Genetics of osteoporosis: accelerating pace in gene identification and validation
- First Online:
- Cite this article as:
- Li, W., Hou, S., Yu, B. et al. Hum Genet (2010) 127: 249. doi:10.1007/s00439-009-0773-z
- 603 Views
Osteoporosis is characterized by low bone mineral density and structural deterioration of bone tissue, leading to an increased risk of fractures. It is the most common metabolic bone disorder worldwide, affecting one in three women and one in eight men over the age of 50. In the past 15 years, a large number of genes have been reported as being associated with osteoporosis. However, only in the past 4 years we have witnessed an accelerated pace in identifying and validating osteoporosis susceptibility loci. This increase in pace is mostly due to large-scale association studies, meta-analyses, and genome-wide association studies of both single nucleotide polymorphisms and copy number variations. A comprehensive review of these developments revealed that, to date, at least 15 genes (VDR, ESR1, ESR2, LRP5, LRP4, SOST, GRP177, OPG, RANK, RANKL, COLIA1, SPP1, ITGA1, SP7, and SOX6) can be reasonably assigned as confirmed osteoporosis susceptibility genes, whereas, another >30 genes are promising candidate genes. Notably, confirmed and promising genes are clustered in three biological pathways, the estrogen endocrine pathway, the Wnt/β-catenin signaling pathway, and the RANKL/RANK/OPG pathway. New biological pathways will certainly emerge when more osteoporosis genes are identified and validated. These genetic findings may provide new routes toward improved therapeutic and preventive interventions of this complex disease.
Osteoporosis is characterized by low bone mineral density (BMD) and structural deterioration of bone tissue, leading to an increased risk of fractures that occur mostly at the hip, spine and wrist. It is the most common metabolic bone disorder worldwide and affects one in three women and one in eight men over the age of 50. The burden of osteoporosis on the healthcare system is extremely large. For example, in the year 2005, approximately 2 million osteoporotic fractures occurred in the USA, which created direct medical costs of US$17 billion. By the year 2025, annual fractures and costs are predicted to grow by 50% (Burge et al. 2007).
BMD, which is measured using dual energy X-ray absorptiometry, is currently the best predictor of osteoporotic fractures; therefore, it is often used as a surrogate phenotype for osteoporosis. Many factors including age, sex, diet, physical activity, medication use and menopausal status influence the risk of osteoporosis. However, one of the most important clinical risk factors is a positive family history, underscoring the importance of genetics in the etiology of the disorder. Heritability of BMD ranges from ~60 to ~90% in twins (Harris et al. 1998; Pocock et al. 1987; Slemenda et al. 1991) and ~45 to ~70% between parents and offspring (Duncan et al. 2003; Gueguen et al. 1995; Krall and Dawson-Hughes 1993). In the general population, BMD is a complex trait that is influenced by many genetic variants with modest effect size and their interactions with environmental factors (Rivadeneira et al. 2009).
In the past 15 years, a large number of genes have been reported to be associated with osteoporosis. In the past 4 years, we have witnessed an accelerating pace in identifying and validating osteoporosis susceptibility loci. This increase in pace is mostly due to large-scale candidate gene association studies (CGASs), meta-analyses, and genome-wide association studies (GWASs). This article reviews the most extensively studied genes during the past 15 years and new promising genes found in the past 4 years.
Summary of all osteoporosis-associated genes reviewed in this article
Osteoporosis susceptibility loci initially analyzed by candidate gene studies
The vitamin D endocrine pathway
Established gene (CGASa + MA-Cb)
Likely a modifier gene (CGAS)
The estrogen endocrine pathway
Established gene (CGAS + MA-C + GWASc + MA-Gd)
Established gene (CGAS)
Promising gene (GWAS)
The Wnt/β-catenin signaling pathway
Established gene (CGAS + MA-C + GWAS + MA-G)
Established gene (GWAS + MA-G + MA-C)
Established gene (CGAS + GWAS)
Promising gene (GWAS)
Promising gene (MA-G)
Promising gene (CGAS + GWAS)
Established gene (MA-G)
The RANKL/RANK/OPG pathway
Established gene (CGAS + MA-C + GWAS + MA-G)
Established gene (CGAS + GWAS + MA-G)
Established gene (CGAS + GWAS + MA-G)
The transforming growth factor-β (TGFB) superfamily
Not associated with the disease
Selected examples of other ‘older’ genes
Established gene (CGAS + MA-C)
Established gene (CGAS + MA-C)
Candidate genes identified inthe past 4 years
DMP1, FLT1, HOXA, IGFBP2, NFATC1, and PTN
Established gene (CGAS + MA-C)
Quantitative trait loci mapping in inbred mouse strains
Likely an osteoporosis susceptibility gene
Linkage analysis in humans
RERE, G1P2, SSU72, and CCDC27
GWASs of SNPs
Novel BMD loci revealed by original GWASs
Established gene (GWAS + MA-G)
Established gene (GWAS + MA-G)
Promising gene (GWAS + MA-G)
Promising gene (GWAS + MA-G)
The MHC region
Putative region (GWAS + MA-G)
New BMD loci revealed by the meta-analysis of five GWASs
MEF2C, STARD3NL, FLJ42280, DCDC5 or DCDC1, CRHR1, MEPE, and HDAC5 or C17orf53
New loci for other osteoporosis-related phenotypes
GWASs of copy number variations
A combination of gene expression profiling and GWAS
21 genes in the EphrinA-EphR pathway
Osteoporosis susceptibility loci initially analyzed by candidate gene studies
Most of the currently established or putative osteoporosis susceptibility loci were first analyzed by the approach of CGASs. A significant fraction of these genes can be classified into well-defined biological pathways. The remaining genes will be addressed in the context of representative studies. Validating findings from GWASs will also be discussed when applicable.
The vitamin D endocrine pathway
The vitamin D endocrine system is pleiotropic and plays an important role in bone metabolism. The effect of vitamin D is mediated through the vitamin D receptor (VDR), a nuclear transcription factor that regulates gene expression by interacting with vitamin D response elements in target genes (Christakos et al. 2003). Missense mutations in the hormone binding domain of VDR cause hereditary vitamin D-resistant rickets (Kristjansson et al. 1993).
Vitamin D receptor (VDR) was the first studied candidate gene with respect to osteoporosis (Morrison et al. 1994). As noted by Ferrari (2008), “VDR association with BMD has been highly controversial, as there are probably as many positive as negative studies”. Here, we focus on several representative studies. A large-scale analysis of 15 haplotype-tagging SNPs in 6,148 elderly whites showed that certain haplotypes in the promoter region and 3′-untranslated region (3′-UTR) were strongly associated with increased fracture risk. Specifically, the associations were independent of BMD, suggesting the existence of other fracture risk-determining mechanisms (Fang et al. 2005). In vitro functional analysis indicated that a promoter risk haplotype was associated with reduced reporter gene expression and a 3′-UTR risk haplotype was associated with increased mRNA decay. In short, these fracture risk alleles resulted in lower VDR mRNA levels (Fang et al. 2005). Subsequent meta-analysis, which examined the Cdx2 (located in the promoter), FokI, BsmI, ApaI, and TaqI polymorphisms in 26,242 participants involving nine European research teams, partially confirmed the above finding. First, none of the studied polymorphisms were associated with BMD. Second, the Cdx2 A-allele was associated with a 9% (P = 0.039) risk reduction for vertebral fractures (Uitterlinden et al. 2006). In vitro functional analysis demonstrated that the Cdx2 A-allele had markedly higher binding affinity for the transcription factor, Cdx2, and significantly higher transcriptional activity compared with the Cdx2-G allele (Arai et al. 2001).
The FokI polymorphism refers to a common C to T polymorphism in exon 2 of the VDR gene (rs10735810). It introduces a new translation start site, resulting in a protein with three additional amino acids (C = 424 aa, T = 427 aa). It is most likely a functional polymorphism due to three lines of evidence. First, the C allele displays higher transactivation activity than the T allele, as demonstrated in reporter constructs under the control of a vitamin D response element in various cell lines (Arai et al. 1997; Whitfield et al. 2001). Second, an engineered shorter protein interacted more efficiently with the basal transcription factor, IIB, and possessed elevated transcriptional activity compared with its longer counterpart in transfected COS-7 monkey kidney epithelial cells (Jurutka et al. 2000). Lastly, the C genotype is more responsive to 1,25-dihydroxyvitamin D3, the principal bioactive form of vitamin D, in cultured human peripheral blood mononuclear cells (Colin et al. 2000). If these results are considered in the context of the findings of Arai et al. (2001), Fang et al. (2005), and Uitterlinden et al. (2006), a protective effect of the C genotype against osteoporosis is implied. However, an analysis of the FokI polymorphism in 6,698 American women aged 65 years or older found an opposing effect; women with the C/C genotype were associated with significantly lower BMD at the distal radius, a modest but significant increase in the risk for non-spine, low traumatic fractures, and a 33% increased risk of wrist fracture as compared with those with the T/T genotype (Moffett et al. 2007).
A new development in the field is the association of VDR genotypes with falls. Onder et al. (2008) analyzed the FokI and BsmI genotypes in 259 subjects aged over 80 years with respect to falls occurring within 90 days of assessment. The bb genotype at BsmI was associated with a reduced rate of falls compared with the BB genotype. By contrast, no effect on falls was observed for the FokI polymorphism (Onder et al. 2008). This finding was supported by a very recent study that involved two larger population cohorts (Barr et al. 2009) and considered to account for some of the VDR-associated fracture risk.
Finally, it is important to note that evidence was observed that supports the association between a SNP (rs2189480) in the VDR gene and femoral neck section modulus and spine BMD in the GWAS of Kiel et al. (2007a), although the association did not achieve genome-wide significance. In addition, SNPs in and around the VDR gene were among the top 1,000 SNPs in the GWAS of Styrkarsdottir et al. (2008).
Vitamin D binding protein (DBP) has two functions in bone metabolism. First, it plays an important role in the maintenance of calcium homeostasis by binding to and transporting vitamin D to target tissues. Second, DBP can be converted to DBP-macrophage activating factor, which mediates bone resorption by directly activating osteoclasts (Fang et al. 2009). Several studies reported association between DBP polymorphisms and BMD or fracture (Al-oanzi et al. 2008; Ezura et al. 2003; Lauridsen et al. 2004; Taes et al. 2006), but only the study of Xiong et al. (2006b) analyzed >1,000 (n = 1,873) subjects. More recently, Fang et al. (2009) genotyped two DBP polymorphisms in 6,181 elderly Caucasians and investigated the interactions of these genotypes with the VDR genotype and dietary calcium intake with respect to fracture risk. The modest effect of the DBP gene on fracture risk became apparent only in the presence of other genetic and environmental factors.
The estrogen endocrine pathway
The estrogen endocrine system has long been known to play an important role in regulating bone mass and the occurrence of osteoporosis. For example, estrogen replacement therapy in postmenopausal women prevents bone loss (Felson et al. 1993) and decreases the risk of osteoporotic fractures (Cauley et al. 1995). Moreover, estrogens bind to and activate estrogen receptors resulting in the up-regulation of the expression of many genes. A 28-year-old man carrying a homozygous inactivating mutation in the estrogen receptor 1 gene (ESR1) had low BMD (Smith et al. 1994). Consistently, Esr1−/− mice show decreased bone mass (Korach 1994).
ESR1 is one of the most extensively studied candidate genes for osteoporosis, exemplified by the publication of three meta-analytic studies. The first study evaluated the XbaI (rs9340799) and PvuII (rs2234693) polymorphisms in intron 1 of 5,834 women from 30 study groups: “XX” homozygotes had higher BMD and also a decreased risk of fractures than carriers of the “x” allele, but the PvuII polymorphism had no effect on either BMD or fracture risk (Ioannidis et al. 2002). The second study analyzed these two polymorphisms together with the promoter TA repeat polymorphism in 18,917 subjects in eight European centers. The only association found was between the Xbal polymorphism and a decreased fracture risk (Ioannidis et al. 2004). The lack of association with BMD in this study was regarded as having potential clinical relevance because it provides information on fracture risk that cannot be obtained by BMD measurements. In this regard, it is pertinent to mention that a ESR1 haplotype was associated with bone quality (Albagha et al. 2005). The third meta-analysis collected data from 4,297 Chinese women reported in 16 eligible studies with respect to the XbaI and PvuII polymorphisms; only a very weak association was found between the PvuII polymorphism and femoral neck BMD (Wang et al. 2007a). Furthermore, the Xbal and PvuII polymorphisms were shown to influence reporter gene expression in vitro (Herrington et al. 2002; Maruyama et al. 2000).
