Sex Hormones Levels and Subclinical Atherosclerosis in Postmenopausal Women: The Multi-Ethnic Study of Atherosclerosis

Abstract

We examined cross-sectional associations between sex hormones and carotid artery intimal-medial thickness (cIMT) and coronary artery calcium in women in the Multi-Ethnic Study of Atherosclerosis.

Serum testosterone, estradiol, sex hormone binding globulin (SHBG), and dehydroepiandrosterone levels were measured in 1,947 postmenopausal women aged 45-84 years (30% White, 14% Chinese-American, 31% Black, and 25% Hispanic) and not on hormone therapy. Using multiple linear regression we evaluated associations between log(sex hormone) levels and log(cIMT) adjusted for age, ethnicity, body mass index (BMI) and cardiac risk factors. Associations between sex hormone levels and the presence and extent of coronary calcium were evaluated.

Total and bioavailable testosterone were positively associated with common cIMT independent of age, BMI, hypertension, smoking, HDL-cholesterol, LDL-cholesterol and insulin sensitivity (p=0.009 and p=0.002 respectively). SHBG was negatively associated with common cIMT (p=0.001) but further adjustment for BMI, cardiovascular risk factors, and LDL- and HDL-cholesterol removed significance. Estradiol and dehydroepiandrosterone were not associated with common cIMT. Sex hormones were not associated with presence of coronary calcium. Among women with measurable coronary calcium, higher SHBG (p=0.012) and lower bioavailable testosterone (p=0.007) were associated with greater coronary calcium score. No heterogeneity by ethnicity was found. In postmenopausal women, testosterone is independently associated with greater common cIMT. SHBG is negatively associated and this may be mediated by LDL- and HDL-cholesterol. In contrast, SHBG and testosterone were associated with extent of coronary calcium but in the opposite direction compared to carotid intimal-medial thickness. These differences warrant further evaluation.

Introduction

Men have higher coronary heart disease (CHD) risk than women. Endogenous sex hormones (SH) may be associated with CHD risk, perhaps through effects on CHD risk factors. However, evidence supports a direct relationship of SH levels with coronary disease without risk factor mediation.¹ The association between testosterone (T) level and atherosclerosis in women is controversial, with reports of both positive associations¹ and lack of association² between T and clinically manifest coronary disease, and inverse associations with carotid intimal-medial thickness (cIMT). ^{3, 4} Women with cardiovascular disease (CVD) have been found to have lower sex hormone binding globulin (SHBG) levels and higher free androgen index (FAI) compared to matched disease-free controls, though this was not independent of body mass index (BMI) and other risk factors. ⁵

The relationship of endogenous estrogens to CHD in women is also controversial. Menopausal age appears inversely associated with CVD mortality⁶, suggesting lifetime exposure to estrogen reduces CVD risk. Lower free estradiol (E2) levels are associated with an atherogenic lipid profile ⁷ while women with impaired glucose tolerance or diabetes have higher E2 levels.⁸ No relationship between endogenous estrogen levels and coronary atherosclerosis or CHD mortality has been reported.^{1, 2}

Many studies were small or did not examine the effect of ethnicity on these relationships. This study examines cross-sectional associations of endogenous SH concentrations with the presence of subclinical atherosclerosis in a large multiethnic sample of postmenopausal female participants of the Multi-Ethnic Study of Atherosclerosis (MESA).

Methods

Study Population

MESA is a six-center study of the prevalence and correlates of subclinical CVD and the factors influencing its progression.⁹ Between July 2000 and August 2002, 6814 men and women, self-identified as white, black, Hispanic or Chinese-American, aged 45-84 years and free of clinical CVD, were recruited in 6 communities: Baltimore City/Baltimore County, MD; Chicago, IL; Forsyth County, NC; Los Angeles County, CA; Northern Manhattan/the Bronx, NY; and St. Paul, MN. Details on the sampling frames and examination procedures have been published.⁹ Each participant provided written informed consent. The study was conducted in accordance with institutional guidelines and Institutional Review Board approval.

