|Home | About | Journals | Submit | Contact Us | Français|
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/us/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Context: Several single-nucleotide polymorphisms (SNPs) have been reliably associated with areal bone mineral density (aBMD) in genome-wide association studies of mostly older subjects.
Objective: We aimed to test those SNPs for an association with peripheral quantitative computed tomography (pQCT) bone measures in two young cohorts.
Design and Study Participants: We genotyped nine SNPs from the most promising aBMD candidates in a cohort of 15-yr-olds [in the Avon Longitudinal Study of Parents and Children (ALSPAC)] and carried out association analysis with several tibial pQCT measures to determine whether these candidates were important during adolescent growth and which particular skeletal parameters each of the candidates were acting upon. We also carried out a metaanalysis of the SNPs for association with cortical bone mineral density (BMDC) in ALSPAC and a similar male-only study (Gothenburg Osteoporosis and Obesity Determinants).
Results: In the ALSPAC cohort, we found a significant association between RANK SNP (rs3018362) and BMDC but not any of the other pQCT bone measures. In the metaanalysis, we found the OPG SNP (rs4355801) and the RANK SNP (rs3018362) to be significantly associated with BMDC. We also found suggestive evidence of an association between the MARK3 SNP (rs2010281) and BMDC but with a direction of effect opposite to that previously reported.
Conclusion: The association of genes from the RANK/RANKL/OPG pathway and BMDC provides new insight into how this system might affect the skeleton, confirming it to be associated with volumetric cortical bone density but observing no relationship with bone size.
Skeletal traits such as areal bone mineral density (aBMD) as measured by dual-energy x-ray absorptiometry (DXA) are highly heritable (1). Although studies examining associations between candidate genes and aBMD have yielded conflicting results, more recent genome-wide association (GWA) studies involving metaanalyses of different cohorts provide more robust evidence that specific genetic influences on bone exist. For example, in a metaanalysis reported by Richards et al. (2), a polymorphism within the LRP5 gene was found to be associated with spinal aBMD, consistent with the previous finding that LRP5 plays a central role in regulating osteoblast function (3,4). Polymorphisms affecting osteoclast function are also likely to contribute to the heritability of bone mass. Metaanalyses have revealed associations between aBMD and genetic markers within the receptor activator of nuclear factor-κB (RANK)/RANK ligand (RANKL)/osteoprotegerin (OPG) system (2,5,6), which plays an essential role in regulating osteoclast activity (7). Since the initiation of this current study, a further genome-wide metaanalysis has been published that incorporates all the previous data and identifies several pathways of importance for aBMD, particularly the Wnt and nuclear factor-κB signaling pathways (8).
The great majority of subjects included in GWA studies based on aBMD, including those cited above, consisted of postmenopausal women and elderly men, and therefore, it is difficult to determine whether genetic effects that were reported relate to the control of peak bone acquisition over the first three decades of life or subsequent age-related bone loss. Because accelerated bone resorption is thought to underlie bone loss, particularly in postmenopausal women, it may be that genetic variation in osteoclast control mechanisms such as the RANK/RANKL/OPG system predominantly influences the rate of bone loss in older subjects. Conversely, bone acquisition in earlier life may be preferentially affected by genetic variation in osteoblast regulatory genes. Consistent with the latter possibility, we recently reported that four markers in the region of the osterix gene (which included rs10876432, previously associated with spinal aBMD in adults) (6) were associated with aBMD derived from total-body DXA scans in approximately 5000 9-yr-old children from the Avon Longitudinal Study of Parents and their Children (ALSPAC) cohort (9). Similarly, an LRP5 gene polymorphism was associated with spinal aBMD in a subgroup of 819 9-yr-old children from the ALSPAC cohort (10).
Previous studies examining genetic influences on skeletal traits have largely focused on aBMD because DXA scans on which this is based are widely available. However, an important limitation of aBMD is that this value is influenced by different skeletal parameters such as periosteal expansion, cortical density, cortical thickness, trabecular number, and trabecular thickness (11), which may be partly under distinct biological and genetic control. For example, whereas periosteal expansion reflects growth and modeling of the skeleton, indices such as cortical density reflect bone remodeling (12). Therefore, although findings from GWA studies suggest that several pathways influence bone mass, aBMD measurements may shed less light on the biological mechanisms that underlie these associations compared with methods like peripheral quantitative computed tomography (pQCT), which analyze the different constituents of bone mass separately.
