Dietary β-Tocopherol and Linoleic Acid, Serum Insulin, and Waist Circumference Predict Circulating Sex Hormone-Binding Globulin in Premenopausal Women

Abstract

Reduced levels of circulating sex hormone-binding globulin (SHBG) are implicated in the etiology of sex steroid-related pathologies and the metabolic syndrome. Dietary correlates of serum SHBG remain unclear and were studied in a convenient cross-sectional sample of healthy 30- to 40-y-old women (n = 255). By univariate analyses, serum SHBG correlated negatively with several indices of the metabolic syndrome, such as BMI, waist circumference, hip circumference (r = −0.36 to −0.44; P < 0.0001), fasting serum insulin (r = −0.41; P < 0.0001), serum triglycerides (r = −0.27; P < 0.0001), serum glucose (r = −0.23; P < 0.001), and plasma testosterone (r = −0.19; P = 0.002). Serum SHBG correlated positively with serum HDL-cholesterol (r = 0.33; P < 0.0001), plasma progesterone (r = 0.17; P = 0.007), and dietary intake of β-tocopherol (r = 0.17; P = 0.006), and negatively with that of fructose (r = −0.13; P = 0.04). Principal component analysis (PCA) extracted 12 nutrient factors with eigenvalues > 1.0 from 54 nutrients and vitamins in food records. Multivariate regression analyses showed that the PCA-extracted nutrient factor most heavily loaded with β-tocopherol and linoleic acid (P = 0.03) was an independent positive predictor of serum SHBG. When individual nutrients were the predictor variables, β-tocopherol (P = 0.002), but not other tocopherols or fatty acids (including linoleic acid), was an independent positive predictor of serum SHBG. Circulating insulin (P = 0.02) and waist circumference (P = 0.002), but not serum lipids, were negative independent predictors of SHBG in all regression models. Additional studies are needed in women of other age groups and men to determine whether consumption of foods rich in β-tocopherol and/or linoleic acid may increase serum SHBG concentrations and may thereby decrease the risk for metabolic syndrome and reproductive organ cancer.

Introduction

Sex hormone-binding globulin (SHBG)⁷, a glycoprotein synthesized in the liver, binds and transports sex steroids in plasma and regulates the bioavailability of sex steroids to target cells (1). Receptors for SHBG (SHBG-R) are present on cell membranes in sex steroid-responsive organs such as the prostate and the breast (2). The SHBG/SHBG-R complex induces cell signaling through the cAMP and protein kinase A pathways, amplifying the action of androgens and antagonizing the action of estradiol (3). The estradiol-induced proliferation rate is lower in breast cancer cells positive for SHBG-R than those negative for this receptor (4). Thus, in addition to being a carrier for sex steroids, SHBG, by interacting with SHBG-R, may also enhance androgen responsiveness and reduce the biological function of estrogens in breast neoplasia and other diseases.

Recent studies indicate that lower concentrations of serum SHBG are associated with increased risk for metabolic syndrome (5,6), a clustering of conditions linked to an increased risk of cardiovascular disease and type 2 diabetes. Features of this syndrome include abdominal obesity, dyslipidemia, insulin resistance, and hyperinsulinemia (7). Obesity and higher insulin concentrations are associated with lower serum SHBG (8,9). Diet has been implicated in the development of metabolic syndrome (10) and intake of fatty acids may influence obesity and insulin sensitivity (11). However, whether diet affects circulating SHBG concentrations is unclear. A low-fat/high-fiber diet was reported to have no effect on serum SHBG in premenopausal women (12,13), but this type of diet, when accompanied by exercise and weight loss, did increase SHBG concentrations in postmenopausal women (14). In a population-based study of postmenopausal women, measures of fat intake or dietary fiber were not associated with serum SHBG (15). Differences in the menopausal status and physical activity of the study participants and in the methods used for assessing consumption of total and specific fats may have contributed to the inconsistency of results in these studies. In this study, we assessed associations between serum SHBG and intakes of a variety of nutrients, including sugars, fats, and vitamins.

Many of these nutrients were correlated with each other, especially those from the same food sources. Therefore, in addition to the classical approach of using individual nutrients as predictor variables, we studied nutrient intakes by principal component analysis (PCA). PCA is a data reduction procedure that clusters a large number of correlated nutrients into a smaller number of uncorrelated principal components or factors (artificial variables) that will account for most of the variance observed in the original variables. This method offers certain advantages over studying individual nutrients, because food is consumed as a combination of nutrients and not individual nutrients. Moreover, patterns of food consumption vary by culture, region, and ethnicity. Thus, the use of dietary patterns has been suggested in understanding the role of diet in predicting disease risk (16). PCA can also reveal synergism between multiple nutrients that may be overlooked when studying individual nutrients (17). We derived PCA factors that were completely orthogonal (uncorrelated) to each other, interpreted them as nutrient components, and investigated the relationship between these nutrient components and serum SHBG concentrations.

Materials and Methods

Study design.