Recently, a GWAS found that a novel SNP (rs1999805), which is located in an intron of the U68068 splice variant of ESR1 and not in LD with the Xbal and PvuII polymorphisms, was associated with BMD (P = 3.8 × 10−7) (Styrkarsdottir et al. 2008). This association was confirmed by a large-scale meta-analysis of five GWASs (Rivadeneira et al. 2009). A notable finding from these two studies suggest that more than one associated signal exist in the ESR1 region. Consistent with this finding, new SNPs in ESR1 have also been reported by CGASs. Wang et al. (2008) genotyped 25 SNPs in ESR1 in 700 elderly Chinese subjects and identified an association between two novel SNPs (rs3020314 and rs1884051) and hip fracture. Moreover, Lai et al. (2008) genotyped a newly described intronic dinucleotide CA repeat polymorphism of ESR1 in 452 pre-, 110 peri- and 622 postmenopausal southern Chinese women; the polymorphism was associated with BMD variation, rate of bone loss and fracture risk in post- but not premenopausal Chinese women.
ESR2 is thought to be less important in mediating estrogen action in bone tissue than ESR1. To date, 15 studies analyzed polymorphisms in the ESR2 gene with respect to BMD or/and fracture risk, all yielding positive results (Geng et al. 2007; Greendale et al. 2006; Ichikawa et al. 2005; Kung et al. 2006; Lau et al. 2002, 2005, 2006; Massart et al. 2009; Moron et al. 2006; Ogawa et al. 2000; Rivadeneira et al. 2006; Scariano et al. 2004; Shearman et al. 2004; Sowers et al. 2006; Wang et al. 2008). Of these studies, examination of six SNPs in the ESR2 gene of 6,343 elderly white individuals demonstrated that variants of ESR2 influence the risk of fracture in postmenopausal women both on their own and by interacting with ESR1 and IGF1 (Rivadeneira et al. 2006); analysis of ESR2 rs4986938 in 641 healthy premenopausal women, aged 20–50 years, showed that the studied polymorphism had an age-specific effect on various skeletal traits (Massart et al. 2009).
ESRRA and ESRRG
The estrogen-related receptors―ERR-α (ESRRA), ERR-β (ESRRB), and ERR-γ (ESRRG)―are a subfamily of orphan nuclear receptors closely related to the estrogen receptor family. They share target genes, coregulatory proteins, ligands and sites of action with the estrogen receptors, show overlap with ESR1 expression, and can actively influence the estrogenic response (Giguere 2002).
In vitro functional studies demonstrated that a higher number of a 23-nucleotide element in the ESRRA promoter is associated with increased gene expression (Laganiere et al. 2004). This functional repeat polymorphism was associated with BMD in a population of premenopausal women from Quebec (Laflamme et al. 2005) but not replicated in a population of premenopausal women from Toronto as shown by the Rousseau group (Giroux et al. 2008). More recently, the Rousseau group analyzed ESRRG as a candidate gene for osteoporosis and found a consistent association of SNPs in the ESRRG region with multiple bone measures in both a sample of 5,144 Quebec women and a sample of 673 Toronto women (Elfassihi et al. 2009).
CYP19A1 encodes a cytochrome P450 enzyme known as aromatase, whose function is to aromatize androgens to estrogens. Its role in regulating bone mass is reflected by (1) significant BMD loss associated with the use of anastrozole, an aromatase inhibitor, in treating estrogen-dependent breast cancer (Eastell et al. 2006); (2) osteoporosis presents in patients with loss-of-function mutations in the CYP19A1 gene (Carani et al. 1997; Morishima et al. 1995); and (3) increased bone mass resulting from estrogen therapy in patients with aromatase deficiency (Bilezikian et al. 1998; Herrmann et al. 2002).
Association between CYP19A1 polymorphisms and BMD or/and osteoporotic fractures has been suggested by a dozen of studies (Dick et al. 2005; Enjuanes et al. 2006; Gennari et al. 2004; Masi et al. 2001; Mendoza et al. 2006; Riancho et al. 2005, 2006, 2007; Somner et al. 2004; Valero et al. 2008; Xiong et al. 2006b; Zarrabeitia et al. 2004). A more recent study analyzed a larger sample comprising 1,163 postmenopausal women and found that three CYP19A1 polymorphisms were associated with BMD (Riancho et al. 2009). The association was age-dependent; statistical significance was observed in the subgroup of subjects aged 67 years or older but not in the subgroup of younger subjects; although the association was in the same direction in the latter subgroup. As noted by Riancho et al. (2009), this finding may account for the lack of association in some studies that analyzed younger women (Moron et al. 2006; Salmen et al. 2003; Tofteng et al. 2004b). In vitro functional analyses suggested that rs1062033, which is located ~12 kb upstream of the translation initiation site, represents a true regulatory polymorphism (Riancho et al. 2009). Finally, significant associations were observed between multiple SNPs in a LD block within CYP19A1 with reduced ultrasound BMD and bone strength in middle-aged and elderly men (Limer et al. 2009).
CYP17A1 plays an important role in the synthesis of androgens and estrogens. Loss-of-function CYP17A1 mutations cause reduced skeletal growth and diffuse osteoporosis (Yanase et al. 1991). The 5′-UTR of CYP17A1 harbors a common T>C polymorphism (rs743572) located 34 bp upstream of the translation initiation site (Carey et al. 1994). The hypothesis that this SNP may affect gene expression by creating a Sp1 binding site (Carey et al. 1994) was not supported by in vitro studies (Nedelcheva Kristensen et al. 1999). Several studies have analyzed whether this SNP is associated with osteoporosis, but they have yielded inconsistent results (Chen et al. 2005; Limer et al. 2009; Somner et al. 2004; Tofteng et al. 2004a; Valero et al. 2005; Yamada et al. 2005b; Zarrabeitia et al. 2007; Zmuda et al. 2001). Here, we focus on the three studies that analyzed >1,000 subjects. In the study of 1,795 Danish perimenopausal women, the CC genotype was associated with lower femoral neck and lumbar spine BMD in lean individuals but not in overweight ones (Tofteng et al. 2004a). In the study of 1,108 postmenopausal and premenopausal Japanese women, the CC genotype was associated with increased femoral neck BMD (Yamada et al. 2005b). In the study of 2,693 men aged 40–79 years, the CC genotype was associated with a significantly lower ultrasound BMD as compared with AA (Limer et al. 2009). In addition, a study of 1,873 subjects from 405 white nuclear families showed a significant association between a haplotype of CYP17A1 and BMD/osteoporotic fracture at the spine and hip (Xiong et al. 2006b).
The Wnt/β-catenin signaling pathway
The canonical Wnt/β-catenin signaling pathway plays an important role in the regulation of bone mass. The pathway is activated when Wnt binds to the frizzled (FZD) family of receptors and to a low-density lipoprotein receptor-related protein 5 (LRP5) or LRP6 co-receptor, leading to the stabilization of β-catenin. The stabilized β-catenin is translocated to the nucleus, where it binds to lymphoid enhancer-binding factor (LEF) and T cell factor (TCF) proteins, ultimately, affecting target gene transcription (Krishnan et al. 2006).
LRP5 and LRP6
The key role of LRP5 in regulating bone mass was first identified from the study of human rare monogenic skeletal diseases. Loss-of-function LRP5 mutations cause autosomal recessive osteoporosis-pseudoglioma syndrome, which is characterized by severe osteoporosis and blindness (Gong et al. 2001). By contrast, an activation point mutation, Gly171Val, causes autosomal dominant high bone mass syndrome (Boyden et al. 2002; Little et al. 2002). In vitro studies demonstrated that Gly171Val resulted in the dysregulation of Dkk1- (Bhat et al. 2007; Boyden et al. 2002; Murrills et al. 2009) and SOST-mediated (Semenov and He 2006) inhibition of Wnt signaling. Other LRP5 missense mutations have been described in different conditions with an increased bone density, including endosteal hyperostosis, Van Buchem disease, autosomal dominant osteosclerosis, and osteopetrosis type I (Van Wesenbeeck et al. 2003). Lrp5−/− mice exhibited low bone density (Kato et al. 2002), whereas transgenic mice expressing the mutant G171V had increased bone mass (Babij et al. 2003).
The aforementioned findings aroused great interest in the potential role of the LRP5 gene in regulating osteoporosis-related traits in the general population. During 2004–2007, numerous studies found an association between common LRP5 SNPs and BMD at the population level (Bollerslev et al. 2005; Brixen et al. 2007; Ezura et al. 2007; Ferrari et al. 2004; Giroux et al. 2007; Kiel et al. 2007b; Koay et al. 2004, 2007; Koh et al. 2004; Koller et al. 2005; Lau et al. 2005, 2006; Mizuguchi et al. 2004; Saarinen et al. 2007; Urano et al. 2004; van Meurs et al. 2006; Xiong et al. 2006b, 2007; Zhang et al. 2005b). Two of these studies reported that common LRP5 SNPs also determine fracture risk in elderly women (Bollerslev et al. 2005; van Meurs et al. 2006). However, these findings were regarded as inconclusive due to limited sample size, variations in the studied polymorphisms and examined phenotypes, and different analytical approaches used in these studies (van Meurs et al. 2008). To overcome these drawbacks, the GENOMOS investigators genotyped the two most frequently studied non-synonymous SNPs, Val667Met and Ala1330Val, in 37,534 individuals from 18 participating teams in Europe and North America (van Meurs et al. 2008). They demonstrated that both SNPs were associated with reduced lumbar spine and femoral neck BMD and increased vertebral fracture risk. The magnitude of the effect was modest, which was consistently found in different populations and was independent of age and sex (van Meurs et al. 2008). Association of LRP5 polymorphisms with BMD was further reported in more recent studies (Agueda et al. 2008; Cheung et al. 2008b; Giroux et al. 2008; Grundberg et al. 2008; Sims et al. 2008; Urano et al. 2009a; Zhang et al. 2009b). Additionally, a Bayesian meta-analysis of ten eligible studies comprising 16,705 individuals revealed a modest effect of the A1330V polymorphism on BMD in the general population (Tran et al. 2008). Finally and most importantly, the association between LRP5 and BMD was confirmed by the GWAS of Richards et al. (2008) and the meta-analysis of five GWASs (Rivadeneira et al. 2009).
Based on the pathophysiological role of LRP5 in bone biology, Val667Met and Ala1330Val should, in principle, result in decreased Wnt signaling. Val667Met resides at the top of the third propeller module in the extracellular domain of the receptor, a domain that is thought to interact with the Wnt-inhibitor Dkk1. Therefore, it is possible that the Val667Met mutant may have a higher binding affinity for Dkk1 as compared with the wild-type protein (van Meurs et al. 2008). Ala1330Val is located within the second low-density lipoprotein (LDL) domain. Although the exact molecular mechanism remains to be clarified, the Ala1330Val mutant was shown to have a significantly reduced Wnt-signaling capacity as compared with the wild-type molecule in two independent TCF-Lef reporter assays (Kiel et al. 2007b; Urano et al. 2009a).
A spontaneous LRP6 missense mutation in mice results in multiple Wnt-deficient phenotypes, including dysmorphologies of the axial skeleton, digits, and the neural tube (Kokubu et al. 2004). In another mouse study, Lrp6 was found to interact with Lrp5 in limb development. Furthermore, heterozygosity for a Lrp6 loss-of-function mutation exacerbates the low BMD phenotype of Lrp5−/− mice (Holmen et al. 2004). In humans, a loss-of-function missense mutation in the LRP6 gene was identified in a family with early coronary disease and severe osteoporosis (Mani et al. 2007).
