Sex Hormone–Binding Globulin, but Not Testosterone, Is Associated Prospectively and Independently With Incident Metabolic Syndrome in Men: The Framingham Heart Study

The Boston University Institutional Review Board approved the study. All subjects provided written consent.

In 1971, the offspring of the original FHS participants and the spouses of the offspring (FHS generation 2) were enrolled into the Framingham Offspring Study (16) and have since completed eight examinations at 4- to 8-year intervals. Generation 2 men who attended examination 7 (1998–2001) were eligible for inclusion in cross-sectional analyses (n = 1,625). The validation sample included children of the offspring cohort (generation 3) who attended examination 1 (n = 1,912). Men receiving testosterone or androgen-deprivation therapy for prostate cancer and those missing testosterone or data necessary for defining metabolic syndrome were excluded, resulting in a sample size of 1,407 subjects for the cross-sectional analyses of the training sample and 1,887 subjects for the validation sample (Fig. 1).

To determine whether hormone levels were associated with incident metabolic syndrome, the men of generation 2 who had hormone measurements at examination 7 were examined, on average, 6.6 years later (2005–2008) at examination 8. We excluded men who were receiving testosterone or androgen-deprivation therapy for prostate cancer or those who had missing testosterone or other data necessary for defining metabolic syndrome at either examination 7 or examination 8. We also excluded men who reported metabolic syndrome at examination 7 (n = 668). Hence, 618 men were available to prospectively examine the association between hormone levels at examination 7 and incident metabolic syndrome at examination 8 (Fig. 1).

Total testosterone and SHBG levels were measured in generation 2 at examination 7 and in generation 3 at examination 1. Samples were obtained after an overnight fast between 7:00 and 9:00 a.m. and stored at −80°C. Total testosterone was measured by LC-MS/MS (17) with a sensitivity of 2 ng/dL and interassay coefficients of variation (CVs) of 7.8, 5.9, and 3.5% at testosterone concentrations of 250, 500, and 1,000 ng/dL, respectively.

SHBG levels were measured using an immunofluorometric assay (DELFIA-Wallac, Turku, Finland) with a sensitivity of 2.5 nmol/L (17). Interassay CVs were 8.3, 7.9, and 10.9%, and intra-assay CVs were 7.3, 7.1, and 8.7%, respectively, in the low, medium, and high pools. Free testosterone was calculated using a law-of-mass-action equation whose dissociation constants and assumptions have been published (17).

Metabolic syndrome was defined using modified Adult Treatment Panel III criteria (18), which defines metabolic syndrome by the presence of at least three of the following: waist circumference >40 inches (102 cm); HDL cholesterol <40 mg/dL; triglycerides ≥150 mg/dL; high blood pressure defined as systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥85 mm Hg or on antihypertensive treatment; and fasting glucose ≥100 mg/dL or on diabetes treatment.

HOMA-IR was calculated as follows: ([insulin] × [fasting glucose × 0.055]/22.5). The men who reported smoking during the past year were categorized as smokers. High-sensitivity C-reactive protein (hsCRP) was measured using a published assay (19).

Descriptive statistics for continuous variables and percentages for dichotomous variables were generated. Cross-sectional associations among sex hormones and metabolic syndrome were assessed using multiple logistic regression. Models were adjusted for age and smoking because these variables were significantly associated with testosterone levels. To determine whether the association of total testosterone with metabolic syndrome was influenced by SHBG, we adjusted the analyses for SHBG. In additional models, BMI and HOMA-IR were added sequentially.

We used multiple logistic regression to examine the relationship between sex hormone levels at examination 7 and incident metabolic syndrome at examination 8 in men free of metabolic syndrome at examination 7, adjusting for age and smoking. Additional models to examine the association between total and free testosterone at examination 7 with incident metabolic syndrome at examination 8 were generated after sequential adjustment for age, smoking, SHBG, BMI, and HOMA-IR. Statistical significance level was set at two-sided P < 0.05.

The original sample for cross-sectional analyses included 1,407 men of generation 2 who attended examination 7, had nonmissing total testosterone, and were not receiving testosterone or androgen-deprivation therapy (Fig. 1). The validation sample included 1,887 men of generation 3 who attended examination 1, had nonmissing testosterone, and were not receiving androgen-deprivation therapy.

For longitudinal analyses, of 1,625 men in generation 2 who attended examination 7, we excluded those who had missing testosterone or metabolic syndrome data, those who received hormonal treatment for prostate cancer, and those who had metabolic syndrome at examination 7; the remaining 618 men constituted the analytic sample for longitudinal analyses. Men who were excluded did not differ from those who were included in longitudinal analyses (not shown).

