Quantitative Trait Loci Affecting Peak Bone Mineral Density in Mice


  • Robert F. Klein,

    Corresponding author
    1. Bone and Mineral Research Unit, Department of Medicine, Oregon Health Sciences University and Portland Veterans Affairs Medical Center, Portland, Oregon, U.S.A.
    • Address reprint requests to: Robert F. Klein, M.D., Portland VA Medical Center (111P), 3710 Southwest U.S. Veterans Hospital Road, Portland, OR 97201 U.S.A.
    Search for more papers by this author
  • Steve R. Mitchell,

    1. Portland Alcohol Research Center and Department of Behavioral Neuroscience, Oregon Health Sciences University and Portland Veterans Affairs Medical Center, Portland, Oregon, U.S.A.
    Search for more papers by this author
  • Tamara J. Phillips,

    1. Portland Alcohol Research Center and Department of Behavioral Neuroscience, Oregon Health Sciences University and Portland Veterans Affairs Medical Center, Portland, Oregon, U.S.A.
    Search for more papers by this author
  • John K. Belknap,

    1. Portland Alcohol Research Center and Department of Behavioral Neuroscience, Oregon Health Sciences University and Portland Veterans Affairs Medical Center, Portland, Oregon, U.S.A.
    Search for more papers by this author
  • Eric S. Orwoll

    1. Bone and Mineral Research Unit, Department of Medicine, Oregon Health Sciences University and Portland Veterans Affairs Medical Center, Portland, Oregon, U.S.A.
    Search for more papers by this author


Peak bone mass is a major determinant of risk of osteoporotic fracture. Family and twin studies have found a strong genetic component to the determination of bone mineral density (BMD). However, BMD is a complex trait whose expression is confounded by environmental influences and polygenic inheritance. The number, locations, and effects of the individual genes contributing to natural variation in this trait are all unknown. Experimental animal models provide a means to circumvent complicating environmental factors, and the development of dense genetic maps based on molecular markers now provides opportunities to resolve quantitative genetic variation into individual regions of the genome influencing a given trait (quantitative trait loci, QTL). To begin to identify the heritable determinants of BMD, we have examined genetically distinct laboratory mouse strains raised under strict environmental control. Mouse whole-body bone mineral content by dual-energy X-ray absorptiometry (DXA) correlated strongly with skeletal calcium content by ashing, and peak whole-body BMD by DXA in female mice occurred at ∼80–90 days of age. We therefore determined mean body weight and peak whole body BMD values in 12-week-old female mice from a panel of 24 recombinant inbred (RI) BXD strains, derived from a cross between C57BL/6 and DBA/2 progenitors. The distribution of body weight and BMD values among the strains clearly indicated the presence of strong genetic influences on both of these traits, with an estimated narrow sense heritability of 60% and 35%, respectively. The patterns of differences in body weight and peak whole body BMD in the BXD strains were then integrated with a large database of genetic markers previously defined in the RI BXD strains to generate chromosome map sites for QTL. After correction for redundancy among the significant correlations, QTL analysis of the BXD RI strain series provisionally identified 10 chromosomal sites linked to peak bone mass development in the female. Several of the identified sites map near genes encoding hormones, structural proteins, and cell surface receptors that are intricately involved in skeletal homeostasis. Four QTL for body weight were also identified. One of these loci was also strongly linked to inherited variation in BMD. This finding suggests that body weight and peak BMD may be influenced by linked genes or perhaps by common genes with pleiotropic effects. Our phenotyping in the RI BXD strains has allowed us to map a number of specific genetic loci strongly related to the acquisition of peak BMD. Confirmation of these findings will likely result in the understanding of which genes control skeletal health.


A MAJOR DETERMINANT of osteoporotic fracture risk, independent of other factors such as falls and aging per se, is bone mass.1–3 The acquisition of bone mass results from bone modeling and linear growth during skeletal development, whereas its maintenance in adults results from a coupling mechanism between the activities of bone formation and bone resorption. The processes governing acquisition of bone mineral have received far less attention than those related to the maintenance of adult bone density. With new emphasis on maximizing peak bone mass as a strategy for reducing fracture risk during the course of adult life, it is of increasing importance that the factors governing acquisition of bone mineral be understood.

