The cause and effect of gut microbiota in development of inflammatory disorders of the breast

The diagrammatic sketch of this research is illustrated in Fig. 1. Briefly, the exposure is the GM, whereas the outcome is IDB. Moreover, reliable results are based on the following 3 assumptions of MR analysis [28]: (1) the closely relationship between the instrumental variables (IVs) and exposure should be a must; (2) IVs should be independent, ensuring no relation with confounding factors; (3) IVs influenced outcome through exposure rather than other factors.

This research related summary-level data were downloaded from openly GWAS database. In detail, the GWAS data on GM originated from MiBioGen consortium [29‐31] and the GWAS data relating IDB were mainly conducted by the Finngen consortium [32, 33]. Ethical approval and consent of GWAS database were achieved, and the summary-level data were publicly available and could be used.

MiBioGen consortium included 24 large cohorts (18,340 participants) from most European countries. 16S rRNA sequencing was used to explore composition of microbial communities and its classification via microbial classification standards [34]. 122,110 variant sites from 211 taxa were obtain in microbiota-GWAS. Owing to 12 unknown genera and 3 unknown families, a total of 196 taxa were included for analysis in the end. In our study, we selected IVs from genus to phylum level of GM taxa. For more detailed information, please refer to original articles [29]. According to the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10), this phenotype is “inflammatory disorders of the breast” (ICD-10 code N61). IDB is defined as the inflammation of breast tissue during lactation or postpartum due to an obstructed duct or infection. IDB can also occur in non-breastfeeding women, and rarely in men. We use this phenotype for the following reasons: firstly, enough types of relating diseases: this phenotype excludes neonatal infective mastitis, includes (1) acute, chronic and nonpuerperal abscess of areola and breast; (2) carbuncle of breast; (3) acute, subacute and nonpuerperal mastitis. Secondly, profound impacts of relating diseases: we have ploughed through relating documents that whatever disease which leading to the inflammation of breast tissue may result in the interruption of lactation and have impact on mother and children health [35, 36]. Therefore, once women develop IDB, this adverse state could inevitably affect women themselves and if women were in lactation period, it could bring breastfeeding crisis. Thirdly, this consortium was large enough to explore the causality between GM and IDB. Above all, we introduced this consortium. A total of 177,446 participants were involved in this GWAS. Among this GWAS, it recruited 177,446 female subjects and divided into 1435 cases and 176,011 controls. A series of corrections have been made during the performance [32].

The selection criteria of IVs were following: (1) previous articles were referred to formulate a relatively more wide-ranging principle (p < 1 × 10^–5) [37, 38]. Therefore, p < 1 × 10^–5 was performed because of the less eligible IVs (p < 5 × 10^–8) [39, 40]. (2) 1000 Genomes project European samples data were referenced to compute the linkage disequilibrium (LD) (R² < 0.001, clumping distance = 10,000 kb) between the single nucleotide polymorphisms (SNPs), these SNPs with the lowest P-values would be eventually reserved. (3) Under the presence of palindromic SNPs circumstances, we used allele frequencies to infer positive strand alleles. (4) During the comparing process, we checked the alleles against Genome Reference Consortium Human Build 38 and removed indeterminate and duplicated SNPs.

During this statistical analysis, several methods were performed to determine the causality between GM and IDB. IVW is a meta-analysis method used by MR to analyze the effects of multiple SNPs at multiple loci. The application premise of IVW is that all SNPs are valid IVs and completely independent of each other. Based on this, the unbiased of IVW results would be presented [41]. MR-Egger regression does not force the regression line to pass through the origin, allowing for targeted gene pleiotropy in the included instrumental variables. When the regression intercept is not zero and p for intercept < 0.05, it indicates the existence of gene pleiotropy [42]. The weighted median is the median of the distribution function obtained after all individual SNP effect size are sorted by weight. When at least 50% of the information comes from effective instrumental variables, weighted median can obtain robust estimates [43]. MR-PRESSO is a method of evaluating horizontal polymorphism using whole genome aggregated association statistical data. MR-PRESSO has three components, including MR-PRESSO overall test, MR-PRESSO outlier test and MR-PRESSO distortion test. Specific SNPs can be excluded by excluding outlier to obtain an estimate closer to the true value [44]. The weighted model and simple model also used to evaluate the effectiveness and correctness of MR calculations [45]. The simple mode takes the largest cluster of SNPs’ causal estimation, and the weighted mode assigns the weights to each SNP [46]. Finally, Cochran's Q statistic was applied to detect heterogeneity. If the Cochran's Q statistic test has statistical significance, it proves that the results were significant heterogeneity.

The leave-one-out method refers to omitting each SNP in turn, calculating the meta effect of the remaining SNPs, and observing whether the results have changed after removing each SNP. If the results change significantly after removing a certain SNP, it indicates that the potential heterogeneous SNPs have a significant impact on the results [47].

