Main

Plant lignans, a major type of phytoestrogens in Nordic countries, are mainly present in cereals, fruit, and vegetables (Adlercreutz, 1998a, 1998b) and are metabolised to mammalian lignans (e.g. enterolactone (ENL)) by the intestinal microflora (Adlercreutz, 2002). Since a preventive action of lignans against breast cancer was suggested (Adlercreutz et al, 1982), this has been evaluated in vitro (Welshons et al, 1987; Hirano et al, 1990; Mousavi and Adlercreutz, 1992), in vivo (Serraino and Thompson, 1991, 1992) and in clinical studies (Adlercreutz et al, 1988, 1991; Phipps et al, 1993; Thompson et al, 2005). Biological plausibility was discussed in a recent review (Adlercreutz, 2007). Hormone-dependent (Adlercreutz et al, 1992, 1993) and other mechanisms (Hirano et al, 1990; Kitts et al, 1999; Mäkelä et al, 1999; Prasad, 2000; Rickard et al, 2000) have been suggested. Six prospective (den Tonkelaar et al, 2001; Keinan-Boker et al, 2004; Kilkkinen et al, 2004; Olsen et al, 2004; Touillaud et al, 2007; Verheus et al, 2007) and six case–control studies (Pietinen et al, 2001; Dai et al, 2002; McCann et al, 2002, 2004, 2006; Fink et al, 2007) have evaluated the issue among postmenopausal women. Of these, only four considered oestrogen and progesterone receptor status of tumours (ER/PR) (den Tonkelaar et al, 2001; Olsen et al, 2004; McCann et al, 2006; Touillaud et al, 2007). We therefore examined the issue in a large population-based cohort study with stratification by family history of breast cancer, level of alcohol intake, body mass index, and use of postmenopausal hormone (PMH).

Materials and methods

The Swedish Mammography Cohort (SMC) was described previously (Wolk et al, 1998; Suzuki et al, 2006). It was established in 1987–90 that all women in Västmanland who were born in 1917–48, and in Uppsala born in 1914–48, were invited. A total of 66 651 women completed a questionnaire including diet. In 1997, a second questionnaire was sent to all cohort members. We excluded those with missing or incorrect data, with previous cancer (except non-melanoma skin cancer), who were not post-menopausal and who were 70+ years old at baseline leaving a cohort of 51 823 women. The information on diet was collected through self-administrated food-frequency questionnaires in 1987 and 1997. Total lignan intake were estimated using published values of following four lignans; secoisolariciresinol, matairesinol, lariciresinol, and pinoresinol (Mazur et al, 1996, 1998a, 1998b, 2000; Adlercreutz and Mazur, 1997; Mazur and Adlercreutz, 1998; Valsta et al, 2003; Milder et al, 2005; Penalvo et al, 2005; Schwartz and Sontag, 2006; Thompson et al, 2006). Other nutrients were calculated based on the Swedish National Food Administration database (Bergström et al, 1991). Cereals (60%), vegetables (27%), and fruits (10%) are the main sources of our lignans. Among a random sample of 137 women from the cohort, the correlation between the FFQ-based estimates of lignan intake and serum ENL levels measured by time-resolved fluoroimmunoassay (Adlercreutz et al, 1998) was r=0.2 (Spearman's rank). Date of breast cancer diagnosis, death, or migration from the study area were identified by linkage of the cohort through the Swedish Registration System. Information about receptor status of breast tumours, measured by an Abbott immunoassay (Pousette et al, 1986) and an immunohistochemical method, was obtained from Uppsala University Hospital and the Regional Oncology Centre. The study was approved by the Regional Ethics Committee at the Uppsala University Hospital and Karolinska Institute. We used time-dependent multivariate Cox proportional hazards regression model to estimate hazard rate ratios and 95% confidence intervals with age as the time scale (Korn et al, 1997). We subdivided lignan intakes into four categories based on approximate quartiles. Trend tests were conducted by using the median value for each category of lignans as a continuous variable. Heterogeneity in the results between the ER+PR+ and other subtypes was evaluated using the Wald statistic (Liao, 2004). P-value for interaction was evaluated by a likelihood ratio test. Analyses were performed by SAS system, version 9.1 (SAS Institute, Cary, NC, USA). Statistical tests were two-sided, and significance levels defined as P<0.05.

Results

Among 51 823 women with an average 8.3-year follow-up, 1284 invasive breast cancer cases were diagnosed, with details of ER/PR status available for 1188 cases. Of these, 716 were ER+PR+, 279 ER+PR−, 50 ER−PR+, and 143 ER−PR− tumours. Women with high lignan intake tended to be older, have more education and have greater use of PMH (Table 1).

Table 1 Age-standardiseda characteristics of risk factors for breast cancer according to the levels of lignan intake among 51 823 postmenopausal women in the Swedish Mammography Cohortb
Full size table

Overall, we observed a statistically significant inverse association between lignan intake and breast cancer risk (Table 2). Compared to women in the lowest quartile (<712 μg day−1), the multivariable adjusted relative risks (RR) for the highest quartile (⩾1036 μg day−1) were 0.83 (95% confidence interval=0.70–0.97; Ptrend=0.042) for overall, 0.86 (0.69–1.08) for ER+PR+, 0.77 (0.54–1.09) for ER+PR−, 0.92(0.56–1.52) for ER−PR−. There was no evidence for heterogeneity in the results between the ER+PR+ and other subtypes (all Pheterogeneities⩾0.65).

