Emerging data indicate that sex hormone-binding globulin (SHBG) as well as adipose tissue function may play important roles in the development of type 2 diabetes[1–3] and cardiovascular diseases[4, 5]. SHBG is synthesized primarily in the liver and binds to androgens and estrogens, thereby regulating the biologically active fraction of sex hormones[6]. Thus, the circulating SHBG level is a major determinant of the metabolic clearance of sex hormones, modulating the access of such sex hormones to their target tissues. The plasma membranes of various kinds of cells have been recently shown to be capable of binding specifically and with a high affinity to SHBG, mediating sex hormone signaling at the cell membrane through SHBG receptors[7, 8].

Recent molecular epidemiologic studies mainly from Western populations have reported that genetically determined levels of SHBG were inversely associated with type 2 diabetes risk[2, 9], lending support to the roles of SHBG in the development of type 2 diabetes. Importantly, fatty liver has been recently suggested to be a major determinant of decreased SHBG levels[10]. Because fatty liver is a strong risk factor of type 2 diabetes[11], investigating whether fatty liver accounts for the SHBG and type 2 diabetes relation may provide further insight into the role of SHBG in the pathogenesis of type 2 diabetes. However, the impact of liver function on the association has not been studied yet. Moreover, the previous studies examining the relation between SHBG and type 2 diabetes are mainly from European populations.

In addition, a sexually-dimorphic relationship between total testosterone levels and type 2 diabetes has been suggested. Studies in rodents reported that male androgen receptor knockout mice manifested late onset obesity with increased lipogenesis, insulin resistance, and impaired glucose tolerance[12–14], while obesity did not develop in the female androgen receptor knockout mice[14]. Further, a systematic review and meta-analysis reported that lower testosterone levels are associated with diabetes in men, whilst higher testosterone levels are associated with diabetes in women[1].

The purpose of this study was 1) to examine the associations of SHBG and testosterone with type 2 diabetes by sex among a Japanese population, independent of traditional diabetes risk factors, if such associations exist, 2) to evaluate whether the associations are independent of the fatty liver index (FLI), a validated measure of fatty liver, and 3) to examine whether sexual-dimorphism exists in the associations.

After multivariable adjustment for age, family history of diabetes, smoking, physical activity, and BMI (Table3; Model 2), SHBG levels were inversely associated with diabetes among women (OR comparing the highest with the lowest quartiles, 0.13 [95% CI, 0.02–0.86]), but not among men (OR, 0.73 [95% CI, 0.38–1.40]). Further adjustment for FLI slightly attenuated the association, but the strong inverse association observed in women remained (OR, 0.13 [95% CI, 0.02–0.96]) (Table3; Model 3), suggesting that the inverse association of SHBG with diabetes is independent of fatty liver. Adjustment for TG, waist, and GGT in Model 3 instead of FLI did not materially alter the results. Spline curves clearly indicated that higher SHBG levels were markedly associated with a reduced chance of having diabetes in women, but not in men (Figure1C and1D). In addition, in a subgroup analysis restricted to postmenopausal women, similar patterns were observed. As compared with the lowest quartile of the intake, the adjusted ORs of diabetes across increasing quartiles were 1.00 (reference), 0.14 (95% CI, 0.01–1.48), 0.20 (95% CI, 0.04–1.07), and 0.12 (95% CI, 0.01–1.17) (P-value for trend = 0.07)among premenopausal women, however, we were not able to observe any pattern due to the small sample size (15 cases and 13 controls). P-value for interaction by sex was 0.007, indicating the importance of considering the sex-specific effects of SHBG. An additional adjustment for fasting insulin and total testosterone levels had little impact on the estimates (Table3; Model 4).