The scatter plot is a plot where the effect of the same SNP on exposure is placed on the horizontal axis, the effect on outcome is placed on the vertical axis, and the slope of the plot represents the causal effect of exposure factors on outcomes. It could visualize the causal effect of exposure on outcomes estimated under different parameter estimation methods [48].

To avoid weak instrument bias, the robustness of IVs could be assessed through F-statistic. We adopt formula F = \(\frac{{R}^{2}\times (N-2)}{(1-{R}^{2})}\) to calculate F-statistic. Among which, we could use R² to represent the degree of exposure explained by IVs with the formula R² = \(\frac{(2\times EAF\times (1 - EAF)\times {beta}^{2})}{(2\times EAF\times (1 - EAF) \times {beta}^{2}) + (2\times EAF\times (1 - EAF)\times N\times SE{(beta)}^{2}},\) where EAF represents the effect allele frequency, beta represents the effect estimate of the genetic variant in the exposure GWAS, SE(beta) represents the standard error of the beta and N represents sample size [46, 49, 50]. In general, when the corresponding F-statistic was > 10, significant weak instrumental bias could be reduced [51].

In the reverse MR analysis, the exposure is the IDB, whereas the outcome is GM. We selected IVs for each IDB phenotypes by using a much stricter threshold, where the significant threshold (p < 5 × 10^–8) [52, 53]. Additionally, the phenotypes, methods and other settings were consistent with those of forward MR. Under the significant threshold (p < 5 × 10^–8), no eligible SNP as IV was selected. A reverse MR analysis was not conducted at last owing to lack of SNPs (related to IDB) satisfying the assumption of the MR study.

Based on the previous selection criteria of IVs (p < 1 × 10^–5), a total of 2370 SNPs were anchored as IVs related to bacterial taxa from phylum to genus for IDB. For further information, Additional file 1: Table S1 is provided for reference.

As shown in Table 1, seven bacterial genera including Eubacterium rectale group, Bifidobacterium, Olsenella, Peptococcus, Prevotella7, Ruminiclostridium-6, RuminococcaceaeUCG003 were found to be associated with IDB in at least one MR method. MR methods found no relevance between bacterial taxa from phylum to family for IDB and detailed results are shown in Additional file 1: Table S2. Among seven bacterial genera, Eubacterium rectale group, Olsenella, Peptococcus and Ruminiclostridium-6 were supported by IVW analysis. Specifically, Eubacterium rectale group (OR = 1.87, 95% CI: 1.02–3.43, p = 4.20E−02), Olsenella (OR = 1.29, 95% CI: 1.02–1.64, p = 3.30E−02), Ruminiclostridium-6 (OR = 1.53, 95% CI: 1.08–2.14, p = 1.60E−02) had an anti-protective effect on IDB. Peptococcus (OR = 0.75, 95% CI: 0.60–0.94, p = 1.30E−02) had a protective effect on IDB. In addition, the F-statistics of seven bacterial genera selected at least one MR method were all above 10, eliminating the possibility of weak instrument bias (more detailed results are shown in Additional file 1: Table S3).

As displayed in Additional file 1: Table S4, sensitivity analysis was employed to identify the pleiotropy and heterogeneity. The results obtained by MR-Egger regression were as follows: Eubacterium rectale group (p = 0.90), Olsenella (p = 0.93), Peptococcus (p = 0.88), Ruminiclostridium-6 (p = 0.65), Prevotella7 (p = 0.87) and RuminococcaceaeUCG003 (p = 0.29), these six bacterial genera showed no horizontal pleiotropy. However, Bifidobacterium (p = 0.02) was removed due to the existence of pleiotropy (Table 2). Meanwhile, Cochran’s IVW Q test suggested Eubacterium rectale group (IVW: p = 0.23; MR Egger: p = 0.15), Olsenella (IVW: p = 0.87; MR Egger: p = 0.80), Peptococcus (IVW: p = 0.92; MR Egger: p = 0.88), Ruminiclostridium-6 (IVW: p = 0.76; MR Egger: p = 0.71) and RuminococcaceaeUCG003 (IVW: p = 0.28; MR Egger: p = 0.31) had no significant heterogeneity except Prevotella7 (IVW: p = 0.04; MR Egger: p = 0.03) (Table 2). Interestingly, although no significant pleiotropy and heterogeneity has been founded in RuminococcaceaeUCG003, RuminococcaceaeUCG003 was still filtered out under the IVW results (p = 0.12).

The leave-one-out plots (Fig. 2) and the scatter plots (Fig. 3) have shown the possible presence of potential outliers. In order to pursue the robustness of MR-Egger regression results, the method of MR-PRESSO method was used. The results were optimistic as no significant outliers were found (all p > 0.05, Table 2).

Finally, the main point is that the outcomes of IVW were assured after checking heterogeneity and pleiotropy. Therefore, Eubacterium rectale group, Olsenella, Peptococcus and Ruminiclostridium-6 were causally related to IDB.