Ectopic pregnancy (EP) is defined as pregnancy occurring outside the uterine cavity (ACOG Practice Bulletin No. 193: Tubal Ectopic Pregnancy, 2018). Over 95% of EP cases occur in the fallopian tube (Marion and Meeks, 2012), which is defined as tubal pregnancy (TP). The incidence of EP in the USA is 1%–2%; however, some studies indicate that this may be underestimated due to the difficulty of tracking patients with EP who have been treated in an outpatient setting (Hendriks et al., 2020, Ucisik-Keser et al., 2021). TP remains the most common cause of maternal morbidity and mortality, deeply affecting patients’ physical and mental state and resulting in approximately 2.7% of pregnancy-related deaths (Creanga et al., 2017). Therefore, early diagnosis and intervention are essential for safeguarding pregnant women against these complications. Although some progress has been made, there are currently no standard measures in place for the early diagnosis of tubal implantation (Carusi, 2019; Chen et al., 2020; Sahin et al., 2021). Although the mechanism of TP is complex, the majority of relevant studies indicate that inflammation of the fallopian tube or the surrounding pelvic tissues, due to factors such as the presence of pelvic inflammatory disease (PID) or Chlamydia trachomatis (CT) infection (Xia et al., 2020), is central to TP development (Shao et al., 2009; Al-Azemi et al., 2010; Li et al., 2013). Therefore, the discovery of inflammation-associated biomarkers could help predict TP development. Moreover, a nested case-control study with a large samples size (N = 2,026) has revealed that abdominal pain [odds ratio (OR): 1.16, 95% confidence interval (CI): 0.92–1.48] and bleeding (OR:1.34, 95% CI: 1.04–1.78) are risk factors for EP (Xia et al., 2020). The incidence of EP among pregnant women experiencing vaginal bleeding or abdominal pain (or both) in the first-trimester of pregnancy has been shown to be as high as 18% (Barnhart et al., 2006). Therefore, in the present study, we focused our attention on symptomatic pregnant women in the early stages of pregnancy.

The human commensal microbiome has evolved to coexist with the human genome and contributes to the maintenance of human health. Recently, the Human Microbiome Project (HMP) has shown that vaginal microbiota (VM) account for approximately 9% of the whole human microbiome (Huttenhower et al., 2012; Methé́ et al., 2012b; Gonzalez et al., 2014). While the most abundant vaginal bacteria are Lactobacilli, other genera, including Gardnerella, Prevotella, Bifidobacterium, Atopobium, Megasphaera, Sneathia, and Anaerococcus also play important roles as part of the vaginal microbiome (Huttenhower et al., 2012a; Methé́ et al., 2012b). 16S rRNA gene sequencing has been widely used to assess bacterial diversity and composition (Woo et al., 2008; Siddiqui et al., 2011; Jiao et al., 2021). This tool can help clinicians to find uncultivable bacteria and evaluate new pathogens that could serve as biomarkers in various disease settings (Yoshimura et al., 2011). As a result, large efforts have been made to assess the bacterial composition and diversity of the healthy versus abnormal vaginal microbiome (Franasiak and Scott, 2015; Moreno and Simon, 2019). In addition, vaginal microbial dysbiosis was shown to be associated with several female reproductive disorders, including premature delivery (Kumar et al., 2021), premature rupture of membranes (Amabebe et al., 2016), infertility (Sharma et al., 2014), PID (Rosca et al., 2020), in vitro fertilization (IVF) failure (Karaer et al., 2021), and recurrent miscarriage (Jiao et al., 2021). It is therefore of outmost importance to understand the vaginal microbiological characteristics of pregnant women, which are especially poorly described early on in pregnancy when fetal location is unknown, and represents a critical period for the early diagnosis of TP.

In this study, we aimed to assess the potential association between vaginal microbial composition and the incidence of TP among symptomatic pregnant women who were initially considered to have a pregnancy of unknown location (PUL), in China. In order to enhance the analytical power of our research, we employed both the 16S rRNA gene sequencing and machine learning methods.

