Bone Mineral Density and Its Change in White Women: Estrogen and Vitamin D Receptor Genotypes and Their Interaction

Authors


Abstract

Low bone mineral density (BMD) is a major risk factor for development of osteoporosis; increasing evidence suggests that attainment and maintenance of peak bone mass as well as bone turnover and bone loss have strong genetic determinants. We examined the association of BMD levels and their change over a 3-year period, and polymorphisms of the estrogen receptor (ER), vitamin D receptor (VDR), type I collagen, osteonectin, osteopontin, and osteocalcin genes in pre- and perimenopausal women who were part of the Michigan Bone Health Study, a population-based longitudinal study of BMD. Body composition measurements, reproductive hormone profiles, bone-related serum protein measurements, and life-style characteristics were also available on each woman. Based on evaluation of women, ER genotypes (identified by PvuII [n = 253] and XbaI [n = 248]) were significantly predictive of both lumbar spine (p < 0.05) and total body BMD level, but not their change over the 3-year period examined. The VDR BsmI restriction fragment length polymorphism was not associated with baseline BMD, change in BMD over time, or any of the bone-related serum and body composition measurements in the 372 women in whom it was evaluated. Likewise, none of the other polymorphic markers was associated with BMD measurements. However, we identified a significant gene × gene interaction effect (p < 0.05) for the VDR locus and PvuII (p < 0.005) and XbaI (p < 0.05) polymorphisms, which impacted BMD levels. Women who had the (−/−) PvuII ER and bb VDR genotype combination had a very high average BMD, while individuals with the (−/−) PvuII ER and BB VDR genotype had significantly lower BMD levels. This contrast was not explained by differences in serum levels of osteocalcin, parathyroid hormone, 1,25-dihydroxyvitamin D, or 25-dihydroxyvitamin D. These data suggest that genetic variation at the ER locus, singly and in relation to the vitamin D receptor gene, influences attainment and maintenance of peak bone mass in younger women, which in turn may render some individuals more susceptible to osteoporosis than others.

INTRODUCTION

Osteoporosis is a major health care problem in the United States and other parts of the world. Low bone mineral density (BMD) is an important risk factor for osteoporosis and related fractures.(1) While reproductive, nutritional, and life-style factors impact BMD, numerous family and twin studies also demonstrate a strong genetic component to its determination.(2,3) Comparison of intrapair differences in BMD between monozygotic and dizygotic adult twin pairs suggests that the heritability of BMD ranges from 45 to 84%, depending on the skeletal site examined.(4) Likewise, the family study by Sowers et al.(3) suggested that ∼50% of variation in BMD at the femoral neck was potentially genetic. Among the genes implicated as determinants of BMD are the genes for the vitamin D receptor (VDR) and the estrogen receptor. The finding of premature osteoporosis in persons with osteogenesis imperfecta suggests a possible role for the type I collagen genes in bone integrity. However, the relative importance of any one gene, particularly that of specific VDR genotypes, has been controversial.(5–9) Moreover, causative mutations have yet to be identified in any genes, except in a few isolated cases.(10–12)

The objective of this study was to determine bone-related genotypes in a population-based study of pre- and perimenopausal women and relate these genotypes to BMD measurements. Because of the longitudinal nature of the parent study as well as the measurement of bone-related proteins, these genotypes could be associated with osteocalcin levels and their 3-year change as well as baseline BMD and its 3-year change.

MATERIALS AND METHODS

Subjects

Data were generated from a subset of the 583 women who constitute the enrollees in the longitudinal Michigan Bone Health Study. This population-based cohort of younger adult women was recruited using the historical family records of the Tecumseh, Michigan Community Health Study (n = 461), which was supplemented with a 1992 community census from which an additional 122 women were recruited. These 583 women (the entire cohort) represent 70% of the age-eligible population who were 24–44 years of age at the time of the 1992 baseline study. Written informed consent was obtained from all participants. This study was approved by the University of Michigan Institutional Review Board.

