Abstract
Context
Many effects of testosterone are mediated through dihydrotestosterone (DHT) and estradiol.
Objective
To determine the relative contributions of each hormone to the observed effects of testosterone treatment in older men with hypogonadism.
Methods
Using data from the Testosterone Trials, we assessed the association of changes in total testosterone, estradiol, and DHT levels over 12 months of testosterone treatment with hemoglobin, high-density lipoprotein (HDL) cholesterol, volumetric bone mineral density (vBMD) of lumbar spine, sexual desire, and prostate-specific antigen (PSA). We used random forests to model the associations of predicted mean changes in outcomes with change in each hormone at low, mean, or high change in the other 2 hormones. Stepwise regression models were run to confirm the findings of random forests.
Result
Predicted increases in hemoglobin and sexual desire were greater with larger increases in estradiol and were larger with high change in DHT compared with low change in DHT. Greater increases in estradiol were associated with larger decreases in HDL cholesterol; this association did not vary according to changes in DHT or testosterone. Change in vBMD was most robustly associated with change in estradiol and was greater with high change in testosterone and DHT. There was no consistent relation between change in PSA and change in any hormone.
Conclusion
Change in estradiol level was the best predictor not only of the change in vBMD and sexual desire but also of the changes in hemoglobin and HDL cholesterol. Consideration of testosterone, estradiol, and DHT together offers a superior prediction of treatment response in older hypogonadal men than testosterone alone.
Keywords: relation of sex hormones with outcomes, the testosterone trials, testosterone and bone mineral density, dihydrotestosterone and bone mineral density, estradiol and hemoglobin, dihydrotestosterone and sexual desire, estradiol and HDL cholesterol
Circulating testosterone is converted in many peripheral tissues to its 2 active metabolites, 5α dihydrotestosterone (DHT) and 17β estradiol (1). In many androgen-responsive tissues, a family of steroid 5α reductase enzymes converts testosterone to DHT, and the aromatase enzyme, a product of the CYP19A1 gene, converts it to estradiol (1). Many tissue-specific biologic effects of testosterone are mediated through DHT and estradiol. The data from preclinical gene-targeting experiments, observational and Mendelian randomization studies, and randomized clinical trials have provided strong evidence of the important role of testosterone’s aromatization to estradiol in mediating its effects on several reproductive and nonreproductive behaviors, fat mass and metabolism, bone mineral density, negative feedback on gonadotropin secretion through the kisspeptin/neurokinin B/dynorphin neurons (KNDy) neuronal network, and plasma lipids (2-5). Mendelian randomization studies have confirmed that genetically determined estradiol levels are more strongly associated with bone mineral density and fracture risk than genetically determined testosterone levels (3). While the effects of testosterone and estradiol on bone may be complementary, some other effects, such as those on erythropoiesis, have been viewed as antagonistic—testosterone is known to stimulate erythropoiesis (6), while estradiol has been reported to inhibit erythropoiesis in some experimental models (7). In contrast, circulating testosterone levels are generally believed to be more robustly associated than estradiol levels with its anabolic effects on skeletal muscle mass, maximal voluntary strength, and some types of male behaviors (2, 6, 8-10). DHT is required for masculinization of the urogenital sinus and the genital tubercle in fetal life and possibly for mediating its effects on the prostate and hair follicle (10) but is not obligatory for mediating testosterone’s effects on the skeletal muscle, bone, or erythropoiesis (11). DHT is also more potent than testosterone in its nongenomic effects on vascular smooth muscle (12).
The rates of conversion of testosterone to DHT and estradiol vary among people due to polymorphisms of genes that encode the steroid 5α reductases and the aromatase enzyme as well as other host-specific factors that affect the activity of these enzymes (3, 13). In hypogonadal men treated with transdermal testosterone gels, the circulating levels of DHT and the ratio of serum DHT to testosterone levels are substantially higher than in hypogonadal men treated with the injectable testosterone esters (14), presumably due to the high activity of steroid 5α reductase enzyme in the skin. Although DHT is recognized as a potent androgen, circulating DHT levels have been ignored in assessing the efficacy of transdermal testosterone, and the circulating DHT levels are not monitored. The mean serum DHT levels in the participants of the Testosterone Trials (TTrials), who were assigned to the testosterone arm of the trial, were 4 to 5 times the mean DHT levels in healthy young men; in some testosterone-treated men, the DHT levels approached the circulating testosterone levels (15). In spite of this recognition, testosterone treatment of men with hypogonadism in clinical practice is guided almost entirely by the monitoring of circulating on-treatment testosterone levels; DHT and estradiol levels are rarely considered in evaluating therapeutic response to testosterone treatment or in dose adjustment (16). It is not known how the circulating concentrations of testosterone’s metabolites—DHT and estradiol—modulate the effects of testosterone on various outcomes and how their circulating levels rank in their contribution to the observed effects of testosterone treatment on physiologic outcomes.
We performed secondary analyses of data from participants in the treatment arm of the TTrials to assess the association of changes in total testosterone, estradiol, and DHT levels over the 12 months of testosterone treatment with 5 quantitative continuous endpoints of the trial—the changes over 12 months in hemoglobin, high-density lipoprotein (HDL) cholesterol levels, volumetric bone mineral density (vBMD) of the trabecular bone of the lumbar spine, sexual desire, and prostate-specific antigen (PSA). We ranked the 3 hormones for their relative contribution in predicting change in these 5 biomarker outcomes of the trial.
Materials and Methods
The TTrials’ protocol and consent forms were approved by the institutional review boards at the University of Pennsylvania and each participating trial site. All participants provided written informed consent. A data and safety monitoring board monitored the study’s progress and safety data every 6 months.
