Genome-Wide Scan Identified QTLs Underlying Femoral Neck Cross-Sectional Geometry That Are Novel Studied Risk Factors of Osteoporosis


  • Dong-Hai Xiong,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Hui Shen,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Peng Xiao,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Yan-Fang Guo,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    2. Laboratory of Molecular and Statistical Genetics, College of Life Sciences, Hunan Normal University, Changsha, Hunan, China
    Search for more papers by this author
  • Ji-Rong Long,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Lan-Juan Zhao,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Yao-Zhong Liu,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Hong-Yi Deng,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Jin-Long Li,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    2. Seattle Biomedical Research Institute, Seattle, Washington, USA
    Search for more papers by this author
  • Robert R Recker,

    1. Osteoporosis Research Center and Department of Biomedical Sciences, Creighton University, Omaha, Nebraska, USA
    Search for more papers by this author
  • Hong-Wen Deng PhD

    Corresponding author
    1. Laboratory of Molecular and Statistical Genetics, College of Life Sciences, Hunan Normal University, Changsha, Hunan, China
    2. The Key Laboratory of Biomedical Information Engineering of Ministry of Education and Institute of Molecular Genetics, School of Life Science and Technology, Xi'an Jiaotong University, Xi'an, China
    3. Department of Orthopedic Surgery and Basic Medical Science, School of Medicine, University of Missouri/Kansas City, Kansas City, Missouri, USA
    • Department of Orthopedic Surgery and Basic Medical Science, School of Medicine, University of Missouri/Kansas City, 2411 Holmes Street, Room M3-C03, Kansas City, MO 64108-2792
    Search for more papers by this author

  • The authors state that they have no conflicts of interest.


A genome-wide screen was conducted using a large white sample to identify QTLs for FNCS geometry. We found significant linkage of FNCS parameters to 20q12 and Xq25, plus significant epistatic interactions and sex-specific QTLs influencing FNCS geometry variation.

Introduction: Bone geometry, a highly heritable trait, is a critical component of bone strength that significantly determines osteoporotic fracture risk. Specifically, femoral neck cross-sectional (FNCS) geometry is significantly associated with hip fracture risk as well as genetic factors. However, genetic research in this respect is still in its infancy.

Materials and Methods: To identify the underlying genomic regions influencing FNCS variables, we performed a remarkably large-scale whole genome linkage scan involving 3998 individuals from 434 pedigrees for four FNCS geometry parameters, namely buckling ratio (BR), cross-sectional area (CSA), cortical thickness (CT), and section modulus (Z). The major statistical approach adopted is the variance component method implemented in SOLAR.

Results: Significant linkage evidence (threshold LOD = 3.72 after correction for tests of multiple phenotypes) was found in the regions of 20q12 and Xq25 for CT (LOD = 4.28 and 3.90, respectively). We also identified eight suggestive linkage signals (threshold LOD = 2.31 after correction for multiple tests) for the respective geometry traits. The above findings were supported by principal component linkage analysis. Of them, 20q12 was of particular interest because it was linked to multiple FNCS geometry traits and significantly interacted with five other genomic loci to influence CSA variation. The effects of 20q12 on FNCS geometry were present in both male and female subgroups. Subgroup analysis also revealed the presence of sex-specific quantitative trait loci (QTLs) for FNCS traits in the regions such as 2p14, 3q26, 7q21 and 15q21.

Conclusions: Our findings laid a foundation for further replication and fine-mapping studies as well as for positional and functional candidate gene studies, aiming at eventually finding the causal genetic variants and hidden mechanisms concerning FNCS geometry variation and the associated hip fractures.


OSTEOPOROSIS IS A common skeletal disease that is becoming one of the biggest public health menaces, with the consequence of millions of fractures annually around the world.(1) Hip fracture is the most severe clinical outcome of age-related osteoporosis because of its high prevalence,(1,2) serious effects on quality of life,(3,4) and excessive therapeutic cost.(5,6) The leading cause of hip fracture is the reduced bone strength at the proximal femur(7) associated with low bone mass and poor bone quality. Clinically, BMD has been used to assess bone strength and predict fracture risk.(1,8) However, recent studies have shown that BMD can only account for, at most, 50-70% of total bone strength.(9,10) Other factors, such as bone geometry, bone remodeling status, and bone microarchitecture also play important roles independent of BMD in determining bone strength and the associated osteoporosis fracture.(10-12) Specifically, hip geometric variables can be used to substantially enhance the identification of people at high risk of hip fracture.(11,13,14) Furthermore, adverse changes in the femoral neck (FN) cross-sectional (FNCS) geometry parameters lead to FN fragility and increased risk of hip fracture in the elderly.(15,16)

In addition to environmental factors, genetic factors also influence hip geometry.(17,18) Heritability of FNCS geometric parameters ranged from 0.37 to 0.62.(19) Candidate genes underlying those variables were proposed.(16,20,21) However, genetic research for bone geometry is still in its infancy compared with that of BMD.

