ORIGINAL ARTICLE

Family-based association study of ROR2 polymorphismswith an array of radiographic hand bonestrength phenotypes

S. Ermakov & I. Malkin & M. Keter & E. Kobyliansky &

G. Livshits

Received: 12 December 2006 /Accepted: 10 May 2007 / Published online: 10 July 2007# International Osteoporosis Foundation and National Osteoporosis Foundation 2007

AbstractSummary For the first time the study provides evidence ofassociation of radiographic hand bone length (BL) andbone mineral density (BMD) with polymorphisms in ROR2gene that plays important role in skeletal development. Thiscontributes to better understanding of bone physiology andmay have application in clinical practice.Introduction and hypothesis Bone size and bone mineraldensity (BMD) are major determinants of bone strength.Identification of genes affecting these traits’ variability isimportant for better understanding of normal and patholog-ical bone physiology and identification of the individuals atrisk for bone fracture. This study tested the hypothesis ofassociation of radiographic hand bone length (BL) andBMD with polymorphisms in ROR2 gene that is importantin skeletal development.Methods Nineteen ROR2 SNPs were genotyped in 705individuals, belonging to 212 nuclear families. The fourtagging SNPs (tSNPs) and the pairwise haplotypes betweenadjacent tSNPs were tested for association with series ofhand BL and BMD measurements, adjusted for covariates,using family-based association tests.Results We observed significant associations with BL andBMD mean values for all 18 studied hand bones (p=0.0080,

0.0030), mean BL and BMD for proximal phalanges(p=0.0218, 0.0060) and metacarpal bones (p=0.0014,0.0004). In the latter, the association remained significantafter correction for multiple testing.Conclusions The region of the first through the secondROR2 introns is most likely to contain the functionalpolymorphism/s responsible for the observed associations.Further studies are required to identify the ROR2 functionalpolymorphism/s affecting bone size and BMD variation.

Keywords BMD . Bone size . SNP. TDT

Introduction

Osteoporosis is one of the major public health concernsworldwide, having enormous social and economic con-sequences [1]. It is characterized by reduced bone strength,which is in turn the main risk factor for fractures. Severalcategories of phenotypes are considered determinants ofbone strength, among them: bone mineral density (BMD),bone size, geometry and structure [2]. A substantialcontribution of genetic factors to variability of bonestrength related traits has been firmly established over thelast two decades [3–5]. Much effort has been put to identifythe particular chromosomal loci and genes involved in thedetermination of these traits variability [5]. However we arestill far from establishing a clear and full view of theelaborate network of genes affecting interindividual vari-ability of bone strength characteristics.

Rapid advances in understanding of bone developmentat the gene and molecular levels provide valuable informa-tion on the new potential candidate genes that might beinvolved in bone strength determination. ROR2, receptortyrosine kinase-like orphan receptor 2, seems to be one of

Osteoporos Int (2007) 18:1683–1692DOI 10.1007/s00198-007-0401-5

S. Ermakov : I. Malkin :M. Keter : E. Kobyliansky :G. Livshits (*)Human Population Biology Research Unit, Department ofAnatomy and Anthropology, Sackler Faculty of Medicine,Tel Aviv University, Ramat Aviv,Tel Aviv 69978, Israele-mail: gregl@post.tau.ac.il

G. LivshitsYoran Institute for Human Genome Research,Sackler Faculty of Medicine, Tel Aviv University,Tel Aviv, Israel

such promising candidate genes. It plays important role inskeletal development. Thus, molecular studies and analysesof mutant mice indicated that ROR2 is essential for properproliferation, maturation, and function of chondrocytes ofall developing cartilage anlagen as well as of mature growthplate [6]. Furthermore, Billiard et al. [7] reported thatROR2 is expressed in human osteoblasts and is stronglyregulated during their differentiation, which suggests itspossible regulatory role in osteoblast survival and differen-tiation. Moreover, ROR2 mutations in humans causeautosomal recessive Robinow syndrome (RS; MIM#268310) and autosomal dominant brachydactyly B1(BDB1; MIM# 113000) [8]. RS is a severe skeletaldysplasia, characterized by short stature, generalized limbbone shortening, segmental defects in spine, brachydactyly,and other abnormalities [9], the skeletal features being verysimilar to the phenotypes observed in Ror2 mutant mice[10]. Main features of BB1 are shortening/hypoplasia of thedistal phalanges, the occurrences of nail dysplasia, hypo-plasia of middle phalanges, and variable degrees of distaland proximal symphalangism [11].

Interestingly, 9q22, the region containing ROR2 gene,was reported to be linked with adult height in a wholegenome linkage study, maximum LOD score 2.74 [12].This finding was confirmed by the consequent genome-wide study with LOD score 4.34 [13].

Despite of the evidence of ROR2 involvement in normaland pathological bone growth and physiology, we are notaware of any published study of possible associationbetween ROR2 polymorphisms and bone strength traits.In light of the above, the aim of the present research was apilot study of DNA polymorphisms in ROR2 gene forpossible association with an array of bone strength relatedphenotypes, namely radiographic measurements of handbone length (BL) and hand BMD, using family-based testsof association.

