1 Introduction

BC is a worldwide public health concern that primarily affects women [1]. Based on the latest GLOBOCAN data, it has been observed that, among female individuals, BC has now surpassed lung cancer to become the leading cancer in terms of global incidence. In the year 2020, the global incidence of BC led to the diagnosis of more than 2.3 million women, with a regrettable 685,000 fatalities reported as an outcome [1,2,3,4]. Recent studies have documented that, beyond hereditary and genetic mutations affecting BC susceptibility [5, 6], acquired dietary and lifestyle factors can exert a substantial influence on the risk of BC [7,8,9]. Dietary and lifestyle modifications may be effective measures to reduce the risk of breast cancer.

Dairy products constitute a pivotal element of the traditional Western dietary pattern and serve as a prominent reservoir of essential nutrients [10]. The association between milk consumption and cancer risk has been subjected to comprehensive scrutiny. Existing observational evidence indicates that milk consumption is linked to a diminished risk of colorectal cancer [11]. In contrast, alternative observational findings propose a positive effect between milk consumption and the risk of prostate cancer [12]. The results of a substantial body of research on the impact of dairy product consumption on BC risk exhibit notable heterogeneity [13, 14]. Several observational investigations have suggested that the intake of dairy products may decreased the risk of BC. This association is postulated to be related to the presence of various compounds, notably calcium, vitamin D, and lactoferrin, all recognized for their anti-cancer properties, which may contribute to a decreased risk of BC [15, 16]. Conversely, other studies have identified a positive effect between dairy consumption and a heightened susceptibility of BC, which may be attributed to the concomitant increased intake of dietary fat, including saturated fat, elements known to have evidence of effect on augmented risk of BC [17, 18]. In addition, dairy products contain a complex array of contaminants and compounds, including estrogen, implicated in aberrant DNA replication and increased mitotic activity [18]. Furthermore, insulin-like growth factor I (IGF-I) found in milk has been linked to the promotion of BC cell proliferation, consequently amplifying BC risk [14].

Besides that, ambiguity surrounds the influence of different categories of dairy products (such as fermented dairy, high-fat dairy, or low-fat dairy) on BC risk, as well as the effects of these products on different subtypes of BC (e.g., estrogen receptor-positive, progesterone receptor-positive) [17,18,19]. Moreover, numerous investigations have postulated that dairy products may have evidence of effect on health-related issues, including obesity and diabetes, which may potentially influence BC development [20, 21]. The overall role of dairy products in BC etiology remains elusive, and the causal relationship between milk consumption and BC risk remains indeterminate due to the inherent limitations of observational studies, including confounding and reverse causation.

MR is a causal inference technique that uses genetic variants as instrumental variables to investigate the causal relationship between an exposure and an outcome [22, 23]. MR provides a powerful tool for overcoming confounding and reversing causality issues that can compromise observational studies [24]. Although existing MR analyses have revealed an association between dairy product intake and breast cancer risk, the causal relationship between different dairy products and different types of breast cancer requires further investigation [25, 26]. To provide more scientific preventive strategies for dietary choices related to BC. The aim of this study was to assess the causal association between different dairy product intakes and the risk of developing different types of BC through two-sample MR and analyze the role of possible mediators in this causal relationship.

2 Method

2.1 Data sources

Our study utilized data from the genome-wide association study (GWAS) database (https://gwas.mrcieu.ac.uk/), which is a publicly available database created by the Integrated Epidemiology Unit (IEU) at the University of Bristol [27]. We used MR to investigate the causal relationship between skimmed milk and full cream milk with BC (including ER+ breast cancer and ER− breast cancer). The GWAS data for skimmed milk, full cream milk, and BC (including ER+ breast cancer and ER− breast cancer) were obtained directly or indirectly from the GWAS summary data through the IEU Open GWAS project. Single nucleotide polymorphisms (SNPs) for ER− breast cancer included 21,468 cases and 105,974 controls. Summary-level data for ER+ breast cancer included 69,501 cases and 105,974 controls.

