This study was designed as a single-center, cross-sectional, retrospective analysis conducted at the Second Hospital of Anhui Medical University in southern China. Clinical records from June 2022 to December 2023 were reviewed, and corresponding stored biospecimens were retrieved to explore associations between gut microbiota composition, circulating inflammatory cytokine profiles, and the severity of endometriosis across different clinical stages. This study was approved by the Institutional Ethics Committee of the Second Hospital of Anhui Medical University (Approval No. YX2024-164, dated October 2024). Biological samples (including serum and stool) were originally collected as part of routine clinical diagnosis and care between 2023 and 2025, with all patients providing written informed consent for biobanking and future research use. The archived and anonymized samples were retrospectively analyzed in the current study. An internal audit confirmed that all samples were collected under institutional protocols that were in effect prior to the formal 2025 study-specific approval.

A total of 150 premenopausal women aged 20 to 45 years were included based on available laparoscopic and histopathological data. Patients were stratified into four groups: healthy controls (n = 40), individuals with benign gynecologic conditions unrelated to endometriosis (e.g., uterine fibroids, ovarian cysts; n = 45), patients diagnosed with stage I–II endometriosis (n = 25), and patients with stage III–IV endometriosis (n = 40). The above classifications were based on criteria from the revised American Society for Reproductive Medicine (rASRM).

Patients with incomplete records, a history of confounding conditions (e.g., autoimmune/metabolic diseases, malignancies, pelvic infections, irritable/inflammatory bowel disease), or documented use of antibiotics or probiotics within six months prior to sample collection were excluded from analysis. Relevant clinical data, anthropometric measurements, and biospecimen information were obtained from institutional medical records and laboratory archives under standardized protocols.

To improve transparency and ensure reproducibility, the inclusion and exclusion criteria for each study group were explicitly defined. Participants in the healthy control group were premenopausal women aged 20 to 45 years who underwent laparoscopic procedures for non-endometriotic indications, such as elective tubal ligation or benign ovarian cysts. They had no history of chronic pelvic pain, dysmenorrhea, or infertility, and no visual or histologic evidence of endometriosis upon surgical evaluation. Individuals were excluded if they had a history of pelvic infection, gynecologic malignancy, or recent use (within 6 months) of antibiotics, probiotics, or hormone therapies.

For the healthy control group, participants were selected among women who underwent laparoscopic procedures for non-endometriotic indications, such as elective tubal ligation or benign ovarian cysts later confirmed to be non-endometriotic. All individuals were evaluated intraoperatively and showed no visible lesions suggestive of endometriosis, and no history of chronic pelvic pain, dysmenorrhea, or infertility. In addition, where indicated, histopathological examination confirmed the absence of ectopic endometrial tissue. This approach was used to minimize the likelihood of undiagnosed asymptomatic endometriosis in the control group.

The benign gynecologic group included patients diagnosed with non-endometriotic gynecologic conditions such as uterine fibroids or functional cysts, confirmed through imaging or surgery, and with no intraoperative or histological signs of endometriosis. Patients were excluded if they had overlapping endometriotic lesions, concurrent endocrine or autoimmune disorders, or had received hormonal or immunosuppressive treatment in the past 6 months.

Patients in the stage I–II endometriosis group were diagnosed according to rASRM criteria and based on laparoscopic findings and histopathological confirmation. They typically presented with symptoms such as pelvic pain or subfertility. Exclusion criteria included recent surgical treatment for endometriosis (within the previous year), incomplete staging, or coexisting severe pelvic pathologies.

The stage III–IV endometriosis group was comprised of women with confirmed advanced disease, as defined by rASRM staging during laparoscopic surgery, along with histological verification. These participants had not received hormonal therapy or antibiotics for at least 3 months prior to biospecimen collection. Women with deeply infiltrating endometriosis involving the bowel or bladder that precluded complete staging were excluded, as well as those receiving systemic immunomodulatory therapies.

The above criteria were designed to minimize confounding variables, while aligning with established diagnostic and microbiome research protocols, as reported in recent literature [12, 13].

Clinical assessments and biospecimen collections had been conducted as part of routine preoperative evaluations, or general gynecological examinations. Blood pressure measurements were recorded using a calibrated automated sphygmomanometer, with patients seated at rest. The documented value was the average of two successive readings.

Peripheral venous blood samples (5 mL) were collected in the early morning (07:00–09:00) under fasting conditions into ethylenediaminetetraacetic acid (EDTA) tubes. After centrifugation at 3000 rpm for 10 minutes at 4 °C, serum was aliquoted into sterile cryotubes and stored at –80 °C until cytokine analysis. Subsequent quantification was focused on pro-inflammatory cytokines, including interleukin-1$\beta$ (IL-1$\beta$), IL-6, IL-8, and TNF-$\alpha$.

