Examining the associations between testosterone and biomarkers as men age

Testosterone concentrations in men decline with advancing age. However, the cause of the decline is yet to be fully elucidated. Therefore, the aims of this study were to examine the associations between chronic diseases such as obesity and type 2 diabetes mellitus (T2DM) with total testosterone (TT) and sex hormone‐binding globulin (SHBG), using a large nationally‐representative data set (National Health and Nutrition Examination Survey; NHANES).


| INTRODUCTION
Testosterone concentrations in men decline with advancing age (Andersson et al., 2007;Araujo et al., 2007;Harman et al., 2001;Liu et al., 2007;Rohrmann et al., 2011).Approximately 38% of men over the age of 45 years have low testosterone concentrations (Araujo et al., 2007;Mulligan et al., 2006;Travison, Araujo, O'Donnell, et al., 2007).Low testosterone concentrations in men are also associated with obesity (Lopez et al., 2018) and comorbidities including type two diabetes mellitus (T2DM) (Grossmann et al., 2010) and metabolic syndrome (Li et al., 2010).Moreover, the relationship between obesity and testosterone concentrations in men is bi-directional, although obesity appears to have a greater influence on testosterone compared to the effect of testosterone on obesity (Grossmann, 2018).For example, findings from the Massachusetts Male Aging Study (MMAS) observed that men who remained or became obese during the 8-year follow up had the lowest concentrations of total testosterone (TT), free testosterone (FT), and sex hormone-binding globulin (SHBG) (Derby et al., 2006).Conversely, 12 months of androgen deprivation therapy (ADT) had minimal effect on body mass index (BMI) (Cheung et al., 2016).Moreover, testosterone treatment only had moderate effects on body composition, and it did not result in significant improvements in insulin resistance (IR), which is strongly associated with BMI (Huang et al., 2018).Consequently, an Endocrine Society statement examining the pathogenesis of obesity did not identify androgen deficiency (AD) as a causative factor (Schwartz et al., 2017).
Despite the general observation that obesity is associated with low testosterone concentrations in men, it is yet to be determined if the decline in testosterone results from obesity per se or one of the comorbidities associated with obesity (Fui et al., 2014;Wang et al., 2011).In particular, the association between BMI and TT concentrations after adjustment for age and obesity-related morbidities such as T2DM, are not known.Therefore, the primary aim of this study was to examine the associations between BMI, measurements associated with T2DM [an oral glucose tolerance test (OGTT), homeostatic model assessment of insulin resistance (HOMA-IR), insulin, and glucose], and age with TT, using a large nationally-representative data set (National Health and Nutrition Examination Survey; NHANES) to determine if the significance of the inverse relationship between BMI and TT remains after adjusting for other variables.A secondary aim of this study was to examine the association of these variables with SHBG, which is understood to regulate the plasma levels and biological actions of testosterone (Hammond et al., 2012).We hypothesized (i) obesity would have an inverse association with TT concentrations; (ii) the measurements associated with T2DM (OGTT, HOMA-IR, insulin, and glucose) would be inversely associated with TT concentrations; (iii) obesity and the measurements associated with T2DM (OGTT, HOMA-IR, insulin, and glucose) would be inversely associated with SHBG; (iv) SHBG would be positively associated with age.

| MATERIALS AND METHODS
NHANES is a cross-sectional survey, physical examination, and laboratory evaluation of a nationallyrepresentative sample of a non-institutionalized United States population.It uses a multi-stage, stratified, probability cluster design.The selected surveys oversampled Hispanics; non-Hispanic blacks; non-Hispanic, non-black Asians; low-income white and other persons; and adults aged 80 and over.Male participants aged ≥18 years during the NHANES 2013-2014 and NHANES 2015-2016 survey periods were selected for this analysis.Due to the small set of observations of males with an Underweight categorization according to BMI (approximately <5% of observations), these observations were not selected for analysis.Fasting sample weights, reflecting the weighting applied to the subsample of participants that provided glucose and insulin data, were employed when estimating descriptive and inferential outcomes in the forthcoming Results.Consent was received from all participants and was approved by the Research Ethics Review Board (U.S.Department of Health and Human Services, n.d.).

