Osteoporos Int (2003) 14: 650–658 DOI 10.1007/s00198-003-1416-1

O R I GI N A L A R T IC L E

Gene polymorphisms, bone mineral density and bone mineral content in young children: the Iowa bone development study Marcia C. Willing Æ James C. Torner Æ Trudy L. Burns Kathleen F. Janz Æ Teresa Marshall Æ Julie Gilmore Sachi P. Deschenes Æ John J. Warren Æ Steven M. Levy

Received: 2 September 2002 / Accepted: 20 March 2003 / Published online: 22 July 2003
International Osteoporosis Foundation and National Osteoporosis Foundation 2003

Abstract We examined the association of candidate gene polymorphisms with bone mineral density (BMD) and bone mineral content (BMC) in a cohort of 428 healthy non-Hispanic white children participating in the Iowa Bone Development Study, a longitudinal study of determinants of bone accrual in childhood. BMD and BMC measurements of the hip, spine and whole body were made using a Hologic 2000 Plus densitometer in 228 girls and 200 boys ages 4.5–6.5 years. Genotypes at 14 loci representing eight candidate genes [type I collagen genes (COL1A1 and COL1A2), osteocalcin, osteonectin, osteopontin, vitamin D receptor (VDR), estrogen receptor (ER), androgen receptor (AR)] were determined. Gender-specific and gender-combined prediction models for bone measures that included age, weight, height (and gender) were developed using multiple linear regression analysis. COL1A2 and osteocalcin genotypes were identified as having the strongest and most consistent association with BMD/BMC measures. Osteonectin,

osteopontin and VDR translation initiation site polymorphisms were associated with some individual bone measures, but none of the associations was as consistent as those identified for the COL1A2 and osteocalcin genes. No association was identified with COL1A1 (RsaI and Sp1), VDR (BsmI) and ER polymorphisms (PvuII, XbaI, TA) and BMD/BMC. However, we identified significant gene-by-gene interaction effects involving the ER and both VDR and osteocalcin, which were associated with BMD/BMC. Our data suggest that genetic variation at multiple genetic loci is important in bone accrual in children. Moreover, the combination of genotypes as several loci may be as important as a single genotype for determining BMD and BMC. Keywords Bone Æ Bone mineral density Æ Children Æ COL1A2 Æ Genetic polymorphisms Æ Osteocalcin

Introduction M.C. Willing (&) Æ S.P. Deschenes Department of Pediatrics, Roy J. and Lucille A. Carver College of Medicine, University of Iowa, Iowa City, IA52242, USA E-mail: [email protected] J.C. Torner Department of Epidemiology, College of Public Health, University of Iowa, Iowa City, IA52242, USA T.L. Burns Department of Biostatistics, College of Public Health, University of Iowa, Iowa City, IA52242, USA K.F. Janz Department of Health, Leisure and Sports Studies, College of Liberal Arts, University of Iowa, Iowa City, IA52242, USA T. Marshall Æ J. Gilmore Æ S.P. Deschenes Æ J.J. Warren S.M. Levy Department of Preventive and Community Dentistry, College of Dentistry, University of Iowa, Iowa City, IA52242, USA M.C. Willing Department of Pediatrics, 2609 JCP, Division of Medical Genetics, University of Iowa, Iowa City, IA52242, USA

Many studies have documented familial aggregation of bone mineral density (BMD) in parents and their children [1,2]. Numerous family and twin studies have demonstrated a strong genetic component in the determination of BMD levels in both males and females [3]. Little is known, however, about the individual genes that influence bone accrual in childhood. The Iowa Bone Development Study is a longitudinal study characterizing biological variation in bone accrual in healthy children. We are collecting information about the contribution of genetic and life-style factors, including diet and physical activity, to BMD and bone mineral content (BMC). The purpose of this paper is to describe the association of BMD and BMC of the hip, spine and whole body and polymorphisms for eight candidate gene loci [vitamin D receptor (VDR), the type I collagen genes (COL1A1 and COL1A2), osteocalcin, osteonectin, osteopontin, estrogen receptor (ER) and the androgen receptor (AR)] in 428 non-Hispanic white children ages 4.5–6.5 years.

651 body, lumbar spine, and the proximal femur, which included the femoral neck, Ward’s triangle and trochanter. BMD is expressed as g/cm2; BMC is expressed in g.

Materials and methods Study population Children were recruited from the Iowa Fluoride Study, a longitudinal study of a birth cohort of 890 families that is examining the relationships between fluoride exposures and the occurrence of dental caries and fluorosis [4]. The original Iowa Fluoride Study recruited mothers with newborns from eight Iowa hospital postpartum wards from March 1992 to February 1995. In addition to serial dental evaluations, detailed dietary questionnaires and 3-day food and beverage diaries were completed on children at ages 6 weeks, 3, 6, 9 and 12 months, and every 4–6 months thereafter. All children in the Iowa Fluoride Study were invited to participate in the Iowa Bone Development Study. A total of 470 children agreed to participate. The cohort is 53% female and 96% white, which is similar to the composition of the Iowa Fluoride Study cohort. Dietary data continue to be collected on these children. Bone mineral density (BMD) and bone mineral content (BMC) are being measured at 2- to 3-year intervals, beginning at age 4.5 years, with the intent to follow these children through adolescence. Physical activity is being assessed longitudinally by questionnaire and accelerometry. All components of the study were approved by the Human Subjects Institutional Review Board at The University of Iowa.

