Introduction

Osteoporosis is a systemic skeletal disease characterized by low bone mineral density (BMD) and microarchitectural deterioration of bone tissue, with a consequent increase in bone fragility and susceptibility to fracture [1]. The World Health Organization (WHO) defines osteoporosis in post-menopausal Caucasian women as a value for bone mineral density (BMD) or bone mineral content that is more than 2.5 SD below the young gender and ethnicity-matched adult mean value [2]. According to the definition, osteoporosis affects 30% of postmenopausal white women in the USA, and the proportion rises to 70% in women over the age of 80 years [3]. The most common clinical outcomes of osteoporosis are fracture of the spine, hip and wrist. Of these, hip fractures are the most severe, leading to a 12–20% reduction of expected survival [4]. The direct cost for hip fractures was around $13.8 billion in the US in 1995 [5], and £942 million in the UK in 1998 [6]. With rapid economic development and aging of the population, the worldwide health and economic burden of osteoporosis will rise further in the future.

It is well established that BMD and other determinants of osteoporotic fracture are under strong genetic control. Identification and characterization of specific loci or genes involved in determining osteoporosis and associated phenotypes not only contribute to a greater understanding of the pathogenesis of osteoporosis, but also lead to the development of better diagnosis, prevention and treatment strategies of the disease. Genetic determination of osteoporosis may be monogenic or polygenic. In this review, we are mainly concerned with the polygenic form, although a limited few monogenic forms will also be briefly mentioned. Genetics research of osteoporosis represents one of the most active areas for bone biology research. Several reviews have nicely summarized the results of the candidate genes research [7,8,9,10]. As a complement to these, in this review, we first give an overview of the evidence that osteoporosis and associated traits have a genetic basis, then briefly summarize the main findings that come from linkage and association studies, with a focus on some promising chromosomal regions, and finally discuss the future challenges and directions.

Evidence for genetic determinants of osteoporosis and associated traits

Fracture is the ultimate consequence of osteoporosis. Ideally, scientists would perform genetic studies with fracture as an endpoint, and search for genes underlying the differential susceptibility to fracture. Genetic epidemiological studies have shown that a family history of fracture is a significant risk factor for fracture [11,12]. However, prospective 25-year follow-up of a nationwide cohort of elderly Finnish twins showed that susceptibility to osteoporotic fracture is not strongly influenced by genetic factors. In women, the pairwise concordance rate for fracture was 9.5% in monozygotic pairs and 7.9% in dizygotic pairs. In men, the figures were 9.9% and 2.3%, respectively [13]. Deng and colleagues [14] estimated that the narrow-sense heritability of Colles' fracture was approximately 0.25 in a cohort of white American women, thus accounting for approximately one-quarter of the variation in total Colles' fracture risk observed. Genetic factors, at most, account for about one third of the variance in the liability to fracture [15]. Fractures are relatively rare and tend to occur late in life. Although vertebral fractures are relatively common compared with hip and wrist fractures, they do not always come to clinical attention and their diagnosis may prove uncertain [16]. The relatively low heritability of osteoporotic fracture and difficulty in recruitment of an adequate sample in which to perform mapping studies of fracture in humans lead investigators to adopt alternative strategies using surrogate traits.

