Sex Hormones Mediate Intensive Lifestyle Intervention–Induced Bone Mineral Density Changes: Look AHEAD Sex Hormone Study

Abstract

Context

The Look AHEAD (Action in Health for Diabetes) randomized controlled trial, which compared an Intensive Lifestyle Intervention (ILI) to Diabetes Support and Education (DSE) in people with overweight/obesity and type 2 diabetes (T2D), showed a greater decline in bone mineral density (BMD) along with greater weight loss in the ILI group. Because weight loss interventions change sex hormones and also affect BMD, understanding the role of sex hormones on these changes has implications for bone health in people with diabetes.

Objective

We assessed sex hormone mediation on BMD changes due to ILI, by sex, among postmenopausal women and older men over 4 years of follow-up.

Methods

In the Look AHEAD Sex Hormone Ancillary Study, we applied structural equation models to estimate the effects of ILI (vs DSE) on hip, femoral neck, and whole-body BMD after 4 years of follow-up, temporally mediated by sex hormones estradiol (E2), total testosterone, bioavailable testosterone, and sex hormone binding globulin (SHBG) measured at 1 year after baseline (nadir of weight loss).

Results

In women, an ILI-associated decrease in estradiol of 14% from baseline to year 1 mediated a decline in whole-body BMD (−1.15 mg/cm², 95% CI: −2.54, −0.21). In men, an ILI-associated increase of 11% in total testosterone mediated a decline of hip BMD (−1.18 mg/cm², 95% CI: −2.68, −0.13).

Conclusion

ILI-associated changes in estradiol and total testosterone resulted in whole-body and hip bone loss, providing insights in the potential role of sex hormones in bone loss accompanying weight loss in older adults with T2D who are at a higher risk of bone fractures.

People with type 2 diabetes (T2D) are at increased risk for fractures (1). T2D impairs bone health and increases fracture risk through mechanisms such as chronic inflammation, formation of advanced glycation end-products, and production of reactive oxygen species (2) The American Diabetes Association recommends screening people with T2D for osteoporosis using dual-energy x-ray absorptiometry (DXA) to assess bone mineral density (BMD) starting at the age of ≥65 years and in those ≥50 years with additional fracture-related risk factors (3).

Look AHEAD (Action in Health for Diabetes) was a randomized controlled trial designed to evaluate the effects of a weight loss targeted intensive lifestyle intervention (ILI) involving restricted caloric intake and increased physical activity, compared to the control, Diabetes Support and Education (DSE), on cardiovascular morbidity and mortality in people with T2D and overweight or obesity (4, 5). The ILI achieved overall benefits including weight loss and improved several cardiovascular risk factors (ie, glycated hemoglobin, systolic blood pressure, and HDL-cholesterol) (4, 5). However, one harm of ILI and the resultant weight loss was greater loss of BMD measured at the hip and femoral neck. Importantly, the ILI effect on bone loss at hip and femoral neck had its nadir at 1 year of intervention in female individuals but persisted for 16 years in male individuals, and also increased the risk of frailty-related fracture (defined as fractures of the hip, pelvis, or upper arm/shoulder fractures) (6-9).

There are several possible reasons why the ILI might have caused a decrease in BMD and fractures. Weight loss itself decreases mechanical loading of bones and reduces BMD (10, 11). Furthermore, weight change and increased physical activity can alter circulating sex hormones that are important for bone metabolism (12-14). Our previous findings from the Look AHEAD Sex Hormone Study demonstrated that after 1 year of the ILI, there were changes in sex hormones, including an ILI-induced increase in sex hormone binding globulin (SHBG) in both postmenopausal female participants and older male participants, a decrease in estradiol (E2) in postmenopausal females, and an increase in total testosterone in males (15). In addition, another prior study from our group found that the ILI-associated increase in total testosterone mediated weight gain in male participants, which may also influence BMD (16).