Analytic Sample
The TTrials protocol and efficacy results have been published (15, 17, 18). Briefly, the TTrials were a multisite, coordinated set of 7 trials to evaluate the effects of testosterone treatment of older men with hypogonadism on various functional outcomes and markers known to decline with age, including sexual function, physical function, vitality, hemoglobin, bone mineral density, and various measures of cardiovascular health. To be eligible for the trial, men had to be 65 years or older, have an average of 2 fasting, early-morning serum testosterone levels less than 275 ng/dL and 1 or more of sexual dysfunction, mobility limitation, or low vitality. The men were randomly allocated to receive testosterone or placebo gel for 1 year, with quarterly visits for assessment of study outcomes. Testosterone levels were measured, and the testosterone dose was adjusted to maintain on-treatment testosterone levels in the range that is mid-normal for healthy young men. This study includes all men in the TTrials who were allocated to the testosterone arm. Of 394 men allocated to the testosterone arm of the TTrials, 111 also participated in the Bone Trial, which was conducted at 9 of the 12 sites for the TTrials.
Outcome Measures
The biomarker outcomes evaluated in the present analyses were change from baseline to 12 months in hemoglobin (g/dL), HDL cholesterol (mg/dL), vBMD of the trabecular bone of the lumbar spine, measured using quantitative computed tomography (CT), performed at baseline and month 12 (19), PSA, and sexual desire, determined using the DeRogatis Interview for Sexual Function-II. Changes from baseline in hemoglobin, HDL cholesterol, PSA, sexual desire, and hormones were measured at 3, 6, 9, and 12 months. Participants provided blood samples at scheduled clinic visits which were stored at –80°C and transferred to central laboratories for assessment of hormones and biomarkers.
The hormones evaluated as predictors of change in outcomes included serum total testosterone, estradiol, and DHT levels, measured using liquid chromatography tandem mass spectrometry (LC-MS/MS) in the Brigham Research Assay Core Laboratory, Boston, MA. As described previously (15), the lower limit of quantitation of the LC-MS/MS assay for testosterone was 1 ng/dL and linear range 1 to 1000 ng/dL (15). The interassay coefficient of variation (CV) was 7.9% at 48.6 ng/dL, 7.7% at 241 ng/dL, 4.4% at 532 ng/dL, and 3.3% at 1016 ng/dL. For the LC-MS/MS assay for serum DHT, the lower limit of quantitation was 1 ng/dL, the linear range 1 to 100 ng/dL, and the interassay CVs in quality control samples with concentrations of 5.2 ng/dL, 22.0 ng/dL, and 44.1 ng/dL were 6.1%, 6.5%, and 8.6%, respectively. For the LC-MS/MS assay for estradiol, the lower limit of quantitation was 1 pg/mL, the linear range 1 to 500 pg/mL, and interassay CVs were 6.9%, 7.0%, and 4.8% at concentrations of 8, 77, and 206 pg/mL, respectively. Free testosterone was measured by a method that uses equilibrium dialysis to separate total and free testosterone, as described (15). A 200-µL serum sample was mixed with 3H-testosterone and dialyzed for 24 hours at 37°C against a buffer mimicking the composition of protein-free plasma using a dialysis membrane with a cut off of 10 000 Daltons. The interassay CVs in low- (free testosterone 4.5 pg/mL), medium- (51 pg/mL) and high-quality (185 pg/mL) control pools were ±14.2%, ±11.2%, and ±11.5%, respectively. Serum PSA was measured by Quest Clinical Trials, Valencia, CA, using the Access Hybritech PSA assay (Brea, CA), a 2-site immunoenzymatic assay (20). The interassay CV was 3.1% for a quality control sample with a mean value of 1.0 ng/mL (20).
For the measurement of the vBMD by quantitative CT at baseline and month 12, the CT technicians at each site were trained by the Central Reading Center (O.N. Diagnostics, Berkeley, CA) that maintained quality control of the CT data collection using phantoms, and performed the image processing, vBMD measurements, and finite element strength analyses, using VirtuOst software, blinded to treatment group (19). HDL cholesterol was assayed at the Laboratory for Clinical Biochemistry Research, University of Vermont, Burlington, VT (21).
The DeRogatis Interview for Sexual Function- II is an updated version of the DISF that includes a sexual desire domain (SDD) and a sexual arousal domain (15, 22). The DISF-II-SDD consists of 4 questions that are answered on a verbal response scale ranging from 0 to 7, with larger values indicating greater desire (22).
Statistical Analysis
We used random forests, a nonparametric regression method, to flexibly model the associations of the average changes in total testosterone, DHT, and estradiol levels over 12 months of testosterone treatment with the changes in hemoglobin, HDL cholesterol, vBMD of the trabecular bone of the lumbar spine, sexual desire, and PSA. We did not include physical function endpoint (6 minute walking distance) as previous analyses did not indicate a strong relationship of testosterone treatment with physical function. Random forests permit the data to specify the shape of the association between predictors and outcomes, including potential nonlinearities. Importantly, the random forest method also enables evaluation of modification of the relationship of changes in 1 hormone (eg, testosterone) with changes in biomarker outcomes by differences in the changes in the levels of other hormones (eg, estradiol and DHT) without imposing strict linearity assumptions on continuous hormone predictors. Separate models were run for each biomarker outcome, and average changes in total testosterone, estradiol, and DHT were included as predictors. The average change was calculated as the mean of change from baseline to 3, 6, 9, and 12 months for each hormone. Baseline values of each biomarker outcome and hormone were additionally included as predictors to account for differences in baseline levels among study participants. Variable importance, a quantitative measure of the predictive utility, was calculated for each predictor as the percent increase in the prediction error when randomly permuting the predictor among study participants, scaled for standard error. Higher values indicate greater predictive value.
We visualized associations by plotting model-based mean changes in markers against respective changes in hormones, holding all other model predictors constant at their mean value. To assess whether the associations of changes in hormones with changes in biomarker outcomes were modified by the extent of changes in other hormones, model-derived mean changes in biomarker outcomes were plotted against respective changes in hormones setting other hormone changes to their 25th (low) and 75th (high) percentiles. Two-stage stepwise selection models were also run for comparison, where the first stage included all possible main effects, and the second stage included all 2-way interactions of selected variables as candidates for entry into the model. The candidate main effects were the same set of predictors as used in the random forests: average change over 12 months in total testosterone, DHT, and estradiol, baseline levels of each hormone, and the baseline value of the respective biomarker. Variables were selected for entry or removal by maximizing the adjusted r-squared. Random forests were run using the randomForest package in R version 3.6.3, and stepwise regression was conducted in SAS version 9.4.
