Estrogen Receptor β (ESR2) Polymorphisms in Interaction With Estrogen Receptor α (ESR1) and Insulin-Like Growth Factor I (IGF1) Variants Influence the Risk of Fracture in Postmenopausal Women


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


In this large population-based cohort study, variants in ESR2 were associated with increased risk of vertebral and incident fragility fracture in postmenopausal women. Interaction of ESR2 with ESR1 and IGF1 was determined and revealed a deleterious genetic combination that enhances the risk of osteoporotic fracture.

Introduction: Osteoporosis is a complex disease with strong genetic influence, but the genes involved are ill-defined. We examined estrogen receptor β (ESR2) polymorphisms in interaction with estrogen receptor α (ESR1) and insulin-like growth factor I (IGF1) variants in relation to the risk of osteoporotic fracture, BMD, and bone geometry.

Materials and Methods: In the Rotterdam study, a prospective population-based cohort of elderly white individuals, we studied six single nucleotide polymorphisms (SNPs) in ESR2 (n = 6343, 60% women). We analyzed the genetic variants in the form of haplotypes reconstructed by a statistical method. Results refer to the most frequent ESR2 haplotype 1 estimated from two SNPs in intron 2 and the 3′-untranslated region (UTR). Outcomes included vertebral and incident nonvertebral fractures, BMD, and hip structural analysis (HSA). We also studied the interaction with (the most frequent) ESR1 haplotype 1 estimated from the PvuII and XbaI polymorphisms and an IGF1 promoter CA-repeat.

Results: Compared with ESR2 haplotype 1 noncarriers, female homozygous carriers had a 1.8- and 1.4-fold increased risk of vertebral and fragility fractures. HSA showed that ESR2 haplotype 1 homozygote women had 2.6% thinner cortices, 1.0% increased neck width, and 4.3% higher bone instability (buckling ratios). For testing the gene interaction, we assumed a recessive model of ESR2 haplotype 1. Female homozygous carriers of ESR2 haplotype 1 and noncarriers of ESR1 haplotype 1 had a 3.5- and 1.8-fold increased risk of vertebral and fragility fractures (pinteraction = 0.10). Such effects and interactions were stronger in women homozygous for the IGF1 192-bp allele, with 9.3-fold increased risk (pinteraction = 0.002) for vertebral and 4.0-fold increased risk (pinteraction = 0.01) for fragility fractures. Multilocus interaction analyses of fracture endured correction for multiple testing using Monte-Carlo simulations (pinteraction = 0.02 for vertebral and pinteraction = 0.03 for fragility fractures). Similar patterns of interaction were observed for BMD, cortical thickness, bone strength (section modulus), and instability (buckling ratio). In men, no such effects were observed.

Conclusions: Variants of ESR2 alone and in interaction with ESR1 and IGF1 influence the risk of fracture in postmenopausal women. These findings reinforce the polygenic and complex character of osteoporosis.


Osteoporosis is a systemic skeletal disease characterized by low BMD and microarchitectural deterioration of bone tissue, with a consequent increase in bone fragility and susceptibility to fracture. From a genetic perspective, osteoporosis is a multifactorial disease resulting from multiple independent gene effects and gene–gene and gene–environment interactions.(1,2) Association analysis of candidate genes is an efficient way to identify the modest but real genetic effects of individual polymorphisms.(3) Even though association studies have to deal frequently with problems of power, false positives, multiple testing, and lack of replication,(4) several variants in candidate genes have been shown to influence independently an individual's genetic susceptibility to fracture, and examples include the VDR,(5–7)COL1A1,(8)ESR1,(9,10)IGF1,(11)SOST,(12)LRP5, and LRP6(13) genes.

Estradiol is a pleitropic sex hormone that regulates many physiological processes, including normal cell growth, development, and tissue-specific gene regulation in the reproductive tract, the central nervous, and the skeletal system.(14) These biological actions are mediated by binding of estradiol to one of the two specific estrogen receptors: α (ERα) and/or β (ERβ). Estrogen receptors regulate gene expression by binding to their cognate response element or through protein–protein interactions with other transcription factors.(15) ERα seems to be the major receptor mediating estrogen action, and its gene (ESR1; 6q25.1) has been an obvious candidate to study in complex diseases. Variants in the ESR1 gene (XbaI, PvuII) are associated with increased susceptibility to cardiovascular disease in both sexes(16,17) and with osteoporotic fracture in women.(9,18) The relationship with fracture has been recently confirmed by a prospective meta-analysis of the GENOMOS consortium, the largest case collection for osteoporosis genetics yet reported.(10)

The discovery of ERβ(19,20) has caused a paradigm shift in the understanding of estrogen action.(21) The DNA-binding domain of ERβ shares high degree of sequence identity with that of ERα, consequently binding estrogen responsive elements with similar specificity and affinity. Co-expression of both receptors has been detected in a number of tissues, including among others, mammary gland, central nervous system, and bone.(22) Importantly, the interplay between the receptors (through heterodimerization) has been shown to result in inhibitory action of ERβ on ERα-mediated gene regulation.(15,22,23)

Variants in the estrogen receptor β gene (ESR2; 14q22–24) have been found to be associated with BMD in white(24–27) and Asian(28,29) populations. Most of these studies found a dinucleotide (CA) repeat polymorphism located in intron 5 (D14S1026), associated with BMD in both men and women. A recent study found three single nucleotide polymorphisms (SNPs; rs1256112 and rs3020444 in the 5′ and rs1152588 in the 3′-untranslated regions [UTRs]) associated with lumbar spine BMD in men.(24) Two other SNPs (rs1256031 in intron 2 and rs1256059 in intron 6) were also associated to femoral neck BMD in 723 men.(27) In the same study, the presence of the inferred haplotype allele C-23CA-T (from D14S1026, rs1256031, and rs1256059) was associated with lower femoral neck BMD in women.