In the third (of four) annual examination of this cohort, whole blood was collected on filter paper cards and dried for subsequent DNA extraction and genotyping. We successfully completed VDR genotyping (defined by the intragenic BsmI polymorphism) for only 372 of the 583 women surveyed. We continued to experience amplification difficulties from the dot blots and the primer pairs for the other candidate genes. Therefore, midway through our fourth and last examination of the population, we altered our specimen collection methodologies and collected 1–2 ml of whole blood from the remaining 262 participants for analysis of the remaining genes. Genotyping was completed for 442 women, 356 with vitamin D genotyping, and 262 for other genes. There were 176 women for whom DNA was available as a result of the dot blots and whole blood. DNA from both the dot blots and the whole blood sample was evaluated for the VDR BsmI in 30 of these women to confirm the consistency of findings. Additionally, there was no difference in BMD, age, or body composition measures of the women who were not genotyped as compared with those who were.

Bone mineral densitometry

BMD and bone mineral content (BMC) were measured using dual-energy X-ray absorptiometry (DPX-L, analysis software version 1.3y, Lunar Corporation, Madison, WI, U.S.A.). Each woman was scanned by one of two certified technicians and BMD/BMC data were determined for total body, lumbar spine, and the proximal femur, which included the femoral neck, Ward's triangle, and trochanter. Calibration was performed daily and a lumbar spine phantom was scanned weekly. The coefficients of variation for dual-energy X-ray absorptiometry were less than 1.0%. For this study, we used BMD variables measured at baseline in 1992–1993 as well as the 3-year change in BMD from 1992–1993 to 1995–1996.

Questionnaires and physical measurements

Participants completed questionnaires that described their medical history, reproductive history, family history of osteoporotic conditions, medication use, smoking and alcohol use, and usual dietary intake, including that of calcium, in the previous year.(13) Physical activity levels were estimated with a recall of minutes of physical activity during the previous July and December. An algorithm based on the Stanford Five-City instrument (measured in metabolic equivalent time [METS]/week) was used to estimate activity level.(14)

The body composition measures, fat tissue (kg), and lean without bone (kg) were available as a result of total body calcium measurement.(15) Height (cm) and weight (kg) were measured in women wearing a single layer of light clothing without shoes using a stadiometer and a calibrated balance-beam scale, respectively. Body mass index (BMI) was calculated as the weight in kilograms divided by the square of the height in meters. Waist (cm) and hip (cm) circumferences were measured with a nonstretchable tape 3 cm above the umbilicus, after a relaxed expiration, and at the maximum girth around the gluteal muscle of the buttocks, respectively. Waist-to-hip ratio was calculated as an indirect estimate of relative abdominal fat by dividing the waist circumference by the hip circumference.

Biochemical assessment

For each annual examination, blood was drawn in days 3–7 of the follicular phase of the menstrual cycle and after participants had been fasting for 8 h. For those women without menses, blood was drawn fasting on the anniversary date of their annual exams. Reproductive hormones were assayed in the Reproductive Sciences Laboratory at the University of Michigan. Testosterone was assessed with a solid-phase125I radioimmunoassay using a testosterone-specific antibody immobilized to the wall of a polypropylene tube. The detection level was 0.05 ng/ml. The precision of the testosterone assay was ± 15.5%. Serum follicle-stimulating hormone (FSH) was measured with a two-site chemiluminometric immunoassay which uses constant amounts of two antibodies with specificity for intact FSH. The test results were based on a calibration curve derived from the standard World Health Organization (WHO) 2nd IRP 78/549. The minimum detectable concentration was 0.3 mIU/l. 1,25-dihydroxyvitamin D (1,25(OH)2D), 25-hydroxyvitamin D (25(OH)D), and parathyroid hormone (PTH) levels were analyzed using commercial radioimmunoassays (Incstar, Stillwater, MN, U.S.A.) in the laboratory of Bruce Hollis. The inter- and intra-assay variation was less than 10% for these assays. Osteocalcin levels were also measured using the Incstar radioimmunoassay (B.H.). For this study, we used osteocalcin measured at baseline in 1992–1993, as well as the 3-year change in osteocalcin from 1992–1993 to 1995–1996.

