Translation of miRNA blood-based discovery to molecular testing for clinical diagnosis of endometriosis

Differentially expressed miRNAs in endometriosis patients In the discovery cohort of samples co llected during the secretory phases (Supplementary Data 1), we identified a total of 85 differentially expressed miRNAs between the patient group and the control group (Fig. 1Aa n d Supplementary Data 2). Among these, 55 miRNAs were signi ficantly upregulated in the disease group, while 30 were downregulated, indicating a distinct molecular signature associated with endometriosis. Interestingly, some subjects exhibited higher expression levels of specificm i R N A sc o m - pared to others within the same group. This variability is likely attributable to the heterogeneous nature of endometriosis, which can manifest differ- ently across individuals in terms of severity, symptom presentation, and molecular profiles. Notably, several miRNAs previously implicated in the pathogenesis and progression of en dometriosis were found to be highly dysregulated in the disease group compared to the control group (Fig.1A, B). These included miR-9-5p and miR-21-5p, both of which have been reported to play critical roles in regulating in flammatory signaling path- ways, a key mechanism underlying the development and progression of endometriosis. Notably,miR-21-5p has been reported by several studies to be dysregulated in women suffering from endometriosis, underscoring its reproducibility as a potential biomarker 26,45. To further investigate the origin a nd functional relevance of these differentially expressed miRNAs, we performed a hypergeometric over- representation analysis using miRNet46,47. This analysis revealed that the majority of these miRNAs predominantly originated from the bone marrow and cervix (Fig.1C). The bone marrow association reflects the serum origin of the miRNA data. The cervix, on the other hand, is relevant as it points to the anatomical regions where endometriosis lesions can be found in proximity, providing confidence that the observed molecular signals are indeed linked to the disease. This suggests that the serum circulating miRNAs may serve as biomarkers for s ystemic changes associated with endometriosis. Additionally, we conducted a hypergeometric test using the gene tar- gets of the differentially expressed miRNAs against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to identify enriched molecular pathways 47. This analysis highlighted that the top non-tumor-speci fic molecular processes included various signaling pathways and focal adhe- sion, which are strongly associated with endometriosis (Fig. 1C). For instance, focal adhesion is a critical process in the pathogenicity of endo- metriosis, as it involves cell –matrix interactions that contribute to the attachment, survival, and invasion of endometrial cells outside the uterus 48. Similarly, dysregulation in signaling pathways, such as the mitogen- activated protein kinase (MAPK signaling pathway, has been observed in endometrial cells, leading to enhan ced cell proliferation, endometriosis lesion establishment, and disease persistence49. Furthermore, MAPK sig- naling is implicated in pain sensitization, suggesting a role in the chronic pelvic pain experienced by manypatients with endometriosis50. Assessment of differentially expressed miRNAs for the predic- tion of endometriosis using machine learning To evaluate whether the identified differentially expressed miRNAs possess predictive value in distinguishing individuals with endometriosis from control subjects, we implemented and assessed three distinct random forest models, each utilizing a different dataset. For the first model (Model #1), we constructed a random forest clas- sifier using all 85 differentially expressed miRNAs. To rigorously assess its performance, we employed a 30-fold repeated subsampling cross-validation approach. Model #1 demonstrated strong predictive capability, achieving an overall sensitivity of 0.91, specificity of 0.88, and an area under the curve (AUC) of 0.95 (Fig. 2A). To further refine the model and enhance inter- pretability, we explored feature selection based on importance scores derived from the initial model. Speci fically, we developed two additional models with reduced feature sets: Model #2, which utilized the top 40 most informative miRNAs, and Model #3, which incorporated only the top 20 (Supplementary Data 3). These refined models were constructed using the same random forest framework and evaluated with the identical repeated subsampling cross-validation procedure. By focusing on the most predictive miRNAs, we aimed to improve classi fication accuracy while eliminating noise introduced by less informative features. The results indicated that both red uced-feature models exhibited slightly improved performance compared to Model #1. Model #2, incor- porating the top 40 miRNAs, achieved a sensitivity of 0.93, a specificity of 0.89, and an AUC of 0.97 (Fig.2B). Model #3, which was restricted to the top 20 miRNAs, further enhanced predictive accuracy, yielding a sensitivity of 0.95, a speci ficity of 0.90, and an AUC of 0.98 (Fig. 2C). These findings suggest that a substantial proportion of the differentially expressed miRNAs may not meaningfully contribute to disease classi fication and instead introduce noise into the predictive model. The improved performance of the reduced-feature models underscores the value of feature selection in enhancing both the robustness and interpretability of machine learning- based biomarker discovery in endometriosis. Selection of miRNA biomarkers and endogenous controls for non-NGS clinical diagnostics The translation of NGS discovery findings into actionable clinical diag- nostics is essential for improving patient care. While NGS provides com- prehensive molecular insights, clinical diagnostic applications often rely on qPCR due to their faster turnaround time, lower costs, and suitability for IVD implementation without requiring the complex bioinformatics infra- structure necessary for NGS analysis 28–30. Given these advantages, we explored the feasibility oftranslating our NGS-basedfindings into a qPCR- based diagnostic assay. One of the primary considerations in this transition is the limit of detection of qPCR. The 85 differentially expressed miRNAs identified in our NGS analysis exhibit a wide range of expression levels across samples, spanning from as few as 10 normalized read counts (e.g.,miR-4710)t oo v e r 1,000,000 (e.g.,miR-21-5p) (Supplementary Data 2). Based on initial qPCR assessments, we determined that reliable detection in qPCR is achieved for miRNAs with an average NGS normalized read count exceeding 500. Applying this threshold, 38 of the dif ferentially expressed miRNAs fell below the cutoff and were deemed unreliable for qPCR-based detection (Fig. 3A). This left 47 differentially expressed miRNAs that met the detection threshold for further investigation (Supplementary Data 4). To assess the predictive potential of these 47 miRNAs in a qPCR setting, we constructed two random forest models. Thefirst model incor- porated all 47 differentially expressed miRNAs, while the second utilized a https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 3 reduced feature set consisting of the20 most informative miRNAs selected from this pool. Notably, many of the miRNAs identified as top contributors in the previous analysis using all 85 markers—such as miR-21-5p, miR-17- 5p,a n dmiR-15b-5p—remained among the top-ranked features, suggesting that key predictive biomarkers identified via NGS may indeed be transla- table to a clinical qPCR assay. Performance evaluation of these models revealed that thefirst model (using all 47 miRNAs) achieved an AUC of 0.78 (Fig. 3B), whereas the second model (using the top 20 miRNAs) exhibited improved performance w i t ha nA U Co f0 . 