In the present study, we examine whether effects on cortical bone growth and/or remodeling in earlier life contribute to associations between genetic markers and aBMD. Therefore, we investigated the relationship between a selection of nine single-nucleotide polymorphisms (SNPs) identified as showing associations with aBMD in recent GWA studies and pQCT parameters as measured at the tibia in adolescents and young adults from the ALSPAC and Gothenburg Osteoporosis and Obesity Determinants (GOOD) cohorts, respectively.
ALSPAC is a longitudinal population-based birth cohort that recruited pregnant women residing in Avon, UK, with an expected delivery data between April 1, 1991, and December 31, 1992. This cohort is described in detail on the website (http://www.alspac.bris.ac.uk) and elsewhere (13). Briefly, 14,541 pregnant women were initially enrolled with 14,062 children born. The total sample of children was increased to 14,610 by recruiting eligible children at 7 yr of age who missed enrollment in the original cohort. Both mothers and children have been extensively followed from the eighth gestational week onward using a combination of self-reported questionnaires, medical records, and physical examinations. Ethical approval was obtained from the ALSPAC Law and Ethics committee and relevant local ethics committees, and written informed was consent provided by all parents. The present study is based on the research clinic at age 15 yr. As well as pQCT scans, height was measured at this clinic using a Harpenden stadiometer (Holtain Ltd., Crymych, Wales, UK), and weight using a Tanita body fat analyzer. Blood samples were taken and DNA extracted as described previously (14).
The GOOD study was initiated to determine both environmental and genetic factors involved in the regulation of bone and fat mass (15). Male study subjects were randomly identified in the greater Gothenburg area in Sweden using national population registers, contacted by telephone, and invited to participate. To be enrolled in the GOOD study, subjects had to be between 18 and 20 yr of age. There were no other exclusion criteria, and 49% of the study candidates agreed to participate (n = 1068). The GOOD study was approved by the local ethics committee at Gothenburg University. Written and oral informed consent was obtained from all study participants. Height was measured using a wall-mounted stadiometer, and weight was measured to the nearest 0.1 kg. All the measurements were carried out by the same trained staff. The coefficient of variation (CV) values were less than 1% for these measurements.
Cortical bone mineral content (BMCC), cortical bone mineral density (BMDC), and cortical bone area (BAC) were measured in the ALSPAC cohort using the Stratec XCT2000L, and BMDC was measured in the GOOD cohort with the Stratec XCT2000 (Pforzheim, Germany). Periosteal circumference (PC), endosteal circumference (EC), and cortical thickness (CT) were derived in ALSPAC using a circular ring model. A threshold routine was used for defining cortical bone, which specified a voxel with a density higher than 650 mg/cm3 (ALSPAC) or higher than 710 mg/cm3 (GOOD) as cortical bone. Of the 4500 scans obtained in ALSPAC, which consisted of a single slice at the midtibia, 89 were rejected as being of insufficient quality. The CV based on 139 ALSPAC subjects scanned a mean of 31 d apart were 2.7, 1.3, and 2.9% for BMCC, BMDC, and BAC, respectively. The GOOD measurements were based on the 25% tibial slice. The CV were less than 1% for all pQCT measurements in GOOD.
We selected SNPs that had been reliably associated with aBMD from genome-wide associations published up until May 2009 (2,5,6), selecting the SNP with the lowest observed P value where multiple SNPs in the same region had been identified (Table 11).). Genotyping of these SNPs was carried out in all ALSPAC children for whom DNA was available (10,121 individuals) by KBioscience (Hoddesdon, UK; http://www.kbioscience.co.uk), who employ a novel form of competitive allele-specific PCR (KASPar) and TaqMan system for genotyping. All SNPs were in Hardy-Weinberg equilibrium (smallest P value = 0.038). The same markers were selected for the present analysis, from results of a GWA performed in GOOD. Genotyping was performed with Illumina HumanHap610 arrays at the Genetic Laboratory, Department of Internal Medicine, Erasmus Medical Center, Rotterdam, The Netherlands. Genotypes were called using the BeadStudio calling algorithm.
Individuals were excluded from the ALSPAC analyses if they had self-reported non-European ancestry or had previously been identified as of non-European ancestry in a genome-wide scan [where data were available (n = 1518)] (9) or if more than one SNP could not be called in the present analysis. For siblings and half-siblings, only the first-born child with available DNA was included. Individuals were excluded from the GOOD analyses if they had more than 2.5% missing genome-wide data, if there was gender discrepancy with genetic data from X-linked markers, if they had excess autosomal heterozygosity more than 0.33, if they were duplicates and/or first-degree relatives identified using IBS probabilities (>97%), and if they were identified as ethnic outliers (3 sd away from the population mean) using multidimensional scaling analysis with four principal components. After data cleaning, there were 8373 individuals in ALSPAC with genotype data, 3113 of whom had pQCT data. In GOOD, there were 935 subjects available with complete genotype data and pQCT measurements.