Healthy premenopausal women of all major ethnicities were recruited from within an 80-km radius of Galveston, Texas, using Web-mail, posted advertisements, and postal mail. The study protocol was approved by the Institutional Review Board of the University of Texas Medical Branch and the Human Research Protection Office of the US Army Medical Research and Materiel Command. Written informed consent was obtained from each participant.

Age was restricted to 30–40 y to avoid inclusion of peri- and postmenopausal women. Pregnancy, breast-feeding, and contraceptive medications (oral, injection, or patch) during the previous 6 mo were criteria for exclusion, because these can affect SHBG levels (18,19). Participants were selected for having 1 menstrual cycle every 24–35 d to time blood collection only during the luteal phase of the cycle, when progesterone levels are increased, thereby allowing measurement of both estradiol and progesterone. Three study visits were scheduled during the luteal phase of each of 2 separate menstrual cycles (6 total visits), usually between cycle d 20 and 24. These participants volunteered for an extended dietary intervention study on biomarkers of breast cancer and their baseline data were the source for these analyses.

Anthropometrics, body composition, and reproductive factors.

Body weight and height were measured at each study visit. At 1 study visit, waist circumference was measured at the umbilicus and hip circumference at the widest point around the buttocks. Total, lean, and fat body mass were measured in duplicate (before and after repositioning) with the participant supine, using dual energy X-ray absorptiometry (Model Discovery A; Model QDR4500A, Hologic). The CV for the 2 readings were <5%. Means of 2 dual energy X-ray absorptiometry measurements were used in all statistical analyses. Reproductive history was obtained using a standard, self-administered gynecologic clinic questionnaire.

Nutrient intake.

Study participants were instructed to record food intake (food item, brand name, and amount) for the 24-h period preceding each scheduled study visit. Each participant provided at least 3 such food records, which were verified for accuracy and analyzed by a research dietitian using the Nutrition Data System for Research software, version v4.05/33 (developed by the Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN; Appendix 22, Sources of Nutrient Data, ppA22.1-A22.12). The mean nutrient intake of the 3 24-h food recall records was used in the analyses of nutritional influences on SHBG levels. A semiquantitative FFQ, version 96/97 GP (Harvard School of Public Health, Boston, MA), was also self-administered by study participants to assess dietary habits during the preceding year. These questionnaires were edited and analyzed by the Nutrition Questionnaire Service Center at the Harvard School of Public Health.

Hormone assays and blood chemistries.

Fasting venous blood samples were drawn between 0800 and 1000 during all study visits. Aliquots of serum and plasma were stored at −80°C until analyzed. Plasma samples from the first 3 visits (all from the luteal phase of 1 menstrual cycle) were analyzed for estradiol, testosterone, and progesterone using commercially available kits (Diagnostic Labs). A RIA kit was used to measure plasma progesterone concentrations (sensitivity, 0.318 nmol/L). ELISA kits were used for measuring plasma testosterone concentrations (sensitivity, 0.014 nmol/L) and plasma 17β-estradiol concentrations (sensitivity, 25.7 pmol/L). Serum samples from the first and 5th study visits were assayed for SHBG by ELISA (DSL-10–7400, sensitivity, 0.61 nmol/L) and for insulin by direct RIA (DSL-1600, sensitivity, 9.03 pmol/L) using commercially available kits (Diagnostic Labs). The intra- and interassay CV for all analytes were <15%. Means of the hormone concentrations from different study visits were used for statistical analyses.

Serum concentrations of glucose, HDL-cholesterol (HDL-C), total cholesterol, and triglycerides were measured in samples from the first and 6th study visit by a certified hospital clinical laboratory using VITROS 5.1 FS (Ortho-Clinical Diagnostics).

Nutrient component derivation.

The correlation matrix of the nutrient variables in this study, as in others, showed strong correlations between many nutrient variables. Therefore, PCA was used to identify nutrient components (factors) that independently influenced serum SHBG. From the entire database, 54 nutrients were entered into the PCA. The FACTOR procedure in SAS (SAS version 9.1, SAS Institute) with the principal axis method was used to extract the independent nutrient components. The components were rotated by orthogonal transformation using the varimax method to identify optimally noncorrelated (orthogonal) measures (20). Factor loadings were used to interpret the resulting factor pattern. Factor loadings > 0.5 were considered important, and the most powerful factors (eigenvalues > 1) were retained for further analyses (Supplemental Table 1).

Statistical analyses.

Data are presented as means and 95% CI of the mean for continuous variables and as frequencies for the categorical variables of ethnicity and parity. Pearson correlation coefficients were computed to assess the linear relationship between SHBG and the predictor variables. Outliers in the data were detected through frequency lists and scatter plots. Outliers from hormone and blood chemistry values (n < 10) were removed from statistical analyses if they were >4 SD away from the mean. The distribution of hormone and blood chemistry data was checked for normality. Log-transformed data were also used for multivariate analyses but the results were the same as untransformed data. Therefore, all results reported herein are from the original untransformed data. ANOVA and t tests were used to examine the effect of categorical variables on serum SHBG. A 2-sided α level of 0.05 was used to determine significance.

