Dietary phyto-oestrogens have been associated with a decreased risk for osteoporosis because of their structural and functional similarities with 17β-oestradiol and their ability to interact with oestrogen receptors(Reference Brynin1, Reference Branca and Lorenzetti2), in particular oestrogen receptor-β(Reference Morito, Hirose and Kinjo3). In vitro and animal studies have shown that phyto-oestrogens can modulate bone turnover and delay bone loss in acute oestrogen deficiency(Reference Setchell and Lydeking-Olsen4). However, a meta-analysis of ten randomised clinical trials lasting at least 1 year found only a weak beneficial effect for larger doses of isoflavones (>80 mg/d), but none for lower doses(Reference Liu, Ho and Su5).
Observational studies have reported a positive association between phyto-oestrogen intake and bone density in Asian populations(Reference Zhang, Shu and Li6, Reference Greendale, FitzGerald and Huang7); however, studies in Caucasian populations are limited due to the lower habitual intake of phyto-oestrogens. In the Shanghai Women's Health study of 75 000 postmenopausal women, the median intake of soya isoflavones (genistein and daidzein) was 38 mg/d, and there was a clear association between soya food consumption and a reduced risk for fractures(Reference Zhang, Shu and Li6). In contrast, the median intake of soya isoflavones in Caucasian women in the Study of Women's Health Across the Nation (SWAN) was 0·04 mg/d, while in the same study, the intake among Chinese (5·3 mg/d) and Japanese (18·9 mg/d) women was much higher(Reference Greendale, FitzGerald and Huang7).
We investigated the association between dietary phyto-oestrogens and bone density in 2580 women and 4973 men in the European Prospective Investigation into Cancer (EPIC)-Norfolk cohort(Reference Day, Oakes and Luben8). In addition to soya isoflavones used in most observational studies conducted thus far, we have also included lignans, enterolignans including equol and coumestrol in our analysis. To our knowledge, this is the largest observational study of the association between phyto-oestrogens and bone health conducted in a European population.
Methods
Study population
Between 1993 and 1997, approximately 25 000 healthy men and women aged between 40 and 75 years were recruited for the EPIC-Norfolk study from general practitioners' databases(Reference Day, Oakes and Luben8). Health and lifestyle, including data on smoking, alcohol consumption, social class, family medical history, physical activity and reproductive history, were assessed by a questionnaire. Postmenopausal status was defined as at least twelve consecutive months of amenorrhoea. Height and weight measurements were collected following a standardised protocol as part of a health check conducted by research nurses(Reference Day, Oakes and Luben8). Diet was assessed by a prospective 7 d diet diary at baseline. Between 1998 and 2000, men and women then aged 42–82 years were invited to attend a second visit, where quantitative ultrasound measurements of the calcaneum were conducted by trained research nurses, according to standardised protocols (n 10 444)(Reference Welch, Camus and Dalzell9–Reference Welch, Bingham and Camus11). Men (n 4973) and postmenopausal women who were not on hormone replacement therapy (n 2897) were eligible for the present study. The study was approved by the Norwich District Health Authority Ethics Committee, and all participants gave signed informed consent.
Analytical methods
Diet was assessed at baseline with a 7 d diary using the in-house dietary assessment software DINER (Data into Nutrients for Epidemiological Research) to calculate average daily nutrient intakes(Reference Welch, McTaggart and Mulligan12). Dietary phyto-oestrogens were determined using the food composition database for foods commonly consumed in the UK(Reference Kuhnle, Dell'Aquila and Aspinall13–Reference Kuhnle, Dell'Aquila and Aspinall16). More than 500 different foods were analysed by liquid chromatography–MS using 13C-labelled internal standards(Reference Kuhnle, Dell'Aquila and Aspinall13–Reference Kuhnle, Dell'Aquila and Low17); the limit of detection of this method was 1·5 g/100 g dry weight (intra-batch CV was between 3 and 14 %; inter-batch CV was between 1 and 6 %). These data were used to calculate the phyto-oestrogens content of 10 708 foods in DINER.
