Follicle-stimulating hormone is associated with low bone mass in postmenopausal women

Abstract

Summary

We evaluated the influence of two endogenous hormones on bone health in older women. Higher FSH was associated with bone disease, especially in lower estradiol environments. FSH attenuated the relationship between estradiol and bone. This may provide a mechanism through which future clinical research intervenes on bone loss.

Introduction/purpose

Despite preclinical evidence for an inverse association of follicle-stimulating hormone (FSH) and bone mineral density (BMD), no large epidemiologic studies have evaluated the separate and joint influences of FSH and estradiol on bone in postmenopausal women.

Methods

In a cross-sectional study of 675 postmenopausal women, we evaluated associations of serum FSH and dual X-ray absorptiometry (DXA)-classified areal BMD as well as low bone mass or osteoporosis (T-score < − 1.0) of the femoral neck and total hip. We stratified this analysis by serum estradiol (cut at the median). We tested whether FSH mediates the association of estradiol and BMD using the Sobel test.

Results

In linear regression models, there was a significant inverse association of serum FSH with both femoral neck and total hip BMD (both p < 0.01) when adjusted for age, hormone therapy (HT) use, and diabetes. In fully adjusted logistic regression models, women in the highest FSH tertile had higher odds of low bone mass/osteoporosis at the femoral neck (OR = 2.98; 95% CI = 1.86–4.77) and at the total hip (OR = 1.74; 95% CI = 1.06–2.84) compared to those in the lowest FSH tertile. We report evidence of effect modification by estradiol in stratified models and an interaction term. FSH met all criteria of a mediator, including an estimated 70% attenuation of the estradiol-BMD relationship (Sobel p value < 0.001).

Conclusions

FSH is associated with higher odds of having low bone mass/osteoporosis even after accounting for HT use. FSH is a mediator of the relationship between estradiol and BMD in healthy postmenopausal women. Larger, prospective studies of FSH concentrations and bone health are needed.

Introduction

Low bone density is a major source of morbidity and mortality in the United States (US) and globally. As the US population ages, the disease burden of osteoporosis is expected to increase [1]. Currently, age-related bone loss is the primary risk factor for bone fracture. In elderly women, the incidence of fractures is greater than that of heart attack, stroke, and breast cancer combined [1].

In women, the menopausal transition is accompanied by a period of decline in bone density concurrent with changing concentrations of circulating endogenous hormones [2, 3]. While changes in circulating estrogen are recognized as important determinants of bone loss, more recent animal-based work supports an estrogen-independent role of follicle stimulating hormone (FSH) in bone maintenance [4–6]. Epidemiologic evidence is conflicting; however, some studies support an inverse association of FSH and bone mineral density (BMD) [7, 8] and others report no association [9, 10].

In postmenopausal women, lower circulating estrogen results in higher FSH secretion via direct action on the pituitary gland [2, 3]; circulating serum FSH and estradiol concentrations are inversely correlated. This in conjunction with the hypothesized estrogen-independent mechanism through which serum FSH acts on bone led us to hypothesize that some of the bone loss attributed to diminished estrogen concentration may in fact be mediated by elevated FSH. No prior studies have evaluated the role of FSH as a mediator in the estradiol-bone relationship.

We evaluated the association of serum FSH concentration with dual X-ray absorptiometry (DXA)-derived areal BMD of the femoral neck and total hip as well as low bone mass and osteoporosis (defined by T-scores) in a cross-sectional study of postmenopausal women with consideration of estradiol concentration. We also examined FSH as a mediator in the estrogen-BMD association.

Methods

Study sample

The Buffalo OsteoPerio Study is an ancillary study to the Women’s Health Initiative Observational Study (WHI OS). The OsteoPerio Study focused on the prevalence, incidence, and progression of osteoporosis and periodontal disease in postmenopausal women [11, 12]. As depicted in Fig. 1, 2249 postmenopausal women were enrolled into the Observational Study of the Women’s Health Initiative at the University at Buffalo from 1993 to 1998. We recruited from this cohort of healthy women into the Buffalo OsteoPerio Study between 1997 and 2000, approximately 3 years after initial WHI OS enrollment. Selection was not based on prevalent low bone mass, osteoporosis, or periodontal disease. Women were excluded from enrolling into the OsteoPerio Study if they had fewer than six teeth, a bilateral hip replacement, a history of bone disease other than osteoporosis, a history of cancer within the previous 10 years, or treatment for serious illness. In total, 1362 women enrolled and completed the baseline visit [11].