Multi-level multiple regression analysis models (forward selection) were constructed using serum SHBG as the dependent variable. Three different approaches were used to present nutrients as predictor variables. The first approach used nutrient components (factors) generated by PCA as dependent variables, whereas the second and 3rd approaches used individual nutrients from food records and FFQ, respectively, as independent predictors.

There were 4 conceptual blocks for our multi-level approach, with the first to the 4th being nutrients, blood chemistries, hormones, and anthropometrics. Multicolinearity diagnostics on final models of regression analyses revealed all variance inflation factors were ≤2, which was far below 10, a cutoff used for assessing multicolinearity. All statistical analyses were performed using the statistical software package SAS (SAS version 9.1, SAS Institute).

Results

Based on the BMI criteria of the WHO (21), 3% of our 255 study participants were underweight (BMI < 18.5), 29% were in the healthy weight range (BMI = 18.5–24.99), 35% were overweight (BMI ≥ 25), and 33% were obese (BMI ≥ 30). The study population was 51% non-Hispanic White, 30% Hispanic, 14.5% African American, 2% Asian, 2.4% unspecified ethnicity, and 1 American Indian. Additional relevant characteristics and nutrient intakes of this convenient group of study participants are summarized (Table 1).

TABLE 1.

General characteristics of 30- to 40–y-old premenopausal women from within an 80-km radius of Galveston, TX¹

Variables
Demographic and anthropometric
Age, y	36.2 (35.9, 36.6)
Height, cm	161.8 (160.9, 162.5)
Weight, kg	74.1 (72.3, 75.9)
BMI, kg/m2	28.3 (27.7, 29.0)
Waist circumference, cm	87.3 (85.9, 88.8)
Hip circumference, cm	109.1 (107.7, 110.6)
Waist:hip ratio	0.8 (0.79, 0.81)
Fat body mass,2kg	28.1 (26.9, 29.4)
Lean body mass,2kg	46.3 (45.5, 47.0)
Reproductive
Parity
Yes, n (%)	221 (87)
No, n (%)	34 (13)
Age of menarche, y	12.6 (12.5, 12.8)
Complete pregnancies, n	2.2 (2.0, 2.3)
Blood chemistry, mmol/L
Serum glucose	4.7 (4.6, 4.8)
Serum triglycerides	1.15 (1.1, 1.2)
Serum cholesterol	4.7 (4.6, 4.8)
Serum HDL-C	1.4 (1.36, 1.43)
Hormone
Serum SHBG, nmol/L	101.1 (96.2, 106.1)
Serum insulin, pmol/L	80.6 (74.3, 87.5)
Plasma testosterone, pmol/L	26.2 (24.7, 27.7)
Plasma progesterone, nmol/L	33.9 (31.9, 35.9)
Plasma 17β-estradiol,3pmol/L	292 (274, 310)
Daily nutrient intake
Energy intake, kJ	7284 (7004, 7564)
Total fats, g	75.4 (71.8, 79.0)
Total proteins, g	67.5 (64.7, 70.3)
Total carbohydrates, g	198.2 (189.4, 207.0)
Total sugars, g	86.3 (81.1, 91.6)
Fructose, g	19.9 (18.1, 21.7)
Linoleic acid, g	13.2 (12.4, 14.0)
Total vitamin E, mg	9.2 (8.5, 9.8)
α-Tocopherol, mg	9.5 (7.0, 12.1)
β-Tocopherol, mg	0.28 (0.26, 0.30)
γ-Tocopherol, mg	16.2 (15.2, 17.2)
δ-Tocopherol, mg	2.5 (2.4, 2.7)

¹Values are means (95% CI), n = 255 or n (%).

²n = 241.

³n = 216.

Serum SHBG concentrations were not associated with age or age of menarche (Table 2). Serum SHBG correlated negatively with anthropometric variables, including weight, BMI, fat body mass, lean body mass, waist circumference, hip circumference, and waist:hip ratio (all P < 0.0001) but not with height. Serum SHBG also correlated negatively with fasting serum insulin (P < 0.0001), plasma testosterone (P = 0.002), serum triglycerides (P < 0.0001), and serum glucose (P = 0.0002). Serum SHBG correlated positively with plasma progesterone (P < 0.01) and serum HDL-C (P < 0.0001) but had no association with luteal phase plasma estradiol. Ethnicity, parity, and number of pregnancies were not associated with serum SHBG in this study sample (results not shown). Intakes of β-tocopherol and (n-6) linoleic acid had weak inverse correlations with serum triglycerides (P = 0.02 and 0.09, respectively) and were not correlated with serum HDL-C (P = 0.99 and 0.70, respectively; results not shown).

TABLE 2.

