Investigation of the Endometrial Microbiome in Recurrent Pregnancy Loss Individuals: Microbial Imbalance and Network Fragility.

Recurrent pregnancy loss (RPL) remains one of the most challenging conditions in reproductive medicine. It affects up to 5% of couples attempting conception 1 and is defined by the European Society of Human Reproduction and Embryology (ESHRE) as the occurrence of at least two pregnancy losses prior to 24 weeks of gestation. 2 Although several risk factors have been identified, including uterine abnormalities, endocrine dysfunction, autoimmune conditions, thrombotic disorders, and chromosomal anomalies, the etiology of RPL remains largely heterogeneous and incompletely understood. Among the proposed factors, infections and imbalances in the reproductive tract microbiota have been implicated in certain cases. 3 Pathogenic microorganisms and their endotoxins may elicit abnormal inflammatory responses in the endometrium, impair decidualization, and disrupt immune regulation at the maternal–fetal interface, potentially leading to abnormal embryonic development or early pregnancy loss. 4 However, no conclusive evidence supports a direct causal role of a specific microbial species in RPL, 5 likely due to complex interspecies interactions and host-specific factors that complicate such investigations. High-throughput sequencing technologies, particularly 16S rRNA sequencing, now allow more detailed characterization of microbial communities in human tissues and offer new perspectives on these interactions. 6 Traditionally, the uterus was considered a sterile environment. This notion was challenged by Mitchell et al, who used qPCR and 16S rRNA sequencing to identify a complex microbial community within the uterine cavity, including Lactobacillus sp., Acinetobacter sp., and Pseudomonas sp. 7 This discovery has prompted increasing interest in the role of the endometrial microbiota in fertility. 8 Studies have shown that women with a Lactobacillus sp.-dominant microbiota (>90%) have significantly higher implantation, pregnancy, and live birth rates during IVF-ET than those with non-dominant microbiota. 9 Recent reviews emphasize a growing body of evidence linking endometrial microbial dysbiosis to adverse reproductive outcomes, particularly RPL. 10 Additionally, one study reported that elevated Ureaplasma urealyticum levels in the endometrium were associated with increased risks of miscarriage and preterm birth. 11 Despite these advances, the comprehensive characteristics of the endometrial microbiota in RPL, including diversity patterns and community interactions, remain poorly understood. The microbiota is not a random collection of bacteria but plays a critical role in regulating immune homeostasis and maintaining the stability of the endometrial microenvironment through intricate ecological networks. Therefore, this study aimed to characterize the endometrial microbiome in patients with RPL, with specific objectives to investigate microbial alterations, identify potential microbial biomarkers, and construct ecological networks of microbial interactions.

A total of 122 patients (89 in the RPL group and 33 in the control group) were included in this study. No statistically significant differences were found in age, BMI, AMH levels, or parity. However, compared with the control group, the RPL group had significantly higher gravidity and number of abortions ( Table 1 ). Table 1 Patient Baseline Characteristics Group Total Control RPL P value Enrolled Population 122 33 89   Age (y) 32.99±3.62 33.03±3.60 32.98±3.65 0.89   BMI (kg/m²) 21.64±2.21 21.79±2.50 21.59±2.10 0.78   AMH (ng/mL) 3.16±2.36 2.24±2.13 2.76±2.45 0.381   Gravidity 2.53±1.47 0.85±0.57 3.15±1.18 <0.001   Parity 0.29±0.49 0.18±0.46 0.32±0.49 0.09   Abortions 2.22±1.31 0.64±0.49 2.81±0.10 <0.001 Abbreviations : RPL, Recurrent pregnancy loss; BMI, Body Mass Index; AMH, Anti-Mullerian Hormone. Patient Baseline Characteristics Abbreviations : RPL, Recurrent pregnancy loss; BMI, Body Mass Index; AMH, Anti-Mullerian Hormone. Before comparing microbial diversity, rarefaction curves were plotted for all samples ( Figure S2a ). The curves showed that the number of species approached saturation as the number of reads increased, indicating sufficient sequencing depth. Alpha diversity measures species richness and evenness within individual samples. The rarefaction curves for both groups were highly similar ( Figure S2b ). Furthermore, no statistically significant differences were observed between the RPL and control groups in several alpha diversity indices, including Chao1, ACE, Shannon, Simpson, and Pielou indices ( Figure 1 ). Figure 1 Alpha Diversity Indices in the Control and RPL Groups. Split violin plots overlaid with boxplots showing the distribution and variation of six alpha diversity indices (Chao1, ACE, Shannon, Simpson, and Pielou indices) in the control and RPL groups. Statistical comparisons between groups were performed using the Wilcoxon rank-sum test. P -values are shown for each pair. RPL: recurrent pregnancy loss. Alpha Diversity Indices in the Control and RPL Groups. Split violin plots overlaid with boxplots showing the distribution and variation of six alpha diversity