The factors known to influence bone mass accumulation during growth include heredity, gender, dietary components (calcium, proteins), endocrine factors (sex steroids, calcitriol, insulin-like growth factor I), and mechanical forces. Quantitatively, the most prominent determinant appears to be heredity. Past research to identify specific genes that influence peak bone mass has focused mainly on the evaluation of candidate genes with identifiable polymorphisms. For example, polymorphisms of the gene for the vitamin D receptor have been associated with low bone mass in studies of some,4–7 but not all,8–11 populations. These inconsistent findings are not surprising considering the genetic heterogeneity and variations in gene frequency and penetrance that can exist between different populations. Moreover, it is very likely that other, as yet unknown, genes also contribute to the determination of bone mass.

We have employed a different genetic approach: quantitative trait locus (QTL) analysis. A quantitative trait (such as bone density) is a phenotypic measure that is continuously distributed and determined by multiple genes. A QTL is therefore defined as a site on a chromosome whose alleles influence a quantitative trait. The overall genetic control of a quantitative trait generally results from the collective influence of many genes, each of which may contribute only a small amount to the genotypic variance, making their identification difficult. This previously daunting task has recently been made feasible through the implementation of technologies to identify genetic variation (polymorphisms) at marker loci throughout the mammalian genome, and the development of statistical methods to detect and genetically map the chromosomal locations of QTLs.12–15 A major strength of this approach is that it enables the provisional identification of candidate genes in the absence of any prior hypothesis about the mechanism by which the phenotype is expressed. The identification of those chromosomal regions where marker allelic and trait variation significantly covary is now a straightforward (although large-scale) enterprise.16–20

Genetically distinct animal strains raised under strict environmental control are critical tools for defining genetic regulation. The availability of inbred strains, combined with its relative fecundity, has established the mouse as the best model system for the study of mammalian genetics and physiology.21 Importantly, genes identified in murine analyses can usually be readily mapped to particular human chromosomal regions because of the high degree of synteny that exists between the mouse and human genomes.22,23

Independently inbred strains of mice frequently exhibit numerous phenotypic differences, reflecting the substantial allelic variability that can exist between laboratory strains. These differences have been accentuated further by the introduction of recombinant inbred (RI) strains, which are derived by systematic inbreeding starting from a cross between two inbred strains. The BXD RI series was developed by Taylor24 from a cross between the C57BL/6 (B6) and DBA/2 (D2) strains. The BXD RI strains have had over 60 brother × sister matings, far in excess of the 20 required to establish inbred strains.24 Because an estimated four cross-over events have occurred per 100 cM chromosome length in the course of their inbreeding,24 a considerable amount of linkage disequilibrium has been fixed in these strains. Thus, each of the surviving 26 BXD RI strains represents chance recombinations of the progenitor chromosomes in a fixed (homozygous) state, and, under controlled laboratory conditions, any phenotypic differences among RI strains will largely reflect their genotypic variance.25 A diagrammatic presentation of the breeding process employed to generate the BXD RI series can be found in Silver's text on mouse genetics.26

The RI strains were originally developed as a tool for detecting and mapping major gene loci.25 The influence of a single major gene on a given trait can be inferred when RI strain means for the trait are found to fall in a bimodal distribution (i.e., some strains resembling the B6 parent, the remaining resembling the D2 parent, and none intermediate). Comparison of the strain distribution pattern (SDP) for that trait (i.e., which strains are “B6-like” and which are “D2-like”) can be made with the SDPs for known marker loci previously mapped to a particular chromosome region. A close match in SDPs between the unknown locus and a marker locus would thus allow provisional mapping to a chromosome region of the latter.24,25 In contrast to this major gene approach emphasizing bimodal distributions, the QTL approach focuses on both major and minor gene loci and is applicable to a broader range of phenotypes, including continuously distributed traits without apparent major gene effects. One version of QTL analysis is specifically tailored to existing mouse RI strains.20,27,28 We maintain a database for over 1500 mapped genetic markers, which are polymorphic in the B6 and D2 strains and therefore in their RI strains. The database also notes the position of each marker on the mouse chromosome, as determined by several mapping methods. A comparison of the RI strain means for a given phenotype with allelic SDPs thus allows one to begin to resolve quantitative genetic variation into individual regions of the genome that may potentially influence those traits. This approach particularly lends itself to the analysis of bone mineral density (BMD), where the major influence of a single gene locus appears to be absent.

