A nomogram prediction model for embryo implantation outcomes based on the cervical microbiota of the infertile patients during IVF-FET.

Infertility is defined as the inability of a couple of childbearing age to become pregnant after having regular unprotected intercourse for at least 1 year ( 1 ). The incidence of infertility among reproductive-aged couples in China has reached 17.6% ( 2 ), making it a global public health issue alongside cancer and cardiovascular disease. In vitro fertilization and embryo transfer (IVF-ET) is widely regarded as a valid method for treating infertility, with the cumulative pregnancy rates after three cycles still hovering around 50% even after nearly half a century of advancements in assisted reproductive technology (ART) ( 3 ). However, recurrent implantation failure (RIF) and recurrent spontaneous abortion (RSA) remain significant challenges in ART as their causes are often unknown and complex. Approximately 10% of women undergoing three cycles of embryo transfer experience RIF, and 12%–15% of women experience miscarriage per transfer cycle, creating a distressing and frustrating situation ( 4 ). Among various factors, the reproductive tract microenvironment, particularly the genital tract microbiome, plays a vital role in establishing and maintaining pregnancy ( 5 , 6 ). The human body harbors a vast number of microbes ( 7 ). Approximately 9% of the microbiome exists in the female reproductive tract and has a significant impact on female reproductive health ( 8 , 9 ). A healthy vaginal microbiome is typically defined as a Lactobacillus -dominant population, while a non- Lactobacillus -dominant vaginal microbiota is generally associated with a high Nugent score and bacterial vaginosis (BV) ( 10 , 11 ). In healthy women, the ratio of Lactobacillus located in the vaginal microbiota is usually greater than 90% and can be classified into four types of Lactobacillus spp. ( Lactobacillus crispatus , Lactobacillus jensenii , Lactobacillus iners , and Lactobacillus gasseri ). Among these, L. crispatus is considered a vaginal genus with healthy outcomes, while Lactobacillus iners is associated with poor outcomes, such as BV, HPV infection, embryo implantation failure, and other disorders ( 10 , 12 – 16 ). More precise research on the microbiota located in the cervix or uterus has drawn public attention. Lactobacillus , which is present in the upper reproductive tract, has been found to promote a healthy reproductive tract environment. Conversely, uterine dysbiosis is indicative of a series of disorders, such as chronic endometritis (CE) and RSA ( 17 – 20 ). The balanced symbiotic barrier formed by the cervico-vaginal epithelium and Lactobacillus helps protect against the invasion of pathogens ( 21 , 22 ). When this balance is disrupted, it can lead to poor reproductive outcomes, such as sexually transmitted diseases, pelvic inflammatory diseases, infertility, and miscarriage ( 23 – 28 ). Furthermore, it has been found to be related to the establishment of pregnancy after IVF-ET. According to Haahr’s research, a negative clinical pregnancy was associated with an aberrant vaginal microbiota due to BV ( 29 ). Microorganisms such as Bifidobacterium , Prevotella , Lactobacillus iners , and Streptococcus were verified to lead to pregnancy failure ( 30 – 32 ). Compared to individuals with non- Lactobacillus -dominant microbiota (NLDM) (< 90% Lactobacillus ), the clinical pregnancy rate of individuals with Lactobacillus -dominant microbiota (LDM) (≥ 90% Lactobacillus ) is significantly higher ( 19 , 33 ). Moreover, Lactobacillus may also serve as a tool for predicting favorable outcomes of IVF-ET ( 20 , 32 , 34 ). However, other researchers have reported controversial findings. Ruiying Wang et al . found no obvious difference in the abundance of Lactobacillus in the cervix and uterus between pregnant and nonpregnant individuals who underwent IVF-ET treatment ( 32 ). According to LelaK. Keburiya’s study, there was also no striking difference in the abundance of Lactobacillus in the uterine microbiota between patients who received embryo transfer for the first time and those who experienced RIF. Additionally, although a significant difference was identified in the prevalence of strict anaerobes and Gardnerella in Keburiya’s study, it had no deleterious effect on pregnancy rates ( 35 ). Therefore, more rigorously designed research studies are necessary to determine whether the microbiota of the female genital tract influences the outcomes of IVF-ET. Advancing our understanding of microbial etiologies in the lower genital tract of females with adverse post-IVF-FET outcomes, this study aimed to explore the role of cervical microbiota anomalies in embryo implantation failure and to develop a predictive model for embryo implantation failure based on cervical microbiota profiles.