Sex hormones play important roles in bone metabolism. E2, in particular, maintains bone health by stimulating osteoblast activity for bone formation and inhibiting osteoclast activity for bone resorption in both female and male subjects (17). Testosterone also contributes to bone health by promoting bone formation via increasing osteoblast activity and suppressing bone resorption by indirectly inhibiting osteoclast differentiation, and it can be converted into E2, thereby exerting additional effects through E2 pathways (18). The bioavailability of both E2 and testosterone are regulated by SHBG, and through binding to SHBG the amount of biologically active hormones available for bone tissue can decrease (19).

While the effects of sex hormones on BMD have been extensively studied in both females (both pre- and postmenopausal) and males (12), their role on BMD in the context of weight loss interventions, particularly among people with T2D, who are at higher risk of fractures, remains underexplored (20). In this study, we assessed the extent to which the intervention-associated changes in sex hormones mediated weight loss-associated bone loss. As an exploratory analysis, we also explored the longitudinal effects of intervention-associated changes in sex hormones, through BMD changes, on bone fracture.

Methods

Study Population

Look AHEAD (ClinicalTrials.gov, NCT00017953) was a randomized controlled trial of 5415 people with T2D and overweight or obesity enrolled from 16 US study centers to compare ILI with DSE on the primary outcome, a composite cardiovascular outcome (cardiovascular death, nonfatal myocardial infarction, nonfatal stroke, and hospitalization for angina) (4). The DSE arm provided educational/social support sessions. The ILI arm included portion-controlled and calorie-restricted diet modification and increased moderate intensity physical activity with a goal of at least 7% of initial weight loss (4).

As previously described (15, 21), in the sex hormone ancillary study, we included participants with serum samples at baseline, year 1, or year 4. For female subjects, we excluded those prior to menopause, age <55 years old and history of hysterectomy without oophorectomy, or taking exogenous hormones or breast cancer treatment. For male subjects, we excluded those on exogenous androgen therapy or prostate cancer anti-androgen medication. We then selected a random sample of 2334, with equal numbers of postmenopausal female and male participants. This analysis was further limited to participants with BMD measurements available for at least one of the time points (ie, baseline, year 1, and year 4) (Supplementary Fig. S1) (22).

The sex hormone ancillary study proposal was approved by the Look AHEAD Steering Committee. Institutional Review Board approval for the original Look AHEAD trial and the ancillary study was granted by the Johns Hopkins School of Medicine. All participants provided informed consent as part of the original trial.

Demographics and Clinical Measurements

As previously described (6, 7), participants’ demographic information and smoking history were collected via questionnaires at baseline. Weight and height at baseline, year 1, and year 4 post-intervention were measured using a digital scale and wall-mounted stadiometer, respectively. At baseline, participants were asked to bring all prescription medications for a medication inventory.

Sex Hormone Measurements

The measurements of SHBG, E2, testosterone, and bioavailable testosterone were described in our previous article (15). SHBG was measured via enzyme-linked immunosorbent assay (ELISA) (RRID: AB_3255149, Cat# K151G9K, Mesoscale Discovery, MD) with both high sensitivity (1.76 nmol/L) and specificity (<0.5% cross-reactivity with known circulating proteins) (21). E2 and testosterone were measured using highly sensitive negative electron capture chemical ionization gas chromatography-mass spectrometry (GCMS) where the limits of detection were 3.67 pmol/L for E2 and 0.0347 nmol/L for testosterone. The intraassay variability was 7% for E2 and 4% for testosterone. Additionally, albumin was measured using the dye-binding method, with an intraassay coefficient of variation (CV) of 1.6% and the interassay CV of 2.3% (23, 24). To calculate bioavailable testosterone, which includes both albumin-bound and free (unbound) testosterone, we used the Vermeulen equation, which used albumin and SHBG concentrations, and albumin was measured using the dye-binding method (25).

Bone Mineral Density Measurements

As previously described (6), BMD was measured by DXA in a substudy conducted at 5 of the 16 study centers. Participants with body weight >300 lbs. (exceeding the DXA scanner limit) were excluded. DXA was performed with a Hologic (QDR-4500A) fan beam densitometer, and any software upgrades were approved by the DXA quality assurance center (San Francisco Coordinating Center, University of California San Francisco). To ensure quality and consistency across sites, longitudinal device performance was monitored using regular scans of spine and whole-body phantoms, with coefficients of variation ranging from 0.36%-0.39% to 1.6%-2.4%. In this study, the BMD measurements included whole-body BMD, total hip BMD, and the femoral neck (of the hip) BMD, all in mg/cm². To align with previous Look AHEAD study examining bone loss after 4 years of follow-up (7), we selected total BMD at the hip as our primary outcome. Other BMD measurements at the femoral neck of the hip and whole body were also explored in this study.