The first whole genome linkage scan (WGS) for human FNCS geometry suggested several potential important genomic regions.(19) Now we report the results of a genome-wide screen for the same geometric variables using a much expanded sample with 3998 whites from 434 pedigrees. The current sample size is more than twice the previous one (1816)(19) and thus this study has much higher statistical power.(22) This sample represents the largest one ever obtained from a single study population of the same ethnicity in the field of linkage studies of osteoporosis. We not only detected a number of FNCS geometry quantitative trait loci (QTLs) in both the entire sample and sex-specific subgroups, but also observed significant epistatic interaction effects.



The study was approved by the Creighton University Institutional Review Board. All the study subjects signed informed consent documents before entering the project. The study subjects came from an expanding database created for ongoing studies in the Osteoporosis Research Center (ORC) of Creighton University to search for genes underlying human complex traits such as BMD and body mass index (BMI). The sampling scheme and exclusion criteria have been detailed elsewhere.(23) Briefly, patients with chronic diseases and conditions that might potentially affect bone mass, structure, or metabolism were excluded. These diseases/conditions included chronic disorders involving vital organs (heart, lung, liver, kidney, brain), serious metabolic diseases (diabetes, hypo- and hyperparathyroidism, hyperthyroidism, etc.), skeletal diseases (Paget's disease, osteogenesis imperfecta, rheumatoid arthritis, etc.), chronic use of drugs affecting bone metabolism (hormone replacement therapy, corticosteroid therapy, anticonvulsant drugs), and malnutrition conditions (such as chronic diarrhea, chronic ulcerative colitis, etc.), and so forth. Blood samples were collected for extracting DNA, medical history and lifestyle questionnaires were administered, anthropometrics were obtained, and BMD and bone size were measured for calculating bone geometry parameters.

All the study subjects were whites of European origin. Specific for FNCS geometry traits, the sample contained a total of 4386 phenotyped subjects from 434 pedigrees (see Table 1 for their basic characteristics), of whom 3998 subjects were genotyped. The sample mainly consisted of pedigrees of median to large size and provided us an exceedingly large number of relative pairs (>150,000) informative for linkage analysis. For example, the number of sibling pairs and first cousin pairs are 6316 and 15,122, respectively. Among the genotyped subjects, 1816 were used in the previous WGS study(19) and the remaining 2182 were newly added into this study.

Table Table 1.. Characteristics of the Femoral Neck Cross-Sectional Geometric Variables in the Study Subjects Stratified by Age and Sex
original image


Areal BMD (g/cm2) and bone size (cm2) of FN were measured by DXA with a Hologic 1000, 2000+, or 4500 scanner (Hologic, Bedford, MA, USA). All scanners are calibrated daily, and long-term precision is monitored with external phantoms. The CVs of FN BMD and bone size measurement obtained on the Hologic 2000 + scanner were 1.87% and 1.94%, respectively. Similar CVs were obtained on Hologic 1000 and 4500 scanners.(24) BMD data obtained from different machines were transformed to a compatible measurement using the formula described in Genant et al.(25) and the algorithm that we developed in-house and used extensively.(26-28) Areal bone size measurements by different scanners in our center were highly compatible with one another and were well within the precision limits.(24) In particular and intuitively, members of the same pedigree were usually measured on the same type of machine, ensuring minimum or no effect on our linkage analysis because of measurements by different scanners.

