Identification of LTBP2 on Chromosome 14Q As a Novel Candidate Gene for Bone Mineral Density Variation and Fracture Risk Association

ORIGINAL ARTICLE

Endocrine Research

Identification of LTBP2 on Chromosome 14q as a Novel Candidate Gene for Bone Mineral Density Variation and Fracture Risk Association

Ching-Lung Cheung, Pak C. Sham, Vivian Chan, Andrew D. Paterson, Keith D. K. Luk, and Annie W. C. Kung

Department of Medicine (C.-L.C., V.C., A.W.C.K.), Genome Research Centre (P.C.S.), Orthopaedics and Traumatology (K.D.K.L.), The University of Hong Kong, Pokfulam, Hong Kong, China; and Program in Genetics and Genomic Biology (A.D.P.), The Hospital for Sick Children Research Institute, University of Toronto, Toronto, Ontario M5G 1L7, Canada

Context: Low bone mineral density (BMD) is a major risk factor for osteoporotic fracture. Chro- mosome 14q has previously been linked to BMD variation in several genome-wide linkage scans in Caucasian populations.

Objective: Our objective was to replicate and identify the novel candidate genes in the quantitative trait loci (QTL) at chromosome 14q QTL.

Subjects and Methods: Eighteen microsatellite markers were genotyped for a 117-cM interval in 306 Southern Chinese pedigrees with 1459 subjects. Successful replication of the QTL was confirmed within this region for trochanter and total hip BMD. Using a gene prioritization approach as implemented in the Endeavour program, we genotyped 65 single-nucleotide polymorphisms in the top five ranking candidate genes within the linkage peak in 706 and 760 case-control subject pairs with extremely high and low trochanter and total hip BMD, respectively.

Results: Single-marker and haplotype analyses revealed that ESR2 and latent TGF-␤ binding protein 2(LTBP2) had significant associations with trochanter and total hip BMD. Multiple logistic regression revealed a strong genetic association between LTBP2 gene locus and total hip BMD variation (P ϭ 0.0004) and prevalent fracture (P ϭ 0.01). Preliminary in vitro study showed differential expression of LTBP2 gene in MC3T3-E1 mouse preosteoblastic cells in culture.

Conclusions: Apart from ESR2, LTBP2 is a novel positional candidate gene in chromosome 14q QTL for BMD variation and fracture. (J Clin Endocrinol Metab 93: 4448–4455, 2008)

steoporosis is an important health problem worldwide as nificant bone loss. In addition, localization of candidate genes for O the prevalence of associated bone fractures is increasing BMD variation aids understanding of the pathogenesis of the due to prolonged life expectancy and aging of the population. disease and development of new therapies. Osteoporosis is a disorder in which there is a reduction in bone More than 20 genome-wide linkage scans (GWLS) have been strength that results in enhanced bone fragility and a consequent published on BMD and osteoporotic fractures. Nonetheless, in- increased fracture risk. Low bone mineral density (BMD) is one consistent results remain the major challenge in the quest for of the major determinants of bone strength and an important risk identification of genes that affect BMD. Suggestive or significant factor for osteoporotic fractures. BMD is under strong genetic linkages for BMD variation at several skeletal sites have been inference with a heritability estimate of 0.63–0.71 in women and detected on chromosome 14q (2–4). In our recent meta-analysis 0.74–0.79 in men (1). Early identification of at-risk subjects study of nine GWLS with 11,842 subjects, several significant allows preventive measures to be implemented before any sig- quantitative trait loci (QTL) were identified, and two neighbor-

0021-972X/08/$15.00/0 Abbreviations: BMD, Bone mineral density; CI, confidence interval; FN, femoral neck; Printed in U.S.A. GWLS, genome-wide linkage scans; LD, linkage disequilibrium; LRT, likelihood ratio test; MAF, minor allele frequency; QTL, quantitative trait loci; SNP, single-nucleotide Copyright © 2008 by The Endocrine Society polymorphism. doi: 10.1210/jc.2007-2836 Received December 26, 2007. Accepted August 5, 2008. First Published Online August 12, 2008

4448 jcem.endojournals.org J Clin Endocrinol Metab. November 2008, 93(11):4448–4455

The Endocrine Society. Downloaded from press.endocrine.org by [${individualUser.displayName}] on 11 September 2015. at 16:25 For personal use only. No other uses without permission. . All rights reserved. J Clin Endocrinol Metab, November 2008, 93(11):4448–4455 jcem.endojournals.org 4449