| Anthropometric assessment
BMI was calculated as weight in kilograms divided by height in meters squared.Scores were rounded to one decimal place.Body measurements for both survey periods were collected in the Mobile Examination Center (MEC) by trained health technicians.A recorder assisted the health technician during the body measures examination.Weight was measured using a Mettler Toledo scale and height was measured using an Inductive Series II stadiometer.

| Sex hormone assessment
Methods, equipment used, and measurements of sex hormones for both survey periods were identical.Serum concentrations of TT were measured in ng/dL and using isotope dilution, high-performance liquid chromatography tandem mass spectrometry as described in the CDC Laboratory Procedure Manual for sex hormones (U.S.Department of Health and Human Services, 2019).This method has demonstrated a high level of accuracy for many years.The coefficient of variation (CV) for TT during the 2015-16 collection was: 1.9%-2.2%.The CV for TT during the 2013-14 collection was unavailable.Serum and plasma SHBG concentrations were measured in nmol/L and determined by the reaction between SHBG and immuno-antibodies and chemo-luminescence measurements of the reaction products.The sample mixture was then exposed to a magnetic field while in a measuring cell as described in the CDC Laboratory Procedure Manual for SHBG (U.S.Department of Health and Human Services, 2019).Samples from 5745 men were collected according to standardized protocols and certified by CDC Hormone Standardization Program (HoSt) (Zhou et al., 2017).The CV for SHBG during the 2013-14 and 2015-16 collections were: 2.0%-4.1%.

| Oral glucose tolerance test
A fasting glucose blood test was performed via venepuncture after fasting for 9 h.Participants were then requested to consume a calibrated dose (75 grams of glucose) of Trutol™.A second venipuncture was conducted 2 h (±15 min) after consumption of Trutol™.Plasma specimens were then processed, stored, and shipped to the University of Missouri (Columbia, MO) where they were analyzed by the Diabetes Diagnostic Laboratory.The 2-h OGTT was measured in mmol/L.

| Homeostatic model assessment of insulin resistance
A calculation to determine the homeostatic model assessment of insulin resistance (HOMA-IR) score was conducted based on the results of the glucose and insulin measurements of survey participants.The calculation was as follows: fasting insulin (pmol/L) Â fasting glucose (nmol/L) / 22.5.

| Glucose and insulin
Fasting glucose and insulin serum and plasma samples were measured in nmol/L and pmol/L, respectively, and collected during morning sessions.Specimens were analyzed by the Diabetes Diagnostic Laboratory at the University of Missouri (Columbia, MO).Fasting glucose was enzymatically measured using the Cobas C311-2017 Analyzer and the Cobas C 501 Analyzer (Roche Diagnostics International AG, Rotkreuz, Switzerland).The CV for glucose during the 2015-16 collection was: 0.8%-1.8%.The CV for glucose during the 2013-14 collection was unavailable.Fasting insulin was measured in pmol/L using a two-site immunoenzymometric assay on Tosoh AIA System analyzer (Tosoh Bioscience, Inc.South San Francisco, CA).The CV for insulin during the 2013-14 and 2015-16 collections were: 2.3%-6.1%.