A common non-synonymous SNP in LRP6 (Ile1062Val) that exhibited decreased β-catenin signaling in HEK293T cells (De Ferrari et al. 2007) was reported to be associated with increased fracture risk in elderly men (van Meurs et al. 2006). In the study of 174 individuals with low BMD and 170 individuals with high BMD, two SNPs in the LRP6 gene (both are in strong LD with the common Ile1062Val polymorphism) were found to be associated with BMD (Sims et al. 2008). However, the large-scale analysis performed by the GENOMOS consortium did not find an association between the LRP6 Ile1062Val polymorphism with either BMD or fracture risk (van Meurs et al. 2008).
SOST encodes sclerostin, a protein expressed exclusively by osteocytes in human bone (van Bezooijen et al. 2004). Although sclerostin shares structural similarity with the DNA family of bone morphogenetic protein (BMP) antagonists, it does not function as a classical BMP antagonist as initially suggested (van Bezooijen et al. 2004, 2007). Instead, sclerostin functions as an antagonist of the canonical Wnt-signaling pathway by binding to LRP5/6 (Ellies et al. 2006; Li et al. 2005; Semenov et al. 2005). This interaction was impaired by LRP5 missense mutations associated with high bone mass in vitro (Balemans et al. 2008; Semenov and He 2006).
SOST was first identified as a key negative regulator of bone mass in a study of rare monogenic diseases. Loss-of-function SOST mutations cause autosomal recessive sclerosteosis, a disease characterized by progressive skeletal overgrowth (Balemans et al. 2001; Brunkow et al. 2001). A homozygous 52-kb deletion, which is located 32-kb downstream of the SOST gene, causes a similar disorder known as van Buchem disease (Balemans et al. 2002b; Staehling-Hampton et al. 2002). The deleted region contains a SOST-specific regulatory element (Loots et al. 2005). Transgenic mice expressing SOST exhibit low bone mass and decreased bone strength (Winkler et al. 2003), whereas Sost−/− mice demonstrate a high bone mass phenotype (Li et al. 2008).
The first attempt to investigate the association between polymorphisms in the SOST gene with BMD yielded negative results (Balemans et al. 2002a) due to both the limited sample size (n = 619) and the relative young age of the participants. Associations between polymorphisms in the SOST promoter and the region deleted in van Buchem disease and BMD were first demonstrated in the Rotterdam study, which analyzed 1,939 elderly whites (Uitterlinden et al. 2004). This finding was supported by more recent studies (Huang et al. 2009a; Sims et al. 2008; Styrkarsdottir et al. 2009; Yerges et al. 2009a). In particular, two SNPs located within the aforementioned van Buchem disease-causing 52-kb deletion (Balemans et al. 2002b; Staehling-Hampton et al. 2002), were reported to be associated with BMD in the GWAS of Styrkarsdottir et al. (2009).
LRP5/6 coreceptor activity is also inhibited by members of the Dickkopf (DKK) family such as DKK1, -2 and -4 (Krishnan et al. 2006). Sims et al. (2008) analyzed multiple Wnt pathway genes including DKK1 and DKK2 in 174 subjects with low BMD and 170 subjects with high BMD and found nominal association between DKK2 and BMD.
Frizzled receptor genes
There are ten known frizzled (FZD) members in humans, and they function as receptors for some or all of the Wnt family members (Angers and Moon 2009). Kim et al. (2009a) genotyped the non-synonymous SNPs (registered in the SNP database) in FZD1 (one SNP), FZD5 (1 SNP), FZD6 (3 SNPs), FZD7 (1 SNP), and FZD9 (4 SNPs) in 371 postmenopausal Korean women. Only two SNPs (both in FZD6) were found in this population; neither of them was associated with BMD at the lumbar spine and femoral neck.
Yerges et al. (2009b) selected FZD1 as a promising candidate gene based upon two observations. First, FZD1 modulates the canonical Wnt-signaling pathway as shown experimentally (Zilberberg et al. 2004). Second, FZD1 is expressed in osteoblast-like cells (Su et al. 2002). Yerges et al. (2009b) sequenced a 6.8 kb region surrounding FZD1 in 48 samples of African ancestry. They then genotyped the three SNPs that had a minor allele frequency of ≥5% in 1,084 men from the Tobago Bone Health Study. A SNP in the promoter region, rs2232158, was significantly associated with lower femoral neck BMD even after correction for multiple testing. The minor C allele in rs2232158 was predicted to generate a new binding site for the Egr1 transcription factor, which was confirmed by electrophoretic mobility shift assay, supershift, and ELISA. The minor C allele was further shown to possess a higher promoter activity than the major G allele in MG63 and SaOS-2 cells (Yerges et al. 2009b).
Secreted frizzled-related protein genes
The secreted frizzled-related protein (SFRP) family has an amino-terminal cysteine-rich domain that shares high homology with the Wnt-binding domain of FZD receptors. Therefore, SFRPs can compete with FZDs to bind to Wnts, which inhibits Wnt signaling (Cho et al. 2008). Sims et al. (2008) analyzed multiple Wnt pathway genes including SFRP1 and SFRP2 in 174 subjects with low BMD and 170 subjects with high BMD; two SNPs in a single haplotype block in the 3′-UTR of SFRP1 were significantly associated with BMD and bone mineral content (BMC). Sfrp1−/− mice exhibit increased trabecular BMD (Bodine et al. 2004).
The WNT family is comprised of 19 members (Angers and Moon 2009). Wnt10b transgenic mice have elevated bone mass and Wnt10b−/− mice have decreased trabecular bone and serum osteocalcin (Bennett et al. 2005); therefore, Zmuda et al. (2009) analyzed the potential involvement of the WNT10B gene in the pathogenesis of osteoporosis. Deep sequencing of the WNT10B gene in 192 individuals (96 African, 96 white) identified 19 SNPs with minor allele frequency ≥0.01. Initial genotyping of seven tagging SNPs and a potentially functional synonymous SNP in exon 5 (rs1051886; functionality inferred from RNA secondary structure prediction) in 1,035 Afro-Caribbean men showed that three SNPs including rs1051886 were associated with hip BMD. rs1051886 and rs3741627 (a 3′ UTR SNP) were replicated in an additional population-based sample of 980 men and a sample of 416 individuals belonging to eight large, multigenerational families, both of African ancestry. The minor allele at rs3741627 was predicted to disrupt a microRNA target site (Zmuda et al. 2009).
Sims et al. (2008) included WNT3A, WNT7B, and WNT10B for analysis in their study using individuals with low (n = 174) or high (n = 170) BMD; a nominal association was found between WNT3A and BMD.
Adenomatous polyposis coli (APC) is a component of the β-catenin degradation protein complex (Angers and Moon 2009). Both APC and β-catenin are expressed in human bone and cartilage (Monaghan et al. 2001). Mice lacking the Apc gene in osteoblasts develop bone in which the marrow component is almost completely absent (Holmen et al. 2005). Yerges et al. analyzed 383 candidate genes, including APC, for association with volumetric BMD (vBMD) at the femoral neck and lumbar spine among 862 white older men. Two SNPs, rs4705573 in the 5′ flanking region and rs6594646 in intron 1 of APC, are associated with lower vBMD at both skeletal sites. A third SNP, rs459552, which is associated with volumetric BMD only at the femoral neck, was presumed to result in a missense change located in the β-catenin down regulation domain of APC (Yerges et al. 2009a).
Forkhead box C2 (FOXC2) is a member of the family of winged helix/forkhead transcription factors. Yamada et al. (2006) first analyzed this gene with respect to osteoporosis due to (1) a close relation between lipid metabolism and bone remodeling, and (2) FOXC2 is a key regulator of adipocyte metabolism (Cederberg et al. 2001). More recent data indicated that FOXC2 also stimulates osteoblast differentiation of mesenchymal cells and preosteoblasts through activation of canonical Wnt–β-catenin signals (Kim et al. 2009b).
Yamada et al. (2006) analyzed the −512C>T polymorphism of FOXC2 in 1,129 men and 1,114 women from Japan and found a significant association with reduced BMD in both sexes. In the study of Yerges et al. (2009a), a SNP (rs3751797) in the 5′ flanking region of FOXC2 was significantly associated with lumbar spine vBMD. In the meta-analysis of 5 GWASs (Rivadeneira et al. 2009), rs10048416, which is located 95 kb downstream from the FOX gene cluster on 16q24.3, was associated with lumbar spine BMD at the genome-wide significance level.
The RANKL/RANK/OPG pathway
The receptor activator of the nuclear factor-κB ligand (RANKL)/receptor activator of the nuclear factor-κB (RANK) signaling regulates the formation, activation and survival of multinucleated osteoclasts in normal bone remodeling. Osteoprotegerin (OPG), a decoy receptor for RANKL, protects the skeleton from excessive bone resorption through binding to RANKL and prevents it from binding to RANK (Boyce and Xing 2007). RANKL, RANK, and OPG belong to the tumor necrosis factor superfamily and are encoded by TNFRSF11, TNFRSF11A, and TNFRSF11B, respectively.
Three lines of evidence demonstrate: (1) mice overexpressing OPG develop osteopetrosis (Simonet et al. 1997), (2) OPG-deficient mice develop severe early onset of osteoporosis (Bucay et al. 1998; Mizuno et al. 1998), and (3) OPG administration blocks bone loss in ovariectomized rats (an animal model of postmenopausal osteoporosis) (Simonet et al. 1997). Four groups independently tested the association between OPG variants and BMD or osteoporotic fractures but yielded inconclusive findings (Arko et al. 2002; Langdahl et al. 2002; Ohmori et al. 2002; Wynne et al. 2002). Recently, Lee et al. (2009b) collated a total of 16 studies with respect to an association between OPG polymorphisms and BMD through an extensive literature search but only included seven studies (Garcia-Unzueta et al. 2008; Kim et al. 2007; Langdahl et al. 2002; Moffett et al. 2008; Ueland et al. 2007; Yamada et al. 2003; Zhao et al. 2005) for the purpose of meta-analysis. The remaining nine studies (Arko et al. 2002, 2005; Brandstrom et al. 2004; Choi et al. 2005; Hsu et al. 2006; Jorgensen et al. 2004; Ohmori et al. 2002; Wynne et al. 2002; Zajickova et al. 2008) were excluded on the basis that either there was a lack of genotype information or BMD data or only a small number of polymorphisms were tested. Of the three polymorphisms [1181G>C (rs2073618), 163A>G (rs3102735) and 950T>C (rs2073617)] under consideration, 1181G>C was shown to be associated with lumbar BMD in both Europeans and Asians, but with femoral neck and total hip BMD only in Europeans (Lee et al. 2009b). In addition, a recent study showed that the 1181G>C polymorphism affects BMD alone or via an interaction with VDR or TNFSF11 polymorphisms (Mencej-Bedrac et al. 2009). However, another recent study, which analyzed nine OPG SNPs, including 1181G>C, in 1,873 subjects from 405 Caucasian nuclear families, did not find any significant association with any of the studied phenotypes (femoral neck compression strength index as well as its three components, femoral neck BMD, femoral neck width, and weight) (Dong et al. 2009).
Finally, it is important to mention the findings from GWASs. Two SNPs, rs64469804 and rs6993813, showed a genome-wide significant association with both spine and hip bone BMD, and they are in LD with the 1181G>C and 163A > G polymorphisms in the OPG gene (Styrkarsdottir et al. 2008). A novel SNP, rs4355801, which is located in the 3′-UTR of the OPG gene was significantly associated BMD at lumbar spine (P = 7.6 × 10−10) and femoral neck (P = 3.3 × 10−8) (Richards et al. 2008). These findings were confirmed by the meta-analysis of five GWASs (Rivadeneira et al. 2009). Additionally, it is worthy to state that the higher BMD-associated G allele at rs4355801 was associated with higher OPG expression (Richards et al. 2008).
