86 , 87 ]. In contrast, the mechanism for a SNP mediated effect on gene regulation has been elegantly described for rs71428439, in the context of myocardial infarction [ 88 ]. Specifically, the risk associated variant rs71428439-G was linked to decreased miR-149 expression. The SNP was predicted to alter the secondary structure of the pre-miR, yielding a less stable pre-miR-149-G. Functional analysis revealed that rs71428439-G modulates the pre- to mature miR processing, resulting in low levels of miR-149. In addition to defining mechanism, this study was unique in that it demonstrated the impact of rs71428439 on myocardial infarction susceptibility in vivo. In a mouse model of myocardial infarction, infarct size and myocyte apoptosis was low in mice injected with precursor for the miR-149- rs71428439-A variant, assigning a protective function against myocardial infarction. This SNP is a common variant, and since miR-149 targets inflammatory cytokines, an association between this SNP and OA phenotype could be possible [ 85 ]. The human miR-27 family is comprised of two family members: miR-27a and b. miR-27a is transcribed within a cluster on chromosome 19, which also contains miR-23a and miR-24-2, whereas miR-27b is transcribed in cluster on chromosome 9, along with miR-23b, miR-3074 and miR-24-1. The mature miR-27a and b isoforms differ by a single nucleotide. During osteoblastic differentiation in vitro, the expression of miR-27a is extensively regulated. miR-27a levels first decrease, as cells are proliferating and laying down extracellular matrix; miR-27a then increases as osteoblasts mature. However, in mineralizing cultures, the expression of miR-27a is nearly undetectable [ 89 ]. Runx2 was shown to negatively regulate the transcription of the miR-27a~ miR-23a~miR-24-2 cluster. Moreover, miR-27a targets SATB2 (special AT-rich sequence binding protein 2), which interacts with Runx2 to promote osteoblast differentiation. Inhibition of the miR-27a~miR-23a~miR-24-2 cluster by Runx2 can promote a feed forward regulatory loop [ 89 ]. Further, miR-27a can target Hoxa10, which promotes osteoblast differentiation through Runx2-dependent and –independent mechanisms [ 74 ]. Whereas these studies suggest that miR-27a has negative effects on osteoblastogenesis, others showed that the miR-27 family can promotes this process by directly targeting inhibitors of Wnt/β-catenin signaling, sFRP1 (soluble frizzled related protein 1) and APC (adenomatous polyposis coli) [ 90 , 91 ]. It is likely that the timing of miR-27 expression is tightly regulated for optimal osteoblastic differentiation ( Figure 5 ). To date, two SNPs have been reported in pre-miR-27a: rs895819 (T>C MAF 0.364 in 1000 Genomes Project) and rs11671784 (G>A MAF 0.007 in 1000 Genomes Project). rs895819 is located in the terminal loop of pre-miR-27a, and increased expression of mature miR-27a has been observed in individuals homozygous for the minor allele [ 92 , 93 ]. This SNP has been studied in the context of several cancers, most notably those of breast and colon. However, there are conflicting reports of the whether the rs895819C variant can be associated with breast or colorectal cancer risk [ 87 , 94 ]. In contrast, rs11671784 was strongly associated with risk of gastric cancer, and the A variant appears to have a protective effect [ 93 , 95 ]. This SNP is located in the stem region of pre-miR-27a, where rs1167178-A introduces a mismatch. The presence of the A allele decreased processing of pre-miR-27a to the mature form by ~50%. Expression of miR-27a is frequently increased in gastric cancer, and patients homozygous or heterozygous for the A allele demonstrated decreased tumor/normal ratio for miR-27a, compared with patients homozygous for the G allele. Decreased miR-27a corresponded to increased levels of the miR-27a target Hoxa10. Moreover, in a mouse xenograft model, gastric cancer