Men in the original sample were, on average, 61.1 years old with mean total testosterone, free testosterone, and SHBG levels of 586 ± 227 ng/dL, 87 ± 32 pg/mL, and 58 ± 27 nmol/L, respectively (Table 1). A total of 47.5% of men in the sample had metabolic syndrome at baseline, 51.8% had abdominal adiposity, 57.3% had fasting glucose >100 mg/dL, 61.4% had hypertension, 33.1% had hypertriglyceridemia, and 36.0% had low HDL cholesterol. In contrast, the men in the validation sample were, on average, younger and had higher total and free testosterone levels and a lower prevalence of metabolic syndrome and all of its components. Because men with metabolic syndrome at examination 7 were excluded from longitudinal analyses, the sample used for longitudinal analyses had a lower prevalence of cardiovascular disease, hypertension, abdominal adiposity, impaired fasting glucose, hypertriglcyeridemia, and low HDL cholesterol.

In cross-sectional analyses of the original sample, total as well as free testosterone levels were significantly negatively associated with the risk of metabolic syndrome after adjusting for age and smoking; the association of metabolic syndrome was stronger with total than with free testosterone (Table 2). For each SD decrease in total and free testosterone levels, the odds of prevalent metabolic syndrome were increased by 83 and 25%, respectively, after adjusting for age and smoking (Table 2). The men in the lowest quartile of total testosterone had 4.5 times the odds of having prevalent metabolic syndrome than men in the highest quartile (P < 0.0001) (Fig. 2A) (Supplementary Table 1). The men in the lowest quartile of free testosterone had 1.9 times the odds of having metabolic syndrome than those in the highest quartile (P < 0.0001).

In addition to age and smoking, total and free testosterone levels were associated with SHBG, BMI, HOMA-IR, and hsCRP levels. SHBG levels also were related to age, smoking, BMI, and HOMA-IR. Accordingly, in sensitivity analyses, we adjusted the analyses sequentially for these covariates in addition to age and smoking. The associations of total and free testosterone with the risk of metabolic syndrome were attenuated after adjusting for SHBG and BMI sequentially; the association was substantially weakened when the analyses were adjusted for all covariates (i.e., age, smoking, SHBG, BMI, and HOMA-IR) simultaneously. Free testosterone was no longer significantly associated with prevalent metabolic syndrome after adjusting for BMI and HOMA-IR (Table 2). Additional adjustment for hsCRP had little effect on the strength of association of hormone levels with metabolic syndrome (not shown).

SHBG levels were significantly negatively associated with the risk of metabolic syndrome; each SD decrease in SHBG was associated with 78% higher odds of prevalent metabolic syndrome after adjusting for age and smoking (Table 2). Men in the lowest quartile of SHBG had nearly five times the odds of having prevalent metabolic syndrome than those in the highest quartile (Fig. 2A) (Supplementary Table 1). Unlike total and free testosterone, the association of SHBG with the risk of metabolic syndrome persisted after adjusting for age, smoking, testosterone, BMI, and HOMA-IR.

The results of the cross-sectional analyses in the original sample were confirmed by analyses of the validation sample (Table 2) (Supplementary Table 2). Total and free testosterone levels in generation 3 were negatively associated with the odds of having metabolic syndrome after adjusting for age and smoking; this association was attenuated after adjusting for SHBG and attenuated further after additional adjustments for BMI and HOMA-IR. Free testosterone was not significantly associated with metabolic syndrome after adjusting for age, smoking, BMI, and HOMA-IR.

Neither total nor free testosterone at examination 7 was significantly associated with incident metabolic syndrome at examination 8 regardless of whether total and free testosterone levels were considered as continuous variables or categorized in quartiles (Table 2) (Fig. 2A) (Supplementary Table 3). Only SHBG at examination 7 was significantly associated with incident metabolic syndrome at examination 8 (Table 2). The association of SHBG with incident metabolic syndrome persisted after adjusting for age, smoking, BMI, and HOMA-IR (Fig. 2A). The men in lower quartiles of SHBG levels had progressively higher risk of incident metabolic syndrome than men in the highest quartile, even after adjusting for total testosterone, BMI, and HOMA-IR.

SHBG levels were significantly associated with each individual component of metabolic syndrome in both the original and the validation samples, although the association was generally stronger for abdominal adiposity and high triglycerides than for other components (Fig. 2B) (Supplementary Tables 4–6). The association of SHBG with individual components of the metabolic syndrome was attenuated after a multivariable adjustment for age, BMI, and HOMA-IR.

Our analyses reveal that SHBG, but not testosterone, is an independent predictor of incident metabolic syndrome. Consistent with a large body of published data (1–9), our cross-sectional analyses also revealed a significant association of total and free testosterone levels with prevalent metabolic syndrome. However, the association of total testosterone and SHBG with prevalent metabolic syndrome was stronger than that of free testosterone. Furthermore, the cross-sectional association of total testosterone with metabolic syndrome was attenuated substantially when the analyses were adjusted for SHBG, age, BMI, and insulin sensitivity. Likewise, the association of free testosterone with metabolic syndrome was no longer significant after adjusting for age, BMI, and insulin sensitivity. In longitudinal analyses, neither total nor free testosterone was significantly associated with incident metabolic syndrome. Only SHBG was significantly associated both cross-sectionally and longitudinally with the risk of metabolic syndrome, independent of testosterone levels.