A flow diagram illustrating the study design is provided in Fig. 1 . Among the 131 patients enrolled, 47 experienced embryo implantation failure, indicating non-pregnancy (NP), and 84 cases achieved clinical pregnancy (CP). The clinical characteristics of the enrolled subjects are detailed in Table 1 . Schematic illustration of the study design. Baseline and transplantation characteristics of enrolled patients d By t -test. By chi-square test. By Fisher exact test. NP, non-pregnancy; CP, clinical pregnancy; values are given as mean ± SD, number (%). NA, not applicable. A total of 4,299 operational taxonomic units (OTUs) were identified in the cervical microbiota. The alpha diversity of the microbiota was calculated by the Shannon index ( P value = 0.15) and Simpson index ( P value = 0.17) ( Fig. 2A and B ). No significant differences in the alpha diversity were observed between the NP and CP groups. Principal component analysis (PCA) and principal coordinate analysis (PCoA) performed at the phylum level showed no significant differences between the two groups ( Fig. S1 ). Further PCA ( Fig. 2C and E ) and PCoA ( Fig. 2D and F ) were applied to illustrate the distribution of the microbial community in the samples. The beta diversity analysis at both the genus and species levels revealed no significant differences between the NP and CP groups. A high level of overlap between the two groups of samples was observed. Diversity and composition of the cervical microbiota. ( A, B ) Alpha-diversity of the cervical microbiota in the two groups was calculated and is shown by the Shannon and Simpson indexes. ( C, E ) Principal component analysis (PCA) illustrated the distribution of the cervical microbiota at the genus and species levels. ( D, F ) Principal coordinate analysis (PCoA) represented the distribution of the cervical microbiota at the genus and species levels. ( G ) Genus-level composition, highlighting the 20 most abundant genera. ( H ) Species-level composition, highlighting the 20 most abundant species. NP, non-pregnancy; CP, clinical pregnancy. Firmicutes was the dominant bacteria of the two groups of cervical microbiota at the phylum level ( Fig. S2A ). No statistically significant differences were shown in Firmicutes ( P = 0.291), with the mean relative abundance of Firmicutes in the NP and CP groups being 83.377% and 84.020%, respectively. Firmicutes was followed by Actinobacteriota , Proteobacteria , Bacteroidota , and other taxa. However, only Fusobacteriota showed significant differences between the NP group and the CP group at the phylum level (0.204% vs. 0.024%, P = 0.008). The composition of other levels of bacteria is shown in Fig. S2 . Comparative results of the relative abundance of bacteria at other levels are also shown in Table S1 . The composition of the two groups is similar at the genus and species levels. Figure 2G displays the top 20 most abundant genera. Among them, Halomonas , Klebsiella , Atopobium , and Ligilactobacillus were obviously different between the two groups. The relative abundances of Lactobacillus , Bifidobacterium , and Streptococcus exceeded 1% in both groups, while those of Prevotella , Halomonas , and Aerococcus were higher than 1% only in the NP group. A total of 21 genera were found to have significant differences between the two groups ( Table S1 ). At the species level, species with the top 20 abundances are shown in Fig. 2H . Among them, Streptococcus agalactiae , Prevotella copri , Halomonas stevensii , Megasphaera lornae, and Atopobium vaginae had obvious significant differences between the two groups. The relative abundances of L. crispatus, Lactobacillus iners, Bifidobacterium leopoldii, Bifidobacterium vaginale, Lactobacillus jensenii, Streptococcus anginosus, and Streptococcus agalactiae were higher than 1% in both groups, while those of Prevotella copri and Aerococcus christensenii were more than 1% only in the NP group. A total of 44 species differed significantly between the two