Bone strength is an ultimate measurement of resistance to fracture. It is mainly determined by BMD, bone size, and bone quality [7,17]. Bone strength cannot be directly measured in vivo in human. As BMD contributes substantially to bone strength and can be practically measured with marked sensitivity and precision, the evaluation of BMD is the most commonly used method for predicting fracture risk in humans. Consequently, the vast majority of genetic studies of osteoporosis to date have used BMD as a surrogate phenotype. BMD is a complex phenotype because it results from remodeling processes affecting bone compartments (endosteal and periosteal), and therefore bone size, and the process differ according to age and gender [18]. BMD in both sons and daughters correlates most closely with the average parental BMD [19]. Twin studies have shown that the heritability estimates of BMD ranged from 0.5–0.9 [20,21,22,23,24,25,26]. Since environmental influences can differ between generations considerably, heritability estimates of BMD in inter-generational studies have generally been lower than those reported in twin studies, ranging from 0.46 to 0.67 [19,27,28,29,30]. Most of the segregation analyses [28,31,32,33] suggest that there exits at least one major gene for population BMD variation. The genetic correlation between lumbar spine and femoral neck BMD is 0.64, and approximately one-third of the genetic influence on variance of femoral neck BMD is mediated through the same gene or genes that influence the lumbar spine [34]. Therefore, there are some common and specific genetic factors underlying the determination of BMD in various skeletal sites. However, a recent study indicated that genetic correlation between fracture and BMD is not significant despite a high phenotypic correlation between fracture and BMD [35]. Thus, all important risk factors for fracture need to be studied in order to find genes for osteoporotic fractures. In addition, there are obvious gender differences in the genetic components of BMD in mice [36]. Men generally have larger bone size and greater cortical mass than women [37], which is associated with considerable fracture risk reduction. Whether the genetic determinants of BMD in humans also would be influenced by gender remains to be elucidated.

Bone size is also an important determinant of osteoporotic fractures. A longer hip axis length is associated with increased hip fracture risk independent of BMD [38]. However, there have been few reports on the estimation of heritability of variation in bone size [39,40,41]. In 49 pedigrees with 703 subjects, after adjusting for sex, age, weight, height, lifestyle factors, and the significant interactions among these factors, heritability estimates were, respectively, 0.48, 0.64, and 0.6 for bone size at the hip, spine, and wrist [41]. In addition, forearm width and hip axis length are also highly heritable with heritability estimates of greater than 0.5 [20,25].

Quantitative ultrasound (QUS) measurements, including broadband ultrasound attenuation (BUA) and speed of sound (SOS), are measurements that reflect the quality aspects of bone. Subjects with lower BUA at baseline have a higher risk of hip and vertebral fractures, possibly independent of BMD [42]. Estimates of heritability based on twin studies for age- and weight-adjusted BUA and SOS are 0.74 and 0.82, respectively [25,26]. Bivariate genetic analysis indicated that the genetic correlation between BUA and BMD ranged between 0.43 and 0.51, whereas the environmental correlation ranged between 0.2 and 0.28 [26].

Bone formation and resorption markers may predict hip fracture in elderly white women [43]. Each standard deviation higher in bone specific alkaline phosphatase (BSAP) values was associated with a 4% lower level of BMD in both lumbar spine and femoral neck [44]. The genetic contribution to variation in bone turnover after attainment of peak bone mass is established [45,46,47], although the genetic effects on bone turnover are smaller than those on BMD and bone size. However, the genetic contribution to variation in rate of bone loss has not been shown consistently.

In order to understand the genetic basis for decreased bone strength, and ultimately osteoporotic fractures, one needs to assess the inheritance of, and identify the specific genes associated with, a multitude of skeletal traits, such as BMD, bone size, QUS, and bone turnover. If osteoporotic fractures are not studied as the phenotype, the genes identified need to be tested for relevance to osteoporotic fracture. The above results consistently support the hypothesis that genetic factors are a major determinant of BMD and, possibly, variance in bone size, bone turnover, and QUS measurements. This hypothesis has fueled most osteoporosis genetics investigations over the last decade.

The search for osteoporosis genes

Three major approaches to identifying genes for osteoporosis have been pursued: animal models, the candidate gene approach (association studies), and genome-wide scans (linkage studies) [42]. One major advantage of using an animal model is that it is possible to control for the heterogeneity of environmental factors in animals, which is otherwise impossible in human studies. The candidate gene approach tests for the association between a particular gene variant and osteoporosis (BMD variation), and depends on linkage disequilibrium of markers with functional mutations. It is generally prone to population admixture/stratification in yielding false positive or false negative results [48]. To overcome this problem, the transmission disequilibrium test (TDT) is employed to test specific candidate genes for both association and linkage [49]. Genome-wide scans test only linkage and are robust to population admixture/stratification. A disadvantage is that they have relatively low statistical power to detect genes with modest effects unless the sample size as reflected by the informative relative pairs is large. Genome scan not only guides candidate gene research by according greater priority to candidates that are located within regions of linkage, but also identifies novel chromosomal regions within which no known candidates have been recognized.

