Higher Serum Free Testosterone Concentration in Older Women Is Associated with Greater Bone Mineral Density, Lean Body Mass, and Total Fat Mass: The Cardiovascular Health Study

Higher serum testosterone is associated with higher bone mineral density, greater lean body mass, and greater total fat mass in women aged 65 and older.

Abstract

Context:

The physiological importance of endogenous testosterone (T) in older women is poorly understood.

Objective:

The aim of the study was to determine the association of higher total and free T levels with bone mineral density (BMD), lean body mass, and fat mass in elderly women.

Design:

Total and free T were measured using sensitive assays in 232 community-dwelling women aged 67–94 yr who were enrolled in the Cardiovascular Health Study and had dual-energy x-ray absorptiometry scans. Cross-sectional analyses were performed to examine associations between total and free T and BMD and body composition.

Results:

In adjusted models, total T was directly associated with BMD at the lumbar spine (P = 0.04) and hip (P = 0.001), but not body composition outcomes, in all women, and after excluding estrogen users and adjusting for estradiol (P = 0.04 and 0.01, respectively). Free T was positively related to hip BMD, lean body mass, and body fat (all P < 0.05), with more than 10% differences in each outcome between women at the highest and lowest ends of the free T range, with attenuation after excluding estrogen users and adjusting for estradiol.

Conclusions:

In the setting of the low estradiol levels found in older women, circulating T levels were associated with bone density. Women with higher free T levels had greater lean body mass, consistent with the anabolic effect of T, and, in contrast to men, greater fat mass. Mechanistic studies are required to determine whether a causal relationship exists between T, bone, and body composition in this population and the degree to which any T effects are estrogen-independent.

Little is known about the role of endogenous testosterone (T) in women over the age of 65. In older men, associations between endogenous T and bone mineral density (BMD) and body composition have been described (1–4). Furthermore, T therapy has been shown to increase vertebral BMD and skeletal muscle mass and decrease fat mass in randomized trials of older men (5–7). Androgen receptors (ARs) are expressed on osteoblasts and osteocytes; T affects osteoblastic bone formation and osteoclastic bone resorption directly through AR signaling, indirectly through its aromatization to estrogen, and through multiple other potential mechanisms (8). T may also promote myogenic differentiation of multipotent mesenchymal progenitor cells (9).

Small studies have shown that T therapy in postmenopausal women increases BMD and lean body mass, while decreasing fat mass (10–13). Furthermore, higher levels of endogenous T in postmenopausal women are associated with a decreased risk of hip fracture (14, 15). Limited data suggest a positive association of T with muscle mass in older women (16), although its association with fat mass may be opposite to that seen in men, with recent reports of associations between higher T and greater adiposity and insulin resistance in older women (17, 18). A syndrome of female androgen insufficiency has been proposed to result in persistent fatigue and sexual dysfunction with potential reductions in BMD, muscle strength, and cognition (19), although experts agree that more data are needed regarding BMD and muscle outcomes (20–22).

This cross-sectional study of 232 U.S. community-dwelling older women was conducted to determine whether T levels in women aged 65 and older are associated with BMD and body composition. We hypothesized that higher T levels would be associated with higher BMD, greater lean body mass, and more fat mass, and we predicted that associations would be stronger for free T than for total T.

Subjects and Methods

Study population

The Cardiovascular Health Study (CHS) is a population-based, longitudinal study of 5888 adults aged 65 and older (23). Enrollment of an original cohort of 5201 adults occurred between 1989 and 1990, and in 1992–1993 an additional 687 predominantly African-American participants were enrolled. Eligible individuals were identified from an age- and gender-stratified random sample of the Medicare eligibility rosters in four U.S. communities: Washington County, Maryland; Pittsburgh, Pennsylvania; Sacramento County, California; and Forsyth County, North Carolina. To be eligible, individuals had to be noninstitutionalized, expecting to remain in the area for the following 3 yr, not under active cancer treatment, not wheelchair-bound in the home, not requiring a proxy respondent at entry, and capable of providing consent. Household members of the sampled individuals were recruited, if eligible. The institutional review boards of all four sites and the coordinating center at the University of Washington in Seattle approved the study, and all participants gave informed consent.

