Direct and indirect associations between dietary magnesium intake and breast cancer risk

This study aimed to explore the effect of dietary magnesium intake on breast cancer risk both directly and indirectly via its effect on inflammatory markers C-reactive protein (CRP) and interleukin-6 (IL-6). This case-control study recruited 1050 case patients and 1229 control subjects. Inflammatory marker levels of 322 cases and 322 controls, randomly selected, were measured using ELISA, and data on dietary magnesium intake were collected using a food frequency questionnaire. Multivariable logistic regression was used to estimate the odds ratio (OR) and 95% confidence interval (CI), and path analysis was used to investigate the mediating effect. A higher magnesium intake was associated with a lower breast cancer risk (adjusted OR = 0.80, 95% CI = 0.65, 0.99). A positive association was found between the CRP level and breast cancer risk (adjusted OR = 1.43, 95% CI = 1.02–2.01). However, IL-6 was not found to be associated with breast cancer risk. Path analysis revealed that dietary magnesium affected breast cancer risk both directly and indirectly by influencing the CRP level. The results indicate that a direct negative association and an indirect association through influencing the CRP level were observed between dietary magnesium intake and breast cancer risk.

Association between dietary magnesium, IL-6, CRP and overall breast cancer risk. The odds ratios (ORs) and 95% confidence interval (CIs) of dietary magnesium intake, IL-6 level, CRP level and overall breast cancer risk are shown in Table 2. A higher dietary magnesium intake was associated with a lower breast cancer risk (crude OR = 0.75, 95% CI = 0.62, 0.92). This association remained significant after adjusting for basic characteristics (OR = 0.78; 95% CI = 0.63, 0.95) in Model 2 and after further adjusting for dietary factors (OR = 0.80; 95% CI = 0.65, 0.99) in Model 3. As shown in Fig. 2, there was a trend of reduced risk of breast cancer associated with increasing magnesium intake (P-trend < 0.001). A marginal positive association was observed between the IL-6 level and breast cancer risk when IL-6 levels were categorised into two groups based on the 1.5-pg/mL cut-off value (crude OR = 1.38; 95% CI = 1.00, 1.90). However, the association became non-significant after adjusting for potential confounders. Compared with participants with CRP values ≤3000 ng/mL (the lower group), those with CRP values >3000 ng/mL (the higher group) had a 1.43 times higher breast cancer risk (95%

Stratified analyses.
The results of the stratified analyses according to menopausal status, BMI and sex hormone status are shown in Table 3. A negative association was found between dietary magnesium intake and breast cancer risk among premenopausal women in Model 3 (adjusted OR = 0.75; 95% CI = 0.57, 0.98), but not among postmenopausal women. Stratified analysis by sex hormone status showed that dietary magnesium intake had a significant negative association with oestrogen receptor (ER)-positive (ER+) and progesterone receptor (PR)-positive (PR+) breast cancer risk in Model 3 (adjusted OR = 0.71; 95% CI = 0.53, 0.95). After adjusting for potential confounders, stratified analysis by BMI showed that a higher dietary magnesium intake was associated with a lower breast cancer risk among overweight or obese women (Model 3: adjusted OR = 0.69, 95% CI = 0.48, 0.99), but not among women with normal weight. Irrespective of the amount of calcium intake, no association was found between dietary magnesium intake and breast cancer risk after stratification. Table 4 shows the total, direct and indirect effects of dietary magnesium and calcium intake on breast cancer risk.