Results
Study Participants
The mean ± SD age among 394 men in the testosterone arm of the TTrials was 72.1 ± 5.7 years, and 251 (63.6%) had a body mass index of 30 kg/m2 or greater (Table 1). Mean ± SD circulating levels of total testosterone, estradiol, and DHT at baseline were 231.8 ± 63.1 ng/dL, 20.3 ± 6.7 pg/mL, and 21.2 ± 11.6 ng/dL, respectively. Mean hemoglobin and HDL cholesterol were in the low ranges of normal for men at 14.0 ± 1.2 g/dL and 44.5 ± 12.7 mg/dL. Mean vBMD of trabecular bone of the lumber spine was 102.4 ± 31.9 mg/cm3. The subset of men participating in the bone trial had similar demographic and clinical characteristics; their mean vBMD of trabecular bone of the lumber spine was 102.4 ± 31.9 mg/cm3. Mean (SD) average changes over 12 months in estradiol, DHT, and total testosterone were 16.2 (13.5) pg/mL, 86.4 (50.5) ng/dL, 323.9 (198.8) ng/dL, respectively, and mean (SD) 12-month changes in hemoglobin, HDL cholesterol, vBMD of trabecular bone of the lumbar spine, sexual desire, and PSA were 0.7 (1.2) g/dL, –1.6 (7.1) mg/dL, 5.9 (6.5), 2.5 (6.4), 0.4 (1.1) ng/mL.
Table 1.
Characteristics of men in the testosterone arm of the testosterone trials: all men enrolled in the TTrials and the men enrolled in the bone trial
| All Men (n = 394) | Bone Trial (n = 110) | |
|---|---|---|
| Age mean (SD) | 72.1 (5.7) | 72.3 (6.3) |
| Race, no. (%) | ||
| White | 348 (88.3) | 93 (84.6) |
| African American | 21 (5.3) | 6 (5.5) |
| Other | 25 (6.4) | 11 (10.0) |
| Ethnicity, no. (%) | ||
| Hispanic | 18 (4.6) | 7 (6.4) |
| Non-Hispanic | 375 (95.4) | 103 (93.6) |
| Concomitant conditions, mean (SD) | ||
| BMI | 31.0 (3.5) | 30.7 (3.7) |
| BMI > 30 (%) | 251 (63.7) | 69 (62.7) |
| Alcohol (drinks/week) | 3.0 (4.3) | 2.5 (3.5) |
| Smoking, no. (%) | ||
| Current (%) | 30 (7.6) | 6 (5.5) |
| Ever (%) | 256 (65.5) | 70 (64.8) |
| Diabetes, no. (%) | 148 (37.6) | 43 (39.1) |
| Hypertension, no. (%) | 286 (72.6) | 77 (70.0) |
| Myocardial Infarction, no. (%) | 53 (13.6) | 19 (17.6) |
| Stroke, no. (%) | 16 (4.1) | 4 (3.7) |
| Sex hormones, mean (SD) | ||
| Testosterone, ng/dL | 231.8 (63.1) | 229.6 (65.3) |
| Free testosteronea, pg/mL | 62.0 (21.4) | 61.2 (20.0) |
| Dihydrotestosterone, ng/dL | 21.2 (11.6) | 20.9 (10.0) |
| Estradiol, pg/mL | 20.3 (6.7) | 20.5 (6.7) |
| Sex hormone–binding globulin, nmol/L | 31.3 (15.2) | 29.1 (12.8) |
| Markers, mean (SD) | ||
| Hemoglobin levels, g/dL | 14.0 (1.2) | 13.9 (1.2) |
| vBMD of trabecular bone of lumber | 102.4 (31.9) | 102.4(31.9) |
| HDL cholesterol, mg/dL | 44.5 (12.7) | 43.7 (12.8) |
| Sexual desireb | 13.9 (7.7) | 13.0 (7.6) |
| Prostate-specific antigen, ng/mL | 1.1 (0.9) | 1.2 (0.9) |
Abbreviations: vBMD, volumetric bone mineral density (mg/cm3), measured using computed tomography; HDL, high-density lipoprotein.
a Determined by equilibrium dialysis.
b Sexual desire was only available in the subset of men who were enrolled in the sexual function trial (n = 230).
Figures 1–5 show predicted trends in biomarker outcomes (hemoglobin, HDL cholesterol, vBMD, sexual desire, and PSA) with changes in each hormone level overall and stratified by low and high changes in other hormones.
Figure 1.
Random forest prediction of mean change in hemoglobin by average change in estradiol (A,B), DHT (C,D), and total testosterone (E,F), fixing changes in other hormones at the mean level in all men (left) and stratifying by high (75th percentile) or low (25th percentile) changes in other hormones (right). (B) Circles, high change in DHT; asterisks, low change in DHT; blue, high change in total testosterone; orange, low change in total testosterone. (D) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in total testosterone; orange, low change in total testosterone. (F) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in estradiol; orange, low change estradiol.
The Relation of Changes in Hemoglobin With Changes in Hormone Levels
The predicted increases in hemoglobin were greater with larger increases in estradiol until approximately + 60 pg/mL (Fig. 1A). For estradiol increases of 60 pg/mL or greater, the estimated mean change in hemoglobin remained stable at + 1.5 g/dL, although sparse data in this range may limit the robustness of conclusions. As shown in Fig. 1B, the increase in hemoglobin associated with increasing estradiol levels was larger with high change in DHT levels compared with low change in DHT but did not differ for high vs low change in total testosterone.
The relation of change in hemoglobin levels with the change in DHT levels was complex, with modest increases in hemoglobin with increases in DHT levels of up to ~100 ng/dL above baseline when other hormone levels were kept constant (Fig. 1 C). Across the range of change in DHT levels, the increase in hemoglobin was higher for larger vs smaller increases in estradiol but did not differ between higher vs lower change in total testosterone level (Fig. 1 B).