2.2 Instrumental variable selection

In the two-sample MR Study, we used SNPs as instrumental variables. When we selected P < 5 × 10–8 as the threshold, there were only a few SNPs had evidence of effect on with skimmed milk and full cream milk, and referring to the criteria of previous articles [28, 29], we choose a more relaxed threshold (P < 5 × 10–6) to get enough SNPs. Additionally, to remove IVs with chained imbalances, we confirmed that these SNPs met the criteria of linkage equilibrium (r2 < 0.001) and clump distance > 10,000 kb [30]. F-statistic > 10 was considered to indicate a strong association between the exposure and outcome variables [31].

2.3 Two-sample MR analysis

We selected three commonly used methods to determine the causal relationship between exposure and outcome. These include: (1) IVW: IVW computes the causal impact of an exposure on the outcome by amalgamating the ratio estimates for each SNP, effectively transforming MR estimates into a weighted regression of SNP-outcome effects concerning SNP-exposure effects [32]. (2) Weighted Median: The weighted median method yields unbiased estimates, even when a substantial portion, up to 50%, of the data originates from invalid instrumental variables (IVs) [33]. (3) MR-Egger: the MR-Egger is a tool that provides a causal effect through the slope coefficient from Egger regression and also detects small study bias [34, 35]. Among these methods, IVW was considered the most important for detecting causality [36]. The Cochran's Q test was used to detect heterogeneity. There is heterogeneity when the Cochran's Q test P < 0.05 [37]. The leave-one-out analysis was used to determine whether the causal estimate was driven by any single SNP [38]. In this step, we removed each SNP from the instrumental variables (IVs) one by one and reassessed the causal estimates. If certain SNPs had significant correlations with the exposure, these SNPs might have dominated the causal estimate. Therefore, we should re-estimate the causal effect by excluding all such SNPs to assess robustness.

2.4 PhenoScanner GWAS analysis

To further analyze the function and phenotype of each SNP, the PhenoScanner database(http://www.phenoscanner.medschl.cam.ac.uk/) was used to find genes that corresponding to each SNP in skimmed milk.

2.5 Two-step MR analysis

We use a two-step MR to analyze potential mediators [39]. (1) Effect of skimmed milk on ER− breast cancer (2) Effect of BMI on ER− breast cancer (3) Regulatory effect of BMI (Fig. 1). The indirect BMI-mediated effect is obtained by multiplying b × c, where b is the MR Effect of skimmed milk on BMI, and c is the MR Effect of mediated BMI on ER-breast cancer, and a is the MR Effect of skimmed milk on ER-breast cancer. we quantified the proportion mediated by dividing indirect effect by the total effect [40, 41]. All analyses were performed using the Two Sample MR package version 0.5.6 in R software.

Fig. 1
figure 1

The flow chart of the two-step MR analysis. In mediation analysis where a represents the total effect, a' represents the direct effect and the indirect effect can be calculated by subtracting a' from a (difference method) or multiplying b × c (product of coefficients method). BMI, Body mass index

Full size image

3 Result

3.1 Exposure and outcome selection

In accordance with the aforementioned criteria for relevance, we have identified 28 SNPs associated with skimmed milk (Supplementary Table 1) and 28 SNPs associated with full cream milk (Supplementary Table 2) as instrumental variables. The statistical F-statistic for full cream milk was calculated to be 232.3, and the corresponding value for skimmed milk was 749.4, suggesting a high degree of instrumental strength. The outcome variables of our study encompassed BC, ER+ breast cancer, and ER− breast cancer. Contributing studies of exposures and outcomes were individuals of European descent The sources were presented in Table 1.