ing bins on chromosome 14, 14q13.1-q24.1 (P ϭ 0.003) and ment. All participants gave informed consent, and the study was ap- 14q23.3-q32.12 (P ϭ 0.022), were shown to be significantly proved by the Ethics Committee of the University of Hong Kong and linked to hip BMD variation (5). These combined findings sug- conducted according to the Declaration of Helsinki. For the family study, probands were identified from subjects with gest that chromosome 14 may harbor multiple candidate genes BMD Z score of less than or equal to Ϫ1.28 (the lowest 10th percentile that contribute to BMD variation at different skeletal sites. Es- of the population) at either the lumbar spine (L1–4) or hip; extended trogen receptor ␤ (ESR2) is one of the positional candidate genes family members were also invited to participate. We estimated family in this region that has been shown to contribute to BMD vari- informativeness for 1021 pedigrees. Based on previous heritability esti- ation (6). Given the QTL size and overall strength of linkage mates of BMD of 70% from a similar population, the expected LOD score for each pedigree was estimated, via regression, on the basis of signals observed on chromosome 14q, it is likely that more than phenotypic values, by using the option rankFamilies in Merlin-re- one susceptibility locus may reside within the QTL. Nonetheless, gress. Families with the highest informativeness were selected for it remains largely unknown which genes in the region also play evaluation (1). Three hundred six families with 1459 subjects (293 a role in BMD regulation. males and 1166 females) spanning two to four generations were an- alyzed. These pedigrees contained 1260 sib pairs, 143 cousin pairs, In this study, we carried out a linkage analysis of the chro- 2356 parent-child pairs, 522 grandparent-grandchild pairs, and 512 mosome 14 region in an independent set of southern Chinese avuncular pairs. Detailed descriptions of these families have been families in attempt to replicate the previously reported linkage reported previously (7, 8). findings. Using a newly developed bioinformatics tool for To increase the power of the association study, a threshold-defined gene prioritization, the five top-ranked genes under the QTL case-control design was adopted, and unrelated subjects from the op- posite extreme of the distribution of BMD were studied for the associ- were selected for association analysis using the tagging ap- ation analysis. Low BMD subjects were arbitrarily defined as those with proach. Our results revealed a novel candidate gene LTBP2 on a BMD Z score of less than or equal to Ϫ1.28 (equivalent to the lowest chromosome 14q that is associated with hip BMD variation 10th percentile of the population) at either the spine or hip. High BMD and fracture prevalence. subjects were sex-matched controls with a BMD Z score higher than ϩ1 (approximately equivalent to the 85th percentile of the population) at the corresponding bone site. We identified 833 unrelated case-control pairs of subjects, with 706 trochanter case-control pairs and 760 total Subjects and Methods hip case-control pairs. A total of 633 pairs constituted both trochanter and total hip cohorts. The case-control cohorts were sex and age Study population matched. Detailed inclusion and exclusion criteria have been described previously (8, 9). The study subjects were extracted from an expanding database being compiled at the Osteoporosis Centre at Queen Mary Hospital, the Uni- versity of Hong Kong, to determine the genetic and environmental risk BMD measurements and prevalent fracture assessment factors for osteoporosis. All study subjects were individuals of southern BMD (grams per square centimeter) at the spine L1–L4, femoral neck Chinese descent resident in the local community. They were recruited at (FN), trochanter, and total hip was measured by dual-energy x-ray ab- road shows and health talks on osteoporosis held between 1998 and sorptiometry (Hologic QDR 4500 plus; Hologic Waltham, MA). The in 2003 and were invited to the Osteoporosis Centre for BMD measure- vivo precision of the machine for spine, FN, and total hip region was 1.2, 1.5, and 1.5%, respectively (10). Weight and height were measured at the same visit. Thoraco- lumbar spine x-rays were assessed for radio- graphic evidence of spine fracture at baseline using the semiquantitative method (11). All low- trauma fractures at the spine, hip, and distal radius and morphometric fracture at the spine were included in the final analysis.

Microsatellite marker genotyping Genomic DNA was extracted from peripheral blood leukocytes using a phenol/chloroform extraction method. A total of 18 high-density microsatellite markers were genotyped in the chromosome 14 region delineated by D14S972 (located at 14q12) and D14S1007 (located at 14q32.2) encompassing a 117-cM interval. This region showed significant linkage to hip BMD variation in a recent meta-analysis. All markers were commercially available through PE Applied Biosystems (ABI PRISM Linkage Mapping Sets, version 2; Norwalk, CT). Marker order and map positions were obtained from the Marshfield map (Fig. 1). The average intermarker distance was 6.5 cM, and average population heterozygosity was 70%. Genotyping was performed on an ABI PRISM 3700 genetic analyzer using the GENES- CAN and GENOTYPER software for allele iden- FIG. 1. Multipoint LOD scores for linkage of spine, FN, trochanter, and total hip BMD to chromosome 14q. tification and sizing.

The Endocrine Society. Downloaded from press.endocrine.org by [${individualUser.displayName}] on 11 September 2015. at 16:25 For personal use only. No other uses without permission. . All rights reserved. 4450 Cheung et al. Chromosome 14q, BMD, and Fracture J Clin Endocrinol Metab, November 2008, 93(11):4448–4455