Bone measures Bone mineral density (BMD) and bone mineral content (BMC) measurements were made in the General Clinical Research Center at The University of Iowa using a Hologic 2000 Plus densitometer (Hologic, Watham, Mass., USA) with fan-beam geometry and a multiple detector array. The densitometer was calibrated daily with the same phantom. The coefficient of variation for dual energy X-ray absorptiometry (DXA) was less than 1%. Each child was scanned between 4.5 and 6.5 years of age by one of two certified technicians. BMD/BMC measurements were determined for whole

Anthropomorphic measurements Anthropomorphic measurements were obtained for each child at the time the bone measures were made. Height was measured using a stadiometer, and recorded in tenths of centimeters. Weight was measured using a standard physician scale, and recorded in tenths of kilograms. The stadiometer and scale were routinely monitored for accuracy and precision.

Genotyping Genomic DNA isolated from either lymphocytes (87%) or buccal epithelial cells (13%) served as the template for PCR-based genotyping. Genotypes at 14 marker loci representing eight candidate genes were determined for each participant. Primer sequences for each polymorphic marker are shown in Table 1. References for primer sequences, amplification conditions and allele identification have been previously published [5,6,7]. Methods for allele standardization and comparison with genotypes reported in other laboratories are discussed in [5] and [6]. For restriction fragment length polymorphisms (RFLPs), genotypes were assigned as (+/+), (+/)) and ()/)), where (+) denotes the presence of the enzyme recognition site and ()) denotes the absence of the site. The VDR BsmI (+/+) and ()/)) genotypes correspond to bb and BB, respectively. The VDR translation initiation site polymorphism (ATG) corresponds to the (+/+) or MM genotype and (ACG) corresponds to the ()/)) or mm genotype. For the COL1A1 Sp1 (C/T) polymorphism in intron 1, the (+/+) and ()/)) genotypes correspond to SS (or sequence GG) and ss (or TT). For the osteocalcin C/T promoter polymorphism, the ()/)) genotype corresponds to the TT genotype, while the genotype denoted (+/+) corresponds to CC. For the variable nucleotide tandem repeats, the genotype was assigned according to the number of repeats.

Table 1 Primers for genotyping. References for primer sequences have been published [5,6,7] Gene name

Polymorphism

Primer sequences

Vitamin D receptor

BsmI

5’-GATTCTGAGGAACTAGATAAGC-3’ 5’-AACCAGCGGGAAGAGGTCAAGGG-3’ 5’-AGCTGGCCCTGGCACTGACTCTGCTCT-3’ 5’-GCAGCCTCCACAGGTCATAGC-3’ 5’-TAT CCA GGG TTA TGT GGC AA-3’ 5’-AAA ATG ACA AAA TGA AAT TAG CTG G-3’ 5’-GACGCATGATATACTTCACC-3’ 5’-GCAGAATCAAATATCCAGATG-3’ 5’-TCC AGA ATC TGT TCC AGA GCG TGC-3’ 5’-GCT GTG AAG GTT GCT GTT CCT CAT-3’ 5’-CAA GAG CAT TCT CTT AAC TGA CCT-3’ 5’-TCC TGG ACT GGA TCC CAG ATT GGG-3’ 5’-TAA CTT CTG GAC TAT TTG CGG ACT TTT TGG-3’ 5’-CAA CCT CAG CCC ATT GGC GCT G-3’ 5’-CTG CTG GAA GTC GTG GTG AT-3’ 5’-CAC CAG GGA AAC CAG TCA TA-3’ 5’-GGA TCC AAA GTC ACA CAT CTA GAG-3’ 5’-CAA TCT ATA TTC TTA TCC TG-3’ 5’-GAT GGT GCG GTG GTT GAT-3’ 5’-ACA GGA AAA AAG AGC CAG CA-3’ 5’-GGG TCT CTG AGG AAG AGT GAG-3’ 5’-CAT GGT GCG GGC GGG TCT AG-3’ 5’-TAT GTT CAC AAG AGG GTG TC-3’ 5’-ATC TCG CCA CTG TAC TCT AC-3’ 5’-TCA GGT GAT GCT TCT GCC TC-3’ 5’-TGA GCC CAG GAG TTT AAG GC-3’

TI Estrogen receptor-a

Pvu II and Xba I 5’ TA repeat

Androgen receptor

(ACG)n

COL1A1

RsaI Sp1

COL1A2

RsaI INT12 VNTR

Osteocalcin

D1S3737 [CA repeat] C/T promoter

Osteonectin

Intragenic CA repeat

Osteopontin

Intragenic CA repeat

652 Statistical methods All statistical analyses were conducted using procedures from the Statistical Analysis System (SAS, version 8). A P-value <0.05 for any single analysis was considered statistically significant, and P<0.10 was considered suggestive. Descriptive statistics were calculated as means and standard deviations (SD) for continuous variables and as proportions for categorical variables. Each RFLP was tested for Hardy-Weinberg equilibrium using a chi-square goodness-of-fit test. The VNTRs were tested for Hardy-Weinberg equilibrium using the program GDA (http://lewis.eeb.uconn.edu/ lewishome/). Age, gender, height and weight were strongly associated with BMD and/or BMC (Table 3), and therefore, multiple linear regression analysis was used to adjust for the effects of these factors prior to consideration of genotype effects. Gender-specific prediction models for bone measures included age, weight and height; gender-combined models included age, weight, height and gender. The association with each RFLP was evaluated using a two degreeof-freedom test (2 df test), which corresponded to adding an RFLP with three genotypic classes to the prediction models. Genotypespecific least squares (LS) means and standard errors (SE) were estimated and the null hypothesis of no difference among the three means was tested. A Bonferroni adjustment was used when multiple pairwise comparisons were made among genotype-specific LS means. For example, three pairwise comparisons were made among the three genotype-specific LS means for each RFLP, using a P-value of 0.05/3=0.017 to determine the significance of each comparison. The coefficient of multiple determination (R2) was also calculated for each model. The association with each dinucleotide and VNTR polymorphism was similarly evaluated. Only those genotypes represented by at least ten children for gender-combined analyses, or by at least ten boys/girls for gender-specific analyses were included in the model. Allele-specific LS means were also determined for each allele that was represented by at least ten children. The mean comparisons were between children that had a specific allele at a dinucleotide or VNTR locus versus children that did not have that allele. A Bonferroni adjustment was used to adjust the significance level for these allele-specific comparisons. Gene-by-gene interaction effects were tested for VDR, osteocalcin, COL1A1, COL1A2 and ER, in models that contained two RFLPs and their product, which represented the interaction effect.