Using DXA-derived FN BMD and bone size, we estimated four FNCS geometric variables. The algorithm and the underlying assumptions regarding the geometry and structure of femoral neck have been well detailed earlier.(15,29) Briefly, the method assumes that the bone within the FN region has the configuration of a uniform right circular cylinder, 60% of the measured bone mass is cortical (i.e., fc = 0.6), and the effective density of bone mineral in fully mineralized bone tissue is 1.05 g/cm3 (i.e., ρm = 1.05 g/cm3), which were well substantiated.(15)

The four estimated FNCS geometric variables are the following: buckling ratio (BR), an index of bone structural instability; cross-sectional area (CSA), an indicator of bone axial compression strength; cortical thickness (CT); and section modulus (Z), an index of bone bending strength. They are computed as follows:

equation image

where W is the FN periosteal diameter and can be approximated by dividing the areal bone size of FN by the width of the region of interest (in Hologic DXA systems, the width of the FN region is standardized at 1.5cm).(16)

equation image


For each subject, DNA was extracted from peripheral blood using the Puregene DNA isolation kit (Gentra Systems, Minneapolis, MN, USA). All the subjects were genotyped with 410 microsatellite markers (including 393 markers for 22 autosomes and 17 markers for the X chromosome) from the Marshfield screening set 14 by Marshfield Center for Medical Genetics. The markers had an average population heterozygosity of 0.75 ± 0.06 and spaced on average 8.9 cM apart. The detailed genotyping protocol is available at A genetic database management system (GenoDB)(30) was used to manage the phenotype and genotype data for linkage analyses. GenoDB was also used for allele binning (including setting up allele binning criteria and converting allele sizes to distinct allele numbers), data quality control, and data formatting for PedCheck(31) and linkage analysis. PedCheck was performed to ensure that the genotype data conform to Mendelian inheritance pattern at all the marker loci. In addition, we used MERLIN(32) to detect genotyping errors of unlikely recombination (e.g., double recombination) in our sample. The genotyping error rate of ∼0.03% was determined.

Statistical analyses

Variance component linkage analyses(33-35) for quantitative traits were performed using SOLAR (Sequential Oligogenic Linkage Analysis Routines),(33) available online ( Two-point and multipoint linkage analyses were performed for each FNCS geometric variable in the 434 pedigrees.

Age, sex, height, weight, and sex-by-age interactions were tested for importance on FNCS geometric phenotypes, and significant factors were adjusted as covariates in linkage analyses. Each phenotype was tested for normality of distribution by kurtosis using Kolmogorov-Smirnov test. In the 434 pedigrees, the kurtosis values of adjusted cross-sectional geometric variables ranged from 0.47 to 0.70. Although the variance component analyses implemented in SOLAR are quite robust to slight deviations from normality (kurtosis < 2.0),(36) we still carried out 10,000 simulations using the procedure “lodadj” implemented in SOLAR(37) to correct for such deviations and to test the robustness of our results. This method is considered to be one of the best ways to deal with normality issues.(38) Thus, estimated correction constants for LOD scores of the four FNCS geometric variables ranged from 0.99 to 1.05. All LOD scores given in the text were empirically adjusted LOD scores. In addition, we calculated empirical pointwise p values for adjusted LOD scores using the “empp” command in SOLAR.

Because the currently available version of SOLAR cannot handle multipoint linkage analysis for the X chromosome, we only calculated two-point LOD scores for X-specific markers. Other software, such as GENEHUNTER,(39) which is capable of multipoint linkage analysis for the X chromosome, however, cannot handle large pedigrees that made up the major part of our sample. Breaking down the large pedigrees into smaller ones might be an option, but this procedure would result in a considerable loss of statistical power.

To aid in interpretation of the linkage results, we performed pairwise correlation analyses among the FNCS geometric variables and hip BMD using SOLAR. Because four correlated FNCS geometric variables were analyzed, correcting for multiple analyses of related phenotypes was performed as described by Camp and Farnham.(40) Briefly, the number of effectively independent tests was estimated to be 2.5, and it was used to establish the genome-wide thresholds of “suggestive” and “significant” evidence for linkage for the four traits, which were 2.31 and 3.72, respectively.(40)

Two-locus analyses to test for epistatic interaction effects on the FNCS geometric variables were performed using the chromosomal region showing the strongest evidence for linkage (for BR, CSA, and CT: 20q12; for Z: 16q24) paired with any other loci harboring linkage peak with LOD score ≥1 for the corresponding trait in the single-locus multipoint genome-wide scan. Two levels of modeling in addition to single-locus modeling were performed: (1) two-locus models with only additive effects for each pair of loci; and (2) two-locus models with additive effects as well as an epistatic term for interaction between the two loci. One-tailed p values were generated using χ2 statistic with one degree of freedom for all hypotheses tested with respect to each geometric variable.(41) Significance for a single test for interaction effect on any one variable was assessed at a type I error rate of 0.05/N according to Bonferroni adjustment for multiple comparison. N was the number of independent tests conducted for each trait. In terms of CSA or CT, N was less than the number of epistatic interactions tested because several testing regions were close in position (<30 cM) and thus should not be regarded as independent.(42) The thresholds of significant interaction were calculated as 0.006 for BR, 0.004 for CSA, 0.005 for CT, and 0.010 for Z, respectively (Supplement Table 1).