Fig. 1.

Flow chart of participant selection

The OsteoPerio Study was designed to assess the relationship between osteoporosis, periodontal disease, and oral bone loss. It included a comprehensive oral examination and assessment of systemic bone density. After about 400 women were enrolled, funding to add a blood collection was secured. As such, blood samples were available on the remaining OsteoPerio Study participants. Of those, 675 women also had a DXA scan at baseline. The Buffalo OsteoPerio Study was reviewed and approved by the Health Sciences Institutional Review Board at the University at Buffalo, and all participants provided informed written consent.

Blood measures

Serum estradiol (E₂) concentration was measured by competitive electrochemiluminescence immunoassay (Roche E Modular system, Roche Diagnostics, Indianapolis, IN). The lowest detection limit (LOD) of the E₂ assay is 5 pg/mL [11]. Measures of FSH were analyzed using a sandwich electrochemiluminescence immunoassay on the Roche Cobas 6000 system (Roche Diagnostics, Indianapolis, IN). The LOD of the FSH assay is 0.10 mIU/mL. Both assays are FDA approved for clinical use.

Twenty-seven pooled quality control (QC) serum samples were included and tested along with participant samples and resulted in an interassay coefficient of variation (CV) of 4.1% for E₂ and 3.3% for FSH. One participant had serum FSH above the upper LOD (the sample was not re-run with dilution), who was excluded from the analysis. For participants with serum E₂ concentration below the lower LOD (n = 154), the midpoint between zero and the lower LOD was imputed (2.5 pg/mL).

Bone mineral density measurements

All participants underwent a dual-energy X-ray absorptiometry (DXA) scan using fan-beam bone densitometer (Hologic 4500A, Waltham, MA) [11] according to standardized protocol by a trained and certified DXA technician at the baseline visit. The DXA scan included measures of total body areal bone mineral density (BMD), as well as fat mass. BMD was also determined for the lumbar anteroposterior (AP) spine (L1–4), the femoral neck, the total hip and at the forearm. DXA scans estimate BMD in grams/cm². The BMD values were converted to T-scores for each skeletal site individually [11]. For these analyses, T-scores [13] were used to classify participants as having normal bone mass (T-score > − 1.0), low bone mass (T-score between − 1.0 and − 2.4), or osteoporosis (T-score ≤ − 2.5) at two sites separately: the femoral neck and the total hip. We did not include other skeletal sites in these analyses.

As part of DXA quality control, twenty scans were repeated after at least a 1-month interval to determine technician analysis error. The coefficients of variation for these were < 1% for duplicate BMD measurements.

Anthropometry

Height was measured to the nearest 0.1 cm using a fixed wall-mounted stadiometer. Weight was measured to the nearest 0.1 kg on a calibrated balance bean scale. BMI was calculated as weight (kg)/height (m²).

Demographics, lifestyle, and medication

Demographic information and potential confounding variables including age, self-identified race and ethnicity, education, marital status, smoking habits, menopausal hormone therapy (HT) use, and medical history were collected via self-administered questionnaires [11]. Dietary intake was collected using a food frequency questionnaire (FFQ) [11].

Statistical analysis

Demographic variables, anthropometric measures, and serum FSH and E₂ measures were characterized according to DXA-classified normal bone mass, low bone mass, and osteoporosis of the femoral neck. Kruskal–Wallis one-way ANOVA tests were used to examine differences for continuous variables, and chi-square tests or Fisher’s exact tests were used as appropriate to analyze differences for categorical variables. The least squares means method was used to calculate age-adjusted femoral neck BMD with 95% confidence intervals according to FSH tertile.

