The aim of this study was to investigate the associations among sex hormone–binding globulin (SHBG), visceral adipose tissue (VAT), liver fat content, and risk of type 2 diabetes (T2D). In the Netherlands Epidemiology of Obesity study, 5,690 women (53%) and men (47%) without preexisting diabetes were included and followed for incident T2D. SHBG concentrations were measured in all participants, VAT was measured using MRI, and liver fat content was measured using proton magnetic resonance spectroscopy in a random subset of 1,822 participants. We examined associations between SHBG and liver fat using linear regression and bidirectional Mendelian randomization analyses and between SHBG and T2D using Cox regression adjusted for confounding and additionally for VAT and liver fat to examine mediation. Mean age was 56 (SD 6) years, mean BMI was 30 (SD 4) kg/m2, median SHBG was 47 (interquartile range [IQR] 34–65) nmol/L in women and 34 (26–43) nmol/L in men, and median liver fat was 3.4% (IQR 1.6–8.2%) in women and 6.0% (2.9–13.5%) in men. Compared with the highest SHBG quartile, liver fat was 2.9-fold (95% CI 2.4, 3.4) increased in women and 1.6-fold (95% CI 1.3, 1.8) increased in men, and the hazard ratio of T2D was 4.9 (95% CI 2.4, 9.9) in women and 1.8 (1.1, 2.9) in men. Genetically predicted SHBG was associated with liver fat content (women: SD −0.45 [95% CI −0.55, −0.35]; men: natural logarithm, −0.25 [95% CI −0.34, −0.16]). VAT and liver fat together mediated 43% (women) and 60% (men) of the SHBG-T2D association. To conclude, in a middle-aged population with overweight, the association between low SHBG and increased risk of T2D was, for a large part, mediated by increased VAT and liver fat.

Article Highlights
  • The objective of this study was to unravel the associations between sex hormone–binding globulin (SHBG) and type 2 diabetes (T2D).

  • The extent to which the association between low SHBG and increased risk of T2D was mediated by visceral adipose tissue (VAT) and liver fat was examined.

  • VAT and liver fat were found to mediate a large part of the association between SHBG and T2D.

  • Both lifestyle and medical interventions that increase SHBG may reduce VAT, liver fat, and the risk of T2D.

Low serum concentrations of sex hormone–binding globulin (SHBG) have been related to an increased risk of type 2 diabetes (T2D) (1–3). This relationship has been confirmed by multiple Mendelian randomization studies suggesting that SHBG is likely involved in the etiology of T2D (4–6). However, the pathophysiology behind this relationship remains unclear.

SHBG has also been associated with liver fat content (7,8), which may be one of the mechanisms underlying the relationship between SHBG and T2D. In observational and Mendelian randomization studies, fatty liver, recently redefined as metabolic dysfunction–associated steatotic liver disease, has been associated with an increased risk of T2D (9). Likewise, excess visceral adipose tissue (VAT), another well-established risk factor for T2D, has been associated with low SHBG concentrations as well (10–12). Moreover, a recent Mendelian randomization study found evidence for a causal relationship between low SHBG and VAT (13).

We hypothesized that VAT and liver fat content play a role in the relationship between SHBG and T2D. However, previous literature on the direction of the association between low serum SHBG and fatty liver has been inconsistent. On the one hand, a low SHBG itself may contribute to the development of fatty liver (14). It has been suggested that SHBG regulates hepatic lipogenesis and that individuals with low serum SHBG are more susceptible to fat accumulation in the liver (1,14–16). This would imply that a low SHBG concentration is a risk factor not only for developing T2D but also for developing fatty liver. On the other hand, excessive liver fat accumulation may downregulate the hepatic expression and concentrations of SHBG since SHBG is primarily produced in the liver (17,18). Mendelian randomization studies using genetic instruments for SHBG and liver fat content may shed light on the direction of this association.

Therefore, the aim of our study was to investigate the relationship between SHBG and liver fat content, both in an observational setting and using a bidirectional Mendelian randomization approach. We aimed to examine to what extent the SHBG-T2D association was mediated by the amount of VAT and liver fat content. The study of these interrelationships may contribute to our understanding of the pathophysiology, prevention, and treatment strategies of T2D and to refining risk stratification.

The Netherlands Epidemiology of Obesity Study Design and Study Population

The Netherlands Epidemiology of Obesity (NEO) study is a population-based, prospective cohort study in 6,671 middle-aged women and men (45–65 years), with an oversampling of individuals meeting criteria for overweight or obesity (19). Between 2008 and 2012, participants were included at the NEO study center of the Leiden University Medical Center and completed questionnaires to report demographic, lifestyle, and clinical information. At the study center, all participants underwent an extensive physical examination, including blood sampling, and were screened for potential health risks and MRI interference (e.g., metallic devices, claustrophobia, body circumference >1.70 m). Of the eligible participants, 2,580 were randomly selected to undergo MRI. After the baseline visit, participants were followed for the occurrence of T2D. For the present analyses, we excluded participants with preexisting diabetes or missing values or who consumed >10 alcoholic beverages per day. The NEO study was approved by the medical ethical committee of the Leiden University Medical Center. All participants gave their written informed consent.

Data Collection

Participants reported ethnicity by self-identification in eight categories (White, Black, Hindustani, Surinamese, Moroccan, Southeast Asian, Turkish, other), which we grouped into White (>95%, reference) and other. Tobacco use was reported as current, former, and never (reference) smoking. Highest education level was reported in 10 categories according to the Dutch education system and grouped into high (higher vocational school, university, and postgraduate education) and other education (reference). The Short Questionnaire to Assess Health-Enhancing Activity was used to report physical activity during leisure time, which is expressed in MET-hours per week. Based on a semiquantitative self-administered 125-item food frequency questionnaire (20), we calculated the Dutch healthy diet index (DHDI) (21) (score range 0–150). Height was measured without shoes using a calibrated, vertically fixed tape measure. Body weight and body fat percentage were estimated by a bioimpedance device (TBF-310; Tanita International Division, Manchester, U.K.), and 1 kg was subtracted from the body weight to compensate for the weight of clothing. BMI was calculated by dividing body weight in kilograms by body height in meters squared. For women, we grouped the use of contraceptives and hormone replacement therapy into current use of estrogens and no current use (reference) of hormones. Menopausal state was categorized as premenopausal and postmenopausal (reference) according to information on oophorectomy, hysterectomy, and self-reported menopausal state.

SHBG Measurements

Fasting blood samples were drawn from the antecubital vein after 5 min of rest, and aliquots were stored at −80°C. In 2021, frozen serum samples were transported to the Endocrine Laboratory of the Amsterdam University Medical Center, where SHBG concentrations were measured using an automated immunoassay (Architect; Abbott Diagnostics), with a lower limit of quantitation of 4.5 nmol/L and an interassay coefficient of variation of <6.5% and 4.9% over the whole concentration range.

VAT and Liver Fat Content

VAT was quantified by a turbo spin echo imaging protocol using MRI. All imaging was performed on a 1.5-T magnetic resonance (MR) system (Philips Medical Systems, Best, the Netherlands). At the level of the fifth lumbar vertebra, three transverse images, each with a slice thickness of 10 mm, were obtained during a breath hold. VAT area was converted from number of pixels to centimeters squared using in-house–developed software (MASS; Medis, Leiden, the Netherlands), and the average of three slices was used in the analyses (19).

Proton MR spectroscopy (MRS) of liver fat content was performed, and spectra were obtained. MRS data were fitted using Java-based MR user interface software (version 2.2; jMRUI, Leuven, Belgium) (22). Liver fat content relative to water was calculated as the sum of signal amplitudes of methyl and methylene divided by the signal amplitude of water and then multiplied by 100. Fatty liver was defined as a liver fat content of ≥5.56% (23).

Incident Diagnoses of T2D

Data on diagnoses of T2D were extracted from the electronic health records of the participants’ general practitioners (24). Health records were screened for the International Classification of Primary Care code T90 (diabetes mellitus) or T90.02 (diabetes mellitus type 2) and for the use of insulin, metformin, and sulfonylurea derivatives. In case of uncertainty, free text in the health records for signs of diabetes was studied, or the participant’s general practitioner was contacted to confirm the diagnosis. The index date was defined by the first date of an International Classification of Primary Care–coded diagnosis or the first date of a prescription for antidiabetic medication. Follow-up time was defined as the number of days between the baseline visit and date of diagnosis or censoring due to death, loss to follow-up, or the end of the follow-up (extraction date in 2013 or 2018), whichever came first.