Fracture Events

As described earlier (9), incident fractures, defined as the first occurrence of fracture after the baseline, were identified through participant self-reports semi-annually. Reports of potential fracture events were followed by trained physicians reviewing hospital and outpatient records. Centrally adjudicated fracture events included in this study were: hand (not fingers), lower arm/wrist, elbow, upper arm/shoulder/clavicle, vertebra, tailbone, pelvis, hip, upper leg, knee, lower leg/ankle, foot (not toes) fractures.

Statistical Analyses

All analyses were stratified by sex. To summarize the treatment arm–specific changes in BMD from baseline to year 4 within the current analytical sample, we calculated medians and interquartile intervals (IQI) and used Wilcoxon's rank-sum test (P < .05) to assess the difference in BMD change between the ILI and DSE groups. For sex hormones, we included SHBG, E2, total testosterone, and bioavailable testosterone. Due to the right-skewed distributions of these variables, natural log transformations were performed before inclusion in all regression-based statistical analyses. To calculate the changes in sex hormones from baseline to year 1, we computed the median and IQIs of percent changes, namely [(year-1 sex hormone − baseline sex hormone)/baseline sex hormone] × 100.

For the temporal mediation analysis, we used structural equation modeling (SEM), including 3 time points: baseline (year 0), year 1, and year 4. We constructed 3 SEM models: Model 1 delineated the temporal sequence of the ILI (vs DSE) effect through sex hormones on BMD change, adjusting for baseline age, race, and study site. Given that part of the ILI effects on year-4 BMD could also be mediated by year-1 weight change, Model 2 expanded on Model 1 by incorporating year-1 weight change as an additional mediator. This approach allowed us to quantify the mediation effects of year-1 sex hormones on year 4 BMD changes independently of year-1 weight change. Model 3 was built to explore the mediation effects of year-1 sex hormones on post-year-4 fracture risk through the year-4 BMD changes.

In Model 1, there were 3 effect pathways to year-4 BMD: cis-mediated effects (ILI to year-1 BMD to year-4 BMD), cross-lagged mediated effects (ILI to year-1 sex hormones to year-4 BMD), and direct effect (ILI to year-4 BMD), and the total effects were the net sum of cis-mediated effects, cross-lagged mediated effects, and direct effects (Fig. 1). In Model 2, there were 2 cross-lagged mediated pathways: ILI to year-1 sex hormones to year-4 BMD (ie, sex hormone–cross-mediated effects) and ILI to year-1 weight to year-4 BMD (ie, weight-cross-mediated effects), and therefore the total effects on year-4 BMD were the net sum of these 4 pathways (ie, the 2 cross-lagged mediated, cis-mediated, and direct effects) (Fig. 2). Since our goal was to assess the mediating role of sex hormones on BMD, the major focus was the sex hormone–mediated pathway from ILI to year 4 BMD (the dashed arrows in Figs. 1 and 2). For Model 3 where the outcome was fracture, the pathway of interest was from ILI to year-1 sex hormones, to year-4 BMD, and then to post-year 4 fractures (Fig. 3).

Figure 1.

The structural diagram for structural modeling equation analysis Model 1. All single-headed arrows represent structural effects. The dashed arrows represent the year-1 sex hormone–mediated pathway. The double-headed curved arrow represents the covariance between baseline sex hormones and BMD. The analyses were conducted separately by sex.

Abbreviations: BMD, bone mineral density; DSE, Diabetes Support and Education; ILI, Intensive Lifestyle Intervention.

Figure 2.

The structural diagram for structural modeling equation analysis Model 2. All single-headed arrows represent structural effects. The dashed arrows represent the year-1 sex hormone–mediated pathway. The double-headed curved arrows represent the covariance between baseline sex hormones and BMD and between weight and sex hormones. The analyses were conducted separately by sex.