| Statistical analysis
All statistical analyses were performed using the R software statistical package (R Core Team, 2021) Version 4.1.0.Population survey-weighted with stratification linear regression analyses were performed using the survey package (Lumley, 2004), model diagnostic analyses employed the svydiags package (Valliant, 2018), and interaction plots were probed using the interactions package (Long, 2019).When predicting TT and SHBG; BMI, age, OGTT, HOMA-IR, glucose, and insulin were included as model predictors.BMI was modeled as three discrete categories: Normal (BMI 18.5-24.99),Overweight (BMI 25-29.99),and Obese (BMI >30).Underweight BMI individuals were excluded from these analyses due to the small count of observations for males older than 18 years of age.Normal BMI was employed as the reference level during modeling, and interaction terms between the BMI categories and age, glucose, and insulin were included during modeling.In adherence to regression model assumptions, observations that reflected notable outliers (i.e., standardized residuals in excess of ±3 following initial model testing) were identified.A residual analysis was conducted.The residual plot confirmed that both normality and homogenous variance assumptions were met.These observations were excluded from model testing by sub-setting them out of the analysis cohort, reflecting a loss of observations less than 5% for each outcome modeled (TT: 1.30%; SHBG: 0.61%).HOMA-IR was modeled as three discrete categories based on research conducted by Horakova et al., (2019): Non-diabetics (<1.82),Pre-diabetics (1.82-3.63),and Diabetics (>3.63).Insulin and glucose concentrations were log-transformed to address the skewness of the data when analyzed with the dependent variables, TT and SHBG.This analysis was applied cautiously to ensure the skewness was not exacerbated and did improve normalization of the data.

| Total testosterone
All variables measured in this study were significantly inversely associated with total testosterone (TT).The variables measured included the BMI categories of overweight and obese; OGTT, HOMA-IR, insulin, glucose, and age.TT concentrations of adult men progressively decreased with advancing age.Obese men (BMI >30) had lower TT concentrations compared to men with a normal BMI (BMI 20-24.99),with overweight men and obese men having TT concentrations approximately 2.7 ng/ dL and 4.8 ng/dL lower than men with a normal BMI, respectively.TT concentrations of adult men decreased as glucose, insulin, and OGTT increased.Adult men with pre-diabetic or diabetic HOMA-IR levels had lower TT concentrations compared to men with nondiabetic HOMA-IR levels.All variables remained significantly inversely associated with TT, except HOMA-IR and glucose, after adjusting for the other variables.Although HOMA-IR and glucose were inversely associated with TT in Model 1 (unadjusted estimates), glucose decreased more rapidly at the lower part of the glucose range before tapering into a flatter relationship.None of the interaction terms with BMI were statistically significant.See Figure 1 for the relationships between age, insulin, BMI, OGTT, HOMA-IR, and TT.

| Sex hormone binding globulin
Age, BMI, a pre-diabetic and diabetic HOMA-IR level, and insulin were significant predictors of sex hormone binding globulin (SHBG) (see Table 2).Age demonstrated a significant positive association with SHBG values, differing from TT, which had a significant negative association with age.Insulin presented a similar negative loglinear relationship with SHBG as the prior TT model.Individuals in the overweight and obese BMI categories had lower SHBG concentrations compared to the normal BMI category individuals, with overweight men and obese men having SHBG concentrations approximately 0.2 and 0.4 nmol/L lower than men with a normal BMI, respectively.Men with pre-diabetic and diabetic HOMA-IR levels had significantly lower SHBG concentrations compared to men with non-diabetic levels.No significant associations between glucose concentrations or OGTT and SHBG were identified in Model 1 (unadjusted estimates) but after adjusting for other predictors, OGTT was significantly inversely associated with SHBG.All variables significantly associated with SHBG from Model 1, remained significantly inversely associated with SHBG, except men with diabetic HOMA-IR levels, after adjusting for the other variables.Although the interactions tested suggested a potential significant interaction between obese BMI and insulin (see Table 2), the change in R 2 between the models was negligible, therefore the interaction effect was of marginal importance.See Figure 2 for the relevant regression plots.

