A bidirectional two-sample Mendelian randomization study was utilized to explore the causal association between periodontitis and endometriosis. We also performed a two-sample MR analysis using the dataset from the UK Biobank (UKB) to validate the causality of periodontitis on endometriosis. As shown in Figure 1 , the study is based on three vital assumptions: firstly, there are strong associations between exposure and IVs. Secondly, IVs are not associated with the confounders. Thirdly, IVs should totally affect outcomes through exposure.

The present study was based on publicly available GWAS summary data, and ethical consent had been obtained in the original studies (21–25). All the data can be used without any restrictions.

We selected the IVs associated with periodontitis from three published GWAS datasets (23–25). All the participants were Europeans, including Dutch, German, and European-American. SNPs associated with the exposure at the genome-wide significance level (P < 5 × 10⁻⁸) were selected.

The datasets of endometriosis and its subgroups were obtained from FinnGen. The FinnGen study was initiated in 2017, which included more than half a million participants (age >18 years) who lived in Finland (26). The FinnGen GWAS summary data were extracted from the MRC-IEU database for endometriosis overall (8,828 cases, 68,969 controls), endometriosis of the ovary (3,231 cases, 68,969 controls), endometriosis of the pelvic peritoneum (2,953 cases, 68,969 controls), and also endometriosis of the rectovaginal septum and vagina (1,360 cases, 68,969 controls). Then, we also chose another dataset of endometriosis from UKB as the validation. UKB is a large prospective study, which aims to investigate the role of genetics, environment, and lifestyle in the causes of leading diseases. Its data come from 500,000 volunteers aged 40–69 in the United Kingdom (27). More details of exposure and outcome datasets are shown in Supplementary Table 1 .

The exposure data of endometriosis and its phenotypes were obtained from the same FinnGen study as mentioned above. The outcome data for the risk of periodontitis were obtained from a GWAS meta-analysis conducted by the Gene Lifestyle Interactions in Dental Endpoints (GLIDE) on Europeans (22). Seven primary studies were included in the analysis. Periodontitis was recognized via the Centers for Disease Control and Prevention/American Academy of Periodontology definitions (28), probing depth (29), or even self-reported (30). The details of the primary studies are shown in Supplementary Table 2 . In our study, we excluded the data of people with Hispanic/Latino ancestry. Data from European ancestry were collected from the original analysis.

SNPs that were strongly associated with exposure at the genome-wide significance level (P < 5 × 10⁻⁸) were selected as IVs. We conducted several quality-control measures to select qualified IVs. First, the independence of SNPs was assessed based on stringent criteria (r ² > 0.001; clumping window < 10,000 kb). Second, we used the PhenoScanner tool to check whether any of the selected SNPs were associated with potential confounders at the outcome. We set the threshold at genome-wide significance (P < 5 × 10⁻⁸) when using the PhenoScanner tool. Third, proxy SNPs were not used as IVs if SNPs were not in the 1000G reference panel. In addition, SNPs with a minor allele frequency of less than 0.01 should be excluded to avoid potential bias from the original GWAS due to the low confidence. The R ² and F statistics were also calculated to avoid bias from weak instruments. SNPs were excluded if the F statistics were less than 10.

For the analyses of periodontitis as exposure, we selected five SNPs as valid IVs ( Supplementary Table 3 ). In reversed analyses, during the selection of IVs associated with endometriosis, rs58502716 was not in the 1000G reference panel, and it was removed from the instruments. In addition, SNPs associated with endometriosis of the rectovaginal septum and vagina at a genome-wide significance level of P < 5 × 10⁻⁶ were selected in order to obtain enough IVs. rs76109112, rs139869063, rs200290589, and rs117783935 were eliminated as their F statistics were no more than 10. Finally, 10, 10, 5, and 13 SNPs were recognized as IVs associated with endometriosis overall, endometriosis of the ovary, endometriosis of the pelvic peritoneum, and endometriosis of the rectovaginal septum and vagina, respectively ( Supplementary Tables 4–7 ).

The inverse variance-weighted (IVW) method in the random-effects model was applied as a dominating approach to analyze the bidirectional causal relationship between periodontitis and endometriosis. Weighted median, MR-Egger, and simple median were also added as complementary approaches to reduce potential horizontal pleiotropy and bias from the IVW method (31). The weighted median method could generate an unbiased estimate if more than half of the weight from effective IVs (32). MR-Egger is more suitable if the P _intercept < 0.05 because it can provide the estimates after pleiotropy is corrected (33).

To test the validity of our findings, sensitivity analyses were conducted using weighted median and MR-Egger regression. Q-tests were performed in both IVW and MR-Egger regression to assess potential heterogeneity. The MR-Egger intercept was used to assess whether the included SNPs had potential horizontal pleiotropy. Weighted median provides consistent estimates when at least 50% of the information is from valid instrumental variables. Leave-one-out analyses were also utilized to estimate the causality and heterogeneity of the study (34). As for the pleiotropy analysis, the P-value of the MR-Egger intercept less than 0.05 is considered to suggest the pleiotropy of the study (35).

Additionally, the statistical power of each analysis was calculated via mRnd, an online calculator (https://shiny.cnsgenomics.com/mRnd/). All data were analyzed by R (version 3.1.5), combined with the R package “TwoSampleMR.” A P-value < 0.05 was considered to be statistically significant. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline ( Supplementary Table 8 ).