8 4( F i g .3C). As observed previously, reducing the feature set led to a slight increase in performance, likely due to the removal of uninformative or noisy variables. However, it is important to note that these qPCR-based models demonstrated lower overall performance compared to the models trained on all 85 differentially expressed miRNAs in the NGS dataset. This suggests that some of the most informative miRNAs are expressed at low levels in the samples, such asmiR-9-5p,w h i c hw a sr a n k e d as the most predictive but had an average normalized read count of only ~350. The inability of qPCR to reliably detect miRNAs with low expression presents a challenge for clinical translation. Addressing this limitation may require optimizing primer designs, adopting ddPCR for enhanced sensi- tivity, or incorporating target capture methods to improve miRNA detection. A second critical factor in transitioning to a qPCR-based assay is the selection of a suitable endogenous co ntrol or reference marker for data normalization. Inappropriate reference markers can introduce variability, leading to unreliable quantification of expression and potentially skewing diagnostic outcomes. To address th is, we implemented a bioinformatic pipeline for identifying disease-specific endogenous controls (Ref. “Meth- ods”). Brie fly, we selected miRNAs that display minimal inter-group variability in expression to be used as endogenous references (Supplementary Data 5). One such candidate, miR-92a-3p (Fig. 3D), demonstrated consistent expression levels between individuals with endo- metriosis and control subjects. Notably,miR-92a-3p has been validated in prior studies as a reliable endogenous control in blood-based miRNA analyses 51. To further refine our models, we performed in silico normalization of the 47 differentially expressed miRNAs againstmiR-92a-3p, simulating the clinical qPCR diagnostic work flow. Following normalization, we re- evaluated the performance of our random forest models. This approach led to noticeable improvements in predictive accuracy, with the model using all 47 miRNAs achieving an AUC of 0.80, and the reduced model using the top 20 miRNAs attaining an AUC of 0.88 (Fig. 3E, F). These findings underscore the importance of proper normalization strategies in enhancing the robustness of qPCR-based diagnostics and further support the potential clinical applicability of our NGS-derived biomarker panel. Experimental assessment for diagnostic qPCR assays To evaluate the potential of translating our NGS discoveryfindings into a clinically viable qPCR diagnostic assay, we selectedfive candidate miRNAs —miR-21-5p, miR-15b-5p, miR-17-5p, miR-19b-3p,a n d miR-23a-3p—for experimental validation using qPCR (Fig.4). These miRNAs were chosen based on their differential expression patterns observed in the NGS dataset, as well as their biological relevance to endometriosis. In addition,miR-92a- 3p w a si n c l u d e da sa ne n d o g e n o u sc o n t r o lf o rn o r m a l i z a t i o n ,g i v e ni t s demonstrated stability across endometriosis and control samples in both our dataset and prior studies. For this validation study, we conducted qPCR assays on 90 serum samples, comprising 65 patients with endometriosis and 25 control subjects whose disease status was confirmed via laparoscopic surgery (Supplemen- tary Data 6). The goal was to determi ne whether the expression patterns A B C AUC Sensitivity Specificity 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 AUC Sensitivity Specificity AUC Sensitivity Specificity 0.00 0.25 0.50 0.75 1.000.95 0.91 0.88 0.97 0.93 0.89 0.98 0.95 0.90 Cross validation #10 (Representative) 0.2 0.4 0.6 1.0 0.8 True positive rate False positive rate 0.2 0.4 0.6 0.8 1.0 Model 1: All DE miRNAs Model 2: Top 40 DE miRNAs Model 2: Top 20 DE miRNAs 0.2 0.4 0.6 1.0 0.8 True positive rate 0.2 0.4 0.6 1.0 0.8 True positive rate False positive rate 0.2 0.4 0.6 0.8 1.0 False positive rate 0.2 0.4 0.6 0.8 1.0 AUC = 0.93 AUC = 0.94 AUC = 0.94 Cross validation #12 (Representative) Cross validation #12 (Representative) Fig. 2 | Diagnostic performance of NGS-identi fied differentially expressed miRNAs using machine learning. The diagnostic utility of differentially expressed miRNAs was assessed through machine learning analysis. Performance metrics represent the mean of 30 iterations of repeated subsampling cross-validation, with error bars indicating variability across iterations. A Using the full set of differentially expressed miRNAs, the model achieved an AUC of 0.95, with a sensitivity of 0.91 and specificity of 0.88. A representative ROC curve from iteration #10 is shown. B Limiting the model to the top 40 most informative miRNAs improved perfor- mance, yielding an AUC of 0.97, sensitivity of 0.93, and specificity of 0.89. The ROC curve displayed corresponds to iteration #12. C Further refinement using the top 20 most informative miRNAs resulted in the highest diagnostic performance, with an AUC of 0.98, sensitivity of 0.95, and speci ficity of 0.90. The representative ROC curve is also from iteration #12. https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 4 observed in the NGS discovery datasetcould be reliably recapitulated using qPCR, a crucial step in transitioning towards a clinically deployable assay. Analysis of the qPCR results revealed that two of thefive selected biomarkers (miR-21-5pand miR-15b-5p) displayed statistically significant differences in normalized expression between the endometriosis and control groups. These findings are consistent with our NGS discovery dataset, reinforcing their potential utility as diagnos tic biomarkers. Additionally, miR-17-5p exhibited a discernible trend of differential expression between disease and control samples, as visualized in the boxplot analysis. However, this dif- ference did not reach statistical signi ficance, suggesting that, while this marker may hold some biological relevance, additional optimization—such as larger sample sizes or re fined qPCR conditions—may be necessary to establish its diagnostic value. In contrast, the remaining two miRNAs,miR- 19b-3p and miR-23a-3p, did not show clear delineation between patients and controls, indicating that their differential expression in the NGS dataset may not translate robustly into a qP CR-based assay. This discrepancy underscores the complexities of biomarker translation and highlights the need for rigorous validation and opti mization at the experimental level. Factors such as primer design, ampli fication efficiency, RNA extraction variability, and technical noise could all contribute to differences between NGS and qPCR results, emphasizing the importance of careful assay development. Notwithstanding, the data suggest that qPCR-based assays could provide meaningful diagnostic utility with proper refinement.

