Chromosome 2q32 May Harbor a QTL Affecting BMD Variation at Different Skeletal Sites

Chromosome 2q32 May Harbor a QTL Affecting BMD Variation atDifferent Skeletal Sites

Liang Wang,1,2 Yong-Jun Liu,2 Peng Xiao,3 Hui Shen,2 Hong-Yi Deng,2 Christopher J Papasian,2 Betty M Drees,2

James J Hamilton,2 Robert R Recker,3 and Hong-Wen Deng1,2,4

ABSTRACT: BMDs at different skeletal sites share some common genetic determinants. Using PCA andbivariate linkage analysis, we identified a QTL on chromosome 2q32 with significant pleiotropic effects onBMDs at different skeletal sites.

Introduction: BMDs at the hip, spine, and forearm are genetically correlated, suggesting the existence ofquantitative trait loci (QTLs) with concurrent effects on BMDs at these three skeletal sites. Consequently, itis important to identify these QTLs in the human genome and, for those implicated QTLs, it is important todifferentiate between pleiotropic effects, caused by a single gene that concurrently effects these traits, andco-incident linkage, caused by multiple, closely linked, genes that independently effect these traits.Materials and Methods: For a sample of 451 American white pedigrees made up of 4498 individuals, weevaluated the correlations between BMDs at the three skeletal sites. We carried out principal componentanalysis (PCA) for the three correlated traits and obtained a major component, PC1, which accounts for >75%of the co-variation of BMDs at the three sites. We subsequently conducted a whole genome linkage scan forPC1 and performed bivariate linkage analysis for pairs of the three traits (i.e., forearm/spine BMD, hip/forearm BMD, and hip/spine BMD).Results: Chromosome region 2q32, near the marker GATA65C03M, showed strong linkage to PC1 (LOD �3.35). Subsequent bivariate linkage analysis substantiated linkage at 2q32 for each trait pair (LOD scores were2.65, 2.42, and 2.13 for forearm/spine BMD, hip/forearm BMD, and hip/spine BMD, respectively). Furtheranalyses rejected the hypothesis of co-incident linkage (p0[forearm/spine] � 0.0005, p0[hip/forearm] � 0.004,p0(hip/spine] � 0.001) but failed to reject the hypothesis of pleiotropy (p1[forearm/spine] � 0.35, p1[hip/forearm] � 0.07, p1[hip/spine] � 0.15).Conclusions: Our results strongly support the conclusion that chromosome region 2q32 may harbor a QTLwith pleiotropic effects on BMDs at different skeletal sites.J Bone Miner Res 2007;22:1672–1678. Published online on August 6, 2007; doi: 10.1359/JBMR.070722

Key words: BMD, pleiotropy, quantitative trait loci, principal component analysis, linkage

INTRODUCTION

OSTEOPOROSIS, MAINLY CHARACTERIZED by low bonemass and microstructure deterioration of bone, is be-

coming a severe problem in the world.(1) Osteoporosis mayprogress painlessly, for years, until a fracture occurs at thehip, spine, forearm, or other skeletal sites. The most pow-erful, measurable determinant of fracture risk is bone mass,measured as BMD.(2–4) Several studies support the conceptthat as much as 60∼90% of BMD variation can be explainedby genetic factors.(5–8)

BMDs at different skeletal sites are highly correlated

with one another. In a study of Australian white twins,Nguyen et al.(9) found that the genetic and environmentalcorrelations between BMD at the spine and femoral neckwere 0.64 and 0.57, respectively. Our earlier study in a U.S.white population(10) found strong genetic correlations(0.58–0.77) between BMDs at the hip, spine, and ultradistalradius (UD) in both women and men. Livshits et al.(11) alsoreported significant correlations (0.50–0.79) betweenBMDs at the hip and spine in three independent popula-tions.

BMD variation at different skeletal sites may be gov-erned by shared genetic factors.(12,13) By performing bivari-ate linkage analyses in a white population, Devoto et al.(14)

found that susceptibility loci may exert pleiotropic effectsThe authors state that they have no conflicts of interest.

1The Key Laboratory of Biomedical Information Engineering of Ministry of Education, and Institute of Molecular Genetics, School ofLife Science and Technology, Xi’an Jiaotong University, Xi’an, Shaanxi, China; 2School of Medicine, University of Missouri-Kansas City,Kansas City, Missouri, USA; 3Department of Biomedical Sciences and Osteoporosis Research Center, School of Medicine, CreightonUniversity, Omaha, Nebraska, USA; 4Laboratory of Molecular and Statistical Genetics, College of Life Sciences Hunan Normal Uni-versity, Changsha, Hunan, China.