At the first follow-up visit (1998–2000), quantitative ultrasound was used to measure broadband ultrasound attenuation (BUA, dB/MHz) of the calcaneum. Using the CUBAClinical machine (McCue Ultrasonics, Winchester, UK), BUA was measured at least twice on each foot as described previously(Reference Welch, Bingham and Camus11). A total of five machines were used, and each machine was calibrated daily with their physical phantom and monthly with a roving phantom and on one operator's calcaneum. As there was no effect of adjustments for temperature or machine, the data in the present study are presented unadjusted for temperature or machine(Reference Welch, Bingham and Reeve18).
Data analysis
Descriptive characteristics of the study population were summarised using means and standard deviations for normally distributed continuous variables, frequency and percentages for categorical variables and medians (interquartile ranges) for non-normally distributed continuous variables. The bottom quintile of Ca intake was defined as ‘low’ intake (women 198–570 mg/d and men 161–699 mg/d); the top quintile as ‘high’ intake (women 958–2363 mg/d and men 1152–3251 mg/d).
Associations between BUA and log-transformed phyto-oestrogen intake were investigated using linear regression models, including age, BMI and total energy intake; potential confounders (age, BMI, energy intake and physical activity) were evaluated through comparison of models with and without the variable under examination, and were retained in the final model if their inclusion resulted in changes to the regression coefficient for total phyto-oestrogens intake of more than 10 %. For the regression analysis, dietary intake was log transformed. Dietary Ca and phyto-oestrogens were correlated using Spearman's rank correlation. Interactions between phyto-oestrogen intake and Ca intake were investigated by including an interaction term in the regression model. All analyses were conducted with STATA (version 10 for UNIX; Stata Corporation, College Station, TX, USA).
Results
BUA, anthropometric and dietary data were available for 4973 men and 2580 women who were postmenopausal and never on hormone replacement therapy. Table 1 shows demographic, anthropometric and dietary information of the study population. Median intake of phyto-oestrogens was lower in women (0·9 mg/d) than among men (1·2 μg/d). Among both sexes, the most common types of phyto-oestrogen were the isoflavones daidzein (205 μg/d in women and 289 μg/d in men) and genistein (352 μg/d in women and 497 μg/d in men), and the lignan secoisolariciresinol (194 μg/d in women and 243 μg/d in men), with lower intakes in women than in men.
(Mean values, standard deviations, medians and interquartile ranges)
The results of the regression analyses are summarised in Table 2. In postmenopausal women, regression analyses with age, BMI and energy intake as covariables showed no significant association for total phyto-oestrogens, total isoflavones, coumestrol or lignans. Only formononetin showed a borderline significant positive association (β = 1·2; P < 0·1), which remained significant when including dietary Ca in the model (β = 1·2; P < 0·1). However, there was a significant association for total enterolignans and equol (β = 1·5; P < 0·05), enterolactone (β = 1·4; P < 0·05) and equol (β = 1·5; P < 0·05). When including dietary Ca as a covariable, these associations became non-significant. In men, only biochanin A was positively associated with BUA (β = 1·2; P < 0·05), and this association remained significant when including dietary Ca in the model (β = 1·2; P < 0·02). Dietary Ca was significantly associated with BUA in women (β = 3·2; P < 0·005) and men (β = 3·5; P < 0·005), and it was significantly correlated with enterolactone and equol intake (Spearman's r 0·88, P < 0·001 in men; 0·90, P < 0·001 in women).
(β-Coefficients and 95 % confidence intervals)
* Adjusted for age (years), BMI (kg/m2) and energy intake (MJ/d).
† Log transformed.
Formal tests for interaction between Ca and phyto-oestrogen intake were conducted but showed no significant result. Following stratification of data by Ca intake, a borderline significant association was found in women between BUA and the intake of total phyto-oestrogens (β = 1·8; P < 0·1) and isoflavones (β = 1·4; P < 0·1), in particular the soya isoflavones daidzein (β = 1·4; P < 0·1), genistein (β = 1·4; P < 0·1) and glycitein (β = 1·9; P < 0·1) in the lowest quintile of Ca intake (Table 3). No significant association was found in the highest quintile and in men. BUA was not significantly different between the highest and lowest quintiles of Ca intake (Table 4); however, participants in the lowest quintile were significantly older and had a higher BMI, but consumed a lower amount of phyto-oestrogens.