indices (Chao1, ACE, Shannon, Simpson, and Pielou indices) in the control and RPL groups. Statistical comparisons between groups were performed using the Wilcoxon rank-sum test. P -values are shown for each pair. RPL: recurrent pregnancy loss. Beta diversity assesses differences in species composition between samples or communities. Among the taxa identified, 48% were shared between the two groups, 12% were unique to the control group, and 40% were unique to the RPL group, indicating greater taxonomic uniqueness in the RPL group ( Figure 2a ). Figure 2 Beta Diversity and PCoA Analysis of the Control and RPL Groups. ( a ) Venn diagram showing the number of taxa identified in the two groups. ( b ) Beta diversity comparison based on Euclidean distance. ( c ) PCoA plot based on Euclidean distance. Abbreviations : PCoA, principal coordinate analysis; RPL, recurrent pregnancy loss. Beta Diversity and PCoA Analysis of the Control and RPL Groups. ( a ) Venn diagram showing the number of taxa identified in the two groups. ( b ) Beta diversity comparison based on Euclidean distance. ( c ) PCoA plot based on Euclidean distance. Euclidean distances were calculated to measure variation in microbial composition in multidimensional space ( Figure S3 ). In the control group, there were no statistically significant differences in Euclidean distances (P = 0.27). In contrast, the RPL group showed borderline significance (P = 0.052), indicating some heterogeneity. The overall variance in Euclidean distances between groups was statistically significant (P = 0.039), suggesting distinct microbial community composition ( Figure 2b ). Principal coordinate analysis (PCoA) revealed that PCo1 and PCo2 accounted for 27.07% and 11.88% of the variance, respectively. Although some overlap was noted in the PCoA plot, a trend toward distinct separation was evident along the PCo1 axis, indicating differences in microbial community composition between the groups ( Figure 2c ). At the phylum level, Firmicutes and Proteobacteria were the dominant phyla. The proportion of Actinobacteria was significantly increased in the RPL group, whereas the abundance of Bacteroidetes was notably reduced ( Figure 3a ). Figure 3 Taxonomic Composition at Different Levels Based on Relative Abundance. Stacked bar plots showing the relative abundance of microbial communities at six taxonomic levels—from phylum to species—in the control and RPL groups. Panels ( a – f ) represent the taxonomic levels: phylum, class, order, family, genus, and species, respectively. Abbreviation : RPL, recurrent pregnancy loss. Taxonomic Composition at Different Levels Based on Relative Abundance. Stacked bar plots showing the relative abundance of microbial communities at six taxonomic levels—from phylum to species—in the control and RPL groups. Panels ( a – f ) represent the taxonomic levels: phylum, class, order, family, genus, and species, respectively. At the class level, Bacilli, Actinobacteria , and Betaproteobacteria accounted for more than 80% of the microbial population. Among these, Actinobacteria was significantly elevated in the RPL group ( Figure 3b ). Lactobacillales  and  Actinomycetales were more abundant in the control group, whereas Bifidobacteriales was more abundant in the RPL group. In addition, the RPL group contained a greater variety of other taxa ( Figure 3c ). At the family level, Lactobacillaceae was more abundant in the control group, whereas the abundance of Bifidobacteriaceae was higher in the RPL group ( Figure 3d ). Due to limitations in taxonomic annotation at the species level, results at the genus and species levels were largely consistent. Lactobacillus was the dominant genus in the endometrial microbiome but was notably reduced in the RPL group. In contrast, Gardnerella and Streptococcus were significantly increased in the RPL group ( Figure 3e and f ). Regarding absolute microbial abundance, all taxonomic levels showed higher absolute abundance in the RPL group than in the control group, suggesting an overall increase in microbial load in the endometrium of patients with RPL, possibly linked to the overgrowth of specific pathogens ( Figure S4 ). While stacked bar plots comparing relative abundances across taxonomic levels can miss low-abundance genera and lack statistical validation, we addressed this by identifying differential taxa across multiple levels, filtered for significance at P < 0.05, and quantified differences using linear discriminant analysis (LDA) scores ( Figure 4 ). At the family level, Cryomorphaceae, Streptococcaceae , and Xanthomonadaceae were identified as differential taxa. Among them, Streptococcaceae and Xanthomonadaceae were significantly elevated in the RPL group, while Cryomorphaceae was enriched in the control group. Figure 4 Differentially Abundant Taxa Between the Control and RPL Groups Across Multiple Taxonomic Levels. ( a ) Cladogram displaying the phylogenetic structure of the endometrial microbiota, organized hierarchically from the center (kingdom) to the outermost level (species), passing through phylum, class, order, family, and genus. The branches are colored based on taxonomic assignment at the phylum level to allow visual differentiation of microbial