JOURNAL OF BONE AND MINERAL RESEARCHVolume 22, Number 11, 2007Published online on August 6, 2007; doi: 10.1359/JBMR.070722© 2007 American Society for Bone and Mineral Research

1672

JO703200 1672 1678 November

(i.e., concurrent effects on multiple phenotypic traits) onBMDs at the femoral neck, lumbar spine, and trochanter.Surprisingly, earlier genome-wide linkage scans revealedlargely site-specific chromosomal regions, suggesting thatgenetic pleiotropy may be modest for BMDs at differentskeletal sites.(15–18) This was somewhat unexpected consid-ering the strong genetic correlations among these traits. Itshould be noted, however, that most of the earlier linkagescans for BMDs were performed separately for individualskeletal sites. Consequently, it is reasonable to hypothesizethat, to identify genetic factors that have concurrent con-tributions to a clustering of bone traits, it may be necessaryto use alternative measures that can capture the commonfeatures of these intercorrelated traits.

In this study, using principal component analysis (PCA)and bivariate linkage analyses for a large sample of whitepedigrees, we aimed (1) to identify linkage regions havingconcurrent effects on BMDs at the three skeletal sites (hip,spine, and forearm) and (2) to differentiate pleiotropic ef-fects (i.e., caused by a single gene that concurrently effectsthese three phenotypic traits) from co-incident linkage (i.e.,caused by multiple, closely linked, genes that independentlyeffect these three phenotypic traits) for those linkage re-gions identified in (1) above.

MATERIALS AND METHODS

Subjects

The study was approved by the Institutional ReviewBoard of both University of Missouri-Kansas City andCreighton University. All study subjects, recruited from thevicinity of Creighton University in Omaha, NE, were whitesof European origin, and signed informed-consent docu-ments before entering the project. The sample used in thisstudy contained 4498 subjects from 451 pedigrees. Amongthem, 4126 were genotyped. The 451 families varied in sizefrom 4 to 416 individuals, with a mean of 11.6 ± 28.5 (SD).The distribution of pedigree sizes is given in Table 1. Theexclusion criteria have been described elsewhere.(19) Ingeneral, only healthy people were included in the analysis;patients with chronic diseases and conditions that mightpotentially affect bone mass, structure, or metabolism wereexcluded.

Measurements

BMDs (g/cm2) at the total hip (femoral neck, trochanter,and intertrochanteric region), lumbar spine (L1–L4), andforearm (UD, mid-distal and one-third distal) were mea-

sured by Hologic 1000, 2000+, or 4500 DXA scanners (Ho-logic, Bedford, MA, USA). All scanners were calibrateddaily, and long-term precision was monitored with externalphantoms. Data obtained from different machines weretransformed to a compatible measurement.(20) Members ofthe same pedigree were usually measured on the same typeof machine. The CVs of the DXA measurements for BMDat the hip, spine, and forearm were 1.4%, 0.9%, and 2.3%,respectively. Weight (kg) and height (m) were measuredusing standard methods at the same visit.

Genotyping

Genomic DNA was extracted from peripheral blood us-ing the Puregene DNA isolation kit (Gentra Systems, Min-neapolis, MN, USA). All subjects were genotyped with 410microsatellite markers (including 392 markers for 22 auto-somes and 18 markers for the X chromosome) by theMarshfield Center for Medical Genetics (Marshfield, WI,USA) using Marshfield screening set 14. These markershave an average population heterozygosity of 0.75 and arespaced 8.9 cM apart on average. The detailed genotypingprotocol is available at http://research.marshfieldclinic.org/genetics/Lab_Methods/methods.html. Marker data for eachpedigree were checked for Mendelian inheritance using thePedCheck program (http://watson.hgen.pitt.edu/register/soft_doc.html). The overall genotyping error rate was∼0.3% in our sample.

Statistical analyses

The narrow sense heritability (h2) estimates of hip, spine,and forearm BMD were calculated using the variance com-ponents methods implemented in Sequential OligogenicLinkage Analysis Routines (SOLAR),(21) which takes intoaccount all familial relations combined and partitions thetotal phenotypic variation of these traits into additive ge-netic and individual environmental components. The heri-tability estimates were calculated as the ratio of the additivegenetic variance to the total phenotypic variance. The samemethod was used to estimate the heritability of PC1. Ge-netic and environmental correlations (�G and �E, respec-tively) between BMDs at the hip, spine, and forearm werecarried out using the bivariate variance decompositionanalyses implemented in SOLAR. Age, age2, sex, height,and weight were used as covariates for the estimation of theabove parameters to increase the genetic signal-to-noise ra-tio by decreasing the proportion of the residual phenotypicvariation. Phenotypic correlation (�P) can be calculated bythe following equation:

�P = ��h12 � �h2

2 � �G�

+ ���1 − h12� � ��1 − h2

2� � �E�

where h12 and h2

2 are heritabilities of each two of the threestudied traits. The significance of �G and �E between anypair of traits was tested using the likelihood ratio statistic.