Statistical Analyses

Because SHBG, VAT, and liver fat content differed by sex, we analyzed women and men separately. Baseline characteristics are expressed as mean (SD), median (interquartile range [IQR]), or percentage and stratified by sex-specific quartiles of SHBG.

First, we visually inspected scatterplots for linearity of the associations. Next, we examined the associations between quartiles of SHBG and liver fat content with linear regression analyses using the highest quartile of SHBG as the reference. Because of a skewed distribution, we used the natural logarithm (ln) of liver fat content as the outcome in the linear regression analyses. For interpretation of the results, we transformed the regression coefficients into ratios, as follows: [exp(β) − 1] if β >0 and −[1 / exp(−β) − 1) if β <0.

Next, we calculated the incidence rates of T2D per quartile of SHBG, expressed per 1,000 person-years (py). Cox proportional hazard models were used to calculate hazard ratios (HRs) with 95% CIs for the association between SHBG and the risk of T2D, using the highest quartile of SHBG as the reference category. All analyses were adjusted for potential confounding (model 2): age (years), smoking (current, former, never), alcohol consumption (g/day), physical activity during leisure time (MET-h/week), menopausal state (pre, post), use of hormone therapy (current, no current use), ethnicity (White, other), education level (high, other), and DHDI (score 0–150). In model 3, we additionally adjusted for total body fat (%).

Mediation Analyses

To examine the role of VAT and liver fat content in the SHBG-T2D association, we performed mediation analysis per Baron and Kenny (25). Figure 1 shows an illustration of our hypothesis. First, we examined whether the associations between SHBG and liver fat content, SHBG and VAT, liver fat content and T2D, and VAT and T2D were present in our study. We tested for exposure-mediator interaction and adjusted for mediator-outcome confounding. We examined the association between VAT and liver fat content and T2D in our population using multivariable Cox regression analyses and adjusted for the aforementioned potential confounding and SHBG.

Figure 1

Hypothesis path diagram illustrating mediation of the association between SHBG and T2D by VAT and liver fat content. Path C represents the total association between SHBG and T2D, path C′ shows the direct association, and paths A and B show the indirect association via VAT and liver fat. Every association was corrected for possible confounding factors (CFs): age (years), smoking (current, former, never), alcohol intake (g/day), DHDI (score 0–150), hormone use (current, past, never), menopausal status (pre, post), ethnicity (White, other), education (high, other), and total body fat (%). The association between VAT and T2D as well as liver fat content and T2D was additionally adjusted for SHBG in nmol/L.

Figure 1

Hypothesis path diagram illustrating mediation of the association between SHBG and T2D by VAT and liver fat content. Path C represents the total association between SHBG and T2D, path C′ shows the direct association, and paths A and B show the indirect association via VAT and liver fat. Every association was corrected for possible confounding factors (CFs): age (years), smoking (current, former, never), alcohol intake (g/day), DHDI (score 0–150), hormone use (current, past, never), menopausal status (pre, post), ethnicity (White, other), education (high, other), and total body fat (%). The association between VAT and T2D as well as liver fat content and T2D was additionally adjusted for SHBG in nmol/L.

Close modal

Finally, in case all assumptions were met, we examined the extent of mediation by VAT and liver fat of the association between SHBG and T2D by adding VAT and liver fat to the fully adjusted Cox regression model. We calculated proportions of mediation using the HR of the total association (HRC) and HR of the direct association (HRC′): (HRC − HRC′) / (HRC − 1) × 100.

We repeated all analyses after excluding women who were using hormone therapy and after excluding participants who were using additional SHBG-increasing medication (ATC codes), including anticonvulsant (N03), tamoxifen (L02BA01), H2 antagonists (A02BA01, A02BA02), lisinopril (C09AA03), tumor necrosis factor-α antagonist (L04AB), morphine (N02AA1), and antidepressants (N06A), or SHBG-decreasing medication, including progesterone (G03D), glucocorticoids (H02), danazol (G03XA), statins (C10AA, C10BA) (26), and after restricting the analyses to postmenopausal women.

Bidirectional Mendelian Randomization Analyses of SHBG and Liver Fat Content

With bidirectional summary-level two-sample Mendelian randomization analyses using publicly available databases, we investigated the direction of the relationship between SHBG and liver fat content. As candidate instruments for SHBG, we selected variants from a sex-specific genome-wide association study (GWAS) of White European UK Biobank participants (5) (189,473 women and 180,726 men), excluding variants located on the X chromosome that had not been considered for the GWAS on liver fat. More specifically, we selected instruments from the author-defined male SHBG and female SHBG variant clusters. As candidate instruments for MRS-derived liver fat content, we selected 12 genome-wide significant independent signals from a nonsex-specific GWAS of MRS-derived liver fat content in 32,858 White British UK Biobank participants (27). For Mendelian randomization analyses in both directions, weights (i.e., regression coefficients and SEs) were extracted from the BMI-unadjusted GWAS result files to limit the potential for collider bias, with summary statistics obtained from the GWAS catalog (https://www.ebi.ac.uk/gwas). We restricted summary statistics to those derived from populations of European ancestry to reduce the likelihood of population structure confounding the association between the instrumental variables and our traits of interest (Mendelian randomization’s independence assumption).

We calculated the median (IQR) and F statistic to gauge whether the instrumental variables associated strongly enough with the exposures of interest (Mendelian randomization’s relevance assumption). Where possible, we used HapMap2 proxies (r2 >0.5) previously identified by Ruth et al. (5) for SHBG instruments not available in the liver fat GWAS data set and identified a European 1,000 Genomes Proxy project in perfect linkage disequilibrium (r2 = 1) for a liver fat percentage instrument absent from the SHBG GWAS data sets. The number of instruments for the univariable Mendelian randomization analyses was 355 (of 368) for SHBG in women, 326 (of 350) for SHBG in men, and 12 for liver fat content. The random-effects inverse-variance–weighted (IVW) estimator was our main estimator.

To test for potential violations of the assumption that the instrumental variables are not related to the outcome other than via the outcome of interest (Mendelian randomization’s exclusion restriction assumption), we performed several sensitivity analyses, which relax the assumptions underlying the IVW estimator in complementary ways. These included Mendelian randomization pleiotropy residual sum and outlier (28), the contamination mixture method (29), and a mixture of experts machine learning approach, which were designed to choose the most appropriate model out of 14 different pleiotropy-robust Mendelian randomization estimation methods before and after Steiger filtering (30,31). In addition, we repeated the SHBG-to-liver fat analyses once with multivariable Mendelian randomization analyses, adjusting separately for genetically predicted BMI and for genetically predicted bioavailable testosterone (5,32). Next, we ran causal analysis using summary effect estimates (CAUSE), a Bayesian-based Mendelian randomization method that accounts for both correlated and uncorrelated pleiotropy and tests whether the association between traits is better explained by causality than by horizontal pleiotropy (33). Finally, we restricted our instruments to three conditionally independent genetic variants (rs1625895, rs12150660, and rs1641537) located in the SHBG gene (34) and calculated a fixed-effects IVW (FE-IVW) causal estimate. For the liver fat content-to-SHBG analyses, we performed one additional sensitivity analysis that restricted liver fat instruments to those with suggestive evidence of diverging effects on SHBG and triglycerides (opposing effects across GWAS at P < 5 × 10−6 for both), which would allude to downstream effects of de novo lipogenesis (17). For this, we performed a look-up of our 12 liver fat instruments in a sex-combined GWAS on log-triglycerides (35).

Data and Resource Availability

The data sets generated during and analyzed in the current study are available from the corresponding author upon reasonable request according to the NEO research procedure.

Baseline Characteristics

Of 6,671 participants, we consecutively excluded those with prevalent diabetes (n = 592); who were lost to follow-up (n = 94); with missing SHBG concentration (n = 75), smoking status (n = 6), alcohol intake (n = 1), physical activity (n = 103), total body fat (n = 27), ethnicity (n = 8), and education (n = 53) data; and who consumed ≥10 alcoholic beverages per day (n = 22), leaving 5,690 participants (53% women) for the analyses. The mean age of the total study population was 56 (SD 6) years, mean BMI was 30 (SD 5) kg/m2 in women and 30 (4) kg/m2 in men, and the median SHBG concentration was 46.7 (IQR 34.1–64.5) nmol/L in women and 33.6 (26.4–42.8) nmol/L in men. Baseline characteristics of the study population are shown in Table 1, stratified by quartiles of SHBG concentrations. Liver fat content was available in a random subset of 1,822 participants. Baseline characteristics of the population with assessment of liver fat content were similar to those without: mean age was 56 (SD 6) vs. 55 (6) years, mean BMI 30 (SD 6) vs. 29 (5) kg/m2in women and 30 (4) vs. 29 (3) kg/m2in men, and median SHBG concentration 53.9 (IQR 34.9–63.9) vs. 46.1 (34.5–67.2) nmol/L in women and 35.7 (13.2–42.7) vs. 33.1 (26.2–43.5) nmol/L in men.

