Phytoestrogen Exposure Correlation with Plasma Estradiol in Postmenopausal Women in European Prospective Investigation of Cancer and Nutrition-Norfolk May Involve Diet-Gene Interactions

Phytoestrogens are diphenolic compounds found in plants, which are thought to exert hormonal effects in the body due to their structural resemblance to the hormone 17β-estradiol. Phytoestrogens in the human diet can be divided into two main groups, the isoflavones (found in soy and other legumes) and the lignans (widely found in various grains, fruits, and vegetables). Epidemiologic evidence has linked high phytoestrogen consumption in Asian populations to lower risk of hormone-related cancers, such as breast cancer, compared with Western populations (1, 2). The mechanisms for the proposed cancer-protective effect of phytoestrogens are not fully understood but may involve modulation of the concentrations of endogenous hormones.

Circulating concentrations of estradiol are strongly and positively related to breast cancer risk in postmenopausal women (3). Sex hormone binding globulin (SHBG) binds circulating estradiol and restricts biological activity. In studies with cell lines, Adlercreutz et al. showed that lignans and isoflavones can increase both the synthesis and the secretion of SHBG in human liver cancer cells (4-6). In cross-sectional studies, they also found that urinary phytoestrogens correlated positively with plasma SHBG (4, 7). These findings led to the hypothesis that phytoestrogens may stimulate the synthesis and release of SHBG, thus reducing the proportion of free estradiol circulating in plasma and indirectly lowering breast cancer risk. Apart from their effects via SHBG, there is also some evidence that phytoestrogens may directly modulate concentrations of circulating estradiol by inhibiting enzymes involved in estrogen biosynthesis and metabolism (8), such as 17β-hydroxysteroid oxidoreductase (9-11), aromatase (12, 13), and steroid sulfatase (14).

However, intervention studies and cross-sectional studies, which investigated such hormonal effects of phytoestrogens in postmenopausal women, have produced mixed results (4, 7, 15-27). To the best of our knowledge, all of the cross-sectional studies conducted thus far used only a single marker quantifying phytoestrogen exposure in either diet or urine. Analytic methods of measuring phytoestrogens in urine and serum at low levels are difficult, and no study of serum phytoestrogens has been reported. In addition, plasma estradiol concentration is regulated by a network of several different enzymes, and polymorphisms in genes encoding for enzymes involved in estradiol metabolism may affect the concentrations of plasma estradiol. In none of the previous studies were the effects of genetic polymorphisms on phytoestrogens and sex hormone levels investigated.

This article presents the first cross-sectional study to investigate the relationship among phytoestrogen exposure, genetic variants involved in estrogen metabolism, and plasma estradiol and SHBG levels in postmenopausal women using three different markers of phytoestrogen exposure (dietary, urinary, and serum). Five single nucleotide polymorphisms (SNP) in four genes (CYP19, SHBG, COMT, and ESR1) involved in estrogen metabolism and signaling were examined in this study. SNPs in the CYP19 and SHBG genes have already been shown to be associated with estradiol and SHBG levels in postmenopausal women in European Prospective Investigation of Cancer and Nutrition (EPIC)-Norfolk (28). Aromatase, encoded by the CYP19 gene, converts and rostenedione and testosterone to estrone and estradiol and is inhibited by phytoestrogens (12, 13). Catechol-O-methyltransferase (encoded by COMT) is involved in the deactivation of estradiol via conversion of 4-hydroxyestradiol to 4-methoxyestradiol. There is potentially a relationship between 4-hydroxyestradiol exposure and breast cancer risk (28). The ESR1 gene is involved in the signaling response to estradiol and possibly also to phytoestrogens, and we noted an interaction between ESR1 PvuII polymorphisms and a reduction in breast density in our previous intervention study (29).

