Genetic loci determining bone density in mice with diet-induced atherosclerosis

THOMAS A. DRAKE1, ERIC SCHADT6,7, KAMBIZ HANNANI2, J. MICHAEL KABO2, KELLY KRASS5, VERONICA COLINAYO5, LLOYD E. GREASER III3, JONATHAN GOLDIN3 and ALDONS J. LUSIS4,5

1 Departments of Pathology and Laboratory Medicine
2 Orthopedic Surgery
3 Radiology
4 Medicine
5 Microbiology, Immunology and Molecular Genetics
6 Biomathematics, University of California, Los Angeles, California 90095
7 Department of Bioinformatics, Rosetta Inpharmatics, Kirkland, Washington 98034


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
This study investigates the phenotypic and genetic relationships among bone-density-related traits and those of adipose tissue and plasma lipids in mice with diet-induced atherosclerosis. Sixteen-month-old female F2 progeny of a C57BL/6J and DBA/2J intercross, which had received an atherogenic diet for 4 mo, were examined for multiple measures of femoral bone mass, density, and biomechanical properties using both computerized tomographic and radiographic methods. In addition, body weight and length, adipose tissue mass, plasma lipids and insulin, and aortic fatty lesions were assessed. Bone mass was inversely correlated with extent of atherosclerosis and with a prooxidant lipid profile and directly correlated with body weight, length, and, most strongly, adipose tissue mass. Quantitative trait locus (QTL) analysis, using composite interval mapping (CIM) and multi-trait analysis, identified six loci with multi-trait CIM LOD scores > 5. Three of these coincided with loci linked with adipose tissue and plasma high-density lipoprotein. Application of statistical tests for distinguishing close linkage vs. pleiotropy supported the presence of a potential pleiotropic effect of two of the loci on these traits. This study shows that bone mass in older female mice with atherosclerosis has multiple genetic determinants and provides phenotypic and genetic evidence linking the regulation of bone density with adipose tissue and plasma lipids.

adipose tissue; plasma lipids; quantitative trait locus analysis; leptin; biomechanics; genetic pleiotropy


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
OSTEOPOROTIC FRACTURES are a major cause of morbidity in the aging population. Susceptibility is multifactorial, with genetic and environmental factors influencing it. Peak bone mass is an important determinant of osteoporosis susceptibility later in life, and numerous twin and family studies in humans have demonstrated significant heritability of peak bone mass at various skeletal sites (48, 50). In these, estimates of the genetic contribution to variability have been between 50 and 80%. Physiological and pathological determinants that may secondarily affect bone mass are prevalent in the aging population and are oftentimes themselves strongly influenced by genetic factors. These include body size, obesity, and diabetes mellitus (9, 10, 43, 46). Clinical studies have provided evidence that atherosclerosis is also associated with reduced bone density (3, 19, 54). Data suggest that lipid oxidation, which plays such a central role in the pathogenesis of atherosclerosis, may also promote osteoporosis, thereby linking lipid metabolism and inflammation to regulation of bone density (3840). Thus complex interrelationships exist among adipose tissue, lipid metabolism, glucose and insulin regulation, and inflammation, which may influence both atherosclerosis and reduction in bone mass.

The mouse has proved to be a useful model for the genetic dissection of complex traits. Quantitative trait locus (QTL) analysis in intercrosses or backcrosses of selected inbred strains allows survey of the entire genome for loci containing genes that influence a particular trait (26). We have previously used this technique to identify genetic loci affecting myocardial injury and calcification, lipid metabolism, and obesity, among other complex traits in mice fed an atherogenic diet (17, 27, 32). Inbred mouse strains have been shown to have significant heritable differences in a variety of bone phenotypes, including bone mass and susceptibility to osteoporosis (2, 6, 12, 22, 35, 55). Application of the technique of QTL analysis in mice for peak bone mass has begun to identify genetic loci associated with this trait (5, 7, 23, 47, 51).

In the above instances, which are typical of most QTL analyses, single or several closely related traits were examined, with studies designed to minimize potentially complicating variables. In the current study we explore the use of QTL and other statistical analyses applied to an F2 intercross population for examining the phenotypic and potential genetic relationships among multiple complex traits. Of particular interest is the potential for assessing whether the presence of pleiotropic genetic loci may account for phenotypic correlations among traits, by analyzing for colocalization of QTLs for different traits, and applying additional statistical tests. Although these analyses cannot prove the presence of pleiotropy, they can generate important testable hypotheses as to whether pleiotropy may be present. We have applied these concepts to the setting described above of determination of bone density in older animals with hyperlipidemia and atherosclerosis, using an intercross between strains C57BL/6J and DBA/2J in which an atherogenic diet was administered. We report here the phenotypic and genetic relationships observed among the above metabolic and morphometric parameters, atherosclerotic lesion formation, and various measures of bone mass and density, including biomechanical properties, in mature (16 mo old) F2 female mice with diet-induced atherosclerosis. Advanced statistical analyses are employed to examine these multiple traits and assess potential pleiotropic effects of genetic loci identified.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

Mice.
Parental mice were purchased from the Jackson Laboratories (Bar Harbor, ME). All mice were housed under conditions meeting the guidelines of the Association for Accreditation of Laboratory Animal Care. Females of strain C57BL/6J (B6) were mated with DBA/2J (DBA) males. F1 progeny were then intercrossed to produce F2 intercross progeny. After weaning at ~21 days, female mice were selected and placed on a diet of rodent chow (Purina 5001) containing 17.5% protein, 11% fat, 0.8% calcium, and 0.5% phosphorus. At 12 mo of age, this was switched to an atherogenic high-fat, high-cholesterol diet for another 16 wk. This consisted of 75% chow supplemented with 7.5% cocoa butter, 2.5% dextrose, 1.625% each of sucrose and dextrin, 1.25% cholesterol, and 0.5% sodium cholate, with 0.73% calcium and 0.68% phosphorus (diet 90221; Harlan Teklad, Madison, WI). The mice were given free access to food and had a 12:12-h light-dark cycle maintained throughout. Body weight and length (nose to anus) were determined prior to death. At death, blood was collected by retro-orbital sampling into EDTA anticoagulant, spleens were removed for DNA isolation, and fat pads (omental, retroperitoneal, parametrial, and subcutaneous) were removed and weighed. Hearts and ascending aortas were removed and embedded in OCT compound and frozen at -70°C. Carcasses were placed in airtight individual bags and frozen at -20°C. Subsequently, the left femurs were dissected and stripped of soft tissue and placed directly in 10% buffered formalin. The right femurs were similarly obtained and were wrapped in gauze soaked in normal saline, placed in airtight plastic bags, frozen at -70°C, and used for biomechanical analyses.

