This was a case-control study conducted at the Department of Obstetrics, Gynecology, and Reproductive Medicine in the two large affiliated hospitals of the University of South China. The present study was approved by the Clinical Research Ethics Committee of the Second Hospital, University of South China ([2023]003, February 27, 2023). All participants willingly participated in the study and gave their written informed consent. We compared the microbiota composition of the endometrium among patients with RIF (Group A, n=17), healthy non-pregnant women (Group B, n=19), and healthy pregnant women who conceived spontaneously and had successful embryo implantation (Group C, n=20) from March 2023 to December 2023. Cases (Group A) and controls (Group B and C) were strictly identified following the established screening procedures, at an approximate ratio of 1:1:1. Special consideration was given to age to ensure that the age distribution in the case group was matched to those of the two control groups. Statistical analysis showed that there was no significant difference in age among the three groups.

Cases included only RIF patients (Group A), who were defined as women under 40 years old and had undergone 3 or more fresh or frozen embryo transfer cycles with a cumulative transfer of at least four high-quality embryos but had failed to achieve a clinical pregnancy (6, 15). Control groups consisted of Groups B and C. Group B comprised healthy women who were not pregnant at the time of endometrial sample collection. Group C included women in early pregnancy (≤9 weeks of gestation) who conceived spontaneously and chose to terminate the pregnancy through artificial abortion following successful embryo implantation.

Exclusion criteria mainly included women with ①abnormal vaginal bleeding; ② abnormal uterine morphology or pathology (such as uterine fibroids, severe pelvic adhesions, endometrial polyps, endometrial cancer, abnormal endometrial hyperplasia, endometriosis, adenomyosis, etc.); ③placed an intrauterine device, untreated hydrosalpinx, or abnormal sex hormone levels; ④acute inflammation, autoimmune diseases, endocrine diseases, sexually transmitted diseases, cancer, severe heart, brain, liver, or kidney diseases; ⑤any antibiotic, hormone, or immunosuppressive treatments within one month; ⑥cervical treatment within one week; ⑦vaginal douche or local vaginal medication within five days; or ⑧sexual intercourse within 48 hours.

The collected data include demographic characteristics (age, BMI, education level, household income), medication utilization evaluation, reproductive history, sexual behavior, number of pregnancies, deliveries, and abortions, as well as history of pelvic surgery. The cases underwent infertility-related examinations, including hysteroscopy at the time of sampling, RNA-seq-based endometrial receptivity analysis, gynecological examination, laboratory testing, imaging studies, and other assessments relevant to assisted reproductive technology treatment.

Endometrial specimens in Group A were collected using an endometrial sampler during RNA-seq-based endometrial receptivity testing, typically on the 5th day of progesterone-induced histological transformation in hormonally controlled cycles. Endometrial specimens in Group B were collected using an endometrial sampler on the 7th day after the surge of luteinizing hormone during the natural menstrual cycle (16). Endometrial specimens in Group C were collected using a disposable negative- pressure suction probe under B-ultrasound guidance, prior to the implementation of induced abortion surgery, to avoid contamination with embryonic villi. During ultrasound examination, accurate differentiation between the endometrium and the fetus can be achieved through careful assessment of their anatomical position, echogenic features, and gestational age, even when their images overlap or are difficult to distinguish.

High-resolution ultrasound equipment facilitated the acquisition of clearer images, while experienced physicians ensured precise sampling through comprehensive analysis of these image features. One endometrial tissue sample from each participant was used for the detection of chronic endometritis(CE). The diagnostic criteria for CE were defined as the presence of ≥5 CD138-positive plasma cells per 10–30 high-power field(HPF) based on immunohistochemical staining (17). The remaining endometrial tissue samples were rinsed with sterile saline and subsequently transferred into disposable sterile cryopreservation tubes (NEST, 607001) for storage at -80°C.