Abbreviations: BMD, bone mineral density; DSE, Diabetes Support and Education; ILI, Intensive Lifestyle Intervention.

Figure 3.

The structural diagram for structural modeling equation analysis of incident bone fracture after 4 years of intervention. All single-headed arrows represent structural effects. The dotted dashed arrows represent the pathway of ILI effects on fracture mediated by year-1 sex hormone and year-4 BMD. The analyses were conducted separately by sex.

Abbreviations: BMD, bone mineral density; DSE, Diabetes Support and Education; ILI, Intensive Lifestyle Intervention.

The sex hormone mediation effects were estimated as the product of regression coefficients of ILI as a predictor of year-1 sex hormone and year-1 sex hormone as a predictor for year-4 BMD. For regression diagnostics, we examined exploratory Lowess-smoother plots to check for linearity assumptions of the structural effects and evaluated the normality of the residuals via Q–Q plots, skewness, and kurtosis. We estimated the empirical CIs of our results using bootstrapping and presented the median of the 1000 iterations as the effect estimate and the 2.5th to 97.5th percentiles as the 95% CIs. We also reported 90% CIs (ie, 5th and 95th percentiles of the 1000-iteration bootstrapped coefficients) to identify the mediation effects that approach but did not achieve the significance level of 95% CIs. We determined graphically that 1000 bootstrap iterations were adequate to achieve stable estimates. For the hypothesized sex hormone mediation effects reaching statistical significance of 95% CIs, we further reported the mediated effects through year-1 BMD itself and weight on year-4 BMD for comparison.

We conducted a sensitivity analysis by further adjusting for the following baseline covariates: smoking status, insulin use, body mass index (BMI), and diabetes duration in both Models 1 and 2.

All analyses were conducted using Stata SE (Version 18.0 Statacorp, College Station, TX).

Results

We included a total sample of 774 Look AHEAD participants who had measured both sex hormones and BMD (n = 356 postmenopausal females; n = 418 males) (Table 1). The mean age at baseline was 60 years for both sexes, and 12.8% and 5.7% were of Black race among female and male participants, respectively.

Table 1.