lineages. Differentially abundant taxa between the control and RPL groups were identified using LEfSe and are labeled (a–s) on the tree. These labels indicate taxa that were statistically significant ( P < 0.05), and their positions in the tree show the taxonomic level at which differences emerge. The outermost layer of the cladogram displays bars representing the effect size (log10-transformed LDA scores) of each taxon. Green bars indicate taxa enriched in the RPL group, while yellow bars represent taxa enriched in the control group. ( b ) Comparative plots of differentially abundant taxa across multiple taxonomic levels between the control and RPL groups. The left panel shows the relative abundance of each taxon in both groups, while the right panel presents the corresponding effect size, measured as the log10-transformed LDA score. This score reflects the magnitude of difference between groups. Dots represent significantly different taxa (P < 0.05), with green indicating enrichment in the RPL group and yellow indicating enrichment in the Control group. Taxa not classified at the species level are labeled as “un” and are annotated using the nearest higher-level taxonomy. Abbreviations : LDA, linear discriminant analysis; LEfSe, linear discriminant analysis effect size; RPL, recurrent pregnancy loss. Differentially Abundant Taxa Between the Control and RPL Groups Across Multiple Taxonomic Levels. ( a ) Cladogram displaying the phylogenetic structure of the endometrial microbiota, organized hierarchically from the center (kingdom) to the outermost level (species), passing through phylum, class, order, family, and genus. The branches are colored based on taxonomic assignment at the phylum level to allow visual differentiation of microbial lineages. Differentially abundant taxa between the control and RPL groups were identified using LEfSe and are labeled (a–s) on the tree. These labels indicate taxa that were statistically significant ( P < 0.05), and their positions in the tree show the taxonomic level at which differences emerge. The outermost layer of the cladogram displays bars representing the effect size (log10-transformed LDA scores) of each taxon. Green bars indicate taxa enriched in the RPL group, while yellow bars represent taxa enriched in the control group. ( b ) Comparative plots of differentially abundant taxa across multiple taxonomic levels between the control and RPL groups. The left panel shows the relative abundance of each taxon in both groups, while the right panel presents the corresponding effect size, measured as the log10-transformed LDA score. This score reflects the magnitude of difference between groups. Dots represent significantly different taxa (P < 0.05), with green indicating enrichment in the RPL group and yellow indicating enrichment in the Control group. Taxa not classified at the species level are labeled as “un” and are annotated using the nearest higher-level taxonomy. In the phylogenetic tree shown in Figure 4a , 16 different taxa were identified at the genus and species levels, with several genera originating from the same family branches. Differential microbiota analysis revealed higher abundance of several potential pathogens in patients with RPL, including Streptococcus, Staphylococcus, Pseudomonadaceae, Campylobacter , and Porphyromonas . These genera are well-known for their association with various infectious diseases, and their enrichment may suggest a link to adverse reproductive outcomes. Three machine learning methods were used for feature selection to identify RPL-associated genera. Figure 5a and b show the selection process in the LASSO model, where increasing the Lambda value reduced the coefficients of less important genera to zero, yielding a refined set of key genera. In the SVM model, the lowest error rate was observed with 20 selected features ( Figure 5c ). In the RF model, increasing the number of decision trees reduced the error rate ( Figure 5d ). The top five important genera identified were Comamonas, Streptococcus, Bacillus, Lactobacillus , and Gardnerella ( Figure 5e ). Figure 5 Identification of RPL-Associated Key Genera Through Multiple Machine Learning Methods. ( a and b ) Lasso regression was used to screen and identify significant genera. ( c ) SVM 5-fold cross-validation error rate plot. ( d ) RF cross-validation error rate plot. ( e ) Top 20 genera identified by RF, ranked by importance. ( f ) Venn diagram showing the overlap of important genera identified by three machine learning methods. ( g ) ROC curve of the regression model based on three key genera. ( h ) Nomogram of the model constructed from the three key genera. Abbreviations : Lasso, least absolute shrinkage and selection operator; RF, random forest; ROC, receiver operating characteristic; RPL, recurrent pregnancy loss; SVM, support vector machine. Identification of RPL-Associated Key Genera Through Multiple Machine Learning Methods. ( a and b ) Lasso regression was used to screen and identify significant genera. ( c ) SVM 5-fold cross-validation error rate plot. ( d ) RF cross-validation error rate plot. ( e ) Top 20 genera identified by RF, ranked by importance. ( f ) Venn diagram showing the overlap of important genera identified by