We performed PCA for BMDs at the hip, spine, andforearm in the 4498 subjects. The analysis was conductedusing the statistical package SPSS (v. 14.0; SPSS, Chicago,IL, USA). PCA is a data reduction method that transforms

TABLE 1. DISTRIBUTION OF PEDIGREE SIZE

Pedigree size(subjects) Number of families Number of subjects

<10 387 183210–19 27 33020–29 12 30730–39 8 270>39 17 1759

Total 451 4498

A WHOLE GENOME SCAN FOR BMD 1673

a number of correlated quantitative variables into feweruncorrelated variables termed principal components (PCs).Each PC is a linear combination of the original variables,with coefficients equal to the eigenvectors of the correlationor covariance matrix. The PC incorporates an element fromeach skeletal site in proportion to the unique informationthe site contains. The obtained first principal component(PC1) was used as a new trait in the following linkage analy-sis. The PC1 loading was extracted for each individual, andthe resulting scores were used in the linkage analysis.

Variance component linkage analyses were performedfor PC1 using SOLAR. Multipoint LOD scores were cal-culated for chromosomes 1 through 22, and two-point LODscores were computed for chromosome X. Age, age2, sex,height, and weight were tested for importance using thepolygenic screen model in SOLAR, and significant factors(p � 0.05) were adjusted as covariates for raw BMD values.Empirical LOD scores for each trait were computed by theprocedure “lodadj” implemented in SOLAR. This proce-dure sampled the null distribution of LOD scores (the dis-tribution obtained under the hypothesis of “no linkage”),which was achieved by permutation of 10,000 replicates,with a fully informative marker unlinked to each trait simu-lated in each replicate.

We further performed bivariate linkage analysis on theidentified linkage regions for PC1. Bivariate linkage analy-sis can be used to determine whether a single locus hasconcurrent effects on the three traits (i.e., pleiotropic ef-fects) or whether each linkage region contains multiple locithat independently influence the three traits (i.e., co-incident linkage). Specifically, bivariate linkage analysiswas performed for the three trait pairs (i.e., forearm/spineBMD, hip/forearm BMD, and hip/spine BMD) using a vari-ance decomposition approach implemented in SOLAR.We tested the null hypothesis of no linkage (�2

q � 0, here�2

q is the additive genetic variance caused by a major locus)by comparing the log likelihood of this restricted modelwith that of a model in which �2

q was estimated for the traitpairs. In linkage analysis, the bivariate model contains 2degrees of freedom (df), whereas the univariate model hasonly 1 df. For the LOD scores to be comparable betweenbivariate and univariate analyses, we associated 2 df bivari-ate LOD score (denoted LOD2) and 1 df effective LODscore (denoted LOD1) with an appropriate p value.(22) Be-cause the bivariate test statistic (2 × Ln10 × LOD2) followsa mixture distribution of 1⁄4�0

2, 1⁄2�12,and 1⁄4�3

2,(23) theLOD2 can be converted to a 1⁄2�1

2 of equivalent p value,which is divided by 2 × ln10 to get the 1 df effective LODscore (LOD1; see handbook for SOLAR).

To test pleiotropy or co-incident linkage, the likelihoodof the linkage model in which �q was estimated was com-pared with the likelihood of the linkage model in which �q

was constrained to 1 (complete pleiotropy) or 0 (completeco-incident linkage). Here, �q is a measure of the sharedmajor genetic effect near the genomic region for which link-age is being assessed. The possibility of co-incident linkageand complete pleiotropy is denoted by p0 and p1 in theresults, respectively. This comparison was made at the lo-cation of the maximum LOD score in the bivariate analy-ses.

RESULTS

Table 2 presents the basic characteristics of the studypopulation. Age, age2, sex, height, and weight were testedfor importance on BMDs at the hip, spine, and forearm,and significant factors (p < 0.05) were adjusted as covariatesfor raw BMD values. The heritability of hip BMD, spineBMD, and forearm BMD was 0.65 ± 0.03 (p < 0.0001),0.62 ± 0.03 (p < 0.0001), and 0.45 ± 0.03 (p < 0.0001), re-spectively (Table 3).

The results of the correlation analyses are summarized inTable 4. The pairwise correlations (including �G, �E, and�P) between BMDs at the spine, hip, and forearm werepositively significant, both genetically and environmentally.The genetic correlations were generally higher than envi-ronmental correlations for each pair of traits. The pairwise�G was 0.76 ± 0.02 for hip BMD × spine BMD, 0.59 ± 0.03for spine BMD × forearm BMD, and 0.70 ± 0.03 for hipBMD × forearm BMD.

PCA generated one major component (PC1), which ac-counted for >75% of the total co-variation of the threeBMD traits. Table 5 shows the loadings of PC1 for BMD at

TABLE 3. NARROW SENSE HERITABILITY OF HIP BMD, SPINE

BMD, FOREARM BMD, AND PC1

Traits* h2 ± SE

Hip BMD 0.65 ± 0.03Spine BMD 0.62 ± 0.03Forearm BMD 0.45 ± 0.03PC1 0.62 ± 0.03

* Adjusted for age, age2, sex, height, and weight.