| DISCUSSION
Obesity demonstrates an inverse association with TT and SHBG; however, it is yet to be determined if the relationship results from obesity per se or one of the comorbidities associated with obesity.Therefore, the purpose of this study was to examine the associations between age, BMI, OGTT, HOMA-IR, insulin, and glucose concentrations with TT and SHBG.The main findings of this study were (i) obesity was inversely associated with TT and SHBG concentrations, even after adjusting for other variables; (ii) the measurements associated with T2DM (OGTT, HOMA-IR, insulin, and glucose) were inversely associated with TT concentrations, and all variables remained significant, except glucose concentrations and HOMA-IR levels, after adjusting for other variables; (iii) insulin was significantly inversely associated with SHBG, even after adjusting for other variables; (iv) HOMA-IR was significantly inversely associated with SHBG; however after adjusting for other variables only the pre-diabetic level remained significant; (v) Glucose concentrations and OGTT were not significantly associated with SHBG concentrations; however, after adjusting for other variables OGTT became significant; (vi) and age was positively associated with SHBG concentrations and this relationship strengthened after adjustment for the other variables.

| Overweight/obese BMI categories
The present study found significant inverse associations between both overweight and obese BMI categories with TT and SHBG concentrations, even after adjusting for the other variables.This means that the relationship between overweight and obese BMI categories with TT and SHBG are independent of other factors, including comorbidities such as T2DM.These results are supported by several studies (Dhindsa et al., 2010;Hall et al., 2008;Tajar et al., 2010), as well as a longitudinal examination of participants in the Massachusetts Male Aging Study (MMAS) (Derby et al., 2006) and the findings of the European Male Aging Study (EMAS) (Wu et al., 2008).
While the exact mechanisms linking obesity with reduced TT and SHBG concentrations are not well understood, several biological mechanisms have been proposed.These include hyperinsulinemia, hyperlipidemia, hyperleptinemia, and chronic inflammation (Gapstur et al., 2007;Hosick et al., 2020;Lainez & Coss, 2019;Osuna et al., 2006).Moreover, testosterone and obesity demonstrate a bi-directional relationship (Kelly & Jones, 2015).Adipocytes reduce testosterone concentrations via several mechanisms, including increased F I G U R E 1 Relationships between total testosterone and the predictors age (A), insulin (B), BMI categories (C), OGTT (D), and HOMA-IR categories (E).Model plots are drawn from the results of the model with no interaction terms (Model 1; see Table 1).
aromatase activity, the enzyme responsible for converting testosterone into estradiol, which in turn, has a negative feedback mechanism on the hypothalamuspituitary-gonadal (HPG) axis; increased concentrations of inflammatory cytokines (adipokines), tumor necrosis factor-alpha (TNF-α) and interleukin-6 (IL-6), which have a suppressive effect on the HPG axis; leptin resistance, which has a negative effect on the HPG axis; and reduced concentrations of adiponectin (adipokine) (Kelly & Jones, 2015).Conversely, elevated testosterone concentrations may reduce obesity via stimulating lipolysis, decreasing lipogenesis, impairing lipid uptake and adipocyte differentiation, and promoting greater β-oxidation (Kelly & Jones, 2015).Despite the clear inverse relationship between obesity and both TT and SHBG concentrations in men, future research is required to elucidate the mechanisms contributing to these relationships.

