Molecular Genetic Studies of Gene Identification for Osteoporosis: A 2004 Update


  • Yong-Jun Liu,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Hui Shen,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Peng Xiao,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Dong-Hai Xiong,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Li-Hua Li,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Robert R Recker,

    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    Search for more papers by this author
  • Hong-Wen Deng

    Corresponding author
    1. Osteoporosis Research Center, Creighton University Medical Center, Omaha, Nebraska, USA
    2. The Key Laboratory of Biomedical Information Engineering of Ministry of Education and Institute of Molecular Genetics, School of Life Science and Technology, Xi'an Jiaotong University, Xi'an, China
    3. Laboratory of Molecular and Statistical Genetics, College of Life Sciences Hunan Normal University, Changsha, Hunan, China
    • Hong-Wen Deng, PhD Osteoporosis Research Center Creighton University North 30th Street, Suite 6787 Omaha, NE 68131, USA
    Search for more papers by this author

  • The authors state that they have no conflicts of interest.

  • Published online on October 10, 2005;


This review summarizes comprehensively the most important and representative molecular genetics studies of gene identification for osteoporosis published up to the end of December 2004. It is intended to constitute a sequential update of our previously published review covering the available data up to the end of 2002. Evidence from candidate gene association studies and genome-wide linkage studies in humans, as well as quantitative trait locus mapping animal models are reviewed separately. Studies of transgenic and knockout mice models relevant to osteoporosis are summarized. An important extension of this update is incorporation of functional genomic studies (including DNA microarrays and proteomics) on osteogenesis and osteoporosis, in light of the rapid advances and the promising prospects of the field. Comments are made on the most notable findings and representative studies for their potential influence and implications on our present understanding of genetics of osteoporosis. The format adopted by this review should be ideal for accommodating future new advances and studies.


We recently reviewed the important findings made in the field of genetics of osteoporosis, which summarized data up to the end of 2002.(1) Since its publication, the review has received considerable attention and is of help for osteoporosis research. This constitutes strong encouragement for us to review the literature in extenso periodically to provide an up-to-date and objective evaluation on the endeavors to identify genetic factors underlying risk of osteoporosis. In this article, we aim to provide an update to our previous review(1) to summarize the molecular genetics studies published within 2003–2004. The data presented in this review were collected by systematic search through PubMed. The key words of “bone,” “BMD,” “osteoporosis,” “polymorphisms,” “linkage,” “QTL,” “osteogenesis,” “microarray,” “proteomics,” etc. were used in the search.

As with our first review,(1) this synthesis includes sections dealing with association studies and linkage studies in human populations, quantitative trait loci (QTL) mapping in animal models, and transgenic and knockout murine models relevant to osteoporosis. A notable extension is inclusion of functional genomic studies, owing to the rapid progress in the field and its promising prospects in understanding biochemistry of proteins, processes, and pathways relevant to osteogenesis and osteoporosis. The results of important studies are entered in the tables for comparison and for ease of reference. Table 1 (cited online) summarizes the major results from ∼170 reported candidate gene association studies. Because transgenic/knockout animal models continue to reveal novel pathways and candidate genes for bone mass regulation, we list in Table 2 (cited online) the studies that have appeared over the past few years. Genome-wide linkage scans in humans and QTL mapping in model organisms are elaborated on and summarized individually in Tables 3 and 4, respectively. Table 5 (cited online) incorporates DNA microarray studies on osteogenesis, osteoporosis, and other bone-related diseases. For findings and studies reported before 2002, the readers may refer to our first review.(1)

Table Table 3.. Evidence for the Presence of Linkage with Osteoporosis-Related Phenotypes*
original image
Table Table 4.. QTLs Identified in Animal Polygenic Models of Osteoporosis with Their Homologous Human Location
original image

Several articles are available that address the strategies(2–4) and status of genetic study of osteoporosis.(5–14) We recently overviewed the powerful and promising methodologies in the study of complex bone disorders at the whole genome level.(15) In addition, we systematically addressed the confounding factors that cause nonreplication in gene mapping of osteoporosis and proposed tentative remedies.(16) Given that, a detailed discussion on the above aspects will not be attempted here. Because of space constraints, we are unable to reference all original studies in this review. With apologies to the omitted authors, the reader is referred to other reviews for detailed descriptions.


During the past two years, ∼170 association studies were published (Table 1, cited online). The candidate genes involve classical genes (e.g., VDR, ER-α, and COL1A1) that have been extensively studied, as well as novel genes (e.g., LRP5 and SOST) recently discovered to be important in bone and mineral metabolism. The newly studied genes include CYP17 (17-α-hydroxylase),(17)CYP1B1 (cytochrome P450),(18)DBP (vitamin D–binding protein),(19)GH1 (growth hormone 1),(20)GnRH (gonadotropin-releasing hormone 1),(21)IGF-II (insulin-like growth factor II),(22)LEPR (leptin receptor),(23)LRP5 (low-density lipoprotein receptor–related protein 5),(24)BMP2 (bone morphogenetic protein 2),(25)CCR2 (chemokine),(26)CLCN7 (chloride channel 7),(27)COMT (catechol-O-methyltransferase),(28)CTSK (cathepsin K),(29)DRD4 (dopamine receptor D4),(30)I-TRAF (TRAF family member-associated NF-κB activator),(31)LCT (lactase),(32)MIF (macrophage migration inhibitory factor),(33)MMP-1 (matrix metalloproteinase 1),(34)MMP-9 (matrix metalloproteinase 9),(35)NCOA3 (nuclear receptor co-activator 3),(36)NPY (neuropeptide Y),(37)OSCAR (osteoclast-associated receptor),(38)PLOD1 (procollagen-lysine, 2-oxoglutarate 5-dioxygenase),(39)PON1 (paraoxonase 1),(40)RIL (LIM domain protein RIL),(41)SERT (serotonin transporter),(42)SOST (sclerostin),(43) and TCIRG1 (T cell immune regulator 1).(44)

In the following, we highlight some studies performed in samples of at least 1000 subjects. This is because statistical power is among the foremost factors for robust and replicable results, and generally, at least 1000 subjects are needed to detect modest genetic effects (e.g., a QTL explaining 5% of phenotypic variation) in a population-based association study.(16) Some meta-analyses were addressed, because meta-analyses, by combining results across studies, are helpful to solve the problems of underpowered studies, revealing unexpected sources of heterogeneity, and resolving discrepancies in genetic studies.(45) Other association studies are cited in Table 1 (cited online). For the classical candidate genes, their potential physiological effects on bone metabolism and pathophysiological implications to osteoporosis have been elaborated elsewhere(1,14); for the novel genes, their potential functions will be briefly outlined.

Classical candidate genes


Association between the VDR gene and osteoporosis-related traits has been extensively studied.(1) The frequently studied markers include BsmI, ApaI, TaqI, and FokI, which currently have unknown functional effects.(1) A novel polymorphism at the Cdx-2 (an intestine-specific homeodomain-containing transcription factor) binding site of the promoter region was identified to be able to activate VDR gene transcription and was associated with BMD variation in a Japanese population.(46)

Yamada et al.(25) studied the 2C-T (i.e., FokI) polymorphism with BMD variation in ∼1100 Japanese women and ∼1100 Japanese men. BMD did not differ among the 2C-T genotypes for all women, whereas men bearing the CT genotype had lower BMD than those with the CC genotype at the distal radius, femoral neck, and Ward's triangle (p ≤ 0.05).(25)

Morita et al.(47) randomly selected 50 women from each of the 5-year age-stratified groups (15–79 years) in three Japanese municipalities, that is, 650 subjects for each area and 1950 in total. After excluding subjects who had medical or menstrual histories affecting BMD, 1434 women were analyzed for ApaI, TaqI, and FokI polymorphisms. TaqI was significantly associated with the distal one-third radius BMD in premenopausal women (p = 0.019). Analyzing three major combined genotypes (aaTT, AaTT, AaTt) of ApaI and TaqI revealed even stronger association for the distal one-third radius BMD in premenopausal women (p = 0.009). However, analyses on major haplotypes (AT or aT) failed to detect any significant association. In addition, they tested the relationship between the three polymorphisms and BMD change over 3 years in 976 subjects.(47) The annual percent changes in lumbar spine BMD of the TaqI tt subjects were different from other genotypes in women who were premenopausal at the follow-up survey and in the women who were postmenopausal at the baseline survey. However, the effect of the tt genotype on BMD change was opposite in the two groups. The authors concluded that none of the individual polymorphisms were consistently associated with baseline BMD or BMD change, and hence, the effect of the VDR genotype on BMD is negligible in Japanese women.