Of the 3,601 MESA women participants, 3176 were considered postmenopausal based on either self-report, history of prior bilateral oophorectomy, age ≥ 55 years, or if the woman had prior hysterectomy with 0 or 1 ovary removed and self-reported as postmenopausal had E2 levels within the range of women aged ≥ 55 years. Women were excluded if on current hormone therapy (n=1084), or they had no available SH levels (n=119) or no cIMT data (n=26). This analysis, therefore, included 1,947 women, including 549 women with prior hormone therapy. The mean time since stopping hormone therapy was 9.1 ± 10 years in the 538 women who provided this information.

At the baseline visit sex, age, race/ethnicity, smoking, medication use, BMI (kg/m²) and resting blood pressure were recorded. Fasting cholesterol and glucose levels were measured at the Collaborative Studies Clinical Laboratory at Fairview-University Medical Center (Minneapolis, MN). Total and high-density lipoprotein cholesterol were measured using cholesterol oxidase methods. Low-density lipoprotein cholesterol (LDL-C) was calculated using the Friedewald equation when triglyceride <400 mg/dL.

Hypertension was defined as receiving antihypertensive medications or a blood pressure ≥ 140 mm Hg systolic or 90 mm Hg diastolic. Diabetes was defined as receiving hypoglycemic medication or a fasting blood glucose of ≥ 126 mg/dl. Smoking status was categorized as never, former, or current.

Endogenous Sex Hormones

Fasting blood samples were drawn between 7:30 am and 10:30 am. Serum was stored at -70°C, then shipped on dry ice for freezer storage to the University of Vermont Central Blood Analysis Laboratory. Serum SH concentrations were measured in the University of Massachusetts Medical Center Sex Hormone Laboratory (Worcester, MA). Total T and dehydroepiandrosterone (DHEA) were measured using radioimmunoassay kits, SHBG by chemiluminescent enzyme immunometric assay, and E2 by an ultra-sensitive radioimmunoassay kit (see Online Supplement). Assay variability was monitored by including approximately 5% blind quality control samples in each batch of analyzed samples. Bioavailable T was calculated from T and SHBG as determined by immunoassay and which has been shown to be comparable to apparent free testosterone concentration obtained by equilibrium dialysis¹⁰. The overall coefficients of variation for total T, SHBG, DHEA and E2 were 12.3%, 9.0%, 11.2% and 10.5% respectively.

Carotid Ultrasound

At the baseline examination B-mode ultrasound images of the near and far walls of the distal common carotid artery and the extracranial portion of the internal carotid artery were obtained.⁹ Digitized scans were read at the New England Medical Center Ultrasound Reading Center. The maximum cIMT was measured for each view. The means of the maximal cIMT from all views of both left and right internal and common carotid arteries were calculated. Intra-class correlation coefficients for intra-reader reproducibility of common and internal carotid IMT both exceeded 0.98 and for inter-reader reproducibility were 0.87 and 0.94, respectively.

Coronary Calcium

Participants underwent computed tomographic chest scans at baseline. Images were obtained during a single breath hold with ECG gating⁹ using either an ECG-triggered electron-beam computed tomography scanner or prospectively ECG- triggered scan acquisition at 50% of the RR interval with a multi-detector computed tomography system. Scans were read centrally at the Harbor-UCLA Research and Education Institute to identify and quantify coronary calcification, including readings adjusted for differences in radiographic densities of standard calcium phantoms. The mean Agatston score obtained from two scans in each patient was used in all analyses.¹¹ Calcification was considered as present when Agatston score was >0. Kappa statistics for intra- and inter-reader reproducibility of coronary calcium prevalence were both 0.92. Intra-class correlation coefficients for intra- and inter-reader reproducibility of coronary calcium scores exceeded 0.99.

Statistial Analyses

The prevalence and levels of cardiovascular risk factors were tabulated by ethnic group. Differences between groups were tested by analysis of variance, nonparametric Kruskal-Wallis tests, and χ² tests where appropriate. We compared SH levels and common and internal cIMT between ethnic groups using Kruskal-Wallis tests. The prevalence of CAC and the distributions of CAC scores were compared across groups using χ² tests and Kruskal-Wallis tests. Log transformation was employed for cIMT and SH measurements that had highly right-skewed distributions. CAC was highly right skewed and many participants had zero values. CAC was analyzed firstly as a binary variable of CAC present/absent, then log-transformed non-zero CAC scores were analyzed separately to assess their association with SH levels.