Correlations between serum SHBG and several independent variables in 30- to 40–y-old premenopausal women¹

Variables	SHBG
Demographic and anthropometric	r	P-value
Age	0.08	0.23
Age at menarche	0.02	0.73
Height	0.07	0.26
Weight	−0.37	<0.0001
BMI	−0.41	<0.0001
Waist circumference	−0.44	<0.0001
Hip circumference	−0.36	<0.0001
Waist:hip ratio	−0.26	<0.0001
Fat body mass2	−0.38	<0.0001
Lean body mass2	−0.27	<0.0001
Blood chemistry/hormone
Serum glucose	−0.23	0.0002
Serum insulin	−0.41	<0.0001
Serum triglycerides	−0.27	<0.0001
Serum HDL-C	0.33	<0.0001
Plasma testosterone	−0.19	0.002
Plasma progesterone	0.17	0.007
Plasma 17β-estradiol3	−0.03	0.65
Daily nutrient intake
Energy intake	−0.03	0.66
Total fats	0.01	0.84
Total proteins	−0.06	0.33
Total carbohydrates	−0.06	0.36
Total sugars	−0.10	0.10
Fructose	−0.13	0.04
Linoleic acid	0.07	0.27
α-Tocopherol	−0.02	0.80
β-Tocopherol	0.17	0.006
γ-Tocopherol	0.07	0.28
δ-Tocopherol	−0.002	0.98

¹n = 255.

²n = 241.

³n = 216.

Among the macronutrients analyzed, only fructose (P = 0.04) had a significant negative correlation with SHBG (Table 2). However, all nutrients with univariate association with serum SHBG (P ≤ 0.1) were included in the regression models. The variable “total sugars” represents the sum of 6 monosaccharides and disaccharides (glucose, fructose, galactose, sucrose, lactose, and maltose). β-Tocopherol intake was positively correlated with serum SHBG (P = 0.006). Dietary intake of α-tocopherol was not associated with serum SHBG (P = 0.8; Table 2) and, after excluding 5 outliers (7–12 SD from the mean), had only a marginal positive association with serum SHBG (r = 0.11; P = 0.09). Dietary γ-tocopherol had no association with serum SHBG (P = 0.28; Table 2); however, after excluding 2 outliers (>5 SD from the mean), this nutrient correlated positively with serum SHBG (r = 0.13; P = 0.04).

PCA resulted in 12 nutrient factors with eigenvalue > 1 that explained 84% of the total variance in the dietary intakes of our participants (Supplemental Table 1). Briefly, factor 1, the most powerful component (eigenvalue = 17.06), was comprised mainly of members of the vitamin B family and a partial loading of α-tocopherol and vegetable protein and was called the vegetable protein and vitamin B factor. Factor 2 consisted of monounsaturated fatty acids 20:1 and 22:1, vitamin D, and all PUFA, except (n-6) PUFA 18:2, and was called the PUFA factor. Factor 3 was loaded with all the SFA and was called the SFA factor. Factor 4 was the trans-fat factor. Factor 5 was the animal protein and cholesterol factor. Factor 6 was the vitamin A factor. Factor 7 was the vitamin K and lutein+zeaxanthine factor. Factor 8 was the retinol factor. Factor 9 had the highest loadings of (n-6) PUFA 18:2 (linoleic acid) and β-tocopherol and a partial loading of γ-tocopherol and was called the linoleic acid and β-tocopherol factor. Factor 10 was loaded exclusively with fructose and glucose and was named the monosaccharide factor. Factors 11 and 12 were the lycopene factor and the vitamin C and β-cryptoxanthine factor, respectively.

Nutrient factors extracted by PCA as predictors of SHBG levels.

Multilevel multivariate regression analysis was used to evaluate the association between the PCA-extracted nutrient factors and serum SHBG (Table 3). Model 1 included nutrient factors 1 through 12 as predictor variables in the regression model. Factor 9 (linoleic acid/β-tocopherol) and factor 10 (monosaccharide) were the only independent predictors of SHBG in model 1, accounting for 7% of the variance. Next, the nutrient model was adjusted for other variables known to be associated with metabolic syndrome, such as serum HDL-C, triglycerides, and glucose (model 2), plasma testosterone (model 3A), and serum insulin (model 3B); and finally the anthropometric variable, waist circumference. The results of the final regression model (model 4) showed that factor 9 (the linoleic acid/β-tocopherol factor) remained a significant independent positive predictor of serum SHBG. Serum insulin and waist circumference were the negative independent predictors of serum SHBG. This final model explained 31% of the variance in circulating SHBG concentrations (model 4; Table 3). Monosaccharide factor was a negative independent predictor of serum SHBG until blood chemistry variables, HDL-C, triglycerides, and glucose, entered the regression analyses (models 1–2; Table 3). Serum triglycerides and glucose were independent predictors (models 2 and 3A) until serum insulin (model 3B) entered the regression analysis model. Serum HDL-C and testosterone were independent predictors of serum SHBG (models 2 and 3A and B) until waist circumference entered the regression model as a predictor variable.