Femoral bone analyses.
Bone mineral mass and indices for bone mineral density (BMD) were measured radiographically and by computerized tomography. For the former, all samples were collectively exposed on a single X-ray film using a Faxitron and subsequently developed under similar conditions. Separate X-rays were obtained for the anterior-posterior (AP) and lateral orientations. The X-rays were digitized and comparatively analyzed. Image 1/AT image processing software on a 486 IBM compatible computer and a CCD 72 video camera (Dage-MTI, Michigan City, IN) with Fujinon Zoom TV lens were used to digitize the X-rays. Image 1/AT was used to analyze the brightness and contrast level as well as the cortical diameter at a fixed axial location of the mouse femur diaphysis. Indices calculated from the digitized measurements for the AP as well as the lateral views were the average gray scale (integral GS) for the entire femur and for cross sections at fixed locations of the diaphysis and intertrochanteric regions [radiographic bone mineral content (BMC-R)]. Each of these was averaged by the total area or respective bone width (D). This method of normalizing for variation in bone size yielded the best correlation with computed tomography (CT) determined mineral content (r = 0.77) and density (r = 0.71) and is analogous to validated measures used for metacarpal bone density determination in human studies (1, 14, 16, 49). Values given are relative densitometry units. Variation in film and radiation exposure at various points on the film was determined to be less than 2%. The coefficients of variation for the primary measurements (integral GS, D) were less than 2%, based on repeated measures by a single observer on two separately prepared radiographs. Only results for the indices derived from the AP views is presented subsequently, as results for lateral views were strongly correlated with these and in almost all instances yielded results for phenotypic and genotypic analyses that were comparable.

CT determination of bone density was performed using electron beam computed tomography (EBCT). This technology has been used clinically for the past decade to detect and quantitate calcium deposits in coronary arteries and was adapted for determination of BMC in mice femurs. Femurs obtained as described above were embedded in a semisolid reusable gel phantom and scanned along with calibration bar phantoms containing known quantities of calcium hydroxyapatite in an Evolution EBCT scanner (Siemens Medical Systems, Iselin, NJ). Bones were oriented perpendicular to the scanning plane, and a total of 40 axial slices were obtained to image the entire phantom. Imaging employed a 1.5-mm collimation continuous volume scan mode, and images were reconstructed using vendor supplied sharp reconstruction kernel, a 12.7-cm field of view, and a 512 x 512 matrix. The calcium score was quantitated using a modification of the vendor supplied scoring software and a calcium threshold of 130 Hounsfield units (HU). Bone volume was determined by summation of the cross-sectional areas of each axial image multiplied by the 1.5-mm collimation thickness. BMD was assessed as the average HU value for each bone, and a measure of BMC was obtained by multiplying the average HU value by the bone volume. This method was demonstrated to have a highly significant correlation of the BMC determination with ash weight (r2 = 0.909), and a between-run coefficient of variation of 1.19%.

Torsion testing.
The proximal and distal mouse femora were potted into square molds using polymethylmethacrylate cement. Torsional testing was subsequently performed using a Burstein-Frankel torsion tester (8, 33). A deadweight pendulum was used to reach torsional failure in less than 0.1 s. The femora were tested in a random sequence and always in external rotation. A torque/rotation curve was recorded using a storage oscilloscope, and the results were photographed. The curve was analyzed using a computer program written in BASIC to measure the failure torque (N-m), angular deformation at failure (rad), stiffness (N-m/rad), and toughness (N-m·rad). These values were calculated using a linear regression model over the linear aspect of the curve. Reproducibility of the technique determined by testing of left and right femurs from 10 mice was 90%. Bones were excluded from the analysis if they did not demonstrate a clean spiral fracture indicating pure torque failure, or the angular displacement was excessively large with low torque indicating slippage. Satisfactory response curves were obtained on 117 of 140 attempted. Unsatisfactory test results were most commonly due to slippage of the ends in the potting material.

Plasma biochemical measurements and aortic fatty lesion quantitation.
Plasma total cholesterol, high-density and low-density lipoprotein (HDL and LDL) cholesterol, triglycerides, and free fatty acids were measured as previously described (32). Plasma leptin and insulin were measured by immunoassay according to manufacturers instructions, using kits specific for mouse proteins (R&D Systems, Minneapolis, MN; and Crystal Chem, Chicago, IL; respectively). Quantitation of aortic fatty lesion formation was performed as previously described. This entails histomorphometric analyses of Oil red-O stained serial frozen sections of the ascending aorta from the aortic valves through the aortic arch (42).