DNA extraction and genotyping

Genomic DNA was extracted from ∼25 μl of dried whole blood collected on filter paper cards using Chelex 100 (Perkin-Elmer, Norwalk, CT, U.S.A.). Blood spot DNA served as the template for VDR genotyping. Lymphocyte DNA, isolated from 1–2 ml of whole blood, served as the template for analysis of allelic variants at the remaining gene loci. Red blood cells were lysed at 4°C in cell lysis buffer (0.32 M sucrose, 10 mM Tris-HCl, pH 7.6, 5 mM MgCl2, 1% Triton X-100). Following three cycles of lysis, the final lymphocyte pellet was resuspended in 10 mM Tris-HCl, pH 8.0, 10 mM NaCl, 10 mM EDTA, and proteinase K (final concentration 1 mg/ml). Samples were incubated at 55°C overnight with periodic mixing to disperse the pellet; an aliquot of 1–3 μl (50–300 ng of DNA) was used for each polymerase chain reaction (PCR). Genotypes were generated by standard PCR-based methodology(16) in the laboratory of M.W. (estrogen receptor [ER], type I collagen A1 (COLI A1), COLI A2, osteonectin, osteopontin, and osteocalcin) or D.A. (VDR). All genotypes were independently verified by two individuals.

Estrogen receptor

Primers designed to amplify intragenic polymorphic PvuII and XbaI sites were derived from Kobayashi et al.(17) Amplification conditions were as follows: 94°C (3 minutes), followed by 35 cycles at 94°C (1 minutes), 58°C or 60°C (45 s), and 72°C (1.5 minutes).

Vitamin D receptor

The polymorphic BsmI restriction site was analyzed for the VDR gene; amplification primers were derived from Morrison et al.(18) PCR conditions were as follows: denaturation at 85°C for 5 minutes, followed by the addition of 2.5 U Taq polymerase; 40 cycles of PCR at 94°C (1 minute), 56°C (1 minute), and 72°C (1 minute). Amplifications were carried out in a 480 Perkin-Elmer thermocycler. Following the PCR, an aliquot of amplified material was cleaved with BsmI, according to the manufacturer's specifications (New England Biolabs, Beverly, MA, U.S.A.). The presence (+) or absence (−) of the enzyme recognition site was identified by ethidium bromide staining of fragments separated in 6% polyacrylamide. Genotypes were assigned as (+/+), (+/−), and (−/−). By convention, the VDR (+/+) and (−/−) genotypes correspond to bb and BB, respectively.(18,19)

COLI A1 and COLI A2

Three polymorphic sites were analyzed for both COLI A1 (MnII, MspI, and RsaI) and COLI A2 (PvuII, RsaI, and an intron 12 variable number tandem repeat [VNTR]). For the COLI A1 gene, amplification primers were derived from Baker et al.(20) (MspI) and Willing et al.(21) (RsaI); MnII primers are shown in Table 1. COLI A2 primers were derived from Baker et al.(20) (RsaI) and Pepe(22) (intron 12 VNTR); COLI A2 PvuII primers are shown in Table 1. A 3 minute hot start at 94°C, followed by 35 cycles of PCR, was used for each primer pair. PCR conditions were as follows: COLI A1 MnII, 94°C (1.5 minutes), 58°C (1.5 minutes), and 72°C (1 minute); COLI A1 MspI, 94°C (1.5 minutes), 60°C (1.5 minutes), and 72°C (1 minute); COLI A1 RsaI, 94°C (1.5 minutes), 55°C (1.5 minutes), and 72°C (1 minutes); COLI A2 PvuII, 94°C (30 s), 60°C (30 s), and 72°C (30 s); COLI A2 RsaI, 94°C (30 s), 55°C (30 s), and 72°C (30 s); COLI A2 VNTR, 94°C (1.5 minutes), 50°C (2 minutes), and 72°C (1 minutes). Amplifications were carried out in either Hybaid Instruments (Holbrook, NY, U.S.A.) or Perkin-Elmer thermocyclers.

Table Table 1. Primers for Polymorphic Markers Analyzed in a Pre- to Perimenopausal Population of Women
original image

Following each PCR, an aliquot of amplified material was cleaved with the appropriate restriction endonuclease, according to manufacturers' specifications (New England Biolabs). For restriction fragment length polymorphisms (RFLPs), genotypes were assigned as described above for VDR. For the intron 12 VNTR, an aliquot of amplified material was denatured at 94°C for 5 minutes and electrophoresed in a 6% polyacrylamide gel containing 7.6 M urea for 3 h at 30 W. Genotypes were identified by silver staining, using Silver Sequence Automatic Processor Compatible film (Promega, Madison, WI, U.S.A.).