S m a l lR N A ss u c ha sm i R N A sh a v eb e e ns h o w nt op l a yp i v o t a lr o l e si n numerous physiological and pathological processes, influencing gene reg- ulation in both normal and disease states. Altered miRNA expression profiles have been extensively documented in the plasma and serum of patients with various conditions, including diffuse large B cell lymphoma, ovarian cancer, and type 2 diabetes 52,53.T h e s efindings underscore the potential of circulating miRNAs as no n-invasive biomarkers for disease detection and monitoring. In this s tudy, we demonstrate that miRNAs circulating in serum can potentially serve as reliable biomarkers for the diagnosis of endometriosis. The ability to analyze serum miRNA levels in a standardized manner presents a promising approach in disease detection. The fluctuations in specific circulating miRNAs offer a quanti fiable and reproducible means of identifying en dometriosis, potentially improving early diagnosis and clinical management. Using serum miRNAs as diagnostic biomarkers offers several key advantages over conventional diagnostic methods in endometriosis. Cur- rently, the gold standard for endometriosis diagnosis relies on laparoscopy A B C D E F Detectable via qPCR Not detectable via qPCR log2(average expression) DE miRNAs 0 5 10 15 20 0.00 0.25 0.50 0.75 1.00 AUC Sensitivity Specificity 0.00 0.25 0.50 0.75 1.00 AUC Sensitivity Specificity 0.00 0.25 0.50 0.75 1.00 AUC Sensitivity Specificity AUC Sensitivity Specificity 0.00 0.25 0.50 0.75 1.00 All qPCR detectable miRNAs Top 20 qPCR detectable miRNAs All qPCR detectable miRNAs normalized with miR-92a-3p Top 20 qPCR detectable miRNAs normalized with miR-92a-3p 0.78 0.66 0.69 0.84 0.68 0.74 0.80 0.70 0.68 0.88 0.77 0.70 Controls Endometriosis 500000 750000 1000000 1250000Normalized read count miR-92a-3p Fig. 3 | Diagnostic performance of differentially expressed miRNAs is reliably quantifiable by qPCR, assessed using machine learning. Performance metrics represent the mean of 30 iterations of repeated subsampling cross-validation, with error bars indicating variability across iterations. A Violin plot illustrating the expression distribution of differentially expressed miRNAs based on normalized NGS read counts. miRNAs shown in blue represent those with expression levels too low for reliable qPCR detection, while those in red indicate miRNAs with suf ficient expression for reliable quanti fication by qPCR. The dotted red line marks the expression threshold of 500 normalized reads, used to distinguish between the two groups. B Predictive performance of a machine-learning model using all 47 miRNAs deemed reliably detectable by qPCR, resulting in an average AUC of 0.78, with a sensitivity of 0.66 and specificity of 0.69.C Model performance using the top 20 most informative miRNAs from the qPCR-detectable set, showing a modest improvement with an average AUC of 0.84, sensitivity of 0.68, and specificity of 0.74. D Boxplot of miR-92a-3p, a potential endogenous control candidate in this disease setting, demonstrating consistent expression with minimal variability between patient and control groups. The box ’s lower and upper hinges correspond to the 25th and 75th percentiles, respectively, with the median indicated by the line inside the box. The whiskers extend to the most extreme data points within 1.5 times the interquartile range below the 25th percentile and above the 75th percentile. E Predictive per- formance using all 47 reliably detectable miRNAs after in silico normalization against miR-92a-3p, yielding an average AUC of 0.80, with sensitivity of 0.70 and specificity of 0.68. F Model performance using the top 20 most informative miRNAs following in silico normalization with miR-92a-3p, showing further improvement with an average AUC of 0.88, sensitivity of 0.77, and speci ficity of 0.68. https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 5 with direct visualization, an invasive surgical procedure. A serum-based miRNA biomarker assay could provide anon-invasive alternative, enabling comprehensive disease assessment without the need for surgery. This is particularly valuable for early detection and for patients who may not have immediate access to specialized surgical evaluation. Secondly, compared to invasive diagnostic procedures, aserum-based miRNA test is significantly more cost-effective. The process involves routine blood collection and standard laboratory processing, making it more accessible for widespread clinical implementation. Additiona lly, standardizing miRNA detection protocols could facilitate large-scal e screening efforts, improving early diagnosis and patient outcomes. In this study, we investigated serum miRNA expression pro files in individuals with endometriosis and identified a distinct set of circulating miRNAs that may serve as potential biomarkers for disease detection. To minimize variability associated with hormonal changes during the men- strual cycle, particularly those impacting female-speci fic physiological processes, serum samples were collected exclusively during the secretory phase. Our results add to the growing evidence supporting the use of serum miRNA signatures as non-invasive diagnostic tools for endometriosis 19–33. In particular, we experimentally validated miR-21-5p and miR-15b-5p, which contain significant information delineating endometriosis patients from the control population. Both miR-21-5p and miR-15-5p have been previously reported to be dysregulated in a variety of pathological condi- tions, particularly those characterized by in flammation or aberrant cell proliferation. For example,miR-21-5pis widely considered as an‘oncomiR’ with roles in cancer, immune activation, and angiogenesis 54,55. Likewise, miR-15-5p has been associated with tissue fibrosis and immune modulation56,57. The presence of