Subjects of this study consisted of 125 postmenopausal women from a nested case-control study on diet and breast cancer in the EPIC-Norfolk cohort (30). EPIC-Norfolk is part of the United Kingdom arm (31) of the EPIC, a large multicenter prospective study initiated in 1993 with the aim of investigating the relationship between diet and cancer (32). Men and women, ages 45 to 75 years, resident in Norfolk, United Kingdom, were recruited in 1993 to 1997 using general practice age-sex registers. At recruitment, subjects completed a health check where they filled in a health questionnaire and completed a 7-day diary of all food and drink consumed. A total of 42 mL blood (not fasting, collected in EDTA, citrated, and plain Monovettes) and a spot urine sample were taken at the health check. All subjects were healthy at the time of recruitment. Details on the study design and sample collection had been reported previously (31).

Subjects of this study were drawn from a nested case-control study on diet and breast cancer (n = 333) in the EPIC-Norfolk cohort (30). Of the 333 eligible women (ages 45-76 years), there were 185 women who were classified as postmenopausal based on self-report of not having menstrual periods for >5 years. Of these, 29 women who were using exogenous sex hormones at the time of the questionnaire or blood and urine collection and 31 women who were found to have sex hormone levels inconsistent with postmenopausal status (either follicle-stimulating hormone <30 IU/L or estradiol >100 pmol/L) were excluded, leaving a final total of 125 eligible subjects in this study. Average (range) age of subjects was 64.1 years (47.3-76.5 years). Subjects had a mean body mass index (BMI) of 27.1 kg/m² [95% confidence interval (95% CI), 26.3-27.9] and were 14.1 years after menopause on average (95% CI, 13.0-15.2).

Spot urine samples were stored at −20°C until analyzed for creatinine and phytoestrogens. Blood samples were processed into straws of plasma, serum, red cells, and buffy coats and stored at −190°C until analysis of plasma for hormones. Serum samples were available for 109 of the 125 subjects and had been stored at −40°C until analyzed for phytoestrogens (32).

Dietary data were obtained from 7-day food diaries. These diaries were given out at the health check after instruction (33) and were completed and returned by post (93% compliance). Dietary isoflavone intakes were determined using a food composition database based on daidzein and genistein concentrations measured in 300 commonly eaten foods. Details on the sampling of foods and the analysis of daidzein and genistein and their contents in different foods have been reported elsewhere (34-37). Isoflavone content of foods gathered from a literature search of published values were also incorporated into the food composition database for use in the analysis. The food composition database of isoflavones used in this study represents United Kingdom's contribution to the Vegetal Estrogens in Nutrition and the Skeleton database, a regional food composition database established to facilitate the estimation of exposure levels to phytoestrogens in four European countries—Italy, the Netherlands, Ireland, and the United Kingdom (38, 39).

Spot urine samples were analyzed for three isoflavones (daidzein, genistein, and glycitein), two metabolites of daidzein [O-desmethylangolensin (O-DMA) and equol], and two lignans (enterodiol and enterolactone). Triply ¹³C-labeled standards in methanol were added to 200 μL sample, and the phytoestrogen conjugates were enzymatically hydrolyzed to the aglycones. These were then extracted on Strata C18-E SPE cartridges (Phenomenex, Macclesfield, United Kingdom) and derivatized to trimethylsilyl derivatives for analysis using isotope dilution gas chromatography/mass spectrometry. Details and information on quality assurance have been reported elsewhere (40). Limits of detection ranged from 1.8 ng/mL (enterodiol) to 8.0 ng/mL (enterolactone). The average intra-assay coefficient of variation ranged from 1.8% (equol) to 6.5% (glycitein). The average interassay coefficient of variation for all analytes were <9%, except for O-DMA (20.2%) and glycitein (26.5%), both of which did not have a corresponding triply ¹³C-labeled standard at the time of analysis.

Urinary creatinine concentrations were measured based on a kinetic modification of the Jaffe reaction using the Roche reagent for creatinine on a Roche Cobas Mira Plus chemistry analyzer (Roche Products Ltd., Hertfordshire, United Kingdom).