three machine learning methods. ( g ) ROC curve of the regression model based on three key genera. ( h ) Nomogram of the model constructed from the three key genera. The three models showed considerable overlap in the selected genera, confirming the robustness of the results. Chryseobacterium, Fusobacterium , and Streptococcus were consistently identified as key genera ( Figure 5f ). A logistic regression model based on these three genera achieved an AUC of 0.762, compared with AUC values of 0.645, 0.572, and 0.644 for Streptococcus, Fusobacterium , and Chryseobacterium , respectively, suggesting that the combined model significantly improved RPL prediction ( Figure 5g ). The nomogram illustrated the contributions of Streptococcus, Fusobacterium , and Chryseobacterium to RPL. Streptococcus and Chryseobacterium were identified as risk factors, while Fusobacterium appeared protective. Genus values were mapped to scores, summed, and used to estimate individual RPL risk, thereby quantifying the link between genus abundance and RPL and offering a foundation for personalized treatment ( Figure 5h ). The microbiome is not merely a sum of different microbial groups; microbial interactions are critical in shaping the endometrial microenvironment and local immune response. Interaction networks constructed using the top 300 genera by abundance revealed a more complex microbial interaction network in the control group than in the RPL group. In the control group, both cooperative and antagonistic interactions were observed, while only a few cooperative interactions were present in the RPL group ( Figure 6a ). Figure 6 Construction of the Endometrial Microbiota Network. ( a ) Microbial interaction network constructed using the top 300 genera ranked by relative abundance. Each node represents a genus, with node colors indicating taxonomic affiliation at the phylum level. Edges represent microbial interactions: brown lines indicate positive (cooperative) associations, and blue lines indicate negative (competitive or antagonistic) associations. ( b ) Topological roles of nodes in the microbial co-occurrence network. Each dot represents an OTU positioned by its within-module connectivity (Z-score) and participation coefficient. Nodes are classified into four categories—peripherals, connectors, module hubs, and network hubs—based on their topological roles, with background shading indicating each role. The control group shows more module hubs and connectors than the RPL group, reflecting higher network complexity and connectivity. ( c ) Natural connectivity under progressive random node removal. The control network shows a slower decline in connectivity compared to the RPL network, indicating greater structural resilience. ( d ) Robustness analysis based on random node removal. Species loss accelerates more rapidly in the RPL network, reflecting lower ecological stability. ( e ) Robustness analysis after targeted removal of module hubs. Disruption of key taxa leads to a sharp collapse in the RPL network, whereas the control network remains more stable. Abbreviation : RPL, recurrent pregnancy loss. Construction of the Endometrial Microbiota Network. ( a ) Microbial interaction network constructed using the top 300 genera ranked by relative abundance. Each node represents a genus, with node colors indicating taxonomic affiliation at the phylum level. Edges represent microbial interactions: brown lines indicate positive (cooperative) associations, and blue lines indicate negative (competitive or antagonistic) associations. ( b ) Topological roles of nodes in the microbial co-occurrence network. Each dot represents an OTU positioned by its within-module connectivity (Z-score) and participation coefficient. Nodes are classified into four categories—peripherals, connectors, module hubs, and network hubs—based on their topological roles, with background shading indicating each role. The control group shows more module hubs and connectors than the RPL group, reflecting higher network complexity and connectivity. ( c ) Natural connectivity under progressive random node removal. The control network shows a slower decline in connectivity compared to the RPL network, indicating greater structural resilience. ( d ) Robustness analysis based on random node removal. Species loss accelerates more rapidly in the RPL network, reflecting lower ecological stability. ( e ) Robustness analysis after targeted removal of module hubs. Disruption of key taxa leads to a sharp collapse in the RPL network, whereas the control network remains more stable. The microbial network in the control group exhibited distinct modular structures, with four module hubs and three connectors. In contrast, the RPL group lacked similar key nodes, indicating reduced network function ( Figure 6b ). Network stability analysis showed that node connectivity was higher in the control group, and the network remained stable even after extensive node removal. Conversely, the RPL group exhibited low connectivity that declined sharply with node removal, suggesting that its microbial network was more fragile and susceptible to collapse ( Figure 6c ). As genera were gradually removed, the number of remaining taxa declined in both groups, but more sharply in the RPL group ( Figure 6d ). Removal of module hubs further accelerated species loss in the RPL network, underscoring its structural vulnerability ( Figure 6e ). Within the top 50 genera, several highly connected modules were identified, with Bacillus, Pontibacillus , and Pelomonas as prominent central nodes. Lactobacillus also occupied a central role in the uterine microbial network, highlighting its importance in maintaining a healthy uterine environment. In the RPL group, the network was sparse, and Lactobacillus had no observable interactions with other genera, indicating a diminished protective role in patients with RPL ( Figure S5 ).