Associations between log(cIMT) and log(SH) were evaluated using linear regression analyses. Model 1 included age, ethnicity and each individual SH; Model 2 = Model 1 plus BMI; Model 3=Model 2 plus hypertension, current smoking, HDL-C and LDL-C and diabetes. The associations between IMT and SH were also adjusted for insulin sensitivity using the quantitative insulin-sensitivity check index (QUICKI=1/{log(insulin) + log(glucose)}) ¹² in subjects who were not receiving insulin. Diabetic subjects on insulin (n=76) were excluded from this analysis. Additional models were assessed where all SH were considered simultaneously and are shown in the Online Supplement. When all SH were considered simultaneously, bioavailable T was not included as it was calculated from total T and SHBG.

We used similar models to determine associations between each SH and the presence (by logistic regression) and extent (by linear regression) of CAC. Interactions of ethnicity with the association between the SH and cIMT or CAC were assessed including age and all SH as covariates in the models. BMI was a continuous variable in models assessing interactions between BMI and SH, and BMI categories were used (<25 kg/m², 25-30 kg/m², 30-40 kg/m², >40 kg/m²) when considering IMT and CAC, Because a total of 16 interaction models were assessed for each ethnicity or BMI category analyses, interactions were considered significant at the Bonferroni corrected p=0.003 level. Dummy variables were used to stratify the BMI groups and the likelihood ratio test (for logistic) and partial f-test for linear regression were used to test the cross-product terms. Statistical analyses were performed using Stata version 9.1 (College Station, TX).

Results

Demographics

Participant characteristics are shown in Table 1. Chinese women had the lowest BMI, LDL-C levels and smoking prevalence. Black and Hispanic women had the highest BMI, yet black women had the lowest triglyceride and Hispanics the highest triglyceride levels. The prevalence of diabetes was lowest in White women. Hypertension was present in over 40% of women in all ethnic groups with the highest prevalence of 65.7% among black women.

Table 1.

Mean or Prevalence Percentages of Selected Characteristics of the MESA Cohort, According to Ethnic Background. 2000-2002.

Risk Factors	White	Chinese	Black	Hispanic	p-value for differences across ethnic groups
	N=581	N=272	N=603	N=491
Age (years)	67.1±9.1	65.6±9.1	65.3±9.1	64.7±9.4	<0.001
Body Mass Index	28.0±5.7	24.1±3.7	31.2±6.4	29.9±5.5	<0.001
Current smokers (%)	65 (11.2%)	4 (1.5%)	97 (16.1%)	45 (9.2%)	<0.001
Hypertension (%)	242 (41.7%)	124 (45.6%)	396 (65.7%)	244 (49.7%)	<0.001
Diabetes Mellitus (%)	39 (6.7%)	50 (18.4%)	120 (19.9%)	100 (15.9%)	<0.001
Total Cholesterol mmol/l	5.31±0.93	5.10±0.88	5.17±0.97	5.26±0.95	<0.001
Triglyceride mmol/l	1.25 [0.87,1.76]	1.42 [1.00,2.02]	0.99 [0.76,1.36]	1.49 [1.07,2.09]	<0.001
HDL-cholesterol mmol/l	1.43±0.36	1.35±0.35	1.44±0.35	1.33±0.35	<0.001
LDL-cholesterol mmol/l	3.21±0.80	2.99±0.81	3.14±0.86	3.42±0.87	0.005

Data expressed as mean ± SD, or median [interquartile range].

Sex hormones and subclinical atherosclerosis

Table 2 shows the median and interquartile ranges of SH, IMT and calcium scores by ethnic background. Levels varied significantly by ethnic group. Total T was lowest in Chinese and highest among Black women. Bioavailable T and E2 were highest among Black women, possibly related to obesity. SHBG was highest among White women and lowest among Hispanics. DHEA was highest among Chinese and lowest among White women.

Table 2.

Median [interquartile range] Sex Hormone Levels and Measures of Subclinical Atherosclerosis According to Ethnic Background, 2000-2002.