Men	Quartiles
	Q1 (lowest)	Q2	Q3	Q4 (highest)	P for trend*
Testosterone (ng/mL)
Median (range)	3.04 (0.06–3.84)	4.42 (3.85–4.88)	5.44 (4.90–6.20)	6.96 (6.21–12.80)
(cases/controls)	69/39	54/55	49/56	43/64
Model 1†	1.00	0.55 (0.31–0.98)	0.48 (0.27–0.84)	0.37 (0.21–0.66)	<0.001
Model 2‡	1.00	0.72 (0.38–1.37)	0.62 (0.32–1.20)	0.44 (0.23–0.84)	0.01
Model 3§	1.00	0.66 (0.34–1.29)	0.64 (0.33–1.26)	0.45 (0.23–0.89)	0.03
Model 4||	1.00	0.64 (0.32–1.30)	0.63 (0.29–1.34)	0.40 (0.18–0.88)	0.03
SHBG (nmol/L)
Median (range)	42.1 (16.8–52.4)	62.2 (52.5–72.4)	81.8 (72.5–93.6)	112.0 (93.8–161.0)
(cases/controls)	58/50	55/54	55/52	47/59
Model 1†	1.00	0.87 (0.50–1.51)	0.87 (0.50–1.54)	0.67 (0.38–1.18)	0.17
Model 2‡	1.00	0.81 (0.43–1.51)	1.04 (0.55–2.00)	0.73 (0.38–1.40)	0.44
Model 3§	1.00	0.85 (0.44–1.64)	1.15 (0.59–2.24)	0.79 (0.41–1.53)	0.59
Model 4||	1.00	1.04 (0.52–2.10)	1.55 (0.73–3.27)	1.22 (0.56–2.67)	0.54
Women
Testosterone (ng/mL)
Median (range)	0.08 (0.04–0.10)	0.13 (0.11–0.15)	0.19 (0.16–0.22)	0.27 (0.23–0.48)
(cases/controls)	20/24	19/22	22/21	21/14
Model 1†	1.00	1.09 (0.49–2.47)	1.33 (0.55–3.17)	1.83 (0.76–4.40)	0.17
Model 2‡	1.00	0.76 (0.22–2.69)	1.05 (0.29–3.81)	1.90 (0.54–6.68)	0.27
Model 3§	1.00	0.72 (0.20–2.62)	0.97 (0.25–3.79)	2.02 (0.54–7.54)	0.25
Model 4||	1.00	0.74 (0.13–4.09)	2.82 (0.36–22.10)	9.20 (0.80–105.15)	0.046
SHBG (nmol/L)
Median (range)	49.5 (12.9–60.8)	76.6 (61.3–89.4)	105.0 (89.5–124.0)	139.5 (125.0–163.0)
(cases/controls)	33/10	20/22	20/23	12/30
Model 1†	1.00	0.35 (0.13–0.91)	0.32 (0.13–0.78)	0.15 (0.05–0.40)	<0.001
Model 2‡	1.00	0.17(0.03–0.91)	0.19 (0.05–0.79)	0.12 (0.02–0.86)	0.03
Model 3§	1.00	0.18 (0.03–0.98)	0.20 (0.05–0.87)	0.13 (0.02–0.96)	0.048
Model 4||	1.00	0.21 (0.02–2.78)	0.23 (0.03–1.93)	0.03 (0.001–0.86)	0.06

Data are ORs (95% CI). Abbreviations: SHBG, sex hormone-binding globulin; ORs, odds ratios; CI, confidence interval * P- values for trend are based on median levels in categories. † Model 1: stratified on matched pairs using conditional logistic regression models. ‡ Model 2: further adjusted for smoking status, physical activity, history of hypertension, family history of diabetes, alcohol use, menopausal status (women only), and BMI. § Model 3: further adjusted for the fatty liver index (quartiles). || Model 4: further adjusted for levels of fasting insulin (quartiles), testosterone (quartiles), and SHBG (quartiles) that were not examined as the primary independent variable.