| Oral glucose tolerance test
The present study found a significant inverse association between the 2-hour OGTT results and TT.This result is supported by several studies (Leutner et al., 2022;Wittert et al., 2021).An OGTT predicts insulin sensitivity with reasonable accuracy (Stumvoll et al., 2001) and may assist with the prediction of T2DM development (Hayashi et al., 2013), one of the comorbidities associated with obesity (Golay & Ybarra, 2005).Moreover, since IR and T2DM are associated with low testosterone concentrations in men (Grossmann et al., 2010; F I G U R E 2 Relationships between sex hormone binding globulin and age (A), insulin (B), BMI categories (C), OGTT (D), and HOMA-IR categories.Model plots are drawn from the results of the model with no interaction terms (Model 1; see Table 2).Ottarsdottir et al., 2018;Wittert & Grossmann, 2022), OGTT results may serve as a surrogate marker for IR and the development of T2DM.It also appears that the relationship between testosterone and IR is bidirectional, since androgen deprivation therapy is associated with increased IR (Basaria, 2008) and the use of testosterone replacement therapy (TRT) improves insulin sensitivity by lowering fasting insulin concentrations (Bhasin, 2003;Dhindsa et al., 2016).Moreover, in the absence of IR, insulin stimulates receptors in the hypothalamus, which stimulate GnRH release into the local network of blood vessels (Sliwowska et al., 2014).The pituitary gland also has receptors for insulin, with in vitro research indicating that it may directly stimulate luteinsing hormone production (Adashi et al., 1981), but is potentially less sensitive than other body tissues (Wu et al., 2012).The Leydig cells of the testes also contain insulin receptors, which has demonstrated a positive effect on testosterone production in vitro (Lin et al., 1986) and IR having a negative effect on testosterone synthesis in the Leydig cells (Pitteloud et al., 2005).
Therefore, insulin appears to be positively associated with testosterone concentrations in men while IR is negatively associated with testosterone concentrations in men.No significant association was found between OGTT and SHBG, except after adjustment for other variables.This result is supported by other research indicating that low SHBG concentrations are associated with hepatic IR (Winters et al., 2014;Ye et al., 2017), while high SHBG concentrations are associated with insulin sensitivity (Peter et al., 2010).

| Homeostatic model assessment for insulin resistance
In the present study, another analysis was conducted, homeostatic model assessment for insulin resistance (HOMA-IR), to determine the relationship between IR and both TT and SHBG concentrations in men.HOMA-IR is a robust tool for assessing IR in people (Antuna-Puente et al., 2011;Lann & LeRoith, 2007).The present study found an inverse association between pre-diabetic and diabetic HOMA-IR categories and both TT and SHBG concentrations in men.This result is supported by other studies with similar findings (Kurniawan et al., 2020;Quan et al., 2021;Reddy & Yadav, 2021;Ye et al., 2017).However, the relationships became nonsignificant, except for the relationship between prediabetic HOMA-IR levels and SHBG, after adjustment for other variables.This result supports the OGTT result, whereby IR is inversely associated with TT and SHBG.

| Insulin
The present study found a significant inverse association between insulin and both TT and SHBG, even after adjusting for other variables.This finding is consistent with several cross-sectional studies (Kurniawan et al., 2020;Leutner et al., 2022;Osuna et al., 2006;Pasquali et al., 1995;Phillips et al., 2003) and an 8-year longitudinal study (Oh et al., 2002), which found testosterone concentrations to be inversely associated with insulin concentrations over the duration of the study.Even though insulin is not a clinical condition, elevated insulin concentrations, hyperinsulinemia, may be indicative of IR and the future development of metabolic syndrome (Wilcox, 2005) (Ling et al., 2016) (Sung et al., 2011).Since SHBG-bound testosterone is a major component of TT, approximately 40% (Faix, 2013;Li et al., 2010), a reduction in its synthesis is likely to result in a significant reduction in TT; SHBG and TT are strongly positively correlated (de Ronde et al., 2005;Winters, 2020).Therefore, the inverse relationship between insulin and both testosterone and SHBG identified in the present study is supported by both in vitro and human studies, demonstrating insulin's potentially inhibitory effect on hepatic SHBG synthesis (Kalme et al., 2003;Plymate et al., 1988;Yki-Jarvinen et al., 1995).However, since low SHBG concentrations are also associated with hepatic IR (Winters et al., 2014;Ye et al., 2017), elevated insulin concentrations may be indicative of IR.IR also suppresses insulin's ability to inhibit lipolysis in adipose tissue, resulting in greater mobilization of free fatty acids (FFA) from adipocytes and greater circulating levels of FFA (Kelly & Jones, 2013).This leads to greater deposition of lipids in hepatic and muscle tissue (Yu & Ginsberg, 2005), resulting in exacerbation of IR in the liver and further suppression of SHBG synthesis; a positive feedback loop (Figure 3).