Statistical analysis for the linkage study to haplotype analysis. Although it does not include phase information, Multipoint regression-based linkage analysis was performed for the power of the two tests is similar (20). In the second stage, backward BMD at the spine, FN, trochanter, and total hip in the whole study logistic regression was applied to identify the most predictive SNPs in population using Merlin-regress (12, 13). Merlin-regress can handle each gene locus. nonrandomly ascertained samples and deviations from multivariate nor- Haplotype association was performed in two ways. In the first, we mality of the observed data but no loss in power when compared with the applied the default block definition in HAPLOVIEW using the method variance component method (13). Based on our previous heritability of Gabriel et al. (21). Haplotype tests of association were run using study, heritability of 0.7 was used for hip BMD in the Merlin-regress. On logistic regression on blocks of SNP markers identified in HAPLOVIEW the basis of the criteria, LOD of at least 1.3 would be considered as with or without adjustment of confounding factors. In the second ap- significant replication of a previous reported QTL (14). To obtain the proach, a three-marker sliding window approach was adopted using empirical P value corresponding to the linkage result, we empirically WHAP (22). For H haplotypes, the omnibus (global) test (HϪ1 degree derived the P values from 1000 simulated data sets using the gene-drop- of freedom test) assessed the overall association of haplotypes in the ping method in Merlin-regress. The P values were calculated for the full haplotype window with the trait. The haplotype-specific test was per- sample. If a LOD score of 1.3 is reached in 10 of the 1000 simulated data formed only if a significant association was observed in the global test. sets, this would indicate an empirical P value of 0.01. Associations between SNPs and prevalent fractures at the hip, spine, and distal forearm were assessed using a single-marker test and multiple Candidate gene selection, single-nucleotide logistic regression as described above. Symptomatic as well as morphometric spine fractures were included for analysis. polymorphism (SNP) selection, and genotyping An arbitrarily defined region of a 1-LOD interval from the linkage peak was selected for fine mapping. We applied a recently developed gene Power estimation of threshold-defined case-control prioritization approach, implemented in software Endeavor (15), to subjects identify the positional candidate genes. Endeavor prioritizes the candi- Power calculations were performed with the quantitative trait case- date genes in a three-step analysis. In the first step, the biological infor- control calculator in the Genetic power calculator (23). We presumed mation about the disease is gathered from a set of training genes, which that the tested marker was the QTL itself or was in complete LD with the are known to affect the disease. The biological information is based on causal allele and that the QTL follows additive inheritance. the literature, gene ontology, gene expression data, expressed sequence The power estimation based on the above method revealed that our tag expression data, protein domain, protein-protein interaction, path- study had over 80% power to detect a marker responsible for at least 1% way, cis-regulatory modules, transcriptional motifs, and sequence sim- of BMD variance at the ␣-level of 1–5%. ilarity. The training genes were selected based on our recent review of the genetics of osteoporosis (16): VDR, ESR1, BMP2, IL-6, IGF-I, CYP19, LTBP2 gene expression in MC3T3-E1 mouse LRP5, CLCN7, TGF-␤, COL1A1, and SOST. In the second step, a set of testing genes is loaded to the software. In our study, the chromosomal preosteoblast cells region 14q22.1-24.3 was loaded for testing and prioritizing the genes. In MC3T3-E1 mouse preosteoblast cells were plated at 1.5 ϫ 105 the last step, the testing genes are ranked based on the similarity with the cells/ml in a 24-well plate. Each well contained ␣-MEM supplemented training properties (i.e. the biological information gathered in the first with 10% fetal calf serum, and growth medium was replaced every 2–3 step). In our study, the top five candidate genes were selected for asso- d. Total RNA was isolated using TRI Reagent (Molecular Research Cen- ciation analysis. ter Inc., Cincinnati, OH) at d 0, 1, 3, 5, 7, 10, 15, and 20 after incubation. To improve the efficiency of association studies, an aggressive tagging The amount of LTBP2 and 18S mRNA expression was quantitated by approach (17) was adopted to select a subset of informative SNPs in each semiquantitative PCR with LTBP2 primers forward GAGCTCATGAT- gene from the HAPMAP Chinese population data Rel 21/Phase II (18), GGCAGTGTG and reverse GCTCCTTCCACTGGGATGTA and 18S with force-tagging of the SNPs located in 5Ј-untranslated region and primers forward TTTCGAGGCCCTGTAATTGGAATGA and reverse coding region. SNPs were genotyped using the high-throughput Seque- TTCAAAGTAAACGCTTCGGGCCC, respectively. nom MassARRAY system (Sequenom, San Diego, CA). Genotyping was repeated in 5% of the samples for verification and quality control; genotype data were confirmed to have an error rate of less than 0.1%. Results Association analyses Multipoint linkage analysis The genotyping quality of each SNP was first checked for the call rate, minor allele frequency (MAF), and Hardy-Weinberg equilibrium using The results of the multipoint linkage analysis in 306 families the HAPLOVIEW (19). SNPs with MAF less than 0.01, call rate less than are summarized in Fig. 1. Demographic data have been previ- 90%, and Hardy-Weinberg equilibrium less than 0.01 were excluded ously described (7). One marker (D14S63) achieved a LOD score from further analysis. Genotype and allele frequencies for each SNP were higher 1.3 with trochanter BMD, with the maximum LOD score determined by gene counting. Pairwise linkage disequilibrium (LD) was of 1.55 (nominal P value ϭ 0.004; empirical P value ϭ 0.006) at calculated as r2 for all SNP-pair combinations using HAPLOVIEW. Odds ratio and 95% confidence interval (CI) were determined using 69.5 cM near the marker D14S63. For total hip BMD, three binary logistic regression to determine the association of the SNPs and markers (D14S63, D14S258, and D14S1036) achieved a LOD SNP haplotype and the BMD status (high vs. low); BMD was adjusted for score higher than 1.3, with the maximum LOD score of 1.78 age, sex, height, and weight in the logistic model. The global association (nominal P ϭ 0.002; empirical P value ϭ 0.009) at 69.5 cM near per candidate gene was assessed by multiple logistic regression with ad- the marker D14S63. A LOD score lower than 1.3 was detected justment for confounding factors (age, height, weight, and sex) while controlling for other loci located at the same gene. This omnibus test for spine and FN BMD. [likelihood ratio test (LRT)] provided a single test for a collection of SNPs per candidate gene locus and an overall evidence of association. It also Candidate gene selection and association analyses ameliorated the multiple comparisons that are usually encountered in a single-marker test, because the LRT has already accounted for all mark- In the region of the 1-LOD interval, i.e. 14q22.1-24.3, 344 ers in the same gene locus and provided a single P value for overall genes were identified (NCBI build 36). We applied the gene-priori- association. In practice, this genotypic multiple regression is comparable tization approach implemented in the Endeavor program to prior-