Certain measures were implemented to reduce the risk of specimen contamination. 1) During the study period, we assigned designated personnel for specimen collection, and the collection process was standardized and consistent. 2) Specimen collection was performed in operating rooms under regular monitoring of environmental hygiene. The microbial community in the air and on the work surfaces of the operating rooms was regularly monitored. These results indicated that no pathogenic bacteria were detected, with a total colony count of 0 CFU/m³ and 0 CFU/cm², respectively. 3) The experimental procedure incorporated nuclease-free water as negative controls, which were processed in parallel with clinical specimens throughout the entire workflow, including DNA extraction, PCR amplification, library preparation, and sequencing. 4) The items utilized for specimen collection were sterilized. 5) Specimen collectors wore sterile gowns and sterile disposable gloves. 6) During the period, microbial communities on the hands of sample collectors and in disinfectants used on the skin or m (18)ucous membranes, such as 0.5% povidone iodine, were regularly tested. The results revealed no detection of pathogenic bacteria, with a total colony count of 0 CFU/cm² and 0 CFU/mL, respectively. 7) Specimens were collected prior to any treatment procedures.

Endometrial microbiota was tested using 16S rRNA gene amplicon sequencing technology. Following thawing, endometrial tissue samples were vigorously vortexed in 2 mL of PBS(pH 7.0) to ensure homogeneous resuspension. The suspensions were then centrifuged at 12,000 rpm for 5 minutes, and the supernatants were discarded.

The remaining sediment were used for genomic DNA extraction using the CTAB method (Nobleryder, China). DNA quality was semi-quantitatively assessed by agarose gel electrophoresis, comparing the brightness of sample bands to that of a known DNA marker to estimate concentration and purity. DNA samples were diluted with sterile water to a concentration of 1ng/μL and then amplified by PCR. Secondly, the extracted DNA and region-specific primers (341F-806R) attached to Illumina paired-end (PE) adapters, barcodes, and sequencing primers were used to amplify the hypervariable V3-V4 region of 16S rRNA. The PCR products were purified using Qiagen Gel Extraction Kit (Qiagen, Germany) and subsequently assessed for their quality via electrophoresis on a 2% agarose gel. Finally, sequencing libraries were constructed using NEBNext^®Ultra™II DNA Library Prep Kit, Cat No. E7645, and their quality was subsequently evaluated using a Qubit@ 2.0 Fluorometer (Thermo Scientific) and an Agilent Bioanalyzer 2100 System. Libraries were sequenced on the NovaSeq6000 platform (Illumina, USA) and 250 bp PE reads were created.

Sequencing data was processed using the Quantitative Insights Into Microbial Ecology 2 (QIIME2) platform (https://qiime2.org/). High-quality clean reads were obtained through splicing the reads using FLASH software (V1.2.11) and controlling their quality using fastp software (V0.20.0). Chimeric sequences were detected and removed by comparing clean reads with the Silva database (https://www.arb-silva.de/) using Vsearch (V2.15.0). Amplicon sequence variants (ASVs) were generated by denoising effective reads using either the DADA2 or Deblur module in QIIME2 software (V QIIME2-202006). The Silva138.1 database was used to generate an ASVs table with taxonomy. ASVs with an abundance of less than 5 were excluded from the following analysis. The relative abundance of each ASV was standardized to the total abundance of all ASVs in each sample.

The microbial analysis platform was used for data analysis (Wekemo Bioincloud, https://www.bioincloud.tech). Alpha and beta diversity of the endometrial microbiota were assessed using Shannon diversity index, Chao index, principal coordinate analysis (PCoA), and permutational multivariate ANOVA (PERMANOVA) in R statistical software (v4.2.1, vegan package). A linear discrimination analysis (LDA) was implemented using LEfSe (R, DESeq2 package, Kruskal Wallis H-tests, two-tailed P-value <0.05, LDA score >4.0) to identify biomarkers at the genus level, which can distinguish the difference of microbiota composition among the three groups. Redundancy analysis (RDA) (R, vegan package, PERMANOVA, 999 permutations, P<0.05) was applied to assess the effects of host factors on microbial community. Microbial co-occurrence networks based on Spearman rank correlation (R, igraph package, Bray-Curtis, complete linkage; SPSS 26.0) was used to analyze the interactions between the microbe, considering taxa with a significant Spearman correlation coefficient (Benjamini-Hochberg corrected P<0.05). Statistical Analysis of Metagenomic Profiles (STAMP) software (V 2.1.3) was utilized to analyse the differences in mean proportions among microbial genera. Extended error bar plots (Welch’s t test or Kruskal-Wallis H test, two-tailed P-value <0.05, Benjamini-Hochberg multiple hypothesis test correction) displayed the microbial taxa exhibiting significant differences.