To our knowledge, no studies have examined the relationship between variants in ESR2 and fracture risk. This and the conflicting associations with BMD across sexes led us to study in elderly individuals whether ESR2 plays a role in the genetic susceptibility to bone fracture and/or in the genetic control of BMD and bone geometry. Furthermore, because ESR2 modulates ESR1 transcriptional activity,(30) biological interaction can be expected between ESR2 and ESR1. Similarly, considering that the IGF-I receptor and ERα have common activation pathways and receptor cross-talk,(31) it can be expected that the ESR1/ESR2 interplay influences IGF1 (12q22–24.1) effects. In addition, we have shown previously how variants in ESR1 and IGF1 are associated independently with postmenopausal osteoporosis. In summary, we found that women who are noncarriers of the XbaI/PvuII haplotype 1 in ESR1 (with suggested higher estrogen receptor sensitivity) had lower risk of vertebral fracture,(9) whereas women homozygous carriers of the wildtype (192-bp) allele in a IGF1 CA-repeat promoter polymorphism had higher IGF-I levels,(32) lower risk of incident fragility fracture, higher BMD, and lower buckling ratios (bone instability).(11)

Therefore, we explored the role of variants in the ESR2 gene in relation to the risk of osteoporotic fracture and to bone parameters in a large prospective population-based cohort of elderly men and women and the interaction with ESR1 and IGF1 polymorphisms.


Study population

Subjects were participants of the Rotterdam Study, a large prospective population-based cohort study of elderly men and women ≥55 years of age. The study was designed to investigate the incidence and determinants of disabling chronic diseases in the elderly. In summary, all 10,275 inhabitants of Ommoord, a district in Rotterdam, The Netherlands, were invited to participate. Of those, 7983 participated at baseline (including extensive physical examination and BMD assessments at the research center performed between 1990 and 1993), bringing the overall response rate to 78%. Both the rationale and the design of the study have been described previously.(33) The Medical Ethics Committee of Erasmus University Medical School approved the Rotterdam Study, and participants provided written informed consent.

Clinical examination and measurements

Height and weight were measured in a standing position wearing indoor clothing without shoes. BMI was computed as weight in kilograms divided by height in meters squared (kg/m2). During the home interview, female participants were asked to recall their age at menopause, and responses were validated as described previously.(34)

Assessment of vertebral fracture

Both at baseline and at the follow-up visit, between 1997 and 1999, thoracolumbar radiographs of the spine were obtained. The follow-up radiographs were available for 3456 individuals, who survived on average 6.4 ± 0.4 (SD) years after the baseline center visit and who were still able to come to our research center. All follow-up radiographs were scored for presence of vertebral fractures using the McCloskey/Kanis method, as described previously.(35)

Assessment of incident nonvertebral fractures

The nonvertebral fracture analysis is based on follow-up data collected from baseline (1990–1993) until January 1, 2002, comprising an average follow-up period of 7.4 ± 3.3 (SD) years. Fracture events were reported either by general practitioners in the research area by means of a computerized system (covering 80% of the cohort) or through hospital records. Research physicians regularly checked participant information in the general practitioners' records outside the research area and independently reviewed and coded the information. Subsequently, for final classification, a medical expert reviewed all coded events. Site-specific incidence rates of fracture have been reported previously,(11,36) and in our population, only 44% of women and 21% of men with osteoporotic fracture had T scores of −2.5 or less.(36) We studied the relationship of the polymorphism to “all types” of nonvertebral fracture occurring in the elderly. Subsequently, we considered “osteoporotic” (all fractures except carpal, metacarpal, phalanxes, foot, face, skull, postprocedural, and pathological) and “fragility fractures occurring at older age” (including hip, pelvic, and proximal humerus fractures with mean age >75 years).

BMD and bone geometry measurements

BMD measurements (g/cm2) of the proximal femur were performed by DXA using a Lunar DPX-L densitometer (Lunar Radiation Corp., Madison, WI, USA) and analyzed with DPX-IQ v.4.7d software. Methods, quality assurance, accuracy, and precision issues of the DXA measurements have been described previously.(37) We used the hip structural analysis software developed by Thomas J Beck(38) to measure hip bone geometry from the DXA scans of the narrow-neck (NN) region across the narrowest point of the femoral neck. BMD, bone width (outer diameter), and cross-sectional moment of inertia (CSMI) were measured directly from mineral mass distributions using algorithms described previously.(38) In addition, estimates of cortical thickness and endocortical diameter were obtained by modeling the NN region as a circular annulus, which assumes a proportion of cortical/trabecular bone of 60/40. For this study, calculations of section moduli (Z), an index of bending strength, and buckling ratios, an index of bone instability, were slightly modified from those reported previously(38) to account for shifts in the center-of-mass. Z was calculated as CSMI/dmax, where dmax is the maximum distance from the center of mass to the medial or lateral surface. Buckling ratios were computed as dmax divided by estimated mean cortical thickness.