Fang et al.(48) examined Cdx-2 with BMD and risk of fracture in a cohort of 2848 Dutch ≥55 years of age. They did not find significant association for BMD, but they detected borderline association with vertebral fracture (p = 0.04) and any type of fractures (p = 0.06) with subjects carrying Cdx-2 A allele having reduced relative risk (RR) of fracture by 20%.(48) The protective effect of the A allele was similar in women and men.

Thakkinstian et al. reported two meta-analyses.(49,50) One study focused on the relationship between VDR BsmI and BMD/osteoporosis at the femoral neck or spine in adult women.(49) In that study, BsmI was associated with spine BMD in postmenopausal (p = 0.028) but not premenopausal women. The association was modest and followed a recessive model, with the BB genotype having lower BMD than Bb/bb genotype. The magnitude of the decrease in spinal BMD by BB genotype was 2.4%, which translated into a population attributable risk of spine fracture of 1.98%.(49) They also studied the association between BsmI and mean percent BMD change over time. Analyses revealed a significant effect (p = 0.017), with BB and Bb genotypes having greater bone loss per year than the bb genotype. The other meta-analysis was conducted with data on BsmI, ApaI, and TaqI, as well as on haplotypes defined by them.(50) Although none of the individual polymorphisms was associated with osteoporosis on their own, they found significant association between spinal osteoporosis and haplotypes BAT (p < 0.001) and BaT (p = 0.031), with ORs of ∼4. For spine BMD, the only association was found for BsmI in postmenopausal Asian women (p < 0.001), suggesting that the genetic effect of VDR BsmI could be population-, menstrual status–, and site-dependent. This was in agreement with their previous meta-analysis.

Based on the available data, Morrison wrote a commentary on association between VDR and BMD.(51) By using the data provided in one meta-analysis mentioned above,(49) the author indicated that the effect size of the BsmI polymorphism on BMD was ∼0.11–0.13 SD in postmenopausal women. To detect such modest effects, large sample sizes are required. For instance, based on the estimated allele frequency, 3046 whites or 4700 Japanese are required to have 80% power to detect the effect of 0.13 SD at p = 0.01.

Recently, two studies assessed the linkage disequilibrium (LD) pattern by examining multiple single nucleotide polymorphisms (SNPs) across the gene.(52,53) Nejentsev et al.(53) resequenced the gene and genotyped 55 common SNPs (minor allele frequency > 10%) in four European populations and one African population. LD patterns were identical (with three block-like regions) in all four European populations, but with two additional LD-breaking spots in an African population. In another study, Fang et al.(52) sequenced 22 kb of the gene (including the promoter, all exons, and the 3′ UTR) and identified 62 SNPs. LD analyses on common SNPs revealed four to eight haplotype blocks, which were less fragmented in whites and Asians than in Africans. They subsequently tested 15 tagging SNPs with BMD and fracture risk in 6535 elderly whites. Two haplotypes (containing the Cdx-2 SNP and 3′UTR, respectively) were associated with increased risk for vertebral fractures or overall osteoporotic fractures. The combined risk alleles showed 46% increased risk for vertebral fracture (p = 0.03) and 34% increased risk for osteoporotic fracture (p = 0.01).(52) This whole gene analysis suggested that VDR polymorphisms in the promoter region and 3′UTR may interactively affect fracture risk. In another study, the same group found tagging SNPs encompassing the 3′UTR of the VDR gene interacted with a SNP in the lactase-phlorizin hydrolase (LPH) gene, a variant correlated to lactose intolerance, to influence height, bone geometry, and BMD.(54)

The LD and haplotype-based association studies have several potential advantages.(55) First, haplotypes reduce the dimension of association tests and may gain statistical power because of increased information by combining multiple SNPs. Second, genetic variation in populations is intrinsically organized into haplotypes. Third, the folding kinetics, stability, or other physical properties of proteins may depend on interactions between pairs or higher-order combinations of amino acid sites. If these interactions are important, haplotypes are of direct biological relevance. For the observed associations with haplotype(s), it is necessary to assess their functions.

To date, >150 studies have been reported on associations between VDR gene polymorphisms and bone-related traits. Based on the available data, a tentative conclusion is that VDR gene polymorphisms, individually or interactively, may have effects on a number of biological endpoints, including BMD variation and bone fractures. Moreover, VDR may interact with nongenetic factors (i.e., calcium intake and estrogens) to modulate BMD. Although the effects of the VDR gene could be modest, they can still confer significant influence in the general population with their high frequencies. A similar example is the PPAR-γ Pro12Ala polymorphism, which is associated with a modest (1.25-fold) increase in diabetes risk.(56) Because the risk allele of Pro12Ala is so common in human populations, its modest effect translated into a large population attributable risk—influencing as much as 25% of type 2 diabetes in the general populations.(56)

The interpretation of VDR polymorphisms is currently hindered by the fact that most studies have investigated only a limited polymorphisms (e.g., BsmI, ApaI, TaqI, and FokI), which have unknown effects. The whole gene analyses that explore all potential sequence variations within/around the gene will be helpful to identify the functional variants. It is also imperative to clarify the molecular mechanisms underlying the observed associations by conducting functional studies using well-defined cell types and/or animal models.


Three ER-α common polymorphisms, PvuII (T→C) and XbaI (A→G) in intron 1 and the TA repeat in the promoter region, have been extensively studied. The TA repeat polymorphism was speculated to affect bone mass by changing mRNA production or stability, whereas the functional relevance of PvuII and XbaI on bone mass remains to be elucidated. A possible mechanism is that the two polymorphisms are in LD with nearby functional variant(s), resulting in positive associations.

Yamada et al.(57) examined the relationship between PvuII and XbaI and BMD in ∼2230 Japanese subjects 40–79 years of age. In women ≥60 years old, XbaI alone or in combination with PvuII showed significant association with femoral neck BMD. Zhao et al.(58) examined eight SNPs spanning the ER-α gene in 405 white nuclear families (1873 subjects). Marginal evidence was observed for femoral neck BMD with rs932477 (p = 0.015) and rs2228480 (p = 0.010). The most common seven-SNP haplotype (TCGCGGG) was associated with higher lumbar spine BMD (p = 0.015). Another study in 401 Chinese nuclear families (1260 subjects) reported a within-family association (p = 0.054) between spine BMD and the XbaI genotypes.(59)

Van Meurs et al.(60) studied the association of PvuII, XbaI, and TA repeat polymorphisms with BMD, vertebral bone area, and fractures in 2042 elderly Dutch whites. In women, subjects homozygous for haplotype px and a low number of TA repeats have significantly lower lumbar spine BMD (p = 0.003 and 0.008, respectively) and decreased vertebral bone area (p = 0.016) than homozygous noncarriers. They also found an increased vertebral fracture risk with an allele dose effect of OR = 2.2 for haplotype px and OR = 2.0 for a low number of TA repeats.

A meta-analysis was reported for PvuII, XbaI, and promoter TA repeats with BMD and fractures in 18,917 individuals from eight European research centers.(61) None of the three polymorphisms or haplotypes thereof had statistically significant association with BMD. However, there was a highly significant protection conferred by the XbaI XX genotype against fracture risk. In women with the XX genotype, the OR was 0.81 (p = 0.002) for any fractures and 0.65 (p = 0.003) for vertebral fractures. The observed effects on fractures were independent of BMD.

As with the VDR gene, the individual contribution of ER-α polymorphisms to osteoporosis remains to be universally confirmed. Future endeavors will be to elucidate their functional molecular relevance and their interaction with the environment in the causation of osteoporosis. A promising application of these polymorphisms is their pharmacogenomic implications, with the possibility of providing better guidance for therapeutic regimens, such as estrogen replacement therapy and selective ER modulators.