Continuous variables were described using means and standard deviation (SD) for data with normal distributions, or using medians and interquartile ranges for data with skewed distributions. Qualitative variables were reported using absolute frequencies and percentages. Intergroup differences were tested using a range of statistical tests, including Student’s t-test, Wilcoxon rank sum test, Kruskal-Wallis H test, chi-square test, or Fisher’s exact test. The relative abundance of the main endometrial microbiota associated with RIF was displayed using means and 95% confidence intervals in the three groups. This approach facilitates establishing a threshold range of the relative abundance for the main microbial community that affects recurrent implantation failure. A P value of <0.05 was considered statistically significant. Redundancy analysis (RDA) (Python 3.8, statsmodels and scikit-learn packages, PERMANOVA with 999 permutations, P<0.05) was performed to evaluate the associations between host factors and endometrial microbes. The relationships between the relative abundance of microbial genera and host factors were examined using linear regression models based on restricted cubic splines (RCSs) in Python (v3.8, statsmodels, patsy packages, P for nonlinearity < 0.05).

A total of 56 participants were included in this study. Among the participants in Group A, all seventeen individuals had CE, with a prevalence rate of 100%. No cases of CE were observed in either Group B or Group C. Table 1 presents a comprehensive report on the demographic characteristics of the study participants. No statistically significant differences were observed among the three groups in terms of key indicators such as age, body mass index, educational level, and family income. However, it is noteworthy that there were significant statistical differences in sexual and reproductive health-related indicators among the three groups. Specifically, Group A exhibited a significantly lower number of deliveries and induced abortions, as 12 women (70.6%) reported being nulliparous and 13 women (76.5%) reported no history of induced abortion. Compared to two control groups, only 1 woman (5.9%) reported engaging in ≥3 sexual activities per week, and 5 women had first intercourse ≤20 years old in Group A.

Diversity of endometrial microbiota was analyzed in Figures 1a-c . The richness of the endometrial microbiota in Group A was significantly lower than that in both control groups ( Figure 1a ). There were no statistically significant differences between Group B and Group C. Based on Shannon’s diversity index, Group A and B exhibited significantly higher microbial diversity than Group C ( Figure 1b ). Figure 1c demonstrated that endometrial microbial composition of Group A differed significantly from those of the other two control groups, while only minor differences were observed between the two control groups. RIF explained 20% of the variations in the composition of endometrial microbiota (PERMANOVA, P=0.001).

Differential analysis of the endometrial microbiota composition was conducted, with the results presented in Figures 1d, e . Among the three groups, the relative abundance of Lactobacillus showed the most significant changes, with the lowest relative abundance in Group A (6.6%) and the highest relative abundance in Group C (16.1%). Pseudomonas, as the top genus, had the highest relative abundance in Group B (27.7%), followed by Group A (19.1%), and finally Group C (5.9%). Compared to the two control groups, Ralstonia (7.9%), Methyloversatilis (7.7%), Vibrio (6.7%), Bacteroides (6.0%), Pseudoalteromonas (5.0%), Stenotrophomonas (4.7%), Methylobacterium (2.4%), Cutibacterium (2.1%), Sphingomonas (0.5%), Escherichia_Shigella (0.2%), and Faecalibacterium (0.2%) had significant relative abundance changes in Group A. According to the criteria of LDA score greater than 4, Methyloversatilis, Bacteroides and Aeromonas were the main differential microbiota in Group A; Ligilactobacillus was the main differential microbiota in Group B; and Sphingomonas, Faecalibacterium, and Escherichia_Shigella were the main differential microbiota in Group C ( Figure 1f ).

This study employed a random forest model to further identify main differential endometrial microbes among the three groups. Sphingomonas, Escherichia-Shigella, Methyloversatilis, and Faecalibacterium were the four most important genera with relatively high Gini coefficients ( Figure 1g ).

In light of the above results, Methyloversatilis, Lactobacillus, Sphingomonas, Faecalibacterium, Escherichia-Shigella, Bacteroides, and Aeromonas were identified as the main differential endometrial microbes associated with RIF. It should be emphasized that Lactobacillus not only represented the second major genus but also exhibited the most pronounced variations in relative abundance among the three groups.