TABLE 3.

Multi-level multiple regression models to predict serum SHBG concentration in 30- to 40–y-old premenopausal women

	Models1
	P-values (standardized estimates, β)
Variables	Model 1	Model 2	Model 3A	Model 3B	Model 4
Nutrient intake factor
Factor 1: Vegetable protein/vitamin B factor	0.38 (0.05)	0.98 (0.001)	0.87 (−0.01)	0.68 (−0.02)	0.44 (−0.04)
Factor 2: PUFA factor	0.18 (−0.08)	0.18 (−0.08)	0.21 (−0.07)	0.11 (−0.09)	0.12 (−0.09)
Factor 3: SFA factor	0.09 (−0.11)	0.03 (−0.12)	0.03(−0.12)	0.02 (−0.13)	0.06 (−0.11)
Factor 4: Trans fat factor	0.55 (0.03)	0.73 (0.02)	0.83 (0.01)	0.56 (0.03)	0.43 (0.04)
Factor 5: Animal protein/cholesterol factor	0.41 (−0.05)	0.53 (−0.04)	0.63 (−0.02)	0.97 (−0.002)	0.48 (0.04)
Factor 6: Vitamin A/carotene factor	0.49 (−0.05)	0.16 (−0.08)	0.20 (−0.07)	0.13 (−0.09)	0.07 (−0.10)
Factor 7: Vitamin K, lutein+zeaxanthine factor	0.55 (−0.04)	0.25 (−0.07)	0.37 (−0.05)	0.66 (−0.02)	0.91 (−0.006)
Factor 8: Retinol factor	0.71 (−0.02)	0.87 (−0.009)	0.88 (−0.01)	0.83 (0.01)	0.75 (−0.02)
Factor 9: Linoleic acid/β-tocopherol factor	0.007 (0.17)	0.03 (0.13)	0.02 (0.13)	0.04 (0.12)	0.03 (0.12)
Factor 10: Monosaccharide factor	0.05 (−0.12)	0.31 (−0.06)	0.22 (−0.07)	0.17 (−0.08)	0.18 (−0.07)
Factor 11: Lycopene	0.59 (−0.02)	0.66 (−0.03)	0.74 (−0.02)	0.77 (−0.02)	0.78 (−0.02)
Factor 12: Vitamin C	0.75 (0.02)	0.74 (−0.02)	0.76 (−0.02)	0.76 (−0.02)	0.66 (−0.02)
Serum concentration
HDL-C, mmol/L	NE2	0.0002 (0.25)	0.0003 (0.24)	0.007 (0.18)	0.16 (0.1)
Triglycerides, mmol/L	NE	0.05 (−0.13)	0.06 (−0.12)	0.58 (−0.04)	0.45 (−0.05)
Glucose, mmol/L	NE	0.03 (−0.14)	0.05 (−0.13)	0.39 (−0.06)	0.56 (−0.04)
Testosterone, pmol/L	NE	NE	0.03 (−0.13)	0.03 (−0.13)	0.08 (−0.10)
Insulin, pmol/L	NE	NE	NE	0.0001 (−0.28)	0.02 (−0.18)
Waist circumference, cm	NE	NE	NE	NE	0.003 (−0.24)
Model R2	0.07	0.21	0.23	0.28	0.31

¹Model 1, The 12 nutrient PCA factors were entered; model 2 (blood chemistry), variables of model 1 plus serum levels of HDL-C, triglycerides, and glucose; model 3A (steroid hormones), variables of model 2 plus serum testosterone; model 3B (insulin), variables of model 3A plus serum insulin; model 4 (anthropometrics), variables of model 3B plus waist circumference.

²NE, Not entered.

Individual nutrients from food recall records as predictors of SHBG levels.

To define the individual nutrient predictors of SHBG, forward selection multilevel multivariate analyses were carried out using nutrients from 3 24-h food records that exhibited univariate association with serum SHBG (P ≤ 0.1; Table 2). Dietary intakes of total sugars and β-tocopherol (but not other tocopherols) were the independent predictors in model 1, accounting for 4.0% of the variance in serum SHBG (Table 4). Sugar intake and β-tocopherol remained independent predictors after sequential adjustment for additional predictor variables known to be risk factors for metabolic syndrome (models 2–4), as described below. First, the nutrient model was adjusted for serum HDL-C, triglycerides, and glucose (model 2), then for plasma testosterone (model 3A) and serum insulin (model 3B), and finally for the waist circumference (model 4). Total sugar intake, serum insulin, and waist circumference were negative independent predictors of serum SHBG, whereas serum HDL-C and β-tocopherol intake were positive independent predictors of serum SHBG (model 4; Table 4). Model 4 explained 28% of the variance in serum SHBG. Plasma estradiol data were available for 216 participants but was not available for 39 participants, because the manufacturer stopped the production of kits used for plasma estradiol analysis. Luteal phase plasma estradiol was not associated with serum SHBG by univariate (Table 2) or multivariate analyses (results not shown).