Table 1 Information of the exposures and outcome datasets
Full size table

3.2 Two-sample MR analysis

IVW analysis indicated a potential inverse association between skimmed milk and BC risk [OR 0.34, 95% CI (0.12–1.00), P = 0.05]. Subgroup analysis revealed that skimmed milk reduces the risk of ER-breast cancer [OR 0.18, 95% CI (0.04–0.90), P = 0.04], but not ER+ breast cancer [OR 0.42, 95% CI (0.15–1.22), P = 0.11]. Furthermore, we found that full cream milk had little evidence of effect on BC risk [P > 0.05]. In addition to the IVW testing method, we also used MR Egger and weighted median methods to verify our results. MR Egger reached similar results, that is, skimmed milk reduces the risk of ER− breast cancer [OR 0.01, 95% CI (0.00–0.70), P = 0.04], but not BC [OR 0.16, 95% CI (0.01–4.66), P = 0.30] and ER+ breast cancer [OR 0.50, 95% CI (0.02–12.61), P = 0.65]. However, weighted median showed that skimmed milk had little evidence of effect on BC (including ER+ breast cancer and ER− breast cancer) risk [P > 0.05]. MR Egger and weighted median showed that full cream milk had little evidence of effect on BC risk (including ER+ breast cancer and ER− breast cancer) [P > 0.05]. The specific MR causal relationships are presented in Table 2.

Table 2 The results of MR analyses
Full size table

3.3 Heterogeneity and pleiotropy test

The Cochran's Q test analysis indicated heterogeneity between skimmed milk and full cream milk regarding their effect with BC (including ER+ breast cancer and ER− breast cancer) (P < 0.05). However, the MR-Egger intercept analysis showed that the evidence of horizontal pleiotropy is limited between skimmed milk, full cream milk, and BC (including ER+ breast cancer and ER− breast cancer) (P > 0.05) (Table 3). The leave-one-out analysis revealed that rs55872725 in skimmed milk plays an important role of the results (Fig. 2).

Table 3 The tests for heterogeneity and horizontal pleiotropy
Full size table
Fig. 2
figure 2

The Leave-one-out analysis of skimmed milk and ER− breast cancer

Full size image

3.4 PhenoScanner GWAS analysis

Using the PhenoScanner database, we analyzed 28 SNPs of skimmed milk associated with ER-breast cancer and analyze the gene corresponding to each SNP. Of note, five SNPs (Table 4) were associated with BMI.

Table 4 The 5 SNPs associated with BMI
Full size table

3.5 Two-step MR analysis of the role of BMI in the causal relationship between skimmed milk and ER-breast cancer

To further investigate the role of BMI in the reduction of ER-breast cancer risk with skimmed milk, we performed two-step MR analysis. IVW analysis indicated a positive effect between skimmed milk and BMI [OR 5.18, 95% CI (1.87–14.36), P = 2.00E−03], and there is an inverse effect between BMI and ER-breast cancer [OR 0.82, 95% CI (0.72–0.93), P = 2.00E-03] (Table 5). By the two-step MR analysis, it was found that the proportion mediated by BMI was 15.86% (Table 6).

Table 5 The result of two-step MR analysis BMI as a mediating factor between skimmed milk and ER-breast cancer
Full size table
Table 6 The result of mediation MR analysis
Full size table

In mediation analysis where a represents the total effect, a' represents the direct effect and the indirect effect can be calculated by subtracting a’ from a (difference method) or multiplying b × c (product of coefficients method).

4 Discussion

In the context of this MR and potential mediator analysis, we scrutinized the causal relationship between milk consumption and the susceptibility to BC. Our findings validated a protective effect between the consumption of skimmed milk and a diminished risk of ER− breast cancer. Additionally, we uncovered no compelling evidence to substantiate an impact of full cream milk consumption on the risk of BC. Further mediation analyses elucidated that consumption of skimmed milk may reduce ER-breast cancer risk partly through BMI.

The question of whether dairy products exert a protective or detrimental influence on the development of BC in women is a subject of ongoing debate. Overall, existing epidemiological research has yielded inconclusive findings concerning the association between dairy consumption and the incidence of BC. While certain constituents of milk are positively linked to an increased risk of BC [42, 43], others seem to have a protective effect [13]. Dairy products are rich in calcium, vitamin D and conjugated linoleic acid. These substances affect cell proliferation and differentiation and inhibit tumor development [44]. In contrast, the high levels of fat, saturated fat, and potentially carcinogenic contaminants such as pesticides, estrogen metabolites, and growth factors (e.g., IGF-1) contained in dairy products are also supported by prospective cohort studies that suggest an increased risk of ER+ breast cancer with milk consumption [45].