In this study, the mean and 95% confidence intervals (CI) of the relative abundance for the main differential microbiota in each group were calculated to estimate their threshold values associated with RIF ( Table 2 ). Differential microbial relative abundance thresholds was defined based on functional characteristics. For embryo implantation-disrupting microbes (e.g., Methyloversatilis, Aeromonas), the higher mean value between the healthy non-pregnant group (Group B) and the early pregnancy group (Group C) was selected as the pathogenic threshold. For embryo implantation-promoting microbes (e.g., Lactobacillus), the mean value in the RIF group (Group A, 6.606%) was adopted as the protective threshold. This strategy ensured statistically significant differences (P<0.05) by comparing 95% confidence intervals( Table 2 ). The threshold values for relative abundance were determined as follows: 3.807% for Methyloversatilis, 6.606% for Lactobacillus, 0.192% for Sphingomonas, 0.193% for Faecalibacterium, 0.618% for Escherichia-Shigella, 2.411% for Bacteroides, and 0.019% for Aeromonas. In Group C, Methyloversatilis was undetectable in 19 individuals and detected at a low level (0.045%) in only one individual. Therefore, the risk of RIF increased when the relative abundance of Methyloversatilis, Bacteroides, Sphingomonas, and Aeromonas exceeded their respective thresholds, or when the relative abundance of Lactobacillus, Faecalibacterium, or Escherichia-Shigella fell below the defined threshold values.

Interaction networks of endometrial microbiota were established for each group ( Figure 2 ). In Group A, Lactobacillus was positively correlated with Sphingomonas (r= 0.64, P= 0.005; Figure 2a ), indicating an enhanced synergistic interaction between Lactobacillus and Sphingomonas. There was a significant positive correlation between Pseudomonas and Methyloversatilis (r=0.86, P<0.001; Figure 2a ). In Group B, Lactobacillus exhibited a negative correlation with Cutibacterium and Streptococcus (r=-0.49, P=0.033; r=-0.53, P=0.018; Figure 2b ). Pseudomonas displayed a significant positive correlation with Vibrio, Methyloversatilis, Pseudoalteromonas, Faecalibacterium, and Escherichia-Shigella (r=0.81, P<0.001; r=0.93, P<0.001; r=0.83, P<0.001; r= 0.64, P=0.003; r=0.65, P=0.003; Figure 2b ). Aeromonas and Methyloversatilis exhibited a positive correlation (r=0.52, P=0.022; Figure 2b ). In Group C, Pseudomonas exhibited a significant positive correlation with Ralstonia, Vibrio, and Pseudoalteromonas (r=0.62, P=0.004; r=0.83, P<0.001; r=0.77, P<0.001; Figure 2c ).

Associations between host factors and the relative abundance of the top 30 microbial genera are presented in Supplementary Table 1 . Statistically significant associations with endometrial microbiota composition were observed for the number of induced abortions (R² = 7.09%, P=0.002), pregnancies (R² = 5.22%, P=0.003), deliveries (R² = 5.22%, P=0.006), and frequency of sexual activity (R² = 4.10%, P= 0.022). History of pelvic surgery(R² = 2.91%, P=0.108) and age at first sexual intercourse(R² = 2.26%, P=0.242) were not significantly associated with endometrial microbiota composition.

As shown in Figure 3a , restricted cubic spline analysis revealed the non-linear relationship between the number of pregnancies and the relative abundance of Ligilactobacillus(P=0.0301). The relative abundance peaked at three pregnancies, declined with both fewer and more pregnancies, and exhibited an increase when the number exceeded five. A significant inverse U-shaped association was observed between the frequency of sexual activity and the relative abundance of Sphingomonas( Figure 3b , P=0.0389). Abundance increased with frequency up to around two times per week, then declined thereafter, indicating a threshold effect in the microbial response. Conversely, a significant U-shaped association was identified between the frequency of sexual activity and the relative abundance of Methyloversatilis ( Figure 3c , P=0.0005). The abundance declined with increasing frequency up to around two times per week, after which it rose sharply, indicating a non-linear microbial response to behavioral exposure. The number of induced abortions had a significant non-linear relationship with the relative abundance of Vibrio ( Figure 3d , P=0.0156). Vibrio abundance was lowest at one induced abortion and increased steadily with a higher number of abortions. This pattern suggested a dose-dependent microbial response potentially linked to cumulative reproductive tract disturbance.

In the present study, there were significant differences in the composition of endometrial microbiota between RIF patients and both healthy non-pregnant women and healthy pregnant women. Methyloversatilis, Lactobacillus, Sphingomonas, Faecalibacterium, Escherichia-Shigella, Bacteroides, and Aeromonas were identified as the main differential endometrial microbes associated with RIF. Preliminary reference thresholds for their relative abundance were defined. These differential microbiota exhibited different interactions at different relative abundances.