Thirteen association studies were reported over the past 2 years for the COL1A1 gene, many focusing on the Sp1 polymorphism, a G→T substitution at the first base of a consensus site in the first intron for the transcriptional factor. It is notable that association of COL1A1 Sp1 with osteoporotic fractures is among those mostly replicated.(62)

In 1044 elderly Swedish women, Gerdhem et al.(63) found association between COL1A1 Sp1 and femoral neck BMD (p = 0.027) and prevalent wrist fracture (p = 0.024). Women carrying at least one copy of the s allele had lower femoral neck BMD and higher prevalence of wrist fracture. The ORs for prevalent wrist fracture were 2.73 and 1.4 for ss and Ss genotypes, respectively, suggesting the pronounced effect of the s allele on increasing risk of wrist fractures. In 401 Chinese nuclear families, Zhang et al.(64) found that a −1997 G/T polymorphism in the COL1A1 upstream regulatory region was associated with hip BMD, explaining 1.6% of the total hip BMD variation. By testing three COL1A1 SNPs in 405 white nuclear families, Long et al.(65) reported that subjects bearing the T allele of SNP2 had, on average, 3.05% smaller wrist size than noncarriers.

Mann and Ralston(66) performed a meta-analysis for COL1A1 Sp1 and BMD/osteoporotic fractures, which involved 7849 subjects from 26 published studies. The Ss genotype group had significantly lower lumbar spine BMD than the SS group (p = 0.00005), but the difference between the SS and ss groups in spine BMD did not reach a significant level (p = 0.13). The femoral neck BMD were lower in the Ss group (p < 0.00001) and in ss group (p = 0.001) versus the SS group. They also found increased OR for any fracture in Ss subjects (OR = 1.26, p = 0.002) and an even greater increase in ss subjects (OR = 1.78, p = 0.003). Subgroup analysis showed that increased risk was largely attributable to vertebral fracture where the OR was 1.37 (p = 0.0004) and 2.48 (p < 0.00001) for Ss and ss subjects, respectively. Their results suggested that the COL1A1 Sp1 alleles contribute to a modest reduction in BMD and a significant increase in risk of osteoporotic fractures, particularly vertebral fractures.

The functional importance of Sp1 has been studied.(67) The s allele had greater affinity for the Sp1 protein compared with the S allele. In the Ss heterozygotes, the RNA transcripts derived from the s allele were three times more abundant than those from the S allele, suggesting allele-specific transcription or a different splicing process for the two alleles. The yield strength of bone derived from Ss individuals was reduced compared with bone derived from SS subjects.(67) With the availability of SNP information of the human genome, together with the ongoing HapMap project,(68) further studies are necessary to identify more causative variants in the COL1A1 gene using the LD mapping approach.

Other classical candidate genes

MTHFR (5,10-@methylenetetrahydrofolate reductase) affects the methylation of homocysteine to methionine, and high serum homocysteine concentrations have adverse effects on bone.(69,70) A MTHFR polymorphism, C677T, causes an alanine to valine substitution and gives rise to a thermolabile variant of MTHFR with reduced activity.(71)MTHFR C677T was associated with elevated levels of circulating homocysteine(72) and lumbar spine BMD.(73) In a study(74) using 1748 healthy postmenopausal Danish women, the TT genotype of the MTHFR C677T polymorphism was associated with lower BMD at the femoral neck, total hip, and spine (p < 0.05), with the effect sizes ranging from 0.1 to 0.3 SD. In consistency with BMD, the fracture incidence was increased >2-fold in subjects with the TT genotype. Although this variant did not affect the response to hormone replacement therapy (HRT), the association of the TT genotype with lower BMD was maintained at the total hip after 5 years of HRT.

Because the effect of C677T on circulating homocysteine levels was dependent on plasma folate concentration,(75) one study in 1632 whites evaluated whether the folate status may modify the association between C677T and BMD or quantitative ultrasonography (QUS) parameters.(76) The results showed that adjusted mean QUS parameters and BMD measurements did not significantly differ between C677T groups. However, suggestive interactions between folate status and the C677T group (CC + CT versus TT) were found for hip BMD (p ≤ 0.05) and for one of the QUS parameters, broadband ultrasound attenuation (BUA; p = 0.11). In subjects with low folate concentration (<4 ng/ml), the TT group had lower mean BUA (p = 0.06) and Ward's area BMD (p = 0.08) compared with the CC + CT group; whereas in subjects with high folate levels (≥4 ng/ml), the TT group had significantly higher hip BMD (p ≤ 0.05). Because dietary B vitamins (folate, vitamin B12, vitamin B6, and riboflavin) can also influence circulating homocysteine levels, a study further exploited the association between C677T and BMD and bone loss, in relation to B vitamin intake in a cohort of perimenopausal and early postmenopausal Scottish women.(77) Although no association was observed for BMD, bone loss, or biochemical markers of bone turnover, there was a significant interaction between C677T and riboflavin intake in relation to BMD. BMD was lower for the TT group at low intakes of riboflavin compared with the other genotype, but at high intakes, BMD was higher in the TT group. The results suggested that the association between C677T and bone phenotypes might depend on B vitamin levels, especially on folate and riboflavin status. The mechanism underlying this modification remains unclear.

ApoE is a lipid-transporting glycoprotein that exists in three isoforms, encoded by three alleles, ApoE2, ApoE3, and ApoE4. A large population-based association study tested the effects of ApoE on BMD, bone loss, and osteoporotic fractures in 5857 Dutch subjects. No significant association was found with baseline BMD at the lumbar spine and femoral neck or with bone loss in a mean follow-up of 2.0 years. During a mean follow-up of 6.6 years, 708 nonvertebral fractures and 149 incident vertebral fractures occurred, but no significant association was observed between ApoE polymorphisms and incidence of these fractures. Another study(78) tested whether ApoE polymorphisms interact with vitamin K transport or with dietary fat to influence BMD and bone loss in 2481 early postmenopausal Scottish women. The only association was found between BMD change at femoral neck (p = 0.023) in alcohol drinkers (median alcohol intake = 7.3 g/day). A study(79) in 387 white nuclear families reported that APOE haplotypes defined by four SNPs may influence BMD in white males but not females.

IGF-I, by stimulating the proliferation of chondrocytes in the growth plate, plays an essential role in longitudinal bone growth. It is also involved in the formation of trabecular bone and is essential for coupling matrix biosynthesis to sustain mineralization. Plasma IGF-I levels were associated with BMD and osteoporotic fractures.(80,81) One study examined the role of a CA repeat polymorphism in the promoter region with hip BMD in a group of elderly Dutch women and men.(82) A total of 5648 and 4134 individuals underwent examination at baseline and a 2-year follow-up, respectively. Lower baseline BMD and higher BMD loss were associated with the absence of the 192-bp allele in women (p = 0.03), but not in men. This polymorphism only explained a minor portion of the variance in BMD (0.2%) and BMD change (0.1%) among females. The same group conducted another study evaluating association of the CA repeat polymorphism with incidence of nonvertebral fractures in 2799 men and 4212 women.(83) They also estimated the effects of this polymorphism on several hip bone geometry parameters, including neck width, cortical thickness, buckling ratio, and section modulus.(83) Women who were noncarriers or heterozygotes of the 192-bp allele had increased risk (1.5 and 1.2, respectively) of osteoporotic fractures compared with homozygotes for this allele (p = 0.0007). This effect was not observed in men. For hip geometry parameters, noncarrier males had narrower femoral neck and lower section modulus (p < 0.05) than homozygote men, whereas noncarrier females had thinner cortices and higher buckling ratios (p < 0.05) but no significant differences in femoral neck width and section modulus. The observed genotype-dependent differences in fracture risk cannot be fully explained by the genotype-dependent effects on hip bone geometry.