We carried out association analyses between the nine SNPs and BMCC, BMDC, BAC, PC, and CT as measured in ALSPAC using linear regression and assuming an additive genotypic model. The same markers were analyzed in relation to BMDC in GOOD. We included sex and age as covariates in the primary analyses and sex, age, height, and weight natural logged [weight(ln)] as covariates in secondary analyses. Analyses were performed using PLINK version 1.04 (16).
We carried out P value metaanalyses of the associations between BMDC and the nine SNPs using METAL (www.sph.umich.edu/csg/abecasis/metal), which uses a weighted Z-method. This method calculates study-specific Z-statistics (which summarize the P values and direction of effect) for each association. The Z-statistics are then summed across studies, using weights proportional to the square root of each study’s sample size, to provide a summary P value for each association.
Because we have performed approximately 24 independent statistical tests [eight genes and assuming approximately three independent traits (BMDC, CT, and PC)], applying a conservative Bonferroni correction would yield α = 0.0021. We therefore refer to 0.0021 < P < 0.05 as suggestive and P < 0.0021 as significant evidence of association.
The characteristics of the ALSPAC subjects with both pQCT and genotype data are shown in Table 22.. The Pearson correlation coefficients between the five pQCT phenotypes are shown in Table 33.. Unsurprisingly, relatively high correlations were observed between many of these traits, reflecting their codependence on cortical bone size. However, weaker associations were observed between PC and CT, and for BMDC, implying that CT, periosteal apposition, and BMDC are at least in part regulated independently.
Table 44 shows the results of association analyses for the nine SNPs in ALSPAC. rs3018362, near the RANK gene, was significantly associated with BMDC (P = 0.001), whereas no association was observed between this marker and any other cortical bone parameter. Each minor allele (A) was associated with a decrease in BMDC of 2.5 mg/cm3, equivalent to a change of 0.07 sd. The association between rs3018362 and BMDC was not attenuated when adjusted for height and weight(ln). No association was observed between other markers within the RANK/RANKL/OPG system and cortical bone measures, with the exception of weak evidence for an association between rs4355801 (within the OPG gene) and BMDC (P = 0.051).
There was suggestive evidence for association between BMDC and rs2010281 within the MARK3 gene (P = 0.006), whereas no associations were observed between this marker and any other cortical bone parameter. Each minor allele (A) of rs2010281 was associated with an increase in BMDC of 2.1 mg/cm3, equivalent to a change of 0.05 sd. The association between rs2010281 and BMDC was partially attenuated by adjustment for height and weight(ln). There was also suggestive evidence of association between the rs1038304 marker within the ESR1 gene and BMDC (P = 0.048), which showed no attenuation after adjustment for height and weight.
We tested each of the nine SNPs for an association with BMDC in a second male-only sample, GOOD (n = 935) (Table 55).). The results for the association (with age as a covariate) are displayed alongside the ALSPAC (model 1, age and sex as covariates) results and the results from the recently published aBMD genome-wide metaanalysis (8). In the GOOD sample, two SNPs (rs4355801 and rs6993813, r2 = 0.60), which flank the OPG gene, showed some evidence of association with BMDC (P = 0.006 and P = 0.002, respectively). After carrying out metaanalysis of the two samples, rs4355801 showed significant evidence (P = 0.002) and rs6993813 showed suggestive evidence (P = 0.003) for overall association with BMDC. Although in the overall ALSPAC analyses these SNPs showed smaller effects than in the GOOD study, when we restrict the ALSPAC analyses to males only, larger effects are seen (rs4355801 β = 1.816; r6993813 β = 2.200). Although rs3018362 near the RANK gene was associated with BMDC in ALSPAC, a similar association was not observed in GOOD alone; however because the direction of effect was consistent between the two, metaanalyses performed on the two samples did show significant evidence for overall association (P = 0.0004). When analyzing the ALSPAC males alone, the effect size of this SNP is even larger (β = −4.177), suggesting the gender composition of the two cohorts does not explain the discrepancy.
Although rs2010281 (MARK3) was suggestively associated with BMDC in ALSPAC (β = 2.122; P = 0.006), a similar association was not seen in GOOD (β = 0.347; P = 0.717), and overall, there was suggestive association (P = 0.009), similar results were seen in ALSPAC when analyzing only the males (β = 2.961; P = 0.016). No association was observed between rs1038304 near the ESR1 gene and BMDC in GOOD, and there was no evidence of an overall association in combined analyses, reflecting the opposing direction of effect in the two cohorts. When analyzing the males alone in ALSPAC, the association with this SNP is even stronger (β = 3.413; P = 0.004), suggesting that the gender composition of the two studies does not explain the discrepancy.