The changes in hemoglobin levels were generally stable across changes in total testosterone holding other hormones constant. Across the range of change in total testosterone, increases in hemoglobin were higher when changes in DHT and estradiol levels were also high and lower when changes in DHT and estradiol levels were both low. Expected change in hemoglobin with change in total testosterone was nearly equivalent if changes in either DHT or estradiol were high while change in the other hormone was low.
The variable importance (a quantitative measure of the predictive utility) was the highest for the change in estradiol (Table 2). Pseudo r-squared, an analog of the ordinary least squares metric r-squared that quantifies the proportion of variability in the outcome explained by the model for explaining variability in changes in hemoglobin, was 0.09. Stepwise regression confirmed average change in estradiol as the hormone explaining the most variation in change in hemoglobin levels during testosterone treatment. The final stepwise selection model included average change in estradiol, baseline hemoglobin, baseline estradiol, average change in DHT, and the 2-way interactions of change in estradiol with change in DHT and change in DHT with change in total testosterone. The r-squared of the final model was 0.22.
Table 2.
Random forest importance: precent increase in mean square error (scaled)
| Biomarker Outcome | |||||
|---|---|---|---|---|---|
| Predictor | Hemoglobin | HDL cholesterol | vBMD of trabecular bone of lumbar spine | Sexual desire | PSA |
| Estradiol | 17.3 | 6.4 | 12.1 | 9.0 | 6.7 |
| DHT | 9.5 | 9.0 | 3.1 | 7.9 | 3.5 |
| Total Testosterone | 5.1 | 7.1 | 2.7 | 3.6 | 7.8 |
The importance from random forest prediction of change from baseline to 12 months in hemoglobin, HDL cholesterol, vBMD, sexual desire, and prostate-specific antigen (PSA) using average change in estradiol, DHT, and total testosterone. Larger values indicate greater predictive value of the indicated hormone in the change in the marker from baseline. Importance is determined by calculating the percent increase in the prediction error (mean square error) when values of the indicated hormone are randomly permuted among participants. Models also include baseline hormones and marker values.
The Relation of Changes in HDL Cholesterol With the Changes in Hormone Levels
The changes in HDL cholesterol were weakly explained by the 3 hormones of interest, with a pseudo r-squared of 0.06. Contrary to our expectation, overall, greater increases in estradiol levels were associated with larger decreases in HDL cholesterol level. This association did not vary according to high or low changes in DHT or total testosterone.
Changes in HDL cholesterol were variable across the range of changes in DHT levels although overall, changes in DHT levels were negatively associated with changes in HDL cholesterol levels (Fig. 2C). Greater suppression of HDL cholesterol across increases in DHT levels was observed with high change in estradiol levels than with low change in estradiol levels. Total testosterone did not alter the association of change in DHT with change in HDL.
Figure 2.
Random forest prediction of mean change in HDL cholesterol by average change in estradiol (A,B), DHT (C,D), and total testosterone (E,F), fixing changes in other hormones at the mean level in all men (left) and stratified by high (75th percentile) or low (25th percentile) changes in other hormones (right). (B) Circles, high change in DHT; asterisks, low change in DHT; blue, high change in total testosterone; orange, low change in total testosterone. (D) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in total testosterone; orange, low change in total testosterone. (F) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in estradiol; orange, low change in estradiol.
HDL cholesterol decreased modestly with increases in total testosterone (Fig. 2 E). A greater decrease in HDL cholesterol levels was observed in participants with high increases in estradiol and DHT levels across the range of change in testosterone levels (Fig. 2 F). Importance metrics indicated changes in DHT as most predictive of changes in HDL cholesterol among the 3 hormones considered (Table 2). The final stepwise regression model selected, in order, baseline HDL, baseline estradiol, average change in estradiol, 2-way interactions of average change in estradiol with average change in DHT and average change in DHT with average change in total testosterone, and total baseline testosterone. The r-squared of the final model was 0.16.
The Relation of Changes in Volumetric Bone Mineral Density With the Changes in Each Hormone Level
The strongest predictor of change in vBMD of trabecular bone of the lumbar spine was the average change in estradiol, with a scaled variable importance of 12.1%; importance of average change in total testosterone and DHT was approximately 3.0% for both hormones. The pseudo r-squared was 0.10. As shown in Fig. 3A, vBMD of trabecular bone of the lumbar spine increased with greater increases in estradiol levels. Across the range of increases in estradiol levels, the change in vBMD was greater in participants with high change in testosterone and DHT levels (Fig. 3 B) than with low change in testosterone and DHT levels.
Figure 3.
Random forest prediction of mean change in volumetric bone mineral density (vBMD) by average change in estradiol (A,B), DHT (C,D), and total testosterone (E,F), fixing changes in other hormones at the mean level in all men (left) and stratified by high (75th percentile) or low (25th percentile) changes in other hormones (right). (B) Circles, high change in DHT; asterisks, low change in DHT; blue, high change in total testosterone; orange, low change in total testosterone. (D) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in total testosterone; orange, low change in total testosterone. (F) Circles, high change in estradiol; asterisks, low change in estradiol;; blue, high change in estradiol; orange, low change estradiol.
Increases in vBMD were higher when the increase in DHT was greater than 150 ng/dL, although relatively few men fell within this range; vBMD increased by approximately 4 percentage points for change in DHT below 150 ng/mL and 7 percentage points when change in DHT was above this level. The increases in vBMD were greater across the range of changes in DHT levels in participants with high change in testosterone and estradiol levels than in those with low change in testosterone and estradiol levels. Stepwise regression first selected average change in estradiol, which explained about 20% of the variation in change in vBMD, followed by baseline vBMD, baseline estradiol, and baseline DHT. The r-squared of the model was 0.27.