Identification of ESR2 polymorphisms and determination of genotypes

We identified 47 potential candidate polymorphisms in the ESR2 gene by consulting polymorphism databases (43 from dbSNP, 4 from Celera) and the literature (11 SNPs).(27–29,39–42) Six validated SNPs distributed across the 61-kb gene were finally selected for genotyping based on frequency (minor allele frequency [MAF] >10%) and location (Fig. 1A). Genomic DNA was isolated from peripheral leukocytes using standard procedures. Taqman allelic discrimination assay was used for genotyping, whereas primer and probe sequences were optimized using the SNP assay-by-design service of Applied Biosystems (Nieuwerkerk aan den IJssel, The Netherlands). Reactions were performed on the Taqman Prism 7900HT 384-well format using 2 ng genomic DNA in a 2-μl reaction volume. Genotyping of the PvuII and XbaI variants in ESR1 (with the same haplotyping methodology used for ESR2)(9) and the promoter microsatellite CA-repeat in IGF1 have been described in detail previously.(43)

Figure Figure 1.

Schematic representation of the (A) ESR2 gene structure (eight exons in gray, UTR elements in black) and localization of the six studied SNPs. The haplotype block structure determined in the Rotterdam Study (six SNPs in black including the two in gray used in the haplotype association study) is compared with that determined with HapMap Phase I (in gray) and Phase II (in black) data. (B) linkage disequilibrium (LD) map of ESR2 SNPs studied in 180 chromosomes and resulting in one major block of 62-kb span: values are pairwise D′ (left) and r2 (right) estimations.

Statistical analysis

We performed the analysis of this study in three stages. In the first stage, we searched for SNPs in the ESR2 gene, determined linkage disequilibrium (LD) of the SNPs, compared haplotype block structure, and reconstructed haplotypes using an statistical method from the selected polymorphisms in 180 chromosomes randomly selected from a blood donor bank. In the second stage, we analyzed the ESR2 haplotypes in relation to incident nonvertebral fractures (n = 1010 individuals with at least one fracture and 5407 individuals without fracture), vertebral fractures (n = 381 fractures identified by X-ray screening in 3456 individuals), and baseline BMD and bone geometry (n = 4418). In the third stage, we studied gene interactions between ESR2, ESR1, and IGF1 based on 6343 individuals (3787 women) with available genotypes in the three genes.

LD and haplotype analyses of ESR2:

Genotype frequencies of the SNPs were tested for Hardy-Weinberg equilibrium proportions using the Haploview V.3.2(44) program. We determined LD among the six selected SNPs in ESR2 using PHASE v.2.0 (University of Washington, Seattle, WA, USA)(45) and Haploview V.3.2(44) programs. LD is the essence of indirect association studies (when tested polymorphisms are not causal) and refers to the occurrence of allelic association at the population level.(46) Two of the most common measures of pairwise LD are Lewontin's D′, a measure used to identify regions with little recombination between alleles, and r2, the square of the correlation coefficient between alleles at two loci that determines directly the power of indirect association studies. They are complementary measures, considering that even when loci are in complete disequilibrium (D′ = 1), the pairwise r2 values can vary widely, because they are related to the allele frequencies and to the positions of the corresponding mutations in the ancestry.(46) Regarding haplotype block structure (i.e., HapMap), D′ and r2 describe the physical extent of the blocks and the haplotype diversity in them. HapMap Phase I and Phase II information was consulted to evaluate correspondence regarding the most common haplotype alleles across the gene. Haplotype alleles based on the intron 2 (rs1256031) and the 3′UTR (rs4986938) SNPs were determined, and all further analyses are presented for the most frequent ESR2 haplotype 1 (CintCutr).

Association analyses:

Means and SDs were computed for all measurements and compared with those of the same sex in the complete Rotterdam Study using t-tests. To estimate the risk of vertebral fracture by ESR2 haplotype 1 genotypes, ORs with 95% CIs were calculated using (multiple) logistic regression models (because the follow-up time to occurrence of vertebral fracture was uncertain). For the analysis of incident nonvertebral fractures, we estimated incident rates by ESR2 haplotype 1 carrier status. For the time-to-event (or fracture-free survival) analysis, we specified age (in years) as the underlying time variable instead of follow-up time to analyze the effect of chronological age on first incident fracture. To do this, we took into account delayed entry (left truncation) by using the counting process notation of S-PLUS V.6.0. We estimated cumulative probabilities of fracture events after 55 years of age using the Kaplan-Meier method. Crude and adjusted hazard ratios were estimated using Cox proportional-hazards models. The assumption of proportionality of hazards was verified for all covariates. Multiple linear regression and analyses of covariance (ANCOVA) were used to model as fixed effect the relation of ESR2 haplotype 1 carrier status with bone geometry parameters, adjusted for age, height, and weight at baseline. We assumed a recessive effect and compared homozygous carriers of ESR2 haplotype 1 to the combined group of heterozygous and noncarriers. Finally, model assumptions were verified, and model residuals were checked for goodness-of-fit. Significance of p values was set at the 0.05 level. If not stated otherwise, SPSS 11.0 (SPSS, Chicago, IL, USA) was used for the statistical analyses.