This retrospective cohort study was conducted using clinical and microbiome data from patients who underwent the EndoMetrial Microbiome Assay (EMMA) at the Reproductive Medicine Clinic of the University of Hong Kong–Shenzhen Hospital between June 1, 2023, and March 31, 2024. Patients were categorized into two groups based on clinical history: the RPL group and the control group. RPL was defined as two or more pregnancy losses occurring before 24 weeks of gestation, regardless of whether conception was spontaneous or assisted. Patients were excluded if they met any of the following criteria: (1) female age ≥ 40 years; (2) body mass index (BMI) outside the range of 18.5–30 kg/m²; (3) chromosomal abnormalities in either partner; and (4) incomplete clinical or microbiome data. Patients with pregnancy loss due to fetal chromosomal abnormalities confirmed by chorionic villus sampling were also excluded from the RPL group. Of the 178 patients initially screened, 122 met the eligibility criteria and were included in the final analysis: 89 in the RPL group and 33 in the control group. A detailed flowchart illustrating participant selection is provided in Figure S1 . The study protocol received approval from the Ethics Committee of the University of Hong Kong–Shenzhen Hospital (Ethical Approval No. [2024]375). Clinical data were extracted from the hospital’s electronic medical records. Variables included age (years), BMI (kg/m²), anti-Müllerian hormone (AMH, ng/mL), gravidity, parity, and the number of prior spontaneous abortions. To protect patient confidentiality, all data were anonymized and coded before analysis, in accordance with ethical guidelines. Endometrial samples were collected during the mid-secretory phase of the menstrual cycle to ensure consistency in hormonal milieu and endometrial receptivity. A standardized protocol was followed to minimize contamination and variability. The vaginal area was first disinfected with povidone-iodine and then thoroughly rinsed with sterile saline to remove any residual antiseptics. Cervical discharge was absorbed using a sterile dry cotton swab. A sterile, single-use endometrial sampling device (Biopsy-Mistogy Tube; Shanghai Jiabao Medical Health Technology Co., Ltd., China) was then gently inserted into the uterine fundus. After withdrawing the inner core to generate negative pressure, the catheter was rotated and maneuvered to aspirate endometrial tissue. Care was taken to avoid contact with the vaginal wall during insertion and withdrawal. All procedures were conducted under strict aseptic conditions. The EMMA was performed to profile the endometrial microbiota and assess whether the uterine environment was conducive to implantation and pregnancy. EMMA utilizes next-generation sequencing (NGS) of the 16S rRNA gene to identify and quantify bacterial DNA with high precision. Genomic DNA was extracted from endometrial tissue using the QIAsymphony DSP DNA Mini Kit (Cat. No. 937236), and DNA concentrations were measured via spectrophotometry (MultiSkan GO, Thermo Fisher). Sequencing libraries were prepared using the Ion Plus Fragment Library Kit (Thermo Fisher), and sequencing was conducted on the Ion GeneStudio S5 Prime platform (Thermo Fisher). Raw reads were quality-controlled and processed using the QIIME2 pipeline. Low-quality sequences were filtered, trimmed, and aligned to the SILVA 16S rRNA reference database for taxonomic annotation. Downstream analysis of 16S rRNA sequencing data was performed using the MicrobiotaProcess R package. This included assessment of α-diversity and β-diversity, sample distance calculations, comparisons of microbial composition across taxonomic levels, and identification of differentially abundant taxa. 12 To identify key genera significantly associated with RPL, three machine learning algorithms were applied: support vector machine (SVM) using the e1071 package, least absolute shrinkage and selection operator (LASSO) regression via the glmnet package, and random forest (RF) classification via the randomForest package. Genera consistently identified across models were used to construct a binary logistic regression model, and a nomogram was generated to visualize predictive outcomes. Model performance was assessed using receiver operating characteristic (ROC) curve analysis. Additionally, microbial co-occurrence networks were constructed and compared between groups using the ggClusterNet package. 13 Baseline characteristics were compared using the Mann–Whitney U -test. All analyses were conducted using R software (version 4.4.1).