TABLE 4.

Multi-level multiple regression models of individual nutrients from food records to predict serum SHBG concentration in 30- to 40–y-old premenopausal women

	Models1
	P-values (standardized estimates, β)
Variables	Model 1	Model 2	Model 3A	Model 3B	Model 4	Energy intake model
Sugar intake, g/d	0.01 (−0.16)	0.05 (−0.12)	0.05 (−0.12)	0.03 (−0.13)	0.04 (−0.12)	0.055 (−0.12)2
β-Tocopherol, mg/d	0.001 (0.22)	0.003 (0.18)	0.005 (0.17)	0.003 (0.18)	0.004 (0.17)	0.003 (0.19)
HDL-C, mmol/L	NE3	<0.0001 (0.25)	0.0001 (0.24)	0.01 (0.17)	0.1 (0.11)	0.04 (0.13)
Triglycerides, mmol/L	NE	0.01 (−0.16)	0.03 (−0.14)	0.72 (−0.02)	0.63 (−0.03)	0.61 (−0.03)
Testosterone, pmolL	NE	NE	0.007 (−0.16)	0.01 (−0.16)	0.03 (−0.12)	0.05 (−0.11)
Insulin, pmol/L	NE	NE	NE	<0.0001 (−0.31)	0.005 (−0.2)	0.007 (−0.20)
Waist circumference, cm	NE	NE	NE	NE	0.004 (−0.21)	0.005 (−0.21)
Model R2	0.05	0.17	0.19	0.26	0.28	0.31

¹Model 1 (nutrients): nutrient variables included total sugars, fructose, glucose, animal protein, cholesterol, α-tocopherol, γ-tocopherol, and β-tocopherol. Independent predictors in model 1 was adjusted for serum levels of glucose, HDL-C, and triglycerides (model 2: blood chemistry), then serum progesterone and testosterone (model 3A: steroid hormones), serum insulin (model 3B: insulin), and finally anthropometrics (model 4: anthropometrics). Energy intake model: total sugar intake was replaced by energy intake in model 4.

²Energy intake as a variable.

³NE, Not entered.

The Pearson correlation coefficients between β-tocopherol and γ-tocopherol (r = 0.62), between β-tocopherol and (n-6) linoleic acid (r = 0.46), and between γ-tocopherol and (n-6) linoleic acid (r = 0.73) were all very high. The nutrient factor most heavily loaded with β-tocopherol and (n-6) linoleic acid also had a partial loading of γ-tocopherol. In 2 separate exploratory models that omitted β-tocopherol as an independent predictor, energy-adjusted intakes of γ-tocopherol (P < 0.01) and linoleic acid (P < 0.07) were predictors of SHBG.

Individual nutrients from the FFQ as predictors of SHBG levels.

To further confirm our results based upon the 3 24-h food records (Tables 3–4), we determined nutrient intakes from the FFQ, which assessed patterns of consumption over 1 y as predictor variables for circulating SHBG. Consistent with the approach used for food records, intakes of nutrients exhibiting univariate association with serum SHBG at P ≤ 0.1 were entered into the first model of the forward selection, multi-level multiple regression analyses (model 1; Table 5). Energy intake and linoleic acid emerged as independent predictors and explained 4% of the variance in serum SHBG. The nutrient model was further adjusted for other variables known to be risk factors for metabolic syndrome in model 2 through model 4, with conceptual approaches similar to those described in Tables 3–4. The final model of the FFQ regression analyses showed that linoleic acid was a positive independent predictor of serum SHBG, whereas energy intake, serum insulin, and waist circumference were negative independent predictors of serum SHBG and explained 28% of the variance. The version of the software used for FFQ analyses did not provide estimation for β-tocopherol intake; therefore, its association with serum SHBG could not be analyzed using FFQ data.

TABLE 5.

Individual nutrients from the FFQ in forward selection multi-level multiple regression models to predict serum SHBG concentration in 30- to 40–y-old premenopausal women

	Models1
	P-values (standardized estimates, β)
Variables	Model 1	Model 2	Model 3A	Model 3B	Model 4
Energy intake, kJ	0.02 (−0.29)	0.04 (−0.24)	0.04 (−0.25)	0.03 (−0.25)	0.05 (−0.22)
PUFA 18:2 or linoleic acid, mg/d	0.003 (0.37)	0.008 (0.31)	0.007 (0.32)	0.006 (0.31)	0.02 (0.27)
HDL-C, mg/d	NE2	0.0004 (0.25)	0.0005 (0.24)	0.007 (0.18)	0.11 (0.11)
Triglycerides, mmol/L	NE	0.006 (−0.19)	0.02 (−0.16)	0.48 (−0.05)	0.43 (−0.06)
Testosterone, nmol/L	NE	NE	0.008 (−0.17)	0.02 (−0.15)	0.04 (−0.13)
Insulin, pmol/L	NE	NE	NE	<0.0001 (−0.29)	0.02 (−0.19)
Waist circumference, cm	NE	NE	NE	NE	0.009 (−0.21)
Model R2	0.04	0.17	0.19	0.25	0.28

