Discussion
In summary, in the DPP, lifestyle intervention exerted effects to increase or attenuate the decline in SHBG levels, manifested differently within each subgroup. In postmenopausal women, for example, ILS resulted in an increase in SHBG, consistent with prior analyses within the DPP,19 whereas in men and in premenopausal women in the ILS arm, the decrement in SHBG was less than that seen in the other treatment arms. Metformin had no impact compared with placebo on SHBG in any of the groups. Through multivariable regression models, we found that a large part of the effect of ILS on SHBG levels was attributable to reductions in adiposity, with modest effects of sex hormones, glucose measures, and their changes. Given that low SHBG levels have been reported to be associated with diabetes risk1–7 and that low SHBG levels are seen in conditions associated with obesity and insulin resistance (eg, diabetes, polycystic ovary syndrome, and fatty liver disease),9–12 25–28 we explored whether the changes in SHBG contributed to reduced diabetes incidence. However, despite these apparently beneficial changes in SHBG, we did not see consistent evidence for SHBG mediating the treatment effects on diabetes prevention. Furthermore, while changes in SHBG were positively correlated with changes in circulating hormones and 1/fasting insulin, and inversely associated with changes in glucose (fasting and 2-hour postchallenge) and HOMA-IR, mediation analyses did not support a centrally mediated effect of SHBG on these outcomes. Taken together, these observations suggest that changes in SHBG reflect the overall metabolic and hormonal milieu, in particular the effects of weight loss, but are not directly influencing risk of development of diabetes.
Our findings are also consistent with our earlier analyses of baseline SHBG and SHBG SNPs associated with diabetes outcomes and risk of diabetes in the DPP. As earlier reported, while baseline SHBG was cross-sectionally associated with some indicators of insulin resistance and diabetes risk (inverse fasting insulin, insulinogenic index, and waist circumference), SHBG concentration at baseline was not associated with diabetes risk in any of the participant groups evaluated. Furthermore, there was no evident association of the SHBG SNPs and diabetes risk in the DPP population.20
To date, it has not been clear whether low SHBG levels represent a biomarker of metabolic abnormality and diabetes risk or are perhaps somehow contributory and causative of disease. Our data would support that dynamic changes in SHBG reflect the changes in the surrounding metabolic environment. Several studies support this line of thought. First, multiple studies have demonstrated changes in SHBG in response to weight loss, independent of mechanism of weight loss (eg, diet type, exercise, and bariatric surgery).16 29–32 In the Sex Hormones and Physical Exercise (SHAPE-2) Study trial, for example, in which overweight inactive women were randomized to diet, exercise or control groups, with a goal of 5–6 kg weight loss over 16 weeks, both the diet and exercise arms achieved weight loss and had significant increases in SHBG, with improvements in sex hormones (increase in free estradiol and decrease in free testosterone) compared with control, yet these effects were attenuated after adjustment for changes in body fat.32 In another study, intentional weight loss followed by intentional weight gain in premenopausal women demonstrated both an initial increase in SHBG followed by the reciprocal decrease, with free androgen index and visceral adipose tissue changing in opposite direction to SHBG changes, again speaking to a dynamic change in SHBG in response to decreases in weight and adiposity.33 With bariatric surgery, resulting in significant and sustained amounts of weight loss,30 31 sustained increases in circulating SHBG have been seen even several years after surgery.31
That we did not see an impact of metformin on circulating SHBG is also of interest. This may be because metformin primarily suppresses hepatic glucose production and is not generally regarded as a potent insulin sensitizer. In contrast, intervention studies with insulin sensitizers (thiazolidinediones), have, like for weight loss described previously, demonstrated responsive increases in SHBG.9 26 34 35 In a prospective randomized controlled trial of women with polycystic ovary syndrome, for example, serum SHBG levels were correlated with glucose disposal rate (insulin sensitivity), as assessed by hyperinsulinemic euglycemic clamp, and increased significantly during treatment with pioglitazone, a treatment known to directly modulate insulin sensitivity. Furthermore, the improvement in glucose disposal rate (ie, insulin sensitivity) was directly associated with treatment-associated increases in serum SHBG,9 thus suggesting the effect of insulin sensitization on increasing SHBG. In another study evaluating rosiglitazone (also an insulin sensitizer) on metabolic and ovarian effects in polycystic ovary syndrome (PCOS), an increase in SHBG was demonstrated in response to rosiglitazone treatment for 12 weeks, with higher levels of SHBG seen in those who ovulated on rosiglitazone. Lower circulating insulin levels were also shown, highlighting once again the dynamic changes in SHBG in response to improving insulin sensitivity.34