cells expressing the miR-27a rs1167178-A allele grew more slowly than those expressing rs1167178-G [ 95 ]. This study elegantly demonstrates the function of a miR-27a variant that could have an impact on osteoblastic differentiation by altering levels of mature miR-27a. In humans, mature miR-124 arises from three different pri-miR structures encoded by independent genes. The genes encoding miR-124-1 and miR-124-2 have been mapped to chromosome 8, while miR-124-3 gene is located on chromosome 20. Mature miR-124 is evolutionarily conserved and highly expressed by neurons in brain [ 96 ]. In the skeleton, miR-124 is expressed by both osteoblasts and osteoclasts [ 97 , 98 ]. In osteoblasts, miR-124 expression decreases during differentiation and it negatively regulates osteoblast formation by targeting Dlx2, Dlx3 and Dlx5, which are critical for osteogenic commitment [ 98 ]. miR-124 has also been shown to suppress expression of chondrocyte marker genes such as Sox9 and Aggrecan [ 99 ]. In contrast, miR-124 has been shown to increase dramatically during commitment of human mesenchymal stem cells to adipocyte lineage [ 98 ]. Thus, miR-124 may promote adipogenenic differentiation at the expense of commitment to the osteo-chondral lineage. In osteoclasts, miR-124 can target NFATc1, key transcription factor; it also reduces osteoclast precursor proliferation and motility by targeting integrin B1 (ITGB1), RhoA and Rac1 [ 97 ]. In a rat inflammatory arthritis model, miR-124 mimic injection was shown to ameliorate progression of the arthritic lesion by decreasing osteoclast formation [ 100 ]. Polymorphism rs531564 (G>C MAF 0.130 in 1000 Genomes Project) was detected in pri-miR-124-1, and examined in a case-control study of Alzheimer’s disease. Although the SNP was not linked to Alzheimer’s disease, the study suggested that the rs531564-C variant might lead to decreased mature miRNA levels, possibly due an alteration in the pri-miR secondary structure [ 101 ]. Since rs531564C is a common variant, and because miR-124 impacts both osteoblast and osteoclast differentiation and function, it will be of interest to determine whether this SNP might be associated with bone phenotype.

Presently, miR-SNPs with a validated impact on bone mass have not been reported. However, a mutation in a miRNA precursor was recently associated with bone mass. The mechanism by which a mutation in miR-2861 affects miRNA processing and bone mass are discussed below. miR-2861 is transcribed in a cluster with miR-3960, from a locus at 9q34.11. BMP2 was shown to induce expression of the miR-3960~2861 cluster and both miRNAs are positive regulators of osteoblast differentiation [ 102 ]. Indeed, systemic administration of a miR-2861 antagomiR decreased bone mass, trabecular bone volume and bone formation rate in mice [ 103 ]. miR-2861 was shown to target histone deacetylase 5 (HDAC5) which is known to deacetylate RUNX2, promoting its ubiquitin mediated degradation. Thus, miR-2861 targeting of Hdac5 increases RUNX2 acetylation and protein stability, providing a positive feedback loop contributing to BMP2-induced osteoblast differentiation and function [ 104 ] ( Figure 6 ). Although SNPs in this miRNA have not been reported, a mutation in pre-miR-2861 was identified in one kindred with low bone mass. Functional analyses revealed that the mutation in this osteopenic kindred mapped to the leading strand in the stem of pre-miR-2861, which caused decreased mature miRNA levels due to reduced miRNA processing. High levels of the miR-2861 target Hdac5 were attributed to decreased RUNX2 protein in the bones of these patients. Although expansion of the linkage study to include unrelated osteoporotic patients and healthy controls did not reveal the presence of pre-miR-2861 mutations in either group, this study highlights a functional miR-SNP variant with a profound impact on bone mass [ 103 ].