The results of MR analyses on the association between periodontitis and the risk of endometriosis are shown in Figure 2 . In the subgroup analyses, we observed that periodontitis was positively associated with endometriosis of the pelvic peritoneum (OR = 1.079, 95% CI = 1.016 to 1.146, P = 0.014). The replication of the main results using UKB data supported the null causality of periodontitis on endometriosis ( Supplementary Figure 1 ). No causal association was indicated between periodontitis and other subtypes of endometriosis. Suggestive estimates were observed in weight median, MR-Egger, and simple median methods similarly.

Although other robust MR methods did not manifest a significant association between periodontitis and endometriosis of the pelvic peritoneum, the sensitivity analysis showed no potential horizontal pleiotropy ( Table 1 ). Meanwhile, no heterogeneity existed in the IVW analyses, and none of the SNPs had a distinctive effect on estimates in the leave-one-out analyses ( Supplementary Figures 2–10 ).

The results of the MR analyses on the association between endometriosis and the risk of periodontitis are shown in Figure 3 . There was no causal effect of endometriosis or its subtypes on periodontitis. Consistent estimates were provided through other robust methods. As for the sensitivity analysis, no heterogeneity was found among SNPs, except for endometriosis of the rectovaginal septum and vagina (P = 0.048) ( Table 1 ). In view of this excess heterogeneity, a random-effects analysis was performed as the main method in this study (36). Furthermore, no evidence for pleiotropy was observed via MR-Egger analyses. No pivotal difference emerged in estimates after we singly removed SNP and repeated the MR analysis.

Our study had a power of > 80% to detect an effect of OR = 1.255 of periodontitis on the risk of endometriosis and a power of > 80% to detect an effect of OR = 1.394 of endometriosis on the risk of periodontitis, on a 5% significance level.

Oral bacteria such as Porphyromonas gingivalis and Fusobacterium nucleatum are prone to transmit into the placenta and induce intrauterine infection (41–43). Similarly, periodontal pathogens might locate in the pelvic peritoneum and then exacerbate endometriosis in situ. Porphyromonas gingivalis could activate peritoneal macrophages and trigger the production of interleukin-1 (IL-1) (44). Subsequent inflammatory cascades, together with increasing IL-1 expression, are pivotal throughout the development of endometriosis in the pelvic peritoneum (45). The frequency of Fusobacterium in endometriosis and endometrial tissues was significantly higher in endometriosis patients than in normal women. Animal studies also demonstrated that Fusobacterium from the oral cavity was a possible source of infecting the uterus through hematogenous transmission (46).

Periodontitis patients usually have a higher count of leukocytes, elevated levels of circulating C-reactive protein, as well as lower levels of hemoglobin and reduced numbers of red blood cells (47). Inflammasome, a multiprotein complex, is also considered as an intermedia. Periodontitis might increase Nlrp7 (a kind of inflammasome) and promote the progression of endometriosis (48). In a word, systemic inflammatory response is conjectured as a mechanism by which periodontitis causes systemic diseases.

Immune dysfunction has been certified as a key cause of endometriosis of the peritoneum. For example, a genomics single-cell RNA-sequencing study first reported that T-cell receptor-positive macrophages, increasing macrophages, and natural killer (NK) dendritic cells existed in the human peritoneal fluid of endometriosis (49). NK cell therapies are also a promising treatment for endometriosis (50). In addition, reducing the immune response induced by P. gingivalis is helpful in alleviating some metabolic diseases associated with periodontitis (51). There is reason to suspect that periodontitis causes endometriosis through autoimmune dysregulation.

Up to now, accepted treatments for endometriosis patients include continuous medical therapies and surgeries to remove pathological tissue (52). However, persistent medications suppress hormone levels and produce side effects, such as hair loss, acne, and mood changes, and operations would lead to inevitable relapse (53). This MR study revealed periodontitis as a novel point to manage endometriosis patients. Targeting a lower risk of periodontitis may therefore improve the prognosis of people with endometriosis of the pelvic peritoneum. In other words, supportive periodontal therapies and routine oral hygiene interventions may produce beneficial effects in treating women with endometriosis of the pelvic peritoneum. In the future, some clinical randomized controlled trials with more samples should be performed to verify the causal relationship. More research is necessary not only to clarify the underlying biological pathways but also to confirm therapeutic targets for endometriosis of the pelvic peritoneum.

The bidirectional MR study has some strengths. Firstly, genetic liability plays a crucial role in the course of periodontitis and endometriosis (18, 38). GWASs allowed larger sample sizes and less bias from population structure. Prevalence rates in the database were almost consistent with those reported in the literature. Hence, MR analysis is suitable for the inference of causality between both (54). Meanwhile, the MR analyses on both sides clarified the direction of causality. Secondly, endometriosis was further subdivided into three subgroups. An interesting finding is that periodontitis might be a cause of endometriosis of the pelvic peritoneum, which has not been realized before.

Some limitations should also be considered. Firstly, IVs had a deficient association with endometriosis of the rectovaginal septum and vagina, since the threshold of P-value was set to 5 × 10⁻⁶. Therefore, more convincing evidence is needed to prove the causality between endometriosis of the rectovaginal septum and vagina and periodontitis. Secondly, a weak power was found in each analysis, due to OR close to 1, so more participants were required to detect the causal effects in the future (55).