Sex Hormones	White (N=581)	Chinese (N=272)	Black (N=603)	Hispanic (N=491)	p-value for differences across ethnic groups
Total Testosterone nmol/L	0.97 [0.66, 1.39]	0.80 [0.56, 1.20]	1.04 [0.69, 1.49]	0.94 [0.54, 1.32]	<0.001
Bioavailable Testosterone nmol/L	0.24 [0.14, 0.38]	0.24 [0.14, 0.38]	0.28 [0.17, 0.42]	0.24 [0.17, 0.42]	<0.003
Estradiol nmol/L	0.05 [0.04, 0.08]	0.05 [0.04, 0.07]	0.07 [0.05, 0.09]	0.06 [0.04, 0.08]	<0.001
SHBG nmol/L	55.0 [39.9, 75.0]	48.5 [34.5, 68.4]	50.4 [37.7, 68.5]	46.2 [33.4, 62.4]	<0.001
DHEA nmol/L	10.3 [6.3, 14.7]	12.1 [8.8, 15.9]	11.0 [7.9, 15.2]	11.1 [7.5, 15.5]	<0.001
Subclinical Atherosclerosis	White	Chinese	Black	Hispanic	P-value
Common cIMT (mm)	0.86 [0.77, 0.99]	0.81 [0.71, 0.92]	0.88 [0.77, 1.00]	0.84 [0.76, 0.96]	<0.001
Internal cIMT (mm)	0.92 [0.70, 1.53]	0.68 [0.55, 0.94]	0.87 [0.67, 1.36]	0.79 [0.63, 1.16]	<0.001
CAC present	55.7% (324)	51.1% (134)	44.3% (267)	39.9% (196)	<0.001
Agatston score if CAC present	101 [27, 305]	60 [21, 163]	77 [21, 238]	55 [16, 160]	0.002

SHBG= sex hormone binding globulin, DHEA = dehydroepiandrosterone

cIMT = carotid intimal-medial thickness

CAC = coronary artery calcium

Chinese women had the lowest internal and common cIMT. CAC score > 0 was found in 44.3% of women. Both the prevalence of any calcium and the CAC score in those with detectable calcium were highest in White and lowest in Hispanic women.

Relationship between SH and cIMT

Table 3 shows the associations between log common cIMT and each log unit increase in SH levels using three models. Using Model 1 (adjusting for age and ethnicity) we found statistically significant positive associations of total T (p=0.001) and bioavailable T (p<0.001) and inverse association of SHBG (p=0.001) with common cIMT. With additional adjustment for BMI (Model 2) there was some attenuation of the associations between SHBG and total T with common cIMT, which nevertheless remained significant. Thus in Model 2, a 2.72-fold difference in T (1 log-unit) was associated with a 1.018-fold (i.e. thicker) geometric mean cIMT (p<0.001) and a 2.72-fold higher SHBG with a 0.979-fold (i.e. thinner) geometric mean common cIMT (p=0.014) as compared to lower levels of those hormones. Addition of the covariates hypertension, lipids, diabetes, insulin sensitivity and smoking (Model 3) did not alter the significant relationship of T with common cIMT. However, further adjustment for LDL-C and HDL-C, resulted in loss of significance for the association of common cIMT with SHBG.

Table 3.

Beta-coefficients and 95% CI for the association between maximal common cIMT, internal cIMT and sex hormone levels