IL-6 is a pleiotropic cytokine that promotes the differentiation of osteoclast precursor cells into mature osteoclasts.(84) Yamada et al.(25) examined the −634C-G polymorphism of the IL-6 gene alone and in combination with the 298C-T polymorphism of the osteocalcin gene in ∼2200 Japanese subjects (∼1100 men and ∼1100 women). Both polymorphisms were associated with BMD of total body and lumbar spine in postmenopausal women (p < 0.05). Analyses with combined genotypes suggested the two polymorphisms exert an additive effect on BMD in postmenopausal women. Another study in 3376 white women ≥65 years of age reported association for distal and proximal radius BMD with the IL-6 G174C polymorphism (p = 0.016 and 0.049, respectively).(85) The risk of wrist fractures decreased by 17% per copy of C allele significantly (p = 0.043). Moreover, compared with women having the GG phenotype, women having the CC genotype also had slower rates of bone loss in the total hip and femoral neck in ∼3.5 years of follow-up and 33% lower risk of wrist fractures over an average of 10.8 years.

Several other classical candidate genes have also been studied. In ∼1300 Australian elderly women, a study examined A986S polymorphism of the CASR gene and T986C polymorphism of the TGF-β gene in relation to BMD, calcaneal QUS, and prevalent and incident fracture.(86,87) The TGF-β C allele was associated with lower BMD of the total hip, femoral neck, and trochanter, as well as QUS parameters. This allele was also associated with an increase in osteoporosis (T score ≤ −2.5 SD) and fracture risks.(87) In contrast, the CASR A986S polymorphism was not associated with any parameters.(86) Shearman et al.(88) assessed four intronic SNPs and one microsatellite marker near the ER-β gene in 723 men and 795 women and observed association with hip BMD in both men and women. OPG (a decoy receptor of RANKL) neutralizes biological activity of RANKL that functions as a potent agonistic ligand through binding to its receptor, RANK, on osteoclast lineage cells.(89) Two polymorphisms in the promoter of the OPG gene were examined in ∼1100 Japanese women and ∼1100 Japanese men.(90) The association was only found in women for BMD at various skeletal sites.(90) In a longitudinal study including 1792 women with or without HRT, two polymorphisms of the CYP19 gene were associated with the magnitude of BMD increase in response to HRT.(91) Albagha et al.(92) assessed the relationship between variants of the TNFRSF1B (i.e., TNFR2) gene and BMD in 1240 perimenopausal women from the United Kingdom. Notably, the TNFRSF1B gene is located on chromosome 1p36, a region linked to BMD in several genome-wide linkage studies.(1) Four polymorphisms, one in exon (T676G) and three in 3′ UTR (G593A, T598G, and T620C), were selected based on their potential functions.(92) No association was found between T676G and spine or hip BMD, but a haplotype (A593-T598-C620) was associated with femoral neck BMD, even after correction for multiple testing (p < 0.0009).(92) The TNF-α G308A polymorphism was evaluated in relation to BMD, bone structural geometry, and fracture risk in 4402 elderly women.(93) The G308A polymorphism was associated with femoral neck subperiosteal width and endocortical diameter, as well as bone bending strength. This polymorphism was also correlated with hip fracture risks during a 12.1-year follow-up, which was independent of its effects on BMD or bone strength.

Novel candidate genes

LRP5 was recently found to be a key regulator of osteoblast proliferation and bone formation. LRP5 mutations resulted in high bone mass phenotypes or osteoporosis-pseudoglioma syndrome.(94,95) In 219 Korean men 20–34 years of age, Koh et al.(96) examined six missense variants (Q89R, A400V, V677M, R1036Q, A1330V, A1525V), which were first identified in whites, with BMD at the lumbar spine and hip. Four of them (A400V, V677M, R1036Q, A1525V) were not polymorphic in Korean men, and the only association was found between Q89R and BMD at Ward's triangle (p = 0.043). Ferrari et al.(97) examined the LD patterns using 13 previously reported SNPs in whites and selected 5 informative SNPs (a minor-allele frequency ≥ 5% and selected only one SNP for each pair of markers in nearly complete LD; r2 ≥ 0.9) for further association analysis. They ruled out potential population stratification using the program STRUCTURE.(98) Significant associations were found between 2047G-A (i.e., V677M) and lumbar spine BMD (p = 0.041), BMC (p = 0.0032), bone area (p = 0.0014), and stature (p = 0.0062). The observed associations with lumbar spine BMC and bone area were driven mainly by men. Haplotype analyses suggested that additional LRP5 genetic variants might also influence lumbar spine bone mass and area. They also found association between LRP5 haplotypes with 1-year bone mass and area changes in prepubertal males but not females. Taken together, the results suggested that LRP5 polymorphisms may influence lumbar spine bone mass and size in men, probably by affecting vertebral bone growth during childhood.(97) However, their claimed replication to a previous linkage reported by Koller et al.(99) might be problematic, because the linkage peak was detected in postmenopausal women,(99) but the association was found in males only.(97) Contrast to the male-specific effects observed by Ferrari et al.,(97) a study in 1301 elderly Australian women showed that the LPR5 gene was associated with hip bone mass and osteoporotic fractures.(100)

SOST is a disease-causing gene for sclerostenosis, a sclerosing bone dysplasia characterized by hyperostosis and overgrowth of normal bone tissue.(101) In a case-control study of 619 Scottish women with either high or low BMD (i.e., those in the upper or lower 16th percentile), Balemans et al.(43) identified 26 polymorphisms and selected 5 for association analyses. The distribution of genotypes and alleles for each tested SNP did not differ in the low or the high BMD group. Another study analyzed eight SOST polymorphisms in 1939 elderly Dutch men and women (55–80 years of age).(102) A 3-bp insertion in the promoter region (SRP3) was associated with decreased BMD in women at the femoral neck (p = 0.05) and lumbar spine (p = 0.01), with evidence of an allele–dose effect in the oldest age group (p = 0.006). The corresponding effect size between extreme genotypes was 0.2 SD. The polymorphism SRP9 was associated with femoral neck BMD (p = 0.007) and lumbar spine BMD (p = 0.02) in men, with the effect size between extreme genotypes of 0.2 SD. However, haplotype analyses failed to confirm the association observed in single SNP analyses. They further studied interactions between SOST polymorphisms and the VDR 3′UTR polymorphisms, as well as the COL1A1 Sp1 polymorphism.(102) An additive effect was observed between the SOST and COL1A1 Sp1 polymorphisms. The SOST–COL1A1 additive effect increased with age and reached 0.5 SD difference in lumbar spine BMD in the oldest age group (p = 0.02).

Giraudeau et al.(29) screened the CTSK gene in the promoters, exons, intron A, and all intron–exon boundaries in 130 random whites. Based on the LD pattern, they examined five CTSK polymorphisms and haplotypes thereof in relation to spine and hip BMD in ∼3000 perimenopausal Scottish women. None of the polymorphisms or haplotypes exhibited significant association.

A number of other newly recognized candidate genes are listed in Table 1 (cited online) including CCR2, COMT, CYP 17, DRD4, IRAK1, MMP-1, MMP-9, PON1, and PON2s. Most of them were evaluated in only one study and await further study.


Association results in the bone field are currently inconsistent/inconclusive. A few articles have addressed this from perspectives of study design, analytical methodology, and others.(16,103–107) In the following, we re-emphasize important issues that undermine the validity of association studies.

First, a large portion of studies are underpowered (Fig. 1), which may yield unreliable/unrepeatable results.(16,108) Meta-analysis is a powerful tool for pooling previous research when individual studies have insufficient power.(49,66) However, meta-analysis is not always a panacea, because it is prone to biases resulting from different ascertainment and diagnostic criteria and population stratification in different studies.(16) Moreover, there exists publication bias favoring positive results, which further exacerbates the situation. We suggest a mechanism to publish negative results in a brief report section of specialty journals, and storage and dissemination of all association data, perhaps in a widely accepted web site, such as the recently developed Genetic Association Database ( Second, limited number of polymorphisms was evaluated in the target gene(s). Given the complexity of LD and haplotype patterns across the human genome and among different populations,(109–112) this may not provide a comprehensive and accurate evaluation of the gene(s) of interest.(16,113) Third, given multiple markers and/or phenotypes tested in a single study, it is too liberal to set the significance threshold at p ≤ 0.05, which, however, has inappropriately been used in many studies. Traditional Bonferroni correction is overconservative, because of nonindependence of the tests. Other methods accounting for multiple testing of correlated hypotheses have been developed.(114–116) Fourth, population stratification, a confounding factor for association analyses,(117) was not appropriately assessed and controlled. Although the actual impact of population stratification on association studies has been a matter of some debate,(118–120) modest amounts of population stratification was indeed detected in case-control and case-cohort studies(121,122) and even in a relatively homogenous genetic isolate.(123) A family-based association approach (e.g., transmission disequilibrium test [TDT]) could be effective to deal with the problem.(124,125) For population-based studies, methods to control population stratification have been developed (e.g., Genomic Control(126) and Structured Association.(127)), which, bearing in mind some assumptions, merit application and further development. Finally, other factors such as heterogeneity, epistasis, and gene–environment interaction have posed special challenges to gene discovery. Endeavor to maximize sample homogeneity by recruiting subjects from the same ethnic group and rigorous controlling possible environmental or confounding factors may increase the chance of success. Studies have suggested that there exists sex-specific genetic contributions to BMD.(128,129) The genes regulating peak bone mass might differ from those regulating bone loss.(130) It is imperative for association studies to be well designed, statistically powerful, and appropriately interpreted.