these miRNAs across multiple disease contexts underscores their lack of disease speci ficity. However, their reproducible dysregulation in endometriosis is biologically plausible given the in flammatory and fibrotic microenvironment that characterizes this disease, and may serve as important components of a multi-biomarker panel, capturing the inflammatory and remodeling milieu of endometriosis when combined with other biomarkers. Nonetheless, despite their promise, several key challenges remain before miRNA-based assays can be translated into clinically reliable diag- nostic applications. One of the primary obstacles lies in the choice of detection platform. While NGS provides a comprehensive assessment of the miRNA landscape, its high cost per sample and dependence on complex bioinformatics infrastructure render it impractical for routine clinical diagnostics and IVD applications. Consequently, there is a need to transi- tion toward more practical methodol o g i e s ,s u c ha sq P C Ro rd d P C R ,w h i c h offer lower costs, faster turnaround times, and compatibility with IVD requirements. Another potential challenge in clinical miRNA diagnostics is the selection of appropriate endogenous controls for normalization, as the commonly usedmiR-16-5pcan exhibit instability in blood-based assays 58.I n our study,miR-92a-3pemerged as a more reliable endogenous control based on empirical comparison betweenthe patient and control groups51,w h e r e a s miR-16-5pshowed greater variability. It should be noted that bothmiR-16- 5p and miR-92a-3pare linked to inflammatory processes59,60,w i t hmiR-92a- 3p implicated in neuroin flammation60,61. While our results support the suitability ofmiR-92a-3pas an endogenous control in this context, valida- tion in larger, independent cohorts is necessary to con firm its broader applicability. Our findings suggest that convertingNGS-based miRNA discoveries into clinically applicable assays is achievable, though it necessitates experi- mental refinement. Despite using a limited qPCR panel with only five miRNA biomarkers, we demonstrated valu a b l ed i a g n o s t i cp o t e n t i a l ,a l b e i t with a need for further optimization. This highlights both the promise and the technical challenges of implementing miRNA-based diagnostics in clinical practice. To improve the reliability and diagnostic accuracy of a qPCR-based test, several key strategies should be c onsidered. (a) Primer optimization: refining primer designs to enhance amplification efficiency and specificity, particularly for low-abundance miRNAs. (b) Adopting ddPCR for enhanced sensitivity: ddPCR offers improved precision and sensitivity, making it a suitable alternative for detecting low-expressed miRNAs that may be missed by qPCR. (c) Expanding sample size: increasing the number of clinical samples analyzed will im prove statistical power and ensure robustness across diverse patient populations. (d) Developing a multiplex assay: creating a multiplexed qPCR panel would allow for the simultaneous detection of multiple miRNA biomarkers, streamlining work flow and improving diagnostic efficiency. While initial qPCR validation of selected miRNAs shows promise, further refinement is necessary and underway to enhance assay reprodu- cibility and clinical performance. Future studi es will focus on validating these biomarkers in larger, independent cohorts, optimizing detection methods, and standardizing protocol s to ensure reproducibility across clinical testing laboratories. These efforts will be critical in bridging the gap between high-throughput discovery research and real-world clinical application, ultimately paving the way for a clinically deployable serum miRNA-based diagnostic test for endometriosis.

Specimen collection Peripheral blood samples were collected prospectively from women aged 18–49 years who presented with mild-to-severe symptoms, including pelvic pain and/or menstrual bleeding. This is a single-center study, and the par- ticipants were enrolled under an approved institutional review board (IRB) of the Women ’s Hospital of Zhejiang University School of Medicine in accordance with the Declaration of Helsinki, with informed consent obtained from all individuals. The IRB number is IRB-20240110-R. The Fig. 4 | Boxplot of qPCR performance for selected differentially expressed miRNAs. All miRNAs were normalized against miR-92a-3p using qPCR. Of the five miRNAs tested, two ( miR-21-5p and miR-15b-5p) showed a signi ficant difference in expression between patients and controls, replicat- ing the direction of effect observed in the NGS dis- covery dataset. miR-17-5p exhibited a concordant trend, with a p value of 0.065, approaching statistical significance. The remaining two miRNAs, miR-19b- 3p and miR-23a-3p, did not show a statistically sig- nificant difference between patients and controls. The boxplot’s lower and upper hinges represent the 25th and 75th percentiles, respectively, with the median indicated by the line inside the box. The whiskers extend to the most extreme data points within 1.5 times the interquartile range below the 25th percentile and above the 75th percentile. miR-21-5p miR-15b-5p miR-17-5p miR-19b-3p miR-23a-3p 2.7 3.0 3.3 3.6 -log2(Delta CT) miRNAs normalized with miR-92a-3p p = 0.029 p = 0.032 p = 0.073 p = 0.729 p = 0.692 Controls n = 25 Endometriosis n = 65 3.9 https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 6 participants included in this study were collected from March 2024 to December 2024. All participants were clinically suspected of a gynecologic abnormal condition and were scheduled to undergo laparoscopy with his- topathological confirmation for endometriosis. A subset of participants in our study cohort presented with co-morbidities such as adenomyosis and leiomyoma. These conditions frequently co-exist with endometriosis