Table 1

Baseline characteristics of middle-aged women and men without a history of diabetes, by quartiles of serum SHBG concentrations

CharacteristicTotalQuartile 1(SHBG <34.1 nmol/L)Quartile 2 (34.1 ≤ SHBG <46.7 nmol/L)Quartile 3(46.7 ≤ SHBG <64.5 nmol/L)Quartile 4(SHBG ≥64.5 nmol/L)
Women      
 Participants, n 3,031 759 759 757 756 
 Age (years) 55 (6) 56 (5) 56 (6) 56 (6) 54 (6) 
 SHBG (nmol/L) 46.7 (34.1–64.5) 27.1 (22.8–31.1) 40.3 (37.5–43.5) 54.8 (51.0–59.0) 81.6 (71.6–102.1) 
 BMI (kg/m230 (5) 32 (5) 31 (5) 29 (5) 28 (5) 
 Obesity (% BMI ≥30 kg/m246 64 52 41 28 
 Total body fat (%) 42 (6) 44 (5) 43 (5) 41 (6) 39 (7) 
 VAT (cm2)* 85 (57, 121) 118 (85, 154) 94 (70, 124) 79 (52, 110) 55 (40, 87) 
 Waist circumference (cm) 97 (13) 103 (12) 99 (12) 95 (13) 91 (13) 
 Hip circumference (cm) 111 (12) 115 (12) 113 (11) 110 911) 107 (11) 
 Waist-to-hip ratio 1.16 (0.09) 1.12 (0.07) 1.15 (0.08) 1.17 (0.09) 1.19 (0.09) 
 Liver fat content (%)* 3.39 (1.60–8.19) 8.13 (4.20–16.80) 4.07 (1.98–8.28) 2.79 (1.52–6.49) 1.49 (0.96–2.89) 
 Fatty liver (>5.56%)* 35 65 38 30 10 
 Postmenopausal 61 67 68 64 46 
 Current use of hormone therapy 23 
 Alcohol consumption (g/day) 6 (1–15) 5 (1–16) 7 (1–16) 5 (1–14) 5 (1–12) 
 Physical activity (MET-h/week) 27 (14–47) 25 (13–44) 27 (13–46) 30 (16–50) 27 (14–46) 
 Current smoker 13 10 12 15 16 
 Former smoker 48 55 51 43 43 
 Ethnicity (White) 95 93 95 95 95 
 Education level (low and medium) 65 70 67 63 59 
 DHDI 72 (14) 72 (14) 71 (14) 74 (14) 73 (14) 
CharacteristicTotalQuartile 1(SHBG <34.1 nmol/L)Quartile 2 (34.1 ≤ SHBG <46.7 nmol/L)Quartile 3(46.7 ≤ SHBG <64.5 nmol/L)Quartile 4(SHBG ≥64.5 nmol/L)
Women      
 Participants, n 3,031 759 759 757 756 
 Age (years) 55 (6) 56 (5) 56 (6) 56 (6) 54 (6) 
 SHBG (nmol/L) 46.7 (34.1–64.5) 27.1 (22.8–31.1) 40.3 (37.5–43.5) 54.8 (51.0–59.0) 81.6 (71.6–102.1) 
 BMI (kg/m230 (5) 32 (5) 31 (5) 29 (5) 28 (5) 
 Obesity (% BMI ≥30 kg/m246 64 52 41 28 
 Total body fat (%) 42 (6) 44 (5) 43 (5) 41 (6) 39 (7) 
 VAT (cm2)* 85 (57, 121) 118 (85, 154) 94 (70, 124) 79 (52, 110) 55 (40, 87) 
 Waist circumference (cm) 97 (13) 103 (12) 99 (12) 95 (13) 91 (13) 
 Hip circumference (cm) 111 (12) 115 (12) 113 (11) 110 911) 107 (11) 
 Waist-to-hip ratio 1.16 (0.09) 1.12 (0.07) 1.15 (0.08) 1.17 (0.09) 1.19 (0.09) 
 Liver fat content (%)* 3.39 (1.60–8.19) 8.13 (4.20–16.80) 4.07 (1.98–8.28) 2.79 (1.52–6.49) 1.49 (0.96–2.89) 
 Fatty liver (>5.56%)* 35 65 38 30 10 
 Postmenopausal 61 67 68 64 46 
 Current use of hormone therapy 23 
 Alcohol consumption (g/day) 6 (1–15) 5 (1–16) 7 (1–16) 5 (1–14) 5 (1–12) 
 Physical activity (MET-h/week) 27 (14–47) 25 (13–44) 27 (13–46) 30 (16–50) 27 (14–46) 
 Current smoker 13 10 12 15 16 
 Former smoker 48 55 51 43 43 
 Ethnicity (White) 95 93 95 95 95 
 Education level (low and medium) 65 70 67 63 59 
 DHDI 72 (14) 72 (14) 71 (14) 74 (14) 73 (14) 
TotalQuartile 1 (SHBG <26.4 nmol/L)Quartile 2 (26.4 ≤ SHBG <33.6 nmol/L)Quartile 3 (33.6 ≤ SHBG <42.8 nmol/L)Quartile 4 (SHBG >42.8 nmol/L)
Men      
 Participants, n 2,659 669 665 653 663 
 Age (years) 56 (6) 53 (7) 55 (6) 56 (6) 58 (6) 
 SHBG (nmol/L) 33.6 (26.4–42.8) 22.0 (18.6–24.4) 30.3 (28.3–32.0) 37.7 (35.7–39.8) 50.9 (46.3–58.0) 
 BMI (kg/m230 (4) 30 (4) 29 (4) 29 (4) 29 (4) 
 Obesity (BMI ≥30 kg/m238 43 39 38 32 
 Total body fat (%) 29 (6) 29 (5) 29 (6) 28 (6) 28 (6) 
 Liver fat content (%)* 5.97 (2.87–13.52) 9.44 (4.06–16.64) 5.91 (3.43–12.84) 5.47 (2.56–10.55) 4.27 (2.13–10.37) 
 VAT (cm2)* 134 (98, 172) 141 (104, 190) 136 (109, 176) 129 (103, 172) 123 (85, 161) 
 Waist circumference (cm) 105 (11) 107 (10) 106 (11) 105 (11) 104 (11) 
 Hip circumference (cm) 108 (7) 109 (7) 108 (7) 108 (7) 107 (7) 
 Waist-to-hip ratio 1.03 (0.07) 1.03 (0.06) 1.03 (0.07) 1.03 (0.07) 1.04 (0.07) 
 Fatty liver (>5.56%)* 53 66 54 49 44 
 Alcohol (g/day) 17 (5–29) 17 (5–30) 17 (7–30) 15 (4–29) 16 (6–28) 
 Physical activity (MET-h/week) 28 (14–49) 24 (12–41) 31 (15–51) 29 (15–49) 33 (15–54) 
 Current smoker 19 14 18 21 24 
 Former smoker 49 50 52 46 48 
 Ethnicity (White) 96 95 96 96 96 
 Education level (low and medium) 57 53 55 59 62 
 DHDI 66 (14) 65 (14) 65 (14) 67 (14) 66 (15) 
TotalQuartile 1 (SHBG <26.4 nmol/L)Quartile 2 (26.4 ≤ SHBG <33.6 nmol/L)Quartile 3 (33.6 ≤ SHBG <42.8 nmol/L)Quartile 4 (SHBG >42.8 nmol/L)
Men      
 Participants, n 2,659 669 665 653 663 
 Age (years) 56 (6) 53 (7) 55 (6) 56 (6) 58 (6) 
 SHBG (nmol/L) 33.6 (26.4–42.8) 22.0 (18.6–24.4) 30.3 (28.3–32.0) 37.7 (35.7–39.8) 50.9 (46.3–58.0) 
 BMI (kg/m230 (4) 30 (4) 29 (4) 29 (4) 29 (4) 
 Obesity (BMI ≥30 kg/m238 43 39 38 32 
 Total body fat (%) 29 (6) 29 (5) 29 (6) 28 (6) 28 (6) 
 Liver fat content (%)* 5.97 (2.87–13.52) 9.44 (4.06–16.64) 5.91 (3.43–12.84) 5.47 (2.56–10.55) 4.27 (2.13–10.37) 
 VAT (cm2)* 134 (98, 172) 141 (104, 190) 136 (109, 176) 129 (103, 172) 123 (85, 161) 
 Waist circumference (cm) 105 (11) 107 (10) 106 (11) 105 (11) 104 (11) 
 Hip circumference (cm) 108 (7) 109 (7) 108 (7) 108 (7) 107 (7) 
 Waist-to-hip ratio 1.03 (0.07) 1.03 (0.06) 1.03 (0.07) 1.03 (0.07) 1.04 (0.07) 
 Fatty liver (>5.56%)* 53 66 54 49 44 
 Alcohol (g/day) 17 (5–29) 17 (5–30) 17 (7–30) 15 (4–29) 16 (6–28) 
 Physical activity (MET-h/week) 28 (14–49) 24 (12–41) 31 (15–51) 29 (15–49) 33 (15–54) 
 Current smoker 19 14 18 21 24 
 Former smoker 49 50 52 46 48 
 Ethnicity (White) 96 95 96 96 96 
 Education level (low and medium) 57 53 55 59 62 
 DHDI 66 (14) 65 (14) 65 (14) 67 (14) 66 (15) 

Data are mean (SD), median (IQR), or percentage unless otherwise indicated.

