Fecal specimens were prospectively collected from June to December 2023 at The Sixth Affiliated Hospital of Sun Yat-Sen University, comprising three patient cohorts: (1) EMS with dysmenorrhea (n = 20; median age, 31.5 years), (2) EMS without dysmenorrhea (n = 13; median age, 28.0 years), and (3) healthy controls (n = 12; median age, 28.5 years).

Ethical approval was obtained from the institutional review board (2024ZSLYEC-020), and written informed consent was obtained from all participants. Pain severity was assessed using the 11-point numerical rating scale, where participants rated their average pain during the past three menstrual cycles on a 0–10 scale.

The specific inclusion and exclusion criteria for the study were as follows:

Inclusion criteria for patients with EMS:

1. Aged 18–45 years, enrolled during the nonmenstrual, nonpregnant, and nonpuerperal periods, with regular menstrual cycles.

2. Diagnostically confirmed by ultrasonography or pelvic magnetic resonance imaging (MRI, both with sensitivity and specificity $>$90%) [15].

3. No hormone therapy in the past six months and did not take antibiotics or probiotics within the past eight weeks.

4. Non-smokers.

5. Absence of inflammatory or nerve-related pain conditions.

6. No history of intestinal surgery, no intestinal treatments within the past six months, and no significant family history of major intestinal diseases (e.g., colorectal cancer or inflammatory bowel disease).

7. Not in the acute stage of vaginitis, with a negative cervical human papillomavirus test.

8. No special dietary habits (e.g., vegan or vegetarian diet, preference for spicy foods, or highly acidic diet).

The diagnosis was based on characteristic findings on transvaginal ultrasonography (unilocular cyst without vascularity and homogeneous ground-glass echogenicity) and/or MRI (high T1 signal, T2 shading, T2 dark spots, and evidence of adhesions), which are widely recognized as reliable alternatives to laparoscopy for this EMS subtype.

Inclusion criteria for healthy controls:

Healthy controls were recruited with the same eligibility criteria as patients with EMS, except that they had regular menstrual cycles without a history of EMS or dysmenorrhea.

The exclusion criteria were as follows:

1. Have other chronic abdominal or kidney issues, ovarian/uterine tumors, liver diseases, inflammatory or immune conditions, or heart/coronary artery diseases.

2. Received hormone therapy within six months before surgery.

3. Regularly use painkillers.

4. Have a history of headaches or other neurological disorders.

5. Experienced severe pelvic inflammation during surgery.

6. Have severe chronic or acute inflammatory diseases.

7. Have adenomyosis.

DNA was extracted using a DNA extraction kit according to the manufacturer’s protocol. The highly variable V3–V4 region of the bacterial 16S rRNA gene, identified by the primer set 338F–806R, was then targeted for amplification by polymerase chain reaction (PCR). The primer sequences employed were as follows: 338F 5^′-ACTCCTACGGGAGGCAGCA-3^′ and 806R 5^′-GGACTACHVGGTWTCTAAT-3^′. The PCR-amplified products were visualized on agarose gels and then purified. The purified PCR amplicons were then pooled in equimolar concentrations for paired-end sequencing (2 $\times$ 250 bp) on the Illumina Novaseq 6000 platform (Illumina, San Diego, CA, USA).

Fecal specimens were initially collected on dry ice and subsequently stored at –80 °C until sequencing to ensure sample integrity. The raw sequencing data were initially processed based on an individual nucleotide quality assessment. The raw data were rigorously filtered using Trimmomatic (version 0.33: http://www.usadellab.org/cms/?page=trimmomatic) [16]. Subsequently, primer sequences were identified and precisely excised with Cutadapt (version 1.9.1; https://cutadapt.readthedocs.io/en/stable/) [17]. The paired-end reads obtained from previous steps were assembled by USEARCH (version 10; https://www.drive5.com/usearch/) [18]. To ensure the reliability of the assembled sequences, chimeras were detected and removed by employing UCHIME (version 8.1; https://www.drive5.com/usearch/manual8.1/uchime_algo.html) [19].