and may present overlapping clinical features, making it dif ficult to fully dis- entangle their individual contributions. It is therefore acknowledged that the potential influence of these co-morbidities on the observed outcomes can- not be ruled out, and this represents alimitation of the study that should be considered when interpreting the results. Detailed clinical information for a l le n r o l l e dp a r t i c i p a n t si sp r o v i d e di nt h eS u p p l e m e n t a r yD a t a .T oa c c o u n t for potential variations in miRNA e xpression due to different menstrual cycle phases, blood samples were collected exclusively from women in the secretory phase of their menstrual cycle. The menstrual phase was initially determined by physicians or surgeons based on self-reported cycle days and clinical assessment. However, relying solely on calendar-based timing may not provide sufficient accuracy. To strengthen the reliability of our sample selection, this secretory classification was further validated through serum progesterone measurements using the protein assay from Kangrun Biotech Co., Ptd (Guangdong, China), with levels exceeding 1.08 ng/mL serving as a biochemical confirmation of the secretory phase according to the manu- facturer’s protocol. This appr oach minimized the in fluence of hormonal fluctuations on biomarker expression, thereby enhancing the reliability of our findings. We acknowledge that serum progesterone levels alone may not precisely distinguish between early, mid, and late secretory phases. For example, a progesterone concentration of 2 ng/mL could represent early or late secretory windows due to the cyclical nature of the hormone level, where dynamic changes in progesterone signaling, immune cell infiltration, and stromal remodeling are well docume nted. Therefore, we recognize that residual heterogeneity due to broad secretory phase classification is a lim- itation of the study and may confound interpretation of the data. Given the constraints of patient recruitment a nd sample availability, we adopted a pragmatic approach that combined calendar-based cycle staging with bio- chemical validation using serum progesterone to ensure all participants were indeed in the secretory phase. Future studies with larger cohorts and additional markers of endometrial dating will be needed to minimize this source of heterogeneity. In this study, a total of 40 symptomatic women were included in the NGS discovery cohort, with 10 mL of blood drawn into standard red-top blood collection tubes prior to the laparoscopic surgery. Among them, 20 women were confirmed to have endometriosis based on both laparoscopic findings and histopathology (disease group), while the remaining 20 had no evidence of endometriosis and served as the control group. Serum was isolated using a two-step centrifugation protocol. First, samples underwent a low-speed centrifugation at 3000 rpm for 10 min at 4 °C to remove cellular components. This was then followed by a second high-speed centrifugation at 16,000 ×g for 10 min at 4 °C to ensure com- plete removal of debris and platelets. The isolated serum was then aliquoted and stored at −80 °C for subsequent RNA extraction and downstream processing. RNA isolation and miRNA-seq Total RNA was isolated from 300 μL of serum using the Norgen RNA extraction kit following the manufacturer’s instructions. Total RNAs were ligated to 3’ adapters by denaturation at 70 °C for 2 min, and then incubated at 16 °C for over 8 h using NEB T4 RNA Ligase 2. Afterwards, 5’ adapters were incubated with the previous product using NEB T4 RNA Ligase 1 at 37 °C for 60 min. Ligated RNA was reverse transcribed in a thermocycler using SuperScript II Reverse Transcriptase from ThermoFisher Inc. under the following conditions: an initial incubation at 50 °C for 60 min, followed by a heat inactivation step at 80 °C for 10 min. Following complementary DNA (cDNA) synthesis, library preparation was performed using the NEB Phusion High-Fidelity DNA Polymerase, adhering strictly to the manu- facturer’s guidelines. The final libraries were then subjected to high- throughput sequencing to profile the miRNA expression by LC Biosciences. NGS miRNA differential expression profiling endogenous control selection The raw FASTQ data obtained from miRNA sequencing underwent pre- processing and analysis using the miRge362 software pipeline. Initially, the sequences were quality-trimmed, and adapter sequences were removed using a Cutadapt63 wrapper integrated within miRge3. The trimmed reads were then aligned to the miRBase 64 reference database using Bowtie 65 optimized for short reads. Following alignment, miRge3 generated a com- prehensive count table summarizing the abundance of each miRNA across the samples. This count table served as the input for downstream differential expression analysis using the DESeq2 66 package in R. DESeq2 was employed to identify miRNAs that exhi bited statistically significant differences in expression between the patient and con trol groups. After identifying the differentially expressed miRNAs, hy pergeometric over-representation analyses were conducted using the miRNet platform 46. In this analysis, two distinct queries were performed: first, the differentially expressed miRNAs were compared against the miRNA-tissue origin database in miRNet to infer potential tissue-speci fic origins or associations of these miRNAs; second, the predicted gene targets of these miRNAs were mapped to the KEGG database 47 to identify statistically enriched biological pathways. This dual-level approach enabled th e contextualization of the miRNA expression patterns in terms of both tissue relevance and functional pathway involvement. Endogenous control selection for in silico normalization and qPCR experimental validation To bridge the findings from the next-generation