Table 2 Description of phenotypes, by gender, for 428 non-Hispanic white children 4.5–6.5 years of age. BMD is expressed as g/cm2; BMC is expressed in g

Age Height (cm) Weight (kg) BMI BMD-whole body BMD-hip BMD-spine BMC-whole body BMC-hip BMC-spine

Girls n=228

Boys n=200

Mean

SD

Mean

SD

5.27 110.78 19.83 16.06 0.710 0.549 0.506 474.36 6.85 14.62

0.39 5.47 3.79 1.96 0.038 0.054 0.061 95.86 1.50 2.45

5.22 112.08 20.48 16.24 0.728 0.579 0.498 505.15 7.27 15.45

0.38 5.33 3.70 1.98 0.039 0.058 0.056 96.08 1.58 2.48

also excluded these children. The remaining 428 nonHispanic white children formed the cohort for our study. Baseline characteristics for the 428 non-Hispanic white children are shown in Table 2. The cohort includes 228 girls and 200 boys. Among the covariates age, gender, height and weight, the strongest predictors of BMD were gender and weight; the strongest predictors of BMC were height and weight (Table 3). Hence, bone accrual at this age is strongly associated with body size. The gender-specific base models for BMD and BMC contained age, height and weight. The gender-combined models also contained an indicator variable for gender. In the gender-combined analysis, the four covariates explained 25% or less of the variability in BMD, but more than 50% of the variability in BMC. BMD and BMC were adjusted for these factors before investigating the association with candidate gene polymorphisms.

Results

Genotype frequencies

Characteristics of the cohort

Genotype frequencies for each RFLP are shown in Table 4 for the cohort. The hypothesis of Hardy-Weinberg equilibrium was not rejected for any one of the RFLPs (P>0.15). Table 5 shows the genotype frequencies for the dinucleotide and VNTR polymorphisms. Only those genotypes present in at least ten children are represented in this table. These genotypes were included in the gender-combined analyses. The osteopontin polymorphism demonstrated significant Hardy-Weinberg

Four hundred and seventy children 4.5–6.5 years of age participated in the first examination for the Iowa Bone Development Study. Of the 470 participants, 13 fell outside the age range and therefore, were excluded. Twenty-nine children represented ethnic/racial minorities. Because of differences in allele frequencies for the candidate genes among the different ethnic groups, we

Table 3 Multiple linear regression base models for 428 non-Hispanic white children Base model

Age (years) Gender Height Weight

Bone measure/b+SE BMD-whole body

BMD-hip

BMD-spine

BMC-whole body

BMC-hip

BMC-spine

0.024+0.006*** 0.019+0.004*** 0.000+0.000 0.001+0.001

0.014+0.008 0.026+0.005*** 0.001+0.001 0.004+0.001***

0.007+0.008 )0.014+0.005* 0.002+0.001** 0.005+0.001***

25.2+7.7** 15.9+5.1** 4.7+0.8*** 14.9+1.0***

0.26+0.16 0.19+0.10 0.12+0.02*** 0.14+0.02***

)0.16+0.26 0.41+0.17* 0.25+0.3*** 0.14+0.03***

*P<0.01; **P<0.005; ***P<0.0001

653 Table 4 Genotype frequencies for candidate gene RFLPs, for 428 non-Hispanic white children. HWE P-test for Hardy-Weinberg equilibrium

COL1A1 RsaI COL1A1 Sp1 COL1A2 RsaI Estrogen receptor PvuII Estrogen receptor XbaI Osteocalcin promoter Vitamin D receptor BsmI Vitamin D receptor TI

n

+/+

+/)

)/)

HWEP

428 426 426 427 426 426 427 426

269 (62.9%) 271 (63.6%) 192 (45.1%) 121 (28.3%) 177 (41.6%) 26 (6.1%) 146 (34.2%) 61 (14.3%)

140 136 182 212 185 152 210 201

19 (4.4%) 19 (4.5%) 52 (12.2%) 94 (22.0%) 64 (15.0%) 248 (58.2%) 71 (16.6%) 164 (38.5%)

>0.80 >0.70 >0.30 >0.90 >0.15 >0.60 >0.70 >0.90

(32.7%) (31.9%) (42.7%) (49.7%) (43.4%) (35.7%) (49.2%) (47.2%)

Table 5 Genotypea frequencies for candidate gene VNTRs, for 428 non-Hispanic white children Osteonectin n=427

Osteocalcin n=426

Osteopontin n=421

Estrogen receptor n=425

COL1A2 intron 12 n=425

Androgen receptor n=426

Genotype

%

Genotype

%

Genotype

%

Genotype

%

Genotype

%

Genotype

1/2 1/3 2/2 2/3 2/4 2/6 3/3 3/4 3/5 3/6 4/4 4/6 – – – – – – – – – – – Other HWP>0.30