Linkage and data analyses.
A complete linkage map for all chromosomes except the Y chromosome was constructed at an average density of 13 cM using microsatellite markers as previously described (17). The constructed linkage map with all markers used is available on request. PCR primers for microsatellite markers were purchased from Research Genetics (Huntsville, AL); methods for use were as described. ANOVA, correlation, regression analyses, and hierarchical clustering were performed using StatView version 5.0 (SAS Institute, Berkeley, CA) and SPLUS 2000 (MathSoft). Linkage maps were constructed and QTL analysis was performed using MapMaker QTL (29) and QTL Cartographer (4). "Log of the odds ratio" (LOD) scores were calculated at 1-cM intervals throughout the genome for each of the traits under consideration. Before analysis, the normal distribution assumption underlying parametric tests for linkage was assessed for each trait by examining the histogram, as well as the skewness and kurtosis statistics for the trait values. In addition, despite its low power, the Kolmogorov-Smirnov test was applied to assist in determining whether a given trait was sampled from a normal distribution. Traits demonstrating obvious departures from normality were log-transformed or square root-transformed and then reassessed for normality. Linkage results for traits in which normality was questionable were verified using nonparametric association tests, such as the Wilcoxon rank sum test, in determining marker/trait associations. Trait values falling outside the mean ± 3 SD were considered outliers and excluded from analysis, accounting for the variation in numbers given in Table 1. A LOD score of 2.8 was the threshold for suggestive evidence of linkage, and a LOD score greater than 4.3 was interpreted as significant evidence for linkage, as proposed by Lander and Kruglyak for an F2 intercross with 2 degrees of freedom (25).


View this table:
[in this window]
[in a new window]
 
Table 1. Femur phenotypes measures in C57BL/6J x DBA/2J intercross F2 mice at 16 mo of age

 
Composite interval mapping and multiple trait analysis.
In addition to the standard interval mapping techniques employed to detect loci affecting the traits of interest, additional analyses were performed to determine whether controlling for genetic background variation using markers outside a putative region of linkage and whether multiple traits considered simultaneously could increase evidence for linkage. Composite interval mapping (CIM) enhances the ability to detect linked QTL over standard interval mapping methods by incorporating markers as cofactors into the statistical model for marker-trait association (30). Explicitly including these cofactors into the model serves to control for genetic background effects that would otherwise be confounded with the experimental error terms, which, in turn, reduce the chances of detecting linked QTL. All CIM analysis was performed using QTL Cartographer (4). Given multiple quantitative traits, CIM analysis can be extended to consider multiple traits simultaneously, potentially dramatically increasing the power to detect loci affecting the traits of interest. Joint CIM analysis was first described by Jiang and Zeng (18) and is currently implemented in the QTL Cartographer software.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

Interrelationships among phenotypic traits.
Construction of an F2 intercross from inbred parental strains generates a genetically heterogeneous population. The phenotypes measured, with the mean and standard deviations for each, are listed in Table 1. As would be expected with complex quantitative traits, these showed a wide range of values that were normally distributed in most cases. Hierarchical clustering analysis was performed to assess the overall relationships among the traits (Fig. 1). All pairwise correlations were computed between the quantitative traits, and then a standard agglomerative hierarchical clustering algorithm was applied to the resulting correlation matrix (21). All of the bone mass and density traits, including cortical thickness, formed a common cluster, suggesting it was reasonable to consider that there may be common biologic and genetic determinants among these. As a group, these were most closely related to another cluster composed of body weight and adipose tissue measures, including leptin. Bone biomechanical measures clustered most closely with several metabolic traits, including plasma lipids and aortic lesions.



View larger version (24K):
[in this window]
[in a new window]
 
Fig. 1. Hierarchical clustering analysis for all quantitative traits discussed in the text. The height along the y-axis represents the distance between two clusters, determined when two clusters are merged (the more similar traits are, with respect to corresponding measurements for each trait, the closer they will appear in the cluster tree). See Table 1 for complete description of abbreviations.

 
In these F2 mice fed an atherogenic diet, there were significant interrelationships among traits potentially associated with bone mass and density (Table 2). Administration of an atherogenic diet to susceptible mice induces a prooxidant state, manifested by aortic fatty lesion development, and hyperlipidemia with increased LDL but reduced HDL (37, 42). Aortic fatty lesions were observed in over 90% of F2 animals and were inversely correlated with plasma HDL/LDL ratios. In mice, adipose tissue mass is positively correlated with plasma HDL, and this was observed here also (32, 36). Consistent with this was an inverse correlation of adipose tissue (measured as total fat pad mass, adiposity, or plasma leptin levels) with aortic fatty lesions. Adipose tissue also influences plasma insulin levels, and positive correlations were found among insulin levels, plasma lipids, and adipose tissue measures. Body weight is influenced by body size and adipose tissue mass, and the expected correlations among these were observed in these animals.


View this table:
[in this window]
[in a new window]
 
Table 2. Correlation matrix for aortic fatty lesions and selected metabolic and morphometric parameters measured in F2 female mice

 
All of these were correlated with CT-measured BMD and with additional bone-related parameters. Body weight, body length, and adipose tissue measurements all showed significant correlations with measures of bone mass, density, and biomechanics. Measures of adipose tissue showed the strongest correlations whether expressed as total fat pad mass or as adiposity (range of r values 0.306–0.584). An independent confirmation of this association was obtained by measuring plasma leptin levels, which had a similar pattern of effect. Body weight showed significant correlations with many of these same measures, though the correlation coefficients were lower (r value range 0.144–0.354). Body length showed the strongest correlations with measures of BMC and BMD for the entire femur (r = 0.410 and 0.353, respectively). It was also significantly correlated with fracture strength and stiffness. Stepwise regression analysis confirmed the independent influence of adipose tissue, leaving adiposity as the only variable in the model for radiographic density or cortical thickness. For total mineral content by CT, stepwise regression identified both adiposity and body length as independent contributors.

Analysis of plasma lipid correlations with bone measures showed that a prooxidant lipid profile was associated with reduced bone mass and density. HDL levels were positively correlated with bone measurements, whereas LDL levels were inversely correlated. Expressing these as a ratio of HDL to LDL increased the correlation coefficients, which were significant for almost all the bone measurements (r values for bone density measurements ranged from 0.202–0.315). Aortic fatty lesions showed significant inverse correlations with bone mass and density. As plasma lipid levels are themselves correlated with body adiposity and morphometry, stepwise regression analyses were performed to determine whether these were independent factors influencing bone or were secondary. Results showed that among the body morphometric and lipid- and lesion-related parameters discussed, adiposity was the major determinant for all, with body length being the only other contributing independent variable for the bone mass parameters. These measured variables together accounted for ~25% of the observed variability in the various bone mass and density parameters.