Acknowledging the pivotal influence of estrogen receptors and lipid metabolism on BC susceptibility, our study is primarily concerned with investigating the causal relationships between the consumption of full cream milk and skimmed milk and the incidence of ER− breast cancer and ER+ breast cancer. The meta-analyses of several previous observational studies have suggested an association between the consumption of milk and dairy products, especially low-fat or fermented dairy products, and a reduced risk of BC [44, 46, 47]. Our study provides further support for this notion by indicating that skimmed milk intake decreases the risk of ER-breast cancer, aligning with prior observational research [43, 47]. We hypothesize that the reason behind this risk reduction may be attributed to the relatively low saturated fat content in skimmed milk and the possibility that individuals with ER− breast cancer may exhibit lower sensitivity to estrogen-related substances found in milk. Evidence suggests that the potentially saturated fat and estrogen in full cream milk increases the risk of breast cancer, while the protective molecules in skimmed milk inhibit the development of breast cancer, which may ultimately be the reason we find that skimmed milk reduces the risk of ER breast cancer.

BMI was found to have an important role in our analysis of potential mediators of skimmed milk in reducing the incidence of ER− breast cancer. It has been reported that there is an inverse relationship between obesity and the risk of premenopausal BC [48, 49]. Seven prospective cohort studies, encompassing 337,819 women and 4,385 cases of invasive BC, reported an inverse association between BMI and the risk of BC [50]. A meta-analysis also indicated that BMI has evidence of effect on a reduced risk of BC incidence [51]. Furthermore, other observational studies have reported an inverse relationship between obesity and the risk of postmenopausal ER− breast cancer and TNBC (triple-negative breast cancer). A breast X-ray cohort study conducted in Sweden involving 51,823 postmenopausal women found an inverse association between obesity and hormone receptor-negative BC [52]. This relationship has also been observed in other studies [53,54,55]. Rs55872725 was located in the intron of the FTO gene(https://www.ncbi.nlm.nih.gov/snp), which had a very strong correlation with BMI [56]. Several previous studies have shown an association between obesity and breast cancer incidence [57, 58], suggesting that BMI may be associated with a causal relationship between skimmd milk and ER-breast cancer. In our analysis of genes associated with SNPs related to skimmed milk, we found that some of these genes were strongly associated with obesity, suggesting that BMI may play a role in the reduction of ER-breast cancer with skimmed milk. We then further confirmed the above hypothesis in mediating factor analysis, indicating that skimmed milk partially reduces the risk of breast cancer through BMI.

Our study has several limitations that should be taken into consideration. Firstly, the use of GWAS data has inherent limitations. For instance, we were unable to evaluate the non-linear relationship between milk consumption and BC risk or assess BC risk at different ages [59]. Secondly, since the number of instrumental variables was insufficient for analysis when using a genome-wide significance threshold, we opted for a P < 5e−06 threshold. This threshold was also based on other literature. Moreover, our Mendelian randomization and pleiotropy results indicate the relative reliability of our findings. However, it is still not possible to completely rule out the potential "contamination" of the genetic instruments for the exposure of interest by these variants, and further analysis is needed. Thirdly, in our mediation analysis, although the primary IVW analysis suggests that BMI may have a potential mediating role in the causal relationship between skimmed milk and the risk of ER-breast cancer, the P-values for the weighted median and MR Egger tests were not statistically significant, requiring further validation with external data. Additionally, the unclear timing of data collection in existing databases means we cannot accurately test mediation behaviors. Moreover, as we analyzed our exposure phenotypes at the genetic level, our results may not be directly comparable to those of traditional observational studies. Additionally, our analysis was limited to European populations, and it is unclear whether the observed association can be extrapolated to other populations.

In summary, we revealed skimmed milk reduces the risk of ER− breast cancer by the MR study. In contrast, there was no significant causal relationship between full cream milk and BC, including both ER+ breast cancer and ER− breast cancer. Further the two-step MR analyses suggest that this protective effect may partly result from BMI. Overall, increasing the intake of skimmed milk could be a viable strategy for preventing ER− breast cancer.