Discussion
This study showed that dietary magnesium intake was inversely associated with breast cancer risk and that higher CRP level was a risk factor for breast cancer development. Path analysis revealed that the serum CRP level, but not the serum IL-6 level, mediated the association between dietary magnesium intake and breast cancer risk.
This study focused on the association between dietary magnesium intake and breast cancer risk because diet is an easy target for intervention. The results showed an inverse association between dietary magnesium intake and overall breast cancer risk. To date, few epidemiological studies have investigated the association between dietary magnesium intake and breast cancer risk. Consistent with our result, an Italian case-control study found that the serum magnesium level was significantly lower among breast cancer patients than among control subjects 2 . Magnesium deficiency has been found to be involved in both the risk and prognosis of cancers, including breast cancer 2,6,40,41 . Some studies have focused on the effect of dietary magnesium on the prognosis of breast cancer 9,10 . Their results suggest that higher dietary magnesium intake is inversely associated with mortality among breast cancer patients.
Several experimental studies have suggested that magnesium affects tumourigenesis through two mechanisms, namely inflammation and free radicals-induced oxidative stress, both of which may cause DNA damage, subsequently leading to tumour onset 41 . In particular, dietary magnesium has been reported to play a crucial role in the regulation of systematic chronic low-grade inflammation, especially the circulating CRP level [14][15][16][17][18] . A meta-analysis including seven cross-sectional studies suggested an inverse association between dietary magnesium intake and serum CRP levels 14 . Another meta-analysis of 11 randomised controlled trials indicated that magnesium supplementation reduces circulating CRP levels in individuals with inflammation (CRP levels >3000 ng/mL) 15 . These findings indicate that magnesium intake may play a protective role in the management of inflammation, which may be one of the mechanisms of tumourigenesis. Therefore, the present study also explored how dietary magnesium affects breast cancer risk through inflammation. This was the first study to explore dietary magnesium together with inflammatory markers as the risk factor for breast cancer. The results showed that dietary magnesium intake affected breast cancer risk both directly and indirectly by modifying the CRP level. This result supports the hypothesis proposed in previous experimental studies that the potential beneficial effect of dietary magnesium intake on breast cancer prevention may, at least in part, be explained by the inhibition of inflammation 41 .
The signalling pathways linked to both STAT3 and NF-κB have been suggested to play important roles in the communication between inflammatory cells and cancer cells 42,43 . In particular, IL-6 has been proven to activate NF-κB and STAT3 pathways to facilitate a tumour micro-environment 36,42,43 . However, there is little evidence from observational studies investigating the effect of IL-6 level on breast cancer risk [37][38][39] . Consistent with our result, a combined analysis of two prospective studies revealed no significant relation between the IL-6 level and overall breast cancer risk 39 . In contrast with our result, a Korean case-control study found that the IL-6 level was significantly higher in breast cancer patients than in control subjects 37 . Given these contradictory results, more epidemiological studies with larger sample sizes should be conducted to explore the effect of IL-6 level on breast cancer risk.
Fundamental experiments have revealed that the moderately high levels of pro-inflammatory markers in chronic inflammatory status are responsible for the formation of an inflammatory micro-environment, which   www.nature.com/scientificreports www.nature.com/scientificreports/ affects the process of epithelial-mesenchymal transition 44,45 . CRP is a classic inflammatory marker that is highly sensitive to inflammatory response [24][25][26] . Several studies have attempted to investigate the association between the circulating CRP level and breast cancer risk, most of which have reported consistent results 30,33,35,38,39,46-52 ; however, few such studies have been conducted on Chinese populations. To date, four meta-analyses have summarised previous studies 30,39,49,51 , three 30,49,51 of which consistently supported a positive association between the CRP level and breast cancer risk as observed in our study. The other meta-analysis including 12 prospective studies suggested a marginal association between the CRP level and breast cancer risk 39 . A cohort study comprising 17 402 Chinese females showed that women with high CRP levels (>3000 ng/mL) at baseline had a significantly higher breast cancer risk (adjusted RR = 1.80; 95% CI = 1.03, 3.15) than women with low CRP levels (<1000 ng/ mL) 52 . Thus, the results of the present study support the hypothesis that high CRP levels (>3000 ng/mL) are a risk factor for breast cancer among Chinese women.
The present study found an inverse relationship between dietary magnesium intake and breast cancer risk among overweight or obese women but not among women with normal weight. It is well-known that obesity is highly correlated with inflammatory response and increased free radical levels [53][54][55][56] . One plausible reason is that women with more adipose tissue are more sensitive to the effect of inflammation; thus, even if dietary magnesium has only a small anti-inflammatory effect, it is likely to affect breast cancer risk in overweight or obese women.
Stratified analyses by menopausal status revealed a negative relationship between dietary magnesium intake and breast cancer risk among premenopausal women. Oestrogen is derived mainly from the ovaries in premenopausal women and from aromatase in postmenopausal women; both of these sources have been found to be regulated by pro-inflammatory cytokines in distinct ways 57,58 . Differences in the oestrogen biosynthesis mechanisms may account for the different effects of inflammatory marker levels on breast cancer risk between pre-and postmenopausal women 57,[59][60][61] . Dietary calcium intake has been suggested to have an interactive effect with magnesium on breast cancer risk. The protective effect of magnesium on breast cancer was stronger among women with adequate calcium intake than among those with calcium deficiency, although the association was non-significant partially due to the small sample size. Magnesium and calcium belong to the same family in the periodic table and share similar metabolic pathways 62 . A previous study on postmenopausal women suggested that compared with an adequate magnesium intake, dietary magnesium deficiency increases calcium retention 63 . The findings on the association between calcium intake and breast cancer risk remain controversial 64 .
To the best of our knowledge, this was the first study to explore the indirect effect of dietary magnesium on breast cancer via its influence on inflammatory marker levels and to assess the association between dietary magnesium intake and breast cancer risk among Chinese women. Despite its strengths, some limitations should be acknowledged. First, the sample size used for measuring inflammatory marker levels was relatively small, which may have affected the power of the test. Second, magnesium from hard water and supplements, which are also regarded as important sources of magnesium conducive to human body 17,65,66 , was not included in the calculation of dietary magnesium values. However, all study participants were natives of Guangdong province or had lived in Guangdong for at least 5 years, indicating that they shared similar water sources. In addition, we collected information about supplements and noted that few patients took magnesium supplements. Thus, magnesium from hard water and supplements may not have significantly affected the result. Third, selection bias and recall bias are inevitable in hospital-based case-control studies. To minimise selection bias, all control subjects were carefully recruited to exclude any diagnosis potentially related to breast cancer or dietary changes. In addition, the relatively high response rate also helped to reduce selection bias. To reduce recall bias, cases were interviewed immediately after breast cancer diagnosis. Moreover, food photographs were used to assist participants with the quantification of dietary intake. Fourth, it should be noted that it was impossible to include all mediators in the path model in which the estimates of direct and indirect effects might be affected. Moreover, the path model was based on the assumption that inflammatory response plays a mediating role, which should occur before the onset of breast cancer. However, this model could not completely avoid inverse causality due to the case-control design of the study. Thus, further prospective studies should be conducted to explore the mediating role of chronic inflammation in the association between dietary magnesium intake and breast cancer risk.
In conclusion, this study indicated that a higher dietary magnesium intake was associated with a lower breast cancer risk both directly and, in part, indirectly via reduction in the CRP level. The results also provided evidence of a positive association between the CRP level and breast cancer risk among Chinese women.