To understand and corroborate our results of single-trait analyses, we also performed principal component analysis (PCA) of the four cross-sectional geometric variables, using the statistical package SAS (SAS v.6.12; SAS Institute, Cary, NC, USA). PCA method is effective for correlated phenotypes regulated by a common locus.(27,38) Individual scores of the first two principal components (PC1 and PC2) were used as phenotypes in the linkage and interaction analyses.

Finally, because of the earlier findings that there may exist gender-specific influences on bone geometry,(43-45) we also conducted linkage analyses for FNCS geometric variables in men and women separately in the 434 pedigrees. In the sex-specific analyses, the phenotype values for individuals of the opposite sex were recorded as missing data. Also, age, height, and weight were tested, and significant factors were adjusted as covariates in the analyses. Moreover, to be more precisely, the Marshfield sex-specific genetic maps were used in the subgroup analyses instead of the sex-average map used elsewhere in this study (The Marshfield electronic database is at


Quantitative genetic analysis

The basic characteristics of the four FNCS geometric variables in the study subjects stratified by age and sex are summarized in Table 1. The data show that males generally have larger values of geometric parameters than females. In both sexes, CSA, CT, and Z decrease with advancing age, whereas BR increases with aging. Such trends are consistent with those reported by previous studies.(15,45-48) Heritability estimates of BR, CSA, CT, and Z were 0.58 ± 0.03 (SE), 0.48 ± 0.03, 0.62 ± 0.03, and 0.38 ± 0.03, respectively, indicating that a substantial proportion of the variation in bone geometry is attributable to genetic effects. All of the four geometric variables were significantly correlated both genetically and phenotypically (Table 2). Also, as expected,(16) these geometric variables are significantly correlated with hip BMD (Table 2). The square of the phenotypic correlation that approximates the proportion of variation of one trait that can be attributable to the other is in the range of 1-77%. The square of the genetic correlation that approximates shared genetic effects between two traits is in the range of 10-88%, indicating the existence of independent gene effects contributing to each phenotype.

Table Table 2.. Correlations Between Femoral Neck Cross-Sectional Geometric Variables, Hip BMD, and Principal Components
original image

Baseline linkage analysis

Using conservative thresholds correcting for multiple analyses, namely “suggestive” evidence for linkage at LOD score 2.31 and “significant” evidence at 3.72, the multipoint genome-wide scan identified one QTL for BR (Fig. 1A), two QTLs for CSA (Fig. 1B), two QTLs for CT (Fig. 1C), and one QTL for Z (Fig. 1D). They are located at four chromosomal regions, which are 1p, 12p, 16q, and 20q, respectively (Fig. 1; Table 3). All of these regions were further plotted in detail in Figs. 2A-2D. We also plotted the two-point LOD score results for X chromosome in Fig. 2E, given that LOD scores >2.31 for CSA and CT were achieved at X-specific markers. Considering the threshold for “suggestive” evidence for a single classic linkage analysis,(49,50) we further summarized in Table 3 the genomic regions (for autosomes) or markers (for X chromosome) with LOD ≥ 1.86 for any one of the studied FNCS geometric variables.

Figure FIG. 1..

Multipoint linkage results for femoral neck cross-sectional geometric variables. The dashed horizontal lines indicate the threshold for suggestive linkage (+2.31) or significant linkage (+3.72) accounting for multiple testing.20

Figure FIG. 2..

Genomic regions showing at least suggestive linkage (LOD ≥ 2.31) for femoral neck cross-sectional geometric variables. BR, solid line; CSA, dotted line; CT, dashed line; Z, dot-dashed line.20

Table Table 3.. Chromosomal Genomic Regions With LOD Scores ≥ 1.86 for Femoral Neck Cross-Sectional Geometric Variables
original image

The most impressive region is 20q12, where suggestive linkage was detected for BR (LOD = 3.53 at marker GATA47F05), CSA (LOD = 2.60 between GATA42A03 and GATA47F05), and significant linkage for CT (LOD = 4.28 at GATA42A03) simultaneously (Fig. 2D; Table 3). This chromosomal region is a QTL cluster according to the definition given in a recent study.(38) Another two autosomal regions, namely 1p12 at marker GATA12A07N and 12p13 at AAC040Z, showed suggestive linkage for CSA (LOD = 2.31) and CT (LOD = 2.80), respectively (Figs. 2A and 2B; Table 3). As for variable Z, 16q24 at marker GATA11C06N gave the suggestive linkage evidence (LOD = 2.80; Fig. 2C; Table 3).