TABLE 4. CORRELATIONS AMONG THE THREE TRAITS

Hip BMD ×spine BMD

Spine BMD ×forearm BMD

Hip BMD ×forearm BMD

�G 0.76 ± 0.02(<0.0001)

0.59 ± 0.03(<0.0001)

0.70 ± 0.03(<0.0001)

�E 0.52 ± 0.03(<0.0001)

0.35 ± 0.03(<0.0001)

0.38 ± 0.03(<0.0001)

�P 0.62 0.41 0.47

p value is given in parentheses.�G, genetic correlation; �E, environmental correlation; �P, phenotypic

correlation.

TABLE 2. BASIC CHARACTERISTICS OF THE STUDY SUBJECTS

Total(n = 4498)

Female(n = 2682)

Male(n = 1816)

Height (m) 1.69 ± 0.10 1.64 ± 0.07 1.72 ± 0.07Weight (kg) 78.50 ± 18.21 71.33 ± 16.01 89.40 ± 15.79Age (yr) 47.7 ± 16.0 47.6 ± 16.0 48.0 ± 16.1BMD (g/cm2)

Hip 0.97 ± 0.16 0.92 ± 0.15 1.04 ± 0.15Spine 1.04 ± 0.16 1.01 ± 0.16 1.07 ± 0.15Forearm 0.61 ± 0.08 0.57 ± 0.07 0.67 ± 0.07

Values are mean ± SD.The reported BMDs are raw phenotype data and normally distributed.

WANG ET AL.1674

the hip (0.92), spine (0.84), and forearm (0.84). Heritabilityfor PC1 was quite high, with narrow sense h2 estimated tobe 0.62 ± 0.03. Table 6 shows the chromosomal regions andthe nearest marker achieving LOD scores �1.9 for PC1.For comparison, we also summarized the LOD scores atthese regions obtained in our earlier univariate linkagestudy using the same sample population.(17) The most sig-nificant linkage (LOD � 3.35) was observed for PC1 onchromosome 2q32 near marker GATA65C03M, where sug-gestive linkage was also found for the three BMD traits inour earlier univariate linkage analysis.(17) Those site-specific linkage regions identified in the study of Xiao etal.(17) did not show significant linkage evidence for PC1 inthis study. Figure 1 presents a plot depicting the multipointLOD scores for PC1 on chromosome 2.

The LOD scores obtained in bivariate linkage analyses(which were converted from 2 df to 1-df LOD scores) of2.65, 2.42, and 2.13 were achieved on 2q32 for trait pairs offorearm/spine BMD, hip/forearm BMD, and hip/spineBMD, respectively (Table 7). Figure 2 shows the bivariatelinkage results for the three trait pairs on chromosome 2.Using a p value of 0.05 as a cut-off point, which is widelyaccepted in the field,(24) we rejected the hypothesis of co-incident linkage (p0[forearm/spine] � 0.0005, p0[hip/forearm] � 0.004, p0[hip/spine] � 0.001) but failed to rejectthe hypothesis of pleiotropy (p1[forearm/spine] � 0.35,p1[hip/forearm] � 0.07, p1[hip/spine] � 0.15; Table 7).

DISCUSSION

It is likely that BMD is controlled by many genes withmodest effects.(25) Consequently, it is important to increase

the power of linkage studies to detect QTLs underlyingBMD variation. Joint analyses of multiple correlated traitshas been repeatedly shown to be preferable and more pow-erful than single-trait analysis.(26,27) This is due primarily tothe capacity of joint multivariate analyses to simultaneouslycapture variation and co-variation of correlated traits. Inaddition, joint analyses of correlated traits may decreasethe measurement error of several related variables. In thisstudy, we attempted to identify QTLs important for a set ofcorrelated BMDs measured at different skeletal sites, whichmay minimize multiple testing of correlated traits. The ex-tracted new trait may help identify a common rather thanspecific genetic component underlying the risk of osteopo-rosis. In the PCA, we obtained a principal component fac-tor, PC1, which explained >75% of the total BMD variationof the hip, spine, and forearm. This factor is a weightedaverage of the contributing BMDs at different skeletal sitesand represents common phenotypic variation of bone masstraits that encompasses shared genetic variation and envi-ronmental variation. Thus, biologically, PC1 can be used asa multivariate phenotype that may, in a different way fromsite-specific BMD, provide evidence of common genetic ef-fects on BMDs at different skeletal sites. Significant linkageto PC1 was identified on chromosome region 2q32. Theseresults are consistent with our previous univariate linkageanalyses,(17) but higher LOD scores were achieved, indicat-ing that PC1 may provide more information for linkageanalysis than independent study of each individual traitstudied. Our results suggest that chromosome region 2q32may have concurrent effects on BMD at the hip, spine, andforearm.

To further determine whether these concurrent effectsare caused by pleiotropic effects or co-incident linkage, weperformed bivariate linkage analysis for pairs of the threeBMD traits on chromosome 2. Our bivariate linkage analy-sis identified suggestive linkages on chromosome 2q32, with

FIG. 1. Results of multipoint linkage analysis for PC1, and hipBMD, spine BMD, and UD BMD (summarized from our previ-ous study(17)) on chromosome 2.

