Genetics of Bone Mineral Density: Evidence for a Major Pleiotropic Effect From an Intercontinental Study
BMD is a primary predictor of osteoporotic fracture, and its genetic determination is still unclear. This study showed that the correlation between BMD at different skeletal sites is caused by an underlying genetic structure of common genetic effects. In addition to possible shared (pleiotropic) genetic and environmental effects, each of the BMD variables may also be determined by site-specific genetic factors.
Introduction: BMD is a primary predictor of osteoporotic fracture and a key phenotype for the genetic study of osteoporosis. The interindividual variation in BMD measured at a given skeletal site is largely regulated by genetic factors. A strong phenotypic covariation exists for BMD at different skeletal sites. This study tests the hypothesis that the covariation is in fact caused by an underlying genetic structure of common genetic effects and that, in addition to possible shared (pleiotropic) genetic effects, each of the BMD variables may also be determined by site-specific genetic factors
Materials and Methods: A bivariate complex segregation analysis as implemented in statistical package PAP was conducted to explore various models of pleiotropic genetic and environmental transmission in lumbar spine and femoral neck BMD, as well as in compact and spongious segments of hand phalanges. The BMD was obtained in three ethnically, culturally, and socially heterogeneous samples of white pedigrees, with 2549 individuals between 18 and 100 years of age, from Australia, Europe, and North America.
Results and Conclusions: The genetic correlation between BMD measures ranged between 0.50 ± 0.09 and 0.79 ± 0.04 in the three samples. In each sample, the model incorporated a major locus pleiotropic effect, and residual correlation was found to be the most parsimonious model. Estimated parameters from the model indicated a significant pleiotropic major gene effect on both lumbar spine and femoral neck BMD, with the existence of a significant residual correlation (0.51 ± 0.07 to 0.66 ± 0.04). These results suggest that the covariation in BMD at different skeletal sites, and between mostly compact versus mostly trabecular bone, was largely determined by common genetic factors that are pleiotropic or in close linkage and linkage disequilibirum, while at the same time, exhibiting considerable evidence of shared environmental effects. The results, for the first time, suggest that the possibility of pleiotropic genetic effect may be controlled by a major genetic locus. Identification of the major locus could open new opportunity to understanding the liability and pathogenic processes in which they are involved in the determination of fracture risk.
THE LIABILITY TO fracture is a function of trauma sustained and bone strength. Bone strength depends on both the density (quantity) of the bone and on the quality of the bone. Bone mass and bone strength can be assessed by BMD. BMD is a primary predictor of fracture risk, with each SD decrease in BMD being associated with at least a 2-fold increase in the risk of fracture.(1) The BMD-fracture relationship generally applies across the skeleton, with some site-specificity; that is, hip fracture risk is more related to BMD measurements at the hip than lumbar spine or forearm.(2) Therefore, BMD is often considered a primary and relevant phenotype in studies of genetics of osteoporosis.
BMD is biologically a complex trait because it is determined by both, and perhaps interaction between, genetic and environmental factors. Several twin and family studies have shown that the difference in BMD among individuals in a population is largely regulated by genetic factors, with genes accounting for between 65% and 92% of the variance of BMD in any population.(3,4) BMD measured in various skeletal regions can be broadly grouped into two types: cortical bone and trabecular bone. For instance, the lumbar spine contains both trabecular and cortical bone, and the femoral neck contains largely trabecular bone. Given their common biological characteristics, it is not surprising that BMD at these sites are correlated, but the correlation is not perfect.(5) For example, the correlation between femoral neck and lumbar spine BMD ranges between 0.5 and 0.7. However, the correlation may reflect a larger underlying phenotype of osteoporosis.