Baseline characteristics by sex and randomization arms, ILI vs DSE

	Female		Male
	ILI N = 181	DSE N = 175	ILI N = 203	DSE N = 215
Age (years), mean (SD)	59.4 (5.4)	59.4 (5.7)	60.2 (6.5)	59.8 (6.5)
Race/ethnicity (%)
Black	21 (11.6%)	23 (13.1%)	8 (3.9%)	16 (7.4%)
White	92 (50.8%)	77 (44.0%)	149 (73.4%)	159 (74.0%)
Hispanic	61 (33.7%)	68 (38.9%)	34 (16.7%)	30 (14.0%)
Othera	7 (3.9%)	7 (4.0%)	12 (5.9%)	10 (4.7%)
Smoking status (%)
Never	116 (64.1%)	121 (69.1%)	81 (39.9%)	86 (40.0%)
Past	57 (31.5%)	49 (28.0%)	110 (54.2%)	121 (56.3%)
Current	8 (4.4%)	5 (2.9%)	12 (5.9%)	8 (3.7%)
Family income
<$20K	41 (24.1%)	47 (28.1%)	20 (10.4%)	13 (6.4%)
$20K–$40K	60 (35.3%)	49 (29.3%)	32 (16.7%)	32 (15.8%)
$40K–$60K	32 (18.8%)	32 (19.2%)	31 (16.1%)	35 (17.2%)
$60K–$80K	14 (8.2%)	21 (12.6%)	25 (13.0%)	38 (18.7%)
>$80K	23 (13.5%)	18 (10.8%)	84 (43.8%)	85 (41.9%)
BMI (kg/m2), mean (SD)	36.0 (5.8)	35.5 (5.1)	33.6 (4.5)	33.8 (4.3)
Weight (kg), mean (SD)	91.4 (17.3)	91.0 (13.9)	102.7 (14.9)	104.4 (14.1)
Waist circumference (cm), mean (SD)	108.2 (12.9)	108.3 (10.8)	114.1 (11.3)	114.6 (11.4)
HbA1c %, mean (SD)	7.2 (1.2)	7.4 (1.3)	7.2 (1.2)	7.1 (1.1)
Diabetes duration (years), mean (SD)	6.0 (6.1)	6.6 (5.9)	7.6 (7.0)	7.2 (6.5)
Insulin use, %	25 (13.8%)	34 (19.5%)	43 (21.2%)	29 (13.5%)
Statin use, %	74 (40.9%)	71 (40.6%)	107 (52.7%)	105 (48.8%)
Antihypertensive medication use, %	132 (72.9%)	127 (73.0%)	140 (69.0%)	139 (64.7%)
SHBG (nmol/L), median (IQI)	36.9 (23.4, 67.5)	36.2 (23.2, 73.6)	32.5 (22.2, 59.4)	33.9 (23.4, 52.7)
E2 (pmol/L), median (IQI)	38.8 (25.3, 61.4)	35.7 (22.5, 59.9)	104.9 (73.5, 135.3)	99.4 (71.7, 131.2)
Total T (nmol/L), median (IQI)	0.8 (0.5, 1.3)	0.8 (0.5, 1.1)	16.7 (12.8, 22.2)	16.4 (12.3, 20.5)
BioT (nmol/L), median (IQI)	0.3 (0.2, 0.5)	0.3 (0.1, 0.4)	7.4 (4.5, 11.3)	7.2 (5.0, 9.8)
Hip BMD (mg/cm2), mean (SD)	1000 (151)	1010 (137)	1081 (124)	1093 (139)
Femoral neck BMD (mg/cm2), mean (SD)	838 (140)	843 (125)	871 (122)	884 (127)
Whole-body BMD (mg/cm2), mean (SD)	1085 (131)	1084 (115)	1172 (100)	1188 (111)

Abbreviations: BioT, bioavailable testosterone; BMI, body mass index; BMD, bone mineral density; DSE, Diabetes Support and Education; E2, estradiol; ILI, Intensive Lifestyle Intervention; IQI, Interquartile interval; SHBG, sex hormone binding globulin; Total T, total testosterone.

^a Other includes American Indian, Native American, Alaskan Native, Asian/Pacific Islander, and Mixed race/ethnicity.

Changes in BMD, Sex Hormones, and Weight From Baseline, by Study Arm and Sex

Females in the ILI (vs DSE) arm experienced a greater decline from baseline to year 4 than the DSE arm in BMD at the hip, femoral neck, and whole body. Similar to females, males in the ILI (vs DSE) arm experienced a greater decline in hip and femoral neck BMD from baseline to year 4, but not the whole-body BMD, which showed greater increase in the DSE group (Supplementary Table S1) (22).

Both female and male participants in the ILI group showed a greater increase in SHBG compared to the DSE group. Furthermore, female participants had a greater decrease in E2 in the ILI group, and they also had a decrease in total and bioavailable testosterone in the ILI group as compared to a slight increase in the DSE group. Male participants, by contrast, had a greater increase in total testosterone among those in the ILI group (Supplementary Table S2) (22).

In females, the mean (SD) weight change from baseline to year 1 was −8.5 (5.1) kg in ILI and −0.6 (4.2) kg in DSE; in males, it was −10.2 (7.4) kg in ILI and −0.7 (4.4) kg in DSE.

Sex Hormone Mediation Effects on BMD

We evaluated the mediation effects of year-1 SHBG, E2, total testosterone, and bioavailable testosterone on year-4 BMD. In females, we observed that ILI-associated change in E2 at year 1 mediated a reduction in whole-body BMD at year 4 by 1.15 mg/cm² (95% CI: 2.54, 0.21), and its mediation on hip and femoral neck BMD also showed a similar direction, although the CIs crossed the null (Fig. 4). Changes in total and bioavailable testosterone also had an overall tendency in mediating decline in hip BMD, despite the CIs overlapping null.

Figure 4.