2.8 8.4 6.1 21.1 8.7 3.5 14.5 12.4 2.8 4.2 3.3 2.6 – – – – – – – – – – – 9.6

2/10 4/10 8/10 8/13 9/10 9/13 10/10 10/12 10/13 12/13 13/13 – – – – – – – – – – – – Other HWP<0.025

2.6 2.4 3.6 2.8 5.7 4.0 19.7 4.8 22.6 4.8 8.8 – – – – – – – – – – – – 18.2

1/2 1/3 1/5 1/8 2/2 2/3 2/4 2/5 2/8 3/5 3/8 4/5 4/8 5/5 5/8 5/17 8/8 8/17 – – – – – Other HWP>0.30

3.0 3.5 4.0 4.7 3.5 4.9 2.8 7.8 8.4 3.8 7.3 2.8 2.6 3.0 8.9 3.3 6.3 2.8 – – – – – 16.6

3/4 4/4 4/5 4/9 4/11 4/12 4/13 5/11 11/12 11/13 – – – – – – – – – – – – – Other HWP>0.90

3.3 10.8 5.6 3.1 6.6 5.6 7.3 2.8 2.4 2.6 – – – – – – – – – – – – – 49.9

AA AB AC AD BB BC BD CC CD – – – – – – – – – – – – – – Other HWP>0.80

5.4 10.4 20.5 5.4 3.1 18.4 4.7 21.9 9.2 – – – – – – – – – – – – – – 1.0

Boys 8 9 10 11 12 13 14 15 Girls 11/14 Allelesb 8 9 10 11 12 13 14 15 16 17 18 – –

% 6.2 11.9 15.5 18.0 12.9 9.3 10.3 6.2 6.1 11.4 21.5 20.2 30.7 17.1 18.4 24.6 13.2 11.4 6.6 4.4 – –

a

Genotypes represented by at least ten children. For the androgen receptor (AR), which is an X-linked gene, males have only one allele. For females, there was only one genotype (11/14) represented by at least ten children Androgen receptor alleles represented in ten of more females are also displayed for comparison

b

disequilibrium (P<0.025) due to the frequencies of some of the rarer genotypes. COL1A2 and osteocalcin gene polymorphisms, BMD and BMC The COL1A2 and osteocalcin gene polymorphisms showed the strongest association with bone measures in this cohort (Tables 6, 7, 8 and 9). Children with the COL1A2 RsaI ()/)) genotype had the lowest whole body BMC, compared to those with either the (+/)) or (+/+) genotypes (P<0.05) (Table 6). A similar trend was observed for whole body BMD and hip BMD (P<0.10; 2 df test). The gender-specific analyses also

identified the COL1A2 RsaI genotype as being associated with BMD/BMC, but only in girls. Consistent with the results in the entire cohort, girls with the RsaI ()/)) genotype (n=31) had significantly lower whole body BMC, compared to those with either the (+/)) or (+/+) genotypes (P<0.025); results for whole body BMD and hip BMC were suggestive as well (P<0.10; 2 df test). We observed a similar trend in boys, but the number of boys with the ()/)) genotype was small (n=21), precluding detection of a statistically significant difference among genotypes because of low power. Neither COL1A1 RsaI nor the COL1A1 Sp1 polymorphism was associated with any of the BMD/BMC measures (data not shown) in either the combined or gender-specific analyses.

654 Table 6 Association between COL1A2 RFLP and BMD and BMC for 428 non-Hispanic white children Boys and girls

n=192 BMD (g/cm2) Whole body Hip Spine BMC (g) Whole body Hip Spine

COL1A2 RsaI genotype

P-valuea

ModelR2

()/))

(+/+)

(+/))

n=182

n=52

0.720±0.003 0.568±0.004 0.506±0.004

0.722±0.003 0.561±0.004 0.500±0.004

0.708±0.005 0.552±0.007 0.492±0.007

<0.10 <0.10 >0.15

0.14 0.25 0.26

491.44±3.75 7.06±0.08 15.02±0.13

491.26±3.84 7.11±0.08 15.08±0.13

471.68±7.19 6.79±0.15 14.73±0.25

<0.05 >0.15 >0.40

0.72 0.55 0.51

H0: no genotype effect after adjustment for age, height, weight (and gender for boys and girls combined) Table 7 Association between COL1A2 VNTR and BMD and BMC for 428 non-Hispanic white children Boys

n

BMC-hip P<0.025

BMD-hip P<0.005

Genotype AA AB AC BC BD CC CD

LS mean+SE 11 22 34 43 10 44 17

LS mean+SE 8.23+0.34 7.39+0.24 7.14+0.19 6.93+0.17 6.97+0.36 7.51+0.17 7.12+0.27

0.626+0.016 0.590+0.011 0.561+0.009 0.567+0.008 0.569+0.017 0.593+0.008 0.570+0.013

Girls

n

BMC-whole body P<0.025

BMD-hip P<0.025

Genotype AA AB AC AD BC BD CC CD

12 22 53 15 35 10 49 22

LS mean+SE 481.67+13.03 469.98+9.61 492.62+6.23 488.28+11.64 466.53+7.72 475.57+14.32 459.83+6.46 468.59+9.61

LS mean+SE 0.541+0.013 0.557+0.010 0.563+0.007 0.561+0.012 0.546+0.008 0.534+0.015 0.531+0.007 0.560+0.010

Table 8 Association between Osteocalcin RFLP and BMD and BMC for 428 non-Hispanic white children Boys and girls

n=26 BMD (g/cm2) Whole body Hip Spine BMC (g) Whole body Hip Spine

Osteocalcin promoter genotype

P-valuea

R2

()/))