To the best of our knowledge, only six studies have tested if an association exists between RANK polymorphisms and osteoporosis. Three studies involved Korean postmenopausal women. The first study analyzed the 575C>T polymorphism in 650 subjects and found no association with BMD (Choi et al. 2005). The second study did not find the 575C>T polymorphism in 385 subjects. In the third study, analysis of 560 subjects found no association between the 575C>T polymorphism and BMD but an association between two intronic polymorphisms with BMD at lumbar spine was reported (P = 0.04 and 0.02, respectively) (Koh et al. 2007). Nevertheless, as already acknowledged by the authors, “if Bonferroni correction were strictly adopted, associated P values could not retain significance” (Koh et al. 2007).
Hsu et al. (2006) analyzed the 575C>T polymorphism in 1,120 Chinese subjects with either extreme low hip BMD (285 men, 270 women) or high hip BMD (290 men, 275 women). The polymorphism was significantly associated with BMD in men but not in women. The remaining two studies were from a same group, who genotyped 18 SNPs in the RANK gene (including none of the aforementioned three polymorphisms) of 1,873 subjects from 405 Caucasian nuclear families. In the earlier study, RANK showed a highly suggestive association with BMD at the spine, hip and ultradistal radius (Xiong et al. 2006b). However, the later study did not find any significant association between any of the studied phenotypes (femoral neck compression strength index as well as its three components viz. femoral neck BMD, femoral neck width, and weight) and any of the studied SNPs (Dong et al. 2009).
In their initial GWAS, Styrkarsdottir and colleagues found that a SNP (rs3018362) located 27 kb downstream of the TNFRSF11A gene and within the same LD block, was consistently associated with hip BMD; however, the association did not reach genome-wide significance (Styrkarsdottir et al. 2008). In their following GWAS with expanded discovery and replication samples, the association became significant genome-wide (P = 5.4 × 10−8) (Styrkarsdottir et al. 2009). This finding was confirmed by the meta-analysis of 5 GWASs (Rivadeneira et al. 2009).
RANKL was first analyzed as a candidate gene by two groups in 2006. Hsu et al. (2006) evaluated RANKL together with OPG and RANK in 1,120 Chinese with either extreme low hip BMD (285 men, 270 women) or high hip BMD (290 men, 275 women). They found that the T>C polymorphism, rs9594782, which is located in the 5′-UTR of the gene was associated with BMD in only men. Men with TC/CC genotypes had a 2.1 times higher risk of having extremely low hip BMD (P = 0.004) and had lower whole body BMD (P < 0.001) (Hsu et al. 2006). The Deng group analyzed RANKL together with 19 other candidate genes in 1,873 subjects from 405 Caucasian nuclear families and provided evidence that RANKL was highly suggestive for hip BMD (Xiong et al. 2006b). More recently, the Deng group reported that three RANKL SNPs rs12585014, rs7988338, and rs2148073 were significantly associated with femoral neck compression strength index (P = 0.0007, 0.0007, and 0.0005, respectively) even after conservative Bonferroni correction. The first two SNPs are in the promoter region and the last SNP is in intron 2 of the gene. All three SNPs are in almost complete LD (Dong et al. 2009). Additionally, a further promoter polymorphism, −290C>T, was associated with lumbar spine BMD in 239 osteoporotic postmenopausal women from Slovenia (Mencej-Bedrac et al. 2009).
rs9594759, located 113 kb upstream of RANKL, was significantly associated with spine BMD in the GWAS of (Styrkarsdottir et al. 2008). This association was replicated in the meta-analysis of five GWASs (Rivadeneira et al. 2009).
The transforming growth factor-β (TGFβ) superfamily
Transforming growth factor-βs (TGFβs) and bone morphogenetic proteins (BMPs) belong to the TGFβ superfamily, a group of multifunctional peptides that control proliferation, differentiation, and other functions in many cell types (http://www.humpath.com/TGFBs).
TGFB1, which encodes TGFβ1, was also one of the most widely studied candidate genes for osteoporosis. TGFβ1 is highly expressed in bone tissue and plays a key role in controlling bone resorption and formation (Bonewald and Mundy 1990). Tgfb1−/− mice show decreased bone mass and bone elasticity (Geiser et al. 1998). Mutations in TGFB1 cause autosomal dominant Camurati-Engelmann disease, a rare condition characterized by increased BMD that predominantly affects the long bones of the arms and legs (Janssens et al. 2000). Several studies prior to 2007 analyzed common TGFB1 SNPs with respect to BMD and osteoporotic fracture but yielded conflicting results (see McGuigan et al. 2007b; Langdahl et al. 2008 for references). Two recent studies genotyped the same five common SNPs in the TGFB1 gene in 2,975 women from the UK (McGuigan et al. 2007b) and 28,924 subjects from 10 European research studies (Langdahl et al. 2008), respectively; both indicated that none of the studied SNPs were generally associated with BMD or fracture risk. In addition, Huang et al. (2009a) genotyped three TGFB1 SNPs in 1,243 Chinese subjects and also did not find an association with BMD.
Note that for ease of discussion, we moved the BMP2 gene to this section. As discussed below, BMP2 was identified by linkage analysis followed by a case–control association study.
One osteoporosis susceptibility locus was mapped to 20p12.3 using linkage analysis in Icelandic families (Styrkarsdottir et al. 2003). This region contains four genes (BMP2, C20orf42, C20orf154, and CHGB) that are expressed in bone marrow or in an osteoblast cell line. BMP2 was regarded as a promising candidate gene due to its critical role in bone formation and osteoblast differentiation (Fujii et al. 1999; Katagiri et al. 1994; Wozney et al. 1988). Subsequent association analysis using more dense markers demonstrated that an association exists between a missense SNP (Ser37Ala) and osteoporotic fractures in two small Icelandic and Danish cohorts. Additionally, this SNP was associated with low BMD in the Icelandic cohort (Styrkarsdottir et al. 2003). In a study of 805 women from Australia, New Zealand, UK, and Belgium, Ser37Ala was found to be associated with lumbar spine BMD (osteoporotic fracture was not studied) (Reneland et al. 2005). However, in the Rotterdam study that involved a large population-based cohort of 6,353 Dutch whites, Ser37Ala and another polymorphism (Arg190Ser) or haplotypes defined by them are not associated with BMD, rates of bone loss, parameters of hip structural analysis, and fractures (Medici et al. 2006). Further studies that analyzed associations of BMP2 polymorphisms with osteoporosis traits yielded inconclusive results (Choi et al. 2006; Freedman et al. 2009; Ichikawa et al. 2006a; McGuigan et al. 2007a; Tranah et al. 2008; Xiong et al. 2006b).
BMP4 also plays an important role in skeletal development and bone formation (Bellusci et al. 1996; Hogan 1996). Ramesh Babu et al. (2005) analyzed three polymorphisms in the BMP4 gene for association with BMD (n = 1,012) and fracture rates (n = 1,232) in postmenopausal women and found that a missense polymorphism was associated with BMD. An association was also found between BMP4 IVS1-160C>T and calcaneus BMD in the study of 237 healthy Korean men (Choi et al. 2006).
Freedman and colleagues analyzed polymorphisms in BMP2, BMP4, and BMP7 for association with vascular calcification and BMD in 920 European Americans from 374 Diabetes Heart Study families. A SNP in BMP7 was found to be associated with several measures of BMD (Freedman et al. 2009).
Signaling of BMPs require the binding of BMP molecules to their receptors. The type IB BMP receptor (BMPR1B) plays an important role in osteoblast commitment and differentiation (Zhao et al. 2002). In their high-density association study of 383 candidate genes for vBMD at the femoral neck and lumbar spine among older men (n = 862), Yerges et al. (2009a) found that two SNPs in the BMPR1B gene, one in the 3′-UTR (rs1434536) and the other in an intron (rs3796443), were associated with BMD at the lumbar spine.
SMAD6, one of the SMAD proteins, is a TGFβ/BMP inducible antagonist of TGFβ/BMP signaling (Imamura et al. 1997). Smad6 transgenic mice showed postnatal dwarfism with osteopenia (Horiki et al. 2004). In the study of 721 Japanese postmenopausal women, Urano et al. (2009b) recently reported that a SMAD6 intronic polymorphism, rs755451, was associated with BMD.
The NOG gene product, noggin, is an extracellular BMP antagonist. Resequencing of 7 kb of the NOG gene region in 24 randomly selected Afro-Caribbean men identified 22 SNPs. Seven of the ten SNPs that have a minor allele frequency greater than 0.05 were genotyped in 2,060 Afro-Caribbean men aged 40 years or older; none of the SNPs were found to be associated with BMD at the proximal femur or lumbar spine (Moffett et al. 2009).
Selected examples of other ‘older’ genes
Only six ‘older’ genes for which the first association report was published before 2006 are discussed here. Interested readers are invited to consult Ferrari (2008), Liu et al. (2006), and Ralston and de Crombrugghe (2006) for more examples.
COL1A1 encodes the α-1 chain of type I collagen, the principal component of bone extracellular matrix. Mutations in the COL1A1 gene cause osteogenesis imperfecta, a rare disease chiefly characterized by multiple bone fractures, usually resulting from minimal trauma (Pope et al. 1985). COL1A1 is also one of the most extensively studied candidate genes for osteoporosis susceptibility. Most studies focused on the +1245G/T polymorphism, rs1800012, which affects a Sp1 binding site in intron 1 (Liu et al. 2006; Ralston and de Crombrugghe 2006). The functional significance of this polymorphism has been well characterized: the T allele is associated with an increased production of COL1A1 mRNA and protein due to its higher binding affinity for Sp1, and the resulting imbalance between the α-1 and α-2 chains contributes to impaired bone strength and reduced mass by affecting bone mineralization (Mann et al. 2001; Stewart et al. 2005). An earlier meta-analysis of 26 published studies including 7,849 participants showed that the presence of the T allele was associated with a modest reduction in BMD and a significant increase in risk of vertebral fractures (Mann and Ralston 2003). A more recent meta-analysis of 20,786 individuals from several European countries observed that only homozygotes for the T allele were associated with BMD and incident vertebral fractures (Ralston et al. 2006). Importantly, both studies indicated that the Sp1 polymorphism may predispose subjects to vertebral fractures independent of an effect on BMD, consistent with the experimentally observed effect of the polymorphism on bone quality (Stewart et al. 2005).
Garcia-Giralt et al. (2002) identified two polymorphisms in the promoter region of the COL1A1 gene, −1997G/T (rs1107946) and −1663IndelT (rs2412298), in a cohort of 256 postmenopausal women. Only −1997G/T was found to be associated with BMD and −1663IndelT was found to be in strong LD with the Sp1 polymorphism; therefore, the −1997G/T polymorphism has been the focus of several subsequent studies. An association with reduced BMD was reported in (Bustamante et al. 2007b; Liu et al. 2004; Yamada et al. 2005a; Zhang et al. 2005a) but not in (Yazdanpanah et al. 2007).
To date, only two studies have examined the effects of the aforementioned three polymorphisms and their haplotypes on BMD. The first study was a population-based association study involving 3,270 women from the UK. Although the three polymorphisms were associated with BMD, the most consistent associations were with haplotypes defined by all of them. Homozygotes of haplotype 2 (−1997G/−1663delT/+1245T) had reduced BMD, whereas homozygotes of haplotype 3 (−1997T/−1663insT/+1245G) had increased BMD (Stewart et al. 2006). The second analysis was a case–control study comprising 462 osteoporotic patients and 336 controls from Denmark. Although only the −1663delT and +1245T alleles were associated with reduced BMD, homozygotes of the aforementioned haplotype 2 also had reduced BMD (Husted et al. 2009). Here, it is also pertinent to mention another work, which analyzed 6,280 individuals from the Netherlands and found no independent effect of the −1997G/T polymorphism on BMD and fracture (Yazdanpanah et al. 2007). These findings, considered together, suggest that the effect of haplotype 2 should be driven by the −1663IndelT and Sp1 polymorphisms. Functional analysis revealed that haplotype 2 increased gene transcription twofold compared with the common haplotype 1 (−1997G/−1663insT/+1245G) (Jin et al. 2009).