Figure Figure 1.

Statistical power for quantitative traits using unrelated population sample, and the sample size distribution of osteoporosis-related association studies published during 2002–2004 (Table 1). In power estimation, QTLs were assumed to be responsible for 2% (♦ or ⋄), 1% (▪ or □), or 0.5% (▴ or ▵) of phenotypic variation, respectively. The power was estimated under an ideal situation, for which the tested marker was the QTL itself with the causal allele frequency of 0.3 and the QTL follows additive inheritance. The significance level was set as α = 0.01 (solid lines) or 0.001 (dashed lines). We set two different α levels because it was suggested that two studies with p < 0.01 or a single study with p < 0.001 is predictive of future replication.(36)

A critical question to be answered is the correlation between osteoporotic fractures (OFs) and osteoporosis risk factors. The search for OF genes should start with significant heritability for OFs and should include risk factors (e.g., BMD) that are genetically correlated with OFs. However, it is unclear to what extent the BMD variation can predict bone fractures, the endpoint of osteoporosis. Studies have suggested that OFs per se can be used as a direct trait for gene hunting of osteoporosis.(131)


A series of candidate genes have been identified by studying monogenic bone diseases/traits. Likewise, using knockout and transgenic animal models, many well-known or novel genes were found to be accountable for abnormal bone phenotypes. Studies on monogenic bone diseases and genetically manipulated animals have provided biologically relevant candidate genes for association studies. They also help isolate and highlight the functions of a particular gene, thus providing further evidence supporting the gene's relevance to bone mass variation. A salient example is the LRP5 gene. Studies on three monogenic traits, autosomal recessive osteoporosis-pseudoglioma,(132) autosomal dominant high bone mass,(95,133) and autosomal recessive osteopetrosis,(134) have brought to light the LRP5 gene for its potential role in osteoporosis. The first two conditions were proven to result from different mutations of the LRP5 gene,(132,133) and the last one was linked to the genomic region, 11q12–13,(134) which contains the gene. To confirm the significance of LRP5 to bone mass variation in the general population, independent studies were performed to test linkage of 11q12–13 to BMD.(99,135) Transgenic mice expressing a unique mutation (G171V) in LRP5 (which was associated with high bone mass in humans(95,133)) had a similar phenotype with high bone mass and enhanced strength(136) and displayed a role in regulating bone biomechanical properties in mice.(137) More recently, studies have investigated the association of the LRP5 gene polymorphisms with osteoporosis-related phenotypes in various populations.(24,96,97,138–141)

Two review articles(9,10) have comprehensively summarized such studies updated to 2002. Over the past 2 years, 28 new studies were reported, which are summarized in Table 2 (cited online).


Twelve studies were reported during 2003–2004, including 10 whole genome linkage scans(142–151) and 2 follow-up studies(152,153) focusing on candidate regions identified in previous genome scans. The first review(1) summarized several whole genome linkage scan results.(154–158) To substantiate the initial findings and to detect novel loci, several groups performed further studies in expanded samples.(142,147–149,152,153) Efforts were also made to explore confounding factors such as heterogeneity(145) and pleiotropy.(143,146) We outline below some interesting studies classified by the studied phenotypes. The newly identified QTLs are added to Table 3, based on our first review update.(1)

BMD and osteoporotic fractures

Styrkársdóttir et al.(150) reported a whole genome linkage scan in 207 Icelandic osteoporotic families (1323 individuals), using phenotypes combining BMD and osteoporotic fractures. The most significant linkage was found on chromosome 20p12.3 (multipoint LOD = 5.10; p = 6.3 × 10−7). By saturating 30 additional markers on 20p12.3, the region was narrowed down to 6.6 cM. In the follow-up LD mapping and association analyses, they found that three variants in the BMP2 gene, including a missense polymorphism Ser37Ala and two haplotypes (hapB and hapC), were associated with osteoporosis. Depending on phenotypes, Ser37Ala yielded RRs in the range of 3.8–6.3. Ser37Ala and hapC were associated with low BMD and osteoporotic fractures. They further replicated their associations in a Danish cohort of postmenopausal women. The identified associations await confirmation across different ethnic groups, and the biological effects of the three putative variants in the BMP2 gene need to be explored.

In 29 Mexican-American families (664 individuals), Kammerer et al.(144) found a QTL affecting forearm (radius midpoint) BMD on chromosome 4p (LOD = 4.33) and suggestive linkage on 12q (LOD = 2.35). They also found suggestive linkage for trochanter BMD on chromosome 6 (LOD = 2.27). In subgroup analysis for men, they obtained linkage for femoral neck BMD on 2p (LOD = 3.98) and trochanter BMD on 13q (LOD = 3.46).

Shen et al.(149) performed their second whole genome scan in a largely expanded sample (79 pedigrees, 1816 subjects). The sample contained >80,000 relative pairs informative for linkage analysis. The strongest linkage was found on Xq27 with two-point LOD scores of 4.30 for wrist BMD and 2.57 for hip BMD. Linkage was also found on 11q23 for spine BMD (LOD = 3.13), confirming findings in two earlier independent studies.(159,160) Another group performed extension studies(142) to confirm their previous finding on chromosome 1q in 464 premenopausal white sister pairs.(158) They reported replication on chromosome 1q in an independent sample of 254 premenopausal white sister pairs. They further fine mapped (4 cM) the region in all white sister pairs (n = 938) and achieved a LOD score of 4.3. In addition, they tested linkage for hip BMD in 570 white sister pairs and 204 black sister pairs(148) to compare the results with their earlier study.(158) Significant linkage was found on chromosomes 14q (LOD = 3.5 for trochanter BMD) and 15q (LOD = 4.3 for femoral BMD). However, their previous highly significant linkage on 11q12–13(99) disappeared in their follow-up efforts,(142,148) implying that underpowered studies may yield unreliable/unstable results. Another follow-up study(161) was conducted to confirm the linkage on chromosome 3p21 for spine BMD.(162) Thirty additional microsatellite markers within the 3p21 region were genotyped for a cohort of extreme discordant and concordant sib pairs (1098 individuals). The maximum evidence of linkage was LOD 3.6 for age-adjusted spine BMD.

To explore potential pleiotropic effects, Karasik et al.(146) performed principal component analyses in 323 pedigrees from the Framingham Osteoporosis Study. For PC1, loci of suggestive linkage were identified on chromosomes 1q21.3 and 8q24.3 with LOD scores of 2.5 and 2.4, respectively. For PC2, multipoint LOD score was 2.1 on 1p36. Suggestive linkage of PC_hip was found on 8q24.3 and 16p13.2 (LODs > 1.9). The study suggested that QTLs underlying bone mass variation are likely to have pleiotropic effects at different skeletal sites.

To estimate potential heterogeneity of their earlier linkage findings,(156) Karasik et al.(145) performed analyses in subsamples stratified by sex, age, and body mass index (BMI). Heterogeneity was found on chromosomes 6p21.2 and 21qter, where linkage findings in the total sample were not supported in the subsample analyses. However, subsample-specific maxima were found on 4q34.1 (males), 9q22–9q31 (younger), 16p13.2 (high BMI), and 17p13.3 (older), which were not reflected by the total sample results. Their results suggested that the genetic effects on bone mass variation could be different between men and women, younger and older, and lean and obese adults. A concern with the study is that subgroup analyses may reduce the statistical power (by reducing sample sizes), although stratified samples could be more homogenous.