TABLE 1. Results of single-marker test for trochanter and total hip BMD

Additive model Dominant model Recessive model Allele Case Ctrl Gene SNP Change MAF MAF Emp-P OR (95% CI) Emp-P OR (95% CI) Emp-P OR (95% CI) Phenotype: trochanter BMD (n ϭ 706 case-control pairs) ESR2 rs1256064 C3G 0.11 0.08 0.006 2.3 (1.2–4.1) 0.004 2.7 (1.4–5.2) 0.865 0.9 (0.1–7.8) ESR2 rs944052 C3T 0.39 0.40 0.185 0.8 (0.6–1.1) 0.857 1.0 (0.6–1.5) 0.019 0.5 (0.3–0.9) ESR2 T-1213C T3C 0.11 0.09 0.023 2.0 (1.1–3.6) 0.006 2.4 (1.3–4.4) 0.260 0.2 (0.0–3.4) LTBP2 rs7569 C3T 0.23 0.27 0.069 0.7 (0.5–1.0) 0.047 0.6 (0.4–1.0) 0.326 0.6 (0.3–1.6) LTBP2 rs2359141 G3A 0.48 0.43 0.041 1.4 (1.0–1.9) 0.234 1.3 (0.8–2.2) 0.032 1.9 (1.1–3.4) ESRRB rs12436385 A3G 0.14 0.14 0.304 0.7 (0.5–1.2) 0.464 0.8 (0.5–1.4) 0.027 0.2 (0.0–0.9) ESRRB rs4903413 T3C 0.17 0.20 0.579 1.2 (0.8–1.8) 1.000 1.0 (0.6–1.6) 0.047 3.8 (1.0–14.0) Phenotype: total hip BMD (n ϭ 760 case-control pairs) ESR2 rs1256064 C3G 0.11 0.08 0.007 2.3 (1.2–4.2) 0.002 2.8 (1.4–5.4) 0.857 0.7 (0.1–6.8) ESR2 rs944052 C3T 0.39 0.40 0.051 0.7 (0.5–1.0) 0.283 0.8 (0.5–1.2) 0.036 0.5 (0.3–0.9) ESR2 T-1213C T3C 0.12 0.08 0.005 2.2 (1.2–3.9) 0.002 2.6 (1.4–4.8) 0.467 0.3 (0.0–3.2) LTBP2 rs2286411 C3T 0.13 0.16 0.032 0.6 (0.4–1.0) 0.060 0.6 (0.4–1.0) 0.138 0.2 (0.0–1.7) LTBP2 rs3825709 G3A 0.10 0.08 0.046 1.7 (1.0–3.1) 0.121 1.6 (0.9–3.1) 0.003 7.8 (0.9–65.7) LTBP2 rs2043948 C3T 0.15 0.20 0.038 0.7 (0.4–1.0) 0.104 0.7 (0.4–1.1) 0.158 0.4 (0.1–1.3) LTBP2 rs1866629 A3G 0.11 0.09 0.039 1.7 (1.0–3.0) 0.050 1.8 (1.0–3.2) 1.000 1.6 (0.1–28.2) ESRRB rs4414418 G3C 0.32 0.32 0.066 1.4 (1.0–2.0) 0.049 1.6 (1.0–2.4) 0.356 1.4 (0.6–3.0) ESRRB rs12436385 A3G 0.15 0.15 0.174 0.8 (0.5–1.2) 1.000 0.9 (0.5–1.4) 0.004 0.2 (0.0–0.7)

Only P values Ͻ 0.05 are shown. Dominant model is Aa ϩ aa vs. AA, and recessive model is aa vs. AA ϩ Aa, where A represents the major allele, and a represents the minor allele. Ctrl, Control; Emp-P, empirical P value obtained through 10,000 permutations; OR, odds ratio. itize these genes within the QTL of 14q. The five top-ranked genes of 0.8 was used to select a set of informative SNPs for each gene were estrogen receptor ␤ (ESR2), transforming growth factor ␤ 3 based on the HapMap Chinese population data Rel 21/Phase II (TGFB3), bone morphogenetic protein 4 (BMP4), estrogen recep- (2005). Three, six, 16, seven, and 26 SNPs were tagged for BMP4, tor related ␤ (ESRRB), and LTBP2. Linkage disequilibrium (LD) ESR2, LTBP2, TGFB3, and ESRRB, respectively, with an inter- pattern investigation of our data revealed no evidence of LD be- marker distance of 4.5 kb. We excluded three SNPs in ESRRB in the tween these five genes on chromosome 14q (data not shown). analysis because they did not pass the genotyping quality control. To assess the association of these five genes with trochanter Table 1 shows the significant results of the single-marker test and total hip BMD variation, a tagging approach with average r2 and BMD variation at either trochanter or total hip. Clinical

TABLE 2. Results of multiple locus analysis of candidate genes using multiple logistic regression and subsequently with backward logistic regression to identify the most predictive SNPs for trochanter BMD and total hip BMD