Estimated year-1 sex hormone–mediating effects on year-4 hip BMD (A), femoral neck BMD (B), and whole-body BMD (C) from Model 1 in females and males. Model 1 was adjusted for baseline age, race, and study site. The median mediation effects and the bootstrapped 95% CIs were to the right of each plot. The mediation effects of all BMDs are in mg/cm².

Abbreviations: BioT, bioavailable testosterone (nmol/L); BMD, bone mineral density; E2, estradiol (pmol/L); SHBG, sex hormone binding globulin (nmol/L); Total T, total testosterone (nmol/L).

In male subjects, we observed that ILI-associated change in total testosterone at year 1 mediated a reduction in hip BMD at year 4 by 1.18 mg/cm² (95% CI: 2.68, 0.13), and it had a similar pattern of mediation in femoral neck BMD (−0.80 mg/cm², 95% CI crossing null; 90% CI: −1.81, −0.03) (Fig. 4). In addition, SHBG also demonstrated a central tendency of mediating decline in hip (−0.65 mg/cm², CIs crossing null) and femoral neck BMD (−0.79 mg/cm², 95% CI crossing null; 90% CI: −1.69, −0.04).

After incorporating weight as an additional mediator in Model 2, the observed mediation effects from sex hormones were similar (Supplementary Fig. S2) (22). For year-4 hip BMD in males, the estimated mediation effect was −1.15 mg/cm² (95% CI: −2.62, −0.09) through year-1 total testosterone, −0.08 mg/cm² (95% CI: −2.80, 2.73) through year-1 weight, and −18.9 mg/cm² (95% CI: −24.5, −13.8) through year-1 hip BMD. For year-4 whole-body BMD in females, the estimated mediation effect was −0.96 mg/cm² (95% CI: −2.22, −0.08) through year-1 E2, −2.31 mg/cm² (95% CI: −5.39, 0.96) through year-1 weight, and 1.53 mg/cm² (95% CI: −4.93, 7.53) through year-1 whole-body BMD.

The sensitivity analysis included additional adjustments for baseline smoking status, insulin use, BMI, and diabetes duration. These analyses were consistent with the main findings.

Sex Hormone Effects on Fractures via BMD

Among the 777 participants in this study, 64 incident fractures occurred in female participants and 31 occurred in male participants after year 4. Despite the high uncertainty reflected as the wide CIs, we observed that in females, ILI-induced changes in year-1 E2 mediated 0.5% higher risk in post-year-4 fracture (95% CI: −0.5%, 2.1%; P value: .236) via year-4 whole-body BMD. In males, year-1 total testosterone mediated a 0.2% higher risk in post-year-4 fracture (95% CI: −0.5%, 1.9%; P value: .43) via year-4 hip BMD (Fig. 5).

Figure 5.

Hazard ratio (HR) and bootstrapped 95% CI for incident fracture comparing ILI vs DSE, mediated by year-1 sex hormone and year-4 hip BMD (A), femoral neck BMD (B), and whole-body BMD (C). Incident fracture is defined as having any fracture at hand (not fingers), lower arm/wrist, elbow, upper arm/shoulder/clavicle, vertebra, tailbone, pelvis, hip, upper leg, knee, lower leg/ankle, foot (not toes) post year 4 of study.

Abbreviations: BioT, bioavailable testosterone (nmol/L); BMD, bone mineral density; DSE, Diabetes Support and Education; E2, estradiol (pmol/L); ILI, Intensive Lifestyle Intervention; SHBG, sex hormone binding globulin (nmol/L); Total T, total testosterone (nmol/L).

Discussion

Our analysis found that, among individuals with T2D, the change in E2 at year 1 associated with ILI (after maximal intervention effect on weight loss) mediated a modest decline in whole-body BMD after 4 years in postmenopausal females. In older males, the ILI-associated change in total testosterone mediated a small decline in hip BMD after 4 years. Notably, although the mediation effects through year-1 sex hormones were modest in magnitude compared to those through year-1 weight or BMD themselves, they remained consistent even after accounting for year-1 weight.