*

Liver fat content and VAT were available in 887 women and 935 men.

SHBG and Liver Fat Content

Higher SHBG concentrations were associated with a lower liver fat content in both women and men. The relationship between SHBG and liver fat content appeared nonlinear, so we further investigated SHBG in sex-specific quartiles. After adjustment for confounding factors, women in the lowest quartile of SHBG had a 2.9-fold (95% CI 2.4, 3.4) higher liver fat content than women in the highest quartile of SHBG (Fig. 2). In men in the lowest quartile of SHBG, this ratio was 1.6 (1.3, 1.8) compared with men in the highest quartile of SHBG.

Figure 2

The relationship between liver fat content and serum SHBG concentration and between VAT and serum SHBG concentration. Results of the association between SHBG and liver fat content were derived from regression coefficients with 95% CIs from linear regression analyses and expressed as relative change, with the highest quartile of SHBG as the reference category. Such relative change can be interpreted as a ratio, for example, women in the lowest quartile of SHBG have a 2.87-fold higher liver fat content compared with women in the highest quartile. The figures on the association between SHBG and VAT present absolute differences, with the highest quartile of SHBG as the reference category, for example, the VAT of women in the lowest quartile of SHBG is 30 cm2 higher compared with women in the highest quartile. All results are adjusted for age (years), smoking (current, former, never), alcohol consumption (g/day), physical activity during leisure time (MET-h/week), menopausal state (pre, post), use of hormone therapy (currently, no current use), ethnicity (White, other), DHDI (score 0–150), education level (high, other), and total body fat (%).

Figure 2

The relationship between liver fat content and serum SHBG concentration and between VAT and serum SHBG concentration. Results of the association between SHBG and liver fat content were derived from regression coefficients with 95% CIs from linear regression analyses and expressed as relative change, with the highest quartile of SHBG as the reference category. Such relative change can be interpreted as a ratio, for example, women in the lowest quartile of SHBG have a 2.87-fold higher liver fat content compared with women in the highest quartile. The figures on the association between SHBG and VAT present absolute differences, with the highest quartile of SHBG as the reference category, for example, the VAT of women in the lowest quartile of SHBG is 30 cm2 higher compared with women in the highest quartile. All results are adjusted for age (years), smoking (current, former, never), alcohol consumption (g/day), physical activity during leisure time (MET-h/week), menopausal state (pre, post), use of hormone therapy (currently, no current use), ethnicity (White, other), DHDI (score 0–150), education level (high, other), and total body fat (%).

Close modal

SHBG and T2D

During a median follow-up of 6.7 (IQR 5.9–7.9) years, 129 women developed T2D, reflecting an incidence rate of 6.4 per 1,000 py (Table 2). In men, 154 incident events of T2D occurred, reflecting an incidence rate of 8.8 per 1,000 py. In women in the highest SHBG quartile, the incidence rate of T2D was 2.0 per 1,000 py, and in the lowest quartile of SHBG, it was 15.4 per 1,000 py. In men, incidence rates were 6.6 per 1,000 py in the highest quartile of SHBG and 11.1 per 1,000 py in the lowest. The HR in women with SHBG concentrations in the lowest quartile compared with the highest quartile, after correction for confounding factors, was 4.89 (95% CI 2.41, 9.90), and in men, it was 1.78 (1.10, 2.88) (Table 2).

Table 2

Incidence rates and risks of T2D by sex-specific quartiles and percentage of mediation by liver fat and VAT in middle-aged women and men