(+/+)

(+/))

n=152

n=248

0.711±0.007 0.544±0.010 0.498±0.010

0.717±0.003 0.562±0.004 0.498±0.004

0.721±0.002 0.567±0.003 0.504±0.003

>0.30 <0.10 >0.50

0.13 0.26 0.26

473.70±10.16 6.55±0.21 14.89±0.35

486.75±4.23 7.15±0.09 15.01±0.14

492.13±3.29 7.05±0.07 15.04±0.11

>0.15 <0.05 >0.90

0.72 0.55 0.51

H0: no genotype effect after adjustment for age, height, weight (and gender for boys and girls combined)

The COL1A2 intron 12 VNTR genotype was also associated with BMD/BMC (Table 7). However, the pattern of genotypic means was different for boys and girls, and therefore, significant assoiciations were only identified in the gender-specific analyses. Among the seven genotypes represented in at least ten boys, there was evidence of an overall association between hip BMD (P<0.005) and hip BMC (P<0.025). After adjustment

for multiple comparisons, the LS mean for genotype AA was significantly higher than the LS mean for genotypes BC (P<0.025) and AC (P<0.01) for hip BMD. In addition, the LS mean for genotype AA was also significantly higher than the LS mean for genotype BC for hip BMC (P<0.025). In girls, eight genotypes were represented in at least ten girls. There was evidence of an association between hip BMD (P<0.025) and whole

655 Table 9 Association between the osteocalcin D1S3737 6 allele and BMD/BMC Bone measure BMD (g/cm2) Whole body Hip Spine BMC (g) Whole body Hip Spine

Without allele 6

With allele 6

0.719+0.002 0.564+0.002 0.502+0.002

0.700+0.012 0.534+0.016 0.468+0.016

489.88+2.55 7.07+0.05 15.04+0.09

463.00+16.44 6.44+0.33 14.69+0.56

body BMC (P<0.025). In contrast to the boys, the AC genotype was associated with the highest LS mean for both bone measures in girls. Moreover, after adjustment for multiple comparisons, the LS mean for genotype AC was significantly higher than the LS mean for genotype CC for whole body BMC (P<0.01). In addition, the LS mean for genotype AC was significantly higher than the LS mean for genotype CC for hip BMD (P<0.025). Although the pattern of genotypic means was different for boys and girls, we consistently found the A allele to be associated with the highest BMD/BMC measures in the gender-combined analyses. One hundred and seventy-seven children had an A allele. These children had higher LS means for all BMD/BMC, when compared to children without an A allele (statistical significance was reached for whole body BMC, P<0.01; hip BMC, P<0.025; and whole body BMD P<0.05, and was suggestive for the other measures). One hundred and fifty-five children had a B allele; 298 had a C allele; and 86 had a D allele. The presence of a B or C allele was associated with consistently but not dramatically lower BMD/BMC measures. There was no clear pattern of LS means associated with the presence of a D allele. The osteocalcin (C/T) promoter polymorphism was also associated with bone measures in our cohort (Table 8). Children with the (+/+) genotype had lower hip BMC, compared to those with the (+/)) (P<0.025) or the ()/)) (P<0.10) genotypes (P<0.05; 2 df test). A similar trend was observed for hip BMD (P<0.10; 2 df test). While there was no overall association between BMD/BMC measures and the osteocalcin marker, D1S13737, in the gender-combined or gender-specific analyses, the allele-specific analyses identified an association with the 6 allele (Table 9). There were only ten children with a 6 allele, but they had dramatically lower LS means for each bone measure when compared to children without a 6 allele. Because of the small sample size, none of the differences was statistically significant after adjustment for multiple comparisons, however. Other candidate gene polymorphisms, BMD and BMC We observed several other associations of candidate gene polymorphisms with BMD/BMC in our cohort.

Although the VDR BsmI genotype has been associated with BMD in some children [8], we did not identify an association with any BMD/BMC measures in our cohort. However, there was an association between the VDR translation initiation site polymorphism (VDRTI) and BMD/BMC. The pattern of association differed between boys and girls. In boys, the VDR-TI genotype was associated with whole body BMC (P<0.05; 2 df test); an association was also suggested for both whole body BMD and spine BMC (P<0.10; 2 df test). Boys with the VDR-TI (+/+) genotype had the highest LS means. The opposite trend was observed in girls; those with the ()/)) genotype tended to have the highest LS means. In contrast to the situation with COL1A2 and osteocalcin gene polymorphisms, however, the LS means did not progress across genotypes, making it difficult to know how to interpret these results. The osteonectin genotype was associated with hip BMD (P<0.05) in the gender-combined analyses. Genotypes 3/6 (n=18) and 4/6 (n=11) were associated with the highest hip BMD (0.599+0.012 for the 3/6 genotype; 0.595+0.015 for the 4/6 genotype). These findings appeared to be driven by the osteonectin 6 allele, since both the 3/3 (n=62) and 4/4 (n=14) genotypes had lower LS means [0.561+0.006 and 0.552+0.013, respectively] than either the (3/6) or (4/6) genotypes. Because there were only three children with the 6,6 genotype, they were not included in this analysis. In contrast, the 3/5 genotype (n=12) was associated with the lowest hip BMD (0.535+0.015). This trend was observed for most of the other bone measures as well. The osteonectin polymorphism was also associated with whole body BMD in girls (P<0.025). After adjustment for multiple comparisons, the whole body BMD LS mean for genotype 1/2 (0.745+0.012) was significantly higher (P<0.05) than the LS mean for genotypes 1/3 (0.698+0.009), 2/4 (0.702+0.008) or 3/3 (0.704+0.006) in the girls. From the allele-specific analysis, it could be seen that children with a 5 allele tended to have lower measures of BMD/BMC and children with a 6 allele tended to have higher measures of BMD/BMC. After adjustment for multiple comparisons, children with a 5 allele had significantly lower (P<0.025) hip BMD (0.535+0.010) than children without a 5 allele (0.565+0.002); children with a 6 allele had significantly higher measures of both hip BMC (P<0.025) and hip BMD (P<0.05), (7.46+0.14; 0.582+0.007, respectively) than children without a 6 allele (6.99+0.05; 0.561+0.003). None of the genotypes for the other candidate genes was associated with BMD/BMC measures in either the gender-combined or gender-specific analyses. However, for the osteopontin dinucleotide polymorphism, allele specific analyses indicated that the 9 allele was associated with spine BMC. For osteopontin, the presence of a 9 allele (n=48) was associated with higher spine BMC compared to children without a 9 allele (15.72+0.25 g versus 14.93+0.09 g) (P<0.05 after adjustment for multiple comparisons).