Our findings further supported that endometrial microbiota was closely associated with the occurrence of RIF (6, 10, 19). Previous studies have shown that the composition of endometrial microbiota in RIF is significantly different from that of other women undergoing assisted reproductive technology treatment (6, 20–22). Based on these studies, this study’s design may help to clarify a potential pathophysiological mechanism linking endometrial microbiota and RIF (10). Compared to healthy women with normal fertility, the endometrial microbiota of RIF patients exhibited reduced richness but increased diversity. Principal Coordinates Analysis (PCoA) further revealed distinct microbial community structures, highlighting the presence of numerous differential taxa in the endometrium of RIF patients. We identified Methyloversatilis, Lactobacillus, Sphingomonas, Faecalibacterium, Escherichia-Shigella, Bacteroides, and Aeromonas as the main differential endometrial microbes associated with RIF. Among them, Lactobacillus, Sphingomonas, and Faecalibacterium appeared to be favorable for pregnancy in women (23, 24). Lactobacillus can impact the function of the endometrium by influencing its immune system, receptivity, and remodeling, thereby affecting female pregnancy (25–27). The effects of Sphingomonas on the endometrium remain unclear. However, a study found that it could inhibit the infection of cervical HPV and the occurrence of cervical cancer (28). Faecalibacterium was a beneficial genus for human health (29). When its relative abundance was high, it had a certain anti-inflammatory effect; however, when its abundance was low, it could easily lead to mucosal inflammation (29–31). Methyloversatilis and Pseudomonas were pathogenic bacteria that contribute to the occurrence of RIF. Methyloversatilis had been found in the female reproductive tract (32), and its effect on the endometrium was currently unclear. Currently, it was believed that Pseudomonas was an opportunistic pathogen in human urogenital system (28). Genomic versatility of Pseudomonas enabled it to encode and produced a diverse array of cell-associated and extracellular virulence factors, thereby facilitating its successful colonization in various host environments and contributing significantly to its elevated levels of antibiotic resistance (33). This factor might pose a challenge in RIF treatment. By comparing the endometrial microbiota of healthy women with normal fertility, we established threshold values for the main differential microbiota for the first time. In the endometrial microenvironment of RIF, there existed a dynamic antagonistic interaction between Lactobacillus and Pseudomonas. In a healthy endometrial environment, Lactobacillus dominates by maintaining acidity, secreting antimicrobial substances, and competitively occupying ecological niches, thereby suppressing the growth of Pseudomonas and limiting it to low-abundance, benign colonization (23, 25). However, in the pathological state of RIF, the abundance of Lactobacillus decreased significantly, compromising the microbiota barrier. Its loss of dominance lifted suppression on Pseudomonas and might facilitate the overgrowth of other microbes, further disrupting the endometrial microenvironment (34). Notably, a decline in Lactobacillus was accompanied by an increase in microbes like Methyloversatilis, which may synergize with Pseudomonas. The increased abundance of Pseudomonas has been associated with inflammatory responses and adverse reproductive outcomes, suggesting its potential transition into a highly virulent pathogen (34). This association indicated that Lactobacillus depletion might be a key trigger for Pseudomonas virulence activation, creating a vicious cycle of “dysbiosis-inflammation-gene suppression”. Importantly, disruption of the endometrial microbiota resulted in marked shifts in community composition and inter-microbial interactions, including the enhanced synergy between Pseudomonas and Methyloversatilis. Such ecological changes might further compromise the integrity of the endometrial microenvironment.

This study has uncovered a significant correlation between RIF, chronic endometritis (CE) and specific microbial communities. Existing research has demonstrated that CE is closely associated with marked alterations in the endometrial microbiota, which may disrupt the endometrial immune environment and consequently influence the success rate of embryo implantation (35). Some evidence indicated that microbial dysbiosis in CE patients, especially the increased presence of non-Lactobacillus microbes, may be associated with endometrial inflammation (18, 22, 34). CE may create a favorable microenvironment for the proliferation of certain pathogenic microbes, potentially contributing to dysbiosis of the endometrial microbiota (18). Further research is needed to fully elucidate the underlying mechanisms of these associations.