Materials and methods

Sample

The study sample consisted of 743 Caucasian individuals(373 males, the mean age was 45.9 years, ranging from 18to 89 years; and 370 females, the mean age was 45.3 years,ranging from 17 to 90 years), belonging to 212 nuclearfamilies (the mean number of offspring in a family was 1.9,ranging from 1 to 4 offspring). The participants wererecruited and enrolled randomly, i.e., regardless of theoutcome of any of the measured variables. DNA sampleswere available for 705 individuals. All studied individualsbelonged to the ethnic group of Chuvashes living innumerous small villages along the Volga river in the

Chuvasha and Bashkortostan Autonomies, Russian Feder-ation. Further details regarding the present sample can befound elsewhere [14]. All subjects who agreed to partici-pate in the study signed an informed consent form, and theTel Aviv University ethics committee approved the project.

Phenotypes measured on hand bone radiographs

Standard plain radiographs of both hands were taken in thepostero-anterior position with the X-ray source positioned60 cm above, at an exposition time of 5–10 s, 100–150 mA,and 50 kv. Both hands and a bone reference control wedgewere placed on the same film-contacting plate to avoidvariation of film development. The index of mineralizationwas measured in units of optical density (1 mg/mm2), and therange of density measurement varied between 0–25 mg/mm2

(further details are given in Malkin et al. [15]).Each radiograph was converted into a digital image

using the computer-attached scanner. All measurements ondigitalized radiographs were carried out using UTHSCSAImageTool Version 3.0 for Windows software package(http://ddsdx.uthscsa.edu/dig/itdesc.html) with the scriptswritten by I. Malkin specifically for this purpose. Aftercalibration of the image, the medial and proximal phalangesand metacarpal bones on the second, third and fourthfingers on both hands (a total of 18 bones) were measuredfor BL and BMD using roentgenographic densitometrymethod [16].

SNP selection and genotyping

As our intention was to cover the whole ROR2 gene, we chosethe set of 19 SNP markers that are relatively equally spreadalong the gene, being less frequent in the extended intron 1region, and more dense in the region of exons 2 through 9.Additional SNP selection criterion was the high level ofmarker’s polymorphism (see Table 1 for markers details). Forthe sake of convenience, SNP markers are further referred toas RM1 through RM19. Additional information aboutselected SNP markers is available on-line at http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=snp.

DNA was prepared from peripheral blood lymphocytesby standard techniques, using Nucleon™ BACC GenomicDNA Extraction Kits (Amersham International plc, UK)according to the manufacturer’s protocol. Genotyping wasperformed using TaqMan® SNP Genotyping Assays (http://www.appliedbiosystems.com) on the ABI PRISM®7900HT Sequence Detection System. The protocol of thegenotyping procedure can be found at http://docs.appliedbiosystems.com/pebiodocs/04332856.pdf. The genotypingwas carried out under the supervision of Dr. M. Korner atThe Center for Genomic Technologies, The AlexanderSilberman Institute of Life Sciences, Hebrew University of

1684 Osteoporos Int (2007) 18:1683–1692

Table 1 Characteristics of the 19 ROR2 SNP markers genotyped in the studied sample

SNP ID and location

Marker Name

MAF Hetero-zygosity

HWP Location(bp)

Distance(bp)

D'

rs10992059(downsream)

RM19 0.402 91,563,3532,130 1.000

rs2230578(untranslated)

RM18 0.160 91,565,4832,472 0.617

rs4073736(intron 8)

RM17 0.411 91,567,95510,150 0.960

rs9409651(intron 5) RM16 0.387 91,578,105

10,150 0.923rs3935601(intron 4)

RM15 0.484 91,588,2559,015 0.962

rs4467996(intron 4)

RM14 0.421 91,597,27013,274 0.899

rs9409461(intron 2)

RM13 0.350 91,610,54413,293 0.970

rs4744098(intron 1)

RM12 0.479 91,623,8372,776 0.977

rs4378021(intron 1)

RM11 0.442 91,626,6133,128 0.994

rs10761130(intron 1) RM10 0.445 91,629,741

32,783 0.417rs2312732(intron 1)

RM9 0.420 91,662,52422,939 0.980

rs2312735(intron 1) RM8 0.455 91,685,463

6,483 0.950rs7048756(intron 1)

RM7 0.467 91,691,94615,938 0.906

rs4744107(intron 1)

RM6 0.448 91,707,88417,642 0.973

rs1534533(intron 1)

RM5 0.464 91,725,52622,762 0.937

rs3905385(intron 1)

RM4 0.331 91,748,28828,119 0.621

rs7038017(intron 1) RM3 0.335 91,776,407

11,551 0.993rs7048699(intron 1)

RM2 0.335 91,787,958

rs7852032(upstream)

RM1 0.335

0.456

0.269

0.441

0.440

0.486

0.476

0.424

0.504

0.484

0.475

0.531

0.508

0.550

0.498

0.516

0.479

0.453

0.451

0.454

0.050

0.742

0.298

0.398

0.877

0.539

0.211

0.619

0.877

0.650

0.073

0.654

0.058

0.861

0.651

0.582

0.458

0.510

0.675 91,800,02112,063 0.993

MAF - minor allele frequency; HWP - p values of tests for Hardy-Weinberg equilibrium; D’ - estimate of linkage disequilibrium; the table rowscontaining characteristics of the selected tSNPs further used in the association analysis are shown in bold. The chromosomal location of themarkers and the linkage disequilibrium pattern between all genotyped markers are shown in Fig. 1.