RUNX2, alternatively known as CBFA1, is an osteoblast-specific transcription factor. Runx2−/− mice show complete absence of bone (Komori et al. 1997; Otto et al. 1997), and Runx2+/− mice show characteristic skeletal abnormalities observed in the human autosomal dominant skeletal disorder, cleidocranial dysplasia (CCD) (Otto et al. 1997). CDD, which is characterized by hypoplasia/aplasia of clavicles, patent fontanelles, supernumerary teeth, short stature and other skeletal anomalies, is caused by inactivating mutations in the RUNX2 gene (Lee et al. 1997; Mundlos et al. 1997; Quack et al. 1999). To date, only a limited number of studies investigated the role of RUNX2 polymorphisms in osteoporosis. The first study analyzed 495 Australian women (the Geelong Osteoporosis Study, GOS) and found an association between a synonymous codon polymorphism in exon 2 (SNP G>A+198) and higher BMD (Vaughan et al. 2002). Furthermore, this finding was replicated in the study of Vaughan et al. (2004), which analyzed 312 Scottish postmenopausal women. Using equal numbers of subjects with extremely high or low femoral neck BMD (n = 132 each), Doecke et al. (2006) found a significant over-representation of SNP G>A+198 in the upper decile of femoral neck BMD. SNP G>A+198 is in near-complete LD with three promoter SNPs that are in complete LD with each other. These four SNPs represent a haplotype block that is significantly associated with increased BMD. When engineered into a luciferase reporter vector, the minor allele-containing sequence block resulted in significantly higher promoter activity than the major allele-containing counterpart. An electrophoretic mobility shift assay revealed that the strongest differential DNA–protein binding occurred at the −1025T/C polymorphic site, rs7771980 (Doecke et al. 2006). The T>C substitution was predicted to affect a potential binding site for the transcription factor NF-kappaB (Lee et al. 2009a). The −1025T/C polymorphism was also associated with increased femoral neck BMD in a cohort of 821 Spanish postmenopausal women (Bustamante et al. 2007a). However, the −1025T/C polymorphism was associated with decreased lumbar spine and proximal femur BMD in 729 postmenopausal Korean women (Lee et al. 2009a).
Promoted by the finding that Cnr2 (cannabinoid receptor 2) knockout mice show decreased bone mass, Karsak et al. (2005) tested whether CNR2 is associated with human osteoporosis in a case–control study involving 168 French postmenopausal osteoporotic women and 220 ethnically, age- and sex-matched controls. Multiple SNPs and haplotypes of the gene were associated with osteoporosis, with rs4649124 and rs2501431 showing the strongest association (Karsak et al. 2005). Moreover, rs2501431 was found to be associated with BMD in the study of a population-based Japanese sample (n = 2,238) (Yamada et al. 2007) and in the study of 1,243 Chinese subjects with low or high BMD (Huang et al. 2009b). Furthermore, SNPs in CNR2 were associated with hand bone strength phenotypes in the study of 574 adults belonging to 126 two- to four-generation pedigrees of Chuvashian descent (Karsak et al. 2009).
Secreted phosphoprotein 1 (SPP1), also known as osteopontin, is involved in bone remodeling (Dodds et al. 1995). Two earlier small CGASs reported association of a dinucleotide repeat polymorphism in the SPP1 gene with baseline femoral neck BMD (Willing et al. 1998) and spine BMC (Willing et al. 2003), respectively. The collaborative meta-analysis of 150 candidate genes in 19,195 subjects of European origin indicated that 14 SPP1 SNPs were significantly associated with lumbar spine BMD, the lowest P value being 6.0 × 10−8 (Richards et al. 2009).
Defects in the chloride channel 7 (CLCN7) gene are the cause of some forms (Cleiren et al. 2001; Kornak et al. 2001) of osteopetrosis, a rare genetic disease characterized by abnormally dense bone due to impaired osteoclast function. Pettersson et al. tested several common SNPs and a variable number tandem repeat (VNTR) polymorphism within intron 8 in the CLCN7 gene for association with lumbar spine and femoral neck BMD in a population-based cohort of 1,077 Scottish women aged 45–55 years. Two common SNPs in exon 15, rs12926089 (Val418Met) and rs12926669, which are separated by only seven base pairs and are in strong LD, were associated with femoral neck BMD (Pettersson et al. 2005). In the study of 425 postmenopausal women aged 64 ± 7 years, the VNTR polymorphism but not rs12926089 was associated with femoral neck BMD (Kornak et al. 2006). In the study of 1,692 healthy premenopausal white sisters (age 33.1 ± 7.2) and 715 health white brothers (age 33.6 ± 10.9) in the US, none of the seven CLCN7 polymorphisms, including rs12926089 and the VNTR polymorphism, were associated with lumbar spine and femoral neck BMD (Chu et al. 2008).
Parathyroid hormone (PTH) is a key regulator of calcium homeostasis and bone remodeling. Administration of PTH increases bone mass by stimulating bone formation in both experimental animals (Hock and Gera 1992; Hock et al. 1988; Jilka et al. 1999) and patients with osteoporosis (Neer et al. 2001). Earlier studies using small sample size yielded inconsistent results regarding association between polymorphisms in PTH and osteoporosis (Deng et al. 2002a; Gong et al. 1999; Katsumata et al. 2002). Tenne et al. (2008) analyzed four PTH pathway genes, PTH, PTHLH (parathyroid hormone-like hormone), PTHR1 (PTH receptor 1), and PTHR2 (PTH receptor 2), in 1,044 women aged 75 years or older from the Malmö Osteoporosis Prospective Risk Assessment study. They found that polymorphisms in PTH may be associated with BMD-independent increased fracture risk. In a very recent study, consistent association was found between PTH and femoral neck BMD in both a GWAS discovery sample of 983 unrelated Caucasian subjects and a family-based replication sample of 2,557 Caucasian subjects (Guo et al. 2009).
Candidate genes identified in the past 4 years
Candidate osteoporosis genes, which were found in the past 4 years but cannot be classified into any of the aforementioned pathways, will be discussed in the context of several representative studies.
Simultaneous analysis of multiple genes involved a common pathway
The study of multiple Wnt pathway genes in a sample of individuals with low and high BMD (Sims et al. 2008) were discussed earlier. Another study selected four genes involved in the growth hormone-insulin-like growth factor I axis for analysis: the growth hormone releasing hormone gene (GHRH), the growth hormone releasing hormone receptor gene (GHRHR), the growth hormone secretagogue receptor gene (GHSR), and the growth hormone receptor gene (GHR). One SNP in each of these genes were genotyped in 498 men and 468 women aged 59–71 years. The GHRH SNP was associated with BMC and BMD at the proximal femur and lumbar spine in both sexes (Dennison et al. 2009).
Mineralization of the extracellular matrix of bone is an essential element of bone development, maintenance and repair. ALPL (alkaline phosphatase, liver/bone/kidney), ANKH [ankylosis, progressive homolog (mouse)], and ENPP1 (ectonucleotide pyrophosphatase/phosphodiesterase 1) are central regulators of phosphate balance and thus play a key role in the matrix mineralization process (Cheung et al. 2009b). The three genes were analyzed for association with hip geometry traits in 1,513 unrelated subjects from the Framingham Offspring cohort (Cheung et al. 2009b). Several SNPs in the three genes showed nominally association but only rs1974201 in ENPP1 remained significant after either Bonferroni corrections or with 1,000,000 permutations. The minor allele of rs1974201 was predicted to disrupt a HOXA7 binding site (Cheung et al. 2009b). In this regard, it is pertinent to mention that a new variant in the Enpp1 gene, which was induced by the use of chemical mutagen N-ethyl-N-nitrosourea (ENU), causes low BMD, crystal-related arthropathy, and vascular calcification in mice (Babij et al. 2009). Another study analyzed ALPL and ENPP1 in 1,253 individuals belonging to 310 Caucasian nuclear families from one ethnically homogenous population, Chuvashians, who live in numerous small villages along the Volga river in the Chuvasha and Bashkortostan Autonomies of the Russian Federation. SNPs in both genes were found to be associated with bone strength related skeletal sites (Ermakov et al. 2009). Of them, a SNP in ALPL, rs3200254 (Tyr263His), had been previously reported to be associated with BMD in the study of 501 Japanese postmenopausal women (Goseki-Sone et al. 2005). rs3200254 was shown to be a functional polymorphism by in vitro studies (Goseki-Sone et al. 2005; Sogabe et al. 2008). Finally, polymorphisms in ANKH were reported to be associated with skeletal size traits in two studies that analyzed only 176 and 212 nuclear families from the Chuvashian population (Malkin et al. 2005, 2006).
Simultaneous analysis of a large number of candidate genes
Having searched several publicly available sources with respect to gene expression and function in normal human trabecular bone cells, genes with a skeletal phenotype in mice and evidence of genes implicated in human skeletal conditions, Yerges et al. (2009a) selected 383 physiologically defined candidate genes for analysis. In the discovery phase, they screened 4,608 tagging and potentially functional SNPs for association with femoral neck and lumbar spine vBMD from quantitative computed tomography among 862 community-dwelling Caucasian men aged 65 years or older in the Osteoporotic Fractures in Men Study (MrOS). In the validation phase, the most promising SNPs, 193 in 56 genes associated with femoral neck vBMD and 173 in 59 genes associated with lumbar spine vBMD, were genotyped in an additional 1,156 Caucasian men from MrOS. They not only confirmed previously identified associations in the SOST (Uitterlinden et al. 2004) and FOXC2 (Yamada et al. 2006) genes but also described new associations in nine genes (i.e. APC, BMPR1B, DMP1, FLT1, HOXA, IGFBP2, NFATC1 and PTN). Specifically, eight SNPs in six genes (i.e. APC, DMP1, FGFR2, FLT1, HOXA and PTN) were associated only with femoral neck vBMD; however, 13 SNPs in 7 genes (i.e. APC, BMPR1B, FOXC2, HOPXA, IGFBP2, NFATC1 and SOST) were associated with only lumbar spine vBMD. In other words, only SNPs in the APC, FOXC2 and HOXA genes were associated with vBMD at both studied sites (Yerges et al. 2009a). APC, FOXC2 and BMPR1B were discussed previously in “APC”, “FOXC2” and “BMPR1B”, respectively.
The above notwithstanding, Yerges et al. did not find associations in the four widely studied genes (i.e. LRP5, ESR1, COL1A1, and VDR). They provided three reasons to account for this discrepancy. First, they analyzed only men whereas most of the past candidate gene studies focused on women. Second, they measured vBMD; however, most of those past studies measured areal BMD. Lastly, they cannot exclude a weaker association between SNPs in the four genes and vMBD at the studied sites due to their statistic power (Yerges et al. 2009a).
Representative studies that analyzed single candidate genes
HMGA2 encodes high mobility group (HMG) AT-hook 2, a member of the HMG DNA-binding family of non-histone architectural transcription factors. A role for HMGA2 in human bone growth and development has been recently suggested by three studies. A common SNP in the 3′-UTR of the HMGA2 gene, rs1042725, was associated with height in the general population (Sanna et al. 2008; Weedon et al. 2007, 2008). The Zmuda group tested whether this SNP was associated with other skeletal measures in two large cohorts of diverse ethnicity: 1,680 Afro-Caribbean men aged 40 years or older and 1,548 Caucasian American men aged 69 years or older. The minor allele of this SNP was associated with decreased tibia trabecular vBMD in both populations with small P values (P = 0.007 in the former and P = 0.0007 in the latter). Analysis of the LD pattern within a one megabase region of HMGA2 revealed that the only other SNPs with significant LD with the tested SNP are in the HMGA2 gene region (Kuipers et al. 2009).
rs1042725 is located within the upstream sequence (USS, sequence between the translational termination codon and the upstream core polyadenylation signal sequence in accordance with Chen et al. (2006a)) of the gene’s 3′-UTR. At present, the best predictor of the putative functionality of USS 3′-UTR variants is their effect on the mRNA secondary structure (Chen et al. 2006b). Using the method described previously (Chen et al. 2006b), the minor T allele was predicted to induce a potentially functional pattern III change that would alter the orientation of the global stem-loop structures as compared with the major C allele. In this regard, it is pertinent to point out that the HMGA2 3′-UTR contains 10 AUUUA motifs that are thought to be involved in mRNA stability (Borrmann et al. 2001). Thus, it is quite plausible, although it remains to be demonstrated experimentally, that the mRNA secondary structural changes induced by the minor T allele affect the binding of trans-acting factors to the AUUUA motifs resulting in deregulated mRNA stability.