On the X chromosome, we detected suggestive linkage evidence for CSA in Xq25 at marker ATCT003 with a two-point LOD of 2.70 (Table 3, Fig. 2E). For CT, two X-specific regions showed suggestive linkage (two-point LOD = 2.38 in Xp11 at GATG011 and LOD = 2.37 in Xq23 at GATA172D05) and one showed significant linkage (two-point LOD = 3.90 in Xq25 at ATCT003).

Epistatic interaction analysis

Supplement Table 1 shows the results of the epistatic interaction analysis. With stringent criteria for statistical significance specific for each geometric variable, we found that there were five significant epistatic interactions influencing CSA variation (Fig. 3). The most striking one was that between 20q12 and 3q27 (Fig. 3B, epistatic 2-locus model LOD = 4.70; p = 0.0002) compared with the additive two-locus model (LOD = 1.98). 20q12 also significantly interacted with 2q33 (epistatic/additive two-locus model LOD: 2.59/0.86; p = 0.002), 2q37 (epistatic/additive two-locus model LOD: 2.40/0.93; p = 0.004), 7p15 (epistatic/additive two-locus model LOD: 2.40/0.88; p = 0.004), and 12q24 (epistatic/additive two-locus model LOD: 3.09/1.29; p = 0.002) on influencing CSA variation (Fig. 3). However, no significant epistatic effects on other femoral neck geometric variables were found.

Figure FIG. 3..

Epistatic interaction analyses for CSA in the entire sample. Single locus model, solid line; additive two locus model, dotted line; epistatic two locus model, dashed line.20

Principal component linkage analysis

Principal components analysis (PCA) transformed the original four FNCS geometric variables into two uncorrelated components, PC1 and PC2, which explained 85.2% and 14.4% of the total variation of all four FN geometric variables, respectively. The principle component loadings (i.e., the correlation coefficients between the variables and factors) suggest that PC1 is a primary factor for CSA variation, whereas PC2 is mainly responsible for BR variation (Table 2).

Linkage analyses using PC1 and PC2 as surrogate phenotypes showed that, except for 17q11, all of the genomic regions with LOD ≥ 1.86 for at least one of the original geometric variables still retained the similar linkage peaks for PC1 or PC2, although the magnitudes were smaller (Table 4). Moreover, we found that 20q12 significantly interacted with 2q33, 2q37, 3q27, 7p15, and 12q24 to exert epistatic effect on PC1 variation (data not shown), which was similar to the case of CSA.

Table Table 4.. Principle Component Analyses for Chromosomal Regions With LOD Scores ≥ 1.86 for Original Femoral Neck Cross-Sectional Geometric Variables
original image

Sex-specific linkage analysis

The results of sex-specific linkage analyses are summarized in Table 5. A major male-specific FNCS geometry QTL may be located in 7q21, where suggestive linkage was detected for CSA (LOD = 2.84 at marker GATA3F01) and CT (LOD = 2.76 at GATA3F01) simultaneously (Fig. 4A). 7q21 also showed a LOD score of 2.27 for BR at marker GATA5D08 (Fig. 4A). In females, three putative FNCS geometry QTLs were identified. First, 2p14 at GATA66D01 showed significant linkage for BR (LOD = 4.76; Fig. 4B) and suggestive linkage for CT (LOD = 2.64; Fig. 4B). Second, 3q26 at GATA3H01 showed suggestive linkage for BR, CSA, and CT (LOD = 2.66, 2.74, and 3.25, respectively; Fig. 4C) in females. The third potential female-specific FNCS geometry QTL was 15q21 that showed suggestive linkage signal for BR and CT (LOD = 3.19 and 2.79, respectively; Fig. 4D). We did not identify any sex-specific QTL for Z (Table 5).

Figure FIG. 4..

