Participant recruitment and sample gathering

We integrated eQTL data encompassing 19,942 genes from the GWAS catalog as the exposure factor, alongside endometriosis datasets ebi-a-GCST90018839 and ukb-d-IBD_ENDOMETRIOSIS, boasting extensive sample populations of 231,771 and 361,194, respectively. To fortify our analysis, we included transcriptomic data from GSE7305, GSE11691, GSE23339, and GSE25268, forming a composite validation cohort of 79 subjects, delineated into 57 endometriosis cases and 22 controls. Further depth was added through GSE214411, which provided single-cell profiles from 128,243 endometrial cells across ten subjects, including six with minimal/mild endometriosis and four controls.

Ectopic and corresponding eutopic endometrial specimens were meticulously collected from patients diagnosed with ovarian endometriotic cysts during laparoscopic surgeries conducted at the Gynecological Department of Xiangya Hospital, within the timeframe of January 2022 to October 2023. Histological examinations post-surgery confirmed the diagnosis of endometriosis. Control specimens were similarly sourced from individuals presenting with benign ovarian cysts unrelated to endometriosis. The study encompassed 38 endometriosis patients with an average age of 31.0 ± 4.4 years, and a control group of 43 individuals with an average age of 29.0 ± 3.2 years. All participants were characterized by regular menstrual cycles and had refrained from hormonal treatments in the three months preceding their surgeries. The timing of sample collection was strategically aligned with the proliferative phase of the menstrual cycle, as corroborated by preoperative assessments and histopathological evaluations. Informed consent was diligently obtained from all participants, and the study protocol received ethical clearance from the Medical Ethics Committee of Xiangya Hospital, Central South University, under approval number 202,109,936.

Elucidating endometriosis risk genes via integrative mendelian randomization, machine learning, and single-cell transcriptomics

We utilized a quintet of methodologies—Inverse Variance Weighted, Weighted Median, MR Egger, Weighted Mode, and Simple Mode—leveraging the TwoSampleMR package to discern endometriosis-associated risk genes (Bowden et al. 2015). We focused on genes from the ebi-a-GCST90018839 and ukb-d-IBD_ENDOMETRIOSIS datasets exhibiting odds ratios (OR) greater than 1, and identified common risk genes through their intersection. Subsequent to batch correction and normalization, transcriptome datasets from GSE7305, GSE11691, GSE23339, and GSE25268 were amalgamated using the sva package to forge a composite validation cohort. Within this cohort, the 11 identified risk genes underwent further scrutiny through machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), and Generalized Linear Model (GLM)—with their efficacy evaluated via Receiver Operating Characteristic (ROC) curves.

Following machine learning validation, the risk genes’ cellular localization was elucidated using single-cell data from GSE214411, adhering to established protocols primarily involving the Seurat and SingleR packages (Fonseca et al. 2023). The integrity of the Mendelian Randomization (MR) outcomes was rigorously assessed through leave-one-out sensitivity analysis and the construction of funnel plots, ensuring the robustness and reliability of our findings in elucidating the genetic underpinnings of endometriosis.

Immunohistochemistry

For the immunohistochemical quantification of AGPAT4, we prepared 4 μm thick paraffin-embedded tissue sections, which were subjected to standard deparaffinization and rehydration protocols. Antigen retrieval was facilitated through microwave heating. To quench endogenous peroxidase activity, sections were immersed in 3% hydrogen peroxide for 10 min. Subsequently, the sections were incubated with a polyclonal rabbit anti-AGPAT4 antibody (Abmart, TD3640, 1:200) at 4 °C overnight, followed by a 30-minute incubation at room temperature with a horseradish peroxidase-conjugated secondary antibody targeting rabbit immunoglobulins. Visualization was achieved using diaminobenzidine (DAB) staining, counterstained with hematoxylin, and the sections were then dehydrated and mounted under coverslips. Imaging was performed with a Leica Upright Metallurgical Microscope (Wetzlar, Germany).

The evaluation of AGPAT4 expression involved assessing both the staining intensity and the proportion of positive cells. Staining intensity was categorized as 0 (no staining), 1 (weak), 2 (moderate), or 3 (strong), and the percentage of AGPAT4-positive cells was scored as 0 (none), 1 (≤ 25%), 2 (> 25% to < 50%), or 3 (≥ 50%) (Akbar et al. 2015). The immunoreactive score was determined by multiplying the intensity of staining by the percentage of positively stained cells.

Enzyme-linked immunosorbent assay

In the ELISA validation cohort, we included 38 endometriosis patients and 43 control subjects. Serum levels of AGPAT4 were quantitatively determined employing an ELISA kit (Abmart, TD3640, China) with a dilution of 1:2000, adhering strictly to the provided manufacturer’s protocol. The optical density at 450 nm (OD450), indicative of AGPAT4 concentration, was measured utilizing a microplate reader (Infinite M200 PRO, TECAN) subsequent to the application of the colorimetric substrate.

Cell isolation and culture

ESCs were cultured from eutopic endometrial samples of women with endometriosis. Samples were collected under aseptic conditions, washed, and transported on ice. ESCs were isolated, passaged using standard trypsinization, and cultivated in phenol red-free DMEM supplemented with 10% FBS at 37 °C with 5% CO2. ESC purity was validated through vimentin immunostaining (Abcam), and only cultures with a purity exceeding 95% were considered for inclusion in the study (Canosa et al. 2017).