Relation of Changes in Sexual Desire With the Changes in Hormone Levels
For estradiol increases below 22 pg/mL, sexual desire increases fluctuated between 0 and 2, whereas for average estradiol increase >2 pg/mL, sexual desire increases ranged from 2 to 4 (Fig. 1 A). Increases in sexual desire with increasing estradiol were the largest with high change in DHT (Fig. 4B and 4C). The increase in sexual desire with increasing DHT was larger with high compared with low change in total testosterone holding change in estradiol constant. Sexual desire did not seem to increase with total testosterone when holding changes in estradiol and DHT constant at their mean value.
Figure 4.
Random forest prediction of mean change in sexual desire (vBMD) by average change in estradiol (A,B), DHT (C,D), and total testosterone (E,F), fixing changes in other hormones at the mean level in all men (left) and stratified by high (75th percentile) or low (25th percentile) changes in other hormones (right). (B) Circles, high change in DHT; asterisks, low change in DHT; blue, high change in total testosterone; orange, low change in total testosterone. (D) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in total testosterone; orange, low change in total testosterone. (F) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in estradiol; orange, low change estradiol.
estradiol change had the highest variable importance, followed by DHT and total testosterone (Table 2). The pseudo r-squared for the random forests model was 0.07. Stepwise selection resulted in a model with 2 predictors: baseline sexual desire, which explained 20% of the variation in 12-month change in sexual desire, and average change in estradiol, for a total r-squared of 0.24.
The Relation of Changes in PSA With the Changes in Hormone Levels
There did not appear to be a consistent relation between average changes in estradiol, DHT, and total testosterone with changes in PSA. There was little difference in the predicted 12-month change in PSA with increasing average change in each of the 3 hormones, irrespective of high vs low change in the other hormones (Fig. 5), and the pseudo r-squared from the random forest model was 0.02. The stepwise regression selected, in order, average change in DHT, the interaction of average change in DHT and baseline total testosterone, and baseline total testosterone, and the final r-squared was 0.06.
Figure 5.
Random forest prediction of mean change in serum PSA by average change in estradiol (A,B), DHT (C,D), and total testosterone (E,F), fixing changes in other hormones at the mean level in all men (left) and stratified by high (75th percentile) or low (25th percentile) changes in other hormones (right). (B) Circles, high change in DHT; asterisks, low change in DHT; blue, high change in total testosterone; orange, low change in total testosterone. (D) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in total testosterone; orange, low change in total testosterone. (F) Circles, high change in estradiol; asterisks, low change in estradiol; blue, high change in estradiol; orange, low change estradiol.
We attempted the random forests and stepwise regression analyses with the noncalcified plaque outcome, but given the small number of men in the TTrials in whom cardiovascular outcomes were measured (n = 73 in the treatment arm and with hormone data), results were unreliable and are not shown. We also performed the analysis of the 5 outcomes with free testosterone in the models. The results were similar for the ranking of variable importance of free and total testosterone for predicting changes in outcomes, with the exception of hemoglobin, for which free testosterone had modestly higher variable importance than total testosterone. Given the high correlation of total and free testosterone, and the similarity of the results, the results of the models that included free testosterone are not shown.
Discussion
We show here that the changes in circulating levels of testosterone and its 2 major active metabolites, DHT and estradiol, are related in a complex manner to the observed effects of testosterone treatment on hemoglobin, HDL cholesterol, vBMD, and sexual desire. Our analyses that considered the relation between change in each of the 3 hormones with changes in hemoglobin, HDL cholesterol, vBMD of the trabecular bone of the spine, and sexual desire at low, mean, or high levels of the other 2 hormones suggest that these relations are far more complex than has been previously appreciated. We found that among the 3 hormones, the change in estradiol level during testosterone treatment was the best predictor not only of the change in vBMD and sexual desire, as was expected, but also of the change in hemoglobin and HDL cholesterol levels. Across the range of changes in testosterone, as well as DHT levels, the participants with high change in estradiol levels generally had the greatest change in hemoglobin, HDL cholesterol, vBMD, and sexual desire. Because estradiol and DHT levels during testosterone treatment with a transdermal gel varied substantially among study participants assigned to the testosterone arm of the trial, these findings suggest that simultaneous consideration of changes in on-treatment estradiol as well DHT levels might offer better prediction of response to testosterone treatment than consideration of the testosterone level alone.
Testosterone treatment is associated with a dose-related increase in hemoglobin and hematocrit (23). In univariate analyses, which considered testosterone alone, the testosterone-induced increases in hemoglobin and hematocrit have been reported to be associated with the increase in circulating testosterone concentrations (23). Administration of testosterone and DHT to mice increases hemoglobin and hematocrit (24). Estradiol is generally believed to antagonize the effects of testosterone on erythropoiesis, but the effects of estradiol on erythropoiesis remain incompletely understood and inconsistent across studies and species. One study found that estrogen loss stimulated hematopoiesis in mice (7), while another study in mice reported enhanced megakaryocytopoiesis and erythropoiesis in the bone marrow and spleen in mice after estradiol administration (25). In steers, estradiol implants did not increase red cell counts (26). In contrast to previous univariate analyses in which only circulating testosterone levels were considered without regard to estradiol or DHT levels and which revealed a robust association of serum testosterone levels with the change in hemoglobin (23, 27), the random forest analyses, which related the change in the levels of 1 hormone with the biomarker outcome, found change in estradiol level to be the most robust predictor of change in hemoglobin level. Across the range of change in serum testosterone levels, the men with high change in estradiol and DHT generally had larger increases in hemoglobin levels during testosterone treatment than those with low change in estradiol and DHT.