TABLE 5. VARIANCE OF BMDS EXPLAINED BY PC1

Traits PC1

Hip BMD 0.92Spine BMD 0.84Forearm BMD 0.84Eigenvalues 2.26Total variance explained 75.17%

The correlation coefficients between the PC1 and the variables of BMDsat the hip, spine, and forearm are shown as principal component loadings.

TABLE 6. RESULTS OF MULTIPOINT GENETIC LINKAGE ANALYSES

FOR PC1 AND BMDS AT THREE SKELETAL SITES SUMMARIZED

FROM OUR EARLIER STUDY

Location(cM)* Marker PC1

HipBMD†

SpineBMD†

UDBMD†

2q32 (187) GATA65C03M 3.35 2.11 1.76 2.233p25 (37) GGAA4B09N 2.093p21 (78) ATA10H11 2.293q27 (200) TTTA040 2.555q23 (128) GATA62A04 3.397p15 (42) GGAA3F06 2.15

LOD > 3.3 (significant linkage) is shown in bold.* Genetic distance from pter.† LOD scores summarized from our earlier univariate linkage study in

the same sample.(17)

A WHOLE GENOME SCAN FOR BMD 1675

multipoint LOD scores of 2.65, 2.42, and 2.13 for pairs offorearm/spine BMD, hip/forearm BMD, and hip/spineBMD, respectively. In subsequent analysis, the hypothesisof co-incident linkage was rejected but we failed to rejectthe hypothesis of pleiotropy. The bivariate linkage resultsprovided further evidence to support the conclusion that aQTL with a pleiotropic effect on BMDs at different skeletalsites exists in chromosome region 2q32.

Biologically, different skeletal sites (e.g., hip, spine, andforearm) have different proportions of cortical and trabecu-lar bones. The variation in behavior of bone at differentskeletal sites may be partially caused by differing surround-ing environments for bone cells in cortical versus cancellousbone.(28) For example, it is likely that osteoblasts and os-teoclasts in cortical bone are controlled primarily by sys-temic osteotropic hormones, such as 1,25-dihydroxyvitaminD3. In contrast, osteoblasts and osteoclasts in cancellousbone are more likely to be controlled by potent osteotropiccytokines. In addition, the origin of osteoclasts is likely todiffer in the peripheral (e.g., forearm) and central skeleton(e.g., spine and hip). For instance, osteoclasts in the periph-eral skeleton originate primarily from circulating mono-cytes,(29,30) whereas osteoclasts in the central skeleton aremost likely derived from red marrow.(31) Thus, bone me-tabolism at these three skeletal sites may have biologicallydifferent regulatory mechanisms.

Our results suggest that the gene(s) underlying the QTLat 2q32 may involve pathways related to metabolism of

both cortical and trabecular bones. The existence of thispleiotropic QTL would provide a biological basis for theobserved phenotypic correlations among BMDs at the hip,spine and forearm.

The importance of chromosome region 2q32 has beensuggested in several previous studies. In a study of white,dizygotic twin sister pairs, Wilson et al.(32) found a sugges-tive linkage for quantitative ultrasound (QUS) of the cal-caneus at 2q33–37. Although QUS is a different parameterfrom BMD for characterizing bone, data showed that asmuch as one third of the genetic influence at the forearmsite is shared by genes controlling BMD and QUS param-eters.(33) In another study in the same population, a linkagesignal (LOD � 1.42) was identified for lumbar spine BMDat 2q33.(34) Hsu et al.(35) reported a QTL for whole bodyBMD on chromosome 2q31 in the Asian population (LOD� 2.71). Similarly, other groups also found linkages forBMD at different skeletal sites in this region in MexicanAmericans(36) and white men.(37)

Chromosome region 2q32 contains several interestingcandidate genes that are known to be involved in bonemineral metabolism. For instance, GDF8 (growth differen-tiation factor 8/myostatin), a member of the TGF-�� super-family, is a prominent candidate gene in 2q32. GDF8 is anegative regulator of skeletal muscle growth, and mice lack-ing the gdf8 gene show a significant increase in muscle massand strength compared with normal mice.(38) Recent stud-ies also showed that gdf8 knockout mice had greater BMDthan normal mice at both the spine and femur.(39,40) Thispositive correlation between muscle and BMD may be par-tially explained by the mechanical loading effects of skeletalmuscle on bone.(41,42) Another important candidate gene in2q32 is STAT1 (signal transducer and activator of transcrip-tion 1). Expression of the STAT1 gene plays an importantrole in developmental events involving fibroblast growthfactor (FGF)/FGF receptor (FGFR)3 endochondral boneformation and chondrocyte differentiation. In a gene ex-pression study in Chinese Han women (X-D Chen, P Xiao,S-F Lei, Y-Z Liu, F-Y Deng, L-J Tan, S-M Xiao, C Jiang, XSun, F Yang, Y-F Guo, S Wu, L-M Li, Y Chen, H Jiang,Z-H Tang, X-L Wang, Y Luo, M-Y Liu, X-Z Zhu, F-RChen, J-G Zhang, H-W Deng, unpublished data, 2007), theSTAT1 gene was found to be expressed at higher levels inthe low BMD than in the high BMD group, implicating thisgene in bone mineral metabolism. The stst1 gene has alsobeen found to contribute to bone metabolism in mice. Forexample, BMC significantly increased in stat1 knockoutmice compared with control mice.(43,44) Ex vivo studies also