A random sample of 368 women in both cohorts seen at the 1992–1993 study visit was later selected for measurement of total and free T, as described previously (17, 24). Total and free T measurements were repeated in women who returned to the 1994–1995 visit and had available serum (n = 311). This visit included a medical history, physical examination, assessment of health status, and phlebotomy. Our analyses are based on data from the 232 women who had dual-energy x-ray absorptiometry (DXA) scans at this visit (performed at Pittsburgh and Sacramento sites only) and were not taking bisphosphonates or corticosteroids. Total T, free T, and health characteristics did not differ between the 232 women with and the 79 women without DXA scans.

Study measures

Serum total T concentrations were measured by RIA using iodinated T as a tracer (25). This assay was developed and validated for the low range of T prevalent in HIV-infected (25–27) and postmenopausal women (28). The sensitivity, defined as hormone concentration corresponding to 90% bound in the presence and absence of analyte point, was 0.22 ng/dl (0.008 nmol/liter). The intraassay coefficient of variation (CV) was 8.2%. Interassay CVs were measured in the low, medium, and high female pool and were 13.2% in the medium pool. In other pools, interassay CVs ranged from 13 to 15%. This assay was validated against liquid chromatography-tandem mass spectrometry (LC-MS/MS) (27). These measurements demonstrated a correlation of 0.997 between the RIA and LC-MS/MS measurements. Free T concentrations were measured by a sensitive equilibrium dialysis assay (25), optimized to precisely and accurately measure low concentrations. The sensitivity of this assay is 0.3 pg/ml (1.0 pmol/liter). Ten replicates of low, medium, and high female pools were used to generate intra- and interassay CVs. The respective intraassay CVs for the low, medium, and high female pools were 5.6% for the low pool, 4.6% for the medium pool, and 2.6% for the high pool (25). The interassay CVs were in the 12–15% range. Cross-reactivity of the major steroids (dehydroepiandrosterone, dehydroepiandrosterone sulfate, dihydrotestosterone, and androstenedione) was less than 1%. Estradiol levels were measured by Esoterix, Inc. (Calabasas Hills, CA) by LC-MS/MS in non-estrogen-using women who had available serum for analysis (n = 184). The sensitivity was 1 pg/ml (3.7 pmol/liter), with an intraassay CV of 2.3% and interassay CV of 4.4% in the low range.

Baseline characteristics included age, race, weight, education, smoking status, alcohol use, activity level (blocks walked last week), and estrogen use. Medication use was determined from examination of medication bottles at the study visit. Body mass index (BMI) (kilograms/meter²), computed from objective measures, was categorized as less than 18.5, 18.5–24.9, 25–29.9, or 30 or greater.

Hip and whole body DXA scans were obtained using Hologic QDR-2000 densitometers (Hologic, Inc., Waltham, MA) with a protocol similar to that in the Study of Osteoporotic Fractures and the Fracture Intervention Trial (29, 30). Scans were read blindly at the University of California, San Francisco Reading Center using Hologic software, version 7.10. The total hip BMD was chosen as the most representative and reproducible of the hip measurements with a CV less than 0.75% (31). Lumbar spine bone density was estimated from whole body scans.

Statistical analysis

Women were categorized as being normal, having low bone mass, or osteoporotic using the World Health Organization osteoporosis classification (32) based on the total BMD of the hip; total and free T levels were compared among the three categories using a trend test for all women and repeated excluding estrogen users. Total and free T were plotted against each BMD and body composition outcome (total hip BMD, lumbar spine BMD, lean body mass, and absolute and percentage total body fat) to detect evidence for nonlinear associations. Paired t tests and one-way ANOVAs were performed to determine factors associated with each BMD and body composition outcome. Factors found to be statistically significant at a P < 0.10 level, based on two-sided tests, were included in multivariable linear regression models predicting BMD and body composition outcomes. Total and free T, BMD measures, and total lean body mass were log-transformed to achieve normality. BMD models were adjusted for age, race, weight, and estrogen use. Lean body mass models were adjusted for age, race, and activity level, whereas both total body fat and percentage body fat models were adjusted for age, race, alcohol use, and activity level. All analyses were repeated in the 189 women who were not taking estrogen and, repeated adjusting for estradiol in the 184 women in whom estradiol levels were performed.