groups ( Table S1 ). Further, the random forest algorithm identified key microbial biomarkers influencing embryo implantation. The model consists of 1,000 decision trees, each built on a different random subset of the training data. Feature importance was assessed by mean decrease accuracy (MDA). The results showed that Halomonas was the most significant contributor to the model at the genus level, with the highest MDA, indicating that it played a key role in the classification process ( Fig. 3A ). Prevotella , Achromobacter , Atopobium , Veillonella , Fenollaria , Pseudomonas , Dialister , Lactobacillus, and Aeromicrobium also showed higher MDA values, indicating that they also contributed to the model ( Fig. 3A ). The box plot showed the top 10 genera with the highest MDA values ( Fig. 3C ). At the species level, the highest contribution to the model was from Halomonas stevensii , followed by Prevotella copri , Achromobacter unclassified , Halomonas sp000246875 , Aeromicrobium halocynthiae , Atopobium vaginae , Pseudomonas veronii , etc. ( Fig. 3B ). Figure 3D shows the box plots of the top 10 species with MDA scores. Random forest analysis. ( A, B ) Random forest analysis was conducted to map the importance of genera or species, with the horizontal axis representing the importance measure and the vertical axis listing the genera or species ranked by their importance. ( C, D ) Boxplots are presented for the genera or species that ranked among the top ten based on the random forest analysis. NP, non-pregnancy; CP, clinical pregnancy. The top 10 genera, based on MDA scores from the random forest analysis, were incorporated as categorical predictors (positive vs negative) in the multivariate binary logistic regression model. The backward LR method was used for stepwise regression, and the relative abundance of Lactobacillus was a must for inclusion in the model. The final prediction model consisted of the classifications of Halomonas , Atopobium , and Veillonella , as well as the relative abundance of Lactobacillus (𝐿𝑜𝑔𝑖𝑡(𝑝) = −1.673 + (−0.257 × 𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑎𝑏𝑢𝑛𝑑𝑎𝑛𝑐𝑒 𝑜𝑓 𝐿𝑎𝑐𝑡𝑜𝑏𝑎𝑐𝑖𝑙𝑙𝑢𝑠) + 1.436 × classification of 𝐻𝑎𝑙𝑜𝑚𝑜𝑛𝑎𝑠 + (−1.136 × classification of 𝐴𝑡𝑜𝑝𝑜𝑏𝑖𝑢𝑚) + 0.92 × classification of 𝑉𝑒𝑖𝑙𝑙𝑜𝑛𝑒𝑙𝑙𝑎)) ( Fig. 4A ). In order to assess the robustness and generalization ability of the model, the bootstrap method was used for internal validation. The average ROC curve of internal validation is displayed in Fig. 4B (AUC = 0.718, 95% CI: 0.628–0.807, P < 0.001). Prediction nomogram for embryo implantation failure and ROC curves for internal and external model validation using bootstrap sampling. ( A ) Nomogram of the embryo implantation failure prediction model. ( B ) ROC curve of internal verification through bootstrap sampling. ( C ) ROC curve of external verification through bootstrap sampling. In order to evaluate the generalization ability of the predictive model, further external validation was carried out. An independent data set from the project PRJNA903403 ( 20 ), containing 16S full-length sequencing data from the cervical microbiota of 120 patients undergoing IVF treatment, was used. A comparison of internal and external data is shown in Table 2 . The bootstrap method was also used to verify the external data, and the results indicated that the mean ROC curve of the external data had an AUC greater than 0.6 (AUC = 0.654, 95% CI: 0.553–0.755, P = 0.037) ( Fig. 4C ). Slight differences existed in the characteristics of the subjects in the external data and the subjects in this study. Despite these discrepancies, the model demonstrated robust predictive accuracy in external validation, underscoring its clinical application. Comparison of internal and external data groups d No severity specified. Three embryo implantation failures were taken as the RIF standard, and RSA is not included. No specific requirements are made, but smokers or alcoholics are not included in the internal data. Y, yes; N, no.