In these analyses, greater increases in estradiol levels, as well as DHT levels, were associated with larger decreases in HDL cholesterol. In randomized intervention trials, administration of testosterone esters has been associated with a modest suppression of HDL cholesterol (28), but trials of transdermal testosterone gels have shown inconsistent changes in HDL cholesterol. However, administration of oral nonaromatizable steroidal androgens, as well as oral nonsteroidal selective androgen receptor modulators, is associated with marked suppression of HDL cholesterol (29, 30), leading to the prevalent view that aromatization of testosterone to estradiol attenuates the suppression of HDL cholesterol by testosterone treatment relative to nonaromatizable androgens. However, in observational studies, the association of serum testosterone and estradiol levels with HDL cholesterol is sexually dimorphic and varies with menopausal status in women. In community dwelling middle-aged and older men, testosterone levels are positively associated with HDL cholesterol levels (31), but in hyperandrogenic women (32), testosterone levels are negatively associated with HDL cholesterol. Furthermore, some aromatase inhibitors which markedly suppresses estradiol levels have been associated with reduction of HDL cholesterol level in postmenopausal women with breast cancer (33), but anastrozole treatment had no effect on HDL cholesterol in healthy men (34). Some other studies have reported that physiologic levels of estradiol stimulate HDL subfractions (35). Our analyses suggest that increases in estradiol as well as DHT and testosterone levels are related to changes in HDL cholesterol levels during testosterone treatment.
Consistent with previous literature, the changes in estradiol levels from baseline during testosterone treatment were the most predictive of changes in vBMD of the trabecular bone of the spine. However, the increases in vBMD across a range of changes in estradiol levels were higher in participants with high change in DHT and testosterone levels, consistent with the view that, in men, androgens and estrogens both play independent roles in regulating bone resorption and bone formation (36, 37). In epidemiologic studies, estradiol levels are more robustly associated with areal as well as vBMD and with fracture risk, but the highest fracture risk exists in men who additionally have low testosterone levels (36, 37). Mendelian randomization studies have confirmed that loci associated with higher estradiol levels as well as with higher testosterone levels are associated with vertebral bone mineral density (3). Estrogens regulate the activation frequency of bone functional basic multicellular units, the duration of the resorption phase and the formation phase, and osteoclast recruitment (36). The effects of testosterone treatment on lumbar spine bone mineral density are blocked by treatment with anastrozole, an aromatase inhibitor, suggesting that aromatization of testosterone to estradiol is required for maintaining bone mineral density (38). Testosterone directly stimulates osteoblastic bone formation by signaling through the classical androgen receptors on osteoblasts and on mesenchymal stem cells (39) and by stimulating the production of several growth factors within the bone, including insulin-like growth factor 1. Testosterone has also been reported to inhibit apoptosis of osteoblasts through nongenotropic mechanisms (40). Thus, our analyses are consistent with the prevalent view that testosterone’s effects on bone mineral density are mediated through estradiol and via direct actions of testosterone and DHT on the bone (36).
The changes in sexual desire were most robustly associated with estradiol, consistent with studies of men with congenital deficiency of the aromatase enzyme (5) and randomized trials in which administration of aromatase inhibitor suppressed sexual desire even in the presence of normal testosterone levels (2). There did not appear to be a consistent relation between the changes in serum PSA levels with changes in any of the 3 hormones. The PSA promoter contains an androgen response element and the expression of PSA gene in the prostate is regulated by androgens (41). The failure to observe a consistent relation between changes in PSA and testosterone levels is likely because of the nature of the dose response relation between circulating testosterone and PSA levels (6, 42). Serum PSA levels increase in response to testosterone treatment of hypogonadal men but reach a plateau at testosterone concentrations that are near the lower limit of the normal male range (6, 42). Because the baseline testosterone levels in the TTrials participants were only slightly below the lower limit of the normal male range (15), the baseline PSA levels were near the upper end of the dose response curve.
These findings should be considered in the context of the study’s strengths and limitations. These secondary analyses utilized data from the TTrials, 1 of the largest controlled clinical trials of testosterone in which subjects were allocated to intervention arms using minimization. The hormone levels were measured using validated LC-MS/MS assays (11, 15), of which testosterone and estradiol assays are certified by the CDC’s Hormone Standardization Program. We selected 5 continuous biomarker outcomes that are known to be testosterone responsive and that could be measured quantitatively with high levels of precision and accuracy. We used random forest analysis, because it imposes fewer restrictions on the relationships of hormones with markers. It works well with continuous values. We also used step-wise regression modeling to complement the findings of the random forest analysis. These analyses were not prespecified and need confirmation in additional studies of other datasets. Our models were modestly predictive, with r-squared values ranging from 0.16 for HDL cholesterol to 0.27 for vBMD, indicating that there is substantial variation in the testosterone response that is not explained by changes in these 3 hormones. The observed findings in older men in the TTrials may not be directly generalizable to younger men or men with severe organic hypogonadism.
In summary, secondary analyses of the TTrials data that considered the association of changes in the circulating level of testosterone, DHT and estradiol during testosterone treatment with 3 testosterone-responsive quantitative biomarker outcomes while concurrently considering the levels of the other 2 hormones unveiled a more complex relation of the changes in each of the 3 sex hormones with the outcome variable than had been previously appreciated. The changes in serum estradiol levels were the most robustly associated with changes in vBMD and sexual desire during testosterone treatment, as expected, but surprisingly also with the changes in hemoglobin and HDL cholesterol levels. The strength of the relation of each hormone with biomarker outcomes was influenced by the additional consideration of the magnitude of changes in the other 2 hormones. Because the serum estradiol and DHT levels of men treated with the transdermal testosterone gel varied substantially relative to their contemporaneous testosterone level, our findings suggest that consideration of all 3 sex hormones—testosterone, estradiol, and DHT—offer a superior prediction of treatment response for at least some of biomarker testosterone-responsive biomarker outcomes than consideration of each hormone alone. These findings have implications for our current practice of guiding the testosterone treatment of men with hypogonadism almost entirely by monitoring the on-treatment testosterone levels. Serum DHT and estradiol levels are rarely considered in evaluating therapeutic response to testosterone treatment or in dose adjustment. Further validation of these findings in other studies could provide the scientific rationale for measuring for DHT and estradiol and using the levels of these hormones as well as that of total testosterone to guide testosterone treatment of hypogonadal men. This may be particularly relevant to transdermal or oral testosterone formulations whose administration is associated with substantially higher DHT levels than with injectable testosterone esters.