Osteoporos Int (2007) 18:1683–1692 1685

Jerusalem, Israel. Genotyped data were checked forMendelian errors by means of the MAN-6 package forWindows [17]. A total of 705 individuals were genotypedfor 19 SNP markers in ROR2 gene.

Haplotype analysis

In order to infer the ROR2 haplotype structure in thestudied Chuvashian population, we used Haploview 3.32Software (http://www.broad.mit.edu/mpg/haploview) to cal-culate the linkage disequilibrium D’ estimates [18] for allpossible pairwise combinations of the 19 studied SNPmarkers (Fig. 1). The data presented at InternationalHapMap Project Site http://www.hapmap.org/cgi-perl/gbrowse/gbrowse was used as a reference for ROR2haplotype structure. With the aim to reduce the number ofpolymorphisms tested in the consequent association analy-ses, and thus to cope with the multiple testing problem, wefurther used the Tagger feature of Haploview 3.2 Softwareto choose the minimal number of the tagging SNPs (tSNP),sufficient for the consequent exploratory association anal-ysis. With the aid of MAN-6 software, we generatedhaplotypes for the pairwise combinations of the adjacentselected tSNPs. Then we considered each pair of adjacenttSNPs as a four-allelic locus with each allele correspondingto the respective haplotype variant.

Statistical and genetic analyses

All original phenotypes measured on digitalized hand radio-graphs were standardized, and the mean values of BMD and

BL for all 18 hand bones and for each of the three bone typesseparately (6 bones in each type) were calculated giving riseto eight composite phenotypes: 1. mean BL for all 18 bones(ML18), 2. mean BL for medial phalanges (ML6M), 3. meanBL for proximal phalanges (ML6P), 4. mean BL formetacarpal bones (ML6C), 5. mean BMD for all 18 bones(MB18), 6. mean BMD for medial phalanges (MB6M), 7.mean BMD for proximal phalanges (MB6P), and 8. meanBMD for metacarpal bones (MB6C).

Variability of each studied phenotype was adjusted forthe effect of potential confounding factors, namely sex andage (including age2 and age3). Since bone size is arecognized additional confounder for BMD [19–20], wealso analyzed the effect of adult height on the studied BMDindices. Adult height was chosen as a proxy for bone sizedue to its universal availability and use.

Adult height was measured in all individuals by the sameinvestigator, using a portable anthropometer (1 mm accu-racy). Participants were asked to hold their breath andmaintain a fully erect position during the measurement.

Accordingly, all studied bone phenotypes were adjustedfor significant confounding effects, implementing the inter-val piecewise aging models as described in Malkin et al.[21]. To test for putative genetic effects, we estimatedfamilial correlations (rsp - for spouses, rpo - for parent-offspring, and rsib - for siblings) and calculated themaximum heritability estimate for each adjusted trait ash2 ¼ rsib þ rpo

� �1þ rsp� ��

1þ rsp þ 2rsprpo� �

[22]. In thenext stage we conducted the joint quantitative trait-DNAmarker association analysis by means of transmission disequi-librium tests (TDTs). Specifically, we used the family-based

Fig. 1 The scheme of the loca-tions of the studied SNPs rela-tive to ROR2 exon-intronstructure, and the pattern of LD.tSNPs are encircled, taggedSNPs are underlined, D’ esti-mates between adjacent tSNPsare marked by squares, SNPgroups in high LD are outlined.Diamonds indicate position oftSNPs with respect to the phys-ical map

1686 Osteoporos Int (2007) 18:1683–1692

association test (FBAT; [23]), the extreme offspring design ttest (EOT; [24]), and two different versions of the orthogonaltest [25], with adjustment for the parent phenotypes (OTP)and without it (OT). The statistical features of these tests havebeen recently discussed elsewhere [26].

In order to combine the results of the four separateassociation tests, we formulated a multiple comparisonprocedure and computed the combined p-values (CPV), theprobability of erroneous rejection of the general nullhypothesis of no linkage disequilibrium, which unites allcertain null hypotheses. The full description of the procedurewas given by us recently [27]. The computed CPVs werefurther tested for significance using the false discovery rate(FDR) approach proposed by Benjamini and Yekutieli [28]for multiple testing under dependency. Each trait-SNP pairproduced only one separate null-hypothesis. The compositemarker generated from two diallelic SNPs generally has fouralleles, represented by haplotypes. Accordingly, the numberof possible dichotomous schemes is four, i.e., each alleleagainst three others. We conducted association tests onlywith haplotypes which frequency exceeded 0.15, otherwisethe alleles were considered not sufficiently informative andwere excluded from subsequent analyses. Additionally, foreach composite marker we tested one 2-to-2 factorization,which cannot be reduced to the testing of initial SNPs.Further details are presented in the Results section.