The purpose of the present study was to investigate the genetic determinants of peak BMD in female mice. Peak whole-body BMD was evaluated in mice from the B6 and D2 parental strains and 24 BXD RI strains derived from their F2 cross. Body weight, another phenotype with a strong genetic component,29–32 is closely associated with bone mass.33,34 We therefore also examined the possibility that the body weight/bone mass relationship might be, at least in part, the result of shared genetic determinants.



All mice used in these experiments were bred under identical conditions at the Portland VA Veterinary Medical Unit from stock originally obtained from The Jackson Laboratory (Bar Harbor, ME, U.S.A.). Breeding mice were maintained for no more than three generations from stock obtained from The Jackson Laboratory. At the time of weaning, the mice were group housed (2–5 animals/cage) and maintained with ad libitum water and laboratory rodent chow (Diet 5001: 23% protein, 10% fat, 0.95% calcium, and 0.67% phosphorus; PMI Feeds, Inc., St. Louis, MO, U.S.A.) in a 12 h light/dark cycle (6:00 a.m. to 6:00 p.m.) at 21 ± 2°C. Female mice from the two progenitor strains, C57BL/6 and DBA/2, and 24 of the BXD RI strains were available in sufficient number (n > 6) for testing. All procedures were approved by the VA Institutional Animal Care and Use Committee and performed in accordance with National Institutes of Health guidelines for the care and use of animals in research.

Bone densitometry

Bone mineral measurements were performed with a pencil beam QDR 1500 densitometer from Hologic (Waltham, MA, U.S.A.) that was calibrated daily with a hydroxyapatite phantom of the human lumbar spine. Analysis was performed using the mouse whole body software (version 3.2), kindly provided by the manufacturer (Hologic). Densitometric analysis was performed on freshly sacrificed mice. Food was withheld the night prior to sacrifice (to eliminate confounding effects of undigested rodent chow on BMD assessment), and the mice were euthanized by CO2 inhalation. The animals were weighed to the nearest 0.1 g and immediately underwent dual-energy X-ray absorptiometry (DXA) scanning. The scans were done with a 1.27 mm diameter collimator, 0.76 mm line spacing, 0.38 mm point resolution, and an acquisition time of 9 minutes. The global window was defined as the whole body image minus the calvarium, mandible, and teeth. Data were gathered as bone mineral content (BMC, mg of hydroxyapatite) and BMD (mg/cm2).

Precision error (expressed as the coefficient of variation) for BMC was 0.99 ± 0.51% and for BMD was 1.71 ± 0.33%. These values are similar to those observed with whole body DXA measurements of rats35 and are superior to values reported with other techniques, such as radiogrammetry, quantitative computerized tomography (QCT), and neutron activation analysis.36,37 The accuracy of DXA measurements was demonstrated by determining whole body BMC and whole body calcium content on the same animals. Mice (n = 12) that were 31–98 days old and weighing 10–24 g were sacrificed. After determination of whole body BMC by DXA, each mouse carcass was ashed at 800°C for 24 h, and the residual material dissolved in hydrochloric acid. Whole body calcium from the ash was measured by automated titration on a Calcette calcium analyzer (Corning, Corning, NY, U.S.A.). We observed a strong linear relationship between whole body calcium content determined chemically and whole body BMC determined densitometrically (r2 = 0.96, p < 0.0001) (Fig. 1).

Figure FIG. 1.

Correlation between DXA-derived and chemically determined whole body calcium content. Bone mineral calcium content (WBCa[DEXA]) was measured in 12 female C57BL/6 mice weighing between 10 and 24 g as described in the Materials and Methods. The carcasses were then ashed and total body calcium (WBCa[chem]) was determined by automated titration on a Corning Calcette calcium analyzer. The best fit line shown was determined by linear regression with WBCa[chem] as the independent variable and WBCa[DEXA] as the dependent variable. There was a strong relationship between the two methods (r2 = 0.96, p < 0.0001).