ARHGEF3, RHOA, and FLNB on chromosome 3p14–p21
The 3p14–p21 region of the human genome has been demonstrated to contain a BMD QTL by four independent studies (Duncan et al. 1999; Streeten et al. 2006; Wilson et al. 2003; Xiao et al. 2006) and two meta-analyses (Ioannidis et al. 2007; Lee et al. 2006). ARHGEF3, which encodes the Rho guanine-nucleotide exchange factor (GEF) 3, was initially analyzed by Mullin et al. (2008) as a strong positional candidate for two reasons. First, ARHGEF3 specifically activates two members of the RhoGTPase family, RHOA implicated in osteoblast differentiation (Meyers et al. 2005) and RHOB involved in osteoarthritis (Mahr et al. 2006). Second, mutations in the FGD1 gene, which encodes another RhoGTPase regulatory protein, cause Aarskog-Scott syndrome which is characterized by multiple skeletal abnormalities (Orrico et al. 2007). They first analyzed 17 SNPs in a discovery cohort of 769 female sibs (mean age 54.2 ± 12.7 years) recruited from Australia and UK. Five SNPs were found to be associated with various measures of age-adjusted BMD (P = 0.0007–0.041), with the intron 1-located rs7646054 showing the strongest association at each site studied (total hip, P = 0.006; femoral neck, P = 0.0007; spine, P = 0.006). Next, rs7646054 was analyzed in a population-based cohort comprising 780 women aged between 45 and 64; all subjects were recruited from a single large general practice in Chingford, North-East London. Significant associations with hip and spine BMD (P = 0.003–0.038) were confirmed, with an additional association with fracture rate (P = 0.02) being demonstrated. In both cohorts, the more common G allele at rs7646054 is associated with lower BMD at each site studied (Mullin et al. 2008). Using the same discovery and replication cohorts, the same group also found an association between variations in the RHOA gene (located on 3p21.3) and the BMD Z score of the spine and hip (P = 0.001–0.036) (Mullin et al. 2009). These findings suggest that the Rho-GTPase-RhoGEF pathway is involved in the etiology of osteoporosis.
More recently, this group further selected the filamin N (FLNB) gene on 3p14–p21 for an association study (Wilson et al. 2009). Mutations in the FLNB gene cause rare human skeletal disorders including autosomal recessive spondylocarpotarsal syndrome, autosomal dominant Larsen syndrome, perinatal lethal atelosteogenesis I and III phenotypes (Krakow et al. 2004) and boomerang dysplasia (Bicknell et al. 2005). Targeted disruption of Flnb in mice results in skeletal malformations similar to the malformations observed in human spondylocarpotarsal syndrome (Farrington-Rock et al. 2008; Lu et al. 2007; Zhou et al. 2007). In addition, variations in FLNB have been associated with human stature variation (Lei et al. 2009a). The most significant finding from the study of Wilson et al. (2009) is the association of a SNP located within intron 1 of the FLNB gene, rs9822918, with femoral neck BMD both in the aforementioned family-based discovery cohort and a population-based cohort of 1,315 Australian women aged between 72 and 85.
Five candidate genes, including FLNB on 3p14–p25 as well the rs7646054 in ARHGEFS3 (Mullin et al. 2008), were analyzed in a case–control cohort of 1,080 Chinese females (Li et al. 2009). No association was found between rs7646054 and a BMD Z score at the spine, femoral neck, or total hip in terms of the whole study population or 533 postmenopausal subjects (Li et al. 2009). Alternatively, rs9822918 in FLNB (Wilson et al. 2009) was found to be significantly associated with spine, femoral neck, or total hip BMD in the Chinese samples (Li et al. 2009). Additionally, Li et al. (2009) also found an association between SNPs in the PTHR1 gene with femoral neck BMD at 3p.
ITGA1 encodes integrin α1 chain, which associates with the β1 chain (ITGB1) to form a heterodimer that functions as a dual laminin/collagen receptor in neural cells and hematopoietic cells. Studies using Itga1-deficieny mice indicated that integrin α1 plays an essential role in fracture healing (Ekholm et al. 2002) and cartilage remodeling (Zemmyo et al. 2003). Lee et al. (2007) analyzed the ITGA1 gene in 946 postmenopausal Korean women and found that several SNPs were associated with BMD at various femur sites. The collaborative meta-analysis of 150 candidate genes in 19,195 participants from five populations of European origin showed that rs13179969, located in one intron of ITGA1, was associated with lumbar spine BMD (P = 9.6 × 10−7) (Richards et al. 2009).
A GWAS has recently identified an association between two common SNPs in the CLDN14 (claudin 14) gene with kidney stones (Thorleifsson et al. 2009). Claudins are tetraspan transmembrane proteins that regulate paracellular passage of ions and small solutes at epithelial tight junctions (Krause et al. 2008). Previous data suggest a risk for bone loss in individuals with kidney stones (Asplin et al. 2003); therefore, the same SNPs were genotyped in a sample of 8,450 Icelandic individuals and 3,601 Danish women and they were found to be associated with reduced BMD at the hip (P = 0.00039) and spine (P = 0.0077). It was postulated that kidney stones and decreased BMD are secondary to metabolic abnormalities (i.e., decreased serum total CO2 and increased urinary calcium) associated with the CLDN14 SNPs (Thorleifsson et al. 2009).
Quantitative trait loci mapping in inbred mouse strains
Quantitative trait loci (QTL) mapping in mice is a powerful tool to dissect the genetic basis of complex diseases. This section discusses two approaches that have directly contributed to the identification of human osteoporosis loci.
Genetic linkage studies using the traditional mapping approach
The principle of the traditional mapping approach used in mouse models has been reviewed by Ralston and de Crombrugghe (2006). Employment of this approach has mapped numerous QTLs for osteoporosis-related traits, particularly, BMD in mice (Liu et al. 2006; Xiong et al. 2009b) and, at the same time, provided many valuable insights into the complex genetic architecture of these traits (Zmuda et al. 2006). However, to date, this approach has only contributed directly to the identification of a very limited number of mouse BMD genes such as Alox15 (Klein et al. 2004), Sfrp4 (Nakanishi et al. 2006), Darc (Edderkaoui et al. 2007) and Ece1 (Saless et al. 2009). The reason for this limitation is that it is always challenging to pinpoint the true ‘culprit’ within a region typically the size of 20–40 cM as defined by genetic mapping (Ralston and de Crombrugghe 2006). For example, the Darc gene was identified by four approaches: (1) fine-mapping of the region of interest using dense polymorphic markers; (2) generation of congenic sublines of mice by repeated successive backcrosses against the parental strains and phenotype characterization; (3) expression profiling of genes in the quantitative QTL region and (4) SNP analyses (Edderkaoui et al. 2007). To date, only human homologs of the Alox15 gene have been analyzed as candidate genes for osteoporosis.
Identification of Alox15 as a negative regulator of peak BMD in mice
Klein et al. (2004) generated a DGA/2 (D2) background congenic mouse, in which an 82 Mb region of chromosome 11 known to harbor a BMD QTL was replaced by the corresponding region of the C57BL/6 (B6) genome. The congenic mice with the B6 chromosome 11 region were characterized with increased peak BMD and improved measures of femoral shaft strength as compared to heterozygous or D2 littermates. Linkage analysis of the B6D2F2 population narrowed the BMD QTL to a 31-Mb region. Next, they analyzed gene expression in B6 and D2 mice by microarray analysis of kidney tissue and identified Alox15 as the only differentially expressed gene within this QTL interval. They further demonstrated that Alox15 expression was also significantly lower in B6 osteoblastic cell cultures than their D2 counterparts (Klein et al. 2004). Alox15 was confirmed to be a negative regulator of peak BMD in mice by three lines of evidence. First, transient overexpression of Alox15 in murine bone marrow stromal cell cultures restricted osteoblast differentiation. Second, Alox15 knockout mice have higher BMD than normal mice. Last, pharmacological inhibitors of Alox15 improved bone density and strength (Klein et al. 2004). Alox15 encodes the enzyme, 12/15-lipoxygenase (12/15-LO) that converts arachidonic and linoleic acids into endogenous ligands for the peroxisome proliferator-activated receptor-γ (PPARγ). Activation of the PPARγ pathway inhibits differentiation of preosteoblasts (Khan and Abu-Amer 2003).
The finding of Klein et al. (2004) was considered to have relevance to human osteoporosis for two specific reasons: (1) a putative association between a SNP of PPARγ and BMD was identified in Japanese postmenopausal women (Ogawa et al. 1999) and (2) linkage was found between spinal BMD (Devoto et al. 1998) [as well as wrist BMD (Deng et al. 2002b)] and human chromosomal region 17p13 that harbors ALOX12 and ALOX15. Nevertheless, overexpression of the human ALOX15 gene in transgenic rabbits protects against leukocyte-mediated bone loss (Serhan et al. 2003).
Inconclusive findings on association between ALOX15 and human osteoporosis-related traits
Studies that analyzed association between ALOX15 and osteoporosis-related traits
Study subjects (years)
Urano et al. (2005)
Healthy postmenopausal women (66.7 ± 8.9)
Allele A heterozygotes and homozygotes (n = 273) had significantly lower Z scores for lumbar spine and total body BMD than allele G homozygotes (n = 46) (P = 0.0014 and 0.048, respectively)
Ichikawa et al. (2006b)
White healthy men (34.6 ± 10.8) and premenopausal women (33.2 ± 7.1)
Only rs9894225, which is in substantial LD with rs748694 (D′ = 0.74), showed marginal association with hip BMD in women (PANOVA = 0.043; PVC = 0.028)
Mullin et al. (2007)
Postmenopausal women (62.5 ± 5.9)
None of the SNPs were significantly associated with any of the BMD parameters or fracture data
Cheung et al. (2008a)
Women with low or high BMD at either the hip or spine
The G allele of rs2619112 was associated with a reduced risk of low BMD at the femoral neck in premenopausal women (OR = 0.442, P = 0.007) but an increased risk in postmenopausal women (OR = 1.727, P = 0.042)
Tranah et al. (2008)
Postmenopausal women (71.7 ± 5)
rs7220870 T/T genotype had a higher rate of hip fracture (hazard ratio [HR] = 1.33; 95% CI = 1.00–1.77) compared with the G/G genotype
ALOX12 is likely a true susceptibility gene for osteoporosis
Lipoxygenases are a family of iron-containing enzymes that catalyze the incorporation of molecular oxygen into polyunsaturated fatty acid substrates such as arachidonic and linoleic acids. Mouse Alox15 shares higher amino acid sequence similarity with human ALOX15 (73%) than with human ALOX12 (57%). However, in terms of function, mouse Alox15 is more similar to human ALOX12 because it preferentially catalyzes the insertion of oxygen into arachidonic acid at carbon-12 rather than at carbon-15 (Ichikawa et al. 2006b). Therefore, Ichikawa et al. (2006b) analyzed ALOX12 for its possible involvement in osteoporosis and found significant associations between several SNPs in the ALOX12 gene and spine BMD in American white men (n = 411) and premenopausal women (n = 1,291). Three of these SNPs were also found to be associated with spine BMD in 779 white postmenopausal women in the UK (Mullin et al. 2007). In addition, a significant association between a different ALOX12 SNP and BMD at the ultradistal radius was observed in 1,873 subjects from 405 white nuclear families in the USA (Xiong et al. 2006b). These consistent findings suggest that ALOX12 is likely a true osteoporosis gene.