This study has several strengths. It provides a detailed characterization of the endometrial microbiome in patients with RPL, encompassing species diversity, microbial abundance, and differential microbiota analysis. The innovative application of machine learning facilitated the identification of key genera strongly associated with RPL. Furthermore, microbial network analysis revealed structural fragility within the endometrial microbiota of patients with RPL, offering new insights into potential microbe-related mechanisms. However, this study also has limitations. As a retrospective analysis, it lacks prospective cohort recruitment, which may introduce selection bias and limit the generalizability of the findings. To mitigate this, strict and consistent inclusion and exclusion criteria were applied. In addition, due to the retrospective design, no environmental or vaginal control samples were collected. Although all samples were obtained using standardized aseptic techniques to minimize contamination, the absence of negative controls remains a limitation. The sample size also limits the ability to explore potential associations between endometrial microbiota features and the occurrence of RPL, underscoring the need for further investigation in large-scale, multicenter studies. Lastly, external factors such as environment, diet, and lifestyle may influence the endometrial microbiome. Future studies should incorporate these variables and employ integrated multi-omics approaches to better elucidate the microbial features and potential mechanisms linked to RPL.

This study presents an in-depth profiling of the endometrial microbiome in women with RPL, revealing significant differences compared to healthy controls. Protective bacteria, such as Lactobacillus spp., were reduced in patients with RPL, whereas pathogenic bacteria, including Gardnerella spp., Streptococcus spp., and Staphylococcus spp., exhibited increased diversity and abundance. Using machine learning, we identified core genera associated with RPL, particularly Streptococcus spp., Chryseobacterium spp., and Fusobacterium spp., which may contribute to miscarriage. Additionally, the endometrial microbial network in patients with RPL appeared more fragile and prone to disruption, suggesting that miscarriage in these patients may be largely correlated with microbial imbalance.

Although the role of microorganisms in human health and disease is receiving increasing attention, many unknowns remain regarding the microbiota under specific pathological conditions. 14 Recently, characteristics of the endometrial microbiome have been uncovered, particularly in the field of reproductive health. 15 Abnormal alterations in the endometrial microbiota are closely correlated with adverse reproductive outcomes. Certain potential pathogens, such as Gardnerella and Atopobium , have been linked to reproductive disorders such as chronic endometritis (CE) and implantation failure. 16–18 However, a comprehensive understanding of the endometrial microbiota in patients with RPL remains limited, especially regarding microbial diversity and functional interaction networks. Therefore, we conducted an in-depth analysis of EMMA data to systematically characterize the endometrial microbiota in patients with RPL, identify key genera, and construct microbial interaction networks to explore their potential association with RPL. Species diversity is a key aspect of endometrial microbiome research. In the present study, multiple alpha diversity indices were not significantly different between the two groups, indicating consistent species richness and evenness within individual samples. This finding aligns with the findings of Lozano et al, 19 who also reported no significant differences in alpha diversity between the RIF and control groups. Similarly, another study found no difference in endometrial alpha diversity between patients with ongoing and non-ongoing pregnancies. 20 However, some studies have indicated that samples dominated by non-Lactobacillus species exhibit higher Shannon indices and that these patients tend to have poorer pregnancy outcomes. 9 These findings suggest that while alpha diversity is important for assessing the overall structure of microbial communities, its impact on pregnancy outcomes may depend on microbial composition and functional state. Beta diversity measures differences in microbial communities between groups. In this study, we observed a statistically significant difference in beta diversity between the RPL and control groups, attributed to a higher number of unique taxa in the RPL group, consistent with previous findings. 19 Notably, within-group differences in the RPL group were also statistically significant, suggesting possible heterogeneity. Although there was some overlap in the PCoA analysis, the observed trend of separation indicated differences in microbial community composition between the groups, warranting further investigation. We next explored differences in microbial composition between groups at various taxonomic levels. Consistent with previous reports, Lactobacillus was the dominant genus in the endometrium. 