(β-Coefficients and 95 % confidence intervals)
* Adjusted for age (years), BMI (kg/m2) and energy intake (MJ/d).
† Bottom quintile (198–570 mg/d).
‡ Top quintile (958–2363 mg/d).
§ Bottom quintile (161–699 mg/d).
∥ Top quintile (1152–3251 mg/d).
(Medians and interquartile ranges)
* Bottom quintile (198–570 mg/d).
† Top quintile (958–2363 mg/d).
‡ Bottom quintile (161–699 mg/d).
§ Top quintile (1152–3251 mg/d).
∥ Wilcoxon's rank-sum test.
Discussion
The present investigation of phyto-oestrogen intake and bone density is the largest prospective study in a Western population undertaken so far. The EPIC-Norfolk cohort was well positioned to investigate the associations between dietary phyto-oestrogens and bone density in a Western population and to improve the observational data available. Whereas most studies conducted previously have investigated only the effect of the so-called soya isoflavones daidzein and genistein(Reference Coxam19), the present study also included other isoflavones and other types of phyto-oestrogens such as lignans, coumestrol, enterolignans and equol using data from a newly developed food composition database(Reference Ward and Kuhnle20). A further strength of the present study is the use of 7 d food diaries. However, a limitation of the study is that only information for two types of lignans, matairesinol and secoisolariciresinol, are available. A further limitation of the present study, as with other studies relying on dietary data to assess exposure to phyto-oestrogens, is that inter-individual differences in metabolism, in particular the conversion of phyto-oestrogens to equol, O-desmethylangolensin and enterolignans by the intestinal microflora, cannot be accounted for. In the present study, we used BUA instead of dual-energy X-ray absorptiometry (DXA); quantitative ultrasound or BUA has been identified as a reliable and accurate tool for the assessment of osteoporosis and fracture risk. A recent comparison has shown that quantitative ultrasound is at least as reliable as DXA for the prediction of fractures in the elderly(Reference Moayyeri, Kaptoge and Dalzell21, Reference Trimpou, Bosaeus and Bengtsson22).
Compared with the dietary intake of genistein and daidzein in other observational studies, the median intake in women of the EPIC-Norfolk cohort was higher than in European women in the SWAN study (0·6 v. 0·04 mg/d) and cohorts in The Netherlands (0·4 mg/d(Reference Keinan-Boker, van Der Schouw and Grobbee23)) and Canada (0·2 mg/d(Reference Cotterchio, Boucher and Kreiger24)), but lower than in a Californian cohort (1·7 mg/d(Reference Horn-Ross, John and Canchola25)) and much lower than in Japanese (7·2 mg/d) or Chinese (3·5 mg/d) participants of the SWAN study(Reference Greendale, FitzGerald and Huang7) or in the Shanghai Women's Health Study (38·0 mg/d(Reference Zhang, Shu and Li6)).
Significant or borderline significant positive associations between dietary phyto-oestrogens and bone density measured as BUA were only found for the non-soya isoflavones formononetin in women and biochanin A in men; high amounts of these compounds are found, for example, in chick peas(Reference Kuhnle, Dell'Aquila and Aspinall16). Formononetin and biochanin A are both 3′-methylated isoflavonoids, which differ only in the presence of an hydroxyl group in the 5′-position; however, it is not possible to investigate whether this structural difference explains the observed associations with the data available.
There was no significant association between any lignan and bone density. Although the coumestan coumestrol was the first phyto-oestrogen for which animal studies have suggested a protective effect on bones(Reference Tsutsumi26), no significant effect was found in the present study.
Strong significant associations were observed for the so-called ‘mammalian phyto-oestrogens’, enterolactone and equol, in postmenopausal women, but not in men. These compounds are formed from lignans and daidzein, respectively, by the gastrointestinal microflora(Reference Webb and McCullough27). So far, it has not been possible to investigate the associations between these compounds and bone density directly, as insufficient information on the dietary content of these compounds was available.