Polymorphisms in miRNA target sites within a mRNA (miR-TS-SNPs) can either abrogate an existing miRNA binding site or create a novel site ( Figure 7 ). Compared with SNPs in miRNA genes, miR-TS-SNPs in 3’ UTRs are more abundant [ 8 , 9 ]. Studies in the field of skeletal diseases such as osteoporosis and chondrodysplasia, as well as cancer genetics have identified the function of miR-TS-SNPs in several target genes with known functions in bone. BMPs are bone anabolic factors critical for development [ 105 ]. They signal through their membrane bound receptors, BMPRI (I-A and I-B) and BMPRII, which form a homo- or heterotrimeric complex upon ligand binding, and activate a canonical SMAD mediated transcriptional program [ 106 ]. Both type I BMP receptors have distinct spatial and temporal expression patterns during embryogenesis [ 107 - 109 ]. In osteoblasts, BMPRI-B is crucial for commitment of mesenchymal precursors towards the osteogenic lineage, while BMPRI-A appears to direct them towards an adipocytic fate [ 110 , 111 ]. Conditional deletion of BMPRI-B in osteoblasts leads to decreased bone formation and growth [ 112 ]. In osteoclasts, inhibition of BMP signaling impairs osteoclastogenesis and bone resorption [ 113 ]. Overall, BMPs directly regulate both arms of the bone remodeling cycle. One case-control study linked SNP rs1434536 (C>T, MAF 0.404 in 1000 Genomes Project) in the BMPRI-B 3’ UTR with increased risk of estrogen receptor-positive (ER+) breast cancer [ 114 ]. Previously, increased BMPRI-B mediated signaling in ER+ breast carcinoma was associated with high tumor grade, greater tumor cell proliferation, cytogenetic instability and poor prognosis [ 115 ]. Interestingly, SNP rs1434536 showed strong linkage with cancer associated SNPs from the CGEMS (Cancer Genetic Markers of Susceptibility) GWAS data set [ 114 , 116 ]. Patients carrying the high risk rs1434536-T allele had increased BMPRI-B expression, and this allele demonstrated impaired binding of miR-125b to the BMPRI-B 3’ UTR [ 114 ]. A second study demonstrated a similar mechanism for the rs1434536-T variant and higher prostate cancer risk [ 117 ]. Remarkably, the rs1434536-T allele was shown to impart BMPR-IB mediated protection against endometriosis, due to diminished miR-125b directed repression [ 118 ]. SNP rs1434536-T is a common variant, and increased BMPRI-B levels in individuals carrying this allele might contribute to higher bone mass.

Presently, there are only handful studies implicating miR-TS-SNPs in skeletal disease. These miR-TS-SNPs reside in the 3’ UTR of genes that play an important role in osteoblast or chondrocyte function, and are discussed below. Fibroblast growth factor 2 (FGF) is a potent mitogenic growth factor that is widely expressed in limb bud mesenchyme, chondrocytes, osteoblasts, osteoclasts and adipocytes. In osteoblasts, FGF2 promotes osteoblast differentiation by mechanisms that include regulation of BMP2 and ATF4 [ 119 - 121 ]. However, sustained FGF2 signaling can inhibit bone formation by maintaining osteoblastic precursors in a proliferative state, thereby preventing their differentiation [ 119 , 120 , 122 ]. Thus, in the osteoblast lineage, appropriate timing and levels of FGF2 signaling are critical for optimal differentiation. In the osteoclast lineage, FGF2 promotes recruitment of osteoclast precursors and their differentiation [ 120 , 121 , 123 ]. FGF2-null mice display osteopenia and decreased bone remodeling [ 124 , 125 ]. Recently, a miR-TS-SNP, rs6854081 (T>G, MAF 0.0855 in 1000 Genomes Project) located in the 3’ UTR of Fgf2 was associated with BMD in a cohort of middle-aged Caucasian women. The minor allele, rs6854081-G, was associated with low BMD, and a subtle increase in FGF2 mRNA was found in monocytes or B cells isolated from the low BMD patients [ 126 ]. Using the poly-miRTS database, SNP rs6854081-G was shown to potentially interfere with miR-146a or miR-146b binding [ 126 ]. It was speculated that rs6854081-G might relieve translational suppression of Fgf2, leading to increased FGF2 levels and enhanced osteoclastogenesis. In this study, Fgf2 was