In this study, the association between the cervical microbiome and embryo implantation fate was investigated. Additionally, we developed a clinical prediction model based on the cervical microbiota for the early identification of high-risk patients with embryo implantation failure. According to our findings, the composition of the cervical microbiota was associated with embryo implantation. Although we found no significant difference in the diversity of the cervical microbiota between the NP and CP groups, specific bacteria, including Halomonas , Klebsiella , Atopobium , and Ligilactobacillus , showed significant differences between the two groups. The genera that significantly contributed to differentiating NP and CP in our model were identified using a random forest algorithm. The prediction model included classifications of Halomonas , Atopobium , and Veillonella , along with the relative abundance of Lactobacillus . Both the internal and external validation of the model showed an AUC greater than 0.6, indicating its potential value in clinical applications. Successful implantation and healthy growth of the embryo are facilitated by potential embryos, a receptive uterus, and the harmonious interaction between them. As is well known, the quality and quantity of embryos are of great importance to the first step of pregnancy. Previous studies showed that the number of high-quality embryos is an important predictor of CP ( 36 ). In addition, embryonic chromosomal abnormalities are important factors that prevent early pregnancy from being established and maintained. A chromosomal error originating from an oocyte can lead to the creation of aneuploid embryos, which are age-related to the female, especially with advanced age over 38 years ( 37 ). In this study, all embryos were of good quality, the stage and number of embryos transferred were equivalent, and the age of women was less than 35 years, eliminating interfering factors as much as possible. Another important factor affecting embryo implantation is the receptive endometrium, and the substantial impacts exerted by the female anatomic structure of the uterus and hormone fluctuations on the endometrium play critical roles in the survival of pregnancy ( 38 , 39 ). Besides, a normal pregnancy is a successful homogeneous hemigenetic transplant, and maternal–fetal immune tolerance allows the embryo to successfully implant and develop until delivery. The substantial changes to the microbial communities inhabiting the mother’s genital tract participate in inducing maternal–fetal immune tolerance. Gardnerella vaginalis activates the production of more NF-κB and IL-8 in the cervico-vaginal epithelium, whereas L. crispatus does not ( 40 ). Patients with embryonic miscarriage have a lower abundance of Lactobacillus in the vagina and higher levels of IL-2 and IL-2/IL-10 in the vaginal lavage fluid ( 41 ). This evidence suggests that microorganisms in the reproductive tract may affect the local immune microenvironment by modulating the host immune response, resulting in an adverse influence on the endometrial receptivity. The interplay between pregnancy and the microbiome has profound implications for maternal and infant wellbeing. Lactobacillus emerges as the sustainer of a balanced reproductive tract microbiota in healthy women. It can prevent genital tract infections in a variety of ways. Lactic acid has been shown to have antibacterial properties, and Lactobacillus is capable of producing large amounts of it ( 42 ). Competing with other bacteria, Lactobacillus can also prevent infection by directly adhering to host cells and regulating the secretion of cytokines by host cells ( 43 , 44 ). In general, a disordered microbiota of the reproductive tract refers to the absence or decrease of Lactobacillus , which can lead to poor reproductive outcomes such as BV, sexually transmitted diseases, pelvic inflammatory diseases, infertility, and miscarriage ( 23 – 28 ). For infertile individuals, some previous studies have found that a Lactobacillus -dominant reproductive tract microbiota is associated with successful pregnancy after IVF treatment. Hong Zeng et al. ( 45 ) showed that the clinical pregnancy rate was greatly reduced in the non- Lactobacillus -dominant vaginal microbiota group (40.98% vs 50.82%, P < 0.01), consistent with the findings in endometrial microbiota from extensive studies ( 31 , 46 , 47 ). According to the studies of Sofia Väinämö ( 48 ) and Wenzheng Guan ( 20 ), L. crispatus may be the main contributor to clinical pregnancy and live birth. In this research, the relative abundance of Lactobacillus did not exhibit a significant variation between the NP and CP groups. This finding is consistent with the studies conducted by Ruiying Wang ( 32 ) and Lela K. Keburiya ( 35 ). Wang reported that the microbiota of both the vagina and cervix did not show significant differences between the pregnant and nonpregnant groups who underwent fresh cleavage-stage embryo transfer (ET) in their first IVF/ICSI cycle. In Keburiya’s research, the relative abundance of Lactobacillus in the cervix did not vary significantly between the pregnant and nonpregnant groups, which included participants who received their first frozen embryo transfer (FET) or suffered from RIF. It remains controversial whether