Quartile 1(SHBG <34.1 nmol/L)Quartile 2(34.1 ≤ SHBG <46.7 nmol/L)Quartile 3(46.7 ≤ SHBG <64.5 nmol/L)Quartile 4(SHBG≥64.5 nmol/L)Total
Women (n = 3,031)      
 Number at risk 759 759 757 756 3,031 
 Follow-up time (py) 4,943 5,120 5,078 4,977 20,119 
 New cases of T2D 76 24 19 10 129 
 Incidence rate per 1,000 py 15.4 4.7 3.7 2.0 6.4 
 Crude model 7.57 (3.91, 14.63) 2.32 (1.11, 4.85) 1.86 (0.86, 3.99)  
 Model 2 7.44 (3.76, 14.71) 2.32 (1.09, 4.96) 1.94 (0.89, 4.25)  
 Model 3 4.89 (2.41, 9.90) 1.74 (0.80, 3.79) 1.63 (0.74, 3.62)  
Women with liver fat measurements available (n = 887)      
 Number at risk 212 226 213 236 887 
 Follow-up time (py) 1,369 1,519 1,422 1,499 5,809 
 New cases of T2D 19 30 
 Incidence rate per 1,000 py 13.9 3.9 2.1 1.3 5.1 
 Crude model 10.11 (2.35, 43.44) 2.88 (0.58, 14.29) 1.55 (0.26, 9.27)  
 Model 2 8.92 (1.96, 40.54) 2.99 (0.57, 15.77) 1.65 (0.26, 10.30)  
 Model 3 (total association) 5.41 (1.12, 26.09) 2.09 (0.37, 11.66) 1.18 (0.18, 7.66)  
 Model 3 + liver fat 4.20 (0.82, 21.53) 2.06 (0.36, 11.96) 1.04 (0.16, 7.03)  
 Mediation by liver fat (%) 27 — —   
 Model 3 + VAT 3.86 (0.75, 19.78) 1.82 (0.32, 10.53) 1.07 (0.16, 7.29)  
 Mediation by VAT (%) 35 — —   
 Model 3 + VAT + liver fat (direct) 3.51 (0.66, 18.65) 1.92 (0.33, 11.31) 0.96 (0.14, 6.80)  
 Proportion of mediation via VAT and liver fat (%) 43     
Quartile 1(SHBG <34.1 nmol/L)Quartile 2(34.1 ≤ SHBG <46.7 nmol/L)Quartile 3(46.7 ≤ SHBG <64.5 nmol/L)Quartile 4(SHBG≥64.5 nmol/L)Total
Women (n = 3,031)      
 Number at risk 759 759 757 756 3,031 
 Follow-up time (py) 4,943 5,120 5,078 4,977 20,119 
 New cases of T2D 76 24 19 10 129 
 Incidence rate per 1,000 py 15.4 4.7 3.7 2.0 6.4 
 Crude model 7.57 (3.91, 14.63) 2.32 (1.11, 4.85) 1.86 (0.86, 3.99)  
 Model 2 7.44 (3.76, 14.71) 2.32 (1.09, 4.96) 1.94 (0.89, 4.25)  
 Model 3 4.89 (2.41, 9.90) 1.74 (0.80, 3.79) 1.63 (0.74, 3.62)  
Women with liver fat measurements available (n = 887)      
 Number at risk 212 226 213 236 887 
 Follow-up time (py) 1,369 1,519 1,422 1,499 5,809 
 New cases of T2D 19 30 
 Incidence rate per 1,000 py 13.9 3.9 2.1 1.3 5.1 
 Crude model 10.11 (2.35, 43.44) 2.88 (0.58, 14.29) 1.55 (0.26, 9.27)  
 Model 2 8.92 (1.96, 40.54) 2.99 (0.57, 15.77) 1.65 (0.26, 10.30)  
 Model 3 (total association) 5.41 (1.12, 26.09) 2.09 (0.37, 11.66) 1.18 (0.18, 7.66)  
 Model 3 + liver fat 4.20 (0.82, 21.53) 2.06 (0.36, 11.96) 1.04 (0.16, 7.03)  
 Mediation by liver fat (%) 27 — —   
 Model 3 + VAT 3.86 (0.75, 19.78) 1.82 (0.32, 10.53) 1.07 (0.16, 7.29)  
 Mediation by VAT (%) 35 — —   
 Model 3 + VAT + liver fat (direct) 3.51 (0.66, 18.65) 1.92 (0.33, 11.31) 0.96 (0.14, 6.80)  
 Proportion of mediation via VAT and liver fat (%) 43     
Quartile 1(SHBG<26.4 nmol/L)Quartile 2(26.4 ≤ SHBG <33.6 nmol/L)Quartile 3(33.6 ≤ SHBG <42.8 nmol/L)Quartile 4(SHBG>42.8 nmol/L)Total
Men (n = 2,659)      
 Number at risk 668 665 663 663 2,659 
 Follow-up time (py) 4,298 4,354 4,397 4,389 17,438 
 New cases of T2D 48 31 46 29 154 
 Incidence rate per 1,000 py 11.1 7.1 10.5 6.6 8.8 
 Crude model 1.70 (1.07, 2.69) 1.08 (0.65, 1.79) 1.59 (1.00, 2.53)  
 Model 2 2.07 (1.28, 3.35) 1.21 (0.73, 2.03) 1.76 (1.10, 2.80)  
 Model 3 1.78 (1.10, 2.88) 1.06 (0.63, 1.77) 1.54 (0.96, 2.47)  
Men with liver fat measurements available (n = 935)      
 Number at risk 241 240 213 241 935 
 Follow-up time (py) 1,557 1,544 1,409 1,586 6,095 
 New cases of T2D 17 10 13 10 50 
 Incidence rate per 1,000 py 10.9 6.5 9.2 6.3 8.2 
 Crude model 1.74 (0.80, 3.81) 1.02 (0.43, 2.46) 1.46 (0.64, 3.34)  
 Model 2 2.11 (0.93, 4.77) 1.21 (0.49, 2.96) 1.60 (0.70, 3.67)  
 Model 3 (total association) 2.01 (0.89, 4.54) 1.18 (0.48, 2.88) 1.68 (0.73, 3.88)  
 Model 3 + liver fat 1.62 (0.71, 3.66) 1.07 (0.44, 2.64) 1.46 (0.63, 3.38)  
 Mediation by liver fat (%) 39 — —   
 Model 3 + VAT 1.69 (0.74, 3.86) 1.05 (0.43, 2.60) 1.49 (0.65, 3.44)  
 Mediation by VAT (%) 32 — —   
 Model 3 + liver fat + VAT (direct) 1.40 (0.61, 3.22) 0.98 (0.40, 2.43) 1.31 (0.56, 3.06)  
 Proportion of mediation via VAT and liver fat (%) 60     
Quartile 1(SHBG<26.4 nmol/L)Quartile 2(26.4 ≤ SHBG <33.6 nmol/L)Quartile 3(33.6 ≤ SHBG <42.8 nmol/L)Quartile 4(SHBG>42.8 nmol/L)Total
Men (n = 2,659)      
 Number at risk 668 665 663 663 2,659 
 Follow-up time (py) 4,298 4,354 4,397 4,389 17,438 
 New cases of T2D 48 31 46 29 154 
 Incidence rate per 1,000 py 11.1 7.1 10.5 6.6 8.8 
 Crude model 1.70 (1.07, 2.69) 1.08 (0.65, 1.79) 1.59 (1.00, 2.53)  
 Model 2 2.07 (1.28, 3.35) 1.21 (0.73, 2.03) 1.76 (1.10, 2.80)  
 Model 3 1.78 (1.10, 2.88) 1.06 (0.63, 1.77) 1.54 (0.96, 2.47)  
Men with liver fat measurements available (n = 935)      
 Number at risk 241 240 213 241 935 
 Follow-up time (py) 1,557 1,544 1,409 1,586 6,095 
 New cases of T2D 17 10 13 10 50 
 Incidence rate per 1,000 py 10.9 6.5 9.2 6.3 8.2 
 Crude model 1.74 (0.80, 3.81) 1.02 (0.43, 2.46) 1.46 (0.64, 3.34)  
 Model 2 2.11 (0.93, 4.77) 1.21 (0.49, 2.96) 1.60 (0.70, 3.67)  
 Model 3 (total association) 2.01 (0.89, 4.54) 1.18 (0.48, 2.88) 1.68 (0.73, 3.88)  
 Model 3 + liver fat 1.62 (0.71, 3.66) 1.07 (0.44, 2.64) 1.46 (0.63, 3.38)  
 Mediation by liver fat (%) 39 — —   
 Model 3 + VAT 1.69 (0.74, 3.86) 1.05 (0.43, 2.60) 1.49 (0.65, 3.44)  
 Mediation by VAT (%) 32 — —   
 Model 3 + liver fat + VAT (direct) 1.40 (0.61, 3.22) 0.98 (0.40, 2.43) 1.31 (0.56, 3.06)  
 Proportion of mediation via VAT and liver fat (%) 60     

Data are HR (95% CI) based on Cox regression analysis unless otherwise indicated. Model 2 is adjusted for age (years), smoking (current, former, never), alcohol consumption (g/day), physical activity during leisure time in (MET-h/week), menopausal state (pre, post), use of hormone therapy (currently, no current use), ethnicity (White, other), DHDI (score 0–150), and education level (high, other). Model 3: model 2 additionally adjusted for total body fat (%), liver fat content (%), and VAT (cm2).

Mediation Analysis

Whereas in the total population the association between SHBG concentrations and the risk of T2D was present when restricting the population to those with VAT and liver fat measurements (n = 1,822) the sample size was too small to detect these associations as statistically significantly (Table 2). Both VAT and liver fat content were associated with risk of T2D. Per SD (60.2 cm2) of VAT the HR of T2D was 2.37 (95% CI 1.59, 3.53) in women and 1.44 (1.07, 1.95) in men. Per percentage of liver fat, the HR of T2D was 1.05 (95% CI 1.02, 1.07) in both women and men, after adjustment for confounding. There was evidence for an interaction between SHBG and VAT (P = 0.01 in women and in men) and between SHBG and liver fat content (P = 0.05 in women, P = 0.00 in men); therefore, we only present percentages of mediation within the lowest quartile of SHBG, which was significantly related to T2D in the total populations of women and men.

First, liver fat was added to the models in the population of participants with liver fat measurements. In women, the HR for the lowest SHBG quartile compared with the highest SHBG quartile changed from 5.41 (95% CI 1.12, 26.09) to 4.20 (0.82, 21.53), reflecting a proportion of mediation of 27%. In men, the HR changed from 2.01 (0.89, 4.54) to 1.62 (0.71, 3.66), demonstrating a proportion of mediation of 39%. The proportions of mediation by VAT was 35% in women and 32% in men. When adding both VAT and liver fat to the model, the proportions of mediation by the combination of VAT and liver fat were 43% in women and 60% in men.

After excluding women using hormone therapy or participants using SHBG-influencing medication, the results remained comparable. Postmenopausal women in the lowest quartile of SHBG had 36 (IQR 25–47) cm2more VAT and a 3.3-fold (95% CI 2.7, 4.0) higher liver fat content, and the HR to develop T2D was 7.27 (95% CI 2.23, 23.73) compared with postmenopausal women in the highest quartile. The median SHBG concentration of postmenopausal women was lower than in premenopausal women (44.5 [IQR 33.4–58.7] vs. 52.7 [36.5–75.2] nmol/L).

Bidirectional Mendelian Randomization Analyses of SHBG and Liver Fat Content

All estimated F statistics were >10, including the conditional F statistics from the multivariable Mendelian randomization analyses. Causal estimates were consistent with higher SHBG concentrations, leading to a lower liver fat percentage in both women (−0.33 [95% CI −0.51, −0.16] SD in liver fat per SD in SHBG) and men (−0.18 [−0.31, −0.05] per ln SHBG). Results from sensitivity analyses were directionally consistent (Table 3). For SHBG in both women and men, CAUSE was unable to distinguish a model of causality from horizontal pleiotropy. However, restricting the genetic instruments to three genetic variants (rs1625895, rs12150660, and rs1641537) located in the SHBG gene was in line with higher SHBG concentrations, leading to lower liver fat percentage (FE-IVW estimates: women, SD −0.16 [95% CI −0.30, −0.02); men, −0.15 [−0.27, −0.02]).