Consistent with our findings that SHBG may reflect the metabolic milieu and changes as such, it is of interest to note that the hepatic environment has also been directly implicated in regulating SHBG. Selva et al36 previously showed that monosaccharide (glucose and fructose) induced hepatic lipogenesis reduced SHBG production by downregulating hepatocyte nuclear factor 4α (HNF-4α) levels, a key transcription factor regulating hepatic expression of SHBG. Supporting this, Winters et al11 evaluated SHBG gene expression in human liver samples and found that SHBG mRNA was a strong predictor of circulating SHBG levels. They described an inverse association between hepatic triglyceride content and SHBG mRNA and serum SHBG, with a suggestion that the low level of SHBG mRNA was largely due to a low level of HNF-4α mRNA expression in the liver, which is also reduced in insulin resistance. Thus, it is plausible that improvements in insulin sensitivity, as seen with weight loss and therapies that directly modify insulin sensitivity, may have an impact on the liver, which then affects expression of SHBG mRNA and production of SHBG.
In contrast, there is emerging evidence that SHBG may play a causative role in disease and may not merely be an ‘innocent bystander’. Sáez-López et al37 recently described a significant inverse relation between SHBG mRNA expression and hepatic triglyceride content, as well as levels of acetyl-coenzyme A carboxylase, a key lipogenic enzyme, in liver samples obtained from obese humans with non-alcoholic fatty liver disease undergoing bariatric surgery. Furthermore, the authors found that SHBG overexpression in cultures of human hepatic (HepG2) cells was able to abrogate the increase in multiple hepatic lipogenic enzymes in the liver when triggered under high-glucose culture conditions. Although this was studied in fatty liver, it is possible that SHBG is not only a biomarker, but it may independently contribute to the pathogenesis or even protection from metabolic disease.
In order to estimate a causal role of circulating SHBG for type 2 diabetes, Wang et al38 applied quantitative nuclear magnetic resonance metabolomics in three Finnish population-based cohorts to profile circulating lipids and metabolites and their association with SHBG. Higher SHBG levels were associated with a more favorable cardiometabolic risk profile, and SHBG was predictive of future insulin resistance and type 2 diabetes. The observed association of SHBG with type 2 diabetes (OR=0.83 per 1 SD) was less than that seen in prospective observational associations by meta-analysis (HR 0.47). These results suggest that circulating SHBG may have a minor direct contributory role in the development of type 2 diabetes but is more likely largely reflective of other factors.
There are several strengths and limitations to these analyses. First, a prospective, controlled evaluation of changes in SHBG by sex and menopausal status in the very well characterized DPP population provided the opportunity to directly assess whether the study interventions affected SHBG and what other factors contributed to those changes. In addition, our sample size is commensurate with other studies in the literature1 2 7 8 for the relevant constituent grouping to have provided meaningful analyses and comparative results. However, given the predefined criteria of the DPP to identify those individuals already at high risk of development of diabetes based on glucose and weight measures, we may have been limited in seeing additional impact of SHBG on risk. Furthermore, SHBG and sex hormone measurements were conducted on samples that were not timed to the endogenous hormonal cycle, and hormone use was assessed by self-report. Randomization may help to minimize potential influence of this and other unanticipated factors that may regulate SHBG.
In summary, in the DPP, ILS was consistently associated with changes in circulating SHBG levels compared with placebo or metformin, specifically increased levels in postmenopausal women, and attenuation of decrease in men and premenopausal women. The effects of ILS on SHBG seemed largely due to changes in adiposity but may also be influenced by other changes (eg, glucose measures and sex steroids). The observed changes in SHBG related to the interventions did not consistently translate to changes in diabetes risk.