Osteonectin, osteopontin, and osteocalcin

The domains of the osteonectin and osteopontin genes which contain an intragenic tandem cytosine adenine (CA) repeat were separately amplified, using previously described oligonucleotide primers.(23,24) Primers, shown in Table 1, amplified a (CA) repeat polymorphism tightly linked to the osteocalcin gene (Raymond, personal communication). This tandem repeat marker was originally identified in a 70 kb bacterial artificial chromosome and maps ∼0.07 cM from the osteocalcin gene. All amplifications were carried out for 35 cycles after a 3 minute denaturation at 94°C. Amplification conditions were as follows: osteonectin, 94°C (1.5 minutes), 55°C or 58°C (1 minute), and 72°C (1 minute); osteopontin, 94°C (1.5 minutes), 62°C (1 minute), and 72°C (1 minute); osteocalcin, 94°C (1.5 minutes), 55°C (1 minute), and 72°C (1 minute). Genotype identification was as described above for the COLI A2 VNTR.

Statistical analysis

Analysis of variance and analysis of covariance models were used to test for an association between phenotypic characteristics and genotypes at each genetic locus. The reported phenotypic means were obtained for each genotype by the method of least squares. We defined that these least squares means (LSMs) demonstrated a pattern of progression across the genotypes at the RFLP loci if they increased from (−/−) to (−/+) to (+/+), or if they decreased from absence of both alleles through presence of both alleles.

All associations were evaluated using a crude analysis as well as an adjustment for age, BMI, and smoking behavior (analysis of covariance). Furthermore, dependent variables included both BMD or BMC. BMC was evaluated in the eventuality that bone size might confound the association between BMD and the genotype of interest(25,26); however, since no evidence for confounding was identified, the BMD results are reported. All analyses used SAS procedures (SAS Proprietary Software, release 6.09, SAS Institute, Cary, NC, U.S.A.). The level of significance for main effects analyses was set at 0.005 to accommodate the number of hypotheses to be tested.

RESULTS

Genetic markers and BMD

Genotype frequencies for each RFLP appear in Table 2. There is no evidence of deviation from Hardy–Weinberg equilibrium for any of the eight polymorphisms (p > 0.05). Table 3 shows the genotype frequencies for the COLI A2 intron 12 VNTR polymorphism, as well as the polymorphisms for osteopontin, osteonectin, and osteocalcin. Because of the large number of alleles for osteonectin, osteopontin, and osteocalcin, we included in the analyses only those genotypes that were identified in 10 or more women (211 women with 8 genotypes for the osteonectin marker, 176 women with 7 genotypes for the osteopontin marker, and 185 women with 11 genotypes for the osteocalcin marker).

Table Table 2. Genotype Frequencies for Restriction Fragment Length Polymorphisms*
original image
Table Table 3. Genotype Frequencies (Percent) for Dinucleotide Repeats and the COLI A2 VTNR Polymorphism Among 262 Women from the Michigan Bone Health Study
original image

No polymorphism was significantly associated with baseline BMD or BMC measured at any site, except for the ER PvuII (p < 0.005 for baseline lumbar spine and baseline total body BMD; Table 4). The association with the ER XbaI genotype approached significance (p < 0.05, spine and 0.1 for total body). Women with the ER PvuII (+/+) genotype had significantly lower lumbar spine BMD (LSM ± SE = 1.228 g/cm2 ± 0.02) than did women with (+/−) (LSM ± SE = 1.305 g/cm2 ± 0.01; pairwise p < 0.001) or with (−/−) (LSM ± SE = 1.312 g/cm2 ± 0.023; pairwise p < 0.005). As with lumbar spine, total body mean BMD progressed across genotypes from (+/+) (LSM = 1.157 g/cm2) to (+/−) (LSM = 1.195 g/cm2) to (−/−) (LSM = 1.194 g/cm2). Total body mean BMD was significantly less for women with the (+/+) genotype than for those with either the (+/−) (pairwise p < 0.001) or (−/−) (pairwise p < 0.05) genotypes. The mean values for the women with (+/−) and (−/−) genotypes were not significantly different from each other (Fig. 1).

Figure FIG. 1.

Femoral neck, lumbar spine, and total body BMD levels in women with the (+/+) genotype of the PvuII ER polymorphism.