Bone structure or bone size

Xu et al.(153) saturated several previously identified regions by genotyping denser markers (∼5 cM apart) in an expanded sample of 79 pedigrees. Significant linkage was achieved for wrist bone size on 17q22 (two-point LOD = 2.27, p = 0.0006; multipoint LOD = 1.78, p = 0.002). The chromosomal region 17q22 contains COL1A1, a strong candidate gene associated with osteoporotic fractures.

Another extension study was performed for femoral structures in 437 white and 201 black healthy premenopausal sister pairs,(147) of which 191 white pairs overlapped with their previous sample.(157) Linkage was replicated on chromosomes 3 (LOD = 5.0 for femoral head width; LOD = 3.6 for femur shaft width), 7 (LOD = 5.0 for femoral head width), and 19 (LOD = 3.2 for femoral neck axis length) in the white sister pairs, with a new locus identified on chromosome 8 (LOD = 6.0 for femoral head width). To identify sex-specific loci, they conducted a genome scan in 257 white brother pairs (18–61 years of age) to compare with their sister pairs.(163) The significant linkage was for pelvic axis length (LOD = 4.1) on chromosome 4p. There was no overlap between the linkage regions identified in men and in women, suggesting possible genetic heterogeneity for bone structure. A potential problem of the study is the small sample (257 brother pairs), which may affect validity of the results.(16)


Wilson et al.(151) performed a genome scan in a cohort of dizygous twin pairs to identify QTLs for bone QUS. Two specific indices, BUA and velocity of sound (VOS), were used as phenotypes. Linkage was found for BUA (LOD 2.1–5.1) on 2q33–37 and for VOS (LOD 2.2–3.4) on 4q12–21. In the Fels Longitudinal Study (453 subjects),(164) linkage was found for calcaneal QUS on chromosome 4p15 near marker D4S419 (LOD = 2.12), a region previously linked to forearm BMD.


Like many other common complex diseases, replication of linkage results for osteoporosis and related traits has been difficult. Reasons could be multiple: complex etiology of osteoporosis (e.g., genetic heterogeneity, incomplete penetrance, epistasis, variable expressivity, and pleiotropy) and poor study design.(16) However, with the increasing number of studies, the pattern of linkage that is beginning to emerge is encouraging. Not only are there strong linkage signals, but some of these are in genomic regions that harbor prominent positional candidate genes. A more encouraging fact is that we are already beginning to observe replications across studies.(142,149,153) However, caution should be taken in interpretation of replication/confirmation results. Actually, the probability an observed linkage is true depends on the power of initial study and the power of replication study.(16) A high LOD score does not always imply a true linkage, especially when studies are underpowered.

As shown in Table 3, >60 QTLs have been identified from ∼20 genome linkage scans. These QTLs were found on all but chromosome Y. Twenty QTLs have been replicated in at least two studies; these QTLs are located on chromosomes 1p36, 1q21–24, 2p21–24, 2q33–37, 3q12–26, 4p15–16, 4q31–34, 5q33–35, 6p21, 8q24-qter, 10q26, 11q23–24, 12q23–24, 13q31–34, 14q12–24, 14q31–32, 16p13, 17p11–13, 19p13-q13, and 21q22-qter. The regions harboring the largest number of replication are 1p36, 1q21–24, 4q31–34, and 12q23–24 in five different studies and 13q31–34 and 17p11–13 in at least three different studies. As the number of linkage studies continues to grow, there is strong reason to anticipate that additional replications will be brought to light. Of course, some genomic regions will eventually be proven to be false positive.


QTL mapping in animals, especially in mice, provides a powerful tool to identify human disease genes. It has greatly facilitated the daunting task facing human researchers because of the high homology (∼75%) between the human and mouse genomes. In the bone field, some pilot QTL mapping studies in mice include (1) the SAMP6 model of osteopenia by Shimizu et al.(165) and by Bene et al.,(166) which uncovered QTLs on chromosomes 2, 7, 11, 13, and 16 for low bone mass; (2) crosses between C57BL/6J (B6) and DBA/2J (D2) inbred strains by Klein et al.,(160,167) which located whole body BMD QTLs on chromosomes 1, 2, 4, and 11; and (3) crosses between B6 and Castaneus (CAST) or C3H/HeJ (C3H) by Beamer et al.,(168,169) which located femoral BMD QTLs on seven chromosomes.

During 2003–2004, 10 studies were published, reporting new QTLs for BMD, bone structure and strength, bone mechanical properties, and biochemical markers. Klein et al.(170) reported their interesting findings by following a region on mouse chromosome 11 that strongly influenced peak BMD in their earlier study.(167) They generated a DGA/2 (D2) background congenic mouse with an 82-Mb region of chromosome 11 replaced by the corresponding region of the C57BL/6 (B6) genome. The congenic mice had increased peak BMD and improved measures of femoral shaft strength (failure load and stiffness) relative to heterozygous or D2 littermates. Linkage analysis of the congenic B6D2F2 population narrowed the BMD QTL to a 31-Mb region. They next analyzed gene expression in B6 and D2 mice. Microarray analysis of kidney tissue showed that Alox15, which is located in the middle of the identified QTL interval, was the only differentially expressed gene within this chromosomal region. Their studies on Alox15 knockout mice confirmed the role of 12/15-lipoxygenase (12/15-LO) in skeletal development. Pharmacological inhibitors of this enzyme improved BMD and bone strength in two rodent models of osteoporosis. The Alox15 gene encodes 12/15-LO, an enzyme that converts arachidonic and linoleic acids into endogenous ligands for PPAR-γ. Their in vitro observations suggested that genetically determined, constitutively high 12/15-LO expression might limit peak bone mass attainment by suppressing osteogenesis through activation of PPAR-γ–dependent pathways. A concern with the study is that gene expression analysis using kidney tissue revealed a gene relevant to bone metabolism. Whereas the results are exciting, Alox15 is subject to further studies to test with the effects on bone mass variation in the general human population.

A previous interval mapping implicated 12 distinct QTLs for peak femoral BMD in (B6 × C3H)F2 and (B6 × CAST)F2 4-month-old female progeny.(168,169) To test the effect of each QTL, Shultz et al.(171) selected two sets of loci (six each from C3H and CAST) to make congenic strains by repeated backcrossing of donor mice carrying a given QTL-containing chromosomal region to recipient mice of the B6 progenitor strain. In addition, they selected the femoral BMD QTL region on chromosome 1 of C3H for congenic subline development to facilitate fine mapping of this locus. In 11 of 12 congenic strains, 6 B6.C3H and 5 B6.CAST, femoral BMD in mice carrying c3h or cast alleles in the QTL regions was significantly different from that of littermates carrying b6 alleles. Analyses of eight sublines derived from the B6.C3H-1T congenic region revealed two QTLs. Their results indicated that many QTLs identified in the F2 analyses exert independent effects when transferred and expressed in a common genetic background, and decomposition of QTL regions by congenic sublines can reveal additional loci for phenotypes assigned to a QTL and can markedly refine QTLs. Using congenic mice, Turner et al.(172) revealed sex-specific genetic regulation of femoral structure.

An earlier study identified QTLs on mouse chromosomes 1, 4, 6, 13, and 18 for femoral BMD in (B6 × C3H)F2 mice.(169) In a subsequent study,(173) the 999 F2 mouse progeny were phenotyped for measures of femoral biomechanics, structure, and more refined femoral midshaft BMD measures. Two novel multivariate phenotypes were derived using principal component analysis. Results of genome-wide analyses provided strong evidence of pleiotropic effects on chromosome 4. Chromosomes 1, 8, 13, and 14 were found to harbor QTLs affecting phenotypes in two of the three aspects of bone properties. Principal component analysis identified pleiotropic QTLs on chromosomes 4 and 14, influencing nearly all the bone phenotypes.