SNP remaining in the final logistic model Gene Omnibus P value Marker P value OR 95% CI Phenotype: trochanter BMD LTBP2 0.011 rs7569 0.046 0.49 0.25 0.99 rs2302114 0.010 2.66 1.27 5.60 rs699370 0.0007 5.21 2.01 13.54 rs862028 0.0007 5.59 2.06 15.15 rs3742794 0.0009 4.86 1.92 12.32 rs934997 0.005 0.31 0.14 0.70 rs2359141 0.016 1.95 1.13 3.35 Phenotype: total hip BMD LTBP2 0.0004a rs7569 0.004 0.35 0.17 0.71 rs2286411 0.0002a 0.27 0.14 0.54 rs2302114 0.086 1.59 0.94 2.70 rs699370 0.001 3.60 1.64 7.94 rs862028 0.0004a 3.47 1.75 6.88 rs2028377 0.054 3.85 0.98 15.21 rs11159089 0.077 5.48 0.83 35.94 rs3825709 0.046 2.11 1.01 4.38 rs12435481 0.038 0.17 0.03 0.91 rs1866629 0.042 6.20 1.07 35.87

Only results with omnibus P value Ͻ 0.05 are shown. OR, Odds ratio. a Remained significant after Bonferroni correction for 55 SNPs and two traits.

The Endocrine Society. Downloaded from press.endocrine.org by [${individualUser.displayName}] on 11 September 2015. at 16:25 For personal use only. No other uses without permission. . All rights reserved. 4452 Cheung et al. Chromosome 14q, BMD, and Fracture J Clin Endocrinol Metab, November 2008, 93(11):4448–4455

TABLE 3. Results of haplotype analysis of candidate genes using a three-marker sliding window approach

Omnibus test Haplotype-specific test Gene Marker 1 Marker 2 Marker 3 P valuea P valueb Haplotype Frequency P valuea P valueb ORc Phenotype: trochanter BMD LTBP2 rs7569 rs2286411 rs862046 0.010 0.021 CCA 0.446 0.014 0.035 1.40 Phenotype: total hip BMD TCA 0.02 0.007 0.008 0.16 LTBP2 rs3825709 rs12435481 rs2043948 0.026 0.131 GTT 0.160 0.003 0.023 0.62 ATC 0.024 0.037 0.082 3.16

Only P values Ͻ 0.05 are shown. a Nominal P value without adjustment. b P value with adjustment of age, sex, height, and weight. c Odds ratio (OR) of more than 1 indicates that the haplotype of interest is associated with low BMD, and OR less than 1 indicates association with high BMD. characteristics of subjects have been previously described (6). analysis revealed significant associations between LTBP2 and Using an empirical P value of 0.05 through 10,000 permutations, trochanter and total hip BMD variation (Table 3). Haplotype the following SNPs were significantly associated with BMD at analysis using block haplotypes revealed similar results (Table the trochanter in either the additive, dominant, or recessive mod- 4), with the block haplotypes in LTBP2 demonstrating a signif- el: rs1256064, rs944052, and T-1213C in ESR2; rs2359141 and icant association with BMD variation at both trochanter and rs7569 SNPs in LTBP2; and rs12436385 and rs4903413 in ES- total hip. The block haplotype in ESR2 showed a significant RRB. There was no significant association between SNPs in association with trochanter BMD variation. TGFB3 and BMP4 with trochanter BMD variation. Similar results were obtained for total hip BMD: rs1256064, rs944052, Association of LTBP2 with prevalent fractures and T-1213C in ESR2; rs2286411, rs3825709, and rs2043948 The association between 16 SNPs in LTBP2 gene locus and in LTBP2; and rs4414418 and rs12436385 in ESRRB (Table 1). prevalent fracture at the spine, hip, and distal radius was exam- The overall gene locus effect of each candidate gene was as- ined. These were the most common osteoporotic fracture sites. sessed using gene-based multiple logistic regression (Table 2) The omnibus test revealed an overall association of LTBP2 gene that included all the SNPs in a gene as predictor variables. The locus with fractures, even after adjustment of BMD at all sites strongest association was observed for LTBP2 (P ϭ 0.011 for (P Ͻ 0.05) (Table 5). Six SNPs remained in the final logistic trochanter BMD and 0.0004 for total hip BMD). The omnibus P model with or without adjustment of BMD. The single-marker value for total hip BMD remained significant after conservative test revealed similar findings; rs862046 and rs2302114 showed Bonferroni correction of multiple testing with 55 SNPs and two significant associations (P Ͻ 0.05) with prevalent fractures with- skeletal sites. out adjustment of BMD. After adjustment of BMD, only In the second stage, backward logistic regression was used to rs862046 remained significant. identify the most predictive SNPs in each gene locus while controlling for other SNPs located at the same gene. The results are LTBP2 gene expression in MC3T3-E1 cells shown in Table 2. Multiple SNPs in LTBP2 remained in the final In vitro study using the MC3T3-E1 mouse preosteoblast logistic model, suggesting they each had a significant main effect cells demonstrated constitutive expression of LTBP2 gene, on trochanter and total hip BMD variation. with differential expression during osteoblastic proliferation In the haplotype analysis, we investigated the regional asso- (first4dofculture) and subsequent differentiation into ma- ciation by performing the three-marker sliding window test. The ture osteoblasts (Fig. 2).