¹Model 1 (nutrients), Eicosapentaenoic acid (EPA) and linoleic acid as predictor variables; model 2 (blood chemistry), independent predictors from model 1 plus serum levels of HDL-C, triglycerides, and glucose; model 3A (steroid hormones), variables of model 2 plus serum testosterone and progesterone; model 3B (insulin), variables of model 3A plus serum insulin; model 4 (anthropometrics), variables of model 3B plus BMI, body weight, waist circumference, and hip circumference. Note that the FFQ did not provide data on β-tocopherol intake.

²NE, Not entered.

Discussion

The primary objective of this cross-sectional study was to determine the dietary components that, along with anthropometric and metabolic factors, were associated with serum SHBG in a multi-ethnic convenient sample of premenopausal women. Of all the nutrients examined, the most significant positive independent predictors were dietary linoleic acid and β-tocopherol, whereas the consumption of sugars was a negative independent predictor of serum SHBG. The major dietary sources for linoleic acid and β- and γ-tocopherols were vegetable oils. β-Tocopherol in our participant was also consumed from popcorn, breakfast cereals, and nuts.

In a recent meta-analysis of 60 controlled intervention trials, linoleic acid was the most effective fatty acid in decreasing the risk of coronary heart disease, as assessed by improvement in the ratio of total:HDL-C (22). Lower intake of linoleic acid, as measured by lipid levels in serum, was associated with a high prevalence of metabolic syndrome in men (23). Linoleic acid consumption improved insulin resistance and was associated with a lower incidence of type 2 diabetes in the Nurses' Health Study (24). Intake of (n-6) fatty acids has doubled in the United States over the last few decades and coronary heart disease mortality has fallen by ∼50% (25). Given the close association of linoleic acid with β-tocopherol in food groups, it is surprising that little attention has been given to the effects of β-tocopherol intake on metabolic syndrome or type 2 diabetes.

The term vitamin E refers to a group of 8 distinct but related chemicals, 4 tocopherols (α-, β-, γ-, and δ-) and 4 tocotrienols (α-, β-, γ-, and δ-) (26). Most studies on vitamin E have focused on the antioxidant and biological activities of α-tocopherol (27). However, some cooking oils in the American diet contain substantial amounts of β-tocopherol (28). Nonetheless, the biological actions of β-tocopherol and other forms of vitamin E have received little attention in in vitro, animal model, or human studies. It is possible that various tocopherols have different cellular functions in the body and these are likely to be independent of their antioxidant activity. For example, the cellular uptake of β-tocopherol was 160–170% higher than the uptake of α-tocopherol in mouse splenocytes and β-tocopherol induced cell proliferation at a much smaller dose than α-tocopherol (29).

Our study demonstrates that dietary intakes of linoleic acid and β-tocopherol, as assessed from the mean of 3 24-h food records as well as FFQ, are positive independent predictors of serum SHBG in premenopausal women. These associations persisted regardless of whether individual nutrients or principal nutrient components were used in regression analyses (Tables 3–5). Although γ-tocopherol or linoleic acid were predictors of SHBG in the exploratory models run without β-tocopherol, intakes of β-tocopherol and sugars were the only independent predictors of SHBG in forward selection models and in nutrient models after mutual adjustment for all other tocopherols. β-Tocopherol was only moderately correlated with linoleic acid (r = 0.46) and γ-tocopherol (r = 0.62), which suggests that the content of these 3 nutrients is not very similar across different food sources. This might explain why β-tocopherol manifests itself as a much stronger independent predictor of SHBG relative to the other tocopherols and linoleic acid. It can be argued that β-tocopherol affected serum SHBG concentrations via hepatic lipids, but our study revealed only a weak inverse correlation between β-tocopherol and serum triglycerides (r = −0.14; P = 0.02) and no correlation with HDL-C (r = 0.001). A lack of attenuation in the association between β-tocopherol and SHBG upon adding serum lipids to the model shows that another pathway may explain this association. However, the underlying mechanism for the observed association between β-tocopherol and SHBG remains to be elucidated.

The reported difference in SHBG levels between participants with and without metabolic syndrome is 7–9 nmol/L (5). Based on the parameter estimates of our regression analyses, an intake increase of 0.1 mg/d β-tocopherol and/or of 5 g linoleic acid may confer an increase in serum SHBG of ∼5 nmol/L, which is a hypothesis testable by future intervention studies.