BMD is not the only risk factor for osteoporosis. Other intermediate phenotypes (e.g., bone structure or size, mechanical properties, and biochemical markers) may offer additional mechanistic insights into the overall processes of peak bone mass acquisition and determinants of bone strength. Volkman et al.(174) studied the genetic determinants of geometric properties of cortical bone (measured by μCT) using a mouse population containing 487 female UM-HET3 mice derived as the progeny of (BALB/cJ × C57BL/6J)F1 females and (C3H/HeJ × DBA/2J)F1 males. Fourteen markers were associated with one or more geometric traits. Because bone strength depends not only on the geometric, but also material properties of the bone, they further exploited QTLs that may affect mechanical and material properties of cortical bone.(175) Femurs from 18-month-old mice were tested to failure in four-point bending to assess mechanical properties of cortical bone. They found QTLs on maternal chromosomes 11 and 13 and paternal chromosomes 2, 4, 7, 10, 11, and 17.

To exploit genetic determinants of vertebral trabecular bone traits, Bouxsein et al.(176) evaluated the fifth lumbar vertebra in 914 adult female mice from the F2 intercross of B6 and C3H progenitor strains. They found a pattern of genetic regulation derived from 13 autosomes, with 5–13 QTLs associated with each of the traits. Using 633 MRL/SJL F2 mice, Masinde et al.(177) identified nine QTLs underlying periosteal circumference (PC), which accounted for 38.6% of phenotype variance. In addition, four epistatic interactions were found that accounted for 37.6% of phenotype variance. In another study, using a B6.C3H-4T (4T) congenic mouse strain, which is genetically 98.4% B6 and carries the C3H chromosome 4 QTL genomic DNA, Robling et al.(178) found evidence for a skeletal mechanosensitivity gene on mouse chromosome 4. Finally, two studies examined the genetic component for two biochemical markers. Srivastava et al.(179) identified three major QTLs (on chromosomes 2, 6, and 14) that influence blood levels of alkaline phosphatase (ALP) in 518 F2 female mice of MRL/MpJ and SJL progenitor strains. Using 633 F2 female mice (MRL/MpJ × SJL), Mohan et al.(180) revealed six QTLs on chromosomes 1, 9, 10, and 11 for serum IGF binding protein-5 levels.


QTL mapping in polygenic mice models continues to yield interesting findings. Among these is discovery of Alox15 as a negative regulator of peak BMD in mice,(170) which suggests traditional gene mapping methods combined with functional studies may provide novel understanding of bone mass regulation. A remarkable investment of effort over the past years was on congenic lines, which can be created by repeated backcrossing into one of the parental strains and recombinants thereof. Congenic lines prove to be a useful tool to confirm the QTL existence and fine map the QTLs and to test the quantitative effect of individual QTLs.(171)


Genetic epidemiology studies have provided valuable data for osteoporosis research. However, genetic epidemiology studies, by exploring relationships between genes and osteoporosis or related traits at the DNA level, cannot tell how the genes contribute to the disease. Functional genomic studies, by look for such relationship at the mRNA and protein levels, may help elucidate intermediate biochemical processes of a disease. The goal of functional genomics is not simply to provide a catalog of all the genes and information about their functions, but to understand how the components work together to comprise functioning cells and organisms.(181) Moreover, functional genomic studies may infer novel candidate genes and relevant pathways for DNA-based studies and confirm the results of DNA-based studies by providing functional evidence. DNA microarray and proteomic technologies, which systemically and quantitatively profile the mRNA and protein expression underlying functions of a cell type, tissue, or organisms at the genome level, are important aspects of functional genomics.

DNA microarray studies

Microarray technology allows simultaneous monitoring of gene expression for tens of thousands of genes. In the bone field, DNA microarray was first applied in 1997, when Heller et al.(182) discovered genes involved in rheumatoid arthritis. Since then, microarrays have been applied to study various aspects of osteogenesis and gene expression profiles in osteoporosis tissues. The results derived from these studies are summarized in Table 5 (cited online).

Several studies exploited the orchestrated gene expression during the process of osteoblast differentiation,(183–189) which provided novel insights into the mechanisms of bone formation and mineralization. For example, microarrays were used to determine genes regulated by different factors, such as BMP2 during differentiation of human marrow stromal cells,(190) and Tbx2 in the mouse NIH3T3 cell line.(191) Changes in expression of many genes were reported to occur during differentiation of the mouse calvarial-derived MC3T3-E1 cell line to an osteoblast-like phenotype.(192) Studies have been done to discover genes regulated in bone-related diseases: rheumatoid arthritis,(193–195) ossification of the posterior longitudinal ligament,(196) and osteosarcoma.(197) Human dental pulp stem cells and bone marrow stromal stem cells were found to have a similar level of gene expression for >4000 known human genes, whereas only a few genes were expressed differentially.(198)

Microarray technique has proven to be a powerful tool to find genes for diseases of interest. For example, based on the C2C12-generated expression data set and a training set containing known members of the osteogenic, myoblastic, and adipocytic pathways, 176 new genes were found as relevant to osteogenesis.(199) Gene array analysis of osteoblast differentiation in the mouse calvarial-derived MC3T3-E1 cell line revealed changes that were not anticipated.(192) In a comparative study of gene expression between a congenic strain that contains a QTL of high BMD and its wildtype strain in mice, ∼40% of the 8734 studied genes and expressed sequence tags (ESTs) were not documented previously.(200) Gene expression profiles during the mineralization process in bone marrow–derived human mesenchymal stem cells disclosed transcriptional stimulation of 55 genes and repression of 82 genes among >20,000 examined.(201) Among >8700 genes studied in the mouse MC3T3-E1 cell line during osteoblast differentiation, 252 were found to be differentially expressed during the proliferation and mineralization phases.(202)

Thus far, most high-throughput gene expression studies have used cultured cell lines of humans, mice, or rats. Using cell cultures has certain advantages, such as virtually no limit to tissue supply, high purity, and homogeneity of desired cell populations, possibility to easily modify experimental conditions. However, it may pose potential problems to the results obtained. Even primary cultures of stromal cells from human bone marrow gradually lose their osteogenic potential,(203,204) which may suggest changes in gene expression profiles. Moreover, cultured tissues lack relationships with other tissues. For complex disease research, it is a significant shortcoming, because it eliminates the influence of genes and/or other factors, which are not expressed in the diseased tissues, but contribute to their development. This may introduce a significant bias into gene expression profiles of cultured cells compared with freshly isolated ones. Given that, studying fresh bone or bone-related tissues may be a promising way to obtain closer to reality data on expression profiles for genes that are tissue specific and differentially expressed locally.

Bone marrow is heterogeneous in terms of cell composition and consists of many cell types, which have a potential to differentiate into various cell lineages. Some cell types of interest, such as bone marrow mesenchymal stem cells, which can differentiate into osteoblastic lineage, comprise only a minor portion of the whole bone marrow cell population. Although bone marrow cell lineages may be ideal for microarray studies, there are currently technical difficulties related to the isolation of the sufficient number of intact cells and, subsequently, RNA for the analysis. Studying gene expression profiles in monocyte/macrophage lineage cells may be of interest, because these cells are early progenitors of osteoclasts and produce cytokines important to bone metabolism,(205) and it is practical to isolate sufficient RNA from circulating monocytes for microarray analysis. Liu et al.(206) reported an in vivo study in humans using circulating monocyte. They performed comparative gene expression studies for circulating monocytes in 10 subjects with high versus 9 with low BMD(206) and identified 66 differentially expressed genes. After real-time RT-PCR validation, they found three very interesting genes, CCR3 (chemokine receptor 3), HDC (histidine decarboxylase), and GCR (glucocorticoid receptor), to be upregulated in subjects with lower BMD. The results suggested a novel pathophysiology mechanism for osteoporosis that is characterized by increased bone monocyte recruitment, increased monocyte differentiation into osteoclasts, and increased osteoclast stimulation through monocyte functional changes. The study provided helpful information for future research using fresh tissue samples such as bone marrow cells that include precursors for osteoblast lineages.