not experimentally validated as a target of miR-146a and b, and it is possible that other miRNAs could be involved. However, the study did report a Fgf2 3’ UTR SNP and provided an implication for the role of miR-TS-SNPs in mediating variations in BMD [ 126 ]. Osteonectin is one of the most abundant non-collagenous extracellular matrix components in bone. It facilitates collagen fibril assembly and organization, and promotes osteoblast commitment, differentiation, and survival [ 127 - 129 ]. Global deficiency of osteonectin leads to development of low turnover osteopenia, with decreased bone formation [ 130 ]. Moreover, osteonectin-null and haploinsufficient mice display an attenuated bone-anabolic response to intermittent PTH treatment [ 131 ]. A candidate gene study associated a miR-TS-SNP rs1054204 (C>G, MAF 0.387 in 1000 Genomes Project) in the 3’ UTR of osteonectin with low bone mass in a cohort of Caucasian men with idiopathic osteoporosis [ 132 , 133 ]. A subsequent study demonstrated a potential mechanism underlying this association, using in vitro and in vivo models. Specifically, a novel set of mouse knock-in models were created, in which the mouse osteonectin 3’ UTR was replaced with the human 3’ UTR, containing either rs1054204-C or-G. Mice carrying the minor allelic variant rs1054204-G had decreased levels of osteonectin in bone tissue, decreased trabecular bone volume, and a reduced bone-anabolic response to intermittent PTH (1-34) treatment [ 132 , 133 ]. Moreover, in vitro studies demonstrated a cell-autonomous effect of the variant on osteoblastic differentiation and mineralized matrix production. Additional in vitro studies showed that the minor allele variant rs1054204-G enhances osteonectin post-transcriptional regulation by creating a novel miR-433 target site in its 3’ UTR [ 133 ]. Other studies demonstrated that miR-433 could inhibit osteoblast differentiation and target Runx2 [ 133 , 134 ]. Overall, these data provided a physiological function to a common miR-TS-SNP, and identified a potential mechanism [ 133 ]. Histone deacetylase 6 (HDAC6) functions as a negative regulator of osteoblast differentiation and a positive regulator of adipogenesis [ 135 ]. In osteoblasts, HDAC6 interacts with RUNX2 and promotes its ability to repress transcription of target genes [ 136 ]. In chondrocytes, HDAC6 may impact differentiation due to its ability to regulate RUNX2 activity. Moreover, since HDAC6 can deacetylate tubulin, it can destabilize ciliary microtubules. In articular chondrocytes, primary cilia responsive to mechanical strain regulate hedgehog signaling, providing an additional potential mechanism for HDAC6 regulation of chondrogenesis [ 137 ]. Although SNPs in Hdac6 have not yet been linked to skeletal disease, a mutation in the 3’ UTR of this gene was suggested to be a likely molecular cause for disease in a kindred presenting with a new form of dominant X-linked chondrodysplasia characterized by platyspondyly, rhizomelic shortening of the members, specific brachydactyly, hydrocephaly, facial dysmorphism and microphtalmia. A linkage study mapped the disease locus to a pericentromeric region within Xp11.3–q13.1 (Lod score = 3.30). The HDAC6 gene was hypothesized to be a candidate gene in this locus, and exon sequencing revealed a variant (c.*281A>T) in the Hdac6 3’ UTR [ 138 ]. In this chondrodysplasia study, miR-433 was demonstrated to target Hdac6. The mutation (c.*281A>T) was localized to a miR-433 seed binding region in the 3’ UTR, and was shown to interfere with miR-433 targeting. Increased HDAC6 levels and activity were demonstrated in tissue from affected family members and in affected limbs of a female child displaying mosaicism, due to selective inactivation of the mutated allele. In the affected patients, upregulation of HDAC6 may indirectly regulate hedgehog signaling, among other pathways, and lead to phenotypic defects. Altogether, this study demonstrates that a Hdac6 3’ UTR variant suppressed miR-433 mediated post-transcriptional regulation, causing overexpression of HDAC6, which could contribute to X-linked chondrodysplasia [ 138 ].