Results

Adjustment of the studied phenotypes for confoundingfactors and analysis of familial correlations

The results of multiple regression analyses, familialcorrelations and maximum heritability estimates for the

studied phenotypes are presented in Table 2. Sex and agewere uniformly significant predictors of all studied pheno-types (p<0.001). For BL traits, the corresponding multipledetermination coefficients of the multiple regression modelsranged from 0.28 for ML6M to 0.36 for ML18. For theBMD phenotypes, adult height had a statistically significanteffect on variation of all four indices. The multipledetermination coefficients of the models that included sex,age, and adult height as confounders, ranged from 0.31 forMB6C to 0.46 for MB6P. For all adjusted phenotypes thespouse correlations were non-significant (except forMB6C) and all were considerably lower than the respectiveparent-offspring and sibling correlations. This patternclearly suggested the presence of the genetic factors effecton the studied traits variability. This was further confirmedby substantial maximum heritability estimates for alladjusted traits (Table 2). Thus, the lowest heritabilityestimate was 0.47 for MB6P, while the highest wasobserved for ML6P (0.96).

Marker properties and tSNPs selection

The chromosomal position, minor allele frequencies andother relevant details of the 19 originally genotyped SNPmarkers are shown in Table 1. The level of heterozygosityin the sample of parents varied from 0.27 to 0.55. Note thatfor two markers, RM7 and RM19, p values wereapproaching the significance level of 0.05 with respect toHardy-Weinberg equilibrium null hypothesis. This may bean indication of the selection acting on these specific SNPsor some other polymorphisms in LD with the studied ones,or may simply be a result of multiple comparison.

Minor allele frequencies of the 19 studied SNPs inChuvasha population showed clear similarity to the featuresof the same SNPs genotyped in Utah residents with ancestry

Table 2 Summaries of multiple regression analyses and familial aggregation of the hand bone phenotypes adjusted for significant covariates inthe studied sample

Primary phenotype Multiple regression analysis Familial correlationsa Maximum heritability estimatea

R2 F p value Sp PO Sib

ML18b 0.36 81.74 <0.001 0.00 0.42 0.48 0.91ML6Mb 0.28 56.86 <0.001 −0.13 0.36 0.42 0.87ML6Pb 0.31 165.18 <0.001 0.03 0.47 0.52 0.96ML6Cb 0.35 194.10 <0.001 0.07 0.43 0.49 0.87MB18c 0.45 84.78 <0.001 0.14 0.23 0.30 0.50MB6Mc 0.40 79.85 <0.001 0.15 0.27 0.28 0.52MB6Pc 0.46 88.62 <0.001 0.15 0.20 0.29 0.47MB6Cc 0.31 46.74 <0.001 0.17 0.29 0.33 0.57

a - the values are estimated after adjustment for covariates; significant correlations are marked in bold.b - covariates were age and sex.c - covariates were age, sex, and adult height. Sp - spouse correlation; PO - parent-offspring correlation; Sib - sibling correlation. ML18 - mean

bone length for all 18 hand bones, ML6M - mean bone length for six medial phalange bones, ML6P - mean bone length for 6 proximalphalange bones, ML6C - mean bone length for six metacarpal bones, MB18 - mean BMD for all 18 hand bones, MB6M - mean BMD for sixmedial phalange bones, MB6P - mean BMD for six proximal phalange bones, MB6C - mean BMD for six metacarpal bones.

Osteoporos Int (2007) 18:1683–1692 1687

from northern and western Europe (CEU population;International HapMap Project data). Additionally, ROR2haplotype structure as inferred by the pattern of D’ estimatesfor the studied markers in Chuvasha population, corres-ponded well to that for CEU population. SNP positionsrelative to ROR2 exon-intron structure and all pairwisedisequilibrium coefficients D’ are presented in Fig. 1,composed using Haploview 3.32 software output image.

As it follows from the D’ pattern, RM18 and RM19(pairwise D’=1) belong to the haploblock, most of whichresides outside the ROR2 gene. Additionally, RM18exhibited low level of heterozygosity, i.e., 0.27 (minorallele frequency=0.16), while RM19 was likely to deviatefrom Hardy-Weinberg equilibrium (p=0.05). This situationcan cause inconsistent and unreliable results of associationanalyses. We therefore excluded these two markers fromthe subsequent analysis in the present study.

With the aid of Haploview’s Tagger we have selectedfour tSNPs: RM3 (tagged RM1 and RM2), RM8 (taggedRM5, RM6, RM7, RM9), RM10 (tagged RM11), andRM16 (tagged RM17). Collectively, these four tSNPstagged eight more SNPs with r2 equal to 0.934. In additionto this, RM8 exhibited high LD with RM4, RM10 was instrong LD with RM12 and RM13, RM16 showed highdisequilibrium with RM13, RM14, and RM15 (D’>0.8).tSNPs, tagged SNPs and the SNP groups exhibiting highLD are marked in Fig. 1. Selected four tSNPs and thecorresponding three composite markers derived from pair-wise combinations of the adjacent tSNPs (RH3-8, RH8-10,and RH10-16) were used in the association analyses.