In contrast, after multivariable adjustment for known risk factors, total testosterone levels were inversely associated with diabetes among men (OR, 0.44 [95% CI, 0.23–0.84]), whereas a strong positive relation was suggested among women, although the estimates were very imprecise (Table3; Model 2). Spline curves suggested a moderate inverse association between total testosterone and diabetes in men, but not in women (Figure1A and1B), suggesting that the total testosterone-diabetes relation observed among women (Table3) was a statistically unstable estimate. The P-value for interaction by sex was 0.29, indicating insufficient evidence to support a sex-difference in the total testosterone and type 2 diabetes relation. Further adjustment for FLI (Table3; Model 3), and SHBG and insulin levels (Table3; Model 4) did not materially alter the estimates. In addition, we examined the relation between free testosterone levels and diabetes by sex. In a manner similar to total testosterone, after multivariable adjustment for known risk factors (Model 2), free testosterone levels were inversely associated with diabetes among men (OR, 0.44 [95% CI, 0.23–0.85]), but the estimates for women were very imprecise (OR, 15.28 [95% CI, 2.25–103.86]) (data not shown). The P-value for interaction by sex was 0.008. This suggests a sexual-dimorphism in the relation between free testosterone and diabetes; however, the effect measure modification might have been mainly driven by the SHBG-diabetes relation. Spline curves suggested a moderate inverse association between free testosterone and diabetes in men, but not in women (data not shown).

In this well-characterized case–control study in a Japanese population, we identified SHBG levels in women and total and free testosterone levels in men to be inversely associated with diabetes after multivariable adjustment for FLI and fasting insulin as well as known risk factors for diabetes. These findings suggest that SHBG in women and testosterone in men are not merely markers of liver function or hyperinsulinemia, adding support to the notion that SHBG and sex hormones may play important roles in the development of type 2 diabetes.

The findings for SHBG levels are in general agreement with previous findings. A systematic review and meta-analysis showed that SHBG was inversely associated with incident and prevalent diabetes in both men and women after adjustment for known risk factors for type 2 diabetes and the inverse association was stronger in women than in men[1]. In addition, previous studies in multi-ethnic populations have reported that the association remained consistent across ethnic groups: African-Americans, Hispanics, and Asians[25, 26]. We did not observe the association in men and the interaction for gender was statistically significant (P = 0.007). Importantly, an earlier study in a largely Hispanic-American population from the San Antonio Heart Study found that the inverse association between SHBG and diabetes risk was strong in pre-menopausal women, weak in postmenopausal women, and negligible in men[27], suggesting a possible sexual-dimorphism in the association. Several studies examined whether the SHBG-diabetes relation was independent of sex hormones and reported that the relation remained[25, 26].We observed that further adjustment for total testosterone levels had little impact on the estimates, suggesting that the mechanisms whereby SHBG contributes to diabetes development may be independent of testosterone[28]. Further, we investigated the impact of fatty liver on the SHBG-diabetes association with the FLI and found that further adjustment for FLI slightly attenuated the SHBG-diabetes association, but the association remained, indicating that SHBG may not merely be a marker of liver function, but may also be associated with diabetes, independent of fatty liver. Furthermore, recent genetic studies provided compelling evidence that genetically determined SHBG levels as well as variants of the SHBG gene are associated with type 2 diabetes risk[2]. Taken together, these findings support the role of SHBG in the development of diabetes.

Although the exact mechanisms by which SHBG influences the diabetes risk remain largely unclear, SHBG may contribute to the impairment of glucose metabolism through modulation of sex hormone bioavailability and direct activation of a specific receptor for SHBG[28]. Compensatory hyperinsulinemia resulting from insulin resistance has been proposed to be responsible for the apparent inverse SHBG-diabetes relationship[29]; however, it has been shown recently that a significant increase rather than decrease occurs in plasma SHBG levels after initiating insulin treatment in type 2 diabetic patients[30].