Animal models

Animal models, which offer controlled exposure, limited and consistent genetic variation, and unlimited size of sib-ships, hold considerable potential for understanding the genetics of osteoporosis and associated traits. A promising approach is to map quantitative traits in experimental animal models and then search syntenic regions of the human genome for genes defining these traits in humans. The genetics of osteoporosis and associated traits have been studied extensively in inbred strains of experimental animals [17,50,51,52,53,54,55,56,57,58]. Li et al. [17] identified six significant QTLs affecting bone breaking strength, of which three influence BMD, two influence bone quality, and one influences bone size. The QTL mapping results for BMD in experimental animals and the associated human homologous regions are summarized in Table 1. Notably, in several cases the same QTLs have been mapped in different crosses, using different but related phenotypes. Examples include cfh-Mit15 [17,50,54,55,58] and Mit291-Mit362 [17,51,58] on chromosome 1, Mit296-Il2ra [50,57] and Mit413-Ncvs42 [50,55,56] on chromosome 2, Mit124-Mit204 on chromosome 4 [17,55], Mit210-Mit80 on chromosome 7 [56,57], Mit242-Mit349 [51,55] and Mit36-Mit160 [50,52,57,58] on chromosome 11, Mit135-Mit16 on chromosome 13 [52,53,54], Mit132-Ptprg [17,50] and Mit160-Mit194 [55,58] on chromosome 14; Mit29-Atf4 on chromosome 15 [50,54], Rik29-Mit39 on chromosome 16 [50,55,57], and Mit185-Ncvs23 [17,50] on chromosome 18. The future challenge is to identify genes responsible for these effects and to determine the relevance of these regions to human osteoporosis and associated traits.

Table 1. Putative QTLs contributing to BMD and osteoporosis in mouse
Full size table

The candidate gene approach

There are three types of candidate genes: functional candidate genes, positional candidate genes, and expressional candidate genes. Functional candidate genes are based upon a priori knowledge of the phenotype and the potential function of the gene involved. Such knowledge may come from clinical observation or physiological studies of affected individuals, from studies of known disease-related process, from animal models of disease, and from pharmacogenetic studies. Positional candidate genes are genes targeted because of their location within regions identified through genetic linkage analyses. Expressional candidate genes are identified through differences in gene expression using genomic arrays. Candidate genes are commonly examined by association studies, using a case-control design. Since the first report of an association study between the α2-HS-glycoprotein (ASHG) gene and bone mass [59], there has been an extensive and growing list of candidate genes investigated for linkage and association with osteoporosis and associated traits. There are currently more than 200 genes that have been proposed as potential candidates for osteoporosis and associated traits [60,61].

Among the multiple candidate genes harboring polymorphic loci so far investigated in relation to BMD and fracture, the vitamin D receptor (VDR) gene has received the greatest attention. The relationship between the VDR genotype and BMD has been studied in Caucasians, East Asian, and Africans. A meta-analysis combining the results of 75 articles and abstracts published between 1994 and 1998 which examined the relationship between the VDR polymorphisms (BsmI, ApaI, TaqI, EcoRV, and FokI) and osteoporosis-related phenotypes (BMD, fracture, and QSU) have shown a highly significant association between VDR polymorphisms and BMD. Positive results were significantly more common in studies that included premenopausal rather than postmenopausal women, and the association may have been missed in some studies because of small sample size and other confounding factors [62].