Table Table 4. Relationship of Estrogen Receptor Polymorphisms PvuII and XbaI and BMD, 3-Year Percent Change in BMD, Serum Osteocalcin Level, Circulating Hormone Levels, and Percent Smoking in Pre- and Perimenopausal Women
original image

There was no association between the VDR polymorphisms and baseline measures of BMD, as shown in Table 5. However, there was a significant gene-by-gene interaction effect (p < 0.05) for the VDR locus and both of the ER polymorphisms (but particularly PvuII) among the 171 women who had both the BsmI and PvuII genotyping. An example of the interaction can be seen in Fig. 2 where women who had the (−/−) PvuII receptor genotype and the bb VDR genotype consistently exhibited a very high average BMD; in contrast, women who had the (−/−) PvuII receptor and the BB VDR genotype had remarkably lower BMD. This pattern was observed at all three bone sites. As shown in Table 6, this contrast was not explained by significant differences in serum osteocalcin, PTH, 1,25(OH)2D, or 25(OH)D.

Figure FIG. 2.

Femoral neck, lumbar spine, and total body BMD levels in women with the (−/−) genotype of the PvuII ER polymorphism, by the VDR BsmI genotype.

Table Table 5. Relationship of the Vitamin D Receptor BSMI Polymorphism and BMD, 3-Year Percent Change in BMD, Serum Osteocalcin Level, Circulating Hormone Levels, and Percent Smoking in Pre- and Perimenopausal Women
original image
Table Table 6. Characteristics of Nine Strata Delineated by their Genotypes at the Estrogen Receptor PvuII and VDR BsmI Loci
original image

The osteopontin genotype was significantly associated with baseline femoral neck BMD (p < 0.01). The association was primarily due to increased femoral neck BMD among women with the 9/13 genotype.

Genetic markers and 3-year BMD change

No genetic locus was associated with the 3-year percentage change in BMD at any site, with the single exception of COLI A1 RsaI. This locus was weakly associated with 3-year femoral neck BMD change alone (p < 0.025) and following adjustment for age and BMI (p < 0.025). It was also associated with 3-year total body BMD change before (p < 0.025) and after adjustment (p < 0.05). However, the mean BMD change (%) did not demonstrate a pattern of progression across the COLI A1 RsaI genotypes in any of these models. The age and BMI-adjusted COLI A1 least squares means were −0.016 g/cm2 for the (−/−), 0.0002 g/cm2 for the (−/+), and −0.033 g/cm2 for the (+/+) in the femoral neck.

Serum osteocalcin levels and 3-year change

The relationship between osteocalcin levels and the genetic loci was evaluated at the baseline and follow-up time points, as well as by 3-year change. Osteocalcin genotype, as defined by the tightly linked dinucleotide repeat, was not associated with either the baseline serum osteocalcin level or its 3-year change. COLI A1 MnII was associated with 3-year change (p < 0.025), but the mean osteocalcin change did not show a pattern of progression across the genotypes. The 3-year change in osteocalcin levels for the (+/+), (+/−), and (−/−) groups was 0.182, 0.497, and −0.162 ng/ml, respectively.

COLI A2 PvuII was associated with 3-year osteocalcin change (p < 0.0005). Women who had the (−/−) genotype showed a decline in osteocalcin levels (LSM ± SE = −0.393 g/cm2 ± 0.213), while women with the (−/+) (LSM ± SE = 0.511 g/cm2 ± 0.117) or (+/+) (LSM ± SE = 0.437 g/cm2 ± 0.127) genotypes showed an increase in osteocalcin levels that were not significantly different from each other.

COLI A2 RsaI was associated with the 3-year change in serum osteocalcin (p < 0.01), a result that was driven by the difference between women with genotype (+/+) (LSM ± SE = 0.527 ng/ml ± 0.150), who had a mean increase in osteocalcin activity, and women with genotype (−/−) (LSM ± SE = −0.209 ng/ml ± 0.213), who had a mean decrease in osteocalcin activity.

Calciotropic hormone levels and FSH

There was no association between any marker and the calciotrophic hormone levels including PTH, serum 25(OH)D, or 1,25(OH)2D levels. These data are presented by VDR and estrogen receptor genotypes in Tables 5 and 6 and by the combination of ER and VDR genotypes in Table 6.