Stool samples (~200 mg) had been collected in sterile containers and delivered under cold-chain conditions to the institutional biobank. Samples were preserved in liquid nitrogen and stored at –80 °C prior to microbial DNA extraction. All biospecimens had been anonymized and linked to relevant clinical data through coded identifiers. The laboratory procedures adhered to institutional protocols for biosafety and biobanking.

For analytical purposes, primary group-wise comparisons were made between the healthy control group and the combined endometriosis cohort. Additional stratified analyses were conducted between stage I–II and stage III–IV endometriosis patients. The benign gynecologic group served as a reference for non-endometriotic perioperative factors, but was not used in the main statistical comparisons.

Approximately 200 mg of stool per sample was used for genomic DNA extraction with the QIAamp Fast DNA Stool Mini Kit (Cat# 51604; Qiagen, Hilden, Germany) following the supplier’s protocol. To enhance lysis of gram-positive bacteria, a bead-beating procedure was incorporated using 0.1 mm zirconia/silica beads. Taxonomic classification was conducted using a naïve Bayes model trained on the SILVA ribosomal RNA gene database project (SILVA) 138 reference dataset at 99% Operational Taxonomic Unit (OTU) similarity. To assess within-sample microbial diversity, alpha diversity metrics including Shannon and Chao1 indices were computed. Differences in microbial composition between groups were evaluated through beta diversity analysis based on Bray-Curtis dissimilarity, and visualized using principal coordinates analysis (PCoA). Group-level variation in microbial communities was statistically tested via permutational multivariate analysis of variance (PERMANOVA) with 999 permutations. Relative abundances of microbial taxa at the genus level were used for downstream comparisons and correlation analyses with clinical and inflammatory parameters.

Systemic inflammatory cytokine levels were measured with enzyme-linked immunosorbent assay (ELISA) kits (MultiSciences Biotech Co., Hangzhou, Zhejiang, China) following the manufacturer’s recommended protocols. After thawing on ice, serum specimens were analyzed in duplicate using 96-well plates. The assays targeted key pro-inflammatory cytokines, specifically IL-1$\beta$, IL-6, IL-8, and TNF-$\alpha$. Standard curves were generated for each plate using recombinant human cytokine standards provided by the manufacturer, with calibration ranges appropriate for physiological serum concentrations (typically 0–1000 pg/mL). Four-parameter logistic regression (4PL) was used to fit the standard curves using GraphPad Prism 9.0 (Dotmatics, Boston, MA, USA), and only assays with an R² $>$0.98 were retained for analysis. Analyte concentrations were calculated based on standard curves and adjusted for sample dilutions when necessary.

To ensure the reliability of results, intra- and inter-assay variability was controlled, with coefficients of variation kept under 10% and 15%, respectively. If duplicate readings varied by $>$20%, or if optical density values fell outside the linear range of the standard curve, the assay was repeated. To minimize batch effects, all cytokines from a single participant were measured on the same plate, and laboratory personnel were blinded to group allocation during processing and data entry.

Statistical processing was performed with R (v4.2.3, R Foundation for Statistical Computing, Vienna, Austria) and Python, version 3.9 (Python Software Foundation, Wilmington, DE, USA), utilizing established libraries such as statsmodels, version 0.14.0 (Seabold, S. & Perktold, J.; https://www.statsmodels.org/), scikit-learn, version 1.3.0 (Pedregosa, F. et al.; https://scikit-learn.org/), ggplot2, version 3.4.2 (Wickham, H.; R package; https://ggplot2.tidyverse.org/), vegan, version 2.6-4 (Oksanen, J. et al.; R package; https://cran.r-project.org/package=vegan), seaborn, version 0.12.2 (Waskom, M.; https://seaborn.pydata.org/), matplotlib, version 3.7.1 (Hunter, J. D.; https://matplotlib.org/). Unless stated otherwise, statistical significance was defined as a two-tailed p value 0.05.

Normality of continuous variables was evaluated using the Shapiro-Wilk test. Data conforming to a normal distribution were summarized as the mean $\pm$ standard deviation (SD) and analyzed using one-way analysis of varianc (ANOVA) for comparisons among multiple groups, followed by Tukey’s honestly significant difference (HSD) post hoc test where appropriate. For two-group comparisons, Student’s t test was applied. For non-normally distributed variables, results were reported as the median with interquartile range (IQR) and compared using the Mann-Whitney U test (for two groups) or Kruskal-Wallis H test (for $\ge$3 groups), followed by Dunn’s post hoc test with Bonferroni correction where indicated. Fig. 1 shows the overall study design, subgroup classification and analytic framework, including ANOVA- and correlation-based assessments of microbiota and cytokines.