Identification of genetic loci controlling bone-related traits.
A complete linkage map for all the chromosomes except the Y chromosome was constructed using microsatellite markers, with an average interval of 13 cM. CIM was performed as implemented by QTL Cartographer software, using four background markers identified by forward/backward regression analysis, in addition to two flanking markers for each trait and locus. This analysis identified six loci showing linkage with multiple bone-related traits (Table 3). These six loci were further analyzed using the multi-trait program (JZMAPQTL) of QTL Cartographer, incorporating up to four bone-related traits showing linkage to the respective loci. At each locus, the LOD score obtained by the multi-trait analysis significantly exceeded that obtained by CIM for any one trait and were highly significant (LOD >5) for all six loci (Table 3; Fig. 2).


View this table:
[in this window]
[in a new window]
 
Table 3. Bone density-related QTLs identified by individual and multitrait CIM, as described in the text and corresponding to Fig. 2

 


View larger version (30K):
[in this window]
[in a new window]
 
Fig. 2. Single and multi-trait composite interval mapping (CIM) log the odds ratio (LOD) score plots for chromosomes with quantitative trait loci (QTL) for bone-related traits, obtained by CIM at 1-cM intervals as described in the text. For each chromosome, the curve for multi-trait CIM is shown along with the single trait CIM for the trait with the highest peak LOD score. In all cases, the joint analysis LOD score curve is represented by the bold line, and a LOD score curve for a representative single trait CIM analysis is also given in each case, represented by the lighter line. These latter are: BMD-R for chromosomes 2 and proximal 15, BMC-CT for chromosome 3, BMC-Rit for chromosome 6, cortex for chromosome 7, and BMC-Rd for distal chromosome 15.

 
Because each of the traits in this analysis had significant associations with QTL on other chromosomes, ANOVA methods, including all cross-terms for two-way and three-way interactions, were applied to determine whether there were significant epistatic effects between QTL identified for each trait. The ANOVA model for radiographic BMD yielded a significant interaction term between the QTL near markers D6Mit198 and D15Mit13 (P = 0.04). No other significant interactions were detected (including all higher-order interactions). The nature of the epistatic interactions could not be precisely determined from ANOVA results, as multiple interaction terms appeared to contribute significantly to the result.

In addition to considering multiple traits simultaneously and detecting epistatic effects, it is of interest to determine the additive and dominance effects for a given QTL. The CIM model and joint CIM models provided in the QTL Cartographer software automatically generate test statistics for assessing additive and dominance effects. Table 3 summarizes the additive and dominance effects for each of the six loci.

Colocalization of QTL for bone and nonbone traits: potential pleiotropy vs. close linkage.
QTL analyses were also performed for the morphometric and metabolic parameters that showed phenotypic correlations with the bone-related traits. One or more QTL each were identified for adipose tissue traits, leptin, body weight and length, plasma lipids, and aortic lesions. Three of the six QTL for bone-related traits were found to be adjacent to or overlapping with QTL for nonbone traits (chromosome 2, adipose tissue and plasma HDL; chromosome 3, plasma LDL and body length; chromosome 6, adipose tissue and plasma HDL) (Figs. 35). Among bone and nonbone traits with adjacent or colocalizing QTL, allele effects were similar in each case.



View larger version (21K):
[in this window]
[in a new window]
 
Fig. 3. Single and multi-trait CIM LOD score plots for bone and non-bone-related traits for chromosome 2. Joint LOD score curves are given in black, bone traits (BMD-Rd and BMD-R) are given in blue, adipose tissue is in green (adipose mass), and plasma high-density lipoprotein (HDL) is in yellow.

 


View larger version (17K):
[in this window]
[in a new window]
 
Fig. 4. CIM LOD score plots for bone and non-bone-related traits for chromosomes 3. The LOD score curves for bone trait (BMC-CT) is given in blue, plasma LDL/VLDL in yellow, and body length in red; the joint LOD score curve for all three traits together is in black. (The gray curve reproduces the joint bone LOD score curve from Fig. 2 for comparison).

 


View larger version (16K):
[in this window]
[in a new window]
 
Fig. 5. CIM LOD score plots for bone and non-bone-related traits for chromosomes 6. The LOD score curve for the bone trait (BMC-Rit) is given in blue, adipose tissue traits are given in bold green (subcutaneous) and green (adipose tissue mass), and plasma HDL is in yellow. The joint LOD score curve for BMC-Rit, adipose tissue mass and plasma HDL taken together, is in black.

 
The colocalization of QTLs and concordant allele effects among traits that had significant phenotypic correlation suggested the possibility that pleiotropy of a common underlying gene rather than close linkage of separate genes may be responsible. Data were evaluated using a two-step process. First, in cases where bone and nonbone traits gave rise to adjacent or colocalizing QTL, multi-trait CIM methods were employed to determine whether the significance of the linkages could be increased by considering bone and nonbone traits simultaneously. Second, if the multi-trait analysis resulted in detecting a significant effect, then the likelihood of potential pleiotropy vs. close linkage was tested to determine whether the significant effect was potentially due to a single QTL having pleiotropic effects across all traits considered simultaneously or to closely linked QTL. This was tested directly by application of the statistical test developed by Jiang and Zeng (18). As set forth by Jiang and Zeng (18), to test the hypotheses of pleiotropy vs. close linkage for two coincident QTL of interest, the multi-trait CIM likelihood function must be reformulated. The hypotheses of interest (H0 and H1) involve the position p1 of the QTL having an effect on trait 1 and position p2 of the QTL having an effect on trait 2 and are given by


The alternative hypothesis indicates that the QTL are nonpleiotropic and are located at different map positions. The likelihood for H0 is the same as that given for the multi-trait CIM model. However, the likelihood for the alternative is that developed by Jiang and Zeng (18). Using the prescription set forth by Jiang and Zeng, calculation of the maximum likelihoods for each hypothesis was carried out using the expectation-conditional maximization (ECM) algorithm. Once the maximum likelihoods under each hypothesis were computed, the log ratio of the likelihoods was computed to serve as the test statistic. This log-likelihood ratio test statistic is asymptotic to a {chi}2 distribution with 1 degree of freedom. The C++ software for testing these particular hypotheses is available upon request.