Bone mass phenotype is a complex and highly heritable trait. Approximately 38 million SNPs have been discovered in the human genome, representing the majority of gene variants [ 1 ]. Linkage studies, in conjunction with candidate gene studies, have identified some rare genetic variants with high impact on bone phenotype, most notably variants of low density lipoprotein-related receptor 5 (LRP5). More recently, GWAS (genome wide association study) identified 62 genome-wide significant loci associated with bone mineral density (BMD) [ 2 , 3 ]. Some of these loci contain candidate genes with known function in pathways important in bone physiology, such as canonical Wnt signaling, the RANK-RANKL-OPG system, and endochondral ossification. However, only chromosomal regions have been determined, and the genes or mechanisms underlying the linkage with BMD are not yet identified. In general, the SNPs identified are common variants with small effect size. Presently, less than 6% of the variance in femoral neck BMD can be explained by the GWAS-identified loci. [ 4 ]. It may be relatively straightforward to determine whether SNPs in the protein-coding region of a gene or within intronic areas that affect splicing might cause changes in protein function or abundance, and thereby, phenotype. However, the functional impact of SNPs in untranslated regions (UTRs), promoter/enhancer regions or regulatory RNAs can be more difficult to evaluate. Remarkably, at conserved sites within the human genome, the largest proportion of variants lies within the UTRs [ 1 ]. Moreover, a study comparing GWAS-SNP sets found an enrichment of trait-associated variants in 3’ UTRs, compared with 5’ UTRs or coding sequences (CDS) [ 5 ]. The 3’ UTR is recognized as a critical control region, with the potential to regulate mRNA stability, translation and localization. Many binding sites for microRNAs (miRNAs, miRs), which are key post-transcriptional regulators, are found in 3’ UTRs [ 6 ].

A growing number of GWAS have associated genomic loci with BMD and fracture risk. Within these intervals, SNPs in protein coding regions as well as UTRs and miRNA genes should be considered for functional validation. To date, only a limited number of studies have identified SNPs or mutations in miRNA genes and target binding sites relevant to BMD. However, miR-associated SNPs (miR-P-SNPs, miR-SNPs and miR-TS-SNPs) validated in other disease models have potential to inform studies of SNPs and bone mass. In particular, data on functional miR-associated SNPs (collated from other diseases), along with information on miRNA-target co-expression and quantitative trait locus (QTL) mapping could provide direction for investigation of miR-SNPs with the potential for relevance in skeletal diseases. Moreover, newer sophisticated bioinformatic tools (i.e. Patrocles Database, PolymiRTS Database 3.0) can be used in conjunction with existing GWAS data to determine whether a polymorphism of interest might also be miR-TS-SNP. Given the polygenic nature of skeletal diseases like osteoporosis, the addition of genetic profiling data that includes miRNA-associated SNPs could enhance the accuracy of fracture prediction tools that currently rely only on BMD and clinical risk factors [ 139 , 140 ]. In addition to genetics, factors including age, lifestyle, hormonal status and co-morbidities can greatly influence BMD. Since a majority of SNP variants are likely to have a small impact on BMD, the evaluation of SNP function can be facilitated by the exclusion of possible confounding variables, such as differences in environment or genetic modifiers. This simplification can be achieved by creating novel in vitro and in vivo models to study SNP function. The increased usage of genome editing technology, such as the CrispR-Cas9 system, for the generation of new mouse models and cell lines will likely be instrumental in driving forward our understanding of SNP function. In this review, we summarized the mechanisms for miR-associated SNP regulation of miRNA function, and provided evidence suggesting of potential impact that variants of several miRNAs might have in regulating skeletal phenotype (summarized Table 1 , Figure 8 ). Lastly, we briefly described the current literature on known miR-associated SNPs or mutations in skeletal diseases. miRNA-based therapeutics are in clinical trials for selected malignancies, and the development of novel miRNA-based therapeutics for bone repair and regeneration is underway. Information on miRNA function in bone and cartilage, as well as information on genetic variants affecting miRNA function could help the clinicians develop a better plan for individualized treatment as well as predict the impact of drug efficacy, adverse reactions or toxicity along with disease risk.