Materials and Methods
Study population. This hospital-based case-control study was conducted from September 2011 to July 2016, the details of which have been described elsewhere 67 . Study cases included patients admitted to the surgical units of the First Affiliated Hospital of Sun Yat-sen University and Guangdong Women and Children Hospital between September 2011 and July 2016 and were recruited based on their fulfilment of the inclusion and exclusion criteria. Patients were included if they were female, 25-70 years old, native of Guangdong province or had lived in Guangdong for at least 5 years and had been diagnosed with incident, primary, histologically confirmed breast cancer no more than 3 months before the study. Patients with a prior history of any cancer or who did not understand or speak Mandarin/Cantonese were excluded. Control subjects without breast cancer were frequency matched by age (5-year interval) to the case patients and were simultaneously recruited from the First Affiliated Hospital of Sun Yat-sen University. The remaining inclusion criteria for control subjects were similar to those for case patients. Control subjects were excluded if they were diagnosed with inflammatory disease, including chronic nasosinusitis, chronic otitis media, chronic tonsillitis or maxillary sinusitis.
This study followed the tenets of the Declaration of Helsinki, and all procedures involving humans were approved by the ethical committee of the School of Public Health, Sun Yat-sen University. All participants provided written informed consent for participation in the study before the interview.
www.nature.com/scientificreports www.nature.com/scientificreports/ Data collection. During hospitalisation, all participants were interviewed in person by trained interviewers using a structured questionnaire that included questions on demographic information, lifestyle factors (e.g. regular smoking, passive smoking, regular drinking and leisure-time physical activity), self-reported weight and height, menopausal status, diseases, reproductive history and family history of cancer. Regular smoking was defined as ever smoking at least one cigarette per day for more than six consecutive months. Passive smoking was defined as exposure to second-hand tobacco smoke for at least 15 min per day during the previous year. Regular drinking was defined as drinking alcohol at least once per week during the previous year. Leisure-time physical activity was classified into never, seldom (1 time/week) and often (≥1 time/week). BMI was calculated by dividing weight (kg) by height squared (m 2 ). Postmenopausal status was defined as at least 1 year since the last menstrual cycle. Relevant medical information, medical diagnosis, histological findings and ER and PR statuses were obtained from the hospital medical records.
A validated 81-item FFQ 68 was used to collect the previous year's dietary information before diagnosis for the cases or before the time of interview for controls. Magnesium, calcium, energy, macro-nutrients, phosphorus, and fibre intake per day was calculated from FFQ based on the frequency of food consumption, food items and serving sizes. Values of nutrients in foods were obtained from the 2002 Chinese Food Composition Table 69 .