Anticipating that the onset of the perimenopause might exert a differential impact on the association between measures of bone mineral change or osteocalcin change, women were dichotomized into two groups. Women were considered premenopausal with serum FSH levels of 10 mIU/l or less, while women with serum FSH levels greater than 10 mIU/l were considered potentially perimenopausal (14% of women with genotyping). Apart from the ER genotypes, there was no significant relationship of any genetic marker with baseline BMD, BMD change, baseline osteocalcin, or osteocalcin change within FSH groups.

DISCUSSION

In this population-based study of women who were pre- or perimenopausal, we found that the PvuII and XbaI ER genotypes were associated with the level of BMD but not with its change over a 3-year period. Notably, we found no association between the BsmI VDR polymorphism (18% prevalence of homozygous (−/−) genotype, i.e., BB) and BMD level or its change over a 3-year period or with osteocalcin levels. However, a significant gene-by-gene interaction effect indicated that women who were homozygous (−/−) at the PvuII or XbaI loci (with prevalences of 18% and 10%, respectively) had significantly different BMD levels according to their VDR status.

ER genotypes were significantly associated with both lumbar spine and total body BMD levels in our cohort. Furthermore, both of the intragenic polymorphic sites for the estrogen receptor, PvuII and XbaI, showed a similar pattern of results. These findings add to other lines of evidence in support of a role for estrogen and its action mediated through the ER in attainment and maintenance of BMD. First, estrogen action has been associated with maintenance of BMD following menopause and in the prevention of fractures.(27) Second, in the ER knockout mouse model system, both males and females have decreased bone mineralization.(28) Third, several other studies have identified an association between ER genotype and BMD or osteoporosis,(17,29–31) although the genotype that predicts low or high BMD may be population specific. In Japanese women, the ER PvuII PP (−/−) genotype is associated with low BMD,(17) while in our population and a Finnish cohort,(30) the PvuII pp (+/+) genotype is associated with low BMD. Finally, in other population groups, there is no obvious relationship between BMD and ER genotype.(32–34) These discrepancies may reflect founder effects which lead to differences in the distribution of alleles among population groups. The variable effect of modifier genes in a given population group may be another confounding factor.

The relationship between VDR genotype, as identified by RFLPs and BMD levels and/or fractures, has been evaluated in numerous population groups since the report of a significant association in Australian Caucasians.(18) Results have been conflicting, with some studies confirming an association between the BB (−/−) genotype and low BMD and other studies not observing the relationship.(6,7,9,19,35–46) However, the interaction between ER genotypes and other candidate genes, e.g., VDR, has not been identified previously.

The mechanism through which the VDR might exert substantial influence on BMD remains elusive and is likely to be modified by the presence of various environmental and genetic factors. These factors have been suggested as younger age,(47,48) older age,(35) and lower calcium intake in some(49) but not other studies.(39) However, when we examined either BMD or BMC and VDR genotype in this population, which included a large number of premenopausal women with a substantial range of calcium intake, we found no such associations, which is in agreement with the work of Garnero et al.(38) Furthermore, when we examined either BMD or BMC and the VDR genotype in our cohort, after adjusting for age, BMI, smoking, and FSH levels in the follicular phase of the menstrual cycle, we found no such association.

We also examined the association of the VDR genotype with serum osteocalcin levels and the 3-year change. Osteocalcin is regulated by the VDR, which binds to its promoter in the presence of 1,25(OH)2D.(50) Studies have reported both significantly lower(42) and higher osteocalcin levels(18) in BB homozygotes, as well as no association.(38) Tokita et al.(43) reported higher osteocalcin levels in premenopausal Japanese women with the BB genotype. Furthermore, we found no such association between the BB genotype and either osteocalcin levels or their change over a 3-year period, or levels of 1,25(OH)2D.