Our study suggested that host factors, including induced abortions, number of pregnancies, and frequency of sexual activity, significantly influence the composition of the endometrial microbiota. The relative abundance of Ligilactobacillus peaked in women with three pregnancies, decreased thereafter, and rebounded after five pregnancies, suggesting a compensatory microbial response to physiological stress. This pattern indicated that the microbiota might adapt to stress through compensatory mechanisms, highlighting the adaptive plasticity of the probiotic ecosystem under pressure (35). Notably, when sexual activity exceeded twice per week, the abundance of Sphingomonas decreased while Methyloversatilis increased, indicating that frequent sexual activity might disrupt the microbial barrier and promote colonization by opportunistic pathogens (36, 37). Additionally, a higher number of induced abortions was positively correlated with increased Vibrio abundance, implying that repeated abortions might exacerbate inflammation and promote pathogen proliferation (38). These findings underscored the role of reproductive factors as modifiable drivers of endometrial dysbiosis and highlighted critical thresholds, such as three pregnancies and twice-weekly sexual activity, that may serve as early intervention targets.

Notably, this study observed a high prevalence of CE at 30.4%, suggesting that CE was associated with the study results through two distinct mechanisms. First, as a key coexisting factor, endometrial microbiota can both cause CE and be affected by it (18, 34, 39). On the one hand, microbial infections may lead to the development of CE, thereby affecting embryo implantation; on the other hand, CE may alter the microbial composition through persistent inflammatory responses, thereby potentially confounding the true associations between host factors and the microbiota (18, 34). Second, as an effect modifier, the chronic inflammatory microenvironment of CE may potentially intensify the impact of reproductive history and sexual behavior on the microbiota by weakening mucosal defenses and shifting local immune tolerance thresholds (19, 22, 40).

This study is the first to investigate and identify the main differential microbiota associated with RIF by comparing healthy non-pregnant women, healthy pregnant women, and RIF patients. These findings, including the identification of the main differential microbiota, the establishment of threshold values for their relative abundance, and the construction of their interaction networks, aim to advance the development of endometrial microbiota-based strategies for the intervention and treatment of RIF. The study’s objectives and scope align closely with global research priorities focused on the role of the female reproductive tract microbiota in reproductive health (23, 41–43).

This study has some limitations. Firstly, this study can not directly prove the causal relationship between endometrial microbiota and RIF. Although a strong correlation between them has been observed, the study’s design and methodological limitations preclude the establishment of a causal relationship between the microbiota and RIF or between the microbiota, CE, and RIF. It is challenging to collect endometrial tissue, detect endometrial microbiota, and follow up on whether RIF occurs in women of childbearing age who have not yet experienced any reproductive problems (which is an obstacle in any study). Secondly, this study was limited by a relatively small sample size. Several factors contributed to this constraint. The collection of endometrial tissue required sterile procedures performed by senior gynecologists under sterile conditions in a hospital setting. For the control groups, endometrial tissue had to be obtained from healthy women who met strict inclusion and exclusion criteria—criteria that most hospital visitors did not fulfill. For the case group, endometrial tissue was collected specifically during RNA-seq-based endometrial receptivity testing, in order to closely mimic the conditions of embryo transfer. Thirdly, while threshold values for the main differential microbiota were preliminarily established, further validation in larger, independent cohorts is needed. Lastly, although multiple confounding factors were controlled for in the study, the potential influence of CE as a confounder in the relationship between host factors and endometrial microbiota cannot be entirely ruled out. The high prevalence of CE may have impacted the observed associations between endometrial microbiota and host factors, potentially contributing to non-linear patterns or threshold effects. These effects may stem from the selective pressure exerted on the microbial community by the chronic inflammatory environment associated with CE. Given the potential impact of CE on endometrial microbiota, further investigation into the interaction between CE and host factors is essential for elucidating the underlying mechanisms of reproductive disorders.

This study could initiate several important next steps. First, it is essential to refine the threshold ranges of relative abundance for key differential endometrial microbiota and to assess their clinical predictive value in relation to reproductive outcomes, particularly recurrent implantation failure (RIF). Establishing standardized cutoff values may enhance the utility of microbial biomarkers in clinical risk stratification and early intervention. Second, mechanistic studies using animal models are warranted to validate the causal roles of these dominant microbial taxa and to explore their interactions within the endometrial microenvironment. These studies could elucidate the underlying pathways through which microbial dysbiosis contributes to implantation failure, thereby informing the development of targeted microbiota-based therapies or immunomodulatory strategies.