TP is still the major cause of hemorrhage-related pregnancy-associated death to date (Segata et al., 2011), and a recent study revealed that the global incidence of TP was on the rise (Goller et al., 2018). The underlying risk factors associated with TP are complex, including previous incidences of PID, EP, tubal surgery, infertility, and Chlamydia trachomatis infection (ACOG Practice Bulletin No. 193: Tubal Ectopic Pregnancy, 2018; Marion and Meeks, 2012; Rana et al., 2013). Numerous studies have indicated that chronic inflammation in the tissues surrounding the female reproductive tract plays a crucial role in the incidence of TP (Marion and Meeks, 2012; Rana et al., 2013; ACOG Practice Bulletin No. 193: Tubal Ectopic Pregnancy, 2018). However, 50% of patients with TP have no definitive risk factors (Ankum et al., 1996; Barnhart et al., 2006; ACOG Practice Bulletin No. 193: Tubal Ectopic Pregnancy, 2018), a statistic which continues to confound gynecologists.

Recently, an increasing number of studies have found that microbes play an important role in human physiology and pathology alike (Filardo et al., 2017; Di Pietro et al., 2018). As an important component of the human microbiome, vaginal microbiota play a critical role in female reproduction and fertility (Franasiak and Scott, 2015; Moreno and Simon, 2019). The development of advanced sequencing analysis has enabled us to expand our knowledge of vaginal microbial composition in order to delineate the relationship between microbial perturbations in pregnancy and adverse pregnancy outcomes. However, few studies have focused on examining the characteristics of the vaginal microbiome in patients with TP. Further characterizing the vaginal microbiome may help gynecologists to identify which women in the PUL period have a higher risk of developing TP. Several studies have shown that vaginal microbiota became less diverse, more stable, and were dominated by Lactobacillus in women with normal pregnancies (DiGiulio et al., 2015). Consistent with the results of our studies, some findings have also indicated that pregnancy complications were associated with the increased relative abundance of anaerobes, including Gardnerella, Prevotella, and Atopobium (Fettweis et al., 2019; Siena et al., 2021). Moreover, the most frequently described harmful change affecting the vaginal microbiome in women of reproductive age, bacterial vaginosis (BV), is characterized by a decrease in relative Lactobacillus abundance and an increase in anaerobic bacteria, primarily Gardnerella vaginalis (Fredricks et al., 2005). The presence of BV in early pregnancy was confirmed to be related to several adverse pregnancy outcomes, including preterm delivery (Petricevic et al., 2014), premature rupture of membranes (Purwar et al., 2001), recurrent miscarriage (Jiao et al., 2021), spontaneous abortion (Isik et al., 2016), and PID (Bagnall and Rizzolo, 2017). Although, the potential link between Gardnerella abundance and the incidence of TP has not been fully elucidated, we propose the following possible mechanism.

Gardnerella vaginalis infection is a large contributor of BV (Cheah et al., 2020), particularly in women of reproductive age (Wong et al., 2018). Approximately 40% of women with BV have no obvious symptoms and remain untreated (Klebanoff et al., 2004). Therefore, the harmful effects of Gardnerella infection are chronic and easily overlooked. It has also been reported that Gardnerella itself is associated with the development of several genital tract conditions, including endometriosis (Khan et al., 2016) and PID (Brunham et al., 2015). These findings support the hypothesis that Gardnerella may ascend into the upper reproductive tract and, thus, affect the incidence of TP. In fact, some studies have also supported this potential underlying mechanism. In women with normal pregnancies, hormonal alterations favor Lactobacillus proliferation, leading to increased level of antibacterial substances (e.g., H₂O₂ and bacteriocins) and lactic acid, which maintains the vaginal pH at 3.5–4.5 (Boskey et al., 1999; Greenbaum et al., 2019). However, when the relative abundance of Gardnerella increases, the neuraminidase enzymes generated by Gardnerella spp. can damage the membrane of cervical epithelial cells and weaken the barrier between the vagina and cervix, thus, raising the risk of cervical infections (Haahr et al., 2016). Nicole et al. reported that Gardnerella vaginalis stimulated group B Streptococcus (GBS) vaginal colonization, leading to ascending uteroplacental infection in pregnant mice (Gilbert et al., 2021). In addition, Gardnerella was shown to activate the local proinflammatory cytokine response and the innate immune system and lead to cervical remodeling (Sierra et al., 2018; Zheng et al., 2019; Florova et al., 2021). Fan and colleagues found that pathogenic bacteria, including Gardnerella, could trigger an imbalance in immune tolerance at the maternal-fetal interface through the perturbation of regulatory chemokine networks (Fan et al., 2020).