Lactobacillus has a significant effect on pregnancy outcomes after fresh or frozen ET, and further research studies are required to validate this conclusion. Here, we included the relative abundance of Lactobacillus in the model despite no prominent difference in its relative abundance between the NP and CP groups. The results showed that the model worked well on both the internal and the external data. In contrast, if the relative abundance of Lactobacillus was deleted from the model, less favorable performance was manifested on the external data. Based on these findings, Lactobacillus contributes to the establishment of pregnancy after IVF treatment, although it is not the primary factor. The presence of other pathogenic bacteria should also be taken into account. In the established prediction models, Halomonas contributed the most and was significantly positively correlated with embryo implantation failure. Halomonas, a class of halophilic bacteria found in high-salt environments, is primarily used in biotechnology and environmental remediation due to its biological properties ( 49 , 50 ). Originally considered an environmental microbe, Halomonas has been found in the human microbiota with the advancement of microbiota studies and has been linked to diseases such as Helicobacter pylori infection, breast cancer, and psychiatric disorders ( 51 – 53 ). Halomonas has also been detected in the female genital tract and is associated with early pregnancy miscarriage ( 54 ). There are few studies on Halomonas and female reproductive health, which may be partly due to the imprecision of the microbiota identification methods used in previous studies. To our knowledge, this study is the first to point out that Halomonas may be involved in embryo implantation failure, and further large-scale research is needed to validate this finding. Veillonella , a common bacterium found in the human oral, intestinal, and respiratory tracts, is associated with various diseases, including periodontal disease, chronic obstructive pulmonary disease (COPD), and inflammatory bowel disease ( 55 – 57 ). Veillonella in the female reproductive tract has been concerned with diseases such as cervical intraepithelial neoplasia, endometriosis, BV, and postpartum fever ( 58 – 61 ). Min Fu et al . found Veillonella as a potential biomarker for RIF development through a study of the vaginal microbiota of patients in the control and RIF groups ( 33 ). Here, we likewise reported the relationship between Veillonella and embryo implantation failure. Atopobium is a common genus in the female genital tract, previously considered the pathogenic bacteria of BV and preterm birth. Since it is also present in the normal microbiota of the female genital tract ( 62 , 63 ), the presence of Atopobium does not mean that it is related to embryo implantation failure, and in most studies, only an abnormal increase in the abundance of Atopobium increased the high risk of embryo implantation failure. In our prediction model, the presence of Atopobium was found to be a protective factor for successful embryo implantation. Notably, the average relative abundance of Atopobium among the enrolled patients was less than 1%. Thus, our findings suggest that Atopobium emerges as a beneficial factor for embryo implantation. However, the potential negative impact of an abnormal increase in its abundance on embryo implantation requires further investigation. Despite these strengths, some limitations of the study should be acknowledged. First, the sample size was relatively small, which may limit the generalizability of the findings. Second, due to the single-center design, selection bias cannot be completely eliminated. Third, although full-length 16S-rDNA provided more comprehensive information than traditional 16S rRNA gene sequencing, it still has limited resolution in identifying bacteria at the species or strain level. Fourth, we only collected cervical samples on the day of embryo transfer and could not assess the temporal changes of the microbiota during the course of pregnancy. Fifth, we did not account for some potentially contradictory factors such as diet and lifestyle, which may influence the cervical–vaginal microenvironment. In the future, large-scale, multicenter prospective studies are needed to validate these findings. More advanced sequencing technologies such as whole-genome shotgun metagenetics may be applied to gain deeper insights into the reproductive microbiota related to pregnancy outcomes. In addition to clinical factors, more comprehensive information on demography, socioeconomic status, and lifestyle should be collected to better control for confounding effects. In conclusion, our study provides preliminary evidence of the correlation between cervical microbiota and embryo implantation failure. A prediction model is developed, which consists of the classification of Halomonas , Atopobium, and Veillonella and the relative abundance of Lactobacillus , with generalizability for external verification. These findings strengthen our understanding of the role of reproductive tract microbes in infertile females attending IVF-FET. The insights gleaned from this study may shed light on the evaluation and early intervention for patients with high risk of embryo transfer failure during IVF-FET.