QTL analysis

To test for markers associated with body weight or BMD, we used the approach first described by Plomin et al. to analyze RI strains.20 Over 1500 informative genetic markers have been genotyped in the BXD RI strains, mostly comprising microsatellite simple sequence repeat polymorphisms determined by polymerase chain reaction, so that the location and the B6- or D2-like repeat length of each of these alleles is known for each strain.38 Because of the derivation of these strains from B6 and D2 progenitors, each strain necessarily possesses either two copies of the B6 allele or two copies of the D2 allele. By convention, D2 alleles were coded with values of 1, and B6 alleles with values of 0 in our database. The QTL analysis involved the computation of Pearson product-moment correlations in which the genetic marker information was correlated with the quantitative phenotype. To identify the putative location of relevant QTLs, point biserial correlation coefficients (r) were calculated between the strain means for the phenotype and each marker in the data bank. Thus, a significant positive correlation between a phenotype and a marker would indicate an association between the D2 allele at that locus and high levels of the phenotype. Alternatively, a significant negative correlation would indicate a similar association of high phenotypic values with the B6 allele. The B6 and D2 progenitor strains were excluded from the QTL analyses because they are not recombinant. Significant correlations suggested associations between markers and the quantitative trait and also indexed the strength of the associations. The statistical significance of the correlation was the same as for the regression of phenotype on gene dosage (number of D2 alleles –0 or 2).39 Moreover, on a trait of interest, the p value of r was the same as that given by a two-tailed t-test between strains bearing the B6 and D2 alleles.

Data analysis

Differences in body weight and whole body BMD between the two progenitor strains, C57BL/6 and DBA/2, were assessed by simple t-test. Phenotype values for all strains are presented as the mean ± SEM. Analysis of variance (ANOVA) was used to detect significant strain differences and provide an estimate of heritability. Multiple regression analyses were performed for each of the two phenotypes (whole body BMD and body weight) to estimate the total amount of genetic variance accounted for by the significant gene marker associations. This analysis corrects for intercorrelations among markers, and in the process, provides an estimate of the number of effective factors (genes) contributing to the genetic variation for which the entire marker set accounts. In other words, due to intercorrelations among markers, some correlations of marker and phenotype are likely to be fortuitous. However, multiple regression analyses do not identify which markers these may be. The markers found in Table 1 were used in these analyses; only one marker from each linkage group was included. This approach has been compared with the same analysis in which every marker was included and has shown comparable results.27 ANOVAs and regression and linkage analyses were performed using the Systat statistical software application for the DOS environment (SPSS, Inc., Chicago, IL, U.S.A.).40 The theoretical rationale underlying the QTL analysis methods have been described previously.20,27 Associations attaining p < 0.01 were interpreted to indicate the provisional presence of a gene near the correlated marker contributing some proportion of trait variance. QTL analysis requires that many correlations be performed and as the number of correlations increases, the type I error rate relative to a single correlation similarly increases. One way to reduce the chance of such errors is to increase the required significance level and consider only those correlations that are significant at a higher probability (e.g., p < 0.001).41 However, in choosing this level of stringency, one risks not considering QTLs which may be important (i.e., type II error). Because we view QTL analysis in BXD RI strains as a preliminary screen for QTLs to be verified using other techniques, such as verification in an F2 population, we report here correlations of a less conservative p < 0.01. Some might view this criterion for significance as too relaxed, but there also appears to be support for the pursuit of hints and hunches by testing for linkage in larger data sets, as we agree must be done.42

original image


Phenotype assessment in the BXD RI strains

Whole body BMD was measured in 36 female B6 progenitor mice sampled between the ages of 31 and 196 days (Fig. 2). As expected, young mice undergo a rapid period of bone mineral acquisition, but by the age of 80–90 days, the accrual rate begins to plateau. Whole body BMD peaks at the same age in D2 progenitor mice (data not shown). Consequently, we examined mice at 84 ± 4 days (or 12 ± 0.5 weeks) of age, a time when the majority of bone mass has been attained but is not yet subject to remodeling processes that likely have their own share of genetic influences. Shown in Fig. 3 are body weight and BMD values for groups of 12-week-old female B6 and D2 mice. Body weights for the two progenitor strains were very similar; however, as has been demonstrated in a previous report,43 B6 animals exhibited considerably greater bone mass than their D2 counterparts (p < 0.001).

Figure FIG. 2.

Cross-sectional study of age-related changes in whole body BMD. Thirty-six female C57BL/6 mice between the ages of 31 and 196 days were examined. Whole body BMD was determined as described in the Materials and Methods. Acquisition of adult peak bone density was achieved by 80–90 days of age.

Figure FIG. 3.