By employing a 97% sequence similarity threshold, operational taxonomic units (OTUs) were derived from nonrepetitive sequences, except for single sequences, and clustered using USEARCH (version 10.0; http://www.drive5.com/usearch/). This stringent clustering approach resulted in a comprehensive OTU abundance table. OTUs were meticulously classified and annotated, leveraging the SILVA database (version 138.1; https://www.arb-silva.de/) [20] and employing the Naive Bayes classifier integrated within QIIME2 (v.2020.6; https://qiime2.org/) [21]. This classification was conducted with a confidence level set at 70%. To assess the $\alpha$-diversity, the abundance-based coverage estimator (ACE), Chao1, and phylogenetic diversity (PD) whole tree indices were calculated. $\beta$-diversity, which quantifies the compositional dissimilarity between microbial communities, was explored using principal component analysis (PCA). To identify differential microorganisms across groups, Python scripts were employed to create ternary phase diagrams, the Wilcoxon rank-sum test was applied to test significant differential microorganisms between groups, and intersection analysis was applied to pinpoint co-regulated differential microbes. The functional potential of the microbial communities was deduced using PICRUSt (version v.2.5.2; https://picrust.github.io/picrust/), a bioinformatics tool that predicts metabolic pathways from 16S rRNA gene sequences. To further delve into the diversity of these metabolic pathways, the Cluster of Orthologous Groups database and the Kyoto Encyclopedia of Genes and Genomes (KEGG) were referred to. Spearman correlation analysis was conducted to investigate the correlation between the gut microbiome and clinical indicators among patients with EMS-related pain. Ultimately, the predictive ability of the differentially abundant microorganisms was evaluated by the receiver operating characteristic (ROC) curve analysis. To identify the optimal threshold for discriminating between groups, sensitivity and specificity for each microbial marker were determined using the Youden index.

Statistical analyses were conducted using GraphPad Prism 10 (GraphPad Software, San Diego, CA, USA). The distribution of continuous variables was assessed using the Shapiro-Wilk test and visual inspection of quantile-quantile (Q-Q) plots (Supplementary Fig. 1). For normally distributed variables (age and body mass index [BMI]), the homogeneity of variances between groups was evaluated using the Brown-Forsythe test. Given that both variables satisfied the assumption of equal variances, they were compared between groups using the one-way analysis of variance. Non-normally distributed continuous variables with three groups (age at menarche [AAM], gravidity, parity, miscarriages, pain level, pain duration, anti-Müllerian hormone [AMH], carbohydrate antigen 125 [CA125], CA19-9, neutrophil count [N], lymphocyte count [L], and neutrophil-to-lymphocyte ratio [NLR]) were compared using the Kruskal-Wallis test, whereas volume of chocolate cysts (VOCC), which involves two groups, was compared using the Mann-Whitney U test. Categorical variables, such as the distribution of endometriotic cysts (unilateral vs. bilateral), were analyzed using Fisher’s exact test. Continuous data are presented as mean $\pm$ standard deviation (SD) for normally distributed variables and as median [interquartile range, IQR] for non-normally distributed variables. A p-value 0.05 was considered significant.

This study enrolled 45 women categorized into three groups: 20 patients with EMS and dysmenorrhea, 13 patients with EMS but without dysmenorrhea, and 12 healthy controls. Their baseline characteristics are shown in Table 1. No significant differences were observed among the three groups in terms of age, BMI, AAM, gravidity, parity, or miscarriage frequency (all p $>$ 0.05). The distribution of endometriotic cysts (unilateral vs. bilateral) was not significantly different between the EMS group with and without dysmenorrhea (p $>$ 0.05). All patients with EMS and dysmenorrhea experienced chronic pelvic pain, with a mean duration of 6.00 [3.25, 10.00] years and an average pain score of 5.00 [3.25, 7.00]. Serum CA125 levels were significantly higher in the EMS patients with dysmenorrhea compared with both the EMS group without dysmenorrhea and the healthy control group (Kruskal-Wallis statistic = 13.4; p = 0.0012). In contrast, no significant differences were found in AMH, CA19-9, N, L, or NLR among the groups (all p $>$ 0.05).

To explore potential relationships between dysmenorrhea and the gut microbiome in EMS, 16S rRNA sequencing was performed on 45 stool samples, identifying 5535 OTUs. Samples were categorized into three groups: control (healthy controls), EMS–Pain (EMS without dysmenorrhea), and EMS+Pain (EMS with dysmenorrhea). Rank-abundance and species accumulation curves indicated sufficient sequencing depth and adequate OTU coverage across the groups (Fig. 1A,B). The $\alpha$-diversity analyses (ACE, Chao1, and PD whole tree) did not show significant pairwise differences among the groups (all p $>$ 0.05; Fig. 1C–E). PCA based on OTU composition showed no clear separation among the three groups; however, some distributional variations were observed (Fig. 1F). Principal component 1 (PC1) and principal component 2 (PC2) explained 13.35% and 12.52% of the variance, respectively. These findings suggest that potential group differences in gut microbiota may be subtle and are not reflected in overall community structure.