Multi-trait analysis was performed on bone and nonbone traits for QTL reported on chromosomes 2, 3, and 6. Incorporation of the non-bone-related trait(s) led to significant increases in the LOD scores for the loci on chromosomes 2, 3, and 6 (Figs. 3 5). The pleiotropy vs. close linkage test was then applied to each of the specific genome regions defined by the QTL resulting from the multi-trait analyses. This test yielded the following results. For chromosome 2, BMD-R and BMD-Rd were tested against adipose mass and HDL cholesterol to determine whether the QTL on chromosome 2 potentially represented the same locus (pleiotropy) or distinct, closely linked loci. The pleiotropy test could not be rejected at the 0.05 significance level for any of the comparisons. For chromosome 3, the chromosome 3 QTL for BMC-CT was tested against the adjacent chromosome 3 QTL for body length and HDL cholesterol, to determine whether they potentially represented the same QTL or closely linked loci. The test was rejected at the P < 0.05 level, for each comparison, indicating that these loci were distinct. Finally, on chromosome 6, total radiographic intertrochanteric density (BMC-Rit) (at position 68 cM) was compared with plasma HDL (at position 73 cM) and subcutaneous fat pad mass (at position 65 cM), and the pleiotropy test could not be rejected at the 0.05 significance level. It is of note that the pleiotropy test could be rejected at the 0.01 significance level when adipose tissue mass (at position 50 cM) was compared against BMC-Rit (at position 68 cM), indicating there may be two QTL controlling for adipose tissue mass and subcutaneous fat pad mass on chromosome 6.

The specificity and sensitivity of the pleiotropy/close linkage test have not been thoroughly explored. However, the position of the QTLs tested affects the power of this test, since the test is dependent on observing recombination in the testing regions defined by the positions under consideration. Therefore, although insufficient support to reject this test is consistent with pleiotropy, it does not prove conclusively that pleiotropic effects of a single gene are controlling for the multiple traits, since the power to detect multiple QTLs at close distances is greatly reduced and since these small distances may contain a large number of genes.


    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
This study has examined genetic and metabolic determinants of bone mass and density in a set of genetically heterogeneous older female mice with diet-induced atherosclerosis. In these mice, there were significant phenotypic correlations between bone parameters and those related to body morphometry and adiposity, lipid metabolism, and atherosclerosis. By the concomitant use of QTL and other genetic analyses for these multiple traits, we sought to determine whether there might be a genetic basis for these relationships. Six distinct loci were identified with significant linkage for multiple bone-related parameters, and three of these coincided closely with loci showing linkage to nonbone traits. Colocalization of QTLs may simply represent close linkage of two or more genes independently influencing the traits, as there are likely hundreds of genes encompassed in the span of each QTL at the current resolution. However, it may also be indicative of the presence of common genetic determinants (genetic pleiotropy), a possibility strengthened by the presence of strong phenotypic correlations, consistent allele effects, and known biologic relationships between the colocalizing traits. We therefore developed and applied novel statistical tools to improve the ability to distinguish closely mapped but independent loci from those that are potentially pleiotropic.

Adipose tissue and plasma HDL traits were correlated with bone mass and density as well as biomechanical properties, and loci on distal chromosome 2 and distal chromosome 6 showed colocalization of QTLs for bone traits with both of these. For the chromosome 2 and 6 QTLs, incorporating adipose tissue and plasma HDL traits into the multi-trait QTL analyses significantly raised the strength of linkage, which is consistent with pleiotropy. Additionally, the test for pleiotropy vs. close linkage supported potential pleiotropy for the chromosome 2 and the distal chromosome 6 loci. The chromosome 2 locus falls within a region identified in other studies as showing linkage with obesity (32).

Clinical studies have suggested that obesity confers a protective effect against osteoporosis and fractures that is independent of its contribution to body weight (43, 44, 46). Traditional explanations for this include the increased peripheral conversion of estrogen and increased insulin levels associated with adipose tissue, each having positive effects on bone mass and density (11, 45). However, recent studies of the role of leptin have shown that the situation is likely much more complicated. Clinical studies have demonstrated that plasma leptin levels are correlated with bone density (31), as we observed in this study, and in vitro leptin stimulates osteoblastic differentiation in progenitor stromal cells (52). It is intriguing that adipocytes and osteoblasts can be derived from common marrow stromal precursor cells and thus might share common cell biologic features (41). On the other hand, leptin is now well documented to be a central neuroendocrine mediator causing inhibition of bone formation, among other effects (13). Also, administration of a high-fat diet to mice can impair both central and peripheral sensitivity to leptin (28) and cause reduced bone density (40).

Similarly, there is observational and experimental evidence linking plasma lipids with bone density. Atherosclerotic lesions and the concomitant lower plasma HDL and higher plasma LDL cholesterol concentrations were associated with reduced bone density, a relationship observed in clinical studies in women (3, 19, 54). Although the relationship could simply be parallel changes occurring with age, recent in vitro and in vivo studies showing inhibitory effects of oxidized lipids on osteoblast differentiation and on bone density have provided evidence for a causative relationship between these (3840). Our findings of two loci showing linkage with bone, adipose tissue, and plasma lipid traits are intriguing in light of the above and allow us to hypothesize that genetic pleiotropy may be responsible. Further studies will be needed to definitively establish whether this is the case or simply close linkage of distinct genes is responsible for the observation.