Common Carotid Intimal-Medial Thickness (Units of beta are log-unit thicker IMT/per 1 log unit greater sex hormone level)
	Model 1 (adjusted for age and ethnicity) N=1947			Model 2 (Model 1 and adjustment for BMI) N=1947			Model 3 (Model 2 and adjustment for hypertension, diabetes, smoking, QUICKI, excluding insulin use) N=1871
	β -coeff	95% CI	p-value	P-coeff	95% CI	p-value	β -coeff	95% CI	p-value
Ln Testosterone	0.021	0.008 to 0.034	0.001	0.018	0.005 to 0.037	0.006	0.018	0.005 to 0.031	0.009
Ln Bioavailable Testosterone	0.027	0.015 to 0.038	<0.001	0.022	0.013 to 0.034	<0.001	0.019	0.007 to 0.032	0.002
Ln Estradiol	0.005	-0.008 to 0.018	0.446	-0.001	-0.14 to 0.012	0.848	-0.006	-0.019 to 0.007	0.389
Ln DHEA	0.003	-0.014 to 0.019	0.761	0.001	-0.15 to 0.017	0.913	-0.007	-0.024 to 0.010	0.398
Ln SHBG	-0.030	-0.048 to -0.12	0.001	-0.021	-0.040 to -0.003	0.026	-0.010	-0.030 to 0.011	0.342
Internal Carotid Intimal-Medial Thickness (Units of beta are log-unit thicker IMT/per 1 log unit greater sex hormone level)
	Model 1 (adjusted for age and ethnicity) N=1914			Model 2 (Model 1 and adjustment for BMI) N=1914			Model 3 (Model 2 and adjustment for hypertension, diabetes, smoking, QUICKI, excluding insulin use) N=1871
	β -coeff	95% CI	p-value	β -coeff	95% CI	p-value	β -coeff	95% CI	p-value
Ln Testosterone	0.041	0.009 to 0.073	0.013	0.038	0.006 to 0.070	0.020	0.038	0.005 to 0.071	0.022
Ln Bioavailable Testosterone	0.028	-0.000 to 0.056	0.053	0.024	-0.005 to 0.053	0.107	0.018	-0.013 to 0.048	0.255
Ln Estradiol	-0.003	-0.035 to 0.029	0.847	-0.009	-0.042 to 0.024	0.582	-0.019	-0.052 to 0.014	0.268
Ln DHEA	-0.025	-0.066 to 0.015	0.222	-0.027	-0.068 to 0.14	0.197	-0.043	-0.085 to 0.000	0.049
Ln SHBG	-0.006	-0.050 to 0.039	0.795	0.004	-0.043 to 0.051	0.859	0.031	-0.020 to 0.082	0.240

The relationship of internal cIMT with T was qualitatively similar to that between T and common cIMT, however, no association was observed for SHBG (Table 3). After adjusting for BMI (Model 2) a 2.72 fold total T was associated with a 1.038 fold (i.e. 0.038 log unit greater) geometric mean internal cIMT. Neither E2 nor DHEA was associated with common or internal cIMT. A negative association was found between DHEA and internal cIMT when the analysis was adjusted for hypertension, diabetes, smoking, cholesterol and insulin sensitivity.

All associations were adjusted for age and ethnicity and did not differ significantly between BMI categories at the specified Bonferroni level. No significant heterogeneity by ethnicity was found for these associations. When all SH were considered simultaneously and adjusted for BMI, T was more strongly associated with internal CIMT (β-coefficient of 0.046, 95% CI of 0.013-0.080, p=0.006). All other analyses showed no qualitative difference when all SH were considered simultaneously. These analyses are shown in the Online Supplement.

To provide a sense of the size of this effect, we compared the calculated difference in common carotid IMT for two women of same age and BMI who had one log difference in testosterone level. The difference would be 0.019 mm in carotid cIMT. The difference in cIMT is statistically significant though relatively small. However, in longitudinal studies a difference of 0.02 mm of carotid IMT at baseline has been associated with an increase in relative risk for myocardial infarction and stroke of 2.4% and 2.8% respectively over a median follow-up of 6.2 yrs after adjustment for age, sex and other risk factors. ¹³

Relationship between SH and CAC

Endogenous SH levels were not significantly associated at the specified Bonferroni level with the presence of CAC, after considering age, race/ethnicity and BMI. In addition, there was no heterogeneity by race/ethnicity. The data is shown on the Online Supplement.

Among women with detectable CAC, total T, E2, and DHEA were not associated with extent of CAC. However bioavailable T was negatively and SHBG positively associated with CAC extent (Table 4). When SH were considered simultaneously, the positive association of SHBG with extent of CAC remained (data in Online Supplement). After full adjustment and considering all SH simultaneously, a 2.72 fold (1 log-unit greater) SHBG was associated with a 1.301 fold geometric mean CAC score. This positive association between SHBG and extent of CAC remained when data were stratified by decade of age (data not shown) reducing the likelihood that it represented survivor bias.

Table 4.