Available serum samples (n = 109 of 125) were analyzed for three isoflavones (daidzein, genistein, and glycitein), two metabolites of daidzein (O-DMA and equol), and two lignans (enterodiol and enterolactone). Triply ¹³C-labeled standards in methanol were added to 200 μL sample, and the phytoestrogen conjugates were enzymatically hydrolyzed to the aglycones. These were then extracted on Strata C18-E SPE cartridges, dried under nitrogen, and redissolved in 40% methanol for analysis using isotope dilution liquid chromatography/tandem mass spectrometry. Details and information on quality assurance have been reported elsewhere (41). Limits of detection ranged from 0.04 ng/mL (daidzein) to 0.11 ng/mL (equol). The average intra-assay coefficient of variation ranged from 2.8% (enterolactone) to 5.7% (glycitein). The average interassay coefficient of variation ranged from 3.0% (genistein) to 4.4% (O-DMA).

Plasma samples were analyzed for estradiol and SHBG. The low postmenopausal levels of plasma estradiol were measured by radioimmunoassay after ether extraction (42). The sensitivity limit was 3.0 pmol/L. All estradiol analyses were conducted in duplicate. Five quality control sera were included at the beginning and end of each assay. The within-assay variability was 9.4% and between-assay variability was 12.8% at a mean level of 26 pmol/L. Plasma SHBG was measured using a liquid-phase immunoradiometric kit from Orion Diagnostica (Espoo, Finland). The sensitivity limit was 0.5 nmol/L, and the within-batch and between-batch coefficients of variation were 2.1% and 7.4%, respectively, at a concentration of 11 nmol/L.

All genotyping was carried out using end point Taqman assays (Applied Biosystems, Warrington, United Kingdom) in 384-well arrays, which included blank wells as negative controls. Assays were run on MJ Tetrad thermal cyclers (Genetics Research Instrumentation, Essex, United Kingdom), and genotypes were subsequently read on a 7900 Sequence Detector (Applied Biosystems) according to the manufacturer's instructions. An automated robotic high-throughput system in a low-volume, 384-well format was used, thereby reducing the chance of errors. The quality of each assay was tested on a specific test set of 96 DNA samples (80 unique, 14 duplicates, and 2 no template controls). The assays were found to be of good quality with clear clustering and showed 100% concordance in the duplicates. Genotype data were obtained on 95 women. The genotype distributions of the five polymorphisms analyzed were found to be in Hardy-Weinberg equilibrium.

The statistical analyses were done using SPSS software version 11.0 (SPSS Ltd., Surrey, United Kingdom). The spot nature of urinary concentrations was corrected using urinary creatinine concentration. Urinary excretion of phytoestrogens was expressed as microgram per millimole of urinary creatinine. Kruskal-Wallis test was used to compare differences in dietary, urinary, and serum phytoestrogen concentrations among subjects with different genotypes for each SNP. All dietary, urinary, and serum phytoestrogen data and plasma estradiol and SHBG data were skewed. Therefore, data were logarithmically transformed to obtain approximately normal frequency distributions, and all subsequent statistical testings were done on log-transformed data. For data sets with 0 values (representing values below limits of detections), 1 was added to the value before log transformation. ANOVA was used to compare differences in plasma estradiol and SHBG concentrations among subjects with different genotypes for each SNP. Trend tests were used to assess any linear associations between plasma estradiol and SHBG concentrations across common homozygotes, heterozygotes, and rare homozygotes for the respective SNPs. Hierarchical multiple regression and partial correlations were used to assess the degree of association between urinary, serum, and dietary phytoestrogens and plasma estradiol and SHBG, controlling for age and BMI. All P-values were two sided, and P < 0.05 was considered statistically significant.

Dietary intake, urinary excretion, and serum levels of phytoestrogens and plasma levels of estradiol and SHBG are shown in Table 1. The median dietary isoflavone intakes, urinary isoflavone concentrations, and serum isoflavone concentrations of the women in this study were several times lower than that reported in Asian populations (43, 44) and reflect the low habitual soy consumption of EPIC-Norfolk women. Despite their low dietary isoflavone intakes, serum isoflavones were still several hundred times higher than plasma estradiol levels in these postmenopausal women. Dietary, urinary, and serum phytoestrogen levels did not differ significantly between women with different genotypes for each SNP (data not shown).