We estimated multivariable linear regression models with serum FSH as the predictor (mIU/mL; continuous) with femoral neck BMD (g/cm²; continuous) or total hip BMD (g/cm²; continuous) as separate outcomes. Results are reported as standardized betas (β) and standard errors (SE) with accompanying p values. Of the 674 participants with serum hormone measures, five were missing outcome data; these were excluded, leading to a final analytic sample of 669 women (see Fig. 1). Covariates were selected based on an a priori review of the confounders. The final list of variables included age (years; continuous), HT use (current user, non-current user; categorical), and diabetes (yes/no; categorical). We report all models adjusted for age and HT use along with a further set of models further adjusting for age, HT use, and diabetes.

Unconditional logistic regression models were used to estimate odds of low bone mass or osteoporosis (T-score < − 1.0) at the femoral neck and total hip according to FSH tertile. Results are reported as odds ratios (ORs) with accompanying 95% confidence intervals (95% CI).

We evaluated modification of the FSH-bone associations by E₂ using a multiplicative interaction term (FSHxE₂) in a logistic regression model with low bone mass or osteoporosis as the binary outcome and FSH as the independent variable (continuous) and in estradiol stratified models (split at the median estradiol value of 19.8 pg/mL). We considered the interaction term significant at p value < 0.10.

Mediation analysis

We evaluated FSH as a mediator of the E₂-bone relationship via the four commonly utilized criteria for identifying mediating variables and the Sobel test [14]. The four criteria are as follows:

1. To fulfill criterion #1, estradiol (continuous) must be statistically significantly associated with femoral neck BMD (continuous) in the absence of FSH (continuous), the putative mediating variable. We utilized an age-adjusted linear regression model to test this criterion.

2. Criterion #2 requires that E₂ is statistically significantly associated with FSH. We used an age-adjusted linear regression model to test this criterion, with FSH as the dependent variable and E₂ as the independent variable.

3. To fulfill criterion #3, FSH must be statistically significantly associated with femoral neck BMD. We used age-adjusted linear regression to assess these criteria in models, where femoral neck BMD was the dependent variable and FSH was the independent variable.

4. Criterion #4 requires that the E₂-BMD association is attenuated after adjustment for FSH. To test criterion 4, FSH was added into the model from step 2 to obtain an age- and FSH-adjusted regression coefficient (β′) representing the association between femoral neck BMD and E₂. The percent attenuation due to FSH was calculated as 100 × (β − β′)/β.

All hypotheses were tested using Wald tests of the regression coefficients at the alpha level of 0.05, apart from the multiplicative interaction term where a threshold of p < 0.10 alpha level was used to indicate statistical significance. Analyses were performed using SAS (release 9.4; SAS Institute, Inc., Cary, NC).

Results

Participant characteristics according to DXA-classified low bone mass and osteoporosis of the femoral neck at baseline are illustrated in Table 1. Women with normal bone mass were younger (63.5 ± 6.1 years) than women with low bone mass (66.5 ± 6.6 years) and osteoporosis (70.6 ± 6.3 years). Women with normal bone mass also tended to have a higher BMI (28.0 ± 5.1 kg/m²) compared to those with low bone mass (25.5 ± 4.5 kg/m²) and osteoporosis (23.8 ± 3.4 kg/m²). Women with osteoporosis had the highest FSH and lowest estradiol concentrations compared to women with normal bone mass.

Table 1.

Participant characteristics by diagnosis of low bone mass and osteoporosis of the femoral neck (n = 669)