TABLE 7. RESULTS OF MULTIPOINT LOD SCORES FROM BIVARIATE LINKAGE ANALYSIS FOR F/S, H/F, AND H/S ON CHROMOSOME 2

BivariateLocation

(cM) Nearest markerChromosome

region LOD2 LOD1 pPleiotropy

(p1)Co-incidentlinkage (p0)

F/S 189 GATA65C03M 2q32 3.60 2.65 0.00024 0.35 0.0005H/F 190 GATA65C03M 2q32 3.34 2.42 0.00042 0.07 0.004H/S 193 GATA65C03M 2q32 3.01 2.13 0.00087 0.15 0.001

F/S, bivariate analyses for forearm BMD × spine BMD; H/F, bivariate analyses for hip BMD × forearm BMD; H/S, bivariate analyses for hip BMD ×spine BMD; LOD1, 1 degree of freedom effective LOD score; LOD2, 2 degree of freedom effective LOD score.

FIG. 2. Results of bivariate linkage analysis for three trait pairs(forearm/spine BMD, hip/forearm BMD, and hip/spine BMD) onchromosome 2.

WANG ET AL.1676

revealed increased osteoblast replication and mineralizedbone nodule formation in mice lacking stat1.(43)

Our results confirm the general finding that joint analysisprovides more power for discovering QTLs than single-traitanalysis when mapping QTL for correlated traits. Our studyclearly showed that PCA and bivariate linkage analysismarkedly improved the power, as reflected by the increasedLOD score at 2q32. For those studies that have collectedmeasures on bone-related traits (such as BMD, bone ge-ometry, bone size, bone strength, fat mass, body mass in-dex, and lean body mass), joint analysis using multivariateanalytical tools (e.g., PCA and bivariate analysis) may pro-vide greater power for identifying QTLs that might bemissed or might not be evident through univariate analysis.

In this study, we did not perform bivariate linkage analy-sis for the BMD trait pairs through the whole genome.Instead, we first obtained PC1 that contains the commoninformation of the three traits by PCA. Bivariate linkageanalysis was only performed on the most significant linkageregion for PC1. By adopting this strategy, we avoided mul-tiple testing problems in whole genome bivariate linkageanalyses and thus, effectively increased the statistical powerfor detecting linkage.

In contrast to our previous study,(17) which focused pri-marily on identifying site-specific QTLs that independentlyeffect regulation of BMDs at several different sites, thisstudy focused on QTLs having concurrent effects on BMDsat multiple skeletal sites. We identified a potential QTLwith pleiotropic effects on chromosome region 2q32 forBMDs at the hip, spine, and forearm. Further replicationstudies and functional assessment of prominent candidategenes at this region is warranted.

ACKNOWLEDGMENTS

Investigators of this work were partially supported bygrants from NIH (R21 AG027110-01A1, R01 AG026564-01A2, K01 AR021170-01, R01 AR45349-01, and R01GM60402-01A1) and an LB595 grant from the State ofNebraska. This study also benefited from grants from Na-tional Science Foundation of China, Xi’an Jiaotong Univer-sity, Huo Ying Dong Education Foundation, Hunan Prov-ince, and the Ministry of Education of China. Thegenotyping experiment was performed by Marshfield Cen-ter for Medical Genetics and supported by NHLBI Mam-malian Genotyping Service (Contract HV48141).

REFERENCES

1. Duncan EL, Brown MA, Sinsheimer J, Bell J, Carr AJ, Words-worth BP, Wass JA 1999 Suggestive linkage of the parathyroidreceptor type 1 to osteoporosis. J Bone Miner Res 14:1993–1999.

2. Black DM, Cummings SR, Melton LJ III 1992 Appendicularbone mineral and a woman’s lifetime risk of hip fracture. JBone Miner Res 7:639–646.

3. Hui SL, Slemenda CW, Johnston CC Jr 1989 Baseline mea-surement of bone mass predicts fracture in white women. AnnIntern Med 111:355–361.

4. Nevitt MC, Johnell O, Black DM, Ensrud K, Genant HK,Cummings SR 1994 Bone mineral density predicts non-spine

fractures in very elderly women. Study of Osteoporotic Frac-tures Research Group. Osteoporos Int 4:325–331.

5. Smith DM, Nance WE, Kang KW, Christian JC, Johnston CCJr 1973 Genetic factors in determining bone mass. J Clin Invest52:2800–2808.