A total of 160 patients were recruited between November 2021 and August 2022. These patients planned to undergo in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI), followed by frozen embryo transfer (FET) treatment at the reproductive center, which is located at the affiliated Suzhou hospital of Nanjing Medical University. Twenty-four patients were excluded because they did not meet the inclusion criteria. Inclusion criteria: Age between 20 and 35 years; First or second FET cycle; Quality and quantity of embryos: one high-quality blastocyst (≥4 BB), two D3 cleavage stage embryos (6–10 cells, with fragments ≤ 10% and symmetric blastomeres), or one D3 cleavage stage embryo with good quality, as determined by time-lapse observation; Endometrial thickness not less than 7 mm on the transfer day; No history of acute reproductive tract infection and no use of antibiotic use in the past 2 months. Exclusion criteria: Severe hydrosalpinx; Stage III–IV endometriosis or severe adenomyosis; Moderate-to-severe intrauterine adhesions without restoration by hysteroscopy; Primary uterine malformation; Body mass index (BMI) >30 kg/m2; RSA or RIF; Autoimmune disease; Diabetes; Thrombotic diseases; Anti-Müllerian hormone <1.1 ng/mL. Cervical mucus was collected on the day of FET by gently rotating a sterile cotton swab in the cervical canal after wiping away any surface secretions. To prevent contamination, the sampling swab should not come into contact with any other parts during this process. The samples were immediately immersed in liquid nitrogen and promptly transported to the laboratory for bacterial DNA extraction (Qiagen, Hilden, Germany). To trace the cervical microbiota, qPCR and 16S rDNA full-length sequencing were performed using this genomic DNA. In this study, 16S full-length assembly sequencing technology (16S-FLAST) was adopted. The fundamental principles and schemes of FLAST are consistent with those depicted in earlier reports on the construction of a full-length single-molecular library of 16S rDNA genes ( 64 ). All samples are sequenced on the NovaSeq 6000 instrument (Illumina). About 496.9 Gb of data (1.7 billion reads) were obtained for 136 vaginal samples. According to the unique tag encoded in both flanks, we obtained 2,014,560 unique 16S rRNA gene sequences (>1,200 bp), among which, five samples were excluded because the sequence number was lower than 1,000 base pairs (bp). The full-length 16S rDNA gene sequences were obtained from the sequenced data by FLAST, and then these sequences were classified into operational taxonomic units (OTUs) using vSearch in QIIME2 with a similarity of 99% ( 65 , 66 ). We used Mothur v1.42.0 ( 67 ) to annotate these OTUs in accordance with the reference database Silva_132_SSURef_Nr99 ( 68 ). Non-pregnancy was defined as a human chorionic gonadotropin (hCG) level <25 mIU/mL 14 days after FET. Conversely, if the hCG value exceeded 25 mIU/mL, biochemical pregnancy was generally accepted. Clinical pregnancy was determined by the presence of a gestational sac revealed by ultrasonography approximately 30 days after FET. Categorical variables were expressed as numbers (relative frequency %), while continuous variables were shown as mean ± standard deviation (SD). Student’s t -test was adopted to compare the differences of continuous variables between groups. For categorical variables, a χ test or Fisher’s exact test was applied at a two-sided significance level of 0.05. For the relative abundance of microorganisms, the Mann–Whitney U test was applied at a two-sided significance level of 0.05. The R package vegan was used for alpha- and beta-diversity analysis. A random forest model between the NP and the CP group was established by using the R package randomForest. The visualization of the above results and the plotting of the box plot were carried out by using the R package ggplot2. A multivariate binary logistic regression model was constructed using SPSS 26.0 with backward stepwise regression. The R package pROC was used to perform bootstrap repeated random sampling and the plotting of receiver operating curves (ROC).