656

Gene-by-gene interaction effects We identified several gene-by-gene interaction effects on BMD and BMC. Our analyses focused on the postulated interaction between pairs of RFLPs for COL1A1, COL1A2, VDR, osteocalcin and ER; none of the dinucleotide or VNTR markers was included in this analysis. Potential interactions were selected for analysis based on biological function and potential interaction of gene products during bone formation. In addition, we had previously identified gene-by-gene interaction effects involving VDR and ER that impacted BMD in pre and perimenopausal women in the Michigan Bone Health Study. Thus, we examined this interaction effect in our cohort as well. None of the VDR BsmI, ER XbaI and ER PvuII genotypes was associated with any of the BMD/BMC measures. However, analysis of the entire cohort revealed a significant VDR BsmI·ERXbaI interaction effect for both whole body BMD (P<0.025) and whole body BMC (P<0.05), and a suggestive effect for hip BMC (P<0.10). Children with the combined VDR BsmI (+/+)/ER XbaI ()/)) genotype (n=17) consistently had the highest LS means for all six BMD/ BMC measures, while those with the VDR BsmI (+/+)/XbaI (+/+) genotype (n=66) had among the lowest LS means (Fig. 1A). A similar trend was observed in the gender-specific analyses. There was a very strong association between the ER XbaI and ER PvuII genotypes (P<0.0001). So, not surprisingly, there was also the suggestion of a VDR BsmI by ER PvuII interaction effect for whole body BMC, spine BMC, hip BMD and spine BMD (P<0.10) (data not shown). Children with the combined VDR BsmI (+/+)/ER PvuII ()/)) genotype (n=27) had the highest LS means for whole body BMC, spine BMC, hip BMD and spine BMD, while children with the VDR

Fig. 1 Gene-by-gene interactions in 428 nonHispanic white children. A VDR (BsmI) by ER (XbaI) interactions for whole body BMC, whole body BMD and hip BMC. B VDR (TI) by ER (XbaI) interactions for whole body BMC, whole body BMD and spine BMC. C Osteocalcin (C/T promoter polymorphism) by ER (XbaI) interaction for whole body BMC, spine BMC and spine BMD. BMD and BMC are expressed in g/cm2 and g, respectively

BsmI (+/+)/ER PvuII (+/+) genotype (n=45) had among the lowest LS means. Gender-specific analyses showed significant interaction effects for spine BMD (P<0.05) in males and hip BMD (P<0.025) in females, with suggestive effects (P<0.10) for hip BMC and spine BMC in females. The pattern of LS means in the genderspecific analyses was consistent with those from the gender-combined analyses (data not shown). There was no association between the VDR BsmI and VDR TI genotypes (P>0.30). However, there was also evidence for VDR-TI by ER XbaI and VDR-TI by ER PvuII interaction effects for spine BMC (P<0.05) in the cohort. Consistent with the univariate analyses with VDR-TI and results from the other VDR by ER interaction effects, the VDR-TI (+/+) genotype, in combination with either ER XbaI ()/)) (n=8) or PvuII ()/)) (n=11) was consistently associated with the highest BMD/BMC LS means (Fig. 1B). In contrast to the VDR BsmI by ER interaction effects, however, the VDR-TI by ER interaction effects appeared to be driven by the VDR-TI genotype. The lowest spine BMC measures were observed for the VDR-TI ()/)) and either ER Xba ()/)) (n=11) or PvuII ()/)) (n=11) genotypes. The same trend was observed for most of the other BMD/BMC measures, although none reached statistical significance. The number of children with the osteocalcin C/T promoter polymorphism (+/+) genotype was small (n=26), and when combined with the ER XbaI genotype, the combinations with the C/T (+/+) genotype were too small for valid analysis. After deleting children with the C/T (+/+) genotype, there was evidence for a significant interaction effect for whole body BMC (P<0.025) and suggestion of an interaction effect for spine BMC and spine BMD (P<0.10) (Fig. 1C). The highest LS means were observed with the C/T ()/))

657

(n=247) genotype in combination with ER XbaI (+/)) (n=111) or ()/)) (n=33).