Table 2 shows the frequencies of the polymorphisms for the five SNPs and the relationship between the SNPs with plasma estradiol and SHBG concentrations. Women with the CC genotype for the CYP19 3′ untranslated region SNP had significantly lower plasma estradiol compared with women with CT and TT genotypes (P_{heterogeneity} = 0.036; P_trend = 0.012), whereas women with the CC genotype for ESR1 PvuII SNP showed a trend toward higher plasma estradiol compared with those with CT or TT genotypes (P_{heterogeneity} = 0.054; P_trend = 0.015). There were no other significant differences.

Table 3 shows the relationship between dietary, urinary, and serum phytoestrogens with plasma estradiol and SHBG adjusted for age and BMI. There were significant inverse relationships (P < 0.03) between urinary concentrations of daidzein, genistein, and glycitein and plasma estradiol levels. Urinary daidzein showed the strongest correlation (R = −0.277). Similar trends were observed between estradiol with serum phytoestrogen levels, although only the correlations with serum daidzein (R = −0.227) and glycitein (R = −0.215) reached statistical significance (P < 0.03). Figure 1 shows plasma estradiol concentrations in relation to serum and urinary daidzein concentrations by tertiles. Measurements of dietary intakes of daidzein and genistein were not correlated with plasma estradiol levels. No correlation was seen among O-DMA, equol, or lignans levels measured in either urine or serum with estradiol levels. No association was found between any measurements of phytoestrogen levels with plasma SHBG levels (Table 3).

Because both the CYP19 3′ untranslated region and ESR1 PvuII SNPs were significantly associated with plasma estradiol levels in our subjects (Table 2), we further investigated whether these two SNPs had any effect on the relationship between different measurements of phytoestrogens and plasma estradiol levels. For women with the ESR1 PvuII CC genotype, all measurements (dietary, urinary, and serum) of daidzein, genistein, and glycitein were negatively correlated (P < 0.03) with plasma estradiol levels (Table 4). Partial correlation coefficients ranged from −0.545 (for dietary genistein) to −0.834 (for urinary genistein). This relationship is illustrated in a partial regression plot in Fig 2. R² values indicate that up to 56.5% of the variance in plasma estradiol levels in this group of women could be explained by urinary genistein concentration alone (Table 4). By contrast, no significant correlation between any of the phytoestrogen measures and plasma estradiol levels was observed in women with the ESR1 PvuII CT and TT genotypes (data not shown). Figure 1 shows that the inverse relationship between plasma estradiol levels across increasing tertiles of serum daidzein was significantly magnified in women with ESR1 CC genotype compared with all women but was absent in women with CT and TT genotypes. Furthermore, the negative correlations observed between estradiol and daidzein, genistein, and glycitein (measured from either urine or serum) in the women as a whole (Table 3) became no longer significant when we excluded women with ESR1 PvuII CC genotype, indicating that the negative correlations observed were due mainly to women with ESR1 PvuII CC genotype. There were no significant correlations between phytoestrogen exposure and CYP19 polymorphisms investigated.

The effect of phytoestrogens on plasma estradiol and SHBG in postmenopausal women has been the subject of investigation in many intervention and cross-sectional studies over the recent years. Four intervention studies have reported increased plasma SHBG levels with phytoestrogen supplementation in postmenopausal women (16-19). However, in one of these studies, the effect on SHBG disappeared after adjusting for body weight changes (19). In addition, at least four intervention studies failed to find any significant effect on plasma SHBG after daily supplementation of soy or other phytoestrogen-rich foods, with supplementation periods ranging from 4 weeks to 6 months (15, 20, 22, 45). As for the effect of phytoestrogen consumption on estradiol levels, only 2 of 10 intervention studies reported a significant decrease in estradiol following supplementation with phytoestrogen-rich foods, whereas others did not find any significant effect (15-24).