It is, therefore, hypothesized that the phenotypic correlation is in fact caused by an underlying genetic structure of common genetic effects and that, in addition to possible shared (pleiotropic) genetic effects, each of the BMD variables may also be determined by site-specific genetic factors. To test these hypotheses, we conducted a bivariate complex segregation analysis to partition the phenotypic correlation between different BMD measures in three groups of white subjects into components caused by genes shared between measures. The studied pedigree samples represent three populations that are drastically different in cultural, social, and ethnic backgrounds and come from three continents: Australia, Europe, and North America. The data were collected independently, and therefore, the obtained results should allow for a generalization of the conclusions with reasonable degree of confidence.
MATERIALS AND METHODS
Setting and study populations
This study used data from three independent populations in Australia, the United States, and Chuvasha (Russian Autonomous Republic). Details of the study design of the studies have been described previously.(6–8) Characteristics of the study subjects are shown in Table 1. Some brief details are as follows.
Table Table 1.. Characteristics of Study Samples
The Australian study
The Dubbo Osteoporosis Genetics Study (DOGS) was designed as a population-based, genetic epidemiological investigation of BMD and osteoporotic fractures.(6) Families were identified and invited to participate in the DOGS study through an index case with the following criteria: (1) moderately high bone density at the femur (Z-score > +1.28 or top 10% of the age-specific BMD distribution) and (2) availability and accessibility of extended families of adult members, identified through these older (age > 60 years) parent index cases and their siblings and families. The probands were identified from the database of the Dubbo Osteoporosis Epidemiology Study. All subjects were white, of Anglo-Saxon background, and ≥18 years of age.
The U.S. study
In the second sample, the study subjects came from previous studies that have been documented earlier.(7,9) The study was approved by the Creighton University Institutional Review Board. All the study subjects signed informed consent documents before entering the project. Only healthy people (defined by the exclusion criteria detailed elsewhere(7)) were included in the analyses. One pedigree was identified through a proband having BMD Z-scores ≥ +1.28 at the hip and spine. The rest pedigrees were identified through a proband having BMD Z-scores ≤ −1.28 at the hip or spine; therefore, the probands were selected from the bottom 10th percentile of the population BMD variation from the midwest United States. Sampling through extreme probands may achieve higher statistical power than random sampling in linkage studies.(10) The study sample contained individuals belonging to 51 complex pedigrees of three and even four generations. All subjects were whites of European origin.
The Israel study
The third sample included ethnically Chuvasha (descendants of Bulgar tribes) individuals who live in many small villages in the Chuvasha and Bashkortostan autonomies, the Russian Federation. The data were retrieved from the 349 nuclear family members ranging from three to seven individuals per family. Some nuclear families (78) were combined in more complex three-generation pedigrees. The Chuvasha population is characterized by a stable family structure with the traditional relations between family members. For several generations, the Chuvashians have lived under the same environmental conditions and have been exposed negligibly to outside influences (such as genetic flow), thus sharing similar living and economic conditions in the majority of families. Complete families have lived in the same villages and have agreed to participate in the study. Although greatly regrettable in an ethical and human sense, the almost complete lack of medical services in the area has generated a naive (i.e., untreated) population over >14 years. This condition allows for a study unmodified by hormonal or other medical interventions. All studied individuals and families were recruited randomly, that is, regardless of the outcome of any of the measured variables. All subjects who agreed to participate in the study signed informed consent and the project was approved by the Tel Aviv University ethnics committee.
Measurement of phenotypes
In the Australian study, BMD (g/cm2) and BMC (g) were measured at the lumbar spine (LS) and femoral neck (FN) by DXA using a Lunar DPX densitometer (Lunar Corp., Madison, WI, USA). The CV with this method at our institution in normal subjects for BMD was 1.5% for the lumbar spine and 1.3% for the femoral neck.
In the U.S. sample, BMD was measured by Hologic 1000, 2000+, or 4500 scanners (Hologic Corp., Waltham, MA, USA). All machines were calibrated daily, and long-term precision was monitored with external spine and hip phantoms. The short-term measurement precision is reflected by a CV of 0.7% for spine BMD, 1.0% for hip BMD, and 1.2% for wrist BMD. For the spine, our quantitative phenotype was combined BMD of L1-L4. For the hip, it was combined BMD of the femoral neck, trochanter, and intertrochanteric region. For the wrist, it was ultradistal BMD. Data obtained from different machines were transformed to a compatible measurement by an algorithm developed by us(11) and members of the same pedigree were usually measured on the same type of machine.