Discussion Increasing evidence suggests that genetic factors play a critical role in bone accrual in childhood and adolescence [1, 2]. We analyzed the relationship between genotypes at candidate gene loci and BMD/BMC in 428 non-Hispanic white children. We included genes that encode proteins with a prominent role in new bone formation (type I collagen, osteocalcin, osteonectin, osteopontin) and mineralization (VDR), reasoning that variation in such genes might be likely to influence BMD and BMC in rapidly growing children. In addition, we included genes that had previously been implicated in BMD/BMC determination in other population-based studies, although most had never been studied in young children. The COL1A2 and osteocalcin genotypes were the strongest predictors of BMD/BMC in our cohort. For the COL1A2 gene, the RsaI ()/)) genotype was consistently associated with the lowest BMD/BMC measures, particularly in girls. We did not have sufficient power to detect an association for the boys, but the trend was consistent with that observed for the girls. Given that 21 boys had the ()/)) genotype, we estimated that there was less than 20% power to detect the adjusted mean differences in boys. The other COL1A2 marker (the intron 12 VNTR) was also associated with BMD/BMC measures, lending additional support to the importance of COL1A2 in bone accrual in our cohort. The A allele was associated with higher bone measures in the gender-combined analyses, although the pattern of genotype-specific means was different in the boys and girls. The gender-specific genotype patterns were consistent among the bone measures, however. The genderspecific pattern differences for the VNTR marker could potentially be related to an interaction effect between physical activity and genotype. We know that the boys in our cohort are more physically active than the girls. In addition, the physical activity differential seems to have the greatest impact on the hip, with males having higher hip bone measures, compared to girls (Janz et al., unpublished data). It may be that physical activity impacts certain genotypes more than others in much the same way that calcium intake and absorption have been shown to modify the effect of VDR genotype on BMD. This hypothesis will be addressed in future analyses. This is the first report of COL1A2 genotypes influencing BMD/BMC levels in a population-based study, although specific nucleotide substitutions in the COL1A2 gene have been previously demonstrated in adults with low BMD [9] and in siblings with juvenile osteoporosis [10]. The COL1A2 markers used in this study are not known to have functional significance. We speculate, however, that they are in linkage disequilibrium with other sequences within or near the gene that confer

functional significance. The dinucleotide repeats in the 5’ flanking region and the first intron of the COL1A2 gene may be important regions to investigate since these polymorphic sequences have been shown to regulate transcriptional activity of the gene, with both the number and combination of repeats at each site contributing to differences in transcriptional activity [11]. We also found that the osteocalcin genotype, identified by the C/T promoter polymorphism was associated with hip BMC, and to a lesser extent, hip BMD in our cohort. These findings are consistent with studies in Asian women [12,13]. In addition, we found that one particular allele of D1S3737 was associated with lower bone measures in our cohort. These results support our previous observations in a case-control study of postmenopausal women [7], and suggest that osteocalcin is an important determinant of both bone accrual and bone loss. It should be noted, however, that the specific allele found to be associated with bone measures in our case-control study is not identical to the ‘‘risk allele’’ identified in the Iowa Bone Development study. The latter cohort is much larger than the postmenopausal cohort (n=428 versus 140), the number of D1S3737 alleles identified is greater and therefore, the numbering of alleles is different. The basic tenant that a specific D1S3737 allele may be associated with either low or high BMD/BMC, however, remains the same. We identified several other associations that were limited to either the gender-combined, gender-specific or allele-specific analyses. For osteonectin and osteopontin, the association appeared to be driven by one specific allele. The association of the osteopontin 9 allele with higher spine BMC supports our previous work from the Michigan Bone Health Study. In the latter cohort, women with the osteopontin 9/13 genotype had significantly higher baseline femoral neck BMD (P<0.01), compared to those without this genotype [5]. Unlike the studies by Sainz [8,14] and Ames [15], we did not identify an association between either the VDR BsmI or the COL1A1 Sp1 genotypes and BMD or BMC. It should be noted, however, that there appears to be a gene by environmental interaction effect between dietary calcium absorption and VDR genotype, which potentially may prove relevant in our cohort [16,17,18]. We are collecting longitudinal dietary data on our children, and therefore, the interaction between calcium and all genotypes will eventually be examined in our model. Preliminary examination of dietary data suggests that calcium intake varies widely among our children. These differences may eventually help to explain some of the discrepancies between our work and studies of Sainz [8] and Ames [15]. In addition, this information may also be relevant for the relationship between the COL1A1 Sp1 genotype and BMD/BMC [14,18]. Although none of the VDR (BsmI) or ER genotypes alone was associated with BMD or BMC, we observed a significant gene by gene interaction effect between ER and both VDR markers. In addition, there was a significant interaction effect for both ER (XbaI and PvuII)

658

polymorphisms and the osteocalcin (C/T) polymorphism, which is a new observation. We previously reported an interaction effect between VDR (BsmI) and both ER (PvuII) and (XbaI) in women from the Michigan Bone Health Study [5]. Similar observations have been reported for other adult cohorts [19], but not for prepubertal children. In contrast to the findings in the Michigan Bone Health Study, we did not observe an association between ER genotypes and BMD/BMC measures in either our gender-combined or gender-specific analyses. While this is probably not surprising since the children in our study are prepubertal, it will be interesting to see if such a relationship is detected during puberty when achievement of peak bone mass parallels the increase in sex steroid hormone levels. One might predict that the genes encoding sex steroid hormone receptors will be important for bone accrual during adolescence in this cohort. Support for this theory comes from studies of the ER genotype in adolescent females [20]. In summary, we have identified the COL1A2 and the osteocalcin genes as determinants of BMD/BMC in our cohort of children. Genes encoding other bone matrix proteins such as osteonectin and osteopontin may also play a role in bone accrual. Moreover, the gene-by-gene interaction effects observed between ER, VDR and osteocalcin suggest that the combination of genotypes at several loci may be just as important as a single genotype in determining BMD/BMC in children. Acknowledgements The authors gratefully acknowledge Joan Grabin and Jennifer Tisch for their organizational assistance, and Marta Tullis and De Frei for their assistance with BMD/BMC measurements. We thank Sheela Koluri, Sonali Patel, Mike Mueller, Barbara Broffitt, and Erica Pugh for assistance with data management and analysis. This work was supported by NIH grants R01-DE12101 and R01-DE09551, and the General Clinical Research Centers Program, National Center for Research Resources RR00059.