Table 2 Relative risks (RRs) and 95% confidence intervals for the association between FFQ-based estimated intake of lignans and postmenopausal breast cancer risk by receptor-defined subtype among 51 823 postmenopausal women in the Swedish Mammography Cohort
Full size table

In the full adjusted analysis stratified by family history of breast cancer, by levels of alcohol intake and by body mass index (<25 or ⩾25 kg/m2), there was no evidence for interaction with lignans in relation to overall risk or of any subtype; all Pvalues for trends were>0.60 and all Pvalues for interaction>0.35. We also observed a significant inverse association of lignans with overall risk among PMH ever-users; the multivariable adjusted RR for the highest quartile of intake compared to the lowest was 46% lower (Ptrend=<0.0001; Table 3). In contrast, among PMH never-users, no association was observed (Pinteraction=0.01). The observed interaction for PMH use seemed to be confined to ER+PR+ tumors (Pinteraction=0.016). There was no heterogeneity in the results between ER+PR+ and other tumors (all Pheterogeneity ⩾0.21). Lignans were positively correlated with intake of fruits and vegetables (r=0.4) and of cereal, fruit and vegetable fibre (r=0.7, 0.2 and 0.4, respectively). After adjusting for these factors, the result for lignans was slightly attenuated but still significant among PMH user (Table 3).

Table 3 Multivariable relative risks (RRs) and 95% confidence intervals (CI) for the association between total lignan intake and all postmenopausal breast cancer risk among 41 795 postmenopausal womena in the Swedish Mammography Cohort with stratified by use of PMH
Full size table

Discussion

In this large population-based prospective cohort of postmenopausal women, we observed a significant inverse association between lignan intake and overall breast cancer risk, especially among PMH user. There was no evidence of heterogeneity across ER/PR tumours. These results are similar to our previous study with a significant inverse association between cereal fibre and breast cancer risk among PMH users (Suzuki et al, 2008). The estimated lignan intake was correlated with cereal fibre (r=0.7) but after adjusting for specific fibres, the association among PMH users was still significant. This inverse association agrees with two previous studies among postmenopausal women (Fink et al, 2007; Touillaud et al, 2007). Non-significant inverse associations (Pietinen et al, 2001; Dai et al, 2002; McCann et al, 2002, 2004; Keinan-Boker et al, 2004; Olsen et al, 2004; Verheus et al, 2007) and no association (den Tonkelaar et al, 2001; Kilkkinen et al, 2004; McCann et al, 2006) have also been reported.

An inverse association of lignans with risk has been reported among premenopausal women (Dai et al, 2002; McCann et al, 2002, 2004, 2006; Linseisen et al, 2004; Piller et al, 2006a), among women with palpable cysts (Boccardo et al, 2004), and high epidermal growth factor concentrations (Boccardo et al, 2003), and among those carrying the A2 allele of CYP17 (McCann et al, 2002; Piller et al, 2006b) possibly associated with increased levels of endogenous hormone (Haiman et al, 1999). Given these findings, an inverse relation of risk with lignans is probable in subgroups of women with high circulating oestrogen level just as discussed with regard to isoflavone (Glazier and Bowman, 2001). The possible biological mechanism is not clear, but in vitro studies also showed that lignan ENL in the presence of oestrogens suppressed the oestrogen-induced proliferation in MCF-7 breast cancer cell (Mousavi and Adlercreutz, 1992) and stimulated the synthesis of sex hormone-binding globulin in liver cells (Adlercreutz et al, 1992).

The lack of association among overweight women may be due to the relatively high circulating oestrogen levels from PMH use having a stronger effect than the endogenous oestrogens formed in peripheral tissues (Cleland et al, 1985; Jurgens et al, 1992; Hankinson et al, 1998). Compared to lean women, obese women tend to have a lower prevalence of PMH use (Suzuki et al, 2006) and lower level of plasma ENL (Kilkkinen et al, 2001; Johnsen et al, 2004). Body fat might attenuate the effect of lignans by suppressing intestinal microflora activity (Nishizawa et al, 1988), or trapping ENL (Johnsen et al, 2004).

Our finding for ER+PR+ tumours among PMH users partly agrees with a prospective study (Touillaud et al, 2007), though these results were not confined to PMH users. No association was reported in two prospective studies (den Tonkelaar et al, 2001; Olsen et al, 2004) and a case–control study (McCann et al, 2006). Some nutrient misclassification and individual variation in intestinal microflora, as well as the lack of detailed information about PMH use are all relevant. Lignan estimates were not highly correlated with plasma ENL, but the observed correlation was comparable to those reported previously (Kilkkinen et al, 2003; Hedelin et al, 2006). In prospective cohort design, this misclassification of exposure tends to be nondifferential which may attenuate the observed association toward null. Further studies need to elucidate this issue with taking the circulating level of oestrogens into consideration.