Phenotype data for BXD RI progenitor strains (C57BL/6 and DBA/2). Body weight and whole body BMD measurements were performed on animals 84 ± 4 days of age as described in Materials and Methods. Individual data points as well as means ± SD are shown: (A) body weight, (B) peak whole body BMD.

The distributions of RI strain means for body weight and peak whole body BMD are shown in Fig. 4. Many individuals of each genotype (i.e., each RI strain) were tested, permitting a high degree of accuracy in determining the relationship between genotype and the two traits. The distributions of both of these traits were continuous and in most cases significantly different from the progenitor strain values. One-way ANOVA, grouped by strain, revealed wide strain variation in whole body BMD (F23,350 = 7.385, p < 0.0001) and body weight (F23,350 = 25.407, p < 0.0001). Peak whole body BMD varied by 26% across the BXD RI panel, ranging from 53.3 ± 1.1 mg/cm2 in BXD-30 to 67.4 ± 1.1 mg/cm2 in BXD-15 (Fig. 4A), and body weight varied by 75%, ranging from 14.1 ± 0.3 g in BXD-30 to 24.8 ± 0.6 g in BXD-9 (Fig. 4B). The coefficient of variation for the individual strain means (a measure of the environmental variance) ranged from 4–13% for BMD and 5–18% for weight. We also estimated the split-half reliability coefficient based on the correlation of odd with even numbered subjects for strain means, which indexes the reliability or consistency of measurement of the phenotype (genotypic differences). The correlation between the two halves of the data using the Spearman-Brown formula was 0.84 (p < 0.0001) for whole body BMD and 0.91 (p < 0.0001) for body weight, thus demonstrating that both whole body BMD and body weight in 12-week-old female mice are reasonably reliable traits. Our BMD results agree well with those reported by Weinstein and Jilka et al.,44,45 who also have used DXA to determine whole body BMD of other strains of laboratory mice. The frequency distribution of the strain means was not significantly different from normal (p > 0.5) for either of these traits, and thus polygenic control of both of these phenotypes was inferred.

Figure FIG. 4.

Phenotype data for the BXD RI series expressed as strain means ± SEM. Body weight and whole body BMD measurements were performed on animals 84 ± 4 days of age as described in the Materials and Methods. Results for the progenitor lines, DBA/2 (D2) and C57BL/6 (B6), are shown as empty and filled columns, respectively. Data for each of the BXD RI strains is shown as cross-hatched columns. A total of 407 mice were examined, with the number of mice examined in each strain recorded above the corresponding error bar. Missing strains either do not exist because they became extinct during the inbreeding process (e.g., BXD-3) or are poor breeders (e.g., BXD-23) so that insufficient numbers were available for testing. (A) Peak whole body BMD, (B) body weight.

As these genetically distinct RI strains were raised in a controlled environment (nutritional intake, physical activity, etc.), the differences observed in peak BMD and body weight are primarily the result of genetic variation. Heritability (h2), or the proportion of the total variation due to additive genetic sources, can be estimated from the r2 from a one-way ANOVA by strain. The heritability reflects the reliability of predicting phenotype from genotype, which is critically important for successful QTL mapping. The estimated narrow sense heritability for body weight was 0.60 (p < 0.0001) and for whole body BMD was 0.35 (p < 0.0001) in the RI series. These estimates are highly significant, thus demonstrating a substantial degree of genetic control of these traits in the RI set. Correcting whole body BMD for variations in body weight resulted in a 25% reduction in the heritability of BMD to 0.26 (p < 0.001). The heritability estimates for bone density derived here are similar to those in human population46,47 in terms of demonstrating that these traits clearly have a genetic basis. Moreover, these heritability estimates are well above those found to lead to successful QTL mapping of other traits.48

QTL analysis

To identify the putative location of relevant QTLs, the strain mean values for body weight and whole body BMD were correlated with the allele (B6 or D2) at each of the 1522 polymorphic loci in our database. QTL analysis of the BXD RI strains identified 10 discrete chromosomal regions on chromosomes 1, 2, 7, 11, 14, 15, 16, 18, and 19 that were associated with peak whole body BMD (Table 1) and 4 discrete chromosomal regions on chromosomes 4, 6, 9, and 14 that were associated with body weight (Table 1). A p < 0.01 (two-tailed, single test) α level of significance was adopted for each correlation. The chromosome locations in centiMorgans were taken from the mouse linkage map of Silver et al.23 At each provisional QTL there was a significant clustering of markers showing similar correlations. For simplicity, only the marker showing the highest correlation with the phenotype is shown, but a list of all significant associations is available from the authors.