Two small cross-sectional studies published in 1987 (n = 50; ref. 7) and 1992 (n = 30; ref. 4) reported positive correlations between urinary phytoestrogens and plasma SHBG, with one of the studies also reporting that urinary phytoestrogens correlated negatively with plasma free estradiol. Dietary data were only available in recent larger cross-sectional studies of postmenopausal women, and these failed to show any significant correlation between phytoestrogen intake and plasma SHBG and/or estradiol (25-27).

There are several possible reasons for the conflicting results reported thus far. In intervention studies, differences in supplementation dose, form, type of phytoestrogen, supplementatisson period, or even differences in gut microflora of subjects could contribute to the disparate results. In cross-sectional studies, errors due to difficulties in accurately quantifying phytoestrogens exposure pose a challenge. We have recently developed fast, sensitive, accurate, and reliable techniques for the analysis of three isoflavones, two metabolites of daidzein, and two lignans in serum and urine (40, 41). Analyses for isoflavone contents of foods have also been published and incorporated into food databases (34-39).

We have reported previously that among free-living women in EPIC-Norfolk, phytoestrogen concentrations in spot urine (adjusted for urinary creatinine) correlated strongly with those in serum, with partial correlation coefficients of >0.8. Trend tests showed significant dose-response relationships (P < 0.02) for both urinary and serum concentrations of isoflavones across increasing tertiles of dietary intakes. Hence, phytoestrogen concentrations in untimed spot urine and serum samples can serve as biomarkers of phytoestrogen intake in this group of women (30).

In this study, we used all three sources of material to quantify phytoestrogen exposure (i.e., dietary isoflavones, urinary phytoestrogens, and serum phytoestrogens). Serum and urinary phytoestrogens were expected to yield similar results due to the high degree of correlation between the two. Indeed, both serum and urinary levels of daidzein, genistein, and glycitein showed negative associations with plasma estradiol levels. There was however no association between equol and estradiol levels despite equol being the most estrogenic metabolite and the one that has been postulated to be most physiologically active (46). Equally, although it has been postulated that phytoestrogens may exert their cancer-protective effect through stimulation of SHBG production, thus reducing estradiol bioavailability (4-7), no correlation was found between phytoestrogens and plasma SHBG concentration after adjusting for BMI and age.

We have shown previously that the relationship between urinary and serum phytoestrogen levels is much stronger than the relationships between diet and either urinary or serum levels (30). This probably accounts for the fact that estimates of dietary intake showed no relation with either plasma estradiol or SHBG in contrast to the findings with urinary and serum measurements of isoflavones. Reviewing the literature, we found five cross-sectional studies that investigated the relationship between phytoestrogen exposure and plasma sex hormones. Interestingly, two studies, which had used urinary phytoestrogens, had found significant correlations between phytoestrogen exposure with plasma estradiol and SHBG (4, 7), whereas three studies, which had used dietary intake to quantify exposure, did not detect any significant relationship (25-27). It seems possible that the choice of markers used may at least partially explain the inconsistent results reported in studies published thus far.

One of the main mechanisms by which phytoestrogens are thought to exert biological effects is by interacting with several of the enzymes involved in estradiol biosynthesis and metabolism. Polymorphisms in the genes encoding for these enzymes could affect circulating levels. We chose to analyze for SNPs in CYP19 and SHBG because these were associated with differences in hormone levels in our previous findings (28). Polymorphisms in CYP17 were not studied here due to their absence of effect on hormone levels (28). In the present smaller study, there was no effect in the SHBG SNPs studied, in contrast to our previous findings, and no association with the SNPs studied in COMT. The previously seen trend of estradiol levels according to genotype for CYP19 3′ untranslated region SNP was also shown (28), but we were unable to show an interaction with phytoestrogens. CYP19 encodes aromatase. Earlier studies have found that concentrations (≥100 μmol/L) of isoflavones that were required to inhibit aromatase in vitro are far more than the serum levels observed in our subjects (47). This could possibly explain why we were unable to show any interaction in our subjects.