References 1. Ferrari S, Rizzoli R, Slosman D, Bonjour, JP (1998) Familial resemblance for bone mineral mass is expressed before puberty. J Clin Endocrinol Metabol 83:358–361 2. Jones G, Nguyen TV (2000) Associations between maternal peak bone mass and bone mass in prepubertal male and female children. J Bone Miner Res 15:1998–2004 3. Slemenda CW, Christian JC, Williams CJ, Norton JA, Johnston CC Jr (1991) Genetic determinants of bone mass in adult women: a reevaluation of the twin model and potential importance of gene interaction on heritability estimates. J Bone Miner Res 6:561–567 4. Levy SM, Kiritsy MC, Slager SL, Warren JJ (1998) Patterns of dietary fluoride supplement use during infancy. J Publ Health Dent 58:228–233 5. Willing MC, Sowers MF, Aron D, Clark MK, Burns T, Bunten C, Crutchfield M, D’Agostino D, Jannausch M (1998) Bone

6.

7.

8.

9. 10.

11. 12.

13.

14.

15.

16.

17.

18.

19.

20.

mineral density and its change in white women: estrogen and vitamin D receptor genotypes and their interaction. J Bone Miner Res 13:695–705 Sowers MF, Willing M, Burns T, Deschenes S, Hollis B, Crutchfield M, Jannausch M (1999) Genetic markers, bone mineral density, and serum osteoclacin levels. J Bone Miner Res 14:1411–1419 Raymond MH, Schutte BC, Torner JC, Burns TL, Willing MC (1999) Osteocalcin: genetic and physical mapping of the human gene (BGLAP) and its potential role in postmenopausal osteoporosis. Genomics 60:210–217 Sainz J, VanTornout JM, Loro ML, Sayre J, Roe TF, Gilsanz V (1997) Vitamin D-receptor gene polymorphisms and bone density in prepubertal American girls of Mexican descent. N Engl J Med 337:77–82 Spotila LD, Colige A, Sereda L, Prockop DJ (1994) Mutation analysis of coding sequences for type I procollagen in individuals with low bone density. J Bone Miner Res 9:923–932 Dawson PA, Kelly TE, Marini JC (1999) Extension of phenotype associated with structural mutations in type I collagen: siblings with juvenile osteoporosis have an a2(I)Gly436 fi arg substitution. J Bone Miner Res 14:449–455 Akai J, Kimua A, Hata RI (1999) Transcriptional regulation of the human type I collagen a2 (COL1A2) gene by the combination of two dinucleotide repeats. Gene 239:65–73 Dohi Y, Iki M, Ohgushi H, Gojo S, Tabata S, Kajita E, Nishino H, Yonemasu K (1998) A novel polymorphism in the promoter region for the human osteocalcin gene: the possibility of a correlation with bone mineral density in postmenopausal Japanese women. J Bone Miner Res 13:1633–1639 Chen HY, Tsai HD, Chen WC, Wu JY, Tsai FJ, Tsai CH (2001) Relation of polymorphism in the promotor region for the human osteocalcin gene to bone mineral density and occurrence of osteoporosis in postmenopausal Chinese women in Taiwan. J Clin Lab Anal 15:251–255 Sainz J, VanTornout JM, Sayre J, Kaufman F, Gilsang V (1999) Association of collagen type I alpha1 gene polymorphism with bone density in early childhood. J Clin Endocrinol Metab 84:853–855 Ames SK, Ellis KJ, Gunn SK, Copeland KC, Abrams SA (1999) Vitamin D receptor gene Fok1 polymorphism predicts calcium absorption and bone mineral density in children. J Bone Miner Res 14:740–746 Kiel DP, Myers RH, Cupples LA, Kong XF, Ordovas J, Schaefer EJ, Felson DT, Rush D, Wilson PWF, Eisman JA, Holick MF (1997) The BsmI vitamin D receptor restriction fragment length polymorphism (bb) influences the effect of calcium intake on bone mineral density. J Bone Miner Res 12:1049–1057 Ferrari SL, Rizzoli R, Slosman DO, Bonjour JP (1998) Do dietary calcium and age explain the controversy surrounding the relationship between bone mineral density and vitamin D receptor gene polymorphisms? J Bone Miner Res 13:363–370 Brown MA, Haughton MA, Grant SFA, Gunnell AS, Henderson NK, Eisman JA (2001) Genetic control of bone density and turnover: role of the collagen 1a1, estrogen receptor, and vitamin D receptor genes. J Bone Miner Res 16:758–764 Gennari L, Becherini L, Masi L, Mansani R, Gonnelli S, Cepollaro C, Tanini A, Martini S (1997) Vitamin D and estrogen receptor allelic variants in postmenopausal women: evidence for an additive effect on bone mineral density (abstract). J Bone Miner Res 12:F581 Gustavvson A, Lorentzon M, Lorentzon R, Nordtrom P (2000) Estrogen receptor alpha gene polymorphism in healthy adolescent girls; relation to bone area and estrogen level dependent relation to bone density. J Bone Miner Res 15: SA161, S262.