sequencing (NGS) dis- covery cohort into clinically applicable diagnostic tools, we further aimed to identify condition-specific endogenous control miRNAs. These controls are essential for normalizing qPCR or ddPCR assays, ensuring accurate and reproducible quantification of target miRNA expression. The selection of endogenous controls was guided by stringent criteria. Speci fically, we evaluated the expression stability of candidate miRNAs by assessing their dispersion estimates and imposing constraints on log2 fold-change (| log2FC| < 0.02) between the patient andcontrol groups. This ensured that the selected controls exhibited minimal variability across conditions. Additionally, candidates were filtered based on an adjusted p-value threshold (≥0.8), ensuring that their expression was not influenced by the experimental conditions or disease state. Disease prediction model construction using a random forest classifier To assess the predictive capability of the differentially expressed miRNAs and determine the extent to which they could accurately classify patients with endometriosis, we constructed a random forest classifier using all the differentially expressed miRNAs identi fied in the discovery cohort. To ensure a robust evaluation of the model’s performance, we performed 30 iterations of repeated random subsampling cross-validation, where in each iteration, the data was split into an 80:20 ratio for training and testing, respectively. This repeated holdout validation approach allowed us to account for variability in model performance due to random data parti- tioning and provided a reliable estimate of the model’s predictive accuracy. During model construction, any missing values were imputed using the median of the corresponding feature. Following the initial model construction, we conducted feature selec- tion to identify the most informative miRNAs for predicting endometriosis. This was achieved by evaluating the feature importance scores generated by the random forest algorithm. Based on these scores, we created two distinct feature sets: one comprising the top 40 most important miRNAs and another consisting of the top 20 most important miRNAs. These reduced feature sets were then used to generate and assess new models using the same repeated random subsampling cross-validation procedure described above. Reducing the feature set is crucial for the assessment of model per- formance and interpretability. A smaller, more informative subset of miRNAs helps prevent overfitting, ensuring the model generalizes well to https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 7 new data. Additionally, focusing on the most important miRNAs reduces noise, leading to more reliable predictions. A reduced feature set also facilitates translation into clinical diagnostics, where the informative miR- NAs can be assayed using qPCR and/or ddPCR instead of NGS. RNA extraction and qPCR experimental validation Total RNA was extracted from 200μL serum using the miRNeasy Serum/ Plasma Advanced Kit from Qiagen, following the manufacturer’s recom- mended protocol. Subsequently, targeted miRNAs were reverse transcribed into cDNA using the FastKing RT Kit II from TianGen Inc. The reverse transcription reaction was carried out in a thermocycler under the following conditions: an initial incubation at 42 °C for 15 min to facilitate cDNA synthesis, followed by a heat inactivation step at 95 °C for 1 min to terminate the reactions. The synthesized cDNA was subjected to qPCR analysis of the five miRNA markers. The PCR reaction mixtures were prepared by miR- CURY LNA miRNA SYBR Green PCR Kit from Qiagen. qPCR ampli fi- cation was performed in the QuantStudio qPCR system following an initial denaturation at 95°C for 2 min, followed by 40 cycles with denaturation at 95 °C for 10 s and annealing at 56 °C for 60 s. Statistical analyses Differential expression analysis was performed using DESeq2, which models NGS count data with a negative binomial distribution to assess statistical differences between patients and controls. For machine learning- based predictions, sensitivity, specificity, and AUC were evaluated using Python’s scikit-learn package. Pairwise expression comparisons between patients and controls were assessed using the Wilcoxon rank-sum test. Study approval The participants were enrolled under an approved institutional review board protocol (IRB-20240110-R) at the Women ’sH o s p i t a lo fZ h e j i a n g University School of Medicine, with i nformed consent obtained from all individuals. Data availability Sequencing data were deposited into Genome Sequence Archive under the accession number HRA011242 and can be accessed viahttps://ngdc.cncb. ac.cn/search/specific?db=hra&q=HRA011242. Code availability Scripts used for feature selection, random forest model construction, and figure plotting have been deposited in the GitHub repository (https://github. com/Heranova-Lifesciences/endometriosis_miRNAseq_manuscript). Received: 4 May 2025; Accepted: 13 November 2025;