The controversies about the relationship between VDR genotype and BMD suggest the possibility of other mechanisms at play. We found a significant gene-by-gene interaction effect between polymorphisms at the ER genotypes and the VDR genotype, genes which reside on different chromosomes. When the population was stratified according to the presence or absence of either restriction site for the ER as well as the BsmI restriction site for the VDR, we observed that BMD levels were consistently higher in women with the VDR bb (+/+) genotype and the (−/−) PvuII and XbaI ER genotypes. We do note that among women with the VDR (−/−) genotype (Table 6), there is substantial variation in FSH levels, suggesting that the menopausal status of women should be included in future studies of genetic characteristics. If such were the case, it would help explain a similar interaction that was recently reported in a postmenopausal Italian population group, where the VDR bb (+/+) and the PvuII and XbaI (+/+) combination predicted higher BMD values, while the VDR BB (−/−) and the (−/−) PvuII and XbaI genotypes predicted lower BMD levels.(51) Although the gene-by-gene interaction may be population specific, both studies suggest that, as with other complex traits, BMD levels are likely to be determined by several genes which act collectively. The combination of alleles at different loci may be more important than any one genotype at a particular locus. In addition, genes that primarily influence bone accretion may be different from those that affect bone resorption.

The mechanism of this gene-by-gene interaction effect is not known. Potentially, there is a physiologic (functional) link between the ER and vitamin D for which these intragenic sites are acting as markers. The P450 aromatase is found in fat tissue, the location usually ascribed to the aromatization of androgens to estrogen; this enzyme has also been found in osteoblasts.(52) Vitamin D is among the hormones that regulate P450 aromatase.(53) Thus, vitamin D and its reception may influence the balance between androgens and estrogens in peripheral tissues, which in turn modulates the availability of steroid hormones for their receptors. We did evaluate total testosterone levels to determine whether these levels conformed to the strata defined by the genotypes, but our data do not indicate that 4-year average testosterone levels are different in strata defined by the combination of genotypes at different loci.

Because osteoporosis is likely to be a polygenic disease,(8) we expanded our study to include other bone-related proteins. Osteocalcin, osteonectin, and osteopontin were selected for analysis because they are the most abundant noncollagenous protein components of bone. We also included COLI A1 and COLI A2 because COLI comprises the major protein component of bone matrix. Alterations in COLI production and structure lead to osteogenesis imperfecta, an inherited brittle-bone disorder characterized by fractures, osteopenia, and abnormal bone matrix,(49) and several reports suggest a clinical overlap between osteogenesis imperfecta and osteoporosis.(10,11) Furthermore, transgenic mice with mutations in COLI genes have been shown to develop bone fragility,(54) some in an age-dependent manner.(55,56)

Our study did not identify a relationship between BMD and polymorphisms at any of the other genetic loci, including COLI A1 and COLI A2. These later findings are in agreement with the recent studies of Spotila et al.,(57) which failed to demonstrate linkage of either COLI A1 or COLI A2 in seven families with familial osteopenia. However, our findings contrast with those of Grant et al.(58) who reported an association between reduced BMD and a polymorphism at an Sp1 binding site in intron 1 of the COLI A1 gene. We did not use this marker in our study, but it would be difficult to explain a relationship between BMD and the Sp1 binding site, considering the lack of association with other intragenic markers. As with the association studies with VDR, there are conflicting results with the COLI A1 marker as well,(54–62) suggesting that the Sp1 polymorphism is predictive of BMD and/or fracture only in certain population groups.

In summary, ER genotypes were associated with baseline lumbar spine and total body BMD in this cohort. None of the other genotypes, including those for the VDR, COLI, osteocalcin, osteonectin, or osteopontin were associated with baseline BMD, change in BMD over time, or the serum markers of osteocalcin or 1,25(OH)2D. However, there was a significant gene-by-gene interaction effect with the ER genotypes and the VDR BsmI genotype. Our data suggest that genetic variation at the ER locus, singly and in relation to the VDR locus, may influence attainment and maintenance of peak bone mass in younger women and may provide insight into the biologic nature of the variation which underlies different susceptibilities to osteoporosis.

Acknowledgements

Support was provided by NIAMS RO1–40888 (Sowers, PI); NIAMS R55–43507 (Willing, Principal Investigator); and Medical Research Service of the Department of Veterans Affairs (Aron, PI). We gratefully acknowledge the laboratory technical support of Sachi Deschenes and Erik Roberts (Iowa City, IA, U.S.A.); the serum analyses for osteocalcin, 1,25(OH)2D, 25(OH)D, and PTH by Bruce Hollis (Medical College of South Carolina, Charleston, SC, U.S.A.); and reproductive hormone analysis by the Reproductive Sciences Program (University of Michigan, Ann Arbor, MI, U.S.A.).

Ancillary