To provide estimates of the clinical significance of findings from the multivariable models, estimates of BMD and body composition outcomes were displayed graphically across a range of total and free T concentrations for a hypothetical woman whose remaining model covariates were at the predominant value for categorical variables or mean values for continuous variables. T levels were back-transformed for interpretation on the original T scales. For example, predicted hip bone densities for a 75-yr-old, 68-kg, non-estrogen-using Caucasian woman across a range of total and free T levels are displayed (see Fig. 2).

Fig. 2.

Predicted lumbar spine (A) and total hip (B) bone density for a 75-yr-old, 68-kg, non-estrogen-using Caucasian woman. Multiply by 0.0347 to convert total T concentrations from nanograms/deciliter to nanomoles/liter. Multiply by 3.47 to convert free T concentrations from picograms/milliliter to picomoles/liter.

Results

Baseline characteristics

The 232 U.S. community-dwelling women ranged in age from 67 to 94 yr, with a mean age of 75.6 yr (Table 1). Seventy-nine percent were Caucasian. The mean BMI in this cohort was 27.2 kg/m². A small number of women (18%) reported estrogen use at the time of the study. On average, the total T level in this cohort was 12 ng/dl (0.42 nmol/liter), and the free T level was 1.6 pg/ml (5.6 pmol/liter).

Table 1.

Baseline characteristics of the study sample (n = 232)

Measurement
Age (yr)	75.6 (4.6)
Race (% Caucasian)	79
Estrogen use (%)	18
Alcohol use (%)
None	48
Any	38
Daily	14
BMI (kg/m2)	27.2 (5)
Median blocks walked in last week	12
Total T (ng/dl)	12 (14)
Free T (pg/ml)	1.6 (1.6)

Data represent mean (sd) unless otherwise stated. Multiply by 0.0347 to convert total T concentrations from ng/dl to nmol/liter. Multiply by 3.47 to convert free T concentrations from pg/ml to pmol/liter.

Mean total and free T levels by osteoporosis classification of the hip

Overall, there was a significant stepwise increase in hip bone mass with higher levels of total and free T (P = 0.003 and 0.009, respectively; Fig. 1). The lowest mean total T [8.6 ng/dl (0.30 nmol/liter)] and free T [1.2 pg/ml (4.2 pmol/liter)] levels were seen in women with osteoporosis of the hip; intermediate levels [11.0 ng/dl (0.38 nmol/liter) and 1.6 pg/ml (5.6 pmol/liter)] were seen in those with low bone mass; and the highest levels [16.3 ng/dl (0.57 nmol/liter) and 2.1 pg/ml (7.29 nmol/liter)] were seen in those with normal BMD. After excluding the 43 women taking estrogen, similar results were seen for osteoporosis, low bone mass, and normal BMD for total T (7.1, 11.2, and 18.5 ng/dl; P < 0.001) and free T (1.2, 1.7, and 2.3 pg/ml; P = 0.02).

Fig. 1.

Mean total T (A) and free T (B) levels by osteoporosis classification of total hip BMD. Osteoporosis was defined as a T-score ≤ 2.5 sd, low bone mass as a T-score of −1 to −2.5 sd, and normal BMD as T-score ≥ −1 sd, per World Health Organization classification (32). Multiply by 0.0347 to convert total T concentrations from nanograms/deciliter to nanomoles/liter. Multiply by 3.47 to convert free T concentrations from picograms/milliliter to picomoles/liter.

Bone density and body composition outcomes

In Table 2, we present crude and adjusted estimates of log-transformed total and free T levels with BMD and body composition outcomes. In adjusted models, total T was directly associated with BMD at the lumbar spine (P = 0.04) and hip (P = 0.001), but not with any of the body composition outcomes. The BMD associations were unchanged after excluding the estrogen users (P = 0.02 at the lumbar spine, and P = 0.004 at the hip) and additionally adjusting for estradiol levels in this subgroup (P = 0.04 and 0.01, respectively). Free T was not significantly associated with lumbar spine BMD (P = 0.12), but was positively associated with total hip BMD (P = 0.03), total lean body mass (P = 0.01), total body fat (P = 0.007), and percentage fat (P = 0.02). After exclusion of the estrogen users, free T remained associated with total body fat (P = 0.04), but not with total hip BMD, total lean body mass, and percentage body fat (P = 0.05–0.12). Additional adjustment for estradiol levels attenuated the magnitude of associations between free T and outcomes, all of which were not statistically significant. No threshold relationships were seen between total or free T and any of the outcomes.