SHBG has been reported to be associated with incident diabetes (20) and fracture risk (21). Polymorphisms of the SHBG gene also have been linked with diabetes (20,22). These observations, when viewed together with our data, highlight the importance of SHBG as an independent predictor of metabolic disorders, such as diabetes and metabolic syndrome.

Our study has several strengths. The FHS cohort included community-dwelling men over a wider age range (aged 19–95 years) than has been included in other studies, which were focused mostly on older men. The findings from the original sample were confirmed in a separate validation sample. The longitudinal analyses lend additional strength to our inferences. Furthermore, the progressive decrease in odds ratios as hormone levels increased from the first to the fourth quartile suggests robust dose-response relationships in these associations. We adjusted our analyses for potential confounders, including age, BMI, smoking, and insulin sensitivity. We measured testosterone levels using LC-MS/MS, the reference method for testosterone measurements (15).

Our study also has some limitations. Epidemiologic studies can define associations but not causality. The FHS population is largely white and of European ancestry; therefore, our findings may not be generalizable to other races/ethnicities. We did not have sex hormones measured at examination 8 to evaluate the relationship between age-related changes in testosterone and SHBG levels and incident metabolic syndrome. We did not measure estradiol and are unable to assess the possible role of aromatization on these outcomes. Testosterone levels are affected by pulsatile, diurnal, and circannual rhythms, and single samples ignore rhythmic hormone secretion. Our analyses show that single, early-morning SHBG levels, obtained in a manner similar to that used by physicians in clinical practice, were associated with incident metabolic syndrome. The FHS cohort was younger and healthier than some other epidemiologic studies, resulting in lower rates of metabolic syndrome.

These data have clinical implications and suggest that previously reported relationships between testosterone and metabolic syndrome should be interpreted cautiously (1–8). Our data do not support the notion that low testosterone is independently related to metabolic syndrome. However, on the basis of the CIs obtained in the regressions of testosterone on incident metabolic syndrome, we cannot exclude the possibility of effects as high as an odds ratio of 1.45 for total testosterone in the age-adjusted model and 1.04 in the fully adjusted model (1.27 and 1.09, respectively, for free testosterone). Testosterone levels are strongly associated with SHBG levels; because of this colinearity between testosterone and SHBG, it is not surprising that total testosterone also is associated with metabolic syndrome. Our findings assert that prevention strategies should focus on factors such as such as adiposity, insulin resistance, and inflammation, which may affect SHBG levels as well as the risk of metabolic syndrome rather than on testosterone therapy.

We do not know whether SHBG is a marker of metabolic syndrome or whether it is causally linked to metabolic syndrome. Additional research is needed to determine whether SHBG is an early predictor of metabolic syndrome and its components or whether it should be considered as an additional component of the metabolic risk assessment. The mechanistic basis of the relationship between SHBG and metabolic syndrome is poorly understood. SHBG can modulate the bioavailability and the biologic effects of testosterone and estradiol on the target tissues (23). SHBG-bound testosterone can be internalized by endocytosis through the megalin receptor (24). SHBG may have testosterone-independent actions (23); it has been reported to regulate the growth of prostate cancer cells and to interact with estrogen receptor.

Our multivariable analyses, consistent with previous reports, suggest that age and BMI independently affect the risk of metabolic syndrome as well as the circulating concentrations of SHBG and total and free testosterone. SHBG levels are independently associated with the risk of metabolic syndrome as well as with total and free testosterone levels. Our analyses suggest that age, BMI, and insulin sensitivity, rather than testosterone, may be the independent pathophysiologic determinants of the risk of metabolic syndrome; among these, BMI and insulin sensitivity are potentially modifiable risk factors. Insulin is known to inhibit hepatic SHBG production (25). Age, BMI, and insulin sensitivity could independently affect SHBG, testosterone, and metabolic risk. Although caution is warranted because of the colinearity and bidirectionality of some of these relationships, overall, our analyses do not reveal an independent prospective relationship between testosterone and incident metabolic syndrome.

This project was supported primarily by National Institutes of Health Grant 1RO1-AG-31206 (to S.B. and R.S.V.). Additional support was provided by the Boston Claude D. Pepper Older Americans Independence Center Grant 5P30AG031679 from the National Institute on Aging and by a grant from the Centers for Disease Control and Prevention Foundation. The Framingham Heart Study is supported by the National Heart, Lung, and Blood Institute’s Framingham Heart Study Contract (N01-HC-25195).

No potential conflicts of interest relevant to this article were reported.

S.B. wrote the manuscript. G.K.J. and M.P. performed the analyses and wrote portions of the manuscript. R.D., A.D.C., and R.S.V. reviewed the manuscript and contributed to discussion. T.G.T. performed the analyses and wrote portions of the manuscript.