Gene interaction analyses:

For the gene interaction analyses, we focused on the occurrence of vertebral and “fragility” fractures occurring at older age (including hip, pelvic, and proximal humerus fractures), BMD, and geometry analyses, in line with the associations observed previously in women for ESR1(9) and IGF1(11) variants. For the analysis of incident fragility fractures, relative risks with 95% CIs were estimated from Cox proportional hazards models as hazard ratios (HRs) and for vertebral fracture as ORs with 95% CIs from multiple logistic regression as described above. Differences in continuous measurements were compared across genotype groups (modeled as fixed effects) using multiple linear regression and ANCOVA. All estimates were adjusted for age, height, and weight. We stratified all interaction analyses by sex and additionally by IGF1 genotypes when we studied the interaction between the estrogen receptor genotypes in relation to IGF1 genotypes. Genetic interaction between ESR2 and ESR1, and the three-gene interaction (of ESR1, ESR2, and IGF1) were modeled assuming a recessive genetic effect for ESR2 haplotype 1 and additive genetics effects of ESR1 haplotype 1 and IGF1 192-bp alleles with an interaction term obtained from the product of fixed main (genotype) effects. Two-way (ESR2recessive × ESR1addititve) and three-way (ESR2recessive × ESR1additive × IGF1additive) interactions were tested, and significance values set at pinteraction = 0.10. Additional multilocus analysis was used to test the three-way interactions in relation to the presence of fracture using the program FAMHAP.(47) FAMHAP implements an expectation-maximization algorithm to estimate the maximum-likelihood haplotype frequencies (for ESR1 and ESR2 variants) and performs a likelihood ratio test across fractures cases and noncases. To obtain simulated empirical p values (more accurate than those obtained from asymptotic theory), FAMHAP permutes the disease status of the individuals for each replicate such that the number of cases and controls are the same as in the original sample. For each p value, one million replicates were simulated. The p value equals the fraction of replicates that yield a test statistic ≥ to that calculated for the actual data. In addition, FAMHAP allows correction for multiple testing by correcting the significance of the marker combination with the strongest association under the global hypothesis that none of the tested interactions is significant.(48)


LD and haplotype analyses

The structure of the ESR2 gene and the localization of the studied SNPs are shown in Fig. 1A. The information provided by the five SNPs analyzed suggested the presence of a unique haplotype block spanning the ESR2 gene. HapMap–Phase I information (available during the period of selection of the 5 SNPs analyzed in the Rotterdam Study presented as gray lines with 20 shown SNPs) identified one long haplotype block spanning most of the gene with a possible small block in the 5′UTR region. HapMap–Phase II (release 19) information (presented as black lines with 47 shown SNPs) identified three main haplotype blocks with the external blocks spanning up- and downstream beyond the region of the ESR2 gene. As shown in Fig. 1B, the rs1256031 SNP, located 692 bp downstream from the start of exon 2, was in high LD with rs1271572 (D′ = 0.98 and r2 = 0.89), rs1256030 (D′ = 1.00 and r2 = 0.98), and rs1256049 (D′ = 1.00 and r2 = 0.01) and somewhat lower LD (D′ = 0.93, r2 = 0.52) with SNP rs4986938, located 38 bp downstream the 3′UTR. The rs1952586 and rs1256049 SNPs had MAF <10% and low pairwise r2 values, reflecting high uncertainty for haplotype determination. Based on the LD results, the haplotype block structure and the previously reported associations, we selected the intron 2 (rs1256031) and the 3′UTR (rs4986938) SNPs for final analysis in the Rotterdam Study in the form of haplotype alleles. Both SNPs are T/C substitutions and their genotype frequencies were in Hardy-Weinberg equilibrium proportions. Haplotype alleles were coded as numbers 1 through 4 in order of decreasing frequency in the population: 1 = CintCutr (f = 45.1%), 2 = TintTutr (f = 37.1%), 3 = TintCutr (f = 17.1%), and 4 = CintTutr (f = 0.7%) (Fig. 1A). Further results are presented for genotype groups based on the presence of (the most common) haplotype 1: noncarriers (f = 31%), heterozygous carriers (f = 49%), and homozygous carriers (f = 20%).

Baseline characteristics

Table 1 shows the baseline characteristics of the reference study population (n = 7983) and the two subsets of genotyped individuals of the fracture analyses. Compared with the total study population, genotyped individuals followed for the occurrence of nonvertebral fractures were on average 1.1 years younger, whereas individuals screened for the presence of vertebral fractures were on average 4.8 years younger. The use of hormone replacement therapy was low (2.3–3.5%) in all of the study population.

Table Table 1.. Baseline Characteristics of the Study Population
original image

Table 2 shows age, anthropometric characteristics, BMD, and bone geometry parameters for the ESR2 haplotype 1 carrier status at baseline. No differences were seen in body height, weight, BMI, or (femoral neck or vertebral) BMD, but we observed differences for parameters of hip bone geometry with evidence for a recessive effect. Female homozygous carriers had 1.0% greater femoral neck width (p = 0.008), 2.6% thinner cortices (p = 0.004), and 4.3% increased buckling ratio (p = 0.0002) compared with noncarriers and heterozygotes combined. In men, no differences between ESR2 genotypes were observed.