Genomic regions with LOD ≥ 2.31 in sex-specific linkage analyses for femoral neck cross-sectional geometric variables. BR, solid line; CSA, dotted line; CT, dashed line; Z, dot-dashed line.20

Table Table 5.. Genomic Regions With LOD Scores ≥ 1.86 in Sex-Specific Linkage Analyses for Femoral Neck Cross-Sectional Geometric Variables
original image

We also summarized in Table 5 the genomic regions with LOD ≥ 1.86 among males and females separately for any one of the studied FNCS geometric variables. One particular region worth mentioning was 20q12, which showed the LOD score of 1.99 for BR in females and LOD of 1.96 for CT in males (Table 5). Plotting of multipoint linkage results of chromosome 20 for BR and CT showed that the linkage patterns in both the entire sample and male/female subgroups were quite alike, although the linkage signals in the latter were weaker than in the former because of the much smaller sample sizes (Fig. 5). For CSA, the situation was similar to BR or CT, although the peak LOD scores within 20q12 in subgroups were <1.86 (data not shown).

Figure FIG. 5..

Linkage analyses of chromosome 20 for BR and CT in the entire sample and subgroups stratified by sex. The entire 434 pedigrees, solid line; males in the 434 pedigrees, dotted line; females in the 434 pedigrees, dashed line.20

Comparison with our previous FNCS geometry study

The results obtained in this study were compared with those of the first WGS for FNCS geometry QTL(19) (Tables 3 and 5). With the exception of Xq25 for CSA and 16q24 for Z, all of our major findings related to the entire sample (Table 3) in this study had the corresponding LOD scores in the earlier study(19) ≥1.0. Of them, the region of note was again 20q12, for which peak LOD scores >1.7 were found for BR, CSA, and CT,(19) consistent with our current findings about 20q12. Other regions of interest were Xp11, Xq23, and Xq25 because they reached suggestive or significant linkage threshold for CT in both studies.

As for sex-specific analyses, the major male-specific QTL detected in this study, 7q21, also had peak LOD scores >1.0 for BR and CSA previously (Table 5). Of the three major findings in females, 3q26 achieved LOD scores over 1.0 for CSA and CT in the former study, whereas the other two regions, 2p14 and 15q21, did not achieve impressive LOD scores previously (Table 5).


The studied four FNCS geometric variables are capable of reflecting bone fragility associated with osteoporotic fractures at the femoral neck.(15,29,51) Specifically, structural stability is reflected by BR.(15,52,53) CSA(29) is an indicator of bone axial strength.(54) CT is proportional to volumetric BMD and negatively correlated with hip fracture risk.(15,51) Finally, Z is an index of bending strength.(29) Significant associations were already reported between these FNCS geometric variables with hip fracture rate,(16) and such effects on fracture might be different from that of BMD according to the previous experience.(15,55-57) Given their clinical significance and high heritability, related studies are being conducted to unravel the underlying genetic factors, with certain progress already made.(16,19-21)

This study, using 3998 subjects from 434 pedigrees, represents the largest linkage study in osteoporosis to date. With more than doubled sample size, we refined one of our major findings in the first WGS for FNCS geometry, which was the suggestive linkage of 20p12-q12 with BR, CSA, and CT.(19) Instead of a broad linkage region of ∼40 cM wide covering 20p12-q12, this time all the three linkage peaks clustered within a 3-cM narrow region that is 20q12 (Fig. 2D). PCAs supported this finding (Table 4). In addition, compared with the first WGS,(19) maximum LOD scores across 20p12-q12 doubled for BR (from 1.94 to 3.53) and CT (from 2.09 to 4.28), plus increased maximum LOD for CSA (from 2.33 to 2.60). Such phenomenon served as a good example showing the importance of large sample sizes to linkage studies in terms of more precisely localizing the QTL-containing regions and greatly increasing the linkage signals,(22) which are unfeasible without sufficient recombination events. Sex-stratified analyses further showed that the effect of 20q12 on FNCS geometry was present in both genders, although with weaker linkage signals caused by the reduced sample size (Fig. 5). It also should be noted that our FNCS geometry QTL identified at 20q12 coincides with previous findings of Mitchell et al.(58) for serum osteocalcin (OC) concentrations, which is a widely used biochemical marker of bone turnover and may predominantly reflect skeletal development or turnover events associated with bone geometry for all ages.