Collagen type I (COL1A1) is the most abundant protein in bone, and mutations in the genes encoding collagen type Iα1 and collagen type Iα2 are estimated to be responsible for up to 90% of cases of the Mendelian disease osteogenesis imperfecta. Polymorphisms affecting the coding regions of the collagen type I genes are rare and do not appear to be associated with osteoporosis [63]. Grant et al. [64] described a G→T polymorphism in intron 1 of the COLIA1 gene at a binding site for the Sp1 transcription factor, and reported decreased BMD and increased fracture risk for carriers of the s allele in an analysis of 205 predominant postmenopausal British women. Since that time, numerous studies have been performed in both Caucasians and Asians. The unfavorable effect of the s allele has not been seen consistently across different studies, and a G→T polymorphism in intron 1 of the COLIA1 gene at a binding site for the Sp1 transcription factor does not appear to exist in Asians [65]. Recently, two meta-analyses about association of COL1A1 Sp1 polymorphism with BMD and/or the risk of prevalent fractures have been performed [66,67]. The main conclusions to emerge from these meta-analyses were that the COLIA1 Sp1 polymorphism showed a dose-response relationship to prevalence of fractures (increases stepwise from SS homozygotes to Ss heterozygotes and from Ss heterozygotes to ss homozygotes). Further, the association with fracture was stronger than expected on the basis of the observed differences in BMD. Because a large part of the inherited predisposition to fracture is due to inherited factors in bone density, and/or material quality of bone, authors concluded that the Sp1 effect on fractures may be mediated in part by its influence on bone quality other than BMD.

The relationship between the estrogen receptor α (ER-α) gene polymorphisms (TA, CA, PvuII, and XbaI) and BMD/fracture has been investigated extensively, and contrasting results were reported. A recent meta-analysis indicated that XX homozygotes (women carrying two copies of the gene variant without an XbaI restriction site) have higher BMD and also a decreased risk of fractures when compared with carriers of the χ allele, whereas the PvuII polymorphism is not associated with either BMD or fracture risk [68]. Notably, a significant gene-gene interaction between VDR and ER-α gene polymorphisms has been suggested by several authors [69,70,71]. In addition, several studies assessed whether genotypes of ER-α are associated with bone changes in women with and without hormone replacement therapy [72,73,74]. Although results are inconsistent, the information obtained should turn out to be helpful in choosing optimum therapy for osteoporosis for these different genotypes (genotype-specific treatment).

Other candidate genes investigated include, but are not limited to, the transformation growth factor β1 (TGFβ1) gene [75,76,77,78,79,80], the parathyroid hormone (PTH) receptor [81,82], the calcitonin receptor [83,84,85,86], the calcium-sensing receptor [87,88], the osteocalcin gene [82,89,90,91,92,93], the interleukin-6 (IL-6) genes [94,95,96,97,98,99], the insulin-like growth factor-I (IGF-I) genes [100,101,102,103,104], the apolipoprotein E gene [105,106,107], alpha 2 HS-glycoprotein (AHSG) gene [108], the interleukin-1 receptor antagonist gene [109], the androgen receptor (ADR) gene [90], the peroxisome proliferator-activated receptor gamma gene [110], tumor necrosis factor receptor 2 gene (TNFR2) [111], calcitonin genes [112], the P57 [113], methylenetetrahydrofolate reductase gene (MTHFR) [114], the aromatase (CYP19) gene [115], the Werner helicase (WRN) gene [116], the CC chemokine receptor-2 (CCR2) gene [117], the Klotho gene [118] and the runt-related gene 2 (RUNX2)/core binding factor A1 (CBFA1) gene [119].

Despite considerable efforts, it is still premature to draw conclusions about the potential influence of these genes on osteoporosis (BMD) and fracture. Results from association studies of candidate genes are often inconsistent. Reasons for this include false positive or negative results [48], small sample sizes and low statistical power, different sets of genes operating in different populations, variable linkage disequilibrium among populations [120], or low prior probability of the involvement of the gene in question in the overall risk of the disease [121]. To deliver robust results, some guidelines have been suggested. These include (1) significantly increased sample sizes; (2) incorporation of diverse study designs including case-control, family-based association studies and intermediate phenotype data sets [122]; (3) replication of findings in additional study groups of similar ethnic origin [123].

Whole-genome scans

Several whole genome-wide linkage studies have been conducted [124,125,126,127,128,129,130,131,132,133,134,135,136]. These results are summarized in Table 2. The following discussion will focus on some promising regions.