	Normal bone mass (N = 300) T-score > −1.0	Low bone mass (N = 314) T-score (−2.4, −1.0)	Osteoporosis (N = 55) T-score ≤ −2.5	p value*
Age (years)	63.5 (6.1)	66.5 (6.6)	70.6 (6.3)	< 0.001
Years since menopause	15.2 (7.6)	18.6 (8.6)	22.6 (7.5)	< 0.001
Total body fat mass (kg)	31.4 (9.2)	26.7 (7.9)	23.6 (6.7)	< 0.001
BMI (kg/m2)	28.0 (5.1)	25.5 (4.5)	23.8 (3.4)	< 0.001
FSH (mIU/mL)	50.4 (27.3)	62.1 (29.2)	63.5 (29.1)	< 0.001
E2 (pg/mL)	37.0 (35.1)	26.1 (28.9)	26.1 (29.9)
Daily vitamin D intake (MCG)	5.1 (2.8)	5.1 (3.6)	4.2 (2.2)	0.100
Daily calcium intake (MG)	796.3 (386.1)	798.7 (479.9)	668.7 (327.8)	0.050
Pack-years of smoking	9.5 (16.5)	8.3 (15.9)	8.9 (15.9)	0.305
Hormone therapy use
Never	69 (23.0%)	105 (33.4%)	27 (49.1%)	< 0.001
Former	51 (17.0%)	71 (22.6%)	11 (20.0%)
Current	180 (60.0%)	138 (44.0%)	17 (30.9%)
Smoking status
Never	156 (52.0%)	178 (56.7%)	32 (58.2%)	0.279+
Former	138 (46.0%)	130 (41.4%)	20 (36.4%)
Current	6 (2.0%)	6 (1.9%)	3 (5.4%)
Education
High school	65 (21.9%)	53 (17.3%)	14 (25.4%)	0.465
College	127 (42.8%)	137 (44.6%)	20 (36.4%)
Post-college	105 (35.3%)	117 (38.1%)	21 (38.2%)
Missing	3	7	0
Race
American Indian/Alaska Native	1 (0.4%)	1 (0.3%)	0 (0.0%)	0.270+
Black/African American	4 (1.3%)	0 (0.0%)	0 (0.0%)
White	295 (98.3%)	311 (99.1%)	55 (100.0%)
Asian American/Pacific Islander	0 (0.0%)	2 (0.6%)	0 (0.0%)
Ethnicity
Hispanic	6 (2.0%)	0 (0.0%)	0 (0.0%)	0.026 +
Non-Hispanic	294 (98.0%)	314 (100.0%)	55 (100.0%)
Family history of fracture
No	191 (65.0%)	162 (52.1%)	28 (50.9%)	0.003
Yes	103 (35.0%)	149 (47.9%)	27 (49.1%)
Missing	6	3	0
Oophorectomy
No	245 (83.1%)	268 (86.4%)	47 (85.5%)	0.502
Yes	50 (16.9%)	42 (13.6%)	8 (14.6%)
Missing	5	4	0
Thyroid disease
No	241 (80.3%)	250 (79.6%)	48 (87.3%)	0.413
Yes	59 (19.7%)	64 (20.4%)	7 (12.7%)
Diabetes status
No	283 (94.3%)	308 (98.1%)	54 (98.2%)	0.037 +
Yes	17 (5.7%)	6 (1.9%)	1 (1.8%)

^*Non-parametric one-way ANOVAs for continuous variables and chi-square or Fisher’s exact tests for all categorical variables

⁺Fishers Exact test used

Age-adjusted mean femoral neck BMD (g/cm²; with 95% CIs) was plotted by FSH tertile in Fig. 2. We observed a significant inverse linear trend (p for trend < 0.001) where the highest mean femoral neck BMD was observed in the lowest FSH tertile (0.75 g/cm²) and the lowest mean femoral neck BMD was observed in the highest FSH tertile (0.70 g/cm²). There was a 6.7% absolute difference in BMD between tertiles 1 and 3.

Fig. 2.

Age-adjusted mean femoral neck BMD (g/cm²) and 95% CI by tertile of FSH concentration (n = 674). *Tertile 1: 0.75 g/cm² (0.74–0.77 g/cm²); tertile 2: 0.73 g/cm² (0.71–0.74 g/cm²); tertile 3: 0.70 g/cm² (0.69–0.72 g/cm.²)

Multivariable linear regression models of serum FSH with femoral neck and total hip BMD are reported in Table 2. In fully adjusted models, there was a significant inverse association between serum FSH concentration and BMD of the femoral neck (stdβ = − 0.142 (0.03); p < 0.001) and total hip (stdβ = − 0.147 (0.04); p = 0.001). For every standard deviation higher serum FSH concentration, there is on average 0.147 g/cm² lower hip BMD and 0.142 g/cm² lower femoral neck BMD.