Transfection experiments

In this investigation, RNA interference was executed via small interfering RNA (siRNA) transfection technique. Targeting the AGPAT4 gene, three distinct siRNAs were synthesized by Ribo Bio, China, supplemented with a control siRNA for comparative purposes. A cohort of 10^4 ESCs were seeded in six-well plate for 24 h prior to the transfection procedure. The transfection process involved both the AGPAT4-specific and control siRNAs using the riboFECT mRNA Transfection Reagent, procured from Ribo Bio, in strict adherence to the manufacturer’s guidelines. Subsequent to a 72-hour incubation post-transfection, cellular samples were subjected to Western blot analysis to ascertain the efficacy of gene suppression. Additional assays were conducted at 48 h following the harvesting of the cells.

Western blotting

Western blotting was performed in accordance with standard protocols. Briefly, proteins were extracted from lysed cells using a radioimmunoprecipitation assay buffer and then clarified by centrifugation at 12,000×g for 15 min at 4 °C. Protein levels in the supernatant were quantified using the bicinchoninic acid method (Themofisher). The proteins were then resolved by electrophoresis on 10% SDS-polyacrylamide gels and transferred to polyvinylidene difluoride membranes (Millipore Billerica). Membranes were blocked with 5% non-fat milk before being incubated with primary antibodies targeting GAPDH (Proteintech, 80570-1-RR,1:10,000), β-Catenin (Cell Signaling Technology, D10A8,1:2,000), MMP-9 (Proteintech, 10375-2-AP, 1:1,000), Wnt3a (Sangon Biotech, D122111,1:2,000), SNAI2 (Sangon Biotech, D221235,1:5,000), and AGPAT4 (Abmart, TD3640, 1:2,000). Following incubation with horseradish peroxidase-conjugated secondary antibodies (goat anti-rabbit) at room temperature for one hour, the blots were washed and developed. Protein bands were quantified using Quantity One software, with GAPDH serving as the normalization control.

Cell proliferation assay

The proliferative capacity of ESCs was evaluated using a Cell Adhesion assay, with a focus on their adhesion characteristics. This assessment was conducted via a CCK-8-based assay, executed in a 96-well plate format. To facilitate the assay, each well was treated with 50 µL of Matrigel, diluted in a serum-free medium at a 1:8 ratio. Following a 48-hour transfection period, a density of 4 × 10^3 cells per 200 µL was cultured in each well, which then underwent an incubation phase for 30 min. Subsequent to this incubation, non-adherent cells were gently washed away. Thereafter, each well received 20 µL of CCK-8 reagent (Biosharp, BS350A, China), followed by an additional incubation period of four hours. The adhesion efficiency of the ESCs was quantitatively analyzed by measuring the optical density (OD) at 450 nm using a spectrophotometric plate reader, with the OD values serving as an indicator of the number of adherent cells.

Wound-healing assay

To assess cellular migratory capabilities, a wound-healing assay was conducted on cells transfected with AGPAT4-specific siRNA. Cells were first cultured in six-well plates to achieve 90% confluency. Subsequently, a standardized wound was introduced into the cellular monolayer using a 200 µL plastic pipette tip. Post-wounding, the cells were rinsed with phosphate-buffered saline (PBS) to remove any detached cellular debris. Digital images of each well were captured at two time points: immediately after wounding (0 h) and 48 h post-wounding. The width of the wound was quantified utilizing Image-Pro Plus software. The migration rate was calculated using the formula: [Cell-free area at 0 h - Cell-free area at 48 h] / Cell-free area at 0 h, effectively measuring the reduction in wound width over the 48-hour period.

Transwell invasion assay

To determine the invasive potential of ESCs, a transwell invasion assay was meticulously conducted. The preparatory phase involved coating the upper chamber of the transwell setup with 60 µL of Matrigel, prepared at a 1:2 ratio with DMEM lacking phenol red, followed by an incubation period of one hour at 37 °C. Subsequently, ESCs were seeded into these upper chambers at a density of 10^3 cells per well. The lower chambers were supplemented with DMEM devoid of phenol red, enriched with 10% fetal bovine serum (FBS). Post a 72-hour incubation interval, cells residing in the upper chamber were carefully removed. The transwell filters underwent fixation using 4% paraformaldehyde for 30 min, followed by a double washing in phosphate-buffered saline (PBS). The staining process involved 0.5% hematoxylin, applied for a duration of 5 min. The invasive cells were then enumerated in three distinct fields, using a Leica Upright Metallurgical Microscope (Wetzlar, Germany). To quantify the invasion, the absorbance at 550 nm was measured utilizing a spectrophotometric plate reader. To bolster the experimental validity, this entire procedure was replicated thrice.

Molecular docking of AGPAT4 and Wnt3a

To investigate the interaction and structural relationship between AGPAT4 and Wnt3a, this study employed a high-precision molecular docking approach. Initially, the protein structures of AGPAT4 and Wnt3a were obtained from the RCSB PDB database (https://www.rcsb.org/). Subsequently, using the Auto-dock software, we conducted ten independent molecular docking simulations to ensure the reliability and accuracy of our results. This method allowed us to analyze the potential interactions between AGPAT4 and Wnt3a at a molecular level in detail.

Statistical analysis

The GEO data was subjected to rigorous analysis using R software (v4.2.1), adhering to the standards of robust data processing. Statistical computations and inferential analyses were conducted using SPSS software, version 22.0 (SPSS, Inc., Chicago, USA), a staple in quantitative research. Descriptive statistics are presented as mean ± standard deviation, providing a clear understanding of data variability and central tendency. The one-way Analysis of Variance (ANOVA) was the chosen statistical method to discern the differences among multiple experimental groups. A threshold of P < 0.05 was set for statistical significance, ensuring that the results were statistically robust and reliable. This level of significance was meticulously maintained throughout the analysis to uphold the integrity of the statistical findings. All experiments were conducted in duplicate.