Laboratory measurement.
Fasting venous blood samples of cases were collected on the second day of hospitalisation prior to any drug treatment or examination. The samples were centrifuged at 3000 rpm for 10 min at 4 °C, and the supernatants were aliquoted into eight parts of 200 μL. All serum samples were stored at −80 °C in an alarmed refrigerator for continuous monitoring until analysis.
High-sensitivity ELISA (Thermo Fisher Scientific, Inc., Carlsbad, CA) was used to measure IL-6 levels, and an ELISA kit (Thermo Fisher Scientific, Inc.) was used to measure CRP levels. Serum samples from both cases and controls were subjected to the same tests, run in the same batch of 96 samples (16 for standard curve, 40 for case samples and 40 for control samples), and assayed in a random order to reduce inter-assay variation and systematic bias. The mean intra-and inter-assay coefficients of variation were 4.9 and 6.0% for the IL-6 level and 5.0 and 5.0% for the CRP level, respectively. 70 , magnesium levels were classified as low (<280 mg/day) or high (≥280 mg/day) based on EARs, and calcium levels were categorised as low (<800 mg/day) or high (≥800 mg/day) based on adequate intake values. According to the Centers for Disease Control and Prevention/American Heart Association criteria, CRP levels were classified as low (≤3000 ng/mL) or high (>3000 ng/mL); this classification was originally created for the risk assessment of cardiovascular disease and has subsequently been used to explore the effect of inflammation on the development of various diseases 14,15,27 . Similarly, IL-6 levels were also categorised as low (≤1.5 pg/mL) or high (>1.5 pg/mL). As no guideline is available for the categorisation of IL-6 levels, the average value among healthy populations obtained from data in the literature was chosen as the cut-off value 37,71 . T-test was used to determine the differences in age, age at menarche and age at first live birth between the cases and controls. Wilcoxon rank-sum test was used to evaluate the differences in magnesium, calcium, phosphorus, dietary fibre, energy, carbohydrate, protein, fat, IL-6 and CRP levels between the two groups. Chi-square test was used to compare the categorical variables between the groups. Multiple unconditional logistic regression analysis was used to estimate the OR and 95% CI of the associations between dietary magnesium intake, serum IL-6 level, serum CRP level and breast cancer risk. The groups of participants with the lowest levels were used as reference groups. The values of Model 1 were shown as crude OR and 95% CI. Model 2 was adjusted for age at menarche, BMI, educational level, passive smoking, physical activity, parity, first-degree relative with cancer and history of benign breast disease to investigate the association between magnesium intake and breast cancer risk. Model 3 was further adjusted for the confounders from Model 2 in addition to the intake of energy, fat, calcium, phosphorus and dietary fibre. The associations between IL-6 levels, CRP levels and breast cancer risk were examined after adjusting for passive smoking, BMI, first-degree relative with cancer, history of benign breast disease and income level. Confounders were selected by comparing baseline characteristics between the cases and controls and between the current study and previous studies that evaluated the risk factors for breast cancer. Dietary magnesium intake was also classified into quartiles to explore the dose-response relationship with breast cancer risk, and then, linear trend was evaluated by entering the median value of magnesium intake for each quartile in the multiple regression model. Stratified analyses by menopausal status (premenopausal and postmenopausal), sex hormone status [ER+, or ER-negative (ER−); PR+ or PR-negative (PR−)], BMI (normal weight: BMI ≥ 18.5 and <24 kg/m 2 ; overweight or obese: BMI ≥ 28 kg/m 2 ) and calcium intake (<800 mg/day and ≥800 mg/day) were also performed 72 . The interactive effect was calculated by including an interaction term in the multiple regression model.

Statistical analysis. Based on Dietary Reference Intakes for Chinese residents
Path analyses were performed to investigate whether the inflammatory factors IL-6 and CRP were potential mediators contributing to the associations between dietary magnesium intake, calcium intake and breast cancer risk in Chinese women. Maximum likelihood is the most popular method in path analysis that is based on the assumption of multivariate normality. In this study, Mardia's coefficient of multivariate kurtosis was 13.66 and the critical ratio was 19.91. Both values less than 1.96 indicated significant non-normality; thus, Bayesian structural equation modelling was used to evaluate the overall presented path analysis, and estimations were conducted based on the Markov Chain Monte Carlo algorithm 73 . The model evaluation criterion was posterior predictive P value ranging from 0 to 1 with an acceptable quantity of 0.5 or close to it 74 .
All P values are two sided, and P values of <0.05 were considered as statistically significant. Statistical analyses were performed using SPSS 20.0 and AMOS 17.0.