6. Dequeker J, Nijs J, Verstraeten A, Geusens P, Gevers G 1987Genetic determinants of bone mineral content at the spine andradius: A twin study. Bone 8:207–209.

7. Pocock NA, Eisman JA, Hopper JL, Yeates MG, SambrookPN, Eberl S 1987 Genetic determinants of bone mass in adults.A twin study. J Clin Invest 80:706–710.

8. Recker RR, Deng HW 2002 Role of genetics in osteoporosis.Endocrine 17:55–66.

9. Nguyen TV, Howard GM, Kelly PJ, Eisman JA 1998 Bonemass, lean mass, and fat mass: Same genes or same environ-ments? Am J Epidemiol 147:3–16.

10. Yang TL, Zhao LJ, Liu YJ, Liu JF, Recker RR, Deng HW2006 Genetic and environmental correlations of bone mineraldensity at different skeletal sites in females and males. CalcifTissue Int 78:212–217.

11. Livshits G, Deng HW, Nguyen TV, Yakovenko K, Recker RR,Eisman JA 2004 Genetics of bone mineral density: Evidencefor a major pleiotropic effect from an intercontinental study. JBone Miner Res 19:914–923.

12. Livshits G, Karasik D, Pavlovsky O, Kobyliansky E 1999 Seg-regation analysis reveals a major gene effect in compact andcancellous bone mineral density in 2 populations. Hum Biol71:155–172.

13. Deng HW, Stegman MR, Davies KM, Conway T, Recker RR1999 Genetic determination of variation and covariation ofpeak bone mass at the hip and spine. J Clin Densitom 2:251–263.

14. Devoto M, Spotila LD, Stabley DL, Wharton GN, Rydbeck H,Korkko J, Kosich R, Prockop D, Tenenhouse A, Sol-Church K2005 Univariate and bivariate variance component linkageanalysis of a whole-genome scan for loci contributing to bonemineral density. Eur J Hum Genet 13:781–788.

15. Karasik D, Myers RH, Hannan MT, Gagnon D, McLean RR,Cupples LA, Kiel DP 2002 Mapping of quantitative ultrasoundof the calcaneus bone to chromosome 1 by genome-wide link-age analysis. Osteoporos Int 13:796–802.

16. Karasik D, Myers RH, Cupples LA, Hannan MT, Gagnon DR,Herbert A, Kiel DP 2002 Genome screen for quantitative traitloci contributing to normal variation in bone mineral density:The Framingham Study. J Bone Miner Res 17:1718–1727.

17. Xiao P, Shen H, Guo YF, Xiong DH, Liu YZ, Liu YJ, Zhao LJ,Long JR, Guo Y, Recker RR, Deng HW 2006 Genomic re-gions identified for BMD in a large sample including epistaticinteractions and gender-specific effects. J Bone Miner Res21:1536–1544.

18. Deng HW, Xu FH, Huang QY, Shen H, Deng H, Conway T,Liu YJ, Liu YZ, Li JL, Zhang HT, Davies KM, Recker RR2002 A whole-genome linkage scan suggests several genomicregions potentially containing quantitative trait Loci for osteo-porosis. J Clin Endocrinol Metab 87:5151–5159.

19. Deng HW, Deng H, Liu YJ, Liu YZ, Xu FH, Shen H, ConwayT, Li JL, Huang QY, Davies KM, Recker RR 2002 A genome-wide linkage scan for quantitative-trait loci for obesity pheno-types. Am J Hum Genet 70:1138–1151.

20. Genant HK, Grampp S, Gluer CC, Faulkner KG, Jergas M,Engelke K, Hagiwara S, Van Kuijk C 1994 Universal standard-ization for dual x-ray absorptiometry: Patient and phantomcross-calibration results. J Bone Miner Res 9:1503–1514.

21. Almasy L, Blangero J 1998 Multipoint quantitative-trait link-age analysis in general pedigrees. Am J Hum Genet 62:1198–1211.

22. Self SGLK-Y 1987 Asymptotic Properties of Maximum Like-lihood Estimators and Likelihood Ratio Tests Under Non-standard Conditions. J Am Stat Assoc 82:605–609.

23. Amos C, de Andrade M, Zhu D 2001 Comparison of multi-variate tests for genetic linkage. Hum Hered 51:133–144.

A WHOLE GENOME SCAN FOR BMD 1677

24. Almasy L, Dyer TD, Blangero J 1997 Bivariate quantitativetrait linkage analysis: Pleiotropy versus co-incident linkages.Genet Epidemiol 14:953–958.

25. Williams FM, Spector TD 2006 Recent advances in the genet-ics of osteoporosis. J Musculoskelet Neuron Interact 6:27–35.

26. Livshits G, Otremski I, Kobyliansky E 1995 Genetics of humanbody size and shape: Complex segregation analysis. Ann HumBiol 22:13–27.

27. Nguyen TV, Blangero J, Eisman JA 2000 Genetic epidemio-logical approaches to the search for osteoporosis genes. J BoneMiner Res 15:392–401.