In the Israel study, standard plain hand radiographs of both hands were obtained from each individual in the posterior-anterior position with an X-ray source. Both hands and reference aluminum wedge were placed on the same film contacting plate to avoid film development variation. Using the computer-attached optic densitometry BMD measurements from the middle and distal phalanges of the third finger of the left and right hand were obtained. The BMD measurements from mostly spongious (trabecular) and mostly compact compartments of each bone were taken separately. Average of standardized measurements of each bone, trabecular and compact separately, was used in this study. Further details on the sample and method are given elsewhere.(8,12)
Before the genetic analysis, each BMD variable in each sample was adjusted for sex, age and weight, and body length, the three most powerful determinants of BMD, in a multiple linear regression model, taking into account nuclear families size and structure.(13) Age was modeled both as a linear term and as a quadratic term. The residual scores from the regression model, which were independent of age, sex, and anthropometrics, were used for testing the hypothesis of inheritance. The results of the respective adjustment reported in our previous papers(6–8) for the U.S., Chuvasha, and Australian samples, respectively. Table 1 provides the corresponding R2 for each sample and variable according to gender.
First, univariate heritability estimates were obtained following the formula of Rice et al.(14): HG2 = (rSB + rPO)(1 + rSP)/(1 + rSP + 2rSPrPO), where rSB,rPO, and rSP are correlations between sibs, parents, and offspring and between spouses, respectively. Classical additive genetic rG(XY) and environmental rE(XY) correlations respectively, as fully described in Falconer and Mackay,(15) were computed. Statistically significant cross-trait parent offspring correlation, rP(X)O(Y) served at the beginning of the analysis as a basic indicator for the possible common genetic factors affecting simultaneously (pleiotropy) both studied traits, X and Y. The magnitude of this effect, which reflects the portion of phenotypic covariation attributable to common genes, is measured in genetic correlation.
At the next stage, a bivariate complex segregation analysis (CSA) was carried out using PAP software. The principles of CSA and parameter definitions were described in details in numerous publications(16,17) and in our previous studies.(6–8,12) Here we will only briefly reiterate the parameter definitions and models tested in the course of analysis.
The general bivariate model assumed that the observed phenotypic correlation between the variables X and Y resulted from a common major genetic locus effect (MG), a common additive polygenic effect (PG), and shared environmental factors (CE), each acting independently and additively. The corresponding parameters of the bivariate model included the gene frequency P(A) for the low levels of both BMD variables in each analysis. Correspondingly, 1 − P(A) = P(B) frequency for the common allele for high levels of X and Y. Under the assumption of random mating, frequencies of the corresponding genotypes, AA, AB, and BB are under the Hardy-Weinberg equilibrium, with population means for each of the three genotypes (γAA, γAB, γBB) and specific for each of the two BMD variables. σg is an SD of the trait in individuals having the same genotype g. It estimates the trait variation caused by all possible environmental factors and potential minor genes, and it has a trait specific value, which was allowed to differ between the BMD variables, X and Y. The general model included also transmission probability parameters τg (τAA, τAB, and τBB). Under the MG effect hypothesis, the model assumes Mendelian transmission of alleles from generation to generation. That is, genotypes AA, AB, and BB transmit allele A with probability of τAA = 1, τAB = 0.5, and τBB = 0, as expected from Mendelian laws. Common additive polygenic and environmental effects were estimated in parameters rPG and rCE. Residual familial resemblance within each trait, not explained in neither common MG, nor rCE was modeled as a residual polygenic heritability (h2x and h2y).