Table 2.

Crude and adjusted models of total and free T and bone density and body composition outcomes

Outcome	Crude β coefficient	P value	Adjusted β coefficient	P value
Log lumbar spine BMD
Log total T
All women	0.033	0.02	0.027	0.04
Excluding estrogen users	0.040	0.01	0.032	0.02
Excluding estrogen users + E2 adjusted			0.030	0.04
Log free T
All women	0.044	0.01	0.025	0.12
Excluding estrogen users	0.045	0.02	0.020	0.25
Excluding estrogen users + E2 adjusted			0.016	0.36
Log total hip BMD
Log total T
All women	0.038	0.002	0.032	0.001
Excluding estrogen users	0.042	0.002	0.032	0.004
Excluding estrogen users + E2 adjusted			0.029	0.01
Log free T
All women	0.046	0.002	0.028	0.03
Excluding estrogen users	0.044	0.007	0.021	0.12
Excluding estrogen users + E2 adjusted			0.020	0.14
Log lean body mass
Log total T
All women	0.005	0.56	−0.001	0.92
Excluding estrogen users	0.005	0.61	−0.006	0.51
Excluding estrogen users + E2 adjusted			−0.012	0.22
Log free T
All women	0.031	0.005	0.024	0.01
Excluding estrogen users	0.031	0.009	0.020	0.06
Excluding estrogen users + E2 adjusted			0.016	0.16
Total body fat (kg)
Log total T
All women	0.61	0.39	0.23	0.73
Excluding estrogen users	0.74	0.36	0.12	0.87
Excluding estrogen users + E2 adjusted			−0.75	0.31
Log free T
All women	2.68	0.002	2.11	0.007
Excluding estrogen users	2.91	0.002	1.88	0.04
Excluding estrogen users + E2 adjusted			1.38	0.11
Percentage body fat
Log total T
All women	0.31	0.54	0.05	0.91
Excluding estrogen users	0.38	0.51	0.08	0.88
Excluding estrogen users + E2 adjusted			−0.38	0.50
Log free T
All women	1.71	0.005	1.39	0.02
Excluding estrogen users	1.80	0.008	1.27	0.05
Excluding estrogen users + E2 adjusted			1.03	0.10

β-Coefficients represent the change in outcome for a one log unit increase in T level and are not back-transformed. BMD models with all women were adjusted for age, race, weight, and estrogen use. BMD models without estrogen users were adjusted for age, race, and weight. Lean body mass models were adjusted for age, race, and activity level. Body fat models were adjusted for age, race, activity level, and alcohol use. Statistically significant values are indicated in bold. E2, Estradiol.

Because the estimated β coefficients are unable to be interpreted clinically, Fig. 2 displays the predicted lumbar spine and hip BMDs for a hypothetical woman whose model covariates were at the predominant value for categorical variables or mean values for continuous variables. This figure shows that predicted lumbar spine and hip BMDs increased nonlinearly and to the same degree with increasing level of total and free T, with a 0.1 g/cm² difference between women at the highest and lowest ends of the total or free T range. A similar effect was seen in the relationship between free T and predicted lean body mass (3.9-kg difference) and fat mass (9.7-kg difference) for a 68-kg woman, but not for total T, with no difference in body composition across the spectrum of total T values (Fig. 3).

Fig. 3.

Predicted lean body mass (A) and fat mass (B) for a 75-yr-old, nondrinking, average activity, Caucasian woman. Multiply by 0.0347 to convert total T concentrations from nanograms/deciliter to nanomoles/liter. Multiply by 3.47 to convert free T concentrations from picograms/milliliter to picomoles/liter.

Discussion

Our findings indicate that older women with higher circulating concentrations of T have significantly greater BMD, independent of weight and other confounding factors, suggesting that circulating androgens may play a role in maintenance of bone density in the setting of the low estradiol levels typical for this age group of women. Women with higher free T levels also had greater lean body mass, consistent with an anabolic effect of T, even at the low levels seen in women of this age group. However, in contrast to what is seen in men, women with higher free T levels had higher absolute and total percentage body fat than their counterparts with lower free T levels.