| Glucose
The present study identified a significant inverse association between glucose concentrations and TT; however, after adjustment for other variables, the association was no longer present, meaning other factors were contributing to that relationship.No significant association was found between glucose concentrations and SHBG concentrations, even after adjusting for other variables.This result contradicts our initial hypothesis that glucose would be inversely associated with SHBG.Research indicates that elevated glucose concentrations may reduce the hepatic synthesis of SHBG via the downregulation of hepatocyte nuclear factor-4alpha (Selva et al., 2007).Like insulin, glucose is not a clinical condition; however, elevated fasting glucose concentrations, impaired glucose tolerance, may be indicative of IR, potentially leading to the development of T2DM (Nichols et al., 2007).Therefore, further research is required to elucidate this finding.

| Age
Age was also found to have a significant inverse association with TT in the present study.This result is supported by both human and animal studies (Chen et al., 2015;Sokanovic et al., 2014;Surampudi et al., 2012;Travison, Araujo, Kupelian, et al., 2007;Veldhuis et al., 2012).Several mechanisms have been proposed to explain this association including HPG axis dysfunction, reduced Leydig cell numbers, reduced sensitivity of Leydig cells to luteinizing hormone, and multiple intracellular molecular dysfunctions within the Leydig cells (Wang et al., 2017).
Conversely, the present study found a significant positive association of age with SHBG concentrations.This finding is supported by several papers including a review paper by Kaufman et al., (2019) examining the relationship between aging and the male reproductive system, a longitudinal examination of participants in the MMAS (Derby et al., 2006), and the findings of the crosssectional study of men in the EMAS (Wu et al., 2008).Moreover, the mechanisms promoting increasing concentrations of SHBG with advancing age is yet to be elucidated, however, a plausible theory exists (Figure 4).Since growth hormone (GH) induces hepatic IR (Forrest et al., 2019;Mercado & Ramirez-Renteria, 2018), resulting in reduced SHBG synthesis, a reduction in GH concentrations due to aging (Chahal & Drake, 2007) may improve hepatic insulin sensitivity and therefore, increase SHBG synthesis (Thaler et al., 2015).Hence, the gradual increase in SHBG concentrations in men as they age.GH also increases peripheral mobilization of free fatty acids, potentially increasing hepatic fatty acid availability (Krag et al., 2007), which exacerbates hepatic IR.Therefore, the age-related reduction in GH potentially decreases hepatic fatty acid availability and further improves hepatic insulin sensitivity, resulting in greater SHBG synthesis and increased SHBG concentrations.

| Limitations
The main limitation of the present study is its crosssectional study design, which means we are unable to determine causality, especially since TT has a bidirectional relationship with obesity and the measurements associated with T2DM (OGTT, HOMA-IR, insulin, and glucose).Moreover, BMI was used as the measure of obesity, whereas waist circumference (Svartberg et al., 2004) or a more direct measure of adiposity, such as dual energy x-ray absorptiometry (DEXA) (Bilsborough et al., 2014), would have provided more detailed and accurate information.Even though BMI does have some limitations, since it does not accurately measure body composition, it does strongly correlate with adiposity and percentage body fat within groups (Flegal et al., 2009;Panuganti et al., 2020).A major limitation of the present study was not identifying the health status of the cohort being examined.However, this was a cross-sectional analysis of a large dataset where multiple health-markers were included.It was simply not feasible nor practical to assess across all known mental (e.g., anxiety, stress, depression, etc.) and physical (e.g., musculoskeletal, endocrine, metabolic, neurological, cardiovascular, etc.) health markers.Indeed, the prevalence of undiagnosed conditions within the cohort likely approaches the numbers of diagnosed individuals.Another potential limitation is that the time of the blood draws (morning, afternoon, or evening) and fasting status were not standardized across participants.According to Vesper et al., (2015), TT concentrations from the NHANES 2011-2012 study population varied between fasted and non-fasted groups in the afternoon ( p = .0183)and evening ( p = .0001)collections, with fasting participants having significantly higher TT compared to the non-fasting participants.However, TT concentrations tend to be consistent in men over the age of 45 irrespective of the blood draw time (Guay et al., 2008).According to our results, the overall model fit (R-squared) indicates that 26% and 34% of the factors investigated in the present study were able to predict the variance in TT and SHBG concentrations, respectively.Therefore, 74% and 66% of the factors contributing to the TT and SHBG concentrations, respectively, remained unexplained and required further investigation.