TABLE 4. Results of haplotype analysis of candidate genes using a block haplotype approach

Omnibus test Haplotype-specific test Block Markers included Gene P valuea P valueb Haplotype Frequency P valuea P valueb ORc Phenotype: trochanter BMD 4 rs1255998 rs8018687 rs928554 ESR2 0.157 0.039 GAACTC 0.098 0.058 0.011 2.12 rs1256064 rs944052 T-1213C CAACTT 0.087 0.008 0.102 0.63 7 rs934997 rs2359141 LTBP2 0.043 0.03 AA 0.141 0.016 0.016 1.78 Phenotype: total hip BMD 5 rs7569 rs2286411 LTBP2 0.024 0.007 CC 0.606 0.007 0.003 1.61 CT 0.014 0.095 0.025 0.58

TABLE 5. Results of multiple locus analysis of LTBP2 using multiple logistic regression and subsequently with backward logistic regression to identify most predictive SNPs for fracture at any site with and without adjustment of BMD

Without adjustment of BMDa With adjustment of BMD§ 95% CI 95% CI SNP P value OR Lower UpperP value OR Lower Upper Global 0.018 0.010 rs2286411 0.005 2.15 1.26 3.66 0.002 2.39 1.38 4.15 rs699370 0.021 0.66 0.46 0.94 0.035 0.68 0.47 0.97 rs862028 0.005 0.35 0.17 0.73 0.003 0.33 0.16 0.69 rs3742794 0.007 0.37 0.18 0.76 0.007 0.36 0.17 0.76 rs2028377 0.019 4.38 1.28 15.00 0.022 4.38 1.23 15.56 rs934997 0.060 1.87 0.97 3.59 0.062 1.89 0.97 3.67

OR, Odds ratio. a P value with adjustment of age, height, weight, sex, and history of fall. b P value was further adjusted with lumbar spine L1–L4, FN, trochanter, and total hip BMD.

Discussion consistent and multiple significant associations at the single- SNP, multiple-SNP, and haplotype levels, suggesting LTBP2 is Previous GWLS provided evidence of linkage to BMD variation another positional QTL gene on chromosome 14q. Using frac- at the spine (3), FN (5), trochanter (2), and total hip (4) and BMD ture per se as the phenotype, LTBP2 also showed a significant at all sites (24). In the present study, successful replication of association with prevalent fractures, even after adjustment of linkage was detected at the trochanter and total hip but not spine BMD, suggesting LTBP2 may exert its independent effect on or FN BMD. According to our linkage findings, the effect size of BMD and fracture. QTL peak observed on chromosome 14q as estimated using LTBP2 is located at 84 cM, in close proximity to the second Ϯ Merlin-regress contributed 15.2 5.4% to the final variance in peak near the marker D14S1036 (Fig. 1). The LTBP2 protein is BMD. In our previous study, ESR2 accounted for only 1% of the known to play a structural role within elastic fiber and affects final variance in women and 4–7% in men (6). Therefore, we extracellular matrix homeostasis (26). It belongs to the family of hypothesized that there may be other candidate genes located in fibrillin/LTBP glycoproteins and contains two domains, epider- the same QTL to explain the strength of signal, although it is well mal growth-factor like domains and eight-cysteine repeats, that recognized that QTL effect-size estimates from linkage studies are important in gene-gene interaction (27). In a recent in vitro are imprecise and often overestimated (25). study of LTBP2 expression during chondrogenic differentiation Over 300 known genes are present under the QTL on chro- of mesenchymal stem cell and chondrocytes, LTBP2 was up- mosome 14q. Prioritization of the candidate gene while focusing regulated during dedifferentiation and down-regulated during on just a few genes was possible by integrating the information chondrocyte differentiation (28). In addition, the level of expres- available from multiple publicly available databases (15). Using sion of LTBP2 in the joint differed between patients and controls this approach to select the top five genes from genotyping, we with osteoarthritis (29) and in the synovium of patients with were able to reconfirm our previous findings of the association systemic lupus erythematosus (26). Although this study did not of ESR2 with trochanter and hip BMD. Besides, LTBP2 showed evaluate the role of LTBP2 in bone metabolism, preliminary in vitro study using the MC3T3-E1 mouse preosteoblast cells demonstrated differential expression of LTBP2 during different stages of osteoblast differentiation (Fig. 2). Based on the common pathway of chondrocyte and osteoblast differentiation from mesenchymal stem cells, and together with our association and preliminary expression data, it is likely that LTBP2 may be involved in osteoblast differentiation, BMD determination, matrix homeostasis, and fracture etiology. The present study, with over 80% power, identified a number of significant markers with the gene-based omnibus approach even after correction for multiple testing for 55 markers and two traits (P Ͻ 0.00045) that warrant further evaluation in other populations (30). Although it could be argued that none of the SNPs were significant in the single-marker test (Table 1) if cor- rected for multiple testing, it should be noted that allelic heterogeneity (i.e. presence of more than one susceptibility allele in a FIG. 2. LTBP2 gene expression in MC3T3–E1 mouse preosteoblast cells at d 0, 1, 3, 5, 7, 10, 15, and 20 of culture; the lower panel shows the semiquantitation of locus or gene) greatly reduced the power for testing of an indi- LTBP2 mRNA expression (mean Ϯ SD) with RT-PCR. vidual SNP (31). Therefore, a single gene-based omnibus test