Total and free testosterone levels were inversely associated with diabetes among men, while a possible positive association was suggested among women after adjustment for known risk factors for type 2 diabetes in our study. These findings are in line with previous studies. In a systematic review and meta-analysis, Ding et al. found an inverse association between testosterone and diabetes in men; they found a positive association between testosterone and diabetes in women[1]. Growing evidence in human and rodents supports the roles of androgen in the glucose metabolism. Testosterone replacement resulted in improved glycemic control in hypogonadal men with type 2 diabetes[31] and androgen receptor knockout male mice manifested insulin resistance and impaired glucose tolerance[12]. Decreased testosterone levels may contribute to the development of diabetes by increasing insulin resistance via decreasing lean mass and increasing fat mass[32]. The inverse testosterone-diabetes relation in men remained after adjusting for BMI and fasting insulin in our study, suggesting that there may be an alternative pathway relating testosterone and diabetes.

Categorical analyses (Models 1–4) in our study suggested that, in contrast to men, total and free testosterone levels appeared to be positively associated with diabetes in women. The interaction for gender was significant for free testosterone (P-value = 0.008), although this might have been driven by the sex-difference in the SHBG-diabetes relation. However, the dose–response relation between testosterone levels and diabetes using cubic spline models indicated that no clear conclusions can be drawn regarding the testosterone-diabetes relation among women. Of importance, the total testosterone levels among women were far smaller than the levels among men; the median values of testosterone in the highest quartile of testosterone levels were 0.27 ng/mL among women and 6.96 ng/mL among men. This difference makes it difficult to directly compare the relation of testosterone with diabetes between men and women. Further experimental and observational studies are needed to better elucidate the sex-specific role of testosterone in the regulation of glucose metabolism.

Our study has several strengths. First, about 95% of the participants who did not self-report diabetes underwent the 75 g oral glucose tolerance test, which minimized the possibility of the inclusion of undiagnosed diabetic patients in the control group. Second, we obtained detailed measures of variables, allowing us to control for a number of potential confounding factors. Third, the control participants were randomly selected from the same population that gave rise to the cases which minimized the selection bias. Nevertheless, some limitations of the present study merit consideration. First, we were unable to establish a temporal relationship between the exposure (i.e., testosterone and SHBG) and diabetes. In addition, the observed associations may vary according to the duration of diabetes. Hyperinsulinemia or insulin resistance typically associated with diabetes might have resulted in decreased testosterone and SHBG levels (i.e., reverse causality). Further controlling for fasting insulin levels did not materially alter the results in our study, but use of sulphonylurea or insulin among participants with diabetes might have influenced the insulin levels. Therefore, the adjustment for fasting insulin levels may not have adequately captured the influence of hyperinsulinemia or insulin resistance. Second, we used the FLI, a validated surrogate maker for fatty liver, to examine the impact of fatty liver on the SHBG-diabetes association. Although this measure has been validated in independent studies, the FLI is not a perfect measure of fatty liver. Thus, residual confounding due to misclassification of fatty liver cannot be ruled out. Third, we did not ask women about the use of exogenous estrogen, which might have confounded the SHBG-diabetes association. The frequency of exogenous estrogen use in Japan is quite low (≈ 2.8%)[33], so such confounding would probably have little impact on our estimates. Finally, measurement errors for the exposures (i.e., testosterone and SHBG) might have biased the results. However, the errors are likely to be nondifferential with respect to the outcome, because case–control pairs were matched and laboratory personnel were blinded to case–control status. Therefore, such misclassifications would lead to an underestimation of the associations. Because most women had low testosterone values in our study and the immune assay in our study may be less sensitive than other assays, such as liquid chromatography–mass spectrometry, the possible inaccurate quantification at low values might have attenuated the association of testosterone with diabetes in women.

Our findings suggest that SHBG in women and testosterone in men are associated with diabetes in a Japanese population. These findings lend support to the notion that SHBG and sex hormones may play important roles in the development of type 2 diabetes.