Family-based association study

Four different TDT-like tests were used to test for theassociation between the specific SNP marker and compositemarker alleles and the studied phenotypes adjusted forcovariates. Additionally, for each trait-marker pair thecombined probability of the erroneous general null hypo-thesis rejection (CPV) was estimated.

Markers RM3 and RM16, as well as RH3-8 T|G andRH8-10 A|A haplotypes, and RH10-16 2-to-2 factorization(united allele A|C or C|T to all the rest) did not show anysignificant association with any of the studied traits. Themost significant and consistent results of the associationtests are presented in Tables 3 and 4, for BL and BMDgroups of phenotypes, respectively.

BL traits generally showed significant association withRM8 and RH3-8 and RH8-10 haplotypes (Table 3, Fig. 2).For RM8, the significant CPVs ranged from 0.0316 forML6P to 0.0016 for ML6C. ML6M didn’t exhibit anynoticeable associations with any of the markers. The rest ofthe phenotypes in BL group showed significant associa-tions. The respective CPVs were lowest for ML6C with

RH8-10 G|A haplotype (CPV=0.0014). ML18 also showedcomparable level of significance, i.e., 0.0080 for RH3-8 C|Ahaplotype. For ML6P, only a modest level of significance(0.0218-0.0316) was observed.

In the analyses of hand BMD phenotypes, we observedsignificant association between MB6C and RM-10 (CPV=0.0121). In testing haplotypes, except for MB6M, all studiedBMD traits were significantly associated with RH3-8 andRH8-10 haplotypes (Table 4, Fig. 2). The most significantCPVs were observed for MB6C, namely 0.0006 withRH3-8 C|G haplotype and 0.0004 with RH8-10 A|C haplotype.

We presented graphically the effects of RH8-10 geno-types on ML6C and MB6C mean values in offspring(Fig. 2). Homozygosity for RH8-10 G|A haplotype in thecase of ML6C and homozygosity for RH8-10 A|Chaplotype in the case of MB6C led to higher genotype-specific mean values of the corresponding phenotypesadjusted for significant covariates.

Correction for multiple testing

In order to correct our TDT results for multiple testing, weallowed for the 0.05 FDR level. We tested four tSNPmarkers and haplotypes of the three adjacent tSNP markerpairs. For RH3-8 and RH8-10 we tested three haplotypes(haplotype frequency>0.15), while only two RH10-16haplotypes were tested. The other two haplotypes hadfrequency below our threshold of 0.15. Additionally, for allthree composite markers studied, we tested 2-to-2 dichot-omous scheme. Correspondingly, the total number of testedSNPs and marker haplotypes was 15. Given eight pheno-types studied (Tables 3 and 4), the total number of thegeneral null-hypotheses tested was 120. After application ofthe FDR procedure, four general null-hypotheses can berejected at FDR=0.05 and the significance threshold of0.0017. By these criteria, at least association of themetacarpal bones length with RM8 and RH8-10, andmetacarpal bones BMD with RH3-8 and RH8-10 arereliably established.

Following reviewers’ suggestion we additionally testedadult height and upper limb length for association withROR2 polymorphisms. Both phenotypes were significantlyassociated with RM8. The corresponding CPVs were0.0459 and 0.0141 for adult height and upper limb length,respectively. Adult height also exhibited significant associ-ations with RH3-8 haplotypes with CPVs ranging from0.0258 (C|G haplotype) to 0.0400 (C|A haplotype). Forupper limb length, CPVs of association tests with RH3-8haplotypes did not reach the formal significance thresholdand were 0.0607 for C|G haplotype and 0.0717 for C|Ahaplotype. Significant CPVs were observed between upperlimb length and RH8-10 haplotypes, i.e., 0.0280 (A|Chaplotype) and 0.0080 (G|A haplotype).

1688 Osteoporos Int (2007) 18:1683–1692

Discussion

The present study provides evidence of a significantassociation between bone strength related traits and poly-morphisms in ROR2 gene.

We genotyped 19 SNP markers, covering the wholeROR2 gene. The properties of these SNPs and the inferredROR2 LD structure in our population were similar to thedata available for Utah residents with ancestry fromnorthern and western Europe (CEU population; Interna-tional HapMap Project). For the purpose of this study, fourtSNPs, that tagged eight more SNPs with r2 equal to 0.934,

were chosen. Additionally, the chosen tSNPs were in strongLD (D’>0.8) with the other six genotyped SNPs (Fig. 1).Thus, the selected tSNPs and the resultant three compositemarkers were considered sufficient for the consequentexploratory association analyses.