The Endocrine Society. Downloaded from press.endocrine.org by [${individualUser.displayName}] on 11 September 2015. at 16:25 For personal use only. No other uses without permission. . All rights reserved. 4454 Cheung et al. Chromosome 14q, BMD, and Fracture J Clin Endocrinol Metab, November 2008, 93(11):4448–4455

(LRT) was used to ameliorate the situation by simply testing the References global null hypothesis about the SNPs located per gene. The gene-based omnibus test is a direct and powerful means of pro- 1. Ng MY, Sham PC, Paterson AD, Chan V, Kung AWC 2006 Effect of environmental factors and gender on the heritability of bone mineral density and tecting the overall false-positive rate when a collection of loci are bone size. Ann Hum Genet 70:428–438 tested, because the P value from the omnibus test has already 2. Koller DL, Econs MJ, Morin PA, Christian JC, Hui SL, Parry P, Curran ME, Rodriguez LA, Conneally PM, Joslyn G, Peacock M, Johnston CC, Foroud T reflected the number of SNPs included in the number of degrees 2000 Genome screen for QTLs contributing to normal variation in bone min- of freedom (32). In this regression framework, the association eral density and osteoporosis. J Clin Endocrinol Metab 85:3116–3120 was examined with controlling for other SNPs at the same gene 3. 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 without stratifying by effects at other loci, hence having more loci contribute to normal variation in bone mineral density in men. J Clin advantage over the simple single-marker test (33). Using this Endocrinol Metab 90:3060–3066 omnibus test approach, a significant enrichment in P value was 4. Streeten EA, McBride DJ, Pollin TI, Ryan K, Shapiro J, Ott S, Mitchell BD, Shuldiner AR, O’Connell JR 2006 Quantitative trait loci for BMD identified observed in the LTBP2 gene when compared with the single- by autosome-wide linkage scan to chromosomes 7q and 21q in men from the marker test, suggesting the presence of untagged susceptibility Amish Family Osteoporosis Study. J Bone Miner Res 21:1433–1442 5. Ioannidis JP, Ng MY, Sham PC, Zintzaras E, Lewis CM, Deng HW, Econs MJ, variants. Conversely, the P value in the omnibus test of ESR2 was Karasik D, Devoto M, Kammerer CM, Spector T, Andrew T, Cupples LA, not enriched when compared with the single-marker test. This Duncan EL, Foroud T, Kiel DP, Koller D, Langdahl B, Mitchell BD, Peacock suggested that either an absence of an additional susceptibility M, Recker R, Shen H, Sol-Church K, Spotila LD, Uitterlinden AG, Wilson SG, Kung AW, Ralston SH 2007 Meta-analysis of genome-wide scans provides variant or the causal variant may already be included or tagged evidence for sex- and site-specific regulation of bone mass. J Bone Miner Res (with very high LD) in the study. 22:173–183 In the haplotype analyses, a number of rare haplotypes were 6. Kung AW, Lai BM, Ng MY, Chan V, Sham PC 2006 T-1213C polymorphism of estrogen receptor ␤ is associated with low bone mineral density and osteo- associated with BMD variation. These observations may be ex- porotic fractures. Bone 39:1097–1106 plained by the existence of an additional rare variant in LD with 7. Cheung CL, Huang QY, Ng MY, Chan V, Sham PC, Kung AW 2006 Con- the rare haplotypes. Although rare haplotypes may not be firmation of linkage to chromosome 1q for spine bone mineral density in southern Chinese. Hum Genet 120:354–359 robust to genotyping error and type I error, the significant 8. Huang QY, Ng MY, Cheung CL, Chan V, Sham PC, Kung AW 2006 Identi- result generated from the haplotype omnibus test in a rela- fication of two sex-specific quantitative trait loci in chromosome 11q for hip bone mineral density in Chinese. Hum Hered 61:237–243 tively large case-control cohort suggested the overall effect of 9. Cheung CL, Chan V, Kung AW 2008 A differential association of ALOX15 the test loci should be valid (34). Of course, future replication polymorphisms with bone mineral density in pre- and post-menopausal study in our population will help to validate our findings. women. Hum Hered 65:1–8 10. Kung AW, Yeung SS, Lau KS 1998 Vitamin D receptor gene polymorphisms Apart from the gene prioritization method, there are other and peak bone mass in southern Chinese women. Bone 22:389–393 approaches for selection of candidate genes, and we cannot 11. Genant HK, Wu CY, van Kuijk C, Nevitt MC 1993 Vertebral fracture assess- preclude the possibility of the presence of other candidate ment using a semiquantitative technique. J Bone Miner Res 8:1137–1148 12. Abecasis GR, Cherny SS, Cookson WO, Cardon LR 2002 Merlin: rapid anal- genes in this QTL. ysis of dense genetic maps using sparse gene flow trees. Nat Genet 30:97–101 In conclusion, QTL in chromosome 14 for hip BMD variation 13. Sham PC, Purcell S, Cherny SS, Abecasis GR 2002 Powerful regression-based quantitative-trait linkage analysis of general pedigrees. Am J Hum Genet 71: was replicated in the Southern Chinese population. In addition 238–253 to ESR2, significant associations between LTBP2 and BMD and 14. Lander E, Kruglyak L 1995 Genetic dissection of complex traits: guidelines for osteoporotic fractures were observed. The association with frac- interpreting and reporting linkage results. Nat Genet 11:241–247 15. Aerts S, Lambrechts D, Maity S, Van Loo P, Coessens B, De Smet F, Tranchevent ture was through mechanisms both dependent and independent LC, De Moor B, Marynen P, Hassan B, Carmeliet P, Moreau Y 2006 Gene of BMD. These results suggest that the LTBP2 gene might be prioritization through genomic data fusion. Nat Biotechnol 24:537–544 clinically important in fracture management. 16. Huang QY, Kung AW 2006 Genetics of osteoporosis. Mol Genet Metab 88: 295–306 17. de Bakker PI, Yelensky R, Pe’er I, Gabriel SB, Daly MJ, Altshuler D 2005 Efficiency and power in genetic association studies. Nat Genet 37:1217–1223 18. International HapMap Consortium 2005 A haplotype map of the human genome. Nature 437:1299–1320 Acknowledgments 19. Barrett JC, Fry B, Maller J, Daly MJ 2005 Haploview: analysis and visual- ization of LD and haplotype maps. Bioinformatics 21:263–265 We thank the staff of the Osteoporosis Centre, The University of Hong 20. Cordell HJ 2006 Estimation and testing of genotype and haplotype effects in Kong, Queen Mary Hospital and K. S. Lau and Benjamin Chan for their case-control studies: comparison of weighted regression and multiple impu- assistance in this project. tation procedures. Genet Epidemiol 30:259–275 21. Gabriel SB, Schaffner SF, Nguyen H, Moore JM, Roy J, Blumenstiel B, Higgins J, DeFelice M, Lochner A, Faggart M, Liu-Cordero SN, Rotimi C, Adeyemo Address all correspondence and requests for reprints to: Annie W. C. A, Cooper R, Ward R, Lander ES, Daly MJ, Altshuler D 2002 The structure Kung, M.D., Department of Medicine, The University of Hong Kong, of haplotype blocks in the human genome. Science 296:2225–2229 Pokfulam, Hong Kong, China. E-mail: [email protected]; or Pak 22. Purcell S, Daly MJ, Sham PC 2007 WHAP: haplotype-based association anal- C. Sham, Ph.D., Genome Research Centre, The University of Hong ysis. Bioinformatics 23:255–256 Kong, Pokfulam, Hong Kong, China. E-mail: [email protected]. 23. Purcell S, Cherny SS, Sham PC 2003 Genetic Power Calculator: design of This project is supported by Hong Kong Research Grant Council; linkage and association genetic mapping studies of complex traits. Bioinfor- The Bone Health Fund, HKU Foundation and Matching Grant, The matics 19:149–150 24. Lee YH, Rho YH, Choi SJ, Ji JD, Song GG 2006 Meta-analysis of genome-wide University of Hong Kong. linkage studies for bone mineral density. J Hum Genet 51:480–486 Current address for C.-L.C.: Institute for Aging Research, Hebrew 25. Goring HH, Terwilliger JD, Blangero J 2001 Large upward bias in esti- SeniorLife and Harvard Medical School, Boston, Massachusetts 02131. mation of locus-specific effects from genomewide scans. Am J Hum Genet Disclosure Statement: The authors have nothing to disclose. 69:1357–1369