28. Favus MJ 2006 Primer on the Metabolic Bone Diseases andDisorders of Mineral Metabolism, 6th ed. American Societyfor Bone and Mineral Research, Washington, DC, USA.

29. Zambonin Zallone A, Teti A, Primavera MV 1984 Monocytesfrom circulating blood fuse in vitro with purified osteoclasts inprimary culture. J Cell Sci 66:335–342.

30. Horton MA, Spragg JH, Bodary SC, Helfrich MH 1994 Rec-ognition of cryptic sites in human and mouse laminins by ratosteoclasts is mediated by beta 3 and beta 1 integrins. Bone15:639–646.

31. Parfitt AM 1998 Osteoclast precursors as leukocytes: Impor-tance of the area code. Bone 23:491–494.

32. Wilson SG, Reed PW, Andrew T, Barber MJ, Lindersson M,Langdown M, Thompson D, Thompson E, Bailey M, ChianoM, Kleyn PW, Spector TD 2004 A genome-screen of a largetwin cohort reveals linkage for quantitative ultrasound of thecalcaneus to 2q33-37 and 4q12-21. J Bone Miner Res 19:270–277.

33. Knapp KM, Andrew T, MacGregor AJ, Blake GM, FogelmanI, Spector TD 2003 An investigation of unique and shared geneeffects on speed of sound and bone density using axial trans-mission quantitative ultrasound and DXA in twins. J BoneMiner Res 18:1525–1530.

34. Wilson SG, Reed PW, Bansal A, Chiano M, Lindersson M,Langdown M, Prince RL, Thompson D, Thompson E, BaileyM, Kleyn PW, Sambrook P, Shi MM, Spector TD 2003 Com-parison of genome screens for two independent cohorts pro-vides replication of suggestive linkage of bone mineral densityto 3p21 and 1p36. Am J Hum Genet 72:144–155.

35. Hsu YH, Xu X, Terwedow HA, Niu T, Hong X, Wu D, WangL, Brain JD, Bouxsein ML, Cummings SR, Rosen CJ, Xu X2007 Large-scale genome-wide linkage analysis for loci linkedto BMD at different skeletal sites in extreme selected sibships.J Bone Miner Res 22:184–194.

36. Kammerer CM, Schneider JL, Cole SA, Hixson JE, SamollowPB, O’Connell JR, Perez R, Dyer TD, Almasy L, Blangero J,Bauer RL, Mitchell BD 2003 Quantitative trait loci on chro-

mosomes 2p, 4p, and 13q influence bone mineral density of theforearm and hip in Mexican Americans. J Bone Miner Res18:2245–2252.

37. Peacock M, Koller DL, Fishburn T, Krishnan S, Lai D, Hui S,Johnston CC, Foroud T, Econs MJ 2005 Sex-specific and non-sex-specific quantitative trait loci contribute to normal varia-tion in bone mineral density in men. J Clin Endocrinol Metab90:3060–3066.

38. McPherron AC, Lawler AM, Lee SJ 1997 Regulation of skel-etal muscle mass in mice by a new TGF-beta superfamily mem-ber. Nature 387:83–90.

39. Hamrick MW, Pennington C, Byron CD 2003 Bone architec-ture and disc degeneration in the lumbar spine of mice lackingGDF-8 (myostatin). J Orthop Res 21:1025–1032.

40. Montgomery E, Pennington C, Isales CM, Hamrick MW 2005Muscle-bone interactions in dystrophin-deficient and myo-statin-deficient mice. Anat Rec A Discov Mol Cell Evol Biol286:814-22.

41. Andreoli A, Monteleone M, Van Loan M, Promenzio L, Tar-antino U, De Lorenzo A 2001 Effects of different sports onbone density and muscle mass in highly trained athletes. MedSci Sports Exerc 33:507–511.

42. Witzke KA, Snow CM 1999 Lean body mass and leg powerbest predict bone mineral density in adolescent girls. Med SciSports Exerc 31:1558–1563.

43. Xiao L, Naganawa T, Obugunde E, Gronowicz G, Ornitz DM,Coffin JD, Hurley MM 2004 Stat1 controls postnatal bone for-mation by regulating fibroblast growth factor signaling in os-teoblasts. J Biol Chem 279:27743–27752.

44. Kim S, Koga T, Isobe M, Kern BE, Yokochi T, Chin YE,Karsenty G, Taniguchi T, Takayanagi H 2003 Stat1 functionsas a cytoplasmic attenuator of Runx2 in the transcriptionalprogram of osteoblast differentiation. Genes Dev 17:1979–1991.

Address reprint requests to:Hong-Wen Deng, PhD

The Key Laboratory of Biomedical InformationEngineering

Ministry of EducationInstitute of Molecular Genetics

School of Life Science and TechnologyXi’an Jiaotong University

Xi’an, Shaanxi 710049, ChinaE-mail: hwdeng@mail.xjtu.edu.cn

Received in original form April 2, 2007; revised form July 15, 2007;accepted July 30, 2007.

WANG ET AL.1678