There was however a clear effect of ESR1 PvuII on plasma estradiol when we analyzed the data stratified according to different genotypes. Women with the CC genotype had significantly higher plasma estradiol levels (Table 2). Furthermore, in women with CC genotype, there were strong negative correlations between daidzein, genistein, and glycitein levels and plasma estradiol levels. This correlation was not seen in women with the ESR1 PvuII CT and TT genotypes. Our data indicate that the negative correlation observed between phytoestrogen measurements and serum estradiol in the women as a whole is mainly due to the small group of women with ESR1 PvuII CC genotype (Fig. 1). This observation has not been reported previously, and the mechanisms by which the ESR1 PvuII polymorphism could modulate the relationship between measured phytoestrogen intake and serum estradiol levels are unknown. The ESR1 PvuII polymorphism is intronic and could potentially affect receptor function via one of several mechanisms: (a) it could alter splicing of the mRNA (48), although this is unlikely because the SNP lies ∼400 bp from the nearest intron-exon boundary; (b) it could lie within a regulatory sequence, such as an enhancer, and affect levels of ESR1 gene expression (49); or (c) it could be a silent variant in linkage disequilibrium with another, as yet unknown, variant in the gene that does have a functional effect.

In this study, blood samples from our subjects were not collected at the same time of the day. Although it is possible that circadian variation may affect estradiol levels, the circadian variation in estradiol is very modest and we did not detect this in a previous study conducted by our own laboratory using highly sensitive assay (50). In addition, we had chosen to analyze plasma estradiol but not estrone. Although estrone is the major circulating estrogen in postmenopausal women, estradiol is a far more potent estrogen. Furthermore, the precision of measurement of estradiol for postmenopausal women using our specialist assays is greater than for estrone. However, this does negate the possibility of us establishing an effect on estrone specifically or on the estradiol/estrone ratio.

In conclusion, we showed that higher isoflavone (daidzein, genistein, and glycitein) exposure was associated with lower plasma estradiol levels among postmenopausal women in EPIC-Norfolk. Isoflavones have also been shown to be estrogenic themselves (47) and may thus partially replace the endogenous estrogens. A recent meta-analysis of nine prospective studies found that the relative risk of breast cancer associated with the doubling of estradiol levels was 1.29 (95% CI, 1.15-1.44; P < 0.001; ref. 3). In our study of 125 postmenopausal women, the geometric means of plasma estradiol concentrations for the highest and lowest tertiles of daidzein exposure (determined via urine and serum levels) were around 19 and 24 pmol/L, respectively. It is therefore unlikely that such values would translate to significant differences in breast cancer risk. However, in the small subgroup of women with the ESR1 PvuII CC genotype, the geometric means of plasma estradiol for the highest and lowest tertiles of serum daidzein were considerably different (17.5 versus 42.7 pmol/L) and would translate to >30% difference in breast cancer risk. This observation raises the interesting possibility of diet-gene interaction where the effect of phytoestrogen exposure may be exceptionally pronounced in women with a particular genotype but was attenuated when women of different genotypes were considered as a whole (Fig. 1). Given the very small numbers of women in this study with the ESR1 PvuII CC genotype, no definitive conclusions can be drawn. However, this intriguing possibility might explain some of the contradictory findings in the literature with respect to phytoestrogen intake and breast cancer risk (30) and this merits further investigation in larger study sets.

We thank the participants of EPIC-Norfolk and EPIC staff for their help with this work, Dr. Liz Folkerd and Debbie Doody for conducting the plasma estradiol and SHBG analyses, and Dr. Nigel Botting (University of St. Andrews, Fife, United Kingdom) for generously donating triply ¹³C-labeled phytoestrogen standards for urinary and serum analyses of phytoestrogens.