Then, composition analyses were conducted at both phylum and genus levels. At the phylum level, Firmicutes and Bacteroidetes were dominant across all groups. The relative abundance of Proteobacteria displayed a trend toward an increase in the EMS group with dysmenorrhea, whereas Fusobacteriota showed relatively lower abundance in the EMS group with dysmenorrhea compared with the other groups (Fig. 1G). At the genus level, Bacteroides and Faecalibacterium were relatively abundant (Fig. 1H). Specifically, the relative abundance levels of Subdoligranulum and Parabacteroides were higher in the EMS group with dysmenorrhea. Healthy controls exhibited higher Roseburia abundance than the EMS group (Fig. 1H). However, none of these microbial abundance variations reached significance. In summary, although overall diversity remained comparable, subtle group-specific shifts in the relative abundance of certain taxa may be associated with the dysmenorrhea status in EMS. Thus, further validation is necessary to confirm these potential associations.

To discern the gut microbes specifically implicated in EMS and EMS-related pain, differential species analysis was conducted using taxonomically annotated data from the five most abundant phyla and other phyla. The ternary diagram showed uniform abundance and distribution of Bacteroidota and Firmicutes across three cohorts (Fig. 2A). Wilcoxon rank-sum testing identified 12 genera with the top smallest p-values, with Colidextribacter exhibiting the highest relative abundance. Compared with the EMS group without dysmenorrhea, the abundances of Colidextribacter and Acinetobacter were significantly decreased in the EMS group with dysmenorrhea. Conversely, compared with the EMS group without dysmenorrhea, the abundance of uncultured_Bacteroidales_bacterium in the EMS group with dysmenorrhea significantly increased (Fig. 2B).

To identify core microbial alterations specifically associated with dysmenorrhea in EMS, “Control vs. EMS+Pain” and “EMS–Pain vs. EMS+Pain” were compared, and the differential microorganisms shared in both comparisons were identified (p 0.05; detailed p-values for each comparison are provided in Supplementary Table 1). At the genus level, there were 15 commonly upregulated and 10 downregulated genera in the EMS group with dysmenorrhea, including the downregulation of Morganella and Streptomyces and upregulation of Clostridioides in the EMS group with dysmenorrhea. At the species level, 27 species were upregulated, and 16 species were commonly downregulated in the EMS group with dysmenorrhea, such as downregulation of Alistipes inops and Gemmobacter aquaticus and upregulation of Parabacteroides faecis among these groups (Fig. 2C). No consistent phylum-level regulation was observed. Notably, several species had expression changes consistent with their corresponding genera. For example, unclassified_Candidatus_Solibacter, a species upregulated in the EMS group with dysmenorrhea, belongs to the genus Candidatus_Solibacter, which was also upregulated at the genus level. Similarly, unclassified_Streptomyces was upregulated, aligning with the upregulation of Streptomyces at the genus level. Likewise, unclassified_Acinetobacter and Faecalicoccus pleomorphus were downregulated at both species and genus levels. These microbes with concordant regulation across taxonomic levels may be crucial in the pathophysiology of EMS-related dysmenorrhea.

Additionally, Spearman’s rank correlation analysis of the top 50 genera revealed a robust co-occurrence network, with UCG_005 and UCG_002 as the central hub taxa (Fig. 2D). The high abundance levels of Bacteroides and Faecalibacterium highlighted their potential regulatory roles in EMS-related phenotypes. Collectively, these exploratory findings implicate multilevel gut microbiota alterations in EMS-related dysmenorrhea, though mechanistic roles must be experimentally validated.

Community-wide gut microbiota functions were predicted using PICRUSt2 based on clusters of orthologous groups (COG) family data to explore potential roles in EMS and associated pain. Analysis revealed significant functional enrichment across 24 distinct COG categories, particularly in signal transduction mechanisms and amino acid transport and metabolism (Supplementary Fig. 2A). Furthermore, intergroup comparisons of COG functional profiles exhibited trends of variation in predicted functions; however, no significant differences were observed (Supplementary Fig. 2B).

Gut microbiota-derived KEGG pathway analysis highlighted enrichment in pathways related to amino acid metabolism, metabolism of cofactors and vitamins, signal transduction, and endocrine and metabolic diseases (Fig. 3A). Pairwise KEGG pathway comparisons among the three groups revealed significantly higher enrichment in pathways related to cofactor and vitamin metabolism (p 0.01) and substance dependence (p 0.05) in the EMS group with dysmenorrhea compared with the EMS group without dysmenorrhea (Fig. 3B). Collectively, these predictive functional analyses suggest that the gut microbiota in EMS may modulate amino acid and cofactor metabolism pathways with potential implications for pain mechanisms.