A previous Look AHEAD analysis reported a greater reduction in bone density loss in the ILI group than DSE group after 1 year of intervention in both females and males, although these effects diminished after 4 years of intervention in females while they persisted up 16 years after intervention in males (6-8). Our findings on BMD changes in this sample are consistent with the prior analysis (6-8). We expanded on this finding by showing that changes in the BMD in females and males had detectable mediation effects from sex hormones, specifically E2 and total testosterone.

In females, we observed that ILI-induced decrease in E2 at year 1 mediated a decline in whole-body BMD at year 4. E2 has been known to support bone health by acting on all key bone cell types including osteoclasts, osteoblasts, and osteocytes to promote bone formation and suppress bone resorption (17, 26, 27). E2 can act directly on osteoclasts to suppress their differentiation by interfering with Receptor Activator of NF-κB Ligand (RANKL) signaling, and it can also act indirectly on them through suppressing RANKL production by other cells (eg, T cells) and modulating bone-resorbing cytokines (eg, interleukin [IL]-1, 6, and tumor necrosis factor (TNF)-α) (17). E2 also inhibits the apoptosis of osteoblasts and prolongs their lifespan by reducing oxidative stress and suppressing NF-κB signaling (17). Furthermore, E2 suppresses the apoptosis of osteocytes and thus limits bone remodeling (17).

Testosterone could also exert protective effects on bone through increased expression of bone anabolic factors such as insulin-like growth factor-1 (IGF-1) and transforming growth factor-β (TGF-β) in osteoblasts, and it suppresses the formation of osteoclasts and shortens their lifespan via increased IL-6 activation (12, 18, 28). In addition, testosterone could be also converted to E2 via aromatase activity, through which it can also impact the bone cells via the pathway from E2 (18). SHBG is also involved in bone metabolism as it regulates the bioavailability of both E2 and testosterone by binding to 20% to 40% and 50% to 60% of them respectively (19, 29). Beyond its role as a binding protein for sex steroids, SHBG could also directly modulate intracellular signaling in osteoblasts, and it was also reported to be independently associated with low BMD, even after accounting for sex steroid concentrations (19, 30, 31).

While the existing literature has shown a bone-protective nature of testosterone (18, 32, 33), we observed that, in males, a decline in hip BMD at year 4 was mediated by an ILI-induced increase in total testosterone at year 1, but this decline was not mediated by changes in bioavailable testosterone (ie, testosterone not binding to SHBG). This discrepancy suggests that the SHBG-bound fraction of testosterone, which is typically considered to be biologically inactive, may play a role in bone metabolism, possibly through the influence of SHBG itself. While current evidence is still limited, SHBG has been identified as an independent predictor of lower BMD, regardless of sex steroid levels (19, 30, 31). Our findings are consistent with this, as SHBG had a negative central tendency in its mediation effect on both hip and femoral BMD in males. Additionally, in the male population of our study, ILI led to a greater elevation in total testosterone and SHBG but not in bioavailable testosterone at year 1, suggesting that the ILI-induced rise in SHBG might offset the increase in total testosterone, which in turn hindered an elevation in bioavailable testosterone, and consequently no mediation was observed from it. Therefore, the concurrent changes in SHBG and total testosterone may explain the observed negative mediation effects on hip BMD from total testosterone rather than bioavailable testosterone.

For the mediation effects of ILI-associated changes in year-1 sex hormones on post-year-4 fractures through year-4 BMD, we observed a central tendency toward higher fracture risk mediated by E2 in females and by total testosterone in males. However, these estimates had high uncertainties due to limited statistical power. Moreover, given that a prior study in Look AHEAD found no significant treatment effects of ILI on all fractures (9), these results should be interpreted with caution and were included here primarily for completeness.