Table Table 2.. Characteristics of the Study Population by ESR2 Haplotype 1 Carrier Status
original image

Association with the risk of fracture

Table 3 shows the risk of fracture by carrier status of the ESR2 haplotype 1 allele in women. Homozygous carriers of haplotype 1 had 1.8 (95% CI, 1.2–2.6) increased risk of vertebral fracture with indication of a recessive genetic effect. Further adjustment for baseline lumbar spine BMD and/or vertebral area did not essentially change these results. In the analysis of incident nonvertebral fractures, homozygous carriers of the ESR2 haplotype 1 allele had 20% increased risk of all-type and osteoporotic-type of fractures and 40% increased risk of fragility fracture compared with “noncarriers and heterozygotes” combined. Adjustment for BMD, hip bone geometry, height, and/or weight measured at baseline did not essentially change these risk estimates. No increased risk of vertebral or nonvertebral fracture was seen in men (data not shown).

Table Table 3.. Risk of Fracture by ESR2 Haplotype 1 Carrier Status in Women
original image

Interaction between ESR2, ESR1, and IGF-I

Figure 2 shows the risk of both vertebral (left) and fragility fracture (right) in women for the combination of haplotype 1 genotypes in both estrogen receptors. The reference group for the six estrogen receptor genotype groups is constituted by “non- and heterozygous carriers” of haplotype 1 in ESR2 who are noncarriers of haplotype 1 in ESR1. Borderline significant interaction (pinteraction = 0.10) between ESR2 and ESR1 was seen for the risk of vertebral fracture (Fig. 2, left). In line with our previous report on ESR1,(9) increased risk of vertebral fracture is consistently seen for women homozygous carriers of ESR1 haplotype 1 (white bars), ranging from 2.1 (95% CI, 1.2–3.6) increased risk in “non- and heterozygous carriers” of ESR2 haplotype 1 (n = 400) to 3.0 (95% CI, 1.5–6.0) in the combined group of homozygous carriers of haplotype 1 in both ESR2 and ESR1 (n = 103). Noncarriers of ESR1 haplotype 1 in (black bars) have no increased risk of vertebral fracture, except when they are homozygous carriers of ESR2 haplotype 1 (n = 88), who have an unexpected 3.3 (95% CI, 1.6–6.9) increased risk of vertebral fracture. When we stratified the vertebral fracture analysis by IGF1 genotypes (Fig. 3A), a similar but more pronounced pattern of ESR2 × ESR1 interaction was observed only in women homozygous for the IGF1 192-bp allele (pinteraction = 0.007). In this IGF1 stratum, the risk of vertebral fracture in women homozygous for ESR1 haplotype 1 was similarly increased across the two ESR2 strata (OR = 3.6 in n = 182 and OR = 3.5 in n = 45). However, noncarriers of haplotype 1 in ESR1 had 9.3 (95% CI, 3.1–28.0) increased risk of vertebral fracture when they were also homozygous carriers of haplotype 1 in ESR2 (n = 32). Genotype effects in the other two IGF1 strata showed the expected additive increase in the risk of fracture with the presence of the ESR1 haplotype 1 allele and augmented in ESR2 haplotype 1 homozygote women. However, risk estimates also increased in heterozygotes and noncarriers of the IGF1 192-bp allele. The three-gene in teraction (ESR2recessive × ESR1additive × IGF1additive) was significant for the risk of vertebral fracture (pinteraction = 0.03).

Figure Figure 2.

Interaction between ESR2 and ESR1 genotypes in women for the risk of vertebral (pinteraction = 0.10) and fragility (pinteraction = 0.10) fracture. Women noncarriers of ESR1 haplotype 1 previously shown not to have increased risk of fracture(8) have increased risk of fracture when they are homozygous carriers of haplotype 1 in ESR2 (black bars with arrows). Significant risk ratios (95% CIs) are shown above the bars, and group sample size is on top of the bars.

Figure Figure 3.

Interaction between ESR2 and ESR1 genotypes for the risk of vertebral and fragiltity fracture across IGF1 192-bp allele genotypes in women. The interaction between ESR1 and ESR2 is stronger in the strata of women homozygous for the IGF1 192-bp allele (previously shown to have reduced fracture risk).(10) In this stratum, women noncarriers of ESR1 haplotype 1 (previously shown to have reduced risk of fracture)(8) who are also homozygous carriers of haplotype 1 in ESR2 (black bars with arrow) have even further increased risk of fracture than in the overall population. Significant risk ratios (95% CIs) are shown above the bars and group sample size is on top of the bars. (A) Interaction for the risk of vertebral fracture was significant only in IGF1 192-bp homozygotes (pinteraction = 0.007). (B) Interaction for the risk of fragility fracture was significant only in IGF1 192-bp homozygotes (pinteraction = 0.01).

In Fig. 2 (right), we also showed a borderline significant interaction (pinteraction = 0.10) between ESR2 and ESR1 for the risk of fragility fractures. Women ESR2 haplotype 1 homozygotes, which are noncarriers of haplotype 1 in ESR1 (n = 158), had 1.8 (95% CI, 1.0–3.0) increased risk of fragility fracture, whereas no other combination of genotypes showed significantly increased risk. When we examined this effects over the risk of fragility fracture across IGF1 genotypes (Fig. 3B), we found a significant interaction between the ESR2 and ESR1 haplotype 1 genotypes (pinteraction = 0.01) only in the group of women homozygous for the IGF1 192-bp allele. The risk for fragility fracture in ESR2 haplotype 1 homozygous women was increased to 4.0 (95% CI, 1.8–8.8) when they were noncarriers of haplotype 1 in ESR1 (n = 59). No increased risk was seen for the combined homozygous carriers of haplotype 1 in both ESR2 and ESR1 genes (n = 96). No such significant interactions were observed in the other IGF1 strata of women heterozygous for or noncarriers of the IGF1 192-bp allele. In the stratum of noncarriers of the IGF1 192-bp allele, a significant increased risk of 4.8 (95% CI, 2.0–11.2) was detected in ESR2 haplotype 1 homozygous women who were heterozygous for ESR1 haplotype 1 (n = 40). The three-gene interaction (ESR2recessive × ESR1additive × IGF1additive) was also significant (pinteraction = 0.02) for the risk of fragility fracture.