1. Zondervan, K. T. et al. Endometriosis. N. Engl. J. Med. 382, 1244–1256 (2020). 2. Bonavina, G. & Taylor, H. S. Endometriosis-associated infertility: from pathophysiology to tailored treatment.Front. Endocrinol. 13, 1020827 (2022). 3. Parasar, P., Ozcan, P. & Terry, K. L. Endometriosis: epidemiology, diagnosis and clinical management. Curr. Obstet. Gynecol. Rep. 6, 34–41 (2017). 4. Fryer, J., Mason-Jones, A. J. & Woodward, A. Understanding diagnostic delay for endometriosis: a scoping review using the social- ecological framework. Health Care Women Int. 46, 335–351 (2025). 5. Kirk, U. B. et al. Understanding endometriosis underfunding and its detrimental impact on awareness and research. npj Women’s Health 2, 45 (2024). 6. Davenport, S., Smith, D. & Green, J. D. Barriers to a timely diagnosis of endometriosis: a qualitative systematic review. Obstet. Gynecol. 142, 571–583 (2023). 7. Pascoal, E. et al. Strengths and limitations of diagnostic tools for endometriosis and relevance in diagnostic test accuracy research. Ultrasound Obstet. Gynecol. 60, 309–327 (2022). 8. Leyland, N., Casper, R., Philippe, L. & Singh, S. S. & SOGC. Endometriosis: diagnosis and management. J. Obstet. Gynaecol. Can. 32,S 1–S32 (2010). 9. Dunselman, G. A. et al. ESHRE guideline: management of women with endometriosis. Hum. Reprod. 29, 400–412 (2014). 10. Taylor, H. S. et al. An evidence-based approach to assessing surgical versus clinical diagnosis of symptomatic endometriosis. Int. J. Gynecol. Obstet. 142, 131–142 (2018). 11. Mak, J., Leonardi, M. & Condous, G. ‘Seeing is believing’: arguing for diagnostic laparoscopy as diagnostic test for endometriosis. Reprod. Fertil. 3, C23–C28 (2022). 12. Agarwal, S. K. et al. Clinical diagnosis of endometriosis: a call to action. Am. J. Obstet. Gynecol. 220, 354.e1–354.e12 (2019). 13. Gerges, B., Li, W., Leonardi, M., Mol, B. W. & Condous, G. Meta- analysis and systematic review to determine the optimal imaging modality for detection of uterosacral ligaments/torus uterinus, rectovaginal septum and vaginal deep endometriosis. Hum. Reprod. Open 2021, hoab041 (2021). 14. Dantkale, K. S. & Agrawal, M. A comprehensive review of the diagnostic landscape of endometriosis: assessing tools, uncovering strengths, and acknowledging limitations. Cureus 16 , e56978 (2024). 15. Albee, R. B. Jr., Sinervo, K. & Fisher, D. T. Laparoscopic excision of lesions suggestive of endometriosis or otherwise atypical in appearance: relationship between visual findings and final histologic diagnosis. J. Minim. Invasive Gynecol. 15,3 2–37 (2008). 16. Gratton, S. et al. Diagnosis of endometriosis at laparoscopy: a validation study comparing surgeon visualization with histologic findings. J. Obstet. Gynaecol. Can. 44, 135–141 (2022). 17. Sutton, C. J., Ewen, S. P., Whitelaw, N. & Haines, P. Prospective, randomized, double-blind, controlled trial of laser laparoscopy in the treatment of pelvic pain associated with minimal, mild, and moderate endometriosis. Fertil. Steril. 62, 696–700 (1994). 18. Afors, K. et al. Employing laparoscopic surgery for endometriosis. Women’s Health 10, 431–443 (2014). 19. Vanhie, A. et al. Plasma miRNAs as biomarkers for endometriosis. Hum. Reprod. 34, 1650–1660 (2019). 20. Avery, J. C. et al. Noninvasive diagnostic imaging for endometriosis part 1: a systematic review of recent developments in ultrasound, combination imaging, and artificial intelligence.Fertil. Steril.121, 164–188 (2024). 21. Oskouei, B. S., Asadi, Z. & Esfahlan, R. J. Non-invasive blood tests for earlier diagnosis and treatment of endometriosis. J. Reprod. Immun. 169, 104521 (2025). 22. Bjorkman, S. & Taylor, H. S. MicroRNAs in endometriosis: biological function and emerging biomarker candidates. Biol. Reprod. 100, 1135–1146 (2019). 23. Bendifallah, S. et al. MicroRNome analysis generates a blood-based signature for endometriosis. Sci. Rep. 12, 4051 (2022). 24. Nazri, H. M. et al. The role of small extracellular vesicle-miRNAs in endometriosis. Hum. Reprod. 38, 2296–2311 (2023). 25. Rekker, K. et al. Circulating miR-200-family micro-RNAs have altered plasma levels in patients with endometriosis and vary with blood collection time. Fertil. Steril. 104, 938–946 (2015). 26. Papari, E., Noruzinia, M., Kashnani, L. & Foster, W. G. Identi fication of candidate microRNA markers of endometriosis with the use of next- generation sequencing and quantitative real-time polymerase chain reaction. Fertil. Steril. 113, 1232–1241 (2020). 27. Perricos, A. et al. Hsa-mir-135a shows potential as a putative diagnostic biomarker in saliva and plasma for endometriosis. Biomolecules 12, 1144 (2022). 28. Vanhie, A. et al. Circulating microRNAs as non-invasive biomarkers in endometriosis diagnosis – a systematic review. Biomedicines 12, 888 (2024). https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 8 29. Leonova, A., Turpin, V. E., Agarwal, S. K., Leonardi, M. & Foster, W. G. A critical appraisal of the circulating levels of differentially expressed microRNA in endometriosis. Biol. Reprod. 105, 1075–1085 (2021). 30. Zafari, N. et al. microRNAs as novel diagnostic biomarkers in endometriosis patients: a systematic review and meta-analysis. Expert Rev. Mol. Diagn. 22, 479–495 (2022). 31. Nothnick, W. B., Falcone, T., Joshi, N., Fazleabas, A. T. & Graham, A. Serum miR-451a levels are signi ficantly elevated in women with endometriosis and recapitulated in baboons ( Papio anubis) with experimentally-induced disease. Reprod. Sci. 24, 1195–1202 (2017). 32. Zhang, L. et al. Serum exosomal MicroRNAs as potential circulating biomarkers for endometriosis. Dis. Markers 2020, 2456340 (2020). 33. Chu, X. et al. Extracellular vesicles in endometriosis: role and potential. Front. Endocrinol. 15, 1365327 (2024). 34. Wang, W., Zhao, Y., Han, B., Hong, S. & Chen, Y. Circulating microRNAs identified in a genome-wide serum microRNA expression analysis as noninvasive biomarkers for endometriosis. J. Clin. Endocrinol. Metab. 98, 281–289 (2013). 35. Hsu, C. Y. et al. miRNA-199a-5p regulates VEGFA in endometrial mesenchymal stem cells and contributes to the pathogenesis of endometriosis. J. Pathol. 232, 330–343 (2014). 36. Petracco, R. et al. Evaluation of miR-135a/b expression in endometriosis lesions. Biomed. Rep. 11, 181–187 (2019). 37. Lamon, S. et al. The effect of the menstrual cycle on the circulating microRNA pool in human plasma: a pilot study. Hum. Reprod. 38, 46–56 (2022). 38. Bendifallah, S. et al. Machine learning algorithms as new screening approach for patients with endometriosis. Sci. Rep. 12, 639 (2022). 39. Ferrier, C. et al. Saliva microRNA signature to diagnose endometriosis: a cost-effectiveness evaluation of the Endotest.BJOG 130, 396–406 (2023). 