Beta-coefficients and 95% CI for the association between extent of coronary artery calcium (CAC) and sex hormone levels

Extent of CAC (Units of beta are log-unit difference in CAC if present per log-unit greater sex hormone level)
	Model 1 (adjusted for age and ethnicity) N=926			Model 2 (Model 1 and adjustment for BMI) N=926			Model 3 (Model 2 and adjustment for hypertension, diabetes, smoking, QUICKI, excluding insulin use) N=877
	β -coeff	95% CI	p-value	β -coeff	95% CI	p-value	β -coeff	95% CI	p-value
Ln Testosterone	-0.098	-0.264 to 0.068	0.249	-0.128	-0.295 to -0.038	0.131	-0.098	-0.271 to 0.075	0.265
Ln Bioavailable Testosterone	-0.141	-0.291 to 0.010	0.066	-0.214	-0.368 to -0.059	0.007	-0.182	-0.345 to -0.018	0.030
Ln Estradiol	0.015	-0.155 to 0.185	0.864	-0.043	-0.218 to 0.131	0.624	-0.037	-0.214 to 0.140	0.683
Ln DHEA	-0.126	-0.325 to 0.073	0.215	-0.142	-0.341 to 0.057	0.161	-0.122	-0.335 to 0.090	0.258
Ln SHBG	0.176	-.059 to 0.410	0.142	0.318	0.071 to 0.565	0.012	0.298	0.024 to 0.571	0.033

Discussion

In this cross-sectional study in postmenopausal women common cIMT was positively associated with T independent of age, BMI, or the risk factors of hypertension, smoking, HDL, LDL and insulin sensitivity. Common cIMT was negatively associated with SHBG when adjusted for age, and BMI. This association was no longer seen after adjustment for HDL and LDL cholesterol suggesting these may be involved in the pathway of association between SHBG and carotid IMT. No association was found for E2 or DHEA. T was also positively associated with increased internal cIMT, while DHEA had a negative association. The ethnic groups differed significantly with regard to body weight. We therefore evaluated the association between SH and cIMT for an interaction between sex hormones and BMI and found none.

CAC was present in 44.3% of the women, with the highest prevalence of 55.7% in white women and lowest (39.9%) in Hispanic women. Although SH levels were not associated with the presence of CAC, among women in whom calcification was present SHBG was positively associated, and bioavailable T negatively associated with extent of CAC. Ethnicity did not modify these associations.

In women, a more androgenic state is associated with an atherogenic lipid profile, higher insulin, glucose, hemostatic factors and inflammatory markers⁷, and increased odds of metabolic syndrome.¹⁴ However, our analyses support an association between testosterone and increased common cIMT that is independent of lipids, BMI, and insulin sensitivity. This would be consistent with direct effects of sex hormones on the vasculature. Hormone receptors are expressed in arterial wall, macrophages, megakaryocytes and platelets.^15-17 Testosterone may increase macrophage acetylated LDL uptake¹⁵ and increase adhesion of monocytes to vascular endothelium.¹⁸ Whether these effects occur in-vivo is unclear.

The relative levels of androgens and estrogens change with menopause.¹⁹ After menopause, ovarian E2 production decreases resulting in very low serum levels. Before menopause, T is produced primarily in the ovary and adrenal cortex. However after menopause, T continues to be produced by ovarian stromal tissue and by peripheral conversion of A4 and DHEA ²⁰ thus T levels change to a lesser degree than E2. The relative androgen/estrogen ratio becomes much more androgenic after menopause and is postulated to be more important than absolute levels in conferring CVD risk/protection.²¹ Our study found that serum T level were associated with cIMT independent of plasma E2 levels. It remains possible that local tissue levels of E2 resulting from peripheral conversion of androgen precursors to estrogens, which may occur in vascular endothelial cells and smooth muscle cells,²², affect the development of subclinical atherosclerosis.