There are several potential candidate genes for bone geometry that localize within or very closely with the region of peak linkage—20q12. The first is CDMP1, which encodes cartilage-derived morphogenetic protein 1, mediating the induction of bone and cartilage formation.(59,60) Deleterious mutations in CDMP1 can cause skeletal abnormalities.(61,62) The second is MMP-9 involved in bone modeling, which has been associated with BMD variation in Japanese.(63) Another candidate gene, NCOA3, as a nuclear receptor coactivator interacting with osteoporosis-related nuclear hormone receptors (e.g., thyroid hormone and vitamin D receptors),(64) was recently associated with lumbar spine BMD.(65)

The importance of 20q12 to FNCS geometry was also embodied in that it interacted with other genomic loci to significantly improve their linkage evidence for CSA (Supplement Table 1; Fig. 3). Interestingly, previous studies have reported linkage in those interacting regions with bone geometry and BMD phenotypes, such as 2q33-37 with quantitative ultrasound parameters,(66) 3q with proximal femur structure,(67) 7p15 with femoral BMD,(68) and 12q24 with femoral neck BMD.(69) It should be noted that most of those linkages were detected at femoral sites, which closely related to our findings.

The candidate genes contained in the above interacting regions include the following: PTH receptor 2 (PTHR2), IGF binding protein 2 (IGFBP2) and Indian hedgehog homolog (IHH) in 2q33; type IV collagen α3 chain (COL4A3) and type VI collagen α3 (COL6A3) in 2q37; α-2-HS-glycoprotein (AHSG) in 3q27; neuropeptide Y (NPY) and homeobox (HOX) genes in 7p15; and IGF-I and matrix metalloproteinase 17 (MMP17) in 12q24. Of these, IGF-I was recently associated with the FNCS geometric variables studied here and fragile fracture in whites,(16) as well as with femoral BMD(70) and bone area.(71) Other genes, such as AHSG and NPY, were also associated with FN BMD previously.(72,73) However, the primary genetic factors underlying CSA variation concerning those regions should be gene-gene interaction effects as suggested by the significantly elevated peak LOD scores conditioned on 20q12. Although the exact epistatic interaction mechanisms remain unknown at present, the identification of such interactions here may be crucial to understanding the contributions of genes, which, by themselves, have relatively small effects on CSA variation.

We also repeatedly found the significance of the X chromosome to CT. Our WGS for BMD variation(74) revealed the linkage of hip BMD to Xq27 that is <10 cM away from Xq25, which may partially supported the significant linkage of CT to Xq25. In addition, we detected the suggestive linkage of CSA to Xq25 this time. We found support in the study of Klein et al.,(43) which linked femoral midshaft CSA to the mouse X chromosome that was syntenic to the human X chromosome. The significance of Xq25 to CSA and CT was also substantiated by PCA (Table 4). However, few clear candidates reside within the identified X-specific loci.

We did not find linkage evidence for 8q24 and 10q26, two regions showing suggestive linkage for CSA and BR, respectively, in the first WGS. This is not unexpected considering the between-study differences in major confounding factors such as genetic heterogeneity, marker allele frequency, genotyping error rate, and statistical power.(22,74) On the other hand, three regions, 1p12, 12p13, and 16q24, were detected for the first time to be linked with CSA, CT, and Z, respectively, with the former two having peak LOD scores >1 in the first WGS(19) (Table 3). Again, PCAs supported our findings concerning those regions by showing similar linkage patterns, although with smaller LOD scores (Table 4). On 12p13, there are two putative candidate genes, namely peroxisome receptor 1 (PXR1) and matrix Gla protein (MGP), with MGP having been associated with BMD by an early study.(75) 16q24, syntenic to mouse chromosome 8, was reported to be linked with femoral biomechanics, structure, and density.(76) Moreover, 16q24 is right next to 16q23, which has been linked to FN BMD.(77)

It is worth mentioning that the data from linkage studies in inbred mice or rats for bone strength and structure are consistent with some of our data. The QTL regions for FNCS geometry detected in this study, such as 1p, 3p, 12q, 13q, and 16q, are homologous to the corresponding QTL regions in mice(43,76,78-80) (Chr 4, 6, 10, 14, and 8, respectively) or rats(81) (Chr 5, 4, 7, 15, and 19, respectively) for certain femur strength and structure phenotypes. These suggest that genetic determinants for bone biomechanics and geometry derived from mouse or rat studies can be extrapolated to humans with caution.