Table 2.

Linear regression models of serum FSH concentration with femoral neck BMD and hip BMD (n = 669)*

b	Model 1 stdβ+ (SE)	p value	Model 2 stdβ (SE)	p value
Femoral neck BMD (g/cm2)	−0.156 (0.04)	<0.001	−0.142 (0.03)	0.001
Total hip BMD (g/cm2)	−0.162 (0.04)	<0.001	−0.147 (0.04)	<0.001

Model 1 adjusted for age and HT use

Model 2 adjusted for age, HT use, and diabetes

In logistic regression models adjusted for age, HT use, and diabetes depicted in Table 3, the odds of low bone mass or osteoporosis of the femoral neck increased in a linear fashion (tertile 2 OR: 1.78, 95% CI: 1.17–2.70; tertile 3 OR: 2.98, 95% CI: 1.86–4.77) compared to the reference tertile. Similar results were reported for the total hip.

Table 3.

Unconditional logistic regression models showing odds of DXA-classified low bone mass and osteoporosis of the femoral neck and total hip according to FSH tertile (N = 669)

	Femoral neck Normal (n = 300), N	T-score < – 1.0 (n = 369), N	Model 1 OR (95% CI)	Model 2 OR (95% CI)
FSH tertile 1 (n = 223) (0.7–42.5 mlU/mL)	133	90	Reference	Reference
FSH tertile 2 (n = 224) (42.6–66.8 mIU/mL)	96	127	1.85 (1.22–2.81)	1.78 (1.17–2.70)
FSH tertile 3 (n = 222) (66.9–188.7 mIU/mL)	71	152	3.17 (1.99–5.06)	2.98 (1.86–4.77)
	Total hip Normal (n = 437), N	T-score < – 1.0 (n = 232), N	Model 1 OR (95% CI)	Model 2 OR (95% CI)
FSH tertile 1 (n = 223) (0.7–42.5 mIU/mL)	173	50	Reference	Reference
FSH tertile 2 (n = 224) (42.6–66.8 mIU/mL)	135	88	1.80 (1.15–2.84)	1.70 (1.08–2.69)
FSH tertile 3 (n = 222) (66.9–188.7 mIU/mL)	129	94	1.89 (1.16–3.05)	1.74 (1.06–2.84)

Model 1 adjusted for age and HT use

Model 2 adjusted for age, HT use, and diabetes

In fully adjusted logistic regression models stratified by estradiol concentration (Table 4), women with lower estradiol (≤ 19.8 pg/mL; median) had significantly higher odds of low bone mass or osteoporosis in the second (OR: 3.01, 95% CI: 1.28–7.04) and third FSH tertiles (OR: 5.08, 95% CI: 2.13–12.10) compared to the first FSH tertile. In women with higher estradiol (> 19.8 pg/mL), we did not observe a significant association between the second tertile of FSH and odds of low bone mass/osteoporosis but did observe higher odds of low bone mass/osteoporosis in the third FSH tertile (OR: 2.43, 95% CI: 1.22–4.84). The interaction term (FSHxE₂) met the threshold for significance (p = 0.083). In sensitivity analyses conducted exclusively among women less than 10 years since menopause at baseline (n = 141), results did not differ appreciably (data not shown).

Table 4.

Unconditional logistic regression showing odds of low bone mass and osteoporosis according to tertile of serum FSH concentration, stratified by serum estradiol concentration cut at the median (N = 669)

	Estradiol below the median (N = 334; < 19.8 pg/mL)			Estradiol above the median (N = 335; > 19.8 pg/mL)
	Normal (N = 115), N	T-score < – 1.0 (N = 219), N	OR (95% CI)	Normal (N = 185), N	T-score < – 1.0 (N = 150), N	OR (95% CI)
FSH tertile 1 (n = 223) (0.7–42.5 mlU/mL)	21	14	-	112	76	-
FSH tertile 2 (n = 224) (42.6–66.8 mIU/mL)	44	81	3.01 (1.28–7.04)	52	47	1.43 (0.86–2.39)
FSH tertile 3 (n = 222) (66.9–188.7 mIU/mL)	50	124	5.08 (2.13–12.10)	21	27	2.43 (1.22–4.84)