Fig. 1.

Study design and microbiota-cytokine association framework. This flowchart illustrates the structure of the study cohort and the analytic workflow. Clinical data and biospecimens from 150 premenopausal women aged 20–45 years were retrospectively analyzed and categorized into four subgroups according to prior laparoscopic and histopathological records: healthy controls (n = 40), patients with benign gynecologic conditions (n = 45), early-stage endometriosis (stage I–II, n = 25), and advanced-stage endometriosis (stage III–IV, n = 40). Microbial alpha diversity (Shannon index) was reduced in the advanced-stage group (p = 0.019), accompanied by increased Prevotella and decreased Blautia abundance. Circulating interleukin-6 (IL-6) and tumor necrosis factor-$\alpha$ (TNF-$\alpha$) levels were significantly elevated in patients with endometriosis (both p 0.0001). Spearman correlation analysis identified positive associations between Prevotella and IL-6 and inverse associations between Blautia and TNF-$\alpha$. Upward arrows (↑) indicate an increase, and downward arrows (↓) indicate a decrease in the corresponding variables.

Differences in the median between groups (e.g., healthy controls vs. stage III–IV endometriosis) were estimated using the Hodges-Lehmann method, a non-parametric approach for assessing central tendency differences. Associated 95% confidence intervals (CIs) were calculated using the exact distribution of Walsh averages, as implemented in the scipy.stats package (Python v3.9; https://www.python.org/). This method was used to complement Kruskal-Wallis tests when interpreting median differences between groups.

The study cohort comprised of 150 premenopausal women, categorized into four groups: healthy controls (n = 40), individuals with benign gynecologic disorders (n = 45), patients diagnosed with stage I–II endometriosis (n = 25), and patients with advanced-stage (stage III–IV) endometriosis (n = 40). Age distribution did not differ significantly across groups (p = 0.234; data not shown).

To address disease-specific effects, the analysis was structured in two tiers. First, comparisons were made between healthy controls and all patients with endometriosis in order to capture the overall differences in gut microbiota composition and inflammatory cytokines. Second, subgroup analyses comparing patients with early-stage (I–II) and advanced-stage (III–IV) endometriosis were performed to explore changes associated with disease progression. Although the benign gynecologic group was included to control for perioperative, hormonal, or hospitalization-related confounders, it was not incorporated in testing of the primary hypothesis. This approach allowed the preservation of data integrity, while focusing analytical power on disease-specific contrasts.

Notable trends were observed for several inflammatory and metabolic markers (Table 1). The median serum IL-6 level increased progressively with disease severity, from 5.5 pg/mL (IQR: 4.1–6.7) in healthy controls to 18.3 pg/mL (14.3–20.5) in patients with stage III–IV endometriosis (p 0.0001). The estimated median difference between these two groups was + 12.8 pg/mL (95% CI: 10.7 to 13.8). This difference was estimated using the Hodges-Lehmann method, based on all pairwise differences between IL-6 values in the two groups. Similarly, TNF-$\alpha$ levels rose from 10.8 pg/mL (7.8–12.0) to 24.9 pg/mL (21.8–28.0), with a corresponding effect size of +14.1 pg/mL (95% CI: 12.0 to 16.1).

Other systemic indicators such as serum creatinine and body mass index (BMI) also showed significant between-group differences. Patients with stage III–IV endometriosis had higher creatinine levels (median: 70.5 µmol/L) and BMI (24.8 kg/m²) compared with healthy controls (58.4 µmol/L and 20.7 kg/m², respectively). Both indicators were significantly different between groups (p 0.0001), with the estimated median differences exceeding clinical thresholds. Diastolic blood pressure varied significantly across groups (Kruskal-Wallis p = 0.0048), with stage III–IV patients showing higher median values than controls (76.2 mmHg vs. 70.6 mmHg, respectively). The estimated median difference and 95% CI were derived using the Hodges-Lehmann method based on pairwise comparisons between groups.

The baseline characteristics and pairwise effect estimates are summarized in Table 1.