As the studied sample included complex multigenera-tional pedigrees, the age range of the examined individualswas considerable (17–90). If not accounted for, thiscondition could be a source of significant bias. To avoidthis, prior to the genetic analyses, all studied phenotypeswere carefully adjusted for the effect of age using advancedmethods of polynomial and piecewise aging model-fitting

Table 3 The results of the association analysis between bone length phenotypes and the selected ROR2 polymorphisms (p-values are shown inthe table)

Marker RM8 RH3-8 RH8-10

Allele A C|A C|G C|A T|G A|C G|ATraita

Frequency 0.55 0.46 0.20 0.71 0.32 0.32EOT 0.0325 0.0099 0.0140 0.0440 0.0479OTP 0.0198 0.0118 0.0535OT 0.0028 0.0089 0.0151 0.0569 0.0080FBAT 0.0115 0.0091 0.0175 0.0631 0.0099

ML18

CPV 0.0087 0.0080 0.0193 0.0688 0.0230EOT 0.0458OTP 0.0652OT 0.0794 0.0451FBAT 0.0451

ML6M

CPV 0.0515

EOT 0.0215 0.0494 0.0225OTP 0.0662OT 0.0090 0.0115 0.0110 0.0323 0.0217FBAT 0.0345 0.0167 0.0193 0.0513 0.0266

ML6P

CPV 0.0316 0.0218 0.0311 0.0502 0.0588

EOT 0.0137 0.0137 0.0186 0.0213 0.0784 0.0049OTP 0.0075 0.0061 0.0740 0.0483 0.0335OT 0.0004 0.0028 0.0385 0.0486 0.0006FBAT 0.0020 0.0021 0.0335 0.0380 0.0010

ML6C

CPV 0.0016 0.0038 0.0328 0.0533 0.0544 0.0014a - the traits were adjusted for sex and age effects. P values, approaching significance level of 0.05 are italic, p values below the 0.05 significance

threshold are bold, and CPV values significant at FDR=0.05 are bold and highlighted with gray. ML18 - mean bone length for all 18 handbones, ML6M - mean bone length for 6 medial phalange bones, ML6P - mean bone length for 6 proximal phalange bones, ML6C - mean bonelength for 6 metacarpal bones.

Osteoporos Int (2007) 18:1683–1692 1689

[15]. It could be mentioned that since only about 1.5% ofthe subjects were younger than 20 years, the data were wellfitted by the corresponding model-fitting curves.

The phenotypes studied represented the two majorcategories of bone strength determinants: bone size, ormore specifically, BL (ML18, ML6M, ML6P, and ML6C)and BMD (MB18, MB6M, MB6P, and MB6C). Inagreement with the published data [3–5], the substantialgenetic effect was observed for all studied bone size andBMD traits (Table 2). For the adjusted phenotypesbelonging to the BL group of traits the maximumheritability estimates ranged from 0.87 to 0.96, while thosefor BMD ranged from 0.47 to 0.57. We observed that traits

belonging to both these categories showed significantassociations with several studied ROR2 polymorphisms(Tables 3 and 4, Fig. 2). Moreover, associations betweenROR2 polymorphisms and mean BL and BMD for metacar-pal bones remained significant even after controlling forFDR at 0.05 level (CPV<0.0017). Additional indirectconfirmation of the observed associations with BL pheno-types comes from significant association results obtained foradult height and upper limb length (see the Results section).Notably, the latter findings on adult height are also inagreement with the results of linkage studies [12, 13].

A common trend of the observed associations wasapparent. BL and BMD indices for medial phalanges didn’t

Table 4 The results of the association analysis between BMD phenotypes and the selected ROR2 polymorphisms (p-values are shown in thetable)

Marker RM10 RH3-8 RH8-10

Allele A C|G C|A T|G A|C A|A G|CTraita

Frequency 0.55 0.21 0.71 0.32 0.36EOT 0.0763 0.0014 0.0030 0.0096 0.0672OTP 0.0103 0.0295 0.0027 0.0040OT 0.0041 0.0115 0.0057 0.0340FBAT 0.0765 0.0083 0.0202 0.0030 0.0355

MB18

CPV 0.0030 0.0072 0.0036 0.0124EOT 0.0459OTP 0.0518 0.0415OT 0.0651

FBAT

MB6M

CPV 0.0674

EOT 0.0058 0.0075 0.0145OTP 0.0229 0.0034 0.0131OT 0.0139 0.0341 0.0125 0.0597

FBAT 0.0199 0.0526 0.0101 0.0696

MB6P

CPV 0.0095 0.0219 0.0060 0.0347EOT 0.0123 0.0008 0.0009 0.0009 0.0224OTP 0.0751 0.0028 0.0183 0.0019 0.0028OT 0.0077 0.0007 0.0084 0.0006 0.0181FBAT 0.0068 0.0016 0.0139 0.0001 0.0278

MB6C

CPV 0.0121 0.0006 0.0026 0.0004 0.0072a - the traits were adjusted for sex, age and adult height effects. P values, approaching significance level of 0.05 are italic, p values below the 0.05

significance threshold are bold, and CPV values significant at FDR=0.05 are bold and highlighted with gray. MB18 - mean BMD for all18 hand bones, MB6M - mean BMD for 6 medial phalange bones, MB6P - mean BMD for 6 proximal phalange bones, MB6C - mean BMDfor 6 metacarpal bones.