Some studies have reported an association between increased bioavailable T and CHD.¹ A nested case-control study from the Women's Health Study found that cases had lower SHBG and higher FAI levels than controls. ⁵ However, this difference was not independent of BMI and other risk factors. A prospective study of 651 postmenopausal women found no relationship between a baseline T level and the development of cardiovascular disease or CHD death over 19 year follow-up.² The cohort studied was different from the MESA cohort being largely Caucasian, leaner, with fewer diabetics and a higher percentage of current smokers. In addition T levels were approximately ten-fold higher than the levels measured in MESA and reported by several other cohorts.²³

Others report an inverse relationship between T and cIMT in a younger cohort.⁴ Golden et al.³ used a case-control design defining cases as having an average IMT ≥ 95^th percentile for all cIMT measurements³ while controls were defined as having <75^th percentile IMT. They also found that SHBG was negatively associated with cIMT but, in contrast to our findings, the top quartile of T had a significantly lower odds ratio for high cIMT. They did not find a significant negative association between the top quartile of total T/SHBG ratio and IMT. Thus, it is possible that the negative association of total T and IMT was driven by SHBG effects more than by T.

In our study the negative association between SHBG and cIMT was no longer significant when adjusted for HDL. Other investigators have reported that SHBG is positively associated with HDL, and negatively associated with BMI, insulin resistance and C-reactive protein ⁷.

We found no association between E2 and cIMT. We did not measure E1, however others have found no association between E1 and carotid IMT.^{3, 4} Although animal studies suggest DHEA has anti-atherosclerotic effects²⁴, we and others^3,25 found no association between DHEA levels and cIMT.

In this study SH was not associated with the presence of CAC, yet among women in whom CAC was detected higher CAC scores were associated with higher SHBG and lower bioavailable T levels. Carotid IMT and plaque are associated with and predict future CAC^{26, 27} and are associated with similar risk factors, so the divergent direction of the associations found between T and SHBG and these markers of subclinical atherosclerosis is surprising. Vascular calcification appears potentially regulated by a number of ion transporters, matrix molecules and signaling pathways. Osteoprotegerin, a signaling molecule involved in osteoblastic and osteoclastic activity in bone remodeling, has been implicated in vascular calcification²⁸ and found to be upregulated in calcified coronary plaques.²⁹ An inverse relationship between serum levels of osteoprotogerin and testosterone has been reported in men³⁰, though not in women.³¹ However, ovariectomy, which would decrease ovarian T and E2 production, has been associated with an increase in vascular osteoprotegerin/RANKL (receptor activator of nuclear fact or-kappa beta ligand) ratio in animal models.³² Inflammation is also considered a prerequisite for vascular calcification. Testosterone may decrease tumor necrosis factor-alpha induced inflammatory response in human endothelial cells. ³³ Recently studies suggest that tissue transglutaminases, cross-linking enzymes, are required for inducing vascular calcification.³⁴ We could find no reports of the effects of sex hormones on vascular tranglutaminases, but it is intriguing that amphibian hepatic tranglutaminase activity decreases with testosterone exposure³⁵ and sex hormones modulate renal expression of apoptotic regulatory proteins that include tissue transglutaminase. ³⁶ It is therefore possible that the pathways by which SH affects the development of coronary calcification differ from those involved in the development of early atherosclerosis.

The strengths of our study include the large sample size of nearly 2000 postmenopausal ethnically diverse women. The limitations of the study include its cross-sectional nature, which prevent assessing temporality. The continuing follow-up of the MESA cohort will allow assessment of the association with subclinical atherosclerosis progression in the future.

Common cIMT is positively associated with serum T levels and inversely associated with SHBG in a multiethnic cohort of postmenopausal women. The inverse association with SHBG appears partially explained by total and HDL-cholesterol. Although the size of the associations between SH and carotid IMT are relatively small, they may be a signal of the effect of exogenous sex hormones on atherosclerosis. A recent analysis among women randomized to 17-β estradiol or placebo showed that the most beneficial hormone profile for reducing cIMT progression were an increase in free estradiol and SHBG with decreased bioavailable T.³⁷ Many established risk factors, including lipids, are important contributors to the development of subclinical atherosclerosis.³⁸ Our analyses show that T is also independently associated with cIMT.

Endogenous SH was not associated with the presence of CAC, while serum T was associated with lower severity of CAC in women with some CAC present. The mechanisms by which endogenous T affects the vascular wall need further study. Longitudinal analyses will confirm whether endogenous SH levels are related to progression of atherosclerosis.