Several recent studies have reported results of QTL analyses for peak bone density in mice, including one by Klein and colleagues (23) that utilized recombinant inbred (RI) strains derived from the same parental strains studied here. Before discussing our findings in relation to these, it should first be noted that the bone density traits examined here are not directly comparable. Rather than reflecting peak bone density, they purposely reflect bone density in the more complicated setting of older age and the concurrent administration of a highly atherogenic diet. Both of these factors could be responsible for some or all of the QTL identified. However, since the traits reflect the cumulative genetic effects over the life of the mice, the QTL may also reflect those affecting peak bone density. In light of the latter, it is of interest that the distal chromosome 2 locus was linked with bone density in both our study and that of Klein et al. (23), suggesting that it is indeed an important locus. Although the same genetic variations should be present in the RI strains and in the F2 intercross mice (since the parental strains are similar), other loci were not found in common, which may be explained by differing study conditions as discussed above and, potentially, the different techniques for measuring bone density. Also, in general, studies with RI strains also have less statistical power than those using intercrosses and are more susceptible to false-positive results. Loci for peak bone density in an intercross between strains C3H/HeJ and C57BL/6J, C57BL/6J and CAST/Ei, and in the osteoporotic SAM-P6 mouse have also been reported recently and do not correspond to loci observed in our study (5, 7, 51). In these cases, strain differences would be expected to yield different loci, in addition to the age and diet effects.

In humans, variations in the genotype for the vitamin D receptor, estrogen receptor, collagen type 1a, and interleukin 6 have been associated with differences in bone density, as has a QTL on chromosome 11, but these are not mapped to any of the loci identified in this study (15, 20, 24, 34, 53). It is important to note that, in studies such as this, lack of linkage is simply noninformative, as there may be no biologically relevant genetic variation between the two strains used. Also, as noted above, the QTL we have detected may be related to age or diet effects rather than peak bone mass. The next steps in the identification of the genes responsible at each of the loci identified here will require finer mapping through the use of strains congenic for each of the loci. Continued identification and mapping of new genes through the various genome projects will also expand the pool of candidate genes to consider.

This study demonstrates the value of combining phenotypic and genetic analysis for elucidating the pathogenic interactions among multiple related traits. Over the past 5 years, a variety of statistical techniques have been developed to resolve coincident QTL, to consider multiple traits simultaneously, and to test whether coincident QTL are closely linked or represent pleiotropic effects from a single QTL. As the number of traits being simultaneously analyzed increases, the power to detect marker/trait associations can potentially dramatically increase if appropriate statistical models are used. The results obtained in this study demonstrate this dramatic increase in power, and the pleiotropy test developed by Jiang and Zeng (18) further demonstrates that an application of more sophisticated statistical tests can go far in elucidating the underlying genetic factors contributing to multiple quantitative traits. We were able to increase the evidence for linkage, given by the LOD score, for six QTL by more than one order of magnitude, on average, and by more than three orders of magnitude in two cases. In addition, we were able to reject the pleiotropy hypothesis for the chromosome 3 loci, but were unable to reject this hypothesis for the chromosome 2 and the distal chromosome 6 QTL, by implementing and then applying the pleiotropy vs. close linkage statistical test. As more phenotype data obtain from a variety of sources (including microarray-measured transcript levels in addition to the more classic traits measured here) there will be increased demand for more sophisticated statistical methods capable of deriving more information from the data, such as those we have employed here. In our own study, we have tried to maximize the available information in our data by making use of the most advanced statistical techniques available to date.


    ACKNOWLEDGMENTS
 
The technical assistance of Larry Castellani, Sarada Charungundla, Mark Cuevas, Lena Dishakjian, Marjon Jahromi, Yu-po Lee, Jian-Hua Qiao, Kathy Salcedo, and Xuping Wang is gratefully acknowledged.

This work was supported in part by National Heart, Lung, and Blood Institute Grants HL-30568 (to A. J. Lusis and T. A. Drake) and HL-28481 (to A. J. Lusis) and by the Laubisch Fund, UCLA. K. Krass was supported by National Research Service Award GM-07014.


    FOOTNOTES
 
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).

Address for reprint requests and other correspondence: T. A. Drake, Dept. of Pathology and Laboratory Medicine, UCLA, Los Angeles, CA 90095-1732 (E-mail: tdrake{at}mednet.ucla.edu).