21 In this study, the relative abundance of Lactobacillales (order level) was higher in the control group than in the RPL group, suggesting a protective role. Peuranpää et al reported a significant reduction in Lactobacillus abundance in patients with RPL, 22 and another study showed significantly higher implantation, pregnancy, and live birth rates during IVF-ET in women with Lactobacillus -dominated endometria. 9 These findings underscore the importance of Lactobacillus in maintaining reproductive health, and its depletion may lead to adverse pregnancy outcomes, particularly in RPL. Homeostasis of the female reproductive tract microbiota relies on Lactobacillus dominance from the vagina and cervix to the uterine cavity. A decline in Lactobacillus abundance increases the risk of local infections and inflammatory responses. 23 First, lactic acid produced by Lactobacillus helps maintain an acidic environment in the reproductive tract, which inhibits pathogen invasion and colonization, thus protecting the reproductive system from potential harmful microorganisms. 24 Second, lactic acid possesses anti-inflammatory properties that help regulate the immune response in the reproductive tract. It can inhibit pro-inflammatory cytokine production via Toll-like receptors and promote IL-10 secretion, thereby reducing the cytotoxicity of natural killer (NK) cells. 25 Another study found that Lactobacillus and its metabolites regulate the immune status of the endometrium by upregulating the anti-inflammatory factor IL-1RA and suppressing pro-inflammatory signals. 26 Therefore, Lactobacillus not only effectively inhibits pathogens through its antibacterial and anti-inflammatory properties but also helps maintain a healthy microbial balance by modulating the local immune microenvironment. In this study, a significant enrichment of Actinobacteria was observed in the RPL group, which may be associated with increased abundance of certain taxa, such as Bifidobacteriaceae and Gardnerella , at lower taxonomic levels in the RPL group. Cariati et al examined the distribution of endometrial microbiota in patients undergoing IVF-ET and reported that Actinobacteria appeared only in the non-pregnant group. 27 Another study revealed an increased abundance of Actinobacteria in the endometrium of non-pregnant patients, 28 suggesting a potential adverse effect of Actinobacteria on pregnancy. Reports of actinomycotic endometritis describe endometrial infiltration by neutrophils and plasma cells. 29 Additionally, two studies demonstrated increased Actinobacteria abundance in the endometrium of patients with inflammation-related diseases, including endometriosis and endometrial polyps, suggesting that it may act as an inducer of immune responses, contributing to pathophysiological changes. 30 , 31 This immune activation could be the mechanism through which Actinobacteria affects pregnancy outcomes. Gardnerella , an anaerobic bacterium linked to bacterial vaginosis, produces virulence factors that allow it to attach to host epithelial cells and form biofilms, which confer treatment resistance and protect other microorganisms, making infections more difficult to clear. 32 Here, we observed a significant increase in Gardnerella in the RPL group, consistent with previous findings. 22 , 33 It has been speculated that Gardnerella ascends from the vagina to the uterus, disrupting the immune balance of the endometrium and negatively affecting pregnancy. Our differential microbiota analysis and machine learning results revealed a significant enrichment of Streptococcus in the endometrium of patients with RPL, suggesting its potential role in the occurrence of RPL. While some Streptococcus species colonize asymptomatically, enrichment has been linked to implantation failure. 34 Moreover, Group B Streptococcus (GBS) is associated with miscarriage and preterm premature rupture of membranes (PPROM) in late pregnancy 35 and may cause intrauterine infections and fetal death through vertical transmission. 36 Flaherty reported that GBS induces immune dysregulation and cell death by activating the mitogen-activated protein kinase (MAPK) and nuclear factor-κB (NF-κB) pathways, leading to miscarriage, preterm birth, and other pregnancy complications. 37 These findings reinforce the notion that Streptococcus acts as a key pathogen contributing to adverse pregnancy outcomes, implying that it may be an important factor in triggering changes in the endometrial microenvironment and immune dysregulation. Pathogens such as Escherichia coli, Streptococcus, Staphylococcus , and Mycoplasma have been implicated in CE. 38 In line with this, we observed increased Staphylococcus and Streptococcus abundance in patients with RPL, suggesting that these individuals may be more susceptible to CE. Growing evidence has shown a strong association between CE and female infertility, RPL, and implantation failure. Imbalances in the endometrial microbiota and pathogenic bacterial overgrowth may lead to the abnormal infiltration of inflammatory cells, ultimately resulting in miscarriage or implantation failure. 