Our study had several notable strengths. First, we focused on a population of older adults with overweight/obesity and T2D who are high risk for bone loss and even frailty fracture due to weight loss (11, 34, 35). This is salient as people living with T2D have a 40% to 70% higher risk of bone fractures compared to the general population (1). Additionally, despite many of its many overwhelming benefits, the Look AHEAD's ILI intervention was associated with one facet of harm, that is, increased frailty fracture risk (9). Second, prior studies examining the relationship between sex hormones and BMD used cross-sectional analyses, but our study assessed the longitudinal effects of hormones on BMD and was able to assess these changes in the context of an ILI targeting weight loss. Our current analysis leveraged the temporality of the trial and utilized structural equation modeling to assess the mediation effects. This approach provides deeper insights into the potential role played by sex hormone changes on bone integrity during weight loss. Third, we used highly sensitive mass spectrometry that enabled E2 and testosterone to be measured at low concentrations that might not be otherwise detected by methods like radioimmunoassay (RIA) (15, 36).

We also acknowledge some study limitations. First, as mentioned in previous Look AHEAD studies on BMD, the use of DXA to measure BMD during weight change may be subject to measurement errors caused by aging-related artifacts and the impacts of fat distribution and soft tissue thickness (6, 7, 10, 37). BMD assessed using quantitative computed tomography is considered a better approach to detect bone loss in people with osteoarthritic changes in the lumber spine, but this method was not utilized in Look AHEAD (38). Nevertheless, DXA is clinically relevant and is recommended by the American Diabetes Association for BMD monitoring in older adults with diabetes (39). Second, we did not include lumbar spine BMD in our study because it was thought to be impacted by aging-related artifacts, such as osteophytes (8). Previously published Look AHEAD articles investigating BMD at 8 years and 12-16 years after intervention found that whole-body BMD in males of both randomization arms increased over time, but the site responsible for this increase was the lumbar spine (8). Given the measurement inaccuracy in lumbar spine, the observed small increase in whole-body BMD in males and the mediation effects on whole-body BMD should be interpreted with caution. Third, our study sample was considerably smaller than previous published Look AHEAD BMD studies (∼1300 participants), which reduced the statistical power (6, 7). Therefore, for some results, although the central tendency was observed, the wide confidence intervals bear the uncertainty of whether they reflect true physiological effects or insufficient statistical power. Fourth, we were unable to account for certain baseline covariates that may be associated with both bone and sex hormone metabolism that were not available in the dataset, such as dietary or supplemental calcium and vitamin D intake and alcohol consumption. Fifth, the mediation effects of bioavailable testosterone should be interpreted with caution due to limitations in its calculation from Vermeulen equation, which is based on linear binding dynamics of testosterone with SHBG and albumin, and such assumption has been questioned in the recent reappraisal (40). Hence, calculated bioavailable testosterone may not accurately reflect the true biologically active fraction.

In conclusion, our study demonstrated that changes in sex hormones, particularly total testosterone, mediated hip bone loss in both males and females undergoing an intensive lifestyle intervention and may help explain our Look AHEAD findings of increased frailty fracture risk in the ILI group. These findings provided a deeper understanding of the potential role of sex hormones underlying bone loss accompanying weight loss, specifically during lifestyle intervention and in older adults with T2D who are at a higher risk of bone fractures.

Abbreviations

BMD

bone mineral density

DSE

Diabetes Support and Education

DXA

dual-energy x-ray absorptiometry

E2

estradiol

IL-

interleukin

ILI

intensive lifestyle intervention

IQI

interquartile interval

SHBG

sex hormone binding globulin

T2D

type 2 diabetes

Funding

This work was funded by NIH/National Institute of Diabetes Digestive & Kidney Diseases grants R01DK127222 and U01DK57149.

Disclosures

The authors had full access to all of the data in this study and they take complete responsibility for the integrity of the data and the accuracy of the data analysis. M.W. reports recent consultancy to Freeline in the last 3 years. Unrelated to this work, E.D.M. has served as a consultant for Amgen, Arrowhead, AstraZeneca, Bayer, Boehringer Ingelheim, Edwards Life Science, Esperion, Ionis, Eli Lilly, Medtronic, Merck, New Amsterdam, Novartis, Novo Nordisk, and Zoll.

Data Availability

Some or all data from this study may be restricted to protect patient confidentiality or due to licensing agreement. The corresponding author can provide details on request about the restrictions and the conditions under which access to certain data may be provided.

Clinical Trials Information

The Look AHEAD (Action for Health in Diabetes) trial's ClinicalTrials.gov number is NCT00017953.