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

  1. Adami S, Zamberlan N, Gatti D, Zanfisi C, Braga V, Broggini M, and Rossini M. Computed radiographic absorptiometry and morphometry in the assessment of postmenopausal bone loss. Osteoporos Int 6: 8–13, 1996.[ISI][Medline]
  2. Akhter MP, Cullen DM, Pedersen EA, Kimmel DB, and Recker RR. Bone response to in vivo mechanical loading in two breeds of mice. Calcif Tissue Int 63: 442–449, 1998.[ISI][Medline]
  3. Barengolts EI, Berman M, Kukreja SC, Kouznetsova T, Lin C, and Chomka EV. Osteoporosis and coronary atherosclerosis in asymptomatic postmenopausal women. Calcif Tissue Int 62: 209–213, 1998.[ISI][Medline]
  4. Basten CA, Weir BS, and Zeng ZB. QTL Cartographer, A Reference Manual and Tutorial for QTL Mapping. Raleigh, NC: Department of Statistics, North Carolina State University, 1999.
  5. Beamer WG, Rosen CJ, Donahue LR, Frankel WN, Churchill GA, Shulz KL, Baylink DJ, and Pettis JL. Location of genes regulating volumetric bone mineral density in C57BL/6J (low) and C3H/HeJ (high) inbred strains of mice (Abstract). Bone Suppl 23: S162, 1998.
  6. Beamer WG, Donahue LR, Rosen CJ, and Baylink DJ. Genetic variability in adult bone density among inbred strains of mice. Bone 18: 397–403, 1996.[ISI][Medline]
  7. Beamer WG, Shultz KL, Churchill GA, Frankel WN, Baylink DJ, Rosen CJ, and Donahue LR. Quantitative trait loci for bone density in C57BL/6J and CAST/EiJ inbred mice. Mamm Genome 10: 1043–1049, 1999.[ISI][Medline]
  8. Burstein AH and Frankel VH. A standard test for laboratory animal bone. J Biomech 4: 155–158, 1971.[ISI][Medline]
  9. Cauley JA, Salamone LM, and Lucas FL. Postmenopausal endogenous and exogenous hormones, degree of obesity, thiazide diuretics, and risk of osteoporosis. In: Osteoporosis, edited by Marcus R, Feldman D, and Kelsey J. San Diego: Academic, 1996, p. 551–576.
  10. Comuzzie AG and Allison DB. The search for human obesity genes. Science 280: 1374–1377, 1998.[Abstract/Free Full Text]
  11. Cornish J, Callon KE, and Reid IR. Insulin increases histomorphometric indices of bone formation in vivo. Calcif Tissue Int 59: 492–495, 1996.[ISI][Medline]
  12. Dimai HP, Linkhart TA, Linkhart SG, Donahue LR, Beamer WG, Rosen CJ, Farley JR, and Baylink DJ. Alkaline phosphatase levels and osteoprogenitor cell numbers suggest bone formation may contribute to peak bone density differences between two inbred strains of mice. Bone 22: 211–216, 1998.[ISI][Medline]
  13. Ducy P, Amling M, Takeda S, Priemel M, Schilling AF, Beil FT, Shen J, Vinson C, Rueger JM, and Karsenty G. Leptin inhibits bone formation through a hypothalamic relay: a central control of bone mass. Cell 100: 197–207, 2000.[ISI][Medline]
  14. Fukunaga M, Tomomitsu T, Otsuka N, Imai H, Morita R, and Nishii Y. Indexes of bone mineral content on second metacarpal bone roentgenogram analyzed by digital image processing: a comparison with other bone mass quantifying methods. Radiat Med 8: 230–235, 1990.[Medline]
  15. Grant SF, Reid DM, Blake G, Herd R, Fogelman I, and Ralston SH. Reduced bone density and osteoporosis associated with a polymorphic Sp1 binding site in the collagen type I alpha 1 gene. Nat Genet 14: 203–205, 1996.[ISI][Medline]
  16. Hayashi Y, Yamamoto K, Fukunaga M, Ishibashi T, Takahashi K, and Nishii Y. Assessment of bone mass by image analysis of metacarpal bone roentgenograms: a quantitative digital image processing (DIP) method. Radiat Med 8: 173–178, 1990.[Medline]
  17. Ivandic BT, Qiao JH, Machleder D, Liao F, Drake TA, and Lusis AJ. A locus on chromosome 7 determines myocardial cell necrosis and calcification (dystrophic cardiac calcinosis) in mice. Proc Natl Acad Sci USA 93: 5483–5488, 1996.[Abstract/Free Full Text]
  18. Jiang C and Zeng ZB. Multiple trait analysis of genetic mapping for quantitative trait loci. Genetics 140: 1111–1127, 1995.[Abstract/Free Full Text]
  19. Jie KG, Bots ML, Vermeer C, Witteman JC, and Grobbee DE. Vitamin K status and bone mass in women with and without aortic atherosclerosis: a population-based study. Calcif Tissue Int 59: 352–356, 1996.[ISI][Medline]
  20. Johnson ML, Gong G, Kimberling W, Recker SM, Kimmel DB, and Recker RB. Linkage of a gene causing high bone mass to human chromosome 11 (11q12–13). Am J Hum Genet 60: 1326–1332, 1997.[ISI][Medline]
  21. Kaufman L and Rousseeuw PJ. Finding Groups in Data: An Introduction to Cluster Analysis. New York: Wiley, 1990.
  22. Kaye M and Kusy RP. Genetic lineage, bone mass, and physical activity in mice. Bone 17: 131–135, 1995.[ISI][Medline]
  23. Klein RF, Mitchell SR, Phillips TJ, Belknap JK, and Orwoll ES. Quantitative trait loci affecting peak bone mineral density in mice. J Bone Miner Res 13: 1648–1656, 1998.[ISI][Medline]
  24. Kobayashi S, Inoue S, Hosoi T, Ouchi Y, Shiraki M, and Orimo H. Association of bone mineral density with polymorphism of the estrogen receptor gene. J Bone Miner Res 11: 306–311, 1996.[ISI][Medline]
  25. Lander E and Kruglyak L. Genetic dissection of complex traits: guidelines for interpreting and reporting linkage results. Nat Genet 11: 241–247, 1995.[ISI][Medline]
  26. Lander ES and Schork NJ. Genetic dissection of complex traits. Science 265: 2037–2048, 1994.[ISI][Medline]
  27. Lembertas AV, Fisher JS, Warden CH, Wen PZ, Xia YR, and Lusis AJ. A locus on the X chromosome is linked to body length in mice. Mamm Genome 7: 171–173, 1996.[ISI][Medline]
  28. Lin S, Thomas TC, Storlien LH, and Huang XF. Development of high fat diet-induced obesity and leptin resistance in C57Bl/6J mice. Int J Obes Relat Metab Disord 24: 639–646, 2000.[Medline]
  29. Lincoln SE, Daly MJ, and Lander ES. Constructing Genetic Linkage Maps with MAPMAKER/EXP Version 3.0: A Tutorial and Reference Manual. Cambridge, MA: Whitehead Institute for Biomedical Research (technical report).