The analysis was started with fitting a general bivariate model in which all parameters were allowed to be estimated. Parameter estimates obtained for each variable and in each sample in previous univariate models(6–8) were used to provide initial values for estimation. Next, the number of restricted submodels was tested in the following order. (1) Sporadic submodel that assumed no genetic (MG and PG) effects on studied traits distribution. Only phenotypic correlation is allowed. This model suggests that genetic factors do not influence interindividual variation in BMD values in the studied sample. (2) Submodel testing hypothesis of only polygenic inheritance of each of the two BMD variables and possible PG correlation between them. (3) Mixed Mendelian submodel, it included presence of two genetic components (MG and PG) as well as possible shared environment for both BMD traits. The reliability of this hypothesis, if not rejected, was tested for departure from the expected Mendelian transmission probabilities, i.e., P(A) = τAA = τAB = τBB. (4) This model assumes existence of the familial aggregation of the given trait, however does not allow for the single gene effect. If hypothesis 3 is accepted, and hypothesis 4 is rejected, the significance of the effect of the MG influencing trait X on trait Y was examined. This hypothesis assumed no MGX effect on trait Y. (5 and 6) Submodels that suggested that if a pleiotropic MG effect is confirmed, there is no statistically significant residual correlation caused by either polygenes or common environment, respectively.
A maximum likelihood ratio test was used as a model fitting technique to compare the general model with each of the restricted submodels. A χ2 statistic (= −2 ln likelihood between the general and the restricted models), with degrees of freedom equal to difference in the number of parameters fit under the models. Our univariate analyses of the BMD traits examined in this study and in the same samples unequally showed that there was no need for ascertainment correction in any of the samples.(6–8)
Estimates of heritability (univariate analysis) ranged between 0.4 to 0.5 (p < 0.001) for femoral neck and lumbar spine BMD as well as for phalangeal BMD. The phenotypic correlations between lumbar spine and femoral neck BMD in the U.S. and Australian samples were comparable; however, the genetic correlation in the Australian samples was slightly higher than that in the U.S. sample. Moreover, both estimates of phenotypic and genetic correlations in the Chuvasha sample were higher than in the U.S. or Australian samples (Table 2).
Table Table 2.. Familial Resemblance for Variation and Covariation of BMD Measurements in Three Ethnically Different Samples
The results of bivariate segregation analysis clearly revealed the common trends in co-transmission of each pair of variables (Tables 3, 4, and 5). Despite the fact that the major gene effect was unequivocally revealed for each of the study variables and in each sample, using the univariate segregation analysis,(6–8) we tested this assumption again in the present bivariate approach. To this aim, in the general model in each sample, transmission probabilities were estimated along with all other parameter estimates. The observed transmission probabilities (tau) estimates were remarkably close to the expected 1.0, 0.5, and 0.0 in all three instances (Tables 3, 4, and 5). Indeed, constraining them to the expected Mendelian probabilities (M1) left the corresponding models likelihood virtually without changes, whereas corresponding submodels denying the existence of the major gene effect were clearly rejected (p < 0.01 in all instances). The fit of the sporadic model (M2) was even worse, and it was rejected with p < 0.001 in all three samples. We therefore constructed all other hypotheses as submodels of the mixed Mendelian model (M1).
Table Table 3.. Bivariate Segregation Analysis of LS_BMD and FN_BMD in Australian Sample
Table Table 4.. Bivariate Segregation Analysis of Spine_BMD and Hip_BMD in U.S. Sample
Table Table 5.. Bivariate Segregation Analysis of Radiographic Hand Compact and Tabecular BMD in Chuvasha Sample
It is of interest to note, however, that the parameter estimates in the respective general models in the U.S. and Australian samples were quite similar. The main difference concerns only the residual multifactorial component. While in Australian pedigrees, significant residual h2 was inferred for both LS and FN BMDs, with corresponding significant genetic correlation between the residuals (Table 3), in the U.S. sample, only LS BMD showed significant residual heritability estimate (Table 4). Correspondingly, no significant residual genetic correlation was detected. The pure polygenic pleiotropic model (M3), compared with the respective mixed Mendelian model, was reliably rejected by likelihood ratio tests (at p < 0.02-0.0001) in all three studied samples. The model, assuming no MG effect of the first variable on the second one (M4), was also rejected (with p < 0.01-0.001).