Considering the multiple gene interactions and genetic models tested, we performed a multilocus analysis of the five markers in the three genes (ESR2: intron 2 and 3′UTR, ESR1: PvuII and XbaI, and IGF1: promoter CA-repeat polymorphisms) using the simulation-based test of association implemented in the FAMHAP program. The corrected p values for interaction adjusted for multiple testing were 0.02 and 0.03 for the risk of vertebral and fragility fracture, respectively. The most significant marker combinations resulted between ESR2 intron 2 and ESR1 PvuII for the risk of vertebral fracture (p = 0.002) and between ESR2 intron 2, ESR1 haplotype 1 (PvuII/XbaI), and the IGF1 promoter polymorphism for the risk of fragility fracture (p = 0.003).

No relationship between such combination of genotypes and the risk of vertebral and/or fragility fracture was seen at all in men.

The interaction between ESR2 and ESR1 haplotype 1 genotypes was also examined in relation to skeletal properties influencing the risk of fragility (hip structural analysis) and vertebral fractures. In the total female study population (Table 4), the gene interaction was significant for BMD and cortical thickness (pinteraction = 0.009), bone strength/section modulus (pinteraction = 0.01), and bone instability/buckling ratio (pinteraction = 0.07). Women who were homozygous carriers of ESR2 haplotype 1 and noncarriers of ESR1 haplotype 1 (n = 107) showed 7% lower femoral neck BMD, 7.5% thinner cortices, 1.5% increased neck width, and 2.3% endocortical diameter (Fig. 4, top) than in the reference group of “non- and heterozygous carriers” of ESR2 haplotype 1 and noncarriers of ESR1 haplotype 1 (n = 420). Such thin cortices in relative wider diameters resulted in 9.6% higher bone instability (as estimated by the buckling ratio), with 6.5% lower bending strength (as estimated by the section modulus) of the femoral neck. Similar significant differences but of lesser magnitude were also seen in the double homozygous carriers of haplotype 1 in both ESR2 and ESR1 (n = 160). In the group of women homozygous for the IGF1 192-bp allele (Table 5; Fig. 4, bottom), differences in skeletal parameters are accentuated in ESR2 haplotype 1 homozygotes, when they are noncarriers of ESR1 haplotype 1 (n = 107). In this strata of women homozygous for the IGF1 192-bp allele, we observed a significant interaction between ESR1 and ESR2 for BMD and cortical thickness (pinteraction = 0.004), bone strength/section modulus (pinteraction = 0.02), and bone instability/buckling ratio (pinteraction = 0.02). We found no evidence of a gene interaction or significant differences across genotypes groups in those skeletal parameters related to the risk of vertebral fracture. However, in women homozygous for the IGF1 192-bp allele (Table 5), ESR2 haplotype 1 homozygotes who are noncarriers of ESR1 haplotype 1 (n = 41) had 5.3% significantly lower vertebral BMD. Again, no such findings were observed in men.

Table Table 4.. Skeletal Parameters by ESR2 and ESR1 Genotypes in Women
original image
Table Table 5.. Skeletal Parameters by ESR2 and ESR1 Genotypes in IGF1 192-bp Homozygous Women
original image
Figure Figure 4.

Schematic presentation (not in scale) of skeletal properties (hip structural analysis) influencing the risk of bone fragility in the total study population (top; n = 2449) and in women homozygous for the IGF1 192-bp allele (bottom; n = 1053). Values are expressed as relative percent difference from reference. Thinner cortices in wider circumferences represent higher bone instability as observed in women homozygous carriers of ESR2 haplotype 1 and noncarriers of ESR1 haplotype 1.


This large population-based study showed for the first time that white postmenopausal women who are homozygous for a common ESR2 haplotype allele have 40–80% increased risk of fragility and vertebral fracture. Associations with other osteoporosis outcomes were consistent with these findings on fracture and showed significant (yet modest) effects. Women who are homozygous carriers of the ESR2 risk haplotype allele had significantly wider femoral necks, thinner femoral neck cortices, and higher femoral cortical instability. We further showed an interaction between ESR2 and ESR1 in relation to the risk of vertebral and fragility fracture in women. Noncarriers of the ESR1 risk haplotype 1 have no increased risk of either type of fracture. However, if noncarriers of the ESR1 risk haplotype 1 are in addition homozygous carriers of the ESR2 risk haplotype 1, the relative risk is 3.3 and 1.8 for vertebral and fragility fracture, respectively. This interaction was stronger in women homozygous for the IGF1 192-bp allele, where the risk of vertebral and fragility fracture rose to 9.3 and 4.0 times increased risk, respectively. The haplotype interaction analyses of fracture endured correction for multiple testing using Monte-Carlo simulations (pinteraction = 0.02 for vertebral and pinteraction = 0.03 for fragility fractures). Furthermore, other osteoporosis-related outcomes including femoral neck BMD, cortical thickness, bone strength, and cortical bone instability also showed such consistent significant interaction between ESR1, ESR2, and IGF1. No such effects were observed in men.