Circulating levels of inflammatory cytokines demonstrated a clear stepwise increase across the four clinical groups. IL-6 concentrations differed significantly among clinical groups (Kruskal-Wallis p 0.0001), with post hoc tests confirming higher levels in patients with advanced endometriosis compared to all other groups. Kruskal-Wallis tests confirmed statistically significant differences for both IL-6 (H = 136.11) and TNF-$\alpha$ (H = 132.05) across all groups. These trends are shown in Fig. 2. The minimal overlap observed between distributions supports their potential utility as systemic inflammatory markers of disease severity, as shown in Table 2.

Fig. 2.

Group-wise comparisons of IL-6 levels and BMI across disease severity. (A) IL-6 concentrations across four participant groups: healthy controls, patients with benign gynecologic conditions, and patients with stage I–II or stage III–IV endometriosis. IL-6 levels showed a progressive increase with disease severity. (B) BMI across the same clinical groups, with higher BMI observed in patients with advanced endometriosis. Each box represents the IQR, with the median denoted by a horizontal line. Whiskers indicate 1.5 $\times$ IQR. Statistical comparisons were performed using the Kruskal-Wallis test followed by post hoc Dunn’s test with Bonferroni correction. All group-wise differences were statistically significant (adjusted p 0.05), as detailed in Table 1. Asterisks indicate statistical significance based on pairwise post hoc comparisons: *p 0.05, **p 0.01, ***p 0.001. All overall p values (Kruskal-Wallis test) were 0.01.

To evaluate disease-associated alterations in gut microbiota, 16S rRNA sequencing was performed on fecal samples from all participants. Alpha diversity, assessed by the Shannon index, showed a progressive reduction across disease severity groups. The median diversity value declined from 3.15 (IQR: 2.95–3.42) in healthy controls to 2.78 (2.60–2.92) in stage III–IV endometriosis patients (p 0.001), indicating a loss of microbial richness and evenness (Fig. 3A).

Fig. 3.

Gut microbiota diversity and composition across disease stages. (A) Boxplot showing decreasing alpha diversity (Shannon index) across four groups: healthy controls, benign gynecologic conditions, stage I–II endometriosis, and stage III–IV endometriosis. (B) Principal Coordinates Analysis (PCoA) based on Bray-Curtis dissimilarity demonstrates distinct clustering of gut microbial communities by disease group. (C) Heatmap displaying the relative abundance of the top 20 most prevalent bacterial genera, averaged across participants within each group. Each box in (A) represents the IQR, with the median indicated by a horizontal line; whiskers extend to 1.5 $\times$ IQR. Statistical comparisons were performed using the Kruskal-Wallis test with post hoc Dunn’s test (adjusted p 0.05), indicating a significant decrease in alpha diversity with increasing disease severity. *p 0.05.

Beta diversity analysis using Bray-Curtis dissimilarity and PCoA revealed distinct clustering by disease stage, with a clear separation between advanced endometriosis and non-disease groups (Fig. 3B). This compositional shift was confirmed by PERMANOVA (R² = 0.12, p = 0.002), suggesting a significant structural divergence in microbial communities.

At the genus level, relative abundance profiles identified several taxa with stage-associated trends. Prevotella was progressively enriched and reached the highest level in stage III–IV disease. In contrast, Blautia, an anti-inflammatory short-chain fatty acid-producing genus, was markedly depleted in more advanced disease. A hierarchical heatmap of the top 20 genera illustrates these shifts (Fig. 3C), highlighting a disease-stage-dependent gradient in microbial signatures.

These results demonstrate that both microbial diversity and taxonomic composition are significantly altered in women with endometriosis, with the most profound dysbiosis observed in patients with advanced-stage disease.

Serum concentrations of key pro-inflammatory cytokines were analyzed across the four clinical groups using Kruskal-Wallis tests, followed by post hoc comparisons to determine group-wise differences. As shown in Fig. 4, a clear upward trend was observed for IL-6, which increased progressively from healthy individuals (median: 5.5 pg/mL, IQR: 4.1–6.7) to patients with stage III–IV endometriosis (18.3 pg/mL, 14.3–20.5). Statistically significant differences in the IL-6 level were detected between all clinical stages based on the Kruskal-Wallis test (p 0.0001).

Fig. 4.

Group-wise distribution of serum inflammatory cytokines. (A) IL-6 and (B) TNF-$\alpha$ levels were significantly elevated in patients with advanced endometriosis (stage III–IV) compared to healthy controls and benign gynecological disease. Box plots represent median and interquartile range; individual data points are overlaid. *p 0.05.