In the “pure polygenic” model (M3), one can note a substantial h2 estimate for both BMD variables and in all three samples, as well as very substantial genetic correlations between each pair of BMDs. All the heritability and genetic correlation estimates were of a quite similar magnitude to corresponding non-model-based estimates (Table 2), and yet by likelihood ratio test, were clearly worse than the model, including the major gene effect. Indeed, while the model assuming that the putative major gene affecting the first BMD variable (e.g., LS BMD) transmission does not affect the transmission of the second one (e.g., FN BMD) was clearly rejected in the studied samples, with p < 0.01 (M4), the only pleiotropic major locus model (M5), which denied the existence of the residual genetic correlation, was easily accepted (p > 0.500!) in all analyses. However, the assumption of nonsignificant residual correlation caused by shared environment (M6) was strongly rejected (p < 0.001) in the studied samples. Thus, our most parsimonious model is model 5, and it consistently suggests significant pleiotropic major gene effect on both LS and FN BMD, COMP and SPNG intergenerational transmission, and the existence of the substantial correlation between the residuals (0.51 ± 0.07 and 0.66 ± 0.04), which originated from the common environment influences. The magnitude of the aforementioned putative pleiotropic gene effect varied between the BMD variables and between the samples. It ranged between about 24% and 41% of the LS and FN BMD variation in the Australian sample and about 26% and 52% in the U.S. pedigrees (Table 6).
Table Table 6.. Major Gene Effect on BMD Variables in Three Studied Samples
Genetic studies of osteoporosis and its major risk factors represent one of the most rapidly developing areas for research in genetic epidemiology and bone biology. It is obvious now that the variation in BMD, a primary predictor of fracture risk, among individuals is largely regulated by genetic factors.(4,18,19) The covariation in BMD measures in different skeletal sites, predominantly at the LS and FN, is also under genetic influence,(3) probably because of a pleiotropic effect or close linkage and linkage disequilibrium of the respective genes. Results of this study confirm that the covariation in BMD at these sites have substantial genetic determinants in common (pleiotropy or close linkage and linkage disequilibirum), while at the same time, exhibiting considerable evidence of shared environmental effects. Furthermore, this study, for the first time, clearly and consistently suggests that the pleiotropic effect, if it exists, may be controlled by a major genetic locus. This major gene is likely involved into a regulation of variation of BMD at different skeletal sites. However, the magnitude of its effect is site dependent (e.g., it explained some 24% of the BMD variation at hip versus 41% at spine in Australian pedigrees; Table 6). In this study, in addition to a common putative major gene effect, site-specific genetic influences were also detected. For instance, in Australian pedigrees, from 16% (spine) to 48% (hip) of the residual BMD variation may be attributable to polygenic component.
However, the real genetic transmission of quantitative traits may even be more complex. Various interactions between genes, genes and age, and environment (including humoral factors) are also likely to be involved in BMD transmission and may complicate the situation substantially.(19) Thus, for example, in the Chuvasha sample, the previous studies revealed that Px haplotype of the ERα gene affect radiographic hand BMD in men and women differently, and it was particularly strongly associated with low BMD in postmenopausal women.(20,21) The latter study also showed that the combination of Px haplotype and allele s of the COLIA1 gene had an even more profound effect on BMD decrease. These genes are likely to influence circulating levels of calciotropic hormones, which in turn modify bone turnover level.(22) However, in segregation analysis, these effects are taken into account in the residual transmission parameters (such as residual familial correlations) or left in the unexplained residual variation.