Table 2. Summary of results of several reported genome-wide scans
Full size table

Chromosome 11q12–13

Significant linkage to chromosome 11q12–13 has been reported for three monogenic bone diseases. The first is osteoporosis-pseudoglioma syndrome (OPS), which is characterized by low bone mass, with childhood fractures and abnormal eye development. It was linked to chromosome 11q12–13 with a maximum LOD score of 5.99 achieved at marker D11S987 [137]. OPS has been shown to be caused by several different mutations in the gene for low-density lipoprotein receptor-related protein 5 (LRP5) [138]. The second monogenic bone disease is autosomal recessive osteopetrosis (arOP) that is characterized by osteosclerosis, deafness, blindness, and severe anaemia that are due to failure of osteoclast-mediated bone resorption. It is linked to chromosome 11q12–13 with a maximum LOD score of 5.9 achieved in two Bedouin pedigrees [139]. The T-cell immune regulator 1 (TCIRG1) gene was identified as one of the genes responsible for arOP [140,141]. Mutations in this gene may account for as many as 50% of the cases of recessive osteopetrosis [142]. Inactivation of TCIRG1 causes osteoclast-rich osteopetrosis in mice [143]. The third monogenic bone trait is an autosomal dominant trait characterized by high bone mass (HBM). Johnson et al. [144] reported it to be linked to a 30-cm region of 11q12–13 with a LOD score of 5.74 achieved at the marker D11S987. A G to T transversion in exon 3 of LRP5 results in the autosomal dominant high-bone-mass trait [145]. Remarkably, this mutation also causes an autosomal dominant syndrome characterized by high bone density, torus palatinus, and a wide, deep mandible [146]. Mice deficient in LRP5 have been reported to have low bone mass, low body weight, and abnormal eye vascularization [147]. This finding supports the critical role of this gene in skeletal integrity. It is of particular interest that variation in bone density in the general population was also linked to the chromosome region containing LRP5 [127,148]. However, Deng et al. [149] genotyped five markers in a genomic region of ~27 cM centering on D11S987 for 630 individuals from 53 human pedigrees, and did not find evidence of linkage of these five markers to BMD at the spine, hip and wrist and total body BMC. The maximum LOD score at these five markers was 0.25 and the maximum LOD score at D11S987 was 0.15. Karasik et al. [129] and others did not report linkage findings on chromosome 11q12–13 in the general population either. Whether common variants that alter the expression or function of LRP5 have a role in the risk of osteoporosis in the general population merits further studies.

Chromosome 1p36

Devoto et al. [124] reported a genome-wide scan in 149 members of seven large pedigrees. The strongest evidence of linkage was on chromosome 1p36, which was identified with two marker loci separated by 13.9 cM (D1S450 and D1S214) that gave LOD scores in the single-point non-parametric analysis of +3.51 and +2.62, respectively, for hip BMD. This finding was confirmed and extended in an expanded sample of 42 families by analyzing nine microsatellite markers spanning a 40 cM interval across the candidate region [150]. In addition, Albagha et al. [151] analyzed allele distribution of microsatellites in 54 women with high BMD and 54 women with low BMD, and found that markers on 1p36 were associated with differences in BMD. Recently, a genome-wide screen of 1097 unselected female UK twin pairs confirms the presence of QTLs for BMD at 1p36 [132]. It is also of interest to note that 1p36 showed suggestive evidence for linkage to BUA [133]. Plausible candidate genes include TNFR2, lysyl hydroxylase (PLOD), and MTHFR. Previous studies have indicated significant association of the polymorphism of the TNFR2 gene, and MTHFR gene with BMD [111,114,152].

Chromosome 1q21–32

An autosomal genome screen in 429 pre-menopausal Caucasian sister pairs found significant evidence of linkage to markers on chromosome 1q. The maximum LOD score (3.11) was attained at the 170 cM position of the Marshfield chromosome 1 map [127]. A genome wide screen of an additional 289 premenopausal Caucasian sister pairs yielded a LOD score of 3.2 at position 188 cM. Linkage analysis in an expanded sample of 570 white, which has partial overlap (281 white pairs) with the previously reported genome screen sample, yielded a maximum LOD score of 6.3 at position 188 cM [131]. In addition, 1q22 showed significant evidence of linkage with one-third distal area with a multi-point LOD score of 4.78 [135].