26. Nzeusseu Toukap A, Galant C, Theate I, Maudoux AL, Lories RJ, Houssiau 30. Lohmueller KE, Pearce CL, Pike M, Lander ES, Hirschhorn JN 2003 Meta- FA, Lauwerys BR 2007 Identification of distinct gene expression profiles in the analysis of genetic association studies supports a contribution of common synovium of patients with systemic lupus erythematosus. Arthritis Rheum variants to susceptibility to common disease. Nat Genet 33:177–182 56:1579–1588 31. Slager SL, Huang J, Vieland VJ 2000 Effect of allelic heterogeneity on the 27. Shipley JM, Mecham RP, Maus E, Bonadio J, Rosenbloom J, McCarthy RT, power of the transmission disequilibrium test. Genet Epidemiol 18:143–156 Baumann ML, Frankfater C, Segade F, Shapiro SD 2000 Developmental ex- 32. Longmate JA 2001 Complexity and power in case-control association studies. pression of latent transforming growth factor ␤ binding protein 2 and its Am J Hum Genet 68:1229–1237 requirement early in mouse development. Mol Cell Biol 20:4879–4887 33. Cordell HJ, Clayton DG 2002 A unified stepwise regression procedure for 28. Goessler UR, Bugert P, Bieback K, Deml M, Sadick H, Hormann K, Riedel F evaluating the relative effects of polymorphisms within a gene using case/ 2005 In-vitro analysis of the expression of TGF␤-superfamily-members during control or family data: application to HLA in type 1 diabetes. Am J Hum Genet chondrogenic differentiation of mesenchymal stem cells and chondrocytes dur- 70:124–141 ing dedifferentiation in cell culture. Cell Mol Biol Lett 10:345–362 34. Moskvina V, Craddock N, Holmans P, Owen MJ, O’Donovan MC 2006 29. Appleton CT, Pitelka V, Henry J, Beier F 2007 Global analyses of gene ex- Effects of differential genotyping error rate on the type I error probability of pression in early experimental osteoarthritis. Arthritis Rheum 56:1854–1868 case-control studies. Hum Hered 61:55–64