The results in the GOOD cohort were very similar when height and weight(ln) were included in the models. The metaanalysis results were therefore almost identical to the simpler model apart from for MARK3, where the slight attenuation in the ALSPAC cohort results in the overall metaanalysis P value rising from 0.009 to 0.018. When metaanalyzing male-only data, broadly similar results are seen to those presented in Table 55 apart from the ESR1 P value rising to 0.035.
The results from the recent large-scale (>19,000 subjects) genome-wide metaanalysis of aBMD (8) include all but one of the SNPs we have studied here. The standardized effect sizes for each alleles association with aBMD are shown in Table 55.. Our results were in the same direction for all SNPs except rs2010281 (MARK3).
We found significant associations between SNPs near OPG (rs43550801) and RANK (rs3018362) and BMDC. The direction and magnitude of these effects were similar to those previously reported in GWA studies based on aBMD in older subjects. For example, there was a 0.07 sd decrease in BMDC per rare allele copy of the rs3018362 marker in ALSPAC compared with a 0.08 sd decrease in hip aBMD in the GWA study reported by Styrkarsdottir et al. (6) and although there was a 0.082 sd increase in lumbar spine aBMD per G allele in the genome-wide metaanalysis (8), the ALSPAC results show a 0.037 sd increase in BMDC and the GOOD results show a 0.125 sd increase. Taking these current BMDC and previous aBMD results together, there is a possibility that OPG/RANK polymorphisms influence aBMD at least in part through an effect on BMDC.
Our conclusions are strengthened by the fact that these analyses were performed on two independent population-based cohorts from different countries, which showed broadly similar genetic effects as judged by the magnitude and direction of β-coefficients. However, some differences were present, because for the OPG markers, β-coefficients were approximately twice as great in GOOD, whereas the converse held for the RANK marker. Although these differences may have arisen from random error, alternatively, they might reflect differences in the relative influence of genetic variation in RANK and OPG on BMDC according to maturity or gender; subjects in the GOOD cohort were approximately 4 yr older than those in ALSPAC, and unlike ALSPAC, GOOD subjects were exclusively male. When restricting the ALSPAC analyses to only males, the effect sizes for the OPG SNPs were closer to those seen in GOOD, but the effect size difference for the RANK SNP (rs3018362) is even more pronounced in the ALSPAC males, suggesting that the importance of this SNP could be dependent on age.
No associations were observed between OPG and RANK markers and other pQCT-derived measures as analyzed in ALSPAC, such as BMCC, BAC, CT, PC, and EC. Because the latter measures, which are closely correlated with each other, are all related to cortical bone size, these findings suggest that actions on cortical geometry do not contribute to genetic effects of RANK/RANKL/OPG on aBMD. In contrast to OPG and RANK markers, there was little evidence for association of the RANKL marker with pQCT parameters in either cohort. This may reflect the fact that in contrast to OPG and RANK, genetic polymorphisms in RANKL have little influence on BMDC. Alternatively, null associations may simply be a reflection of limited power of the study (see below). But this SNP had the largest effect size (of those studied here) in the aBMD genome-wide metaanalysis (8). However, the RANKL marker identified in previous studies is located some distance from the RANKL gene, and our preliminary analyses suggest that with denser coverage, it may be possible to detect associations between other RANKL markers and BMDC.
The RANK/RANKL/OPG pathway plays an essential role in regulating osteoclast function, and osteoclast activity in turn affects BMDC, for example, by determining the extent of cortical porosity. Therefore, our findings that suggest genetic variability in the RANK/RANKL/OPG pathway is an important determinant of BMDC, at least during the first two decades of life, seems biologically plausible. We are not aware of any previous study to have analyzed relationships between genetic polymorphisms in the RANK/RANKL/OPG pathway and BMDC, and so the present findings point to new ways as to how changes in this system might affect the skeleton. For example, based on our results, it is tempting to speculate that changes in BMDC contribute to the increase in aBMD and reduction in fracture risk recently reported after administration of the RANKL inhibitor denosumab to postmenopausal women (17). Consistent with this possibility, administration of denosumab has been found to increase femoral BMDC in mice with a knock-in of humanized RANKL (18).