Because of the many correlation coefficients that were calculated for each measure in these QTL analyses (n = 1522), it is expected that some of these associations will be significant merely due to chance at the p < 0.01 level. However, because many of these markers are linked, and thus are not independent, the number of fortuitous correlations expected for linked regions will be much less than the expected 1% because of linkage disequilibrium fixed by inbreeding. To judge the global significance of our results, computer simulations41 were used to calculate the number of regions that would be expected to reach these levels of significance simply by chance (that is, the number of false-positive linkages). This procedure has advantages over traditional approaches to correcting nominal significance levels because it allows for the number, informativeness, and distance between markers that were actually utilized in our search. The expected number of regions of false-positive linkage was 1 for p < 0.002 (compared with three observed linkages for BMD) and ∼4 for p < 0.01 (compared with 10 observed linkages for BMD), so the observed number of regions meeting these statistical criteria for linkage are statistically larger (p < 0.05) than would be expected by chance.41


Peak bone mass is a major determinant of risk of osteoporotic fracture. However, BMD is a complex trait whose expression is complicated by environmental influences and polygenic inheritance. Experimental animal models furnish a means to largely circumvent confounding environmental factors, and the availability of dense genetic maps based on molecular markers now provides the opportunity to resolve quantitative genetic variation into individual regions of the genome (QTLs) influencing a given trait. Thus, investigations with animal models should provide a general means for the identification of genetic factors contributing to the acquisition and maintenance of skeletal mass.

Recent studies of various genetically homogeneous inbred mouse strains have revealed significant differences in skeletal phenotypes.43–45,49–51 We have taken advantage of the large panel of extensively genotyped RI strains in the BXD series to begin to map genes influencing bone mass. For two important reasons, the BXD RI series is particularly useful for this analysis. First, the progenitor B6 and D2 strains differ substantially in their peak whole body BMD values, which practically ensures that this trait will be substantially heritable. Indeed, this was the observed result (see Fig. 4). Second, the B6 and D2 strains are highly polymorphic14 and the BXD series (24 strains were examined here) is sufficiently large to detect the larger gene effects.41 There are over 1500 polymorphic SDPs for marker loci in this series. The resulting genetic map provides adequate to excellent coverage for all chromosomes. After correction for redundancy among the significant correlations, 10 provisional QTLs for peak whole body BMD in female mice were detected on nine different chromosomes (Table 1). We can estimate that with α = 0.01, up to four of the QTLs in Table 1 may be false positives. However, making the criteria more stringent would increase the likelihood of type II errors (i.e., missing actual influential loci). Our strategy therefore is to use the BXD strain mean analysis as an initial screen, suggesting provisional regions of influence that await confirmation or rejection in subsequent analyses in additional mapping populations derived from the same progenitors. The approach for subsequent verification of the QTLs provisionally reported here has been discussed previously.41,52 We plan to examine other populations to follow-up the results reported here with the goal of attaining the aggregate levels of statistical significance required to confirm linkage that have been proposed by Lander and Kruglyak.42

It is worth noting the somewhat contradictory reports in the literature concerning the skeletal phenotype of the B6 parental strain. Kaye and Kusy examined five inbred strains of mice, including D2 and B6, and found bone mass as well as bone strength to be highest in the B6 animals.43 This study found a 25% difference in tibial bone mass between the D2 and B6 strains, a value remarkably similar to the 21% difference in whole body BMD observed in the current study (see Fig. 3). In contrast, Beamer et al. examined femoral BMD in 11 inbred strains of mice, including D2 and B6 with peripheral quantitative computed tomography (pQCT).50 Femoral BMD of B6 mice was found to be significantly lower than all others examined. The disparity among studies may result from differences in animal husbandry (nutritional factors, activity, housing, etc.), any of which could exert significant effects on bone mass acquisition. Alternatively, the different methods employed for assessing bone mass (pQCT vs. DXA), may be involved. As discrepancies between environmental conditions and/or measurement techniques could have a major impact on the identification of genetic loci associated with BMD, it is clear that the specific methodology for skeletal phenotyping must be carefully considered. However, the fact that our BMD determinations correlate so highly with direct bone calcium measurements (Fig. 1) increases our confidence in the present methods.