Table 3

Results from bidirectional Mendelian randomization analyses between SHBG and liver fat content

Women, SD (95% CI)Men, ln (95% CI)
Causal effect of SHBG on liver fat percentage: estimates from 2SMR   
Main analyses   
 IVW −0.33 (−0.51, −0.16) −0.18 (−0.31, −0.05) 
 MR-PRESSO −0.45 (−0.55, −0.35) −0.25 (−0.34, −0.16) 
 Contamination mixture −0.55 (−0.70, −0.43) −0.10 (−0.23, 0.00) 
 Mixture of experts −0.33 (−0.51, −0.16) −0.08 (−0.47, 0.31) 
Additional sensitivity analyses   
SHBG SNPs −0.16 (−0.30, −0.02) −0.15 (−0.27, −0.02) 
 MVMR: BMI adjusted −0.36 (−0.54, −0.18) −0.12 (−0.34, 0.10) 
 MVMR: bioT adjusted −0.27 (−0.51, −0.04) −0.14 (−0.28, 0.00) 
 CAUSE −0.32 (−0.70, −0.19), P = 0.20* −0.31 (−0.43, −0.16), P = 0.09* 
Causal effect of liver fat percentage on SHBG: estimates from 2SMR Women, SD (95% CI) Men, SD (95% CI) 
Main analyses   
 IVW 0.00 (−0.06, 0.06) 0.06 (−0.01, 0.13) 
 MR-PRESSO 0.00 (−0.04, 0.05) 0.10 (0.07, 0.12) 
 Contamination mixture −0.02 (NA) 0.13 (0.12, 0.14) 
 Mixture of experts 0.00 (−0.03, 0.03) 0.09 (0.06, 0.12) 
Additional sensitivity analyses   
 DNL susceptibility SNPs 0.06 (−0.12, 0.24) 0.08 (−0.05, 0.20) 
Women, SD (95% CI)Men, ln (95% CI)
Causal effect of SHBG on liver fat percentage: estimates from 2SMR   
Main analyses   
 IVW −0.33 (−0.51, −0.16) −0.18 (−0.31, −0.05) 
 MR-PRESSO −0.45 (−0.55, −0.35) −0.25 (−0.34, −0.16) 
 Contamination mixture −0.55 (−0.70, −0.43) −0.10 (−0.23, 0.00) 
 Mixture of experts −0.33 (−0.51, −0.16) −0.08 (−0.47, 0.31) 
Additional sensitivity analyses   
SHBG SNPs −0.16 (−0.30, −0.02) −0.15 (−0.27, −0.02) 
 MVMR: BMI adjusted −0.36 (−0.54, −0.18) −0.12 (−0.34, 0.10) 
 MVMR: bioT adjusted −0.27 (−0.51, −0.04) −0.14 (−0.28, 0.00) 
 CAUSE −0.32 (−0.70, −0.19), P = 0.20* −0.31 (−0.43, −0.16), P = 0.09* 
Causal effect of liver fat percentage on SHBG: estimates from 2SMR Women, SD (95% CI) Men, SD (95% CI) 
Main analyses   
 IVW 0.00 (−0.06, 0.06) 0.06 (−0.01, 0.13) 
 MR-PRESSO 0.00 (−0.04, 0.05) 0.10 (0.07, 0.12) 
 Contamination mixture −0.02 (NA) 0.13 (0.12, 0.14) 
 Mixture of experts 0.00 (−0.03, 0.03) 0.09 (0.06, 0.12) 
Additional sensitivity analyses   
 DNL susceptibility SNPs 0.06 (−0.12, 0.24) 0.08 (−0.05, 0.20) 

Sensitivity analyses represent FE-IVW estimators. 2SMR, two-sample Mendelian randomization; bioT, bioavailable testosterone; DNL, de novo lipogenesis; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier; MVMR, multivariable Mendelian randomization; NA, not applicable; SNP, single nucleotide polymorphism.

*

P value for CAUSE denotes whether a causal model describes the data better than a noncausal model. The units for the underlying GWAS: liver fat, SD; SHBG women, SD; SHBG men, ln.

We did not find evidence that higher liver fat content decreases SHBG (estimates from two-sample Mendelian randomization: women, SD 0.00 [95% CI −0.06, 0.06]; men, 0.06 [0.01, 0.13]). The sensitivity analysis restricted to variants suggestive for downstream effects of de novo lipogenesis resulted in imprecise FE-IVW estimates (women, SD 0.06 [95% CI −0.12, 0.24]; men, 0.08 [−0.05, 0.20]).

In this study, we investigated the relationships among SHBG, VAT, liver fat content, and the development of T2D. In our population, with an oversampling of participants meeting criteria for overweight, we noted an inverse association between SHBG and VAT and between SHBG and liver fat content, and these associations were stronger in women than in men. The results of our Mendelian randomization analyses supported our hypothesized direction that low SHBG concentrations may lead to liver fat accumulation. Furthermore, we observed that the lowest quartile of SHBG in women was associated with a 4.9-fold increased risk of T2D compared with the highest quartile, and this association was mediated by liver fat content for 27%. In men, the lowest quartile of SHBG was associated with a 1.8-fold increased risk of T2D compared with the highest quartile, and 39% of this association was mediated by liver fat content. The combination of VAT and liver fat content mediated 43% (women) and 60% (men) of the SHBG-T2D association.

Our finding regarding the association between low SHBG and an increased risk of T2D is supported by previous literature. Although previous studies described that hyperinsulinemia itself may lead to low SHBG concentrations by suppressing hepatic SHBG production (36), Mendelian randomization studies have indicated that low SHBG itself is involved in the development of T2D (4–6). Possible underlying mechanisms include effects of SHBG on lipolysis and alteration of glucose homeostasis (8,37). In addition, we hypothesized that liver fat accumulation is a potential underlying mechanism.

Previous literature on the direction of the association between low SHBG concentrations and liver fat remained inconsistent (38). A large population-based study recently reported that per percent liver fat content, SHBG was 0.008 nmol/L lower in men and 0.017 nmol/L lower in women (39). The authors argued that this is in line with experimental studies suggesting that hepatic de novo lipogenesis could downregulate SHBG (18,40). However, results from our Mendelian randomization analyses are in favor of the opposite direction of the association, namely from low concentrations of SHBG leading to higher liver fat percentage. This direction is supported by previous studies in mice showing that overexpression of SHBG led to a reduction of liver fat (16,37). Experiments in HepG2 cells (human hepatocyte study model) revealed a possible underlying mechanism: Treatment with SHBG activated the extracellular signal–regulated kinase 1/2-mitogen-activated protein kinase pathway, which reduced peroxisome proliferator–activated receptor-γ mRNA and protein levels, consequently leading to a downregulation of key lipogenic enzymes, such as acetyl CoA carboxylase (15). Another possible mechanism was described in an in vitro study on macrophages and adipocytes that were exposed to physiologic concentrations of SHBG. Inflammation and lipid accumulation were suppressed in both cell types (1), suggesting that low SHBG concentrations could result in lipid storage in the liver.

Although women had higher SHBG concentrations, less VAT, lower liver fat percentage, and lower incidence of T2D than men, it must be noted that the associations between SHBG and VAT, SHBG and liver fat content, and SHBG and risk of T2D were more pronounced in women than in men. This is consistent with previous literature (7). Associations were even stronger after restricting analyses to postmenopausal women. The biological mechanisms accounting for this sex difference remain unclear but could be related to sex-specific changes in bioavailability of sex steroid concentrations (8,41). A high free and total testosterone concentration is associated with an increased risk of T2D in women (42,43). It could be argued that high SHBG concentrations in women are protective against exposure to higher free testosterone concentrations, while low SHBG concentrations lead to higher free testosterone concentrations (8) and indirectly may increase the risk of T2D. In men, low testosterone, high estradiol, and low SHBG are risk factors for T2D (2,44,45). When SHBG levels decrease, a temporarily increased free testosterone concentration leads to decreased production of testosterone, and consequently, free testosterone concentration will remain normal (8). This feedback mechanism is only present in men and ensures that a change in SHBG concentration does not influence free testosterone concentrations. If the SHBG-T2D association is partially explained by free testosterone concentrations, this might explain the weaker associations in men. The fact that the association is observed in men despite this feedback mechanism advocates for direct effects of SHBG on T2D, independent of free testosterone concentrations. This is also supported by a previous study in hepatic steatosis mouse models where even in castrated mice, overexpression of SHBG reduced liver fat accumulation by reducing key lipogenic enzymes (15).

Strengths of our study include the prospective design with a follow-up of 6.7 years, the assessment of liver fat content by MRS, the availability of many confounding factors, and the combination of observational with bidirectional Mendelian randomization analyses to examine the direction of the SHBG-liver fat content association. Our study also has several limitations. Despite the size of the total study population, the subset with liver fat measurements was too small (n = 1,822) to detect the associations between SHBG and T2D as statistically significant. Nevertheless, given that the associations that we observed in the total population were significant and are supported by the literature, we are confident that the estimated percentages of mediation of the SHBG-T2D association by liver fat content are robust.