Glossary
Abbreviations
- CT
computed tomography
- CV
coefficient of variation
- DHT
dihydrotestosterone
- HDL
high-density lipoprotein
- LC-MS/MS
liquid chromatography tandem mass spectrometry
- PSA
prostate-specific antigen
- SDD
sexual desire domain
- TTrials
Testosterone Trials
- vBMD
volumetric bone mineral density
Financial Support
The Testosterone Trials were supported by a grant from the National Institute on Aging, National Institutes of Health (U01 AG030644), supplemented by funds from the National Heart, Lung and Blood Institute, National Institute of Neurological Diseases and Stroke, and National Institute of Child Health and Human Development. AbbVie (formerly Solvay and Abbott Laboratories) provided funding and donated the study medication and the placebo gel. Dr. Bhasin was supported partially by the Boston Claude D. Pepper Older Americans Independence Center grant 5P30AG031679 (SB, PI).
Sponsor’s Role
The funding agencies played no role in the design of the trial, analyses of data, preparation of the manuscript, or in the decision to publish.
Clinical Trial Information
ClinicalTrials.gov number: NCT00799617 (registered December 1, 2008).
Conflicts of Interest
Dr. Bhasin reports receiving research grants from NIA, NCMRR, PCORI, FNIH, AbbVie, Metro International Technology, Transition Therapeutics, and Function Promoting Therapies, LLC; equity interest in Function Promoting Therapies, LLC; and personal consulting fees from OPKO and Aditum. These conflicts are managed in accordance with the institutional polices and overseen by the Officer of Industry Interaction of the Mass General Brigham. Dr. Snyder reports research grants from AbbVie and Crinetics and personal consulting fees from Teva. Dr. Stephens-Shields reports research grants from NIDDK, NCI, NIAID, NHLBI, PCORI, and the Bloomberg Family Fund.
Data Availability Statement
Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.
References
- 1. Bhasin S, Jameson JL.In: Jameson JL, Fauci AS, Kasper D, Hauser S, Longo D, Loscalzo J. eds. Harrison’s Textbook of Internal Medicine. McGraw Hills; 2021:2538-63. [Google Scholar]
- 2. Finkelstein JS, Lee H, Burnett-Bowie SA, et al. Gonadal steroids and body composition, strength, and sexual function in men. N Engl J Med. 2013;369(11):1011-1022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Eriksson AL, Perry JRB, Coviello AD, et al. Genetic determinants of circulating estrogen levels and evidence of a causal effect of estradiol on bone density in men. J Clin Endocrinol Metab. 2018;103(3):991-1004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Falahati-Nini A, Riggs BL, Atkinson EJ, O’Fallon WM, Eastell R, Khosla S. Relative contributions of testosterone and estrogen in regulating bone resorption and formation in normal elderly men. J Clin Invest. 2000;106(12):1553-1560. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Bulun SE. Aromatase and estrogen receptor alpha deficiency. Fertil Steril. 2014;101(2):323-329. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Bhasin S, Woodhouse L, Casaburi R, et al. Testosterone dose-response relationships in healthy young men. Am J Physiol Endocrinol Metab. 2001;281(6):E1172-E1181. [DOI] [PubMed] [Google Scholar]
- 7. Jilka RL, Passeri G, Girasole G, et al. Estrogen loss upregulates hematopoiesis in the mouse: a mediating role of IL-6. Exp Hematol. 1995;23(6):500-506. [PubMed] [Google Scholar]
- 8. Bhasin S, Fielder T, Peacock N, Sod-Moriah UA, Swerdloff RS. Dissociating antifertility effects of GnRH-antagonist from its adverse effects on mating behavior in male rats. Am J Physiol. 1988;254(1 Pt 1):E84-E91. [DOI] [PubMed] [Google Scholar]
- 9. Storer TW, Magliano L, Woodhouse L, et al. Testosterone dose-dependently increases maximal voluntary strength and leg power, but does not affect fatigability or specific tension. J Clin Endocrinol Metab. 2003;88(4):1478-1485. [DOI] [PubMed] [Google Scholar]
- 10. Fielder TJ, Peacock NR, McGivern RF, Swerdloff RS, Bhasin S. Testosterone dose-dependency of sexual and nonsexual behaviors in the gonadotropin-releasing hormone antagonist-treated male rat. J Androl. 1989;10(3):167-173. [DOI] [PubMed] [Google Scholar]
- 11. Bhasin S, Travison TG, Storer TW, et al. Effect of testosterone supplementation with and without a dual 5alpha-reductase inhibitor on fat-free mass in men with suppressed testosterone production: a randomized controlled trial. JAMA. 2012;307(9):931-939. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Perusquia M, Stallone JN. Do androgens play a beneficial role in the regulation of vascular tone? Nongenomic vascular effects of testosterone metabolites. Am J Physiol Heart Circ Physiol. 2010;298(5):H1301-H1307. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Li J, Coates RJ, Gwinn M, Khoury MJ. Steroid 5-α-reductase Type 2 (SRD5a2) gene polymorphisms and risk of prostate cancer: a HuGE review. Am J Epidemiol. 2010;171(1):1-13. [DOI] [PubMed] [Google Scholar]
- 14. Swerdloff RS, Wang C, Cunningham G, et al. Long-term pharmacokinetics of transdermal testosterone gel in hypogonadal men. J Clin Endocrinol Metab. 2000;85(12):4500-4510. [DOI] [PubMed] [Google Scholar]
- 15. Snyder PJ, Bhasin S, Cunningham GR, et al. Effects of testosterone treatment in older men. N Engl J Med. 2016;374(7):611-624. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Bhasin S, Brito JP, Cunningham GR, et al. Testosterone therapy in men with hypogonadism: an endocrine society clinical practice guideline. J Clin Endocrinol Metab. 2018;103(5):1715-1744. [DOI] [PubMed] [Google Scholar]