Fully adjusted; age, HT use, and diabetes status

T-score < − 1.0 = low bone mass or osteoporosis

In mediation analyses (Table 5), we observed a significant positive relationship between estradiol and femoral neck BMD, as well as a significant inverse association between estradiol and FSH, fulfilling criteria 1 and 2, respectively. We also observed a significant inverse relationship between FSH and femoral neck BMD (criterion 3). Finally, in step 4, we estimated a 70% attenuation of the relationship between E₂ and femoral neck BMD via the mediating effect of FSH. The significant p value of the Sobel test provides additional evidence that FSH mediates the association of estradiol and BMD.

Table 5.

Mediation analysis using the Sobel method (n = 669) [14]

Step 1 β1 (se)	Step 1 p value	Step 2 β2 (se)	Step 2 p value	Step 3 β3 (se)	Step 3 p value	Step 4 β3 (se)	Step 4 p value	Step 4 (%atten.)	Sobel test p value
0.001 (0.0001)	< 0.001	−0.43 (0.03)	< 0.001	−0.001 (0.0002)	< 0.001	0.0003 (0.0002)	0.064	70%	< 0.001

Step 1 is linear regression of E2 (IV) and BMD (DV), model (criterion 1)

Step 2 is linear regression of FSH (mediator) and E2 (criterion 2)

Step 3 is linear regression of the BMD and FSH (criterion 3)

Step 4 is linear regression of E2 and BMD adjusted for FSH (criterion 4)

Percent attenuation (% atten.) = 100*(β1 − β4/β1)

Discussion

The findings herein support an inverse association of serum FSH concentration with low bone mass and osteoporosis, especially in women with low serum estradiol concentrations. Our results provide the most robust evidence to date of an inverse association between FSH and BMD in postmenopausal women.

Two recent studies of postmenopausal women are in agreement with our findings [7, 8]. In a study of 248 Chinese women over age 65, serum FSH concentration was significantly higher in women with osteoporosis or low bone mass (50.1 ± 4.9 mIU/mL) compared to those with normal bone mass (47.2 ± 4.32 mIU/mL) [8]. These findings mirror our observed FSH differences; women with osteoporosis had significantly higher FSH (63.5 ± 29.1 mIU/mL) compared to women with normal bone mass (50.4 ± 27.3 mIU/mL). In a cross-sectional study of 238 postmenopausal Icelandic women (mean age 80.8 ± 4.2 years), those in the highest FSH quartile had significantly lower femoral neck BMD (0.755 g/cm²) compared to women in the lowest FSH quartile (0.798 g/cm²; p for trend = 0.04) when adjusted for age, subgroup (data collected at different timepoints for each subgroup), serum estradiol concentration, and serum testosterone concentration [7].

In a smaller cross-sectional study of serum FSH concentration and DXA-derived measures of areal BMD conducted in 111 postmenopausal women (mean age 57.5 ± 3.7 years), FSH was not significantly associated with BMD at any skeletal site in models adjusted for age, weight, race, bioavailable E₂ concentration, bioavailable testosterone concentration, luteinizing hormone (LH) concentration, thyroid-stimulating hormone concentration, sex hormone binding globulin (SHBG) concentration, and bone specific alkaline phosphatase [9]. In a follow-up analysis in 94 non-current HT users from the original sample, there were no associations of serum FSH concentration and BMD in models adjusted for LH, SHBG, and urine N-telopeptide [10]. While the authors removed confounding due to HT use via restricting to non-current HT users in the follow-up analysis, they did not consider measured estradiol concentration in their study. These smaller studies may have had inadequate statistical power to detect associations.

