2.1

Data sources

This study employs a 2-sample Mendelian randomization (MR) analysis to explore the causal link between gut flora and BCa. Data on gut flora and BCa are derived from a genome-wide association study (GWAS). The flowchart of this study is shown in Fig. 1 . Genetic variations in gut flora are sourced from the MiBioGen consortium and the Dutch Microbiome Project (DMP) on gut microbiota composition. The MiBioGen study supplied a dataset of genome-wide linked 16S rRNA gene sequencing from 24 cohorts, encompassing 18,340 participants across diverse ethnicities and age groups from Europe, Africa, and Asia, for the MR analysis of gut microbiota [ ]. 211 taxonomic units have been identified in these cohorts, including 9 phylums, 16 classes, 20 orders, 35 families, and 131 genera of associated genetic variation [ , ]. DMP possesses the largest species-level gut microbiome database to date, including metagenomic sequencing data from 7,738 participants from northern Netherlands. It has identified 207 taxa and 205 pathways, illustrating microbial composition and functionality. The GWAS catalog provides summary statistics for the metabolic pathways (entry numbers range from GCST90027446 to GCST90027857, https://dutchmicrobiomeproject.molgeniscloud.org ) [ ]. BCa-related data, involving 2,264 cases and 454,084 controls, were obtained from the public database GWAS ( https://gwas.mrcieu.ac.uk/ ), and genetic variation associated with BCa was analyzed by pooling the datasets GCST90041857 ( https://www.ebi.ac.uk/gwas/studies/GCST90041857 ) [ ]. See Table 1 .

An overview of the study design. Randomization (MR) analysis. GWAS = genome-wide association study; IVs = instrumental variables; IVW = inverse variance weighted; MR-PRESSO = Mendelian randomization pleiotropy residual sum and outlier; SNPs = single nucleotide polymorphisms.

2.2

The selection of instrumental variables (IVs)

To ensure the accuracy of the results of the causal effect of gut flora on BCa, appropriate SNPs were selected according to the basic principles of Mendelian randomization, and stringent quality control procedures were performed on the GWAS pooled data. Firstly, valid instrumental variables (IVs) must meet three essential assumptions: (1) IVs are strongly and reliably linked to the gut microbiota; (2) IVs are not associated with other confounding factors that might affect risk factors and outcomes; and (3) IVs can affect BCa solely through the gut flora. Second, for quality control of SNPs, we chose SNPs that were statistically significantly associated with the gut microbiome as IVs ( P < 5 × 10 ⁻⁸ ). Unfortunately, only a small number of SNPs were chosen as IVs. To explore more comprehensive results, we relaxed the association p-value threshold between the instrument and the exposure, selecting SNPs with P < 1 × 10 ⁻⁵ as IVs to obtain more microbial data [ ]. Consistent with previous studies, to avoid strong linkage disequilibrium (LD) that could lead to biased results, we used reference data from the 1000 Genomes Project European samples, set an LD threshold of r ² <0.001, and set an aggregation window of 10,000 kb for our analyses [ , ]. Finally, to assess the strength of the SNPs, we calculated F statistics for each SNP to quantify the statistical strength, and F statistics <10 were excluded [ ].

2.3

Mendelian randomization (MR) analyses

To determine the correlation between intestinal flora and BCa, we used five methods: inverse variance weighting (IVW), the MR-Egger method, the weighted median method (WMe), the weighted mode method (WMo), and the simple mode method (SM). Specifically regarding the five methods under study, IVW stands out as the primary analytical approach for MR. It assumes that all IVs share a common causal effect on outcomes through risk exposure, yielding unbiased estimates of effects without horizontal pleiotropy. Unlike the MR-Egger regression, which is a method used for horizontal pleiotropy, it is assumed that pleiotropy is present in >50% of the SNPs [ ]. And, assuming polytropy in <50% of SNPs, WMe and WMo can be used as robust scenarios [ ]. After IVW confirmed a causal relationship between exposure and outcome, if the direction of the OR values of MR-Egger and WMe were in agreement with them (both: OR > 1 or OR < 1), the results were considered to be reliable, and the next step of sensitivity and specificity assessment was performed.

We used the MR-Egger and MR Polytomous RESidual Sum and Outlier (MR-PRESSO) tests for horizontal polytomousness and outliers. MR-PRESSO is divided into three parts: the global test, the outlier test, and the distortion test. Compared with MR-Egger, MR-PRESSO can more accurately identify horizontal pleiotropy and outliers and screen for abnormal SNPs [ ]. And MR-Egger was used to explain the presence of directional genetic pleiotropy; when the P -value was > 0.05, we considered that there was no statistical significance of genetic pleiotropy. Subsequently, we performed a Cochran’s Q test to test for heterogeneity between instrumental variables, and when P < 0.05, we observed heterogeneity. Finally, to infer the reliability of causal direction, we used the MR Steiger directionality test to examine whether gut flora is directionally causally related to BCa [ ].

2.4

Statistical analysis

R software (version 4.1.2) was used for all statistical and data visualization analyses in MR. MR analyses of causality between gut flora and BCa were performed using the “TwoSampleMR” package and the “MRPRESSO” package, with P < 0.05 indicating a statistically significant potential causal relationship.