Recently developments in cell and molecular biology make it possible to obtain a genome-wide expression profile of a single cell.(207) Single cell gene expression analysis is currently used to generate data within the fundamental unit, the single cell, thereby freeing the analysis from assumptions or questions regarding cell population homogeneity. The data obtained with this approach suggest that even morphologically identical cells may have quite distinct transcriptomes, which gives deeper insights into cell differentiation and functional assignment.(208,209) Given that the processes of osteogenesis and bone metabolism are multistage and involve many cell types (osteogenic precursors, various bone cellular components, T and B cells, etc.), including those that are able to transdifferentiate, this approach may be useful for fine characterization of the distinct bone-related cell types. Single cell expression profiling may also offer a highly parallel view of the workings of a gene regulatory network at one specific point in time and will hopefully provide insights that could lead to an improved ability to interpret gene expression patterns. However, its direct relevance and significance in relation to disease gene identification (e.g., osteoporosis) remains to be determined.

Advances in functional genomics and computer sciences make it possible to analyze the gene expression changes in the context of known biological pathways.(184,185,187) However, many challenges remain in the use of functional genomic approach for complex bone disorders. The questions to be answered include how various environmental, lifestyles or inherent regulating factors contribute to the BMD variation, onset, and development of osteoporosis, and risk of osteoporotic fractures. Another question is identification of relationships and interaction between genes that have been reported to be either associated with or linked to disorders. Do their expression levels correlate with the degree of association or linkage? Do these genes interact in developing the trait?

Proteomics studies

Because of complicated processes such as alternative mRNA splicing and post-translational modifications, the correlation between mRNA and protein expression level(184,185,210–213) could be low. There are ∼30,000 genes in the human genome, but the estimated number of proteins in human cells is between ∼300,000 and ∼1,000,000.(214) Thus, the proteomics approach, complementary to DNA microarrays, is an indispensable component of functional genomics.

Proteomics can be classified into three major types: expression proteomics, functional proteomics, and structural proteomics.(215) Expression proteomics quantitatively analyzes and identifies differentially expressed proteins from protein profiles between case and control groups. Those differentially expressed proteins provide biomarkers or important proteins in pathways underlying different biomedical conditions. Functional proteomics involves the global understanding of protein–protein interactions. Because key proteins involved in disease development generally interact with other proteins, functional proteomics is a favorable approach for unveiling whole pathways participating in the etiology of diseases. To better understand and even predict protein function, one should figure out 3D structures of the proteome. Structural proteomics may prospectively fulfill this goal by mapping out the structures of protein complexes or the proteins in a specific cellular organelle.(216)

Application of proteomics in the bone field has a relatively short history, and more results from their use are yet to come. Applying an expression proteomics approach on cultured cells, pilot studies identified some novel proteins important for the development of bone marrow hematopoietic cells,(217) mesenchymal chondroblasts,(218) and osteoblasts.(219) Combining 2-dimensional electrophoresis (2-DE) and isotype coded affinity tag (ICAT) techniques, a study suggested novel proteins related to osteoclast differentiation.(220) A proteomics approach was used in seeking inhibitors of osteoclast-mediated bone resorption and is currently screening for bone anabolic agents.(221) Some in vivo studies were performed to characterize global scale molecular profiling of a variety of bone-related diseases.(222) For example, a study showed comparative molecular characterization at the transcriptome (microarray with 12,526 gene specificities) and proteome levels (multi-Western blot PowerBlot with 791 antibodies) of synovial tissue from patients with rheumatoid arthritis (RA) compared with those with osteoarthritis (OA).(222) Several new candidate molecules (e.g., Cathepsin D and Stat1) displayed reproducible differences of expression in patients with RA versus patients with OA.

Proteomics represents one of the most promising fields poised to boost our understanding of systems level cellular behavior and the fundamental etiology of osteoporosis that can be related to specific genes. It is anticipated that a huge amount of data will be produced for years to come, and the wealth of information will undoubtedly benefit osteoporosis research communities


Since our first update by the end of 2002, remarkable progress has been made in revealing the molecular genetic basis of osteoporosis. The numbers of QTLs, genes, and other markers linked and/or associated with osteoporosis-related traits continue to expand and become significantly more detailed and complex. There are now several promising chromosomal regions and candidate genes that are supported by multiple studies. On the other hand, the majority of findings are still inconclusive pending further study, which calls for new approaches and strategies with both sensitivity and robustness to accommodate confounding effects from various sources. With the rapid development in human and model organisms genome sequences, and the progress in molecular technologies, analytical tools, bioinformatics, and functional genomics, one can expect that it will be possible to define the genes and mutations and their functions in the predisposition or the resistance to osteoporosis.


HWD was partially supported by NIH and the State of Nebraska (LB595). The study was also supported by grants from CNSF, Huo Ying Dong Education Foundation, the Ministry of Education of China, and Xi'An Jiao Tong University.


Allelic heterogeneity: The same phenotypic outcome can be caused by different variants within the same gene. For example, vitamin D-dependent rickets type IIA could result from any of eight different mutations in the vitamin D receptor (VDR) gene.

Association analysis: Population-based or family-based genetic studies that examine whether an allele of a certain gene or marker co-occurs with a phenotype (e.g., a disease) at a significantly higher rate than predicted by chance alone.

Complex trait: A measured phenotype, such as disease status or a quantitative character, which is influenced by many environmental and genetic factors, and potentially by interactions among them.

Effect-size estimates: Tests of a null hypothesis can show that an effect is significant but not how large the effect is. Measures of effect size are based on the proportion of variance in the data that can be attributed to the experimental variables.

Epistasis: A form gene interaction whereby one gene masks or interferes with the phenotypic expression of one or more genes at other loci.

Gene-environment interaction: A phenomenon whereby an individual's genotype interacts with environment factors to determine his/her risk of disease. In practice, this interaction is shown as that the effect of a variant manifests only in populations with a certain environmental exposure.

Genomic control: A method to assess population stratification by using the independent marker loci to adjust the distribution of a standard test statistic.

Haplotype: A combination of alleles at different sites on a single chromosome.

Heritability: The proportion of the phenotypic variance caused by genetic variance. It reflects the degree to which the phenotype is inherited.

Linkage analysis: the analysis of family-based genotipic data to detect cosegregation of a disease locus with one or more loci within a family.

Linkage disequilibrium (LD): Two loci that are in linkage disequilibrium are inherited together more often than would be expected by chance. LD depends heavily on population history.

Locus heterogeneity: The same phenotypic outcome can be caused by variants at different genetic loci. For example, alleles at both the BRCA1 and BRCA2 locus can increase susceptibility to breast cancer.

LOD (log of the odds) score: A measure of the likelihood of genetic linkage between loci. The log (base 10) of the odds that the loci are linked (with recombination fraction θ) rather than unlinked. In classical genetics, a LOD score greater than +3 is evidence of significant linkage; one that is greater than +1.9 is evidence of suggestive linkage.

Odds ratio (OR): This is closely related to the relative risk and is defined as the odds of possessing the phenotype in those with the variant allele divided by the odds of possessing the phenotype in those without the variant allele. Odds ratios are simply a different way of expressing this association than relative risk because they compare odds rather than risk of an event.

Penetrance: The probability of an individual expresses the character of a trait/disease in the phenotype, given that he/she has a certain genotype. If the phenotype is always expressed in the presence of the genotype, the genotype is completely penetrant. If it is not always expressed, it is incompletely penetrant.

Pleiotropy: A genetic variant can affect more than one trait.

Population stratification: The presence of multiple subgroups with different allele frequencies within a population. The different underlying allele frequencies in sampled subgroups might be independent of the disease within each group, and they can lead to erroneous conclusions of linkage disequilibrium or disease relevance.

Power: The power of a statistical test is the probability that the test will correctly reject the null hypothesis when it is false. The higher the power, the greater the chance of obtaining a statistical significant result when the null hypothesis is false.

Relative risk (RR): The ratio of the incidence of the phenotype under consideration in subjects with the variant allele to the incidence in those without the variant allele.

Quantitative trait loci (QTL): Genetic loci where allelic variation is associated with variation in a quantitative trait.

Structured association: A method to infer the details of population structure to testing for association.

Transmission disequilibrium test (TDT): A method of detecting genetic association that avoids problems of population stratification. Instead of comparing unrelated cases and controls, the test determines whether given the parental genotypes, the alleles that are transmitted from parent to child and the child's affectation status are independent.

Type I error: The probability of rejecting the null hypothesis when it is true. For association or linkage studies, type I errors are manifest as false-positive reports.