The observation that estrogen deficiency itself does not fully explain the bone loss accompanying hypogonadism has led some to hypothesize an alternative pituitary-driven cause of postmenopausal bone loss [15–18]. During early perimenopause, a time when circulating FSH increases exponentially, there is a marked increase in the rate of bone loss despite no meaningful reduction in circulating estrogen [19, 20]. In fact, the first study of FSH and bone in women reported age-related decline in both lumbar spine and femoral neck bone density associated with elevated serum FSH concentration in 281 perimenopausal women, despite mean serum estradiol concentrations similar to those of premenopausal women [21]. In animal studies, administering FSH to mice has been shown to increase the formation of resorption pits (lacunae) of human peripheral blood mononuclear cell-derived osteoclasts [4, 5]. This prevents osteoclast apoptosis and implicates FSH in survival of bone-resorbing osteoclasts [4, 5]. However, replication of this study has been inconsistent [6]. Another study demonstrated FSH induced the production of tumor necrosis factor-α (TNFα) in mouse monocytes and macrophages which directly led to the promotion of osteoclast precursors [22, 23]. Finally, one study demonstrated that ovariectomized FSH receptor (FSHR) null mice do not experience bone loss despite severe hypogonadism [24]. This suggests that the skeletal action of FSH may be estrogen independent.

In addition to the existing animal studies and observational studies of postmenopausal women described above, a randomized trial including 41 postmenopausal women was conducted to test the influence of a GnRH inhibitor (an FSH suppressant) on bone turnover markers over a 3-month period [25]. Participants were randomized to receive either the GnRH inhibitor (leuprolide acetate) resulting in suppressed FSH or placebo. Both groups received an aromatase inhibitor to suppress estrogen concentrations. FSH suppression did not induce a change in bone turnover markers including serum C-terminal telopeptide of type I collagen (CTX), tartrate-resistant acid phosphatase isoform type 5b (TRAP-5b), or osteocalcin. In addition to concerns regarding the utility of the surrogate biomarkers of bone remodeling [26–31], there is ample evidence that low estrogen is a cause of hypogonadal bone loss independent of FSH; however, it does not exclude a role of FSH in skeletal homeostasis [31]. There is inherent difficultly in separating the action of FSH from that of estrogen on bone in vivo; however, there is biologic evidence that they act through separate mechanisms.

In estradiol-stratified analyses, we observed a more pronounced relationship between FSH and bone in the low estradiol stratum after accounting for HT use. Taken together to the significant interaction term indicating multiplicative interaction of FSH and estradiol, these findings illustrate that estradiol is not a confounder, as it has been treated in previous studies (if considered at all), but instead acts as an effect modifier in the FSH-bone relationship. These results corroborate two previous studies conducted in women across the menopausal transition which demonstrate that each halving of serum E₂ and doubling of serum FSH over at least 1 year of follow-up is associated with greater risk of significant bone loss when compared to women with increasing E2 concentration or diminishing FSH concentration over follow-up [32, 33]. In high-estrogen environments, the protective effect of estradiol may mitigate or mask the negative effect of FSH on bone while in low-estrogen environments, the protective effect of estradiol is lost, leaving the opportunity for adverse action of elevated FSH.

The separate mediation analysis illustrates FSH as a mediator of the relationship between the E₂ and bone, where approximately 70% of the E₂-BMD association was mediated through FSH. This evidence demonstrates the interconnected E₂-FSH feedback mechanism on bone health and the difficulty of teasing them apart via statistical methodology.

Our study has several strengths including the ability to evaluate the estrogen-independent association as we have measured serum estradiol concentration and information on HT use. This is a relatively large sample of postmenopausal women who represent the group at highest risk of age-related bone loss and the largest sample of postmenopausal women to evaluate the association to date. We used a variety of statistical methods to analyze the complicated biological relations of these hormones in an epidemiologic context to help tease out and improve interpretations of the individual hormone influence on bone loss and improve on published literature. Due to the cross-sectional nature of the study, we are unable to make conclusions about temporality of these associations. The Buffalo OsteoPerio Study is a cohort of predominately non-Hispanic White women (98.7%); therefore, our findings cannot be generalized to other race and ethnic groups.

To our knowledge, this study constitutes the largest examination of FSH and bone mineral density in postmenopausal women. Our findings suggest that elevated circulating FSH is adversely associated with skeletal homeostasis, and the effects may be more pronounced in a low-estrogen environment. Longitudinal studies in large peri- and postmenopausal samples are needed to further study the joint and independent influence of estradiol and FSH on bone loss.