Chromosome 2p23-p24

Devoto et al. [124] reported a multi-point LOD score of 2.25 on 2p23-p24 for spinal BMD. Niu and colleagues [126] found linkage evidence of 2p23–24 with forearm BMD with LOD scores of 2.15 in a Chinese population. In addition, 2p25 also showed evidence of linkage with L1 area with a multi-point LOD score of 2.98 [135]. This region contains two genes of potential interest, pro-opiomelanocortin (POMC) and serine threonine kinase (STK).

Chromosome 4q25-q32

Devoto et al. [124] reported a LOD score of 2.28 near D4S1535 for hip BMD and a LOD score of 2.95 near D4S1539 for spine BMD. Deng et al. [130] reported a genomewide scan in 53 Caucasian pedigrees. They observed significant evidence of linkage to 4q25–32 for spine BMD with multi-point LOD scores of 3.08, and suggestive evidence of linkage for ultra-distal forearm BMD. Some additional support came from Duncan et al. [153], who found some evidence of linkage to 4q26 for femoral neck BMD.

Chromosome 6p11–21

Koller et al. [127] reported a LOD score of 1.94 near D6S462 for spine BMD. After a genome scan of 330 Caucasian families, Karasik et al. [129] have reported evidence of linkage to 6pter for both femoral neck and spine BMD at D6S2427.

Chromosome 12q12–24

Preliminary results of a genome scan in 286 members of ten large Mexican-American families identified low BMD loci on 12q24 [125]. Again in the Framingham osteoporosis study, Karasik et al. [129] have reported a LOD score of 2.08 (137 cM) for lumbar spine BMD at 12q23 in humans. Further support for linkage to 12q24 was reported by Deng et al. [130], with a multi-point LOD score of 2.96 for spine BMD. Drake et al. [56] detected a QTL for femur BMD with a LOD score greater than 2.3 in mice in regions homologous to human chromosome12q24. Plausible candidate genes include IGF1, T-box 3 (TBX3), TBX5 and nuclear transcription factor Y β (NFYB). IGF1 has previously been associated with BMD or osteoporosis in human populations [100,101]. Duncan et al. [153] reported a two-point LOD score of 1.7 at D12S83 for lumbar spine BMD. 12q13 achieved a LOD of 1.69 at D12S368 for hip BMD in two-point analysis [130]. This region contains a number of candidate genes, including VDR, integrin α7 (ITGA7), collagen type II α1 (COL2A1), and a cluster of homeobox (HOX). Interestingly, a QTL affecting osteochondrosis was located to a position between the interferon-γ (IFNG) and IGF-1 genes at pig chromosome 5 [154]. This region is homologous to human chromosome 12q14–24.

Chromosome 13q31–34

The genomic region 13q33–34 has previously been linked to forearm BMD with a LOD score of 1.67 [126]. 13q33 showed suggestive evidence of linkage with spine BMD [130]. Interestingly, in men, but not women, 13q34 (near marker D13S800) showed linkage with intertrochanter BMD with a LOD score of 3.1 [155]. Potential candidate genes in this region include collagen type IV α1 (COL4A1) and COL4A2.

Chromosome 17p12–13

Devoto et al. [124] reported a LOD score of 2.34 near D17S261 for hip BMD in humans. 17p12–13 showed suggestive evidence of linkage with ultra-distal forearm BMD [130]. In the region homologous to human chromosome 17p11–12 in mice, Beamer et al. [55] identified a QTL for femoral and vertebral BMD variation with LOD scores of 6.76 and 2.98, respectively, and Shimizu et al. [52] identified a QTL for femur peak bone mass with a LOD score of 10.8. Also, Benes et al. [57] reported a QTL for spine BMD (P=0.0001). A candidate gene, GLI3, was located in this region.

Chromosome 18q21–23

Devoto et al. [124] found linkage of hip BMD to D18S70 and D18S42 with a LOD score of 2.14 and 2.58, respectively. The gene responsible for familial expansile osteolysis (FEO), a rare, autosomal dominant bone disorder, has been linked to a region of chromosome 18q21.2–18q21.3 [156]. An 18-bp insertion in exon 1 of TNFRSF11A segregates with patients with FEO [157]. Cody et al. [158] documented a maximum two-point LOD score of 3.4, at marker D18S42, in a large pedigree with Paget disease of bone (PDB). Positive linkage of PDB to this region has also reported by Haslam et al. [159]. Recently, Good et al. [160] has identified a novel susceptibility locus of PDB at 18q23 (multipoint LOD score of 4.71 at marker D18S70) in a large subpedigree.