  30. Lynch M and Walsh JB. Genetics and Analysis of Quantitative Traits. Sunderland, MA: Sinauer Associates, 1998.
  31. Matkovic V, Ilich JZ, Skugor M, Badenhop NE, Goel P, Clairmont A, Klisovic D, Nahhas RW, and Landoll JD. Leptin is inversely related to age at menarche in human females. J Clin Endocrinol Metab 82: 3239–3245, 1997.[Abstract/Free Full Text]
  32. Mehrabian M, Wen PZ, Fisler J, Davis RC, and Lusis AJ. Genetic loci controlling body fat, lipoprotein metabolism, and insulin levels in a multifactorial mouse model. J Clin Invest 101: 2485–2496, 1998.[Abstract/Free Full Text]
  33. Mikic B, Van der Meulen MC, Kingsley DM, and Carter DR. Mechanical and geometric changes in the growing femora of BMP-5 deficient mice. Bone 18: 601–607, 1996.[ISI][Medline]
  34. Morrison NA, Qi JC, Tokita A, Kelly PJ, Crofts L, Nguyen TV, Sambrook PN, and Eisman JA. Prediction of bone density from vitamin D receptor alleles. Nature 367: 284–287, 1994.[ISI][Medline]
  35. Murray EJ, Song MK, Laird EC, and Murray SS. Strain-dependent differences in vertebral bone mass, serum osteocalcin, and calcitonin in calcium-replete and -deficient mice. Proc Soc Exp Biol Med 203: 64–73, 1993.[Abstract]
  36. Nishina PM, Naggert JK, Verstuyft J, and Paigen B. Atherosclerosis in genetically obese mice: the mutants obese, diabetes, fat, tubby, and lethal yellow. Metabolism 43: 554–558, 1994.[ISI][Medline]
  37. Paigen B, Morrow A, Brandon C, Mitchell D, and Holmes P. Variation in susceptibility to atherosclerosis among inbred strains of mice. Atherosclerosis 57: 65–73, 1985.[ISI][Medline]
  38. Parhami F, Jackson SM, Tintut Y, Le V, Balucan JP, Territo M, and Demer LL. Atherogenic diet and minimally oxidized low density lipoprotein inhibit osteogenic and promote adipogenic differentiation of marrow stromal cells. J Bone Miner Res 14: 2067–2078, 1999.[ISI][Medline]
  39. Parhami F, Morrow AD, Balucan J, Leitinger N, Watson AD, Tintut Y, Berliner JA, and Demer LL. Lipid oxidation products have opposite effects on calcifying vascular cell and bone cell differentiation. A possible explanation for the paradox of arterial calcification in osteoporotic patients. Arterioscler Thromb Vasc Biol 17: 680–687, 1997.[Abstract/Free Full Text]
  40. Parhami F, Tintut Y, Beamer WG, Gharavi N, Goodman W, and Demer LL. Atherogenic high fat diet reduces bone mineralization in mice. J Bone Mineral Res. 16: 182–188, 2001.[ISI][Medline]
  41. Prockop DJ. Marrow stromal cells as stem cells for nonhematopoietic tissues. Science 276: 71–74, 1997.[Abstract/Free Full Text]
  42. Qiao JH, Xie PZ, Fishbein MC, Kreuzer J, Drake TA, Demer LL, and Lusis AJ. Pathology of atheromatous lesions in inbred and genetically engineered mice. Genetic determination of arterial calcification. Arterioscler Thromb 14: 1480–1497, 1994.[Abstract]
  43. Reid IR, Ames R, Evans MC, Sharpe S, Gamble G, France JT, Lim TM, and Cundy TF. Determinants of total body and regional bone mineral density in normal postmenopausal women: a key role for fat mass. J Clin Endocrinol Metab 75: 45–51, 1992.[Abstract]
  44. Reid IR, Evans MC, and Ames RW. Volumetric bone density of the lumbar spine is related to fat mass but not lean mass in normal postmenopausal women. Osteoporos Int 4: 362–367, 1994.[ISI][Medline]
  45. Reid IR, Evans MC, Cooper GJ, Ames RW, and Stapleton J. Circulating insulin levels are related to bone density in normal postmenopausal women. Am J Physiol Endocrinol Metab 265: E655–E659, 1993.[Abstract/Free Full Text]
  46. Reid IR, Plank LD, and Evans MC. Fat mass is an important determinant of whole body bone density in premenopausal women but not in men. J Clin Endocrinol Metab 75: 779–782, 1992.[Abstract]
  47. Rosen CJ, Churchill GA, Donahue LR, Shultz KL, Burgess JK, Powell DR, and Beamer WG. Mapping quantitative trait loci for serum insulin-like growth factor-1 levels in mice. Bone 27: 521–528, 2000.[ISI][Medline]
  48. Sambrook PN, Kelly PJ, White CP, Morrison NA, and Eisman JA. Genetic determinants of bone mass. In: Osteoporosis, edited by Marcus R, Feldman D, and Kelsey J. San Diego: Academic Press, 1996, p. 477–482.
  49. Sanada T, Tomomitsu T, Otsuka N, Fukunaga M, and Togawa H. A fundamental study of a computed X-ray densitometry system for bone mineral measurement in the second metacarpal bone. Radiat Med 12: 143–146, 1994.[Medline]
  50. Seeman E and Hopper JL. Genetic and environmental components of the population variance in bone density. Osteoporos Int Suppl 7: S10–S16, 1997.
  51. Shimizu M, Higuchi K, Bennett B, Xia C, Tsuboyama T, Kasai S, Chiba T, Fujisawa H, Kogishi K, Kitado H, Kimoto M, Takeda N, Matsushita M, Okumura H, Serikawa T, Nakamura T, Johnson TE, and Hosokawa M. Identification of peak bone mass QTL in a spontaneously osteoporotic mouse strain. Mamm Genome 10: 81–87, 1999.[ISI][Medline]
  52. Thomas T, Gori F, Khosla S, Jensen MD, Burguera B, and Riggs BL. Leptin acts on human marrow stromal cells to enhance differentiation to osteoblasts and to inhibit differentiation to adipocytes. Endocrinology 140: 1630–1638, 1999.[Abstract/Free Full Text]
  53. Uitterlinden AG, Burger H, Huang Q, Yue F, McGuigan FE, Grant SF, Hofman A, van Leeuwen JP, Pols HA, and Ralston SH. Relation of alleles of the collagen type I{alpha}1 gene to bone density and the risk of osteoporotic fractures in postmenopausal women. N Engl J Med 338: 1016–1021, 1998.[Abstract/Free Full Text]
  54. Uyama O, Yoshimoto Y, Yamamoto Y, and Kawai A. Bone changes and carotid atherosclerosis in postmenopausal women. Stroke 28: 1730–1732, 1997.[Abstract/Free Full Text]
  55. Weiss A, Arbell I, Steinhagen-Thiessen E, and Silbermann M. Structural changes in aging bone: osteopenia in the proximal femurs of female mice. Bone 12: 165–172, 1991.[ISI][Medline]