To investigate the relationship between differential gut microbes in EMS and clinical indicators of pain and inflammation, correlation analyses were conducted across phylum, genus, and species levels. Clinical parameters included age, BMI, pain severity, AAM, VOCC, AMH, CA125, CA19-9, and NLR.

At the phylum level, Spirochaetota and Deferribacterota showed negative and positive correlations with NLR, respectively, in the comparison between the healthy control group and EMS group without dysmenorrhea (Fig. 4A). In the comparison between the EMS group without dysmenorrhea and the group with dysmenorrhea, Crenarchaeota and Cyanobacteria showed significant positive correlations with pain severity (Fig. 4B). Conversely, in the comparison between the healthy control group and the EMS group with dysmenorrhea, Spirochaetota, Deferribacterota, and Halobacterota were negatively associated with pain severity (Fig. 4C).

At the genus level, Gordonibacter and Luteolibacter showed a negative correlation with NLR, whereas Mucispirillum and Butyricimonas demonstrated a positive correlation in the comparison of the healthy control group with the EMS group without dysmenorrhea (Fig. 4D). In the analysis between the EMS group without dysmenorrhea and the dysmenorrhea group, Colidextribacter, Faecalicoccus, and Acinetobacter were significantly negatively correlated with the pain severity of dysmenorrhea in EMS. Oscillibacter, Streptomyces, and Candidatus_Solibacter exhibited a significant positive correlation with the degree of pain in EMS (Fig. 4E). In the analysis between the healthy control group and the EMS group with dysmenorrhea, Acinetobacter and Faecalicoccus showed a significant negative correlation with the degree of dysmenorrhea of EMS, whereas Candidatus_Solibacter and Streptomyces demonstrated a positive correlation (Fig. 4F).

At the species level, in the comparison between the healthy control group and the EMS group without dysmenorrhea, Limosilactobacillus_mucosae and Bacteroides sp. showed a negative correlation with NLR, whereas Megamonas funiformis and unclassified_Prevotella demonstrated a positive correlation with NLR (Fig. 4G). In the analysis between the EMS group without dysmenorrhea and the dysmenorrhea group, Faecalicoccus_pleomorphus showed a negative correlation with pain, whereas unclassified_Candidatus_Solibacter and unclassified_Streptomyces showed a positive correlation with pain (Fig. 4H). In the analysis between the healthy control group and the EMS group with dysmenorrhea, unclassified_Acinetobacter and Faecalicoccus_pleomorphus displayed a negative correlation with pain, whereas Alistipes_onderdonkii, Alistipes_indistinctus, and unclassified_Streptomyces showed a positive correlation with pain (Fig. 4I). These findings identify microbiota–clinical indicator correlations that indicate potential roles of specific taxa in EMS-related inflammation and pain.

To validate discriminatory microbial markers, ROC curve analysis was performed using an area under the curve (AUC) $>$0.7 threshold. At the genus level, Ligilactobacillus (AUC = 0.7308, sensitivity = 46.15%, specificity = 100.00%) and Butyricimonas (AUC = 0.7212, sensitivity = 61.54%, specificity = 91.67%) effectively distinguished the EMS group from the healthy control group (Fig. 5A). Additionally, four genera demonstrated strong discriminatory ability between the EMS group with and without dysmenorrhea: unclassified_Cyanobacteriales (AUC = 0.7269, sensitivity = 65.00%, specificity = 76.92%), Oscillibacter (AUC = 0.7154, sensitivity = 90.00%, specificity = 46.15%), [Eubacterium]_oxidoreducens_group (AUC = 0.7462, sensitivity = 90.00%, specificity = 61.54%), and Colidextribacter (AUC = 0.7577, sensitivity = 75.00%, specificity = 76.92%) (Fig. 5B). At the species level, Alistipes_onderdonkii (AUC = 0.7276, sensitivity = 46.15%, specificity = 100.00%) and Alistipes_indistinctus (AUC = 0.75, sensitivity = 46.15%, specificity = 100.00%) showed excellent ability to differentiate the EMS group from the healthy control group (Fig. 5C). Similarly, unclassified_Cyanobacteriales (AUC = 0.7269, sensitivity = 65.00%, specificity = 76.92%), unclassified_Butyricimonas (AUC = 0.7596, sensitivity = 95.00%, specificity = 61.54%), and unclassified_[Eubacterium]_oxidoreducens_group (AUC = 0.7462, sensitivity = 90.00%, specificity = 61.54%) discriminated between the EMS groups with and without dysmenorrhea (Fig. 5D). Collectively, these findings highlight Colidextribacter and unclassified_Butyricimonas as potential microbial markers with robust discriminatory power for EMS-associated dysmenorrhea.