Prospects for gene discovery in osteoporosis

What next? The construction of a dense single nucleotide polymorphisms (SNPs) linkage map [161] and development of new technologies such as microarrays greatly facilitate identification of osteoporosis genes. The following aspects could be advanced.

Fine mapping

The most crucial future aim is positional cloning of causal genes and identification of sequence variants within the coding or controlling regions of such genes. To achieve this, it will be essential to refine and to narrow the existing QTL to ~1 cM, a requisite size at which positional cloning becomes feasible. The chromosomal regions described so far are quite broad. It is recognized that the saturation of a candidate interval with ever more markers contribute very little to its narrowing by linkage [162,163]. Now, increased attention is turning to techniques of linkage disequilibrium (LD)-based association mapping with SNPs. SNPs allow the unification of the candidate gene approach and association-based fine mapping to identify gene(s) of interest. However, it is important to keep in mind that, even in the region narrowed, there is still the challenge of identifying the actual gene involved. There may be lack of LD even between polymorphic loci that mapped to the same gene [164]. On the contrary, even where association is demonstrated it might not indicate a contribution of that gene, but might rather reflect LD with polymorphisms in a neighboring gene [165]. Recently, haplotype blocks were found in the human genome [166,167,168]. The existence of haplotype blocks raises hope that whole genome association studies can be carried out with reasonable cost by genotyping only a small fraction of SNPs that represent most haplotype diversity within the blocks. However, haplotype blocks make identification of a true causal variant more difficult due to the underlying haplotype effect.

DNA microarray analysis and proteomics

Oligonucleotide and cDNA microarrays have revolutionized the study of differential gene expression in cells and tissues, enabling genome-wide screening of gene transcript variations. Failure to find mutations in candidate gene coding regions does not rule out a possible contribution of altered gene expression contributing to osteoporosis. The identification of differentially expressed transcripts in normal versus affected tissues may add to the process of gene discovery in osteoporosis. In the simplest case, the target gene of interest might be identified directly by characteristic changes in expression levels across a series of samples. Alternatively, statistical analysis of microarray data might aid gene discovery by detecting new metabolic disease pathways related to the target gene and facilitating identification of candidate genes [169]. Of course, some of the gene expression changes identified in this way may be a result of environmental factors, chance, or other confounding variables (false positive). Nevertheless, combining positional information and expression information will simplify the process of moving from putative linkage to gene identification. For example, microarray analysis led to the generation of a list of 175 cDNAs underexpressed by 2.5-fold or more in the fibroblasts of an affected individual (the Tangier disease, TD). By combining these data with linkage information that localized the disease gene to chromosome 9q between the markers WI-14706 and WI-4062, the candidate list was narrowed sufficiently to identify the gene ABC1, which did carry mutations [170]. DNA microarray technology has begun to be utilized to identify differentially expressed genes associated with osteoporosis [171]. It can be anticipated that more of these data will emerge in the near future.

Likewise, comparison of protein expression between normal and disease states would identify proteins relevant to the disease process, and provides obvious candidate genes as the source of inherited variation in susceptibility. DNA microarrays have limited utility for the analysis of biological fluids and for uncovering assayable biomarkers directly in the fluids. Numerous alterations may occur in proteins that are not reflected in changes at the RNA level. Since genes ultimately influence disease states through the protein products they encode, the field of proteomics could be a powerful means to help identify candidate genes that underlie genetic variation. The correlation among DNA sequence, mRNA and protein is low due to transcriptional control, translational control and post-translational modification. Strategies to incorporate DNA microarray and proteomics data into traditional linkage or candidate gene studies would improve the efficiency and capability of gene discovery in osteoporosis and in illuminating the functions of the genes and the pathways the genes and/or their products involved [172,173].