- 17. Snyder PJ, Ellenberg SS, Cunningham GR, et al. The Testosterone Trials: Seven coordinated trials of testosterone treatment in elderly men. Clin Trials. 2014;11(3):362-375. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Snyder PJ, Bhasin S, Cunningham GR, et al. Lessons From the Testosterone Trials. Endocr Rev. 2018;39(3):369-386. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Snyder PJ, Kopperdahl DL, Stephens-Shields AJ, et al. Effect of testosterone treatment on volumetric bone density and strength in older men with low testosterone: a controlled clinical trial. JAMA Intern Med. 2017;177(4):471-479. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Cunningham GR, Ellenberg SS, Bhasin S, et al. Prostate-specific antigen levels during testosterone treatment of hypogonadal older men: data from a controlled trial. J Clin Endocrinol Metab. 2019;104(12):6238-6246. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Mohler ER 3rd, Ellenberg SS, Lewis CE, et al. The effect of testosterone on cardiovascular biomarkers in the testosterone trials. J Clin Endocrinol Metab. 2018;103(2):681-688. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Derogatis LR. The Derogatis interview for sexual functioning (DISF/DISF-SR): an introductory report. J Sex Marital Ther. 1997;23(4):291-304. [DOI] [PubMed] [Google Scholar]
- 23. Coviello AD, Kaplan B, Lakshman KM, Chen T, Singh AB, Bhasin S. Effects of graded doses of testosterone on erythropoiesis in healthy young and older men. J Clin Endocrinol Metab. 2008;93(3):914-919. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Guo W, Bachman E, Li M, et al. Testosterone administration inhibits hepcidin transcription and is associated with increased iron incorporation into red blood cells. Aging Cell. 2013;12(2):280-291. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Landshman N, Bleiberg I. Effect of estradiol on erythropoiesis and megakaryocytopoiesis in mice. Isr J Med Sci. 1979;15(2):140-146. [PubMed] [Google Scholar]
- 26. Angel KL, Schumacher J, Wolfe DF, Klesius PH, Tyler JW, Carson RL. Effects of estradiol 17 beta implants on hematologic values and the chemiluminescence response of neutrophils of steers. Am J Vet Res. 1992;53(1):78-82. [PubMed] [Google Scholar]
- 27. Roy CN, Snyder PJ, Stephens-Shields AJ, et al. Association of testosterone levels with anemia in older men: a controlled clinical trial. JAMA Intern Med. 2017;177(4):480-490. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Whitsel EA, Boyko EJ, Matsumoto AM, Anawalt BD, Siscovick DS. Intramuscular testosterone esters and plasma lipids in hypogonadal men: a meta-analysis. Am J Med. 2001;111(4):261-269. [DOI] [PubMed] [Google Scholar]
- 29. Basaria S, Collins L, Dillon EL, et al. The safety, pharmacokinetics, and effects of LGD-4033, a novel nonsteroidal oral, selective androgen receptor modulator, in healthy young men. J Gerontol A Biol Sci Med Sci. 2013;68(1):87-95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Davis SM, Cox-Martin MG, Bardsley MZ, Kowal K, Zeitler PS, Ross JL. Effects of oxandrolone on cardiometabolic health in boys with Klinefelter syndrome: a randomized controlled trial. J Clin Endocrinol Metab. 2017;102(1):176-184. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Haffner SM, Mykkanen L, Valdez RA, Katz MS. Relationship of sex hormones to lipids and lipoproteins in nondiabetic men. J Clin Endocrinol Metab. 1993;77(6):1610-1615. [DOI] [PubMed] [Google Scholar]
- 32. Senoz S, Ozaksit G, Turhan NO, Gulekli B, Gokmen O. Lipid profiles in women with hirsutism and polycystic ovaries. Gynecol Endocrinol. 1994;8(1):33-37. [DOI] [PubMed] [Google Scholar]
- 33. Wang X, Zhu A, Wang J, et al. Steroidal aromatase inhibitors have a more favorable effect on lipid profiles than nonsteroidal aromatase inhibitors in postmenopausal women with early breast cancer: a prospective cohort study. Ther Adv Med Oncol. 2020;12:1758835920925991. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Dias JP, Shardell MD, Carlson OD, et al. Testosterone vs. aromatase inhibitor in older men with low testosterone: effects on cardiometabolic parameters. Andrology. 2017;5(1):31-40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Bagatell CJ, Knopp RH, Rivier JE, Bremner WJ. Physiological levels of estradiol stimulate plasma high density lipoprotein2 cholesterol levels in normal men. J Clin Endocrinol Metab. 1994;78(4):855-861. [DOI] [PubMed] [Google Scholar]
- 36. Vanderschueren D, Laurent MR, Claessens F, et al. Sex steroid actions in male bone. Endocr Rev. 2014;35(6):906-960. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Khosla S, Amin S, Orwoll E. Osteoporosis in men. Endocr Rev. 2008;29(4):441-464. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Leder BZ, LeBlanc KM, Schoenfeld DA, Eastell R, Finkelstein JS. Differential effects of androgens and estrogens on bone turnover in normal men. J Clin Endocrinol Metab. 2003;88(1):204-210. [DOI] [PubMed] [Google Scholar]
- 39. Vandenput L, Swinnen JV, Boonen S, et al. Role of the androgen receptor in skeletal homeostasis: the androgen-resistant testicular feminized male mouse model. J Bone Miner Res. 2004;19(9):1462-1470. [DOI] [PubMed] [Google Scholar]
- 40. Kousteni S, Chen JR, Bellido T, et al. Reversal of bone loss in mice by nongenotropic signaling of sex steroids. Science. 2002;298(5594):843-846. [DOI] [PubMed] [Google Scholar]
- 41. Riegman PH, Vlietstra RJ, van der Korput J, A, Brinkmann AO, Trapman J. The promoter of the prostate-specific antigen gene contains a functional androgen responsive element. Mol Endocrinol. 1991;5(12):1921-1930. [DOI] [PubMed] [Google Scholar]
- 42. Bhasin S, Woodhouse L, Casaburi R, et al. Older men are as responsive as young men to the anabolic effects of graded doses of testosterone on the skeletal muscle. J Clin Endocrinol Metab. 2005;90(2):678-688. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.