Our study revealed the potential association between vaginal microbiota and pregnancy location. However, translating these complex ecological metrics into the clinical setting is a challenge. Classification of vaginal microbiota effectively simplifies the complex biological dataset and may be helpful in epidemiological investigations and in disease diagnosis (Kindinger et al., 2016; Ravel et al., 2021). Several studies have indicated that the vaginal microbiota can be clustered by community state type (CST), which was developed by characterizing the diversity of bacterial taxa in the vaginal microbiome of ~400 multiethnic women of reproductive age (Ravel et al., 2011; Gajer et al., 2012). Four of the CSTs are dominated by Lactobacillus, including Lactobacillus crispatus (CST I), Lactobacillus gasseri (CST II), Lactobacillus iners (CST III), and Lactobacillus jensenii (CST V) (Ravel et al., 2011; Drell et al., 2013). Among these, only CST IV comprises many subsets of anaerobic bacteria, including Gardnerella, Atopobium, Prevotella, Mobiluncus, and Sneathia. Therefore, CSTs have their limitations, one of which is that most of the CSTs are built around the presence of four definitive dominant Lactobacillus species. Crucially, approximately 25% of the women sampled clustered into CST IV (Ravel et al., 2011; Drell et al., 2013), therefore masking the detrimental effect of Gardnerella. Classification according to the relative abundances at the genus level could classify samples more definitively and may more effectively reveal associations between the vaginal microbiome and pregnancy outcome (Al-Memar et al., 2020). Robinson et al. pointed out that instead of clustering by classification, machine-learning algorithms could overcome such disadvantages (Robinson et al., 2016). Machine learning could effectively improve bioinformatics analysis for making the microbial-community groups independent of samples and supporting comparability between studies (LeCun et al., 2015). Nevertheless, few studies have applied machine learning to analyze vaginal microbiota. Extreme Gradient Boosting (XGBoost) is an efficient algorithm used for machine learning (Taylor et al., 2018; Ye et al., 2018) that has been widely applied to predict the onset of disease and even for providing genetic counseling for individuals. The predictive testing process employed by the XGBoost model is beneficial for evaluating the expecting onset, which is helpful for optimizing future medical plans (Peng et al., 2021). Therefore, we integrated LEfSe analysis and the XGBoost algorithm to efficiently process large-scale microbiological data, with the aim of predicting TP.

Few studies have compared the vaginal microbiota between IUP and TP groups. The population we included consisted of symptomatic pregnant women with PUL, who were at high-risk for TP and required special clinical attention. Our study revealed a significant difference in the relative abundance of Gardnerella in the vaginal microbiome of women with TP and IUP, identifying Gardnerella as a potential biomarker for predicting the incidence of TP in early pregnancy. Moreover, we generated a model to reliably predict the incidence of TP in women with PUL based on 16S rDNA gene sequencing and machine learning. Nevertheless, our study still has the following limitations. Firstly, we included a well-defined study population, so the results would be limited to pregnant women with abdominal pain and/or uterine bleeding. Secondly, because the composition of vaginal microbiota can be affected by ethnicity (Fettweis et al., 2019; Dunlop et al., 2021), further studies should be performed to verify whether the results of our study are applicable to women with symptomatic early pregnancies outside of China. Thirdly, our study involved association analysis research and focused specifically on describing the association between vaginal microbiota and the incidence of TP. Although we found that the relative abundance of Gardnerella could be a potential and valuable biomarker to identify TP, the underlying mechanism needs to be further explored.

Compared with the IUP group, the vaginal microbiome of TP patients exhibited higher microbial diversity and a higher relative abundance of Gardnerella, which could represent a potential diagnostic biomarker for TP. Our results provide new insights into the intricate relationship between the vaginal microbiome in early pregnancy and incidence of TP.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: NCBI SRA BioProject, Accession No. PRJNA737055.

The studies involving human participants were reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Guangzhou University of Chinese Medicine (No. ZYYECK2017-060). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

JG, GD, and SL designed and founded the study. YG, XC, XH, XX, YG, XR and XL recruited participants. SC, YG, XC, XH, XX and YG collected clinical data and samples as well as analyzed and interpreted the data. SC and YZ generated the figures and tables. SC wrote the first draft, which was further developed by YZ. All authors have read the final manuscript and have agreed to be held accountable for its content.

The study was funded by the National Natural Science Foundation of China (grant number 81774358), the Guangzhou University of Chinese Medicine (grant numbers XK2019016 and 2019IIT33), the National administration of Traditional Chinese Medicine (National Chinese Medicine People’s Education Development (2018) No. 12), and Department of Finance of Guangdong province (grant number 2020B1111100003).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.