Genome-wide haplotype association mapping
Genome-wide haplotype association mapping (GWHAP) represents a quick in silico method to map QTLs in mice by using a large panel of inbred strains wherein dense sets of genotyped SNPs are available (Grupe et al. 2001; Liao et al. 2004; McClurg et al. 2006; Pletcher et al. 2004). At present, the application of this approach in finding QTLs for complex traits is still in its initial stage (Webb et al. 2009).
Identification of Cer1 as a putative BMD gene in female mice
Using in silico analysis of whole body BMD variation and SNP data retrieved from 30 inbred mouse strains (Bogue et al. 2007), Tang and colleagues identified 22 blocks that may contain genes for BMD in female mice (Tang et al. 2009). They focused on two chromosomal regions, each showing two peaks in close proximity. Of the 10 known genes located within these two regions, they only selected the Cer1 gene on chromosome 4 for further analysis because (1) it is in close proximity to the peak, and (2) Cer1 may act as an inhibitor of BMP and Wnt signals (Piccolo et al. 1999). A non-synonymous SNP, Met232Ile (rs32341805), was shown to be strongly associated with lower whole body BMD in female mice. Met232Ile is located in the cystine knot, potentially affecting the interaction of Cer1 with BMPs and other proteins. Moreover, Cer1 was shown to be expressed in mouse bone and growth plate by RT-PCR, immunohistochemistry and in situ hybridization (Tang et al. 2009).
Putative association of a genetic variant in CER1 with BMD and fracture in Chinese women
After analysis of a cohort of 1,083 high and low BMD Chinese subjects, Tang et al. ( 2009) found that a non-synonymous SNP (rs3747532; Ala65Gly) in CER1 was associated with increased risk of low BMD in premenopausal women and increased risk of vertebrate fractures in postmenopausal women. However, the association signal observed in the human study became insignificant after Bonferroni correction for multiple traits tested.
Linkage analysis in humans
Linkage analysis has been successfully used for identifying many monogenetic diseases including multiple monogenic skeletal diseases affecting BMD (Duncan and Brown 2008). Although its employment has identified numerous QTLs that regulate osteoporosis-related traits on virtually any chromosome in the normal population (Ferrari 2008; Ioannidis et al. 2007; Kaufman et al. 2008; Liu et al. 2006; Zhang et al. 2009c), limited success was achieved in identifying the underlying genes. Indeed, it is always challenging to identify the responsible gene in a large region mapped by linkage analysis, particularly when the tested SNPs are in extensive LD (Ichikawa et al. 2008). Consequently, in practice, linkage analysis was often complemented by a CGAS, exemplified by the BMP2 gene (“BMP genes”). Here, we review several osteoporosis candidate genes that were discovered through fine-mapping previously known QTL loci.
Cheung and colleagues fine-mapped a 6 Mb spine BMD QTL on 1q21–q23 in 610 sib pairs from 231 Chinese families using 380 SNPs (Cheung et al. 2009a). Only two SNPs in the pre-B-cell leukemia homeobox-1 (PBX1) gene, rs2800791 and rs9661977, showed consistent associations with spine BMD. rs2800791 was replicated in a Chinese case–control cohort comprising 835 subjects with extremely high and extremely low BMD (P = 0.007) and a Japanese case–control cohort comprising 703 osteoporotic subjects and 565 healthy subjects (P = 0.05). PBX1 was shown to be expressed in both human bone-derived cells and mouse MC3T3-E1 preosteoblast cells by RT-PCR analysis. Transient silencing of Pbx1 expression with RNAi in MC3T3-E1 cells resulted in the downregulation of Runx2 and Osterix, two master genes that control osteoblast differentiation. rs2800791, located in intron 2, was predicted to affect a binding site for the transcription factor, c-myc (Cheung et al. 2009a).
Having genotyped 18 microsatellite markers within a 117-cM interval on 14q in 1,459 subjects from 306 Chinese families, Cheung et al. (2008c) confirmed that this region contains a QTL for trochanter and total hip BMD. Of the >300 known genes present in this QTL region, they genotyped 65 SNPs in the top five candidate genes [ranked by means of gene prioritization (Aerts et al. 2006)] within the linkage peak, in 706 and 760 Chinese subjects with extremely high and low trochanter and total hip BMD, respectively. ESR2 and a new gene, LTBP2 (latent TGF-β binding protein 2), were found to be significantly associated with the studied phenotypes. LTBP2 was also shown to be associated with prevalent fractures, independent of variations on BMD. Ltbp2 is expressed in mouse MC3T3-E1 cells (Cheung et al. 2008c).
Several other genes
High resolution linkage and LD analyses of a previously known QTL on 1p36, identified several candidate genes for low BMD including RERE, G1P2, SSU72 and CCDC27 (Zhang et al. 2009a).
GWASs of SNPs
Taking advantage of the much more extensive knowledge of the complex genetic basis of osteoporosis, the larger number of DNA samples collected over the years, the increased knowledge of human genome structure (as exemplified by the HapMap project), and technological advances in high-throughput genotyping, the first GWAS on common genetic risk factors for osteoporosis was published in 2007 (Kiel et al. 2007a). This study analyzed ~100,000 markers in 1,141 subjects in the Framingham Heart study and identified 40 SNPs that may be potentially associated with several bone phenotypes. However, none of these associations met the standard of genome-wide significance. By contrast, the next two GWASs genotyped >300,000 SNPs in much larger cohorts and achieved genome-wide significant associations (Richards et al. 2008; Styrkarsdottir et al. 2008). However, these studies were only the beginning and were followed by more GWASs (Cho et al. 2009; Guo et al. 2009; Liu et al. 2008b, 2009; Styrkarsdottir et al. 2009; Timpson et al. 2009; Xiong et al. 2009a) in a short time period. Very recently, a meta-analysis of five GWASs of femoral neck and lumbar spine BMD in 19,195 subjects of Northern European descent has also been published (Rivadeneira et al. 2009).
GWASs validated or provided the strongest evidence to date that LRP5, ESR1, SOST, OPG, RANKL, and RANK are true osteoporosis genes. This section focuses on new loci revealed by the original GWASs and the subsequent meta-analytic study.
Novel BMD loci revealed by original GWASs
rs10876432, located near the SP7 (also known as OSTERIX) locus on 12q13, showed significant association with spine BMD (P = 1.3 × 10−7) in the GWAS of Styrkarsdottir et al. (2009). Common SNPs near the SP7 region were also found to be associated with total body BMD in a GWAS that analyzed a population of children (Timpson et al. 2009). These findings were supported by the meta-analysis of Rivadeneira et al. (2009). SP7 is a zinc finger-containing transcription factor that is essential for osteoblast differentiation and bone formation (Baek et al. 2009; Nakashima et al. 2002).
Liu et al. (2009) performed the first bivariate GWAS of obesity and osteoporosis. Body mass index (BMI) and hip BMD were used as the obesity and osteoporosis phenotypes, respectively. They scanned ~380,000 SNPs in 1,000 homogenous unrelated Caucasians (female n = 501, male n = 499) and identified two SNPs in intron 1 of the SOX6 (sex determining region Y-box 6) gene that were bivariately associated with both BMI and hip BMD in the male subjects. They further replicated their finding in the 1,370 male subjects of the Framingham Heart Study cohort (Liu et al. 2009). The meta-analysis of Rivadeneira et al. (2009) identified the SNP, re7117858, which is located 297 kb upstream of SOX6, as being associated with femoral neck BMD at genome-wide significance level. SOX6 is a member of the SOX gene family that encodes transcription factors defined by the conserved HMG DNA-binding domain (Cohen-Barak et al. 2001). Sox6+/− mice present with mild skeletal abnormalities and Sox6−/− fetuses die with severe, generalized chondrodysplasia (Smits et al. 2001).
FAM3C and SFRP4
Cho et al. (2009) analyzed 352,228 eligible SNPs in 8,842 subjects of the Korean Genome Epidemiology Study for eight quantitative traits and identified two loci of interest for BMD. On chromosome 7q31, rs7776725 was associated with BMD at the radius (P = 1.0 × 10−11), tibia (P = 1.6 × 10−6) and heel (P = 1.9 × 10−10). This SNP was located within FAM3C, a gene that has no known role in bone biology. On chromosome 7p14, rs1721400 was consistently associated with BMD at the aforementioned three sites (P = 2.2 × 10−3, 1.4 × 10−7 and 6.0 × 10−4, respectively). Of the three genes in the immediate vicinity (TXNDC3, SFRP4, and EPDR1), SFRP4 emerged as the obvious candidate (refer to “Secreted frizzled-related protein genes”). Moreover, sfrp4 has been shown to be a negative regulator of BMD in mice (Nakanishi et al. 2006, 2008).
ADAMTS18 and TGFBR3
Xiong et al. (2009a) tested ~380,000 SNPs in 1,000 unrelated white US subjects for association with BMD. For replication, they genotyped the most significant SNPs in 1,972 subjects from white US pedigrees, a Chinese hip fracture sample comprising 350 cases and controls, a Chinese BMD sample with 2,955 subjects and a Tobago cohort of 908 males of African ancestry. Two genes, ADAMTS18 (ADAM metallopeptidase with thrombospondin type 1 motif, 18) and TGFBR3 (transforming growth factor, beta receptor III), were significantly associated with BMD variation in the three major ethnic groups. These associations were further supported by a meta-analysis of the publicly available Framingham GWAS data from 2,953 whites. In addition, ADAMTS18 SNPs were significantly associated with hip fracture in the Chinese hip fracture sample (Xiong et al. 2009a).
The minor C allele of one significant ADAMTS18 SNP, rs16945612, was predicted to generate a binding site for TEL2, a member of the E26 transformation-specific family of transcription factors. Given that TEL2 is a transcriptional repressor of two genes, BMP-6 and PARa, involved in osteoblast differentiation and bone remodeling (Gu et al. 2001), the minor C allele was presumed to repress ADAMTS18 expression as well. This postulate was supported by electrophoretic mobility shift analysis. In addition, NCBI GEO (Gene Expression Omnibus) expression profiles showed that subjects with non-union skeletal fractures (unhealed at 6 months) had significantly lower ADAMTS18 levels than those with normal-healing fractures (Xiong et al. 2009a).
TGFBR3 binds to various members of the TGF-β superfamily including the BMP family. Notably, TGFBR3 has been shown to modulate the biological function of BMP2 (Kirkbride et al. 2008). In addition, Tgfbr3−/− mice show severe skeletal abnormalities (Stenvers et al. 2003). Furthermore, NCBI GEO expression profiles showed that subjects with non-union skeletal fractures had significantly higher TGFBR3 levels than those with normal-healing fractures (Xiong et al. 2009a).
Two SNPs, rs7524102 and rs6696981, in LD on chromosome 1p36 showed associations with both hip and spine BMD. rs7524102, the stronger marker, exhibited similar effects on hip BMD among the studied Icelandic, Danish and Australian subjects. The closest gene is ZBTB40 (zinc finger and BTB domain containing 40), which is located 80 kb downstream from the signal (Styrkarsdottir et al. 2008). This association was confirmed by the meta-analysis of Rivadeneira et al. (2009). ZBTB40 is expressed in bone although its function is unknown (Styrkarsdottir et al. 2008).
rs2010281, located in intron 1 of the MARK3 (MAP/microtubule affinity-regulating kinase 3) gene, was significantly associated with total hip BMD (P = 1.8 × 10−9) in the GWAS of Styrkarsdottir et al. (2009). This SNP was significantly associated with femoral neck BMD in the meta-analysis of Rivadeneira et al. (2009) but not at the genome-wide significance level. As noted by Rivadeneira et al. (2009), this discrepancy may be due to the different phenotypes used in the two studies. The MARK3 gene product phosphorylates microtubule-associated proteins and plays a role in determining cell polarity (Styrkarsdottir et al. 2009).