Our findings of associations between T and BMD in older women are consistent with several other studies showing a positive relationship with BMD (1, 2, 16, 33, 34) and decreased incidence of fracture (14, 15), although not with three studies reporting an absence of association between T and fracture risk in women (35–37). Our findings are also concordant with four small controlled T administration trials, three in postmenopausal women and the other in women with hypopituitarism, that demonstrated gains in BMD with T use (10–12, 38). Our observational study extends the association between T and BMD to elderly women, as the only published analysis conducted exclusively in women aged 65 and older. Furthermore, our study employed a more sensitive T assay than previous studies have, and it is the only one with measurements of free T by dialysis.

T can potentially affect bone density through multiple mechanisms, including through its direct effects on the ARs in trabecular and cortical bone (8). Aromatization of T to estradiol by adipose tissue or locally at bone, with subsequent stimulation of estrogen receptors in bone, plays an important role in the action of T on the bone (39–41). In women of the advanced age of our study participants, T is the major source of circulating estradiol. T may also regulate local production of cytokines and growth factors in bone, including IL-6, IL-1β, TGF-β, and IGFs (8). The anabolic effects of T on muscle mass and strength could also indirectly affect bone mass. Additional studies are needed to elucidate the independence of the roles of physical activity and sex steroids on bone (42).

Additionally, in our investigation, we found that older women with higher free T levels had significantly greater lean body mass, which is consistent with the anabolic effects of T. Two other cross-sectional studies of postmenopausal women have shown correlations between free T and lean body mass (16, 43), although these analyses were not adjusted for relevant covariates. Interestingly, in contrast to what has been shown in men, our study found that higher free T levels were associated with greater fat mass. This is consistent with other data showing that postmenopausal women with higher free T levels have greater BMI and are more insulin resistant (17, 18, 24), and it raises the question of whether the effects of T on body composition are gender dimorphic. We prefer an alternative explanation, that the increase in androgen production with increasing adiposity originally described in polycystic ovary syndrome occurs across a woman's life span. Women with greater adiposity are more likely to have hyperinsulinemia due to insulin resistance, which could then lead to insulin-stimulated T production by the postmenopausal ovary. We have recently published a mechanistic study in postmenopausal women that supports this explanation (44). Data from this study suggest that T is a marker of adiposity and insulin resistance in older women and does not play a causal role in fat synthesis.

Because estrogen supplementation may have effects on bone and body composition that are incompletely accounted for by adjustment, we repeated all of our analyses after excluding estrogen users. Interestingly, this had no effect on the findings of a relationship between total T and BMD, but it did result in a decrease in both the magnitude and statistical significance of the relationships between free T and outcomes. The diminishment of associations, by 9–25%, is likely due to a true difference between estrogen users and non-users, although there was also a decrease in statistical power to detect associations due to a reduced sample size. Additional adjustment for estradiol, which has been shown to have important effects on analyses of T and outcomes (45), led to further decreases in the magnitude of the associations. However, an important issue should be considered in the models with both free T and estradiol. Estradiol concentrations are a marker of T concentrations, particularly in women with high adiposity, because the major source of estradiol in the postmenopausal woman is aromatization of T. In addition, the free T assay, due to the low concentration that the assay is attempting to measure, has more assay variability than a tandem mass spectrometry assay of estradiol. When modeled together, the more precise assay, in this case estradiol, may predominate due to assay characteristics, not underlying biology. Additional studies are required to determine the specific roles of estrogens vs. androgens in bone density and body composition in elderly women.

The strengths of our study include its population-based sample of older women, available measured covariates, and the use of highly sensitive assays of T and estradiol. However, our analyses are cross-sectional in nature, which limits inferences about causality and directionality of association. In addition, DXA measurement of lean body mass includes body water and internal organs in addition to muscle mass and is not the “gold standard” for assessing this tissue compartment.

In conclusion, our study demonstrated that higher endogenous free T levels are associated with higher BMD, greater lean body mass, and greater total fat mass in women aged 65 and older. The differences in BMD and muscle mass across the spectrum of free T levels were not only statistically significant, but also clinically significant, at greater than 10%. There are now a multitude of medications to prevent or treat osteoporosis, but no pharmacological therapies for age-related sarcopenia or frailty. The hypothesis that T therapy might potentially improve bone density and sarcopenia across a range of T levels that are physiological for women, either through T supplementation or through administration of selective AR modulators, should be explored. Further studies are needed to refine the appropriate target populations and to examine the risks and benefits of exogenous T in older women.