Investigation of gene-gene and gene-environment interactions

For complex human diseases such as osteoporosis, which are determined by the joint action of multiple genes and environmental factors, most current models treat separate disease loci as if they were independent of each other. Even though the individual effect of a gene may appear to be small, interactions with other genes and/or environments could make a substantial contribution to the final manifestation of the disease. Failure to recognize and accommodate such interactions may often mask the effects of individual gene. For example, Cox et al. [174] described an approach to assessing statistical interactions between different chromosomal regions where evidence for linkage at one region is taken into account in assessing the evidence for linkage elsewhere in the genome. Using this approach, they showed an interaction between loci on chromosomes 2 and 15 that increases susceptibility to non-insulin-dependent diabetes (NIDD1). Interestingly, conventional linkage analysis failed to detect linkage to chromosome 15 in the initial genome scan. In addition, Cordell et al. [175] described a multi-locus linkage method. They showed that multi-locus analysis not only increased power to detect linkage, but also assisted in determining the nature of the relation between disease loci (i.e. genetic heterogeneity versus epistasis). One of the most important goals of the next generation of genetic studies of osteoporosis is to determine which multi-locus genotypes create the highest risk for development of osteoporosis.

The examples of gene-gene and gene-environment interaction for the VDR gene have been described by others [69,70,71] and us [10]. Bone density at any age is the end result of peak bone density and subsequent loss, and thus reflects the sum of responses to various environmental exposures. If genetic factors modulate those responses to environments, these gene-enviroment interactions presumably accumulate over time with aging [7]. Strength and direction of the VDR allelic effects may relate to the genetic backgrounds in different studies and environmental factors such as calcium and vitamin D intakes. This at least partially explains inconsistent results of the relationship between VDR polymorphism and BMD across multiple studies.

International collaborations

Several whole genome-wide linkage studies have been conducted [124,125,126,127,128,129,130,131,132,133,134,135,136]. Sample sizes have generally been modest. The relatively small numbers studied would have tended both to limit the power of these genome screens to detect linkage and to increase the possibility of false positive errors. A common approach to enhance the power of any study is to utilize a larger sample size. Large samples may augment weak linkage signals found in small data sets and are less susceptible to random statistical fluctuations that may lead to false positive results in smaller samples. The most expedient approach to further progress in the identification of genes for osteoporosis and associated traits would be through the efforts of a consortium to merge and jointly analyze all extant data sets for linkage. The feasibility of this approach has been demonstrated in search of genes for type 1 diabetes [176]. However, since there may be considerable differences in sampling strategies, in phenotypic measurement, marker sets, and expected etiologic heterogeneity, sometimes it is not possible to directly pool the data from studies that are conducted independently without standardization. In this case, meta-analyses of multiple independent data sets have been proposed as an alternative [177]. We advocate that both significant and non-significant results of whole genome scans should be published to facilitate meta-analyses. On the other hand, multicenter genetic and family studies are rapidly evolving as a means of generating large samples of family data collected by using standardized protocols, for example, the Family Blood Pressure Program (FBPP) [178].

Conclusions

Considerable efforts have been made recently to investigate the genetic basis of osteoporosis. Numerous candidate genes have been tested for association and linkage with osteoporosis. However, these candidate genes are often neither essential nor sufficient to produce osteoporosis on their own. Whole genome studies have identified some regions that may harbor QTLs contributing to osteoporosis. However, there is currently little consensus about loci or identity of specific genes that confer genetic susceptibility to the development of osteoporosis in different populations. Furthermore, the transition from QTL detection to gene identification has proven difficult. Nevertheless, the successful identification of NOD2 [179,180] and calpain-10 susceptibility loci [181] for Crohn's disease and type II diabetes, as well as replication of some of linkage findings across multiple studies is encouraging. With the anatomy of the human genome at hand, sequence-based gene discovery is complementing, and will eventually replace, map-based gene discovery. Identifying sequence variations responsible for osteoporosis and understanding how these variations regulate the phenotypes will still be the major challenges in the future. We can be optimistic concerning the future of gene discovery for osteoporosis by the use of a combination of functional, positional, and expression information.