In terms of other markers, whereas we observed a suggestive association between the rs2010281 SNP of MARK3 and BMDC in combined analyses [which attenuates slightly when including height and weight(ln) as covariates], this effect was in the opposite direction to that previously reported for hip aBMD in adults (6,8). Although the basis for this apparent discrepancy is unclear, taken together, these results suggest that changes in BMDC do not contribute to any relationship that exists between MARK3 polymorphisms and hip aBMD. Presumably, the lack of association between the rs3736228 LRP5 polymorphism and pQCT parameters, despite previous reports of associations between this marker and spinal aBMD in adults (2,8) and related SNPs and spinal aBMD in a subgroup of ALSPAC (10), reflects the fact that genetic influences of LRP5 polymorphisms largely involve the trabecular bone compartment.
Because the magnitude of specific genetic influences on skeletal traits is generally relatively small, the absence of association between pQCT parameters and other genetic markers that we examined may be a consequence of limited statistical power. For example, the proportions of variance of BMD that the associated SNPs explained in the original genome-wide papers varied from 0.24–0.67% with a mean of 0.38%. With a combined sample size of about 4000, we had 98% power to detect a suggestive association with a variant explaining 0.38% of the variance (α = 0.05). This power dropped to 88% for a variant explaining 0.24% of the variance. This indicates that for the SNPs for which we found no association, there is unlikely to be effect sizes larger than this, but we cannot rule out that they may have smaller effects. The winner’s curse will probably have resulted in the effect sizes of these variants being overestimated in the original studies, and so associations with these variants, but with smaller effects, are still quite possible.
In terms of weaknesses of our study, within the genes studied here, the actual markers that we have examined were identified from previous GWA studies and, as such, were selected purely on the basis of statistical association; it is therefore likely that these SNPs are just tagging the real variation of functional relevance and have no functional importance themselves. In addition, the aBMD genome-wide metaanalysis (8) that was published after our SNP selection has identified other SNPs that show a stronger association than the ones studied here (both in the regions studied here and other regions). As for strengths, we are not aware of any previous studies where the genetic determinants of BMDC have been assessed using pQCT. In particular, our finding that OPG and RANK markers are associated with BMDC, but not pQCT parameters related to cortical bone size, helps to advance understanding of the relationship between these markers and aBMD reported in previous studies (2,5,6,8); aBMD reflects both bone size and volumetric bone density, and genetic influences on its constituents can be identified only by techniques like pQCT that are able to distinguish these.
In summary, we investigated associations between nine SNPs related to aBMD in recent GWA studies and cortical bone traits as measured at the tibia by pQCT in ALSPAC and GOOD. Metaanalyses combining results from both cohorts revealed significant associations between the rs4355801 marker of the OPG gene, the rs3018362 marker of the RANK gene, and BMDC. Based on these findings, we propose that genetic variation within the RANK/RANKL/OPG system influences aBMD in adults at least in part by affecting BMDC during the process of peak bone mass attainment. Further studies are justified to establish whether mechanisms such as altered intracortical remodeling explain this association and whether changes in BMDC underlie changes in aBMD and fracture risk caused by perturbations in the RANK/RANKL/OPG systems in other contexts such as after pharmacological manipulation.
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.
L.P. and D.M.E. are supported by a Medical Research Council New Investigator Award (MRC G0800582 to D.M.E.). This work was supported by the Wellcome Trust (G079960 funded the ALSPAC pQCT scans). The GOOD study was supported by the Swedish Research Council, the Swedish Foundation for Strategic Research, European Commission, the Lundberg Foundation, the Torsten and Ragnar Söderberg’s Foundation, Petrus and Augusta Hedlund’s Foundation, the ALF/LUA Grant from the Sahlgrenska University Hospital, and the Novo Nordisk Foundation. The United Kingdom Medical Research Council, the Wellcome Trust and the University of Bristol provide core support for ALSPAC.
This publication is the work of the authors, and they will serve as guarantors for the contents of this paper.
Disclosure Summary: The authors have nothing to disclose.
First Published Online June 9, 2010
Abbreviations: aBMD, Areal bone mineral density; ALSPAC, Avon Longitudinal Study of Parents and their Children; BAC, cortical bone area; BMCC, cortical bone mineral content; BMDC, cortical bone mineral density; CT, cortical thickness; CV, coefficient of variation; DXA, dual-energy x-ray absorptiometry; EC, endosteal circumference; GOOD, Gothenburg Osteoporosis and Obesity Determinants; GWA, genome-wide association; OPG, osteoprotegerin; PC, periosteal circumference; pQCT, peripheral quantitative computed tomography; RANK, receptor activator of nuclear factor-κB; RANKL, RANK ligand; SNP, single-nucleotide polymorphism; weight(In), weight natural logged.