Regarding our Mendelian randomization analysis, a potential limitation lies in using sex-combined summary statistics for liver fat content. However, previous studies found no evidence of sex specificity for common risk loci for fatty liver (46), which largely overlap with those found for MRS-derived liver fat. In addition, we could only use 12 genetic instruments for liver fat, which impeded bidirectional Mendelian randomization analyses aimed at mediation by liver fat between SHBG and T2D. As the UK Biobank imaging substudy continues to work toward its goal of scanning 100,000 individuals, a larger number of instruments from future GWAS will further increase power and disentangle potential heterogeneity of causal effects. This may include causal effects between women and men or metabolically distinct paths contributing to excess liver fat and more advanced stages of metabolic dysfunction–associated steatotic liver disease (47). Assuming that all individuals from the liver fat percentage GWAS contributed to the SHBG GWAS, a potential participant overlap of up to ∼9% for both women and men is possible. However, given the strength of our instruments, marked weak instrument bias in the context of sample overlap is unlikely for our study. For SHBG in relation to liver fat, the results from our CAUSE analyses suggest that we cannot rule out that correlated horizontal pleiotropy underlies our findings. However, when restricting the genetic instruments to those specific to the SHBG gene, we found directionally consistent, though attenuated, causal estimates.

Finally, it is important to note that 95% of the population in the NEO study identified as White, and there was an oversampling of participants with a BMI >27 kg/m2. Therefore, our findings apply to a White European population that is overweight, and our results need to be confirmed in other populations.

Our results may be relevant for people receiving therapies that influence SHBG levels, such as gender-affirming hormone therapy. A recent study showed that in transgender women receiving gonadotropin-releasing hormone analogs (suppressing testosterone production) and estradiol agents, liver fat content was decreased after 18 months of treatment (48). Due to the estradiol supplementation, SHBG concentrations rise, and concentrations >100 nmol/L in users of feminizing hormone treatment are not rare. Transgender men receiving testosterone treatment usually show a decrease of SHBG. In the same study, an increase in not only liver fat content but also in VAT after the start of masculinizing hormone treatment was reported, supporting our results.

Altogether, our results in combination with those of others support a causal role of low SHBG in both VAT and liver fat accumulation. Furthermore, the association between low SHBG and risk of T2D was, for a large part, mediated by the amount of VAT and liver fat. Therefore, lifestyle interventions are warranted for reducing VAT, liver fat, and the risk of T2D. Our findings further suggest that interventions targeted at increasing SHBG may contribute to reducing VAT, liver fat, and, as such, the risk of T2D. Future research should focus on the underlying mechanisms of sex differences in the associations among SHBG, VAT, liver fat, and T2D for the development of methods to safely modulate SHBG concentrations in individuals and for subsequent changes in VAT, liver fat, and T2D.

Acknowledgments. The authors thank all the individuals who participated in the NEO study and all participating general practitioners for inviting eligible participants. The authors also thank, from the Department of Clinical Epidemiology, Leiden University Medical Center, P.R. van Beelen and all research nurses for collection of the data, P.J. Noordijk and team for sample handling and storage, and I. de Jonge for data management of the NEO study, and, from the Department of Experimental Psychology, University College London, London, U.K., M.A. Cottam for assistance with proofreading the manuscript.

Funding. The NEO study was supported by the participating divisions and departments and board of directors of the Leiden University Medical Center and by the Leiden University Research Profile Area Vascular and Regenerative Medicine.

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

Author Contributions. T.A.S. contributed to the data analysis and interpretation and writing of the manuscript. C.M.W. provided supervision and reviewed and approved the final version of the manuscript. R.A.J.S. executed the Mendelian randomization analysis and contributed to the writing and approval of the final version of the manuscript. A.v.H.V. and M.C.G.J.B. reviewed and approved the final version of the manuscript. H.J.L. contributed to the study design and data collection and approved the final version of the manuscript. J.H.P.M.v.d.V. and S.C.B. contributed to the data collection and reviewed and approved final version of the manuscript. E.W.-v.E. contributed to the study design and data collection and reviewed and approved the final version of the manuscript. F.R.R. contributed to the study design and data collection and approved the final version of the manuscript. M.d.H. contributed to the study conceptualization and approved the final version of the manuscript. A.C.H. contributed to the study conceptualization and analysis of the blood samples and approved the final version of the manuscript. R.d.M. contributed to the study conceptualization and design and data collection, provided supervision, and reviewed and approved the final version of the manuscript. R.d.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in poster form at the Dutch Endocrine Meeting, Noordwijkerhout, the Netherlands, 2–3 February 2023, and in oral form at the Annual Dutch Diabetes Research Meeting, Wageningen, the Netherlands, 3–4 November 2022.