The MHC region
Three interleukin 21 receptor (IL21R) gene SNPs, rs8057551, rs8061992, and rs7199138, showed consistent association with femoral neck BMD in both a GWAS discovery sample of 983 unrelated Caucasian subjects and a family-based replication sample of 2,557 Caucasian subjects, with combined P values of 2.31 × 10−6, 8.62 × 10−6, and 1.41 × 10−5, respectively (Guo et al. 2009). Expression of IL21R correlates negatively with the destruction of cartilage and bone (Jüngel et al. 2004).
New BMD loci revealed by the meta-analysis of five GWASs
The meta-analysis of five GWASs confirmed eight known BMD loci (LRP5, ESR1, OPG, RANK, RANKL,SP7, ZBTB40, and FOXC2) at the genome-wide significance level. In addition, it identified 12 new BMD loci; 8 have not been previously associated with BMD [1p31.3 (GRP177), 3p22 (CTNNB1), 5q14 (MEF2C), 7p14-p13 (STARD3NL), 7q21.3 (FLJ42280), 11p15 (SOX6), 11p14.1 (DCDC5; DCDC1), and 17q12-q22 (CRHR1)] and 4 have been suggested to associate with BMD in previous GWASs [2p21 (SPTBN1), 4q21.1 (MEPE), 11p11.2 (LRP4), and 17q21 (HDAC5; C17orf53)] [SOX6 was discussed in “SOX6”]. Related pathways, monogenic syndromes, knockout mouse models and functional data with respect to these novel candidate genes have been described in the original report (Rivadeneira et al. 2009). Four genes that are involved in the biological pathways described in “Osteoporosis susceptibility loci initially analyzed by candidate gene studies” will be highlighted below.
SNPs within and close to the LRP4 locus at 11p11.2 showed suggestive association with hip BMD in the GWAS of Styrkarsdottir et al. (2008). This locus contains many genes in the same LD block, including LRP4, F2, ARHGAP1, and CKAP5. SNPs within this block were associated with femoral neck BMD at the genome-wide significance level in both the meta-analysis of five GWASs (Rivadeneira et al. 2009) and the collaborative meta-analysis of Richards et al. (2009).
Two common SNPs in complete LD were associated with both femoral neck and lumbar spine BMD. The two SNPs are located within an intron of the GPR177 (G-protein-coupled receptor 177, alternatively known as WNTLESS homolog) gene, a component of the Wnt-signaling pathway (see “The Wnt/β-catenin signaling pathway”).
rs87939, which is located 103 kb upstream of the CTNNB1 gene, was associated with femoral neck BMD. CTNNB1 encodes β-catenin, an integral part of the Wnt/β-catenin signaling pathway (see “The Wnt/β-catenin signaling pathway”).
rs11898505, located in an intron of the SPTBN1 (spectrin, β, non-erythrocytic 1) gene, was associated with lumbar spine BMD. Disruption of the adaptor protein ELF, a beta-spectrin, results in the disruption of TGFβ signaling by Smad proteins (see also “The transforming growth factor-β (TGFβ) superfamily”) in mice (Tang et al. 2003).
New loci for other osteoporosis-related phenotypes
Liu et al. (2008b) performed the first GWAS for hip bone size, one of the risk factors for hip fracture. After testing ~380,000 SNPs in 1,000 homogenous unrelated Caucasians comprised of 501 females and 499 males, they found that the phospholipase c-like 1 (PLCL1) gene had four SNPs that achieved or approached the genome-wide significance level in the female subjects. “Gene-wise” rather than “SNP-wise” replication was achieved in both an independent UK cohort comprising 1,216 Caucasian females and in a Chinese sample with 403 females. The PLCL1 gene product can inhibit IP3 (inositol 1,4,5-triphosphate)-mediated calcium signaling, an important pathway that regulates the response of bone cells to mechanical signals (Liu et al. 2008b).
After analyzing ~379,000 eligible SNPs in 1,000 Caucasians, Zhao et al. (2009) identified a common intronic SNP, rs7430431, in the RTP3 gene that was strongly associated with buckling ratio and femoral cortical thickness (CT), two indices of femoral neck bone geometry. The association was successfully replicated in 1,488 independent Caucasians and 2,118 Chinese subjects (Zhao et al. 2009). The SNP is located in 3p21, a region previously linked with CT (LOD = 2.19, P = 0.0006) in 3,998 subjects from 434 pedigrees (Xiong et al. 2006a). A role of RTP3 in regulating bone geometry has yet to be established.
GWASs of copy number variations
Copy number variation (CNV) refers to a DNA segment of 1 kb or larger that is present in different copy numbers with respect to a reference genome sequence (Scherer et al. 2007). CNVs affect ~12% of the human genome (Redon et al. 2006) and capture ~18% of the total detected genetic variation in gene expression (Stranger et al. 2007). They have also been implicated in the etiology of human complex conditions such as HIV-1/AIDS susceptibility (Gonzalez et al. 2005), glomerulonephritis (Aitman et al. 2006), and autism (Glessner et al. 2009; Sebat et al. 2007). Two new osteoporosis candidate genes, UGT2B17 and VPS13B, have recently been identified by means of GWAS of CNVs.
Higher UGT2B17 copy number was found to be associated with lower BMD, thinner cortical thickness, higher buckling ratio and increased risk of osteoporotic fractures at the hip in both Chinese and white populations by means of the Affymetrix 500 K Array Set (Yang et al. 2008). UGT2B17 encodes a critical enzyme for the local inactivation of androgens, a major source for estrogen, which directly stimulates bone formation. Furthermore, subjects with zero copies of UGT2B17 had significantly higher total testosterone and estradiol concentrations than subjects with one or two copies of the gene. Therefore, Yang et al. (2008) proposed a physiologically plausible mechanism suggesting that increased UGT2B17 gene dosage may impair bone formation via its inhibitory effect on androgen.
A GWAS of CNVs for BMD and femoral neck cross-sectional genometric parameters was performed in 1,000 unrelated Caucasian subjects. A CNV involving the VPS13B gene was associated with higher bone strength (Deng et al. 2009). In support of this genetic finding, inactivating mutations in the VPS13B gene including an intragenic deletion spanning exons 6–16 cause Cohen syndrome, which involves multiple skeletal abnormalities (Bugiani et al. 2008; Hennies et al. 2004; Kolehmainen et al. 2003).
A combination of gene expression profiling and GWAS
Gene expression analysis proved to be critical in identifying Alox15 as an important negative regulator of bone mass in mice (Klein et al. 2004). Studies of gene expression in human circulating monocytes or B cells isolated from subjects with low and high BMD or peak bone mass revealed a dozen of differentially expressed genes that may potentially contribute to bone metabolism (Lei et al. 2009b; Liu et al. 2005; Xiao et al. 2008). Microarray analysis of gene expression in primary cultures of osteoblasts isolated from osteoporotic and non-osteoporotic human bone tissue samples identified a novel list of genes and metabolic pathways that may have relevance for the pathogenesis of osteoporosis (Trost et al. 2009). Differentially expressed genes were also found in bone biopsies related to BMD and/or osteoporosis (Balla et al. 2008, 2009; Hopwood et al. 2009; Reppe et al. 2009). The first combination of gene expression analysis with GWAS has recently led to the identification of a new candidate gene for osteoporosis (Chen et al. 2009a). The expression of 168 genes related to cytokines, chemokines, osteoclast formation factors and corresponding receptors was investigated in monocytes from 26 Chinese and 20 Caucasian premenopausal women with extremely discordant BMD. Only the STAT1 (signal transducer and activator of transcription 1) gene was found to be significantly upregulated in the low versus the high BMD groups. Genotyping of 1,000 unrelated Caucasians with the Affymetrix Mapping 250 k Nsp and 250 k Sty arrays found that two SNPs in the STAT1 gene were associated with spine BMD (Chen et al. 2009a). Based on these findings and previous knowledge, the authors proposed a novel mechanism for osteoclastogenesis: in peripheral blood, the STAT1-mediated interferon (IFN) pathway may stimulate circulating monocytes to produce cytokines such as IL1, TNF, CXCL10 and IL15, and these molecules then increase bone resorption function of osteoclasts (Chen et al. 2009a).
Using the recently proposed “pathway-based approaches for analysis of GWASs (Wang et al. 2007b)”, Chen et al. (2009b) studied the joint effects of genes involved in given biological pathways on femoral neck bone geometry variations in 1,000 unrelated US whites. Seventy-six pathways showed nominal significant association with section modulus (Z), an index of bone bending strength. The most significant association was with the EphrinA–EphR pathway, which plays an important role in bone homeostasis (Edwards and Mundy 2008). The association remained significant after multiple testing adjustments and was replicated in samples from the Framingham Osteoporosis Study. Of the 35 genes in the EphrinA–EphR pathway, 21 were associated with femoral neck Z at a nominal significance level (Chen et al. 2009b). Notably, none of the 21 genes were shown to be associated with femoral bone geometry at the genome-wide significance level in the GWAS that analyzed the same 1,000 unrelated US whites (Zhao et al. 2009). This finding demonstrated the usefulness of the pathway-based GWAS in detecting moderate genetic risk factors that might have often been missed by the conventional GWAS approach.
Conclusions and perspectives
The achievements in osteoporosis genetics in a period of 15 years, particularly in the past 4 years, are impressive. Although the majority of associations remain to be replicated and validated, it is clear that osteoporosis susceptibility is conferred by a large number of genetic variants and each one has a modest effect. This is best illustrated by the study of Rivadeneira et al. (2009): 15 top SNPs associated with lumbar spine BMD and 10 top SNPs associated with femoral neck BMD explained only ~2.9% and ~1.9% of the BMD variances at these two skeletal sites, respectively.
We classified the reported genes into the following different categories: established genes, promising genes, and inconclusive or putative associations based upon a combined consideration of the currently available data (Table 1). Nevertheless, this classification will be subjected to constant revision and amendments due to the availability of new association, replication, and functional analytic data. Interestingly, established as well as promising susceptibility genes are clustered in three biological pathways, the estrogen endocrine pathway, the Wnt/β-catenin signaling pathway, and the RANKL/RANK/OPG pathway. Other genes may be classified into these biological pathways once a clearer picture of their functional roles is developed. New biological pathways may also emerge if an extensive meta-analytic work is performed.
Although GWAS is the most powerful approach for investigating the genetic basis of osteoporosis, well-designed CGASs (e.g., study of well-selected cases and controls) remain an attractive and efficient way to identify new susceptibility genes/variations. We surmise that the candidate gene approach may be used more frequently either along or in combination with fine-mapping of known osteoporosis loci once the costs of high-throughput sequencing are reduced. Integration of linkage, gene expression, biological function, association data and meta-analytic analysis should help resolve inconsistencies pertaining to the role of some genes in the pathogenesis of osteoporosis.
In most studies, the associated variants have not been functionally characterized. Functional analysis of these variants may help trace down the truly causative genes/variants and provide new mechanistic insights. In addition, most studies, in particular GWASs, only genotyped variants with minor allele frequency ≥0.05. This left the “common disease, rare variants” theory almost completely untested in the etiology of osteoporosis. While deep resequencing of some confirmed osteoporosis genes in patients may provide a quick answer to this question, we surmise that both common and rare variants contribute to osteoporosis. Furthermore, gene–gene interactions, gene-environment interactions, and CNV-SNP interactions still need to be tackled.
The ultimate promise of osteoporosis genetics is not only to better understand the disease process, but more importantly, to lead to better therapeutic and preventive interventions. In this regard, denosumab (a fully human monoclonal antibody to RANKL) was recently reported to significantly reduce the risk of vertebral, non-vertebral, and hip fractures in women with osteoporosis (Cummings et al. 2009).
We regret that, owing to space limitations, we could not include all of the relevant work and references. This work was supported by the National Natural Science Foundation of China (Grant number 30672133); and the INSERM (Institut National de la Santé et de la Recherche Médicale), France.