1690 Osteoporos Int (2007) 18:1683–1692

show any noticeable association with the studied poly-morphisms. The highest levels of significance were reachedin the association tests of BL and BMD traits in metacarpalbones. Thus, for example the association between RH8-10G|A haplotype and ML6C was significant at CPV=0.0014,while MB6C was associated with RH8-10 A|C allele atCPV=0.0006. Homozygosity for the RH8-10 G|A and A|Chaplotypes corresponded to higher genotype-specific meanvalues of the adjusted ML6C and MB6C, respectively(Fig. 2). ML6P and MB6P were also significantly associ-ated with several studied polymorphisms, the levels ofsignificance, however, were generally considerably lowerwhen compared to the corresponding ones observed inmetacarpal bones. The associations observed for ML18 andMB18 reached intermediate levels of significance, comparedto BMD and BL in proximal phalanges and metacarpalbones. This trend implies that the effect of ROR2 poly-morphisms on bone strength phenotypes may be stronger orbetter detected in metacarpal bones, and may get weaker inproximal to distal direction. One possible explanation of thismay be the gradient way of ROR2 distribution and action,which is common for developmental factors.

Notably, RM3, which is in high LD with promoter andexon1 regions, as well as RM16, being in high LD with theinterval where the third and more distal exons reside, didn’tshow any significant effect on the studied phenotypes. Onthe other hand, RM8 and haplotypes of two pairs of tSNPs,namely RH3-8 and RH8-10, were significantly associatedwith both, BL and BMD characteristics of hand bones.Additionally, we observed significant associations of adult

height and upper limb length with the above polymor-phisms. This indicates that the potential polymorphism/sresponsible for the observed association is/are likely toreside in the region, containing the first intron, the secondexon, and the second intron of the ROR2 gene. It is alsopossible that there are two different ROR2 functionalpolymorphisms, independently affecting BL and BMDcharacteristics, which is suggested by somewhat differentpattern of associations observed for these two phenotypecategories. This difference is particularly noticeable withrespect to RM8. BL traits showed association with thisSNP, while no significant association with this marker wasdetected for any of the studied BMD characteristics. Theother examples are RH3-8 C|A haplotype and RH8-10 2-to-2factorization (united allele A|A or G|C to all the rest). Herewe observed that the former was associated with BMDphenotypes, but not with BL traits, while RH8-10 2-to-2factorization showed association specifically with BL, butnot with BMD indices.

As any research project the present study has itsadvantages and limitations. The main limitation of ourresearch is that the studied bone phenotypes have no directclinical relevance, and the present sample does not includeindividuals suffering from osteoporotic fractures. However,we implemented highly precise method of bone phenotypesassessment, and obtained reliable new data contributing toour understanding of the genetic mechanisms affectingnormal bone size and mass variation, which in turn laysfoundation for further clinically oriented studies. It is ofimportance to mention that we examined ethnically andgenetically homogeneous sample, appropriate for theassociation studies, and the assessed bone phenotypes werenot modified by hormone replacement therapy, medicine orfood supplements. On the other hand, we limited ourselvesto association study of only four chosen tSNPs in ROR2gene. This was done in attempt to cope with multipletesting problem, that becomes a more and more seriousissue as the number of studied markers and/or traits grows.The four tSNPs and the derived haplotypes explainsubstantial portion of the ROR2 genetic variation. Yet, thewhole gene is some 387 Kbp long, which leaves plenty ofspace for further in-depth studies.

While the structure and the functions of the ROR2 geneproduct have been extensively studied [6, 7, 10], theresearch of ROR2 gene organization, e.g., the presenceand location of potential regulatory elements, has been verylimited. This situation makes it difficult to offer seriousevidence-based speculations on the actual source of theobserved associations. Our findings pose the new questionsand provide a motivation and a possible new direction forthe future research of the ROR2 gene structure and themechanisms of the ROR2 polymorphisms effect on bonestrength related traits.

Fig. 2 Mean metacarpal bone length and BMD genotype-specificmeans in offspring population. The corresponding standard errors ofmeans are also shown. The data were adjusted for significantcovariates

Osteoporos Int (2007) 18:1683–1692 1691

In conclusion, this pilot study provides convincingevidence of association between hand BL and BMDphenotypes normal variability and ROR2 genetic poly-morphisms. These results were further confirmed bysignificant associations of adult height and upper limblength with several studied ROR2 polymorphisms. Associ-ations with BL and BMD in metacarpal bones weresignificant even after controlling for FDR at 0.05 level(CPV<0.0017). The results of this study point to the regionthat is most likely to contain the functional polymorphism/sresponsible for the observed association signal, namely theregion of the first through the second ROR2 introns.Further studies are required to confirm our findings inother populations and in different skeletal sites, as well asto further explore ROR2 gene structure, and to eventuallyidentify the functional polymorphism/s affecting bone sizeand BMD variation.

Acknowledgements The study was performed in partial fulfillmentof the doctoral degree requirements of Sergey Ermakov. We thankDr. Svetlana Trofimov (Department of Anatomy and Anthropology,Sackler Faculty of Medicine, Tel Aviv University) for help in DNApreparation, Dr. Mira Korner and her staff (The Center for GenomicTechnologies, The Institute of Life Sciences, The Hebrew University ofJerusalem, Israel) for genotyping of the samples, and Galit Schwartz(Applied Biosystems, Agentek-Israel) for her assistance. This study wassupported by Israel National Science Foundation (Grant No. 1042/04).