Several potential sex-specific QTLs for certain geometric variables were identified, such as 7q21 in males and 2p14, 3q26, and 15q21 in females (Table 5; Fig. 4). At these genomic regions, the linkage signals detected in the entire sample and the sex-specific subgroups exhibited similar patterns, but the former was much lower than the latter. This could be caused by the increased sample homogeneity after removing the influence of mixed genders. Conversely, these results might also be treated with caution because of the inflated false-positive and/or false-negative rates caused by increased multiple comparisons and insufficient power in individual subgroups. The putative male-specific QTL-7q21 contained the calcitonin receptor gene CTR, which was recently associated with various bone phenotypes including hip BMD and fracture in white men.(82) Of the female-specific QTLs, 3q26 was previously linked with femoral structure(67,83); 15q21 harbors the aromatase gene, CYP19, that was associated with hip BMD, fracture, and a number of estrogen-related bone phenotypes in white women.(84-86) These studies corroborated our sex-specific findings.

The exceedingly large sample size and the presence of complex pedigrees that make up the majority of our sample led to the huge number of informative relative pairs (>150,000) in this study. Thus, this work possessed much higher statistical power than most other linkage studies of complex traits, ensuring the robustness of our linkage results.(22) Furthermore, to correct for multiple analyses caused by four correlated phenotypes tested, we used the method of Camp and Farnham(40) to get the corrected genome-wide p values that were consistent with the original paradigm used for determining significance thresholds for the single scan.(50) Thresholds calculated this way had been proven to be conservative compared with those empirically determined by simulation.(40) Multiple testing is also a major problem in the epistasis analyses. To control for untoward effects of multiple comparisons, we only examined the possible interactions between the regions showing maximum LOD scores with regions having peak LODs ≥1.0. The corrected significant levels were obtained through the conservative Bonferroni method.

Several issues need to be addressed for this study. The first is concerning the limitations of this study, which has been well discussed before(19) and will not be repeated here. Second, this study should not be regarded as a “replication study” of our previous one(19) according to Lander's guidelines.(50) To preserve power to detect robust significant linkage, we analyzed the entire sample combining both formerly and newly recruited subjects and compared the results between two studies without claiming “replication” or “confirmation.” Third, we did not estimate the locus-specific variance using SOLAR because the implemented variance components method was well known to result in a large upward bias in such estimation.(87,88) Fourth, although PCA supported our findings concerning FNCS geometry traits by showing similar linkage patterns, compared with the single-trait analyses, the peak LOD scores decreased below the respective linkage thresholds except for 20q12 in PCA. However, this is normal because PCA is only effective in detecting common genetic components underlying correlated traits.(38) The inclusion of Z that shared no common linkage regions with the other three traits in PCA reduced the linkage signal to certain extent. On the other hand, the PCA data supported the clustering of maximum peak LOD scores for BR, CSA, and CT on 20q12, which was clearly observed in single-trait analyses. Finally, at the loci with highest LOD scores for FNCS geometry traits, families with >50 members comprised ∼35% of the entire sample and contributed ∼42% of the linkage signal, showing that large families account for only part of the linkage evidence.

Our effort resulted in the discovery of 20q12 and Xq25, which are of particular importance to the variation of FNCS geometry because of their high LOD scores exceeding the “significance” threshold corrected for multiple analyses. 20q12 is also likely to be a pleiotropic locus linked to multiple FNCS geometric traits. These two loci can serve as a starting point for further replication or fine-mapping study. Furthermore, we detected significant epistatic interaction effects between 20q12 and several promising genomic regions influencing CSA variation. Sex-specific QTL underlying FNCS geometry were identified as well. The knowledge obtained will turn out to be crucial to the identification of genetic polymorphisms associated with hip bone geometry and fractures, for which molecular biological studies are needed to eventually unravel the corresponding specific mechanisms. Those polymorphisms can also be used as genetic markers for femoral neck fragility as well as life-long indicators of hip fracture risk, with the aim of improving the diagnosis and treatment of osteoporotic fractures in humans.


Investigators of this work were partially supported by grants from NIH (K01 AR02170-01, R01 AR45349-01, R01 GM60402-01A1) and an LB595 grant from the State of Nebraska. The study was also benefited from grants from National Science Foundation of China, Huo Ying Dong Education Foundation, HuNan Province, Xi'an Jiaotong University, and the Ministry of Education of China. The genotyping experiment was performed by Marshfield Center for Medical Genetics and supported by NHLBI Mammalian Genotyping Service (Contract HV48141).