ESR2 is a relatively small gene (encompassing 61 kb) with high LD between the polymorphisms identified thus far in the region. Even though the SNPs we have genotyped in ESR2 are unlikely to be functional, the haplotype block structure descriptions in the HapMap project (Fig. 1A) confirmed the very high LD across the ESR2 region we also observed in our study. In fact, the three LD blocks identified with the extensive Phase II HapMap data show that the blocks have very high LD between them (multimarker D′ estimations exceeding 0.94). This way, the haplotype analyses we performed in our study possibly captured a large proportion of the common variation in the ESR2 gene. Because the r2 measured can be considered as a standard measure of whether the LD between two markers is sufficient for detecting associations (indicator of power to detect the causal genetic variant using LD), we can conclude that causal common variants between the intron 2 (rs1256031) and 3′UTR (rs4986938) polymorphisms (r2 = 0.92) should be detected by our haplotypes. Considering the high LD (D′ = 0.95 and r2 = 0.89) between the intron 2 SNP (rs1256031) and the 5′ promoter SNP (rs1271572), this LD possibly extends at least 15.7 kb upstream of intron 2 of the ESR2 gene. We focused our results on the analysis of the most frequent haplotype 1 because it is the more stable haplotype across the depicted LD blocks in Phase I and II HapMap data. The influence of the other haplotypes and of other rare and uncommon (MAF < 10%) variants determining susceptibility to disease are probably not possible to identify using LD mapping with these two SNPs alone. Future studies of ESR2 in our population will target genotyping additional SNPs to increase resolution and power to identify less frequent variants.

In addition to ESR2, ESR1 and IGF1 are known susceptibility genes for osteoporosis.(9,11) Previously, we have shown that carriers of ESR1 haplotype 1 seem to be less sensitive to estrogen, probably because of differential expression of the receptor. This was suggested by some functional studies(17,49) and by the multiple associations already documented for this gene variant. For example, body height in pre- and postmenopausal women(50) and estradiol levels in premenopausal women were lower,(51) whereas age at menopause was higher,(34) with the number of copies of ESR1 haplotype 1 in their genotype. Similarly, we have shown previously that the presence of the 192-bp allele in the IGF1 promoter genotype is associated with higher circulating IGF-I levels(32) and with increased body height in postmenopausal women(11,32) with evidence for an allele–dose effect. However, the XbaI and PvuII variants in ESR1 and the (microsatellite) CA-repeat promoter polymorphism of the IGF1 gene have no known functional effect on ERα (expression) and IGF-I production and activity. Thus, further genetic and functional studies to identify additional polymorphisms and the LD structure across these genes are needed to determine how these variants (or others in LD) may interact at the molecular level and contribute to explain our findings at the population level.

The interaction we report between variants in ESR2, ESR1, and IGF1 is plausible from a biological perspective. ERα and the IGF-I receptors share common signaling through the mitogen-activated protein kinase and the phosphoinositide 3 kinase pathways.(52) In addition, ESR1 is needed for some IGF-I–induced responses.(31) Similarly, it has been shown that ESR1 can interact in an indirect manner with some promoters by physically associating with prebound AP1 or SP1 protein complexes, mechanisms that seem to explain how estrogens regulate the cyclin D, c-Myc, and IGF1 promoters.(53) Several studies have shown the regulatory action that ESR2 exerts over ESR1 transcriptional activity and the ability of ESR2 to heterodimerize with and inactivate ESR1. These findings have positioned ESR2 as the dominant regulator of estrogen signaling.(30,54)

This study also showed, contrary to expectation, that a deleterious combination of genotypes does not necessarily arise from the combination of “risk” genotypes. Only one “risk” genotype in the presence of other “favorable” genotypes may result in an even more detrimental outcome—in this case the risk of fracture/osteoporosis. Such a finding highlights not only the multigenic nature of osteoporosis, but also depicts the complexity that should be considered in the genetic studies of multifactorial diseases. In fact, the lack of accounting for gene–gene interactions has been widely suggested to explain difficulties in replicating significant associations across studies.(55)

In this complex perspective of genetic interaction, the combination of ESR2 haplotype 1 homozygotes who are noncarriers of ESR1 haplotype 1 may reflect women with higher (ESR1-mediated) sensitivity to estrogen with apparent compromised regulatory capacity of ESR2 on bone. In addition, individuals with the 192/192 genotype of the IGF1 promoter polymorphism seem to have the optimum configuration for IGF1 transcription, higher IGF-I levels, and potentially higher anabolic activity on bone, which could explain why the effects of such a genotypic combination of estrogen receptors is exacerbated in women who are (in addition) homozygous for the 192-bp allele of the IGF1 promoter polymorphism. Even though we saw an isolated large increase in the risk for fragility fracture in the stratum of noncarriers of the IGF1 192-bp allele, who were ESR1 haplotype 1 heterozygotes and ESR2 haplotype 1 homozygotes (Fig. 3B), no evidence of statistical interaction was observed in this group. Nevertheless, we can not conclude this is a spurious finding because the group of IGF1 192-bp allele noncarriers is a combination of multiple alleles, one of which could explain the observed increased risk.