References

1. Lane NE (2006) Epidemiology, etiology, and diagnosis ofosteoporosis. Am J Obstet Gynecol 194:3–11

2. Eisman JA (1999) Genetics of osteoporosis. Endocr Rev 20:788–8043. Peacock M, Turner CH, Econs MJ et al (2002) Genetics of

osteoporosis. Endocr Rev 23:303–3264. Livshits G (2005) Genetic epidemiology of skeletal system aging

in apparently healthy human population. Mech Ageing Dev126:269–279

5. Uitterlinden AG, van Meurs JB, Rivadeneira F et al (2006)Identifying genetic risk factors for osteoporosis. J MusculoskeletNeuronal Interact 6:16–26

6. DeChiara TM, Kimble RB, Poueymirou WT et al (2000) Ror2,encoding a receptor-like tyrosine kinase, is required for cartilageand growth plate development. Nat Genet 24:271–274

7. Billiard J, Way DS, Seestaller-Wehr LM et al (2005) The orphanreceptor tyrosine kinase Ror2 modulates canonical Wnt signalingin osteoblastic cells. Mol Endocrinol 19:90–101

8. Afzal AR, Jeffery S (2003) One gene, two phenotypes: ROR2mutations in autosomal recessive Robinow syndrome and autosomaldominant brachydactyly type B. Hum Mutat 22:1–11

9. Patton MA, Afzal AR (2002) Robinow syndrome. J Med Genet39:305–310

10. Takeuchi S, Takeda K, Oishi I et al (2000) Mouse Ror2 receptortyrosine kinase is required for the heart development and limbformation. Genes Cells 5:71–78

11. Schwabe GC, Tinschert S, Buschow C et al (2000) Distinctmutations in the receptor tyrosine kinase gene ROR2 causebrachydactyly type B. Am J Hum Genet 67:822–831

12. Liu YZ, Xu FH, Shen H et al (2004) Genetic dissection of humanstature in a large sample of multiplex pedigrees. Ann Hum Genet68:472–488

13. Liu YZ, Xiao P, Guo YF et al (2006) Genetic linkage of humanheight is confirmed to 9q22 and Xq24. Hum Genet 119:295–304

14. Livshits G, Karasik D, Kobyliansky E (2002) Complex segrega-tion analysis of the radiographic phalanges bone mineral densityand their age related changes. J Bone Miner Res 17:1–10

15. Malkin I, Bigman G, Matias R et al (2006) Age-related changes ofbone strength phenotypes: observational follow-up study of handbone mineral density. Osteoporos Int 12; Epub ahead of print

16. Bouxsein ML, Michaeli DA, Plass DB et al (1997) Precision andaccuracy of computed digital absorptiometry for assessment ofbone density of the hand. Osteoporos Int 7:444–449

17. Malkin I, Ginsburg E (2003) Program package for Mendeliananalysis of pedigree data (MAN, Version 6). Department ofAnatomy and Anthropology, Sackler Faculty of Medicine, TelAviv University; Technical Report

18. Lewontin R (1974) The genetic basis of evolutionary change.Columbia University Press, New York

19. Melton LJ 3rd, Khosla S, Achenbach SJ et al (2000) Effects ofbody size and skeletal site on the estimated prevalence ofosteoporosis in women and men. Osteoporos Int 11:977–983

20. Schoenau E, Saggese G, Peter F et al (2004) From bone biology tobone analysis. Horm Res 61:257–269

21. Malkin I, Karasik D, Livshits G et al (2002) Modeling of agerelated bone loss using cross-sectional data. Ann Hum Biol29:256–270

22. Rice T, Després JP, Daw EW et al (1997) Familial resemblance forabdominal visceral fat: the HERITAGE family study. Int J ObesRelat Metab Disord 21:1024–1031

23. Horvath S, Xu X, Laird NM (2001) The family based associationtest method: strategies for studying general genotype-phenotypeassociations. Eur J Hum Genet 9:301–306

24. Malkin I, Ginsburg E, Elston RC (2002) Increase in power oftransmission disequilibrium tests for quantitative traits. GenetEpidemiol 23:234–244

25. Abecasis GR, Cardon LR, Cookson WS (2000) A general test ofassociation for quantitative traits in nuclear families. Am J HumGenet 66:279–292

26. Malkin I, Dahm S, Suk A et al (2005) Association of ANKH genepolymorphisms with radiographic hand bone size and geometry ina Chuvasha population. Bone 36:365–373

27. Malkin I, Ermakov S, Kobyliansky E et al (2006) Strongassociation between polymorphisms in ANKH locus and skeletalsize traits. Hum Genet 120:42–51

28. Benjamini Y, Yekutieli D (2001) The control of the falsediscovery rate in multiple testing under dependency. Ann Stat29:1165–1188

1692 Osteoporos Int (2007) 18:1683–1692