| Future directions
The current study is simply an observational study, meaning causality cannot be demonstrated irrespective of the sample size.Therefore, interventional studies should be conducted that identify causality.For example, obesity is hypothesized to reduce TT and SHBG concentrations via several proposed mechanisms, including inflammation.Therefore, an interventional study could be conducted using an intradermal lipopolysaccharide (LPS) challenge, with a similar study design to Buters et al., (2022), which would induce inflammation.TT and SHBG concentrations could be tested at various timepoints after administration, and perhaps compared with a placebo intradermal challenge.
Identifying relationships between other hormones and markers in future studies may also be useful.Measuring estradiol concentrations in men and comparing them to testosterone concentrations and/or a body composition variable, BMI or DEXA, may potentially identify levels of aromatase activity.

| CONCLUSION
Analysis of the NHANES 2013-2014 and NHANES 2015-2016 data are consistent with previous research, indicating that an overweight or obese condition have significant inverse associations with TT and SHBG concentrations.Moreover, the present study determined that even after adjusting for other variables, the significance of the inverse relationships remained.Several variables associated with T2DM (OGTT, HOMA-IR, insulin, and glucose) and their relationship with TT and SHBG were also examined.The findings of the present study were consistent with the findings of previous research, with OGTT, HOMA-IR, insulin, and glucose being inversely associated with TT.The present study also found that the association between OGTT and insulin with TT remained significant after adjusting for the other variables.However, the association between glucose and HOMA-IR with TT became non-significant after adjusting for the other variables.The T2DM variables significantly inversely associated with SHBG were insulin and HOMA-IR; however, only the significance of the association between SHBG and pre-diabetic HOMA-IR levels remained after adjusting for the other variables.Interestingly, OGTT became significantly associated with SHBG after adjusting for the other variables.Age was significantly inversely associated with TT, but positively associated with SHBG, even after adjusting for other variables.Contrary to our initial hypotheses, (i) blood glucose was not significantly associated with TT after adjusting for other variables or with SHBG concentrations.The present study represents the largest study to date examining the relationship between insulin, age, obesity, and glucose concentrations with TT and SHBG concentrations.

AUTHORS CONTRIBUTIONS
I can confirm that all authors have met all the Criteria for Authorship credit.I, Stephen Smith, contributed to the study design, the acquisition of the data, the interpretation of the data, and the drafting of the article.Daniel Bekele analyzed the data and drafted the article.Adrian Lopresti interpreted the data and reviewed and revised the article.Timothy Fairchild conceptualized the study design, interpreted the data, and reviewed and revised the article.

F
I G U R E 3 Proposed mechanism for insulin's effect on total testosterone concentrations.F I G U R E 4 Proposed mechanism for increased SHBG concentrations in older men.
T A B L E 1Note: R 2 = 0.26.Robust standard errors are presented.The reference category for BMI is Normal.The reference category for HOMA_IR is non-diabetics.Total testosterone converted to square root.Glucose, OGTT and insulin were converted using a natural log function when specifying the model, in order to approximate a normal distribution of data for these predictors.* p < .05;** p < .01;*** p < .001.