It is feasible to identify potential candidate genes for some of the QTLs identified here by extrapolation from map positions of markers and BMD-related loci given in the Mouse Genome Database.23 Many of the identified sites map near genes encoding hormones, structural proteins and cell surface receptors that are intricately involved in skeletal homeostasis (Table 2). The absence of candidate genes at three loci (chromosomes 1, 18, and 19) raises the prospect that this analysis may result in the identification of novel genes that make important contributions to skeletal metabolism. In the vicinity of the three most significant loci are candidate genes of considerable interest—two members of the bone morphogenetic protein (BMP) family: BMP2 on chromosome 2 and BMP4 on chromosome 14, as well as two calciotropic hormones (parathyroid hormone and calcitonin) and the type I insulin-like growth factor receptor which reside in close proximity to the provisional QTL on chromosome 7. Whether any of these candidate genes are actually responsible for the QTL effects measured here will require further research.

original image

Comparative gene mapping in humans and rodents has revealed evidence for substantial conservation of gene order during mammalian evolution. Based on the considerable linkage conservation (estimated to be ∼80%) between mouse and human genomes, findings in this animal model system should aid in the identification of specific candidate genes for study in humans. Shown in Table 2 are the homologous sites in the human genome for the 10 provisional QTLs identified in this analysis of the BXD RI set. It is interesting to note that the segment of mouse chromosome 7 containing marker D7Mit234 is thought to be homologous with a section of human chromosome 11. A recent study in humans has linked a gene causing high bone mass to this same region of human chromosome 11 (11q12–13).53 Studies in the baboon also suggest this same region harbors one or more QTLs that influence BMD acquisition.54 It is important to point out though, that these kinds of linkage studies can only identify relatively broad chromosomal regions that may contain QTLs regulating BMD and should therefore be regarded with caution. The possibility that mouse chromosome 7 contains a QTL regulating bone mass acquisition, however, can now be directly pursued by examining whole body BMD in congenic strains of mice that are genetically identical except for selected segments of this chromosome.

Epidemiological studies have clearly demonstrated that body weight is a very strong predictor of BMD.55–58 However, the mechanism underlying the strong association of weight with BMD is poorly understood. The coincidence of increased body weight with increased BMD could stem from environmental factors such as complementary nutritional effects on body composition and skeletal mass or the association could largely be the result of mechanical loading.59 In addition to environmental causes, body weight and BMD may be modulated by linked genes or perhaps even the same genes. In the current analysis, 4 QTLs for body weight were identified in the BXD RI series (Table 1). All four of these loci had been previously identified by Keightley et al.60 in a QTL analysis of mouse lines divergently selected for body weight from a base population derived from B6 and D2 parental strains. Interestingly, the Ptprg locus that is linked to body weight in both studies is also strongly linked in the current study to inherited variation in BMD. These findings raise the intriguing possibility that body weight and peak BMD may be influenced by linked genes or perhaps by common genes with pleiotropic effects. Although it is difficult to distinguish between these two possibilities using RI strains, further studies in congenic lines should be useful for linked genes that independently regulate body weight and BMD.

In summary, QTL analysis of the BXD RI strain series has provisionally identified 10 chromosomal sites linked to peak bone mass development. Several of these sites are associated with candidate genes of interest in skeletal biology. The presence of an association between bone density and a site on mouse chromosome 7 is particularly interesting in view of a similar association recently reported between bone density and the syntenic region on chromosome 11 in both humans53 and baboons.54 The finding that bone density and weight are both associated with a similar region on chromosome 14 may reflect a common genetic determinant. The present report constitutes meaningful progress toward the detection and mapping of the QTLs that influence attainment of peak BMD. In addition to providing an efficient screening tool for mapping studies, the experimental system utilized here should be useful in validating these provisional findings, and in eventually identifying the specific genes which influence skeletal physiology and disease.


The authors thank Elizabeth Allen, Virginia Chambers, Amy Carlos, Carrie McKinnon, Shelia Orwoll, and Sandra Veith for skilled technical assistance and John Crabbe for encouragement and advice. This work was supported by funds from the National Institutes of Health (AA 10760 and AR 44659), Medical Research Service of the V.A. and the Medical Research Foundation of Oregon.