A similar trend was noted for TNF-$\alpha$, with the median level increasing from 10.8 pg/mL in healthy individuals to 24.9 pg/mL in patients with advanced disease (p 0.0001). The IL-8 concentration also rose across the disease spectrum, although with a smaller effect size, reaching 22.7 pg/mL (19.4–26.1) in the stage III–IV group compared to 16.9 pg/mL (14.3–19.8) in healthy controls (p = 0.004). A non-significant upward trend (p = 0.073) was observed for IL-1$\beta$, suggesting potential low-grade activation.

Post hoc comparisons with the Kruskal-Wallis test confirmed that IL-6 and TNF-$\alpha$ levels were significantly higher in stage III–IV patients compared to both the healthy and benign groups (all p 0.001 after Bonferroni correction). These findings support the hypothesis that advanced endometriosis is characterized by a systemic pro-inflammatory phenotype involving multiple cytokine pathways.

Based on Linear Discriminant Analysis Effect Size (LEfSe) analysis, 8 genera demonstrated significant group-level differences in relative abundance across the disease stages. Prevotella and Escherichia were progressively enriched in patients with advanced-stage endometriosis, while Blautia, Ruminococcus, Subdoligranulum and Bacteroides showed decreasing trends. These taxa met the threshold for false discovery rate (FDR) correction (q 0.06) and are detailed in Supplementary Table 1.

To explore the interplay between gut microbiota and systemic inflammation, Spearman correlation analyses were performed between genus-level relative abundances, Shannon diversity, and serum cytokine levels in all participants. The resulting correlation matrix is presented in Fig. 5A.

Fig. 5.

Correlation between gut microbiota, alpha diversity, and systemic inflammation. (A) Spearman correlation matrix showing pairwise associations between serum inflammatory cytokines (IL-6, TNF-$\alpha$), Shannon diversity, and the relative abundance of representative microbial genera (Prevotella, Blautia). Cell color represents correlation strength and direction (Spearman’s $\rho$), with positive associations in red and negative in blue. (B,C) Representative scatter plots illustrating significant correlations between IL-6 and Prevotella ($\rho$ = 0.33, p 0.001) and TNF-$\alpha$ and Blautia ($\rho$ = –0.32, p 0.001), respectively. False discovery rate (FDR) correction was applied to correlation analyses; values presented are unadjusted for visual clarity, but only relationships that remained significant after FDR correction (q 0.05) are discussed in the text.

Among the significant associations, the abundance of prevotella showed a moderate positive correlation with IL-6 ($\rho$ = 0.33, p 0.001, q = 0.018 after FDR correction), while Blautia was negatively correlated with TNF-$\alpha$ ($\rho$ = –0.32, p 0.001, q = 0.026). These associations remained statistically significant after controlling for the FDR and are shown in Fig. 5B,C. Participants with higher Prevotella levels tended to exhibit elevated IL-6 concentrations, whereas lower Blautia abundance was associated with increased TNF-$\alpha$ levels.

Other genera such as Ruminococcus and Faecalibacterium displayed weaker, non-significant correlations with IL-8 and IL-1$\beta$, but these associations disappeared after correcting for FDR.

We next examined whether the association between endometriosis severity and inflammatory burden was independent of metabolic and lifestyle factors by constructing generalized linear models (GLMs) using IL-6 concentration and microbial diversity (Shannon index) as continuous outcomes. These models were adjusted for BMI, dietary pattern, and history of antibiotic use (Fig. 6A,B).

Fig. 6.

Adjusted IL-6 levels and predictive model estimates across disease groups. (A) Adjusted serum IL-6 concentrations across four participant groups (healthy controls, benign gynecologic conditions, stage I–II endometriosis, and stage III–IV endometriosis), based on generalized linear model controlling for BMI, dietary pattern, and antibiotic use history. Error bars represent 95% confidence intervals. (B) Model-predicted IL-6 values across disease stages derived from the same adjusted model. The shaded area indicates the 95% confidence band for predicted means. Open circles in panel A represent individual outlier values identified by boxplot analysis.

After multivariable adjustment, IL-6 remained significantly elevated in both the stage I–II ($\beta$ = 4.82, 95% CI: 3.25 to 6.40, p 0.001) and stage III–IV endometriosis groups ($\beta$ = 9.84, 95% CI: 8.13 to 11.56, p 0.001) compared to healthy controls. BMI was independently associated with IL-6 concentration ($\beta$ = 0.47 per 1 kg/m² increase, p = 0.005), while dietary and antibiotic history did not significantly affect cytokine levels.

The IL-6 values predicted by the model across different disease stages (Fig. 6B) mirrored the observed trends, demonstrating progressive inflammatory activation with increasing disease severity, even after adjustment for covariates (Table 3).