1.
Yamazaki
H
,
Kushiyama
A
,
Sakoda
H
, et al
.
Protective Effect of Sex Hormone-Binding Globulin against Metabolic Syndrome: In Vitro Evidence Showing Anti-Inflammatory and Lipolytic Effects on Adipocytes and Macrophages
.
Mediators Inflamm
2018
;
2018
:
3062319
2018/06/25 2018
2.
Le
TN
,
Nestler
JE
,
Strauss
JF
,
Wickham
EP
.
Sex hormone-binding globulin and type 2 diabetes mellitus
.
Trends Endocrinol Metab
2012
;
23
:
32
40
3.
Chen
BH
,
Brennan
K
,
Goto
A
, et al
.
Sex hormone-binding globulin and risk of clinical diabetes in American black, Hispanic, and Asian/Pacific Islander postmenopausal women
.
Clin Chem
2012
;
58
:
1457
1466
4.
Ding
EL
,
Song
Y
,
Manson
JE
, et al
.
Sex hormone–binding globulin and risk of type 2 diabetes in women and men
.
N Engl J Med
2009
;
361
:
1152
1163
5.
Ruth
KS
,
Day
FR
,
Tyrrell
J
, et al.;
Endometrial Cancer Association Consortium
.
Using human genetics to understand the disease impacts of testosterone in men and women
.
Nat Med
2020
;
26
:
252
258
6.
Perry
JRB
,
Weedon
MN
,
Langenberg
C
, et al.;
MAGIC
.
Genetic evidence that raised sex hormone binding globulin (SHBG) levels reduce the risk of type 2 diabetes
.
Hum Mol Genet
2010
;
19
:
535
544
7.
Jaruvongvanich
V
,
Sanguankeo
A
,
Riangwiwat
T
,
Upala
S
.
Testosterone, Sex Hormone-Binding Globulin and Nonalcoholic Fatty Liver Disease: a Systematic Review and Meta-Analysis
.
Ann Hepatol
2017
;
16
:
382
394
8.
Simons
PIHG
,
Valkenburg
O
,
Stehouwer
CDA
,
Brouwers
MCGJ
.
Sex hormone-binding globulin: biomarker and hepatokine?
Trends Endocrinol Metab
2021
;
32
:
544
553
9.
Martin
S
,
Sorokin
EP
,
Thomas
EL
, et al
.
Estimating the Effect of Liver and Pancreas Volume and Fat Content on Risk of Diabetes: A Mendelian Randomization Study
.
Diabetes Care
2022
;
45
:
460
468
10.
Nielsen
TL
,
Hagen
C
,
Wraae
K
, et al
.
Visceral and Subcutaneous Adipose Tissue Assessed by Magnetic Resonance Imaging in Relation to Circulating Androgens, Sex Hormone-Binding Globulin, and Luteinizing Hormone in Young Men
.
J Clin Endocrinol Metab
2007
;
92
:
2696
2705
11.
Neeland
IJ
,
Ross
R
,
Després
J-P
, et al.;
International Chair on Cardiometabolic Risk Working Group on Visceral Obesity
.
Visceral and ectopic fat, atherosclerosis, and cardiometabolic disease: a position statement
.
Lancet Diabetes Endocrinol
2019
;
7
:
715
725
12.
Després
J-P
.
Is visceral obesity the cause of the metabolic syndrome?
Ann Med
2006
;
38
:
52
63
13.
Cao
Y-t
,
Xiang
L-l
,
Nie
L-j
, et al
.
The sex-specific causal role of visceral adipose tissue in non-alcoholic fatty liver disease susceptibility: results from Mendelian Randomization and mediation analysis considering sex hormones
. 9 November
2023
[preprint].
SSRN
:4625736
14.
Wang
X
,
Xie
J
,
Pang
J
, et al
.
Serum SHBG is associated with the development and regression of nonalcoholic fatty liver disease: a prospective study
.
J Clin Endocrinol Metab
2020
;
105
:
e791
e804
15.
Saez-Lopez
C
,
Barbosa-Desongles
A
,
Hernandez
C
, et al
.
Sex hormone-binding globulin reduction in metabolic disorders may play a role in NAFLD development
.
Endocrinology
2017
;
158
:
545
559
16.
Sáez-López
C
,
Salcedo-Allende
MT
,
Hernandez
C
,
Simó-Servat
O
,
Simó
R
,
Selva
DM
.
Sex hormone-binding globulin expression correlates with acetyl-coenzyme A carboxylase and triglyceride content in human liver
.
J Clin Endocrinol Metab
2019
;
104
:
1500
1507
17.
Simons
PIHG
,
Valkenburg
O
,
Stehouwer
CDA
,
Brouwers
MCGJ
.
Association between de novo lipogenesis susceptibility genes and coronary artery disease
.
Nutr Metab Cardiovasc Dis
2022
;
32
:
2883
2889
18.
Selva
DM
,
Hogeveen
KN
,
Innis
SM
,
Hammond
GL
.
Monosaccharide-induced lipogenesis regulates the human hepatic sex hormone–binding globulin gene
.
J Clin Invest
2007
;
117
:
3979
3987
19.
de Mutsert
R
,
den Heijer
M
,
Rabelink
TJ
, et al
.
The Netherlands Epidemiology of Obesity (NEO) study: study design and data collection
.
Eur J Epidemiol
2013
;
28
:
513
523
20.
Siebelink
E
,
Geelen
A
,
de Vries
JHM
.
Self-reported energy intake by FFQ compared with actual energy intake to maintain body weight in 516 adults
.
Br J Nutr
2011
;
106
:
274
281
21.
Looman
M
,
Feskens
EJ
,
de Rijk
M
, et al
.
Development and evaluation of the Dutch Healthy Diet Index 2015
.
Public Health Nutr
2017
;
20
:
2289
2299
22.
Naressi
A
,
Couturier
C
,
Devos
JM
, et al
.
Java-based graphical user interface for the MRUI quantitation package
.
MAGMA
2001
;
12
:
141
152
23.
Szczepaniak
LS
,
Nurenberg
P
,
Leonard
D
, et al
.
Magnetic resonance spectroscopy to measure hepatic triglyceride content: prevalence of hepatic steatosis in the general population
.
Am J Physiol Endocrinol Metab
2005
;
288
:
E462
E468
24.
de Boer
AW
,
Blom
JW
,
de Waal
MWM
, et al
.
Coded diagnoses from general practice electronic health records are a feasible and valid alternative to self-report to define diabetes cases in research
.
Prim Care Diabetes
2021
;
15
:
234
239
25.
Baron
RM
,
Kenny
DA
.
The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations
.
J Pers Soc Psychol
1986
;
51
:
1173
1182
26.
Narinx
N
,
David
K
,
Walravens
J
, et al
.
Role of sex hormone-binding globulin in the free hormone hypothesis and the relevance of free testosterone in androgen physiology
.
Cell Mol Life Sci
2022
;
79
:
1
30
27.
Liu
Y
,
Basty
N
,
Whitcher
B
, et al
.
Genetic architecture of 11 organ traits derived from abdominal MRI using deep learning
.
Elife
2021
;
10
:
e65554
28.
Verbanck
M
,
Chen
C-Y
,
Neale
B
,
Do
R
.
Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases
.
Nat Genet
May
2018
;
50
:
693
698
29.
Burgess
S
,
Foley
CN
,
Allara
E
,
Staley
JR
,
Howson
JMM
.
A robust and efficient method for Mendelian randomization with hundreds of genetic variants
.
Nat Commun
2020
;
11
:
376
30.
Hemani
G
,
Tilling
K
,
Davey Smith
G
.
Orienting the causal relationship between imprecisely measured traits using GWAS summary data
.
PLoS Genet
2017
;
13
:
e1007081
31.
Hemani
G
,
Bowden
J
,
Haycock
P
, et al
.
Automating Mendelian randomization through machine learning to construct a putative causal map of the human phenome
. 23 August
2017
[preprint].
bioRxiv
:
173682
32.
Locke
AE
,
Kahali
B
,
Berndt
SI
, et al.;
International Endogene Consortium
.
Genetic studies of body mass index yield new insights for obesity biology
.
Nature
2015
;
518
:
197
206
33.
Morrison
J
,
Knoblauch
N
,
Marcus
JH
,
Stephens
M
,
He
X
.
Mendelian randomization accounting for correlated and uncorrelated pleiotropic effects using genome-wide summary statistics
.
Nat Genet
2020
;
52
:
740
747
34.
Coviello
AD
,
Haring
R
,
Wellons
M
, et al
.
A genome-wide association meta-analysis of circulating sex hormone-binding globulin reveals multiple Loci implicated in sex steroid hormone regulation
.
PLoS Genet
2012
;
8
:
e1002805
35.
Graham
SE
,
Clarke
SL
,
Wu
K-HH
, et al.;
Global Lipids Genetics Consortium
.
The power of genetic diversity in genome-wide association studies of lipids
.
Nature
2021
;
600
:
675
679
36.
Pasquali
R
,
Casimirri
F
,
De Iasio
R
, et al
.
Insulin regulates testosterone and sex hormone-binding globulin concentrations in adult normal weight and obese men
.
J Clin Endocrinol Metab
1995
;
80
:
654
658
37.
Saez-Lopez
C
,
Villena
JA
,
Simó
R
,
Selva
DM
.
Sex hormone-binding globulin overexpression protects against high-fat diet-induced obesity in transgenic male mice
.
J Nutr Biochem
2020
;
85
:
108480
38.
Grossmann
M
,
Wierman
ME
,
Angus
P
,
Handelsman
DJ
.
Reproductive endocrinology of nonalcoholic fatty liver disease
.
Endocr Rev
2019
;
40
:
417
446
39.
Simons
PIHG
,
Valkenburg
O
,
van de Waarenburg
MPH
, et al
.
Serum sex hormone-binding globulin is a mediator of the association between intrahepatic lipid content and type 2 diabetes: the Maastricht Study
.
Diabetologia
2022
;
66
:
213
222
40.
Simons
PIHG
,
Valkenburg
O
,
Telgenkamp
I
, et al
.
Relationship between de novo lipogenesis and serum sex hormone binding globulin in humans
.
Clin Endocrinol (Oxf)
2021
;
95
:
101
106
41.
Mody
A
,
White
D
,
Kanwal
F
,
Garcia
JM
.
Relevance of low testosterone to non-alcoholic fatty liver disease
.
Cardiovasc Endocrinol
2015
;
4
:
83
89
42.
Goto
A
,
Morita
A
,
Goto
M
, et al.;
Saku Cohort Study Group
.
Associations of sex hormone-binding globulin and testosterone with diabetes among men and women (the Saku Diabetes study): a case control study
.
Cardiovasc Diabetol
2012
;
11
:
130
139
43.
Rasmussen
JJ
,
Selmer
C
,
Frøssing
S
, et al
.
Endogenous testosterone levels are associated with risk of type 2 diabetes in women without established comorbidity
.
J Endocr Soc
2020
;
4
:
bvaa050
44.
Hu
J
,
Zhang
A
,
Yang
S
, et al
.
Combined effects of sex hormone-binding globulin and sex hormones on risk of incident type 2 diabetes
.
J Diabetes
2016
;
8
:
508
515
45.
Huang
R
,
Wang
Y
,
Yan
R
,
Ding
B
,
Ma
J
.
Sex hormone binding globulin is an independent predictor for insulin resistance in male patients with newly diagnosed type 2 diabetes mellitus
.
Diabetes Ther
2023
;
14
:
1627
1637
46.
Lefebvre
P
,
Staels
B
.
Hepatic sexual dimorphism - implications for non-alcoholic fatty liver disease
.
Nat Rev Endocrinol
2021
;
17
:
662
670
47.
Sliz
E
,
Sebert
S
,
Würtz
P
, et al
.
NAFLD risk alleles in PNPLA3, TM6SF2, GCKR and LYPLAL1 show divergent metabolic effects
.
Hum Mol Genet
2018
;
27
:
2214
2223
48.
Tebbens
M
,
Schutte
M
,
Troelstra
MA
, et al
.
Sex steroids regulate liver fat content and body fat distribution in both men and women: a study in transgender persons
.
J Clin Endocrinol Metab
2023
;
109
:
e280
e290
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.