The analysis software of the 24-h food records estimated intakes of both β-tocopherol and linoleic acid, whereas that of the FFQ estimated intake of only linoleic acid but not of β-tocopherol. Linoleic acid and β-tocopherol are derived from similar food groups and had similar loading patterns in PCA and may be considered surrogate markers for each other. When both β-tocopherol and linoleic acid were entered into regression models simultaneously, β-tocopherol was a stronger independent predictor of serum SHBG than linoleic acid (Table 4). Antioxidants, including vitamin E, are reported to ameliorate the effects of an environmental pollutant on rat testes by increasing the production of androgen-binding protein by Sertoli cells (30). Androgen-binding protein and SHBG are closely related proteins with only a slight difference in their sugar moieties (31). An effect of β-tocopherol on SHBG may not have been studied or reported previously, because tocopherols are minor constituents of vitamin E. However, only intervention studies can determine the true effect of tocopherols and/or linoleic acid on serum SHBG.

The negative association of dietary sugars with SHBG levels in our study is supported by the experimental findings of Selva et al. (32), who demonstrated an inhibitory effect of monosaccharides on SHBG production in HepG2 cells and in mice expressing human SHBG transgenes. Monosaccharide-induced reduction in SHBG production occurred via a downregulation of hepatocyte nuclear factor-4α and a concomitant increase in cellular palmitate levels. Furthermore, Selva et al. (32) showed that the inhibitory effect of monosaccharides on SHBG synthesis by HepG2 cells could be mimicked by treating the cells with palmitoyl-CoA alone, while it was antagonized by cotreatment of cells with the lipogenesis inhibitor cerulinin. These data suggest that lipogenesis is the key regulator of SHBG synthesis, as evidenced by our study and other studies showing that risk factors of metabolic syndrome such as sugar consumption and energy intake, especially excess energy intake (for which waist circumference is an excellent proxy), are negative independent predictors of serum SHBG. Strategies to minimize excess energy intake as carbohydrates and its storage as fat may increase serum SHBG and reduce the risk for the metabolic syndrome and other diseases.

In this study, we also demonstrated that several other known risk factors for metabolic syndrome, such as serum HDL-C, serum insulin, plasma testosterone, and waist circumference, were independent predictors of serum SHBG. This is consistent with recent epidemiologic observations that low serum SHBG concentrations are associated with metabolic syndrome and cardiovascular disease (33,34). The positive correlation of serum HDL-C and the negative correlations of serum triglycerides and glucose with serum SHBG noted in our study have been observed by others (35). However, we further showed that the association of serum SHBG with serum triglycerides and glucose was not significant after adjusting for serum insulin, which may be explained by insulin's ability to regulate blood glucose and triglycerides (36).

Of the 3 sex steroids examined, only testosterone was a predictor of serum SHBG, an association independent of serum insulin and waist circumference. The negative association of testosterone with serum SHBG has been previously observed in women with polycystic ovary syndrome (37). Estradiol has been suggested to be a major regulator of SHBG (38,39). The association between circulating estradiol and SHBG has been difficult to demonstrate conclusively in all previous studies. Oral, but not trans-dermal, administration of estradiol stimulated the synthesis of SHBG (40). Serum SHBG concentrations did not vary during the follicular and luteal phases of the menstrual cycle (41), suggesting that high levels of estradiol in the liver, achievable through the first-pass effect of oral estradiol, may be needed to influence serum SHBG concentrations. Luteal phase plasma progesterone had a univariate correlation with serum SHBG but not by multivariate regression.

The strengths of this study include the selection of a well-defined group of premenopausal women with regular menstrual cycles, none of whom were taking contraceptive medications. Blood was sampled only during the luteal phase of the menstrual cycle so that associations of serum SHBG with both plasma estradiol and progesterone could be studied. Dietary intake was assessed by 3 24-h food recall records as well as FFQ. We used individual nutrients, as well as principal nutrient components that minimized the threat of colinearity, to evaluate associations between nutrients and serum SHBG concentrations. There are several weaknesses of our study, including the lack of data on physical activity of the participants, the lack of estimation of β-tocopherol in food frequency data by the version of software used, and the lack of β-tocopherol measurement in blood samples. Other common issues associated with dietary assessment, e.g. under-reporting of fat and oil intake by study participants, are well-known in nutritional epidemiology studies. Causal inference is limited in a cross-sectional study of volunteers from a small geographical area and the exclusion of postmenopausal women and men limits the extension of our results to these groups.

In summary, in this cross-sectional study, serum SHBG concentrations were negatively associated with all major factors implicated in the metabolic syndrome, including dietary sugars, serum concentrations of insulin, triglycerides, glucose, and testosterone, and waist circumference. In addition, we found that intake of β-tocopherol and/or linoleic acid was positively associated with serum SHBG. Our observations may have public health implications, because low serum SHBG concentrations are considered risk factors for metabolic syndrome and coronary heart disease. Therefore, it is important to understand the molecular mechanisms by which β-tocopherol and/or linoleic acid influence serum SHBG concentrations and to understand the biological actions of SHBG beyond its more classical role as a carrier protein for sex steroids. Current findings in 30- to 40-y-old women are from a single geographical area. Additional studies in women of other age groups and men will be needed to evaluate the validity of these associations and determine their clinical importance.