However, how could such combination of genotypes be so detrimental to bone health, in particular when noncarriers of ESR1 haplotype 1 and carriers of IGF1 192-bp alleles have lower risk of fracture? In women, the presence of estrogen since the onset of puberty limits periosteal apposition and favors bone formation in the inner border of the cortices (endocortical apposition).(56) This process has been postulated (by Fuller Albright in the early 1940s) to be an evolutionary response of the female skeleton to pack extra bone mineral that is easily available for the needs of pregnancy and lactation.(57) Such an effect of estrogen is thought to be achieved by altering the mechanostat in cells around the bone marrow.(58) It has also been proposed that this process is mediated by specific action of both its receptors, with ERα favoring endocortical apposition and ERβ restraining periosteal apposition, the latter functioning as an “antimechanostat.”(59) Such estrogen-mediated restrain on bone expansion disappears after menopause, when women undergo a rapid phase of cortical thinning (consequence of endocortical expansion) with accompanying increase in bone volume (caused by periosteal apposition).(60,61) If the process of cortical thinning and bone expansion is carried too far, the section modulus no longer governs bending strength, because the cortex and the trabeculae that internally support it become locally unstable and prone to buckle and fracture.(38,56,60,62,63)

Our results on the risk for fragility fracture and the findings on hip bone geometry at baseline could support such a biomechanical and endocrinologic framework based on the principle of inadequate regulation of ERβ in female homozygous carriers of ESR2 haplotype 1. Greater bone instability (higher fracture susceptibility) will occur, especially in women with enhanced cortical thinning and bone expansion arising from sensitive ERα− and IGF-I–mediated stimuli. This concept of local bone instability (buckling)(63) is consistent with our observation that women who are homozygous carriers of ESR2 haplotype 1 were more likely to suffer fragility fractures than noncarriers. Similarly, our interaction analysis (Fig. 4; Table 4) showed that female homozygous carriers of ESR2 haplotype 1 and noncarriers of ESR1 haplotype 1 had greater endocortical diameters (cortical thinning) and neck widths (bone volume), with lower bending strength (section modulus) and greater bone instability (buckling ratios). If, in addition, women of this group were homozygous for the IGF1 192-bp allele (Fig. 4; Table 5), bending strength (section modulus) was further decreased and cortical instability was further increased (buckling ratios). Nevertheless, because adjustment for BMD, bone geometry, or vertebral bone area at baseline did not modify the genotype effect on risk of fracture, these geometric parameters measured at baseline are clearly neither a necessary nor a sufficient cause to explain the genotype-dependent fragility fracture risk. Even though such parameters are strongly correlated with the risk of fracture at the phenotypic level, there is limited genetic correlation with the risk of fracture.(64) In addition, because the risk of fracture is not only determined by bone properties, other extrinsic factors like increased risk of falling could also be genetically determined (i.e., sarcopenia, cognition, and neurological integrity) and could help explain the observed increase of fragility fracture. We did not find evidence of effect modification with age, but this should not be completely disregarded because it could be the result of limited power for subgroup analysis given the high number of genetic groups.

The absence of ESR2 effects and/or interactions observed in men could represent the higher estradiol levels elderly men have compared with postmenopausal women of similar age. Alternatively, periosteal apposition in men is a gradual life-long process that confers greater bone strength and seems to explain the lower incidence of bone fracture in men.(56) Even though associations of variants in ESR2 with BMD have been described in men,(24,27) androgens (by themselves or in interplay with estrogen receptors and growth factors) seem to be the main regulators of periosteal apposition in men and will be the target of future studies. Sex-specific genetic effects of autosomal genes like those in our findings on ESR2 in women or the case of LRP5 and LRP6 in men(13) show the sex-specific and complex genetic architecture of osteoporosis traits.(65) On the other hand, we cannot exclude that sex differences could reflect lack of power, considering the modest genotypic effects and the fact that the number of fractures that occur in men is considerably lower than in women.

Given the high number of statistical interactions that have been tested across the multiple genotypic combinations, multiple testing could play a role in our results. Nevertheless, our multilocus analysis of the variants in the three genes using Monte-Carlo simulations as implemented in the software FAMHAP(47) endured correction for multiple interaction testing(48) in relation to the risks of vertebral and fragility fracture. In addition, the consistency in the occurrence and directions of (significant) findings for all these osteoporosis-related outcomes makes it unlikely that our results could be explained by chance alone. Nevertheless, single study association designs like ours can not completely distinguish false-positive findings from true association,(4) so replication in other population-based studies is warranted.

In summary, this large population-based study in elderly men and women ≥55 years of age has shown that the presence of the ESR2 haplotype 1 influences the risk of osteoporotic fracture in postmenopausal white women. Interlocus interaction of variants in ESR2 with variants of the ESR1 and IGF1 genes suggest that a deleterious genetic combination for bone health may arise from a risk genotype in ESR2 that compromises estrogen response regulation, particularly in the presence of higher (ESR1-mediated) estrogen receptor sensitivity and enhanced genetically determined IGF-I activity. These findings provide some more insight into the etiology of bone fracture, but most importantly, they reinforce the polygenic and especially the complex character of osteoporosis.


This study is supported by the Netherlands Organization of Scientific Research (NWO)-Research Institute for Diseases in the Elderly (Grant 014-90-001; RIDE) and the European Commision (Grant QLK6-CT-2002-02629; GENOMOS). The authors are very grateful to the participants of the Rotterdam study and to the DXA and radiograph technicians, L Buist and HWM Mathot. Furthermore, we acknowledge all participating general practitioners and the many field workers in the research center in Ommoord, Rotterdam, The Netherlands.