Dietary Ca is an important contributor to bone health(Reference Dawson-Hughes28, Reference New, Robins and Campbell29) and positively associated with bone density. We have therefore investigated the effect of dietary Ca on the associations observed. Whereas no significant effect was observed for phyto-oestrogens of plant origin, the associations between BUA and enterolignans and equol in postmenopausal women were attenuated and became non-significant. The main sources of enterolignans and equol in the diet of participants of the EPIC-Norfolk study are cheese and other dairy products(Reference Kuhnle, Dell'Aquila and Aspinall13), which are also a main source of Ca, and the strong correlation between these compounds and dietary Ca is a possible explanation. The high correlation limits our ability to identify which dietary factor exerted greater influence, and it is possible that enterolignan and equol intake does affect bone density but is attenuated on adjustment for dietary Ca as a result of collinearity.
We have further investigated the effect of Ca by stratifying the cohort by Ca intake: in the lowest quintile of Ca intake (less than 570 mg/d) in postmenopausal women, total phyto-oestrogens, total isoflavones and soya isoflavones were all positively associated with BUA, although the association was only borderline significant; the association was non-significant for higher dietary Ca intake; no significant associations were found in men. The differences between postmenopausal women in the top and bottom quintiles of Ca intake do not provide an obvious explanation for the observed associations. These results confirm observations from previous studies, which have shown that the beneficial effect of soya on bone density is stronger in subjects with lower dietary Ca intake(Reference Chen, Ho and Lam30–Reference Alekel, Van Loan and Koehler32). The mean Ca intake in cohort studies, which showed a positive association between soya intake and bone health, was below 650 mg/d in the Shanghai Women's Health Study(Reference Zhang, Shu and Li6) and the Chinese and Japanese participants of the SWAN study(Reference Greendale, FitzGerald and Huang7). These data suggest that with low Ca intake, soya intake can affect bone density; the strong association in populations with high habitual soya consumption, and the weak association in this cohort, also suggest a dose–response effect. However, there are still insufficient data available to explain this association.
The results in postmenopausal women clearly suggest an interaction between dietary Ca, soya and bone density. Conversely, the results for men do not suggest such an interaction: enterolignans and equol were not associated with BUA, and no significant associations were observed after stratification for Ca intake.
The magnitude of the relationship observed (difference of adjusted means between the top and bottom quintiles of intake) was 1·5 dB/MHz of BUA in men, and between 1·2 dB/MHz (for formononetin) and 2·1 dB/MHz (total phyto-oestrogens in the lowest quintile of Ca intake) in postmenopausal women. Although this estimate would be attenuated due to the random measurement error in dietary intake, it is of interest in terms of the physiological relationship between dietary phyto-oestrogen intake and BUA. A reduction of 1 sd of BUA (20 dB/MHz) is associated with about doubling in the risk of fracture(Reference Khaw, Reeve and Luben10), so 2 units might be associated with a 10 % difference in fracture risk.
In conclusion, our data indicate an independent association between the non-soya isoflavones formononetin and biochanin A and bone density in postmenopausal women and men, respectively. Positive associations observed for enterolignans and equol in postmenopausal women became non-significant following adjustment for Ca intake, and the likely explanation for most compounds is that these compounds act as a surrogate marker for Ca. In postmenopausal women with low habitual Ca intake, soya isoflavones were borderline significantly associated with increased BUA in postmenopausal women, confirming reports from previous studies. Our results therefore suggest that non-soya isoflavones are associated with bone density independently of Ca, whereas the association with soya or soya isoflavones is affected by dietary Ca.
Acknowledgements
The present study was supported by the Medical Research Council, UK. The authors thank the EPIC-Norfolk study participants, general practitioners and study teams. The authors also acknowledge gratefully the support and expert advice by Dr Christina C. Dahm. The authors have no conflict of interest to declare. The authors' responsibilities were as follows: G. G. C. K. conducted the statistical analyses and wrote the manuscript; H. A. W. and A. V. conducted the statistical analyses and contributed to the manuscript; R. N. L., A. M. and N. G. F. contributed to the manuscript; K.-T. K. and N. J. W. (Principal Investigators of EPIC-Norfolk) contributed to the manuscript.