39 In patients with unexplained RPL, imbalanced states in the endometrial microbiota, particularly the increase in pathogenic bacteria, may play a critical role in the mechanisms underlying their miscarriages. In this study, we also identified some taxa that have been less frequently reported in prior studies, such as Cupriavidus, Mendocina, Campylobacter , and Porphyromonas , all of which were significantly more abundant in the control group, suggesting that these microorganisms may be essential for maintaining endometrial microbiota homeostasis. A study in mice detected Campylobacter in the uterus post-ovulation, suggesting that it may participate in reproductive regulation through metabolic pathways and steroid biosynthesis. 40 Another study found the presence of Porphyromonas in the endometrium but reported no association with inflammatory responses. 41 Jain et al revealed that Porphyromonas in the endometrium is associated with changes in Lactobacillus abundance and may be linked to endometrial cancer. 42 We also observed increased Bacillus abundance in the RPL group. A study co-culturing primary endometrial epithelial cells with Bacillus showed that it could induce epithelial cell death within 24 h and significantly increase the expression of pro-inflammatory cytokines. 43 Another study also found that Bacillus can infect the placenta and fetal membranes, leading to adverse pregnancy outcomes such as miscarriage. 44 Chang et al demonstrated that intrauterine infusion of platelet-rich plasma (PRP) improved endometrial receptivity by altering the composition of endometrial microbiota, with the most notable microbial shift being a reduction in Bacillus . 45 These findings suggest that Bacillus may contribute to inflammation and pathogenic changes in the endometrium, adversely affecting pregnancy. This study is the first to use machine learning to identify core genera associated with RPL, namely Chryseobacterium, Fusobacterium , and Streptococcus . Incorporating their relative abundance allowed effective RPL classification. Streptococcus , which is enriched as a pathogen, may play a key role in RPL, whereas Fusobacterium showed an inverse trend, suggesting a protective role. Overall, the combination of these three genera adds value to the RPL. Research on the roles of Chryseobacterium and Fusobacterium in reproductive outcomes is limited; however, some studies have revealed their potential association with endometrial-related diseases. One study suggested that Fusobacterium infection enhances the proliferation, adhesion, and migration of endometrial fibroblasts, contributing to endometriosis. 46 Another study found abnormal enrichment of Chryseobacterium and Fusobacterium in the microbiota of patients with CE and endometrial polyps. This suggests that these genera may influence the local immune environment of the endometrium, potentially affecting the reproductive health. Although the current data are limited, it is advisable to better understand how these genera will specifically contribute to conditions such as RPL in the future. 47 In this study, Fusobacterium showed a reverse scoring trend, suggesting a different mechanism. Microbial pathogenicity is modulated by environmental factors, community dynamics, and host interactions. Under certain conditions, Fusobacterium may compete with or interact with other microbes, inhibit pathogenic processes, and exert protective effects. Microbial interactions are essential for regulating the endometrial immune microenvironment. Network analysis offers a powerful tool to elucidate these relationships and understand how microbial communities contribute to endometrial health through coordinated ecological effects. 48 In this study, the microbial network in the RPL group was sparse and fragile, lacking key module hubs and connectors, and exhibiting low natural connectivity. It rapidly collapsed when a few nodes were removed. This indicates that the microbial community in the RPL group was ecologically less stable than that in the control group, suggesting a link between RPL and the degradation of the endometrial microbial network structure. Low connectivity and fragility of the microbial community may exacerbate immune dysregulation in the endometrium, further impairing its ability to support embryo implantation and pregnancy maintenance. Therefore, changes in the network structure of the endometrial microbiota are not only markers of RPL but may also play a key role in its pathological process. 49