Although the existence of extensive pleiotropy between the two BMD measures is in itself probably not surprising, the rather consistent and significant genetic correlation between these variables across three independent populations is remarkable. Indeed, the present investigation that used pedigree data from populations living in three continents clearly consistently demonstrated that the involvement of major genes in variation and covariation of BMD measures is very likely. The Mendelian transmission in all analyses was accepted, whereas alternative hypotheses were unequivocally rejected. Because the three studied samples are independent, the probability that the non-Mendelian hypothesis was rejected by chance is only p < 10−5.(23) However, it is important to note here that genetic correlation may not be caused only by pleiotropy but also by close linkage and linkage disequilibrium (LD) of genes with similar functions The segregation analysis cannot distinguish between these two situations, because LD groups of genes expected to follow simple Mendelian transmission rules. If linkage and LD are not perfect, this may increase the genotype specific variance (σ2g) but will unlikely affect transmission probability parameters (τg).
The present results do not prove, but suggest, the likelihood of the particular genetic model and mode of transmission of BMD. However, there are a number of findings that support or at least do not contradict our conclusions on major gene effect, in particular, recent genome scans. Of course, it is quite likely that different polymorphisms (or genes) are responsible for the variation of BMD in different populations. That is, the existence of numerous potentially relevant genes involved in BMD determination cannot be excluded. However, the different mutations with major effect were likely accumulated in various populations. Thus, apart from the possible false positive results obtained in a “whole-genome” scan, different loci are expected to be associated with BMD in different populations.
Indeed, whole-genome linkage scans to date(9,19,24,25) have yielded inconsistent results across populations. In the majority of studies, the different DNA markers were associated with BMD at various skeletal sites even within the same pedigrees. For example, femoral trochanter was linked to a 21qter chromosome, and the lumbar spine was linked to a 14q31 chromosome in the Framingham cohort.(24) On the other hand, in most studies, for the given BMD, as a rule, significant linkage to one specific chromosomal segment and a few more (if any) with substantially lower probability were identified. The best example is the recent study by Styrkarsdottir et al.,(26) in which 950 DNA markers were examined in 281 Icelandic families identified through individuals with the lowest 10th percentile of BMD at either the hip or the spine. Only a single chromosomal area, 20p12 with significant evidence for linkage (LOD 3.6), was found in that investigation. The inclusion of additional individuals and DNA markers only confirmed these results, proposing that the BMP-2 gene is the most likely candidate.
It should also be noted that most current whole genome scans are underpowered. This, plus the involvement of a residual polygenic inheritance, may easily render that some true QTLs detected in different studies can be difficult to replicate across studies.(27) This is because even if a sample has low power to detect one specific QTL, the power to detect one of several QTLs can still be substantially higher. This may also partially explain the findings that, in whole genome scans reported so far, different QTLs were associated with different skeletal sites.
A common feature of past whole-genome linkage studies is that the analysis was based on univariate models. Given the presence of pleiotropic effect as observed in this study, such models may be less optimal for the search of osteoporosis genes. Perhaps, a multivariate model based on a linear combination of multiple BMD measures may increase the signal of linkage.(28,29) It has been shown recently that some associations may be detected using, for example, principal components of several bone traits. These associations were not detectable when individual traits were used.(30) Moreover, the simultaneous linkage analysis of a number of intercorrelated phenotypes (multivariate linkage analysis) may be significantly more powerful and much less redundant than the sequential univariate linkage analysis.(31) The recent simulation studies using multivariate continuous phenotypes showed that multivariate test for linkage, with application to pleiotropy, is superior to univariate linkage analysis.(32)
In summary, this study highlights the complex genetic relationships between BMD measures. The recognition and unraveling of such genetic interactions could open new opportunity to understanding the liability and pathogenic processes in which they are involved in the determination of fracture risk.
Dr Eisman has received corporate appointments and research support, and served as a consultant for Aventis, Eli Lilly and Company, Merck & Co., Inc., NPS Pharmaceuticals, Organon, Roche, and Servier. All other authors have no conflict of interest.