40. Zalis, M. et al. Next-generation sequencing impact on cancer care: applications, challenges, and future directions. Front. Genet. 15, 1420190 (2024). 41. Desai, K. et al. Real-world trends in costs of next-generation sequencing (NGS) testing in U.S. setting.J. Clin. Oncol. 39, e18824 (2021) 42. Karlovich, C. A. & Williams, P. M. Clinical applications of next- generation sequencing in precision oncology. Cancer J. 25, 264–271 (2019). 43. Chapman, J. R. & Waldenstrom, J. With reference to reference genes: a systematic review of endogenous controls in gene expression studies. PLoS ONE 10, e0141853 (2015). 44. Taylor, S. C. et al. The ultimate qPCR experiment: producing publication quality, reproducible data the first time. Trends Biotechnol. 37, 761–774 (2019). 45. Kupec, T. et al. Diagnostic potential of serum circulating miRNAs for endometriosis in patients with chronic pelvic pain. J. Clin. Med. 14, 5154 (2025). 46. Chang, L., Zhou, G., Soufan, O. & Xia, J. miRNet 2.0: network-based visual analytics for miRNA functional analysis and systems biology. Nucleic Acids Res. 48, W244–W251 (2020). 47. Kanehisa, M. & Goto, S. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. 28,2 7–30 (2000). 48. Nagai, T. et al. Focal adhesion kinase-mediated sequences, including cell adhesion, inflammatory response, and fibrosis, as a therapeutic target in endometriosis. Reprod. Sci. 27, 1400–1410 (2020). 49. Bora, G. & Yaba, A. The role of mitogen-activated protein kinase signaling pathway in endometriosis. J. Obstet. Gynaecol. Res. 47, 1610–1623 (2021). 50. Uimari, O. et al. Genome-wide genetic analyses highlight mitogen- activated protein kinase (MAPK) signaling in the pathogenesis of endometriosis. Hum. Reprod. 32, 780–793 (2017). 51. Solayman, M. H. M. et al. Identi fication of suitable endogenous normalizers for qRT-PCR analysis of plasma microRNA expression in essential hypertension. Mol. Biotechnol. 58, 179–187 (2016). 52. Leva, G. D. & Croce, C. M. miRNA pro filing of cancer. Curr. Opin. Genet. Dev. 23,3 –11 (2013). 53. Vasu, S. et al. MicroRNA signatures as future biomarkers for diagnosis of diabetes states. Cells 8, 1533 (2019). 54. Yan, L. et al. miR-21-5p induces cell proliferation by targeting TGFBI in non-small cell lung cancer cells. Exp. Ther. Med. 16, 4655–4663 (2018). 55. Park, S., Yang, H. D., Seo, J., Nam, J. & Nam, S. hnRNPC induces isoform shifts in miR-21-5p leading to cancer development. Exp. Mol. Med. 54, 812– 824 (2022). 56. González-López, P. et al. Role of miR-15a-5p and miR-199a-3p in the inflammatory pathway regulated by NF-kB in experimental and human atherosclerosis. Clin. Transl. Med. 13, e1362 (2023). 57. Liu, D. & Guan, Y. Mechanism of action of miR-15a-5p and miR-152- 3p in paraquat-induced pulmonary fibrosis through Wnt/β-catenin signaling mediation. PeerJ 12, e17662 (2024). 58. Pritchard, C. C. et al. Blood cell origin of circulating microRNAs: a cautionary note for cancer biomarker studies. Cancer Prev. Res. 5, 492–497 (2012). 59. Eyileten, C. et al. Thrombosis-related circulating miR-16-5p is associated with disease severity in patients hospitalized for COVID- 19. RNA Biol. 19, 963–979 (2022). 60. Geng, X., Guo, X., Wang, T., Xu, J. & Jiang, L. miR-92a-3p regulates neuropathic pain and neuroinflammation by regulating the expression of WNT5A. J. Neuroimmunol. 407, 578695 (2025). 61. Fujiwara, M. et al. microRNA-92a promotes CNS autoimmunity by modulating the regulatory and in flammatory T cell balance. J. Clin. Invest. 132, e155693 (2022). 62. Patil, A. H. & Halushka, M. K. miRge3.0: a comprehensive microRNA and tRF sequencing analysis pipeline. NAR Genom. Bioinform. 3, lqab068 (2021). 63. Marcel, M. Cutadapt removes adapter sequences from high- throughput sequencing reads. EMBnet.journal 17,1 0–12 (2011). 64. Kozomara, A., Birgaoanu, M. & Grif fiths-Jones, S. miRbase: from microRNA sequences to function. Nucleic Acid Res. 47, D155–D162 (2019). 65. Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 10, R25 (2009). 66. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq with DESeq2. Genome Biol. 15, 550 (2014).

The authors wish to express their gratitude to Jonathan Zhao and Frank Zhang (both from Heranova Lifesciences) for their valuable input on the study design and contributions to the development of the manuscript. This study was funded by the internal R&D budget of Heranova Lifesciences. Author contributions Y. Yu and W.H. Wong led the research group, analyzed the data, and wrote the manuscript. Y. Yu, W.H. Wong, X. Zhang, and F.Z. Bischoff conceptualized the study. X. Zhang and L. Zhu provided the samples while S. Lu coordinated sample collection between the hospital and the laboratory. W.H. Wong and Y. Hu performed random forest analysis. Y. Yu, Y. Shen, and X. Xu performed molecular experiments. Competing interests All authors, except X.Z. and L.Z., are employees of Heranova Lifesciences, a company engaged in the commercial development of a non-invasive test for endometriosis. F.Z.B. holds stock options of Heranova Lifesciences. The authors declare no other conflicts of interest, financial or otherwise. https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 9 Additional information Supplementary informationThe online version contains supplementary material available at https://doi.org/10.1038/s44294-025-00116-5 . Correspondenceand requests for materials should be addressed to Xinmei Zhang or Farideh Z. Bischoff. Reprints and permissions informationis available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modi fied the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party

in this article are included in the article ’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by- nc-nd/4.0/ . © The Author(s) 2025 https://doi.org/10.1038/s44294-025-00116-5 Article npj Women's Health | (2025) 3:67 10